COFF: Implement safe ICF on rodata using address-significance tables.

[rnk]

lld/COFF/Driver.cpp

761

This should be checked in case of corrupt inputs.

[pcc]

• Add error handling

[rnk]

Lgtm

lld/COFF/Chunks.h

protected:
   uint64_t RVA = 0;
 
   // The output section for this chunk.
   OutputSection *Out = nullptr;
 
 public:
   // The offset from beginning of the output section. The writer sets a value.
   uint64_t OutputSectionOff = 0;
+
+  // Whether this section needs to be kept distinct from other sections during
+  // ICF. This is set by the driver using address-significance tables.
+  bool KeepUnique = false;
 };
 
 // A chunk corresponding a section of an input file.
 class SectionChunk final : public Chunk {
   // Identical COMDAT Folding feature accesses section internal data.
   friend class ICF;
 
 public:

lld/COFF/Driver.cpp

 #include "llvm/BinaryFormat/Magic.h"
 #include "llvm/Object/ArchiveWriter.h"
 #include "llvm/Object/COFFImportFile.h"
 #include "llvm/Object/COFFModuleDefinition.h"
 #include "llvm/Option/Arg.h"
 #include "llvm/Option/ArgList.h"
 #include "llvm/Option/Option.h"
 #include "llvm/Support/Debug.h"
+#include "llvm/Support/LEB128.h"
 #include "llvm/Support/Path.h"
 #include "llvm/Support/Process.h"
 #include "llvm/Support/TarWriter.h"
 #include "llvm/Support/TargetSelect.h"
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/ToolDrivers/llvm-lib/LibDriver.h"
 #include <algorithm>
 #include <future>
if (Set.count(S) == 0) {
       if (Config->WarnMissingOrderSymbol)
         warn("/order:" + Arg + ": missing symbol: " + S + " [LNK4037]");
     }
     else
       Config->Order[S] = INT_MIN + Config->Order.size();
   }
 }
 
+static void markAddrsig(Symbol *S) {
+  if (auto *D = dyn_cast_or_null<Defined>(S))
+    if (Chunk *C = D->getChunk())
+      C->KeepUnique = true;
+}
+
+static void findKeepUniqueSections() {
+  // Exported symbols could be address-significant in other executables or DSOs,
+  // so we conservatively mark them as address-significant.
+  for (Export &R : Config->Exports)
+    markAddrsig(R.Sym);
+
+  // Visit the address-significance table in each object file and mark each
+  // referenced symbol as address-significant.
+  for (ObjFile *Obj : ObjFile::Instances) {
+    ArrayRef<Symbol *> Syms = Obj->getSymbols();
+    if (Obj->AddrsigSec) {
+      ArrayRef<uint8_t> Contents;
+      Obj->getCOFFObj()->getSectionContents(Obj->AddrsigSec, Contents);
+      const uint8_t *Cur = Contents.begin();
+      while (Cur != Contents.end()) {
+        unsigned Size;
+        const char *Err;
+        uint64_t SymIndex = decodeULEB128(Cur, &Size, Contents.end(), &Err);
+        if (Err)
+          fatal(toString(Obj) + ": could not decode addrsig section: " + Err);
+        markAddrsig(Syms[SymIndex]);
+        Cur += Size;
+      }
+    } else {
+      // If an object file does not have an address-significance table,
+      // conservatively mark all of its symbols as address-significant.
+      for (Symbol *S : Syms)
+        markAddrsig(S);
+    }
+  }
+}
+
 void LinkerDriver::link(ArrayRef<const char *> ArgsArr) {
   // If the first command line argument is "/lib", link.exe acts like lib.exe.
   // We call our own implementation of lib.exe that understands bitcode files.
   if (ArgsArr.size() > 1 && StringRef(ArgsArr[1]).equals_lower("/lib")) {
     if (llvm::libDriverMain(ArgsArr.slice(1)) != 0)
       fatal("lib failed");
     return;
   }
void LinkerDriver::link(ArrayRef<const char *> ArgsArr) {
   if (auto *Arg = Args.getLastArg(OPT_order))
     parseOrderFile(Arg->getValue());
 
   // Identify unreferenced COMDAT sections.
   if (Config->DoGC)
     markLive(Symtab->getChunks());
 
   // Identify identical COMDAT sections to merge them.
-  if (Config->DoICF)
+  if (Config->DoICF) {
+    findKeepUniqueSections();
     doICF(Symtab->getChunks());
+  }
 
