llvm-6502/lib/MC/MachObjectWriter.cpp
Rafael Espindola 712c27f870 Move HasInstructions to MCSection.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@238150 91177308-0d34-0410-b5e6-96231b3b80d8
2015-05-25 18:34:26 +00:00

1008 lines
36 KiB
C++

//===- lib/MC/MachObjectWriter.cpp - Mach-O File Writer -------------------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
#include "llvm/MC/MCMachObjectWriter.h"
#include "llvm/ADT/StringMap.h"
#include "llvm/ADT/Twine.h"
#include "llvm/MC/MCAsmBackend.h"
#include "llvm/MC/MCAsmLayout.h"
#include "llvm/MC/MCAssembler.h"
#include "llvm/MC/MCExpr.h"
#include "llvm/MC/MCFixupKindInfo.h"
#include "llvm/MC/MCMachOSymbolFlags.h"
#include "llvm/MC/MCObjectWriter.h"
#include "llvm/MC/MCSectionMachO.h"
#include "llvm/MC/MCSymbol.h"
#include "llvm/MC/MCValue.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/MachO.h"
#include "llvm/Support/raw_ostream.h"
#include <vector>
using namespace llvm;
#define DEBUG_TYPE "mc"
void MachObjectWriter::reset() {
Relocations.clear();
IndirectSymBase.clear();
StringTable.clear();
LocalSymbolData.clear();
ExternalSymbolData.clear();
UndefinedSymbolData.clear();
MCObjectWriter::reset();
}
bool MachObjectWriter::doesSymbolRequireExternRelocation(const MCSymbol &S) {
// Undefined symbols are always extern.
if (S.isUndefined())
return true;
// References to weak definitions require external relocation entries; the
// definition may not always be the one in the same object file.
if (S.getData().getFlags() & SF_WeakDefinition)
return true;
// Otherwise, we can use an internal relocation.
return false;
}
bool MachObjectWriter::
MachSymbolData::operator<(const MachSymbolData &RHS) const {
return Symbol->getName() < RHS.Symbol->getName();
}
bool MachObjectWriter::isFixupKindPCRel(const MCAssembler &Asm, unsigned Kind) {
const MCFixupKindInfo &FKI = Asm.getBackend().getFixupKindInfo(
(MCFixupKind) Kind);
return FKI.Flags & MCFixupKindInfo::FKF_IsPCRel;
}
uint64_t MachObjectWriter::getFragmentAddress(const MCFragment *Fragment,
const MCAsmLayout &Layout) const {
return getSectionAddress(Fragment->getParent()) +
Layout.getFragmentOffset(Fragment);
}
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.getData().getFragment()->getParent()) +
Layout.getSymbolOffset(S);
}
uint64_t MachObjectWriter::getPaddingSize(const MCSectionData *SD,
const MCAsmLayout &Layout) const {
uint64_t EndAddr = getSectionAddress(SD) + Layout.getSectionAddressSize(SD);
unsigned Next = SD->getSection().getLayoutOrder() + 1;
if (Next >= Layout.getSectionOrder().size())
return 0;
const MCSectionData &NextSD = *Layout.getSectionOrder()[Next];
if (NextSD.getSection().isVirtualSection())
return 0;
return OffsetToAlignment(EndAddr, NextSD.getSection().getAlignment());
}
void MachObjectWriter::WriteHeader(unsigned NumLoadCommands,
unsigned LoadCommandsSize,
bool SubsectionsViaSymbols) {
uint32_t Flags = 0;
if (SubsectionsViaSymbols)
Flags |= MachO::MH_SUBSECTIONS_VIA_SYMBOLS;
// struct mach_header (28 bytes) or
// struct mach_header_64 (32 bytes)
uint64_t Start = OS.tell();
(void) Start;
Write32(is64Bit() ? MachO::MH_MAGIC_64 : MachO::MH_MAGIC);
Write32(TargetObjectWriter->getCPUType());
Write32(TargetObjectWriter->getCPUSubtype());
Write32(MachO::MH_OBJECT);
Write32(NumLoadCommands);
Write32(LoadCommandsSize);
Write32(Flags);
if (is64Bit())
Write32(0); // reserved
assert(OS.tell() - Start ==
(is64Bit()?sizeof(MachO::mach_header_64): sizeof(MachO::mach_header)));
}
/// WriteSegmentLoadCommand - Write a segment load command.
///
/// \param NumSections The number of sections in this segment.
