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>> Buff =
214 MemoryBuffer::getFileOrSTDIN(Filename);
215 if (std::error_code EC = Buff.getError()) {
216 errs() << "llvm-objdump: " << Filename << ": " << EC.message() << "\n";
220 std::unique_ptr<MachOObjectFile> MachOOF =
221 std::move(ObjectFile::createMachOObjectFile(Buff.get()).get());
223 DisassembleInputMachO2(Filename, MachOOF.get());
226 static void DisassembleInputMachO2(StringRef Filename,
227 MachOObjectFile *MachOOF) {
228 const char *McpuDefault = nullptr;
229 const Target *ThumbTarget = nullptr;
230 const Target *TheTarget = GetTarget(MachOOF, &McpuDefault, &ThumbTarget);
232 // GetTarget prints out stuff.
235 if (MCPU.empty() && McpuDefault)
238 std::unique_ptr<const MCInstrInfo> InstrInfo(TheTarget->createMCInstrInfo());
239 std::unique_ptr<MCInstrAnalysis> InstrAnalysis(
240 TheTarget->createMCInstrAnalysis(InstrInfo.get()));
241 std::unique_ptr<const MCInstrInfo> ThumbInstrInfo;
242 std::unique_ptr<MCInstrAnalysis> ThumbInstrAnalysis;
244 ThumbInstrInfo.reset(ThumbTarget->createMCInstrInfo());
245 ThumbInstrAnalysis.reset(
246 ThumbTarget->createMCInstrAnalysis(ThumbInstrInfo.get()));
249 // Package up features to be passed to target/subtarget
250 std::string FeaturesStr;
252 SubtargetFeatures Features;
253 for (unsigned i = 0; i != MAttrs.size(); ++i)
254 Features.AddFeature(MAttrs[i]);
255 FeaturesStr = Features.getString();
258 // Set up disassembler.
259 std::unique_ptr<const MCRegisterInfo> MRI(
260 TheTarget->createMCRegInfo(TripleName));
261 std::unique_ptr<const MCAsmInfo> AsmInfo(
262 TheTarget->createMCAsmInfo(*MRI, TripleName));
263 std::unique_ptr<const MCSubtargetInfo> STI(
264 TheTarget->createMCSubtargetInfo(TripleName, MCPU, FeaturesStr));
265 MCContext Ctx(AsmInfo.get(), MRI.get(), nullptr);
266 std::unique_ptr<const MCDisassembler> DisAsm(
267 TheTarget->createMCDisassembler(*STI, Ctx));
268 int AsmPrinterVariant = AsmInfo->getAssemblerDialect();
269 std::unique_ptr<MCInstPrinter> IP(TheTarget->createMCInstPrinter(
270 AsmPrinterVariant, *AsmInfo, *InstrInfo, *MRI, *STI));
272 if (!InstrAnalysis || !AsmInfo || !STI || !DisAsm || !IP) {
273 errs() << "error: couldn't initialize disassembler for target "
274 << TripleName << '\n';
278 // Set up thumb disassembler.
279 std::unique_ptr<const MCRegisterInfo> ThumbMRI;
280 std::unique_ptr<const MCAsmInfo> ThumbAsmInfo;
281 std::unique_ptr<const MCSubtargetInfo> ThumbSTI;
282 std::unique_ptr<const MCDisassembler> ThumbDisAsm;
283 std::unique_ptr<MCInstPrinter> ThumbIP;
284 std::unique_ptr<MCContext> ThumbCtx;
286 ThumbMRI.reset(ThumbTarget->createMCRegInfo(ThumbTripleName));
288 ThumbTarget->createMCAsmInfo(*ThumbMRI, ThumbTripleName));
290 ThumbTarget->createMCSubtargetInfo(ThumbTripleName, MCPU, FeaturesStr));
291 ThumbCtx.reset(new MCContext(ThumbAsmInfo.get(), ThumbMRI.get(), nullptr));
292 ThumbDisAsm.reset(ThumbTarget->createMCDisassembler(*ThumbSTI, *ThumbCtx));
293 int ThumbAsmPrinterVariant = ThumbAsmInfo->getAssemblerDialect();
294 ThumbIP.reset(ThumbTarget->createMCInstPrinter(
295 ThumbAsmPrinterVariant, *ThumbAsmInfo, *ThumbInstrInfo, *ThumbMRI,
299 if (ThumbTarget && (!ThumbInstrAnalysis || !ThumbAsmInfo || !ThumbSTI ||
300 !ThumbDisAsm || !ThumbIP)) {
301 errs() << "error: couldn't initialize disassembler for target "
302 << ThumbTripleName << '\n';
306 outs() << '\n' << Filename << ":\n\n";
308 MachO::mach_header Header = MachOOF->getHeader();
310 // FIXME: FoundFns isn't used anymore. Using symbols/LC_FUNCTION_STARTS to
311 // determine function locations will eventually go in MCObjectDisassembler.
