uint64_t MachObjectWriter::getSymbolAddress(const MCSymbolData* SD,
const MCAsmLayout &Layout) const {
const MCSymbol &S = SD->getSymbol();
// If this is a variable, then recursively evaluate now.
if (S.isVariable()) {
MCValue Target;
if (!S.getVariableValue()->EvaluateAsRelocatable(Target, Layout))
report_fatal_error("unable to evaluate offset for variable '" +
S.getName() + "'");
// Verify that any used symbols are defined.
if (Target.getSymA() && Target.getSymA()->getSymbol().isUndefined())
report_fatal_error("unable to evaluate offset to undefined symbol '" +
Target.getSymA()->getSymbol().getName() + "'");
if (Target.getSymB() && Target.getSymB()->getSymbol().isUndefined())
report_fatal_error("unable to evaluate offset to undefined symbol '" +
Target.getSymB()->getSymbol().getName() + "'");
uint64_t Address = Target.getConstant();
if (Target.getSymA())
Address += getSymbolAddress(&Layout.getAssembler().getSymbolData(
Target.getSymA()->getSymbol()), Layout);
if (Target.getSymB())
Address += getSymbolAddress(&Layout.getAssembler().getSymbolData(
Target.getSymB()->getSymbol()), Layout);
return Address;
}
return getSectionAddress(SD->getFragment()->getParent()) +
Layout.getSymbolOffset(SD);
}
void MCAssembler::AddSectionToTheEnd(MCSectionData &SD, MCAsmLayout &Layout) {
// Create dummy fragments and assign section ordinals.
unsigned SectionIndex = 0;
for (MCAssembler::iterator it = begin(), ie = end(); it != ie; ++it)
SectionIndex++;
SD.setOrdinal(SectionIndex);
// Assign layout order indices to sections and fragments.
unsigned FragmentIndex = 0;
unsigned i = 0;
for (unsigned e = Layout.getSectionOrder().size(); i != e; ++i) {
MCSectionData *SD = Layout.getSectionOrder()[i];
for (MCSectionData::iterator it2 = SD->begin(),
ie2 = SD->end(); it2 != ie2; ++it2)
FragmentIndex++;
}
SD.setLayoutOrder(i);
for (MCSectionData::iterator it2 = SD.begin(),
ie2 = SD.end(); it2 != ie2; ++it2) {
it2->setLayoutOrder(FragmentIndex++);
}
Layout.getSectionOrder().push_back(&SD);
Layout.LayoutSection(&SD);
// Layout until everything fits.
while (LayoutOnce(Layout))
continue;
}
uint64_t MCCodePaddingPolicy::getNextFragmentOffset(const MCFragment *Fragment,
const MCAsmLayout &Layout) {
assert(Fragment != nullptr && "Fragment cannot be null");
MCFragment const *NextFragment = Fragment->getNextNode();
return NextFragment == nullptr
? Layout.getSectionAddressSize(Fragment->getParent())
: Layout.getFragmentOffset(NextFragment);
}
void MCAssembler::writeSectionData(const MCSectionData *SD,
const MCAsmLayout &Layout) const {
// Ignore virtual sections.
if (SD->getSection().isVirtualSection()) {
assert(Layout.getSectionFileSize(SD) == 0 && "Invalid size for section!");
// Check that contents are only things legal inside a virtual section.
for (MCSectionData::const_iterator it = SD->begin(),
ie = SD->end(); it != ie; ++it) {
switch (it->getKind()) {
default: llvm_unreachable("Invalid fragment in virtual section!");
case MCFragment::FT_Data: {
// Check that we aren't trying to write a non-zero contents (or fixups)
// into a virtual section. This is to support clients which use standard
// directives to fill the contents of virtual sections.
const MCDataFragment &DF = cast<MCDataFragment>(*it);
assert(DF.fixup_begin() == DF.fixup_end() &&
"Cannot have fixups in virtual section!");
for (unsigned i = 0, e = DF.getContents().size(); i != e; ++i)
if (DF.getContents()[i]) {
if (auto *ELFSec = dyn_cast<const MCSectionELF>(&SD->getSection()))
report_fatal_error("non-zero initializer found in section '" +
ELFSec->getSectionName() + "'");
else
report_fatal_error("non-zero initializer found in virtual section");
}
break;
}
case MCFragment::FT_Align:
// Check that we aren't trying to write a non-zero value into a virtual
// section.
assert((cast<MCAlignFragment>(it)->getValueSize() == 0 ||
cast<MCAlignFragment>(it)->getValue() == 0) &&
"Invalid align in virtual section!");
break;
case MCFragment::FT_Fill:
assert((cast<MCFillFragment>(it)->getValueSize() == 0 ||
cast<MCFillFragment>(it)->getValue() == 0) &&
"Invalid fill in virtual section!");
break;
}
}
return;
}
uint64_t Start = getWriter().getStream().tell();
(void)Start;
for (MCSectionData::const_iterator it = SD->begin(), ie = SD->end();
it != ie; ++it)
writeFragment(*this, Layout, *it);
assert(getWriter().getStream().tell() - Start ==
Layout.getSectionAddressSize(SD));
}
bool MCCodePadder::relaxFragment(MCPaddingFragment *Fragment,
MCAsmLayout &Layout) {
if (!Fragment->isInsertionPoint())
return false;
uint64_t OldSize = Fragment->getSize();
uint64_t MaxWindowSize = getMaxWindowSize(Fragment, Layout);
if (MaxWindowSize == UINT64_C(0))
return false;
assert(isPowerOf2_64(MaxWindowSize) &&
"MaxWindowSize must be an integer power of 2");
uint64_t SectionAlignment = Fragment->getParent()->getAlignment();
assert(isPowerOf2_64(SectionAlignment) &&
"SectionAlignment must be an integer power of 2");
MCPFRange &Jurisdiction = getJurisdiction(Fragment, Layout);
uint64_t OptimalSize = UINT64_C(0);
double OptimalWeight = std::numeric_limits<double>::max();
uint64_t MaxFragmentSize = MaxWindowSize - UINT16_C(1);
for (uint64_t Size = UINT64_C(0); Size <= MaxFragmentSize; ++Size) {
Fragment->setSize(Size);
Layout.invalidateFragmentsFrom(Fragment);
double SizeWeight = 0.0;
// The section is guaranteed to be aligned to SectionAlignment, but that
// doesn't guarantee the exact section offset w.r.t. the policies window
// size.
