From 2d207ed1eccae17ca5ba29920841d0e31c27e3f0 Mon Sep 17 00:00:00 2001 From: Mark Mentovai Date: Wed, 28 May 2025 16:17:37 -0400 Subject: [PATCH 001/250] t: run tests from a normalized working directory Some tests make git perform actions that produce observable pathnames, and have expectations on those paths. Tests run with $HOME set to a $TRASH_DIRECTORY, and with their working directory the same $TRASH_DIRECTORY, although these paths are logically identical, they do not observe the same pathname canonicalization rules and thus might not be represented by strings that compare equal. In particular, no pathname normalization is applied to $TRASH_DIRECTORY or $HOME, while tests change their working directory with `cd -P`, which normalizes the working directory's path by fully resolving symbolic links. t7900's macOS maintenance tests (which are not limited to running on macOS) have an expectation on a path that `git maintenance` forms by using abspath.c strbuf_realpath() to resolve a canonical absolute path based on $HOME. When t7900 runs from a working directory that contains symbolic links in its pathname, $HOME will also contain symbolic links, which `git maintenance` resolves but the test's expectation does not, causing a test failure. Align $TRASH_DIRECTORY and $HOME with the normalized path as used for the working directory by resetting them to match the working directory after it's established by `cd -P`. With all paths in agreement and symbolic links resolved, pathname expectations can be set and met based on string comparison without regard to external environmental factors such as the presence of symbolic links in a path. Suggested-by: Junio C Hamano Signed-off-by: Mark Mentovai Signed-off-by: Junio C Hamano --- t/test-lib.sh | 2 ++ 1 file changed, 2 insertions(+) diff --git a/t/test-lib.sh b/t/test-lib.sh index 9001ed3a6470a2..e2e91b5b24df90 100644 --- a/t/test-lib.sh +++ b/t/test-lib.sh @@ -1567,6 +1567,8 @@ fi # Use -P to resolve symlinks in our working directory so that the cwd # in subprocesses like git equals our $PWD (for pathname comparisons). cd -P "$TRASH_DIRECTORY" || BAIL_OUT "cannot cd -P to \"$TRASH_DIRECTORY\"" +TRASH_DIRECTORY=$(pwd) +HOME="$TRASH_DIRECTORY" start_test_output "$0" From d584219107440825e19a13a4d9cd607f51b5e575 Mon Sep 17 00:00:00 2001 From: Brad Smith Date: Sun, 1 Jun 2025 23:14:43 -0400 Subject: [PATCH 002/250] config.mak.uname: update settings for OpenBSD OpenBSD requires DIR_HAS_BSD_GROUP_SEMANTICS. OpenBSD has never had the BSD sysctl KERN_PROC_PATHNAME nor does it support or use the /proc filesystem. OpenBSD has had strcasestr() since 3.8. OpenBSD has had memmem() since 5.4. Signed-off-by: Brad Smith Signed-off-by: Junio C Hamano --- config.mak.uname | 5 +---- 1 file changed, 1 insertion(+), 4 deletions(-) diff --git a/config.mak.uname b/config.mak.uname index b12d4e168ae119..362c34aa764a28 100644 --- a/config.mak.uname +++ b/config.mak.uname @@ -294,17 +294,14 @@ ifeq ($(uname_S),FreeBSD) FILENO_IS_A_MACRO = UnfortunatelyYes endif ifeq ($(uname_S),OpenBSD) - NO_STRCASESTR = YesPlease - NO_MEMMEM = YesPlease + DIR_HAS_BSD_GROUP_SEMANTICS = YesPlease USE_ST_TIMESPEC = YesPlease NEEDS_LIBICONV = YesPlease BASIC_CFLAGS += -I/usr/local/include BASIC_LDFLAGS += -L/usr/local/lib HAVE_PATHS_H = YesPlease HAVE_BSD_SYSCTL = YesPlease - HAVE_BSD_KERN_PROC_SYSCTL = YesPlease CSPRNG_METHOD = arc4random - PROCFS_EXECUTABLE_PATH = /proc/curproc/file FREAD_READS_DIRECTORIES = UnfortunatelyYes FILENO_IS_A_MACRO = UnfortunatelyYes endif From db170e18262ed10e5e9b7f3cbca6a0ac43267b20 Mon Sep 17 00:00:00 2001 From: Collin Funk Date: Mon, 2 Jun 2025 15:31:32 -0700 Subject: [PATCH 003/250] completion: make sed command that generates config-list.h portable. The OpenBSD 'sed' command does not support '\n' to represent newlines in sed expressions. This leads to the follow compiler error: In file included from builtin/help.c:15: ./config-list.h:282:18: error: use of undeclared identifier 'n' "gitcvs.dbUser",n "gitcvs.dbPass", ^ 1 error generated. gmake: *** [Makefile:2821: builtin/help.o] Error 1 We can fix this by documenting related configuration variables one-per-line instead of listing them separated by commas. This allows us to remove the unportable part of the sed expression in generate-configlist.sh. Signed-off-by: Collin Funk Reviewed-by: Jacob Keller Signed-off-by: Junio C Hamano --- Documentation/config/gitcvs.adoc | 3 ++- Documentation/config/http.adoc | 3 ++- generate-configlist.sh | 1 - 3 files changed, 4 insertions(+), 3 deletions(-) diff --git a/Documentation/config/gitcvs.adoc b/Documentation/config/gitcvs.adoc index 02da427fd9743f..31d7be3992b572 100644 --- a/Documentation/config/gitcvs.adoc +++ b/Documentation/config/gitcvs.adoc @@ -47,7 +47,8 @@ gitcvs.dbDriver:: May not contain double colons (`:`). Default: 'SQLite'. See linkgit:git-cvsserver[1]. -gitcvs.dbUser, gitcvs.dbPass:: +gitcvs.dbUser:: +gitcvs.dbPass:: Database user and password. Only useful if setting `gitcvs.dbDriver`, since SQLite has no concept of database users and/or passwords. 'gitcvs.dbUser' supports variable substitution (see diff --git a/Documentation/config/http.adoc b/Documentation/config/http.adoc index 67393282fad57b..9da5c298cc1d5e 100644 --- a/Documentation/config/http.adoc +++ b/Documentation/config/http.adoc @@ -289,7 +289,8 @@ for most push problems, but can increase memory consumption significantly since the entire buffer is allocated even for small pushes. -http.lowSpeedLimit, http.lowSpeedTime:: +http.lowSpeedLimit:: +http.lowSpeedTime:: If the HTTP transfer speed, in bytes per second, is less than 'http.lowSpeedLimit' for longer than 'http.lowSpeedTime' seconds, the transfer is aborted. diff --git a/generate-configlist.sh b/generate-configlist.sh index b06da53c891819..9d2ad6165d60cd 100755 --- a/generate-configlist.sh +++ b/generate-configlist.sh @@ -19,7 +19,6 @@ EOF s/::$//; s/`//g; s/^.*$/ "&",/; - s/, */",\n "/g; p;}; d' \ "$SOURCE_DIR"/Documentation/*config.adoc \ From 35ec1e22452fa9e56e81f53bd91ef688ad3b6624 Mon Sep 17 00:00:00 2001 From: Brad Smith Date: Mon, 2 Jun 2025 03:29:02 -0400 Subject: [PATCH 004/250] compat: fixes for header handling with OpenBSD / NetBSD Handle OpenBSD and NetBSD as FreeBSD / DragonFly are. OpenBSD would need _XOPEN_SOURCE to be set to 700. Its simpler to just not set _XOPEN_SOURCE. CC strbuf.o strbuf.c:645:6: warning: call to undeclared function 'getdelim'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration] r = getdelim(&sb->buf, &sb->alloc, term, fp); ^ 1 warning generated. Signed-off-by: Brad Smith Reviewed-by: Collin Funk Signed-off-by: Junio C Hamano --- compat/posix.h | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) diff --git a/compat/posix.h b/compat/posix.h index f4c71f9427d774..067a00f33b83f3 100644 --- a/compat/posix.h +++ b/compat/posix.h @@ -60,11 +60,11 @@ # else # define _XOPEN_SOURCE 500 # endif -#elif !defined(__APPLE__) && !defined(__FreeBSD__) && !defined(__USLC__) && \ - !defined(_M_UNIX) && !defined(__sgi) && !defined(__DragonFly__) && \ - !defined(__TANDEM) && !defined(__QNX__) && !defined(__MirBSD__) && \ - !defined(__CYGWIN__) -#define _XOPEN_SOURCE 600 /* glibc2 and AIX 5.3L need 500, OpenBSD needs 600 for S_ISLNK() */ +#elif !defined(__APPLE__) && !defined(__FreeBSD__) && !defined(__NetBSD__) && \ + !defined(__OpenBSD__) && !defined(__DragonFly__) && !defined(__MirBSD__) && \ + !defined(__USLC__) && !defined(_M_UNIX) && !defined(__sgi) && \ + !defined(__TANDEM) && !defined(__QNX__) && !defined(__CYGWIN__) +#define _XOPEN_SOURCE 600 /* glibc2 and AIX 5.3L need 500 */ #define _XOPEN_SOURCE_EXTENDED 1 /* AIX 5.3L needs this */ #endif #define _ALL_SOURCE 1 From 6f11c42e8edc5cf7d65156c9dd68e720f1b92229 Mon Sep 17 00:00:00 2001 From: Jeff King Date: Wed, 4 Jun 2025 16:55:13 -0400 Subject: [PATCH 005/250] curl: fix integer constant typechecks with curl_easy_setopt() The curl documentation specifies that curl_easy_setopt() takes either: ...a long, a function pointer, an object pointer or a curl_off_t, depending on what the specific option expects. But when we pass an integer constant like "0", it will by default be a regular non-long int. This has always been wrong, but seemed to work in practice (I didn't dig into curl's implementation to see whether this might actually be triggering undefined behavior, but it seems likely and regardless we should do what the docs say). This is especially important since curl has a type-checking macro that causes building against curl 8.14 to produce many warnings. The specific commit is due to their 79b4e56b3 (typecheck-gcc.h: fix the typechecks, 2025-04-22). Curiously, it does only seem to trigger when compiled with -O2 for me. We can fix it by just marking the constants with a long "L". Signed-off-by: Jeff King Signed-off-by: Junio C Hamano --- http-push.c | 2 +- http.c | 14 +++++++------- remote-curl.c | 6 +++--- 3 files changed, 11 insertions(+), 11 deletions(-) diff --git a/http-push.c b/http-push.c index f9e67cabd4bee8..591e46ab260d57 100644 --- a/http-push.c +++ b/http-push.c @@ -195,7 +195,7 @@ static char *xml_entities(const char *s) static void curl_setup_http_get(CURL *curl, const char *url, const char *custom_req) { - curl_easy_setopt(curl, CURLOPT_HTTPGET, 1); + curl_easy_setopt(curl, CURLOPT_HTTPGET, 1L); curl_easy_setopt(curl, CURLOPT_URL, url); curl_easy_setopt(curl, CURLOPT_CUSTOMREQUEST, custom_req); curl_easy_setopt(curl, CURLOPT_WRITEFUNCTION, fwrite_null); diff --git a/http.c b/http.c index 3c029cf8947df7..cce2ea72873687 100644 --- a/http.c +++ b/http.c @@ -1019,13 +1019,13 @@ static CURL *get_curl_handle(void) die("curl_easy_init failed"); if (!curl_ssl_verify) { - curl_easy_setopt(result, CURLOPT_SSL_VERIFYPEER, 0); - curl_easy_setopt(result, CURLOPT_SSL_VERIFYHOST, 0); + curl_easy_setopt(result, CURLOPT_SSL_VERIFYPEER, 0L); + curl_easy_setopt(result, CURLOPT_SSL_VERIFYHOST, 0L); } else { /* Verify authenticity of the peer's certificate */ - curl_easy_setopt(result, CURLOPT_SSL_VERIFYPEER, 1); + curl_easy_setopt(result, CURLOPT_SSL_VERIFYPEER, 1L); /* The name in the cert must match whom we tried to connect */ - curl_easy_setopt(result, CURLOPT_SSL_VERIFYHOST, 2); + curl_easy_setopt(result, CURLOPT_SSL_VERIFYHOST, 2L); } if (curl_http_version) { @@ -1117,7 +1117,7 @@ static CURL *get_curl_handle(void) curl_low_speed_time); } - curl_easy_setopt(result, CURLOPT_MAXREDIRS, 20); + curl_easy_setopt(result, CURLOPT_MAXREDIRS, 20L); curl_easy_setopt(result, CURLOPT_POSTREDIR, CURL_REDIR_POST_ALL); #ifdef GIT_CURL_HAVE_CURLOPT_PROTOCOLS_STR @@ -1151,7 +1151,7 @@ static CURL *get_curl_handle(void) user_agent ? user_agent : git_user_agent()); if (curl_ftp_no_epsv) - curl_easy_setopt(result, CURLOPT_FTP_USE_EPSV, 0); + curl_easy_setopt(result, CURLOPT_FTP_USE_EPSV, 0L); if (curl_ssl_try) curl_easy_setopt(result, CURLOPT_USE_SSL, CURLUSESSL_TRY); @@ -1254,7 +1254,7 @@ static CURL *get_curl_handle(void) } init_curl_proxy_auth(result); - curl_easy_setopt(result, CURLOPT_TCP_KEEPALIVE, 1); + curl_easy_setopt(result, CURLOPT_TCP_KEEPALIVE, 1L); if (curl_tcp_keepidle > -1) curl_easy_setopt(result, CURLOPT_TCP_KEEPIDLE, diff --git a/remote-curl.c b/remote-curl.c index 590b228f67fcbc..6183772191f214 100644 --- a/remote-curl.c +++ b/remote-curl.c @@ -877,12 +877,12 @@ static int probe_rpc(struct rpc_state *rpc, struct slot_results *results) headers = curl_slist_append(headers, rpc->hdr_content_type); headers = curl_slist_append(headers, rpc->hdr_accept); - curl_easy_setopt(slot->curl, CURLOPT_NOBODY, 0); - curl_easy_setopt(slot->curl, CURLOPT_POST, 1); + curl_easy_setopt(slot->curl, CURLOPT_NOBODY, 0L); + curl_easy_setopt(slot->curl, CURLOPT_POST, 1L); curl_easy_setopt(slot->curl, CURLOPT_URL, rpc->service_url); curl_easy_setopt(slot->curl, CURLOPT_ENCODING, NULL); curl_easy_setopt(slot->curl, CURLOPT_POSTFIELDS, "0000"); - curl_easy_setopt(slot->curl, CURLOPT_POSTFIELDSIZE, 4); + curl_easy_setopt(slot->curl, CURLOPT_POSTFIELDSIZE, 4L); curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, headers); curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION, fwrite_buffer); curl_easy_setopt(slot->curl, CURLOPT_WRITEDATA, &buf); From 30325e23ba0d40567cc4ef78e4ba0c3776ef0c06 Mon Sep 17 00:00:00 2001 From: Jeff King Date: Wed, 4 Jun 2025 16:55:52 -0400 Subject: [PATCH 006/250] curl: fix integer variable typechecks with curl_easy_setopt() As discussed in the previous commit, we should be passing long integers, not regular ones, to curl_easy_setopt(), and compiling against curl 8.14 loudly complains if we don't. That patch fixed integer constants by adding an "L". This one deals with actual variables. Arguably these variables could just be declared as "long" in the first place. But it's actually kind of awkward due to other code which uses them: - port is conceptually a short, and we even call htons() on it (though weirdly it is defined as a regular int). - ssl_verify is conceptually a bool, and we assign to it from git_config_bool(). So I think we could probably switch these out for longs without hurting anything, but it just feels a bit weird. Doubly so because if you don't set USE_CURL_FOR_IMAP_SEND set, then the current types are fine! So let's just cast these to longs in the curl calls, which makes what's going on obvious. There aren't that many spots to modify (and as you can see from the context, we already have some similar casts). Signed-off-by: Jeff King Signed-off-by: Junio C Hamano --- imap-send.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/imap-send.c b/imap-send.c index 27dc033c7f8e33..2e812f5a6e9e10 100644 --- a/imap-send.c +++ b/imap-send.c @@ -1420,7 +1420,7 @@ static CURL *setup_curl(struct imap_server_conf *srvc, struct credential *cred) curl_easy_setopt(curl, CURLOPT_URL, path.buf); strbuf_release(&path); - curl_easy_setopt(curl, CURLOPT_PORT, srvc->port); + curl_easy_setopt(curl, CURLOPT_PORT, (long)srvc->port); if (srvc->auth_method) { struct strbuf auth = STRBUF_INIT; @@ -1433,8 +1433,8 @@ static CURL *setup_curl(struct imap_server_conf *srvc, struct credential *cred) if (!srvc->use_ssl) curl_easy_setopt(curl, CURLOPT_USE_SSL, (long)CURLUSESSL_TRY); - curl_easy_setopt(curl, CURLOPT_SSL_VERIFYPEER, srvc->ssl_verify); - curl_easy_setopt(curl, CURLOPT_SSL_VERIFYHOST, srvc->ssl_verify); + curl_easy_setopt(curl, CURLOPT_SSL_VERIFYPEER, (long)srvc->ssl_verify); + curl_easy_setopt(curl, CURLOPT_SSL_VERIFYHOST, (long)srvc->ssl_verify); curl_easy_setopt(curl, CURLOPT_READFUNCTION, fread_buffer); From 4558c8f84b2f8d3ba1483727bcb49935ae8ff595 Mon Sep 17 00:00:00 2001 From: Jeff King Date: Wed, 4 Jun 2025 16:56:22 -0400 Subject: [PATCH 007/250] curl: fix symbolic constant typechecks with curl_easy_setopt() As with the previous two commits, we should be passing long integers, not regular ones, to curl_easy_setopt(), and compiling against curl 8.14 loudly complains if we don't. This patch catches the remaining cases, which are ones where we pass curl's own symbolic constants. We'll cast them to long manually in each call. It seems kind of weird to me that curl doesn't define these constants as longs, since the point of them is to pass to curl_easy_setopt(). But in the curl documentation and examples, they clearly show casting them as part of the setopt calls. It may be that there is some reason not to push the type into the macro, like backwards compatibility. I didn't dig, as it doesn't really matter: we have to follow what existing curl versions ask for anyway. Signed-off-by: Jeff King Signed-off-by: Junio C Hamano --- http.c | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-) diff --git a/http.c b/http.c index cce2ea72873687..ecbc47ea4b3f27 100644 --- a/http.c +++ b/http.c @@ -1057,7 +1057,7 @@ static CURL *get_curl_handle(void) if (http_ssl_backend && !strcmp("schannel", http_ssl_backend) && !http_schannel_check_revoke) { - curl_easy_setopt(result, CURLOPT_SSL_OPTIONS, CURLSSLOPT_NO_REVOKE); + curl_easy_setopt(result, CURLOPT_SSL_OPTIONS, (long)CURLSSLOPT_NO_REVOKE); } if (http_proactive_auth != PROACTIVE_AUTH_NONE) @@ -1118,7 +1118,7 @@ static CURL *get_curl_handle(void) } curl_easy_setopt(result, CURLOPT_MAXREDIRS, 20L); - curl_easy_setopt(result, CURLOPT_POSTREDIR, CURL_REDIR_POST_ALL); + curl_easy_setopt(result, CURLOPT_POSTREDIR, (long)CURL_REDIR_POST_ALL); #ifdef GIT_CURL_HAVE_CURLOPT_PROTOCOLS_STR { @@ -1193,18 +1193,18 @@ static CURL *get_curl_handle(void) if (starts_with(curl_http_proxy, "socks5h")) curl_easy_setopt(result, - CURLOPT_PROXYTYPE, CURLPROXY_SOCKS5_HOSTNAME); + CURLOPT_PROXYTYPE, (long)CURLPROXY_SOCKS5_HOSTNAME); else if (starts_with(curl_http_proxy, "socks5")) curl_easy_setopt(result, - CURLOPT_PROXYTYPE, CURLPROXY_SOCKS5); + CURLOPT_PROXYTYPE, (long)CURLPROXY_SOCKS5); else if (starts_with(curl_http_proxy, "socks4a")) curl_easy_setopt(result, - CURLOPT_PROXYTYPE, CURLPROXY_SOCKS4A); + CURLOPT_PROXYTYPE, (long)CURLPROXY_SOCKS4A); else if (starts_with(curl_http_proxy, "socks")) curl_easy_setopt(result, - CURLOPT_PROXYTYPE, CURLPROXY_SOCKS4); + CURLOPT_PROXYTYPE, (long)CURLPROXY_SOCKS4); else if (starts_with(curl_http_proxy, "https")) { - curl_easy_setopt(result, CURLOPT_PROXYTYPE, CURLPROXY_HTTPS); + curl_easy_setopt(result, CURLOPT_PROXYTYPE, (long)CURLPROXY_HTTPS); if (http_proxy_ssl_cert) curl_easy_setopt(result, CURLOPT_PROXY_SSLCERT, http_proxy_ssl_cert); From 52a86dd26d998394e2f9b82a2097c76a701138b2 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Thu, 5 Jun 2025 10:16:45 +0000 Subject: [PATCH 008/250] t5410: avoid hangs in CI runs in the win+Meson test jobs In the GitHub workflow used in Git's CI builds, the `vs test` jobs use a subset of a specific revision of Git for Windows' SDK to run Git's test suite. This revision is validated by another CI workflow to ensure that said revision _can_ run Git's test suite successfully, skipping buggy updates in Git for Windows' SDK. The `win+Meson test` jobs do things differently, quite differently. They use the Bash of the Git for Windows version that is installed on the runners to run Git's test suite. This difference has consequences. When 68cb0b5253a0 (builtin/receive-pack: add option to skip connectivity check, 2025-05-20) introduced a test case that uses `tee | git receive-pack` as `--receive-pack` parameter (imitating an existing pattern in the same test script), it hit just the sweet spot to trigger a bug in the MSYS2 runtime shipped in Git for Windows v2.49.0. This version is the one currently installed on GitHub's runners. The problem is that the `git receive-pack` process finishes while the `tee` process does not need to write anything anymore and therefore does not receive an EOF. Instead, it should receive a SIGPIPE, but the bug in the MSYS2 runtime prevents that from working as intended. As a consequence, the `tee` process waits for more input from the `git.exe send-pack` process but none is coming, and the test script patiently waits until the 6h timeout hits. Only every once in a while, the `git receive-pack` process manages to send an EOF to the `tee` process and no hang occurs. Therefore, the problem can be worked around by cancelling the clearly-hanging job after twenty or so minutes and re-running it, repeating the process about half a dozen times, until the hang was successfully avoided. This bug in the MSYS2 runtime has been fixed in the meantime, which is the reason why the same test case causes no problems in the `win test` and the `vs test` jobs. This will continue to be the case until the Git for Windows version on the GitHub runners is upgraded to a version that distributes a newer MSYS2 runtime version. However, as of time of writing, this _is_ the latest Git for Windows version, and will be for another 1.5 weeks, until Git v2.50.0 is scheduled to appear (and shortly thereafter Git for Windows v2.50.0). Traditionally it takes a while before the runners pick up the new version. We could just wait it out, six hours at a time. Here, I opt for an alternative: Detect the buggy MSYS2 runtime and simply skip the test case. It's not like the `receive-pack` test cases are specific to Windows, and even then, to my chagrin the CI runs in git-for-windows/git spend around ten hours of compute time each and every time to run the entire test suite on all the platforms, even the tests that cover cross-platform code, and for Windows alone we do that three times: with GCC, with MSVC, and with MSVC via Meson. Therefore, I deem it more than acceptable to skip this test case in one of those matrices. For good luck, also the preceding test case is skipped in that scenario, as it uses the same `--receive-pack=tee | git receive-pack` pattern, even though I never observed that test case to hang in practice. Signed-off-by: Johannes Schindelin Signed-off-by: Junio C Hamano --- t/t5410-receive-pack.sh | 17 +++++++++++++++-- 1 file changed, 15 insertions(+), 2 deletions(-) diff --git a/t/t5410-receive-pack.sh b/t/t5410-receive-pack.sh index f76a22943ef0d1..09d6bfd2a10f5a 100755 --- a/t/t5410-receive-pack.sh +++ b/t/t5410-receive-pack.sh @@ -41,7 +41,19 @@ test_expect_success 'with core.alternateRefsPrefixes' ' test_cmp expect actual.haves ' -test_expect_success 'receive-pack missing objects fails connectivity check' ' +# The `tee.exe` shipped in Git for Windows v2.49.0 is known to hang frequently +# when spawned from `git.exe` and piping its output to `git.exe`. This seems +# related to MSYS2 runtime bug fixes regarding the signal handling; Let's just +# skip the tests that need to exercise this when the faulty MSYS2 runtime is +# detected; The test cases are exercised enough in other matrix jobs of the CI +# runs. +test_lazy_prereq TEE_DOES_NOT_HANG ' + test_have_prereq !MINGW && + case "$(uname -a)" in *3.5.7-463ebcdc.x86_64*) false;; esac +' + +test_expect_success TEE_DOES_NOT_HANG \ + 'receive-pack missing objects fails connectivity check' ' test_when_finished rm -rf repo remote.git setup.git && git init repo && @@ -62,7 +74,8 @@ test_expect_success 'receive-pack missing objects fails connectivity check' ' test_must_fail git -C remote.git cat-file -e $(git -C repo rev-parse HEAD) ' -test_expect_success 'receive-pack missing objects bypasses connectivity check' ' +test_expect_success TEE_DOES_NOT_HANG \ + 'receive-pack missing objects bypasses connectivity check' ' test_when_finished rm -rf repo remote.git setup.git && git init repo && From 765f1db2b5a890c0fa01f8976f197a8483357afe Mon Sep 17 00:00:00 2001 From: Johannes Sixt Date: Fri, 6 Jun 2025 07:41:42 +0200 Subject: [PATCH 009/250] git-gui: don't delete source files when auto_mkindex fails Commit 2cc5b0facfa4 (git-gui: extract script to generate "tclIndex", 2025-03-11) converted commands in a Makefile rule to a shell script. In this process, the Makefile variable $@ had to be replaced by the file name that it represents, 'lib/tclIndex'. However, the occurrence in `rm -f $@` was missed. In a shell script, $@ expands to all command line arguments, which happen to be the source files lib/*.tcl in this case. Needless to say that we do not want to remove source files during a build. Replace $@ by the intended 'lib/tclIndex'. Reported-by: Randall S. Becker Signed-off-by: Johannes Sixt --- generate-tclindex.sh | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/generate-tclindex.sh b/generate-tclindex.sh index 36e3a0bd90eefa..0b031d8339c443 100755 --- a/generate-tclindex.sh +++ b/generate-tclindex.sh @@ -23,7 +23,7 @@ then : ok else echo >&2 " * $TCL_PATH failed; using unoptimized loading" - rm -f $@ + rm -f lib/tclIndex echo '# Autogenerated by git-gui Makefile' >lib/tclIndex echo >>lib/tclIndex echo "class.tcl" >>lib/tclIndex From 229d12665edb555df26653cf4a4731ae71c71a6d Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Fri, 6 Jun 2025 09:29:24 +0000 Subject: [PATCH 010/250] curl: pass `long` values where expected As of Homebrew's update to cURL v8.14.0, there are new compile errors to be observed in the `osx-gcc` job of Git's CI builds: In file included from http.h:8, from imap-send.c:36: In function 'setup_curl', inlined from 'curl_append_msgs_to_imap' at imap-send.c:1460:9, inlined from 'cmd_main' at imap-send.c:1581:9: /usr/local/Cellar/curl/8.14.0/include/curl/typecheck-gcc.h:50:15: error: call to '_curl_easy_setopt_err_long' declared with attribute warning: curl_easy_setopt expects a long argument [-Werror=attribute-warning] 50 | _curl_easy_setopt_err_long(); \ | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~ /usr/local/Cellar/curl/8.14.0/include/curl/curl.h:54:7: note: in definition of macro 'CURL_IGNORE_DEPRECATION' 54 | statements \ | ^~~~~~~~~~ imap-send.c:1423:9: note: in expansion of macro 'curl_easy_setopt' 1423 | curl_easy_setopt(curl, CURLOPT_PORT, srvc->port); | ^~~~~~~~~~~~~~~~ [... many more instances of nearly identical warnings...] See for example this CI workflow run: https://github.com/git/git/actions/runs/15454602308/job/43504278284#step:4:307 The most likely explanation is the entry "typecheck-gcc.h: fix the typechecks" in cURL's release notes (https://curl.se/ch/8.14.0.html). Nearly identical compile errors afflicted recently-updated Debian setups, which have been addressed by `jk/curl-easy-setopt-typefix`. However, on macOS Git is built with different build options, which uncovered more instances of `int` values that need to be cast to constants, which were not covered by 6f11c42e8edc (curl: fix integer constant typechecks with curl_easy_setopt(), 2025-06-04). Let's explicitly convert even those remaining `int` constants in `curl_easy_setopt()` calls to `long` parameters. In addition to looking at the compile errors of the `osx-gcc` job, I verified that there are no other instances of the same issue that need to be handled in this manner (and that might not be caught by our CI builds because of yet other build options that might skip those code parts), I ran the following command and inspected all 23 results manually to ensure that the fix is now actually complete: git grep -n curl_easy_setopt | grep -ve ',.*, *[A-Za-z_"&]' \ -e ',.*, *[-0-9]*L)' \ -e ',.*,.* (long)' Signed-off-by: Johannes Schindelin Signed-off-by: Junio C Hamano --- http-push.c | 6 +++--- http.c | 22 +++++++++++----------- remote-curl.c | 6 +++--- 3 files changed, 17 insertions(+), 17 deletions(-) diff --git a/http-push.c b/http-push.c index 591e46ab260d57..f5a92529a8406b 100644 --- a/http-push.c +++ b/http-push.c @@ -205,7 +205,7 @@ static void curl_setup_http(CURL *curl, const char *url, const char *custom_req, struct buffer *buffer, curl_write_callback write_fn) { - curl_easy_setopt(curl, CURLOPT_UPLOAD, 1); + curl_easy_setopt(curl, CURLOPT_UPLOAD, 1L); curl_easy_setopt(curl, CURLOPT_URL, url); curl_easy_setopt(curl, CURLOPT_INFILE, buffer); curl_easy_setopt(curl, CURLOPT_INFILESIZE, buffer->buf.len); @@ -213,9 +213,9 @@ static void curl_setup_http(CURL *curl, const char *url, curl_easy_setopt(curl, CURLOPT_SEEKFUNCTION, seek_buffer); curl_easy_setopt(curl, CURLOPT_SEEKDATA, buffer); curl_easy_setopt(curl, CURLOPT_WRITEFUNCTION, write_fn); - curl_easy_setopt(curl, CURLOPT_NOBODY, 0); + curl_easy_setopt(curl, CURLOPT_NOBODY, 0L); curl_easy_setopt(curl, CURLOPT_CUSTOMREQUEST, custom_req); - curl_easy_setopt(curl, CURLOPT_UPLOAD, 1); + curl_easy_setopt(curl, CURLOPT_UPLOAD, 1L); } static struct curl_slist *get_dav_token_headers(struct remote_lock *lock, enum dav_header_flag options) diff --git a/http.c b/http.c index ecbc47ea4b3f27..d88e79fbde9c4c 100644 --- a/http.c +++ b/http.c @@ -1540,9 +1540,9 @@ struct active_request_slot *get_active_slot(void) curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION, NULL); curl_easy_setopt(slot->curl, CURLOPT_POSTFIELDS, NULL); curl_easy_setopt(slot->curl, CURLOPT_POSTFIELDSIZE, -1L); - curl_easy_setopt(slot->curl, CURLOPT_UPLOAD, 0); - curl_easy_setopt(slot->curl, CURLOPT_HTTPGET, 1); - curl_easy_setopt(slot->curl, CURLOPT_FAILONERROR, 1); + curl_easy_setopt(slot->curl, CURLOPT_UPLOAD, 0L); + curl_easy_setopt(slot->curl, CURLOPT_HTTPGET, 1L); + curl_easy_setopt(slot->curl, CURLOPT_FAILONERROR, 1L); curl_easy_setopt(slot->curl, CURLOPT_RANGE, NULL); /* @@ -1551,9 +1551,9 @@ struct active_request_slot *get_active_slot(void) * HTTP_FOLLOW_* cases themselves. */ if (http_follow_config == HTTP_FOLLOW_ALWAYS) - curl_easy_setopt(slot->curl, CURLOPT_FOLLOWLOCATION, 1); + curl_easy_setopt(slot->curl, CURLOPT_FOLLOWLOCATION, 1L); else - curl_easy_setopt(slot->curl, CURLOPT_FOLLOWLOCATION, 0); + curl_easy_setopt(slot->curl, CURLOPT_FOLLOWLOCATION, 0L); curl_easy_setopt(slot->curl, CURLOPT_IPRESOLVE, git_curl_ipresolve); curl_easy_setopt(slot->curl, CURLOPT_HTTPAUTH, http_auth_methods); @@ -2120,12 +2120,12 @@ static int http_request(const char *url, int ret; slot = get_active_slot(); - curl_easy_setopt(slot->curl, CURLOPT_HTTPGET, 1); + curl_easy_setopt(slot->curl, CURLOPT_HTTPGET, 1L); if (!result) { - curl_easy_setopt(slot->curl, CURLOPT_NOBODY, 1); + curl_easy_setopt(slot->curl, CURLOPT_NOBODY, 1L); } else { - curl_easy_setopt(slot->curl, CURLOPT_NOBODY, 0); + curl_easy_setopt(slot->curl, CURLOPT_NOBODY, 0L); curl_easy_setopt(slot->curl, CURLOPT_WRITEDATA, result); if (target == HTTP_REQUEST_FILE) { @@ -2151,7 +2151,7 @@ static int http_request(const char *url, strbuf_addstr(&buf, " no-cache"); if (options && options->initial_request && http_follow_config == HTTP_FOLLOW_INITIAL) - curl_easy_setopt(slot->curl, CURLOPT_FOLLOWLOCATION, 1); + curl_easy_setopt(slot->curl, CURLOPT_FOLLOWLOCATION, 1L); headers = curl_slist_append(headers, buf.buf); @@ -2170,7 +2170,7 @@ static int http_request(const char *url, curl_easy_setopt(slot->curl, CURLOPT_URL, url); curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, headers); curl_easy_setopt(slot->curl, CURLOPT_ENCODING, ""); - curl_easy_setopt(slot->curl, CURLOPT_FAILONERROR, 0); + curl_easy_setopt(slot->curl, CURLOPT_FAILONERROR, 0L); ret = run_one_slot(slot, &results); @@ -2750,7 +2750,7 @@ struct http_object_request *new_http_object_request(const char *base_url, freq->headers = object_request_headers(); curl_easy_setopt(freq->slot->curl, CURLOPT_WRITEDATA, freq); - curl_easy_setopt(freq->slot->curl, CURLOPT_FAILONERROR, 0); + curl_easy_setopt(freq->slot->curl, CURLOPT_FAILONERROR, 0L); curl_easy_setopt(freq->slot->curl, CURLOPT_WRITEFUNCTION, fwrite_sha1_file); curl_easy_setopt(freq->slot->curl, CURLOPT_ERRORBUFFER, freq->errorstr); curl_easy_setopt(freq->slot->curl, CURLOPT_URL, freq->url); diff --git a/remote-curl.c b/remote-curl.c index 6183772191f214..b8bc3a80cf4142 100644 --- a/remote-curl.c +++ b/remote-curl.c @@ -970,8 +970,8 @@ static int post_rpc(struct rpc_state *rpc, int stateless_connect, int flush_rece slot = get_active_slot(); - curl_easy_setopt(slot->curl, CURLOPT_NOBODY, 0); - curl_easy_setopt(slot->curl, CURLOPT_POST, 1); + curl_easy_setopt(slot->curl, CURLOPT_NOBODY, 0L); + curl_easy_setopt(slot->curl, CURLOPT_POST, 1L); curl_easy_setopt(slot->curl, CURLOPT_URL, rpc->service_url); curl_easy_setopt(slot->curl, CURLOPT_ENCODING, ""); @@ -1058,7 +1058,7 @@ static int post_rpc(struct rpc_state *rpc, int stateless_connect, int flush_rece rpc_in_data.check_pktline = stateless_connect; memset(&rpc_in_data.pktline_state, 0, sizeof(rpc_in_data.pktline_state)); curl_easy_setopt(slot->curl, CURLOPT_WRITEDATA, &rpc_in_data); - curl_easy_setopt(slot->curl, CURLOPT_FAILONERROR, 0); + curl_easy_setopt(slot->curl, CURLOPT_FAILONERROR, 0L); rpc->any_written = 0; From 8db3019401f8093fb895f581be641fe7e05aa755 Mon Sep 17 00:00:00 2001 From: Junio C Hamano Date: Sat, 7 Jun 2025 10:46:34 -0700 Subject: [PATCH 011/250] A bit more before -rc2 Signed-off-by: Junio C Hamano --- Documentation/RelNotes/2.50.0.adoc | 6 ++++++ 1 file changed, 6 insertions(+) diff --git a/Documentation/RelNotes/2.50.0.adoc b/Documentation/RelNotes/2.50.0.adoc index 95349ea50c001e..c9adfed33999e9 100644 --- a/Documentation/RelNotes/2.50.0.adoc +++ b/Documentation/RelNotes/2.50.0.adoc @@ -201,6 +201,12 @@ Performance, Internal Implementation, Development Support etc. we ended up checking for these non-existent files repeatedly, which has been optimized by memoizing the non-existence. + * Build settings have been improved for BSD based systems. + + * Newer version of libcURL detected curl_easy_setopt() calls we made + with platform-natural "int" when we should have used "long", which + all have been corrected. + Fixes since v2.49 ----------------- From 4c0e625c091d4c648cec7319bafaed3cc81658e5 Mon Sep 17 00:00:00 2001 From: Junio C Hamano Date: Mon, 9 Jun 2025 07:17:31 -0700 Subject: [PATCH 012/250] Git 2.50-rc2 Signed-off-by: Junio C Hamano --- Documentation/RelNotes/2.50.0.adoc | 5 +++++ GIT-VERSION-GEN | 2 +- 2 files changed, 6 insertions(+), 1 deletion(-) diff --git a/Documentation/RelNotes/2.50.0.adoc b/Documentation/RelNotes/2.50.0.adoc index c9adfed33999e9..99acf814875246 100644 --- a/Documentation/RelNotes/2.50.0.adoc +++ b/Documentation/RelNotes/2.50.0.adoc @@ -207,6 +207,11 @@ Performance, Internal Implementation, Development Support etc. with platform-natural "int" when we should have used "long", which all have been corrected. + * Tests that compare $HOME and $(pwd), which should be the same + directory unless the tests chdir's around, would fail when the user + enters the test directory via symbolic links, which has been + corrected. + Fixes since v2.49 ----------------- diff --git a/GIT-VERSION-GEN b/GIT-VERSION-GEN index 1047b8d11de767..208e91a17f0455 100755 --- a/GIT-VERSION-GEN +++ b/GIT-VERSION-GEN @@ -1,6 +1,6 @@ #!/bin/sh -DEF_VER=v2.50.0-rc1 +DEF_VER=v2.50.0-rc2 LF=' ' From e73fefa143ab4e368ea08b81bdd4f2e10e388296 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Wed, 6 Nov 2024 20:34:50 +0100 Subject: [PATCH 013/250] sideband: mask control characters The output of `git clone` is a vital component for understanding what has happened when things go wrong. However, these logs are partially under the control of the remote server (via the "sideband", which typically contains what the remote `git pack-objects` process sends to `stderr`), and is currently not sanitized by Git. This makes Git susceptible to ANSI escape sequence injection (see CWE-150, https://cwe.mitre.org/data/definitions/150.html), which allows attackers to corrupt terminal state, to hide information, and even to insert characters into the input buffer (i.e. as if the user had typed those characters). To plug this vulnerability, disallow any control character in the sideband, replacing them instead with the common `^` (e.g. `^[` for `\x1b`, `^A` for `\x01`). There is likely a need for more fine-grained controls instead of using a "heavy hammer" like this, which will be introduced subsequently. Signed-off-by: Johannes Schindelin --- sideband.c | 17 +++++++++++++++-- t/t5409-colorize-remote-messages.sh | 12 ++++++++++++ 2 files changed, 27 insertions(+), 2 deletions(-) diff --git a/sideband.c b/sideband.c index 251e9615ed0239..81b1ff0805b3f6 100644 --- a/sideband.c +++ b/sideband.c @@ -66,6 +66,19 @@ void list_config_color_sideband_slots(struct string_list *list, const char *pref list_config_item(list, prefix, keywords[i].keyword); } +static void strbuf_add_sanitized(struct strbuf *dest, const char *src, int n) +{ + strbuf_grow(dest, n); + for (; n && *src; src++, n--) { + if (!iscntrl(*src) || *src == '\t' || *src == '\n') + strbuf_addch(dest, *src); + else { + strbuf_addch(dest, '^'); + strbuf_addch(dest, 0x40 + *src); + } + } +} + /* * Optionally highlight one keyword in remote output if it appears at the start * of the line. This should be called for a single line only, which is @@ -81,7 +94,7 @@ static void maybe_colorize_sideband(struct strbuf *dest, const char *src, int n) int i; if (!want_color_stderr(use_sideband_colors())) { - strbuf_add(dest, src, n); + strbuf_add_sanitized(dest, src, n); return; } @@ -114,7 +127,7 @@ static void maybe_colorize_sideband(struct strbuf *dest, const char *src, int n) } } - strbuf_add(dest, src, n); + strbuf_add_sanitized(dest, src, n); } diff --git a/t/t5409-colorize-remote-messages.sh b/t/t5409-colorize-remote-messages.sh index fa5de4500a4f50..d0745c391b2625 100755 --- a/t/t5409-colorize-remote-messages.sh +++ b/t/t5409-colorize-remote-messages.sh @@ -98,4 +98,16 @@ test_expect_success 'fallback to color.ui' ' grep "error: error" decoded ' +test_expect_success 'disallow (color) control sequences in sideband' ' + write_script .git/color-me-surprised <<-\EOF && + printf "error: Have you \\033[31mread\\033[m this?\\n" >&2 + exec "$@" + EOF + test_config_global uploadPack.packObjectshook ./color-me-surprised && + test_commit need-at-least-one-commit && + git clone --no-local . throw-away 2>stderr && + test_decode_color decoded && + test_grep ! RED decoded +' + test_done From e9c50c30d95102c5f4bb970efebd28322127d731 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Wed, 6 Nov 2024 21:07:51 +0100 Subject: [PATCH 014/250] sideband: introduce an "escape hatch" to allow control characters The preceding commit fixed the vulnerability whereas sideband messages (that are under the control of the remote server) could contain ANSI escape sequences that would be sent to the terminal verbatim. However, this fix may not be desirable under all circumstances, e.g. when remote servers deliberately add coloring to their messages to increase their urgency. To help with those use cases, give users a way to opt-out of the protections: `sideband.allowControlCharacters`. Signed-off-by: Johannes Schindelin --- Documentation/config.adoc | 2 ++ Documentation/config/sideband.adoc | 5 +++++ sideband.c | 10 ++++++++++ t/t5409-colorize-remote-messages.sh | 8 +++++++- 4 files changed, 24 insertions(+), 1 deletion(-) create mode 100644 Documentation/config/sideband.adoc diff --git a/Documentation/config.adoc b/Documentation/config.adoc index cc769251be339b..a8b04c4e5183d2 100644 --- a/Documentation/config.adoc +++ b/Documentation/config.adoc @@ -522,6 +522,8 @@ include::config/sequencer.adoc[] include::config/showbranch.adoc[] +include::config/sideband.adoc[] + include::config/sparse.adoc[] include::config/splitindex.adoc[] diff --git a/Documentation/config/sideband.adoc b/Documentation/config/sideband.adoc new file mode 100644 index 00000000000000..3fb5045cd79581 --- /dev/null +++ b/Documentation/config/sideband.adoc @@ -0,0 +1,5 @@ +sideband.allowControlCharacters:: + By default, control characters that are delivered via the sideband + are masked, to prevent potentially unwanted ANSI escape sequences + from being sent to the terminal. Use this config setting to override + this behavior. diff --git a/sideband.c b/sideband.c index 81b1ff0805b3f6..d1c326fa191068 100644 --- a/sideband.c +++ b/sideband.c @@ -26,6 +26,8 @@ static struct keyword_entry keywords[] = { { "error", GIT_COLOR_BOLD_RED }, }; +static int allow_control_characters; + /* Returns a color setting (GIT_COLOR_NEVER, etc). */ static int use_sideband_colors(void) { @@ -39,6 +41,9 @@ static int use_sideband_colors(void) if (use_sideband_colors_cached >= 0) return use_sideband_colors_cached; + git_config_get_bool("sideband.allowcontrolcharacters", + &allow_control_characters); + if (!git_config_get_string_tmp(key, &value)) use_sideband_colors_cached = git_config_colorbool(key, value); else if (!git_config_get_string_tmp("color.ui", &value)) @@ -68,6 +73,11 @@ void list_config_color_sideband_slots(struct string_list *list, const char *pref static void strbuf_add_sanitized(struct strbuf *dest, const char *src, int n) { + if (allow_control_characters) { + strbuf_add(dest, src, n); + return; + } + strbuf_grow(dest, n); for (; n && *src; src++, n--) { if (!iscntrl(*src) || *src == '\t' || *src == '\n') diff --git a/t/t5409-colorize-remote-messages.sh b/t/t5409-colorize-remote-messages.sh index d0745c391b2625..fb31e8525418a1 100755 --- a/t/t5409-colorize-remote-messages.sh +++ b/t/t5409-colorize-remote-messages.sh @@ -105,9 +105,15 @@ test_expect_success 'disallow (color) control sequences in sideband' ' EOF test_config_global uploadPack.packObjectshook ./color-me-surprised && test_commit need-at-least-one-commit && + git clone --no-local . throw-away 2>stderr && test_decode_color decoded && - test_grep ! RED decoded + test_grep ! RED decoded && + + rm -rf throw-away && + git -c sideband.allowControlCharacters clone --no-local . throw-away 2>stderr && + test_decode_color decoded && + test_grep RED decoded ' test_done From 8b0a078df8a4ca24dca854ad8c456787a55b26da Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Mon, 18 Nov 2024 21:42:57 +0100 Subject: [PATCH 015/250] sideband: do allow ANSI color sequences by default The preceding two commits introduced special handling of the sideband channel to neutralize ANSI escape sequences before sending the payload to the terminal, and `sideband.allowControlCharacters` to override that behavior. However, some `pre-receive` hooks that are actively used in practice want to color their messages and therefore rely on the fact that Git passes them through to the terminal. In contrast to other ANSI escape sequences, it is highly unlikely that coloring sequences can be essential tools in attack vectors that mislead Git users e.g. by hiding crucial information. Therefore we can have both: Continue to allow ANSI coloring sequences to be passed to the terminal, and neutralize all other ANSI escape sequences. Signed-off-by: Johannes Schindelin --- Documentation/config/sideband.adoc | 17 ++++++-- sideband.c | 61 ++++++++++++++++++++++++++--- t/t5409-colorize-remote-messages.sh | 16 +++++++- 3 files changed, 84 insertions(+), 10 deletions(-) diff --git a/Documentation/config/sideband.adoc b/Documentation/config/sideband.adoc index 3fb5045cd79581..f347fd6b33004a 100644 --- a/Documentation/config/sideband.adoc +++ b/Documentation/config/sideband.adoc @@ -1,5 +1,16 @@ sideband.allowControlCharacters:: By default, control characters that are delivered via the sideband - are masked, to prevent potentially unwanted ANSI escape sequences - from being sent to the terminal. Use this config setting to override - this behavior. + are masked, except ANSI color sequences. This prevents potentially + unwanted ANSI escape sequences from being sent to the terminal. Use + this config setting to override this behavior: ++ +-- + color:: + Allow ANSI color sequences, line feeds and horizontal tabs, + but mask all other control characters. This is the default. + false:: + Mask all control characters other than line feeds and + horizontal tabs. + true:: + Allow all control characters to be sent to the terminal. +-- diff --git a/sideband.c b/sideband.c index d1c326fa191068..9084ca234dcecc 100644 --- a/sideband.c +++ b/sideband.c @@ -26,7 +26,11 @@ static struct keyword_entry keywords[] = { { "error", GIT_COLOR_BOLD_RED }, }; -static int allow_control_characters; +static enum { + ALLOW_NO_CONTROL_CHARACTERS = 0, + ALLOW_ALL_CONTROL_CHARACTERS = 1, + ALLOW_ANSI_COLOR_SEQUENCES = 2 +} allow_control_characters = ALLOW_ANSI_COLOR_SEQUENCES; /* Returns a color setting (GIT_COLOR_NEVER, etc). */ static int use_sideband_colors(void) @@ -41,8 +45,24 @@ static int use_sideband_colors(void) if (use_sideband_colors_cached >= 0) return use_sideband_colors_cached; - git_config_get_bool("sideband.allowcontrolcharacters", - &allow_control_characters); + switch (git_config_get_maybe_bool("sideband.allowcontrolcharacters", &i)) { + case 0: /* Boolean value */ + allow_control_characters = i ? ALLOW_ALL_CONTROL_CHARACTERS : + ALLOW_NO_CONTROL_CHARACTERS; + break; + case -1: /* non-Boolean value */ + if (git_config_get_string_tmp("sideband.allowcontrolcharacters", + &value)) + ; /* huh? `get_maybe_bool()` returned -1 */ + else if (!strcmp(value, "color")) + allow_control_characters = ALLOW_ANSI_COLOR_SEQUENCES; + else + warning(_("unrecognized value for `sideband." + "allowControlCharacters`: '%s'"), value); + break; + default: + break; /* not configured */ + } if (!git_config_get_string_tmp(key, &value)) use_sideband_colors_cached = git_config_colorbool(key, value); @@ -71,9 +91,37 @@ void list_config_color_sideband_slots(struct string_list *list, const char *pref list_config_item(list, prefix, keywords[i].keyword); } +static int handle_ansi_color_sequence(struct strbuf *dest, const char *src, int n) +{ + int i; + + /* + * Valid ANSI color sequences are of the form + * + * ESC [ [ [; ]*] m + */ + + if (allow_control_characters != ALLOW_ANSI_COLOR_SEQUENCES || + n < 3 || src[0] != '\x1b' || src[1] != '[') + return 0; + + for (i = 2; i < n; i++) { + if (src[i] == 'm') { + strbuf_add(dest, src, i + 1); + return i; + } + if (!isdigit(src[i]) && src[i] != ';') + break; + } + + return 0; +} + static void strbuf_add_sanitized(struct strbuf *dest, const char *src, int n) { - if (allow_control_characters) { + int i; + + if (allow_control_characters == ALLOW_ALL_CONTROL_CHARACTERS) { strbuf_add(dest, src, n); return; } @@ -82,7 +130,10 @@ static void strbuf_add_sanitized(struct strbuf *dest, const char *src, int n) for (; n && *src; src++, n--) { if (!iscntrl(*src) || *src == '\t' || *src == '\n') strbuf_addch(dest, *src); - else { + else if ((i = handle_ansi_color_sequence(dest, src, n))) { + src += i; + n -= i; + } else { strbuf_addch(dest, '^'); strbuf_addch(dest, 0x40 + *src); } diff --git a/t/t5409-colorize-remote-messages.sh b/t/t5409-colorize-remote-messages.sh index fb31e8525418a1..a755c49a74e634 100755 --- a/t/t5409-colorize-remote-messages.sh +++ b/t/t5409-colorize-remote-messages.sh @@ -100,7 +100,7 @@ test_expect_success 'fallback to color.ui' ' test_expect_success 'disallow (color) control sequences in sideband' ' write_script .git/color-me-surprised <<-\EOF && - printf "error: Have you \\033[31mread\\033[m this?\\n" >&2 + printf "error: Have you \\033[31mread\\033[m this?\\a\\n" >&2 exec "$@" EOF test_config_global uploadPack.packObjectshook ./color-me-surprised && @@ -108,12 +108,24 @@ test_expect_success 'disallow (color) control sequences in sideband' ' git clone --no-local . throw-away 2>stderr && test_decode_color decoded && + test_grep RED decoded && + test_grep "\\^G" stderr && + tr -dc "\\007" actual && + test_must_be_empty actual && + + rm -rf throw-away && + git -c sideband.allowControlCharacters=false \ + clone --no-local . throw-away 2>stderr && + test_decode_color decoded && test_grep ! RED decoded && + test_grep "\\^G" stderr && rm -rf throw-away && git -c sideband.allowControlCharacters clone --no-local . throw-away 2>stderr && test_decode_color decoded && - test_grep RED decoded + test_grep RED decoded && + tr -dc "\\007" actual && + test_file_not_empty actual ' test_done From 1465d4c374e667962c9e5b0fb8c11f1cf9cf7bf8 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Wed, 30 Oct 2024 19:48:46 +0100 Subject: [PATCH 016/250] unix-socket: avoid leak when initialization fails When a Unix socket is initialized, the current directory's path is stored so that the cleanup code can `chdir()` back to where it was before exit. If the path that needs to be stored exceeds the default size of the `sun_path` attribute of `struct sockaddr_un` (which is defined as a 108-sized byte array on Linux), a larger buffer needs to be allocated so that it can hold the path, and it is the responsibility of the `unix_sockaddr_cleanup()` function to release that allocated memory. In Git's CI, this stack allocation is not necessary because the code is checked out to `/home/runner/work/git/git`. Concatenate the path `t/trash directory.t0301-credential-cache/.cache/git/credential/socket` and a terminating NUL, and you end up with 96 bytes, 12 shy of the default `sun_path` size. However, I use worktrees with slightly longer paths: `/home/me/projects/git/yes/i/nest/worktrees/to/organize/them/` is more in line with what I have. When I recently tried to locally reproduce a failure of the `linux-leaks` CI job, this t0301 test failed (where it had not failed in CI). The reason: When `credential-cache` tries to reach its daemon initially by calling `unix_sockaddr_init()`, it is expected that the daemon cannot be reached (the idea is to spin up the daemon in that case and try again). However, when this first call to `unix_sockaddr_init()` fails, the code returns early from the `unix_stream_connect()` function _without_ giving the cleanup code a chance to run, skipping the deallocation of above-mentioned path. The fix is easy: do not return early but instead go directly to the cleanup code. Signed-off-by: Johannes Schindelin --- unix-socket.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/unix-socket.c b/unix-socket.c index 8860203c3f46dc..1fa0cf6c15c721 100644 --- a/unix-socket.c +++ b/unix-socket.c @@ -84,7 +84,7 @@ int unix_stream_connect(const char *path, int disallow_chdir) struct unix_sockaddr_context ctx; if (unix_sockaddr_init(&sa, path, &ctx, disallow_chdir) < 0) - return -1; + goto fail; fd = socket(AF_UNIX, SOCK_STREAM, 0); if (fd < 0) goto fail; From 2ebb94f9e12891832fe56516061e4439fa84f24a Mon Sep 17 00:00:00 2001 From: Sverre Rabbelier Date: Sun, 24 Jul 2011 15:54:04 +0200 Subject: [PATCH 017/250] t9350: point out that refs are not updated correctly This happens only when the corresponding commits are not exported in the current fast-export run. This can happen either when the relevant commit is already marked, or when the commit is explicitly marked as UNINTERESTING with a negative ref by another argument. This breaks fast-export basec remote helpers. Signed-off-by: Sverre Rabbelier --- t/t9350-fast-export.sh | 11 +++++++++++ 1 file changed, 11 insertions(+) diff --git a/t/t9350-fast-export.sh b/t/t9350-fast-export.sh index 76619765fcbc98..da9cfc161f7900 100755 --- a/t/t9350-fast-export.sh +++ b/t/t9350-fast-export.sh @@ -905,4 +905,15 @@ test_expect_success 'fast-export handles --end-of-options' ' test_cmp expect actual ' +cat > expected << EOF +reset refs/heads/master +from $(git rev-parse master) + +EOF + +test_expect_failure 'refs are updated even if no commits need to be exported' ' + git fast-export master..master > actual && + test_cmp expected actual +' + test_done From f84d8facfd7d1999d3dd945d76327797293f1190 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Mon, 2 Jun 2025 14:45:24 +0200 Subject: [PATCH 018/250] mingw: avoid relative `#include`s We want to make them relative to the top-level directory. Signed-off-by: Johannes Schindelin --- compat/mingw.c | 20 ++++++++++---------- 1 file changed, 10 insertions(+), 10 deletions(-) diff --git a/compat/mingw.c b/compat/mingw.c index 8a9972a1ca19f7..cec19048617005 100644 --- a/compat/mingw.c +++ b/compat/mingw.c @@ -1,22 +1,22 @@ #define USE_THE_REPOSITORY_VARIABLE #define DISABLE_SIGN_COMPARE_WARNINGS -#include "../git-compat-util.h" +#include "git-compat-util.h" #include "win32.h" #include #include #include #include -#include "../strbuf.h" -#include "../run-command.h" -#include "../abspath.h" -#include "../alloc.h" +#include "strbuf.h" +#include "run-command.h" +#include "abspath.h" +#include "alloc.h" #include "win32/lazyload.h" -#include "../config.h" -#include "../environment.h" -#include "../trace2.h" -#include "../symlinks.h" -#include "../wrapper.h" +#include "config.h" +#include "environment.h" +#include "trace2.h" +#include "symlinks.h" +#include "wrapper.h" #include "dir.h" #include "gettext.h" #define SECURITY_WIN32 From 675644f428b8c8b7370cecb2fcef0c34591f2fa6 Mon Sep 17 00:00:00 2001 From: Jeff King Date: Mon, 13 Jan 2025 01:26:01 -0500 Subject: [PATCH 019/250] grep: prevent `^$` false match at end of file In some implementations, `regexec_buf()` assumes that it is fed lines; Without `REG_NOTEOL` it thinks the end of the buffer is the end of a line. Which makes sense, but trips up this case because we are not feeding lines, but rather a whole buffer. So the final newline is not the start of an empty line, but the true end of the buffer. This causes an interesting bug: $ echo content >file.txt $ git grep --no-index -n '^$' file.txt file.txt:2: This bug is fixed by making the end of the buffer consistently the end of the final line. The patch was applied from https://lore.kernel.org/git/20250113062601.GD767856@coredump.intra.peff.net/ Reported-by: Olly Betts Signed-off-by: Johannes Schindelin --- grep.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/grep.c b/grep.c index f8d535182c36d3..ba88dc1ccc35a4 100644 --- a/grep.c +++ b/grep.c @@ -1646,6 +1646,8 @@ static int grep_source_1(struct grep_opt *opt, struct grep_source *gs, int colle bol = gs->buf; left = gs->size; + if (left && gs->buf[left-1] == '\n') + left--; while (left) { const char *eol; int hit; From 93f431836378862df3d8be76882936f6e66d5882 Mon Sep 17 00:00:00 2001 From: Sverre Rabbelier Date: Sat, 28 Aug 2010 20:49:01 -0500 Subject: [PATCH 020/250] transport-helper: add trailing -- [PT: ensure we add an additional element to the argv array] Signed-off-by: Sverre Rabbelier Signed-off-by: Johannes Schindelin --- transport-helper.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/transport-helper.c b/transport-helper.c index 0789e5bca53282..3e3f9e66c39f02 100644 --- a/transport-helper.c +++ b/transport-helper.c @@ -499,6 +499,8 @@ static int get_exporter(struct transport *transport, for (size_t i = 0; i < revlist_args->nr; i++) strvec_push(&fastexport->args, revlist_args->items[i].string); + strvec_push(&fastexport->args, "--"); + fastexport->git_cmd = 1; return start_command(fastexport); } From 7063803a9d2392456208451c96b3b08b3f65e7c7 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Mon, 2 Jun 2025 14:46:43 +0200 Subject: [PATCH 021/250] mingw: order `#include`s alphabetically It allows for more consistent patches that way. Signed-off-by: Johannes Schindelin --- compat/mingw.c | 22 +++++++++++----------- 1 file changed, 11 insertions(+), 11 deletions(-) diff --git a/compat/mingw.c b/compat/mingw.c index cec19048617005..cf65ab4482e5fa 100644 --- a/compat/mingw.c +++ b/compat/mingw.c @@ -2,25 +2,25 @@ #define DISABLE_SIGN_COMPARE_WARNINGS #include "git-compat-util.h" -#include "win32.h" -#include -#include -#include -#include -#include "strbuf.h" -#include "run-command.h" #include "abspath.h" #include "alloc.h" -#include "win32/lazyload.h" #include "config.h" +#include "dir.h" #include "environment.h" -#include "trace2.h" +#include "gettext.h" +#include "run-command.h" +#include "strbuf.h" #include "symlinks.h" +#include "trace2.h" +#include "win32.h" +#include "win32/lazyload.h" #include "wrapper.h" -#include "dir.h" -#include "gettext.h" +#include +#include +#include #define SECURITY_WIN32 #include +#include #include #define STATUS_DELETE_PENDING ((NTSTATUS) 0xC0000056) From 7b644fa91e0d5bacc55c30be469e51a52f1ac492 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Sun, 20 Oct 2019 22:08:58 +0200 Subject: [PATCH 022/250] win32/pthread: avoid name clashes with winpthread When asking the mingw-w64 variant of GCC to compile C11 code, it seems to link implicitly to libwinpthread, which does implement a pthread emulation (that is more complete than Git's). In preparation for vendoring in mimalloc (which requires C11 support), let's keep preferring Git's own pthread emulation. To avoid linker errors where it thinks that the `pthread_self` and the `pthread_create` symbols are defined twice, let's give our version a `win32_` prefix, just like we already do for `pthread_join()`. Signed-off-by: Johannes Schindelin --- compat/win32/pthread.c | 6 +++--- compat/win32/pthread.h | 8 +++++--- 2 files changed, 8 insertions(+), 6 deletions(-) diff --git a/compat/win32/pthread.c b/compat/win32/pthread.c index 58980a529c3eb9..4cc6356f757d80 100644 --- a/compat/win32/pthread.c +++ b/compat/win32/pthread.c @@ -21,8 +21,8 @@ static unsigned __stdcall win32_start_routine(void *arg) return 0; } -int pthread_create(pthread_t *thread, const void *attr UNUSED, - void *(*start_routine)(void *), void *arg) +int win32_pthread_create(pthread_t *thread, const void *attr UNUSED, + void *(*start_routine)(void *), void *arg) { thread->arg = arg; thread->start_routine = start_routine; @@ -53,7 +53,7 @@ int win32_pthread_join(pthread_t *thread, void **value_ptr) } } -pthread_t pthread_self(void) +pthread_t win32_pthread_self(void) { pthread_t t = { NULL }; t.tid = GetCurrentThreadId(); diff --git a/compat/win32/pthread.h b/compat/win32/pthread.h index e2b5c4f64c9b91..476e78b911b857 100644 --- a/compat/win32/pthread.h +++ b/compat/win32/pthread.h @@ -50,8 +50,9 @@ typedef struct { DWORD tid; } pthread_t; -int pthread_create(pthread_t *thread, const void *unused, - void *(*start_routine)(void*), void *arg); +int win32_pthread_create(pthread_t *thread, const void *unused, + void *(*start_routine)(void*), void *arg); +#define pthread_create win32_pthread_create /* * To avoid the need of copying a struct, we use small macro wrapper to pass @@ -62,7 +63,8 @@ int pthread_create(pthread_t *thread, const void *unused, int win32_pthread_join(pthread_t *thread, void **value_ptr); #define pthread_equal(t1, t2) ((t1).tid == (t2).tid) -pthread_t pthread_self(void); +pthread_t win32_pthread_self(void); +#define pthread_self win32_pthread_self static inline void NORETURN pthread_exit(void *ret) { From b3374559706885dd61e456be4366dc040ff0cde2 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Fri, 12 Aug 2022 12:44:15 +0200 Subject: [PATCH 023/250] git-compat-util: avoid redeclaring _DEFAULT_SOURCE We are about to vendor in `mimalloc`'s source code which we will want to include `compat/posix.h` after defining that constant. Signed-off-by: Johannes Schindelin --- compat/posix.h | 2 ++ 1 file changed, 2 insertions(+) diff --git a/compat/posix.h b/compat/posix.h index 067a00f33b83f3..626d71256f8428 100644 --- a/compat/posix.h +++ b/compat/posix.h @@ -70,7 +70,9 @@ #define _ALL_SOURCE 1 #define _GNU_SOURCE 1 #define _BSD_SOURCE 1 +#ifndef _DEFAULT_SOURCE #define _DEFAULT_SOURCE 1 +#endif #define _NETBSD_SOURCE 1 #define _SGI_SOURCE 1 From 21b9b932a935a59c20ec974a969aba5eeb644cba Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Mon, 24 Jun 2019 21:31:30 +0200 Subject: [PATCH 024/250] Import the source code of mimalloc v2.2.3 Update to newer mimalloc versions like this: update_mimalloc () { test $# = 1 || { echo "Need a mimalloc version" 1>&2; return 1 }; for oneline in 'mimalloc: adjust for building inside Git' 'Import the source code of mimalloc'; do git revert -n HEAD^{/^"$oneline"} && git checkout HEAD -- Makefile && git commit -sm "Temporarily revert \"$oneline\"" -m 'In preparation for upgrading to a newer mimalloc version.' || return 1; done; for file in $(git show --format='%n' --name-only --diff-filter=A HEAD^{/^"Import the source code of mimalloc "}) compat/mimalloc/arena-abandon.c compat/mimalloc/free.c compat/mimalloc/libc.c compat/mimalloc/prim/prim.c compat/mimalloc/mimalloc-stats.h; do file2=${file#compat/mimalloc/}; case "$file2" in segment-cache.c) : no longer needed; continue ;; bitmap.h | *.c) file2=src/$file2 ;; *.h) file2=include/$file2 ;; esac; mkdir -p "${file%/*}" && git -C /usr/src/mimalloc/ show "$1":$file2 > "$file" && git add "$file" || { echo "Failed: $file2 -> $file" 1>&2; return 1 }; done; conv_sed='sed -n "/^ *eval/d;/ /p"' && git commit -sm "Import the source code of mimalloc $1" -m "Update to newer mimalloc versions like this:" -m "$(set | sed -n '/^update_mimalloc *() *$/,/^}/{s/^./ &/;p}')" -m ' update_mimalloc $MIMALLOC_VERSION' -m 'For convenience, you can set `MIMALLOC_VERSION` and then run:' -m ' eval "$(git show -s | '"$conv_sed"')"' || return 1; git cherry-pick HEAD^{/^'mimalloc: adjust for building inside Git'} || return 1 } update_mimalloc $MIMALLOC_VERSION For convenience, you can set `MIMALLOC_VERSION` and then run: eval "$(git show -s | sed -n "/^ *eval/d;/ /p")" Signed-off-by: Johannes Schindelin --- compat/mimalloc/LICENSE | 21 + compat/mimalloc/alloc-aligned.c | 360 ++++++ compat/mimalloc/alloc.c | 692 +++++++++++ compat/mimalloc/arena-abandon.c | 346 ++++++ compat/mimalloc/arena.c | 1044 ++++++++++++++++ compat/mimalloc/bitmap.c | 441 +++++++ compat/mimalloc/bitmap.h | 119 ++ compat/mimalloc/free.c | 569 +++++++++ compat/mimalloc/heap.c | 735 ++++++++++++ compat/mimalloc/init.c | 720 ++++++++++++ compat/mimalloc/libc.c | 334 ++++++ compat/mimalloc/mimalloc-stats.h | 103 ++ compat/mimalloc/mimalloc.h | 608 ++++++++++ compat/mimalloc/mimalloc/atomic.h | 554 +++++++++ compat/mimalloc/mimalloc/internal.h | 1120 ++++++++++++++++++ compat/mimalloc/mimalloc/prim.h | 419 +++++++ compat/mimalloc/mimalloc/track.h | 145 +++ compat/mimalloc/mimalloc/types.h | 707 +++++++++++ compat/mimalloc/options.c | 663 +++++++++++ compat/mimalloc/os.c | 713 +++++++++++ compat/mimalloc/page-queue.c | 390 ++++++ compat/mimalloc/page.c | 1043 ++++++++++++++++ compat/mimalloc/prim/osx/prim.c | 9 + compat/mimalloc/prim/prim.c | 76 ++ compat/mimalloc/prim/unix/prim.c | 907 ++++++++++++++ compat/mimalloc/prim/windows/prim.c | 825 +++++++++++++ compat/mimalloc/random.c | 258 ++++ compat/mimalloc/segment-map.c | 142 +++ compat/mimalloc/segment.c | 1701 +++++++++++++++++++++++++++ compat/mimalloc/stats.c | 629 ++++++++++ 30 files changed, 16393 insertions(+) create mode 100644 compat/mimalloc/LICENSE create mode 100644 compat/mimalloc/alloc-aligned.c create mode 100644 compat/mimalloc/alloc.c create mode 100644 compat/mimalloc/arena-abandon.c create mode 100644 compat/mimalloc/arena.c create mode 100644 compat/mimalloc/bitmap.c create mode 100644 compat/mimalloc/bitmap.h create mode 100644 compat/mimalloc/free.c create mode 100644 compat/mimalloc/heap.c create mode 100644 compat/mimalloc/init.c create mode 100644 compat/mimalloc/libc.c create mode 100644 compat/mimalloc/mimalloc-stats.h create mode 100644 compat/mimalloc/mimalloc.h create mode 100644 compat/mimalloc/mimalloc/atomic.h create mode 100644 compat/mimalloc/mimalloc/internal.h create mode 100644 compat/mimalloc/mimalloc/prim.h create mode 100644 compat/mimalloc/mimalloc/track.h create mode 100644 compat/mimalloc/mimalloc/types.h create mode 100644 compat/mimalloc/options.c create mode 100644 compat/mimalloc/os.c create mode 100644 compat/mimalloc/page-queue.c create mode 100644 compat/mimalloc/page.c create mode 100644 compat/mimalloc/prim/osx/prim.c create mode 100644 compat/mimalloc/prim/prim.c create mode 100644 compat/mimalloc/prim/unix/prim.c create mode 100644 compat/mimalloc/prim/windows/prim.c create mode 100644 compat/mimalloc/random.c create mode 100644 compat/mimalloc/segment-map.c create mode 100644 compat/mimalloc/segment.c create mode 100644 compat/mimalloc/stats.c diff --git a/compat/mimalloc/LICENSE b/compat/mimalloc/LICENSE new file mode 100644 index 00000000000000..53315ebee557ac --- /dev/null +++ b/compat/mimalloc/LICENSE @@ -0,0 +1,21 @@ +MIT License + +Copyright (c) 2018-2025 Microsoft Corporation, Daan Leijen + +Permission is hereby granted, free of charge, to any person obtaining a copy +of this software and associated documentation files (the "Software"), to deal +in the Software without restriction, including without limitation the rights +to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +copies of the Software, and to permit persons to whom the Software is +furnished to do so, subject to the following conditions: + +The above copyright notice and this permission notice shall be included in all +copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +SOFTWARE. diff --git a/compat/mimalloc/alloc-aligned.c b/compat/mimalloc/alloc-aligned.c new file mode 100644 index 00000000000000..3d3202eb574971 --- /dev/null +++ b/compat/mimalloc/alloc-aligned.c @@ -0,0 +1,360 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2021, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ + +#include "mimalloc.h" +#include "mimalloc/internal.h" +#include "mimalloc/prim.h" // mi_prim_get_default_heap + +#include // memset + +// ------------------------------------------------------ +// Aligned Allocation +// ------------------------------------------------------ + +static bool mi_malloc_is_naturally_aligned( size_t size, size_t alignment ) { + // objects up to `MI_MAX_ALIGN_GUARANTEE` are allocated aligned to their size (see `segment.c:_mi_segment_page_start`). + mi_assert_internal(_mi_is_power_of_two(alignment) && (alignment > 0)); + if (alignment > size) return false; + if (alignment <= MI_MAX_ALIGN_SIZE) return true; + const size_t bsize = mi_good_size(size); + return (bsize <= MI_MAX_ALIGN_GUARANTEE && (bsize & (alignment-1)) == 0); +} + +#if MI_GUARDED +static mi_decl_restrict void* mi_heap_malloc_guarded_aligned(mi_heap_t* heap, size_t size, size_t alignment, bool zero) mi_attr_noexcept { + // use over allocation for guarded blocksl + mi_assert_internal(alignment > 0 && alignment < MI_BLOCK_ALIGNMENT_MAX); + const size_t oversize = size + alignment - 1; + void* base = _mi_heap_malloc_guarded(heap, oversize, zero); + void* p = mi_align_up_ptr(base, alignment); + mi_track_align(base, p, (uint8_t*)p - (uint8_t*)base, size); + mi_assert_internal(mi_usable_size(p) >= size); + mi_assert_internal(_mi_is_aligned(p, alignment)); + return p; +} + +static void* mi_heap_malloc_zero_no_guarded(mi_heap_t* heap, size_t size, bool zero) { + const size_t rate = heap->guarded_sample_rate; + // only write if `rate!=0` so we don't write to the constant `_mi_heap_empty` + if (rate != 0) { heap->guarded_sample_rate = 0; } + void* p = _mi_heap_malloc_zero(heap, size, zero); + if (rate != 0) { heap->guarded_sample_rate = rate; } + return p; +} +#else +static void* mi_heap_malloc_zero_no_guarded(mi_heap_t* heap, size_t size, bool zero) { + return _mi_heap_malloc_zero(heap, size, zero); +} +#endif + +// Fallback aligned allocation that over-allocates -- split out for better codegen +static mi_decl_noinline void* mi_heap_malloc_zero_aligned_at_overalloc(mi_heap_t* const heap, const size_t size, const size_t alignment, const size_t offset, const bool zero) mi_attr_noexcept +{ + mi_assert_internal(size <= (MI_MAX_ALLOC_SIZE - MI_PADDING_SIZE)); + mi_assert_internal(alignment != 0 && _mi_is_power_of_two(alignment)); + + void* p; + size_t oversize; + if mi_unlikely(alignment > MI_BLOCK_ALIGNMENT_MAX) { + // use OS allocation for very large alignment and allocate inside a huge page (dedicated segment with 1 page) + // This can support alignments >= MI_SEGMENT_SIZE by ensuring the object can be aligned at a point in the + // first (and single) page such that the segment info is `MI_SEGMENT_SIZE` bytes before it (so it can be found by aligning the pointer down) + if mi_unlikely(offset != 0) { + // todo: cannot support offset alignment for very large alignments yet +#if MI_DEBUG > 0 + _mi_error_message(EOVERFLOW, "aligned allocation with a very large alignment cannot be used with an alignment offset (size %zu, alignment %zu, offset %zu)\n", size, alignment, offset); +#endif + return NULL; + } + oversize = (size <= MI_SMALL_SIZE_MAX ? MI_SMALL_SIZE_MAX + 1 /* ensure we use generic malloc path */ : size); + // note: no guarded as alignment > 0 + p = _mi_heap_malloc_zero_ex(heap, oversize, false, alignment); // the page block size should be large enough to align in the single huge page block + // zero afterwards as only the area from the aligned_p may be committed! + if (p == NULL) return NULL; + } + else { + // otherwise over-allocate + oversize = (size < MI_MAX_ALIGN_SIZE ? MI_MAX_ALIGN_SIZE : size) + alignment - 1; // adjust for size <= 16; with size 0 and aligment 64k, we would allocate a 64k block and pointing just beyond that. + p = mi_heap_malloc_zero_no_guarded(heap, oversize, zero); + if (p == NULL) return NULL; + } + mi_page_t* page = _mi_ptr_page(p); + + // .. and align within the allocation + const uintptr_t align_mask = alignment - 1; // for any x, `(x & align_mask) == (x % alignment)` + const uintptr_t poffset = ((uintptr_t)p + offset) & align_mask; + const uintptr_t adjust = (poffset == 0 ? 0 : alignment - poffset); + mi_assert_internal(adjust < alignment); + void* aligned_p = (void*)((uintptr_t)p + adjust); + if (aligned_p != p) { + mi_page_set_has_aligned(page, true); + #if MI_GUARDED + // set tag to aligned so mi_usable_size works with guard pages + if (adjust >= sizeof(mi_block_t)) { + mi_block_t* const block = (mi_block_t*)p; + block->next = MI_BLOCK_TAG_ALIGNED; + } + #endif + _mi_padding_shrink(page, (mi_block_t*)p, adjust + size); + } + // todo: expand padding if overallocated ? + + mi_assert_internal(mi_page_usable_block_size(page) >= adjust + size); + mi_assert_internal(((uintptr_t)aligned_p + offset) % alignment == 0); + mi_assert_internal(mi_usable_size(aligned_p)>=size); + mi_assert_internal(mi_usable_size(p) == mi_usable_size(aligned_p)+adjust); + #if MI_DEBUG > 1 + mi_page_t* const apage = _mi_ptr_page(aligned_p); + void* unalign_p = _mi_page_ptr_unalign(apage, aligned_p); + mi_assert_internal(p == unalign_p); + #endif + + // now zero the block if needed + if (alignment > MI_BLOCK_ALIGNMENT_MAX) { + // for the tracker, on huge aligned allocations only the memory from the start of the large block is defined + mi_track_mem_undefined(aligned_p, size); + if (zero) { + _mi_memzero_aligned(aligned_p, mi_usable_size(aligned_p)); + } + } + + if (p != aligned_p) { + mi_track_align(p,aligned_p,adjust,mi_usable_size(aligned_p)); + #if MI_GUARDED + mi_track_mem_defined(p, sizeof(mi_block_t)); + #endif + } + return aligned_p; +} + +// Generic primitive aligned allocation -- split out for better codegen +static mi_decl_noinline void* mi_heap_malloc_zero_aligned_at_generic(mi_heap_t* const heap, const size_t size, const size_t alignment, const size_t offset, const bool zero) mi_attr_noexcept +{ + mi_assert_internal(alignment != 0 && _mi_is_power_of_two(alignment)); + // we don't allocate more than MI_MAX_ALLOC_SIZE (see ) + if mi_unlikely(size > (MI_MAX_ALLOC_SIZE - MI_PADDING_SIZE)) { + #if MI_DEBUG > 0 + _mi_error_message(EOVERFLOW, "aligned allocation request is too large (size %zu, alignment %zu)\n", size, alignment); + #endif + return NULL; + } + + // use regular allocation if it is guaranteed to fit the alignment constraints. + // this is important to try as the fast path in `mi_heap_malloc_zero_aligned` only works when there exist + // a page with the right block size, and if we always use the over-alloc fallback that would never happen. + if (offset == 0 && mi_malloc_is_naturally_aligned(size,alignment)) { + void* p = mi_heap_malloc_zero_no_guarded(heap, size, zero); + mi_assert_internal(p == NULL || ((uintptr_t)p % alignment) == 0); + const bool is_aligned_or_null = (((uintptr_t)p) & (alignment-1))==0; + if mi_likely(is_aligned_or_null) { + return p; + } + else { + // this should never happen if the `mi_malloc_is_naturally_aligned` check is correct.. + mi_assert(false); + mi_free(p); + } + } + + // fall back to over-allocation + return mi_heap_malloc_zero_aligned_at_overalloc(heap,size,alignment,offset,zero); +} + + +// Primitive aligned allocation +static void* mi_heap_malloc_zero_aligned_at(mi_heap_t* const heap, const size_t size, const size_t alignment, const size_t offset, const bool zero) mi_attr_noexcept +{ + // note: we don't require `size > offset`, we just guarantee that the address at offset is aligned regardless of the allocated size. + if mi_unlikely(alignment == 0 || !_mi_is_power_of_two(alignment)) { // require power-of-two (see ) + #if MI_DEBUG > 0 + _mi_error_message(EOVERFLOW, "aligned allocation requires the alignment to be a power-of-two (size %zu, alignment %zu)\n", size, alignment); + #endif + return NULL; + } + + #if MI_GUARDED + if (offset==0 && alignment < MI_BLOCK_ALIGNMENT_MAX && mi_heap_malloc_use_guarded(heap,size)) { + return mi_heap_malloc_guarded_aligned(heap, size, alignment, zero); + } + #endif + + // try first if there happens to be a small block available with just the right alignment + if mi_likely(size <= MI_SMALL_SIZE_MAX && alignment <= size) { + const uintptr_t align_mask = alignment-1; // for any x, `(x & align_mask) == (x % alignment)` + const size_t padsize = size + MI_PADDING_SIZE; + mi_page_t* page = _mi_heap_get_free_small_page(heap, padsize); + if mi_likely(page->free != NULL) { + const bool is_aligned = (((uintptr_t)page->free + offset) & align_mask)==0; + if mi_likely(is_aligned) + { + void* p = (zero ? _mi_page_malloc_zeroed(heap,page,padsize) : _mi_page_malloc(heap,page,padsize)); // call specific page malloc for better codegen + mi_assert_internal(p != NULL); + mi_assert_internal(((uintptr_t)p + offset) % alignment == 0); + mi_track_malloc(p,size,zero); + return p; + } + } + } + + // fallback to generic aligned allocation + return mi_heap_malloc_zero_aligned_at_generic(heap, size, alignment, offset, zero); +} + + +// ------------------------------------------------------ +// Optimized mi_heap_malloc_aligned / mi_malloc_aligned +// ------------------------------------------------------ + +mi_decl_nodiscard mi_decl_restrict void* mi_heap_malloc_aligned_at(mi_heap_t* heap, size_t size, size_t alignment, size_t offset) mi_attr_noexcept { + return mi_heap_malloc_zero_aligned_at(heap, size, alignment, offset, false); +} + +mi_decl_nodiscard mi_decl_restrict void* mi_heap_malloc_aligned(mi_heap_t* heap, size_t size, size_t alignment) mi_attr_noexcept { + return mi_heap_malloc_aligned_at(heap, size, alignment, 0); +} + +// ensure a definition is emitted +#if defined(__cplusplus) +void* _mi_extern_heap_malloc_aligned = (void*)&mi_heap_malloc_aligned; +#endif + +// ------------------------------------------------------ +// Aligned Allocation +// ------------------------------------------------------ + +mi_decl_nodiscard mi_decl_restrict void* mi_heap_zalloc_aligned_at(mi_heap_t* heap, size_t size, size_t alignment, size_t offset) mi_attr_noexcept { + return mi_heap_malloc_zero_aligned_at(heap, size, alignment, offset, true); +} + +mi_decl_nodiscard mi_decl_restrict void* mi_heap_zalloc_aligned(mi_heap_t* heap, size_t size, size_t alignment) mi_attr_noexcept { + return mi_heap_zalloc_aligned_at(heap, size, alignment, 0); +} + +mi_decl_nodiscard mi_decl_restrict void* mi_heap_calloc_aligned_at(mi_heap_t* heap, size_t count, size_t size, size_t alignment, size_t offset) mi_attr_noexcept { + size_t total; + if (mi_count_size_overflow(count, size, &total)) return NULL; + return mi_heap_zalloc_aligned_at(heap, total, alignment, offset); +} + +mi_decl_nodiscard mi_decl_restrict void* mi_heap_calloc_aligned(mi_heap_t* heap, size_t count, size_t size, size_t alignment) mi_attr_noexcept { + return mi_heap_calloc_aligned_at(heap,count,size,alignment,0); +} + +mi_decl_nodiscard mi_decl_restrict void* mi_malloc_aligned_at(size_t size, size_t alignment, size_t offset) mi_attr_noexcept { + return mi_heap_malloc_aligned_at(mi_prim_get_default_heap(), size, alignment, offset); +} + +mi_decl_nodiscard mi_decl_restrict void* mi_malloc_aligned(size_t size, size_t alignment) mi_attr_noexcept { + return mi_heap_malloc_aligned(mi_prim_get_default_heap(), size, alignment); +} + +mi_decl_nodiscard mi_decl_restrict void* mi_zalloc_aligned_at(size_t size, size_t alignment, size_t offset) mi_attr_noexcept { + return mi_heap_zalloc_aligned_at(mi_prim_get_default_heap(), size, alignment, offset); +} + +mi_decl_nodiscard mi_decl_restrict void* mi_zalloc_aligned(size_t size, size_t alignment) mi_attr_noexcept { + return mi_heap_zalloc_aligned(mi_prim_get_default_heap(), size, alignment); +} + +mi_decl_nodiscard mi_decl_restrict void* mi_calloc_aligned_at(size_t count, size_t size, size_t alignment, size_t offset) mi_attr_noexcept { + return mi_heap_calloc_aligned_at(mi_prim_get_default_heap(), count, size, alignment, offset); +} + +mi_decl_nodiscard mi_decl_restrict void* mi_calloc_aligned(size_t count, size_t size, size_t alignment) mi_attr_noexcept { + return mi_heap_calloc_aligned(mi_prim_get_default_heap(), count, size, alignment); +} + + +// ------------------------------------------------------ +// Aligned re-allocation +// ------------------------------------------------------ + +static void* mi_heap_realloc_zero_aligned_at(mi_heap_t* heap, void* p, size_t newsize, size_t alignment, size_t offset, bool zero) mi_attr_noexcept { + mi_assert(alignment > 0); + if (alignment <= sizeof(uintptr_t)) return _mi_heap_realloc_zero(heap,p,newsize,zero); + if (p == NULL) return mi_heap_malloc_zero_aligned_at(heap,newsize,alignment,offset,zero); + size_t size = mi_usable_size(p); + if (newsize <= size && newsize >= (size - (size / 2)) + && (((uintptr_t)p + offset) % alignment) == 0) { + return p; // reallocation still fits, is aligned and not more than 50% waste + } + else { + // note: we don't zero allocate upfront so we only zero initialize the expanded part + void* newp = mi_heap_malloc_aligned_at(heap,newsize,alignment,offset); + if (newp != NULL) { + if (zero && newsize > size) { + // also set last word in the previous allocation to zero to ensure any padding is zero-initialized + size_t start = (size >= sizeof(intptr_t) ? size - sizeof(intptr_t) : 0); + _mi_memzero((uint8_t*)newp + start, newsize - start); + } + _mi_memcpy_aligned(newp, p, (newsize > size ? size : newsize)); + mi_free(p); // only free if successful + } + return newp; + } +} + +static void* mi_heap_realloc_zero_aligned(mi_heap_t* heap, void* p, size_t newsize, size_t alignment, bool zero) mi_attr_noexcept { + mi_assert(alignment > 0); + if (alignment <= sizeof(uintptr_t)) return _mi_heap_realloc_zero(heap,p,newsize,zero); + size_t offset = ((uintptr_t)p % alignment); // use offset of previous allocation (p can be NULL) + return mi_heap_realloc_zero_aligned_at(heap,p,newsize,alignment,offset,zero); +} + +mi_decl_nodiscard void* mi_heap_realloc_aligned_at(mi_heap_t* heap, void* p, size_t newsize, size_t alignment, size_t offset) mi_attr_noexcept { + return mi_heap_realloc_zero_aligned_at(heap,p,newsize,alignment,offset,false); +} + +mi_decl_nodiscard void* mi_heap_realloc_aligned(mi_heap_t* heap, void* p, size_t newsize, size_t alignment) mi_attr_noexcept { + return mi_heap_realloc_zero_aligned(heap,p,newsize,alignment,false); +} + +mi_decl_nodiscard void* mi_heap_rezalloc_aligned_at(mi_heap_t* heap, void* p, size_t newsize, size_t alignment, size_t offset) mi_attr_noexcept { + return mi_heap_realloc_zero_aligned_at(heap, p, newsize, alignment, offset, true); +} + +mi_decl_nodiscard void* mi_heap_rezalloc_aligned(mi_heap_t* heap, void* p, size_t newsize, size_t alignment) mi_attr_noexcept { + return mi_heap_realloc_zero_aligned(heap, p, newsize, alignment, true); +} + +mi_decl_nodiscard void* mi_heap_recalloc_aligned_at(mi_heap_t* heap, void* p, size_t newcount, size_t size, size_t alignment, size_t offset) mi_attr_noexcept { + size_t total; + if (mi_count_size_overflow(newcount, size, &total)) return NULL; + return mi_heap_rezalloc_aligned_at(heap, p, total, alignment, offset); +} + +mi_decl_nodiscard void* mi_heap_recalloc_aligned(mi_heap_t* heap, void* p, size_t newcount, size_t size, size_t alignment) mi_attr_noexcept { + size_t total; + if (mi_count_size_overflow(newcount, size, &total)) return NULL; + return mi_heap_rezalloc_aligned(heap, p, total, alignment); +} + +mi_decl_nodiscard void* mi_realloc_aligned_at(void* p, size_t newsize, size_t alignment, size_t offset) mi_attr_noexcept { + return mi_heap_realloc_aligned_at(mi_prim_get_default_heap(), p, newsize, alignment, offset); +} + +mi_decl_nodiscard void* mi_realloc_aligned(void* p, size_t newsize, size_t alignment) mi_attr_noexcept { + return mi_heap_realloc_aligned(mi_prim_get_default_heap(), p, newsize, alignment); +} + +mi_decl_nodiscard void* mi_rezalloc_aligned_at(void* p, size_t newsize, size_t alignment, size_t offset) mi_attr_noexcept { + return mi_heap_rezalloc_aligned_at(mi_prim_get_default_heap(), p, newsize, alignment, offset); +} + +mi_decl_nodiscard void* mi_rezalloc_aligned(void* p, size_t newsize, size_t alignment) mi_attr_noexcept { + return mi_heap_rezalloc_aligned(mi_prim_get_default_heap(), p, newsize, alignment); +} + +mi_decl_nodiscard void* mi_recalloc_aligned_at(void* p, size_t newcount, size_t size, size_t alignment, size_t offset) mi_attr_noexcept { + return mi_heap_recalloc_aligned_at(mi_prim_get_default_heap(), p, newcount, size, alignment, offset); +} + +mi_decl_nodiscard void* mi_recalloc_aligned(void* p, size_t newcount, size_t size, size_t alignment) mi_attr_noexcept { + return mi_heap_recalloc_aligned(mi_prim_get_default_heap(), p, newcount, size, alignment); +} + + diff --git a/compat/mimalloc/alloc.c b/compat/mimalloc/alloc.c new file mode 100644 index 00000000000000..0fed5e754c1918 --- /dev/null +++ b/compat/mimalloc/alloc.c @@ -0,0 +1,692 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2024, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ +#ifndef _DEFAULT_SOURCE +#define _DEFAULT_SOURCE // for realpath() on Linux +#endif + +#include "mimalloc.h" +#include "mimalloc/internal.h" +#include "mimalloc/atomic.h" +#include "mimalloc/prim.h" // _mi_prim_thread_id() + +#include // memset, strlen (for mi_strdup) +#include // malloc, abort + +#define MI_IN_ALLOC_C +#include "alloc-override.c" +#include "free.c" +#undef MI_IN_ALLOC_C + +// ------------------------------------------------------ +// Allocation +// ------------------------------------------------------ + +// Fast allocation in a page: just pop from the free list. +// Fall back to generic allocation only if the list is empty. +// Note: in release mode the (inlined) routine is about 7 instructions with a single test. +extern inline void* _mi_page_malloc_zero(mi_heap_t* heap, mi_page_t* page, size_t size, bool zero) mi_attr_noexcept +{ + mi_assert_internal(size >= MI_PADDING_SIZE); + mi_assert_internal(page->block_size == 0 /* empty heap */ || mi_page_block_size(page) >= size); + + // check the free list + mi_block_t* const block = page->free; + if mi_unlikely(block == NULL) { + return _mi_malloc_generic(heap, size, zero, 0); + } + mi_assert_internal(block != NULL && _mi_ptr_page(block) == page); + + // pop from the free list + page->free = mi_block_next(page, block); + page->used++; + mi_assert_internal(page->free == NULL || _mi_ptr_page(page->free) == page); + mi_assert_internal(page->block_size < MI_MAX_ALIGN_SIZE || _mi_is_aligned(block, MI_MAX_ALIGN_SIZE)); + + #if MI_DEBUG>3 + if (page->free_is_zero && size > sizeof(*block)) { + mi_assert_expensive(mi_mem_is_zero(block+1,size - sizeof(*block))); + } + #endif + + // allow use of the block internally + // note: when tracking we need to avoid ever touching the MI_PADDING since + // that is tracked by valgrind etc. as non-accessible (through the red-zone, see `mimalloc/track.h`) + mi_track_mem_undefined(block, mi_page_usable_block_size(page)); + + // zero the block? note: we need to zero the full block size (issue #63) + if mi_unlikely(zero) { + mi_assert_internal(page->block_size != 0); // do not call with zero'ing for huge blocks (see _mi_malloc_generic) + mi_assert_internal(!mi_page_is_huge(page)); + #if MI_PADDING + mi_assert_internal(page->block_size >= MI_PADDING_SIZE); + #endif + if (page->free_is_zero) { + block->next = 0; + mi_track_mem_defined(block, page->block_size - MI_PADDING_SIZE); + } + else { + _mi_memzero_aligned(block, page->block_size - MI_PADDING_SIZE); + } + } + + #if (MI_DEBUG>0) && !MI_TRACK_ENABLED && !MI_TSAN + if (!zero && !mi_page_is_huge(page)) { + memset(block, MI_DEBUG_UNINIT, mi_page_usable_block_size(page)); + } + #elif (MI_SECURE!=0) + if (!zero) { block->next = 0; } // don't leak internal data + #endif + + #if (MI_STAT>0) + const size_t bsize = mi_page_usable_block_size(page); + if (bsize <= MI_MEDIUM_OBJ_SIZE_MAX) { + mi_heap_stat_increase(heap, malloc_normal, bsize); + mi_heap_stat_counter_increase(heap, malloc_normal_count, 1); + #if (MI_STAT>1) + const size_t bin = _mi_bin(bsize); + mi_heap_stat_increase(heap, malloc_bins[bin], 1); + mi_heap_stat_increase(heap, malloc_requested, size - MI_PADDING_SIZE); + #endif + } + #endif + + #if MI_PADDING // && !MI_TRACK_ENABLED + mi_padding_t* const padding = (mi_padding_t*)((uint8_t*)block + mi_page_usable_block_size(page)); + ptrdiff_t delta = ((uint8_t*)padding - (uint8_t*)block - (size - MI_PADDING_SIZE)); + #if (MI_DEBUG>=2) + mi_assert_internal(delta >= 0 && mi_page_usable_block_size(page) >= (size - MI_PADDING_SIZE + delta)); + #endif + mi_track_mem_defined(padding,sizeof(mi_padding_t)); // note: re-enable since mi_page_usable_block_size may set noaccess + padding->canary = mi_ptr_encode_canary(page,block,page->keys); + padding->delta = (uint32_t)(delta); + #if MI_PADDING_CHECK + if (!mi_page_is_huge(page)) { + uint8_t* fill = (uint8_t*)padding - delta; + const size_t maxpad = (delta > MI_MAX_ALIGN_SIZE ? MI_MAX_ALIGN_SIZE : delta); // set at most N initial padding bytes + for (size_t i = 0; i < maxpad; i++) { fill[i] = MI_DEBUG_PADDING; } + } + #endif + #endif + + return block; +} + +// extra entries for improved efficiency in `alloc-aligned.c`. +extern void* _mi_page_malloc(mi_heap_t* heap, mi_page_t* page, size_t size) mi_attr_noexcept { + return _mi_page_malloc_zero(heap,page,size,false); +} +extern void* _mi_page_malloc_zeroed(mi_heap_t* heap, mi_page_t* page, size_t size) mi_attr_noexcept { + return _mi_page_malloc_zero(heap,page,size,true); +} + +#if MI_GUARDED +mi_decl_restrict void* _mi_heap_malloc_guarded(mi_heap_t* heap, size_t size, bool zero) mi_attr_noexcept; +#endif + +static inline mi_decl_restrict void* mi_heap_malloc_small_zero(mi_heap_t* heap, size_t size, bool zero) mi_attr_noexcept { + mi_assert(heap != NULL); + mi_assert(size <= MI_SMALL_SIZE_MAX); + #if MI_DEBUG + const uintptr_t tid = _mi_thread_id(); + mi_assert(heap->thread_id == 0 || heap->thread_id == tid); // heaps are thread local + #endif + #if (MI_PADDING || MI_GUARDED) + if (size == 0) { size = sizeof(void*); } + #endif + #if MI_GUARDED + if (mi_heap_malloc_use_guarded(heap,size)) { + return _mi_heap_malloc_guarded(heap, size, zero); + } + #endif + + // get page in constant time, and allocate from it + mi_page_t* page = _mi_heap_get_free_small_page(heap, size + MI_PADDING_SIZE); + void* const p = _mi_page_malloc_zero(heap, page, size + MI_PADDING_SIZE, zero); + mi_track_malloc(p,size,zero); + + #if MI_DEBUG>3 + if (p != NULL && zero) { + mi_assert_expensive(mi_mem_is_zero(p, size)); + } + #endif + return p; +} + +// allocate a small block +mi_decl_nodiscard extern inline mi_decl_restrict void* mi_heap_malloc_small(mi_heap_t* heap, size_t size) mi_attr_noexcept { + return mi_heap_malloc_small_zero(heap, size, false); +} + +mi_decl_nodiscard extern inline mi_decl_restrict void* mi_malloc_small(size_t size) mi_attr_noexcept { + return mi_heap_malloc_small(mi_prim_get_default_heap(), size); +} + +// The main allocation function +extern inline void* _mi_heap_malloc_zero_ex(mi_heap_t* heap, size_t size, bool zero, size_t huge_alignment) mi_attr_noexcept { + // fast path for small objects + if mi_likely(size <= MI_SMALL_SIZE_MAX) { + mi_assert_internal(huge_alignment == 0); + return mi_heap_malloc_small_zero(heap, size, zero); + } + #if MI_GUARDED + else if (huge_alignment==0 && mi_heap_malloc_use_guarded(heap,size)) { + return _mi_heap_malloc_guarded(heap, size, zero); + } + #endif + else { + // regular allocation + mi_assert(heap!=NULL); + mi_assert(heap->thread_id == 0 || heap->thread_id == _mi_thread_id()); // heaps are thread local + void* const p = _mi_malloc_generic(heap, size + MI_PADDING_SIZE, zero, huge_alignment); // note: size can overflow but it is detected in malloc_generic + mi_track_malloc(p,size,zero); + + #if MI_DEBUG>3 + if (p != NULL && zero) { + mi_assert_expensive(mi_mem_is_zero(p, size)); + } + #endif + return p; + } +} + +extern inline void* _mi_heap_malloc_zero(mi_heap_t* heap, size_t size, bool zero) mi_attr_noexcept { + return _mi_heap_malloc_zero_ex(heap, size, zero, 0); +} + +mi_decl_nodiscard extern inline mi_decl_restrict void* mi_heap_malloc(mi_heap_t* heap, size_t size) mi_attr_noexcept { + return _mi_heap_malloc_zero(heap, size, false); +} + +mi_decl_nodiscard extern inline mi_decl_restrict void* mi_malloc(size_t size) mi_attr_noexcept { + return mi_heap_malloc(mi_prim_get_default_heap(), size); +} + +// zero initialized small block +mi_decl_nodiscard mi_decl_restrict void* mi_zalloc_small(size_t size) mi_attr_noexcept { + return mi_heap_malloc_small_zero(mi_prim_get_default_heap(), size, true); +} + +mi_decl_nodiscard extern inline mi_decl_restrict void* mi_heap_zalloc(mi_heap_t* heap, size_t size) mi_attr_noexcept { + return _mi_heap_malloc_zero(heap, size, true); +} + +mi_decl_nodiscard mi_decl_restrict void* mi_zalloc(size_t size) mi_attr_noexcept { + return mi_heap_zalloc(mi_prim_get_default_heap(),size); +} + + +mi_decl_nodiscard extern inline mi_decl_restrict void* mi_heap_calloc(mi_heap_t* heap, size_t count, size_t size) mi_attr_noexcept { + size_t total; + if (mi_count_size_overflow(count,size,&total)) return NULL; + return mi_heap_zalloc(heap,total); +} + +mi_decl_nodiscard mi_decl_restrict void* mi_calloc(size_t count, size_t size) mi_attr_noexcept { + return mi_heap_calloc(mi_prim_get_default_heap(),count,size); +} + +// Uninitialized `calloc` +mi_decl_nodiscard extern mi_decl_restrict void* mi_heap_mallocn(mi_heap_t* heap, size_t count, size_t size) mi_attr_noexcept { + size_t total; + if (mi_count_size_overflow(count, size, &total)) return NULL; + return mi_heap_malloc(heap, total); +} + +mi_decl_nodiscard mi_decl_restrict void* mi_mallocn(size_t count, size_t size) mi_attr_noexcept { + return mi_heap_mallocn(mi_prim_get_default_heap(),count,size); +} + +// Expand (or shrink) in place (or fail) +void* mi_expand(void* p, size_t newsize) mi_attr_noexcept { + #if MI_PADDING + // we do not shrink/expand with padding enabled + MI_UNUSED(p); MI_UNUSED(newsize); + return NULL; + #else + if (p == NULL) return NULL; + const size_t size = _mi_usable_size(p,"mi_expand"); + if (newsize > size) return NULL; + return p; // it fits + #endif +} + +void* _mi_heap_realloc_zero(mi_heap_t* heap, void* p, size_t newsize, bool zero) mi_attr_noexcept { + // if p == NULL then behave as malloc. + // else if size == 0 then reallocate to a zero-sized block (and don't return NULL, just as mi_malloc(0)). + // (this means that returning NULL always indicates an error, and `p` will not have been freed in that case.) + const size_t size = _mi_usable_size(p,"mi_realloc"); // also works if p == NULL (with size 0) + if mi_unlikely(newsize <= size && newsize >= (size / 2) && newsize > 0) { // note: newsize must be > 0 or otherwise we return NULL for realloc(NULL,0) + mi_assert_internal(p!=NULL); + // todo: do not track as the usable size is still the same in the free; adjust potential padding? + // mi_track_resize(p,size,newsize) + // if (newsize < size) { mi_track_mem_noaccess((uint8_t*)p + newsize, size - newsize); } + return p; // reallocation still fits and not more than 50% waste + } + void* newp = mi_heap_malloc(heap,newsize); + if mi_likely(newp != NULL) { + if (zero && newsize > size) { + // also set last word in the previous allocation to zero to ensure any padding is zero-initialized + const size_t start = (size >= sizeof(intptr_t) ? size - sizeof(intptr_t) : 0); + _mi_memzero((uint8_t*)newp + start, newsize - start); + } + else if (newsize == 0) { + ((uint8_t*)newp)[0] = 0; // work around for applications that expect zero-reallocation to be zero initialized (issue #725) + } + if mi_likely(p != NULL) { + const size_t copysize = (newsize > size ? size : newsize); + mi_track_mem_defined(p,copysize); // _mi_useable_size may be too large for byte precise memory tracking.. + _mi_memcpy(newp, p, copysize); + mi_free(p); // only free the original pointer if successful + } + } + return newp; +} + +mi_decl_nodiscard void* mi_heap_realloc(mi_heap_t* heap, void* p, size_t newsize) mi_attr_noexcept { + return _mi_heap_realloc_zero(heap, p, newsize, false); +} + +mi_decl_nodiscard void* mi_heap_reallocn(mi_heap_t* heap, void* p, size_t count, size_t size) mi_attr_noexcept { + size_t total; + if (mi_count_size_overflow(count, size, &total)) return NULL; + return mi_heap_realloc(heap, p, total); +} + + +// Reallocate but free `p` on errors +mi_decl_nodiscard void* mi_heap_reallocf(mi_heap_t* heap, void* p, size_t newsize) mi_attr_noexcept { + void* newp = mi_heap_realloc(heap, p, newsize); + if (newp==NULL && p!=NULL) mi_free(p); + return newp; +} + +mi_decl_nodiscard void* mi_heap_rezalloc(mi_heap_t* heap, void* p, size_t newsize) mi_attr_noexcept { + return _mi_heap_realloc_zero(heap, p, newsize, true); +} + +mi_decl_nodiscard void* mi_heap_recalloc(mi_heap_t* heap, void* p, size_t count, size_t size) mi_attr_noexcept { + size_t total; + if (mi_count_size_overflow(count, size, &total)) return NULL; + return mi_heap_rezalloc(heap, p, total); +} + + +mi_decl_nodiscard void* mi_realloc(void* p, size_t newsize) mi_attr_noexcept { + return mi_heap_realloc(mi_prim_get_default_heap(),p,newsize); +} + +mi_decl_nodiscard void* mi_reallocn(void* p, size_t count, size_t size) mi_attr_noexcept { + return mi_heap_reallocn(mi_prim_get_default_heap(),p,count,size); +} + +// Reallocate but free `p` on errors +mi_decl_nodiscard void* mi_reallocf(void* p, size_t newsize) mi_attr_noexcept { + return mi_heap_reallocf(mi_prim_get_default_heap(),p,newsize); +} + +mi_decl_nodiscard void* mi_rezalloc(void* p, size_t newsize) mi_attr_noexcept { + return mi_heap_rezalloc(mi_prim_get_default_heap(), p, newsize); +} + +mi_decl_nodiscard void* mi_recalloc(void* p, size_t count, size_t size) mi_attr_noexcept { + return mi_heap_recalloc(mi_prim_get_default_heap(), p, count, size); +} + + + +// ------------------------------------------------------ +// strdup, strndup, and realpath +// ------------------------------------------------------ + +// `strdup` using mi_malloc +mi_decl_nodiscard mi_decl_restrict char* mi_heap_strdup(mi_heap_t* heap, const char* s) mi_attr_noexcept { + if (s == NULL) return NULL; + size_t len = _mi_strlen(s); + char* t = (char*)mi_heap_malloc(heap,len+1); + if (t == NULL) return NULL; + _mi_memcpy(t, s, len); + t[len] = 0; + return t; +} + +mi_decl_nodiscard mi_decl_restrict char* mi_strdup(const char* s) mi_attr_noexcept { + return mi_heap_strdup(mi_prim_get_default_heap(), s); +} + +// `strndup` using mi_malloc +mi_decl_nodiscard mi_decl_restrict char* mi_heap_strndup(mi_heap_t* heap, const char* s, size_t n) mi_attr_noexcept { + if (s == NULL) return NULL; + const size_t len = _mi_strnlen(s,n); // len <= n + char* t = (char*)mi_heap_malloc(heap, len+1); + if (t == NULL) return NULL; + _mi_memcpy(t, s, len); + t[len] = 0; + return t; +} + +mi_decl_nodiscard mi_decl_restrict char* mi_strndup(const char* s, size_t n) mi_attr_noexcept { + return mi_heap_strndup(mi_prim_get_default_heap(),s,n); +} + +#ifndef __wasi__ +// `realpath` using mi_malloc +#ifdef _WIN32 +#ifndef PATH_MAX +#define PATH_MAX MAX_PATH +#endif + +mi_decl_nodiscard mi_decl_restrict char* mi_heap_realpath(mi_heap_t* heap, const char* fname, char* resolved_name) mi_attr_noexcept { + // todo: use GetFullPathNameW to allow longer file names + char buf[PATH_MAX]; + DWORD res = GetFullPathNameA(fname, PATH_MAX, (resolved_name == NULL ? buf : resolved_name), NULL); + if (res == 0) { + errno = GetLastError(); return NULL; + } + else if (res > PATH_MAX) { + errno = EINVAL; return NULL; + } + else if (resolved_name != NULL) { + return resolved_name; + } + else { + return mi_heap_strndup(heap, buf, PATH_MAX); + } +} +#else +/* +#include // pathconf +static size_t mi_path_max(void) { + static size_t path_max = 0; + if (path_max <= 0) { + long m = pathconf("/",_PC_PATH_MAX); + if (m <= 0) path_max = 4096; // guess + else if (m < 256) path_max = 256; // at least 256 + else path_max = m; + } + return path_max; +} +*/ +char* mi_heap_realpath(mi_heap_t* heap, const char* fname, char* resolved_name) mi_attr_noexcept { + if (resolved_name != NULL) { + return realpath(fname,resolved_name); + } + else { + char* rname = realpath(fname, NULL); + if (rname == NULL) return NULL; + char* result = mi_heap_strdup(heap, rname); + mi_cfree(rname); // use checked free (which may be redirected to our free but that's ok) + // note: with ASAN realpath is intercepted and mi_cfree may leak the returned pointer :-( + return result; + } + /* + const size_t n = mi_path_max(); + char* buf = (char*)mi_malloc(n+1); + if (buf == NULL) { + errno = ENOMEM; + return NULL; + } + char* rname = realpath(fname,buf); + char* result = mi_heap_strndup(heap,rname,n); // ok if `rname==NULL` + mi_free(buf); + return result; + } + */ +} +#endif + +mi_decl_nodiscard mi_decl_restrict char* mi_realpath(const char* fname, char* resolved_name) mi_attr_noexcept { + return mi_heap_realpath(mi_prim_get_default_heap(),fname,resolved_name); +} +#endif + +/*------------------------------------------------------- +C++ new and new_aligned +The standard requires calling into `get_new_handler` and +throwing the bad_alloc exception on failure. If we compile +with a C++ compiler we can implement this precisely. If we +use a C compiler we cannot throw a `bad_alloc` exception +but we call `exit` instead (i.e. not returning). +-------------------------------------------------------*/ + +#ifdef __cplusplus +#include +static bool mi_try_new_handler(bool nothrow) { + #if defined(_MSC_VER) || (__cplusplus >= 201103L) + std::new_handler h = std::get_new_handler(); + #else + std::new_handler h = std::set_new_handler(); + std::set_new_handler(h); + #endif + if (h==NULL) { + _mi_error_message(ENOMEM, "out of memory in 'new'"); + #if defined(_CPPUNWIND) || defined(__cpp_exceptions) // exceptions are not always enabled + if (!nothrow) { + throw std::bad_alloc(); + } + #else + MI_UNUSED(nothrow); + #endif + return false; + } + else { + h(); + return true; + } +} +#else +typedef void (*std_new_handler_t)(void); + +#if (defined(__GNUC__) || (defined(__clang__) && !defined(_MSC_VER))) // exclude clang-cl, see issue #631 +std_new_handler_t __attribute__((weak)) _ZSt15get_new_handlerv(void) { + return NULL; +} +static std_new_handler_t mi_get_new_handler(void) { + return _ZSt15get_new_handlerv(); +} +#else +// note: on windows we could dynamically link to `?get_new_handler@std@@YAP6AXXZXZ`. +static std_new_handler_t mi_get_new_handler() { + return NULL; +} +#endif + +static bool mi_try_new_handler(bool nothrow) { + std_new_handler_t h = mi_get_new_handler(); + if (h==NULL) { + _mi_error_message(ENOMEM, "out of memory in 'new'"); + if (!nothrow) { + abort(); // cannot throw in plain C, use abort + } + return false; + } + else { + h(); + return true; + } +} +#endif + +mi_decl_export mi_decl_noinline void* mi_heap_try_new(mi_heap_t* heap, size_t size, bool nothrow ) { + void* p = NULL; + while(p == NULL && mi_try_new_handler(nothrow)) { + p = mi_heap_malloc(heap,size); + } + return p; +} + +static mi_decl_noinline void* mi_try_new(size_t size, bool nothrow) { + return mi_heap_try_new(mi_prim_get_default_heap(), size, nothrow); +} + + +mi_decl_nodiscard mi_decl_restrict void* mi_heap_alloc_new(mi_heap_t* heap, size_t size) { + void* p = mi_heap_malloc(heap,size); + if mi_unlikely(p == NULL) return mi_heap_try_new(heap, size, false); + return p; +} + +mi_decl_nodiscard mi_decl_restrict void* mi_new(size_t size) { + return mi_heap_alloc_new(mi_prim_get_default_heap(), size); +} + + +mi_decl_nodiscard mi_decl_restrict void* mi_heap_alloc_new_n(mi_heap_t* heap, size_t count, size_t size) { + size_t total; + if mi_unlikely(mi_count_size_overflow(count, size, &total)) { + mi_try_new_handler(false); // on overflow we invoke the try_new_handler once to potentially throw std::bad_alloc + return NULL; + } + else { + return mi_heap_alloc_new(heap,total); + } +} + +mi_decl_nodiscard mi_decl_restrict void* mi_new_n(size_t count, size_t size) { + return mi_heap_alloc_new_n(mi_prim_get_default_heap(), count, size); +} + + +mi_decl_nodiscard mi_decl_restrict void* mi_new_nothrow(size_t size) mi_attr_noexcept { + void* p = mi_malloc(size); + if mi_unlikely(p == NULL) return mi_try_new(size, true); + return p; +} + +mi_decl_nodiscard mi_decl_restrict void* mi_new_aligned(size_t size, size_t alignment) { + void* p; + do { + p = mi_malloc_aligned(size, alignment); + } + while(p == NULL && mi_try_new_handler(false)); + return p; +} + +mi_decl_nodiscard mi_decl_restrict void* mi_new_aligned_nothrow(size_t size, size_t alignment) mi_attr_noexcept { + void* p; + do { + p = mi_malloc_aligned(size, alignment); + } + while(p == NULL && mi_try_new_handler(true)); + return p; +} + +mi_decl_nodiscard void* mi_new_realloc(void* p, size_t newsize) { + void* q; + do { + q = mi_realloc(p, newsize); + } while (q == NULL && mi_try_new_handler(false)); + return q; +} + +mi_decl_nodiscard void* mi_new_reallocn(void* p, size_t newcount, size_t size) { + size_t total; + if mi_unlikely(mi_count_size_overflow(newcount, size, &total)) { + mi_try_new_handler(false); // on overflow we invoke the try_new_handler once to potentially throw std::bad_alloc + return NULL; + } + else { + return mi_new_realloc(p, total); + } +} + +#if MI_GUARDED +// We always allocate a guarded allocation at an offset (`mi_page_has_aligned` will be true). +// We then set the first word of the block to `0` for regular offset aligned allocations (in `alloc-aligned.c`) +// and the first word to `~0` for guarded allocations to have a correct `mi_usable_size` + +static void* mi_block_ptr_set_guarded(mi_block_t* block, size_t obj_size) { + // TODO: we can still make padding work by moving it out of the guard page area + mi_page_t* const page = _mi_ptr_page(block); + mi_page_set_has_aligned(page, true); + block->next = MI_BLOCK_TAG_GUARDED; + + // set guard page at the end of the block + mi_segment_t* const segment = _mi_page_segment(page); + const size_t block_size = mi_page_block_size(page); // must use `block_size` to match `mi_free_local` + const size_t os_page_size = _mi_os_page_size(); + mi_assert_internal(block_size >= obj_size + os_page_size + sizeof(mi_block_t)); + if (block_size < obj_size + os_page_size + sizeof(mi_block_t)) { + // should never happen + mi_free(block); + return NULL; + } + uint8_t* guard_page = (uint8_t*)block + block_size - os_page_size; + mi_assert_internal(_mi_is_aligned(guard_page, os_page_size)); + if (segment->allow_decommit && _mi_is_aligned(guard_page, os_page_size)) { + _mi_os_protect(guard_page, os_page_size); + } + else { + _mi_warning_message("unable to set a guard page behind an object due to pinned memory (large OS pages?) (object %p of size %zu)\n", block, block_size); + } + + // align pointer just in front of the guard page + size_t offset = block_size - os_page_size - obj_size; + mi_assert_internal(offset > sizeof(mi_block_t)); + if (offset > MI_BLOCK_ALIGNMENT_MAX) { + // give up to place it right in front of the guard page if the offset is too large for unalignment + offset = MI_BLOCK_ALIGNMENT_MAX; + } + void* p = (uint8_t*)block + offset; + mi_track_align(block, p, offset, obj_size); + mi_track_mem_defined(block, sizeof(mi_block_t)); + return p; +} + +mi_decl_restrict void* _mi_heap_malloc_guarded(mi_heap_t* heap, size_t size, bool zero) mi_attr_noexcept +{ + #if defined(MI_PADDING_SIZE) + mi_assert(MI_PADDING_SIZE==0); + #endif + // allocate multiple of page size ending in a guard page + // ensure minimal alignment requirement? + const size_t os_page_size = _mi_os_page_size(); + const size_t obj_size = (mi_option_is_enabled(mi_option_guarded_precise) ? size : _mi_align_up(size, MI_MAX_ALIGN_SIZE)); + const size_t bsize = _mi_align_up(_mi_align_up(obj_size, MI_MAX_ALIGN_SIZE) + sizeof(mi_block_t), MI_MAX_ALIGN_SIZE); + const size_t req_size = _mi_align_up(bsize + os_page_size, os_page_size); + mi_block_t* const block = (mi_block_t*)_mi_malloc_generic(heap, req_size, zero, 0 /* huge_alignment */); + if (block==NULL) return NULL; + void* const p = mi_block_ptr_set_guarded(block, obj_size); + + // stats + mi_track_malloc(p, size, zero); + if (p != NULL) { + if (!mi_heap_is_initialized(heap)) { heap = mi_prim_get_default_heap(); } + #if MI_STAT>1 + mi_heap_stat_adjust_decrease(heap, malloc_requested, req_size); + mi_heap_stat_increase(heap, malloc_requested, size); + #endif + _mi_stat_counter_increase(&heap->tld->stats.malloc_guarded_count, 1); + } + #if MI_DEBUG>3 + if (p != NULL && zero) { + mi_assert_expensive(mi_mem_is_zero(p, size)); + } + #endif + return p; +} +#endif + +// ------------------------------------------------------ +// ensure explicit external inline definitions are emitted! +// ------------------------------------------------------ + +#ifdef __cplusplus +void* _mi_externs[] = { + (void*)&_mi_page_malloc, + (void*)&_mi_page_malloc_zero, + (void*)&_mi_heap_malloc_zero, + (void*)&_mi_heap_malloc_zero_ex, + (void*)&mi_malloc, + (void*)&mi_malloc_small, + (void*)&mi_zalloc_small, + (void*)&mi_heap_malloc, + (void*)&mi_heap_zalloc, + (void*)&mi_heap_malloc_small, + // (void*)&mi_heap_alloc_new, + // (void*)&mi_heap_alloc_new_n +}; +#endif diff --git a/compat/mimalloc/arena-abandon.c b/compat/mimalloc/arena-abandon.c new file mode 100644 index 00000000000000..460c80fc22782f --- /dev/null +++ b/compat/mimalloc/arena-abandon.c @@ -0,0 +1,346 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2019-2024, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ + +#if !defined(MI_IN_ARENA_C) +#error "this file should be included from 'arena.c' (so mi_arena_t is visible)" +// add includes help an IDE +#include "mimalloc.h" +#include "mimalloc/internal.h" +#include "bitmap.h" +#endif + +// Minimal exports for arena-abandoned. +size_t mi_arena_id_index(mi_arena_id_t id); +mi_arena_t* mi_arena_from_index(size_t idx); +size_t mi_arena_get_count(void); +void* mi_arena_block_start(mi_arena_t* arena, mi_bitmap_index_t bindex); +bool mi_arena_memid_indices(mi_memid_t memid, size_t* arena_index, mi_bitmap_index_t* bitmap_index); + +/* ----------------------------------------------------------- + Abandoned blocks/segments: + + _mi_arena_segment_clear_abandoned + _mi_arena_segment_mark_abandoned + + This is used to atomically abandon/reclaim segments + (and crosses the arena API but it is convenient to have here). + + Abandoned segments still have live blocks; they get reclaimed + when a thread frees a block in it, or when a thread needs a fresh + segment. + + Abandoned segments are atomically marked in the `block_abandoned` + bitmap of arenas. Any segments allocated outside arenas are put + in the sub-process `abandoned_os_list`. This list is accessed + using locks but this should be uncommon and generally uncontended. + Reclaim and visiting either scan through the `block_abandoned` + bitmaps of the arena's, or visit the `abandoned_os_list` + + A potentially nicer design is to use arena's for everything + and perhaps have virtual arena's to map OS allocated memory + but this would lack the "density" of our current arena's. TBC. +----------------------------------------------------------- */ + + +// reclaim a specific OS abandoned segment; `true` on success. +// sets the thread_id. +static bool mi_arena_segment_os_clear_abandoned(mi_segment_t* segment, bool take_lock) { + mi_assert(segment->memid.memkind != MI_MEM_ARENA); + // not in an arena, remove from list of abandoned os segments + mi_subproc_t* const subproc = segment->subproc; + if (take_lock && !mi_lock_try_acquire(&subproc->abandoned_os_lock)) { + return false; // failed to acquire the lock, we just give up + } + // remove atomically from the abandoned os list (if possible!) + bool reclaimed = false; + mi_segment_t* const next = segment->abandoned_os_next; + mi_segment_t* const prev = segment->abandoned_os_prev; + if (next != NULL || prev != NULL || subproc->abandoned_os_list == segment) { + #if MI_DEBUG>3 + // find ourselves in the abandoned list (and check the count) + bool found = false; + size_t count = 0; + for (mi_segment_t* current = subproc->abandoned_os_list; current != NULL; current = current->abandoned_os_next) { + if (current == segment) { found = true; } + count++; + } + mi_assert_internal(found); + mi_assert_internal(count == mi_atomic_load_relaxed(&subproc->abandoned_os_list_count)); + #endif + // remove (atomically) from the list and reclaim + if (prev != NULL) { prev->abandoned_os_next = next; } + else { subproc->abandoned_os_list = next; } + if (next != NULL) { next->abandoned_os_prev = prev; } + else { subproc->abandoned_os_list_tail = prev; } + segment->abandoned_os_next = NULL; + segment->abandoned_os_prev = NULL; + mi_atomic_decrement_relaxed(&subproc->abandoned_count); + mi_atomic_decrement_relaxed(&subproc->abandoned_os_list_count); + if (take_lock) { // don't reset the thread_id when iterating + mi_atomic_store_release(&segment->thread_id, _mi_thread_id()); + } + reclaimed = true; + } + if (take_lock) { mi_lock_release(&segment->subproc->abandoned_os_lock); } + return reclaimed; +} + +// reclaim a specific abandoned segment; `true` on success. +// sets the thread_id. +bool _mi_arena_segment_clear_abandoned(mi_segment_t* segment) { + if mi_unlikely(segment->memid.memkind != MI_MEM_ARENA) { + return mi_arena_segment_os_clear_abandoned(segment, true /* take lock */); + } + // arena segment: use the blocks_abandoned bitmap. + size_t arena_idx; + size_t bitmap_idx; + mi_arena_memid_indices(segment->memid, &arena_idx, &bitmap_idx); + mi_arena_t* arena = mi_arena_from_index(arena_idx); + mi_assert_internal(arena != NULL); + // reclaim atomically + bool was_marked = _mi_bitmap_unclaim(arena->blocks_abandoned, arena->field_count, 1, bitmap_idx); + if (was_marked) { + mi_assert_internal(mi_atomic_load_acquire(&segment->thread_id) == 0); + mi_atomic_decrement_relaxed(&segment->subproc->abandoned_count); + mi_atomic_store_release(&segment->thread_id, _mi_thread_id()); + } + // mi_assert_internal(was_marked); + mi_assert_internal(!was_marked || _mi_bitmap_is_claimed(arena->blocks_inuse, arena->field_count, 1, bitmap_idx)); + //mi_assert_internal(arena->blocks_committed == NULL || _mi_bitmap_is_claimed(arena->blocks_committed, arena->field_count, 1, bitmap_idx)); + return was_marked; +} + + +// mark a specific OS segment as abandoned +static void mi_arena_segment_os_mark_abandoned(mi_segment_t* segment) { + mi_assert(segment->memid.memkind != MI_MEM_ARENA); + // not in an arena; we use a list of abandoned segments + mi_subproc_t* const subproc = segment->subproc; + mi_lock(&subproc->abandoned_os_lock) { + // push on the tail of the list (important for the visitor) + mi_segment_t* prev = subproc->abandoned_os_list_tail; + mi_assert_internal(prev == NULL || prev->abandoned_os_next == NULL); + mi_assert_internal(segment->abandoned_os_prev == NULL); + mi_assert_internal(segment->abandoned_os_next == NULL); + if (prev != NULL) { prev->abandoned_os_next = segment; } + else { subproc->abandoned_os_list = segment; } + subproc->abandoned_os_list_tail = segment; + segment->abandoned_os_prev = prev; + segment->abandoned_os_next = NULL; + mi_atomic_increment_relaxed(&subproc->abandoned_os_list_count); + mi_atomic_increment_relaxed(&subproc->abandoned_count); + // and release the lock + } + return; +} + +// mark a specific segment as abandoned +// clears the thread_id. +void _mi_arena_segment_mark_abandoned(mi_segment_t* segment) +{ + mi_assert_internal(segment->used == segment->abandoned); + mi_atomic_store_release(&segment->thread_id, (uintptr_t)0); // mark as abandoned for multi-thread free's + if mi_unlikely(segment->memid.memkind != MI_MEM_ARENA) { + mi_arena_segment_os_mark_abandoned(segment); + return; + } + // segment is in an arena, mark it in the arena `blocks_abandoned` bitmap + size_t arena_idx; + size_t bitmap_idx; + mi_arena_memid_indices(segment->memid, &arena_idx, &bitmap_idx); + mi_arena_t* arena = mi_arena_from_index(arena_idx); + mi_assert_internal(arena != NULL); + // set abandonment atomically + mi_subproc_t* const subproc = segment->subproc; // don't access the segment after setting it abandoned + const bool was_unmarked = _mi_bitmap_claim(arena->blocks_abandoned, arena->field_count, 1, bitmap_idx, NULL); + if (was_unmarked) { mi_atomic_increment_relaxed(&subproc->abandoned_count); } + mi_assert_internal(was_unmarked); + mi_assert_internal(_mi_bitmap_is_claimed(arena->blocks_inuse, arena->field_count, 1, bitmap_idx)); +} + + +/* ----------------------------------------------------------- + Iterate through the abandoned blocks/segments using a cursor. + This is used for reclaiming and abandoned block visiting. +----------------------------------------------------------- */ + +// start a cursor at a randomized arena +void _mi_arena_field_cursor_init(mi_heap_t* heap, mi_subproc_t* subproc, bool visit_all, mi_arena_field_cursor_t* current) { + mi_assert_internal(heap == NULL || heap->tld->segments.subproc == subproc); + current->bitmap_idx = 0; + current->subproc = subproc; + current->visit_all = visit_all; + current->hold_visit_lock = false; + const size_t abandoned_count = mi_atomic_load_relaxed(&subproc->abandoned_count); + const size_t abandoned_list_count = mi_atomic_load_relaxed(&subproc->abandoned_os_list_count); + const size_t max_arena = mi_arena_get_count(); + if (heap != NULL && heap->arena_id != _mi_arena_id_none()) { + // for a heap that is bound to one arena, only visit that arena + current->start = mi_arena_id_index(heap->arena_id); + current->end = current->start + 1; + current->os_list_count = 0; + } + else { + // otherwise visit all starting at a random location + if (abandoned_count > abandoned_list_count && max_arena > 0) { + current->start = (heap == NULL || max_arena == 0 ? 