/// WriteFragmentData - Write the \arg F data to the output file.
static void WriteFragmentData(const MCAssembler &Asm, const MCAsmLayout &Layout,
const MCFragment &F, MCObjectWriter *OW) {
uint64_t Start = OW->getStream().tell();
(void) Start;
++stats::EmittedFragments;
// FIXME: Embed in fragments instead?
uint64_t FragmentSize = Layout.getFragmentEffectiveSize(&F);
switch (F.getKind()) {
case MCFragment::FT_Align: {
MCAlignFragment &AF = cast<MCAlignFragment>(F);
uint64_t Count = FragmentSize / AF.getValueSize();
assert(AF.getValueSize() && "Invalid virtual align in concrete fragment!");
// FIXME: This error shouldn't actually occur (the front end should emit
// 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() != FragmentSize)
report_fatal_error("undefined .align directive, value size '" +
Twine(AF.getValueSize()) +
"' is not a divisor of padding size '" +
Twine(FragmentSize) + "'");
// 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 we are aligning with nops, ask that target to emit the right data.
if (AF.hasEmitNops()) {
if (!Asm.getBackend().WriteNopData(Count, OW))
report_fatal_error("unable to write nop sequence of " +
Twine(Count) + " bytes");
break;
}
// Otherwise, write out in multiples of the value size.
for (uint64_t i = 0; i != Count; ++i) {
switch (AF.getValueSize()) {
default:
assert(0 && "Invalid size!");
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);
assert(FragmentSize == DF.getContents().size() && "Invalid size!");
OW->WriteBytes(DF.getContents().str());
break;
}
case MCFragment::FT_Fill: {
MCFillFragment &FF = cast<MCFillFragment>(F);
assert(FF.getValueSize() && "Invalid virtual align in concrete fragment!");
for (uint64_t i = 0, e = FF.getSize() / FF.getValueSize(); i != e; ++i) {
switch (FF.getValueSize()) {
default:
assert(0 && "Invalid size!");
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;
}
case MCFragment::FT_Inst:
llvm_unreachable("unexpected inst fragment after lowering");
break;
case MCFragment::FT_Org: {
MCOrgFragment &OF = cast<MCOrgFragment>(F);
for (uint64_t i = 0, e = FragmentSize; i != e; ++i)
OW->Write8(uint8_t(OF.getValue()));
break;
}
}
assert(OW->getStream().tell() - Start == FragmentSize);
}
bool MachObjectWriter::isFixupKindPCRel(const MCAssembler &Asm, unsigned Kind) {
const MCFixupKindInfo &FKI = Asm.getBackend().getFixupKindInfo(
(MCFixupKind) Kind);
return FKI.Flags & MCFixupKindInfo::FKF_IsPCRel;
}
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 = 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.
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);
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 (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();
}
}
/// This function takes a symbol data object from the assembler
/// and creates the associated COFF symbol staging object.
void WinCOFFObjectWriter::DefineSymbol(MCSymbolData const &SymbolData,
MCAssembler &Assembler,
const MCAsmLayout &Layout) {
MCSymbol const &Symbol = SymbolData.getSymbol();
COFFSymbol *coff_symbol = GetOrCreateCOFFSymbol(&Symbol);
SymbolMap[&Symbol] = coff_symbol;
if (SymbolData.getFlags() & COFF::SF_WeakExternal) {
coff_symbol->Data.StorageClass = COFF::IMAGE_SYM_CLASS_WEAK_EXTERNAL;
if (Symbol.isVariable()) {
const MCSymbolRefExpr *SymRef =
dyn_cast<MCSymbolRefExpr>(Symbol.getVariableValue());
if (!SymRef)
report_fatal_error("Weak externals may only alias symbols");
coff_symbol->Other = GetOrCreateCOFFSymbol(&SymRef->getSymbol());
} else {
std::string WeakName = std::string(".weak.")
+ Symbol.getName().str()
+ ".default";
COFFSymbol *WeakDefault = createSymbol(WeakName);
WeakDefault->Data.SectionNumber = COFF::IMAGE_SYM_ABSOLUTE;
WeakDefault->Data.StorageClass = COFF::IMAGE_SYM_CLASS_EXTERNAL;
WeakDefault->Data.Type = 0;
WeakDefault->Data.Value = 0;
coff_symbol->Other = WeakDefault;
}
// Setup the Weak External auxiliary symbol.
coff_symbol->Aux.resize(1);
memset(&coff_symbol->Aux[0], 0, sizeof(coff_symbol->Aux[0]));
coff_symbol->Aux[0].AuxType = ATWeakExternal;
coff_symbol->Aux[0].Aux.WeakExternal.TagIndex = 0;
coff_symbol->Aux[0].Aux.WeakExternal.Characteristics =
COFF::IMAGE_WEAK_EXTERN_SEARCH_LIBRARY;
coff_symbol->MCData = &SymbolData;
} else {
const MCSymbolData &ResSymData =
Assembler.getSymbolData(Symbol.AliasedSymbol());
if (Symbol.isVariable()) {
int64_t Addr;
if (Symbol.getVariableValue()->EvaluateAsAbsolute(Addr, Layout))
coff_symbol->Data.Value = Addr;
} else if (SymbolData.isExternal() && SymbolData.isCommon()) {
coff_symbol->Data.Value = SymbolData.getCommonSize();
}
coff_symbol->Data.Type = (ResSymData.getFlags() & 0x0000FFFF) >> 0;
coff_symbol->Data.StorageClass = (ResSymData.getFlags() & 0x00FF0000) >> 16;
// If no storage class was specified in the streamer, define it here.
if (coff_symbol->Data.StorageClass == 0) {
bool external = ResSymData.isExternal() || !ResSymData.Fragment;
coff_symbol->Data.StorageClass =
external ? COFF::IMAGE_SYM_CLASS_EXTERNAL : COFF::IMAGE_SYM_CLASS_STATIC;
}
if (Symbol.isAbsolute() || Symbol.AliasedSymbol().isVariable())
coff_symbol->Data.SectionNumber = COFF::IMAGE_SYM_ABSOLUTE;
else if (ResSymData.Fragment)
coff_symbol->Section =
SectionMap[&ResSymData.Fragment->getParent()->getSection()];
coff_symbol->MCData = &ResSymData;
}
}
void WinCOFFObjectWriter::WriteObject(MCAssembler &Asm,
const MCAsmLayout &Layout) {
// Assign symbol and section indexes and offsets.
Header.NumberOfSections = 0;
DenseMap<COFFSection *, uint16_t> SectionIndices;
for (sections::iterator i = Sections.begin(),
e = Sections.end(); i != e; i++) {
if (Layout.getSectionAddressSize((*i)->MCData) > 0) {
size_t Number = ++Header.NumberOfSections;
SectionIndices[i->get()] = Number;
MakeSectionReal(**i, Number);
} else {
(*i)->Number = -1;
}
}
Header.NumberOfSymbols = 0;
for (symbols::iterator i = Symbols.begin(), e = Symbols.end(); i != e; i++) {
COFFSymbol &coff_symbol = **i;
MCSymbolData const *SymbolData = coff_symbol.MCData;
// Update section number & offset for symbols that have them.
if (SymbolData && SymbolData->Fragment) {
assert(coff_symbol.Section != nullptr);
coff_symbol.Data.SectionNumber = coff_symbol.Section->Number;
coff_symbol.Data.Value = Layout.getFragmentOffset(SymbolData->Fragment)
+ SymbolData->Offset;
}
if (coff_symbol.should_keep()) {
MakeSymbolReal(coff_symbol, Header.NumberOfSymbols++);
// Update auxiliary symbol info.
coff_symbol.Data.NumberOfAuxSymbols = coff_symbol.Aux.size();
Header.NumberOfSymbols += coff_symbol.Data.NumberOfAuxSymbols;
} else
coff_symbol.Index = -1;
}
// Fixup weak external references.
for (symbols::iterator i = Symbols.begin(), e = Symbols.end(); i != e; i++) {
COFFSymbol &coff_symbol = **i;
if (coff_symbol.Other) {
assert(coff_symbol.Index != -1);
assert(coff_symbol.Aux.size() == 1 &&
"Symbol must contain one aux symbol!");
assert(coff_symbol.Aux[0].AuxType == ATWeakExternal &&
"Symbol's aux symbol must be a Weak External!");
coff_symbol.Aux[0].Aux.WeakExternal.TagIndex = coff_symbol.Other->Index;
}
}
// Fixup associative COMDAT sections.
for (sections::iterator i = Sections.begin(),
e = Sections.end(); i != e; i++) {
if ((*i)->Symbol->Aux[0].Aux.SectionDefinition.Selection !=
COFF::IMAGE_COMDAT_SELECT_ASSOCIATIVE)
continue;
const MCSectionCOFF &MCSec = static_cast<const MCSectionCOFF &>(
(*i)->MCData->getSection());
COFFSection *Assoc = SectionMap.lookup(MCSec.getAssocSection());
if (!Assoc) {
report_fatal_error(Twine("Missing associated COMDAT section ") +
MCSec.getAssocSection()->getSectionName() +
" for section " + MCSec.getSectionName());
}
// Skip this section if the associated section is unused.
if (Assoc->Number == -1)
continue;
(*i)->Symbol->Aux[0].Aux.SectionDefinition.Number = SectionIndices[Assoc];
}
// Assign file offsets to COFF object file structures.
unsigned offset = 0;
offset += COFF::HeaderSize;
offset += COFF::SectionSize * Header.NumberOfSections;
for (MCAssembler::const_iterator i = Asm.begin(),
e = Asm.end();
i != e; i++) {
COFFSection *Sec = SectionMap[&i->getSection()];
if (Sec->Number == -1)
continue;
Sec->Header.SizeOfRawData = Layout.getSectionAddressSize(i);
if (IsPhysicalSection(Sec)) {
Sec->Header.PointerToRawData = offset;
offset += Sec->Header.SizeOfRawData;
}
if (Sec->Relocations.size() > 0) {
bool RelocationsOverflow = Sec->Relocations.size() >= 0xffff;
if (RelocationsOverflow) {
// Signal overflow by setting NumberOfSections to max value. Actual
// size is found in reloc #0. Microsoft tools understand this.
Sec->Header.NumberOfRelocations = 0xffff;
} else {
Sec->Header.NumberOfRelocations = Sec->Relocations.size();
}
Sec->Header.PointerToRelocations = offset;
if (RelocationsOverflow) {
// Reloc #0 will contain actual count, so make room for it.
offset += COFF::RelocationSize;
}
offset += COFF::RelocationSize * Sec->Relocations.size();
for (relocations::iterator cr = Sec->Relocations.begin(),
er = Sec->Relocations.end();
cr != er; ++cr) {
assert((*cr).Symb->Index != -1);
(*cr).Data.SymbolTableIndex = (*cr).Symb->Index;
}
}
assert(Sec->Symbol->Aux.size() == 1
&& "Section's symbol must have one aux!");
AuxSymbol &Aux = Sec->Symbol->Aux[0];
assert(Aux.AuxType == ATSectionDefinition &&
"Section's symbol's aux symbol must be a Section Definition!");
Aux.Aux.SectionDefinition.Length = Sec->Header.SizeOfRawData;
Aux.Aux.SectionDefinition.NumberOfRelocations =
Sec->Header.NumberOfRelocations;
Aux.Aux.SectionDefinition.NumberOfLinenumbers =
Sec->Header.NumberOfLineNumbers;
}
Header.PointerToSymbolTable = offset;
// We want a deterministic output. It looks like GNU as also writes 0 in here.
Header.TimeDateStamp = 0;
// Write it all to disk...
WriteFileHeader(Header);
{
sections::iterator i, ie;
MCAssembler::const_iterator j, je;
for (i = Sections.begin(), ie = Sections.end(); i != ie; i++)
if ((*i)->Number != -1) {
if ((*i)->Relocations.size() >= 0xffff) {
(*i)->Header.Characteristics |= COFF::IMAGE_SCN_LNK_NRELOC_OVFL;
}
WriteSectionHeader((*i)->Header);
}
for (i = Sections.begin(), ie = Sections.end(),
j = Asm.begin(), je = Asm.end();
(i != ie) && (j != je); ++i, ++j) {
if ((*i)->Number == -1)
continue;
if ((*i)->Header.PointerToRawData != 0) {
assert(OS.tell() == (*i)->Header.PointerToRawData &&
"Section::PointerToRawData is insane!");
Asm.writeSectionData(j, Layout);
}
if ((*i)->Relocations.size() > 0) {
assert(OS.tell() == (*i)->Header.PointerToRelocations &&
"Section::PointerToRelocations is insane!");
if ((*i)->Relocations.size() >= 0xffff) {
// In case of overflow, write actual relocation count as first
// relocation. Including the synthetic reloc itself (+ 1).
