1 //===-- MachODump.cpp - Object file dumping utility for llvm --------------===//
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 // This file implements the MachO-specific dumper for llvm-objdump.
12 //===----------------------------------------------------------------------===//
14 #include "llvm-objdump.h"
15 #include "llvm/ADT/STLExtras.h"
16 #include "llvm/ADT/StringExtras.h"
17 #include "llvm/ADT/Triple.h"
18 #include "llvm/DebugInfo/DIContext.h"
19 #include "llvm/MC/MCAsmInfo.h"
20 #include "llvm/MC/MCContext.h"
21 #include "llvm/MC/MCDisassembler.h"
22 #include "llvm/MC/MCInst.h"
23 #include "llvm/MC/MCInstPrinter.h"
24 #include "llvm/MC/MCInstrAnalysis.h"
25 #include "llvm/MC/MCInstrDesc.h"
26 #include "llvm/MC/MCInstrInfo.h"
27 #include "llvm/MC/MCRegisterInfo.h"
28 #include "llvm/MC/MCSubtargetInfo.h"
29 #include "llvm/Object/MachO.h"
30 #include "llvm/Support/Casting.h"
31 #include "llvm/Support/CommandLine.h"
32 #include "llvm/Support/Debug.h"
33 #include "llvm/Support/Endian.h"
34 #include "llvm/Support/Format.h"
35 #include "llvm/Support/GraphWriter.h"
36 #include "llvm/Support/MachO.h"
37 #include "llvm/Support/MemoryBuffer.h"
38 #include "llvm/Support/TargetRegistry.h"
39 #include "llvm/Support/TargetSelect.h"
40 #include "llvm/Support/raw_ostream.h"
43 #include <system_error>
45 using namespace object;
48 UseDbg("g", cl::desc("Print line information from debug info if available"));
50 static cl::opt<std::string>
51 DSYMFile("dsym", cl::desc("Use .dSYM file for debug info"));
53 static std::string ThumbTripleName;
55 static const Target *GetTarget(const MachOObjectFile *MachOObj,
56 const char **McpuDefault,
57 const Target **ThumbTarget) {
58 // Figure out the target triple.
59 if (TripleName.empty()) {
60 llvm::Triple TT("unknown-unknown-unknown");
61 llvm::Triple ThumbTriple = Triple();
62 TT = MachOObj->getArch(McpuDefault, &ThumbTriple);
63 TripleName = TT.str();
64 ThumbTripleName = ThumbTriple.str();
67 // Get the target specific parser.
69 const Target *TheTarget = TargetRegistry::lookupTarget(TripleName, Error);
70 if (TheTarget && ThumbTripleName.empty())
73 *ThumbTarget = TargetRegistry::lookupTarget(ThumbTripleName, Error);
77 errs() << "llvm-objdump: error: unable to get target for '";
81 errs() << ThumbTripleName;
82 errs() << "', see --version and --triple.\n";
87 bool operator()(const SymbolRef &A, const SymbolRef &B) {
88 SymbolRef::Type AType, BType;
92 uint64_t AAddr, BAddr;
93 if (AType != SymbolRef::ST_Function)
97 if (BType != SymbolRef::ST_Function)
101 return AAddr < BAddr;
105 // Types for the storted data in code table that is built before disassembly
106 // and the predicate function to sort them.
107 typedef std::pair<uint64_t, DiceRef> DiceTableEntry;
108 typedef std::vector<DiceTableEntry> DiceTable;
109 typedef DiceTable::iterator dice_table_iterator;
112 compareDiceTableEntries(const DiceTableEntry i,
113 const DiceTableEntry j) {
114 return i.first == j.first;
117 static void DumpDataInCode(const char *bytes, uint64_t Size,
118 unsigned short Kind) {
122 case MachO::DICE_KIND_DATA:
125 Value = bytes[3] << 24 |
129 outs() << "\t.long " << Value;
132 Value = bytes[1] << 8 |
134 outs() << "\t.short " << Value;
138 outs() << "\t.byte " << Value;
141 outs() << "\t@ KIND_DATA\n";
143 case MachO::DICE_KIND_JUMP_TABLE8:
145 outs() << "\t.byte " << Value << "\t@ KIND_JUMP_TABLE8";
147 case MachO::DICE_KIND_JUMP_TABLE16:
148 Value = bytes[1] << 8 |
150 outs() << "\t.short " << Value << "\t@ KIND_JUMP_TABLE16";
152 case MachO::DICE_KIND_JUMP_TABLE32:
153 Value = bytes[3] << 24 |
157 outs() << "\t.long " << Value << "\t@ KIND_JUMP_TABLE32";
160 outs() << "\t@ data in code kind = " << Kind << "\n";
165 static void getSectionsAndSymbols(const MachO::mach_header Header,
166 MachOObjectFile *MachOObj,
167 std::vector<SectionRef> &Sections,
168 std::vector<SymbolRef> &Symbols,
169 SmallVectorImpl<uint64_t> &FoundFns,
170 uint64_t &BaseSegmentAddress) {
171 for (const SymbolRef &Symbol : MachOObj->symbols())
172 Symbols.push_back(Symbol);
174 for (const SectionRef &Section : MachOObj->sections()) {
176 Section.getName(SectName);
177 Sections.push_back(Section);
180 MachOObjectFile::LoadCommandInfo Command =
181 MachOObj->getFirstLoadCommandInfo();
182 bool BaseSegmentAddressSet = false;
183 for (unsigned i = 0; ; ++i) {
184 if (Command.C.cmd == MachO::LC_FUNCTION_STARTS) {
185 // We found a function starts segment, parse the addresses for later
187 MachO::linkedit_data_command LLC =
188 MachOObj->getLinkeditDataLoadCommand(Command);
190 MachOObj->ReadULEB128s(LLC.dataoff, FoundFns);
192 else if (Command.C.cmd == MachO::LC_SEGMENT) {
193 MachO::segment_command SLC =
194 MachOObj->getSegmentLoadCommand(Command);
195 StringRef SegName = SLC.segname;
196 if(!BaseSegmentAddressSet && SegName != "__PAGEZERO") {
197 BaseSegmentAddressSet = true;
198 BaseSegmentAddress = SLC.vmaddr;
202 if (i == Header.ncmds - 1)
205 Command = MachOObj->getNextLoadCommandInfo(Command);
209 static void DisassembleInputMachO2(StringRef Filename,
210 MachOObjectFile *MachOOF);
212 void llvm::DisassembleInputMachO(StringRef Filename) {
213 ErrorOr<std::unique_ptr<MemoryBuffer>> BuffOrErr =
214 MemoryBuffer::getFileOrSTDIN(Filename);
215 if (std::error_code EC = BuffOrErr.getError()) {
216 errs() << "llvm-objdump: " << Filename << ": " << EC.message() << "\n";
219 std::unique_ptr<MemoryBuffer> Buff = std::move(BuffOrErr.get());
221 std::unique_ptr<MachOObjectFile> MachOOF = std::move(
222 ObjectFile::createMachOObjectFile(Buff.get()->getMemBufferRef()).get());
224 DisassembleInputMachO2(Filename, MachOOF.get());
227 static void DisassembleInputMachO2(StringRef Filename,
228 MachOObjectFile *MachOOF) {
229 const char *McpuDefault = nullptr;
230 const Target *ThumbTarget = nullptr;
231 const Target *TheTarget = GetTarget(MachOOF, &McpuDefault, &ThumbTarget);
233 // GetTarget prints out stuff.