0 : (mi_arena_id_t)(_mi_heap_random_next(heap) % max_arena)); + current->end = current->start + max_arena; + } + else { + current->start = 0; + current->end = 0; + } + current->os_list_count = abandoned_list_count; // max entries to visit in the os abandoned list + } + mi_assert_internal(current->start <= max_arena); +} + +void _mi_arena_field_cursor_done(mi_arena_field_cursor_t* current) { + if (current->hold_visit_lock) { + mi_lock_release(¤t->subproc->abandoned_os_visit_lock); + current->hold_visit_lock = false; + } +} + +static mi_segment_t* mi_arena_segment_clear_abandoned_at(mi_arena_t* arena, mi_subproc_t* subproc, mi_bitmap_index_t bitmap_idx) { + // try to reclaim an abandoned segment in the arena atomically + if (!_mi_bitmap_unclaim(arena->blocks_abandoned, arena->field_count, 1, bitmap_idx)) return NULL; + mi_assert_internal(_mi_bitmap_is_claimed(arena->blocks_inuse, arena->field_count, 1, bitmap_idx)); + mi_segment_t* segment = (mi_segment_t*)mi_arena_block_start(arena, bitmap_idx); + mi_assert_internal(mi_atomic_load_relaxed(&segment->thread_id) == 0); + // check that the segment belongs to our sub-process + // note: this is the reason we need the `abandoned_visit` lock in the case abandoned visiting is enabled. + // without the lock an abandoned visit may otherwise fail to visit all abandoned segments in the sub-process. + // for regular reclaim it is fine to miss one sometimes so without abandoned visiting we don't need the `abandoned_visit` lock. + if (segment->subproc != subproc) { + // it is from another sub-process, re-mark it and continue searching + const bool was_zero = _mi_bitmap_claim(arena->blocks_abandoned, arena->field_count, 1, bitmap_idx, NULL); + mi_assert_internal(was_zero); MI_UNUSED(was_zero); + return NULL; + } + else { + // success, we unabandoned a segment in our sub-process + mi_atomic_decrement_relaxed(&subproc->abandoned_count); + return segment; + } +} + +static mi_segment_t* mi_arena_segment_clear_abandoned_next_field(mi_arena_field_cursor_t* previous) { + const size_t max_arena = mi_arena_get_count(); + size_t field_idx = mi_bitmap_index_field(previous->bitmap_idx); + size_t bit_idx = mi_bitmap_index_bit_in_field(previous->bitmap_idx); + // visit arena's (from the previous cursor) + for (; previous->start < previous->end; previous->start++, field_idx = 0, bit_idx = 0) { + // index wraps around + size_t arena_idx = (previous->start >= max_arena ? previous->start % max_arena : previous->start); + mi_arena_t* arena = mi_arena_from_index(arena_idx); + if (arena != NULL) { + bool has_lock = false; + // visit the abandoned fields (starting at previous_idx) + for (; field_idx < arena->field_count; field_idx++, bit_idx = 0) { + size_t field = mi_atomic_load_relaxed(&arena->blocks_abandoned[field_idx]); + if mi_unlikely(field != 0) { // skip zero fields quickly + // we only take the arena lock if there are actually abandoned segments present + if (!has_lock && mi_option_is_enabled(mi_option_visit_abandoned)) { + has_lock = (previous->visit_all ? (mi_lock_acquire(&arena->abandoned_visit_lock),true) : mi_lock_try_acquire(&arena->abandoned_visit_lock)); + if (!has_lock) { + if (previous->visit_all) { + _mi_error_message(EFAULT, "internal error: failed to visit all abandoned segments due to failure to acquire the visitor lock"); + } + // skip to next arena + break; + } + } + mi_assert_internal(has_lock || !mi_option_is_enabled(mi_option_visit_abandoned)); + // visit each set bit in the field (todo: maybe use `ctz` here?) + for (; bit_idx < MI_BITMAP_FIELD_BITS; bit_idx++) { + // pre-check if the bit is set + size_t mask = ((size_t)1 << bit_idx); + if mi_unlikely((field & mask) == mask) { + mi_bitmap_index_t bitmap_idx = mi_bitmap_index_create(field_idx, bit_idx); + mi_segment_t* const segment = mi_arena_segment_clear_abandoned_at(arena, previous->subproc, bitmap_idx); + if (segment != NULL) { + //mi_assert_internal(arena->blocks_committed == NULL || _mi_bitmap_is_claimed(arena->blocks_committed, arena->field_count, 1, bitmap_idx)); + if (has_lock) { mi_lock_release(&arena->abandoned_visit_lock); } + previous->bitmap_idx = mi_bitmap_index_create_ex(field_idx, bit_idx + 1); // start at next one for the next iteration + return segment; + } + } + } + } + } + if (has_lock) { mi_lock_release(&arena->abandoned_visit_lock); } + } + } + return NULL; +} + +static mi_segment_t* mi_arena_segment_clear_abandoned_next_list(mi_arena_field_cursor_t* previous) { + // go through the abandoned_os_list + // we only allow one thread per sub-process to do to visit guarded by the `abandoned_os_visit_lock`. + // The lock is released when the cursor is released. + if (!previous->hold_visit_lock) { + previous->hold_visit_lock = (previous->visit_all ? (mi_lock_acquire(&previous->subproc->abandoned_os_visit_lock),true) + : mi_lock_try_acquire(&previous->subproc->abandoned_os_visit_lock)); + if (!previous->hold_visit_lock) { + if (previous->visit_all) { + _mi_error_message(EFAULT, "internal error: failed to visit all abandoned segments due to failure to acquire the OS visitor lock"); + } + return NULL; // we cannot get the lock, give up + } + } + // One list entry at a time + while (previous->os_list_count > 0) { + previous->os_list_count--; + mi_lock_acquire(&previous->subproc->abandoned_os_lock); // this could contend with concurrent OS block abandonment and reclaim from `free` + mi_segment_t* segment = previous->subproc->abandoned_os_list; + // pop from head of the list, a subsequent mark will push at the end (and thus we iterate through os_list_count entries) + if (segment == NULL || mi_arena_segment_os_clear_abandoned(segment, false /* we already have the lock */)) { + mi_lock_release(&previous->subproc->abandoned_os_lock); + return segment; + } + // already abandoned, try again + mi_lock_release(&previous->subproc->abandoned_os_lock); + } + // done + mi_assert_internal(previous->os_list_count == 0); + return NULL; +} + + +// reclaim abandoned segments +// this does not set the thread id (so it appears as still abandoned) +mi_segment_t* _mi_arena_segment_clear_abandoned_next(mi_arena_field_cursor_t* previous) { + if (previous->start < previous->end) { + // walk the arena + mi_segment_t* segment = mi_arena_segment_clear_abandoned_next_field(previous); + if (segment != NULL) { return segment; } + } + // no entries in the arena's anymore, walk the abandoned OS list + mi_assert_internal(previous->start == previous->end); + return mi_arena_segment_clear_abandoned_next_list(previous); +} + + +bool mi_abandoned_visit_blocks(mi_subproc_id_t subproc_id, int heap_tag, bool visit_blocks, mi_block_visit_fun* visitor, void* arg) { + // (unfortunately) the visit_abandoned option must be enabled from the start. + // This is to avoid taking locks if abandoned list visiting is not required (as for most programs) + if (!mi_option_is_enabled(mi_option_visit_abandoned)) { + _mi_error_message(EFAULT, "internal error: can only visit abandoned blocks when MIMALLOC_VISIT_ABANDONED=ON"); + return false; + } + mi_arena_field_cursor_t current; + _mi_arena_field_cursor_init(NULL, _mi_subproc_from_id(subproc_id), true /* visit all (blocking) */, ¤t); + mi_segment_t* segment; + bool ok = true; + while (ok && (segment = _mi_arena_segment_clear_abandoned_next(¤t)) != NULL) { + ok = _mi_segment_visit_blocks(segment, heap_tag, visit_blocks, visitor, arg); + _mi_arena_segment_mark_abandoned(segment); + } + _mi_arena_field_cursor_done(¤t); + return ok; +} diff --git a/compat/mimalloc/arena.c b/compat/mimalloc/arena.c new file mode 100644 index 00000000000000..01102c275d3ad1 --- /dev/null +++ b/compat/mimalloc/arena.c @@ -0,0 +1,1044 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2019-2024, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ + +/* ---------------------------------------------------------------------------- +"Arenas" are fixed area's of OS memory from which we can allocate +large blocks (>= MI_ARENA_MIN_BLOCK_SIZE, 4MiB). +In contrast to the rest of mimalloc, the arenas are shared between +threads and need to be accessed using atomic operations. + +Arenas are also used to for huge OS page (1GiB) reservations or for reserving +OS memory upfront which can be improve performance or is sometimes needed +on embedded devices. We can also employ this with WASI or `sbrk` systems +to reserve large arenas upfront and be able to reuse the memory more effectively. + +The arena allocation needs to be thread safe and we use an atomic bitmap to allocate. +-----------------------------------------------------------------------------*/ + +#include "mimalloc.h" +#include "mimalloc/internal.h" +#include "mimalloc/atomic.h" +#include "bitmap.h" + + +/* ----------------------------------------------------------- + Arena allocation +----------------------------------------------------------- */ + +// A memory arena descriptor +typedef struct mi_arena_s { + mi_arena_id_t id; // arena id; 0 for non-specific + mi_memid_t memid; // memid of the memory area + _Atomic(uint8_t*) start; // the start of the memory area + size_t block_count; // size of the area in arena blocks (of `MI_ARENA_BLOCK_SIZE`) + size_t field_count; // number of bitmap fields (where `field_count * MI_BITMAP_FIELD_BITS >= block_count`) + size_t meta_size; // size of the arena structure itself (including its bitmaps) + mi_memid_t meta_memid; // memid of the arena structure itself (OS or static allocation) + int numa_node; // associated NUMA node + bool exclusive; // only allow allocations if specifically for this arena + bool is_large; // memory area consists of large- or huge OS pages (always committed) + mi_lock_t abandoned_visit_lock; // lock is only used when abandoned segments are being visited + _Atomic(size_t) search_idx; // optimization to start the search for free blocks + _Atomic(mi_msecs_t) purge_expire; // expiration time when blocks should be purged from `blocks_purge`. + + mi_bitmap_field_t* blocks_dirty; // are the blocks potentially non-zero? + mi_bitmap_field_t* blocks_committed; // are the blocks committed? (can be NULL for memory that cannot be decommitted) + mi_bitmap_field_t* blocks_purge; // blocks that can be (reset) decommitted. (can be NULL for memory that cannot be (reset) decommitted) + mi_bitmap_field_t* blocks_abandoned; // blocks that start with an abandoned segment. (This crosses API's but it is convenient to have here) + mi_bitmap_field_t blocks_inuse[1]; // in-place bitmap of in-use blocks (of size `field_count`) + // do not add further fields here as the dirty, committed, purged, and abandoned bitmaps follow the inuse bitmap fields. +} mi_arena_t; + + +#define MI_ARENA_BLOCK_SIZE (MI_SEGMENT_SIZE) // 64MiB (must be at least MI_SEGMENT_ALIGN) +#define MI_ARENA_MIN_OBJ_SIZE (MI_ARENA_BLOCK_SIZE/2) // 32MiB +#define MI_MAX_ARENAS (132) // Limited as the reservation exponentially increases (and takes up .bss) + +// The available arenas +static mi_decl_cache_align _Atomic(mi_arena_t*) mi_arenas[MI_MAX_ARENAS]; +static mi_decl_cache_align _Atomic(size_t) mi_arena_count; // = 0 +static mi_decl_cache_align _Atomic(int64_t) mi_arenas_purge_expire; // set if there exist purgeable arenas + +#define MI_IN_ARENA_C +#include "arena-abandon.c" +#undef MI_IN_ARENA_C + +/* ----------------------------------------------------------- + Arena id's + id = arena_index + 1 +----------------------------------------------------------- */ + +size_t mi_arena_id_index(mi_arena_id_t id) { + return (size_t)(id <= 0 ? MI_MAX_ARENAS : id - 1); +} + +static mi_arena_id_t mi_arena_id_create(size_t arena_index) { + mi_assert_internal(arena_index < MI_MAX_ARENAS); + return (int)arena_index + 1; +} + +mi_arena_id_t _mi_arena_id_none(void) { + return 0; +} + +static bool mi_arena_id_is_suitable(mi_arena_id_t arena_id, bool arena_is_exclusive, mi_arena_id_t req_arena_id) { + return ((!arena_is_exclusive && req_arena_id == _mi_arena_id_none()) || + (arena_id == req_arena_id)); +} + +bool _mi_arena_memid_is_suitable(mi_memid_t memid, mi_arena_id_t request_arena_id) { + if (memid.memkind == MI_MEM_ARENA) { + return mi_arena_id_is_suitable(memid.mem.arena.id, memid.mem.arena.is_exclusive, request_arena_id); + } + else { + return mi_arena_id_is_suitable(_mi_arena_id_none(), false, request_arena_id); + } +} + +bool _mi_arena_memid_is_os_allocated(mi_memid_t memid) { + return (memid.memkind == MI_MEM_OS); +} + +size_t mi_arena_get_count(void) { + return mi_atomic_load_relaxed(&mi_arena_count); +} + +mi_arena_t* mi_arena_from_index(size_t idx) { + mi_assert_internal(idx < mi_arena_get_count()); + return mi_atomic_load_ptr_acquire(mi_arena_t, &mi_arenas[idx]); +} + + +/* ----------------------------------------------------------- + Arena allocations get a (currently) 16-bit memory id where the + lower 8 bits are the arena id, and the upper bits the block index. +----------------------------------------------------------- */ + +static size_t mi_block_count_of_size(size_t size) { + return _mi_divide_up(size, MI_ARENA_BLOCK_SIZE); +} + +static size_t mi_arena_block_size(size_t bcount) { + return (bcount * MI_ARENA_BLOCK_SIZE); +} + +static size_t mi_arena_size(mi_arena_t* arena) { + return mi_arena_block_size(arena->block_count); +} + +static mi_memid_t mi_memid_create_arena(mi_arena_id_t id, bool is_exclusive, mi_bitmap_index_t bitmap_index) { + mi_memid_t memid = _mi_memid_create(MI_MEM_ARENA); + memid.mem.arena.id = id; + memid.mem.arena.block_index = bitmap_index; + memid.mem.arena.is_exclusive = is_exclusive; + return memid; +} + +bool mi_arena_memid_indices(mi_memid_t memid, size_t* arena_index, mi_bitmap_index_t* bitmap_index) { + mi_assert_internal(memid.memkind == MI_MEM_ARENA); + *arena_index = mi_arena_id_index(memid.mem.arena.id); + *bitmap_index = memid.mem.arena.block_index; + return memid.mem.arena.is_exclusive; +} + + + +/* ----------------------------------------------------------- + Special static area for mimalloc internal structures + to avoid OS calls (for example, for the arena metadata (~= 256b)) +----------------------------------------------------------- */ + +#define MI_ARENA_STATIC_MAX ((MI_INTPTR_SIZE/2)*MI_KiB) // 4 KiB on 64-bit + +static mi_decl_cache_align uint8_t mi_arena_static[MI_ARENA_STATIC_MAX]; // must be cache aligned, see issue #895 +static mi_decl_cache_align _Atomic(size_t) mi_arena_static_top; + +static void* mi_arena_static_zalloc(size_t size, size_t alignment, mi_memid_t* memid) { + *memid = _mi_memid_none(); + if (size == 0 || size > MI_ARENA_STATIC_MAX) return NULL; + const size_t toplow = mi_atomic_load_relaxed(&mi_arena_static_top); + if ((toplow + size) > MI_ARENA_STATIC_MAX) return NULL; + + // try to claim space + if (alignment < MI_MAX_ALIGN_SIZE) { alignment = MI_MAX_ALIGN_SIZE; } + const size_t oversize = size + alignment - 1; + if (toplow + oversize > MI_ARENA_STATIC_MAX) return NULL; + const size_t oldtop = mi_atomic_add_acq_rel(&mi_arena_static_top, oversize); + size_t top = oldtop + oversize; + if (top > MI_ARENA_STATIC_MAX) { + // try to roll back, ok if this fails + mi_atomic_cas_strong_acq_rel(&mi_arena_static_top, &top, oldtop); + return NULL; + } + + // success + *memid = _mi_memid_create(MI_MEM_STATIC); + memid->initially_zero = true; + const size_t start = _mi_align_up(oldtop, alignment); + uint8_t* const p = &mi_arena_static[start]; + _mi_memzero_aligned(p, size); + return p; +} + +void* _mi_arena_meta_zalloc(size_t size, mi_memid_t* memid) { + *memid = _mi_memid_none(); + + // try static + void* p = mi_arena_static_zalloc(size, MI_MAX_ALIGN_SIZE, memid); + if (p != NULL) return p; + + // or fall back to the OS + p = _mi_os_alloc(size, memid); + if (p == NULL) return NULL; + + // zero the OS memory if needed + if (!memid->initially_zero) { + _mi_memzero_aligned(p, size); + memid->initially_zero = true; + } + return p; +} + +void _mi_arena_meta_free(void* p, mi_memid_t memid, size_t size) { + if (mi_memkind_is_os(memid.memkind)) { + _mi_os_free(p, size, memid); + } + else { + mi_assert(memid.memkind == MI_MEM_STATIC); + } +} + +void* mi_arena_block_start(mi_arena_t* arena, mi_bitmap_index_t bindex) { + return (arena->start + mi_arena_block_size(mi_bitmap_index_bit(bindex))); +} + + +/* ----------------------------------------------------------- + Thread safe allocation in an arena +----------------------------------------------------------- */ + +// claim the `blocks_inuse` bits +static bool mi_arena_try_claim(mi_arena_t* arena, size_t blocks, mi_bitmap_index_t* bitmap_idx) +{ + size_t idx = 0; // mi_atomic_load_relaxed(&arena->search_idx); // start from last search; ok to be relaxed as the exact start does not matter + if (_mi_bitmap_try_find_from_claim_across(arena->blocks_inuse, arena->field_count, idx, blocks, bitmap_idx)) { + mi_atomic_store_relaxed(&arena->search_idx, mi_bitmap_index_field(*bitmap_idx)); // start search from found location next time around + return true; + }; + return false; +} + + +/* ----------------------------------------------------------- + Arena Allocation +----------------------------------------------------------- */ + +static mi_decl_noinline void* mi_arena_try_alloc_at(mi_arena_t* arena, size_t arena_index, size_t needed_bcount, + bool commit, mi_memid_t* memid) +{ + MI_UNUSED(arena_index); + mi_assert_internal(mi_arena_id_index(arena->id) == arena_index); + + mi_bitmap_index_t bitmap_index; + if (!mi_arena_try_claim(arena, needed_bcount, &bitmap_index)) return NULL; + + // claimed it! + void* p = mi_arena_block_start(arena, bitmap_index); + *memid = mi_memid_create_arena(arena->id, arena->exclusive, bitmap_index); + memid->is_pinned = arena->memid.is_pinned; + + // none of the claimed blocks should be scheduled for a decommit + if (arena->blocks_purge != NULL) { + // this is thread safe as a potential purge only decommits parts that are not yet claimed as used (in `blocks_inuse`). + _mi_bitmap_unclaim_across(arena->blocks_purge, arena->field_count, needed_bcount, bitmap_index); + } + + // set the dirty bits (todo: no need for an atomic op here?) + if (arena->memid.initially_zero && arena->blocks_dirty != NULL) { + memid->initially_zero = _mi_bitmap_claim_across(arena->blocks_dirty, arena->field_count, needed_bcount, bitmap_index, NULL, NULL); + } + + // set commit state + if (arena->blocks_committed == NULL) { + // always committed + memid->initially_committed = true; + } + else if (commit) { + // commit requested, but the range may not be committed as a whole: ensure it is committed now + memid->initially_committed = true; + bool any_uncommitted; + size_t already_committed = 0; + _mi_bitmap_claim_across(arena->blocks_committed, arena->field_count, needed_bcount, bitmap_index, &any_uncommitted, &already_committed); + if (any_uncommitted) { + mi_assert_internal(already_committed < needed_bcount); + const size_t commit_size = mi_arena_block_size(needed_bcount); + const size_t stat_commit_size = commit_size - mi_arena_block_size(already_committed); + bool commit_zero = false; + if (!_mi_os_commit_ex(p, commit_size, &commit_zero, stat_commit_size)) { + memid->initially_committed = false; + } + else { + if (commit_zero) { memid->initially_zero = true; } + } + } + } + else { + // no need to commit, but check if already fully committed + size_t already_committed = 0; + memid->initially_committed = _mi_bitmap_is_claimed_across(arena->blocks_committed, arena->field_count, needed_bcount, bitmap_index, &already_committed); + if (!memid->initially_committed && already_committed > 0) { + // partially committed: as it will be committed at some time, adjust the stats and pretend the range is fully uncommitted. + mi_assert_internal(already_committed < needed_bcount); + _mi_stat_decrease(&_mi_stats_main.committed, mi_arena_block_size(already_committed)); + _mi_bitmap_unclaim_across(arena->blocks_committed, arena->field_count, needed_bcount, bitmap_index); + } + } + + return p; +} + +// allocate in a specific arena +static void* mi_arena_try_alloc_at_id(mi_arena_id_t arena_id, bool match_numa_node, int numa_node, size_t size, size_t alignment, + bool commit, bool allow_large, mi_arena_id_t req_arena_id, mi_memid_t* memid ) +{ + MI_UNUSED_RELEASE(alignment); + mi_assert(alignment <= MI_SEGMENT_ALIGN); + const size_t bcount = mi_block_count_of_size(size); + const size_t arena_index = mi_arena_id_index(arena_id); + mi_assert_internal(arena_index < mi_atomic_load_relaxed(&mi_arena_count)); + mi_assert_internal(size <= mi_arena_block_size(bcount)); + + // Check arena suitability + mi_arena_t* arena = mi_arena_from_index(arena_index); + if (arena == NULL) return NULL; + if (!allow_large && arena->is_large) return NULL; + if (!mi_arena_id_is_suitable(arena->id, arena->exclusive, req_arena_id)) return NULL; + if (req_arena_id == _mi_arena_id_none()) { // in not specific, check numa affinity + const bool numa_suitable = (numa_node < 0 || arena->numa_node < 0 || arena->numa_node == numa_node); + if (match_numa_node) { if (!numa_suitable) return NULL; } + else { if (numa_suitable) return NULL; } + } + + // try to allocate + void* p = mi_arena_try_alloc_at(arena, arena_index, bcount, commit, memid); + mi_assert_internal(p == NULL || _mi_is_aligned(p, alignment)); + return p; +} + + +// allocate from an arena with fallback to the OS +static mi_decl_noinline void* mi_arena_try_alloc(int numa_node, size_t size, size_t alignment, + bool commit, bool allow_large, + mi_arena_id_t req_arena_id, mi_memid_t* memid ) +{ + MI_UNUSED(alignment); + mi_assert_internal(alignment <= MI_SEGMENT_ALIGN); + const size_t max_arena = mi_atomic_load_relaxed(&mi_arena_count); + if mi_likely(max_arena == 0) return NULL; + + if (req_arena_id != _mi_arena_id_none()) { + // try a specific arena if requested + if (mi_arena_id_index(req_arena_id) < max_arena) { + void* p = mi_arena_try_alloc_at_id(req_arena_id, true, numa_node, size, alignment, commit, allow_large, req_arena_id, memid); + if (p != NULL) return p; + } + } + else { + // try numa affine allocation + for (size_t i = 0; i < max_arena; i++) { + void* p = mi_arena_try_alloc_at_id(mi_arena_id_create(i), true, numa_node, size, alignment, commit, allow_large, req_arena_id, memid); + if (p != NULL) return p; + } + + // try from another numa node instead.. + if (numa_node >= 0) { // if numa_node was < 0 (no specific affinity requested), all arena's have been tried already + for (size_t i = 0; i < max_arena; i++) { + void* p = mi_arena_try_alloc_at_id(mi_arena_id_create(i), false /* only proceed if not numa local */, numa_node, size, alignment, commit, allow_large, req_arena_id, memid); + if (p != NULL) return p; + } + } + } + return NULL; +} + +// try to reserve a fresh arena space +static bool mi_arena_reserve(size_t req_size, bool allow_large, mi_arena_id_t *arena_id) +{ + if (_mi_preloading()) return false; // use OS only while pre loading + + const size_t arena_count = mi_atomic_load_acquire(&mi_arena_count); + if (arena_count > (MI_MAX_ARENAS - 4)) return false; + + size_t arena_reserve = mi_option_get_size(mi_option_arena_reserve); + if (arena_reserve == 0) return false; + + if (!_mi_os_has_virtual_reserve()) { + arena_reserve = arena_reserve/4; // be conservative if virtual reserve is not supported (for WASM for example) + } + arena_reserve = _mi_align_up(arena_reserve, MI_ARENA_BLOCK_SIZE); + arena_reserve = _mi_align_up(arena_reserve, MI_SEGMENT_SIZE); + if (arena_count >= 8 && arena_count <= 128) { + // scale up the arena sizes exponentially every 8 entries (128 entries get to 589TiB) + const size_t multiplier = (size_t)1 << _mi_clamp(arena_count/8, 0, 16 ); + size_t reserve = 0; + if (!mi_mul_overflow(multiplier, arena_reserve, &reserve)) { + arena_reserve = reserve; + } + } + if (arena_reserve < req_size) return false; // should be able to at least handle the current allocation size + + // commit eagerly? + bool arena_commit = false; + if (mi_option_get(mi_option_arena_eager_commit) == 2) { arena_commit = _mi_os_has_overcommit(); } + else if (mi_option_get(mi_option_arena_eager_commit) == 1) { arena_commit = true; } + + return (mi_reserve_os_memory_ex(arena_reserve, arena_commit, allow_large, false /* exclusive? */, arena_id) == 0); +} + + +void* _mi_arena_alloc_aligned(size_t size, size_t alignment, size_t align_offset, bool commit, bool allow_large, + mi_arena_id_t req_arena_id, mi_memid_t* memid) +{ + mi_assert_internal(memid != NULL); + mi_assert_internal(size > 0); + *memid = _mi_memid_none(); + + const int numa_node = _mi_os_numa_node(); // current numa node + + // try to allocate in an arena if the alignment is small enough and the object is not too small (as for heap meta data) + if (!mi_option_is_enabled(mi_option_disallow_arena_alloc)) { // is arena allocation allowed? + if (size >= MI_ARENA_MIN_OBJ_SIZE && alignment <= MI_SEGMENT_ALIGN && align_offset == 0) + { + void* p = mi_arena_try_alloc(numa_node, size, alignment, commit, allow_large, req_arena_id, memid); + if (p != NULL) return p; + + // otherwise, try to first eagerly reserve a new arena + if (req_arena_id == _mi_arena_id_none()) { + mi_arena_id_t arena_id = 0; + if (mi_arena_reserve(size, allow_large, &arena_id)) { + // and try allocate in there + mi_assert_internal(req_arena_id == _mi_arena_id_none()); + p = mi_arena_try_alloc_at_id(arena_id, true, numa_node, size, alignment, commit, allow_large, req_arena_id, memid); + if (p != NULL) return p; + } + } + } + } + + // if we cannot use OS allocation, return NULL + if (mi_option_is_enabled(mi_option_disallow_os_alloc) || req_arena_id != _mi_arena_id_none()) { + errno = ENOMEM; + return NULL; + } + + // finally, fall back to the OS + if (align_offset > 0) { + return _mi_os_alloc_aligned_at_offset(size, alignment, align_offset, commit, allow_large, memid); + } + else { + return _mi_os_alloc_aligned(size, alignment, commit, allow_large, memid); + } +} + +void* _mi_arena_alloc(size_t size, bool commit, bool allow_large, mi_arena_id_t req_arena_id, mi_memid_t* memid) +{ + return _mi_arena_alloc_aligned(size, MI_ARENA_BLOCK_SIZE, 0, commit, allow_large, req_arena_id, memid); +} + + +void* mi_arena_area(mi_arena_id_t arena_id, size_t* size) { + if (size != NULL) *size = 0; + size_t arena_index = mi_arena_id_index(arena_id); + if (arena_index >= MI_MAX_ARENAS) return NULL; + mi_arena_t* arena = mi_atomic_load_ptr_acquire(mi_arena_t, &mi_arenas[arena_index]); + if (arena == NULL) return NULL; + if (size != NULL) { *size = mi_arena_block_size(arena->block_count); } + return arena->start; +} + + +/* ----------------------------------------------------------- + Arena purge +----------------------------------------------------------- */ + +static long mi_arena_purge_delay(void) { + // <0 = no purging allowed, 0=immediate purging, >0=milli-second delay + return (mi_option_get(mi_option_purge_delay) * mi_option_get(mi_option_arena_purge_mult)); +} + +// reset or decommit in an arena and update the committed/decommit bitmaps +// assumes we own the area (i.e. blocks_in_use is claimed by us) +static void mi_arena_purge(mi_arena_t* arena, size_t bitmap_idx, size_t blocks) { + mi_assert_internal(arena->blocks_committed != NULL); + mi_assert_internal(arena->blocks_purge != NULL); + mi_assert_internal(!arena->memid.is_pinned); + const size_t size = mi_arena_block_size(blocks); + void* const p = mi_arena_block_start(arena, bitmap_idx); + bool needs_recommit; + size_t already_committed = 0; + if (_mi_bitmap_is_claimed_across(arena->blocks_committed, arena->field_count, blocks, bitmap_idx, &already_committed)) { + // all blocks are committed, we can purge freely + mi_assert_internal(already_committed == blocks); + needs_recommit = _mi_os_purge(p, size); + } + else { + // some blocks are not committed -- this can happen when a partially committed block is freed + // in `_mi_arena_free` and it is conservatively marked as uncommitted but still scheduled for a purge + // we need to ensure we do not try to reset (as that may be invalid for uncommitted memory). + mi_assert_internal(already_committed < blocks); + mi_assert_internal(mi_option_is_enabled(mi_option_purge_decommits)); + needs_recommit = _mi_os_purge_ex(p, size, false /* allow reset? */, mi_arena_block_size(already_committed)); + } + + // clear the purged blocks + _mi_bitmap_unclaim_across(arena->blocks_purge, arena->field_count, blocks, bitmap_idx); + // update committed bitmap + if (needs_recommit) { + _mi_bitmap_unclaim_across(arena->blocks_committed, arena->field_count, blocks, bitmap_idx); + } +} + +// Schedule a purge. This is usually delayed to avoid repeated decommit/commit calls. +// Note: assumes we (still) own the area as we may purge immediately +static void mi_arena_schedule_purge(mi_arena_t* arena, size_t bitmap_idx, size_t blocks) { + mi_assert_internal(arena->blocks_purge != NULL); + const long delay = mi_arena_purge_delay(); + if (delay < 0) return; // is purging allowed at all? + + if (_mi_preloading() || delay == 0) { + // decommit directly + mi_arena_purge(arena, bitmap_idx, blocks); + } + else { + // schedule purge + const mi_msecs_t expire = _mi_clock_now() + delay; + mi_msecs_t expire0 = 0; + if (mi_atomic_casi64_strong_acq_rel(&arena->purge_expire, &expire0, expire)) { + // expiration was not yet set + // maybe set the global arenas expire as well (if it wasn't set already) + mi_atomic_casi64_strong_acq_rel(&mi_arenas_purge_expire, &expire0, expire); + } + else { + // already an expiration was set + } + _mi_bitmap_claim_across(arena->blocks_purge, arena->field_count, blocks, bitmap_idx, NULL, NULL); + } +} + +// purge a range of blocks +// return true if the full range was purged. +// assumes we own the area (i.e. blocks_in_use is claimed by us) +static bool mi_arena_purge_range(mi_arena_t* arena, size_t idx, size_t startidx, size_t bitlen, size_t purge) { + const size_t endidx = startidx + bitlen; + size_t bitidx = startidx; + bool all_purged = false; + while (bitidx < endidx) { + // count consecutive ones in the purge mask + size_t count = 0; + while (bitidx + count < endidx && (purge & ((size_t)1 << (bitidx + count))) != 0) { + count++; + } + if (count > 0) { + // found range to be purged + const mi_bitmap_index_t range_idx = mi_bitmap_index_create(idx, bitidx); + mi_arena_purge(arena, range_idx, count); + if (count == bitlen) { + all_purged = true; + } + } + bitidx += (count+1); // +1 to skip the zero bit (or end) + } + return all_purged; +} + +// returns true if anything was purged +static bool mi_arena_try_purge(mi_arena_t* arena, mi_msecs_t now, bool force) +{ + // check pre-conditions + if (arena->memid.is_pinned) return false; + + // expired yet? + mi_msecs_t expire = mi_atomic_loadi64_relaxed(&arena->purge_expire); + if (!force && (expire == 0 || expire > now)) return false; + + // reset expire (if not already set concurrently) + mi_atomic_casi64_strong_acq_rel(&arena->purge_expire, &expire, (mi_msecs_t)0); + _mi_stat_counter_increase(&_mi_stats_main.arena_purges, 1); + + // potential purges scheduled, walk through the bitmap + bool any_purged = false; + bool full_purge = true; + for (size_t i = 0; i < arena->field_count; i++) { + size_t purge = mi_atomic_load_relaxed(&arena->blocks_purge[i]); + if (purge != 0) { + size_t bitidx = 0; + while (bitidx < MI_BITMAP_FIELD_BITS) { + // find consecutive range of ones in the purge mask + size_t bitlen = 0; + while (bitidx + bitlen < MI_BITMAP_FIELD_BITS && (purge & ((size_t)1 << (bitidx + bitlen))) != 0) { + bitlen++; + } + // temporarily claim the purge range as "in-use" to be thread-safe with allocation + // try to claim the longest range of corresponding in_use bits + const mi_bitmap_index_t bitmap_index = mi_bitmap_index_create(i, bitidx); + while( bitlen > 0 ) { + if (_mi_bitmap_try_claim(arena->blocks_inuse, arena->field_count, bitlen, bitmap_index)) { + break; + } + bitlen--; + } + // actual claimed bits at `in_use` + if (bitlen > 0) { + // read purge again now that we have the in_use bits + purge = mi_atomic_load_acquire(&arena->blocks_purge[i]); + if (!mi_arena_purge_range(arena, i, bitidx, bitlen, purge)) { + full_purge = false; + } + any_purged = true; + // release the claimed `in_use` bits again + _mi_bitmap_unclaim(arena->blocks_inuse, arena->field_count, bitlen, bitmap_index); + } + bitidx += (bitlen+1); // +1 to skip the zero (or end) + } // while bitidx + } // purge != 0 + } + // if not fully purged, make sure to purge again in the future + if (!full_purge) { + const long delay = mi_arena_purge_delay(); + mi_msecs_t expected = 0; + mi_atomic_casi64_strong_acq_rel(&arena->purge_expire,&expected,_mi_clock_now() + delay); + } + return any_purged; +} + +static void mi_arenas_try_purge( bool force, bool visit_all ) +{ + if (_mi_preloading() || mi_arena_purge_delay() <= 0) return; // nothing will be scheduled + + // check if any arena needs purging? + const mi_msecs_t now = _mi_clock_now(); + mi_msecs_t arenas_expire = mi_atomic_loadi64_acquire(&mi_arenas_purge_expire); + if (!force && (arenas_expire == 0 || arenas_expire < now)) return; + + const size_t max_arena = mi_atomic_load_acquire(&mi_arena_count); + if (max_arena == 0) return; + + // allow only one thread to purge at a time + static mi_atomic_guard_t purge_guard; + mi_atomic_guard(&purge_guard) + { + // increase global expire: at most one purge per delay cycle + mi_atomic_storei64_release(&mi_arenas_purge_expire, now + mi_arena_purge_delay()); + size_t max_purge_count = (visit_all ? max_arena : 2); + bool all_visited = true; + for (size_t i = 0; i < max_arena; i++) { + mi_arena_t* arena = mi_atomic_load_ptr_acquire(mi_arena_t, &mi_arenas[i]); + if (arena != NULL) { + if (mi_arena_try_purge(arena, now, force)) { + if (max_purge_count <= 1) { + all_visited = false; + break; + } + max_purge_count--; + } + } + } + if (all_visited) { + // all arena's were visited and purged: reset global expire + mi_atomic_storei64_release(&mi_arenas_purge_expire, 0); + } + } +} + + +/* ----------------------------------------------------------- + Arena free +----------------------------------------------------------- */ + +void _mi_arena_free(void* p, size_t size, size_t committed_size, mi_memid_t memid) { + mi_assert_internal(size > 0); + mi_assert_internal(committed_size <= size); + if (p==NULL) return; + if (size==0) return; + const bool all_committed = (committed_size == size); + const size_t decommitted_size = (committed_size <= size ? size - committed_size : 0); + + // need to set all memory to undefined as some parts may still be marked as no_access (like padding etc.) + mi_track_mem_undefined(p,size); + + if (mi_memkind_is_os(memid.memkind)) { + // was a direct OS allocation, pass through + if (!all_committed && decommitted_size > 0) { + // if partially committed, adjust the committed stats (as `_mi_os_free` will decrease commit by the full size) + _mi_stat_increase(&_mi_stats_main.committed, decommitted_size); + } + _mi_os_free(p, size, memid); + } + else if (memid.memkind == MI_MEM_ARENA) { + // allocated in an arena + size_t arena_idx; + size_t bitmap_idx; + mi_arena_memid_indices(memid, &arena_idx, &bitmap_idx); + mi_assert_internal(arena_idx < MI_MAX_ARENAS); + mi_arena_t* arena = mi_atomic_load_ptr_acquire(mi_arena_t,&mi_arenas[arena_idx]); + mi_assert_internal(arena != NULL); + const size_t blocks = mi_block_count_of_size(size); + + // checks + if (arena == NULL) { + _mi_error_message(EINVAL, "trying to free from an invalid arena: %p, size %zu, memid: 0x%zx\n", p, size, memid); + return; + } + mi_assert_internal(arena->field_count > mi_bitmap_index_field(bitmap_idx)); + if (arena->field_count <= mi_bitmap_index_field(bitmap_idx)) { + _mi_error_message(EINVAL, "trying to free from an invalid arena block: %p, size %zu, memid: 0x%zx\n", p, size, memid); + return; + } + + // potentially decommit + if (arena->memid.is_pinned || arena->blocks_committed == NULL) { + mi_assert_internal(all_committed); + } + else { + mi_assert_internal(arena->blocks_committed != NULL); + mi_assert_internal(arena->blocks_purge != NULL); + + if (!all_committed) { + // mark the entire range as no longer committed (so we will recommit the full range when re-using) + _mi_bitmap_unclaim_across(arena->blocks_committed, arena->field_count, blocks, bitmap_idx); + mi_track_mem_noaccess(p,size); + //if (committed_size > 0) { + // if partially committed, adjust the committed stats (is it will be recommitted when re-using) + // in the delayed purge, we do no longer decrease the commit if the range is not marked entirely as committed. + _mi_stat_decrease(&_mi_stats_main.committed, committed_size); + //} + // note: if not all committed, it may be that the purge will reset/decommit the entire range + // that contains already decommitted parts. Since purge consistently uses reset or decommit that + // works (as we should never reset decommitted parts). + } + // (delay) purge the entire range + mi_arena_schedule_purge(arena, bitmap_idx, blocks); + } + + // and make it available to others again + bool all_inuse = _mi_bitmap_unclaim_across(arena->blocks_inuse, arena->field_count, blocks, bitmap_idx); + if (!all_inuse) { + _mi_error_message(EAGAIN, "trying to free an already freed arena block: %p, size %zu\n", p, size); + return; + }; + } + else { + // arena was none, external, or static; nothing to do + mi_assert_internal(memid.memkind < MI_MEM_OS); + } + + // purge expired decommits + mi_arenas_try_purge(false, false); +} + +// destroy owned arenas; this is unsafe and should only be done using `mi_option_destroy_on_exit` +// for dynamic libraries that are unloaded and need to release all their allocated memory. +static void mi_arenas_unsafe_destroy(void) { + const size_t max_arena = mi_atomic_load_relaxed(&mi_arena_count); + size_t new_max_arena = 0; + for (size_t i = 0; i < max_arena; i++) { + mi_arena_t* arena = mi_atomic_load_ptr_acquire(mi_arena_t, &mi_arenas[i]); + if (arena != NULL) { + mi_lock_done(&arena->abandoned_visit_lock); + if (arena->start != NULL && mi_memkind_is_os(arena->memid.memkind)) { + mi_atomic_store_ptr_release(mi_arena_t, &mi_arenas[i], NULL); + _mi_os_free(arena->start, mi_arena_size(arena), arena->memid); + } + else { + new_max_arena = i; + } + _mi_arena_meta_free(arena, arena->meta_memid, arena->meta_size); + } + } + + // try to lower the max arena. + size_t expected = max_arena; + mi_atomic_cas_strong_acq_rel(&mi_arena_count, &expected, new_max_arena); +} + +// Purge the arenas; if `force_purge` is true, amenable parts are purged even if not yet expired +void _mi_arenas_collect(bool force_purge) { + mi_arenas_try_purge(force_purge, force_purge /* visit all? */); +} + +// destroy owned arenas; this is unsafe and should only be done using `mi_option_destroy_on_exit` +// for dynamic libraries that are unloaded and need to release all their allocated memory. +void _mi_arena_unsafe_destroy_all(void) { + mi_arenas_unsafe_destroy(); + _mi_arenas_collect(true /* force purge */); // purge non-owned arenas +} + +// Is a pointer inside any of our arenas? +bool _mi_arena_contains(const void* p) { + const size_t max_arena = mi_atomic_load_relaxed(&mi_arena_count); + for (size_t i = 0; i < max_arena; i++) { + mi_arena_t* arena = mi_atomic_load_ptr_relaxed(mi_arena_t, &mi_arenas[i]); + if (arena != NULL && arena->start <= (const uint8_t*)p && arena->start + mi_arena_block_size(arena->block_count) > (const uint8_t*)p) { + return true; + } + } + return false; +} + +/* ----------------------------------------------------------- + Add an arena. +----------------------------------------------------------- */ + +static bool mi_arena_add(mi_arena_t* arena, mi_arena_id_t* arena_id, mi_stats_t* stats) { + mi_assert_internal(arena != NULL); + mi_assert_internal((uintptr_t)mi_atomic_load_ptr_relaxed(uint8_t,&arena->start) % MI_SEGMENT_ALIGN == 0); + mi_assert_internal(arena->block_count > 0); + if (arena_id != NULL) { *arena_id = -1; } + + size_t i = mi_atomic_increment_acq_rel(&mi_arena_count); + if (i >= MI_MAX_ARENAS) { + mi_atomic_decrement_acq_rel(&mi_arena_count); + return false; + } + _mi_stat_counter_increase(&stats->arena_count,1); + arena->id = mi_arena_id_create(i); + mi_atomic_store_ptr_release(mi_arena_t,&mi_arenas[i], arena); + if (arena_id != NULL) { *arena_id = arena->id; } + return true; +} + +static bool mi_manage_os_memory_ex2(void* start, size_t size, bool is_large, int numa_node, bool exclusive, mi_memid_t memid, mi_arena_id_t* arena_id) mi_attr_noexcept +{ + if (arena_id != NULL) *arena_id = _mi_arena_id_none(); + if (size < MI_ARENA_BLOCK_SIZE) { + _mi_warning_message("the arena size is too small (memory at %p with size %zu)\n", start, size); + return false; + } + if (is_large) { + mi_assert_internal(memid.initially_committed && memid.is_pinned); + } + if (!_mi_is_aligned(start, MI_SEGMENT_ALIGN)) { + void* const aligned_start = mi_align_up_ptr(start, MI_SEGMENT_ALIGN); + const size_t diff = (uint8_t*)aligned_start - (uint8_t*)start; + if (diff >= size || (size - diff) < MI_ARENA_BLOCK_SIZE) { + _mi_warning_message("after alignment, the size of the arena becomes too small (memory at %p with size %zu)\n", start, size); + return false; + } + start = aligned_start; + size = size - diff; + } + + const size_t bcount = size / MI_ARENA_BLOCK_SIZE; + const size_t fields = _mi_divide_up(bcount, MI_BITMAP_FIELD_BITS); + const size_t bitmaps = (memid.is_pinned ? 3 : 5); + const size_t asize = sizeof(mi_arena_t) + (bitmaps*fields*sizeof(mi_bitmap_field_t)); + mi_memid_t meta_memid; + mi_arena_t* arena = (mi_arena_t*)_mi_arena_meta_zalloc(asize, &meta_memid); + if (arena == NULL) return false; + + // already zero'd due to zalloc + // _mi_memzero(arena, asize); + arena->id = _mi_arena_id_none(); + arena->memid = memid; + arena->exclusive = exclusive; + arena->meta_size = asize; + arena->meta_memid = meta_memid; + arena->block_count = bcount; + arena->field_count = fields; + arena->start = (uint8_t*)start; + arena->numa_node = numa_node; // TODO: or get the current numa node if -1? (now it allows anyone to allocate on -1) + arena->is_large = is_large; + arena->purge_expire = 0; + arena->search_idx = 0; + mi_lock_init(&arena->abandoned_visit_lock); + // consecutive bitmaps + arena->blocks_dirty = &arena->blocks_inuse[fields]; // just after inuse bitmap + arena->blocks_abandoned = &arena->blocks_inuse[2 * fields]; // just after dirty bitmap + arena->blocks_committed = (arena->memid.is_pinned ? NULL : &arena->blocks_inuse[3*fields]); // just after abandoned bitmap + arena->blocks_purge = (arena->memid.is_pinned ? NULL : &arena->blocks_inuse[4*fields]); // just after committed bitmap + // initialize committed bitmap? + if (arena->blocks_committed != NULL && arena->memid.initially_committed) { + memset((void*)arena->blocks_committed, 0xFF, fields*sizeof(mi_bitmap_field_t)); // cast to void* to avoid atomic warning + } + + // and claim leftover blocks if needed (so we never allocate there) + ptrdiff_t post = (fields * MI_BITMAP_FIELD_BITS) - bcount; + mi_assert_internal(post >= 0); + if (post > 0) { + // don't use leftover bits at the end + mi_bitmap_index_t postidx = mi_bitmap_index_create(fields - 1, MI_BITMAP_FIELD_BITS - post); + _mi_bitmap_claim(arena->blocks_inuse, fields, post, postidx, NULL); + } + return mi_arena_add(arena, arena_id, &_mi_stats_main); + +} + +bool mi_manage_os_memory_ex(void* start, size_t size, bool is_committed, bool is_large, bool is_zero, int numa_node, bool exclusive, mi_arena_id_t* arena_id) mi_attr_noexcept { + mi_memid_t memid = _mi_memid_create(MI_MEM_EXTERNAL); + memid.initially_committed = is_committed; + memid.initially_zero = is_zero; + memid.is_pinned = is_large; + return mi_manage_os_memory_ex2(start,size,is_large,numa_node,exclusive,memid, arena_id); +} + +// Reserve a range of regular OS memory +int mi_reserve_os_memory_ex(size_t size, bool commit, bool allow_large, bool exclusive, mi_arena_id_t* arena_id) mi_attr_noexcept { + if (arena_id != NULL) *arena_id = _mi_arena_id_none(); + size = _mi_align_up(size, MI_ARENA_BLOCK_SIZE); // at least one block + mi_memid_t memid; + void* start = _mi_os_alloc_aligned(size, MI_SEGMENT_ALIGN, commit, allow_large, &memid); + if (start == NULL) return ENOMEM; + const bool is_large = memid.is_pinned; // todo: use separate is_large field? + if (!mi_manage_os_memory_ex2(start, size, is_large, -1 /* numa node */, exclusive, memid, arena_id)) { + _mi_os_free_ex(start, size, commit, memid); + _mi_verbose_message("failed to reserve %zu KiB memory\n", _mi_divide_up(size, 1024)); + return ENOMEM; + } + _mi_verbose_message("reserved %zu KiB memory%s\n", _mi_divide_up(size, 1024), is_large ? " (in large os pages)" : ""); + return 0; +} + + +// Manage a range of regular OS memory +bool mi_manage_os_memory(void* start, size_t size, bool is_committed, bool is_large, bool is_zero, int numa_node) mi_attr_noexcept { + return mi_manage_os_memory_ex(start, size, is_committed, is_large, is_zero, numa_node, false /* exclusive? */, NULL); +} + +// Reserve a range of regular OS memory +int mi_reserve_os_memory(size_t size, bool commit, bool allow_large) mi_attr_noexcept { + return mi_reserve_os_memory_ex(size, commit, allow_large, false, NULL); +} + + +/* ----------------------------------------------------------- + Debugging +----------------------------------------------------------- */ + +static size_t mi_debug_show_bitmap(const char* prefix, const char* header, size_t block_count, mi_bitmap_field_t* fields, size_t field_count ) { + _mi_message("%s%s:\n", prefix, header); + size_t bcount = 0; + size_t inuse_count = 0; + for (size_t i = 0; i < field_count; i++) { + char buf[MI_BITMAP_FIELD_BITS + 1]; + uintptr_t field = mi_atomic_load_relaxed(&fields[i]); + for (size_t bit = 0; bit < MI_BITMAP_FIELD_BITS; bit++, bcount++) { + if (bcount < block_count) { + bool inuse = ((((uintptr_t)1 << bit) & field) != 0); + if (inuse) inuse_count++; + buf[bit] = (inuse ? 'x' : '.'); + } + else { + buf[bit] = ' '; + } + } + buf[MI_BITMAP_FIELD_BITS] = 0; + _mi_message("%s %s\n", prefix, buf); + } + _mi_message("%s total ('x'): %zu\n", prefix, inuse_count); + return inuse_count; +} + +void mi_debug_show_arenas(void) mi_attr_noexcept { + const bool show_inuse = true; + size_t max_arenas = mi_atomic_load_relaxed(&mi_arena_count); + size_t inuse_total = 0; + //size_t abandoned_total = 0; + //size_t purge_total = 0; + for (size_t i = 0; i < max_arenas; i++) { + mi_arena_t* arena = mi_atomic_load_ptr_relaxed(mi_arena_t, &mi_arenas[i]); + if (arena == NULL) break; + _mi_message("arena %zu: %zu blocks of size %zuMiB (in %zu fields) %s\n", i, arena->block_count, MI_ARENA_BLOCK_SIZE / MI_MiB, arena->field_count, (arena->memid.is_pinned ? ", pinned" : "")); + if (show_inuse) { + inuse_total += mi_debug_show_bitmap(" ", "inuse blocks", arena->block_count, arena->blocks_inuse, arena->field_count); + } + if (arena->blocks_committed != NULL) { + mi_debug_show_bitmap(" ", "committed blocks", arena->block_count, arena->blocks_committed, arena->field_count); + } + //if (show_abandoned) { + // abandoned_total += mi_debug_show_bitmap(" ", "abandoned blocks", arena->block_count, arena->blocks_abandoned, arena->field_count); + //} + //if (show_purge && arena->blocks_purge != NULL) { + // purge_total += mi_debug_show_bitmap(" ", "purgeable blocks", arena->block_count, arena->blocks_purge, arena->field_count); + //} + } + if (show_inuse) _mi_message("total inuse blocks : %zu\n", inuse_total); + //if (show_abandoned) _mi_message("total abandoned blocks: %zu\n", abandoned_total); + //if (show_purge) _mi_message("total purgeable blocks: %zu\n", purge_total); +} + + +void mi_arenas_print(void) mi_attr_noexcept { + mi_debug_show_arenas(); +} + + +/* ----------------------------------------------------------- + Reserve a huge page arena. +----------------------------------------------------------- */ +// reserve at a specific numa node +int mi_reserve_huge_os_pages_at_ex(size_t pages, int numa_node, size_t timeout_msecs, bool exclusive, mi_arena_id_t* arena_id) mi_attr_noexcept { + if (arena_id != NULL) *arena_id = -1; + if (pages==0) return 0; + if (numa_node < -1) numa_node = -1; + if (numa_node >= 0) numa_node = numa_node % _mi_os_numa_node_count(); + size_t hsize = 0; + size_t pages_reserved = 0; + mi_memid_t memid; + void* p = _mi_os_alloc_huge_os_pages(pages, numa_node, timeout_msecs, &pages_reserved, &hsize, &memid); + if (p==NULL || pages_reserved==0) { + _mi_warning_message("failed to reserve %zu GiB huge pages\n", pages); + return ENOMEM; + } + _mi_verbose_message("numa node %i: reserved %zu GiB huge pages (of the %zu GiB requested)\n", numa_node, pages_reserved, pages); + + if (!mi_manage_os_memory_ex2(p, hsize, true, numa_node, exclusive, memid, arena_id)) { + _mi_os_free(p, hsize, memid); + return ENOMEM; + } + return 0; +} + +int mi_reserve_huge_os_pages_at(size_t pages, int numa_node, size_t timeout_msecs) mi_attr_noexcept { + return mi_reserve_huge_os_pages_at_ex(pages, numa_node, timeout_msecs, false, NULL); +} + +// reserve huge pages evenly among the given number of numa nodes (or use the available ones as detected) +int mi_reserve_huge_os_pages_interleave(size_t pages, size_t numa_nodes, size_t timeout_msecs) mi_attr_noexcept { + if (pages == 0) return 0; + + // pages per numa node + size_t numa_count = (numa_nodes > 0 ? numa_nodes : _mi_os_numa_node_count()); + if (numa_count <= 0) numa_count = 1; + const size_t pages_per = pages / numa_count; + const size_t pages_mod = pages % numa_count; + const size_t timeout_per = (timeout_msecs==0 ? 0 : (timeout_msecs / numa_count) + 50); + + // reserve evenly among numa nodes + for (size_t numa_node = 0; numa_node < numa_count && pages > 0; numa_node++) { + size_t node_pages = pages_per; // can be 0 + if (numa_node < pages_mod) node_pages++; + int err = mi_reserve_huge_os_pages_at(node_pages, (int)numa_node, timeout_per); + if (err) return err; + if (pages < node_pages) { + pages = 0; + } + else { + pages -= node_pages; + } + } + + return 0; +} + +int mi_reserve_huge_os_pages(size_t pages, double max_secs, size_t* pages_reserved) mi_attr_noexcept { + MI_UNUSED(max_secs); + _mi_warning_message("mi_reserve_huge_os_pages is deprecated: use mi_reserve_huge_os_pages_interleave/at instead\n"); + if (pages_reserved != NULL) *pages_reserved = 0; + int err = mi_reserve_huge_os_pages_interleave(pages, 0, (size_t)(max_secs * 1000.0)); + if (err==0 && pages_reserved!=NULL) *pages_reserved = pages; + return err; +} diff --git a/compat/mimalloc/bitmap.c b/compat/mimalloc/bitmap.c new file mode 100644 index 00000000000000..32d1e9548d3e3b --- /dev/null +++ b/compat/mimalloc/bitmap.c @@ -0,0 +1,441 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2019-2023 Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ + +/* ---------------------------------------------------------------------------- +Concurrent bitmap that can set/reset sequences of bits atomically, +represented as an array of fields where each field is a machine word (`size_t`) + +There are two api's; the standard one cannot have sequences that cross +between the bitmap fields (and a sequence must be <= MI_BITMAP_FIELD_BITS). + +The `_across` postfixed functions do allow sequences that can cross over +between the fields. (This is used in arena allocation) +---------------------------------------------------------------------------- */ + +#include "mimalloc.h" +#include "mimalloc/internal.h" +#include "bitmap.h" + +/* ----------------------------------------------------------- + Bitmap definition +----------------------------------------------------------- */ + +// The bit mask for a given number of blocks at a specified bit index. +static inline size_t mi_bitmap_mask_(size_t count, size_t bitidx) { + mi_assert_internal(count + bitidx <= MI_BITMAP_FIELD_BITS); + mi_assert_internal(count > 0); + if (count >= MI_BITMAP_FIELD_BITS) return MI_BITMAP_FIELD_FULL; + if (count == 0) return 0; + return ((((size_t)1 << count) - 1) << bitidx); +} + + +/* ----------------------------------------------------------- + Claim a bit sequence atomically +----------------------------------------------------------- */ + +// Try to atomically claim a sequence of `count` bits in a single +// field at `idx` in `bitmap`. Returns `true` on success. +inline bool _mi_bitmap_try_find_claim_field(mi_bitmap_t bitmap, size_t idx, const size_t count, mi_bitmap_index_t* bitmap_idx) +{ + mi_assert_internal(bitmap_idx != NULL); + mi_assert_internal(count <= MI_BITMAP_FIELD_BITS); + mi_assert_internal(count > 0); + mi_bitmap_field_t* field = &bitmap[idx]; + size_t map = mi_atomic_load_relaxed(field); + if (map==MI_BITMAP_FIELD_FULL) return false; // short cut + + // search for 0-bit sequence of length count + const size_t mask = mi_bitmap_mask_(count, 0); + const size_t bitidx_max = MI_BITMAP_FIELD_BITS - count; + +#ifdef MI_HAVE_FAST_BITSCAN + size_t bitidx = mi_ctz(~map); // quickly find the first zero bit if possible +#else + size_t bitidx = 0; // otherwise start at 0 +#endif + size_t m = (mask << bitidx); // invariant: m == mask shifted by bitidx + + // scan linearly for a free range of zero bits + while (bitidx <= bitidx_max) { + const size_t mapm = (map & m); + if (mapm == 0) { // are the mask bits free at bitidx? + mi_assert_internal((m >> bitidx) == mask); // no overflow? + const size_t newmap = (map | m); + mi_assert_internal((newmap^map) >> bitidx == mask); + if (!mi_atomic_cas_strong_acq_rel(field, &map, newmap)) { // TODO: use weak cas here? + // no success, another thread claimed concurrently.. keep going (with updated `map`) + continue; + } + else { + // success, we claimed the bits! + *bitmap_idx = mi_bitmap_index_create(idx, bitidx); + return true; + } + } + else { + // on to the next bit range +#ifdef MI_HAVE_FAST_BITSCAN + mi_assert_internal(mapm != 0); + const size_t shift = (count == 1 ? 1 : (MI_SIZE_BITS - mi_clz(mapm) - bitidx)); + mi_assert_internal(shift > 0 && shift <= count); +#else + const size_t shift = 1; +#endif + bitidx += shift; + m <<= shift; + } + } + // no bits found + return false; +} + +// Find `count` bits of 0 and set them to 1 atomically; returns `true` on success. +// Starts at idx, and wraps around to search in all `bitmap_fields` fields. +// `count` can be at most MI_BITMAP_FIELD_BITS and will never cross fields. +bool _mi_bitmap_try_find_from_claim(mi_bitmap_t bitmap, const size_t bitmap_fields, const size_t start_field_idx, const size_t count, mi_bitmap_index_t* bitmap_idx) { + size_t idx = start_field_idx; + for (size_t visited = 0; visited < bitmap_fields; visited++, idx++) { + if (idx >= bitmap_fields) { idx = 0; } // wrap + if (_mi_bitmap_try_find_claim_field(bitmap, idx, count, bitmap_idx)) { + return true; + } + } + return false; +} + +// Like _mi_bitmap_try_find_from_claim but with an extra predicate that must be fullfilled +bool _mi_bitmap_try_find_from_claim_pred(mi_bitmap_t bitmap, const size_t bitmap_fields, + const size_t start_field_idx, const size_t count, + mi_bitmap_pred_fun_t pred_fun, void* pred_arg, + mi_bitmap_index_t* bitmap_idx) { + size_t idx = start_field_idx; + for (size_t visited = 0; visited < bitmap_fields; visited++, idx++) { + if (idx >= bitmap_fields) idx = 0; // wrap + if (_mi_bitmap_try_find_claim_field(bitmap, idx, count, bitmap_idx)) { + if (pred_fun == NULL || pred_fun(*bitmap_idx, pred_arg)) { + return true; + } + // predicate returned false, unclaim and look further + _mi_bitmap_unclaim(bitmap, bitmap_fields, count, *bitmap_idx); + } + } + return false; +} + +// Set `count` bits at `bitmap_idx` to 0 atomically +// Returns `true` if all `count` bits were 1 previously. +bool _mi_bitmap_unclaim(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx) { + const size_t idx = mi_bitmap_index_field(bitmap_idx); + const size_t bitidx = mi_bitmap_index_bit_in_field(bitmap_idx); + const size_t mask = mi_bitmap_mask_(count, bitidx); + mi_assert_internal(bitmap_fields > idx); MI_UNUSED(bitmap_fields); + // mi_assert_internal((bitmap[idx] & mask) == mask); + const size_t prev = mi_atomic_and_acq_rel(&bitmap[idx], ~mask); + return ((prev & mask) == mask); +} + + +// Set `count` bits at `bitmap_idx` to 1 atomically +// Returns `true` if all `count` bits were 0 previously. `any_zero` is `true` if there was at least one zero bit. +bool _mi_bitmap_claim(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, bool* any_zero) { + const size_t idx = mi_bitmap_index_field(bitmap_idx); + const size_t bitidx = mi_bitmap_index_bit_in_field(bitmap_idx); + const size_t mask = mi_bitmap_mask_(count, bitidx); + mi_assert_internal(bitmap_fields > idx); MI_UNUSED(bitmap_fields); + //mi_assert_internal(any_zero != NULL || (bitmap[idx] & mask) == 0); + size_t prev = mi_atomic_or_acq_rel(&bitmap[idx], mask); + if (any_zero != NULL) { *any_zero = ((prev & mask) != mask); } + return ((prev & mask) == 0); +} + +// Returns `true` if all `count` bits were 1. `any_ones` is `true` if there was at least one bit set to one. +static bool mi_bitmap_is_claimedx(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, bool* any_ones) { + const size_t idx = mi_bitmap_index_field(bitmap_idx); + const size_t bitidx = mi_bitmap_index_bit_in_field(bitmap_idx); + const size_t mask = mi_bitmap_mask_(count, bitidx); + mi_assert_internal(bitmap_fields > idx); MI_UNUSED(bitmap_fields); + const size_t field = mi_atomic_load_relaxed(&bitmap[idx]); + if (any_ones != NULL) { *any_ones = ((field & mask) != 0); } + return ((field & mask) == mask); +} + +// Try to set `count` bits at `bitmap_idx` from 0 to 1 atomically. +// Returns `true` if successful when all previous `count` bits were 0. +bool _mi_bitmap_try_claim(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx) { + const size_t idx = mi_bitmap_index_field(bitmap_idx); + const size_t bitidx = mi_bitmap_index_bit_in_field(bitmap_idx); + const size_t mask = mi_bitmap_mask_(count, bitidx); + mi_assert_internal(bitmap_fields > idx); MI_UNUSED(bitmap_fields); + size_t expected = mi_atomic_load_relaxed(&bitmap[idx]); + do { + if ((expected & mask) != 0) return false; + } + while (!mi_atomic_cas_strong_acq_rel(&bitmap[idx], &expected, expected | mask)); + mi_assert_internal((expected & mask) == 0); + return true; +} + + +bool _mi_bitmap_is_claimed(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx) { + return mi_bitmap_is_claimedx(bitmap, bitmap_fields, count, bitmap_idx, NULL); +} + +bool _mi_bitmap_is_any_claimed(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx) { + bool any_ones; + mi_bitmap_is_claimedx(bitmap, bitmap_fields, count, bitmap_idx, &any_ones); + return any_ones; +} + + +//-------------------------------------------------------------------------- +// the `_across` functions work on bitmaps where sequences can cross over +// between the fields. This is used in arena allocation +//-------------------------------------------------------------------------- + +// Try to atomically claim a sequence of `count` bits starting from the field +// at `idx` in `bitmap` and crossing into subsequent fields. Returns `true` on success. +// Only needs to consider crossing into the next fields (see `mi_bitmap_try_find_from_claim_across`) +static bool mi_bitmap_try_find_claim_field_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t idx, const size_t count, const size_t retries, mi_bitmap_index_t* bitmap_idx) +{ + mi_assert_internal(bitmap_idx != NULL); + + // check initial trailing zeros + mi_bitmap_field_t* field = &bitmap[idx]; + size_t map = mi_atomic_load_relaxed(field); + const size_t initial = mi_clz(map); // count of initial zeros starting at idx + mi_assert_internal(initial <= MI_BITMAP_FIELD_BITS); + if (initial == 0) return false; + if (initial >= count) return _mi_bitmap_try_find_claim_field(bitmap, idx, count, bitmap_idx); // no need to cross fields (this case won't happen for us) + if (_mi_divide_up(count - initial, MI_BITMAP_FIELD_BITS) >= (bitmap_fields - idx)) return false; // not enough entries + + // scan ahead + size_t found = initial; + size_t mask = 0; // mask bits for the final field + while(found < count) { + field++; + map = mi_atomic_load_relaxed(field); + const size_t mask_bits = (found + MI_BITMAP_FIELD_BITS <= count ? MI_BITMAP_FIELD_BITS : (count - found)); + mi_assert_internal(mask_bits > 0 && mask_bits <= MI_BITMAP_FIELD_BITS); + mask = mi_bitmap_mask_(mask_bits, 0); + if ((map & mask) != 0) return false; // some part is already claimed + found += mask_bits; + } + mi_assert_internal(field < &bitmap[bitmap_fields]); + + // we found a range of contiguous zeros up to the final field; mask contains mask in the final field + // now try to claim the range atomically + mi_bitmap_field_t* const final_field = field; + const size_t final_mask = mask; + mi_bitmap_field_t* const initial_field = &bitmap[idx]; + const size_t initial_idx = MI_BITMAP_FIELD_BITS - initial; + const size_t initial_mask = mi_bitmap_mask_(initial, initial_idx); + + // initial field + size_t newmap; + field = initial_field; + map = mi_atomic_load_relaxed(field); + do { + newmap = (map | initial_mask); + if ((map & initial_mask) != 0) { goto rollback; }; + } while (!mi_atomic_cas_strong_acq_rel(field, &map, newmap)); + + // intermediate fields + while (++field < final_field) { + newmap = MI_BITMAP_FIELD_FULL; + map = 0; + if (!mi_atomic_cas_strong_acq_rel(field, &map, newmap)) { goto rollback; } + } + + // final field + mi_assert_internal(field == final_field); + map = mi_atomic_load_relaxed(field); + do { + newmap = (map | final_mask); + if ((map & final_mask) != 0) { goto rollback; } + } while (!mi_atomic_cas_strong_acq_rel(field, &map, newmap)); + + // claimed! + *bitmap_idx = mi_bitmap_index_create(idx, initial_idx); + return true; + +rollback: + // roll back intermediate fields + // (we just failed to claim `field` so decrement first) + while (--field > initial_field) { + newmap = 0; + map = MI_BITMAP_FIELD_FULL; + mi_assert_internal(mi_atomic_load_relaxed(field) == map); + mi_atomic_store_release(field, newmap); + } + if (field == initial_field) { // (if we failed on the initial field, `field + 1 == initial_field`) + map = mi_atomic_load_relaxed(field); + do { + mi_assert_internal((map & initial_mask) == initial_mask); + newmap = (map & ~initial_mask); + } while (!mi_atomic_cas_strong_acq_rel(field, &map, newmap)); + } + mi_stat_counter_increase(_mi_stats_main.arena_rollback_count,1); + // retry? (we make a recursive call instead of goto to be able to use const declarations) + if (retries <= 2) { + return mi_bitmap_try_find_claim_field_across(bitmap, bitmap_fields, idx, count, retries+1, bitmap_idx); + } + else { + return false; + } +} + + +// Find `count` bits of zeros and set them to 1 atomically; returns `true` on success. +// Starts at idx, and wraps around to search in all `bitmap_fields` fields. +bool _mi_bitmap_try_find_from_claim_across(mi_bitmap_t bitmap, const size_t bitmap_fields, const size_t start_field_idx, const size_t count, mi_bitmap_index_t* bitmap_idx) { + mi_assert_internal(count > 0); + if (count <= 2) { + // we don't bother with crossover fields for small counts + return _mi_bitmap_try_find_from_claim(bitmap, bitmap_fields, start_field_idx, count, bitmap_idx); + } + + // visit the fields + size_t idx = start_field_idx; + for (size_t visited = 0; visited < bitmap_fields; visited++, idx++) { + if (idx >= bitmap_fields) { idx = 0; } // wrap + // first try to claim inside a field + /* + if (count <= MI_BITMAP_FIELD_BITS) { + if (_mi_bitmap_try_find_claim_field(bitmap, idx, count, bitmap_idx)) { + return true; + } + } + */ + // if that fails, then try to claim across fields + if (mi_bitmap_try_find_claim_field_across(bitmap, bitmap_fields, idx, count, 0, bitmap_idx)) { + return true; + } + } + return false; +} + +// Helper for masks across fields; returns the mid count, post_mask may be 0 +static size_t mi_bitmap_mask_across(mi_bitmap_index_t bitmap_idx, size_t bitmap_fields, size_t count, size_t* pre_mask, size_t* mid_mask, size_t* post_mask) { + MI_UNUSED(bitmap_fields); + const size_t bitidx = mi_bitmap_index_bit_in_field(bitmap_idx); + if mi_likely(bitidx + count <= MI_BITMAP_FIELD_BITS) { + *pre_mask = mi_bitmap_mask_(count, bitidx); + *mid_mask = 0; + *post_mask = 0; + mi_assert_internal(mi_bitmap_index_field(bitmap_idx) < bitmap_fields); + return 0; + } + else { + const size_t pre_bits = MI_BITMAP_FIELD_BITS - bitidx; + mi_assert_internal(pre_bits < count); + *pre_mask = mi_bitmap_mask_(pre_bits, bitidx); + count -= pre_bits; + const size_t mid_count = (count / MI_BITMAP_FIELD_BITS); + *mid_mask = MI_BITMAP_FIELD_FULL; + count %= MI_BITMAP_FIELD_BITS; + *post_mask = (count==0 ? 0 : mi_bitmap_mask_(count, 0)); + mi_assert_internal(mi_bitmap_index_field(bitmap_idx) + mid_count + (count==0 ? 0 : 1) < bitmap_fields); + return mid_count; + } +} + +// Set `count` bits at `bitmap_idx` to 0 atomically +// Returns `true` if all `count` bits were 1 previously. +bool _mi_bitmap_unclaim_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx) { + size_t idx = mi_bitmap_index_field(bitmap_idx); + size_t pre_mask; + size_t mid_mask; + size_t post_mask; + size_t mid_count = mi_bitmap_mask_across(bitmap_idx, bitmap_fields, count, &pre_mask, &mid_mask, &post_mask); + bool all_one = true; + mi_bitmap_field_t* field = &bitmap[idx]; + size_t prev = mi_atomic_and_acq_rel(field++, ~pre_mask); // clear first part + if ((prev & pre_mask) != pre_mask) all_one = false; + while(mid_count-- > 0) { + prev = mi_atomic_and_acq_rel(field++, ~mid_mask); // clear mid part + if ((prev & mid_mask) != mid_mask) all_one = false; + } + if (post_mask!=0) { + prev = mi_atomic_and_acq_rel(field, ~post_mask); // clear end part + if ((prev & post_mask) != post_mask) all_one = false; + } + return all_one; +} + +// Set `count` bits at `bitmap_idx` to 1 atomically +// Returns `true` if all `count` bits were 0 previously. `any_zero` is `true` if there was at least one zero bit. +bool _mi_bitmap_claim_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, bool* pany_zero, size_t* already_set) { + size_t idx = mi_bitmap_index_field(bitmap_idx); + size_t pre_mask; + size_t mid_mask; + size_t post_mask; + size_t mid_count = mi_bitmap_mask_across(bitmap_idx, bitmap_fields, count, &pre_mask, &mid_mask, &post_mask); + bool all_zero = true; + bool any_zero = false; + size_t one_count = 0; + _Atomic(size_t)*field = &bitmap[idx]; + size_t prev = mi_atomic_or_acq_rel(field++, pre_mask); + if ((prev & pre_mask) != 0) { all_zero = false; one_count += mi_popcount(prev & pre_mask); } + if ((prev & pre_mask) != pre_mask) any_zero = true; + while (mid_count-- > 0) { + prev = mi_atomic_or_acq_rel(field++, mid_mask); + if ((prev & mid_mask) != 0) { all_zero = false; one_count += mi_popcount(prev & mid_mask); } + if ((prev & mid_mask) != mid_mask) any_zero = true; + } + if (post_mask!=0) { + prev = mi_atomic_or_acq_rel(field, post_mask); + if ((prev & post_mask) != 0) { all_zero = false; one_count += mi_popcount(prev & post_mask); } + if ((prev & post_mask) != post_mask) any_zero = true; + } + if (pany_zero != NULL) { *pany_zero = any_zero; } + if (already_set != NULL) { *already_set = one_count; }; + mi_assert_internal(all_zero ? one_count == 0 : one_count <= count); + return all_zero; +} + + +// Returns `true` if all `count` bits were 1. +// `any_ones` is `true` if there was at least one bit set to one. +static bool mi_bitmap_is_claimedx_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, bool* pany_ones, size_t* already_set) { + size_t idx = mi_bitmap_index_field(bitmap_idx); + size_t pre_mask; + size_t mid_mask; + size_t post_mask; + size_t mid_count = mi_bitmap_mask_across(bitmap_idx, bitmap_fields, count, &pre_mask, &mid_mask, &post_mask); + bool all_ones = true; + bool any_ones = false; + size_t one_count = 0; + mi_bitmap_field_t* field = &bitmap[idx]; + size_t prev = mi_atomic_load_relaxed(field++); + if ((prev & pre_mask) != pre_mask) all_ones = false; + if ((prev & pre_mask) != 0) { any_ones = true; one_count += mi_popcount(prev & pre_mask); } + while (mid_count-- > 0) { + prev = mi_atomic_load_relaxed(field++); + if ((prev & mid_mask) != mid_mask) all_ones = false; + if ((prev & mid_mask) != 0) { any_ones = true; one_count += mi_popcount(prev & mid_mask); } + } + if (post_mask!=0) { + prev = mi_atomic_load_relaxed(field); + if ((prev & post_mask) != post_mask) all_ones = false; + if ((prev & post_mask) != 0) { any_ones = true; one_count += mi_popcount(prev & post_mask); } + } + if (pany_ones != NULL) { *pany_ones = any_ones; } + if (already_set != NULL) { *already_set = one_count; } + mi_assert_internal(all_ones ? one_count == count : one_count < count); + return all_ones; +} + +bool _mi_bitmap_is_claimed_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, size_t* already_set) { + return mi_bitmap_is_claimedx_across(bitmap, bitmap_fields, count, bitmap_idx, NULL, already_set); +} + +bool _mi_bitmap_is_any_claimed_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx) { + bool any_ones; + mi_bitmap_is_claimedx_across(bitmap, bitmap_fields, count, bitmap_idx, &any_ones, NULL); + return any_ones; +} diff --git a/compat/mimalloc/bitmap.h b/compat/mimalloc/bitmap.h new file mode 100644 index 00000000000000..0f4744f4fc3ffd --- /dev/null +++ b/compat/mimalloc/bitmap.h @@ -0,0 +1,119 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2019-2023 Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ + +/* ---------------------------------------------------------------------------- +Concurrent bitmap that can set/reset sequences of bits atomically, +represented as an array of fields where each field is a machine word (`size_t`) + +There are two api's; the standard one cannot have sequences that cross +between the bitmap fields (and a sequence must be <= MI_BITMAP_FIELD_BITS). +(this is used in region allocation) + +The `_across` postfixed functions do allow sequences that can cross over +between the fields. (This is used in arena allocation) +---------------------------------------------------------------------------- */ +#pragma once +#ifndef MI_BITMAP_H +#define MI_BITMAP_H + +/* ----------------------------------------------------------- + Bitmap definition +----------------------------------------------------------- */ + +#define MI_BITMAP_FIELD_BITS (8*MI_SIZE_SIZE) +#define MI_BITMAP_FIELD_FULL (~((size_t)0)) // all bits set + +// An atomic bitmap of `size_t` fields +typedef _Atomic(size_t) mi_bitmap_field_t; +typedef mi_bitmap_field_t* mi_bitmap_t; + +// A bitmap index is the index of the bit in a bitmap. +typedef size_t mi_bitmap_index_t; + +// Create a bit index. +static inline mi_bitmap_index_t mi_bitmap_index_create_ex(size_t idx, size_t bitidx) { + mi_assert_internal(bitidx <= MI_BITMAP_FIELD_BITS); + return (idx*MI_BITMAP_FIELD_BITS) + bitidx; +} +static inline mi_bitmap_index_t mi_bitmap_index_create(size_t idx, size_t bitidx) { + mi_assert_internal(bitidx < MI_BITMAP_FIELD_BITS); + return mi_bitmap_index_create_ex(idx,bitidx); +} + +// Create a bit index. +static inline mi_bitmap_index_t mi_bitmap_index_create_from_bit(size_t full_bitidx) { + return mi_bitmap_index_create(full_bitidx / MI_BITMAP_FIELD_BITS, full_bitidx % MI_BITMAP_FIELD_BITS); +} + +// Get the field index from a bit index. +static inline size_t mi_bitmap_index_field(mi_bitmap_index_t bitmap_idx) { + return (bitmap_idx / MI_BITMAP_FIELD_BITS); +} + +// Get the bit index in a bitmap field +static inline size_t mi_bitmap_index_bit_in_field(mi_bitmap_index_t bitmap_idx) { + return (bitmap_idx % MI_BITMAP_FIELD_BITS); +} + +// Get the full bit index +static inline size_t mi_bitmap_index_bit(mi_bitmap_index_t bitmap_idx) { + return bitmap_idx; +} + +/* ----------------------------------------------------------- + Claim a bit sequence atomically +----------------------------------------------------------- */ + +// Try to atomically claim a sequence of `count` bits in a single +// field at `idx` in `bitmap`. Returns `true` on success. +bool _mi_bitmap_try_find_claim_field(mi_bitmap_t bitmap, size_t idx, const size_t count, mi_bitmap_index_t* bitmap_idx); + +// Starts at idx, and wraps around to search in all `bitmap_fields` fields. +// For now, `count` can be at most MI_BITMAP_FIELD_BITS and will never cross fields. +bool _mi_bitmap_try_find_from_claim(mi_bitmap_t bitmap, const size_t bitmap_fields, const size_t start_field_idx, const size_t count, mi_bitmap_index_t* bitmap_idx); + +// Like _mi_bitmap_try_find_from_claim but with an extra predicate that must be fullfilled +typedef bool (mi_cdecl *mi_bitmap_pred_fun_t)(mi_bitmap_index_t bitmap_idx, void* pred_arg); +bool _mi_bitmap_try_find_from_claim_pred(mi_bitmap_t bitmap, const size_t bitmap_fields, const size_t start_field_idx, const size_t count, mi_bitmap_pred_fun_t pred_fun, void* pred_arg, mi_bitmap_index_t* bitmap_idx); + +// Set `count` bits at `bitmap_idx` to 0 atomically +// Returns `true` if all `count` bits were 1 previously. +bool _mi_bitmap_unclaim(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx); + +// Try to set `count` bits at `bitmap_idx` from 0 to 1 atomically. +// Returns `true` if successful when all previous `count` bits were 0. +bool _mi_bitmap_try_claim(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx); + +// Set `count` bits at `bitmap_idx` to 1 atomically +// Returns `true` if all `count` bits were 0 previously. `any_zero` is `true` if there was at least one zero bit. +bool _mi_bitmap_claim(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, bool* any_zero); + +bool _mi_bitmap_is_claimed(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx); +bool _mi_bitmap_is_any_claimed(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx); + + +//-------------------------------------------------------------------------- +// the `_across` functions work on bitmaps where sequences can cross over +// between the fields. This is used in arena allocation +//-------------------------------------------------------------------------- + +// Find `count` bits of zeros and set them to 1 atomically; returns `true` on success. +// Starts at idx, and wraps around to search in all `bitmap_fields` fields. +bool _mi_bitmap_try_find_from_claim_across(mi_bitmap_t bitmap, const size_t bitmap_fields, const size_t start_field_idx, const size_t count, mi_bitmap_index_t* bitmap_idx); + +// Set `count` bits at `bitmap_idx` to 0 atomically +// Returns `true` if all `count` bits were 1 previously. +bool _mi_bitmap_unclaim_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx); + +// Set `count` bits at `bitmap_idx` to 1 atomically +// Returns `true` if all `count` bits were 0 previously. `any_zero` is `true` if there was at least one zero bit. +bool _mi_bitmap_claim_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, bool* pany_zero, size_t* already_set); + +bool _mi_bitmap_is_claimed_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx, size_t* already_set); +bool _mi_bitmap_is_any_claimed_across(mi_bitmap_t bitmap, size_t bitmap_fields, size_t count, mi_bitmap_index_t bitmap_idx); + +#endif diff --git a/compat/mimalloc/free.c b/compat/mimalloc/free.c new file mode 100644 index 00000000000000..3b9067383f1fb6 --- /dev/null +++ b/compat/mimalloc/free.c @@ -0,0 +1,569 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2024, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ +#if !defined(MI_IN_ALLOC_C) +#error "this file should be included from 'alloc.c' (so aliases can work from alloc-override)" +// add includes help an IDE +#include "mimalloc.h" +#include "mimalloc/internal.h" +#include "mimalloc/prim.h" // _mi_prim_thread_id() +#endif + +// forward declarations +static void mi_check_padding(const mi_page_t* page, const mi_block_t* block); +static bool mi_check_is_double_free(const mi_page_t* page, const mi_block_t* block); +static size_t mi_page_usable_size_of(const mi_page_t* page, const mi_block_t* block); +static void mi_stat_free(const mi_page_t* page, const mi_block_t* block); + + +// ------------------------------------------------------ +// Free +// ------------------------------------------------------ + +// forward declaration of multi-threaded free (`_mt`) (or free in huge block if compiled with MI_HUGE_PAGE_ABANDON) +static mi_decl_noinline void mi_free_block_mt(mi_page_t* page, mi_segment_t* segment, mi_block_t* block); + +// regular free of a (thread local) block pointer +// fast path written carefully to prevent spilling on the stack +static inline void mi_free_block_local(mi_page_t* page, mi_block_t* block, bool track_stats, bool check_full) +{ + // checks + if mi_unlikely(mi_check_is_double_free(page, block)) return; + mi_check_padding(page, block); + if (track_stats) { mi_stat_free(page, block); } + #if (MI_DEBUG>0) && !MI_TRACK_ENABLED && !MI_TSAN && !MI_GUARDED + if (!mi_page_is_huge(page)) { // huge page content may be already decommitted + memset(block, MI_DEBUG_FREED, mi_page_block_size(page)); + } + #endif + if (track_stats) { mi_track_free_size(block, mi_page_usable_size_of(page, block)); } // faster then mi_usable_size as we already know the page and that p is unaligned + + // actual free: push on the local free list + mi_block_set_next(page, block, page->local_free); + page->local_free = block; + if mi_unlikely(--page->used == 0) { + _mi_page_retire(page); + } + else if mi_unlikely(check_full && mi_page_is_in_full(page)) { + _mi_page_unfull(page); + } +} + +// Adjust a block that was allocated aligned, to the actual start of the block in the page. +// note: this can be called from `mi_free_generic_mt` where a non-owning thread accesses the +// `page_start` and `block_size` fields; however these are constant and the page won't be +// deallocated (as the block we are freeing keeps it alive) and thus safe to read concurrently. +mi_block_t* _mi_page_ptr_unalign(const mi_page_t* page, const void* p) { + mi_assert_internal(page!=NULL && p!=NULL); + + size_t diff = (uint8_t*)p - page->page_start; + size_t adjust; + if mi_likely(page->block_size_shift != 0) { + adjust = diff & (((size_t)1 << page->block_size_shift) - 1); + } + else { + adjust = diff % mi_page_block_size(page); + } + + return (mi_block_t*)((uintptr_t)p - adjust); +} + +// forward declaration for a MI_GUARDED build +#if MI_GUARDED +static void mi_block_unguard(mi_page_t* page, mi_block_t* block, void* p); // forward declaration +static inline void mi_block_check_unguard(mi_page_t* page, mi_block_t* block, void* p) { + if (mi_block_ptr_is_guarded(block, p)) { mi_block_unguard(page, block, p); } +} +#else +static inline void mi_block_check_unguard(mi_page_t* page, mi_block_t* block, void* p) { + MI_UNUSED(page); MI_UNUSED(block); MI_UNUSED(p); +} +#endif + +// free a local pointer (page parameter comes first for better codegen) +static void mi_decl_noinline mi_free_generic_local(mi_page_t* page, mi_segment_t* segment, void* p) mi_attr_noexcept { + MI_UNUSED(segment); + mi_block_t* const block = (mi_page_has_aligned(page) ? _mi_page_ptr_unalign(page, p) : (mi_block_t*)p); + mi_block_check_unguard(page, block, p); + mi_free_block_local(page, block, true /* track stats */, true /* check for a full page */); +} + +// free a pointer owned by another thread (page parameter comes first for better codegen) +static void mi_decl_noinline mi_free_generic_mt(mi_page_t* page, mi_segment_t* segment, void* p) mi_attr_noexcept { + mi_block_t* const block = _mi_page_ptr_unalign(page, p); // don't check `has_aligned` flag to avoid a race (issue #865) + mi_block_check_unguard(page, block, p); + mi_free_block_mt(page, segment, block); +} + +// generic free (for runtime integration) +void mi_decl_noinline _mi_free_generic(mi_segment_t* segment, mi_page_t* page, bool is_local, void* p) mi_attr_noexcept { + if (is_local) mi_free_generic_local(page,segment,p); + else mi_free_generic_mt(page,segment,p); +} + +// Get the segment data belonging to a pointer +// This is just a single `and` in release mode but does further checks in debug mode +// (and secure mode) to see if this was a valid pointer. +static inline mi_segment_t* mi_checked_ptr_segment(const void* p, const char* msg) +{ + MI_UNUSED(msg); + + #if (MI_DEBUG>0) + if mi_unlikely(((uintptr_t)p & (MI_INTPTR_SIZE - 1)) != 0 && !mi_option_is_enabled(mi_option_guarded_precise)) { + _mi_error_message(EINVAL, "%s: invalid (unaligned) pointer: %p\n", msg, p); + return NULL; + } + #endif + + mi_segment_t* const segment = _mi_ptr_segment(p); + if mi_unlikely(segment==NULL) return segment; + + #if (MI_DEBUG>0) + if mi_unlikely(!mi_is_in_heap_region(p)) { + #if (MI_INTPTR_SIZE == 8 && defined(__linux__)) + if (((uintptr_t)p >> 40) != 0x7F) { // linux tends to align large blocks above 0x7F000000000 (issue #640) + #else + { + #endif + _mi_warning_message("%s: pointer might not point to a valid heap region: %p\n" + "(this may still be a valid very large allocation (over 64MiB))\n", msg, p); + if mi_likely(_mi_ptr_cookie(segment) == segment->cookie) { + _mi_warning_message("(yes, the previous pointer %p was valid after all)\n", p); + } + } + } + #endif + #if (MI_DEBUG>0 || MI_SECURE>=4) + if mi_unlikely(_mi_ptr_cookie(segment) != segment->cookie) { + _mi_error_message(EINVAL, "%s: pointer does not point to a valid heap space: %p\n", msg, p); + return NULL; + } + #endif + + return segment; +} + +// Free a block +// Fast path written carefully to prevent register spilling on the stack +void mi_free(void* p) mi_attr_noexcept +{ + mi_segment_t* const segment = mi_checked_ptr_segment(p,"mi_free"); + if mi_unlikely(segment==NULL) return; + + const bool is_local = (_mi_prim_thread_id() == mi_atomic_load_relaxed(&segment->thread_id)); + mi_page_t* const page = _mi_segment_page_of(segment, p); + + if mi_likely(is_local) { // thread-local free? + if mi_likely(page->flags.full_aligned == 0) { // and it is not a full page (full pages need to move from the full bin), nor has aligned blocks (aligned blocks need to be unaligned) + // thread-local, aligned, and not a full page + mi_block_t* const block = (mi_block_t*)p; + mi_free_block_local(page, block, true /* track stats */, false /* no need to check if the page is full */); + } + else { + // page is full or contains (inner) aligned blocks; use generic path + mi_free_generic_local(page, segment, p); + } + } + else { + // not thread-local; use generic path + mi_free_generic_mt(page, segment, p); + } +} + +// return true if successful +bool _mi_free_delayed_block(mi_block_t* block) { + // get segment and page + mi_assert_internal(block!=NULL); + const mi_segment_t* const segment = _mi_ptr_segment(block); + mi_assert_internal(_mi_ptr_cookie(segment) == segment->cookie); + mi_assert_internal(_mi_thread_id() == segment->thread_id); + mi_page_t* const page = _mi_segment_page_of(segment, block); + + // Clear the no-delayed flag so delayed freeing is used again for this page. + // This must be done before collecting the free lists on this page -- otherwise + // some blocks may end up in the page `thread_free` list with no blocks in the + // heap `thread_delayed_free` list which may cause the page to be never freed! + // (it would only be freed if we happen to scan it in `mi_page_queue_find_free_ex`) + if (!_mi_page_try_use_delayed_free(page, MI_USE_DELAYED_FREE, false /* dont overwrite never delayed */)) { + return false; + } + + // collect all other non-local frees (move from `thread_free` to `free`) to ensure up-to-date `used` count + _mi_page_free_collect(page, false); + + // and free the block (possibly freeing the page as well since `used` is updated) + mi_free_block_local(page, block, false /* stats have already been adjusted */, true /* check for a full page */); + return true; +} + +// ------------------------------------------------------ +// Multi-threaded Free (`_mt`) +// ------------------------------------------------------ + +// Push a block that is owned by another thread on its page-local thread free +// list or it's heap delayed free list. Such blocks are later collected by +// the owning thread in `_mi_free_delayed_block`. +static void mi_decl_noinline mi_free_block_delayed_mt( mi_page_t* page, mi_block_t* block ) +{ + // Try to put the block on either the page-local thread free list, + // or the heap delayed free list (if this is the first non-local free in that page) + mi_thread_free_t tfreex; + bool use_delayed; + mi_thread_free_t tfree = mi_atomic_load_relaxed(&page->xthread_free); + do { + use_delayed = (mi_tf_delayed(tfree) == MI_USE_DELAYED_FREE); + if mi_unlikely(use_delayed) { + // unlikely: this only happens on the first concurrent free in a page that is in the full list + tfreex = mi_tf_set_delayed(tfree,MI_DELAYED_FREEING); + } + else { + // usual: directly add to page thread_free list + mi_block_set_next(page, block, mi_tf_block(tfree)); + tfreex = mi_tf_set_block(tfree,block); + } + } while (!mi_atomic_cas_weak_release(&page->xthread_free, &tfree, tfreex)); + + // If this was the first non-local free, we need to push it on the heap delayed free list instead + if mi_unlikely(use_delayed) { + // racy read on `heap`, but ok because MI_DELAYED_FREEING is set (see `mi_heap_delete` and `mi_heap_collect_abandon`) + mi_heap_t* const heap = (mi_heap_t*)(mi_atomic_load_acquire(&page->xheap)); //mi_page_heap(page); + mi_assert_internal(heap != NULL); + if (heap != NULL) { + // add to the delayed free list of this heap. (do this atomically as the lock only protects heap memory validity) + mi_block_t* dfree = mi_atomic_load_ptr_relaxed(mi_block_t, &heap->thread_delayed_free); + do { + mi_block_set_nextx(heap,block,dfree, heap->keys); + } while (!mi_atomic_cas_ptr_weak_release(mi_block_t,&heap->thread_delayed_free, &dfree, block)); + } + + // and reset the MI_DELAYED_FREEING flag + tfree = mi_atomic_load_relaxed(&page->xthread_free); + do { + tfreex = tfree; + mi_assert_internal(mi_tf_delayed(tfree) == MI_DELAYED_FREEING); + tfreex = mi_tf_set_delayed(tfree,MI_NO_DELAYED_FREE); + } while (!mi_atomic_cas_weak_release(&page->xthread_free, &tfree, tfreex)); + } +} + +// Multi-threaded free (`_mt`) (or free in huge block if compiled with MI_HUGE_PAGE_ABANDON) +static void mi_decl_noinline mi_free_block_mt(mi_page_t* page, mi_segment_t* segment, mi_block_t* block) +{ + // first see if the segment was abandoned and if we can reclaim it into our thread + if (_mi_option_get_fast(mi_option_abandoned_reclaim_on_free) != 0 && + #if MI_HUGE_PAGE_ABANDON + segment->page_kind != MI_PAGE_HUGE && + #endif + mi_atomic_load_relaxed(&segment->thread_id) == 0 && // segment is abandoned? + mi_prim_get_default_heap() != (mi_heap_t*)&_mi_heap_empty) // and we did not already exit this thread (without this check, a fresh heap will be initalized (issue #944)) + { + // the segment is abandoned, try to reclaim it into our heap + if (_mi_segment_attempt_reclaim(mi_heap_get_default(), segment)) { + mi_assert_internal(_mi_thread_id() == mi_atomic_load_relaxed(&segment->thread_id)); + mi_assert_internal(mi_heap_get_default()->tld->segments.subproc == segment->subproc); + mi_free(block); // recursively free as now it will be a local free in our heap + return; + } + } + + // The padding check may access the non-thread-owned page for the key values. + // that is safe as these are constant and the page won't be freed (as the block is not freed yet). + mi_check_padding(page, block); + + // adjust stats (after padding check and potentially recursive `mi_free` above) + mi_stat_free(page, block); // stat_free may access the padding + mi_track_free_size(block, mi_page_usable_size_of(page,block)); + + // for small size, ensure we can fit the delayed thread pointers without triggering overflow detection + _mi_padding_shrink(page, block, sizeof(mi_block_t)); + + if (segment->kind == MI_SEGMENT_HUGE) { + #if MI_HUGE_PAGE_ABANDON + // huge page segments are always abandoned and can be freed immediately + _mi_segment_huge_page_free(segment, page, block); + return; + #else + // huge pages are special as they occupy the entire segment + // as these are large we reset the memory occupied by the page so it is available to other threads + // (as the owning thread needs to actually free the memory later). + _mi_segment_huge_page_reset(segment, page, block); + #endif + } + else { + #if (MI_DEBUG>0) && !MI_TRACK_ENABLED && !MI_TSAN // note: when tracking, cannot use mi_usable_size with multi-threading + memset(block, MI_DEBUG_FREED, mi_usable_size(block)); + #endif + } + + // and finally free the actual block by pushing it on the owning heap + // thread_delayed free list (or heap delayed free list) + mi_free_block_delayed_mt(page,block); +} + + +// ------------------------------------------------------ +// Usable size +// ------------------------------------------------------ + +// Bytes available in a block +static size_t mi_decl_noinline mi_page_usable_aligned_size_of(const mi_page_t* page, const void* p) mi_attr_noexcept { + const mi_block_t* block = _mi_page_ptr_unalign(page, p); + const size_t size = mi_page_usable_size_of(page, block); + const ptrdiff_t adjust = (uint8_t*)p - (uint8_t*)block; + mi_assert_internal(adjust >= 0 && (size_t)adjust <= size); + const size_t aligned_size = (size - adjust); + #if MI_GUARDED + if (mi_block_ptr_is_guarded(block, p)) { + return aligned_size - _mi_os_page_size(); + } + #endif + return aligned_size; +} + +static inline size_t _mi_usable_size(const void* p, const char* msg) mi_attr_noexcept { + const mi_segment_t* const segment = mi_checked_ptr_segment(p, msg); + if mi_unlikely(segment==NULL) return 0; + const mi_page_t* const page = _mi_segment_page_of(segment, p); + if mi_likely(!mi_page_has_aligned(page)) { + const mi_block_t* block = (const mi_block_t*)p; + return mi_page_usable_size_of(page, block); + } + else { + // split out to separate routine for improved code generation + return mi_page_usable_aligned_size_of(page, p); + } +} + +mi_decl_nodiscard size_t mi_usable_size(const void* p) mi_attr_noexcept { + return _mi_usable_size(p, "mi_usable_size"); +} + + +// ------------------------------------------------------ +// Free variants +// ------------------------------------------------------ + +void mi_free_size(void* p, size_t size) mi_attr_noexcept { + MI_UNUSED_RELEASE(size); + mi_assert(p == NULL || size <= _mi_usable_size(p,"mi_free_size")); + mi_free(p); +} + +void mi_free_size_aligned(void* p, size_t size, size_t alignment) mi_attr_noexcept { + MI_UNUSED_RELEASE(alignment); + mi_assert(((uintptr_t)p % alignment) == 0); + mi_free_size(p,size); +} + +void mi_free_aligned(void* p, size_t alignment) mi_attr_noexcept { + MI_UNUSED_RELEASE(alignment); + mi_assert(((uintptr_t)p % alignment) == 0); + mi_free(p); +} + + +// ------------------------------------------------------ +// Check for double free in secure and debug mode +// This is somewhat expensive so only enabled for secure mode 4 +// ------------------------------------------------------ + +#if (MI_ENCODE_FREELIST && (MI_SECURE>=4 || MI_DEBUG!=0)) +// linear check if the free list contains a specific element +static bool mi_list_contains(const mi_page_t* page, const mi_block_t* list, const mi_block_t* elem) { + while (list != NULL) { + if (elem==list) return true; + list = mi_block_next(page, list); + } + return false; +} + +static mi_decl_noinline bool mi_check_is_double_freex(const mi_page_t* page, const mi_block_t* block) { + // The decoded value is in the same page (or NULL). + // Walk the free lists to verify positively if it is already freed + if (mi_list_contains(page, page->free, block) || + mi_list_contains(page, page->local_free, block) || + mi_list_contains(page, mi_page_thread_free(page), block)) + { + _mi_error_message(EAGAIN, "double free detected of block %p with size %zu\n", block, mi_page_block_size(page)); + return true; + } + return false; +} + +#define mi_track_page(page,access) { size_t psize; void* pstart = _mi_page_start(_mi_page_segment(page),page,&psize); mi_track_mem_##access( pstart, psize); } + +static inline bool mi_check_is_double_free(const mi_page_t* page, const mi_block_t* block) { + bool is_double_free = false; + mi_block_t* n = mi_block_nextx(page, block, page->keys); // pretend it is freed, and get the decoded first field + if (((uintptr_t)n & (MI_INTPTR_SIZE-1))==0 && // quick check: aligned pointer? + (n==NULL || mi_is_in_same_page(block, n))) // quick check: in same page or NULL? + { + // Suspicious: decoded value a in block is in the same page (or NULL) -- maybe a double free? + // (continue in separate function to improve code generation) + is_double_free = mi_check_is_double_freex(page, block); + } + return is_double_free; +} +#else +static inline bool mi_check_is_double_free(const mi_page_t* page, const mi_block_t* block) { + MI_UNUSED(page); + MI_UNUSED(block); + return false; +} +#endif + + +// --------------------------------------------------------------------------- +// Check for heap block overflow by setting up padding at the end of the block +// --------------------------------------------------------------------------- + +#if MI_PADDING // && !MI_TRACK_ENABLED +static bool mi_page_decode_padding(const mi_page_t* page, const mi_block_t* block, size_t* delta, size_t* bsize) { + *bsize = mi_page_usable_block_size(page); + const mi_padding_t* const padding = (mi_padding_t*)((uint8_t*)block + *bsize); + mi_track_mem_defined(padding,sizeof(mi_padding_t)); + *delta = padding->delta; + uint32_t canary = padding->canary; + uintptr_t keys[2]; + keys[0] = page->keys[0]; + keys[1] = page->keys[1]; + bool ok = (mi_ptr_encode_canary(page,block,keys) == canary && *delta <= *bsize); + mi_track_mem_noaccess(padding,sizeof(mi_padding_t)); + return ok; +} + +// Return the exact usable size of a block. +static size_t mi_page_usable_size_of(const mi_page_t* page, const mi_block_t* block) { + size_t bsize; + size_t delta; + bool ok = mi_page_decode_padding(page, block, &delta, &bsize); + mi_assert_internal(ok); mi_assert_internal(delta <= bsize); + return (ok ? bsize - delta : 0); +} + +// When a non-thread-local block is freed, it becomes part of the thread delayed free +// list that is freed later by the owning heap. If the exact usable size is too small to +// contain the pointer for the delayed list, then shrink the padding (by decreasing delta) +// so it will later not trigger an overflow error in `mi_free_block`. +void _mi_padding_shrink(const mi_page_t* page, const mi_block_t* block, const size_t min_size) { + size_t bsize; + size_t delta; + bool ok = mi_page_decode_padding(page, block, &delta, &bsize); + mi_assert_internal(ok); + if (!ok || (bsize - delta) >= min_size) return; // usually already enough space + mi_assert_internal(bsize >= min_size); + if (bsize < min_size) return; // should never happen + size_t new_delta = (bsize - min_size); + mi_assert_internal(new_delta < bsize); + mi_padding_t* padding = (mi_padding_t*)((uint8_t*)block + bsize); + mi_track_mem_defined(padding,sizeof(mi_padding_t)); + padding->delta = (uint32_t)new_delta; + mi_track_mem_noaccess(padding,sizeof(mi_padding_t)); +} +#else +static size_t mi_page_usable_size_of(const mi_page_t* page, const mi_block_t* block) { + MI_UNUSED(block); + return mi_page_usable_block_size(page); +} + +void _mi_padding_shrink(const mi_page_t* page, const mi_block_t* block, const size_t min_size) { + MI_UNUSED(page); + MI_UNUSED(block); + MI_UNUSED(min_size); +} +#endif + +#if MI_PADDING && MI_PADDING_CHECK + +static bool mi_verify_padding(const mi_page_t* page, const mi_block_t* block, size_t* size, size_t* wrong) { + size_t bsize; + size_t delta; + bool ok = mi_page_decode_padding(page, block, &delta, &bsize); + *size = *wrong = bsize; + if (!ok) return false; + mi_assert_internal(bsize >= delta); + *size = bsize - delta; + if (!mi_page_is_huge(page)) { + uint8_t* fill = (uint8_t*)block + bsize - delta; + const size_t maxpad = (delta > MI_MAX_ALIGN_SIZE ? MI_MAX_ALIGN_SIZE : delta); // check at most the first N padding bytes + mi_track_mem_defined(fill, maxpad); + for (size_t i = 0; i < maxpad; i++) { + if (fill[i] != MI_DEBUG_PADDING) { + *wrong = bsize - delta + i; + ok = false; + break; + } + } + mi_track_mem_noaccess(fill, maxpad); + } + return ok; +} + +static void mi_check_padding(const mi_page_t* page, const mi_block_t* block) { + size_t size; + size_t wrong; + if (!mi_verify_padding(page,block,&size,&wrong)) { + _mi_error_message(EFAULT, "buffer overflow in heap block %p of size %zu: write after %zu bytes\n", block, size, wrong ); + } +} + +#else + +static void mi_check_padding(const mi_page_t* page, const mi_block_t* block) { + MI_UNUSED(page); + MI_UNUSED(block); +} + +#endif + +// only maintain stats for smaller objects if requested +#if (MI_STAT>0) +static void mi_stat_free(const mi_page_t* page, const mi_block_t* block) { + MI_UNUSED(block); + mi_heap_t* const heap = mi_heap_get_default(); + const size_t bsize = mi_page_usable_block_size(page); + // #if (MI_STAT>1) + // const size_t usize = mi_page_usable_size_of(page, block); + // mi_heap_stat_decrease(heap, malloc_requested, usize); + // #endif + if (bsize <= MI_MEDIUM_OBJ_SIZE_MAX) { + mi_heap_stat_decrease(heap, malloc_normal, bsize); + #if (MI_STAT > 1) + mi_heap_stat_decrease(heap, malloc_bins[_mi_bin(bsize)], 1); + #endif + } + //else if (bsize <= MI_LARGE_OBJ_SIZE_MAX) { + // mi_heap_stat_decrease(heap, malloc_large, bsize); + //} + else { + mi_heap_stat_decrease(heap, malloc_huge, bsize); + } +} +#else +static void mi_stat_free(const mi_page_t* page, const mi_block_t* block) { + MI_UNUSED(page); MI_UNUSED(block); +} +#endif + + +// Remove guard page when building with MI_GUARDED +#if MI_GUARDED +static void mi_block_unguard(mi_page_t* page, mi_block_t* block, void* p) { + MI_UNUSED(p); + mi_assert_internal(mi_block_ptr_is_guarded(block, p)); + mi_assert_internal(mi_page_has_aligned(page)); + mi_assert_internal((uint8_t*)p - (uint8_t*)block >= (ptrdiff_t)sizeof(mi_block_t)); + mi_assert_internal(block->next == MI_BLOCK_TAG_GUARDED); + + const size_t bsize = mi_page_block_size(page); + const size_t psize = _mi_os_page_size(); + mi_assert_internal(bsize > psize); + mi_assert_internal(_mi_page_segment(page)->allow_decommit); + void* gpage = (uint8_t*)block + bsize - psize; + mi_assert_internal(_mi_is_aligned(gpage, psize)); + _mi_os_unprotect(gpage, psize); +} +#endif diff --git a/compat/mimalloc/heap.c b/compat/mimalloc/heap.c new file mode 100644 index 00000000000000..cbfee560b8c959 --- /dev/null +++ b/compat/mimalloc/heap.c @@ -0,0 +1,735 @@ +/*---------------------------------------------------------------------------- +Copyright (c) 2018-2021, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ + +#include "mimalloc.h" +#include "mimalloc/internal.h" +#include "mimalloc/atomic.h" +#include "mimalloc/prim.h" // mi_prim_get_default_heap + +#include // memset, memcpy + +#if defined(_MSC_VER) && (_MSC_VER < 1920) +#pragma warning(disable:4204) // non-constant aggregate initializer +#endif + +/* ----------------------------------------------------------- + Helpers +----------------------------------------------------------- */ + +// return `true` if ok, `false` to break +typedef bool (heap_page_visitor_fun)(mi_heap_t* heap, mi_page_queue_t* pq, mi_page_t* page, void* arg1, void* arg2); + +// Visit all pages in a heap; returns `false` if break was called. +static bool mi_heap_visit_pages(mi_heap_t* heap, heap_page_visitor_fun* fn, void* arg1, void* arg2) +{ + if (heap==NULL || heap->page_count==0) return 0; + + // visit all pages + #if MI_DEBUG>1 + size_t total = heap->page_count; + size_t count = 0; + #endif + + for (size_t i = 0; i <= MI_BIN_FULL; i++) { + mi_page_queue_t* pq = &heap->pages[i]; + mi_page_t* page = pq->first; + while(page != NULL) { + mi_page_t* next = page->next; // save next in case the page gets removed from the queue + mi_assert_internal(mi_page_heap(page) == heap); + #if MI_DEBUG>1 + count++; + #endif + if (!fn(heap, pq, page, arg1, arg2)) return false; + page = next; // and continue + } + } + mi_assert_internal(count == total); + return true; +} + + +#if MI_DEBUG>=2 +static bool mi_heap_page_is_valid(mi_heap_t* heap, mi_page_queue_t* pq, mi_page_t* page, void* arg1, void* arg2) { + MI_UNUSED(arg1); + MI_UNUSED(arg2); + MI_UNUSED(pq); + mi_assert_internal(mi_page_heap(page) == heap); + mi_segment_t* segment = _mi_page_segment(page); + mi_assert_internal(mi_atomic_load_relaxed(&segment->thread_id) == heap->thread_id); + mi_assert_expensive(_mi_page_is_valid(page)); + return true; +} +#endif +#if MI_DEBUG>=3 +static bool mi_heap_is_valid(mi_heap_t* heap) { + mi_assert_internal(heap!=NULL); + mi_heap_visit_pages(heap, &mi_heap_page_is_valid, NULL, NULL); + return true; +} +#endif + + + + +/* ----------------------------------------------------------- + "Collect" pages by migrating `local_free` and `thread_free` + lists and freeing empty pages. This is done when a thread + stops (and in that case abandons pages if there are still + blocks alive) +----------------------------------------------------------- */ + +typedef enum mi_collect_e { + MI_NORMAL, + MI_FORCE, + MI_ABANDON +} mi_collect_t; + + +static bool mi_heap_page_collect(mi_heap_t* heap, mi_page_queue_t* pq, mi_page_t* page, void* arg_collect, void* arg2 ) { + MI_UNUSED(arg2); + MI_UNUSED(heap); + mi_assert_internal(mi_heap_page_is_valid(heap, pq, page, NULL, NULL)); + mi_collect_t collect = *((mi_collect_t*)arg_collect); + _mi_page_free_collect(page, collect >= MI_FORCE); + if (collect == MI_FORCE) { + // note: call before a potential `_mi_page_free` as the segment may be freed if this was the last used page in that segment. + mi_segment_t* segment = _mi_page_segment(page); + _mi_segment_collect(segment, true /* force? */); + } + if (mi_page_all_free(page)) { + // no more used blocks, free the page. + // note: this will free retired pages as well. + _mi_page_free(page, pq, collect >= MI_FORCE); + } + else if (collect == MI_ABANDON) { + // still used blocks but the thread is done; abandon the page + _mi_page_abandon(page, pq); + } + return true; // don't break +} + +static bool mi_heap_page_never_delayed_free(mi_heap_t* heap, mi_page_queue_t* pq, mi_page_t* page, void* arg1, void* arg2) { + MI_UNUSED(arg1); + MI_UNUSED(arg2); + MI_UNUSED(heap); + MI_UNUSED(pq); + _mi_page_use_delayed_free(page, MI_NEVER_DELAYED_FREE, false); + return true; // don't break +} + +static void mi_heap_collect_ex(mi_heap_t* heap, mi_collect_t collect) +{ + if (heap==NULL || !mi_heap_is_initialized(heap)) return; + + const bool force = (collect >= MI_FORCE); + _mi_deferred_free(heap, force); + + // python/cpython#112532: we may be called from a thread that is not the owner of the heap + const bool is_main_thread = (_mi_is_main_thread() && heap->thread_id == _mi_thread_id()); + + // note: never reclaim on collect but leave it to threads that need storage to reclaim + const bool force_main = + #ifdef NDEBUG + collect == MI_FORCE + #else + collect >= MI_FORCE + #endif + && is_main_thread && mi_heap_is_backing(heap) && !heap->no_reclaim; + + if (force_main) { + // the main thread is abandoned (end-of-program), try to reclaim all abandoned segments. + // if all memory is freed by now, all segments should be freed. + // note: this only collects in the current subprocess + _mi_abandoned_reclaim_all(heap, &heap->tld->segments); + } + + // if abandoning, mark all pages to no longer add to delayed_free + if (collect == MI_ABANDON) { + mi_heap_visit_pages(heap, &mi_heap_page_never_delayed_free, NULL, NULL); + } + + // free all current thread delayed blocks. + // (if abandoning, after this there are no more thread-delayed references into the pages.) + _mi_heap_delayed_free_all(heap); + + // collect retired pages + _mi_heap_collect_retired(heap, force); + + // collect all pages owned by this thread + mi_heap_visit_pages(heap, &mi_heap_page_collect, &collect, NULL); + mi_assert_internal( collect != MI_ABANDON || mi_atomic_load_ptr_acquire(mi_block_t,&heap->thread_delayed_free) == NULL ); + + // collect abandoned segments (in particular, purge expired parts of segments in the abandoned segment list) + // note: forced purge can be quite expensive if many threads are created/destroyed so we do not force on abandonment + _mi_abandoned_collect(heap, collect == MI_FORCE /* force? */, &heap->tld->segments); + + // if forced, collect thread data cache on program-exit (or shared library unload) + if (force && is_main_thread && mi_heap_is_backing(heap)) { + _mi_thread_data_collect(); // collect thread data cache + } + + // collect arenas (this is program wide so don't force purges on abandonment of threads) + _mi_arenas_collect(collect == MI_FORCE /* force purge? */); + + // merge statistics + if (collect <= MI_FORCE) { + mi_stats_merge(); + } +} + +void _mi_heap_collect_abandon(mi_heap_t* heap) { + mi_heap_collect_ex(heap, MI_ABANDON); +} + +void mi_heap_collect(mi_heap_t* heap, bool force) mi_attr_noexcept { + mi_heap_collect_ex(heap, (force ? MI_FORCE : MI_NORMAL)); +} + +void mi_collect(bool force) mi_attr_noexcept { + mi_heap_collect(mi_prim_get_default_heap(), force); +} + + +/* ----------------------------------------------------------- + Heap new +----------------------------------------------------------- */ + +mi_heap_t* mi_heap_get_default(void) { + mi_thread_init(); + return mi_prim_get_default_heap(); +} + +static bool mi_heap_is_default(const mi_heap_t* heap) { + return (heap == mi_prim_get_default_heap()); +} + + +mi_heap_t* mi_heap_get_backing(void) { + mi_heap_t* heap = mi_heap_get_default(); + mi_assert_internal(heap!=NULL); + mi_heap_t* bheap = heap->tld->heap_backing; + mi_assert_internal(bheap!=NULL); + mi_assert_internal(bheap->thread_id == _mi_thread_id()); + return bheap; +} + +void _mi_heap_init(mi_heap_t* heap, mi_tld_t* tld, mi_arena_id_t arena_id, bool noreclaim, uint8_t tag) { + _mi_memcpy_aligned(heap, &_mi_heap_empty, sizeof(mi_heap_t)); + heap->tld = tld; + heap->thread_id = _mi_thread_id(); + heap->arena_id = arena_id; + heap->no_reclaim = noreclaim; + heap->tag = tag; + if (heap == tld->heap_backing) { + _mi_random_init(&heap->random); + } + else { + _mi_random_split(&tld->heap_backing->random, &heap->random); + } + heap->cookie = _mi_heap_random_next(heap) | 1; + heap->keys[0] = _mi_heap_random_next(heap); + heap->keys[1] = _mi_heap_random_next(heap); + _mi_heap_guarded_init(heap); + // push on the thread local heaps list + heap->next = heap->tld->heaps; + heap->tld->heaps = heap; +} + +mi_decl_nodiscard mi_heap_t* mi_heap_new_ex(int heap_tag, bool allow_destroy, mi_arena_id_t arena_id) { + mi_heap_t* bheap = mi_heap_get_backing(); + mi_heap_t* heap = mi_heap_malloc_tp(bheap, mi_heap_t); // todo: OS allocate in secure mode? + if (heap == NULL) return NULL; + mi_assert(heap_tag >= 0 && heap_tag < 256); + _mi_heap_init(heap, bheap->tld, arena_id, allow_destroy /* no reclaim? */, (uint8_t)heap_tag /* heap tag */); + return heap; +} + +mi_decl_nodiscard mi_heap_t* mi_heap_new_in_arena(mi_arena_id_t arena_id) { + return mi_heap_new_ex(0 /* default heap tag */, false /* don't allow `mi_heap_destroy` */, arena_id); +} + +mi_decl_nodiscard mi_heap_t* mi_heap_new(void) { + // don't reclaim abandoned memory or otherwise destroy is unsafe + return mi_heap_new_ex(0 /* default heap tag */, true /* no reclaim */, _mi_arena_id_none()); +} + +bool _mi_heap_memid_is_suitable(mi_heap_t* heap, mi_memid_t memid) { + return _mi_arena_memid_is_suitable(memid, heap->arena_id); +} + +uintptr_t _mi_heap_random_next(mi_heap_t* heap) { + return _mi_random_next(&heap->random); +} + +// zero out the page queues +static void mi_heap_reset_pages(mi_heap_t* heap) { + mi_assert_internal(heap != NULL); + mi_assert_internal(mi_heap_is_initialized(heap)); + // TODO: copy full empty heap instead? + memset(&heap->pages_free_direct, 0, sizeof(heap->pages_free_direct)); + _mi_memcpy_aligned(&heap->pages, &_mi_heap_empty.pages, sizeof(heap->pages)); + heap->thread_delayed_free = NULL; + heap->page_count = 0; +} + +// called from `mi_heap_destroy` and `mi_heap_delete` to free the internal heap resources. +static void mi_heap_free(mi_heap_t* heap) { + mi_assert(heap != NULL); + mi_assert_internal(mi_heap_is_initialized(heap)); + if (heap==NULL || !mi_heap_is_initialized(heap)) return; + if (mi_heap_is_backing(heap)) return; // dont free the backing heap + + // reset default + if (mi_heap_is_default(heap)) { + _mi_heap_set_default_direct(heap->tld->heap_backing); + } + + // remove ourselves from the thread local heaps list + // linear search but we expect the number of heaps to be relatively small + mi_heap_t* prev = NULL; + mi_heap_t* curr = heap->tld->heaps; + while (curr != heap && curr != NULL) { + prev = curr; + curr = curr->next; + } + mi_assert_internal(curr == heap); + if (curr == heap) { + if (prev != NULL) { prev->next = heap->next; } + else { heap->tld->heaps = heap->next; } + } + mi_assert_internal(heap->tld->heaps != NULL); + + // and free the used memory + mi_free(heap); +} + +// return a heap on the same thread as `heap` specialized for the specified tag (if it exists) +mi_heap_t* _mi_heap_by_tag(mi_heap_t* heap, uint8_t tag) { + if (heap->tag == tag) { + return heap; + } + for (mi_heap_t *curr = heap->tld->heaps; curr != NULL; curr = curr->next) { + if (curr->tag == tag) { + return curr; + } + } + return NULL; +} + +/* ----------------------------------------------------------- + Heap destroy +----------------------------------------------------------- */ + +static bool _mi_heap_page_destroy(mi_heap_t* heap, mi_page_queue_t* pq, mi_page_t* page, void* arg1, void* arg2) { + MI_UNUSED(arg1); + MI_UNUSED(arg2); + MI_UNUSED(heap); + MI_UNUSED(pq); + + // ensure no more thread_delayed_free will be added + _mi_page_use_delayed_free(page, MI_NEVER_DELAYED_FREE, false); + + // stats + const size_t bsize = mi_page_block_size(page); + if (bsize > MI_MEDIUM_OBJ_SIZE_MAX) { + //if (bsize <= MI_LARGE_OBJ_SIZE_MAX) { + // mi_heap_stat_decrease(heap, malloc_large, bsize); + //} + //else + { + mi_heap_stat_decrease(heap, malloc_huge, bsize); + } + } + #if (MI_STAT>0) + _mi_page_free_collect(page, false); // update used count + const size_t inuse = page->used; + if (bsize <= MI_LARGE_OBJ_SIZE_MAX) { + mi_heap_stat_decrease(heap, malloc_normal, bsize * inuse); + #if (MI_STAT>1) + mi_heap_stat_decrease(heap, malloc_bins[_mi_bin(bsize)], inuse); + #endif + } + // mi_heap_stat_decrease(heap, malloc_requested, bsize * inuse); // todo: off for aligned blocks... + #endif + + /// pretend it is all free now + mi_assert_internal(mi_page_thread_free(page) == NULL); + page->used = 0; + + // and free the page + // mi_page_free(page,false); + page->next = NULL; + page->prev = NULL; + _mi_segment_page_free(page,false /* no force? */, &heap->tld->segments); + + return true; // keep going +} + +void _mi_heap_destroy_pages(mi_heap_t* heap) { + mi_heap_visit_pages(heap, &_mi_heap_page_destroy, NULL, NULL); + mi_heap_reset_pages(heap); +} + +#if MI_TRACK_HEAP_DESTROY +static bool mi_cdecl mi_heap_track_block_free(const mi_heap_t* heap, const mi_heap_area_t* area, void* block, size_t block_size, void* arg) { + MI_UNUSED(heap); MI_UNUSED(area); MI_UNUSED(arg); MI_UNUSED(block_size); + mi_track_free_size(block,mi_usable_size(block)); + return true; +} +#endif + +void mi_heap_destroy(mi_heap_t* heap) { + mi_assert(heap != NULL); + mi_assert(mi_heap_is_initialized(heap)); + mi_assert(heap->no_reclaim); + mi_assert_expensive(mi_heap_is_valid(heap)); + if (heap==NULL || !mi_heap_is_initialized(heap)) return; + #if MI_GUARDED + // _mi_warning_message("'mi_heap_destroy' called but MI_GUARDED is enabled -- using `mi_heap_delete` instead (heap at %p)\n", heap); + mi_heap_delete(heap); + return; + #else + if (!heap->no_reclaim) { + _mi_warning_message("'mi_heap_destroy' called but ignored as the heap was not created with 'allow_destroy' (heap at %p)\n", heap); + // don't free in case it may contain reclaimed pages + mi_heap_delete(heap); + } + else { + // track all blocks as freed + #if MI_TRACK_HEAP_DESTROY + mi_heap_visit_blocks(heap, true, mi_heap_track_block_free, NULL); + #endif + // free all pages + _mi_heap_destroy_pages(heap); + mi_heap_free(heap); + } + #endif +} + +// forcefully destroy all heaps in the current thread +void _mi_heap_unsafe_destroy_all(mi_heap_t* heap) { + mi_assert_internal(heap != NULL); + if (heap == NULL) return; + mi_heap_t* curr = heap->tld->heaps; + while (curr != NULL) { + mi_heap_t* next = curr->next; + if (curr->no_reclaim) { + mi_heap_destroy(curr); + } + else { + _mi_heap_destroy_pages(curr); + } + curr = next; + } +} + +/* ----------------------------------------------------------- + Safe Heap delete +----------------------------------------------------------- */ + +// Transfer the pages from one heap to the other +static void mi_heap_absorb(mi_heap_t* heap, mi_heap_t* from) { + mi_assert_internal(heap!