[llvm][dsymutil] Use the DW_AT_name of the uniqued DIE for insertion into .debug_names

The root cause of the issue is that we have a DW_AT_subprogram definition whose DW_AT_specification DIE got deduplicated. But the DW_AT_name of the original specification is different than the one it got uniqued to. That’s technically fine because dsymutil uniques by linkage name, which uniquely identifies any function with non-internal linkage.

But we insert the definition DIE into the debug-names table using the DW_AT_name of the original specification (we call getDIENames(InputDIE…)). But what we really want to do is use the name of the adjusted DW_AT_specifcation (i.e., the DW_AT_specification of the output DIE). That’s not as simple as it sounds because we can’t just get ahold of the DIE in the output CU. We have to grab the ODR DeclContext of the input DIE’s specification. That is the only link back to the canonical specification DIE. For that to be of any use, we have to stash the DW_AT_name into DeclContext so we can use it in getDIENames.

We have to account for the possibility of multiple levels of DW_AT_specification/DW_AT_abstract_origin. So my proposed solution is to recursively scan the referenced DIE’s, grab the canonical DIE for those and get the name from the DeclContext (if none exists then use the DW_AT_name of the DIE itself).

Author: Michael137

llvm/include/llvm/DWARFLinker/Classic/DWARFLinker.h

@@ -708,7 +708,11 @@ class LLVM_ABI DWARFLinker : public DWARFLinkerBase {
     /// already there.
     /// \returns is a name was found.
     bool getDIENames(const DWARFDie &Die, AttributesInfo &Info,
-                     OffsetsStringPool &StringPool, bool StripTemplate = false);
+                     OffsetsStringPool &StringPool, const DWARFFile &File,
+                     CompileUnit &Unit, bool StripTemplate = false);
+
+    llvm::StringRef getCanonicalDIEName(DWARFDie Die, const DWARFFile &File,
+                                        CompileUnit *Unit);
 
     uint32_t hashFullyQualifiedName(DWARFDie DIE, CompileUnit &U,
                                     const DWARFFile &File,

llvm/include/llvm/DWARFLinker/Classic/DWARFLinkerDeclContext.h

@@ -84,12 +84,13 @@ class DeclContext {
   DeclContext() : DefinedInClangModule(0), Parent(*this) {}
 
   DeclContext(unsigned Hash, uint32_t Line, uint32_t ByteSize, uint16_t Tag,
-              StringRef NameForUniquing, StringRef File,
+              StringRef Name, StringRef NameForUniquing, StringRef File,
               const DeclContext &Parent, DWARFDie LastSeenDIE = DWARFDie(),
               unsigned CUId = 0)
       : QualifiedNameHash(Hash), Line(Line), ByteSize(ByteSize), Tag(Tag),
-        DefinedInClangModule(0), NameForUniquing(NameForUniquing), File(File),
-        Parent(Parent), LastSeenDIE(LastSeenDIE), LastSeenCompileUnitID(CUId) {}
+        DefinedInClangModule(0), Name(Name), NameForUniquing(NameForUniquing),
+        File(File), Parent(Parent), LastSeenDIE(LastSeenDIE),
+        LastSeenCompileUnitID(CUId) {}
 
   uint32_t getQualifiedNameHash() const { return QualifiedNameHash; }
 
@@ -101,6 +102,7 @@ class DeclContext {
 
   uint32_t getCanonicalDIEOffset() const { return CanonicalDIEOffset; }
   void setCanonicalDIEOffset(uint32_t Offset) { CanonicalDIEOffset = Offset; }
+  llvm::StringRef getCanonicalName() const { return Name; }
 
   bool isDefinedInClangModule() const { return DefinedInClangModule; }
   void setDefinedInClangModule(bool Val) { DefinedInClangModule = Val; }
@@ -115,6 +117,7 @@ class DeclContext {
   uint32_t ByteSize = 0;
   uint16_t Tag = dwarf::DW_TAG_compile_unit;
   unsigned DefinedInClangModule : 1;
+  StringRef Name;
   StringRef NameForUniquing;
   StringRef File;
   const DeclContext &Parent;