236 if (MCPU.empty() && McpuDefault)
239 std::unique_ptr<const MCInstrInfo> InstrInfo(TheTarget->createMCInstrInfo());
240 std::unique_ptr<MCInstrAnalysis> InstrAnalysis(
241 TheTarget->createMCInstrAnalysis(InstrInfo.get()));
242 std::unique_ptr<const MCInstrInfo> ThumbInstrInfo;
243 std::unique_ptr<MCInstrAnalysis> ThumbInstrAnalysis;
245 ThumbInstrInfo.reset(ThumbTarget->createMCInstrInfo());
246 ThumbInstrAnalysis.reset(
247 ThumbTarget->createMCInstrAnalysis(ThumbInstrInfo.get()));
250 // Package up features to be passed to target/subtarget
251 std::string FeaturesStr;
253 SubtargetFeatures Features;
254 for (unsigned i = 0; i != MAttrs.size(); ++i)
255 Features.AddFeature(MAttrs[i]);
256 FeaturesStr = Features.getString();
259 // Set up disassembler.
260 std::unique_ptr<const MCRegisterInfo> MRI(
261 TheTarget->createMCRegInfo(TripleName));
262 std::unique_ptr<const MCAsmInfo> AsmInfo(
263 TheTarget->createMCAsmInfo(*MRI, TripleName));
264 std::unique_ptr<const MCSubtargetInfo> STI(
265 TheTarget->createMCSubtargetInfo(TripleName, MCPU, FeaturesStr));
266 MCContext Ctx(AsmInfo.get(), MRI.get(), nullptr);
267 std::unique_ptr<const MCDisassembler> DisAsm(
268 TheTarget->createMCDisassembler(*STI, Ctx));
269 int AsmPrinterVariant = AsmInfo->getAssemblerDialect();
270 std::unique_ptr<MCInstPrinter> IP(TheTarget->createMCInstPrinter(
271 AsmPrinterVariant, *AsmInfo, *InstrInfo, *MRI, *STI));
273 if (!InstrAnalysis || !AsmInfo || !STI || !DisAsm || !IP) {
274 errs() << "error: couldn't initialize disassembler for target "
275 << TripleName << '\n';
279 // Set up thumb disassembler.
280 std::unique_ptr<const MCRegisterInfo> ThumbMRI;
281 std::unique_ptr<const MCAsmInfo> ThumbAsmInfo;
282 std::unique_ptr<const MCSubtargetInfo> ThumbSTI;
283 std::unique_ptr<const MCDisassembler> ThumbDisAsm;
284 std::unique_ptr<MCInstPrinter> ThumbIP;
285 std::unique_ptr<MCContext> ThumbCtx;
287 ThumbMRI.reset(ThumbTarget->createMCRegInfo(ThumbTripleName));
289 ThumbTarget->createMCAsmInfo(*ThumbMRI, ThumbTripleName));
291 ThumbTarget->createMCSubtargetInfo(ThumbTripleName, MCPU, FeaturesStr));
292 ThumbCtx.reset(new MCContext(ThumbAsmInfo.get(), ThumbMRI.get(), nullptr));
293 ThumbDisAsm.reset(ThumbTarget->createMCDisassembler(*ThumbSTI, *ThumbCtx));
294 int ThumbAsmPrinterVariant = ThumbAsmInfo->getAssemblerDialect();
295 ThumbIP.reset(ThumbTarget->createMCInstPrinter(
296 ThumbAsmPrinterVariant, *ThumbAsmInfo, *ThumbInstrInfo, *ThumbMRI,
300 if (ThumbTarget && (!ThumbInstrAnalysis || !ThumbAsmInfo || !ThumbSTI ||
301 !ThumbDisAsm || !ThumbIP)) {
302 errs() << "error: couldn't initialize disassembler for target "
303 << ThumbTripleName << '\n';
307 outs() << '\n' << Filename << ":\n\n";
309 MachO::mach_header Header = MachOOF->getHeader();
311 // FIXME: FoundFns isn't used anymore. Using symbols/LC_FUNCTION_STARTS to
312 // determine function locations will eventually go in MCObjectDisassembler.
313 // FIXME: Using the -cfg command line option, this code used to be able to
314 // annotate relocations with the referenced symbol's name, and if this was
315 // inside a __[cf]string section, the data it points to. This is now replaced
316 // by the upcoming MCSymbolizer, which needs the appropriate setup done above.