312 // FIXME: Using the -cfg command line option, this code used to be able to
313 // annotate relocations with the referenced symbol's name, and if this was
314 // inside a __[cf]string section, the data it points to. This is now replaced
315 // by the upcoming MCSymbolizer, which needs the appropriate setup done above.
316 std::vector<SectionRef> Sections;
317 std::vector<SymbolRef> Symbols;
318 SmallVector<uint64_t, 8> FoundFns;
319 uint64_t BaseSegmentAddress;
321 getSectionsAndSymbols(Header, MachOOF, Sections, Symbols, FoundFns,
324 // Sort the symbols by address, just in case they didn't come in that way.
325 std::sort(Symbols.begin(), Symbols.end(), SymbolSorter());
327 // Build a data in code table that is sorted on by the address of each entry.
328 uint64_t BaseAddress = 0;
329 if (Header.filetype == MachO::MH_OBJECT)
330 Sections[0].getAddress(BaseAddress);
332 BaseAddress = BaseSegmentAddress;
334 for (dice_iterator DI = MachOOF->begin_dices(), DE = MachOOF->end_dices();
337 DI->getOffset(Offset);
338 Dices.push_back(std::make_pair(BaseAddress + Offset, *DI));
340 array_pod_sort(Dices.begin(), Dices.end());
343 raw_ostream &DebugOut = DebugFlag ? dbgs() : nulls();
345 raw_ostream &DebugOut = nulls();
348 std::unique_ptr<DIContext> diContext;
349 ObjectFile *DbgObj = MachOOF;
350 // Try to find debug info and set up the DIContext for it.
352 // A separate DSym file path was specified, parse it as a macho file,
353 // get the sections and supply it to the section name parsing machinery.
354 if (!DSYMFile.empty()) {
355 ErrorOr<std::unique_ptr<MemoryBuffer>> Buf =
356 MemoryBuffer::getFileOrSTDIN(DSYMFile);
357 if (std::error_code EC = Buf.getError()) {
358 errs() << "llvm-objdump: " << Filename << ": " << EC.message() << '\n';
361 DbgObj = ObjectFile::createMachOObjectFile(Buf.get()).get().release();
364 // Setup the DIContext
365 diContext.reset(DIContext::getDWARFContext(*DbgObj));
368 for (unsigned SectIdx = 0; SectIdx != Sections.size(); SectIdx++) {
370 bool SectIsText = false;
371 Sections[SectIdx].isText(SectIsText);
372 if (SectIsText == false)
376 if (Sections[SectIdx].getName(SectName) ||
377 SectName != "__text")
378 continue; // Skip non-text sections
380 DataRefImpl DR = Sections[SectIdx].getRawDataRefImpl();
382 StringRef SegmentName = MachOOF->getSectionFinalSegmentName(DR);
383 if (SegmentName != "__TEXT")
387 Sections[SectIdx].getContents(Bytes);
388 StringRefMemoryObject memoryObject(Bytes);
389 bool symbolTableWorked = false;
391 // Parse relocations.
392 std::vector<std::pair<uint64_t, SymbolRef>> Relocs;
393 for (const RelocationRef &Reloc : Sections[SectIdx].relocations()) {
394 uint64_t RelocOffset, SectionAddress;
395 Reloc.getOffset(RelocOffset);
396 Sections[SectIdx].getAddress(SectionAddress);
397 RelocOffset -= SectionAddress;
399 symbol_iterator RelocSym = Reloc.getSymbol();
401 Relocs.push_back(std::make_pair(RelocOffset, *RelocSym));
403 array_pod_sort(Relocs.begin(), Relocs.end());
405 // Disassemble symbol by symbol.
