1 //===- lib/MC/MCAssembler.cpp - Assembler Backend Implementation ----------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 #define DEBUG_TYPE "assembler"
11 #include "llvm/MC/MCAssembler.h"
12 #include "llvm/MC/MCExpr.h"
13 #include "llvm/MC/MCSectionMachO.h"
14 #include "llvm/MC/MCSymbol.h"
15 #include "llvm/MC/MCValue.h"
16 #include "llvm/ADT/DenseMap.h"
17 #include "llvm/ADT/SmallString.h"
18 #include "llvm/ADT/Statistic.h"
19 #include "llvm/ADT/StringExtras.h"
20 #include "llvm/ADT/StringMap.h"
21 #include "llvm/ADT/Twine.h"
22 #include "llvm/Support/ErrorHandling.h"
23 #include "llvm/Support/MachO.h"
24 #include "llvm/Support/raw_ostream.h"
25 #include "llvm/Support/Debug.h"
29 class MachObjectWriter;
31 STATISTIC(EmittedFragments, "Number of emitted assembler fragments");
33 // FIXME FIXME FIXME: There are number of places in this file where we convert
34 // what is a 64-bit assembler value used for computation into a value in the
35 // object file, which may truncate it. We should detect that truncation where
36 // invalid and report errors back.
38 static void WriteFileData(raw_ostream &OS, const MCSectionData &SD,
39 MachObjectWriter &MOW);
41 /// isVirtualSection - Check if this is a section which does not actually exist
42 /// in the object file.
43 static bool isVirtualSection(const MCSection &Section) {
45 const MCSectionMachO &SMO = static_cast<const MCSectionMachO&>(Section);
46 unsigned Type = SMO.getTypeAndAttributes() & MCSectionMachO::SECTION_TYPE;
47 return (Type == MCSectionMachO::S_ZEROFILL);
50 class MachObjectWriter {
51 // See <mach-o/loader.h>.
53 Header_Magic32 = 0xFEEDFACE,
54 Header_Magic64 = 0xFEEDFACF
57 static const unsigned Header32Size = 28;
58 static const unsigned Header64Size = 32;
59 static const unsigned SegmentLoadCommand32Size = 56;
60 static const unsigned Section32Size = 68;
61 static const unsigned SymtabLoadCommandSize = 24;
62 static const unsigned DysymtabLoadCommandSize = 80;
63 static const unsigned Nlist32Size = 12;
64 static const unsigned RelocationInfoSize = 8;
71 HF_SubsectionsViaSymbols = 0x2000
74 enum LoadCommandType {
80 // See <mach-o/nlist.h>.
87 enum SymbolTypeFlags {
88 // If any of these bits are set, then the entry is a stab entry number (see
89 // <mach-o/stab.h>. Otherwise the other masks apply.
90 STF_StabsEntryMask = 0xe0,
94 STF_PrivateExtern = 0x10
97 /// IndirectSymbolFlags - Flags for encoding special values in the indirect
99 enum IndirectSymbolFlags {
100 ISF_Local = 0x80000000,
101 ISF_Absolute = 0x40000000
104 /// RelocationFlags - Special flags for addresses.
105 enum RelocationFlags {
106 RF_Scattered = 0x80000000
109 enum RelocationInfoType {
113 RIT_PreboundLazyPointer = 3,
114 RIT_LocalDifference = 4
117 /// MachSymbolData - Helper struct for containing some precomputed information
119 struct MachSymbolData {
120 MCSymbolData *SymbolData;
121 uint64_t StringIndex;
122 uint8_t SectionIndex;
124 // Support lexicographic sorting.
125 bool operator<(const MachSymbolData &RHS) const {
126 const std::string &Name = SymbolData->getSymbol().getName();
127 return Name < RHS.SymbolData->getSymbol().getName();
135 MachObjectWriter(raw_ostream &_OS, bool _IsLSB = true)
136 : OS(_OS), IsLSB(_IsLSB) {
139 /// @name Helper Methods
142 void Write8(uint8_t Value) {
146 void Write16(uint16_t Value) {
148 Write8(uint8_t(Value >> 0));
149 Write8(uint8_t(Value >> 8));
151 Write8(uint8_t(Value >> 8));
152 Write8(uint8_t(Value >> 0));
156 void Write32(uint32_t Value) {
158 Write16(uint16_t(Value >> 0));
159 Write16(uint16_t(Value >> 16));
161 Write16(uint16_t(Value >> 16));
162 Write16(uint16_t(Value >> 0));
166 void Write64(uint64_t Value) {
168 Write32(uint32_t(Value >> 0));
169 Write32(uint32_t(Value >> 32));
171 Write32(uint32_t(Value >> 32));
172 Write32(uint32_t(Value >> 0));
176 void WriteZeros(unsigned N) {
177 const char Zeros[16] = { 0 };
179 for (unsigned i = 0, e = N / 16; i != e; ++i)
180 OS << StringRef(Zeros, 16);
182 OS << StringRef(Zeros, N % 16);
185 void WriteString(StringRef Str, unsigned ZeroFillSize = 0) {
188 WriteZeros(ZeroFillSize - Str.size());
193 void WriteHeader32(unsigned NumLoadCommands, unsigned LoadCommandsSize,
194 bool SubsectionsViaSymbols) {
197 if (SubsectionsViaSymbols)
198 Flags |= HF_SubsectionsViaSymbols;
200 // struct mach_header (28 bytes)
202 uint64_t Start = OS.tell();
205 Write32(Header_Magic32);
207 // FIXME: Support cputype.