llvm/lib/DWARFLinker/Classic/DWARFLinker.cpp

@@ -151,22 +151,84 @@ static bool isTypeTag(uint16_t Tag) {
   return false;
 }
 
-bool DWARFLinker::DIECloner::getDIENames(const DWARFDie &Die,
-                                         AttributesInfo &Info,
-                                         OffsetsStringPool &StringPool,
-                                         bool StripTemplate) {
+/// Recurse through the input DIE's canonical references until we find a
+/// DW_AT_name.
+llvm::StringRef
+DWARFLinker::DIECloner::getCanonicalDIEName(DWARFDie Die, const DWARFFile &File,
+                                            CompileUnit *Unit) {
+  if (!Die)
+    return {};
+
+  std::optional<DWARFFormValue> Ref;
+
+  auto GetDieName = [](const DWARFDie &D) -> llvm::StringRef {
+    auto NameForm = D.find(llvm::dwarf::DW_AT_name);
+    if (!NameForm)
+      return {};
+
+    auto NameOrErr = NameForm->getAsCString();
+    if (!NameOrErr) {
+      llvm::consumeError(NameOrErr.takeError());
+      return {};
+    }
+
+    return *NameOrErr;
+  };
+
+  llvm::StringRef Name = GetDieName(Die);
+  if (!Name.empty())
+    return Name;
+
+  while (true) {
+    if (!(Ref = Die.find(llvm::dwarf::DW_AT_specification)) &&
+        !(Ref = Die.find(llvm::dwarf::DW_AT_abstract_origin)))
+      break;
+
+    Die = Linker.resolveDIEReference(File, CompileUnits, *Ref, Die, Unit);
+    if (!Die)
+      break;
+
+    assert(Unit);
+
+    unsigned SpecIdx = Unit->getOrigUnit().getDIEIndex(Die);
+    CompileUnit::DIEInfo &SpecInfo = Unit->getInfo(SpecIdx);
+    if (SpecInfo.Ctxt && SpecInfo.Ctxt->hasCanonicalDIE()) {
+      if (!SpecInfo.Ctxt->getCanonicalName().empty()) {
+        Name = SpecInfo.Ctxt->getCanonicalName();
+        break;
+      }
+    }
+
+    Name = GetDieName(Die);
+    if (!Name.empty())
+      break;
+  }
+
+  return Name;
+}
+
+bool DWARFLinker::DIECloner::getDIENames(
+    const DWARFDie &Die, AttributesInfo &Info, OffsetsStringPool &StringPool,
+    const DWARFFile &File, CompileUnit &Unit, bool StripTemplate) {
   // This function will be called on DIEs having low_pcs and
   // ranges. As getting the name might be more expansive, filter out
   // blocks directly.
   if (Die.getTag() == dwarf::DW_TAG_lexical_block)
     return false;
 
+  // The mangled name of an specification DIE will by virtue of the
+  // uniquing algorithm be the same as the one it got uniqued into.
+  // So just use the input DIE's linkage name.
   if (!Info.MangledName)
     if (const char *MangledName = Die.getLinkageName())
       Info.MangledName = StringPool.getEntry(MangledName);
 
+  // For subprograms with linkage names, we unique on the linkage name,
+  // so DW_AT_name's may differ between the input and canonical DIEs.
+  // Use the name of the canonical DIE.
   if (!Info.Name)
-    if (const char *Name = Die.getShortName())
+    if (llvm::StringRef Name = getCanonicalDIEName(Die, File, &Unit);
+        !Name.empty())
       Info.Name = StringPool.getEntry(Name);
 
   if (!Info.MangledName)
@@ -1939,7 +2001,7 @@ DIE *DWARFLinker::DIECloner::cloneDIE(const DWARFDie &InputDIE,
   // accelerator tables too. For now stick with dsymutil's behavior.
   if ((Info.InDebugMap || AttrInfo.HasLowPc || AttrInfo.HasRanges) &&
       Tag != dwarf::DW_TAG_compile_unit &&
-      getDIENames(InputDIE, AttrInfo, DebugStrPool,
+      getDIENames(InputDIE, AttrInfo, DebugStrPool, File, Unit,
                   Tag != dwarf::DW_TAG_inlined_subroutine)) {
     if (AttrInfo.MangledName && AttrInfo.MangledName != AttrInfo.Name)
       Unit.addNameAccelerator(Die, AttrInfo.MangledName,
@@ -1962,7 +2024,7 @@ DIE *DWARFLinker::DIECloner::cloneDIE(const DWARFDie &InputDIE,
   } else if (Tag == dwarf::DW_TAG_imported_declaration && AttrInfo.Name) {
     Unit.addNamespaceAccelerator(Die, AttrInfo.Name);
   } else if (isTypeTag(Tag) && !AttrInfo.IsDeclaration) {
-    bool Success = getDIENames(InputDIE, AttrInfo, DebugStrPool);
+    bool Success = getDIENames(InputDIE, AttrInfo, DebugStrPool, File, Unit);
     uint64_t RuntimeLang =
         dwarf::toUnsigned(InputDIE.find(dwarf::DW_AT_APPLE_runtime_class))
             .value_or(0);

llvm/lib/DWARFLinker/Classic/DWARFLinkerDeclContext.cpp

@@ -84,13 +84,14 @@ DeclContextTree::getChildDeclContext(DeclContext &Context, const DWARFDie &DIE,
     break;
   }
 
+  StringRef Name = DIE.getShortName();
   StringRef NameForUniquing;
   StringRef FileRef;
 
   if (const char *LinkageName = DIE.getLinkageName())
     NameForUniquing = StringPool.internString(LinkageName);
-  else if (const char *ShortName = DIE.getShortName())
-    NameForUniquing = StringPool.internString(ShortName);
+  else if (!Name.empty())
+    NameForUniquing = StringPool.internString(Name);
 
   bool IsAnonymousNamespace =
       NameForUniquing.empty() && Tag == dwarf::DW_TAG_namespace;
@@ -163,15 +164,16 @@ DeclContextTree::getChildDeclContext(DeclContext &Context, const DWARFDie &DIE,
     Hash = hash_combine(Hash, FileRef);
 