406 for (unsigned SymIdx = 0; SymIdx != Symbols.size(); SymIdx++) {
408 Symbols[SymIdx].getName(SymName);
411 Symbols[SymIdx].getType(ST);
412 if (ST != SymbolRef::ST_Function)
415 // Make sure the symbol is defined in this section.
416 bool containsSym = false;
417 Sections[SectIdx].containsSymbol(Symbols[SymIdx], containsSym);
421 // Start at the address of the symbol relative to the section's address.
422 uint64_t SectionAddress = 0;
424 Sections[SectIdx].getAddress(SectionAddress);
425 Symbols[SymIdx].getAddress(Start);
426 Start -= SectionAddress;
428 // Stop disassembling either at the beginning of the next symbol or at
429 // the end of the section.
430 bool containsNextSym = false;
431 uint64_t NextSym = 0;
432 uint64_t NextSymIdx = SymIdx+1;
433 while (Symbols.size() > NextSymIdx) {
434 SymbolRef::Type NextSymType;
435 Symbols[NextSymIdx].getType(NextSymType);
436 if (NextSymType == SymbolRef::ST_Function) {
437 Sections[SectIdx].containsSymbol(Symbols[NextSymIdx],
439 Symbols[NextSymIdx].getAddress(NextSym);
440 NextSym -= SectionAddress;
447 Sections[SectIdx].getSize(SectSize);
448 uint64_t End = containsNextSym ? NextSym : SectSize;
451 symbolTableWorked = true;
453 DataRefImpl Symb = Symbols[SymIdx].getRawDataRefImpl();
455 (MachOOF->getSymbolFlags(Symb) & SymbolRef::SF_Thumb) && ThumbTarget;
457 outs() << SymName << ":\n";
459 for (uint64_t Index = Start; Index < End; Index += Size) {
462 uint64_t SectAddress = 0;
463 Sections[SectIdx].getAddress(SectAddress);
464 outs() << format("%8" PRIx64 ":\t", SectAddress + Index);
466 // Check the data in code table here to see if this is data not an
467 // instruction to be disassembled.
469 Dice.push_back(std::make_pair(SectAddress + Index, DiceRef()));
470 dice_table_iterator DTI = std::search(Dices.begin(), Dices.end(),
471 Dice.begin(), Dice.end(),
472 compareDiceTableEntries);
473 if (DTI != Dices.end()){
475 DTI->second.getLength(Length);
476 DumpBytes(StringRef(Bytes.data() + Index, Length));
478 DTI->second.getKind(Kind);
479 DumpDataInCode(Bytes.data() + Index, Length, Kind);
485 gotInst = ThumbDisAsm->getInstruction(Inst, Size, memoryObject, Index,
488 gotInst = DisAsm->getInstruction(Inst, Size, memoryObject, Index,
491 DumpBytes(StringRef(Bytes.data() + Index, Size));
493 ThumbIP->printInst(&Inst, outs(), "");
495 IP->printInst(&Inst, outs(), "");
500 diContext->getLineInfoForAddress(SectAddress + Index);
501 // Print valid line info if it changed.
502 if (dli != lastLine && dli.Line != 0)
503 outs() << "\t## " << dli.FileName << ':' << dli.Line << ':'
509 errs() << "llvm-objdump: warning: invalid instruction encoding\n";
511 Size = 1; // skip illegible bytes
515 if (!symbolTableWorked) {
516 // Reading the symbol table didn't work, disassemble the whole section.
