Auto merge of rust-lang#127726 - RalfJung:miri-sync, r=RalfJung

Miri subtree update

r? `@ghost`

Author: bors

src/tools/miri/cargo-miri/src/phases.rs

@@ -104,9 +104,17 @@ pub fn phase_cargo_miri(mut args: impl Iterator<Item = String>) {
             miri_for_host()
         )
     });
-    let host = &rustc_version.host;
-    let target = get_arg_flag_value("--target");
-    let target = target.as_ref().unwrap_or(host);
+    let mut targets = get_arg_flag_values("--target").collect::<Vec<_>>();
+    // If `targets` is empty, we need to add a `--target $HOST` flag ourselves, and also ensure
+    // that the host target is indeed setup.
+    let target_flag = if targets.is_empty() {
+        let host = &rustc_version.host;
+        targets.push(host.clone());
+        Some(host)
+    } else {
+        // We don't need to add a `--target` flag, we just forward the user's flags.
+        None
+    };
 
     // If cleaning the target directory & sysroot cache,
     // delete them then exit. There is no reason to setup a new
@@ -118,8 +126,11 @@ pub fn phase_cargo_miri(mut args: impl Iterator<Item = String>) {
         return;
     }
 
-    // We always setup.
-    let miri_sysroot = setup(&subcommand, target, &rustc_version, verbose, quiet);
+    for target in &targets {
+        // We always setup.
+        setup(&subcommand, target.as_str(), &rustc_version, verbose, quiet);
+    }
+    let miri_sysroot = get_sysroot_dir();
 
     // Invoke actual cargo for the job, but with different flags.
     // We re-use `cargo test` and `cargo run`, which makes target and binary handling very easy but
@@ -155,10 +166,9 @@ pub fn phase_cargo_miri(mut args: impl Iterator<Item = String>) {
     // This is needed to make the `target.runner` settings do something,
     // and it later helps us detect which crates are proc-macro/build-script
     // (host crates) and which crates are needed for the program itself.
-    if get_arg_flag_value("--target").is_none() {
-        // No target given. Explicitly pick the host.
+    if let Some(target_flag) = target_flag {
         cmd.arg("--target");
-        cmd.arg(host);
+        cmd.arg(target_flag);
     }
 
     // Set ourselves as runner for al binaries invoked by cargo.

src/tools/miri/ci/ci.sh

@@ -41,9 +41,11 @@ function run_tests {
   if [ -n "${TEST_TARGET-}" ]; then
     begingroup "Testing foreign architecture $TEST_TARGET"
     TARGET_FLAG="--target $TEST_TARGET"
+    MULTI_TARGET_FLAG=""
   else
     begingroup "Testing host architecture"
     TARGET_FLAG=""
+    MULTI_TARGET_FLAG="--multi-target"
   fi
 
   ## ui test suite
@@ -93,7 +95,7 @@ function run_tests {
     echo 'build.rustc-wrapper = "thisdoesnotexist"' > .cargo/config.toml
   fi
   # Run the actual test
-  time ${PYTHON} test-cargo-miri/run-test.py $TARGET_FLAG
+  time ${PYTHON} test-cargo-miri/run-test.py $TARGET_FLAG $MULTI_TARGET_FLAG
   # Clean up
   unset RUSTC MIRI
   rm -rf .cargo

src/tools/miri/rust-version

@@ -1 +1 @@
-66b4f0021bfb11a8c20d084c99a40f4a78ce1d38
+99b7134389e9766462601a2fc4013840b9d31745

src/tools/miri/src/bin/miri.rs

@@ -592,6 +592,9 @@ fn main() {
             let num_cpus = param
                 .parse::<u32>()
                 .unwrap_or_else(|err| show_error!("-Zmiri-num-cpus requires a `u32`: {}", err));
+            if !(1..=miri::MAX_CPUS).contains(&usize::try_from(num_cpus).unwrap()) {
+                show_error!("-Zmiri-num-cpus must be in the range 1..={}", miri::MAX_CPUS);
+            }
             miri_config.num_cpus = num_cpus;
         } else if let Some(param) = arg.strip_prefix("-Zmiri-force-page-size=") {
             let page_size = param.parse::<u64>().unwrap_or_else(|err| {