317 std::vector<SectionRef> Sections;
318 std::vector<SymbolRef> Symbols;
319 SmallVector<uint64_t, 8> FoundFns;
320 uint64_t BaseSegmentAddress;
322 getSectionsAndSymbols(Header, MachOOF, Sections, Symbols, FoundFns,
325 // Sort the symbols by address, just in case they didn't come in that way.
326 std::sort(Symbols.begin(), Symbols.end(), SymbolSorter());
328 // Build a data in code table that is sorted on by the address of each entry.
329 uint64_t BaseAddress = 0;
330 if (Header.filetype == MachO::MH_OBJECT)
331 Sections[0].getAddress(BaseAddress);
333 BaseAddress = BaseSegmentAddress;
335 for (dice_iterator DI = MachOOF->begin_dices(), DE = MachOOF->end_dices();
338 DI->getOffset(Offset);
339 Dices.push_back(std::make_pair(BaseAddress + Offset, *DI));
341 array_pod_sort(Dices.begin(), Dices.end());
344 raw_ostream &DebugOut = DebugFlag ? dbgs() : nulls();
346 raw_ostream &DebugOut = nulls();
349 std::unique_ptr<DIContext> diContext;
350 ObjectFile *DbgObj = MachOOF;
351 // Try to find debug info and set up the DIContext for it.
353 // A separate DSym file path was specified, parse it as a macho file,
354 // get the sections and supply it to the section name parsing machinery.
355 if (!DSYMFile.empty()) {
356 ErrorOr<std::unique_ptr<MemoryBuffer>> BufOrErr =
357 MemoryBuffer::getFileOrSTDIN(DSYMFile);
358 if (std::error_code EC = BufOrErr.getError()) {
359 errs() << "llvm-objdump: " << Filename << ": " << EC.message() << '\n';
363 ObjectFile::createMachOObjectFile(BufOrErr.get()->getMemBufferRef())
368 // Setup the DIContext
369 diContext.reset(DIContext::getDWARFContext(*DbgObj));
372 for (unsigned SectIdx = 0; SectIdx != Sections.size(); SectIdx++) {
374 bool SectIsText = false;
375 Sections[SectIdx].isText(SectIsText);
376 if (SectIsText == false)
380 if (Sections[SectIdx].getName(SectName) ||
381 SectName != "__text")
382 continue; // Skip non-text sections
384 DataRefImpl DR = Sections[SectIdx].getRawDataRefImpl();
386 StringRef SegmentName = MachOOF->getSectionFinalSegmentName(DR);
387 if (SegmentName != "__TEXT")
391 Sections[SectIdx].getContents(Bytes);
392 StringRefMemoryObject memoryObject(Bytes);
393 bool symbolTableWorked = false;
395 // Parse relocations.
396 std::vector<std::pair<uint64_t, SymbolRef>> Relocs;
397 for (const RelocationRef &Reloc : Sections[SectIdx].relocations()) {
398 uint64_t RelocOffset, SectionAddress;
399 Reloc.getOffset(RelocOffset);
400 Sections[SectIdx].getAddress(SectionAddress);
401 RelocOffset -= SectionAddress;
403 symbol_iterator RelocSym = Reloc.getSymbol();
405 Relocs.push_back(std::make_pair(RelocOffset, *RelocSym));
407 array_pod_sort(Relocs.begin(), Relocs.end());
409 // Disassemble symbol by symbol.
410 for (unsigned SymIdx = 0; SymIdx != Symbols.size(); SymIdx++) {
412 Symbols[SymIdx].getName(SymName);
415 Symbols[SymIdx].getType(ST);
416 if (ST != SymbolRef::ST_Function)
419 // Make sure the symbol is defined in this section.
420 bool containsSym = false;
421 Sections[SectIdx].containsSymbol(Symbols[SymIdx], containsSym);
425 // Start at the address of the symbol relative to the section's address.
426 uint64_t SectionAddress = 0;
428 Sections[SectIdx].getAddress(SectionAddress);
429 Symbols[SymIdx].getAddress(Start);
430 Start -= SectionAddress;
432 // Stop disassembling either at the beginning of the next symbol or at
433 // the end of the section.
434 bool containsNextSym = false;
435 uint64_t NextSym = 0;
436 uint64_t NextSymIdx = SymIdx+1;
437 while (Symbols.size() > NextSymIdx) {
438 SymbolRef::Type NextSymType;
439 Symbols[NextSymIdx].getType(NextSymType);
440 if (NextSymType == SymbolRef::ST_Function) {
441 Sections[SectIdx].containsSymbol(Symbols[NextSymIdx],
443 Symbols[NextSymIdx].getAddress(NextSym);
444 NextSym -= SectionAddress;
451 Sections[SectIdx].getSize(SectSize);
452 uint64_t End = containsNextSym ? NextSym : SectSize;
455 symbolTableWorked = true;
457 DataRefImpl Symb = Symbols[SymIdx].getRawDataRefImpl();
459 (MachOOF->getSymbolFlags(Symb) & SymbolRef::SF_Thumb) && ThumbTarget;
461 outs() << SymName << ":\n";
463 for (uint64_t Index = Start; Index < End; Index += Size) {
466 uint64_t SectAddress = 0;
467 Sections[SectIdx].getAddress(SectAddress);
468 outs() << format("%8" PRIx64 ":\t", SectAddress + Index);
470 // Check the data in code table here to see if this is data not an
471 // instruction to be disassembled.
473 Dice.push_back(std::make_pair(SectAddress + Index, DiceRef()));
474 dice_table_iterator DTI = std::search(Dices.begin(), Dices.end(),
475 Dice.begin(), Dice.end(),
476 compareDiceTableEntries);
477 if (DTI != Dices.end()){
479 DTI->second.getLength(Length);
480 DumpBytes(StringRef(Bytes.data() + Index, Length));
482 DTI->second.getKind(Kind);
483 DumpDataInCode(Bytes.data() + Index, Length, Kind);
489 gotInst = ThumbDisAsm->getInstruction(Inst, Size, memoryObject, Index,
492 gotInst = DisAsm->getInstruction(Inst, Size, memoryObject, Index,
495 DumpBytes(StringRef(Bytes.data() + Index, Size));
497 ThumbIP->printInst(&Inst, outs(), "");
499 IP->printInst(&Inst, outs(), "");
504 diContext->getLineInfoForAddress(SectAddress + Index);
505 // Print valid line info if it changed.
