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 "SourcePrinter.h" 24 #include "WasmDump.h" 25 #include "XCOFFDump.h" 26 #include "llvm/ADT/IndexedMap.h" 27 #include "llvm/ADT/Optional.h" 28 #include "llvm/ADT/STLExtras.h" 29 #include "llvm/ADT/SetOperations.h" 30 #include "llvm/ADT/SmallSet.h" 31 #include "llvm/ADT/StringExtras.h" 32 #include "llvm/ADT/StringSet.h" 33 #include "llvm/ADT/Triple.h" 34 #include "llvm/ADT/Twine.h" 35 #include "llvm/DebugInfo/DWARF/DWARFContext.h" 36 #include "llvm/DebugInfo/Symbolize/Symbolize.h" 37 #include "llvm/Demangle/Demangle.h" 38 #include "llvm/MC/MCAsmInfo.h" 39 #include "llvm/MC/MCContext.h" 40 #include "llvm/MC/MCDisassembler/MCDisassembler.h" 41 #include "llvm/MC/MCDisassembler/MCRelocationInfo.h" 42 #include "llvm/MC/MCInst.h" 43 #include "llvm/MC/MCInstPrinter.h" 44 #include "llvm/MC/MCInstrAnalysis.h" 45 #include "llvm/MC/MCInstrInfo.h" 46 #include "llvm/MC/MCObjectFileInfo.h" 47 #include "llvm/MC/MCRegisterInfo.h" 48 #include "llvm/MC/MCSubtargetInfo.h" 49 #include "llvm/MC/MCTargetOptions.h" 50 #include "llvm/Object/Archive.h" 51 #include "llvm/Object/COFF.h" 52 #include "llvm/Object/COFFImportFile.h" 53 #include "llvm/Object/ELFObjectFile.h" 54 #include "llvm/Object/FaultMapParser.h" 55 #include "llvm/Object/MachO.h" 56 #include "llvm/Object/MachOUniversal.h" 57 #include "llvm/Object/ObjectFile.h" 58 #include "llvm/Object/Wasm.h" 59 #include "llvm/Option/Arg.h" 60 #include "llvm/Option/ArgList.h" 61 #include "llvm/Option/Option.h" 62 #include "llvm/Support/Casting.h" 63 #include "llvm/Support/Debug.h" 64 #include "llvm/Support/Errc.h" 65 #include "llvm/Support/FileSystem.h" 66 #include "llvm/Support/Format.h" 67 #include "llvm/Support/FormatVariadic.h" 68 #include "llvm/Support/GraphWriter.h" 69 #include "llvm/Support/Host.h" 70 #include "llvm/Support/InitLLVM.h" 71 #include "llvm/Support/MemoryBuffer.h" 72 #include "llvm/Support/SourceMgr.h" 73 #include "llvm/Support/StringSaver.h" 74 #include "llvm/Support/TargetRegistry.h" 75 #include "llvm/Support/TargetSelect.h" 76 #include "llvm/Support/WithColor.h" 77 #include "llvm/Support/raw_ostream.h" 78 #include <algorithm> 79 #include <cctype> 80 #include <cstring> 81 #include <system_error> 82 #include <unordered_map> 83 #include <utility> 84 85 using namespace llvm; 86 using namespace llvm::object; 87 using namespace llvm::objdump; 88 using namespace llvm::opt; 89 90 namespace { 91 92 class CommonOptTable : public opt::OptTable { 93 public: 94 CommonOptTable(ArrayRef<Info> OptionInfos, const char *Usage, 95 const char *Description) 96 : OptTable(OptionInfos), Usage(Usage), Description(Description) { 97 setGroupedShortOptions(true); 98 } 99 100 void printHelp(StringRef Argv0, bool ShowHidden = false) const { 101 Argv0 = sys::path::filename(Argv0); 102 opt::OptTable::printHelp(outs(), (Argv0 + Usage).str().c_str(), Description, 103 ShowHidden, ShowHidden); 104 // TODO Replace this with OptTable API once it adds extrahelp support. 105 outs() << "\nPass @FILE as argument to read options from FILE.