#define DEBUG_TYPE "assembler"
#include "llvm/MC/MCAssembler.h"
+#include "llvm/MC/MCExpr.h"
#include "llvm/MC/MCSectionMachO.h"
-#include "llvm/Target/TargetMachOWriterInfo.h"
+#include "llvm/MC/MCSymbol.h"
+#include "llvm/MC/MCValue.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/SmallString.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"
+
+// FIXME: Gross.
+#include "../Target/X86/X86FixupKinds.h"
+
#include <vector>
using namespace llvm;
STATISTIC(EmittedFragments, "Number of emitted assembler fragments");
+// FIXME FIXME FIXME: There are number of places in this file where we convert
+// what is a 64-bit assembler value used for computation into a value in the
+// 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);
+ unsigned Type = SMO.getTypeAndAttributes() & MCSectionMachO::SECTION_TYPE;
+ return (Type == 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
};
-
+
static const unsigned Header32Size = 28;
static const unsigned Header64Size = 32;
static const unsigned SegmentLoadCommand32Size = 56;
static const unsigned SymtabLoadCommandSize = 24;
static const unsigned DysymtabLoadCommandSize = 80;
static const unsigned Nlist32Size = 12;
+ static const unsigned RelocationInfoSize = 8;
enum HeaderFileType {
HFT_Object = 0x1
};
+ enum HeaderFlags {
+ HF_SubsectionsViaSymbols = 0x2000
+ };
+
enum LoadCommandType {
LCT_Segment = 0x1,
LCT_Symtab = 0x2,
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 {
bool IsLSB;
public:
- MachObjectWriter(raw_ostream &_OS, bool _IsLSB = true)
+ MachObjectWriter(raw_ostream &_OS, bool _IsLSB = true)
: OS(_OS), IsLSB(_IsLSB) {
}
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(const StringRef &Str, unsigned ZeroFillSize = 0) {
+ void WriteString(StringRef Str, unsigned ZeroFillSize = 0) {
OS << Str;
if (ZeroFillSize)
WriteZeros(ZeroFillSize - Str.size());
}
/// @}
-
- void WriteHeader32(unsigned NumLoadCommands, unsigned LoadCommandsSize) {
+
+ void WriteHeader32(unsigned NumLoadCommands, unsigned LoadCommandsSize,
+ bool SubsectionsViaSymbols) {
+ uint32_t Flags = 0;
+
+ if (SubsectionsViaSymbols)
+ Flags |= HF_SubsectionsViaSymbols;
+
// struct mach_header (28 bytes)
uint64_t Start = OS.tell();
Write32(Header_Magic32);
// FIXME: Support cputype.
- Write32(TargetMachOWriterInfo::HDR_CPU_TYPE_I386);
-
+ Write32(MachO::CPUTypeI386);
// FIXME: Support cpusubtype.
- Write32(TargetMachOWriterInfo::HDR_CPU_SUBTYPE_I386_ALL);
-
+ Write32(MachO::CPUSubType_I386_ALL);
Write32(HFT_Object);
-
- // Object files have a single load command, the segment.
- Write32(NumLoadCommands);
+ Write32(NumLoadCommands); // Object files have a single load command, the
+ // segment.
Write32(LoadCommandsSize);
- Write32(0); // Flags
+ Write32(Flags);
assert(OS.tell() - Start == Header32Size);
}
/// \arg NumSections - The number of sections in this segment.