   // Write the result.
   writeResult();
 
   // Stop early so we can print the results.
   Timer::root().stop();
   if (Config->ShowTiming)
     Timer::root().print();
 }
 
 } // namespace coff
 } // namespace lld

lld/COFF/ICF.cpp

if (C->getOutputCharacteristics() & llvm::COFF::IMAGE_SCN_MEM_EXECUTE)
     return true;
 
   // .pdata and .xdata unwind info sections are eligible.
   StringRef OutSecName = C->getSectionName().split('$').first;
   if (OutSecName == ".pdata" || OutSecName == ".xdata")
     return true;
 
   // So are vtables.
-  return C->Sym && C->Sym->getName().startswith("??_7");
+  if (C->Sym && C->Sym->getName().startswith("??_7"))
+    return true;
+
+  // Anything else not in an address-significance table is eligible.
+  return !C->KeepUnique;
 }
 
 // Split an equivalence class into smaller classes.
 void ICF::segregate(size_t Begin, size_t End, bool Constant) {
   while (Begin < End) {
     // Divide [Begin, End) into two. Let Mid be the start index of the
     // second group.
     auto Bound = std::stable_partition(

lld/COFF/InputFiles.h

public:
   bool hasGuardCF() { return Feat00Flags & 0x800; }
 
   // Pointer to the PDB module descriptor builder. Various debug info records
   // will reference object files by "module index", which is here. Things like
   // source files and section contributions are also recorded here. Will be null
   // if we are not producing a PDB.
   llvm::pdb::DbiModuleDescriptorBuilder *ModuleDBI = nullptr;
 
+  const coff_section *AddrsigSec = nullptr;
+
 private:
   void initializeChunks();
   void initializeSymbols();
 
   SectionChunk *
   readSection(uint32_t SectionNumber,
               const llvm::object::coff_aux_section_definition *Def,
               StringRef LeaderName);

lld/COFF/InputFiles.cpp

SectionChunk *ObjFile::readSection(uint32_t SectionNumber,
 
   if (Name == ".drectve") {
     ArrayRef<uint8_t> Data;
     COFFObj->getSectionContents(Sec, Data);
     Directives = std::string((const char *)Data.data(), Data.size());
     return nullptr;
   }
 
+  if (Name == ".llvm_addrsig") {
+    AddrsigSec = Sec;
+    return nullptr;
+  }
+
   // Object files may have DWARF debug info or MS CodeView debug info
   // (or both).
   //
   // DWARF sections don't need any special handling from the perspective
   // of the linker; they are just a data section containing relocations.
   // We can just link them to complete debug info.
   //
   // CodeView needs a linker support. We need to interpret and debug

lld/test/COFF/Inputs/icf-safe.s

+.section .rdata,"dr",one_only,non_addrsig1
+.globl non_addrsig1
+non_addrsig1:
+.byte 3
+
+.section .rdata,"dr",one_only,non_addrsig2
+.globl non_addrsig2
+non_addrsig2:
+.byte 3

lld/test/COFF/icf-safe.s

+# RUN: llvm-mc -filetype=obj -triple=x86_64-pc-win32 %s -o %t1.obj
+# RUN: llvm-mc -filetype=obj -triple=x86_64-pc-win32 %S/Inputs/icf-safe.s -o %t2.obj
+# RUN: lld-link /dll /noentry /out:%t.dll /verbose /opt:noref,icf %t1.obj %t2.obj 2>&1 | FileCheck %s
+# RUN: lld-link /dll /noentry /out:%t.dll /verbose /opt:noref,icf /export:g3 /export:g4 %t1.obj %t2.obj 2>&1 | FileCheck --check-prefix=EXPORT %s
+
+# CHECK-NOT: Selected
+# CHECK: Selected g3
+# CHECK-NEXT:   Removed g4
+# CHECK-NOT: Removed
+# CHECK-NOT: Selected
+
+# EXPORT-NOT: Selected
+
+.section .rdata,"dr",one_only,g1
+.globl g1
+g1:
+.byte 1
+
+.section .rdata,"dr",one_only,g2
+.globl g2
+g2:
+.byte 1
+
+.section .rdata,"dr",one_only,g3
+.globl g3
+g3:
+.byte 2
+
+.section .rdata,"dr",one_only,g4
+.globl g4
+g4:
+.byte 2
+
+.addrsig
+.addrsig_sym g1
+.addrsig_sym g2