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/Demangle/Demangle.h" 40 #include "llvm/MC/MCAsmInfo.h" 41 #include "llvm/MC/MCContext.h" 42 #include "llvm/MC/MCDisassembler/MCDisassembler.h" 43 #include "llvm/MC/MCDisassembler/MCRelocationInfo.h" 44 #include "llvm/MC/MCInst.h" 45 #include "llvm/MC/MCInstPrinter.h" 46 #include "llvm/MC/MCInstrAnalysis.h" 47 #include "llvm/MC/MCInstrInfo.h" 48 #include "llvm/MC/MCObjectFileInfo.h" 49 #include "llvm/MC/MCRegisterInfo.h" 50 #include "llvm/MC/MCSubtargetInfo.h" 51 #include "llvm/MC/MCTargetOptions.h" 52 #include "llvm/MC/TargetRegistry.h" 53 #include "llvm/Object/Archive.h" 54 #include "llvm/Object/COFF.h" 55 #include "llvm/Object/COFFImportFile.h" 56 #include "llvm/Object/ELFObjectFile.h" 57 #include "llvm/Object/ELFTypes.h" 58 #include "llvm/Object/FaultMapParser.h" 59 #include "llvm/Object/MachO.h" 60 #include "llvm/Object/MachOUniversal.h" 61 #include "llvm/Object/ObjectFile.h" 62 #include "llvm/Object/OffloadBinary.h" 63 #include "llvm/Object/Wasm.h" 64 #include "llvm/Option/Arg.h" 65 #include "llvm/Option/ArgList.h" 66 #include "llvm/Option/Option.h" 67 #include "llvm/Support/Casting.h" 68 #include "llvm/Support/Debug.h" 69 #include "llvm/Support/Errc.h" 70 #include "llvm/Support/FileSystem.h" 71 #include "llvm/Support/Format.h" 72 #include "llvm/Support/FormatVariadic.h" 73 #include "llvm/Support/GraphWriter.h" 74 #include "llvm/Support/Host.h" 75 #include "llvm/Support/InitLLVM.h" 76 #include "llvm/Support/MemoryBuffer.h" 77 #include "llvm/Support/SourceMgr.h" 78 #include "llvm/Support/StringSaver.h" 79 #include "llvm/Support/TargetSelect.h" 80 #include "llvm/Support/WithColor.h" 81 #include "llvm/Support/raw_ostream.h" 82 #include <algorithm> 83 #include <cctype> 84 #include <cstring> 85 #include <system_error> 86 #include <unordered_map> 87 #include <utility> 88 89 using namespace llvm; 90 using namespace llvm::object; 91 using namespace llvm::objdump; 92 using namespace llvm::opt; 93 94 namespace { 95 96 class CommonOptTable : public opt::OptTable { 97 public: 98 CommonOptTable(ArrayRef<Info> OptionInfos, const char *Usage, 99 const char *Description) 100 : OptTable(OptionInfos), Usage(Usage), Description(Description) { 101 setGroupedShortOptions(true); 102 } 103 104 void printHelp(StringRef Argv0, bool ShowHidden = false) const { 105 Argv0 = sys::path::filename(Argv0); 106 opt::OptTable::printHelp(outs(), (Argv0 + Usage).str().c_str(), Description, 107 ShowHidden, ShowHidden); 108 // TODO Replace this with OptTable API once it adds extrahelp support. 109 outs() << "\nPass @FILE as argument to read options from FILE.\n"; 110 } 111 112 private: 113 const char *Usage; 114 const char *Description; 115 }; 116 117 // ObjdumpOptID is in ObjdumpOptID.h 118 119 #define PREFIX(NAME, VALUE) const char *const OBJDUMP_##NAME[] = VALUE; 120 #include "ObjdumpOpts.inc" 121 #undef PREFIX 122 123 static constexpr opt::OptTable::Info ObjdumpInfoTable[] = { 124 #define OBJDUMP_nullptr nullptr 125 #define OPTION(PREFIX, NAME, ID, KIND, GROUP, ALIAS, ALIASARGS, FLAGS, PARAM, \ 126 HELPTEXT, METAVAR, VALUES) \ 127 {OBJDUMP_##PREFIX, NAME, HELPTEXT, \ 128 METAVAR, OBJDUMP_##ID, opt::Option::KIND##Class, \ 129 PARAM, FLAGS, OBJDUMP_##GROUP, \ 130 OBJDUMP_##ALIAS, ALIASARGS, VALUES}, 131 #include "ObjdumpOpts.inc" 132 #undef OPTION 133 #undef OBJDUMP_nullptr 134 }; 135 136 class ObjdumpOptTable : public CommonOptTable { 137 public: 138 ObjdumpOptTable() 139 : CommonOptTable(ObjdumpInfoTable, " [options] <input object files>", 140 "llvm object file dumper") {} 141 }; 142 143 enum OtoolOptID { 144 OTOOL_INVALID = 0, // This is not an option ID. 145 #define OPTION(PREFIX, NAME, ID, KIND, GROUP, ALIAS, ALIASARGS, FLAGS, PARAM, \ 146 HELPTEXT, METAVAR, VALUES) \ 147 OTOOL_##ID, 148 #include "OtoolOpts.inc" 149 #undef OPTION 150 }; 151 152 #define PREFIX(NAME, VALUE) const char *const OTOOL_##NAME[] = VALUE; 153 #include "OtoolOpts.inc" 154 #undef PREFIX 155 156 static constexpr opt::OptTable::Info OtoolInfoTable[] = { 157 #define OTOOL_nullptr nullptr 158 #define OPTION(PREFIX, NAME, ID, KIND, GROUP, ALIAS, ALIASARGS, FLAGS, PARAM, \ 159 HELPTEXT, METAVAR, VALUES) \ 160 {OTOOL_##PREFIX, NAME, HELPTEXT, \ 161 METAVAR, OTOOL_##ID, opt::Option::KIND##Class, \ 162 PARAM, FLAGS, OTOOL_##GROUP, \ 163 OTOOL_##ALIAS, ALIASARGS, VALUES}, 164 #include "OtoolOpts.inc" 165 #undef OPTION 166 #undef OTOOL_nullptr 167 }; 168 169 class OtoolOptTable : public CommonOptTable { 170 public: 171 OtoolOptTable() 172 : CommonOptTable(OtoolInfoTable, " [option...] [file...]", 173 "Mach-O object file displaying tool") {} 174 }; 175 176 } // namespace 177 178 #define DEBUG_TYPE "objdump" 179 180 static uint64_t AdjustVMA; 181 static bool AllHeaders; 182 static std::string ArchName; 183 bool objdump::ArchiveHeaders; 184 bool objdump::Demangle; 185 bool objdump::Disassemble; 186 bool objdump::DisassembleAll; 187 bool objdump::SymbolDescription; 188 static std::vector<std::string> DisassembleSymbols; 189 static bool DisassembleZeroes; 190 static std::vector<std::string> DisassemblerOptions; 191 DIDumpType objdump::DwarfDumpType; 192 static bool DynamicRelocations; 193 static bool FaultMapSection; 194 static bool FileHeaders; 195 bool objdump::SectionContents; 196 static std::vector<std::string> InputFilenames; 197 bool objdump::PrintLines; 198 static bool MachOOpt; 199 std::string objdump::MCPU; 200 std::vector<std::string> objdump::MAttrs; 201 bool objdump::ShowRawInsn; 202 bool objdump::LeadingAddr; 203 static bool Offloading; 204 static bool RawClangAST; 205 bool objdump::Relocations; 206 bool objdump::PrintImmHex; 207 bool objdump::PrivateHeaders; 208 std::vector<std::string> objdump::FilterSections; 209 bool objdump::SectionHeaders; 210 static bool ShowLMA; 211 bool objdump::PrintSource; 212 213 static uint64_t StartAddress; 214 static bool HasStartAddressFlag; 215 static uint64_t StopAddress = UINT64_MAX; 216 static bool HasStopAddressFlag; 217 218 bool objdump::SymbolTable; 219 static bool SymbolizeOperands; 220 static bool DynamicSymbolTable; 221 std::string objdump::TripleName; 222 bool objdump::UnwindInfo; 223 static bool Wide; 224 std::string objdump::Prefix; 225 uint32_t objdump::PrefixStrip; 226 227 DebugVarsFormat objdump::DbgVariables = DVDisabled; 228 229 int objdump::DbgIndent = 52; 230 231 static StringSet<> DisasmSymbolSet; 232 StringSet<> objdump::FoundSectionSet; 233 static StringRef ToolName; 234 235 namespace { 236 struct FilterResult { 237 // True if the section should not be skipped. 238 bool Keep; 239 240 // True if the index counter should be incremented, even if the section should 241 // be skipped. For example, sections may be skipped if they are not included 242 // in the --section flag, but we still want those to count toward the section 243 // count. 244 bool IncrementIndex; 245 }; 246 } // namespace 247 248 static FilterResult checkSectionFilter(object::SectionRef S) { 249 if (FilterSections.empty()) 250 return {/*Keep=*/true, /*IncrementIndex=*/true}; 251 252 Expected<StringRef> SecNameOrErr = S.getName(); 253 if (!SecNameOrErr) { 254 consumeError(SecNameOrErr.takeError()); 255 return {/*Keep=*/false, /*IncrementIndex=*/false}; 256 } 257 StringRef SecName = *SecNameOrErr; 258 259 // StringSet does not allow empty key so avoid adding sections with 260 // no name (such as the section with index 0) here. 261 if (!SecName.empty()) 262 FoundSectionSet.insert(SecName); 263 264 // Only show the section if it's in the FilterSections list, but always 265 // increment so the indexing is stable. 266 return {/*Keep=*/is_contained(FilterSections, SecName), 267 /*IncrementIndex=*/true}; 268 } 269 270 SectionFilter objdump::ToolSectionFilter(object::ObjectFile const &O, 271 uint64_t *Idx) { 272 // Start at UINT64_MAX so that the first index returned after an increment is 273 // zero (after the unsigned wrap). 274 if (Idx) 275 *Idx = UINT64_MAX; 276 return SectionFilter( 277 [Idx](object::SectionRef S) { 278 FilterResult Result = checkSectionFilter(S); 279 if (Idx != nullptr && Result.IncrementIndex) 280 *Idx += 1; 281 return Result.Keep; 282 }, 283 O); 284 } 285 286 std::string objdump::getFileNameForError(const object::Archive::Child &C, 287 unsigned Index) { 288 Expected<StringRef> NameOrErr = C.getName(); 289 if (NameOrErr) 290 return std::string(NameOrErr.get()); 291 // If we have an error getting the name then we print the index of the archive 292 // member. Since we are already in an error state, we just ignore this error. 293 consumeError(NameOrErr.takeError()); 294 return "<file index: " + std::to_string(Index) + ">"; 295 } 296 297 void objdump::reportWarning(const Twine &Message, StringRef File) { 298 // Output order between errs() and outs() matters especially for archive 299 // files where the output is per member object. 300 outs().flush(); 301 WithColor::warning(errs(), ToolName) 302 << "'" << File << "': " << Message << "\n"; 303 } 304 305 [[noreturn]] void objdump::reportError(StringRef File, const Twine &Message) { 306 outs().flush(); 307 WithColor::error(errs(), ToolName) << "'" << File << "': " << Message << "\n"; 308 exit(1); 309 } 310 311 [[noreturn]] void objdump::reportError(Error E, StringRef FileName, 312 StringRef ArchiveName, 313 StringRef ArchitectureName) { 314 assert(E); 315 outs().flush(); 316 WithColor::error(errs(), ToolName); 317 if (ArchiveName != "") 318 errs() << ArchiveName << "(" << FileName << ")"; 319 else 320 errs() << "'" << FileName << "'"; 321 if (!ArchitectureName.empty()) 322 errs() << " (for architecture " << ArchitectureName << ")"; 323 errs() << ": "; 324 logAllUnhandledErrors(std::move(E), errs()); 325 exit(1); 326 } 327 328 static void reportCmdLineWarning(const Twine &Message) { 329 WithColor::warning(errs(), ToolName) << Message << "\n"; 330 } 331 332 [[noreturn]] static void reportCmdLineError(const Twine &Message) { 333 WithColor::error(errs(), ToolName) << Message << "\n"; 334 exit(1); 335 } 336 337 static void warnOnNoMatchForSections() { 338 SetVector<StringRef> MissingSections; 339 for (StringRef S : FilterSections) { 340 if (FoundSectionSet.count(S)) 341 return; 342 // User may specify a unnamed section. Don't warn for it. 343 if (!S.empty()) 344 MissingSections.insert(S); 345 } 346 347 // Warn only if no section in FilterSections is matched. 348 for (StringRef S : MissingSections) 349 reportCmdLineWarning("section '" + S + 350 "' mentioned in a -j/--section option, but not " 351 "found in any input file"); 352 } 353 354 static const Target *getTarget(const ObjectFile *Obj) { 355 // Figure out the target triple. 356 Triple TheTriple("unknown-unknown-unknown"); 357 if (TripleName.empty()) { 358 TheTriple = Obj->makeTriple(); 359 } else { 360 TheTriple.setTriple(Triple::normalize(TripleName)); 361 auto Arch = Obj->getArch(); 362 if (Arch == Triple::arm || Arch == Triple::armeb) 363 Obj->setARMSubArch(TheTriple); 364 } 365 366 // Get the target specific parser. 367 std::string Error; 368 const Target *TheTarget = TargetRegistry::lookupTarget(ArchName, TheTriple, 369 Error); 370 if (!TheTarget) 371 reportError(Obj->getFileName(), "can't find target: " + Error); 372 373 // Update the triple name and return the found target. 374 TripleName = TheTriple.getTriple(); 375 return TheTarget; 376 } 377 378 bool objdump::isRelocAddressLess(RelocationRef A, RelocationRef B) { 379 return A.getOffset() < B.getOffset(); 380 } 381 382 static Error getRelocationValueString(const RelocationRef &Rel, 383 SmallVectorImpl<char> &Result) { 384 const ObjectFile *Obj = Rel.getObject(); 385 if (auto *ELF = dyn_cast<ELFObjectFileBase>(Obj)) 386 return getELFRelocationValueString(ELF, Rel, Result); 387 if (auto *COFF = dyn_cast<COFFObjectFile>(Obj)) 388 return getCOFFRelocationValueString(COFF, Rel, Result); 389 if (auto *Wasm = dyn_cast<WasmObjectFile>(Obj)) 390 return getWasmRelocationValueString(Wasm, Rel, Result); 391 if (auto *MachO = dyn_cast<MachOObjectFile>(Obj)) 392 return getMachORelocationValueString(MachO, Rel, Result); 393 if (auto *XCOFF = dyn_cast<XCOFFObjectFile>(Obj)) 394 return getXCOFFRelocationValueString(*XCOFF, Rel, Result); 395 llvm_unreachable("unknown object file format"); 396 } 397 398 /// Indicates whether this relocation should hidden when listing 399 /// relocations, usually because it is the trailing part of a multipart 400 /// relocation that will be printed as part of the leading relocation. 401 static bool getHidden(RelocationRef RelRef) { 402 auto *MachO = dyn_cast<MachOObjectFile>(RelRef.getObject()); 403 if (!MachO) 404 return false; 405 406 unsigned Arch = MachO->getArch(); 407 DataRefImpl Rel = RelRef.getRawDataRefImpl(); 408 uint64_t Type = MachO->getRelocationType(Rel); 409 410 // On arches that use the generic relocations, GENERIC_RELOC_PAIR 411 // is always hidden. 412 if (Arch == Triple::x86 || Arch == Triple::arm || Arch == Triple::ppc) 413 return Type == MachO::GENERIC_RELOC_PAIR; 414 415 if (Arch == Triple::x86_64) { 416 // On x86_64, X86_64_RELOC_UNSIGNED is hidden only when it follows 417 // an X86_64_RELOC_SUBTRACTOR. 