COFF::relocation r;
r.VirtualAddress = (*i)->Relocations.size() + 1;
r.SymbolTableIndex = 0;
r.Type = 0;
WriteRelocation(r);
}
for (relocations::const_iterator k = (*i)->Relocations.begin(),
ke = (*i)->Relocations.end();
k != ke; k++) {
WriteRelocation(k->Data);
}
} else
assert((*i)->Header.PointerToRelocations == 0 &&
"Section::PointerToRelocations is insane!");
}
}
assert(OS.tell() == Header.PointerToSymbolTable &&
"Header::PointerToSymbolTable is insane!");
for (symbols::iterator i = Symbols.begin(), e = Symbols.end(); i != e; i++)
if ((*i)->Index != -1)
WriteSymbol(**i);
OS.write((char const *)&Strings.Data.front(), Strings.Data.size());
}
void WinCOFFObjectWriter::RecordRelocation(const MCAssembler &Asm,
const MCAsmLayout &Layout,
const MCFragment *Fragment,
const MCFixup &Fixup,
MCValue Target,
uint64_t &FixedValue) {
assert(Target.getSymA() != NULL && "Relocation must reference a symbol!");
const MCSymbol *A = &Target.getSymA()->getSymbol();
MCSymbolData &A_SD = Asm.getSymbolData(*A);
MCSectionData const *SectionData = Fragment->getParent();
// Mark this symbol as requiring an entry in the symbol table.
assert(SectionMap.find(&SectionData->getSection()) != SectionMap.end() &&
"Section must already have been defined in ExecutePostLayoutBinding!");
assert(SymbolMap.find(&A_SD.getSymbol()) != SymbolMap.end() &&
"Symbol must already have been defined in ExecutePostLayoutBinding!");
COFFSection *coff_section = SectionMap[&SectionData->getSection()];
COFFSymbol *coff_symbol = SymbolMap[&A_SD.getSymbol()];
const MCSymbolRefExpr *SymA = Target.getSymA();
const MCSymbolRefExpr *SymB = Target.getSymB();
const bool CrossSection = SymB &&
&SymA->getSymbol().getSection() != &SymB->getSymbol().getSection();
if (Target.getSymB()) {
const MCSymbol *B = &Target.getSymB()->getSymbol();
MCSymbolData &B_SD = Asm.getSymbolData(*B);
// Offset of the symbol in the section
int64_t a = Layout.getSymbolOffset(&B_SD);
// Ofeset of the relocation in the section
int64_t b = Layout.getFragmentOffset(Fragment) + Fixup.getOffset();
FixedValue = b - a;
// In the case where we have SymbA and SymB, we just need to store the delta
// between the two symbols. Update FixedValue to account for the delta, and
// skip recording the relocation.
if (!CrossSection)
return;
} else {
FixedValue = Target.getConstant();
}
COFFRelocation Reloc;
Reloc.Data.SymbolTableIndex = 0;
Reloc.Data.VirtualAddress = Layout.getFragmentOffset(Fragment);
// Turn relocations for temporary symbols into section relocations.
if (coff_symbol->MCData->getSymbol().isTemporary() || CrossSection) {
Reloc.Symb = coff_symbol->Section->Symbol;
FixedValue += Layout.getFragmentOffset(coff_symbol->MCData->Fragment)
+ coff_symbol->MCData->getOffset();
} else
Reloc.Symb = coff_symbol;
++Reloc.Symb->Relocations;
Reloc.Data.VirtualAddress += Fixup.getOffset();
unsigned FixupKind = Fixup.getKind();
if (CrossSection)
FixupKind = FK_PCRel_4;
Reloc.Data.Type = TargetObjectWriter->getRelocType(FixupKind);
// FIXME: Can anyone explain what this does other than adjust for the size
// of the offset?
if (Reloc.Data.Type == COFF::IMAGE_REL_AMD64_REL32 ||
Reloc.Data.Type == COFF::IMAGE_REL_I386_REL32)
FixedValue += 4;
coff_section->Relocations.push_back(Reloc);
}
void AArch64MachObjectWriter::recordRelocation(
MachObjectWriter *Writer, MCAssembler &Asm, const MCAsmLayout &Layout,
const MCFragment *Fragment, const MCFixup &Fixup, MCValue Target,
uint64_t &FixedValue) {
unsigned IsPCRel = Writer->isFixupKindPCRel(Asm, Fixup.getKind());
// See <reloc.h>.
uint32_t FixupOffset = Layout.getFragmentOffset(Fragment);
unsigned Log2Size = 0;
int64_t Value = 0;
unsigned Index = 0;
unsigned Type = 0;
unsigned Kind = Fixup.getKind();
const MCSymbol *RelSymbol = nullptr;
FixupOffset += Fixup.getOffset();
// AArch64 pcrel relocation addends do not include the section offset.
if (IsPCRel)
FixedValue += FixupOffset;
// ADRP fixups use relocations for the whole symbol value and only
// put the addend in the instruction itself. Clear out any value the
// generic code figured out from the sybmol definition.
if (Kind == AArch64::fixup_aarch64_pcrel_adrp_imm21)
FixedValue = 0;
// imm19 relocations are for conditional branches, which require
// assembler local symbols. If we got here, that's not what we have,
// so complain loudly.
if (Kind == AArch64::fixup_aarch64_pcrel_branch19) {
Asm.getContext().reportError(Fixup.getLoc(),
"conditional branch requires assembler-local"
" label. '" +
Target.getSymA()->getSymbol().getName() +
"' is external.");
return;
}
// 14-bit branch relocations should only target internal labels, and so
// should never get here.
if (Kind == AArch64::fixup_aarch64_pcrel_branch14) {
Asm.getContext().reportError(Fixup.getLoc(),
"Invalid relocation on conditional branch!");
return;
}
if (!getAArch64FixupKindMachOInfo(Fixup, Type, Target.getSymA(), Log2Size,
Asm)) {
Asm.getContext().reportError(Fixup.getLoc(), "unknown AArch64 fixup kind!");
return;
}
Value = Target.getConstant();
if (Target.isAbsolute()) { // constant
// FIXME: Should this always be extern?
// SymbolNum of 0 indicates the absolute section.
Type = MachO::ARM64_RELOC_UNSIGNED;
if (IsPCRel) {
Asm.getContext().reportError(Fixup.getLoc(),
"PC relative absolute relocation!");
return;
// FIXME: x86_64 sets the type to a branch reloc here. Should we do
// something similar?
}
} else if (Target.getSymB()) { // A - B + constant
const MCSymbol *A = &Target.getSymA()->getSymbol();
const MCSymbol *A_Base = Asm.getAtom(*A);
const MCSymbol *B = &Target.getSymB()->getSymbol();
const MCSymbol *B_Base = Asm.getAtom(*B);
// Check for "_foo@got - .", which comes through here as:
// Ltmp0:
// ... _foo@got - Ltmp0
if (Target.getSymA()->getKind() == MCSymbolRefExpr::VK_GOT &&
Target.getSymB()->getKind() == MCSymbolRefExpr::VK_None &&
Layout.getSymbolOffset(*B) ==
Layout.getFragmentOffset(Fragment) + Fixup.getOffset()) {
// SymB is the PC, so use a PC-rel pointer-to-GOT relocation.
Type = MachO::ARM64_RELOC_POINTER_TO_GOT;
IsPCRel = 1;
MachO::any_relocation_info MRE;
MRE.r_word0 = FixupOffset;
MRE.r_word1 = (IsPCRel << 24) | (Log2Size << 25) | (Type << 28);
Writer->addRelocation(A_Base, Fragment->getParent(), MRE);
return;
} else if (Target.getSymA()->getKind() != MCSymbolRefExpr::VK_None ||
Target.getSymB()->getKind() != MCSymbolRefExpr::VK_None) {
// Otherwise, neither symbol can be modified.
Asm.getContext().reportError(Fixup.getLoc(),
"unsupported relocation of modified symbol");
return;
}
// We don't support PCrel relocations of differences.
if (IsPCRel) {
Asm.getContext().reportError(Fixup.getLoc(),
"unsupported pc-relative relocation of "
"difference");
return;
}
// AArch64 always uses external relocations. If there is no symbol to use as
// a base address (a local symbol with no preceding non-local symbol),
// error out.
//
// FIXME: We should probably just synthesize an external symbol and use
// that.
if (!A_Base) {
Asm.getContext().reportError(
Fixup.getLoc(),
"unsupported relocation of local symbol '" + A->getName() +
"'. Must have non-local symbol earlier in section.");
return;
}
if (!B_Base) {
Asm.getContext().reportError(
Fixup.getLoc(),
"unsupported relocation of local symbol '" + B->getName() +
"'. Must have non-local symbol earlier in section.");
return;
}
if (A_Base == B_Base && A_Base) {
Asm.getContext().reportError(
Fixup.getLoc(), "unsupported relocation with identical base");
return;
}
Value += (!A->getFragment() ? 0 : Writer->getSymbolAddress(*A, Layout)) -
(!A_Base || !A_Base->getFragment() ? 0 : Writer->getSymbolAddress(
*A_Base, Layout));
Value -= (!B->getFragment() ? 0 : Writer->getSymbolAddress(*B, Layout)) -
(!B_Base || !B_Base->getFragment() ? 0 : Writer->getSymbolAddress(
*B_Base, Layout));
Type = MachO::ARM64_RELOC_UNSIGNED;
MachO::any_relocation_info MRE;
MRE.r_word0 = FixupOffset;
MRE.r_word1 = (IsPCRel << 24) | (Log2Size << 25) | (Type << 28);
Writer->addRelocation(A_Base, Fragment->getParent(), MRE);
RelSymbol = B_Base;
Type = MachO::ARM64_RELOC_SUBTRACTOR;
} else { // A + constant
const MCSymbol *Symbol = &Target.getSymA()->getSymbol();
const MCSectionMachO &Section =
static_cast<const MCSectionMachO &>(*Fragment->getParent());
bool CanUseLocalRelocation =
canUseLocalRelocation(Section, *Symbol, Log2Size);
if (Symbol->isTemporary() && (Value || !CanUseLocalRelocation)) {
const MCSection &Sec = Symbol->getSection();
if (!Asm.getContext().getAsmInfo()->isSectionAtomizableBySymbols(Sec))
Symbol->setUsedInReloc();
}
const MCSymbol *Base = Asm.getAtom(*Symbol);
// If the symbol is a variable and we weren't able to get a Base for it
// (i.e., it's not in the symbol table associated with a section) resolve
// the relocation based its expansion instead.
if (Symbol->isVariable() && !Base) {
// If the evaluation is an absolute value, just use that directly
// to keep things easy.
int64_t Res;
if (Symbol->getVariableValue()->evaluateAsAbsolute(
Res, Layout, Writer->getSectionAddressMap())) {
FixedValue = Res;
return;
}
// FIXME: Will the Target we already have ever have any data in it
// we need to preserve and merge with the new Target? How about
// the FixedValue?
if (!Symbol->getVariableValue()->evaluateAsRelocatable(Target, &Layout,
&Fixup)) {
Asm.getContext().reportError(Fixup.getLoc(),
"unable to resolve variable '" +
Symbol->getName() + "'");
return;
}
return recordRelocation(Writer, Asm, Layout, Fragment, Fixup, Target,
FixedValue);
}
// Relocations inside debug sections always use local relocations when
// possible. This seems to be done because the debugger doesn't fully
// understand relocation entries and expects to find values that
// have already been fixed up.
if (Symbol->isInSection()) {
if (Section.hasAttribute(MachO::S_ATTR_DEBUG))
Base = nullptr;
}
// AArch64 uses external relocations as much as possible. For debug
// sections, and for pointer-sized relocations (.quad), we allow section
// relocations. It's code sections that run into trouble.
if (Base) {
RelSymbol = Base;
// Add the local offset, if needed.
if (Base != Symbol)
Value +=
Layout.getSymbolOffset(*Symbol) - Layout.getSymbolOffset(*Base);
} else if (Symbol->isInSection()) {
if (!CanUseLocalRelocation) {
Asm.getContext().reportError(
Fixup.getLoc(),
"unsupported relocation of local symbol '" + Symbol->getName() +
"'. Must have non-local symbol earlier in section.");
return;
}
// Adjust the relocation to be section-relative.
// The index is the section ordinal (1-based).
const MCSection &Sec = Symbol->getSection();
Index = Sec.getOrdinal() + 1;
Value += Writer->getSymbolAddress(*Symbol, Layout);
if (IsPCRel)
Value -= Writer->getFragmentAddress(Fragment, Layout) +
Fixup.getOffset() + (1ULL << Log2Size);
} else {
// Resolve constant variables.
if (Symbol->isVariable()) {
int64_t Res;
if (Symbol->getVariableValue()->evaluateAsAbsolute(
Res, Layout, Writer->getSectionAddressMap())) {
FixedValue = Res;
return;
}
}
Asm.getContext().reportError(Fixup.getLoc(),
"unsupported relocation of variable '" +
Symbol->getName() + "'");
return;
}
}
// If the relocation kind is Branch26, Page21, or Pageoff12, any addend
// is represented via an Addend relocation, not encoded directly into
// the instruction.
if ((Type == MachO::ARM64_RELOC_BRANCH26 ||
Type == MachO::ARM64_RELOC_PAGE21 ||
Type == MachO::ARM64_RELOC_PAGEOFF12) &&
Value) {
assert((Value & 0xff000000) == 0 && "Added relocation out of range!");
MachO::any_relocation_info MRE;
MRE.r_word0 = FixupOffset;
MRE.r_word1 =
(Index << 0) | (IsPCRel << 24) | (Log2Size << 25) | (Type << 28);
Writer->addRelocation(RelSymbol, Fragment->getParent(), MRE);
// Now set up the Addend relocation.