/// \arg SectionDataSize - The total size of the sections.
void WriteSegmentLoadCommand32(unsigned NumSections,
+ uint64_t VMSize,
uint64_t SectionDataStartOffset,
uint64_t SectionDataSize) {
// struct segment_command (56 bytes)
WriteString("", 16);
Write32(0); // vmaddr
- Write32(SectionDataSize); // vmsize
+ Write32(VMSize); // vmsize
Write32(SectionDataStartOffset); // file offset
Write32(SectionDataSize); // file size
Write32(0x7); // maxprot
assert(OS.tell() - Start == SegmentLoadCommand32Size);
}
- void WriteSection32(const MCSectionData &SD, uint64_t FileOffset) {
+ void WriteSection32(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)
uint64_t Start = OS.tell();
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(0); // file offset of relocation entries
- Write32(0); // number of relocation entrions
- Write32(Section.getTypeAndAttributes());
+ Write32(NumRelocations ? RelocationsStart : 0);
+ Write32(NumRelocations);
+ Write32(Flags);
Write32(0); // reserved1
Write32(Section.getStubSize()); // reserved2
void WriteNlist32(MachSymbolData &MSD) {
MCSymbolData &Data = *MSD.SymbolData;
- MCSymbol &Symbol = Data.getSymbol();
+ 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>.
//
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.getFragment()->getAddress() + Data.getOffset();
+ }
+ } 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(Data.getFlags() & 0xFFFF);
+ Write16(Flags);
+ Write32(Address);
+ }
- // Write the symbol address.
- uint32_t Address = 0;
- if (Symbol.isDefined()) {
- if (Symbol.isAbsolute()) {
- llvm_unreachable("FIXME: Not yet implemented!");
+ struct MachRelocationEntry {
+ uint32_t Word0;
+ uint32_t Word1;
+ };
+ void ComputeScatteredRelocationInfo(MCAssembler &Asm, MCFragment &Fragment,
+ MCAsmFixup &Fixup,
+ const MCValue &Target,
+ DenseMap<const MCSymbol*,MCSymbolData*> &SymbolMap,
+ std::vector<MachRelocationEntry> &Relocs) {
+ uint32_t Address = Fragment.getOffset() + Fixup.Offset;
+ unsigned IsPCRel = 0;
+ unsigned Log2Size = getFixupKindLog2Size(Fixup.Kind);
+ unsigned Type = RIT_Vanilla;
+
+ // See <reloc.h>.
+ const MCSymbol *A = Target.getSymA();
+ MCSymbolData *SD = SymbolMap.lookup(A);
+ uint32_t Value = SD->getFragment()->getAddress() + SD->getOffset();
+ uint32_t Value2 = 0;
+
+ if (const MCSymbol *B = Target.getSymB()) {
+ Type = RIT_LocalDifference;
+
+ MCSymbolData *SD = SymbolMap.lookup(B);
+ Value2 = SD->getFragment()->getAddress() + SD->getOffset();
+ }
+
+ // The value which goes in the fixup is current value of the expression.
+ Fixup.FixedValue = Value - Value2 + Target.getConstant();
+ if (isFixupKindPCRel(Fixup.Kind)) {
+ Fixup.FixedValue -= Address;
+ IsPCRel = 1;
+ }
+
+ MachRelocationEntry MRE;
+ MRE.Word0 = ((Address << 0) |
+ (Type << 24) |
+ (Log2Size << 28) |
+ (IsPCRel << 30) |
+ RF_Scattered);
+ MRE.Word1 = Value;
+ Relocs.push_back(MRE);
+
+ if (Type == RIT_LocalDifference) {
+ Type = RIT_Pair;
+
+ MachRelocationEntry MRE;
+ MRE.Word0 = ((0 << 0) |
+ (Type << 24) |
+ (Log2Size << 28) |
+ (0 << 30) |
+ RF_Scattered);
+ MRE.Word1 = Value2;
+ Relocs.push_back(MRE);
+ }
+ }
+
+ void ComputeRelocationInfo(MCAssembler &Asm, MCDataFragment &Fragment,
+ MCAsmFixup &Fixup,
+ DenseMap<const MCSymbol*,MCSymbolData*> &SymbolMap,
+ std::vector<MachRelocationEntry> &Relocs) {
+ MCValue Target;
+ if (!Fixup.Value->EvaluateAsRelocatable(Target))
+ llvm_report_error("expected relocatable expression");
+
+ // If this is a difference or a local symbol plus an offset, then we need a
+ // scattered relocation entry.
