[dwarfdump] Skip 'stripped' sections

[JDevlieghere]

Summary

When dsymutil generates the companion file, its strips all unnecessary
sections by omitting their body and setting the offset in their
corresponding load command to zero.

One such section is the .eh_frame section, as it contains runtime
information rather than debug information and is part of the __TEXT
segment. When reading this section, we would just read the number of
bytes specified in the load command, starting from offset 0 (i.e. the
beginning of the file).

Rather than trying to parse this obviously invalid section, dwarfdump
now skips this.

PS: The same thing occurs for tools like objdump when trying to dump or
disassemble stripped sections in the __TEXT segment. I will address this
in a separate patch.

[aprantl]

lib/Object/MachOObjectFile.cpp

1931 ↗

(On Diff #116188)

Can you add a big comment that contains the information explaining what/why this happens similar to what you wrote in the introduction of this review?

[JDevlieghere]

• Add comment which explains the zero offset.

[aprantl]

lib/Object/MachOObjectFile.cpp

1932 ↗

(On Diff #116323)

s/its/it/

1932 ↗

(On Diff #116323)

I think it might be better to move the comment into the header so doxygen can pick it up.

llvm/trunk/include/llvm/Object/MachO.h

std::error_code getSectionContents(DataRefImpl Sec,
                                      StringRef &Res) const override;
   uint64_t getSectionAlignment(DataRefImpl Sec) const override;
   bool isSectionCompressed(DataRefImpl Sec) const override;
   bool isSectionText(DataRefImpl Sec) const override;
   bool isSectionData(DataRefImpl Sec) const override;
   bool isSectionBSS(DataRefImpl Sec) const override;
   bool isSectionVirtual(DataRefImpl Sec) const override;
   bool isSectionBitcode(DataRefImpl Sec) const override;
+
+  /// When dsymutil generates the companion file, it strips all unnecessary
+  /// sections (e.g. everything in the _TEXT segment) by omitting their body
+  /// and setting the offset in their corresponding load command to zero.
+  ///
+  /// While the load command itself is valid, reading the section corresponds
+  /// to reading the number of bytes specified in the load command, starting
+  /// from offset 0 (i.e. the Mach-O header at the beginning of the file).
+  bool isSectionStripped(DataRefImpl Sec) const override;
+
   relocation_iterator section_rel_begin(DataRefImpl Sec) const override;
   relocation_iterator section_rel_end(DataRefImpl Sec) const override;
 
   relocation_iterator extrel_begin() const;
   relocation_iterator extrel_end() const;
   iterator_range<relocation_iterator> external_relocations() const {
     return make_range(extrel_begin(), extrel_end());
   }

llvm/trunk/include/llvm/Object/ObjectFile.h

public:
   uint64_t getAlignment() const;
 
   bool isCompressed() const;
   bool isText() const;
   bool isData() const;
   bool isBSS() const;
   bool isVirtual() const;
   bool isBitcode() const;
+  bool isStripped() const;
 
   bool containsSymbol(SymbolRef S) const;
 
   relocation_iterator relocation_begin() const;
   relocation_iterator relocation_end() const;
   iterator_range<relocation_iterator> relocations() const {
     return make_range(relocation_begin(), relocation_end());
   }
protected:
   virtual uint64_t getSectionAlignment(DataRefImpl Sec) const = 0;
   virtual bool isSectionCompressed(DataRefImpl Sec) const = 0;
   virtual bool isSectionText(DataRefImpl Sec) const = 0;
   virtual bool isSectionData(DataRefImpl Sec) const = 0;
   virtual bool isSectionBSS(DataRefImpl Sec) const = 0;
   // A section is 'virtual' if its contents aren't present in the object image.
   virtual bool isSectionVirtual(DataRefImpl Sec) const = 0;
   virtual bool isSectionBitcode(DataRefImpl Sec) const;
+  virtual bool isSectionStripped(DataRefImpl Sec) const;
   virtual relocation_iterator section_rel_begin(DataRefImpl Sec) const = 0;
   virtual relocation_iterator section_rel_end(DataRefImpl Sec) const = 0;
   virtual section_iterator getRelocatedSection(DataRefImpl Sec) const;
 