Type = MachO::ARM64_RELOC_ADDEND;
Index = Value;
RelSymbol = nullptr;
IsPCRel = 0;
Log2Size = 2;
// Put zero into the instruction itself. The addend is in the relocation.
Value = 0;
}
// If there's any addend left to handle, encode it in the instruction.
FixedValue = Value;
// struct relocation_info (8 bytes)
MachO::any_relocation_info MRE;
MRE.r_word0 = FixupOffset;
MRE.r_word1 =
(Index << 0) | (IsPCRel << 24) | (Log2Size << 25) | (Type << 28);
Writer->addRelocation(RelSymbol, Fragment->getParent(), MRE);
}
bool MachObjectWriter::
IsSymbolRefDifferenceFullyResolvedImpl(const MCAssembler &Asm,
const MCSymbolData &DataA,
const MCFragment &FB,
bool InSet,
bool IsPCRel) const {
if (InSet)
return true;
// The effective address is
// addr(atom(A)) + offset(A)
// - addr(atom(B)) - offset(B)
// and the offsets are not relocatable, so the fixup is fully resolved when
// addr(atom(A)) - addr(atom(B)) == 0.
const MCSymbolData *A_Base = 0, *B_Base = 0;
const MCSymbol &SA = DataA.getSymbol().AliasedSymbol();
const MCSection &SecA = SA.getSection();
const MCSection &SecB = FB.getParent()->getSection();
if (IsPCRel) {
// 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 considering symbol differences fully resolved.
if (!Asm.getBackend().hasReliableSymbolDifference()) {
if (!SA.isTemporary() || !SA.isInSection() || &SecA != &SecB)
return false;
return true;
}
// For Darwin x86_64, there is one special case when the reference IsPCRel.
// If the fragment with the reference does not have a base symbol but meets
// the simple way of dealing with this, in that it is a temporary symbol in
// the same atom then it is assumed to be fully resolved. This is needed so
// a relocation entry is not created and so the static linker does not
// mess up the reference later.
else if(!FB.getAtom() &&
SA.isTemporary() && SA.isInSection() && &SecA == &SecB){
return true;
}
} else {
if (!TargetObjectWriter->useAggressiveSymbolFolding())
return false;
}
const MCFragment *FA = Asm.getSymbolData(SA).getFragment();
// Bail if the symbol has no fragment.
if (!FA)
return false;
A_Base = FA->getAtom();
if (!A_Base)
return false;
B_Base = FB.getAtom();
if (!B_Base)
return false;
// If the atoms are the same, they are guaranteed to have the same address.
if (A_Base == B_Base)
return true;
// Otherwise, we can't prove this is fully resolved.
return false;
}
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 symbol table load command sizes, if used.
unsigned NumSymbols = LocalSymbolData.size() + ExternalSymbolData.size() +
UndefinedSymbolData.size();
if (NumSymbols) {
NumLoadCommands += 2;
LoadCommandsSize += (macho::SymtabLoadCommandSize +
macho::DysymtabLoadCommandSize);
}
// 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 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() ? macho::Nlist64Size :
macho::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, 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 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();
}
}
// If bundle alignment is used and there are any instructions in the section, it
// needs to be aligned to at least the bundle size.
static void setSectionAlignmentForBundling(const MCAssembler &Assembler,
MCSection *Section) {
if (Section && Assembler.isBundlingEnabled() && Section->hasInstructions() &&
Section->getAlignment() < Assembler.getBundleAlignSize())
Section->setAlignment(Assembler.getBundleAlignSize());
}
/// 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 MachObjectWriter::
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->getSymbol()))
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->getSymbol()))
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 ARMMachObjectWriter::
RecordARMScatteredHalfRelocation(MachObjectWriter *Writer,
const MCAssembler &Asm,
const MCAsmLayout &Layout,
const MCFragment *Fragment,
const MCFixup &Fixup,
MCValue Target,
uint64_t &FixedValue) {
uint32_t FixupOffset = Layout.getFragmentOffset(Fragment)+Fixup.getOffset();
unsigned IsPCRel = Writer->isFixupKindPCRel(Asm, Fixup.getKind());
unsigned Type = MachO::ARM_RELOC_HALF;
// See <reloc.h>.
const MCSymbol *A = &Target.getSymA()->getSymbol();
const MCSymbolData *A_SD = &Asm.getSymbolData(*A);
if (!A_SD->getFragment())
Asm.getContext().FatalError(Fixup.getLoc(),
"symbol '" + A->getName() +
"' can not be undefined in a subtraction expression");
uint32_t Value = Writer->getSymbolAddress(A_SD, Layout);
uint32_t Value2 = 0;
uint64_t SecAddr =
Writer->getSectionAddress(A_SD->getFragment()->getParent());
FixedValue += SecAddr;
if (const MCSymbolRefExpr *B = Target.getSymB()) {
const MCSymbolData *B_SD = &Asm.getSymbolData(B->getSymbol());
if (!B_SD->getFragment())
Asm.getContext().FatalError(Fixup.getLoc(),
"symbol '" + B->getSymbol().getName() +
"' can not be undefined in a subtraction expression");
// Select the appropriate difference relocation type.
Type = MachO::ARM_RELOC_HALF_SECTDIFF;
Value2 = Writer->getSymbolAddress(B_SD, Layout);
FixedValue -= Writer->getSectionAddress(B_SD->getFragment()->getParent());
}
// Relocations are written out in reverse order, so the PAIR comes first.
// ARM_RELOC_HALF and ARM_RELOC_HALF_SECTDIFF abuse the r_length field:
//
// For these two r_type relocations they always have a pair following them and
// the r_length bits are used differently. The encoding of the r_length is as
// follows:
// low bit of r_length:
// 0 - :lower16: for movw instructions
// 1 - :upper16: for movt instructions
// high bit of r_length:
// 0 - arm instructions
// 1 - thumb instructions
// the other half of the relocated expression is in the following pair
// relocation entry in the low 16 bits of r_address field.
unsigned ThumbBit = 0;
unsigned MovtBit = 0;
switch ((unsigned)Fixup.getKind()) {
default: break;
case ARM::fixup_arm_movt_hi16:
MovtBit = 1;
// The thumb bit shouldn't be set in the 'other-half' bit of the
// relocation, but it will be set in FixedValue if the base symbol
// is a thumb function. Clear it out here.
if (Asm.isThumbFunc(A))
FixedValue &= 0xfffffffe;
break;
case ARM::fixup_t2_movt_hi16:
if (Asm.isThumbFunc(A))
FixedValue &= 0xfffffffe;
MovtBit = 1;
// Fallthrough
case ARM::fixup_t2_movw_lo16:
ThumbBit = 1;
break;
}
if (Type == MachO::ARM_RELOC_HALF_SECTDIFF) {
uint32_t OtherHalf = MovtBit
? (FixedValue & 0xffff) : ((FixedValue & 0xffff0000) >> 16);
MachO::any_relocation_info MRE;
MRE.r_word0 = ((OtherHalf << 0) |
(MachO::ARM_RELOC_PAIR << 24) |
(MovtBit << 28) |
(ThumbBit << 29) |
(IsPCRel << 30) |
MachO::R_SCATTERED);
MRE.r_word1 = Value2;
Writer->addRelocation(nullptr, Fragment->getParent(), MRE);
}
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() ?
sizeof(MachO::segment_command_64) + NumSections * sizeof(MachO::section_64):
sizeof(MachO::segment_command) + NumSections * sizeof(MachO::section);
// 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 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 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 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() ?
sizeof(MachO::nlist_64) :
sizeof(MachO::nlist));
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::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 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::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.str();
}
}
void MachObjectWriter::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.
// Report errors for use of .indirect_symbol not in a symbol pointer section
// or stub section.
for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
ie = Asm.indirect_symbol_end(); it != ie; ++it) {
const MCSectionMachO &Section =
cast<MCSectionMachO>(it->SectionData->getSection());
if (Section.getType() != MCSectionMachO::S_NON_LAZY_SYMBOL_POINTERS &&
Section.getType() != MCSectionMachO::S_LAZY_SYMBOL_POINTERS &&
Section.getType() != MCSectionMachO::S_SYMBOL_STUBS) {
MCSymbol &Symbol = *it->Symbol;
report_fatal_error("indirect symbol '" + Symbol.getName() +
"' not in a symbol pointer or stub section");
}
}
// Bind non-lazy symbol pointers first.
unsigned IndirectIndex = 0;
for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) {
const MCSectionMachO &Section =
cast<MCSectionMachO>(it->SectionData->getSection());
if (Section.getType() != MCSectionMachO::S_NON_LAZY_SYMBOL_POINTERS)
continue;
// Initialize the section indirect symbol base, if necessary.
IndirectSymBase.insert(std::make_pair(it->SectionData, IndirectIndex));
Asm.getOrCreateSymbolData(*it->Symbol);
}
// Then lazy symbol pointers and symbol stubs.
IndirectIndex = 0;
for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
ie = Asm.indirect_symbol_end(); it != ie; ++it, ++IndirectIndex) {
const MCSectionMachO &Section =
cast<MCSectionMachO>(it->SectionData->getSection());
if (Section.getType() != MCSectionMachO::S_LAZY_SYMBOL_POINTERS &&
Section.getType() != MCSectionMachO::S_SYMBOL_STUBS)
continue;
// Initialize the section indirect symbol base, if necessary.
IndirectSymBase.insert(std::make_pair(it->SectionData, IndirectIndex));
// 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);
}
}
/// \brief Write the fragment \p F to the output file.
static void writeFragment(const MCAssembler &Asm, const MCAsmLayout &Layout,
const MCFragment &F) {
MCObjectWriter *OW = &Asm.getWriter();
// FIXME: Embed in fragments instead?
uint64_t FragmentSize = Asm.computeFragmentSize(Layout, F);
Asm.writeFragmentPadding(F, FragmentSize, OW);
// This variable (and its dummy usage) is to participate in the assert at
// the end of the function.
uint64_t Start = OW->getStream().tell();
(void) Start;
++stats::EmittedFragments;
switch (F.getKind()) {
case MCFragment::FT_Align: {
++stats::EmittedAlignFragments;
const MCAlignFragment &AF = cast<MCAlignFragment>(F);
assert(AF.getValueSize() && "Invalid virtual align in concrete fragment!");
uint64_t Count = FragmentSize / AF.getValueSize();
// FIXME: This error shouldn't actually occur (the front end should emit
// 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() != FragmentSize)
report_fatal_error("undefined .align directive, value size '" +
Twine(AF.getValueSize()) +
"' is not a divisor of padding size '" +
Twine(FragmentSize) + "'");
// 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 Value and ValueSize to fill the rest.