208 Write32(MachO::CPUTypeI386);
209 // FIXME: Support cpusubtype.
210 Write32(MachO::CPUSubType_I386_ALL);
212 Write32(NumLoadCommands); // Object files have a single load command, the
214 Write32(LoadCommandsSize);
217 assert(OS.tell() - Start == Header32Size);
220 /// WriteSegmentLoadCommand32 - Write a 32-bit segment load command.
222 /// \arg NumSections - The number of sections in this segment.
223 /// \arg SectionDataSize - The total size of the sections.
224 void WriteSegmentLoadCommand32(unsigned NumSections,
226 uint64_t SectionDataStartOffset,
227 uint64_t SectionDataSize) {
228 // struct segment_command (56 bytes)
230 uint64_t Start = OS.tell();
233 Write32(LCT_Segment);
234 Write32(SegmentLoadCommand32Size + NumSections * Section32Size);
237 Write32(0); // vmaddr
238 Write32(VMSize); // vmsize
239 Write32(SectionDataStartOffset); // file offset
240 Write32(SectionDataSize); // file size
241 Write32(0x7); // maxprot
242 Write32(0x7); // initprot
243 Write32(NumSections);
246 assert(OS.tell() - Start == SegmentLoadCommand32Size);
249 void WriteSection32(const MCSectionData &SD, uint64_t FileOffset,
250 uint64_t RelocationsStart, unsigned NumRelocations) {
251 // The offset is unused for virtual sections.
252 if (isVirtualSection(SD.getSection())) {
253 assert(SD.getFileSize() == 0 && "Invalid file size!");
257 // struct section (68 bytes)
259 uint64_t Start = OS.tell();
262 // FIXME: cast<> support!
263 const MCSectionMachO &Section =
264 static_cast<const MCSectionMachO&>(SD.getSection());
265 WriteString(Section.getSectionName(), 16);
266 WriteString(Section.getSegmentName(), 16);
267 Write32(SD.getAddress()); // address
268 Write32(SD.getSize()); // size
271 unsigned Flags = Section.getTypeAndAttributes();
272 if (SD.hasInstructions())
273 Flags |= MCSectionMachO::S_ATTR_SOME_INSTRUCTIONS;
275 assert(isPowerOf2_32(SD.getAlignment()) && "Invalid alignment!");
276 Write32(Log2_32(SD.getAlignment()));
277 Write32(NumRelocations ? RelocationsStart : 0);
278 Write32(NumRelocations);
280 Write32(0); // reserved1
281 Write32(Section.getStubSize()); // reserved2
283 assert(OS.tell() - Start == Section32Size);
286 void WriteSymtabLoadCommand(uint32_t SymbolOffset, uint32_t NumSymbols,
287 uint32_t StringTableOffset,
288 uint32_t StringTableSize) {
289 // struct symtab_command (24 bytes)
291 uint64_t Start = OS.tell();
295 Write32(SymtabLoadCommandSize);
296 Write32(SymbolOffset);
298 Write32(StringTableOffset);
299 Write32(StringTableSize);
301 assert(OS.tell() - Start == SymtabLoadCommandSize);
304 void WriteDysymtabLoadCommand(uint32_t FirstLocalSymbol,
305 uint32_t NumLocalSymbols,
306 uint32_t FirstExternalSymbol,
307 uint32_t NumExternalSymbols,
308 uint32_t FirstUndefinedSymbol,
309 uint32_t NumUndefinedSymbols,
310 uint32_t IndirectSymbolOffset,
311 uint32_t NumIndirectSymbols) {
312 // struct dysymtab_command (80 bytes)
314 uint64_t Start = OS.tell();
317 Write32(LCT_Dysymtab);
318 Write32(DysymtabLoadCommandSize);
319 Write32(FirstLocalSymbol);
320 Write32(NumLocalSymbols);
321 Write32(FirstExternalSymbol);
322 Write32(NumExternalSymbols);
323 Write32(FirstUndefinedSymbol);
324 Write32(NumUndefinedSymbols);
325 Write32(0); // tocoff
327 Write32(0); // modtaboff
328 Write32(0); // nmodtab
329 Write32(0); // extrefsymoff
330 Write32(0); // nextrefsyms
331 Write32(IndirectSymbolOffset);
332 Write32(NumIndirectSymbols);
333 Write32(0); // extreloff
334 Write32(0); // nextrel
335 Write32(0); // locreloff
336 Write32(0); // nlocrel
338 assert(OS.tell() - Start == DysymtabLoadCommandSize);
341 void WriteNlist32(MachSymbolData &MSD) {
342 MCSymbolData &Data = *MSD.SymbolData;
343 const MCSymbol &Symbol = Data.getSymbol();
345 uint16_t Flags = Data.getFlags();
346 uint32_t Address = 0;
348 // Set the N_TYPE bits. See <mach-o/nlist.h>.
350 // FIXME: Are the prebound or indirect fields possible here?
351 if (Symbol.isUndefined())
352 Type = STT_Undefined;
353 else if (Symbol.isAbsolute())
358 // FIXME: Set STAB bits.
360 if (Data.isPrivateExtern())
361 Type |= STF_PrivateExtern;
364 if (Data.isExternal() || Symbol.isUndefined())
365 Type |= STF_External;
367 // Compute the symbol address.
