void MCAssembler::WriteSectionData(const MCSectionData *SD,
const MCAsmLayout &Layout,
MCObjectWriter *OW) const {
// Ignore virtual sections.
if (getBackend().isVirtualSection(SD->getSection())) {
assert(Layout.getSectionFileSize(SD) == 0 && "Invalid size for section!");
// Check that contents are only things legal inside a virtual section.
for (MCSectionData::const_iterator it = SD->begin(),
ie = SD->end(); it != ie; ++it) {
switch (it->getKind()) {
default:
assert(0 && "Invalid fragment in virtual section!");
case MCFragment::FT_Data: {
// Check that we aren't trying to write a non-zero contents (or fixups)
// into a virtual section. This is to support clients which use standard
// directives to fill the contents of virtual sections.
MCDataFragment &DF = cast<MCDataFragment>(*it);
assert(DF.fixup_begin() == DF.fixup_end() &&
"Cannot have fixups in virtual section!");
for (unsigned i = 0, e = DF.getContents().size(); i != e; ++i)
assert(DF.getContents()[i] == 0 &&
"Invalid data value for virtual section!");
break;
}
case MCFragment::FT_Align:
// Check that we aren't trying to write a non-zero value into a virtual
// section.
assert((!cast<MCAlignFragment>(it)->getValueSize() ||
!cast<MCAlignFragment>(it)->getValue()) &&
"Invalid align in virtual section!");
break;
case MCFragment::FT_Fill:
assert(!cast<MCFillFragment>(it)->getValueSize() &&
"Invalid fill in virtual section!");
break;
}
}
return;
}
uint64_t Start = OW->getStream().tell();
(void) Start;
for (MCSectionData::const_iterator it = SD->begin(),
ie = SD->end(); it != ie; ++it)
WriteFragmentData(*this, Layout, *it, OW);
assert(OW->getStream().tell() - Start == Layout.getSectionFileSize(SD));
}
void SVMELFProgramWriter::writeSectionHeader(const MCAsmLayout &Layout,
const MCSectionData *SD)
{
const MCSectionELF *SE = dyn_cast<MCSectionELF>(&SD->getSection());
assert(SE);
uint32_t sh_type = SE->getType();
uint32_t sh_link = 0;
uint32_t sh_info = 0;
// Type-specific data finds itself in sh_link and sh_info
switch (sh_type) {
case ELF::SHT_SYMTAB:
case ELF::SHT_DYNSYM:
sh_link = EMB.getStringTableIndex();
sh_info = EMB.getLastLocalSymbolIndex();
break;
default:
break;
}
Write32(EMB.getSectionStringTableIndex(SE)); // sh_name
Write32(sh_type); // sh_type
Write32(SE->getFlags()); // sh_flags
Write32(ML.getSectionMemAddress(SD)); // sh_addr
Write32(ML.getSectionDiskOffset(SD)); // sh_offset
Write32(Layout.getSectionFileSize(SD)); // sh_size
Write32(sh_link); // sh_link
Write32(sh_info); // sh_info
Write32(SD->getAlignment()); // sh_addralign
Write32(SE->getEntrySize()); // sh_entsize
}
void MCAssembler::writeSectionData(const MCSectionData *SD,
const MCAsmLayout &Layout) const {
// Ignore virtual sections.
if (SD->getSection().isVirtualSection()) {
assert(Layout.getSectionFileSize(SD) == 0 && "Invalid size for section!");
// Check that contents are only things legal inside a virtual section.
for (MCSectionData::const_iterator it = SD->begin(),
ie = SD->end(); it != ie; ++it) {
switch (it->getKind()) {
default: llvm_unreachable("Invalid fragment in virtual section!");
case MCFragment::FT_Data: {
// Check that we aren't trying to write a non-zero contents (or fixups)
// into a virtual section. This is to support clients which use standard
// directives to fill the contents of virtual sections.