\n"; 106 } 107 108 private: 109 const char *Usage; 110 const char *Description; 111 }; 112 113 // ObjdumpOptID is in ObjdumpOptID.h 114 115 #define PREFIX(NAME, VALUE) const char *const OBJDUMP_##NAME[] = VALUE; 116 #include "ObjdumpOpts.inc" 117 #undef PREFIX 118 119 static constexpr opt::OptTable::Info ObjdumpInfoTable[] = { 120 #define OBJDUMP_nullptr nullptr 121 #define OPTION(PREFIX, NAME, ID, KIND, GROUP, ALIAS, ALIASARGS, FLAGS, PARAM, \ 122 HELPTEXT, METAVAR, VALUES) \ 123 {OBJDUMP_##PREFIX, NAME, HELPTEXT, \ 124 METAVAR, OBJDUMP_##ID, opt::Option::KIND##Class, \ 125 PARAM, FLAGS, OBJDUMP_##GROUP, \ 126 OBJDUMP_##ALIAS, ALIASARGS, VALUES}, 127 #include "ObjdumpOpts.inc" 128 #undef OPTION 129 #undef OBJDUMP_nullptr 130 }; 131 132 class ObjdumpOptTable : public CommonOptTable { 133 public: 134 ObjdumpOptTable() 135 : CommonOptTable(ObjdumpInfoTable, " [options] <input object files>", 136 "llvm object file dumper") {} 137 }; 138 139 enum OtoolOptID { 140 OTOOL_INVALID = 0, // This is not an option ID. 141 #define OPTION(PREFIX, NAME, ID, KIND, GROUP, ALIAS, ALIASARGS, FLAGS, PARAM, \ 142 HELPTEXT, METAVAR, VALUES) \ 143 OTOOL_##ID, 144 #include "OtoolOpts.inc" 145 #undef OPTION 146 }; 147 148 #define PREFIX(NAME, VALUE) const char *const OTOOL_##NAME[] = VALUE; 149 #include "OtoolOpts.inc" 150 #undef PREFIX 151 152 static constexpr opt::OptTable::Info OtoolInfoTable[] = { 153 #define OTOOL_nullptr nullptr 154 #define OPTION(PREFIX, NAME, ID, KIND, GROUP, ALIAS, ALIASARGS, FLAGS, PARAM, \ 155 HELPTEXT, METAVAR, VALUES) \ 156 {OTOOL_##PREFIX, NAME, HELPTEXT, \ 157 METAVAR, OTOOL_##ID, opt::Option::KIND##Class, \ 158 PARAM, FLAGS, OTOOL_##GROUP, \ 159 OTOOL_##ALIAS, ALIASARGS, VALUES}, 160 #include "OtoolOpts.inc" 161 #undef OPTION 162 #undef OTOOL_nullptr 163 }; 164 165 class OtoolOptTable : public CommonOptTable { 166 public: 167 OtoolOptTable() 168 : CommonOptTable(OtoolInfoTable, " [option...] [file...]", 169 "Mach-O object file displaying tool") {} 170 }; 171 172 } // namespace 173 174 #define DEBUG_TYPE "objdump" 175 176 static uint64_t AdjustVMA; 177 static bool AllHeaders; 178 static std::string ArchName; 179 bool objdump::ArchiveHeaders; 180 bool objdump::Demangle; 181 bool objdump::Disassemble; 182 bool objdump::DisassembleAll; 183 bool objdump::SymbolDescription; 184 static std::vector<std::string> DisassembleSymbols; 185 static bool DisassembleZeroes; 186 static std::vector<std::string> DisassemblerOptions; 187 DIDumpType objdump::DwarfDumpType; 188 static bool DynamicRelocations; 189 static bool FaultMapSection; 190 static bool FileHeaders; 191 bool objdump::SectionContents; 192 static std::vector<std::string> InputFilenames; 193 bool objdump::PrintLines; 194 static bool MachOOpt; 195 std::string objdump::MCPU; 196 std::vector<std::string> objdump::MAttrs; 197 bool objdump::ShowRawInsn; 198 bool objdump::LeadingAddr; 199 static bool RawClangAST; 200 bool objdump::Relocations; 201 bool objdump::PrintImmHex; 202 bool objdump::PrivateHeaders; 203 std::vector<std::string> objdump::FilterSections; 204 bool objdump::SectionHeaders; 205 static bool ShowLMA; 206 bool objdump::PrintSource; 207 208 static uint64_t StartAddress; 209 static bool HasStartAddressFlag; 210 static uint64_t StopAddress = UINT64_MAX; 211 static bool HasStopAddressFlag; 212 213 bool objdump::SymbolTable; 214 static bool SymbolizeOperands; 215 static bool DynamicSymbolTable; 216 std::string objdump::TripleName; 217 bool objdump::UnwindInfo; 218 static bool Wide; 219 std::string objdump::Prefix; 220 uint32_t objdump::PrefixStrip; 221 222 DebugVarsFormat objdump::DbgVariables = DVDisabled; 223 224 int objdump::DbgIndent = 52; 225 226 static StringSet<> DisasmSymbolSet; 227 StringSet<> objdump::FoundSectionSet; 228 static StringRef ToolName; 229 230 namespace { 231 struct FilterResult { 232 // True if the section should not be skipped. 