=NULL); + if (from==NULL || from->page_count == 0) return; + + // reduce the size of the delayed frees + _mi_heap_delayed_free_partial(from); + + // transfer all pages by appending the queues; this will set a new heap field + // so threads may do delayed frees in either heap for a while. + // note: appending waits for each page to not be in the `MI_DELAYED_FREEING` state + // so after this only the new heap will get delayed frees + for (size_t i = 0; i <= MI_BIN_FULL; i++) { + mi_page_queue_t* pq = &heap->pages[i]; + mi_page_queue_t* append = &from->pages[i]; + size_t pcount = _mi_page_queue_append(heap, pq, append); + heap->page_count += pcount; + from->page_count -= pcount; + } + mi_assert_internal(from->page_count == 0); + + // and do outstanding delayed frees in the `from` heap + // note: be careful here as the `heap` field in all those pages no longer point to `from`, + // turns out to be ok as `_mi_heap_delayed_free` only visits the list and calls a + // the regular `_mi_free_delayed_block` which is safe. + _mi_heap_delayed_free_all(from); + #if !defined(_MSC_VER) || (_MSC_VER > 1900) // somehow the following line gives an error in VS2015, issue #353 + mi_assert_internal(mi_atomic_load_ptr_relaxed(mi_block_t,&from->thread_delayed_free) == NULL); + #endif + + // and reset the `from` heap + mi_heap_reset_pages(from); +} + +// are two heaps compatible with respect to heap-tag, exclusive arena etc. +static bool mi_heaps_are_compatible(mi_heap_t* heap1, mi_heap_t* heap2) { + return (heap1->tag == heap2->tag && // store same kind of objects + heap1->arena_id == heap2->arena_id); // same arena preference +} + +// Safe delete a heap without freeing any still allocated blocks in that heap. +void mi_heap_delete(mi_heap_t* heap) +{ + mi_assert(heap != NULL); + mi_assert(mi_heap_is_initialized(heap)); + mi_assert_expensive(mi_heap_is_valid(heap)); + if (heap==NULL || !mi_heap_is_initialized(heap)) return; + + mi_heap_t* bheap = heap->tld->heap_backing; + if (bheap != heap && mi_heaps_are_compatible(bheap,heap)) { + // transfer still used pages to the backing heap + mi_heap_absorb(bheap, heap); + } + else { + // the backing heap abandons its pages + _mi_heap_collect_abandon(heap); + } + mi_assert_internal(heap->page_count==0); + mi_heap_free(heap); +} + +mi_heap_t* mi_heap_set_default(mi_heap_t* heap) { + mi_assert(heap != NULL); + mi_assert(mi_heap_is_initialized(heap)); + if (heap==NULL || !mi_heap_is_initialized(heap)) return NULL; + mi_assert_expensive(mi_heap_is_valid(heap)); + mi_heap_t* old = mi_prim_get_default_heap(); + _mi_heap_set_default_direct(heap); + return old; +} + + + + +/* ----------------------------------------------------------- + Analysis +----------------------------------------------------------- */ + +// static since it is not thread safe to access heaps from other threads. +static mi_heap_t* mi_heap_of_block(const void* p) { + if (p == NULL) return NULL; + mi_segment_t* segment = _mi_ptr_segment(p); + bool valid = (_mi_ptr_cookie(segment) == segment->cookie); + mi_assert_internal(valid); + if mi_unlikely(!valid) return NULL; + return mi_page_heap(_mi_segment_page_of(segment,p)); +} + +bool mi_heap_contains_block(mi_heap_t* heap, const void* p) { + mi_assert(heap != NULL); + if (heap==NULL || !mi_heap_is_initialized(heap)) return false; + return (heap == mi_heap_of_block(p)); +} + + +static bool mi_heap_page_check_owned(mi_heap_t* heap, mi_page_queue_t* pq, mi_page_t* page, void* p, void* vfound) { + MI_UNUSED(heap); + MI_UNUSED(pq); + bool* found = (bool*)vfound; + void* start = mi_page_start(page); + void* end = (uint8_t*)start + (page->capacity * mi_page_block_size(page)); + *found = (p >= start && p < end); + return (!*found); // continue if not found +} + +bool mi_heap_check_owned(mi_heap_t* heap, const void* p) { + mi_assert(heap != NULL); + if (heap==NULL || !mi_heap_is_initialized(heap)) return false; + if (((uintptr_t)p & (MI_INTPTR_SIZE - 1)) != 0) return false; // only aligned pointers + bool found = false; + mi_heap_visit_pages(heap, &mi_heap_page_check_owned, (void*)p, &found); + return found; +} + +bool mi_check_owned(const void* p) { + return mi_heap_check_owned(mi_prim_get_default_heap(), p); +} + +/* ----------------------------------------------------------- + Visit all heap blocks and areas + Todo: enable visiting abandoned pages, and + enable visiting all blocks of all heaps across threads +----------------------------------------------------------- */ + +void _mi_heap_area_init(mi_heap_area_t* area, mi_page_t* page) { + const size_t bsize = mi_page_block_size(page); + const size_t ubsize = mi_page_usable_block_size(page); + area->reserved = page->reserved * bsize; + area->committed = page->capacity * bsize; + area->blocks = mi_page_start(page); + area->used = page->used; // number of blocks in use (#553) + area->block_size = ubsize; + area->full_block_size = bsize; + area->heap_tag = page->heap_tag; +} + + +static void mi_get_fast_divisor(size_t divisor, uint64_t* magic, size_t* shift) { + mi_assert_internal(divisor > 0 && divisor <= UINT32_MAX); + *shift = MI_SIZE_BITS - mi_clz(divisor - 1); + *magic = ((((uint64_t)1 << 32) * (((uint64_t)1 << *shift) - divisor)) / divisor + 1); +} + +static size_t mi_fast_divide(size_t n, uint64_t magic, size_t shift) { + mi_assert_internal(n <= UINT32_MAX); + const uint64_t hi = ((uint64_t)n * magic) >> 32; + return (size_t)((hi + n) >> shift); +} + +bool _mi_heap_area_visit_blocks(const mi_heap_area_t* area, mi_page_t* page, mi_block_visit_fun* visitor, void* arg) { + mi_assert(area != NULL); + if (area==NULL) return true; + mi_assert(page != NULL); + if (page == NULL) return true; + + _mi_page_free_collect(page,true); // collect both thread_delayed and local_free + mi_assert_internal(page->local_free == NULL); + if (page->used == 0) return true; + + size_t psize; + uint8_t* const pstart = _mi_segment_page_start(_mi_page_segment(page), page, &psize); + mi_heap_t* const heap = mi_page_heap(page); + const size_t bsize = mi_page_block_size(page); + const size_t ubsize = mi_page_usable_block_size(page); // without padding + + // optimize page with one block + if (page->capacity == 1) { + mi_assert_internal(page->used == 1 && page->free == NULL); + return visitor(mi_page_heap(page), area, pstart, ubsize, arg); + } + mi_assert(bsize <= UINT32_MAX); + + // optimize full pages + if (page->used == page->capacity) { + uint8_t* block = pstart; + for (size_t i = 0; i < page->capacity; i++) { + if (!visitor(heap, area, block, ubsize, arg)) return false; + block += bsize; + } + return true; + } + + // create a bitmap of free blocks. + #define MI_MAX_BLOCKS (MI_SMALL_PAGE_SIZE / sizeof(void*)) + uintptr_t free_map[MI_MAX_BLOCKS / MI_INTPTR_BITS]; + const uintptr_t bmapsize = _mi_divide_up(page->capacity, MI_INTPTR_BITS); + memset(free_map, 0, bmapsize * sizeof(intptr_t)); + if (page->capacity % MI_INTPTR_BITS != 0) { + // mark left-over bits at the end as free + size_t shift = (page->capacity % MI_INTPTR_BITS); + uintptr_t mask = (UINTPTR_MAX << shift); + free_map[bmapsize - 1] = mask; + } + + // fast repeated division by the block size + uint64_t magic; + size_t shift; + mi_get_fast_divisor(bsize, &magic, &shift); + + #if MI_DEBUG>1 + size_t free_count = 0; + #endif + for (mi_block_t* block = page->free; block != NULL; block = mi_block_next(page, block)) { + #if MI_DEBUG>1 + free_count++; + #endif + mi_assert_internal((uint8_t*)block >= pstart && (uint8_t*)block < (pstart + psize)); + size_t offset = (uint8_t*)block - pstart; + mi_assert_internal(offset % bsize == 0); + mi_assert_internal(offset <= UINT32_MAX); + size_t blockidx = mi_fast_divide(offset, magic, shift); + mi_assert_internal(blockidx == offset / bsize); + mi_assert_internal(blockidx < MI_MAX_BLOCKS); + size_t bitidx = (blockidx / MI_INTPTR_BITS); + size_t bit = blockidx - (bitidx * MI_INTPTR_BITS); + free_map[bitidx] |= ((uintptr_t)1 << bit); + } + mi_assert_internal(page->capacity == (free_count + page->used)); + + // walk through all blocks skipping the free ones + #if MI_DEBUG>1 + size_t used_count = 0; + #endif + uint8_t* block = pstart; + for (size_t i = 0; i < bmapsize; i++) { + if (free_map[i] == 0) { + // every block is in use + for (size_t j = 0; j < MI_INTPTR_BITS; j++) { + #if MI_DEBUG>1 + used_count++; + #endif + if (!visitor(heap, area, block, ubsize, arg)) return false; + block += bsize; + } + } + else { + // visit the used blocks in the mask + uintptr_t m = ~free_map[i]; + while (m != 0) { + #if MI_DEBUG>1 + used_count++; + #endif + size_t bitidx = mi_ctz(m); + if (!visitor(heap, area, block + (bitidx * bsize), ubsize, arg)) return false; + m &= m - 1; // clear least significant bit + } + block += bsize * MI_INTPTR_BITS; + } + } + mi_assert_internal(page->used == used_count); + return true; +} + + + +// Separate struct to keep `mi_page_t` out of the public interface +typedef struct mi_heap_area_ex_s { + mi_heap_area_t area; + mi_page_t* page; +} mi_heap_area_ex_t; + +typedef bool (mi_heap_area_visit_fun)(const mi_heap_t* heap, const mi_heap_area_ex_t* area, void* arg); + +static bool mi_heap_visit_areas_page(mi_heap_t* heap, mi_page_queue_t* pq, mi_page_t* page, void* vfun, void* arg) { + MI_UNUSED(heap); + MI_UNUSED(pq); + mi_heap_area_visit_fun* fun = (mi_heap_area_visit_fun*)vfun; + mi_heap_area_ex_t xarea; + xarea.page = page; + _mi_heap_area_init(&xarea.area, page); + return fun(heap, &xarea, arg); +} + +// Visit all heap pages as areas +static bool mi_heap_visit_areas(const mi_heap_t* heap, mi_heap_area_visit_fun* visitor, void* arg) { + if (visitor == NULL) return false; + return mi_heap_visit_pages((mi_heap_t*)heap, &mi_heap_visit_areas_page, (void*)(visitor), arg); // note: function pointer to void* :-{ +} + +// Just to pass arguments +typedef struct mi_visit_blocks_args_s { + bool visit_blocks; + mi_block_visit_fun* visitor; + void* arg; +} mi_visit_blocks_args_t; + +static bool mi_heap_area_visitor(const mi_heap_t* heap, const mi_heap_area_ex_t* xarea, void* arg) { + mi_visit_blocks_args_t* args = (mi_visit_blocks_args_t*)arg; + if (!args->visitor(heap, &xarea->area, NULL, xarea->area.block_size, args->arg)) return false; + if (args->visit_blocks) { + return _mi_heap_area_visit_blocks(&xarea->area, xarea->page, args->visitor, args->arg); + } + else { + return true; + } +} + +// Visit all blocks in a heap +bool mi_heap_visit_blocks(const mi_heap_t* heap, bool visit_blocks, mi_block_visit_fun* visitor, void* arg) { + mi_visit_blocks_args_t args = { visit_blocks, visitor, arg }; + return mi_heap_visit_areas(heap, &mi_heap_area_visitor, &args); +} diff --git a/compat/mimalloc/init.c b/compat/mimalloc/init.c new file mode 100644 index 00000000000000..ddded152a33240 --- /dev/null +++ b/compat/mimalloc/init.c @@ -0,0 +1,720 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2022, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ +#include "mimalloc.h" +#include "mimalloc/internal.h" +#include "mimalloc/prim.h" + +#include // memcpy, memset +#include // atexit + + +// Empty page used to initialize the small free pages array +const mi_page_t _mi_page_empty = { + 0, + false, false, false, false, + 0, // capacity + 0, // reserved capacity + { 0 }, // flags + false, // is_zero + 0, // retire_expire + NULL, // free + NULL, // local_free + 0, // used + 0, // block size shift + 0, // heap tag + 0, // block_size + NULL, // page_start + #if (MI_PADDING || MI_ENCODE_FREELIST) + { 0, 0 }, + #endif + MI_ATOMIC_VAR_INIT(0), // xthread_free + MI_ATOMIC_VAR_INIT(0), // xheap + NULL, NULL + , { 0 } // padding +}; + +#define MI_PAGE_EMPTY() ((mi_page_t*)&_mi_page_empty) + +#if (MI_SMALL_WSIZE_MAX==128) +#if (MI_PADDING>0) && (MI_INTPTR_SIZE >= 8) +#define MI_SMALL_PAGES_EMPTY { MI_INIT128(MI_PAGE_EMPTY), MI_PAGE_EMPTY(), MI_PAGE_EMPTY() } +#elif (MI_PADDING>0) +#define MI_SMALL_PAGES_EMPTY { MI_INIT128(MI_PAGE_EMPTY), MI_PAGE_EMPTY(), MI_PAGE_EMPTY(), MI_PAGE_EMPTY() } +#else +#define MI_SMALL_PAGES_EMPTY { MI_INIT128(MI_PAGE_EMPTY), MI_PAGE_EMPTY() } +#endif +#else +#error "define right initialization sizes corresponding to MI_SMALL_WSIZE_MAX" +#endif + +// Empty page queues for every bin +#define QNULL(sz) { NULL, NULL, (sz)*sizeof(uintptr_t) } +#define MI_PAGE_QUEUES_EMPTY \ + { QNULL(1), \ + QNULL( 1), QNULL( 2), QNULL( 3), QNULL( 4), QNULL( 5), QNULL( 6), QNULL( 7), QNULL( 8), /* 8 */ \ + QNULL( 10), QNULL( 12), QNULL( 14), QNULL( 16), QNULL( 20), QNULL( 24), QNULL( 28), QNULL( 32), /* 16 */ \ + QNULL( 40), QNULL( 48), QNULL( 56), QNULL( 64), QNULL( 80), QNULL( 96), QNULL( 112), QNULL( 128), /* 24 */ \ + QNULL( 160), QNULL( 192), QNULL( 224), QNULL( 256), QNULL( 320), QNULL( 384), QNULL( 448), QNULL( 512), /* 32 */ \ + QNULL( 640), QNULL( 768), QNULL( 896), QNULL( 1024), QNULL( 1280), QNULL( 1536), QNULL( 1792), QNULL( 2048), /* 40 */ \ + QNULL( 2560), QNULL( 3072), QNULL( 3584), QNULL( 4096), QNULL( 5120), QNULL( 6144), QNULL( 7168), QNULL( 8192), /* 48 */ \ + QNULL( 10240), QNULL( 12288), QNULL( 14336), QNULL( 16384), QNULL( 20480), QNULL( 24576), QNULL( 28672), QNULL( 32768), /* 56 */ \ + QNULL( 40960), QNULL( 49152), QNULL( 57344), QNULL( 65536), QNULL( 81920), QNULL( 98304), QNULL(114688), QNULL(131072), /* 64 */ \ + QNULL(163840), QNULL(196608), QNULL(229376), QNULL(262144), QNULL(327680), QNULL(393216), QNULL(458752), QNULL(524288), /* 72 */ \ + QNULL(MI_MEDIUM_OBJ_WSIZE_MAX + 1 /* 655360, Huge queue */), \ + QNULL(MI_MEDIUM_OBJ_WSIZE_MAX + 2) /* Full queue */ } + +#define MI_STAT_COUNT_NULL() {0,0,0} + +// Empty statistics +#define MI_STATS_NULL \ + MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), \ + MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), \ + MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), MI_STAT_COUNT_NULL(), \ + { 0 }, { 0 }, { 0 }, { 0 }, \ + { 0 }, { 0 }, { 0 }, { 0 }, \ + \ + { 0 }, { 0 }, { 0 }, { 0 }, { 0 }, \ + MI_INIT4(MI_STAT_COUNT_NULL), \ + { 0 }, { 0 }, { 0 }, { 0 }, \ + \ + { MI_INIT4(MI_STAT_COUNT_NULL) }, \ + { { 0 }, { 0 }, { 0 }, { 0 } }, \ + \ + { MI_INIT74(MI_STAT_COUNT_NULL) }, \ + { MI_INIT74(MI_STAT_COUNT_NULL) } + + +// Empty slice span queues for every bin +#define SQNULL(sz) { NULL, NULL, sz } +#define MI_SEGMENT_SPAN_QUEUES_EMPTY \ + { SQNULL(1), \ + SQNULL( 1), SQNULL( 2), SQNULL( 3), SQNULL( 4), SQNULL( 5), SQNULL( 6), SQNULL( 7), SQNULL( 10), /* 8 */ \ + SQNULL( 12), SQNULL( 14), SQNULL( 16), SQNULL( 20), SQNULL( 24), SQNULL( 28), SQNULL( 32), SQNULL( 40), /* 16 */ \ + SQNULL( 48), SQNULL( 56), SQNULL( 64), SQNULL( 80), SQNULL( 96), SQNULL( 112), SQNULL( 128), SQNULL( 160), /* 24 */ \ + SQNULL( 192), SQNULL( 224), SQNULL( 256), SQNULL( 320), SQNULL( 384), SQNULL( 448), SQNULL( 512), SQNULL( 640), /* 32 */ \ + SQNULL( 768), SQNULL( 896), SQNULL( 1024) /* 35 */ } + + +// -------------------------------------------------------- +// Statically allocate an empty heap as the initial +// thread local value for the default heap, +// and statically allocate the backing heap for the main +// thread so it can function without doing any allocation +// itself (as accessing a thread local for the first time +// may lead to allocation itself on some platforms) +// -------------------------------------------------------- + +mi_decl_cache_align const mi_heap_t _mi_heap_empty = { + NULL, + MI_ATOMIC_VAR_INIT(NULL), + 0, // tid + 0, // cookie + 0, // arena id + { 0, 0 }, // keys + { {0}, {0}, 0, true }, // random + 0, // page count + MI_BIN_FULL, 0, // page retired min/max + 0, 0, // generic count + NULL, // next + false, // can reclaim + 0, // tag + #if MI_GUARDED + 0, 0, 0, 0, 1, // count is 1 so we never write to it (see `internal.h:mi_heap_malloc_use_guarded`) + #endif + MI_SMALL_PAGES_EMPTY, + MI_PAGE_QUEUES_EMPTY +}; + +static mi_decl_cache_align mi_subproc_t mi_subproc_default; + +#define tld_empty_stats ((mi_stats_t*)((uint8_t*)&tld_empty + offsetof(mi_tld_t,stats))) + +mi_decl_cache_align static const mi_tld_t tld_empty = { + 0, + false, + NULL, NULL, + { MI_SEGMENT_SPAN_QUEUES_EMPTY, 0, 0, 0, 0, 0, &mi_subproc_default, tld_empty_stats }, // segments + { MI_STAT_VERSION, MI_STATS_NULL } // stats +}; + +mi_threadid_t _mi_thread_id(void) mi_attr_noexcept { + return _mi_prim_thread_id(); +} + +// the thread-local default heap for allocation +mi_decl_thread mi_heap_t* _mi_heap_default = (mi_heap_t*)&_mi_heap_empty; + +extern mi_decl_hidden mi_heap_t _mi_heap_main; + +static mi_decl_cache_align mi_tld_t tld_main = { + 0, false, + &_mi_heap_main, & _mi_heap_main, + { MI_SEGMENT_SPAN_QUEUES_EMPTY, 0, 0, 0, 0, 0, &mi_subproc_default, &tld_main.stats }, // segments + { MI_STAT_VERSION, MI_STATS_NULL } // stats +}; + +mi_decl_cache_align mi_heap_t _mi_heap_main = { + &tld_main, + MI_ATOMIC_VAR_INIT(NULL), + 0, // thread id + 0, // initial cookie + 0, // arena id + { 0, 0 }, // the key of the main heap can be fixed (unlike page keys that need to be secure!) + { {0x846ca68b}, {0}, 0, true }, // random + 0, // page count + MI_BIN_FULL, 0, // page retired min/max + 0, 0, // generic count + NULL, // next heap + false, // can reclaim + 0, // tag + #if MI_GUARDED + 0, 0, 0, 0, 0, + #endif + MI_SMALL_PAGES_EMPTY, + MI_PAGE_QUEUES_EMPTY +}; + +bool _mi_process_is_initialized = false; // set to `true` in `mi_process_init`. + +mi_stats_t _mi_stats_main = { MI_STAT_VERSION, MI_STATS_NULL }; + +#if MI_GUARDED +mi_decl_export void mi_heap_guarded_set_sample_rate(mi_heap_t* heap, size_t sample_rate, size_t seed) { + heap->guarded_sample_seed = seed; + if (heap->guarded_sample_seed == 0) { + heap->guarded_sample_seed = _mi_heap_random_next(heap); + } + heap->guarded_sample_rate = sample_rate; + if (heap->guarded_sample_rate >= 1) { + heap->guarded_sample_seed = heap->guarded_sample_seed % heap->guarded_sample_rate; + } + heap->guarded_sample_count = heap->guarded_sample_seed; // count down samples +} + +mi_decl_export void mi_heap_guarded_set_size_bound(mi_heap_t* heap, size_t min, size_t max) { + heap->guarded_size_min = min; + heap->guarded_size_max = (min > max ? min : max); +} + +void _mi_heap_guarded_init(mi_heap_t* heap) { + mi_heap_guarded_set_sample_rate(heap, + (size_t)mi_option_get_clamp(mi_option_guarded_sample_rate, 0, LONG_MAX), + (size_t)mi_option_get(mi_option_guarded_sample_seed)); + mi_heap_guarded_set_size_bound(heap, + (size_t)mi_option_get_clamp(mi_option_guarded_min, 0, LONG_MAX), + (size_t)mi_option_get_clamp(mi_option_guarded_max, 0, LONG_MAX) ); +} +#else +mi_decl_export void mi_heap_guarded_set_sample_rate(mi_heap_t* heap, size_t sample_rate, size_t seed) { + MI_UNUSED(heap); MI_UNUSED(sample_rate); MI_UNUSED(seed); +} + +mi_decl_export void mi_heap_guarded_set_size_bound(mi_heap_t* heap, size_t min, size_t max) { + MI_UNUSED(heap); MI_UNUSED(min); MI_UNUSED(max); +} +void _mi_heap_guarded_init(mi_heap_t* heap) { + MI_UNUSED(heap); +} +#endif + + +static void mi_heap_main_init(void) { + if (_mi_heap_main.cookie == 0) { + _mi_heap_main.thread_id = _mi_thread_id(); + _mi_heap_main.cookie = 1; + #if defined(_WIN32) && !defined(MI_SHARED_LIB) + _mi_random_init_weak(&_mi_heap_main.random); // prevent allocation failure during bcrypt dll initialization with static linking + #else + _mi_random_init(&_mi_heap_main.random); + #endif + _mi_heap_main.cookie = _mi_heap_random_next(&_mi_heap_main); + _mi_heap_main.keys[0] = _mi_heap_random_next(&_mi_heap_main); + _mi_heap_main.keys[1] = _mi_heap_random_next(&_mi_heap_main); + mi_lock_init(&mi_subproc_default.abandoned_os_lock); + mi_lock_init(&mi_subproc_default.abandoned_os_visit_lock); + _mi_heap_guarded_init(&_mi_heap_main); + } +} + +mi_heap_t* _mi_heap_main_get(void) { + mi_heap_main_init(); + return &_mi_heap_main; +} + + +/* ----------------------------------------------------------- + Sub process +----------------------------------------------------------- */ + +mi_subproc_id_t mi_subproc_main(void) { + return NULL; +} + +mi_subproc_id_t mi_subproc_new(void) { + mi_memid_t memid = _mi_memid_none(); + mi_subproc_t* subproc = (mi_subproc_t*)_mi_arena_meta_zalloc(sizeof(mi_subproc_t), &memid); + if (subproc == NULL) return NULL; + subproc->memid = memid; + subproc->abandoned_os_list = NULL; + mi_lock_init(&subproc->abandoned_os_lock); + mi_lock_init(&subproc->abandoned_os_visit_lock); + return subproc; +} + +mi_subproc_t* _mi_subproc_from_id(mi_subproc_id_t subproc_id) { + return (subproc_id == NULL ? &mi_subproc_default : (mi_subproc_t*)subproc_id); +} + +void mi_subproc_delete(mi_subproc_id_t subproc_id) { + if (subproc_id == NULL) return; + mi_subproc_t* subproc = _mi_subproc_from_id(subproc_id); + // check if there are no abandoned segments still.. + bool safe_to_delete = false; + mi_lock(&subproc->abandoned_os_lock) { + if (subproc->abandoned_os_list == NULL) { + safe_to_delete = true; + } + } + if (!safe_to_delete) return; + // safe to release + // todo: should we refcount subprocesses? + mi_lock_done(&subproc->abandoned_os_lock); + mi_lock_done(&subproc->abandoned_os_visit_lock); + _mi_arena_meta_free(subproc, subproc->memid, sizeof(mi_subproc_t)); +} + +void mi_subproc_add_current_thread(mi_subproc_id_t subproc_id) { + mi_heap_t* heap = mi_heap_get_default(); + if (heap == NULL) return; + mi_assert(heap->tld->segments.subproc == &mi_subproc_default); + if (heap->tld->segments.subproc != &mi_subproc_default) return; + heap->tld->segments.subproc = _mi_subproc_from_id(subproc_id); +} + + + +/* ----------------------------------------------------------- + Initialization and freeing of the thread local heaps +----------------------------------------------------------- */ + +// note: in x64 in release build `sizeof(mi_thread_data_t)` is under 4KiB (= OS page size). +typedef struct mi_thread_data_s { + mi_heap_t heap; // must come first due to cast in `_mi_heap_done` + mi_tld_t tld; + mi_memid_t memid; // must come last due to zero'ing +} mi_thread_data_t; + + +// Thread meta-data is allocated directly from the OS. For +// some programs that do not use thread pools and allocate and +// destroy many OS threads, this may causes too much overhead +// per thread so we maintain a small cache of recently freed metadata. + +#define TD_CACHE_SIZE (32) +static _Atomic(mi_thread_data_t*) td_cache[TD_CACHE_SIZE]; + +static mi_thread_data_t* mi_thread_data_zalloc(void) { + // try to find thread metadata in the cache + bool is_zero = false; + mi_thread_data_t* td = NULL; + for (int i = 0; i < TD_CACHE_SIZE; i++) { + td = mi_atomic_load_ptr_relaxed(mi_thread_data_t, &td_cache[i]); + if (td != NULL) { + // found cached allocation, try use it + td = mi_atomic_exchange_ptr_acq_rel(mi_thread_data_t, &td_cache[i], NULL); + if (td != NULL) { + break; + } + } + } + + // if that fails, allocate as meta data + if (td == NULL) { + mi_memid_t memid; + td = (mi_thread_data_t*)_mi_os_alloc(sizeof(mi_thread_data_t), &memid); + if (td == NULL) { + // if this fails, try once more. (issue #257) + td = (mi_thread_data_t*)_mi_os_alloc(sizeof(mi_thread_data_t), &memid); + if (td == NULL) { + // really out of memory + _mi_error_message(ENOMEM, "unable to allocate thread local heap metadata (%zu bytes)\n", sizeof(mi_thread_data_t)); + } + } + if (td != NULL) { + td->memid = memid; + is_zero = memid.initially_zero; + } + } + + if (td != NULL && !is_zero) { + _mi_memzero_aligned(td, offsetof(mi_thread_data_t,memid)); + } + return td; +} + +static void mi_thread_data_free( mi_thread_data_t* tdfree ) { + // try to add the thread metadata to the cache + for (int i = 0; i < TD_CACHE_SIZE; i++) { + mi_thread_data_t* td = mi_atomic_load_ptr_relaxed(mi_thread_data_t, &td_cache[i]); + if (td == NULL) { + mi_thread_data_t* expected = NULL; + if (mi_atomic_cas_ptr_weak_acq_rel(mi_thread_data_t, &td_cache[i], &expected, tdfree)) { + return; + } + } + } + // if that fails, just free it directly + _mi_os_free(tdfree, sizeof(mi_thread_data_t), tdfree->memid); +} + +void _mi_thread_data_collect(void) { + // free all thread metadata from the cache + for (int i = 0; i < TD_CACHE_SIZE; i++) { + mi_thread_data_t* td = mi_atomic_load_ptr_relaxed(mi_thread_data_t, &td_cache[i]); + if (td != NULL) { + td = mi_atomic_exchange_ptr_acq_rel(mi_thread_data_t, &td_cache[i], NULL); + if (td != NULL) { + _mi_os_free(td, sizeof(mi_thread_data_t), td->memid); + } + } + } +} + +// Initialize the thread local default heap, called from `mi_thread_init` +static bool _mi_thread_heap_init(void) { + if (mi_heap_is_initialized(mi_prim_get_default_heap())) return true; + if (_mi_is_main_thread()) { + // mi_assert_internal(_mi_heap_main.thread_id != 0); // can happen on freeBSD where alloc is called before any initialization + // the main heap is statically allocated + mi_heap_main_init(); + _mi_heap_set_default_direct(&_mi_heap_main); + //mi_assert_internal(_mi_heap_default->tld->heap_backing == mi_prim_get_default_heap()); + } + else { + // use `_mi_os_alloc` to allocate directly from the OS + mi_thread_data_t* td = mi_thread_data_zalloc(); + if (td == NULL) return false; + + mi_tld_t* tld = &td->tld; + mi_heap_t* heap = &td->heap; + _mi_tld_init(tld, heap); // must be before `_mi_heap_init` + _mi_heap_init(heap, tld, _mi_arena_id_none(), false /* can reclaim */, 0 /* default tag */); + _mi_heap_set_default_direct(heap); + } + return false; +} + +// initialize thread local data +void _mi_tld_init(mi_tld_t* tld, mi_heap_t* bheap) { + _mi_memcpy_aligned(tld, &tld_empty, sizeof(mi_tld_t)); + tld->heap_backing = bheap; + tld->heaps = NULL; + tld->segments.subproc = &mi_subproc_default; + tld->segments.stats = &tld->stats; +} + +// Free the thread local default heap (called from `mi_thread_done`) +static bool _mi_thread_heap_done(mi_heap_t* heap) { + if (!mi_heap_is_initialized(heap)) return true; + + // reset default heap + _mi_heap_set_default_direct(_mi_is_main_thread() ? &_mi_heap_main : (mi_heap_t*)&_mi_heap_empty); + + // switch to backing heap + heap = heap->tld->heap_backing; + if (!mi_heap_is_initialized(heap)) return false; + + // delete all non-backing heaps in this thread + mi_heap_t* curr = heap->tld->heaps; + while (curr != NULL) { + mi_heap_t* next = curr->next; // save `next` as `curr` will be freed + if (curr != heap) { + mi_assert_internal(!mi_heap_is_backing(curr)); + mi_heap_delete(curr); + } + curr = next; + } + mi_assert_internal(heap->tld->heaps == heap && heap->next == NULL); + mi_assert_internal(mi_heap_is_backing(heap)); + + // collect if not the main thread + if (heap != &_mi_heap_main) { + _mi_heap_collect_abandon(heap); + } + + // merge stats + _mi_stats_done(&heap->tld->stats); + + // free if not the main thread + if (heap != &_mi_heap_main) { + // the following assertion does not always hold for huge segments as those are always treated + // as abondened: one may allocate it in one thread, but deallocate in another in which case + // the count can be too large or negative. todo: perhaps not count huge segments? see issue #363 + // mi_assert_internal(heap->tld->segments.count == 0 || heap->thread_id != _mi_thread_id()); + mi_thread_data_free((mi_thread_data_t*)heap); + } + else { + #if 0 + // never free the main thread even in debug mode; if a dll is linked statically with mimalloc, + // there may still be delete/free calls after the mi_fls_done is called. Issue #207 + _mi_heap_destroy_pages(heap); + mi_assert_internal(heap->tld->heap_backing == &_mi_heap_main); + #endif + } + return false; +} + + + +// -------------------------------------------------------- +// Try to run `mi_thread_done()` automatically so any memory +// owned by the thread but not yet released can be abandoned +// and re-owned by another thread. +// +// 1. windows dynamic library: +// call from DllMain on DLL_THREAD_DETACH +// 2. windows static library: +// use `FlsAlloc` to call a destructor when the thread is done +// 3. unix, pthreads: +// use a pthread key to call a destructor when a pthread is done +// +// In the last two cases we also need to call `mi_process_init` +// to set up the thread local keys. +// -------------------------------------------------------- + +// Set up handlers so `mi_thread_done` is called automatically +static void mi_process_setup_auto_thread_done(void) { + static bool tls_initialized = false; // fine if it races + if (tls_initialized) return; + tls_initialized = true; + _mi_prim_thread_init_auto_done(); + _mi_heap_set_default_direct(&_mi_heap_main); +} + + +bool _mi_is_main_thread(void) { + return (_mi_heap_main.thread_id==0 || _mi_heap_main.thread_id == _mi_thread_id()); +} + +static _Atomic(size_t) thread_count = MI_ATOMIC_VAR_INIT(1); + +size_t _mi_current_thread_count(void) { + return mi_atomic_load_relaxed(&thread_count); +} + +// This is called from the `mi_malloc_generic` +void mi_thread_init(void) mi_attr_noexcept +{ + // ensure our process has started already + mi_process_init(); + + // initialize the thread local default heap + // (this will call `_mi_heap_set_default_direct` and thus set the + // fiber/pthread key to a non-zero value, ensuring `_mi_thread_done` is called) + if (_mi_thread_heap_init()) return; // returns true if already initialized + + _mi_stat_increase(&_mi_stats_main.threads, 1); + mi_atomic_increment_relaxed(&thread_count); + //_mi_verbose_message("thread init: 0x%zx\n", _mi_thread_id()); +} + +void mi_thread_done(void) mi_attr_noexcept { + _mi_thread_done(NULL); +} + +void _mi_thread_done(mi_heap_t* heap) +{ + // calling with NULL implies using the default heap + if (heap == NULL) { + heap = mi_prim_get_default_heap(); + if (heap == NULL) return; + } + + // prevent re-entrancy through heap_done/heap_set_default_direct (issue #699) + if (!mi_heap_is_initialized(heap)) { + return; + } + + // adjust stats + mi_atomic_decrement_relaxed(&thread_count); + _mi_stat_decrease(&_mi_stats_main.threads, 1); + + // check thread-id as on Windows shutdown with FLS the main (exit) thread may call this on thread-local heaps... + if (heap->thread_id != _mi_thread_id()) return; + + // abandon the thread local heap + if (_mi_thread_heap_done(heap)) return; // returns true if already ran +} + +void _mi_heap_set_default_direct(mi_heap_t* heap) { + mi_assert_internal(heap != NULL); + #if defined(MI_TLS_SLOT) + mi_prim_tls_slot_set(MI_TLS_SLOT,heap); + #elif defined(MI_TLS_PTHREAD_SLOT_OFS) + *mi_prim_tls_pthread_heap_slot() = heap; + #elif defined(MI_TLS_PTHREAD) + // we use _mi_heap_default_key + #else + _mi_heap_default = heap; + #endif + + // ensure the default heap is passed to `_mi_thread_done` + // setting to a non-NULL value also ensures `mi_thread_done` is called. + _mi_prim_thread_associate_default_heap(heap); +} + +void mi_thread_set_in_threadpool(void) mi_attr_noexcept { + // nothing +} + +// -------------------------------------------------------- +// Run functions on process init/done, and thread init/done +// -------------------------------------------------------- +static bool os_preloading = true; // true until this module is initialized + +// Returns true if this module has not been initialized; Don't use C runtime routines until it returns false. +bool mi_decl_noinline _mi_preloading(void) { + return os_preloading; +} + +// Returns true if mimalloc was redirected +mi_decl_nodiscard bool mi_is_redirected(void) mi_attr_noexcept { + return _mi_is_redirected(); +} + +// Called once by the process loader from `src/prim/prim.c` +void _mi_process_load(void) { + mi_heap_main_init(); + #if defined(__APPLE__) || defined(MI_TLS_RECURSE_GUARD) + volatile mi_heap_t* dummy = _mi_heap_default; // access TLS to allocate it before setting tls_initialized to true; + if (dummy == NULL) return; // use dummy or otherwise the access may get optimized away (issue #697) + #endif + os_preloading = false; + mi_assert_internal(_mi_is_main_thread()); + _mi_options_init(); + mi_process_setup_auto_thread_done(); + mi_process_init(); + if (_mi_is_redirected()) _mi_verbose_message("malloc is redirected.\n"); + + // show message from the redirector (if present) + const char* msg = NULL; + _mi_allocator_init(&msg); + if (msg != NULL && (mi_option_is_enabled(mi_option_verbose) || mi_option_is_enabled(mi_option_show_errors))) { + _mi_fputs(NULL,NULL,NULL,msg); + } + + // reseed random + _mi_random_reinit_if_weak(&_mi_heap_main.random); +} + +#if defined(_WIN32) && (defined(_M_IX86) || defined(_M_X64)) +#include +mi_decl_cache_align bool _mi_cpu_has_fsrm = false; +mi_decl_cache_align bool _mi_cpu_has_erms = false; + +static void mi_detect_cpu_features(void) { + // FSRM for fast short rep movsb/stosb support (AMD Zen3+ (~2020) or Intel Ice Lake+ (~2017)) + // EMRS for fast enhanced rep movsb/stosb support + int32_t cpu_info[4]; + __cpuid(cpu_info, 7); + _mi_cpu_has_fsrm = ((cpu_info[3] & (1 << 4)) != 0); // bit 4 of EDX : see + _mi_cpu_has_erms = ((cpu_info[1] & (1 << 9)) != 0); // bit 9 of EBX : see +} +#else +static void mi_detect_cpu_features(void) { + // nothing +} +#endif + +// Initialize the process; called by thread_init or the process loader +void mi_process_init(void) mi_attr_noexcept { + // ensure we are called once + static mi_atomic_once_t process_init; + #if _MSC_VER < 1920 + mi_heap_main_init(); // vs2017 can dynamically re-initialize _mi_heap_main + #endif + if (!mi_atomic_once(&process_init)) return; + _mi_process_is_initialized = true; + _mi_verbose_message("process init: 0x%zx\n", _mi_thread_id()); + mi_process_setup_auto_thread_done(); + + mi_detect_cpu_features(); + _mi_os_init(); + mi_heap_main_init(); + mi_thread_init(); + + #if defined(_WIN32) + // On windows, when building as a static lib the FLS cleanup happens to early for the main thread. + // To avoid this, set the FLS value for the main thread to NULL so the fls cleanup + // will not call _mi_thread_done on the (still executing) main thread. See issue #508. + _mi_prim_thread_associate_default_heap(NULL); + #endif + + mi_stats_reset(); // only call stat reset *after* thread init (or the heap tld == NULL) + mi_track_init(); + + if (mi_option_is_enabled(mi_option_reserve_huge_os_pages)) { + size_t pages = mi_option_get_clamp(mi_option_reserve_huge_os_pages, 0, 128*1024); + long reserve_at = mi_option_get(mi_option_reserve_huge_os_pages_at); + if (reserve_at != -1) { + mi_reserve_huge_os_pages_at(pages, reserve_at, pages*500); + } else { + mi_reserve_huge_os_pages_interleave(pages, 0, pages*500); + } + } + if (mi_option_is_enabled(mi_option_reserve_os_memory)) { + long ksize = mi_option_get(mi_option_reserve_os_memory); + if (ksize > 0) { + mi_reserve_os_memory((size_t)ksize*MI_KiB, true /* commit? */, true /* allow large pages? */); + } + } +} + +// Called when the process is done (through `at_exit`) +void mi_cdecl _mi_process_done(void) { + // only shutdown if we were initialized + if (!_mi_process_is_initialized) return; + // ensure we are called once + static bool process_done = false; + if (process_done) return; + process_done = true; + + // get the default heap so we don't need to acces thread locals anymore + mi_heap_t* heap = mi_prim_get_default_heap(); // use prim to not initialize any heap + mi_assert_internal(heap != NULL); + + // release any thread specific resources and ensure _mi_thread_done is called on all but the main thread + _mi_prim_thread_done_auto_done(); + + + #ifndef MI_SKIP_COLLECT_ON_EXIT + #if (MI_DEBUG || !defined(MI_SHARED_LIB)) + // free all memory if possible on process exit. This is not needed for a stand-alone process + // but should be done if mimalloc is statically linked into another shared library which + // is repeatedly loaded/unloaded, see issue #281. + mi_heap_collect(heap, true /* force */ ); + #endif + #endif + + // Forcefully release all retained memory; this can be dangerous in general if overriding regular malloc/free + // since after process_done there might still be other code running that calls `free` (like at_exit routines, + // or C-runtime termination code. + if (mi_option_is_enabled(mi_option_destroy_on_exit)) { + mi_heap_collect(heap, true /* force */); + _mi_heap_unsafe_destroy_all(heap); // forcefully release all memory held by all heaps (of this thread only!) + _mi_arena_unsafe_destroy_all(); + _mi_segment_map_unsafe_destroy(); + } + + if (mi_option_is_enabled(mi_option_show_stats) || mi_option_is_enabled(mi_option_verbose)) { + mi_stats_print(NULL); + } + _mi_allocator_done(); + _mi_verbose_message("process done: 0x%zx\n", _mi_heap_main.thread_id); + os_preloading = true; // don't call the C runtime anymore +} + diff --git a/compat/mimalloc/libc.c b/compat/mimalloc/libc.c new file mode 100644 index 00000000000000..52d095eb240dc1 --- /dev/null +++ b/compat/mimalloc/libc.c @@ -0,0 +1,334 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2023, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ + +// -------------------------------------------------------- +// This module defines various std libc functions to reduce +// the dependency on libc, and also prevent errors caused +// by some libc implementations when called before `main` +// executes (due to malloc redirection) +// -------------------------------------------------------- + +#include "mimalloc.h" +#include "mimalloc/internal.h" +#include "mimalloc/prim.h" // mi_prim_getenv + +char _mi_toupper(char c) { + if (c >= 'a' && c <= 'z') return (c - 'a' + 'A'); + else return c; +} + +int _mi_strnicmp(const char* s, const char* t, size_t n) { + if (n == 0) return 0; + for (; *s != 0 && *t != 0 && n > 0; s++, t++, n--) { + if (_mi_toupper(*s) != _mi_toupper(*t)) break; + } + return (n == 0 ? 0 : *s - *t); +} + +void _mi_strlcpy(char* dest, const char* src, size_t dest_size) { + if (dest==NULL || src==NULL || dest_size == 0) return; + // copy until end of src, or when dest is (almost) full + while (*src != 0 && dest_size > 1) { + *dest++ = *src++; + dest_size--; + } + // always zero terminate + *dest = 0; +} + +void _mi_strlcat(char* dest, const char* src, size_t dest_size) { + if (dest==NULL || src==NULL || dest_size == 0) return; + // find end of string in the dest buffer + while (*dest != 0 && dest_size > 1) { + dest++; + dest_size--; + } + // and catenate + _mi_strlcpy(dest, src, dest_size); +} + +size_t _mi_strlen(const char* s) { + if (s==NULL) return 0; + size_t len = 0; + while(s[len] != 0) { len++; } + return len; +} + +size_t _mi_strnlen(const char* s, size_t max_len) { + if (s==NULL) return 0; + size_t len = 0; + while(s[len] != 0 && len < max_len) { len++; } + return len; +} + +#ifdef MI_NO_GETENV +bool _mi_getenv(const char* name, char* result, size_t result_size) { + MI_UNUSED(name); + MI_UNUSED(result); + MI_UNUSED(result_size); + return false; +} +#else +bool _mi_getenv(const char* name, char* result, size_t result_size) { + if (name==NULL || result == NULL || result_size < 64) return false; + return _mi_prim_getenv(name,result,result_size); +} +#endif + +// -------------------------------------------------------- +// Define our own limited `_mi_vsnprintf` and `_mi_snprintf` +// This is mostly to avoid calling these when libc is not yet +// initialized (and to reduce dependencies) +// +// format: d i, p x u, s +// prec: z l ll L +// width: 10 +// align-left: - +// fill: 0 +// plus: + +// -------------------------------------------------------- + +static void mi_outc(char c, char** out, char* end) { + char* p = *out; + if (p >= end) return; + *p = c; + *out = p + 1; +} + +static void mi_outs(const char* s, char** out, char* end) { + if (s == NULL) return; + char* p = *out; + while (*s != 0 && p < end) { + *p++ = *s++; + } + *out = p; +} + +static void mi_out_fill(char fill, size_t len, char** out, char* end) { + char* p = *out; + for (size_t i = 0; i < len && p < end; i++) { + *p++ = fill; + } + *out = p; +} + +static void mi_out_alignright(char fill, char* start, size_t len, size_t extra, char* end) { + if (len == 0 || extra == 0) return; + if (start + len + extra >= end) return; + // move `len` characters to the right (in reverse since it can overlap) + for (size_t i = 1; i <= len; i++) { + start[len + extra - i] = start[len - i]; + } + // and fill the start + for (size_t i = 0; i < extra; i++) { + start[i] = fill; + } +} + + +static void mi_out_num(uintmax_t x, size_t base, char prefix, char** out, char* end) +{ + if (x == 0 || base == 0 || base > 16) { + if (prefix != 0) { mi_outc(prefix, out, end); } + mi_outc('0',out,end); + } + else { + // output digits in reverse + char* start = *out; + while (x > 0) { + char digit = (char)(x % base); + mi_outc((digit <= 9 ? '0' + digit : 'A' + digit - 10),out,end); + x = x / base; + } + if (prefix != 0) { + mi_outc(prefix, out, end); + } + size_t len = *out - start; + // and reverse in-place + for (size_t i = 0; i < (len / 2); i++) { + char c = start[len - i - 1]; + start[len - i - 1] = start[i]; + start[i] = c; + } + } +} + + +#define MI_NEXTC() c = *in; if (c==0) break; in++; + +int _mi_vsnprintf(char* buf, size_t bufsize, const char* fmt, va_list args) { + if (buf == NULL || bufsize == 0 || fmt == NULL) return 0; + buf[bufsize - 1] = 0; + char* const end = buf + (bufsize - 1); + const char* in = fmt; + char* out = buf; + while (true) { + if (out >= end) break; + char c; + MI_NEXTC(); + if (c != '%') { + if ((c >= ' ' && c <= '~') || c=='\n' || c=='\r' || c=='\t') { // output visible ascii or standard control only + mi_outc(c, &out, end); + } + } + else { + MI_NEXTC(); + char fill = ' '; + size_t width = 0; + char numtype = 'd'; + char numplus = 0; + bool alignright = true; + if (c == '+' || c == ' ') { numplus = c; MI_NEXTC(); } + if (c == '-') { alignright = false; MI_NEXTC(); } + if (c == '0') { fill = '0'; MI_NEXTC(); } + if (c >= '1' && c <= '9') { + width = (c - '0'); MI_NEXTC(); + while (c >= '0' && c <= '9') { + width = (10 * width) + (c - '0'); MI_NEXTC(); + } + if (c == 0) break; // extra check due to while + } + if (c == 'z' || c == 't' || c == 'L') { numtype = c; MI_NEXTC(); } + else if (c == 'l') { + numtype = c; MI_NEXTC(); + if (c == 'l') { numtype = 'L'; MI_NEXTC(); } + } + + char* start = out; + if (c == 's') { + // string + const char* s = va_arg(args, const char*); + mi_outs(s, &out, end); + } + else if (c == 'p' || c == 'x' || c == 'u') { + // unsigned + uintmax_t x = 0; + if (c == 'x' || c == 'u') { + if (numtype == 'z') x = va_arg(args, size_t); + else if (numtype == 't') x = va_arg(args, uintptr_t); // unsigned ptrdiff_t + else if (numtype == 'L') x = va_arg(args, unsigned long long); + else if (numtype == 'l') x = va_arg(args, unsigned long); + else x = va_arg(args, unsigned int); + } + else if (c == 'p') { + x = va_arg(args, uintptr_t); + mi_outs("0x", &out, end); + start = out; + width = (width >= 2 ? width - 2 : 0); + } + if (width == 0 && (c == 'x' || c == 'p')) { + if (c == 'p') { width = 2 * (x <= UINT32_MAX ? 4 : ((x >> 16) <= UINT32_MAX ? 6 : sizeof(void*))); } + if (width == 0) { width = 2; } + fill = '0'; + } + mi_out_num(x, (c == 'x' || c == 'p' ? 16 : 10), numplus, &out, end); + } + else if (c == 'i' || c == 'd') { + // signed + intmax_t x = 0; + if (numtype == 'z') x = va_arg(args, intptr_t ); + else if (numtype == 't') x = va_arg(args, ptrdiff_t); + else if (numtype == 'L') x = va_arg(args, long long); + else if (numtype == 'l') x = va_arg(args, long); + else x = va_arg(args, int); + char pre = 0; + if (x < 0) { + pre = '-'; + if (x > INTMAX_MIN) { x = -x; } + } + else if (numplus != 0) { + pre = numplus; + } + mi_out_num((uintmax_t)x, 10, pre, &out, end); + } + else if (c >= ' ' && c <= '~') { + // unknown format + mi_outc('%', &out, end); + mi_outc(c, &out, end); + } + + // fill & align + mi_assert_internal(out <= end); + mi_assert_internal(out >= start); + const size_t len = out - start; + if (len < width) { + mi_out_fill(fill, width - len, &out, end); + if (alignright && out <= end) { + mi_out_alignright(fill, start, len, width - len, end); + } + } + } + } + mi_assert_internal(out <= end); + *out = 0; + return (int)(out - buf); +} + +int _mi_snprintf(char* buf, size_t buflen, const char* fmt, ...) { + va_list args; + va_start(args, fmt); + const int written = _mi_vsnprintf(buf, buflen, fmt, args); + va_end(args); + return written; +} + + +#if MI_SIZE_SIZE == 4 +#define mi_mask_even_bits32 (0x55555555) +#define mi_mask_even_pairs32 (0x33333333) +#define mi_mask_even_nibbles32 (0x0F0F0F0F) + +// sum of all the bytes in `x` if it is guaranteed that the sum < 256! +static size_t mi_byte_sum32(uint32_t x) { + // perform `x * 0x01010101`: the highest byte contains the sum of all bytes. + x += (x << 8); + x += (x << 16); + return (size_t)(x >> 24); +} + +static size_t mi_popcount_generic32(uint32_t x) { + // first count each 2-bit group `a`, where: a==0b00 -> 00, a==0b01 -> 01, a==0b10 -> 01, a==0b11 -> 10 + // in other words, `a - (a>>1)`; to do this in parallel, we need to mask to prevent spilling a bit pair + // into the lower bit-pair: + x = x - ((x >> 1) & mi_mask_even_bits32); + // add the 2-bit pair results + x = (x & mi_mask_even_pairs32) + ((x >> 2) & mi_mask_even_pairs32); + // add the 4-bit nibble results + x = (x + (x >> 4)) & mi_mask_even_nibbles32; + // each byte now has a count of its bits, we can sum them now: + return mi_byte_sum32(x); +} + +mi_decl_noinline size_t _mi_popcount_generic(size_t x) { + return mi_popcount_generic32(x); +} + +#else +#define mi_mask_even_bits64 (0x5555555555555555) +#define mi_mask_even_pairs64 (0x3333333333333333) +#define mi_mask_even_nibbles64 (0x0F0F0F0F0F0F0F0F) + +// sum of all the bytes in `x` if it is guaranteed that the sum < 256! +static size_t mi_byte_sum64(uint64_t x) { + x += (x << 8); + x += (x << 16); + x += (x << 32); + return (size_t)(x >> 56); +} + +static size_t mi_popcount_generic64(uint64_t x) { + x = x - ((x >> 1) & mi_mask_even_bits64); + x = (x & mi_mask_even_pairs64) + ((x >> 2) & mi_mask_even_pairs64); + x = (x + (x >> 4)) & mi_mask_even_nibbles64; + return mi_byte_sum64(x); +} + +mi_decl_noinline size_t _mi_popcount_generic(size_t x) { + return mi_popcount_generic64(x); +} +#endif + diff --git a/compat/mimalloc/mimalloc-stats.h b/compat/mimalloc/mimalloc-stats.h new file mode 100644 index 00000000000000..44c4886f88a0c7 --- /dev/null +++ b/compat/mimalloc/mimalloc-stats.h @@ -0,0 +1,103 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2025, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ +#pragma once +#ifndef MIMALLOC_STATS_H +#define MIMALLOC_STATS_H + +#include +#include + +#define MI_STAT_VERSION 1 // increased on every backward incompatible change + +// count allocation over time +typedef struct mi_stat_count_s { + int64_t total; // total allocated + int64_t peak; // peak allocation + int64_t current; // current allocation +} mi_stat_count_t; + +// counters only increase +typedef struct mi_stat_counter_s { + int64_t total; // total count +} mi_stat_counter_t; + +#define MI_STAT_FIELDS() \ + MI_STAT_COUNT(pages) /* count of mimalloc pages */ \ + MI_STAT_COUNT(reserved) /* reserved memory bytes */ \ + MI_STAT_COUNT(committed) /* committed bytes */ \ + MI_STAT_COUNT(reset) /* reset bytes */ \ + MI_STAT_COUNT(purged) /* purged bytes */ \ + MI_STAT_COUNT(page_committed) /* committed memory inside pages */ \ + MI_STAT_COUNT(pages_abandoned) /* abandonded pages count */ \ + MI_STAT_COUNT(threads) /* number of threads */ \ + MI_STAT_COUNT(malloc_normal) /* allocated bytes <= MI_LARGE_OBJ_SIZE_MAX */ \ + MI_STAT_COUNT(malloc_huge) /* allocated bytes in huge pages */ \ + MI_STAT_COUNT(malloc_requested) /* malloc requested bytes */ \ + \ + MI_STAT_COUNTER(mmap_calls) \ + MI_STAT_COUNTER(commit_calls) \ + MI_STAT_COUNTER(reset_calls) \ + MI_STAT_COUNTER(purge_calls) \ + MI_STAT_COUNTER(arena_count) /* number of memory arena's */ \ + MI_STAT_COUNTER(malloc_normal_count) /* number of blocks <= MI_LARGE_OBJ_SIZE_MAX */ \ + MI_STAT_COUNTER(malloc_huge_count) /* number of huge bloks */ \ + MI_STAT_COUNTER(malloc_guarded_count) /* number of allocations with guard pages */ \ + \ + /* internal statistics */ \ + MI_STAT_COUNTER(arena_rollback_count) \ + MI_STAT_COUNTER(arena_purges) \ + MI_STAT_COUNTER(pages_extended) /* number of page extensions */ \ + MI_STAT_COUNTER(pages_retire) /* number of pages that are retired */ \ + MI_STAT_COUNTER(page_searches) /* searches for a fresh page */ \ + /* only on v1 and v2 */ \ + MI_STAT_COUNT(segments) \ + MI_STAT_COUNT(segments_abandoned) \ + MI_STAT_COUNT(segments_cache) \ + MI_STAT_COUNT(_segments_reserved) \ + /* only on v3 */ \ + MI_STAT_COUNTER(pages_reclaim_on_alloc) \ + MI_STAT_COUNTER(pages_reclaim_on_free) \ + MI_STAT_COUNTER(pages_reabandon_full) \ + MI_STAT_COUNTER(pages_unabandon_busy_wait) \ + + +// Define the statistics structure +#define MI_BIN_HUGE (73U) // see types.h +#define MI_STAT_COUNT(stat) mi_stat_count_t stat; +#define MI_STAT_COUNTER(stat) mi_stat_counter_t stat; + +typedef struct mi_stats_s +{ + int version; + + MI_STAT_FIELDS() + + // future extension + mi_stat_count_t _stat_reserved[4]; + mi_stat_counter_t _stat_counter_reserved[4]; + + // size segregated statistics + mi_stat_count_t malloc_bins[MI_BIN_HUGE+1]; // allocation per size bin + mi_stat_count_t page_bins[MI_BIN_HUGE+1]; // pages allocated per size bin +} mi_stats_t; + +#undef MI_STAT_COUNT +#undef MI_STAT_COUNTER + +// Exported definitions +#ifdef __cplusplus +extern "C" { +#endif + +mi_decl_export void mi_stats_get( size_t stats_size, mi_stats_t* stats ) mi_attr_noexcept; +mi_decl_export char* mi_stats_get_json( size_t buf_size, char* buf ) mi_attr_noexcept; // use mi_free to free the result if the input buf == NULL + +#ifdef __cplusplus +} +#endif + +#endif // MIMALLOC_STATS_H diff --git a/compat/mimalloc/mimalloc.h b/compat/mimalloc/mimalloc.h new file mode 100644 index 00000000000000..ff6f0568301db4 --- /dev/null +++ b/compat/mimalloc/mimalloc.h @@ -0,0 +1,608 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2025, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ +#pragma once +#ifndef MIMALLOC_H +#define MIMALLOC_H + +#define MI_MALLOC_VERSION 223 // major + 2 digits minor + +// ------------------------------------------------------ +// Compiler specific attributes +// ------------------------------------------------------ + +#ifdef __cplusplus + #if (__cplusplus >= 201103L) || (_MSC_VER > 1900) // C++11 + #define mi_attr_noexcept noexcept + #else + #define mi_attr_noexcept throw() + #endif +#else + #define mi_attr_noexcept +#endif + +#if defined(__cplusplus) && (__cplusplus >= 201703) + #define mi_decl_nodiscard [[nodiscard]] +#elif (defined(__GNUC__) && (__GNUC__ >= 4)) || defined(__clang__) // includes clang, icc, and clang-cl + #define mi_decl_nodiscard __attribute__((warn_unused_result)) +#elif defined(_HAS_NODISCARD) + #define mi_decl_nodiscard _NODISCARD +#elif (_MSC_VER >= 1700) + #define mi_decl_nodiscard _Check_return_ +#else + #define mi_decl_nodiscard +#endif + +#if defined(_MSC_VER) || defined(__MINGW32__) + #if !defined(MI_SHARED_LIB) + #define mi_decl_export + #elif defined(MI_SHARED_LIB_EXPORT) + #define mi_decl_export __declspec(dllexport) + #else + #define mi_decl_export __declspec(dllimport) + #endif + #if defined(__MINGW32__) + #define mi_decl_restrict + #define mi_attr_malloc __attribute__((malloc)) + #else + #if (_MSC_VER >= 1900) && !defined(__EDG__) + #define mi_decl_restrict __declspec(allocator) __declspec(restrict) + #else + #define mi_decl_restrict __declspec(restrict) + #endif + #define mi_attr_malloc + #endif + #define mi_cdecl __cdecl + #define mi_attr_alloc_size(s) + #define mi_attr_alloc_size2(s1,s2) + #define mi_attr_alloc_align(p) +#elif defined(__GNUC__) // includes clang and icc + #if defined(MI_SHARED_LIB) && defined(MI_SHARED_LIB_EXPORT) + #define mi_decl_export __attribute__((visibility("default"))) + #else + #define mi_decl_export + #endif + #define mi_cdecl // leads to warnings... __attribute__((cdecl)) + #define mi_decl_restrict + #define mi_attr_malloc __attribute__((malloc)) + #if (defined(__clang_major__) && (__clang_major__ < 4)) || (__GNUC__ < 5) + #define mi_attr_alloc_size(s) + #define mi_attr_alloc_size2(s1,s2) + #define mi_attr_alloc_align(p) + #elif defined(__INTEL_COMPILER) + #define mi_attr_alloc_size(s) __attribute__((alloc_size(s))) + #define mi_attr_alloc_size2(s1,s2) __attribute__((alloc_size(s1,s2))) + #define mi_attr_alloc_align(p) + #else + #define mi_attr_alloc_size(s) __attribute__((alloc_size(s))) + #define mi_attr_alloc_size2(s1,s2) __attribute__((alloc_size(s1,s2))) + #define mi_attr_alloc_align(p) __attribute__((alloc_align(p))) + #endif +#else + #define mi_cdecl + #define mi_decl_export + #define mi_decl_restrict + #define mi_attr_malloc + #define mi_attr_alloc_size(s) + #define mi_attr_alloc_size2(s1,s2) + #define mi_attr_alloc_align(p) +#endif + +// ------------------------------------------------------ +// Includes +// ------------------------------------------------------ + +#include // size_t +#include // bool +#include // INTPTR_MAX + +#ifdef __cplusplus +extern "C" { +#endif + +// ------------------------------------------------------ +// Standard malloc interface +// ------------------------------------------------------ + +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_malloc(size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_calloc(size_t count, size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size2(1,2); +mi_decl_nodiscard mi_decl_export void* mi_realloc(void* p, size_t newsize) mi_attr_noexcept mi_attr_alloc_size(2); +mi_decl_export void* mi_expand(void* p, size_t newsize) mi_attr_noexcept mi_attr_alloc_size(2); + +mi_decl_export void mi_free(void* p) mi_attr_noexcept; +mi_decl_nodiscard mi_decl_export mi_decl_restrict char* mi_strdup(const char* s) mi_attr_noexcept mi_attr_malloc; +mi_decl_nodiscard mi_decl_export mi_decl_restrict char* mi_strndup(const char* s, size_t n) mi_attr_noexcept mi_attr_malloc; +mi_decl_nodiscard mi_decl_export mi_decl_restrict char* mi_realpath(const char* fname, char* resolved_name) mi_attr_noexcept mi_attr_malloc; + +// ------------------------------------------------------ +// Extended functionality +// ------------------------------------------------------ +#define MI_SMALL_WSIZE_MAX (128) +#define MI_SMALL_SIZE_MAX (MI_SMALL_WSIZE_MAX*sizeof(void*)) + +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_malloc_small(size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_zalloc_small(size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_zalloc(size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1); + +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_mallocn(size_t count, size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size2(1,2); +mi_decl_nodiscard mi_decl_export void* mi_reallocn(void* p, size_t count, size_t size) mi_attr_noexcept mi_attr_alloc_size2(2,3); +mi_decl_nodiscard mi_decl_export void* mi_reallocf(void* p, size_t newsize) mi_attr_noexcept mi_attr_alloc_size(2); + +mi_decl_nodiscard mi_decl_export size_t mi_usable_size(const void* p) mi_attr_noexcept; +mi_decl_nodiscard mi_decl_export size_t mi_good_size(size_t size) mi_attr_noexcept; + + +// ------------------------------------------------------ +// Internals +// ------------------------------------------------------ + +typedef void (mi_cdecl mi_deferred_free_fun)(bool force, unsigned long long heartbeat, void* arg); +mi_decl_export void mi_register_deferred_free(mi_deferred_free_fun* deferred_free, void* arg) mi_attr_noexcept; + +typedef void (mi_cdecl mi_output_fun)(const char* msg, void* arg); +mi_decl_export void mi_register_output(mi_output_fun* out, void* arg) mi_attr_noexcept; + +typedef void (mi_cdecl mi_error_fun)(int err, void* arg); +mi_decl_export void mi_register_error(mi_error_fun* fun, void* arg); + +mi_decl_export void mi_collect(bool force) mi_attr_noexcept; +mi_decl_export int mi_version(void) mi_attr_noexcept; +mi_decl_export void mi_stats_reset(void) mi_attr_noexcept; +mi_decl_export void mi_stats_merge(void) mi_attr_noexcept; +mi_decl_export void mi_stats_print(void* out) mi_attr_noexcept; // backward compatibility: `out` is ignored and should be NULL +mi_decl_export void mi_stats_print_out(mi_output_fun* out, void* arg) mi_attr_noexcept; +mi_decl_export void mi_options_print(void) mi_attr_noexcept; + +mi_decl_export void mi_process_init(void) mi_attr_noexcept; +mi_decl_export void mi_thread_init(void) mi_attr_noexcept; +mi_decl_export void mi_thread_done(void) mi_attr_noexcept; +mi_decl_export void mi_thread_stats_print_out(mi_output_fun* out, void* arg) mi_attr_noexcept; + +mi_decl_export void mi_process_info(size_t* elapsed_msecs, size_t* user_msecs, size_t* system_msecs, + size_t* current_rss, size_t* peak_rss, + size_t* current_commit, size_t* peak_commit, size_t* page_faults) mi_attr_noexcept; + +// ------------------------------------------------------------------------------------- +// Aligned allocation +// Note that `alignment` always follows `size` for consistency with unaligned +// allocation, but unfortunately this differs from `posix_memalign` and `aligned_alloc`. +// ------------------------------------------------------------------------------------- + +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_malloc_aligned(size_t size, size_t alignment) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1) mi_attr_alloc_align(2); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_malloc_aligned_at(size_t size, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_zalloc_aligned(size_t size, size_t alignment) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1) mi_attr_alloc_align(2); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_zalloc_aligned_at(size_t size, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_calloc_aligned(size_t count, size_t size, size_t alignment) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size2(1,2) mi_attr_alloc_align(3); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_calloc_aligned_at(size_t count, size_t size, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size2(1,2); +mi_decl_nodiscard mi_decl_export void* mi_realloc_aligned(void* p, size_t newsize, size_t alignment) mi_attr_noexcept mi_attr_alloc_size(2) mi_attr_alloc_align(3); +mi_decl_nodiscard mi_decl_export void* mi_realloc_aligned_at(void* p, size_t newsize, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_alloc_size(2); + + +// ------------------------------------------------------------------------------------- +// Heaps: first-class, but can only allocate from the same thread that created it. +// ------------------------------------------------------------------------------------- + +struct mi_heap_s; +typedef struct mi_heap_s mi_heap_t; + +mi_decl_nodiscard mi_decl_export mi_heap_t* mi_heap_new(void); +mi_decl_export void mi_heap_delete(mi_heap_t* heap); +mi_decl_export void mi_heap_destroy(mi_heap_t* heap); +mi_decl_export mi_heap_t* mi_heap_set_default(mi_heap_t* heap); +mi_decl_export mi_heap_t* mi_heap_get_default(void); +mi_decl_export mi_heap_t* mi_heap_get_backing(void); +mi_decl_export void mi_heap_collect(mi_heap_t* heap, bool force) mi_attr_noexcept; + +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_malloc(mi_heap_t* heap, size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(2); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_zalloc(mi_heap_t* heap, size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(2); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_calloc(mi_heap_t* heap, size_t count, size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size2(2, 3); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_mallocn(mi_heap_t* heap, size_t count, size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size2(2, 3); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_malloc_small(mi_heap_t* heap, size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(2); + +mi_decl_nodiscard mi_decl_export void* mi_heap_realloc(mi_heap_t* heap, void* p, size_t newsize) mi_attr_noexcept mi_attr_alloc_size(3); +mi_decl_nodiscard mi_decl_export void* mi_heap_reallocn(mi_heap_t* heap, void* p, size_t count, size_t size) mi_attr_noexcept mi_attr_alloc_size2(3,4); +mi_decl_nodiscard mi_decl_export void* mi_heap_reallocf(mi_heap_t* heap, void* p, size_t newsize) mi_attr_noexcept mi_attr_alloc_size(3); + +mi_decl_nodiscard mi_decl_export mi_decl_restrict char* mi_heap_strdup(mi_heap_t* heap, const char* s) mi_attr_noexcept mi_attr_malloc; +mi_decl_nodiscard mi_decl_export mi_decl_restrict char* mi_heap_strndup(mi_heap_t* heap, const char* s, size_t n) mi_attr_noexcept mi_attr_malloc; +mi_decl_nodiscard mi_decl_export mi_decl_restrict char* mi_heap_realpath(mi_heap_t* heap, const char* fname, char* resolved_name) mi_attr_noexcept mi_attr_malloc; + +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_malloc_aligned(mi_heap_t* heap, size_t size, size_t alignment) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(2) mi_attr_alloc_align(3); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_malloc_aligned_at(mi_heap_t* heap, size_t size, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(2); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_zalloc_aligned(mi_heap_t* heap, size_t size, size_t alignment) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(2) mi_attr_alloc_align(3); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_zalloc_aligned_at(mi_heap_t* heap, size_t size, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(2); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_calloc_aligned(mi_heap_t* heap, size_t count, size_t size, size_t alignment) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size2(2, 3) mi_attr_alloc_align(4); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_calloc_aligned_at(mi_heap_t* heap, size_t count, size_t size, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size2(2, 3); +mi_decl_nodiscard mi_decl_export void* mi_heap_realloc_aligned(mi_heap_t* heap, void* p, size_t newsize, size_t alignment) mi_attr_noexcept mi_attr_alloc_size(3) mi_attr_alloc_align(4); +mi_decl_nodiscard mi_decl_export void* mi_heap_realloc_aligned_at(mi_heap_t* heap, void* p, size_t newsize, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_alloc_size(3); + + +// -------------------------------------------------------------------------------- +// Zero initialized re-allocation. +// Only valid on memory that was originally allocated with zero initialization too. +// e.g. `mi_calloc`, `mi_zalloc`, `mi_zalloc_aligned` etc. +// see +// -------------------------------------------------------------------------------- + +mi_decl_nodiscard mi_decl_export void* mi_rezalloc(void* p, size_t newsize) mi_attr_noexcept mi_attr_alloc_size(2); +mi_decl_nodiscard mi_decl_export void* mi_recalloc(void* p, size_t newcount, size_t size) mi_attr_noexcept mi_attr_alloc_size2(2,3); + +mi_decl_nodiscard mi_decl_export void* mi_rezalloc_aligned(void* p, size_t newsize, size_t alignment) mi_attr_noexcept mi_attr_alloc_size(2) mi_attr_alloc_align(3); +mi_decl_nodiscard mi_decl_export void* mi_rezalloc_aligned_at(void* p, size_t newsize, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_alloc_size(2); +mi_decl_nodiscard mi_decl_export void* mi_recalloc_aligned(void* p, size_t newcount, size_t size, size_t alignment) mi_attr_noexcept mi_attr_alloc_size2(2,3) mi_attr_alloc_align(4); +mi_decl_nodiscard mi_decl_export void* mi_recalloc_aligned_at(void* p, size_t newcount, size_t size, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_alloc_size2(2,3); + +mi_decl_nodiscard mi_decl_export void* mi_heap_rezalloc(mi_heap_t* heap, void* p, size_t newsize) mi_attr_noexcept mi_attr_alloc_size(3); +mi_decl_nodiscard mi_decl_export void* mi_heap_recalloc(mi_heap_t* heap, void* p, size_t newcount, size_t size) mi_attr_noexcept mi_attr_alloc_size2(3,4); + +mi_decl_nodiscard mi_decl_export void* mi_heap_rezalloc_aligned(mi_heap_t* heap, void* p, size_t newsize, size_t alignment) mi_attr_noexcept mi_attr_alloc_size(3) mi_attr_alloc_align(4); +mi_decl_nodiscard mi_decl_export void* mi_heap_rezalloc_aligned_at(mi_heap_t* heap, void* p, size_t newsize, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_alloc_size(3); +mi_decl_nodiscard mi_decl_export void* mi_heap_recalloc_aligned(mi_heap_t* heap, void* p, size_t newcount, size_t size, size_t alignment) mi_attr_noexcept mi_attr_alloc_size2(3,4) mi_attr_alloc_align(5); +mi_decl_nodiscard mi_decl_export void* mi_heap_recalloc_aligned_at(mi_heap_t* heap, void* p, size_t newcount, size_t size, size_t alignment, size_t offset) mi_attr_noexcept mi_attr_alloc_size2(3,4); + + +// ------------------------------------------------------ +// Analysis +// ------------------------------------------------------ + +mi_decl_export bool mi_heap_contains_block(mi_heap_t* heap, const void* p); +mi_decl_export bool mi_heap_check_owned(mi_heap_t* heap, const void* p); +mi_decl_export bool mi_check_owned(const void* p); + +// An area of heap space contains blocks of a single size. +typedef struct mi_heap_area_s { + void* blocks; // start of the area containing heap blocks + size_t reserved; // bytes reserved for this area (virtual) + size_t committed; // current available bytes for this area + size_t used; // number of allocated blocks + size_t block_size; // size in bytes of each block + size_t full_block_size; // size in bytes of a full block including padding and metadata. + int heap_tag; // heap tag associated with this area +} mi_heap_area_t; + +typedef bool (mi_cdecl mi_block_visit_fun)(const mi_heap_t* heap, const mi_heap_area_t* area, void* block, size_t block_size, void* arg); + +mi_decl_export bool mi_heap_visit_blocks(const mi_heap_t* heap, bool visit_blocks, mi_block_visit_fun* visitor, void* arg); + +// Experimental +mi_decl_nodiscard mi_decl_export bool mi_is_in_heap_region(const void* p) mi_attr_noexcept; +mi_decl_nodiscard mi_decl_export bool mi_is_redirected(void) mi_attr_noexcept; + +mi_decl_export int mi_reserve_huge_os_pages_interleave(size_t pages, size_t numa_nodes, size_t timeout_msecs) mi_attr_noexcept; +mi_decl_export int mi_reserve_huge_os_pages_at(size_t pages, int numa_node, size_t timeout_msecs) mi_attr_noexcept; + +mi_decl_export int mi_reserve_os_memory(size_t size, bool commit, bool allow_large) mi_attr_noexcept; +mi_decl_export bool mi_manage_os_memory(void* start, size_t size, bool is_committed, bool is_large, bool is_zero, int numa_node) mi_attr_noexcept; + +mi_decl_export void mi_debug_show_arenas(void) mi_attr_noexcept; +mi_decl_export void mi_arenas_print(void) mi_attr_noexcept; + +// Experimental: heaps associated with specific memory arena's +typedef int mi_arena_id_t; +mi_decl_export void* mi_arena_area(mi_arena_id_t arena_id, size_t* size); +mi_decl_export int mi_reserve_huge_os_pages_at_ex(size_t pages, int numa_node, size_t timeout_msecs, bool exclusive, mi_arena_id_t* arena_id) mi_attr_noexcept; +mi_decl_export int mi_reserve_os_memory_ex(size_t size, bool commit, bool allow_large, bool exclusive, mi_arena_id_t* arena_id) mi_attr_noexcept; +mi_decl_export bool mi_manage_os_memory_ex(void* start, size_t size, bool is_committed, bool is_large, bool is_zero, int numa_node, bool exclusive, mi_arena_id_t* arena_id) mi_attr_noexcept; + +#if MI_MALLOC_VERSION >= 182 +// Create a heap that only allocates in the specified arena +mi_decl_nodiscard mi_decl_export mi_heap_t* mi_heap_new_in_arena(mi_arena_id_t arena_id); +#endif + + +// Experimental: allow sub-processes whose memory areas stay separated (and no reclamation between them) +// Used for example for separate interpreters in one process. +typedef void* mi_subproc_id_t; +mi_decl_export mi_subproc_id_t mi_subproc_main(void); +mi_decl_export mi_subproc_id_t mi_subproc_new(void); +mi_decl_export void mi_subproc_delete(mi_subproc_id_t subproc); +mi_decl_export void mi_subproc_add_current_thread(mi_subproc_id_t subproc); // this should be called right after a thread is created (and no allocation has taken place yet) + +// Experimental: visit abandoned heap areas (that are not owned by a specific heap) +mi_decl_export bool mi_abandoned_visit_blocks(mi_subproc_id_t subproc_id, int heap_tag, bool visit_blocks, mi_block_visit_fun* visitor, void* arg); + +// Experimental: objects followed by a guard page. +// A sample rate of 0 disables guarded objects, while 1 uses a guard page for every object. +// A seed of 0 uses a random start point. Only objects within the size bound are eligable for guard pages. +mi_decl_export void mi_heap_guarded_set_sample_rate(mi_heap_t* heap, size_t sample_rate, size_t seed); +mi_decl_export void mi_heap_guarded_set_size_bound(mi_heap_t* heap, size_t min, size_t max); + +// Experimental: communicate that the thread is part of a threadpool +mi_decl_export void mi_thread_set_in_threadpool(void) mi_attr_noexcept; + +// Experimental: create a new heap with a specified heap tag. Set `allow_destroy` to false to allow the thread +// to reclaim abandoned memory (with a compatible heap_tag and arena_id) but in that case `mi_heap_destroy` will +// fall back to `mi_heap_delete`. +mi_decl_nodiscard mi_decl_export mi_heap_t* mi_heap_new_ex(int heap_tag, bool allow_destroy, mi_arena_id_t arena_id); + +// deprecated +mi_decl_export int mi_reserve_huge_os_pages(size_t pages, double max_secs, size_t* pages_reserved) mi_attr_noexcept; +mi_decl_export void mi_collect_reduce(size_t target_thread_owned) mi_attr_noexcept; + + + +// ------------------------------------------------------ +// Convenience +// ------------------------------------------------------ + +#define mi_malloc_tp(tp) ((tp*)mi_malloc(sizeof(tp))) +#define mi_zalloc_tp(tp) ((tp*)mi_zalloc(sizeof(tp))) +#define mi_calloc_tp(tp,n) ((tp*)mi_calloc(n,sizeof(tp))) +#define mi_mallocn_tp(tp,n) ((tp*)mi_mallocn(n,sizeof(tp))) +#define mi_reallocn_tp(p,tp,n) ((tp*)mi_reallocn(p,n,sizeof(tp))) +#define mi_recalloc_tp(p,tp,n) ((tp*)mi_recalloc(p,n,sizeof(tp))) + +#define mi_heap_malloc_tp(hp,tp) ((tp*)mi_heap_malloc(hp,sizeof(tp))) +#define mi_heap_zalloc_tp(hp,tp) ((tp*)mi_heap_zalloc(hp,sizeof(tp))) +#define mi_heap_calloc_tp(hp,tp,n) ((tp*)mi_heap_calloc(hp,n,sizeof(tp))) +#define mi_heap_mallocn_tp(hp,tp,n) ((tp*)mi_heap_mallocn(hp,n,sizeof(tp))) +#define mi_heap_reallocn_tp(hp,p,tp,n) ((tp*)mi_heap_reallocn(hp,p,n,sizeof(tp))) +#define mi_heap_recalloc_tp(hp,p,tp,n) ((tp*)mi_heap_recalloc(hp,p,n,sizeof(tp))) + + +// ------------------------------------------------------ +// Options +// ------------------------------------------------------ + +typedef enum mi_option_e { + // stable options + mi_option_show_errors, // print error messages + mi_option_show_stats, // print statistics on termination + mi_option_verbose, // print verbose messages + // advanced options + mi_option_eager_commit, // eager commit segments? (after `eager_commit_delay` segments) (=1) + mi_option_arena_eager_commit, // eager commit arenas? Use 2 to enable just on overcommit systems (=2) + mi_option_purge_decommits, // should a memory purge decommit? (=1). Set to 0 to use memory reset on a purge (instead of decommit) + mi_option_allow_large_os_pages, // allow large (2 or 4 MiB) OS pages, implies eager commit. If false, also disables THP for the process. + mi_option_reserve_huge_os_pages, // reserve N huge OS pages (1GiB pages) at startup + mi_option_reserve_huge_os_pages_at, // reserve huge OS pages at a specific NUMA node + mi_option_reserve_os_memory, // reserve specified amount of OS memory in an arena at startup (internally, this value is in KiB; use `mi_option_get_size`) + mi_option_deprecated_segment_cache, + mi_option_deprecated_page_reset, + mi_option_abandoned_page_purge, // immediately purge delayed purges on thread termination + mi_option_deprecated_segment_reset, + mi_option_eager_commit_delay, // the first N segments per thread are not eagerly committed (but per page in the segment on demand) + mi_option_purge_delay, // memory purging is delayed by N milli seconds; use 0 for immediate purging or -1 for no purging at all. (=10) + mi_option_use_numa_nodes, // 0 = use all available numa nodes, otherwise use at most N nodes. + mi_option_disallow_os_alloc, // 1 = do not use OS memory for allocation (but only programmatically reserved arenas) + mi_option_os_tag, // tag used for OS logging (macOS only for now) (=100) + mi_option_max_errors, // issue at most N error messages + mi_option_max_warnings, // issue at most N warning messages + mi_option_max_segment_reclaim, // max. percentage of the abandoned segments can be reclaimed per try (=10%) + mi_option_destroy_on_exit, // if set, release all memory on exit; sometimes used for dynamic unloading but can be unsafe + mi_option_arena_reserve, // initial memory size for arena reservation (= 1 GiB on 64-bit) (internally, this value is in KiB; use `mi_option_get_size`) + mi_option_arena_purge_mult, // multiplier for `purge_delay` for the purging delay for arenas (=10) + mi_option_purge_extend_delay, + mi_option_abandoned_reclaim_on_free, // allow to reclaim an abandoned segment on a free (=1) + mi_option_disallow_arena_alloc, // 1 = do not use arena's for allocation (except if using specific arena id's) + mi_option_retry_on_oom, // retry on out-of-memory for N milli seconds (=400), set to 0 to disable retries. (only on windows) + mi_option_visit_abandoned, // allow visiting heap blocks from abandoned threads (=0) + mi_option_guarded_min, // only used when building with MI_GUARDED: minimal rounded object size for guarded objects (=0) + mi_option_guarded_max, // only used when building with MI_GUARDED: maximal rounded object size for guarded objects (=0) + mi_option_guarded_precise, // disregard minimal alignment requirement to always place guarded blocks exactly in front of a guard page (=0) + mi_option_guarded_sample_rate, // 1 out of N allocations in the min/max range will be guarded (=1000) + mi_option_guarded_sample_seed, // can be set to allow for a (more) deterministic re-execution when a guard page is triggered (=0) + mi_option_target_segments_per_thread, // experimental (=0) + mi_option_generic_collect, // collect heaps every N (=10000) generic allocation calls + _mi_option_last, + // legacy option names + mi_option_large_os_pages = mi_option_allow_large_os_pages, + mi_option_eager_region_commit = mi_option_arena_eager_commit, + mi_option_reset_decommits = mi_option_purge_decommits, + mi_option_reset_delay = mi_option_purge_delay, + mi_option_abandoned_page_reset = mi_option_abandoned_page_purge, + mi_option_limit_os_alloc = mi_option_disallow_os_alloc +} mi_option_t; + + +mi_decl_nodiscard mi_decl_export bool mi_option_is_enabled(mi_option_t option); +mi_decl_export void mi_option_enable(mi_option_t option); +mi_decl_export void mi_option_disable(mi_option_t option); +mi_decl_export void mi_option_set_enabled(mi_option_t option, bool enable); +mi_decl_export void mi_option_set_enabled_default(mi_option_t option, bool enable); + +mi_decl_nodiscard mi_decl_export long mi_option_get(mi_option_t option); +mi_decl_nodiscard mi_decl_export long mi_option_get_clamp(mi_option_t option, long min, long max); +mi_decl_nodiscard mi_decl_export size_t mi_option_get_size(mi_option_t option); +mi_decl_export void mi_option_set(mi_option_t option, long value); +mi_decl_export void mi_option_set_default(mi_option_t option, long value); + + +// ------------------------------------------------------------------------------------------------------- +// "mi" prefixed implementations of various posix, Unix, Windows, and C++ allocation functions. +// (This can be convenient when providing overrides of these functions as done in `mimalloc-override.h`.) +// note: we use `mi_cfree` as "checked free" and it checks if the pointer is in our heap before free-ing. +// ------------------------------------------------------------------------------------------------------- + +mi_decl_export void mi_cfree(void* p) mi_attr_noexcept; +mi_decl_export void* mi__expand(void* p, size_t newsize) mi_attr_noexcept; +mi_decl_nodiscard mi_decl_export size_t mi_malloc_size(const void* p) mi_attr_noexcept; +mi_decl_nodiscard mi_decl_export size_t mi_malloc_good_size(size_t size) mi_attr_noexcept; +mi_decl_nodiscard mi_decl_export size_t mi_malloc_usable_size(const void *p) mi_attr_noexcept; + +mi_decl_export int mi_posix_memalign(void** p, size_t alignment, size_t size) mi_attr_noexcept; +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_memalign(size_t alignment, size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(2) mi_attr_alloc_align(1); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_valloc(size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_pvalloc(size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_aligned_alloc(size_t alignment, size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(2) mi_attr_alloc_align(1); + +mi_decl_nodiscard mi_decl_export void* mi_reallocarray(void* p, size_t count, size_t size) mi_attr_noexcept mi_attr_alloc_size2(2,3); +mi_decl_nodiscard mi_decl_export int mi_reallocarr(void* p, size_t count, size_t size) mi_attr_noexcept; +mi_decl_nodiscard mi_decl_export void* mi_aligned_recalloc(void* p, size_t newcount, size_t size, size_t alignment) mi_attr_noexcept; +mi_decl_nodiscard mi_decl_export void* mi_aligned_offset_recalloc(void* p, size_t newcount, size_t size, size_t alignment, size_t offset) mi_attr_noexcept; + +mi_decl_nodiscard mi_decl_export mi_decl_restrict unsigned short* mi_wcsdup(const unsigned short* s) mi_attr_noexcept mi_attr_malloc; +mi_decl_nodiscard mi_decl_export mi_decl_restrict unsigned char* mi_mbsdup(const unsigned char* s) mi_attr_noexcept mi_attr_malloc; +mi_decl_export int mi_dupenv_s(char** buf, size_t* size, const char* name) mi_attr_noexcept; +mi_decl_export int mi_wdupenv_s(unsigned short** buf, size_t* size, const unsigned short* name) mi_attr_noexcept; + +mi_decl_export void mi_free_size(void* p, size_t size) mi_attr_noexcept; +mi_decl_export void mi_free_size_aligned(void* p, size_t size, size_t alignment) mi_attr_noexcept; +mi_decl_export void mi_free_aligned(void* p, size_t alignment) mi_attr_noexcept; + +// The `mi_new` wrappers implement C++ semantics on out-of-memory instead of directly returning `NULL`. +// (and call `std::get_new_handler` and potentially raise a `std::bad_alloc` exception). +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_new(size_t size) mi_attr_malloc mi_attr_alloc_size(1); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_new_aligned(size_t size, size_t alignment) mi_attr_malloc mi_attr_alloc_size(1) mi_attr_alloc_align(2); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_new_nothrow(size_t size) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_new_aligned_nothrow(size_t size, size_t alignment) mi_attr_noexcept mi_attr_malloc mi_attr_alloc_size(1) mi_attr_alloc_align(2); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_new_n(size_t count, size_t size) mi_attr_malloc mi_attr_alloc_size2(1, 2); +mi_decl_nodiscard mi_decl_export void* mi_new_realloc(void* p, size_t newsize) mi_attr_alloc_size(2); +mi_decl_nodiscard mi_decl_export void* mi_new_reallocn(void* p, size_t newcount, size_t size) mi_attr_alloc_size2(2, 3); + +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_alloc_new(mi_heap_t* heap, size_t size) mi_attr_malloc mi_attr_alloc_size(2); +mi_decl_nodiscard mi_decl_export mi_decl_restrict void* mi_heap_alloc_new_n(mi_heap_t* heap, size_t count, size_t size) mi_attr_malloc mi_attr_alloc_size2(2, 3); + +#ifdef __cplusplus +} +#endif + +// --------------------------------------------------------------------------------------------- +// Implement the C++ std::allocator interface for use in STL containers. +// (note: see `mimalloc-new-delete.h` for overriding the new/delete operators globally) +// --------------------------------------------------------------------------------------------- +#ifdef __cplusplus + +#include // std::size_t +#include // PTRDIFF_MAX +#if (__cplusplus >= 201103L) || (_MSC_VER > 1900) // C++11 +#include // std::true_type +#include // std::forward +#endif + +template struct _mi_stl_allocator_common { + typedef T value_type; + typedef std::size_t size_type; + typedef std::ptrdiff_t difference_type; + typedef value_type& reference; + typedef value_type const& const_reference; + typedef value_type* pointer; + typedef value_type const* const_pointer; + + #if ((__cplusplus >= 201103L) || (_MSC_VER > 1900)) // C++11 + using propagate_on_container_copy_assignment = std::true_type; + using propagate_on_container_move_assignment = std::true_type; + using propagate_on_container_swap = std::true_type; + template void construct(U* p, Args&& ...args) { ::new(p) U(std::forward(args)...); } + template void destroy(U* p) mi_attr_noexcept { p->~U(); } + #else + void construct(pointer p, value_type const& val) { ::new(p) value_type(val); } + void destroy(pointer p) { p->~value_type(); } + #endif + + size_type max_size() const mi_attr_noexcept { return (PTRDIFF_MAX/sizeof(value_type)); } + pointer address(reference x) const { return &x; } + const_pointer address(const_reference x) const { return &x; } +}; + +template struct mi_stl_allocator : public _mi_stl_allocator_common { + using typename _mi_stl_allocator_common::size_type; + using typename _mi_stl_allocator_common::value_type; + using typename _mi_stl_allocator_common::pointer; + template struct rebind { typedef mi_stl_allocator other; }; + + mi_stl_allocator() mi_attr_noexcept = default; + mi_stl_allocator(const mi_stl_allocator&) mi_attr_noexcept = default; + template mi_stl_allocator(const mi_stl_allocator&) mi_attr_noexcept { } + mi_stl_allocator select_on_container_copy_construction() const { return *this; } + void deallocate(T* p, size_type) { mi_free(p); } + + #if (__cplusplus >= 201703L) // C++17 + mi_decl_nodiscard T* allocate(size_type count) { return static_cast(mi_new_n(count, sizeof(T))); } + mi_decl_nodiscard T* allocate(size_type count, const void*) { return allocate(count); } + #else + mi_decl_nodiscard pointer allocate(size_type count, const void* = 0) { return static_cast(mi_new_n(count, sizeof(value_type))); } + #endif + + #if ((__cplusplus >= 201103L) || (_MSC_VER > 1900)) // C++11 + using is_always_equal = std::true_type; + #endif +}; + +template bool operator==(const mi_stl_allocator& , const mi_stl_allocator& ) mi_attr_noexcept { return true; } +template bool operator!=(const mi_stl_allocator& , const mi_stl_allocator& ) mi_attr_noexcept { return false; } + + +#if (__cplusplus >= 201103L) || (_MSC_VER >= 1900) // C++11 +#define MI_HAS_HEAP_STL_ALLOCATOR 1 + +#include // std::shared_ptr + +// Common base class for STL allocators in a specific heap +template struct _mi_heap_stl_allocator_common : public _mi_stl_allocator_common { + using typename _mi_stl_allocator_common::size_type; + using typename _mi_stl_allocator_common::value_type; + using typename _mi_stl_allocator_common::pointer; + + _mi_heap_stl_allocator_common(mi_heap_t* hp) : heap(hp, [](mi_heap_t*) {}) {} /* will not delete nor destroy the passed in heap */ + + #if (__cplusplus >= 201703L) // C++17 + mi_decl_nodiscard T* allocate(size_type count) { return static_cast(mi_heap_alloc_new_n(this->heap.get(), count, sizeof(T))); } + mi_decl_nodiscard T* allocate(size_type count, const void*) { return allocate(count); } + #else + mi_decl_nodiscard pointer allocate(size_type count, const void* = 0) { return static_cast(mi_heap_alloc_new_n(this->heap.get(), count, sizeof(value_type))); } + #endif + + #if ((__cplusplus >= 201103L) || (_MSC_VER > 1900)) // C++11 + using is_always_equal = std::false_type; + #endif + + void collect(bool force) { mi_heap_collect(this->heap.get(), force); } + template bool is_equal(const _mi_heap_stl_allocator_common& x) const { return (this->heap == x.heap); } + +protected: + std::shared_ptr heap; + template friend struct _mi_heap_stl_allocator_common; + + _mi_heap_stl_allocator_common() { + mi_heap_t* hp = mi_heap_new(); + this->heap.reset(hp, (_mi_destroy ? &heap_destroy : &heap_delete)); /* calls heap_delete/destroy when the refcount drops to zero */ + } + _mi_heap_stl_allocator_common(const _mi_heap_stl_allocator_common& x) mi_attr_noexcept : heap(x.heap) { } + template _mi_heap_stl_allocator_common(const _mi_heap_stl_allocator_common& x) mi_attr_noexcept : heap(x.heap) { } + +private: + static void heap_delete(mi_heap_t* hp) { if (hp != NULL) { mi_heap_delete(hp); } } + static void heap_destroy(mi_heap_t* hp) { if (hp != NULL) { mi_heap_destroy(hp); } } +}; + +// STL allocator allocation in a specific heap +template struct mi_heap_stl_allocator : public _mi_heap_stl_allocator_common { + using typename _mi_heap_stl_allocator_common::size_type; + mi_heap_stl_allocator() : _mi_heap_stl_allocator_common() { } // creates fresh heap that is deleted when the destructor is called + mi_heap_stl_allocator(mi_heap_t* hp) : _mi_heap_stl_allocator_common(hp) { } // no delete nor destroy on the passed in heap + template mi_heap_stl_allocator(const mi_heap_stl_allocator& x) mi_attr_noexcept : _mi_heap_stl_allocator_common(x) { } + + mi_heap_stl_allocator select_on_container_copy_construction() const { return *this; } + void deallocate(T* p, size_type) { mi_free(p); } + template struct rebind { typedef mi_heap_stl_allocator other; }; +}; + +template bool operator==(const mi_heap_stl_allocator& x, const mi_heap_stl_allocator& y) mi_attr_noexcept { return (x.is_equal(y)); } +template bool operator!=(const mi_heap_stl_allocator& x, const mi_heap_stl_allocator& y) mi_attr_noexcept { return (!x.is_equal(y)); } + + +// STL allocator allocation in a specific heap, where `free` does nothing and +// the heap is destroyed in one go on destruction -- use with care! +template struct mi_heap_destroy_stl_allocator : public _mi_heap_stl_allocator_common { + using typename _mi_heap_stl_allocator_common::size_type; + mi_heap_destroy_stl_allocator() : _mi_heap_stl_allocator_common() { } // creates fresh heap that is destroyed when the destructor is called + mi_heap_destroy_stl_allocator(mi_heap_t* hp) : _mi_heap_stl_allocator_common(hp) { } // no delete nor destroy on the passed in heap + template mi_heap_destroy_stl_allocator(const mi_heap_destroy_stl_allocator& x) mi_attr_noexcept : _mi_heap_stl_allocator_common(x) { } + + mi_heap_destroy_stl_allocator select_on_container_copy_construction() const { return *this; } + void deallocate(T*, size_type) { /* do nothing as we destroy the heap on destruct. */ } + template struct rebind { typedef mi_heap_destroy_stl_allocator other; }; +}; + +template bool operator==(const mi_heap_destroy_stl_allocator& x, const mi_heap_destroy_stl_allocator& y) mi_attr_noexcept { return (x.is_equal(y)); } +template bool operator!=(const mi_heap_destroy_stl_allocator& x, const mi_heap_destroy_stl_allocator& y) mi_attr_noexcept { return (!x.is_equal(y)); } + +#endif // C++11 + +#endif // __cplusplus + +#endif diff --git a/compat/mimalloc/mimalloc/atomic.h b/compat/mimalloc/mimalloc/atomic.h new file mode 100644 index 00000000000000..39ff5c90a194dc --- /dev/null +++ b/compat/mimalloc/mimalloc/atomic.h @@ -0,0 +1,554 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2024 Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ +#pragma once +#ifndef MIMALLOC_ATOMIC_H +#define MIMALLOC_ATOMIC_H + +// include windows.h or pthreads.h +#if defined(_WIN32) +#ifndef WIN32_LEAN_AND_MEAN +#define WIN32_LEAN_AND_MEAN +#endif +#include +#elif !defined(__wasi__) && (!defined(__EMSCRIPTEN__) || defined(__EMSCRIPTEN_PTHREADS__)) +#define MI_USE_PTHREADS +#include +#endif + +// -------------------------------------------------------------------------------------------- +// Atomics +// We need to be portable between C, C++, and MSVC. +// We base the primitives on the C/C++ atomics and create a minimal wrapper for MSVC in C compilation mode. +// This is why we try to use only `uintptr_t` and `*` as atomic types. +// To gain better insight in the range of used atomics, we use explicitly named memory order operations +// instead of passing the memory order as a parameter. +// ----------------------------------------------------------------------------------------------- + +#if defined(__cplusplus) +// Use C++ atomics +#include +#define _Atomic(tp) std::atomic +#define mi_atomic(name) std::atomic_##name +#define mi_memory_order(name) std::memory_order_##name +#if (__cplusplus >= 202002L) // c++20, see issue #571 + #define MI_ATOMIC_VAR_INIT(x) x +#elif !defined(ATOMIC_VAR_INIT) + #define MI_ATOMIC_VAR_INIT(x) x +#else + #define MI_ATOMIC_VAR_INIT(x) ATOMIC_VAR_INIT(x) +#endif +#elif defined(_MSC_VER) +// Use MSVC C wrapper for C11 atomics +#define _Atomic(tp) tp +#define MI_ATOMIC_VAR_INIT(x) x +#define mi_atomic(name) mi_atomic_##name +#define mi_memory_order(name) mi_memory_order_##name +#else +// Use C11 atomics +#include +#define mi_atomic(name) atomic_##name +#define mi_memory_order(name) memory_order_##name +#if (__STDC_VERSION__ >= 201710L) // c17, see issue #735 + #define MI_ATOMIC_VAR_INIT(x) x +#elif !defined(ATOMIC_VAR_INIT) + #define MI_ATOMIC_VAR_INIT(x) x +#else + #define MI_ATOMIC_VAR_INIT(x) ATOMIC_VAR_INIT(x) +#endif +#endif + +// Various defines for all used memory orders in mimalloc +#define mi_atomic_cas_weak(p,expected,desired,mem_success,mem_fail) \ + mi_atomic(compare_exchange_weak_explicit)(p,expected,desired,mem_success,mem_fail) + +#define mi_atomic_cas_strong(p,expected,desired,mem_success,mem_fail) \ + mi_atomic(compare_exchange_strong_explicit)(p,expected,desired,mem_success,mem_fail) + +#define mi_atomic_load_acquire(p) mi_atomic(load_explicit)(p,mi_memory_order(acquire)) +#define mi_atomic_load_relaxed(p) mi_atomic(load_explicit)(p,mi_memory_order(relaxed)) +#define mi_atomic_store_release(p,x) mi_atomic(store_explicit)(p,x,mi_memory_order(release)) +#define mi_atomic_store_relaxed(p,x) mi_atomic(store_explicit)(p,x,mi_memory_order(relaxed)) +#define mi_atomic_exchange_relaxed(p,x) mi_atomic(exchange_explicit)(p,x,mi_memory_order(relaxed)) +#define mi_atomic_exchange_release(p,x) mi_atomic(exchange_explicit)(p,x,mi_memory_order(release)) +#define mi_atomic_exchange_acq_rel(p,x) mi_atomic(exchange_explicit)(p,x,mi_memory_order(acq_rel)) +#define mi_atomic_cas_weak_release(p,exp,des) mi_atomic_cas_weak(p,exp,des,mi_memory_order(release),mi_memory_order(relaxed)) +#define mi_atomic_cas_weak_acq_rel(p,exp,des) mi_atomic_cas_weak(p,exp,des,mi_memory_order(acq_rel),mi_memory_order(acquire)) +#define mi_atomic_cas_strong_release(p,exp,des) mi_atomic_cas_strong(p,exp,des,mi_memory_order(release),mi_memory_order(relaxed)) +#define mi_atomic_cas_strong_acq_rel(p,exp,des) mi_atomic_cas_strong(p,exp,des,mi_memory_order(acq_rel),mi_memory_order(acquire)) + +#define mi_atomic_add_relaxed(p,x) mi_atomic(fetch_add_explicit)(p,x,mi_memory_order(relaxed)) +#define mi_atomic_sub_relaxed(p,x) mi_atomic(fetch_sub_explicit)(p,x,mi_memory_order(relaxed)) +#define mi_atomic_add_acq_rel(p,x) mi_atomic(fetch_add_explicit)(p,x,mi_memory_order(acq_rel)) +#define mi_atomic_sub_acq_rel(p,x) mi_atomic(fetch_sub_explicit)(p,x,mi_memory_order(acq_rel)) +#define mi_atomic_and_acq_rel(p,x) mi_atomic(fetch_and_explicit)(p,x,mi_memory_order(acq_rel)) +#define mi_atomic_or_acq_rel(p,x) mi_atomic(fetch_or_explicit)(p,x,mi_memory_order(acq_rel)) + +#define mi_atomic_increment_relaxed(p) mi_atomic_add_relaxed(p,(uintptr_t)1) +#define mi_atomic_decrement_relaxed(p) mi_atomic_sub_relaxed(p,(uintptr_t)1) +#define mi_atomic_increment_acq_rel(p) mi_atomic_add_acq_rel(p,(uintptr_t)1) +#define mi_atomic_decrement_acq_rel(p) mi_atomic_sub_acq_rel(p,(uintptr_t)1) + +static inline void mi_atomic_yield(void); +static inline intptr_t mi_atomic_addi(_Atomic(intptr_t)*p, intptr_t add); +static inline intptr_t mi_atomic_subi(_Atomic(intptr_t)*p, intptr_t sub); + + +#if defined(__cplusplus) || !defined(_MSC_VER) + +// In C++/C11 atomics we have polymorphic atomics so can use the typed `ptr` variants (where `tp` is the type of atomic value) +// We use these macros so we can provide a typed wrapper in MSVC in C compilation mode as well +#define mi_atomic_load_ptr_acquire(tp,p) mi_atomic_load_acquire(p) +#define mi_atomic_load_ptr_relaxed(tp,p) mi_atomic_load_relaxed(p) + +// In C++ we need to add casts to help resolve templates if NULL is passed +#if defined(__cplusplus) +#define mi_atomic_store_ptr_release(tp,p,x) mi_atomic_store_release(p,(tp*)x) +#define mi_atomic_store_ptr_relaxed(tp,p,x) mi_atomic_store_relaxed(p,(tp*)x) +#define mi_atomic_cas_ptr_weak_release(tp,p,exp,des) mi_atomic_cas_weak_release(p,exp,(tp*)des) +#define mi_atomic_cas_ptr_weak_acq_rel(tp,p,exp,des) mi_atomic_cas_weak_acq_rel(p,exp,(tp*)des) +#define mi_atomic_cas_ptr_strong_release(tp,p,exp,des) mi_atomic_cas_strong_release(p,exp,(tp*)des) +#define mi_atomic_exchange_ptr_relaxed(tp,p,x) mi_atomic_exchange_relaxed(p,(tp*)x) +#define mi_atomic_exchange_ptr_release(tp,p,x) mi_atomic_exchange_release(p,(tp*)x) +#define mi_atomic_exchange_ptr_acq_rel(tp,p,x) mi_atomic_exchange_acq_rel(p,(tp*)x) +#else +#define mi_atomic_store_ptr_release(tp,p,x) mi_atomic_store_release(p,x) +#define mi_atomic_store_ptr_relaxed(tp,p,x) mi_atomic_store_relaxed(p,x) +#define mi_atomic_cas_ptr_weak_release(tp,p,exp,des) mi_atomic_cas_weak_release(p,exp,des) +#define mi_atomic_cas_ptr_weak_acq_rel(tp,p,exp,des) mi_atomic_cas_weak_acq_rel(p,exp,des) +#define mi_atomic_cas_ptr_strong_release(tp,p,exp,des) mi_atomic_cas_strong_release(p,exp,des) +#define mi_atomic_exchange_ptr_relaxed(tp,p,x) mi_atomic_exchange_relaxed(p,x) +#define mi_atomic_exchange_ptr_release(tp,p,x) mi_atomic_exchange_release(p,x) +#define mi_atomic_exchange_ptr_acq_rel(tp,p,x) mi_atomic_exchange_acq_rel(p,x) +#endif + +// These are used by the statistics +static inline int64_t mi_atomic_addi64_relaxed(volatile int64_t* p, int64_t add) { + return mi_atomic(fetch_add_explicit)((_Atomic(int64_t)*)p, add, mi_memory_order(relaxed)); +} +static inline void mi_atomic_void_addi64_relaxed(volatile int64_t* p, const volatile int64_t* padd) { + const int64_t add = mi_atomic_load_relaxed((_Atomic(int64_t)*)padd); + if (add != 0) { + mi_atomic(fetch_add_explicit)((_Atomic(int64_t)*)p, add, mi_memory_order(relaxed)); + } +} +static inline void mi_atomic_maxi64_relaxed(volatile int64_t* p, int64_t x) { + int64_t current = mi_atomic_load_relaxed((_Atomic(int64_t)*)p); + while (current < x && !mi_atomic_cas_weak_release((_Atomic(int64_t)*)p, ¤t, x)) { /* nothing */ }; +} + +// Used by timers +#define mi_atomic_loadi64_acquire(p) mi_atomic(load_explicit)(p,mi_memory_order(acquire)) +#define mi_atomic_loadi64_relaxed(p) mi_atomic(load_explicit)(p,mi_memory_order(relaxed)) +#define mi_atomic_storei64_release(p,x) mi_atomic(store_explicit)(p,x,mi_memory_order(release)) +#define mi_atomic_storei64_relaxed(p,x) mi_atomic(store_explicit)(p,x,mi_memory_order(relaxed)) + +#define mi_atomic_casi64_strong_acq_rel(p,e,d) mi_atomic_cas_strong_acq_rel(p,e,d) +#define mi_atomic_addi64_acq_rel(p,i) mi_atomic_add_acq_rel(p,i) + + +#elif defined(_MSC_VER) + +// Legacy MSVC plain C compilation wrapper that uses Interlocked operations to model C11 atomics. +#include +#ifdef _WIN64 +typedef LONG64 msc_intptr_t; +#define MI_64(f) f##64 +#else +typedef LONG msc_intptr_t; +#define MI_64(f) f +#endif + +typedef enum mi_memory_order_e { + mi_memory_order_relaxed, + mi_memory_order_consume, + mi_memory_order_acquire, + mi_memory_order_release, + mi_memory_order_acq_rel, + mi_memory_order_seq_cst +} mi_memory_order; + +static inline uintptr_t mi_atomic_fetch_add_explicit(_Atomic(uintptr_t)*p, uintptr_t add, mi_memory_order mo) { + (void)(mo); + return (uintptr_t)MI_64(_InterlockedExchangeAdd)((volatile msc_intptr_t*)p, (msc_intptr_t)add); +} +static inline uintptr_t mi_atomic_fetch_sub_explicit(_Atomic(uintptr_t)*p, uintptr_t sub, mi_memory_order mo) { + (void)(mo); + return (uintptr_t)MI_64(_InterlockedExchangeAdd)((volatile msc_intptr_t*)p, -((msc_intptr_t)sub)); +} +static inline uintptr_t mi_atomic_fetch_and_explicit(_Atomic(uintptr_t)*p, uintptr_t x, mi_memory_order mo) { + (void)(mo); + return (uintptr_t)MI_64(_InterlockedAnd)((volatile msc_intptr_t*)p, (msc_intptr_t)x); +} +static inline uintptr_t mi_atomic_fetch_or_explicit(_Atomic(uintptr_t)*p, uintptr_t x, mi_memory_order mo) { + (void)(mo); + return (uintptr_t)MI_64(_InterlockedOr)((volatile msc_intptr_t*)p, (msc_intptr_t)x); +} +static inline bool mi_atomic_compare_exchange_strong_explicit(_Atomic(uintptr_t)*p, uintptr_t* expected, uintptr_t desired, mi_memory_order mo1, mi_memory_order mo2) { + (void)(mo1); (void)(mo2); + uintptr_t read = (uintptr_t)MI_64(_InterlockedCompareExchange)((volatile msc_intptr_t*)p, (msc_intptr_t)desired, (msc_intptr_t)(*expected)); + if (read == *expected) { + return true; + } + else { + *expected = read; + return false; + } +} +static inline bool mi_atomic_compare_exchange_weak_explicit(_Atomic(uintptr_t)*p, uintptr_t* expected, uintptr_t desired, mi_memory_order mo1, mi_memory_order mo2) { + return mi_atomic_compare_exchange_strong_explicit(p, expected, desired, mo1, mo2); +} +static inline uintptr_t mi_atomic_exchange_explicit(_Atomic(uintptr_t)*p, uintptr_t exchange, mi_memory_order mo) { + (void)(mo); + return (uintptr_t)MI_64(_InterlockedExchange)((volatile msc_intptr_t*)p, (msc_intptr_t)exchange); +} +static inline void mi_atomic_thread_fence(mi_memory_order mo) { + (void)(mo); + _Atomic(uintptr_t) x = 0; + mi_atomic_exchange_explicit(&x, 1, mo); +} +static inline uintptr_t mi_atomic_load_explicit(_Atomic(uintptr_t) const* p, mi_memory_order mo) { + (void)(mo); +#if defined(_M_IX86) || defined(_M_X64) + return *p; +#else + uintptr_t x = *p; + if (mo > mi_memory_order_relaxed) { + while (!mi_atomic_compare_exchange_weak_explicit((_Atomic(uintptr_t)*)p, &x, x, mo, mi_memory_order_relaxed)) { /* nothing */ }; + } + return x; +#endif +} +static inline void mi_atomic_store_explicit(_Atomic(uintptr_t)*p, uintptr_t x, mi_memory_order mo) { + (void)(mo); +#if defined(_M_IX86) || defined(_M_X64) + *p = x; +#else + mi_atomic_exchange_explicit(p, x, mo); +#endif +} +static inline int64_t mi_atomic_loadi64_explicit(_Atomic(int64_t)*p, mi_memory_order mo) { + (void)(mo); +#if defined(_M_X64) + return *p; +#else + int64_t old = *p; + int64_t x = old; + while ((old = InterlockedCompareExchange64(p, x, old)) != x) { + x = old; + } + return x; +#endif +} +static inline void mi_atomic_storei64_explicit(_Atomic(int64_t)*p, int64_t x, mi_memory_order mo) { + (void)(mo); +#if defined(x_M_IX86) || defined(_M_X64) + *p = x; +#else + InterlockedExchange64(p, x); +#endif +} + +// These are used by the statistics +static inline int64_t mi_atomic_addi64_relaxed(volatile _Atomic(int64_t)*p, int64_t add) { +#ifdef _WIN64 + return (int64_t)mi_atomic_addi((int64_t*)p, add); +#else + int64_t current; + int64_t sum; + do { + current = *p; + sum = current + add; + } while (_InterlockedCompareExchange64(p, sum, current) != current); + return current; +#endif +} +static inline void mi_atomic_void_addi64_relaxed(volatile int64_t* p, const volatile int64_t* padd) { + const int64_t add = *padd; + if (add != 0) { + mi_atomic_addi64_relaxed((volatile _Atomic(int64_t)*)p, add); + } +} + +static inline void mi_atomic_maxi64_relaxed(volatile _Atomic(int64_t)*p, int64_t x) { + int64_t current; + do { + current = *p; + } while (current < x && _InterlockedCompareExchange64(p, x, current) != current); +} + +static inline void mi_atomic_addi64_acq_rel(volatile _Atomic(int64_t*)p, int64_t i) { + mi_atomic_addi64_relaxed(p, i); +} + +static inline bool mi_atomic_casi64_strong_acq_rel(volatile _Atomic(int64_t*)p, int64_t* exp, int64_t des) { + int64_t read = _InterlockedCompareExchange64(p, des, *exp); + if (read == *exp) { + return true; + } + else { + *exp = read; + return false; + } +} + +// The pointer macros cast to `uintptr_t`. +#define mi_atomic_load_ptr_acquire(tp,p) (tp*)mi_atomic_load_acquire((_Atomic(uintptr_t)*)(p)) +#define mi_atomic_load_ptr_relaxed(tp,p) (tp*)mi_atomic_load_relaxed((_Atomic(uintptr_t)*)(p)) +#define mi_atomic_store_ptr_release(tp,p,x) mi_atomic_store_release((_Atomic(uintptr_t)*)(p),(uintptr_t)(x)) +#define mi_atomic_store_ptr_relaxed(tp,p,x) mi_atomic_store_relaxed((_Atomic(uintptr_t)*)(p),(uintptr_t)(x)) +#define mi_atomic_cas_ptr_weak_release(tp,p,exp,des) mi_atomic_cas_weak_release((_Atomic(uintptr_t)*)(p),(uintptr_t*)exp,(uintptr_t)des) +#define mi_atomic_cas_ptr_weak_acq_rel(tp,p,exp,des) mi_atomic_cas_weak_acq_rel((_Atomic(uintptr_t)*)(p),(uintptr_t*)exp,(uintptr_t)des) +#define mi_atomic_cas_ptr_strong_release(tp,p,exp,des) mi_atomic_cas_strong_release((_Atomic(uintptr_t)*)(p),(uintptr_t*)exp,(uintptr_t)des) +#define mi_atomic_exchange_ptr_relaxed(tp,p,x) (tp*)mi_atomic_exchange_relaxed((_Atomic(uintptr_t)*)(p),(uintptr_t)x) +#define mi_atomic_exchange_ptr_release(tp,p,x) (tp*)mi_atomic_exchange_release((_Atomic(uintptr_t)*)(p),(uintptr_t)x) +#define mi_atomic_exchange_ptr_acq_rel(tp,p,x) (tp*)mi_atomic_exchange_acq_rel((_Atomic(uintptr_t)*)(p),(uintptr_t)x) + +#define mi_atomic_loadi64_acquire(p) mi_atomic(loadi64_explicit)(p,mi_memory_order(acquire)) +#define mi_atomic_loadi64_relaxed(p) mi_atomic(loadi64_explicit)(p,mi_memory_order(relaxed)) +#define mi_atomic_storei64_release(p,x) mi_atomic(storei64_explicit)(p,x,mi_memory_order(release)) +#define mi_atomic_storei64_relaxed(p,x) mi_atomic(storei64_explicit)(p,x,mi_memory_order(relaxed)) + + +#endif + + +// Atomically add a signed value; returns the previous value. +static inline intptr_t mi_atomic_addi(_Atomic(intptr_t)*p, intptr_t add) { + return (intptr_t)mi_atomic_add_acq_rel((_Atomic(uintptr_t)*)p, (uintptr_t)add); +} + +// Atomically subtract a signed value; returns the previous value. +static inline intptr_t mi_atomic_subi(_Atomic(intptr_t)*p, intptr_t sub) { + return (intptr_t)mi_atomic_addi(p, -sub); +} + + +// ---------------------------------------------------------------------- +// Once and Guard +// ---------------------------------------------------------------------- + +typedef _Atomic(uintptr_t) mi_atomic_once_t; + +// Returns true only on the first invocation +static inline bool mi_atomic_once( mi_atomic_once_t* once ) { + if (mi_atomic_load_relaxed(once) != 0) return false; // quick test + uintptr_t expected = 0; + return mi_atomic_cas_strong_acq_rel(once, &expected, (uintptr_t)1); // try to set to 1 +} + +typedef _Atomic(uintptr_t) mi_atomic_guard_t; + +// Allows only one thread to execute at a time +#define mi_atomic_guard(guard) \ + uintptr_t _mi_guard_expected = 0; \ + for(bool _mi_guard_once = true; \ + _mi_guard_once && mi_atomic_cas_strong_acq_rel(guard,&_mi_guard_expected,(uintptr_t)1); \ + (mi_atomic_store_release(guard,(uintptr_t)0), _mi_guard_once = false) ) + + + +// ---------------------------------------------------------------------- +// Yield +// ---------------------------------------------------------------------- + +#if defined(__cplusplus) +#include +static inline void mi_atomic_yield(void) { + std::this_thread::yield(); +} +#elif defined(_WIN32) +static inline void mi_atomic_yield(void) { + YieldProcessor(); +} +#elif defined(__SSE2__) +#include +static inline void mi_atomic_yield(void) { + _mm_pause(); +} +#elif (defined(__GNUC__) || defined(__clang__)) && \ + (defined(__x86_64__) || defined(__i386__) || \ + defined(__aarch64__) || defined(__arm__) || \ + defined(__powerpc__) || defined(__ppc__) || defined(__PPC__) || defined(__POWERPC__)) +#if defined(__x86_64__) || defined(__i386__) +static inline void mi_atomic_yield(void) { + __asm__ volatile ("pause" ::: "memory"); +} +#elif defined(__aarch64__) +static inline void mi_atomic_yield(void) { + __asm__ volatile("wfe"); +} +#elif defined(__arm__) +#if __ARM_ARCH >= 7 +static inline void mi_atomic_yield(void) { + __asm__ volatile("yield" ::: "memory"); +} +#else +static inline void mi_atomic_yield(void) { + __asm__ volatile ("nop" ::: "memory"); +} +#endif +#elif defined(__powerpc__) || defined(__ppc__) || defined(__PPC__) || defined(__POWERPC__) +#ifdef __APPLE__ +static inline void mi_atomic_yield(void) { + __asm__ volatile ("or r27,r27,r27" ::: "memory"); +} +#else +static inline void mi_atomic_yield(void) { + __asm__ __volatile__ ("or 27,27,27" ::: "memory"); +} +#endif +#endif +#elif defined(__sun) +// Fallback for other archs +#include +static inline void mi_atomic_yield(void) { + smt_pause(); +} +#elif defined(__wasi__) +#include +static inline void mi_atomic_yield(void) { + sched_yield(); +} +#else +#include +static inline void mi_atomic_yield(void) { + sleep(0); +} +#endif + + +// ---------------------------------------------------------------------- +// Locks +// These do not have to be recursive and should be light-weight +// in-process only locks. Only used for reserving arena's and to +// maintain the abandoned list. +// ---------------------------------------------------------------------- +#if _MSC_VER +#pragma warning(disable:26110) // unlock with holding lock +#endif + +#define mi_lock(lock) for(bool _go = (mi_lock_acquire(lock),true); _go; (mi_lock_release(lock), _go=false) ) + +#if defined(_WIN32) + +#if 1 +#define mi_lock_t SRWLOCK // slim reader-writer lock + +static inline bool mi_lock_try_acquire(mi_lock_t* lock) { + return TryAcquireSRWLockExclusive(lock); +} +static inline void mi_lock_acquire(mi_lock_t* lock) { + AcquireSRWLockExclusive(lock); +} +static inline void mi_lock_release(mi_lock_t* lock) { + ReleaseSRWLockExclusive(lock); +} +static inline void mi_lock_init(mi_lock_t* lock) { + InitializeSRWLock(lock); +} +static inline void mi_lock_done(mi_lock_t* lock) { + (void)(lock); +} + +#else +#define mi_lock_t CRITICAL_SECTION + +static inline bool mi_lock_try_acquire(mi_lock_t* lock) { + return TryEnterCriticalSection(lock); +} +static inline void mi_lock_acquire(mi_lock_t* lock) { + EnterCriticalSection(lock); +} +static inline void mi_lock_release(mi_lock_t* lock) { + LeaveCriticalSection(lock); +} +static inline void mi_lock_init(mi_lock_t* lock) { + InitializeCriticalSection(lock); +} +static inline void mi_lock_done(mi_lock_t* lock) { + DeleteCriticalSection(lock); +} + +#endif + +#elif defined(MI_USE_PTHREADS) + +void _mi_error_message(int err, const char* fmt, ...); + +#define mi_lock_t pthread_mutex_t + +static inline bool mi_lock_try_acquire(mi_lock_t* lock) { + return (pthread_mutex_trylock(lock) == 0); +} +static inline void mi_lock_acquire(mi_lock_t* lock) { + const int err = pthread_mutex_lock(lock); + if (err != 0) { + _mi_error_message(err, "internal error: lock cannot be acquired\n"); + } +} +static inline void mi_lock_release(mi_lock_t* lock) { + pthread_mutex_unlock(lock); +} +static inline void mi_lock_init(mi_lock_t* lock) { + pthread_mutex_init(lock, NULL); +} +static inline void mi_lock_done(mi_lock_t* lock) { + pthread_mutex_destroy(lock); +} + +#elif defined(__cplusplus) + +#include +#define mi_lock_t std::mutex + +static inline bool mi_lock_try_acquire(mi_lock_t* lock) { + return lock->try_lock(); +} +static inline void mi_lock_acquire(mi_lock_t* lock) { + lock->lock(); +} +static inline void mi_lock_release(mi_lock_t* lock) { + lock->unlock(); +} +static inline void mi_lock_init(mi_lock_t* lock) { + (void)(lock); +} +static inline void mi_lock_done(mi_lock_t* lock) { + (void)(lock); +} + +#else + +// fall back to poor man's locks. +// this should only be the case in a single-threaded environment (like __wasi__) + +#define mi_lock_t _Atomic(uintptr_t) + +static inline bool mi_lock_try_acquire(mi_lock_t* lock) { + uintptr_t expected = 0; + return mi_atomic_cas_strong_acq_rel(lock, &expected, (uintptr_t)1); +} +static inline void mi_lock_acquire(mi_lock_t* lock) { + for (int i = 0; i < 1000; i++) { // for at most 1000 tries? + if (mi_lock_try_acquire(lock)) return; + mi_atomic_yield(); + } +} +static inline void mi_lock_release(mi_lock_t* lock) { + mi_atomic_store_release(lock, (uintptr_t)0); +} +static inline void mi_lock_init(mi_lock_t* lock) { + mi_lock_release(lock); +} +static inline void mi_lock_done(mi_lock_t* lock) { + (void)(lock); +} + +#endif + + +#endif // __MIMALLOC_ATOMIC_H diff --git a/compat/mimalloc/mimalloc/internal.h b/compat/mimalloc/mimalloc/internal.h new file mode 100644 index 00000000000000..eae85ab6e0a3f2 --- /dev/null +++ b/compat/mimalloc/mimalloc/internal.h @@ -0,0 +1,1120 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2023, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ +#pragma once +#ifndef MIMALLOC_INTERNAL_H +#define MIMALLOC_INTERNAL_H + + +// -------------------------------------------------------------------------- +// This file contains the internal API's of mimalloc and various utility +// functions and macros. +// -------------------------------------------------------------------------- + +#include "types.h" +#include "track.h" + +#if (MI_DEBUG>0) +#define mi_trace_message(...) _mi_trace_message(__VA_ARGS__) +#else +#define mi_trace_message(...) +#endif + +#define MI_CACHE_LINE 64 +#if defined(_MSC_VER) +#pragma warning(disable:4127) // suppress constant conditional warning (due to MI_SECURE paths) +#pragma warning(disable:26812) // unscoped enum warning +#define mi_decl_noinline __declspec(noinline) +#define mi_decl_thread __declspec(thread) +#define mi_decl_cache_align __declspec(align(MI_CACHE_LINE)) +#define mi_decl_weak +#define mi_decl_hidden +#elif (defined(__GNUC__) && (__GNUC__ >= 3)) || defined(__clang__) // includes clang and icc +#define mi_decl_noinline __attribute__((noinline)) +#define mi_decl_thread __thread +#define mi_decl_cache_align __attribute__((aligned(MI_CACHE_LINE))) +#define mi_decl_weak __attribute__((weak)) +#define mi_decl_hidden __attribute__((visibility("hidden"))) +#elif __cplusplus >= 201103L // c++11 +#define mi_decl_noinline +#define mi_decl_thread thread_local +#define mi_decl_cache_align alignas(MI_CACHE_LINE) +#define mi_decl_weak +#define mi_decl_hidden +#else +#define mi_decl_noinline +#define mi_decl_thread __thread // hope for the best :-) +#define mi_decl_cache_align +#define mi_decl_weak +#define mi_decl_hidden +#endif + +#if defined(__EMSCRIPTEN__) && !defined(__wasi__) +#define __wasi__ +#endif + +#if defined(__cplusplus) +#define mi_decl_externc extern "C" +#else +#define mi_decl_externc +#endif + +// "libc.c" +#include +int _mi_vsnprintf(char* buf, size_t bufsize, const char* fmt, va_list args); +int _mi_snprintf(char* buf, size_t buflen, const char* fmt, ...); +char _mi_toupper(char c); +int _mi_strnicmp(const char* s, const char* t, size_t n); +void _mi_strlcpy(char* dest, const char* src, size_t dest_size); +void _mi_strlcat(char* dest, const char* src, size_t dest_size); +size_t _mi_strlen(const char* s); +size_t _mi_strnlen(const char* s, size_t max_len); +bool _mi_getenv(const char* name, char* result, size_t result_size); + +// "options.c" +void _mi_fputs(mi_output_fun* out, void* arg, const char* prefix, const char* message); +void _mi_fprintf(mi_output_fun* out, void* arg, const char* fmt, ...); +void _mi_message(const char* fmt, ...); +void _mi_warning_message(const char* fmt, ...); +void _mi_verbose_message(const char* fmt, ...); +void _mi_trace_message(const char* fmt, ...); +void _mi_options_init(void); +long _mi_option_get_fast(mi_option_t option); +void _mi_error_message(int err, const char* fmt, ...); + +// random.c +void _mi_random_init(mi_random_ctx_t* ctx); +void _mi_random_init_weak(mi_random_ctx_t* ctx); +void _mi_random_reinit_if_weak(mi_random_ctx_t * ctx); +void _mi_random_split(mi_random_ctx_t* ctx, mi_random_ctx_t* new_ctx); +uintptr_t _mi_random_next(mi_random_ctx_t* ctx); +uintptr_t _mi_heap_random_next(mi_heap_t* heap); +uintptr_t _mi_os_random_weak(uintptr_t extra_seed); +static inline uintptr_t _mi_random_shuffle(uintptr_t x); + +// init.c +extern mi_decl_cache_align mi_stats_t _mi_stats_main; +extern mi_decl_hidden mi_decl_cache_align const mi_page_t _mi_page_empty; +void _mi_process_load(void); +void mi_cdecl _mi_process_done(void); +bool _mi_is_redirected(void); +bool _mi_allocator_init(const char** message); +void _mi_allocator_done(void); +bool _mi_is_main_thread(void); +size_t _mi_current_thread_count(void); +bool _mi_preloading(void); // true while the C runtime is not initialized yet +void _mi_thread_done(mi_heap_t* heap); +void _mi_thread_data_collect(void); +void _mi_tld_init(mi_tld_t* tld, mi_heap_t* bheap); +mi_threadid_t _mi_thread_id(void) mi_attr_noexcept; +mi_heap_t* _mi_heap_main_get(void); // statically allocated main backing heap +mi_subproc_t* _mi_subproc_from_id(mi_subproc_id_t subproc_id); +void _mi_heap_guarded_init(mi_heap_t* heap); + +// os.c +void _mi_os_init(void); // called from process init +void* _mi_os_alloc(size_t size, mi_memid_t* memid); +void _mi_os_free(void* p, size_t size, mi_memid_t memid); +void _mi_os_free_ex(void* p, size_t size, bool still_committed, mi_memid_t memid); + +size_t _mi_os_page_size(void); +size_t _mi_os_good_alloc_size(size_t size); +bool _mi_os_has_overcommit(void); +bool _mi_os_has_virtual_reserve(void); + +bool _mi_os_reset(void* addr, size_t size); +bool _mi_os_commit(void* p, size_t size, bool* is_zero); +bool _mi_os_commit_ex(void* addr, size_t size, bool* is_zero, size_t stat_size); +bool _mi_os_decommit(void* addr, size_t size); +bool _mi_os_protect(void* addr, size_t size); +bool _mi_os_unprotect(void* addr, size_t size); +bool _mi_os_purge(void* p, size_t size); +bool _mi_os_purge_ex(void* p, size_t size, bool allow_reset, size_t stat_size); + +void* _mi_os_alloc_aligned(size_t size, size_t alignment, bool commit, bool allow_large, mi_memid_t* memid); +void* _mi_os_alloc_aligned_at_offset(size_t size, size_t alignment, size_t align_offset, bool commit, bool allow_large, mi_memid_t* memid); + +void* _mi_os_get_aligned_hint(size_t try_alignment, size_t size); +bool _mi_os_use_large_page(size_t size, size_t alignment); +size_t _mi_os_large_page_size(void); + +void* _mi_os_alloc_huge_os_pages(size_t pages, int numa_node, mi_msecs_t max_secs, size_t* pages_reserved, size_t* psize, mi_memid_t* memid); + +// arena.c +mi_arena_id_t _mi_arena_id_none(void); +void _mi_arena_free(void* p, size_t size, size_t still_committed_size, mi_memid_t memid); +void* _mi_arena_alloc(size_t size, bool commit, bool allow_large, mi_arena_id_t req_arena_id, mi_memid_t* memid); +void* _mi_arena_alloc_aligned(size_t size, size_t alignment, size_t align_offset, bool commit, bool allow_large, mi_arena_id_t req_arena_id, mi_memid_t* memid); +bool _mi_arena_memid_is_suitable(mi_memid_t memid, mi_arena_id_t request_arena_id); +bool _mi_arena_contains(const void* p); +void _mi_arenas_collect(bool force_purge); +void _mi_arena_unsafe_destroy_all(void); + +bool _mi_arena_segment_clear_abandoned(mi_segment_t* segment); +void _mi_arena_segment_mark_abandoned(mi_segment_t* segment); + +void* _mi_arena_meta_zalloc(size_t size, mi_memid_t* memid); +void _mi_arena_meta_free(void* p, mi_memid_t memid, size_t size); + +typedef struct mi_arena_field_cursor_s { // abstract struct + size_t os_list_count; // max entries to visit in the OS abandoned list + size_t start; // start arena idx (may need to be wrapped) + size_t end; // end arena idx (exclusive, may need to be wrapped) + size_t bitmap_idx; // current bit idx for an arena + mi_subproc_t* subproc; // only visit blocks in this sub-process + bool visit_all; // ensure all abandoned blocks are seen (blocking) + bool hold_visit_lock; // if the subproc->abandoned_os_visit_lock is held +} mi_arena_field_cursor_t; +void _mi_arena_field_cursor_init(mi_heap_t* heap, mi_subproc_t* subproc, bool visit_all, mi_arena_field_cursor_t* current); +mi_segment_t* _mi_arena_segment_clear_abandoned_next(mi_arena_field_cursor_t* previous); +void _mi_arena_field_cursor_done(mi_arena_field_cursor_t* current); + +// "segment-map.c" +void _mi_segment_map_allocated_at(const mi_segment_t* segment); +void _mi_segment_map_freed_at(const mi_segment_t* segment); +void _mi_segment_map_unsafe_destroy(void); + +// "segment.c" +mi_page_t* _mi_segment_page_alloc(mi_heap_t* heap, size_t block_size, size_t page_alignment, mi_segments_tld_t* tld); +void _mi_segment_page_free(mi_page_t* page, bool force, mi_segments_tld_t* tld); +void _mi_segment_page_abandon(mi_page_t* page, mi_segments_tld_t* tld); +bool _mi_segment_try_reclaim_abandoned( mi_heap_t* heap, bool try_all, mi_segments_tld_t* tld); +void _mi_segment_collect(mi_segment_t* segment, bool force); + +#if MI_HUGE_PAGE_ABANDON +void _mi_segment_huge_page_free(mi_segment_t* segment, mi_page_t* page, mi_block_t* block); +#else +void _mi_segment_huge_page_reset(mi_segment_t* segment, mi_page_t* page, mi_block_t* block); +#endif + +uint8_t* _mi_segment_page_start(const mi_segment_t* segment, const mi_page_t* page, size_t* page_size); // page start for any page +void _mi_abandoned_reclaim_all(mi_heap_t* heap, mi_segments_tld_t* tld); +void _mi_abandoned_collect(mi_heap_t* heap, bool force, mi_segments_tld_t* tld); +bool _mi_segment_attempt_reclaim(mi_heap_t* heap, mi_segment_t* segment); +bool _mi_segment_visit_blocks(mi_segment_t* segment, int heap_tag, bool visit_blocks, mi_block_visit_fun* visitor, void* arg); + +// "page.c" +void* _mi_malloc_generic(mi_heap_t* heap, size_t size, bool zero, size_t huge_alignment) mi_attr_noexcept mi_attr_malloc; + +void _mi_page_retire(mi_page_t* page) mi_attr_noexcept; // free the page if there are no other pages with many free blocks +void _mi_page_unfull(mi_page_t* page); +void _mi_page_free(mi_page_t* page, mi_page_queue_t* pq, bool force); // free the page +void _mi_page_abandon(mi_page_t* page, mi_page_queue_t* pq); // abandon the page, to be picked up by another thread... +void _mi_page_force_abandon(mi_page_t* page); + +void _mi_heap_delayed_free_all(mi_heap_t* heap); +bool _mi_heap_delayed_free_partial(mi_heap_t* heap); +void _mi_heap_collect_retired(mi_heap_t* heap, bool force); + +void _mi_page_use_delayed_free(mi_page_t* page, mi_delayed_t delay, bool override_never); +bool _mi_page_try_use_delayed_free(mi_page_t* page, mi_delayed_t delay, bool override_never); +size_t _mi_page_queue_append(mi_heap_t* heap, mi_page_queue_t* pq, mi_page_queue_t* append); +void _mi_deferred_free(mi_heap_t* heap, bool force); + +void _mi_page_free_collect(mi_page_t* page,bool force); +void _mi_page_reclaim(mi_heap_t* heap, mi_page_t* page); // callback from segments + +size_t _mi_bin_size(size_t bin); // for stats +size_t _mi_bin(size_t size); // for stats + +// "heap.c" +void _mi_heap_init(mi_heap_t* heap, mi_tld_t* tld, mi_arena_id_t arena_id, bool noreclaim, uint8_t tag); +void _mi_heap_destroy_pages(mi_heap_t* heap); +void _mi_heap_collect_abandon(mi_heap_t* heap); +void _mi_heap_set_default_direct(mi_heap_t* heap); +bool _mi_heap_memid_is_suitable(mi_heap_t* heap, mi_memid_t memid); +void _mi_heap_unsafe_destroy_all(mi_heap_t* heap); +mi_heap_t* _mi_heap_by_tag(mi_heap_t* heap, uint8_t tag); +void _mi_heap_area_init(mi_heap_area_t* area, mi_page_t* page); +bool _mi_heap_area_visit_blocks(const mi_heap_area_t* area, mi_page_t* page, mi_block_visit_fun* visitor, void* arg); + +// "stats.c" +void _mi_stats_done(mi_stats_t* stats); +mi_msecs_t _mi_clock_now(void); +mi_msecs_t _mi_clock_end(mi_msecs_t start); +mi_msecs_t _mi_clock_start(void); + +// "alloc.c" +void* _mi_page_malloc_zero(mi_heap_t* heap, mi_page_t* page, size_t size, bool zero) mi_attr_noexcept; // called from `_mi_malloc_generic` +void* _mi_page_malloc(mi_heap_t* heap, mi_page_t* page, size_t size) mi_attr_noexcept; // called from `_mi_heap_malloc_aligned` +void* _mi_page_malloc_zeroed(mi_heap_t* heap, mi_page_t* page, size_t size) mi_attr_noexcept; // called from `_mi_heap_malloc_aligned` +void* _mi_heap_malloc_zero(mi_heap_t* heap, size_t size, bool zero) mi_attr_noexcept; +void* _mi_heap_malloc_zero_ex(mi_heap_t* heap, size_t size, bool zero, size_t huge_alignment) mi_attr_noexcept; // called from `_mi_heap_malloc_aligned` +void* _mi_heap_realloc_zero(mi_heap_t* heap, void* p, size_t newsize, bool zero) mi_attr_noexcept; +mi_block_t* _mi_page_ptr_unalign(const mi_page_t* page, const void* p); +bool _mi_free_delayed_block(mi_block_t* block); +void _mi_free_generic(mi_segment_t* segment, mi_page_t* page, bool is_local, void* p) mi_attr_noexcept; // for runtime integration +void _mi_padding_shrink(const mi_page_t* page, const mi_block_t* block, const size_t min_size); + +#if MI_DEBUG>1 +bool _mi_page_is_valid(mi_page_t* page); +#endif + + +// ------------------------------------------------------ +// Branches +// ------------------------------------------------------ + +#if defined(__GNUC__) || defined(__clang__) +#define mi_unlikely(x) (__builtin_expect(!!