diff options
author | rdivacky <rdivacky@FreeBSD.org> | 2010-03-21 10:49:05 +0000 |
---|---|---|
committer | rdivacky <rdivacky@FreeBSD.org> | 2010-03-21 10:49:05 +0000 |
commit | 2f2afc1aae898651e26987a5c71f3febb19bca98 (patch) | |
tree | 2caca31db4facdc95c23930c0c745c8ef0dee97d /lib/MC | |
parent | 0f448b841684305c051796982f300c9bff959307 (diff) | |
download | FreeBSD-src-2f2afc1aae898651e26987a5c71f3febb19bca98.zip FreeBSD-src-2f2afc1aae898651e26987a5c71f3febb19bca98.tar.gz |
Update LLVM to r99115.
Diffstat (limited to 'lib/MC')
-rw-r--r-- | lib/MC/CMakeLists.txt | 2 | ||||
-rw-r--r-- | lib/MC/MCAsmStreamer.cpp | 3 | ||||
-rw-r--r-- | lib/MC/MCAssembler.cpp | 1234 | ||||
-rw-r--r-- | lib/MC/MCCodeEmitter.cpp | 8 | ||||
-rw-r--r-- | lib/MC/MCContext.cpp | 11 | ||||
-rw-r--r-- | lib/MC/MCExpr.cpp | 44 | ||||
-rw-r--r-- | lib/MC/MCMachOStreamer.cpp | 8 | ||||
-rw-r--r-- | lib/MC/MCObjectWriter.cpp | 15 | ||||
-rw-r--r-- | lib/MC/MCValue.cpp | 9 | ||||
-rw-r--r-- | lib/MC/MachObjectWriter.cpp | 1109 | ||||
-rw-r--r-- | lib/MC/TargetAsmBackend.cpp | 5 |
11 files changed, 1413 insertions, 1035 deletions
diff --git a/lib/MC/CMakeLists.txt b/lib/MC/CMakeLists.txt index 4cf71dc..dba0e14 100644 --- a/lib/MC/CMakeLists.txt +++ b/lib/MC/CMakeLists.txt @@ -12,11 +12,13 @@ add_llvm_library(LLVMMC MCInstPrinter.cpp MCMachOStreamer.cpp MCNullStreamer.cpp + MCObjectWriter.cpp MCSection.cpp MCSectionELF.cpp MCSectionMachO.cpp MCStreamer.cpp MCSymbol.cpp MCValue.cpp + MachObjectWriter.cpp TargetAsmBackend.cpp ) diff --git a/lib/MC/MCAsmStreamer.cpp b/lib/MC/MCAsmStreamer.cpp index 7f39471..2025463 100644 --- a/lib/MC/MCAsmStreamer.cpp +++ b/lib/MC/MCAsmStreamer.cpp @@ -16,6 +16,7 @@ #include "llvm/MC/MCInstPrinter.h" #include "llvm/MC/MCSectionMachO.h" #include "llvm/MC/MCSymbol.h" +#include "llvm/ADT/OwningPtr.h" #include "llvm/ADT/SmallString.h" #include "llvm/ADT/Twine.h" #include "llvm/Support/ErrorHandling.h" @@ -29,7 +30,7 @@ namespace { class MCAsmStreamer : public MCStreamer { formatted_raw_ostream &OS; const MCAsmInfo &MAI; - MCInstPrinter *InstPrinter; + OwningPtr<MCInstPrinter> InstPrinter; MCCodeEmitter *Emitter; SmallString<128> CommentToEmit; diff --git a/lib/MC/MCAssembler.cpp b/lib/MC/MCAssembler.cpp index 4cf8b7e..beecf7e 100644 --- a/lib/MC/MCAssembler.cpp +++ b/lib/MC/MCAssembler.cpp @@ -10,18 +10,16 @@ #define DEBUG_TYPE "assembler" #include "llvm/MC/MCAssembler.h" #include "llvm/MC/MCAsmLayout.h" +#include "llvm/MC/MCCodeEmitter.h" #include "llvm/MC/MCExpr.h" -#include "llvm/MC/MCSectionMachO.h" +#include "llvm/MC/MCObjectWriter.h" #include "llvm/MC/MCSymbol.h" #include "llvm/MC/MCValue.h" -#include "llvm/ADT/DenseMap.h" -#include "llvm/ADT/SmallString.h" +#include "llvm/ADT/OwningPtr.h" #include "llvm/ADT/Statistic.h" #include "llvm/ADT/StringExtras.h" -#include "llvm/ADT/StringMap.h" #include "llvm/ADT/Twine.h" #include "llvm/Support/ErrorHandling.h" -#include "llvm/Support/MachO.h" #include "llvm/Support/raw_ostream.h" #include "llvm/Support/Debug.h" #include "llvm/Target/TargetRegistry.h" @@ -33,8 +31,6 @@ #include <vector> using namespace llvm; -class MachObjectWriter; - STATISTIC(EmittedFragments, "Number of emitted assembler fragments"); // FIXME FIXME FIXME: There are number of places in this file where we convert @@ -42,917 +38,6 @@ STATISTIC(EmittedFragments, "Number of emitted assembler fragments"); // object file, which may truncate it. We should detect that truncation where // invalid and report errors back. -static void WriteFileData(raw_ostream &OS, const MCSectionData &SD, - MachObjectWriter &MOW); - -static uint64_t WriteNopData(uint64_t Count, MachObjectWriter &MOW); - -/// isVirtualSection - Check if this is a section which does not actually exist -/// in the object file. -static bool isVirtualSection(const MCSection &Section) { - // FIXME: Lame. - const MCSectionMachO &SMO = static_cast<const MCSectionMachO&>(Section); - return (SMO.getType() == MCSectionMachO::S_ZEROFILL); -} - -static unsigned getFixupKindLog2Size(unsigned Kind) { - switch (Kind) { - default: llvm_unreachable("invalid fixup kind!"); - case X86::reloc_pcrel_1byte: - case FK_Data_1: return 0; - case FK_Data_2: return 1; - case X86::reloc_pcrel_4byte: - case X86::reloc_riprel_4byte: - case FK_Data_4: return 2; - case FK_Data_8: return 3; - } -} - -static bool isFixupKindPCRel(unsigned Kind) { - switch (Kind) { - default: - return false; - case X86::reloc_pcrel_1byte: - case X86::reloc_pcrel_4byte: - case X86::reloc_riprel_4byte: - return true; - } -} - -class MachObjectWriter { - // See <mach-o/loader.h>. - enum { - Header_Magic32 = 0xFEEDFACE, - Header_Magic64 = 0xFEEDFACF - }; - - enum { - Header32Size = 28, - Header64Size = 32, - SegmentLoadCommand32Size = 56, - SegmentLoadCommand64Size = 72, - Section32Size = 68, - Section64Size = 80, - SymtabLoadCommandSize = 24, - DysymtabLoadCommandSize = 80, - Nlist32Size = 12, - Nlist64Size = 16, - RelocationInfoSize = 8 - }; - - enum HeaderFileType { - HFT_Object = 0x1 - }; - - enum HeaderFlags { - HF_SubsectionsViaSymbols = 0x2000 - }; - - enum LoadCommandType { - LCT_Segment = 0x1, - LCT_Symtab = 0x2, - LCT_Dysymtab = 0xb, - LCT_Segment64 = 0x19 - }; - - // See <mach-o/nlist.h>. - enum SymbolTypeType { - STT_Undefined = 0x00, - STT_Absolute = 0x02, - STT_Section = 0x0e - }; - - enum SymbolTypeFlags { - // If any of these bits are set, then the entry is a stab entry number (see - // <mach-o/stab.h>. Otherwise the other masks apply. - STF_StabsEntryMask = 0xe0, - - STF_TypeMask = 0x0e, - STF_External = 0x01, - STF_PrivateExtern = 0x10 - }; - - /// IndirectSymbolFlags - Flags for encoding special values in the indirect - /// symbol entry. - enum IndirectSymbolFlags { - ISF_Local = 0x80000000, - ISF_Absolute = 0x40000000 - }; - - /// RelocationFlags - Special flags for addresses. - enum RelocationFlags { - RF_Scattered = 0x80000000 - }; - - enum RelocationInfoType { - RIT_Vanilla = 0, - RIT_Pair = 1, - RIT_Difference = 2, - RIT_PreboundLazyPointer = 3, - RIT_LocalDifference = 4 - }; - - /// MachSymbolData - Helper struct for containing some precomputed information - /// on symbols. - struct MachSymbolData { - MCSymbolData *SymbolData; - uint64_t StringIndex; - uint8_t SectionIndex; - - // Support lexicographic sorting. - bool operator<(const MachSymbolData &RHS) const { - const std::string &Name = SymbolData->getSymbol().getName(); - return Name < RHS.SymbolData->getSymbol().getName(); - } - }; - - raw_ostream &OS; - unsigned Is64Bit : 1; - unsigned IsLSB : 1; - -public: - MachObjectWriter(raw_ostream &_OS, bool _Is64Bit, bool _IsLSB = true) - : OS(_OS), Is64Bit(_Is64Bit), IsLSB(_IsLSB) { - } - - /// @name Helper Methods - /// @{ - - void Write8(uint8_t Value) { - OS << char(Value); - } - - void Write16(uint16_t Value) { - if (IsLSB) { - Write8(uint8_t(Value >> 0)); - Write8(uint8_t(Value >> 8)); - } else { - Write8(uint8_t(Value >> 8)); - Write8(uint8_t(Value >> 0)); - } - } - - void Write32(uint32_t Value) { - if (IsLSB) { - Write16(uint16_t(Value >> 0)); - Write16(uint16_t(Value >> 16)); - } else { - Write16(uint16_t(Value >> 16)); - Write16(uint16_t(Value >> 0)); - } - } - - void Write64(uint64_t Value) { - if (IsLSB) { - Write32(uint32_t(Value >> 0)); - Write32(uint32_t(Value >> 32)); - } else { - Write32(uint32_t(Value >> 32)); - Write32(uint32_t(Value >> 0)); - } - } - - void WriteZeros(unsigned N) { - const char Zeros[16] = { 0 }; - - for (unsigned i = 0, e = N / 16; i != e; ++i) - OS << StringRef(Zeros, 16); - - OS << StringRef(Zeros, N % 16); - } - - void WriteString(StringRef Str, unsigned ZeroFillSize = 0) { - OS << Str; - if (ZeroFillSize) - WriteZeros(ZeroFillSize - Str.size()); - } - - /// @} - - void WriteHeader(unsigned NumLoadCommands, unsigned LoadCommandsSize, - bool SubsectionsViaSymbols) { - uint32_t Flags = 0; - - if (SubsectionsViaSymbols) - Flags |= HF_SubsectionsViaSymbols; - - // struct mach_header (28 bytes) or - // struct mach_header_64 (32 bytes) - - uint64_t Start = OS.tell(); - (void) Start; - - Write32(Is64Bit ? Header_Magic64 : Header_Magic32); - - // FIXME: Support cputype. - Write32(Is64Bit ? MachO::CPUTypeX86_64 : MachO::CPUTypeI386); - // FIXME: Support cpusubtype. - Write32(MachO::CPUSubType_I386_ALL); - Write32(HFT_Object); - Write32(NumLoadCommands); // Object files have a single load command, the - // segment. - Write32(LoadCommandsSize); - Write32(Flags); - if (Is64Bit) - Write32(0); // reserved - - assert(OS.tell() - Start == Is64Bit ? Header64Size : Header32Size); - } - - /// WriteSegmentLoadCommand - Write a segment load command. - /// - /// \arg NumSections - The number of sections in this segment. - /// \arg SectionDataSize - The total size of the sections. - void WriteSegmentLoadCommand(unsigned NumSections, - uint64_t VMSize, - uint64_t SectionDataStartOffset, - uint64_t SectionDataSize) { - // struct segment_command (56 bytes) or - // struct segment_command_64 (72 bytes) - - uint64_t Start = OS.tell(); - (void) Start; - - unsigned SegmentLoadCommandSize = Is64Bit ? SegmentLoadCommand64Size : - SegmentLoadCommand32Size; - Write32(Is64Bit ? LCT_Segment64 : LCT_Segment); - Write32(SegmentLoadCommandSize + - NumSections * (Is64Bit ? Section64Size : Section32Size)); - - WriteString("", 16); - if (Is64Bit) { - Write64(0); // vmaddr - Write64(VMSize); // vmsize - Write64(SectionDataStartOffset); // file offset - Write64(SectionDataSize); // file size - } else { - Write32(0); // vmaddr - Write32(VMSize); // vmsize - Write32(SectionDataStartOffset); // file offset - Write32(SectionDataSize); // file size - } - Write32(0x7); // maxprot - Write32(0x7); // initprot - Write32(NumSections); - Write32(0); // flags - - assert(OS.tell() - Start == SegmentLoadCommandSize); - } - - void WriteSection(const MCSectionData &SD, uint64_t FileOffset, - uint64_t RelocationsStart, unsigned NumRelocations) { - // The offset is unused for virtual sections. - if (isVirtualSection(SD.getSection())) { - assert(SD.getFileSize() == 0 && "Invalid file size!"); - FileOffset = 0; - } - - // struct section (68 bytes) or - // struct section_64 (80 bytes) - - uint64_t Start = OS.tell(); - (void) Start; - - // FIXME: cast<> support! - const MCSectionMachO &Section = - static_cast<const MCSectionMachO&>(SD.getSection()); - WriteString(Section.getSectionName(), 16); - WriteString(Section.getSegmentName(), 16); - if (Is64Bit) { - Write64(SD.getAddress()); // address - Write64(SD.getSize()); // size - } else { - Write32(SD.getAddress()); // address - Write32(SD.getSize()); // size - } - Write32(FileOffset); - - unsigned Flags = Section.getTypeAndAttributes(); - if (SD.hasInstructions()) - Flags |= MCSectionMachO::S_ATTR_SOME_INSTRUCTIONS; - - assert(isPowerOf2_32(SD.getAlignment()) && "Invalid alignment!"); - Write32(Log2_32(SD.getAlignment())); - Write32(NumRelocations ? RelocationsStart : 0); - Write32(NumRelocations); - Write32(Flags); - Write32(0); // reserved1 - Write32(Section.getStubSize()); // reserved2 - if (Is64Bit) - Write32(0); // reserved3 - - assert(OS.tell() - Start == Is64Bit ? Section64Size : Section32Size); - } - - void WriteSymtabLoadCommand(uint32_t SymbolOffset, uint32_t NumSymbols, - uint32_t StringTableOffset, - uint32_t StringTableSize) { - // struct symtab_command (24 bytes) - - uint64_t Start = OS.tell(); - (void) Start; - - Write32(LCT_Symtab); - Write32(SymtabLoadCommandSize); - Write32(SymbolOffset); - Write32(NumSymbols); - Write32(StringTableOffset); - Write32(StringTableSize); - - assert(OS.tell() - Start == SymtabLoadCommandSize); - } - - void WriteDysymtabLoadCommand(uint32_t FirstLocalSymbol, - uint32_t NumLocalSymbols, - uint32_t FirstExternalSymbol, - uint32_t NumExternalSymbols, - uint32_t FirstUndefinedSymbol, - uint32_t NumUndefinedSymbols, - uint32_t IndirectSymbolOffset, - uint32_t NumIndirectSymbols) { - // struct dysymtab_command (80 bytes) - - uint64_t Start = OS.tell(); - (void) Start; - - Write32(LCT_Dysymtab); - Write32(DysymtabLoadCommandSize); - Write32(FirstLocalSymbol); - Write32(NumLocalSymbols); - Write32(FirstExternalSymbol); - Write32(NumExternalSymbols); - Write32(FirstUndefinedSymbol); - Write32(NumUndefinedSymbols); - Write32(0); // tocoff - Write32(0); // ntoc - Write32(0); // modtaboff - Write32(0); // nmodtab - Write32(0); // extrefsymoff - Write32(0); // nextrefsyms - Write32(IndirectSymbolOffset); - Write32(NumIndirectSymbols); - Write32(0); // extreloff - Write32(0); // nextrel - Write32(0); // locreloff - Write32(0); // nlocrel - - assert(OS.tell() - Start == DysymtabLoadCommandSize); - } - - void WriteNlist(MachSymbolData &MSD) { - MCSymbolData &Data = *MSD.SymbolData; - const MCSymbol &Symbol = Data.getSymbol(); - uint8_t Type = 0; - uint16_t Flags = Data.getFlags(); - uint32_t Address = 0; - - // Set the N_TYPE bits. See <mach-o/nlist.h>. - // - // FIXME: Are the prebound or indirect fields possible here? - if (Symbol.isUndefined()) - Type = STT_Undefined; - else if (Symbol.isAbsolute()) - Type = STT_Absolute; - else - Type = STT_Section; - - // FIXME: Set STAB bits. - - if (Data.isPrivateExtern()) - Type |= STF_PrivateExtern; - - // Set external bit. - if (Data.isExternal() || Symbol.isUndefined()) - Type |= STF_External; - - // Compute the symbol address. - if (Symbol.isDefined()) { - if (Symbol.isAbsolute()) { - llvm_unreachable("FIXME: Not yet implemented!"); - } else { - Address = Data.getAddress(); - } - } else if (Data.isCommon()) { - // Common symbols are encoded with the size in the address - // field, and their alignment in the flags. - Address = Data.getCommonSize(); - - // Common alignment is packed into the 'desc' bits. - if (unsigned Align = Data.getCommonAlignment()) { - unsigned Log2Size = Log2_32(Align); - assert((1U << Log2Size) == Align && "Invalid 'common' alignment!"); - if (Log2Size > 15) - llvm_report_error("invalid 'common' alignment '" + - Twine(Align) + "'"); - // FIXME: Keep this mask with the SymbolFlags enumeration. - Flags = (Flags & 0xF0FF) | (Log2Size << 8); - } - } - - // struct nlist (12 bytes) - - Write32(MSD.StringIndex); - Write8(Type); - Write8(MSD.SectionIndex); - - // The Mach-O streamer uses the lowest 16-bits of the flags for the 'desc' - // value. - Write16(Flags); - if (Is64Bit) - Write64(Address); - else - Write32(Address); - } - - struct MachRelocationEntry { - uint32_t Word0; - uint32_t Word1; - }; - void ComputeScatteredRelocationInfo(MCAssembler &Asm, MCFragment &Fragment, - MCAsmFixup &Fixup, - const MCValue &Target, - std::vector<MachRelocationEntry> &Relocs) { - uint32_t Address = Fragment.getOffset() + Fixup.Offset; - unsigned IsPCRel = isFixupKindPCRel(Fixup.Kind); - unsigned Log2Size = getFixupKindLog2Size(Fixup.Kind); - unsigned Type = RIT_Vanilla; - - // See <reloc.h>. - const MCSymbol *A = Target.getSymA(); - MCSymbolData *A_SD = &Asm.getSymbolData(*A); - - if (!A_SD->getFragment()) - llvm_report_error("symbol '" + A->getName() + - "' can not be undefined in a subtraction expression"); - - uint32_t Value = A_SD->getAddress(); - uint32_t Value2 = 0; - - if (const MCSymbol *B = Target.getSymB()) { - MCSymbolData *B_SD = &Asm.getSymbolData(*B); - - if (!B_SD->getFragment()) - llvm_report_error("symbol '" + B->getName() + - "' can not be undefined in a subtraction expression"); - - // Select the appropriate difference relocation type. - // - // Note that there is no longer any semantic difference between these two - // relocation types from the linkers point of view, this is done solely - // for pedantic compatibility with 'as'. - Type = A_SD->isExternal() ? RIT_Difference : RIT_LocalDifference; - Value2 = B_SD->getAddress(); - } - - MachRelocationEntry MRE; - MRE.Word0 = ((Address << 0) | - (Type << 24) | - (Log2Size << 28) | - (IsPCRel << 30) | - RF_Scattered); - MRE.Word1 = Value; - Relocs.push_back(MRE); - - if (Type == RIT_Difference || Type == RIT_LocalDifference) { - MachRelocationEntry MRE; - MRE.Word0 = ((0 << 0) | - (RIT_Pair << 24) | - (Log2Size << 28) | - (IsPCRel << 30) | - RF_Scattered); - MRE.Word1 = Value2; - Relocs.push_back(MRE); - } - } - - void ComputeRelocationInfo(MCAssembler &Asm, MCDataFragment &Fragment, - MCAsmFixup &Fixup, - std::vector<MachRelocationEntry> &Relocs) { - unsigned IsPCRel = isFixupKindPCRel(Fixup.Kind); - unsigned Log2Size = getFixupKindLog2Size(Fixup.Kind); - - // FIXME: Share layout object. - MCAsmLayout Layout(Asm); - - // Evaluate the fixup; if the value was resolved, no relocation is needed. - MCValue Target; - if (Asm.EvaluateFixup(Layout, Fixup, &Fragment, Target, Fixup.FixedValue)) - return; - - // If this is a difference or a defined symbol plus an offset, then we need - // a scattered relocation entry. - uint32_t Offset = Target.getConstant(); - if (IsPCRel) - Offset += 1 << Log2Size; - if (Target.getSymB() || - (Target.getSymA() && !Target.getSymA()->isUndefined() && - Offset)) - return ComputeScatteredRelocationInfo(Asm, Fragment, Fixup, Target, - Relocs); - - // See <reloc.h>. - uint32_t Address = Fragment.getOffset() + Fixup.Offset; - uint32_t Value = 0; - unsigned Index = 0; - unsigned IsExtern = 0; - unsigned Type = 0; - - if (Target.isAbsolute()) { // constant - // SymbolNum of 0 indicates the absolute section. - // - // FIXME: Currently, these are never generated (see code below). I cannot - // find a case where they are actually emitted. - Type = RIT_Vanilla; - Value = 0; - } else { - const MCSymbol *Symbol = Target.getSymA(); - MCSymbolData *SD = &Asm.getSymbolData(*Symbol); - - if (Symbol->isUndefined()) { - IsExtern = 1; - Index = SD->getIndex(); - Value = 0; - } else { - // The index is the section ordinal. - // - // FIXME: O(N) - Index = 1; - MCAssembler::iterator it = Asm.begin(), ie = Asm.end(); - for (; it != ie; ++it, ++Index) - if (&*it == SD->getFragment()->getParent()) - break; - assert(it != ie && "Unable to find section index!"); - Value = SD->getAddress(); - } - - Type = RIT_Vanilla; - } - - // struct relocation_info (8 bytes) - MachRelocationEntry MRE; - MRE.Word0 = Address; - MRE.Word1 = ((Index << 0) | - (IsPCRel << 24) | - (Log2Size << 25) | - (IsExtern << 27) | - (Type << 28)); - Relocs.push_back(MRE); - } - - void BindIndirectSymbols(MCAssembler &Asm) { - // This is the point where 'as' creates actual symbols for indirect symbols - // (in the following two passes). It would be easier for us to do this - // sooner when we see the attribute, but that makes getting the order in the - // symbol table much more complicated than it is worth. - // - // FIXME: Revisit this when the dust settles. - - // Bind non lazy symbol pointers first. - for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(), - ie = Asm.indirect_symbol_end(); it != ie; ++it) { - // FIXME: cast<> support! - const MCSectionMachO &Section = - static_cast<const MCSectionMachO&>(it->SectionData->getSection()); - - if (Section.getType() != MCSectionMachO::S_NON_LAZY_SYMBOL_POINTERS) - continue; - - Asm.getOrCreateSymbolData(*it->Symbol); - } - - // Then lazy symbol pointers and symbol stubs. - for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(), - ie = Asm.indirect_symbol_end(); it != ie; ++it) { - // FIXME: cast<> support! - const MCSectionMachO &Section = - static_cast<const MCSectionMachO&>(it->SectionData->getSection()); - - if (Section.getType() != MCSectionMachO::S_LAZY_SYMBOL_POINTERS && - Section.getType() != MCSectionMachO::S_SYMBOL_STUBS) - continue; - - // Set the symbol type to undefined lazy, but only on construction. - // - // FIXME: Do not hardcode. - bool Created; - MCSymbolData &Entry = Asm.getOrCreateSymbolData(*it->Symbol, &Created); - if (Created) - Entry.setFlags(Entry.getFlags() | 0x0001); - } - } - - /// ComputeSymbolTable - Compute the symbol table data - /// - /// \param StringTable [out] - The string table data. - /// \param StringIndexMap [out] - Map from symbol names to offsets in the - /// string table. - void ComputeSymbolTable(MCAssembler &Asm, SmallString<256> &StringTable, - std::vector<MachSymbolData> &LocalSymbolData, - std::vector<MachSymbolData> &ExternalSymbolData, - std::vector<MachSymbolData> &UndefinedSymbolData) { - // Build section lookup table. - DenseMap<const MCSection*, uint8_t> SectionIndexMap; - unsigned Index = 1; - for (MCAssembler::iterator it = Asm.begin(), - ie = Asm.end(); it != ie; ++it, ++Index) - SectionIndexMap[&it->getSection()] = Index; - assert(Index <= 256 && "Too many sections!"); - - // Index 0 is always the empty string. - StringMap<uint64_t> StringIndexMap; - StringTable += '\x00'; - - // Build the symbol arrays and the string table, but only for non-local - // symbols. - // - // The particular order that we collect the symbols and create the string - // table, then sort the symbols is chosen to match 'as'. Even though it - // doesn't matter for correctness, this is important for letting us diff .o - // files. - for (MCAssembler::symbol_iterator it = Asm.symbol_begin(), - ie = Asm.symbol_end(); it != ie; ++it) { - const MCSymbol &Symbol = it->getSymbol(); - - // Ignore assembler temporaries. - if (it->getSymbol().isTemporary() && - (!it->getFragment() || - !Asm.getBackend().doesSectionRequireSymbols( - it->getFragment()->getParent()->getSection()))) - continue; - - if (!it->isExternal() && !Symbol.isUndefined()) - continue; - - uint64_t &Entry = StringIndexMap[Symbol.getName()]; - if (!Entry) { - Entry = StringTable.size(); - StringTable += Symbol.getName(); - StringTable += '\x00'; - } - - MachSymbolData MSD; - MSD.SymbolData = it; - MSD.StringIndex = Entry; - - if (Symbol.isUndefined()) { - MSD.SectionIndex = 0; - UndefinedSymbolData.push_back(MSD); - } else if (Symbol.isAbsolute()) { - MSD.SectionIndex = 0; - ExternalSymbolData.push_back(MSD); - } else { - MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection()); - assert(MSD.SectionIndex && "Invalid section index!"); - ExternalSymbolData.push_back(MSD); - } - } - - // Now add the data for local symbols. - for (MCAssembler::symbol_iterator it = Asm.symbol_begin(), - ie = Asm.symbol_end(); it != ie; ++it) { - const MCSymbol &Symbol = it->getSymbol(); - - // Ignore assembler temporaries. - if (it->getSymbol().isTemporary() && - (!it->getFragment() || - !Asm.getBackend().doesSectionRequireSymbols( - it->getFragment()->getParent()->getSection()))) - continue; - - if (it->isExternal() || Symbol.isUndefined()) - continue; - - uint64_t &Entry = StringIndexMap[Symbol.getName()]; - if (!Entry) { - Entry = StringTable.size(); - StringTable += Symbol.getName(); - StringTable += '\x00'; - } - - MachSymbolData MSD; - MSD.SymbolData = it; - MSD.StringIndex = Entry; - - if (Symbol.isAbsolute()) { - MSD.SectionIndex = 0; - LocalSymbolData.push_back(MSD); - } else { - MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection()); - assert(MSD.SectionIndex && "Invalid section index!"); - LocalSymbolData.push_back(MSD); - } - } - - // External and undefined symbols are required to be in lexicographic order. - std::sort(ExternalSymbolData.begin(), ExternalSymbolData.end()); - std::sort(UndefinedSymbolData.begin(), UndefinedSymbolData.end()); - - // Set the symbol indices. - Index = 0; - for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i) - LocalSymbolData[i].SymbolData->setIndex(Index++); - for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i) - ExternalSymbolData[i].SymbolData->setIndex(Index++); - for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i) - UndefinedSymbolData[i].SymbolData->setIndex(Index++); - - // The string table is padded to a multiple of 4. - while (StringTable.size() % 4) - StringTable += '\x00'; - } - - void WriteObject(MCAssembler &Asm) { - unsigned NumSections = Asm.size(); - - // Create symbol data for any indirect symbols. - BindIndirectSymbols(Asm); - - // Compute symbol table information. - SmallString<256> StringTable; - std::vector<MachSymbolData> LocalSymbolData; - std::vector<MachSymbolData> ExternalSymbolData; - std::vector<MachSymbolData> UndefinedSymbolData; - unsigned NumSymbols = Asm.symbol_size(); - - // No symbol table command is written if there are no symbols. - if (NumSymbols) - ComputeSymbolTable(Asm, StringTable, LocalSymbolData, ExternalSymbolData, - UndefinedSymbolData); - - // The section data starts after the header, the segment load command (and - // section headers) and the symbol table. - unsigned NumLoadCommands = 1; - uint64_t LoadCommandsSize = Is64Bit ? - SegmentLoadCommand64Size + NumSections * Section64Size : - SegmentLoadCommand32Size + NumSections * Section32Size; - - // Add the symbol table load command sizes, if used. - if (NumSymbols) { - NumLoadCommands += 2; - LoadCommandsSize += SymtabLoadCommandSize + DysymtabLoadCommandSize; - } - - // Compute the total size of the section data, as well as its file size and - // vm size. - uint64_t SectionDataStart = (Is64Bit ? Header64Size : Header32Size) - + LoadCommandsSize; - uint64_t SectionDataSize = 0; - uint64_t SectionDataFileSize = 0; - uint64_t VMSize = 0; - for (MCAssembler::iterator it = Asm.begin(), - ie = Asm.end(); it != ie; ++it) { - MCSectionData &SD = *it; - - VMSize = std::max(VMSize, SD.getAddress() + SD.getSize()); - - if (isVirtualSection(SD.getSection())) - continue; - - SectionDataSize = std::max(SectionDataSize, - SD.getAddress() + SD.getSize()); - SectionDataFileSize = std::max(SectionDataFileSize, - SD.getAddress() + SD.getFileSize()); - } - - // The section data is padded to 4 bytes. - // - // FIXME: Is this machine dependent? - unsigned SectionDataPadding = OffsetToAlignment(SectionDataFileSize, 4); - SectionDataFileSize += SectionDataPadding; - - // Write the prolog, starting with the header and load command... - WriteHeader(NumLoadCommands, LoadCommandsSize, - Asm.getSubsectionsViaSymbols()); - WriteSegmentLoadCommand(NumSections, VMSize, - SectionDataStart, SectionDataSize); - - // ... and then the section headers. - // - // We also compute the section relocations while we do this. Note that - // computing relocation info will also update the fixup to have the correct - // value; this will overwrite the appropriate data in the fragment when it - // is written. - std::vector<MachRelocationEntry> RelocInfos; - uint64_t RelocTableEnd = SectionDataStart + SectionDataFileSize; - for (MCAssembler::iterator it = Asm.begin(), - ie = Asm.end(); it != ie; ++it) { - MCSectionData &SD = *it; - - // The assembler writes relocations in the reverse order they were seen. - // - // FIXME: It is probably more complicated than this. - unsigned NumRelocsStart = RelocInfos.size(); - for (MCSectionData::reverse_iterator it2 = SD.rbegin(), - ie2 = SD.rend(); it2 != ie2; ++it2) - if (MCDataFragment *DF = dyn_cast<MCDataFragment>(&*it2)) - for (unsigned i = 0, e = DF->fixup_size(); i != e; ++i) - ComputeRelocationInfo(Asm, *DF, DF->getFixups()[e - i - 1], - RelocInfos); - - unsigned NumRelocs = RelocInfos.size() - NumRelocsStart; - uint64_t SectionStart = SectionDataStart + SD.getAddress(); - WriteSection(SD, SectionStart, RelocTableEnd, NumRelocs); - RelocTableEnd += NumRelocs * RelocationInfoSize; - } - - // Write the symbol table load command, if used. - if (NumSymbols) { - unsigned FirstLocalSymbol = 0; - unsigned NumLocalSymbols = LocalSymbolData.size(); - unsigned FirstExternalSymbol = FirstLocalSymbol + NumLocalSymbols; - unsigned NumExternalSymbols = ExternalSymbolData.size(); - unsigned FirstUndefinedSymbol = FirstExternalSymbol + NumExternalSymbols; - unsigned NumUndefinedSymbols = UndefinedSymbolData.size(); - unsigned NumIndirectSymbols = Asm.indirect_symbol_size(); - unsigned NumSymTabSymbols = - NumLocalSymbols + NumExternalSymbols + NumUndefinedSymbols; - uint64_t IndirectSymbolSize = NumIndirectSymbols * 4; - uint64_t IndirectSymbolOffset = 0; - - // If used, the indirect symbols are written after the section data. - if (NumIndirectSymbols) - IndirectSymbolOffset = RelocTableEnd; - - // The symbol table is written after the indirect symbol data. - uint64_t SymbolTableOffset = RelocTableEnd + IndirectSymbolSize; - - // The string table is written after symbol table. - uint64_t StringTableOffset = - SymbolTableOffset + NumSymTabSymbols * (Is64Bit ? Nlist64Size : - Nlist32Size); - WriteSymtabLoadCommand(SymbolTableOffset, NumSymTabSymbols, - StringTableOffset, StringTable.size()); - - WriteDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols, - FirstExternalSymbol, NumExternalSymbols, - FirstUndefinedSymbol, NumUndefinedSymbols, - IndirectSymbolOffset, NumIndirectSymbols); - } - - // Write the actual section data. - for (MCAssembler::iterator it = Asm.begin(), ie = Asm.end(); it != ie; ++it) - WriteFileData(OS, *it, *this); - - // Write the extra padding. - WriteZeros(SectionDataPadding); - - // Write the relocation entries. - for (unsigned i = 0, e = RelocInfos.size(); i != e; ++i) { - Write32(RelocInfos[i].Word0); - Write32(RelocInfos[i].Word1); - } - - // Write the symbol table data, if used. - if (NumSymbols) { - // Write the indirect symbol entries. - for (MCAssembler::indirect_symbol_iterator - it = Asm.indirect_symbol_begin(), - ie = Asm.indirect_symbol_end(); it != ie; ++it) { - // Indirect symbols in the non lazy symbol pointer section have some - // special handling. - const MCSectionMachO &Section = - static_cast<const MCSectionMachO&>(it->SectionData->getSection()); - if (Section.getType() == MCSectionMachO::S_NON_LAZY_SYMBOL_POINTERS) { - // If this symbol is defined and internal, mark it as such. - if (it->Symbol->isDefined() && - !Asm.getSymbolData(*it->Symbol).isExternal()) { - uint32_t Flags = ISF_Local; - if (it->Symbol->isAbsolute()) - Flags |= ISF_Absolute; - Write32(Flags); - continue; - } - } - - Write32(Asm.getSymbolData(*it->Symbol).getIndex()); - } - - // FIXME: Check that offsets match computed ones. - - // Write the symbol table entries. - for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i) - WriteNlist(LocalSymbolData[i]); - for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i) - WriteNlist(ExternalSymbolData[i]); - for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i) - WriteNlist(UndefinedSymbolData[i]); - - // Write the string table. - OS << StringTable.str(); - } - } - - void ApplyFixup(const MCAsmFixup &Fixup, MCDataFragment &DF) { - unsigned Size = 1 << getFixupKindLog2Size(Fixup.Kind); - - // FIXME: Endianness assumption. - assert(Fixup.Offset + Size <= DF.getContents().size() && - "Invalid fixup offset!"); - for (unsigned i = 0; i != Size; ++i) - DF.getContents()[Fixup.Offset + i] = uint8_t(Fixup.FixedValue >> (i * 8)); - } -}; - /* *** */ MCFragment::MCFragment() : Kind(FragmentType(~0)) { @@ -1008,14 +93,149 @@ MCSymbolData::MCSymbolData(const MCSymbol &_Symbol, MCFragment *_Fragment, /* *** */ MCAssembler::MCAssembler(MCContext &_Context, TargetAsmBackend &_Backend, - raw_ostream &_OS) - : Context(_Context), Backend(_Backend), OS(_OS), SubsectionsViaSymbols(false) + MCCodeEmitter &_Emitter, raw_ostream &_OS) + : Context(_Context), Backend(_Backend), Emitter(_Emitter), + OS(_OS), SubsectionsViaSymbols(false) { } MCAssembler::~MCAssembler() { } +static bool isScatteredFixupFullyResolvedSimple(const MCAssembler &Asm, + const MCAsmFixup &Fixup, + const MCDataFragment *DF, + const MCValue Target, + const MCSection *BaseSection) { + // The effective fixup address is + // addr(atom(A)) + offset(A) + // - addr(atom(B)) - offset(B) + // - addr(<base symbol>) + <fixup offset from base symbol> + // and the offsets are not relocatable, so the fixup is fully resolved when + // addr(atom(A)) - addr(atom(B)) - addr(<base symbol>)) == 0. + // + // The simple (Darwin, except on x86_64) way of dealing with this was to + // assume that any reference to a temporary symbol *must* be a temporary + // symbol in the same atom, unless the sections differ. Therefore, any PCrel + // relocation to a temporary symbol (in the same section) is fully + // resolved. This also works in conjunction with absolutized .set, which + // requires the compiler to use .set to absolutize the differences between + // symbols which the compiler knows to be assembly time constants, so we don't + // need to worry about consider symbol differences fully resolved. + + // Non-relative fixups are only resolved if constant. + if (!BaseSection) + return Target.isAbsolute(); + + // Otherwise, relative fixups are only resolved if not a difference and the + // target is a temporary in the same section. + if (Target.isAbsolute() || Target.getSymB()) + return false; + + const MCSymbol *A = &Target.getSymA()->getSymbol(); + if (!A->isTemporary() || !A->isInSection() || + &A->getSection() != BaseSection) + return false; + + return true; +} + +static bool isScatteredFixupFullyResolved(const MCAssembler &Asm, + const MCAsmFixup &Fixup, + const MCDataFragment *DF, + const MCValue Target, + const MCSymbolData *BaseSymbol) { + // The effective fixup address is + // addr(atom(A)) + offset(A) + // - addr(atom(B)) - offset(B) + // - addr(BaseSymbol) + <fixup offset from base symbol> + // and the offsets are not relocatable, so the fixup is fully resolved when + // addr(atom(A)) - addr(atom(B)) - addr(BaseSymbol) == 0. + // + // Note that "false" is almost always conservatively correct (it means we emit + // a relocation