// As a concrete example, the section could be aligned to 16B, but a
// policy's window size can be 32B. That means that the section actual start
// address can either be 0mod32 or 16mod32. The said policy will act
// differently for each case, so we need to take both into consideration.
for (uint64_t Offset = UINT64_C(0); Offset < MaxWindowSize;
Offset += SectionAlignment) {
double OffsetWeight = std::accumulate(
CodePaddingPolicies.begin(), CodePaddingPolicies.end(), 0.0,
[&Jurisdiction, &Offset, &Layout](
double Weight, const MCCodePaddingPolicy *Policy) -> double {
double PolicyWeight =
Policy->computeRangePenaltyWeight(Jurisdiction, Offset, Layout);
assert(PolicyWeight >= 0.0 && "A penalty weight must be positive");
return Weight + PolicyWeight;
});
SizeWeight = std::max(SizeWeight, OffsetWeight);
}
if (SizeWeight < OptimalWeight) {
OptimalWeight = SizeWeight;
OptimalSize = Size;
}
if (OptimalWeight == 0.0)
break;
}
Fragment->setSize(OptimalSize);
Layout.invalidateFragmentsFrom(Fragment);
return OldSize != OptimalSize;
}
uint64_t MachObjectWriter::getPaddingSize(const MCSection *Sec,
const MCAsmLayout &Layout) const {
uint64_t EndAddr = getSectionAddress(Sec) + Layout.getSectionAddressSize(Sec);
unsigned Next = Sec->getLayoutOrder() + 1;
if (Next >= Layout.getSectionOrder().size())
return 0;
const MCSection &NextSec = *Layout.getSectionOrder()[Next];
if (NextSec.isVirtualSection())
return 0;
return OffsetToAlignment(EndAddr, NextSec.getAlignment());
}
void MCAssembler::WriteSectionData(const MCSectionData *SD,
const MCAsmLayout &Layout,
MCObjectWriter *OW) const {
// Ignore virtual sections.
if (getBackend().isVirtualSection(SD->getSection())) {
assert(Layout.getSectionFileSize(SD) == 0 && "Invalid size for section!");
// Check that contents are only things legal inside a virtual section.
for (MCSectionData::const_iterator it = SD->begin(),
ie = SD->end(); it != ie; ++it) {
switch (it->getKind()) {
default:
assert(0 && "Invalid fragment in virtual section!");
case MCFragment::FT_Data: {
// Check that we aren't trying to write a non-zero contents (or fixups)
// into a virtual section. This is to support clients which use standard
// directives to fill the contents of virtual sections.
MCDataFragment &DF = cast<MCDataFragment>(*it);
assert(DF.fixup_begin() == DF.fixup_end() &&
"Cannot have fixups in virtual section!");
for (unsigned i = 0, e = DF.getContents().size(); i != e; ++i)
assert(DF.getContents()[i] == 0 &&
"Invalid data value for virtual section!");
break;
}
case MCFragment::FT_Align:
// Check that we aren't trying to write a non-zero value into a virtual
// section.
assert((!cast<MCAlignFragment>(it)->getValueSize() ||
!cast<MCAlignFragment>(it)->getValue()) &&
"Invalid align in virtual section!");
break;
case MCFragment::FT_Fill:
assert(!cast<MCFillFragment>(it)->getValueSize() &&
"Invalid fill in virtual section!");
break;
}
}
return;
}
uint64_t Start = OW->getStream().tell();
(void) Start;
for (MCSectionData::const_iterator it = SD->begin(),
ie = SD->end(); it != ie; ++it)
WriteFragmentData(*this, Layout, *it, OW);
assert(OW->getStream().tell() - Start == Layout.getSectionFileSize(SD));
}
uint64_t MCAssembler::computeFragmentSize(const MCAsmLayout &Layout,
const MCFragment &F) const {
switch (F.getKind()) {
case MCFragment::FT_Data:
case MCFragment::FT_Relaxable:
case MCFragment::FT_CompactEncodedInst:
return cast<MCEncodedFragment>(F).getContents().size();
case MCFragment::FT_Fill:
return cast<MCFillFragment>(F).getSize();
case MCFragment::FT_LEB:
return cast<MCLEBFragment>(F).getContents().size();
case MCFragment::FT_Align: {
const MCAlignFragment &AF = cast<MCAlignFragment>(F);
unsigned Offset = Layout.getFragmentOffset(&AF);
unsigned Size = OffsetToAlignment(Offset, AF.getAlignment());
// If we are padding with nops, force the padding to be larger than the
// minimum nop size.
if (Size > 0 && AF.hasEmitNops()) {
while (Size % getBackend().getMinimumNopSize())
Size += AF.getAlignment();
}
if (Size > AF.getMaxBytesToEmit())
return 0;
return Size;
}
case MCFragment::FT_Org: {
const MCOrgFragment &OF = cast<MCOrgFragment>(F);
int64_t TargetLocation;
if (!OF.getOffset().EvaluateAsAbsolute(TargetLocation, Layout))
report_fatal_error("expected assembly-time absolute expression");
// FIXME: We need a way to communicate this error.
uint64_t FragmentOffset = Layout.getFragmentOffset(&OF);
int64_t Size = TargetLocation - FragmentOffset;
if (Size < 0 || Size >= 0x40000000)
report_fatal_error("invalid .org offset '" + Twine(TargetLocation) +
"' (at offset '" + Twine(FragmentOffset) + "')");
return Size;
}
case MCFragment::FT_Dwarf:
return cast<MCDwarfLineAddrFragment>(F).getContents().size();
case MCFragment::FT_DwarfFrame:
return cast<MCDwarfCallFrameFragment>(F).getContents().size();
}
llvm_unreachable("invalid fragment kind");
}
void MachObjectWriter::computeSectionAddresses(const MCAssembler &Asm,
const MCAsmLayout &Layout) {
uint64_t StartAddress = 0;
for (const MCSection *Sec : Layout.getSectionOrder()) {
StartAddress = alignTo(StartAddress, Sec->getAlignment());
SectionAddress[Sec] = StartAddress;
StartAddress += Layout.getSectionAddressSize(Sec);
// Explicitly pad the section to match the alignment requirements of the
// following one. This is for 'gas' compatibility, it shouldn't
/// strictly be necessary.
