Implement support for __llvm_addrsig for MachO in llvm-mc

[alx32]

Summary

The __llvm_addrsig section is a section that the linker needs for safe icf.
This was not yet implemented for MachO - this is the implementation.
It has been tested with a safe deduplication implementation inside lld.

[smeenai]

Adding some more MC reviewers.

You're missing a test case. You should also upload the patch with context (arc does this for you automatically, otherwise use git diff -U99999 if uploading a patch manually).

[MaskRay]

llvm/include/llvm/BinaryFormat/MachO.h

178

The values may be defined by Apple. We cannot randomly grab a value.

[tschuett]

llvm/include/llvm/BinaryFormat/MachO.h

176

This one is in my local loader.h. S_ADDRSIG is not there.

[smeenai]

D112160 is a similar recent example of adding a custom Mach-O section, and it might be a good reference for adding a test and assigning a section identifier.

[alx32]

Add unit test & address section ID comments

[int3]

llvm/test/MC/MachO/addrsig-macho.ll

5 ↗

(On Diff #423450)

similar to my comment on D123752, let's minimize this test case

117 ↗

(On Diff #423450)

would be nice to follow up with a diff that teaches objdump to properly dump this section

[int3]

@MaskRay do you feel comfortable reviewing this?

[MaskRay]

Placing code gen test/MC

llvm/include/llvm/MC/MCObjectWriter.h

37

Moving the small boolean variable after the vector. If someone adds more smaller variables to the class in the future, it will make the class size smaller.

llvm/lib/MC/MachObjectWriter.cpp

767

*fragmentList.begin() the dereference has implied that it can't be nullptr, so assert can be removed.

773

drop brace

llvm/test/MC/MachO/addrsig-macho.ll

55 ↗

(On Diff #424049)

CodeGen/X86/addrsig.ll has some interesting IR objects which may be worth adding here. It seems to me that CodeGen may be a better directory.

[ayermolo]

llvm/lib/MC/MCMachOStreamer.cpp

34

I think this snuck in?

llvm/lib/MC/MachObjectWriter.cpp

760

replace with actual type

[alx32]

@MaskRay, @ayermolo - Addressed all feedback, could you have a look again ? Thanks!

[MaskRay]

llvm/lib/MC/MachObjectWriter.cpp

765

delete the blank line

llvm/test/CodeGen/AArch64/addrsig-macho.ll

5 ↗

(On Diff #425054)

[ayermolo]

lgtm

llvm/include/llvm/BinaryFormat/MachO.h

   /// S_THREAD_LOCAL_INIT_FUNCTION_POINTERS - Section with thread local
   /// variable initialization pointers to functions.
   S_THREAD_LOCAL_INIT_FUNCTION_POINTERS = 0x15u,
+  /// S_ADDRSIG - Section with list of sybols marked as address significant.
+  S_ADDRSIG = 0x16u,
 
-  LAST_KNOWN_SECTION_TYPE = S_THREAD_LOCAL_INIT_FUNCTION_POINTERS
+  LAST_KNOWN_SECTION_TYPE = S_ADDRSIG
 };
 
 enum : uint32_t {

llvm/include/llvm/MC/MCMachObjectWriter.h

                                               bool IsPCRel) const override;
 
   uint64_t writeObject(MCAssembler &Asm, const MCAsmLayout &Layout) override;
+
+  void writeAddrsigSection(MCAssembler &Asm);
 };
 
 /// Construct a new Mach-O writer instance.

llvm/include/llvm/MC/MCObjectFileInfo.h

   MCSection *LazySymbolPointerSection = nullptr;
   MCSection *NonLazySymbolPointerSection = nullptr;
   MCSection *ThreadLocalPointerSection = nullptr;
+  MCSection *AddrSigSection = nullptr;
 
   /// COFF specific sections.
   MCSection *DrectveSection = nullptr;

   MCSection *getThreadLocalPointerSection() const {
     return ThreadLocalPointerSection;
   }
+  MCSection *getAddrSigSection() const { return AddrSigSection; }
 
   // COFF specific sections.
   MCSection *getDrectveSection() const { return DrectveSection; }

llvm/include/llvm/MC/MCObjectWriter.h

 #define LLVM_MC_MCOBJECTWRITER_H
 
 #include "llvm/ADT/Triple.h"
+#include "llvm/MC/MCSymbol.h"
 #include <cstdint>
 
 namespace llvm {

 /// should be emitted as part of writeObject().
 class MCObjectWriter {
 protected:
+  bool EmitAddrsigSection = false;
+  std::vector<const MCSymbol *> AddrsigSyms;
+
   MCObjectWriter() = default;
 
 public:

   /// Tell the object writer to emit an address-significance table during
   /// writeObject(). If this function is not called, all symbols are treated as
   /// address-significant.
-  virtual void emitAddrsigSection() {}
+  void emitAddrsigSection() { EmitAddrsigSection = true; }
+
+  bool getEmitAddrsigSection() { return EmitAddrsigSection; }
 
   /// Record the given symbol in the address-significance table to be written
   /// diring writeObject().
-  virtual void addAddrsigSymbol(const MCSymbol *Sym) {}
+  void addAddrsigSymbol(const MCSymbol *Sym) { AddrsigSyms.push_back(Sym); }
+
+  std::vector<const MCSymbol *> &getAddrsigSyms() { return AddrsigSyms; }
 