418 if (Type == MachO::X86_64_RELOC_UNSIGNED && Rel.d.a > 0) { 419 DataRefImpl RelPrev = Rel; 420 RelPrev.d.a--; 421 uint64_t PrevType = MachO->getRelocationType(RelPrev); 422 if (PrevType == MachO::X86_64_RELOC_SUBTRACTOR) 423 return true; 424 } 425 } 426 427 return false; 428 } 429 430 namespace { 431 432 /// Get the column at which we want to start printing the instruction 433 /// disassembly, taking into account anything which appears to the left of it. 434 unsigned getInstStartColumn(const MCSubtargetInfo &STI) { 435 return !ShowRawInsn ? 16 : STI.getTargetTriple().isX86() ? 40 : 24; 436 } 437 438 static bool isAArch64Elf(const ObjectFile &Obj) { 439 const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj); 440 return Elf && Elf->getEMachine() == ELF::EM_AARCH64; 441 } 442 443 static bool isArmElf(const ObjectFile &Obj) { 444 const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj); 445 return Elf && Elf->getEMachine() == ELF::EM_ARM; 446 } 447 448 static bool isCSKYElf(const ObjectFile &Obj) { 449 const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj); 450 return Elf && Elf->getEMachine() == ELF::EM_CSKY; 451 } 452 453 static bool hasMappingSymbols(const ObjectFile &Obj) { 454 return isArmElf(Obj) || isAArch64Elf(Obj) || isCSKYElf(Obj) ; 455 } 456 457 static void printRelocation(formatted_raw_ostream &OS, StringRef FileName, 458 const RelocationRef &Rel, uint64_t Address, 459 bool Is64Bits) { 460 StringRef Fmt = Is64Bits ? "\t\t%016" PRIx64 ": " : "\t\t\t%08" PRIx64 ": "; 461 SmallString<16> Name; 462 SmallString<32> Val; 463 Rel.getTypeName(Name); 464 if (Error E = getRelocationValueString(Rel, Val)) 465 reportError(std::move(E), FileName); 466 OS << format(Fmt.data(), Address) << Name << "\t" << Val; 467 } 468 469 static void AlignToInstStartColumn(size_t Start, const MCSubtargetInfo &STI, 470 raw_ostream &OS) { 471 // The output of printInst starts with a tab. Print some spaces so that 472 // the tab has 1 column and advances to the target tab stop. 473 unsigned TabStop = getInstStartColumn(STI); 474 unsigned Column = OS.tell() - Start; 475 OS.indent(Column < TabStop - 1 ? TabStop - 1 - Column : 7 - Column % 8); 476 } 477 478 class PrettyPrinter { 479 public: 480 virtual ~PrettyPrinter() = default; 481 virtual void 482 printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 483 object::SectionedAddress Address, formatted_raw_ostream &OS, 484 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 485 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 486 LiveVariablePrinter &LVP) { 487 if (SP && (PrintSource || PrintLines)) 488 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 489 LVP.printBetweenInsts(OS, false); 490 491 size_t Start = OS.tell(); 492 if (LeadingAddr) 493 OS << format("%8" PRIx64 ":", Address.Address); 494 if (ShowRawInsn) { 495 OS << ' '; 496 dumpBytes(Bytes, OS); 497 } 498 499 AlignToInstStartColumn(Start, STI, OS); 500 501 if (MI) { 502 // See MCInstPrinter::printInst. On targets where a PC relative immediate 503 // is relative to the next instruction and the length of a MCInst is 504 // difficult to measure (x86), this is the address of the next 505 // instruction. 506 uint64_t Addr = 507 Address.Address + (STI.getTargetTriple().isX86() ? Bytes.size() : 0); 508 IP.printInst(MI, Addr, "", STI, OS); 509 } else 510 OS << "\t<unknown>"; 511 } 512 }; 513 PrettyPrinter PrettyPrinterInst; 514 515 class HexagonPrettyPrinter : public PrettyPrinter { 516 public: 517 void printLead(ArrayRef<uint8_t> Bytes, uint64_t Address, 518 formatted_raw_ostream &OS) { 519 uint32_t opcode = 520 (Bytes[3] << 24) | (Bytes[2] << 16) | (Bytes[1] << 8) | Bytes[0]; 521 if (LeadingAddr) 522 OS << format("%8" PRIx64 ":", Address); 523 if (ShowRawInsn) { 524 OS << "\t"; 525 dumpBytes(Bytes.slice(0, 4), OS); 526 OS << format("\t%08" PRIx32, opcode); 527 } 528 } 529 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 530 object::SectionedAddress Address, formatted_raw_ostream &OS, 531 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 532 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 533 LiveVariablePrinter &LVP) override { 534 if (SP && (PrintSource || PrintLines)) 535 SP->printSourceLine(OS, Address, ObjectFilename, LVP, ""); 536 if (!MI) { 537 printLead(Bytes, Address.Address, OS); 538 OS << " <unknown>"; 539 return; 540 } 541 std::string Buffer; 542 { 543 raw_string_ostream TempStream(Buffer); 544 IP.printInst(MI, Address.Address, "", STI, TempStream); 545 } 546 StringRef Contents(Buffer); 547 // Split off bundle attributes 548 auto PacketBundle = Contents.rsplit('\n'); 549 // Split off first instruction from the rest 550 auto HeadTail = PacketBundle.first.split('\n'); 551 auto Preamble = " { "; 552 auto Separator = ""; 553 554 // Hexagon's packets require relocations to be inline rather than 555 // clustered at the end of the packet. 556 std::vector<RelocationRef>::const_iterator RelCur = Rels->begin(); 557 std::vector<RelocationRef>::const_iterator RelEnd = Rels->end(); 558 auto PrintReloc = [&]() -> void { 559 while ((RelCur != RelEnd) && (RelCur->getOffset() <= Address.Address)) { 560 if (RelCur->getOffset() == Address.Address) { 561 printRelocation(OS, ObjectFilename, *RelCur, Address.Address, false); 562 return; 563 } 564 ++RelCur; 565 } 566 }; 567 568 while (!HeadTail.first.empty()) { 569 OS << Separator; 570 Separator = "\n"; 571 if (SP && (PrintSource || PrintLines)) 572 SP->printSourceLine(OS, Address, ObjectFilename, LVP, ""); 573 printLead(Bytes, Address.Address, OS); 574 OS << Preamble; 575 Preamble = " "; 576 StringRef Inst; 577 auto Duplex = HeadTail.first.split('\v'); 578 if (!Duplex.second.empty()) { 579 OS << Duplex.first; 580 OS << "; "; 581 Inst = Duplex.second; 582 } 583 else 584 Inst = HeadTail.first; 585 OS << Inst; 586 HeadTail = HeadTail.second.split('\n'); 587 if (HeadTail.first.empty()) 588 OS << " } " << PacketBundle.second; 589 PrintReloc(); 590 Bytes = Bytes.slice(4); 591 Address.Address += 4; 592 } 593 } 594 }; 595 HexagonPrettyPrinter HexagonPrettyPrinterInst; 596 597 class AMDGCNPrettyPrinter : public PrettyPrinter { 598 public: 599 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 600 object::SectionedAddress Address, formatted_raw_ostream &OS, 601 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 602 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 603 LiveVariablePrinter &LVP) override { 604 if (SP && (PrintSource || PrintLines)) 605 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 606 607 if (MI) { 608 SmallString<40> InstStr; 609 raw_svector_ostream IS(InstStr); 610 611 IP.printInst(MI, Address.Address, "", STI, IS); 612 613 OS << left_justify(IS.str(), 60); 614 } else { 615 // an unrecognized encoding - this is probably data so represent it 616 // using the .long directive, or .byte directive if fewer than 4 bytes 617 // remaining 618 if (Bytes.size() >= 4) { 619 OS << format("\t.long 0x%08" PRIx32 " ", 620 support::endian::read32<support::little>(Bytes.data())); 621 OS.indent(42); 622 } else { 623 OS << format("\t.byte 0x%02" PRIx8, Bytes[0]); 624 for (unsigned int i = 1; i < Bytes.size(); i++) 625 OS << format(", 0x%02" PRIx8, Bytes[i]); 626 OS.indent(55 - (6 * Bytes.size())); 627 } 628 } 629 630 OS << format("// %012" PRIX64 ":", Address.Address); 631 if (Bytes.size() >= 4) { 632 // D should be casted to uint32_t here as it is passed by format to 633 // snprintf as vararg. 634 for (uint32_t D : makeArrayRef( 635 reinterpret_cast<const support::little32_t *>(Bytes.data()), 636 Bytes.size() / 4)) 637 OS << format(" %08" PRIX32, D); 638 } else { 639 for (unsigned char B : Bytes) 640 OS << format(" %02" PRIX8, B); 641 } 642 643 if (!Annot.empty()) 644 OS << " // " << Annot; 645 } 646 }; 647 AMDGCNPrettyPrinter AMDGCNPrettyPrinterInst; 648 649 class BPFPrettyPrinter : public PrettyPrinter { 650 public: 651 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 652 object::SectionedAddress Address, formatted_raw_ostream &OS, 653 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 654 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 655 LiveVariablePrinter &LVP) override { 656 if (SP && (PrintSource || PrintLines)) 657 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 658 if (LeadingAddr) 659 OS << format("%8" PRId64 ":", Address.Address / 8); 660 if (ShowRawInsn) { 661 OS << "\t"; 662 dumpBytes(Bytes, OS); 663 } 664 if (MI) 665 IP.printInst(MI, Address.Address, "", STI, OS); 666 else 667 OS << "\t<unknown>"; 668 } 669 }; 670 BPFPrettyPrinter BPFPrettyPrinterInst; 671 672 class ARMPrettyPrinter : public PrettyPrinter { 673 public: 674 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 675 object::SectionedAddress Address, formatted_raw_ostream &OS, 676 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 677 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 678 LiveVariablePrinter &LVP) override { 679 if (SP && (PrintSource || PrintLines)) 680 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 681 LVP.printBetweenInsts(OS, false); 682 683 size_t Start = OS.tell(); 684 if (LeadingAddr) 685 OS << format("%8" PRIx64 ":", Address.Address); 686 if (ShowRawInsn) { 687 size_t Pos = 0, End = Bytes.size(); 688 if (STI.checkFeatures("+thumb-mode")) { 689 for (; Pos + 2 <= End; Pos += 2) 690 OS << ' ' 691 << format_hex_no_prefix( 692 llvm::support::endian::read<uint16_t>( 693 Bytes.data() + Pos, llvm::support::little), 694 4); 695 } else { 696 for (; Pos + 4 <= End; Pos += 4) 697 OS << ' ' 698 << format_hex_no_prefix( 699 llvm::support::endian::read<uint32_t>( 700 Bytes.data() + Pos, llvm::support::little), 701 8); 702 } 703 if (Pos < End) { 704 OS << ' '; 705 dumpBytes(Bytes.slice(Pos), OS); 706 } 707 } 708 709 AlignToInstStartColumn(Start, STI, OS); 710 711 if (MI) { 712 IP.printInst(MI, Address.Address, "", STI, OS); 713 } else 714 OS << "\t<unknown>"; 715 } 716 }; 717 ARMPrettyPrinter ARMPrettyPrinterInst; 718 719 class AArch64PrettyPrinter : public PrettyPrinter { 720 public: 721 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 722 object::SectionedAddress Address, formatted_raw_ostream &OS, 723 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 724 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 725 LiveVariablePrinter &LVP) override { 726 if (SP && (PrintSource || PrintLines)) 727 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 728 LVP.printBetweenInsts(OS, false); 729 730 size_t Start = OS.tell(); 731 if (LeadingAddr) 732 OS << format("%8" PRIx64 ":", Address.Address); 733 if (ShowRawInsn) { 734 size_t Pos = 0, End = Bytes.size(); 735 for (; Pos + 4 <= End; Pos += 4) 736 OS << ' ' 737 << format_hex_no_prefix( 738 llvm::support::endian::read<uint32_t>(Bytes.data() + Pos, 739 llvm::support::little), 740 8); 741 if (Pos < End) { 742 OS << ' '; 743 dumpBytes(Bytes.slice(Pos), OS); 744 } 745 } 746 747 AlignToInstStartColumn(Start, STI, OS); 748 749 if (MI) { 750 IP.printInst(MI, Address.Address, "", STI, OS); 751 } else 752 OS << "\t<unknown>"; 753 } 754 }; 755 AArch64PrettyPrinter AArch64PrettyPrinterInst; 756 757 PrettyPrinter &selectPrettyPrinter(Triple const &Triple) { 758 switch(Triple.getArch()) { 759 default: 760 return PrettyPrinterInst; 761 case Triple::hexagon: 762 return HexagonPrettyPrinterInst; 763 case Triple::amdgcn: 764 return AMDGCNPrettyPrinterInst; 765 case Triple::bpfel: 766 case Triple::bpfeb: 767 return BPFPrettyPrinterInst; 768 case Triple::arm: 769 case Triple::armeb: 770 case Triple::thumb: 771 case Triple::thumbeb: 772 return ARMPrettyPrinterInst; 773 case Triple::aarch64: 774 case Triple::aarch64_be: 775 case Triple::aarch64_32: 776 return AArch64PrettyPrinterInst; 777 } 778 } 779 } 780 781 static uint8_t getElfSymbolType(const ObjectFile &Obj, const SymbolRef &Sym) { 782 assert(Obj.isELF()); 783 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj)) 784 return unwrapOrError(Elf32LEObj->getSymbol(Sym.getRawDataRefImpl()), 785 Obj.getFileName()) 786 ->getType(); 787 if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj)) 788 return unwrapOrError(Elf64LEObj->getSymbol(Sym.getRawDataRefImpl()), 789 Obj.getFileName()) 790 ->getType(); 791 if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj)) 792 return unwrapOrError(Elf32BEObj->getSymbol(Sym.getRawDataRefImpl()), 793 Obj.getFileName()) 794 ->getType(); 795 if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj)) 796 return unwrapOrError(Elf64BEObj->getSymbol(Sym.getRawDataRefImpl()), 797 Obj.getFileName()) 798 ->getType(); 799 llvm_unreachable("Unsupported binary format"); 800 } 801 802 template <class ELFT> 803 static void 804 addDynamicElfSymbols(const ELFObjectFile<ELFT> &Obj, 805 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 806 for (auto Symbol : Obj.getDynamicSymbolIterators()) { 807 uint8_t SymbolType = Symbol.getELFType(); 808 if (SymbolType == ELF::STT_SECTION) 809 continue; 810 811 uint64_t Address = unwrapOrError(Symbol.getAddress(), Obj.getFileName()); 812 // ELFSymbolRef::getAddress() returns size instead of value for common 813 // symbols which is not desirable for disassembly output. Overriding. 