const MCDataFragment &DF = cast<MCDataFragment>(*it);
assert(DF.fixup_begin() == DF.fixup_end() &&
"Cannot have fixups in virtual section!");
for (unsigned i = 0, e = DF.getContents().size(); i != e; ++i)
if (DF.getContents()[i]) {
if (auto *ELFSec = dyn_cast<const MCSectionELF>(&SD->getSection()))
report_fatal_error("non-zero initializer found in section '" +
ELFSec->getSectionName() + "'");
else
report_fatal_error("non-zero initializer found in virtual section");
}
break;
}
case MCFragment::FT_Align:
// Check that we aren't trying to write a non-zero value into a virtual
// section.
assert((cast<MCAlignFragment>(it)->getValueSize() == 0 ||
cast<MCAlignFragment>(it)->getValue() == 0) &&
"Invalid align in virtual section!");
break;
case MCFragment::FT_Fill:
assert((cast<MCFillFragment>(it)->getValueSize() == 0 ||
cast<MCFillFragment>(it)->getValue() == 0) &&
"Invalid fill in virtual section!");
break;
}
}
return;
}
uint64_t Start = getWriter().getStream().tell();
(void)Start;
for (MCSectionData::const_iterator it = SD->begin(), ie = SD->end();
it != ie; ++it)
writeFragment(*this, Layout, *it);
assert(getWriter().getStream().tell() - Start ==
Layout.getSectionAddressSize(SD));
}
void MachObjectWriter::WriteSection(const MCAssembler &Asm,
const MCAsmLayout &Layout,
const MCSectionData &SD,
uint64_t FileOffset,
uint64_t RelocationsStart,
unsigned NumRelocations) {
uint64_t SectionSize = Layout.getSectionAddressSize(&SD);
// The offset is unused for virtual sections.
if (SD.getSection().isVirtualSection()) {
assert(Layout.getSectionFileSize(&SD) == 0 && "Invalid file size!");
FileOffset = 0;
}
// struct section (68 bytes) or
// struct section_64 (80 bytes)
uint64_t Start = OS.tell();
(void) Start;
const MCSectionMachO &Section = cast<MCSectionMachO>(SD.getSection());
WriteBytes(Section.getSectionName(), 16);
WriteBytes(Section.getSegmentName(), 16);
if (is64Bit()) {
Write64(getSectionAddress(&SD)); // address
Write64(SectionSize); // size
} else {
Write32(getSectionAddress(&SD)); // address
Write32(SectionSize); // 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(IndirectSymBase.lookup(&SD)); // reserved1
Write32(Section.getStubSize()); // reserved2
if (is64Bit())
Write32(0); // reserved3
assert(OS.tell() - Start == (is64Bit() ? macho::Section64Size :
macho::Section32Size));
}
void MachObjectWriter::writeSection(const MCAsmLayout &Layout,
const MCSection &Sec, uint64_t VMAddr,
uint64_t FileOffset, unsigned Flags,
uint64_t RelocationsStart,
unsigned NumRelocations) {
uint64_t SectionSize = Layout.getSectionAddressSize(&Sec);
const MCSectionMachO &Section = cast<MCSectionMachO>(Sec);
// The offset is unused for virtual sections.
if (Section.isVirtualSection()) {
assert(Layout.getSectionFileSize(&Sec) == 0 && "Invalid file size!");
FileOffset = 0;
}
// struct section (68 bytes) or
// struct section_64 (80 bytes)
uint64_t Start = getStream().tell();
(void) Start;
writeBytes(Section.getSectionName(), 16);
writeBytes(Section.getSegmentName(), 16);
if (is64Bit()) {
write64(VMAddr); // address
write64(SectionSize); // size
} else {
write32(VMAddr); // address
write32(SectionSize); // size
}
write32(FileOffset);
assert(isPowerOf2_32(Section.getAlignment()) && "Invalid alignment!");
write32(Log2_32(Section.getAlignment()));
write32(NumRelocations ? RelocationsStart : 0);
write32(NumRelocations);
write32(Flags);
write32(IndirectSymBase.lookup(&Sec)); // reserved1
write32(Section.getStubSize()); // reserved2
if (is64Bit())
write32(0); // reserved3
assert(getStream().tell() - Start ==
(is64Bit() ? sizeof(MachO::section_64) : sizeof(MachO::section)));
}
void MachObjectWriter::writeObject(MCAssembler &Asm,
const MCAsmLayout &Layout) {
// Compute symbol table information and bind symbol indices.