/// \param SectionDataSize The total size of the sections.
void MachObjectWriter::WriteSegmentLoadCommand(unsigned NumSections,
uint64_t VMSize,
uint64_t SectionDataStartOffset,
uint64_t SectionDataSize) {
// struct segment_command (56 bytes) or
// struct segment_command_64 (72 bytes)
uint64_t Start = OS.tell();
(void) Start;
unsigned SegmentLoadCommandSize =
is64Bit() ? sizeof(MachO::segment_command_64):
sizeof(MachO::segment_command);
Write32(is64Bit() ? MachO::LC_SEGMENT_64 : MachO::LC_SEGMENT);
Write32(SegmentLoadCommandSize +
NumSections * (is64Bit() ? sizeof(MachO::section_64) :
sizeof(MachO::section)));
WriteBytes("", 16);
if (is64Bit()) {
Write64(0); // vmaddr
Write64(VMSize); // vmsize
Write64(SectionDataStartOffset); // file offset
Write64(SectionDataSize); // file size
} else {
Write32(0); // vmaddr
Write32(VMSize); // vmsize
Write32(SectionDataStartOffset); // file offset
Write32(SectionDataSize); // file size
}
// maxprot
Write32(MachO::VM_PROT_READ | MachO::VM_PROT_WRITE | MachO::VM_PROT_EXECUTE);
// initprot
Write32(MachO::VM_PROT_READ | MachO::VM_PROT_WRITE | MachO::VM_PROT_EXECUTE);
Write32(NumSections);
Write32(0); // flags
assert(OS.tell() - Start == SegmentLoadCommandSize);
}
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);
const MCSectionMachO &Section = cast<MCSectionMachO>(SD.getSection());
// The offset is unused for virtual sections.
if (Section.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;
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 (Section.hasInstructions())
Flags |= MachO::S_ATTR_SOME_INSTRUCTIONS;
assert(isPowerOf2_32(Section.getAlignment()) && "Invalid alignment!");
Write32(Log2_32(Section.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() ? sizeof(MachO::section_64) :
sizeof(MachO::section)));
}
void MachObjectWriter::WriteSymtabLoadCommand(uint32_t SymbolOffset,
uint32_t NumSymbols,
uint32_t StringTableOffset,
uint32_t StringTableSize) {
// struct symtab_command (24 bytes)
uint64_t Start = OS.tell();
(void) Start;
Write32(MachO::LC_SYMTAB);
Write32(sizeof(MachO::symtab_command));
Write32(SymbolOffset);
Write32(NumSymbols);
Write32(StringTableOffset);
Write32(StringTableSize);
assert(OS.tell() - Start == sizeof(MachO::symtab_command));
}
void MachObjectWriter::WriteDysymtabLoadCommand(uint32_t FirstLocalSymbol,
uint32_t NumLocalSymbols,
uint32_t FirstExternalSymbol,
uint32_t NumExternalSymbols,
uint32_t FirstUndefinedSymbol,
uint32_t NumUndefinedSymbols,
uint32_t IndirectSymbolOffset,
uint32_t NumIndirectSymbols) {
// struct dysymtab_command (80 bytes)
uint64_t Start = OS.tell();
(void) Start;
Write32(MachO::LC_DYSYMTAB);
Write32(sizeof(MachO::dysymtab_command));
Write32(FirstLocalSymbol);
Write32(NumLocalSymbols);
Write32(FirstExternalSymbol);
Write32(NumExternalSymbols);
Write32(FirstUndefinedSymbol);
Write32(NumUndefinedSymbols);
Write32(0); // tocoff
Write32(0); // ntoc
Write32(0); // modtaboff
Write32(0); // nmodtab
Write32(0); // extrefsymoff
Write32(0); // nextrefsyms
Write32(IndirectSymbolOffset);
Write32(NumIndirectSymbols);
Write32(0); // extreloff
Write32(0); // nextrel
Write32(0); // locreloff
Write32(0); // nlocrel
assert(OS.tell() - Start == sizeof(MachO::dysymtab_command));
}
MachObjectWriter::MachSymbolData *
MachObjectWriter::findSymbolData(const MCSymbol &Sym) {
for (auto &Entry : LocalSymbolData)
if (Entry.Symbol == &Sym)
return &Entry;
for (auto &Entry : ExternalSymbolData)
if (Entry.Symbol == &Sym)
return &Entry;
for (auto &Entry : UndefinedSymbolData)
if (Entry.Symbol == &Sym)
return &Entry;
return nullptr;
}
const MCSymbol &MachObjectWriter::findAliasedSymbol(const MCSymbol &Sym) const {
const MCSymbol *S = &Sym;