// If we are aligning with nops, ask that target to emit the right data.
if (AF.hasEmitNops()) {
if (!Asm.getBackend().writeNopData(Count, OW))
report_fatal_error("unable to write nop sequence of " +
Twine(Count) + " bytes");
break;
}
// Otherwise, write out in multiples of the value size.
for (uint64_t i = 0; i != Count; ++i) {
switch (AF.getValueSize()) {
default: llvm_unreachable("Invalid size!");
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:
++stats::EmittedDataFragments;
OW->writeBytes(cast<MCDataFragment>(F).getContents());
break;
case MCFragment::FT_Relaxable:
++stats::EmittedRelaxableFragments;
OW->writeBytes(cast<MCRelaxableFragment>(F).getContents());
break;
case MCFragment::FT_CompactEncodedInst:
++stats::EmittedCompactEncodedInstFragments;
OW->writeBytes(cast<MCCompactEncodedInstFragment>(F).getContents());
break;
case MCFragment::FT_Fill: {
++stats::EmittedFillFragments;
const MCFillFragment &FF = cast<MCFillFragment>(F);
uint8_t V = FF.getValue();
const unsigned MaxChunkSize = 16;
char Data[MaxChunkSize];
memcpy(Data, &V, 1);
for (unsigned I = 1; I < MaxChunkSize; ++I)
Data[I] = Data[0];
uint64_t Size = FF.getSize();
for (unsigned ChunkSize = MaxChunkSize; ChunkSize; ChunkSize /= 2) {
StringRef Ref(Data, ChunkSize);
for (uint64_t I = 0, E = Size / ChunkSize; I != E; ++I)
OW->writeBytes(Ref);
Size = Size % ChunkSize;
}
break;
}
case MCFragment::FT_LEB: {
const MCLEBFragment &LF = cast<MCLEBFragment>(F);
OW->writeBytes(LF.getContents());
break;
}
case MCFragment::FT_SafeSEH: {
const MCSafeSEHFragment &SF = cast<MCSafeSEHFragment>(F);
OW->write32(SF.getSymbol()->getIndex());
break;
}
case MCFragment::FT_Org: {
++stats::EmittedOrgFragments;
const MCOrgFragment &OF = cast<MCOrgFragment>(F);
for (uint64_t i = 0, e = FragmentSize; i != e; ++i)
OW->write8(uint8_t(OF.getValue()));
break;
}
case MCFragment::FT_Dwarf: {
const MCDwarfLineAddrFragment &OF = cast<MCDwarfLineAddrFragment>(F);
OW->writeBytes(OF.getContents());
break;
}
case MCFragment::FT_DwarfFrame: {
const MCDwarfCallFrameFragment &CF = cast<MCDwarfCallFrameFragment>(F);
OW->writeBytes(CF.getContents());
break;
}
case MCFragment::FT_CVInlineLines: {
const auto &OF = cast<MCCVInlineLineTableFragment>(F);
OW->writeBytes(OF.getContents());
break;
}
case MCFragment::FT_CVDefRange: {
const auto &DRF = cast<MCCVDefRangeFragment>(F);
OW->writeBytes(DRF.getContents());
break;
}
case MCFragment::FT_Dummy:
llvm_unreachable("Should not have been added");
}
assert(OW->getStream().tell() - Start == FragmentSize &&
"The stream should advance by fragment size");
}
void ELFObjectWriter::writeObject(MCAssembler &Asm,
const MCAsmLayout &Layout) {
MCContext &Ctx = Asm.getContext();
MCSectionELF *StrtabSection =
Ctx.getELFSection(".strtab", ELF::SHT_STRTAB, 0);
StringTableIndex = addToSectionTable(StrtabSection);
RevGroupMapTy RevGroupMap;
SectionIndexMapTy SectionIndexMap;
std::map<const MCSymbol *, std::vector<const MCSectionELF *>> GroupMembers;
// Write out the ELF header ...
writeHeader(Asm);
// ... then the sections ...
SectionOffsetsTy SectionOffsets;
std::vector<MCSectionELF *> Groups;
std::vector<MCSectionELF *> Relocations;
for (MCSection &Sec : Asm) {
MCSectionELF &Section = static_cast<MCSectionELF &>(Sec);
align(Section.getAlignment());
// Remember the offset into the file for this section.
uint64_t SecStart = getStream().tell();
const MCSymbolELF *SignatureSymbol = Section.getGroup();
writeSectionData(Asm, Section, Layout);
uint64_t SecEnd = getStream().tell();
SectionOffsets[&Section] = std::make_pair(SecStart, SecEnd);
MCSectionELF *RelSection = createRelocationSection(Ctx, Section);
if (SignatureSymbol) {
Asm.registerSymbol(*SignatureSymbol);
unsigned &GroupIdx = RevGroupMap[SignatureSymbol];
if (!GroupIdx) {
MCSectionELF *Group = Ctx.createELFGroupSection(SignatureSymbol);
GroupIdx = addToSectionTable(Group);
Group->setAlignment(4);
Groups.push_back(Group);
}
std::vector<const MCSectionELF *> &Members =
GroupMembers[SignatureSymbol];
Members.push_back(&Section);
if (RelSection)
Members.push_back(RelSection);
}
SectionIndexMap[&Section] = addToSectionTable(&Section);
if (RelSection) {
SectionIndexMap[RelSection] = addToSectionTable(RelSection);
Relocations.push_back(RelSection);
}
}
for (MCSectionELF *Group : Groups) {
align(Group->getAlignment());
// Remember the offset into the file for this section.
uint64_t SecStart = getStream().tell();
const MCSymbol *SignatureSymbol = Group->getGroup();
assert(SignatureSymbol);
write(uint32_t(ELF::GRP_COMDAT));
for (const MCSectionELF *Member : GroupMembers[SignatureSymbol]) {
uint32_t SecIndex = SectionIndexMap.lookup(Member);
write(SecIndex);
}
uint64_t SecEnd = getStream().tell();
SectionOffsets[Group] = std::make_pair(SecStart, SecEnd);
}
// Compute symbol table information.
computeSymbolTable(Asm, Layout, SectionIndexMap, RevGroupMap, SectionOffsets);
for (MCSectionELF *RelSection : Relocations) {
align(RelSection->getAlignment());
// Remember the offset into the file for this section.
uint64_t SecStart = getStream().tell();
writeRelocations(Asm, *RelSection->getAssociatedSection());
uint64_t SecEnd = getStream().tell();
SectionOffsets[RelSection] = std::make_pair(SecStart, SecEnd);
}
{
uint64_t SecStart = getStream().tell();
const MCSectionELF *Sec = createStringTable(Ctx);
uint64_t SecEnd = getStream().tell();
SectionOffsets[Sec] = std::make_pair(SecStart, SecEnd);
}
uint64_t NaturalAlignment = is64Bit() ? 8 : 4;
align(NaturalAlignment);
const uint64_t SectionHeaderOffset = getStream().tell();
// ... then the section header table ...
writeSectionHeader(Layout, SectionIndexMap, SectionOffsets);
uint16_t NumSections = (SectionTable.size() + 1 >= ELF::SHN_LORESERVE)
? (uint16_t)ELF::SHN_UNDEF
: SectionTable.size() + 1;
if (sys::IsLittleEndianHost != IsLittleEndian)
sys::swapByteOrder(NumSections);
unsigned NumSectionsOffset;
if (is64Bit()) {
uint64_t Val = SectionHeaderOffset;
if (sys::IsLittleEndianHost != IsLittleEndian)
sys::swapByteOrder(Val);
getStream().pwrite(reinterpret_cast<char *>(&Val), sizeof(Val),
offsetof(ELF::Elf64_Ehdr, e_shoff));
NumSectionsOffset = offsetof(ELF::Elf64_Ehdr, e_shnum);
} else {
uint32_t Val = SectionHeaderOffset;
if (sys::IsLittleEndianHost != IsLittleEndian)
sys::swapByteOrder(Val);
getStream().pwrite(reinterpret_cast<char *>(&Val), sizeof(Val),
offsetof(ELF::Elf32_Ehdr, e_shoff));
NumSectionsOffset = offsetof(ELF::Elf32_Ehdr, e_shnum);
}
getStream().pwrite(reinterpret_cast<char *>(&NumSections),
sizeof(NumSections), NumSectionsOffset);
}
void WinCOFFObjectWriter::RecordRelocation(const MCAssembler &Asm,
const MCAsmLayout &Layout,
const MCFragment *Fragment,
const MCFixup &Fixup,
MCValue Target,
bool &IsPCRel,
uint64_t &FixedValue) {
assert(Target.getSymA() && "Relocation must reference a symbol!");
const MCSymbol &Symbol = Target.getSymA()->getSymbol();
const MCSymbol &A = Symbol.AliasedSymbol();
if (!Asm.hasSymbolData(A))
Asm.getContext().FatalError(
Fixup.getLoc(),
Twine("symbol '") + A.getName() + "' can not be undefined");
MCSymbolData &A_SD = Asm.getSymbolData(A);
MCSectionData const *SectionData = Fragment->getParent();
// Mark this symbol as requiring an entry in the symbol table.
assert(SectionMap.find(&SectionData->getSection()) != SectionMap.end() &&
"Section must already have been defined in ExecutePostLayoutBinding!");
assert(SymbolMap.find(&A_SD.getSymbol()) != SymbolMap.end() &&
"Symbol must already have been defined in ExecutePostLayoutBinding!");
COFFSection *coff_section = SectionMap[&SectionData->getSection()];
COFFSymbol *coff_symbol = SymbolMap[&A_SD.getSymbol()];
const MCSymbolRefExpr *SymB = Target.getSymB();
bool CrossSection = false;
if (SymB) {
const MCSymbol *B = &SymB->getSymbol();
MCSymbolData &B_SD = Asm.getSymbolData(*B);
if (!B_SD.getFragment())
Asm.getContext().FatalError(
Fixup.getLoc(),
Twine("symbol '") + B->getName() +
"' can not be undefined in a subtraction expression");
if (!A_SD.getFragment())
Asm.getContext().FatalError(
Fixup.getLoc(),
Twine("symbol '") + Symbol.getName() +
"' can not be undefined in a subtraction expression");
CrossSection = &Symbol.getSection() != &B->getSection();
// Offset of the symbol in the section
int64_t a = Layout.getSymbolOffset(&B_SD);
// Ofeset of the relocation in the section
int64_t b = Layout.getFragmentOffset(Fragment) + Fixup.getOffset();
FixedValue = b - a;
// In the case where we have SymbA and SymB, we just need to store the delta
// between the two symbols. Update FixedValue to account for the delta, and
// skip recording the relocation.
if (!CrossSection)
return;
} else {
FixedValue = Target.getConstant();
}
COFFRelocation Reloc;
Reloc.Data.SymbolTableIndex = 0;
Reloc.Data.VirtualAddress = Layout.getFragmentOffset(Fragment);
// Turn relocations for temporary symbols into section relocations.
if (coff_symbol->MCData->getSymbol().isTemporary() || CrossSection) {
Reloc.Symb = coff_symbol->Section->Symbol;
FixedValue += Layout.getFragmentOffset(coff_symbol->MCData->Fragment)
+ coff_symbol->MCData->getOffset();
} else
Reloc.Symb = coff_symbol;
++Reloc.Symb->Relocations;
Reloc.Data.VirtualAddress += Fixup.getOffset();
Reloc.Data.Type = TargetObjectWriter->getRelocType(Target, Fixup,
CrossSection);
// FIXME: Can anyone explain what this does other than adjust for the size
// of the offset?
if ((Header.Machine == COFF::IMAGE_FILE_MACHINE_AMD64 &&
Reloc.Data.Type == COFF::IMAGE_REL_AMD64_REL32) ||
(Header.Machine == COFF::IMAGE_FILE_MACHINE_I386 &&
Reloc.Data.Type == COFF::IMAGE_REL_I386_REL32))
FixedValue += 4;
coff_section->Relocations.push_back(Reloc);
}
// It is always valid to create a relocation with a symbol. It is preferable
// to use a relocation with a section if that is possible. Using the section
// allows us to omit some local symbols from the symbol table.
bool ELFObjectWriter::shouldRelocateWithSymbol(const MCAssembler &Asm,
const MCSymbolRefExpr *RefA,
const MCSymbol *S, uint64_t C,
unsigned Type) const {
const auto *Sym = cast_or_null<MCSymbolELF>(S);
// A PCRel relocation to an absolute value has no symbol (or section). We
// represent that with a relocation to a null section.
if (!RefA)
return false;
MCSymbolRefExpr::VariantKind Kind = RefA->getKind();
switch (Kind) {
default:
break;
// The .odp creation emits a relocation against the symbol ".TOC." which
// create a R_PPC64_TOC relocation. However the relocation symbol name
// in final object creation should be NULL, since the symbol does not
// really exist, it is just the reference to TOC base for the current
// object file. Since the symbol is undefined, returning false results
// in a relocation with a null section which is the desired result.
case MCSymbolRefExpr::VK_PPC_TOCBASE:
return false;
// These VariantKind cause the relocation to refer to something other than
// the symbol itself, like a linker generated table. Since the address of
// symbol is not relevant, we cannot replace the symbol with the
// section and patch the difference in the addend.
case MCSymbolRefExpr::VK_GOT:
case MCSymbolRefExpr::VK_PLT:
case MCSymbolRefExpr::VK_GOTPCREL:
case MCSymbolRefExpr::VK_PPC_GOT_LO:
case MCSymbolRefExpr::VK_PPC_GOT_HI:
case MCSymbolRefExpr::VK_PPC_GOT_HA:
return true;
}
// An undefined symbol is not in any section, so the relocation has to point
// to the symbol itself.
assert(Sym && "Expected a symbol");
if (Sym->isUndefined())
return true;
unsigned Binding = Sym->getBinding();
switch(Binding) {
default:
llvm_unreachable("Invalid Binding");
case ELF::STB_LOCAL:
break;
case ELF::STB_WEAK:
// If the symbol is weak, it might be overridden by a symbol in another
// file. The relocation has to point to the symbol so that the linker
// can update it.
return true;
case ELF::STB_GLOBAL:
// Global ELF symbols can be preempted by the dynamic linker. The relocation
// has to point to the symbol for a reason analogous to the STB_WEAK case.
return true;
}
// If a relocation points to a mergeable section, we have to be careful.
// If the offset is zero, a relocation with the section will encode the
// same information. With a non-zero offset, the situation is different.
// For example, a relocation can point 42 bytes past the end of a string.