   // Same as above for RelocationRef.
   friend class RelocationRef;
   virtual void moveRelocationNext(DataRefImpl &Rel) const = 0;
   virtual uint64_t getRelocationOffset(DataRefImpl Rel) const = 0;

 inline bool SectionRef::isVirtual() const {
   return OwningObject->isSectionVirtual(SectionPimpl);
 }
 
 inline bool SectionRef::isBitcode() const {
   return OwningObject->isSectionBitcode(SectionPimpl);
 }
 
+inline bool SectionRef::isStripped() const {
+  return OwningObject->isSectionStripped(SectionPimpl);
+}
+
 inline relocation_iterator SectionRef::relocation_begin() const {
   return OwningObject->section_rel_begin(SectionPimpl);
 }
 
 inline relocation_iterator SectionRef::relocation_end() const {
   return OwningObject->section_rel_end(SectionPimpl);
 }
 

llvm/trunk/lib/DebugInfo/DWARF/DWARFContext.cpp

for (const SectionRef &Section : Obj.sections()) {
       Section.getName(Name);
       ++SectionAmountMap[Name];
       SectionNames.push_back({ Name, true });
 
       // Skip BSS and Virtual sections, they aren't interesting.
       if (Section.isBSS() || Section.isVirtual())
         continue;
 
+      // Skip sections stripped by dsymutil.
+      if (Section.isStripped())
+        continue;
+
       StringRef Data;
       section_iterator RelocatedSection = Section.getRelocatedSection();
       // Try to obtain an already relocated version of this section.
       // Else use the unrelocated section from the object file. We'll have to
       // apply relocations ourselves later.
       if (!L || !L->getLoadedSectionContents(*RelocatedSection, Data))
         Section.getContents(Data);
 

llvm/trunk/lib/Object/MachOObjectFile.cpp

 bool MachOObjectFile::isSectionBitcode(DataRefImpl Sec) const {
   StringRef SegmentName = getSectionFinalSegmentName(Sec);
   StringRef SectName;
   if (!getSectionName(Sec, SectName))
     return (SegmentName == "__LLVM" && SectName == "__bitcode");
   return false;
 }
 
+bool MachOObjectFile::isSectionStripped(DataRefImpl Sec) const {
+  if (is64Bit())
+    return getSection64(Sec).offset == 0;
+  return getSection(Sec).offset == 0;
+}
+
 relocation_iterator MachOObjectFile::section_rel_begin(DataRefImpl Sec) const {
   DataRefImpl Ret;
   Ret.d.a = Sec.d.a;
   Ret.d.b = 0;
   return relocation_iterator(RelocationRef(Ret, this));
 }
 
 relocation_iterator

llvm/trunk/lib/Object/ObjectFile.cpp

 
 bool ObjectFile::isSectionBitcode(DataRefImpl Sec) const {
   StringRef SectName;
   if (!getSectionName(Sec, SectName))
     return SectName == ".llvmbc";
   return false;
 }
 
+bool ObjectFile::isSectionStripped(DataRefImpl Sec) const { return false; }
+
 section_iterator ObjectFile::getRelocatedSection(DataRefImpl Sec) const {
   return section_iterator(SectionRef(Sec, this));
 }
 
 Triple ObjectFile::makeTriple() const {
   Triple TheTriple;
   auto Arch = getArch();
   TheTriple.setArch(Triple::ArchType(Arch));

llvm/trunk/test/tools/llvm-dwarfdump/X86/stripped.test

+RUN: llvm-dwarfdump -debug-frame %S/Inputs/empty.dSYM | FileCheck %s
+
+You can generate an empty dSYM companion file by invoking dsymutil on any
+object file.
+
+CHECK: .debug_frame contents:
+CHECK-NOT: CIE
+CHECK-NOT: FDE
+CHECK: .eh_frame contents:
+CHECK-NOT: CIE
+CHECK-NOT: FDE