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/SmallSet.h" 31 #include "llvm/ADT/StringExtras.h" 32 #include "llvm/ADT/StringSet.h" 33 #include "llvm/ADT/Twine.h" 34 #include "llvm/DebugInfo/BTF/BTFParser.h" 35 #include "llvm/DebugInfo/DWARF/DWARFContext.h" 36 #include "llvm/DebugInfo/Symbolize/SymbolizableModule.h" 37 #include "llvm/DebugInfo/Symbolize/Symbolize.h" 38 #include "llvm/Debuginfod/BuildIDFetcher.h" 39 #include "llvm/Debuginfod/Debuginfod.h" 40 #include "llvm/Debuginfod/HTTPClient.h" 41 #include "llvm/Demangle/Demangle.h" 42 #include "llvm/MC/MCAsmInfo.h" 43 #include "llvm/MC/MCContext.h" 44 #include "llvm/MC/MCDisassembler/MCDisassembler.h" 45 #include "llvm/MC/MCDisassembler/MCRelocationInfo.h" 46 #include "llvm/MC/MCInst.h" 47 #include "llvm/MC/MCInstPrinter.h" 48 #include "llvm/MC/MCInstrAnalysis.h" 49 #include "llvm/MC/MCInstrInfo.h" 50 #include "llvm/MC/MCObjectFileInfo.h" 51 #include "llvm/MC/MCRegisterInfo.h" 52 #include "llvm/MC/MCTargetOptions.h" 53 #include "llvm/MC/TargetRegistry.h" 54 #include "llvm/Object/Archive.h" 55 #include "llvm/Object/BuildID.h" 56 #include "llvm/Object/COFF.h" 57 #include "llvm/Object/COFFImportFile.h" 58 #include "llvm/Object/ELFObjectFile.h" 59 #include "llvm/Object/ELFTypes.h" 60 #include "llvm/Object/FaultMapParser.h" 61 #include "llvm/Object/MachO.h" 62 #include "llvm/Object/MachOUniversal.h" 63 #include "llvm/Object/ObjectFile.h" 64 #include "llvm/Object/OffloadBinary.h" 65 #include "llvm/Object/Wasm.h" 66 #include "llvm/Option/Arg.h" 67 #include "llvm/Option/ArgList.h" 68 #include "llvm/Option/Option.h" 69 #include "llvm/Support/Casting.h" 70 #include "llvm/Support/Debug.h" 71 #include "llvm/Support/Errc.h" 72 #include "llvm/Support/FileSystem.h" 73 #include "llvm/Support/Format.h" 74 #include "llvm/Support/FormatVariadic.h" 75 #include "llvm/Support/GraphWriter.h" 76 #include "llvm/Support/InitLLVM.h" 77 #include "llvm/Support/LLVMDriver.h" 78 #include "llvm/Support/MemoryBuffer.h" 79 #include "llvm/Support/SourceMgr.h" 80 #include "llvm/Support/StringSaver.h" 81 #include "llvm/Support/TargetSelect.h" 82 #include "llvm/Support/WithColor.h" 83 #include "llvm/Support/raw_ostream.h" 84 #include "llvm/TargetParser/Host.h" 85 #include "llvm/TargetParser/Triple.h" 86 #include <algorithm> 87 #include <cctype> 88 #include <cstring> 89 #include <optional> 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<const 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.BBEntries) { 1279 uint64_t BBAddress = BBEntry.Offset + Iter->second.Addr; 1280 if (BBAddress >= EndAddress) 1281 continue; 1282 Labels[BBAddress].push_back(("BB" + Twine(BBEntry.ID)).str()); 1283 } 1284 } 1285 1286 static void collectLocalBranchTargets( 1287 ArrayRef<uint8_t> Bytes, const MCInstrAnalysis *MIA, MCDisassembler *DisAsm, 1288 MCInstPrinter *IP, const MCSubtargetInfo *STI, uint64_t SectionAddr, 1289 uint64_t Start, uint64_t End, std::unordered_map<uint64_t, std::string> &Labels) { 1290 // So far only supports PowerPC and X86. 1291 if (!STI->getTargetTriple().isPPC() && !STI->getTargetTriple().isX86()) 1292 return; 1293 1294 Labels.clear(); 1295 unsigned LabelCount = 0; 1296 Start += SectionAddr; 1297 End += SectionAddr; 1298 uint64_t Index = Start; 1299 while (Index < End) { 1300 // Disassemble a real instruction and record function-local branch labels. 1301 MCInst Inst; 1302 uint64_t Size; 1303 ArrayRef<uint8_t> ThisBytes = Bytes.slice(Index - SectionAddr); 1304 bool Disassembled = 1305 DisAsm->getInstruction(Inst, Size, ThisBytes, Index, nulls()); 1306 if (Size == 0) 1307 Size = std::min<uint64_t>(ThisBytes.size(), 1308 DisAsm->suggestBytesToSkip(ThisBytes, Index)); 1309 1310 if (Disassembled && MIA) { 1311 uint64_t Target; 1312 bool TargetKnown = MIA->evaluateBranch(Inst, Index, Size, Target); 1313 // On PowerPC, if the address of a branch is the same as the target, it 1314 // means that it's a function call. Do not mark the label for this case. 1315 if (TargetKnown && (Target >= Start && Target < End) && 1316 !Labels.count(Target) && 1317 !(STI->getTargetTriple().isPPC() && Target == Index)) 1318 Labels[Target] = ("L" + Twine(LabelCount++)).str(); 1319 } 1320 Index += Size; 1321 } 1322 } 1323 1324 // Create an MCSymbolizer for the target and add it to the MCDisassembler. 1325 // This is currently only used on AMDGPU, and assumes the format of the 1326 // void * argument passed to AMDGPU's createMCSymbolizer. 1327 static void addSymbolizer( 1328 MCContext &Ctx, const Target *Target, StringRef TripleName, 1329 MCDisassembler *DisAsm, uint64_t SectionAddr, ArrayRef<uint8_t> Bytes, 1330 SectionSymbolsTy &Symbols, 1331 std::vector<std::unique_ptr<std::string>> &SynthesizedLabelNames) { 1332 1333 std::unique_ptr<MCRelocationInfo> RelInfo( 1334 Target->createMCRelocationInfo(TripleName, Ctx)); 1335 if (!RelInfo) 1336 return; 1337 std::unique_ptr<MCSymbolizer> Symbolizer(Target->createMCSymbolizer( 1338 TripleName, nullptr, nullptr, &Symbols, &Ctx, std::move(RelInfo))); 1339 MCSymbolizer *SymbolizerPtr = &*Symbolizer; 1340 DisAsm->setSymbolizer(std::move(Symbolizer)); 1341 1342 if (!SymbolizeOperands) 1343 return; 1344 1345 // Synthesize labels referenced by branch instructions by 1346 // disassembling, discarding the output, and collecting the referenced 1347 // addresses from the symbolizer. 1348 for (size_t Index = 0; Index != Bytes.size();) { 1349 MCInst Inst; 1350 uint64_t Size; 1351 ArrayRef<uint8_t> ThisBytes = Bytes.slice(Index); 1352 const uint64_t ThisAddr = SectionAddr + Index; 1353 DisAsm->getInstruction(Inst, Size, ThisBytes, ThisAddr, nulls()); 1354 if (Size == 0) 1355 Size = std::min<uint64_t>(ThisBytes.size(), 1356 DisAsm->suggestBytesToSkip(ThisBytes, Index)); 1357 Index += Size; 1358 } 1359 ArrayRef<uint64_t> LabelAddrsRef = SymbolizerPtr->getReferencedAddresses(); 1360 // Copy and sort to remove duplicates. 1361 std::vector<uint64_t> LabelAddrs; 1362 LabelAddrs.insert(LabelAddrs.end(), LabelAddrsRef.begin(), 1363 LabelAddrsRef.end()); 1364 llvm::sort(LabelAddrs); 1365 LabelAddrs.resize(std::unique(LabelAddrs.begin(), LabelAddrs.end()) - 1366 LabelAddrs.begin()); 1367 // Add the labels. 1368 for (unsigned LabelNum = 0; LabelNum != LabelAddrs.size(); ++LabelNum) { 1369 auto Name = std::make_unique<std::string>(); 1370 *Name = (Twine("L") + Twine(LabelNum)).str(); 1371 SynthesizedLabelNames.push_back(std::move(Name)); 1372 Symbols.push_back(SymbolInfoTy( 1373 LabelAddrs[LabelNum], *SynthesizedLabelNames.back(), ELF::STT_NOTYPE)); 1374 } 1375 llvm::stable_sort(Symbols); 1376 // Recreate the symbolizer with the new symbols list. 1377 RelInfo.reset(Target->createMCRelocationInfo(TripleName, Ctx)); 1378 Symbolizer.reset(Target->createMCSymbolizer( 1379 TripleName, nullptr, nullptr, &Symbols, &Ctx, std::move(RelInfo))); 1380 DisAsm->setSymbolizer(std::move(Symbolizer)); 1381 } 1382 1383 static StringRef getSegmentName(const MachOObjectFile *MachO, 1384 const SectionRef &Section) { 1385 if (MachO) { 1386 DataRefImpl DR = Section.getRawDataRefImpl(); 1387 StringRef SegmentName = MachO->getSectionFinalSegmentName(DR); 1388 return SegmentName; 1389 } 1390 return ""; 1391 } 1392 1393 static void emitPostInstructionInfo(formatted_raw_ostream &FOS, 1394 const MCAsmInfo &MAI, 1395 const MCSubtargetInfo &STI, 1396 StringRef Comments, 1397 LiveVariablePrinter &LVP) { 1398 do { 1399 if (!Comments.empty()) { 1400 // Emit a line of comments. 1401 StringRef Comment; 1402 std::tie(Comment, Comments) = Comments.split('\n'); 1403 // MAI.getCommentColumn() assumes that instructions are printed at the 1404 // position of 8, while getInstStartColumn() returns the actual position. 1405 unsigned CommentColumn = 1406 MAI.getCommentColumn() - 8 + getInstStartColumn(STI); 1407 FOS.PadToColumn(CommentColumn); 1408 FOS << MAI.getCommentString() << ' ' << Comment; 1409 } 1410 LVP.printAfterInst(FOS); 1411 FOS << '\n'; 1412 } while (!Comments.empty()); 1413 FOS.flush(); 1414 } 1415 1416 static void createFakeELFSections(ObjectFile &Obj) { 1417 assert(Obj.isELF()); 1418 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj)) 1419 Elf32LEObj->createFakeSections(); 1420 else if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj)) 1421 Elf64LEObj->createFakeSections(); 1422 else if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj)) 1423 Elf32BEObj->createFakeSections(); 1424 else if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj)) 1425 Elf64BEObj->createFakeSections(); 1426 else 1427 llvm_unreachable("Unsupported binary format"); 1428 } 1429 1430 // Tries to fetch a more complete version of the given object file using its 1431 // Build ID. Returns std::nullopt if nothing was found. 1432 static std::optional<OwningBinary<Binary>> 1433 fetchBinaryByBuildID(const ObjectFile &Obj) { 1434 object::BuildIDRef BuildID = getBuildID(&Obj); 1435 if (BuildID.empty()) 1436 return std::nullopt; 1437 std::optional<std::string> Path = BIDFetcher->fetch(BuildID); 1438 if (!Path) 1439 return std::nullopt; 1440 Expected<OwningBinary<Binary>> DebugBinary = createBinary(*Path); 1441 if (!DebugBinary) { 1442 reportWarning(toString(DebugBinary.takeError()), *Path); 1443 return std::nullopt; 1444 } 1445 return std::move(*DebugBinary); 1446 } 1447 1448 static void 1449 disassembleObject(ObjectFile &Obj, const ObjectFile &DbgObj, 1450 DisassemblerTarget &PrimaryTarget, 1451 std::optional<DisassemblerTarget> &SecondaryTarget, 1452 SourcePrinter &SP, bool InlineRelocs) { 1453 DisassemblerTarget *DT = &PrimaryTarget; 1454 bool PrimaryIsThumb = false; 1455 SmallVector<std::pair<uint64_t, uint64_t>, 0> CHPECodeMap; 1456 1457 if (SecondaryTarget) { 1458 if (isArmElf(Obj)) { 1459 PrimaryIsThumb = 1460 PrimaryTarget.SubtargetInfo->checkFeatures("+thumb-mode"); 1461 } else if (const auto *COFFObj = dyn_cast<COFFObjectFile>(&Obj)) { 1462 const chpe_metadata *CHPEMetadata = COFFObj->getCHPEMetadata(); 1463 if (CHPEMetadata && CHPEMetadata->CodeMapCount) { 1464 uintptr_t CodeMapInt; 1465 cantFail(COFFObj->getRvaPtr(CHPEMetadata->CodeMap, CodeMapInt)); 1466 auto CodeMap = reinterpret_cast<const chpe_range_entry *>(CodeMapInt); 1467 1468 for (uint32_t i = 0; i < CHPEMetadata->CodeMapCount; ++i) { 1469 if (CodeMap[i].getType() == chpe_range_type::Amd64 && 1470 CodeMap[i].Length) { 1471 // Store x86_64 CHPE code ranges. 