// If we change such a relocation to use the section, the linker would think
// that it pointed to another string and subtracting 42 at runtime will
// produce the wrong value.
auto &Sec = cast<MCSectionELF>(Sym->getSection());
unsigned Flags = Sec.getFlags();
if (Flags & ELF::SHF_MERGE) {
if (C != 0)
return true;
// It looks like gold has a bug (http://sourceware.org/PR16794) and can
// only handle section relocations to mergeable sections if using RELA.
if (!hasRelocationAddend())
return true;
}
// Most TLS relocations use a got, so they need the symbol. Even those that
// are just an offset (@tpoff), require a symbol in gold versions before
// 5efeedf61e4fe720fd3e9a08e6c91c10abb66d42 (2014-09-26) which fixed
// http://sourceware.org/PR16773.
if (Flags & ELF::SHF_TLS)
return true;
// If the symbol is a thumb function the final relocation must set the lowest
// bit. With a symbol that is done by just having the symbol have that bit
// set, so we would lose the bit if we relocated with the section.
// FIXME: We could use the section but add the bit to the relocation value.
if (Asm.isThumbFunc(Sym))
return true;
if (TargetObjectWriter->needsRelocateWithSymbol(*Sym, Type))
return true;
return false;
}
void WinCOFFObjectWriter::WriteObject(MCAssembler &Asm,
const MCAsmLayout &Layout) {
// Assign symbol and section indexes and offsets.
Header.NumberOfSections = 0;
for (sections::iterator i = Sections.begin(),
e = Sections.end(); i != e; i++) {
if (Layout.getSectionAddressSize((*i)->MCData) > 0) {
MakeSectionReal(**i, ++Header.NumberOfSections);
} else {
(*i)->Number = -1;
}
}
Header.NumberOfSymbols = 0;
for (symbols::iterator i = Symbols.begin(), e = Symbols.end(); i != e; i++) {
COFFSymbol *coff_symbol = *i;
MCSymbolData const *SymbolData = coff_symbol->MCData;
// Update section number & offset for symbols that have them.
if ((SymbolData != NULL) && (SymbolData->Fragment != NULL)) {
assert(coff_symbol->Section != NULL);
coff_symbol->Data.SectionNumber = coff_symbol->Section->Number;
coff_symbol->Data.Value = Layout.getFragmentOffset(SymbolData->Fragment)
+ SymbolData->Offset;
}
if (coff_symbol->should_keep()) {
MakeSymbolReal(*coff_symbol, Header.NumberOfSymbols++);
// Update auxiliary symbol info.
coff_symbol->Data.NumberOfAuxSymbols = coff_symbol->Aux.size();
Header.NumberOfSymbols += coff_symbol->Data.NumberOfAuxSymbols;
} else
coff_symbol->Index = -1;
}
// Fixup weak external references.
for (symbols::iterator i = Symbols.begin(), e = Symbols.end(); i != e; i++) {
COFFSymbol *coff_symbol = *i;
if (coff_symbol->Other != NULL) {
assert(coff_symbol->Index != -1);
assert(coff_symbol->Aux.size() == 1 &&
"Symbol must contain one aux symbol!");
assert(coff_symbol->Aux[0].AuxType == ATWeakExternal &&
"Symbol's aux symbol must be a Weak External!");
coff_symbol->Aux[0].Aux.WeakExternal.TagIndex = coff_symbol->Other->Index;
}
}
// Assign file offsets to COFF object file structures.
unsigned offset = 0;
offset += COFF::HeaderSize;
offset += COFF::SectionSize * Header.NumberOfSections;
for (MCAssembler::const_iterator i = Asm.begin(),
e = Asm.end();
i != e; i++) {
COFFSection *Sec = SectionMap[&i->getSection()];
if (Sec->Number == -1)
continue;
Sec->Header.SizeOfRawData = Layout.getSectionAddressSize(i);
if (IsPhysicalSection(Sec)) {
Sec->Header.PointerToRawData = offset;
offset += Sec->Header.SizeOfRawData;
}
if (Sec->Relocations.size() > 0) {
Sec->Header.NumberOfRelocations = Sec->Relocations.size();
Sec->Header.PointerToRelocations = offset;
offset += COFF::RelocationSize * Sec->Relocations.size();
for (relocations::iterator cr = Sec->Relocations.begin(),
er = Sec->Relocations.end();
cr != er; ++cr) {
assert((*cr).Symb->Index != -1);
(*cr).Data.SymbolTableIndex = (*cr).Symb->Index;
}
}
assert(Sec->Symbol->Aux.size() == 1
&& "Section's symbol must have one aux!");
AuxSymbol &Aux = Sec->Symbol->Aux[0];
assert(Aux.AuxType == ATSectionDefinition &&
"Section's symbol's aux symbol must be a Section Definition!");
Aux.Aux.SectionDefinition.Length = Sec->Header.SizeOfRawData;
Aux.Aux.SectionDefinition.NumberOfRelocations =
Sec->Header.NumberOfRelocations;
Aux.Aux.SectionDefinition.NumberOfLinenumbers =
Sec->Header.NumberOfLineNumbers;
}
Header.PointerToSymbolTable = offset;
Header.TimeDateStamp = sys::TimeValue::now().toEpochTime();
// Write it all to disk...
WriteFileHeader(Header);
{
sections::iterator i, ie;
MCAssembler::const_iterator j, je;
for (i = Sections.begin(), ie = Sections.end(); i != ie; i++)
if ((*i)->Number != -1)
WriteSectionHeader((*i)->Header);
for (i = Sections.begin(), ie = Sections.end(),
j = Asm.begin(), je = Asm.end();
(i != ie) && (j != je); ++i, ++j) {
if ((*i)->Number == -1)
continue;
if ((*i)->Header.PointerToRawData != 0) {
assert(OS.tell() == (*i)->Header.PointerToRawData &&
"Section::PointerToRawData is insane!");
Asm.writeSectionData(j, Layout);
}
if ((*i)->Relocations.size() > 0) {
assert(OS.tell() == (*i)->Header.PointerToRelocations &&
"Section::PointerToRelocations is insane!");
for (relocations::const_iterator k = (*i)->Relocations.begin(),
ke = (*i)->Relocations.end();
k != ke; k++) {
WriteRelocation(k->Data);
}
} else
assert((*i)->Header.PointerToRelocations == 0 &&
"Section::PointerToRelocations is insane!");
}
}
assert(OS.tell() == Header.PointerToSymbolTable &&
"Header::PointerToSymbolTable is insane!");
for (symbols::iterator i = Symbols.begin(), e = Symbols.end(); i != e; i++)
if ((*i)->Index != -1)
WriteSymbol(*i);
OS.write((char const *)&Strings.Data.front(), Strings.Data.size());
}
void ELFObjectWriter::recordRelocation(MCAssembler &Asm,
const MCAsmLayout &Layout,
const MCFragment *Fragment,
const MCFixup &Fixup, MCValue Target,
bool &IsPCRel, uint64_t &FixedValue) {
const MCSectionELF &FixupSection = cast<MCSectionELF>(*Fragment->getParent());
uint64_t C = Target.getConstant();
uint64_t FixupOffset = Layout.getFragmentOffset(Fragment) + Fixup.getOffset();
MCContext &Ctx = Asm.getContext();
if (const MCSymbolRefExpr *RefB = Target.getSymB()) {
assert(RefB->getKind() == MCSymbolRefExpr::VK_None &&
"Should not have constructed this");
// Let A, B and C being the components of Target and R be the location of
// the fixup. If the fixup is not pcrel, we want to compute (A - B + C).
// If it is pcrel, we want to compute (A - B + C - R).
// In general, ELF has no relocations for -B. It can only represent (A + C)
// or (A + C - R). If B = R + K and the relocation is not pcrel, we can
// replace B to implement it: (A - R - K + C)
if (IsPCRel) {
Ctx.reportError(
Fixup.getLoc(),
"No relocation available to represent this relative expression");
return;
}
const auto &SymB = cast<MCSymbolELF>(RefB->getSymbol());
if (SymB.isUndefined()) {
Ctx.reportError(Fixup.getLoc(),
Twine("symbol '") + SymB.getName() +
"' can not be undefined in a subtraction expression");
return;
}
assert(!SymB.isAbsolute() && "Should have been folded");
const MCSection &SecB = SymB.getSection();
if (&SecB != &FixupSection) {
Ctx.reportError(Fixup.getLoc(),
"Cannot represent a difference across sections");
return;
}
uint64_t SymBOffset = Layout.getSymbolOffset(SymB);
uint64_t K = SymBOffset - FixupOffset;
IsPCRel = true;
C -= K;
}
// We either rejected the fixup or folded B into C at this point.
const MCSymbolRefExpr *RefA = Target.getSymA();
const auto *SymA = RefA ? cast<MCSymbolELF>(&RefA->getSymbol()) : nullptr;
bool ViaWeakRef = false;
if (SymA && SymA->isVariable()) {
const MCExpr *Expr = SymA->getVariableValue();
if (const auto *Inner = dyn_cast<MCSymbolRefExpr>(Expr)) {
if (Inner->getKind() == MCSymbolRefExpr::VK_WEAKREF) {
SymA = cast<MCSymbolELF>(&Inner->getSymbol());
ViaWeakRef = true;
}
}
}
unsigned Type = getRelocType(Ctx, Target, Fixup, IsPCRel);
uint64_t OriginalC = C;
bool RelocateWithSymbol = shouldRelocateWithSymbol(Asm, RefA, SymA, C, Type);
if (!RelocateWithSymbol && SymA && !SymA->isUndefined())
C += Layout.getSymbolOffset(*SymA);
uint64_t Addend = 0;
if (hasRelocationAddend()) {
Addend = C;
C = 0;
}
FixedValue = C;
if (!RelocateWithSymbol) {
const MCSection *SecA =
(SymA && !SymA->isUndefined()) ? &SymA->getSection() : nullptr;
auto *ELFSec = cast_or_null<MCSectionELF>(SecA);
const auto *SectionSymbol =
ELFSec ? cast<MCSymbolELF>(ELFSec->getBeginSymbol()) : nullptr;
if (SectionSymbol)
SectionSymbol->setUsedInReloc();
ELFRelocationEntry Rec(FixupOffset, SectionSymbol, Type, Addend, SymA,
OriginalC);
Relocations[&FixupSection].push_back(Rec);
return;
}
const auto *RenamedSymA = SymA;
if (SymA) {
if (const MCSymbolELF *R = Renames.lookup(SymA))
RenamedSymA = R;
if (ViaWeakRef)
RenamedSymA->setIsWeakrefUsedInReloc();
else
RenamedSymA->setUsedInReloc();
}
ELFRelocationEntry Rec(FixupOffset, RenamedSymA, Type, Addend, SymA,
OriginalC);
Relocations[&FixupSection].push_back(Rec);
}
bool AArch64MachObjectWriter::getAArch64FixupKindMachOInfo(
const MCFixup &Fixup, unsigned &RelocType, const MCSymbolRefExpr *Sym,
unsigned &Log2Size, const MCAssembler &Asm) {
RelocType = unsigned(MachO::ARM64_RELOC_UNSIGNED);
Log2Size = ~0U;
switch ((unsigned)Fixup.getKind()) {
default:
return false;
case FK_Data_1:
Log2Size = Log2_32(1);
return true;
case FK_Data_2:
Log2Size = Log2_32(2);
return true;
case FK_Data_4:
Log2Size = Log2_32(4);
if (Sym->getKind() == MCSymbolRefExpr::VK_GOT)
RelocType = unsigned(MachO::ARM64_RELOC_POINTER_TO_GOT);
return true;
case FK_Data_8:
Log2Size = Log2_32(8);
if (Sym->getKind() == MCSymbolRefExpr::VK_GOT)
RelocType = unsigned(MachO::ARM64_RELOC_POINTER_TO_GOT);
return true;
case AArch64::fixup_aarch64_add_imm12:
case AArch64::fixup_aarch64_ldst_imm12_scale1:
case AArch64::fixup_aarch64_ldst_imm12_scale2:
case AArch64::fixup_aarch64_ldst_imm12_scale4:
case AArch64::fixup_aarch64_ldst_imm12_scale8:
case AArch64::fixup_aarch64_ldst_imm12_scale16:
Log2Size = Log2_32(4);
switch (Sym->getKind()) {
default:
return false;
case MCSymbolRefExpr::VK_PAGEOFF:
RelocType = unsigned(MachO::ARM64_RELOC_PAGEOFF12);
return true;
case MCSymbolRefExpr::VK_GOTPAGEOFF:
RelocType = unsigned(MachO::ARM64_RELOC_GOT_LOAD_PAGEOFF12);
return true;
case MCSymbolRefExpr::VK_TLVPPAGEOFF:
RelocType = unsigned(MachO::ARM64_RELOC_TLVP_LOAD_PAGEOFF12);
return true;
}
case AArch64::fixup_aarch64_pcrel_adrp_imm21:
Log2Size = Log2_32(4);
// This encompasses the relocation for the whole 21-bit value.
switch (Sym->getKind()) {
default:
Asm.getContext().reportError(Fixup.getLoc(),
"ADR/ADRP relocations must be GOT relative");
return false;
case MCSymbolRefExpr::VK_PAGE:
RelocType = unsigned(MachO::ARM64_RELOC_PAGE21);
return true;
case MCSymbolRefExpr::VK_GOTPAGE:
RelocType = unsigned(MachO::ARM64_RELOC_GOT_LOAD_PAGE21);
return true;
case MCSymbolRefExpr::VK_TLVPPAGE:
RelocType = unsigned(MachO::ARM64_RELOC_TLVP_LOAD_PAGE21);
return true;
}
return true;
case AArch64::fixup_aarch64_pcrel_branch26:
case AArch64::fixup_aarch64_pcrel_call26:
Log2Size = Log2_32(4);
RelocType = unsigned(MachO::ARM64_RELOC_BRANCH26);
return true;
}
}
void ELFObjectWriter::computeSymbolTable(
MCAssembler &Asm, const MCAsmLayout &Layout,
const SectionIndexMapTy &SectionIndexMap, const RevGroupMapTy &RevGroupMap,
SectionOffsetsTy &SectionOffsets) {
MCContext &Ctx = Asm.getContext();
SymbolTableWriter Writer(*this, is64Bit());
// Symbol table
unsigned EntrySize = is64Bit() ? ELF::SYMENTRY_SIZE64 : ELF::SYMENTRY_SIZE32;
MCSectionELF *SymtabSection =
Ctx.getELFSection(".symtab", ELF::SHT_SYMTAB, 0, EntrySize, "");
SymtabSection->setAlignment(is64Bit() ? 8 : 4);
SymbolTableIndex = addToSectionTable(SymtabSection);
align(SymtabSection->getAlignment());
uint64_t SecStart = getStream().tell();
// The first entry is the undefined symbol entry.
