1 //===-- llvm-objdump.cpp - Object file dumping utility for llvm -----------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 // This program is a utility that works like binutils "objdump", that is, it 10 // dumps out a plethora of information about an object file depending on the 11 // flags. 12 // 13 // The flags and output of this program should be near identical to those of 14 // binutils objdump. 15 // 16 //===----------------------------------------------------------------------===// 17 18 #include "llvm-objdump.h" 19 #include "COFFDump.h" 20 #include "ELFDump.h" 21 #include "MachODump.h" 22 #include "ObjdumpOptID.h" 23 #include "OffloadDump.h" 24 #include "SourcePrinter.h" 25 #include "WasmDump.h" 26 #include "XCOFFDump.h" 27 #include "llvm/ADT/STLExtras.h" 28 #include "llvm/ADT/SetOperations.h" 29 #include "llvm/ADT/StringExtras.h" 30 #include "llvm/ADT/StringSet.h" 31 #include "llvm/ADT/Twine.h" 32 #include "llvm/BinaryFormat/Wasm.h" 33 #include "llvm/DebugInfo/BTF/BTFParser.h" 34 #include "llvm/DebugInfo/DWARF/DWARFContext.h" 35 #include "llvm/DebugInfo/Symbolize/SymbolizableModule.h" 36 #include "llvm/DebugInfo/Symbolize/Symbolize.h" 37 #include "llvm/Debuginfod/BuildIDFetcher.h" 38 #include "llvm/Debuginfod/Debuginfod.h" 39 #include "llvm/Debuginfod/HTTPClient.h" 40 #include "llvm/Demangle/Demangle.h" 41 #include "llvm/MC/MCAsmInfo.h" 42 #include "llvm/MC/MCContext.h" 43 #include "llvm/MC/MCDisassembler/MCDisassembler.h" 44 #include "llvm/MC/MCDisassembler/MCRelocationInfo.h" 45 #include "llvm/MC/MCInst.h" 46 #include "llvm/MC/MCInstPrinter.h" 47 #include "llvm/MC/MCInstrAnalysis.h" 48 #include "llvm/MC/MCInstrInfo.h" 49 #include "llvm/MC/MCObjectFileInfo.h" 50 #include "llvm/MC/MCRegisterInfo.h" 51 #include "llvm/MC/MCTargetOptions.h" 52 #include "llvm/MC/TargetRegistry.h" 53 #include "llvm/Object/Archive.h" 54 #include "llvm/Object/BuildID.h" 55 #include "llvm/Object/COFF.h" 56 #include "llvm/Object/COFFImportFile.h" 57 #include "llvm/Object/ELFObjectFile.h" 58 #include "llvm/Object/ELFTypes.h" 59 #include "llvm/Object/FaultMapParser.h" 60 #include "llvm/Object/MachO.h" 61 #include "llvm/Object/MachOUniversal.h" 62 #include "llvm/Object/ObjectFile.h" 63 #include "llvm/Object/OffloadBinary.h" 64 #include "llvm/Object/Wasm.h" 65 #include "llvm/Option/Arg.h" 66 #include "llvm/Option/ArgList.h" 67 #include "llvm/Option/Option.h" 68 #include "llvm/Support/Casting.h" 69 #include "llvm/Support/Debug.h" 70 #include "llvm/Support/Errc.h" 71 #include "llvm/Support/FileSystem.h" 72 #include "llvm/Support/Format.h" 73 #include "llvm/Support/FormatVariadic.h" 74 #include "llvm/Support/GraphWriter.h" 75 #include "llvm/Support/LLVMDriver.h" 76 #include "llvm/Support/MemoryBuffer.h" 77 #include "llvm/Support/SourceMgr.h" 78 #include "llvm/Support/StringSaver.h" 79 #include "llvm/Support/TargetSelect.h" 80 #include "llvm/Support/WithColor.h" 81 #include "llvm/Support/raw_ostream.h" 82 #include "llvm/TargetParser/Host.h" 83 #include "llvm/TargetParser/Triple.h" 84 #include <algorithm> 85 #include <cctype> 86 #include <cstring> 87 #include <optional> 88 #include <set> 89 #include <system_error> 90 #include <unordered_map> 91 #include <utility> 92 93 using namespace llvm; 94 using namespace llvm::object; 95 using namespace llvm::objdump; 96 using namespace llvm::opt; 97 98 namespace { 99 100 class CommonOptTable : public opt::GenericOptTable { 101 public: 102 CommonOptTable(ArrayRef<Info> OptionInfos, const char *Usage, 103 const char *Description) 104 : opt::GenericOptTable(OptionInfos), Usage(Usage), 105 Description(Description) { 106 setGroupedShortOptions(true); 107 } 108 109 void printHelp(StringRef Argv0, bool ShowHidden = false) const { 110 Argv0 = sys::path::filename(Argv0); 111 opt::GenericOptTable::printHelp(outs(), (Argv0 + Usage).str().c_str(), 112 Description, ShowHidden, ShowHidden); 113 // TODO Replace this with OptTable API once it adds extrahelp support. 114 outs() << "\nPass @FILE as argument to read options from FILE.\n"; 115 } 116 117 private: 118 const char *Usage; 119 const char *Description; 120 }; 121 122 // ObjdumpOptID is in ObjdumpOptID.h 123 namespace objdump_opt { 124 #define PREFIX(NAME, VALUE) \ 125 static constexpr StringLiteral NAME##_init[] = VALUE; \ 126 static constexpr ArrayRef<StringLiteral> NAME(NAME##_init, \ 127 std::size(NAME##_init) - 1); 128 #include "ObjdumpOpts.inc" 129 #undef PREFIX 130 131 static constexpr opt::OptTable::Info ObjdumpInfoTable[] = { 132 #define OPTION(...) \ 133 LLVM_CONSTRUCT_OPT_INFO_WITH_ID_PREFIX(OBJDUMP_, __VA_ARGS__), 134 #include "ObjdumpOpts.inc" 135 #undef OPTION 136 }; 137 } // namespace objdump_opt 138 139 class ObjdumpOptTable : public CommonOptTable { 140 public: 141 ObjdumpOptTable() 142 : CommonOptTable(objdump_opt::ObjdumpInfoTable, 143 " [options] <input object files>", 144 "llvm object file dumper") {} 145 }; 146 147 enum OtoolOptID { 148 OTOOL_INVALID = 0, // This is not an option ID. 149 #define OPTION(...) LLVM_MAKE_OPT_ID_WITH_ID_PREFIX(OTOOL_, __VA_ARGS__), 150 #include "OtoolOpts.inc" 151 #undef OPTION 152 }; 153 154 namespace otool { 155 #define PREFIX(NAME, VALUE) \ 156 static constexpr StringLiteral NAME##_init[] = VALUE; \ 157 static constexpr ArrayRef<StringLiteral> NAME(NAME##_init, \ 158 std::size(NAME##_init) - 1); 159 #include "OtoolOpts.inc" 160 #undef PREFIX 161 162 static constexpr opt::OptTable::Info OtoolInfoTable[] = { 163 #define OPTION(...) LLVM_CONSTRUCT_OPT_INFO_WITH_ID_PREFIX(OTOOL_, __VA_ARGS__), 164 #include "OtoolOpts.inc" 165 #undef OPTION 166 }; 167 } // namespace otool 168 169 class OtoolOptTable : public CommonOptTable { 170 public: 171 OtoolOptTable() 172 : CommonOptTable(otool::OtoolInfoTable, " [option...] [file...]", 173 "Mach-O object file displaying tool") {} 174 }; 175 176 struct BBAddrMapLabel { 177 std::string BlockLabel; 178 std::string PGOAnalysis; 179 }; 180 181 // This class represents the BBAddrMap and PGOMap associated with a single 182 // function. 183 class BBAddrMapFunctionEntry { 184 public: 185 BBAddrMapFunctionEntry(BBAddrMap AddrMap, PGOAnalysisMap PGOMap) 186 : AddrMap(std::move(AddrMap)), PGOMap(std::move(PGOMap)) {} 187 188 const BBAddrMap &getAddrMap() const { return AddrMap; } 189 190 // Returns the PGO string associated with the entry of index `PGOBBEntryIndex` 191 // in `PGOMap`. 192 std::string constructPGOLabelString(size_t PGOBBEntryIndex) const { 193 if (!PGOMap.FeatEnable.hasPGOAnalysis()) 194 return ""; 195 std::string PGOString; 196 raw_string_ostream PGOSS(PGOString); 197 198 PGOSS << " ("; 199 if (PGOMap.FeatEnable.FuncEntryCount && PGOBBEntryIndex == 0) { 200 PGOSS << "Entry count: " << Twine(PGOMap.FuncEntryCount); 201 if (PGOMap.FeatEnable.hasPGOAnalysisBBData()) { 202 PGOSS << ", "; 203 } 204 } 205 206 if (PGOMap.FeatEnable.hasPGOAnalysisBBData()) { 207 208 assert(PGOBBEntryIndex < PGOMap.BBEntries.size() && 209 "Expected PGOAnalysisMap and BBAddrMap to have the same entries"); 210 const PGOAnalysisMap::PGOBBEntry &PGOBBEntry = 211 PGOMap.BBEntries[PGOBBEntryIndex]; 212 213 if (PGOMap.FeatEnable.BBFreq) { 214 PGOSS << "Frequency: " << Twine(PGOBBEntry.BlockFreq.getFrequency()); 215 if (PGOMap.FeatEnable.BrProb && PGOBBEntry.Successors.size() > 0) { 216 PGOSS << ", "; 217 } 218 } 219 if (PGOMap.FeatEnable.BrProb && PGOBBEntry.Successors.size() > 0) { 220 PGOSS << "Successors: "; 221 interleaveComma( 222 PGOBBEntry.Successors, PGOSS, 223 [&PGOSS](const PGOAnalysisMap::PGOBBEntry::SuccessorEntry &SE) { 224 PGOSS << "BB" << SE.ID << ":"; 225 PGOSS.write_hex(SE.Prob.getNumerator()); 226 }); 227 } 228 } 229 PGOSS << ")"; 230 231 return PGOString; 232 } 233 234 private: 235 const BBAddrMap AddrMap; 236 const PGOAnalysisMap PGOMap; 237 }; 238 239 // This class represents the BBAddrMap and PGOMap of potentially multiple 240 // functions in a section. 241 class BBAddrMapInfo { 242 public: 243 void clear() { 244 FunctionAddrToMap.clear(); 245 RangeBaseAddrToFunctionAddr.clear(); 246 } 247 248 bool empty() const { return FunctionAddrToMap.empty(); } 249 250 void AddFunctionEntry(BBAddrMap AddrMap, PGOAnalysisMap PGOMap) { 251 uint64_t FunctionAddr = AddrMap.getFunctionAddress(); 252 for (size_t I = 1; I < AddrMap.BBRanges.size(); ++I) 253 RangeBaseAddrToFunctionAddr.emplace(AddrMap.BBRanges[I].BaseAddress, 254 FunctionAddr); 255 [[maybe_unused]] auto R = FunctionAddrToMap.try_emplace( 256 FunctionAddr, std::move(AddrMap), std::move(PGOMap)); 257 assert(R.second && "duplicate function address"); 258 } 259 260 // Returns the BBAddrMap entry for the function associated with `BaseAddress`. 261 // `BaseAddress` could be the function address or the address of a range 262 // associated with that function. Returns `nullptr` if `BaseAddress` is not 263 // mapped to any entry. 264 const BBAddrMapFunctionEntry *getEntryForAddress(uint64_t BaseAddress) const { 265 uint64_t FunctionAddr = BaseAddress; 266 auto S = RangeBaseAddrToFunctionAddr.find(BaseAddress); 267 if (S != RangeBaseAddrToFunctionAddr.end()) 268 FunctionAddr = S->second; 269 auto R = FunctionAddrToMap.find(FunctionAddr); 270 if (R == FunctionAddrToMap.end()) 271 return nullptr; 272 return &R->second; 273 } 274 275 private: 276 std::unordered_map<uint64_t, BBAddrMapFunctionEntry> FunctionAddrToMap; 277 std::unordered_map<uint64_t, uint64_t> RangeBaseAddrToFunctionAddr; 278 }; 279 280 } // namespace 281 282 #define DEBUG_TYPE "objdump" 283 284 enum class ColorOutput { 285 Auto, 286 Enable, 287 Disable, 288 Invalid, 289 }; 290 291 static uint64_t AdjustVMA; 292 static bool AllHeaders; 293 static std::string ArchName; 294 bool objdump::ArchiveHeaders; 295 bool objdump::Demangle; 296 bool objdump::Disassemble; 297 bool objdump::DisassembleAll; 298 bool objdump::SymbolDescription; 299 bool objdump::TracebackTable; 300 static std::vector<std::string> DisassembleSymbols; 301 static bool DisassembleZeroes; 302 static std::vector<std::string> DisassemblerOptions; 303 static ColorOutput DisassemblyColor; 304 DIDumpType objdump::DwarfDumpType; 305 static bool DynamicRelocations; 306 static bool FaultMapSection; 307 static bool FileHeaders; 308 bool objdump::SectionContents; 309 static std::vector<std::string> InputFilenames; 310 bool objdump::PrintLines; 311 static bool MachOOpt; 312 std::string objdump::MCPU; 313 std::vector<std::string> objdump::MAttrs; 314 bool objdump::ShowRawInsn; 315 bool objdump::LeadingAddr; 316 static bool Offloading; 317 static bool RawClangAST; 318 bool objdump::Relocations; 319 bool objdump::PrintImmHex; 320 bool objdump::PrivateHeaders; 321 std::vector<std::string> objdump::FilterSections; 322 bool objdump::SectionHeaders; 323 static bool ShowAllSymbols; 324 static bool ShowLMA; 325 bool objdump::PrintSource; 326 327 static uint64_t StartAddress; 328 static bool HasStartAddressFlag; 329 static uint64_t StopAddress = UINT64_MAX; 330 static bool HasStopAddressFlag; 331 332 bool objdump::SymbolTable; 333 static bool SymbolizeOperands; 334 static bool DynamicSymbolTable; 335 std::string objdump::TripleName; 336 bool objdump::UnwindInfo; 337 static bool Wide; 338 std::string objdump::Prefix; 339 uint32_t objdump::PrefixStrip; 340 341 DebugVarsFormat objdump::DbgVariables = DVDisabled; 342 343 int objdump::DbgIndent = 52; 344 345 static StringSet<> DisasmSymbolSet; 346 StringSet<> objdump::FoundSectionSet; 347 static StringRef ToolName; 348 349 std::unique_ptr<BuildIDFetcher> BIDFetcher; 350 351 Dumper::Dumper(const object::ObjectFile &O) : O(O) { 352 WarningHandler = [this](const Twine &Msg) { 353 if (Warnings.insert(Msg.str()).second) 354 reportWarning(Msg, this->O.getFileName()); 355 return Error::success(); 356 }; 357 } 358 359 void Dumper::reportUniqueWarning(Error Err) { 360 reportUniqueWarning(toString(std::move(Err))); 361 } 362 363 void Dumper::reportUniqueWarning(const Twine &Msg) { 364 cantFail(WarningHandler(Msg)); 365 } 366 367 static Expected<std::unique_ptr<Dumper>> createDumper(const ObjectFile &Obj) { 368 if (const auto *O = dyn_cast<COFFObjectFile>(&Obj)) 369 return createCOFFDumper(*O); 370 if (const auto *O = dyn_cast<ELFObjectFileBase>(&Obj)) 371 return createELFDumper(*O); 372 if (const auto *O = dyn_cast<MachOObjectFile>(&Obj)) 373 return createMachODumper(*O); 374 if (const auto *O = dyn_cast<WasmObjectFile>(&Obj)) 375 return createWasmDumper(*O); 376 if (const auto *O = dyn_cast<XCOFFObjectFile>(&Obj)) 377 return createXCOFFDumper(*O); 378 379 return createStringError(errc::invalid_argument, 380 "unsupported object file format"); 381 } 382 383 namespace { 384 struct FilterResult { 385 // True if the section should not be skipped. 386 bool Keep; 387 388 // True if the index counter should be incremented, even if the section should 389 // be skipped. For example, sections may be skipped if they are not included 390 // in the --section flag, but we still want those to count toward the section 391 // count. 392 bool IncrementIndex; 393 }; 394 } // namespace 395 396 static FilterResult checkSectionFilter(object::SectionRef S) { 397 if (FilterSections.empty()) 398 return {/*Keep=*/true, /*IncrementIndex=*/true}; 399 400 Expected<StringRef> SecNameOrErr = S.getName(); 401 if (!SecNameOrErr) { 402 consumeError(SecNameOrErr.takeError()); 403 return {/*Keep=*/false, /*IncrementIndex=*/false}; 404 } 405 StringRef SecName = *SecNameOrErr; 406 407 // StringSet does not allow empty key so avoid adding sections with 408 // no name (such as the section with index 0) here. 409 if (!SecName.empty()) 410 FoundSectionSet.insert(SecName); 411 412 // Only show the section if it's in the FilterSections list, but always 413 // increment so the indexing is stable. 414 return {/*Keep=*/is_contained(FilterSections, SecName), 415 /*IncrementIndex=*/true}; 416 } 417 418 SectionFilter objdump::ToolSectionFilter(object::ObjectFile const &O, 419 uint64_t *Idx) { 420 // Start at UINT64_MAX so that the first index returned after an increment is 421 // zero (after the unsigned wrap). 422 if (Idx) 423 *Idx = UINT64_MAX; 424 return SectionFilter( 425 [Idx](object::SectionRef S) { 426 FilterResult Result = checkSectionFilter(S); 427 if (Idx != nullptr && Result.IncrementIndex) 428 *Idx += 1; 429 return Result.Keep; 430 }, 431 O); 432 } 433 434 std::string objdump::getFileNameForError(const object::Archive::Child &C, 435 unsigned Index) { 436 Expected<StringRef> NameOrErr = C.getName(); 437 if (NameOrErr) 438 return std::string(NameOrErr.get()); 439 // If we have an error getting the name then we print the index of the archive 440 // member. Since we are already in an error state, we just ignore this error. 441 consumeError(NameOrErr.takeError()); 442 return "<file index: " + std::to_string(Index) + ">"; 443 } 444 445 void objdump::reportWarning(const Twine &Message, StringRef File) { 446 // Output order between errs() and outs() matters especially for archive 447 // files where the output is per member object. 448 outs().flush(); 449 WithColor::warning(errs(), ToolName) 450 << "'" << File << "': " << Message << "\n"; 451 } 452 453 [[noreturn]] void objdump::reportError(StringRef File, const Twine &Message) { 454 outs().flush(); 455 WithColor::error(errs(), ToolName) << "'" << File << "': " << Message << "\n"; 456 exit(1); 457 } 458 459 [[noreturn]] void objdump::reportError(Error E, StringRef FileName, 460 StringRef ArchiveName, 461 StringRef ArchitectureName) { 462 assert(E); 463 outs().flush(); 464 WithColor::error(errs(), ToolName); 465 if (ArchiveName != "") 466 errs() << ArchiveName << "(" << FileName << ")"; 467 else 468 errs() << "'" << FileName << "'"; 469 if (!ArchitectureName.empty()) 470 errs() << " (for architecture " << ArchitectureName << ")"; 471 errs() << ": "; 472 logAllUnhandledErrors(std::move(E), errs()); 473 exit(1); 474 } 475 476 static void reportCmdLineWarning(const Twine &Message) { 477 WithColor::warning(errs(), ToolName) << Message << "\n"; 478 } 479 480 [[noreturn]] static void reportCmdLineError(const Twine &Message) { 481 WithColor::error(errs(), ToolName) << Message << "\n"; 482 exit(1); 483 } 484 485 static void warnOnNoMatchForSections() { 486 SetVector<StringRef> MissingSections; 487 for (StringRef S : FilterSections) { 488 if (FoundSectionSet.count(S)) 489 return; 490 // User may specify a unnamed section. Don't warn for it. 491 if (!S.empty()) 492 MissingSections.insert(S); 493 } 494 495 // Warn only if no section in FilterSections is matched. 496 for (StringRef S : MissingSections) 497 reportCmdLineWarning("section '" + S + 498 "' mentioned in a -j/--section option, but not " 499 "found in any input file"); 500 } 501 502 static const Target *getTarget(const ObjectFile *Obj) { 503 // Figure out the target triple. 