computeSymbolTable(Asm, LocalSymbolData, ExternalSymbolData,
UndefinedSymbolData);
unsigned NumSections = Asm.size();
const MCAssembler::VersionMinInfoType &VersionInfo =
Layout.getAssembler().getVersionMinInfo();
// 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() ?
sizeof(MachO::segment_command_64) + NumSections * sizeof(MachO::section_64):
sizeof(MachO::segment_command) + NumSections * sizeof(MachO::section);
// Add the deployment target version info load command size, if used.
if (VersionInfo.Major != 0) {
++NumLoadCommands;
LoadCommandsSize += sizeof(MachO::version_min_command);
}
// Add the data-in-code load command size, if used.
unsigned NumDataRegions = Asm.getDataRegions().size();
if (NumDataRegions) {
++NumLoadCommands;
LoadCommandsSize += sizeof(MachO::linkedit_data_command);
}
// Add the loh load command size, if used.
uint64_t LOHRawSize = Asm.getLOHContainer().getEmitSize(*this, Layout);
uint64_t LOHSize = alignTo(LOHRawSize, is64Bit() ? 8 : 4);
if (LOHSize) {
++NumLoadCommands;
LoadCommandsSize += sizeof(MachO::linkedit_data_command);
}
// Add the symbol table load command sizes, if used.
unsigned NumSymbols = LocalSymbolData.size() + ExternalSymbolData.size() +
UndefinedSymbolData.size();
if (NumSymbols) {
NumLoadCommands += 2;
LoadCommandsSize += (sizeof(MachO::symtab_command) +
sizeof(MachO::dysymtab_command));
}
// Add the linker option load commands sizes.
for (const auto &Option : Asm.getLinkerOptions()) {
++NumLoadCommands;
LoadCommandsSize += ComputeLinkerOptionsLoadCommandSize(Option, is64Bit());
}
// Compute the total size of the section data, as well as its file size and vm
// size.
uint64_t SectionDataStart = (is64Bit() ? sizeof(MachO::mach_header_64) :
sizeof(MachO::mach_header)) + LoadCommandsSize;
uint64_t SectionDataSize = 0;
uint64_t SectionDataFileSize = 0;
uint64_t VMSize = 0;
for (const MCSection &Sec : Asm) {
uint64_t Address = getSectionAddress(&Sec);
uint64_t Size = Layout.getSectionAddressSize(&Sec);
uint64_t FileSize = Layout.getSectionFileSize(&Sec);
FileSize += getPaddingSize(&Sec, Layout);
VMSize = std::max(VMSize, Address + Size);
if (Sec.isVirtualSection())
continue;
SectionDataSize = std::max(SectionDataSize, Address + Size);
SectionDataFileSize = std::max(SectionDataFileSize, Address + FileSize);
}
// 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(MachO::MH_OBJECT, NumLoadCommands, LoadCommandsSize,
Asm.getSubsectionsViaSymbols());
uint32_t Prot =
MachO::VM_PROT_READ | MachO::VM_PROT_WRITE | MachO::VM_PROT_EXECUTE;
writeSegmentLoadCommand("", NumSections, 0, VMSize, SectionDataStart,
SectionDataSize, Prot, Prot);
// ... and then the section headers.
uint64_t RelocTableEnd = SectionDataStart + SectionDataFileSize;
for (const MCSection &Section : Asm) {
const auto &Sec = cast<MCSectionMachO>(Section);
std::vector<RelAndSymbol> &Relocs = Relocations[&Sec];
unsigned NumRelocs = Relocs.size();
uint64_t SectionStart = SectionDataStart + getSectionAddress(&Sec);
unsigned Flags = Sec.getTypeAndAttributes();
if (Sec.hasInstructions())
Flags |= MachO::S_ATTR_SOME_INSTRUCTIONS;
writeSection(Layout, Sec, getSectionAddress(&Sec), SectionStart, Flags,
RelocTableEnd, NumRelocs);
RelocTableEnd += NumRelocs * sizeof(MachO::any_relocation_info);
}
// Write out the deployment target information, if it's available.