StartAddress += getPaddingSize(Sec, Layout);
}
}
uint64_t ELFObjectWriter::SymbolValue(const MCSymbol &Sym,
const MCAsmLayout &Layout) {
if (Sym.isCommon() && Sym.isExternal())
return Sym.getCommonAlignment();
uint64_t Res;
if (!Layout.getSymbolOffset(Sym, Res))
return 0;
if (Layout.getAssembler().isThumbFunc(&Sym))
Res |= 1;
return Res;
}
void MachObjectWriter::WriteSection(const MCAssembler &Asm,
const MCAsmLayout &Layout,
const MCSectionData &SD,
uint64_t FileOffset,
uint64_t RelocationsStart,
unsigned NumRelocations) {
uint64_t SectionSize = Layout.getSectionAddressSize(&SD);
// The offset is unused for virtual sections.
if (SD.getSection().isVirtualSection()) {
assert(Layout.getSectionFileSize(&SD) == 0 && "Invalid file size!");
FileOffset = 0;
}
// struct section (68 bytes) or
// struct section_64 (80 bytes)
uint64_t Start = OS.tell();
(void) Start;
const MCSectionMachO &Section = cast<MCSectionMachO>(SD.getSection());
WriteBytes(Section.getSectionName(), 16);
WriteBytes(Section.getSegmentName(), 16);
if (is64Bit()) {
Write64(getSectionAddress(&SD)); // address
Write64(SectionSize); // size
} else {
Write32(getSectionAddress(&SD)); // address
Write32(SectionSize); // size
}
Write32(FileOffset);
unsigned Flags = Section.getTypeAndAttributes();
if (SD.hasInstructions())
Flags |= MCSectionMachO::S_ATTR_SOME_INSTRUCTIONS;
assert(isPowerOf2_32(SD.getAlignment()) && "Invalid alignment!");
Write32(Log2_32(SD.getAlignment()));
Write32(NumRelocations ? RelocationsStart : 0);
Write32(NumRelocations);
Write32(Flags);
Write32(IndirectSymBase.lookup(&SD)); // reserved1
Write32(Section.getStubSize()); // reserved2
if (is64Bit())
Write32(0); // reserved3
assert(OS.tell() - Start == (is64Bit() ? macho::Section64Size :
macho::Section32Size));
}
uint64_t MachObjectWriter::getSymbolAddress(const MCSymbol &S,
const MCAsmLayout &Layout) const {
// If this is a variable, then recursively evaluate now.
if (S.isVariable()) {
if (const MCConstantExpr *C =
dyn_cast<const MCConstantExpr>(S.getVariableValue()))
return C->getValue();
MCValue Target;
if (!S.getVariableValue()->evaluateAsRelocatable(Target, &Layout, nullptr))
report_fatal_error("unable to evaluate offset for variable '" +
S.getName() + "'");
// Verify that any used symbols are defined.
if (Target.getSymA() && Target.getSymA()->getSymbol().isUndefined())
report_fatal_error("unable to evaluate offset to undefined symbol '" +
Target.getSymA()->getSymbol().getName() + "'");
if (Target.getSymB() && Target.getSymB()->getSymbol().isUndefined())
report_fatal_error("unable to evaluate offset to undefined symbol '" +
Target.getSymB()->getSymbol().getName() + "'");
uint64_t Address = Target.getConstant();
if (Target.getSymA())
Address += getSymbolAddress(Target.getSymA()->getSymbol(), Layout);
if (Target.getSymB())
Address += getSymbolAddress(Target.getSymB()->getSymbol(), Layout);
return Address;
}
return getSectionAddress(S.getFragment()->getParent()) +
Layout.getSymbolOffset(S);
}
void ELFObjectWriter::writeSectionHeader(
const MCAsmLayout &Layout, const SectionIndexMapTy &SectionIndexMap,
const SectionOffsetsTy &SectionOffsets) {
const unsigned NumSections = SectionTable.size();
// Null section first.
uint64_t FirstSectionSize =
(NumSections + 1) >= ELF::SHN_LORESERVE ? NumSections + 1 : 0;
WriteSecHdrEntry(0, 0, 0, 0, 0, FirstSectionSize, 0, 0, 0, 0);
for (const MCSectionELF *Section : SectionTable) {
uint32_t GroupSymbolIndex;
unsigned Type = Section->getType();
if (Type != ELF::SHT_GROUP)
GroupSymbolIndex = 0;
else
GroupSymbolIndex = Section->getGroup()->getIndex();
const std::pair<uint64_t, uint64_t> &Offsets =
SectionOffsets.find(Section)->second;
uint64_t Size;
if (Type == ELF::SHT_NOBITS)
Size = Layout.getSectionAddressSize(Section);
else
Size = Offsets.second - Offsets.first;
writeSection(SectionIndexMap, GroupSymbolIndex, Offsets.first, Size,
*Section);
}
}
bool ELFObjectWriter::isInSymtab(const MCAsmLayout &Layout,
const MCSymbolELF &Symbol, bool Used,
bool Renamed) {
if (Symbol.isVariable()) {
const MCExpr *Expr = Symbol.getVariableValue();
if (const MCSymbolRefExpr *Ref = dyn_cast<MCSymbolRefExpr>(Expr)) {
if (Ref->getKind() == MCSymbolRefExpr::VK_WEAKREF)
return false;
}
}
if (Used)
return true;
if (Renamed)
return false;
if (Symbol.isVariable() && Symbol.isUndefined()) {
// FIXME: this is here just to diagnose the case of a var = commmon_sym.
Layout.getBaseSymbol(Symbol);
return false;
}
if (Symbol.isUndefined() && !Symbol.isBindingSet())
return false;
if (Symbol.isTemporary())
return false;
if (Symbol.getType() == ELF::STT_SECTION)
return false;
return true;
}
void MachObjectWriter::computeSectionAddresses(const MCAssembler &Asm,
const MCAsmLayout &Layout) {
uint64_t StartAddress = 0;
const SmallVectorImpl<MCSectionData*> &Order = Layout.getSectionOrder();
for (int i = 0, n = Order.size(); i != n ; ++i) {
const MCSectionData *SD = Order[i];
StartAddress = RoundUpToAlignment(StartAddress, SD->getAlignment());
SectionAddress[SD] = StartAddress;
StartAddress += Layout.getSectionAddressSize(SD);
// Explicitly pad the section to match the alignment requirements of the
// following one. This is for 'gas' compatibility, it shouldn't
/// strictly be necessary.