504 Triple TheTriple("unknown-unknown-unknown"); 505 if (TripleName.empty()) { 506 TheTriple = Obj->makeTriple(); 507 } else { 508 TheTriple.setTriple(Triple::normalize(TripleName)); 509 auto Arch = Obj->getArch(); 510 if (Arch == Triple::arm || Arch == Triple::armeb) 511 Obj->setARMSubArch(TheTriple); 512 } 513 514 // Get the target specific parser. 515 std::string Error; 516 const Target *TheTarget = TargetRegistry::lookupTarget(ArchName, TheTriple, 517 Error); 518 if (!TheTarget) 519 reportError(Obj->getFileName(), "can't find target: " + Error); 520 521 // Update the triple name and return the found target. 522 TripleName = TheTriple.getTriple(); 523 return TheTarget; 524 } 525 526 bool objdump::isRelocAddressLess(RelocationRef A, RelocationRef B) { 527 return A.getOffset() < B.getOffset(); 528 } 529 530 static Error getRelocationValueString(const RelocationRef &Rel, 531 bool SymbolDescription, 532 SmallVectorImpl<char> &Result) { 533 const ObjectFile *Obj = Rel.getObject(); 534 if (auto *ELF = dyn_cast<ELFObjectFileBase>(Obj)) 535 return getELFRelocationValueString(ELF, Rel, Result); 536 if (auto *COFF = dyn_cast<COFFObjectFile>(Obj)) 537 return getCOFFRelocationValueString(COFF, Rel, Result); 538 if (auto *Wasm = dyn_cast<WasmObjectFile>(Obj)) 539 return getWasmRelocationValueString(Wasm, Rel, Result); 540 if (auto *MachO = dyn_cast<MachOObjectFile>(Obj)) 541 return getMachORelocationValueString(MachO, Rel, Result); 542 if (auto *XCOFF = dyn_cast<XCOFFObjectFile>(Obj)) 543 return getXCOFFRelocationValueString(*XCOFF, Rel, SymbolDescription, 544 Result); 545 llvm_unreachable("unknown object file format"); 546 } 547 548 /// Indicates whether this relocation should hidden when listing 549 /// relocations, usually because it is the trailing part of a multipart 550 /// relocation that will be printed as part of the leading relocation. 551 static bool getHidden(RelocationRef RelRef) { 552 auto *MachO = dyn_cast<MachOObjectFile>(RelRef.getObject()); 553 if (!MachO) 554 return false; 555 556 unsigned Arch = MachO->getArch(); 557 DataRefImpl Rel = RelRef.getRawDataRefImpl(); 558 uint64_t Type = MachO->getRelocationType(Rel); 559 560 // On arches that use the generic relocations, GENERIC_RELOC_PAIR 561 // is always hidden. 562 if (Arch == Triple::x86 || Arch == Triple::arm || Arch == Triple::ppc) 563 return Type == MachO::GENERIC_RELOC_PAIR; 564 565 if (Arch == Triple::x86_64) { 566 // On x86_64, X86_64_RELOC_UNSIGNED is hidden only when it follows 567 // an X86_64_RELOC_SUBTRACTOR. 568 if (Type == MachO::X86_64_RELOC_UNSIGNED && Rel.d.a > 0) { 569 DataRefImpl RelPrev = Rel; 570 RelPrev.d.a--; 571 uint64_t PrevType = MachO->getRelocationType(RelPrev); 572 if (PrevType == MachO::X86_64_RELOC_SUBTRACTOR) 573 return true; 574 } 575 } 576 577 return false; 578 } 579 580 /// Get the column at which we want to start printing the instruction 581 /// disassembly, taking into account anything which appears to the left of it. 582 unsigned objdump::getInstStartColumn(const MCSubtargetInfo &STI) { 583 return !ShowRawInsn ? 16 : STI.getTargetTriple().isX86() ? 40 : 24; 584 } 585 586 static void AlignToInstStartColumn(size_t Start, const MCSubtargetInfo &STI, 587 raw_ostream &OS) { 588 // The output of printInst starts with a tab. Print some spaces so that 589 // the tab has 1 column and advances to the target tab stop. 590 unsigned TabStop = getInstStartColumn(STI); 591 unsigned Column = OS.tell() - Start; 592 OS.indent(Column < TabStop - 1 ? TabStop - 1 - Column : 7 - Column % 8); 593 } 594 595 void objdump::printRawData(ArrayRef<uint8_t> Bytes, uint64_t Address, 596 formatted_raw_ostream &OS, 597 MCSubtargetInfo const &STI) { 598 size_t Start = OS.tell(); 599 if (LeadingAddr) 600 OS << format("%8" PRIx64 ":", Address); 601 if (ShowRawInsn) { 602 OS << ' '; 603 dumpBytes(Bytes, OS); 604 } 605 AlignToInstStartColumn(Start, STI, OS); 606 } 607 608 namespace { 609 610 static bool isAArch64Elf(const ObjectFile &Obj) { 611 const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj); 612 return Elf && Elf->getEMachine() == ELF::EM_AARCH64; 613 } 614 615 static bool isArmElf(const ObjectFile &Obj) { 616 const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj); 617 return Elf && Elf->getEMachine() == ELF::EM_ARM; 618 } 619 620 static bool isCSKYElf(const ObjectFile &Obj) { 621 const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj); 622 return Elf && Elf->getEMachine() == ELF::EM_CSKY; 623 } 624 625 static bool hasMappingSymbols(const ObjectFile &Obj) { 626 return isArmElf(Obj) || isAArch64Elf(Obj) || isCSKYElf(Obj) ; 627 } 628 629 static void printRelocation(formatted_raw_ostream &OS, StringRef FileName, 630 const RelocationRef &Rel, uint64_t Address, 631 bool Is64Bits) { 632 StringRef Fmt = Is64Bits ? "%016" PRIx64 ": " : "%08" PRIx64 ": "; 633 SmallString<16> Name; 634 SmallString<32> Val; 635 Rel.getTypeName(Name); 636 if (Error E = getRelocationValueString(Rel, SymbolDescription, Val)) 637 reportError(std::move(E), FileName); 638 OS << (Is64Bits || !LeadingAddr ? "\t\t" : "\t\t\t"); 639 if (LeadingAddr) 640 OS << format(Fmt.data(), Address); 641 OS << Name << "\t" << Val; 642 } 643 644 static void printBTFRelocation(formatted_raw_ostream &FOS, llvm::BTFParser &BTF, 645 object::SectionedAddress Address, 646 LiveVariablePrinter &LVP) { 647 const llvm::BTF::BPFFieldReloc *Reloc = BTF.findFieldReloc(Address); 648 if (!Reloc) 649 return; 650 651 SmallString<64> Val; 652 BTF.symbolize(Reloc, Val); 653 FOS << "\t\t"; 654 if (LeadingAddr) 655 FOS << format("%016" PRIx64 ": ", Address.Address + AdjustVMA); 656 FOS << "CO-RE " << Val; 657 LVP.printAfterOtherLine(FOS, true); 658 } 659 660 class PrettyPrinter { 661 public: 662 virtual ~PrettyPrinter() = default; 663 virtual void 664 printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 665 object::SectionedAddress Address, formatted_raw_ostream &OS, 666 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 667 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 668 LiveVariablePrinter &LVP) { 669 if (SP && (PrintSource || PrintLines)) 670 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 671 LVP.printBetweenInsts(OS, false); 672 673 printRawData(Bytes, Address.Address, OS, STI); 674 675 if (MI) { 676 // See MCInstPrinter::printInst. On targets where a PC relative immediate 677 // is relative to the next instruction and the length of a MCInst is 678 // difficult to measure (x86), this is the address of the next 679 // instruction. 680 uint64_t Addr = 681 Address.Address + (STI.getTargetTriple().isX86() ? Bytes.size() : 0); 682 IP.printInst(MI, Addr, "", STI, OS); 683 } else 684 OS << "\t<unknown>"; 685 } 686 }; 687 PrettyPrinter PrettyPrinterInst; 688 689 class HexagonPrettyPrinter : public PrettyPrinter { 690 public: 691 void printLead(ArrayRef<uint8_t> Bytes, uint64_t Address, 692 formatted_raw_ostream &OS) { 693 uint32_t opcode = 694 (Bytes[3] << 24) | (Bytes[2] << 16) | (Bytes[1] << 8) | Bytes[0]; 695 if (LeadingAddr) 696 OS << format("%8" PRIx64 ":", Address); 697 if (ShowRawInsn) { 698 OS << "\t"; 699 dumpBytes(Bytes.slice(0, 4), OS); 700 OS << format("\t%08" PRIx32, opcode); 701 } 702 } 703 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 704 object::SectionedAddress Address, formatted_raw_ostream &OS, 705 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 706 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 707 LiveVariablePrinter &LVP) override { 708 if (SP && (PrintSource || PrintLines)) 709 SP->printSourceLine(OS, Address, ObjectFilename, LVP, ""); 710 if (!MI) { 711 printLead(Bytes, Address.Address, OS); 712 OS << " <unknown>"; 713 return; 714 } 715 std::string Buffer; 716 { 717 raw_string_ostream TempStream(Buffer); 718 IP.printInst(MI, Address.Address, "", STI, TempStream); 719 } 720 StringRef Contents(Buffer); 721 // Split off bundle attributes 722 auto PacketBundle = Contents.rsplit('\n'); 723 // Split off first instruction from the rest 724 auto HeadTail = PacketBundle.first.split('\n'); 725 auto Preamble = " { "; 726 auto Separator = ""; 727 728 // Hexagon's packets require relocations to be inline rather than 729 // clustered at the end of the packet. 730 std::vector<RelocationRef>::const_iterator RelCur = Rels->begin(); 731 std::vector<RelocationRef>::const_iterator RelEnd = Rels->end(); 732 auto PrintReloc = [&]() -> void { 733 while ((RelCur != RelEnd) && (RelCur->getOffset() <= Address.Address)) { 734 if (RelCur->getOffset() == Address.Address) { 735 printRelocation(OS, ObjectFilename, *RelCur, Address.Address, false); 736 return; 737 } 738 ++RelCur; 739 } 740 }; 741 742 while (!HeadTail.first.empty()) { 743 OS << Separator; 744 Separator = "\n"; 745 if (SP && (PrintSource || PrintLines)) 746 SP->printSourceLine(OS, Address, ObjectFilename, LVP, ""); 747 printLead(Bytes, Address.Address, OS); 748 OS << Preamble; 749 Preamble = " "; 750 StringRef Inst; 751 auto Duplex = HeadTail.first.split('\v'); 752 if (!Duplex.second.empty()) { 753 OS << Duplex.first; 754 OS << "; "; 755 Inst = Duplex.second; 756 } 757 else 758 Inst = HeadTail.first; 759 OS << Inst; 760 HeadTail = HeadTail.second.split('\n'); 761 if (HeadTail.first.empty()) 762 OS << " } " << PacketBundle.second; 763 PrintReloc(); 764 Bytes = Bytes.slice(4); 765 Address.Address += 4; 766 } 767 } 768 }; 769 HexagonPrettyPrinter HexagonPrettyPrinterInst; 770 771 class AMDGCNPrettyPrinter : public PrettyPrinter { 772 public: 773 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 774 object::SectionedAddress Address, formatted_raw_ostream &OS, 775 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 776 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 777 LiveVariablePrinter &LVP) override { 778 if (SP && (PrintSource || PrintLines)) 779 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 780 781 if (MI) { 782 SmallString<40> InstStr; 783 raw_svector_ostream IS(InstStr); 784 785 IP.printInst(MI, Address.Address, "", STI, IS); 786 787 OS << left_justify(IS.str(), 60); 788 } else { 789 // an unrecognized encoding - this is probably data so represent it 790 // using the .long directive, or .byte directive if fewer than 4 bytes 791 // remaining 792 if (Bytes.size() >= 4) { 793 OS << format( 794 "\t.long 0x%08" PRIx32 " ", 795 support::endian::read32<llvm::endianness::little>(Bytes.data())); 796 OS.indent(42); 797 } else { 798 OS << format("\t.byte 0x%02" PRIx8, Bytes[0]); 799 for (unsigned int i = 1; i < Bytes.size(); i++) 800 OS << format(", 0x%02" PRIx8, Bytes[i]); 801 OS.indent(55 - (6 * Bytes.size())); 802 } 803 } 804 805 OS << format("// %012" PRIX64 ":", Address.Address); 806 if (Bytes.size() >= 4) { 807 // D should be casted to uint32_t here as it is passed by format to 808 // snprintf as vararg. 809 for (uint32_t D : 810 ArrayRef(reinterpret_cast<const support::little32_t *>(Bytes.data()), 811 Bytes.size() / 4)) 812 OS << format(" %08" PRIX32, D); 813 } else { 814 for (unsigned char B : Bytes) 815 OS << format(" %02" PRIX8, B); 816 } 817 818 if (!Annot.empty()) 819 OS << " // " << Annot; 820 } 821 }; 822 AMDGCNPrettyPrinter AMDGCNPrettyPrinterInst; 823 824 class BPFPrettyPrinter : public PrettyPrinter { 825 public: 826 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 827 object::SectionedAddress Address, formatted_raw_ostream &OS, 828 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 829 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 830 LiveVariablePrinter &LVP) override { 831 if (SP && (PrintSource || PrintLines)) 832 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 833 if (LeadingAddr) 834 OS << format("%8" PRId64 ":", Address.Address / 8); 835 if (ShowRawInsn) { 836 OS << "\t"; 837 dumpBytes(Bytes, OS); 838 } 839 if (MI) 840 IP.printInst(MI, Address.Address, "", STI, OS); 841 else 842 OS << "\t<unknown>"; 843 } 844 }; 845 BPFPrettyPrinter BPFPrettyPrinterInst; 846 847 class ARMPrettyPrinter : public PrettyPrinter { 848 public: 849 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 850 object::SectionedAddress Address, formatted_raw_ostream &OS, 851 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 852 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 853 LiveVariablePrinter &LVP) override { 854 if (SP && (PrintSource || PrintLines)) 855 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 856 LVP.printBetweenInsts(OS, false); 857 858 size_t Start = OS.tell(); 859 if (LeadingAddr) 860 OS << format("%8" PRIx64 ":", Address.Address); 861 if (ShowRawInsn) { 862 size_t Pos = 0, End = Bytes.size(); 863 if (STI.checkFeatures("+thumb-mode")) { 864 for (; Pos + 2 <= End; Pos += 2) 865 OS << ' ' 866 << format_hex_no_prefix( 867 llvm::support::endian::read<uint16_t>( 868 Bytes.data() + Pos, InstructionEndianness), 869 4); 870 } else { 871 for (; Pos + 4 <= End; Pos += 4) 872 OS << ' ' 873 << format_hex_no_prefix( 874 llvm::support::endian::read<uint32_t>( 875 Bytes.data() + Pos, InstructionEndianness), 876 8); 877 } 878 if (Pos < End) { 879 OS << ' '; 880 dumpBytes(Bytes.slice(Pos), OS); 881 } 882 } 883 884 AlignToInstStartColumn(Start, STI, OS); 885 886 if (MI) { 887 IP.printInst(MI, Address.Address, "", STI, OS); 888 } else 889 OS << "\t<unknown>"; 890 } 891 892 void setInstructionEndianness(llvm::endianness Endianness) { 893 InstructionEndianness = Endianness; 894 } 895 896 private: 897 llvm::endianness InstructionEndianness = llvm::endianness::little; 898 }; 899 ARMPrettyPrinter ARMPrettyPrinterInst; 900 901 class AArch64PrettyPrinter : public PrettyPrinter { 902 public: 903 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 904 object::SectionedAddress Address, formatted_raw_ostream &OS, 905 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 906 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 907 LiveVariablePrinter &LVP) override { 908 if (SP && (PrintSource || PrintLines)) 909 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 910 LVP.printBetweenInsts(OS, false); 911 912 size_t Start = OS.tell(); 913 if (LeadingAddr) 914 OS << format("%8" PRIx64 ":", Address.Address); 915 if (ShowRawInsn) { 916 size_t Pos = 0, End = Bytes.size(); 917 for (; Pos + 4 <= End; Pos += 4) 918 OS << ' ' 919 << format_hex_no_prefix( 920 llvm::support::endian::read<uint32_t>( 921 Bytes.data() + Pos, llvm::endianness::little), 922 8); 923 if (Pos < End) { 924 OS << ' '; 925 dumpBytes(Bytes.slice(Pos), OS); 926 } 927 } 928 929 AlignToInstStartColumn(Start, STI, OS); 930 931 if (MI) { 932 IP.printInst(MI, Address.Address, "", STI, OS); 933 } else 934 OS << "\t<unknown>"; 935 } 936 }; 937 AArch64PrettyPrinter AArch64PrettyPrinterInst; 938 939 PrettyPrinter &selectPrettyPrinter(Triple const &Triple) { 940 switch(Triple.getArch()) { 941 default: 942 return PrettyPrinterInst; 943 case Triple::hexagon: 944 return HexagonPrettyPrinterInst; 945 case Triple::amdgcn: 946 return AMDGCNPrettyPrinterInst; 947 case Triple::bpfel: 948 case Triple::bpfeb: 949 return BPFPrettyPrinterInst; 950 case Triple::arm: 951 case Triple::armeb: 952 case Triple::thumb: 953 case Triple::thumbeb: 954 return ARMPrettyPrinterInst; 955 case Triple::aarch64: 956 case Triple::aarch64_be: 957 case Triple::aarch64_32: 958 return AArch64PrettyPrinterInst; 959 } 960 } 961 962 class DisassemblerTarget { 963 public: 964 const Target *TheTarget; 965 std::unique_ptr<const MCSubtargetInfo> SubtargetInfo; 966 std::shared_ptr<MCContext> Context; 967 std::unique_ptr<MCDisassembler> DisAsm; 968 std::shared_ptr<MCInstrAnalysis> InstrAnalysis; 969 std::shared_ptr<MCInstPrinter> InstPrinter; 970 PrettyPrinter *Printer; 971 972 DisassemblerTarget(const Target *TheTarget, ObjectFile &Obj, 973 StringRef TripleName, StringRef MCPU, 974 SubtargetFeatures &Features); 975 DisassemblerTarget(DisassemblerTarget &Other, SubtargetFeatures &Features); 976 977 private: 978 MCTargetOptions Options; 979 std::shared_ptr<const MCRegisterInfo> RegisterInfo; 980 std::shared_ptr<const MCAsmInfo> AsmInfo; 981 std::shared_ptr<const MCInstrInfo> InstrInfo; 982 std::shared_ptr<MCObjectFileInfo> ObjectFileInfo; 983 }; 984 985 DisassemblerTarget::DisassemblerTarget(const Target *TheTarget, ObjectFile &Obj, 986 StringRef TripleName, StringRef MCPU, 987 SubtargetFeatures &Features) 988 : TheTarget(TheTarget), 989 Printer(&selectPrettyPrinter(Triple(TripleName))), 990 RegisterInfo(TheTarget->createMCRegInfo(TripleName)) { 991 if (!RegisterInfo) 992 reportError(Obj.getFileName(), "no register info for target " + TripleName); 993 994 // Set up disassembler. 