233 bool Keep; 234 235 // True if the index counter should be incremented, even if the section should 236 // be skipped. For example, sections may be skipped if they are not included 237 // in the --section flag, but we still want those to count toward the section 238 // count. 239 bool IncrementIndex; 240 }; 241 } // namespace 242 243 static FilterResult checkSectionFilter(object::SectionRef S) { 244 if (FilterSections.empty()) 245 return {/*Keep=*/true, /*IncrementIndex=*/true}; 246 247 Expected<StringRef> SecNameOrErr = S.getName(); 248 if (!SecNameOrErr) { 249 consumeError(SecNameOrErr.takeError()); 250 return {/*Keep=*/false, /*IncrementIndex=*/false}; 251 } 252 StringRef SecName = *SecNameOrErr; 253 254 // StringSet does not allow empty key so avoid adding sections with 255 // no name (such as the section with index 0) here. 256 if (!SecName.empty()) 257 FoundSectionSet.insert(SecName); 258 259 // Only show the section if it's in the FilterSections list, but always 260 // increment so the indexing is stable. 261 return {/*Keep=*/is_contained(FilterSections, SecName), 262 /*IncrementIndex=*/true}; 263 } 264 265 SectionFilter objdump::ToolSectionFilter(object::ObjectFile const &O, 266 uint64_t *Idx) { 267 // Start at UINT64_MAX so that the first index returned after an increment is 268 // zero (after the unsigned wrap). 269 if (Idx) 270 *Idx = UINT64_MAX; 271 return SectionFilter( 272 [Idx](object::SectionRef S) { 273 FilterResult Result = checkSectionFilter(S); 274 if (Idx != nullptr && Result.IncrementIndex) 275 *Idx += 1; 276 return Result.Keep; 277 }, 278 O); 279 } 280 281 std::string objdump::getFileNameForError(const object::Archive::Child &C, 282 unsigned Index) { 283 Expected<StringRef> NameOrErr = C.getName(); 284 if (NameOrErr) 285 return std::string(NameOrErr.get()); 286 // If we have an error getting the name then we print the index of the archive 287 // member. Since we are already in an error state, we just ignore this error. 288 consumeError(NameOrErr.takeError()); 289 return "<file index: " + std::to_string(Index) + ">"; 290 } 291 292 void objdump::reportWarning(const Twine &Message, StringRef File) { 293 // Output order between errs() and outs() matters especially for archive 294 // files where the output is per member object. 295 outs().flush(); 296 WithColor::warning(errs(), ToolName) 297 << "'" << File << "': " << Message << "\n"; 298 } 299 300 LLVM_ATTRIBUTE_NORETURN void objdump::reportError(StringRef File, 301 const Twine &Message) { 302 outs().flush(); 303 WithColor::error(errs(), ToolName) << "'" << File << "': " << Message << "\n"; 304 exit(1); 305 } 306 307 LLVM_ATTRIBUTE_NORETURN void objdump::reportError(Error E, StringRef FileName, 308 StringRef ArchiveName, 309 StringRef ArchitectureName) { 310 assert(E); 311 outs().flush(); 312 WithColor::error(errs(), ToolName); 313 if (ArchiveName != "") 314 errs() << ArchiveName << "(" << FileName << ")"; 315 else 316 errs() << "'" << FileName << "'"; 317 if (!ArchitectureName.empty()) 318 errs() << " (for architecture " << ArchitectureName << ")"; 319 errs() << ": "; 320 logAllUnhandledErrors(std::move(E), errs()); 321 exit(1); 322 } 323 324 static void reportCmdLineWarning(const Twine &Message) { 325 WithColor::warning(errs(), ToolName) << Message << "\n"; 