StartAddress += getPaddingSize(SD, Layout);
}
}
void SVMELFProgramWriter::writeSectionHeader(const MCAsmLayout &Layout,
const MCSectionData *SD)
{
const MCSectionELF *SE = dyn_cast<MCSectionELF>(&SD->getSection());
assert(SE);
uint32_t sh_type = SE->getType();
uint32_t sh_link = 0;
uint32_t sh_info = 0;
// Type-specific data finds itself in sh_link and sh_info
switch (sh_type) {
case ELF::SHT_SYMTAB:
case ELF::SHT_DYNSYM:
sh_link = EMB.getStringTableIndex();
sh_info = EMB.getLastLocalSymbolIndex();
break;
default:
break;
}
Write32(EMB.getSectionStringTableIndex(SE)); // sh_name
Write32(sh_type); // sh_type
Write32(SE->getFlags()); // sh_flags
Write32(ML.getSectionMemAddress(SD)); // sh_addr
Write32(ML.getSectionDiskOffset(SD)); // sh_offset
Write32(Layout.getSectionFileSize(SD)); // sh_size
Write32(sh_link); // sh_link
Write32(sh_info); // sh_info
Write32(SD->getAlignment()); // sh_addralign
Write32(SE->getEntrySize()); // sh_entsize
}
uint32_t SVMMemoryLayout::getSectionMemSize(const MCSectionData *SD, const MCAsmLayout &Layout) const
{
SectionMemSizeOverrides_t::const_iterator it = SectionMemSizeOverrides.find(SD);
if (it == SectionMemSizeOverrides.end())
return Layout.getSectionAddressSize(SD);
else
return it->second;
}
void MachObjectWriter::writeSection(const MCAsmLayout &Layout,
const MCSection &Sec, uint64_t VMAddr,
uint64_t FileOffset, unsigned Flags,
uint64_t RelocationsStart,
unsigned NumRelocations) {
uint64_t SectionSize = Layout.getSectionAddressSize(&Sec);
const MCSectionMachO &Section = cast<MCSectionMachO>(Sec);
// The offset is unused for virtual sections.
if (Section.isVirtualSection()) {
assert(Layout.getSectionFileSize(&Sec) == 0 && "Invalid file size!");
FileOffset = 0;
}
// struct section (68 bytes) or
// struct section_64 (80 bytes)
uint64_t Start = getStream().tell();
(void) Start;
writeBytes(Section.getSectionName(), 16);
writeBytes(Section.getSegmentName(), 16);
if (is64Bit()) {
write64(VMAddr); // address
write64(SectionSize); // size
} else {
write32(VMAddr); // address
write32(SectionSize); // size
}
write32(FileOffset);
assert(isPowerOf2_32(Section.getAlignment()) && "Invalid alignment!");
write32(Log2_32(Section.getAlignment()));
write32(NumRelocations ? RelocationsStart : 0);
write32(NumRelocations);
write32(Flags);
write32(IndirectSymBase.lookup(&Sec)); // reserved1
write32(Section.getStubSize()); // reserved2
if (is64Bit())
write32(0); // reserved3
assert(getStream().tell() - Start ==
(is64Bit() ? sizeof(MachO::section_64) : sizeof(MachO::section)));
}
// Each LOH is composed by, in this order (each field is encoded using ULEB128):
// - Its kind.
// - Its number of arguments (let say N).
// - Its arg1.
// - ...
// - Its argN.
// <arg1> to <argN> are absolute addresses in the object file, i.e.,
// relative addresses from the beginning of the object file.
void MCLOHDirective::Emit_impl(raw_ostream &OutStream,
const MachObjectWriter &ObjWriter,
const MCAsmLayout &Layout) const {
const MCAssembler &Asm = Layout.getAssembler();
encodeULEB128(Kind, OutStream);
encodeULEB128(Args.size(), OutStream);
for (LOHArgs::const_iterator It = Args.begin(), EndIt = Args.end();
It != EndIt; ++It)
encodeULEB128(ObjWriter.getSymbolAddress(&Asm.getSymbolData(**It), Layout),
OutStream);
}
static uint64_t getSymbolValue(const MCSymbolData &Data,
const MCAsmLayout &Layout) {
if (Data.isCommon() && Data.isExternal())
return Data.getCommonSize();
uint64_t Res;
if (!Layout.getSymbolOffset(&Data, Res))
return 0;
return Res;
}
// Simple getSymbolOffset helper for the non-varibale case.
static bool getLabelOffset(const MCAsmLayout &Layout, const MCSymbol &S,
bool ReportError, uint64_t &Val) {
if (!S.getFragment()) {
if (ReportError)
report_fatal_error("unable to evaluate offset to undefined symbol '" +
S.getName() + "'");
return false;
}
Val = Layout.getFragmentOffset(S.getFragment()) + S.getOffset();
return true;
}
void X86MachObjectWriter::RecordTLVPRelocation(MachObjectWriter *Writer,
const MCAssembler &Asm,
const MCAsmLayout &Layout,
const MCFragment *Fragment,
const MCFixup &Fixup,
MCValue Target,
uint64_t &FixedValue) {
assert(Target.getSymA()->getKind() == MCSymbolRefExpr::VK_TLVP &&
!is64Bit() &&
"Should only be called with a 32-bit TLVP relocation!");
unsigned Log2Size = getFixupKindLog2Size(Fixup.getKind());
uint32_t Value = Layout.getFragmentOffset(Fragment)+Fixup.getOffset();
unsigned IsPCRel = 0;
// Get the symbol data.
const MCSymbolData *SD_A = &Asm.getSymbolData(Target.getSymA()->getSymbol());
unsigned Index = SD_A->getIndex();
// We're only going to have a second symbol in pic mode and it'll be a
// subtraction from the picbase. For 32-bit pic the addend is the difference
// between the picbase and the next address. For 32-bit static the addend is
// zero.
if (Target.getSymB()) {
// If this is a subtraction then we're pcrel.