   /// Write the object file and returns the number of bytes written.
   ///

llvm/lib/MC/ELFObjectWriter.cpp

   DenseMap<const MCSymbolELF *, const MCSymbolELF *> Renames;
 
   bool SeenGnuAbi = false;
-  bool EmitAddrsigSection = false;
-  std::vector<const MCSymbol *> AddrsigSyms;
 
   bool hasRelocationAddend() const;
 

 
   void markGnuAbi() override { SeenGnuAbi = true; }
   bool seenGnuAbi() const { return SeenGnuAbi; }
-  void emitAddrsigSection() override { EmitAddrsigSection = true; }
-  void addAddrsigSymbol(const MCSymbol *Sym) override {
-    AddrsigSyms.push_back(Sym);
-  }
 
   friend struct ELFWriter;
 };

llvm/lib/MC/MCMachOStreamer.cpp

 #include "llvm/MC/TargetRegistry.h"
 #include "llvm/Support/Casting.h"
 #include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/LEB128.h"
 #include "llvm/Support/raw_ostream.h"
 #include <cassert>
 #include <vector>

 
   void finalizeCGProfileEntry(const MCSymbolRefExpr *&SRE);
   void finalizeCGProfile();
+  void createAddrSigSection();
 };
 
 } // end anonymous namespace.

 
   finalizeCGProfile();
 
+  createAddrSigSection();
   this->MCObjectStreamer::finishImpl();
 }
 

     S->getAssembler().setRelaxAll(true);
   return S;
 }
+
+// Create the AddrSig section and first data fragment here as its layout needs
+// to be computed immediately after in order for it to be exported correctly.
+void MCMachOStreamer::createAddrSigSection() {
+  MCAssembler &Asm = getAssembler();
+  MCObjectWriter &writer = Asm.getWriter();
+  if (!writer.getEmitAddrsigSection())
+    return;
+  MCSection *AddrSigSection =
+      Asm.getContext().getObjectFileInfo()->getAddrSigSection();
+  Asm.registerSection(*AddrSigSection);
+  new MCDataFragment(AddrSigSection);
+}
 No newline at end of file

llvm/lib/MC/MCObjectFileInfo.cpp

                            MachO::S_THREAD_LOCAL_VARIABLE_POINTERS,
                            SectionKind::getMetadata());
 
+  AddrSigSection = Ctx->getMachOSection(
+      "__DATA", "__llvm_addrsig", MachO::S_ADDRSIG, SectionKind::getData());
+
   // Exception Handling.
   LSDASection = Ctx->getMachOSection("__TEXT", "__gcc_except_tab", 0,
                                      SectionKind::getReadOnlyWithRel());

llvm/lib/MC/MCSectionMachO.cpp

      StringLiteral("S_THREAD_LOCAL_VARIABLE_POINTERS")}, // 0x14
     {StringLiteral("thread_local_init_function_pointers"),
      StringLiteral("S_THREAD_LOCAL_INIT_FUNCTION_POINTERS")}, // 0x15
+    {StringLiteral("addrsig"), StringLiteral("S_ADDRSIG")},   // 0x16
 };
 
 /// SectionAttrDescriptors - This is an array of descriptors for section

llvm/lib/MC/MachObjectWriter.cpp

 #include "llvm/MC/MCFixupKindInfo.h"
 #include "llvm/MC/MCFragment.h"
 #include "llvm/MC/MCMachObjectWriter.h"
+#include "llvm/MC/MCObjectFileInfo.h"
 #include "llvm/MC/MCObjectWriter.h"
 #include "llvm/MC/MCSection.h"
 #include "llvm/MC/MCSectionMachO.h"

 #include "llvm/Support/Casting.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/LEB128.h"
 #include "llvm/Support/MathExtras.h"
 #include "llvm/Support/raw_ostream.h"
 #include <algorithm>

   llvm_unreachable("Invalid mc version min type");
 }
 
+// Encode addrsig data as symbol indexes in variable length encoding.
+void MachObjectWriter::writeAddrsigSection(MCAssembler &Asm) {
+  MCSection *AddrSigSection =
+      Asm.getContext().getObjectFileInfo()->getAddrSigSection();
+  auto &fragmentList = AddrSigSection->getFragmentList();
+  assert(fragmentList.size() <= 1);
+
+  MCFragment *pFragment = &*fragmentList.begin();
+  assert(pFragment);
+  MCDataFragment *pDataFragment = dyn_cast_or_null<MCDataFragment>(pFragment);
+  assert(pDataFragment);
+
+  raw_svector_ostream OS(pDataFragment->getContents());
+  for (const MCSymbol *sym : this->getAddrsigSyms()) {
+    encodeULEB128(sym->getIndex(), OS);
+  }
+}
+
 uint64_t MachObjectWriter::writeObject(MCAssembler &Asm,
                                        const MCAsmLayout &Layout) {
   uint64_t StartOffset = W.OS.tell();

   // Compute symbol table information and bind symbol indices.
   computeSymbolTable(Asm, LocalSymbolData, ExternalSymbolData,
                      UndefinedSymbolData);
+  writeAddrsigSection(Asm);
 
   if (!Asm.CGProfile.empty()) {
     MCSection *CGProfileSection = Asm.getContext().getMachOSection(

llvm/lib/MC/WinCOFFObjectWriter.cpp

   bool UseBigObj;
   bool UseOffsetLabels = false;
 
-  bool EmitAddrsigSection = false;
   MCSectionCOFF *AddrsigSection;
-  std::vector<const MCSymbol *> AddrsigSyms;
 
   MCSectionCOFF *CGProfileSection = nullptr;
 

   void assignSectionNumbers();
   void assignFileOffsets(MCAssembler &Asm, const MCAsmLayout &Layout);
 
-  void emitAddrsigSection() override { EmitAddrsigSection = true; }
-  void addAddrsigSymbol(const MCSymbol *Sym) override {
-    AddrsigSyms.push_back(Sym);
-  }
-
   uint64_t writeObject(MCAssembler &Asm, const MCAsmLayout &Layout) override;
 };