MC: Introduce an ELF dwo object writer and teach llvm-mc about it.

[pcc]

Summary

Part of PR37466.

Depends on D47049

Depends on D47050

[echristo]

And it even comes with testcases. :)

LGTM.

-eric

llvm/lib/MC/ELFObjectWriter.cpp

165

It feels a little awkward to special case this for dwo files rather than just be a general object emitter. That said, generalization should be possible in the future.

[pcc]

llvm/lib/MC/ELFObjectWriter.cpp

165

That's true, but right now it feels premature to be generalizing this code, as there aren't enough users yet. Probably something to think about if/when we add another user.

llvm/include/llvm/MC/MCAsmBackend.h

public:
   /// lifetime management
   virtual void reset() {}
 
   /// Create a new MCObjectWriter instance for use by the assembler backend to
   /// emit the final object file.
   std::unique_ptr<MCObjectWriter>
   createObjectWriter(raw_pwrite_stream &OS) const;
 
+  /// Create an MCObjectWriter that writes two object files: a .o file which is
+  /// linked into the final program and a .dwo file which is used by debuggers.
+  /// This function is only supported with ELF targets.
+  std::unique_ptr<MCObjectWriter>
+  createDwoObjectWriter(raw_pwrite_stream &OS, raw_pwrite_stream &DwoOS) const;
+
   virtual std::unique_ptr<MCObjectTargetWriter>
   createObjectTargetWriter() const = 0;
 
   /// \name Target Fixup Interfaces
   /// @{
 
   /// Get the number of target specific fixup kinds.
   virtual unsigned getNumFixupKinds() const = 0;

llvm/include/llvm/MC/MCELFObjectWriter.h

 ///
 /// \param MOTW - The target specific ELF writer subclass.
 /// \param OS - The stream to write to.
 /// \returns The constructed object writer.
 std::unique_ptr<MCObjectWriter>
 createELFObjectWriter(std::unique_ptr<MCELFObjectTargetWriter> MOTW,
                       raw_pwrite_stream &OS, bool IsLittleEndian);
 
+std::unique_ptr<MCObjectWriter>
+createELFDwoObjectWriter(std::unique_ptr<MCELFObjectTargetWriter> MOTW,
+                         raw_pwrite_stream &OS, raw_pwrite_stream &DwoOS,
+                         bool IsLittleEndian);
+
 } // end namespace llvm
 
 #endif // LLVM_MC_MCELFOBJECTWRITER_H

llvm/lib/MC/ELFObjectWriter.cpp

 
 namespace {
 
 using SectionIndexMapTy = DenseMap<const MCSectionELF *, uint32_t>;
 
 class ELFObjectWriter;
 struct ELFWriter;
 
+bool isDwoSection(const MCSectionELF &Sec) {
+  return Sec.getSectionName().endswith(".dwo");
+}
+
 class SymbolTableWriter {
   ELFWriter &EWriter;
   bool Is64Bit;
 
   // indexes we are going to write to .symtab_shndx.
   std::vector<uint32_t> ShndxIndexes;
 
   // The numbel of symbols written so far.
public:
 
   ArrayRef<uint32_t> getShndxIndexes() const { return ShndxIndexes; }
 };
 
 struct ELFWriter {
   ELFObjectWriter &OWriter;
   support::endian::Writer W;
 
+  enum DwoMode {
+    AllSections,
+    NonDwoOnly,
+    DwoOnly,
+  } Mode;
+
   static uint64_t SymbolValue(const MCSymbol &Sym, const MCAsmLayout &Layout);
   static bool isInSymtab(const MCAsmLayout &Layout, const MCSymbolELF &Symbol,
                          bool Used, bool Renamed);
 