(x),false)) +#define mi_likely(x) (__builtin_expect(!!(x),true)) +#elif (defined(__cplusplus) && (__cplusplus >= 202002L)) || (defined(_MSVC_LANG) && _MSVC_LANG >= 202002L) +#define mi_unlikely(x) (x) [[unlikely]] +#define mi_likely(x) (x) [[likely]] +#else +#define mi_unlikely(x) (x) +#define mi_likely(x) (x) +#endif + +#ifndef __has_builtin +#define __has_builtin(x) 0 +#endif + + +/* ----------------------------------------------------------- + Error codes passed to `_mi_fatal_error` + All are recoverable but EFAULT is a serious error and aborts by default in secure mode. + For portability define undefined error codes using common Unix codes: + +----------------------------------------------------------- */ +#include +#ifndef EAGAIN // double free +#define EAGAIN (11) +#endif +#ifndef ENOMEM // out of memory +#define ENOMEM (12) +#endif +#ifndef EFAULT // corrupted free-list or meta-data +#define EFAULT (14) +#endif +#ifndef EINVAL // trying to free an invalid pointer +#define EINVAL (22) +#endif +#ifndef EOVERFLOW // count*size overflow +#define EOVERFLOW (75) +#endif + + +/* ----------------------------------------------------------- + Inlined definitions +----------------------------------------------------------- */ +#define MI_UNUSED(x) (void)(x) +#if (MI_DEBUG>0) +#define MI_UNUSED_RELEASE(x) +#else +#define MI_UNUSED_RELEASE(x) MI_UNUSED(x) +#endif + +#define MI_INIT4(x) x(),x(),x(),x() +#define MI_INIT8(x) MI_INIT4(x),MI_INIT4(x) +#define MI_INIT16(x) MI_INIT8(x),MI_INIT8(x) +#define MI_INIT32(x) MI_INIT16(x),MI_INIT16(x) +#define MI_INIT64(x) MI_INIT32(x),MI_INIT32(x) +#define MI_INIT128(x) MI_INIT64(x),MI_INIT64(x) +#define MI_INIT256(x) MI_INIT128(x),MI_INIT128(x) +#define MI_INIT74(x) MI_INIT64(x),MI_INIT8(x),x(),x() + +#include +// initialize a local variable to zero; use memset as compilers optimize constant sized memset's +#define _mi_memzero_var(x) memset(&x,0,sizeof(x)) + +// Is `x` a power of two? (0 is considered a power of two) +static inline bool _mi_is_power_of_two(uintptr_t x) { + return ((x & (x - 1)) == 0); +} + +// Is a pointer aligned? +static inline bool _mi_is_aligned(void* p, size_t alignment) { + mi_assert_internal(alignment != 0); + return (((uintptr_t)p % alignment) == 0); +} + +// Align upwards +static inline uintptr_t _mi_align_up(uintptr_t sz, size_t alignment) { + mi_assert_internal(alignment != 0); + uintptr_t mask = alignment - 1; + if ((alignment & mask) == 0) { // power of two? + return ((sz + mask) & ~mask); + } + else { + return (((sz + mask)/alignment)*alignment); + } +} + +// Align downwards +static inline uintptr_t _mi_align_down(uintptr_t sz, size_t alignment) { + mi_assert_internal(alignment != 0); + uintptr_t mask = alignment - 1; + if ((alignment & mask) == 0) { // power of two? + return (sz & ~mask); + } + else { + return ((sz / alignment) * alignment); + } +} + +// Align a pointer upwards +static inline void* mi_align_up_ptr(void* p, size_t alignment) { + return (void*)_mi_align_up((uintptr_t)p, alignment); +} + +// Align a pointer downwards +static inline void* mi_align_down_ptr(void* p, size_t alignment) { + return (void*)_mi_align_down((uintptr_t)p, alignment); +} + + +// Divide upwards: `s <= _mi_divide_up(s,d)*d < s+d`. +static inline uintptr_t _mi_divide_up(uintptr_t size, size_t divider) { + mi_assert_internal(divider != 0); + return (divider == 0 ? size : ((size + divider - 1) / divider)); +} + + +// clamp an integer +static inline size_t _mi_clamp(size_t sz, size_t min, size_t max) { + if (sz < min) return min; + else if (sz > max) return max; + else return sz; +} + +// Is memory zero initialized? +static inline bool mi_mem_is_zero(const void* p, size_t size) { + for (size_t i = 0; i < size; i++) { + if (((uint8_t*)p)[i] != 0) return false; + } + return true; +} + + +// Align a byte size to a size in _machine words_, +// i.e. byte size == `wsize*sizeof(void*)`. +static inline size_t _mi_wsize_from_size(size_t size) { + mi_assert_internal(size <= SIZE_MAX - sizeof(uintptr_t)); + return (size + sizeof(uintptr_t) - 1) / sizeof(uintptr_t); +} + +// Overflow detecting multiply +#if __has_builtin(__builtin_umul_overflow) || (defined(__GNUC__) && (__GNUC__ >= 5)) +#include // UINT_MAX, ULONG_MAX +#if defined(_CLOCK_T) // for Illumos +#undef _CLOCK_T +#endif +static inline bool mi_mul_overflow(size_t count, size_t size, size_t* total) { + #if (SIZE_MAX == ULONG_MAX) + return __builtin_umull_overflow(count, size, (unsigned long *)total); + #elif (SIZE_MAX == UINT_MAX) + return __builtin_umul_overflow(count, size, (unsigned int *)total); + #else + return __builtin_umulll_overflow(count, size, (unsigned long long *)total); + #endif +} +#else /* __builtin_umul_overflow is unavailable */ +static inline bool mi_mul_overflow(size_t count, size_t size, size_t* total) { + #define MI_MUL_COULD_OVERFLOW ((size_t)1 << (4*sizeof(size_t))) // sqrt(SIZE_MAX) + *total = count * size; + // note: gcc/clang optimize this to directly check the overflow flag + return ((size >= MI_MUL_COULD_OVERFLOW || count >= MI_MUL_COULD_OVERFLOW) && size > 0 && (SIZE_MAX / size) < count); +} +#endif + +// Safe multiply `count*size` into `total`; return `true` on overflow. +static inline bool mi_count_size_overflow(size_t count, size_t size, size_t* total) { + if (count==1) { // quick check for the case where count is one (common for C++ allocators) + *total = size; + return false; + } + else if mi_unlikely(mi_mul_overflow(count, size, total)) { + #if MI_DEBUG > 0 + _mi_error_message(EOVERFLOW, "allocation request is too large (%zu * %zu bytes)\n", count, size); + #endif + *total = SIZE_MAX; + return true; + } + else return false; +} + + +/*---------------------------------------------------------------------------------------- + Heap functions +------------------------------------------------------------------------------------------- */ + +extern mi_decl_hidden const mi_heap_t _mi_heap_empty; // read-only empty heap, initial value of the thread local default heap + +static inline bool mi_heap_is_backing(const mi_heap_t* heap) { + return (heap->tld->heap_backing == heap); +} + +static inline bool mi_heap_is_initialized(mi_heap_t* heap) { + mi_assert_internal(heap != NULL); + return (heap != NULL && heap != &_mi_heap_empty); +} + +static inline uintptr_t _mi_ptr_cookie(const void* p) { + extern mi_decl_hidden mi_heap_t _mi_heap_main; + mi_assert_internal(_mi_heap_main.cookie != 0); + return ((uintptr_t)p ^ _mi_heap_main.cookie); +} + +/* ----------------------------------------------------------- + Pages +----------------------------------------------------------- */ + +static inline mi_page_t* _mi_heap_get_free_small_page(mi_heap_t* heap, size_t size) { + mi_assert_internal(size <= (MI_SMALL_SIZE_MAX + MI_PADDING_SIZE)); + const size_t idx = _mi_wsize_from_size(size); + mi_assert_internal(idx < MI_PAGES_DIRECT); + return heap->pages_free_direct[idx]; +} + +// Segment that contains the pointer +// Large aligned blocks may be aligned at N*MI_SEGMENT_SIZE (inside a huge segment > MI_SEGMENT_SIZE), +// and we need align "down" to the segment info which is `MI_SEGMENT_SIZE` bytes before it; +// therefore we align one byte before `p`. +// We check for NULL afterwards on 64-bit systems to improve codegen for `mi_free`. +static inline mi_segment_t* _mi_ptr_segment(const void* p) { + mi_segment_t* const segment = (mi_segment_t*)(((uintptr_t)p - 1) & ~MI_SEGMENT_MASK); + #if MI_INTPTR_SIZE <= 4 + return (p==NULL ? NULL : segment); + #else + return ((intptr_t)segment <= 0 ? NULL : segment); + #endif +} + +static inline mi_page_t* mi_slice_to_page(mi_slice_t* s) { + mi_assert_internal(s->slice_offset== 0 && s->slice_count > 0); + return (mi_page_t*)(s); +} + +static inline mi_slice_t* mi_page_to_slice(mi_page_t* p) { + mi_assert_internal(p->slice_offset== 0 && p->slice_count > 0); + return (mi_slice_t*)(p); +} + +// Segment belonging to a page +static inline mi_segment_t* _mi_page_segment(const mi_page_t* page) { + mi_assert_internal(page!=NULL); + mi_segment_t* segment = _mi_ptr_segment(page); + mi_assert_internal(segment == NULL || ((mi_slice_t*)page >= segment->slices && (mi_slice_t*)page < segment->slices + segment->slice_entries)); + return segment; +} + +static inline mi_slice_t* mi_slice_first(const mi_slice_t* slice) { + mi_slice_t* start = (mi_slice_t*)((uint8_t*)slice - slice->slice_offset); + mi_assert_internal(start >= _mi_ptr_segment(slice)->slices); + mi_assert_internal(start->slice_offset == 0); + mi_assert_internal(start + start->slice_count > slice); + return start; +} + +// Get the page containing the pointer (performance critical as it is called in mi_free) +static inline mi_page_t* _mi_segment_page_of(const mi_segment_t* segment, const void* p) { + mi_assert_internal(p > (void*)segment); + ptrdiff_t diff = (uint8_t*)p - (uint8_t*)segment; + mi_assert_internal(diff > 0 && diff <= (ptrdiff_t)MI_SEGMENT_SIZE); + size_t idx = (size_t)diff >> MI_SEGMENT_SLICE_SHIFT; + mi_assert_internal(idx <= segment->slice_entries); + mi_slice_t* slice0 = (mi_slice_t*)&segment->slices[idx]; + mi_slice_t* slice = mi_slice_first(slice0); // adjust to the block that holds the page data + mi_assert_internal(slice->slice_offset == 0); + mi_assert_internal(slice >= segment->slices && slice < segment->slices + segment->slice_entries); + return mi_slice_to_page(slice); +} + +// Quick page start for initialized pages +static inline uint8_t* mi_page_start(const mi_page_t* page) { + mi_assert_internal(page->page_start != NULL); + mi_assert_expensive(_mi_segment_page_start(_mi_page_segment(page),page,NULL) == page->page_start); + return page->page_start; +} + +// Get the page containing the pointer +static inline mi_page_t* _mi_ptr_page(void* p) { + mi_assert_internal(p!=NULL); + return _mi_segment_page_of(_mi_ptr_segment(p), p); +} + +// Get the block size of a page (special case for huge objects) +static inline size_t mi_page_block_size(const mi_page_t* page) { + mi_assert_internal(page->block_size > 0); + return page->block_size; +} + +static inline bool mi_page_is_huge(const mi_page_t* page) { + mi_assert_internal((page->is_huge && _mi_page_segment(page)->kind == MI_SEGMENT_HUGE) || + (!page->is_huge && _mi_page_segment(page)->kind != MI_SEGMENT_HUGE)); + return page->is_huge; +} + +// Get the usable block size of a page without fixed padding. +// This may still include internal padding due to alignment and rounding up size classes. +static inline size_t mi_page_usable_block_size(const mi_page_t* page) { + return mi_page_block_size(page) - MI_PADDING_SIZE; +} + +// size of a segment +static inline size_t mi_segment_size(mi_segment_t* segment) { + return segment->segment_slices * MI_SEGMENT_SLICE_SIZE; +} + +static inline uint8_t* mi_segment_end(mi_segment_t* segment) { + return (uint8_t*)segment + mi_segment_size(segment); +} + +// Thread free access +static inline mi_block_t* mi_page_thread_free(const mi_page_t* page) { + return (mi_block_t*)(mi_atomic_load_relaxed(&((mi_page_t*)page)->xthread_free) & ~3); +} + +static inline mi_delayed_t mi_page_thread_free_flag(const mi_page_t* page) { + return (mi_delayed_t)(mi_atomic_load_relaxed(&((mi_page_t*)page)->xthread_free) & 3); +} + +// Heap access +static inline mi_heap_t* mi_page_heap(const mi_page_t* page) { + return (mi_heap_t*)(mi_atomic_load_relaxed(&((mi_page_t*)page)->xheap)); +} + +static inline void mi_page_set_heap(mi_page_t* page, mi_heap_t* heap) { + mi_assert_internal(mi_page_thread_free_flag(page) != MI_DELAYED_FREEING); + mi_atomic_store_release(&page->xheap,(uintptr_t)heap); + if (heap != NULL) { page->heap_tag = heap->tag; } +} + +// Thread free flag helpers +static inline mi_block_t* mi_tf_block(mi_thread_free_t tf) { + return (mi_block_t*)(tf & ~0x03); +} +static inline mi_delayed_t mi_tf_delayed(mi_thread_free_t tf) { + return (mi_delayed_t)(tf & 0x03); +} +static inline mi_thread_free_t mi_tf_make(mi_block_t* block, mi_delayed_t delayed) { + return (mi_thread_free_t)((uintptr_t)block | (uintptr_t)delayed); +} +static inline mi_thread_free_t mi_tf_set_delayed(mi_thread_free_t tf, mi_delayed_t delayed) { + return mi_tf_make(mi_tf_block(tf),delayed); +} +static inline mi_thread_free_t mi_tf_set_block(mi_thread_free_t tf, mi_block_t* block) { + return mi_tf_make(block, mi_tf_delayed(tf)); +} + +// are all blocks in a page freed? +// note: needs up-to-date used count, (as the `xthread_free` list may not be empty). see `_mi_page_collect_free`. +static inline bool mi_page_all_free(const mi_page_t* page) { + mi_assert_internal(page != NULL); + return (page->used == 0); +} + +// are there any available blocks? +static inline bool mi_page_has_any_available(const mi_page_t* page) { + mi_assert_internal(page != NULL && page->reserved > 0); + return (page->used < page->reserved || (mi_page_thread_free(page) != NULL)); +} + +// are there immediately available blocks, i.e. blocks available on the free list. +static inline bool mi_page_immediate_available(const mi_page_t* page) { + mi_assert_internal(page != NULL); + return (page->free != NULL); +} + +// is more than 7/8th of a page in use? +static inline bool mi_page_is_mostly_used(const mi_page_t* page) { + if (page==NULL) return true; + uint16_t frac = page->reserved / 8U; + return (page->reserved - page->used <= frac); +} + +static inline mi_page_queue_t* mi_page_queue(const mi_heap_t* heap, size_t size) { + return &((mi_heap_t*)heap)->pages[_mi_bin(size)]; +} + + + +//----------------------------------------------------------- +// Page flags +//----------------------------------------------------------- +static inline bool mi_page_is_in_full(const mi_page_t* page) { + return page->flags.x.in_full; +} + +static inline void mi_page_set_in_full(mi_page_t* page, bool in_full) { + page->flags.x.in_full = in_full; +} + +static inline bool mi_page_has_aligned(const mi_page_t* page) { + return page->flags.x.has_aligned; +} + +static inline void mi_page_set_has_aligned(mi_page_t* page, bool has_aligned) { + page->flags.x.has_aligned = has_aligned; +} + +/* ------------------------------------------------------------------- + Guarded objects +------------------------------------------------------------------- */ +#if MI_GUARDED +static inline bool mi_block_ptr_is_guarded(const mi_block_t* block, const void* p) { + const ptrdiff_t offset = (uint8_t*)p - (uint8_t*)block; + return (offset >= (ptrdiff_t)(sizeof(mi_block_t)) && block->next == MI_BLOCK_TAG_GUARDED); +} + +static inline bool mi_heap_malloc_use_guarded(mi_heap_t* heap, size_t size) { + // this code is written to result in fast assembly as it is on the hot path for allocation + const size_t count = heap->guarded_sample_count - 1; // if the rate was 0, this will underflow and count for a long time.. + if mi_likely(count != 0) { + // no sample + heap->guarded_sample_count = count; + return false; + } + else if (size >= heap->guarded_size_min && size <= heap->guarded_size_max) { + // use guarded allocation + heap->guarded_sample_count = heap->guarded_sample_rate; // reset + return (heap->guarded_sample_rate != 0); + } + else { + // failed size criteria, rewind count (but don't write to an empty heap) + if (heap->guarded_sample_rate != 0) { heap->guarded_sample_count = 1; } + return false; + } +} + +mi_decl_restrict void* _mi_heap_malloc_guarded(mi_heap_t* heap, size_t size, bool zero) mi_attr_noexcept; + +#endif + + +/* ------------------------------------------------------------------- +Encoding/Decoding the free list next pointers + +This is to protect against buffer overflow exploits where the +free list is mutated. Many hardened allocators xor the next pointer `p` +with a secret key `k1`, as `p^k1`. This prevents overwriting with known +values but might be still too weak: if the attacker can guess +the pointer `p` this can reveal `k1` (since `p^k1^p == k1`). +Moreover, if multiple blocks can be read as well, the attacker can +xor both as `(p1^k1) ^ (p2^k1) == p1^p2` which may reveal a lot +about the pointers (and subsequently `k1`). + +Instead mimalloc uses an extra key `k2` and encodes as `((p^k2)<<> (MI_INTPTR_BITS - shift)))); +} +static inline uintptr_t mi_rotr(uintptr_t x, uintptr_t shift) { + shift %= MI_INTPTR_BITS; + return (shift==0 ? x : ((x >> shift) | (x << (MI_INTPTR_BITS - shift)))); +} + +static inline void* mi_ptr_decode(const void* null, const mi_encoded_t x, const uintptr_t* keys) { + void* p = (void*)(mi_rotr(x - keys[0], keys[0]) ^ keys[1]); + return (p==null ? NULL : p); +} + +static inline mi_encoded_t mi_ptr_encode(const void* null, const void* p, const uintptr_t* keys) { + uintptr_t x = (uintptr_t)(p==NULL ? null : p); + return mi_rotl(x ^ keys[1], keys[0]) + keys[0]; +} + +static inline uint32_t mi_ptr_encode_canary(const void* null, const void* p, const uintptr_t* keys) { + const uint32_t x = (uint32_t)(mi_ptr_encode(null,p,keys)); + // make the lowest byte 0 to prevent spurious read overflows which could be a security issue (issue #951) + #ifdef MI_BIG_ENDIAN + return (x & 0x00FFFFFF); + #else + return (x & 0xFFFFFF00); + #endif +} + +static inline mi_block_t* mi_block_nextx( const void* null, const mi_block_t* block, const uintptr_t* keys ) { + mi_track_mem_defined(block,sizeof(mi_block_t)); + mi_block_t* next; + #ifdef MI_ENCODE_FREELIST + next = (mi_block_t*)mi_ptr_decode(null, block->next, keys); + #else + MI_UNUSED(keys); MI_UNUSED(null); + next = (mi_block_t*)block->next; + #endif + mi_track_mem_noaccess(block,sizeof(mi_block_t)); + return next; +} + +static inline void mi_block_set_nextx(const void* null, mi_block_t* block, const mi_block_t* next, const uintptr_t* keys) { + mi_track_mem_undefined(block,sizeof(mi_block_t)); + #ifdef MI_ENCODE_FREELIST + block->next = mi_ptr_encode(null, next, keys); + #else + MI_UNUSED(keys); MI_UNUSED(null); + block->next = (mi_encoded_t)next; + #endif + mi_track_mem_noaccess(block,sizeof(mi_block_t)); +} + +static inline mi_block_t* mi_block_next(const mi_page_t* page, const mi_block_t* block) { + #ifdef MI_ENCODE_FREELIST + mi_block_t* next = mi_block_nextx(page,block,page->keys); + // check for free list corruption: is `next` at least in the same page? + // TODO: check if `next` is `page->block_size` aligned? + if mi_unlikely(next!=NULL && !mi_is_in_same_page(block, next)) { + _mi_error_message(EFAULT, "corrupted free list entry of size %zub at %p: value 0x%zx\n", mi_page_block_size(page), block, (uintptr_t)next); + next = NULL; + } + return next; + #else + MI_UNUSED(page); + return mi_block_nextx(page,block,NULL); + #endif +} + +static inline void mi_block_set_next(const mi_page_t* page, mi_block_t* block, const mi_block_t* next) { + #ifdef MI_ENCODE_FREELIST + mi_block_set_nextx(page,block,next, page->keys); + #else + MI_UNUSED(page); + mi_block_set_nextx(page,block,next,NULL); + #endif +} + + +// ------------------------------------------------------------------- +// commit mask +// ------------------------------------------------------------------- + +static inline void mi_commit_mask_create_empty(mi_commit_mask_t* cm) { + for (size_t i = 0; i < MI_COMMIT_MASK_FIELD_COUNT; i++) { + cm->mask[i] = 0; + } +} + +static inline void mi_commit_mask_create_full(mi_commit_mask_t* cm) { + for (size_t i = 0; i < MI_COMMIT_MASK_FIELD_COUNT; i++) { + cm->mask[i] = ~((size_t)0); + } +} + +static inline bool mi_commit_mask_is_empty(const mi_commit_mask_t* cm) { + for (size_t i = 0; i < MI_COMMIT_MASK_FIELD_COUNT; i++) { + if (cm->mask[i] != 0) return false; + } + return true; +} + +static inline bool mi_commit_mask_is_full(const mi_commit_mask_t* cm) { + for (size_t i = 0; i < MI_COMMIT_MASK_FIELD_COUNT; i++) { + if (cm->mask[i] != ~((size_t)0)) return false; + } + return true; +} + +// defined in `segment.c`: +size_t _mi_commit_mask_committed_size(const mi_commit_mask_t* cm, size_t total); +size_t _mi_commit_mask_next_run(const mi_commit_mask_t* cm, size_t* idx); + +#define mi_commit_mask_foreach(cm,idx,count) \ + idx = 0; \ + while ((count = _mi_commit_mask_next_run(cm,&idx)) > 0) { + +#define mi_commit_mask_foreach_end() \ + idx += count; \ + } + + + +/* ----------------------------------------------------------- + memory id's +----------------------------------------------------------- */ + +static inline mi_memid_t _mi_memid_create(mi_memkind_t memkind) { + mi_memid_t memid; + _mi_memzero_var(memid); + memid.memkind = memkind; + return memid; +} + +static inline mi_memid_t _mi_memid_none(void) { + return _mi_memid_create(MI_MEM_NONE); +} + +static inline mi_memid_t _mi_memid_create_os(bool committed, bool is_zero, bool is_large) { + mi_memid_t memid = _mi_memid_create(MI_MEM_OS); + memid.initially_committed = committed; + memid.initially_zero = is_zero; + memid.is_pinned = is_large; + return memid; +} + + +// ------------------------------------------------------------------- +// Fast "random" shuffle +// ------------------------------------------------------------------- + +static inline uintptr_t _mi_random_shuffle(uintptr_t x) { + if (x==0) { x = 17; } // ensure we don't get stuck in generating zeros +#if (MI_INTPTR_SIZE>=8) + // by Sebastiano Vigna, see: + x ^= x >> 30; + x *= 0xbf58476d1ce4e5b9UL; + x ^= x >> 27; + x *= 0x94d049bb133111ebUL; + x ^= x >> 31; +#elif (MI_INTPTR_SIZE==4) + // by Chris Wellons, see: + x ^= x >> 16; + x *= 0x7feb352dUL; + x ^= x >> 15; + x *= 0x846ca68bUL; + x ^= x >> 16; +#endif + return x; +} + +// ------------------------------------------------------------------- +// Optimize numa node access for the common case (= one node) +// ------------------------------------------------------------------- + +int _mi_os_numa_node_get(void); +size_t _mi_os_numa_node_count_get(void); + +extern mi_decl_hidden _Atomic(size_t) _mi_numa_node_count; +static inline int _mi_os_numa_node(void) { + if mi_likely(mi_atomic_load_relaxed(&_mi_numa_node_count) == 1) { return 0; } + else return _mi_os_numa_node_get(); +} +static inline size_t _mi_os_numa_node_count(void) { + const size_t count = mi_atomic_load_relaxed(&_mi_numa_node_count); + if mi_likely(count > 0) { return count; } + else return _mi_os_numa_node_count_get(); +} + + + +// ----------------------------------------------------------------------- +// Count bits: trailing or leading zeros (with MI_INTPTR_BITS on all zero) +// ----------------------------------------------------------------------- + +#if defined(__GNUC__) + +#include // LONG_MAX +#define MI_HAVE_FAST_BITSCAN +static inline size_t mi_clz(size_t x) { + if (x==0) return MI_SIZE_BITS; + #if (SIZE_MAX == ULONG_MAX) + return __builtin_clzl(x); + #else + return __builtin_clzll(x); + #endif +} +static inline size_t mi_ctz(size_t x) { + if (x==0) return MI_SIZE_BITS; + #if (SIZE_MAX == ULONG_MAX) + return __builtin_ctzl(x); + #else + return __builtin_ctzll(x); + #endif +} + +#elif defined(_MSC_VER) + +#include // LONG_MAX +#include // BitScanReverse64 +#define MI_HAVE_FAST_BITSCAN +static inline size_t mi_clz(size_t x) { + if (x==0) return MI_SIZE_BITS; + unsigned long idx; + #if (SIZE_MAX == ULONG_MAX) + _BitScanReverse(&idx, x); + #else + _BitScanReverse64(&idx, x); + #endif + return ((MI_SIZE_BITS - 1) - idx); +} +static inline size_t mi_ctz(size_t x) { + if (x==0) return MI_SIZE_BITS; + unsigned long idx; + #if (SIZE_MAX == ULONG_MAX) + _BitScanForward(&idx, x); + #else + _BitScanForward64(&idx, x); + #endif + return idx; +} + +#else + +static inline size_t mi_ctz_generic32(uint32_t x) { + // de Bruijn multiplication, see + static const uint8_t debruijn[32] = { + 0, 1, 28, 2, 29, 14, 24, 3, 30, 22, 20, 15, 25, 17, 4, 8, + 31, 27, 13, 23, 21, 19, 16, 7, 26, 12, 18, 6, 11, 5, 10, 9 + }; + if (x==0) return 32; + return debruijn[(uint32_t)((x & -(int32_t)x) * (uint32_t)(0x077CB531U)) >> 27]; +} + +static inline size_t mi_clz_generic32(uint32_t x) { + // de Bruijn multiplication, see + static const uint8_t debruijn[32] = { + 31, 22, 30, 21, 18, 10, 29, 2, 20, 17, 15, 13, 9, 6, 28, 1, + 23, 19, 11, 3, 16, 14, 7, 24, 12, 4, 8, 25, 5, 26, 27, 0 + }; + if (x==0) return 32; + x |= x >> 1; + x |= x >> 2; + x |= x >> 4; + x |= x >> 8; + x |= x >> 16; + return debruijn[(uint32_t)(x * (uint32_t)(0x07C4ACDDU)) >> 27]; +} + +static inline size_t mi_ctz(size_t x) { + if (x==0) return MI_SIZE_BITS; + #if (MI_SIZE_BITS <= 32) + return mi_ctz_generic32((uint32_t)x); + #else + const uint32_t lo = (uint32_t)x; + if (lo != 0) { + return mi_ctz_generic32(lo); + } + else { + return (32 + mi_ctz_generic32((uint32_t)(x>>32))); + } + #endif +} + +static inline size_t mi_clz(size_t x) { + if (x==0) return MI_SIZE_BITS; + #if (MI_SIZE_BITS <= 32) + return mi_clz_generic32((uint32_t)x); + #else + const uint32_t hi = (uint32_t)(x>>32); + if (hi != 0) { + return mi_clz_generic32(hi); + } + else { + return 32 + mi_clz_generic32((uint32_t)x); + } + #endif +} + +#endif + +// "bit scan reverse": Return index of the highest bit (or MI_SIZE_BITS if `x` is zero) +static inline size_t mi_bsr(size_t x) { + return (x==0 ? MI_SIZE_BITS : MI_SIZE_BITS - 1 - mi_clz(x)); +} + +size_t _mi_popcount_generic(size_t x); + +static inline size_t mi_popcount(size_t x) { + if (x<=1) return x; + if (x==SIZE_MAX) return MI_SIZE_BITS; + #if defined(__GNUC__) + #if (SIZE_MAX == ULONG_MAX) + return __builtin_popcountl(x); + #else + return __builtin_popcountll(x); + #endif + #else + return _mi_popcount_generic(x); + #endif +} + +// --------------------------------------------------------------------------------- +// Provide our own `_mi_memcpy` for potential performance optimizations. +// +// For now, only on Windows with msvc/clang-cl we optimize to `rep movsb` if +// we happen to run on x86/x64 cpu's that have "fast short rep movsb" (FSRM) support +// (AMD Zen3+ (~2020) or Intel Ice Lake+ (~2017). See also issue #201 and pr #253. +// --------------------------------------------------------------------------------- + +#if !MI_TRACK_ENABLED && defined(_WIN32) && (defined(_M_IX86) || defined(_M_X64)) +#include +extern bool _mi_cpu_has_fsrm; +extern bool _mi_cpu_has_erms; +static inline void _mi_memcpy(void* dst, const void* src, size_t n) { + if ((_mi_cpu_has_fsrm && n <= 128) || (_mi_cpu_has_erms && n > 128)) { + __movsb((unsigned char*)dst, (const unsigned char*)src, n); + } + else { + memcpy(dst, src, n); + } +} +static inline void _mi_memzero(void* dst, size_t n) { + if ((_mi_cpu_has_fsrm && n <= 128) || (_mi_cpu_has_erms && n > 128)) { + __stosb((unsigned char*)dst, 0, n); + } + else { + memset(dst, 0, n); + } +} +#else +static inline void _mi_memcpy(void* dst, const void* src, size_t n) { + memcpy(dst, src, n); +} +static inline void _mi_memzero(void* dst, size_t n) { + memset(dst, 0, n); +} +#endif + +// ------------------------------------------------------------------------------- +// The `_mi_memcpy_aligned` can be used if the pointers are machine-word aligned +// This is used for example in `mi_realloc`. +// ------------------------------------------------------------------------------- + +#if (defined(__GNUC__) && (__GNUC__ >= 4)) || defined(__clang__) +// On GCC/CLang we provide a hint that the pointers are word aligned. +static inline void _mi_memcpy_aligned(void* dst, const void* src, size_t n) { + mi_assert_internal(((uintptr_t)dst % MI_INTPTR_SIZE == 0) && ((uintptr_t)src % MI_INTPTR_SIZE == 0)); + void* adst = __builtin_assume_aligned(dst, MI_INTPTR_SIZE); + const void* asrc = __builtin_assume_aligned(src, MI_INTPTR_SIZE); + _mi_memcpy(adst, asrc, n); +} + +static inline void _mi_memzero_aligned(void* dst, size_t n) { + mi_assert_internal((uintptr_t)dst % MI_INTPTR_SIZE == 0); + void* adst = __builtin_assume_aligned(dst, MI_INTPTR_SIZE); + _mi_memzero(adst, n); +} +#else +// Default fallback on `_mi_memcpy` +static inline void _mi_memcpy_aligned(void* dst, const void* src, size_t n) { + mi_assert_internal(((uintptr_t)dst % MI_INTPTR_SIZE == 0) && ((uintptr_t)src % MI_INTPTR_SIZE == 0)); + _mi_memcpy(dst, src, n); +} + +static inline void _mi_memzero_aligned(void* dst, size_t n) { + mi_assert_internal((uintptr_t)dst % MI_INTPTR_SIZE == 0); + _mi_memzero(dst, n); +} +#endif + + +#endif diff --git a/compat/mimalloc/mimalloc/prim.h b/compat/mimalloc/mimalloc/prim.h new file mode 100644 index 00000000000000..bddd66e9465c28 --- /dev/null +++ b/compat/mimalloc/mimalloc/prim.h @@ -0,0 +1,419 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2024, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ +#pragma once +#ifndef MIMALLOC_PRIM_H +#define MIMALLOC_PRIM_H + + +// -------------------------------------------------------------------------- +// This file specifies the primitive portability API. +// Each OS/host needs to implement these primitives, see `src/prim` +// for implementations on Window, macOS, WASI, and Linux/Unix. +// +// note: on all primitive functions, we always have result parameters != NULL, and: +// addr != NULL and page aligned +// size > 0 and page aligned +// the return value is an error code as an `int` where 0 is success +// -------------------------------------------------------------------------- + +// OS memory configuration +typedef struct mi_os_mem_config_s { + size_t page_size; // default to 4KiB + size_t large_page_size; // 0 if not supported, usually 2MiB (4MiB on Windows) + size_t alloc_granularity; // smallest allocation size (usually 4KiB, on Windows 64KiB) + size_t physical_memory_in_kib; // physical memory size in KiB + size_t virtual_address_bits; // usually 48 or 56 bits on 64-bit systems. (used to determine secure randomization) + bool has_overcommit; // can we reserve more memory than can be actually committed? + bool has_partial_free; // can allocated blocks be freed partially? (true for mmap, false for VirtualAlloc) + bool has_virtual_reserve; // supports virtual address space reservation? (if true we can reserve virtual address space without using commit or physical memory) +} mi_os_mem_config_t; + +// Initialize +void _mi_prim_mem_init( mi_os_mem_config_t* config ); + +// Free OS memory +int _mi_prim_free(void* addr, size_t size ); + +// Allocate OS memory. Return NULL on error. +// The `try_alignment` is just a hint and the returned pointer does not have to be aligned. +// If `commit` is false, the virtual memory range only needs to be reserved (with no access) +// which will later be committed explicitly using `_mi_prim_commit`. +// `is_zero` is set to true if the memory was zero initialized (as on most OS's) +// The `hint_addr` address is either `NULL` or a preferred allocation address but can be ignored. +// pre: !commit => !allow_large +// try_alignment >= _mi_os_page_size() and a power of 2 +int _mi_prim_alloc(void* hint_addr, size_t size, size_t try_alignment, bool commit, bool allow_large, bool* is_large, bool* is_zero, void** addr); + +// Commit memory. Returns error code or 0 on success. +// For example, on Linux this would make the memory PROT_READ|PROT_WRITE. +// `is_zero` is set to true if the memory was zero initialized (e.g. on Windows) +int _mi_prim_commit(void* addr, size_t size, bool* is_zero); + +// Decommit memory. Returns error code or 0 on success. The `needs_recommit` result is true +// if the memory would need to be re-committed. For example, on Windows this is always true, +// but on Linux we could use MADV_DONTNEED to decommit which does not need a recommit. +// pre: needs_recommit != NULL +int _mi_prim_decommit(void* addr, size_t size, bool* needs_recommit); + +// Reset memory. The range keeps being accessible but the content might be reset. +// Returns error code or 0 on success. +int _mi_prim_reset(void* addr, size_t size); + +// Protect memory. Returns error code or 0 on success. +int _mi_prim_protect(void* addr, size_t size, bool protect); + +// Allocate huge (1GiB) pages possibly associated with a NUMA node. +// `is_zero` is set to true if the memory was zero initialized (as on most OS's) +// pre: size > 0 and a multiple of 1GiB. +// numa_node is either negative (don't care), or a numa node number. +int _mi_prim_alloc_huge_os_pages(void* hint_addr, size_t size, int numa_node, bool* is_zero, void** addr); + +// Return the current NUMA node +size_t _mi_prim_numa_node(void); + +// Return the number of logical NUMA nodes +size_t _mi_prim_numa_node_count(void); + +// Clock ticks +mi_msecs_t _mi_prim_clock_now(void); + +// Return process information (only for statistics) +typedef struct mi_process_info_s { + mi_msecs_t elapsed; + mi_msecs_t utime; + mi_msecs_t stime; + size_t current_rss; + size_t peak_rss; + size_t current_commit; + size_t peak_commit; + size_t page_faults; +} mi_process_info_t; + +void _mi_prim_process_info(mi_process_info_t* pinfo); + +// Default stderr output. (only for warnings etc. with verbose enabled) +// msg != NULL && _mi_strlen(msg) > 0 +void _mi_prim_out_stderr( const char* msg ); + +// Get an environment variable. (only for options) +// name != NULL, result != NULL, result_size >= 64 +bool _mi_prim_getenv(const char* name, char* result, size_t result_size); + + +// Fill a buffer with strong randomness; return `false` on error or if +// there is no strong randomization available. +bool _mi_prim_random_buf(void* buf, size_t buf_len); + +// Called on the first thread start, and should ensure `_mi_thread_done` is called on thread termination. +void _mi_prim_thread_init_auto_done(void); + +// Called on process exit and may take action to clean up resources associated with the thread auto done. +void _mi_prim_thread_done_auto_done(void); + +// Called when the default heap for a thread changes +void _mi_prim_thread_associate_default_heap(mi_heap_t* heap); + + + + + +//------------------------------------------------------------------- +// Access to TLS (thread local storage) slots. +// We need fast access to both a unique thread id (in `free.c:mi_free`) and +// to a thread-local heap pointer (in `alloc.c:mi_malloc`). +// To achieve this we use specialized code for various platforms. +//------------------------------------------------------------------- + +// On some libc + platform combinations we can directly access a thread-local storage (TLS) slot. +// The TLS layout depends on both the OS and libc implementation so we use specific tests for each main platform. +// If you test on another platform and it works please send a PR :-) +// see also https://akkadia.org/drepper/tls.pdf for more info on the TLS register. +// +// Note: we would like to prefer `__builtin_thread_pointer()` nowadays instead of using assembly, +// but unfortunately we can not detect support reliably (see issue #883) +// We also use it on Apple OS as we use a TLS slot for the default heap there. +#if defined(__GNUC__) && ( \ + (defined(__GLIBC__) && (defined(__x86_64__) || defined(__i386__) || (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__))) \ + || (defined(__APPLE__) && (defined(__x86_64__) || defined(__aarch64__) || defined(__POWERPC__))) \ + || (defined(__BIONIC__) && (defined(__x86_64__) || defined(__i386__) || (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__))) \ + || (defined(__FreeBSD__) && (defined(__x86_64__) || defined(__i386__) || defined(__aarch64__))) \ + || (defined(__OpenBSD__) && (defined(__x86_64__) || defined(__i386__) || defined(__aarch64__))) \ + ) + +#define MI_HAS_TLS_SLOT 1 + +static inline void* mi_prim_tls_slot(size_t slot) mi_attr_noexcept { + void* res; + const size_t ofs = (slot*sizeof(void*)); + #if defined(__i386__) + __asm__("movl %%gs:%1, %0" : "=r" (res) : "m" (*((void**)ofs)) : ); // x86 32-bit always uses GS + #elif defined(__APPLE__) && defined(__x86_64__) + __asm__("movq %%gs:%1, %0" : "=r" (res) : "m" (*((void**)ofs)) : ); // x86_64 macOSX uses GS + #elif defined(__x86_64__) && (MI_INTPTR_SIZE==4) + __asm__("movl %%fs:%1, %0" : "=r" (res) : "m" (*((void**)ofs)) : ); // x32 ABI + #elif defined(__x86_64__) + __asm__("movq %%fs:%1, %0" : "=r" (res) : "m" (*((void**)ofs)) : ); // x86_64 Linux, BSD uses FS + #elif defined(__arm__) + void** tcb; MI_UNUSED(ofs); + __asm__ volatile ("mrc p15, 0, %0, c13, c0, 3\nbic %0, %0, #3" : "=r" (tcb)); + res = tcb[slot]; + #elif defined(__aarch64__) + void** tcb; MI_UNUSED(ofs); + #if defined(__APPLE__) // M1, issue #343 + __asm__ volatile ("mrs %0, tpidrro_el0\nbic %0, %0, #7" : "=r" (tcb)); + #else + __asm__ volatile ("mrs %0, tpidr_el0" : "=r" (tcb)); + #endif + res = tcb[slot]; + #elif defined(__APPLE__) && defined(__POWERPC__) // ppc, issue #781 + MI_UNUSED(ofs); + res = pthread_getspecific(slot); + #endif + return res; +} + +// setting a tls slot is only used on macOS for now +static inline void mi_prim_tls_slot_set(size_t slot, void* value) mi_attr_noexcept { + const size_t ofs = (slot*sizeof(void*)); + #if defined(__i386__) + __asm__("movl %1,%%gs:%0" : "=m" (*((void**)ofs)) : "rn" (value) : ); // 32-bit always uses GS + #elif defined(__APPLE__) && defined(__x86_64__) + __asm__("movq %1,%%gs:%0" : "=m" (*((void**)ofs)) : "rn" (value) : ); // x86_64 macOS uses GS + #elif defined(__x86_64__) && (MI_INTPTR_SIZE==4) + __asm__("movl %1,%%fs:%0" : "=m" (*((void**)ofs)) : "rn" (value) : ); // x32 ABI + #elif defined(__x86_64__) + __asm__("movq %1,%%fs:%0" : "=m" (*((void**)ofs)) : "rn" (value) : ); // x86_64 Linux, BSD uses FS + #elif defined(__arm__) + void** tcb; MI_UNUSED(ofs); + __asm__ volatile ("mrc p15, 0, %0, c13, c0, 3\nbic %0, %0, #3" : "=r" (tcb)); + tcb[slot] = value; + #elif defined(__aarch64__) + void** tcb; MI_UNUSED(ofs); + #if defined(__APPLE__) // M1, issue #343 + __asm__ volatile ("mrs %0, tpidrro_el0\nbic %0, %0, #7" : "=r" (tcb)); + #else + __asm__ volatile ("mrs %0, tpidr_el0" : "=r" (tcb)); + #endif + tcb[slot] = value; + #elif defined(__APPLE__) && defined(__POWERPC__) // ppc, issue #781 + MI_UNUSED(ofs); + pthread_setspecific(slot, value); + #endif +} + +#elif _WIN32 && MI_WIN_USE_FIXED_TLS && !defined(MI_WIN_USE_FLS) + +// On windows we can store the thread-local heap at a fixed TLS slot to avoid +// thread-local initialization checks in the fast path. This uses a fixed location +// in the TCB though (last user-reserved slot by default) which may clash with other applications. + +#define MI_HAS_TLS_SLOT 2 // 2 = we can reliably initialize the slot (saving a test on each malloc) + +#if MI_WIN_USE_FIXED_TLS > 1 +#define MI_TLS_SLOT (MI_WIN_USE_FIXED_TLS) +#elif MI_SIZE_SIZE == 4 +#define MI_TLS_SLOT (0x710) // Last user-reserved slot +// #define MI_TLS_SLOT (0xF0C) // Last TlsSlot (might clash with other app reserved slot) +#else +#define MI_TLS_SLOT (0x888) // Last user-reserved slot +// #define MI_TLS_SLOT (0x1678) // Last TlsSlot (might clash with other app reserved slot) +#endif + +static inline void* mi_prim_tls_slot(size_t slot) mi_attr_noexcept { + #if (_M_X64 || _M_AMD64) && !defined(_M_ARM64EC) + return (void*)__readgsqword((unsigned long)slot); // direct load at offset from gs + #elif _M_IX86 && !defined(_M_ARM64EC) + return (void*)__readfsdword((unsigned long)slot); // direct load at offset from fs + #else + return ((void**)NtCurrentTeb())[slot / sizeof(void*)]; + #endif +} +static inline void mi_prim_tls_slot_set(size_t slot, void* value) mi_attr_noexcept { + ((void**)NtCurrentTeb())[slot / sizeof(void*)] = value; +} + +#endif + + + +//------------------------------------------------------------------- +// Get a fast unique thread id. +// +// Getting the thread id should be performant as it is called in the +// fast path of `_mi_free` and we specialize for various platforms as +// inlined definitions. Regular code should call `init.c:_mi_thread_id()`. +// We only require _mi_prim_thread_id() to return a unique id +// for each thread (unequal to zero). +//------------------------------------------------------------------- + + +// Do we have __builtin_thread_pointer? This would be the preferred way to get a unique thread id +// but unfortunately, it seems we cannot test for this reliably at this time (see issue #883) +// Nevertheless, it seems needed on older graviton platforms (see issue #851). +// For now, we only enable this for specific platforms. +#if !defined(__APPLE__) /* on apple (M1) the wrong register is read (tpidr_el0 instead of tpidrro_el0) so fall back to TLS slot assembly ()*/ \ + && !defined(__CYGWIN__) \ + && !defined(MI_LIBC_MUSL) \ + && (!defined(__clang_major__) || __clang_major__ >= 14) /* older clang versions emit bad code; fall back to using the TLS slot () */ + #if (defined(__GNUC__) && (__GNUC__ >= 7) && defined(__aarch64__)) /* aarch64 for older gcc versions (issue #851) */ \ + || (defined(__GNUC__) && (__GNUC__ >= 11) && defined(__x86_64__)) \ + || (defined(__clang_major__) && (__clang_major__ >= 14) && (defined(__aarch64__) || defined(__x86_64__))) + #define MI_USE_BUILTIN_THREAD_POINTER 1 + #endif +#endif + + + +// defined in `init.c`; do not use these directly +extern mi_decl_thread mi_heap_t* _mi_heap_default; // default heap to allocate from +extern bool _mi_process_is_initialized; // has mi_process_init been called? + +static inline mi_threadid_t _mi_prim_thread_id(void) mi_attr_noexcept; + +// Get a unique id for the current thread. +#if defined(MI_PRIM_THREAD_ID) + +static inline mi_threadid_t _mi_prim_thread_id(void) mi_attr_noexcept { + return MI_PRIM_THREAD_ID(); // used for example by CPython for a free threaded build (see python/cpython#115488) +} + +#elif defined(_WIN32) + +static inline mi_threadid_t _mi_prim_thread_id(void) mi_attr_noexcept { + // Windows: works on Intel and ARM in both 32- and 64-bit + return (uintptr_t)NtCurrentTeb(); +} + +#elif MI_USE_BUILTIN_THREAD_POINTER + +static inline mi_threadid_t _mi_prim_thread_id(void) mi_attr_noexcept { + // Works on most Unix based platforms with recent compilers + return (uintptr_t)__builtin_thread_pointer(); +} + +#elif MI_HAS_TLS_SLOT + +static inline mi_threadid_t _mi_prim_thread_id(void) mi_attr_noexcept { + #if defined(__BIONIC__) + // issue #384, #495: on the Bionic libc (Android), slot 1 is the thread id + // see: https://github.com/aosp-mirror/platform_bionic/blob/c44b1d0676ded732df4b3b21c5f798eacae93228/libc/platform/bionic/tls_defines.h#L86 + return (uintptr_t)mi_prim_tls_slot(1); + #else + // in all our other targets, slot 0 is the thread id + // glibc: https://sourceware.org/git/?p=glibc.git;a=blob_plain;f=sysdeps/x86_64/nptl/tls.h + // apple: https://github.com/apple/darwin-xnu/blob/main/libsyscall/os/tsd.h#L36 + return (uintptr_t)mi_prim_tls_slot(0); + #endif +} + +#else + +// otherwise use portable C, taking the address of a thread local variable (this is still very fast on most platforms). +static inline mi_threadid_t _mi_prim_thread_id(void) mi_attr_noexcept { + return (uintptr_t)&_mi_heap_default; +} + +#endif + + + +/* ---------------------------------------------------------------------------------------- +Get the thread local default heap: `_mi_prim_get_default_heap()` + +This is inlined here as it is on the fast path for allocation functions. + +On most platforms (Windows, Linux, FreeBSD, NetBSD, etc), this just returns a +__thread local variable (`_mi_heap_default`). With the initial-exec TLS model this ensures +that the storage will always be available (allocated on the thread stacks). + +On some platforms though we cannot use that when overriding `malloc` since the underlying +TLS implementation (or the loader) will call itself `malloc` on a first access and recurse. +We try to circumvent this in an efficient way: +- macOSX : we use an unused TLS slot from the OS allocated slots (MI_TLS_SLOT). On OSX, the + loader itself calls `malloc` even before the modules are initialized. +- OpenBSD: we use an unused slot from the pthread block (MI_TLS_PTHREAD_SLOT_OFS). +- DragonFly: defaults are working but seem slow compared to freeBSD (see PR #323) +------------------------------------------------------------------------------------------- */ + +static inline mi_heap_t* mi_prim_get_default_heap(void); + +#if defined(MI_MALLOC_OVERRIDE) +#if defined(__APPLE__) // macOS + #define MI_TLS_SLOT 89 // seems unused? + // other possible unused ones are 9, 29, __PTK_FRAMEWORK_JAVASCRIPTCORE_KEY4 (94), __PTK_FRAMEWORK_GC_KEY9 (112) and __PTK_FRAMEWORK_OLDGC_KEY9 (89) + // see +#elif defined(__OpenBSD__) + // use end bytes of a name; goes wrong if anyone uses names > 23 characters (ptrhread specifies 16) + // see + #define MI_TLS_PTHREAD_SLOT_OFS (6*sizeof(int) + 4*sizeof(void*) + 24) + // #elif defined(__DragonFly__) + // #warning "mimalloc is not working correctly on DragonFly yet." + // #define MI_TLS_PTHREAD_SLOT_OFS (4 + 1*sizeof(void*)) // offset `uniqueid` (also used by gdb?) +#elif defined(__ANDROID__) + // See issue #381 + #define MI_TLS_PTHREAD +#endif +#endif + + +#if MI_TLS_SLOT +# if !defined(MI_HAS_TLS_SLOT) +# error "trying to use a TLS slot for the default heap, but the mi_prim_tls_slot primitives are not defined" +# endif + +static inline mi_heap_t* mi_prim_get_default_heap(void) { + mi_heap_t* heap = (mi_heap_t*)mi_prim_tls_slot(MI_TLS_SLOT); + #if MI_HAS_TLS_SLOT == 1 // check if the TLS slot is initialized + if mi_unlikely(heap == NULL) { + #ifdef __GNUC__ + __asm(""); // prevent conditional load of the address of _mi_heap_empty + #endif + heap = (mi_heap_t*)&_mi_heap_empty; + } + #endif + return heap; +} + +#elif defined(MI_TLS_PTHREAD_SLOT_OFS) + +static inline mi_heap_t** mi_prim_tls_pthread_heap_slot(void) { + pthread_t self = pthread_self(); + #if defined(__DragonFly__) + if (self==NULL) return NULL; + #endif + return (mi_heap_t**)((uint8_t*)self + MI_TLS_PTHREAD_SLOT_OFS); +} + +static inline mi_heap_t* mi_prim_get_default_heap(void) { + mi_heap_t** pheap = mi_prim_tls_pthread_heap_slot(); + if mi_unlikely(pheap == NULL) return _mi_heap_main_get(); + mi_heap_t* heap = *pheap; + if mi_unlikely(heap == NULL) return (mi_heap_t*)&_mi_heap_empty; + return heap; +} + +#elif defined(MI_TLS_PTHREAD) + +extern pthread_key_t _mi_heap_default_key; +static inline mi_heap_t* mi_prim_get_default_heap(void) { + mi_heap_t* heap = (mi_unlikely(_mi_heap_default_key == (pthread_key_t)(-1)) ? _mi_heap_main_get() : (mi_heap_t*)pthread_getspecific(_mi_heap_default_key)); + return (mi_unlikely(heap == NULL) ? (mi_heap_t*)&_mi_heap_empty : heap); +} + +#else // default using a thread local variable; used on most platforms. + +static inline mi_heap_t* mi_prim_get_default_heap(void) { + #if defined(MI_TLS_RECURSE_GUARD) + if (mi_unlikely(!_mi_process_is_initialized)) return _mi_heap_main_get(); + #endif + return _mi_heap_default; +} + +#endif // mi_prim_get_default_heap() + + +#endif // MIMALLOC_PRIM_H diff --git a/compat/mimalloc/mimalloc/track.h b/compat/mimalloc/mimalloc/track.h new file mode 100644 index 00000000000000..4b5709e2b54110 --- /dev/null +++ b/compat/mimalloc/mimalloc/track.h @@ -0,0 +1,145 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2023, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ +#pragma once +#ifndef MIMALLOC_TRACK_H +#define MIMALLOC_TRACK_H + +/* ------------------------------------------------------------------------------------------------------ +Track memory ranges with macros for tools like Valgrind address sanitizer, or other memory checkers. +These can be defined for tracking allocation: + + #define mi_track_malloc_size(p,reqsize,size,zero) + #define mi_track_free_size(p,_size) + +The macros are set up such that the size passed to `mi_track_free_size` +always matches the size of `mi_track_malloc_size`. (currently, `size == mi_usable_size(p)`). +The `reqsize` is what the user requested, and `size >= reqsize`. +The `size` is either byte precise (and `size==reqsize`) if `MI_PADDING` is enabled, +or otherwise it is the usable block size which may be larger than the original request. +Use `_mi_block_size_of(void* p)` to get the full block size that was allocated (including padding etc). +The `zero` parameter is `true` if the allocated block is zero initialized. + +Optional: + + #define mi_track_align(p,alignedp,offset,size) + #define mi_track_resize(p,oldsize,newsize) + #define mi_track_init() + +The `mi_track_align` is called right after a `mi_track_malloc` for aligned pointers in a block. +The corresponding `mi_track_free` still uses the block start pointer and original size (corresponding to the `mi_track_malloc`). +The `mi_track_resize` is currently unused but could be called on reallocations within a block. +`mi_track_init` is called at program start. + +The following macros are for tools like asan and valgrind to track whether memory is +defined, undefined, or not accessible at all: + + #define mi_track_mem_defined(p,size) + #define mi_track_mem_undefined(p,size) + #define mi_track_mem_noaccess(p,size) + +-------------------------------------------------------------------------------------------------------*/ + +#if MI_TRACK_VALGRIND +// valgrind tool + +#define MI_TRACK_ENABLED 1 +#define MI_TRACK_HEAP_DESTROY 1 // track free of individual blocks on heap_destroy +#define MI_TRACK_TOOL "valgrind" + +#include +#include + +#define mi_track_malloc_size(p,reqsize,size,zero) VALGRIND_MALLOCLIKE_BLOCK(p,size,MI_PADDING_SIZE /*red zone*/,zero) +#define mi_track_free_size(p,_size) VALGRIND_FREELIKE_BLOCK(p,MI_PADDING_SIZE /*red zone*/) +#define mi_track_resize(p,oldsize,newsize) VALGRIND_RESIZEINPLACE_BLOCK(p,oldsize,newsize,MI_PADDING_SIZE /*red zone*/) +#define mi_track_mem_defined(p,size) VALGRIND_MAKE_MEM_DEFINED(p,size) +#define mi_track_mem_undefined(p,size) VALGRIND_MAKE_MEM_UNDEFINED(p,size) +#define mi_track_mem_noaccess(p,size) VALGRIND_MAKE_MEM_NOACCESS(p,size) + +#elif MI_TRACK_ASAN +// address sanitizer + +#define MI_TRACK_ENABLED 1 +#define MI_TRACK_HEAP_DESTROY 0 +#define MI_TRACK_TOOL "asan" + +#include + +#define mi_track_malloc_size(p,reqsize,size,zero) ASAN_UNPOISON_MEMORY_REGION(p,size) +#define mi_track_free_size(p,size) ASAN_POISON_MEMORY_REGION(p,size) +#define mi_track_mem_defined(p,size) ASAN_UNPOISON_MEMORY_REGION(p,size) +#define mi_track_mem_undefined(p,size) ASAN_UNPOISON_MEMORY_REGION(p,size) +#define mi_track_mem_noaccess(p,size) ASAN_POISON_MEMORY_REGION(p,size) + +#elif MI_TRACK_ETW +// windows event tracing + +#define MI_TRACK_ENABLED 1 +#define MI_TRACK_HEAP_DESTROY 1 +#define MI_TRACK_TOOL "ETW" + +#include "../src/prim/windows/etw.h" + +#define mi_track_init() EventRegistermicrosoft_windows_mimalloc(); +#define mi_track_malloc_size(p,reqsize,size,zero) EventWriteETW_MI_ALLOC((UINT64)(p), size) +#define mi_track_free_size(p,size) EventWriteETW_MI_FREE((UINT64)(p), size) + +#else +// no tracking + +#define MI_TRACK_ENABLED 0 +#define MI_TRACK_HEAP_DESTROY 0 +#define MI_TRACK_TOOL "none" + +#define mi_track_malloc_size(p,reqsize,size,zero) +#define mi_track_free_size(p,_size) + +#endif + +// ------------------- +// Utility definitions + +#ifndef mi_track_resize +#define mi_track_resize(p,oldsize,newsize) mi_track_free_size(p,oldsize); mi_track_malloc(p,newsize,false) +#endif + +#ifndef mi_track_align +#define mi_track_align(p,alignedp,offset,size) mi_track_mem_noaccess(p,offset) +#endif + +#ifndef mi_track_init +#define mi_track_init() +#endif + +#ifndef mi_track_mem_defined +#define mi_track_mem_defined(p,size) +#endif + +#ifndef mi_track_mem_undefined +#define mi_track_mem_undefined(p,size) +#endif + +#ifndef mi_track_mem_noaccess +#define mi_track_mem_noaccess(p,size) +#endif + + +#if MI_PADDING +#define mi_track_malloc(p,reqsize,zero) \ + if ((p)!=NULL) { \ + mi_assert_internal(mi_usable_size(p)==(reqsize)); \ + mi_track_malloc_size(p,reqsize,reqsize,zero); \ + } +#else +#define mi_track_malloc(p,reqsize,zero) \ + if ((p)!=NULL) { \ + mi_assert_internal(mi_usable_size(p)>=(reqsize)); \ + mi_track_malloc_size(p,reqsize,mi_usable_size(p),zero); \ + } +#endif + +#endif diff --git a/compat/mimalloc/mimalloc/types.h b/compat/mimalloc/mimalloc/types.h new file mode 100644 index 00000000000000..5a3f5fe2acac71 --- /dev/null +++ b/compat/mimalloc/mimalloc/types.h @@ -0,0 +1,707 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2024, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ +#pragma once +#ifndef MIMALLOC_TYPES_H +#define MIMALLOC_TYPES_H + +// -------------------------------------------------------------------------- +// This file contains the main type definitions for mimalloc: +// mi_heap_t : all data for a thread-local heap, contains +// lists of all managed heap pages. +// mi_segment_t : a larger chunk of memory (32GiB) from where pages +// are allocated. A segment is divided in slices (64KiB) from +// which pages are allocated. +// mi_page_t : a "mimalloc" page (usually 64KiB or 512KiB) from +// where objects are allocated. +// Note: we write "OS page" for OS memory pages while +// using plain "page" for mimalloc pages (`mi_page_t`). +// -------------------------------------------------------------------------- + + +#include +#include // ptrdiff_t +#include // uintptr_t, uint16_t, etc +#include "atomic.h" // _Atomic + +#ifdef _MSC_VER +#pragma warning(disable:4214) // bitfield is not int +#endif + +// Minimal alignment necessary. On most platforms 16 bytes are needed +// due to SSE registers for example. This must be at least `sizeof(void*)` +#ifndef MI_MAX_ALIGN_SIZE +#define MI_MAX_ALIGN_SIZE 16 // sizeof(max_align_t) +#endif + +// ------------------------------------------------------ +// Variants +// ------------------------------------------------------ + +// Define NDEBUG in the release version to disable assertions. +// #define NDEBUG + +// Define MI_TRACK_ to enable tracking support +// #define MI_TRACK_VALGRIND 1 +// #define MI_TRACK_ASAN 1 +// #define MI_TRACK_ETW 1 + +// Define MI_STAT as 1 to maintain statistics; set it to 2 to have detailed statistics (but costs some performance). +// #define MI_STAT 1 + +// Define MI_SECURE to enable security mitigations +// #define MI_SECURE 1 // guard page around metadata +// #define MI_SECURE 2 // guard page around each mimalloc page +// #define MI_SECURE 3 // encode free lists (detect corrupted free list (buffer overflow), and invalid pointer free) +// #define MI_SECURE 4 // checks for double free. (may be more expensive) + +#if !defined(MI_SECURE) +#define MI_SECURE 0 +#endif + +// Define MI_DEBUG for debug mode +// #define MI_DEBUG 1 // basic assertion checks and statistics, check double free, corrupted free list, and invalid pointer free. +// #define MI_DEBUG 2 // + internal assertion checks +// #define MI_DEBUG 3 // + extensive internal invariant checking (cmake -DMI_DEBUG_FULL=ON) +#if !defined(MI_DEBUG) +#if defined(MI_BUILD_RELEASE) || defined(NDEBUG) +#define MI_DEBUG 0 +#else +#define MI_DEBUG 2 +#endif +#endif + +// Use guard pages behind objects of a certain size (set by the MIMALLOC_DEBUG_GUARDED_MIN/MAX options) +// Padding should be disabled when using guard pages +// #define MI_GUARDED 1 +#if defined(MI_GUARDED) +#define MI_PADDING 0 +#endif + +// Reserve extra padding at the end of each block to be more resilient against heap block overflows. +// The padding can detect buffer overflow on free. +#if !defined(MI_PADDING) && (MI_SECURE>=3 || MI_DEBUG>=1 || (MI_TRACK_VALGRIND || MI_TRACK_ASAN || MI_TRACK_ETW)) +#define MI_PADDING 1 +#endif + +// Check padding bytes; allows byte-precise buffer overflow detection +#if !defined(MI_PADDING_CHECK) && MI_PADDING && (MI_SECURE>=3 || MI_DEBUG>=1) +#define MI_PADDING_CHECK 1 +#endif + + +// Encoded free lists allow detection of corrupted free lists +// and can detect buffer overflows, modify after free, and double `free`s. +#if (MI_SECURE>=3 || MI_DEBUG>=1) +#define MI_ENCODE_FREELIST 1 +#endif + + +// We used to abandon huge pages in order to eagerly deallocate it if freed from another thread. +// Unfortunately, that makes it not possible to visit them during a heap walk or include them in a +// `mi_heap_destroy`. We therefore instead reset/decommit the huge blocks nowadays if freed from +// another thread so the memory becomes "virtually" available (and eventually gets properly freed by +// the owning thread). +// #define MI_HUGE_PAGE_ABANDON 1 + + +// ------------------------------------------------------ +// Platform specific values +// ------------------------------------------------------ + +// ------------------------------------------------------ +// Size of a pointer. +// We assume that `sizeof(void*)==sizeof(intptr_t)` +// and it holds for all platforms we know of. +// +// However, the C standard only requires that: +// p == (void*)((intptr_t)p)) +// but we also need: +// i == (intptr_t)((void*)i) +// or otherwise one might define an intptr_t type that is larger than a pointer... +// ------------------------------------------------------ + +#if INTPTR_MAX > INT64_MAX +# define MI_INTPTR_SHIFT (4) // assume 128-bit (as on arm CHERI for example) +#elif INTPTR_MAX == INT64_MAX +# define MI_INTPTR_SHIFT (3) +#elif INTPTR_MAX == INT32_MAX +# define MI_INTPTR_SHIFT (2) +#else +#error platform pointers must be 32, 64, or 128 bits +#endif + +#if SIZE_MAX == UINT64_MAX +# define MI_SIZE_SHIFT (3) +typedef int64_t mi_ssize_t; +#elif SIZE_MAX == UINT32_MAX +# define MI_SIZE_SHIFT (2) +typedef int32_t mi_ssize_t; +#else +#error platform objects must be 32 or 64 bits +#endif + +#if (SIZE_MAX/2) > LONG_MAX +# define MI_ZU(x) x##ULL +# define MI_ZI(x) x##LL +#else +# define MI_ZU(x) x##UL +# define MI_ZI(x) x##L +#endif + +#define MI_INTPTR_SIZE (1< 4 +#define MI_SEGMENT_SHIFT ( 9 + MI_SEGMENT_SLICE_SHIFT) // 32MiB +#else +#define MI_SEGMENT_SHIFT ( 7 + MI_SEGMENT_SLICE_SHIFT) // 4MiB on 32-bit +#endif +#endif + +#ifndef MI_SMALL_PAGE_SHIFT +#define MI_SMALL_PAGE_SHIFT (MI_SEGMENT_SLICE_SHIFT) // 64KiB +#endif +#ifndef MI_MEDIUM_PAGE_SHIFT +#define MI_MEDIUM_PAGE_SHIFT ( 3 + MI_SMALL_PAGE_SHIFT) // 512KiB +#endif + +// Derived constants +#define MI_SEGMENT_SIZE (MI_ZU(1)<= 655360) +#error "mimalloc internal: define more bins" +#endif + +// Maximum block size for which blocks are guaranteed to be block size aligned. (see `segment.c:_mi_segment_page_start`) +#define MI_MAX_ALIGN_GUARANTEE (MI_MEDIUM_OBJ_SIZE_MAX) + +// Alignments over MI_BLOCK_ALIGNMENT_MAX are allocated in dedicated huge page segments +#define MI_BLOCK_ALIGNMENT_MAX (MI_SEGMENT_SIZE >> 1) + +// Maximum slice count (255) for which we can find the page for interior pointers +#define MI_MAX_SLICE_OFFSET_COUNT ((MI_BLOCK_ALIGNMENT_MAX / MI_SEGMENT_SLICE_SIZE) - 1) + +// we never allocate more than PTRDIFF_MAX (see also ) +// on 64-bit+ systems we also limit the maximum allocation size such that the slice count fits in 32-bits. (issue #877) +#if (PTRDIFF_MAX > INT32_MAX) && (PTRDIFF_MAX >= (MI_SEGMENT_SLIZE_SIZE * UINT32_MAX)) +#define MI_MAX_ALLOC_SIZE (MI_SEGMENT_SLICE_SIZE * (UINT32_MAX-1)) +#else +#define MI_MAX_ALLOC_SIZE PTRDIFF_MAX +#endif + + +// ------------------------------------------------------ +// Mimalloc pages contain allocated blocks +// ------------------------------------------------------ + +// The free lists use encoded next fields +// (Only actually encodes when MI_ENCODED_FREELIST is defined.) +typedef uintptr_t mi_encoded_t; + +// thread id's +typedef size_t mi_threadid_t; + +// free lists contain blocks +typedef struct mi_block_s { + mi_encoded_t next; +} mi_block_t; + +#if MI_GUARDED +// we always align guarded pointers in a block at an offset +// the block `next` field is then used as a tag to distinguish regular offset aligned blocks from guarded ones +#define MI_BLOCK_TAG_ALIGNED ((mi_encoded_t)(0)) +#define MI_BLOCK_TAG_GUARDED (~MI_BLOCK_TAG_ALIGNED) +#endif + + +// The delayed flags are used for efficient multi-threaded free-ing +typedef enum mi_delayed_e { + MI_USE_DELAYED_FREE = 0, // push on the owning heap thread delayed list + MI_DELAYED_FREEING = 1, // temporary: another thread is accessing the owning heap + MI_NO_DELAYED_FREE = 2, // optimize: push on page local thread free queue if another block is already in the heap thread delayed free list + MI_NEVER_DELAYED_FREE = 3 // sticky: used for abandoned pages without a owning heap; this only resets on page reclaim +} mi_delayed_t; + + +// The `in_full` and `has_aligned` page flags are put in a union to efficiently +// test if both are false (`full_aligned == 0`) in the `mi_free` routine. +#if !MI_TSAN +typedef union mi_page_flags_s { + uint8_t full_aligned; + struct { + uint8_t in_full : 1; + uint8_t has_aligned : 1; + } x; +} mi_page_flags_t; +#else +// under thread sanitizer, use a byte for each flag to suppress warning, issue #130 +typedef union mi_page_flags_s { + uint32_t full_aligned; + struct { + uint8_t in_full; + uint8_t has_aligned; + } x; +} mi_page_flags_t; +#endif + +// Thread free list. +// We use the bottom 2 bits of the pointer for mi_delayed_t flags +typedef uintptr_t mi_thread_free_t; + +// A page contains blocks of one specific size (`block_size`). +// Each page has three list of free blocks: +// `free` for blocks that can be allocated, +// `local_free` for freed blocks that are not yet available to `mi_malloc` +// `thread_free` for freed blocks by other threads +// The `local_free` and `thread_free` lists are migrated to the `free` list +// when it is exhausted. The separate `local_free` list is necessary to +// implement a monotonic heartbeat. The `thread_free` list is needed for +// avoiding atomic operations in the common case. +// +// `used - |thread_free|` == actual blocks that are in use (alive) +// `used - |thread_free| + |free| + |local_free| == capacity` +// +// We don't count `freed` (as |free|) but use `used` to reduce +// the number of memory accesses in the `mi_page_all_free` function(s). +// +// Notes: +// - Access is optimized for `free.c:mi_free` and `alloc.c:mi_page_alloc` +// - Using `uint16_t` does not seem to slow things down +// - The size is 12 words on 64-bit which helps the page index calculations +// (and 14 words on 32-bit, and encoded free lists add 2 words) +// - `xthread_free` uses the bottom bits as a delayed-free flags to optimize +// concurrent frees where only the first concurrent free adds to the owning +// heap `thread_delayed_free` list (see `free.c:mi_free_block_mt`). +// The invariant is that no-delayed-free is only set if there is +// at least one block that will be added, or as already been added, to +// the owning heap `thread_delayed_free` list. This guarantees that pages +// will be freed correctly even if only other threads free blocks. +typedef struct mi_page_s { + // "owned" by the segment + uint32_t slice_count; // slices in this page (0 if not a page) + uint32_t slice_offset; // distance from the actual page data slice (0 if a page) + uint8_t is_committed:1; // `true` if the page virtual memory is committed + uint8_t is_zero_init:1; // `true` if the page was initially zero initialized + uint8_t is_huge:1; // `true` if the page is in a huge segment (`segment->kind == MI_SEGMENT_HUGE`) + // padding + // layout like this to optimize access in `mi_malloc` and `mi_free` + uint16_t capacity; // number of blocks committed, must be the first field, see `segment.c:page_clear` + uint16_t reserved; // number of blocks reserved in memory + mi_page_flags_t flags; // `in_full` and `has_aligned` flags (8 bits) + uint8_t free_is_zero:1; // `true` if the blocks in the free list are zero initialized + uint8_t retire_expire:7; // expiration count for retired blocks + + mi_block_t* free; // list of available free blocks (`malloc` allocates from this list) + mi_block_t* local_free; // list of deferred free blocks by this thread (migrates to `free`) + uint16_t used; // number of blocks in use (including blocks in `thread_free`) + uint8_t block_size_shift; // if not zero, then `(1 << block_size_shift) == block_size` (only used for fast path in `free.c:_mi_page_ptr_unalign`) + uint8_t heap_tag; // tag of the owning heap, used to separate heaps by object type + // padding + size_t block_size; // size available in each block (always `>0`) + uint8_t* page_start; // start of the page area containing the blocks + + #if (MI_ENCODE_FREELIST || MI_PADDING) + uintptr_t keys[2]; // two random keys to encode the free lists (see `_mi_block_next`) or padding canary + #endif + + _Atomic(mi_thread_free_t) xthread_free; // list of deferred free blocks freed by other threads + _Atomic(uintptr_t) xheap; + + struct mi_page_s* next; // next page owned by this thread with the same `block_size` + struct mi_page_s* prev; // previous page owned by this thread with the same `block_size` + + // 64-bit 11 words, 32-bit 13 words, (+2 for secure) + void* padding[1]; +} mi_page_t; + + + +// ------------------------------------------------------ +// Mimalloc segments contain mimalloc pages +// ------------------------------------------------------ + +typedef enum mi_page_kind_e { + MI_PAGE_SMALL, // small blocks go into 64KiB pages inside a segment + MI_PAGE_MEDIUM, // medium blocks go into 512KiB pages inside a segment + MI_PAGE_LARGE, // larger blocks go into a single page spanning a whole segment + MI_PAGE_HUGE // a huge page is a single page in a segment of variable size + // used for blocks `> MI_LARGE_OBJ_SIZE_MAX` or an aligment `> MI_BLOCK_ALIGNMENT_MAX`. +} mi_page_kind_t; + +typedef enum mi_segment_kind_e { + MI_SEGMENT_NORMAL, // MI_SEGMENT_SIZE size with pages inside. + MI_SEGMENT_HUGE, // segment with just one huge page inside. +} mi_segment_kind_t; + +// ------------------------------------------------------ +// A segment holds a commit mask where a bit is set if +// the corresponding MI_COMMIT_SIZE area is committed. +// The MI_COMMIT_SIZE must be a multiple of the slice +// size. If it is equal we have the most fine grained +// decommit (but setting it higher can be more efficient). +// The MI_MINIMAL_COMMIT_SIZE is the minimal amount that will +// be committed in one go which can be set higher than +// MI_COMMIT_SIZE for efficiency (while the decommit mask +// is still tracked in fine-grained MI_COMMIT_SIZE chunks) +// ------------------------------------------------------ + +#define MI_MINIMAL_COMMIT_SIZE (1*MI_SEGMENT_SLICE_SIZE) +#define MI_COMMIT_SIZE (MI_SEGMENT_SLICE_SIZE) // 64KiB +#define MI_COMMIT_MASK_BITS (MI_SEGMENT_SIZE / MI_COMMIT_SIZE) +#define MI_COMMIT_MASK_FIELD_BITS MI_SIZE_BITS +#define MI_COMMIT_MASK_FIELD_COUNT (MI_COMMIT_MASK_BITS / MI_COMMIT_MASK_FIELD_BITS) + +#if (MI_COMMIT_MASK_BITS != (MI_COMMIT_MASK_FIELD_COUNT * MI_COMMIT_MASK_FIELD_BITS)) +#error "the segment size must be exactly divisible by the (commit size * size_t bits)" +#endif + +typedef struct mi_commit_mask_s { + size_t mask[MI_COMMIT_MASK_FIELD_COUNT]; +} mi_commit_mask_t; + +typedef mi_page_t mi_slice_t; +typedef int64_t mi_msecs_t; + + +// --------------------------------------------------------------- +// a memory id tracks the provenance of arena/OS allocated memory +// --------------------------------------------------------------- + +// Memory can reside in arena's, direct OS allocated, or statically allocated. The memid keeps track of this. +typedef enum mi_memkind_e { + MI_MEM_NONE, // not allocated + MI_MEM_EXTERNAL, // not owned by mimalloc but provided externally (via `mi_manage_os_memory` for example) + MI_MEM_STATIC, // allocated in a static area and should not be freed (for arena meta data for example) + MI_MEM_OS, // allocated from the OS + MI_MEM_OS_HUGE, // allocated as huge OS pages (usually 1GiB, pinned to physical memory) + MI_MEM_OS_REMAP, // allocated in a remapable area (i.e. using `mremap`) + MI_MEM_ARENA // allocated from an arena (the usual case) +} mi_memkind_t; + +static inline bool mi_memkind_is_os(mi_memkind_t memkind) { + return (memkind >= MI_MEM_OS && memkind <= MI_MEM_OS_REMAP); +} + +typedef struct mi_memid_os_info { + void* base; // actual base address of the block (used for offset aligned allocations) + size_t size; // full allocation size +} mi_memid_os_info_t; + +typedef struct mi_memid_arena_info { + size_t block_index; // index in the arena + mi_arena_id_t id; // arena id (>= 1) + bool is_exclusive; // this arena can only be used for specific arena allocations +} mi_memid_arena_info_t; + +typedef struct mi_memid_s { + union { + mi_memid_os_info_t os; // only used for MI_MEM_OS + mi_memid_arena_info_t arena; // only used for MI_MEM_ARENA + } mem; + bool is_pinned; // `true` if we cannot decommit/reset/protect in this memory (e.g. when allocated using large (2Mib) or huge (1GiB) OS pages) + bool initially_committed;// `true` if the memory was originally allocated as committed + bool initially_zero; // `true` if the memory was originally zero initialized + mi_memkind_t memkind; +} mi_memid_t; + + +// ----------------------------------------------------------------------------------------- +// Segments are large allocated memory blocks (32mb on 64 bit) from arenas or the OS. +// +// Inside segments we allocated fixed size mimalloc pages (`mi_page_t`) that contain blocks. +// The start of a segment is this structure with a fixed number of slice entries (`slices`) +// usually followed by a guard OS page and the actual allocation area with pages. +// While a page is not allocated, we view it's data as a `mi_slice_t` (instead of a `mi_page_t`). +// Of any free area, the first slice has the info and `slice_offset == 0`; for any subsequent +// slices part of the area, the `slice_offset` is the byte offset back to the first slice +// (so we can quickly find the page info on a free, `internal.h:_mi_segment_page_of`). +// For slices, the `block_size` field is repurposed to signify if a slice is used (`1`) or not (`0`). +// Small and medium pages use a fixed amount of slices to reduce slice fragmentation, while +// large and huge pages span a variable amount of slices. + +typedef struct mi_subproc_s mi_subproc_t; + +typedef struct mi_segment_s { + // constant fields + mi_memid_t memid; // memory id for arena/OS allocation + bool allow_decommit; // can we decommmit the memory + bool allow_purge; // can we purge the memory (reset or decommit) + size_t segment_size; + mi_subproc_t* subproc; // segment belongs to sub process + + // segment fields + mi_msecs_t purge_expire; // purge slices in the `purge_mask` after this time + mi_commit_mask_t purge_mask; // slices that can be purged + mi_commit_mask_t commit_mask; // slices that are currently committed + + // from here is zero initialized + struct mi_segment_s* next; // the list of freed segments in the cache (must be first field, see `segment.c:mi_segment_init`) + bool was_reclaimed; // true if it was reclaimed (used to limit on-free reclamation) + bool dont_free; // can be temporarily true to ensure the segment is not freed + + size_t abandoned; // abandoned pages (i.e. the original owning thread stopped) (`abandoned <= used`) + size_t abandoned_visits; // count how often this segment is visited during abondoned reclamation (to force reclaim if it takes too long) + size_t used; // count of pages in use + uintptr_t cookie; // verify addresses in debug mode: `mi_ptr_cookie(segment) == segment->cookie` + + struct mi_segment_s* abandoned_os_next; // only used for abandoned segments outside arena's, and only if `mi_option_visit_abandoned` is enabled + struct mi_segment_s* abandoned_os_prev; + + size_t segment_slices; // for huge segments this may be different from `MI_SLICES_PER_SEGMENT` + size_t segment_info_slices; // initial count of slices that we are using for segment info and possible guard pages. + + // layout like this to optimize access in `mi_free` + mi_segment_kind_t kind; + size_t slice_entries; // entries in the `slices` array, at most `MI_SLICES_PER_SEGMENT` + _Atomic(mi_threadid_t) thread_id; // unique id of the thread owning this segment + + mi_slice_t slices[MI_SLICES_PER_SEGMENT+1]; // one extra final entry for huge blocks with large alignment +} mi_segment_t; + + +// ------------------------------------------------------ +// Heaps +// Provide first-class heaps to allocate from. +// A heap just owns a set of pages for allocation and +// can only be allocate/reallocate from the thread that created it. +// Freeing blocks can be done from any thread though. +// Per thread, the segments are shared among its heaps. +// Per thread, there is always a default heap that is +// used for allocation; it is initialized to statically +// point to an empty heap to avoid initialization checks +// in the fast path. +// ------------------------------------------------------ + +// Thread local data +typedef struct mi_tld_s mi_tld_t; + +// Pages of a certain block size are held in a queue. +typedef struct mi_page_queue_s { + mi_page_t* first; + mi_page_t* last; + size_t block_size; +} mi_page_queue_t; + +#define MI_BIN_FULL (MI_BIN_HUGE+1) + +// Random context +typedef struct mi_random_cxt_s { + uint32_t input[16]; + uint32_t output[16]; + int output_available; + bool weak; +} mi_random_ctx_t; + + +// In debug mode there is a padding structure at the end of the blocks to check for buffer overflows +#if (MI_PADDING) +typedef struct mi_padding_s { + uint32_t canary; // encoded block value to check validity of the padding (in case of overflow) + uint32_t delta; // padding bytes before the block. (mi_usable_size(p) - delta == exact allocated bytes) +} mi_padding_t; +#define MI_PADDING_SIZE (sizeof(mi_padding_t)) +#define MI_PADDING_WSIZE ((MI_PADDING_SIZE + MI_INTPTR_SIZE - 1) / MI_INTPTR_SIZE) +#else +#define MI_PADDING_SIZE 0 +#define MI_PADDING_WSIZE 0 +#endif + +#define MI_PAGES_DIRECT (MI_SMALL_WSIZE_MAX + MI_PADDING_WSIZE + 1) + + +// A heap owns a set of pages. +struct mi_heap_s { + mi_tld_t* tld; + _Atomic(mi_block_t*) thread_delayed_free; + mi_threadid_t thread_id; // thread this heap belongs too + mi_arena_id_t arena_id; // arena id if the heap belongs to a specific arena (or 0) + uintptr_t cookie; // random cookie to verify pointers (see `_mi_ptr_cookie`) + uintptr_t keys[2]; // two random keys used to encode the `thread_delayed_free` list + mi_random_ctx_t random; // random number context used for secure allocation + size_t page_count; // total number of pages in the `pages` queues. + size_t page_retired_min; // smallest retired index (retired pages are fully free, but still in the page queues) + size_t page_retired_max; // largest retired index into the `pages` array. + long generic_count; // how often is `_mi_malloc_generic` called? + long generic_collect_count; // how often is `_mi_malloc_generic` called without collecting? + mi_heap_t* next; // list of heaps per thread + bool no_reclaim; // `true` if this heap should not reclaim abandoned pages + uint8_t tag; // custom tag, can be used for separating heaps based on the object types + #if MI_GUARDED + size_t guarded_size_min; // minimal size for guarded objects + size_t guarded_size_max; // maximal size for guarded objects + size_t guarded_sample_rate; // sample rate (set to 0 to disable guarded pages) + size_t guarded_sample_seed; // starting sample count + size_t guarded_sample_count; // current sample count (counting down to 0) + #endif + mi_page_t* pages_free_direct[MI_PAGES_DIRECT]; // optimize: array where every entry points a page with possibly free blocks in the corresponding queue for that size. + mi_page_queue_t pages[MI_BIN_FULL + 1]; // queue of pages for each size class (or "bin") +}; + + +// ------------------------------------------------------ +// Sub processes do not reclaim or visit segments +// from other sub processes. These are essentially the +// static variables of a process. +// ------------------------------------------------------ + +struct mi_subproc_s { + _Atomic(size_t) abandoned_count; // count of abandoned segments for this sub-process + _Atomic(size_t) abandoned_os_list_count; // count of abandoned segments in the os-list + mi_lock_t abandoned_os_lock; // lock for the abandoned os segment list (outside of arena's) (this lock protect list operations) + mi_lock_t abandoned_os_visit_lock; // ensure only one thread per subproc visits the abandoned os list + mi_segment_t* abandoned_os_list; // doubly-linked list of abandoned segments outside of arena's (in OS allocated memory) + mi_segment_t* abandoned_os_list_tail; // the tail-end of the list + mi_memid_t memid; // provenance of this memory block +}; + + +// ------------------------------------------------------ +// Thread Local data +// ------------------------------------------------------ + +// A "span" is is an available range of slices. The span queues keep +// track of slice spans of at most the given `slice_count` (but more than the previous size class). +typedef struct mi_span_queue_s { + mi_slice_t* first; + mi_slice_t* last; + size_t slice_count; +} mi_span_queue_t; + +#define MI_SEGMENT_BIN_MAX (35) // 35 == mi_segment_bin(MI_SLICES_PER_SEGMENT) + +// Segments thread local data +typedef struct mi_segments_tld_s { + mi_span_queue_t spans[MI_SEGMENT_BIN_MAX+1]; // free slice spans inside segments + size_t count; // current number of segments; + size_t peak_count; // peak number of segments + size_t current_size; // current size of all segments + size_t peak_size; // peak size of all segments + size_t reclaim_count;// number of reclaimed (abandoned) segments + mi_subproc_t* subproc; // sub-process this thread belongs to. + mi_stats_t* stats; // points to tld stats +} mi_segments_tld_t; + +// Thread local data +struct mi_tld_s { + unsigned long long heartbeat; // monotonic heartbeat count + bool recurse; // true if deferred was called; used to prevent infinite recursion. + mi_heap_t* heap_backing; // backing heap of this thread (cannot be deleted) + mi_heap_t* heaps; // list of heaps in this thread (so we can abandon all when the thread terminates) + mi_segments_tld_t segments; // segment tld + mi_stats_t stats; // statistics +}; + + + +// ------------------------------------------------------ +// Debug +// ------------------------------------------------------ + +#if !defined(MI_DEBUG_UNINIT) +#define MI_DEBUG_UNINIT (0xD0) +#endif +#if !defined(MI_DEBUG_FREED) +#define MI_DEBUG_FREED (0xDF) +#endif +#if !defined(MI_DEBUG_PADDING) +#define MI_DEBUG_PADDING (0xDE) +#endif + +#if (MI_DEBUG) +// use our own assertion to print without memory allocation +void _mi_assert_fail(const char* assertion, const char* fname, unsigned int line, const char* func ); +#define mi_assert(expr) ((expr) ? (void)0 : _mi_assert_fail(#expr,__FILE__,__LINE__,__func__)) +#else +#define mi_assert(x) +#endif + +#if (MI_DEBUG>1) +#define mi_assert_internal mi_assert +#else +#define mi_assert_internal(x) +#endif + +#if (MI_DEBUG>2) +#define mi_assert_expensive mi_assert +#else +#define mi_assert_expensive(x) +#endif + + +// ------------------------------------------------------ +// Statistics +// ------------------------------------------------------ +#ifndef MI_STAT +#if (MI_DEBUG>0) +#define MI_STAT 2 +#else +#define MI_STAT 0 +#endif +#endif + +// add to stat keeping track of the peak +void _mi_stat_increase(mi_stat_count_t* stat, size_t amount); +void _mi_stat_decrease(mi_stat_count_t* stat, size_t amount); +void _mi_stat_adjust_decrease(mi_stat_count_t* stat, size_t amount); +// counters can just be increased +void _mi_stat_counter_increase(mi_stat_counter_t* stat, size_t amount); + +#if (MI_STAT) +#define mi_stat_increase(stat,amount) _mi_stat_increase( &(stat), amount) +#define mi_stat_decrease(stat,amount) _mi_stat_decrease( &(stat), amount) +#define mi_stat_adjust_decrease(stat,amount) _mi_stat_adjust_decrease( &(stat), amount) +#define mi_stat_counter_increase(stat,amount) _mi_stat_counter_increase( &(stat), amount) +#else +#define mi_stat_increase(stat,amount) ((void)0) +#define mi_stat_decrease(stat,amount) ((void)0) +#define mi_stat_adjust_decrease(stat,amount) ((void)0) +#define mi_stat_counter_increase(stat,amount) ((void)0) +#endif + +#define mi_heap_stat_counter_increase(heap,stat,amount) mi_stat_counter_increase( (heap)->tld->stats.stat, amount) +#define mi_heap_stat_increase(heap,stat,amount) mi_stat_increase( (heap)->tld->stats.stat, amount) +#define mi_heap_stat_decrease(heap,stat,amount) mi_stat_decrease( (heap)->tld->stats.stat, amount) +#define mi_heap_stat_adjust_decrease(heap,stat,amount) mi_stat_adjust_decrease( (heap)->tld->stats.stat, amount) + +#endif diff --git a/compat/mimalloc/options.c b/compat/mimalloc/options.c new file mode 100644 index 00000000000000..4759e0b03bb779 --- /dev/null +++ b/compat/mimalloc/options.c @@ -0,0 +1,663 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2021, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ +#include "mimalloc.h" +#include "mimalloc/internal.h" +#include "mimalloc/atomic.h" +#include "mimalloc/prim.h" // mi_prim_out_stderr + +#include // stdin/stdout +#include // abort + + + +static long mi_max_error_count = 16; // stop outputting errors after this (use < 0 for no limit) +static long mi_max_warning_count = 16; // stop outputting warnings after this (use < 0 for no limit) + +static void mi_add_stderr_output(void); + +int mi_version(void) mi_attr_noexcept { + return MI_MALLOC_VERSION; +} + + +// -------------------------------------------------------- +// Options +// These can be accessed by multiple threads and may be +// concurrently initialized, but an initializing data race +// is ok since they resolve to the same value. +// -------------------------------------------------------- +typedef enum mi_init_e { + UNINIT, // not yet initialized + DEFAULTED, // not found in the environment, use default value + INITIALIZED // found in environment or set explicitly +} mi_init_t; + +typedef struct mi_option_desc_s { + long value; // the value + mi_init_t init; // is it initialized yet? (from the environment) + mi_option_t option; // for debugging: the option index should match the option + const char* name; // option name without `mimalloc_` prefix + const char* legacy_name; // potential legacy option name +} mi_option_desc_t; + +#define MI_OPTION(opt) mi_option_##opt, #opt, NULL +#define MI_OPTION_LEGACY(opt,legacy) mi_option_##opt, #opt, #legacy + +// Some options can be set at build time for statically linked libraries +// (use `-DMI_EXTRA_CPPDEFS="opt1=val1;opt2=val2"`) +// +// This is useful if we cannot pass them as environment variables +// (and setting them programmatically would be too late) + +#ifndef MI_DEFAULT_VERBOSE +#define MI_DEFAULT_VERBOSE 0 +#endif + +#ifndef MI_DEFAULT_EAGER_COMMIT +#define MI_DEFAULT_EAGER_COMMIT 1 +#endif + +#ifndef MI_DEFAULT_ARENA_EAGER_COMMIT +#define MI_DEFAULT_ARENA_EAGER_COMMIT 2 +#endif + +// in KiB +#ifndef MI_DEFAULT_ARENA_RESERVE + #if (MI_INTPTR_SIZE>4) + #define MI_DEFAULT_ARENA_RESERVE 1024L*1024L + #else + #define MI_DEFAULT_ARENA_RESERVE 128L*1024L + #endif +#endif + +#ifndef MI_DEFAULT_DISALLOW_ARENA_ALLOC +#define MI_DEFAULT_DISALLOW_ARENA_ALLOC 0 +#endif + +#ifndef MI_DEFAULT_ALLOW_LARGE_OS_PAGES +#if defined(__linux__) && !defined(__ANDROID__) +#define MI_DEFAULT_ALLOW_LARGE_OS_PAGES 2 // enabled, but only use transparent huge pages through madvise +#else +#define MI_DEFAULT_ALLOW_LARGE_OS_PAGES 0 +#endif +#endif + +#ifndef MI_DEFAULT_RESERVE_HUGE_OS_PAGES +#define MI_DEFAULT_RESERVE_HUGE_OS_PAGES 0 +#endif + +#ifndef MI_DEFAULT_RESERVE_OS_MEMORY +#define MI_DEFAULT_RESERVE_OS_MEMORY 0 +#endif + +#ifndef MI_DEFAULT_GUARDED_SAMPLE_RATE +#if MI_GUARDED +#define MI_DEFAULT_GUARDED_SAMPLE_RATE 4000 +#else +#define MI_DEFAULT_GUARDED_SAMPLE_RATE 0 +#endif +#endif + + +static mi_option_desc_t options[_mi_option_last] = +{ + // stable options + #if MI_DEBUG || defined(MI_SHOW_ERRORS) + { 1, UNINIT, MI_OPTION(show_errors) }, + #else + { 0, UNINIT, MI_OPTION(show_errors) }, + #endif + { 0, UNINIT, MI_OPTION(show_stats) }, + { MI_DEFAULT_VERBOSE, UNINIT, MI_OPTION(verbose) }, + + // some of the following options are experimental and not all combinations are allowed. + { MI_DEFAULT_EAGER_COMMIT, + UNINIT, MI_OPTION(eager_commit) }, // commit per segment directly (4MiB) (but see also `eager_commit_delay`) + { MI_DEFAULT_ARENA_EAGER_COMMIT, + UNINIT, MI_OPTION_LEGACY(arena_eager_commit,eager_region_commit) }, // eager commit arena's? 2 is used to enable this only on an OS that has overcommit (i.e. linux) + { 1, UNINIT, MI_OPTION_LEGACY(purge_decommits,reset_decommits) }, // purge decommits memory (instead of reset) (note: on linux this uses MADV_DONTNEED for decommit) + { MI_DEFAULT_ALLOW_LARGE_OS_PAGES, + UNINIT, MI_OPTION_LEGACY(allow_large_os_pages,large_os_pages) }, // use large OS pages, use only with eager commit to prevent fragmentation of VMA's + { MI_DEFAULT_RESERVE_HUGE_OS_PAGES, + UNINIT, MI_OPTION(reserve_huge_os_pages) }, // per 1GiB huge pages + {-1, UNINIT, MI_OPTION(reserve_huge_os_pages_at) }, // reserve huge pages at node N + { MI_DEFAULT_RESERVE_OS_MEMORY, + UNINIT, MI_OPTION(reserve_os_memory) }, // reserve N KiB OS memory in advance (use `option_get_size`) + { 0, UNINIT, MI_OPTION(deprecated_segment_cache) }, // cache N segments per thread + { 0, UNINIT, MI_OPTION(deprecated_page_reset) }, // reset page memory on free + { 0, UNINIT, MI_OPTION_LEGACY(abandoned_page_purge,abandoned_page_reset) }, // reset free page memory when a thread terminates + { 0, UNINIT, MI_OPTION(deprecated_segment_reset) }, // reset segment memory on free (needs eager commit) +#if defined(__NetBSD__) + { 0, UNINIT, MI_OPTION(eager_commit_delay) }, // the first N segments per thread are not eagerly committed +#else + { 1, UNINIT, MI_OPTION(eager_commit_delay) }, // the first N segments per thread are not eagerly committed (but per page in the segment on demand) +#endif + { 10, UNINIT, MI_OPTION_LEGACY(purge_delay,reset_delay) }, // purge delay in milli-seconds + { 0, UNINIT, MI_OPTION(use_numa_nodes) }, // 0 = use available numa nodes, otherwise use at most N nodes. + { 0, UNINIT, MI_OPTION_LEGACY(disallow_os_alloc,limit_os_alloc) }, // 1 = do not use OS memory for allocation (but only reserved arenas) + { 100, UNINIT, MI_OPTION(os_tag) }, // only apple specific for now but might serve more or less related purpose + { 32, UNINIT, MI_OPTION(max_errors) }, // maximum errors that are output + { 32, UNINIT, MI_OPTION(max_warnings) }, // maximum warnings that are output + { 10, UNINIT, MI_OPTION(max_segment_reclaim)}, // max. percentage of the abandoned segments to be reclaimed per try. + { 0, UNINIT, MI_OPTION(destroy_on_exit)}, // release all OS memory on process exit; careful with dangling pointer or after-exit frees! + { MI_DEFAULT_ARENA_RESERVE, UNINIT, MI_OPTION(arena_reserve) }, // reserve memory N KiB at a time (=1GiB) (use `option_get_size`) + { 10, UNINIT, MI_OPTION(arena_purge_mult) }, // purge delay multiplier for arena's + { 1, UNINIT, MI_OPTION_LEGACY(purge_extend_delay, decommit_extend_delay) }, + { 0, UNINIT, MI_OPTION(abandoned_reclaim_on_free) },// reclaim an abandoned segment on a free + { MI_DEFAULT_DISALLOW_ARENA_ALLOC, UNINIT, MI_OPTION(disallow_arena_alloc) }, // 1 = do not use arena's for allocation (except if using specific arena id's) + { 400, UNINIT, MI_OPTION(retry_on_oom) }, // windows only: retry on out-of-memory for N milli seconds (=400), set to 0 to disable retries. +#if defined(MI_VISIT_ABANDONED) + { 1, INITIALIZED, MI_OPTION(visit_abandoned) }, // allow visiting heap blocks in abandoned segments; requires taking locks during reclaim. +#else + { 0, UNINIT, MI_OPTION(visit_abandoned) }, +#endif + { 0, UNINIT, MI_OPTION(guarded_min) }, // only used when building with MI_GUARDED: minimal rounded object size for guarded objects + { MI_GiB, UNINIT, MI_OPTION(guarded_max) }, // only used when building with MI_GUARDED: maximal rounded object size for guarded objects + { 0, UNINIT, MI_OPTION(guarded_precise) }, // disregard minimal alignment requirement to always place guarded blocks exactly in front of a guard page (=0) + { MI_DEFAULT_GUARDED_SAMPLE_RATE, + UNINIT, MI_OPTION(guarded_sample_rate)}, // 1 out of N allocations in the min/max range will be guarded (=4000) + { 0, UNINIT, MI_OPTION(guarded_sample_seed)}, + { 0, UNINIT, MI_OPTION(target_segments_per_thread) }, // abandon segments beyond this point, or 0 to disable. + { 10000, UNINIT, MI_OPTION(generic_collect) }, // collect heaps every N (=10000) generic allocation calls +}; + +static void mi_option_init(mi_option_desc_t* desc); + +static bool mi_option_has_size_in_kib(mi_option_t option) { + return (option == mi_option_reserve_os_memory || option == mi_option_arena_reserve); +} + +void _mi_options_init(void) { + // called on process load + mi_add_stderr_output(); // now it safe to use stderr for output + for(int i = 0; i < _mi_option_last; i++ ) { + mi_option_t option = (mi_option_t)i; + long l = mi_option_get(option); MI_UNUSED(l); // initialize + } + mi_max_error_count = mi_option_get(mi_option_max_errors); + mi_max_warning_count = mi_option_get(mi_option_max_warnings); + #if MI_GUARDED + if (mi_option_get(mi_option_guarded_sample_rate) > 0) { + if (mi_option_is_enabled(mi_option_allow_large_os_pages)) { + mi_option_disable(mi_option_allow_large_os_pages); + _mi_warning_message("option 'allow_large_os_pages' is disabled to allow for guarded objects\n"); + } + } + #endif + if (mi_option_is_enabled(mi_option_verbose)) { mi_options_print(); } +} + +#define mi_stringifyx(str) #str // and stringify +#define mi_stringify(str) mi_stringifyx(str) // expand + +void mi_options_print(void) mi_attr_noexcept +{ + // show version + const int vermajor = MI_MALLOC_VERSION/100; + const int verminor = (MI_MALLOC_VERSION%100)/10; + const int verpatch = (MI_MALLOC_VERSION%10); + _mi_message("v%i.%i.%i%s%s (built on %s, %s)\n", vermajor, verminor, verpatch, + #if defined(MI_CMAKE_BUILD_TYPE) + ", " mi_stringify(MI_CMAKE_BUILD_TYPE) + #else + "" + #endif + , + #if defined(MI_GIT_DESCRIBE) + ", git " mi_stringify(MI_GIT_DESCRIBE) + #else + "" + #endif + , __DATE__, __TIME__); + + // show options + for (int i = 0; i < _mi_option_last; i++) { + mi_option_t option = (mi_option_t)i; + long l = mi_option_get(option); MI_UNUSED(l); // possibly initialize + mi_option_desc_t* desc = &options[option]; + _mi_message("option '%s': %ld %s\n", desc->name, desc->value, (mi_option_has_size_in_kib(option) ? "KiB" : "")); + } + + // show build configuration + _mi_message("debug level : %d\n", MI_DEBUG ); + _mi_message("secure level: %d\n", MI_SECURE ); + _mi_message("mem tracking: %s\n", MI_TRACK_TOOL); + #if MI_GUARDED + _mi_message("guarded build: %s\n", mi_option_get(mi_option_guarded_sample_rate) != 0 ? "enabled" : "disabled"); + #endif + #if MI_TSAN + _mi_message("thread santizer enabled\n"); + #endif +} + +long _mi_option_get_fast(mi_option_t option) { + mi_assert(option >= 0 && option < _mi_option_last); + mi_option_desc_t* desc = &options[option]; + mi_assert(desc->option == option); // index should match the option + //mi_assert(desc->init != UNINIT); + return desc->value; +} + + +mi_decl_nodiscard long mi_option_get(mi_option_t option) { + mi_assert(option >= 0 && option < _mi_option_last); + if (option < 0 || option >= _mi_option_last) return 0; + mi_option_desc_t* desc = &options[option]; + mi_assert(desc->option == option); // index should match the option + if mi_unlikely(desc->init == UNINIT) { + mi_option_init(desc); + } + return desc->value; +} + +mi_decl_nodiscard long mi_option_get_clamp(mi_option_t option, long min, long max) { + long x = mi_option_get(option); + return (x < min ? min : (x > max ? max : x)); +} + +mi_decl_nodiscard size_t mi_option_get_size(mi_option_t option) { + const long x = mi_option_get(option); + size_t size = (x < 0 ? 0 : (size_t)x); + if (mi_option_has_size_in_kib(option)) { + size *= MI_KiB; + } + return size; +} + +void mi_option_set(mi_option_t option, long value) { + mi_assert(option >= 0 && option < _mi_option_last); + if (option < 0 || option >= _mi_option_last) return; + mi_option_desc_t* desc = &options[option]; + mi_assert(desc->option == option); // index should match the option + desc->value = value; + desc->init = INITIALIZED; + // ensure min/max range; be careful to not recurse. + if (desc->option == mi_option_guarded_min && _mi_option_get_fast(mi_option_guarded_max) < value) { + mi_option_set(mi_option_guarded_max, value); + } + else if (desc->option == mi_option_guarded_max && _mi_option_get_fast(mi_option_guarded_min) > value) { + mi_option_set(mi_option_guarded_min, value); + } +} + +void mi_option_set_default(mi_option_t option, long value) { + mi_assert(option >= 0 && option < _mi_option_last); + if (option < 0 || option >= _mi_option_last) return; + mi_option_desc_t* desc = &options[option]; + if (desc->init != INITIALIZED) { + desc->value = value; + } +} + +mi_decl_nodiscard bool mi_option_is_enabled(mi_option_t option) { + return (mi_option_get(option) != 0); +} + +void mi_option_set_enabled(mi_option_t option, bool enable) { + mi_option_set(option, (enable ? 1 : 0)); +} + +void mi_option_set_enabled_default(mi_option_t option, bool enable) { + mi_option_set_default(option, (enable ? 1 : 0)); +} + +void mi_option_enable(mi_option_t option) { + mi_option_set_enabled(option,true); +} + +void mi_option_disable(mi_option_t option) { + mi_option_set_enabled(option,false); +} + +static void mi_cdecl mi_out_stderr(const char* msg, void* arg) { + MI_UNUSED(arg); + if (msg != NULL && msg[0] != 0) { + _mi_prim_out_stderr(msg); + } +} + +// Since an output function can be registered earliest in the `main` +// function we also buffer output that happens earlier. When +// an output function is registered it is called immediately with +// the output up to that point. +#ifndef MI_MAX_DELAY_OUTPUT +#define MI_MAX_DELAY_OUTPUT ((size_t)(16*1024)) +#endif +static char out_buf[MI_MAX_DELAY_OUTPUT+1]; +static _Atomic(size_t) out_len; + +static void mi_cdecl mi_out_buf(const char* msg, void* arg) { + MI_UNUSED(arg); + if (msg==NULL) return; + if (mi_atomic_load_relaxed(&out_len)>=MI_MAX_DELAY_OUTPUT) return; + size_t n = _mi_strlen(msg); + if (n==0) return; + // claim space + size_t start = mi_atomic_add_acq_rel(&out_len, n); + if (start >= MI_MAX_DELAY_OUTPUT) return; + // check bound + if (start+n >= MI_MAX_DELAY_OUTPUT) { + n = MI_MAX_DELAY_OUTPUT-start-1; + } + _mi_memcpy(&out_buf[start], msg, n); +} + +static void mi_out_buf_flush(mi_output_fun* out, bool no_more_buf, void* arg) { + if (out==NULL) return; + // claim (if `no_more_buf == true`, no more output will be added after this point) + size_t count = mi_atomic_add_acq_rel(&out_len, (no_more_buf ? MI_MAX_DELAY_OUTPUT : 1)); + // and output the current contents + if (count>MI_MAX_DELAY_OUTPUT) count = MI_MAX_DELAY_OUTPUT; + out_buf[count] = 0; + out(out_buf,arg); + if (!no_more_buf) { + out_buf[count] = '\n'; // if continue with the buffer, insert a newline + } +} + + +// Once this module is loaded, switch to this routine +// which outputs to stderr and the delayed output buffer. +static void mi_cdecl mi_out_buf_stderr(const char* msg, void* arg) { + mi_out_stderr(msg,arg); + mi_out_buf(msg,arg); +} + + + +// -------------------------------------------------------- +// Default output handler +// -------------------------------------------------------- + +// Should be atomic but gives errors on many platforms as generally we cannot cast a function pointer to a uintptr_t. +// For now, don't register output from multiple threads. +static mi_output_fun* volatile mi_out_default; // = NULL +static _Atomic(void*) mi_out_arg; // = NULL + +static mi_output_fun* mi_out_get_default(void** parg) { + if (parg != NULL) { *parg = mi_atomic_load_ptr_acquire(void,&mi_out_arg); } + mi_output_fun* out = mi_out_default; + return (out == NULL ? &mi_out_buf : out); +} + +void mi_register_output(mi_output_fun* out, void* arg) mi_attr_noexcept { + mi_out_default = (out == NULL ? &mi_out_stderr : out); // stop using the delayed output buffer + mi_atomic_store_ptr_release(void,&mi_out_arg, arg); + if (out!=NULL) mi_out_buf_flush(out,true,arg); // output all the delayed output now +} + +// add stderr to the delayed output after the module is loaded +static void mi_add_stderr_output(void) { + mi_assert_internal(mi_out_default == NULL); + mi_out_buf_flush(&mi_out_stderr, false, NULL); // flush current contents to stderr + mi_out_default = &mi_out_buf_stderr; // and add stderr to the delayed output +} + +// -------------------------------------------------------- +// Messages, all end up calling `_mi_fputs`. +// -------------------------------------------------------- +static _Atomic(size_t) error_count; // = 0; // when >= max_error_count stop emitting errors +static _Atomic(size_t) warning_count; // = 0; // when >= max_warning_count stop emitting warnings + +// When overriding malloc, we may recurse into mi_vfprintf if an allocation +// inside the C runtime causes another message. +// In some cases (like on macOS) the loader already allocates which +// calls into mimalloc; if we then access thread locals (like `recurse`) +// this may crash as the access may call _tlv_bootstrap that tries to +// (recursively) invoke malloc again to allocate space for the thread local +// variables on demand. This is why we use a _mi_preloading test on such +// platforms. However, C code generator may move the initial thread local address +// load before the `if` and we therefore split it out in a separate function. +static mi_decl_thread bool recurse = false; + +static mi_decl_noinline bool mi_recurse_enter_prim(void) { + if (recurse) return false; + recurse = true; + return true; +} + +static mi_decl_noinline void mi_recurse_exit_prim(void) { + recurse = false; +} + +static bool mi_recurse_enter(void) { + #if defined(__APPLE__) || defined(MI_TLS_RECURSE_GUARD) + if (_mi_preloading()) return false; + #endif + return mi_recurse_enter_prim(); +} + +static void mi_recurse_exit(void) { + #if defined(__APPLE__) || defined(MI_TLS_RECURSE_GUARD) + if (_mi_preloading()) return; + #endif + mi_recurse_exit_prim(); +} + +void _mi_fputs(mi_output_fun* out, void* arg, const char* prefix, const char* message) { + if (out==NULL || (void*)out==(void*)stdout || (void*)out==(void*)stderr) { // TODO: use mi_out_stderr for stderr? + if (!mi_recurse_enter()) return; + out = mi_out_get_default(&arg); + if (prefix != NULL) out(prefix, arg); + out(message, arg); + mi_recurse_exit(); + } + else { + if (prefix != NULL) out(prefix, arg); + out(message, arg); + } +} + +// Define our own limited `fprintf` that avoids memory allocation. +// We do this using `_mi_vsnprintf` with a limited buffer. +static void mi_vfprintf( mi_output_fun* out, void* arg, const char* prefix, const char* fmt, va_list args ) { + char buf[512]; + if (fmt==NULL) return; + if (!mi_recurse_enter()) return; + _mi_vsnprintf(buf, sizeof(buf)-1, fmt, args); + mi_recurse_exit(); + _mi_fputs(out,arg,prefix,buf); +} + +void _mi_fprintf( mi_output_fun* out, void* arg, const char* fmt, ... ) { + va_list args; + va_start(args,fmt); + mi_vfprintf(out,arg,NULL,fmt,args); + va_end(args); +} + +static void mi_vfprintf_thread(mi_output_fun* out, void* arg, const char* prefix, const char* fmt, va_list args) { + if (prefix != NULL && _mi_strnlen(prefix,33) <= 32 && !_mi_is_main_thread()) { + char tprefix[64]; + _mi_snprintf(tprefix, sizeof(tprefix), "%sthread 0x%tx: ", prefix, (uintptr_t)_mi_thread_id()); + mi_vfprintf(out, arg, tprefix, fmt, args); + } + else { + mi_vfprintf(out, arg, prefix, fmt, args); + } +} + +void _mi_message(const char* fmt, ...) { + va_list args; + va_start(args, fmt); + mi_vfprintf_thread(NULL, NULL, "mimalloc: ", fmt, args); + va_end(args); +} + +void _mi_trace_message(const char* fmt, ...) { + if (mi_option_get(mi_option_verbose) <= 1) return; // only with verbose level 2 or higher + va_list args; + va_start(args, fmt); + mi_vfprintf_thread(NULL, NULL, "mimalloc: ", fmt, args); + va_end(args); +} + +void _mi_verbose_message(const char* fmt, ...) { + if (!mi_option_is_enabled(mi_option_verbose)) return; + va_list args; + va_start(args,fmt); + mi_vfprintf(NULL, NULL, "mimalloc: ", fmt, args); + va_end(args); +} + +static void mi_show_error_message(const char* fmt, va_list args) { + if (!mi_option_is_enabled(mi_option_verbose)) { + if (!mi_option_is_enabled(mi_option_show_errors)) return; + if (mi_max_error_count >= 0 && (long)mi_atomic_increment_acq_rel(&error_count) > mi_max_error_count) return; + } + mi_vfprintf_thread(NULL, NULL, "mimalloc: error: ", fmt, args); +} + +void _mi_warning_message(const char* fmt, ...) { + if (!mi_option_is_enabled(mi_option_verbose)) { + if (!mi_option_is_enabled(mi_option_show_errors)) return; + if (mi_max_warning_count >= 0 && (long)mi_atomic_increment_acq_rel(&warning_count) > mi_max_warning_count) return; + } + va_list args; + va_start(args,fmt); + mi_vfprintf_thread(NULL, NULL, "mimalloc: warning: ", fmt, args); + va_end(args); +} + + +#if MI_DEBUG +void _mi_assert_fail(const char* assertion, const char* fname, unsigned line, const char* func ) { + _mi_fprintf(NULL, NULL, "mimalloc: assertion failed: at \"%s\":%u, %s\n assertion: \"%s\"\n", fname, line, (func==NULL?"":func), assertion); + abort(); +} +#endif + +// -------------------------------------------------------- +// Errors +// -------------------------------------------------------- + +static mi_error_fun* volatile mi_error_handler; // = NULL +static _Atomic(void*) mi_error_arg; // = NULL + +static void mi_error_default(int err) { + MI_UNUSED(err); +#if (MI_DEBUG>0) + if (err==EFAULT) { + #ifdef _MSC_VER + __debugbreak(); + #endif + abort(); + } +#endif +#if (MI_SECURE>0) + if (err==EFAULT) { // abort on serious errors in secure mode (corrupted meta-data) + abort(); + } +#endif +#if defined(MI_XMALLOC) + if (err==ENOMEM || err==EOVERFLOW) { // abort on memory allocation fails in xmalloc mode + abort(); + } +#endif +} + +void mi_register_error(mi_error_fun* fun, void* arg) { + mi_error_handler = fun; // can be NULL + mi_atomic_store_ptr_release(void,&mi_error_arg, arg); +} + +void _mi_error_message(int err, const char* fmt, ...) { + // show detailed error message + va_list args; + va_start(args, fmt); + mi_show_error_message(fmt, args); + va_end(args); + // and call the error handler which may abort (or return normally) + if (mi_error_handler != NULL) { + mi_error_handler(err, mi_atomic_load_ptr_acquire(void,&mi_error_arg)); + } + else { + mi_error_default(err); + } +} + +// -------------------------------------------------------- +// Initialize options by checking the environment +// -------------------------------------------------------- + +// TODO: implement ourselves to reduce dependencies on the C runtime +#include // strtol +#include // strstr + + +static void mi_option_init(mi_option_desc_t* desc) { + // Read option value from the environment + char s[64 + 1]; + char buf[64+1]; + _mi_strlcpy(buf, "mimalloc_", sizeof(buf)); + _mi_strlcat(buf, desc->name, sizeof(buf)); + bool found = _mi_getenv(buf, s, sizeof(s)); + if (!found && desc->legacy_name != NULL) { + _mi_strlcpy(buf, "mimalloc_", sizeof(buf)); + _mi_strlcat(buf, desc->legacy_name, sizeof(buf)); + found = _mi_getenv(buf, s, sizeof(s)); + if (found) { + _mi_warning_message("environment option \"mimalloc_%s\" is deprecated -- use \"mimalloc_%s\" instead.\n", desc->legacy_name, desc->name); + } + } + + if (found) { + size_t len = _mi_strnlen(s, sizeof(buf) - 1); + for (size_t i = 0; i < len; i++) { + buf[i] = _mi_toupper(s[i]); + } + buf[len] = 0; + if (buf[0] == 0 || strstr("1;TRUE;YES;ON", buf) != NULL) { + desc->value = 1; + desc->init = INITIALIZED; + } + else if (strstr("0;FALSE;NO;OFF", buf) != NULL) { + desc->value = 0; + desc->init = INITIALIZED; + } + else { + char* end = buf; + long value = strtol(buf, &end, 10); + if (mi_option_has_size_in_kib(desc->option)) { + // this option is interpreted in KiB to prevent overflow of `long` for large allocations + // (long is 32-bit on 64-bit windows, which allows for 4TiB max.) + size_t size = (value < 0 ? 0 : (size_t)value); + bool overflow = false; + if (*end == 'K') { end++; } + else if (*end == 'M') { overflow = mi_mul_overflow(size,MI_KiB,&size); end++; } + else if (*end == 'G') { overflow = mi_mul_overflow(size,MI_MiB,&size); end++; } + else if (*end == 'T') { overflow = mi_mul_overflow(size,MI_GiB,&size); end++; } + else { size = (size + MI_KiB - 1) / MI_KiB; } + if (end[0] == 'I' && end[1] == 'B') { end += 2; } // KiB, MiB, GiB, TiB + else if (*end == 'B') { end++; } // Kb, Mb, Gb, Tb + if (overflow || size > MI_MAX_ALLOC_SIZE) { size = (MI_MAX_ALLOC_SIZE / MI_KiB); } + value = (size > LONG_MAX ? LONG_MAX : (long)size); + } + if (*end == 0) { + mi_option_set(desc->option, value); + } + else { + // set `init` first to avoid recursion through _mi_warning_message on mimalloc_verbose. + desc->init = DEFAULTED; + if (desc->option == mi_option_verbose && desc->value == 0) { + // if the 'mimalloc_verbose' env var has a bogus value we'd never know + // (since the value defaults to 'off') so in that case briefly enable verbose + desc->value = 1; + _mi_warning_message("environment option mimalloc_%s has an invalid value.\n", desc->name); + desc->value = 0; + } + else { + _mi_warning_message("environment option mimalloc_%s has an invalid value.\n", desc->name); + } + } + } + mi_assert_internal(desc->init != UNINIT); + } + else if (!