517 uint64_t SectAddress;
518 Sections[SectIdx].getAddress(SectAddress);
520 Sections[SectIdx].getSize(SectSize);
522 for (uint64_t Index = 0; Index < SectSize; Index += InstSize) {
525 if (DisAsm->getInstruction(Inst, InstSize, memoryObject, Index,
526 DebugOut, nulls())) {
527 outs() << format("%8" PRIx64 ":\t", SectAddress + Index);
528 DumpBytes(StringRef(Bytes.data() + Index, InstSize));
529 IP->printInst(&Inst, outs(), "");
532 errs() << "llvm-objdump: warning: invalid instruction encoding\n";
534 InstSize = 1; // skip illegible bytes
542 //===----------------------------------------------------------------------===//
543 // __compact_unwind section dumping
544 //===----------------------------------------------------------------------===//
548 template <typename T> static uint64_t readNext(const char *&Buf) {
549 using llvm::support::little;
550 using llvm::support::unaligned;
552 uint64_t Val = support::endian::read<T, little, unaligned>(Buf);
557 struct CompactUnwindEntry {
558 uint32_t OffsetInSection;
560 uint64_t FunctionAddr;
562 uint32_t CompactEncoding;
563 uint64_t PersonalityAddr;
566 RelocationRef FunctionReloc;
567 RelocationRef PersonalityReloc;
568 RelocationRef LSDAReloc;
570 CompactUnwindEntry(StringRef Contents, unsigned Offset, bool Is64)
571 : OffsetInSection(Offset) {
573 read<uint64_t>(Contents.data() + Offset);
575 read<uint32_t>(Contents.data() + Offset);
579 template<typename UIntPtr>
580 void read(const char *Buf) {
581 FunctionAddr = readNext<UIntPtr>(Buf);
582 Length = readNext<uint32_t>(Buf);
583 CompactEncoding = readNext<uint32_t>(Buf);
584 PersonalityAddr = readNext<UIntPtr>(Buf);
585 LSDAAddr = readNext<UIntPtr>(Buf);
590 /// Given a relocation from __compact_unwind, consisting of the RelocationRef
591 /// and data being relocated, determine the best base Name and Addend to use for
592 /// display purposes.
594 /// 1. An Extern relocation will directly reference a symbol (and the data is
595 /// then already an addend), so use that.
596 /// 2. Otherwise the data is an offset in the object file's layout; try to find
597 // a symbol before it in the same section, and use the offset from there.
598 /// 3. Finally, if all that fails, fall back to an offset from the start of the
599 /// referenced section.
600 static void findUnwindRelocNameAddend(const MachOObjectFile *Obj,
601 std::map<uint64_t, SymbolRef> &Symbols,
602 const RelocationRef &Reloc,
604 StringRef &Name, uint64_t &Addend) {
605 if (Reloc.getSymbol() != Obj->symbol_end()) {
606 Reloc.getSymbol()->getName(Name);
611 auto RE = Obj->getRelocation(Reloc.getRawDataRefImpl());
612 SectionRef RelocSection = Obj->getRelocationSection(RE);
614 uint64_t SectionAddr;
615 RelocSection.getAddress(SectionAddr);
617 auto Sym = Symbols.upper_bound(Addr);
618 if (Sym == Symbols.begin()) {
619 // The first symbol in the object is after this reference, the best we can
620 // do is section-relative notation.
621 RelocSection.getName(Name);
622 Addend = Addr - SectionAddr;
626 // Go back one so that SymbolAddress <= Addr.
629 section_iterator SymSection = Obj->section_end();
630 Sym->second.getSection(SymSection);
631 if (RelocSection == *SymSection) {
632 // There's a valid symbol in the same section before this reference.
633 Sym->second.getName(Name);
634 Addend = Addr - Sym->first;
638 // There is a symbol before this reference, but it's in a different
639 // section. Probably not helpful to mention it, so use the section name.
640 RelocSection.getName(Name);
641 Addend = Addr - SectionAddr;
644 static void printUnwindRelocDest(const MachOObjectFile *Obj,
645 std::map<uint64_t, SymbolRef> &Symbols,
646 const RelocationRef &Reloc,
651 findUnwindRelocNameAddend(Obj, Symbols, Reloc, Addr, Name, Addend);
655 outs() << " + " << format("0x%" PRIx64, Addend);
659 printMachOCompactUnwindSection(const MachOObjectFile *Obj,
660 std::map<uint64_t, SymbolRef> &Symbols,
661 const SectionRef &CompactUnwind) {
663 assert(Obj->isLittleEndian() &&
664 "There should not be a big-endian .o with __compact_unwind");
666 bool Is64 = Obj->is64Bit();
667 uint32_t PointerSize = Is64 ? sizeof(uint64_t) : sizeof(uint32_t);
668 uint32_t EntrySize = 3 * PointerSize + 2 * sizeof(uint32_t);
671 CompactUnwind.getContents(Contents);
673 SmallVector<CompactUnwindEntry, 4> CompactUnwinds;
675 // First populate the initial raw offsets, encodings and so on from the entry.