995 AsmInfo.reset(TheTarget->createMCAsmInfo(*RegisterInfo, TripleName, Options)); 996 if (!AsmInfo) 997 reportError(Obj.getFileName(), "no assembly info for target " + TripleName); 998 999 SubtargetInfo.reset( 1000 TheTarget->createMCSubtargetInfo(TripleName, MCPU, Features.getString())); 1001 if (!SubtargetInfo) 1002 reportError(Obj.getFileName(), 1003 "no subtarget info for target " + TripleName); 1004 InstrInfo.reset(TheTarget->createMCInstrInfo()); 1005 if (!InstrInfo) 1006 reportError(Obj.getFileName(), 1007 "no instruction info for target " + TripleName); 1008 Context = 1009 std::make_shared<MCContext>(Triple(TripleName), AsmInfo.get(), 1010 RegisterInfo.get(), SubtargetInfo.get()); 1011 1012 // FIXME: for now initialize MCObjectFileInfo with default values 1013 ObjectFileInfo.reset( 1014 TheTarget->createMCObjectFileInfo(*Context, /*PIC=*/false)); 1015 Context->setObjectFileInfo(ObjectFileInfo.get()); 1016 1017 DisAsm.reset(TheTarget->createMCDisassembler(*SubtargetInfo, *Context)); 1018 if (!DisAsm) 1019 reportError(Obj.getFileName(), "no disassembler for target " + TripleName); 1020 1021 if (auto *ELFObj = dyn_cast<ELFObjectFileBase>(&Obj)) 1022 DisAsm->setABIVersion(ELFObj->getEIdentABIVersion()); 1023 1024 InstrAnalysis.reset(TheTarget->createMCInstrAnalysis(InstrInfo.get())); 1025 1026 int AsmPrinterVariant = AsmInfo->getAssemblerDialect(); 1027 InstPrinter.reset(TheTarget->createMCInstPrinter(Triple(TripleName), 1028 AsmPrinterVariant, *AsmInfo, 1029 *InstrInfo, *RegisterInfo)); 1030 if (!InstPrinter) 1031 reportError(Obj.getFileName(), 1032 "no instruction printer for target " + TripleName); 1033 InstPrinter->setPrintImmHex(PrintImmHex); 1034 InstPrinter->setPrintBranchImmAsAddress(true); 1035 InstPrinter->setSymbolizeOperands(SymbolizeOperands); 1036 InstPrinter->setMCInstrAnalysis(InstrAnalysis.get()); 1037 1038 switch (DisassemblyColor) { 1039 case ColorOutput::Enable: 1040 InstPrinter->setUseColor(true); 1041 break; 1042 case ColorOutput::Auto: 1043 InstPrinter->setUseColor(outs().has_colors()); 1044 break; 1045 case ColorOutput::Disable: 1046 case ColorOutput::Invalid: 1047 InstPrinter->setUseColor(false); 1048 break; 1049 }; 1050 } 1051 1052 DisassemblerTarget::DisassemblerTarget(DisassemblerTarget &Other, 1053 SubtargetFeatures &Features) 1054 : TheTarget(Other.TheTarget), 1055 SubtargetInfo(TheTarget->createMCSubtargetInfo(TripleName, MCPU, 1056 Features.getString())), 1057 Context(Other.Context), 1058 DisAsm(TheTarget->createMCDisassembler(*SubtargetInfo, *Context)), 1059 InstrAnalysis(Other.InstrAnalysis), InstPrinter(Other.InstPrinter), 1060 Printer(Other.Printer), RegisterInfo(Other.RegisterInfo), 1061 AsmInfo(Other.AsmInfo), InstrInfo(Other.InstrInfo), 1062 ObjectFileInfo(Other.ObjectFileInfo) {} 1063 } // namespace 1064 1065 static uint8_t getElfSymbolType(const ObjectFile &Obj, const SymbolRef &Sym) { 1066 assert(Obj.isELF()); 1067 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj)) 1068 return unwrapOrError(Elf32LEObj->getSymbol(Sym.getRawDataRefImpl()), 1069 Obj.getFileName()) 1070 ->getType(); 1071 if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj)) 1072 return unwrapOrError(Elf64LEObj->getSymbol(Sym.getRawDataRefImpl()), 1073 Obj.getFileName()) 1074 ->getType(); 1075 if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj)) 1076 return unwrapOrError(Elf32BEObj->getSymbol(Sym.getRawDataRefImpl()), 1077 Obj.getFileName()) 1078 ->getType(); 1079 if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj)) 1080 return unwrapOrError(Elf64BEObj->getSymbol(Sym.getRawDataRefImpl()), 1081 Obj.getFileName()) 1082 ->getType(); 1083 llvm_unreachable("Unsupported binary format"); 1084 } 1085 1086 template <class ELFT> 1087 static void 1088 addDynamicElfSymbols(const ELFObjectFile<ELFT> &Obj, 1089 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 1090 for (auto Symbol : Obj.getDynamicSymbolIterators()) { 1091 uint8_t SymbolType = Symbol.getELFType(); 1092 if (SymbolType == ELF::STT_SECTION) 1093 continue; 1094 1095 uint64_t Address = unwrapOrError(Symbol.getAddress(), Obj.getFileName()); 1096 // ELFSymbolRef::getAddress() returns size instead of value for common 1097 // symbols which is not desirable for disassembly output. Overriding. 1098 if (SymbolType == ELF::STT_COMMON) 1099 Address = unwrapOrError(Obj.getSymbol(Symbol.getRawDataRefImpl()), 1100 Obj.getFileName()) 1101 ->st_value; 1102 1103 StringRef Name = unwrapOrError(Symbol.getName(), Obj.getFileName()); 1104 if (Name.empty()) 1105 continue; 1106 1107 section_iterator SecI = 1108 unwrapOrError(Symbol.getSection(), Obj.getFileName()); 1109 if (SecI == Obj.section_end()) 1110 continue; 1111 1112 AllSymbols[*SecI].emplace_back(Address, Name, SymbolType); 1113 } 1114 } 1115 1116 static void 1117 addDynamicElfSymbols(const ELFObjectFileBase &Obj, 1118 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 1119 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj)) 1120 addDynamicElfSymbols(*Elf32LEObj, AllSymbols); 1121 else if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj)) 1122 addDynamicElfSymbols(*Elf64LEObj, AllSymbols); 1123 else if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj)) 1124 addDynamicElfSymbols(*Elf32BEObj, AllSymbols); 1125 else if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj)) 1126 addDynamicElfSymbols(*Elf64BEObj, AllSymbols); 1127 else 1128 llvm_unreachable("Unsupported binary format"); 1129 } 1130 1131 static std::optional<SectionRef> getWasmCodeSection(const WasmObjectFile &Obj) { 1132 for (auto SecI : Obj.sections()) { 1133 const WasmSection &Section = Obj.getWasmSection(SecI); 1134 if (Section.Type == wasm::WASM_SEC_CODE) 1135 return SecI; 1136 } 1137 return std::nullopt; 1138 } 1139 1140 static void 1141 addMissingWasmCodeSymbols(const WasmObjectFile &Obj, 1142 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 1143 std::optional<SectionRef> Section = getWasmCodeSection(Obj); 1144 if (!Section) 1145 return; 1146 SectionSymbolsTy &Symbols = AllSymbols[*Section]; 1147 1148 std::set<uint64_t> SymbolAddresses; 1149 for (const auto &Sym : Symbols) 1150 SymbolAddresses.insert(Sym.Addr); 1151 1152 for (const wasm::WasmFunction &Function : Obj.functions()) { 1153 // This adjustment mirrors the one in WasmObjectFile::getSymbolAddress. 1154 uint32_t Adjustment = Obj.isRelocatableObject() || Obj.isSharedObject() 1155 ? 0 1156 : Section->getAddress(); 1157 uint64_t Address = Function.CodeSectionOffset + Adjustment; 1158 // Only add fallback symbols for functions not already present in the symbol 1159 // table. 1160 if (SymbolAddresses.count(Address)) 1161 continue; 1162 // This function has no symbol, so it should have no SymbolName. 1163 assert(Function.SymbolName.empty()); 1164 // We use DebugName for the name, though it may be empty if there is no 1165 // "name" custom section, or that section is missing a name for this 1166 // function. 1167 StringRef Name = Function.DebugName; 1168 Symbols.emplace_back(Address, Name, ELF::STT_NOTYPE); 1169 } 1170 } 1171 1172 static void addPltEntries(const ObjectFile &Obj, 1173 std::map<SectionRef, SectionSymbolsTy> &AllSymbols, 1174 StringSaver &Saver) { 1175 auto *ElfObj = dyn_cast<ELFObjectFileBase>(&Obj); 1176 if (!ElfObj) 1177 return; 1178 DenseMap<StringRef, SectionRef> Sections; 1179 for (SectionRef Section : Obj.sections()) { 1180 Expected<StringRef> SecNameOrErr = Section.getName(); 1181 if (!SecNameOrErr) { 1182 consumeError(SecNameOrErr.takeError()); 1183 continue; 1184 } 1185 Sections[*SecNameOrErr] = Section; 1186 } 1187 for (auto Plt : ElfObj->getPltEntries()) { 1188 if (Plt.Symbol) { 1189 SymbolRef Symbol(*Plt.Symbol, ElfObj); 1190 uint8_t SymbolType = getElfSymbolType(Obj, Symbol); 1191 if (Expected<StringRef> NameOrErr = Symbol.getName()) { 1192 if (!NameOrErr->empty()) 1193 AllSymbols[Sections[Plt.Section]].emplace_back( 1194 Plt.Address, Saver.save((*NameOrErr + "@plt").str()), SymbolType); 1195 continue; 1196 } else { 1197 // The warning has been reported in disassembleObject(). 1198 consumeError(NameOrErr.takeError()); 1199 } 1200 } 1201 reportWarning("PLT entry at 0x" + Twine::utohexstr(Plt.Address) + 1202 " references an invalid symbol", 1203 Obj.getFileName()); 1204 } 1205 } 1206 1207 // Normally the disassembly output will skip blocks of zeroes. This function 1208 // returns the number of zero bytes that can be skipped when dumping the 1209 // disassembly of the instructions in Buf. 1210 static size_t countSkippableZeroBytes(ArrayRef<uint8_t> Buf) { 1211 // Find the number of leading zeroes. 1212 size_t N = 0; 1213 while (N < Buf.size() && !Buf[N]) 1214 ++N; 1215 1216 // We may want to skip blocks of zero bytes, but unless we see 1217 // at least 8 of them in a row. 1218 if (N < 8) 1219 return 0; 1220 1221 // We skip zeroes in multiples of 4 because do not want to truncate an 1222 // instruction if it starts with a zero byte. 1223 return N & ~0x3; 1224 } 1225 1226 // Returns a map from sections to their relocations. 1227 static std::map<SectionRef, std::vector<RelocationRef>> 1228 getRelocsMap(object::ObjectFile const &Obj) { 1229 std::map<SectionRef, std::vector<RelocationRef>> Ret; 1230 uint64_t I = (uint64_t)-1; 1231 for (SectionRef Sec : Obj.sections()) { 1232 ++I; 1233 Expected<section_iterator> RelocatedOrErr = Sec.getRelocatedSection(); 1234 if (!RelocatedOrErr) 1235 reportError(Obj.getFileName(), 1236 "section (" + Twine(I) + 1237 "): failed to get a relocated section: " + 1238 toString(RelocatedOrErr.takeError())); 1239 1240 section_iterator Relocated = *RelocatedOrErr; 1241 if (Relocated == Obj.section_end() || !checkSectionFilter(*Relocated).Keep) 1242 continue; 1243 std::vector<RelocationRef> &V = Ret[*Relocated]; 1244 append_range(V, Sec.relocations()); 1245 // Sort relocations by address. 1246 llvm::stable_sort(V, isRelocAddressLess); 1247 } 1248 return Ret; 1249 } 1250 1251 // Used for --adjust-vma to check if address should be adjusted by the 1252 // specified value for a given section. 1253 // For ELF we do not adjust non-allocatable sections like debug ones, 1254 // because they are not loadable. 1255 // TODO: implement for other file formats. 1256 static bool shouldAdjustVA(const SectionRef &Section) { 1257 const ObjectFile *Obj = Section.getObject(); 1258 if (Obj->isELF()) 1259 return ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC; 1260 return false; 1261 } 1262 1263 1264 typedef std::pair<uint64_t, char> MappingSymbolPair; 1265 static char getMappingSymbolKind(ArrayRef<MappingSymbolPair> MappingSymbols, 1266 uint64_t Address) { 1267 auto It = 1268 partition_point(MappingSymbols, [Address](const MappingSymbolPair &Val) { 1269 return Val.first <= Address; 1270 }); 1271 // Return zero for any address before the first mapping symbol; this means 1272 // we should use the default disassembly mode, depending on the target. 1273 if (It == MappingSymbols.begin()) 1274 return '\x00'; 1275 return (It - 1)->second; 1276 } 1277 1278 static uint64_t dumpARMELFData(uint64_t SectionAddr, uint64_t Index, 1279 uint64_t End, const ObjectFile &Obj, 1280 ArrayRef<uint8_t> Bytes, 1281 ArrayRef<MappingSymbolPair> MappingSymbols, 1282 const MCSubtargetInfo &STI, raw_ostream &OS) { 1283 llvm::endianness Endian = 1284 Obj.isLittleEndian() ? llvm::endianness::little : llvm::endianness::big; 1285 size_t Start = OS.tell(); 1286 OS << format("%8" PRIx64 ": ", SectionAddr + Index); 1287 if (Index + 4 <= End) { 1288 dumpBytes(Bytes.slice(Index, 4), OS); 1289 AlignToInstStartColumn(Start, STI, OS); 1290 OS << "\t.word\t" 1291 << format_hex(support::endian::read32(Bytes.data() + Index, Endian), 1292 10); 1293 return 4; 1294 } 1295 if (Index + 2 <= End) { 1296 dumpBytes(Bytes.slice(Index, 2), OS); 1297 AlignToInstStartColumn(Start, STI, OS); 1298 OS << "\t.short\t" 1299 << format_hex(support::endian::read16(Bytes.data() + Index, Endian), 6); 1300 return 2; 1301 } 1302 dumpBytes(Bytes.slice(Index, 1), OS); 1303 AlignToInstStartColumn(Start, STI, OS); 1304 OS << "\t.byte\t" << format_hex(Bytes[Index], 4); 1305 return 1; 1306 } 1307 1308 static void dumpELFData(uint64_t SectionAddr, uint64_t Index, uint64_t End, 1309 ArrayRef<uint8_t> Bytes) { 1310 // print out data up to 8 bytes at a time in hex and ascii 1311 uint8_t AsciiData[9] = {'\0'}; 1312 uint8_t Byte; 1313 int NumBytes = 0; 1314 1315 for (; Index < End; ++Index) { 1316 if (NumBytes == 0) 1317 outs() << format("%8" PRIx64 ":", SectionAddr + Index); 1318 Byte = Bytes.slice(Index)[0]; 1319 outs() << format(" %02x", Byte); 1320 AsciiData[NumBytes] = isPrint(Byte) ? Byte : '.'; 1321 1322 uint8_t IndentOffset = 0; 1323 NumBytes++; 1324 if (Index == End - 1 || NumBytes > 8) { 1325 // Indent the space for less than 8 bytes data. 1326 // 2 spaces for byte and one for space between bytes 1327 IndentOffset = 3 * (8 - NumBytes); 1328 for (int Excess = NumBytes; Excess < 8; Excess++) 1329 AsciiData[Excess] = '\0'; 1330 NumBytes = 8; 1331 } 1332 if (NumBytes == 8) { 1333 AsciiData[8] = '\0'; 1334 outs() << std::string(IndentOffset, ' ') << " "; 1335 outs() << reinterpret_cast<char *>(AsciiData); 1336 outs() << '\n'; 1337 NumBytes = 0; 1338 } 1339 } 1340 } 1341 1342 SymbolInfoTy objdump::createSymbolInfo(const ObjectFile &Obj, 1343 const SymbolRef &Symbol, 1344 bool IsMappingSymbol) { 1345 const StringRef FileName = Obj.getFileName(); 1346 const uint64_t Addr = unwrapOrError(Symbol.getAddress(), FileName); 1347 const StringRef Name = unwrapOrError(Symbol.getName(), FileName); 1348 1349 if (Obj.isXCOFF() && (SymbolDescription || TracebackTable)) { 1350 const auto &XCOFFObj = cast<XCOFFObjectFile>(Obj); 1351 DataRefImpl SymbolDRI = Symbol.getRawDataRefImpl(); 1352 1353 const uint32_t SymbolIndex = XCOFFObj.getSymbolIndex(SymbolDRI.p); 1354 std::optional<XCOFF::StorageMappingClass> Smc = 1355 getXCOFFSymbolCsectSMC(XCOFFObj, Symbol); 1356 return SymbolInfoTy(Smc, Addr, Name, SymbolIndex, 1357 isLabel(XCOFFObj, Symbol)); 1358 } else if (Obj.isXCOFF()) { 1359 const SymbolRef::Type SymType = unwrapOrError(Symbol.getType(), FileName); 1360 return SymbolInfoTy(Addr, Name, SymType, /*IsMappingSymbol=*/false, 1361 /*IsXCOFF=*/true); 1362 } else if (Obj.isWasm()) { 1363 uint8_t SymType = 1364 cast<WasmObjectFile>(&Obj)->getWasmSymbol(Symbol).Info.Kind; 1365 return SymbolInfoTy(Addr, Name, SymType, false); 1366 } else { 1367 uint8_t Type = 1368 Obj.isELF() ? getElfSymbolType(Obj, Symbol) : (uint8_t)ELF::STT_NOTYPE; 1369 return SymbolInfoTy(Addr, Name, Type, IsMappingSymbol); 1370 } 1371 } 1372 1373 static SymbolInfoTy createDummySymbolInfo(const ObjectFile &Obj, 1374 const uint64_t Addr, StringRef &Name, 1375 uint8_t Type) { 1376 if (Obj.isXCOFF() && (SymbolDescription || TracebackTable)) 1377 return SymbolInfoTy(std::nullopt, Addr, Name, std::nullopt, false); 1378 if (Obj.isWasm()) 1379 return SymbolInfoTy(Addr, Name, wasm::WASM_SYMBOL_TYPE_SECTION); 1380 return SymbolInfoTy(Addr, Name, Type); 1381 } 1382 1383 static void collectBBAddrMapLabels( 1384 const BBAddrMapInfo &FullAddrMap, uint64_t SectionAddr, uint64_t Start, 1385 uint64_t End, 1386 std::unordered_map<uint64_t, std::vector<BBAddrMapLabel>> &Labels) { 1387 if (FullAddrMap.empty()) 1388 return; 1389 Labels.clear(); 1390 uint64_t StartAddress = SectionAddr + Start; 1391 uint64_t EndAddress = SectionAddr + End; 1392 const BBAddrMapFunctionEntry *FunctionMap = 1393 FullAddrMap.getEntryForAddress(StartAddress); 1394 if (!FunctionMap) 1395 return; 1396 std::optional<size_t> BBRangeIndex = 1397 FunctionMap->getAddrMap().getBBRangeIndexForBaseAddress(StartAddress); 1398 if (!BBRangeIndex) 1399 return; 1400 size_t NumBBEntriesBeforeRange = 0; 1401 for (size_t I = 0; I < *BBRangeIndex; ++I) 1402 NumBBEntriesBeforeRange += 1403 FunctionMap->getAddrMap().BBRanges[I].BBEntries.size(); 1404 const auto &BBRange = FunctionMap->getAddrMap().BBRanges[*BBRangeIndex]; 1405 for (size_t I = 0; I < BBRange.BBEntries.size(); ++I) { 1406 const BBAddrMap::BBEntry &BBEntry = BBRange.BBEntries[I]; 1407 uint64_t BBAddress = BBEntry.Offset + BBRange.BaseAddress; 1408 if (BBAddress >= EndAddress) 1409 continue; 1410 1411 std::string LabelString = ("BB" + Twine(BBEntry.ID)).str(); 1412 Labels[BBAddress].push_back( 1413 {LabelString, 1414 FunctionMap->constructPGOLabelString(NumBBEntriesBeforeRange + I)}); 1415 } 1416 } 1417 1418 static void 1419 collectLocalBranchTargets(ArrayRef<uint8_t> Bytes, MCInstrAnalysis *MIA, 1420 MCDisassembler *DisAsm, MCInstPrinter *IP, 1421 const MCSubtargetInfo *STI, uint64_t SectionAddr, 1422 uint64_t Start, uint64_t End, 1423 std::unordered_map<uint64_t, std::string> &Labels) { 1424 // So far only supports PowerPC and X86. 1425 const bool isPPC = STI->getTargetTriple().isPPC(); 1426 if (!isPPC && !STI->getTargetTriple().isX86()) 1427 return; 1428 1429 if (MIA) 1430 MIA->resetState(); 1431 1432 Labels.clear(); 1433 unsigned LabelCount = 0; 1434 Start += SectionAddr; 1435 End += SectionAddr; 1436 const bool isXCOFF = STI->getTargetTriple().isOSBinFormatXCOFF(); 1437 for (uint64_t Index = Start; Index < End;) { 1438 // Disassemble a real instruction and record function-local branch labels. 