_mi_preloading()) { + desc->init = DEFAULTED; + } +} diff --git a/compat/mimalloc/os.c b/compat/mimalloc/os.c new file mode 100644 index 00000000000000..2472b8030b3c61 --- /dev/null +++ b/compat/mimalloc/os.c @@ -0,0 +1,713 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2023, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ +#include "mimalloc.h" +#include "mimalloc/internal.h" +#include "mimalloc/atomic.h" +#include "mimalloc/prim.h" + +#define mi_os_stat_increase(stat,amount) _mi_stat_increase(&_mi_stats_main.stat, amount) +#define mi_os_stat_decrease(stat,amount) _mi_stat_decrease(&_mi_stats_main.stat, amount) +#define mi_os_stat_counter_increase(stat,inc) _mi_stat_counter_increase(&_mi_stats_main.stat, inc) + +/* ----------------------------------------------------------- + Initialization. +----------------------------------------------------------- */ +#ifndef MI_DEFAULT_VIRTUAL_ADDRESS_BITS +#if MI_INTPTR_SIZE < 8 +#define MI_DEFAULT_VIRTUAL_ADDRESS_BITS 32 +#else +#define MI_DEFAULT_VIRTUAL_ADDRESS_BITS 48 +#endif +#endif + +#ifndef MI_DEFAULT_PHYSICAL_MEMORY_IN_KIB +#if MI_INTPTR_SIZE < 8 +#define MI_DEFAULT_PHYSICAL_MEMORY_IN_KIB 4*MI_MiB // 4 GiB +#else +#define MI_DEFAULT_PHYSICAL_MEMORY_IN_KIB 32*MI_MiB // 32 GiB +#endif +#endif + +static mi_os_mem_config_t mi_os_mem_config = { + 4096, // page size + 0, // large page size (usually 2MiB) + 4096, // allocation granularity + MI_DEFAULT_PHYSICAL_MEMORY_IN_KIB, + MI_DEFAULT_VIRTUAL_ADDRESS_BITS, + true, // has overcommit? (if true we use MAP_NORESERVE on mmap systems) + false, // can we partially free allocated blocks? (on mmap systems we can free anywhere in a mapped range, but on Windows we must free the entire span) + true // has virtual reserve? (if true we can reserve virtual address space without using commit or physical memory) +}; + +bool _mi_os_has_overcommit(void) { + return mi_os_mem_config.has_overcommit; +} + +bool _mi_os_has_virtual_reserve(void) { + return mi_os_mem_config.has_virtual_reserve; +} + + +// OS (small) page size +size_t _mi_os_page_size(void) { + return mi_os_mem_config.page_size; +} + +// if large OS pages are supported (2 or 4MiB), then return the size, otherwise return the small page size (4KiB) +size_t _mi_os_large_page_size(void) { + return (mi_os_mem_config.large_page_size != 0 ? mi_os_mem_config.large_page_size : _mi_os_page_size()); +} + +bool _mi_os_use_large_page(size_t size, size_t alignment) { + // if we have access, check the size and alignment requirements + if (mi_os_mem_config.large_page_size == 0 || !mi_option_is_enabled(mi_option_allow_large_os_pages)) return false; + return ((size % mi_os_mem_config.large_page_size) == 0 && (alignment % mi_os_mem_config.large_page_size) == 0); +} + +// round to a good OS allocation size (bounded by max 12.5% waste) +size_t _mi_os_good_alloc_size(size_t size) { + size_t align_size; + if (size < 512*MI_KiB) align_size = _mi_os_page_size(); + else if (size < 2*MI_MiB) align_size = 64*MI_KiB; + else if (size < 8*MI_MiB) align_size = 256*MI_KiB; + else if (size < 32*MI_MiB) align_size = 1*MI_MiB; + else align_size = 4*MI_MiB; + if mi_unlikely(size >= (SIZE_MAX - align_size)) return size; // possible overflow? + return _mi_align_up(size, align_size); +} + +void _mi_os_init(void) { + _mi_prim_mem_init(&mi_os_mem_config); +} + + +/* ----------------------------------------------------------- + Util +-------------------------------------------------------------- */ +bool _mi_os_decommit(void* addr, size_t size); +bool _mi_os_commit(void* addr, size_t size, bool* is_zero); + + +/* ----------------------------------------------------------- + aligned hinting +-------------------------------------------------------------- */ + +// On systems with enough virtual address bits, we can do efficient aligned allocation by using +// the 2TiB to 30TiB area to allocate those. If we have at least 46 bits of virtual address +// space (64TiB) we use this technique. (but see issue #939) +#if (MI_INTPTR_SIZE >= 8) && !defined(MI_NO_ALIGNED_HINT) +static mi_decl_cache_align _Atomic(uintptr_t)aligned_base; + +// Return a MI_SEGMENT_SIZE aligned address that is probably available. +// If this returns NULL, the OS will determine the address but on some OS's that may not be +// properly aligned which can be more costly as it needs to be adjusted afterwards. +// For a size > 1GiB this always returns NULL in order to guarantee good ASLR randomization; +// (otherwise an initial large allocation of say 2TiB has a 50% chance to include (known) addresses +// in the middle of the 2TiB - 6TiB address range (see issue #372)) + +#define MI_HINT_BASE ((uintptr_t)2 << 40) // 2TiB start +#define MI_HINT_AREA ((uintptr_t)4 << 40) // upto 6TiB (since before win8 there is "only" 8TiB available to processes) +#define MI_HINT_MAX ((uintptr_t)30 << 40) // wrap after 30TiB (area after 32TiB is used for huge OS pages) + +void* _mi_os_get_aligned_hint(size_t try_alignment, size_t size) +{ + if (try_alignment <= 1 || try_alignment > MI_SEGMENT_SIZE) return NULL; + if (mi_os_mem_config.virtual_address_bits < 46) return NULL; // < 64TiB virtual address space + size = _mi_align_up(size, MI_SEGMENT_SIZE); + if (size > 1*MI_GiB) return NULL; // guarantee the chance of fixed valid address is at most 1/(MI_HINT_AREA / 1<<30) = 1/4096. + #if (MI_SECURE>0) + size += MI_SEGMENT_SIZE; // put in `MI_SEGMENT_SIZE` virtual gaps between hinted blocks; this splits VLA's but increases guarded areas. + #endif + + uintptr_t hint = mi_atomic_add_acq_rel(&aligned_base, size); + if (hint == 0 || hint > MI_HINT_MAX) { // wrap or initialize + uintptr_t init = MI_HINT_BASE; + #if (MI_SECURE>0 || MI_DEBUG==0) // security: randomize start of aligned allocations unless in debug mode + uintptr_t r = _mi_heap_random_next(mi_prim_get_default_heap()); + init = init + ((MI_SEGMENT_SIZE * ((r>>17) & 0xFFFFF)) % MI_HINT_AREA); // (randomly 20 bits)*4MiB == 0 to 4TiB + #endif + uintptr_t expected = hint + size; + mi_atomic_cas_strong_acq_rel(&aligned_base, &expected, init); + hint = mi_atomic_add_acq_rel(&aligned_base, size); // this may still give 0 or > MI_HINT_MAX but that is ok, it is a hint after all + } + if (hint%try_alignment != 0) return NULL; + return (void*)hint; +} +#else +void* _mi_os_get_aligned_hint(size_t try_alignment, size_t size) { + MI_UNUSED(try_alignment); MI_UNUSED(size); + return NULL; +} +#endif + +/* ----------------------------------------------------------- + Free memory +-------------------------------------------------------------- */ + +static void mi_os_free_huge_os_pages(void* p, size_t size); + +static void mi_os_prim_free(void* addr, size_t size, size_t commit_size) { + mi_assert_internal((size % _mi_os_page_size()) == 0); + if (addr == NULL || size == 0) return; // || _mi_os_is_huge_reserved(addr) + int err = _mi_prim_free(addr, size); + if (err != 0) { + _mi_warning_message("unable to free OS memory (error: %d (0x%x), size: 0x%zx bytes, address: %p)\n", err, err, size, addr); + } + if (commit_size > 0) { + mi_os_stat_decrease(committed, commit_size); + } + mi_os_stat_decrease(reserved, size); +} + +void _mi_os_free_ex(void* addr, size_t size, bool still_committed, mi_memid_t memid) { + if (mi_memkind_is_os(memid.memkind)) { + size_t csize = memid.mem.os.size; + if (csize==0) { _mi_os_good_alloc_size(size); } + size_t commit_size = (still_committed ? csize : 0); + void* base = addr; + // different base? (due to alignment) + if (memid.mem.os.base != base) { + mi_assert(memid.mem.os.base <= addr); + base = memid.mem.os.base; + const size_t diff = (uint8_t*)addr - (uint8_t*)memid.mem.os.base; + if (memid.mem.os.size==0) { + csize += diff; + } + if (still_committed) { + commit_size -= diff; // the (addr-base) part was already un-committed + } + } + // free it + if (memid.memkind == MI_MEM_OS_HUGE) { + mi_assert(memid.is_pinned); + mi_os_free_huge_os_pages(base, csize); + } + else { + mi_os_prim_free(base, csize, (still_committed ? commit_size : 0)); + } + } + else { + // nothing to do + mi_assert(memid.memkind < MI_MEM_OS); + } +} + +void _mi_os_free(void* p, size_t size, mi_memid_t memid) { + _mi_os_free_ex(p, size, true, memid); +} + + +/* ----------------------------------------------------------- + Primitive allocation from the OS. +-------------------------------------------------------------- */ + +// Note: the `try_alignment` is just a hint and the returned pointer is not guaranteed to be aligned. +// Also `hint_addr` is a hint and may be ignored. +static void* mi_os_prim_alloc_at(void* hint_addr, size_t size, size_t try_alignment, bool commit, bool allow_large, bool* is_large, bool* is_zero) { + mi_assert_internal(size > 0 && (size % _mi_os_page_size()) == 0); + mi_assert_internal(is_zero != NULL); + mi_assert_internal(is_large != NULL); + if (size == 0) return NULL; + if (!commit) { allow_large = false; } + if (try_alignment == 0) { try_alignment = 1; } // avoid 0 to ensure there will be no divide by zero when aligning + *is_zero = false; + void* p = NULL; + int err = _mi_prim_alloc(hint_addr, size, try_alignment, commit, allow_large, is_large, is_zero, &p); + if (err != 0) { + _mi_warning_message("unable to allocate OS memory (error: %d (0x%x), addr: %p, size: 0x%zx bytes, align: 0x%zx, commit: %d, allow large: %d)\n", err, err, hint_addr, size, try_alignment, commit, allow_large); + } + + + + mi_os_stat_counter_increase(mmap_calls, 1); + if (p != NULL) { + mi_os_stat_increase(reserved, size); + if (commit) { + mi_os_stat_increase(committed, size); + // seems needed for asan (or `mimalloc-test-api` fails) + #ifdef MI_TRACK_ASAN + if (*is_zero) { mi_track_mem_defined(p,size); } + else { mi_track_mem_undefined(p,size); } + #endif + } + } + return p; +} + +static void* mi_os_prim_alloc(size_t size, size_t try_alignment, bool commit, bool allow_large, bool* is_large, bool* is_zero) { + return mi_os_prim_alloc_at(NULL, size, try_alignment, commit, allow_large, is_large, is_zero); +} + + +// Primitive aligned allocation from the OS. +// This function guarantees the allocated memory is aligned. +static void* mi_os_prim_alloc_aligned(size_t size, size_t alignment, bool commit, bool allow_large, bool* is_large, bool* is_zero, void** base) { + mi_assert_internal(alignment >= _mi_os_page_size() && ((alignment & (alignment - 1)) == 0)); + mi_assert_internal(size > 0 && (size % _mi_os_page_size()) == 0); + mi_assert_internal(is_large != NULL); + mi_assert_internal(is_zero != NULL); + mi_assert_internal(base != NULL); + if (!commit) allow_large = false; + if (!(alignment >= _mi_os_page_size() && ((alignment & (alignment - 1)) == 0))) return NULL; + size = _mi_align_up(size, _mi_os_page_size()); + + // try first with a requested alignment hint (this will usually be aligned directly on Win 10+ or BSD) + void* p = mi_os_prim_alloc(size, alignment, commit, allow_large, is_large, is_zero); + if (p == NULL) return NULL; + + // aligned already? + if (((uintptr_t)p % alignment) == 0) { + *base = p; + } + else { + // if not aligned, free it, overallocate, and unmap around it + #if !MI_TRACK_ASAN + _mi_warning_message("unable to allocate aligned OS memory directly, fall back to over-allocation (size: 0x%zx bytes, address: %p, alignment: 0x%zx, commit: %d)\n", size, p, alignment, commit); + #endif + if (p != NULL) { mi_os_prim_free(p, size, (commit ? size : 0)); } + if (size >= (SIZE_MAX - alignment)) return NULL; // overflow + const size_t over_size = size + alignment; + + if (!mi_os_mem_config.has_partial_free) { // win32 virtualAlloc cannot free parts of an allocated block + // over-allocate uncommitted (virtual) memory + p = mi_os_prim_alloc(over_size, 1 /*alignment*/, false /* commit? */, false /* allow_large */, is_large, is_zero); + if (p == NULL) return NULL; + + // set p to the aligned part in the full region + // note: this is dangerous on Windows as VirtualFree needs the actual base pointer + // this is handled though by having the `base` field in the memid's + *base = p; // remember the base + p = mi_align_up_ptr(p, alignment); + + // explicitly commit only the aligned part + if (commit) { + _mi_os_commit(p, size, NULL); + } + } + else { // mmap can free inside an allocation + // overallocate... + p = mi_os_prim_alloc(over_size, 1, commit, false, is_large, is_zero); + if (p == NULL) return NULL; + + // and selectively unmap parts around the over-allocated area. + void* aligned_p = mi_align_up_ptr(p, alignment); + size_t pre_size = (uint8_t*)aligned_p - (uint8_t*)p; + size_t mid_size = _mi_align_up(size, _mi_os_page_size()); + size_t post_size = over_size - pre_size - mid_size; + mi_assert_internal(pre_size < over_size&& post_size < over_size&& mid_size >= size); + if (pre_size > 0) { mi_os_prim_free(p, pre_size, (commit ? pre_size : 0)); } + if (post_size > 0) { mi_os_prim_free((uint8_t*)aligned_p + mid_size, post_size, (commit ? post_size : 0)); } + // we can return the aligned pointer on `mmap` systems + p = aligned_p; + *base = aligned_p; // since we freed the pre part, `*base == p`. + } + } + + mi_assert_internal(p == NULL || (p != NULL && *base != NULL && ((uintptr_t)p % alignment) == 0)); + return p; +} + + +/* ----------------------------------------------------------- + OS API: alloc and alloc_aligned +----------------------------------------------------------- */ + +void* _mi_os_alloc(size_t size, mi_memid_t* memid) { + *memid = _mi_memid_none(); + if (size == 0) return NULL; + size = _mi_os_good_alloc_size(size); + bool os_is_large = false; + bool os_is_zero = false; + void* p = mi_os_prim_alloc(size, 0, true, false, &os_is_large, &os_is_zero); + if (p != NULL) { + *memid = _mi_memid_create_os(true, os_is_zero, os_is_large); + } + return p; +} + +void* _mi_os_alloc_aligned(size_t size, size_t alignment, bool commit, bool allow_large, mi_memid_t* memid) +{ + MI_UNUSED(&_mi_os_get_aligned_hint); // suppress unused warnings + *memid = _mi_memid_none(); + if (size == 0) return NULL; + size = _mi_os_good_alloc_size(size); + alignment = _mi_align_up(alignment, _mi_os_page_size()); + + bool os_is_large = false; + bool os_is_zero = false; + void* os_base = NULL; + void* p = mi_os_prim_alloc_aligned(size, alignment, commit, allow_large, &os_is_large, &os_is_zero, &os_base ); + if (p != NULL) { + *memid = _mi_memid_create_os(commit, os_is_zero, os_is_large); + memid->mem.os.base = os_base; + // memid->mem.os.alignment = alignment; + memid->mem.os.size += ((uint8_t*)p - (uint8_t*)os_base); // todo: return from prim_alloc_aligned + } + return p; +} + +/* ----------------------------------------------------------- + OS aligned allocation with an offset. This is used + for large alignments > MI_BLOCK_ALIGNMENT_MAX. We use a large mimalloc + page where the object can be aligned at an offset from the start of the segment. + As we may need to overallocate, we need to free such pointers using `mi_free_aligned` + to use the actual start of the memory region. +----------------------------------------------------------- */ + +void* _mi_os_alloc_aligned_at_offset(size_t size, size_t alignment, size_t offset, bool commit, bool allow_large, mi_memid_t* memid) { + mi_assert(offset <= MI_SEGMENT_SIZE); + mi_assert(offset <= size); + mi_assert((alignment % _mi_os_page_size()) == 0); + *memid = _mi_memid_none(); + if (offset > MI_SEGMENT_SIZE) return NULL; + if (offset == 0) { + // regular aligned allocation + return _mi_os_alloc_aligned(size, alignment, commit, allow_large, memid); + } + else { + // overallocate to align at an offset + const size_t extra = _mi_align_up(offset, alignment) - offset; + const size_t oversize = size + extra; + void* const start = _mi_os_alloc_aligned(oversize, alignment, commit, allow_large, memid); + if (start == NULL) return NULL; + + void* const p = (uint8_t*)start + extra; + mi_assert(_mi_is_aligned((uint8_t*)p + offset, alignment)); + // decommit the overallocation at the start + if (commit && extra > _mi_os_page_size()) { + _mi_os_decommit(start, extra); + } + return p; + } +} + +/* ----------------------------------------------------------- + OS memory API: reset, commit, decommit, protect, unprotect. +----------------------------------------------------------- */ + +// OS page align within a given area, either conservative (pages inside the area only), +// or not (straddling pages outside the area is possible) +static void* mi_os_page_align_areax(bool conservative, void* addr, size_t size, size_t* newsize) { + mi_assert(addr != NULL && size > 0); + if (newsize != NULL) *newsize = 0; + if (size == 0 || addr == NULL) return NULL; + + // page align conservatively within the range + void* start = (conservative ? mi_align_up_ptr(addr, _mi_os_page_size()) + : mi_align_down_ptr(addr, _mi_os_page_size())); + void* end = (conservative ? mi_align_down_ptr((uint8_t*)addr + size, _mi_os_page_size()) + : mi_align_up_ptr((uint8_t*)addr + size, _mi_os_page_size())); + ptrdiff_t diff = (uint8_t*)end - (uint8_t*)start; + if (diff <= 0) return NULL; + + mi_assert_internal((conservative && (size_t)diff <= size) || (!conservative && (size_t)diff >= size)); + if (newsize != NULL) *newsize = (size_t)diff; + return start; +} + +static void* mi_os_page_align_area_conservative(void* addr, size_t size, size_t* newsize) { + return mi_os_page_align_areax(true, addr, size, newsize); +} + +bool _mi_os_commit_ex(void* addr, size_t size, bool* is_zero, size_t stat_size) { + if (is_zero != NULL) { *is_zero = false; } + mi_os_stat_increase(committed, stat_size); // use size for precise commit vs. decommit + mi_os_stat_counter_increase(commit_calls, 1); + + // page align range + size_t csize; + void* start = mi_os_page_align_areax(false /* conservative? */, addr, size, &csize); + if (csize == 0) return true; + + // commit + bool os_is_zero = false; + int err = _mi_prim_commit(start, csize, &os_is_zero); + if (err != 0) { + _mi_warning_message("cannot commit OS memory (error: %d (0x%x), address: %p, size: 0x%zx bytes)\n", err, err, start, csize); + return false; + } + if (os_is_zero && is_zero != NULL) { + *is_zero = true; + mi_assert_expensive(mi_mem_is_zero(start, csize)); + } + // note: the following seems required for asan (otherwise `mimalloc-test-stress` fails) + #ifdef MI_TRACK_ASAN + if (os_is_zero) { mi_track_mem_defined(start,csize); } + else { mi_track_mem_undefined(start,csize); } + #endif + return true; +} + +bool _mi_os_commit(void* addr, size_t size, bool* is_zero) { + return _mi_os_commit_ex(addr, size, is_zero, size); +} + +static bool mi_os_decommit_ex(void* addr, size_t size, bool* needs_recommit, size_t stat_size) { + mi_assert_internal(needs_recommit!=NULL); + mi_os_stat_decrease(committed, stat_size); + + // page align + size_t csize; + void* start = mi_os_page_align_area_conservative(addr, size, &csize); + if (csize == 0) return true; + + // decommit + *needs_recommit = true; + int err = _mi_prim_decommit(start,csize,needs_recommit); + if (err != 0) { + _mi_warning_message("cannot decommit OS memory (error: %d (0x%x), address: %p, size: 0x%zx bytes)\n", err, err, start, csize); + } + mi_assert_internal(err == 0); + return (err == 0); +} + +bool _mi_os_decommit(void* addr, size_t size) { + bool needs_recommit; + return mi_os_decommit_ex(addr, size, &needs_recommit, size); +} + + +// Signal to the OS that the address range is no longer in use +// but may be used later again. This will release physical memory +// pages and reduce swapping while keeping the memory committed. +// We page align to a conservative area inside the range to reset. +bool _mi_os_reset(void* addr, size_t size) { + // page align conservatively within the range + size_t csize; + void* start = mi_os_page_align_area_conservative(addr, size, &csize); + if (csize == 0) return true; // || _mi_os_is_huge_reserved(addr) + mi_os_stat_increase(reset, csize); + mi_os_stat_counter_increase(reset_calls, 1); + + #if (MI_DEBUG>1) && !MI_SECURE && !MI_TRACK_ENABLED // && !MI_TSAN + memset(start, 0, csize); // pretend it is eagerly reset + #endif + + int err = _mi_prim_reset(start, csize); + if (err != 0) { + _mi_warning_message("cannot reset OS memory (error: %d (0x%x), address: %p, size: 0x%zx bytes)\n", err, err, start, csize); + } + return (err == 0); +} + + +// either resets or decommits memory, returns true if the memory needs +// to be recommitted if it is to be re-used later on. +bool _mi_os_purge_ex(void* p, size_t size, bool allow_reset, size_t stat_size) +{ + if (mi_option_get(mi_option_purge_delay) < 0) return false; // is purging allowed? + mi_os_stat_counter_increase(purge_calls, 1); + mi_os_stat_increase(purged, size); + + if (mi_option_is_enabled(mi_option_purge_decommits) && // should decommit? + !_mi_preloading()) // don't decommit during preloading (unsafe) + { + bool needs_recommit = true; + mi_os_decommit_ex(p, size, &needs_recommit, stat_size); + return needs_recommit; + } + else { + if (allow_reset) { // this can sometimes be not allowed if the range is not fully committed + _mi_os_reset(p, size); + } + return false; // needs no recommit + } +} + +// either resets or decommits memory, returns true if the memory needs +// to be recommitted if it is to be re-used later on. +bool _mi_os_purge(void* p, size_t size) { + return _mi_os_purge_ex(p, size, true, size); +} + +// Protect a region in memory to be not accessible. +static bool mi_os_protectx(void* addr, size_t size, bool protect) { + // page align conservatively within the range + size_t csize = 0; + void* start = mi_os_page_align_area_conservative(addr, size, &csize); + if (csize == 0) return false; + /* + if (_mi_os_is_huge_reserved(addr)) { + _mi_warning_message("cannot mprotect memory allocated in huge OS pages\n"); + } + */ + int err = _mi_prim_protect(start,csize,protect); + if (err != 0) { + _mi_warning_message("cannot %s OS memory (error: %d (0x%x), address: %p, size: 0x%zx bytes)\n", (protect ? "protect" : "unprotect"), err, err, start, csize); + } + return (err == 0); +} + +bool _mi_os_protect(void* addr, size_t size) { + return mi_os_protectx(addr, size, true); +} + +bool _mi_os_unprotect(void* addr, size_t size) { + return mi_os_protectx(addr, size, false); +} + + + +/* ---------------------------------------------------------------------------- +Support for allocating huge OS pages (1Gib) that are reserved up-front +and possibly associated with a specific NUMA node. (use `numa_node>=0`) +-----------------------------------------------------------------------------*/ +#define MI_HUGE_OS_PAGE_SIZE (MI_GiB) + + +#if (MI_INTPTR_SIZE >= 8) +// To ensure proper alignment, use our own area for huge OS pages +static mi_decl_cache_align _Atomic(uintptr_t) mi_huge_start; // = 0 + +// Claim an aligned address range for huge pages +static uint8_t* mi_os_claim_huge_pages(size_t pages, size_t* total_size) { + if (total_size != NULL) *total_size = 0; + const size_t size = pages * MI_HUGE_OS_PAGE_SIZE; + + uintptr_t start = 0; + uintptr_t end = 0; + uintptr_t huge_start = mi_atomic_load_relaxed(&mi_huge_start); + do { + start = huge_start; + if (start == 0) { + // Initialize the start address after the 32TiB area + start = ((uintptr_t)32 << 40); // 32TiB virtual start address + #if (MI_SECURE>0 || MI_DEBUG==0) // security: randomize start of huge pages unless in debug mode + uintptr_t r = _mi_heap_random_next(mi_prim_get_default_heap()); + start = start + ((uintptr_t)MI_HUGE_OS_PAGE_SIZE * ((r>>17) & 0x0FFF)); // (randomly 12bits)*1GiB == between 0 to 4TiB + #endif + } + end = start + size; + mi_assert_internal(end % MI_SEGMENT_SIZE == 0); + } while (!mi_atomic_cas_strong_acq_rel(&mi_huge_start, &huge_start, end)); + + if (total_size != NULL) *total_size = size; + return (uint8_t*)start; +} +#else +static uint8_t* mi_os_claim_huge_pages(size_t pages, size_t* total_size) { + MI_UNUSED(pages); + if (total_size != NULL) *total_size = 0; + return NULL; +} +#endif + +// Allocate MI_SEGMENT_SIZE aligned huge pages +void* _mi_os_alloc_huge_os_pages(size_t pages, int numa_node, mi_msecs_t max_msecs, size_t* pages_reserved, size_t* psize, mi_memid_t* memid) { + *memid = _mi_memid_none(); + if (psize != NULL) *psize = 0; + if (pages_reserved != NULL) *pages_reserved = 0; + size_t size = 0; + uint8_t* start = mi_os_claim_huge_pages(pages, &size); + if (start == NULL) return NULL; // or 32-bit systems + + // Allocate one page at the time but try to place them contiguously + // We allocate one page at the time to be able to abort if it takes too long + // or to at least allocate as many as available on the system. + mi_msecs_t start_t = _mi_clock_start(); + size_t page = 0; + bool all_zero = true; + while (page < pages) { + // allocate a page + bool is_zero = false; + void* addr = start + (page * MI_HUGE_OS_PAGE_SIZE); + void* p = NULL; + int err = _mi_prim_alloc_huge_os_pages(addr, MI_HUGE_OS_PAGE_SIZE, numa_node, &is_zero, &p); + if (!is_zero) { all_zero = false; } + if (err != 0) { + _mi_warning_message("unable to allocate huge OS page (error: %d (0x%x), address: %p, size: %zx bytes)\n", err, err, addr, MI_HUGE_OS_PAGE_SIZE); + break; + } + + // Did we succeed at a contiguous address? + if (p != addr) { + // no success, issue a warning and break + if (p != NULL) { + _mi_warning_message("could not allocate contiguous huge OS page %zu at %p\n", page, addr); + mi_os_prim_free(p, MI_HUGE_OS_PAGE_SIZE, MI_HUGE_OS_PAGE_SIZE); + } + break; + } + + // success, record it + page++; // increase before timeout check (see issue #711) + mi_os_stat_increase(committed, MI_HUGE_OS_PAGE_SIZE); + mi_os_stat_increase(reserved, MI_HUGE_OS_PAGE_SIZE); + + // check for timeout + if (max_msecs > 0) { + mi_msecs_t elapsed = _mi_clock_end(start_t); + if (page >= 1) { + mi_msecs_t estimate = ((elapsed / (page+1)) * pages); + if (estimate > 2*max_msecs) { // seems like we are going to timeout, break + elapsed = max_msecs + 1; + } + } + if (elapsed > max_msecs) { + _mi_warning_message("huge OS page allocation timed out (after allocating %zu page(s))\n", page); + break; + } + } + } + mi_assert_internal(page*MI_HUGE_OS_PAGE_SIZE <= size); + if (pages_reserved != NULL) { *pages_reserved = page; } + if (psize != NULL) { *psize = page * MI_HUGE_OS_PAGE_SIZE; } + if (page != 0) { + mi_assert(start != NULL); + *memid = _mi_memid_create_os(true /* is committed */, all_zero, true /* is_large */); + memid->memkind = MI_MEM_OS_HUGE; + mi_assert(memid->is_pinned); + #ifdef MI_TRACK_ASAN + if (all_zero) { mi_track_mem_defined(start,size); } + #endif + } + return (page == 0 ? NULL : start); +} + +// free every huge page in a range individually (as we allocated per page) +// note: needed with VirtualAlloc but could potentially be done in one go on mmap'd systems. +static void mi_os_free_huge_os_pages(void* p, size_t size) { + if (p==NULL || size==0) return; + uint8_t* base = (uint8_t*)p; + while (size >= MI_HUGE_OS_PAGE_SIZE) { + mi_os_prim_free(base, MI_HUGE_OS_PAGE_SIZE, MI_HUGE_OS_PAGE_SIZE); + size -= MI_HUGE_OS_PAGE_SIZE; + base += MI_HUGE_OS_PAGE_SIZE; + } +} + +/* ---------------------------------------------------------------------------- +Support NUMA aware allocation +-----------------------------------------------------------------------------*/ + +_Atomic(size_t) _mi_numa_node_count; // = 0 // cache the node count + +size_t _mi_os_numa_node_count_get(void) { + size_t count = mi_atomic_load_acquire(&_mi_numa_node_count); + if (count <= 0) { + long ncount = mi_option_get(mi_option_use_numa_nodes); // given explicitly? + if (ncount > 0) { + count = (size_t)ncount; + } + else { + count = _mi_prim_numa_node_count(); // or detect dynamically + if (count == 0) count = 1; + } + mi_atomic_store_release(&_mi_numa_node_count, count); // save it + _mi_verbose_message("using %zd numa regions\n", count); + } + return count; +} + +int _mi_os_numa_node_get(void) { + size_t numa_count = _mi_os_numa_node_count(); + if (numa_count<=1) return 0; // optimize on single numa node systems: always node 0 + // never more than the node count and >= 0 + size_t numa_node = _mi_prim_numa_node(); + if (numa_node >= numa_count) { numa_node = numa_node % numa_count; } + return (int)numa_node; +} diff --git a/compat/mimalloc/page-queue.c b/compat/mimalloc/page-queue.c new file mode 100644 index 00000000000000..469e57d52be383 --- /dev/null +++ b/compat/mimalloc/page-queue.c @@ -0,0 +1,390 @@ +/*---------------------------------------------------------------------------- +Copyright (c) 2018-2024, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ + +/* ----------------------------------------------------------- + Definition of page queues for each block size +----------------------------------------------------------- */ + +#ifndef MI_IN_PAGE_C +#error "this file should be included from 'page.c'" +// include to help an IDE +#include "mimalloc.h" +#include "mimalloc/internal.h" +#include "mimalloc/atomic.h" +#endif + +/* ----------------------------------------------------------- + Minimal alignment in machine words (i.e. `sizeof(void*)`) +----------------------------------------------------------- */ + +#if (MI_MAX_ALIGN_SIZE > 4*MI_INTPTR_SIZE) + #error "define alignment for more than 4x word size for this platform" +#elif (MI_MAX_ALIGN_SIZE > 2*MI_INTPTR_SIZE) + #define MI_ALIGN4W // 4 machine words minimal alignment +#elif (MI_MAX_ALIGN_SIZE > MI_INTPTR_SIZE) + #define MI_ALIGN2W // 2 machine words minimal alignment +#else + // ok, default alignment is 1 word +#endif + + +/* ----------------------------------------------------------- + Queue query +----------------------------------------------------------- */ + + +static inline bool mi_page_queue_is_huge(const mi_page_queue_t* pq) { + return (pq->block_size == (MI_MEDIUM_OBJ_SIZE_MAX+sizeof(uintptr_t))); +} + +static inline bool mi_page_queue_is_full(const mi_page_queue_t* pq) { + return (pq->block_size == (MI_MEDIUM_OBJ_SIZE_MAX+(2*sizeof(uintptr_t)))); +} + +static inline bool mi_page_queue_is_special(const mi_page_queue_t* pq) { + return (pq->block_size > MI_MEDIUM_OBJ_SIZE_MAX); +} + +/* ----------------------------------------------------------- + Bins +----------------------------------------------------------- */ + +// Return the bin for a given field size. +// Returns MI_BIN_HUGE if the size is too large. +// We use `wsize` for the size in "machine word sizes", +// i.e. byte size == `wsize*sizeof(void*)`. +static inline size_t mi_bin(size_t size) { + size_t wsize = _mi_wsize_from_size(size); +#if defined(MI_ALIGN4W) + if mi_likely(wsize <= 4) { + return (wsize <= 1 ? 1 : (wsize+1)&~1); // round to double word sizes + } +#elif defined(MI_ALIGN2W) + if mi_likely(wsize <= 8) { + return (wsize <= 1 ? 1 : (wsize+1)&~1); // round to double word sizes + } +#else + if mi_likely(wsize <= 8) { + return (wsize == 0 ? 1 : wsize); + } +#endif + else if mi_unlikely(wsize > MI_MEDIUM_OBJ_WSIZE_MAX) { + return MI_BIN_HUGE; + } + else { + #if defined(MI_ALIGN4W) + if (wsize <= 16) { wsize = (wsize+3)&~3; } // round to 4x word sizes + #endif + wsize--; + // find the highest bit + const size_t b = (MI_SIZE_BITS - 1 - mi_clz(wsize)); // note: wsize != 0 + // and use the top 3 bits to determine the bin (~12.5% worst internal fragmentation). + // - adjust with 3 because we use do not round the first 8 sizes + // which each get an exact bin + const size_t bin = ((b << 2) + ((wsize >> (b - 2)) & 0x03)) - 3; + mi_assert_internal(bin > 0 && bin < MI_BIN_HUGE); + return bin; + } +} + + + +/* ----------------------------------------------------------- + Queue of pages with free blocks +----------------------------------------------------------- */ + +size_t _mi_bin(size_t size) { + return mi_bin(size); +} + +size_t _mi_bin_size(size_t bin) { + return _mi_heap_empty.pages[bin].block_size; +} + +// Good size for allocation +size_t mi_good_size(size_t size) mi_attr_noexcept { + if (size <= MI_MEDIUM_OBJ_SIZE_MAX) { + return _mi_bin_size(mi_bin(size + MI_PADDING_SIZE)); + } + else { + return _mi_align_up(size + MI_PADDING_SIZE,_mi_os_page_size()); + } +} + +#if (MI_DEBUG>1) +static bool mi_page_queue_contains(mi_page_queue_t* queue, const mi_page_t* page) { + mi_assert_internal(page != NULL); + mi_page_t* list = queue->first; + while (list != NULL) { + mi_assert_internal(list->next == NULL || list->next->prev == list); + mi_assert_internal(list->prev == NULL || list->prev->next == list); + if (list == page) break; + list = list->next; + } + return (list == page); +} + +#endif + +#if (MI_DEBUG>1) +static bool mi_heap_contains_queue(const mi_heap_t* heap, const mi_page_queue_t* pq) { + return (pq >= &heap->pages[0] && pq <= &heap->pages[MI_BIN_FULL]); +} +#endif + +static inline bool mi_page_is_large_or_huge(const mi_page_t* page) { + return (mi_page_block_size(page) > MI_MEDIUM_OBJ_SIZE_MAX || mi_page_is_huge(page)); +} + +static size_t mi_page_bin(const mi_page_t* page) { + const size_t bin = (mi_page_is_in_full(page) ? MI_BIN_FULL : (mi_page_is_huge(page) ? MI_BIN_HUGE : mi_bin(mi_page_block_size(page)))); + mi_assert_internal(bin <= MI_BIN_FULL); + return bin; +} + +static mi_page_queue_t* mi_heap_page_queue_of(mi_heap_t* heap, const mi_page_t* page) { + mi_assert_internal(heap!=NULL); + const size_t bin = mi_page_bin(page); + mi_page_queue_t* pq = &heap->pages[bin]; + mi_assert_internal((mi_page_block_size(page) == pq->block_size) || + (mi_page_is_large_or_huge(page) && mi_page_queue_is_huge(pq)) || + (mi_page_is_in_full(page) && mi_page_queue_is_full(pq))); + return pq; +} + +static mi_page_queue_t* mi_page_queue_of(const mi_page_t* page) { + mi_heap_t* heap = mi_page_heap(page); + mi_page_queue_t* pq = mi_heap_page_queue_of(heap, page); + mi_assert_expensive(mi_page_queue_contains(pq, page)); + return pq; +} + +// The current small page array is for efficiency and for each +// small size (up to 256) it points directly to the page for that +// size without having to compute the bin. This means when the +// current free page queue is updated for a small bin, we need to update a +// range of entries in `_mi_page_small_free`. +static inline void mi_heap_queue_first_update(mi_heap_t* heap, const mi_page_queue_t* pq) { + mi_assert_internal(mi_heap_contains_queue(heap,pq)); + size_t size = pq->block_size; + if (size > MI_SMALL_SIZE_MAX) return; + + mi_page_t* page = pq->first; + if (pq->first == NULL) page = (mi_page_t*)&_mi_page_empty; + + // find index in the right direct page array + size_t start; + size_t idx = _mi_wsize_from_size(size); + mi_page_t** pages_free = heap->pages_free_direct; + + if (pages_free[idx] == page) return; // already set + + // find start slot + if (idx<=1) { + start = 0; + } + else { + // find previous size; due to minimal alignment upto 3 previous bins may need to be skipped + size_t bin = mi_bin(size); + const mi_page_queue_t* prev = pq - 1; + while( bin == mi_bin(prev->block_size) && prev > &heap->pages[0]) { + prev--; + } + start = 1 + _mi_wsize_from_size(prev->block_size); + if (start > idx) start = idx; + } + + // set size range to the right page + mi_assert(start <= idx); + for (size_t sz = start; sz <= idx; sz++) { + pages_free[sz] = page; + } +} + +/* +static bool mi_page_queue_is_empty(mi_page_queue_t* queue) { + return (queue->first == NULL); +} +*/ + +static void mi_page_queue_remove(mi_page_queue_t* queue, mi_page_t* page) { + mi_assert_internal(page != NULL); + mi_assert_expensive(mi_page_queue_contains(queue, page)); + mi_assert_internal(mi_page_block_size(page) == queue->block_size || + (mi_page_is_large_or_huge(page) && mi_page_queue_is_huge(queue)) || + (mi_page_is_in_full(page) && mi_page_queue_is_full(queue))); + mi_heap_t* heap = mi_page_heap(page); + + if (page->prev != NULL) page->prev->next = page->next; + if (page->next != NULL) page->next->prev = page->prev; + if (page == queue->last) queue->last = page->prev; + if (page == queue->first) { + queue->first = page->next; + // update first + mi_assert_internal(mi_heap_contains_queue(heap, queue)); + mi_heap_queue_first_update(heap,queue); + } + heap->page_count--; + page->next = NULL; + page->prev = NULL; + // mi_atomic_store_ptr_release(mi_atomic_cast(void*, &page->heap), NULL); + mi_page_set_in_full(page,false); +} + + +static void mi_page_queue_push(mi_heap_t* heap, mi_page_queue_t* queue, mi_page_t* page) { + mi_assert_internal(mi_page_heap(page) == heap); + mi_assert_internal(!mi_page_queue_contains(queue, page)); + #if MI_HUGE_PAGE_ABANDON + mi_assert_internal(_mi_page_segment(page)->kind != MI_SEGMENT_HUGE); + #endif + mi_assert_internal(mi_page_block_size(page) == queue->block_size || + (mi_page_is_large_or_huge(page) && mi_page_queue_is_huge(queue)) || + (mi_page_is_in_full(page) && mi_page_queue_is_full(queue))); + + mi_page_set_in_full(page, mi_page_queue_is_full(queue)); + // mi_atomic_store_ptr_release(mi_atomic_cast(void*, &page->heap), heap); + page->next = queue->first; + page->prev = NULL; + if (queue->first != NULL) { + mi_assert_internal(queue->first->prev == NULL); + queue->first->prev = page; + queue->first = page; + } + else { + queue->first = queue->last = page; + } + + // update direct + mi_heap_queue_first_update(heap, queue); + heap->page_count++; +} + +static void mi_page_queue_move_to_front(mi_heap_t* heap, mi_page_queue_t* queue, mi_page_t* page) { + mi_assert_internal(mi_page_heap(page) == heap); + mi_assert_internal(mi_page_queue_contains(queue, page)); + if (queue->first == page) return; + mi_page_queue_remove(queue, page); + mi_page_queue_push(heap, queue, page); + mi_assert_internal(queue->first == page); +} + +static void mi_page_queue_enqueue_from_ex(mi_page_queue_t* to, mi_page_queue_t* from, bool enqueue_at_end, mi_page_t* page) { + mi_assert_internal(page != NULL); + mi_assert_expensive(mi_page_queue_contains(from, page)); + mi_assert_expensive(!mi_page_queue_contains(to, page)); + const size_t bsize = mi_page_block_size(page); + MI_UNUSED(bsize); + mi_assert_internal((bsize == to->block_size && bsize == from->block_size) || + (bsize == to->block_size && mi_page_queue_is_full(from)) || + (bsize == from->block_size && mi_page_queue_is_full(to)) || + (mi_page_is_large_or_huge(page) && mi_page_queue_is_huge(to)) || + (mi_page_is_large_or_huge(page) && mi_page_queue_is_full(to))); + + mi_heap_t* heap = mi_page_heap(page); + + // delete from `from` + if (page->prev != NULL) page->prev->next = page->next; + if (page->next != NULL) page->next->prev = page->prev; + if (page == from->last) from->last = page->prev; + if (page == from->first) { + from->first = page->next; + // update first + mi_assert_internal(mi_heap_contains_queue(heap, from)); + mi_heap_queue_first_update(heap, from); + } + + // insert into `to` + if (enqueue_at_end) { + // enqueue at the end + page->prev = to->last; + page->next = NULL; + if (to->last != NULL) { + mi_assert_internal(heap == mi_page_heap(to->last)); + to->last->next = page; + to->last = page; + } + else { + to->first = page; + to->last = page; + mi_heap_queue_first_update(heap, to); + } + } + else { + if (to->first != NULL) { + // enqueue at 2nd place + mi_assert_internal(heap == mi_page_heap(to->first)); + mi_page_t* next = to->first->next; + page->prev = to->first; + page->next = next; + to->first->next = page; + if (next != NULL) { + next->prev = page; + } + else { + to->last = page; + } + } + else { + // enqueue at the head (singleton list) + page->prev = NULL; + page->next = NULL; + to->first = page; + to->last = page; + mi_heap_queue_first_update(heap, to); + } + } + + mi_page_set_in_full(page, mi_page_queue_is_full(to)); +} + +static void mi_page_queue_enqueue_from(mi_page_queue_t* to, mi_page_queue_t* from, mi_page_t* page) { + mi_page_queue_enqueue_from_ex(to, from, true /* enqueue at the end */, page); +} + +static void mi_page_queue_enqueue_from_full(mi_page_queue_t* to, mi_page_queue_t* from, mi_page_t* page) { + // note: we could insert at the front to increase reuse, but it slows down certain benchmarks (like `alloc-test`) + mi_page_queue_enqueue_from_ex(to, from, true /* enqueue at the end of the `to` queue? */, page); +} + +// Only called from `mi_heap_absorb`. +size_t _mi_page_queue_append(mi_heap_t* heap, mi_page_queue_t* pq, mi_page_queue_t* append) { + mi_assert_internal(mi_heap_contains_queue(heap,pq)); + mi_assert_internal(pq->block_size == append->block_size); + + if (append->first==NULL) return 0; + + // set append pages to new heap and count + size_t count = 0; + for (mi_page_t* page = append->first; page != NULL; page = page->next) { + // inline `mi_page_set_heap` to avoid wrong assertion during absorption; + // in this case it is ok to be delayed freeing since both "to" and "from" heap are still alive. + mi_atomic_store_release(&page->xheap, (uintptr_t)heap); + // set the flag to delayed free (not overriding NEVER_DELAYED_FREE) which has as a + // side effect that it spins until any DELAYED_FREEING is finished. This ensures + // that after appending only the new heap will be used for delayed free operations. + _mi_page_use_delayed_free(page, MI_USE_DELAYED_FREE, false); + count++; + } + + if (pq->last==NULL) { + // take over afresh + mi_assert_internal(pq->first==NULL); + pq->first = append->first; + pq->last = append->last; + mi_heap_queue_first_update(heap, pq); + } + else { + // append to end + mi_assert_internal(pq->last!=NULL); + mi_assert_internal(append->first!=NULL); + pq->last->next = append->first; + append->first->prev = pq->last; + pq->last = append->last; + } + return count; +} diff --git a/compat/mimalloc/page.c b/compat/mimalloc/page.c new file mode 100644 index 00000000000000..8db2463f4ab745 --- /dev/null +++ b/compat/mimalloc/page.c @@ -0,0 +1,1043 @@ +/*---------------------------------------------------------------------------- +Copyright (c) 2018-2024, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ + +/* ----------------------------------------------------------- + The core of the allocator. Every segment contains + pages of a certain block size. The main function + exported is `mi_malloc_generic`. +----------------------------------------------------------- */ + +#include "mimalloc.h" +#include "mimalloc/internal.h" +#include "mimalloc/atomic.h" + +/* ----------------------------------------------------------- + Definition of page queues for each block size +----------------------------------------------------------- */ + +#define MI_IN_PAGE_C +#include "page-queue.c" +#undef MI_IN_PAGE_C + + +/* ----------------------------------------------------------- + Page helpers +----------------------------------------------------------- */ + +// Index a block in a page +static inline mi_block_t* mi_page_block_at(const mi_page_t* page, void* page_start, size_t block_size, size_t i) { + MI_UNUSED(page); + mi_assert_internal(page != NULL); + mi_assert_internal(i <= page->reserved); + return (mi_block_t*)((uint8_t*)page_start + (i * block_size)); +} + +static void mi_page_init(mi_heap_t* heap, mi_page_t* page, size_t size, mi_tld_t* tld); +static void mi_page_extend_free(mi_heap_t* heap, mi_page_t* page, mi_tld_t* tld); + +#if (MI_DEBUG>=3) +static size_t mi_page_list_count(mi_page_t* page, mi_block_t* head) { + size_t count = 0; + while (head != NULL) { + mi_assert_internal(page == _mi_ptr_page(head)); + count++; + head = mi_block_next(page, head); + } + return count; +} + +/* +// Start of the page available memory +static inline uint8_t* mi_page_area(const mi_page_t* page) { + return _mi_page_start(_mi_page_segment(page), page, NULL); +} +*/ + +static bool mi_page_list_is_valid(mi_page_t* page, mi_block_t* p) { + size_t psize; + uint8_t* page_area = _mi_segment_page_start(_mi_page_segment(page), page, &psize); + mi_block_t* start = (mi_block_t*)page_area; + mi_block_t* end = (mi_block_t*)(page_area + psize); + while(p != NULL) { + if (p < start || p >= end) return false; + p = mi_block_next(page, p); + } +#if MI_DEBUG>3 // generally too expensive to check this + if (page->free_is_zero) { + const size_t ubsize = mi_page_usable_block_size(page); + for (mi_block_t* block = page->free; block != NULL; block = mi_block_next(page, block)) { + mi_assert_expensive(mi_mem_is_zero(block + 1, ubsize - sizeof(mi_block_t))); + } + } +#endif + return true; +} + +static bool mi_page_is_valid_init(mi_page_t* page) { + mi_assert_internal(mi_page_block_size(page) > 0); + mi_assert_internal(page->used <= page->capacity); + mi_assert_internal(page->capacity <= page->reserved); + + uint8_t* start = mi_page_start(page); + mi_assert_internal(start == _mi_segment_page_start(_mi_page_segment(page), page, NULL)); + mi_assert_internal(page->is_huge == (_mi_page_segment(page)->kind == MI_SEGMENT_HUGE)); + //mi_assert_internal(start + page->capacity*page->block_size == page->top); + + mi_assert_internal(mi_page_list_is_valid(page,page->free)); + mi_assert_internal(mi_page_list_is_valid(page,page->local_free)); + + #if MI_DEBUG>3 // generally too expensive to check this + if (page->free_is_zero) { + const size_t ubsize = mi_page_usable_block_size(page); + for(mi_block_t* block = page->free; block != NULL; block = mi_block_next(page,block)) { + mi_assert_expensive(mi_mem_is_zero(block + 1, ubsize - sizeof(mi_block_t))); + } + } + #endif + + #if !MI_TRACK_ENABLED && !MI_TSAN + mi_block_t* tfree = mi_page_thread_free(page); + mi_assert_internal(mi_page_list_is_valid(page, tfree)); + //size_t tfree_count = mi_page_list_count(page, tfree); + //mi_assert_internal(tfree_count <= page->thread_freed + 1); + #endif + + size_t free_count = mi_page_list_count(page, page->free) + mi_page_list_count(page, page->local_free); + mi_assert_internal(page->used + free_count == page->capacity); + + return true; +} + +extern bool _mi_process_is_initialized; // has mi_process_init been called? + +bool _mi_page_is_valid(mi_page_t* page) { + mi_assert_internal(mi_page_is_valid_init(page)); + #if MI_SECURE + mi_assert_internal(page->keys[0] != 0); + #endif + if (mi_page_heap(page)!=NULL) { + mi_segment_t* segment = _mi_page_segment(page); + + mi_assert_internal(!_mi_process_is_initialized || segment->thread_id==0 || segment->thread_id == mi_page_heap(page)->thread_id); + #if MI_HUGE_PAGE_ABANDON + if (segment->kind != MI_SEGMENT_HUGE) + #endif + { + mi_page_queue_t* pq = mi_page_queue_of(page); + mi_assert_internal(mi_page_queue_contains(pq, page)); + mi_assert_internal(pq->block_size==mi_page_block_size(page) || mi_page_block_size(page) > MI_MEDIUM_OBJ_SIZE_MAX || mi_page_is_in_full(page)); + mi_assert_internal(mi_heap_contains_queue(mi_page_heap(page),pq)); + } + } + return true; +} +#endif + +void _mi_page_use_delayed_free(mi_page_t* page, mi_delayed_t delay, bool override_never) { + while (!_mi_page_try_use_delayed_free(page, delay, override_never)) { + mi_atomic_yield(); + } +} + +bool _mi_page_try_use_delayed_free(mi_page_t* page, mi_delayed_t delay, bool override_never) { + mi_thread_free_t tfreex; + mi_delayed_t old_delay; + mi_thread_free_t tfree; + size_t yield_count = 0; + do { + tfree = mi_atomic_load_acquire(&page->xthread_free); // note: must acquire as we can break/repeat this loop and not do a CAS; + tfreex = mi_tf_set_delayed(tfree, delay); + old_delay = mi_tf_delayed(tfree); + if mi_unlikely(old_delay == MI_DELAYED_FREEING) { + if (yield_count >= 4) return false; // give up after 4 tries + yield_count++; + mi_atomic_yield(); // delay until outstanding MI_DELAYED_FREEING are done. + // tfree = mi_tf_set_delayed(tfree, MI_NO_DELAYED_FREE); // will cause CAS to busy fail + } + else if (delay == old_delay) { + break; // avoid atomic operation if already equal + } + else if (!override_never && old_delay == MI_NEVER_DELAYED_FREE) { + break; // leave never-delayed flag set + } + } while ((old_delay == MI_DELAYED_FREEING) || + !mi_atomic_cas_weak_release(&page->xthread_free, &tfree, tfreex)); + + return true; // success +} + +/* ----------------------------------------------------------- + Page collect the `local_free` and `thread_free` lists +----------------------------------------------------------- */ + +// Collect the local `thread_free` list using an atomic exchange. +// Note: The exchange must be done atomically as this is used right after +// moving to the full list in `mi_page_collect_ex` and we need to +// ensure that there was no race where the page became unfull just before the move. +static void _mi_page_thread_free_collect(mi_page_t* page) +{ + mi_block_t* head; + mi_thread_free_t tfreex; + mi_thread_free_t tfree = mi_atomic_load_relaxed(&page->xthread_free); + do { + head = mi_tf_block(tfree); + tfreex = mi_tf_set_block(tfree,NULL); + } while (!mi_atomic_cas_weak_acq_rel(&page->xthread_free, &tfree, tfreex)); + + // return if the list is empty + if (head == NULL) return; + + // find the tail -- also to get a proper count (without data races) + size_t max_count = page->capacity; // cannot collect more than capacity + size_t count = 1; + mi_block_t* tail = head; + mi_block_t* next; + while ((next = mi_block_next(page,tail)) != NULL && count <= max_count) { + count++; + tail = next; + } + // if `count > max_count` there was a memory corruption (possibly infinite list due to double multi-threaded free) + if (count > max_count) { + _mi_error_message(EFAULT, "corrupted thread-free list\n"); + return; // the thread-free items cannot be freed + } + + // and append the current local free list + mi_block_set_next(page,tail, page->local_free); + page->local_free = head; + + // update counts now + page->used -= (uint16_t)count; +} + +void _mi_page_free_collect(mi_page_t* page, bool force) { + mi_assert_internal(page!=NULL); + + // collect the thread free list + if (force || mi_page_thread_free(page) != NULL) { // quick test to avoid an atomic operation + _mi_page_thread_free_collect(page); + } + + // and the local free list + if (page->local_free != NULL) { + if mi_likely(page->free == NULL) { + // usual case + page->free = page->local_free; + page->local_free = NULL; + page->free_is_zero = false; + } + else if (force) { + // append -- only on shutdown (force) as this is a linear operation + mi_block_t* tail = page->local_free; + mi_block_t* next; + while ((next = mi_block_next(page, tail)) != NULL) { + tail = next; + } + mi_block_set_next(page, tail, page->free); + page->free = page->local_free; + page->local_free = NULL; + page->free_is_zero = false; + } + } + + mi_assert_internal(!force || page->local_free == NULL); +} + + + +/* ----------------------------------------------------------- + Page fresh and retire +----------------------------------------------------------- */ + +// called from segments when reclaiming abandoned pages +void _mi_page_reclaim(mi_heap_t* heap, mi_page_t* page) { + mi_assert_expensive(mi_page_is_valid_init(page)); + + mi_assert_internal(mi_page_heap(page) == heap); + mi_assert_internal(mi_page_thread_free_flag(page) != MI_NEVER_DELAYED_FREE); + #if MI_HUGE_PAGE_ABANDON + mi_assert_internal(_mi_page_segment(page)->kind != MI_SEGMENT_HUGE); + #endif + + // TODO: push on full queue immediately if it is full? + mi_page_queue_t* pq = mi_page_queue(heap, mi_page_block_size(page)); + mi_page_queue_push(heap, pq, page); + mi_assert_expensive(_mi_page_is_valid(page)); +} + +// allocate a fresh page from a segment +static mi_page_t* mi_page_fresh_alloc(mi_heap_t* heap, mi_page_queue_t* pq, size_t block_size, size_t page_alignment) { + #if !MI_HUGE_PAGE_ABANDON + mi_assert_internal(pq != NULL); + mi_assert_internal(mi_heap_contains_queue(heap, pq)); + mi_assert_internal(page_alignment > 0 || block_size > MI_MEDIUM_OBJ_SIZE_MAX || block_size == pq->block_size); + #endif + mi_page_t* page = _mi_segment_page_alloc(heap, block_size, page_alignment, &heap->tld->segments); + if (page == NULL) { + // this may be out-of-memory, or an abandoned page was reclaimed (and in our queue) + return NULL; + } + #if MI_HUGE_PAGE_ABANDON + mi_assert_internal(pq==NULL || _mi_page_segment(page)->page_kind != MI_PAGE_HUGE); + #endif + mi_assert_internal(page_alignment >0 || block_size > MI_MEDIUM_OBJ_SIZE_MAX || _mi_page_segment(page)->kind != MI_SEGMENT_HUGE); + mi_assert_internal(pq!=NULL || mi_page_block_size(page) >= block_size); + // a fresh page was found, initialize it + const size_t full_block_size = (pq == NULL || mi_page_is_huge(page) ? mi_page_block_size(page) : block_size); // see also: mi_segment_huge_page_alloc + mi_assert_internal(full_block_size >= block_size); + mi_page_init(heap, page, full_block_size, heap->tld); + mi_heap_stat_increase(heap, pages, 1); + mi_heap_stat_increase(heap, page_bins[mi_page_bin(page)], 1); + if (pq != NULL) { mi_page_queue_push(heap, pq, page); } + mi_assert_expensive(_mi_page_is_valid(page)); + return page; +} + +// Get a fresh page to use +static mi_page_t* mi_page_fresh(mi_heap_t* heap, mi_page_queue_t* pq) { + mi_assert_internal(mi_heap_contains_queue(heap, pq)); + mi_page_t* page = mi_page_fresh_alloc(heap, pq, pq->block_size, 0); + if (page==NULL) return NULL; + mi_assert_internal(pq->block_size==mi_page_block_size(page)); + mi_assert_internal(pq==mi_page_queue(heap, mi_page_block_size(page))); + return page; +} + +/* ----------------------------------------------------------- + Do any delayed frees + (put there by other threads if they deallocated in a full page) +----------------------------------------------------------- */ +void _mi_heap_delayed_free_all(mi_heap_t* heap) { + while (!_mi_heap_delayed_free_partial(heap)) { + mi_atomic_yield(); + } +} + +// returns true if all delayed frees were processed +bool _mi_heap_delayed_free_partial(mi_heap_t* heap) { + // take over the list (note: no atomic exchange since it is often NULL) + mi_block_t* block = mi_atomic_load_ptr_relaxed(mi_block_t, &heap->thread_delayed_free); + while (block != NULL && !mi_atomic_cas_ptr_weak_acq_rel(mi_block_t, &heap->thread_delayed_free, &block, NULL)) { /* nothing */ }; + bool all_freed = true; + + // and free them all + while(block != NULL) { + mi_block_t* next = mi_block_nextx(heap,block, heap->keys); + // use internal free instead of regular one to keep stats etc correct + if (!_mi_free_delayed_block(block)) { + // we might already start delayed freeing while another thread has not yet + // reset the delayed_freeing flag; in that case delay it further by reinserting the current block + // into the delayed free list + all_freed = false; + mi_block_t* dfree = mi_atomic_load_ptr_relaxed(mi_block_t, &heap->thread_delayed_free); + do { + mi_block_set_nextx(heap, block, dfree, heap->keys); + } while (!mi_atomic_cas_ptr_weak_release(mi_block_t,&heap->thread_delayed_free, &dfree, block)); + } + block = next; + } + return all_freed; +} + +/* ----------------------------------------------------------- + Unfull, abandon, free and retire +----------------------------------------------------------- */ + +// Move a page from the full list back to a regular list +void _mi_page_unfull(mi_page_t* page) { + mi_assert_internal(page != NULL); + mi_assert_expensive(_mi_page_is_valid(page)); + mi_assert_internal(mi_page_is_in_full(page)); + if (!mi_page_is_in_full(page)) return; + + mi_heap_t* heap = mi_page_heap(page); + mi_page_queue_t* pqfull = &heap->pages[MI_BIN_FULL]; + mi_page_set_in_full(page, false); // to get the right queue + mi_page_queue_t* pq = mi_heap_page_queue_of(heap, page); + mi_page_set_in_full(page, true); + mi_page_queue_enqueue_from_full(pq, pqfull, page); +} + +static void mi_page_to_full(mi_page_t* page, mi_page_queue_t* pq) { + mi_assert_internal(pq == mi_page_queue_of(page)); + mi_assert_internal(!mi_page_immediate_available(page)); + mi_assert_internal(!mi_page_is_in_full(page)); + + if (mi_page_is_in_full(page)) return; + mi_page_queue_enqueue_from(&mi_page_heap(page)->pages[MI_BIN_FULL], pq, page); + _mi_page_free_collect(page,false); // try to collect right away in case another thread freed just before MI_USE_DELAYED_FREE was set +} + + +// Abandon a page with used blocks at the end of a thread. +// Note: only call if it is ensured that no references exist from +// the `page->heap->thread_delayed_free` into this page. +// Currently only called through `mi_heap_collect_ex` which ensures this. +void _mi_page_abandon(mi_page_t* page, mi_page_queue_t* pq) { + mi_assert_internal(page != NULL); + mi_assert_expensive(_mi_page_is_valid(page)); + mi_assert_internal(pq == mi_page_queue_of(page)); + mi_assert_internal(mi_page_heap(page) != NULL); + + mi_heap_t* pheap = mi_page_heap(page); + + // remove from our page list + mi_segments_tld_t* segments_tld = &pheap->tld->segments; + mi_page_queue_remove(pq, page); + + // page is no longer associated with our heap + mi_assert_internal(mi_page_thread_free_flag(page)==MI_NEVER_DELAYED_FREE); + mi_page_set_heap(page, NULL); + +#if (MI_DEBUG>1) && !MI_TRACK_ENABLED + // check there are no references left.. + for (mi_block_t* block = (mi_block_t*)pheap->thread_delayed_free; block != NULL; block = mi_block_nextx(pheap, block, pheap->keys)) { + mi_assert_internal(_mi_ptr_page(block) != page); + } +#endif + + // and abandon it + mi_assert_internal(mi_page_heap(page) == NULL); + _mi_segment_page_abandon(page,segments_tld); +} + +// force abandon a page +void _mi_page_force_abandon(mi_page_t* page) { + mi_heap_t* heap = mi_page_heap(page); + // mark page as not using delayed free + _mi_page_use_delayed_free(page, MI_NEVER_DELAYED_FREE, false); + + // ensure this page is no longer in the heap delayed free list + _mi_heap_delayed_free_all(heap); + // We can still access the page meta-info even if it is freed as we ensure + // in `mi_segment_force_abandon` that the segment is not freed (yet) + if (page->capacity == 0) return; // it may have been freed now + + // and now unlink it from the page queue and abandon (or free) + mi_page_queue_t* pq = mi_heap_page_queue_of(heap, page); + if (mi_page_all_free(page)) { + _mi_page_free(page, pq, false); + } + else { + _mi_page_abandon(page, pq); + } +} + + +// Free a page with no more free blocks +void _mi_page_free(mi_page_t* page, mi_page_queue_t* pq, bool force) { + mi_assert_internal(page != NULL); + mi_assert_expensive(_mi_page_is_valid(page)); + mi_assert_internal(pq == mi_page_queue_of(page)); + mi_assert_internal(mi_page_all_free(page)); + mi_assert_internal(mi_page_thread_free_flag(page)!=MI_DELAYED_FREEING); + + // no more aligned blocks in here + mi_page_set_has_aligned(page, false); + + // remove from the page list + // (no need to do _mi_heap_delayed_free first as all blocks are already free) + mi_heap_t* heap = mi_page_heap(page); + mi_segments_tld_t* segments_tld = &heap->tld->segments; + mi_page_queue_remove(pq, page); + + // and free it + mi_heap_stat_decrease(heap, page_bins[mi_page_bin(page)], 1); + mi_page_set_heap(page,NULL); + _mi_segment_page_free(page, force, segments_tld); +} + +#define MI_MAX_RETIRE_SIZE MI_MEDIUM_OBJ_SIZE_MAX // should be less than size for MI_BIN_HUGE +#define MI_RETIRE_CYCLES (16) + +// Retire a page with no more used blocks +// Important to not retire too quickly though as new +// allocations might coming. +// Note: called from `mi_free` and benchmarks often +// trigger this due to freeing everything and then +// allocating again so careful when changing this. +void _mi_page_retire(mi_page_t* page) mi_attr_noexcept { + mi_assert_internal(page != NULL); + mi_assert_expensive(_mi_page_is_valid(page)); + mi_assert_internal(mi_page_all_free(page)); + + mi_page_set_has_aligned(page, false); + + // don't retire too often.. + // (or we end up retiring and re-allocating most of the time) + // NOTE: refine this more: we should not retire if this + // is the only page left with free blocks. It is not clear + // how to check this efficiently though... + // for now, we don't retire if it is the only page left of this size class. + mi_page_queue_t* pq = mi_page_queue_of(page); + #if MI_RETIRE_CYCLES > 0 + const size_t bsize = mi_page_block_size(page); + if mi_likely( /* bsize < MI_MAX_RETIRE_SIZE && */ !mi_page_queue_is_special(pq)) { // not full or huge queue? + if (pq->last==page && pq->first==page) { // the only page in the queue? + mi_stat_counter_increase(_mi_stats_main.pages_retire,1); + page->retire_expire = (bsize <= MI_SMALL_OBJ_SIZE_MAX ? MI_RETIRE_CYCLES : MI_RETIRE_CYCLES/4); + mi_heap_t* heap = mi_page_heap(page); + mi_assert_internal(pq >= heap->pages); + const size_t index = pq - heap->pages; + mi_assert_internal(index < MI_BIN_FULL && index < MI_BIN_HUGE); + if (index < heap->page_retired_min) heap->page_retired_min = index; + if (index > heap->page_retired_max) heap->page_retired_max = index; + mi_assert_internal(mi_page_all_free(page)); + return; // don't free after all + } + } + #endif + _mi_page_free(page, pq, false); +} + +// free retired pages: we don't need to look at the entire queues +// since we only retire pages that are at the head position in a queue. +void _mi_heap_collect_retired(mi_heap_t* heap, bool force) { + size_t min = MI_BIN_FULL; + size_t max = 0; + for(size_t bin = heap->page_retired_min; bin <= heap->page_retired_max; bin++) { + mi_page_queue_t* pq = &heap->pages[bin]; + mi_page_t* page = pq->first; + if (page != NULL && page->retire_expire != 0) { + if (mi_page_all_free(page)) { + page->retire_expire--; + if (force || page->retire_expire == 0) { + _mi_page_free(pq->first, pq, force); + } + else { + // keep retired, update min/max + if (bin < min) min = bin; + if (bin > max) max = bin; + } + } + else { + page->retire_expire = 0; + } + } + } + heap->page_retired_min = min; + heap->page_retired_max = max; +} + + +/* ----------------------------------------------------------- + Initialize the initial free list in a page. + In secure mode we initialize a randomized list by + alternating between slices. +----------------------------------------------------------- */ + +#define MI_MAX_SLICE_SHIFT (6) // at most 64 slices +#define MI_MAX_SLICES (1UL << MI_MAX_SLICE_SHIFT) +#define MI_MIN_SLICES (2) + +static void mi_page_free_list_extend_secure(mi_heap_t* const heap, mi_page_t* const page, const size_t bsize, const size_t extend, mi_stats_t* const stats) { + MI_UNUSED(stats); + #if (MI_SECURE<=2) + mi_assert_internal(page->free == NULL); + mi_assert_internal(page->local_free == NULL); + #endif + mi_assert_internal(page->capacity + extend <= page->reserved); + mi_assert_internal(bsize == mi_page_block_size(page)); + void* const page_area = mi_page_start(page); + + // initialize a randomized free list + // set up `slice_count` slices to alternate between + size_t shift = MI_MAX_SLICE_SHIFT; + while ((extend >> shift) == 0) { + shift--; + } + const size_t slice_count = (size_t)1U << shift; + const size_t slice_extend = extend / slice_count; + mi_assert_internal(slice_extend >= 1); + mi_block_t* blocks[MI_MAX_SLICES]; // current start of the slice + size_t counts[MI_MAX_SLICES]; // available objects in the slice + for (size_t i = 0; i < slice_count; i++) { + blocks[i] = mi_page_block_at(page, page_area, bsize, page->capacity + i*slice_extend); + counts[i] = slice_extend; + } + counts[slice_count-1] += (extend % slice_count); // final slice holds the modulus too (todo: distribute evenly?) + + // and initialize the free list by randomly threading through them + // set up first element + const uintptr_t r = _mi_heap_random_next(heap); + size_t current = r % slice_count; + counts[current]--; + mi_block_t* const free_start = blocks[current]; + // and iterate through the rest; use `random_shuffle` for performance + uintptr_t rnd = _mi_random_shuffle(r|1); // ensure not 0 + for (size_t i = 1; i < extend; i++) { + // call random_shuffle only every INTPTR_SIZE rounds + const size_t round = i%MI_INTPTR_SIZE; + if (round == 0) rnd = _mi_random_shuffle(rnd); + // select a random next slice index + size_t next = ((rnd >> 8*round) & (slice_count-1)); + while (counts[next]==0) { // ensure it still has space + next++; + if (next==slice_count) next = 0; + } + // and link the current block to it + counts[next]--; + mi_block_t* const block = blocks[current]; + blocks[current] = (mi_block_t*)((uint8_t*)block + bsize); // bump to the following block + mi_block_set_next(page, block, blocks[next]); // and set next; note: we may have `current == next` + current = next; + } + // prepend to the free list (usually NULL) + mi_block_set_next(page, blocks[current], page->free); // end of the list + page->free = free_start; +} + +static mi_decl_noinline void mi_page_free_list_extend( mi_page_t* const page, const size_t bsize, const size_t extend, mi_stats_t* const stats) +{ + MI_UNUSED(stats); + #if (MI_SECURE <= 2) + mi_assert_internal(page->free == NULL); + mi_assert_internal(page->local_free == NULL); + #endif + mi_assert_internal(page->capacity + extend <= page->reserved); + mi_assert_internal(bsize == mi_page_block_size(page)); + void* const page_area = mi_page_start(page); + + mi_block_t* const start = mi_page_block_at(page, page_area, bsize, page->capacity); + + // initialize a sequential free list + mi_block_t* const last = mi_page_block_at(page, page_area, bsize, page->capacity + extend - 1); + mi_block_t* block = start; + while(block <= last) { + mi_block_t* next = (mi_block_t*)((uint8_t*)block + bsize); + mi_block_set_next(page,block,next); + block = next; + } + // prepend to free list (usually `NULL`) + mi_block_set_next(page, last, page->free); + page->free = start; +} + +/* ----------------------------------------------------------- + Page initialize and extend the capacity +----------------------------------------------------------- */ + +#define MI_MAX_EXTEND_SIZE (4*1024) // heuristic, one OS page seems to work well. +#if (MI_SECURE>0) +#define MI_MIN_EXTEND (8*MI_SECURE) // extend at least by this many +#else +#define MI_MIN_EXTEND (4) +#endif + +// Extend the capacity (up to reserved) by initializing a free list +// We do at most `MI_MAX_EXTEND` to avoid touching too much memory +// Note: we also experimented with "bump" allocation on the first +// allocations but this did not speed up any benchmark (due to an +// extra test in malloc? or cache effects?) +static void mi_page_extend_free(mi_heap_t* heap, mi_page_t* page, mi_tld_t* tld) { + MI_UNUSED(tld); + mi_assert_expensive(mi_page_is_valid_init(page)); + #if (MI_SECURE<=2) + mi_assert(page->free == NULL); + mi_assert(page->local_free == NULL); + if (page->free != NULL) return; + #endif + if (page->capacity >= page->reserved) return; + + mi_stat_counter_increase(tld->stats.pages_extended, 1); + + // calculate the extend count + const size_t bsize = mi_page_block_size(page); + size_t extend = page->reserved - page->capacity; + mi_assert_internal(extend > 0); + + size_t max_extend = (bsize >= MI_MAX_EXTEND_SIZE ? MI_MIN_EXTEND : MI_MAX_EXTEND_SIZE/bsize); + if (max_extend < MI_MIN_EXTEND) { max_extend = MI_MIN_EXTEND; } + mi_assert_internal(max_extend > 0); + + if (extend > max_extend) { + // ensure we don't touch memory beyond the page to reduce page commit. + // the `lean` benchmark tests this. Going from 1 to 8 increases rss by 50%. + extend = max_extend; + } + + mi_assert_internal(extend > 0 && extend + page->capacity <= page->reserved); + mi_assert_internal(extend < (1UL<<16)); + + // and append the extend the free list + if (extend < MI_MIN_SLICES || MI_SECURE==0) { //!mi_option_is_enabled(mi_option_secure)) { + mi_page_free_list_extend(page, bsize, extend, &tld->stats ); + } + else { + mi_page_free_list_extend_secure(heap, page, bsize, extend, &tld->stats); + } + // enable the new free list + page->capacity += (uint16_t)extend; + mi_stat_increase(tld->stats.page_committed, extend * bsize); + mi_assert_expensive(mi_page_is_valid_init(page)); +} + +// Initialize a fresh page +static void mi_page_init(mi_heap_t* heap, mi_page_t* page, size_t block_size, mi_tld_t* tld) { + mi_assert(page != NULL); + mi_segment_t* segment = _mi_page_segment(page); + mi_assert(segment != NULL); + mi_assert_internal(block_size > 0); + // set fields + mi_page_set_heap(page, heap); + page->block_size = block_size; + size_t page_size; + page->page_start = _mi_segment_page_start(segment, page, &page_size); + mi_track_mem_noaccess(page->page_start,page_size); + mi_assert_internal(mi_page_block_size(page) <= page_size); + mi_assert_internal(page_size <= page->slice_count*MI_SEGMENT_SLICE_SIZE); + mi_assert_internal(page_size / block_size < (1L<<16)); + page->reserved = (uint16_t)(page_size / block_size); + mi_assert_internal(page->reserved > 0); + #if (MI_PADDING || MI_ENCODE_FREELIST) + page->keys[0] = _mi_heap_random_next(heap); + page->keys[1] = _mi_heap_random_next(heap); + #endif + page->free_is_zero = page->is_zero_init; + #if MI_DEBUG>2 + if (page->is_zero_init) { + mi_track_mem_defined(page->page_start, page_size); + mi_assert_expensive(mi_mem_is_zero(page->page_start, page_size)); + } + #endif + mi_assert_internal(page->is_committed); + if (block_size > 0 && _mi_is_power_of_two(block_size)) { + page->block_size_shift = (uint8_t)(mi_ctz((uintptr_t)block_size)); + } + else { + page->block_size_shift = 0; + } + + mi_assert_internal(page->capacity == 0); + mi_assert_internal(page->free == NULL); + mi_assert_internal(page->used == 0); + mi_assert_internal(page->xthread_free == 0); + mi_assert_internal(page->next == NULL); + mi_assert_internal(page->prev == NULL); + mi_assert_internal(page->retire_expire == 0); + mi_assert_internal(!mi_page_has_aligned(page)); + #if (MI_PADDING || MI_ENCODE_FREELIST) + mi_assert_internal(page->keys[0] != 0); + mi_assert_internal(page->keys[1] != 0); + #endif + mi_assert_internal(page->block_size_shift == 0 || (block_size == ((size_t)1 << page->block_size_shift))); + mi_assert_expensive(mi_page_is_valid_init(page)); + + // initialize an initial free list + mi_page_extend_free(heap,page,tld); + mi_assert(mi_page_immediate_available(page)); +} + + +/* ----------------------------------------------------------- + Find pages with free blocks +-------------------------------------------------------------*/ + +// search for a best next page to use for at most N pages (often cut short if immediate blocks are available) +#define MI_MAX_CANDIDATE_SEARCH (4) + +// is the page not yet used up to its reserved space? +static bool mi_page_is_expandable(const mi_page_t* page) { + mi_assert_internal(page != NULL); + mi_assert_internal(page->capacity <= page->reserved); + return (page->capacity < page->reserved); +} + + +// Find a page with free blocks of `page->block_size`. +static mi_page_t* mi_page_queue_find_free_ex(mi_heap_t* heap, mi_page_queue_t* pq, bool first_try) +{ + // search through the pages in "next fit" order + #if MI_STAT + size_t count = 0; + #endif + size_t candidate_count = 0; // we reset this on the first candidate to limit the search + mi_page_t* page_candidate = NULL; // a page with free space + mi_page_t* page = pq->first; + + while (page != NULL) + { + mi_page_t* next = page->next; // remember next + #if MI_STAT + count++; + #endif + candidate_count++; + + // collect freed blocks by us and other threads + _mi_page_free_collect(page, false); + + #if MI_MAX_CANDIDATE_SEARCH > 1 + // search up to N pages for a best candidate + + // is the local free list non-empty? + const bool immediate_available = mi_page_immediate_available(page); + + // if the page is completely full, move it to the `mi_pages_full` + // queue so we don't visit long-lived pages too often. + if (!immediate_available && !mi_page_is_expandable(page)) { + mi_assert_internal(!mi_page_is_in_full(page) && !mi_page_immediate_available(page)); + mi_page_to_full(page, pq); + } + else { + // the page has free space, make it a candidate + // we prefer non-expandable pages with high usage as candidates (to reduce commit, and increase chances of free-ing up pages) + if (page_candidate == NULL) { + page_candidate = page; + candidate_count = 0; + } + // prefer to reuse fuller pages (in the hope the less used page gets freed) + else if (page->used >= page_candidate->used && !mi_page_is_mostly_used(page) && !mi_page_is_expandable(page)) { + page_candidate = page; + } + // if we find a non-expandable candidate, or searched for N pages, return with the best candidate + if (immediate_available || candidate_count > MI_MAX_CANDIDATE_SEARCH) { + mi_assert_internal(page_candidate!=NULL); + break; + } + } + #else + // first-fit algorithm + // If the page contains free blocks, we are done + if (mi_page_immediate_available(page) || mi_page_is_expandable(page)) { + break; // pick this one + } + + // If the page is completely full, move it to the `mi_pages_full` + // queue so we don't visit long-lived pages too often. + mi_assert_internal(!mi_page_is_in_full(page) && !mi_page_immediate_available(page)); + mi_page_to_full(page, pq); + #endif + + page = next; + } // for each page + + mi_heap_stat_counter_increase(heap, page_searches, count); + + // set the page to the best candidate + if (page_candidate != NULL) { + page = page_candidate; + } + if (page != NULL && !mi_page_immediate_available(page)) { + mi_assert_internal(mi_page_is_expandable(page)); + mi_page_extend_free(heap, page, heap->tld); + } + + if (page == NULL) { + _mi_heap_collect_retired(heap, false); // perhaps make a page available? + page = mi_page_fresh(heap, pq); + if (page == NULL && first_try) { + // out-of-memory _or_ an abandoned page with free blocks was reclaimed, try once again + page = mi_page_queue_find_free_ex(heap, pq, false); + } + } + else { + // move the page to the front of the queue + mi_page_queue_move_to_front(heap, pq, page); + page->retire_expire = 0; + // _mi_heap_collect_retired(heap, false); // update retire counts; note: increases rss on MemoryLoad bench so don't do this + } + mi_assert_internal(page == NULL || mi_page_immediate_available(page)); + + + return page; +} + + + +// Find a page with free blocks of `size`. +static inline mi_page_t* mi_find_free_page(mi_heap_t* heap, size_t size) { + mi_page_queue_t* pq = mi_page_queue(heap, size); + + // check the first page: we even do this with candidate search or otherwise we re-search every time + mi_page_t* page = pq->first; + if (page != NULL) { + #if (MI_SECURE>=3) // in secure mode, we extend half the time to increase randomness + if (page->capacity < page->reserved && ((_mi_heap_random_next(heap) & 1) == 1)) { + mi_page_extend_free(heap, page, heap->tld); + mi_assert_internal(mi_page_immediate_available(page)); + } + else + #endif + { + _mi_page_free_collect(page,false); + } + + if (mi_page_immediate_available(page)) { + page->retire_expire = 0; + return page; // fast path + } + } + + return mi_page_queue_find_free_ex(heap, pq, true); +} + + +/* ----------------------------------------------------------- + Users can register a deferred free function called + when the `free` list is empty. Since the `local_free` + is separate this is deterministically called after + a certain number of allocations. +----------------------------------------------------------- */ + +static mi_deferred_free_fun* volatile deferred_free = NULL; +static _Atomic(void*) deferred_arg; // = NULL + +void _mi_deferred_free(mi_heap_t* heap, bool force) { + heap->tld->heartbeat++; + if (deferred_free != NULL && !heap->tld->recurse) { + heap->tld->recurse = true; + deferred_free(force, heap->tld->heartbeat, mi_atomic_load_ptr_relaxed(void,&deferred_arg)); + heap->tld->recurse = false; + } +} + +void mi_register_deferred_free(mi_deferred_free_fun* fn, void* arg) mi_attr_noexcept { + deferred_free = fn; + mi_atomic_store_ptr_release(void,&deferred_arg, arg); +} + + +/* ----------------------------------------------------------- + General allocation +----------------------------------------------------------- */ + +// Large and huge page allocation. +// Huge pages contain just one block, and the segment contains just that page (as `MI_SEGMENT_HUGE`). +// Huge pages are also use if the requested alignment is very large (> MI_BLOCK_ALIGNMENT_MAX) +// so their size is not always `> MI_LARGE_OBJ_SIZE_MAX`. +static mi_page_t* mi_large_huge_page_alloc(mi_heap_t* heap, size_t size, size_t page_alignment) { + size_t block_size = _mi_os_good_alloc_size(size); + mi_assert_internal(mi_bin(block_size) == MI_BIN_HUGE || page_alignment > 0); + bool is_huge = (block_size > MI_LARGE_OBJ_SIZE_MAX || page_alignment > 0); + #if MI_HUGE_PAGE_ABANDON + mi_page_queue_t* pq = (is_huge ? NULL : mi_page_queue(heap, block_size)); + #else + mi_page_queue_t* pq = mi_page_queue(heap, is_huge ? MI_LARGE_OBJ_SIZE_MAX+1 : block_size); + mi_assert_internal(!is_huge || mi_page_queue_is_huge(pq)); + #endif + mi_page_t* page = mi_page_fresh_alloc(heap, pq, block_size, page_alignment); + if (page != NULL) { + mi_assert_internal(mi_page_immediate_available(page)); + + if (is_huge) { + mi_assert_internal(mi_page_is_huge(page)); + mi_assert_internal(_mi_page_segment(page)->kind == MI_SEGMENT_HUGE); + mi_assert_internal(_mi_page_segment(page)->used==1); + #if MI_HUGE_PAGE_ABANDON + mi_assert_internal(_mi_page_segment(page)->thread_id==0); // abandoned, not in the huge queue + mi_page_set_heap(page, NULL); + #endif + } + else { + mi_assert_internal(!mi_page_is_huge(page)); + } + + const size_t bsize = mi_page_usable_block_size(page); // note: not `mi_page_block_size` to account for padding + /*if (bsize <= MI_LARGE_OBJ_SIZE_MAX) { + mi_heap_stat_increase(heap, malloc_large, bsize); + mi_heap_stat_counter_increase(heap, malloc_large_count, 1); + } + else */ + { + _mi_stat_increase(&heap->tld->stats.malloc_huge, bsize); + _mi_stat_counter_increase(&heap->tld->stats.malloc_huge_count, 1); + } + } + return page; +} + + +// Allocate a page +// Note: in debug mode the size includes MI_PADDING_SIZE and might have overflowed. +static mi_page_t* mi_find_page(mi_heap_t* heap, size_t size, size_t huge_alignment) mi_attr_noexcept { + // huge allocation? + const size_t req_size = size - MI_PADDING_SIZE; // correct for padding_size in case of an overflow on `size` + if mi_unlikely(req_size > (MI_MEDIUM_OBJ_SIZE_MAX - MI_PADDING_SIZE) || huge_alignment > 0) { + if mi_unlikely(req_size > MI_MAX_ALLOC_SIZE) { + _mi_error_message(EOVERFLOW, "allocation request is too large (%zu bytes)\n", req_size); + return NULL; + } + else { + return mi_large_huge_page_alloc(heap,size,huge_alignment); + } + } + else { + // otherwise find a page with free blocks in our size segregated queues + #if MI_PADDING + mi_assert_internal(size >= MI_PADDING_SIZE); + #endif + return mi_find_free_page(heap, size); + } +} + +// Generic allocation routine if the fast path (`alloc.c:mi_page_malloc`) does not succeed. +// Note: in debug mode the size includes MI_PADDING_SIZE and might have overflowed. +// The `huge_alignment` is normally 0 but is set to a multiple of MI_SEGMENT_SIZE for +// very large requested alignments in which case we use a huge segment. +void* _mi_malloc_generic(mi_heap_t* heap, size_t size, bool zero, size_t huge_alignment) mi_attr_noexcept +{ + mi_assert_internal(heap != NULL); + + // initialize if necessary + if mi_unlikely(!mi_heap_is_initialized(heap)) { + heap = mi_heap_get_default(); // calls mi_thread_init + if mi_unlikely(!mi_heap_is_initialized(heap)) { return NULL; } + } + mi_assert_internal(mi_heap_is_initialized(heap)); + + // do administrative tasks every N generic mallocs + if mi_unlikely(++heap->generic_count >= 100) { + heap->generic_collect_count += heap->generic_count; + heap->generic_count = 0; + // call potential deferred free routines + _mi_deferred_free(heap, false); + + // free delayed frees from other threads (but skip contended ones) + _mi_heap_delayed_free_partial(heap); + + // collect every once in a while (10000 by default) + const long generic_collect = mi_option_get_clamp(mi_option_generic_collect, 1, 1000000L); + if (heap->generic_collect_count >= generic_collect) { + heap->generic_collect_count = 0; + mi_heap_collect(heap, false /* force? */); + } + } + + // find (or allocate) a page of the right size + mi_page_t* page = mi_find_page(heap, size, huge_alignment); + if mi_unlikely(page == NULL) { // first time out of memory, try to collect and retry the allocation once more + mi_heap_collect(heap, true /* force */); + page = mi_find_page(heap, size, huge_alignment); + } + + if mi_unlikely(page == NULL) { // out of memory + const size_t req_size = size - MI_PADDING_SIZE; // correct for padding_size in case of an overflow on `size` + _mi_error_message(ENOMEM, "unable to allocate memory (%zu bytes)\n", req_size); + return NULL; + } + + mi_assert_internal(mi_page_immediate_available(page)); + mi_assert_internal(mi_page_block_size(page) >= size); + + // and try again, this time succeeding! (i.e. this should never recurse through _mi_page_malloc) + void* p; + if mi_unlikely(zero && mi_page_is_huge(page)) { + // note: we cannot call _mi_page_malloc with zeroing for huge blocks; we zero it afterwards in that case. + p = _mi_page_malloc(heap, page, size); + mi_assert_internal(p != NULL); + _mi_memzero_aligned(p, mi_page_usable_block_size(page)); + } + else { + p = _mi_page_malloc_zero(heap, page, size, zero); + mi_assert_internal(p != NULL); + } + // move singleton pages to the full queue + if (page->reserved == page->used) { + mi_page_to_full(page, mi_page_queue_of(page)); + } + return p; +} diff --git a/compat/mimalloc/prim/osx/prim.c b/compat/mimalloc/prim/osx/prim.c new file mode 100644 index 00000000000000..8a2f4e8aa47316 --- /dev/null +++ b/compat/mimalloc/prim/osx/prim.c @@ -0,0 +1,9 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2023, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ + +// We use the unix/prim.c with the mmap API on macOSX +#include "../unix/prim.c" diff --git a/compat/mimalloc/prim/prim.c b/compat/mimalloc/prim/prim.c new file mode 100644 index 00000000000000..2002853f2854be --- /dev/null +++ b/compat/mimalloc/prim/prim.c @@ -0,0 +1,76 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2023, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ + +// Select the implementation of the primitives +// depending on the OS. + +#if defined(_WIN32) +#include "windows/prim.c" // VirtualAlloc (Windows) + +#elif defined(__APPLE__) +#include "osx/prim.c" // macOSX (actually defers to mmap in unix/prim.c) + +#elif defined(__wasi__) +#define MI_USE_SBRK +#include "wasi/prim.c" // memory-grow or sbrk (Wasm) + +#elif defined(__EMSCRIPTEN__) +#include "emscripten/prim.c" // emmalloc_*, + pthread support + +#else +#include "unix/prim.c" // mmap() (Linux, macOSX, BSD, Illumnos, Haiku, DragonFly, etc.) + +#endif + +// Generic process initialization +#ifndef MI_PRIM_HAS_PROCESS_ATTACH +#if defined(__GNUC__) || defined(__clang__) + // gcc,clang: use the constructor/destructor attribute + // which for both seem to run before regular constructors/destructors + #if defined(__clang__) + #define mi_attr_constructor __attribute__((constructor(101))) + #define mi_attr_destructor __attribute__((destructor(101))) + #else + #define mi_attr_constructor __attribute__((constructor)) + #define mi_attr_destructor __attribute__((destructor)) + #endif + static void mi_attr_constructor mi_process_attach(void) { + _mi_process_load(); + } + static void mi_attr_destructor mi_process_detach(void) { + _mi_process_done(); + } +#elif defined(__cplusplus) + // C++: use static initialization to detect process start/end + // This is not guaranteed to be first/last but the best we can generally do? + struct mi_init_done_t { + mi_init_done_t() { + _mi_process_load(); + } + ~mi_init_done_t() { + _mi_process_done(); + } + }; + static mi_init_done_t mi_init_done; + #else + #pragma message("define a way to call _mi_process_load/done on your platform") +#endif +#endif + +// Generic allocator init/done callback +#ifndef MI_PRIM_HAS_ALLOCATOR_INIT +bool _mi_is_redirected(void) { + return false; +} +bool _mi_allocator_init(const char** message) { + if (message != NULL) { *message = NULL; } + return true; +} +void _mi_allocator_done(void) { + // nothing to do +} +#endif diff --git a/compat/mimalloc/prim/unix/prim.c b/compat/mimalloc/prim/unix/prim.c new file mode 100644 index 00000000000000..ad6ca2a9e5db57 --- /dev/null +++ b/compat/mimalloc/prim/unix/prim.c @@ -0,0 +1,907 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2023, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ + +// This file is included in `src/prim/prim.c` + +#ifndef _DEFAULT_SOURCE +#define _DEFAULT_SOURCE // ensure mmap flags and syscall are defined +#endif + +#if defined(__sun) +// illumos provides new mman.h api when any of these are defined +// otherwise the old api based on caddr_t which predates the void pointers one. +// stock solaris provides only the former, chose to atomically to discard those +// flags only here rather than project wide tough. +#undef _XOPEN_SOURCE +#undef _POSIX_C_SOURCE +#endif + +#include "mimalloc.h" +#include "mimalloc/internal.h" +#include "mimalloc/prim.h" + +#include // mmap +#include // sysconf +#include // open, close, read, access +#include // getenv, arc4random_buf + +#if defined(__linux__) + #include + #include // PR_SET_VMA + //#if defined(MI_NO_THP) + #include // THP disable + //#endif + #if defined(__GLIBC__) + #include // linux mmap flags + #else + #include + #endif +#elif defined(__APPLE__) + #include + #include + #if !defined(TARGET_OS_OSX) || TARGET_OS_OSX // see issue #879, used to be (!TARGET_IOS_IPHONE && !TARGET_IOS_SIMULATOR) + #include // VM_MAKE_TAG, VM_FLAGS_SUPERPAGE_SIZE_2MB, etc. + #endif + #if !defined(MAC_OS_X_VERSION_10_7) + #define MAC_OS_X_VERSION_10_7 1070 + #endif +#elif defined(__FreeBSD__) || defined(__DragonFly__) + #include + #if __FreeBSD_version >= 1200000 + #include + #include + #endif + #include +#endif + +#if (defined(__linux__) && !defined(__ANDROID__)) || defined(__FreeBSD__) + #define MI_HAS_SYSCALL_H + #include +#endif + +#if !defined(MADV_DONTNEED) && defined(POSIX_MADV_DONTNEED) // QNX +#define MADV_DONTNEED POSIX_MADV_DONTNEED +#endif +#if !defined(MADV_FREE) && defined(POSIX_MADV_FREE) // QNX +#define MADV_FREE POSIX_MADV_FREE +#endif + + +//------------------------------------------------------------------------------------ +// Use syscalls for some primitives to allow for libraries that override open/read/close etc. +// and do allocation themselves; using syscalls prevents recursion when mimalloc is +// still initializing (issue #713) +// Declare inline to avoid unused function warnings. +//------------------------------------------------------------------------------------ + +#if defined(MI_HAS_SYSCALL_H) && defined(SYS_open) && defined(SYS_close) && defined(SYS_read) && defined(SYS_access) + +static inline int mi_prim_open(const char* fpath, int open_flags) { + return syscall(SYS_open,fpath,open_flags,0); +} +static inline ssize_t mi_prim_read(int fd, void* buf, size_t bufsize) { + return syscall(SYS_read,fd,buf,bufsize); +} +static inline int mi_prim_close(int fd) { + return syscall(SYS_close,fd); +} +static inline int mi_prim_access(const char *fpath, int mode) { + return syscall(SYS_access,fpath,mode); +} + +#else + +static inline int mi_prim_open(const char* fpath, int open_flags) { + return open(fpath,open_flags); +} +static inline ssize_t mi_prim_read(int fd, void* buf, size_t bufsize) { + return read(fd,buf,bufsize); +} +static inline int mi_prim_close(int fd) { + return close(fd); +} +static inline int mi_prim_access(const char *fpath, int mode) { + return access(fpath,mode); +} + +#endif + + + +//--------------------------------------------- +// init +//--------------------------------------------- + +static bool unix_detect_overcommit(void) { + bool os_overcommit = true; +#if defined(__linux__) + int fd = mi_prim_open("/proc/sys/vm/overcommit_memory", O_RDONLY); + if (fd >= 0) { + char buf[32]; + ssize_t nread = mi_prim_read(fd, &buf, sizeof(buf)); + mi_prim_close(fd); + // + // 0: heuristic overcommit, 1: always overcommit, 2: never overcommit (ignore NORESERVE) + if (nread >= 1) { + os_overcommit = (buf[0] == '0' || buf[0] == '1'); + } + } +#elif defined(__FreeBSD__) + int val = 0; + size_t olen = sizeof(val); + if (sysctlbyname("vm.overcommit", &val, &olen, NULL, 0) == 0) { + os_overcommit = (val != 0); + } +#else + // default: overcommit is true +#endif + return os_overcommit; +} + +void _mi_prim_mem_init( mi_os_mem_config_t* config ) +{ + long psize = sysconf(_SC_PAGESIZE); + if (psize > 0) { + config->page_size = (size_t)psize; + config->alloc_granularity = (size_t)psize; + #if defined(_SC_PHYS_PAGES) + long pphys = sysconf(_SC_PHYS_PAGES); + const size_t psize_in_kib = (size_t)psize / MI_KiB; + if (psize_in_kib > 0 && pphys > 0 && (size_t)pphys <= (SIZE_MAX/psize_in_kib)) { + config->physical_memory_in_kib = (size_t)pphys * psize_in_kib; + } + #endif + } + config->large_page_size = 2*MI_MiB; // TODO: can we query the OS for this? + config->has_overcommit = unix_detect_overcommit(); + config->has_partial_free = true; // mmap can free in parts + config->has_virtual_reserve = true; // todo: check if this true for NetBSD? (for anonymous mmap with PROT_NONE) + + // disable transparent huge pages for this process? + #if (defined(__linux__) || defined(__ANDROID__)) && defined(PR_GET_THP_DISABLE) + #if defined(MI_NO_THP) + if (true) + #else + if (!mi_option_is_enabled(mi_option_allow_large_os_pages)) // disable THP also if large OS pages are not allowed in the options + #endif + { + int val = 0; + if (prctl(PR_GET_THP_DISABLE, &val, 0, 0, 0) != 0) { + // Most likely since distros often come with always/madvise settings. + val = 1; + // Disabling only for mimalloc process rather than touching system wide settings + (void)prctl(PR_SET_THP_DISABLE, &val, 0, 0, 0); + } + } + #endif +} + + +//--------------------------------------------- +// free +//--------------------------------------------- + +int _mi_prim_free(void* addr, size_t size ) { + bool err = (munmap(addr, size) == -1); + return (err ? errno : 0); +} + + +//--------------------------------------------- +// mmap +//--------------------------------------------- + +static int unix_madvise(void* addr, size_t size, int advice) { + #if defined(__sun) + int res = madvise((caddr_t)addr, size, advice); // Solaris needs cast (issue #520) + #elif defined(__QNX__) + int res = posix_madvise(addr, size, advice); + #else + int res = madvise(addr, size, advice); + #endif + return (res==0 ? 0 : errno); +} + +static void* unix_mmap_prim(void* addr, size_t size, int protect_flags, int flags, int fd) { + void* p = mmap(addr, size, protect_flags, flags, fd, 0 /* offset */); + #if (defined(__linux__) && defined(PR_SET_VMA)) + if (p!=MAP_FAILED && p!=NULL) { + prctl(PR_SET_VMA, PR_SET_VMA_ANON_NAME, p, size, "mimalloc"); + } + #endif + return p; +} + +static void* unix_mmap_prim_aligned(void* addr, size_t size, size_t try_alignment, int protect_flags, int flags, int fd) { + MI_UNUSED(try_alignment); + void* p = NULL; + #if defined(MAP_ALIGNED) // BSD + if (addr == NULL && try_alignment > 1 && (try_alignment % _mi_os_page_size()) == 0) { + size_t n = mi_bsr(try_alignment); + if (((size_t)1 << n) == try_alignment && n >= 12 && n <= 30) { // alignment is a power of 2 and 4096 <= alignment <= 1GiB + p = unix_mmap_prim(addr, size, protect_flags, flags | MAP_ALIGNED(n), fd); + if (p==MAP_FAILED || !_mi_is_aligned(p,try_alignment)) { + int err = errno; + _mi_trace_message("unable to directly request aligned OS memory (error: %d (0x%x), size: 0x%zx bytes, alignment: 0x%zx, hint address: %p)\n", err, err, size, try_alignment, addr); + } + if (p!=MAP_FAILED) return p; + // fall back to regular mmap + } + } + #elif defined(MAP_ALIGN) // Solaris + if (addr == NULL && try_alignment > 1 && (try_alignment % _mi_os_page_size()) == 0) { + p = unix_mmap_prim((void*)try_alignment, size, protect_flags, flags | MAP_ALIGN, fd); // addr parameter is the required alignment + if (p!=MAP_FAILED) return p; + // fall back to regular mmap + } + #endif + #if (MI_INTPTR_SIZE >= 8) && !defined(MAP_ALIGNED) + // on 64-bit systems, use the virtual address area after 2TiB for 4MiB aligned allocations + if (addr == NULL) { + void* hint = _mi_os_get_aligned_hint(try_alignment, size); + if (hint != NULL) { + p = unix_mmap_prim(hint, size, protect_flags, flags, fd); + if (p==MAP_FAILED || !_mi_is_aligned(p,try_alignment)) { + #if MI_TRACK_ENABLED // asan sometimes does not instrument errno correctly? + int err = 0; + #else + int err = errno; + #endif + _mi_trace_message("unable to directly request hinted aligned OS memory (error: %d (0x%x), size: 0x%zx bytes, alignment: 0x%zx, hint address: %p)\n", err, err, size, try_alignment, hint); + } + if (p!=MAP_FAILED) return p; + // fall back to regular mmap + } + } + #endif + // regular mmap + p = unix_mmap_prim(addr, size, protect_flags, flags, fd); + if (p!=MAP_FAILED) return p; + // failed to allocate + return NULL; +} + +static int unix_mmap_fd(void) { + #if defined(VM_MAKE_TAG) + // macOS: tracking anonymous page with a specific ID. (All up to 98 are taken officially but LLVM sanitizers had taken 99) + int os_tag = (int)mi_option_get(mi_option_os_tag); + if (os_tag < 100 || os_tag > 255) { os_tag = 254; } + return VM_MAKE_TAG(os_tag); + #else + return -1; + #endif +} + +static void* unix_mmap(void* addr, size_t size, size_t try_alignment, int protect_flags, bool large_only, bool allow_large, bool* is_large) { + #if !defined(MAP_ANONYMOUS) + #define MAP_ANONYMOUS MAP_ANON + #endif + #if !defined(MAP_NORESERVE) + #define MAP_NORESERVE 0 + #endif + void* p = NULL; + const int fd = unix_mmap_fd(); + int flags = MAP_PRIVATE | MAP_ANONYMOUS; + if (_mi_os_has_overcommit()) { + flags |= MAP_NORESERVE; + } + #if defined(PROT_MAX) + protect_flags |= PROT_MAX(PROT_READ | PROT_WRITE); // BSD + #endif + // huge page allocation + if (allow_large && (large_only || (_mi_os_use_large_page(size, try_alignment) && mi_option_get(mi_option_allow_large_os_pages) == 1))) { + static _Atomic(size_t) large_page_try_ok; // = 0; + size_t try_ok = mi_atomic_load_acquire(&large_page_try_ok); + if (!large_only && try_ok > 0) { + // If the OS is not configured for large OS pages, or the user does not have + // enough permission, the `mmap` will always fail (but it might also fail for other reasons). + // Therefore, once a large page allocation failed, we don't try again for `large_page_try_ok` times + // to avoid too many failing calls to mmap. + mi_atomic_cas_strong_acq_rel(&large_page_try_ok, &try_ok, try_ok - 1); + } + else { + int lflags = flags & ~MAP_NORESERVE; // using NORESERVE on huge pages seems to fail on Linux + int lfd = fd; + #ifdef MAP_ALIGNED_SUPER + lflags |= MAP_ALIGNED_SUPER; + #endif + #ifdef MAP_HUGETLB + lflags |= MAP_HUGETLB; + #endif + #ifdef MAP_HUGE_1GB + static bool mi_huge_pages_available = true; + if (large_only && (size % MI_GiB) == 0 && mi_huge_pages_available) { + lflags |= MAP_HUGE_1GB; + } + else + #endif + { + #ifdef MAP_HUGE_2MB + lflags |= MAP_HUGE_2MB; + #endif + } + #ifdef VM_FLAGS_SUPERPAGE_SIZE_2MB + lfd |= VM_FLAGS_SUPERPAGE_SIZE_2MB; + #endif + if (large_only || lflags != flags) { + // try large OS page allocation + *is_large = true; + p = unix_mmap_prim_aligned(addr, size, try_alignment, protect_flags, lflags, lfd); + #ifdef MAP_HUGE_1GB + if (p == NULL && (lflags & MAP_HUGE_1GB) == MAP_HUGE_1GB) { + mi_huge_pages_available = false; // don't try huge 1GiB pages again + if (large_only) { + _mi_warning_message("unable to allocate huge (1GiB) page, trying large (2MiB) pages instead (errno: %i)\n", errno); + } + lflags = ((lflags & ~MAP_HUGE_1GB) | MAP_HUGE_2MB); + p = unix_mmap_prim_aligned(addr, size, try_alignment, protect_flags, lflags, lfd); + } + #endif + if (large_only) return p; + if (p == NULL) { + mi_atomic_store_release(&large_page_try_ok, (size_t)8); // on error, don't try again for the next N allocations + } + } + } + } + // regular allocation + if (p == NULL) { + *is_large = false; + p = unix_mmap_prim_aligned(addr, size, try_alignment, protect_flags, flags, fd); + if (p != NULL) { + #if defined(MADV_HUGEPAGE) + // Many Linux systems don't allow MAP_HUGETLB but they support instead + // transparent huge pages (THP). Generally, it is not required to call `madvise` with MADV_HUGE + // though since properly aligned allocations will already use large pages if available + // in that case -- in particular for our large regions (in `memory.c`). + // However, some systems only allow THP if called with explicit `madvise`, so + // when large OS pages are enabled for mimalloc, we call `madvise` anyways. + if (allow_large && _mi_os_use_large_page(size, try_alignment)) { + if (unix_madvise(p, size, MADV_HUGEPAGE) == 0) { + // *is_large = true; // possibly + }; + } + #elif defined(__sun) + if (allow_large && _mi_os_use_large_page(size, try_alignment)) { + struct memcntl_mha cmd = {0}; + cmd.mha_pagesize = _mi_os_large_page_size(); + cmd.mha_cmd = MHA_MAPSIZE_VA; + if (memcntl((caddr_t)p, size, MC_HAT_ADVISE, (caddr_t)&cmd, 0, 0) == 0) { + // *is_large = true; // possibly + } + } + #endif + } + } + return p; +} + +// Note: the `try_alignment` is just a hint and the returned pointer is not guaranteed to be aligned. +int _mi_prim_alloc(void* hint_addr, size_t size, size_t try_alignment, bool commit, bool allow_large, bool* is_large, bool* is_zero, void** addr) { + mi_assert_internal(size > 0 && (size % _mi_os_page_size()) == 0); + mi_assert_internal(commit || !allow_large); + mi_assert_internal(try_alignment > 0); + + *is_zero = true; + int protect_flags = (commit ? (PROT_WRITE | PROT_READ) : PROT_NONE); + *addr = unix_mmap(hint_addr, size, try_alignment, protect_flags, false, allow_large, is_large); + return (*addr != NULL ? 0 : errno); +} + + +//--------------------------------------------- +// Commit/Reset +//--------------------------------------------- + +static void unix_mprotect_hint(int err) { + #if defined(__linux__) && (MI_SECURE>=2) // guard page around every mimalloc page + if (err == ENOMEM) { + _mi_warning_message("The next warning may be caused by a low memory map limit.\n" + " On Linux this is controlled by the vm.max_map_count -- maybe increase it?\n" + " For example: sudo sysctl -w vm.max_map_count=262144\n"); + } + #else + MI_UNUSED(err); + #endif +} + +int _mi_prim_commit(void* start, size_t size, bool* is_zero) { + // commit: ensure we can access the area + // note: we may think that *is_zero can be true since the memory + // was either from mmap PROT_NONE, or from decommit MADV_DONTNEED, but + // we sometimes call commit on a range with still partially committed + // memory and `mprotect` does not zero the range. + *is_zero = false; + int err = mprotect(start, size, (PROT_READ | PROT_WRITE)); + if (err != 0) { + err = errno; + unix_mprotect_hint(err); + } + return err; +} + +int _mi_prim_decommit(void* start, size_t size, bool* needs_recommit) { + int err = 0; + // decommit: use MADV_DONTNEED as it decreases rss immediately (unlike MADV_FREE) + err = unix_madvise(start, size, MADV_DONTNEED); + #if !MI_DEBUG && !MI_SECURE + *needs_recommit = false; + #else + *needs_recommit = true; + mprotect(start, size, PROT_NONE); + #endif + /* + // decommit: use mmap with MAP_FIXED and PROT_NONE to discard the existing memory (and reduce rss) + *needs_recommit = true; + const int fd = unix_mmap_fd(); + void* p = mmap(start, size, PROT_NONE, (MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS | MAP_NORESERVE), fd, 0); + if (p != start) { err = errno; } + */ + return err; +} + +int _mi_prim_reset(void* start, size_t size) { + // We try to use `MADV_FREE` as that is the fastest. A drawback though is that it + // will not reduce the `rss` stats in tools like `top` even though the memory is available + // to other processes. With the default `MIMALLOC_PURGE_DECOMMITS=1` we ensure that by + // default `MADV_DONTNEED` is used though. + #if defined(MADV_FREE) + static _Atomic(size_t) advice = MI_ATOMIC_VAR_INIT(MADV_FREE); + int oadvice = (int)mi_atomic_load_relaxed(&advice); + int err; + while ((err = unix_madvise(start, size, oadvice)) != 0 && errno == EAGAIN) { errno = 0; }; + if (err != 0 && errno == EINVAL && oadvice == MADV_FREE) { + // if MADV_FREE is not supported, fall back to MADV_DONTNEED from now on + mi_atomic_store_release(&advice, (size_t)MADV_DONTNEED); + err = unix_madvise(start, size, MADV_DONTNEED); + } + #else + int err = unix_madvise(start, size, MADV_DONTNEED); + #endif + return err; +} + +int _mi_prim_protect(void* start, size_t size, bool protect) { + int err = mprotect(start, size, protect ? PROT_NONE : (PROT_READ | PROT_WRITE)); + if (err != 0) { err = errno; } + unix_mprotect_hint(err); + return err; +} + + + +//--------------------------------------------- +// Huge page allocation +//--------------------------------------------- + +#if (MI_INTPTR_SIZE >= 8) && !defined(__HAIKU__) && !defined(__CYGWIN__) + +#ifndef MPOL_PREFERRED +#define MPOL_PREFERRED 1 +#endif + +#if defined(MI_HAS_SYSCALL_H) && defined(SYS_mbind) +static long mi_prim_mbind(void* start, unsigned long len, unsigned long mode, const unsigned long* nmask, unsigned long maxnode, unsigned flags) { + return syscall(SYS_mbind, start, len, mode, nmask, maxnode, flags); +} +#else +static long mi_prim_mbind(void* start, unsigned long len, unsigned long mode, const unsigned long* nmask, unsigned long maxnode, unsigned flags) { + MI_UNUSED(start); MI_UNUSED(len); MI_UNUSED(mode); MI_UNUSED(nmask); MI_UNUSED(maxnode); MI_UNUSED(flags); + return 0; +} +#endif + +int _mi_prim_alloc_huge_os_pages(void* hint_addr, size_t size, int numa_node, bool* is_zero, void** addr) { + bool is_large = true; + *is_zero = true; + *addr = unix_mmap(hint_addr, size, MI_SEGMENT_SIZE, PROT_READ | PROT_WRITE, true, true, &is_large); + if (*addr != NULL && numa_node >= 0 && numa_node < 8*MI_INTPTR_SIZE) { // at most 64 nodes + unsigned long numa_mask = (1UL << numa_node); + // TODO: does `mbind` work correctly for huge OS pages? should we + // use `set_mempolicy` before calling mmap instead? + // see: + long err = mi_prim_mbind(*addr, size, MPOL_PREFERRED, &numa_mask, 8*MI_INTPTR_SIZE, 0); + if (err != 0) { + err = errno; + _mi_warning_message("failed to bind huge (1GiB) pages to numa node %d (error: %d (0x%x))\n", numa_node, err, err); + } + } + return (*addr != NULL ? 0 : errno); +} + +#else + +int _mi_prim_alloc_huge_os_pages(void* hint_addr, size_t size, int numa_node, bool* is_zero, void** addr) { + MI_UNUSED(hint_addr); MI_UNUSED(size); MI_UNUSED(numa_node); + *is_zero = false; + *addr = NULL; + return ENOMEM; +} + +#endif + +//--------------------------------------------- +// NUMA nodes +//--------------------------------------------- + +#if defined(__linux__) + +size_t _mi_prim_numa_node(void) { + #if defined(MI_HAS_SYSCALL_H) && defined(SYS_getcpu) + unsigned long node = 0; + unsigned long ncpu = 0; + long err = syscall(SYS_getcpu, &ncpu, &node, NULL); + if (err != 0) return 0; + return node; + #else + return 0; + #endif +} + +size_t _mi_prim_numa_node_count(void) { + char buf[128]; + unsigned node = 0; + for(node = 0; node < 256; node++) { + // enumerate node entries -- todo: it there a more efficient way to do this? (but ensure there is no allocation) + _mi_snprintf(buf, 127, "/sys/devices/system/node/node%u", node + 1); + if (mi_prim_access(buf,R_OK) != 0) break; + } + return (node+1); +} + +#elif defined(__FreeBSD__) && __FreeBSD_version >= 1200000 + +size_t _mi_prim_numa_node(void) { + domainset_t dom; + size_t node; + int policy; + if (cpuset_getdomain(CPU_LEVEL_CPUSET, CPU_WHICH_PID, -1, sizeof(dom), &dom, &policy) == -1) return 0ul; + for (node = 0; node < MAXMEMDOM; node++) { + if (DOMAINSET_ISSET(node, &dom)) return node; + } + return 0ul; +} + +size_t _mi_prim_numa_node_count(void) { + size_t ndomains = 0; + size_t len = sizeof(ndomains); + if (sysctlbyname("vm.ndomains", &ndomains, &len, NULL, 0) == -1) return 0ul; + return ndomains; +} + +#elif defined(__DragonFly__) + +size_t _mi_prim_numa_node(void) { + // TODO: DragonFly does not seem to provide any userland means to get this information. + return 0ul; +} + +size_t _mi_prim_numa_node_count(void) { + size_t ncpus = 0, nvirtcoresperphys = 0; + size_t len = sizeof(size_t); + if (sysctlbyname("hw.ncpu", &ncpus, &len, NULL, 0) == -1) return 0ul; + if (sysctlbyname("hw.cpu_topology_ht_ids", &nvirtcoresperphys, &len, NULL, 0) == -1) return 0ul; + return nvirtcoresperphys * ncpus; +} + +#else + +size_t _mi_prim_numa_node(void) { + return 0; +} + +size_t _mi_prim_numa_node_count(void) { + return 1; +} + +#endif + +// ---------------------------------------------------------------- +// Clock +// ---------------------------------------------------------------- + +#include + +#if defined(CLOCK_REALTIME) || defined(CLOCK_MONOTONIC) + +mi_msecs_t _mi_prim_clock_now(void) { + struct timespec t; + #ifdef CLOCK_MONOTONIC + clock_gettime(CLOCK_MONOTONIC, &t); + #else + clock_gettime(CLOCK_REALTIME, &t); + #endif + return ((mi_msecs_t)t.tv_sec * 1000) + ((mi_msecs_t)t.tv_nsec / 1000000); +} + +#else + +// low resolution timer +mi_msecs_t _mi_prim_clock_now(void) { + #if !defined(CLOCKS_PER_SEC) || (CLOCKS_PER_SEC == 1000) || (CLOCKS_PER_SEC == 0) + return (mi_msecs_t)clock(); + #elif (CLOCKS_PER_SEC < 1000) + return (mi_msecs_t)clock() * (1000 / (mi_msecs_t)CLOCKS_PER_SEC); + #else + return (mi_msecs_t)clock() / ((mi_msecs_t)CLOCKS_PER_SEC / 1000); + #endif +} + +#endif + + + + +//---------------------------------------------------------------- +// Process info +//---------------------------------------------------------------- + +#if defined(__unix__) || defined(__unix) || defined(unix) || defined(__APPLE__) || defined(__HAIKU__) +#include +#include +#include + +#if defined(__APPLE__) +#include +#endif + +#if defined(__HAIKU__) +#include +#endif + +static mi_msecs_t timeval_secs(const struct timeval* tv) { + return ((mi_msecs_t)tv->tv_sec * 1000L) + ((mi_msecs_t)tv->tv_usec / 1000L); +} + +void _mi_prim_process_info(mi_process_info_t* pinfo) +{ + struct rusage rusage; + getrusage(RUSAGE_SELF, &rusage); + pinfo->utime = timeval_secs(&rusage.ru_utime); + pinfo->stime = timeval_secs(&rusage.ru_stime); +#if !defined(__HAIKU__) + pinfo->page_faults = rusage.ru_majflt; +#endif +#if defined(__HAIKU__) + // Haiku does not have (yet?) a way to + // get these stats per process + thread_info tid; + area_info mem; + ssize_t c; + get_thread_info(find_thread(0), &tid); + while (get_next_area_info(tid.team, &c, &mem) == B_OK) { + pinfo->peak_rss += mem.ram_size; + } + pinfo->page_faults = 0; +#elif defined(__APPLE__) + pinfo->peak_rss = rusage.ru_maxrss; // macos reports in bytes + #ifdef MACH_TASK_BASIC_INFO + struct mach_task_basic_info info; + mach_msg_type_number_t infoCount = MACH_TASK_BASIC_INFO_COUNT; + if (task_info(mach_task_self(), MACH_TASK_BASIC_INFO, (task_info_t)&info, &infoCount) == KERN_SUCCESS) { + pinfo->current_rss = (size_t)info.resident_size; + } + #else + struct task_basic_info info; + mach_msg_type_number_t infoCount = TASK_BASIC_INFO_COUNT; + if (task_info(mach_task_self(), TASK_BASIC_INFO, (task_info_t)&info, &infoCount) == KERN_SUCCESS) { + pinfo->current_rss = (size_t)info.resident_size; + } + #endif +#else + pinfo->peak_rss = rusage.ru_maxrss * 1024; // Linux/BSD report in KiB +#endif + // use defaults for commit +} + +#else + +#ifndef __wasi__ +// WebAssembly instances are not processes +#pragma message("define a way to get process info") +#endif + +void _mi_prim_process_info(mi_process_info_t* pinfo) +{ + // use defaults + MI_UNUSED(pinfo); +} + +#endif + + +//---------------------------------------------------------------- +// Output +//---------------------------------------------------------------- + +void _mi_prim_out_stderr( const char* msg ) { + fputs(msg,stderr); +} + + +//---------------------------------------------------------------- +// Environment +//---------------------------------------------------------------- + +#if !defined(MI_USE_ENVIRON) || (MI_USE_ENVIRON!=0) +// On Posix systemsr use `environ` to access environment variables +// even before the C runtime is initialized. +#if defined(__APPLE__) && defined(__has_include) && __has_include() +#include +static char** mi_get_environ(void) { + return (*_NSGetEnviron()); +} +#else +extern char** environ; +static char** mi_get_environ(void) { + return environ; +} +#endif +bool _mi_prim_getenv(const char* name, char* result, size_t result_size) { + if (name==NULL) return false; + const size_t len = _mi_strlen(name); + if (len == 0) return false; + char** env = mi_get_environ(); + if (env == NULL) return false; + // compare up to 10000 entries + for (int i = 0; i < 10000 && env[i] != NULL; i++) { + const char* s = env[i]; + if (_mi_strnicmp(name, s, len) == 0 && s[len] == '=') { // case insensitive + // found it + _mi_strlcpy(result, s + len + 1, result_size); + return true; + } + } + return false; +} +#else +// fallback: use standard C `getenv` but this cannot be used while initializing the C runtime +bool _mi_prim_getenv(const char* name, char* result, size_t result_size) { + // cannot call getenv() when still initializing the C runtime. + if (_mi_preloading()) return false; + const char* s = getenv(name); + if (s == NULL) { + // we check the upper case name too. + char buf[64+1]; + size_t len = _mi_strnlen(name,sizeof(buf)-1); + for (size_t i = 0; i < len; i++) { + buf[i] = _mi_toupper(name[i]); + } + buf[len] = 0; + s = getenv(buf); + } + if (s == NULL || _mi_strnlen(s,result_size) >= result_size) return false; + _mi_strlcpy(result, s, result_size); + return true; +} +#endif // !MI_USE_ENVIRON + + +//---------------------------------------------------------------- +// Random +//---------------------------------------------------------------- + +#if defined(__APPLE__) && defined(MAC_OS_X_VERSION_10_15) && (MAC_OS_X_VERSION_MIN_REQUIRED >= MAC_OS_X_VERSION_10_15) +#include +#include + +bool _mi_prim_random_buf(void* buf, size_t buf_len) { + // We prefer CCRandomGenerateBytes as it returns an error code while arc4random_buf + // may fail silently on macOS. See PR #390, and + return (CCRandomGenerateBytes(buf, buf_len) == kCCSuccess); +} + +#elif defined(__ANDROID__) || defined(__DragonFly__) || \ + defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) || \ + defined(__sun) || \ + (defined(__APPLE__) && (MAC_OS_X_VERSION_MIN_REQUIRED >= MAC_OS_X_VERSION_10_7)) + +bool _mi_prim_random_buf(void* buf, size_t buf_len) { + arc4random_buf(buf, buf_len); + return true; +} + +#elif defined(__APPLE__) || defined(__linux__) || defined(__HAIKU__) // also for old apple versions < 10.7 (issue #829) + +#include +#include +#include + +bool _mi_prim_random_buf(void* buf, size_t buf_len) { + // Modern Linux provides `getrandom` but different distributions either use `sys/random.h` or `linux/random.h` + // and for the latter the actual `getrandom` call is not always defined. + // (see ) + // We therefore use a syscall directly and fall back dynamically to /dev/urandom when needed. + #if defined(MI_HAS_SYSCALL_H) && defined(SYS_getrandom) + #ifndef GRND_NONBLOCK + #define GRND_NONBLOCK (1) + #endif + static _Atomic(uintptr_t) no_getrandom; // = 0 + if (mi_atomic_load_acquire(&no_getrandom)==0) { + ssize_t ret = syscall(SYS_getrandom, buf, buf_len, GRND_NONBLOCK); + if (ret >= 0) return (buf_len == (size_t)ret); + if (errno != ENOSYS) return false; + mi_atomic_store_release(&no_getrandom, (uintptr_t)1); // don't call again, and fall back to /dev/urandom + } + #endif + int flags = O_RDONLY; + #if defined(O_CLOEXEC) + flags |= O_CLOEXEC; + #endif + int fd = mi_prim_open("/dev/urandom", flags); + if (fd < 0) return false; + size_t count = 0; + while(count < buf_len) { + ssize_t ret = mi_prim_read(fd, (char*)buf + count, buf_len - count); + if (ret<=0) { + if (errno!=EAGAIN && errno!=EINTR) break; + } + else { + count += ret; + } + } + mi_prim_close(fd); + return (count==buf_len); +} + +#else + +bool _mi_prim_random_buf(void* buf, size_t buf_len) { + return false; +} + +#endif + + +//---------------------------------------------------------------- +// Thread init/done +//---------------------------------------------------------------- + +#if defined(MI_USE_PTHREADS) + +// use pthread local storage keys to detect thread ending +// (and used with MI_TLS_PTHREADS for the default heap) +pthread_key_t _mi_heap_default_key = (pthread_key_t)(-1); + +static void mi_pthread_done(void* value) { + if (value!=NULL) { + _mi_thread_done((mi_heap_t*)value); + } +} + +void _mi_prim_thread_init_auto_done(void) { + mi_assert_internal(_mi_heap_default_key == (pthread_key_t)(-1)); + pthread_key_create(&_mi_heap_default_key, &mi_pthread_done); +} + +void _mi_prim_thread_done_auto_done(void) { + if (_mi_heap_default_key != (pthread_key_t)(-1)) { // do not leak the key, see issue #809 + pthread_key_delete(_mi_heap_default_key); + } +} + +void _mi_prim_thread_associate_default_heap(mi_heap_t* heap) { + if (_mi_heap_default_key != (pthread_key_t)(-1)) { // can happen during recursive invocation on freeBSD + pthread_setspecific(_mi_heap_default_key, heap); + } +} + +#else + +void _mi_prim_thread_init_auto_done(void) { + // nothing +} + +void _mi_prim_thread_done_auto_done(void) { + // nothing +} + +void _mi_prim_thread_associate_default_heap(mi_heap_t* heap) { + MI_UNUSED(heap); +} + +#endif diff --git a/compat/mimalloc/prim/windows/prim.c b/compat/mimalloc/prim/windows/prim.c new file mode 100644 index 00000000000000..a080f4bc362bcb --- /dev/null +++ b/compat/mimalloc/prim/windows/prim.c @@ -0,0 +1,825 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2023, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ + +// This file is included in `src/prim/prim.c` + +#include "mimalloc.h" +#include "mimalloc/internal.h" +#include "mimalloc/prim.h" +#include // fputs, stderr + + +//--------------------------------------------- +// Dynamically bind Windows API points for portability +//--------------------------------------------- + +// We use VirtualAlloc2 for aligned allocation, but it is only supported on Windows 10 and Windows Server 2016. +// So, we need to look it up dynamically to run on older systems. (use __stdcall for 32-bit compatibility) +// NtAllocateVirtualAllocEx is used for huge OS page allocation (1GiB) +// We define a minimal MEM_EXTENDED_PARAMETER ourselves in order to be able to compile with older SDK's. +typedef enum MI_MEM_EXTENDED_PARAMETER_TYPE_E { + MiMemExtendedParameterInvalidType = 0, + MiMemExtendedParameterAddressRequirements, + MiMemExtendedParameterNumaNode, + MiMemExtendedParameterPartitionHandle, + MiMemExtendedParameterUserPhysicalHandle, + MiMemExtendedParameterAttributeFlags, + MiMemExtendedParameterMax +} MI_MEM_EXTENDED_PARAMETER_TYPE; + +typedef struct DECLSPEC_ALIGN(8) MI_MEM_EXTENDED_PARAMETER_S { + struct { DWORD64 Type : 8; DWORD64 Reserved : 56; } Type; + union { DWORD64 ULong64; PVOID Pointer; SIZE_T Size; HANDLE Handle; DWORD ULong; } Arg; +} MI_MEM_EXTENDED_PARAMETER; + +typedef struct MI_MEM_ADDRESS_REQUIREMENTS_S { + PVOID LowestStartingAddress; + PVOID HighestEndingAddress; + SIZE_T Alignment; +} MI_MEM_ADDRESS_REQUIREMENTS; + +#define MI_MEM_EXTENDED_PARAMETER_NONPAGED_HUGE 0x00000010 + +#include +typedef PVOID (__stdcall *PVirtualAlloc2)(HANDLE, PVOID, SIZE_T, ULONG, ULONG, MI_MEM_EXTENDED_PARAMETER*, ULONG); +typedef NTSTATUS (__stdcall *PNtAllocateVirtualMemoryEx)(HANDLE, PVOID*, SIZE_T*, ULONG, ULONG, MI_MEM_EXTENDED_PARAMETER*, ULONG); +static PVirtualAlloc2 pVirtualAlloc2 = NULL; +static PNtAllocateVirtualMemoryEx pNtAllocateVirtualMemoryEx = NULL; + +// Similarly, GetNumaProcessorNodeEx is only supported since Windows 7 +typedef struct MI_PROCESSOR_NUMBER_S { WORD Group; BYTE Number; BYTE Reserved; } MI_PROCESSOR_NUMBER; + +typedef VOID (__stdcall *PGetCurrentProcessorNumberEx)(MI_PROCESSOR_NUMBER* ProcNumber); +typedef BOOL (__stdcall *PGetNumaProcessorNodeEx)(MI_PROCESSOR_NUMBER* Processor, PUSHORT NodeNumber); +typedef BOOL (__stdcall* PGetNumaNodeProcessorMaskEx)(USHORT Node, PGROUP_AFFINITY ProcessorMask); +typedef BOOL (__stdcall *PGetNumaProcessorNode)(UCHAR Processor, PUCHAR NodeNumber); +static PGetCurrentProcessorNumberEx pGetCurrentProcessorNumberEx = NULL; +static PGetNumaProcessorNodeEx pGetNumaProcessorNodeEx = NULL; +static PGetNumaNodeProcessorMaskEx pGetNumaNodeProcessorMaskEx = NULL; +static PGetNumaProcessorNode pGetNumaProcessorNode = NULL; + +// Available after Windows XP +typedef BOOL (__stdcall *PGetPhysicallyInstalledSystemMemory)( PULONGLONG TotalMemoryInKilobytes ); + +//--------------------------------------------- +// Enable large page support dynamically (if possible) +//--------------------------------------------- + +static bool win_enable_large_os_pages(size_t* large_page_size) +{ + static bool large_initialized = false; + if (large_initialized) return (_mi_os_large_page_size() > 0); + large_initialized = true; + + // Try to see if large OS pages are supported + // To use large pages on Windows, we first need access permission + // Set "Lock pages in memory" permission in the group policy editor + // + unsigned long err = 0; + HANDLE token = NULL; + BOOL ok = OpenProcessToken(GetCurrentProcess(), TOKEN_ADJUST_PRIVILEGES | TOKEN_QUERY, &token); + if (ok) { + TOKEN_PRIVILEGES tp; + ok = LookupPrivilegeValue(NULL, TEXT("SeLockMemoryPrivilege"), &tp.Privileges[0].Luid); + if (ok) { + tp.PrivilegeCount = 1; + tp.Privileges[0].Attributes = SE_PRIVILEGE_ENABLED; + ok = AdjustTokenPrivileges(token, FALSE, &tp, 0, (PTOKEN_PRIVILEGES)NULL, 0); + if (ok) { + err = GetLastError(); + ok = (err == ERROR_SUCCESS); + if (ok && large_page_size != NULL) { + *large_page_size = GetLargePageMinimum(); + } + } + } + CloseHandle(token); + } + if (!ok) { + if (err == 0) err = GetLastError(); + _mi_warning_message("cannot enable large OS page support, error %lu\n", err); + } + return (ok!