506 if (dli != lastLine && dli.Line != 0)
507 outs() << "\t## " << dli.FileName << ':' << dli.Line << ':'
513 errs() << "llvm-objdump: warning: invalid instruction encoding\n";
515 Size = 1; // skip illegible bytes
519 if (!symbolTableWorked) {
520 // Reading the symbol table didn't work, disassemble the whole section.
521 uint64_t SectAddress;
522 Sections[SectIdx].getAddress(SectAddress);
524 Sections[SectIdx].getSize(SectSize);
526 for (uint64_t Index = 0; Index < SectSize; Index += InstSize) {
529 if (DisAsm->getInstruction(Inst, InstSize, memoryObject, Index,
530 DebugOut, nulls())) {
531 outs() << format("%8" PRIx64 ":\t", SectAddress + Index);
532 DumpBytes(StringRef(Bytes.data() + Index, InstSize));
533 IP->printInst(&Inst, outs(), "");
536 errs() << "llvm-objdump: warning: invalid instruction encoding\n";
538 InstSize = 1; // skip illegible bytes
546 //===----------------------------------------------------------------------===//
547 // __compact_unwind section dumping
548 //===----------------------------------------------------------------------===//
552 template <typename T> static uint64_t readNext(const char *&Buf) {
553 using llvm::support::little;
554 using llvm::support::unaligned;
556 uint64_t Val = support::endian::read<T, little, unaligned>(Buf);
561 struct CompactUnwindEntry {
562 uint32_t OffsetInSection;
564 uint64_t FunctionAddr;
566 uint32_t CompactEncoding;
567 uint64_t PersonalityAddr;
570 RelocationRef FunctionReloc;
571 RelocationRef PersonalityReloc;
572 RelocationRef LSDAReloc;
574 CompactUnwindEntry(StringRef Contents, unsigned Offset, bool Is64)
575 : OffsetInSection(Offset) {
577 read<uint64_t>(Contents.data() + Offset);
579 read<uint32_t>(Contents.data() + Offset);
583 template<typename UIntPtr>
584 void read(const char *Buf) {
585 FunctionAddr = readNext<UIntPtr>(Buf);
586 Length = readNext<uint32_t>(Buf);
587 CompactEncoding = readNext<uint32_t>(Buf);
588 PersonalityAddr = readNext<UIntPtr>(Buf);
589 LSDAAddr = readNext<UIntPtr>(Buf);
594 /// Given a relocation from __compact_unwind, consisting of the RelocationRef
595 /// and data being relocated, determine the best base Name and Addend to use for
596 /// display purposes.
598 /// 1. An Extern relocation will directly reference a symbol (and the data is
599 /// then already an addend), so use that.
600 /// 2. Otherwise the data is an offset in the object file's layout; try to find
601 // a symbol before it in the same section, and use the offset from there.
602 /// 3. Finally, if all that fails, fall back to an offset from the start of the
603 /// referenced section.
604 static void findUnwindRelocNameAddend(const MachOObjectFile *Obj,
605 std::map<uint64_t, SymbolRef> &Symbols,
606 const RelocationRef &Reloc,
608 StringRef &Name, uint64_t &Addend) {
609 if (Reloc.getSymbol() != Obj->symbol_end()) {
610 Reloc.getSymbol()->getName(Name);
615 auto RE = Obj->getRelocation(Reloc.getRawDataRefImpl());
616 SectionRef RelocSection = Obj->getRelocationSection(RE);
618 uint64_t SectionAddr;
619 RelocSection.getAddress(SectionAddr);
621 auto Sym = Symbols.upper_bound(Addr);
622 if (Sym == Symbols.begin()) {
623 // The first symbol in the object is after this reference, the best we can
624 // do is section-relative notation.
625 RelocSection.getName(Name);
626 Addend = Addr - SectionAddr;
630 // Go back one so that SymbolAddress <= Addr.
633 section_iterator SymSection = Obj->section_end();
634 Sym->second.getSection(SymSection);
635 if (RelocSection == *SymSection) {
636 // There's a valid symbol in the same section before this reference.
637 Sym->second.getName(Name);
638 Addend = Addr - Sym->first;
642 // There is a symbol before this reference, but it's in a different
643 // section. Probably not helpful to mention it, so use the section name.
644 RelocSection.getName(Name);
645 Addend = Addr - SectionAddr;
648 static void printUnwindRelocDest(const MachOObjectFile *Obj,
649 std::map<uint64_t, SymbolRef> &Symbols,
650 const RelocationRef &Reloc,
655 findUnwindRelocNameAddend(Obj, Symbols, Reloc, Addr, Name, Addend);
659 outs() << " + " << format("0x%" PRIx64, Addend);
663 printMachOCompactUnwindSection(const MachOObjectFile *Obj,
664 std::map<uint64_t, SymbolRef> &Symbols,
665 const SectionRef &CompactUnwind) {
667 assert(Obj->isLittleEndian() &&
668 "There should not be a big-endian .o with __compact_unwind");
670 bool Is64 = Obj->is64Bit();
671 uint32_t PointerSize = Is64 ? sizeof(uint64_t) : sizeof(uint32_t);
672 uint32_t EntrySize = 3 * PointerSize + 2 * sizeof(uint32_t);
675 CompactUnwind.getContents(Contents);
677 SmallVector<CompactUnwindEntry, 4> CompactUnwinds;
679 // First populate the initial raw offsets, encodings and so on from the entry.
680 for (unsigned Offset = 0; Offset < Contents.size(); Offset += EntrySize) {
681 CompactUnwindEntry Entry(Contents.data(), Offset, Is64);
682 CompactUnwinds.push_back(Entry);
685 // Next we need to look at the relocations to find out what objects are
686 // actually being referred to.