1439 MCInst Inst; 1440 uint64_t Size; 1441 ArrayRef<uint8_t> ThisBytes = Bytes.slice(Index - SectionAddr); 1442 bool Disassembled = 1443 DisAsm->getInstruction(Inst, Size, ThisBytes, Index, nulls()); 1444 if (Size == 0) 1445 Size = std::min<uint64_t>(ThisBytes.size(), 1446 DisAsm->suggestBytesToSkip(ThisBytes, Index)); 1447 1448 if (MIA) { 1449 if (Disassembled) { 1450 uint64_t Target; 1451 bool TargetKnown = MIA->evaluateBranch(Inst, Index, Size, Target); 1452 if (TargetKnown && (Target >= Start && Target < End) && 1453 !Labels.count(Target)) { 1454 // On PowerPC and AIX, a function call is encoded as a branch to 0. 1455 // On other PowerPC platforms (ELF), a function call is encoded as 1456 // a branch to self. Do not add a label for these cases. 1457 if (!(isPPC && 1458 ((Target == 0 && isXCOFF) || (Target == Index && !isXCOFF)))) 1459 Labels[Target] = ("L" + Twine(LabelCount++)).str(); 1460 } 1461 MIA->updateState(Inst, Index); 1462 } else 1463 MIA->resetState(); 1464 } 1465 Index += Size; 1466 } 1467 } 1468 1469 // Create an MCSymbolizer for the target and add it to the MCDisassembler. 1470 // This is currently only used on AMDGPU, and assumes the format of the 1471 // void * argument passed to AMDGPU's createMCSymbolizer. 1472 static void addSymbolizer( 1473 MCContext &Ctx, const Target *Target, StringRef TripleName, 1474 MCDisassembler *DisAsm, uint64_t SectionAddr, ArrayRef<uint8_t> Bytes, 1475 SectionSymbolsTy &Symbols, 1476 std::vector<std::unique_ptr<std::string>> &SynthesizedLabelNames) { 1477 1478 std::unique_ptr<MCRelocationInfo> RelInfo( 1479 Target->createMCRelocationInfo(TripleName, Ctx)); 1480 if (!RelInfo) 1481 return; 1482 std::unique_ptr<MCSymbolizer> Symbolizer(Target->createMCSymbolizer( 1483 TripleName, nullptr, nullptr, &Symbols, &Ctx, std::move(RelInfo))); 1484 MCSymbolizer *SymbolizerPtr = &*Symbolizer; 1485 DisAsm->setSymbolizer(std::move(Symbolizer)); 1486 1487 if (!SymbolizeOperands) 1488 return; 1489 1490 // Synthesize labels referenced by branch instructions by 1491 // disassembling, discarding the output, and collecting the referenced 1492 // addresses from the symbolizer. 1493 for (size_t Index = 0; Index != Bytes.size();) { 1494 MCInst Inst; 1495 uint64_t Size; 1496 ArrayRef<uint8_t> ThisBytes = Bytes.slice(Index); 1497 const uint64_t ThisAddr = SectionAddr + Index; 1498 DisAsm->getInstruction(Inst, Size, ThisBytes, ThisAddr, nulls()); 1499 if (Size == 0) 1500 Size = std::min<uint64_t>(ThisBytes.size(), 1501 DisAsm->suggestBytesToSkip(ThisBytes, Index)); 1502 Index += Size; 1503 } 1504 ArrayRef<uint64_t> LabelAddrsRef = SymbolizerPtr->getReferencedAddresses(); 1505 // Copy and sort to remove duplicates. 1506 std::vector<uint64_t> LabelAddrs; 1507 LabelAddrs.insert(LabelAddrs.end(), LabelAddrsRef.begin(), 1508 LabelAddrsRef.end()); 1509 llvm::sort(LabelAddrs); 1510 LabelAddrs.resize(std::unique(LabelAddrs.begin(), LabelAddrs.end()) - 1511 LabelAddrs.begin()); 1512 // Add the labels. 1513 for (unsigned LabelNum = 0; LabelNum != LabelAddrs.size(); ++LabelNum) { 1514 auto Name = std::make_unique<std::string>(); 1515 *Name = (Twine("L") + Twine(LabelNum)).str(); 1516 SynthesizedLabelNames.push_back(std::move(Name)); 1517 Symbols.push_back(SymbolInfoTy( 1518 LabelAddrs[LabelNum], *SynthesizedLabelNames.back(), ELF::STT_NOTYPE)); 1519 } 1520 llvm::stable_sort(Symbols); 1521 // Recreate the symbolizer with the new symbols list. 1522 RelInfo.reset(Target->createMCRelocationInfo(TripleName, Ctx)); 1523 Symbolizer.reset(Target->createMCSymbolizer( 1524 TripleName, nullptr, nullptr, &Symbols, &Ctx, std::move(RelInfo))); 1525 DisAsm->setSymbolizer(std::move(Symbolizer)); 1526 } 1527 1528 static StringRef getSegmentName(const MachOObjectFile *MachO, 1529 const SectionRef &Section) { 1530 if (MachO) { 1531 DataRefImpl DR = Section.getRawDataRefImpl(); 1532 StringRef SegmentName = MachO->getSectionFinalSegmentName(DR); 1533 return SegmentName; 1534 } 1535 return ""; 1536 } 1537 1538 static void emitPostInstructionInfo(formatted_raw_ostream &FOS, 1539 const MCAsmInfo &MAI, 1540 const MCSubtargetInfo &STI, 1541 StringRef Comments, 1542 LiveVariablePrinter &LVP) { 1543 do { 1544 if (!Comments.empty()) { 1545 // Emit a line of comments. 1546 StringRef Comment; 1547 std::tie(Comment, Comments) = Comments.split('\n'); 1548 // MAI.getCommentColumn() assumes that instructions are printed at the 1549 // position of 8, while getInstStartColumn() returns the actual position. 1550 unsigned CommentColumn = 1551 MAI.getCommentColumn() - 8 + getInstStartColumn(STI); 1552 FOS.PadToColumn(CommentColumn); 1553 FOS << MAI.getCommentString() << ' ' << Comment; 1554 } 1555 LVP.printAfterInst(FOS); 1556 FOS << '\n'; 1557 } while (!Comments.empty()); 1558 FOS.flush(); 1559 } 1560 1561 static void createFakeELFSections(ObjectFile &Obj) { 1562 assert(Obj.isELF()); 1563 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj)) 1564 Elf32LEObj->createFakeSections(); 1565 else if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj)) 1566 Elf64LEObj->createFakeSections(); 1567 else if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj)) 1568 Elf32BEObj->createFakeSections(); 1569 else if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj)) 1570 Elf64BEObj->createFakeSections(); 1571 else 1572 llvm_unreachable("Unsupported binary format"); 1573 } 1574 1575 // Tries to fetch a more complete version of the given object file using its 1576 // Build ID. Returns std::nullopt if nothing was found. 1577 static std::optional<OwningBinary<Binary>> 1578 fetchBinaryByBuildID(const ObjectFile &Obj) { 1579 object::BuildIDRef BuildID = getBuildID(&Obj); 1580 if (BuildID.empty()) 1581 return std::nullopt; 1582 std::optional<std::string> Path = BIDFetcher->fetch(BuildID); 1583 if (!Path) 1584 return std::nullopt; 1585 Expected<OwningBinary<Binary>> DebugBinary = createBinary(*Path); 1586 if (!DebugBinary) { 1587 reportWarning(toString(DebugBinary.takeError()), *Path); 1588 return std::nullopt; 1589 } 1590 return std::move(*DebugBinary); 1591 } 1592 1593 static void 1594 disassembleObject(ObjectFile &Obj, const ObjectFile &DbgObj, 1595 DisassemblerTarget &PrimaryTarget, 1596 std::optional<DisassemblerTarget> &SecondaryTarget, 1597 SourcePrinter &SP, bool InlineRelocs) { 1598 DisassemblerTarget *DT = &PrimaryTarget; 1599 bool PrimaryIsThumb = false; 1600 SmallVector<std::pair<uint64_t, uint64_t>, 0> CHPECodeMap; 1601 1602 if (SecondaryTarget) { 1603 if (isArmElf(Obj)) { 1604 PrimaryIsThumb = 1605 PrimaryTarget.SubtargetInfo->checkFeatures("+thumb-mode"); 1606 } else if (const auto *COFFObj = dyn_cast<COFFObjectFile>(&Obj)) { 1607 const chpe_metadata *CHPEMetadata = COFFObj->getCHPEMetadata(); 1608 if (CHPEMetadata && CHPEMetadata->CodeMapCount) { 1609 uintptr_t CodeMapInt; 1610 cantFail(COFFObj->getRvaPtr(CHPEMetadata->CodeMap, CodeMapInt)); 1611 auto CodeMap = reinterpret_cast<const chpe_range_entry *>(CodeMapInt); 1612 1613 for (uint32_t i = 0; i < CHPEMetadata->CodeMapCount; ++i) { 1614 if (CodeMap[i].getType() == chpe_range_type::Amd64 && 1615 CodeMap[i].Length) { 1616 // Store x86_64 CHPE code ranges. 1617 uint64_t Start = CodeMap[i].getStart() + COFFObj->getImageBase(); 1618 CHPECodeMap.emplace_back(Start, Start + CodeMap[i].Length); 1619 } 1620 } 1621 llvm::sort(CHPECodeMap); 1622 } 1623 } 1624 } 1625 1626 std::map<SectionRef, std::vector<RelocationRef>> RelocMap; 1627 if (InlineRelocs || Obj.isXCOFF()) 1628 RelocMap = getRelocsMap(Obj); 1629 bool Is64Bits = Obj.getBytesInAddress() > 4; 1630 1631 // Create a mapping from virtual address to symbol name. This is used to 1632 // pretty print the symbols while disassembling. 1633 std::map<SectionRef, SectionSymbolsTy> AllSymbols; 1634 std::map<SectionRef, SmallVector<MappingSymbolPair, 0>> AllMappingSymbols; 1635 SectionSymbolsTy AbsoluteSymbols; 1636 const StringRef FileName = Obj.getFileName(); 1637 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(&Obj); 1638 for (const SymbolRef &Symbol : Obj.symbols()) { 1639 Expected<StringRef> NameOrErr = Symbol.getName(); 1640 if (!NameOrErr) { 1641 reportWarning(toString(NameOrErr.takeError()), FileName); 1642 continue; 1643 } 1644 if (NameOrErr->empty() && !(Obj.isXCOFF() && SymbolDescription)) 1645 continue; 1646 1647 if (Obj.isELF() && 1648 (cantFail(Symbol.getFlags()) & SymbolRef::SF_FormatSpecific)) { 1649 // Symbol is intended not to be displayed by default (STT_FILE, 1650 // STT_SECTION, or a mapping symbol). Ignore STT_SECTION symbols. We will 1651 // synthesize a section symbol if no symbol is defined at offset 0. 1652 // 1653 // For a mapping symbol, store it within both AllSymbols and 1654 // AllMappingSymbols. If --show-all-symbols is unspecified, its label will 1655 // not be printed in disassembly listing. 1656 if (getElfSymbolType(Obj, Symbol) != ELF::STT_SECTION && 1657 hasMappingSymbols(Obj)) { 1658 section_iterator SecI = unwrapOrError(Symbol.getSection(), FileName); 1659 if (SecI != Obj.section_end()) { 1660 uint64_t SectionAddr = SecI->getAddress(); 1661 uint64_t Address = cantFail(Symbol.getAddress()); 1662 StringRef Name = *NameOrErr; 1663 if (Name.consume_front("$") && Name.size() && 1664 strchr("adtx", Name[0])) { 1665 AllMappingSymbols[*SecI].emplace_back(Address - SectionAddr, 1666 Name[0]); 1667 AllSymbols[*SecI].push_back( 1668 createSymbolInfo(Obj, Symbol, /*MappingSymbol=*/true)); 1669 } 1670 } 1671 } 1672 continue; 1673 } 1674 1675 if (MachO) { 1676 // __mh_(execute|dylib|dylinker|bundle|preload|object)_header are special 1677 // symbols that support MachO header introspection. They do not bind to 1678 // code locations and are irrelevant for disassembly. 1679 if (NameOrErr->starts_with("__mh_") && NameOrErr->ends_with("_header")) 1680 continue; 1681 // Don't ask a Mach-O STAB symbol for its section unless you know that 1682 // STAB symbol's section field refers to a valid section index. Otherwise 1683 // the symbol may error trying to load a section that does not exist. 1684 DataRefImpl SymDRI = Symbol.getRawDataRefImpl(); 1685 uint8_t NType = (MachO->is64Bit() ? 1686 MachO->getSymbol64TableEntry(SymDRI).n_type: 1687 MachO->getSymbolTableEntry(SymDRI).n_type); 1688 if (NType & MachO::N_STAB) 1689 continue; 1690 } 1691 1692 section_iterator SecI = unwrapOrError(Symbol.getSection(), FileName); 1693 if (SecI != Obj.section_end()) 1694 AllSymbols[*SecI].push_back(createSymbolInfo(Obj, Symbol)); 1695 else 1696 AbsoluteSymbols.push_back(createSymbolInfo(Obj, Symbol)); 1697 } 1698 1699 if (AllSymbols.empty() && Obj.isELF()) 1700 addDynamicElfSymbols(cast<ELFObjectFileBase>(Obj), AllSymbols); 1701 1702 if (Obj.isWasm()) 1703 addMissingWasmCodeSymbols(cast<WasmObjectFile>(Obj), AllSymbols); 1704 1705 if (Obj.isELF() && Obj.sections().empty()) 1706 createFakeELFSections(Obj); 1707 1708 BumpPtrAllocator A; 1709 StringSaver Saver(A); 1710 addPltEntries(Obj, AllSymbols, Saver); 1711 1712 // Create a mapping from virtual address to section. An empty section can 1713 // cause more than one section at the same address. Sort such sections to be 1714 // before same-addressed non-empty sections so that symbol lookups prefer the 1715 // non-empty section. 1716 std::vector<std::pair<uint64_t, SectionRef>> SectionAddresses; 1717 for (SectionRef Sec : Obj.sections()) 1718 SectionAddresses.emplace_back(Sec.getAddress(), Sec); 1719 llvm::stable_sort(SectionAddresses, [](const auto &LHS, const auto &RHS) { 1720 if (LHS.first != RHS.first) 1721 return LHS.first < RHS.first; 1722 return LHS.second.getSize() < RHS.second.getSize(); 1723 }); 1724 1725 // Linked executables (.exe and .dll files) typically don't include a real 1726 // symbol table but they might contain an export table. 1727 if (const auto *COFFObj = dyn_cast<COFFObjectFile>(&Obj)) { 1728 for (const auto &ExportEntry : COFFObj->export_directories()) { 1729 StringRef Name; 1730 if (Error E = ExportEntry.getSymbolName(Name)) 1731 reportError(std::move(E), Obj.getFileName()); 1732 if (Name.empty()) 1733 continue; 1734 1735 uint32_t RVA; 1736 if (Error E = ExportEntry.getExportRVA(RVA)) 1737 reportError(std::move(E), Obj.getFileName()); 1738 1739 uint64_t VA = COFFObj->getImageBase() + RVA; 1740 auto Sec = partition_point( 1741 SectionAddresses, [VA](const std::pair<uint64_t, SectionRef> &O) { 1742 return O.first <= VA; 1743 }); 1744 if (Sec != SectionAddresses.begin()) { 1745 --Sec; 1746 AllSymbols[Sec->second].emplace_back(VA, Name, ELF::STT_NOTYPE); 1747 } else 1748 AbsoluteSymbols.emplace_back(VA, Name, ELF::STT_NOTYPE); 1749 } 1750 } 1751 1752 // Sort all the symbols, this allows us to use a simple binary search to find 1753 // Multiple symbols can have the same address. Use a stable sort to stabilize 1754 // the output. 1755 StringSet<> FoundDisasmSymbolSet; 1756 for (std::pair<const SectionRef, SectionSymbolsTy> &SecSyms : AllSymbols) 1757 llvm::stable_sort(SecSyms.second); 1758 llvm::stable_sort(AbsoluteSymbols); 1759 1760 std::unique_ptr<DWARFContext> DICtx; 1761 LiveVariablePrinter LVP(*DT->Context->getRegisterInfo(), *DT->SubtargetInfo); 1762 1763 if (DbgVariables != DVDisabled) { 1764 DICtx = DWARFContext::create(DbgObj); 1765 for (const std::unique_ptr<DWARFUnit> &CU : DICtx->compile_units()) 1766 LVP.addCompileUnit(CU->getUnitDIE(false)); 1767 } 1768 1769 LLVM_DEBUG(LVP.dump()); 1770 1771 BBAddrMapInfo FullAddrMap; 1772 auto ReadBBAddrMap = [&](std::optional<unsigned> SectionIndex = 1773 std::nullopt) { 1774 FullAddrMap.clear(); 1775 if (const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj)) { 1776 std::vector<PGOAnalysisMap> PGOAnalyses; 1777 auto BBAddrMapsOrErr = Elf->readBBAddrMap(SectionIndex, &PGOAnalyses); 1778 if (!BBAddrMapsOrErr) { 1779 reportWarning(toString(BBAddrMapsOrErr.takeError()), Obj.getFileName()); 1780 return; 1781 } 1782 for (auto &&[FunctionBBAddrMap, FunctionPGOAnalysis] : 1783 zip_equal(*std::move(BBAddrMapsOrErr), std::move(PGOAnalyses))) { 1784 FullAddrMap.AddFunctionEntry(std::move(FunctionBBAddrMap), 1785 std::move(FunctionPGOAnalysis)); 1786 } 1787 } 1788 }; 1789 1790 // For non-relocatable objects, Read all LLVM_BB_ADDR_MAP sections into a 1791 // single mapping, since they don't have any conflicts. 1792 if (SymbolizeOperands && !Obj.isRelocatableObject()) 1793 ReadBBAddrMap(); 1794 1795 std::optional<llvm::BTFParser> BTF; 1796 if (InlineRelocs && BTFParser::hasBTFSections(Obj)) { 1797 BTF.emplace(); 1798 BTFParser::ParseOptions Opts = {}; 1799 Opts.LoadTypes = true; 1800 Opts.LoadRelocs = true; 1801 if (Error E = BTF->parse(Obj, Opts)) 1802 WithColor::defaultErrorHandler(std::move(E)); 1803 } 1804 1805 for (const SectionRef &Section : ToolSectionFilter(Obj)) { 1806 if (FilterSections.empty() && !DisassembleAll && 1807 (!Section.isText() || Section.isVirtual())) 1808 continue; 1809 1810 uint64_t SectionAddr = Section.getAddress(); 1811 uint64_t SectSize = Section.getSize(); 1812 if (!SectSize) 1813 continue; 1814 1815 // For relocatable object files, read the LLVM_BB_ADDR_MAP section 1816 // corresponding to this section, if present. 1817 if (SymbolizeOperands && Obj.isRelocatableObject()) 1818 ReadBBAddrMap(Section.getIndex()); 1819 1820 // Get the list of all the symbols in this section. 1821 SectionSymbolsTy &Symbols = AllSymbols[Section]; 1822 auto &MappingSymbols = AllMappingSymbols[Section]; 1823 llvm::sort(MappingSymbols); 1824 1825 ArrayRef<uint8_t> Bytes = arrayRefFromStringRef( 1826 unwrapOrError(Section.getContents(), Obj.getFileName())); 1827 1828 std::vector<std::unique_ptr<std::string>> SynthesizedLabelNames; 1829 if (Obj.isELF() && Obj.getArch() == Triple::amdgcn) { 1830 // AMDGPU disassembler uses symbolizer for printing labels 1831 addSymbolizer(*DT->Context, DT->TheTarget, TripleName, DT->DisAsm.get(), 1832 SectionAddr, Bytes, Symbols, SynthesizedLabelNames); 1833 } 1834 1835 StringRef SegmentName = getSegmentName(MachO, Section); 1836 StringRef SectionName = unwrapOrError(Section.getName(), Obj.getFileName()); 1837 // If the section has no symbol at the start, just insert a dummy one. 1838 // Without --show-all-symbols, also insert one if all symbols at the start 1839 // are mapping symbols. 1840 bool CreateDummy = Symbols.empty(); 1841 if (!CreateDummy) { 1842 CreateDummy = true; 1843 for (auto &Sym : Symbols) { 1844 if (Sym.Addr != SectionAddr) 1845 break; 1846 if (!Sym.IsMappingSymbol || ShowAllSymbols) 1847 CreateDummy = false; 1848 } 1849 } 1850 if (CreateDummy) { 1851 SymbolInfoTy Sym = createDummySymbolInfo( 1852 Obj, SectionAddr, SectionName, 1853 Section.isText() ? ELF::STT_FUNC : ELF::STT_OBJECT); 1854 if (Obj.isXCOFF()) 1855 Symbols.insert(Symbols.begin(), Sym); 1856 else 1857 Symbols.insert(llvm::lower_bound(Symbols, Sym), Sym); 1858 } 1859 1860 SmallString<40> Comments; 1861 raw_svector_ostream CommentStream(Comments); 1862 1863 uint64_t VMAAdjustment = 0; 1864 if (shouldAdjustVA(Section)) 1865 VMAAdjustment = AdjustVMA; 1866 1867 // In executable and shared objects, r_offset holds a virtual address. 