=0); +} + + +//--------------------------------------------- +// Initialize +//--------------------------------------------- + +void _mi_prim_mem_init( mi_os_mem_config_t* config ) +{ + config->has_overcommit = false; + config->has_partial_free = false; + config->has_virtual_reserve = true; + // get the page size + SYSTEM_INFO si; + GetSystemInfo(&si); + if (si.dwPageSize > 0) { config->page_size = si.dwPageSize; } + if (si.dwAllocationGranularity > 0) { config->alloc_granularity = si.dwAllocationGranularity; } + // get virtual address bits + if ((uintptr_t)si.lpMaximumApplicationAddress > 0) { + const size_t vbits = MI_SIZE_BITS - mi_clz((uintptr_t)si.lpMaximumApplicationAddress); + config->virtual_address_bits = vbits; + } + + // get the VirtualAlloc2 function + HINSTANCE hDll; + hDll = LoadLibrary(TEXT("kernelbase.dll")); + if (hDll != NULL) { + // use VirtualAlloc2FromApp if possible as it is available to Windows store apps + pVirtualAlloc2 = (PVirtualAlloc2)(void (*)(void))GetProcAddress(hDll, "VirtualAlloc2FromApp"); + if (pVirtualAlloc2==NULL) pVirtualAlloc2 = (PVirtualAlloc2)(void (*)(void))GetProcAddress(hDll, "VirtualAlloc2"); + FreeLibrary(hDll); + } + // NtAllocateVirtualMemoryEx is used for huge page allocation + hDll = LoadLibrary(TEXT("ntdll.dll")); + if (hDll != NULL) { + pNtAllocateVirtualMemoryEx = (PNtAllocateVirtualMemoryEx)(void (*)(void))GetProcAddress(hDll, "NtAllocateVirtualMemoryEx"); + FreeLibrary(hDll); + } + // Try to use Win7+ numa API + hDll = LoadLibrary(TEXT("kernel32.dll")); + if (hDll != NULL) { + pGetCurrentProcessorNumberEx = (PGetCurrentProcessorNumberEx)(void (*)(void))GetProcAddress(hDll, "GetCurrentProcessorNumberEx"); + pGetNumaProcessorNodeEx = (PGetNumaProcessorNodeEx)(void (*)(void))GetProcAddress(hDll, "GetNumaProcessorNodeEx"); + pGetNumaNodeProcessorMaskEx = (PGetNumaNodeProcessorMaskEx)(void (*)(void))GetProcAddress(hDll, "GetNumaNodeProcessorMaskEx"); + pGetNumaProcessorNode = (PGetNumaProcessorNode)(void (*)(void))GetProcAddress(hDll, "GetNumaProcessorNode"); + // Get physical memory (not available on XP, so check dynamically) + PGetPhysicallyInstalledSystemMemory pGetPhysicallyInstalledSystemMemory = (PGetPhysicallyInstalledSystemMemory)(void (*)(void))GetProcAddress(hDll,"GetPhysicallyInstalledSystemMemory"); + if (pGetPhysicallyInstalledSystemMemory != NULL) { + ULONGLONG memInKiB = 0; + if ((*pGetPhysicallyInstalledSystemMemory)(&memInKiB)) { + if (memInKiB > 0 && memInKiB <= SIZE_MAX) { + config->physical_memory_in_kib = (size_t)memInKiB; + } + } + } + FreeLibrary(hDll); + } + // Enable large/huge OS page support? + if (mi_option_is_enabled(mi_option_allow_large_os_pages) || mi_option_is_enabled(mi_option_reserve_huge_os_pages)) { + win_enable_large_os_pages(&config->large_page_size); + } +} + + +//--------------------------------------------- +// Free +//--------------------------------------------- + +int _mi_prim_free(void* addr, size_t size ) { + MI_UNUSED(size); + DWORD errcode = 0; + bool err = (VirtualFree(addr, 0, MEM_RELEASE) == 0); + if (err) { errcode = GetLastError(); } + if (errcode == ERROR_INVALID_ADDRESS) { + // In mi_os_mem_alloc_aligned the fallback path may have returned a pointer inside + // the memory region returned by VirtualAlloc; in that case we need to free using + // the start of the region. + MEMORY_BASIC_INFORMATION info; _mi_memzero_var(info); + VirtualQuery(addr, &info, sizeof(info)); + if (info.AllocationBase < addr && ((uint8_t*)addr - (uint8_t*)info.AllocationBase) < (ptrdiff_t)MI_SEGMENT_SIZE) { + errcode = 0; + err = (VirtualFree(info.AllocationBase, 0, MEM_RELEASE) == 0); + if (err) { errcode = GetLastError(); } + } + } + return (int)errcode; +} + + +//--------------------------------------------- +// VirtualAlloc +//--------------------------------------------- + +static void* win_virtual_alloc_prim_once(void* addr, size_t size, size_t try_alignment, DWORD flags) { + #if (MI_INTPTR_SIZE >= 8) + // on 64-bit systems, try to use the virtual address area after 2TiB for 4MiB aligned allocations + if (addr == NULL) { + void* hint = _mi_os_get_aligned_hint(try_alignment,size); + if (hint != NULL) { + void* p = VirtualAlloc(hint, size, flags, PAGE_READWRITE); + if (p != NULL) return p; + _mi_verbose_message("warning: unable to allocate hinted aligned OS memory (%zu bytes, error code: 0x%x, address: %p, alignment: %zu, flags: 0x%x)\n", size, GetLastError(), hint, try_alignment, flags); + // fall through on error + } + } + #endif + // on modern Windows try use VirtualAlloc2 for aligned allocation + if (addr == NULL && try_alignment > 1 && (try_alignment % _mi_os_page_size()) == 0 && pVirtualAlloc2 != NULL) { + MI_MEM_ADDRESS_REQUIREMENTS reqs = { 0, 0, 0 }; + reqs.Alignment = try_alignment; + MI_MEM_EXTENDED_PARAMETER param = { {0, 0}, {0} }; + param.Type.Type = MiMemExtendedParameterAddressRequirements; + param.Arg.Pointer = &reqs; + void* p = (*pVirtualAlloc2)(GetCurrentProcess(), addr, size, flags, PAGE_READWRITE, ¶m, 1); + if (p != NULL) return p; + _mi_warning_message("unable to allocate aligned OS memory (0x%zx bytes, error code: 0x%x, address: %p, alignment: 0x%zx, flags: 0x%x)\n", size, GetLastError(), addr, try_alignment, flags); + // fall through on error + } + // last resort + return VirtualAlloc(addr, size, flags, PAGE_READWRITE); +} + +static bool win_is_out_of_memory_error(DWORD err) { + switch (err) { + case ERROR_COMMITMENT_MINIMUM: + case ERROR_COMMITMENT_LIMIT: + case ERROR_PAGEFILE_QUOTA: + case ERROR_NOT_ENOUGH_MEMORY: + return true; + default: + return false; + } +} + +static void* win_virtual_alloc_prim(void* addr, size_t size, size_t try_alignment, DWORD flags) { + long max_retry_msecs = mi_option_get_clamp(mi_option_retry_on_oom, 0, 2000); // at most 2 seconds + if (max_retry_msecs == 1) { max_retry_msecs = 100; } // if one sets the option to "true" + for (long tries = 1; tries <= 10; tries++) { // try at most 10 times (=2200ms) + void* p = win_virtual_alloc_prim_once(addr, size, try_alignment, flags); + if (p != NULL) { + // success, return the address + return p; + } + else if (max_retry_msecs > 0 && (try_alignment <= 2*MI_SEGMENT_ALIGN) && + (flags&MEM_COMMIT) != 0 && (flags&MEM_LARGE_PAGES) == 0 && + win_is_out_of_memory_error(GetLastError())) { + // if committing regular memory and being out-of-memory, + // keep trying for a bit in case memory frees up after all. See issue #894 + _mi_warning_message("out-of-memory on OS allocation, try again... (attempt %lu, 0x%zx bytes, error code: 0x%x, address: %p, alignment: 0x%zx, flags: 0x%x)\n", tries, size, GetLastError(), addr, try_alignment, flags); + long sleep_msecs = tries*40; // increasing waits + if (sleep_msecs > max_retry_msecs) { sleep_msecs = max_retry_msecs; } + max_retry_msecs -= sleep_msecs; + Sleep(sleep_msecs); + } + else { + // otherwise return with an error + break; + } + } + return NULL; +} + +static void* win_virtual_alloc(void* addr, size_t size, size_t try_alignment, DWORD flags, bool large_only, bool allow_large, bool* is_large) { + mi_assert_internal(!(large_only && !allow_large)); + static _Atomic(size_t) large_page_try_ok; // = 0; + void* p = NULL; + // Try to allocate large OS pages (2MiB) if allowed or required. + if ((large_only || _mi_os_use_large_page(size, try_alignment)) + && allow_large && (flags&MEM_COMMIT)!=0 && (flags&MEM_RESERVE)!=0) { + size_t try_ok = mi_atomic_load_acquire(&large_page_try_ok); + if (!large_only && try_ok > 0) { + // if a large page allocation fails, it seems the calls to VirtualAlloc get very expensive. + // therefore, once a large page allocation failed, we don't try again for `large_page_try_ok` times. + mi_atomic_cas_strong_acq_rel(&large_page_try_ok, &try_ok, try_ok - 1); + } + else { + // large OS pages must always reserve and commit. + *is_large = true; + p = win_virtual_alloc_prim(addr, size, try_alignment, flags | MEM_LARGE_PAGES); + if (large_only) return p; + // fall back to non-large page allocation on error (`p == NULL`). + if (p == NULL) { + mi_atomic_store_release(&large_page_try_ok,10UL); // on error, don't try again for the next N allocations + } + } + } + // Fall back to regular page allocation + if (p == NULL) { + *is_large = ((flags&MEM_LARGE_PAGES) != 0); + p = win_virtual_alloc_prim(addr, size, try_alignment, flags); + } + //if (p == NULL) { _mi_warning_message("unable to allocate OS memory (%zu bytes, error code: 0x%x, address: %p, alignment: %zu, flags: 0x%x, large only: %d, allow large: %d)\n", size, GetLastError(), addr, try_alignment, flags, large_only, allow_large); } + return p; +} + +int _mi_prim_alloc(void* hint_addr, size_t size, size_t try_alignment, bool commit, bool allow_large, bool* is_large, bool* is_zero, void** addr) { + mi_assert_internal(size > 0 && (size % _mi_os_page_size()) == 0); + mi_assert_internal(commit || !allow_large); + mi_assert_internal(try_alignment > 0); + *is_zero = true; + int flags = MEM_RESERVE; + if (commit) { flags |= MEM_COMMIT; } + *addr = win_virtual_alloc(hint_addr, size, try_alignment, flags, false, allow_large, is_large); + return (*addr != NULL ? 0 : (int)GetLastError()); +} + + +//--------------------------------------------- +// Commit/Reset/Protect +//--------------------------------------------- +#ifdef _MSC_VER +#pragma warning(disable:6250) // suppress warning calling VirtualFree without MEM_RELEASE (for decommit) +#endif + +int _mi_prim_commit(void* addr, size_t size, bool* is_zero) { + *is_zero = false; + /* + // zero'ing only happens on an initial commit... but checking upfront seems expensive.. + _MEMORY_BASIC_INFORMATION meminfo; _mi_memzero_var(meminfo); + if (VirtualQuery(addr, &meminfo, size) > 0) { + if ((meminfo.State & MEM_COMMIT) == 0) { + *is_zero = true; + } + } + */ + // commit + void* p = VirtualAlloc(addr, size, MEM_COMMIT, PAGE_READWRITE); + if (p == NULL) return (int)GetLastError(); + return 0; +} + +int _mi_prim_decommit(void* addr, size_t size, bool* needs_recommit) { + BOOL ok = VirtualFree(addr, size, MEM_DECOMMIT); + *needs_recommit = true; // for safety, assume always decommitted even in the case of an error. + return (ok ? 0 : (int)GetLastError()); +} + +int _mi_prim_reset(void* addr, size_t size) { + void* p = VirtualAlloc(addr, size, MEM_RESET, PAGE_READWRITE); + mi_assert_internal(p == addr); + #if 0 + if (p != NULL) { + VirtualUnlock(addr,size); // VirtualUnlock after MEM_RESET removes the memory directly from the working set + } + #endif + return (p != NULL ? 0 : (int)GetLastError()); +} + +int _mi_prim_protect(void* addr, size_t size, bool protect) { + DWORD oldprotect = 0; + BOOL ok = VirtualProtect(addr, size, protect ? PAGE_NOACCESS : PAGE_READWRITE, &oldprotect); + return (ok ? 0 : (int)GetLastError()); +} + + +//--------------------------------------------- +// Huge page allocation +//--------------------------------------------- + +static void* _mi_prim_alloc_huge_os_pagesx(void* hint_addr, size_t size, int numa_node) +{ + const DWORD flags = MEM_LARGE_PAGES | MEM_COMMIT | MEM_RESERVE; + + win_enable_large_os_pages(NULL); + + MI_MEM_EXTENDED_PARAMETER params[3] = { {{0,0},{0}},{{0,0},{0}},{{0,0},{0}} }; + // on modern Windows try use NtAllocateVirtualMemoryEx for 1GiB huge pages + static bool mi_huge_pages_available = true; + if (pNtAllocateVirtualMemoryEx != NULL && mi_huge_pages_available) { + params[0].Type.Type = MiMemExtendedParameterAttributeFlags; + params[0].Arg.ULong64 = MI_MEM_EXTENDED_PARAMETER_NONPAGED_HUGE; + ULONG param_count = 1; + if (numa_node >= 0) { + param_count++; + params[1].Type.Type = MiMemExtendedParameterNumaNode; + params[1].Arg.ULong = (unsigned)numa_node; + } + SIZE_T psize = size; + void* base = hint_addr; + NTSTATUS err = (*pNtAllocateVirtualMemoryEx)(GetCurrentProcess(), &base, &psize, flags, PAGE_READWRITE, params, param_count); + if (err == 0 && base != NULL) { + return base; + } + else { + // fall back to regular large pages + mi_huge_pages_available = false; // don't try further huge pages + _mi_warning_message("unable to allocate using huge (1GiB) pages, trying large (2MiB) pages instead (status 0x%lx)\n", err); + } + } + // on modern Windows try use VirtualAlloc2 for numa aware large OS page allocation + if (pVirtualAlloc2 != NULL && numa_node >= 0) { + params[0].Type.Type = MiMemExtendedParameterNumaNode; + params[0].Arg.ULong = (unsigned)numa_node; + return (*pVirtualAlloc2)(GetCurrentProcess(), hint_addr, size, flags, PAGE_READWRITE, params, 1); + } + + // otherwise use regular virtual alloc on older windows + return VirtualAlloc(hint_addr, size, flags, PAGE_READWRITE); +} + +int _mi_prim_alloc_huge_os_pages(void* hint_addr, size_t size, int numa_node, bool* is_zero, void** addr) { + *is_zero = true; + *addr = _mi_prim_alloc_huge_os_pagesx(hint_addr,size,numa_node); + return (*addr != NULL ? 0 : (int)GetLastError()); +} + + +//--------------------------------------------- +// Numa nodes +//--------------------------------------------- + +size_t _mi_prim_numa_node(void) { + USHORT numa_node = 0; + if (pGetCurrentProcessorNumberEx != NULL && pGetNumaProcessorNodeEx != NULL) { + // Extended API is supported + MI_PROCESSOR_NUMBER pnum; + (*pGetCurrentProcessorNumberEx)(&pnum); + USHORT nnode = 0; + BOOL ok = (*pGetNumaProcessorNodeEx)(&pnum, &nnode); + if (ok) { numa_node = nnode; } + } + else if (pGetNumaProcessorNode != NULL) { + // Vista or earlier, use older API that is limited to 64 processors. Issue #277 + DWORD pnum = GetCurrentProcessorNumber(); + UCHAR nnode = 0; + BOOL ok = pGetNumaProcessorNode((UCHAR)pnum, &nnode); + if (ok) { numa_node = nnode; } + } + return numa_node; +} + +size_t _mi_prim_numa_node_count(void) { + ULONG numa_max = 0; + GetNumaHighestNodeNumber(&numa_max); + // find the highest node number that has actual processors assigned to it. Issue #282 + while(numa_max > 0) { + if (pGetNumaNodeProcessorMaskEx != NULL) { + // Extended API is supported + GROUP_AFFINITY affinity; + if ((*pGetNumaNodeProcessorMaskEx)((USHORT)numa_max, &affinity)) { + if (affinity.Mask != 0) break; // found the maximum non-empty node + } + } + else { + // Vista or earlier, use older API that is limited to 64 processors. + ULONGLONG mask; + if (GetNumaNodeProcessorMask((UCHAR)numa_max, &mask)) { + if (mask != 0) break; // found the maximum non-empty node + }; + } + // max node was invalid or had no processor assigned, try again + numa_max--; + } + return ((size_t)numa_max + 1); +} + + +//---------------------------------------------------------------- +// Clock +//---------------------------------------------------------------- + +static mi_msecs_t mi_to_msecs(LARGE_INTEGER t) { + static LARGE_INTEGER mfreq; // = 0 + if (mfreq.QuadPart == 0LL) { + LARGE_INTEGER f; + QueryPerformanceFrequency(&f); + mfreq.QuadPart = f.QuadPart/1000LL; + if (mfreq.QuadPart == 0) mfreq.QuadPart = 1; + } + return (mi_msecs_t)(t.QuadPart / mfreq.QuadPart); +} + +mi_msecs_t _mi_prim_clock_now(void) { + LARGE_INTEGER t; + QueryPerformanceCounter(&t); + return mi_to_msecs(t); +} + + +//---------------------------------------------------------------- +// Process Info +//---------------------------------------------------------------- + +#include + +static mi_msecs_t filetime_msecs(const FILETIME* ftime) { + ULARGE_INTEGER i; + i.LowPart = ftime->dwLowDateTime; + i.HighPart = ftime->dwHighDateTime; + mi_msecs_t msecs = (i.QuadPart / 10000); // FILETIME is in 100 nano seconds + return msecs; +} + +typedef BOOL (WINAPI *PGetProcessMemoryInfo)(HANDLE, PPROCESS_MEMORY_COUNTERS, DWORD); +static PGetProcessMemoryInfo pGetProcessMemoryInfo = NULL; + +void _mi_prim_process_info(mi_process_info_t* pinfo) +{ + FILETIME ct; + FILETIME ut; + FILETIME st; + FILETIME et; + GetProcessTimes(GetCurrentProcess(), &ct, &et, &st, &ut); + pinfo->utime = filetime_msecs(&ut); + pinfo->stime = filetime_msecs(&st); + + // load psapi on demand + if (pGetProcessMemoryInfo == NULL) { + HINSTANCE hDll = LoadLibrary(TEXT("psapi.dll")); + if (hDll != NULL) { + pGetProcessMemoryInfo = (PGetProcessMemoryInfo)(void (*)(void))GetProcAddress(hDll, "GetProcessMemoryInfo"); + } + } + + // get process info + PROCESS_MEMORY_COUNTERS info; _mi_memzero_var(info); + if (pGetProcessMemoryInfo != NULL) { + pGetProcessMemoryInfo(GetCurrentProcess(), &info, sizeof(info)); + } + pinfo->current_rss = (size_t)info.WorkingSetSize; + pinfo->peak_rss = (size_t)info.PeakWorkingSetSize; + pinfo->current_commit = (size_t)info.PagefileUsage; + pinfo->peak_commit = (size_t)info.PeakPagefileUsage; + pinfo->page_faults = (size_t)info.PageFaultCount; +} + +//---------------------------------------------------------------- +// Output +//---------------------------------------------------------------- + +void _mi_prim_out_stderr( const char* msg ) +{ + // on windows with redirection, the C runtime cannot handle locale dependent output + // after the main thread closes so we use direct console output. + if (!_mi_preloading()) { + // _cputs(msg); // _cputs cannot be used as it aborts when failing to lock the console + static HANDLE hcon = INVALID_HANDLE_VALUE; + static bool hconIsConsole; + if (hcon == INVALID_HANDLE_VALUE) { + CONSOLE_SCREEN_BUFFER_INFO sbi; + hcon = GetStdHandle(STD_ERROR_HANDLE); + hconIsConsole = ((hcon != INVALID_HANDLE_VALUE) && GetConsoleScreenBufferInfo(hcon, &sbi)); + } + const size_t len = _mi_strlen(msg); + if (len > 0 && len < UINT32_MAX) { + DWORD written = 0; + if (hconIsConsole) { + WriteConsoleA(hcon, msg, (DWORD)len, &written, NULL); + } + else if (hcon != INVALID_HANDLE_VALUE) { + // use direct write if stderr was redirected + WriteFile(hcon, msg, (DWORD)len, &written, NULL); + } + else { + // finally fall back to fputs after all + fputs(msg, stderr); + } + } + } +} + + +//---------------------------------------------------------------- +// Environment +//---------------------------------------------------------------- + +// On Windows use GetEnvironmentVariable instead of getenv to work +// reliably even when this is invoked before the C runtime is initialized. +// i.e. when `_mi_preloading() == true`. +// Note: on windows, environment names are not case sensitive. +bool _mi_prim_getenv(const char* name, char* result, size_t result_size) { + result[0] = 0; + size_t len = GetEnvironmentVariableA(name, result, (DWORD)result_size); + return (len > 0 && len < result_size); +} + + +//---------------------------------------------------------------- +// Random +//---------------------------------------------------------------- + +#if defined(MI_USE_RTLGENRANDOM) // || defined(__cplusplus) +// We prefer to use BCryptGenRandom instead of (the unofficial) RtlGenRandom but when using +// dynamic overriding, we observed it can raise an exception when compiled with C++, and +// sometimes deadlocks when also running under the VS debugger. +// In contrast, issue #623 implies that on Windows Server 2019 we need to use BCryptGenRandom. +// To be continued.. +#pragma comment (lib,"advapi32.lib") +#define RtlGenRandom SystemFunction036 +mi_decl_externc BOOLEAN NTAPI RtlGenRandom(PVOID RandomBuffer, ULONG RandomBufferLength); + +bool _mi_prim_random_buf(void* buf, size_t buf_len) { + return (RtlGenRandom(buf, (ULONG)buf_len) != 0); +} + +#else + +#ifndef BCRYPT_USE_SYSTEM_PREFERRED_RNG +#define BCRYPT_USE_SYSTEM_PREFERRED_RNG 0x00000002 +#endif + +typedef LONG (NTAPI *PBCryptGenRandom)(HANDLE, PUCHAR, ULONG, ULONG); +static PBCryptGenRandom pBCryptGenRandom = NULL; + +bool _mi_prim_random_buf(void* buf, size_t buf_len) { + if (pBCryptGenRandom == NULL) { + HINSTANCE hDll = LoadLibrary(TEXT("bcrypt.dll")); + if (hDll != NULL) { + pBCryptGenRandom = (PBCryptGenRandom)(void (*)(void))GetProcAddress(hDll, "BCryptGenRandom"); + } + if (pBCryptGenRandom == NULL) return false; + } + return (pBCryptGenRandom(NULL, (PUCHAR)buf, (ULONG)buf_len, BCRYPT_USE_SYSTEM_PREFERRED_RNG) >= 0); +} + +#endif // MI_USE_RTLGENRANDOM + + + +//---------------------------------------------------------------- +// Process & Thread Init/Done +//---------------------------------------------------------------- + +static void NTAPI mi_win_main(PVOID module, DWORD reason, LPVOID reserved) { + MI_UNUSED(reserved); + MI_UNUSED(module); + #if MI_TLS_SLOT >= 2 + if ((reason==DLL_PROCESS_ATTACH || reason==DLL_THREAD_ATTACH) && mi_prim_get_default_heap() == NULL) { + _mi_heap_set_default_direct((mi_heap_t*)&_mi_heap_empty); + } + #endif + if (reason==DLL_PROCESS_ATTACH) { + _mi_process_load(); + } + else if (reason==DLL_PROCESS_DETACH) { + _mi_process_done(); + } + else if (reason==DLL_THREAD_DETACH && !_mi_is_redirected()) { + _mi_thread_done(NULL); + } +} + + +#if defined(MI_SHARED_LIB) + #define MI_PRIM_HAS_PROCESS_ATTACH 1 + + // Windows DLL: easy to hook into process_init and thread_done + BOOL WINAPI DllMain(HINSTANCE inst, DWORD reason, LPVOID reserved) { + mi_win_main((PVOID)inst,reason,reserved); + return TRUE; + } + + // nothing to do since `_mi_thread_done` is handled through the DLL_THREAD_DETACH event. + void _mi_prim_thread_init_auto_done(void) { } + void _mi_prim_thread_done_auto_done(void) { } + void _mi_prim_thread_associate_default_heap(mi_heap_t* heap) { + MI_UNUSED(heap); + } + +#elif !defined(MI_WIN_USE_FLS) + #define MI_PRIM_HAS_PROCESS_ATTACH 1 + + static void NTAPI mi_win_main_attach(PVOID module, DWORD reason, LPVOID reserved) { + if (reason == DLL_PROCESS_ATTACH || reason == DLL_THREAD_ATTACH) { + mi_win_main(module, reason, reserved); + } + } + static void NTAPI mi_win_main_detach(PVOID module, DWORD reason, LPVOID reserved) { + if (reason == DLL_PROCESS_DETACH || reason == DLL_THREAD_DETACH) { + mi_win_main(module, reason, reserved); + } + } + + // Set up TLS callbacks in a statically linked library by using special data sections. + // See + // We use 2 entries to ensure we call attach events before constructors + // are called, and detach events after destructors are called. + #if defined(__cplusplus) + extern "C" { + #endif + + #if defined(_WIN64) + #pragma comment(linker, "/INCLUDE:_tls_used") + #pragma comment(linker, "/INCLUDE:_mi_tls_callback_pre") + #pragma comment(linker, "/INCLUDE:_mi_tls_callback_post") + #pragma const_seg(".CRT$XLB") + extern const PIMAGE_TLS_CALLBACK _mi_tls_callback_pre[]; + const PIMAGE_TLS_CALLBACK _mi_tls_callback_pre[] = { &mi_win_main_attach }; + #pragma const_seg() + #pragma const_seg(".CRT$XLY") + extern const PIMAGE_TLS_CALLBACK _mi_tls_callback_post[]; + const PIMAGE_TLS_CALLBACK _mi_tls_callback_post[] = { &mi_win_main_detach }; + #pragma const_seg() + #else + #pragma comment(linker, "/INCLUDE:__tls_used") + #pragma comment(linker, "/INCLUDE:__mi_tls_callback_pre") + #pragma comment(linker, "/INCLUDE:__mi_tls_callback_post") + #pragma data_seg(".CRT$XLB") + PIMAGE_TLS_CALLBACK _mi_tls_callback_pre[] = { &mi_win_main_attach }; + #pragma data_seg() + #pragma data_seg(".CRT$XLY") + PIMAGE_TLS_CALLBACK _mi_tls_callback_post[] = { &mi_win_main_detach }; + #pragma data_seg() + #endif + + #if defined(__cplusplus) + } + #endif + + // nothing to do since `_mi_thread_done` is handled through the DLL_THREAD_DETACH event. + void _mi_prim_thread_init_auto_done(void) { } + void _mi_prim_thread_done_auto_done(void) { } + void _mi_prim_thread_associate_default_heap(mi_heap_t* heap) { + MI_UNUSED(heap); + } + +#else // deprecated: statically linked, use fiber api + + #if defined(_MSC_VER) // on clang/gcc use the constructor attribute (in `src/prim/prim.c`) + // MSVC: use data section magic for static libraries + // See + #define MI_PRIM_HAS_PROCESS_ATTACH 1 + + static int mi_process_attach(void) { + mi_win_main(NULL,DLL_PROCESS_ATTACH,NULL); + atexit(&_mi_process_done); + return 0; + } + typedef int(*mi_crt_callback_t)(void); + #if defined(_WIN64) + #pragma comment(linker, "/INCLUDE:_mi_tls_callback") + #pragma section(".CRT$XIU", long, read) + #else + #pragma comment(linker, "/INCLUDE:__mi_tls_callback") + #endif + #pragma data_seg(".CRT$XIU") + mi_decl_externc mi_crt_callback_t _mi_tls_callback[] = { &mi_process_attach }; + #pragma data_seg() + #endif + + // use the fiber api for calling `_mi_thread_done`. + #include + #if (_WIN32_WINNT < 0x600) // before Windows Vista + WINBASEAPI DWORD WINAPI FlsAlloc( _In_opt_ PFLS_CALLBACK_FUNCTION lpCallback ); + WINBASEAPI PVOID WINAPI FlsGetValue( _In_ DWORD dwFlsIndex ); + WINBASEAPI BOOL WINAPI FlsSetValue( _In_ DWORD dwFlsIndex, _In_opt_ PVOID lpFlsData ); + WINBASEAPI BOOL WINAPI FlsFree(_In_ DWORD dwFlsIndex); + #endif + + static DWORD mi_fls_key = (DWORD)(-1); + + static void NTAPI mi_fls_done(PVOID value) { + mi_heap_t* heap = (mi_heap_t*)value; + if (heap != NULL) { + _mi_thread_done(heap); + FlsSetValue(mi_fls_key, NULL); // prevent recursion as _mi_thread_done may set it back to the main heap, issue #672 + } + } + + void _mi_prim_thread_init_auto_done(void) { + mi_fls_key = FlsAlloc(&mi_fls_done); + } + + void _mi_prim_thread_done_auto_done(void) { + // call thread-done on all threads (except the main thread) to prevent + // dangling callback pointer if statically linked with a DLL; Issue #208 + FlsFree(mi_fls_key); + } + + void _mi_prim_thread_associate_default_heap(mi_heap_t* heap) { + mi_assert_internal(mi_fls_key != (DWORD)(-1)); + FlsSetValue(mi_fls_key, heap); + } +#endif + +// ---------------------------------------------------- +// Communicate with the redirection module on Windows +// ---------------------------------------------------- +#if defined(MI_SHARED_LIB) && !defined(MI_WIN_NOREDIRECT) + #define MI_PRIM_HAS_ALLOCATOR_INIT 1 + + static bool mi_redirected = false; // true if malloc redirects to mi_malloc + + bool _mi_is_redirected(void) { + return mi_redirected; + } + + #ifdef __cplusplus + extern "C" { + #endif + mi_decl_export void _mi_redirect_entry(DWORD reason) { + // called on redirection; careful as this may be called before DllMain + #if MI_TLS_SLOT >= 2 + if ((reason==DLL_PROCESS_ATTACH || reason==DLL_THREAD_ATTACH) && mi_prim_get_default_heap() == NULL) { + _mi_heap_set_default_direct((mi_heap_t*)&_mi_heap_empty); + } + #endif + if (reason == DLL_PROCESS_ATTACH) { + mi_redirected = true; + } + else if (reason == DLL_PROCESS_DETACH) { + mi_redirected = false; + } + else if (reason == DLL_THREAD_DETACH) { + _mi_thread_done(NULL); + } + } + __declspec(dllimport) bool mi_cdecl mi_allocator_init(const char** message); + __declspec(dllimport) void mi_cdecl mi_allocator_done(void); + #ifdef __cplusplus + } + #endif + bool _mi_allocator_init(const char** message) { + return mi_allocator_init(message); + } + void _mi_allocator_done(void) { + mi_allocator_done(); + } +#endif diff --git a/compat/mimalloc/random.c b/compat/mimalloc/random.c new file mode 100644 index 00000000000000..f17698ba8a6d08 --- /dev/null +++ b/compat/mimalloc/random.c @@ -0,0 +1,258 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2019-2021, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ +#include "mimalloc.h" +#include "mimalloc/internal.h" +#include "mimalloc/prim.h" // _mi_prim_random_buf +#include // memset + +/* ---------------------------------------------------------------------------- +We use our own PRNG to keep predictable performance of random number generation +and to avoid implementations that use a lock. We only use the OS provided +random source to initialize the initial seeds. Since we do not need ultimate +performance but we do rely on the security (for secret cookies in secure mode) +we use a cryptographically secure generator (chacha20). +-----------------------------------------------------------------------------*/ + +#define MI_CHACHA_ROUNDS (20) // perhaps use 12 for better performance? + + +/* ---------------------------------------------------------------------------- +Chacha20 implementation as the original algorithm with a 64-bit nonce +and counter: https://en.wikipedia.org/wiki/Salsa20 +The input matrix has sixteen 32-bit values: +Position 0 to 3: constant key +Position 4 to 11: the key +Position 12 to 13: the counter. +Position 14 to 15: the nonce. + +The implementation uses regular C code which compiles very well on modern compilers. +(gcc x64 has no register spills, and clang 6+ uses SSE instructions) +-----------------------------------------------------------------------------*/ + +static inline uint32_t rotl(uint32_t x, uint32_t shift) { + return (x << shift) | (x >> (32 - shift)); +} + +static inline void qround(uint32_t x[16], size_t a, size_t b, size_t c, size_t d) { + x[a] += x[b]; x[d] = rotl(x[d] ^ x[a], 16); + x[c] += x[d]; x[b] = rotl(x[b] ^ x[c], 12); + x[a] += x[b]; x[d] = rotl(x[d] ^ x[a], 8); + x[c] += x[d]; x[b] = rotl(x[b] ^ x[c], 7); +} + +static void chacha_block(mi_random_ctx_t* ctx) +{ + // scramble into `x` + uint32_t x[16]; + for (size_t i = 0; i < 16; i++) { + x[i] = ctx->input[i]; + } + for (size_t i = 0; i < MI_CHACHA_ROUNDS; i += 2) { + qround(x, 0, 4, 8, 12); + qround(x, 1, 5, 9, 13); + qround(x, 2, 6, 10, 14); + qround(x, 3, 7, 11, 15); + qround(x, 0, 5, 10, 15); + qround(x, 1, 6, 11, 12); + qround(x, 2, 7, 8, 13); + qround(x, 3, 4, 9, 14); + } + + // add scrambled data to the initial state + for (size_t i = 0; i < 16; i++) { + ctx->output[i] = x[i] + ctx->input[i]; + } + ctx->output_available = 16; + + // increment the counter for the next round + ctx->input[12] += 1; + if (ctx->input[12] == 0) { + ctx->input[13] += 1; + if (ctx->input[13] == 0) { // and keep increasing into the nonce + ctx->input[14] += 1; + } + } +} + +static uint32_t chacha_next32(mi_random_ctx_t* ctx) { + if (ctx->output_available <= 0) { + chacha_block(ctx); + ctx->output_available = 16; // (assign again to suppress static analysis warning) + } + const uint32_t x = ctx->output[16 - ctx->output_available]; + ctx->output[16 - ctx->output_available] = 0; // reset once the data is handed out + ctx->output_available--; + return x; +} + +static inline uint32_t read32(const uint8_t* p, size_t idx32) { + const size_t i = 4*idx32; + return ((uint32_t)p[i+0] | (uint32_t)p[i+1] << 8 | (uint32_t)p[i+2] << 16 | (uint32_t)p[i+3] << 24); +} + +static void chacha_init(mi_random_ctx_t* ctx, const uint8_t key[32], uint64_t nonce) +{ + // since we only use chacha for randomness (and not encryption) we + // do not _need_ to read 32-bit values as little endian but we do anyways + // just for being compatible :-) + memset(ctx, 0, sizeof(*ctx)); + for (size_t i = 0; i < 4; i++) { + const uint8_t* sigma = (uint8_t*)"expand 32-byte k"; + ctx->input[i] = read32(sigma,i); + } + for (size_t i = 0; i < 8; i++) { + ctx->input[i + 4] = read32(key,i); + } + ctx->input[12] = 0; + ctx->input[13] = 0; + ctx->input[14] = (uint32_t)nonce; + ctx->input[15] = (uint32_t)(nonce >> 32); +} + +static void chacha_split(mi_random_ctx_t* ctx, uint64_t nonce, mi_random_ctx_t* ctx_new) { + memset(ctx_new, 0, sizeof(*ctx_new)); + _mi_memcpy(ctx_new->input, ctx->input, sizeof(ctx_new->input)); + ctx_new->input[12] = 0; + ctx_new->input[13] = 0; + ctx_new->input[14] = (uint32_t)nonce; + ctx_new->input[15] = (uint32_t)(nonce >> 32); + mi_assert_internal(ctx->input[14] != ctx_new->input[14] || ctx->input[15] != ctx_new->input[15]); // do not reuse nonces! + chacha_block(ctx_new); +} + + +/* ---------------------------------------------------------------------------- +Random interface +-----------------------------------------------------------------------------*/ + +#if MI_DEBUG>1 +static bool mi_random_is_initialized(mi_random_ctx_t* ctx) { + return (ctx != NULL && ctx->input[0] != 0); +} +#endif + +void _mi_random_split(mi_random_ctx_t* ctx, mi_random_ctx_t* ctx_new) { + mi_assert_internal(mi_random_is_initialized(ctx)); + mi_assert_internal(ctx != ctx_new); + chacha_split(ctx, (uintptr_t)ctx_new /*nonce*/, ctx_new); +} + +uintptr_t _mi_random_next(mi_random_ctx_t* ctx) { + mi_assert_internal(mi_random_is_initialized(ctx)); + uintptr_t r; + do { + #if MI_INTPTR_SIZE <= 4 + r = chacha_next32(ctx); + #elif MI_INTPTR_SIZE == 8 + r = (((uintptr_t)chacha_next32(ctx) << 32) | chacha_next32(ctx)); + #else + # error "define mi_random_next for this platform" + #endif + } while (r==0); + return r; +} + + +/* ---------------------------------------------------------------------------- +To initialize a fresh random context. +If we cannot get good randomness, we fall back to weak randomness based on a timer and ASLR. +-----------------------------------------------------------------------------*/ + +uintptr_t _mi_os_random_weak(uintptr_t extra_seed) { + uintptr_t x = (uintptr_t)&_mi_os_random_weak ^ extra_seed; // ASLR makes the address random + x ^= _mi_prim_clock_now(); + // and do a few randomization steps + uintptr_t max = ((x ^ (x >> 17)) & 0x0F) + 1; + for (uintptr_t i = 0; i < max || x==0; i++, x++) { + x = _mi_random_shuffle(x); + } + mi_assert_internal(x != 0); + return x; +} + +static void mi_random_init_ex(mi_random_ctx_t* ctx, bool use_weak) { + uint8_t key[32]; + if (use_weak || !_mi_prim_random_buf(key, sizeof(key))) { + // if we fail to get random data from the OS, we fall back to a + // weak random source based on the current time + #if !defined(__wasi__) + if (!use_weak) { _mi_warning_message("unable to use secure randomness\n"); } + #endif + uintptr_t x = _mi_os_random_weak(0); + for (size_t i = 0; i < 8; i++, x++) { // key is eight 32-bit words. + x = _mi_random_shuffle(x); + ((uint32_t*)key)[i] = (uint32_t)x; + } + ctx->weak = true; + } + else { + ctx->weak = false; + } + chacha_init(ctx, key, (uintptr_t)ctx /*nonce*/ ); +} + +void _mi_random_init(mi_random_ctx_t* ctx) { + mi_random_init_ex(ctx, false); +} + +void _mi_random_init_weak(mi_random_ctx_t * ctx) { + mi_random_init_ex(ctx, true); +} + +void _mi_random_reinit_if_weak(mi_random_ctx_t * ctx) { + if (ctx->weak) { + _mi_random_init(ctx); + } +} + +/* -------------------------------------------------------- +test vectors from +----------------------------------------------------------- */ +/* +static bool array_equals(uint32_t* x, uint32_t* y, size_t n) { + for (size_t i = 0; i < n; i++) { + if (x[i] != y[i]) return false; + } + return true; +} +static void chacha_test(void) +{ + uint32_t x[4] = { 0x11111111, 0x01020304, 0x9b8d6f43, 0x01234567 }; + uint32_t x_out[4] = { 0xea2a92f4, 0xcb1cf8ce, 0x4581472e, 0x5881c4bb }; + qround(x, 0, 1, 2, 3); + mi_assert_internal(array_equals(x, x_out, 4)); + + uint32_t y[16] = { + 0x879531e0, 0xc5ecf37d, 0x516461b1, 0xc9a62f8a, + 0x44c20ef3, 0x3390af7f, 0xd9fc690b, 0x2a5f714c, + 0x53372767, 0xb00a5631, 0x974c541a, 0x359e9963, + 0x5c971061, 0x3d631689, 0x2098d9d6, 0x91dbd320 }; + uint32_t y_out[16] = { + 0x879531e0, 0xc5ecf37d, 0xbdb886dc, 0xc9a62f8a, + 0x44c20ef3, 0x3390af7f, 0xd9fc690b, 0xcfacafd2, + 0xe46bea80, 0xb00a5631, 0x974c541a, 0x359e9963, + 0x5c971061, 0xccc07c79, 0x2098d9d6, 0x91dbd320 }; + qround(y, 2, 7, 8, 13); + mi_assert_internal(array_equals(y, y_out, 16)); + + mi_random_ctx_t r = { + { 0x61707865, 0x3320646e, 0x79622d32, 0x6b206574, + 0x03020100, 0x07060504, 0x0b0a0908, 0x0f0e0d0c, + 0x13121110, 0x17161514, 0x1b1a1918, 0x1f1e1d1c, + 0x00000001, 0x09000000, 0x4a000000, 0x00000000 }, + {0}, + 0 + }; + uint32_t r_out[16] = { + 0xe4e7f110, 0x15593bd1, 0x1fdd0f50, 0xc47120a3, + 0xc7f4d1c7, 0x0368c033, 0x9aaa2204, 0x4e6cd4c3, + 0x466482d2, 0x09aa9f07, 0x05d7c214, 0xa2028bd9, + 0xd19c12b5, 0xb94e16de, 0xe883d0cb, 0x4e3c50a2 }; + chacha_block(&r); + mi_assert_internal(array_equals(r.output, r_out, 16)); +} +*/ diff --git a/compat/mimalloc/segment-map.c b/compat/mimalloc/segment-map.c new file mode 100644 index 00000000000000..2f68f8c411e1b7 --- /dev/null +++ b/compat/mimalloc/segment-map.c @@ -0,0 +1,142 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2019-2023, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ + +/* ----------------------------------------------------------- + The following functions are to reliably find the segment or + block that encompasses any pointer p (or NULL if it is not + in any of our segments). + We maintain a bitmap of all memory with 1 bit per MI_SEGMENT_SIZE (64MiB) + set to 1 if it contains the segment meta data. +----------------------------------------------------------- */ +#include "mimalloc.h" +#include "mimalloc/internal.h" +#include "mimalloc/atomic.h" + +// Reduce total address space to reduce .bss (due to the `mi_segment_map`) +#if (MI_INTPTR_SIZE > 4) && MI_TRACK_ASAN +#define MI_SEGMENT_MAP_MAX_ADDRESS (128*1024ULL*MI_GiB) // 128 TiB (see issue #881) +#elif (MI_INTPTR_SIZE > 4) +#define MI_SEGMENT_MAP_MAX_ADDRESS (48*1024ULL*MI_GiB) // 48 TiB +#else +#define MI_SEGMENT_MAP_MAX_ADDRESS (UINT32_MAX) +#endif + +#define MI_SEGMENT_MAP_PART_SIZE (MI_INTPTR_SIZE*MI_KiB - 128) // 128 > sizeof(mi_memid_t) ! +#define MI_SEGMENT_MAP_PART_BITS (8*MI_SEGMENT_MAP_PART_SIZE) +#define MI_SEGMENT_MAP_PART_ENTRIES (MI_SEGMENT_MAP_PART_SIZE / MI_INTPTR_SIZE) +#define MI_SEGMENT_MAP_PART_BIT_SPAN (MI_SEGMENT_ALIGN) // memory area covered by 1 bit + +#if (MI_SEGMENT_MAP_PART_BITS < (MI_SEGMENT_MAP_MAX_ADDRESS / MI_SEGMENT_MAP_PART_BIT_SPAN)) // prevent overflow on 32-bit (issue #1017) +#define MI_SEGMENT_MAP_PART_SPAN (MI_SEGMENT_MAP_PART_BITS * MI_SEGMENT_MAP_PART_BIT_SPAN) +#else +#define MI_SEGMENT_MAP_PART_SPAN MI_SEGMENT_MAP_MAX_ADDRESS +#endif + +#define MI_SEGMENT_MAP_MAX_PARTS ((MI_SEGMENT_MAP_MAX_ADDRESS / MI_SEGMENT_MAP_PART_SPAN) + 1) + +// A part of the segment map. +typedef struct mi_segmap_part_s { + mi_memid_t memid; + _Atomic(uintptr_t) map[MI_SEGMENT_MAP_PART_ENTRIES]; +} mi_segmap_part_t; + +// Allocate parts on-demand to reduce .bss footprint +static _Atomic(mi_segmap_part_t*) mi_segment_map[MI_SEGMENT_MAP_MAX_PARTS]; // = { NULL, .. } + +static mi_segmap_part_t* mi_segment_map_index_of(const mi_segment_t* segment, bool create_on_demand, size_t* idx, size_t* bitidx) { + // note: segment can be invalid or NULL. + mi_assert_internal(_mi_ptr_segment(segment + 1) == segment); // is it aligned on MI_SEGMENT_SIZE? + *idx = 0; + *bitidx = 0; + if ((uintptr_t)segment >= MI_SEGMENT_MAP_MAX_ADDRESS) return NULL; + const uintptr_t segindex = ((uintptr_t)segment) / MI_SEGMENT_MAP_PART_SPAN; + if (segindex >= MI_SEGMENT_MAP_MAX_PARTS) return NULL; + mi_segmap_part_t* part = mi_atomic_load_ptr_relaxed(mi_segmap_part_t, &mi_segment_map[segindex]); + + // allocate on demand to reduce .bss footprint + if mi_unlikely(part == NULL) { + if (!create_on_demand) return NULL; + mi_memid_t memid; + part = (mi_segmap_part_t*)_mi_os_alloc(sizeof(mi_segmap_part_t), &memid); + if (part == NULL) return NULL; + part->memid = memid; + mi_segmap_part_t* expected = NULL; + if (!mi_atomic_cas_ptr_strong_release(mi_segmap_part_t, &mi_segment_map[segindex], &expected, part)) { + _mi_os_free(part, sizeof(mi_segmap_part_t), memid); + part = expected; + if (part == NULL) return NULL; + } + } + mi_assert(part != NULL); + const uintptr_t offset = ((uintptr_t)segment) % MI_SEGMENT_MAP_PART_SPAN; + const uintptr_t bitofs = offset / MI_SEGMENT_MAP_PART_BIT_SPAN; + *idx = bitofs / MI_INTPTR_BITS; + *bitidx = bitofs % MI_INTPTR_BITS; + return part; +} + +void _mi_segment_map_allocated_at(const mi_segment_t* segment) { + if (segment->memid.memkind == MI_MEM_ARENA) return; // we lookup segments first in the arena's and don't need the segment map + size_t index; + size_t bitidx; + mi_segmap_part_t* part = mi_segment_map_index_of(segment, true /* alloc map if needed */, &index, &bitidx); + if (part == NULL) return; // outside our address range.. + uintptr_t mask = mi_atomic_load_relaxed(&part->map[index]); + uintptr_t newmask; + do { + newmask = (mask | ((uintptr_t)1 << bitidx)); + } while (!mi_atomic_cas_weak_release(&part->map[index], &mask, newmask)); +} + +void _mi_segment_map_freed_at(const mi_segment_t* segment) { + if (segment->memid.memkind == MI_MEM_ARENA) return; + size_t index; + size_t bitidx; + mi_segmap_part_t* part = mi_segment_map_index_of(segment, false /* don't alloc if not present */, &index, &bitidx); + if (part == NULL) return; // outside our address range.. + uintptr_t mask = mi_atomic_load_relaxed(&part->map[index]); + uintptr_t newmask; + do { + newmask = (mask & ~((uintptr_t)1 << bitidx)); + } while (!mi_atomic_cas_weak_release(&part->map[index], &mask, newmask)); +} + +// Determine the segment belonging to a pointer or NULL if it is not in a valid segment. +static mi_segment_t* _mi_segment_of(const void* p) { + if (p == NULL) return NULL; + mi_segment_t* segment = _mi_ptr_segment(p); // segment can be NULL + size_t index; + size_t bitidx; + mi_segmap_part_t* part = mi_segment_map_index_of(segment, false /* dont alloc if not present */, &index, &bitidx); + if (part == NULL) return NULL; + const uintptr_t mask = mi_atomic_load_relaxed(&part->map[index]); + if mi_likely((mask & ((uintptr_t)1 << bitidx)) != 0) { + bool cookie_ok = (_mi_ptr_cookie(segment) == segment->cookie); + mi_assert_internal(cookie_ok); MI_UNUSED(cookie_ok); + return segment; // yes, allocated by us + } + return NULL; +} + +// Is this a valid pointer in our heap? +static bool mi_is_valid_pointer(const void* p) { + // first check if it is in an arena, then check if it is OS allocated + return (_mi_arena_contains(p) || _mi_segment_of(p) != NULL); +} + +mi_decl_nodiscard mi_decl_export bool mi_is_in_heap_region(const void* p) mi_attr_noexcept { + return mi_is_valid_pointer(p); +} + +void _mi_segment_map_unsafe_destroy(void) { + for (size_t i = 0; i < MI_SEGMENT_MAP_MAX_PARTS; i++) { + mi_segmap_part_t* part = mi_atomic_exchange_ptr_relaxed(mi_segmap_part_t, &mi_segment_map[i], NULL); + if (part != NULL) { + _mi_os_free(part, sizeof(mi_segmap_part_t), part->memid); + } + } +} diff --git a/compat/mimalloc/segment.c b/compat/mimalloc/segment.c new file mode 100644 index 00000000000000..29502bcd68cc73 --- /dev/null +++ b/compat/mimalloc/segment.c @@ -0,0 +1,1701 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2024, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ +#include "mimalloc.h" +#include "mimalloc/internal.h" +#include "mimalloc/atomic.h" + +#include // memset +#include + +// ------------------------------------------------------------------- +// Segments +// mimalloc pages reside in segments. See `mi_segment_valid` for invariants. +// ------------------------------------------------------------------- + + +static void mi_segment_try_purge(mi_segment_t* segment, bool force); + + +// ------------------------------------------------------------------- +// commit mask +// ------------------------------------------------------------------- + +static bool mi_commit_mask_all_set(const mi_commit_mask_t* commit, const mi_commit_mask_t* cm) { + for (size_t i = 0; i < MI_COMMIT_MASK_FIELD_COUNT; i++) { + if ((commit->mask[i] & cm->mask[i]) != cm->mask[i]) return false; + } + return true; +} + +static bool mi_commit_mask_any_set(const mi_commit_mask_t* commit, const mi_commit_mask_t* cm) { + for (size_t i = 0; i < MI_COMMIT_MASK_FIELD_COUNT; i++) { + if ((commit->mask[i] & cm->mask[i]) != 0) return true; + } + return false; +} + +static void mi_commit_mask_create_intersect(const mi_commit_mask_t* commit, const mi_commit_mask_t* cm, mi_commit_mask_t* res) { + for (size_t i = 0; i < MI_COMMIT_MASK_FIELD_COUNT; i++) { + res->mask[i] = (commit->mask[i] & cm->mask[i]); + } +} + +static void mi_commit_mask_clear(mi_commit_mask_t* res, const mi_commit_mask_t* cm) { + for (size_t i = 0; i < MI_COMMIT_MASK_FIELD_COUNT; i++) { + res->mask[i] &= ~(cm->mask[i]); + } +} + +static void mi_commit_mask_set(mi_commit_mask_t* res, const mi_commit_mask_t* cm) { + for (size_t i = 0; i < MI_COMMIT_MASK_FIELD_COUNT; i++) { + res->mask[i] |= cm->mask[i]; + } +} + +static void mi_commit_mask_create(size_t bitidx, size_t bitcount, mi_commit_mask_t* cm) { + mi_assert_internal(bitidx < MI_COMMIT_MASK_BITS); + mi_assert_internal((bitidx + bitcount) <= MI_COMMIT_MASK_BITS); + if (bitcount == MI_COMMIT_MASK_BITS) { + mi_assert_internal(bitidx==0); + mi_commit_mask_create_full(cm); + } + else if (bitcount == 0) { + mi_commit_mask_create_empty(cm); + } + else { + mi_commit_mask_create_empty(cm); + size_t i = bitidx / MI_COMMIT_MASK_FIELD_BITS; + size_t ofs = bitidx % MI_COMMIT_MASK_FIELD_BITS; + while (bitcount > 0) { + mi_assert_internal(i < MI_COMMIT_MASK_FIELD_COUNT); + size_t avail = MI_COMMIT_MASK_FIELD_BITS - ofs; + size_t count = (bitcount > avail ? avail : bitcount); + size_t mask = (count >= MI_COMMIT_MASK_FIELD_BITS ? ~((size_t)0) : (((size_t)1 << count) - 1) << ofs); + cm->mask[i] = mask; + bitcount -= count; + ofs = 0; + i++; + } + } +} + +size_t _mi_commit_mask_committed_size(const mi_commit_mask_t* cm, size_t total) { + mi_assert_internal((total%MI_COMMIT_MASK_BITS)==0); + size_t count = 0; + for (size_t i = 0; i < MI_COMMIT_MASK_FIELD_COUNT; i++) { + size_t mask = cm->mask[i]; + if (~mask == 0) { + count += MI_COMMIT_MASK_FIELD_BITS; + } + else { + for (; mask != 0; mask >>= 1) { // todo: use popcount + if ((mask&1)!=0) count++; + } + } + } + // we use total since for huge segments each commit bit may represent a larger size + return ((total / MI_COMMIT_MASK_BITS) * count); +} + + +size_t _mi_commit_mask_next_run(const mi_commit_mask_t* cm, size_t* idx) { + size_t i = (*idx) / MI_COMMIT_MASK_FIELD_BITS; + size_t ofs = (*idx) % MI_COMMIT_MASK_FIELD_BITS; + size_t mask = 0; + // find first ones + while (i < MI_COMMIT_MASK_FIELD_COUNT) { + mask = cm->mask[i]; + mask >>= ofs; + if (mask != 0) { + while ((mask&1) == 0) { + mask >>= 1; + ofs++; + } + break; + } + i++; + ofs = 0; + } + if (i >= MI_COMMIT_MASK_FIELD_COUNT) { + // not found + *idx = MI_COMMIT_MASK_BITS; + return 0; + } + else { + // found, count ones + size_t count = 0; + *idx = (i*MI_COMMIT_MASK_FIELD_BITS) + ofs; + do { + mi_assert_internal(ofs < MI_COMMIT_MASK_FIELD_BITS && (mask&1) == 1); + do { + count++; + mask >>= 1; + } while ((mask&1) == 1); + if ((((*idx + count) % MI_COMMIT_MASK_FIELD_BITS) == 0)) { + i++; + if (i >= MI_COMMIT_MASK_FIELD_COUNT) break; + mask = cm->mask[i]; + ofs = 0; + } + } while ((mask&1) == 1); + mi_assert_internal(count > 0); + return count; + } +} + + +/* -------------------------------------------------------------------------------- + Segment allocation + We allocate pages inside bigger "segments" (32 MiB on 64-bit). This is to avoid + splitting VMA's on Linux and reduce fragmentation on other OS's. + Each thread owns its own segments. + + Currently we have: + - small pages (64KiB) + - medium pages (512KiB) + - large pages (4MiB), + - huge segments have 1 page in one segment that can be larger than `MI_SEGMENT_SIZE`. + it is used for blocks `> MI_LARGE_OBJ_SIZE_MAX` or with alignment `> MI_BLOCK_ALIGNMENT_MAX`. + + The memory for a segment is usually committed on demand. + (i.e. we are careful to not touch the memory until we actually allocate a block there) + + If a thread ends, it "abandons" pages that still contain live blocks. + Such segments are abandoned and these can be reclaimed by still running threads, + (much like work-stealing). +-------------------------------------------------------------------------------- */ + + +/* ----------------------------------------------------------- + Slices +----------------------------------------------------------- */ + + +static const mi_slice_t* mi_segment_slices_end(const mi_segment_t* segment) { + return &segment->slices[segment->slice_entries]; +} + +static uint8_t* mi_slice_start(const mi_slice_t* slice) { + mi_segment_t* segment = _mi_ptr_segment(slice); + mi_assert_internal(slice >= segment->slices && slice < mi_segment_slices_end(segment)); + return ((uint8_t*)segment + ((slice - segment->slices)*MI_SEGMENT_SLICE_SIZE)); +} + + +/* ----------------------------------------------------------- + Bins +----------------------------------------------------------- */ +// Use bit scan forward to quickly find the first zero bit if it is available + +static inline size_t mi_slice_bin8(size_t slice_count) { + if (slice_count<=1) return slice_count; + mi_assert_internal(slice_count <= MI_SLICES_PER_SEGMENT); + slice_count--; + size_t s = mi_bsr(slice_count); // slice_count > 1 + if (s <= 2) return slice_count + 1; + size_t bin = ((s << 2) | ((slice_count >> (s - 2))&0x03)) - 4; + return bin; +} + +static inline size_t mi_slice_bin(size_t slice_count) { + mi_assert_internal(slice_count*MI_SEGMENT_SLICE_SIZE <= MI_SEGMENT_SIZE); + mi_assert_internal(mi_slice_bin8(MI_SLICES_PER_SEGMENT) <= MI_SEGMENT_BIN_MAX); + size_t bin = mi_slice_bin8(slice_count); + mi_assert_internal(bin <= MI_SEGMENT_BIN_MAX); + return bin; +} + +static inline size_t mi_slice_index(const mi_slice_t* slice) { + mi_segment_t* segment = _mi_ptr_segment(slice); + ptrdiff_t index = slice - segment->slices; + mi_assert_internal(index >= 0 && index < (ptrdiff_t)segment->slice_entries); + return index; +} + + +/* ----------------------------------------------------------- + Slice span queues +----------------------------------------------------------- */ + +static void mi_span_queue_push(mi_span_queue_t* sq, mi_slice_t* slice) { + // todo: or push to the end? + mi_assert_internal(slice->prev == NULL && slice->next==NULL); + slice->prev = NULL; // paranoia + slice->next = sq->first; + sq->first = slice; + if (slice->next != NULL) slice->next->prev = slice; + else sq->last = slice; + slice->block_size = 0; // free +} + +static mi_span_queue_t* mi_span_queue_for(size_t slice_count, mi_segments_tld_t* tld) { + size_t bin = mi_slice_bin(slice_count); + mi_span_queue_t* sq = &tld->spans[bin]; + mi_assert_internal(sq->slice_count >= slice_count); + return sq; +} + +static void mi_span_queue_delete(mi_span_queue_t* sq, mi_slice_t* slice) { + mi_assert_internal(slice->block_size==0 && slice->slice_count>0 && slice->slice_offset==0); + // should work too if the queue does not contain slice (which can happen during reclaim) + if (slice->prev != NULL) slice->prev->next = slice->next; + if (slice == sq->first) sq->first = slice->next; + if (slice->next != NULL) slice->next->prev = slice->prev; + if (slice == sq->last) sq->last = slice->prev; + slice->prev = NULL; + slice->next = NULL; + slice->block_size = 1; // no more free +} + + +/* ----------------------------------------------------------- + Invariant checking +----------------------------------------------------------- */ + +static bool mi_slice_is_used(const mi_slice_t* slice) { + return (slice->block_size > 0); +} + + +#if (MI_DEBUG>=3) +static bool mi_span_queue_contains(mi_span_queue_t* sq, mi_slice_t* slice) { + for (mi_slice_t* s = sq->first; s != NULL; s = s->next) { + if (s==slice) return true; + } + return false; +} + +static bool mi_segment_is_valid(mi_segment_t* segment, mi_segments_tld_t* tld) { + mi_assert_internal(segment != NULL); + mi_assert_internal(_mi_ptr_cookie(segment) == segment->cookie); + mi_assert_internal(segment->abandoned <= segment->used); + mi_assert_internal(segment->thread_id == 0 || segment->thread_id == _mi_thread_id()); + mi_assert_internal(mi_commit_mask_all_set(&segment->commit_mask, &segment->purge_mask)); // can only decommit committed blocks + //mi_assert_internal(segment->segment_info_size % MI_SEGMENT_SLICE_SIZE == 0); + mi_slice_t* slice = &segment->slices[0]; + const mi_slice_t* end = mi_segment_slices_end(segment); + size_t used_count = 0; + mi_span_queue_t* sq; + while(slice < end) { + mi_assert_internal(slice->slice_count > 0); + mi_assert_internal(slice->slice_offset == 0); + size_t index = mi_slice_index(slice); + size_t maxindex = (index + slice->slice_count >= segment->slice_entries ? segment->slice_entries : index + slice->slice_count) - 1; + if (mi_slice_is_used(slice)) { // a page in use, we need at least MAX_SLICE_OFFSET_COUNT valid back offsets + used_count++; + mi_assert_internal(slice->is_huge == (segment->kind == MI_SEGMENT_HUGE)); + for (size_t i = 0; i <= MI_MAX_SLICE_OFFSET_COUNT && index + i <= maxindex; i++) { + mi_assert_internal(segment->slices[index + i].slice_offset == i*sizeof(mi_slice_t)); + mi_assert_internal(i==0 || segment->slices[index + i].slice_count == 0); + mi_assert_internal(i==0 || segment->slices[index + i].block_size == 1); + } + // and the last entry as well (for coalescing) + const mi_slice_t* last = slice + slice->slice_count - 1; + if (last > slice && last < mi_segment_slices_end(segment)) { + mi_assert_internal(last->slice_offset == (slice->slice_count-1)*sizeof(mi_slice_t)); + mi_assert_internal(last->slice_count == 0); + mi_assert_internal(last->block_size == 1); + } + } + else { // free range of slices; only last slice needs a valid back offset + mi_slice_t* last = &segment->slices[maxindex]; + if (segment->kind != MI_SEGMENT_HUGE || slice->slice_count <= (segment->slice_entries - segment->segment_info_slices)) { + mi_assert_internal((uint8_t*)slice == (uint8_t*)last - last->slice_offset); + } + mi_assert_internal(slice == last || last->slice_count == 0 ); + mi_assert_internal(last->block_size == 0 || (segment->kind==MI_SEGMENT_HUGE && last->block_size==1)); + if (segment->kind != MI_SEGMENT_HUGE && segment->thread_id != 0) { // segment is not huge or abandoned + sq = mi_span_queue_for(slice->slice_count,tld); + mi_assert_internal(mi_span_queue_contains(sq,slice)); + } + } + slice = &segment->slices[maxindex+1]; + } + mi_assert_internal(slice == end); + mi_assert_internal(used_count == segment->used + 1); + return true; +} +#endif + +/* ----------------------------------------------------------- + Segment size calculations +----------------------------------------------------------- */ + +static size_t mi_segment_info_size(mi_segment_t* segment) { + return segment->segment_info_slices * MI_SEGMENT_SLICE_SIZE; +} + +static uint8_t* _mi_segment_page_start_from_slice(const mi_segment_t* segment, const mi_slice_t* slice, size_t block_size, size_t* page_size) +{ + const ptrdiff_t idx = slice - segment->slices; + const size_t psize = (size_t)slice->slice_count * MI_SEGMENT_SLICE_SIZE; + uint8_t* const pstart = (uint8_t*)segment + (idx*MI_SEGMENT_SLICE_SIZE); + // make the start not OS page aligned for smaller blocks to avoid page/cache effects + // note: the offset must always be a block_size multiple since we assume small allocations + // are aligned (see `mi_heap_malloc_aligned`). + size_t start_offset = 0; + if (block_size > 0 && block_size <= MI_MAX_ALIGN_GUARANTEE) { + // for small objects, ensure the page start is aligned with the block size (PR#66 by kickunderscore) + const size_t adjust = block_size - ((uintptr_t)pstart % block_size); + if (adjust < block_size && psize >= block_size + adjust) { + start_offset += adjust; + } + } + if (block_size >= MI_INTPTR_SIZE) { + if (block_size <= 64) { start_offset += 3*block_size; } + else if (block_size <= 512) { start_offset += block_size; } + } + start_offset = _mi_align_up(start_offset, MI_MAX_ALIGN_SIZE); + mi_assert_internal(_mi_is_aligned(pstart + start_offset, MI_MAX_ALIGN_SIZE)); + mi_assert_internal(block_size == 0 || block_size > MI_MAX_ALIGN_GUARANTEE || _mi_is_aligned(pstart + start_offset,block_size)); + if (page_size != NULL) { *page_size = psize - start_offset; } + return (pstart + start_offset); +} + +// Start of the page available memory; can be used on uninitialized pages +uint8_t* _mi_segment_page_start(const mi_segment_t* segment, const mi_page_t* page, size_t* page_size) +{ + const mi_slice_t* slice = mi_page_to_slice((mi_page_t*)page); + uint8_t* p = _mi_segment_page_start_from_slice(segment, slice, mi_page_block_size(page), page_size); + mi_assert_internal(mi_page_block_size(page) > 0 || _mi_ptr_page(p) == page); + mi_assert_internal(_mi_ptr_segment(p) == segment); + return p; +} + + +static size_t mi_segment_calculate_slices(size_t required, size_t* info_slices) { + size_t page_size = _mi_os_page_size(); + size_t isize = _mi_align_up(sizeof(mi_segment_t), page_size); + size_t guardsize = 0; + + if (MI_SECURE>0) { + // in secure mode, we set up a protected page in between the segment info + // and the page data (and one at the end of the segment) + guardsize = page_size; + if (required > 0) { + required = _mi_align_up(required, MI_SEGMENT_SLICE_SIZE) + page_size; + } + } + + isize = _mi_align_up(isize + guardsize, MI_SEGMENT_SLICE_SIZE); + if (info_slices != NULL) *info_slices = isize / MI_SEGMENT_SLICE_SIZE; + size_t segment_size = (required==0 ? MI_SEGMENT_SIZE : _mi_align_up( required + isize + guardsize, MI_SEGMENT_SLICE_SIZE) ); + mi_assert_internal(segment_size % MI_SEGMENT_SLICE_SIZE == 0); + return (segment_size / MI_SEGMENT_SLICE_SIZE); +} + + +/* ---------------------------------------------------------------------------- +Segment caches +We keep a small segment cache per thread to increase local +reuse and avoid setting/clearing guard pages in secure mode. +------------------------------------------------------------------------------- */ + +static void mi_segments_track_size(long segment_size, mi_segments_tld_t* tld) { + if (segment_size>=0) _mi_stat_increase(&tld->stats->segments,1); + else _mi_stat_decrease(&tld->stats->segments,1); + tld->count += (segment_size >= 0 ? 1 : -1); + if (tld->count > tld->peak_count) tld->peak_count = tld->count; + tld->current_size += segment_size; + if (tld->current_size > tld->peak_size) tld->peak_size = tld->current_size; +} + +static void mi_segment_os_free(mi_segment_t* segment, mi_segments_tld_t* tld) { + segment->thread_id = 0; + _mi_segment_map_freed_at(segment); + mi_segments_track_size(-((long)mi_segment_size(segment)),tld); + if (segment->was_reclaimed) { + tld->reclaim_count--; + segment->was_reclaimed = false; + } + if (MI_SECURE>0) { + // _mi_os_unprotect(segment, mi_segment_size(segment)); // ensure no more guard pages are set + // unprotect the guard pages; we cannot just unprotect the whole segment size as part may be decommitted + size_t os_pagesize = _mi_os_page_size(); + _mi_os_unprotect((uint8_t*)segment + mi_segment_info_size(segment) - os_pagesize, os_pagesize); + uint8_t* end = (uint8_t*)segment + mi_segment_size(segment) - os_pagesize; + _mi_os_unprotect(end, os_pagesize); + } + + // purge delayed decommits now? (no, leave it to the arena) + // mi_segment_try_purge(segment,true,tld->stats); + + const size_t size = mi_segment_size(segment); + const size_t csize = _mi_commit_mask_committed_size(&segment->commit_mask, size); + + _mi_arena_free(segment, mi_segment_size(segment), csize, segment->memid); +} + +/* ----------------------------------------------------------- + Commit/Decommit ranges +----------------------------------------------------------- */ + +static void mi_segment_commit_mask(mi_segment_t* segment, bool conservative, uint8_t* p, size_t size, uint8_t** start_p, size_t* full_size, mi_commit_mask_t* cm) { + mi_assert_internal(_mi_ptr_segment(p + 1) == segment); + mi_assert_internal(segment->kind != MI_SEGMENT_HUGE); + mi_commit_mask_create_empty(cm); + if (size == 0 || size > MI_SEGMENT_SIZE || segment->kind == MI_SEGMENT_HUGE) return; + const size_t segstart = mi_segment_info_size(segment); + const size_t segsize = mi_segment_size(segment); + if (p >= (uint8_t*)segment + segsize) return; + + size_t pstart = (p - (uint8_t*)segment); + mi_assert_internal(pstart + size <= segsize); + + size_t start; + size_t end; + if (conservative) { + // decommit conservative + start = _mi_align_up(pstart, MI_COMMIT_SIZE); + end = _mi_align_down(pstart + size, MI_COMMIT_SIZE); + mi_assert_internal(start >= segstart); + mi_assert_internal(end <= segsize); + } + else { + // commit liberal + start = _mi_align_down(pstart, MI_MINIMAL_COMMIT_SIZE); + end = _mi_align_up(pstart + size, MI_MINIMAL_COMMIT_SIZE); + } + if (pstart >= segstart && start < segstart) { // note: the mask is also calculated for an initial commit of the info area + start = segstart; + } + if (end > segsize) { + end = segsize; + } + + mi_assert_internal(start <= pstart && (pstart + size) <= end); + mi_assert_internal(start % MI_COMMIT_SIZE==0 && end % MI_COMMIT_SIZE == 0); + *start_p = (uint8_t*)segment + start; + *full_size = (end > start ? end - start : 0); + if (*full_size == 0) return; + + size_t bitidx = start / MI_COMMIT_SIZE; + mi_assert_internal(bitidx < MI_COMMIT_MASK_BITS); + + size_t bitcount = *full_size / MI_COMMIT_SIZE; // can be 0 + if (bitidx + bitcount > MI_COMMIT_MASK_BITS) { + _mi_warning_message("commit mask overflow: idx=%zu count=%zu start=%zx end=%zx p=0x%p size=%zu fullsize=%zu\n", bitidx, bitcount, start, end, p, size, *full_size); + } + mi_assert_internal((bitidx + bitcount) <= MI_COMMIT_MASK_BITS); + mi_commit_mask_create(bitidx, bitcount, cm); +} + +static bool mi_segment_commit(mi_segment_t* segment, uint8_t* p, size_t size) { + mi_assert_internal(mi_commit_mask_all_set(&segment->commit_mask, &segment->purge_mask)); + + // commit liberal + uint8_t* start = NULL; + size_t full_size = 0; + mi_commit_mask_t mask; + mi_segment_commit_mask(segment, false /* conservative? */, p, size, &start, &full_size, &mask); + if (mi_commit_mask_is_empty(&mask) || full_size == 0) return true; + + if (!mi_commit_mask_all_set(&segment->commit_mask, &mask)) { + // committing + bool is_zero = false; + mi_commit_mask_t cmask; + mi_commit_mask_create_intersect(&segment->commit_mask, &mask, &cmask); + _mi_stat_decrease(&_mi_stats_main.committed, _mi_commit_mask_committed_size(&cmask, MI_SEGMENT_SIZE)); // adjust for overlap + if (!_mi_os_commit(start, full_size, &is_zero)) return false; + mi_commit_mask_set(&segment->commit_mask, &mask); + } + + // increase purge expiration when using part of delayed purges -- we assume more allocations are coming soon. + if (mi_commit_mask_any_set(&segment->purge_mask, &mask)) { + segment->purge_expire = _mi_clock_now() + mi_option_get(mi_option_purge_delay); + } + + // always clear any delayed purges in our range (as they are either committed now) + mi_commit_mask_clear(&segment->purge_mask, &mask); + return true; +} + +static bool mi_segment_ensure_committed(mi_segment_t* segment, uint8_t* p, size_t size) { + mi_assert_internal(mi_commit_mask_all_set(&segment->commit_mask, &segment->purge_mask)); + // note: assumes commit_mask is always full for huge segments as otherwise the commit mask bits can overflow + if (mi_commit_mask_is_full(&segment->commit_mask) && mi_commit_mask_is_empty(&segment->purge_mask)) return true; // fully committed + mi_assert_internal(segment->kind != MI_SEGMENT_HUGE); + return mi_segment_commit(segment, p, size); +} + +static bool mi_segment_purge(mi_segment_t* segment, uint8_t* p, size_t size) { + mi_assert_internal(mi_commit_mask_all_set(&segment->commit_mask, &segment->purge_mask)); + if (!segment->allow_purge) return true; + + // purge conservative + uint8_t* start = NULL; + size_t full_size = 0; + mi_commit_mask_t mask; + mi_segment_commit_mask(segment, true /* conservative? */, p, size, &start, &full_size, &mask); + if (mi_commit_mask_is_empty(&mask) || full_size==0) return true; + + if (mi_commit_mask_any_set(&segment->commit_mask, &mask)) { + // purging + mi_assert_internal((void*)start != (void*)segment); + mi_assert_internal(segment->allow_decommit); + const bool decommitted = _mi_os_purge(start, full_size); // reset or decommit + if (decommitted) { + mi_commit_mask_t cmask; + mi_commit_mask_create_intersect(&segment->commit_mask, &mask, &cmask); + _mi_stat_increase(&_mi_stats_main.committed, full_size - _mi_commit_mask_committed_size(&cmask, MI_SEGMENT_SIZE)); // adjust for double counting + mi_commit_mask_clear(&segment->commit_mask, &mask); + } + } + + // always clear any scheduled purges in our range + mi_commit_mask_clear(&segment->purge_mask, &mask); + return true; +} + +static void mi_segment_schedule_purge(mi_segment_t* segment, uint8_t* p, size_t size) { + if (!segment->allow_purge) return; + + if (mi_option_get(mi_option_purge_delay) == 0) { + mi_segment_purge(segment, p, size); + } + else { + // register for future purge in the purge mask + uint8_t* start = NULL; + size_t full_size = 0; + mi_commit_mask_t mask; + mi_segment_commit_mask(segment, true /*conservative*/, p, size, &start, &full_size, &mask); + if (mi_commit_mask_is_empty(&mask) || full_size==0) return; + + // update delayed commit + mi_assert_internal(segment->purge_expire > 0 || mi_commit_mask_is_empty(&segment->purge_mask)); + mi_commit_mask_t cmask; + mi_commit_mask_create_intersect(&segment->commit_mask, &mask, &cmask); // only purge what is committed; span_free may try to decommit more + mi_commit_mask_set(&segment->purge_mask, &cmask); + mi_msecs_t now = _mi_clock_now(); + if (segment->purge_expire == 0) { + // no previous purgess, initialize now + segment->purge_expire = now + mi_option_get(mi_option_purge_delay); + } + else if (segment->purge_expire <= now) { + // previous purge mask already expired + if (segment->purge_expire + mi_option_get(mi_option_purge_extend_delay) <= now) { + mi_segment_try_purge(segment, true); + } + else { + segment->purge_expire = now + mi_option_get(mi_option_purge_extend_delay); // (mi_option_get(mi_option_purge_delay) / 8); // wait a tiny bit longer in case there is a series of free's + } + } + else { + // previous purge mask is not yet expired, increase the expiration by a bit. + segment->purge_expire += mi_option_get(mi_option_purge_extend_delay); + } + } +} + +static void mi_segment_try_purge(mi_segment_t* segment, bool force) { + if (!segment->allow_purge || segment->purge_expire == 0 || mi_commit_mask_is_empty(&segment->purge_mask)) return; + mi_msecs_t now = _mi_clock_now(); + if (!force && now < segment->purge_expire) return; + + mi_commit_mask_t mask = segment->purge_mask; + segment->purge_expire = 0; + mi_commit_mask_create_empty(&segment->purge_mask); + + size_t idx; + size_t count; + mi_commit_mask_foreach(&mask, idx, count) { + // if found, decommit that sequence + if (count > 0) { + uint8_t* p = (uint8_t*)segment + (idx*MI_COMMIT_SIZE); + size_t size = count * MI_COMMIT_SIZE; + mi_segment_purge(segment, p, size); + } + } + mi_commit_mask_foreach_end() + mi_assert_internal(mi_commit_mask_is_empty(&segment->purge_mask)); +} + +// called from `mi_heap_collect_ex` +// this can be called per-page so it is important that try_purge has fast exit path +void _mi_segment_collect(mi_segment_t* segment, bool force) { + mi_segment_try_purge(segment, force); +} + +/* ----------------------------------------------------------- + Span free +----------------------------------------------------------- */ + +static bool mi_segment_is_abandoned(mi_segment_t* segment) { + return (mi_atomic_load_relaxed(&segment->thread_id) == 0); +} + +// note: can be called on abandoned segments +static void mi_segment_span_free(mi_segment_t* segment, size_t slice_index, size_t slice_count, bool allow_purge, mi_segments_tld_t* tld) { + mi_assert_internal(slice_index < segment->slice_entries); + mi_span_queue_t* sq = (segment->kind == MI_SEGMENT_HUGE || mi_segment_is_abandoned(segment) + ? NULL : mi_span_queue_for(slice_count,tld)); + if (slice_count==0) slice_count = 1; + mi_assert_internal(slice_index + slice_count - 1 < segment->slice_entries); + + // set first and last slice (the intermediates can be undetermined) + mi_slice_t* slice = &segment->slices[slice_index]; + slice->slice_count = (uint32_t)slice_count; + mi_assert_internal(slice->slice_count == slice_count); // no overflow? + slice->slice_offset = 0; + if (slice_count > 1) { + mi_slice_t* last = slice + slice_count - 1; + mi_slice_t* end = (mi_slice_t*)mi_segment_slices_end(segment); + if (last > end) { last = end; } + last->slice_count = 0; + last->slice_offset = (uint32_t)(sizeof(mi_page_t)*(slice_count - 1)); + last->block_size = 0; + } + + // perhaps decommit + if (allow_purge) { + mi_segment_schedule_purge(segment, mi_slice_start(slice), slice_count * MI_SEGMENT_SLICE_SIZE); + } + + // and push it on the free page queue (if it was not a huge page) + if (sq != NULL) mi_span_queue_push( sq, slice ); + else slice->block_size = 0; // mark huge page as free anyways +} + +/* +// called from reclaim to add existing free spans +static void mi_segment_span_add_free(mi_slice_t* slice, mi_segments_tld_t* tld) { + mi_segment_t* segment = _mi_ptr_segment(slice); + mi_assert_internal(slice->xblock_size==0 && slice->slice_count>0 && slice->slice_offset==0); + size_t slice_index = mi_slice_index(slice); + mi_segment_span_free(segment,slice_index,slice->slice_count,tld); +} +*/ + +static void mi_segment_span_remove_from_queue(mi_slice_t* slice, mi_segments_tld_t* tld) { + mi_assert_internal(slice->slice_count > 0 && slice->slice_offset==0 && slice->block_size==0); + mi_assert_internal(_mi_ptr_segment(slice)->kind != MI_SEGMENT_HUGE); + mi_span_queue_t* sq = mi_span_queue_for(slice->slice_count, tld); + mi_span_queue_delete(sq, slice); +} + +// note: can be called on abandoned segments +static mi_slice_t* mi_segment_span_free_coalesce(mi_slice_t* slice, mi_segments_tld_t* tld) { + mi_assert_internal(slice != NULL && slice->slice_count > 0 && slice->slice_offset == 0); + mi_segment_t* const segment = _mi_ptr_segment(slice); + + // for huge pages, just mark as free but don't add to the queues + if (segment->kind == MI_SEGMENT_HUGE) { + // issue #691: segment->used can be 0 if the huge page block was freed while abandoned (reclaim will get here in that case) + mi_assert_internal((segment->used==0 && slice->block_size==0) || segment->used == 1); // decreased right after this call in `mi_segment_page_clear` + slice->block_size = 0; // mark as free anyways + // we should mark the last slice `xblock_size=0` now to maintain invariants but we skip it to + // avoid a possible cache miss (and the segment is about to be freed) + return slice; + } + + // otherwise coalesce the span and add to the free span queues + const bool is_abandoned = (segment->thread_id == 0); // mi_segment_is_abandoned(segment); + size_t slice_count = slice->slice_count; + mi_slice_t* next = slice + slice->slice_count; + mi_assert_internal(next <= mi_segment_slices_end(segment)); + if (next < mi_segment_slices_end(segment) && next->block_size==0) { + // free next block -- remove it from free and merge + mi_assert_internal(next->slice_count > 0 && next->slice_offset==0); + slice_count += next->slice_count; // extend + if (!is_abandoned) { mi_segment_span_remove_from_queue(next, tld); } + } + if (slice > segment->slices) { + mi_slice_t* prev = mi_slice_first(slice - 1); + mi_assert_internal(prev >= segment->slices); + if (prev->block_size==0) { + // free previous slice -- remove it from free and merge + mi_assert_internal(prev->slice_count > 0 && prev->slice_offset==0); + slice_count += prev->slice_count; + slice->slice_count = 0; + slice->slice_offset = (uint32_t)((uint8_t*)slice - (uint8_t*)prev); // set the slice offset for `segment_force_abandon` (in case the previous free block is very large). + if (!is_abandoned) { mi_segment_span_remove_from_queue(prev, tld); } + slice = prev; + } + } + + // and add the new free page + mi_segment_span_free(segment, mi_slice_index(slice), slice_count, true, tld); + return slice; +} + + + +/* ----------------------------------------------------------- + Page allocation +----------------------------------------------------------- */ + +// Note: may still return NULL if committing the memory failed +static mi_page_t* mi_segment_span_allocate(mi_segment_t* segment, size_t slice_index, size_t slice_count) { + mi_assert_internal(slice_index < segment->slice_entries); + mi_slice_t* const slice = &segment->slices[slice_index]; + mi_assert_internal(slice->block_size==0 || slice->block_size==1); + + // commit before changing the slice data + if (!mi_segment_ensure_committed(segment, _mi_segment_page_start_from_slice(segment, slice, 0, NULL), slice_count * MI_SEGMENT_SLICE_SIZE)) { + return NULL; // commit failed! + } + + // convert the slices to a page + slice->slice_offset = 0; + slice->slice_count = (uint32_t)slice_count; + mi_assert_internal(slice->slice_count == slice_count); + const size_t bsize = slice_count * MI_SEGMENT_SLICE_SIZE; + slice->block_size = bsize; + mi_page_t* page = mi_slice_to_page(slice); + mi_assert_internal(mi_page_block_size(page) == bsize); + + // set slice back pointers for the first MI_MAX_SLICE_OFFSET_COUNT entries + size_t extra = slice_count-1; + if (extra > MI_MAX_SLICE_OFFSET_COUNT) extra = MI_MAX_SLICE_OFFSET_COUNT; + if (slice_index + extra >= segment->slice_entries) extra = segment->slice_entries - slice_index - 1; // huge objects may have more slices than avaiable entries in the segment->slices + + mi_slice_t* slice_next = slice + 1; + for (size_t i = 1; i <= extra; i++, slice_next++) { + slice_next->slice_offset = (uint32_t)(sizeof(mi_slice_t)*i); + slice_next->slice_count = 0; + slice_next->block_size = 1; + } + + // and also for the last one (if not set already) (the last one is needed for coalescing and for large alignments) + // note: the cast is needed for ubsan since the index can be larger than MI_SLICES_PER_SEGMENT for huge allocations (see #543) + mi_slice_t* last = slice + slice_count - 1; + mi_slice_t* end = (mi_slice_t*)mi_segment_slices_end(segment); + if (last > end) last = end; + if (last > slice) { + last->slice_offset = (uint32_t)(sizeof(mi_slice_t) * (last - slice)); + last->slice_count = 0; + last->block_size = 1; + } + + // and initialize the page + page->is_committed = true; + page->is_huge = (segment->kind == MI_SEGMENT_HUGE); + segment->used++; + return page; +} + +static void mi_segment_slice_split(mi_segment_t* segment, mi_slice_t* slice, size_t slice_count, mi_segments_tld_t* tld) { + mi_assert_internal(_mi_ptr_segment(slice) == segment); + mi_assert_internal(slice->slice_count >= slice_count); + mi_assert_internal(slice->block_size > 0); // no more in free queue + if (slice->slice_count <= slice_count) return; + mi_assert_internal(segment->kind != MI_SEGMENT_HUGE); + size_t next_index = mi_slice_index(slice) + slice_count; + size_t next_count = slice->slice_count - slice_count; + mi_segment_span_free(segment, next_index, next_count, false /* don't purge left-over part */, tld); + slice->slice_count = (uint32_t)slice_count; +} + +static mi_page_t* mi_segments_page_find_and_allocate(size_t slice_count, mi_arena_id_t req_arena_id, mi_segments_tld_t* tld) { + mi_assert_internal(slice_count*MI_SEGMENT_SLICE_SIZE <= MI_LARGE_OBJ_SIZE_MAX); + // search from best fit up + mi_span_queue_t* sq = mi_span_queue_for(slice_count, tld); + if (slice_count == 0) slice_count = 1; + while (sq <= &tld->spans[MI_SEGMENT_BIN_MAX]) { + for (mi_slice_t* slice = sq->first; slice != NULL; slice = slice->next) { + if (slice->slice_count >= slice_count) { + // found one + mi_segment_t* segment = _mi_ptr_segment(slice); + if (_mi_arena_memid_is_suitable(segment->memid, req_arena_id)) { + // found a suitable page span + mi_span_queue_delete(sq, slice); + + if (slice->slice_count > slice_count) { + mi_segment_slice_split(segment, slice, slice_count, tld); + } + mi_assert_internal(slice != NULL && slice->slice_count == slice_count && slice->block_size > 0); + mi_page_t* page = mi_segment_span_allocate(segment, mi_slice_index(slice), slice->slice_count); + if (page == NULL) { + // commit failed; return NULL but first restore the slice + mi_segment_span_free_coalesce(slice, tld); + return NULL; + } + return page; + } + } + } + sq++; + } + // could not find a page.. + return NULL; +} + + +/* ----------------------------------------------------------- + Segment allocation +----------------------------------------------------------- */ + +static mi_segment_t* mi_segment_os_alloc( size_t required, size_t page_alignment, bool eager_delayed, mi_arena_id_t req_arena_id, + size_t* psegment_slices, size_t* pinfo_slices, + bool commit, mi_segments_tld_t* tld) + +{ + mi_memid_t memid; + bool allow_large = (!eager_delayed && (MI_SECURE == 0)); // only allow large OS pages once we are no longer lazy + size_t align_offset = 0; + size_t alignment = MI_SEGMENT_ALIGN; + + if (page_alignment > 0) { + // mi_assert_internal(huge_page != NULL); + mi_assert_internal(page_alignment >= MI_SEGMENT_ALIGN); + alignment = page_alignment; + const size_t info_size = (*pinfo_slices) * MI_SEGMENT_SLICE_SIZE; + align_offset = _mi_align_up( info_size, MI_SEGMENT_ALIGN ); + const size_t extra = align_offset - info_size; + // recalculate due to potential guard pages + *psegment_slices = mi_segment_calculate_slices(required + extra, pinfo_slices); + mi_assert_internal(*psegment_slices > 0 && *psegment_slices <= UINT32_MAX); + } + + const size_t segment_size = (*psegment_slices) * MI_SEGMENT_SLICE_SIZE; + mi_segment_t* segment = (mi_segment_t*)_mi_arena_alloc_aligned(segment_size, alignment, align_offset, commit, allow_large, req_arena_id, &memid); + if (segment == NULL) { + return NULL; // failed to allocate + } + + // ensure metadata part of the segment is committed + mi_commit_mask_t commit_mask; + if (memid.initially_committed) { + mi_commit_mask_create_full(&commit_mask); + } + else { + // at least commit the info slices + const size_t commit_needed = _mi_divide_up((*pinfo_slices)*MI_SEGMENT_SLICE_SIZE, MI_COMMIT_SIZE); + mi_assert_internal(commit_needed>0); + mi_commit_mask_create(0, commit_needed, &commit_mask); + mi_assert_internal(commit_needed*MI_COMMIT_SIZE >= (*pinfo_slices)*MI_SEGMENT_SLICE_SIZE); + if (!_mi_os_commit(segment, commit_needed*MI_COMMIT_SIZE, NULL)) { + _mi_arena_free(segment,segment_size,0,memid); + return NULL; + } + } + mi_assert_internal(segment != NULL && (uintptr_t)segment % MI_SEGMENT_SIZE == 0); + + segment->memid = memid; + segment->allow_decommit = !memid.is_pinned; + segment->allow_purge = segment->allow_decommit && (mi_option_get(mi_option_purge_delay) >= 0); + segment->segment_size = segment_size; + segment->subproc = tld->subproc; + segment->commit_mask = commit_mask; + segment->purge_expire = 0; + mi_commit_mask_create_empty(&segment->purge_mask); + + mi_segments_track_size((long)(segment_size), tld); + _mi_segment_map_allocated_at(segment); + return segment; +} + + +// Allocate a segment from the OS aligned to `MI_SEGMENT_SIZE` . +static mi_segment_t* mi_segment_alloc(size_t required, size_t page_alignment, mi_arena_id_t req_arena_id, mi_segments_tld_t* tld, mi_page_t** huge_page) +{ + mi_assert_internal((required==0 && huge_page==NULL) || (required>0 && huge_page != NULL)); + + // calculate needed sizes first + size_t info_slices; + size_t segment_slices = mi_segment_calculate_slices(required, &info_slices); + mi_assert_internal(segment_slices > 0 && segment_slices <= UINT32_MAX); + + // Commit eagerly only if not the first N lazy segments (to reduce impact of many threads that allocate just a little) + const bool eager_delay = (// !_mi_os_has_overcommit() && // never delay on overcommit systems + _mi_current_thread_count() > 1 && // do not delay for the first N threads + tld->peak_count < (size_t)mi_option_get(mi_option_eager_commit_delay)); + const bool eager = !eager_delay && mi_option_is_enabled(mi_option_eager_commit); + bool commit = eager || (required > 0); + + // Allocate the segment from the OS + mi_segment_t* segment = mi_segment_os_alloc(required, page_alignment, eager_delay, req_arena_id, + &segment_slices, &info_slices, commit, tld); + if (segment == NULL) return NULL; + + // zero the segment info? -- not always needed as it may be zero initialized from the OS + if (!segment->memid.initially_zero) { + ptrdiff_t ofs = offsetof(mi_segment_t, next); + size_t prefix = offsetof(mi_segment_t, slices) - ofs; + size_t zsize = prefix + (sizeof(mi_slice_t) * (segment_slices + 1)); // one more + _mi_memzero((uint8_t*)segment + ofs, zsize); + } + + // initialize the rest of the segment info + const size_t slice_entries = (segment_slices > MI_SLICES_PER_SEGMENT ? MI_SLICES_PER_SEGMENT : segment_slices); + segment->segment_slices = segment_slices; + segment->segment_info_slices = info_slices; + segment->thread_id = _mi_thread_id(); + segment->cookie = _mi_ptr_cookie(segment); + segment->slice_entries = slice_entries; + segment->kind = (required == 0 ? MI_SEGMENT_NORMAL : MI_SEGMENT_HUGE); + + // _mi_memzero(segment->slices, sizeof(mi_slice_t)*(info_slices+1)); + _mi_stat_increase(&tld->stats->page_committed, mi_segment_info_size(segment)); + + // set up guard pages + size_t guard_slices = 0; + if (MI_SECURE>0) { + // in secure mode, we set up a protected page in between the segment info + // and the page data, and at the end of the segment. + size_t os_pagesize = _mi_os_page_size(); + _mi_os_protect((uint8_t*)segment + mi_segment_info_size(segment) - os_pagesize, os_pagesize); + uint8_t* end = (uint8_t*)segment + mi_segment_size(segment) - os_pagesize; + mi_segment_ensure_committed(segment, end, os_pagesize); + _mi_os_protect(end, os_pagesize); + if (slice_entries == segment_slices) segment->slice_entries--; // don't use the last slice :-( + guard_slices = 1; + } + + // reserve first slices for segment info + mi_page_t* page0 = mi_segment_span_allocate(segment, 0, info_slices); + mi_assert_internal(page0!=NULL); if (page0==NULL) return NULL; // cannot fail as we always commit in advance + mi_assert_internal(segment->used == 1); + segment->used = 0; // don't count our internal slices towards usage + + // initialize initial free pages + if (segment->kind == MI_SEGMENT_NORMAL) { // not a huge page + mi_assert_internal(huge_page==NULL); + mi_segment_span_free(segment, info_slices, segment->slice_entries - info_slices, false /* don't purge */, tld); + } + else { + mi_assert_internal(huge_page!=NULL); + mi_assert_internal(mi_commit_mask_is_empty(&segment->purge_mask)); + mi_assert_internal(mi_commit_mask_is_full(&segment->commit_mask)); + *huge_page = mi_segment_span_allocate(segment, info_slices, segment_slices - info_slices - guard_slices); + mi_assert_internal(*huge_page != NULL); // cannot fail as we commit in advance + } + + mi_assert_expensive(mi_segment_is_valid(segment,tld)); + return segment; +} + + +static void mi_segment_free(mi_segment_t* segment, bool force, mi_segments_tld_t* tld) { + MI_UNUSED(force); + mi_assert_internal(segment != NULL); + mi_assert_internal(segment->next == NULL); + mi_assert_internal(segment->used == 0); + + // in `mi_segment_force_abandon` we set this to true to ensure the segment's memory stays valid + if (segment->dont_free) return; + + // Remove the free pages + mi_slice_t* slice = &segment->slices[0]; + const mi_slice_t* end = mi_segment_slices_end(segment); + #if MI_DEBUG>1 + size_t page_count = 0; + #endif + while (slice < end) { + mi_assert_internal(slice->slice_count > 0); + mi_assert_internal(slice->slice_offset == 0); + mi_assert_internal(mi_slice_index(slice)==0 || slice->block_size == 0); // no more used pages .. + if (slice->block_size == 0 && segment->kind != MI_SEGMENT_HUGE) { + mi_segment_span_remove_from_queue(slice, tld); + } + #if MI_DEBUG>1 + page_count++; + #endif + slice = slice + slice->slice_count; + } + mi_assert_internal(page_count == 2); // first page is allocated by the segment itself + + // stats + // _mi_stat_decrease(&tld->stats->page_committed, mi_segment_info_size(segment)); + + // return it to the OS + mi_segment_os_free(segment, tld); +} + + +/* ----------------------------------------------------------- + Page Free +----------------------------------------------------------- */ + +static void mi_segment_abandon(mi_segment_t* segment, mi_segments_tld_t* tld); + +// note: can be called on abandoned pages +static mi_slice_t* mi_segment_page_clear(mi_page_t* page, mi_segments_tld_t* tld) { + mi_assert_internal(page->block_size > 0); + mi_assert_internal(mi_page_all_free(page)); + mi_segment_t* segment = _mi_ptr_segment(page); + mi_assert_internal(segment->used > 0); + + size_t inuse = page->capacity * mi_page_block_size(page); + _mi_stat_decrease(&tld->stats->page_committed, inuse); + _mi_stat_decrease(&tld->stats->pages, 1); + + // reset the page memory to reduce memory pressure? + if (segment->allow_decommit && mi_option_is_enabled(mi_option_deprecated_page_reset)) { + size_t psize; + uint8_t* start = _mi_segment_page_start(segment, page, &psize); + _mi_os_reset(start, psize); + } + + // zero the page data, but not the segment fields and heap tag + page->is_zero_init = false; + uint8_t heap_tag = page->heap_tag; + ptrdiff_t ofs = offsetof(mi_page_t, capacity); + _mi_memzero((uint8_t*)page + ofs, sizeof(*page) - ofs); + page->block_size = 1; + page->heap_tag = heap_tag; + + // and free it + mi_slice_t* slice = mi_segment_span_free_coalesce(mi_page_to_slice(page), tld); + segment->used--; + // cannot assert segment valid as it is called during reclaim + // mi_assert_expensive(mi_segment_is_valid(segment, tld)); + return slice; +} + +void _mi_segment_page_free(mi_page_t* page, bool force, mi_segments_tld_t* tld) +{ + mi_assert(page != NULL); + mi_segment_t* segment = _mi_page_segment(page); + mi_assert_expensive(mi_segment_is_valid(segment,tld)); + + // mark it as free now + mi_segment_page_clear(page, tld); + mi_assert_expensive(mi_segment_is_valid(segment, tld)); + + if (segment->used == 0) { + // no more used pages; remove from the free list and free the segment + mi_segment_free(segment, force, tld); + } + else if (segment->used == segment->abandoned) { + // only abandoned pages; remove from free list and abandon + mi_segment_abandon(segment,tld); + } + else { + // perform delayed purges + mi_segment_try_purge(segment, false /* force? */); + } +} + + +/* ----------------------------------------------------------- +Abandonment + +When threads terminate, they can leave segments with +live blocks (reachable through other threads). Such segments +are "abandoned" and will be reclaimed by other threads to +reuse their pages and/or free them eventually. The +`thread_id` of such segments is 0. + +When a block is freed in an abandoned segment, the segment +is reclaimed into that thread. + +Moreover, if threads are looking for a fresh segment, they +will first consider abandoned segments -- these can be found +by scanning the arena memory +(segments outside arena memoryare only reclaimed by a free). +----------------------------------------------------------- */ + +/* ----------------------------------------------------------- + Abandon segment/page +----------------------------------------------------------- */ + +static void mi_segment_abandon(mi_segment_t* segment, mi_segments_tld_t* tld) { + mi_assert_internal(segment->used == segment->abandoned); + mi_assert_internal(segment->used > 0); + mi_assert_internal(segment->abandoned_visits == 0); + mi_assert_expensive(mi_segment_is_valid(segment,tld)); + + // remove the free pages from the free page queues + mi_slice_t* slice = &segment->slices[0]; + const mi_slice_t* end = mi_segment_slices_end(segment); + while (slice < end) { + mi_assert_internal(slice->slice_count > 0); + mi_assert_internal(slice->slice_offset == 0); + if (slice->block_size == 0) { // a free page + mi_segment_span_remove_from_queue(slice,tld); + slice->block_size = 0; // but keep it free + } + slice = slice + slice->slice_count; + } + + // perform delayed decommits (forcing is much slower on mstress) + // Only abandoned segments in arena memory can be reclaimed without a free + // so if a segment is not from an arena we force purge here to be conservative. + const bool force_purge = (segment->memid.memkind != MI_MEM_ARENA) || mi_option_is_enabled(mi_option_abandoned_page_purge); + mi_segment_try_purge(segment, force_purge); + + // all pages in the segment are abandoned; add it to the abandoned list + _mi_stat_increase(&tld->stats->segments_abandoned, 1); + mi_segments_track_size(-((long)mi_segment_size(segment)), tld); + segment->thread_id = 0; + segment->abandoned_visits = 1; // from 0 to 1 to signify it is abandoned + if (segment->was_reclaimed) { + tld->reclaim_count--; + segment->was_reclaimed = false; + } + _mi_arena_segment_mark_abandoned(segment); +} + +void _mi_segment_page_abandon(mi_page_t* page, mi_segments_tld_t* tld) { + mi_assert(page != NULL); + mi_assert_internal(mi_page_thread_free_flag(page)==MI_NEVER_DELAYED_FREE); + mi_assert_internal(mi_page_heap(page) == NULL); + mi_segment_t* segment = _mi_page_segment(page); + + mi_assert_expensive(mi_segment_is_valid(segment,tld)); + segment->abandoned++; + + _mi_stat_increase(&tld->stats->pages_abandoned, 1); + mi_assert_internal(segment->abandoned <= segment->used); + if (segment->used == segment->abandoned) { + // all pages are abandoned, abandon the entire segment + mi_segment_abandon(segment, tld); + } +} + +/* ----------------------------------------------------------- + Reclaim abandoned pages +----------------------------------------------------------- */ + +static mi_slice_t* mi_slices_start_iterate(mi_segment_t* segment, const mi_slice_t** end) { + mi_slice_t* slice = &segment->slices[0]; + *end = mi_segment_slices_end(segment); + mi_assert_internal(slice->slice_count>0 && slice->block_size>0); // segment allocated page + slice = slice + slice->slice_count; // skip the first segment allocated page + return slice; +} + +// Possibly free pages and check if free space is available +static bool mi_segment_check_free(mi_segment_t* segment, size_t slices_needed, size_t block_size, mi_segments_tld_t* tld) +{ + mi_assert_internal(mi_segment_is_abandoned(segment)); + bool has_page = false; + + // for all slices + const mi_slice_t* end; + mi_slice_t* slice = mi_slices_start_iterate(segment, &end); + while (slice < end) { + mi_assert_internal(slice->slice_count > 0); + mi_assert_internal(slice->slice_offset == 0); + if (mi_slice_is_used(slice)) { // used page + // ensure used count is up to date and collect potential concurrent frees + mi_page_t* const page = mi_slice_to_page(slice); + _mi_page_free_collect(page, false); + if (mi_page_all_free(page)) { + // if this page is all free now, free it without adding to any queues (yet) + mi_assert_internal(page->next == NULL && page->prev==NULL); + _mi_stat_decrease(&tld->stats->pages_abandoned, 1); + segment->abandoned--; + slice = mi_segment_page_clear(page, tld); // re-assign slice due to coalesce! + mi_assert_internal(!mi_slice_is_used(slice)); + if (slice->slice_count >= slices_needed) { + has_page = true; + } + } + else if (mi_page_block_size(page) == block_size && mi_page_has_any_available(page)) { + // a page has available free blocks of the right size + has_page = true; + } + } + else { + // empty span + if (slice->slice_count >= slices_needed) { + has_page = true; + } + } + slice = slice + slice->slice_count; + } + return has_page; +} + +// Reclaim an abandoned segment; returns NULL if the segment was freed +// set `right_page_reclaimed` to `true` if it reclaimed a page of the right `block_size` that was not full. +static mi_segment_t* mi_segment_reclaim(mi_segment_t* segment, mi_heap_t* heap, size_t requested_block_size, bool* right_page_reclaimed, mi_segments_tld_t* tld) { + if (right_page_reclaimed != NULL) { *right_page_reclaimed = false; } + // can be 0 still with abandoned_next, or already a thread id for segments outside an arena that are reclaimed on a free. + mi_assert_internal(mi_atomic_load_relaxed(&segment->thread_id) == 0 || mi_atomic_load_relaxed(&segment->thread_id) == _mi_thread_id()); + mi_assert_internal(segment->subproc == heap->tld->segments.subproc); // only reclaim within the same subprocess + mi_atomic_store_release(&segment->thread_id, _mi_thread_id()); + segment->abandoned_visits = 0; + segment->was_reclaimed = true; + tld->reclaim_count++; + mi_segments_track_size((long)mi_segment_size(segment), tld); + mi_assert_internal(segment->next == NULL); + _mi_stat_decrease(&tld->stats->segments_abandoned, 1); + + // for all slices + const mi_slice_t* end; + mi_slice_t* slice = mi_slices_start_iterate(segment, &end); + while (slice < end) { + mi_assert_internal(slice->slice_count > 0); + mi_assert_internal(slice->slice_offset == 0); + if (mi_slice_is_used(slice)) { + // in use: reclaim the page in our heap + mi_page_t* page = mi_slice_to_page(slice); + mi_assert_internal(page->is_committed); + mi_assert_internal(mi_page_thread_free_flag(page)==MI_NEVER_DELAYED_FREE); + mi_assert_internal(mi_page_heap(page) == NULL); + mi_assert_internal(page->next == NULL && page->prev==NULL); + _mi_stat_decrease(&tld->stats->pages_abandoned, 1); + segment->abandoned--; + // get the target heap for this thread which has a matching heap tag (so we reclaim into a matching heap) + mi_heap_t* target_heap = _mi_heap_by_tag(heap, page->heap_tag); // allow custom heaps to separate objects + if (target_heap == NULL) { + target_heap = heap; + _mi_error_message(EFAULT, "page with tag %u cannot be reclaimed by a heap with the same tag (using heap tag %u instead)\n", page->heap_tag, heap->tag ); + } + // associate the heap with this page, and allow heap thread delayed free again. + mi_page_set_heap(page, target_heap); + _mi_page_use_delayed_free(page, MI_USE_DELAYED_FREE, true); // override never (after heap is set) + _mi_page_free_collect(page, false); // ensure used count is up to date + if (mi_page_all_free(page)) { + // if everything free by now, free the page + slice = mi_segment_page_clear(page, tld); // set slice again due to coalesceing + } + else { + // otherwise reclaim it into the heap + _mi_page_reclaim(target_heap, page); + if (requested_block_size == mi_page_block_size(page) && mi_page_has_any_available(page) && heap == target_heap) { + if (right_page_reclaimed != NULL) { *right_page_reclaimed = true; } + } + } + } + else { + // the span is free, add it to our page queues + slice = mi_segment_span_free_coalesce(slice, tld); // set slice again due to coalesceing + } + mi_assert_internal(slice->slice_count>0 && slice->slice_offset==0); + slice = slice + slice->slice_count; + } + + mi_assert(segment->abandoned == 0); + mi_assert_expensive(mi_segment_is_valid(segment, tld)); + if (segment->used == 0) { // due to page_clear + mi_assert_internal(right_page_reclaimed == NULL || !(*right_page_reclaimed)); + mi_segment_free(segment, false, tld); + return NULL; + } + else { + return segment; + } +} + + +// attempt to reclaim a particular segment (called from multi threaded free `alloc.c:mi_free_block_mt`) +bool _mi_segment_attempt_reclaim(mi_heap_t* heap, mi_segment_t* segment) { + if (mi_atomic_load_relaxed(&segment->thread_id) != 0) return false; // it is not abandoned + if (segment->subproc != heap->tld->segments.subproc) return false; // only reclaim within the same subprocess + if (!_mi_heap_memid_is_suitable(heap,segment->memid)) return false; // don't reclaim between exclusive and non-exclusive arena's + const long target = _mi_option_get_fast(mi_option_target_segments_per_thread); + if (target > 0 && (size_t)target <= heap->tld->segments.count) return false; // don't reclaim if going above the target count + + // don't reclaim more from a `free` call than half the current segments + // this is to prevent a pure free-ing thread to start owning too many segments + // (but not for out-of-arena segments as that is the main way to be reclaimed for those) + if (segment->memid.memkind == MI_MEM_ARENA && heap->tld->segments.reclaim_count * 2 > heap->tld->segments.count) { + return false; + } + if (_mi_arena_segment_clear_abandoned(segment)) { // atomically unabandon + mi_segment_t* res = mi_segment_reclaim(segment, heap, 0, NULL, &heap->tld->segments); + mi_assert_internal(res == segment); + return (res != NULL); + } + return false; +} + +void _mi_abandoned_reclaim_all(mi_heap_t* heap, mi_segments_tld_t* tld) { + mi_segment_t* segment; + mi_arena_field_cursor_t current; + _mi_arena_field_cursor_init(heap, tld->subproc, true /* visit all, blocking */, ¤t); + while ((segment = _mi_arena_segment_clear_abandoned_next(¤t)) != NULL) { + mi_segment_reclaim(segment, heap, 0, NULL, tld); + } + _mi_arena_field_cursor_done(¤t); +} + + +static bool segment_count_is_within_target(mi_segments_tld_t* tld, size_t* ptarget) { + const size_t target = (size_t)mi_option_get_clamp(mi_option_target_segments_per_thread, 0, 1024); + if (ptarget != NULL) { *ptarget = target; } + return (target == 0 || tld->count < target); +} + +static long mi_segment_get_reclaim_tries(mi_segments_tld_t* tld) { + // limit the tries to 10% (default) of the abandoned segments with at least 8 and at most 1024 tries. + const size_t perc = (size_t)mi_option_get_clamp(mi_option_max_segment_reclaim, 0, 100); + if (perc <= 0) return 0; + const size_t total_count = mi_atomic_load_relaxed(&tld->subproc->abandoned_count); + if (total_count == 0) return 0; + const size_t relative_count = (total_count > 10000 ? (total_count / 100) * perc : (total_count * perc) / 100); // avoid overflow + long max_tries = (long)(relative_count <= 1 ? 1 : (relative_count > 1024 ? 1024 : relative_count)); + if (max_tries < 8 && total_count > 8) { max_tries = 8; } + return max_tries; +} + +static mi_segment_t* mi_segment_try_reclaim(mi_heap_t* heap, size_t needed_slices, size_t block_size, bool* reclaimed, mi_segments_tld_t* tld) +{ + *reclaimed = false; + long max_tries = mi_segment_get_reclaim_tries(tld); + if (max_tries <= 0) return NULL; + + mi_segment_t* result = NULL; + mi_segment_t* segment = NULL; + mi_arena_field_cursor_t current; + _mi_arena_field_cursor_init(heap, tld->subproc, false /* non-blocking */, ¤t); + while (segment_count_is_within_target(tld,NULL) && (max_tries-- > 0) && ((segment = _mi_arena_segment_clear_abandoned_next(¤t)) != NULL)) + { + mi_assert(segment->subproc == heap->tld->segments.subproc); // cursor only visits segments in our sub-process + segment->abandoned_visits++; + // todo: should we respect numa affinity for abandoned reclaim? perhaps only for the first visit? + // todo: an arena exclusive heap will potentially visit many abandoned unsuitable segments and use many tries + // Perhaps we can skip non-suitable ones in a better way? + bool is_suitable = _mi_heap_memid_is_suitable(heap, segment->memid); + bool has_page = mi_segment_check_free(segment,needed_slices,block_size,tld); // try to free up pages (due to concurrent frees) + if (segment->used == 0) { + // free the segment (by forced reclaim) to make it available to other threads. + // note1: we prefer to free a segment as that might lead to reclaiming another + // segment that is still partially used. + // note2: we could in principle optimize this by skipping reclaim and directly + // freeing but that would violate some invariants temporarily) + mi_segment_reclaim(segment, heap, 0, NULL, tld); + } + else if (has_page && is_suitable) { + // found a large enough free span, or a page of the right block_size with free space + // we return the result of reclaim (which is usually `segment`) as it might free + // the segment due to concurrent frees (in which case `NULL` is returned). + result = mi_segment_reclaim(segment, heap, block_size, reclaimed, tld); + break; + } + else if (segment->abandoned_visits > 3 && is_suitable) { + // always reclaim on 3rd visit to limit the abandoned segment count. + mi_segment_reclaim(segment, heap, 0, NULL, tld); + } + else { + // otherwise, push on the visited list so it gets not looked at too quickly again + max_tries++; // don't count this as a try since it was not suitable + mi_segment_try_purge(segment, false /* true force? */); // force purge if needed as we may not visit soon again + _mi_arena_segment_mark_abandoned(segment); + } + } + _mi_arena_field_cursor_done(¤t); + return result; +} + +// collect abandoned segments +void _mi_abandoned_collect(mi_heap_t* heap, bool force, mi_segments_tld_t* tld) +{ + mi_segment_t* segment; + mi_arena_field_cursor_t current; _mi_arena_field_cursor_init(heap, tld->subproc, force /* blocking? */, ¤t); + long max_tries = (force ? (long)mi_atomic_load_relaxed(&tld->subproc->abandoned_count) : 1024); // limit latency + while ((max_tries-- > 0) && ((segment = _mi_arena_segment_clear_abandoned_next(¤t)) != NULL)) { + mi_segment_check_free(segment,0,0,tld); // try to free up pages (due to concurrent frees) + if (segment->used == 0) { + // free the segment (by forced reclaim) to make it available to other threads. + // note: we could in principle optimize this by skipping reclaim and directly + // freeing but that would violate some invariants temporarily) + mi_segment_reclaim(segment, heap, 0, NULL, tld); + } + else { + // otherwise, purge if needed and push on the visited list + // note: forced purge can be expensive if many threads are destroyed/created as in mstress. + mi_segment_try_purge(segment, force); + _mi_arena_segment_mark_abandoned(segment); + } + } + _mi_arena_field_cursor_done(¤t); +} + +/* ----------------------------------------------------------- + Force abandon a segment that is in use by our thread +----------------------------------------------------------- */ + +// force abandon a segment +static void mi_segment_force_abandon(mi_segment_t* segment, mi_segments_tld_t* tld) +{ + mi_assert_internal(!mi_segment_is_abandoned(segment)); + mi_assert_internal(!segment->dont_free); + + // ensure the segment does not get free'd underneath us (so we can check if a page has been freed in `mi_page_force_abandon`) + segment->dont_free = true; + + // for all slices + const mi_slice_t* end; + mi_slice_t* slice = mi_slices_start_iterate(segment, &end); + while (slice < end) { + mi_assert_internal(slice->slice_count > 0); + mi_assert_internal(slice->slice_offset == 0); + if (mi_slice_is_used(slice)) { + // ensure used count is up to date and collect potential concurrent frees + mi_page_t* const page = mi_slice_to_page(slice); + _mi_page_free_collect(page, false); + { + // abandon the page if it is still in-use (this will free it if possible as well) + mi_assert_internal(segment->used > 0); + if (segment->used == segment->abandoned+1) { + // the last page.. abandon and return as the segment will be abandoned after this + // and we should no longer access it. + segment->dont_free = false; + _mi_page_force_abandon(page); + return; + } + else { + // abandon and continue + _mi_page_force_abandon(page); + // it might be freed, reset the slice (note: relies on coalesce setting the slice_offset) + slice = mi_slice_first(slice); + } + } + } + slice = slice + slice->slice_count; + } + segment->dont_free = false; + mi_assert(segment->used == segment->abandoned); + mi_assert(segment->used == 0); + if (segment->used == 0) { // paranoia + // all free now + mi_segment_free(segment, false, tld); + } + else { + // perform delayed purges + mi_segment_try_purge(segment, false /* force? */); + } +} + + +// try abandon segments. +// this should be called from `reclaim_or_alloc` so we know all segments are (about) fully in use. +static void mi_segments_try_abandon_to_target(mi_heap_t* heap, size_t target, mi_segments_tld_t* tld) { + if (target <= 1) return; + const size_t min_target = (target > 4 ? (target*3)/4 : target); // 75% + // todo: we should maintain a list of segments per thread; for now, only consider segments from the heap full pages + for (int i = 0; i < 64 && tld->count >= min_target; i++) { + mi_page_t* page = heap->pages[MI_BIN_FULL].first; + while (page != NULL && mi_page_block_size(page) > MI_LARGE_OBJ_SIZE_MAX) { + page = page->next; + } + if (page==NULL) { + break; + } + mi_segment_t* segment = _mi_page_segment(page); + mi_segment_force_abandon(segment, tld); + mi_assert_internal(page != heap->pages[MI_BIN_FULL].first); // as it is just abandoned + } +} + +// try abandon segments. +// this should be called from `reclaim_or_alloc` so we know all segments are (about) fully in use. +static void mi_segments_try_abandon(mi_heap_t* heap, mi_segments_tld_t* tld) { + // we call this when we are about to add a fresh segment so we should be under our target segment count. + size_t target = 0; + if (segment_count_is_within_target(tld, &target)) return; + mi_segments_try_abandon_to_target(heap, target, tld); +} + +void mi_collect_reduce(size_t target_size) mi_attr_noexcept { + mi_collect(true); + mi_heap_t* heap = mi_heap_get_default(); + mi_segments_tld_t* tld = &heap->tld->segments; + size_t target = target_size / MI_SEGMENT_SIZE; + if (target == 0) { + target = (size_t)mi_option_get_clamp(mi_option_target_segments_per_thread, 1, 1024); + } + mi_segments_try_abandon_to_target(heap, target, tld); +} + +/* ----------------------------------------------------------- + Reclaim or allocate +----------------------------------------------------------- */ + +static mi_segment_t* mi_segment_reclaim_or_alloc(mi_heap_t* heap, size_t needed_slices, size_t block_size, mi_segments_tld_t* tld) +{ + mi_assert_internal(block_size <= MI_LARGE_OBJ_SIZE_MAX); + + // try to abandon some segments to increase reuse between threads + mi_segments_try_abandon(heap,tld); + + // 1. try to reclaim an abandoned segment + bool reclaimed; + mi_segment_t* segment = mi_segment_try_reclaim(heap, needed_slices, block_size, &reclaimed, tld); + if (reclaimed) { + // reclaimed the right page right into the heap + mi_assert_internal(segment != NULL); + return NULL; // pretend out-of-memory as the page will be in the page queue of the heap with available blocks + } + else if (segment != NULL) { + // reclaimed a segment with a large enough empty span in it + return segment; + } + // 2. otherwise allocate a fresh segment + return mi_segment_alloc(0, 0, heap->arena_id, tld, NULL); +} + + +/* ----------------------------------------------------------- + Page allocation +----------------------------------------------------------- */ + +static mi_page_t* mi_segments_page_alloc(mi_heap_t* heap, mi_page_kind_t page_kind, size_t required, size_t block_size, mi_segments_tld_t* tld) +{ + mi_assert_internal(required <= MI_LARGE_OBJ_SIZE_MAX && page_kind <= MI_PAGE_LARGE); + + // find a free page + size_t page_size = _mi_align_up(required, (required > MI_MEDIUM_PAGE_SIZE ? MI_MEDIUM_PAGE_SIZE : MI_SEGMENT_SLICE_SIZE)); + size_t slices_needed = page_size / MI_SEGMENT_SLICE_SIZE; + mi_assert_internal(slices_needed * MI_SEGMENT_SLICE_SIZE == page_size); + mi_page_t* page = mi_segments_page_find_and_allocate(slices_needed, heap->arena_id, tld); //(required <= MI_SMALL_SIZE_MAX ? 0 : slices_needed), tld); + if (page==NULL) { + // no free page, allocate a new segment and try again + if (mi_segment_reclaim_or_alloc(heap, slices_needed, block_size, tld) == NULL) { + // OOM or reclaimed a good page in the heap + return NULL; + } + else { + // otherwise try again + return mi_segments_page_alloc(heap, page_kind, required, block_size, tld); + } + } + mi_assert_internal(page != NULL && page->slice_count*MI_SEGMENT_SLICE_SIZE == page_size); + mi_assert_internal(_mi_ptr_segment(page)->thread_id == _mi_thread_id()); + mi_segment_try_purge(_mi_ptr_segment(page), false); + return page; +} + + + +/* ----------------------------------------------------------- + Huge page allocation +----------------------------------------------------------- */ + +static mi_page_t* mi_segment_huge_page_alloc(size_t size, size_t page_alignment, mi_arena_id_t req_arena_id, mi_segments_tld_t* tld) +{ + mi_page_t* page = NULL; + mi_segment_t* segment = mi_segment_alloc(size,page_alignment,req_arena_id,tld,&page); + if (segment == NULL || page==NULL) return NULL; + mi_assert_internal(segment->used==1); + mi_assert_internal(mi_page_block_size(page) >= size); + #if MI_HUGE_PAGE_ABANDON + segment->thread_id = 0; // huge segments are immediately abandoned + #endif + + // for huge pages we initialize the block_size as we may + // overallocate to accommodate large alignments. + size_t psize; + uint8_t* start = _mi_segment_page_start(segment, page, &psize); + page->block_size = psize; + mi_assert_internal(page->is_huge); + + // decommit the part of the prefix of a page that will not be used; this can be quite large (close to MI_SEGMENT_SIZE) + if (page_alignment > 0 && segment->allow_decommit) { + uint8_t* aligned_p = (uint8_t*)_mi_align_up((uintptr_t)start, page_alignment); + mi_assert_internal(_mi_is_aligned(aligned_p, page_alignment)); + mi_assert_internal(psize - (aligned_p - start) >= size); + uint8_t* decommit_start = start + sizeof(mi_block_t); // for the free list + ptrdiff_t decommit_size = aligned_p - decommit_start; + _mi_os_reset(decommit_start, decommit_size); // note: cannot use segment_decommit on huge segments + } + + return page; +} + +#if MI_HUGE_PAGE_ABANDON +// free huge block from another thread +void _mi_segment_huge_page_free(mi_segment_t* segment, mi_page_t* page, mi_block_t* block) { + // huge page segments are always abandoned and can be freed immediately by any thread + mi_assert_internal(segment->kind==MI_SEGMENT_HUGE); + mi_assert_internal(segment == _mi_page_segment(page)); + mi_assert_internal(mi_atomic_load_relaxed(&segment->thread_id)==0); + + // claim it and free + mi_heap_t* heap = mi_heap_get_default(); // issue #221; don't use the internal get_default_heap as we need to ensure the thread is initialized. + // paranoia: if this it the last reference, the cas should always succeed + size_t expected_tid = 0; + if (mi_atomic_cas_strong_acq_rel(&segment->thread_id, &expected_tid, heap->thread_id)) { + mi_block_set_next(page, block, page->free); + page->free = block; + page->used--; + page->is_zero_init = false; + mi_assert(page->used == 0); + mi_tld_t* tld = heap->tld; + _mi_segment_page_free(page, true, &tld->segments); + } +#if (MI_DEBUG!