368 if (Symbol.isDefined()) {
369 if (Symbol.isAbsolute()) {
370 llvm_unreachable("FIXME: Not yet implemented!");
372 Address = Data.getFragment()->getAddress() + Data.getOffset();
374 } else if (Data.isCommon()) {
375 // Common symbols are encoded with the size in the address
376 // field, and their alignment in the flags.
377 Address = Data.getCommonSize();
379 // Common alignment is packed into the 'desc' bits.
380 if (unsigned Align = Data.getCommonAlignment()) {
381 unsigned Log2Size = Log2_32(Align);
382 assert((1U << Log2Size) == Align && "Invalid 'common' alignment!");
384 llvm_report_error("invalid 'common' alignment '" +
386 // FIXME: Keep this mask with the SymbolFlags enumeration.
387 Flags = (Flags & 0xF0FF) | (Log2Size << 8);
391 // struct nlist (12 bytes)
393 Write32(MSD.StringIndex);
395 Write8(MSD.SectionIndex);
397 // The Mach-O streamer uses the lowest 16-bits of the flags for the 'desc'
403 struct MachRelocationEntry {
407 void ComputeScatteredRelocationInfo(MCAssembler &Asm, MCFragment &Fragment,
409 const MCValue &Target,
410 DenseMap<const MCSymbol*,MCSymbolData*> &SymbolMap,
411 std::vector<MachRelocationEntry> &Relocs) {
412 uint32_t Address = Fragment.getOffset() + Fixup.Offset;
413 unsigned IsPCRel = 0;
414 unsigned Type = RIT_Vanilla;
417 const MCSymbol *A = Target.getSymA();
418 MCSymbolData *SD = SymbolMap.lookup(A);
419 uint32_t Value = SD->getFragment()->getAddress() + SD->getOffset();
422 if (const MCSymbol *B = Target.getSymB()) {
423 Type = RIT_LocalDifference;
425 MCSymbolData *SD = SymbolMap.lookup(B);
426 Value2 = SD->getFragment()->getAddress() + SD->getOffset();
429 unsigned Log2Size = Log2_32(Fixup.Size);
430 assert((1U << Log2Size) == Fixup.Size && "Invalid fixup size!");
432 // The value which goes in the fixup is current value of the expression.
433 Fixup.FixedValue = Value - Value2 + Target.getConstant();
435 MachRelocationEntry MRE;
436 MRE.Word0 = ((Address << 0) |
442 Relocs.push_back(MRE);
444 if (Type == RIT_LocalDifference) {
447 MachRelocationEntry MRE;
448 MRE.Word0 = ((0 << 0) |
454 Relocs.push_back(MRE);
458 void ComputeRelocationInfo(MCAssembler &Asm, MCFragment &Fragment,
460 DenseMap<const MCSymbol*,MCSymbolData*> &SymbolMap,
461 std::vector<MachRelocationEntry> &Relocs) {
463 if (!Fixup.Value->EvaluateAsRelocatable(Target))
464 llvm_report_error("expected relocatable expression");
466 // If this is a difference or a local symbol plus an offset, then we need a
467 // scattered relocation entry.
468 if (Target.getSymB() ||
469 (Target.getSymA() && !Target.getSymA()->isUndefined() &&
470 Target.getConstant()))
471 return ComputeScatteredRelocationInfo(Asm, Fragment, Fixup, Target,
475 uint32_t Address = Fragment.getOffset() + Fixup.Offset;
478 unsigned IsPCRel = 0;
479 unsigned IsExtern = 0;
482 if (Target.isAbsolute()) { // constant
483 // SymbolNum of 0 indicates the absolute section.
485 // FIXME: When is this generated?
488 llvm_unreachable("FIXME: Not yet implemented!");
490 const MCSymbol *Symbol = Target.getSymA();
491 MCSymbolData *SD = SymbolMap.lookup(Symbol);
493 if (Symbol->isUndefined()) {
495 Index = SD->getIndex();
498 // The index is the section ordinal.
502 for (MCAssembler::iterator it = Asm.begin(),
503 ie = Asm.end(); it != ie; ++it, ++Index)
504 if (&*it == SD->getFragment()->getParent())
506 Value = SD->getFragment()->getAddress() + SD->getOffset();
512 // The value which goes in the fixup is current value of the expression.
513 Fixup.FixedValue = Value + Target.getConstant();
515 unsigned Log2Size = Log2_32(Fixup.Size);
516 assert((1U << Log2Size) == Fixup.Size && "Invalid fixup size!");
518 // struct relocation_info (8 bytes)
519 MachRelocationEntry MRE;
521 MRE.Word1 = ((Index << 0) |
526 Relocs.push_back(MRE);
529 void BindIndirectSymbols(MCAssembler &Asm,
530 DenseMap<const MCSymbol*,MCSymbolData*> &SymbolMap) {
531 // This is the point where 'as' creates actual symbols for indirect symbols
532 // (in the following two passes). It would be easier for us to do this
533 // sooner when we see the attribute, but that makes getting the order in the
534 // symbol table much more complicated than it is worth.
536 // FIXME: Revisit this when the dust settles.
538 // Bind non lazy symbol pointers first.
539 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
540 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
541 // FIXME: cast<> support!
542 const MCSectionMachO &Section =
543 static_cast<const MCSectionMachO&>(it->SectionData->getSection());
546 Section.getTypeAndAttributes() & MCSectionMachO::SECTION_TYPE;
547 if (Type != MCSectionMachO::S_NON_LAZY_SYMBOL_POINTERS)
550 MCSymbolData *&Entry = SymbolMap[it->Symbol];
552 Entry = new MCSymbolData(*it->Symbol, 0, 0, &Asm);
555 // Then lazy symbol pointers and symbol stubs.
