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