814 if (SymbolType == ELF::STT_COMMON) 815 Address = unwrapOrError(Obj.getSymbol(Symbol.getRawDataRefImpl()), 816 Obj.getFileName()) 817 ->st_value; 818 819 StringRef Name = unwrapOrError(Symbol.getName(), Obj.getFileName()); 820 if (Name.empty()) 821 continue; 822 823 section_iterator SecI = 824 unwrapOrError(Symbol.getSection(), Obj.getFileName()); 825 if (SecI == Obj.section_end()) 826 continue; 827 828 AllSymbols[*SecI].emplace_back(Address, Name, SymbolType); 829 } 830 } 831 832 static void 833 addDynamicElfSymbols(const ELFObjectFileBase &Obj, 834 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 835 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj)) 836 addDynamicElfSymbols(*Elf32LEObj, AllSymbols); 837 else if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj)) 838 addDynamicElfSymbols(*Elf64LEObj, AllSymbols); 839 else if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj)) 840 addDynamicElfSymbols(*Elf32BEObj, AllSymbols); 841 else if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj)) 842 addDynamicElfSymbols(*Elf64BEObj, AllSymbols); 843 else 844 llvm_unreachable("Unsupported binary format"); 845 } 846 847 static Optional<SectionRef> getWasmCodeSection(const WasmObjectFile &Obj) { 848 for (auto SecI : Obj.sections()) { 849 const WasmSection &Section = Obj.getWasmSection(SecI); 850 if (Section.Type == wasm::WASM_SEC_CODE) 851 return SecI; 852 } 853 return None; 854 } 855 856 static void 857 addMissingWasmCodeSymbols(const WasmObjectFile &Obj, 858 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 859 Optional<SectionRef> Section = getWasmCodeSection(Obj); 860 if (!Section) 861 return; 862 SectionSymbolsTy &Symbols = AllSymbols[*Section]; 863 864 std::set<uint64_t> SymbolAddresses; 865 for (const auto &Sym : Symbols) 866 SymbolAddresses.insert(Sym.Addr); 867 868 for (const wasm::WasmFunction &Function : Obj.functions()) { 869 uint64_t Address = Function.CodeSectionOffset; 870 // Only add fallback symbols for functions not already present in the symbol 871 // table. 872 if (SymbolAddresses.count(Address)) 873 continue; 874 // This function has no symbol, so it should have no SymbolName. 875 assert(Function.SymbolName.empty()); 876 // We use DebugName for the name, though it may be empty if there is no 877 // "name" custom section, or that section is missing a name for this 878 // function. 879 StringRef Name = Function.DebugName; 880 Symbols.emplace_back(Address, Name, ELF::STT_NOTYPE); 881 } 882 } 883 884 static void addPltEntries(const ObjectFile &Obj, 885 std::map<SectionRef, SectionSymbolsTy> &AllSymbols, 886 StringSaver &Saver) { 887 Optional<SectionRef> Plt = None; 888 for (const SectionRef &Section : Obj.sections()) { 889 Expected<StringRef> SecNameOrErr = Section.getName(); 890 if (!SecNameOrErr) { 891 consumeError(SecNameOrErr.takeError()); 892 continue; 893 } 894 if (*SecNameOrErr == ".plt") 895 Plt = Section; 896 } 897 if (!Plt) 898 return; 899 if (auto *ElfObj = dyn_cast<ELFObjectFileBase>(&Obj)) { 900 for (auto PltEntry : ElfObj->getPltAddresses()) { 901 if (PltEntry.first) { 902 SymbolRef Symbol(*PltEntry.first, ElfObj); 903 uint8_t SymbolType = getElfSymbolType(Obj, Symbol); 904 if (Expected<StringRef> NameOrErr = Symbol.getName()) { 905 if (!NameOrErr->empty()) 906 AllSymbols[*Plt].emplace_back( 907 PltEntry.second, Saver.save((*NameOrErr + "@plt").str()), 908 SymbolType); 909 continue; 910 } else { 911 // The warning has been reported in disassembleObject(). 912 consumeError(NameOrErr.takeError()); 913 } 914 } 915 reportWarning("PLT entry at 0x" + Twine::utohexstr(PltEntry.second) + 916 " references an invalid symbol", 917 Obj.getFileName()); 918 } 919 } 920 } 921 922 // Normally the disassembly output will skip blocks of zeroes. This function 923 // returns the number of zero bytes that can be skipped when dumping the 924 // disassembly of the instructions in Buf. 925 static size_t countSkippableZeroBytes(ArrayRef<uint8_t> Buf) { 926 // Find the number of leading zeroes. 927 size_t N = 0; 928 while (N < Buf.size() && !Buf[N]) 929 ++N; 930 931 // We may want to skip blocks of zero bytes, but unless we see 932 // at least 8 of them in a row. 933 if (N < 8) 934 return 0; 935 936 // We skip zeroes in multiples of 4 because do not want to truncate an 937 // instruction if it starts with a zero byte. 938 return N & ~0x3; 939 } 940 941 // Returns a map from sections to their relocations. 942 static std::map<SectionRef, std::vector<RelocationRef>> 943 getRelocsMap(object::ObjectFile const &Obj) { 944 std::map<SectionRef, std::vector<RelocationRef>> Ret; 945 uint64_t I = (uint64_t)-1; 946 for (SectionRef Sec : Obj.sections()) { 947 ++I; 948 Expected<section_iterator> RelocatedOrErr = Sec.getRelocatedSection(); 949 if (!RelocatedOrErr) 950 reportError(Obj.getFileName(), 951 "section (" + Twine(I) + 952 "): failed to get a relocated section: " + 953 toString(RelocatedOrErr.takeError())); 954 955 section_iterator Relocated = *RelocatedOrErr; 956 if (Relocated == Obj.section_end() || !checkSectionFilter(*Relocated).Keep) 957 continue; 958 std::vector<RelocationRef> &V = Ret[*Relocated]; 959 append_range(V, Sec.relocations()); 960 // Sort relocations by address. 961 llvm::stable_sort(V, isRelocAddressLess); 962 } 963 return Ret; 964 } 965 966 // Used for --adjust-vma to check if address should be adjusted by the 967 // specified value for a given section. 968 // For ELF we do not adjust non-allocatable sections like debug ones, 969 // because they are not loadable. 970 // TODO: implement for other file formats. 971 static bool shouldAdjustVA(const SectionRef &Section) { 972 const ObjectFile *Obj = Section.getObject(); 973 if (Obj->isELF()) 974 return ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC; 975 return false; 976 } 977 978 979 typedef std::pair<uint64_t, char> MappingSymbolPair; 980 static char getMappingSymbolKind(ArrayRef<MappingSymbolPair> MappingSymbols, 981 uint64_t Address) { 982 auto It = 983 partition_point(MappingSymbols, [Address](const MappingSymbolPair &Val) { 984 return Val.first <= Address; 985 }); 986 // Return zero for any address before the first mapping symbol; this means 987 // we should use the default disassembly mode, depending on the target. 988 if (It == MappingSymbols.begin()) 989 return '\x00'; 990 return (It - 1)->second; 991 } 992 993 static uint64_t dumpARMELFData(uint64_t SectionAddr, uint64_t Index, 994 uint64_t End, const ObjectFile &Obj, 995 ArrayRef<uint8_t> Bytes, 996 ArrayRef<MappingSymbolPair> MappingSymbols, 997 raw_ostream &OS) { 998 support::endianness Endian = 999 Obj.isLittleEndian() ? support::little : support::big; 1000 OS << format("%8" PRIx64 ":\t", SectionAddr + Index); 1001 if (Index + 4 <= End) { 1002 dumpBytes(Bytes.slice(Index, 4), OS); 1003 OS << "\t.word\t" 1004 << format_hex(support::endian::read32(Bytes.data() + Index, Endian), 1005 10); 1006 return 4; 1007 } 1008 if (Index + 2 <= End) { 1009 dumpBytes(Bytes.slice(Index, 2), OS); 1010 OS << "\t\t.short\t" 1011 << format_hex(support::endian::read16(Bytes.data() + Index, Endian), 1012 6); 1013 return 2; 1014 } 1015 dumpBytes(Bytes.slice(Index, 1), OS); 1016 OS << "\t\t.byte\t" << format_hex(Bytes[Index], 4); 1017 return 1; 1018 } 1019 1020 static void dumpELFData(uint64_t SectionAddr, uint64_t Index, uint64_t End, 1021 ArrayRef<uint8_t> Bytes) { 1022 // print out data up to 8 bytes at a time in hex and ascii 1023 uint8_t AsciiData[9] = {'\0'}; 1024 uint8_t Byte; 1025 int NumBytes = 0; 1026 1027 for (; Index < End; ++Index) { 1028 if (NumBytes == 0) 1029 outs() << format("%8" PRIx64 ":", SectionAddr + Index); 1030 Byte = Bytes.slice(Index)[0]; 1031 outs() << format(" %02x", Byte); 1032 AsciiData[NumBytes] = isPrint(Byte) ? Byte : '.'; 1033 1034 uint8_t IndentOffset = 0; 1035 NumBytes++; 1036 if (Index == End - 1 || NumBytes > 8) { 1037 // Indent the space for less than 8 bytes data. 1038 // 2 spaces for byte and one for space between bytes 1039 IndentOffset = 3 * (8 - NumBytes); 1040 for (int Excess = NumBytes; Excess < 8; Excess++) 1041 AsciiData[Excess] = '\0'; 1042 NumBytes = 8; 1043 } 1044 if (NumBytes == 8) { 1045 AsciiData[8] = '\0'; 1046 outs() << std::string(IndentOffset, ' ') << " "; 1047 outs() << reinterpret_cast<char *>(AsciiData); 1048 outs() << '\n'; 1049 NumBytes = 0; 1050 } 1051 } 1052 } 1053 1054 SymbolInfoTy objdump::createSymbolInfo(const ObjectFile &Obj, 1055 const SymbolRef &Symbol) { 1056 const StringRef FileName = Obj.getFileName(); 1057 const uint64_t Addr = unwrapOrError(Symbol.getAddress(), FileName); 1058 const StringRef Name = unwrapOrError(Symbol.getName(), FileName); 1059 1060 if (Obj.isXCOFF() && SymbolDescription) { 1061 const auto &XCOFFObj = cast<XCOFFObjectFile>(Obj); 1062 DataRefImpl SymbolDRI = Symbol.getRawDataRefImpl(); 1063 1064 const uint32_t SymbolIndex = XCOFFObj.getSymbolIndex(SymbolDRI.p); 1065 Optional<XCOFF::StorageMappingClass> Smc = 1066 getXCOFFSymbolCsectSMC(XCOFFObj, Symbol); 1067 return SymbolInfoTy(Addr, Name, Smc, SymbolIndex, 1068 isLabel(XCOFFObj, Symbol)); 1069 } else if (Obj.isXCOFF()) { 1070 const SymbolRef::Type SymType = unwrapOrError(Symbol.getType(), FileName); 1071 return SymbolInfoTy(Addr, Name, SymType, true); 1072 } else 1073 return SymbolInfoTy(Addr, Name, 1074 Obj.isELF() ? getElfSymbolType(Obj, Symbol) 1075 : (uint8_t)ELF::STT_NOTYPE); 1076 } 1077 1078 static SymbolInfoTy createDummySymbolInfo(const ObjectFile &Obj, 1079 const uint64_t Addr, StringRef &Name, 1080 uint8_t Type) { 1081 if (Obj.isXCOFF() && SymbolDescription) 1082 return SymbolInfoTy(Addr, Name, None, None, false); 1083 else 1084 return SymbolInfoTy(Addr, Name, Type); 1085 } 1086 1087 static void 1088 collectBBAddrMapLabels(const std::unordered_map<uint64_t, BBAddrMap> &AddrToBBAddrMap, 1089 uint64_t SectionAddr, uint64_t Start, uint64_t End, 1090 std::unordered_map<uint64_t, std::vector<std::string>> &Labels) { 1091 if (AddrToBBAddrMap.empty()) 1092 return; 1093 Labels.clear(); 1094 uint64_t StartAddress = SectionAddr + Start; 1095 uint64_t EndAddress = SectionAddr + End; 1096 auto Iter = AddrToBBAddrMap.find(StartAddress); 1097 if (Iter == AddrToBBAddrMap.end()) 1098 return; 1099 for (unsigned I = 0, Size = Iter->second.BBEntries.size(); I < Size; ++I) { 1100 uint64_t BBAddress = Iter->second.BBEntries[I].Offset + Iter->second.Addr; 1101 if (BBAddress >= EndAddress) 1102 continue; 1103 Labels[BBAddress].push_back(("BB" + Twine(I)).str()); 1104 } 1105 } 1106 1107 static void collectLocalBranchTargets( 1108 ArrayRef<uint8_t> Bytes, const MCInstrAnalysis *MIA, MCDisassembler *DisAsm, 1109 MCInstPrinter *IP, const MCSubtargetInfo *STI, uint64_t SectionAddr, 1110 uint64_t Start, uint64_t End, std::unordered_map<uint64_t, std::string> &Labels) { 1111 // So far only supports PowerPC and X86. 1112 if (!STI->getTargetTriple().isPPC() && !STI->getTargetTriple().isX86()) 1113 return; 1114 1115 Labels.clear(); 1116 unsigned LabelCount = 0; 1117 Start += SectionAddr; 1118 End += SectionAddr; 1119 uint64_t Index = Start; 1120 while (Index < End) { 1121 // Disassemble a real instruction and record function-local branch labels. 1122 MCInst Inst; 1123 uint64_t Size; 1124 ArrayRef<uint8_t> ThisBytes = Bytes.slice(Index - SectionAddr); 1125 bool Disassembled = 1126 DisAsm->getInstruction(Inst, Size, ThisBytes, Index, nulls()); 1127 if (Size == 0) 1128 Size = std::min(ThisBytes.size(), 1129 DisAsm->suggestBytesToSkip(ThisBytes, Index)); 1130 1131 if (Disassembled && MIA) { 1132 uint64_t Target; 1133 bool TargetKnown = MIA->evaluateBranch(Inst, Index, Size, Target); 1134 // On PowerPC, if the address of a branch is the same as the target, it 1135 // means that it's a function call. Do not mark the label for this case. 1136 if (TargetKnown && (Target >= Start && Target < End) && 1137 !Labels.count(Target) && 1138 !(STI->getTargetTriple().isPPC() && Target == Index)) 1139 Labels[Target] = ("L" + Twine(LabelCount++)).str(); 1140 } 1141 Index += Size; 1142 } 1143 } 1144 1145 // Create an MCSymbolizer for the target and add it to the MCDisassembler. 1146 // This is currently only used on AMDGPU, and assumes the format of the 1147 // void * argument passed to AMDGPU's createMCSymbolizer. 1148 static void addSymbolizer( 1149 MCContext &Ctx, const Target *Target, StringRef TripleName, 1150 MCDisassembler *DisAsm, uint64_t SectionAddr, ArrayRef<uint8_t> Bytes, 1151 SectionSymbolsTy &Symbols, 1152 std::vector<std::unique_ptr<std::string>> &SynthesizedLabelNames) { 1153 1154 std::unique_ptr<MCRelocationInfo> RelInfo( 1155 Target->createMCRelocationInfo(TripleName, Ctx)); 1156 if (!RelInfo) 1157 return; 1158 std::unique_ptr<MCSymbolizer> Symbolizer(Target->createMCSymbolizer( 1159 TripleName, nullptr, nullptr, &Symbols, &Ctx, std::move(RelInfo))); 1160 MCSymbolizer *SymbolizerPtr = &*Symbolizer; 1161 DisAsm->setSymbolizer(std::move(Symbolizer)); 1162 1163 if (!SymbolizeOperands) 1164 return; 1165 1166 // Synthesize labels referenced by branch instructions by 1167 // disassembling, discarding the output, and collecting the referenced 1168 // addresses from the symbolizer. 1169 for (size_t Index = 0; Index != Bytes.size();) { 1170 MCInst Inst; 1171 uint64_t Size; 1172 ArrayRef<uint8_t> ThisBytes = Bytes.slice(Index - SectionAddr); 1173 DisAsm->getInstruction(Inst, Size, ThisBytes, Index, nulls()); 1174 if (Size == 0) 1175 Size = std::min(ThisBytes.size(), 1176 DisAsm->suggestBytesToSkip(ThisBytes, Index)); 1177 Index += Size; 1178 } 1179 ArrayRef<uint64_t> LabelAddrsRef = SymbolizerPtr->getReferencedAddresses(); 1180 // Copy and sort to remove duplicates. 