326 } 327 328 LLVM_ATTRIBUTE_NORETURN static void reportCmdLineError(const Twine &Message) { 329 WithColor::error(errs(), ToolName) << Message << "\n"; 330 exit(1); 331 } 332 333 static void warnOnNoMatchForSections() { 334 SetVector<StringRef> MissingSections; 335 for (StringRef S : FilterSections) { 336 if (FoundSectionSet.count(S)) 337 return; 338 // User may specify a unnamed section. Don't warn for it. 339 if (!S.empty()) 340 MissingSections.insert(S); 341 } 342 343 // Warn only if no section in FilterSections is matched. 344 for (StringRef S : MissingSections) 345 reportCmdLineWarning("section '" + S + 346 "' mentioned in a -j/--section option, but not " 347 "found in any input file"); 348 } 349 350 static const Target *getTarget(const ObjectFile *Obj) { 351 // Figure out the target triple. 352 Triple TheTriple("unknown-unknown-unknown"); 353 if (TripleName.empty()) { 354 TheTriple = Obj->makeTriple(); 355 } else { 356 TheTriple.setTriple(Triple::normalize(TripleName)); 357 auto Arch = Obj->getArch(); 358 if (Arch == Triple::arm || Arch == Triple::armeb) 359 Obj->setARMSubArch(TheTriple); 360 } 361 362 // Get the target specific parser. 363 std::string Error; 364 const Target *TheTarget = TargetRegistry::lookupTarget(ArchName, TheTriple, 365 Error); 366 if (!TheTarget) 367 reportError(Obj->getFileName(), "can't find target: " + Error); 368 369 // Update the triple name and return the found target. 370 TripleName = TheTriple.getTriple(); 371 return TheTarget; 372 } 373 374 bool objdump::isRelocAddressLess(RelocationRef A, RelocationRef B) { 375 return A.getOffset() < B.getOffset(); 376 } 377 378 static Error getRelocationValueString(const RelocationRef &Rel, 379 SmallVectorImpl<char> &Result) { 380 const ObjectFile *Obj = Rel.getObject(); 381 if (auto *ELF = dyn_cast<ELFObjectFileBase>(Obj)) 382 return getELFRelocationValueString(ELF, Rel, Result); 383 if (auto *COFF = dyn_cast<COFFObjectFile>(Obj)) 384 return getCOFFRelocationValueString(COFF, Rel, Result); 385 if (auto *Wasm = dyn_cast<WasmObjectFile>(Obj)) 386 return getWasmRelocationValueString(Wasm, Rel, Result); 387 if (auto *MachO = dyn_cast<MachOObjectFile>(Obj)) 388 return getMachORelocationValueString(MachO, Rel, Result); 389 if (auto *XCOFF = dyn_cast<XCOFFObjectFile>(Obj)) 390 return getXCOFFRelocationValueString(XCOFF, Rel, Result); 391 llvm_unreachable("unknown object file format"); 392 } 393 394 /// Indicates whether this relocation should hidden when listing 395 /// relocations, usually because it is the trailing part of a multipart 396 /// relocation that will be printed as part of the leading relocation. 397 static bool getHidden(RelocationRef RelRef) { 398 auto *MachO = dyn_cast<MachOObjectFile>(RelRef.getObject()); 399 if (!MachO) 400 return false; 401 402 unsigned Arch = MachO->getArch(); 403 DataRefImpl Rel = RelRef.getRawDataRefImpl(); 404 uint64_t Type = MachO->getRelocationType(Rel); 405 406 // On arches that use the generic relocations, GENERIC_RELOC_PAIR 407 // is always hidden. 408 if (Arch == Triple::x86 || Arch == Triple::arm || Arch == Triple::ppc) 409 return Type == MachO::GENERIC_RELOC_PAIR; 410 411 if (Arch == Triple::x86_64) { 412 // On x86_64, X86_64_RELOC_UNSIGNED is hidden only when it follows 413 // an X86_64_RELOC_SUBTRACTOR. 