[lld][AArch64][Build Attributes] Add support for AArch64 Build Attributes

lld/ELF/InputFiles.cpp

@@ -21,6 +21,7 @@
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/LTO/LTO.h"
 #include "llvm/Object/IRObjectFile.h"
+#include "llvm/Support/AArch64AttributeParser.h"
 #include "llvm/Support/ARMAttributeParser.h"
 #include "llvm/Support/ARMBuildAttributes.h"
 #include "llvm/Support/Endian.h"
@@ -537,6 +538,52 @@ uint32_t ObjFile<ELFT>::getSectionIndex(const Elf_Sym &sym) const {
       this);
 }
 
+template <class ELFT>
+static void
+handleAArch64BAAndGnuProperties(ObjFile<ELFT> *file, Ctx &ctx, bool hasGP,
+                                const AArch64BuildAttrSubsections &baInfo,
+                                const GnuPropertiesInfo &gpInfo) {
+  if (hasGP) {
+    // Check for data mismatch
+    if (gpInfo.pauthAbiCoreInfo) {
+      if (baInfo.Pauth.TagPlatform != gpInfo.pauthAbiCoreInfo->platform ||
+          baInfo.Pauth.TagSchema != gpInfo.pauthAbiCoreInfo->version)
+        Err(ctx)
+            << file
+            << " Pauth Data mismatch: file contains both GNU properties and "
+               "AArch64 build attributes sections with different Pauth data";
+    }
+    if (baInfo.AndFeatures != gpInfo.andFeatures)
+      Err(ctx) << file
+               << " Features Data mismatch: file contains both GNU "
+                  "properties and AArch64 build attributes sections with "
+                  "different And Features data";
+  } else {
+    // Write missing data
+    // We can only know when Pauth is missing.
+    // Unlike AArch64 Build Attributes, GNU properties does not give a way to
+    // distinguish between no-value given to value of '0' given.
+    if (baInfo.Pauth.TagPlatform || baInfo.Pauth.TagSchema) {
+      // According to the BuildAttributes specification Build Attributes
+      // default to a value of 0 when not present. A (TagPlatform, TagSchema) of
+      // (0, 0) maps to 'no PAuth property present'. A (TagPlatform, TagSchema)
+      // of (0, 1) maps to an explicit PAuth property of platform = 0, version =
+      // 0 ('Invalid').
+      if (baInfo.Pauth.TagPlatform == 0 && baInfo.Pauth.TagSchema == 1) {
+        file->aarch64PauthAbiCoreInfo = {0, 0};
+      } else {
+        file->aarch64PauthAbiCoreInfo = {baInfo.Pauth.TagPlatform,
+                                         baInfo.Pauth.TagSchema};
+      }
+    }
+    file->andFeatures = baInfo.AndFeatures;
+  }
+}
+
+template <typename ELFT>
+static GnuPropertiesInfo readGnuProperty(Ctx &, const InputSection &,
+                                         ObjFile<ELFT> &);
+
 template <class ELFT> void ObjFile<ELFT>::parse(bool ignoreComdats) {
   object::ELFFile<ELFT> obj = this->getObj();
   // Read a section table. justSymbols is usually false.
@@ -552,8 +599,31 @@ template <class ELFT> void ObjFile<ELFT>::parse(bool ignoreComdats) {
   StringRef shstrtab = CHECK2(obj.getSectionStringTable(objSections), this);
   uint64_t size = objSections.size();
   sections.resize(size);
+
+  // For handling AArch64 Build attributes and GNU properties
+  AArch64BuildAttrSubsections aarch64BAsubSections;
+  GnuPropertiesInfo gnuProperty;
+  bool hasAArch64BuildAttributes = false;
+  bool hasGNUProperties = false;
+
   for (size_t i = 0; i != size; ++i) {
     const Elf_Shdr &sec = objSections[i];
+    // Object files that use processor features such as Intel Control-Flow
+    // Enforcement (CET) or AArch64 Branch Target Identification BTI, use a
+    // .note.gnu.property section containing a bitfield of feature bits like the
+    // GNU_PROPERTY_X86_FEATURE_1_IBT flag. Read a bitmap containing the flag.
+    if (check(obj.getSectionName(sec, shstrtab)) == ".note.gnu.property") {
+      gnuProperty = readGnuProperty(
+          ctx,
+          InputSection(*this, sec, check(obj.getSectionName(sec, shstrtab))),
+          *this);
+      hasGNUProperties = true;
+      // Since we merge bitmaps from multiple object files to create a new
+      // .note.gnu.property containing a single AND'ed bitmap, we discard an
+      // input file's .note.gnu.property section.
+      sections[i] = &InputSection::discarded;
+    }
+
     if (LLVM_LIKELY(sec.sh_type == SHT_PROGBITS))
       continue;
     if (LLVM_LIKELY(sec.sh_type == SHT_GROUP)) {
@@ -637,13 +707,27 @@ template <class ELFT> void ObjFile<ELFT>::parse(bool ignoreComdats) {
       }
       break;
     case EM_AARCH64:
-      // FIXME: BuildAttributes have been implemented in llvm, but not yet in
-      // lld. Remove the section so that it does not accumulate in the output
-      // file. When support is implemented we expect not to output a build
-      // attributes section in files of type ET_EXEC or ET_SHARED, but ld -r
-      // ouptut will need a single merged attributes section.
-      if (sec.sh_type == SHT_AARCH64_ATTRIBUTES)
+      // At this stage AArch64 Build Attributes does not replace GNU Properties.
+      // When both exists, their values must match.
+      // When both exists and contain different attributes, they complement each
+      // other. Currently attributes are represented in the linked object file
+      // as GNU properties, which are already supported by the Linux kernel and
+      // the dynamic loader. In the future, when relocatable linking (`-r` flag)
+      // is performed, a single merged AArch64 Build Attributes section will be
+      // emitted.
+      if (sec.sh_type == SHT_AARCH64_ATTRIBUTES) {
+        ArrayRef<uint8_t> contents = check(obj.getSectionContents(sec));
+        AArch64AttributeParser attributes;
+        StringRef name = check(obj.getSectionName(sec, shstrtab));
+        InputSection isec(*this, sec, name);
+        if (Error e = attributes.parse(contents, ELFT::Endianness)) {
+          Warn(ctx) << &isec << ": " << std::move(e);
+        } else {
+          aarch64BAsubSections = extractBuildAttributesSubsections(attributes);
+          hasAArch64BuildAttributes = true;
+        }
         sections[i] = &InputSection::discarded;
+      }
       // Producing a static binary with MTE globals is not currently supported,
       // remove all SHT_AARCH64_MEMTAG_GLOBALS_STATIC sections as they're unused
       // medatada, and we don't want them to end up in the output file for
@@ -655,6 +739,14 @@ template <class ELFT> void ObjFile<ELFT>::parse(bool ignoreComdats) {
     }
   }
 