556 for (MCAssembler::indirect_symbol_iterator it = Asm.indirect_symbol_begin(),
557 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
558 // FIXME: cast<> support!
559 const MCSectionMachO &Section =
560 static_cast<const MCSectionMachO&>(it->SectionData->getSection());
563 Section.getTypeAndAttributes() & MCSectionMachO::SECTION_TYPE;
564 if (Type != MCSectionMachO::S_LAZY_SYMBOL_POINTERS &&
565 Type != MCSectionMachO::S_SYMBOL_STUBS)
568 MCSymbolData *&Entry = SymbolMap[it->Symbol];
570 Entry = new MCSymbolData(*it->Symbol, 0, 0, &Asm);
572 // Set the symbol type to undefined lazy, but only on construction.
574 // FIXME: Do not hardcode.
575 Entry->setFlags(Entry->getFlags() | 0x0001);
580 /// ComputeSymbolTable - Compute the symbol table data
582 /// \param StringTable [out] - The string table data.
583 /// \param StringIndexMap [out] - Map from symbol names to offsets in the
585 void ComputeSymbolTable(MCAssembler &Asm, SmallString<256> &StringTable,
586 std::vector<MachSymbolData> &LocalSymbolData,
587 std::vector<MachSymbolData> &ExternalSymbolData,
588 std::vector<MachSymbolData> &UndefinedSymbolData) {
589 // Build section lookup table.
590 DenseMap<const MCSection*, uint8_t> SectionIndexMap;
592 for (MCAssembler::iterator it = Asm.begin(),
593 ie = Asm.end(); it != ie; ++it, ++Index)
594 SectionIndexMap[&it->getSection()] = Index;
595 assert(Index <= 256 && "Too many sections!");
597 // Index 0 is always the empty string.
598 StringMap<uint64_t> StringIndexMap;
599 StringTable += '\x00';
601 // Build the symbol arrays and the string table, but only for non-local
604 // The particular order that we collect the symbols and create the string
605 // table, then sort the symbols is chosen to match 'as'. Even though it
606 // doesn't matter for correctness, this is important for letting us diff .o
608 for (MCAssembler::symbol_iterator it = Asm.symbol_begin(),
609 ie = Asm.symbol_end(); it != ie; ++it) {
610 const MCSymbol &Symbol = it->getSymbol();
612 // Ignore assembler temporaries.
613 if (it->getSymbol().isTemporary())
616 if (!it->isExternal() && !Symbol.isUndefined())
619 uint64_t &Entry = StringIndexMap[Symbol.getName()];
621 Entry = StringTable.size();
622 StringTable += Symbol.getName();
623 StringTable += '\x00';
628 MSD.StringIndex = Entry;
630 if (Symbol.isUndefined()) {
631 MSD.SectionIndex = 0;
632 UndefinedSymbolData.push_back(MSD);
633 } else if (Symbol.isAbsolute()) {
634 MSD.SectionIndex = 0;
635 ExternalSymbolData.push_back(MSD);
637 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
638 assert(MSD.SectionIndex && "Invalid section index!");
639 ExternalSymbolData.push_back(MSD);
643 // Now add the data for local symbols.
644 for (MCAssembler::symbol_iterator it = Asm.symbol_begin(),
645 ie = Asm.symbol_end(); it != ie; ++it) {
646 const MCSymbol &Symbol = it->getSymbol();
648 // Ignore assembler temporaries.
649 if (it->getSymbol().isTemporary())
652 if (it->isExternal() || Symbol.isUndefined())
655 uint64_t &Entry = StringIndexMap[Symbol.getName()];
657 Entry = StringTable.size();
658 StringTable += Symbol.getName();
659 StringTable += '\x00';
664 MSD.StringIndex = Entry;
666 if (Symbol.isAbsolute()) {
667 MSD.SectionIndex = 0;
668 LocalSymbolData.push_back(MSD);
670 MSD.SectionIndex = SectionIndexMap.lookup(&Symbol.getSection());
671 assert(MSD.SectionIndex && "Invalid section index!");
672 LocalSymbolData.push_back(MSD);
676 // External and undefined symbols are required to be in lexicographic order.
677 std::sort(ExternalSymbolData.begin(), ExternalSymbolData.end());
678 std::sort(UndefinedSymbolData.begin(), UndefinedSymbolData.end());
680 // Set the symbol indices.
682 for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i)
683 LocalSymbolData[i].SymbolData->setIndex(Index++);
684 for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i)
685 ExternalSymbolData[i].SymbolData->setIndex(Index++);
686 for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i)
687 UndefinedSymbolData[i].SymbolData->setIndex(Index++);
689 // The string table is padded to a multiple of 4.
690 while (StringTable.size() % 4)
691 StringTable += '\x00';
694 void WriteObject(MCAssembler &Asm) {
695 unsigned NumSections = Asm.size();
697 // Compute the symbol -> symbol data map.
699 // FIXME: This should not be here.
700 DenseMap<const MCSymbol*, MCSymbolData *> SymbolMap;
701 for (MCAssembler::symbol_iterator it = Asm.symbol_begin(),
702 ie = Asm.symbol_end(); it != ie; ++it)
703 SymbolMap[&it->getSymbol()] = it;
705 // Create symbol data for any indirect symbols.
706 BindIndirectSymbols(Asm, SymbolMap);
708 // Compute symbol table information.