=0) + else { + mi_assert_internal(false); + } +#endif +} + +#else +// reset memory of a huge block from another thread +void _mi_segment_huge_page_reset(mi_segment_t* segment, mi_page_t* page, mi_block_t* block) { + MI_UNUSED(page); + mi_assert_internal(segment->kind == MI_SEGMENT_HUGE); + mi_assert_internal(segment == _mi_page_segment(page)); + mi_assert_internal(page->used == 1); // this is called just before the free + mi_assert_internal(page->free == NULL); + if (segment->allow_decommit) { + size_t csize = mi_usable_size(block); + if (csize > sizeof(mi_block_t)) { + csize = csize - sizeof(mi_block_t); + uint8_t* p = (uint8_t*)block + sizeof(mi_block_t); + _mi_os_reset(p, csize); // note: cannot use segment_decommit on huge segments + } + } +} +#endif + +/* ----------------------------------------------------------- + Page allocation and free +----------------------------------------------------------- */ +mi_page_t* _mi_segment_page_alloc(mi_heap_t* heap, size_t block_size, size_t page_alignment, mi_segments_tld_t* tld) { + mi_page_t* page; + if mi_unlikely(page_alignment > MI_BLOCK_ALIGNMENT_MAX) { + mi_assert_internal(_mi_is_power_of_two(page_alignment)); + mi_assert_internal(page_alignment >= MI_SEGMENT_SIZE); + if (page_alignment < MI_SEGMENT_SIZE) { page_alignment = MI_SEGMENT_SIZE; } + page = mi_segment_huge_page_alloc(block_size,page_alignment,heap->arena_id,tld); + } + else if (block_size <= MI_SMALL_OBJ_SIZE_MAX) { + page = mi_segments_page_alloc(heap,MI_PAGE_SMALL,block_size,block_size,tld); + } + else if (block_size <= MI_MEDIUM_OBJ_SIZE_MAX) { + page = mi_segments_page_alloc(heap,MI_PAGE_MEDIUM,MI_MEDIUM_PAGE_SIZE,block_size,tld); + } + else if (block_size <= MI_LARGE_OBJ_SIZE_MAX) { + page = mi_segments_page_alloc(heap,MI_PAGE_LARGE,block_size,block_size,tld); + } + else { + page = mi_segment_huge_page_alloc(block_size,page_alignment,heap->arena_id,tld); + } + mi_assert_internal(page == NULL || _mi_heap_memid_is_suitable(heap, _mi_page_segment(page)->memid)); + mi_assert_expensive(page == NULL || mi_segment_is_valid(_mi_page_segment(page),tld)); + mi_assert_internal(page == NULL || _mi_page_segment(page)->subproc == tld->subproc); + return page; +} + + +/* ----------------------------------------------------------- + Visit blocks in a segment (only used for abandoned segments) +----------------------------------------------------------- */ + +static bool mi_segment_visit_page(mi_page_t* page, bool visit_blocks, mi_block_visit_fun* visitor, void* arg) { + mi_heap_area_t area; + _mi_heap_area_init(&area, page); + if (!visitor(NULL, &area, NULL, area.block_size, arg)) return false; + if (visit_blocks) { + return _mi_heap_area_visit_blocks(&area, page, visitor, arg); + } + else { + return true; + } +} + +bool _mi_segment_visit_blocks(mi_segment_t* segment, int heap_tag, bool visit_blocks, mi_block_visit_fun* visitor, void* arg) { + const mi_slice_t* end; + mi_slice_t* slice = mi_slices_start_iterate(segment, &end); + while (slice < end) { + if (mi_slice_is_used(slice)) { + mi_page_t* const page = mi_slice_to_page(slice); + if (heap_tag < 0 || (int)page->heap_tag == heap_tag) { + if (!mi_segment_visit_page(page, visit_blocks, visitor, arg)) return false; + } + } + slice = slice + slice->slice_count; + } + return true; +} diff --git a/compat/mimalloc/stats.c b/compat/mimalloc/stats.c new file mode 100644 index 00000000000000..dec74f70c1fffd --- /dev/null +++ b/compat/mimalloc/stats.c @@ -0,0 +1,629 @@ +/* ---------------------------------------------------------------------------- +Copyright (c) 2018-2021, Microsoft Research, Daan Leijen +This is free software; you can redistribute it and/or modify it under the +terms of the MIT license. A copy of the license can be found in the file +"LICENSE" at the root of this distribution. +-----------------------------------------------------------------------------*/ +#include "mimalloc.h" +#include "mimalloc/internal.h" +#include "mimalloc/atomic.h" +#include "mimalloc/prim.h" + +#include // memset + +#if defined(_MSC_VER) && (_MSC_VER < 1920) +#pragma warning(disable:4204) // non-constant aggregate initializer +#endif + +/* ----------------------------------------------------------- + Statistics operations +----------------------------------------------------------- */ + +static bool mi_is_in_main(void* stat) { + return ((uint8_t*)stat >= (uint8_t*)&_mi_stats_main + && (uint8_t*)stat < ((uint8_t*)&_mi_stats_main + sizeof(mi_stats_t))); +} + +static void mi_stat_update(mi_stat_count_t* stat, int64_t amount) { + if (amount == 0) return; + if mi_unlikely(mi_is_in_main(stat)) + { + // add atomically (for abandoned pages) + int64_t current = mi_atomic_addi64_relaxed(&stat->current, amount); + // if (stat == &_mi_stats_main.committed) { mi_assert_internal(current + amount >= 0); }; + mi_atomic_maxi64_relaxed(&stat->peak, current + amount); + if (amount > 0) { + mi_atomic_addi64_relaxed(&stat->total,amount); + } + } + else { + // add thread local + stat->current += amount; + if (stat->current > stat->peak) { stat->peak = stat->current; } + if (amount > 0) { stat->total += amount; } + } +} + +void _mi_stat_counter_increase(mi_stat_counter_t* stat, size_t amount) { + if (mi_is_in_main(stat)) { + mi_atomic_addi64_relaxed( &stat->total, (int64_t)amount ); + } + else { + stat->total += amount; + } +} + +void _mi_stat_increase(mi_stat_count_t* stat, size_t amount) { + mi_stat_update(stat, (int64_t)amount); +} + +void _mi_stat_decrease(mi_stat_count_t* stat, size_t amount) { + mi_stat_update(stat, -((int64_t)amount)); +} + + +static void mi_stat_adjust(mi_stat_count_t* stat, int64_t amount) { + if (amount == 0) return; + if mi_unlikely(mi_is_in_main(stat)) + { + // adjust atomically + mi_atomic_addi64_relaxed(&stat->current, amount); + mi_atomic_addi64_relaxed(&stat->total,amount); + } + else { + // adjust local + stat->current += amount; + stat->total += amount; + } +} + +void _mi_stat_adjust_decrease(mi_stat_count_t* stat, size_t amount) { + mi_stat_adjust(stat, -((int64_t)amount)); +} + + +// must be thread safe as it is called from stats_merge +static void mi_stat_count_add_mt(mi_stat_count_t* stat, const mi_stat_count_t* src) { + if (stat==src) return; + mi_atomic_void_addi64_relaxed(&stat->total, &src->total); + mi_atomic_void_addi64_relaxed(&stat->current, &src->current); + // peak scores do really not work across threads .. we just add them + mi_atomic_void_addi64_relaxed( &stat->peak, &src->peak); + // or, take the max? + // mi_atomic_maxi64_relaxed(&stat->peak, src->peak); +} + +static void mi_stat_counter_add_mt(mi_stat_counter_t* stat, const mi_stat_counter_t* src) { + if (stat==src) return; + mi_atomic_void_addi64_relaxed(&stat->total, &src->total); +} + +#define MI_STAT_COUNT(stat) mi_stat_count_add_mt(&stats->stat, &src->stat); +#define MI_STAT_COUNTER(stat) mi_stat_counter_add_mt(&stats->stat, &src->stat); + +// must be thread safe as it is called from stats_merge +static void mi_stats_add(mi_stats_t* stats, const mi_stats_t* src) { + if (stats==src) return; + + // copy all fields + MI_STAT_FIELDS() + + #if MI_STAT>1 + for (size_t i = 0; i <= MI_BIN_HUGE; i++) { + mi_stat_count_add_mt(&stats->malloc_bins[i], &src->malloc_bins[i]); + } + #endif + for (size_t i = 0; i <= MI_BIN_HUGE; i++) { + mi_stat_count_add_mt(&stats->page_bins[i], &src->page_bins[i]); + } +} + +#undef MI_STAT_COUNT +#undef MI_STAT_COUNTER + +/* ----------------------------------------------------------- + Display statistics +----------------------------------------------------------- */ + +// unit > 0 : size in binary bytes +// unit == 0: count as decimal +// unit < 0 : count in binary +static void mi_printf_amount(int64_t n, int64_t unit, mi_output_fun* out, void* arg, const char* fmt) { + char buf[32]; buf[0] = 0; + int len = 32; + const char* suffix = (unit <= 0 ? " " : "B"); + const int64_t base = (unit == 0 ? 1000 : 1024); + if (unit>0) n *= unit; + + const int64_t pos = (n < 0 ? -n : n); + if (pos < base) { + if (n!=1 || suffix[0] != 'B') { // skip printing 1 B for the unit column + _mi_snprintf(buf, len, "%lld %-3s", (long long)n, (n==0 ? "" : suffix)); + } + } + else { + int64_t divider = base; + const char* magnitude = "K"; + if (pos >= divider*base) { divider *= base; magnitude = "M"; } + if (pos >= divider*base) { divider *= base; magnitude = "G"; } + const int64_t tens = (n / (divider/10)); + const long whole = (long)(tens/10); + const long frac1 = (long)(tens%10); + char unitdesc[8]; + _mi_snprintf(unitdesc, 8, "%s%s%s", magnitude, (base==1024 ? "i" : ""), suffix); + _mi_snprintf(buf, len, "%ld.%ld %-3s", whole, (frac1 < 0 ? -frac1 : frac1), unitdesc); + } + _mi_fprintf(out, arg, (fmt==NULL ? "%12s" : fmt), buf); +} + + +static void mi_print_amount(int64_t n, int64_t unit, mi_output_fun* out, void* arg) { + mi_printf_amount(n,unit,out,arg,NULL); +} + +static void mi_print_count(int64_t n, int64_t unit, mi_output_fun* out, void* arg) { + if (unit==1) _mi_fprintf(out, arg, "%12s"," "); + else mi_print_amount(n,0,out,arg); +} + +static void mi_stat_print_ex(const mi_stat_count_t* stat, const char* msg, int64_t unit, mi_output_fun* out, void* arg, const char* notok ) { + _mi_fprintf(out, arg,"%10s:", msg); + if (unit != 0) { + if (unit > 0) { + mi_print_amount(stat->peak, unit, out, arg); + mi_print_amount(stat->total, unit, out, arg); + // mi_print_amount(stat->freed, unit, out, arg); + mi_print_amount(stat->current, unit, out, arg); + mi_print_amount(unit, 1, out, arg); + mi_print_count(stat->total, unit, out, arg); + } + else { + mi_print_amount(stat->peak, -1, out, arg); + mi_print_amount(stat->total, -1, out, arg); + // mi_print_amount(stat->freed, -1, out, arg); + mi_print_amount(stat->current, -1, out, arg); + if (unit == -1) { + _mi_fprintf(out, arg, "%24s", ""); + } + else { + mi_print_amount(-unit, 1, out, arg); + mi_print_count((stat->total / -unit), 0, out, arg); + } + } + if (stat->current != 0) { + _mi_fprintf(out, arg, " "); + _mi_fprintf(out, arg, (notok == NULL ? "not all freed" : notok)); + _mi_fprintf(out, arg, "\n"); + } + else { + _mi_fprintf(out, arg, " ok\n"); + } + } + else { + mi_print_amount(stat->peak, 1, out, arg); + mi_print_amount(stat->total, 1, out, arg); + _mi_fprintf(out, arg, "%11s", " "); // no freed + mi_print_amount(stat->current, 1, out, arg); + _mi_fprintf(out, arg, "\n"); + } +} + +static void mi_stat_print(const mi_stat_count_t* stat, const char* msg, int64_t unit, mi_output_fun* out, void* arg) { + mi_stat_print_ex(stat, msg, unit, out, arg, NULL); +} + +static void mi_stat_peak_print(const mi_stat_count_t* stat, const char* msg, int64_t unit, mi_output_fun* out, void* arg) { + _mi_fprintf(out, arg, "%10s:", msg); + mi_print_amount(stat->peak, unit, out, arg); + _mi_fprintf(out, arg, "\n"); +} + +static void mi_stat_total_print(const mi_stat_count_t* stat, const char* msg, int64_t unit, mi_output_fun* out, void* arg) { + _mi_fprintf(out, arg, "%10s:", msg); + _mi_fprintf(out, arg, "%12s", " "); // no peak + mi_print_amount(stat->total, unit, out, arg); + _mi_fprintf(out, arg, "\n"); +} + +static void mi_stat_counter_print(const mi_stat_counter_t* stat, const char* msg, mi_output_fun* out, void* arg ) { + _mi_fprintf(out, arg, "%10s:", msg); + mi_print_amount(stat->total, -1, out, arg); + _mi_fprintf(out, arg, "\n"); +} + + +static void mi_stat_counter_print_avg(const mi_stat_counter_t* stat, const char* msg, mi_output_fun* out, void* arg) { + const int64_t avg_tens = (stat->total == 0 ? 0 : (stat->total*10 / stat->total)); + const long avg_whole = (long)(avg_tens/10); + const long avg_frac1 = (long)(avg_tens%10); + _mi_fprintf(out, arg, "%10s: %5ld.%ld avg\n", msg, avg_whole, avg_frac1); +} + + +static void mi_print_header(mi_output_fun* out, void* arg ) { + _mi_fprintf(out, arg, "%10s: %11s %11s %11s %11s %11s\n", "heap stats", "peak ", "total ", "current ", "block ", "total# "); +} + +#if MI_STAT>1 +static void mi_stats_print_bins(const mi_stat_count_t* bins, size_t max, const char* fmt, mi_output_fun* out, void* arg) { + bool found = false; + char buf[64]; + for (size_t i = 0; i <= max; i++) { + if (bins[i].total > 0) { + found = true; + int64_t unit = _mi_bin_size((uint8_t)i); + _mi_snprintf(buf, 64, "%s %3lu", fmt, (long)i); + mi_stat_print(&bins[i], buf, unit, out, arg); + } + } + if (found) { + _mi_fprintf(out, arg, "\n"); + mi_print_header(out, arg); + } +} +#endif + + + +//------------------------------------------------------------ +// Use an output wrapper for line-buffered output +// (which is nice when using loggers etc.) +//------------------------------------------------------------ +typedef struct buffered_s { + mi_output_fun* out; // original output function + void* arg; // and state + char* buf; // local buffer of at least size `count+1` + size_t used; // currently used chars `used <= count` + size_t count; // total chars available for output +} buffered_t; + +static void mi_buffered_flush(buffered_t* buf) { + buf->buf[buf->used] = 0; + _mi_fputs(buf->out, buf->arg, NULL, buf->buf); + buf->used = 0; +} + +static void mi_cdecl mi_buffered_out(const char* msg, void* arg) { + buffered_t* buf = (buffered_t*)arg; + if (msg==NULL || buf==NULL) return; + for (const char* src = msg; *src != 0; src++) { + char c = *src; + if (buf->used >= buf->count) mi_buffered_flush(buf); + mi_assert_internal(buf->used < buf->count); + buf->buf[buf->used++] = c; + if (c == '\n') mi_buffered_flush(buf); + } +} + +//------------------------------------------------------------ +// Print statistics +//------------------------------------------------------------ + +static void _mi_stats_print(mi_stats_t* stats, mi_output_fun* out0, void* arg0) mi_attr_noexcept { + // wrap the output function to be line buffered + char buf[256]; + buffered_t buffer = { out0, arg0, NULL, 0, 255 }; + buffer.buf = buf; + mi_output_fun* out = &mi_buffered_out; + void* arg = &buffer; + + // and print using that + mi_print_header(out,arg); + #if MI_STAT>1 + mi_stats_print_bins(stats->malloc_bins, MI_BIN_HUGE, "bin",out,arg); + #endif + #if MI_STAT + mi_stat_print(&stats->malloc_normal, "binned", (stats->malloc_normal_count.total == 0 ? 1 : -1), out, arg); + // mi_stat_print(&stats->malloc_large, "large", (stats->malloc_large_count.total == 0 ? 1 : -1), out, arg); + mi_stat_print(&stats->malloc_huge, "huge", (stats->malloc_huge_count.total == 0 ? 1 : -1), out, arg); + mi_stat_count_t total = { 0,0,0 }; + mi_stat_count_add_mt(&total, &stats->malloc_normal); + // mi_stat_count_add(&total, &stats->malloc_large); + mi_stat_count_add_mt(&total, &stats->malloc_huge); + mi_stat_print_ex(&total, "total", 1, out, arg, ""); + #endif + #if MI_STAT>1 + mi_stat_total_print(&stats->malloc_requested, "malloc req", 1, out, arg); + _mi_fprintf(out, arg, "\n"); + #endif + mi_stat_print_ex(&stats->reserved, "reserved", 1, out, arg, ""); + mi_stat_print_ex(&stats->committed, "committed", 1, out, arg, ""); + mi_stat_peak_print(&stats->reset, "reset", 1, out, arg ); + mi_stat_peak_print(&stats->purged, "purged", 1, out, arg ); + mi_stat_print_ex(&stats->page_committed, "touched", 1, out, arg, ""); + mi_stat_print(&stats->segments, "segments", -1, out, arg); + mi_stat_print(&stats->segments_abandoned, "-abandoned", -1, out, arg); + mi_stat_print(&stats->segments_cache, "-cached", -1, out, arg); + mi_stat_print(&stats->pages, "pages", -1, out, arg); + mi_stat_print(&stats->pages_abandoned, "-abandoned", -1, out, arg); + mi_stat_counter_print(&stats->pages_extended, "-extended", out, arg); + mi_stat_counter_print(&stats->pages_retire, "-retire", out, arg); + mi_stat_counter_print(&stats->arena_count, "arenas", out, arg); + // mi_stat_counter_print(&stats->arena_crossover_count, "-crossover", out, arg); + mi_stat_counter_print(&stats->arena_rollback_count, "-rollback", out, arg); + mi_stat_counter_print(&stats->mmap_calls, "mmaps", out, arg); + mi_stat_counter_print(&stats->commit_calls, "commits", out, arg); + mi_stat_counter_print(&stats->reset_calls, "resets", out, arg); + mi_stat_counter_print(&stats->purge_calls, "purges", out, arg); + mi_stat_counter_print(&stats->malloc_guarded_count, "guarded", out, arg); + mi_stat_print(&stats->threads, "threads", -1, out, arg); + mi_stat_counter_print_avg(&stats->page_searches, "searches", out, arg); + _mi_fprintf(out, arg, "%10s: %5zu\n", "numa nodes", _mi_os_numa_node_count()); + + size_t elapsed; + size_t user_time; + size_t sys_time; + size_t current_rss; + size_t peak_rss; + size_t current_commit; + size_t peak_commit; + size_t page_faults; + mi_process_info(&elapsed, &user_time, &sys_time, ¤t_rss, &peak_rss, ¤t_commit, &peak_commit, &page_faults); + _mi_fprintf(out, arg, "%10s: %5ld.%03ld s\n", "elapsed", elapsed/1000, elapsed%1000); + _mi_fprintf(out, arg, "%10s: user: %ld.%03ld s, system: %ld.%03ld s, faults: %lu, rss: ", "process", + user_time/1000, user_time%1000, sys_time/1000, sys_time%1000, (unsigned long)page_faults ); + mi_printf_amount((int64_t)peak_rss, 1, out, arg, "%s"); + if (peak_commit > 0) { + _mi_fprintf(out, arg, ", commit: "); + mi_printf_amount((int64_t)peak_commit, 1, out, arg, "%s"); + } + _mi_fprintf(out, arg, "\n"); +} + +static mi_msecs_t mi_process_start; // = 0 + +static mi_stats_t* mi_stats_get_default(void) { + mi_heap_t* heap = mi_heap_get_default(); + return &heap->tld->stats; +} + +static void mi_stats_merge_from(mi_stats_t* stats) { + if (stats != &_mi_stats_main) { + mi_stats_add(&_mi_stats_main, stats); + memset(stats, 0, sizeof(mi_stats_t)); + } +} + +void mi_stats_reset(void) mi_attr_noexcept { + mi_stats_t* stats = mi_stats_get_default(); + if (stats != &_mi_stats_main) { memset(stats, 0, sizeof(mi_stats_t)); } + memset(&_mi_stats_main, 0, sizeof(mi_stats_t)); + if (mi_process_start == 0) { mi_process_start = _mi_clock_start(); }; +} + +void mi_stats_merge(void) mi_attr_noexcept { + mi_stats_merge_from( mi_stats_get_default() ); +} + +void _mi_stats_done(mi_stats_t* stats) { // called from `mi_thread_done` + mi_stats_merge_from(stats); +} + +void mi_stats_print_out(mi_output_fun* out, void* arg) mi_attr_noexcept { + mi_stats_merge_from(mi_stats_get_default()); + _mi_stats_print(&_mi_stats_main, out, arg); +} + +void mi_stats_print(void* out) mi_attr_noexcept { + // for compatibility there is an `out` parameter (which can be `stdout` or `stderr`) + mi_stats_print_out((mi_output_fun*)out, NULL); +} + +void mi_thread_stats_print_out(mi_output_fun* out, void* arg) mi_attr_noexcept { + _mi_stats_print(mi_stats_get_default(), out, arg); +} + + +// ---------------------------------------------------------------- +// Basic timer for convenience; use milli-seconds to avoid doubles +// ---------------------------------------------------------------- + +static mi_msecs_t mi_clock_diff; + +mi_msecs_t _mi_clock_now(void) { + return _mi_prim_clock_now(); +} + +mi_msecs_t _mi_clock_start(void) { + if (mi_clock_diff == 0.0) { + mi_msecs_t t0 = _mi_clock_now(); + mi_clock_diff = _mi_clock_now() - t0; + } + return _mi_clock_now(); +} + +mi_msecs_t _mi_clock_end(mi_msecs_t start) { + mi_msecs_t end = _mi_clock_now(); + return (end - start - mi_clock_diff); +} + + +// -------------------------------------------------------- +// Basic process statistics +// -------------------------------------------------------- + +mi_decl_export void mi_process_info(size_t* elapsed_msecs, size_t* user_msecs, size_t* system_msecs, size_t* current_rss, size_t* peak_rss, size_t* current_commit, size_t* peak_commit, size_t* page_faults) mi_attr_noexcept +{ + mi_process_info_t pinfo; + _mi_memzero_var(pinfo); + pinfo.elapsed = _mi_clock_end(mi_process_start); + pinfo.current_commit = (size_t)(mi_atomic_loadi64_relaxed((_Atomic(int64_t)*)&_mi_stats_main.committed.current)); + pinfo.peak_commit = (size_t)(mi_atomic_loadi64_relaxed((_Atomic(int64_t)*)&_mi_stats_main.committed.peak)); + pinfo.current_rss = pinfo.current_commit; + pinfo.peak_rss = pinfo.peak_commit; + pinfo.utime = 0; + pinfo.stime = 0; + pinfo.page_faults = 0; + + _mi_prim_process_info(&pinfo); + + if (elapsed_msecs!=NULL) *elapsed_msecs = (pinfo.elapsed < 0 ? 0 : (pinfo.elapsed < (mi_msecs_t)PTRDIFF_MAX ? (size_t)pinfo.elapsed : PTRDIFF_MAX)); + if (user_msecs!=NULL) *user_msecs = (pinfo.utime < 0 ? 0 : (pinfo.utime < (mi_msecs_t)PTRDIFF_MAX ? (size_t)pinfo.utime : PTRDIFF_MAX)); + if (system_msecs!=NULL) *system_msecs = (pinfo.stime < 0 ? 0 : (pinfo.stime < (mi_msecs_t)PTRDIFF_MAX ? (size_t)pinfo.stime : PTRDIFF_MAX)); + if (current_rss!=NULL) *current_rss = pinfo.current_rss; + if (peak_rss!=NULL) *peak_rss = pinfo.peak_rss; + if (current_commit!=NULL) *current_commit = pinfo.current_commit; + if (peak_commit!=NULL) *peak_commit = pinfo.peak_commit; + if (page_faults!=NULL) *page_faults = pinfo.page_faults; +} + + +// -------------------------------------------------------- +// Return statistics +// -------------------------------------------------------- + +void mi_stats_get(size_t stats_size, mi_stats_t* stats) mi_attr_noexcept { + if (stats == NULL || stats_size == 0) return; + _mi_memzero(stats, stats_size); + const size_t size = (stats_size > sizeof(mi_stats_t) ? sizeof(mi_stats_t) : stats_size); + _mi_memcpy(stats, &_mi_stats_main, size); + stats->version = MI_STAT_VERSION; +} + + +// -------------------------------------------------------- +// Statics in json format +// -------------------------------------------------------- + +typedef struct mi_heap_buf_s { + char* buf; + size_t size; + size_t used; + bool can_realloc; +} mi_heap_buf_t; + +static bool mi_heap_buf_expand(mi_heap_buf_t* hbuf) { + if (hbuf==NULL) return false; + if (hbuf->buf != NULL && hbuf->size>0) { + hbuf->buf[hbuf->size-1] = 0; + } + if (hbuf->size > SIZE_MAX/2 || !hbuf->can_realloc) return false; + const size_t newsize = (hbuf->size == 0 ? 2*MI_KiB : 2*hbuf->size); + char* const newbuf = (char*)mi_rezalloc(hbuf->buf, newsize); + if (newbuf == NULL) return false; + hbuf->buf = newbuf; + hbuf->size = newsize; + return true; +} + +static void mi_heap_buf_print(mi_heap_buf_t* hbuf, const char* msg) { + if (msg==NULL || hbuf==NULL) return; + if (hbuf->used + 1 >= hbuf->size && !hbuf->can_realloc) return; + for (const char* src = msg; *src != 0; src++) { + char c = *src; + if (hbuf->used + 1 >= hbuf->size) { + if (!mi_heap_buf_expand(hbuf)) return; + } + mi_assert_internal(hbuf->used < hbuf->size); + hbuf->buf[hbuf->used++] = c; + } + mi_assert_internal(hbuf->used < hbuf->size); + hbuf->buf[hbuf->used] = 0; +} + +static void mi_heap_buf_print_count_bin(mi_heap_buf_t* hbuf, const char* prefix, mi_stat_count_t* stat, size_t bin, bool add_comma) { + const size_t binsize = _mi_bin_size(bin); + const size_t pagesize = (binsize <= MI_SMALL_OBJ_SIZE_MAX ? MI_SMALL_PAGE_SIZE : + (binsize <= MI_MEDIUM_OBJ_SIZE_MAX ? MI_MEDIUM_PAGE_SIZE : + #if MI_LARGE_PAGE_SIZE + (binsize <= MI_LARGE_OBJ_SIZE_MAX ? MI_LARGE_PAGE_SIZE : 0) + #else + 0 + #endif + )); + char buf[128]; + _mi_snprintf(buf, 128, "%s{ \"total\": %lld, \"peak\": %lld, \"current\": %lld, \"block_size\": %zu, \"page_size\": %zu }%s\n", prefix, stat->total, stat->peak, stat->current, binsize, pagesize, (add_comma ? "," : "")); + buf[127] = 0; + mi_heap_buf_print(hbuf, buf); +} + +static void mi_heap_buf_print_count(mi_heap_buf_t* hbuf, const char* prefix, mi_stat_count_t* stat, bool add_comma) { + char buf[128]; + _mi_snprintf(buf, 128, "%s{ \"total\": %lld, \"peak\": %lld, \"current\": %lld }%s\n", prefix, stat->total, stat->peak, stat->current, (add_comma ? "," : "")); + buf[127] = 0; + mi_heap_buf_print(hbuf, buf); +} + +static void mi_heap_buf_print_count_value(mi_heap_buf_t* hbuf, const char* name, mi_stat_count_t* stat) { + char buf[128]; + _mi_snprintf(buf, 128, " \"%s\": ", name); + buf[127] = 0; + mi_heap_buf_print(hbuf, buf); + mi_heap_buf_print_count(hbuf, "", stat, true); +} + +static void mi_heap_buf_print_value(mi_heap_buf_t* hbuf, const char* name, int64_t val) { + char buf[128]; + _mi_snprintf(buf, 128, " \"%s\": %lld,\n", name, val); + buf[127] = 0; + mi_heap_buf_print(hbuf, buf); +} + +static void mi_heap_buf_print_size(mi_heap_buf_t* hbuf, const char* name, size_t val, bool add_comma) { + char buf[128]; + _mi_snprintf(buf, 128, " \"%s\": %zu%s\n", name, val, (add_comma ? "," : "")); + buf[127] = 0; + mi_heap_buf_print(hbuf, buf); +} + +static void mi_heap_buf_print_counter_value(mi_heap_buf_t* hbuf, const char* name, mi_stat_counter_t* stat) { + mi_heap_buf_print_value(hbuf, name, stat->total); +} + +#define MI_STAT_COUNT(stat) mi_heap_buf_print_count_value(&hbuf, #stat, &stats->stat); +#define MI_STAT_COUNTER(stat) mi_heap_buf_print_counter_value(&hbuf, #stat, &stats->stat); + +char* mi_stats_get_json(size_t output_size, char* output_buf) mi_attr_noexcept { + mi_heap_buf_t hbuf = { NULL, 0, 0, true }; + if (output_size > 0 && output_buf != NULL) { + _mi_memzero(output_buf, output_size); + hbuf.buf = output_buf; + hbuf.size = output_size; + hbuf.can_realloc = false; + } + else { + if (!mi_heap_buf_expand(&hbuf)) return NULL; + } + mi_heap_buf_print(&hbuf, "{\n"); + mi_heap_buf_print_value(&hbuf, "version", MI_STAT_VERSION); + mi_heap_buf_print_value(&hbuf, "mimalloc_version", MI_MALLOC_VERSION); + + // process info + mi_heap_buf_print(&hbuf, " \"process\": {\n"); + size_t elapsed; + size_t user_time; + size_t sys_time; + size_t current_rss; + size_t peak_rss; + size_t current_commit; + size_t peak_commit; + size_t page_faults; + mi_process_info(&elapsed, &user_time, &sys_time, ¤t_rss, &peak_rss, ¤t_commit, &peak_commit, &page_faults); + mi_heap_buf_print_size(&hbuf, "elapsed_msecs", elapsed, true); + mi_heap_buf_print_size(&hbuf, "user_msecs", user_time, true); + mi_heap_buf_print_size(&hbuf, "system_msecs", sys_time, true); + mi_heap_buf_print_size(&hbuf, "page_faults", page_faults, true); + mi_heap_buf_print_size(&hbuf, "rss_current", current_rss, true); + mi_heap_buf_print_size(&hbuf, "rss_peak", peak_rss, true); + mi_heap_buf_print_size(&hbuf, "commit_current", current_commit, true); + mi_heap_buf_print_size(&hbuf, "commit_peak", peak_commit, false); + mi_heap_buf_print(&hbuf, " },\n"); + + // statistics + mi_stats_t* stats = &_mi_stats_main; + MI_STAT_FIELDS() + + // size bins + mi_heap_buf_print(&hbuf, " \"malloc_bins\": [\n"); + for (size_t i = 0; i <= MI_BIN_HUGE; i++) { + mi_heap_buf_print_count_bin(&hbuf, " ", &stats->malloc_bins[i], i, i!=MI_BIN_HUGE); + } + mi_heap_buf_print(&hbuf, " ],\n"); + mi_heap_buf_print(&hbuf, " \"page_bins\": [\n"); + for (size_t i = 0; i <= MI_BIN_HUGE; i++) { + mi_heap_buf_print_count_bin(&hbuf, " ", &stats->page_bins[i], i, i!=MI_BIN_HUGE); + } + mi_heap_buf_print(&hbuf, " ]\n"); + mi_heap_buf_print(&hbuf, "}\n"); + return hbuf.buf; +} From 9f7db14c2197966ef2c0e9a798dc70b417a12328 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Wed, 4 Jun 2025 15:25:51 +0200 Subject: [PATCH 025/250] mimalloc: avoid excessive memory retention In commit cfe73e9d (wip: merging from upstream, 2024-12-25), mimalloc introduced a bug where memory is retained excessively. This leads to a problem e.g. when fetching Git for Windows' `main` branch using the i686 variant of Git for Windows because it simply runs out of address space. Fix this as suggested in https://github.com/microsoft/mimalloc/issues/1025#issuecomment-2768863283 Signed-off-by: Johannes Schindelin --- compat/mimalloc/os.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/compat/mimalloc/os.c b/compat/mimalloc/os.c index 2472b8030b3c61..2a3dd09b89d4d9 100644 --- a/compat/mimalloc/os.c +++ b/compat/mimalloc/os.c @@ -166,7 +166,7 @@ static void mi_os_prim_free(void* addr, size_t size, size_t commit_size) { void _mi_os_free_ex(void* addr, size_t size, bool still_committed, mi_memid_t memid) { if (mi_memkind_is_os(memid.memkind)) { size_t csize = memid.mem.os.size; - if (csize==0) { _mi_os_good_alloc_size(size); } + if (csize==0) { csize = _mi_os_good_alloc_size(size); } size_t commit_size = (still_committed ? csize : 0); void* base = addr; // different base? (due to alignment) From 0f70bdba9356da8903a6875a5063d019a24c626b Mon Sep 17 00:00:00 2001 From: Sverre Rabbelier Date: Sun, 24 Jul 2011 00:06:00 +0200 Subject: [PATCH 026/250] remote-helper: check helper status after import/export Signed-off-by: Johannes Schindelin Signed-off-by: Sverre Rabbelier --- t/t5801-remote-helpers.sh | 2 +- transport-helper.c | 15 +++++++++++++++ 2 files changed, 16 insertions(+), 1 deletion(-) diff --git a/t/t5801-remote-helpers.sh b/t/t5801-remote-helpers.sh index d21877150ed82e..3917da47276825 100755 --- a/t/t5801-remote-helpers.sh +++ b/t/t5801-remote-helpers.sh @@ -262,7 +262,7 @@ test_expect_success 'push update refs failure' ' echo "update fail" >>file && git commit -a -m "update fail" && git rev-parse --verify testgit/origin/heads/update >expect && - test_expect_code 1 env GIT_REMOTE_TESTGIT_FAILURE="non-fast forward" \ + test_must_fail env GIT_REMOTE_TESTGIT_FAILURE="non-fast forward" \ git push origin update && git rev-parse --verify testgit/origin/heads/update >actual && test_cmp expect actual diff --git a/transport-helper.c b/transport-helper.c index 3e3f9e66c39f02..7545e50d27ed63 100644 --- a/transport-helper.c +++ b/transport-helper.c @@ -505,6 +505,19 @@ static int get_exporter(struct transport *transport, return start_command(fastexport); } +static void check_helper_status(struct helper_data *data) +{ + int pid, status; + + pid = waitpid(data->helper->pid, &status, WNOHANG); + if (pid < 0) + die("Could not retrieve status of remote helper '%s'", + data->name); + if (pid > 0 && WIFEXITED(status)) + die("Remote helper '%s' died with %d", + data->name, WEXITSTATUS(status)); +} + static int fetch_with_import(struct transport *transport, int nr_heads, struct ref **to_fetch) { @@ -541,6 +554,7 @@ static int fetch_with_import(struct transport *transport, if (finish_command(&fastimport)) die(_("error while running fast-import")); + check_helper_status(data); /* * The fast-import stream of a remote helper that advertises @@ -1160,6 +1174,7 @@ static int push_refs_with_export(struct transport *transport, if (finish_command(&exporter)) die(_("error while running fast-export")); + check_helper_status(data); if (push_update_refs_status(data, remote_refs, flags)) return 1; From 42dd6c7983539d6a69ad448fd6793bd6126954ad Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Tue, 18 Apr 2017 12:09:08 +0200 Subject: [PATCH 027/250] mingw: demonstrate a problem with certain absolute paths On Windows, there are several categories of absolute paths. One such category starts with a backslash and is implicitly relative to the drive associated with the current working directory. Example: c: git clone https://github.com/git-for-windows/git \G4W should clone into C:\G4W. There is currently a problem with that, in that mingw_mktemp() does not expect the _wmktemp() function to prefix the absolute path with the drive prefix, and as a consequence, the resulting path does not fit into the originally-passed string buffer. The symptom is a "Result too large" error. Reported by Juan Carlos Arevalo Baeza. Signed-off-by: Johannes Schindelin --- t/t5580-unc-paths.sh | 19 ++++++++++++++----- 1 file changed, 14 insertions(+), 5 deletions(-) diff --git a/t/t5580-unc-paths.sh b/t/t5580-unc-paths.sh index 65ef1a3628ee94..2c30f3b51d3183 100755 --- a/t/t5580-unc-paths.sh +++ b/t/t5580-unc-paths.sh @@ -20,14 +20,11 @@ fi UNCPATH="$(winpwd)" case "$UNCPATH" in [A-Z]:*) + WITHOUTDRIVE="${UNCPATH#?:}" # Use administrative share e.g. \\localhost\C$\git-sdk-64\usr\src\git # (we use forward slashes here because MSYS2 and Git accept them, and # they are easier on the eyes) - UNCPATH="//localhost/${UNCPATH%%:*}\$/${UNCPATH#?:}" - test -d "$UNCPATH" || { - skip_all='could not access administrative share; skipping' - test_done - } + UNCPATH="//localhost/${UNCPATH%%:*}\$$WITHOUTDRIVE" ;; *) skip_all='skipping UNC path tests, cannot determine current path as UNC' @@ -35,6 +32,18 @@ case "$UNCPATH" in ;; esac +test_expect_failure 'clone into absolute path lacking a drive prefix' ' + USINGBACKSLASHES="$(echo "$WITHOUTDRIVE"/without-drive-prefix | + tr / \\\\)" && + git clone . "$USINGBACKSLASHES" && + test -f without-drive-prefix/.git/HEAD +' + +test -d "$UNCPATH" || { + skip_all='could not access administrative share; skipping' + test_done +} + test_expect_success setup ' test_commit initial ' From 72dc1a5d55fa69f649db998e6c5e652af8def133 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Fri, 7 Dec 2018 13:39:30 +0100 Subject: [PATCH 028/250] clean: do not traverse mount points It seems to be not exactly rare on Windows to install NTFS junction points (the equivalent of "bind mounts" on Linux/Unix) in worktrees, e.g. to map some development tools into a subdirectory. In such a scenario, it is pretty horrible if `git clean -dfx` traverses into the mapped directory and starts to "clean up". Let's just not do that. Let's make sure before we traverse into a directory that it is not a mount point (or junction). This addresses https://github.com/git-for-windows/git/issues/607 Signed-off-by: Johannes Schindelin --- builtin/clean.c | 14 ++++++++++++++ compat/mingw.c | 22 ++++++++++++++++++++++ compat/mingw.h | 3 +++ git-compat-util.h | 4 ++++ path.c | 39 +++++++++++++++++++++++++++++++++++++++ path.h | 1 + t/t7300-clean.sh | 9 +++++++++ 7 files changed, 92 insertions(+) diff --git a/builtin/clean.c b/builtin/clean.c index 053c94fc6bd12a..aeb7eb2d1ed2c3 100644 --- a/builtin/clean.c +++ b/builtin/clean.c @@ -40,6 +40,8 @@ static const char *msg_remove = N_("Removing %s\n"); static const char *msg_would_remove = N_("Would remove %s\n"); static const char *msg_skip_git_dir = N_("Skipping repository %s\n"); static const char *msg_would_skip_git_dir = N_("Would skip repository %s\n"); +static const char *msg_skip_mount_point = N_("Skipping mount point %s\n"); +static const char *msg_would_skip_mount_point = N_("Would skip mount point %s\n"); static const char *msg_warn_remove_failed = N_("failed to remove %s"); static const char *msg_warn_lstat_failed = N_("could not lstat %s\n"); static const char *msg_skip_cwd = N_("Refusing to remove current working directory\n"); @@ -184,6 +186,18 @@ static int remove_dirs(struct strbuf *path, const char *prefix, int force_flag, goto out; } + if (is_mount_point(path)) { + if (!quiet) { + quote_path(path->buf, prefix, "ed, 0); + printf(dry_run ? + _(msg_would_skip_mount_point) : + _(msg_skip_mount_point), quoted.buf); + } + *dir_gone = 0; + + goto out; + } + dir = opendir(path->buf); if (!dir) { /* an empty dir could be removed even if it is unreadble */ diff --git a/compat/mingw.c b/compat/mingw.c index 8a9972a1ca19f7..b830621c372c0e 100644 --- a/compat/mingw.c +++ b/compat/mingw.c @@ -2711,6 +2711,28 @@ pid_t waitpid(pid_t pid, int *status, int options) return -1; } +int mingw_is_mount_point(struct strbuf *path) +{ + WIN32_FIND_DATAW findbuf = { 0 }; + HANDLE handle; + wchar_t wfilename[MAX_PATH]; + int wlen = xutftowcs_path(wfilename, path->buf); + if (wlen < 0) + die(_("could not get long path for '%s'"), path->buf); + + /* remove trailing slash, if any */ + if (wlen > 0 && wfilename[wlen - 1] == L'/') + wfilename[--wlen] = L'\0'; + + handle = FindFirstFileW(wfilename, &findbuf); + if (handle == INVALID_HANDLE_VALUE) + return 0; + FindClose(handle); + + return (findbuf.dwFileAttributes & FILE_ATTRIBUTE_REPARSE_POINT) && + (findbuf.dwReserved0 == IO_REPARSE_TAG_MOUNT_POINT); +} + int xutftowcsn(wchar_t *wcs, const char *utfs, size_t wcslen, int utflen) { int upos = 0, wpos = 0; diff --git a/compat/mingw.h b/compat/mingw.h index 444daedfa52469..af6fc3f12970bf 100644 --- a/compat/mingw.h +++ b/compat/mingw.h @@ -36,6 +36,9 @@ static inline void convert_slashes(char *path) if (*path == '\\') *path = '/'; } +struct strbuf; +int mingw_is_mount_point(struct strbuf *path); +#define is_mount_point mingw_is_mount_point #define PATH_SEP ';' char *mingw_query_user_email(void); #define query_user_email mingw_query_user_email diff --git a/git-compat-util.h b/git-compat-util.h index 4678e21c4cb80b..3d7af4b529fb17 100644 --- a/git-compat-util.h +++ b/git-compat-util.h @@ -383,6 +383,10 @@ static inline int git_has_dir_sep(const char *path) #define has_dir_sep(path) git_has_dir_sep(path) #endif +#ifndef is_mount_point +#define is_mount_point is_mount_point_via_stat +#endif + #ifndef query_user_email #define query_user_email() NULL #endif diff --git a/path.c b/path.c index 3b598b2847ff03..0968f505503af7 100644 --- a/path.c +++ b/path.c @@ -1423,6 +1423,45 @@ char *strip_path_suffix(const char *path, const char *suffix) return offset == -1 ? NULL : xstrndup(path, offset); } +int is_mount_point_via_stat(struct strbuf *path) +{ + size_t len = path->len; + dev_t current_dev; + struct stat st; + + if (!strcmp("/", path->buf)) + return 1; + + strbuf_addstr(path, "/."); + if (lstat(path->buf, &st)) { + /* + * If we cannot access the current directory, we cannot say + * that it is a bind mount. + */ + strbuf_setlen(path, len); + return 0; + } + current_dev = st.st_dev; + + /* Now look at the parent directory */ + strbuf_addch(path, '.'); + if (lstat(path->buf, &st)) { + /* + * If we cannot access the parent directory, we cannot say + * that it is a bind mount. + */ + strbuf_setlen(path, len); + return 0; + } + strbuf_setlen(path, len); + + /* + * If the device ID differs between current and parent directory, + * then it is a bind mount. + */ + return current_dev != st.st_dev; +} + int daemon_avoid_alias(const char *p) { int sl, ndot; diff --git a/path.h b/path.h index e67348f25397cc..7a46488f8d706d 100644 --- a/path.h +++ b/path.h @@ -172,6 +172,7 @@ int normalize_path_copy(char *dst, const char *src); int strbuf_normalize_path(struct strbuf *src); int longest_ancestor_length(const char *path, struct string_list *prefixes); char *strip_path_suffix(const char *path, const char *suffix); +int is_mount_point_via_stat(struct strbuf *path); int daemon_avoid_alias(const char *path); /* diff --git a/t/t7300-clean.sh b/t/t7300-clean.sh index 00d4070156243b..7c3a1ca91df534 100755 --- a/t/t7300-clean.sh +++ b/t/t7300-clean.sh @@ -800,4 +800,13 @@ test_expect_success 'traverse into directories that may have ignored entries' ' ) ' +test_expect_success MINGW 'clean does not traverse mount points' ' + mkdir target && + >target/dont-clean-me && + git init with-mountpoint && + cmd //c "mklink /j with-mountpoint\\mountpoint target" && + git -C with-mountpoint clean -dfx && + test_path_is_file target/dont-clean-me +' + test_done From c8d01e19be022f271eb530f1d2aec5c901dee089 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Mon, 24 Jun 2019 23:41:27 +0200 Subject: [PATCH 029/250] mimalloc: adjust for building inside Git We want to compile mimalloc's source code as part of Git, rather than requiring the code to be built as an external library: mimalloc uses a CMake-based build, which is not necessarily easy to integrate into the flavors of Git for Windows (which will be the main benefitting port). Signed-off-by: Johannes Schindelin --- compat/mimalloc/alloc.c | 1 - compat/mimalloc/mimalloc.h | 3 ++- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/compat/mimalloc/alloc.c b/compat/mimalloc/alloc.c index 0fed5e754c1918..f975a92b6b6d1a 100644 --- a/compat/mimalloc/alloc.c +++ b/compat/mimalloc/alloc.c @@ -17,7 +17,6 @@ terms of the MIT license. A copy of the license can be found in the file #include // malloc, abort #define MI_IN_ALLOC_C -#include "alloc-override.c" #include "free.c" #undef MI_IN_ALLOC_C diff --git a/compat/mimalloc/mimalloc.h b/compat/mimalloc/mimalloc.h index ff6f0568301db4..c319bc2bf79603 100644 --- a/compat/mimalloc/mimalloc.h +++ b/compat/mimalloc/mimalloc.h @@ -95,7 +95,8 @@ terms of the MIT license. A copy of the license can be found in the file // Includes // ------------------------------------------------------ -#include // size_t +#include "compat/posix.h" + #include // bool #include // INTPTR_MAX From 49fd7a5d10ae8a6fdac4be5cae30cc0e30536cdc Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Mon, 9 Apr 2012 13:04:35 -0500 Subject: [PATCH 030/250] Always auto-gc after calling a fast-import transport After importing anything with fast-import, we should always let the garbage collector do its job, since the objects are written to disk inefficiently. This brings down an initial import of http://selenic.com/hg from about 230 megabytes to about 14. In the future, we may want to make this configurable on a per-remote basis, or maybe teach fast-import about it in the first place. Signed-off-by: Johannes Schindelin --- transport-helper.c | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/transport-helper.c b/transport-helper.c index 7545e50d27ed63..af4a717ad538d2 100644 --- a/transport-helper.c +++ b/transport-helper.c @@ -22,6 +22,8 @@ #include "packfile.h" static int debug; +/* TODO: put somewhere sensible, e.g. git_transport_options? */ +static int auto_gc = 1; struct helper_data { char *name; @@ -588,6 +590,13 @@ static int fetch_with_import(struct transport *transport, } } strbuf_release(&buf); + if (auto_gc) { + struct child_process cmd = CHILD_PROCESS_INIT; + + cmd.git_cmd = 1; + strvec_pushl(&cmd.args, "gc", "--auto", "--quiet", NULL); + run_command(&cmd); + } return 0; } From afd54e4c5189df619799632b9c58c5f9a60d2cb3 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Tue, 18 Apr 2017 12:38:30 +0200 Subject: [PATCH 031/250] mingw: allow absolute paths without drive prefix When specifying an absolute path without a drive prefix, we convert that path internally. Let's make sure that we handle that case properly, too ;-) This fixes the command git clone https://github.com/git-for-windows/git \G4W Signed-off-by: Johannes Schindelin --- compat/mingw.c | 10 +++++++++- t/t5580-unc-paths.sh | 2 +- 2 files changed, 10 insertions(+), 2 deletions(-) diff --git a/compat/mingw.c b/compat/mingw.c index 8a9972a1ca19f7..24c7f268183e2d 100644 --- a/compat/mingw.c +++ b/compat/mingw.c @@ -1164,11 +1164,19 @@ unsigned int sleep (unsigned int seconds) char *mingw_mktemp(char *template) { wchar_t wtemplate[MAX_PATH]; + int offset = 0; + if (xutftowcs_path(wtemplate, template) < 0) return NULL; + + if (is_dir_sep(template[0]) && !is_dir_sep(template[1]) && + iswalpha(wtemplate[0]) && wtemplate[1] == L':') { + /* We have an absolute path missing the drive prefix */ + offset = 2; + } if (!_wmktemp(wtemplate)) return NULL; - if (xwcstoutf(template, wtemplate, strlen(template) + 1) < 0) + if (xwcstoutf(template, wtemplate + offset, strlen(template) + 1) < 0) return NULL; return template; } diff --git a/t/t5580-unc-paths.sh b/t/t5580-unc-paths.sh index 2c30f3b51d3183..e9df367d5777fd 100755 --- a/t/t5580-unc-paths.sh +++ b/t/t5580-unc-paths.sh @@ -32,7 +32,7 @@ case "$UNCPATH" in ;; esac -test_expect_failure 'clone into absolute path lacking a drive prefix' ' +test_expect_success 'clone into absolute path lacking a drive prefix' ' USINGBACKSLASHES="$(echo "$WITHOUTDRIVE"/without-drive-prefix | tr / \\\\)" && git clone . "$USINGBACKSLASHES" && From 11467abb92d518b0c000cf8bcf6a14e3e77432e9 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Tue, 11 Dec 2018 12:55:26 +0100 Subject: [PATCH 032/250] clean: remove mount points when possible Windows' equivalent to "bind mounts", NTFS junction points, can be unlinked without affecting the mount target. This is clearly what users expect to happen when they call `git clean -dfx` in a worktree that contains NTFS junction points: the junction should be removed, and the target directory of said junction should be left alone (unless it is inside the worktree). Signed-off-by: Johannes Schindelin --- builtin/clean.c | 13 +++++++++++++ compat/mingw.h | 1 + t/t7300-clean.sh | 1 + 3 files changed, 15 insertions(+) diff --git a/builtin/clean.c b/builtin/clean.c index aeb7eb2d1ed2c3..df39efcc50fbc4 100644 --- a/builtin/clean.c +++ b/builtin/clean.c @@ -40,8 +40,10 @@ static const char *msg_remove = N_("Removing %s\n"); static const char *msg_would_remove = N_("Would remove %s\n"); static const char *msg_skip_git_dir = N_("Skipping repository %s\n"); static const char *msg_would_skip_git_dir = N_("Would skip repository %s\n"); +#ifndef CAN_UNLINK_MOUNT_POINTS static const char *msg_skip_mount_point = N_("Skipping mount point %s\n"); static const char *msg_would_skip_mount_point = N_("Would skip mount point %s\n"); +#endif static const char *msg_warn_remove_failed = N_("failed to remove %s"); static const char *msg_warn_lstat_failed = N_("could not lstat %s\n"); static const char *msg_skip_cwd = N_("Refusing to remove current working directory\n"); @@ -187,6 +189,7 @@ static int remove_dirs(struct strbuf *path, const char *prefix, int force_flag, } if (is_mount_point(path)) { +#ifndef CAN_UNLINK_MOUNT_POINTS if (!quiet) { quote_path(path->buf, prefix, "ed, 0); printf(dry_run ? @@ -194,6 +197,16 @@ static int remove_dirs(struct strbuf *path, const char *prefix, int force_flag, _(msg_skip_mount_point), quoted.buf); } *dir_gone = 0; +#else + if (!dry_run && unlink(path->buf)) { + int saved_errno = errno; + quote_path(path->buf, prefix, "ed, 0); + errno = saved_errno; + warning_errno(_(msg_warn_remove_failed), quoted.buf); + *dir_gone = 0; + ret = -1; + } +#endif goto out; } diff --git a/compat/mingw.h b/compat/mingw.h index af6fc3f12970bf..fb83cdaf4e982c 100644 --- a/compat/mingw.h +++ b/compat/mingw.h @@ -39,6 +39,7 @@ static inline void convert_slashes(char *path) struct strbuf; int mingw_is_mount_point(struct strbuf *path); #define is_mount_point mingw_is_mount_point +#define CAN_UNLINK_MOUNT_POINTS 1 #define PATH_SEP ';' char *mingw_query_user_email(void); #define query_user_email mingw_query_user_email diff --git a/t/t7300-clean.sh b/t/t7300-clean.sh index 7c3a1ca91df534..6f16f3893191e7 100755 --- a/t/t7300-clean.sh +++ b/t/t7300-clean.sh @@ -806,6 +806,7 @@ test_expect_success MINGW 'clean does not traverse mount points' ' git init with-mountpoint && cmd //c "mklink /j with-mountpoint\\mountpoint target" && git -C with-mountpoint clean -dfx && + test_path_is_missing with-mountpoint/mountpoint && test_path_is_file target/dont-clean-me ' From a834a75d68eaa4832d731a3ab9ed89c504ff96e9 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Mon, 16 Feb 2015 14:06:59 +0100 Subject: [PATCH 033/250] mingw: include the Python parts in the build While Git for Windows does not _ship_ Python (in order to save on bandwidth), MSYS2 provides very fine Python interpreters that users can easily take advantage of, by using Git for Windows within its SDK. Signed-off-by: Johannes Schindelin --- config.mak.uname | 1 + 1 file changed, 1 insertion(+) diff --git a/config.mak.uname b/config.mak.uname index 3e26bb074a4b51..b1b106c223a3fe 100644 --- a/config.mak.uname +++ b/config.mak.uname @@ -753,6 +753,7 @@ ifeq ($(uname_S),MINGW) ifneq (CLANGARM64,$(MSYSTEM)) USE_NED_ALLOCATOR = YesPlease endif + NO_PYTHON = ifeq (/mingw64,$(subst 32,64,$(subst clangarm,mingw,$(prefix)))) # Move system config into top-level /etc/ ETC_GITCONFIG = ../etc/gitconfig From 6e243986daa838604195c994f9583b8d96ea70ed Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Fri, 31 Jan 2020 12:02:47 +0100 Subject: [PATCH 034/250] mingw: demonstrate a `git add` issue with NTFS junctions NTFS junctions are somewhat similar in spirit to Unix bind mounts: they point to a different directory and are resolved by the filesystem driver. As such, they appear to `lstat()` as if they are directories, not as if they are symbolic links. _Any_ user can create junctions, while symbolic links can only be created by non-administrators in Developer Mode on Windows 10. Hence NTFS junctions are much more common "in the wild" than NTFS symbolic links. It was reported in https://github.com/git-for-windows/git/issues/2481 that adding files via an absolute path that traverses an NTFS junction: since 1e64d18 (mingw: do resolve symlinks in `getcwd()`), we resolve not only symbolic links but also NTFS junctions when determining the absolute path of the current directory. The same is not true for `git add `, where symbolic links are resolved in ``, but not NTFS junctions. Signed-off-by: Johannes Schindelin --- t/t3700-add.sh | 11 +++++++++++ 1 file changed, 11 insertions(+) diff --git a/t/t3700-add.sh b/t/t3700-add.sh index df580a5806b4f1..d934f3417faa45 100755 --- a/t/t3700-add.sh +++ b/t/t3700-add.sh @@ -548,4 +548,15 @@ test_expect_success CASE_INSENSITIVE_FS 'path is case-insensitive' ' git add "$downcased" ' +test_expect_failure MINGW 'can add files via NTFS junctions' ' + test_when_finished "cmd //c rmdir junction && rm -rf target" && + test_create_repo target && + cmd //c "mklink /j junction target" && + >target/via-junction && + git -C junction add "$(pwd)/junction/via-junction" && + echo via-junction >expect && + git -C target diff --cached --name-only >actual && + test_cmp expect actual +' + test_done From efca11072f3a182fa8c69b19130396b59c554e50 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Tue, 21 Feb 2017 13:28:58 +0100 Subject: [PATCH 035/250] mingw: ensure valid CTYPE A change between versions 2.4.1 and 2.6.0 of the MSYS2 runtime modified how Cygwin's runtime (and hence Git for Windows' MSYS2 runtime derivative) handles locales: d16a56306d (Consolidate wctomb/mbtowc calls for POSIX-1.2008, 2016-07-20). An unintended side-effect is that "cold-calling" into the POSIX emulation will start with a locale based on the current code page, something that Git for Windows is very ill-prepared for, as it expects to be able to pass a command-line containing non-ASCII characters to the shell without having those characters munged. One symptom of this behavior: when `git clone` or `git fetch` shell out to call `git-upload-pack` with a path that contains non-ASCII characters, the shell tried to interpret the entire command-line (including command-line parameters) as executable path, which obviously must fail. This fixes https://github.com/git-for-windows/git/issues/1036 Signed-off-by: Johannes Schindelin --- compat/mingw.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/compat/mingw.c b/compat/mingw.c index 8a9972a1ca19f7..f1571e977e4092 100644 --- a/compat/mingw.c +++ b/compat/mingw.c @@ -2847,6 +2847,9 @@ static void setup_windows_environment(void) if (!tmp && (tmp = getenv("USERPROFILE"))) setenv("HOME", tmp, 1); } + + if (!getenv("LC_ALL") && !getenv("LC_CTYPE") && !getenv("LANG")) + setenv("LC_CTYPE", "C.UTF-8", 1); } static void get_current_user_sid(PSID *sid, HANDLE *linked_token) From 84380fffd6e519d870df5fde445fcefdfeeae7fa Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Mon, 24 Jun 2019 23:43:06 +0200 Subject: [PATCH 036/250] mimalloc: offer a build-time option to enable it By defining `USE_MIMALLOC`, Git can now be compiled with that nicely-fast and small allocator. Note that we have to disable a couple `DEVELOPER` options to build mimalloc's source code, as it makes heavy use of declarations after statements, among other things that disagree with Git's conventions. We even have to silence some GCC warnings in non-DEVELOPER mode. For example, the `-Wno-array-bounds` flag is needed because in `-O2` builds, trying to call `NtCurrentTeb()` (which `_mi_thread_id()` does on Windows) causes the bogus warning about a system header, likely related to https://sourceforge.net/p/mingw-w64/mailman/message/37674519/ and to https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99578: C:/git-sdk-64-minimal/mingw64/include/psdk_inc/intrin-impl.h:838:1: error: array subscript 0 is outside array bounds of 'long long unsigned int[0]' [-Werror=array-bounds] 838 | __buildreadseg(__readgsqword, unsigned __int64, "gs", "q") | ^~~~~~~~~~~~~~ Also: The `mimalloc` library uses C11-style atomics, therefore we must require that standard when compiling with GCC if we want to use `mimalloc` (instead of requiring "only" C99). This is what we do in the CMake definition already, therefore this commit does not need to touch `contrib/buildsystems/`. Signed-off-by: Johannes Schindelin --- Makefile | 41 +++++++++++++++++++++++++++++++++++++++++ compat/.gitattributes | 1 + compat/posix.h | 10 ++++++++++ config.mak.dev | 2 ++ config.mak.uname | 2 +- 5 files changed, 55 insertions(+), 1 deletion(-) diff --git a/Makefile b/Makefile index 70d1543b6b8688..bdbe91d4ebbcb4 100644 --- a/Makefile +++ b/Makefile @@ -1346,6 +1346,7 @@ BUILTIN_OBJS += builtin/write-tree.o # upstream unnecessarily (making merging in future changes easier). THIRD_PARTY_SOURCES += compat/inet_ntop.c THIRD_PARTY_SOURCES += compat/inet_pton.c +THIRD_PARTY_SOURCES += compat/mimalloc/% THIRD_PARTY_SOURCES += compat/nedmalloc/% THIRD_PARTY_SOURCES += compat/obstack.% THIRD_PARTY_SOURCES += compat/poll/% @@ -2149,6 +2150,46 @@ ifdef USE_NED_ALLOCATOR OVERRIDE_STRDUP = YesPlease endif +ifdef USE_MIMALLOC + MIMALLOC_OBJS = \ + compat/mimalloc/alloc-aligned.o \ + compat/mimalloc/alloc.o \ + compat/mimalloc/arena.o \ + compat/mimalloc/bitmap.o \ + compat/mimalloc/heap.o \ + compat/mimalloc/init.o \ + compat/mimalloc/libc.o \ + compat/mimalloc/options.o \ + compat/mimalloc/os.o \ + compat/mimalloc/page.o \ + compat/mimalloc/random.o \ + compat/mimalloc/prim/prim.o \ + compat/mimalloc/segment.o \ + compat/mimalloc/segment-map.o \ + compat/mimalloc/stats.o + + COMPAT_CFLAGS += -Icompat/mimalloc -DMI_DEBUG=0 -DUSE_MIMALLOC --std=gnu11 + COMPAT_OBJS += $(MIMALLOC_OBJS) + +$(MIMALLOC_OBJS): COMPAT_CFLAGS += -DBANNED_H + +$(MIMALLOC_OBJS): COMPAT_CFLAGS += \ + -DMI_WIN_USE_FLS \ + -Wno-attributes \ + -Wno-unknown-pragmas \ + -Wno-unused-function \ + -Wno-array-bounds + +ifdef DEVELOPER +$(MIMALLOC_OBJS): COMPAT_CFLAGS += \ + -Wno-pedantic \ + -Wno-declaration-after-statement \ + -Wno-old-style-definition \ + -Wno-missing-prototypes \ + -Wno-implicit-function-declaration +endif +endif + ifdef OVERRIDE_STRDUP COMPAT_CFLAGS += -DOVERRIDE_STRDUP COMPAT_OBJS += compat/strdup.o diff --git a/compat/.gitattributes b/compat/.gitattributes index 40dbfb170dabc5..2b5a66a3b34bda 100644 --- a/compat/.gitattributes +++ b/compat/.gitattributes @@ -1 +1,2 @@ /zlib-uncompress2.c whitespace=-indent-with-non-tab,-trailing-space +/mimalloc/**/* whitespace=-trailing-space diff --git a/compat/posix.h b/compat/posix.h index 626d71256f8428..106b16c49eaf62 100644 --- a/compat/posix.h +++ b/compat/posix.h @@ -176,6 +176,16 @@ typedef unsigned long uintptr_t; #define _ALL_SOURCE 1 #endif +#ifdef USE_MIMALLOC +#include "mimalloc.h" +#define malloc mi_malloc +#define calloc mi_calloc +#define realloc mi_realloc +#define free mi_free +#define strdup mi_strdup +#define strndup mi_strndup +#endif + #ifdef MKDIR_WO_TRAILING_SLASH #define mkdir(a,b) compat_mkdir_wo_trailing_slash((a),(b)) int compat_mkdir_wo_trailing_slash(const char*, mode_t); diff --git a/config.mak.dev b/config.mak.dev index e86b6e1b34a2d7..b63797ef509333 100644 --- a/config.mak.dev +++ b/config.mak.dev @@ -22,8 +22,10 @@ endif ifneq ($(uname_S),FreeBSD) ifneq ($(or $(filter gcc6,$(COMPILER_FEATURES)),$(filter clang7,$(COMPILER_FEATURES))),) +ifndef USE_MIMALLOC DEVELOPER_CFLAGS += -std=gnu99 endif +endif else # FreeBSD cannot limit to C99 because its system headers unconditionally # rely on C11 features. diff --git a/config.mak.uname b/config.mak.uname index b1b106c223a3fe..189549e3af7d1c 100644 --- a/config.mak.uname +++ b/config.mak.uname @@ -502,7 +502,7 @@ endif CC = compat/vcbuild/scripts/clink.pl AR = compat/vcbuild/scripts/lib.pl CFLAGS = - BASIC_CFLAGS = -nologo -I. -Icompat/vcbuild/include -DWIN32 -D_CONSOLE -DHAVE_STRING_H -D_CRT_SECURE_NO_WARNINGS -D_CRT_NONSTDC_NO_DEPRECATE + BASIC_CFLAGS = -nologo -I. -Icompat/vcbuild/include -DWIN32 -D_CONSOLE -DHAVE_STRING_H -D_CRT_SECURE_NO_WARNINGS -D_CRT_NONSTDC_NO_DEPRECATE -MP -std:c11 COMPAT_OBJS = compat/msvc.o compat/winansi.o \ compat/win32/flush.o \ compat/win32/path-utils.o \ From 88b3a214811d37f4ea59ad61d83a081a2bc128f0 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Fri, 31 Jan 2020 11:44:31 +0100 Subject: [PATCH 037/250] strbuf_realpath(): use platform-dependent API if available Some platforms (e.g. Windows) provide API functions to resolve paths much quicker. Let's offer a way to short-cut `strbuf_realpath()` on those platforms. Signed-off-by: Johannes Schindelin --- abspath.c | 3 +++ git-compat-util.h | 4 ++++ 2 files changed, 7 insertions(+) diff --git a/abspath.c b/abspath.c index 1202cde23dbc9b..0c17e98654e4b0 100644 --- a/abspath.c +++ b/abspath.c @@ -93,6 +93,9 @@ static char *strbuf_realpath_1(struct strbuf *resolved, const char *path, goto error_out; } + if (platform_strbuf_realpath(resolved, path)) + return resolved->buf; + strbuf_addstr(&remaining, path); get_root_part(resolved, &remaining); diff --git a/git-compat-util.h b/git-compat-util.h index 4678e21c4cb80b..cd1d21a1894dfd 100644 --- a/git-compat-util.h +++ b/git-compat-util.h @@ -387,6 +387,10 @@ static inline int git_has_dir_sep(const char *path) #define query_user_email() NULL #endif +#ifndef platform_strbuf_realpath +#define platform_strbuf_realpath(resolved, path) NULL +#endif + #ifdef __TANDEM #include #include From 2cc06a7cf8e5e2d4bfe676f0ccc264bef210590f Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Sat, 1 Feb 2020 00:31:16 +0100 Subject: [PATCH 038/250] mingw: allow `git.exe` to be used instead of the "Git wrapper" Git for Windows wants to add `git.exe` to the users' `PATH`, without cluttering the latter with unnecessary executables such as `wish.exe`. To that end, it invented the concept of its "Git wrapper", i.e. a tiny executable located in `C:\Program Files\Git\cmd\git.exe` (originally a CMD script) whose sole purpose is to set up a couple of environment variables and then spawn the _actual_ `git.exe` (which nowadays lives in `C:\Program Files\Git\mingw64\bin\git.exe` for 64-bit, and the obvious equivalent for 32-bit installations). Currently, the following environment variables are set unless already initialized: - `MSYSTEM`, to make sure that the MSYS2 Bash and the MSYS2 Perl interpreter behave as expected, and - `PLINK_PROTOCOL`, to force PuTTY's `plink.exe` to use the SSH protocol instead of Telnet, - `PATH`, to make sure that the `bin` folder in the user's home directory, as well as the `/mingw64/bin` and the `/usr/bin` directories are included. The trick here is that the `/mingw64/bin/` and `/usr/bin/` directories are relative to the top-level installation directory of Git for Windows (which the included Bash interprets as `/`, i.e. as the MSYS pseudo root directory). Using the absence of `MSYSTEM` as a tell-tale, we can detect in `git.exe` whether these environment variables have been initialized properly. Therefore we can call `C:\Program Files\Git\mingw64\bin\git` in-place after this change, without having to call Git through the Git wrapper. Obviously, above-mentioned directories must be _prepended_ to the `PATH` variable, otherwise we risk picking up executables from unrelated Git installations. We do that by constructing the new `PATH` value from scratch, appending `$HOME/bin` (if `HOME` is set), then the MSYS2 system directories, and then appending the original `PATH`. Side note: this modification of the `PATH` variable is independent of the modification necessary to reach the executables and scripts in `/mingw64/libexec/git-core/`, i.e. the `GIT_EXEC_PATH`. That modification is still performed by Git, elsewhere, long after making the changes described above. While we _still_ cannot simply hard-link `mingw64\bin\git.exe` to `cmd` (because the former depends on a couple of `.dll` files that are only in `mingw64\bin`, i.e. calling `...\cmd\git.exe` would fail to load due to missing dependencies), at least we can now avoid that extra process of running the Git wrapper (which then has to wait for the spawned `git.exe` to finish) by calling `...\mingw64\bin\git.exe` directly, via its absolute path. Testing this is in Git's test suite tricky: we set up a "new" MSYS pseudo-root and copy the `git.exe` file into the appropriate location, then verify that `MSYSTEM` is set properly, and also that the `PATH` is modified so that scripts can be found in `$HOME/bin`, `/mingw64/bin/` and `/usr/bin/`. This addresses https://github.com/git-for-windows/git/issues/2283 Signed-off-by: Johannes Schindelin --- compat/mingw.c | 80 +++++++++++++++++++++++++++++++++++++++++++ config.mak.uname | 4 +-- t/t0060-path-utils.sh | 33 +++++++++++++++++- 3 files changed, 114 insertions(+), 3 deletions(-) diff --git a/compat/mingw.c b/compat/mingw.c index f1571e977e4092..b758203fc4dfb3 100644 --- a/compat/mingw.c +++ b/compat/mingw.c @@ -2796,6 +2796,54 @@ int xwcstoutf(char *utf, const wchar_t *wcs, size_t utflen) return -1; } +#ifdef ENSURE_MSYSTEM_IS_SET +#if !defined(RUNTIME_PREFIX) || !defined(HAVE_WPGMPTR) +static size_t append_system_bin_dirs(char *path UNUSED, size_t size UNUSED) +{ + return 0; +} +#else +static size_t append_system_bin_dirs(char *path, size_t size) +{ + char prefix[32768]; + const char *slash; + size_t len = xwcstoutf(prefix, _wpgmptr, sizeof(prefix)), off = 0; + + if (len == 0 || len >= sizeof(prefix) || + !(slash = find_last_dir_sep(prefix))) + return 0; + /* strip trailing `git.exe` */ + len = slash - prefix; + + /* strip trailing `cmd` or `mingw64\bin` or `mingw32\bin` or `bin` or `libexec\git-core` */ + if (strip_suffix_mem(prefix, &len, "\\mingw64\\libexec\\git-core") || + strip_suffix_mem(prefix, &len, "\\mingw64\\bin")) + off += xsnprintf(path + off, size - off, + "%.*s\\mingw64\\bin;", (int)len, prefix); + else if (strip_suffix_mem(prefix, &len, "\\clangarm64\\libexec\\git-core") || + strip_suffix_mem(prefix, &len, "\\clangarm64\\bin")) + off += xsnprintf(path + off, size - off, + "%.*s\\clangarm64\\bin;", (int)len, prefix); + else if (strip_suffix_mem(prefix, &len, "\\mingw32\\libexec\\git-core") || + strip_suffix_mem(prefix, &len, "\\mingw32\\bin")) + off += xsnprintf(path + off, size - off, + "%.*s\\mingw32\\bin;", (int)len, prefix); + else if (strip_suffix_mem(prefix, &len, "\\cmd") || + strip_suffix_mem(prefix, &len, "\\bin") || + strip_suffix_mem(prefix, &len, "\\libexec\\git-core")) + off += xsnprintf(path + off, size - off, + "%.*s\\mingw%d\\bin;", (int)len, prefix, + (int)(sizeof(void *) * 8)); + else + return 0; + + off += xsnprintf(path + off, size - off, + "%.*s\\usr\\bin;", (int)len, prefix); + return off; +} +#endif +#endif + static void setup_windows_environment(void) { char *tmp = getenv("TMPDIR"); @@ -2848,6 +2896,38 @@ static void setup_windows_environment(void) setenv("HOME", tmp, 1); } + if (!getenv("PLINK_PROTOCOL")) + setenv("PLINK_PROTOCOL", "ssh", 0); + +#ifdef ENSURE_MSYSTEM_IS_SET + if (!(tmp = getenv("MSYSTEM")) || !tmp[0]) { + const char *home = getenv("HOME"), *path = getenv("PATH"); + char buf[32768]; + size_t off = 0; + +#if defined(__aarch64__) || defined(_M_ARM64) || defined(_M_ARM64EC) + setenv("MSYSTEM", "CLANGARM64", 1); +#elif defined(__MINGW64__) || defined(_M_AMD64) + setenv("MSYSTEM", "MINGW64", 1); +#else + setenv("MSYSTEM", "MINGW32", 1); +#endif + + if (home) + off += xsnprintf(buf + off, sizeof(buf) - off, + "%s\\bin;", home); + off += append_system_bin_dirs(buf + off, sizeof(buf) - off); + if (path) + off += xsnprintf(buf + off, sizeof(buf) - off, + "%s", path); + else if (off > 0) + buf[off - 1] = '\0'; + else + buf[0] = '\0'; + setenv("PATH", buf, 1); + } +#endif + if (!getenv("LC_ALL") && !getenv("LC_CTYPE") && !getenv("LANG")) setenv("LC_CTYPE", "C.UTF-8", 1); } diff --git a/config.mak.uname b/config.mak.uname index 3e26bb074a4b51..1ebaa8b17692b9 100644 --- a/config.mak.uname +++ b/config.mak.uname @@ -509,7 +509,7 @@ endif compat/win32/pthread.o compat/win32/syslog.o \ compat/win32/trace2_win32_process_info.o \ compat/win32/dirent.o - COMPAT_CFLAGS = -D__USE_MINGW_ACCESS -DDETECT_MSYS_TTY -DNOGDI -DHAVE_STRING_H -Icompat -Icompat/regex -Icompat/win32 -DSTRIP_EXTENSION=\".exe\" + COMPAT_CFLAGS = -D__USE_MINGW_ACCESS -DDETECT_MSYS_TTY -DENSURE_MSYSTEM_IS_SET -DNOGDI -DHAVE_STRING_H -Icompat -Icompat/regex -Icompat/win32 -DSTRIP_EXTENSION=\".exe\" BASIC_LDFLAGS = -IGNORE:4217 -IGNORE:4049 -NOLOGO -ENTRY:wmainCRTStartup -SUBSYSTEM:CONSOLE # invalidcontinue.obj allows Git's source code to close the same file # handle twice, or to access the osfhandle of an already-closed stdout @@ -742,7 +742,7 @@ ifeq ($(uname_S),MINGW) endif CC = gcc COMPAT_CFLAGS += -D__USE_MINGW_ANSI_STDIO=0 -DDETECT_MSYS_TTY \ - -fstack-protector-strong + -DENSURE_MSYSTEM_IS_SET -fstack-protector-strong EXTLIBS += -lntdll EXTRA_PROGRAMS += headless-git$X INSTALL = /bin/install diff --git a/t/t0060-path-utils.sh b/t/t0060-path-utils.sh index 8545cdfab559b4..56faf5fe732ee0 100755 --- a/t/t0060-path-utils.sh +++ b/t/t0060-path-utils.sh @@ -602,7 +602,8 @@ test_expect_success !VALGRIND,RUNTIME_PREFIX,CAN_EXEC_IN_PWD 'RUNTIME_PREFIX wor echo "echo HERE" | write_script pretend/libexec/git-core/git-here && GIT_EXEC_PATH= ./pretend/bin/git here >actual && echo HERE >expect && - test_cmp expect actual' + test_cmp expect actual +' test_expect_success !VALGRIND,RUNTIME_PREFIX,CAN_EXEC_IN_PWD '%(prefix)/ works' ' git config yes.path "%(prefix)/yes" && @@ -611,4 +612,34 @@ test_expect_success !VALGRIND,RUNTIME_PREFIX,CAN_EXEC_IN_PWD '%(prefix)/ works' test_cmp expect actual ' +test_expect_success MINGW,RUNTIME_PREFIX 'MSYSTEM/PATH is adjusted if necessary' ' + if test -z "$MINGW_PREFIX" + then + MINGW_PREFIX="/$(echo "${MSYSTEM:-MINGW64}" | tr A-Z a-z)" + fi && + mkdir -p "$HOME"/bin pretend"$MINGW_PREFIX"/bin \ + pretend"$MINGW_PREFIX"/libexec/git-core pretend/usr/bin && + cp "$GIT_EXEC_PATH"/git.exe pretend"$MINGW_PREFIX"/bin/ && + cp "$GIT_EXEC_PATH"/git.exe pretend"$MINGW_PREFIX"/libexec/git-core/ && + # copy the .dll files, if any (happens when building via CMake) + if test -n "$(ls "$GIT_EXEC_PATH"/*.dll 2>/dev/null)" + then + cp "$GIT_EXEC_PATH"/*.dll pretend"$MINGW_PREFIX"/bin/ && + cp "$GIT_EXEC_PATH"/*.dll pretend"$MINGW_PREFIX"/libexec/git-core/ + fi && + echo "env | grep MSYSTEM=" | write_script "$HOME"/bin/git-test-home && + echo "echo ${MINGW_PREFIX#/}" | write_script pretend"$MINGW_PREFIX"/bin/git-test-bin && + echo "echo usr" | write_script pretend/usr/bin/git-test-bin2 && + + ( + MSYSTEM= && + GIT_EXEC_PATH= && + pretend"$MINGW_PREFIX"/libexec/git-core/git.exe test-home >actual && + pretend"$MINGW_PREFIX"/libexec/git-core/git.exe test-bin >>actual && + pretend"$MINGW_PREFIX"/bin/git.exe test-bin2 >>actual + ) && + test_write_lines MSYSTEM=$MSYSTEM "${MINGW_PREFIX#/}" usr >expect && + test_cmp expect actual +' + test_done From 2a0619a9a528258b4ccfde69e9ccda38376567f2 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Mon, 24 Jun 2019 23:45:21 +0200 Subject: [PATCH 039/250] mingw: use mimalloc Thorough benchmarking with repacking a subset of linux.git (the commit history reachable from 93a6fefe2f ([PATCH] fix the SYSCTL=n compilation, 2007-02-28), to be precise) suggest that this allocator is on par, in multi-threaded situations maybe even better than nedmalloc: `git repack -adfq` with mimalloc, 8 threads: 31.166991900 27.576763800 28.712311000 27.373859000 27.163141900 `git repack -adfq` with nedmalloc, 8 threads: 31.915032900 27.149883100 28.244933700 27.240188800 28.580849500 In a different test using GitHub Actions build agents (probably single-threaded, a core-strength of nedmalloc)): `git repack -q -d -l -A --unpack-unreachable=2.weeks.ago` with mimalloc: 943.426 978.500 939.709 959.811 954.605 `git repack -q -d -l -A --unpack-unreachable=2.weeks.ago` with nedmalloc: 995.383 952.179 943.253 963.043 980.468 While these measurements were not executed with complete scientific rigor, as no hardware was set aside specifically for these benchmarks, it shows that mimalloc and nedmalloc perform almost the same, nedmalloc with a bit higher variance and also slightly higher average (further testing suggests that nedmalloc performs worse in multi-threaded situations than in single-threaded ones). In short: mimalloc seems to be slightly better suited for our purposes than nedmalloc. Seeing that mimalloc is developed actively, while nedmalloc ceased to see any updates in eight years, let's use mimalloc on Windows instead. Signed-off-by: Johannes Schindelin --- config.mak.uname | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-) diff --git a/config.mak.uname b/config.mak.uname index 189549e3af7d1c..71209def6f0fc4 100644 --- a/config.mak.uname +++ b/config.mak.uname @@ -750,9 +750,7 @@ ifeq ($(uname_S),MINGW) HAVE_LIBCHARSET_H = YesPlease USE_GETTEXT_SCHEME = fallthrough USE_LIBPCRE = YesPlease - ifneq (CLANGARM64,$(MSYSTEM)) - USE_NED_ALLOCATOR = YesPlease - endif + USE_MIMALLOC = YesPlease NO_PYTHON = ifeq (/mingw64,$(subst 32,64,$(subst clangarm,mingw,$(prefix)))) # Move system config into top-level /etc/ From ab04fcc9a2aea740c46a5b8ff7a6e4008af04929 Mon Sep 17 00:00:00 2001 From: Thomas Braun Date: Thu, 8 May 2014 21:43:24 +0200 Subject: [PATCH 040/250] transport: optionally disable side-band-64k Since commit 0c499ea60fda (send-pack: demultiplex a sideband stream with status data, 2010-02-05) the send-pack builtin uses the side-band-64k capability if advertised by the server. Unfortunately this breaks pushing over the dump git protocol if used over a network connection. The detailed reasons for this breakage are (by courtesy of Jeff Preshing, quoted from https://groups.google.com/d/msg/msysgit/at8D7J-h7mw/eaLujILGUWoJ): MinGW wraps Windows sockets in CRT file descriptors in order to mimic the functionality of POSIX sockets. This causes msvcrt.dll to treat sockets as Installable File System (IFS) handles, calling ReadFile, WriteFile, DuplicateHandle and CloseHandle on them. This approach works well in simple cases on recent versions of Windows, but does not support all usage patterns. In particular, using this approach, any attempt to read & write concurrently on the same socket (from one or more processes) will deadlock in a scenario where the read waits for a response from the server which is only invoked after the write. This is what send_pack currently attempts to do in the use_sideband codepath. The new config option `sendpack.sideband` allows to override the side-band-64k capability of the server, and thus makes the dumb git protocol work. Other transportation methods like ssh and http/https still benefit from the sideband channel, therefore the default value of `sendpack.sideband` is still true. Signed-off-by: Thomas Braun Signed-off-by: Oliver Schneider Signed-off-by: Johannes Schindelin --- Documentation/config.adoc | 2 ++ Documentation/config/sendpack.adoc | 5 +++++ send-pack.c | 6 +++--- 3 files changed, 10 insertions(+), 3 deletions(-) create mode 100644 Documentation/config/sendpack.adoc diff --git a/Documentation/config.adoc b/Documentation/config.adoc index a8b04c4e5183d2..644216f02a322a 100644 --- a/Documentation/config.adoc +++ b/Documentation/config.adoc @@ -518,6 +518,8 @@ include::config/safe.adoc[] include::config/sendemail.adoc[] +include::config/sendpack.adoc[] + include::config/sequencer.adoc[] include::config/showbranch.adoc[] diff --git a/Documentation/config/sendpack.adoc b/Documentation/config/sendpack.adoc new file mode 100644 index 00000000000000..e306f657fba7dd --- /dev/null +++ b/Documentation/config/sendpack.adoc @@ -0,0 +1,5 @@ +sendpack.sideband:: + Allows to disable the side-band-64k capability for send-pack even + when it is advertised by the server. Makes it possible to work + around a limitation in the git for windows implementation together + with the dump git protocol. Defaults to true. diff --git a/send-pack.c b/send-pack.c index 86592ce526db95..360b89be197041 100644 --- a/send-pack.c +++ b/send-pack.c @@ -494,7 +494,7 @@ int send_pack(struct repository *r, int need_pack_data = 0; int allow_deleting_refs = 0; int status_report = 0; - int use_sideband = 0; + int use_sideband = 1; int quiet_supported = 0; int agent_supported = 0; int advertise_sid = 0; @@ -518,6 +518,7 @@ int send_pack(struct repository *r, goto out; } + repo_config_get_bool(r, "sendpack.sideband", &use_sideband); repo_config_get_bool(r, "push.negotiate", &push_negotiate); if (push_negotiate) { trace2_region_enter("send_pack", "push_negotiate", r); @@ -539,8 +540,7 @@ int send_pack(struct repository *r, allow_deleting_refs = 1; if (server_supports("ofs-delta")) args->use_ofs_delta = 1; - if (server_supports("side-band-64k")) - use_sideband = 1; + use_sideband = use_sideband && server_supports("side-band-64k"); if (server_supports("quiet")) quiet_supported = 1; if (server_supports("agent")) From 14986ffe5b4aaec5fffe2b17eb0e4de311c4eba4 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Wed, 1 Jan 2020 21:07:22 +0100 Subject: [PATCH 041/250] mingw: do resolve symlinks in `getcwd()` As pointed out in https://github.com/git-for-windows/git/issues/1676, the `git rev-parse --is-inside-work-tree` command currently fails when the current directory's path contains symbolic links. The underlying reason for this bug is that `getcwd()` is supposed to resolve symbolic links, but our `mingw_getcwd()` implementation did not. We do have all the building blocks for that, though: the `GetFinalPathByHandleW()` function will resolve symbolic links. However, we only called that function if `GetLongPathNameW()` failed, for historical reasons: the latter function was supported for a long time, but the former API function was introduced only with Windows Vista, and we used to support also Windows XP. With that support having been dropped, we are free to call the symbolic link-resolving function right away. Signed-off-by: Johannes Schindelin --- compat/mingw.c | 18 +++++++----------- 1 file changed, 7 insertions(+), 11 deletions(-) diff --git a/compat/mingw.c b/compat/mingw.c index 8a9972a1ca19f7..8233b0627ec79f 100644 --- a/compat/mingw.c +++ b/compat/mingw.c @@ -1234,18 +1234,16 @@ char *mingw_getcwd(char *pointer, int len) { wchar_t cwd[MAX_PATH], wpointer[MAX_PATH]; DWORD ret = GetCurrentDirectoryW(ARRAY_SIZE(cwd), cwd); + HANDLE hnd; if (!ret || ret >= ARRAY_SIZE(cwd)) { errno = ret ? ENAMETOOLONG : err_win_to_posix(GetLastError()); return NULL; } - ret = GetLongPathNameW(cwd, wpointer, ARRAY_SIZE(wpointer)); - if (!ret && GetLastError() == ERROR_ACCESS_DENIED) { - HANDLE hnd = CreateFileW(cwd, 0, - FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE, NULL, - OPEN_EXISTING, FILE_FLAG_BACKUP_SEMANTICS, NULL); - if (hnd == INVALID_HANDLE_VALUE) - return NULL; + hnd = CreateFileW(cwd, 0, + FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE, NULL, + OPEN_EXISTING, FILE_FLAG_BACKUP_SEMANTICS, NULL); + if (hnd != INVALID_HANDLE_VALUE) { ret = GetFinalPathNameByHandleW(hnd, wpointer, ARRAY_SIZE(wpointer), 0); CloseHandle(hnd); if (!ret || ret >= ARRAY_SIZE(wpointer)) @@ -1254,13 +1252,11 @@ char *mingw_getcwd(char *pointer, int len) return NULL; return pointer; } - if (!ret || ret >= ARRAY_SIZE(wpointer)) - return NULL; - if (GetFileAttributesW(wpointer) == INVALID_FILE_ATTRIBUTES) { + if (GetFileAttributesW(cwd) == INVALID_FILE_ATTRIBUTES) { errno = ENOENT; return NULL; } - if (xwcstoutf(pointer, wpointer, len) < 0) + if (xwcstoutf(pointer, cwd, len) < 0) return NULL; convert_slashes(pointer); return pointer; From 3ff40b11dcc31552c7b106c406130691eaef6251 Mon Sep 17 00:00:00 2001 From: Bjoern Mueller Date: Wed, 22 Jan 2020 13:49:13 +0100 Subject: [PATCH 042/250] mingw: fix fatal error working on mapped network drives on Windows In 1e64d18 (mingw: do resolve symlinks in `getcwd()`) a problem was introduced that causes git for Windows to stop working with certain mapped network drives (in particular, drives that are mapped to locations with long path names). Error message was "fatal: Unable to read current working directory: No such file or directory". Present change fixes this issue as discussed in https://github.com/git-for-windows/git/issues/2480 Signed-off-by: Bjoern Mueller --- compat/mingw.c | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-) diff --git a/compat/mingw.c b/compat/mingw.c index 8a9972a1ca19f7..181fdb91ef8eeb 100644 --- a/compat/mingw.c +++ b/compat/mingw.c @@ -1248,8 +1248,13 @@ char *mingw_getcwd(char *pointer, int len) return NULL; ret = GetFinalPathNameByHandleW(hnd, wpointer, ARRAY_SIZE(wpointer), 0); CloseHandle(hnd); - if (!ret || ret >= ARRAY_SIZE(wpointer)) - return NULL; + if (!ret || ret >= ARRAY_SIZE(wpointer)) { + ret = GetLongPathNameW(cwd, wpointer, ARRAY_SIZE(wpointer)); + if (!ret || ret >= ARRAY_SIZE(wpointer)) { + errno = ret ? ENAMETOOLONG : err_win_to_posix(GetLastError()); + return NULL; + } + } if (xwcstoutf(pointer, normalize_ntpath(wpointer), len) < 0) return NULL; return pointer; From 7ccce5222e741bf4d9d505903594da05fe8a50d5 Mon Sep 17 00:00:00 2001 From: Jeff Hostetler Date: Thu, 30 Jan 2020 14:22:27 -0500 Subject: [PATCH 043/250] clink.pl: fix MSVC compile script to handle libcurl-d.lib Update clink.pl to link with either libcurl.lib or libcurl-d.lib depending on whether DEBUG=1 is set. Signed-off-by: Jeff Hostetler Signed-off-by: Johannes Schindelin --- compat/vcbuild/scripts/clink.pl | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/compat/vcbuild/scripts/clink.pl b/compat/vcbuild/scripts/clink.pl index 3bd824154be381..c4c99d1a11f18c 100755 --- a/compat/vcbuild/scripts/clink.pl +++ b/compat/vcbuild/scripts/clink.pl @@ -56,7 +56,8 @@ # need to use that instead? foreach my $flag (@lflags) { if ($flag =~ /^-LIBPATH:(.*)/) { - foreach my $l ("libcurl_imp.lib", "libcurl.lib") { + my $libcurl = $is_debug ? "libcurl-d.lib" : "libcurl.lib"; + foreach my $l ("libcurl_imp.lib", $libcurl) { if (-f "$1/$l") { $lib = $l; last; From 36775844adddeb2b79947ccc7e753c747a0a7e61 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Fri, 31 Jan 2020 11:49:04 +0100 Subject: [PATCH 044/250] mingw: implement a platform-specific `strbuf_realpath()` There is a Win32 API function to resolve symbolic links, and we can use that instead of resolving them manually. Even better, this function also resolves NTFS junction points (which are somewhat similar to bind mounts). This fixes https://github.com/git-for-windows/git/issues/2481. Signed-off-by: Johannes Schindelin --- compat/mingw.c | 76 +++++++++++++++++++++++++++++++++++++++++++ compat/mingw.h | 3 ++ t/t0060-path-utils.sh | 8 +++++ t/t3700-add.sh | 2 +- t/t5601-clone.sh | 7 ++++ 5 files changed, 95 insertions(+), 1 deletion(-) diff --git a/compat/mingw.c b/compat/mingw.c index 8a9972a1ca19f7..2b5cfd63463ac4 100644 --- a/compat/mingw.c +++ b/compat/mingw.c @@ -1230,6 +1230,82 @@ struct tm *localtime_r(const time_t *timep, struct tm *result) } #endif +char *mingw_strbuf_realpath(struct strbuf *resolved, const char *path) +{ + wchar_t wpath[MAX_PATH]; + HANDLE h; + DWORD ret; + int len; + const char *last_component = NULL; + char *append = NULL; + + if (xutftowcs_path(wpath, path) < 0) + return NULL; + + h = CreateFileW(wpath, 0, + FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE, NULL, + OPEN_EXISTING, FILE_FLAG_BACKUP_SEMANTICS, NULL); + + /* + * strbuf_realpath() allows the last path component to not exist. If + * that is the case, now it's time to try without last component. + */ + if (h == INVALID_HANDLE_VALUE && + GetLastError() == ERROR_FILE_NOT_FOUND) { + /* cut last component off of `wpath` */ + wchar_t *p = wpath + wcslen(wpath); + + while (p != wpath) + if (*(--p) == L'/' || *p == L'\\') + break; /* found start of last component */ + + if (p != wpath && (last_component = find_last_dir_sep(path))) { + append = xstrdup(last_component + 1); /* skip directory separator */ + /* + * Do not strip the trailing slash at the drive root, otherwise + * the path would be e.g. `C:` (which resolves to the + * _current_ directory on that drive). + */ + if (p[-1] == L':') + p[1] = L'\0'; + else + *p = L'\0'; + h = CreateFileW(wpath, 0, FILE_SHARE_READ | + FILE_SHARE_WRITE | FILE_SHARE_DELETE, + NULL, OPEN_EXISTING, + FILE_FLAG_BACKUP_SEMANTICS, NULL); + } + } + + if (h == INVALID_HANDLE_VALUE) { +realpath_failed: + FREE_AND_NULL(append); + return NULL; + } + + ret = GetFinalPathNameByHandleW(h, wpath, ARRAY_SIZE(wpath), 0); + CloseHandle(h); + if (!ret || ret >= ARRAY_SIZE(wpath)) + goto realpath_failed; + + len = wcslen(wpath) * 3; + strbuf_grow(resolved, len); + len = xwcstoutf(resolved->buf, normalize_ntpath(wpath), len); + if (len < 0) + goto realpath_failed; + resolved->len = len; + + if (append) { + /* Use forward-slash, like `normalize_ntpath()` */ + strbuf_complete(resolved, '/'); + strbuf_addstr(resolved, append); + FREE_AND_NULL(append); + } + + return resolved->buf; + +} + char *mingw_getcwd(char *pointer, int len) { wchar_t cwd[MAX_PATH], wpointer[MAX_PATH]; diff --git a/compat/mingw.h b/compat/mingw.h index 444daedfa52469..f6daf47ee4e0a7 100644 --- a/compat/mingw.h +++ b/compat/mingw.h @@ -39,6 +39,9 @@ static inline void convert_slashes(char *path) #define PATH_SEP ';' char *mingw_query_user_email(void); #define query_user_email mingw_query_user_email +struct strbuf; +char *mingw_strbuf_realpath(struct strbuf *resolved, const char *path); +#define platform_strbuf_realpath mingw_strbuf_realpath /** * Verifies that the specified path is owned by the user running the diff --git a/t/t0060-path-utils.sh b/t/t0060-path-utils.sh index 8545cdfab559b4..eb2ab9d437ea8e 100755 --- a/t/t0060-path-utils.sh +++ b/t/t0060-path-utils.sh @@ -281,6 +281,14 @@ test_expect_success SYMLINKS 'real path works on symlinks' ' test_cmp expect actual ' +test_expect_success MINGW 'real path works near drive root' ' + # we need a non-existing path at the drive root; simply skip if C:/xyz exists + if test ! -e C:/xyz + then + test C:/xyz = $(test-tool path-utils real_path C:/xyz) + fi +' + test_expect_success SYMLINKS 'prefix_path works with absolute paths to work tree symlinks' ' ln -s target symlink && echo "symlink" >expect && diff --git a/t/t3700-add.sh b/t/t3700-add.sh index d934f3417faa45..10185f31f019a9 100755 --- a/t/t3700-add.sh +++ b/t/t3700-add.sh @@ -548,7 +548,7 @@ test_expect_success CASE_INSENSITIVE_FS 'path is case-insensitive' ' git add "$downcased" ' -test_expect_failure MINGW 'can add files via NTFS junctions' ' +test_expect_success MINGW 'can add files via NTFS junctions' ' test_when_finished "cmd //c rmdir junction && rm -rf target" && test_create_repo target && cmd //c "mklink /j junction target" && diff --git a/t/t5601-clone.sh b/t/t5601-clone.sh index d743d986c401a0..f70d99016ea2f7 100755 --- a/t/t5601-clone.sh +++ b/t/t5601-clone.sh @@ -78,6 +78,13 @@ test_expect_success 'clone respects GIT_WORK_TREE' ' ' +test_expect_success CASE_INSENSITIVE_FS 'core.worktree is not added due to path case' ' + + mkdir UPPERCASE && + git clone src "$(pwd)/uppercase" && + test "unset" = "$(git -C UPPERCASE config --default unset core.worktree)" +' + test_expect_success 'clone from hooks' ' test_create_repo r0 && From 853be8515dea4c7568fdf73bd684244bd1bf053d Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Tue, 25 Aug 2020 12:13:26 +0200 Subject: [PATCH 045/250] mingw: ignore HOMEDRIVE/HOMEPATH if it points to Windows' system directory Internally, Git expects the environment variable `HOME` to be set, and to point to the current user's home directory. This environment variable is not set by default on Windows, and therefore Git tries its best to construct one if it finds `HOME` unset. There are actually two different approaches Git tries: first, it looks at `HOMEDRIVE`/`HOMEPATH` because this is widely used in corporate environments with roaming profiles, and a user generally wants their global Git settings to be in a roaming profile. Only when `HOMEDRIVE`/`HOMEPATH` is either unset or does not point to a valid location, Git will fall back to using `USERPROFILE` instead. However, starting with Windows Vista, for secondary logons and services, the environment variables `HOMEDRIVE`/`HOMEPATH` point to Windows' system directory (usually `C:\Windows\system32`). That is undesirable, and that location is usually write-protected anyway. So let's verify that the `HOMEDRIVE`/`HOMEPATH` combo does not point to Windows' system directory before using it, falling back to `USERPROFILE` if it does. This fixes git-for-windows#2709 Initial-Path-by: Ivan Pozdeev Signed-off-by: Johannes Schindelin --- compat/mingw.c | 15 ++++++++++++++- 1 file changed, 14 insertions(+), 1 deletion(-) diff --git a/compat/mingw.c b/compat/mingw.c index b758203fc4dfb3..d26f2c116335cd 100644 --- a/compat/mingw.c +++ b/compat/mingw.c @@ -2844,6 +2844,18 @@ static size_t append_system_bin_dirs(char *path, size_t size) #endif #endif +static int is_system32_path(const char *path) +{ + WCHAR system32[MAX_PATH], wpath[MAX_PATH]; + + if (xutftowcs_path(wpath, path) < 0 || + !GetSystemDirectoryW(system32, ARRAY_SIZE(system32)) || + _wcsicmp(system32, wpath)) + return 0; + + return 1; +} + static void setup_windows_environment(void) { char *tmp = getenv("TMPDIR"); @@ -2884,7 +2896,8 @@ static void setup_windows_environment(void) strbuf_addstr(&buf, tmp); if ((tmp = getenv("HOMEPATH"))) { strbuf_addstr(&buf, tmp); - if (is_directory(buf.buf)) + if (!is_system32_path(buf.buf) && + is_directory(buf.buf)) setenv("HOME", buf.buf, 1); else tmp = NULL; /* use $USERPROFILE */ From 8a624d1df0f182ff0c9d0c27483408df1fd232f9 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Wed, 4 Mar 2020 21:55:28 +0100 Subject: [PATCH 046/250] http: use new "best effort" strategy for Secure Channel revoke checking MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The native Windows HTTPS backend is based on Secure Channel which lets the caller decide how to handle revocation checking problems caused by missing information in the certificate or offline CRL distribution points. Unfortunately, cURL chose to handle these problems differently than OpenSSL by default: while OpenSSL happily ignores those problems (essentially saying "¯\_(ツ)_/¯"), the Secure Channel backend will error out instead. As a remedy, the "no revoke" mode was introduced, which turns off revocation checking altogether. This is a bit heavy-handed. We support this via the `http.schannelCheckRevoke` setting. In https://github.com/curl/curl/pull/4981, we contributed an opt-in "best effort" strategy that emulates what OpenSSL seems to do. In Git for Windows, we actually want this to be the default. This patch makes it so, introducing it as a new value for the `http.schannelCheckRevoke" setting, which now becmes a tristate: it accepts the values "false", "true" or "best-effort" (defaulting to the last one). Signed-off-by: Johannes Schindelin --- Documentation/config/http.adoc | 12 +++++++----- http.c | 26 ++++++++++++++++++++++---- 2 files changed, 29 insertions(+), 9 deletions(-) diff --git a/Documentation/config/http.adoc b/Documentation/config/http.adoc index 9da5c298cc1d5e..9122c5dc23ea1a 100644 --- a/Documentation/config/http.adoc +++ b/Documentation/config/http.adoc @@ -233,11 +233,13 @@ http.sslKeyType:: http.schannelCheckRevoke:: Used to enforce or disable certificate revocation checks in cURL - when http.sslBackend is set to "schannel". Defaults to `true` if - unset. Only necessary to disable this if Git consistently errors - and the message is about checking the revocation status of a - certificate. This option is ignored if cURL lacks support for - setting the relevant SSL option at runtime. + when http.sslBackend is set to "schannel" via "true" and "false", + respectively. Another accepted value is "best-effort" (the default) + in which case revocation checks are performed, but errors due to + revocation list distribution points that are offline are silently + ignored, as well as errors due to certificates missing revocation + list distribution points. This option is ignored if cURL lacks + support for setting the relevant SSL option at runtime. http.schannelUseSSLCAInfo:: As of cURL v7.60.0, the Secure Channel backend can use the diff --git a/http.c b/http.c index d88e79fbde9c4c..808069ec29a220 100644 --- a/http.c +++ b/http.c @@ -147,7 +147,13 @@ static char *cached_accept_language; static char *http_ssl_backend; -static int http_schannel_check_revoke = 1; +static long http_schannel_check_revoke_mode = +#ifdef CURLSSLOPT_REVOKE_BEST_EFFORT + CURLSSLOPT_REVOKE_BEST_EFFORT; +#else + CURLSSLOPT_NO_REVOKE; +#endif + /* * With the backend being set to `schannel`, setting sslCAinfo would override * the Certificate Store in cURL v7.60.0 and later, which is not what we want @@ -422,7 +428,19 @@ static int http_options(const char *var, const char *value, } if (!strcmp("http.schannelcheckrevoke", var)) { - http_schannel_check_revoke = git_config_bool(var, value); + if (value && !strcmp(value, "best-effort")) { + http_schannel_check_revoke_mode = +#ifdef CURLSSLOPT_REVOKE_BEST_EFFORT + CURLSSLOPT_REVOKE_BEST_EFFORT; +#else + CURLSSLOPT_NO_REVOKE; + warning(_("%s=%s unsupported by current cURL"), + var, value); +#endif + } else + http_schannel_check_revoke_mode = + (git_config_bool(var, value) ? + 0 : CURLSSLOPT_NO_REVOKE); return 0; } @@ -1056,8 +1074,8 @@ static CURL *get_curl_handle(void) #endif if (http_ssl_backend && !strcmp("schannel", http_ssl_backend) && - !http_schannel_check_revoke) { - curl_easy_setopt(result, CURLOPT_SSL_OPTIONS, (long)CURLSSLOPT_NO_REVOKE); + http_schannel_check_revoke_mode) { + curl_easy_setopt(result, CURLOPT_SSL_OPTIONS, http_schannel_check_revoke_mode); } if (http_proactive_auth != PROACTIVE_AUTH_NONE) From e3c52b6bbbf74291b0e590dedf2b2c6442900740 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Sat, 9 May 2020 16:19:06 +0200 Subject: [PATCH 047/250] t5505/t5516: allow running without `.git/branches/` in the templates When we commit the template directory as part of `make vcxproj`, the `branches/` directory is not actually commited, as it is empty. Two tests were not prepared for that situation. This developer tried to get rid of the support for `.git/branches/` a long time ago, but that effort did not bear fruit, so the best we can do is work around in these here tests. Signed-off-by: Johannes Schindelin --- t/t5505-remote.sh | 4 ++-- t/t5516-fetch-push.sh | 8 ++++---- 2 files changed, 6 insertions(+), 6 deletions(-) diff --git a/t/t5505-remote.sh b/t/t5505-remote.sh index bef0250e8921c6..a94ce02c7b89b3 100755 --- a/t/t5505-remote.sh +++ b/t/t5505-remote.sh @@ -1155,7 +1155,7 @@ test_expect_success !WITH_BREAKING_CHANGES 'migrate a remote from named file in ( cd six && git remote rm origin && - mkdir .git/branches && + mkdir -p .git/branches && echo "$origin_url#main" >.git/branches/origin && git remote rename origin origin && test_path_is_missing .git/branches/origin && @@ -1170,7 +1170,7 @@ test_expect_success !WITH_BREAKING_CHANGES 'migrate a remote from named file in ( cd seven && git remote rm origin && - mkdir .git/branches && + mkdir -p .git/branches && echo "quux#foom" > .git/branches/origin && git remote rename origin origin && test_path_is_missing .git/branches/origin && diff --git a/t/t5516-fetch-push.sh b/t/t5516-fetch-push.sh index dabcc5f8117645..048bd0f286405a 100755 --- a/t/t5516-fetch-push.sh +++ b/t/t5516-fetch-push.sh @@ -979,7 +979,7 @@ test_expect_success !WITH_BREAKING_CHANGES 'fetch with branches' ' mk_empty testrepo && git branch second $the_first_commit && git checkout second && - mkdir testrepo/.git/branches && + mkdir -p testrepo/.git/branches && echo ".." > testrepo/.git/branches/branch1 && ( cd testrepo && @@ -993,7 +993,7 @@ test_expect_success !WITH_BREAKING_CHANGES 'fetch with branches' ' test_expect_success !WITH_BREAKING_CHANGES 'fetch with branches containing #' ' mk_empty testrepo && - mkdir testrepo/.git/branches && + mkdir -p testrepo/.git/branches && echo "..#second" > testrepo/.git/branches/branch2 && ( cd testrepo && @@ -1010,7 +1010,7 @@ test_expect_success !WITH_BREAKING_CHANGES 'push with branches' ' git checkout second && test_when_finished "rm -rf .git/branches" && - mkdir .git/branches && + mkdir -p .git/branches && echo "testrepo" > .git/branches/branch1 && git push branch1 && @@ -1026,7 +1026,7 @@ test_expect_success !WITH_BREAKING_CHANGES 'push with branches containing #' ' mk_empty testrepo && test_when_finished "rm -rf .git/branches" && - mkdir .git/branches && + mkdir -p .git/branches && echo "testrepo#branch3" > .git/branches/branch2 && git push branch2 && From ecd89df1a09d2a484261387677ec2d014770cccb Mon Sep 17 00:00:00 2001 From: Jeff Hostetler Date: Tue, 30 Mar 2021 14:25:31 -0400 Subject: [PATCH 048/250] clink.pl: fix libexpatd.lib link error when using MSVC When building with `make MSVC=1 DEBUG=1`, link to `libexpatd.lib` rather than `libexpat.lib`. It appears that the `vcpkg` package for "libexpat" has changed and now creates `libexpatd.lib` for debug mode builds. Previously, both debug and release builds created a ".lib" with the same basename. Signed-off-by: Jeff Hostetler --- compat/vcbuild/scripts/clink.pl | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/compat/vcbuild/scripts/clink.pl b/compat/vcbuild/scripts/clink.pl index 3bd824154be381..2768ae15f1879f 100755 --- a/compat/vcbuild/scripts/clink.pl +++ b/compat/vcbuild/scripts/clink.pl @@ -66,7 +66,11 @@ } push(@args, $lib); } elsif ("$arg" eq "-lexpat") { + if ($is_debug) { + push(@args, "libexpatd.lib"); + } else { push(@args, "libexpat.lib"); + } } elsif ("$arg" =~ /^-L/ && "$arg" ne "-LTCG") { $arg =~ s/^-L/-LIBPATH:/; push(@lflags, $arg); From 4ce4d733000e14fbb93c6b114ea31585e6410d71 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Sat, 9 May 2020 19:24:23 +0200 Subject: [PATCH 049/250] t5505/t5516: fix white-space around redirectors The convention in Git project's shell scripts is to have white-space _before_, but not _after_ the `>` (or `<`). Signed-off-by: Johannes Schindelin --- t/t5505-remote.sh | 6 +++--- t/t5516-fetch-push.sh | 10 +++++----- 2 files changed, 8 insertions(+), 8 deletions(-) diff --git a/t/t5505-remote.sh b/t/t5505-remote.sh index a94ce02c7b89b3..d9eaee6b98fa82 100755 --- a/t/t5505-remote.sh +++ b/t/t5505-remote.sh @@ -951,8 +951,8 @@ test_expect_success '"remote show" does not show symbolic refs' ' ( cd three && git remote show origin >output && - ! grep "^ *HEAD$" < output && - ! grep -i stale < output + ! grep "^ *HEAD$" .git/branches/origin && + echo "quux#foom" >.git/branches/origin && git remote rename origin origin && test_path_is_missing .git/branches/origin && test "$(git config remote.origin.url)" = "quux" && diff --git a/t/t5516-fetch-push.sh b/t/t5516-fetch-push.sh index 048bd0f286405a..0b3a6219f444d1 100755 --- a/t/t5516-fetch-push.sh +++ b/t/t5516-fetch-push.sh @@ -980,7 +980,7 @@ test_expect_success !WITH_BREAKING_CHANGES 'fetch with branches' ' git branch second $the_first_commit && git checkout second && mkdir -p testrepo/.git/branches && - echo ".." > testrepo/.git/branches/branch1 && + echo ".." >testrepo/.git/branches/branch1 && ( cd testrepo && git fetch branch1 && @@ -994,7 +994,7 @@ test_expect_success !WITH_BREAKING_CHANGES 'fetch with branches' ' test_expect_success !WITH_BREAKING_CHANGES 'fetch with branches containing #' ' mk_empty testrepo && mkdir -p testrepo/.git/branches && - echo "..#second" > testrepo/.git/branches/branch2 && + echo "..#second" >testrepo/.git/branches/branch2 && ( cd testrepo && git fetch branch2 && @@ -1011,7 +1011,7 @@ test_expect_success !WITH_BREAKING_CHANGES 'push with branches' ' test_when_finished "rm -rf .git/branches" && mkdir -p .git/branches && - echo "testrepo" > .git/branches/branch1 && + echo "testrepo" >.git/branches/branch1 && git push branch1 && ( @@ -1027,7 +1027,7 @@ test_expect_success !WITH_BREAKING_CHANGES 'push with branches containing #' ' test_when_finished "rm -rf .git/branches" && mkdir -p .git/branches && - echo "testrepo#branch3" > .git/branches/branch2 && + echo "testrepo#branch3" >.git/branches/branch2 && git push branch2 && ( @@ -1557,7 +1557,7 @@ EOF git init no-thin && git --git-dir=no-thin/.git config receive.unpacklimit 0 && git push no-thin/.git refs/heads/main:refs/heads/foo && - echo modified >> path1 && + echo modified >>path1 && git commit -am modified && git repack -adf && rcvpck="git receive-pack --reject-thin-pack-for-testing" && From 580bf73cb9518eb6089e62d65876ab8bb7172ee8 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Sat, 12 Sep 2015 12:25:47 +0200 Subject: [PATCH 050/250] t3701: verify that we can add *lots* of files interactively Signed-off-by: Johannes Schindelin --- t/t3701-add-interactive.sh | 21 +++++++++++++++++++++ 1 file changed, 21 insertions(+) diff --git a/t/t3701-add-interactive.sh b/t/t3701-add-interactive.sh index b8a05d95f3f130..81d0487b3b90eb 100755 --- a/t/t3701-add-interactive.sh +++ b/t/t3701-add-interactive.sh @@ -1166,6 +1166,27 @@ test_expect_success 'checkout -p patch editing of added file' ' ) ' +test_expect_success EXPENSIVE 'add -i with a lot of files' ' + git reset --hard && + x160=0123456789012345678901234567890123456789 && + x160=$x160$x160$x160$x160 && + y= && + i=0 && + while test $i -le 200 + do + name=$(printf "%s%03d" $x160 $i) && + echo $name >$name && + git add -N $name && + y="${y}y$LF" && + i=$(($i+1)) || + exit 1 + done && + echo "$y" | git add -p -- . && + git diff --cached >staged && + test_line_count = 1407 staged && + git reset --hard +' + test_expect_success 'show help from add--helper' ' git reset --hard && cat >expect <<-EOF && From 70600ccf33999420e795b8a3c6298928db3a1544 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Thu, 2 Jul 2020 16:35:05 +0200 Subject: [PATCH 051/250] git add -i: handle CR/LF line endings in the interactive input As of Git for Windows v2.27.0, there is an option to use Windows' newly-introduced Pseudo Console support. When running an interactive add operation with this support enabled, Git will receive CR/LF line endings. Therefore, let's not pretend that we are expecting Unix line endings. This fixes https://github.com/git-for-windows/git/issues/2729 Signed-off-by: Johannes Schindelin --- prompt.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/prompt.c b/prompt.c index f21c5bf1c7e42f..a07b135050b9c2 100644 --- a/prompt.c +++ b/prompt.c @@ -80,7 +80,7 @@ int git_read_line_interactively(struct strbuf *line) int ret; fflush(stdout); - ret = strbuf_getline_lf(line, stdin); + ret = strbuf_getline(line, stdin); if (ret != EOF) strbuf_trim_trailing_newline(line); From 3c8ffbf7b335da7cff7d925924c26997743ddf95 Mon Sep 17 00:00:00 2001 From: Luke Bonanomi Date: Wed, 24 Jun 2020 07:45:52 -0400 Subject: [PATCH 052/250] commit: accept "scissors" with CR/LF line endings This change enhances `git commit --cleanup=scissors` by detecting scissors lines ending in either LF (UNIX-style) or CR/LF (DOS-style). Regression tests are included to specifically test for trailing comments after a CR/LF-terminated scissors line. Signed-off-by: Luke Bonanomi Signed-off-by: Johannes Schindelin --- t/t7502-commit-porcelain.sh | 42 +++++++++++++++++++++++++++++++++++++ wt-status.c | 13 +++++++++--- 2 files changed, 52 insertions(+), 3 deletions(-) diff --git a/t/t7502-commit-porcelain.sh b/t/t7502-commit-porcelain.sh index b37e2018a74a7b..c38b96b66cd20a 100755 --- a/t/t7502-commit-porcelain.sh +++ b/t/t7502-commit-porcelain.sh @@ -623,6 +623,48 @@ test_expect_success 'cleanup commit messages (scissors option,-F,-e, scissors on test_must_be_empty actual ' +test_expect_success 'helper-editor' ' + + write_script lf-to-crlf.sh <<-\EOF + sed "s/\$/Q/" <"$1" | tr Q "\\015" >"$1".new && + mv -f "$1".new "$1" + EOF +' + +test_expect_success 'cleanup commit messages (scissors option,-F,-e, CR/LF line endings)' ' + + test_config core.editor "\"$PWD/lf-to-crlf.sh\"" && + scissors="# ------------------------ >8 ------------------------" && + + test_write_lines >text \ + "# Keep this comment" "" " $scissors" \ + "# Keep this comment, too" "$scissors" \ + "# Remove this comment" "$scissors" \ + "Remove this comment, too" && + + test_write_lines >expect \ + "# Keep this comment" "" " $scissors" \ + "# Keep this comment, too" && + + git commit --cleanup=scissors -e -F text --allow-empty && + git cat-file -p HEAD >raw && + sed -e "1,/^\$/d" raw >actual && + test_cmp expect actual +' + +test_expect_success 'cleanup commit messages (scissors option,-F,-e, scissors on first line, CR/LF line endings)' ' + + scissors="# ------------------------ >8 ------------------------" && + test_write_lines >text \ + "$scissors" \ + "# Remove this comment and any following lines" && + cp text /tmp/test2-text && + git commit --cleanup=scissors -e -F text --allow-empty --allow-empty-message && + git cat-file -p HEAD >raw && + sed -e "1,/^\$/d" raw >actual && + test_must_be_empty actual +' + test_expect_success 'cleanup commit messages (strip option,-F)' ' echo >>negative && diff --git a/wt-status.c b/wt-status.c index 454601afa15a95..c1f46e42757f9f 100644 --- a/wt-status.c +++ b/wt-status.c @@ -40,7 +40,7 @@ #define UF_DELAY_WARNING_IN_MS (2 * 1000) static const char cut_line[] = -"------------------------ >8 ------------------------\n"; +"------------------------ >8 ------------------------"; static char default_wt_status_colors[][COLOR_MAXLEN] = { GIT_COLOR_NORMAL, /* WT_STATUS_HEADER */ @@ -1096,15 +1096,22 @@ static void wt_longstatus_print_other(struct wt_status *s, status_printf_ln(s, GIT_COLOR_NORMAL, "%s", ""); } +static inline int starts_with_newline(const char *p) +{ + return *p == '\n' || (*p == '\r' && p[1] == '\n'); +} + size_t wt_status_locate_end(const char *s, size_t len) { const char *p; struct strbuf pattern = STRBUF_INIT; strbuf_addf(&pattern, "\n%s %s", comment_line_str, cut_line); - if (starts_with(s, pattern.buf + 1)) + if (starts_with(s, pattern.buf + 1) && + starts_with_newline(s + pattern.len - 1)) len = 0; - else if ((p = strstr(s, pattern.buf))) { + else if ((p = strstr(s, pattern.buf)) && + starts_with_newline(p + pattern.len)) { size_t newlen = p - s + 1; if (newlen < len) len = newlen; From 0fb397b9cd780c72321cc0b423d8b4a98165bd14 Mon Sep 17 00:00:00 2001 From: Jens Glathe Date: Tue, 2 Jun 2020 12:12:25 +0200 Subject: [PATCH 053/250] t0014: fix indentation For some reason, this test case was indented with 4 spaces instead of 1 horizontal tab. The other test cases in the same test script are fine. Signed-off-by: Jens Glathe Signed-off-by: Johannes Schindelin --- t/t0014-alias.sh | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/t/t0014-alias.sh b/t/t0014-alias.sh index 854d59ec58c25a..30708146887d19 100755 --- a/t/t0014-alias.sh +++ b/t/t0014-alias.sh @@ -38,10 +38,10 @@ test_expect_success 'looping aliases - internal execution' ' #' test_expect_success 'run-command formats empty args properly' ' - test_must_fail env GIT_TRACE=1 git frotz a "" b " " c 2>actual.raw && - sed -ne "/run_command:/s/.*trace: run_command: //p" actual.raw >actual && - echo "git-frotz a '\'''\'' b '\'' '\'' c" >expect && - test_cmp expect actual + test_must_fail env GIT_TRACE=1 git frotz a "" b " " c 2>actual.raw && + sed -ne "/run_command:/s/.*trace: run_command: //p" actual.raw >actual && + echo "git-frotz a '\'''\'' b '\'' '\'' c" >expect && + test_cmp expect actual ' test_expect_success 'tracing a shell alias with arguments shows trace of prepared command' ' From 1ed143af999a83168f801bc1189e6170d7bc04ef Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Wed, 12 Aug 2020 15:06:17 +0000 Subject: [PATCH 054/250] git-gui: accommodate for intent-to-add files As of Git v2.28.0, the diff for files staged via `git add -N` marks them as new files. Git GUI was ill-prepared for that, and this patch teaches Git GUI about them. Please note that this will not even fix things with v2.28.0, as the `rp/apply-cached-with-i-t-a` patches are required on Git's side, too. This fixes https://github.com/git-for-windows/git/issues/2779 Signed-off-by: Johannes Schindelin Signed-off-by: Pratyush Yadav --- git-gui/git-gui.sh | 2 ++ git-gui/lib/diff.tcl | 12 ++++++++---- 2 files changed, 10 insertions(+), 4 deletions(-) diff --git a/git-gui/git-gui.sh b/git-gui/git-gui.sh index 28572c889c0e2b..5149c4bab26cc2 100755 --- a/git-gui/git-gui.sh +++ b/git-gui/git-gui.sh @@ -2086,6 +2086,7 @@ set all_icons(U$ui_index) file_merge set all_icons(T$ui_index) file_statechange set all_icons(_$ui_workdir) file_plain +set all_icons(A$ui_workdir) file_plain set all_icons(M$ui_workdir) file_mod set all_icons(D$ui_workdir) file_question set all_icons(U$ui_workdir) file_merge @@ -2112,6 +2113,7 @@ foreach i { {A_ {mc "Staged for commit"}} {AM {mc "Portions staged for commit"}} {AD {mc "Staged for commit, missing"}} + {AA {mc "Intended to be added"}} {_D {mc "Missing"}} {D_ {mc "Staged for removal"}} diff --git a/git-gui/lib/diff.tcl b/git-gui/lib/diff.tcl index d657bfec05b498..9a593a3f6e1547 100644 --- a/git-gui/lib/diff.tcl +++ b/git-gui/lib/diff.tcl @@ -568,7 +568,8 @@ proc apply_or_revert_hunk {x y revert} { if {$current_diff_side eq $ui_index} { set failed_msg [mc "Failed to unstage selected hunk."] lappend apply_cmd --reverse --cached - if {[string index $mi 0] ne {M}} { + set file_state [string index $mi 0] + if {$file_state ne {M} && $file_state ne {A}} { unlock_index return } @@ -581,7 +582,8 @@ proc apply_or_revert_hunk {x y revert} { lappend apply_cmd --cached } - if {[string index $mi 1] ne {M}} { + set file_state [string index $mi 1] + if {$file_state ne {M} && $file_state ne {A}} { unlock_index return } @@ -673,7 +675,8 @@ proc apply_or_revert_range_or_line {x y revert} { set failed_msg [mc "Failed to unstage selected line."] set to_context {+} lappend apply_cmd --reverse --cached - if {[string index $mi 0] ne {M}} { + set file_state [string index $mi 0] + if {$file_state ne {M} && $file_state ne {A}} { unlock_index return } @@ -688,7 +691,8 @@ proc apply_or_revert_range_or_line {x y revert} { lappend apply_cmd --cached } - if {[string index $mi 1] ne {M}} { + set file_state [string index $mi 1] + if {$file_state ne {M} && $file_state ne {A}} { unlock_index return } From 3e0bf9baa348a55369b552f047196bec7fb77ba7 Mon Sep 17 00:00:00 2001 From: Jeff Hostetler Date: Mon, 5 Apr 2021 15:27:38 -0400 Subject: [PATCH 055/250] Makefile: clean up .ilk files when MSVC=1 Signed-off-by: Jeff Hostetler --- Makefile | 3 +++ 1 file changed, 3 insertions(+) diff --git a/Makefile b/Makefile index 70d1543b6b8688..5152d855054717 100644 --- a/Makefile +++ b/Makefile @@ -3799,12 +3799,15 @@ ifdef MSVC $(RM) $(patsubst %.o,%.o.pdb,$(OBJECTS)) $(RM) headless-git.o.pdb $(RM) $(patsubst %.exe,%.pdb,$(OTHER_PROGRAMS)) + $(RM) $(patsubst %.exe,%.ilk,$(OTHER_PROGRAMS)) $(RM) $(patsubst %.exe,%.iobj,$(OTHER_PROGRAMS)) $(RM) $(patsubst %.exe,%.ipdb,$(OTHER_PROGRAMS)) $(RM) $(patsubst %.exe,%.pdb,$(PROGRAMS)) + $(RM) $(patsubst %.exe,%.ilk,$(PROGRAMS)) $(RM) $(patsubst %.exe,%.iobj,$(PROGRAMS)) $(RM) $(patsubst %.exe,%.ipdb,$(PROGRAMS)) $(RM) $(patsubst %.exe,%.pdb,$(TEST_PROGRAMS)) + $(RM) $(patsubst %.exe,%.ilk,$(TEST_PROGRAMS)) $(RM) $(patsubst %.exe,%.iobj,$(TEST_PROGRAMS)) $(RM) $(patsubst %.exe,%.ipdb,$(TEST_PROGRAMS)) $(RM) compat/vcbuild/MSVC-DEFS-GEN From 817fe8f836c55c714b8e25adbc27a6007a157029 Mon Sep 17 00:00:00 2001 From: Ian Bearman Date: Fri, 31 Jan 2020 16:00:25 -0800 Subject: [PATCH 056/250] vcbuild: install ARM64 dependencies when building ARM64 binaries Co-authored-by: Dennis Ameling Signed-off-by: Ian Bearman Signed-off-by: Dennis Ameling Signed-off-by: Johannes Schindelin --- compat/vcbuild/README | 6 +++++- compat/vcbuild/vcpkg_copy_dlls.bat | 7 ++++++- compat/vcbuild/vcpkg_install.bat | 9 +++++++-- 3 files changed, 18 insertions(+), 4 deletions(-) diff --git a/compat/vcbuild/README b/compat/vcbuild/README index 29ec1d0f104b80..1df1cabb1ebbbd 100644 --- a/compat/vcbuild/README +++ b/compat/vcbuild/README @@ -6,7 +6,11 @@ The Steps to Build Git with VS2015 or VS2017 from the command line. Prompt or from an SDK bash window: $ cd - $ ./compat/vcbuild/vcpkg_install.bat + $ ./compat/vcbuild/vcpkg_install.bat x64-windows + + or + + $ ./compat/vcbuild/vcpkg_install.bat arm64-windows The vcpkg tools and all of the third-party sources will be installed in this folder: diff --git a/compat/vcbuild/vcpkg_copy_dlls.bat b/compat/vcbuild/vcpkg_copy_dlls.bat index 13661c14f8705c..8bea0cbf83b6cf 100644 --- a/compat/vcbuild/vcpkg_copy_dlls.bat +++ b/compat/vcbuild/vcpkg_copy_dlls.bat @@ -15,7 +15,12 @@ REM ================================================================ @FOR /F "delims=" %%D IN ("%~dp0") DO @SET cwd=%%~fD cd %cwd% - SET arch=x64-windows + SET arch=%2 + IF NOT DEFINED arch ( + echo defaulting to 'x64-windows`. Invoke %0 with 'x86-windows', 'x64-windows', or 'arm64-windows' + set arch=x64-windows + ) + SET inst=%cwd%vcpkg\installed\%arch% IF [%1]==[release] ( diff --git a/compat/vcbuild/vcpkg_install.bat b/compat/vcbuild/vcpkg_install.bat index 8330d8120fb511..cacef18c11dc79 100644 --- a/compat/vcbuild/vcpkg_install.bat +++ b/compat/vcbuild/vcpkg_install.bat @@ -31,6 +31,12 @@ REM ================================================================ SETLOCAL EnableDelayedExpansion + SET arch=%1 + IF NOT DEFINED arch ( + echo defaulting to 'x64-windows`. Invoke %0 with 'x86-windows', 'x64-windows', or 'arm64-windows' + set arch=x64-windows + ) + @FOR /F "delims=" %%D IN ("%~dp0") DO @SET cwd=%%~fD cd %cwd% @@ -55,9 +61,8 @@ REM ================================================================ echo Successfully installed %cwd%vcpkg\vcpkg.exe :install_libraries - SET arch=x64-windows - echo Installing third-party libraries... + echo Installing third-party libraries(%arch%)... FOR %%i IN (zlib expat libiconv openssl libssh2 curl) DO ( cd %cwd%vcpkg IF NOT EXIST "packages\%%i_%arch%" CALL :sub__install_one %%i From 5967d401c2b38634682fdeb7d27f087265616e46 Mon Sep 17 00:00:00 2001 From: Jeff Hostetler Date: Mon, 5 Apr 2021 14:08:22 -0400 Subject: [PATCH 057/250] vcbuild: add support for compiling Windows resource files Create a wrapper for the Windows Resource Compiler (RC.EXE) for use by the MSVC=1 builds. This is similar to the CL.EXE and LIB.EXE wrappers used for the MSVC=1 builds. Signed-off-by: Jeff Hostetler --- compat/vcbuild/find_vs_env.bat | 7 ++++++ compat/vcbuild/scripts/rc.pl | 46 ++++++++++++++++++++++++++++++++++ config.mak.uname | 3 ++- 3 files changed, 55 insertions(+), 1 deletion(-) create mode 100644 compat/vcbuild/scripts/rc.pl diff --git a/compat/vcbuild/find_vs_env.bat b/compat/vcbuild/find_vs_env.bat index b35d264c0e6bed..379b16296e09c2 100644 --- a/compat/vcbuild/find_vs_env.bat +++ b/compat/vcbuild/find_vs_env.bat @@ -99,6 +99,7 @@ REM ================================================================ SET sdk_dir=%WindowsSdkDir% SET sdk_ver=%WindowsSDKVersion% + SET sdk_ver_bin_dir=%WindowsSdkVerBinPath%%tgt% SET si=%sdk_dir%Include\%sdk_ver% SET sdk_includes=-I"%si%ucrt" -I"%si%um" -I"%si%shared" SET sl=%sdk_dir%lib\%sdk_ver% @@ -130,6 +131,7 @@ REM ================================================================ SET sdk_dir=%WindowsSdkDir% SET sdk_ver=%WindowsSDKVersion% + SET sdk_ver_bin_dir=%WindowsSdkVerBinPath%bin\amd64 SET si=%sdk_dir%Include\%sdk_ver% SET sdk_includes=-I"%si%ucrt" -I"%si%um" -I"%si%shared" -I"%si%winrt" SET sl=%sdk_dir%lib\%sdk_ver% @@ -160,6 +162,11 @@ REM ================================================================ echo msvc_includes=%msvc_includes% echo msvc_libs=%msvc_libs% + echo sdk_ver_bin_dir=%sdk_ver_bin_dir% + SET X1=%sdk_ver_bin_dir:C:=/C% + SET X2=%X1:\=/% + echo sdk_ver_bin_dir_msys=%X2% + echo sdk_includes=%sdk_includes% echo sdk_libs=%sdk_libs% diff --git a/compat/vcbuild/scripts/rc.pl b/compat/vcbuild/scripts/rc.pl new file mode 100644 index 00000000000000..7bca4cd81c6c63 --- /dev/null +++ b/compat/vcbuild/scripts/rc.pl @@ -0,0 +1,46 @@ +#!/usr/bin/perl -w +###################################################################### +# Compile Resources on Windows +# +# This is a wrapper to facilitate the compilation of Git with MSVC +# using GNU Make as the build system. So, instead of manipulating the +# Makefile into something nasty, just to support non-space arguments +# etc, we use this wrapper to fix the command line options +# +###################################################################### +use strict; +my @args = (); +my @input = (); + +while (@ARGV) { + my $arg = shift @ARGV; + if ("$arg" =~ /^-[dD]/) { + # GIT_VERSION gets passed with too many + # layers of dquote escaping. + $arg =~ s/\\"/"/g; + + push(@args, $arg); + + } elsif ("$arg" eq "-i") { + my $arg = shift @ARGV; + # TODO complain if NULL or is dashed ?? + push(@input, $arg); + + } elsif ("$arg" eq "-o") { + my $arg = shift @ARGV; + # TODO complain if NULL or is dashed ?? + push(@args, "-fo$arg"); + + } else { + push(@args, $arg); + } +} + +push(@args, "-nologo"); +push(@args, "-v"); +push(@args, @input); + +unshift(@args, "rc.exe"); +printf("**** @args\n"); + +exit (system(@args) != 0); diff --git a/config.mak.uname b/config.mak.uname index 1ebaa8b17692b9..4fe429463bc901 100644 --- a/config.mak.uname +++ b/config.mak.uname @@ -448,7 +448,7 @@ ifeq ($(uname_S),Windows) # link.exe next to, and required by, cl.exe, we have to prepend this # onto the existing $PATH. # - SANE_TOOL_PATH ?= $(msvc_bin_dir_msys) + SANE_TOOL_PATH ?= $(msvc_bin_dir_msys):$(sdk_ver_bin_dir_msys) HAVE_ALLOCA_H = YesPlease NO_PREAD = YesPlease NEEDS_CRYPTO_WITH_SSL = YesPlease @@ -516,6 +516,7 @@ endif # See https://msdn.microsoft.com/en-us/library/ms235330.aspx EXTLIBS = user32.lib advapi32.lib shell32.lib wininet.lib ws2_32.lib invalidcontinue.obj kernel32.lib ntdll.lib PTHREAD_LIBS = + RC = compat/vcbuild/scripts/rc.pl lib = BASIC_CFLAGS += $(vcpkg_inc) $(sdk_includes) $(msvc_includes) ifndef DEBUG From 5acde99e74dba49ecbad2b9e65d59b9b6019e2b8 Mon Sep 17 00:00:00 2001 From: Ian Bearman Date: Tue, 4 Feb 2020 10:34:40 -0800 Subject: [PATCH 058/250] vcbuild: add an option to install individual 'features' In this context, a "feature" is a dependency combined with its own dependencies. Signed-off-by: Ian Bearman Signed-off-by: Johannes Schindelin --- compat/vcbuild/vcpkg_install.bat | 35 +++++++++++++++++++++++++++++++- 1 file changed, 34 insertions(+), 1 deletion(-) diff --git a/compat/vcbuild/vcpkg_install.bat b/compat/vcbuild/vcpkg_install.bat index cacef18c11dc79..8da212487ae97d 100644 --- a/compat/vcbuild/vcpkg_install.bat +++ b/compat/vcbuild/vcpkg_install.bat @@ -85,14 +85,47 @@ REM ================================================================ :sub__install_one echo Installing package %1... + call :%1_features + REM vcpkg may not be reliable on slow, intermittent or proxy REM connections, see e.g. REM https://social.msdn.microsoft.com/Forums/windowsdesktop/en-US/4a8f7be5-5e15-4213-a7bb-ddf424a954e6/winhttpsendrequest-ends-with-12002-errorhttptimeout-after-21-seconds-no-matter-what-timeout?forum=windowssdk REM which explains the hidden 21 second timeout REM (last post by Dave : Microsoft - Windows Networking team) - .\vcpkg.exe install %1:%arch% + .