if (VersionInfo.Major != 0) {
assert(VersionInfo.Update < 256 && "unencodable update target version");
assert(VersionInfo.Minor < 256 && "unencodable minor target version");
assert(VersionInfo.Major < 65536 && "unencodable major target version");
uint32_t EncodedVersion = VersionInfo.Update | (VersionInfo.Minor << 8) |
(VersionInfo.Major << 16);
MachO::LoadCommandType LCType;
switch (VersionInfo.Kind) {
case MCVM_OSXVersionMin:
LCType = MachO::LC_VERSION_MIN_MACOSX;
break;
case MCVM_IOSVersionMin:
LCType = MachO::LC_VERSION_MIN_IPHONEOS;
break;
case MCVM_TvOSVersionMin:
LCType = MachO::LC_VERSION_MIN_TVOS;
break;
case MCVM_WatchOSVersionMin:
LCType = MachO::LC_VERSION_MIN_WATCHOS;
break;
}
write32(LCType);
write32(sizeof(MachO::version_min_command));
write32(EncodedVersion);
write32(0); // reserved.
}
// Write the data-in-code load command, if used.
uint64_t DataInCodeTableEnd = RelocTableEnd + NumDataRegions * 8;
if (NumDataRegions) {
uint64_t DataRegionsOffset = RelocTableEnd;
uint64_t DataRegionsSize = NumDataRegions * 8;
writeLinkeditLoadCommand(MachO::LC_DATA_IN_CODE, DataRegionsOffset,
DataRegionsSize);
}
// Write the loh load command, if used.
uint64_t LOHTableEnd = DataInCodeTableEnd + LOHSize;
if (LOHSize)
writeLinkeditLoadCommand(MachO::LC_LINKER_OPTIMIZATION_HINT,
DataInCodeTableEnd, LOHSize);
// 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 = LOHTableEnd;
// The symbol table is written after the indirect symbol data.
uint64_t SymbolTableOffset = LOHTableEnd + IndirectSymbolSize;
// The string table is written after symbol table.
uint64_t StringTableOffset =
SymbolTableOffset + NumSymTabSymbols * (is64Bit() ?
sizeof(MachO::nlist_64) :
sizeof(MachO::nlist));
writeSymtabLoadCommand(SymbolTableOffset, NumSymTabSymbols,
StringTableOffset, StringTable.data().size());
writeDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols,
FirstExternalSymbol, NumExternalSymbols,
FirstUndefinedSymbol, NumUndefinedSymbols,
IndirectSymbolOffset, NumIndirectSymbols);
}
// Write the linker options load commands.
for (const auto &Option : Asm.getLinkerOptions())
writeLinkerOptionsLoadCommand(Option);
// Write the actual section data.
for (const MCSection &Sec : Asm) {
Asm.writeSectionData(&Sec, Layout);
uint64_t Pad = getPaddingSize(&Sec, Layout);
WriteZeros(Pad);
}
// Write the extra padding.
WriteZeros(SectionDataPadding);
// Write the relocation entries.
for (const MCSection &Sec : Asm) {
// Write the section relocation entries, in reverse order to match 'as'
// (approximately, the exact algorithm is more complicated than this).
std::vector<RelAndSymbol> &Relocs = Relocations[&Sec];
for (const RelAndSymbol &Rel : make_range(Relocs.rbegin(), Relocs.rend())) {
write32(Rel.MRE.r_word0);
write32(Rel.MRE.r_word1);
}
}
// Write out the data-in-code region payload, if there is one.
for (MCAssembler::const_data_region_iterator
it = Asm.data_region_begin(), ie = Asm.data_region_end();
it != ie; ++it) {
const DataRegionData *Data = &(*it);
uint64_t Start = getSymbolAddress(*Data->Start, Layout);
uint64_t End = getSymbolAddress(*Data->End, Layout);
DEBUG(dbgs() << "data in code region-- kind: " << Data->Kind
<< " start: " << Start << "(" << Data->Start->getName() << ")"
<< " end: " << End << "(" << Data->End->getName() << ")"
<< " size: " << End - Start
<< "\n");
write32(Start);
write16(End - Start);
write16(Data->Kind);
}
// Write out the loh commands, if there is one.