+ if (Target.getSymB() ||
+ (Target.getSymA() && !Target.getSymA()->isUndefined() &&
+ Target.getConstant()))
+ return ComputeScatteredRelocationInfo(Asm, Fragment, Fixup, Target,
+ SymbolMap, Relocs);
+
+ // See <reloc.h>.
+ uint32_t Address = Fragment.getOffset() + Fixup.Offset;
+ uint32_t Value = 0;
+ unsigned Index = 0;
+ unsigned IsPCRel = 0;
+ unsigned Log2Size = getFixupKindLog2Size(Fixup.Kind);
+ unsigned IsExtern = 0;
+ unsigned Type = 0;
+
+ if (Target.isAbsolute()) { // constant
+ // SymbolNum of 0 indicates the absolute section.
+ //
+ // FIXME: When is this generated?
+ Type = RIT_Vanilla;
+ Value = 0;
+ llvm_unreachable("FIXME: Not yet implemented!");
+ } else {
+ const MCSymbol *Symbol = Target.getSymA();
+ MCSymbolData *SD = SymbolMap.lookup(Symbol);
+
+ if (Symbol->isUndefined()) {
+ IsExtern = 1;
+ Index = SD->getIndex();
+ Value = 0;
} else {
- Address = Data.getFragment()->getAddress() + Data.getOffset();
+ // 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->getFragment()->getAddress() + SD->getOffset();
}
+
+ Type = RIT_Vanilla;
}
- Write32(Address);
+
+ // The value which goes in the fixup is current value of the expression.
+ Fixup.FixedValue = Value + Target.getConstant();
+
+ if (isFixupKindPCRel(Fixup.Kind)) {
+ Fixup.FixedValue -= Address;
+ IsPCRel = 1;
+ }
+
+ // 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,
- DenseMap<MCSymbol*, MCSymbolData*> &SymbolMap) {
+ DenseMap<const MCSymbol*,MCSymbolData*> &SymbolMap) {
// 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
// files.
for (MCAssembler::symbol_iterator it = Asm.symbol_begin(),
ie = Asm.symbol_end(); it != ie; ++it) {
- MCSymbol &Symbol = it->getSymbol();
+ const MCSymbol &Symbol = it->getSymbol();
+
+ // Ignore assembler temporaries.
+ if (it->getSymbol().isTemporary())
+ continue;
if (!it->isExternal() && !Symbol.isUndefined())
continue;
// Now add the data for local symbols.
for (MCAssembler::symbol_iterator it = Asm.symbol_begin(),
ie = Asm.symbol_end(); it != ie; ++it) {
- MCSymbol &Symbol = it->getSymbol();
+ const MCSymbol &Symbol = it->getSymbol();
+
+ // Ignore assembler temporaries.
+ if (it->getSymbol().isTemporary())
+ continue;
if (it->isExternal() || Symbol.isUndefined())
continue;
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.
- //
- // FIXME: Check to see if this varies per arch.
while (StringTable.size() % 4)
StringTable += '\x00';
}
// Compute the symbol -> symbol data map.
//
// FIXME: This should not be here.
- DenseMap<MCSymbol*, MCSymbolData *> SymbolMap;
+ DenseMap<const MCSymbol*, MCSymbolData *> SymbolMap;
for (MCAssembler::symbol_iterator it = Asm.symbol_begin(),
ie = Asm.symbol_end(); it != ie; ++it)
SymbolMap[&it->getSymbol()] = it;
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;
LoadCommandsSize += SymtabLoadCommandSize + DysymtabLoadCommandSize;
}
+ // Compute the total size of the section data, as well as its file size and
+ // vm size.
uint64_t SectionDataStart = Header32Size + LoadCommandsSize;
- uint64_t SectionDataEnd = SectionDataStart;
uint64_t SectionDataSize = 0;
- if (!Asm.getSectionList().empty()) {
- MCSectionData &SD = Asm.getSectionList().back();
- SectionDataSize = SD.getAddress() + SD.getSize();
- SectionDataEnd = SectionDataStart + SD.getAddress() + SD.getFileSize();
+ 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...