1181 std::vector<uint64_t> LabelAddrs; 1182 LabelAddrs.insert(LabelAddrs.end(), LabelAddrsRef.begin(), 1183 LabelAddrsRef.end()); 1184 llvm::sort(LabelAddrs); 1185 LabelAddrs.resize(std::unique(LabelAddrs.begin(), LabelAddrs.end()) - 1186 LabelAddrs.begin()); 1187 // Add the labels. 1188 for (unsigned LabelNum = 0; LabelNum != LabelAddrs.size(); ++LabelNum) { 1189 auto Name = std::make_unique<std::string>(); 1190 *Name = (Twine("L") + Twine(LabelNum)).str(); 1191 SynthesizedLabelNames.push_back(std::move(Name)); 1192 Symbols.push_back(SymbolInfoTy( 1193 LabelAddrs[LabelNum], *SynthesizedLabelNames.back(), ELF::STT_NOTYPE)); 1194 } 1195 llvm::stable_sort(Symbols); 1196 // Recreate the symbolizer with the new symbols list. 1197 RelInfo.reset(Target->createMCRelocationInfo(TripleName, Ctx)); 1198 Symbolizer.reset(Target->createMCSymbolizer( 1199 TripleName, nullptr, nullptr, &Symbols, &Ctx, std::move(RelInfo))); 1200 DisAsm->setSymbolizer(std::move(Symbolizer)); 1201 } 1202 1203 static StringRef getSegmentName(const MachOObjectFile *MachO, 1204 const SectionRef &Section) { 1205 if (MachO) { 1206 DataRefImpl DR = Section.getRawDataRefImpl(); 1207 StringRef SegmentName = MachO->getSectionFinalSegmentName(DR); 1208 return SegmentName; 1209 } 1210 return ""; 1211 } 1212 1213 static void emitPostInstructionInfo(formatted_raw_ostream &FOS, 1214 const MCAsmInfo &MAI, 1215 const MCSubtargetInfo &STI, 1216 StringRef Comments, 1217 LiveVariablePrinter &LVP) { 1218 do { 1219 if (!Comments.empty()) { 1220 // Emit a line of comments. 1221 StringRef Comment; 1222 std::tie(Comment, Comments) = Comments.split('\n'); 1223 // MAI.getCommentColumn() assumes that instructions are printed at the 1224 // position of 8, while getInstStartColumn() returns the actual position. 1225 unsigned CommentColumn = 1226 MAI.getCommentColumn() - 8 + getInstStartColumn(STI); 1227 FOS.PadToColumn(CommentColumn); 1228 FOS << MAI.getCommentString() << ' ' << Comment; 1229 } 1230 LVP.printAfterInst(FOS); 1231 FOS << '\n'; 1232 } while (!Comments.empty()); 1233 FOS.flush(); 1234 } 1235 1236 static void createFakeELFSections(ObjectFile &Obj) { 1237 assert(Obj.isELF()); 1238 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj)) 1239 Elf32LEObj->createFakeSections(); 1240 else if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj)) 1241 Elf64LEObj->createFakeSections(); 1242 else if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj)) 1243 Elf32BEObj->createFakeSections(); 1244 else if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj)) 1245 Elf64BEObj->createFakeSections(); 1246 else 1247 llvm_unreachable("Unsupported binary format"); 1248 } 1249 1250 static void disassembleObject(const Target *TheTarget, ObjectFile &Obj, 1251 MCContext &Ctx, MCDisassembler *PrimaryDisAsm, 1252 MCDisassembler *SecondaryDisAsm, 1253 const MCInstrAnalysis *MIA, MCInstPrinter *IP, 1254 const MCSubtargetInfo *PrimarySTI, 1255 const MCSubtargetInfo *SecondarySTI, 1256 PrettyPrinter &PIP, SourcePrinter &SP, 1257 bool InlineRelocs) { 1258 const MCSubtargetInfo *STI = PrimarySTI; 1259 MCDisassembler *DisAsm = PrimaryDisAsm; 1260 bool PrimaryIsThumb = false; 1261 if (isArmElf(Obj)) 1262 PrimaryIsThumb = STI->checkFeatures("+thumb-mode"); 1263 1264 std::map<SectionRef, std::vector<RelocationRef>> RelocMap; 1265 if (InlineRelocs) 1266 RelocMap = getRelocsMap(Obj); 1267 bool Is64Bits = Obj.getBytesInAddress() > 4; 1268 1269 // Create a mapping from virtual address to symbol name. This is used to 1270 // pretty print the symbols while disassembling. 1271 std::map<SectionRef, SectionSymbolsTy> AllSymbols; 1272 SectionSymbolsTy AbsoluteSymbols; 1273 const StringRef FileName = Obj.getFileName(); 1274 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(&Obj); 1275 for (const SymbolRef &Symbol : Obj.symbols()) { 1276 Expected<StringRef> NameOrErr = Symbol.getName(); 1277 if (!NameOrErr) { 1278 reportWarning(toString(NameOrErr.takeError()), FileName); 1279 continue; 1280 } 1281 if (NameOrErr->empty() && !(Obj.isXCOFF() && SymbolDescription)) 1282 continue; 1283 1284 if (Obj.isELF() && getElfSymbolType(Obj, Symbol) == ELF::STT_SECTION) 1285 continue; 1286 1287 if (MachO) { 1288 // __mh_(execute|dylib|dylinker|bundle|preload|object)_header are special 1289 // symbols that support MachO header introspection. They do not bind to 1290 // code locations and are irrelevant for disassembly. 1291 if (NameOrErr->startswith("__mh_") && NameOrErr->endswith("_header")) 1292 continue; 1293 // Don't ask a Mach-O STAB symbol for its section unless you know that 1294 // STAB symbol's section field refers to a valid section index. Otherwise 1295 // the symbol may error trying to load a section that does not exist. 1296 DataRefImpl SymDRI = Symbol.getRawDataRefImpl(); 1297 uint8_t NType = (MachO->is64Bit() ? 1298 MachO->getSymbol64TableEntry(SymDRI).n_type: 1299 MachO->getSymbolTableEntry(SymDRI).n_type); 1300 if (NType & MachO::N_STAB) 1301 continue; 1302 } 1303 1304 section_iterator SecI = unwrapOrError(Symbol.getSection(), FileName); 1305 if (SecI != Obj.section_end()) 1306 AllSymbols[*SecI].push_back(createSymbolInfo(Obj, Symbol)); 1307 else 1308 AbsoluteSymbols.push_back(createSymbolInfo(Obj, Symbol)); 1309 } 1310 1311 if (AllSymbols.empty() && Obj.isELF()) 1312 addDynamicElfSymbols(cast<ELFObjectFileBase>(Obj), AllSymbols); 1313 1314 if (Obj.isWasm()) 1315 addMissingWasmCodeSymbols(cast<WasmObjectFile>(Obj), AllSymbols); 1316 1317 if (Obj.isELF() && Obj.sections().empty()) 1318 createFakeELFSections(Obj); 1319 1320 BumpPtrAllocator A; 1321 StringSaver Saver(A); 1322 addPltEntries(Obj, AllSymbols, Saver); 1323 1324 // Create a mapping from virtual address to section. An empty section can 1325 // cause more than one section at the same address. Sort such sections to be 1326 // before same-addressed non-empty sections so that symbol lookups prefer the 1327 // non-empty section. 1328 std::vector<std::pair<uint64_t, SectionRef>> SectionAddresses; 1329 for (SectionRef Sec : Obj.sections()) 1330 SectionAddresses.emplace_back(Sec.getAddress(), Sec); 1331 llvm::stable_sort(SectionAddresses, [](const auto &LHS, const auto &RHS) { 1332 if (LHS.first != RHS.first) 1333 return LHS.first < RHS.first; 1334 return LHS.second.getSize() < RHS.second.getSize(); 1335 }); 1336 1337 // Linked executables (.exe and .dll files) typically don't include a real 1338 // symbol table but they might contain an export table. 1339 if (const auto *COFFObj = dyn_cast<COFFObjectFile>(&Obj)) { 1340 for (const auto &ExportEntry : COFFObj->export_directories()) { 1341 StringRef Name; 1342 if (Error E = ExportEntry.getSymbolName(Name)) 1343 reportError(std::move(E), Obj.getFileName()); 1344 if (Name.empty()) 1345 continue; 1346 1347 uint32_t RVA; 1348 if (Error E = ExportEntry.getExportRVA(RVA)) 1349 reportError(std::move(E), Obj.getFileName()); 1350 1351 uint64_t VA = COFFObj->getImageBase() + RVA; 1352 auto Sec = partition_point( 1353 SectionAddresses, [VA](const std::pair<uint64_t, SectionRef> &O) { 1354 return O.first <= VA; 1355 }); 1356 if (Sec != SectionAddresses.begin()) { 1357 --Sec; 1358 AllSymbols[Sec->second].emplace_back(VA, Name, ELF::STT_NOTYPE); 1359 } else 1360 AbsoluteSymbols.emplace_back(VA, Name, ELF::STT_NOTYPE); 1361 } 1362 } 1363 1364 // Sort all the symbols, this allows us to use a simple binary search to find 1365 // Multiple symbols can have the same address. Use a stable sort to stabilize 1366 // the output. 1367 StringSet<> FoundDisasmSymbolSet; 1368 for (std::pair<const SectionRef, SectionSymbolsTy> &SecSyms : AllSymbols) 1369 llvm::stable_sort(SecSyms.second); 1370 llvm::stable_sort(AbsoluteSymbols); 1371 1372 std::unique_ptr<DWARFContext> DICtx; 1373 LiveVariablePrinter LVP(*Ctx.getRegisterInfo(), *STI); 1374 1375 if (DbgVariables != DVDisabled) { 1376 DICtx = DWARFContext::create(Obj); 1377 for (const std::unique_ptr<DWARFUnit> &CU : DICtx->compile_units()) 1378 LVP.addCompileUnit(CU->getUnitDIE(false)); 1379 } 1380 1381 LLVM_DEBUG(LVP.dump()); 1382 1383 std::unordered_map<uint64_t, BBAddrMap> AddrToBBAddrMap; 1384 auto ReadBBAddrMap = [&](Optional<unsigned> SectionIndex = None) { 1385 AddrToBBAddrMap.clear(); 1386 if (const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj)) { 1387 auto BBAddrMapsOrErr = Elf->readBBAddrMap(SectionIndex); 1388 if (!BBAddrMapsOrErr) 1389 reportWarning(toString(BBAddrMapsOrErr.takeError()), 1390 Obj.getFileName()); 1391 for (auto &FunctionBBAddrMap : *BBAddrMapsOrErr) 1392 AddrToBBAddrMap.emplace(FunctionBBAddrMap.Addr, 1393 std::move(FunctionBBAddrMap)); 1394 } 1395 }; 1396 1397 // For non-relocatable objects, Read all LLVM_BB_ADDR_MAP sections into a 1398 // single mapping, since they don't have any conflicts. 1399 if (SymbolizeOperands && !Obj.isRelocatableObject()) 1400 ReadBBAddrMap(); 1401 1402 for (const SectionRef &Section : ToolSectionFilter(Obj)) { 1403 if (FilterSections.empty() && !DisassembleAll && 1404 (!Section.isText() || Section.isVirtual())) 1405 continue; 1406 1407 uint64_t SectionAddr = Section.getAddress(); 1408 uint64_t SectSize = Section.getSize(); 1409 if (!SectSize) 1410 continue; 1411 1412 // For relocatable object files, read the LLVM_BB_ADDR_MAP section 1413 // corresponding to this section, if present. 1414 if (SymbolizeOperands && Obj.isRelocatableObject()) 1415 ReadBBAddrMap(Section.getIndex()); 1416 1417 // Get the list of all the symbols in this section. 1418 SectionSymbolsTy &Symbols = AllSymbols[Section]; 1419 std::vector<MappingSymbolPair> MappingSymbols; 1420 if (hasMappingSymbols(Obj)) { 1421 for (const auto &Symb : Symbols) { 1422 uint64_t Address = Symb.Addr; 1423 StringRef Name = Symb.Name; 1424 if (Name.startswith("$d")) 1425 MappingSymbols.emplace_back(Address - SectionAddr, 'd'); 1426 if (Name.startswith("$x")) 1427 MappingSymbols.emplace_back(Address - SectionAddr, 'x'); 1428 if (Name.startswith("$a")) 1429 MappingSymbols.emplace_back(Address - SectionAddr, 'a'); 1430 if (Name.startswith("$t")) 1431 MappingSymbols.emplace_back(Address - SectionAddr, 't'); 1432 } 1433 } 1434 1435 llvm::sort(MappingSymbols); 1436 1437 ArrayRef<uint8_t> Bytes = arrayRefFromStringRef( 1438 unwrapOrError(Section.getContents(), Obj.getFileName())); 1439 1440 std::vector<std::unique_ptr<std::string>> SynthesizedLabelNames; 1441 if (Obj.isELF() && Obj.getArch() == Triple::amdgcn) { 1442 // AMDGPU disassembler uses symbolizer for printing labels 1443 addSymbolizer(Ctx, TheTarget, TripleName, DisAsm, SectionAddr, Bytes, 1444 Symbols, SynthesizedLabelNames); 1445 } 1446 1447 StringRef SegmentName = getSegmentName(MachO, Section); 1448 StringRef SectionName = unwrapOrError(Section.getName(), Obj.getFileName()); 1449 // If the section has no symbol at the start, just insert a dummy one. 1450 if (Symbols.empty() || Symbols[0].Addr != 0) { 1451 Symbols.insert(Symbols.begin(), 1452 createDummySymbolInfo(Obj, SectionAddr, SectionName, 1453 Section.isText() ? ELF::STT_FUNC 1454 : ELF::STT_OBJECT)); 1455 } 1456 1457 SmallString<40> Comments; 1458 raw_svector_ostream CommentStream(Comments); 1459 1460 uint64_t VMAAdjustment = 0; 1461 if (shouldAdjustVA(Section)) 1462 VMAAdjustment = AdjustVMA; 1463 1464 // In executable and shared objects, r_offset holds a virtual address. 1465 // Subtract SectionAddr from the r_offset field of a relocation to get 1466 // the section offset. 1467 uint64_t RelAdjustment = Obj.isRelocatableObject() ? 0 : SectionAddr; 1468 uint64_t Size; 1469 uint64_t Index; 1470 bool PrintedSection = false; 1471 std::vector<RelocationRef> Rels = RelocMap[Section]; 1472 std::vector<RelocationRef>::const_iterator RelCur = Rels.begin(); 1473 std::vector<RelocationRef>::const_iterator RelEnd = Rels.end(); 1474 // Disassemble symbol by symbol. 1475 for (unsigned SI = 0, SE = Symbols.size(); SI != SE; ++SI) { 1476 std::string SymbolName = Symbols[SI].Name.str(); 1477 if (Demangle) 1478 SymbolName = demangle(SymbolName); 1479 1480 // Skip if --disassemble-symbols is not empty and the symbol is not in 1481 // the list. 1482 if (!DisasmSymbolSet.empty() && !DisasmSymbolSet.count(SymbolName)) 1483 continue; 1484 1485 uint64_t Start = Symbols[SI].Addr; 1486 if (Start < SectionAddr || StopAddress <= Start) 1487 continue; 1488 else 1489 FoundDisasmSymbolSet.insert(SymbolName); 1490 1491 // The end is the section end, the beginning of the next symbol, or 1492 // --stop-address. 1493 uint64_t End = std::min<uint64_t>(SectionAddr + SectSize, StopAddress); 1494 if (SI + 1 < SE) 1495 End = std::min(End, Symbols[SI + 1].Addr); 1496 if (Start >= End || End <= StartAddress) 1497 continue; 1498 Start -= SectionAddr; 1499 End -= SectionAddr; 1500 1501 if (!PrintedSection) { 1502 PrintedSection = true; 1503 outs() << "\nDisassembly of section "; 1504 if (!SegmentName.empty()) 1505 outs() << SegmentName << ","; 1506 outs() << SectionName << ":\n"; 1507 } 1508 1509 outs() << '\n'; 1510 if (LeadingAddr) 1511 outs() << format(Is64Bits ? "%016" PRIx64 " " : "%08" PRIx64 " ", 1512 SectionAddr + Start + VMAAdjustment); 1513 if (Obj.isXCOFF() && SymbolDescription) { 1514 outs() << getXCOFFSymbolDescription(Symbols[SI], SymbolName) << ":\n"; 1515 } else 1516 outs() << '<' << SymbolName << ">:\n"; 1517 1518 // Don't print raw contents of a virtual section. A virtual section 1519 // doesn't have any contents in the file. 1520 if (Section.isVirtual()) { 1521 outs() << "...