if (LOHSize) {
#ifndef NDEBUG
unsigned Start = getStream().tell();
#endif
Asm.getLOHContainer().emit(*this, Layout);
// Pad to a multiple of the pointer size.
writeBytes("", OffsetToAlignment(LOHRawSize, is64Bit() ? 8 : 4));
assert(getStream().tell() - Start == LOHSize);
}
// 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->Section);
if (Section.getType() == MachO::S_NON_LAZY_SYMBOL_POINTERS) {
// If this symbol is defined and internal, mark it as such.
if (it->Symbol->isDefined() && !it->Symbol->isExternal()) {
uint32_t Flags = MachO::INDIRECT_SYMBOL_LOCAL;
if (it->Symbol->isAbsolute())
Flags |= MachO::INDIRECT_SYMBOL_ABS;
write32(Flags);
continue;
}
}
write32(it->Symbol->getIndex());
}
// FIXME: Check that offsets match computed ones.
// Write the symbol table entries.
for (auto *SymbolData :
{&LocalSymbolData, &ExternalSymbolData, &UndefinedSymbolData})
for (MachSymbolData &Entry : *SymbolData)
writeNlist(Entry, Layout);
// Write the string table.
getStream() << StringTable.data();
}
}
void MachObjectWriter::WriteObject(MCAssembler &Asm,
const MCAsmLayout &Layout) {
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() ?
macho::SegmentLoadCommand64Size + NumSections * macho::Section64Size :
macho::SegmentLoadCommand32Size + NumSections * macho::Section32Size;
// Add the data-in-code load command size, if used.
unsigned NumDataRegions = Asm.getDataRegions().size();
if (NumDataRegions) {
++NumLoadCommands;
LoadCommandsSize += macho::LinkeditLoadCommandSize;
}
// Add the symbol table load command sizes, if used.
unsigned NumSymbols = LocalSymbolData.size() + ExternalSymbolData.size() +
UndefinedSymbolData.size();
if (NumSymbols) {
NumLoadCommands += 2;
LoadCommandsSize += (macho::SymtabLoadCommandSize +
macho::DysymtabLoadCommandSize);
}
// Add the linker option load commands sizes.
const std::vector<std::vector<std::string> > &LinkerOptions =
Asm.getLinkerOptions();
for (unsigned i = 0, e = LinkerOptions.size(); i != e; ++i) {
++NumLoadCommands;
LoadCommandsSize += ComputeLinkerOptionsLoadCommandSize(LinkerOptions[i],
is64Bit());
}
// Compute the total size of the section data, as well as its file size and vm
// size.
uint64_t SectionDataStart = (is64Bit() ? macho::Header64Size :
macho::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;
uint64_t Address = getSectionAddress(&SD);
uint64_t Size = Layout.getSectionAddressSize(&SD);
uint64_t FileSize = Layout.getSectionFileSize(&SD);
FileSize += getPaddingSize(&SD, Layout);
VMSize = std::max(VMSize, Address + Size);
if (SD.getSection().isVirtualSection())
continue;
SectionDataSize = std::max(SectionDataSize, Address + Size);
SectionDataFileSize = std::max(SectionDataFileSize, Address + FileSize);
}
// 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<macho::RelocationEntry> &Relocs = Relocations[it];
unsigned NumRelocs = Relocs.size();
uint64_t SectionStart = SectionDataStart + getSectionAddress(it);
WriteSection(Asm, Layout, *it, SectionStart, RelocTableEnd, NumRelocs);
RelocTableEnd += NumRelocs * macho::RelocationInfoSize;
}
// Write the data-in-code load command, if used.
uint64_t DataInCodeTableEnd = RelocTableEnd + NumDataRegions * 8;
if (NumDataRegions) {
uint64_t DataRegionsOffset = RelocTableEnd;
uint64_t DataRegionsSize = NumDataRegions * 8;
WriteLinkeditLoadCommand(macho::LCT_DataInCode, DataRegionsOffset,
DataRegionsSize);
}
// 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 = DataInCodeTableEnd;
// The symbol table is written after the indirect symbol data.