676 for (unsigned Offset = 0; Offset < Contents.size(); Offset += EntrySize) {
677 CompactUnwindEntry Entry(Contents.data(), Offset, Is64);
678 CompactUnwinds.push_back(Entry);
681 // Next we need to look at the relocations to find out what objects are
682 // actually being referred to.
683 for (const RelocationRef &Reloc : CompactUnwind.relocations()) {
684 uint64_t RelocAddress;
685 Reloc.getOffset(RelocAddress);
687 uint32_t EntryIdx = RelocAddress / EntrySize;
688 uint32_t OffsetInEntry = RelocAddress - EntryIdx * EntrySize;
689 CompactUnwindEntry &Entry = CompactUnwinds[EntryIdx];
691 if (OffsetInEntry == 0)
692 Entry.FunctionReloc = Reloc;
693 else if (OffsetInEntry == PointerSize + 2 * sizeof(uint32_t))
694 Entry.PersonalityReloc = Reloc;
695 else if (OffsetInEntry == 2 * PointerSize + 2 * sizeof(uint32_t))
696 Entry.LSDAReloc = Reloc;
698 llvm_unreachable("Unexpected relocation in __compact_unwind section");
701 // Finally, we're ready to print the data we've gathered.
702 outs() << "Contents of __compact_unwind section:\n";
703 for (auto &Entry : CompactUnwinds) {
704 outs() << " Entry at offset "
705 << format("0x%" PRIx32, Entry.OffsetInSection) << ":\n";
707 // 1. Start of the region this entry applies to.
709 << format("0x%" PRIx64, Entry.FunctionAddr) << ' ';
710 printUnwindRelocDest(Obj, Symbols, Entry.FunctionReloc,
714 // 2. Length of the region this entry applies to.
715 outs() << " length: "
716 << format("0x%" PRIx32, Entry.Length) << '\n';
717 // 3. The 32-bit compact encoding.
718 outs() << " compact encoding: "
719 << format("0x%08" PRIx32, Entry.CompactEncoding) << '\n';
721 // 4. The personality function, if present.
722 if (Entry.PersonalityReloc.getObjectFile()) {
723 outs() << " personality function: "
724 << format("0x%" PRIx64, Entry.PersonalityAddr) << ' ';
725 printUnwindRelocDest(Obj, Symbols, Entry.PersonalityReloc,
726 Entry.PersonalityAddr);
730 // 5. This entry's language-specific data area.
731 if (Entry.LSDAReloc.getObjectFile()) {
733 << format("0x%" PRIx64, Entry.LSDAAddr) << ' ';
734 printUnwindRelocDest(Obj, Symbols, Entry.LSDAReloc, Entry.LSDAAddr);
740 //===----------------------------------------------------------------------===//
741 // __unwind_info section dumping
742 //===----------------------------------------------------------------------===//
744 static void printRegularSecondLevelUnwindPage(const char *PageStart) {
745 const char *Pos = PageStart;
746 uint32_t Kind = readNext<uint32_t>(Pos);
748 assert(Kind == 2 && "kind for a regular 2nd level index should be 2");
750 uint16_t EntriesStart = readNext<uint16_t>(Pos);
751 uint16_t NumEntries = readNext<uint16_t>(Pos);
753 Pos = PageStart + EntriesStart;
754 for (unsigned i = 0; i < NumEntries; ++i) {
755 uint32_t FunctionOffset = readNext<uint32_t>(Pos);
756 uint32_t Encoding = readNext<uint32_t>(Pos);
758 outs() << " [" << i << "]: "
759 << "function offset="
760 << format("0x%08" PRIx32, FunctionOffset) << ", "
762 << format("0x%08" PRIx32, Encoding)
767 static void printCompressedSecondLevelUnwindPage(
768 const char *PageStart, uint32_t FunctionBase,
769 const SmallVectorImpl<uint32_t> &CommonEncodings) {
770 const char *Pos = PageStart;
771 uint32_t Kind = readNext<uint32_t>(Pos);
773 assert(Kind == 3 && "kind for a compressed 2nd level index should be 3");
775 uint16_t EntriesStart = readNext<uint16_t>(Pos);
776 uint16_t NumEntries = readNext<uint16_t>(Pos);
778 uint16_t EncodingsStart = readNext<uint16_t>(Pos);
779 readNext<uint16_t>(Pos);
780 const auto *PageEncodings = reinterpret_cast<const support::ulittle32_t *>(
781 PageStart + EncodingsStart);
783 Pos = PageStart + EntriesStart;
784 for (unsigned i = 0; i < NumEntries; ++i) {
785 uint32_t Entry = readNext<uint32_t>(Pos);