\n"; 1522 continue; 1523 } 1524 1525 auto Status = DisAsm->onSymbolStart(Symbols[SI], Size, 1526 Bytes.slice(Start, End - Start), 1527 SectionAddr + Start, CommentStream); 1528 // To have round trippable disassembly, we fall back to decoding the 1529 // remaining bytes as instructions. 1530 // 1531 // If there is a failure, we disassemble the failed region as bytes before 1532 // falling back. The target is expected to print nothing in this case. 1533 // 1534 // If there is Success or SoftFail i.e no 'real' failure, we go ahead by 1535 // Size bytes before falling back. 1536 // So if the entire symbol is 'eaten' by the target: 1537 // Start += Size // Now Start = End and we will never decode as 1538 // // instructions 1539 // 1540 // Right now, most targets return None i.e ignore to treat a symbol 1541 // separately. But WebAssembly decodes preludes for some symbols. 1542 // 1543 if (Status) { 1544 if (Status.value() == MCDisassembler::Fail) { 1545 outs() << "// Error in decoding " << SymbolName 1546 << " : Decoding failed region as bytes.\n"; 1547 for (uint64_t I = 0; I < Size; ++I) { 1548 outs() << "\t.byte\t " << format_hex(Bytes[I], 1, /*Upper=*/true) 1549 << "\n"; 1550 } 1551 } 1552 } else { 1553 Size = 0; 1554 } 1555 1556 Start += Size; 1557 1558 Index = Start; 1559 if (SectionAddr < StartAddress) 1560 Index = std::max<uint64_t>(Index, StartAddress - SectionAddr); 1561 1562 // If there is a data/common symbol inside an ELF text section and we are 1563 // only disassembling text (applicable all architectures), we are in a 1564 // situation where we must print the data and not disassemble it. 1565 if (Obj.isELF() && !DisassembleAll && Section.isText()) { 1566 uint8_t SymTy = Symbols[SI].Type; 1567 if (SymTy == ELF::STT_OBJECT || SymTy == ELF::STT_COMMON) { 1568 dumpELFData(SectionAddr, Index, End, Bytes); 1569 Index = End; 1570 } 1571 } 1572 1573 bool CheckARMELFData = hasMappingSymbols(Obj) && 1574 Symbols[SI].Type != ELF::STT_OBJECT && 1575 !DisassembleAll; 1576 bool DumpARMELFData = false; 1577 formatted_raw_ostream FOS(outs()); 1578 1579 std::unordered_map<uint64_t, std::string> AllLabels; 1580 std::unordered_map<uint64_t, std::vector<std::string>> BBAddrMapLabels; 1581 if (SymbolizeOperands) { 1582 collectLocalBranchTargets(Bytes, MIA, DisAsm, IP, PrimarySTI, 1583 SectionAddr, Index, End, AllLabels); 1584 collectBBAddrMapLabels(AddrToBBAddrMap, SectionAddr, Index, End, 1585 BBAddrMapLabels); 1586 } 1587 1588 while (Index < End) { 1589 // ARM and AArch64 ELF binaries can interleave data and text in the 1590 // same section. We rely on the markers introduced to understand what 1591 // we need to dump. If the data marker is within a function, it is 1592 // denoted as a word/short etc. 1593 if (CheckARMELFData) { 1594 char Kind = getMappingSymbolKind(MappingSymbols, Index); 1595 DumpARMELFData = Kind == 'd'; 1596 if (SecondarySTI) { 1597 if (Kind == 'a') { 1598 STI = PrimaryIsThumb ? SecondarySTI : PrimarySTI; 1599 DisAsm = PrimaryIsThumb ? SecondaryDisAsm : PrimaryDisAsm; 1600 } else if (Kind == 't') { 1601 STI = PrimaryIsThumb ? PrimarySTI : SecondarySTI; 1602 DisAsm = PrimaryIsThumb ? PrimaryDisAsm : SecondaryDisAsm; 1603 } 1604 } 1605 } 1606 1607 if (DumpARMELFData) { 1608 Size = dumpARMELFData(SectionAddr, Index, End, Obj, Bytes, 1609 MappingSymbols, FOS); 1610 } else { 1611 // When -z or --disassemble-zeroes are given we always dissasemble 1612 // them. Otherwise we might want to skip zero bytes we see. 1613 if (!DisassembleZeroes) { 1614 uint64_t MaxOffset = End - Index; 1615 // For --reloc: print zero blocks patched by relocations, so that 1616 // relocations can be shown in the dump. 1617 if (RelCur != RelEnd) 1618 MaxOffset = std::min(RelCur->getOffset() - RelAdjustment - Index, 1619 MaxOffset); 1620 1621 if (size_t N = 1622 countSkippableZeroBytes(Bytes.slice(Index, MaxOffset))) { 1623 FOS << "\t\t..." << '\n'; 1624 Index += N; 1625 continue; 1626 } 1627 } 1628 1629 // Print local label if there's any. 1630 auto Iter1 = BBAddrMapLabels.find(SectionAddr + Index); 1631 if (Iter1 != BBAddrMapLabels.end()) { 1632 for (StringRef Label : Iter1->second) 1633 FOS << "<" << Label << ">:\n"; 1634 } else { 1635 auto Iter2 = AllLabels.find(SectionAddr + Index); 1636 if (Iter2 != AllLabels.end()) 1637 FOS << "<" << Iter2->second << ">:\n"; 1638 } 1639 1640 // Disassemble a real instruction or a data when disassemble all is 1641 // provided 1642 MCInst Inst; 1643 ArrayRef<uint8_t> ThisBytes = Bytes.slice(Index); 1644 uint64_t ThisAddr = SectionAddr + Index; 1645 bool Disassembled = DisAsm->getInstruction(Inst, Size, ThisBytes, 1646 ThisAddr, CommentStream); 1647 if (Size == 0) 1648 Size = std::min(ThisBytes.size(), 1649 DisAsm->suggestBytesToSkip(ThisBytes, ThisAddr)); 1650 1651 LVP.update({Index, Section.getIndex()}, 1652 {Index + Size, Section.getIndex()}, Index + Size != End); 1653 1654 IP->setCommentStream(CommentStream); 1655 1656 PIP.printInst( 1657 *IP, Disassembled ? &Inst : nullptr, Bytes.slice(Index, Size), 1658 {SectionAddr + Index + VMAAdjustment, Section.getIndex()}, FOS, 1659 "", *STI, &SP, Obj.getFileName(), &Rels, LVP); 1660 1661 IP->setCommentStream(llvm::nulls()); 1662 1663 // If disassembly has failed, avoid analysing invalid/incomplete 1664 // instruction information. Otherwise, try to resolve the target 1665 // address (jump target or memory operand address) and print it on the 1666 // right of the instruction. 1667 if (Disassembled && MIA) { 1668 // Branch targets are printed just after the instructions. 1669 llvm::raw_ostream *TargetOS = &FOS; 1670 uint64_t Target; 1671 bool PrintTarget = 1672 MIA->evaluateBranch(Inst, SectionAddr + Index, Size, Target); 1673 if (!PrintTarget) 1674 if (Optional<uint64_t> MaybeTarget = 1675 MIA->evaluateMemoryOperandAddress( 1676 Inst, STI, SectionAddr + Index, Size)) { 1677 Target = *MaybeTarget; 1678 PrintTarget = true; 1679 // Do not print real address when symbolizing. 1680 if (!SymbolizeOperands) { 1681 // Memory operand addresses are printed as comments. 1682 TargetOS = &CommentStream; 1683 *TargetOS << "0x" << Twine::utohexstr(Target); 1684 } 1685 } 1686 if (PrintTarget) { 1687 // In a relocatable object, the target's section must reside in 1688 // the same section as the call instruction or it is accessed 1689 // through a relocation. 1690 // 1691 // In a non-relocatable object, the target may be in any section. 1692 // In that case, locate the section(s) containing the target 1693 // address and find the symbol in one of those, if possible. 1694 // 1695 // N.B. We don't walk the relocations in the relocatable case yet. 1696 std::vector<const SectionSymbolsTy *> TargetSectionSymbols; 1697 if (!Obj.isRelocatableObject()) { 1698 auto It = llvm::partition_point( 1699 SectionAddresses, 1700 [=](const std::pair<uint64_t, SectionRef> &O) { 1701 return O.first <= Target; 1702 }); 1703 uint64_t TargetSecAddr = 0; 1704 while (It != SectionAddresses.begin()) { 1705 --It; 1706 if (TargetSecAddr == 0) 1707 TargetSecAddr = It->first; 1708 if (It->first != TargetSecAddr) 1709 break; 1710 TargetSectionSymbols.push_back(&AllSymbols[It->second]); 1711 } 1712 } else { 1713 TargetSectionSymbols.push_back(&Symbols); 1714 } 1715 TargetSectionSymbols.push_back(&AbsoluteSymbols); 1716 1717 // Find the last symbol in the first candidate section whose 1718 // offset is less than or equal to the target. If there are no 1719 // such symbols, try in the next section and so on, before finally 1720 // using the nearest preceding absolute symbol (if any), if there 1721 // are no other valid symbols. 1722 const SymbolInfoTy *TargetSym = nullptr; 1723 for (const SectionSymbolsTy *TargetSymbols : 1724 TargetSectionSymbols) { 1725 auto It = llvm::partition_point( 1726 *TargetSymbols, 1727 [=](const SymbolInfoTy &O) { return O.Addr <= Target; }); 1728 if (It != TargetSymbols->begin()) { 1729 TargetSym = &*(It - 1); 1730 break; 1731 } 1732 } 1733 1734 // Print the labels corresponding to the target if there's any. 1735 bool BBAddrMapLabelAvailable = BBAddrMapLabels.count(Target); 1736 bool LabelAvailable = AllLabels.count(Target); 1737 if (TargetSym != nullptr) { 1738 uint64_t TargetAddress = TargetSym->Addr; 1739 uint64_t Disp = Target - TargetAddress; 1740 std::string TargetName = TargetSym->Name.str(); 1741 if (Demangle) 1742 TargetName = demangle(TargetName); 1743 1744 *TargetOS << " <"; 1745 if (!Disp) { 1746 // Always Print the binary symbol precisely corresponding to 1747 // the target address. 1748 *TargetOS << TargetName; 1749 } else if (BBAddrMapLabelAvailable) { 1750 *TargetOS << BBAddrMapLabels[Target].front(); 1751 } else if (LabelAvailable) { 1752 *TargetOS << AllLabels[Target]; 1753 } else { 1754 // Always Print the binary symbol plus an offset if there's no 1755 // local label corresponding to the target address. 1756 *TargetOS << TargetName << "+0x" << Twine::utohexstr(Disp); 1757 } 1758 *TargetOS << ">"; 1759 } else if (BBAddrMapLabelAvailable) { 1760 *TargetOS << " <" << BBAddrMapLabels[Target].front() << ">"; 1761 } else if (LabelAvailable) { 1762 *TargetOS << " <" << AllLabels[Target] << ">"; 1763 } 1764 // By convention, each record in the comment stream should be 1765 // terminated. 1766 if (TargetOS == &CommentStream) 1767 *TargetOS << "\n"; 1768 } 1769 } 1770 } 1771 1772 assert(Ctx.getAsmInfo()); 1773 emitPostInstructionInfo(FOS, *Ctx.getAsmInfo(), *STI, 1774 CommentStream.str(), LVP); 1775 Comments.clear(); 1776 1777 // Hexagon does this in pretty printer 1778 if (Obj.getArch() != Triple::hexagon) { 1779 // Print relocation for instruction and data. 1780 while (RelCur != RelEnd) { 1781 uint64_t Offset = RelCur->getOffset() - RelAdjustment; 1782 // If this relocation is hidden, skip it. 1783 if (getHidden(*RelCur) || SectionAddr + Offset < StartAddress) { 1784 ++RelCur; 1785 continue; 1786 } 1787 1788 // Stop when RelCur's offset is past the disassembled 1789 // instruction/data. Note that it's possible the disassembled data 1790 // is not the complete data: we might see the relocation printed in 1791 // the middle of the data, but this matches the binutils objdump 1792 // output. 1793 if (Offset >= Index + Size) 1794 break; 1795 1796 // When --adjust-vma is used, update the address printed. 1797 if (RelCur->getSymbol() != Obj.symbol_end()) { 1798 Expected<section_iterator> SymSI = 1799 RelCur->getSymbol()->getSection(); 1800 if (SymSI && *SymSI != Obj.section_end() && 1801 shouldAdjustVA(**SymSI)) 1802 Offset += AdjustVMA; 1803 } 1804 1805 printRelocation(FOS, Obj.getFileName(), *RelCur, 1806 SectionAddr + Offset, Is64Bits); 1807 LVP.printAfterOtherLine(FOS, true); 1808 ++RelCur; 1809 } 1810 } 1811 1812 Index += Size; 1813 } 1814 } 1815 } 1816 StringSet<> MissingDisasmSymbolSet = 1817 set_difference(DisasmSymbolSet, FoundDisasmSymbolSet); 1818 for (StringRef Sym : MissingDisasmSymbolSet.keys()) 1819 reportWarning("failed to disassemble missing symbol " + Sym, FileName); 1820 } 1821 1822 static void disassembleObject(ObjectFile *Obj, bool InlineRelocs) { 1823 const Target *TheTarget = getTarget(Obj); 1824 1825 // Package up features to be passed to target/subtarget 1826 SubtargetFeatures Features = Obj->getFeatures(); 1827 if (!MAttrs.empty()) { 1828 for (unsigned I = 0; I != MAttrs.size(); ++I) 1829 Features.AddFeature(MAttrs[I]); 1830 } else if (MCPU.empty() && Obj->getArch() == llvm::Triple::aarch64) { 1831 Features.AddFeature("+all"); 1832 } 1833 1834 std::unique_ptr<const MCRegisterInfo> MRI( 1835 TheTarget->createMCRegInfo(TripleName)); 1836 if (!MRI) 1837 reportError(Obj->getFileName(), 1838 "no register info for target " + TripleName); 1839 1840 // Set up disassembler. 1841 MCTargetOptions MCOptions; 1842 std::unique_ptr<const MCAsmInfo> AsmInfo( 1843 TheTarget->createMCAsmInfo(*MRI, TripleName, MCOptions)); 1844 if (!AsmInfo) 1845 reportError(Obj->getFileName(), 1846 "no assembly info for target " + TripleName); 1847 1848 if (MCPU.empty()) 1849 MCPU = Obj->tryGetCPUName().value_or("").str(); 1850 1851 std::unique_ptr<const MCSubtargetInfo> STI( 1852 TheTarget->createMCSubtargetInfo(TripleName, MCPU, Features.getString())); 1853 if (!STI) 1854 reportError(Obj->getFileName(), 1855 "no subtarget info for target " + TripleName); 1856 std::unique_ptr<const MCInstrInfo> MII(TheTarget->createMCInstrInfo()); 1857 if (!MII) 1858 reportError(Obj->getFileName(), 1859 "no instruction info for target " + TripleName); 1860 MCContext Ctx(Triple(TripleName), AsmInfo.get(), MRI.get(), STI.get()); 1861 // FIXME: for now initialize MCObjectFileInfo with default values 1862 std::unique_ptr<MCObjectFileInfo> MOFI( 1863 TheTarget->createMCObjectFileInfo(Ctx, /*PIC=*/false)); 1864 Ctx.setObjectFileInfo(MOFI.get()); 1865 1866 std::unique_ptr<MCDisassembler> DisAsm( 1867 TheTarget->createMCDisassembler(*STI, Ctx)); 1868 if (!DisAsm) 1869 reportError(Obj->getFileName(), "no disassembler for target " + TripleName); 1870 1871 // If we have an ARM object file, we need a second disassembler, because 1872 // ARM CPUs have two different instruction sets: ARM mode, and Thumb mode. 1873 // We use mapping symbols to switch between the two assemblers, where 1874 // appropriate. 