687 for (const RelocationRef &Reloc : CompactUnwind.relocations()) {
688 uint64_t RelocAddress;
689 Reloc.getOffset(RelocAddress);
691 uint32_t EntryIdx = RelocAddress / EntrySize;
692 uint32_t OffsetInEntry = RelocAddress - EntryIdx * EntrySize;
693 CompactUnwindEntry &Entry = CompactUnwinds[EntryIdx];
695 if (OffsetInEntry == 0)
696 Entry.FunctionReloc = Reloc;
697 else if (OffsetInEntry == PointerSize + 2 * sizeof(uint32_t))
698 Entry.PersonalityReloc = Reloc;
699 else if (OffsetInEntry == 2 * PointerSize + 2 * sizeof(uint32_t))
700 Entry.LSDAReloc = Reloc;
702 llvm_unreachable("Unexpected relocation in __compact_unwind section");
705 // Finally, we're ready to print the data we've gathered.
706 outs() << "Contents of __compact_unwind section:\n";
707 for (auto &Entry : CompactUnwinds) {
708 outs() << " Entry at offset "
709 << format("0x%" PRIx32, Entry.OffsetInSection) << ":\n";
711 // 1. Start of the region this entry applies to.
713 << format("0x%" PRIx64, Entry.FunctionAddr) << ' ';
714 printUnwindRelocDest(Obj, Symbols, Entry.FunctionReloc,
718 // 2. Length of the region this entry applies to.
719 outs() << " length: "
720 << format("0x%" PRIx32, Entry.Length) << '\n';
721 // 3. The 32-bit compact encoding.
722 outs() << " compact encoding: "
723 << format("0x%08" PRIx32, Entry.CompactEncoding) << '\n';
725 // 4. The personality function, if present.
726 if (Entry.PersonalityReloc.getObjectFile()) {
727 outs() << " personality function: "
728 << format("0x%" PRIx64, Entry.PersonalityAddr) << ' ';
729 printUnwindRelocDest(Obj, Symbols, Entry.PersonalityReloc,
730 Entry.PersonalityAddr);
734 // 5. This entry's language-specific data area.
735 if (Entry.LSDAReloc.getObjectFile()) {
737 << format("0x%" PRIx64, Entry.LSDAAddr) << ' ';
738 printUnwindRelocDest(Obj, Symbols, Entry.LSDAReloc, Entry.LSDAAddr);
744 //===----------------------------------------------------------------------===//
745 // __unwind_info section dumping
746 //===----------------------------------------------------------------------===//
748 static void printRegularSecondLevelUnwindPage(const char *PageStart) {
749 const char *Pos = PageStart;
750 uint32_t Kind = readNext<uint32_t>(Pos);
752 assert(Kind == 2 && "kind for a regular 2nd level index should be 2");
754 uint16_t EntriesStart = readNext<uint16_t>(Pos);
755 uint16_t NumEntries = readNext<uint16_t>(Pos);
757 Pos = PageStart + EntriesStart;
758 for (unsigned i = 0; i < NumEntries; ++i) {
759 uint32_t FunctionOffset = readNext<uint32_t>(Pos);
760 uint32_t Encoding = readNext<uint32_t>(Pos);
762 outs() << " [" << i << "]: "
763 << "function offset="
764 << format("0x%08" PRIx32, FunctionOffset) << ", "
766 << format("0x%08" PRIx32, Encoding)
771 static void printCompressedSecondLevelUnwindPage(
772 const char *PageStart, uint32_t FunctionBase,
773 const SmallVectorImpl<uint32_t> &CommonEncodings) {
774 const char *Pos = PageStart;
775 uint32_t Kind = readNext<uint32_t>(Pos);
777 assert(Kind == 3 && "kind for a compressed 2nd level index should be 3");
779 uint16_t EntriesStart = readNext<uint16_t>(Pos);
780 uint16_t NumEntries = readNext<uint16_t>(Pos);
782 uint16_t EncodingsStart = readNext<uint16_t>(Pos);
783 readNext<uint16_t>(Pos);
784 const auto *PageEncodings = reinterpret_cast<const support::ulittle32_t *>(
785 PageStart + EncodingsStart);
787 Pos = PageStart + EntriesStart;
788 for (unsigned i = 0; i < NumEntries; ++i) {
789 uint32_t Entry = readNext<uint32_t>(Pos);
790 uint32_t FunctionOffset = FunctionBase + (Entry & 0xffffff);
791 uint32_t EncodingIdx = Entry >> 24;
794 if (EncodingIdx < CommonEncodings.size())
795 Encoding = CommonEncodings[EncodingIdx];
797 Encoding = PageEncodings[EncodingIdx - CommonEncodings.size()];
799 outs() << " [" << i << "]: "
800 << "function offset="
801 << format("0x%08" PRIx32, FunctionOffset) << ", "
802 << "encoding[" << EncodingIdx << "]="
803 << format("0x%08" PRIx32, Encoding)
809 printMachOUnwindInfoSection(const MachOObjectFile *Obj,
810 std::map<uint64_t, SymbolRef> &Symbols,
811 const SectionRef &UnwindInfo) {
813 assert(Obj->isLittleEndian() &&
814 "There should not be a big-endian .o with __unwind_info");
816 outs() << "Contents of __unwind_info section:\n";
819 UnwindInfo.getContents(Contents);
820 const char *Pos = Contents.data();
822 //===----------------------------------
824 //===----------------------------------
826 uint32_t Version = readNext<uint32_t>(Pos);
827 outs() << " Version: "
828 << format("0x%" PRIx32, Version) << '\n';
829 assert(Version == 1 && "only understand version 1");
831 uint32_t CommonEncodingsStart = readNext<uint32_t>(Pos);
832 outs() << " Common encodings array section offset: "
833 << format("0x%" PRIx32, CommonEncodingsStart) << '\n';
834 uint32_t NumCommonEncodings = readNext<uint32_t>(Pos);
835 outs() << " Number of common encodings in array: "
836 << format("0x%" PRIx32, NumCommonEncodings) << '\n';
838 uint32_t PersonalitiesStart = readNext<uint32_t>(Pos);
839 outs() << " Personality function array section offset: "
840 << format("0x%" PRIx32, PersonalitiesStart) << '\n';
841 uint32_t NumPersonalities = readNext<uint32_t>(Pos);
842 outs() << " Number of personality functions in array: "
843 << format("0x%" PRIx32, NumPersonalities) << '\n';
845 uint32_t IndicesStart = readNext<uint32_t>(Pos);
846 outs() << " Index array section offset: "
847 << format("0x%" PRIx32, IndicesStart) << '\n';
848 uint32_t NumIndices = readNext<uint32_t>(Pos);
849 outs() << " Number of indices in array: "
850 << format("0x%" PRIx32, NumIndices) << '\n';
852 //===----------------------------------
853 // A shared list of common encodings
854 //===----------------------------------
856 // These occupy indices in the range [0, N] whenever an encoding is referenced
857 // from a compressed 2nd level index table. In practice the linker only
858 // creates ~128 of these, so that indices are available to embed encodings in
859 // the 2nd level index.