414 if (Type == MachO::X86_64_RELOC_UNSIGNED && Rel.d.a > 0) { 415 DataRefImpl RelPrev = Rel; 416 RelPrev.d.a--; 417 uint64_t PrevType = MachO->getRelocationType(RelPrev); 418 if (PrevType == MachO::X86_64_RELOC_SUBTRACTOR) 419 return true; 420 } 421 } 422 423 return false; 424 } 425 426 namespace { 427 428 /// Get the column at which we want to start printing the instruction 429 /// disassembly, taking into account anything which appears to the left of it. 430 unsigned getInstStartColumn(const MCSubtargetInfo &STI) { 431 return !ShowRawInsn ? 16 : STI.getTargetTriple().isX86() ? 40 : 24; 432 } 433 434 static bool isAArch64Elf(const ObjectFile *Obj) { 435 const auto *Elf = dyn_cast<ELFObjectFileBase>(Obj); 436 return Elf && Elf->getEMachine() == ELF::EM_AARCH64; 437 } 438 439 static bool isArmElf(const ObjectFile *Obj) { 440 const auto *Elf = dyn_cast<ELFObjectFileBase>(Obj); 441 return Elf && Elf->getEMachine() == ELF::EM_ARM; 442 } 443 444 static bool hasMappingSymbols(const ObjectFile *Obj) { 445 return isArmElf(Obj) || isAArch64Elf(Obj); 446 } 447 448 static void printRelocation(formatted_raw_ostream &OS, StringRef FileName, 449 const RelocationRef &Rel, uint64_t Address, 450 bool Is64Bits) { 451 StringRef Fmt = Is64Bits ? "\t\t%016" PRIx64 ": " : "\t\t\t%08" PRIx64 ": "; 452 SmallString<16> Name; 453 SmallString<32> Val; 454 Rel.getTypeName(Name); 455 if (Error E = getRelocationValueString(Rel, Val)) 456 reportError(std::move(E), FileName); 457 OS << format(Fmt.data(), Address) << Name << "\t" << Val; 458 } 459 460 class PrettyPrinter { 461 public: 462 virtual ~PrettyPrinter() = default; 463 virtual void 464 printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 465 object::SectionedAddress Address, formatted_raw_ostream &OS, 466 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 467 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 468 LiveVariablePrinter &LVP) { 469 if (SP && (PrintSource || PrintLines)) 470 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 471 LVP.printBetweenInsts(OS, false); 472 473 size_t Start = OS.tell(); 474 if (LeadingAddr) 475 OS << format("%8" PRIx64 ":", Address.Address); 476 if (ShowRawInsn) { 477 OS << ' '; 478 dumpBytes(Bytes, OS); 479 } 480 481 // The output of printInst starts with a tab. Print some spaces so that 482 // the tab has 1 column and advances to the target tab stop. 483 unsigned TabStop = getInstStartColumn(STI); 484 unsigned Column = OS.tell() - Start; 485 OS.indent(Column < TabStop - 1 ? TabStop - 1 - Column : 7 - Column % 8); 486 487 if (MI) { 488 // See MCInstPrinter::printInst. On targets where a PC relative immediate 489 // is relative to the next instruction and the length of a MCInst is 490 // difficult to measure (x86), this is the address of the next 491 // instruction. 492 uint64_t Addr = 493 Address.Address + (STI.getTargetTriple().isX86() ? Bytes.size() : 0); 494 IP.printInst(MI, Addr, "", STI, OS); 495 } else 496 OS << "\t<unknown>"; 497 } 498 }; 499 PrettyPrinter PrettyPrinterInst; 500 501 class HexagonPrettyPrinter : public PrettyPrinter { 502 public: 503 void printLead(ArrayRef<uint8_t> Bytes, uint64_t Address, 504 formatted_raw_ostream &OS) { 505 uint32_t opcode = 506 (Bytes[3] << 24) | (Bytes[2] << 16) | (Bytes[1] << 8) | Bytes[0]; 507 if (LeadingAddr) 508 OS << format("%8" PRIx64 ":", Address); 509 if (ShowRawInsn) { 510 OS << "\t"; 511 dumpBytes(Bytes.slice(0, 4), OS); 512 OS << format("\t%08" PRIx32, opcode); 513 } 514 } 515 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 516 