which is unnecessary), except when it would force us to emit a + // relocation which the target cannot encode. + + const MCSymbolData *A_Base = 0, *B_Base = 0; + if (const MCSymbolRefExpr *A = Target.getSymA()) { + // Modified symbol references cannot be resolved. + if (A->getKind() != MCSymbolRefExpr::VK_None) + return false; + + A_Base = Asm.getAtom(&Asm.getSymbolData(A->getSymbol())); + if (!A_Base) + return false; + } + + if (const MCSymbolRefExpr *B = Target.getSymB()) { + // Modified symbol references cannot be resolved. + if (B->getKind() != MCSymbolRefExpr::VK_None) + return false; + + B_Base = Asm.getAtom(&Asm.getSymbolData(B->getSymbol())); + if (!B_Base) + return false; + } + + // If there is no base, A and B have to be the same atom for this fixup to be + // fully resolved. + if (!BaseSymbol) + return A_Base == B_Base; + + // Otherwise, B must be missing and A must be the base. + return !B_Base && BaseSymbol == A_Base; +} + +bool MCAssembler::isSymbolLinkerVisible(const MCSymbolData *SD) const { + // Non-temporary labels should always be visible to the linker. + if (!SD->getSymbol().isTemporary()) + return true; + + // Absolute temporary labels are never visible. + if (!SD->getFragment()) + return false; + + // Otherwise, check if the section requires symbols even for temporary labels. + return getBackend().doesSectionRequireSymbols( + SD->getFragment()->getParent()->getSection()); +} + +const MCSymbolData *MCAssembler::getAtomForAddress(const MCSectionData *Section, + uint64_t Address) const { + const MCSymbolData *Best = 0; + for (MCAssembler::const_symbol_iterator it = symbol_begin(), + ie = symbol_end(); it != ie; ++it) { + // Ignore non-linker visible symbols. + if (!isSymbolLinkerVisible(it)) + continue; + + // Ignore symbols not in the same section. + if (!it->getFragment() || it->getFragment()->getParent() != Section) + continue; + + // Otherwise, find the closest symbol preceding this address (ties are + // resolved in favor of the last defined symbol). + if (it->getAddress() <= Address && + (!Best || it->getAddress() >= Best->getAddress())) + Best = it; + } + + return Best; +} + +const MCSymbolData *MCAssembler::getAtom(const MCSymbolData *SD) const { + // Linker visible symbols define atoms. + if (isSymbolLinkerVisible(SD)) + return SD; + + // Absolute and undefined symbols have no defining atom. + if (!SD->getFragment()) + return 0; + + // Otherwise, search by address. + return getAtomForAddress(SD->getFragment()->getParent(), SD->getAddress()); +} + bool MCAssembler::EvaluateFixup(const MCAsmLayout &Layout, MCAsmFixup &Fixup, MCDataFragment *DF, MCValue &Target, uint64_t &Value) const { @@ -1028,34 +248,47 @@ bool MCAssembler::EvaluateFixup(const MCAsmLayout &Layout, MCAsmFixup &Fixup, Value = Target.getConstant(); - // FIXME: This "resolved" check isn't quite right. The assumption is that if - // we have a PCrel access to a temporary, then that temporary is in the same - // atom, and so the value is resolved. We need explicit atom's to implement - // this more precisely. - bool IsResolved = true, IsPCRel = isFixupKindPCRel(Fixup.Kind); - if (const MCSymbol *Symbol = Target.getSymA()) { - if (Symbol->isDefined()) - Value += getSymbolData(*Symbol).getAddress(); + bool IsPCRel = + Emitter.getFixupKindInfo(Fixup.Kind).Flags & MCFixupKindInfo::FKF_IsPCRel; + bool IsResolved = true; + if (const MCSymbolRefExpr *A = Target.getSymA()) { + if (A->getSymbol().isDefined()) + Value += getSymbolData(A->getSymbol()).getAddress(); else IsResolved = false; - - // With scattered symbols, we assume anything that isn't a PCrel temporary - // access can have an arbitrary value. - if (getBackend().hasScatteredSymbols() && - (!IsPCRel || !Symbol->isTemporary())) - IsResolved = false; } - if (const MCSymbol *Symbol = Target.getSymB()) { - if (Symbol->isDefined()) - Value -= getSymbolData(*Symbol).getAddress(); + if (const MCSymbolRefExpr *B = Target.getSymB()) { + if (B->getSymbol().isDefined()) + Value -= getSymbolData(B->getSymbol()).getAddress(); else IsResolved = false; + } - // With scattered symbols, we assume anything that isn't a PCrel temporary - // access can have an arbitrary value. - if (getBackend().hasScatteredSymbols() && - (!IsPCRel || !Symbol->isTemporary())) - IsResolved = false; + // If we are using scattered symbols, determine whether this value is actually + // resolved; scattering may cause atoms to move. + if (IsResolved && getBackend().hasScatteredSymbols()) { + if (getBackend().hasReliableSymbolDifference()) { + // If this is a PCrel relocation, find the base atom (identified by its + // symbol) that the fixup value is relative to. + const MCSymbolData *BaseSymbol = 0; + if (IsPCRel) { + BaseSymbol = getAtomForAddress( + DF->getParent(), DF->getAddress() + Fixup.Offset); + if (!BaseSymbol) + IsResolved = false; + } + + if (IsResolved) + IsResolved = isScatteredFixupFullyResolved(*this, Fixup, DF, Target, + BaseSymbol); + } else { + const MCSection *BaseSection = 0; + if (IsPCRel) + BaseSection = &DF->getParent()->getSection(); + + IsResolved = isScatteredFixupFullyResolvedSimple(*this, Fixup, DF, Target, + BaseSection); + } } if (IsPCRel) @@ -1127,7 +360,7 @@ void MCAssembler::LayoutSection(MCSectionData &SD) { // Set the section sizes. SD.setSize(Address - SD.getAddress()); - if (isVirtualSection(SD.getSection())) + if (getBackend().isVirtualSection(SD.getSection())) SD.setFileSize(0); else SD.setFileSize(Address - SD.getAddress()); @@ -1138,7 +371,7 @@ void MCAssembler::LayoutSection(MCSectionData &SD) { /// the \arg Count is more than the maximum optimal nops. /// /// FIXME this is X86 32-bit specific and should move to a better place. -static uint64_t WriteNopData(uint64_t Count, MachObjectWriter &MOW) { +static uint64_t WriteNopData(uint64_t Count, MCObjectWriter *OW) { static const uint8_t Nops[16][16] = { // nop {0x90}, @@ -1186,15 +419,14 @@ static uint64_t WriteNopData(uint64_t Count, MachObjectWriter &MOW) { return 0; for (uint64_t i = 0; i < Count; i++) - MOW.Write8 (uint8_t(Nops[Count - 1][i])); + OW->Write8(uint8_t(Nops[Count - 1][i])); return Count; } -/// WriteFileData - Write the \arg F data to the output file. -static void WriteFileData(raw_ostream &OS, const MCFragment &F, - MachObjectWriter &MOW) { - uint64_t Start = OS.tell(); +/// WriteFragmentData - Write the \arg F data to the output file. +static void WriteFragmentData(const MCFragment &F, MCObjectWriter *OW) { + uint64_t Start = OW->getStream().tell(); (void) Start; ++EmittedFragments; @@ -1218,7 +450,7 @@ static void WriteFileData(raw_ostream &OS, const MCFragment &F, // the Count bytes. Then if that did not fill any bytes or there are any // bytes left to fill use the the Value and ValueSize to fill the rest. if (AF.getEmitNops()) { - uint64_t NopByteCount = WriteNopData(Count, MOW); + uint64_t NopByteCount = WriteNopData(Count, OW); Count -= NopByteCount; } @@ -1226,26 +458,17 @@ static void WriteFileData(raw_ostream &OS, const MCFragment &F, switch (AF.getValueSize()) { default: assert(0 && "Invalid size!"); - case 1: MOW.Write8 (uint8_t (AF.getValue())); break; - case 2: MOW.Write16(uint16_t(AF.getValue())); break; - case 4: MOW.Write32(uint32_t(AF.getValue())); break; - case 8: MOW.Write64(uint64_t(AF.getValue())); break; + case 1: OW->Write8 (uint8_t (AF.getValue())); break; + case 2: OW->Write16(uint16_t(AF.getValue())); break; + case 4: OW->Write32(uint32_t(AF.getValue())); break; + case 8: OW->Write64(uint64_t(AF.getValue())); break; } } break; } case MCFragment::FT_Data: { - MCDataFragment &DF = cast<MCDataFragment>(F); - - // Apply the fixups. - // - // FIXME: Move elsewhere. - for (MCDataFragment::const_fixup_iterator it = DF.fixup_begin(), - ie = DF.fixup_end(); it != ie; ++it) - MOW.ApplyFixup(*it, DF); - - OS << cast<MCDataFragment>(F).getContents().str(); + OW->WriteBytes(cast<MCDataFragment>(F).getContents().str()); break; } @@ -1255,10 +478,10 @@ static void WriteFileData(raw_ostream &OS, const MCFragment &F, switch (FF.getValueSize()) { default: assert(0 && "Invalid size!"); - case 1: MOW.Write8 (uint8_t (FF.getValue())); break; - case 2: MOW.Write16(uint16_t(FF.getValue())); break; - case 4: MOW.Write32(uint32_t(FF.getValue())); break; - case 8: MOW.Write64(uint64_t(FF.getValue())); break; + case 1: OW->Write8 (uint8_t (FF.getValue())); break; + case 2: OW->Write16(uint16_t(FF.getValue())); break; + case 4: OW->Write32(uint32_t(FF.getValue())); break; + case 8: OW->Write64(uint64_t(FF.getValue())); break; } } break; @@ -1268,7 +491,7 @@ static void WriteFileData(raw_ostream &OS, const MCFragment &F, MCOrgFragment &OF = cast<MCOrgFragment>(F); for (uint64_t i = 0, e = OF.getFileSize(); i != e; ++i) - MOW.Write8(uint8_t(OF.getValue())); + OW->Write8(uint8_t(OF.getValue())); break; } @@ -1279,30 +502,29 @@ static void WriteFileData(raw_ostream &OS, const MCFragment &F, } } - assert(OS.tell() - Start == F.getFileSize()); + assert(OW->getStream().tell() - Start == F.getFileSize()); } -/// WriteFileData - Write the \arg SD data to the output file. -static void WriteFileData(raw_ostream &OS, const MCSectionData &SD, - MachObjectWriter &MOW) { +void MCAssembler::WriteSectionData(const MCSectionData *SD, + MCObjectWriter *OW) const { // Ignore virtual sections. - if (isVirtualSection(SD.getSection())) { - assert(SD.getFileSize() == 0); + if (getBackend().isVirtualSection(SD->getSection())) { + assert(SD->getFileSize() == 0); return; } - uint64_t Start = OS.tell(); + uint64_t Start = OW->getStream().tell(); (void) Start; - for (MCSectionData::const_iterator it = SD.begin(), - ie = SD.end(); it != ie; ++it) - WriteFileData(OS, *it, MOW); + for (MCSectionData::const_iterator it = SD->begin(), + ie = SD->end(); it != ie; ++it) + WriteFragmentData(*it, OW); // Add section padding. - assert(SD.getFileSize() >= SD.getSize() && "Invalid section sizes!"); - MOW.WriteZeros(SD.getFileSize() - SD.getSize()); + assert(SD->getFileSize() >= SD->getSize() && "Invalid section sizes!"); + OW->WriteZeros(SD->getFileSize() - SD->getSize()); - assert(OS.tell() - Start == SD.getFileSize()); + assert(OW->getStream().tell() - Start == SD->getFileSize()); } void MCAssembler::Finish() { @@ -1318,13 +540,47 @@ void MCAssembler::Finish() { llvm::errs() << "assembler backend - post-layout\n--\n"; dump(); }); - // Write the object file. - // // FIXME: Factor out MCObjectWriter. - bool Is64Bit = StringRef(getBackend().getTarget().getName()) == "x86-64"; - MachObjectWriter MOW(OS, Is64Bit); - MOW.WriteObject(*this); + llvm::OwningPtr<MCObjectWriter> Writer(getBackend().createObjectWriter(OS)); + if (!Writer) + llvm_report_error("unable to create object writer!"); + + // Allow the object writer a chance to perform post-layout binding (for + // example, to set the index fields in the symbol data). + Writer->ExecutePostLayoutBinding(*this); + + // Evaluate and apply the fixups, generating relocation entries as necessary. + // + // FIXME: Share layout object. + MCAsmLayout Layout(*this); + for (MCAssembler::iterator it = begin(), ie = end(); it != ie; ++it) { + for (MCSectionData::iterator it2 = it->begin(), + ie2 = it->end(); it2 != ie2; ++it2) { + MCDataFragment *DF = dyn_cast<MCDataFragment>(it2); + if (!DF) + continue; + + for (MCDataFragment::fixup_iterator it3 = DF->fixup_begin(), + ie3 = DF->fixup_end(); it3 != ie3; ++it3) { + MCAsmFixup &Fixup = *it3; + // Evaluate the fixup. + MCValue Target; + uint64_t FixedValue; + if (!EvaluateFixup(Layout, Fixup, DF, Target, FixedValue)) { + // The fixup was unresolved, we need a relocation. Inform the object + // writer of the relocation, and give it an opportunity to adjust the + // fixup value if need be. + Writer->RecordRelocation(*this, *DF, Fixup, Target, FixedValue); + } + + getBackend().ApplyFixup(Fixup, *DF, FixedValue); + } + } + } + + // Write the object file. + Writer->WriteObject(*this); OS.flush(); } @@ -1354,7 +610,7 @@ bool MCAssembler::LayoutOnce() { MCSectionData &SD = *it; // Skip virtual sections. - if (isVirtualSection(SD.getSection())) + if (getBackend().isVirtualSection(SD.getSection())) continue; // Align this section if necessary by adding padding bytes to the previous @@ -1377,7 +633,7 @@ bool MCAssembler::LayoutOnce() { for (iterator it = begin(), ie = end(); it != ie; ++it) { MCSectionData &SD = *it; - if (!isVirtualSection(SD.getSection())) + if (!getBackend().isVirtualSection(SD.getSection())) continue; // Align this section if necessary by adding padding bytes to the previous diff --git a/lib/MC/MCCodeEmitter.cpp b/lib/MC/MCCodeEmitter.cpp index accb06c..d513237 100644 --- a/lib/MC/MCCodeEmitter.cpp +++ b/lib/MC/MCCodeEmitter.cpp @@ -19,10 +19,10 @@ MCCodeEmitter::~MCCodeEmitter() { const MCFixupKindInfo &MCCodeEmitter::getFixupKindInfo(MCFixupKind Kind) const { static const MCFixupKindInfo Builtins[] = { - { "FK_Data_1", 0, 8 }, - { "FK_Data_2", 0, 16 }, - { "FK_Data_4", 0, 32 }, - { "FK_Data_8", 0, 64 } + { "FK_Data_1", 0, 8, 0 }, + { "FK_Data_2", 0, 16, 0 }, + { "FK_Data_4", 0, 32, 0 }, + { "FK_Data_8", 0, 64, 0 } }; assert(Kind <= 3 && "Unknown fixup kind"); diff --git a/lib/MC/MCContext.cpp b/lib/MC/MCContext.cpp index 70c89a2..37e8282 100644 --- a/lib/MC/MCContext.cpp +++ b/lib/MC/MCContext.cpp @@ -49,17 +49,6 @@ MCSymbol *MCContext::CreateTempSymbol() { "tmp" + Twine(NextUniqueID++)); } - -MCSymbol *MCContext::GetOrCreateTemporarySymbol(StringRef Name) { - // If there is no name, create a new anonymous symbol. - // FIXME: Remove this. This form of the method should always take a name. - if (Name.empty()) - return GetOrCreateTemporarySymbol(Twine(MAI.getPrivateGlobalPrefix()) + - "tmp" + Twine(NextUniqueID++)); - - return GetOrCreateSymbol(Name, true); -} - MCSymbol *MCContext::GetOrCreateTemporarySymbol(const Twine &Name) { SmallString<128> NameSV; Name.toVector(NameSV); diff --git a/lib/MC/MCExpr.cpp b/lib/MC/MCExpr.cpp index a2ed20b..2759944 100644 --- a/lib/MC/MCExpr.cpp +++ b/lib/MC/MCExpr.cpp @@ -30,7 +30,7 @@ void MCExpr::print(raw_ostream &OS) const { case MCExpr::SymbolRef: { const MCSymbolRefExpr &SRE = cast<MCSymbolRefExpr>(*this); const MCSymbol &Sym = SRE.getSymbol(); - + // Parenthesize names that start with $ so that they don't look like // absolute names. if (Sym.getName()[0] == '$') @@ -59,14 +59,14 @@ void MCExpr::print(raw_ostream &OS) const { case MCExpr::Binary: { const MCBinaryExpr &BE = cast<MCBinaryExpr>(*this); - + // Only print parens around the LHS if it is non-trivial. if (isa<MCConstantExpr>(BE.getLHS()) || isa<MCSymbolRefExpr>(BE.getLHS())) { OS << *BE.getLHS(); } else { OS << '(' << *BE.getLHS() << ')'; } - + switch (BE.getOpcode()) { default: assert(0 && "Invalid opcode!"); case MCBinaryExpr::Add: @@ -77,7 +77,7 @@ void MCExpr::print(raw_ostream &OS) const { return; } } - + OS << '+'; break; case MCBinaryExpr::And: OS << '&'; break; @@ -98,7 +98,7 @@ void MCExpr::print(raw_ostream &OS) const { case MCBinaryExpr::Sub: OS << '-'; break; case MCBinaryExpr::Xor: OS << '^'; break; } - + // Only print parens around the LHS if it is non-trivial. if (isa<MCConstantExpr>(BE.getRHS()) || isa<MCSymbolRefExpr>(BE.getRHS())) { OS << *BE.getRHS(); @@ -193,7 +193,7 @@ void MCTargetExpr::Anchor() {} bool MCExpr::EvaluateAsAbsolute(int64_t &Res, const MCAsmLayout *Layout) const { MCValue Value; - + if (!EvaluateAsRelocatable(Value, Layout) || !Value.isAbsolute()) return false; @@ -201,16 +201,16 @@ bool MCExpr::EvaluateAsAbsolute(int64_t &Res, const MCAsmLayout *Layout) const { return true; } -static bool EvaluateSymbolicAdd(const MCValue &LHS, const MCSymbol *RHS_A, - const MCSymbol *RHS_B, int64_t RHS_Cst, +static bool EvaluateSymbolicAdd(const MCValue &LHS,const MCSymbolRefExpr *RHS_A, + const MCSymbolRefExpr *RHS_B, int64_t RHS_Cst, MCValue &Res) { // We can't add or subtract two symbols. if ((LHS.getSymA() && RHS_A) || (LHS.getSymB() && RHS_B)) return false; - const MCSymbol *A = LHS.getSymA() ? LHS.getSymA() : RHS_A; - const MCSymbol *B = LHS.getSymB() ? LHS.getSymB() : RHS_B; + const MCSymbolRefExpr *A = LHS.getSymA() ? LHS.getSymA() : RHS_A; + const MCSymbolRefExpr *B = LHS.getSymB() ? LHS.getSymB() : RHS_B; if (B) { // If we have a negated symbol, then we must have also have a non-negated // symbol in order to encode the expression. We can do this check later to @@ -228,13 +228,14 @@ bool MCExpr::EvaluateAsRelocatable(MCValue &Res, switch (getKind()) { case Target: return cast<MCTargetExpr>(this)->EvaluateAsRelocatableImpl(Res, Layout); - + case Constant: Res = MCValue::get(cast<MCConstantExpr>(this)->getValue()); return true; case SymbolRef: { - const MCSymbol &Sym = cast<MCSymbolRefExpr>(this)->getSymbol(); + const MCSymbolRefExpr *SRE = cast<MCSymbolRefExpr>(this); + const MCSymbol &Sym = SRE->getSymbol(); // Evaluate recursively if this is a variable. if (Sym.isVariable()) { @@ -245,9 +246,12 @@ bool MCExpr::EvaluateAsRelocatable(MCValue &Res, // layout object and the target requests it. if (Layout && Res.getSymB() && Layout->getAssembler().getBackend().hasAbsolutizedSet() && - Res.getSymA()->isDefined() && Res.getSymB()->isDefined()) { - MCSymbolData &A = Layout->getAssembler().getSymbolData(*Res.getSymA()); - MCSymbolData &B = Layout->getAssembler().getSymbolData(*Res.getSymB()); + Res.getSymA()->getSymbol().isDefined() && + Res.getSymB()->getSymbol().isDefined()) { + MCSymbolData &A = + Layout->getAssembler().getSymbolData(Res.getSymA()->getSymbol()); + MCSymbolData &B = + Layout->getAssembler().getSymbolData(Res.getSymB()->getSymbol()); Res = MCValue::get(+ A.getFragment()->getAddress() + A.getOffset() - B.getFragment()->getAddress() - B.getOffset() + Res.getConstant()); @@ -256,7 +260,7 @@ bool MCExpr::EvaluateAsRelocatable(MCValue &Res, return true; } - Res = MCValue::get(&Sym, 0, 0); + Res = MCValue::get(SRE, 0, 0); return true; } @@ -277,13 +281,13 @@ bool MCExpr::EvaluateAsRelocatable(MCValue &Res, /// -(a - b + const) ==> (b - a - const) if (Value.getSymA() && !Value.getSymB()) return false; - Res = MCValue::get(Value.getSymB(), Value.getSymA(), - -Value.getConstant()); + Res = MCValue::get(Value.getSymB(), Value.getSymA(), + -Value.getConstant()); break; case MCUnaryExpr::Not: if (!Value.isAbsolute()) return false; - Res = MCValue::get(~Value.getConstant()); + Res = MCValue::get(~Value.getConstant()); break; case MCUnaryExpr::Plus: Res = Value; @@ -296,7 +300,7 @@ bool MCExpr::EvaluateAsRelocatable(MCValue &Res, case Binary: { const MCBinaryExpr *ABE = cast<MCBinaryExpr>(this); MCValue LHSValue, RHSValue; - + if (!ABE->getLHS()->EvaluateAsRelocatable(LHSValue, Layout) || !ABE->getRHS()->EvaluateAsRelocatable(RHSValue, Layout)) return false; diff --git a/lib/MC/MCMachOStreamer.cpp b/lib/MC/MCMachOStreamer.cpp index 73b1074..9504392 100644 --- a/lib/MC/MCMachOStreamer.cpp +++ b/lib/MC/MCMachOStreamer.cpp @@ -45,7 +45,6 @@ class MCMachOStreamer : public MCStreamer { private: MCAssembler Assembler; - MCCodeEmitter *Emitter; MCSectionData *CurSectionData; private: @@ -61,7 +60,7 @@ private: public: MCMachOStreamer(MCContext &Context, TargetAsmBackend &TAB, raw_ostream &_OS, MCCodeEmitter *_Emitter) - : MCStreamer(Context), Assembler(Context, TAB, _OS), Emitter(_Emitter), + : MCStreamer(Context), Assembler(Context, TAB, *_Emitter, _OS), CurSectionData(0) {} ~MCMachOStreamer() {} @@ -370,15 +369,12 @@ void MCMachOStreamer::EmitInstruction(const MCInst &Inst) { if (Inst.getOperand(i).isExpr()) AddValueSymbols(Inst.getOperand(i).getExpr()); - if (!Emitter) - llvm_unreachable("no code emitter available!"); - CurSectionData->setHasInstructions(true); SmallVector<MCFixup, 4> Fixups; SmallString<256> Code; raw_svector_ostream VecOS(Code); - Emitter->EncodeInstruction(Inst, VecOS, Fixups); + Assembler.getEmitter().EncodeInstruction(Inst, VecOS, Fixups); VecOS.flush(); // Add the fixups and data. diff --git a/lib/MC/MCObjectWriter.cpp b/lib/MC/MCObjectWriter.cpp new file mode 100644 index 0000000..d117e82 --- /dev/null +++ b/lib/MC/MCObjectWriter.cpp @@ -0,0 +1,15 @@ +//===- lib/MC/MCObjectWriter.cpp - MCObjectWriter implementation ----------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// + +#include "llvm/MC/MCObjectWriter.h" + +using namespace llvm; + +MCObjectWriter::~MCObjectWriter() { +} diff --git a/lib/MC/MCValue.cpp b/lib/MC/MCValue.cpp index 043a49d..c6ea16c 100644 --- a/lib/MC/MCValue.cpp +++ b/lib/MC/MCValue.cpp @@ -8,6 +8,7 @@ //===----------------------------------------------------------------------===// #include "llvm/MC/MCValue.h" +#include "llvm/MC/MCExpr.h" #include "llvm/Support/Debug.h" #include "llvm/Support/raw_ostream.h" @@ -19,10 +20,12 @@ void MCValue::print(raw_ostream &OS, const MCAsmInfo *MAI) const { return; } - OS << *getSymA(); + getSymA()->print(OS); - if (getSymB()) - OS << " - " << *getSymB(); + if (getSymB()) { + OS << " - "; + getSymB()->print(OS); + } if (getConstant()) OS << " + " << getConstant(); diff --git a/lib/MC/MachObjectWriter.cpp b/lib/MC/MachObjectWriter.cpp new file mode 100644 index 0000000..4b08c22 --- /dev/null +++ b/lib/MC/MachObjectWriter.cpp @@ -0,0 +1,1109 @@ +//===- lib/MC/MachObjectWriter.cpp - Mach-O File Writer -------------------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// + +#include "llvm/MC/MachObjectWriter.h" +#include "llvm/ADT/StringMap.h" +#include "llvm/ADT/Twine.h" +#include "llvm/MC/MCAssembler.h" +#include "llvm/MC/MCExpr.h" +#include "llvm/MC/MCObjectWriter.h" +#include "llvm/MC/MCSectionMachO.h" +#include "llvm/MC/MCSymbol.h" +#include "llvm/MC/MCValue.h" +#include "llvm/Support/ErrorHandling.h" +#include "llvm/Support/MachO.h" +#include "llvm/Target/TargetAsmBackend.h" + +// FIXME: Gross. +#include "../Target/X86/X86FixupKinds.h" + +#include <vector> +using namespace llvm; + +static unsigned getFixupKindLog2Size(unsigned Kind) { + switch (Kind) { + default: llvm_unreachable("invalid fixup kind!"); + case X86::reloc_pcrel_1byte: + case FK_Data_1: return 0; + case FK_Data_2: return 1; + case X86::reloc_pcrel_4byte: + case X86::reloc_riprel_4byte: + case X86::reloc_riprel_4byte_movq_load: + case FK_Data_4: return 2; + case FK_Data_8: return 3; + } +} + +static bool