   /// Helper struct for containing some precomputed information on symbols.
   struct ELFSymbolData {
     const MCSymbolELF *Symbol;
     uint32_t SectionIndex;
struct ELFWriter {
   void align(unsigned Alignment);
 
   bool maybeWriteCompression(uint64_t Size,
                              SmallVectorImpl<char> &CompressedContents,
                              bool ZLibStyle, unsigned Alignment);
 
 public:
   ELFWriter(ELFObjectWriter &OWriter, raw_pwrite_stream &OS,
-            bool IsLittleEndian)
+            bool IsLittleEndian, DwoMode Mode)
       : OWriter(OWriter),
-        W(OS, IsLittleEndian ? support::little : support::big) {}
+        W(OS, IsLittleEndian ? support::little : support::big), Mode(Mode) {}
 
   void WriteWord(uint64_t Word) {
     if (is64Bit())
       W.write<uint64_t>(Word);
     else
       W.write<uint32_t>(Word);
   }
 
void reset() override {
     MCObjectWriter::reset();
   }
 
   bool isSymbolRefDifferenceFullyResolvedImpl(const MCAssembler &Asm,
                                               const MCSymbol &SymA,
                                               const MCFragment &FB, bool InSet,
                                               bool IsPCRel) const override;
 
+  virtual bool checkRelocation(MCContext &Ctx, SMLoc Loc,
+                               const MCSectionELF *From,
+                               const MCSectionELF *To) {
+    return true;
+  }
+
   void recordRelocation(MCAssembler &Asm, const MCAsmLayout &Layout,
                         const MCFragment *Fragment, const MCFixup &Fixup,
                         MCValue Target, uint64_t &FixedValue) override;
 
   void executePostLayoutBinding(MCAssembler &Asm,
                                 const MCAsmLayout &Layout) override;
 
   friend struct ELFWriter;
 };
 
 class ELFSingleObjectWriter : public ELFObjectWriter {
   raw_pwrite_stream &OS;
   bool IsLittleEndian;
 
 public:
   ELFSingleObjectWriter(std::unique_ptr<MCELFObjectTargetWriter> MOTW,
                         raw_pwrite_stream &OS, bool IsLittleEndian)
       : ELFObjectWriter(std::move(MOTW)), OS(OS),
         IsLittleEndian(IsLittleEndian) {}
 
   uint64_t writeObject(MCAssembler &Asm, const MCAsmLayout &Layout) override {
-    return ELFWriter(*this, OS, IsLittleEndian).writeObject(Asm, Layout);
+    return ELFWriter(*this, OS, IsLittleEndian, ELFWriter::AllSections)
+        .writeObject(Asm, Layout);
+  }
+
+  friend struct ELFWriter;
+};
+
+class ELFDwoObjectWriter : public ELFObjectWriter {
+  raw_pwrite_stream &OS, &DwoOS;
+  bool IsLittleEndian;
+
+public:
+  ELFDwoObjectWriter(std::unique_ptr<MCELFObjectTargetWriter> MOTW,
+                     raw_pwrite_stream &OS, raw_pwrite_stream &DwoOS,
+                     bool IsLittleEndian)
+      : ELFObjectWriter(std::move(MOTW)), OS(OS), DwoOS(DwoOS),
+        IsLittleEndian(IsLittleEndian) {}
+
+  virtual bool checkRelocation(MCContext &Ctx, SMLoc Loc,
+                               const MCSectionELF *From,
+                               const MCSectionELF *To) override {
+    if (isDwoSection(*From)) {
+      Ctx.reportError(Loc, "A dwo section may not contain relocations");
+      return false;
+    }
+    if (To && isDwoSection(*To)) {
+      Ctx.reportError(Loc, "A relocation may not refer to a dwo section");
+      return false;
+    }
+    return true;
+  }
+
+  uint64_t writeObject(MCAssembler &Asm, const MCAsmLayout &Layout) override {
+    uint64_t Size = ELFWriter(*this, OS, IsLittleEndian, ELFWriter::NonDwoOnly)
+                        .writeObject(Asm, Layout);
+    Size += ELFWriter(*this, DwoOS, IsLittleEndian, ELFWriter::DwoOnly)
+                .writeObject(Asm, Layout);
+    return Size;
   }
 };
 
 } // end anonymous namespace
 
 void ELFWriter::align(unsigned Alignment) {
   uint64_t Padding = OffsetToAlignment(W.OS.tell(), Alignment);
   W.OS.write_zeros(Padding);
if (Symbol.isAbsolute()) {
         if (MSD.SectionIndex >= ELF::SHN_LORESERVE)
           HasLargeSectionIndex = true;
       } else {
         MSD.SectionIndex = ELF::SHN_UNDEF;
       }
     } else {
       const MCSectionELF &Section =
           static_cast<const MCSectionELF &>(Symbol.getSection());
+      if (Mode == NonDwoOnly && isDwoSection(Section))
+        continue;
       MSD.SectionIndex = SectionIndexMap.lookup(&Section);
       assert(MSD.SectionIndex && "Invalid section index!");
       if (MSD.SectionIndex >= ELF::SHN_LORESERVE)
         HasLargeSectionIndex = true;
     }
 