while (S->isVariable()) {
const MCExpr *Value = S->getVariableValue();
const auto *Ref = dyn_cast<MCSymbolRefExpr>(Value);
if (!Ref)
return *S;
S = &Ref->getSymbol();
}
return *S;
}
void MachObjectWriter::WriteNlist(MachSymbolData &MSD,
const MCAsmLayout &Layout) {
const MCSymbol *Symbol = MSD.Symbol;
MCSymbolData &Data = Symbol->getData();
const MCSymbol *AliasedSymbol = &findAliasedSymbol(*Symbol);
uint8_t SectionIndex = MSD.SectionIndex;
uint8_t Type = 0;
uint16_t Flags = Data.getFlags();
uint64_t Address = 0;
bool IsAlias = Symbol != AliasedSymbol;
const MCSymbol &OrigSymbol = *Symbol;
MachSymbolData *AliaseeInfo;
if (IsAlias) {
AliaseeInfo = findSymbolData(*AliasedSymbol);
if (AliaseeInfo)
SectionIndex = AliaseeInfo->SectionIndex;
Symbol = AliasedSymbol;
// FIXME: Should this update Data as well? Do we need OrigSymbol at all?
}
// Set the N_TYPE bits. See <mach-o/nlist.h>.
//
// FIXME: Are the prebound or indirect fields possible here?
if (IsAlias && Symbol->isUndefined())
Type = MachO::N_INDR;
else if (Symbol->isUndefined())
Type = MachO::N_UNDF;
else if (Symbol->isAbsolute())
Type = MachO::N_ABS;
else
Type = MachO::N_SECT;
// FIXME: Set STAB bits.
if (Data.isPrivateExtern())
Type |= MachO::N_PEXT;
// Set external bit.
if (Data.isExternal() || (!IsAlias && Symbol->isUndefined()))
Type |= MachO::N_EXT;
// Compute the symbol address.
if (IsAlias && Symbol->isUndefined())
Address = AliaseeInfo->StringIndex;
else if (Symbol->isDefined())
Address = getSymbolAddress(OrigSymbol, Layout);
else if (Data.isCommon()) {
// Common symbols are encoded with the size in the address
// field, and their alignment in the flags.
Address = Data.getCommonSize();
// Common alignment is packed into the 'desc' bits.
if (unsigned Align = Data.getCommonAlignment()) {
unsigned Log2Size = Log2_32(Align);
assert((1U << Log2Size) == Align && "Invalid 'common' alignment!");
if (Log2Size > 15)
report_fatal_error("invalid 'common' alignment '" +
Twine(Align) + "' for '" + Symbol->getName() + "'",
false);
// FIXME: Keep this mask with the SymbolFlags enumeration.
Flags = (Flags & 0xF0FF) | (Log2Size << 8);
}
}
if (Layout.getAssembler().isThumbFunc(Symbol))
Flags |= SF_ThumbFunc;
// struct nlist (12 bytes)
Write32(MSD.StringIndex);
Write8(Type);
Write8(SectionIndex);
// The Mach-O streamer uses the lowest 16-bits of the flags for the 'desc'
// value.
Write16(Flags);
if (is64Bit())
Write64(Address);
else
Write32(Address);
}
void MachObjectWriter::WriteLinkeditLoadCommand(uint32_t Type,
uint32_t DataOffset,
uint32_t DataSize) {
uint64_t Start = OS.tell();
(void) Start;
Write32(Type);
Write32(sizeof(MachO::linkedit_data_command));
Write32(DataOffset);
Write32(DataSize);
assert(OS.tell() - Start == sizeof(MachO::linkedit_data_command));
}
static unsigned ComputeLinkerOptionsLoadCommandSize(
const std::vector<std::string> &Options, bool is64Bit)
{
unsigned Size = sizeof(MachO::linker_option_command);
for (unsigned i = 0, e = Options.size(); i != e; ++i)
Size += Options[i].size() + 1;
return RoundUpToAlignment(Size, is64Bit ? 8 : 4);
}
void MachObjectWriter::WriteLinkerOptionsLoadCommand(
const std::vector<std::string> &Options)
{
unsigned Size = ComputeLinkerOptionsLoadCommandSize(Options, is64Bit());
uint64_t Start = OS.tell();
(void) Start;
Write32(MachO::LC_LINKER_OPTION);
Write32(Size);
Write32(Options.size());
uint64_t BytesWritten = sizeof(MachO::linker_option_command);
for (unsigned i = 0, e = Options.size(); i != e; ++i) {
// Write each string, including the null byte.
const std::string &Option = Options[i];
WriteBytes(Option.c_str(), Option.size() + 1);
BytesWritten += Option.size() + 1;
}
// Pad to a multiple of the pointer size.