1875 std::unique_ptr<MCDisassembler> SecondaryDisAsm; 1876 std::unique_ptr<const MCSubtargetInfo> SecondarySTI; 1877 if (isArmElf(*Obj) && !STI->checkFeatures("+mclass")) { 1878 if (STI->checkFeatures("+thumb-mode")) 1879 Features.AddFeature("-thumb-mode"); 1880 else 1881 Features.AddFeature("+thumb-mode"); 1882 SecondarySTI.reset(TheTarget->createMCSubtargetInfo(TripleName, MCPU, 1883 Features.getString())); 1884 SecondaryDisAsm.reset(TheTarget->createMCDisassembler(*SecondarySTI, Ctx)); 1885 } 1886 1887 std::unique_ptr<const MCInstrAnalysis> MIA( 1888 TheTarget->createMCInstrAnalysis(MII.get())); 1889 1890 int AsmPrinterVariant = AsmInfo->getAssemblerDialect(); 1891 std::unique_ptr<MCInstPrinter> IP(TheTarget->createMCInstPrinter( 1892 Triple(TripleName), AsmPrinterVariant, *AsmInfo, *MII, *MRI)); 1893 if (!IP) 1894 reportError(Obj->getFileName(), 1895 "no instruction printer for target " + TripleName); 1896 IP->setPrintImmHex(PrintImmHex); 1897 IP->setPrintBranchImmAsAddress(true); 1898 IP->setSymbolizeOperands(SymbolizeOperands); 1899 IP->setMCInstrAnalysis(MIA.get()); 1900 1901 PrettyPrinter &PIP = selectPrettyPrinter(Triple(TripleName)); 1902 SourcePrinter SP(Obj, TheTarget->getName()); 1903 1904 for (StringRef Opt : DisassemblerOptions) 1905 if (!IP->applyTargetSpecificCLOption(Opt)) 1906 reportError(Obj->getFileName(), 1907 "Unrecognized disassembler option: " + Opt); 1908 1909 disassembleObject(TheTarget, *Obj, Ctx, DisAsm.get(), SecondaryDisAsm.get(), 1910 MIA.get(), IP.get(), STI.get(), SecondarySTI.get(), PIP, SP, 1911 InlineRelocs); 1912 } 1913 1914 void objdump::printRelocations(const ObjectFile *Obj) { 1915 StringRef Fmt = Obj->getBytesInAddress() > 4 ? "%016" PRIx64 : 1916 "%08" PRIx64; 1917 1918 // Build a mapping from relocation target to a vector of relocation 1919 // sections. Usually, there is an only one relocation section for 1920 // each relocated section. 1921 MapVector<SectionRef, std::vector<SectionRef>> SecToRelSec; 1922 uint64_t Ndx; 1923 for (const SectionRef &Section : ToolSectionFilter(*Obj, &Ndx)) { 1924 if (Obj->isELF() && (ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC)) 1925 continue; 1926 if (Section.relocation_begin() == Section.relocation_end()) 1927 continue; 1928 Expected<section_iterator> SecOrErr = Section.getRelocatedSection(); 1929 if (!SecOrErr) 1930 reportError(Obj->getFileName(), 1931 "section (" + Twine(Ndx) + 1932 "): unable to get a relocation target: " + 1933 toString(SecOrErr.takeError())); 1934 SecToRelSec[**SecOrErr].push_back(Section); 1935 } 1936 1937 for (std::pair<SectionRef, std::vector<SectionRef>> &P : SecToRelSec) { 1938 StringRef SecName = unwrapOrError(P.first.getName(), Obj->getFileName()); 1939 outs() << "\nRELOCATION RECORDS FOR [" << SecName << "]:\n"; 1940 uint32_t OffsetPadding = (Obj->getBytesInAddress() > 4 ? 16 : 8); 1941 uint32_t TypePadding = 24; 1942 outs() << left_justify("OFFSET", OffsetPadding) << " " 1943 << left_justify("TYPE", TypePadding) << " " 1944 << "VALUE\n"; 1945 1946 for (SectionRef Section : P.second) { 1947 for (const RelocationRef &Reloc : Section.relocations()) { 1948 uint64_t Address = Reloc.getOffset(); 1949 SmallString<32> RelocName; 1950 SmallString<32> ValueStr; 1951 if (Address < StartAddress || Address > StopAddress || getHidden(Reloc)) 1952 continue; 1953 Reloc.getTypeName(RelocName); 1954 if (Error E = getRelocationValueString(Reloc, ValueStr)) 1955 reportError(std::move(E), Obj->getFileName()); 1956 1957 outs() << format(Fmt.data(), Address) << " " 1958 << left_justify(RelocName, TypePadding) << " " << ValueStr 1959 << "\n"; 1960 } 1961 } 1962 } 1963 } 1964 1965 void objdump::printDynamicRelocations(const ObjectFile *Obj) { 1966 // For the moment, this option is for ELF only 1967 if (!Obj->isELF()) 1968 return; 1969 1970 const auto *Elf = dyn_cast<ELFObjectFileBase>(Obj); 1971 if (!Elf || !any_of(Elf->sections(), [](const ELFSectionRef Sec) { 1972 return Sec.getType() == ELF::SHT_DYNAMIC; 1973 })) { 1974 reportError(Obj->getFileName(), "not a dynamic object"); 1975 return; 1976 } 1977 1978 std::vector<SectionRef> DynRelSec = Obj->dynamic_relocation_sections(); 1979 if (DynRelSec.empty()) 1980 return; 1981 1982 outs() << "\nDYNAMIC RELOCATION RECORDS\n"; 1983 const uint32_t OffsetPadding = (Obj->getBytesInAddress() > 4 ? 16 : 8); 1984 const uint32_t TypePadding = 24; 1985 outs() << left_justify("OFFSET", OffsetPadding) << ' ' 1986 << left_justify("TYPE", TypePadding) << " VALUE\n"; 1987 1988 StringRef Fmt = Obj->getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 1989 for (const SectionRef &Section : DynRelSec) 1990 for (const RelocationRef &Reloc : Section.relocations()) { 1991 uint64_t Address = Reloc.getOffset(); 1992 SmallString<32> RelocName; 1993 SmallString<32> ValueStr; 1994 Reloc.getTypeName(RelocName); 1995 if (Error E = getRelocationValueString(Reloc, ValueStr)) 1996 reportError(std::move(E), Obj->getFileName()); 1997 outs() << format(Fmt.data(), Address) << ' ' 1998 << left_justify(RelocName, TypePadding) << ' ' << ValueStr << '\n'; 1999 } 2000 } 2001 2002 // Returns true if we need to show LMA column when dumping section headers. We 2003 // show it only when the platform is ELF and either we have at least one section 2004 // whose VMA and LMA are different and/or when --show-lma flag is used. 2005 static bool shouldDisplayLMA(const ObjectFile &Obj) { 2006 if (!Obj.isELF()) 2007 return false; 2008 for (const SectionRef &S : ToolSectionFilter(Obj)) 2009 if (S.getAddress() != getELFSectionLMA(S)) 2010 return true; 2011 return ShowLMA; 2012 } 2013 2014 static size_t getMaxSectionNameWidth(const ObjectFile &Obj) { 2015 // Default column width for names is 13 even if no names are that long. 2016 size_t MaxWidth = 13; 2017 for (const SectionRef &Section : ToolSectionFilter(Obj)) { 2018 StringRef Name = unwrapOrError(Section.getName(), Obj.getFileName()); 2019 MaxWidth = std::max(MaxWidth, Name.size()); 2020 } 2021 return MaxWidth; 2022 } 2023 2024 void objdump::printSectionHeaders(ObjectFile &Obj) { 2025 size_t NameWidth = getMaxSectionNameWidth(Obj); 2026 size_t AddressWidth = 2 * Obj.getBytesInAddress(); 2027 bool HasLMAColumn = shouldDisplayLMA(Obj); 2028 outs() << "\nSections:\n"; 2029 if (HasLMAColumn) 2030 outs() << "Idx " << left_justify("Name", NameWidth) << " Size " 2031 << left_justify("VMA", AddressWidth) << " " 2032 << left_justify("LMA", AddressWidth) << " Type\n"; 2033 else 2034 outs() << "Idx " << left_justify("Name", NameWidth) << " Size " 2035 << left_justify("VMA", AddressWidth) << " Type\n"; 2036 2037 if (Obj.isELF() && Obj.sections().empty()) 2038 createFakeELFSections(Obj); 2039 2040 uint64_t Idx; 2041 for (const SectionRef &Section : ToolSectionFilter(Obj, &Idx)) { 2042 StringRef Name = unwrapOrError(Section.getName(), Obj.getFileName()); 2043 uint64_t VMA = Section.getAddress(); 2044 if (shouldAdjustVA(Section)) 2045 VMA += AdjustVMA; 2046 2047 uint64_t Size = Section.getSize(); 2048 2049 std::string Type = Section.isText() ? "TEXT" : ""; 2050 if (Section.isData()) 2051 Type += Type.empty() ? "DATA" : ", DATA"; 2052 if (Section.isBSS()) 2053 Type += Type.empty() ? "BSS" : ", BSS"; 2054 if (Section.isDebugSection()) 2055 Type += Type.empty() ? "DEBUG" : ", DEBUG"; 2056 2057 if (HasLMAColumn) 2058 outs() << format("%3" PRIu64 " %-*s %08" PRIx64 " ", Idx, NameWidth, 2059 Name.str().c_str(), Size) 2060 << format_hex_no_prefix(VMA, AddressWidth) << " " 2061 << format_hex_no_prefix(getELFSectionLMA(Section), AddressWidth) 2062 << " " << Type << "\n"; 2063 else 2064 outs() << format("%3" PRIu64 " %-*s %08" PRIx64 " ", Idx, NameWidth, 2065 Name.str().c_str(), Size) 2066 << format_hex_no_prefix(VMA, AddressWidth) << " " << Type << "\n"; 2067 } 2068 } 2069 2070 void objdump::printSectionContents(const ObjectFile *Obj) { 2071 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(Obj); 2072 2073 for (const SectionRef &Section : ToolSectionFilter(*Obj)) { 2074 StringRef Name = unwrapOrError(Section.getName(), Obj->getFileName()); 2075 uint64_t BaseAddr = Section.getAddress(); 2076 uint64_t Size = Section.getSize(); 2077 if (!Size) 2078 continue; 2079 2080 outs() << "Contents of section "; 2081 StringRef SegmentName = getSegmentName(MachO, Section); 2082 if (!SegmentName.empty()) 2083 outs() << SegmentName << ","; 2084 outs() << Name << ":\n"; 2085 if (Section.isBSS()) { 2086 outs() << format("<skipping contents of bss section at [%04" PRIx64 2087 ", %04" PRIx64 ")>\n", 2088 BaseAddr, BaseAddr + Size); 2089 continue; 2090 } 2091 2092 StringRef Contents = unwrapOrError(Section.getContents(), Obj->getFileName()); 2093 2094 // Dump out the content as hex and printable ascii characters. 2095 for (std::size_t Addr = 0, End = Contents.size(); Addr < End; Addr += 16) { 2096 outs() << format(" %04" PRIx64 " ", BaseAddr + Addr); 2097 // Dump line of hex. 2098 for (std::size_t I = 0; I < 16; ++I) { 2099 if (I != 0 && I % 4 == 0) 2100 outs() << ' '; 2101 if (Addr + I < End) 2102 outs() << hexdigit((Contents[Addr + I] >> 4) & 0xF, true) 2103 << hexdigit(Contents[Addr + I] & 0xF, true); 2104 else 2105 outs() << " "; 2106 } 2107 // Print ascii. 2108 outs() << " "; 2109 for (std::size_t I = 0; I < 16 && Addr + I < End; ++I) { 2110 if (isPrint(static_cast<unsigned char>(Contents[Addr + I]) & 0xFF)) 2111 outs() << Contents[Addr + I]; 2112 else 2113 outs() << "."; 2114 } 2115 outs() << "\n"; 2116 } 2117 } 2118 } 2119 2120 void objdump::printSymbolTable(const ObjectFile &O, StringRef ArchiveName, 2121 StringRef ArchitectureName, bool DumpDynamic) { 2122 if (O.isCOFF() && !DumpDynamic) { 2123 outs() << "\nSYMBOL TABLE:\n"; 2124 printCOFFSymbolTable(cast<const COFFObjectFile>(O)); 2125 return; 2126 } 2127 2128 const StringRef FileName = O.getFileName(); 2129 2130 if (!DumpDynamic) { 2131 outs() << "\nSYMBOL TABLE:\n"; 2132 for (auto I = O.symbol_begin(); I != O.symbol_end(); ++I) 2133 printSymbol(O, *I, {}, FileName, ArchiveName, ArchitectureName, 2134 DumpDynamic); 2135 return; 2136 } 2137 2138 outs() << "\nDYNAMIC SYMBOL TABLE:\n"; 2139 if (!O.isELF()) { 2140 reportWarning( 2141 "this operation is not currently supported for this file format", 2142 FileName); 2143 return; 2144 } 2145 2146 const ELFObjectFileBase *ELF = cast<const ELFObjectFileBase>(&O); 2147 auto Symbols = ELF->getDynamicSymbolIterators(); 2148 Expected<std::vector<VersionEntry>> SymbolVersionsOrErr = 2149 ELF->readDynsymVersions(); 2150 if (!SymbolVersionsOrErr) { 2151 reportWarning(toString(SymbolVersionsOrErr.takeError()), FileName); 2152 SymbolVersionsOrErr = std::vector<VersionEntry>(); 2153 (void)!SymbolVersionsOrErr; 2154 } 2155 for (auto &Sym : Symbols) 2156 printSymbol(O, Sym, *SymbolVersionsOrErr, FileName, ArchiveName, 2157 ArchitectureName, DumpDynamic); 2158 } 2159 2160 void objdump::printSymbol(const ObjectFile &O, const SymbolRef &Symbol, 2161 ArrayRef<VersionEntry> SymbolVersions, 2162 StringRef FileName, StringRef ArchiveName, 2163 StringRef ArchitectureName, bool DumpDynamic) { 2164 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(&O); 2165 uint64_t Address = unwrapOrError(Symbol.getAddress(), FileName, ArchiveName, 2166 ArchitectureName); 2167 if ((Address < StartAddress) || (Address > StopAddress)) 2168 return; 2169 SymbolRef::Type Type = 2170 unwrapOrError(Symbol.getType(), FileName, ArchiveName, ArchitectureName); 2171 uint32_t Flags = 2172 unwrapOrError(Symbol.getFlags(), FileName, ArchiveName, ArchitectureName); 2173 2174 // Don't ask a Mach-O STAB symbol for its section unless you know that 2175 // STAB symbol's section field refers to a valid section index. Otherwise 2176 // the symbol may error trying to load a section that does not exist. 2177 bool IsSTAB = false; 2178 if (MachO) { 2179 DataRefImpl SymDRI = Symbol.getRawDataRefImpl(); 2180 uint8_t NType = 2181 (MachO->is64Bit() ? MachO->getSymbol64TableEntry(SymDRI).n_type 2182 : MachO->getSymbolTableEntry(SymDRI).n_type); 2183 if (NType & MachO::N_STAB) 2184 IsSTAB = true; 2185 } 2186 section_iterator Section = IsSTAB 2187 ? O.section_end() 2188 : unwrapOrError(Symbol.getSection(), FileName, 2189 ArchiveName, ArchitectureName); 2190 2191 StringRef Name; 2192 if (Type == SymbolRef::ST_Debug && Section != O.section_end()) { 2193 if (Expected<StringRef> NameOrErr = Section->getName()) 2194 Name = *NameOrErr; 2195 else 2196 consumeError(NameOrErr.takeError()); 2197 2198 } else { 2199 Name = unwrapOrError(Symbol.getName(), FileName, ArchiveName, 2200 ArchitectureName); 2201 } 2202 2203 bool Global = Flags & SymbolRef::SF_Global; 2204 bool Weak = Flags & SymbolRef::SF_Weak; 2205 bool Absolute = Flags & SymbolRef::SF_Absolute; 2206 bool Common = Flags & SymbolRef::SF_Common; 2207 bool Hidden = Flags & SymbolRef::SF_Hidden; 2208 2209 char GlobLoc = ' '; 2210 if ((Section != O.section_end() || Absolute) && !Weak) 2211 GlobLoc = Global ? 