- WriteHeader32(NumLoadCommands, LoadCommandsSize);
- WriteSegmentLoadCommand32(NumSections, SectionDataStart, SectionDataSize);
-
+ WriteHeader32(NumLoadCommands, LoadCommandsSize,
+ Asm.getSubsectionsViaSymbols());
+ WriteSegmentLoadCommand32(NumSections, VMSize,
+ SectionDataStart, SectionDataSize);
+
// ... and then the section headers.
- for (MCAssembler::iterator it = Asm.begin(), ie = Asm.end(); it != ie; ++it)
- WriteSection32(*it, SectionDataStart + it->getAddress());
+ //
+ // 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],
+ SymbolMap, RelocInfos);
+
+ unsigned NumRelocs = RelocInfos.size() - NumRelocsStart;
+ uint64_t SectionStart = SectionDataStart + SD.getAddress();
+ WriteSection32(SD, SectionStart, RelocTableEnd, NumRelocs);
+ RelocTableEnd += NumRelocs * RelocationInfoSize;
+ }
// Write the symbol table load command, if used.
if (NumSymbols) {
// If used, the indirect symbols are written after the section data.
if (NumIndirectSymbols)
- IndirectSymbolOffset = SectionDataEnd;
+ IndirectSymbolOffset = RelocTableEnd;
// The symbol table is written after the indirect symbol data.
- uint64_t SymbolTableOffset = SectionDataEnd + IndirectSymbolSize;
+ uint64_t SymbolTableOffset = RelocTableEnd + IndirectSymbolSize;
// The string table is written after symbol table.
uint64_t StringTableOffset =
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) {
- // FIXME: We shouldn't need this index table.
- DenseMap<MCSymbol*, unsigned> SymbolIndexMap;
- for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i) {
- MCSymbol &Symbol = LocalSymbolData[i].SymbolData->getSymbol();
- SymbolIndexMap.insert(std::make_pair(&Symbol, SymbolIndexMap.size()));
- }
- for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i) {
- MCSymbol &Symbol = ExternalSymbolData[i].SymbolData->getSymbol();
- SymbolIndexMap.insert(std::make_pair(&Symbol, SymbolIndexMap.size()));
- }
- for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i) {
- MCSymbol &Symbol = UndefinedSymbolData[i].SymbolData->getSymbol();
- SymbolIndexMap.insert(std::make_pair(&Symbol, SymbolIndexMap.size()));
- }
-
// Write the indirect symbol entries.
- //
- // FIXME: We need the symbol index map for this.
for (MCAssembler::indirect_symbol_iterator
it = Asm.indirect_symbol_begin(),
ie = Asm.indirect_symbol_end(); it != ie; ++it) {
}
}
- Write32(SymbolIndexMap[it->Symbol]);
+ Write32(SymbolMap[it->Symbol]->getIndex());
}
// FIXME: Check that offsets match computed ones.