1868 // Subtract SectionAddr from the r_offset field of a relocation to get 1869 // the section offset. 1870 uint64_t RelAdjustment = Obj.isRelocatableObject() ? 0 : SectionAddr; 1871 uint64_t Size; 1872 uint64_t Index; 1873 bool PrintedSection = false; 1874 std::vector<RelocationRef> Rels = RelocMap[Section]; 1875 std::vector<RelocationRef>::const_iterator RelCur = Rels.begin(); 1876 std::vector<RelocationRef>::const_iterator RelEnd = Rels.end(); 1877 1878 // Loop over each chunk of code between two points where at least 1879 // one symbol is defined. 1880 for (size_t SI = 0, SE = Symbols.size(); SI != SE;) { 1881 // Advance SI past all the symbols starting at the same address, 1882 // and make an ArrayRef of them. 1883 unsigned FirstSI = SI; 1884 uint64_t Start = Symbols[SI].Addr; 1885 ArrayRef<SymbolInfoTy> SymbolsHere; 1886 while (SI != SE && Symbols[SI].Addr == Start) 1887 ++SI; 1888 SymbolsHere = ArrayRef<SymbolInfoTy>(&Symbols[FirstSI], SI - FirstSI); 1889 1890 // Get the demangled names of all those symbols. We end up with a vector 1891 // of StringRef that holds the names we're going to use, and a vector of 1892 // std::string that stores the new strings returned by demangle(), if 1893 // any. If we don't call demangle() then that vector can stay empty. 1894 std::vector<StringRef> SymNamesHere; 1895 std::vector<std::string> DemangledSymNamesHere; 1896 if (Demangle) { 1897 // Fetch the demangled names and store them locally. 1898 for (const SymbolInfoTy &Symbol : SymbolsHere) 1899 DemangledSymNamesHere.push_back(demangle(Symbol.Name)); 1900 // Now we've finished modifying that vector, it's safe to make 1901 // a vector of StringRefs pointing into it. 1902 SymNamesHere.insert(SymNamesHere.begin(), DemangledSymNamesHere.begin(), 1903 DemangledSymNamesHere.end()); 1904 } else { 1905 for (const SymbolInfoTy &Symbol : SymbolsHere) 1906 SymNamesHere.push_back(Symbol.Name); 1907 } 1908 1909 // Distinguish ELF data from code symbols, which will be used later on to 1910 // decide whether to 'disassemble' this chunk as a data declaration via 1911 // dumpELFData(), or whether to treat it as code. 1912 // 1913 // If data _and_ code symbols are defined at the same address, the code 1914 // takes priority, on the grounds that disassembling code is our main 1915 // purpose here, and it would be a worse failure to _not_ interpret 1916 // something that _was_ meaningful as code than vice versa. 1917 // 1918 // Any ELF symbol type that is not clearly data will be regarded as code. 1919 // In particular, one of the uses of STT_NOTYPE is for branch targets 1920 // inside functions, for which STT_FUNC would be inaccurate. 1921 // 1922 // So here, we spot whether there's any non-data symbol present at all, 1923 // and only set the DisassembleAsELFData flag if there isn't. Also, we use 1924 // this distinction to inform the decision of which symbol to print at 1925 // the head of the section, so that if we're printing code, we print a 1926 // code-related symbol name to go with it. 1927 bool DisassembleAsELFData = false; 1928 size_t DisplaySymIndex = SymbolsHere.size() - 1; 1929 if (Obj.isELF() && !DisassembleAll && Section.isText()) { 1930 DisassembleAsELFData = true; // unless we find a code symbol below 1931 1932 for (size_t i = 0; i < SymbolsHere.size(); ++i) { 1933 uint8_t SymTy = SymbolsHere[i].Type; 1934 if (SymTy != ELF::STT_OBJECT && SymTy != ELF::STT_COMMON) { 1935 DisassembleAsELFData = false; 1936 DisplaySymIndex = i; 1937 } 1938 } 1939 } 1940 1941 // Decide which symbol(s) from this collection we're going to print. 1942 std::vector<bool> SymsToPrint(SymbolsHere.size(), false); 1943 // If the user has given the --disassemble-symbols option, then we must 1944 // display every symbol in that set, and no others. 1945 if (!DisasmSymbolSet.empty()) { 1946 bool FoundAny = false; 1947 for (size_t i = 0; i < SymbolsHere.size(); ++i) { 1948 if (DisasmSymbolSet.count(SymNamesHere[i])) { 1949 SymsToPrint[i] = true; 1950 FoundAny = true; 1951 } 1952 } 1953 1954 // And if none of the symbols here is one that the user asked for, skip 1955 // disassembling this entire chunk of code. 1956 if (!FoundAny) 1957 continue; 1958 } else if (!SymbolsHere[DisplaySymIndex].IsMappingSymbol) { 1959 // Otherwise, print whichever symbol at this location is last in the 1960 // Symbols array, because that array is pre-sorted in a way intended to 1961 // correlate with priority of which symbol to display. 1962 SymsToPrint[DisplaySymIndex] = true; 1963 } 1964 1965 // Now that we know we're disassembling this section, override the choice 1966 // of which symbols to display by printing _all_ of them at this address 1967 // if the user asked for all symbols. 1968 // 1969 // That way, '--show-all-symbols --disassemble-symbol=foo' will print 1970 // only the chunk of code headed by 'foo', but also show any other 1971 // symbols defined at that address, such as aliases for 'foo', or the ARM 1972 // mapping symbol preceding its code. 1973 if (ShowAllSymbols) { 1974 for (size_t i = 0; i < SymbolsHere.size(); ++i) 1975 SymsToPrint[i] = true; 1976 } 1977 1978 if (Start < SectionAddr || StopAddress <= Start) 1979 continue; 1980 1981 for (size_t i = 0; i < SymbolsHere.size(); ++i) 1982 FoundDisasmSymbolSet.insert(SymNamesHere[i]); 1983 1984 // The end is the section end, the beginning of the next symbol, or 1985 // --stop-address. 1986 uint64_t End = std::min<uint64_t>(SectionAddr + SectSize, StopAddress); 1987 if (SI < SE) 1988 End = std::min(End, Symbols[SI].Addr); 1989 if (Start >= End || End <= StartAddress) 1990 continue; 1991 Start -= SectionAddr; 1992 End -= SectionAddr; 1993 1994 if (!PrintedSection) { 1995 PrintedSection = true; 1996 outs() << "\nDisassembly of section "; 1997 if (!SegmentName.empty()) 1998 outs() << SegmentName << ","; 1999 outs() << SectionName << ":\n"; 2000 } 2001 2002 bool PrintedLabel = false; 2003 for (size_t i = 0; i < SymbolsHere.size(); ++i) { 2004 if (!SymsToPrint[i]) 2005 continue; 2006 2007 const SymbolInfoTy &Symbol = SymbolsHere[i]; 2008 const StringRef SymbolName = SymNamesHere[i]; 2009 2010 if (!PrintedLabel) { 2011 outs() << '\n'; 2012 PrintedLabel = true; 2013 } 2014 if (LeadingAddr) 2015 outs() << format(Is64Bits ? "%016" PRIx64 " " : "%08" PRIx64 " ", 2016 SectionAddr + Start + VMAAdjustment); 2017 if (Obj.isXCOFF() && SymbolDescription) { 2018 outs() << getXCOFFSymbolDescription(Symbol, SymbolName) << ":\n"; 2019 } else 2020 outs() << '<' << SymbolName << ">:\n"; 2021 } 2022 2023 // Don't print raw contents of a virtual section. A virtual section 2024 // doesn't have any contents in the file. 2025 if (Section.isVirtual()) { 2026 outs() << "...\n"; 2027 continue; 2028 } 2029 2030 // See if any of the symbols defined at this location triggers target- 2031 // specific disassembly behavior, e.g. of special descriptors or function 2032 // prelude information. 2033 // 2034 // We stop this loop at the first symbol that triggers some kind of 2035 // interesting behavior (if any), on the assumption that if two symbols 2036 // defined at the same address trigger two conflicting symbol handlers, 2037 // the object file is probably confused anyway, and it would make even 2038 // less sense to present the output of _both_ handlers, because that 2039 // would describe the same data twice. 2040 for (size_t SHI = 0; SHI < SymbolsHere.size(); ++SHI) { 2041 SymbolInfoTy Symbol = SymbolsHere[SHI]; 2042 2043 auto Status = DT->DisAsm->onSymbolStart( 2044 Symbol, Size, Bytes.slice(Start, End - Start), SectionAddr + Start, 2045 CommentStream); 2046 2047 if (!Status) { 2048 // If onSymbolStart returns std::nullopt, that means it didn't trigger 2049 // any interesting handling for this symbol. Try the other symbols 2050 // defined at this address. 2051 continue; 2052 } 2053 2054 if (*Status == MCDisassembler::Fail) { 2055 // If onSymbolStart returns Fail, that means it identified some kind 2056 // of special data at this address, but wasn't able to disassemble it 2057 // meaningfully. So we fall back to disassembling the failed region 2058 // as bytes, assuming that the target detected the failure before 2059 // printing anything. 2060 // 2061 // Return values Success or SoftFail (i.e no 'real' failure) are 2062 // expected to mean that the target has emitted its own output. 2063 // 2064 // Either way, 'Size' will have been set to the amount of data 2065 // covered by whatever prologue the target identified. So we advance 2066 // our own position to beyond that. Sometimes that will be the entire 2067 // distance to the next symbol, and sometimes it will be just a 2068 // prologue and we should start disassembling instructions from where 2069 // it left off. 2070 outs() << DT->Context->getAsmInfo()->getCommentString() 2071 << " error in decoding " << SymNamesHere[SHI] 2072 << " : decoding failed region as bytes.\n"; 2073 for (uint64_t I = 0; I < Size; ++I) { 2074 outs() << "\t.byte\t " << format_hex(Bytes[I], 1, /*Upper=*/true) 2075 << "\n"; 2076 } 2077 } 2078 Start += Size; 2079 break; 2080 } 2081 2082 Index = Start; 2083 if (SectionAddr < StartAddress) 2084 Index = std::max<uint64_t>(Index, StartAddress - SectionAddr); 2085 2086 if (DisassembleAsELFData) { 2087 dumpELFData(SectionAddr, Index, End, Bytes); 2088 Index = End; 2089 continue; 2090 } 2091 2092 // Skip relocations from symbols that are not dumped. 2093 for (; RelCur != RelEnd; ++RelCur) { 2094 uint64_t Offset = RelCur->getOffset() - RelAdjustment; 2095 if (Index <= Offset) 2096 break; 2097 } 2098 2099 bool DumpARMELFData = false; 2100 bool DumpTracebackTableForXCOFFFunction = 2101 Obj.isXCOFF() && Section.isText() && TracebackTable && 2102 Symbols[SI - 1].XCOFFSymInfo.StorageMappingClass && 2103 (*Symbols[SI - 1].XCOFFSymInfo.StorageMappingClass == XCOFF::XMC_PR); 2104 2105 formatted_raw_ostream FOS(outs()); 2106 2107 // FIXME: Workaround for bug in formatted_raw_ostream. Color escape codes 2108 // are (incorrectly) written directly to the unbuffered raw_ostream 2109 // wrapped by the formatted_raw_ostream. 2110 if (DisassemblyColor == ColorOutput::Enable || 2111 DisassemblyColor == ColorOutput::Auto) 2112 FOS.SetUnbuffered(); 2113 2114 std::unordered_map<uint64_t, std::string> AllLabels; 2115 std::unordered_map<uint64_t, std::vector<BBAddrMapLabel>> BBAddrMapLabels; 2116 if (SymbolizeOperands) { 2117 collectLocalBranchTargets(Bytes, DT->InstrAnalysis.get(), 2118 DT->DisAsm.get(), DT->InstPrinter.get(), 2119 PrimaryTarget.SubtargetInfo.get(), 2120 SectionAddr, Index, End, AllLabels); 2121 collectBBAddrMapLabels(FullAddrMap, SectionAddr, Index, End, 2122 BBAddrMapLabels); 2123 } 2124 2125 if (DT->InstrAnalysis) 2126 DT->InstrAnalysis->resetState(); 2127 2128 while (Index < End) { 2129 uint64_t RelOffset; 2130 2131 // ARM and AArch64 ELF binaries can interleave data and text in the 2132 // same section. We rely on the markers introduced to understand what 2133 // we need to dump. If the data marker is within a function, it is 2134 // denoted as a word/short etc. 2135 if (!MappingSymbols.empty()) { 2136 char Kind = getMappingSymbolKind(MappingSymbols, Index); 2137 DumpARMELFData = Kind == 'd'; 2138 if (SecondaryTarget) { 2139 if (Kind == 'a') { 2140 DT = PrimaryIsThumb ? &*SecondaryTarget : &PrimaryTarget; 2141 } else if (Kind == 't') { 2142 DT = PrimaryIsThumb ? &PrimaryTarget : &*SecondaryTarget; 2143 } 2144 } 2145 } else if (!CHPECodeMap.empty()) { 2146 uint64_t Address = SectionAddr + Index; 2147 auto It = partition_point( 2148 CHPECodeMap, 2149 [Address](const std::pair<uint64_t, uint64_t> &Entry) { 2150 return Entry.first <= Address; 2151 }); 2152 if (It != CHPECodeMap.begin() && Address < (It - 1)->second) { 2153 DT = &*SecondaryTarget; 2154 } else { 2155 DT = &PrimaryTarget; 2156 // X64 disassembler range may have left Index unaligned, so 2157 // make sure that it's aligned when we switch back to ARM64 2158 // code. 2159 Index = llvm::alignTo(Index, 4); 2160 if (Index >= End) 2161 break; 2162 } 2163 } 2164 2165 auto findRel = [&]() { 2166 while (RelCur != RelEnd) { 2167 RelOffset = RelCur->getOffset() - RelAdjustment; 2168 // If this relocation is hidden, skip it. 2169 if (getHidden(*RelCur) || SectionAddr + RelOffset < StartAddress) { 2170 ++RelCur; 2171 continue; 2172 } 2173 2174 // Stop when RelCur's offset is past the disassembled 2175 // instruction/data. 2176 if (RelOffset >= Index + Size) 2177 return false; 2178 if (RelOffset >= Index) 2179 return true; 2180 ++RelCur; 2181 } 2182 return false; 2183 }; 2184 2185 if (DumpARMELFData) { 2186 Size = dumpARMELFData(SectionAddr, Index, End, Obj, Bytes, 2187 MappingSymbols, *DT->SubtargetInfo, FOS); 2188 } else { 2189 // When -z or --disassemble-zeroes are given we always dissasemble 2190 // them. Otherwise we might want to skip zero bytes we see. 2191 if (!DisassembleZeroes) { 2192 uint64_t MaxOffset = End - Index; 2193 // For --reloc: print zero blocks patched by relocations, so that 2194 // relocations can be shown in the dump. 2195 if (InlineRelocs && RelCur != RelEnd) 2196 MaxOffset = std::min(RelCur->getOffset() - RelAdjustment - Index, 2197 MaxOffset); 2198 2199 if (size_t N = 2200 countSkippableZeroBytes(Bytes.slice(Index, MaxOffset))) { 2201 FOS << "\t\t..." << '\n'; 2202 Index += N; 2203 continue; 2204 } 2205 } 2206 2207 if (DumpTracebackTableForXCOFFFunction && 2208 doesXCOFFTracebackTableBegin(Bytes.slice(Index, 4))) { 2209 dumpTracebackTable(Bytes.slice(Index), 2210 SectionAddr + Index + VMAAdjustment, FOS, 2211 SectionAddr + End + VMAAdjustment, 2212 *DT->SubtargetInfo, cast<XCOFFObjectFile>(&Obj)); 2213 Index = End; 2214 continue; 2215 } 2216 2217 // Print local label if there's any. 2218 auto Iter1 = BBAddrMapLabels.find(SectionAddr + Index); 2219 if (Iter1 != BBAddrMapLabels.end()) { 2220 for (const auto &BBLabel : Iter1->second) 2221 FOS << "<" << BBLabel.BlockLabel << ">" << BBLabel.PGOAnalysis 2222 << ":\n"; 2223 } else { 2224 auto Iter2 = AllLabels.find(SectionAddr + Index); 2225 if (Iter2 != AllLabels.end()) 2226 FOS << "<" << Iter2->second << ">:\n"; 2227 } 2228 2229 // Disassemble a real instruction or a data when disassemble all is 2230 // provided 2231 MCInst Inst; 2232 ArrayRef<uint8_t> ThisBytes = Bytes.slice(Index); 2233 uint64_t ThisAddr = SectionAddr + Index; 2234 bool Disassembled = DT->DisAsm->getInstruction( 2235 Inst, Size, ThisBytes, ThisAddr, CommentStream); 2236 if (Size == 0) 2237 Size = std::min<uint64_t>( 2238 ThisBytes.size(), 2239 DT->DisAsm->suggestBytesToSkip(ThisBytes, ThisAddr)); 2240 2241 LVP.update({Index, Section.getIndex()}, 2242 {Index + Size, Section.getIndex()}, Index + Size != End); 2243 2244 DT->InstPrinter->setCommentStream(CommentStream); 2245 2246 DT->Printer->printInst( 2247 *DT->InstPrinter, Disassembled ? &Inst : nullptr, 2248 Bytes.slice(Index, Size), 2249 {SectionAddr + Index + VMAAdjustment, Section.getIndex()}, FOS, 2250 "", *DT->SubtargetInfo, &SP, Obj.getFileName(), &Rels, LVP); 2251 2252 DT->InstPrinter->setCommentStream(llvm::nulls()); 2253 2254 // If disassembly succeeds, we try to resolve the target address 2255 // (jump target or memory operand address) and print it to the 2256 // right of the instruction. 2257 // 2258 // Otherwise, we don't print anything else so that we avoid 2259 // analyzing invalid or incomplete instruction information. 2260 if (Disassembled && DT->InstrAnalysis) { 2261 llvm::raw_ostream *TargetOS = &FOS; 2262 uint64_t Target; 2263 bool PrintTarget = DT->InstrAnalysis->evaluateBranch( 2264 Inst, SectionAddr + Index, Size, Target); 2265 2266 if (!PrintTarget) { 2267 if (std::optional<uint64_t> MaybeTarget = 2268 DT->InstrAnalysis->evaluateMemoryOperandAddress( 2269 Inst, DT->SubtargetInfo.get(), SectionAddr + Index, 2270 Size)) { 2271 Target = *MaybeTarget; 2272 PrintTarget = true; 2273 // Do not print real address when symbolizing. 2274 if (!SymbolizeOperands) { 2275 // Memory operand addresses are printed as comments. 2276 TargetOS = &CommentStream; 2277 *TargetOS << "0x" << Twine::utohexstr(Target); 2278 } 2279 } 2280 } 2281 2282 if (PrintTarget) { 2283 // In a relocatable object, the target's section must reside in 2284 // the same section as the call instruction or it is accessed 2285 // through a relocation. 2286 // 2287 // In a non-relocatable object, the target may be in any section. 2288 // In that case, locate the section(s) containing the target 2289 // address and find the symbol in one of those, if possible. 