1472 uint64_t Start = CodeMap[i].getStart() + COFFObj->getImageBase(); 1473 CHPECodeMap.emplace_back(Start, Start + CodeMap[i].Length); 1474 } 1475 } 1476 llvm::sort(CHPECodeMap); 1477 } 1478 } 1479 } 1480 1481 std::map<SectionRef, std::vector<RelocationRef>> RelocMap; 1482 if (InlineRelocs) 1483 RelocMap = getRelocsMap(Obj); 1484 bool Is64Bits = Obj.getBytesInAddress() > 4; 1485 1486 // Create a mapping from virtual address to symbol name. This is used to 1487 // pretty print the symbols while disassembling. 1488 std::map<SectionRef, SectionSymbolsTy> AllSymbols; 1489 std::map<SectionRef, SmallVector<MappingSymbolPair, 0>> AllMappingSymbols; 1490 SectionSymbolsTy AbsoluteSymbols; 1491 const StringRef FileName = Obj.getFileName(); 1492 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(&Obj); 1493 for (const SymbolRef &Symbol : Obj.symbols()) { 1494 Expected<StringRef> NameOrErr = Symbol.getName(); 1495 if (!NameOrErr) { 1496 reportWarning(toString(NameOrErr.takeError()), FileName); 1497 continue; 1498 } 1499 if (NameOrErr->empty() && !(Obj.isXCOFF() && SymbolDescription)) 1500 continue; 1501 1502 if (Obj.isELF() && 1503 (cantFail(Symbol.getFlags()) & SymbolRef::SF_FormatSpecific)) { 1504 // Symbol is intended not to be displayed by default (STT_FILE, 1505 // STT_SECTION, or a mapping symbol). Ignore STT_SECTION symbols. We will 1506 // synthesize a section symbol if no symbol is defined at offset 0. 1507 // 1508 // For a mapping symbol, store it within both AllSymbols and 1509 // AllMappingSymbols. If --show-all-symbols is unspecified, its label will 1510 // not be printed in disassembly listing. 1511 if (getElfSymbolType(Obj, Symbol) != ELF::STT_SECTION && 1512 hasMappingSymbols(Obj)) { 1513 section_iterator SecI = unwrapOrError(Symbol.getSection(), FileName); 1514 if (SecI != Obj.section_end()) { 1515 uint64_t SectionAddr = SecI->getAddress(); 1516 uint64_t Address = cantFail(Symbol.getAddress()); 1517 StringRef Name = *NameOrErr; 1518 if (Name.consume_front("$") && Name.size() && 1519 strchr("adtx", Name[0])) { 1520 AllMappingSymbols[*SecI].emplace_back(Address - SectionAddr, 1521 Name[0]); 1522 AllSymbols[*SecI].push_back( 1523 createSymbolInfo(Obj, Symbol, /*MappingSymbol=*/true)); 1524 } 1525 } 1526 } 1527 continue; 1528 } 1529 1530 if (MachO) { 1531 // __mh_(execute|dylib|dylinker|bundle|preload|object)_header are special 1532 // symbols that support MachO header introspection. They do not bind to 1533 // code locations and are irrelevant for disassembly. 1534 if (NameOrErr->startswith("__mh_") && NameOrErr->endswith("_header")) 1535 continue; 1536 // Don't ask a Mach-O STAB symbol for its section unless you know that 1537 // STAB symbol's section field refers to a valid section index. Otherwise 1538 // the symbol may error trying to load a section that does not exist. 1539 DataRefImpl SymDRI = Symbol.getRawDataRefImpl(); 1540 uint8_t NType = (MachO->is64Bit() ? 1541 MachO->getSymbol64TableEntry(SymDRI).n_type: 1542 MachO->getSymbolTableEntry(SymDRI).n_type); 1543 if (NType & MachO::N_STAB) 1544 continue; 1545 } 1546 1547 section_iterator SecI = unwrapOrError(Symbol.getSection(), FileName); 1548 if (SecI != Obj.section_end()) 1549 AllSymbols[*SecI].push_back(createSymbolInfo(Obj, Symbol)); 1550 else 1551 AbsoluteSymbols.push_back(createSymbolInfo(Obj, Symbol)); 1552 } 1553 1554 if (AllSymbols.empty() && Obj.isELF()) 1555 addDynamicElfSymbols(cast<ELFObjectFileBase>(Obj), AllSymbols); 1556 1557 if (Obj.isWasm()) 1558 addMissingWasmCodeSymbols(cast<WasmObjectFile>(Obj), AllSymbols); 1559 1560 if (Obj.isELF() && Obj.sections().empty()) 1561 createFakeELFSections(Obj); 1562 1563 BumpPtrAllocator A; 1564 StringSaver Saver(A); 1565 addPltEntries(Obj, AllSymbols, Saver); 1566 1567 // Create a mapping from virtual address to section. An empty section can 1568 // cause more than one section at the same address. Sort such sections to be 1569 // before same-addressed non-empty sections so that symbol lookups prefer the 1570 // non-empty section. 1571 std::vector<std::pair<uint64_t, SectionRef>> SectionAddresses; 1572 for (SectionRef Sec : Obj.sections()) 1573 SectionAddresses.emplace_back(Sec.getAddress(), Sec); 1574 llvm::stable_sort(SectionAddresses, [](const auto &LHS, const auto &RHS) { 1575 if (LHS.first != RHS.first) 1576 return LHS.first < RHS.first; 1577 return LHS.second.getSize() < RHS.second.getSize(); 1578 }); 1579 1580 // Linked executables (.exe and .dll files) typically don't include a real 1581 // symbol table but they might contain an export table. 1582 if (const auto *COFFObj = dyn_cast<COFFObjectFile>(&Obj)) { 1583 for (const auto &ExportEntry : COFFObj->export_directories()) { 1584 StringRef Name; 1585 if (Error E = ExportEntry.getSymbolName(Name)) 1586 reportError(std::move(E), Obj.getFileName()); 1587 if (Name.empty()) 1588 continue; 1589 1590 uint32_t RVA; 1591 if (Error E = ExportEntry.getExportRVA(RVA)) 1592 reportError(std::move(E), Obj.getFileName()); 1593 1594 uint64_t VA = COFFObj->getImageBase() + RVA; 1595 auto Sec = partition_point( 1596 SectionAddresses, [VA](const std::pair<uint64_t, SectionRef> &O) { 1597 return O.first <= VA; 1598 }); 1599 if (Sec != SectionAddresses.begin()) { 1600 --Sec; 1601 AllSymbols[Sec->second].emplace_back(VA, Name, ELF::STT_NOTYPE); 1602 } else 1603 AbsoluteSymbols.emplace_back(VA, Name, ELF::STT_NOTYPE); 1604 } 1605 } 1606 1607 // Sort all the symbols, this allows us to use a simple binary search to find 1608 // Multiple symbols can have the same address. Use a stable sort to stabilize 1609 // the output. 1610 StringSet<> FoundDisasmSymbolSet; 1611 for (std::pair<const SectionRef, SectionSymbolsTy> &SecSyms : AllSymbols) 1612 llvm::stable_sort(SecSyms.second); 1613 llvm::stable_sort(AbsoluteSymbols); 1614 1615 std::unique_ptr<DWARFContext> DICtx; 1616 LiveVariablePrinter LVP(*DT->Context->getRegisterInfo(), *DT->SubtargetInfo); 1617 1618 if (DbgVariables != DVDisabled) { 1619 DICtx = DWARFContext::create(DbgObj); 1620 for (const std::unique_ptr<DWARFUnit> &CU : DICtx->compile_units()) 1621 LVP.addCompileUnit(CU->getUnitDIE(false)); 1622 } 1623 1624 LLVM_DEBUG(LVP.dump()); 1625 1626 std::unordered_map<uint64_t, BBAddrMap> AddrToBBAddrMap; 1627 auto ReadBBAddrMap = [&](std::optional<unsigned> SectionIndex = 1628 std::nullopt) { 1629 AddrToBBAddrMap.clear(); 1630 if (const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj)) { 1631 auto BBAddrMapsOrErr = Elf->readBBAddrMap(SectionIndex); 1632 if (!BBAddrMapsOrErr) { 1633 reportWarning(toString(BBAddrMapsOrErr.takeError()), Obj.getFileName()); 1634 return; 1635 } 1636 for (auto &FunctionBBAddrMap : *BBAddrMapsOrErr) 1637 AddrToBBAddrMap.emplace(FunctionBBAddrMap.Addr, 1638 std::move(FunctionBBAddrMap)); 1639 } 1640 }; 1641 1642 // For non-relocatable objects, Read all LLVM_BB_ADDR_MAP sections into a 1643 // single mapping, since they don't have any conflicts. 1644 if (SymbolizeOperands && !Obj.isRelocatableObject()) 1645 ReadBBAddrMap(); 1646 1647 std::optional<llvm::BTFParser> BTF; 1648 if (InlineRelocs && BTFParser::hasBTFSections(Obj)) { 1649 BTF.emplace(); 1650 BTFParser::ParseOptions Opts = {}; 1651 Opts.LoadTypes = true; 1652 Opts.LoadRelocs = true; 1653 if (Error E = BTF->parse(Obj, Opts)) 1654 WithColor::defaultErrorHandler(std::move(E)); 1655 } 1656 1657 for (const SectionRef &Section : ToolSectionFilter(Obj)) { 1658 if (FilterSections.empty() && !DisassembleAll && 1659 (!Section.isText() || Section.isVirtual())) 1660 continue; 1661 1662 uint64_t SectionAddr = Section.getAddress(); 1663 uint64_t SectSize = Section.getSize(); 1664 if (!SectSize) 1665 continue; 1666 1667 // For relocatable object files, read the LLVM_BB_ADDR_MAP section 1668 // corresponding to this section, if present. 1669 if (SymbolizeOperands && Obj.isRelocatableObject()) 1670 ReadBBAddrMap(Section.getIndex()); 1671 1672 // Get the list of all the symbols in this section. 1673 SectionSymbolsTy &Symbols = AllSymbols[Section]; 1674 auto &MappingSymbols = AllMappingSymbols[Section]; 1675 llvm::sort(MappingSymbols); 1676 1677 ArrayRef<uint8_t> Bytes = arrayRefFromStringRef( 1678 unwrapOrError(Section.getContents(), Obj.getFileName())); 1679 1680 std::vector<std::unique_ptr<std::string>> SynthesizedLabelNames; 1681 if (Obj.isELF() && Obj.getArch() == Triple::amdgcn) { 1682 // AMDGPU disassembler uses symbolizer for printing labels 1683 addSymbolizer(*DT->Context, DT->TheTarget, TripleName, DT->DisAsm.get(), 1684 SectionAddr, Bytes, Symbols, SynthesizedLabelNames); 1685 } 1686 1687 StringRef SegmentName = getSegmentName(MachO, Section); 1688 StringRef SectionName = unwrapOrError(Section.getName(), Obj.getFileName()); 1689 // If the section has no symbol at the start, just insert a dummy one. 1690 // Without --show-all-symbols, also insert one if all symbols at the start 1691 // are mapping symbols. 1692 bool CreateDummy = Symbols.empty(); 1693 if (!CreateDummy) { 1694 CreateDummy = true; 1695 for (auto &Sym : Symbols) { 1696 if (Sym.Addr != SectionAddr) 1697 break; 1698 if (!Sym.IsMappingSymbol || ShowAllSymbols) 1699 CreateDummy = false; 1700 } 1701 } 1702 if (CreateDummy) { 1703 SymbolInfoTy Sym = createDummySymbolInfo( 1704 Obj, SectionAddr, SectionName, 1705 Section.isText() ? ELF::STT_FUNC : ELF::STT_OBJECT); 1706 if (Obj.isXCOFF()) 1707 Symbols.insert(Symbols.begin(), Sym); 1708 else 1709 Symbols.insert(llvm::lower_bound(Symbols, Sym), Sym); 1710 } 1711 1712 SmallString<40> Comments; 1713 raw_svector_ostream CommentStream(Comments); 1714 1715 uint64_t VMAAdjustment = 0; 1716 if (shouldAdjustVA(Section)) 1717 VMAAdjustment = AdjustVMA; 1718 1719 // In executable and shared objects, r_offset holds a virtual address. 1720 // Subtract SectionAddr from the r_offset field of a relocation to get 1721 // the section offset. 1722 uint64_t RelAdjustment = Obj.isRelocatableObject() ? 0 : SectionAddr; 1723 uint64_t Size; 1724 uint64_t Index; 1725 bool PrintedSection = false; 1726 std::vector<RelocationRef> Rels = RelocMap[Section]; 1727 std::vector<RelocationRef>::const_iterator RelCur = Rels.begin(); 1728 std::vector<RelocationRef>::const_iterator RelEnd = Rels.end(); 1729 1730 // Loop over each chunk of code between two points where at least 1731 // one symbol is defined. 1732 for (size_t SI = 0, SE = Symbols.size(); SI != SE;) { 1733 // Advance SI past all the symbols starting at the same address, 1734 // and make an ArrayRef of them. 1735 unsigned FirstSI = SI; 1736 uint64_t Start = Symbols[SI].Addr; 1737 ArrayRef<SymbolInfoTy> SymbolsHere; 1738 while (SI != SE && Symbols[SI].Addr == Start) 1739 ++SI; 1740 SymbolsHere = ArrayRef<SymbolInfoTy>(&Symbols[FirstSI], SI - FirstSI); 1741 1742 // Get the demangled names of all those symbols. We end up with a vector 1743 // of StringRef that holds the names we're going to use, and a vector of 1744 // std::string that stores the new strings returned by demangle(), if 1745 // any. If we don't call demangle() then that vector can stay empty. 