     StringRef Name = Symbol.getName();
 
uint64_t ELFWriter::writeObject(MCAssembler &Asm, const MCAsmLayout &Layout) {
   writeHeader(Asm);
 
   // ... then the sections ...
   SectionOffsetsTy SectionOffsets;
   std::vector<MCSectionELF *> Groups;
   std::vector<MCSectionELF *> Relocations;
   for (MCSection &Sec : Asm) {
     MCSectionELF &Section = static_cast<MCSectionELF &>(Sec);
+    if (Mode == NonDwoOnly && isDwoSection(Section))
+      continue;
+    if (Mode == DwoOnly && !isDwoSection(Section))
+      continue;
 
     align(Section.getAlignment());
 
     // Remember the offset into the file for this section.
     uint64_t SecStart = W.OS.tell();
 
     const MCSymbolELF *SignatureSymbol = Section.getGroup();
     writeSectionData(Asm, Section, Layout);
for (const MCSectionELF *Member : GroupMembers[SignatureSymbol]) {
       uint32_t SecIndex = SectionIndexMap.lookup(Member);
       write(SecIndex);
     }
 
     uint64_t SecEnd = W.OS.tell();
     SectionOffsets[Group] = std::make_pair(SecStart, SecEnd);
   }
 
+  if (Mode == DwoOnly) {
+    // dwo files don't have symbol tables or relocations, but they do have
+    // string tables.
+    StrTabBuilder.finalize();
+  } else {
     // Compute symbol table information.
-  computeSymbolTable(Asm, Layout, SectionIndexMap, RevGroupMap, SectionOffsets);
+    computeSymbolTable(Asm, Layout, SectionIndexMap, RevGroupMap,
+                       SectionOffsets);
 
     for (MCSectionELF *RelSection : Relocations) {
       align(RelSection->getAlignment());
 
       // Remember the offset into the file for this section.
       uint64_t SecStart = W.OS.tell();
 
       writeRelocations(Asm,
                        cast<MCSectionELF>(*RelSection->getAssociatedSection()));
 
       uint64_t SecEnd = W.OS.tell();
       SectionOffsets[RelSection] = std::make_pair(SecStart, SecEnd);
     }
+  }
 
   {
     uint64_t SecStart = W.OS.tell();
     const MCSectionELF *Sec = createStringTable(Ctx);
     uint64_t SecEnd = W.OS.tell();
     SectionOffsets[Sec] = std::make_pair(SecStart, SecEnd);
   }
 
void ELFObjectWriter::recordRelocation(MCAssembler &Asm,
   uint64_t Addend = 0;
   if (hasRelocationAddend()) {
     Addend = C;
     C = 0;
   }
 
   FixedValue = C;
 
+  const MCSectionELF *SecA = (SymA && SymA->isInSection())
+                                 ? cast<MCSectionELF>(&SymA->getSection())
+                                 : nullptr;
+  if (!checkRelocation(Ctx, Fixup.getLoc(), &FixupSection, SecA))
+    return;
+
   if (!RelocateWithSymbol) {
-    const MCSection *SecA =
-        (SymA && !SymA->isUndefined()) ? &SymA->getSection() : nullptr;
     const auto *SectionSymbol =
         SecA ? cast<MCSymbolELF>(SecA->getBeginSymbol()) : nullptr;
     if (SectionSymbol)
       SectionSymbol->setUsedInReloc();
     ELFRelocationEntry Rec(FixupOffset, SectionSymbol, Type, Addend, SymA,
                            OriginalC);
     Relocations[&FixupSection].push_back(Rec);
     return;