uint64_t SymbolTableOffset = DataInCodeTableEnd + IndirectSymbolSize;
// The string table is written after symbol table.
uint64_t StringTableOffset =
SymbolTableOffset + NumSymTabSymbols * (is64Bit() ? macho::Nlist64Size :
macho::Nlist32Size);
WriteSymtabLoadCommand(SymbolTableOffset, NumSymTabSymbols,
StringTableOffset, StringTable.size());
WriteDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols,
FirstExternalSymbol, NumExternalSymbols,
FirstUndefinedSymbol, NumUndefinedSymbols,
IndirectSymbolOffset, NumIndirectSymbols);
}
// Write the linker options load commands.
for (unsigned i = 0, e = LinkerOptions.size(); i != e; ++i) {
WriteLinkerOptionsLoadCommand(LinkerOptions[i]);
}
// Write the actual section data.
for (MCAssembler::const_iterator it = Asm.begin(),
ie = Asm.end(); it != ie; ++it) {
Asm.writeSectionData(it, Layout);
uint64_t Pad = getPaddingSize(it, Layout);
for (unsigned int i = 0; i < Pad; ++i)
Write8(0);
}
// 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<macho::RelocationEntry> &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 out the data-in-code region payload, if there is one.
for (MCAssembler::const_data_region_iterator
it = Asm.data_region_begin(), ie = Asm.data_region_end();
it != ie; ++it) {
const DataRegionData *Data = &(*it);
uint64_t Start =
getSymbolAddress(&Layout.getAssembler().getSymbolData(*Data->Start),
Layout);
uint64_t End =
getSymbolAddress(&Layout.getAssembler().getSymbolData(*Data->End),
Layout);
DEBUG(dbgs() << "data in code region-- kind: " << Data->Kind
<< " start: " << Start << "(" << Data->Start->getName() << ")"
<< " end: " << End << "(" << Data->End->getName() << ")"
<< " size: " << End - Start
<< "\n");
Write32(Start);
Write16(End - Start);
Write16(Data->Kind);
}
// 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 = macho::ISF_Local;
if (it->Symbol->isAbsolute())
Flags |= macho::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], Layout);
for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i)
WriteNlist(ExternalSymbolData[i], Layout);
for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i)
WriteNlist(UndefinedSymbolData[i], Layout);
// Write the string table.
OS << StringTable.str();
}
}
void MachObjectWriter::WriteObject(MCAssembler &Asm,
const MCAsmLayout &Layout) {
unsigned NumSections = Asm.size();
const MCAssembler::VersionMinInfoType &VersionInfo =
Layout.getAssembler().getVersionMinInfo();
// 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() ?
sizeof(MachO::segment_command_64) + NumSections * sizeof(MachO::section_64):
sizeof(MachO::segment_command) + NumSections * sizeof(MachO::section);
// Add the deployment target version info load command size, if used.
if (VersionInfo.Major != 0) {
++NumLoadCommands;
LoadCommandsSize += sizeof(MachO::version_min_command);
}
// Add the data-in-code load command size, if used.
unsigned NumDataRegions = Asm.getDataRegions().size();
if (NumDataRegions) {
++NumLoadCommands;
LoadCommandsSize += sizeof(MachO::linkedit_data_command);
}
// Add the loh load command size, if used.
uint64_t LOHRawSize = Asm.getLOHContainer().getEmitSize(*this, Layout);
uint64_t LOHSize = RoundUpToAlignment(LOHRawSize, is64Bit() ? 8 : 4);
if (LOHSize) {
++NumLoadCommands;
LoadCommandsSize += sizeof(MachO::linkedit_data_command);
}
// Add the symbol table load command sizes, if used.
unsigned NumSymbols = LocalSymbolData.size() + ExternalSymbolData.size() +
UndefinedSymbolData.size();
if (NumSymbols) {
NumLoadCommands += 2;
LoadCommandsSize += (sizeof(MachO::symtab_command) +
sizeof(MachO::dysymtab_command));
}
// Add the linker option load commands sizes.