Writer.writeSymbol(0, 0, 0, 0, 0, 0, false);
std::vector<ELFSymbolData> LocalSymbolData;
std::vector<ELFSymbolData> ExternalSymbolData;
// Add the data for the symbols.
bool HasLargeSectionIndex = false;
for (const MCSymbol &S : Asm.symbols()) {
const auto &Symbol = cast<MCSymbolELF>(S);
bool Used = Symbol.isUsedInReloc();
bool WeakrefUsed = Symbol.isWeakrefUsedInReloc();
bool isSignature = Symbol.isSignature();
if (!isInSymtab(Layout, Symbol, Used || WeakrefUsed || isSignature,
Renames.count(&Symbol)))
continue;
if (Symbol.isTemporary() && Symbol.isUndefined()) {
Ctx.reportError(SMLoc(), "Undefined temporary symbol");
continue;
}
ELFSymbolData MSD;
MSD.Symbol = cast<MCSymbolELF>(&Symbol);
bool Local = Symbol.getBinding() == ELF::STB_LOCAL;
assert(Local || !Symbol.isTemporary());
if (Symbol.isAbsolute()) {
MSD.SectionIndex = ELF::SHN_ABS;
} else if (Symbol.isCommon()) {
assert(!Local);
MSD.SectionIndex = ELF::SHN_COMMON;
} else if (Symbol.isUndefined()) {
if (isSignature && !Used) {
MSD.SectionIndex = RevGroupMap.lookup(&Symbol);
if (MSD.SectionIndex >= ELF::SHN_LORESERVE)
HasLargeSectionIndex = true;
} else {
MSD.SectionIndex = ELF::SHN_UNDEF;
}
} else {
const MCSectionELF &Section =
static_cast<const MCSectionELF &>(Symbol.getSection());
MSD.SectionIndex = SectionIndexMap.lookup(&Section);
assert(MSD.SectionIndex && "Invalid section index!");
if (MSD.SectionIndex >= ELF::SHN_LORESERVE)
HasLargeSectionIndex = true;
}
// The @@@ in symbol version is replaced with @ in undefined symbols and @@
// in defined ones.
//
// FIXME: All name handling should be done before we get to the writer,
// including dealing with GNU-style version suffixes. Fixing this isn't
// trivial.
//
// We thus have to be careful to not perform the symbol version replacement
// blindly:
//
// The ELF format is used on Windows by the MCJIT engine. Thus, on
// Windows, the ELFObjectWriter can encounter symbols mangled using the MS
// Visual Studio C++ name mangling scheme. Symbols mangled using the MSVC
// C++ name mangling can legally have "@@@" as a sub-string. In that case,
// the EFLObjectWriter should not interpret the "@@@" sub-string as
// specifying GNU-style symbol versioning. The ELFObjectWriter therefore
// checks for the MSVC C++ name mangling prefix which is either "?", "@?",
// "__imp_?" or "__imp_@?".
//
// It would have been interesting to perform the MS mangling prefix check
// only when the target triple is of the form *-pc-windows-elf. But, it
// seems that this information is not easily accessible from the
// ELFObjectWriter.
StringRef Name = Symbol.getName();
SmallString<32> Buf;
if (!Name.startswith("?") && !Name.startswith("@?") &&
!Name.startswith("__imp_?") && !Name.startswith("__imp_@?")) {
// This symbol isn't following the MSVC C++ name mangling convention. We
// can thus safely interpret the @@@ in symbol names as specifying symbol
// versioning.
size_t Pos = Name.find("@@@");
if (Pos != StringRef::npos) {
Buf += Name.substr(0, Pos);
unsigned Skip = MSD.SectionIndex == ELF::SHN_UNDEF ? 2 : 1;
Buf += Name.substr(Pos + Skip);
Name = VersionSymSaver.save(Buf.c_str());
}
}
// Sections have their own string table
if (Symbol.getType() != ELF::STT_SECTION) {
MSD.Name = Name;
StrTabBuilder.add(Name);
}
if (Local)
LocalSymbolData.push_back(MSD);
else
ExternalSymbolData.push_back(MSD);
}
// This holds the .symtab_shndx section index.
unsigned SymtabShndxSectionIndex = 0;
if (HasLargeSectionIndex) {
MCSectionELF *SymtabShndxSection =
Ctx.getELFSection(".symtab_shndxr", ELF::SHT_SYMTAB_SHNDX, 0, 4, "");
SymtabShndxSectionIndex = addToSectionTable(SymtabShndxSection);
SymtabShndxSection->setAlignment(4);
}
ArrayRef<std::string> FileNames = Asm.getFileNames();
for (const std::string &Name : FileNames)
StrTabBuilder.add(Name);
StrTabBuilder.finalize();
for (const std::string &Name : FileNames)
Writer.writeSymbol(StrTabBuilder.getOffset(Name),
ELF::STT_FILE | ELF::STB_LOCAL, 0, 0, ELF::STV_DEFAULT,
ELF::SHN_ABS, true);
// Symbols are required to be in lexicographic order.
array_pod_sort(LocalSymbolData.begin(), LocalSymbolData.end());
array_pod_sort(ExternalSymbolData.begin(), ExternalSymbolData.end());
// Set the symbol indices. Local symbols must come before all other
// symbols with non-local bindings.
unsigned Index = FileNames.size() + 1;
for (ELFSymbolData &MSD : LocalSymbolData) {
unsigned StringIndex = MSD.Symbol->getType() == ELF::STT_SECTION
? 0
: StrTabBuilder.getOffset(MSD.Name);
MSD.Symbol->setIndex(Index++);
writeSymbol(Writer, StringIndex, MSD, Layout);
}
// Write the symbol table entries.
LastLocalSymbolIndex = Index;
for (ELFSymbolData &MSD : ExternalSymbolData) {
unsigned StringIndex = StrTabBuilder.getOffset(MSD.Name);
MSD.Symbol->setIndex(Index++);
writeSymbol(Writer, StringIndex, MSD, Layout);
assert(MSD.Symbol->getBinding() != ELF::STB_LOCAL);
}
uint64_t SecEnd = getStream().tell();
SectionOffsets[SymtabSection] = std::make_pair(SecStart, SecEnd);
ArrayRef<uint32_t> ShndxIndexes = Writer.getShndxIndexes();
if (ShndxIndexes.empty()) {
assert(SymtabShndxSectionIndex == 0);
return;
}
assert(SymtabShndxSectionIndex != 0);
SecStart = getStream().tell();
const MCSectionELF *SymtabShndxSection =
SectionTable[SymtabShndxSectionIndex - 1];
for (uint32_t Index : ShndxIndexes)
write(Index);
SecEnd = getStream().tell();
SectionOffsets[SymtabShndxSection] = std::make_pair(SecStart, SecEnd);
}
void WinCOFFObjectWriter::RecordRelocation(const MCAssembler &Asm,
const MCAsmLayout &Layout,
const MCFragment *Fragment,
const MCFixup &Fixup,
MCValue Target,
bool &IsPCRel,
uint64_t &FixedValue) {
assert(Target.getSymA() && "Relocation must reference a symbol!");
const MCSymbol &Symbol = Target.getSymA()->getSymbol();
const MCSymbol &A = Symbol.AliasedSymbol();
if (!Asm.hasSymbolData(A))
Asm.getContext().FatalError(
Fixup.getLoc(),
Twine("symbol '") + A.getName() + "' can not be undefined");
const MCSymbolData &A_SD = Asm.getSymbolData(A);
MCSectionData const *SectionData = Fragment->getParent();
// Mark this symbol as requiring an entry in the symbol table.
assert(SectionMap.find(&SectionData->getSection()) != SectionMap.end() &&
"Section must already have been defined in ExecutePostLayoutBinding!");
assert(SymbolMap.find(&A_SD.getSymbol()) != SymbolMap.end() &&
"Symbol must already have been defined in ExecutePostLayoutBinding!");
COFFSection *coff_section = SectionMap[&SectionData->getSection()];
COFFSymbol *coff_symbol = SymbolMap[&A_SD.getSymbol()];
const MCSymbolRefExpr *SymB = Target.getSymB();
bool CrossSection = false;
if (SymB) {
const MCSymbol *B = &SymB->getSymbol();
const MCSymbolData &B_SD = Asm.getSymbolData(*B);
if (!B_SD.getFragment())
Asm.getContext().FatalError(
Fixup.getLoc(),
Twine("symbol '") + B->getName() +
"' can not be undefined in a subtraction expression");
if (!A_SD.getFragment())
Asm.getContext().FatalError(
Fixup.getLoc(),
Twine("symbol '") + Symbol.getName() +
"' can not be undefined in a subtraction expression");
CrossSection = &Symbol.getSection() != &B->getSection();
// Offset of the symbol in the section
int64_t a = Layout.getSymbolOffset(&B_SD);
// Ofeset of the relocation in the section
int64_t b = Layout.getFragmentOffset(Fragment) + Fixup.getOffset();
FixedValue = b - a;
// In the case where we have SymbA and SymB, we just need to store the delta
// between the two symbols. Update FixedValue to account for the delta, and
// skip recording the relocation.
if (!CrossSection)
return;
} else {
FixedValue = Target.getConstant();
}
COFFRelocation Reloc;
Reloc.Data.SymbolTableIndex = 0;
Reloc.Data.VirtualAddress = Layout.getFragmentOffset(Fragment);
// Turn relocations for temporary symbols into section relocations.
if (coff_symbol->MCData->getSymbol().isTemporary() || CrossSection) {
Reloc.Symb = coff_symbol->Section->Symbol;
FixedValue += Layout.getFragmentOffset(coff_symbol->MCData->Fragment)
+ coff_symbol->MCData->getOffset();
} else
Reloc.Symb = coff_symbol;
++Reloc.Symb->Relocations;
Reloc.Data.VirtualAddress += Fixup.getOffset();
Reloc.Data.Type = TargetObjectWriter->getRelocType(Target, Fixup,
CrossSection);
// FIXME: Can anyone explain what this does other than adjust for the size
// of the offset?
if ((Header.Machine == COFF::IMAGE_FILE_MACHINE_AMD64 &&
Reloc.Data.Type == COFF::IMAGE_REL_AMD64_REL32) ||
(Header.Machine == COFF::IMAGE_FILE_MACHINE_I386 &&
Reloc.Data.Type == COFF::IMAGE_REL_I386_REL32))
FixedValue += 4;
if (Header.Machine == COFF::IMAGE_FILE_MACHINE_ARMNT) {
switch (Reloc.Data.Type) {
case COFF::IMAGE_REL_ARM_ABSOLUTE:
case COFF::IMAGE_REL_ARM_ADDR32:
case COFF::IMAGE_REL_ARM_ADDR32NB:
case COFF::IMAGE_REL_ARM_TOKEN:
case COFF::IMAGE_REL_ARM_SECTION:
case COFF::IMAGE_REL_ARM_SECREL:
break;
case COFF::IMAGE_REL_ARM_BRANCH11:
case COFF::IMAGE_REL_ARM_BLX11:
// IMAGE_REL_ARM_BRANCH11 and IMAGE_REL_ARM_BLX11 are only used for
// pre-ARMv7, which implicitly rules it out of ARMNT (it would be valid
// for Windows CE).
case COFF::IMAGE_REL_ARM_BRANCH24:
case COFF::IMAGE_REL_ARM_BLX24:
case COFF::IMAGE_REL_ARM_MOV32A:
// IMAGE_REL_ARM_BRANCH24, IMAGE_REL_ARM_BLX24, IMAGE_REL_ARM_MOV32A are
// only used for ARM mode code, which is documented as being unsupported
// by Windows on ARM. Emperical proof indicates that masm is able to
// generate the relocations however the rest of the MSVC toolchain is
// unable to handle it.
llvm_unreachable("unsupported relocation");
break;
case COFF::IMAGE_REL_ARM_MOV32T:
break;
case COFF::IMAGE_REL_ARM_BRANCH20T:
case COFF::IMAGE_REL_ARM_BRANCH24T:
case COFF::IMAGE_REL_ARM_BLX23T:
// IMAGE_REL_BRANCH20T, IMAGE_REL_ARM_BRANCH24T, IMAGE_REL_ARM_BLX23T all
// perform a 4 byte adjustment to the relocation. Relative branches are
// offset by 4 on ARM, however, because there is no RELA relocations, all
// branches are offset by 4.