+  if (hasAArch64BuildAttributes) {
+    // Handle AArch64 Build Attributes and GNU properties:
+    // - Err on mismatched values.
+    // - Store missing values as GNU properties.
+    handleAArch64BAAndGnuProperties<ELFT>(this, ctx, hasGNUProperties,
+                                          aarch64BAsubSections, gnuProperty);
+  }
+
   // Read a symbol table.
   initializeSymbols(obj);
 }
@@ -974,8 +1066,8 @@ static void parseGnuPropertyNote(Ctx &ctx, ELFFileBase &f,
 //   hardware-assisted call flow control;
 // - AArch64 PAuth ABI core info (16 bytes).
 template <class ELFT>
-static void readGnuProperty(Ctx &ctx, const InputSection &sec,
-                            ObjFile<ELFT> &f) {
+static GnuPropertiesInfo readGnuProperty(Ctx &ctx, const InputSection &sec,
+                                         ObjFile<ELFT> &f) {
   using Elf_Nhdr = typename ELFT::Nhdr;
   using Elf_Note = typename ELFT::Note;
 
@@ -992,7 +1084,7 @@ static void readGnuProperty(Ctx &ctx, const InputSection &sec,
     featureAndType = GNU_PROPERTY_RISCV_FEATURE_1_AND;
     break;
   default:
-    return;
+    return GnuPropertiesInfo{};
   }
 
   ArrayRef<uint8_t> data = sec.content();
@@ -1007,7 +1099,7 @@ static void readGnuProperty(Ctx &ctx, const InputSection &sec,
     auto *nhdr = reinterpret_cast<const Elf_Nhdr *>(data.data());
     if (data.size() < sizeof(Elf_Nhdr) ||
         data.size() < nhdr->getSize(sec.addralign))
-      return void(err(data.data()) << "data is too short");
+      return (err(data.data()) << "data is too short", GnuPropertiesInfo{});
 