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));
+ }
};
/* *** */
/* *** */
-MCSectionData::MCSectionData() : Section(*(MCSection*)0) {}
+MCSectionData::MCSectionData() : Section(0) {}
MCSectionData::MCSectionData(const MCSection &_Section, MCAssembler *A)
- : Section(_Section),
+ : Section(&_Section),
Alignment(1),
Address(~UINT64_C(0)),
Size(~UINT64_C(0)),
- FileSize(~UINT64_C(0))
+ FileSize(~UINT64_C(0)),
+ HasInstructions(false)
{
if (A)
A->getSectionList().push_back(this);
/* *** */
-MCSymbolData::MCSymbolData() : Symbol(*(MCSymbol*)0) {}
+MCSymbolData::MCSymbolData() : Symbol(0) {}
-MCSymbolData::MCSymbolData(MCSymbol &_Symbol, MCFragment *_Fragment,
+MCSymbolData::MCSymbolData(const MCSymbol &_Symbol, MCFragment *_Fragment,
uint64_t _Offset, MCAssembler *A)
- : Symbol(_Symbol), Fragment(_Fragment), Offset(_Offset),
- IsExternal(false), IsPrivateExtern(false), Flags(0)
+ : Symbol(&_Symbol), Fragment(_Fragment), Offset(_Offset),
+ IsExternal(false), IsPrivateExtern(false),
+ CommonSize(0), CommonAlign(0), Flags(0), Index(0)
{
if (A)
A->getSymbolList().push_back(this);
/* *** */
-MCAssembler::MCAssembler(raw_ostream &_OS) : OS(_OS) {}
+MCAssembler::MCAssembler(MCContext &_Context, raw_ostream &_OS)
+ : Context(_Context), OS(_OS), SubsectionsViaSymbols(false)
+{
+}
MCAssembler::~MCAssembler() {
}
-void MCAssembler::LayoutSection(MCSectionData &SD, unsigned NextAlign) {
+void MCAssembler::LayoutSection(MCSectionData &SD) {
uint64_t Address = SD.getAddress();
for (MCSectionData::iterator it = SD.begin(), ie = SD.end(); it != ie; ++it) {
switch (F.getKind()) {
case MCFragment::FT_Align: {
MCAlignFragment &AF = cast<MCAlignFragment>(F);
-
- uint64_t Size = RoundUpToAlignment(Address, AF.getAlignment()) - Address;
+
+ uint64_t Size = OffsetToAlignment(Address, AF.getAlignment());
if (Size > AF.getMaxBytesToEmit())
AF.setFileSize(0);
else
case MCFragment::FT_Org: {
MCOrgFragment &OF = cast<MCOrgFragment>(F);
- if (!OF.getOffset().isAbsolute())
+ MCValue Target;
+ if (!OF.getOffset().EvaluateAsRelocatable(Target))
+ llvm_report_error("expected relocatable expression");
+
+ if (!Target.isAbsolute())
llvm_unreachable("FIXME: Not yet implemented!");
- uint64_t OrgOffset = OF.getOffset().getConstant();
+ uint64_t OrgOffset = Target.getConstant();
uint64_t Offset = Address - SD.getAddress();
// FIXME: We need a way to communicate this error.
if (OrgOffset < Offset)
- llvm_report_error("invalid .org offset '" + Twine(OrgOffset) +
+ llvm_report_error("invalid .org offset '" + Twine(OrgOffset) +
"' (at offset '" + Twine(Offset) + "'");
-
+
F.setFileSize(OrgOffset - Offset);
break;
- }
+ }
+
+ case MCFragment::FT_ZeroFill: {
+ MCZeroFillFragment &ZFF = cast<MCZeroFillFragment>(F);
+
+ // Align the fragment offset; it is safe to adjust the offset freely since
+ // this is only in virtual sections.
+ uint64_t Aligned = RoundUpToAlignment(Address, ZFF.getAlignment());
+ F.setOffset(Aligned - SD.getAddress());
+
+ // FIXME: This is misnamed.
+ F.setFileSize(ZFF.getSize());
+ break;
+ }
}
Address += F.getFileSize();
// Set the section sizes.