uint32_t FixupAddress =
Writer->getFragmentAddress(Fragment, Layout) + Fixup.getOffset();
const MCSymbolData *SD_B =
&Asm.getSymbolData(Target.getSymB()->getSymbol());
IsPCRel = 1;
FixedValue = (FixupAddress - Writer->getSymbolAddress(SD_B, Layout) +
Target.getConstant());
FixedValue += 1ULL << Log2Size;
} else {
FixedValue = 0;
}
// struct relocation_info (8 bytes)
MachO::any_relocation_info MRE;
MRE.r_word0 = Value;
MRE.r_word1 = ((Index << 0) |
(IsPCRel << 24) |
(Log2Size << 25) |
(1 << 27) | // r_extern
(MachO::GENERIC_RELOC_TLV << 28)); // r_type
Writer->addRelocation(Fragment->getParent(), MRE);
}
void SVMMemoryLayout::AllocateSections(MCAssembler &Asm, const MCAsmLayout &Layout)
{
memset(spsMemSize, 0, sizeof spsMemSize);
memset(spsDiskSize, 0, sizeof spsDiskSize);
bssAlign = 1;
// Leave one free block at the beginning of DEBUG, for a special message.
spsMemSize[SPS_DEBUG] = SVMTargetMachine::getBlockSize();
spsDiskSize[SPS_DEBUG] = spsMemSize[SPS_DEBUG];
for (MCAssembler::const_iterator it = Asm.begin(), ie = Asm.end();
it != ie; ++it) {
const MCSectionData *SD = &*it;
SVMProgramSection sps = getSectionKind(SD);
switch (sps) {
case SPS_BSS:
// Also track BSS segment alignment, then fall through...
bssAlign = std::max(bssAlign, SD->getAlignment());
case SPS_RO:
case SPS_RW_Z:
case SPS_RW_PLAIN:
case SPS_DEBUG:
case SPS_META:
spsMemSize[sps] = RoundUpToAlignment(spsMemSize[sps], SD->getAlignment());
spsDiskSize[sps] = RoundUpToAlignment(spsDiskSize[sps], SD->getAlignment());
SectionOffsetMap[SD] = spsDiskSize[sps];
spsMemSize[sps] += getSectionMemSize(SD, Layout);
spsDiskSize[sps] += Layout.getSectionAddressSize(SD);
break;
default:
break;
}
}
unsigned memUsed = getSectionMemAddress(SPS_END) - SVMTargetMachine::getRAMBase();
unsigned memMax = SVMTargetMachine::getRAMSize();
if (memUsed > memMax)
report_fatal_error("Application is too large to fit in RAM! Need "
+ Twine(memUsed) + " bytes, which exceeds the maximum of " + Twine(memMax));
}
void ELFObjectWriter::writeSymbol(SymbolTableWriter &Writer,
uint32_t StringIndex, ELFSymbolData &MSD,
const MCAsmLayout &Layout) {
const auto &Symbol = cast<MCSymbolELF>(*MSD.Symbol);
const MCSymbolELF *Base =
cast_or_null<MCSymbolELF>(Layout.getBaseSymbol(Symbol));
// This has to be in sync with when computeSymbolTable uses SHN_ABS or
// SHN_COMMON.
bool IsReserved = !Base || Symbol.isCommon();
// Binding and Type share the same byte as upper and lower nibbles
uint8_t Binding = Symbol.getBinding();
uint8_t Type = Symbol.getType();
if (Base) {
Type = mergeTypeForSet(Type, Base->getType());
}
uint8_t Info = (Binding << 4) | Type;
// Other and Visibility share the same byte with Visibility using the lower
// 2 bits
uint8_t Visibility = Symbol.getVisibility();
uint8_t Other = Symbol.getOther() | Visibility;
uint64_t Value = SymbolValue(*MSD.Symbol, Layout);
uint64_t Size = 0;
const MCExpr *ESize = MSD.Symbol->getSize();
if (!ESize && Base)
ESize = Base->getSize();
if (ESize) {
int64_t Res;
if (!ESize->evaluateKnownAbsolute(Res, Layout))
report_fatal_error("Size expression must be absolute.");
Size = Res;
}
// Write out the symbol table entry
Writer.writeSymbol(StringIndex, Info, Value, Size, Other, MSD.SectionIndex,
IsReserved);
}
static bool getSymbolOffsetImpl(const MCAsmLayout &Layout,
const MCSymbolData *SD, bool ReportError,
uint64_t &Val) {
const MCSymbol &S = SD->getSymbol();
if (!S.isVariable())
return getLabelOffset(Layout, *SD, ReportError, Val);
// If SD is a variable, evaluate it.
MCValue Target;
if (!S.getVariableValue()->EvaluateAsValue(Target, &Layout, nullptr))
report_fatal_error("unable to evaluate offset for variable '" +
S.getName() + "'");
uint64_t Offset = Target.getConstant();
const MCAssembler &Asm = Layout.getAssembler();
const MCSymbolRefExpr *A = Target.getSymA();
if (A) {
uint64_t ValA;
if (!getLabelOffset(Layout, Asm.getSymbolData(A->getSymbol()), ReportError,
ValA))
return false;
Offset += ValA;
}
const MCSymbolRefExpr *B = Target.getSymB();
if (B) {
uint64_t ValB;
if (!getLabelOffset(Layout, Asm.getSymbolData(B->getSymbol()), ReportError,
ValB))
return false;
Offset -= ValB;
}
Val = Offset;
return true;
}
void SVMMemoryLayout::RecordRelocation(const MCAssembler &Asm,
const MCAsmLayout &Layout, const MCFragment *Fragment,
const MCFixup &Fixup, MCValue Target, uint64_t &FixedValue)
{
SVM::Fixups kind = (SVM::Fixups) Fixup.getKind();
switch (kind) {
case SVM::fixup_fnstack: {
/*
* Function stack adjustment annotation. The adjustment amount is
* stored in the first argument of a binary op.
* (See SVMMCCodeEmitter::EncodeInstruction)
*
* We store this offset for later, in our FNStackMap, indexed
* by the address of the FNStack pseudo-op itself.
*/
FNStackMap[std::make_pair(Fragment->getParent(),
Layout.getFragmentOffset(Fragment))] = (int)Target.getConstant();
break;
}
default: {
/*
* All other fixups are applied later, in ApplyLateFixups().
* This is necessary because other fixup types can depend on
* the FNStack values collected above.
*/
SVMLateFixup LF((MCFragment *) Fragment, Fixup, Target);
LateFixupList.push_back(LF);
// This gets OR'ed with the fixup later. Zero it.