     Elf_Note note(*nhdr);
     if (nhdr->n_type != NT_GNU_PROPERTY_TYPE_0 || note.getName() != "GNU") {
@@ -1023,6 +1115,7 @@ static void readGnuProperty(Ctx &ctx, const InputSection &sec,
     // Go to next NOTE record to look for more FEATURE_1_AND descriptions.
     data = data.slice(nhdr->getSize(sec.addralign));
   }
+  return GnuPropertiesInfo{f.andFeatures, f.aarch64PauthAbiCoreInfo};
 }
 
 template <class ELFT>

lld/ELF/InputFiles.h

@@ -244,6 +244,11 @@ class ELFFileBase : public InputFile {
   std::optional<AArch64PauthAbiCoreInfo> aarch64PauthAbiCoreInfo;
 };
 
+struct GnuPropertiesInfo {
+  uint32_t andFeatures = 0;
+  std::optional<AArch64PauthAbiCoreInfo> pauthAbiCoreInfo;
+};
+
 // .o file.
 template <class ELFT> class ObjFile : public ELFFileBase {
   LLVM_ELF_IMPORT_TYPES_ELFT(ELFT)

lld/test/ELF/Inputs/aarch64-func3-pac-replace.s

@@ -0,0 +1,10 @@
+// Declare file properties exclusively with aarch64 build attributes.
+
+.aeabi_subsection aeabi_feature_and_bits, optional, uleb128
+.aeabi_attribute Tag_Feature_PAC, 1
+
+.text
+.globl func3
+.type func3,@function
+func3:
+  ret

lld/test/ELF/Inputs/aarch64-pac1-replace.s

@@ -0,0 +1,13 @@
+// This file replace gnu properties with aarch64 build attributes.
+
+.aeabi_subsection aeabi_feature_and_bits, optional, uleb128
+.aeabi_attribute Tag_Feature_PAC, 1
+
+.text
+.globl func2
+.type func2,@function
+func2:
+  .globl func3
+  .type func3, @function
+  bl func3
+  ret

lld/test/ELF/aarch64-build-attributes-be.s

@@ -0,0 +1,50 @@
+// REQUIRES: aarch64
+// RUN: llvm-mc -triple=aarch64_be %s -filetype=obj -o %t.o
+// RUN: ld.lld %t.o --shared -o %t.so
+// RUN: llvm-readelf -n %t.so | FileCheck %s --check-prefix=NOTE
+
+/// The Build attributes section appearing in the output of
+/// llvm-mc should not appear in the output of lld, because
+/// AArch64 build attributes are being transformed into .gnu.properties.
+
+/// Test mc -> big endian, lld -> little endian
+// RUN: llvm-mc -triple=aarch64_be %s -filetype=obj -o %t.o
+// RUN: ld.lld %t.o --shared -o %t.so
+// RUN: llvm-readelf -n %t.so | FileCheck %s --check-prefix=NOTE
+// RUN: ld.lld %t.o -o %t
+// RUN: llvm-readelf -n %t.so | FileCheck %s --check-prefix=NOTE
+// RUN: ld.lld -r %t.o -o %t2.o
+// RUN: llvm-readelf -n %t.so | FileCheck %s --check-prefix=NOTE
+
+/// Test mc -> little endian, lld -> big endian
+// RUN: llvm-mc -triple=aarch64 %s -filetype=obj -o %t.o
+// RUN: ld.lld --EB %t.o --shared -o %t.so
+// RUN: llvm-readelf -n %t.so | FileCheck %s --check-prefix=NOTE
+// RUN: ld.lld --EB %t.o -o %t
+// RUN: llvm-readelf -n %t.so | FileCheck %s --check-prefix=NOTE
+// RUN: ld.lld --EB -r %t.o -o %t2.o
+// RUN: llvm-readelf -n %t.so | FileCheck %s --check-prefix=NOTE
+
+/// Test mc -> big endian, lld -> big endian
+// RUN: llvm-mc -triple=aarch64_be %s -filetype=obj -o %t.o
+// RUN: ld.lld --EB %t.o --shared -o %t.so
+// RUN: llvm-readelf -n %t.so | FileCheck %s --check-prefix=NOTE
+// RUN: ld.lld --EB %t.o -o %t
+// RUN: llvm-readelf -n %t.so | FileCheck %s --check-prefix=NOTE
+// RUN: ld.lld --EB -r %t.o -o %t2.o
+// RUN: llvm-readelf -n %t.so | FileCheck %s --check-prefix=NOTE
+
+// NOTE: Displaying notes found in: .note.gnu.property
+// NOTE-NEXT: Owner Data size Description
+// NOTE-NEXT: GNU 0x00000028 NT_GNU_PROPERTY_TYPE_0 (property note)
+// NOTE-NEXT: Properties: aarch64 feature: BTI, PAC, GCS
+// NOTE-NEXT: AArch64 PAuth ABI core info: platform 0x89abcdef (unknown), version 0x89abcdef
+
+
+.aeabi_subsection aeabi_pauthabi, required, uleb128
+.aeabi_attribute Tag_PAuth_Platform, 81985529216486895
+.aeabi_attribute Tag_PAuth_Schema, 81985529216486895
+.aeabi_subsection aeabi_feature_and_bits, optional, uleb128
+.aeabi_attribute Tag_Feature_BTI, 1
+.aeabi_attribute Tag_Feature_PAC, 1
+.aeabi_attribute Tag_Feature_GCS, 1