isFixupKindPCRel(unsigned Kind) { + switch (Kind) { + default: + return false; + case X86::reloc_pcrel_1byte: + case X86::reloc_pcrel_4byte: + case X86::reloc_riprel_4byte: + case X86::reloc_riprel_4byte_movq_load: + return true; + } +} + +static bool isFixupKindRIPRel(unsigned Kind) { + return Kind == X86::reloc_riprel_4byte || + Kind == X86::reloc_riprel_4byte_movq_load; +} + +namespace { + +class MachObjectWriterImpl { + // See <mach-o/loader.h>. + enum { + Header_Magic32 = 0xFEEDFACE, + Header_Magic64 = 0xFEEDFACF + }; + + enum { + Header32Size = 28, + Header64Size = 32, + SegmentLoadCommand32Size = 56, + SegmentLoadCommand64Size = 72, + Section32Size = 68, + Section64Size = 80, + SymtabLoadCommandSize = 24, + DysymtabLoadCommandSize = 80, + Nlist32Size = 12, + Nlist64Size = 16, + RelocationInfoSize = 8 + }; + + enum HeaderFileType { + HFT_Object = 0x1 + }; + + enum HeaderFlags { + HF_SubsectionsViaSymbols = 0x2000 + }; + + enum LoadCommandType { + LCT_Segment = 0x1, + LCT_Symtab = 0x2, + LCT_Dysymtab = 0xb, + LCT_Segment64 = 0x19 + }; + + // See <mach-o/nlist.h>. + enum SymbolTypeType { + STT_Undefined = 0x00, + STT_Absolute = 0x02, + STT_Section = 0x0e + }; + + enum SymbolTypeFlags { + // If any of these bits are set, then the entry is a stab entry number (see + // <mach-o/stab.h>. Otherwise the other masks apply. + STF_StabsEntryMask = 0xe0, + + STF_TypeMask = 0x0e, + STF_External = 0x01, + STF_PrivateExtern = 0x10 + }; + + /// IndirectSymbolFlags - Flags for encoding special values in the indirect + /// symbol entry. + enum IndirectSymbolFlags { + ISF_Local = 0x80000000, + ISF_Absolute = 0x40000000 + }; + + /// RelocationFlags - Special flags for addresses. + enum RelocationFlags { + RF_Scattered = 0x80000000 + }; + + enum RelocationInfoType { + RIT_Vanilla = 0, + RIT_Pair = 1, + RIT_Difference = 2, + RIT_PreboundLazyPointer = 3, + RIT_LocalDifference = 4 + }; + + /// X86_64 uses its own relocation types. + enum RelocationInfoTypeX86_64 { + RIT_X86_64_Unsigned = 0, + RIT_X86_64_Signed = 1, + RIT_X86_64_Branch = 2, + RIT_X86_64_GOTLoad = 3, + RIT_X86_64_GOT = 4, + RIT_X86_64_Subtractor = 5, + RIT_X86_64_Signed1 = 6, + RIT_X86_64_Signed2 = 7, + RIT_X86_64_Signed4 = 8 + }; + + /// MachSymbolData - Helper struct for containing some precomputed information + /// on symbols. + struct MachSymbolData { + MCSymbolData *SymbolData; + uint64_t StringIndex; + uint8_t SectionIndex; + + // Support lexicographic sorting. + bool operator<(const MachSymbolData &RHS) const { + const std::string &Name = SymbolData->getSymbol().getName(); + return Name < RHS.SymbolData->getSymbol().getName(); + } + }; + + /// @name Relocation Data + /// @{ + + struct MachRelocationEntry { + uint32_t Word0; + uint32_t Word1; + }; + + llvm::DenseMap<const MCSectionData*, + std::vector<MachRelocationEntry> > Relocations; + + /// @} + /// @name Symbol Table Data + /// @{ + + SmallString<256> StringTable; + std::vector<MachSymbolData> LocalSymbolData; + std::vector<MachSymbolData> ExternalSymbolData; + std::vector<MachSymbolData> UndefinedSymbolData; + + /// @} + + MachObjectWriter *Writer; + + raw_ostream &OS; + + unsigned Is64Bit : 1; + +public: + MachObjectWriterImpl(MachObjectWriter *_Writer, bool _Is64Bit) + : Writer(_Writer), OS(Writer->getStream()), Is64Bit(_Is64Bit) { + } + + void Write8(uint8_t Value) { Writer->Write8(Value); } + void Write16(uint16_t Value) { Writer->Write16(Value); } + void Write32(uint32_t Value) { Writer->Write32(Value); } + void Write64(uint64_t Value) { Writer->Write64(Value); } + void WriteZeros(unsigned N) { Writer->WriteZeros(N); } + void WriteBytes(StringRef Str, unsigned ZeroFillSize = 0) { + Writer->WriteBytes(Str, ZeroFillSize); + } + + void WriteHeader(unsigned NumLoadCommands, unsigned LoadCommandsSize, + bool SubsectionsViaSymbols) { + uint32_t Flags = 0; + + if (SubsectionsViaSymbols) + Flags |= HF_SubsectionsViaSymbols; + + // struct mach_header (28 bytes) or + // struct mach_header_64 (32 bytes) + + uint64_t Start = OS.tell(); + (void) Start; + + Write32(Is64Bit ? Header_Magic64 : Header_Magic32); + + // FIXME: Support cputype. + Write32(Is64Bit ? MachO::CPUTypeX86_64 : MachO::CPUTypeI386); + // FIXME: Support cpusubtype. + Write32(MachO::CPUSubType_I386_ALL); + Write32(HFT_Object); + Write32(NumLoadCommands); // Object files have a single load command, the + // segment. + Write32(LoadCommandsSize); + Write32(Flags); + if (Is64Bit) + Write32(0); // reserved + + assert(OS.tell() - Start == Is64Bit ? Header64Size : Header32Size); + } + + /// WriteSegmentLoadCommand - Write a segment load command. + /// + /// \arg NumSections - The number of sections in this segment. + /// \arg SectionDataSize - The total size of the sections. + void WriteSegmentLoadCommand(unsigned NumSections, + uint64_t VMSize, + uint64_t SectionDataStartOffset, + uint64_t SectionDataSize) { + // struct segment_command (56 bytes) or + // struct segment_command_64 (72 bytes) + + uint64_t Start = OS.tell(); + (void) Start; + + unsigned SegmentLoadCommandSize = Is64Bit ? SegmentLoadCommand64Size : + SegmentLoadCommand32Size; + Write32(Is64Bit ? LCT_Segment64 : LCT_Segment); + Write32(SegmentLoadCommandSize + + NumSections * (Is64Bit ? Section64Size : Section32Size)); + + WriteBytes("", 16); + if (Is64Bit) { + Write64(0); // vmaddr + Write64(VMSize); // vmsize + Write64(SectionDataStartOffset); // file offset + Write64(SectionDataSize); // file size + } else { + Write32(0); // vmaddr + Write32(VMSize); // vmsize + Write32(SectionDataStartOffset); // file offset + Write32(SectionDataSize); // file size + } + Write32(0x7); // maxprot + Write32(0x7); // initprot + Write32(NumSections); + Write32(0); // flags + + assert(OS.tell() - Start == SegmentLoadCommandSize); + } + + void WriteSection(const MCAssembler &Asm, const MCSectionData &SD, + uint64_t FileOffset, uint64_t RelocationsStart, + unsigned NumRelocations) { + // The offset is unused for virtual sections. + if (Asm.getBackend().isVirtualSection(SD.getSection())) { + assert(SD.getFileSize() == 0 && "Invalid file size!"); + FileOffset = 0; + } + + // struct section (68 bytes) or + // struct section_64 (80 bytes) + + uint64_t Start = OS.tell(); + (void) Start; + + // FIXME: cast<> support! + const MCSectionMachO &Section = + static_cast<const MCSectionMachO&>(SD.getSection()); + WriteBytes(Section.getSectionName(), 16); + WriteBytes(Section.getSegmentName(), 16); + if (Is64Bit) { + Write64(SD.getAddress()); // address + Write64(SD.getSize()); // size + } else { + Write32(SD.getAddress()); // address + Write32(SD.getSize()); // size + } + Write32(FileOffset); + + unsigned Flags = Section.getTypeAndAttributes(); + if (SD.hasInstructions()) + Flags |= MCSectionMachO::S_ATTR_SOME_INSTRUCTIONS; + + assert(isPowerOf2_32(SD.getAlignment()) && "Invalid alignment!"); + Write32(Log2_32(SD.getAlignment())); + Write32(NumRelocations ? RelocationsStart : 0); + Write32(NumRelocations); + Write32(Flags); + Write32(0); // reserved1 + Write32(Section.getStubSize()); // reserved2 + if (Is64Bit) + Write32(0); // reserved3 + + assert(OS.tell() - Start == Is64Bit ? Section64Size : Section32Size); + } + + void WriteSymtabLoadCommand(uint32_t SymbolOffset, uint32_t NumSymbols, + uint32_t StringTableOffset, + uint32_t StringTableSize) { + // struct symtab_command (24 bytes) + + uint64_t Start = OS.tell(); + (void) Start; + + Write32(LCT_Symtab); + Write32(SymtabLoadCommandSize); + Write32(SymbolOffset); + Write32(NumSymbols); + Write32(StringTableOffset); + Write32(StringTableSize); + + assert(OS.tell() - Start == SymtabLoadCommandSize); + } + + void WriteDysymtabLoadCommand(uint32_t FirstLocalSymbol, + uint32_t NumLocalSymbols, + uint32_t FirstExternalSymbol, + uint32_t NumExternalSymbols, + uint32_t FirstUndefinedSymbol, + uint32_t NumUndefinedSymbols, + uint32_t IndirectSymbolOffset, + uint32_t NumIndirectSymbols) { + // struct dysymtab_command (80 bytes) + + uint64_t Start = OS.tell(); + (void) Start; + + Write32(LCT_Dysymtab); + Write32(DysymtabLoadCommandSize); + Write32(FirstLocalSymbol); + Write32(NumLocalSymbols); + Write32(FirstExternalSymbol); + Write32(NumExternalSymbols); + Write32(FirstUndefinedSymbol); + Write32(NumUndefinedSymbols); + Write32(0); // tocoff + Write32(0); // ntoc + Write32(0); // modtaboff + Write32(0); // nmodtab + Write32(0); // extrefsymoff + Write32(0); // nextrefsyms + Write32(IndirectSymbolOffset); + Write32(NumIndirectSymbols); + Write32(0); // extreloff + Write32(0); // nextrel + Write32(0); // locreloff + Write32(0); // nlocrel + + assert(OS.tell() - Start == DysymtabLoadCommandSize); + } + + void WriteNlist(MachSymbolData &MSD) { + MCSymbolData &Data = *MSD.SymbolData; + const MCSymbol &Symbol = Data.getSymbol(); + uint8_t Type = 0; + uint16_t Flags = Data.getFlags(); + uint32_t Address = 0; + + // Set the N_TYPE bits. See <mach-o/nlist.h>. + // + // FIXME: Are the prebound or indirect fields possible here? + if (Symbol.isUndefined()) + Type = STT_Undefined; + else if (Symbol.isAbsolute()) + Type = STT_Absolute; + else + Type = STT_Section; + + // FIXME: Set STAB bits. + + if (Data.isPrivateExtern()) + Type |= STF_PrivateExtern; + + // Set external bit. + if (Data.isExternal() || Symbol.isUndefined()) + Type |= STF_External; + + // Compute the symbol address. + if (Symbol.isDefined()) { + if (Symbol.isAbsolute()) { + llvm_unreachable("FIXME: Not yet implemented!"); + } else { + Address = Data.getAddress(); + } + } else if (Data.isCommon()) { + // Common symbols are encoded with the size in the address + // field, and their alignment in the flags. + Address = Data.getCommonSize(); + + // Common alignment is packed into the 'desc' bits. + if (unsigned Align = Data.getCommonAlignment()) { + unsigned Log2Size = Log2_32(Align); + assert((1U << Log2Size) == Align && "Invalid 'common' alignment!"); + if (Log2Size > 15) + llvm_report_error("invalid 'common' alignment '" + + Twine(Align) + "'"); + // FIXME: Keep this mask with the SymbolFlags enumeration. + Flags = (Flags & 0xF0FF) | (Log2Size << 8); + } + } + + // struct nlist (12 bytes) + + Write32(MSD.StringIndex); + Write8(Type); + Write8(MSD.SectionIndex); + + // The Mach-O streamer uses the lowest 16-bits of the flags for the 'desc' + // value. + Write16(Flags); + if (Is64Bit) + Write64(Address); + else + Write32(Address); + } + + void RecordX86_64Relocation(const MCAssembler &Asm, + const MCDataFragment &Fragment, + const MCAsmFixup &Fixup, MCValue Target, + uint64_t &FixedValue) { + unsigned IsPCRel = isFixupKindPCRel(Fixup.Kind); + unsigned IsRIPRel = isFixupKindRIPRel(Fixup.Kind); + unsigned Log2Size = getFixupKindLog2Size(Fixup.Kind); + + // See <reloc.h>. + uint32_t Address = Fragment.getOffset() + Fixup.Offset; + int64_t Value = 0; + unsigned Index = 0; + unsigned IsExtern = 0; + unsigned Type = 0; + + Value = Target.getConstant(); + + if (IsPCRel) { + // Compensate for the relocation offset, Darwin x86_64 relocations only + // have the addend and appear to have attempted to define it to be the + // actual expression addend without the