709 SmallString<256> StringTable;
710 std::vector<MachSymbolData> LocalSymbolData;
711 std::vector<MachSymbolData> ExternalSymbolData;
712 std::vector<MachSymbolData> UndefinedSymbolData;
713 unsigned NumSymbols = Asm.symbol_size();
715 // No symbol table command is written if there are no symbols.
717 ComputeSymbolTable(Asm, StringTable, LocalSymbolData, ExternalSymbolData,
718 UndefinedSymbolData);
720 // The section data starts after the header, the segment load command (and
721 // section headers) and the symbol table.
722 unsigned NumLoadCommands = 1;
723 uint64_t LoadCommandsSize =
724 SegmentLoadCommand32Size + NumSections * Section32Size;
726 // Add the symbol table load command sizes, if used.
728 NumLoadCommands += 2;
729 LoadCommandsSize += SymtabLoadCommandSize + DysymtabLoadCommandSize;
732 // Compute the total size of the section data, as well as its file size and
734 uint64_t SectionDataStart = Header32Size + LoadCommandsSize;
735 uint64_t SectionDataSize = 0;
736 uint64_t SectionDataFileSize = 0;
738 for (MCAssembler::iterator it = Asm.begin(),
739 ie = Asm.end(); it != ie; ++it) {
740 MCSectionData &SD = *it;
742 VMSize = std::max(VMSize, SD.getAddress() + SD.getSize());
744 if (isVirtualSection(SD.getSection()))
747 SectionDataSize = std::max(SectionDataSize,
748 SD.getAddress() + SD.getSize());
749 SectionDataFileSize = std::max(SectionDataFileSize,
750 SD.getAddress() + SD.getFileSize());
753 // The section data is padded to 4 bytes.
755 // FIXME: Is this machine dependent?
756 unsigned SectionDataPadding = OffsetToAlignment(SectionDataFileSize, 4);
757 SectionDataFileSize += SectionDataPadding;
759 // Write the prolog, starting with the header and load command...
760 WriteHeader32(NumLoadCommands, LoadCommandsSize,
761 Asm.getSubsectionsViaSymbols());
762 WriteSegmentLoadCommand32(NumSections, VMSize,
763 SectionDataStart, SectionDataSize);
765 // ... and then the section headers.
767 // We also compute the section relocations while we do this. Note that
768 // computing relocation info will also update the fixup to have the correct
769 // value; this will overwrite the appropriate data in the fragment when it
771 std::vector<MachRelocationEntry> RelocInfos;
772 uint64_t RelocTableEnd = SectionDataStart + SectionDataFileSize;
773 for (MCAssembler::iterator it = Asm.begin(),
774 ie = Asm.end(); it != ie; ++it) {
775 MCSectionData &SD = *it;
777 // The assembler writes relocations in the reverse order they were seen.
779 // FIXME: It is probably more complicated than this.
780 unsigned NumRelocsStart = RelocInfos.size();
781 for (MCSectionData::reverse_iterator it2 = SD.rbegin(),
782 ie2 = SD.rend(); it2 != ie2; ++it2)
783 for (unsigned i = 0, e = it2->fixup_size(); i != e; ++i)
784 ComputeRelocationInfo(Asm, *it2, it2->getFixups()[e - i - 1],
785 SymbolMap, RelocInfos);
787 unsigned NumRelocs = RelocInfos.size() - NumRelocsStart;
788 uint64_t SectionStart = SectionDataStart + SD.getAddress();
789 WriteSection32(SD, SectionStart, RelocTableEnd, NumRelocs);
790 RelocTableEnd += NumRelocs * RelocationInfoSize;
793 // Write the symbol table load command, if used.
795 unsigned FirstLocalSymbol = 0;
796 unsigned NumLocalSymbols = LocalSymbolData.size();
797 unsigned FirstExternalSymbol = FirstLocalSymbol + NumLocalSymbols;
798 unsigned NumExternalSymbols = ExternalSymbolData.size();
799 unsigned FirstUndefinedSymbol = FirstExternalSymbol + NumExternalSymbols;
800 unsigned NumUndefinedSymbols = UndefinedSymbolData.size();
801 unsigned NumIndirectSymbols = Asm.indirect_symbol_size();
802 unsigned NumSymTabSymbols =
803 NumLocalSymbols + NumExternalSymbols + NumUndefinedSymbols;
804 uint64_t IndirectSymbolSize = NumIndirectSymbols * 4;
805 uint64_t IndirectSymbolOffset = 0;
807 // If used, the indirect symbols are written after the section data.
808 if (NumIndirectSymbols)
809 IndirectSymbolOffset = RelocTableEnd;
811 // The symbol table is written after the indirect symbol data.
812 uint64_t SymbolTableOffset = RelocTableEnd + IndirectSymbolSize;
814 // The string table is written after symbol table.
815 uint64_t StringTableOffset =
816 SymbolTableOffset + NumSymTabSymbols * Nlist32Size;
817 WriteSymtabLoadCommand(SymbolTableOffset, NumSymTabSymbols,
818 StringTableOffset, StringTable.size());
820 WriteDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols,
821 FirstExternalSymbol, NumExternalSymbols,
822 FirstUndefinedSymbol, NumUndefinedSymbols,
823 IndirectSymbolOffset, NumIndirectSymbols);
826 // Write the actual section data.
827 for (MCAssembler::iterator it = Asm.begin(), ie = Asm.end(); it != ie; ++it)
828 WriteFileData(OS, *it, *this);
830 // Write the extra padding.