const std::vector<std::vector<std::string> > &LinkerOptions =
Asm.getLinkerOptions();
for (unsigned i = 0, e = LinkerOptions.size(); i != e; ++i) {
++NumLoadCommands;
LoadCommandsSize += ComputeLinkerOptionsLoadCommandSize(LinkerOptions[i],
is64Bit());
}
// Compute the total size of the section data, as well as its file size and vm
// size.
uint64_t SectionDataStart = (is64Bit() ? sizeof(MachO::mach_header_64) :
sizeof(MachO::mach_header)) + 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;
uint64_t Address = getSectionAddress(&SD);
uint64_t Size = Layout.getSectionAddressSize(&SD);
uint64_t FileSize = Layout.getSectionFileSize(&SD);
FileSize += getPaddingSize(&SD, Layout);
VMSize = std::max(VMSize, Address + Size);
if (SD.getSection().isVirtualSection())
continue;
SectionDataSize = std::max(SectionDataSize, Address + Size);
SectionDataFileSize = std::max(SectionDataFileSize, Address + FileSize);
}
// 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<MachO::any_relocation_info> &Relocs = Relocations[it];
unsigned NumRelocs = Relocs.size();
uint64_t SectionStart = SectionDataStart + getSectionAddress(it);
WriteSection(Asm, Layout, *it, SectionStart, RelocTableEnd, NumRelocs);
RelocTableEnd += NumRelocs * sizeof(MachO::any_relocation_info);
}
// Write out the deployment target information, if it's available.
if (VersionInfo.Major != 0) {
assert(VersionInfo.Update < 256 && "unencodable update target version");
assert(VersionInfo.Minor < 256 && "unencodable minor target version");
assert(VersionInfo.Major < 65536 && "unencodable major target version");
uint32_t EncodedVersion = VersionInfo.Update | (VersionInfo.Minor << 8) |
(VersionInfo.Major << 16);
Write32(VersionInfo.Kind == MCVM_OSXVersionMin ? MachO::LC_VERSION_MIN_MACOSX :
MachO::LC_VERSION_MIN_IPHONEOS);
Write32(sizeof(MachO::version_min_command));
Write32(EncodedVersion);
Write32(0); // reserved.
}
// Write the data-in-code load command, if used.
uint64_t DataInCodeTableEnd = RelocTableEnd + NumDataRegions * 8;
if (NumDataRegions) {
uint64_t DataRegionsOffset = RelocTableEnd;
uint64_t DataRegionsSize = NumDataRegions * 8;
WriteLinkeditLoadCommand(MachO::LC_DATA_IN_CODE, DataRegionsOffset,
DataRegionsSize);
}
// Write the loh load command, if used.
uint64_t LOHTableEnd = DataInCodeTableEnd + LOHSize;
if (LOHSize)
WriteLinkeditLoadCommand(MachO::LC_LINKER_OPTIMIZATION_HINT,
DataInCodeTableEnd, LOHSize);
// 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 = LOHTableEnd;
// The symbol table is written after the indirect symbol data.
uint64_t SymbolTableOffset = LOHTableEnd + IndirectSymbolSize;
// The string table is written after symbol table.
uint64_t StringTableOffset =
SymbolTableOffset + NumSymTabSymbols * (is64Bit() ?
sizeof(MachO::nlist_64) :
sizeof(MachO::nlist));
WriteSymtabLoadCommand(SymbolTableOffset, NumSymTabSymbols,
StringTableOffset, StringTable.data().size());
WriteDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols,
FirstExternalSymbol, NumExternalSymbols,
FirstUndefinedSymbol, NumUndefinedSymbols,
IndirectSymbolOffset, NumIndirectSymbols);
}
// Write the linker options load commands.
for (unsigned i = 0, e = LinkerOptions.size(); i != e; ++i) {
WriteLinkerOptionsLoadCommand(LinkerOptions[i]);
}
// Write the actual section data.