\vcpkg.exe install %1%features%:%arch% IF ERRORLEVEL 1 ( EXIT /B 1 ) echo Finished %1 goto :EOF + +:: +:: features for each vcpkg to install +:: there should be an entry here for each package to install +:: 'set features=' means use the default otherwise +:: 'set features=[comma-delimited-feature-set]' is the syntax +:: + +:zlib_features +set features= +goto :EOF + +:expat_features +set features= +goto :EOF + +:libiconv_features +set features= +goto :EOF + +:openssl_features +set features= +goto :EOF + +:libssh2_features +set features= +goto :EOF + +:curl_features +set features=[core,openssl] +goto :EOF From b57cf141f5ef6fdfa26027e1eb36d1470dd80250 Mon Sep 17 00:00:00 2001 From: Jeff Hostetler Date: Mon, 5 Apr 2021 14:12:14 -0400 Subject: [PATCH 059/250] config.mak.uname: add git.rc to MSVC builds Teach MSVC=1 builds to depend on the `git.rc` file so that the resulting executables have Windows-style resources and version number information within them. Signed-off-by: Jeff Hostetler --- config.mak.uname | 1 + 1 file changed, 1 insertion(+) diff --git a/config.mak.uname b/config.mak.uname index 4fe429463bc901..9daf49b1f055f8 100644 --- a/config.mak.uname +++ b/config.mak.uname @@ -515,6 +515,7 @@ endif # handle twice, or to access the osfhandle of an already-closed stdout # See https://msdn.microsoft.com/en-us/library/ms235330.aspx EXTLIBS = user32.lib advapi32.lib shell32.lib wininet.lib ws2_32.lib invalidcontinue.obj kernel32.lib ntdll.lib + GITLIBS += git.res PTHREAD_LIBS = RC = compat/vcbuild/scripts/rc.pl lib = From 06e6a98d22a64da25e0606d34c0e416ad75716f8 Mon Sep 17 00:00:00 2001 From: Dennis Ameling Date: Fri, 4 Dec 2020 14:11:34 +0100 Subject: [PATCH 060/250] cmake: allow building for Windows/ARM64 Signed-off-by: Dennis Ameling Signed-off-by: Johannes Schindelin --- contrib/buildsystems/CMakeLists.txt | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/contrib/buildsystems/CMakeLists.txt b/contrib/buildsystems/CMakeLists.txt index 25b495fa737d07..aaaee1bc4b7943 100644 --- a/contrib/buildsystems/CMakeLists.txt +++ b/contrib/buildsystems/CMakeLists.txt @@ -65,9 +65,9 @@ if(USE_VCPKG) set(VCPKG_DIR "${CMAKE_SOURCE_DIR}/compat/vcbuild/vcpkg") if(NOT EXISTS ${VCPKG_DIR}) message("Initializing vcpkg and building the Git's dependencies (this will take a while...)") - execute_process(COMMAND ${CMAKE_SOURCE_DIR}/compat/vcbuild/vcpkg_install.bat) + execute_process(COMMAND ${CMAKE_SOURCE_DIR}/compat/vcbuild/vcpkg_install.bat ${VCPKG_ARCH}) endif() - list(APPEND CMAKE_PREFIX_PATH "${VCPKG_DIR}/installed/x64-windows") + list(APPEND CMAKE_PREFIX_PATH "${VCPKG_DIR}/installed/${VCPKG_ARCH}") # In the vcpkg edition, we need this to be able to link to libcurl set(CURL_NO_CURL_CMAKE ON) @@ -1211,7 +1211,7 @@ string(REPLACE "@USE_LIBPCRE2@" "" git_build_options "${git_build_options}") string(REPLACE "@WITH_BREAKING_CHANGES@" "" git_build_options "${git_build_options}") string(REPLACE "@X@" "${EXE_EXTENSION}" git_build_options "${git_build_options}") if(USE_VCPKG) - string(APPEND git_build_options "PATH=\"$PATH:$TEST_DIRECTORY/../compat/vcbuild/vcpkg/installed/x64-windows/bin\"\n") + string(APPEND git_build_options "PATH=\"$PATH:$TEST_DIRECTORY/../compat/vcbuild/vcpkg/installed/${VCPKG_ARCH}/bin\"\n") endif() file(WRITE ${CMAKE_BINARY_DIR}/GIT-BUILD-OPTIONS ${git_build_options}) From 07a014398040c12a0f62f51935f3d845267dd7c5 Mon Sep 17 00:00:00 2001 From: Jeff Hostetler Date: Mon, 5 Apr 2021 14:24:52 -0400 Subject: [PATCH 061/250] clink.pl: ignore no-stack-protector arg on MSVC=1 builds Ignore the `-fno-stack-protector` compiler argument when building with MSVC. This will be used in a later commit that needs to build a Win32 GUI app. Signed-off-by: Jeff Hostetler --- compat/vcbuild/scripts/clink.pl | 2 ++ 1 file changed, 2 insertions(+) diff --git a/compat/vcbuild/scripts/clink.pl b/compat/vcbuild/scripts/clink.pl index 2768ae15f1879f..73c8a2b184f38b 100755 --- a/compat/vcbuild/scripts/clink.pl +++ b/compat/vcbuild/scripts/clink.pl @@ -122,6 +122,8 @@ push(@cflags, "-wd4996"); } elsif ("$arg" =~ /^-W[a-z]/) { # let's ignore those + } elsif ("$arg" eq "-fno-stack-protector") { + # eat this } else { push(@args, $arg); } From 372ccc705be8e5cf6a741fbf5ce685c2a4ceafbd Mon Sep 17 00:00:00 2001 From: Philip Oakley Date: Sun, 6 Oct 2019 18:40:55 +0100 Subject: [PATCH 062/250] vcpkg_install: detect lack of Git The vcpkg_install batch file depends on the availability of a working Git on the CMD path. This may not be present if the user has selected the 'bash only' option during Git-for-Windows install. Detect and tell the user about their lack of a working Git in the CMD window. Fixes #2348. A separate PR https://github.com/git-for-windows/build-extra/pull/258 now highlights the recommended path setting during install. Signed-off-by: Philip Oakley --- compat/vcbuild/vcpkg_install.bat | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/compat/vcbuild/vcpkg_install.bat b/compat/vcbuild/vcpkg_install.bat index ebd0bad242a8ca..bcbbf536af3141 100644 --- a/compat/vcbuild/vcpkg_install.bat +++ b/compat/vcbuild/vcpkg_install.bat @@ -36,6 +36,13 @@ REM ================================================================ dir vcpkg\vcpkg.exe >nul 2>nul && GOTO :install_libraries + git.exe version 2>nul + IF ERRORLEVEL 1 ( + echo "***" + echo "Git not found. Please adjust your CMD path or Git install option." + echo "***" + EXIT /B 1 ) + echo Fetching vcpkg in %cwd%vcpkg git.exe clone https://github.com/Microsoft/vcpkg vcpkg IF ERRORLEVEL 1 ( EXIT /B 1 ) From 4f49d9e93025d189c0fa9f5cdee14d3f001090d6 Mon Sep 17 00:00:00 2001 From: Dennis Ameling Date: Sun, 29 Nov 2020 00:12:26 +0100 Subject: [PATCH 063/250] ci(vs-build) also build Windows/ARM64 artifacts There are no Windows/ARM64 agents in GitHub Actions yet, therefore we just skip adjusting the `vs-test` job for now. Signed-off-by: Dennis Ameling Signed-off-by: Johannes Schindelin --- .github/workflows/main.yml | 17 ++++++++++------- 1 file changed, 10 insertions(+), 7 deletions(-) diff --git a/.github/workflows/main.yml b/.github/workflows/main.yml index 7dbf9f7f123c89..78c42a589d24bf 100644 --- a/.github/workflows/main.yml +++ b/.github/workflows/main.yml @@ -169,8 +169,11 @@ jobs: NO_PERL: 1 GIT_CONFIG_PARAMETERS: "'user.name=CI' 'user.email=ci@git'" runs-on: windows-latest + strategy: + matrix: + arch: [x64, arm64] concurrency: - group: vs-build-${{ github.ref }} + group: vs-build-${{ github.ref }}-${{ matrix.arch }} cancel-in-progress: ${{ needs.ci-config.outputs.skip_concurrent == 'yes' }} steps: - uses: actions/checkout@v4 @@ -189,14 +192,14 @@ jobs: uses: microsoft/setup-msbuild@v2 - name: copy dlls to root shell: cmd - run: compat\vcbuild\vcpkg_copy_dlls.bat release + run: compat\vcbuild\vcpkg_copy_dlls.bat release ${{ matrix.arch }}-windows - name: generate Visual Studio solution shell: bash run: | - cmake `pwd`/contrib/buildsystems/ -DCMAKE_PREFIX_PATH=`pwd`/compat/vcbuild/vcpkg/installed/x64-windows \ - -DNO_GETTEXT=YesPlease -DPERL_TESTS=OFF -DPYTHON_TESTS=OFF -DCURL_NO_CURL_CMAKE=ON + cmake `pwd`/contrib/buildsystems/ -DCMAKE_PREFIX_PATH=`pwd`/compat/vcbuild/vcpkg/installed/${{ matrix.arch }}-windows \ + -DNO_GETTEXT=YesPlease -DPERL_TESTS=OFF -DPYTHON_TESTS=OFF -DCURL_NO_CURL_CMAKE=ON -DCMAKE_GENERATOR_PLATFORM=${{ matrix.arch }} -DVCPKG_ARCH=${{ matrix.arch }}-windows - name: MSBuild - run: msbuild git.sln -property:Configuration=Release -property:Platform=x64 -maxCpuCount:4 -property:PlatformToolset=v142 + run: msbuild git.sln -property:Configuration=Release -property:Platform=${{ matrix.arch }} -maxCpuCount:4 -property:PlatformToolset=v142 - name: bundle artifact tar shell: bash env: @@ -210,7 +213,7 @@ jobs: - name: upload tracked files and build artifacts uses: actions/upload-artifact@v4 with: - name: vs-artifacts + name: vs-artifacts-${{ matrix.arch }} path: artifacts vs-test: name: win+VS test @@ -228,7 +231,7 @@ jobs: - name: download tracked files and build artifacts uses: actions/download-artifact@v4 with: - name: vs-artifacts + name: vs-artifacts-x64 path: ${{github.workspace}} - name: extract tracked files and build artifacts shell: bash From ceda56d552e3d1d017028acdb7e44d6eb5201fc7 Mon Sep 17 00:00:00 2001 From: Jeff Hostetler Date: Mon, 5 Apr 2021 14:39:33 -0400 Subject: [PATCH 064/250] clink.pl: move default linker options for MSVC=1 builds Move the default `-ENTRY` and `-SUBSYSTEM` arguments for MSVC=1 builds from `config.mak.uname` into `clink.pl`. These args are constant for console-mode executables. Add support to `clink.pl` for generating a Win32 GUI application using the `-mwindows` argument (to match how GCC does it). This changes the `-ENTRY` and `-SUBSYSTEM` arguments accordingly. Signed-off-by: Jeff Hostetler --- compat/vcbuild/scripts/clink.pl | 11 +++++++++++ config.mak.uname | 2 +- 2 files changed, 12 insertions(+), 1 deletion(-) diff --git a/compat/vcbuild/scripts/clink.pl b/compat/vcbuild/scripts/clink.pl index 73c8a2b184f38b..a38b360015ece9 100755 --- a/compat/vcbuild/scripts/clink.pl +++ b/compat/vcbuild/scripts/clink.pl @@ -15,6 +15,7 @@ my @lflags = (); my $is_linking = 0; my $is_debug = 0; +my $is_gui = 0; while (@ARGV) { my $arg = shift @ARGV; if ("$arg" eq "-DDEBUG") { @@ -124,11 +125,21 @@ # let's ignore those } elsif ("$arg" eq "-fno-stack-protector") { # eat this + } elsif ("$arg" eq "-mwindows") { + $is_gui = 1; } else { push(@args, $arg); } } if ($is_linking) { + if ($is_gui) { + push(@args, "-ENTRY:wWinMainCRTStartup"); + push(@args, "-SUBSYSTEM:WINDOWS"); + } else { + push(@args, "-ENTRY:wmainCRTStartup"); + push(@args, "-SUBSYSTEM:CONSOLE"); + } + push(@args, @lflags); unshift(@args, "link.exe"); } else { diff --git a/config.mak.uname b/config.mak.uname index 9daf49b1f055f8..4e21f43d233ac9 100644 --- a/config.mak.uname +++ b/config.mak.uname @@ -510,7 +510,7 @@ endif compat/win32/trace2_win32_process_info.o \ compat/win32/dirent.o COMPAT_CFLAGS = -D__USE_MINGW_ACCESS -DDETECT_MSYS_TTY -DENSURE_MSYSTEM_IS_SET -DNOGDI -DHAVE_STRING_H -Icompat -Icompat/regex -Icompat/win32 -DSTRIP_EXTENSION=\".exe\" - BASIC_LDFLAGS = -IGNORE:4217 -IGNORE:4049 -NOLOGO -ENTRY:wmainCRTStartup -SUBSYSTEM:CONSOLE + BASIC_LDFLAGS = -IGNORE:4217 -IGNORE:4049 -NOLOGO # invalidcontinue.obj allows Git's source code to close the same file # handle twice, or to access the osfhandle of an already-closed stdout # See https://msdn.microsoft.com/en-us/library/ms235330.aspx From 68f7e595ed5e1af004993dbb7e19ea458bf2acac Mon Sep 17 00:00:00 2001 From: Yuyi Wang Date: Sat, 11 Mar 2023 17:51:18 +0800 Subject: [PATCH 065/250] cmake: install headless-git. headless-git is a git executable without opening a console window. It is useful when other GUI executables want to call git. We should install it together with git on Windows. Signed-off-by: Yuyi Wang --- contrib/buildsystems/CMakeLists.txt | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/contrib/buildsystems/CMakeLists.txt b/contrib/buildsystems/CMakeLists.txt index 25b495fa737d07..709f51120a683a 100644 --- a/contrib/buildsystems/CMakeLists.txt +++ b/contrib/buildsystems/CMakeLists.txt @@ -745,6 +745,7 @@ if(WIN32) endif() add_executable(headless-git ${CMAKE_SOURCE_DIR}/compat/win32/headless.c) + list(APPEND PROGRAMS_BUILT headless-git) if(CMAKE_C_COMPILER_ID STREQUAL "GNU" OR CMAKE_C_COMPILER_ID STREQUAL "Clang") target_link_options(headless-git PUBLIC -municode -Wl,-subsystem,windows) elseif(CMAKE_C_COMPILER_ID STREQUAL "MSVC") @@ -945,7 +946,7 @@ list(TRANSFORM git_perl_scripts PREPEND "${CMAKE_BINARY_DIR}/") #install foreach(program ${PROGRAMS_BUILT}) -if(program MATCHES "^(git|git-shell|scalar)$") +if(program MATCHES "^(git|git-shell|headless-git|scalar)$") install(TARGETS ${program} RUNTIME DESTINATION bin) else() From 459ec1e455241430fec7c244ac69d0e6d9bb275f Mon Sep 17 00:00:00 2001 From: Philip Oakley Date: Sun, 6 Oct 2019 18:43:57 +0100 Subject: [PATCH 066/250] vcpkg_install: add comment regarding slow network connections The vcpkg downloads may not succeed. Warn careful readers of the time out. A simple retry will usually resolve the issue. Signed-off-by: Philip Oakley Signed-off-by: Johannes Schindelin --- compat/vcbuild/vcpkg_install.bat | 6 ++++++ 1 file changed, 6 insertions(+) diff --git a/compat/vcbuild/vcpkg_install.bat b/compat/vcbuild/vcpkg_install.bat index bcbbf536af3141..8330d8120fb511 100644 --- a/compat/vcbuild/vcpkg_install.bat +++ b/compat/vcbuild/vcpkg_install.bat @@ -80,6 +80,12 @@ REM ================================================================ :sub__install_one echo Installing package %1... + REM vcpkg may not be reliable on slow, intermittent or proxy + REM connections, see e.g. + REM https://social.msdn.microsoft.com/Forums/windowsdesktop/en-US/4a8f7be5-5e15-4213-a7bb-ddf424a954e6/winhttpsendrequest-ends-with-12002-errorhttptimeout-after-21-seconds-no-matter-what-timeout?forum=windowssdk + REM which explains the hidden 21 second timeout + REM (last post by Dave : Microsoft - Windows Networking team) + .\vcpkg.exe install %1:%arch% IF ERRORLEVEL 1 ( EXIT /B 1 ) From acdbf8d7fdec2b2523eb2cf75595a39fee07073a Mon Sep 17 00:00:00 2001 From: Dennis Ameling Date: Sun, 6 Dec 2020 18:39:26 +0100 Subject: [PATCH 067/250] Add schannel to curl installation Signed-off-by: Dennis Ameling --- compat/vcbuild/vcpkg_install.bat | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/compat/vcbuild/vcpkg_install.bat b/compat/vcbuild/vcpkg_install.bat index 8da212487ae97d..575c65c20ba307 100644 --- a/compat/vcbuild/vcpkg_install.bat +++ b/compat/vcbuild/vcpkg_install.bat @@ -127,5 +127,5 @@ set features= goto :EOF :curl_features -set features=[core,openssl] +set features=[core,openssl,schannel] goto :EOF From 74c1ba8fdac20fa38e686fead51ad7d9629c8235 Mon Sep 17 00:00:00 2001 From: Dennis Ameling Date: Mon, 19 Jul 2021 13:02:16 +0200 Subject: [PATCH 068/250] cmake(): allow setting HOST_CPU for cross-compilation Git's regular Makefile mentions that HOST_CPU should be defined when cross-compiling Git: https://github.com/git-for-windows/git/blob/37796bca76ef4180c39ee508ca3e42c0777ba444/Makefile#L438-L439 This is then used to set the GIT_HOST_CPU variable when compiling Git: https://github.com/git-for-windows/git/blob/37796bca76ef4180c39ee508ca3e42c0777ba444/Makefile#L1337-L1341 Then, when the user runs `git version --build-options`, it returns that value: https://github.com/git-for-windows/git/blob/37796bca76ef4180c39ee508ca3e42c0777ba444/help.c#L658 This commit adds the same functionality to the CMake configuration. Users can now set -DHOST_CPU= to set the target architecture. Signed-off-by: Dennis Ameling --- .github/workflows/main.yml | 2 +- contrib/buildsystems/CMakeLists.txt | 9 ++++++++- 2 files changed, 9 insertions(+), 2 deletions(-) diff --git a/.github/workflows/main.yml b/.github/workflows/main.yml index 78c42a589d24bf..14e598bac16818 100644 --- a/.github/workflows/main.yml +++ b/.github/workflows/main.yml @@ -197,7 +197,7 @@ jobs: shell: bash run: | cmake `pwd`/contrib/buildsystems/ -DCMAKE_PREFIX_PATH=`pwd`/compat/vcbuild/vcpkg/installed/${{ matrix.arch }}-windows \ - -DNO_GETTEXT=YesPlease -DPERL_TESTS=OFF -DPYTHON_TESTS=OFF -DCURL_NO_CURL_CMAKE=ON -DCMAKE_GENERATOR_PLATFORM=${{ matrix.arch }} -DVCPKG_ARCH=${{ matrix.arch }}-windows + -DNO_GETTEXT=YesPlease -DPERL_TESTS=OFF -DPYTHON_TESTS=OFF -DCURL_NO_CURL_CMAKE=ON -DCMAKE_GENERATOR_PLATFORM=${{ matrix.arch }} -DVCPKG_ARCH=${{ matrix.arch }}-windows -DHOST_CPU=${{ matrix.arch }} - name: MSBuild run: msbuild git.sln -property:Configuration=Release -property:Platform=${{ matrix.arch }} -maxCpuCount:4 -property:PlatformToolset=v142 - name: bundle artifact tar diff --git a/contrib/buildsystems/CMakeLists.txt b/contrib/buildsystems/CMakeLists.txt index aaaee1bc4b7943..d61e54a7f6dd4e 100644 --- a/contrib/buildsystems/CMakeLists.txt +++ b/contrib/buildsystems/CMakeLists.txt @@ -212,7 +212,14 @@ endif() #default behaviour include_directories(${CMAKE_SOURCE_DIR}) -add_compile_definitions(GIT_HOST_CPU="${CMAKE_SYSTEM_PROCESSOR}") + +# When cross-compiling, define HOST_CPU as the canonical name of the CPU on +# which the built Git will run (for instance "x86_64"). +if(NOT HOST_CPU) + add_compile_definitions(GIT_HOST_CPU="${CMAKE_SYSTEM_PROCESSOR}") +else() + add_compile_definitions(GIT_HOST_CPU="${HOST_CPU}") +endif() add_compile_definitions(SHA256_BLK INTERNAL_QSORT RUNTIME_PREFIX) add_compile_definitions(NO_OPENSSL SHA1_DC SHA1DC_NO_STANDARD_INCLUDES SHA1DC_INIT_SAFE_HASH_DEFAULT=0 From babfd659f66e7b63cda9ae473bba5df5628fe728 Mon Sep 17 00:00:00 2001 From: Philip Oakley Date: Fri, 2 Jul 2021 00:30:24 +0100 Subject: [PATCH 069/250] CMake: default Visual Studio generator has changed Correct some wording and inform users regarding the Visual Studio changes (from V16.6) to the default generator. Subsequent commits ensure that Git for Windows can be directly opened in modern Visual Studio without needing special configuration of the CMakeLists settings. It appeares that internally Visual Studio creates it's own version of the .sln file (etc.) for extension tools that expect them. The large number of references below document the shifting of Visual Studio default and CMake setting options. refs: https://docs.microsoft.com/en-us/search/?scope=C%2B%2B&view=msvc-150&terms=Ninja 1. https://docs.microsoft.com/en-us/cpp/linux/cmake-linux-configure?view=msvc-160 (note the linux bit) "In Visual Studio 2019 version 16.6 or later ***, Ninja is the default generator for configurations targeting a remote system or WSL. For more information, see this post on the C++ Team Blog [https://devblogs.microsoft.com/cppblog/linux-development-with-visual-studio-first-class-support-for-gdbserver-improved-build-times-with-ninja-and-updates-to-the-connection-manager/]. For more information about these settings, see CMakeSettings.json reference [https://docs.microsoft.com/en-us/cpp/build/cmakesettings-reference?view=msvc-160]." 2. https://docs.microsoft.com/en-us/cpp/build/cmake-presets-vs?view=msvc-160 "CMake supports two files that allow users to specify common configure, build, and test options and share them with others: CMakePresets.json and CMakeUserPresets.json." " Both files are supported in Visual Studio 2019 version 16.10 or later. ***" 3. https://devblogs.microsoft.com/cppblog/linux-development-with-visual-studio-first-class-support-for-gdbserver-improved-build-times-with-ninja-and-updates-to-the-connection-manager/ " Ninja has been the default generator (underlying build system) for CMake configurations targeting Windows for some time***, but in Visual Studio 2019 version 16.6 Preview 3*** we added support for Ninja on Linux." 4. https://docs.microsoft.com/en-us/cpp/build/cmakesettings-reference?view=msvc-160 " `generator`: specifies CMake generator to use for this configuration. May be one of: Visual Studio 2019 only: Visual Studio 16 2019 Visual Studio 16 2019 Win64 Visual Studio 16 2019 ARM Visual Studio 2017 and later: Visual Studio 15 2017 Visual Studio 15 2017 Win64 Visual Studio 15 2017 ARM Visual Studio 14 2015 Visual Studio 14 2015 Win64 Visual Studio 14 2015 ARM Unix Makefiles Ninja Because Ninja is designed for fast build speeds instead of flexibility and function, it is set as the default. However, some CMake projects may be unable to correctly build using Ninja. If this occurs, you can instruct CMake to generate Visual Studio projects instead. To specify a Visual Studio generator in Visual Studio 2017, open the settings editor from the main menu by choosing CMake | Change CMake Settings. Delete "Ninja" and type "V". This activates IntelliSense, which enables you to choose the generator you want." "To specify a Visual Studio generator in Visual Studio 2019, right-click on the CMakeLists.txt file in Solution Explorer and choose CMake Settings for project > Show Advanced Settings > CMake Generator. When the active configuration specifies a Visual Studio generator, by default MSBuild.exe is invoked with` -m -v:minimal` arguments." 5. https://docs.microsoft.com/en-us/cpp/build/cmake-presets-vs?view=msvc-160#enable-cmakepresetsjson-integration-in-visual-studio-2019 "Enable CMakePresets.json integration in Visual Studio 2019 CMakePresets.json integration isn't enabled by default in Visual Studio 2019. You can enable it for all CMake projects in Tools > Options > CMake > General: (tick a box)" ... see more. 6. https://docs.microsoft.com/en-us/cpp/build/cmakesettings-reference?view=msvc-140 (whichever v140 is..) "CMake projects are supported in Visual Studio 2017 and later." 7. https://docs.microsoft.com/en-us/cpp/overview/what-s-new-for-cpp-2017?view=msvc-150 "Support added for the CMake Ninja generator." 8. https://docs.microsoft.com/en-us/cpp/overview/what-s-new-for-cpp-2017?view=msvc-150#cmake-support-via-open-folder "CMake support via Open Folder Visual Studio 2017 introduces support for using CMake projects without converting to MSBuild project files (.vcxproj). For more information, see CMake projects in Visual Studio[https://docs.microsoft.com/en-us/cpp/build/cmake-projects-in-visual-studio?view=msvc-150]. Opening CMake projects with Open Folder automatically configures the environment for C++ editing, building, and debugging." ... +more! 9. https://docs.microsoft.com/en-us/cpp/build/cmake-presets-vs?view=msvc-160#supported-cmake-and-cmakepresetsjson-versions "Visual Studio reads and evaluates CMakePresets.json and CMakeUserPresets.json itself and doesn't invoke CMake directly with the --preset option. So, CMake version 3.20 or later isn't strictly required when you're building with CMakePresets.json inside Visual Studio. We recommend using CMake version 3.14 or later." 10. https://docs.microsoft.com/en-us/cpp/build/cmake-presets-vs?view=msvc-160#enable-cmakepresetsjson-integration-in-visual-studio-2019 "If you don't want to enable CMakePresets.json integration for all CMake projects, you can enable CMakePresets.json integration for a single CMake project by adding a CMakePresets.json file to the root of the open folder. You must close and reopen the folder in Visual Studio to activate the integration. 11. https://docs.microsoft.com/en-us/cpp/build/cmake-presets-vs?view=msvc-160#default-configure-presets ***(doesn't actually say which version..) "Default Configure Presets If no CMakePresets.json or CMakeUserPresets.json file exists, or if CMakePresets.json or CMakeUserPresets.json is invalid, Visual Studio will fall back*** on the following default Configure Presets: Windows example JSON { "name": "windows-default", "displayName": "Windows x64 Debug", "description": "Sets Ninja generator, compilers, x64 architecture, build and install directory, debug build type", "generator": "Ninja", "binaryDir": "${sourceDir}/out/build/${presetName}", "architecture": { "value": "x64", "strategy": "external" }, "cacheVariables": { "CMAKE_BUILD_TYPE": "Debug", "CMAKE_INSTALL_PREFIX": "${sourceDir}/out/install/${presetName}" }, "vendor": { "microsoft.com/VisualStudioSettings/CMake/1.0": { "hostOS": [ "Windows" ] } } }, " Signed-off-by: Philip Oakley --- contrib/buildsystems/CMakeLists.txt | 16 ++++++++++------ 1 file changed, 10 insertions(+), 6 deletions(-) diff --git a/contrib/buildsystems/CMakeLists.txt b/contrib/buildsystems/CMakeLists.txt index d61e54a7f6dd4e..30fa3250c75bf1 100644 --- a/contrib/buildsystems/CMakeLists.txt +++ b/contrib/buildsystems/CMakeLists.txt @@ -14,6 +14,11 @@ Note: Visual Studio also has the option of opening `CMakeLists.txt` directly; Using this option, Visual Studio will not find the source code, though, therefore the `File>Open>Folder...` option is preferred. +Visual Studio does not produce a .sln solution file nor the .vcxproj files +that may be required by VS extension tools. + +To generate the .sln/.vcxproj files run CMake manually, as described below. + Instructions to run CMake manually: mkdir -p contrib/buildsystems/out @@ -22,7 +27,7 @@ Instructions to run CMake manually: This will build the git binaries in contrib/buildsystems/out directory (our top-level .gitignore file knows to ignore contents of -this directory). +this directory). The project .sln and .vcxproj files are also generated. Possible build configurations(-DCMAKE_BUILD_TYPE) with corresponding compiler flags @@ -35,17 +40,16 @@ empty(default) : NOTE: -DCMAKE_BUILD_TYPE is optional. For multi-config generators like Visual Studio this option is ignored -This process generates a Makefile(Linux/*BSD/MacOS) , Visual Studio solution(Windows) by default. +This process generates a Makefile(Linux/*BSD/MacOS), Visual Studio solution(Windows) by default. Run `make` to build Git on Linux/*BSD/MacOS. Open git.sln on Windows and build Git. -NOTE: By default CMake uses Makefile as the build tool on Linux and Visual Studio in Windows, -to use another tool say `ninja` add this to the command line when configuring. -`-G Ninja` - NOTE: By default CMake will install vcpkg locally to your source tree on configuration, to avoid this, add `-DNO_VCPKG=TRUE` to the command line when configuring. +The Visual Studio default generator changed in v16.6 from its Visual Studio +implemenation to `Ninja` This required changes to many CMake scripts. + ]] cmake_minimum_required(VERSION 3.14) From 127ef69e3d20667b9cd824829bb542e8c6ec6e54 Mon Sep 17 00:00:00 2001 From: Philip Oakley Date: Sun, 31 Oct 2021 23:15:13 +0000 Subject: [PATCH 070/250] hash-object: demonstrate a >4GB/LLP64 problem On LLP64 systems, such as Windows, the size of `long`, `int`, etc. is only 32 bits (for backward compatibility). Git's use of `unsigned long` for file memory sizes in many places, rather than size_t, limits the handling of large files on LLP64 systems (commonly given as `>4GB`). Provide a minimum test for handling a >4GB file. The `hash-object` command, with the `--literally` and without `-w` option avoids writing the object, either loose or packed. This avoids the code paths hitting the `bigFileThreshold` config test code, the zlib code, and the pack code. Subsequent patches will walk the test's call chain, converting types to `size_t` (which is larger in LLP64 data models) where appropriate. Signed-off-by: Philip Oakley Signed-off-by: Johannes Schindelin --- t/t1007-hash-object.sh | 11 +++++++++++ 1 file changed, 11 insertions(+) diff --git a/t/t1007-hash-object.sh b/t/t1007-hash-object.sh index dbbe9fb0d4b19b..09741ea3d6c4fe 100755 --- a/t/t1007-hash-object.sh +++ b/t/t1007-hash-object.sh @@ -49,6 +49,9 @@ test_expect_success 'setup' ' example sha1:ddd3f836d3e3fbb7ae289aa9ae83536f76956399 example sha256:b44fe1fe65589848253737db859bd490453510719d7424daab03daf0767b85ae + + large5GB sha1:0be2be10a4c8764f32c4bf372a98edc731a4b204 + large5GB sha256:dc18ca621300c8d3cfa505a275641ebab00de189859e022a975056882d313e64 EOF ' @@ -258,4 +261,12 @@ test_expect_success '--stdin outside of repository (uses SHA-1)' ' test_cmp expect actual ' +test_expect_failure EXPENSIVE,SIZE_T_IS_64BIT,!LONG_IS_64BIT \ + 'files over 4GB hash literally' ' + test-tool genzeros $((5*1024*1024*1024)) >big && + test_oid large5GB >expect && + git hash-object --stdin --literally actual && + test_cmp expect actual +' + test_done From c65a9ecb430fb8ec085f107af52f9ae22fd2c1dc Mon Sep 17 00:00:00 2001 From: Philip Oakley Date: Sat, 24 Apr 2021 11:09:58 +0100 Subject: [PATCH 071/250] .gitignore: add Visual Studio CMakeSetting.json file The CMakeSettings.json file is tool generated. Developers may track it should they provide additional settings. Signed-off-by: Philip Oakley --- .gitignore | 1 + 1 file changed, 1 insertion(+) diff --git a/.gitignore b/.gitignore index 04c444404e4ba8..735bb036415bd5 100644 --- a/.gitignore +++ b/.gitignore @@ -252,5 +252,6 @@ Release/ /git.VC.db *.dSYM /contrib/buildsystems/out +CMakeSettings.json /contrib/libgit-rs/target /contrib/libgit-sys/target From 3e12b5d748ad37c3005128c4b965c39d02a7d902 Mon Sep 17 00:00:00 2001 From: Philip Oakley Date: Fri, 12 Nov 2021 21:14:50 +0000 Subject: [PATCH 072/250] object-file.c: use size_t for header lengths Continue walking the code path for the >4GB `hash-object --literally` test. The `hash_object_file_literally()` function internally uses both `hash_object_file()` and `write_object_file_prepare()`. Both function signatures use `unsigned long` rather than `size_t` for the mem buffer sizes. Use `size_t` instead, for LLP64 compatibility. While at it, convert those function's object's header buffer length to `size_t` for consistency. The value is already upcast to `uintmax_t` for print format compatibility. Note: The hash-object test still does not pass. A subsequent commit continues to walk the call tree's lower level hash functions to identify further fixes. Signed-off-by: Philip Oakley Signed-off-by: Johannes Schindelin --- object-file.c | 14 +++++++------- object-file.h | 4 ++-- 2 files changed, 9 insertions(+), 9 deletions(-) diff --git a/object-file.c b/object-file.c index 1ac04c2891634a..ae5e01efd5b63d 100644 --- a/object-file.c +++ b/object-file.c @@ -530,7 +530,7 @@ int loose_object_info(struct repository *r, static void hash_object_body(const struct git_hash_algo *algo, struct git_hash_ctx *c, const void *buf, unsigned long len, struct object_id *oid, - char *hdr, int *hdrlen) + char *hdr, size_t *hdrlen) { algo->init_fn(c); git_hash_update(c, hdr, *hdrlen); @@ -539,9 +539,9 @@ static void hash_object_body(const struct git_hash_algo *algo, struct git_hash_c } static void write_object_file_prepare(const struct git_hash_algo *algo, - const void *buf, unsigned long len, + const void *buf, size_t len, enum object_type type, struct object_id *oid, - char *hdr, int *hdrlen) + char *hdr, size_t *hdrlen) { struct git_hash_ctx c; @@ -682,11 +682,11 @@ int finalize_object_file_flags(const char *tmpfile, const char *filename, } void hash_object_file(const struct git_hash_algo *algo, const void *buf, - unsigned long len, enum object_type type, + size_t len, enum object_type type, struct object_id *oid) { char hdr[MAX_HEADER_LEN]; - int hdrlen = sizeof(hdr); + size_t hdrlen = sizeof(hdr); write_object_file_prepare(algo, buf, len, type, oid, hdr, &hdrlen); } @@ -1050,7 +1050,7 @@ int stream_loose_object(struct input_stream *in_stream, size_t len, return err; } -int write_object_file_flags(const void *buf, unsigned long len, +int write_object_file_flags(const void *buf, size_t len, enum object_type type, struct object_id *oid, struct object_id *compat_oid_in, unsigned flags) { @@ -1059,7 +1059,7 @@ int write_object_file_flags(const void *buf, unsigned long len, const struct git_hash_algo *compat = repo->compat_hash_algo; struct object_id compat_oid; char hdr[MAX_HEADER_LEN]; - int hdrlen = sizeof(hdr); + size_t hdrlen = sizeof(hdr); /* Generate compat_oid */ if (compat) { diff --git a/object-file.h b/object-file.h index 6f411424523932..185a3a3381ee38 100644 --- a/object-file.h +++ b/object-file.h @@ -172,7 +172,7 @@ enum { WRITE_OBJECT_FILE_SILENT = (1 << 1), }; -int write_object_file_flags(const void *buf, unsigned long len, +int write_object_file_flags(const void *buf, size_t len, enum object_type type, struct object_id *oid, struct object_id *compat_oid_in, unsigned flags); static inline int write_object_file(const void *buf, unsigned long len, @@ -223,7 +223,7 @@ int finalize_object_file_flags(const char *tmpfile, const char *filename, enum finalize_object_file_flags flags); void hash_object_file(const struct git_hash_algo *algo, const void *buf, - unsigned long len, enum object_type type, + size_t len, enum object_type type, struct object_id *oid); /* Helper to check and "touch" a file */ From 9aa87372b5454a878a8efefaf2bfcb65b7e4d728 Mon Sep 17 00:00:00 2001 From: Victoria Dye Date: Thu, 5 Aug 2021 19:04:13 -0400 Subject: [PATCH 073/250] subtree: update `contrib/subtree` `test` target The intention of this change is to align with how the top-level git `Makefile` defines its own test target (which also internally calls `$(MAKE) -C t/ all`). This change also ensures the consistency of `make -C contrib/subtree test` with other testing in CI executions (which rely on `$DEFAULT_TEST_TARGET` being defined as `prove`). Signed-off-by: Victoria Dye --- contrib/subtree/Makefile | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/contrib/subtree/Makefile b/contrib/subtree/Makefile index c0c9f21cb78022..dab2dfc08ee222 100644 --- a/contrib/subtree/Makefile +++ b/contrib/subtree/Makefile @@ -95,7 +95,7 @@ $(GIT_SUBTREE_TEST): $(GIT_SUBTREE) cp $< $@ test: $(GIT_SUBTREE_TEST) - $(MAKE) -C t/ test + $(MAKE) -C t/ all clean: $(RM) $(GIT_SUBTREE) From d49d8a06e4a5079393bfcaee80999feae906da3a Mon Sep 17 00:00:00 2001 From: Philip Oakley Date: Thu, 22 Apr 2021 11:11:38 +0100 Subject: [PATCH 074/250] CMakeLists: add default "x64-windows" arch for Visual Studio In Git-for-Windows, work on using ARM64 has progressed. The commit 2d94b77b27 (cmake: allow building for Windows/ARM64, 2020-12-04) failed to notice that /compat/vcbuild/vcpkg_install.bat will default to using the "x64-windows" architecture for the vcpkg installation if not set, but CMake is not told of this default. Commit 635b6d99b3 (vcbuild: install ARM64 dependencies when building ARM64 binaries, 2020-01-31) later updated vcpkg_install.bat to accept an arch (%1) parameter, but retained the default. This default is neccessary for the use case where the project directory is opened directly in Visual Studio, which will find and build a CMakeLists.txt file without any parameters, thus expecting use of the default setting. Also Visual studio will generate internal .sln solution and .vcxproj project files needed for some extension tools. Inform users of the additional .sln/.vcxproj generation. ** How to test: rm -rf '.vs' # remove old visual studio settings rm -rf 'compat/vcbuild/vcpkg' # remove any vcpkg downloads rm -rf 'contrib/buildsystems/out' # remove builds & CMake artifacts with a fresh Visual Studio Community Edition, File>>Open>>(git *folder*) to load the project (which will take some time!). check for successful compilation. The implicit .sln (etc.) are in the hidden .vs directory created by Visual Studio. Signed-off-by: Philip Oakley --- contrib/buildsystems/CMakeLists.txt | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/contrib/buildsystems/CMakeLists.txt b/contrib/buildsystems/CMakeLists.txt index 30fa3250c75bf1..d21e22835643d0 100644 --- a/contrib/buildsystems/CMakeLists.txt +++ b/contrib/buildsystems/CMakeLists.txt @@ -71,6 +71,10 @@ if(USE_VCPKG) message("Initializing vcpkg and building the Git's dependencies (this will take a while...)") execute_process(COMMAND ${CMAKE_SOURCE_DIR}/compat/vcbuild/vcpkg_install.bat ${VCPKG_ARCH}) endif() + if(NOT EXISTS ${VCPKG_ARCH}) + message("VCPKG_ARCH: unset, using 'x64-windows'") + set(VCPKG_ARCH "x64-windows") # default from vcpkg_install.bat + endif() list(APPEND CMAKE_PREFIX_PATH "${VCPKG_DIR}/installed/${VCPKG_ARCH}") # In the vcpkg edition, we need this to be able to link to libcurl From 8475931875429268ff9f5a1e8a6bf4d8fbc2831e Mon Sep 17 00:00:00 2001 From: Philip Oakley Date: Fri, 12 Nov 2021 21:16:51 +0000 Subject: [PATCH 075/250] hash algorithms: use size_t for section lengths Continue walking the code path for the >4GB `hash-object --literally` test to the hash algorithm step for LLP64 systems. This patch lets the SHA1DC code use `size_t`, making it compatible with LLP64 data models (as used e.g. by Windows). The interested reader of this patch will note that we adjust the signature of the `git_SHA1DCUpdate()` function without updating _any_ call site. This certainly puzzled at least one reviewer already, so here is an explanation: This function is never called directly, but always via the macro `platform_SHA1_Update`, which is usually called via the macro `git_SHA1_Update`. However, we never call `git_SHA1_Update()` directly in `struct git_hash_algo`. Instead, we call `git_hash_sha1_update()`, which is defined thusly: static void git_hash_sha1_update(git_hash_ctx *ctx, const void *data, size_t len) { git_SHA1_Update(&ctx->sha1, data, len); } i.e. it contains an implicit downcast from `size_t` to `unsigned long` (before this here patch). With this patch, there is no downcast anymore. With this patch, finally, the t1007-hash-object.sh "files over 4GB hash literally" test case is fixed. Signed-off-by: Philip Oakley Signed-off-by: Johannes Schindelin --- object-file.c | 4 ++-- sha1dc_git.c | 3 +-- sha1dc_git.h | 2 +- t/t1007-hash-object.sh | 2 +- 4 files changed, 5 insertions(+), 6 deletions(-) diff --git a/object-file.c b/object-file.c index ae5e01efd5b63d..1644a0d98533da 100644 --- a/object-file.c +++ b/object-file.c @@ -528,7 +528,7 @@ int loose_object_info(struct repository *r, } static void hash_object_body(const struct git_hash_algo *algo, struct git_hash_ctx *c, - const void *buf, unsigned long len, + const void *buf, size_t len, struct object_id *oid, char *hdr, size_t *hdrlen) { @@ -548,7 +548,7 @@ static void write_object_file_prepare(const struct git_hash_algo *algo, /* Generate the header */ *hdrlen = format_object_header(hdr, *hdrlen, type, len); - /* Sha1.. */ + /* Hash (function pointers) computation */ hash_object_body(algo, &c, buf, len, oid, hdr, hdrlen); } diff --git a/sha1dc_git.c b/sha1dc_git.c index 9b675a046ee699..fe58d7962a30c9 100644 --- a/sha1dc_git.c +++ b/sha1dc_git.c @@ -27,10 +27,9 @@ void git_SHA1DCFinal(unsigned char hash[20], SHA1_CTX *ctx) /* * Same as SHA1DCUpdate, but adjust types to match git's usual interface. */ -void git_SHA1DCUpdate(SHA1_CTX *ctx, const void *vdata, unsigned long len) +void git_SHA1DCUpdate(SHA1_CTX *ctx, const void *vdata, size_t len) { const char *data = vdata; - /* We expect an unsigned long, but sha1dc only takes an int */ while (len > INT_MAX) { SHA1DCUpdate(ctx, data, INT_MAX); data += INT_MAX; diff --git a/sha1dc_git.h b/sha1dc_git.h index f6f880cabea382..0bcf1aa84b7241 100644 --- a/sha1dc_git.h +++ b/sha1dc_git.h @@ -15,7 +15,7 @@ void git_SHA1DCInit(SHA1_CTX *); #endif void git_SHA1DCFinal(unsigned char [20], SHA1_CTX *); -void git_SHA1DCUpdate(SHA1_CTX *ctx, const void *data, unsigned long len); +void git_SHA1DCUpdate(SHA1_CTX *ctx, const void *data, size_t len); #define platform_SHA_IS_SHA1DC /* used by "test-tool sha1-is-sha1dc" */ diff --git a/t/t1007-hash-object.sh b/t/t1007-hash-object.sh index 09741ea3d6c4fe..290a69aadfc199 100755 --- a/t/t1007-hash-object.sh +++ b/t/t1007-hash-object.sh @@ -261,7 +261,7 @@ test_expect_success '--stdin outside of repository (uses SHA-1)' ' test_cmp expect actual ' -test_expect_failure EXPENSIVE,SIZE_T_IS_64BIT,!LONG_IS_64BIT \ +test_expect_success EXPENSIVE,SIZE_T_IS_64BIT,!LONG_IS_64BIT \ 'files over 4GB hash literally' ' test-tool genzeros $((5*1024*1024*1024)) >big && test_oid large5GB >expect && From c8333df47f2d1eb8a2645a31bba563ed13f15cc3 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Fri, 2 Apr 2021 22:50:54 +0200 Subject: [PATCH 076/250] mingw: allow for longer paths in `parse_interpreter()` MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit As reported in https://github.com/newren/git-filter-repo/pull/225, it looks like 99 bytes is not really sufficient to represent e.g. the full path to Python when installed via Windows Store (and this path is used in the hasb bang line when installing scripts via `pip`). Let's increase it to what is probably the maximum sensible path size: MAX_PATH. This makes `parse_interpreter()` in line with what `lookup_prog()` handles. Signed-off-by: Johannes Schindelin Signed-off-by: Vilius Šumskas --- compat/mingw.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/compat/mingw.c b/compat/mingw.c index 8a9972a1ca19f7..be485d8dbcb8d3 100644 --- a/compat/mingw.c +++ b/compat/mingw.c @@ -1362,7 +1362,7 @@ static const char *quote_arg_msys2(const char *arg) static const char *parse_interpreter(const char *cmd) { - static char buf[100]; + static char buf[MAX_PATH]; char *p, *opt; ssize_t n; /* read() can return negative values */ int fd; From 0dba464c32d1e23997beb8cceef7dbb30ed19646 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Mon, 17 May 2021 10:46:52 +0200 Subject: [PATCH 077/250] compat/vcbuild: document preferred way to build in Visual Studio We used to have that `make vcxproj` hack, but a hack it is. In the meantime, we have a much cleaner solution: using CMake, either explicitly, or even more conveniently via Visual Studio's built-in CMake support (simply open Git's top-level directory via File>Open>Folder...). Let's let the `README` reflect this. Signed-off-by: Johannes Schindelin --- compat/vcbuild/README | 28 +++++++++------------------- 1 file changed, 9 insertions(+), 19 deletions(-) diff --git a/compat/vcbuild/README b/compat/vcbuild/README index 29ec1d0f104b80..5c71ea2daa4017 100644 --- a/compat/vcbuild/README +++ b/compat/vcbuild/README @@ -37,27 +37,17 @@ The Steps to Build Git with VS2015 or VS2017 from the command line. ================================================================ -Alternatively, run `make vcxproj` and then load the generated `git.sln` in -Visual Studio. The initial build will install the vcpkg system and build the +Alternatively, just open Git's top-level directory in Visual Studio, via +`File>Open>Folder...`. This will use CMake internally to generate the +project definitions. It will also install the vcpkg system and build the dependencies automatically. This will take a while. -Instead of generating the `git.sln` file yourself (which requires a full Git -for Windows SDK), you may want to consider fetching the `vs/master` branch of -https://github.com/git-for-windows/git instead (which is updated automatically -via CI running `make vcxproj`). The `vs/master` branch does not require a Git -for Windows to build, but you can run the test scripts in a regular Git Bash. - -Note that `make vcxproj` will automatically add and commit the generated `.sln` -and `.vcxproj` files to the repo. This is necessary to allow building a -fully-testable Git in Visual Studio, where a regular Git Bash can be used to -run the test scripts (as opposed to a full Git for Windows SDK): a number of -build targets, such as Git commands implemented as Unix shell scripts (where -`@@SHELL_PATH@@` and other placeholders are interpolated) require a full-blown -Git for Windows SDK (which is about 10x the size of a regular Git for Windows -installation). - -If your plan is to open a Pull Request with Git for Windows, it is a good idea -to drop this commit before submitting. +You can also generate the Visual Studio solution manually by downloading +and running CMake explicitly rather than letting Visual Studio doing +that implicitly. + +Another, deprecated option is to run `make vcxproj`. This option is +superseded by the CMake-based build, and will be removed at some point. ================================================================ The Steps of Build Git with VS2008 From 9b58f03884db6f07f84653f4e8abbffc7056c184 Mon Sep 17 00:00:00 2001 From: Pascal Muller Date: Wed, 23 Jun 2021 21:21:10 +0200 Subject: [PATCH 078/250] http: optionally send SSL client certificate This adds support for a new http.sslAutoClientCert config value. In cURL 7.77 or later the schannel backend does not automatically send client certificates from the Windows Certificate Store anymore. This config value is only used if http.sslBackend is set to "schannel", and can be used to opt in to the old behavior and force cURL to send client certificates. This fixes https://github.com/git-for-windows/git/issues/3292 Signed-off-by: Pascal Muller --- Documentation/config/http.adoc | 5 +++++ git-curl-compat.h | 8 ++++++++ http.c | 24 +++++++++++++++++++++--- 3 files changed, 34 insertions(+), 3 deletions(-) diff --git a/Documentation/config/http.adoc b/Documentation/config/http.adoc index 9122c5dc23ea1a..7fd001206ded22 100644 --- a/Documentation/config/http.adoc +++ b/Documentation/config/http.adoc @@ -249,6 +249,11 @@ http.schannelUseSSLCAInfo:: when the `schannel` backend was configured via `http.sslBackend`, unless `http.schannelUseSSLCAInfo` overrides this behavior. +http.sslAutoClientCert:: + As of cURL v7.77.0, the Secure Channel backend won't automatically + send client certificates from the Windows Certificate Store anymore. + To opt in to the old behavior, http.sslAutoClientCert can be set. + http.pinnedPubkey:: Public key of the https service. It may either be the filename of a PEM or DER encoded public key file or a string starting with diff --git a/git-curl-compat.h b/git-curl-compat.h index aa8eed7ed2b5e3..04496113fb7b83 100644 --- a/git-curl-compat.h +++ b/git-curl-compat.h @@ -37,6 +37,14 @@ #define GIT_CURL_NEED_TRANSFER_ENCODING_HEADER #endif +/** + * CURLSSLOPT_AUTO_CLIENT_CERT was added in 7.77.0, released in May + * 2021. + */ +#if LIBCURL_VERSION_NUM >= 0x074d00 +#define GIT_CURL_HAVE_CURLSSLOPT_AUTO_CLIENT_CERT +#endif + /** * CURLOPT_PROTOCOLS_STR and CURLOPT_REDIR_PROTOCOLS_STR were added in 7.85.0, * released in August 2022. diff --git a/http.c b/http.c index 808069ec29a220..6d06c6c463f9f8 100644 --- a/http.c +++ b/http.c @@ -161,6 +161,8 @@ static long http_schannel_check_revoke_mode = */ static int http_schannel_use_ssl_cainfo; +static int http_auto_client_cert; + static int always_auth_proactively(void) { return http_proactive_auth != PROACTIVE_AUTH_NONE && @@ -449,6 +451,11 @@ static int http_options(const char *var, const char *value, return 0; } + if (!strcmp("http.sslautoclientcert", var)) { + http_auto_client_cert = git_config_bool(var, value); + return 0; + } + if (!strcmp("http.minsessions", var)) { min_curl_sessions = git_config_int(var, value, ctx->kvi); if (min_curl_sessions > 1) @@ -1073,9 +1080,20 @@ static CURL *get_curl_handle(void) } #endif - if (http_ssl_backend && !strcmp("schannel", http_ssl_backend) && - http_schannel_check_revoke_mode) { - curl_easy_setopt(result, CURLOPT_SSL_OPTIONS, http_schannel_check_revoke_mode); + if (http_ssl_backend && !strcmp("schannel", http_ssl_backend)) { + long ssl_options = 0; + if (http_schannel_check_revoke_mode) { + ssl_options |= http_schannel_check_revoke_mode; + } + + if (http_auto_client_cert) { +#ifdef GIT_CURL_HAVE_CURLSSLOPT_AUTO_CLIENT_CERT + ssl_options |= CURLSSLOPT_AUTO_CLIENT_CERT; +#endif + } + + if (ssl_options) + curl_easy_setopt(result, CURLOPT_SSL_OPTIONS, ssl_options); } if (http_proactive_auth != PROACTIVE_AUTH_NONE) From d2b4c193cc9c47fae391cf816676edf858caf62f Mon Sep 17 00:00:00 2001 From: Victoria Dye Date: Thu, 5 Aug 2021 19:11:59 -0400 Subject: [PATCH 079/250] ci: run `contrib/subtree` tests in CI builds Because `git subtree` (unlike most other `contrib` modules) is included as part of the standard release of Git for Windows, its stability should be verified as consistently as it is for the rest of git. By including the `git subtree` tests in the CI workflow, these tests are as much of a gate to merging and indicator of stability as the standard test suite. Signed-off-by: Victoria Dye --- ci/run-build-and-tests.sh | 4 ++++ ci/run-test-slice.sh | 3 +++ 2 files changed, 7 insertions(+) diff --git a/ci/run-build-and-tests.sh b/ci/run-build-and-tests.sh index 01823fd0f140bb..5e94ff33830a16 100755 --- a/ci/run-build-and-tests.sh +++ b/ci/run-build-and-tests.sh @@ -72,5 +72,9 @@ case "$jobname" in ;; esac +case " $MAKE_TARGETS " in +*" all "*) make -C contrib/subtree test;; +esac + check_unignored_build_artifacts save_good_tree diff --git a/ci/run-test-slice.sh b/ci/run-test-slice.sh index 0444c79c023c82..6e21260e17543b 100755 --- a/ci/run-test-slice.sh +++ b/ci/run-test-slice.sh @@ -15,4 +15,7 @@ if [ "$1" == "0" ] ; then group "Run unit tests" make --quiet -C t unit-tests-test-tool fi +# Run the git subtree tests only if main tests succeeded +test 0 != "$1" || make -C contrib/subtree test + check_unignored_build_artifacts From bee24969d319fb504bbb2d7b404090042a75c77c Mon Sep 17 00:00:00 2001 From: Philip Oakley Date: Mon, 10 May 2021 16:47:40 +0100 Subject: [PATCH 080/250] CMake: show Win32 and Generator_platform build-option values Ensure key CMake option values are part of the CMake output to facilitate user support when tool updates impact the wider CMake actions, particularly ongoing 'improvements' in Visual Studio. These CMake displays perform the same function as the build-options.txt provided in the main Git for Windows. CMake is already chatty. The setting of CMAKE_EXPORT_COMPILE_COMMANDS is also reported. Include the environment's CMAKE_EXPORT_COMPILE_COMMANDS value which may have been propogated to CMake's internal value. Testing the CMAKE_EXPORT_COMPILE_COMMANDS processing can be difficult in the Visual Studio environment, as it may be cached in many places. The 'environment' may include the OS, the user shell, CMake's own environment, along with the Visual Studio presets and caches. See previous commit for arefacts that need removing for a clean test. Signed-off-by: Philip Oakley --- contrib/buildsystems/CMakeLists.txt | 10 ++++++++++ 1 file changed, 10 insertions(+) diff --git a/contrib/buildsystems/CMakeLists.txt b/contrib/buildsystems/CMakeLists.txt index d21e22835643d0..fa3885b9bc99b0 100644 --- a/contrib/buildsystems/CMakeLists.txt +++ b/contrib/buildsystems/CMakeLists.txt @@ -63,10 +63,20 @@ endif() if(NOT DEFINED CMAKE_EXPORT_COMPILE_COMMANDS) set(CMAKE_EXPORT_COMPILE_COMMANDS TRUE) + message("settting CMAKE_EXPORT_COMPILE_COMMANDS: ${CMAKE_EXPORT_COMPILE_COMMANDS}") endif() if(USE_VCPKG) set(VCPKG_DIR "${CMAKE_SOURCE_DIR}/compat/vcbuild/vcpkg") + message("WIN32: ${WIN32}") # show its underlying text values + message("VCPKG_DIR: ${VCPKG_DIR}") + message("VCPKG_ARCH: ${VCPKG_ARCH}") # maybe unset + message("MSVC: ${MSVC}") + message("CMAKE_GENERATOR: ${CMAKE_GENERATOR}") + message("CMAKE_CXX_COMPILER_ID: ${CMAKE_CXX_COMPILER_ID}") + message("CMAKE_GENERATOR_PLATFORM: ${CMAKE_GENERATOR_PLATFORM}") + message("CMAKE_EXPORT_COMPILE_COMMANDS: ${CMAKE_EXPORT_COMPILE_COMMANDS}") + message("ENV(CMAKE_EXPORT_COMPILE_COMMANDS): $ENV{CMAKE_EXPORT_COMPILE_COMMANDS}") if(NOT EXISTS ${VCPKG_DIR}) message("Initializing vcpkg and building the Git's dependencies (this will take a while...)") execute_process(COMMAND ${CMAKE_SOURCE_DIR}/compat/vcbuild/vcpkg_install.bat ${VCPKG_ARCH}) From 514b16226f38cbf4040689390343b12faf349e47 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Wed, 8 Sep 2021 13:05:42 +0200 Subject: [PATCH 081/250] init: do parse _all_ core.* settings early In Git for Windows, `has_symlinks` is set to 0 by default. Therefore, we need to parse the config setting `core.symlinks` to know if it has been set to `true`. In `git init`, we must do that before copying the templates because they might contain symbolic links. Even if the support for symbolic links on Windows has not made it to upstream Git yet, we really should make sure that all the `core.*` settings are parsed before proceeding, as they might very well change the behavior of `git init` in a way the user intended. This fixes https://github.com/git-for-windows/git/issues/3414 Signed-off-by: Johannes Schindelin --- config.c | 4 ++-- config.h | 2 ++ setup.c | 2 +- 3 files changed, 5 insertions(+), 3 deletions(-) diff --git a/config.c b/config.c index b18b5617fcd05d..fee2c33c8c2d7e 100644 --- a/config.c +++ b/config.c @@ -1393,8 +1393,8 @@ int git_config_color(char *dest, const char *var, const char *value) return 0; } -static int git_default_core_config(const char *var, const char *value, - const struct config_context *ctx, void *cb) +int git_default_core_config(const char *var, const char *value, + const struct config_context *ctx, void *cb) { /* This needs a better name */ if (!strcmp(var, "core.filemode")) { diff --git a/config.h b/config.h index 29a027748375f1..36ab27a4ace709 100644 --- a/config.h +++ b/config.h @@ -167,6 +167,8 @@ typedef int (*config_fn_t)(const char *, const char *, int git_default_config(const char *, const char *, const struct config_context *, void *); +int git_default_core_config(const char *var, const char *value, + const struct config_context *ctx, void *cb); /** * Read a specific file in git-config format. diff --git a/setup.c b/setup.c index f93bd6a24a5d9c..554b3e3751390d 100644 --- a/setup.c +++ b/setup.c @@ -2590,7 +2590,7 @@ int init_db(const char *git_dir, const char *real_git_dir, * have set up the repository format such that we can evaluate * includeIf conditions correctly in the case of re-initialization. */ - git_config(platform_core_config, NULL); + git_config(git_default_core_config, NULL); safe_create_dir(the_repository, git_dir, 0); From 6b71feb31baa0205cb47a5c1a7d3b96c8613a070 Mon Sep 17 00:00:00 2001 From: Philip Oakley Date: Mon, 6 Dec 2021 22:26:50 +0000 Subject: [PATCH 082/250] hash-object --stdin: verify that it works with >4GB/LLP64 Just like the `hash-object --literally` code path, the `--stdin` code path also needs to use `size_t` instead of `unsigned long` to represent memory sizes, otherwise it would cause problems on platforms using the LLP64 data model (such as Windows). To limit the scope of the test case, the object is explicitly not written to the object store, nor are any filters applied. The `big` file from the previous test case is reused to save setup time; To avoid relying on that side effect, it is generated if it does not exist (e.g. when running via `sh t1007-*.sh --long --run=1,41`). Signed-off-by: Philip Oakley Signed-off-by: Johannes Schindelin --- t/t1007-hash-object.sh | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/t/t1007-hash-object.sh b/t/t1007-hash-object.sh index 290a69aadfc199..b2e8d3890c849a 100755 --- a/t/t1007-hash-object.sh +++ b/t/t1007-hash-object.sh @@ -269,4 +269,12 @@ test_expect_success EXPENSIVE,SIZE_T_IS_64BIT,!LONG_IS_64BIT \ test_cmp expect actual ' +test_expect_success EXPENSIVE,SIZE_T_IS_64BIT,!LONG_IS_64BIT \ + 'files over 4GB hash correctly via --stdin' ' + { test -f big || test-tool genzeros $((5*1024*1024*1024)) >big; } && + test_oid large5GB >expect && + git hash-object --stdin actual && + test_cmp expect actual +' + test_done From 937caaf6cff9c30a56acaa1a80e39a293b2a81ea Mon Sep 17 00:00:00 2001 From: Philip Oakley Date: Mon, 6 Dec 2021 22:42:46 +0000 Subject: [PATCH 083/250] hash-object: add another >4GB/LLP64 test case To complement the `--stdin` and `--literally` test cases that verify that we can hash files larger than 4GB on 64-bit platforms using the LLP64 data model, here is a test case that exercises `hash-object` _without_ any options. Just as before, we use the `big` file from the previous test case if it exists to save on setup time, otherwise generate it. Signed-off-by: Philip Oakley Signed-off-by: Johannes Schindelin --- t/t1007-hash-object.sh | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/t/t1007-hash-object.sh b/t/t1007-hash-object.sh index b2e8d3890c849a..b306cd01f9f488 100755 --- a/t/t1007-hash-object.sh +++ b/t/t1007-hash-object.sh @@ -277,4 +277,12 @@ test_expect_success EXPENSIVE,SIZE_T_IS_64BIT,!LONG_IS_64BIT \ test_cmp expect actual ' +test_expect_success EXPENSIVE,SIZE_T_IS_64BIT,!LONG_IS_64BIT \ + 'files over 4GB hash correctly' ' + { test -f big || test-tool genzeros $((5*1024*1024*1024)) >big; } && + test_oid large5GB >expect && + git hash-object -- big >actual && + test_cmp expect actual +' + test_done From 3a4578ce14a3d2548a0bbc35625df41631329b8a Mon Sep 17 00:00:00 2001 From: Derrick Stolee Date: Wed, 13 Apr 2022 14:49:17 -0400 Subject: [PATCH 084/250] setup: properly use "%(prefix)/" when in WSL Signed-off-by: Derrick Stolee --- setup.c | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-) diff --git a/setup.c b/setup.c index f93bd6a24a5d9c..6791111992c5bb 100644 --- a/setup.c +++ b/setup.c @@ -1781,10 +1781,19 @@ const char *setup_git_directory_gently(int *nongit_ok) break; case GIT_DIR_INVALID_OWNERSHIP: if (!nongit_ok) { + struct strbuf prequoted = STRBUF_INIT; struct strbuf quoted = STRBUF_INIT; strbuf_complete(&report, '\n'); - sq_quote_buf_pretty("ed, dir.buf); + +#ifdef __MINGW32__ + if (dir.buf[0] == '/') + strbuf_addstr(&prequoted, "%(prefix)/"); +#endif + + strbuf_add(&prequoted, dir.buf, dir.len); + sq_quote_buf_pretty("ed, prequoted.buf); + die(_("detected dubious ownership in repository at '%s'\n" "%s" "To add an exception for this directory, call:\n" From e77d7a10059cf08ed7b0f3f526df7925f502eaf2 Mon Sep 17 00:00:00 2001 From: Philip Oakley Date: Tue, 7 Dec 2021 09:53:41 +0000 Subject: [PATCH 085/250] hash-object: add a >4GB/LLP64 test case using filtered input To verify that the `clean` side of the `clean`/`smudge` filter code is correct with regards to LLP64 (read: to ensure that `size_t` is used instead of `unsigned long`), here is a test case using a trivial filter, specifically _not_ writing anything to the object store to limit the scope of the test case. As in previous commits, the `big` file from previous test cases is reused if available, to save setup time, otherwise re-generated. Signed-off-by: Philip Oakley Signed-off-by: Johannes Schindelin --- t/t1007-hash-object.sh | 12 ++++++++++++ 1 file changed, 12 insertions(+) diff --git a/t/t1007-hash-object.sh b/t/t1007-hash-object.sh index b306cd01f9f488..05cc5f7954b3f6 100755 --- a/t/t1007-hash-object.sh +++ b/t/t1007-hash-object.sh @@ -285,4 +285,16 @@ test_expect_success EXPENSIVE,SIZE_T_IS_64BIT,!LONG_IS_64BIT \ test_cmp expect actual ' +# This clean filter does nothing, other than excercising the interface. +# We ensure that cleaning doesn't mangle large files on 64-bit Windows. +test_expect_success EXPENSIVE,SIZE_T_IS_64BIT,!LONG_IS_64BIT \ + 'hash filtered files over 4GB correctly' ' + { test -f big || test-tool genzeros $((5*1024*1024*1024)) >big; } && + test_oid large5GB >expect && + test_config filter.null-filter.clean "cat" && + echo "big filter=null-filter" >.gitattributes && + git hash-object -- big >actual && + test_cmp expect actual +' + test_done From 3c103b5fe8d59b02f2b9ef11734326294a2c3087 Mon Sep 17 00:00:00 2001 From: Derrick Stolee Date: Wed, 13 Apr 2022 14:54:43 -0400 Subject: [PATCH 086/250] compat/mingw.c: do not warn when failing to get owner In the case of Git for Windows (say, in a Git Bash window) running in a Windows Subsystem for Linux (WSL) directory, the GetNamedSecurityInfoW() call in is_path_owned_By_current_side() returns an error code other than ERROR_SUCCESS. This is consistent behavior across this boundary. In these cases, the owner would always be different because the WSL owner is a different entity than the Windows user. The change here is to suppress the error message that looks like this: error: failed to get owner for '//wsl.localhost/...' (1) Before this change, this warning happens for every Git command, regardless of whether the directory is marked with safe.directory. Signed-off-by: Derrick Stolee --- compat/mingw.c | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-) diff --git a/compat/mingw.c b/compat/mingw.c index 8a9972a1ca19f7..200bb0a2c2fc12 100644 --- a/compat/mingw.c +++ b/compat/mingw.c @@ -2962,9 +2962,7 @@ int is_path_owned_by_current_sid(const char *path, struct strbuf *report) DACL_SECURITY_INFORMATION, &sid, NULL, NULL, NULL, &descriptor); - if (err != ERROR_SUCCESS) - error(_("failed to get owner for '%s' (%ld)"), path, err); - else if (sid && IsValidSid(sid)) { + if (err == ERROR_SUCCESS && sid && IsValidSid(sid)) { /* Now, verify that the SID matches the current user's */ static PSID current_user_sid; static HANDLE linked_token; From 0a76eca34d93ec863b8b6afa2574474e2aa8af05 Mon Sep 17 00:00:00 2001 From: Rafael Kitover Date: Tue, 12 Apr 2022 19:53:33 +0000 Subject: [PATCH 087/250] mingw: $env:TERM="xterm-256color" for newer OSes For Windows builds >= 15063 set $env:TERM to "xterm-256color" instead of "cygwin" because they have a more capable console system that supports this. Also set $env:COLORTERM="truecolor" if unset. $env:TERM is initialized so that ANSI colors in color.c work, see 29a3963484 (Win32: patch Windows environment on startup, 2012-01-15). See git-for-windows/git#3629 regarding problems caused by always setting $env:TERM="cygwin". This is the same heuristic used by the Cygwin runtime. Signed-off-by: Rafael Kitover Signed-off-by: Johannes Schindelin --- compat/mingw.c | 17 ++++++++++++++--- 1 file changed, 14 insertions(+), 3 deletions(-) diff --git a/compat/mingw.c b/compat/mingw.c index 8a9972a1ca19f7..e5c9d8b7b9842a 100644 --- a/compat/mingw.c +++ b/compat/mingw.c @@ -2820,9 +2820,20 @@ static void setup_windows_environment(void) convert_slashes(tmp); } - /* simulate TERM to enable auto-color (see color.c) */ - if (!getenv("TERM")) - setenv("TERM", "cygwin", 1); + + /* + * Make sure TERM is set up correctly to enable auto-color + * (see color.c .) Use "cygwin" for older OS releases which + * works correctly with MSYS2 utilities on older consoles. + */ + if (!getenv("TERM")) { + if ((GetVersion() >> 16) < 15063) + setenv("TERM", "cygwin", 0); + else { + setenv("TERM", "xterm-256color", 0); + setenv("COLORTERM", "truecolor", 0); + } + } /* calculate HOME if not set */ if (!getenv("HOME")) { From 70ee9bf1494c94e5a61a297491545905534eb633 Mon Sep 17 00:00:00 2001 From: Christopher Degawa Date: Sat, 28 May 2022 14:53:54 -0500 Subject: [PATCH 088/250] winansi: check result and Buffer before using Name NtQueryObject under Wine can return a success but fill out no name. In those situations, Wine will set Buffer to NULL, and set result to the sizeof(OBJECT_NAME_INFORMATION). Running a command such as echo "$(git.exe --version 2>/dev/null)" will crash due to a NULL pointer dereference when the code attempts to null terminate the buffer, although, weirdly, removing the subshell or redirecting stdout to a file will not trigger the crash. Code has been added to also check Buffer and Length to ensure the check is as robust as possible due to the current behavior being fragile at best, and could potentially change in the future This code is based on the behavior of NtQueryObject under wine and reactos. Signed-off-by: Christopher Degawa --- compat/winansi.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/compat/winansi.c b/compat/winansi.c index ac2ffb78691a7d..d28137a20b0bcc 100644 --- a/compat/winansi.c +++ b/compat/winansi.c @@ -575,6 +575,9 @@ static void detect_msys_tty(int fd) if (!NT_SUCCESS(NtQueryObject(h, ObjectNameInformation, buffer, sizeof(buffer) - 2, &result))) return; + if (result < sizeof(*nameinfo) || !nameinfo->Name.Buffer || + !nameinfo->Name.Length) + return; name = nameinfo->Name.Buffer; name[nameinfo->Name.Length / sizeof(*name)] = 0; From dea0430c8d675c12382f4c175908bcf200868be9 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E5=AD=99=E5=8D=93=E8=AF=86?= Date: Sun, 16 Jan 2022 03:38:33 +0800 Subject: [PATCH 089/250] Add config option `windows.appendAtomically` MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Atomic append on windows is only supported on local disk files, and it may cause errors in other situations, e.g. network file system. If that is the case, this config option should be used to turn atomic append off. Co-Authored-By: Johannes Schindelin Signed-off-by: 孙卓识 Signed-off-by: Johannes Schindelin --- Documentation/config.adoc | 2 ++ Documentation/config/windows.adoc | 4 ++++ compat/mingw.c | 36 ++++++++++++++++++++++++++++--- 3 files changed, 39 insertions(+), 3 deletions(-) create mode 100644 Documentation/config/windows.adoc diff --git a/Documentation/config.adoc b/Documentation/config.adoc index a8b04c4e5183d2..30e11203c87d1c 100644 --- a/Documentation/config.adoc +++ b/Documentation/config.adoc @@ -558,4 +558,6 @@ include::config/versionsort.adoc[] include::config/web.adoc[] +include::config/windows.adoc[] + include::config/worktree.adoc[] diff --git a/Documentation/config/windows.adoc b/Documentation/config/windows.adoc new file mode 100644 index 00000000000000..fdaaf1c65504f3 --- /dev/null +++ b/Documentation/config/windows.adoc @@ -0,0 +1,4 @@ +windows.appendAtomically:: + By default, append atomic API is used on windows. But it works only with + local disk files, if you're working on a network file system, you should + set it false to turn it off. diff --git a/compat/mingw.c b/compat/mingw.c index cf65ab4482e5fa..c3c4f89447c00d 100644 --- a/compat/mingw.c +++ b/compat/mingw.c @@ -8,6 +8,7 @@ #include "dir.h" #include "environment.h" #include "gettext.h" +#include "repository.h" #include "run-command.h" #include "strbuf.h" #include "symlinks.h" @@ -620,6 +621,7 @@ static int is_local_named_pipe_path(const char *filename) int mingw_open (const char *filename, int oflags, ...) { + static int append_atomically = -1; typedef int (*open_fn_t)(wchar_t const *wfilename, int oflags, ...); va_list args; unsigned mode; @@ -638,7 +640,16 @@ int mingw_open (const char *filename, int oflags, ...) return -1; } - if ((oflags & O_APPEND) && !is_local_named_pipe_path(filename)) + /* + * Only set append_atomically to default value(1) when repo is initialized + * and fail to get config value + */ + if (append_atomically < 0 && the_repository && the_repository->commondir && + git_config_get_bool("windows.appendatomically", &append_atomically)) + append_atomically = 1; + + if (append_atomically && (oflags & O_APPEND) && + !is_local_named_pipe_path(filename)) open_fn = mingw_open_append; else if (!(oflags & ~(O_ACCMODE | O_NOINHERIT))) open_fn = mingw_open_existing; @@ -804,9 +815,28 @@ ssize_t mingw_write(int fd, const void *buf, size_t len) /* check if fd is a pipe */ HANDLE h = (HANDLE) _get_osfhandle(fd); - if (GetFileType(h) != FILE_TYPE_PIPE) + if (GetFileType(h) != FILE_TYPE_PIPE) { + if (orig == EINVAL) { + wchar_t path[MAX_PATH]; + DWORD ret = GetFinalPathNameByHandleW(h, path, + ARRAY_SIZE(path), 0); + UINT drive_type = ret > 0 && ret < ARRAY_SIZE(path) ? + GetDriveTypeW(path) : DRIVE_UNKNOWN; + + /* + * The default atomic append causes such an error on + * network file systems, in such a case, it should be + * turned off via config. + * + * `drive_type` of UNC path: DRIVE_NO_ROOT_DIR + */ + if (DRIVE_NO_ROOT_DIR == drive_type || DRIVE_REMOTE == drive_type) + warning("invalid write operation detected; you may try:\n" + "\n\tgit config windows.appendAtomically false"); + } + errno = orig; - else if (orig == EINVAL) + } else if (orig == EINVAL) errno = EPIPE; else { DWORD buf_size; From 0285ddec7d04a2564530e7c247016a5d52b6f99a Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Mon, 4 Sep 2017 11:59:45 +0200 Subject: [PATCH 090/250] mingw: change core.fsyncObjectFiles = 1 by default MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit From the documentation of said setting: This boolean will enable fsync() when writing object files. This is a total waste of time and effort on a filesystem that orders data writes properly, but can be useful for filesystems that do not use journalling (traditional UNIX filesystems) or that only journal metadata and not file contents (OS X’s HFS+, or Linux ext3 with "data=writeback"). The most common file system on Windows (NTFS) does not guarantee that order, therefore a sudden loss of power (or any other event causing an unclean shutdown) would cause corrupt files (i.e. files filled with NULs). Therefore we need to change the default. Note that the documentation makes it sound as if this causes really bad performance. In reality, writing loose objects is something that is done only rarely, and only a handful of files at a time. Signed-off-by: Johannes Schindelin --- compat/mingw.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/compat/mingw.c b/compat/mingw.c index c3c4f89447c00d..4363789bd0f80b 100644 --- a/compat/mingw.c +++ b/compat/mingw.c @@ -16,6 +16,7 @@ #include "win32.h" #include "win32/lazyload.h" #include "wrapper.h" +#include "write-or-die.h" #include #include #include @@ -3325,6 +3326,7 @@ int wmain(int argc, const wchar_t **wargv) #endif maybe_redirect_std_handles(); + fsync_object_files = 1; /* determine size of argv and environ conversion buffer */ maxlen = wcslen(wargv[0]); From 531aa1758bc7c6c79b8eac0575205c08e9dbd406 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Matthias=20A=C3=9Fhauer?= Date: Sun, 10 Jul 2022 11:27:25 +0200 Subject: [PATCH 091/250] MinGW: link as terminal server aware MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Whith Windows 2000, Microsoft introduced a flag to the PE header to mark executables as "terminal server aware". Windows terminal servers provide a redirected Windows directory and redirected registry hives when launching legacy applications without this flag set. Since we do not use any INI files in the Windows directory and don't write to the registry, we don't need this additional preparation. Telling the OS that we don't need this should provide slightly improved startup times in terminal server environments. When building for supported Windows Versions with MSVC the /TSAWARE linker flag is automatically set, but MinGW requires us to set the --tsaware flag manually. This partially addresses https://github.com/git-for-windows/git/issues/3935. Signed-off-by: Matthias Aßhauer --- config.mak.uname | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/config.mak.uname b/config.mak.uname index 3e26bb074a4b51..c524a2e434b3e8 100644 --- a/config.mak.uname +++ b/config.mak.uname @@ -699,7 +699,7 @@ ifeq ($(uname_S),MINGW) DEFAULT_HELP_FORMAT = html HAVE_PLATFORM_PROCINFO = YesPlease CSPRNG_METHOD = rtlgenrandom - BASIC_LDFLAGS += -municode + BASIC_LDFLAGS += -municode -Wl,--tsaware COMPAT_CFLAGS += -DNOGDI -Icompat -Icompat/win32 COMPAT_CFLAGS += -DSTRIP_EXTENSION=\".exe\" COMPAT_OBJS += compat/mingw.o compat/winansi.o \ From 9e7ccf8c9b3c7b6b135509cbf0a2adb781ecad54 Mon Sep 17 00:00:00 2001 From: Kiel Hurley Date: Wed, 2 Nov 2022 22:56:16 +1300 Subject: [PATCH 092/250] Fix Windows version resources Add FileVersion, which is a required field As not all required fields were present, none were being included Fixes #4090 Signed-off-by: Kiel Hurley --- git.rc.in | 1 + 1 file changed, 1 insertion(+) diff --git a/git.rc.in b/git.rc.in index e69444eef3f0c5..460ea39561b87f 100644 --- a/git.rc.in +++ b/git.rc.in @@ -12,6 +12,7 @@ BEGIN VALUE "OriginalFilename", "git.exe\0" VALUE "ProductName", "Git\0" VALUE "ProductVersion", "@GIT_VERSION@\0" + VALUE "FileVersion", "@GIT_VERSION@\0" END END From 1dddbdb73f7303599da8dfde592c8c9c85c5009b Mon Sep 17 00:00:00 2001 From: Andrey Zabavnikov Date: Fri, 28 Oct 2022 17:12:06 +0300 Subject: [PATCH 093/250] status: fix for old-style submodules with commondir In f9b7573f6b00 (repository: free fields before overwriting them, 2017-09-05), Git was taught to release memory before overwriting it, but 357a03ebe9e0 (repository.c: move env-related setup code back to environment.c, 2018-03-03) changed the code so that it would not _always_ be overwritten. As a consequence, the `commondir` attribute would point to already-free()d memory. This seems not to cause problems in core Git, but there are add-on patches in Git for Windows where the `commondir` attribute is subsequently used and causing invalid memory accesses e.g. in setups containing old-style submodules (i.e. the ones with a `.git` directory within theirs worktrees) that have `commondir` configured. This fixes https://github.com/git-for-windows/git/pull/4083. Signed-off-by: Andrey Zabavnikov Signed-off-by: Johannes Schindelin --- repository.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/repository.c b/repository.c index 9b3d6665fc6d4f..59a95cdfef5cfb 100644 --- a/repository.c +++ b/repository.c @@ -136,7 +136,7 @@ static void repo_set_commondir(struct repository *repo, { struct strbuf sb = STRBUF_INIT; - free(repo->commondir); + FREE_AND_NULL(repo->commondir); if (commondir) { repo->different_commondir = 1; From 5338ae8644c5cbf246957529fb4fbc56b750841a Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Sat, 6 May 2023 22:26:15 +0200 Subject: [PATCH 094/250] http: optionally load libcurl lazily This compile-time option allows to ask Git to load libcurl dynamically at runtime. Together with a follow-up patch that optionally overrides the file name depending on the `http.sslBackend` setting, this kicks open the door for installing multiple libcurl flavors side by side, and load the one corresponding to the (runtime-)configured SSL/TLS backend. Signed-off-by: Johannes Schindelin --- Makefile | 28 +++- compat/lazyload-curl.c | 354 +++++++++++++++++++++++++++++++++++++++++ 2 files changed, 375 insertions(+), 7 deletions(-) create mode 100644 compat/lazyload-curl.c diff --git a/Makefile b/Makefile index 70d1543b6b8688..8e37b102c94aca 100644 --- a/Makefile +++ b/Makefile @@ -472,6 +472,11 @@ include shared.mak # # CURL_LDFLAGS=-lcurl # +# Define LAZYLOAD_LIBCURL to dynamically load the libcurl; This can be useful +# if Multiple libcurl versions exist (with different file names) that link to +# various SSL/TLS backends, to support the `http.sslBackend` runtime switch in +# such a scenario. +# # === Optional library: libpcre2 === # # Define USE_LIBPCRE if you have and want to use libpcre. Various @@ -1665,10 +1670,19 @@ else CURL_LIBCURL = endif - ifndef CURL_LDFLAGS - CURL_LDFLAGS = $(eval CURL_LDFLAGS := $$(shell $$(CURL_CONFIG) --libs))$(CURL_LDFLAGS) + ifdef LAZYLOAD_LIBCURL + LAZYLOAD_LIBCURL_OBJ = compat/lazyload-curl.o + OBJECTS += $(LAZYLOAD_LIBCURL_OBJ) + # The `CURL_STATICLIB` constant must be defined to avoid seeing the functions + # declared as DLL imports + CURL_CFLAGS = -DCURL_STATICLIB + CURL_LIBCURL = -ldl + else + ifndef CURL_LDFLAGS + CURL_LDFLAGS = $(eval CURL_LDFLAGS := $$(shell $$(CURL_CONFIG) --libs))$(CURL_LDFLAGS) + endif + CURL_LIBCURL += $(CURL_LDFLAGS) endif - CURL_LIBCURL += $(CURL_LDFLAGS) ifndef CURL_CFLAGS CURL_CFLAGS = $(eval CURL_CFLAGS := $$(shell $$(CURL_CONFIG) --cflags))$(CURL_CFLAGS) @@ -1689,7 +1703,7 @@ else endif ifdef USE_CURL_FOR_IMAP_SEND BASIC_CFLAGS += -DUSE_CURL_FOR_IMAP_SEND - IMAP_SEND_BUILDDEPS = http.o + IMAP_SEND_BUILDDEPS = http.o $(LAZYLOAD_LIBCURL_OBJ) IMAP_SEND_LDFLAGS += $(CURL_LIBCURL) endif ifndef NO_EXPAT @@ -2898,10 +2912,10 @@ git-imap-send$X: imap-send.o $(IMAP_SEND_BUILDDEPS) GIT-LDFLAGS $(GITLIBS) $(QUIET_LINK)$(CC) $(ALL_CFLAGS) -o $@ $(ALL_LDFLAGS) $(filter %.o,$^) \ $(IMAP_SEND_LDFLAGS) $(LIBS) -git-http-fetch$X: http.o http-walker.o http-fetch.o GIT-LDFLAGS $(GITLIBS) +git-http-fetch$X: http.o http-walker.o http-fetch.o $(LAZYLOAD_LIBCURL_OBJ) GIT-LDFLAGS $(GITLIBS) $(QUIET_LINK)$(CC) $(ALL_CFLAGS) -o $@ $(ALL_LDFLAGS) $(filter %.o,$^) \ $(CURL_LIBCURL) $(LIBS) -git-http-push$X: http.o http-push.o GIT-LDFLAGS $(GITLIBS) +git-http-push$X: http.o http-push.o $(LAZYLOAD_LIBCURL_OBJ) GIT-LDFLAGS $(GITLIBS) $(QUIET_LINK)$(CC) $(ALL_CFLAGS) -o $@ $(ALL_LDFLAGS) $(filter %.o,$^) \ $(CURL_LIBCURL) $(EXPAT_LIBEXPAT) $(LIBS) @@ -2911,7 +2925,7 @@ $(REMOTE_CURL_ALIASES): $(REMOTE_CURL_PRIMARY) ln -s $< $@ 2>/dev/null || \ cp $< $@ -$(REMOTE_CURL_PRIMARY): remote-curl.o http.o http-walker.o GIT-LDFLAGS $(GITLIBS) +$(REMOTE_CURL_PRIMARY): remote-curl.o http.o http-walker.o $(LAZYLOAD_LIBCURL_OBJ) GIT-LDFLAGS $(GITLIBS) $(QUIET_LINK)$(CC) $(ALL_CFLAGS) -o $@ $(ALL_LDFLAGS) $(filter %.o,$^) \ $(CURL_LIBCURL) $(EXPAT_LIBEXPAT) $(LIBS) diff --git a/compat/lazyload-curl.c b/compat/lazyload-curl.c new file mode 100644 index 00000000000000..19aa2b6d4b6942 --- /dev/null +++ b/compat/lazyload-curl.c @@ -0,0 +1,354 @@ +#include "../git-compat-util.h" +#include "../git-curl-compat.h" +#include + +/* + * The ABI version of libcurl is encoded in its shared libraries' file names. + * This ABI version has not changed since October 2006 and is unlikely to be + * changed in the future. See https://curl.se/libcurl/abi.html for details. + */ +#define LIBCURL_ABI_VERSION "4" + +typedef void (*func_t)(void); + +#ifdef __APPLE__ +#define LIBCURL_FILE_NAME(base) base "." LIBCURL_ABI_VERSION ".dylib" +#else +#define LIBCURL_FILE_NAME(base) base ".so." LIBCURL_ABI_VERSION +#endif + +static void *load_library(const char *name) +{ + return dlopen(name, RTLD_LAZY); +} + +static func_t load_function(void *handle, const char *name) +{ + /* + * Casting the return value of `dlsym()` to a function pointer is + * explicitly allowed in recent POSIX standards, but GCC complains + * about this in pedantic mode nevertheless. For more about this issue, + * see https://stackoverflow.com/q/31526876/1860823 and + * http://stackoverflow.com/a/36385690/1905491. + */ + func_t f; + *(void **)&f = dlsym(handle, name); + return f; +} + +typedef struct curl_version_info_data *(*curl_version_info_type)(CURLversion version); +static curl_version_info_type curl_version_info_func; + +typedef char *(*curl_easy_escape_type)(CURL *handle, const char *string, int length); +static curl_easy_escape_type curl_easy_escape_func; + +typedef void (*curl_free_type)(void *p); +static curl_free_type curl_free_func; + +typedef CURLcode (*curl_global_init_type)(long flags); +static curl_global_init_type curl_global_init_func; + +typedef CURLsslset (*curl_global_sslset_type)(curl_sslbackend id, const char *name, const curl_ssl_backend ***avail); +static curl_global_sslset_type curl_global_sslset_func; + +typedef void (*curl_global_cleanup_type)(void); +static curl_global_cleanup_type curl_global_cleanup_func; + +typedef struct curl_slist *(*curl_slist_append_type)(struct curl_slist *list, const char *data); +static curl_slist_append_type curl_slist_append_func; + +typedef void (*curl_slist_free_all_type)(struct curl_slist *list); +static curl_slist_free_all_type curl_slist_free_all_func; + +typedef const char *(*curl_easy_strerror_type)(CURLcode error); +static curl_easy_strerror_type curl_easy_strerror_func; + +typedef CURLM *(*curl_multi_init_type)(void); +static curl_multi_init_type curl_multi_init_func; + +typedef CURLMcode (*curl_multi_add_handle_type)(CURLM *multi_handle, CURL *curl_handle); +static curl_multi_add_handle_type curl_multi_add_handle_func; + +typedef CURLMcode (*curl_multi_remove_handle_type)(CURLM *multi_handle, CURL *curl_handle); +static curl_multi_remove_handle_type curl_multi_remove_handle_func; + +typedef CURLMcode (*curl_multi_fdset_type)(CURLM *multi_handle, fd_set *read_fd_set, fd_set *write_fd_set, fd_set *exc_fd_set, int *max_fd); +static curl_multi_fdset_type curl_multi_fdset_func; + +typedef CURLMcode (*curl_multi_perform_type)(CURLM *multi_handle, int *running_handles); +static curl_multi_perform_type curl_multi_perform_func; + +typedef CURLMcode (*curl_multi_cleanup_type)(CURLM *multi_handle); +static curl_multi_cleanup_type curl_multi_cleanup_func; + +typedef CURLMsg *(*curl_multi_info_read_type)(CURLM *multi_handle, int *msgs_in_queue); +static curl_multi_info_read_type curl_multi_info_read_func; + +typedef const char *(*curl_multi_strerror_type)(CURLMcode error); +static curl_multi_strerror_type curl_multi_strerror_func; + +typedef CURLMcode (*curl_multi_timeout_type)(CURLM *multi_handle, long *milliseconds); +static curl_multi_timeout_type curl_multi_timeout_func; + +typedef CURL *(*curl_easy_init_type)(void); +static curl_easy_init_type curl_easy_init_func; + +typedef CURLcode (*curl_easy_perform_type)(CURL *curl); +static curl_easy_perform_type curl_easy_perform_func; + +typedef void (*curl_easy_cleanup_type)(CURL *curl); +static curl_easy_cleanup_type curl_easy_cleanup_func; + +typedef CURL *(*curl_easy_duphandle_type)(CURL *curl); +static curl_easy_duphandle_type curl_easy_duphandle_func; + +typedef CURLcode (*curl_easy_getinfo_long_type)(CURL *curl, CURLINFO info, long *value); +static curl_easy_getinfo_long_type curl_easy_getinfo_long_func; + +typedef CURLcode (*curl_easy_getinfo_pointer_type)(CURL *curl, CURLINFO info, void **value); +static curl_easy_getinfo_pointer_type curl_easy_getinfo_pointer_func; + +typedef CURLcode (*curl_easy_getinfo_off_t_type)(CURL *curl, CURLINFO info, curl_off_t *value); +static curl_easy_getinfo_off_t_type curl_easy_getinfo_off_t_func; + +typedef CURLcode (*curl_easy_setopt_long_type)(CURL *curl, CURLoption opt, long value); +static curl_easy_setopt_long_type curl_easy_setopt_long_func; + +typedef CURLcode (*curl_easy_setopt_pointer_type)(CURL *curl, CURLoption opt, void *value); +static curl_easy_setopt_pointer_type curl_easy_setopt_pointer_func; + +typedef CURLcode (*curl_easy_setopt_off_t_type)(CURL *curl, CURLoption opt, curl_off_t value); +static curl_easy_setopt_off_t_type curl_easy_setopt_off_t_func; + +static void lazy_load_curl(void) +{ + static int initialized; + void *libcurl; + func_t curl_easy_getinfo_func, curl_easy_setopt_func; + + if (initialized) + return; + + initialized = 1; + libcurl = load_library(LIBCURL_FILE_NAME("libcurl")); + if (!libcurl) + die("failed to load library '%s'", LIBCURL_FILE_NAME("libcurl")); + + curl_version_info_func = (curl_version_info_type)load_function(libcurl, "curl_version_info"); + curl_easy_escape_func = (curl_easy_escape_type)load_function(libcurl, "curl_easy_escape"); + curl_free_func = (curl_free_type)load_function(libcurl, "curl_free"); + curl_global_init_func = (curl_global_init_type)load_function(libcurl, "curl_global_init"); + curl_global_sslset_func = (curl_global_sslset_type)load_function(libcurl, "curl_global_sslset"); + curl_global_cleanup_func = (curl_global_cleanup_type)load_function(libcurl, "curl_global_cleanup"); + curl_slist_append_func = (curl_slist_append_type)load_function(libcurl, "curl_slist_append"); + curl_slist_free_all_func = (curl_slist_free_all_type)load_function(libcurl, "curl_slist_free_all"); + curl_easy_strerror_func = (curl_easy_strerror_type)load_function(libcurl, "curl_easy_strerror"); + curl_multi_init_func = (curl_multi_init_type)load_function(libcurl, "curl_multi_init"); + curl_multi_add_handle_func = (curl_multi_add_handle_type)load_function(libcurl, "curl_multi_add_handle"); + curl_multi_remove_handle_func = (curl_multi_remove_handle_type)load_function(libcurl, "curl_multi_remove_handle"); + curl_multi_fdset_func = (curl_multi_fdset_type)load_function(libcurl, "curl_multi_fdset"); + curl_multi_perform_func = (curl_multi_perform_type)load_function(libcurl, "curl_multi_perform"); + curl_multi_cleanup_func = (curl_multi_cleanup_type)load_function(libcurl, "curl_multi_cleanup"); + curl_multi_info_read_func = (curl_multi_info_read_type)load_function(libcurl, "curl_multi_info_read"); + curl_multi_strerror_func = (curl_multi_strerror_type)load_function(libcurl, "curl_multi_strerror"); + curl_multi_timeout_func = (curl_multi_timeout_type)load_function(libcurl, "curl_multi_timeout"); + curl_easy_init_func = (curl_easy_init_type)load_function(libcurl, "curl_easy_init"); + curl_easy_perform_func = (curl_easy_perform_type)load_function(libcurl, "curl_easy_perform"); + curl_easy_cleanup_func = (curl_easy_cleanup_type)load_function(libcurl, "curl_easy_cleanup"); + curl_easy_duphandle_func = (curl_easy_duphandle_type)load_function(libcurl, "curl_easy_duphandle"); + + curl_easy_getinfo_func = load_function(libcurl, "curl_easy_getinfo"); + curl_easy_getinfo_long_func = (curl_easy_getinfo_long_type)curl_easy_getinfo_func; + curl_easy_getinfo_pointer_func = (curl_easy_getinfo_pointer_type)curl_easy_getinfo_func; + curl_easy_getinfo_off_t_func = (curl_easy_getinfo_off_t_type)curl_easy_getinfo_func; + + curl_easy_setopt_func = load_function(libcurl, "curl_easy_setopt"); + curl_easy_setopt_long_func = (curl_easy_setopt_long_type)curl_easy_setopt_func; + curl_easy_setopt_pointer_func = (curl_easy_setopt_pointer_type)curl_easy_setopt_func; + curl_easy_setopt_off_t_func = (curl_easy_setopt_off_t_type)curl_easy_setopt_func; +} + +struct curl_version_info_data *curl_version_info(CURLversion version) +{ + lazy_load_curl(); + return curl_version_info_func(version); +} + +char *curl_easy_escape(CURL *handle, const char *string, int length) +{ + lazy_load_curl(); + return curl_easy_escape_func(handle, string, length); +} + +void curl_free(void *p) +{ + lazy_load_curl(); + curl_free_func(p); +} + +CURLcode curl_global_init(long flags) +{ + lazy_load_curl(); + return curl_global_init_func(flags); +} + +CURLsslset curl_global_sslset(curl_sslbackend id, const char *name, const curl_ssl_backend ***avail) +{ + lazy_load_curl(); + return curl_global_sslset_func(id, name, avail); +} + +void curl_global_cleanup(void) +{ + lazy_load_curl(); + curl_global_cleanup_func(); +} + +struct curl_slist *curl_slist_append(struct curl_slist *list, const char *data) +{ + lazy_load_curl(); + return curl_slist_append_func(list, data); +} + +void curl_slist_free_all(struct curl_slist *list) +{ + lazy_load_curl(); + curl_slist_free_all_func(list); +} + +const char *curl_easy_strerror(CURLcode error) +{ + lazy_load_curl(); + return curl_easy_strerror_func(error); +} + +CURLM *curl_multi_init(void) +{ + lazy_load_curl(); + return curl_multi_init_func(); +} + +CURLMcode curl_multi_add_handle(CURLM *multi_handle, CURL *curl_handle) +{ + lazy_load_curl(); + return curl_multi_add_handle_func(multi_handle, curl_handle); +} + +CURLMcode curl_multi_remove_handle(CURLM *multi_handle, CURL *curl_handle) +{ + lazy_load_curl(); + return curl_multi_remove_handle_func(multi_handle, curl_handle); +} + +CURLMcode curl_multi_fdset(CURLM *multi_handle, fd_set *read_fd_set, fd_set *write_fd_set, fd_set *exc_fd_set, int *max_fd) +{ + lazy_load_curl(); + return curl_multi_fdset_func(multi_handle, read_fd_set, write_fd_set, exc_fd_set, max_fd); +} + +CURLMcode curl_multi_perform(CURLM *multi_handle, int *running_handles) +{ + lazy_load_curl(); + return curl_multi_perform_func(multi_handle, running_handles); +} + +CURLMcode curl_multi_cleanup(CURLM *multi_handle) +{ + lazy_load_curl(); + return curl_multi_cleanup_func(multi_handle); +} + +CURLMsg *curl_multi_info_read(CURLM *multi_handle, int *msgs_in_queue) +{ + lazy_load_curl(); + return curl_multi_info_read_func(multi_handle, msgs_in_queue); +} + +const char *curl_multi_strerror(CURLMcode error) +{ + lazy_load_curl(); + return curl_multi_strerror_func(error); +} + +CURLMcode curl_multi_timeout(CURLM *multi_handle, long *milliseconds) +{ + lazy_load_curl(); + return curl_multi_timeout_func(multi_handle, milliseconds); +} + +CURL *curl_easy_init(void) +{ + lazy_load_curl(); + return curl_easy_init_func(); +} + +CURLcode curl_easy_perform(CURL *curl) +{ + lazy_load_curl(); + return curl_easy_perform_func(curl); +} + +void curl_easy_cleanup(CURL *curl) +{ + lazy_load_curl(); + curl_easy_cleanup_func(curl); +} + +CURL *curl_easy_duphandle(CURL *curl) +{ + lazy_load_curl(); + return curl_easy_duphandle_func(curl); +} + +#ifndef CURL_IGNORE_DEPRECATION +#define CURL_IGNORE_DEPRECATION(x) x +#endif + +#ifndef CURLOPTTYPE_BLOB +#define CURLOPTTYPE_BLOB 40000 +#endif + +#undef curl_easy_getinfo +CURLcode curl_easy_getinfo(CURL *curl, CURLINFO info, ...) +{ + va_list ap; + CURLcode res; + + va_start(ap, info); + lazy_load_curl(); + CURL_IGNORE_DEPRECATION( + if (info >= CURLINFO_LONG && info < CURLINFO_DOUBLE) + res = curl_easy_getinfo_long_func(curl, info, va_arg(ap, long *)); + else if ((info >= CURLINFO_STRING && info < CURLINFO_LONG) || + (info >= CURLINFO_SLIST && info < CURLINFO_SOCKET)) + res = curl_easy_getinfo_pointer_func(curl, info, va_arg(ap, void **)); + else if (info >= CURLINFO_OFF_T) + res = curl_easy_getinfo_off_t_func(curl, info, va_arg(ap, curl_off_t *)); + else + die("%s:%d: TODO (info: %d)!", __FILE__, __LINE__, info); + ) + va_end(ap); + return res; +} + +#undef curl_easy_setopt +CURLcode curl_easy_setopt(CURL *curl, CURLoption opt, ...) +{ + va_list ap; + CURLcode res; + + va_start(ap, opt); + lazy_load_curl(); + CURL_IGNORE_DEPRECATION( + if (opt >= CURLOPTTYPE_LONG && opt < CURLOPTTYPE_OBJECTPOINT) + res = curl_easy_setopt_long_func(curl, opt, va_arg(ap, long)); + else if (opt >= CURLOPTTYPE_OBJECTPOINT && opt < CURLOPTTYPE_OFF_T) + res = curl_easy_setopt_pointer_func(curl, opt, va_arg(ap, void *)); + else if (opt >= CURLOPTTYPE_OFF_T && opt < CURLOPTTYPE_BLOB) + res = curl_easy_setopt_off_t_func(curl, opt, va_arg(ap, curl_off_t)); + else + die("%s:%d: TODO (opt: %d)!", __FILE__, __LINE__, opt); + ) + va_end(ap); + return res; +} From 425d33f59e3b5c3bf5d5d4ad04bb08cac4d2d9cc Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Sun, 7 May 2023 22:51:52 +0200 Subject: [PATCH 095/250] http: support lazy-loading libcurl also on Windows This implements the Windows-specific support code, because everything is slightly different on Windows, even loading shared libraries. Note: I specifically do _not_ use the code from `compat/win32/lazyload.h` here because that code is optimized for loading individual functions from various system DLLs, while we specifically want to load _many_ functions from _one_ DLL here, and distinctly not a system DLL (we expect libcurl to be located outside `C:\Windows\system32`, something `INIT_PROC_ADDR` refuses to work with). Also, the `curl_easy_getinfo()`/`curl_easy_setopt()` functions are declared as vararg functions, which `lazyload.h` cannot handle. Finally, we are about to optionally override the exact file name that is to be loaded, which is a goal contrary to `lazyload.h`'s design. Signed-off-by: Johannes Schindelin --- Makefile | 4 ++++ compat/lazyload-curl.c | 52 ++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 56 insertions(+) diff --git a/Makefile b/Makefile index 8e37b102c94aca..508d2d11cd706f 100644 --- a/Makefile +++ b/Makefile @@ -1676,7 +1676,11 @@ else # The `CURL_STATICLIB` constant must be defined to avoid seeing the functions # declared as DLL imports CURL_CFLAGS = -DCURL_STATICLIB +ifneq ($(uname_S),MINGW) +ifneq ($(uname_S),Windows) CURL_LIBCURL = -ldl +endif +endif else ifndef CURL_LDFLAGS CURL_LDFLAGS = $(eval CURL_LDFLAGS := $$(shell $$(CURL_CONFIG) --libs))$(CURL_LDFLAGS) diff --git a/compat/lazyload-curl.c b/compat/lazyload-curl.c index 19aa2b6d4b6942..98d73fb0f2a66f 100644 --- a/compat/lazyload-curl.c +++ b/compat/lazyload-curl.c @@ -1,6 +1,8 @@ #include "../git-compat-util.h" #include "../git-curl-compat.h" +#ifndef WIN32 #include +#endif /* * The ABI version of libcurl is encoded in its shared libraries' file names. @@ -11,6 +13,7 @@ typedef void (*func_t)(void); +#ifndef WIN32 #ifdef __APPLE__ #define LIBCURL_FILE_NAME(base) base "." LIBCURL_ABI_VERSION ".dylib" #else @@ -35,6 +38,55 @@ static func_t load_function(void *handle, const char *name) *(void **)&f = dlsym(handle, name); return f; } +#else +#define LIBCURL_FILE_NAME(base) base "-" LIBCURL_ABI_VERSION ".dll" + +static void *load_library(const char *name) +{ + size_t name_size = strlen(name) + 1; + const char *path = getenv("PATH"); + char dll_path[MAX_PATH]; + + while (path && *path) { + const char *sep = strchrnul(path, ';'); + size_t len = sep - path; + + if (len && len + name_size < sizeof(dll_path)) { + memcpy(dll_path, path, len); + dll_path[len] = '/'; + memcpy(dll_path + len + 1, name, name_size); + + if (!access(dll_path, R_OK)) { + wchar_t wpath[MAX_PATH]; + int wlen = MultiByteToWideChar(CP_UTF8, 0, dll_path, -1, wpath, ARRAY_SIZE(wpath)); + void *res = wlen ? (void *)LoadLibraryExW(wpath, NULL, 0) : NULL; + if (!res) { + DWORD err = GetLastError(); + char buf[1024]; + + if (!FormatMessageA(FORMAT_MESSAGE_FROM_SYSTEM | + FORMAT_MESSAGE_ARGUMENT_ARRAY | + FORMAT_MESSAGE_IGNORE_INSERTS, + NULL, err, LANG_NEUTRAL, + buf, sizeof(buf) - 1, NULL)) + xsnprintf(buf, sizeof(buf), "last error: %ld", err); + error("LoadLibraryExW() failed with: %s", buf); + } + return res; + } + } + + path = *sep ? sep + 1 : NULL; + } + + return NULL; +} + +static func_t load_function(void *handle, const char *name) +{ + return (func_t)GetProcAddress((HANDLE)handle, name); +} +#endif typedef struct curl_version_info_data *(*curl_version_info_type)(CURLversion version); static curl_version_info_type curl_version_info_func; From b6b2717bf50d921c7a47c78ea354c5eee6926efb Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Sun, 7 May 2023 22:05:33 +0200 Subject: [PATCH 096/250] http: when loading libcurl lazily, allow for multiple SSL backends The previous commits introduced a compile-time option to load libcurl lazily, but it uses the hard-coded name "libcurl-4.dll" (or equivalent on platforms other than Windows). To allow for installing multiple libcurl flavors side by side, where each supports one specific SSL/TLS backend, let's first look whether `libcurl--4.dll` exists, and only use `libcurl-4.dll` as a fall back. That will allow us to ship with a libcurl by default that only supports the Secure Channel backend for the `https://` protocol. This libcurl won't suffer from any dependency problem when upgrading OpenSSL to a new major version (which will change the DLL name, and hence break every program and library that depends on it). This is crucial because Git for Windows relies on libcurl to keep working when building and deploying a new OpenSSL package because that library is used by `git fetch` and `git clone`. Note that this feature is by no means specific to Windows. On Ubuntu, for example, a `git` built using `LAZY_LOAD_LIBCURL` will use `libcurl.so.4` for `http.sslbackend=openssl` and `libcurl-gnutls.so.4` for `http.sslbackend=gnutls`. Signed-off-by: Johannes Schindelin --- compat/lazyload-curl.c | 16 ++++++++++++++-- 1 file changed, 14 insertions(+), 2 deletions(-) diff --git a/compat/lazyload-curl.c b/compat/lazyload-curl.c index 98d73fb0f2a66f..2f2b2ebc27993e 100644 --- a/compat/lazyload-curl.c +++ b/compat/lazyload-curl.c @@ -172,17 +172,26 @@ static curl_easy_setopt_pointer_type curl_easy_setopt_pointer_func; typedef CURLcode (*curl_easy_setopt_off_t_type)(CURL *curl, CURLoption opt, curl_off_t value); static curl_easy_setopt_off_t_type curl_easy_setopt_off_t_func; +static char ssl_backend[64]; + static void lazy_load_curl(void) { static int initialized; - void *libcurl; + void *libcurl = NULL; func_t curl_easy_getinfo_func, curl_easy_setopt_func; if (initialized) return; initialized = 1; - libcurl = load_library(LIBCURL_FILE_NAME("libcurl")); + if (ssl_backend[0]) { + char dll_name[64 + 16]; + snprintf(dll_name, sizeof(dll_name) - 1, + LIBCURL_FILE_NAME("libcurl-%s"), ssl_backend); + libcurl = load_library(dll_name); + } + if (!libcurl) + libcurl = load_library(LIBCURL_FILE_NAME("libcurl")); if (!libcurl) die("failed to load library '%s'", LIBCURL_FILE_NAME("libcurl")); @@ -246,6 +255,9 @@ CURLcode curl_global_init(long flags) CURLsslset curl_global_sslset(curl_sslbackend id, const char *name, const curl_ssl_backend ***avail) { + if (name && strlen(name) < sizeof(ssl_backend)) + strlcpy(ssl_backend, name, sizeof(ssl_backend)); + lazy_load_curl(); return curl_global_sslset_func(id, name, avail); } From b08bb4394abd83140692ebf3a53ea9ac5194c118 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Fri, 27 Jan 2023 08:55:21 +0100 Subject: [PATCH 097/250] windows: skip linking `git-` for built-ins It is merely a historical wart that, say, `git-commit` exists in the `libexec/git-core/` directory, a tribute to the original idea to let Git be essentially a bunch of Unix shell scripts revolving around very few "plumbing" (AKA low-level) commands. Git has evolved a lot from there. These days, most of Git's functionality is contained within the `git` executable, in the form of "built-in" commands. To accommodate for scripts that use the "dashed" form of Git commands, even today, Git provides hard-links that make the `git` executable available as, say, `git-commit`, just in case that an old script has not been updated to invoke `git commit`. Those hard-links do not come cheap: they take about half a minute for every build of Git on Windows, they are mistaken for taking up huge amounts of space by some Windows Explorer versions that do not understand hard-links, and therefore many a "bug" report had to be addressed. The "dashed form" has been officially deprecated in Git version 1.5.4, which was released on February 2nd, 2008, i.e. a very long time ago. This deprecation was never finalized by skipping these hard-links, but we can start the process now, in Git for Windows. Signed-off-by: Johannes Schindelin --- config.mak.uname | 2 ++ 1 file changed, 2 insertions(+) diff --git a/config.mak.uname b/config.mak.uname index 71209def6f0fc4..73acf907e32af2 100644 --- a/config.mak.uname +++ b/config.mak.uname @@ -493,6 +493,7 @@ ifeq ($(uname_S),Windows) NO_POSIX_GOODIES = UnfortunatelyYes NATIVE_CRLF = YesPlease DEFAULT_HELP_FORMAT = html + SKIP_DASHED_BUILT_INS = YabbaDabbaDoo ifeq (/mingw64,$(subst 32,64,$(subst clangarm,mingw,$(prefix)))) # Move system config into top-level /etc/ ETC_GITCONFIG = ../etc/gitconfig @@ -685,6 +686,7 @@ ifeq ($(uname_S),MINGW) FSMONITOR_DAEMON_BACKEND = win32 FSMONITOR_OS_SETTINGS = win32 + SKIP_DASHED_BUILT_INS = YabbaDabbaDoo RUNTIME_PREFIX = YesPlease HAVE_WPGMPTR = YesWeDo NO_ST_BLOCKS_IN_STRUCT_STAT = YesPlease From ba9743042dd5216eb242d16c12b58944fd131b60 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Tue, 21 Mar 2023 16:14:44 +0100 Subject: [PATCH 098/250] windows: fix Repository>Explore Working Copy Since Git v2.39.1, we are a bit more stringent in searching the PATH. In particular, we specifically require the `.exe` suffix. However, the `Repository>Explore Working Copy` command asks for `explorer.exe` to be found on the `PATH`, which _already_ has that suffix. Let's unstartle the PATH-finding logic about this scenario. This fixes https://github.com/git-for-windows/git/issues/4356 Signed-off-by: Johannes Schindelin --- git-gui/git-gui.sh | 3 +++ 1 file changed, 3 insertions(+) diff --git a/git-gui/git-gui.sh b/git-gui/git-gui.sh index 28572c889c0e2b..93555c7a9268c1 100755 --- a/git-gui/git-gui.sh +++ b/git-gui/git-gui.sh @@ -101,6 +101,9 @@ proc _which {what args} { if {[is_Windows] && [lsearch -exact $args -script] >= 0} { set suffix {} + } elseif {[string match *$_search_exe $what]} { + # The search string already has the file extension + set suffix {} } else { set suffix $_search_exe } From 073762f8e6acba311b1baf0fdb8f6dad167d927a Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Sun, 7 May 2023 22:43:37 +0200 Subject: [PATCH 099/250] mingw: do load libcurl dynamically by default This will help with Git for Windows' maintenance going forward: It allows Git for Windows to switch its primary libcurl to a variant without the OpenSSL backend, while still loading an alternate when setting `http.sslBackend = openssl`. This is necessary to avoid maintenance headaches with upgrading OpenSSL: its major version name is encoded in the shared library's file name and hence major version updates (temporarily) break libraries that are linked against the OpenSSL library. Signed-off-by: Johannes Schindelin --- config.mak.uname | 1 + 1 file changed, 1 insertion(+) diff --git a/config.mak.uname b/config.mak.uname index c524a2e434b3e8..7ddd6d86326208 100644 --- a/config.mak.uname +++ b/config.mak.uname @@ -700,6 +700,7 @@ ifeq ($(uname_S),MINGW) HAVE_PLATFORM_PROCINFO = YesPlease CSPRNG_METHOD = rtlgenrandom BASIC_LDFLAGS += -municode -Wl,--tsaware + LAZYLOAD_LIBCURL = YesDoThatPlease COMPAT_CFLAGS += -DNOGDI -Icompat -Icompat/win32 COMPAT_CFLAGS += -DSTRIP_EXTENSION=\".exe\" COMPAT_OBJS += compat/mingw.o compat/winansi.o \ From 85e3ebd1e71321cc72011140761d852e92986036 Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Wed, 2 Nov 2022 16:23:58 +0100 Subject: [PATCH 100/250] Add a GitHub workflow to verify that Git/Scalar work in Nano Server In Git for Windows v2.39.0, we fixed a regression where `git.exe` would no longer work in Windows Nano Server (frequently used in Docker containers). This GitHub workflow can be used to verify manually that the Git/Scalar executables work in Nano Server. Signed-off-by: Johannes Schindelin --- .github/workflows/nano-server.yml | 76 +++++++++++++++++++++++++++++++ 1 file changed, 76 insertions(+) create mode 100644 .github/workflows/nano-server.yml diff --git a/.github/workflows/nano-server.yml b/.github/workflows/nano-server.yml new file mode 100644 index 00000000000000..3d943c23d2616d --- /dev/null +++ b/.github/workflows/nano-server.yml @@ -0,0 +1,76 @@ +name: Windows Nano Server tests + +on: + workflow_dispatch: + +env: + DEVELOPER: 1 + +jobs: + test-nano-server: + runs-on: windows-2022 + env: + WINDBG_DIR: "C:/Program Files (x86)/Windows Kits/10/Debuggers/x64" + IMAGE: mcr.microsoft.com/powershell:nanoserver-ltsc2022 + + steps: + - uses: actions/checkout@v4 + - uses: git-for-windows/setup-git-for-windows-sdk@v1 + - name: build Git + shell: bash + run: make -j15 + - name: pull nanoserver image + shell: bash + run: docker pull $IMAGE + - name: run nano-server test + shell: bash + run: | + docker run \ + --user "ContainerAdministrator" \ + -v "$WINDBG_DIR:C:/dbg" \ + -v "$(cygpath -aw /mingw64/bin):C:/mingw64-bin" \ + -v "$(cygpath -aw .):C:/test" \ + $IMAGE pwsh.exe -Command ' + # Extend the PATH to include the `.dll` files in /mingw64/bin/ + $env:PATH += ";C:\mingw64-bin" + + # For each executable to test pick some no-operation set of + # flags/subcommands or something that should quickly result in an + # error with known exit code that is not a negative 32-bit + # number, and set the expected return code appropriately. + # + # Only test executables that could be expected to run in a UI + # less environment. + # + # ( Executable path, arguments, expected return code ) + # also note space is required before close parenthesis (a + # powershell quirk when defining nested arrays like this) + + $executables_to_test = @( + ("C:\test\git.exe", "", 1 ), + ("C:\test\scalar.exe", "version", 0 ) + ) + + foreach ($executable in $executables_to_test) + { + Write-Output "Now testing $($executable[0])" + &$executable[0] $executable[1] + if ($LASTEXITCODE -ne $executable[2]) { + # if we failed, run the debugger to find out what function + # or DLL could not be found and then exit the script with + # failure The missing DLL or EXE will be referenced near + # the end of the output + + # Set a flag to have the debugger show loader stub + # diagnostics. This requires running as administrator, + # otherwise the flag will be ignored. + C:\dbg\gflags -i $executable[0] +SLS + + C:\dbg\cdb.exe -c "g" -c "q" $executable[0] $executable[1] + + exit 1 + } + } + + exit 0 + ' From a2df6ee45deb3000cdfdc29e98cd0b005badc172 Mon Sep 17 00:00:00 2001 From: David Lomas Date: Fri, 28 Jul 2023 15:31:25 +0100 Subject: [PATCH 101/250] mingw: suggest `windows.appendAtomically` in more cases When running Git for Windows on a remote APFS filesystem, it would appear that the `mingw_open_append()`/`write()` combination would fail almost exactly like on some CIFS-mounted shares as had been reported in https://github.com/git-for-windows/git/issues/2753, albeit with a different `errno` value. Let's handle that `errno` value just the same, by suggesting to set `windows.appendAtomically=false`. Signed-off-by: David Lomas Signed-off-by: Johannes Schindelin --- compat/mingw.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/compat/mingw.c b/compat/mingw.c index c3c4f89447c00d..23e97a8c6e865d 100644 --- a/compat/mingw.c +++ b/compat/mingw.c @@ -810,7 +810,7 @@ ssize_t mingw_write(int fd, const void *buf, size_t len) { ssize_t result = write(fd, buf, len); - if (result < 0 && (errno == EINVAL || errno == ENOSPC) && buf) { + if (result < 0 && (errno == EINVAL || errno == EBADF || errno == ENOSPC) && buf) { int orig = errno; /* check if fd is a pipe */ @@ -836,7 +836,7 @@ ssize_t mingw_write(int fd, const void *buf, size_t len) } errno = orig; - } else if (orig == EINVAL) + } else if (orig == EINVAL || errno == EBADF) errno = EPIPE; else { DWORD buf_size; From 3bb9c2fad3afdc714580158c48b269648843f53e Mon Sep 17 00:00:00 2001 From: Johannes Schindelin Date: Wed, 22 Nov 2023 22:57:38 +0100 Subject: [PATCH 102/250] win32: use native ANSI sequence processing, if possible Windows 10 version 1511 (also known as Anniversary Update), according to https://learn.microsoft.com/en-us/windows/console/console-virtual-terminal-sequences introduced native support for ANSI sequence processing. This allows using colors from the entire 24-bit color range. All we need to do is test whether the console's "virtual processing support" can be enabled. If it can, we do not even need to start the `console_thread` to handle ANSI sequences. Or, almost all we need to do: When `console_thread()` does its work, it uses the Unicode-aware `write_console()` function to write to the Win32 Console, which supports Git for Windows' implicit convention that all text that is written is encoded in UTF-8. The same is not necessarily true if native ANSI sequence processing is used, as the output is then subject to the current code page. Let's ensure that the code page is set to `CP_UTF8` as long as Git writes to it. Signed-off-by: Johannes Schindelin --- compat/winansi.c | 46 ++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 46 insertions(+) diff --git a/compat/winansi.c b/compat/winansi.c index ac2ffb78691a7d..a83a7f47ada9b2 100644 --- a/compat/winansi.c +++ b/compat/winansi.c @@ -593,6 +593,49 @@ static void detect_msys_tty(int fd) #endif +static HANDLE std_console_handle; +static DWORD std_console_mode = ENABLE_VIRTUAL_TERMINAL_PROCESSING; +static UINT std_console_code_page = CP_UTF8; + +static void reset_std_console(void) +{ + if (std_console_mode != ENABLE_VIRTUAL_TERMINAL_PROCESSING) + SetConsoleMode(std_console_handle, std_console_mode); + if (std_console_code_page != CP_UTF8) + SetConsoleOutputCP(std_console_code_page); +} + +static int enable_virtual_processing(void) +{ + std_console_handle = GetStdHandle(STD_OUTPUT_HANDLE); + if (std_console_handle == INVALID_HANDLE_VALUE || + !GetConsoleMode(std_console_handle, &std_console_mode)) { + std_console_handle = GetStdHandle(STD_ERROR_HANDLE); + if (std_console_handle == INVALID_HANDLE_VALUE || + !GetConsoleMode(std_console_handle, &std_console_mode)) + return 0; + } + + std_console_code_page = GetConsoleOutputCP(); + if (std_console_code_page != CP_UTF8) + SetConsoleOutputCP(CP_UTF8); + if (!std_console_code_page) + std_console_code_page = CP_UTF8; + + atexit(reset_std_console); + + if (std_console_mode & ENABLE_VIRTUAL_TERMINAL_PROCESSING) + return 1; + + if (!SetConsoleMode(std_console_handle, + std_console_mode | + ENABLE_PROCESSED_OUTPUT | + ENABLE_VIRTUAL_TERMINAL_PROCESSING)) + return 0; + + return 1; +} + /* * Wrapper for isatty(). Most calls in the main git code * call isatty(1 or 2) to see if the instance is interactive @@ -631,6 +674,9 @@ void winansi_init(void) return; } + if (enable_virtual_processing()) + return; + /* create a named pipe to communicate with the console thread */ if (swprintf(name, ARRAY_SIZE(name) - 1, L"\\\\.\\pipe\\winansi%lu", GetCurrentProcessId()) < 0) From a693ad9e5bb3f8e1c2a07dcc50c670b024956c8c Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Matthias=20A=C3=9Fhauer?= Date: Sat, 2 Dec 2023 12:10:00 +0100 Subject: [PATCH 103/250] git.rc: include winuser.h MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit winuser.h contains the definition of RT_MANIFEST that our LLVM based toolchain needs to understand that we want to embed compat/win32/git.manifest as an application manifest. It currently just embeds it as additional data that Windows doesn't understand. This also helps our GCC based toolchain understand that we only want one copy embedded. It currently embeds one working assembly manifest and one nearly identical, but useless copy as additional data. This also teaches our Visual Studio based buildsystems to pick up the manifest file from git.rc. This means we don't have to explicitly specify it in contrib/buildsystems/Generators/Vcxproj.pm anymore. Slightly counter-intuitively this also means we have to explicitly tell Cmake not to embed a default manifest. This fixes https://github.com/git-for-windows/git/issues/4707 Signed-off-by: Matthias Aßhauer Signed-off-by: Johannes Schindelin --- contrib/buildsystems/CMakeLists.txt | 1 + git.rc.in | 1 + 2 files changed, 2 insertions(+) diff --git a/contrib/buildsystems/CMakeLists.txt b/contrib/buildsystems/CMakeLists.txt index 25b495fa737d07..5ac5850100033c 100644 --- a/contrib/buildsystems/CMakeLists.txt +++ b/contrib/buildsystems/CMakeLists.txt @@ -208,6 +208,7 @@ if(CMAKE_C_COMPILER_ID STREQUAL "MSVC") set(CMAKE_RUNTIME_OUTPUT_DIRECTORY_DEBUG ${CMAKE_BINARY_DIR}) set(CMAKE_RUNTIME_OUTPUT_DIRECTORY_RELEASE ${CMAKE_BINARY_DIR}) add_compile_options(/MP /std:c11) + add_link_options(/MANIFEST:NO) endif() #default behaviour diff --git a/git.rc.in b/git.rc.in index e69444eef3f0c5..1d5b627b610549 100644 --- a/git.rc.in +++ b/git.rc.in @@ -1,3 +1,4 @@ +#include 1 VERSIONINFO FILEVERSION @GIT_MAJOR_VERSION@,@GIT_MINOR_VERSION@,@GIT_MICRO_VERSION@,@GIT_PATCH_LEVEL@ PRODUCTVERSION @GIT_MAJOR_VERSION@,@GIT_MINOR_VERSION@,@GIT_MICRO_VERSION@,@GIT_PATCH_LEVEL@ From bc719e22c4cad6d8a978b9f62d1afb1a176bc881 Mon Sep 17 00:00:00 2001 From: MinarKotonoha Date: Mon, 8 Apr 2024 16:41:10 +0800 Subject: [PATCH 104/250] common-main.c: fflush stdout buffer upon exit By default, the buffer type of Windows' `stdout` is unbuffered (_IONBF), and there is no need to manually fflush `stdout`. But some programs, such as the Windows Filtering Platform driver provided by the security software, may change the buffer type of `stdout` to full buffering. This nees `fflush(stdout)` to be called manually, otherwise there will be no output to `stdout`. Signed-off-by: MinarKotonoha Signed-off-by: Johannes Schindelin --- common-exit.c | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/common-exit.c b/common-exit.c index 1aaa538be3ed67..609f32abed8b53 100644 --- a/common-exit.c +++ b/common-exit.c @@ -11,6 +11,13 @@ static void check_bug_if_BUG(void) /* We wrap exit() to call common_exit() in git-compat-util.h */ int common_exit(const char *file, int line, int code) { + /* + * Windows Filtering Platform driver provided by the security software + * may change buffer type of stdout from _IONBF to _IOFBF. + * It will no output without fflush manually. + */ + fflush(stdout); + /* * For non-POSIX systems: Take the lowest 8 bits of the "code" * to e.g. turn -1 into 255. On a POSIX system this is From f2090107338ce7bfc3493ba072a900a39f61e322 Mon Sep 17 00:00:00 2001 From: Derrick Stolee Date: Fri, 6 Sep 2024 14:16:13 -0400 Subject: [PATCH 105/250] revision: create mark_trees_uninteresting_dense() The sparse tree walk algorithm was created in d5d2e93577e (revision: implement sparse algorithm, 2019-01-16) and involves using the mark_trees_uninteresting_sparse() method. This method takes a repository and an oidset of tree IDs, some of which have the UNINTERESTING flag and some of which do not. Create a method that has an equivalent set of preconditions but uses a "dense" walk (recursively visits all reachable trees, as long as they have not previously been marked UNINTERESTING). This is an important difference from mark_tree_uninteresting(), which short-circuits if the given tree has the UNINTERESTING flag. A use of this method will be added in a later change, with a condition set whether the sparse or dense approach should be used. Signed-off-by: Derrick Stolee --- revision.c | 15 +++++++++++++++ revision.h | 1 + 2 files changed, 16 insertions(+) diff --git a/revision.c b/revision.c index 2c36a9c179efb5..9a203ed7fdd2ec 100644 --- a/revision.c +++ b/revision.c @@ -212,6 +212,21 @@ static void add_children_by_path(struct repository *r, free_tree_buffer(tree); } +void mark_trees_uninteresting_dense(struct repository *r, + struct oidset *trees) +{ + struct object_id *oid; + struct oidset_iter iter; + + oidset_iter_init(trees, &iter); + while ((oid = oidset_iter_next(&iter))) { + struct tree *tree = lookup_tree(r, oid); + + if (tree->object.flags & UNINTERESTING) + mark_tree_contents_uninteresting(r, tree); + } +} + void mark_trees_uninteresting_sparse(struct repository *r, struct oidset *trees) { diff --git a/revision.h b/revision.h index 6d369cdad6a473..4fb276cbd7c8ae 100644 --- a/revision.h +++ b/revision.h @@ -486,6 +486,7 @@ void put_revision_mark(const struct rev_info *revs, void mark_parents_uninteresting(struct rev_info *revs, struct commit *commit); void mark_tree_uninteresting(struct repository *r, struct tree *tree); +void mark_trees_uninteresting_dense(struct repository *r, struct oidset *trees); void mark_trees_uninteresting_sparse(struct repository *r, struct oidset *trees); /** From 2faba742663f1bdee127d954b800baeb2f2465ac Mon Sep 17 00:00:00 2001 From: Derrick Stolee Date: Tue, 24 Sep 2024 13:45:19 -0400 Subject: [PATCH 106/250] pack-objects: extract should_attempt_deltas() This will be helpful in a future change. Signed-off-by: Derrick Stolee --- builtin/pack-objects.c | 53 +++++++++++++++++++++++------------------- 1 file changed, 29 insertions(+), 24 deletions(-) diff --git a/builtin/pack-objects.c b/builtin/pack-objects.c index 8b33edc2ff525a..ce64cdd08336d4 100644 --- a/builtin/pack-objects.c +++ b/builtin/pack-objects.c @@ -3283,6 +3283,33 @@ static int add_ref_tag(const char *tag UNUSED, const char *referent UNUSED, cons return 0; } +static int should_attempt_deltas(struct object_entry *entry) +{ + if (DELTA(entry)) + return 0; + + if (!entry->type_valid || + oe_size_less_than(&to_pack, entry, 50)) + return 0; + + if (entry->no_try_delta) + return 0; + + if (!entry->preferred_base) { + if (oe_type(entry) < 0) + die(_("unable to get type of object %s"), + oid_to_hex(&entry->idx.oid)); + } else if (oe_type(entry) < 0) { + /* + * This object is not found, but we + * don't have to include it anyway. + */ + return 0; + } + + return 1; +} + static void prepare_pack(int window, int depth) { struct object_entry **delta_list; @@ -3313,33 +3340,11 @@ static void prepare_pack(int window, int depth) for (i = 0; i < to_pack.nr_objects; i++) { struct object_entry *entry = to_pack.objects + i; - if (DELTA(entry)) - /* This happens if we decided to reuse existing - * delta from a pack. "reuse_delta &&" is implied. - */ - continue; - - if (!entry->type_valid || - oe_size_less_than(&to_pack, entry, 50)) + if (!should_attempt_deltas(entry)) continue; - if (entry->no_try_delta) - continue; - - if (!entry->preferred_base) { + if (!entry->preferred_base) nr_deltas++; - if (oe_type(entry) < 0) - die(_("unable to get type of object %s"), - oid_to_hex(&entry->idx.oid)); - } else { - if (oe_type(entry) < 0) { - /* - * This object is not found, but we - * don't have to include it anyway. - */ - continue; - } - } delta_list[n++] = entry; } From b12b4368b1579226acfcb8da0458f4dfe378f006 Mon Sep 17 00:00:00 2001 From: Derrick Stolee Date: Thu, 5 Sep 2024 10:04:51 -0400 Subject: [PATCH 107/250] pack-objects: add --path-walk option In order to more easily compute delta bases among objects that appear at the exact same path, add a --path-walk option to 'git pack-objects'. This option will use the path-walk API instead of the object walk given by the revision machinery. Since objects will be provided in batches representing a common path, those objects can be tested for delta bases immediately instead of waiting for a sort of the full object list by name-hash. This has multiple benefits, including avoiding collisions by name-hash. The objects marked as UNINTERESTING are included in these batches, so we are guaranteeing some locality to find good delta bases. After the individual passes are done on a per-path basis, the default name-hash is used to find other opportunistic delta bases that did not match exactly by the full path name. RFC TODO: It is important to note that this option is inherently incompatible with using a bitmap index. This walk probably also does not work with other advanced features, such as delta islands. Getting ahead of myself, this option compares well with --full-name-hash when the packfile is large enough, but also performs at least as well as the default in all cases that I've seen. RFC TODO: this should probably be recording the batch locations to another list so they could be processed in a second phase using threads. RFC TODO: list some examples of how this outperforms previous pack-objects strategies. (This is coming in later commits that include performance test changes.) Signed-off-by: Derrick Stolee --- Documentation/git-pack-objects.adoc | 12 +- Documentation/technical/api-path-walk.adoc | 1 + builtin/pack-objects.c | 147 +++++++++++++++++++-- t/t5300-pack-object.sh | 17 +++ 4 files changed, 167 insertions(+), 10 deletions(-) diff --git a/Documentation/git-pack-objects.adoc b/Documentation/git-pack-objects.adoc index 7f69ae4855f6ce..a91aa7989cbeff 100644 --- a/Documentation/git-pack-objects.adoc +++ b/Documentation/git-pack-objects.adoc @@ -16,7 +16,7 @@ SYNOPSIS [--cruft] [--cruft-expiration=