1746 std::vector<StringRef> SymNamesHere; 1747 std::vector<std::string> DemangledSymNamesHere; 1748 if (Demangle) { 1749 // Fetch the demangled names and store them locally. 1750 for (const SymbolInfoTy &Symbol : SymbolsHere) 1751 DemangledSymNamesHere.push_back(demangle(Symbol.Name)); 1752 // Now we've finished modifying that vector, it's safe to make 1753 // a vector of StringRefs pointing into it. 1754 SymNamesHere.insert(SymNamesHere.begin(), DemangledSymNamesHere.begin(), 1755 DemangledSymNamesHere.end()); 1756 } else { 1757 for (const SymbolInfoTy &Symbol : SymbolsHere) 1758 SymNamesHere.push_back(Symbol.Name); 1759 } 1760 1761 // Distinguish ELF data from code symbols, which will be used later on to 1762 // decide whether to 'disassemble' this chunk as a data declaration via 1763 // dumpELFData(), or whether to treat it as code. 1764 // 1765 // If data _and_ code symbols are defined at the same address, the code 1766 // takes priority, on the grounds that disassembling code is our main 1767 // purpose here, and it would be a worse failure to _not_ interpret 1768 // something that _was_ meaningful as code than vice versa. 1769 // 1770 // Any ELF symbol type that is not clearly data will be regarded as code. 1771 // In particular, one of the uses of STT_NOTYPE is for branch targets 1772 // inside functions, for which STT_FUNC would be inaccurate. 1773 // 1774 // So here, we spot whether there's any non-data symbol present at all, 1775 // and only set the DisassembleAsData flag if there isn't. Also, we use 1776 // this distinction to inform the decision of which symbol to print at 1777 // the head of the section, so that if we're printing code, we print a 1778 // code-related symbol name to go with it. 1779 bool DisassembleAsData = false; 1780 size_t DisplaySymIndex = SymbolsHere.size() - 1; 1781 if (Obj.isELF() && !DisassembleAll && Section.isText()) { 1782 DisassembleAsData = true; // unless we find a code symbol below 1783 1784 for (size_t i = 0; i < SymbolsHere.size(); ++i) { 1785 uint8_t SymTy = SymbolsHere[i].Type; 1786 if (SymTy != ELF::STT_OBJECT && SymTy != ELF::STT_COMMON) { 1787 DisassembleAsData = false; 1788 DisplaySymIndex = i; 1789 } 1790 } 1791 } 1792 1793 // Decide which symbol(s) from this collection we're going to print. 1794 std::vector<bool> SymsToPrint(SymbolsHere.size(), false); 1795 // If the user has given the --disassemble-symbols option, then we must 1796 // display every symbol in that set, and no others. 1797 if (!DisasmSymbolSet.empty()) { 1798 bool FoundAny = false; 1799 for (size_t i = 0; i < SymbolsHere.size(); ++i) { 1800 if (DisasmSymbolSet.count(SymNamesHere[i])) { 1801 SymsToPrint[i] = true; 1802 FoundAny = true; 1803 } 1804 } 1805 1806 // And if none of the symbols here is one that the user asked for, skip 1807 // disassembling this entire chunk of code. 1808 if (!FoundAny) 1809 continue; 1810 } else if (!SymbolsHere[DisplaySymIndex].IsMappingSymbol) { 1811 // Otherwise, print whichever symbol at this location is last in the 1812 // Symbols array, because that array is pre-sorted in a way intended to 1813 // correlate with priority of which symbol to display. 1814 SymsToPrint[DisplaySymIndex] = true; 1815 } 1816 1817 // Now that we know we're disassembling this section, override the choice 1818 // of which symbols to display by printing _all_ of them at this address 1819 // if the user asked for all symbols. 1820 // 1821 // That way, '--show-all-symbols --disassemble-symbol=foo' will print 1822 // only the chunk of code headed by 'foo', but also show any other 1823 // symbols defined at that address, such as aliases for 'foo', or the ARM 1824 // mapping symbol preceding its code. 1825 if (ShowAllSymbols) { 1826 for (size_t i = 0; i < SymbolsHere.size(); ++i) 1827 SymsToPrint[i] = true; 1828 } 1829 1830 if (Start < SectionAddr || StopAddress <= Start) 1831 continue; 1832 1833 for (size_t i = 0; i < SymbolsHere.size(); ++i) 1834 FoundDisasmSymbolSet.insert(SymNamesHere[i]); 1835 1836 // The end is the section end, the beginning of the next symbol, or 1837 // --stop-address. 1838 uint64_t End = std::min<uint64_t>(SectionAddr + SectSize, StopAddress); 1839 if (SI < SE) 1840 End = std::min(End, Symbols[SI].Addr); 1841 if (Start >= End || End <= StartAddress) 1842 continue; 1843 Start -= SectionAddr; 1844 End -= SectionAddr; 1845 1846 if (!PrintedSection) { 1847 PrintedSection = true; 1848 outs() << "\nDisassembly of section "; 1849 if (!SegmentName.empty()) 1850 outs() << SegmentName << ","; 1851 outs() << SectionName << ":\n"; 1852 } 1853 1854 bool PrintedLabel = false; 1855 for (size_t i = 0; i < SymbolsHere.size(); ++i) { 1856 if (!SymsToPrint[i]) 1857 continue; 1858 1859 const SymbolInfoTy &Symbol = SymbolsHere[i]; 1860 const StringRef SymbolName = SymNamesHere[i]; 1861 1862 if (!PrintedLabel) { 1863 outs() << '\n'; 1864 PrintedLabel = true; 1865 } 1866 if (LeadingAddr) 1867 outs() << format(Is64Bits ? "%016" PRIx64 " " : "%08" PRIx64 " ", 1868 SectionAddr + Start + VMAAdjustment); 1869 if (Obj.isXCOFF() && SymbolDescription) { 1870 outs() << getXCOFFSymbolDescription(Symbol, SymbolName) << ":\n"; 1871 } else 1872 outs() << '<' << SymbolName << ">:\n"; 1873 } 1874 1875 // Don't print raw contents of a virtual section. A virtual section 1876 // doesn't have any contents in the file. 1877 if (Section.isVirtual()) { 1878 outs() << "...\n"; 1879 continue; 1880 } 1881 1882 // See if any of the symbols defined at this location triggers target- 1883 // specific disassembly behavior, e.g. of special descriptors or function 1884 // prelude information. 1885 // 1886 // We stop this loop at the first symbol that triggers some kind of 1887 // interesting behavior (if any), on the assumption that if two symbols 1888 // defined at the same address trigger two conflicting symbol handlers, 1889 // the object file is probably confused anyway, and it would make even 1890 // less sense to present the output of _both_ handlers, because that 1891 // would describe the same data twice. 1892 for (size_t SHI = 0; SHI < SymbolsHere.size(); ++SHI) { 1893 SymbolInfoTy Symbol = SymbolsHere[SHI]; 1894 1895 auto Status = DT->DisAsm->onSymbolStart( 1896 Symbol, Size, Bytes.slice(Start, End - Start), SectionAddr + Start, 1897 CommentStream); 1898 1899 if (!Status) { 1900 // If onSymbolStart returns std::nullopt, that means it didn't trigger 1901 // any interesting handling for this symbol. Try the other symbols 1902 // defined at this address. 1903 continue; 1904 } 1905 1906 if (*Status == MCDisassembler::Fail) { 1907 // If onSymbolStart returns Fail, that means it identified some kind 1908 // of special data at this address, but wasn't able to disassemble it 1909 // meaningfully. So we fall back to disassembling the failed region 1910 // as bytes, assuming that the target detected the failure before 1911 // printing anything. 1912 // 1913 // Return values Success or SoftFail (i.e no 'real' failure) are 1914 // expected to mean that the target has emitted its own output. 1915 // 1916 // Either way, 'Size' will have been set to the amount of data 1917 // covered by whatever prologue the target identified. So we advance 1918 // our own position to beyond that. Sometimes that will be the entire 1919 // distance to the next symbol, and sometimes it will be just a 1920 // prologue and we should start disassembling instructions from where 1921 // it left off. 1922 outs() << DT->Context->getAsmInfo()->getCommentString() 1923 << " error in decoding " << SymNamesHere[SHI] 1924 << " : decoding failed region as bytes.\n"; 1925 for (uint64_t I = 0; I < Size; ++I) { 1926 outs() << "\t.byte\t " << format_hex(Bytes[I], 1, /*Upper=*/true) 1927 << "\n"; 1928 } 1929 } 1930 Start += Size; 1931 break; 1932 } 1933 1934 Index = Start; 1935 if (SectionAddr < StartAddress) 1936 Index = std::max<uint64_t>(Index, StartAddress - SectionAddr); 1937 1938 if (DisassembleAsData) { 1939 dumpELFData(SectionAddr, Index, End, Bytes); 1940 Index = End; 1941 continue; 1942 } 1943 1944 bool DumpARMELFData = false; 1945 bool DumpTracebackTableForXCOFFFunction = 1946 Obj.isXCOFF() && Section.isText() && TracebackTable && 1947 Symbols[SI - 1].XCOFFSymInfo.StorageMappingClass && 1948 (*Symbols[SI - 1].XCOFFSymInfo.StorageMappingClass == XCOFF::XMC_PR); 1949 1950 formatted_raw_ostream FOS(outs()); 1951 1952 // FIXME: Workaround for bug in formatted_raw_ostream. Color escape codes 1953 // are (incorrectly) written directly to the unbuffered raw_ostream 1954 // wrapped by the formatted_raw_ostream. 1955 if (DisassemblyColor == ColorOutput::Enable || 1956 DisassemblyColor == ColorOutput::Auto) 1957 FOS.SetUnbuffered(); 1958 1959 std::unordered_map<uint64_t, std::string> AllLabels; 1960 std::unordered_map<uint64_t, std::vector<std::string>> BBAddrMapLabels; 1961 if (SymbolizeOperands) { 1962 collectLocalBranchTargets(Bytes, DT->InstrAnalysis.get(), 1963 DT->DisAsm.get(), DT->InstPrinter.get(), 1964 PrimaryTarget.SubtargetInfo.get(), 1965 SectionAddr, Index, End, AllLabels); 1966 collectBBAddrMapLabels(AddrToBBAddrMap, SectionAddr, Index, End, 1967 BBAddrMapLabels); 1968 } 1969 1970 while (Index < End) { 1971 // ARM and AArch64 ELF binaries can interleave data and text in the 1972 // same section. We rely on the markers introduced to understand what 1973 // we need to dump. If the data marker is within a function, it is 1974 // denoted as a word/short etc. 1975 if (!MappingSymbols.empty()) { 1976 char Kind = getMappingSymbolKind(MappingSymbols, Index); 1977 DumpARMELFData = Kind == 'd'; 1978 if (SecondaryTarget) { 1979 if (Kind == 'a') { 1980 DT = PrimaryIsThumb ? &*SecondaryTarget : &PrimaryTarget; 1981 } else if (Kind == 't') { 1982 DT = PrimaryIsThumb ? &PrimaryTarget : &*SecondaryTarget; 1983 } 1984 } 1985 } else if (!CHPECodeMap.empty()) { 1986 uint64_t Address = SectionAddr + Index; 1987 auto It = partition_point( 1988 CHPECodeMap, 1989 [Address](const std::pair<uint64_t, uint64_t> &Entry) { 1990 return Entry.first <= Address; 1991 }); 1992 if (It != CHPECodeMap.begin() && Address < (It - 1)->second) { 1993 DT = &*SecondaryTarget; 1994 } else { 1995 DT = &PrimaryTarget; 1996 // X64 disassembler range may have left