861 SmallVector<uint32_t, 64> CommonEncodings;
862 outs() << " Common encodings: (count = " << NumCommonEncodings << ")\n";
863 Pos = Contents.data() + CommonEncodingsStart;
864 for (unsigned i = 0; i < NumCommonEncodings; ++i) {
865 uint32_t Encoding = readNext<uint32_t>(Pos);
866 CommonEncodings.push_back(Encoding);
868 outs() << " encoding[" << i << "]: " << format("0x%08" PRIx32, Encoding)
873 //===----------------------------------
874 // Personality functions used in this executable
875 //===----------------------------------
877 // There should be only a handful of these (one per source language,
878 // roughly). Particularly since they only get 2 bits in the compact encoding.
880 outs() << " Personality functions: (count = " << NumPersonalities << ")\n";
881 Pos = Contents.data() + PersonalitiesStart;
882 for (unsigned i = 0; i < NumPersonalities; ++i) {
883 uint32_t PersonalityFn = readNext<uint32_t>(Pos);
884 outs() << " personality[" << i + 1
885 << "]: " << format("0x%08" PRIx32, PersonalityFn) << '\n';
888 //===----------------------------------
889 // The level 1 index entries
890 //===----------------------------------
892 // These specify an approximate place to start searching for the more detailed
893 // information, sorted by PC.
896 uint32_t FunctionOffset;
897 uint32_t SecondLevelPageStart;
901 SmallVector<IndexEntry, 4> IndexEntries;
903 outs() << " Top level indices: (count = " << NumIndices << ")\n";
904 Pos = Contents.data() + IndicesStart;
905 for (unsigned i = 0; i < NumIndices; ++i) {
908 Entry.FunctionOffset = readNext<uint32_t>(Pos);
909 Entry.SecondLevelPageStart = readNext<uint32_t>(Pos);
910 Entry.LSDAStart = readNext<uint32_t>(Pos);
911 IndexEntries.push_back(Entry);
913 outs() << " [" << i << "]: "
914 << "function offset="
915 << format("0x%08" PRIx32, Entry.FunctionOffset) << ", "
916 << "2nd level page offset="
917 << format("0x%08" PRIx32, Entry.SecondLevelPageStart) << ", "
919 << format("0x%08" PRIx32, Entry.LSDAStart) << '\n';
923 //===----------------------------------
924 // Next come the LSDA tables
925 //===----------------------------------
927 // The LSDA layout is rather implicit: it's a contiguous array of entries from
928 // the first top-level index's LSDAOffset to the last (sentinel).
930 outs() << " LSDA descriptors:\n";
931 Pos = Contents.data() + IndexEntries[0].LSDAStart;
932 int NumLSDAs = (IndexEntries.back().LSDAStart - IndexEntries[0].LSDAStart) /
933 (2 * sizeof(uint32_t));
934 for (int i = 0; i < NumLSDAs; ++i) {
935 uint32_t FunctionOffset = readNext<uint32_t>(Pos);
936 uint32_t LSDAOffset = readNext<uint32_t>(Pos);
937 outs() << " [" << i << "]: "
938 << "function offset="
939 << format("0x%08" PRIx32, FunctionOffset) << ", "
941 << format("0x%08" PRIx32, LSDAOffset) << '\n';
944 //===----------------------------------
945 // Finally, the 2nd level indices
946 //===----------------------------------
948 // Generally these are 4K in size, and have 2 possible forms:
949 // + Regular stores up to 511 entries with disparate encodings
950 // + Compressed stores up to 1021 entries if few enough compact encoding
952 outs() << " Second level indices:\n";
953 for (unsigned i = 0; i < IndexEntries.size() - 1; ++i) {
954 // The final sentinel top-level index has no associated 2nd level page
955 if (IndexEntries[i].SecondLevelPageStart == 0)
958 outs() << " Second level index[" << i << "]: "
959 << "offset in section="
960 << format("0x%08" PRIx32, IndexEntries[i].SecondLevelPageStart)
962 << "base function offset="
963 << format("0x%08" PRIx32, IndexEntries[i].FunctionOffset) << '\n';
965 Pos = Contents.data() + IndexEntries[i].SecondLevelPageStart;
966 uint32_t Kind = *reinterpret_cast<const support::ulittle32_t *>(Pos);
968 printRegularSecondLevelUnwindPage(Pos);
970 printCompressedSecondLevelUnwindPage(Pos, IndexEntries[i].FunctionOffset,
973 llvm_unreachable("Do not know how to print this kind of 2nd level page");
978 void llvm::printMachOUnwindInfo(const MachOObjectFile *Obj) {
979 std::map<uint64_t, SymbolRef> Symbols;
980 for (const SymbolRef &SymRef : Obj->symbols()) {
981 // Discard any undefined or absolute symbols. They're not going to take part
982 // in the convenience lookup for unwind info and just take up resources.