src/tools/miri/src/borrow_tracker/stacked_borrows/stack.rs

@@ -136,8 +136,16 @@ impl StackCache {
 
 impl PartialEq for Stack {
     fn eq(&self, other: &Self) -> bool {
-        // All the semantics of Stack are in self.borrows, everything else is caching
-        self.borrows == other.borrows
+        let Stack {
+            borrows,
+            unknown_bottom,
+            // The cache is ignored for comparison.
+            #[cfg(feature = "stack-cache")]
+                cache: _,
+            #[cfg(feature = "stack-cache")]
+                unique_range: _,
+        } = self;
+        *borrows == other.borrows && *unknown_bottom == other.unknown_bottom
     }
 }
 

src/tools/miri/src/concurrency/cpu_affinity.rs

@@ -0,0 +1,90 @@
+use rustc_middle::ty::layout::LayoutOf;
+use rustc_target::abi::Endian;
+
+use crate::*;
+
+/// The maximum number of CPUs supported by miri.
+///
+/// This value is compatible with the libc `CPU_SETSIZE` constant and corresponds to the number
+/// of CPUs that a `cpu_set_t` can contain.
+///
+/// Real machines can have more CPUs than this number, and there exist APIs to set their affinity,
+/// but this is not currently supported by miri.
+pub const MAX_CPUS: usize = 1024;
+
+/// A thread's CPU affinity mask determines the set of CPUs on which it is eligible to run.
+// the actual representation depends on the target's endianness and pointer width.
+// See CpuAffinityMask::set for details
+#[derive(Clone)]
+pub(crate) struct CpuAffinityMask([u8; Self::CPU_MASK_BYTES]);
+
+impl CpuAffinityMask {
+    pub(crate) const CPU_MASK_BYTES: usize = MAX_CPUS / 8;
+
+    pub fn new<'tcx>(cx: &impl LayoutOf<'tcx>, cpu_count: u32) -> Self {
+        let mut this = Self([0; Self::CPU_MASK_BYTES]);
+
+        // the default affinity mask includes only the available CPUs
+        for i in 0..cpu_count as usize {
+            this.set(cx, i);
+        }
+
+        this
+    }
+
+    pub fn chunk_size<'tcx>(cx: &impl LayoutOf<'tcx>) -> u64 {
+        // The actual representation of the CpuAffinityMask is [c_ulong; _].
+        let ulong = helpers::path_ty_layout(cx, &["core", "ffi", "c_ulong"]);
+        ulong.size.bytes()
+    }
+
+    fn set<'tcx>(&mut self, cx: &impl LayoutOf<'tcx>, cpu: usize) {
+        // we silently ignore CPUs that are out of bounds. This matches the behavior of
+        // `sched_setaffinity` with a mask that specifies more than `CPU_SETSIZE` CPUs.
+        if cpu >= MAX_CPUS {
+            return;
+        }
+
+        // The actual representation of the CpuAffinityMask is [c_ulong; _].
+        // Within the array elements, we need to use the endianness of the target.
+        let target = &cx.tcx().sess.target;
+        match Self::chunk_size(cx) {
+            4 => {
+                let start = cpu / 32 * 4; // first byte of the correct u32
+                let chunk = self.0[start..].first_chunk_mut::<4>().unwrap();
+                let offset = cpu % 32;
+                *chunk = match target.options.endian {
+                    Endian::Little => (u32::from_le_bytes(*chunk) | 1 << offset).to_le_bytes(),
+                    Endian::Big => (u32::from_be_bytes(*chunk) | 1 << offset).to_be_bytes(),
+                };
+            }
+            8 => {
+                let start = cpu / 64 * 8; // first byte of the correct u64
+                let chunk = self.0[start..].first_chunk_mut::<8>().unwrap();
+                let offset = cpu % 64;
+                *chunk = match target.options.endian {
+                    Endian::Little => (u64::from_le_bytes(*chunk) | 1 << offset).to_le_bytes(),
+                    Endian::Big => (u64::from_be_bytes(*chunk) | 1 << offset).to_be_bytes(),
+                };
+            }
+            other => bug!("chunk size not supported: {other}"),
+        };
+    }
+
+    pub fn as_slice(&self) -> &[u8] {
+        self.0.as_slice()
+    }
+
+    pub fn from_array<'tcx>(
+        cx: &impl LayoutOf<'tcx>,
+        cpu_count: u32,
+        bytes: [u8; Self::CPU_MASK_BYTES],
+    ) -> Option<Self> {
+        // mask by what CPUs are actually available
+        let default = Self::new(cx, cpu_count);
+        let masked = std::array::from_fn(|i| bytes[i] & default.0[i]);
+
+        // at least one thread must be set for the input to be valid
+        masked.iter().any(|b| *b != 0).then_some(Self(masked))
+    }
+}