lld/test/ELF/aarch64-build-attributes-err.s

@@ -0,0 +1,35 @@
+// REQUIRES: aarch64
+
+// RUN: llvm-mc -triple=aarch64 %s -filetype=obj -o %t.o
+// RUN: not ld.lld %t.o -o /dev/null 2>&1 | FileCheck %s --check-prefix=ERR
+
+// ERR: Pauth Data mismatch: file contains both GNU properties and AArch64 build attributes sections with different Pauth data
+// ERR-NEXT: Features Data mismatch: file contains both GNU properties and AArch64 build attributes sections with different And Features data
+
+.aeabi_subsection aeabi_pauthabi, required, uleb128
+.aeabi_attribute Tag_PAuth_Platform, 5
+.aeabi_attribute Tag_PAuth_Schema, 5
+.aeabi_subsection aeabi_feature_and_bits, optional, uleb128
+.aeabi_attribute Tag_Feature_BTI, 1
+.aeabi_attribute Tag_Feature_PAC, 1
+.aeabi_attribute Tag_Feature_GCS, 1
+
+.section ".note.gnu.property", "a"
+.long 4
+.long 0x10
+.long 0x5
+.asciz "GNU"
+.long 0xc0000000 // GNU_PROPERTY_AARCH64_FEATURE_1_AND
+.long 4
+.long 2          // GNU_PROPERTY_AARCH64_FEATURE_1_PAC
+.long 0
+
+.section ".note.gnu.property", "a"
+.long 4
+.long 24
+.long 5
+.asciz "GNU"
+.long 0xc0000001
+.long 16
+.quad 305419896 // platform
+.quad 2271560481  // version

lld/test/ELF/aarch64-build-attributes-invalid.s

@@ -0,0 +1,18 @@
+// REQUIRES: aarch64
+
+// RUN: llvm-mc -triple=aarch64 -filetype=obj %s -o %t.o
+// RUN: ld.lld -r %t.o -o %t.invalid.o
+// RUN: llvm-readelf -n %t.invalid.o | FileCheck %s
+
+/// According to the BuildAttributes specification Build Attributes
+/// A (TagPlatform, TagSchema)of (0, 1) maps to an explicit PAuth property 
+/// of platform = 0, version = 0 ('Invalid').
+
+// CHECK:      Displaying notes found in: .note.gnu.property
+// CHECK-NEXT:  Owner                Data size 	Description
+// CHECK-NEXT:  GNU                  0x00000018	NT_GNU_PROPERTY_TYPE_0 (property note)
+// CHECK-NEXT:    Properties:        AArch64 PAuth ABI core info: platform 0x0 (invalid), version 0x0
+
+.aeabi_subsection aeabi_pauthabi, required, uleb128
+.aeabi_attribute Tag_PAuth_Platform, 0
+.aeabi_attribute Tag_PAuth_Schema, 1

lld/test/ELF/aarch64-build-attributes-malformed.s

@@ -0,0 +1,16 @@
+# RUN: llvm-mc -triple=aarch64 -filetype=obj %s -o %t.o
+# RUN: ld.lld %t.o /dev/null 2>&1 | FileCheck %s
+
+# CHECK: (.ARM.attributes): unexpected end of data at offset 0x3f while reading [0x3d, 0x41)
+
+.section .ARM.attributes,"",%0x70000003
+.byte 0x41                               // Tag 'A' (format version)
+.long 0x00000019                         // Subsection length
+.asciz "aeabi_pauthabi"                  // Subsection name
+.byte 0x00, 0x00                         // Optionality and Type
+.byte 0x01, 0x01, 0x02, 0x01             // PAuth_Platform and PAuth_Schema
+.long 0x00000023                         // Subsection length
+.asciz "aeabi_feature_and_bits"          // Subsection name
+.byte 0x01, 0x00                         // Optionality and Type
+.byte 0x00, 0x01, 0x01, 0x01, 0x02, 0x01 // BTI, PAC, GCS
+.byte 0x00, 0x00                         // This is the malformation, data is too long.