FixedValue = 0;
break;
}
}
}
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 MCSymbolData &A_SD = Asm.getSymbolData(*A);
const MCSymbol *A_Base = Asm.getAtom(&A_SD);
const MCSymbol *B = &Target.getSymB()->getSymbol();
if (B->isTemporary())
B = &Writer->findAliasedSymbol(*B);
const MCSymbolData &B_SD = Asm.getSymbolData(*B);
const MCSymbol *B_Base = Asm.getAtom(&B_SD);
// Neither symbol can be modified.
if (Target.getSymA()->getKind() != MCSymbolRefExpr::VK_None ||
Target.getSymB()->getKind() != MCSymbolRefExpr::VK_None)
report_fatal_error("unsupported relocation of modified symbol", false);
// We don't support PCrel relocations of differences. Darwin 'as' doesn't
// implement most of these correctly.
if (IsPCRel)
report_fatal_error("unsupported pc-relative relocation of difference",
false);
// 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)
report_fatal_error("unsupported relocation with identical base", false);
// 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().FatalError(Fixup.getLoc(),
"unsupported relocation with subtraction expression, symbol '" +
Name + "' can not be undefined in a subtraction expression");
}
Value +=
Writer->getSymbolAddress(&A_SD, Layout) -
(!A_Base ? 0 : Writer->getSymbolAddress(&A_Base->getData(), Layout));
Value -=
Writer->getSymbolAddress(&B_SD, Layout) -
(!B_Base ? 0 : Writer->getSymbolAddress(&B_Base->getData(), Layout));
if (!A_Base)
Index = A_SD.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_SD.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))
Asm.addLocalUsedInReloc(*Symbol);
}
const MCSymbolData &SD = Asm.getSymbolData(*Symbol);
RelSymbol = Asm.getAtom(&SD);
// 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()->getSection());
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->getData() != &SD)
Value += Layout.getSymbolOffset(&SD) -
Layout.getSymbolOffset(&RelSymbol->getData());
} else if (Symbol->isInSection() && !Symbol->isVariable()) {
// The index is the section ordinal (1-based).
Index = SD.getFragment()->getParent()->getOrdinal() + 1;
Value += Writer->getSymbolAddress(&SD, 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 {
report_fatal_error("unsupported relocation of variable '" +
Symbol->getName() + "'", false);
}
} else {
report_fatal_error("unsupported relocation of undefined symbol '" +
Symbol->getName() + "'", false);
}
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) {
report_fatal_error("unsupported symbol modifier in relocation",
false);
} 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)
report_fatal_error("unsupported symbol modifier in branch "
"relocation", false);
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) {
report_fatal_error("TLVP symbol modifier should have been rip-rel",
false);
} else if (Modifier != MCSymbolRefExpr::VK_None)
report_fatal_error("unsupported symbol modifier in relocation", false);
else {
Type = MachO::X86_64_RELOC_UNSIGNED;
unsigned Kind = Fixup.getKind();
if (Kind == X86::reloc_signed_4byte)
report_fatal_error("32-bit absolute addressing is not supported in "
"64-bit mode", false);
}
}
}
// 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);
}
void X86MachObjectWriter::RecordX86Relocation(MachObjectWriter *Writer,
const MCAssembler &Asm,
const MCAsmLayout &Layout,
const MCFragment *Fragment,
const MCFixup &Fixup,
MCValue Target,
uint64_t &FixedValue) {
unsigned IsPCRel = Writer->isFixupKindPCRel(Asm, Fixup.getKind());
unsigned Log2Size = getFixupKindLog2Size(Fixup.getKind());
// If this is a 32-bit TLVP reloc it's handled a bit differently.
if (Target.getSymA() &&
Target.getSymA()->getKind() == MCSymbolRefExpr::VK_TLVP) {
RecordTLVPRelocation(Writer, Asm, Layout, Fragment, Fixup, Target,
FixedValue);
return;
}
// If this is a difference or a defined symbol plus an offset, then we need a
// scattered relocation entry. Differences always require scattered
// relocations.
if (Target.getSymB()) {
RecordScatteredRelocation(Writer, Asm, Layout, Fragment, Fixup,
Target, Log2Size, FixedValue);
return;
}
// Get the symbol data, if any.
const MCSymbolData *SD = nullptr;
if (Target.getSymA())
SD = &Asm.getSymbolData(Target.getSymA()->getSymbol());
// If this is an internal relocation with an offset, it also needs a scattered
// relocation entry.
uint32_t Offset = Target.getConstant();
if (IsPCRel)
Offset += 1 << Log2Size;
// Try to record the scattered relocation if needed. Fall back to non
// scattered if necessary (see comments in RecordScatteredRelocation()
// for details).
if (Offset && SD && !Writer->doesSymbolRequireExternRelocation(SD) &&
RecordScatteredRelocation(Writer, Asm, Layout, Fragment, Fixup,
Target, Log2Size, FixedValue))
return;
// See <reloc.h>.
uint32_t FixupOffset = Layout.getFragmentOffset(Fragment)+Fixup.getOffset();
unsigned Index = 0;
unsigned Type = 0;
const MCSymbol *RelSymbol = nullptr;
if (Target.isAbsolute()) { // constant
// SymbolNum of 0 indicates the absolute section.
//
// FIXME: Currently, these are never generated (see code below). I cannot
// find a case where they are actually emitted.
Type = MachO::GENERIC_RELOC_VANILLA;
} else {
// Resolve constant variables.
if (SD->getSymbol().isVariable()) {
int64_t Res;
if (SD->getSymbol().getVariableValue()->EvaluateAsAbsolute(
Res, Layout, Writer->getSectionAddressMap())) {
FixedValue = Res;
return;
}
}
// Check whether we need an external or internal relocation.
if (Writer->doesSymbolRequireExternRelocation(SD)) {
RelSymbol = &SD->getSymbol();
// For external relocations, make sure to offset the fixup value to
// compensate for the addend of the symbol address, if it was
// undefined. This occurs with weak definitions, for example.
if (!SD->getSymbol().isUndefined())
FixedValue -= Layout.getSymbolOffset(SD);
} else {
// The index is the section ordinal (1-based).
