[lld][AArch64][Build Attributes] Add support for AArch64 Build Attributes (#147970)

This patch enables lld to read AArch64 Build Attributes and convert them
into GNU Properties.

Changes:
    - Parses AArch64 Build Attributes from input object files.
    - Converts known attributes into corresponding GNU Properties.
    - Merges attributes when linking multiple objects.

Spec reference:
    https://github.com/ARM-software/abi-aa/pull/230/files#r1030

Co-authored-by: Sivan Shani <sivan.shani@arm.com>

---------

Co-authored-by: Sivan Shani <sivan.shani@arm.com>

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 +538,41 @@ uint32_t ObjFile<ELFT>::getSectionIndex(const Elf_Sym &sym) const {
       this);
 }
 
+template <class ELFT>
+static void
+handleAArch64BAAndGnuProperties(ObjFile<ELFT> *file, Ctx &ctx,
+                                const AArch64BuildAttrSubsections &baInfo) {
+  if (file->aarch64PauthAbiCoreInfo) {
+    // Check for data mismatch.
+    if (file->aarch64PauthAbiCoreInfo) {
+      if (baInfo.Pauth.TagPlatform != file->aarch64PauthAbiCoreInfo->platform ||
+          baInfo.Pauth.TagSchema != file->aarch64PauthAbiCoreInfo->version)
+        Err(ctx) << file
+                 << " GNU properties and build attributes have conflicting "
+                    "AArch64 PAuth data";
+    }
+    if (baInfo.AndFeatures != file->andFeatures)
+      Err(ctx) << file
+               << " GNU properties and build attributes have conflicting "
+                  "AArch64 PAuth data";
+  } else {
+    // When BuildAttributes are missing, PauthABI value defaults to (TagPlatform
+    // = 0, TagSchema = 0). GNU properties do not write PAuthAbiCoreInfo if GNU
+    // property is not present. To match this behaviour, we only write
+    // PAuthAbiCoreInfo when there is at least one non-zero value. The
+    // specification reserves TagPlatform = 0, TagSchema = 1 values to match the
+    // 'Invalid' GNU property section with platform = 0, version = 0.
+    if (baInfo.Pauth.TagPlatform || baInfo.Pauth.TagSchema) {
+      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 <class ELFT> void ObjFile<ELFT>::parse(bool ignoreComdats) {
   object::ELFFile<ELFT> obj = this->getObj();
   // Read a section table. justSymbols is usually false.
@@ -554,6 +590,7 @@ template <class ELFT> void ObjFile<ELFT>::parse(bool ignoreComdats) {
   sections.resize(size);
   for (size_t i = 0; i != size; ++i) {
     const Elf_Shdr &sec = objSections[i];
+
     if (LLVM_LIKELY(sec.sh_type == SHT_PROGBITS))
       continue;
     if (LLVM_LIKELY(sec.sh_type == SHT_GROUP)) {
@@ -637,13 +674,6 @@ 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)
-        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
@@ -744,6 +774,8 @@ void ObjFile<ELFT>::initializeSections(bool ignoreComdats,
   StringRef shstrtab = CHECK2(obj.getSectionStringTable(objSections), this);
   uint64_t size = objSections.size();
   SmallVector<ArrayRef<Elf_Word>, 0> selectedGroups;
+  AArch64BuildAttrSubsections aarch64BAsubSections;
+  bool hasAArch64BuildAttributes = false;
   for (size_t i = 0; i != size; ++i) {
     if (this->sections[i] == &InputSection::discarded)
       continue;
@@ -775,6 +807,26 @@ void ObjFile<ELFT>::initializeSections(bool ignoreComdats,
       continue;
     }
 