 }
 
 std::unique_ptr<MCObjectWriter>
 llvm::createELFObjectWriter(std::unique_ptr<MCELFObjectTargetWriter> MOTW,
                             raw_pwrite_stream &OS, bool IsLittleEndian) {
   return llvm::make_unique<ELFSingleObjectWriter>(std::move(MOTW), OS,
                                                   IsLittleEndian);
 }
+
+std::unique_ptr<MCObjectWriter>
+llvm::createELFDwoObjectWriter(std::unique_ptr<MCELFObjectTargetWriter> MOTW,
+                               raw_pwrite_stream &OS, raw_pwrite_stream &DwoOS,
+                               bool IsLittleEndian) {
+  return llvm::make_unique<ELFDwoObjectWriter>(std::move(MOTW), OS, DwoOS,
+                                               IsLittleEndian);
+}

llvm/lib/MC/MCAsmBackend.cpp

MCAsmBackend::createObjectWriter(raw_pwrite_stream &OS) const {
   case Triple::Wasm:
     return createWasmObjectWriter(cast<MCWasmObjectTargetWriter>(std::move(TW)),
                                   OS);
   default:
     llvm_unreachable("unexpected object format");
   }
 }
 
+std::unique_ptr<MCObjectWriter>
+MCAsmBackend::createDwoObjectWriter(raw_pwrite_stream &OS,
+                                    raw_pwrite_stream &DwoOS) const {
+  auto TW = createObjectTargetWriter();
+  if (TW->getFormat() != Triple::ELF)
+    report_fatal_error("dwo only supported with ELF");
+  return createELFDwoObjectWriter(cast<MCELFObjectTargetWriter>(std::move(TW)),
+                                  OS, DwoOS, Endian == support::little);
+}
+
 Optional<MCFixupKind> MCAsmBackend::getFixupKind(StringRef Name) const {
   return None;
 }
 
 const MCFixupKindInfo &MCAsmBackend::getFixupKindInfo(MCFixupKind Kind) const {
   static const MCFixupKindInfo Builtins[] = {
       {"FK_Data_1", 0, 8, 0},
       {"FK_Data_2", 0, 16, 0},

llvm/test/MC/ELF/dwo-restrict-relocs.s

+// RUN: not llvm-mc -filetype=obj -triple x86_64-pc-linux-gnu %s -o %t1 -split-dwarf-file %t2 2>&1 | FileCheck %s
+
+// CHECK: error: A relocation may not refer to a dwo section
+.quad .foo.dwo
+
+.section .foo.dwo
+// CHECK: error: A dwo section may not contain relocations
+.quad .text
+// CHECK: error: A dwo section may not contain relocations
+.quad .foo.dwo

llvm/test/MC/ELF/dwo-sections.s

+// RUN: llvm-mc -filetype=obj -triple x86_64-pc-linux-gnu %s -o %t1 -split-dwarf-file %t2
+// RUN: llvm-objdump -s %t1 | FileCheck --check-prefix=O %s
+// RUN: llvm-objdump -s %t2 | FileCheck --check-prefix=DWO %s
+
+// O-NOT: Contents of section
+// O: Contents of section .strtab:
+// O-NOT: Contents of section
+// O: Contents of section .text:
+// O-NEXT: 0000 c3
+// O-NEXT: Contents of section .symtab:
+// O-NOT: Contents of section
+.globl main
+main:
+.Ltmp1:
+ret
+.Ltmp2:
+
+// DWO-NOT: Contents of section
+// DWO: Contents of section .strtab:
+// DWO-NOT: Contents of section
+// DWO: Contents of section .foo.dwo:
+// DWO-NEXT: 0000 01000000
+// DWO-NOT: Contents of section
+.section .foo.dwo
+.long .Ltmp2-.Ltmp1

llvm/tools/llvm-mc/llvm-mc.cpp

 #include "llvm/Support/ToolOutputFile.h"
 #include "llvm/Support/WithColor.h"
 
 using namespace llvm;
 
 static cl::opt<std::string>
 InputFilename(cl::Positional, cl::desc("<input file>"), cl::init("-"));
 