'g' : 'l'; 2212 char IFunc = ' '; 2213 if (O.isELF()) { 2214 if (ELFSymbolRef(Symbol).getELFType() == ELF::STT_GNU_IFUNC) 2215 IFunc = 'i'; 2216 if (ELFSymbolRef(Symbol).getBinding() == ELF::STB_GNU_UNIQUE) 2217 GlobLoc = 'u'; 2218 } 2219 2220 char Debug = ' '; 2221 if (DumpDynamic) 2222 Debug = 'D'; 2223 else if (Type == SymbolRef::ST_Debug || Type == SymbolRef::ST_File) 2224 Debug = 'd'; 2225 2226 char FileFunc = ' '; 2227 if (Type == SymbolRef::ST_File) 2228 FileFunc = 'f'; 2229 else if (Type == SymbolRef::ST_Function) 2230 FileFunc = 'F'; 2231 else if (Type == SymbolRef::ST_Data) 2232 FileFunc = 'O'; 2233 2234 const char *Fmt = O.getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 2235 2236 outs() << format(Fmt, Address) << " " 2237 << GlobLoc // Local -> 'l', Global -> 'g', Neither -> ' ' 2238 << (Weak ? 'w' : ' ') // Weak? 2239 << ' ' // Constructor. Not supported yet. 2240 << ' ' // Warning. Not supported yet. 2241 << IFunc // Indirect reference to another symbol. 2242 << Debug // Debugging (d) or dynamic (D) symbol. 2243 << FileFunc // Name of function (F), file (f) or object (O). 2244 << ' '; 2245 if (Absolute) { 2246 outs() << "*ABS*"; 2247 } else if (Common) { 2248 outs() << "*COM*"; 2249 } else if (Section == O.section_end()) { 2250 if (O.isXCOFF()) { 2251 XCOFFSymbolRef XCOFFSym = cast<const XCOFFObjectFile>(O).toSymbolRef( 2252 Symbol.getRawDataRefImpl()); 2253 if (XCOFF::N_DEBUG == XCOFFSym.getSectionNumber()) 2254 outs() << "*DEBUG*"; 2255 else 2256 outs() << "*UND*"; 2257 } else 2258 outs() << "*UND*"; 2259 } else { 2260 StringRef SegmentName = getSegmentName(MachO, *Section); 2261 if (!SegmentName.empty()) 2262 outs() << SegmentName << ","; 2263 StringRef SectionName = unwrapOrError(Section->getName(), FileName); 2264 outs() << SectionName; 2265 if (O.isXCOFF()) { 2266 Optional<SymbolRef> SymRef = 2267 getXCOFFSymbolContainingSymbolRef(cast<XCOFFObjectFile>(O), Symbol); 2268 if (SymRef) { 2269 2270 Expected<StringRef> NameOrErr = SymRef->getName(); 2271 2272 if (NameOrErr) { 2273 outs() << " (csect:"; 2274 std::string SymName(NameOrErr.get()); 2275 2276 if (Demangle) 2277 SymName = demangle(SymName); 2278 2279 if (SymbolDescription) 2280 SymName = getXCOFFSymbolDescription( 2281 createSymbolInfo(O, SymRef.value()), SymName); 2282 2283 outs() << ' ' << SymName; 2284 outs() << ") "; 2285 } else 2286 reportWarning(toString(NameOrErr.takeError()), FileName); 2287 } 2288 } 2289 } 2290 2291 if (Common) 2292 outs() << '\t' << format(Fmt, static_cast<uint64_t>(Symbol.getAlignment())); 2293 else if (O.isXCOFF()) 2294 outs() << '\t' 2295 << format(Fmt, cast<XCOFFObjectFile>(O).getSymbolSize( 2296 Symbol.getRawDataRefImpl())); 2297 else if (O.isELF()) 2298 outs() << '\t' << format(Fmt, ELFSymbolRef(Symbol).getSize()); 2299 2300 if (O.isELF()) { 2301 if (!SymbolVersions.empty()) { 2302 const VersionEntry &Ver = 2303 SymbolVersions[Symbol.getRawDataRefImpl().d.b - 1]; 2304 std::string Str; 2305 if (!Ver.Name.empty()) 2306 Str = Ver.IsVerDef ? ' ' + Ver.Name : '(' + Ver.Name + ')'; 2307 outs() << ' ' << left_justify(Str, 12); 2308 } 2309 2310 uint8_t Other = ELFSymbolRef(Symbol).getOther(); 2311 switch (Other) { 2312 case ELF::STV_DEFAULT: 2313 break; 2314 case ELF::STV_INTERNAL: 2315 outs() << " .internal"; 2316 break; 2317 case ELF::STV_HIDDEN: 2318 outs() << " .hidden"; 2319 break; 2320 case ELF::STV_PROTECTED: 2321 outs() << " .protected"; 2322 break; 2323 default: 2324 outs() << format(" 0x%02x", Other); 2325 break; 2326 } 2327 } else if (Hidden) { 2328 outs() << " .hidden"; 2329 } 2330 2331 std::string SymName(Name); 2332 if (Demangle) 2333 SymName = demangle(SymName); 2334 2335 if (O.isXCOFF() && SymbolDescription) 2336 SymName = getXCOFFSymbolDescription(createSymbolInfo(O, Symbol), SymName); 2337 2338 outs() << ' ' << SymName << '\n'; 2339 } 2340 2341 static void printUnwindInfo(const ObjectFile *O) { 2342 outs() << "Unwind info:\n\n"; 2343 2344 if (const COFFObjectFile *Coff = dyn_cast<COFFObjectFile>(O)) 2345 printCOFFUnwindInfo(Coff); 2346 else if (const MachOObjectFile *MachO = dyn_cast<MachOObjectFile>(O)) 2347 printMachOUnwindInfo(MachO); 2348 else 2349 // TODO: Extract DWARF dump tool to objdump. 2350 WithColor::error(errs(), ToolName) 2351 << "This operation is only currently supported " 2352 "for COFF and MachO object files.\n"; 2353 } 2354 2355 /// Dump the raw contents of the __clangast section so the output can be piped 2356 /// into llvm-bcanalyzer. 2357 static void printRawClangAST(const ObjectFile *Obj) { 2358 if (outs().is_displayed()) { 2359 WithColor::error(errs(), ToolName) 2360 << "The -raw-clang-ast option will dump the raw binary contents of " 2361 "the clang ast section.\n" 2362 "Please redirect the output to a file or another program such as " 2363 "llvm-bcanalyzer.\n"; 2364 return; 2365 } 2366 2367 StringRef ClangASTSectionName("__clangast"); 2368 if (Obj->isCOFF()) { 2369 ClangASTSectionName = "clangast"; 2370 } 2371 2372 Optional<object::SectionRef> ClangASTSection; 2373 for (auto Sec : ToolSectionFilter(*Obj)) { 2374 StringRef Name; 2375 if (Expected<StringRef> NameOrErr = Sec.getName()) 2376 Name = *NameOrErr; 2377 else 2378 consumeError(NameOrErr.takeError()); 2379 2380 if (Name == ClangASTSectionName) { 2381 ClangASTSection = Sec; 2382 break; 2383 } 2384 } 2385 if (!ClangASTSection) 2386 return; 2387 2388 StringRef ClangASTContents = 2389 unwrapOrError(ClangASTSection.value().getContents(), Obj->getFileName()); 2390 outs().write(ClangASTContents.data(), ClangASTContents.size()); 2391 } 2392 2393 static void printFaultMaps(const ObjectFile *Obj) { 2394 StringRef FaultMapSectionName; 2395 2396 if (Obj->isELF()) { 2397 FaultMapSectionName = ".llvm_faultmaps"; 2398 } else if (Obj->isMachO()) { 2399 FaultMapSectionName = "__llvm_faultmaps"; 2400 } else { 2401 WithColor::error(errs(), ToolName) 2402 << "This operation is only currently supported " 2403 "for ELF and Mach-O executable files.\n"; 2404 return; 2405 } 2406 2407 Optional<object::SectionRef> FaultMapSection; 2408 2409 for (auto Sec : ToolSectionFilter(*Obj)) { 2410 StringRef Name; 2411 if (Expected<StringRef> NameOrErr = Sec.getName()) 2412 Name = *NameOrErr; 2413 else 2414 consumeError(NameOrErr.takeError()); 2415 2416 if (Name == FaultMapSectionName) { 2417 FaultMapSection = Sec; 2418 break; 2419 } 2420 } 2421 2422 outs() << "FaultMap table:\n"; 2423 2424 if (!FaultMapSection) { 2425 outs() << "<not found>\n"; 2426 return; 2427 } 2428 2429 StringRef FaultMapContents = 2430 unwrapOrError(FaultMapSection->getContents(), Obj->getFileName()); 2431 FaultMapParser FMP(FaultMapContents.bytes_begin(), 2432 FaultMapContents.bytes_end()); 2433 2434 outs() << FMP; 2435 } 2436 2437 static void printPrivateFileHeaders(const ObjectFile *O, bool OnlyFirst) { 2438 if (O->isELF()) { 2439 printELFFileHeader(O); 2440 printELFDynamicSection(O); 2441 printELFSymbolVersionInfo(O); 2442 return; 2443 } 2444 if (O->isCOFF()) 2445 return printCOFFFileHeader(cast<object::COFFObjectFile>(*O)); 2446 if (O->isWasm()) 2447 return printWasmFileHeader(O); 2448 if (O->isMachO()) { 2449 printMachOFileHeader(O); 2450 if (!OnlyFirst) 2451 printMachOLoadCommands(O); 2452 return; 2453 } 2454 reportError(O->getFileName(), "Invalid/Unsupported object file format"); 2455 } 2456 2457 static void printFileHeaders(const ObjectFile *O) { 2458 if (!O->isELF() && !O->isCOFF()) 2459 reportError(O->getFileName(), "Invalid/Unsupported object file format"); 2460 2461 Triple::ArchType AT = O->getArch(); 2462 outs() << "architecture: " << Triple::getArchTypeName(AT) << "\n"; 2463 uint64_t Address = unwrapOrError(O->getStartAddress(), O->getFileName()); 2464 2465 StringRef Fmt = O->getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 2466 outs() << "start address: " 2467 << "0x" << format(Fmt.data(), Address) << "\n"; 2468 } 2469 2470 static void printArchiveChild(StringRef Filename, const Archive::Child &C) { 2471 Expected<sys::fs::perms> ModeOrErr = C.getAccessMode(); 2472 if (!ModeOrErr) { 2473 WithColor::error(errs(), ToolName) << "ill-formed archive entry.\n"; 2474 consumeError(ModeOrErr.takeError()); 2475 return; 2476 } 2477 sys::fs::perms Mode = ModeOrErr.get(); 2478 outs() << ((Mode & sys::fs::owner_read) ? "r" : "-"); 2479 outs() << ((Mode & sys::fs::owner_write) ? "w" : "-"); 2480 outs() << ((Mode & sys::fs::owner_exe) ? "x" : "-"); 2481 outs() << ((Mode & sys::fs::group_read) ? "r" : "-"); 2482 outs() << ((Mode & sys::fs::group_write) ? "w" : "-"); 2483 outs() << ((Mode & sys::fs::group_exe) ? "x" : "-"); 2484 outs() << ((Mode & sys::fs::others_read) ? "r" : "-"); 2485 outs() << ((Mode & sys::fs::others_write) ? "w" : "-"); 2486 outs() << ((Mode & sys::fs::others_exe) ? "x" : "-"); 2487 2488 outs() << " "; 2489 2490 outs() << format("%d/%d %6" PRId64 " ", unwrapOrError(C.getUID(), Filename), 2491 unwrapOrError(C.getGID(), Filename), 2492 unwrapOrError(C.getRawSize(), Filename)); 2493 2494 StringRef RawLastModified = C.getRawLastModified(); 2495 unsigned Seconds; 2496 if (RawLastModified.getAsInteger(10, Seconds)) 2497 outs() << "(date: \"" << RawLastModified 2498 << "\" contains non-decimal chars) "; 2499 else { 2500 // Since ctime(3) returns a 26 character string of the form: 2501 // "Sun Sep 16 01:03:52 1973\n\0" 2502 // just print 24 characters. 2503 time_t t = Seconds; 2504 outs() << format("%.24s ", ctime(&t)); 2505 } 2506 2507 StringRef Name = ""; 2508 Expected<StringRef> NameOrErr = C.getName(); 2509 if (!NameOrErr) { 2510 consumeError(NameOrErr.takeError()); 2511 Name = unwrapOrError(C.getRawName(), Filename); 2512 } else { 2513 Name = NameOrErr.get(); 2514 } 2515 outs() << Name << "\n"; 2516 } 2517 2518 // For ELF only now. 2519 static bool shouldWarnForInvalidStartStopAddress(ObjectFile *Obj) { 2520 if (const auto *Elf = dyn_cast<ELFObjectFileBase>(Obj)) { 2521 if (Elf->getEType() != ELF::ET_REL) 2522 return true; 2523 } 2524 return false; 2525 } 2526 2527 static void checkForInvalidStartStopAddress(ObjectFile *Obj, 2528 uint64_t Start, uint64_t Stop) { 2529 if (!shouldWarnForInvalidStartStopAddress(Obj)) 2530 return; 2531 2532 for (const SectionRef &Section : Obj->sections()) 2533 if (ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC) { 2534 uint64_t BaseAddr = Section.getAddress(); 2535 uint64_t Size = Section.getSize(); 2536 if ((Start < BaseAddr + Size) && Stop > BaseAddr) 2537 return; 2538 } 2539 2540 if (!HasStartAddressFlag) 2541 reportWarning("no section has address less than 0x" + 2542 Twine::utohexstr(Stop) + " specified by --stop-address", 2543 Obj->getFileName()); 2544 else if (!HasStopAddressFlag) 2545 reportWarning("no section has address greater than or equal to 0x" + 2546 Twine::utohexstr(Start) + " specified by --start-address", 2547 Obj->getFileName()); 2548 else 2549 reportWarning("no section overlaps the range [0x" + 2550 Twine::utohexstr(Start) + ",0x" + Twine::utohexstr(Stop) + 2551 ") specified by --start-address/--stop-address", 2552 Obj->getFileName()); 2553 } 2554 2555 static void dumpObject(ObjectFile *O, const Archive *A = nullptr, 2556 const Archive::Child *C = nullptr) { 2557 // Avoid other output when using a raw option. 2558 if (!RawClangAST) { 2559 outs() << '\n'; 2560 if (A) 2561 outs() << A->getFileName() << "(" << O->getFileName() << ")"; 2562 else 2563 outs() << O->getFileName(); 2564 outs() << ":\tfile format " << O->getFileFormatName().lower() << "\n"; 2565 } 2566 2567 if (HasStartAddressFlag || HasStopAddressFlag) 2568 checkForInvalidStartStopAddress(O, StartAddress, StopAddress); 2569 2570 // Note: the order here matches GNU objdump for compatability. 2571 StringRef ArchiveName = A ? A->getFileName() : ""; 2572 if (ArchiveHeaders && !MachOOpt && C) 2573 printArchiveChild(ArchiveName, *C); 2574 if (FileHeaders) 2575 printFileHeaders(O); 2576 if (PrivateHeaders || FirstPrivateHeader) 2577 printPrivateFileHeaders(O, FirstPrivateHeader); 2578 if (SectionHeaders) 2579 printSectionHeaders(*O); 2580 if (SymbolTable) 2581 printSymbolTable(*O, ArchiveName); 2582 if (DynamicSymbolTable) 2583 printSymbolTable(*O, ArchiveName, /*ArchitectureName=*/"", 2584 /*DumpDynamic=*/true); 2585 if (DwarfDumpType != DIDT_Null) { 2586 std::unique_ptr<DIContext> DICtx = DWARFContext::create(*O); 2587 // Dump the complete DWARF structure. 2588 DIDumpOptions DumpOpts; 2589 DumpOpts.DumpType = DwarfDumpType; 2590 DICtx->dump(outs(), DumpOpts); 2591 } 2592 if (Relocations && !Disassemble) 2593 printRelocations(O); 2594 if (DynamicRelocations) 2595 printDynamicRelocations(O); 2596 if (SectionContents) 2597 printSectionContents(O); 2598 if (Disassemble) 2599 disassembleObject(O, Relocations); 2600 if (UnwindInfo) 2601 printUnwindInfo(O); 2602 2603 // Mach-O specific options: 2604 if (ExportsTrie) 2605 printExportsTrie(O); 2606 if (Rebase) 2607 printRebaseTable(O); 2608 if (Bind) 2609 printBindTable(O); 2610 if (LazyBind) 2611 printLazyBindTable(O); 2612 if (WeakBind) 2613 printWeakBindTable(O); 2614 2615 // Other special sections: 2616 if (RawClangAST) 2617 printRawClangAST(O); 2618 if (FaultMapSection) 2619 printFaultMaps(O); 2620 if (Offloading) 2621 dumpOffloadBinary(*O); 2622 } 2623 2624 static void dumpObject(const COFFImportFile *I, const Archive *A, 2625 const Archive::Child *C = nullptr) { 2626 StringRef ArchiveName = A ? A->getFileName() : ""; 2627 2628 // Avoid other output when using a raw option. 