983 section_iterator Section = Obj->section_end();
984 SymRef.getSection(Section);
985 if (Section == Obj->section_end())
989 SymRef.getAddress(Addr);
990 Symbols.insert(std::make_pair(Addr, SymRef));
993 for (const SectionRef &Section : Obj->sections()) {
995 Section.getName(SectName);
996 if (SectName == "__compact_unwind")
997 printMachOCompactUnwindSection(Obj, Symbols, Section);
998 else if (SectName == "__unwind_info")
999 printMachOUnwindInfoSection(Obj, Symbols, Section);
1000 else if (SectName == "__eh_frame")
1001 outs() << "llvm-objdump: warning: unhandled __eh_frame section\n";
1006 static void PrintMachHeader(uint32_t magic, uint32_t cputype,
1007 uint32_t cpusubtype, uint32_t filetype,
1008 uint32_t ncmds, uint32_t sizeofcmds, uint32_t flags,
1010 outs() << "Mach header\n";
1011 outs() << " magic cputype cpusubtype caps filetype ncmds "
1012 "sizeofcmds flags\n";
1014 if (magic == MachO::MH_MAGIC)
1015 outs() << " MH_MAGIC";
1016 else if (magic == MachO::MH_MAGIC_64)
1017 outs() << "MH_MAGIC_64";
1019 outs() << format(" 0x%08" PRIx32, magic);
1021 case MachO::CPU_TYPE_I386:
1023 switch (cpusubtype & ~MachO::CPU_SUBTYPE_MASK) {
1024 case MachO::CPU_SUBTYPE_I386_ALL:
1028 outs() << format(" %10d", cpusubtype & ~MachO::CPU_SUBTYPE_MASK);
1032 case MachO::CPU_TYPE_X86_64:
1033 outs() << " X86_64";
1034 case MachO::CPU_SUBTYPE_X86_64_ALL:
1037 case MachO::CPU_SUBTYPE_X86_64_H:
1038 outs() << " Haswell";
1040 switch (cpusubtype & ~MachO::CPU_SUBTYPE_MASK) {
1042 outs() << format(" %10d", cpusubtype & ~MachO::CPU_SUBTYPE_MASK);
1046 case MachO::CPU_TYPE_ARM:
1048 switch (cpusubtype & ~MachO::CPU_SUBTYPE_MASK) {
1049 case MachO::CPU_SUBTYPE_ARM_ALL:
1052 case MachO::CPU_SUBTYPE_ARM_V4T:
1055 case MachO::CPU_SUBTYPE_ARM_V5TEJ:
1058 case MachO::CPU_SUBTYPE_ARM_XSCALE:
1059 outs() << " XSCALE";
1061 case MachO::CPU_SUBTYPE_ARM_V6:
1064 case MachO::CPU_SUBTYPE_ARM_V6M:
1067 case MachO::CPU_SUBTYPE_ARM_V7:
1070 case MachO::CPU_SUBTYPE_ARM_V7EM:
1073 case MachO::CPU_SUBTYPE_ARM_V7K:
1076 case MachO::CPU_SUBTYPE_ARM_V7M:
1079 case MachO::CPU_SUBTYPE_ARM_V7S:
1083 outs() << format(" %10d", cpusubtype & ~MachO::CPU_SUBTYPE_MASK);
1087 case MachO::CPU_TYPE_ARM64:
1089 switch (cpusubtype & ~MachO::CPU_SUBTYPE_MASK) {
1090 case MachO::CPU_SUBTYPE_ARM64_ALL:
1094 outs() << format(" %10d", cpusubtype & ~MachO::CPU_SUBTYPE_MASK);
1098 case MachO::CPU_TYPE_POWERPC:
1100 switch (cpusubtype & ~MachO::CPU_SUBTYPE_MASK) {
1101 case MachO::CPU_SUBTYPE_POWERPC_ALL:
1105 outs() << format(" %10d", cpusubtype & ~MachO::CPU_SUBTYPE_MASK);
1109 case MachO::CPU_TYPE_POWERPC64:
1111 switch (cpusubtype & ~MachO::CPU_SUBTYPE_MASK) {
1112 case MachO::CPU_SUBTYPE_POWERPC_ALL:
1116 outs() << format(" %10d", cpusubtype & ~MachO::CPU_SUBTYPE_MASK);
1121 if ((cpusubtype & MachO::CPU_SUBTYPE_MASK) == MachO::CPU_SUBTYPE_LIB64) {
1122 outs() << " LIB64 ";
1124 outs() << format(" 0x%02" PRIx32,
1125 (cpusubtype & MachO::CPU_SUBTYPE_MASK) >> 24);
1128 case MachO::MH_OBJECT:
1129 outs() << " OBJECT";
1131 case MachO::MH_EXECUTE:
1132 outs() << " EXECUTE";
1134 case MachO::MH_FVMLIB:
1135 outs() << " FVMLIB";
1137 case MachO::MH_CORE:
1140 case MachO::MH_PRELOAD:
1141 outs() << " PRELOAD";
1143 case MachO::MH_DYLIB:
1146 case MachO::MH_DYLIB_STUB:
1147 outs() << " DYLIB_STUB";
1149 case MachO::MH_DYLINKER:
1150 outs() << " DYLINKER";
1152 case MachO::MH_BUNDLE:
1153 outs() << " BUNDLE";
1155 case MachO::MH_DSYM:
1158 case MachO::MH_KEXT_BUNDLE:
1159 outs() << " KEXTBUNDLE";
1162 outs() << format(" %10u", filetype);
1165 outs() << format(" %5u", ncmds);
1166 outs() << format(" %10u", sizeofcmds);
1168 if (f & MachO::MH_NOUNDEFS) {
1169 outs() << " NOUNDEFS";