831 WriteZeros(SectionDataPadding);
833 // Write the relocation entries.
834 for (unsigned i = 0, e = RelocInfos.size(); i != e; ++i) {
835 Write32(RelocInfos[i].Word0);
836 Write32(RelocInfos[i].Word1);
839 // Write the symbol table data, if used.
841 // Write the indirect symbol entries.
842 for (MCAssembler::indirect_symbol_iterator
843 it = Asm.indirect_symbol_begin(),
844 ie = Asm.indirect_symbol_end(); it != ie; ++it) {
845 // Indirect symbols in the non lazy symbol pointer section have some
847 const MCSectionMachO &Section =
848 static_cast<const MCSectionMachO&>(it->SectionData->getSection());
850 Section.getTypeAndAttributes() & MCSectionMachO::SECTION_TYPE;
851 if (Type == MCSectionMachO::S_NON_LAZY_SYMBOL_POINTERS) {
852 // If this symbol is defined and internal, mark it as such.
853 if (it->Symbol->isDefined() &&
854 !SymbolMap.lookup(it->Symbol)->isExternal()) {
855 uint32_t Flags = ISF_Local;
856 if (it->Symbol->isAbsolute())
857 Flags |= ISF_Absolute;
863 Write32(SymbolMap[it->Symbol]->getIndex());
866 // FIXME: Check that offsets match computed ones.
868 // Write the symbol table entries.
869 for (unsigned i = 0, e = LocalSymbolData.size(); i != e; ++i)
870 WriteNlist32(LocalSymbolData[i]);
871 for (unsigned i = 0, e = ExternalSymbolData.size(); i != e; ++i)
872 WriteNlist32(ExternalSymbolData[i]);
873 for (unsigned i = 0, e = UndefinedSymbolData.size(); i != e; ++i)
874 WriteNlist32(UndefinedSymbolData[i]);
876 // Write the string table.
877 OS << StringTable.str();
881 void ApplyFixup(const MCAsmFixup &Fixup, MCDataFragment &DF) {
882 // FIXME: Endianness assumption.
883 for (unsigned i = 0; i != Fixup.Size; ++i)
884 DF.getContents()[Fixup.Offset + i] = uint8_t(Fixup.FixedValue >> (i * 8));
890 MCFragment::MCFragment() : Kind(FragmentType(~0)) {
893 MCFragment::MCFragment(FragmentType _Kind, MCSectionData *_Parent)
896 FileSize(~UINT64_C(0))
899 Parent->getFragmentList().push_back(this);
902 MCFragment::~MCFragment() {
905 uint64_t MCFragment::getAddress() const {
906 assert(getParent() && "Missing Section!");
907 return getParent()->getAddress() + Offset;
912 MCSectionData::MCSectionData() : Section(0) {}
914 MCSectionData::MCSectionData(const MCSection &_Section, MCAssembler *A)
915 : Section(&_Section),
917 Address(~UINT64_C(0)),
919 FileSize(~UINT64_C(0)),
920 HasInstructions(false)
923 A->getSectionList().push_back(this);
928 MCSymbolData::MCSymbolData() : Symbol(0) {}
930 MCSymbolData::MCSymbolData(const MCSymbol &_Symbol, MCFragment *_Fragment,
931 uint64_t _Offset, MCAssembler *A)
932 : Symbol(&_Symbol), Fragment(_Fragment), Offset(_Offset),
933 IsExternal(false), IsPrivateExtern(false),
934 CommonSize(0), CommonAlign(0), Flags(0), Index(0)
937 A->getSymbolList().push_back(this);
942 MCAssembler::MCAssembler(MCContext &_Context, raw_ostream &_OS)
943 : Context(_Context), OS(_OS), SubsectionsViaSymbols(false)
947 MCAssembler::~MCAssembler() {
950 void MCAssembler::LayoutSection(MCSectionData &SD) {
951 uint64_t Address = SD.getAddress();
953 for (MCSectionData::iterator it = SD.begin(), ie = SD.end(); it != ie; ++it) {
956 F.setOffset(Address - SD.getAddress());
958 // Evaluate fragment size.
959 switch (F.getKind()) {
960 case MCFragment::FT_Align: {
961 MCAlignFragment &AF = cast<MCAlignFragment>(F);
963 uint64_t Size = OffsetToAlignment(Address, AF.getAlignment());
964 if (Size > AF.getMaxBytesToEmit())
967 AF.setFileSize(Size);
971 case MCFragment::FT_Data:
972 F.setFileSize(F.getMaxFileSize());
975 case MCFragment::FT_Fill: {
976 MCFillFragment &FF = cast<MCFillFragment>(F);
978 F.setFileSize(F.getMaxFileSize());
981 if (!FF.getValue().EvaluateAsRelocatable(Target))
982 llvm_report_error("expected relocatable expression");
984 // If the fill value is constant, thats it.
985 if (Target.isAbsolute())
988 // Otherwise, add fixups for the values.
990 // FIXME: What we want to do here is lower this to a data fragment once we
991 // realize it will need relocations. This means that the only place we
992 // need to worry about relocations and fixing is on data fragments.
993 for (uint64_t i = 0, e = FF.getCount(); i != e; ++i)
994 FF.getFixups().push_back(MCAsmFixup(i*FF.getValueSize(), FF.getValue(),
999 case MCFragment::FT_Org: {
1000 MCOrgFragment &OF = cast<MCOrgFragment>(F);
1003 if (!OF.getOffset().EvaluateAsRelocatable(Target))
1004 llvm_report_error("expected relocatable expression");
1006 if (!Target.isAbsolute())
1007 llvm_unreachable("FIXME: Not yet implemented!");
1008 uint64_t OrgOffset = Target.getConstant();
1009 uint64_t Offset = Address - SD.getAddress();
1011 // FIXME: We need a way to communicate this error.