src/tools/miri/src/concurrency/mod.rs

@@ -1,3 +1,4 @@
+pub mod cpu_affinity;
 pub mod data_race;
 pub mod init_once;
 mod range_object_map;

src/tools/miri/src/concurrency/sync.rs

@@ -269,7 +269,7 @@ pub(super) trait EvalContextExtPriv<'tcx>: crate::MiriInterpCxExt<'tcx> {
         let this = self.eval_context_mut();
         if this.mutex_is_locked(mutex) {
             assert_ne!(this.mutex_get_owner(mutex), this.active_thread());
-            this.mutex_enqueue_and_block(mutex, retval, dest);
+            this.mutex_enqueue_and_block(mutex, Some((retval, dest)));
         } else {
             // We can have it right now!
             this.mutex_lock(mutex);
@@ -390,9 +390,15 @@ pub trait EvalContextExt<'tcx>: crate::MiriInterpCxExt<'tcx> {
     }
 
     /// Put the thread into the queue waiting for the mutex.
-    /// Once the Mutex becomes available, `retval` will be written to `dest`.
+    ///
+    /// Once the Mutex becomes available and if it exists, `retval_dest.0` will
+    /// be written to `retval_dest.1`.
     #[inline]
-    fn mutex_enqueue_and_block(&mut self, id: MutexId, retval: Scalar, dest: MPlaceTy<'tcx>) {
+    fn mutex_enqueue_and_block(
+        &mut self,
+        id: MutexId,
+        retval_dest: Option<(Scalar, MPlaceTy<'tcx>)>,
+    ) {
         let this = self.eval_context_mut();
         assert!(this.mutex_is_locked(id), "queing on unlocked mutex");
         let thread = this.active_thread();
@@ -403,13 +409,16 @@ pub trait EvalContextExt<'tcx>: crate::MiriInterpCxExt<'tcx> {
             callback!(
                 @capture<'tcx> {
                     id: MutexId,
-                    retval: Scalar,
-                    dest: MPlaceTy<'tcx>,
+                    retval_dest: Option<(Scalar, MPlaceTy<'tcx>)>,
                 }
                 @unblock = |this| {
                     assert!(!this.mutex_is_locked(id));
                     this.mutex_lock(id);
-                    this.write_scalar(retval, &dest)?;
+
+                    if let Some((retval, dest)) = retval_dest {
+                        this.write_scalar(retval, &dest)?;
+                    }
+
                     Ok(())
                 }
             ),

src/tools/miri/src/concurrency/thread.rs

@@ -936,6 +936,11 @@ pub trait EvalContextExt<'tcx>: crate::MiriInterpCxExt<'tcx> {
         // After this all accesses will be treated as occurring in the new thread.
         let old_thread_id = this.machine.threads.set_active_thread_id(new_thread_id);
 
+        // The child inherits its parent's cpu affinity.
+        if let Some(cpuset) = this.machine.thread_cpu_affinity.get(&old_thread_id).cloned() {
+            this.machine.thread_cpu_affinity.insert(new_thread_id, cpuset);
+        }
+
         // Perform the function pointer load in the new thread frame.
         let instance = this.get_ptr_fn(start_routine)?.as_instance()?;
 

src/tools/miri/src/eval.rs

@@ -282,7 +282,8 @@ pub fn create_ecx<'tcx>(
     })?;
 
     // Make sure we have MIR. We check MIR for some stable monomorphic function in libcore.
-    let sentinel = ecx.try_resolve_path(&["core", "ascii", "escape_default"], Namespace::ValueNS);
+    let sentinel =
+        helpers::try_resolve_path(tcx, &["core", "ascii", "escape_default"], Namespace::ValueNS);
     if !matches!(sentinel, Some(s) if tcx.is_mir_available(s.def.def_id())) {
         tcx.dcx().fatal(
             "the current sysroot was built without `-Zalways-encode-mir`, or libcore seems missing. \