2290 // 2291 // N.B. Except for XCOFF, we don't walk the relocations in the 2292 // relocatable case yet. 2293 std::vector<const SectionSymbolsTy *> TargetSectionSymbols; 2294 if (!Obj.isRelocatableObject()) { 2295 auto It = llvm::partition_point( 2296 SectionAddresses, 2297 [=](const std::pair<uint64_t, SectionRef> &O) { 2298 return O.first <= Target; 2299 }); 2300 uint64_t TargetSecAddr = 0; 2301 while (It != SectionAddresses.begin()) { 2302 --It; 2303 if (TargetSecAddr == 0) 2304 TargetSecAddr = It->first; 2305 if (It->first != TargetSecAddr) 2306 break; 2307 TargetSectionSymbols.push_back(&AllSymbols[It->second]); 2308 } 2309 } else { 2310 TargetSectionSymbols.push_back(&Symbols); 2311 } 2312 TargetSectionSymbols.push_back(&AbsoluteSymbols); 2313 2314 // Find the last symbol in the first candidate section whose 2315 // offset is less than or equal to the target. If there are no 2316 // such symbols, try in the next section and so on, before finally 2317 // using the nearest preceding absolute symbol (if any), if there 2318 // are no other valid symbols. 2319 const SymbolInfoTy *TargetSym = nullptr; 2320 for (const SectionSymbolsTy *TargetSymbols : 2321 TargetSectionSymbols) { 2322 auto It = llvm::partition_point( 2323 *TargetSymbols, 2324 [=](const SymbolInfoTy &O) { return O.Addr <= Target; }); 2325 while (It != TargetSymbols->begin()) { 2326 --It; 2327 // Skip mapping symbols to avoid possible ambiguity as they 2328 // do not allow uniquely identifying the target address. 2329 if (!It->IsMappingSymbol) { 2330 TargetSym = &*It; 2331 break; 2332 } 2333 } 2334 if (TargetSym) 2335 break; 2336 } 2337 2338 // Branch targets are printed just after the instructions. 2339 // Print the labels corresponding to the target if there's any. 2340 bool BBAddrMapLabelAvailable = BBAddrMapLabels.count(Target); 2341 bool LabelAvailable = AllLabels.count(Target); 2342 2343 if (TargetSym != nullptr) { 2344 uint64_t TargetAddress = TargetSym->Addr; 2345 uint64_t Disp = Target - TargetAddress; 2346 std::string TargetName = Demangle ? demangle(TargetSym->Name) 2347 : TargetSym->Name.str(); 2348 bool RelFixedUp = false; 2349 SmallString<32> Val; 2350 2351 *TargetOS << " <"; 2352 // On XCOFF, we use relocations, even without -r, so we 2353 // can print the correct name for an extern function call. 2354 if (Obj.isXCOFF() && findRel()) { 2355 // Check for possible branch relocations and 2356 // branches to fixup code. 2357 bool BranchRelocationType = true; 2358 XCOFF::RelocationType RelocType; 2359 if (Obj.is64Bit()) { 2360 const XCOFFRelocation64 *Reloc = 2361 reinterpret_cast<XCOFFRelocation64 *>( 2362 RelCur->getRawDataRefImpl().p); 2363 RelFixedUp = Reloc->isFixupIndicated(); 2364 RelocType = Reloc->Type; 2365 } else { 2366 const XCOFFRelocation32 *Reloc = 2367 reinterpret_cast<XCOFFRelocation32 *>( 2368 RelCur->getRawDataRefImpl().p); 2369 RelFixedUp = Reloc->isFixupIndicated(); 2370 RelocType = Reloc->Type; 2371 } 2372 BranchRelocationType = 2373 RelocType == XCOFF::R_BA || RelocType == XCOFF::R_BR || 2374 RelocType == XCOFF::R_RBA || RelocType == XCOFF::R_RBR; 2375 2376 // If we have a valid relocation, try to print its 2377 // corresponding symbol name. Multiple relocations on the 2378 // same instruction are not handled. 2379 // Branches to fixup code will have the RelFixedUp flag set in 2380 // the RLD. For these instructions, we print the correct 2381 // branch target, but print the referenced symbol as a 2382 // comment. 2383 if (Error E = getRelocationValueString(*RelCur, false, Val)) { 2384 // If -r was used, this error will be printed later. 2385 // Otherwise, we ignore the error and print what 2386 // would have been printed without using relocations. 2387 consumeError(std::move(E)); 2388 *TargetOS << TargetName; 2389 RelFixedUp = false; // Suppress comment for RLD sym name 2390 } else if (BranchRelocationType && !RelFixedUp) 2391 *TargetOS << Val; 2392 else 2393 *TargetOS << TargetName; 2394 if (Disp) 2395 *TargetOS << "+0x" << Twine::utohexstr(Disp); 2396 } else if (!Disp) { 2397 *TargetOS << TargetName; 2398 } else if (BBAddrMapLabelAvailable) { 2399 *TargetOS << BBAddrMapLabels[Target].front().BlockLabel; 2400 } else if (LabelAvailable) { 2401 *TargetOS << AllLabels[Target]; 2402 } else { 2403 // Always Print the binary symbol plus an offset if there's no 2404 // local label corresponding to the target address. 2405 *TargetOS << TargetName << "+0x" << Twine::utohexstr(Disp); 2406 } 2407 *TargetOS << ">"; 2408 if (RelFixedUp && !InlineRelocs) { 2409 // We have fixup code for a relocation. We print the 2410 // referenced symbol as a comment. 2411 *TargetOS << "\t# " << Val; 2412 } 2413 2414 } else if (BBAddrMapLabelAvailable) { 2415 *TargetOS << " <" << BBAddrMapLabels[Target].front().BlockLabel 2416 << ">"; 2417 } else if (LabelAvailable) { 2418 *TargetOS << " <" << AllLabels[Target] << ">"; 2419 } 2420 // By convention, each record in the comment stream should be 2421 // terminated. 2422 if (TargetOS == &CommentStream) 2423 *TargetOS << "\n"; 2424 } 2425 2426 DT->InstrAnalysis->updateState(Inst, SectionAddr + Index); 2427 } else if (!Disassembled && DT->InstrAnalysis) { 2428 DT->InstrAnalysis->resetState(); 2429 } 2430 } 2431 2432 assert(DT->Context->getAsmInfo()); 2433 emitPostInstructionInfo(FOS, *DT->Context->getAsmInfo(), 2434 *DT->SubtargetInfo, CommentStream.str(), LVP); 2435 Comments.clear(); 2436 2437 if (BTF) 2438 printBTFRelocation(FOS, *BTF, {Index, Section.getIndex()}, LVP); 2439 2440 // Hexagon handles relocs in pretty printer 2441 if (InlineRelocs && Obj.getArch() != Triple::hexagon) { 2442 while (findRel()) { 2443 // When --adjust-vma is used, update the address printed. 2444 if (RelCur->getSymbol() != Obj.symbol_end()) { 2445 Expected<section_iterator> SymSI = 2446 RelCur->getSymbol()->getSection(); 2447 if (SymSI && *SymSI != Obj.section_end() && 2448 shouldAdjustVA(**SymSI)) 2449 RelOffset += AdjustVMA; 2450 } 2451 2452 printRelocation(FOS, Obj.getFileName(), *RelCur, 2453 SectionAddr + RelOffset, Is64Bits); 2454 LVP.printAfterOtherLine(FOS, true); 2455 ++RelCur; 2456 } 2457 } 2458 2459 Index += Size; 2460 } 2461 } 2462 } 2463 StringSet<> MissingDisasmSymbolSet = 2464 set_difference(DisasmSymbolSet, FoundDisasmSymbolSet); 2465 for (StringRef Sym : MissingDisasmSymbolSet.keys()) 2466 reportWarning("failed to disassemble missing symbol " + Sym, FileName); 2467 } 2468 2469 static void disassembleObject(ObjectFile *Obj, bool InlineRelocs) { 2470 // If information useful for showing the disassembly is missing, try to find a 2471 // more complete binary and disassemble that instead. 2472 OwningBinary<Binary> FetchedBinary; 2473 if (Obj->symbols().empty()) { 2474 if (std::optional<OwningBinary<Binary>> FetchedBinaryOpt = 2475 fetchBinaryByBuildID(*Obj)) { 2476 if (auto *O = dyn_cast<ObjectFile>(FetchedBinaryOpt->getBinary())) { 2477 if (!O->symbols().empty() || 2478 (!O->sections().empty() && Obj->sections().empty())) { 2479 FetchedBinary = std::move(*FetchedBinaryOpt); 2480 Obj = O; 2481 } 2482 } 2483 } 2484 } 2485 2486 const Target *TheTarget = getTarget(Obj); 2487 2488 // Package up features to be passed to target/subtarget 2489 Expected<SubtargetFeatures> FeaturesValue = Obj->getFeatures(); 2490 if (!FeaturesValue) 2491 reportError(FeaturesValue.takeError(), Obj->getFileName()); 2492 SubtargetFeatures Features = *FeaturesValue; 2493 if (!MAttrs.empty()) { 2494 for (unsigned I = 0; I != MAttrs.size(); ++I) 2495 Features.AddFeature(MAttrs[I]); 2496 } else if (MCPU.empty() && Obj->getArch() == llvm::Triple::aarch64) { 2497 Features.AddFeature("+all"); 2498 } 2499 2500 if (MCPU.empty()) 2501 MCPU = Obj->tryGetCPUName().value_or("").str(); 2502 2503 if (isArmElf(*Obj)) { 2504 // When disassembling big-endian Arm ELF, the instruction endianness is 2505 // determined in a complex way. In relocatable objects, AAELF32 mandates 2506 // that instruction endianness matches the ELF file endianness; in 2507 // executable images, that's true unless the file header has the EF_ARM_BE8 2508 // flag, in which case instructions are little-endian regardless of data 2509 // endianness. 2510 // 2511 // We must set the big-endian-instructions SubtargetFeature to make the 2512 // disassembler read the instructions the right way round, and also tell 2513 // our own prettyprinter to retrieve the encodings the same way to print in 2514 // hex. 2515 const auto *Elf32BE = dyn_cast<ELF32BEObjectFile>(Obj); 2516 2517 if (Elf32BE && (Elf32BE->isRelocatableObject() || 2518 !(Elf32BE->getPlatformFlags() & ELF::EF_ARM_BE8))) { 2519 Features.AddFeature("+big-endian-instructions"); 2520 ARMPrettyPrinterInst.setInstructionEndianness(llvm::endianness::big); 2521 } else { 2522 ARMPrettyPrinterInst.setInstructionEndianness(llvm::endianness::little); 2523 } 2524 } 2525 2526 DisassemblerTarget PrimaryTarget(TheTarget, *Obj, TripleName, MCPU, Features); 2527 2528 // If we have an ARM object file, we need a second disassembler, because 2529 // ARM CPUs have two different instruction sets: ARM mode, and Thumb mode. 2530 // We use mapping symbols to switch between the two assemblers, where 2531 // appropriate. 2532 std::optional<DisassemblerTarget> SecondaryTarget; 2533 2534 if (isArmElf(*Obj)) { 2535 if (!PrimaryTarget.SubtargetInfo->checkFeatures("+mclass")) { 2536 if (PrimaryTarget.SubtargetInfo->checkFeatures("+thumb-mode")) 2537 Features.AddFeature("-thumb-mode"); 2538 else 2539 Features.AddFeature("+thumb-mode"); 2540 SecondaryTarget.emplace(PrimaryTarget, Features); 2541 } 2542 } else if (const auto *COFFObj = dyn_cast<COFFObjectFile>(Obj)) { 2543 const chpe_metadata *CHPEMetadata = COFFObj->getCHPEMetadata(); 2544 if (CHPEMetadata && CHPEMetadata->CodeMapCount) { 2545 // Set up x86_64 disassembler for ARM64EC binaries. 2546 Triple X64Triple(TripleName); 2547 X64Triple.setArch(Triple::ArchType::x86_64); 2548 2549 std::string Error; 2550 const Target *X64Target = 2551 TargetRegistry::lookupTarget("", X64Triple, Error); 2552 if (X64Target) { 2553 SubtargetFeatures X64Features; 2554 SecondaryTarget.emplace(X64Target, *Obj, X64Triple.getTriple(), "", 2555 X64Features); 2556 } else { 2557 reportWarning(Error, Obj->getFileName()); 2558 } 2559 } 2560 } 2561 2562 const ObjectFile *DbgObj = Obj; 2563 if (!FetchedBinary.getBinary() && !Obj->hasDebugInfo()) { 2564 if (std::optional<OwningBinary<Binary>> DebugBinaryOpt = 2565 fetchBinaryByBuildID(*Obj)) { 2566 if (auto *FetchedObj = 2567 dyn_cast<const ObjectFile>(DebugBinaryOpt->getBinary())) { 2568 if (FetchedObj->hasDebugInfo()) { 2569 FetchedBinary = std::move(*DebugBinaryOpt); 2570 DbgObj = FetchedObj; 2571 } 2572 } 2573 } 2574 } 2575 2576 std::unique_ptr<object::Binary> DSYMBinary; 2577 std::unique_ptr<MemoryBuffer> DSYMBuf; 2578 if (!DbgObj->hasDebugInfo()) { 2579 if (const MachOObjectFile *MachOOF = dyn_cast<MachOObjectFile>(&*Obj)) { 2580 DbgObj = objdump::getMachODSymObject(MachOOF, Obj->getFileName(), 2581 DSYMBinary, DSYMBuf); 2582 if (!DbgObj) 2583 return; 2584 } 2585 } 2586 2587 SourcePrinter SP(DbgObj, TheTarget->getName()); 2588 2589 for (StringRef Opt : DisassemblerOptions) 2590 if (!PrimaryTarget.InstPrinter->applyTargetSpecificCLOption(Opt)) 2591 reportError(Obj->getFileName(), 2592 "Unrecognized disassembler option: " + Opt); 2593 2594 disassembleObject(*Obj, *DbgObj, PrimaryTarget, SecondaryTarget, SP, 2595 InlineRelocs); 2596 } 2597 2598 void Dumper::printRelocations() { 2599 StringRef Fmt = O.getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 2600 2601 // Build a mapping from relocation target to a vector of relocation 2602 // sections. Usually, there is an only one relocation section for 2603 // each relocated section. 2604 MapVector<SectionRef, std::vector<SectionRef>> SecToRelSec; 2605 uint64_t Ndx; 2606 for (const SectionRef &Section : ToolSectionFilter(O, &Ndx)) { 2607 if (O.isELF() && (ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC)) 2608 continue; 2609 if (Section.relocation_begin() == Section.relocation_end()) 2610 continue; 2611 Expected<section_iterator> SecOrErr = Section.getRelocatedSection(); 2612 if (!SecOrErr) 2613 reportError(O.getFileName(), 2614 "section (" + Twine(Ndx) + 2615 "): unable to get a relocation target: " + 2616 toString(SecOrErr.takeError())); 2617 SecToRelSec[**SecOrErr].push_back(Section); 2618 } 2619 2620 for (std::pair<SectionRef, std::vector<SectionRef>> &P : SecToRelSec) { 2621 StringRef SecName = unwrapOrError(P.first.getName(), O.getFileName()); 2622 outs() << "\nRELOCATION RECORDS FOR [" << SecName << "]:\n"; 2623 uint32_t OffsetPadding = (O.getBytesInAddress() > 4 ? 16 : 8); 2624 uint32_t TypePadding = 24; 2625 outs() << left_justify("OFFSET", OffsetPadding) << " " 2626 << left_justify("TYPE", TypePadding) << " " 2627 << "VALUE\n"; 2628 2629 for (SectionRef Section : P.second) { 2630 for (const RelocationRef &Reloc : Section.relocations()) { 2631 uint64_t Address = Reloc.getOffset(); 2632 SmallString<32> RelocName; 2633 SmallString<32> ValueStr; 2634 if (Address < StartAddress || Address > StopAddress || getHidden(Reloc)) 2635 continue; 2636 Reloc.getTypeName(RelocName); 2637 if (Error E = 2638 getRelocationValueString(Reloc, SymbolDescription, ValueStr)) 2639 reportUniqueWarning(std::move(E)); 2640 2641 outs() << format(Fmt.data(), Address) << " " 2642 << left_justify(RelocName, TypePadding) << " " << ValueStr 2643 << "\n"; 2644 } 2645 } 2646 } 2647 } 2648 2649 // Returns true if we need to show LMA column when dumping section headers. We 2650 // show it only when the platform is ELF and either we have at least one section 2651 // whose VMA and LMA are different and/or when --show-lma flag is used. 2652 static bool shouldDisplayLMA(const ObjectFile &Obj) { 2653 if (!Obj.isELF()) 2654 return false; 2655 for (const SectionRef &S : ToolSectionFilter(Obj)) 2656 if (S.getAddress() != getELFSectionLMA(S)) 2657 return true; 2658 return ShowLMA; 2659 } 2660 2661 static size_t getMaxSectionNameWidth(const ObjectFile &Obj) { 2662 // Default column width for names is 13 even if no names are that long. 2663 size_t MaxWidth = 13; 2664 for (const SectionRef &Section : ToolSectionFilter(Obj)) { 2665 StringRef Name = unwrapOrError(Section.getName(), Obj.getFileName()); 2666 MaxWidth = std::max(MaxWidth, Name.size()); 2667 } 2668 return MaxWidth; 2669 } 2670 2671 void objdump::printSectionHeaders(ObjectFile &Obj) { 2672 if (Obj.isELF() && Obj.sections().empty()) 2673 createFakeELFSections(Obj); 2674 2675 size_t NameWidth = getMaxSectionNameWidth(Obj); 2676 size_t AddressWidth = 2 * Obj.getBytesInAddress(); 2677 bool HasLMAColumn = shouldDisplayLMA(Obj); 2678 outs() << "\nSections:\n"; 2679 if (HasLMAColumn) 2680 outs() << "Idx " << left_justify("Name", NameWidth) << " Size " 2681 << left_justify("VMA", AddressWidth) << " " 2682 << left_justify("LMA", AddressWidth) << " Type\n"; 2683 else 2684 outs() << "Idx " << left_justify("Name", NameWidth) << " Size " 2685 << left_justify("VMA", AddressWidth) << " Type\n"; 2686 2687 uint64_t Idx; 2688 for (const SectionRef &Section : ToolSectionFilter(Obj, &Idx)) { 2689 StringRef Name = unwrapOrError(Section.getName(), Obj.getFileName()); 2690 uint64_t VMA = Section.getAddress(); 2691 if (shouldAdjustVA(Section)) 2692 VMA += AdjustVMA; 2693 2694 uint64_t Size = Section.getSize(); 2695 2696 std::string Type = Section.isText() ? "TEXT" : ""; 2697 if (Section.isData()) 2698 Type += Type.empty() ? "DATA" : ", DATA"; 2699 if (Section.isBSS()) 2700 Type += Type.empty() ? "BSS" : ", BSS"; 2701 if (Section.isDebugSection()) 2702 Type += Type.empty() ? "DEBUG" : ", DEBUG"; 2703 2704 if (HasLMAColumn) 2705 outs() << format("%3" PRIu64 " %-*s %08" PRIx64 " ", Idx, NameWidth, 2706 Name.str().c_str(), Size) 2707 << format_hex_no_prefix(VMA, AddressWidth) << " " 2708 << format_hex_no_prefix(getELFSectionLMA(Section), AddressWidth) 2709 << " " << Type << "\n"; 2710 else 2711 outs() << format("%3" PRIu64 " %-*s %08" PRIx64 " ", Idx, NameWidth, 2712 Name.str().c_str(), Size) 2713 << format_hex_no_prefix(VMA, AddressWidth) << " " << Type << "\n"; 2714 } 2715 } 2716 2717 void objdump::printSectionContents(const ObjectFile *Obj) { 2718 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(Obj); 2719 2720 for (const SectionRef &Section : ToolSectionFilter(*Obj)) { 2721 StringRef Name = unwrapOrError(Section.getName(), Obj->getFileName()); 2722 uint64_t BaseAddr = Section.getAddress(); 2723 uint64_t Size = Section.getSize(); 2724 if (!Size) 2725 continue; 2726 2727 outs() << "Contents of section "; 2728 StringRef SegmentName = getSegmentName(MachO, Section); 2729 if (!SegmentName.empty()) 2730 outs() << SegmentName << ","; 2731 outs() << Name << ":\n"; 2732 if (Section.isBSS()) { 2733 outs() << format("<skipping contents of bss section at [%04" PRIx64 2734 ", %04" PRIx64 ")>\n", 2735 BaseAddr, BaseAddr + Size); 2736 continue; 2737 } 2738 2739 StringRef Contents = unwrapOrError(Section.getContents(), Obj->getFileName()); 2740 2741 // Dump out the content as hex and printable ascii characters. 