Index unaligned, so 1997 // make sure that it's aligned when we switch back to ARM64 1998 // code. 1999 Index = llvm::alignTo(Index, 4); 2000 if (Index >= End) 2001 break; 2002 } 2003 } 2004 2005 if (DumpARMELFData) { 2006 Size = dumpARMELFData(SectionAddr, Index, End, Obj, Bytes, 2007 MappingSymbols, *DT->SubtargetInfo, FOS); 2008 } else { 2009 // When -z or --disassemble-zeroes are given we always dissasemble 2010 // them. Otherwise we might want to skip zero bytes we see. 2011 if (!DisassembleZeroes) { 2012 uint64_t MaxOffset = End - Index; 2013 // For --reloc: print zero blocks patched by relocations, so that 2014 // relocations can be shown in the dump. 2015 if (RelCur != RelEnd) 2016 MaxOffset = std::min(RelCur->getOffset() - RelAdjustment - Index, 2017 MaxOffset); 2018 2019 if (size_t N = 2020 countSkippableZeroBytes(Bytes.slice(Index, MaxOffset))) { 2021 FOS << "\t\t..." << '\n'; 2022 Index += N; 2023 continue; 2024 } 2025 } 2026 2027 if (DumpTracebackTableForXCOFFFunction && 2028 doesXCOFFTracebackTableBegin(Bytes.slice(Index, 4))) { 2029 dumpTracebackTable(Bytes.slice(Index), 2030 SectionAddr + Index + VMAAdjustment, FOS, 2031 SectionAddr + End + VMAAdjustment, 2032 *DT->SubtargetInfo, cast<XCOFFObjectFile>(&Obj)); 2033 Index = End; 2034 continue; 2035 } 2036 2037 // Print local label if there's any. 2038 auto Iter1 = BBAddrMapLabels.find(SectionAddr + Index); 2039 if (Iter1 != BBAddrMapLabels.end()) { 2040 for (StringRef Label : Iter1->second) 2041 FOS << "<" << Label << ">:\n"; 2042 } else { 2043 auto Iter2 = AllLabels.find(SectionAddr + Index); 2044 if (Iter2 != AllLabels.end()) 2045 FOS << "<" << Iter2->second << ">:\n"; 2046 } 2047 2048 // Disassemble a real instruction or a data when disassemble all is 2049 // provided 2050 MCInst Inst; 2051 ArrayRef<uint8_t> ThisBytes = Bytes.slice(Index); 2052 uint64_t ThisAddr = SectionAddr + Index; 2053 bool Disassembled = DT->DisAsm->getInstruction( 2054 Inst, Size, ThisBytes, ThisAddr, CommentStream); 2055 if (Size == 0) 2056 Size = std::min<uint64_t>( 2057 ThisBytes.size(), 2058 DT->DisAsm->suggestBytesToSkip(ThisBytes, ThisAddr)); 2059 2060 LVP.update({Index, Section.getIndex()}, 2061 {Index + Size, Section.getIndex()}, Index + Size != End); 2062 2063 DT->InstPrinter->setCommentStream(CommentStream); 2064 2065 DT->Printer->printInst( 2066 *DT->InstPrinter, Disassembled ? &Inst : nullptr, 2067 Bytes.slice(Index, Size), 2068 {SectionAddr + Index + VMAAdjustment, Section.getIndex()}, FOS, 2069 "", *DT->SubtargetInfo, &SP, Obj.getFileName(), &Rels, LVP); 2070 2071 DT->InstPrinter->setCommentStream(llvm::nulls()); 2072 2073 // If disassembly has failed, avoid analysing invalid/incomplete 2074 // instruction information. Otherwise, try to resolve the target 2075 // address (jump target or memory operand address) and print it on the 2076 // right of the instruction. 2077 if (Disassembled && DT->InstrAnalysis) { 2078 // Branch targets are printed just after the instructions. 2079 llvm::raw_ostream *TargetOS = &FOS; 2080 uint64_t Target; 2081 bool PrintTarget = DT->InstrAnalysis->evaluateBranch( 2082 Inst, SectionAddr + Index, Size, Target); 2083 if (!PrintTarget) 2084 if (std::optional<uint64_t> MaybeTarget = 2085 DT->InstrAnalysis->evaluateMemoryOperandAddress( 2086 Inst, DT->SubtargetInfo.get(), SectionAddr + Index, 2087 Size)) { 2088 Target = *MaybeTarget; 2089 PrintTarget = true; 2090 // Do not print real address when symbolizing. 2091 if (!SymbolizeOperands) { 2092 // Memory operand addresses are printed as comments. 2093 TargetOS = &CommentStream; 2094 *TargetOS << "0x" << Twine::utohexstr(Target); 2095 } 2096 } 2097 if (PrintTarget) { 2098 // In a relocatable object, the target's section must reside in 2099 // the same section as the call instruction or it is accessed 2100 // through a relocation. 2101 // 2102 // In a non-relocatable object, the target may be in any section. 2103 // In that case, locate the section(s) containing the target 2104 // address and find the symbol in one of those, if possible. 2105 // 2106 // N.B. We don't walk the relocations in the relocatable case yet. 2107 std::vector<const SectionSymbolsTy *> TargetSectionSymbols; 2108 if (!Obj.isRelocatableObject()) { 2109 auto It = llvm::partition_point( 2110 SectionAddresses, 2111 [=](const std::pair<uint64_t, SectionRef> &O) { 2112 return O.first <= Target; 2113 }); 2114 uint64_t TargetSecAddr = 0; 2115 while (It != SectionAddresses.begin()) { 2116 --It; 2117 if (TargetSecAddr == 0) 2118 TargetSecAddr = It->first; 2119 if (It->first != TargetSecAddr) 2120 break; 2121 TargetSectionSymbols.push_back(&AllSymbols[It->second]); 2122 } 2123 } else { 2124 TargetSectionSymbols.push_back(&Symbols); 2125 } 2126 TargetSectionSymbols.push_back(&AbsoluteSymbols); 2127 2128 // Find the last symbol in the first candidate section whose 2129 // offset is less than or equal to the target. If there are no 2130 // such symbols, try in the next section and so on, before finally 2131 // using the nearest preceding absolute symbol (if any), if there 2132 // are no other valid symbols. 2133 const SymbolInfoTy *TargetSym = nullptr; 2134 for (const SectionSymbolsTy *TargetSymbols : 2135 TargetSectionSymbols) { 2136 auto It = llvm::partition_point( 2137 *TargetSymbols, 2138 [=](const SymbolInfoTy &O) { return O.Addr <= Target; }); 2139 while (It != TargetSymbols->begin()) { 2140 --It; 2141 // Skip mapping symbols to avoid possible ambiguity as they 2142 // do not allow uniquely identifying the target address. 2143 if (!It->IsMappingSymbol) { 2144 TargetSym = &*It; 2145 break; 2146 } 2147 } 2148 if (TargetSym) 2149 break; 2150 } 2151 2152 // Print the labels corresponding to the target if there's any. 2153 bool BBAddrMapLabelAvailable = BBAddrMapLabels.count(Target); 2154 bool LabelAvailable = AllLabels.count(Target); 2155 if (TargetSym != nullptr) { 2156 uint64_t TargetAddress = TargetSym->Addr; 2157 uint64_t Disp = Target - TargetAddress; 2158 std::string TargetName = Demangle ? demangle(TargetSym->Name) 2159 : TargetSym->Name.str(); 2160 2161 *TargetOS << " <"; 2162 if (!Disp) { 2163 // Always Print the binary symbol precisely corresponding to 2164 // the target address. 2165 *TargetOS << TargetName; 2166 } else if (BBAddrMapLabelAvailable) { 2167 *TargetOS << BBAddrMapLabels[Target].front(); 2168 } else if (LabelAvailable) { 2169 *TargetOS << AllLabels[Target]; 2170 } else { 2171 // Always Print the binary symbol plus an offset if there's no 2172 // local label corresponding to the target address. 2173 *TargetOS << TargetName << "+0x" << Twine::utohexstr(Disp); 2174 } 2175 *TargetOS << ">"; 2176 } else if (BBAddrMapLabelAvailable) { 2177 *TargetOS << " <" << BBAddrMapLabels[Target].front() << ">"; 2178 } else if (LabelAvailable) { 2179 *TargetOS << " <" << AllLabels[Target] << ">"; 2180 } 2181 // By convention, each record in the comment stream should be 2182 // terminated. 2183 if (TargetOS == &CommentStream) 2184 *TargetOS << "\n"; 2185 } 2186 } 2187 } 2188 2189 assert(DT->Context->getAsmInfo()); 2190 emitPostInstructionInfo(FOS, *DT->Context->getAsmInfo(), 2191 *DT->SubtargetInfo, CommentStream.str(), LVP); 2192 Comments.clear(); 2193 2194 if (BTF) 2195 printBTFRelocation(FOS, *BTF, {Index, Section.getIndex()}, LVP); 2196 2197 // Hexagon does this in pretty printer 2198 if (Obj.getArch() != Triple::hexagon) { 2199 // Print relocation for instruction and data. 2200 while (RelCur != RelEnd) { 2201 uint64_t Offset = RelCur->getOffset() - RelAdjustment; 2202 // If this relocation is hidden, skip it. 2203 if (getHidden(*RelCur) || SectionAddr + Offset < StartAddress) { 2204 ++RelCur; 2205 continue; 2206 } 2207 2208 // Stop when RelCur's offset is past the disassembled 2209 // instruction/data. Note that it's possible the disassembled data 2210 // is not the complete data: we might see the relocation printed in 2211 // the middle of the data, but this matches the binutils objdump 2212 // output. 2213 if (Offset >= Index + Size) 2214 break; 2215 2216 // When --adjust-vma is used, update the address printed. 2217 if (RelCur->getSymbol() != Obj.symbol_end()) { 2218 Expected<section_iterator> SymSI = 2219 RelCur->getSymbol()->getSection(); 2220 if (SymSI && *SymSI != Obj.section_end() && 2221 shouldAdjustVA(**SymSI)) 2222 Offset += AdjustVMA; 2223 } 2224 2225 printRelocation(FOS, Obj.getFileName(), *RelCur, 2226 SectionAddr + Offset, Is64Bits); 2227 LVP.printAfterOtherLine(FOS, true); 2228 ++RelCur; 2229 } 2230 } 2231 2232 Index += Size; 2233 } 2234 } 2235 } 2236 StringSet<> MissingDisasmSymbolSet = 2237 set_difference(DisasmSymbolSet, FoundDisasmSymbolSet); 2238 for (StringRef Sym : MissingDisasmSymbolSet.keys()) 2239 reportWarning("failed to disassemble missing symbol " + Sym, FileName); 2240 } 2241 2242 static void disassembleObject(ObjectFile *Obj, bool InlineRelocs) { 2243 // If information useful for showing the disassembly is missing, try to find a 2244 // more complete binary and disassemble that instead. 2245 OwningBinary<Binary> FetchedBinary; 2246 if (Obj->symbols().empty()) { 2247 if (std::optional<OwningBinary<Binary>> FetchedBinaryOpt = 2248 fetchBinaryByBuildID(*Obj)) { 2249 if (auto *O = dyn_cast<ObjectFile>(FetchedBinaryOpt->getBinary())) { 2250 if (!O->symbols().empty() || 2251 (!O->sections().empty() && Obj->sections().empty())) { 2252 FetchedBinary = std::move(*FetchedBinaryOpt); 2253 Obj = O; 2254 } 2255 } 2256 } 2257 } 2258 2259 const Target *TheTarget = getTarget(Obj); 2260 2261 // Package up features to be passed to target/subtarget 2262 Expected<SubtargetFeatures> FeaturesValue = Obj->getFeatures(); 2263 if (!FeaturesValue) 2264 reportError(FeaturesValue.takeError(), Obj->getFileName()); 2265 SubtargetFeatures Features = *FeaturesValue; 2266 if (!MAttrs.empty()) { 2267 for (unsigned I = 0; I != MAttrs.size(); ++I) 2268 Features.AddFeature(MAttrs[I]); 2269 } else if (MCPU.empty() && Obj->getArch() == llvm::Triple::aarch64) { 2270 Features.AddFeature("+all"); 2271 } 2272 2273 if (MCPU.empty()) 2274 MCPU = Obj->tryGetCPUName().value_or("").str(); 2275 2276 if (isArmElf(*Obj)) { 2277 // When disassembling big-endian Arm ELF, the instruction endianness is 2278 // determined in a complex way. In relocatable objects, AAELF32 mandates 2279 // that instruction endianness matches the ELF file endianness; in 2280 // executable images, that's true unless the file header has the EF_ARM_BE8 2281 // flag, in which case instructions are little-endian regardless of data 2282 // endianness. 