WriteBytes("", OffsetToAlignment(BytesWritten, is64Bit() ? 8 : 4));
assert(OS.tell() - Start == Size);
}
void MachObjectWriter::RecordRelocation(MCAssembler &Asm,
const MCAsmLayout &Layout,
const MCFragment *Fragment,
const MCFixup &Fixup, MCValue Target,
bool &IsPCRel, uint64_t &FixedValue) {
TargetObjectWriter->RecordRelocation(this, Asm, Layout, Fragment, Fixup,
Target, FixedValue);
}
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() != MachO::S_NON_LAZY_SYMBOL_POINTERS &&
Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS &&
Section.getType() != MachO::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() != MachO::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() != MachO::S_LAZY_SYMBOL_POINTERS &&
Section.getType() != MachO::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);
}
}
/// 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->getSection()] = 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(StringTableBuilder::MachO);
// 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()) {
MCSymbolData &SD = Symbol.getData();
// Ignore non-linker visible symbols.
if (!Asm.isSymbolLinkerVisible(Symbol))
continue;
if (!SD.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()) {
MCSymbolData &SD = Symbol.getData();
// Ignore non-linker visible symbols.
if (!Asm.isSymbolLinkerVisible(Symbol))
continue;
if (SD.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 (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i)
LocalSymbolData[i].Symbol->setIndex(Index++);
for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i)
ExternalSymbolData[i].Symbol->setIndex(Index++);
for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i)
UndefinedSymbolData[i].Symbol->setIndex(Index++);
for (const MCSectionData &SD : Asm) {
std::vector<RelAndSymbol> &Relocs = Relocations[&SD];
for (RelAndSymbol &Rel : Relocs) {
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 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->getSection().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 MachObjectWriter::ExecutePostLayoutBinding(MCAssembler &Asm,
const MCAsmLayout &Layout) {
computeSectionAddresses(Asm, Layout);
// Create symbol data for any indirect symbols.
BindIndirectSymbols(Asm);
}
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()->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 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() != Asm.getSymbolData(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;
}
} else {
if (!TargetObjectWriter->useAggressiveSymbolFolding())
return false;
}
// If they are not in the same section, we can't compute the diff.
if (&SecA != &SecB)
return false;
const MCFragment *FA = Asm.getSymbolData(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;
}
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.
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<RelAndSymbol> &Relocs = Relocations[it];
unsigned NumRelocs = Relocs.size();
uint64_t SectionStart = SectionDataStart + getSectionAddress(it);
WriteSection(Asm, Layout, *it, SectionStart, RelocTableEnd, NumRelocs);
RelocTableEnd += NumRelocs * sizeof(MachO::any_relocation_info);
}
// Write out the deployment target information, if it's available.
if (VersionInfo.Major != 0) {
assert(VersionInfo.Update < 256 && "unencodable update target version");
assert(VersionInfo.Minor < 256 && "unencodable minor target version");
assert(VersionInfo.Major < 65536 && "unencodable major target version");
uint32_t EncodedVersion = VersionInfo.Update | (VersionInfo.Minor << 8) |
(VersionInfo.Major << 16);
Write32(VersionInfo.Kind == MCVM_OSXVersionMin ? MachO::LC_VERSION_MIN_MACOSX :
MachO::LC_VERSION_MIN_IPHONEOS);
Write32(sizeof(MachO::version_min_command));
Write32(EncodedVersion);
Write32(0); // reserved.