const MCSectionData &SymSD = Asm.getSectionData(
SD->getSymbol().getSection());
Index = SymSD.getOrdinal() + 1;
FixedValue += Writer->getSectionAddress(&SymSD);
}
if (IsPCRel)
FixedValue -= Writer->getSectionAddress(Fragment->getParent());
Type = MachO::GENERIC_RELOC_VANILLA;
}
// 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 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();
const MCSymbolData *A_SD = &Asm.getSymbolData(*A);
if (!A_SD->getFragment())
report_fatal_error("symbol '" + A->getName() +
"' can not be undefined in a subtraction expression",
false);
uint32_t Value = Writer->getSymbolAddress(A_SD, Layout);
uint64_t SecAddr = Writer->getSectionAddress(A_SD->getFragment()->getParent());
FixedValue += SecAddr;
uint32_t Value2 = 0;
if (const MCSymbolRefExpr *B = Target.getSymB()) {
const MCSymbolData *B_SD = &Asm.getSymbolData(B->getSymbol());
if (!B_SD->getFragment())
report_fatal_error("symbol '" + B->getSymbol().getName() +
"' can not be undefined in a subtraction expression",
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_SD->isExternal() ? (unsigned)MachO::GENERIC_RELOC_SECTDIFF :
(unsigned)MachO::GENERIC_RELOC_LOCAL_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.
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().FatalError(Fixup.getLoc(),
Twine("Section too large, can't encode "
"r_address (") + Buffer +
") into 24 bits of scattered "
"relocation entry.");
llvm_unreachable("fatal error returned?!");
}
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;
}
void MachObjectWriter::writeObject(MCAssembler &Asm,
const MCAsmLayout &Layout) {
// Compute symbol table information and bind symbol indices.
computeSymbolTable(Asm, LocalSymbolData, ExternalSymbolData,
UndefinedSymbolData);
unsigned NumSections = Asm.size();
const MCAssembler::VersionMinInfoType &VersionInfo =
Layout.getAssembler().getVersionMinInfo();
// The section data starts after the header, the segment load command (and
// section headers) and the symbol table.
unsigned NumLoadCommands = 1;
uint64_t LoadCommandsSize = is64Bit() ?
sizeof(MachO::segment_command_64) + NumSections * sizeof(MachO::section_64):
sizeof(MachO::segment_command) + NumSections * sizeof(MachO::section);
// Add the deployment target version info load command size, if used.
if (VersionInfo.Major != 0) {
++NumLoadCommands;
LoadCommandsSize += sizeof(MachO::version_min_command);
}
// Add the data-in-code load command size, if used.
unsigned NumDataRegions = Asm.getDataRegions().size();
if (NumDataRegions) {
++NumLoadCommands;
LoadCommandsSize += sizeof(MachO::linkedit_data_command);
}
// Add the loh load command size, if used.
uint64_t LOHRawSize = Asm.getLOHContainer().getEmitSize(*this, Layout);
uint64_t LOHSize = alignTo(LOHRawSize, is64Bit() ? 8 : 4);
if (LOHSize) {
++NumLoadCommands;
LoadCommandsSize += sizeof(MachO::linkedit_data_command);
}
// Add the symbol table load command sizes, if used.
unsigned NumSymbols = LocalSymbolData.size() + ExternalSymbolData.size() +
UndefinedSymbolData.size();
if (NumSymbols) {
NumLoadCommands += 2;
LoadCommandsSize += (sizeof(MachO::symtab_command) +
sizeof(MachO::dysymtab_command));
}
// Add the linker option load commands sizes.
for (const auto &Option : Asm.getLinkerOptions()) {
++NumLoadCommands;
LoadCommandsSize += ComputeLinkerOptionsLoadCommandSize(Option, is64Bit());
}
// Compute the total size of the section data, as well as its file size and vm
// size.
uint64_t SectionDataStart = (is64Bit() ? sizeof(MachO::mach_header_64) :
sizeof(MachO::mach_header)) + LoadCommandsSize;
uint64_t SectionDataSize = 0;
uint64_t SectionDataFileSize = 0;
uint64_t VMSize = 0;
for (const MCSection &Sec : Asm) {
uint64_t Address = getSectionAddress(&Sec);
uint64_t Size = Layout.getSectionAddressSize(&Sec);
uint64_t FileSize = Layout.getSectionFileSize(&Sec);
FileSize += getPaddingSize(&Sec, Layout);
VMSize = std::max(VMSize, Address + Size);
if (Sec.isVirtualSection())
continue;
SectionDataSize = std::max(SectionDataSize, Address + Size);
SectionDataFileSize = std::max(SectionDataFileSize, Address + FileSize);
}
// The section data is padded to 4 bytes.
//
// FIXME: Is this machine dependent?
unsigned SectionDataPadding = OffsetToAlignment(SectionDataFileSize, 4);
SectionDataFileSize += SectionDataPadding;
// Write the prolog, starting with the header and load command...
writeHeader(MachO::MH_OBJECT, NumLoadCommands, LoadCommandsSize,
Asm.getSubsectionsViaSymbols());
uint32_t Prot =
MachO::VM_PROT_READ | MachO::VM_PROT_WRITE | MachO::VM_PROT_EXECUTE;
writeSegmentLoadCommand("", NumSections, 0, VMSize, SectionDataStart,
SectionDataSize, Prot, Prot);
// ... and then the section headers.
uint64_t RelocTableEnd = SectionDataStart + SectionDataFileSize;
for (const MCSection &Section : Asm) {
const auto &Sec = cast<MCSectionMachO>(Section);
std::vector<RelAndSymbol> &Relocs = Relocations[&Sec];
unsigned NumRelocs = Relocs.size();
uint64_t SectionStart = SectionDataStart + getSectionAddress(&Sec);
unsigned Flags = Sec.getTypeAndAttributes();
if (Sec.hasInstructions())
Flags |= MachO::S_ATTR_SOME_INSTRUCTIONS;
writeSection(Layout, Sec, getSectionAddress(&Sec), SectionStart, Flags,
RelocTableEnd, NumRelocs);
RelocTableEnd += NumRelocs * sizeof(MachO::any_relocation_info);
}
// Write out the deployment target information, if it's available.
if (VersionInfo.Major != 0) {
assert(VersionInfo.Update < 256 && "unencodable update target version");
assert(VersionInfo.Minor < 256 && "unencodable minor target version");
assert(VersionInfo.Major < 65536 && "unencodable major target version");
uint32_t EncodedVersion = VersionInfo.Update | (VersionInfo.Minor << 8) |
(VersionInfo.Major << 16);
MachO::LoadCommandType LCType;
switch (VersionInfo.Kind) {
case MCVM_OSXVersionMin:
LCType = MachO::LC_VERSION_MIN_MACOSX;
break;
case MCVM_IOSVersionMin:
LCType = MachO::LC_VERSION_MIN_IPHONEOS;
break;
case MCVM_TvOSVersionMin:
LCType = MachO::LC_VERSION_MIN_TVOS;
break;
case MCVM_WatchOSVersionMin:
LCType = MachO::LC_VERSION_MIN_WATCHOS;
break;
}
write32(LCType);
write32(sizeof(MachO::version_min_command));
write32(EncodedVersion);
write32(0); // reserved.