2283 // 2284 // We must set the big-endian-instructions SubtargetFeature to make the 2285 // disassembler read the instructions the right way round, and also tell 2286 // our own prettyprinter to retrieve the encodings the same way to print in 2287 // hex. 2288 const auto *Elf32BE = dyn_cast<ELF32BEObjectFile>(Obj); 2289 2290 if (Elf32BE && (Elf32BE->isRelocatableObject() || 2291 !(Elf32BE->getPlatformFlags() & ELF::EF_ARM_BE8))) { 2292 Features.AddFeature("+big-endian-instructions"); 2293 ARMPrettyPrinterInst.setInstructionEndianness(llvm::endianness::big); 2294 } else { 2295 ARMPrettyPrinterInst.setInstructionEndianness(llvm::endianness::little); 2296 } 2297 } 2298 2299 DisassemblerTarget PrimaryTarget(TheTarget, *Obj, TripleName, MCPU, Features); 2300 2301 // If we have an ARM object file, we need a second disassembler, because 2302 // ARM CPUs have two different instruction sets: ARM mode, and Thumb mode. 2303 // We use mapping symbols to switch between the two assemblers, where 2304 // appropriate. 2305 std::optional<DisassemblerTarget> SecondaryTarget; 2306 2307 if (isArmElf(*Obj)) { 2308 if (!PrimaryTarget.SubtargetInfo->checkFeatures("+mclass")) { 2309 if (PrimaryTarget.SubtargetInfo->checkFeatures("+thumb-mode")) 2310 Features.AddFeature("-thumb-mode"); 2311 else 2312 Features.AddFeature("+thumb-mode"); 2313 SecondaryTarget.emplace(PrimaryTarget, Features); 2314 } 2315 } else if (const auto *COFFObj = dyn_cast<COFFObjectFile>(Obj)) { 2316 const chpe_metadata *CHPEMetadata = COFFObj->getCHPEMetadata(); 2317 if (CHPEMetadata && CHPEMetadata->CodeMapCount) { 2318 // Set up x86_64 disassembler for ARM64EC binaries. 2319 Triple X64Triple(TripleName); 2320 X64Triple.setArch(Triple::ArchType::x86_64); 2321 2322 std::string Error; 2323 const Target *X64Target = 2324 TargetRegistry::lookupTarget("", X64Triple, Error); 2325 if (X64Target) { 2326 SubtargetFeatures X64Features; 2327 SecondaryTarget.emplace(X64Target, *Obj, X64Triple.getTriple(), "", 2328 X64Features); 2329 } else { 2330 reportWarning(Error, Obj->getFileName()); 2331 } 2332 } 2333 } 2334 2335 const ObjectFile *DbgObj = Obj; 2336 if (!FetchedBinary.getBinary() && !Obj->hasDebugInfo()) { 2337 if (std::optional<OwningBinary<Binary>> DebugBinaryOpt = 2338 fetchBinaryByBuildID(*Obj)) { 2339 if (auto *FetchedObj = 2340 dyn_cast<const ObjectFile>(DebugBinaryOpt->getBinary())) { 2341 if (FetchedObj->hasDebugInfo()) { 2342 FetchedBinary = std::move(*DebugBinaryOpt); 2343 DbgObj = FetchedObj; 2344 } 2345 } 2346 } 2347 } 2348 2349 std::unique_ptr<object::Binary> DSYMBinary; 2350 std::unique_ptr<MemoryBuffer> DSYMBuf; 2351 if (!DbgObj->hasDebugInfo()) { 2352 if (const MachOObjectFile *MachOOF = dyn_cast<MachOObjectFile>(&*Obj)) { 2353 DbgObj = objdump::getMachODSymObject(MachOOF, Obj->getFileName(), 2354 DSYMBinary, DSYMBuf); 2355 if (!DbgObj) 2356 return; 2357 } 2358 } 2359 2360 SourcePrinter SP(DbgObj, TheTarget->getName()); 2361 2362 for (StringRef Opt : DisassemblerOptions) 2363 if (!PrimaryTarget.InstPrinter->applyTargetSpecificCLOption(Opt)) 2364 reportError(Obj->getFileName(), 2365 "Unrecognized disassembler option: " + Opt); 2366 2367 disassembleObject(*Obj, *DbgObj, PrimaryTarget, SecondaryTarget, SP, 2368 InlineRelocs); 2369 } 2370 2371 void Dumper::printRelocations() { 2372 StringRef Fmt = O.getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 2373 2374 // Build a mapping from relocation target to a vector of relocation 2375 // sections. Usually, there is an only one relocation section for 2376 // each relocated section. 2377 MapVector<SectionRef, std::vector<SectionRef>> SecToRelSec; 2378 uint64_t Ndx; 2379 for (const SectionRef &Section : ToolSectionFilter(O, &Ndx)) { 2380 if (O.isELF() && (ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC)) 2381 continue; 2382 if (Section.relocation_begin() == Section.relocation_end()) 2383 continue; 2384 Expected<section_iterator> SecOrErr = Section.getRelocatedSection(); 2385 if (!SecOrErr) 2386 reportError(O.getFileName(), 2387 "section (" + Twine(Ndx) + 2388 "): unable to get a relocation target: " + 2389 toString(SecOrErr.takeError())); 2390 SecToRelSec[**SecOrErr].push_back(Section); 2391 } 2392 2393 for (std::pair<SectionRef, std::vector<SectionRef>> &P : SecToRelSec) { 2394 StringRef SecName = unwrapOrError(P.first.getName(), O.getFileName()); 2395 outs() << "\nRELOCATION RECORDS FOR [" << SecName << "]:\n"; 2396 uint32_t OffsetPadding = (O.getBytesInAddress() > 4 ? 16 : 8); 2397 uint32_t TypePadding = 24; 2398 outs() << left_justify("OFFSET", OffsetPadding) << " " 2399 << left_justify("TYPE", TypePadding) << " " 2400 << "VALUE\n"; 2401 2402 for (SectionRef Section : P.second) { 2403 for (const RelocationRef &Reloc : Section.relocations()) { 2404 uint64_t Address = Reloc.getOffset(); 2405 SmallString<32> RelocName; 2406 SmallString<32> ValueStr; 2407 if (Address < StartAddress || Address > StopAddress || getHidden(Reloc)) 2408 continue; 2409 Reloc.getTypeName(RelocName); 2410 if (Error E = getRelocationValueString(Reloc, ValueStr)) 2411 reportUniqueWarning(std::move(E)); 2412 2413 outs() << format(Fmt.data(), Address) << " " 2414 << left_justify(RelocName, TypePadding) << " " << ValueStr 2415 << "\n"; 2416 } 2417 } 2418 } 2419 } 2420 2421 // Returns true if we need to show LMA column when dumping section headers. We 2422 // show it only when the platform is ELF and either we have at least one section 2423 // whose VMA and LMA are different and/or when --show-lma flag is used. 2424 static bool shouldDisplayLMA(const ObjectFile &Obj) { 2425 if (!Obj.isELF()) 2426 return false; 2427 for (const SectionRef &S : ToolSectionFilter(Obj)) 2428 if (S.getAddress() != getELFSectionLMA(S)) 2429 return true; 2430 return ShowLMA; 2431 } 2432 2433 static size_t getMaxSectionNameWidth(const ObjectFile &Obj) { 2434 // Default column width for names is 13 even if no names are that long. 2435 size_t MaxWidth = 13; 2436 for (const SectionRef &Section : ToolSectionFilter(Obj)) { 2437 StringRef Name = unwrapOrError(Section.getName(), Obj.getFileName()); 2438 MaxWidth = std::max(MaxWidth, Name.size()); 2439 } 2440 return MaxWidth; 2441 } 2442 2443 void objdump::printSectionHeaders(ObjectFile &Obj) { 2444 if (Obj.isELF() && Obj.sections().empty()) 2445 createFakeELFSections(Obj); 2446 2447 size_t NameWidth = getMaxSectionNameWidth(Obj); 2448 size_t AddressWidth = 2 * Obj.getBytesInAddress(); 2449 bool HasLMAColumn = shouldDisplayLMA(Obj); 2450 outs() << "\nSections:\n"; 2451 if (HasLMAColumn) 2452 outs() << "Idx " << left_justify("Name", NameWidth) << " Size " 2453 << left_justify("VMA", AddressWidth) << " " 2454 << left_justify("LMA", AddressWidth) << " Type\n"; 2455 else 2456 outs() << "Idx " << left_justify("Name", NameWidth) << " Size " 2457 << left_justify("VMA", AddressWidth) << " Type\n"; 2458 2459 uint64_t Idx; 2460 for (const SectionRef &Section : ToolSectionFilter(Obj, &Idx)) { 2461 StringRef Name = unwrapOrError(Section.getName(), Obj.getFileName()); 2462 uint64_t VMA = Section.getAddress(); 2463 if (shouldAdjustVA(Section)) 2464 VMA += AdjustVMA; 2465 2466 uint64_t Size = Section.getSize(); 2467 2468 std::string Type = Section.isText() ? "TEXT" : ""; 2469 if (Section.isData()) 2470 Type += Type.empty() ? "DATA" : ", DATA"; 2471 if (Section.isBSS()) 2472 Type += Type.empty() ? "BSS" : ", BSS"; 2473 if (Section.isDebugSection()) 2474 Type += Type.empty() ? "DEBUG" : ", DEBUG"; 2475 2476 if (HasLMAColumn) 2477 outs() << format("%3" PRIu64 " %-*s %08" PRIx64 " ", Idx, NameWidth, 2478 Name.str().c_str(), Size) 2479 << format_hex_no_prefix(VMA, AddressWidth) << " " 2480 << format_hex_no_prefix(getELFSectionLMA(Section), AddressWidth) 2481 << " " << Type << "\n"; 2482 else 2483 outs() << format("%3" PRIu64 " %-*s %08" PRIx64 " ", Idx, NameWidth, 2484 Name.str().c_str(), Size) 2485 << format_hex_no_prefix(VMA, AddressWidth) << " " << Type << "\n"; 2486 } 2487 } 2488 2489 void objdump::printSectionContents(const ObjectFile *Obj) { 2490 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(Obj); 2491 2492 for (const SectionRef &Section : ToolSectionFilter(*Obj)) { 2493 StringRef Name = unwrapOrError(Section.getName(), Obj->getFileName()); 2494 uint64_t BaseAddr = Section.getAddress(); 2495 uint64_t Size = Section.getSize(); 2496 if (!Size) 2497 continue; 2498 2499 outs() << "Contents of section "; 2500 StringRef SegmentName = getSegmentName(MachO, Section); 2501 if (!SegmentName.empty()) 2502 outs() << SegmentName << ","; 2503 outs() << Name << ":\n"; 2504 if (Section.isBSS()) { 2505 outs() << format("<skipping contents of bss section at [%04" PRIx64 2506 ", %04" PRIx64 ")>\n", 2507 BaseAddr, BaseAddr + Size); 2508 continue; 2509 } 2510 2511 StringRef Contents = unwrapOrError(Section.getContents(), Obj->getFileName()); 2512 2513 // Dump out the content as hex and printable ascii characters. 2514 for (std::size_t Addr = 0, End = Contents.size(); Addr < End; Addr += 16) { 2515 outs() << format(" %04" PRIx64 " ", BaseAddr + Addr); 2516 // Dump line of hex. 2517 for (std::size_t I = 0; I < 16; ++I) { 2518 if (I != 0 && I % 4 == 0) 2519 outs() << ' '; 2520 if (Addr + I < End) 2521 outs() << hexdigit((Contents[Addr + I] >> 4) & 0xF, true) 2522 << hexdigit(Contents[Addr + I] & 0xF, true); 2523 else 2524 outs() << " "; 2525 } 2526 // Print ascii. 