+    // Processor-specific types that do not use the following switch statement.
+    //
+    // Extract Build Attributes section contents into aarch64BAsubSections.
+    // Input objects may contain both build Build Attributes and GNU
+    // properties. We delay processing Build Attributes until we have finished
+    // reading all sections so that we can check that these are consistent.
+    if (type == SHT_AARCH64_ATTRIBUTES && ctx.arg.emachine == EM_AARCH64) {
+      ArrayRef<uint8_t> contents = check(obj.getSectionContents(sec));
+      AArch64AttributeParser attributes;
+      if (Error e = attributes.parse(contents, ELFT::Endianness)) {
+        StringRef name = check(obj.getSectionName(sec, shstrtab));
+        InputSection isec(*this, sec, name);
+        Warn(ctx) << &isec << ": " << std::move(e);
+      } else {
+        aarch64BAsubSections = extractBuildAttributesSubsections(attributes);
+        hasAArch64BuildAttributes = true;
+      }
+      this->sections[i] = &InputSection::discarded;
+      continue;
+    }
     switch (type) {
     case SHT_GROUP: {
       if (!ctx.arg.relocatable)
@@ -912,6 +964,12 @@ void ObjFile<ELFT>::initializeSections(bool ignoreComdats,
           << linkSec;
   }
 
+  // Handle AArch64 Build Attributes and GNU properties:
+  // - Err on mismatched values.
+  // - Store missing values as GNU properties.
+  if (hasAArch64BuildAttributes)
+    handleAArch64BAAndGnuProperties<ELFT>(this, ctx, aarch64BAsubSections);
+
   for (ArrayRef<Elf_Word> entries : selectedGroups)
     handleSectionGroup<ELFT>(this->sections, entries);
 }

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

@@ -0,0 +1,29 @@
+// 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
+
+// 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 that lld can read big-endian build-attributes.
+
+// 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, 0x123456789ABCDEF
+.aeabi_attribute Tag_PAuth_Schema, 0x123456789ABCDEF
+.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: GNU properties and build attributes have conflicting AArch64 PAuth data
+// ERR-NEXT: GNU properties and build attributes have conflicting AArch64 PAuth 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 0x4
+.long 0x10
+.long 0x5
+.asciz "GNU"
+.long 0xc0000000 // GNU_PROPERTY_AARCH64_FEATURE_1_AND
+.long 0x4
+.long 0x2          // GNU_PROPERTY_AARCH64_FEATURE_1_PAC
+.long 0x0
+
+.section ".note.gnu.property", "a"
+.long 0x4
+.long 0x18
+.long 0x5
+.asciz "GNU"
+.long 0xc0000001
+.long 0x10
+.quad 0x12345678 // platform
+.quad 0x87654321  // 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 +1,18 @@
+# REQUIRES: aarch64
+
+# 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.