1012 if (OrgOffset < Offset)
1013 llvm_report_error("invalid .org offset '" + Twine(OrgOffset) +
1014 "' (at offset '" + Twine(Offset) + "'");
1016 F.setFileSize(OrgOffset - Offset);
1020 case MCFragment::FT_ZeroFill: {
1021 MCZeroFillFragment &ZFF = cast<MCZeroFillFragment>(F);
1023 // Align the fragment offset; it is safe to adjust the offset freely since
1024 // this is only in virtual sections.
1025 uint64_t Aligned = RoundUpToAlignment(Address, ZFF.getAlignment());
1026 F.setOffset(Aligned - SD.getAddress());
1028 // FIXME: This is misnamed.
1029 F.setFileSize(ZFF.getSize());
1034 Address += F.getFileSize();
1037 // Set the section sizes.
1038 SD.setSize(Address - SD.getAddress());
1039 if (isVirtualSection(SD.getSection()))
1042 SD.setFileSize(Address - SD.getAddress());
1045 /// WriteFileData - Write the \arg F data to the output file.
1046 static void WriteFileData(raw_ostream &OS, const MCFragment &F,
1047 MachObjectWriter &MOW) {
1048 uint64_t Start = OS.tell();
1053 // FIXME: Embed in fragments instead?
1054 switch (F.getKind()) {
1055 case MCFragment::FT_Align: {
1056 MCAlignFragment &AF = cast<MCAlignFragment>(F);
1057 uint64_t Count = AF.getFileSize() / AF.getValueSize();
1059 // FIXME: This error shouldn't actually occur (the front end should emit
1060 // multiple .align directives to enforce the semantics it wants), but is
1061 // severe enough that we want to report it. How to handle this?
1062 if (Count * AF.getValueSize() != AF.getFileSize())
1063 llvm_report_error("undefined .align directive, value size '" +
1064 Twine(AF.getValueSize()) +
1065 "' is not a divisor of padding size '" +
1066 Twine(AF.getFileSize()) + "'");
1068 for (uint64_t i = 0; i != Count; ++i) {
1069 switch (AF.getValueSize()) {
1071 assert(0 && "Invalid size!");
1072 case 1: MOW.Write8 (uint8_t (AF.getValue())); break;
1073 case 2: MOW.Write16(uint16_t(AF.getValue())); break;
1074 case 4: MOW.Write32(uint32_t(AF.getValue())); break;
1075 case 8: MOW.Write64(uint64_t(AF.getValue())); break;
1081 case MCFragment::FT_Data: {
1082 MCDataFragment &DF = cast<MCDataFragment>(F);
1084 // Apply the fixups.
1086 // FIXME: Move elsewhere.
1087 for (MCDataFragment::const_fixup_iterator it = DF.fixup_begin(),
1088 ie = DF.fixup_end(); it != ie; ++it)
1089 MOW.ApplyFixup(*it, DF);
1091 OS << cast<MCDataFragment>(F).getContents().str();
1095 case MCFragment::FT_Fill: {
1096 MCFillFragment &FF = cast<MCFillFragment>(F);
1101 if (!FF.getValue().EvaluateAsRelocatable(Target))
1102 llvm_report_error("expected relocatable expression");
1104 if (Target.isAbsolute())
1105 Value = Target.getConstant();
1106 for (uint64_t i = 0, e = FF.getCount(); i != e; ++i) {
1107 if (!Target.isAbsolute()) {
1110 // FIXME: Find a better way to write in the fixes (move to
1112 const MCAsmFixup *Fixup = FF.LookupFixup(i * FF.getValueSize());
1113 assert(Fixup && "Missing fixup for fill value!");
1114 Value = Fixup->FixedValue;
1117 switch (FF.getValueSize()) {
1119 assert(0 && "Invalid size!");
1120 case 1: MOW.Write8 (uint8_t (Value)); break;
1121 case 2: MOW.Write16(uint16_t(Value)); break;
1122 case 4: MOW.Write32(uint32_t(Value)); break;
1123 case 8: MOW.Write64(uint64_t(Value)); break;
1129 case MCFragment::FT_Org: {
1130 MCOrgFragment &OF = cast<MCOrgFragment>(F);
1132 for (uint64_t i = 0, e = OF.getFileSize(); i != e; ++i)
1133 MOW.Write8(uint8_t(OF.getValue()));
1138 case MCFragment::FT_ZeroFill: {
1139 assert(0 && "Invalid zero fill fragment in concrete section!");
1144 assert(OS.tell() - Start == F.getFileSize());
1147 /// WriteFileData - Write the \arg SD data to the output file.
1148 static void WriteFileData(raw_ostream &OS, const MCSectionData &SD,
1149 MachObjectWriter &MOW) {
1150 // Ignore virtual sections.
1151 if (isVirtualSection(SD.getSection())) {
1152 assert(SD.getFileSize() == 0);
1156 uint64_t Start = OS.tell();
1159 for (MCSectionData::const_iterator it = SD.begin(),
1160 ie = SD.end(); it != ie; ++it)
1161 WriteFileData(OS, *it, MOW);
1163 // Add section padding.