786 uint32_t FunctionOffset = FunctionBase + (Entry & 0xffffff);
787 uint32_t EncodingIdx = Entry >> 24;
790 if (EncodingIdx < CommonEncodings.size())
791 Encoding = CommonEncodings[EncodingIdx];
793 Encoding = PageEncodings[EncodingIdx - CommonEncodings.size()];
795 outs() << " [" << i << "]: "
796 << "function offset="
797 << format("0x%08" PRIx32, FunctionOffset) << ", "
798 << "encoding[" << EncodingIdx << "]="
799 << format("0x%08" PRIx32, Encoding)
805 printMachOUnwindInfoSection(const MachOObjectFile *Obj,
806 std::map<uint64_t, SymbolRef> &Symbols,
807 const SectionRef &UnwindInfo) {
809 assert(Obj->isLittleEndian() &&
810 "There should not be a big-endian .o with __unwind_info");
812 outs() << "Contents of __unwind_info section:\n";
815 UnwindInfo.getContents(Contents);
816 const char *Pos = Contents.data();
818 //===----------------------------------
820 //===----------------------------------
822 uint32_t Version = readNext<uint32_t>(Pos);
823 outs() << " Version: "
824 << format("0x%" PRIx32, Version) << '\n';
825 assert(Version == 1 && "only understand version 1");
827 uint32_t CommonEncodingsStart = readNext<uint32_t>(Pos);
828 outs() << " Common encodings array section offset: "
829 << format("0x%" PRIx32, CommonEncodingsStart) << '\n';
830 uint32_t NumCommonEncodings = readNext<uint32_t>(Pos);
831 outs() << " Number of common encodings in array: "
832 << format("0x%" PRIx32, NumCommonEncodings) << '\n';
834 uint32_t PersonalitiesStart = readNext<uint32_t>(Pos);
835 outs() << " Personality function array section offset: "
836 << format("0x%" PRIx32, PersonalitiesStart) << '\n';
837 uint32_t NumPersonalities = readNext<uint32_t>(Pos);
838 outs() << " Number of personality functions in array: "
839 << format("0x%" PRIx32, NumPersonalities) << '\n';
841 uint32_t IndicesStart = readNext<uint32_t>(Pos);
842 outs() << " Index array section offset: "
843 << format("0x%" PRIx32, IndicesStart) << '\n';
844 uint32_t NumIndices = readNext<uint32_t>(Pos);
845 outs() << " Number of indices in array: "
846 << format("0x%" PRIx32, NumIndices) << '\n';
848 //===----------------------------------
849 // A shared list of common encodings
850 //===----------------------------------
852 // These occupy indices in the range [0, N] whenever an encoding is referenced
853 // from a compressed 2nd level index table. In practice the linker only
854 // creates ~128 of these, so that indices are available to embed encodings in
855 // the 2nd level index.
857 SmallVector<uint32_t, 64> CommonEncodings;
858 outs() << " Common encodings: (count = " << NumCommonEncodings << ")\n";
859 Pos = Contents.data() + CommonEncodingsStart;
860 for (unsigned i = 0; i < NumCommonEncodings; ++i) {
861 uint32_t Encoding = readNext<uint32_t>(Pos);
862 CommonEncodings.push_back(Encoding);
864 outs() << " encoding[" << i << "]: " << format("0x%08" PRIx32, Encoding)
869 //===----------------------------------
870 // Personality functions used in this executable
871 //===----------------------------------
873 // There should be only a handful of these (one per source language,
874 // roughly). Particularly since they only get 2 bits in the compact encoding.
876 outs() << " Personality functions: (count = " << NumPersonalities << ")\n";
877 Pos = Contents.data() + PersonalitiesStart;
878 for (unsigned i = 0; i < NumPersonalities; ++i) {
879 uint32_t PersonalityFn = readNext<uint32_t>(Pos);
880 outs() << " personality[" << i + 1
881 << "]: " << format("0x%08" PRIx32, PersonalityFn) << '\n';
884 //===----------------------------------
885 // The level 1 index entries
886 //===----------------------------------
888 // These specify an approximate place to start searching for the more detailed
889 // information, sorted by PC.