SD.setSize(Address - SD.getAddress());
- SD.setFileSize(RoundUpToAlignment(Address, NextAlign) - SD.getAddress());
+ if (isVirtualSection(SD.getSection()))
+ SD.setFileSize(0);
+ else
+ SD.setFileSize(Address - SD.getAddress());
+}
+
+/// WriteNopData - Write optimal nops to the output file for the \arg Count
+/// bytes. This returns the number of bytes written. It may return 0 if
+/// 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 const uint8_t Nops[16][16] = {
+ // nop
+ {0x90},
+ // xchg %ax,%ax
+ {0x66, 0x90},
+ // nopl (%[re]ax)
+ {0x0f, 0x1f, 0x00},
+ // nopl 0(%[re]ax)
+ {0x0f, 0x1f, 0x40, 0x00},
+ // nopl 0(%[re]ax,%[re]ax,1)
+ {0x0f, 0x1f, 0x44, 0x00, 0x00},
+ // nopw 0(%[re]ax,%[re]ax,1)
+ {0x66, 0x0f, 0x1f, 0x44, 0x00, 0x00},
+ // nopl 0L(%[re]ax)
+ {0x0f, 0x1f, 0x80, 0x00, 0x00, 0x00, 0x00},
+ // nopl 0L(%[re]ax,%[re]ax,1)
+ {0x0f, 0x1f, 0x84, 0x00, 0x00, 0x00, 0x00, 0x00},
+ // nopw 0L(%[re]ax,%[re]ax,1)
+ {0x66, 0x0f, 0x1f, 0x84, 0x00, 0x00, 0x00, 0x00, 0x00},
+ // nopw %cs:0L(%[re]ax,%[re]ax,1)
+ {0x66, 0x2e, 0x0f, 0x1f, 0x84, 0x00, 0x00, 0x00, 0x00, 0x00},
+ // nopl 0(%[re]ax,%[re]ax,1)
+ // nopw 0(%[re]ax,%[re]ax,1)
+ {0x0f, 0x1f, 0x44, 0x00, 0x00,
+ 0x66, 0x0f, 0x1f, 0x44, 0x00, 0x00},
+ // nopw 0(%[re]ax,%[re]ax,1)
+ // nopw 0(%[re]ax,%[re]ax,1)
+ {0x66, 0x0f, 0x1f, 0x44, 0x00, 0x00,
+ 0x66, 0x0f, 0x1f, 0x44, 0x00, 0x00},
+ // nopw 0(%[re]ax,%[re]ax,1)
+ // nopl 0L(%[re]ax) */
+ {0x66, 0x0f, 0x1f, 0x44, 0x00, 0x00,
+ 0x0f, 0x1f, 0x80, 0x00, 0x00, 0x00, 0x00},
+ // nopl 0L(%[re]ax)
+ // nopl 0L(%[re]ax)
+ {0x0f, 0x1f, 0x80, 0x00, 0x00, 0x00, 0x00,
+ 0x0f, 0x1f, 0x80, 0x00, 0x00, 0x00, 0x00},
+ // nopl 0L(%[re]ax)
+ // nopl 0L(%[re]ax,%[re]ax,1)
+ {0x0f, 0x1f, 0x80, 0x00, 0x00, 0x00, 0x00,
+ 0x0f, 0x1f, 0x84, 0x00, 0x00, 0x00, 0x00, 0x00}
+ };
+
+ if (Count > 15)
+ return 0;
+
+ for (uint64_t i = 0; i < Count; i++)
+ MOW.Write8 (uint8_t(Nops[Count - 1][i]));
+
+ return Count;
}
/// WriteFileData - Write the \arg F data to the output file.
MachObjectWriter &MOW) {
uint64_t Start = OS.tell();
(void) Start;
-
+
++EmittedFragments;
// FIXME: Embed in fragments instead?
// multiple .align directives to enforce the semantics it wants), but is
// severe enough that we want to report it. How to handle this?
