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