1164 assert(SD.getFileSize() >= SD.getSize() && "Invalid section sizes!");
1165 MOW.WriteZeros(SD.getFileSize() - SD.getSize());
1167 assert(OS.tell() - Start == SD.getFileSize());
1170 void MCAssembler::Finish() {
1171 DEBUG_WITH_TYPE("mc-dump", {
1172 llvm::errs() << "assembler backend - pre-layout\n--\n";
1175 // Layout the concrete sections and fragments.
1176 uint64_t Address = 0;
1177 MCSectionData *Prev = 0;
1178 for (iterator it = begin(), ie = end(); it != ie; ++it) {
1179 MCSectionData &SD = *it;
1181 // Skip virtual sections.
1182 if (isVirtualSection(SD.getSection()))
1185 // Align this section if necessary by adding padding bytes to the previous
1187 if (uint64_t Pad = OffsetToAlignment(Address, it->getAlignment())) {
1188 assert(Prev && "Missing prev section!");
1189 Prev->setFileSize(Prev->getFileSize() + Pad);
1193 // Layout the section fragments and its size.
1194 SD.setAddress(Address);
1196 Address += SD.getFileSize();
1201 // Layout the virtual sections.
1202 for (iterator it = begin(), ie = end(); it != ie; ++it) {
1203 MCSectionData &SD = *it;
1205 if (!isVirtualSection(SD.getSection()))
1208 SD.setAddress(Address);
1210 Address += SD.getSize();
1213 DEBUG_WITH_TYPE("mc-dump", {
1214 llvm::errs() << "assembler backend - post-layout\n--\n";
1217 // Write the object file.
1218 MachObjectWriter MOW(OS);
1219 MOW.WriteObject(*this);
1225 // Debugging methods
1229 raw_ostream &operator<<(raw_ostream &OS, const MCAsmFixup &AF) {
1230 OS << "<MCAsmFixup" << " Offset:" << AF.Offset << " Value:" << AF.Value
1231 << " Size:" << AF.Size << ">";
1237 void MCFragment::dump() {
1238 raw_ostream &OS = llvm::errs();
1240 OS << "<MCFragment " << (void*) this << " Offset:" << Offset
1241 << " FileSize:" << FileSize;
1243 if (!Fixups.empty()) {
1246 for (fixup_iterator it = fixup_begin(), ie = fixup_end(); it != ie; ++it) {
1247 if (it != fixup_begin()) OS << ",\n ";
1256 void MCAlignFragment::dump() {
1257 raw_ostream &OS = llvm::errs();
1259 OS << "<MCAlignFragment ";
1260 this->MCFragment::dump();
1262 OS << " Alignment:" << getAlignment()
1263 << " Value:" << getValue() << " ValueSize:" << getValueSize()
1264 << " MaxBytesToEmit:" << getMaxBytesToEmit() << ">";
1267 void MCDataFragment::dump() {
1268 raw_ostream &OS = llvm::errs();
1270 OS << "<MCDataFragment ";
1271 this->MCFragment::dump();
1273 OS << " Contents:[";
1274 for (unsigned i = 0, e = getContents().size(); i != e; ++i) {
1276 OS << hexdigit((Contents[i] >> 4) & 0xF) << hexdigit(Contents[i] & 0xF);
1281 void MCFillFragment::dump() {
1282 raw_ostream &OS = llvm::errs();
1284 OS << "<MCFillFragment ";
1285 this->MCFragment::dump();
1287 OS << " Value:" << getValue() << " ValueSize:" << getValueSize()
1288 << " Count:" << getCount() << ">";
1291 void MCOrgFragment::dump() {
1292 raw_ostream &OS = llvm::errs();
1294 OS << "<MCOrgFragment ";
1295 this->MCFragment::dump();
1297 OS << " Offset:" << getOffset() << " Value:" << getValue() << ">";
1300 void MCZeroFillFragment::dump() {
1301 raw_ostream &OS = llvm::errs();
1303 OS << "<MCZeroFillFragment ";
1304 this->MCFragment::dump();
1306 OS << " Size:" << getSize() << " Alignment:" << getAlignment() << ">";
1309 void MCSectionData::dump() {
1310 raw_ostream &OS = llvm::errs();
1312 OS << "<MCSectionData";
1313 OS << " Alignment:" << getAlignment() << " Address:" << Address
1314 << " Size:" << Size << " FileSize:" << FileSize
1316 for (iterator it = begin(), ie = end(); it != ie; ++it) {
1317 if (it != begin()) OS << ",\n ";
1323 void MCSymbolData::dump() {
1324 raw_ostream &OS = llvm::errs();
1326 OS << "<MCSymbolData Symbol:" << getSymbol()
1327 << " Fragment:" << getFragment() << " Offset:" << getOffset()
1328 << " Flags:" << getFlags() << " Index:" << getIndex();
1330 OS << " (common, size:" << getCommonSize()
1331 << " align: " << getCommonAlignment() << ")";
1333 OS << " (external)";
1334 if (isPrivateExtern())
1335 OS << " (private extern)";
1339 void MCAssembler::dump() {
1340 raw_ostream &OS = llvm::errs();
1342 OS << "<MCAssembler\n";
1343 OS << " Sections:[";
1344 for (iterator it = begin(), ie = end(); it != ie; ++it) {
1345 if (it != begin()) OS << ",\n ";
1351 for (symbol_iterator it = symbol_begin(), ie = symbol_end(); it != ie; ++it) {
1352 if (it != symbol_begin()) OS << ",\n ";