FixedValue = FixedValue + 4;
break;
}
}
coff_section->Relocations.push_back(Reloc);
}
void WinCOFFObjectWriter::RecordRelocation(const MCAssembler &Asm,
const MCAsmLayout &Layout,
const MCFragment *Fragment,
const MCFixup &Fixup,
MCValue Target,
uint64_t &FixedValue) {
assert(Target.getSymA() != NULL && "Relocation must reference a symbol!");
const MCSymbol *A = &Target.getSymA()->getSymbol();
MCSymbolData &A_SD = Asm.getSymbolData(*A);
MCSectionData const *SectionData = Fragment->getParent();
// Mark this symbol as requiring an entry in the symbol table.
assert(SectionMap.find(&SectionData->getSection()) != SectionMap.end() &&
"Section must already have been defined in ExecutePostLayoutBinding!");
assert(SymbolMap.find(&A_SD.getSymbol()) != SymbolMap.end() &&
"Symbol must already have been defined in ExecutePostLayoutBinding!");
COFFSection *coff_section = SectionMap[&SectionData->getSection()];
COFFSymbol *coff_symbol = SymbolMap[&A_SD.getSymbol()];
if (Target.getSymB()) {
if (&Target.getSymA()->getSymbol().getSection()
!= &Target.getSymB()->getSymbol().getSection()) {
llvm_unreachable("Symbol relative relocations are only allowed between "
"symbols in the same section");
}
const MCSymbol *B = &Target.getSymB()->getSymbol();
MCSymbolData &B_SD = Asm.getSymbolData(*B);
FixedValue = Layout.getSymbolOffset(&A_SD) - Layout.getSymbolOffset(&B_SD);
// In the case where we have SymbA and SymB, we just need to store the delta
// between the two symbols. Update FixedValue to account for the delta, and
// skip recording the relocation.
return;
} else {
FixedValue = Target.getConstant();
}
COFFRelocation Reloc;
Reloc.Data.SymbolTableIndex = 0;
Reloc.Data.VirtualAddress = Layout.getFragmentOffset(Fragment);
// Turn relocations for temporary symbols into section relocations.
if (coff_symbol->MCData->getSymbol().isTemporary()) {
Reloc.Symb = coff_symbol->Section->Symbol;
FixedValue += Layout.getFragmentOffset(coff_symbol->MCData->Fragment)
+ coff_symbol->MCData->getOffset();
} else
Reloc.Symb = coff_symbol;
++Reloc.Symb->Relocations;
Reloc.Data.VirtualAddress += Fixup.getOffset();
switch ((unsigned)Fixup.getKind()) {
case FK_PCRel_4:
case X86::reloc_riprel_4byte:
case X86::reloc_riprel_4byte_movq_load:
Reloc.Data.Type = Is64Bit ? COFF::IMAGE_REL_AMD64_REL32
: COFF::IMAGE_REL_I386_REL32;
// FIXME: Can anyone explain what this does other than adjust for the size
// of the offset?
FixedValue += 4;
break;
case FK_Data_4:
case X86::reloc_signed_4byte:
Reloc.Data.Type = Is64Bit ? COFF::IMAGE_REL_AMD64_ADDR32
: COFF::IMAGE_REL_I386_DIR32;
break;
case FK_Data_8:
if (Is64Bit)
Reloc.Data.Type = COFF::IMAGE_REL_AMD64_ADDR64;
else
llvm_unreachable("unsupported relocation type");
break;
default:
llvm_unreachable("unsupported relocation type");
}
coff_section->Relocations.push_back(Reloc);
}
/// computeSymbolTable - Compute the symbol table data
void MachObjectWriter::computeSymbolTable(
MCAssembler &Asm, 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] = Index;
assert(Index <= 256 && "Too many sections!");
// Build the string table.
for (const MCSymbol &Symbol : Asm.symbols()) {
if (!Asm.isSymbolLinkerVisible(Symbol))
continue;
StringTable.add(Symbol.getName());
}
StringTable.finalize();
// Build the symbol arrays but only for non-local symbols.
//
// The particular order that we collect and 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 (const MCSymbol &Symbol : Asm.symbols()) {
// Ignore non-linker visible symbols.
if (!Asm.isSymbolLinkerVisible(Symbol))
continue;
if (!Symbol.isExternal() && !Symbol.isUndefined())
continue;
MachSymbolData MSD;
MSD.Symbol = &Symbol;
MSD.StringIndex = StringTable.getOffset(Symbol.getName());
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 (const MCSymbol &Symbol : Asm.symbols()) {
// Ignore non-linker visible symbols.
if (!Asm.isSymbolLinkerVisible(Symbol))
continue;
if (Symbol.isExternal() || Symbol.isUndefined())
continue;
MachSymbolData MSD;
MSD.Symbol = &Symbol;
MSD.StringIndex = StringTable.getOffset(Symbol.getName());
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 (auto *SymbolData :
{&LocalSymbolData, &ExternalSymbolData, &UndefinedSymbolData})
for (MachSymbolData &Entry : *SymbolData)
Entry.Symbol->setIndex(Index++);
for (const MCSection &Section : Asm) {
for (RelAndSymbol &Rel : Relocations[&Section]) {
if (!Rel.Sym)
continue;
// Set the Index and the IsExtern bit.
unsigned Index = Rel.Sym->getIndex();
assert(isInt<24>(Index));
if (IsLittleEndian)
Rel.MRE.r_word1 = (Rel.MRE.r_word1 & (~0U << 24)) | Index | (1 << 27);
else
Rel.MRE.r_word1 = (Rel.MRE.r_word1 & 0xff) | Index << 8 | (1 << 4);
}
}
}
void X86MachObjectWriter::RecordX86_64Relocation(
MachObjectWriter *Writer, MCAssembler &Asm, const MCAsmLayout &Layout,
const MCFragment *Fragment, const MCFixup &Fixup, MCValue Target,
uint64_t &FixedValue) {
unsigned IsPCRel = Writer->isFixupKindPCRel(Asm, Fixup.getKind());
unsigned IsRIPRel = isFixupKindRIPRel(Fixup.getKind());
unsigned Log2Size = getFixupKindLog2Size(Fixup.getKind());
// See <reloc.h>.
uint32_t FixupOffset =
Layout.getFragmentOffset(Fragment) + Fixup.getOffset();
uint32_t FixupAddress =
Writer->getFragmentAddress(Fragment, Layout) + Fixup.getOffset();
int64_t Value = 0;
unsigned Index = 0;
unsigned IsExtern = 0;
unsigned Type = 0;
const MCSymbol *RelSymbol = nullptr;
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 accommodated for (see comment below
// regarding SIGNED{1,2,4}), so it isn't exactly that either.
Value += 1LL << Log2Size;
}
if (Target.isAbsolute()) { // constant
// SymbolNum of 0 indicates the absolute section.
Type = MachO::X86_64_RELOC_UNSIGNED;
// 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 = MachO::X86_64_RELOC_BRANCH;
}
} else if (Target.getSymB()) { // A - B + constant
const MCSymbol *A = &Target.getSymA()->getSymbol();
if (A->isTemporary())
A = &Writer->findAliasedSymbol(*A);
const MCSymbol *A_Base = Asm.getAtom(*A);
const MCSymbol *B = &Target.getSymB()->getSymbol();
if (B->isTemporary())
B = &Writer->findAliasedSymbol(*B);
const MCSymbol *B_Base = Asm.getAtom(*B);
// Neither symbol can be modified.
if (Target.getSymA()->getKind() != MCSymbolRefExpr::VK_None) {
Asm.getContext().reportError(Fixup.getLoc(),
"unsupported relocation of modified symbol");
return;
}
// We don't support PCrel relocations of differences. Darwin 'as' doesn't
// implement most of these correctly.
if (IsPCRel) {
Asm.getContext().reportError(
Fixup.getLoc(), "unsupported pc-relative relocation of difference");
return;
}
// The support for the situation where one or both of the symbols would
// require a local relocation is handled just like if the symbols were
// external. This is certainly used in the case of debug sections where the
// section has only temporary symbols and thus the symbols don't have base
// symbols. This is encoded using the section ordinal and non-extern
// relocation entries.
// Darwin 'as' doesn't emit correct relocations for this (it ends up with a
// single SIGNED relocation); reject it for now. Except the case where both
// symbols don't have a base, equal but both NULL.
if (A_Base == B_Base && A_Base) {
Asm.getContext().reportError(
Fixup.getLoc(), "unsupported relocation with identical base");
return;
}
// A subtraction expression where either symbol is undefined is a
// non-relocatable expression.
if (A->isUndefined() || B->isUndefined()) {
StringRef Name = A->isUndefined() ? A->getName() : B->getName();
Asm.getContext().reportError(Fixup.getLoc(),
"unsupported relocation with subtraction expression, symbol '" +
Name + "' can not be undefined in a subtraction expression");
return;
}
Value += Writer->getSymbolAddress(*A, Layout) -
(!A_Base ? 0 : Writer->getSymbolAddress(*A_Base, Layout));
Value -= Writer->getSymbolAddress(*B, Layout) -
(!B_Base ? 0 : Writer->getSymbolAddress(*B_Base, Layout));
if (!A_Base)
Index = A->getFragment()->getParent()->getOrdinal() + 1;
Type = MachO::X86_64_RELOC_UNSIGNED;
MachO::any_relocation_info MRE;
MRE.r_word0 = FixupOffset;
MRE.r_word1 =
(Index << 0) | (IsPCRel << 24) | (Log2Size << 25) | (Type << 28);
Writer->addRelocation(A_Base, Fragment->getParent(), MRE);
if (B_Base)
RelSymbol = B_Base;
else
Index = B->getFragment()->getParent()->getOrdinal() + 1;
Type = MachO::X86_64_RELOC_SUBTRACTOR;
} else {
const MCSymbol *Symbol = &Target.getSymA()->getSymbol();
if (Symbol->isTemporary() && Value) {
const MCSection &Sec = Symbol->getSection();
if (!Asm.getContext().getAsmInfo()->isSectionAtomizableBySymbols(Sec))
Symbol->setUsedInReloc();
}
RelSymbol = Asm.getAtom(*Symbol);
// Relocations inside debug sections always use local relocations when
// possible. This seems to be done because the debugger doesn't fully
// understand x86_64 relocation entries, and expects to find values that
// have already been fixed up.
if (Symbol->isInSection()) {
const MCSectionMachO &Section =
static_cast<const MCSectionMachO &>(*Fragment->getParent());
if (Section.hasAttribute(MachO::S_ATTR_DEBUG))
RelSymbol = nullptr;
}
// x86_64 almost always uses external relocations, except when there is no
// symbol to use as a base address (a local symbol with no preceding
// non-local symbol).
if (RelSymbol) {
// Add the local offset, if needed.
if (RelSymbol != Symbol)
Value += Layout.getSymbolOffset(*Symbol) -
Layout.getSymbolOffset(*RelSymbol);
} else if (Symbol->isInSection() && !Symbol->isVariable()) {
// The index is the section ordinal (1-based).