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

@@ -0,0 +1,67 @@
+// REQUIRES: aarch64
+
+// RUN: rm -rf %t && split-file %s %t && cd %t
+
+// RUN: llvm-mc -triple=aarch64 -filetype=obj %s -o %t11.o
+// RUN: llvm-mc -triple=aarch64 -filetype=obj merged-property.s -o %t12.o
+// RUN: llvm-mc -triple=aarch64 -filetype=obj merged-property2.s -o %t13.o
+// RUN: ld.lld -r %t11.o %t12.o %t13.o -o %t.merged1.o
+// RUN: llvm-readelf -n %t.merged1.o | FileCheck %s --check-prefix=NOTE-MIXED
+
+/// This test verifies merging of AArch64 build attributes and GNU property notes.
+/// Three object files are combined: one with build attributes (PAuth information, BTI, PAC, GCS),
+/// and two with GNU property notes encoding the same feature bits. 
+/// PAuth ABI info is provided in one of the files and it is expected to be preserved in the merged output.
+
+// NOTE-MIXED: Displaying notes found in: .note.gnu.property
+// NOTE-MIXED-NEXT:   Owner                Data size 	Description
+// NOTE-MIXED-NEXT:   GNU                  0x00000028	NT_GNU_PROPERTY_TYPE_0 (property note)
+// NOTE-MIXED-NEXT:     Properties:    aarch64 feature: BTI, PAC
+// NOTE-MIXED-NEXT:         AArch64 PAuth ABI core info: platform 0x31 (unknown), version 0x13
+
+// CHECK: .note.gnu.property
+// CHECK-NOT: .ARM.attributes
+
+.aeabi_subsection aeabi_pauthabi, required, uleb128
+.aeabi_attribute Tag_PAuth_Platform, 49
+.aeabi_attribute Tag_PAuth_Schema, 19
+.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
+
+
+//--- merged-property.s
+.section ".note.gnu.property", "a"
+  .long 0x4           // Name length is always 4 ("GNU")
+  .long end - begin   // Data length
+  .long 0x5           // Type: NT_GNU_PROPERTY_TYPE_0
+  .asciz "GNU"        // Name
+  .p2align 0x3
+begin:
+  .long 0xc0000000    // GNU_PROPERTY_AARCH64_FEATURE_1_AND
+  .long 0x4
+  .long 0x7           // pr_data: BTI (1), PAC (2), GCS (4) = 0b111 = 7
+  .long 0x0
+  // PAuth ABI property note
+  .long 0xc0000001    // GNU_PROPERTY_AARCH64_FEATURE_PAUTH
+  .long 0x10          // Data length
+  .quad 0x31          // PAuth ABI platform
+  .quad 0x13          // PAuth ABI version
+  .p2align 0x3        // Align to 8 byte for 64 bit
+end:
+
+//--- merged-property2.s
+.section .note.gnu.property, "a"
+  .align 0x4
+  .long 0x4           // Name length is always 4 ("GNU")
+  .long end2 - begin2 // Data length
+  .long 0x5           // Type: NT_GNU_PROPERTY_TYPE_0
+  .asciz "GNU"        // Name
+begin2:
+  .align 0x4
+  .long 0xc0000000    // Type: GNU_PROPERTY_AARCH64_FEATURE_1_AND
+  .long 0x4           // Data length
+  .long 0x7           // pr_data: BTI (1), PAC (2), GCS (4) = 0b111 = 7
+  .long 0x0
+end2:

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

@@ -1,26 +1,50 @@
 // REQUIRES: aarch64
-// RUN: llvm-mc -triple=aarch64 %s -filetype=obj -o %t.o
+// RUN: rm -rf %t && split-file %s %t && cd %t
-// RUN: ld.lld %t.o --shared -o %t.so
-// RUN: llvm-readelf --sections %t.so | FileCheck %s
-// RUN: ld.lld %t.o -o %t
-// RUN: llvm-readelf --sections %t | FileCheck %s
-// RUN: ld.lld -r %t.o -o %t2.o
-// RUN: llvm-readelf --sections %t2.o | FileCheck %s
 
-/// File has a Build attributes section. This should not appear in
+// RUN: llvm-mc -triple=aarch64 -filetype=obj %s -o %t1.o
-/// ET_EXEC or ET_SHARED files as there is no requirement for it to
+// RUN: llvm-mc -triple=aarch64 -filetype=obj pauth-bti-gcs.s -o %t2.o
-/// do so. FIXME, the ld -r (relocatable link) should output a single
+// RUN: llvm-mc -triple=aarch64 -filetype=obj pauth-bti-pac.s -o %t3.o
-/// merged build attributes section. When full support is added in
+// RUN: ld.lld -r %t1.o %t2.o %t3.o -o %t.merged.o
-/// ld.lld this test should be updated.
+// RUN: llvm-readelf -n %t.merged.o | FileCheck %s --check-prefix=NOTE
 
+/// This test merges three object files with AArch64 build attributes.
+/// All contain identical PAuth ABI info (platform/version), which must be preserved.
+/// Only BTI is common across all three in the AND feature set, so the merged output
+/// must show BTI only. PAC and GCS are present in subsets and should not appear.
+
+// 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
+// NOTE-NEXT:        AArch64 PAuth ABI core info: platform 0x31 (unknown), version 0x13
+
+// CHECK: .note.gnu.property
 // CHECK-NOT: .ARM.attributes
 