2527 outs() << " "; 2528 for (std::size_t I = 0; I < 16 && Addr + I < End; ++I) { 2529 if (isPrint(static_cast<unsigned char>(Contents[Addr + I]) & 0xFF)) 2530 outs() << Contents[Addr + I]; 2531 else 2532 outs() << "."; 2533 } 2534 outs() << "\n"; 2535 } 2536 } 2537 } 2538 2539 void Dumper::printSymbolTable(StringRef ArchiveName, StringRef ArchitectureName, 2540 bool DumpDynamic) { 2541 if (O.isCOFF() && !DumpDynamic) { 2542 outs() << "\nSYMBOL TABLE:\n"; 2543 printCOFFSymbolTable(cast<const COFFObjectFile>(O)); 2544 return; 2545 } 2546 2547 const StringRef FileName = O.getFileName(); 2548 2549 if (!DumpDynamic) { 2550 outs() << "\nSYMBOL TABLE:\n"; 2551 for (auto I = O.symbol_begin(); I != O.symbol_end(); ++I) 2552 printSymbol(*I, {}, FileName, ArchiveName, ArchitectureName, DumpDynamic); 2553 return; 2554 } 2555 2556 outs() << "\nDYNAMIC SYMBOL TABLE:\n"; 2557 if (!O.isELF()) { 2558 reportWarning( 2559 "this operation is not currently supported for this file format", 2560 FileName); 2561 return; 2562 } 2563 2564 const ELFObjectFileBase *ELF = cast<const ELFObjectFileBase>(&O); 2565 auto Symbols = ELF->getDynamicSymbolIterators(); 2566 Expected<std::vector<VersionEntry>> SymbolVersionsOrErr = 2567 ELF->readDynsymVersions(); 2568 if (!SymbolVersionsOrErr) { 2569 reportWarning(toString(SymbolVersionsOrErr.takeError()), FileName); 2570 SymbolVersionsOrErr = std::vector<VersionEntry>(); 2571 (void)!SymbolVersionsOrErr; 2572 } 2573 for (auto &Sym : Symbols) 2574 printSymbol(Sym, *SymbolVersionsOrErr, FileName, ArchiveName, 2575 ArchitectureName, DumpDynamic); 2576 } 2577 2578 void Dumper::printSymbol(const SymbolRef &Symbol, 2579 ArrayRef<VersionEntry> SymbolVersions, 2580 StringRef FileName, StringRef ArchiveName, 2581 StringRef ArchitectureName, bool DumpDynamic) { 2582 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(&O); 2583 Expected<uint64_t> AddrOrErr = Symbol.getAddress(); 2584 if (!AddrOrErr) { 2585 reportUniqueWarning(AddrOrErr.takeError()); 2586 return; 2587 } 2588 uint64_t Address = *AddrOrErr; 2589 section_iterator SecI = unwrapOrError(Symbol.getSection(), FileName); 2590 if (SecI != O.section_end() && shouldAdjustVA(*SecI)) 2591 Address += AdjustVMA; 2592 if ((Address < StartAddress) || (Address > StopAddress)) 2593 return; 2594 SymbolRef::Type Type = 2595 unwrapOrError(Symbol.getType(), FileName, ArchiveName, ArchitectureName); 2596 uint32_t Flags = 2597 unwrapOrError(Symbol.getFlags(), FileName, ArchiveName, ArchitectureName); 2598 2599 // Don't ask a Mach-O STAB symbol for its section unless you know that 2600 // STAB symbol's section field refers to a valid section index. Otherwise 2601 // the symbol may error trying to load a section that does not exist. 2602 bool IsSTAB = false; 2603 if (MachO) { 2604 DataRefImpl SymDRI = Symbol.getRawDataRefImpl(); 2605 uint8_t NType = 2606 (MachO->is64Bit() ? MachO->getSymbol64TableEntry(SymDRI).n_type 2607 : MachO->getSymbolTableEntry(SymDRI).n_type); 2608 if (NType & MachO::N_STAB) 2609 IsSTAB = true; 2610 } 2611 section_iterator Section = IsSTAB 2612 ? O.section_end() 2613 : unwrapOrError(Symbol.getSection(), FileName, 2614 ArchiveName, ArchitectureName); 2615 2616 StringRef Name; 2617 if (Type == SymbolRef::ST_Debug && Section != O.section_end()) { 2618 if (Expected<StringRef> NameOrErr = Section->getName()) 2619 Name = *NameOrErr; 2620 else 2621 consumeError(NameOrErr.takeError()); 2622 2623 } else { 2624 Name = unwrapOrError(Symbol.getName(), FileName, ArchiveName, 2625 ArchitectureName); 2626 } 2627 2628 bool Global = Flags & SymbolRef::SF_Global; 2629 bool Weak = Flags & SymbolRef::SF_Weak; 2630 bool Absolute = Flags & SymbolRef::SF_Absolute; 2631 bool Common = Flags & SymbolRef::SF_Common; 2632 bool Hidden = Flags & SymbolRef::SF_Hidden; 2633 2634 char GlobLoc = ' '; 2635 if ((Section != O.section_end() || Absolute) && !Weak) 2636 GlobLoc = Global ? 'g' : 'l'; 2637 char IFunc = ' '; 2638 if (O.isELF()) { 2639 if (ELFSymbolRef(Symbol).getELFType() == ELF::STT_GNU_IFUNC) 2640 IFunc = 'i'; 2641 if (ELFSymbolRef(Symbol).getBinding() == ELF::STB_GNU_UNIQUE) 2642 GlobLoc = 'u'; 2643 } 2644 2645 char Debug = ' '; 2646 if (DumpDynamic) 2647 Debug = 'D'; 2648 else if (Type == SymbolRef::ST_Debug || Type == SymbolRef::ST_File) 2649 Debug = 'd'; 2650 2651 char FileFunc = ' '; 2652 if (Type == SymbolRef::ST_File) 2653 FileFunc = 'f'; 2654 else if (Type == SymbolRef::ST_Function) 2655 FileFunc = 'F'; 2656 else if (Type == SymbolRef::ST_Data) 2657 FileFunc = 'O'; 2658 2659 const char *Fmt = O.getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 2660 2661 outs() << format(Fmt, Address) << " " 2662 << GlobLoc // Local -> 'l', Global -> 'g', Neither -> ' ' 2663 << (Weak ? 'w' : ' ') // Weak? 2664 << ' ' // Constructor. Not supported yet. 2665 << ' ' // Warning. Not supported yet. 2666 << IFunc // Indirect reference to another symbol. 2667 << Debug // Debugging (d) or dynamic (D) symbol. 2668 << FileFunc // Name of function (F), file (f) or object (O). 2669 << ' '; 2670 if (Absolute) { 2671 outs() << "*ABS*"; 2672 } else if (Common) { 2673 outs() << "*COM*"; 2674 } else if (Section == O.section_end()) { 2675 if (O.isXCOFF()) { 2676 XCOFFSymbolRef XCOFFSym = cast<const XCOFFObjectFile>(O).toSymbolRef( 2677 Symbol.getRawDataRefImpl()); 2678 if (XCOFF::N_DEBUG == XCOFFSym.getSectionNumber()) 2679 outs() << "*DEBUG*"; 2680 else 2681 outs() << "*UND*"; 2682 } else 2683 outs() << "*UND*"; 2684 } else { 2685 StringRef SegmentName = getSegmentName(MachO, *Section); 2686 if (!SegmentName.empty()) 2687 outs() << SegmentName << ","; 2688 StringRef SectionName = unwrapOrError(Section->getName(), FileName); 2689 outs() << SectionName; 2690 if (O.isXCOFF()) { 2691 std::optional<SymbolRef> SymRef = 2692 getXCOFFSymbolContainingSymbolRef(cast<XCOFFObjectFile>(O), Symbol); 2693 if (SymRef) { 2694 2695 Expected<StringRef> NameOrErr = SymRef->getName(); 2696 2697 if (NameOrErr) { 2698 outs() << " (csect:"; 2699 std::string SymName = 2700 Demangle ? demangle(*NameOrErr) : NameOrErr->str(); 2701 2702 if (SymbolDescription) 2703 SymName = getXCOFFSymbolDescription(createSymbolInfo(O, *SymRef), 2704 SymName); 2705 2706 outs() << ' ' << SymName; 2707 outs() << ") "; 2708 } else 2709 reportWarning(toString(NameOrErr.takeError()), FileName); 2710 } 2711 } 2712 } 2713 2714 if (Common) 2715 outs() << '\t' << format(Fmt, static_cast<uint64_t>(Symbol.getAlignment())); 2716 else if (O.isXCOFF()) 2717 outs() << '\t' 2718 << format(Fmt, cast<XCOFFObjectFile>(O).getSymbolSize( 2719 Symbol.getRawDataRefImpl())); 2720 else if (O.isELF()) 2721 outs() << '\t' << format(Fmt, ELFSymbolRef(Symbol).getSize()); 2722 2723 if (O.isELF()) { 2724 if (!SymbolVersions.empty()) { 2725 const VersionEntry &Ver = 2726 SymbolVersions[Symbol.getRawDataRefImpl().d.b - 1]; 2727 std::string Str; 2728 if (!Ver.Name.empty()) 2729 Str = Ver.IsVerDef ? ' ' + Ver.Name : '(' + Ver.Name + ')'; 2730 outs() << ' ' << left_justify(Str, 12); 2731 } 2732 2733 uint8_t Other = ELFSymbolRef(Symbol).getOther(); 2734 switch (Other) { 2735 case ELF::STV_DEFAULT: 2736 break; 2737 case ELF::STV_INTERNAL: 2738 outs() << " .internal"; 2739 break; 2740 case ELF::STV_HIDDEN: 2741 outs() << " .hidden"; 2742 break; 2743 case ELF::STV_PROTECTED: 2744 outs() << " .protected"; 2745 break; 2746 default: 2747 outs() << format(" 0x%02x", Other); 2748 break; 2749 } 2750 } else if (Hidden) { 2751 outs() << " .hidden"; 2752 } 2753 2754 std::string SymName = Demangle ? demangle(Name) : Name.str(); 2755 if (O.isXCOFF() && SymbolDescription) 2756 SymName = getXCOFFSymbolDescription(createSymbolInfo(O, Symbol), SymName); 2757 2758 outs() << ' ' << SymName << '\n'; 2759 } 2760 2761 static void printUnwindInfo(const ObjectFile *O) { 2762 outs() << "Unwind info:\n\n"; 2763 2764 if (const COFFObjectFile *Coff = dyn_cast<COFFObjectFile>(O)) 2765 printCOFFUnwindInfo(Coff); 2766 else if (const MachOObjectFile *MachO = dyn_cast<MachOObjectFile>(O)) 2767 printMachOUnwindInfo(MachO); 2768 else 2769 // TODO: Extract DWARF dump tool to objdump. 2770 WithColor::error(errs(), ToolName) 2771 << "This operation is only currently supported " 2772 "for COFF and MachO object files.\n"; 2773 } 2774 2775 /// Dump the raw contents of the __clangast section so the output can be piped 2776 /// into llvm-bcanalyzer. 2777 static void printRawClangAST(const ObjectFile *Obj) { 2778 if (outs().is_displayed()) { 2779 WithColor::error(errs(), ToolName) 2780 << "The -raw-clang-ast option will dump the raw binary contents of " 2781 "the clang ast section.\n" 2782 "Please redirect the output to a file or another program such as " 2783 "llvm-bcanalyzer.\n"; 2784 return; 2785 } 2786 2787 StringRef ClangASTSectionName("__clangast"); 2788 if (Obj->isCOFF()) { 2789 ClangASTSectionName = "clangast"; 2790 } 2791 2792 std::optional<object::SectionRef> ClangASTSection; 2793 for (auto Sec : ToolSectionFilter(*Obj)) { 2794 StringRef Name; 2795 if (Expected<StringRef> NameOrErr = Sec.getName()) 2796 Name = *NameOrErr; 2797 else 2798 consumeError(NameOrErr.takeError()); 2799 2800 if (Name == ClangASTSectionName) { 2801 ClangASTSection = Sec; 2802 break; 2803 } 2804 } 2805 if (!ClangASTSection) 2806 return; 2807 2808 StringRef ClangASTContents = 2809 unwrapOrError(ClangASTSection->getContents(), Obj->getFileName()); 2810 outs().write(ClangASTContents.data(), ClangASTContents.size()); 2811 } 2812 2813 static void printFaultMaps(const ObjectFile *Obj) { 2814 StringRef FaultMapSectionName; 2815 2816 if (Obj->isELF()) { 2817 FaultMapSectionName = ".llvm_faultmaps"; 2818 } else if (Obj->isMachO()) { 2819 FaultMapSectionName = "__llvm_faultmaps"; 2820 } else { 2821 WithColor::error(errs(), ToolName) 2822 << "This operation is only currently supported " 2823 "for ELF and Mach-O executable files.\n"; 2824 return; 2825 } 2826 2827 std::optional<object::SectionRef> FaultMapSection; 2828 2829 for (auto Sec : ToolSectionFilter(*Obj)) { 2830 StringRef Name; 2831 if (Expected<StringRef> NameOrErr = Sec.getName()) 2832 Name = *NameOrErr; 2833 else 2834 consumeError(NameOrErr.takeError()); 2835 2836 if (Name == FaultMapSectionName) { 2837 FaultMapSection = Sec; 2838 break; 2839 } 2840 } 2841 2842 outs() << "FaultMap table:\n"; 2843 2844 if (!FaultMapSection) { 2845 outs() << "<not found>\n"; 2846 return; 2847 } 2848 2849 StringRef FaultMapContents = 2850 unwrapOrError(FaultMapSection->getContents(), Obj->getFileName()); 2851 FaultMapParser FMP(FaultMapContents.bytes_begin(), 2852 FaultMapContents.bytes_end()); 2853 2854 outs() << FMP; 2855 } 2856 2857 void Dumper::printPrivateHeaders() { 2858 reportError(O.getFileName(), "Invalid/Unsupported object file format"); 2859 } 2860 2861 static void printFileHeaders(const ObjectFile *O) { 2862 if (!O->isELF() && !O->isCOFF()) 2863 reportError(O->getFileName(), "Invalid/Unsupported object file format"); 2864 2865 Triple::ArchType AT = O->getArch(); 2866 outs() << "architecture: " << Triple::getArchTypeName(AT) << "\n"; 2867 uint64_t Address = unwrapOrError(O->getStartAddress(), O->getFileName()); 2868 2869 StringRef Fmt = O->getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 2870 outs() << "start address: " 2871 << "0x" << format(Fmt.data(), Address) << "\n"; 2872 } 2873 2874 static void printArchiveChild(StringRef Filename, const Archive::Child &C) { 2875 Expected<sys::fs::perms> ModeOrErr = C.getAccessMode(); 2876 if (!ModeOrErr) { 2877 WithColor::error(errs(), ToolName) << "ill-formed archive entry.