if (Count * AF.getValueSize() != AF.getFileSize())
- llvm_report_error("undefined .align directive, value size '" +
- Twine(AF.getValueSize()) +
+ llvm_report_error("undefined .align directive, value size '" +
+ Twine(AF.getValueSize()) +
"' is not a divisor of padding size '" +
Twine(AF.getFileSize()) + "'");
+ // See if we are aligning with nops, and if so do that first to try to fill
+ // 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);
+ Count -= NopByteCount;
+ }
+
for (uint64_t i = 0; i != Count; ++i) {
switch (AF.getValueSize()) {
default:
break;
}
- case MCFragment::FT_Data:
+ 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();
break;
+ }
case MCFragment::FT_Fill: {
MCFillFragment &FF = cast<MCFillFragment>(F);
-
- if (!FF.getValue().isAbsolute())
- llvm_unreachable("FIXME: Not yet implemented!");
- int64_t Value = FF.getValue().getConstant();
-
for (uint64_t i = 0, e = FF.getCount(); i != e; ++i) {
switch (FF.getValueSize()) {
default:
assert(0 && "Invalid size!");
- case 1: MOW.Write8 (uint8_t (Value)); break;
- case 2: MOW.Write16(uint16_t(Value)); break;
- case 4: MOW.Write32(uint32_t(Value)); break;
- case 8: MOW.Write64(uint64_t(Value)); break;
+ 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;
}
}
break;
}
-
+
case MCFragment::FT_Org: {
MCOrgFragment &OF = cast<MCOrgFragment>(F);
break;
}
+
+ case MCFragment::FT_ZeroFill: {
+ assert(0 && "Invalid zero fill fragment in concrete section!");
+ break;
+ }
}
assert(OS.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) {
+ // Ignore virtual sections.
+ if (isVirtualSection(SD.getSection())) {
+ assert(SD.getFileSize() == 0);
+ return;
+ }
+
uint64_t Start = OS.tell();
(void) Start;
-
+
for (MCSectionData::const_iterator it = SD.begin(),
ie = SD.end(); it != ie; ++it)
WriteFileData(OS, *it, MOW);
}
void MCAssembler::Finish() {
- // Layout the sections and fragments.
+ DEBUG_WITH_TYPE("mc-dump", {
+ llvm::errs() << "assembler backend - pre-layout\n--\n";
+ dump(); });
+
+ // Layout the concrete sections and fragments.
uint64_t Address = 0;
- for (iterator it = begin(), ie = end(); it != ie;) {
+ MCSectionData *Prev = 0;
+ for (iterator it = begin(), ie = end(); it != ie; ++it) {
MCSectionData &SD = *it;
- // Select the amount of padding alignment we need, based on either the next
- // sections alignment or the default alignment.
- //
- // FIXME: This should probably match the native word size.
- unsigned NextAlign = 4;
- ++it;
- if (it != ie)
- NextAlign = it->getAlignment();
+ // Skip virtual sections.
+ if (isVirtualSection(SD.getSection()))
+ continue;
+
+ // Align this section if necessary by adding padding bytes to the previous
+ // section.
+ if (uint64_t Pad = OffsetToAlignment(Address, it->getAlignment())) {
+ assert(Prev && "Missing prev section!");
+ Prev->setFileSize(Prev->getFileSize() + Pad);
+ Address += Pad;
+ }
// Layout the section fragments and its size.
SD.setAddress(Address);
- LayoutSection(SD, NextAlign);
+ LayoutSection(SD);
Address += SD.getFileSize();
+
+ Prev = &SD;
+ }
+
+ // Layout the virtual sections.
+ for (iterator it = begin(), ie = end(); it != ie; ++it) {
+ MCSectionData &SD = *it;
+
+ if (!isVirtualSection(SD.getSection()))
+ continue;
+
+ SD.setAddress(Address);
+ LayoutSection(SD);
+ Address += SD.getSize();
}
+ DEBUG_WITH_TYPE("mc-dump", {
+ llvm::errs() << "assembler backend - post-layout\n--\n";
+ dump(); });
+
// Write the object file.