+.aeabi_subsection aeabi_pauthabi, required, uleb128
+.aeabi_attribute Tag_PAuth_Platform, 49
+.aeabi_attribute Tag_PAuth_Schema, 19
 .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
 
-.global _start
+
-.type _start, %function
+//--- pauth-bti-gcs.s
-_start:
+.aeabi_subsection aeabi_pauthabi, required, uleb128
-ret
+.aeabi_attribute Tag_PAuth_Platform, 49
+.aeabi_attribute Tag_PAuth_Schema, 19
+.aeabi_subsection aeabi_feature_and_bits, optional, uleb128
+.aeabi_attribute Tag_Feature_BTI, 1
+.aeabi_attribute Tag_Feature_PAC, 0
+.aeabi_attribute Tag_Feature_GCS, 1
+
+
+//--- pauth-bti-pac.s
+.aeabi_subsection aeabi_pauthabi, required, uleb128
+.aeabi_attribute Tag_PAuth_Platform, 49
+.aeabi_attribute Tag_PAuth_Schema, 19
+.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, 0

llvm/include/llvm/Support/AArch64AttributeParser.h

@@ -25,6 +25,17 @@ public:
       : ELFExtendedAttrParser(nullptr, returnTagsNamesMap()) {}
 };
 
+// Used for extracting AArch64 Build Attributes
+struct AArch64BuildAttrSubsections {
+  struct PauthSubSection {
+    uint64_t TagPlatform = 0;
+    uint64_t TagSchema = 0;
+  } Pauth;
+  uint32_t AndFeatures = 0;
+};
+
+AArch64BuildAttrSubsections
+extractBuildAttributesSubsections(const llvm::AArch64AttributeParser &);
 } // namespace llvm
 
 #endif // LLVM_SUPPORT_AARCH64ATTRIBUTEPARSER_H

llvm/lib/Support/AArch64AttributeParser.cpp

@@ -8,6 +8,7 @@
 //===---------------------------------------------------------------------===//
 
 #include "llvm/Support/AArch64AttributeParser.h"
+#include "llvm/Support/AArch64BuildAttributes.h"
 
 std::vector<llvm::SubsectionAndTagToTagName> &
 llvm::AArch64AttributeParser::returnTagsNamesMap() {
@@ -19,3 +20,29 @@ llvm::AArch64AttributeParser::returnTagsNamesMap() {
       {"aeabi_feature_and_bits", 2, "Tag_Feature_GCS"}};
   return TagsNamesMap;
 }
+
+llvm::AArch64BuildAttrSubsections llvm::extractBuildAttributesSubsections(
+    const llvm::AArch64AttributeParser &Attributes) {
+
+  llvm::AArch64BuildAttrSubsections SubSections;
+  auto GetPauthValue = [&Attributes](unsigned Tag) {
+    return Attributes.getAttributeValue("aeabi_pauthabi", Tag).value_or(0);
+  };
+  SubSections.Pauth.TagPlatform =
+      GetPauthValue(llvm::AArch64BuildAttributes::TAG_PAUTH_PLATFORM);
+  SubSections.Pauth.TagSchema =
+      GetPauthValue(llvm::AArch64BuildAttributes::TAG_PAUTH_SCHEMA);
+
+  auto GetFeatureValue = [&Attributes](unsigned Tag) {
+    return Attributes.getAttributeValue("aeabi_feature_and_bits", Tag)
+        .value_or(0);
+  };
+  SubSections.AndFeatures |=
+      GetFeatureValue(llvm::AArch64BuildAttributes::TAG_FEATURE_BTI);
+  SubSections.AndFeatures |=
+      GetFeatureValue(llvm::AArch64BuildAttributes::TAG_FEATURE_PAC) << 1;
+  SubSections.AndFeatures |=
+      GetFeatureValue(llvm::AArch64BuildAttributes::TAG_FEATURE_GCS) << 2;
+
+  return SubSections;
+}