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