}
// Write the data-in-code load command, if used.
uint64_t DataInCodeTableEnd = RelocTableEnd + NumDataRegions * 8;
if (NumDataRegions) {
uint64_t DataRegionsOffset = RelocTableEnd;
uint64_t DataRegionsSize = NumDataRegions * 8;
writeLinkeditLoadCommand(MachO::LC_DATA_IN_CODE, DataRegionsOffset,
DataRegionsSize);
}
// Write the loh load command, if used.
uint64_t LOHTableEnd = DataInCodeTableEnd + LOHSize;
if (LOHSize)
writeLinkeditLoadCommand(MachO::LC_LINKER_OPTIMIZATION_HINT,
DataInCodeTableEnd, LOHSize);
// Write the symbol table load command, if used.
if (NumSymbols) {
unsigned FirstLocalSymbol = 0;
unsigned NumLocalSymbols = LocalSymbolData.size();
unsigned FirstExternalSymbol = FirstLocalSymbol + NumLocalSymbols;
unsigned NumExternalSymbols = ExternalSymbolData.size();
unsigned FirstUndefinedSymbol = FirstExternalSymbol + NumExternalSymbols;
unsigned NumUndefinedSymbols = UndefinedSymbolData.size();
unsigned NumIndirectSymbols = Asm.indirect_symbol_size();
unsigned NumSymTabSymbols =
NumLocalSymbols + NumExternalSymbols + NumUndefinedSymbols;
uint64_t IndirectSymbolSize = NumIndirectSymbols * 4;
uint64_t IndirectSymbolOffset = 0;
// If used, the indirect symbols are written after the section data.
if (NumIndirectSymbols)
IndirectSymbolOffset = LOHTableEnd;
// The symbol table is written after the indirect symbol data.
uint64_t SymbolTableOffset = LOHTableEnd + IndirectSymbolSize;
// The string table is written after symbol table.
uint64_t StringTableOffset =
SymbolTableOffset + NumSymTabSymbols * (is64Bit() ?
sizeof(MachO::nlist_64) :
sizeof(MachO::nlist));
writeSymtabLoadCommand(SymbolTableOffset, NumSymTabSymbols,
StringTableOffset, StringTable.data().size());
writeDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols,
FirstExternalSymbol, NumExternalSymbols,
FirstUndefinedSymbol, NumUndefinedSymbols,
IndirectSymbolOffset, NumIndirectSymbols);
}
// Write the linker options load commands.
for (const auto &Option : Asm.getLinkerOptions())
writeLinkerOptionsLoadCommand(Option);
// Write the actual section data.
for (const MCSection &Sec : Asm) {
Asm.writeSectionData(&Sec, Layout);
uint64_t Pad = getPaddingSize(&Sec, Layout);
WriteZeros(Pad);
}
// Write the extra padding.
WriteZeros(SectionDataPadding);
// Write the relocation entries.
for (const MCSection &Sec : Asm) {
// Write the section relocation entries, in reverse order to match 'as'
// (approximately, the exact algorithm is more complicated than this).
std::vector<RelAndSymbol> &Relocs = Relocations[&Sec];
for (const RelAndSymbol &Rel : make_range(Relocs.rbegin(), Relocs.rend())) {
write32(Rel.MRE.r_word0);
write32(Rel.MRE.r_word1);
}
}
// Write out the data-in-code region payload, if there is one.
for (MCAssembler::const_data_region_iterator
it = Asm.data_region_begin(), ie = Asm.data_region_end();
it != ie; ++it) {
const DataRegionData *Data = &(*it);
uint64_t Start = getSymbolAddress(*Data->Start, Layout);
uint64_t End = getSymbolAddress(*Data->End, Layout);
DEBUG(dbgs() << "data in code region-- kind: " << Data->Kind
<< " start: " << Start << "(" << Data->Start->getName() << ")"
<< " end: " << End << "(" << Data->End->getName() << ")"
<< " size: " << End - Start
<< "\n");
write32(Start);
write16(End - Start);
write16(Data->Kind);
}
// Write out the loh commands, if there is one.
if (LOHSize) {
#ifndef NDEBUG
unsigned Start = getStream().tell();
#endif
Asm.getLOHContainer().emit(*this, Layout);
// Pad to a multiple of the pointer size.
writeBytes("", OffsetToAlignment(LOHRawSize, is64Bit() ? 8 : 4));
assert(getStream().tell() - Start == LOHSize);
}
// Write the symbol table data, if used.
if (NumSymbols) {
// Write the indirect symbol entries.
for (MCAssembler::const_indirect_symbol_iterator
it = Asm.indirect_symbol_begin(),
ie = Asm.indirect_symbol_end(); it != ie; ++it) {
// Indirect symbols in the non-lazy symbol pointer section have some
// special handling.
const MCSectionMachO &Section =
static_cast<const MCSectionMachO &>(*it->Section);
if (Section.getType() == MachO::S_NON_LAZY_SYMBOL_POINTERS) {
// If this symbol is defined and internal, mark it as such.
if (it->Symbol->isDefined() && !it->Symbol->isExternal()) {
uint32_t Flags = MachO::INDIRECT_SYMBOL_LOCAL;
if (it->Symbol->isAbsolute())
Flags |= MachO::INDIRECT_SYMBOL_ABS;
write32(Flags);
continue;
}
}
write32(it->Symbol->getIndex());
}
// FIXME: Check that offsets match computed ones.
// Write the symbol table entries.
for (auto *SymbolData :
{&LocalSymbolData, &ExternalSymbolData, &UndefinedSymbolData})
for (MachSymbolData &Entry : *SymbolData)
writeNlist(Entry, Layout);
// Write the string table.
getStream() << StringTable.data();
}
}