892 uint32_t FunctionOffset;
893 uint32_t SecondLevelPageStart;
897 SmallVector<IndexEntry, 4> IndexEntries;
899 outs() << " Top level indices: (count = " << NumIndices << ")\n";
900 Pos = Contents.data() + IndicesStart;
901 for (unsigned i = 0; i < NumIndices; ++i) {
904 Entry.FunctionOffset = readNext<uint32_t>(Pos);
905 Entry.SecondLevelPageStart = readNext<uint32_t>(Pos);
906 Entry.LSDAStart = readNext<uint32_t>(Pos);
907 IndexEntries.push_back(Entry);
909 outs() << " [" << i << "]: "
910 << "function offset="
911 << format("0x%08" PRIx32, Entry.FunctionOffset) << ", "
912 << "2nd level page offset="
913 << format("0x%08" PRIx32, Entry.SecondLevelPageStart) << ", "
915 << format("0x%08" PRIx32, Entry.LSDAStart) << '\n';
919 //===----------------------------------
920 // Next come the LSDA tables
921 //===----------------------------------
923 // The LSDA layout is rather implicit: it's a contiguous array of entries from
924 // the first top-level index's LSDAOffset to the last (sentinel).
926 outs() << " LSDA descriptors:\n";
927 Pos = Contents.data() + IndexEntries[0].LSDAStart;
928 int NumLSDAs = (IndexEntries.back().LSDAStart - IndexEntries[0].LSDAStart) /
929 (2 * sizeof(uint32_t));
930 for (int i = 0; i < NumLSDAs; ++i) {
931 uint32_t FunctionOffset = readNext<uint32_t>(Pos);
932 uint32_t LSDAOffset = readNext<uint32_t>(Pos);
933 outs() << " [" << i << "]: "
934 << "function offset="
935 << format("0x%08" PRIx32, FunctionOffset) << ", "
937 << format("0x%08" PRIx32, LSDAOffset) << '\n';
940 //===----------------------------------
941 // Finally, the 2nd level indices
942 //===----------------------------------
944 // Generally these are 4K in size, and have 2 possible forms:
945 // + Regular stores up to 511 entries with disparate encodings
946 // + Compressed stores up to 1021 entries if few enough compact encoding
948 outs() << " Second level indices:\n";
949 for (unsigned i = 0; i < IndexEntries.size() - 1; ++i) {
950 // The final sentinel top-level index has no associated 2nd level page
951 if (IndexEntries[i].SecondLevelPageStart == 0)
954 outs() << " Second level index[" << i << "]: "
955 << "offset in section="
956 << format("0x%08" PRIx32, IndexEntries[i].SecondLevelPageStart)
958 << "base function offset="
959 << format("0x%08" PRIx32, IndexEntries[i].FunctionOffset) << '\n';
961 Pos = Contents.data() + IndexEntries[i].SecondLevelPageStart;
962 uint32_t Kind = *reinterpret_cast<const support::ulittle32_t *>(Pos);
964 printRegularSecondLevelUnwindPage(Pos);
966 printCompressedSecondLevelUnwindPage(Pos, IndexEntries[i].FunctionOffset,
969 llvm_unreachable("Do not know how to print this kind of 2nd level page");
974 void llvm::printMachOUnwindInfo(const MachOObjectFile *Obj) {
975 std::map<uint64_t, SymbolRef> Symbols;
976 for (const SymbolRef &SymRef : Obj->symbols()) {
977 // Discard any undefined or absolute symbols. They're not going to take part
978 // in the convenience lookup for unwind info and just take up resources.
979 section_iterator Section = Obj->section_end();
980 SymRef.getSection(Section);
981 if (Section == Obj->section_end())
985 SymRef.getAddress(Addr);
986 Symbols.insert(std::make_pair(Addr, SymRef));
989 for (const SectionRef &Section : Obj->sections()) {
991 Section.getName(SectName);
992 if (SectName == "__compact_unwind")
993 printMachOCompactUnwindSection(Obj, Symbols, Section);
994 else if (SectName == "__unwind_info")
995 printMachOUnwindInfoSection(Obj, Symbols, Section);
996 else if (SectName == "__eh_frame")
997 outs() << "llvm-objdump: warning: unhandled __eh_frame section\n";