MCAsmLayout::MCAsmLayout(MCAssembler &Asm)
: Assembler(Asm), LastValidFragment(0)
{
// Compute the section layout order. Virtual sections must go last.
for (MCAssembler::iterator it = Asm.begin(), ie = Asm.end(); it != ie; ++it)
if (!Asm.getBackend().isVirtualSection(it->getSection()))
SectionOrder.push_back(&*it);
for (MCAssembler::iterator it = Asm.begin(), ie = Asm.end(); it != ie; ++it)
if (Asm.getBackend().isVirtualSection(it->getSection()))
SectionOrder.push_back(&*it);
}
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;
}
} else {
if (!TargetObjectWriter->useAggressiveSymbolFolding())
return false;
}
const MCFragment &FA = *Asm.getSymbolData(SA).getFragment();
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;
}
bool MachObjectWriter::isFixupKindPCRel(const MCAssembler &Asm, unsigned Kind) {
const MCFixupKindInfo &FKI = Asm.getBackend().getFixupKindInfo(
(MCFixupKind) Kind);
return FKI.Flags & MCFixupKindInfo::FKF_IsPCRel;
}
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 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.
if (!Asm.getBackend().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;
}
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 ELFObjectWriter::recordRelocation(MCAssembler &Asm,
const MCAsmLayout &Layout,
const MCFragment *Fragment,
const MCFixup &Fixup, MCValue Target,
uint64_t &FixedValue) {
MCAsmBackend &Backend = Asm.getBackend();
bool IsPCRel = Backend.getFixupKindInfo(Fixup.getKind()).Flags &
MCFixupKindInfo::FKF_IsPCRel;
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()) {
// 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);
}
/// \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);
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());
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_Dummy:
llvm_unreachable("Should not have been added");
}
assert(OW->getStream().tell() - Start == FragmentSize &&
"The stream should advance by fragment size");
}
/// \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");
}
/// WriteFragmentData - Write the \p F data to the output file.
static void WriteFragmentData(const MCAssembler &Asm, const MCAsmLayout &Layout,
const MCFragment &F) {
MCObjectWriter *OW = &Asm.getWriter();
uint64_t Start = OW->getStream().tell();
(void) Start;
++stats::EmittedFragments;
// FIXME: Embed in fragments instead?
uint64_t FragmentSize = Asm.computeFragmentSize(Layout, 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 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: {
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: 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_Inst: {
MCInstFragment &IF = cast<MCInstFragment>(F);
OW->WriteBytes(StringRef(IF.getCode().begin(), IF.getCode().size()));
break;
}
case MCFragment::FT_LEB: {
MCLEBFragment &LF = cast<MCLEBFragment>(F);
OW->WriteBytes(LF.getContents().str());
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;
}
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);
}