}
// Write the data-in-code load command, if used.
uint64_t DataInCodeTableEnd = RelocTableEnd + NumDataRegions * 8;
if (NumDataRegions) {
uint64_t DataRegionsOffset = RelocTableEnd;
uint64_t DataRegionsSize = NumDataRegions * 8;
WriteLinkeditLoadCommand(MachO::LC_DATA_IN_CODE, DataRegionsOffset,
DataRegionsSize);
}
// Write the loh load command, if used.
uint64_t LOHTableEnd = DataInCodeTableEnd + LOHSize;
if (LOHSize)
WriteLinkeditLoadCommand(MachO::LC_LINKER_OPTIMIZATION_HINT,
DataInCodeTableEnd, LOHSize);
// Write the symbol table load command, if used.
if (NumSymbols) {
unsigned FirstLocalSymbol = 0;
unsigned NumLocalSymbols = LocalSymbolData.size();
unsigned FirstExternalSymbol = FirstLocalSymbol + NumLocalSymbols;
unsigned NumExternalSymbols = ExternalSymbolData.size();
unsigned FirstUndefinedSymbol = FirstExternalSymbol + NumExternalSymbols;
unsigned NumUndefinedSymbols = UndefinedSymbolData.size();
unsigned NumIndirectSymbols = Asm.indirect_symbol_size();
unsigned NumSymTabSymbols =
NumLocalSymbols + NumExternalSymbols + NumUndefinedSymbols;
uint64_t IndirectSymbolSize = NumIndirectSymbols * 4;
uint64_t IndirectSymbolOffset = 0;
// If used, the indirect symbols are written after the section data.
if (NumIndirectSymbols)
IndirectSymbolOffset = LOHTableEnd;
// The symbol table is written after the indirect symbol data.
uint64_t SymbolTableOffset = LOHTableEnd + IndirectSymbolSize;
// The string table is written after symbol table.
uint64_t StringTableOffset =
SymbolTableOffset + NumSymTabSymbols * (is64Bit() ?
sizeof(MachO::nlist_64) :
sizeof(MachO::nlist));
WriteSymtabLoadCommand(SymbolTableOffset, NumSymTabSymbols,
StringTableOffset, StringTable.data().size());
WriteDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols,
FirstExternalSymbol, NumExternalSymbols,
FirstUndefinedSymbol, NumUndefinedSymbols,
IndirectSymbolOffset, NumIndirectSymbols);
}
// Write the linker options load commands.
for (unsigned i = 0, e = LinkerOptions.size(); i != e; ++i) {
WriteLinkerOptionsLoadCommand(LinkerOptions[i]);
}
// Write the actual section data.
for (MCAssembler::const_iterator it = Asm.begin(),
ie = Asm.end(); it != ie; ++it) {
Asm.writeSectionData(it, Layout);
uint64_t Pad = getPaddingSize(it, Layout);
WriteZeros(Pad);
}
// 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<RelAndSymbol> &Relocs = Relocations[it];
for (unsigned i = 0, e = Relocs.size(); i != e; ++i) {
Write32(Relocs[e - i - 1].MRE.r_word0);
Write32(Relocs[e - i - 1].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 = OS.tell();
#endif
Asm.getLOHContainer().Emit(*this, Layout);
// Pad to a multiple of the pointer size.
WriteBytes("", OffsetToAlignment(LOHRawSize, is64Bit() ? 8 : 4));
assert(OS.tell() - Start == LOHSize);
}
// Write the symbol table data, if used.
if (NumSymbols) {
// Write the indirect symbol entries.
for (MCAssembler::const_indirect_symbol_iterator
it = Asm.indirect_symbol_begin(),
ie = Asm.indirect_symbol_end(); it != ie; ++it) {
// Indirect symbols in the non-lazy symbol pointer section have some
// special handling.
const MCSectionMachO &Section =
static_cast<const MCSectionMachO&>(it->SectionData->getSection());
if (Section.getType() == MachO::S_NON_LAZY_SYMBOL_POINTERS) {
// If this symbol is defined and internal, mark it as such.
if (it->Symbol->isDefined() &&
!Asm.getSymbolData(*it->Symbol).isExternal()) {
uint32_t Flags = MachO::INDIRECT_SYMBOL_LOCAL;
if (it->Symbol->isAbsolute())
Flags |= MachO::INDIRECT_SYMBOL_ABS;
Write32(Flags);
continue;
}
}
Write32(it->Symbol->getIndex());
}
// FIXME: Check that offsets match computed ones.
// Write the symbol table entries.
for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i)
WriteNlist(LocalSymbolData[i], Layout);
for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i)
WriteNlist(ExternalSymbolData[i], Layout);
for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i)
WriteNlist(UndefinedSymbolData[i], Layout);
// Write the string table.
OS << StringTable.data();
}
}
MCObjectWriter *llvm::createMachObjectWriter(MCMachObjectTargetWriter *MOTW,
raw_pwrite_stream &OS,
bool IsLittleEndian) {
return new MachObjectWriter(MOTW, OS, IsLittleEndian);
}