Index = Symbol->getFragment()->getParent()->getOrdinal() + 1;
Value += Writer->getSymbolAddress(*Symbol, Layout);
if (IsPCRel)
Value -= FixupAddress + (1 << Log2Size);
} else if (Symbol->isVariable()) {
const MCExpr *Value = Symbol->getVariableValue();
int64_t Res;
bool isAbs = Value->evaluateAsAbsolute(Res, Layout,
Writer->getSectionAddressMap());
if (isAbs) {
FixedValue = Res;
return;
} else {
Asm.getContext().reportError(Fixup.getLoc(),
"unsupported relocation of variable '" +
Symbol->getName() + "'");
return;
}
} else {
Asm.getContext().reportError(
Fixup.getLoc(), "unsupported relocation of undefined symbol '" +
Symbol->getName() + "'");
return;
}
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.getKind()) == X86::reloc_riprel_4byte_movq_load)
Type = MachO::X86_64_RELOC_GOT_LOAD;
else
Type = MachO::X86_64_RELOC_GOT;
} else if (Modifier == MCSymbolRefExpr::VK_TLVP) {
Type = MachO::X86_64_RELOC_TLV;
} else if (Modifier != MCSymbolRefExpr::VK_None) {
Asm.getContext().reportError(
Fixup.getLoc(), "unsupported symbol modifier in relocation");
return;
} else {
Type = MachO::X86_64_RELOC_SIGNED;
// 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.
switch (-(Target.getConstant() + (1LL << Log2Size))) {
case 1: Type = MachO::X86_64_RELOC_SIGNED_1; break;
case 2: Type = MachO::X86_64_RELOC_SIGNED_2; break;
case 4: Type = MachO::X86_64_RELOC_SIGNED_4; break;
}
}
} else {
if (Modifier != MCSymbolRefExpr::VK_None) {
Asm.getContext().reportError(
Fixup.getLoc(),
"unsupported symbol modifier in branch relocation");
return;
}
Type = MachO::X86_64_RELOC_BRANCH;
}
} else {
if (Modifier == MCSymbolRefExpr::VK_GOT) {
Type = MachO::X86_64_RELOC_GOT;
} else if (Modifier == MCSymbolRefExpr::VK_GOTPCREL) {
// GOTPCREL is allowed as a modifier on non-PCrel instructions, in which
// case all we do is set the PCrel bit in the relocation entry; this is
// used with exception handling, for example. The source is required to
// include any necessary offset directly.
Type = MachO::X86_64_RELOC_GOT;
IsPCRel = 1;
} else if (Modifier == MCSymbolRefExpr::VK_TLVP) {
Asm.getContext().reportError(
Fixup.getLoc(), "TLVP symbol modifier should have been rip-rel");
return;
} else if (Modifier != MCSymbolRefExpr::VK_None) {
Asm.getContext().reportError(
Fixup.getLoc(), "unsupported symbol modifier in relocation");
return;
} else {
Type = MachO::X86_64_RELOC_UNSIGNED;
unsigned Kind = Fixup.getKind();
if (Kind == X86::reloc_signed_4byte) {
Asm.getContext().reportError(
Fixup.getLoc(),
"32-bit absolute addressing is not supported in 64-bit mode");
return;
}
}
}
}
// x86_64 always writes custom values into the fixups.
FixedValue = Value;
// struct relocation_info (8 bytes)
MachO::any_relocation_info MRE;
MRE.r_word0 = FixupOffset;
MRE.r_word1 = (Index << 0) | (IsPCRel << 24) | (Log2Size << 25) |
(IsExtern << 27) | (Type << 28);
Writer->addRelocation(RelSymbol, Fragment->getParent(), MRE);
}
bool MachObjectWriter::isSymbolRefDifferenceFullyResolvedImpl(
const MCAssembler &Asm, const MCSymbol &SymA, const MCFragment &FB,
bool InSet, bool IsPCRel) const {
if (InSet)
return true;
// The effective address is
// addr(atom(A)) + offset(A)
// - addr(atom(B)) - offset(B)
// and the offsets are not relocatable, so the fixup is fully resolved when
// addr(atom(A)) - addr(atom(B)) == 0.
const MCSymbol &SA = findAliasedSymbol(SymA);
const MCSection &SecA = SA.getSection();
const MCSection &SecB = *FB.getParent();
if (IsPCRel) {
// 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 considering symbol differences fully resolved.
//
// If the file isn't using sub-sections-via-symbols, we can make the
// same assumptions about any symbol that we normally make about
// assembler locals.
bool hasReliableSymbolDifference = isX86_64();
if (!hasReliableSymbolDifference) {
if (!SA.isInSection() || &SecA != &SecB ||
(!SA.isTemporary() && FB.getAtom() != SA.getFragment()->getAtom() &&
Asm.getSubsectionsViaSymbols()))
return false;
return true;
}
// For Darwin x86_64, there is one special case when the reference IsPCRel.
// If the fragment with the reference does not have a base symbol but meets
// the simple way of dealing with this, in that it is a temporary symbol in
// the same atom then it is assumed to be fully resolved. This is needed so
// a relocation entry is not created and so the static linker does not
// mess up the reference later.
else if(!FB.getAtom() &&
SA.isTemporary() && SA.isInSection() && &SecA == &SecB){
return true;
}
}
// If they are not in the same section, we can't compute the diff.
if (&SecA != &SecB)
return false;
const MCFragment *FA = SA.getFragment();
// Bail if the symbol has no fragment.
if (!FA)
return false;
// If the atoms are the same, they are guaranteed to have the same address.
if (FA->getAtom() == FB.getAtom())
return true;
// Otherwise, we can't prove this is fully resolved.
return false;
}
bool X86MachObjectWriter::recordScatteredRelocation(MachObjectWriter *Writer,
const MCAssembler &Asm,
const MCAsmLayout &Layout,
const MCFragment *Fragment,
const MCFixup &Fixup,
MCValue Target,
unsigned Log2Size,
uint64_t &FixedValue) {
uint64_t OriginalFixedValue = FixedValue;
uint32_t FixupOffset = Layout.getFragmentOffset(Fragment)+Fixup.getOffset();
unsigned IsPCRel = Writer->isFixupKindPCRel(Asm, Fixup.getKind());
unsigned Type = MachO::GENERIC_RELOC_VANILLA;
// See <reloc.h>.
const MCSymbol *A = &Target.getSymA()->getSymbol();
if (!A->getFragment()) {
Asm.getContext().reportError(
Fixup.getLoc(),
"symbol '" + A->getName() +
"' can not be undefined in a subtraction expression");
return false;
}
uint32_t Value = Writer->getSymbolAddress(*A, Layout);
uint64_t SecAddr = Writer->getSectionAddress(A->getFragment()->getParent());
FixedValue += SecAddr;
uint32_t Value2 = 0;
if (const MCSymbolRefExpr *B = Target.getSymB()) {
const MCSymbol *SB = &B->getSymbol();
if (!SB->getFragment()) {
Asm.getContext().reportError(
Fixup.getLoc(),
"symbol '" + SB->getName() +
"' can not be undefined in a subtraction expression");
return false;
}
// 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->isExternal() ? (unsigned)MachO::GENERIC_RELOC_SECTDIFF
: (unsigned)MachO::GENERIC_RELOC_LOCAL_SECTDIFF;
Value2 = Writer->getSymbolAddress(*SB, Layout);
FixedValue -= Writer->getSectionAddress(SB->getFragment()->getParent());
}
// Relocations are written out in reverse order, so the PAIR comes first.
if (Type == MachO::GENERIC_RELOC_SECTDIFF ||
Type == MachO::GENERIC_RELOC_LOCAL_SECTDIFF) {
// If the offset is too large to fit in a scattered relocation,
// we're hosed. It's an unfortunate limitation of the MachO format.
if (FixupOffset > 0xffffff) {
char Buffer[32];
format("0x%x", FixupOffset).print(Buffer, sizeof(Buffer));
Asm.getContext().reportError(Fixup.getLoc(),
Twine("Section too large, can't encode "
"r_address (") + Buffer +
") into 24 bits of scattered "
"relocation entry.");
return false;
}
MachO::any_relocation_info MRE;
MRE.r_word0 = ((0 << 0) | // r_address
(MachO::GENERIC_RELOC_PAIR << 24) | // r_type
(Log2Size << 28) |
(IsPCRel << 30) |
MachO::R_SCATTERED);
MRE.r_word1 = Value2;
Writer->addRelocation(nullptr, Fragment->getParent(), MRE);
} else {
// If the offset is more than 24-bits, it won't fit in a scattered
// relocation offset field, so we fall back to using a non-scattered
// relocation. This is a bit risky, as if the offset reaches out of
// the block and the linker is doing scattered loading on this
// symbol, things can go badly.
//
// Required for 'as' compatibility.
if (FixupOffset > 0xffffff) {
FixedValue = OriginalFixedValue;
return false;
}
}
MachO::any_relocation_info MRE;
MRE.r_word0 = ((FixupOffset << 0) |
(Type << 24) |
(Log2Size << 28) |
(IsPCRel << 30) |
MachO::R_SCATTERED);
MRE.r_word1 = Value;
Writer->addRelocation(nullptr, Fragment->getParent(), MRE);
return true;
}
/// \brief Write the fragment \p F to the output file.
static void writeFragment(const MCAssembler &Asm, const MCAsmLayout &Layout,
const MCFragment &F) {
MCObjectWriter *OW = &Asm.getWriter();
// FIXME: Embed in fragments instead?
uint64_t FragmentSize = Asm.computeFragmentSize(Layout, F);
// Should NOP padding be written out before this fragment?
unsigned BundlePadding = F.getBundlePadding();
if (BundlePadding > 0) {
assert(Asm.isBundlingEnabled() &&
"Writing bundle padding with disabled bundling");
assert(F.hasInstructions() &&
"Writing bundle padding for a fragment without instructions");
unsigned TotalLength = BundlePadding + static_cast<unsigned>(FragmentSize);
if (F.alignToBundleEnd() && TotalLength > Asm.getBundleAlignSize()) {
// If the padding itself crosses a bundle boundary, it must be emitted
// in 2 pieces, since even nop instructions must not cross boundaries.
// v--------------v <- BundleAlignSize
// v---------v <- BundlePadding
// ----------------------------
// | Prev |####|####| F |
// ----------------------------
// ^-------------------^ <- TotalLength
unsigned DistanceToBoundary = TotalLength - Asm.getBundleAlignSize();
if (!Asm.getBackend().writeNopData(DistanceToBoundary, OW))
report_fatal_error("unable to write NOP sequence of " +
Twine(DistanceToBoundary) + " bytes");
BundlePadding -= DistanceToBoundary;
}
if (!Asm.getBackend().writeNopData(BundlePadding, OW))
report_fatal_error("unable to write NOP sequence of " +
Twine(BundlePadding) + " bytes");
}
// This variable (and its dummy usage) is to participate in the assert at
// the end of the function.
uint64_t Start = OW->getStream().tell();
(void) Start;
++stats::EmittedFragments;
switch (F.getKind()) {
case MCFragment::FT_Align: {
++stats::EmittedAlignFragments;
const MCAlignFragment &AF = cast<MCAlignFragment>(F);
assert(AF.getValueSize() && "Invalid virtual align in concrete fragment!");
uint64_t Count = FragmentSize / AF.getValueSize();
// FIXME: This error shouldn't actually occur (the front end should emit
// 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() != FragmentSize)
report_fatal_error("undefined .align directive, value size '" +
Twine(AF.getValueSize()) +
"' is not a divisor of padding size '" +
Twine(FragmentSize) + "'");
// 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 Value and ValueSize to fill the rest.
// If we are aligning with nops, ask that target to emit the right data.
if (AF.hasEmitNops()) {
if (!Asm.getBackend().writeNopData(Count, OW))
report_fatal_error("unable to write nop sequence of " +
Twine(Count) + " bytes");
break;
}
// Otherwise, write out in multiples of the value size.
for (uint64_t i = 0; i != Count; ++i) {
switch (AF.getValueSize()) {
default: llvm_unreachable("Invalid size!");
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:
++stats::EmittedDataFragments;
writeFragmentContents(F, OW);
break;
case MCFragment::FT_Relaxable:
++stats::EmittedRelaxableFragments;
writeFragmentContents(F, OW);
break;
case MCFragment::FT_CompactEncodedInst:
++stats::EmittedCompactEncodedInstFragments;
writeFragmentContents(F, OW);
break;
case MCFragment::FT_Fill: {
++stats::EmittedFillFragments;
const MCFillFragment &FF = cast<MCFillFragment>(F);
assert(FF.getValueSize() && "Invalid virtual align in concrete fragment!");
for (uint64_t i = 0, e = FF.getSize() / FF.getValueSize(); i != e; ++i) {
switch (FF.getValueSize()) {
default: llvm_unreachable("Invalid size!");
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;
}
case MCFragment::FT_LEB: {
const MCLEBFragment &LF = cast<MCLEBFragment>(F);
OW->WriteBytes(LF.getContents().str());
break;
}
case MCFragment::FT_Org: {
++stats::EmittedOrgFragments;
const MCOrgFragment &OF = cast<MCOrgFragment>(F);
for (uint64_t i = 0, e = FragmentSize; i != e; ++i)
OW->Write8(uint8_t(OF.getValue()));
break;
}
case MCFragment::FT_Dwarf: {
const MCDwarfLineAddrFragment &OF = cast<MCDwarfLineAddrFragment>(F);
OW->WriteBytes(OF.getContents().str());
break;
}
case MCFragment::FT_DwarfFrame: {
const MCDwarfCallFrameFragment &CF = cast<MCDwarfCallFrameFragment>(F);
OW->WriteBytes(CF.getContents().str());
break;
}
}
assert(OW->getStream().tell() - Start == FragmentSize &&
"The stream should advance by fragment size");
}
void AMDGPUMCObjectWriter::writeObject(MCAssembler &Asm,
const MCAsmLayout &Layout) {
for (MCAssembler::iterator I = Asm.begin(), E = Asm.end(); I != E; ++I) {
Asm.writeSectionData(&*I, Layout);
}
}