\n"; 2878 consumeError(ModeOrErr.takeError()); 2879 return; 2880 } 2881 sys::fs::perms Mode = ModeOrErr.get(); 2882 outs() << ((Mode & sys::fs::owner_read) ? "r" : "-"); 2883 outs() << ((Mode & sys::fs::owner_write) ? "w" : "-"); 2884 outs() << ((Mode & sys::fs::owner_exe) ? "x" : "-"); 2885 outs() << ((Mode & sys::fs::group_read) ? "r" : "-"); 2886 outs() << ((Mode & sys::fs::group_write) ? "w" : "-"); 2887 outs() << ((Mode & sys::fs::group_exe) ? "x" : "-"); 2888 outs() << ((Mode & sys::fs::others_read) ? "r" : "-"); 2889 outs() << ((Mode & sys::fs::others_write) ? "w" : "-"); 2890 outs() << ((Mode & sys::fs::others_exe) ? "x" : "-"); 2891 2892 outs() << " "; 2893 2894 outs() << format("%d/%d %6" PRId64 " ", unwrapOrError(C.getUID(), Filename), 2895 unwrapOrError(C.getGID(), Filename), 2896 unwrapOrError(C.getRawSize(), Filename)); 2897 2898 StringRef RawLastModified = C.getRawLastModified(); 2899 unsigned Seconds; 2900 if (RawLastModified.getAsInteger(10, Seconds)) 2901 outs() << "(date: \"" << RawLastModified 2902 << "\" contains non-decimal chars) "; 2903 else { 2904 // Since ctime(3) returns a 26 character string of the form: 2905 // "Sun Sep 16 01:03:52 1973\n\0" 2906 // just print 24 characters. 2907 time_t t = Seconds; 2908 outs() << format("%.24s ", ctime(&t)); 2909 } 2910 2911 StringRef Name = ""; 2912 Expected<StringRef> NameOrErr = C.getName(); 2913 if (!NameOrErr) { 2914 consumeError(NameOrErr.takeError()); 2915 Name = unwrapOrError(C.getRawName(), Filename); 2916 } else { 2917 Name = NameOrErr.get(); 2918 } 2919 outs() << Name << "\n"; 2920 } 2921 2922 // For ELF only now. 2923 static bool shouldWarnForInvalidStartStopAddress(ObjectFile *Obj) { 2924 if (const auto *Elf = dyn_cast<ELFObjectFileBase>(Obj)) { 2925 if (Elf->getEType() != ELF::ET_REL) 2926 return true; 2927 } 2928 return false; 2929 } 2930 2931 static void checkForInvalidStartStopAddress(ObjectFile *Obj, 2932 uint64_t Start, uint64_t Stop) { 2933 if (!shouldWarnForInvalidStartStopAddress(Obj)) 2934 return; 2935 2936 for (const SectionRef &Section : Obj->sections()) 2937 if (ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC) { 2938 uint64_t BaseAddr = Section.getAddress(); 2939 uint64_t Size = Section.getSize(); 2940 if ((Start < BaseAddr + Size) && Stop > BaseAddr) 2941 return; 2942 } 2943 2944 if (!HasStartAddressFlag) 2945 reportWarning("no section has address less than 0x" + 2946 Twine::utohexstr(Stop) + " specified by --stop-address", 2947 Obj->getFileName()); 2948 else if (!HasStopAddressFlag) 2949 reportWarning("no section has address greater than or equal to 0x" + 2950 Twine::utohexstr(Start) + " specified by --start-address", 2951 Obj->getFileName()); 2952 else 2953 reportWarning("no section overlaps the range [0x" + 2954 Twine::utohexstr(Start) + ",0x" + Twine::utohexstr(Stop) + 2955 ") specified by --start-address/--stop-address", 2956 Obj->getFileName()); 2957 } 2958 2959 static void dumpObject(ObjectFile *O, const Archive *A = nullptr, 2960 const Archive::Child *C = nullptr) { 2961 Expected<std::unique_ptr<Dumper>> DumperOrErr = createDumper(*O); 2962 if (!DumperOrErr) { 2963 reportError(DumperOrErr.takeError(), O->getFileName(), 2964 A ? A->getFileName() : ""); 2965 return; 2966 } 2967 Dumper &D = **DumperOrErr; 2968 2969 // Avoid other output when using a raw option. 2970 if (!RawClangAST) { 2971 outs() << '\n'; 2972 if (A) 2973 outs() << A->getFileName() << "(" << O->getFileName() << ")"; 2974 else 2975 outs() << O->getFileName(); 2976 outs() << ":\tfile format " << O->getFileFormatName().lower() << "\n"; 2977 } 2978 2979 if (HasStartAddressFlag || HasStopAddressFlag) 2980 checkForInvalidStartStopAddress(O, StartAddress, StopAddress); 2981 2982 // TODO: Change print* free functions to Dumper member functions to utilitize 2983 // stateful functions like reportUniqueWarning. 2984 2985 // Note: the order here matches GNU objdump for compatability. 2986 StringRef ArchiveName = A ? A->getFileName() : ""; 2987 if (ArchiveHeaders && !MachOOpt && C) 2988 printArchiveChild(ArchiveName, *C); 2989 if (FileHeaders) 2990 printFileHeaders(O); 2991 if (PrivateHeaders || FirstPrivateHeader) 2992 D.printPrivateHeaders(); 2993 if (SectionHeaders) 2994 printSectionHeaders(*O); 2995 if (SymbolTable) 2996 D.printSymbolTable(ArchiveName); 2997 if (DynamicSymbolTable) 2998 D.printSymbolTable(ArchiveName, /*ArchitectureName=*/"", 2999 /*DumpDynamic=*/true); 3000 if (DwarfDumpType != DIDT_Null) { 3001 std::unique_ptr<DIContext> DICtx = DWARFContext::create(*O); 3002 // Dump the complete DWARF structure. 3003 DIDumpOptions DumpOpts; 3004 DumpOpts.DumpType = DwarfDumpType; 3005 DICtx->dump(outs(), DumpOpts); 3006 } 3007 if (Relocations && !Disassemble) 3008 D.printRelocations(); 3009 if (DynamicRelocations) 3010 D.printDynamicRelocations(); 3011 if (SectionContents) 3012 printSectionContents(O); 3013 if (Disassemble) 3014 disassembleObject(O, Relocations); 3015 if (UnwindInfo) 3016 printUnwindInfo(O); 3017 3018 // Mach-O specific options: 3019 if (ExportsTrie) 3020 printExportsTrie(O); 3021 if (Rebase) 3022 printRebaseTable(O); 3023 if (Bind) 3024 printBindTable(O); 3025 if (LazyBind) 3026 printLazyBindTable(O); 3027 if (WeakBind) 3028 printWeakBindTable(O); 3029 3030 // Other special sections: 3031 if (RawClangAST) 3032 printRawClangAST(O); 3033 if (FaultMapSection) 3034 printFaultMaps(O); 3035 if (Offloading) 3036 dumpOffloadBinary(*O); 3037 } 3038 3039 static void dumpObject(const COFFImportFile *I, const Archive *A, 3040 const Archive::Child *C = nullptr) { 3041 StringRef ArchiveName = A ? A->getFileName() : ""; 3042 3043 // Avoid other output when using a raw option. 3044 if (!RawClangAST) 3045 outs() << '\n' 3046 << ArchiveName << "(" << I->getFileName() << ")" 3047 << ":\tfile format COFF-import-file" 3048 << "\n\n"; 3049 3050 if (ArchiveHeaders && !MachOOpt && C) 3051 printArchiveChild(ArchiveName, *C); 3052 if (SymbolTable) 3053 printCOFFSymbolTable(*I); 3054 } 3055 3056 /// Dump each object file in \a a; 3057 static void dumpArchive(const Archive *A) { 3058 Error Err = Error::success(); 3059 unsigned I = -1; 3060 for (auto &C : A->children(Err)) { 3061 ++I; 3062 Expected<std::unique_ptr<Binary>> ChildOrErr = C.getAsBinary(); 3063 if (!ChildOrErr) { 3064 if (auto E = isNotObjectErrorInvalidFileType(ChildOrErr.takeError())) 3065 reportError(std::move(E), getFileNameForError(C, I), A->getFileName()); 3066 continue; 3067 } 3068 if (ObjectFile *O = dyn_cast<ObjectFile>(&*ChildOrErr.get())) 3069 dumpObject(O, A, &C); 3070 else if (COFFImportFile *I = dyn_cast<COFFImportFile>(&*ChildOrErr.get())) 3071 dumpObject(I, A, &C); 3072 else 3073 reportError(errorCodeToError(object_error::invalid_file_type), 3074 A->getFileName()); 3075 } 3076 if (Err) 3077 reportError(std::move(Err), A->getFileName()); 3078 } 3079 3080 /// Open file and figure out how to dump it. 3081 static void dumpInput(StringRef file) { 3082 // If we are using the Mach-O specific object file parser, then let it parse 3083 // the file and process the command line options. So the -arch flags can 3084 // be used to select specific slices, etc. 3085 if (MachOOpt) { 3086 parseInputMachO(file); 3087 return; 3088 } 3089 3090 // Attempt to open the binary. 3091 OwningBinary<Binary> OBinary = unwrapOrError(createBinary(file), file); 3092 Binary &Binary = *OBinary.getBinary(); 3093 3094 if (Archive *A = dyn_cast<Archive>(&Binary)) 3095 dumpArchive(A); 3096 else if (ObjectFile *O = dyn_cast<ObjectFile>(&Binary)) 3097 dumpObject(O); 3098 else if (MachOUniversalBinary *UB = dyn_cast<MachOUniversalBinary>(&Binary)) 3099 parseInputMachO(UB); 3100 else if (OffloadBinary *OB = dyn_cast<OffloadBinary>(&Binary)) 3101 dumpOffloadSections(*OB); 3102 else 3103 reportError(errorCodeToError(object_error::invalid_file_type), file); 3104 } 3105 3106 template <typename T> 3107 static void parseIntArg(const llvm::opt::InputArgList &InputArgs, int ID, 3108 T &Value) { 3109 if (const opt::Arg *A = InputArgs.getLastArg(ID)) { 3110 StringRef V(A->getValue()); 3111 if (!llvm::to_integer(V, Value, 0)) { 3112 reportCmdLineError(A->getSpelling() + 3113 ": expected a non-negative integer, but got '" + V + 3114 "'"); 3115 } 3116 } 3117 } 3118 3119 static object::BuildID parseBuildIDArg(const opt::Arg *A) { 3120 StringRef V(A->getValue()); 3121 object::BuildID BID = parseBuildID(V); 3122 if (BID.empty()) 3123 reportCmdLineError(A->getSpelling() + ": expected a build ID, but got '" + 3124 V + "'"); 3125 return BID; 3126 } 3127 3128 void objdump::invalidArgValue(const opt::Arg *A) { 3129 reportCmdLineError("'" + StringRef(A->getValue()) + 3130 "' is not a valid value for '" + A->getSpelling() + "'"); 3131 } 3132 3133 static std::vector<std::string> 3134 commaSeparatedValues(const llvm::opt::InputArgList &InputArgs, int ID) { 3135 std::vector<std::string> Values; 3136 for (StringRef Value : InputArgs.getAllArgValues(ID)) { 3137 llvm::SmallVector<StringRef, 2> SplitValues; 3138 llvm::SplitString(Value, SplitValues, ","); 3139 for (StringRef SplitValue : SplitValues) 3140 Values.push_back(SplitValue.str()); 3141 } 3142 return Values; 3143 } 3144 3145 static void parseOtoolOptions(const llvm::opt::InputArgList &InputArgs) { 3146 MachOOpt = true; 3147 FullLeadingAddr = true; 3148 PrintImmHex = true; 3149 3150 ArchName = InputArgs.getLastArgValue(OTOOL_arch).str(); 3151 LinkOptHints = InputArgs.hasArg(OTOOL_C); 3152 if (InputArgs.hasArg(OTOOL_d)) 3153 FilterSections.push_back("__DATA,__data"); 3154 DylibId = InputArgs.hasArg(OTOOL_D); 3155 UniversalHeaders = InputArgs.hasArg(OTOOL_f); 3156 DataInCode = InputArgs.hasArg(OTOOL_G); 3157 FirstPrivateHeader = InputArgs.hasArg(OTOOL_h); 3158 IndirectSymbols = InputArgs.hasArg(OTOOL_I); 3159 ShowRawInsn = InputArgs.hasArg(OTOOL_j); 3160 PrivateHeaders = InputArgs.hasArg(OTOOL_l); 3161 DylibsUsed = InputArgs.hasArg(OTOOL_L); 3162 MCPU = InputArgs.getLastArgValue(OTOOL_mcpu_EQ).str(); 3163 ObjcMetaData = InputArgs.hasArg(OTOOL_o); 3164 DisSymName = InputArgs.getLastArgValue(OTOOL_p).str(); 3165 InfoPlist = InputArgs.hasArg(OTOOL_P); 3166 Relocations = InputArgs.hasArg(OTOOL_r); 3167 if (const Arg *A = InputArgs.getLastArg(OTOOL_s)) { 3168 auto Filter = (A->getValue(0) + StringRef(",") + A->getValue(1)).str(); 3169 FilterSections.push_back(Filter); 3170 } 3171 if (InputArgs.hasArg(OTOOL_t)) 3172 FilterSections.push_back("__TEXT,__text"); 3173 Verbose = InputArgs.hasArg(OTOOL_v) || InputArgs.hasArg(OTOOL_V) || 3174 InputArgs.hasArg(OTOOL_o); 3175 SymbolicOperands = InputArgs.hasArg(OTOOL_V); 3176 if (InputArgs.hasArg(OTOOL_x)) 3177 FilterSections.push_back(",__text"); 3178 LeadingAddr = LeadingHeaders = !InputArgs.hasArg(OTOOL_X); 