object::SectionedAddress Address, formatted_raw_ostream &OS, 517 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 518 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 519 LiveVariablePrinter &LVP) override { 520 if (SP && (PrintSource || PrintLines)) 521 SP->printSourceLine(OS, Address, ObjectFilename, LVP, ""); 522 if (!MI) { 523 printLead(Bytes, Address.Address, OS); 524 OS << " <unknown>"; 525 return; 526 } 527 std::string Buffer; 528 { 529 raw_string_ostream TempStream(Buffer); 530 IP.printInst(MI, Address.Address, "", STI, TempStream); 531 } 532 StringRef Contents(Buffer); 533 // Split off bundle attributes 534 auto PacketBundle = Contents.rsplit('\n'); 535 // Split off first instruction from the rest 536 auto HeadTail = PacketBundle.first.split('\n'); 537 auto Preamble = " { "; 538 auto Separator = ""; 539 540 // Hexagon's packets require relocations to be inline rather than 541 // clustered at the end of the packet. 542 std::vector<RelocationRef>::const_iterator RelCur = Rels->begin(); 543 std::vector<RelocationRef>::const_iterator RelEnd = Rels->end(); 544 auto PrintReloc = [&]() -> void { 545 while ((RelCur != RelEnd) && (RelCur->getOffset() <= Address.Address)) { 546 if (RelCur->getOffset() == Address.Address) { 547 printRelocation(OS, ObjectFilename, *RelCur, Address.Address, false); 548 return; 549 } 550 ++RelCur; 551 } 552 }; 553 554 while (!HeadTail.first.empty()) { 555 OS << Separator; 556 Separator = "\n"; 557 if (SP && (PrintSource || PrintLines)) 558 SP->printSourceLine(OS, Address, ObjectFilename, LVP, ""); 559 printLead(Bytes, Address.Address, OS); 560 OS << Preamble; 561 Preamble = " "; 562 StringRef Inst; 563 auto Duplex = HeadTail.first.split('\v'); 564 if (!Duplex.second.empty()) { 565 OS << Duplex.first; 566 OS << "; "; 567 Inst = Duplex.second; 568 } 569 else 570 Inst = HeadTail.first; 571 OS << Inst; 572 HeadTail = HeadTail.second.split('\n'); 573 if (HeadTail.first.empty()) 574 OS << " } " << PacketBundle.second; 575 PrintReloc(); 576 Bytes = Bytes.slice(4); 577 Address.Address += 4; 578 } 579 } 580 }; 581 HexagonPrettyPrinter HexagonPrettyPrinterInst; 582 583 class AMDGCNPrettyPrinter : public PrettyPrinter { 584 public: 585 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 586 object::SectionedAddress Address, formatted_raw_ostream &OS, 587 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 588 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 589 LiveVariablePrinter &LVP) override { 590 if (SP && (PrintSource || PrintLines)) 591 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 592 593 if (MI) { 594 SmallString<40> InstStr; 595 raw_svector_ostream IS(InstStr); 596 597 IP.printInst(MI, Address.Address, "", STI, IS); 598 599 OS << left_justify(IS.str(), 60); 600 } else { 601 // an unrecognized encoding - this is probably data so represent it 602 // using the .long directive, or .byte directive if fewer than 4 bytes 603 // remaining 604 if (Bytes.size() >= 4) { 605 OS << format("\t.long 0x%08" PRIx32 " ", 606 support::endian::read32<support::little>(Bytes.data())); 607 OS.indent(42); 608 } else { 609 OS << format("\t.byte 0x%02" PRIx8, Bytes[0]); 610 for (unsigned int i = 1; i < Bytes.size(); i++) 611 OS << format(", 0x%02" PRIx8, Bytes[i]); 612 OS.indent(55 - (6 * Bytes.size())); 613 } 614 } 615 616 OS << format("// %012" PRIX64 ":", Address.Address); 617 if (Bytes.size() >= 4) { 618 // D should be casted to uint32_t here as it is passed by format to 619 // snprintf as vararg. 