2742 for (std::size_t Addr = 0, End = Contents.size(); Addr < End; Addr += 16) { 2743 outs() << format(" %04" PRIx64 " ", BaseAddr + Addr); 2744 // Dump line of hex. 2745 for (std::size_t I = 0; I < 16; ++I) { 2746 if (I != 0 && I % 4 == 0) 2747 outs() << ' '; 2748 if (Addr + I < End) 2749 outs() << hexdigit((Contents[Addr + I] >> 4) & 0xF, true) 2750 << hexdigit(Contents[Addr + I] & 0xF, true); 2751 else 2752 outs() << " "; 2753 } 2754 // Print ascii. 2755 outs() << " "; 2756 for (std::size_t I = 0; I < 16 && Addr + I < End; ++I) { 2757 if (isPrint(static_cast<unsigned char>(Contents[Addr + I]) & 0xFF)) 2758 outs() << Contents[Addr + I]; 2759 else 2760 outs() << "."; 2761 } 2762 outs() << "\n"; 2763 } 2764 } 2765 } 2766 2767 void Dumper::printSymbolTable(StringRef ArchiveName, StringRef ArchitectureName, 2768 bool DumpDynamic) { 2769 if (O.isCOFF() && !DumpDynamic) { 2770 outs() << "\nSYMBOL TABLE:\n"; 2771 printCOFFSymbolTable(cast<const COFFObjectFile>(O)); 2772 return; 2773 } 2774 2775 const StringRef FileName = O.getFileName(); 2776 2777 if (!DumpDynamic) { 2778 outs() << "\nSYMBOL TABLE:\n"; 2779 for (auto I = O.symbol_begin(); I != O.symbol_end(); ++I) 2780 printSymbol(*I, {}, FileName, ArchiveName, ArchitectureName, DumpDynamic); 2781 return; 2782 } 2783 2784 outs() << "\nDYNAMIC SYMBOL TABLE:\n"; 2785 if (!O.isELF()) { 2786 reportWarning( 2787 "this operation is not currently supported for this file format", 2788 FileName); 2789 return; 2790 } 2791 2792 const ELFObjectFileBase *ELF = cast<const ELFObjectFileBase>(&O); 2793 auto Symbols = ELF->getDynamicSymbolIterators(); 2794 Expected<std::vector<VersionEntry>> SymbolVersionsOrErr = 2795 ELF->readDynsymVersions(); 2796 if (!SymbolVersionsOrErr) { 2797 reportWarning(toString(SymbolVersionsOrErr.takeError()), FileName); 2798 SymbolVersionsOrErr = std::vector<VersionEntry>(); 2799 (void)!SymbolVersionsOrErr; 2800 } 2801 for (auto &Sym : Symbols) 2802 printSymbol(Sym, *SymbolVersionsOrErr, FileName, ArchiveName, 2803 ArchitectureName, DumpDynamic); 2804 } 2805 2806 void Dumper::printSymbol(const SymbolRef &Symbol, 2807 ArrayRef<VersionEntry> SymbolVersions, 2808 StringRef FileName, StringRef ArchiveName, 2809 StringRef ArchitectureName, bool DumpDynamic) { 2810 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(&O); 2811 Expected<uint64_t> AddrOrErr = Symbol.getAddress(); 2812 if (!AddrOrErr) { 2813 reportUniqueWarning(AddrOrErr.takeError()); 2814 return; 2815 } 2816 uint64_t Address = *AddrOrErr; 2817 section_iterator SecI = unwrapOrError(Symbol.getSection(), FileName); 2818 if (SecI != O.section_end() && shouldAdjustVA(*SecI)) 2819 Address += AdjustVMA; 2820 if ((Address < StartAddress) || (Address > StopAddress)) 2821 return; 2822 SymbolRef::Type Type = 2823 unwrapOrError(Symbol.getType(), FileName, ArchiveName, ArchitectureName); 2824 uint32_t Flags = 2825 unwrapOrError(Symbol.getFlags(), FileName, ArchiveName, ArchitectureName); 2826 2827 // Don't ask a Mach-O STAB symbol for its section unless you know that 2828 // STAB symbol's section field refers to a valid section index. Otherwise 2829 // the symbol may error trying to load a section that does not exist. 2830 bool IsSTAB = false; 2831 if (MachO) { 2832 DataRefImpl SymDRI = Symbol.getRawDataRefImpl(); 2833 uint8_t NType = 2834 (MachO->is64Bit() ? MachO->getSymbol64TableEntry(SymDRI).n_type 2835 : MachO->getSymbolTableEntry(SymDRI).n_type); 2836 if (NType & MachO::N_STAB) 2837 IsSTAB = true; 2838 } 2839 section_iterator Section = IsSTAB 2840 ? O.section_end() 2841 : unwrapOrError(Symbol.getSection(), FileName, 2842 ArchiveName, ArchitectureName); 2843 2844 StringRef Name; 2845 if (Type == SymbolRef::ST_Debug && Section != O.section_end()) { 2846 if (Expected<StringRef> NameOrErr = Section->getName()) 2847 Name = *NameOrErr; 2848 else 2849 consumeError(NameOrErr.takeError()); 2850 2851 } else { 2852 Name = unwrapOrError(Symbol.getName(), FileName, ArchiveName, 2853 ArchitectureName); 2854 } 2855 2856 bool Global = Flags & SymbolRef::SF_Global; 2857 bool Weak = Flags & SymbolRef::SF_Weak; 2858 bool Absolute = Flags & SymbolRef::SF_Absolute; 2859 bool Common = Flags & SymbolRef::SF_Common; 2860 bool Hidden = Flags & SymbolRef::SF_Hidden; 2861 2862 char GlobLoc = ' '; 2863 if ((Section != O.section_end() || Absolute) && !Weak) 2864 GlobLoc = Global ? 'g' : 'l'; 2865 char IFunc = ' '; 2866 if (O.isELF()) { 2867 if (ELFSymbolRef(Symbol).getELFType() == ELF::STT_GNU_IFUNC) 2868 IFunc = 'i'; 2869 if (ELFSymbolRef(Symbol).getBinding() == ELF::STB_GNU_UNIQUE) 2870 GlobLoc = 'u'; 2871 } 2872 2873 char Debug = ' '; 2874 if (DumpDynamic) 2875 Debug = 'D'; 2876 else if (Type == SymbolRef::ST_Debug || Type == SymbolRef::ST_File) 2877 Debug = 'd'; 2878 2879 char FileFunc = ' '; 2880 if (Type == SymbolRef::ST_File) 2881 FileFunc = 'f'; 2882 else if (Type == SymbolRef::ST_Function) 2883 FileFunc = 'F'; 2884 else if (Type == SymbolRef::ST_Data) 2885 FileFunc = 'O'; 2886 2887 const char *Fmt = O.getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 2888 2889 outs() << format(Fmt, Address) << " " 2890 << GlobLoc // Local -> 'l', Global -> 'g', Neither -> ' ' 2891 << (Weak ? 'w' : ' ') // Weak? 2892 << ' ' // Constructor. Not supported yet. 2893 << ' ' // Warning. Not supported yet. 2894 << IFunc // Indirect reference to another symbol. 2895 << Debug // Debugging (d) or dynamic (D) symbol. 2896 << FileFunc // Name of function (F), file (f) or object (O). 2897 << ' '; 2898 if (Absolute) { 2899 outs() << "*ABS*"; 2900 } else if (Common) { 2901 outs() << "*COM*"; 2902 } else if (Section == O.section_end()) { 2903 if (O.isXCOFF()) { 2904 XCOFFSymbolRef XCOFFSym = cast<const XCOFFObjectFile>(O).toSymbolRef( 2905 Symbol.getRawDataRefImpl()); 2906 if (XCOFF::N_DEBUG == XCOFFSym.getSectionNumber()) 2907 outs() << "*DEBUG*"; 2908 else 2909 outs() << "*UND*"; 2910 } else 2911 outs() << "*UND*"; 2912 } else { 2913 StringRef SegmentName = getSegmentName(MachO, *Section); 2914 if (!SegmentName.empty()) 2915 outs() << SegmentName << ","; 2916 StringRef SectionName = unwrapOrError(Section->getName(), FileName); 2917 outs() << SectionName; 2918 if (O.isXCOFF()) { 2919 std::optional<SymbolRef> SymRef = 2920 getXCOFFSymbolContainingSymbolRef(cast<XCOFFObjectFile>(O), Symbol); 2921 if (SymRef) { 2922 2923 Expected<StringRef> NameOrErr = SymRef->getName(); 2924 2925 if (NameOrErr) { 2926 outs() << " (csect:"; 2927 std::string SymName = 2928 Demangle ? demangle(*NameOrErr) : NameOrErr->str(); 2929 2930 if (SymbolDescription) 2931 SymName = getXCOFFSymbolDescription(createSymbolInfo(O, *SymRef), 2932 SymName); 2933 2934 outs() << ' ' << SymName; 2935 outs() << ") "; 2936 } else 2937 reportWarning(toString(NameOrErr.takeError()), FileName); 2938 } 2939 } 2940 } 2941 2942 if (Common) 2943 outs() << '\t' << format(Fmt, static_cast<uint64_t>(Symbol.getAlignment())); 2944 else if (O.isXCOFF()) 2945 outs() << '\t' 2946 << format(Fmt, cast<XCOFFObjectFile>(O).getSymbolSize( 2947 Symbol.getRawDataRefImpl())); 2948 else if (O.isELF()) 2949 outs() << '\t' << format(Fmt, ELFSymbolRef(Symbol).getSize()); 2950 else if (O.isWasm()) 2951 outs() << '\t' 2952 << format(Fmt, static_cast<uint64_t>( 2953 cast<WasmObjectFile>(O).getSymbolSize(Symbol))); 2954 2955 if (O.isELF()) { 2956 if (!SymbolVersions.empty()) { 2957 const VersionEntry &Ver = 2958 SymbolVersions[Symbol.getRawDataRefImpl().d.b - 1]; 2959 std::string Str; 2960 if (!Ver.Name.empty()) 2961 Str = Ver.IsVerDef ? ' ' + Ver.Name : '(' + Ver.Name + ')'; 2962 outs() << ' ' << left_justify(Str, 12); 2963 } 2964 2965 uint8_t Other = ELFSymbolRef(Symbol).getOther(); 2966 switch (Other) { 2967 case ELF::STV_DEFAULT: 2968 break; 2969 case ELF::STV_INTERNAL: 2970 outs() << " .internal"; 2971 break; 2972 case ELF::STV_HIDDEN: 2973 outs() << " .hidden"; 2974 break; 2975 case ELF::STV_PROTECTED: 2976 outs() << " .protected"; 2977 break; 2978 default: 2979 outs() << format(" 0x%02x", Other); 2980 break; 2981 } 2982 } else if (Hidden) { 2983 outs() << " .hidden"; 2984 } 2985 2986 std::string SymName = Demangle ? demangle(Name) : Name.str(); 2987 if (O.isXCOFF() && SymbolDescription) 2988 SymName = getXCOFFSymbolDescription(createSymbolInfo(O, Symbol), SymName); 2989 2990 outs() << ' ' << SymName << '\n'; 2991 } 2992 2993 static void printUnwindInfo(const ObjectFile *O) { 2994 outs() << "Unwind info:\n\n"; 2995 2996 if (const COFFObjectFile *Coff = dyn_cast<COFFObjectFile>(O)) 2997 printCOFFUnwindInfo(Coff); 2998 else if (const MachOObjectFile *MachO = dyn_cast<MachOObjectFile>(O)) 2999 printMachOUnwindInfo(MachO); 3000 else 3001 // TODO: Extract DWARF dump tool to objdump. 3002 WithColor::error(errs(), ToolName) 3003 << "This operation is only currently supported " 3004 "for COFF and MachO object files.\n"; 3005 } 3006 3007 /// Dump the raw contents of the __clangast section so the output can be piped 3008 /// into llvm-bcanalyzer. 3009 static void printRawClangAST(const ObjectFile *Obj) { 3010 if (outs().is_displayed()) { 3011 WithColor::error(errs(), ToolName) 3012 << "The -raw-clang-ast option will dump the raw binary contents of " 3013 "the clang ast section.\n" 3014 "Please redirect the output to a file or another program such as " 3015 "llvm-bcanalyzer.\n"; 3016 return; 3017 } 3018 3019 StringRef ClangASTSectionName("__clangast"); 3020 if (Obj->isCOFF()) { 3021 ClangASTSectionName = "clangast"; 3022 } 3023 3024 std::optional<object::SectionRef> ClangASTSection; 3025 for (auto Sec : ToolSectionFilter(*Obj)) { 3026 StringRef Name; 3027 if (Expected<StringRef> NameOrErr = Sec.getName()) 3028 Name = *NameOrErr; 3029 else 3030 consumeError(NameOrErr.takeError()); 3031 3032 if (Name == ClangASTSectionName) { 3033 ClangASTSection = Sec; 3034 break; 3035 } 3036 } 3037 if (!ClangASTSection) 3038 return; 3039 3040 StringRef ClangASTContents = 3041 unwrapOrError(ClangASTSection->getContents(), Obj->getFileName()); 3042 outs().write(ClangASTContents.data(), ClangASTContents.size()); 3043 } 3044 3045 static void printFaultMaps(const ObjectFile *Obj) { 3046 StringRef FaultMapSectionName; 3047 3048 if (Obj->isELF()) { 3049 FaultMapSectionName = ".llvm_faultmaps"; 3050 } else if (Obj->isMachO()) { 3051 FaultMapSectionName = "__llvm_faultmaps"; 3052 } else { 3053 WithColor::error(errs(), ToolName) 3054 << "This operation is only currently supported " 3055 "for ELF and Mach-O executable files.\n"; 3056 return; 3057 } 3058 3059 std::optional<object::SectionRef> FaultMapSection; 3060 3061 for (auto Sec : ToolSectionFilter(*Obj)) { 3062 StringRef Name; 3063 if (Expected<StringRef> NameOrErr = Sec.getName()) 3064 Name = *NameOrErr; 3065 else 3066 consumeError(NameOrErr.takeError()); 3067 3068 if (Name == FaultMapSectionName) { 3069 FaultMapSection = Sec; 3070 break; 3071 } 3072 } 3073 3074 outs() << "FaultMap table:\n"; 3075 3076 if (!FaultMapSection) { 3077 outs() << "<not found>\n"; 3078 return; 3079 } 3080 3081 StringRef FaultMapContents = 3082 unwrapOrError(FaultMapSection->getContents(), Obj->getFileName()); 3083 FaultMapParser FMP(FaultMapContents.bytes_begin(), 3084 FaultMapContents.bytes_end()); 3085 3086 outs() << FMP; 3087 } 3088 3089 void Dumper::printPrivateHeaders() { 3090 reportError(O.getFileName(), "Invalid/Unsupported object file format"); 3091 } 3092 3093 static void printFileHeaders(const ObjectFile *O) { 3094 if (!O->isELF() && !O->isCOFF()) 3095 reportError(O->getFileName(), "Invalid/Unsupported object file format"); 3096 3097 Triple::ArchType AT = O->getArch(); 3098 outs() << "architecture: " << Triple::getArchTypeName(AT) << "\n"; 3099 uint64_t Address = unwrapOrError(O->getStartAddress(), O->getFileName()); 3100 3101 StringRef Fmt = O->getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 3102 outs() << "start address: " 3103 << "0x" << format(Fmt.data(), Address) << "\n"; 3104 } 3105 3106 static void printArchiveChild(StringRef Filename, const Archive::Child &C) { 3107 Expected<sys::fs::perms> ModeOrErr = C.getAccessMode(); 3108 if (!ModeOrErr) { 3109 WithColor::error(errs(), ToolName) << "ill-formed archive entry.\n"; 3110 consumeError(ModeOrErr.takeError()); 3111 return; 3112 } 3113 sys::fs::perms Mode = ModeOrErr.get(); 3114 outs() << ((Mode & sys::fs::owner_read) ? "r" : "-"); 3115 outs() << ((Mode & sys::fs::owner_write) ? "w" : "-"); 3116 outs() << ((Mode & sys::fs::owner_exe) ? "x" : "-"); 3117 outs() << ((Mode & sys::fs::group_read) ? "r" : "-"); 3118 outs() << ((Mode & sys::fs::group_write) ? "w" : "-"); 3119 outs() << ((Mode & sys::fs::group_exe) ? "x" : "-"); 3120 outs() << ((Mode & sys::fs::others_read) ? "r" : "-"); 3121 outs() << ((Mode & sys::fs::others_write) ? "w" : "-"); 3122 outs() << ((Mode & sys::fs::others_exe) ? "x" : "-"); 3123 3124 outs() << " "; 3125 3126 outs() << format("%d/%d %6" PRId64 " ", unwrapOrError(C.getUID(), Filename), 3127 unwrapOrError(C.getGID(), Filename), 3128 unwrapOrError(C.getRawSize(), Filename)); 3129 3130 StringRef RawLastModified = C.getRawLastModified(); 3131 unsigned Seconds; 3132 if (RawLastModified.getAsInteger(10, Seconds)) 3133 outs() << "(date: \"" << RawLastModified 3134 << "\" contains non-decimal chars) "; 3135 else { 3136 // Since ctime(3) returns a 26 character string of the form: 3137 // "Sun Sep 16 01:03:52 1973\n\0" 3138 // just print 24 characters. 3139 time_t t = Seconds; 3140 outs() << format("%.24s ", ctime(&t)); 3141 } 3142 3143 StringRef Name = ""; 3144 Expected<StringRef> NameOrErr = C.getName(); 3145 if (!NameOrErr) { 3146 consumeError(NameOrErr.takeError()); 3147 Name = unwrapOrError(C.getRawName(), Filename); 3148 } else { 3149 Name = NameOrErr.get(); 3150 } 3151 outs() << Name << "\n"; 3152 } 3153 3154 // For ELF only now. 3155 static bool shouldWarnForInvalidStartStopAddress(ObjectFile *Obj) { 3156 if (const auto *Elf = dyn_cast<ELFObjectFileBase>(Obj)) { 3157 if (Elf->getEType() != ELF::ET_REL) 3158 return true; 3159 } 3160 return false; 3161 } 3162 3163 static void checkForInvalidStartStopAddress(ObjectFile *Obj, 3164 uint64_t Start, uint64_t Stop) { 3165 if (!shouldWarnForInvalidStartStopAddress(Obj)) 3166 return; 3167 3168 for (const SectionRef &Section : Obj->sections()) 3169 if (ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC) { 3170 uint64_t BaseAddr = Section.getAddress(); 3171 uint64_t Size = Section.getSize(); 3172 if ((Start < BaseAddr + Size) && Stop > BaseAddr) 3173 return; 3174 } 3175 3176 if (!HasStartAddressFlag) 3177 reportWarning("no section has address less than 0x" + 3178 Twine::utohexstr(Stop) + " specified by --stop-address", 3179 Obj->getFileName()); 3180 else if (!HasStopAddressFlag) 3181 reportWarning("no section has address greater than or equal to 0x" + 3182 Twine::utohexstr(Start) + " specified by --start-address", 3183 Obj->getFileName()); 3184 else 3185 reportWarning("no section overlaps the range [0x" + 3186 Twine::utohexstr(Start) + ",0x" + Twine::utohexstr(Stop) + 3187 ") specified by --start-address/--stop-address", 3188 Obj->getFileName()); 3189 } 3190 3191 static void dumpObject(ObjectFile *O, const Archive *A = nullptr, 3192 const Archive::Child *C = nullptr) { 3193 Expected<std::unique_ptr<Dumper>> DumperOrErr = createDumper(*O); 3194 if (!DumperOrErr) { 3195 reportError(DumperOrErr.takeError(), O->getFileName(), 3196 A ? A->getFileName() : ""); 3197 return; 3198 } 3199 Dumper &D = **DumperOrErr; 3200 3201 // Avoid other output when using a raw option. 3202 if (!RawClangAST) { 3203 outs() << '\n'; 3204 if (A) 3205 outs() << A->getFileName() << "(" << O->getFileName() << ")"; 3206 else 3207 outs() << O->getFileName(); 3208 outs() << ":\tfile format " << O->getFileFormatName().lower() << "\n"; 3209 } 3210 3211 if (HasStartAddressFlag || HasStopAddressFlag) 3212 checkForInvalidStartStopAddress(O, StartAddress, StopAddress); 3213 3214 // TODO: Change print* free functions to Dumper member functions to utilitize 3215 // stateful functions like reportUniqueWarning. 3216 3217 // Note: the order here matches GNU objdump for compatability. 3218 StringRef ArchiveName = A ? A->getFileName() : ""; 3219 if (ArchiveHeaders && !MachOOpt && C) 3220 printArchiveChild(ArchiveName, *C); 3221 if (FileHeaders) 3222 printFileHeaders(O); 3223 if (PrivateHeaders || FirstPrivateHeader) 3224 D.printPrivateHeaders(); 3225 if (SectionHeaders) 3226 printSectionHeaders(*O); 3227 if (SymbolTable) 3228 D.printSymbolTable(ArchiveName); 3229 if (DynamicSymbolTable) 3230 D.printSymbolTable(ArchiveName, /*ArchitectureName=*/"", 3231 /*DumpDynamic=*/true); 3232 if (DwarfDumpType != DIDT_Null) { 3233 std::unique_ptr<DIContext> DICtx = DWARFContext::create(*O); 3234 // Dump the complete DWARF structure. 