MachObjectWriter MOW(OS);
MOW.WriteObject(*this);
OS.flush();
}
+
+
+// Debugging methods
+
+namespace llvm {
+
+raw_ostream &operator<<(raw_ostream &OS, const MCAsmFixup &AF) {
+ OS << "<MCAsmFixup" << " Offset:" << AF.Offset << " Value:" << *AF.Value
+ << " Kind:" << AF.Kind << ">";
+ return OS;
+}
+
+}
+
+void MCFragment::dump() {
+ raw_ostream &OS = llvm::errs();
+
+ OS << "<MCFragment " << (void*) this << " Offset:" << Offset
+ << " FileSize:" << FileSize;
+
+ OS << ">";
+}
+
+void MCAlignFragment::dump() {
+ raw_ostream &OS = llvm::errs();
+
+ OS << "<MCAlignFragment ";
+ this->MCFragment::dump();
+ OS << "\n ";
+ OS << " Alignment:" << getAlignment()
+ << " Value:" << getValue() << " ValueSize:" << getValueSize()
+ << " MaxBytesToEmit:" << getMaxBytesToEmit() << ">";
+}
+
+void MCDataFragment::dump() {
+ raw_ostream &OS = llvm::errs();
+
+ OS << "<MCDataFragment ";
+ this->MCFragment::dump();
+ OS << "\n ";
+ OS << " Contents:[";
+ for (unsigned i = 0, e = getContents().size(); i != e; ++i) {
+ if (i) OS << ",";
+ OS << hexdigit((Contents[i] >> 4) & 0xF) << hexdigit(Contents[i] & 0xF);
+ }
+ OS << "] (" << getContents().size() << " bytes)";
+
+ if (!getFixups().empty()) {
+ OS << ",\n ";
+ OS << " Fixups:[";
+ for (fixup_iterator it = fixup_begin(), ie = fixup_end(); it != ie; ++it) {
+ if (it != fixup_begin()) OS << ",\n ";
+ OS << *it;
+ }
+ OS << "]";
+ }
+
+ OS << ">";
+}
+
+void MCFillFragment::dump() {
+ raw_ostream &OS = llvm::errs();
+
+ OS << "<MCFillFragment ";
+ this->MCFragment::dump();
+ OS << "\n ";
+ OS << " Value:" << getValue() << " ValueSize:" << getValueSize()
+ << " Count:" << getCount() << ">";
+}
+
+void MCOrgFragment::dump() {
+ raw_ostream &OS = llvm::errs();
+
+ OS << "<MCOrgFragment ";
+ this->MCFragment::dump();
+ OS << "\n ";
+ OS << " Offset:" << getOffset() << " Value:" << getValue() << ">";
+}
+
+void MCZeroFillFragment::dump() {
+ raw_ostream &OS = llvm::errs();
+
+ OS << "<MCZeroFillFragment ";
+ this->MCFragment::dump();
+ OS << "\n ";
+ OS << " Size:" << getSize() << " Alignment:" << getAlignment() << ">";
+}
+
+void MCSectionData::dump() {
+ raw_ostream &OS = llvm::errs();
+
+ OS << "<MCSectionData";
+ OS << " Alignment:" << getAlignment() << " Address:" << Address
+ << " Size:" << Size << " FileSize:" << FileSize
+ << " Fragments:[";
+ for (iterator it = begin(), ie = end(); it != ie; ++it) {
+ if (it != begin()) OS << ",\n ";
+ it->dump();
+ }
+ OS << "]>";
+}
+
+void MCSymbolData::dump() {
+ raw_ostream &OS = llvm::errs();
+
+ OS << "<MCSymbolData Symbol:" << getSymbol()
+ << " Fragment:" << getFragment() << " Offset:" << getOffset()
+ << " Flags:" << getFlags() << " Index:" << getIndex();
+ if (isCommon())
+ OS << " (common, size:" << getCommonSize()
+ << " align: " << getCommonAlignment() << ")";
+ if (isExternal())
+ OS << " (external)";
+ if (isPrivateExtern())
+ OS << " (private extern)";
+ OS << ">";
+}
+
+void MCAssembler::dump() {
+ raw_ostream &OS = llvm::errs();
+
+ OS << "<MCAssembler\n";
+ OS << " Sections:[";
+ for (iterator it = begin(), ie = end(); it != ie; ++it) {
+ if (it != begin()) OS << ",\n ";
+ it->dump();
+ }
+ OS << "],\n";
+ OS << " Symbols:[";
+
+ for (symbol_iterator it = symbol_begin(), ie = symbol_end(); it != ie; ++it) {
+ if (it != symbol_begin()) OS << ",\n ";
+ it->dump();
+ }
+ OS << "]>\n";
+}