3179 3180 ChainedFixups = InputArgs.hasArg(OTOOL_chained_fixups); 3181 DyldInfo = InputArgs.hasArg(OTOOL_dyld_info); 3182 3183 InputFilenames = InputArgs.getAllArgValues(OTOOL_INPUT); 3184 if (InputFilenames.empty()) 3185 reportCmdLineError("no input file"); 3186 3187 for (const Arg *A : InputArgs) { 3188 const Option &O = A->getOption(); 3189 if (O.getGroup().isValid() && O.getGroup().getID() == OTOOL_grp_obsolete) { 3190 reportCmdLineWarning(O.getPrefixedName() + 3191 " is obsolete and not implemented"); 3192 } 3193 } 3194 } 3195 3196 static void parseObjdumpOptions(const llvm::opt::InputArgList &InputArgs) { 3197 parseIntArg(InputArgs, OBJDUMP_adjust_vma_EQ, AdjustVMA); 3198 AllHeaders = InputArgs.hasArg(OBJDUMP_all_headers); 3199 ArchName = InputArgs.getLastArgValue(OBJDUMP_arch_name_EQ).str(); 3200 ArchiveHeaders = InputArgs.hasArg(OBJDUMP_archive_headers); 3201 Demangle = InputArgs.hasArg(OBJDUMP_demangle); 3202 Disassemble = InputArgs.hasArg(OBJDUMP_disassemble); 3203 DisassembleAll = InputArgs.hasArg(OBJDUMP_disassemble_all); 3204 SymbolDescription = InputArgs.hasArg(OBJDUMP_symbol_description); 3205 TracebackTable = InputArgs.hasArg(OBJDUMP_traceback_table); 3206 DisassembleSymbols = 3207 commaSeparatedValues(InputArgs, OBJDUMP_disassemble_symbols_EQ); 3208 DisassembleZeroes = InputArgs.hasArg(OBJDUMP_disassemble_zeroes); 3209 if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_dwarf_EQ)) { 3210 DwarfDumpType = StringSwitch<DIDumpType>(A->getValue()) 3211 .Case("frames", DIDT_DebugFrame) 3212 .Default(DIDT_Null); 3213 if (DwarfDumpType == DIDT_Null) 3214 invalidArgValue(A); 3215 } 3216 DynamicRelocations = InputArgs.hasArg(OBJDUMP_dynamic_reloc); 3217 FaultMapSection = InputArgs.hasArg(OBJDUMP_fault_map_section); 3218 Offloading = InputArgs.hasArg(OBJDUMP_offloading); 3219 FileHeaders = InputArgs.hasArg(OBJDUMP_file_headers); 3220 SectionContents = InputArgs.hasArg(OBJDUMP_full_contents); 3221 PrintLines = InputArgs.hasArg(OBJDUMP_line_numbers); 3222 InputFilenames = InputArgs.getAllArgValues(OBJDUMP_INPUT); 3223 MachOOpt = InputArgs.hasArg(OBJDUMP_macho); 3224 MCPU = InputArgs.getLastArgValue(OBJDUMP_mcpu_EQ).str(); 3225 MAttrs = commaSeparatedValues(InputArgs, OBJDUMP_mattr_EQ); 3226 ShowRawInsn = !InputArgs.hasArg(OBJDUMP_no_show_raw_insn); 3227 LeadingAddr = !InputArgs.hasArg(OBJDUMP_no_leading_addr); 3228 RawClangAST = InputArgs.hasArg(OBJDUMP_raw_clang_ast); 3229 Relocations = InputArgs.hasArg(OBJDUMP_reloc); 3230 PrintImmHex = 3231 InputArgs.hasFlag(OBJDUMP_print_imm_hex, OBJDUMP_no_print_imm_hex, true); 3232 PrivateHeaders = InputArgs.hasArg(OBJDUMP_private_headers); 3233 FilterSections = InputArgs.getAllArgValues(OBJDUMP_section_EQ); 3234 SectionHeaders = InputArgs.hasArg(OBJDUMP_section_headers); 3235 ShowAllSymbols = InputArgs.hasArg(OBJDUMP_show_all_symbols); 3236 ShowLMA = InputArgs.hasArg(OBJDUMP_show_lma); 3237 PrintSource = InputArgs.hasArg(OBJDUMP_source); 3238 parseIntArg(InputArgs, OBJDUMP_start_address_EQ, StartAddress); 3239 HasStartAddressFlag = InputArgs.hasArg(OBJDUMP_start_address_EQ); 3240 parseIntArg(InputArgs, OBJDUMP_stop_address_EQ, StopAddress); 3241 HasStopAddressFlag = InputArgs.hasArg(OBJDUMP_stop_address_EQ); 3242 SymbolTable = InputArgs.hasArg(OBJDUMP_syms); 3243 SymbolizeOperands = InputArgs.hasArg(OBJDUMP_symbolize_operands); 3244 DynamicSymbolTable = InputArgs.hasArg(OBJDUMP_dynamic_syms); 3245 TripleName = InputArgs.getLastArgValue(OBJDUMP_triple_EQ).str(); 3246 UnwindInfo = InputArgs.hasArg(OBJDUMP_unwind_info); 3247 Wide = InputArgs.hasArg(OBJDUMP_wide); 3248 Prefix = InputArgs.getLastArgValue(OBJDUMP_prefix).str(); 3249 parseIntArg(InputArgs, OBJDUMP_prefix_strip, PrefixStrip); 3250 if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_debug_vars_EQ)) { 3251 DbgVariables = StringSwitch<DebugVarsFormat>(A->getValue()) 3252 .Case("ascii", DVASCII) 3253 .Case("unicode", DVUnicode) 3254 .Default(DVInvalid); 3255 if (DbgVariables == DVInvalid) 3256 invalidArgValue(A); 3257 } 3258 if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_disassembler_color_EQ)) { 3259 DisassemblyColor = StringSwitch<ColorOutput>(A->getValue()) 3260 .Case("on", ColorOutput::Enable) 3261 .Case("off", ColorOutput::Disable) 3262 .Case("terminal", ColorOutput::Auto) 3263 .Default(ColorOutput::Invalid); 3264 if (DisassemblyColor == ColorOutput::Invalid) 3265 invalidArgValue(A); 3266 } 3267 3268 parseIntArg(InputArgs, OBJDUMP_debug_vars_indent_EQ, DbgIndent); 3269 3270 parseMachOOptions(InputArgs); 3271 3272 // Parse -M (--disassembler-options) and deprecated 3273 // --x86-asm-syntax={att,intel}. 3274 // 3275 // Note, for x86, the asm dialect (AssemblerDialect) is initialized when the 3276 // MCAsmInfo is constructed. MCInstPrinter::applyTargetSpecificCLOption is 3277 // called too late. For now we have to use the internal cl::opt option. 3278 const char *AsmSyntax = nullptr; 3279 for (const auto *A : InputArgs.filtered(OBJDUMP_disassembler_options_EQ, 3280 OBJDUMP_x86_asm_syntax_att, 3281 OBJDUMP_x86_asm_syntax_intel)) { 3282 switch (A->getOption().getID()) { 3283 case OBJDUMP_x86_asm_syntax_att: 3284 AsmSyntax = "--x86-asm-syntax=att"; 3285 continue; 3286 case OBJDUMP_x86_asm_syntax_intel: 3287 AsmSyntax = "--x86-asm-syntax=intel"; 3288 continue; 3289 } 3290 3291 SmallVector<StringRef, 2> Values; 3292 llvm::SplitString(A->getValue(), Values, ","); 3293 for (StringRef V : Values) { 3294 if (V == "att") 3295 AsmSyntax = "--x86-asm-syntax=att"; 3296 else if (V == "intel") 3297 AsmSyntax = "--x86-asm-syntax=intel"; 3298 else 3299 DisassemblerOptions.push_back(V.str()); 3300 } 3301 } 3302 if (AsmSyntax) { 3303 const char *Argv[] = {"llvm-objdump", AsmSyntax}; 3304 llvm::cl::ParseCommandLineOptions(2, Argv); 3305 } 3306 3307 // Look up any provided build IDs, then append them to the input filenames. 3308 for (const opt::Arg *A : InputArgs.filtered(OBJDUMP_build_id)) { 3309 object::BuildID BuildID = parseBuildIDArg(A); 3310 std::optional<std::string> Path = BIDFetcher->fetch(BuildID); 3311 if (!Path) { 3312 reportCmdLineError(A->getSpelling() + ": could not find build ID '" + 3313 A->getValue() + "'"); 3314 } 3315 InputFilenames.push_back(std::move(*Path)); 3316 } 3317 3318 // objdump defaults to a.out if no filenames specified. 3319 if (InputFilenames.empty()) 3320 InputFilenames.push_back("a.out"); 3321 } 3322 3323 int llvm_objdump_main(int argc, char **argv, const llvm::ToolContext &) { 3324 using namespace llvm; 3325 InitLLVM X(argc, argv); 3326 3327 ToolName = argv[0]; 3328 std::unique_ptr<CommonOptTable> T; 3329 OptSpecifier Unknown, HelpFlag, HelpHiddenFlag, VersionFlag; 3330 3331 StringRef Stem = sys::path::stem(ToolName); 3332 auto Is = [=](StringRef Tool) { 3333 // We need to recognize the following filenames: 3334 // 3335 // llvm-objdump -> objdump 3336 // llvm-otool-10.exe -> otool 3337 // powerpc64-unknown-freebsd13-objdump -> objdump 3338 auto I = Stem.rfind_insensitive(Tool); 3339 return I != StringRef::npos && 3340 (I + Tool.size() == Stem.size() || !isAlnum(Stem[I + Tool.size()])); 3341 }; 3342 if (Is("otool")) { 3343 T = std::make_unique<OtoolOptTable>(); 3344 Unknown = OTOOL_UNKNOWN; 3345 HelpFlag = OTOOL_help; 3346 HelpHiddenFlag = OTOOL_help_hidden; 3347 VersionFlag = OTOOL_version; 3348 } else { 3349 T = std::make_unique<ObjdumpOptTable>(); 3350 Unknown = OBJDUMP_UNKNOWN; 3351 HelpFlag = OBJDUMP_help; 3352 HelpHiddenFlag = OBJDUMP_help_hidden; 3353 VersionFlag = OBJDUMP_version; 3354 } 3355 3356 BumpPtrAllocator A; 3357 StringSaver Saver(A); 3358 opt::InputArgList InputArgs = 3359 T->parseArgs(argc, argv, Unknown, Saver, 3360 [&](StringRef Msg) { reportCmdLineError(Msg); }); 3361 3362 if (InputArgs.size() == 0 || InputArgs.hasArg(HelpFlag)) { 3363 T->printHelp(ToolName); 3364 return 0; 3365 } 3366 if (InputArgs.hasArg(HelpHiddenFlag)) { 3367 T->printHelp(ToolName, /*ShowHidden=*/true); 3368 return 0; 3369 } 3370 3371 // Initialize targets and assembly printers/parsers. 3372 InitializeAllTargetInfos(); 3373 InitializeAllTargetMCs(); 3374 InitializeAllDisassemblers(); 3375 3376 if (InputArgs.hasArg(VersionFlag)) { 3377 cl::PrintVersionMessage(); 3378 if (!Is("otool")) { 3379 outs() << '\n'; 3380 TargetRegistry::printRegisteredTargetsForVersion(outs()); 3381 } 3382 return 0; 3383 } 3384 3385 // Initialize debuginfod. 3386 const bool ShouldUseDebuginfodByDefault = 3387 InputArgs.hasArg(OBJDUMP_build_id) || canUseDebuginfod(); 3388 std::vector<std::string> DebugFileDirectories = 3389 InputArgs.getAllArgValues(OBJDUMP_debug_file_directory); 3390 if (InputArgs.hasFlag(OBJDUMP_debuginfod, OBJDUMP_no_debuginfod, 3391 ShouldUseDebuginfodByDefault)) { 3392 HTTPClient::initialize(); 3393 BIDFetcher = 3394 std::make_unique<DebuginfodFetcher>(std::move(DebugFileDirectories)); 3395 } else { 3396 BIDFetcher = 3397 std::make_unique<BuildIDFetcher>(std::move(DebugFileDirectories)); 3398 } 3399 3400 if (Is("otool")) 3401 parseOtoolOptions(InputArgs); 3402 else 3403 parseObjdumpOptions(InputArgs); 3404 3405 if (StartAddress >= StopAddress) 3406 reportCmdLineError("start address should be less than stop address"); 3407 3408 // Removes trailing separators from prefix. 3409 while (!Prefix.empty() && sys::path::is_separator(Prefix.back())) 3410 Prefix.pop_back(); 3411 3412 if (AllHeaders) 3413 ArchiveHeaders = FileHeaders = PrivateHeaders = Relocations = 3414 SectionHeaders = SymbolTable = true; 3415 3416 if (DisassembleAll || PrintSource || PrintLines || TracebackTable || 3417 !DisassembleSymbols.empty()) 3418 Disassemble = true; 3419 3420 if (!ArchiveHeaders && !Disassemble && DwarfDumpType == DIDT_Null && 3421 !DynamicRelocations && !FileHeaders && !PrivateHeaders && !RawClangAST && 3422 !Relocations && !SectionHeaders && !SectionContents && !SymbolTable && 3423 !DynamicSymbolTable && !UnwindInfo && !FaultMapSection && !Offloading && 3424 !(MachOOpt && 3425 (Bind || DataInCode || ChainedFixups || DyldInfo || DylibId || 3426 DylibsUsed || ExportsTrie || FirstPrivateHeader || 3427 FunctionStartsType != FunctionStartsMode::None || IndirectSymbols || 3428 InfoPlist || LazyBind || LinkOptHints || ObjcMetaData || Rebase || 3429 Rpaths || UniversalHeaders || WeakBind || !FilterSections.empty()))) { 3430 T->printHelp(ToolName); 3431 return 2; 3432 } 3433 3434 DisasmSymbolSet.insert(DisassembleSymbols.begin(), DisassembleSymbols.end()); 3435 3436 llvm::for_each(InputFilenames, dumpInput); 3437 3438 warnOnNoMatchForSections(); 3439 3440 return EXIT_SUCCESS; 3441 } 3442