3235 DIDumpOptions DumpOpts; 3236 DumpOpts.DumpType = DwarfDumpType; 3237 DICtx->dump(outs(), DumpOpts); 3238 } 3239 if (Relocations && !Disassemble) 3240 D.printRelocations(); 3241 if (DynamicRelocations) 3242 D.printDynamicRelocations(); 3243 if (SectionContents) 3244 printSectionContents(O); 3245 if (Disassemble) 3246 disassembleObject(O, Relocations); 3247 if (UnwindInfo) 3248 printUnwindInfo(O); 3249 3250 // Mach-O specific options: 3251 if (ExportsTrie) 3252 printExportsTrie(O); 3253 if (Rebase) 3254 printRebaseTable(O); 3255 if (Bind) 3256 printBindTable(O); 3257 if (LazyBind) 3258 printLazyBindTable(O); 3259 if (WeakBind) 3260 printWeakBindTable(O); 3261 3262 // Other special sections: 3263 if (RawClangAST) 3264 printRawClangAST(O); 3265 if (FaultMapSection) 3266 printFaultMaps(O); 3267 if (Offloading) 3268 dumpOffloadBinary(*O); 3269 } 3270 3271 static void dumpObject(const COFFImportFile *I, const Archive *A, 3272 const Archive::Child *C = nullptr) { 3273 StringRef ArchiveName = A ? A->getFileName() : ""; 3274 3275 // Avoid other output when using a raw option. 3276 if (!RawClangAST) 3277 outs() << '\n' 3278 << ArchiveName << "(" << I->getFileName() << ")" 3279 << ":\tfile format COFF-import-file" 3280 << "\n\n"; 3281 3282 if (ArchiveHeaders && !MachOOpt && C) 3283 printArchiveChild(ArchiveName, *C); 3284 if (SymbolTable) 3285 printCOFFSymbolTable(*I); 3286 } 3287 3288 /// Dump each object file in \a a; 3289 static void dumpArchive(const Archive *A) { 3290 Error Err = Error::success(); 3291 unsigned I = -1; 3292 for (auto &C : A->children(Err)) { 3293 ++I; 3294 Expected<std::unique_ptr<Binary>> ChildOrErr = C.getAsBinary(); 3295 if (!ChildOrErr) { 3296 if (auto E = isNotObjectErrorInvalidFileType(ChildOrErr.takeError())) 3297 reportError(std::move(E), getFileNameForError(C, I), A->getFileName()); 3298 continue; 3299 } 3300 if (ObjectFile *O = dyn_cast<ObjectFile>(&*ChildOrErr.get())) 3301 dumpObject(O, A, &C); 3302 else if (COFFImportFile *I = dyn_cast<COFFImportFile>(&*ChildOrErr.get())) 3303 dumpObject(I, A, &C); 3304 else 3305 reportError(errorCodeToError(object_error::invalid_file_type), 3306 A->getFileName()); 3307 } 3308 if (Err) 3309 reportError(std::move(Err), A->getFileName()); 3310 } 3311 3312 /// Open file and figure out how to dump it. 3313 static void dumpInput(StringRef file) { 3314 // If we are using the Mach-O specific object file parser, then let it parse 3315 // the file and process the command line options. So the -arch flags can 3316 // be used to select specific slices, etc. 3317 if (MachOOpt) { 3318 parseInputMachO(file); 3319 return; 3320 } 3321 3322 // Attempt to open the binary. 3323 OwningBinary<Binary> OBinary = unwrapOrError(createBinary(file), file); 3324 Binary &Binary = *OBinary.getBinary(); 3325 3326 if (Archive *A = dyn_cast<Archive>(&Binary)) 3327 dumpArchive(A); 3328 else if (ObjectFile *O = dyn_cast<ObjectFile>(&Binary)) 3329 dumpObject(O); 3330 else if (MachOUniversalBinary *UB = dyn_cast<MachOUniversalBinary>(&Binary)) 3331 parseInputMachO(UB); 3332 else if (OffloadBinary *OB = dyn_cast<OffloadBinary>(&Binary)) 3333 dumpOffloadSections(*OB); 3334 else 3335 reportError(errorCodeToError(object_error::invalid_file_type), file); 3336 } 3337 3338 template <typename T> 3339 static void parseIntArg(const llvm::opt::InputArgList &InputArgs, int ID, 3340 T &Value) { 3341 if (const opt::Arg *A = InputArgs.getLastArg(ID)) { 3342 StringRef V(A->getValue()); 3343 if (!llvm::to_integer(V, Value, 0)) { 3344 reportCmdLineError(A->getSpelling() + 3345 ": expected a non-negative integer, but got '" + V + 3346 "'"); 3347 } 3348 } 3349 } 3350 3351 static object::BuildID parseBuildIDArg(const opt::Arg *A) { 3352 StringRef V(A->getValue()); 3353 object::BuildID BID = parseBuildID(V); 3354 if (BID.empty()) 3355 reportCmdLineError(A->getSpelling() + ": expected a build ID, but got '" + 3356 V + "'"); 3357 return BID; 3358 } 3359 3360 void objdump::invalidArgValue(const opt::Arg *A) { 3361 reportCmdLineError("'" + StringRef(A->getValue()) + 3362 "' is not a valid value for '" + A->getSpelling() + "'"); 3363 } 3364 3365 static std::vector<std::string> 3366 commaSeparatedValues(const llvm::opt::InputArgList &InputArgs, int ID) { 3367 std::vector<std::string> Values; 3368 for (StringRef Value : InputArgs.getAllArgValues(ID)) { 3369 llvm::SmallVector<StringRef, 2> SplitValues; 3370 llvm::SplitString(Value, SplitValues, ","); 3371 for (StringRef SplitValue : SplitValues) 3372 Values.push_back(SplitValue.str()); 3373 } 3374 return Values; 3375 } 3376 3377 static void parseOtoolOptions(const llvm::opt::InputArgList &InputArgs) { 3378 MachOOpt = true; 3379 FullLeadingAddr = true; 3380 PrintImmHex = true; 3381 3382 ArchName = InputArgs.getLastArgValue(OTOOL_arch).str(); 3383 LinkOptHints = InputArgs.hasArg(OTOOL_C); 3384 if (InputArgs.hasArg(OTOOL_d)) 3385 FilterSections.push_back("__DATA,__data"); 3386 DylibId = InputArgs.hasArg(OTOOL_D); 3387 UniversalHeaders = InputArgs.hasArg(OTOOL_f); 3388 DataInCode = InputArgs.hasArg(OTOOL_G); 3389 FirstPrivateHeader = InputArgs.hasArg(OTOOL_h); 3390 IndirectSymbols = InputArgs.hasArg(OTOOL_I); 3391 ShowRawInsn = InputArgs.hasArg(OTOOL_j); 3392 PrivateHeaders = InputArgs.hasArg(OTOOL_l); 3393 DylibsUsed = InputArgs.hasArg(OTOOL_L); 3394 MCPU = InputArgs.getLastArgValue(OTOOL_mcpu_EQ).str(); 3395 ObjcMetaData = InputArgs.hasArg(OTOOL_o); 3396 DisSymName = InputArgs.getLastArgValue(OTOOL_p).str(); 3397 InfoPlist = InputArgs.hasArg(OTOOL_P); 3398 Relocations = InputArgs.hasArg(OTOOL_r); 3399 if (const Arg *A = InputArgs.getLastArg(OTOOL_s)) { 3400 auto Filter = (A->getValue(0) + StringRef(",") + A->getValue(1)).str(); 3401 FilterSections.push_back(Filter); 3402 } 3403 if (InputArgs.hasArg(OTOOL_t)) 3404 FilterSections.push_back("__TEXT,__text"); 3405 Verbose = InputArgs.hasArg(OTOOL_v) || InputArgs.hasArg(OTOOL_V) || 3406 InputArgs.hasArg(OTOOL_o); 3407 SymbolicOperands = InputArgs.hasArg(OTOOL_V); 3408 if (InputArgs.hasArg(OTOOL_x)) 3409 FilterSections.push_back(",__text"); 3410 LeadingAddr = LeadingHeaders = !InputArgs.hasArg(OTOOL_X); 3411 3412 ChainedFixups = InputArgs.hasArg(OTOOL_chained_fixups); 3413 DyldInfo = InputArgs.hasArg(OTOOL_dyld_info); 3414 3415 InputFilenames = InputArgs.getAllArgValues(OTOOL_INPUT); 3416 if (InputFilenames.empty()) 3417 reportCmdLineError("no input file"); 3418 3419 for (const Arg *A : InputArgs) { 3420 const Option &O = A->getOption(); 3421 if (O.getGroup().isValid() && O.getGroup().getID() == OTOOL_grp_obsolete) { 3422 reportCmdLineWarning(O.getPrefixedName() + 3423 " is obsolete and not implemented"); 3424 } 3425 } 3426 } 3427 3428 static void parseObjdumpOptions(const llvm::opt::InputArgList &InputArgs) { 3429 parseIntArg(InputArgs, OBJDUMP_adjust_vma_EQ, AdjustVMA); 3430 AllHeaders = InputArgs.hasArg(OBJDUMP_all_headers); 3431 ArchName = InputArgs.getLastArgValue(OBJDUMP_arch_name_EQ).str(); 3432 ArchiveHeaders = InputArgs.hasArg(OBJDUMP_archive_headers); 3433 Demangle = InputArgs.hasArg(OBJDUMP_demangle); 3434 Disassemble = InputArgs.hasArg(OBJDUMP_disassemble); 3435 DisassembleAll = InputArgs.hasArg(OBJDUMP_disassemble_all); 3436 SymbolDescription = InputArgs.hasArg(OBJDUMP_symbol_description); 3437 TracebackTable = InputArgs.hasArg(OBJDUMP_traceback_table); 3438 DisassembleSymbols = 3439 commaSeparatedValues(InputArgs, OBJDUMP_disassemble_symbols_EQ); 3440 DisassembleZeroes = InputArgs.hasArg(OBJDUMP_disassemble_zeroes); 3441 if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_dwarf_EQ)) { 3442 DwarfDumpType = StringSwitch<DIDumpType>(A->getValue()) 3443 .Case("frames", DIDT_DebugFrame) 3444 .Default(DIDT_Null); 3445 if (DwarfDumpType == DIDT_Null) 3446 invalidArgValue(A); 3447 } 3448 DynamicRelocations = InputArgs.hasArg(OBJDUMP_dynamic_reloc); 3449 FaultMapSection = InputArgs.hasArg(OBJDUMP_fault_map_section); 3450 Offloading = InputArgs.hasArg(OBJDUMP_offloading); 3451 FileHeaders = InputArgs.hasArg(OBJDUMP_file_headers); 3452 SectionContents = InputArgs.hasArg(OBJDUMP_full_contents); 3453 PrintLines = InputArgs.hasArg(OBJDUMP_line_numbers); 3454 InputFilenames = InputArgs.getAllArgValues(OBJDUMP_INPUT); 3455 MachOOpt = InputArgs.hasArg(OBJDUMP_macho); 3456 MCPU = InputArgs.getLastArgValue(OBJDUMP_mcpu_EQ).str(); 3457 MAttrs = commaSeparatedValues(InputArgs, OBJDUMP_mattr_EQ); 3458 ShowRawInsn = !InputArgs.hasArg(OBJDUMP_no_show_raw_insn); 3459 LeadingAddr = !InputArgs.hasArg(OBJDUMP_no_leading_addr); 3460 RawClangAST = InputArgs.hasArg(OBJDUMP_raw_clang_ast); 3461 Relocations = InputArgs.hasArg(OBJDUMP_reloc); 3462 PrintImmHex = 3463 InputArgs.hasFlag(OBJDUMP_print_imm_hex, OBJDUMP_no_print_imm_hex, true); 3464 PrivateHeaders = InputArgs.hasArg(OBJDUMP_private_headers); 3465 FilterSections = InputArgs.getAllArgValues(OBJDUMP_section_EQ); 3466 SectionHeaders = InputArgs.hasArg(OBJDUMP_section_headers); 3467 ShowAllSymbols = InputArgs.hasArg(OBJDUMP_show_all_symbols); 3468 ShowLMA = InputArgs.hasArg(OBJDUMP_show_lma); 3469 PrintSource = InputArgs.hasArg(OBJDUMP_source); 3470 parseIntArg(InputArgs, OBJDUMP_start_address_EQ, StartAddress); 3471 HasStartAddressFlag = InputArgs.hasArg(OBJDUMP_start_address_EQ); 3472 parseIntArg(InputArgs, OBJDUMP_stop_address_EQ, StopAddress); 3473 HasStopAddressFlag = InputArgs.hasArg(OBJDUMP_stop_address_EQ); 3474 SymbolTable = InputArgs.hasArg(OBJDUMP_syms); 3475 SymbolizeOperands = InputArgs.hasArg(OBJDUMP_symbolize_operands); 3476 DynamicSymbolTable = InputArgs.hasArg(OBJDUMP_dynamic_syms); 3477 TripleName = InputArgs.getLastArgValue(OBJDUMP_triple_EQ).str(); 3478 UnwindInfo = InputArgs.hasArg(OBJDUMP_unwind_info); 3479 Wide = InputArgs.hasArg(OBJDUMP_wide); 3480 Prefix = InputArgs.getLastArgValue(OBJDUMP_prefix).str(); 3481 parseIntArg(InputArgs, OBJDUMP_prefix_strip, PrefixStrip); 3482 if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_debug_vars_EQ)) { 3483 DbgVariables = StringSwitch<DebugVarsFormat>(A->getValue()) 3484 .Case("ascii", DVASCII) 3485 .Case("unicode", DVUnicode) 3486 .Default(DVInvalid); 3487 if (DbgVariables == DVInvalid) 3488 invalidArgValue(A); 3489 } 3490 if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_disassembler_color_EQ)) { 3491 DisassemblyColor = StringSwitch<ColorOutput>(A->getValue()) 3492 .Case("on", ColorOutput::Enable) 3493 .Case("off", ColorOutput::Disable) 3494 .Case("terminal", ColorOutput::Auto) 3495 .Default(ColorOutput::Invalid); 3496 if (DisassemblyColor == ColorOutput::Invalid) 3497 invalidArgValue(A); 3498 } 3499 3500 parseIntArg(InputArgs, OBJDUMP_debug_vars_indent_EQ, DbgIndent); 3501 3502 parseMachOOptions(InputArgs); 3503 3504 // Parse -M (--disassembler-options) and deprecated 3505 // --x86-asm-syntax={att,intel}. 3506 // 3507 // Note, for x86, the asm dialect (AssemblerDialect) is initialized when the 3508 // MCAsmInfo is constructed. MCInstPrinter::applyTargetSpecificCLOption is 3509 // called too late. For now we have to use the internal cl::opt option. 3510 const char *AsmSyntax = nullptr; 3511 for (const auto *A : InputArgs.filtered(OBJDUMP_disassembler_options_EQ, 3512 OBJDUMP_x86_asm_syntax_att, 3513 OBJDUMP_x86_asm_syntax_intel)) { 3514 switch (A->getOption().getID()) { 3515 case OBJDUMP_x86_asm_syntax_att: 3516 AsmSyntax = "--x86-asm-syntax=att"; 3517 continue; 3518 case OBJDUMP_x86_asm_syntax_intel: 3519 AsmSyntax = "--x86-asm-syntax=intel"; 3520 continue; 3521 } 3522 3523 SmallVector<StringRef, 2> Values; 3524 llvm::SplitString(A->getValue(), Values, ","); 3525 for (StringRef V : Values) { 3526 if (V == "att") 3527 AsmSyntax = "--x86-asm-syntax=att"; 3528 else if (V == "intel") 3529 AsmSyntax = "--x86-asm-syntax=intel"; 3530 else 3531 DisassemblerOptions.push_back(V.str()); 3532 } 3533 } 3534 SmallVector<const char *> Args = {"llvm-objdump"}; 3535 for (const opt::Arg *A : InputArgs.filtered(OBJDUMP_mllvm)) 3536 Args.push_back(A->getValue()); 3537 if (AsmSyntax) 3538 Args.push_back(AsmSyntax); 3539 if (Args.size() > 1) 3540 llvm::cl::ParseCommandLineOptions(Args.size(), Args.data()); 3541 3542 // Look up any provided build IDs, then append them to the input filenames. 3543 for (const opt::Arg *A : InputArgs.filtered(OBJDUMP_build_id)) { 3544 object::BuildID BuildID = parseBuildIDArg(A); 3545 std::optional<std::string> Path = BIDFetcher->fetch(BuildID); 3546 if (!Path) { 3547 reportCmdLineError(A->getSpelling() + ": could not find build ID '" + 3548 A->getValue() + "'"); 3549 } 3550 InputFilenames.push_back(std::move(*Path)); 3551 } 3552 3553 // objdump defaults to a.out if no filenames specified. 3554 if (InputFilenames.empty()) 3555 InputFilenames.push_back("a.out"); 3556 } 3557 3558 int llvm_objdump_main(int argc, char **argv, const llvm::ToolContext &) { 3559 using namespace llvm; 3560 3561 ToolName = argv[0]; 3562 std::unique_ptr<CommonOptTable> T; 3563 OptSpecifier Unknown, HelpFlag, HelpHiddenFlag, VersionFlag; 3564 3565 StringRef Stem = sys::path::stem(ToolName); 3566 auto Is = [=](StringRef Tool) { 3567 // We need to recognize the following filenames: 3568 // 3569 // llvm-objdump -> objdump 3570 // llvm-otool-10.exe -> otool 3571 // powerpc64-unknown-freebsd13-objdump -> objdump 3572 auto I = Stem.rfind_insensitive(Tool); 3573 return I != StringRef::npos && 3574 (I + Tool.size() == Stem.size() || !isAlnum(Stem[I + Tool.size()])); 3575 }; 3576 if (Is("otool")) { 3577 T = std::make_unique<OtoolOptTable>(); 3578 Unknown = OTOOL_UNKNOWN; 3579 HelpFlag = OTOOL_help; 3580 HelpHiddenFlag = OTOOL_help_hidden; 3581 VersionFlag = OTOOL_version; 3582 } else { 3583 T = std::make_unique<ObjdumpOptTable>(); 3584 Unknown = OBJDUMP_UNKNOWN; 3585 HelpFlag = OBJDUMP_help; 3586 HelpHiddenFlag = OBJDUMP_help_hidden; 3587 VersionFlag = OBJDUMP_version; 3588 } 3589 3590 BumpPtrAllocator A; 3591 StringSaver Saver(A); 3592 opt::InputArgList InputArgs = 3593 T->parseArgs(argc, argv, Unknown, Saver, 3594 [&](StringRef Msg) { reportCmdLineError(Msg); }); 3595 3596 if (InputArgs.size() == 0 || InputArgs.hasArg(HelpFlag)) { 3597 T->printHelp(ToolName); 3598 return 0; 3599 } 3600 if (InputArgs.hasArg(HelpHiddenFlag)) { 3601 T->printHelp(ToolName, /*ShowHidden=*/true); 3602 return 0; 3603 } 3604 3605 // Initialize targets and assembly printers/parsers. 3606 InitializeAllTargetInfos(); 3607 InitializeAllTargetMCs(); 3608 InitializeAllDisassemblers(); 3609 3610 if (InputArgs.hasArg(VersionFlag)) { 3611 cl::PrintVersionMessage(); 3612 if (!Is("otool")) { 3613 outs() << '\n'; 3614 TargetRegistry::printRegisteredTargetsForVersion(outs()); 3615 } 3616 return 0; 3617 } 3618 3619 // Initialize debuginfod. 3620 const bool ShouldUseDebuginfodByDefault = 3621 InputArgs.hasArg(OBJDUMP_build_id) || canUseDebuginfod(); 3622 std::vector<std::string> DebugFileDirectories = 3623 InputArgs.getAllArgValues(OBJDUMP_debug_file_directory); 3624 if (InputArgs.hasFlag(OBJDUMP_debuginfod, OBJDUMP_no_debuginfod, 3625 ShouldUseDebuginfodByDefault)) { 3626 HTTPClient::initialize(); 3627 BIDFetcher = 3628 std::make_unique<DebuginfodFetcher>(std::move(DebugFileDirectories)); 3629 } else { 3630 BIDFetcher = 3631 std::make_unique<BuildIDFetcher>(std::move(DebugFileDirectories)); 3632 } 3633 3634 if (Is("otool")) 3635 parseOtoolOptions(InputArgs); 3636 else 3637 parseObjdumpOptions(InputArgs); 3638 3639 if (StartAddress >= StopAddress) 3640 reportCmdLineError("start address should be less than stop address"); 3641 3642 // Removes trailing separators from prefix. 3643 while (!Prefix.empty() && sys::path::is_separator(Prefix.back())) 3644 Prefix.pop_back(); 3645 3646 if (AllHeaders) 3647 ArchiveHeaders = FileHeaders = PrivateHeaders = Relocations = 3648 SectionHeaders = SymbolTable = true; 3649 3650 if (DisassembleAll || PrintSource || PrintLines || TracebackTable || 3651 !DisassembleSymbols.empty()) 3652 Disassemble = true; 3653 3654 if (!ArchiveHeaders && !Disassemble && DwarfDumpType == DIDT_Null && 3655 !DynamicRelocations && !FileHeaders && !PrivateHeaders && !RawClangAST && 3656 !Relocations && !SectionHeaders && !SectionContents && !SymbolTable && 3657 !DynamicSymbolTable && !UnwindInfo && !FaultMapSection && !Offloading && 3658 !(MachOOpt && 3659 (Bind || DataInCode || ChainedFixups || DyldInfo || DylibId || 3660 DylibsUsed || ExportsTrie || FirstPrivateHeader || 3661 FunctionStartsType != FunctionStartsMode::None || IndirectSymbols || 3662 InfoPlist || LazyBind || LinkOptHints || ObjcMetaData || Rebase || 3663 Rpaths || UniversalHeaders || WeakBind || !FilterSections.empty()))) { 3664 T->printHelp(ToolName); 3665 return 2; 3666 } 3667 3668 DisasmSymbolSet.insert(DisassembleSymbols.begin(), DisassembleSymbols.end()); 3669 3670 llvm::for_each(InputFilenames, dumpInput); 3671 3672 warnOnNoMatchForSections(); 3673 3674 return EXIT_SUCCESS; 3675 } 3676