for (MCAssembler::const_iterator it = Asm.begin(),
ie = Asm.end(); it != ie; ++it) {
Asm.writeSectionData(it, Layout);
uint64_t Pad = getPaddingSize(it, Layout);
for (unsigned int i = 0; i < Pad; ++i)
Write8(0);
}
// 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<MachO::any_relocation_info> &Relocs = Relocations[it];
for (unsigned i = 0, e = Relocs.size(); i != e; ++i) {
Write32(Relocs[e - i - 1].r_word0);
Write32(Relocs[e - i - 1].r_word1);
}
}
// Write out the data-in-code region payload, if there is one.
for (MCAssembler::const_data_region_iterator
it = Asm.data_region_begin(), ie = Asm.data_region_end();
it != ie; ++it) {
const DataRegionData *Data = &(*it);
uint64_t Start =
getSymbolAddress(&Layout.getAssembler().getSymbolData(*Data->Start),
Layout);
uint64_t End =
getSymbolAddress(&Layout.getAssembler().getSymbolData(*Data->End),
Layout);
DEBUG(dbgs() << "data in code region-- kind: " << Data->Kind
<< " start: " << Start << "(" << Data->Start->getName() << ")"
<< " end: " << End << "(" << Data->End->getName() << ")"
<< " size: " << End - Start
<< "\n");
Write32(Start);
Write16(End - Start);
Write16(Data->Kind);
}
// Write out the loh commands, if there is one.
if (LOHSize) {
#ifndef NDEBUG
unsigned Start = OS.tell();
#endif
Asm.getLOHContainer().Emit(*this, Layout);
// Pad to a multiple of the pointer size.
WriteBytes("", OffsetToAlignment(LOHRawSize, is64Bit() ? 8 : 4));
assert(OS.tell() - Start == LOHSize);
}
// 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() == MachO::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 = MachO::INDIRECT_SYMBOL_LOCAL;
if (it->Symbol->isAbsolute())
Flags |= MachO::INDIRECT_SYMBOL_ABS;
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], Layout);
for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i)
WriteNlist(ExternalSymbolData[i], Layout);
for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i)
WriteNlist(UndefinedSymbolData[i], Layout);
// Write the string table.
OS << StringTable.data();
}
}
void SVMELFProgramWriter::WriteObject(MCAssembler &Asm,
const MCAsmLayout &Layout)
{
// First pass, allocate all non-debug sections and compute the
// initial layout of the plaintext RWDATA segments.
ML.AllocateSections(Asm, Layout);
// Apply fixups that were stored in RecordRelocation
ML.ApplyLateFixups(Asm, Layout);
// Now we can know the final binary image of the RWDATA segments. Compress them.
rwCompress(Asm, Layout, ML);
ML.AllocateSections(Asm, Layout);
if (ELFDebug) {
// Allocate all debug sections last
EMB.BuildSections(Asm, Layout, ML);
ML.AllocateSections(Asm, Layout);
}
// Write header blocks
writeELFHeader(Asm, Layout);
for (int S = 0; S < SPS_DEBUG; ++S)
writeProgramHeader((SVMProgramSection) S);
// Write program data, sorted by SPS section
int endS = ELFDebug ? SPS_NUM_SECTIONS : SPS_DEBUG;
for (int S = 0; S < endS; ++S) {
if (S == SPS_DEBUG)
writeDebugMessage();
for (MCAssembler::const_iterator it = Asm.begin(), ie = Asm.end();
it != ie; ++it) {
const MCSectionData *SD = &*it;
if (ML.getSectionKind(SD) != S)
continue;
if (Layout.getSectionFileSize(SD) == 0)
continue;
padToOffset(ML.getSectionDiskOffset(SD));
Asm.WriteSectionData(SD, Layout);
}
}
if (ELFDebug) {
// On debug binaries, generate section headers last
padToOffset(SHOffset);
// Dummy NULL section header (index 0)
WriteZeros(sizeof(ELF::Elf32_Shdr));
for (MCAssembler::const_iterator it = Asm.begin(), ie = Asm.end();
it != ie; ++it) {
const MCSectionData *SD = &*it;
writeSectionHeader(Layout, SD);
}
}
}