PCrel bias. However, instructions + // with data following the relocation are not accomodated for (see comment + // below regarding SIGNED{1,2,4}), so it isn't exactly that either. + Value += 1 << Log2Size; + } + + if (Target.isAbsolute()) { // constant + // SymbolNum of 0 indicates the absolute section. + Type = RIT_X86_64_Unsigned; + Index = 0; + + // FIXME: I believe this is broken, I don't think the linker can + // understand it. I think it would require a local relocation, but I'm not + // sure if that would work either. The official way to get an absolute + // PCrel relocation is to use an absolute symbol (which we don't support + // yet). + if (IsPCRel) { + IsExtern = 1; + Type = RIT_X86_64_Branch; + } + } else if (Target.getSymB()) { // A - B + constant + const MCSymbol *A = &Target.getSymA()->getSymbol(); + MCSymbolData &A_SD = Asm.getSymbolData(*A); + const MCSymbolData *A_Base = Asm.getAtom(&A_SD); + + const MCSymbol *B = &Target.getSymB()->getSymbol(); + MCSymbolData &B_SD = Asm.getSymbolData(*B); + const MCSymbolData *B_Base = Asm.getAtom(&B_SD); + + // Neither symbol can be modified. + if (Target.getSymA()->getKind() != MCSymbolRefExpr::VK_None || + Target.getSymB()->getKind() != MCSymbolRefExpr::VK_None) + llvm_report_error("unsupported relocation of modified symbol"); + + // We don't support PCrel relocations of differences. Darwin 'as' doesn't + // implement most of these correctly. + if (IsPCRel) + llvm_report_error("unsupported pc-relative relocation of difference"); + + // We don't currently support any situation where one or both of the + // symbols would require a local relocation. This is almost certainly + // unused and may not be possible to encode correctly. + if (!A_Base || !B_Base) + llvm_report_error("unsupported local relocations in difference"); + + // Darwin 'as' doesn't emit correct relocations for this (it ends up with + // a single SIGNED relocation); reject it for now. + if (A_Base == B_Base) + llvm_report_error("unsupported relocation with identical base"); + + Value += A_SD.getAddress() - A_Base->getAddress(); + Value -= B_SD.getAddress() - B_Base->getAddress(); + + Index = A_Base->getIndex(); + IsExtern = 1; + Type = RIT_X86_64_Unsigned; + + MachRelocationEntry MRE; + MRE.Word0 = Address; + MRE.Word1 = ((Index << 0) | + (IsPCRel << 24) | + (Log2Size << 25) | + (IsExtern << 27) | + (Type << 28)); + Relocations[Fragment.getParent()].push_back(MRE); + + Index = B_Base->getIndex(); + IsExtern = 1; + Type = RIT_X86_64_Subtractor; + } else { + const MCSymbol *Symbol = &Target.getSymA()->getSymbol(); + MCSymbolData &SD = Asm.getSymbolData(*Symbol); + const MCSymbolData *Base = Asm.getAtom(&SD); + + // x86_64 almost always uses external relocations, except when there is no + // symbol to use as a base address (a local symbol with no preceeding + // non-local symbol). + if (Base) { + Index = Base->getIndex(); + IsExtern = 1; + + // Add the local offset, if needed. + if (Base != &SD) + Value += SD.getAddress() - Base->getAddress(); + } else { + // The index is the section ordinal. + // + // FIXME: O(N) + Index = 1; + MCAssembler::const_iterator it = Asm.begin(), ie = Asm.end(); + for (; it != ie; ++it, ++Index) + if (&*it == SD.getFragment()->getParent()) + break; + assert(it != ie && "Unable to find section index!"); + IsExtern = 0; + Value += SD.getAddress(); + + if (IsPCRel) + Value -= Address + (1 << Log2Size); + } + + MCSymbolRefExpr::VariantKind Modifier = Target.getSymA()->getKind(); + if (IsPCRel) { + if (IsRIPRel) { + if (Modifier == MCSymbolRefExpr::VK_GOTPCREL) { + // x86_64 distinguishes movq foo@GOTPCREL so that the linker can + // rewrite the movq to an leaq at link time if the symbol ends up in + // the same linkage unit. + if (unsigned(Fixup.Kind) == X86::reloc_riprel_4byte_movq_load) + Type = RIT_X86_64_GOTLoad; + else + Type = RIT_X86_64_GOT; + } else if (Modifier != MCSymbolRefExpr::VK_None) + llvm_report_error("unsupported symbol modifier in relocation"); + else + Type = RIT_X86_64_Signed; + } else { + if (Modifier != MCSymbolRefExpr::VK_None) + llvm_report_error("unsupported symbol modifier in branch " + "relocation"); + + Type = RIT_X86_64_Branch; + } + + // The Darwin x86_64 relocation format has a problem where it cannot + // encode an address (L<foo> + <constant>) which is outside the atom + // containing L<foo>. Generally, this shouldn't occur but it does happen + // when we have a RIPrel instruction with data following the relocation + // entry (e.g., movb $012, L0(%rip)). Even with the PCrel adjustment + // Darwin x86_64 uses, the offset is still negative and the linker has + // no way to recognize this. + // + // To work around this, Darwin uses several special relocation types to + // indicate the offsets. However, the specification or implementation of + // these seems to also be incomplete; they should adjust the addend as + // well based on the actual encoded instruction (the additional bias), + // but instead appear to just look at the final offset. + if (IsRIPRel) { + switch (-(Target.getConstant() + (1 << Log2Size))) { + case 1: Type = RIT_X86_64_Signed1; break; + case 2: Type = RIT_X86_64_Signed2; break; + case 4: Type = RIT_X86_64_Signed4; break; + } + } + } else { + if (Modifier == MCSymbolRefExpr::VK_GOT) + Type = RIT_X86_64_GOT; + else if (Modifier != MCSymbolRefExpr::VK_None) + llvm_report_error("unsupported symbol modifier in relocation"); + else + Type = RIT_X86_64_Unsigned; + } + } + + // x86_64 always writes custom values into the fixups. + FixedValue = Value; + + // struct relocation_info (8 bytes) + MachRelocationEntry MRE; + MRE.Word0 = Address; + MRE.Word1 = ((Index << 0) | + (IsPCRel << 24) | + (Log2Size << 25) | + (IsExtern << 27) | + (Type << 28)); + Relocations[Fragment.getParent()].push_back(MRE); + } + + void RecordScatteredRelocation(const MCAssembler &Asm, + const MCFragment &Fragment, + const MCAsmFixup &Fixup, MCValue Target, + uint64_t &FixedValue) { + uint32_t Address = Fragment.getOffset() + Fixup.Offset; + unsigned IsPCRel = isFixupKindPCRel(Fixup.Kind); + unsigned Log2Size = getFixupKindLog2Size(Fixup.Kind); + unsigned Type = RIT_Vanilla; + + // See <reloc.h>. + const MCSymbol *A = &Target.getSymA()->getSymbol(); + MCSymbolData *A_SD = &Asm.getSymbolData(*A); + + if (!A_SD->getFragment()) + llvm_report_error("symbol '" + A->getName() + + "' can not be undefined in a subtraction expression"); + + uint32_t Value = A_SD->getAddress(); + uint32_t Value2 = 0; + + if (const MCSymbolRefExpr *B = Target.getSymB()) { + MCSymbolData *B_SD = &Asm.getSymbolData(B->getSymbol()); + + if (!B_SD->getFragment()) + llvm_report_error("symbol '" + B->getSymbol().getName() + + "' can not be undefined in a subtraction expression"); + + // Select the appropriate difference relocation type. + // + // Note that there is no longer any semantic difference between these two + // relocation types from the linkers point of view, this is done solely + // for pedantic compatibility with 'as'. + Type = A_SD->isExternal() ? RIT_Difference : RIT_LocalDifference; + Value2 = B_SD->getAddress(); + } + + // Relocations are written out in reverse order, so the PAIR comes first. + if (Type == RIT_Difference || Type == RIT_LocalDifference) { + MachRelocationEntry MRE; + MRE.Word0 = ((0 << 0) | + (RIT_Pair << 24) | + (Log2Size << 28) | + (IsPCRel << 30) | + RF_Scattered); + MRE.Word1 = Value2; + Relocations[Fragment.getParent()].push_back(MRE); + } + + MachRelocationEntry MRE; + MRE.Word0 = ((Address << 0) | + (Type << 24) | + (Log2Size << 28) | + (IsPCRel << 30) | + RF_Scattered); + MRE.Word1 = Value; + Relocations[Fragment.getParent()].push_back(MRE); + } + + void RecordRelocation(const MCAssembler &Asm, const MCDataFragment &Fragment, + const MCAsmFixup &Fixup, MCValue Target, + uint64_t &FixedValue) { + if (Is64Bit) { + RecordX86_64Relocation(Asm, Fragment, Fixup, Target, FixedValue); + return; + } + + unsigned IsPCRel = isFixupKindPCRel(Fixup.Kind); + unsigned Log2Size = getFixupKindLog2Size(Fixup.Kind); + + // If this is a difference or a defined symbol plus an offset, then we need + // a scattered relocation entry. + uint32_t Offset = Target.getConstant(); + if (IsPCRel) + Offset += 1 << Log2Size; + if (Target.getSymB() || + (Target.getSymA() && !Target.getSymA()->getSymbol().isUndefined() && + Offset)) { + RecordScatteredRelocation(Asm, Fragment, Fixup, Target, FixedValue); + return; + } + + // See <reloc.h>. + uint32_t Address = Fragment.getOffset() + Fixup.Offset; + uint32_t Value = 0; + unsigned Index = 0; + unsigned IsExtern = 0; + unsigned Type = 0; + + if (Target.isAbsolute()) { // constant + // SymbolNum of 0 indicates the absolute section. + // + // FIXME: Currently, these are never generated (see code below). I cannot + // find a case where they are actually emitted. + Type = RIT_Vanilla; + Value = 0; + } else { + const MCSymbol *Symbol = &Target.getSymA()->getSymbol(); + MCSymbolData *SD = &Asm.getSymbolData(*Symbol); + + if (Symbol->isUndefined()) { + IsExtern = 1; + Index = SD->getIndex(); + Value = 0; + } else { + // The index is the section ordinal. + // + // FIXME: O(N) + Index = 1; + MCAssembler::const_iterator it = Asm.begin(), ie = Asm.end(); + for (; it != ie; ++it, ++Index) + if (&*it == SD->getFragment()->getParent()) + break; + assert(it != ie && "Unable to find section index!"); + Value = SD->getAddress(); + } + + Type = RIT_Vanilla; + } + + // struct relocation_info (8 bytes) + MachRelocationEntry MRE; + MRE.Word0 = Address; + MRE.Word1 = ((Index << 0) | + (IsPCRel << 24) | + (Log2Size << 25) | + (IsExtern << 27) | + (Type << 28)); + Relocations[Fragment.getParent()].push_back(MRE); + } + + void BindIndirectSymbols(MCAssembler &Asm) { + // This is the point where 'as' creates actual symbols for indirect symbols + // (in the following two passes). It would be easier for us to do this + // sooner when we see the attribute, but that makes getting the order in the + // symbol table much more complicated than it is worth. + // + // FIXME: Revisit this when the dust settles. + + // Bind non lazy symbol pointers first. + for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(), + ie = Asm.indirect_symbol_end(); it != ie; ++it) { + // FIXME: cast<> support! + const MCSectionMachO &Section = + static_cast<const MCSectionMachO&>(it->SectionData->getSection()); + + if (Section.getType() != MCSectionMachO::S_NON_LAZY_SYMBOL_POINTERS) + continue; + + Asm.getOrCreateSymbolData(*it->Symbol); + } + + // Then lazy symbol pointers and symbol stubs. + for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(), + ie = Asm.indirect_symbol_end(); it != ie; ++it) { + // FIXME: cast<> support! + const MCSectionMachO &Section = + static_cast<const MCSectionMachO&>(it->SectionData->getSection()); + + if (Section.getType() != MCSectionMachO::S_LAZY_SYMBOL_POINTERS && + Section.getType() != MCSectionMachO::S_SYMBOL_STUBS) + continue; + + // Set the symbol type to undefined lazy, but only on construction. + // + // FIXME: Do not hardcode. + bool Created; + MCSymbolData &Entry = Asm.getOrCreateSymbolData(*it->Symbol, &Created); + if (Created) + Entry.setFlags(Entry.getFlags() | 0x0001); + } + } + + /// ComputeSymbolTable - Compute the symbol table data + /// + /// \param StringTable [out] - The string table data. + /// \param StringIndexMap [out] - Map from symbol names to offsets in the + /// string table. + void ComputeSymbolTable(MCAssembler &Asm, SmallString<256> &StringTable, + std::vector<MachSymbolData> &LocalSymbolData, + std::vector<MachSymbolData> &ExternalSymbolData, + std::vector<MachSymbolData> &UndefinedSymbolData) { + // Build section lookup table. + DenseMap<const MCSection*, uint8_t> SectionIndexMap; + unsigned Index = 1; + for (MCAssembler::iterator it = Asm.begin(), + ie = Asm.end(); it != ie; ++it, ++Index) + SectionIndexMap[&it->getSection()] = Index; + assert(Index <= 256 && "Too many sections!"); + + // Index 0 is always the empty string. + StringMap<uint64_t> StringIndexMap; + StringTable += '\x00'; + + // Build the symbol arrays and the string table, but only for non-local + // symbols. + // + // The particular order that we collect the symbols and create the string + // table, then sort the symbols is chosen to match 'as'. Even though it + // doesn't matter for correctness, this is important for letting us diff .o + // files. + for (MCAssembler::symbol_iterator it = Asm.symbol_begin(), + ie = Asm.symbol_end(); it != ie; ++it) { + const MCSymbol &Symbol = it->getSymbol(); + + // Ignore non-linker visible symbols. + if (!Asm.isSymbolLinkerVisible(it)) + continue; + + if (!it->isExternal() && !Symbol.isUndefined()) + continue; + + uint64_t &Entry = StringIndexMap[Symbol.getName()]; + if (!Entry) { + Entry = StringTable.size(); + StringTable += Symbol.getName(); + StringTable += '\x00'; + } + + MachSymbolData MSD; + MSD.SymbolData = it; + MSD.StringIndex = Entry; + + if (Symbol.isUndefined()) { + MSD.SectionIndex = 0; + UndefinedSymbolData.push_back(MSD); + } else if (Symbol.isAbsolute()) { + MSD.SectionIndex = 0; + ExternalSymbolData.push_back(MSD); + } else { + MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection()); + assert(MSD.SectionIndex && "Invalid section index!"); + ExternalSymbolData.push_back(MSD); + } + } + + // Now add the data for local symbols. + for (MCAssembler::symbol_iterator it = Asm.symbol_begin(), + ie = Asm.symbol_end(); it != ie; ++it) { + const MCSymbol &Symbol = it->getSymbol(); + + // Ignore non-linker visible symbols. + if (!Asm.isSymbolLinkerVisible(it)) + continue; + + if (it->isExternal() || Symbol.isUndefined()) + continue; + + uint64_t &Entry = StringIndexMap[Symbol.getName()]; + if (!Entry) { + Entry = StringTable.size(); + StringTable += Symbol.getName(); + StringTable += '\x00'; + } + + MachSymbolData MSD; + MSD.SymbolData = it; + MSD.StringIndex = Entry; + + if (Symbol.isAbsolute()) { + MSD.SectionIndex = 0; + LocalSymbolData.push_back(MSD); + } else { + MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection()); + assert(MSD.SectionIndex && "Invalid section index!"); + LocalSymbolData.push_back(MSD); + } + } + + // External and undefined symbols are required to be in lexicographic order. + std::sort(ExternalSymbolData.begin(), ExternalSymbolData.end()); + std::sort(UndefinedSymbolData.begin(), UndefinedSymbolData.end()); + + // Set the symbol indices. + Index = 0; + for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i) + LocalSymbolData[i].SymbolData->setIndex(Index++); + for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i) + ExternalSymbolData[i].SymbolData->setIndex(Index++); + for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i) + UndefinedSymbolData[i].SymbolData->setIndex(Index++); + + // The string table is padded to a multiple of 4. + while (StringTable.size() % 4) + StringTable += '\x00'; + } + + void ExecutePostLayoutBinding(MCAssembler &Asm) { + // Create symbol data for any indirect symbols. + BindIndirectSymbols(Asm); + + // Compute symbol table information and bind symbol indices. + ComputeSymbolTable(Asm, StringTable, LocalSymbolData, ExternalSymbolData, + UndefinedSymbolData); + } + + void WriteObject(const MCAssembler &Asm) { + unsigned NumSections = Asm.size(); + + // The section data starts after the header, the segment load command (and + // section headers) and the symbol table. + unsigned NumLoadCommands = 1; + uint64_t LoadCommandsSize = Is64Bit ? + SegmentLoadCommand64Size + NumSections * Section64Size : + SegmentLoadCommand32Size + NumSections * Section32Size; + + // Add the symbol table load command sizes, if used. + unsigned NumSymbols = LocalSymbolData.size() + ExternalSymbolData.size() + + UndefinedSymbolData.size(); + if (NumSymbols) { + NumLoadCommands += 2; + LoadCommandsSize += SymtabLoadCommandSize + DysymtabLoadCommandSize; + } + + // Compute the total size of the section data, as well as its file size and + // vm size. + uint64_t SectionDataStart = (Is64Bit ? Header64Size : Header32Size) + + LoadCommandsSize; + uint64_t SectionDataSize = 0; + uint64_t SectionDataFileSize = 0; + uint64_t VMSize = 0; + for (MCAssembler::const_iterator it = Asm.begin(), + ie = Asm.end(); it != ie; ++it) { + const MCSectionData &SD = *it; + + VMSize = std::max(VMSize, SD.getAddress() + SD.getSize()); + + if (Asm.getBackend().isVirtualSection(SD.getSection())) + continue; + + SectionDataSize = std::max(SectionDataSize, + SD.getAddress() + SD.getSize()); + SectionDataFileSize = std::max(SectionDataFileSize, + SD.getAddress() + SD.getFileSize()); + } + + // The section data is padded to 4 bytes. + // + // FIXME: Is this machine dependent? + unsigned SectionDataPadding = OffsetToAlignment(SectionDataFileSize, 4); + SectionDataFileSize += SectionDataPadding; + + // Write the prolog, starting with the header and load command... + WriteHeader(NumLoadCommands, LoadCommandsSize, + Asm.getSubsectionsViaSymbols()); + WriteSegmentLoadCommand(NumSections, VMSize, + SectionDataStart, SectionDataSize); + + // ... and then the section headers. + uint64_t RelocTableEnd = SectionDataStart + SectionDataFileSize; + for (MCAssembler::const_iterator it = Asm.begin(), + ie = Asm.end(); it != ie; ++it) { + std::vector<MachRelocationEntry> &Relocs = Relocations[it]; + unsigned NumRelocs = Relocs.size(); + uint64_t SectionStart = SectionDataStart + it->getAddress(); + WriteSection(Asm, *it, SectionStart, RelocTableEnd, NumRelocs); + RelocTableEnd += NumRelocs * RelocationInfoSize; + } + + // Write the symbol table load command, if used. + if (NumSymbols) { + unsigned FirstLocalSymbol = 0; + unsigned NumLocalSymbols = LocalSymbolData.size(); + unsigned FirstExternalSymbol = FirstLocalSymbol + NumLocalSymbols; + unsigned NumExternalSymbols = ExternalSymbolData.size(); + unsigned FirstUndefinedSymbol = FirstExternalSymbol + NumExternalSymbols; + unsigned NumUndefinedSymbols = UndefinedSymbolData.size(); + unsigned NumIndirectSymbols = Asm.indirect_symbol_size(); + unsigned NumSymTabSymbols = + NumLocalSymbols + NumExternalSymbols + NumUndefinedSymbols; + uint64_t IndirectSymbolSize = NumIndirectSymbols * 4; + uint64_t IndirectSymbolOffset = 0; + + // If used, the indirect symbols are written after the section data. + if (NumIndirectSymbols) + IndirectSymbolOffset = RelocTableEnd; + + // The symbol table is written after the indirect symbol data. + uint64_t SymbolTableOffset = RelocTableEnd + IndirectSymbolSize; + + // The string table is written after symbol table. + uint64_t StringTableOffset = + SymbolTableOffset + NumSymTabSymbols * (Is64Bit ? Nlist64Size : + Nlist32Size); + WriteSymtabLoadCommand(SymbolTableOffset, NumSymTabSymbols, + StringTableOffset, StringTable.size()); + + WriteDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols, + FirstExternalSymbol, NumExternalSymbols, + FirstUndefinedSymbol, NumUndefinedSymbols, + IndirectSymbolOffset, NumIndirectSymbols); + } + + // Write the actual section data. + for (MCAssembler::const_iterator it = Asm.begin(), + ie = Asm.end(); it != ie; ++it) + Asm.WriteSectionData(it, Writer); + + // Write the extra padding. + WriteZeros(SectionDataPadding); + + // Write the relocation entries. + for (MCAssembler::const_iterator it = Asm.begin(), + ie = Asm.end(); it != ie; ++it) { + // Write the section relocation entries, in reverse order to match 'as' + // (approximately, the exact algorithm is more complicated than this). + std::vector<MachRelocationEntry> &Relocs = Relocations[it]; + for (unsigned i = 0, e = Relocs.size(); i != e; ++i) { + Write32(Relocs[e - i - 1].Word0); + Write32(Relocs[e - i - 1].Word1); + } + } + + // Write the symbol table data, if used. + if (NumSymbols) { + // Write the indirect symbol entries. + for (MCAssembler::const_indirect_symbol_iterator + it = Asm.indirect_symbol_begin(), + ie = Asm.indirect_symbol_end(); it != ie; ++it) { + // Indirect symbols in the non lazy symbol pointer section have some + // special handling. + const MCSectionMachO &Section = + static_cast<const MCSectionMachO&>(it->SectionData->getSection()); + if (Section.getType() == MCSectionMachO::S_NON_LAZY_SYMBOL_POINTERS) { + // If this symbol is defined and internal, mark it as such. + if (it->Symbol->isDefined() && + !Asm.getSymbolData(*it->Symbol).isExternal()) { + uint32_t Flags = ISF_Local; + if (it->Symbol->isAbsolute()) + Flags |= ISF_Absolute; + Write32(Flags); + continue; + } + } + + Write32(Asm.getSymbolData(*it->Symbol).getIndex()); + } + + // FIXME: Check that offsets match computed ones. + + // Write the symbol table entries. + for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i) + WriteNlist(LocalSymbolData[i]); + for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i) + WriteNlist(ExternalSymbolData[i]); + for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i) + WriteNlist(UndefinedSymbolData[i]); + + // Write the string table. + OS << StringTable.str(); + } + } +}; + +} + +MachObjectWriter::MachObjectWriter(raw_ostream &OS, + bool Is64Bit, + bool IsLittleEndian) + : MCObjectWriter(OS, IsLittleEndian) +{ + Impl = new MachObjectWriterImpl(this, Is64Bit); +} + +MachObjectWriter::~MachObjectWriter() { + delete (MachObjectWriterImpl*) Impl; +} + +void MachObjectWriter::ExecutePostLayoutBinding(MCAssembler &Asm) { + ((MachObjectWriterImpl*) Impl)->ExecutePostLayoutBinding(Asm); +} + +void MachObjectWriter::RecordRelocation(const MCAssembler &Asm, + const MCDataFragment &Fragment, + const MCAsmFixup &Fixup, MCValue Target, + uint64_t &FixedValue) { + ((MachObjectWriterImpl*) Impl)->RecordRelocation(Asm, Fragment, Fixup, + Target, FixedValue); +} + +void MachObjectWriter::WriteObject(const MCAssembler &Asm) { + ((MachObjectWriterImpl*) Impl)->WriteObject(Asm); +} diff --git a/lib/MC/TargetAsmBackend.cpp b/lib/MC/TargetAsmBackend.cpp index 918d272..bbfddbe 100644 --- a/lib/MC/TargetAsmBackend.cpp +++ b/lib/MC/TargetAsmBackend.cpp @@ -11,7 +11,10 @@ using namespace llvm; TargetAsmBackend::TargetAsmBackend(const Target &T) - : TheTarget(T) + : TheTarget(T), + HasAbsolutizedSet(false), + HasReliableSymbolDifference(false), + HasScatteredSymbols(false) { } |