620 for (uint32_t D : makeArrayRef( 621 reinterpret_cast<const support::little32_t *>(Bytes.data()), 622 Bytes.size() / 4)) 623 OS << format(" %08" PRIX32, D); 624 } else { 625 for (unsigned char B : Bytes) 626 OS << format(" %02" PRIX8, B); 627 } 628 629 if (!Annot.empty()) 630 OS << " // " << Annot; 631 } 632 }; 633 AMDGCNPrettyPrinter AMDGCNPrettyPrinterInst; 634 635 class BPFPrettyPrinter : public PrettyPrinter { 636 public: 637 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 638 object::SectionedAddress Address, formatted_raw_ostream &OS, 639 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 640 StringRef ObjectFilename, std::vector<RelocationRef> *Rels, 641 LiveVariablePrinter &LVP) override { 642 if (SP && (PrintSource || PrintLines)) 643 SP->printSourceLine(OS, Address, ObjectFilename, LVP); 644 if (LeadingAddr) 645 OS << format("%8" PRId64 ":", Address.Address / 8); 646 if (ShowRawInsn) { 647 OS << "\t"; 648 dumpBytes(Bytes, OS); 649 } 650 if (MI) 651 IP.printInst(MI, Address.Address, "", STI, OS); 652 else 653 OS << "\t<unknown>"; 654 } 655 }; 656 BPFPrettyPrinter BPFPrettyPrinterInst; 657 658 PrettyPrinter &selectPrettyPrinter(Triple const &Triple) { 659 switch(Triple.getArch()) { 660 default: 661 return PrettyPrinterInst; 662 case Triple::hexagon: 663 return HexagonPrettyPrinterInst; 664 case Triple::amdgcn: 665 return AMDGCNPrettyPrinterInst; 666 case Triple::bpfel: 667 case Triple::bpfeb: 668 return BPFPrettyPrinterInst; 669 } 670 } 671 } 672 673 static uint8_t getElfSymbolType(const ObjectFile *Obj, const SymbolRef &Sym) { 674 assert(Obj->isELF()); 675 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(Obj)) 676 return unwrapOrError(Elf32LEObj->getSymbol(Sym.getRawDataRefImpl()), 677 Obj->getFileName()) 678 ->getType(); 679 if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(Obj)) 680 return unwrapOrError(Elf64LEObj->getSymbol(Sym.getRawDataRefImpl()), 681 Obj->getFileName()) 682 ->getType(); 683 if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(Obj)) 684 return unwrapOrError(Elf32BEObj->getSymbol(Sym.getRawDataRefImpl()), 685 Obj->getFileName()) 686 ->getType(); 687 if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(Obj)) 688 return unwrapOrError(Elf64BEObj->getSymbol(Sym.getRawDataRefImpl()), 689 Obj->getFileName()) 690 ->getType(); 691 llvm_unreachable("Unsupported binary format"); 692 } 693 694 template <class ELFT> static void 695 addDynamicElfSymbols(const ELFObjectFile<ELFT> *Obj, 696 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 697 for (auto Symbol : Obj->getDynamicSymbolIterators()) { 698 uint8_t SymbolType = Symbol.getELFType(); 699 if (SymbolType == ELF::STT_SECTION) 700 continue; 701 702 uint64_t Address = unwrapOrError(Symbol.getAddress(), Obj->getFileName()); 703 // ELFSymbolRef::getAddress() returns size instead of value for common 704 // symbols which is not desirable for disassembly output. Overriding. 