   // Now look if this context already exists.
-  DeclContext Key(Hash, Line, ByteSize, Tag, NameForUniquing, FileRef, Context);
+  DeclContext Key(Hash, Line, ByteSize, Tag, Name, NameForUniquing, FileRef,
+                  Context);
   auto ContextIter = Contexts.find(&Key);
 
   if (ContextIter == Contexts.end()) {
     // The context wasn't found.
     bool Inserted;
-    DeclContext *NewContext =
-        new (Allocator) DeclContext(Hash, Line, ByteSize, Tag, NameForUniquing,
-                                    FileRef, Context, DIE, U.getUniqueID());
+    DeclContext *NewContext = new (Allocator)
+        DeclContext(Hash, Line, ByteSize, Tag, Name, NameForUniquing, FileRef,
+                    Context, DIE, U.getUniqueID());
     std::tie(ContextIter, Inserted) = Contexts.insert(NewContext);
     assert(Inserted && "Failed to insert DeclContext");
     (void)Inserted;

llvm/test/tools/dsymutil/AArch64/dummy-debug-map-arm64.map

@@ -11,9 +11,13 @@ objects:
   - filename: 1.o
     symbols:
       - { sym: _bar, objAddr: 0x0, binAddr: 0x10000, size: 0x10 }
+      - { sym: __Z13lib1_internalv, objAddr: 0x0, binAddr: 0x10020, size: 0x20 }
+      - { sym: __ZN3Foo4funcIZ13lib1_internalvE3$_0EEvv, objAddr: 0x0, binAddr: 0x10040, size: 0x20 }
   - filename: 2.o
     symbols:
       - { sym: __Z3foov, objAddr: 0x0, binAddr: 0x20000, size: 0x10 }
+      - { sym: __Z13lib1_internalv, objAddr: 0x0, binAddr: 0x20020, size: 0x20 }
+      - { sym: __ZN3Foo4funcIZ13lib1_internalvE3$_0EEvv, objAddr: 0x0, binAddr: 0x20040, size: 0x20 }
   - filename: 3.o
     symbols:
       - { sym: __Z3foov, objAddr: 0x0, binAddr: 0x30000, size: 0x10 }

llvm/test/tools/dsymutil/AArch64/odr-uniquing-DW_AT_name-conflict.test

@@ -0,0 +1,28 @@
+# Tests the case where a DW_TAG_subprogram for a method declaration
+# got uniqued into a DW_TAG_subprogram with the same linkage name (but
+# different DW_AT_name). Make sure the DW_TAG_subprogram DIE for the
+# definition, which previously pointed to the now de-deduplicated declaration,
+# gets inserted into the .debug_names table using the DW_AT_name of the canonical
+# declaration DW_TAG_subprogram.
+#
+# Object files compiled as follows:
+#   clang -g -c -o 1.o Inputs/odr-uniquing-DW_AT_name-conflict/lib1.cpp
+#   clang -g -c -o 2.o Inputs/odr-uniquing-DW_AT_name-conflict/lib2.cpp
+
+# RUN: dsymutil -f -oso-prepend-path=%p/../Inputs/odr-uniquing-DW_AT_name-conflict -y %p/dummy-debug-map-arm64.map -o - \
+# RUN:     | llvm-dwarfdump --verify - | FileCheck %s
+
+# RUN: dsymutil --linker parallel -f -oso-prepend-path=%p/../Inputs/odr-uniquing-DW_AT_name-conflict -y %p/dummy-debug-map-arm64.map -o - \
+# RUN:     | not llvm-dwarfdump --verify - | FileCheck %s --check-prefix=PARALLEL-ODR
+
+# RUN: dsymutil -f -oso-prepend-path=%p/../Inputs/odr-uniquing-DW_AT_name-conflict -y %p/dummy-debug-map-arm64.map -no-odr -o - \
+# RUN:     | llvm-dwarfdump --verify - | FileCheck %s
+
+# RUN: dsymutil --linker parallel -f -oso-prepend-path=%p/../Inputs/odr-uniquing-DW_AT_name-conflict -y %p/dummy-debug-map-arm64.map -no-odr -o - \
+# RUN:     | llvm-dwarfdump --verify - | FileCheck %s
+
+# CHECK: No errors.
+
+# FIXME: parallel DWARFLinker uses wrong DW_AT_name when inserting uniqued subprogram into .debug_names
+# PARALLEL-ODR:      Verifying .debug_names...
+# PARALLEL-ODR-NEXT: error: Name Index {{.*}} mismatched Name of DIE

llvm/test/tools/dsymutil/Inputs/odr-uniquing-DW_AT_name-conflict/1.o

@@ -0,0 +1,5 @@
+#include "lib1.h"
+
+[[gnu::weak]] void lib1_internal() {
+  Foo{}.func<decltype([]{})>();
+}

llvm/test/tools/dsymutil/Inputs/odr-uniquing-DW_AT_name-conflict/2.o

@@ -0,0 +1,3 @@
+struct Foo {
+  template<typename T> void func() {}
+};