2629 if (!RawClangAST) 2630 outs() << '\n' 2631 << ArchiveName << "(" << I->getFileName() << ")" 2632 << ":\tfile format COFF-import-file" 2633 << "\n\n"; 2634 2635 if (ArchiveHeaders && !MachOOpt && C) 2636 printArchiveChild(ArchiveName, *C); 2637 if (SymbolTable) 2638 printCOFFSymbolTable(*I); 2639 } 2640 2641 /// Dump each object file in \a a; 2642 static void dumpArchive(const Archive *A) { 2643 Error Err = Error::success(); 2644 unsigned I = -1; 2645 for (auto &C : A->children(Err)) { 2646 ++I; 2647 Expected<std::unique_ptr<Binary>> ChildOrErr = C.getAsBinary(); 2648 if (!ChildOrErr) { 2649 if (auto E = isNotObjectErrorInvalidFileType(ChildOrErr.takeError())) 2650 reportError(std::move(E), getFileNameForError(C, I), A->getFileName()); 2651 continue; 2652 } 2653 if (ObjectFile *O = dyn_cast<ObjectFile>(&*ChildOrErr.get())) 2654 dumpObject(O, A, &C); 2655 else if (COFFImportFile *I = dyn_cast<COFFImportFile>(&*ChildOrErr.get())) 2656 dumpObject(I, A, &C); 2657 else 2658 reportError(errorCodeToError(object_error::invalid_file_type), 2659 A->getFileName()); 2660 } 2661 if (Err) 2662 reportError(std::move(Err), A->getFileName()); 2663 } 2664 2665 /// Open file and figure out how to dump it. 2666 static void dumpInput(StringRef file) { 2667 // If we are using the Mach-O specific object file parser, then let it parse 2668 // the file and process the command line options. So the -arch flags can 2669 // be used to select specific slices, etc. 2670 if (MachOOpt) { 2671 parseInputMachO(file); 2672 return; 2673 } 2674 2675 // Attempt to open the binary. 2676 OwningBinary<Binary> OBinary = unwrapOrError(createBinary(file), file); 2677 Binary &Binary = *OBinary.getBinary(); 2678 2679 if (Archive *A = dyn_cast<Archive>(&Binary)) 2680 dumpArchive(A); 2681 else if (ObjectFile *O = dyn_cast<ObjectFile>(&Binary)) 2682 dumpObject(O); 2683 else if (MachOUniversalBinary *UB = dyn_cast<MachOUniversalBinary>(&Binary)) 2684 parseInputMachO(UB); 2685 else if (OffloadBinary *OB = dyn_cast<OffloadBinary>(&Binary)) 2686 dumpOffloadSections(*OB); 2687 else 2688 reportError(errorCodeToError(object_error::invalid_file_type), file); 2689 } 2690 2691 template <typename T> 2692 static void parseIntArg(const llvm::opt::InputArgList &InputArgs, int ID, 2693 T &Value) { 2694 if (const opt::Arg *A = InputArgs.getLastArg(ID)) { 2695 StringRef V(A->getValue()); 2696 if (!llvm::to_integer(V, Value, 0)) { 2697 reportCmdLineError(A->getSpelling() + 2698 ": expected a non-negative integer, but got '" + V + 2699 "'"); 2700 } 2701 } 2702 } 2703 2704 static void invalidArgValue(const opt::Arg *A) { 2705 reportCmdLineError("'" + StringRef(A->getValue()) + 2706 "' is not a valid value for '" + A->getSpelling() + "'"); 2707 } 2708 2709 static std::vector<std::string> 2710 commaSeparatedValues(const llvm::opt::InputArgList &InputArgs, int ID) { 2711 std::vector<std::string> Values; 2712 for (StringRef Value : InputArgs.getAllArgValues(ID)) { 2713 llvm::SmallVector<StringRef, 2> SplitValues; 2714 llvm::SplitString(Value, SplitValues, ","); 2715 for (StringRef SplitValue : SplitValues) 2716 Values.push_back(SplitValue.str()); 2717 } 2718 return Values; 2719 } 2720 2721 static void parseOtoolOptions(const llvm::opt::InputArgList &InputArgs) { 2722 MachOOpt = true; 2723 FullLeadingAddr = true; 2724 PrintImmHex = true; 2725 2726 ArchName = InputArgs.getLastArgValue(OTOOL_arch).str(); 2727 LinkOptHints = InputArgs.hasArg(OTOOL_C); 2728 if (InputArgs.hasArg(OTOOL_d)) 2729 FilterSections.push_back("__DATA,__data"); 2730 DylibId = InputArgs.hasArg(OTOOL_D); 2731 UniversalHeaders = InputArgs.hasArg(OTOOL_f); 2732 DataInCode = InputArgs.hasArg(OTOOL_G); 2733 FirstPrivateHeader = InputArgs.hasArg(OTOOL_h); 2734 IndirectSymbols = InputArgs.hasArg(OTOOL_I); 2735 ShowRawInsn = InputArgs.hasArg(OTOOL_j); 2736 PrivateHeaders = InputArgs.hasArg(OTOOL_l); 2737 DylibsUsed = InputArgs.hasArg(OTOOL_L); 2738 MCPU = InputArgs.getLastArgValue(OTOOL_mcpu_EQ).str(); 2739 ObjcMetaData = InputArgs.hasArg(OTOOL_o); 2740 DisSymName = InputArgs.getLastArgValue(OTOOL_p).str(); 2741 InfoPlist = InputArgs.hasArg(OTOOL_P); 2742 Relocations = InputArgs.hasArg(OTOOL_r); 2743 if (const Arg *A = InputArgs.getLastArg(OTOOL_s)) { 2744 auto Filter = (A->getValue(0) + StringRef(",") + A->getValue(1)).str(); 2745 FilterSections.push_back(Filter); 2746 } 2747 if (InputArgs.hasArg(OTOOL_t)) 2748 FilterSections.push_back("__TEXT,__text"); 2749 Verbose = InputArgs.hasArg(OTOOL_v) || InputArgs.hasArg(OTOOL_V) || 2750 InputArgs.hasArg(OTOOL_o); 2751 SymbolicOperands = InputArgs.hasArg(OTOOL_V); 2752 if (InputArgs.hasArg(OTOOL_x)) 2753 FilterSections.push_back(",__text"); 2754 LeadingAddr = LeadingHeaders = !InputArgs.hasArg(OTOOL_X); 2755 2756 InputFilenames = InputArgs.getAllArgValues(OTOOL_INPUT); 2757 if (InputFilenames.empty()) 2758 reportCmdLineError("no input file"); 2759 2760 for (const Arg *A : InputArgs) { 2761 const Option &O = A->getOption(); 2762 if (O.getGroup().isValid() && O.getGroup().getID() == OTOOL_grp_obsolete) { 2763 reportCmdLineWarning(O.getPrefixedName() + 2764 " is obsolete and not implemented"); 2765 } 2766 } 2767 } 2768 2769 static void parseObjdumpOptions(const llvm::opt::InputArgList &InputArgs) { 2770 parseIntArg(InputArgs, OBJDUMP_adjust_vma_EQ, AdjustVMA); 2771 AllHeaders = InputArgs.hasArg(OBJDUMP_all_headers); 2772 ArchName = InputArgs.getLastArgValue(OBJDUMP_arch_name_EQ).str(); 2773 ArchiveHeaders = InputArgs.hasArg(OBJDUMP_archive_headers); 2774 Demangle = InputArgs.hasArg(OBJDUMP_demangle); 2775 Disassemble = InputArgs.hasArg(OBJDUMP_disassemble); 2776 DisassembleAll = InputArgs.hasArg(OBJDUMP_disassemble_all); 2777 SymbolDescription = InputArgs.hasArg(OBJDUMP_symbol_description); 2778 DisassembleSymbols = 2779 commaSeparatedValues(InputArgs, OBJDUMP_disassemble_symbols_EQ); 2780 DisassembleZeroes = InputArgs.hasArg(OBJDUMP_disassemble_zeroes); 2781 if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_dwarf_EQ)) { 2782 DwarfDumpType = StringSwitch<DIDumpType>(A->getValue()) 2783 .Case("frames", DIDT_DebugFrame) 2784 .Default(DIDT_Null); 2785 if (DwarfDumpType == DIDT_Null) 2786 invalidArgValue(A); 2787 } 2788 DynamicRelocations = InputArgs.hasArg(OBJDUMP_dynamic_reloc); 2789 FaultMapSection = InputArgs.hasArg(OBJDUMP_fault_map_section); 2790 Offloading = InputArgs.hasArg(OBJDUMP_offloading); 2791 FileHeaders = InputArgs.hasArg(OBJDUMP_file_headers); 2792 SectionContents = InputArgs.hasArg(OBJDUMP_full_contents); 2793 PrintLines = InputArgs.hasArg(OBJDUMP_line_numbers); 2794 InputFilenames = InputArgs.getAllArgValues(OBJDUMP_INPUT); 2795 MachOOpt = InputArgs.hasArg(OBJDUMP_macho); 2796 MCPU = InputArgs.getLastArgValue(OBJDUMP_mcpu_EQ).str(); 2797 MAttrs = commaSeparatedValues(InputArgs, OBJDUMP_mattr_EQ); 2798 ShowRawInsn = !InputArgs.hasArg(OBJDUMP_no_show_raw_insn); 2799 LeadingAddr = !InputArgs.hasArg(OBJDUMP_no_leading_addr); 2800 RawClangAST = InputArgs.hasArg(OBJDUMP_raw_clang_ast); 2801 Relocations = InputArgs.hasArg(OBJDUMP_reloc); 2802 PrintImmHex = 2803 InputArgs.hasFlag(OBJDUMP_print_imm_hex, OBJDUMP_no_print_imm_hex, false); 2804 PrivateHeaders = InputArgs.hasArg(OBJDUMP_private_headers); 2805 FilterSections = InputArgs.getAllArgValues(OBJDUMP_section_EQ); 2806 SectionHeaders = InputArgs.hasArg(OBJDUMP_section_headers); 2807 ShowLMA = InputArgs.hasArg(OBJDUMP_show_lma); 2808 PrintSource = InputArgs.hasArg(OBJDUMP_source); 2809 parseIntArg(InputArgs, OBJDUMP_start_address_EQ, StartAddress); 2810 HasStartAddressFlag = InputArgs.hasArg(OBJDUMP_start_address_EQ); 2811 parseIntArg(InputArgs, OBJDUMP_stop_address_EQ, StopAddress); 2812 HasStopAddressFlag = InputArgs.hasArg(OBJDUMP_stop_address_EQ); 2813 SymbolTable = InputArgs.hasArg(OBJDUMP_syms); 2814 SymbolizeOperands = InputArgs.hasArg(OBJDUMP_symbolize_operands); 2815 DynamicSymbolTable = InputArgs.hasArg(OBJDUMP_dynamic_syms); 2816 TripleName = InputArgs.getLastArgValue(OBJDUMP_triple_EQ).str(); 2817 UnwindInfo = InputArgs.hasArg(OBJDUMP_unwind_info); 2818 Wide = InputArgs.hasArg(OBJDUMP_wide); 2819 Prefix = InputArgs.getLastArgValue(OBJDUMP_prefix).str(); 2820 parseIntArg(InputArgs, OBJDUMP_prefix_strip, PrefixStrip); 2821 if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_debug_vars_EQ)) { 2822 DbgVariables = StringSwitch<DebugVarsFormat>(A->getValue()) 2823 .Case("ascii", DVASCII) 2824 .Case("unicode", DVUnicode) 2825 .Default(DVInvalid); 2826 if (DbgVariables == DVInvalid) 2827 invalidArgValue(A); 2828 } 2829 parseIntArg(InputArgs, OBJDUMP_debug_vars_indent_EQ, DbgIndent); 2830 2831 parseMachOOptions(InputArgs); 2832 2833 // Parse -M (--disassembler-options) and deprecated 2834 // --x86-asm-syntax={att,intel}. 2835 // 2836 // Note, for x86, the asm dialect (AssemblerDialect) is initialized when the 2837 // MCAsmInfo is constructed. MCInstPrinter::applyTargetSpecificCLOption is 2838 // called too late. For now we have to use the internal cl::opt option. 2839 const char *AsmSyntax = nullptr; 2840 for (const auto *A : InputArgs.filtered(OBJDUMP_disassembler_options_EQ, 2841 OBJDUMP_x86_asm_syntax_att, 2842 OBJDUMP_x86_asm_syntax_intel)) { 2843 switch (A->getOption().getID()) { 2844 case OBJDUMP_x86_asm_syntax_att: 2845 AsmSyntax = "--x86-asm-syntax=att"; 2846 continue; 2847 case OBJDUMP_x86_asm_syntax_intel: 2848 AsmSyntax = "--x86-asm-syntax=intel"; 2849 continue; 2850 } 2851 2852 SmallVector<StringRef, 2> Values; 2853 llvm::SplitString(A->getValue(), Values, ","); 2854 for (StringRef V : Values) { 2855 if (V == "att") 2856 AsmSyntax = "--x86-asm-syntax=att"; 2857 else if (V == "intel") 2858 AsmSyntax = "--x86-asm-syntax=intel"; 2859 else 2860 DisassemblerOptions.push_back(V.str()); 2861 } 2862 } 2863 if (AsmSyntax) { 2864 const char *Argv[] = {"llvm-objdump", AsmSyntax}; 2865 llvm::cl::ParseCommandLineOptions(2, Argv); 2866 } 2867 2868 // objdump defaults to a.out if no filenames specified. 2869 if (InputFilenames.empty()) 2870 InputFilenames.push_back("a.out"); 2871 } 2872 2873 int main(int argc, char **argv) { 2874 using namespace llvm; 2875 InitLLVM X(argc, argv); 2876 2877 ToolName = argv[0]; 2878 std::unique_ptr<CommonOptTable> T; 2879 OptSpecifier Unknown, HelpFlag, HelpHiddenFlag, VersionFlag; 2880 2881 StringRef Stem = sys::path::stem(ToolName); 2882 auto Is = [=](StringRef Tool) { 2883 // We need to recognize the following filenames: 2884 // 2885 // llvm-objdump -> objdump 2886 // llvm-otool-10.exe -> otool 2887 // powerpc64-unknown-freebsd13-objdump -> objdump 2888 auto I = Stem.rfind_insensitive(Tool); 2889 return I != StringRef::npos && 2890 (I + Tool.size() == Stem.size() || !isAlnum(Stem[I + Tool.size()])); 2891 }; 2892 if (Is("otool")) { 2893 T = std::make_unique<OtoolOptTable>(); 2894 Unknown = OTOOL_UNKNOWN; 2895 HelpFlag = OTOOL_help; 2896 HelpHiddenFlag = OTOOL_help_hidden; 2897 VersionFlag = OTOOL_version; 2898 } else { 2899 T = std::make_unique<ObjdumpOptTable>(); 2900 Unknown = OBJDUMP_UNKNOWN; 2901 HelpFlag = OBJDUMP_help; 2902 HelpHiddenFlag = OBJDUMP_help_hidden; 2903 VersionFlag = OBJDUMP_version; 2904 } 2905 2906 BumpPtrAllocator A; 2907 StringSaver Saver(A); 2908 opt::InputArgList InputArgs = 2909 T->parseArgs(argc, argv, Unknown, Saver, 2910 [&](StringRef Msg) { reportCmdLineError(Msg); }); 2911 2912 if (InputArgs.size() == 0 || InputArgs.hasArg(HelpFlag)) { 2913 T->printHelp(ToolName); 2914 return 0; 2915 } 2916 if (InputArgs.hasArg(HelpHiddenFlag)) { 2917 T->printHelp(ToolName, /*ShowHidden=*/true); 2918 return 0; 2919 } 2920 2921 // Initialize targets and assembly printers/parsers. 2922 InitializeAllTargetInfos(); 2923 InitializeAllTargetMCs(); 2924 InitializeAllDisassemblers(); 2925 2926 if (InputArgs.hasArg(VersionFlag)) { 2927 cl::PrintVersionMessage(); 2928 if (!Is("otool")) { 2929 outs() << '\n'; 2930 TargetRegistry::printRegisteredTargetsForVersion(outs()); 2931 } 2932 return 0; 2933 } 2934 2935 if (Is("otool")) 2936 parseOtoolOptions(InputArgs); 2937 else 2938 parseObjdumpOptions(InputArgs); 2939 2940 if (StartAddress >= StopAddress) 2941 reportCmdLineError("start address should be less than stop address"); 2942 2943 // Removes trailing separators from prefix. 2944 while (!Prefix.empty() && sys::path::is_separator(Prefix.back())) 2945 Prefix.pop_back(); 2946 2947 if (AllHeaders) 2948 ArchiveHeaders = FileHeaders = PrivateHeaders = Relocations = 2949 SectionHeaders = SymbolTable = true; 2950 2951 if (DisassembleAll || PrintSource || PrintLines || 2952 !DisassembleSymbols.empty()) 2953 Disassemble = true; 2954 2955 if (!ArchiveHeaders && !Disassemble && DwarfDumpType == DIDT_Null && 2956 !DynamicRelocations && !FileHeaders && !PrivateHeaders && !RawClangAST && 2957 !Relocations && !SectionHeaders && !SectionContents && !SymbolTable && 2958 !DynamicSymbolTable && !UnwindInfo && !FaultMapSection && !Offloading && 2959 !(MachOOpt && (Bind || DataInCode || DyldInfo || DylibId || DylibsUsed || 2960 ExportsTrie || FirstPrivateHeader || FunctionStarts || 2961 IndirectSymbols || InfoPlist || LazyBind || LinkOptHints || 2962 ObjcMetaData || Rebase || Rpaths || UniversalHeaders || 2963 WeakBind || !FilterSections.empty()))) { 2964 T->printHelp(ToolName); 2965 return 2; 2966 } 2967 2968 DisasmSymbolSet.insert(DisassembleSymbols.begin(), DisassembleSymbols.end()); 2969 2970 llvm::for_each(InputFilenames, dumpInput); 2971 2972 warnOnNoMatchForSections(); 2973 2974 return EXIT_SUCCESS; 2975 } 2976