-static cl::opt<std::string>
-OutputFilename("o", cl::desc("Output filename"),
+static cl::opt<std::string> OutputFilename("o", cl::desc("Output filename"),
+                                           cl::value_desc("filename"),
+                                           cl::init("-"));
+
+static cl::opt<std::string> SplitDwarfFile("split-dwarf-file",
+                                           cl::desc("DWO output filename"),
                                            cl::value_desc("filename"));
 
 static cl::opt<bool>
 ShowEncoding("show-encoding", cl::desc("Show instruction encodings"));
 
 static cl::opt<bool> RelaxELFRel(
     "relax-relocations", cl::init(true),
     cl::desc("Emit R_X86_64_GOTPCRELX instead of R_X86_64_GOTPCREL"));
 
if (!TheTarget) {
     return nullptr;
   }
 
   // Update the triple name and return the found target.
   TripleName = TheTriple.getTriple();
   return TheTarget;
 }
 
-static std::unique_ptr<ToolOutputFile> GetOutputStream() {
-  if (OutputFilename == "")
-    OutputFilename = "-";
-
+static std::unique_ptr<ToolOutputFile> GetOutputStream(StringRef Path) {
   std::error_code EC;
-  auto Out =
-      llvm::make_unique<ToolOutputFile>(OutputFilename, EC, sys::fs::F_None);
+  auto Out = llvm::make_unique<ToolOutputFile>(Path, EC, sys::fs::F_None);
   if (EC) {
     WithColor::error() << EC.message() << '\n';
     return nullptr;
   }
 
   return Out;
 }
 
int main(int argc, char **argv) {
   std::string FeaturesStr;
   if (MAttrs.size()) {
     SubtargetFeatures Features;
     for (unsigned i = 0; i != MAttrs.size(); ++i)
       Features.AddFeature(MAttrs[i]);
     FeaturesStr = Features.getString();
   }
 
-  std::unique_ptr<ToolOutputFile> Out = GetOutputStream();
+  std::unique_ptr<ToolOutputFile> Out = GetOutputStream(OutputFilename);
   if (!Out)
     return 1;
 
+  std::unique_ptr<ToolOutputFile> DwoOut;
+  if (!SplitDwarfFile.empty()) {
+    if (FileType != OFT_ObjectFile) {
+      WithColor::error() << "dwo output only supported with object files\n";
+      return 1;
+    }
+    DwoOut = GetOutputStream(SplitDwarfFile);
+    if (!DwoOut)
+      return 1;
+  }
+
   std::unique_ptr<buffer_ostream> BOS;
   raw_pwrite_stream *OS = &Out->os();
   std::unique_ptr<MCStreamer> Str;
 
   std::unique_ptr<MCInstrInfo> MCII(TheTarget->createMCInstrInfo());
   std::unique_ptr<MCSubtargetInfo> STI(
       TheTarget->createMCSubtargetInfo(TripleName, MCPU, FeaturesStr));
 
if (!Out->os().supportsSeeking()) {
       BOS = make_unique<buffer_ostream>(Out->os());
       OS = BOS.get();
     }
 
     MCCodeEmitter *CE = TheTarget->createMCCodeEmitter(*MCII, *MRI, Ctx);
     MCAsmBackend *MAB = TheTarget->createMCAsmBackend(*STI, *MRI, MCOptions);
     Str.reset(TheTarget->createMCObjectStreamer(
         TheTriple, Ctx, std::unique_ptr<MCAsmBackend>(MAB),
-        MAB->createObjectWriter(*OS), std::unique_ptr<MCCodeEmitter>(CE), *STI,
-        MCOptions.MCRelaxAll, MCOptions.MCIncrementalLinkerCompatible,
+        DwoOut ? MAB->createDwoObjectWriter(*OS, DwoOut->os())
+               : MAB->createObjectWriter(*OS),
+        std::unique_ptr<MCCodeEmitter>(CE), *STI, MCOptions.MCRelaxAll,
+        MCOptions.MCIncrementalLinkerCompatible,
         /*DWARFMustBeAtTheEnd*/ false));
     if (NoExecStack)
       Str->InitSections(true);
   }
 
   // Use Assembler information for parsing.
   Str->setUseAssemblerInfoForParsing(true);
 
case AC_Disassemble:
     disassemble = true;
     break;
   }
   if (disassemble)
     Res = Disassembler::disassemble(*TheTarget, TripleName, *STI, *Str,
                                     *Buffer, SrcMgr, Out->os());
 
   // Keep output if no errors.
-  if (Res == 0) Out->keep();
+  if (Res == 0) {
+    Out->keep();
+    if (DwoOut)
+      DwoOut->keep();
+  }
   return Res;
 }