1170 f &= ~MachO::MH_NOUNDEFS;
1172 if (f & MachO::MH_INCRLINK) {
1173 outs() << " INCRLINK";
1174 f &= ~MachO::MH_INCRLINK;
1176 if (f & MachO::MH_DYLDLINK) {
1177 outs() << " DYLDLINK";
1178 f &= ~MachO::MH_DYLDLINK;
1180 if (f & MachO::MH_BINDATLOAD) {
1181 outs() << " BINDATLOAD";
1182 f &= ~MachO::MH_BINDATLOAD;
1184 if (f & MachO::MH_PREBOUND) {
1185 outs() << " PREBOUND";
1186 f &= ~MachO::MH_PREBOUND;
1188 if (f & MachO::MH_SPLIT_SEGS) {
1189 outs() << " SPLIT_SEGS";
1190 f &= ~MachO::MH_SPLIT_SEGS;
1192 if (f & MachO::MH_LAZY_INIT) {
1193 outs() << " LAZY_INIT";
1194 f &= ~MachO::MH_LAZY_INIT;
1196 if (f & MachO::MH_TWOLEVEL) {
1197 outs() << " TWOLEVEL";
1198 f &= ~MachO::MH_TWOLEVEL;
1200 if (f & MachO::MH_FORCE_FLAT) {
1201 outs() << " FORCE_FLAT";
1202 f &= ~MachO::MH_FORCE_FLAT;
1204 if (f & MachO::MH_NOMULTIDEFS) {
1205 outs() << " NOMULTIDEFS";
1206 f &= ~MachO::MH_NOMULTIDEFS;
1208 if (f & MachO::MH_NOFIXPREBINDING) {
1209 outs() << " NOFIXPREBINDING";
1210 f &= ~MachO::MH_NOFIXPREBINDING;
1212 if (f & MachO::MH_PREBINDABLE) {
1213 outs() << " PREBINDABLE";
1214 f &= ~MachO::MH_PREBINDABLE;
1216 if (f & MachO::MH_ALLMODSBOUND) {
1217 outs() << " ALLMODSBOUND";
1218 f &= ~MachO::MH_ALLMODSBOUND;
1220 if (f & MachO::MH_SUBSECTIONS_VIA_SYMBOLS) {
1221 outs() << " SUBSECTIONS_VIA_SYMBOLS";
1222 f &= ~MachO::MH_SUBSECTIONS_VIA_SYMBOLS;
1224 if (f & MachO::MH_CANONICAL) {
1225 outs() << " CANONICAL";
1226 f &= ~MachO::MH_CANONICAL;
1228 if (f & MachO::MH_WEAK_DEFINES) {
1229 outs() << " WEAK_DEFINES";
1230 f &= ~MachO::MH_WEAK_DEFINES;
1232 if (f & MachO::MH_BINDS_TO_WEAK) {
1233 outs() << " BINDS_TO_WEAK";
1234 f &= ~MachO::MH_BINDS_TO_WEAK;
1236 if (f & MachO::MH_ALLOW_STACK_EXECUTION) {
1237 outs() << " ALLOW_STACK_EXECUTION";
1238 f &= ~MachO::MH_ALLOW_STACK_EXECUTION;
1240 if (f & MachO::MH_DEAD_STRIPPABLE_DYLIB) {
1241 outs() << " DEAD_STRIPPABLE_DYLIB";
1242 f &= ~MachO::MH_DEAD_STRIPPABLE_DYLIB;
1244 if (f & MachO::MH_PIE) {
1246 f &= ~MachO::MH_PIE;
1248 if (f & MachO::MH_NO_REEXPORTED_DYLIBS) {
1249 outs() << " NO_REEXPORTED_DYLIBS";
1250 f &= ~MachO::MH_NO_REEXPORTED_DYLIBS;
1252 if (f & MachO::MH_HAS_TLV_DESCRIPTORS) {
1253 outs() << " MH_HAS_TLV_DESCRIPTORS";
1254 f &= ~MachO::MH_HAS_TLV_DESCRIPTORS;
1256 if (f & MachO::MH_NO_HEAP_EXECUTION) {
1257 outs() << " MH_NO_HEAP_EXECUTION";
1258 f &= ~MachO::MH_NO_HEAP_EXECUTION;
1260 if (f & MachO::MH_APP_EXTENSION_SAFE) {
1261 outs() << " APP_EXTENSION_SAFE";
1262 f &= ~MachO::MH_APP_EXTENSION_SAFE;
1264 if (f != 0 || flags == 0)
1265 outs() << format(" 0x%08" PRIx32, f);
1267 outs() << format(" 0x%08" PRIx32, magic);
1268 outs() << format(" %7d", cputype);
1269 outs() << format(" %10d", cpusubtype & ~MachO::CPU_SUBTYPE_MASK);
1270 outs() << format(" 0x%02" PRIx32,
1271 (cpusubtype & MachO::CPU_SUBTYPE_MASK) >> 24);
1272 outs() << format(" %10u", filetype);
1273 outs() << format(" %5u", ncmds);
1274 outs() << format(" %10u", sizeofcmds);
1275 outs() << format(" 0x%08" PRIx32, flags);
1280 static void getAndPrintMachHeader(const MachOObjectFile *Obj, bool verbose) {
1281 if (Obj->is64Bit()) {
1282 MachO::mach_header_64 H_64;
1283 H_64 = Obj->getHeader64();
1284 PrintMachHeader(H_64.magic, H_64.cputype, H_64.cpusubtype, H_64.filetype,
1285 H_64.ncmds, H_64.sizeofcmds, H_64.flags, verbose);
1287 MachO::mach_header H;
1288 H = Obj->getHeader();
1289 PrintMachHeader(H.magic, H.cputype, H.cpusubtype, H.filetype, H.ncmds,
1290 H.sizeofcmds, H.flags, verbose);
1294 void llvm::printMachOFileHeader(const object::ObjectFile *Obj) {
1295 const MachOObjectFile *file = dyn_cast<const MachOObjectFile>(Obj);
1296 getAndPrintMachHeader(file, true);
1297 // TODO: next get and print the load commands.