705 if (SymbolType == ELF::STT_COMMON) 706 Address = unwrapOrError(Obj->getSymbol(Symbol.getRawDataRefImpl()), 707 Obj->getFileName()) 708 ->st_value; 709 710 StringRef Name = unwrapOrError(Symbol.getName(), Obj->getFileName()); 711 if (Name.empty()) 712 continue; 713 714 section_iterator SecI = 715 unwrapOrError(Symbol.getSection(), Obj->getFileName()); 716 if (SecI == Obj->section_end()) 717 continue; 718 719 AllSymbols[*SecI].emplace_back(Address, Name, SymbolType); 720 } 721 } 722 723 static void 724 addDynamicElfSymbols(const ObjectFile *Obj, 725 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 726 assert(Obj->isELF()); 727 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(Obj)) 728 addDynamicElfSymbols(Elf32LEObj, AllSymbols); 729 else if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(Obj)) 730 addDynamicElfSymbols(Elf64LEObj, AllSymbols); 731 else if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(Obj)) 732 addDynamicElfSymbols(Elf32BEObj, AllSymbols); 733 else if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(Obj)) 734 addDynamicElfSymbols(Elf64BEObj, AllSymbols); 735 else 736 llvm_unreachable("Unsupported binary format"); 737 } 738 739 static Optional<SectionRef> getWasmCodeSection(const WasmObjectFile *Obj) { 740 for (auto SecI : Obj->sections()) { 741 const WasmSection &Section = Obj->getWasmSection(SecI); 742 if (Section.Type == wasm::WASM_SEC_CODE) 743 return SecI; 744 } 745 return None; 746 } 747 748 static void 749 addMissingWasmCodeSymbols(const WasmObjectFile *Obj, 750 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 751 Optional<SectionRef> Section = getWasmCodeSection(Obj); 752 if (!Section) 753 return; 754 SectionSymbolsTy &Symbols = AllSymbols[*Section]; 755 756 std::set<uint64_t> SymbolAddresses; 757 for (const auto &Sym : Symbols) 758 SymbolAddresses.insert(Sym.Addr); 759 760 for (const wasm::WasmFunction &Function : Obj->functions()) { 761 uint64_t Address = Function.CodeSectionOffset; 762 // Only add fallback symbols for functions not already present in the symbol 763 // table. 764 if (SymbolAddresses.count(Address)) 765 continue; 766 // This function has no symbol, so it should have no SymbolName. 767 assert(Function.SymbolName.empty()); 768 // We use DebugName for the name, though it may be empty if there is no 769 // "name" custom section, or that section is missing a name for this 770 // function. 771 StringRef Name = Function.DebugName; 772 Symbols.emplace_back(Address, Name, ELF::STT_NOTYPE); 773 } 774 } 775 776 static void addPltEntries(const ObjectFile *Obj, 777 std::map<SectionRef, SectionSymbolsTy> &AllSymbols, 778 StringSaver &Saver) { 779 Optional<SectionRef> Plt = None; 780 for (const SectionRef &Section : Obj->sections()) { 781 Expected<StringRef> SecNameOrErr = Section.getName(); 782 if (!SecNameOrErr) { 783 consumeError(SecNameOrErr.takeError()); 784 continue; 785 } 786 if (*SecNameOrErr == ".plt") 787 Plt = Section; 788 } 789 if (!Plt) 790 return; 791 if (auto *ElfObj = dyn_cast<ELFObjectFileBase>(Obj)) { 792 for (auto PltEntry : ElfObj->getPltAddresses()) { 793 if (PltEntry.first) { 794 SymbolRef Symbol(*PltEntry.first, ElfObj); 795 uint8_t SymbolType = getElfSymbolType(Obj, Symbol); 796 if (Expected<StringRef> NameOrErr = Symbol.getName()) { 797 if (!NameOrErr->empty()) 798 AllSymbols[*Plt].emplace_back( 799 PltEntry.second, Saver.save((*NameOrErr + "@plt").str()), 800 SymbolType); 801 continue; 802 } else { 803 // The warning has been reported in disassembleObject(). 804 consumeError(NameOrErr.takeError()); 805 } 806 } 807 reportWarning("PLT entry at 0x" + Twine::utohexstr(PltEntry.second) + 808 " references an invalid symbol", 809 Obj->getFileName()); 810 } 811 } 812 } 813 814 // Normally the disassembly output will skip blocks of zeroes. This function 815 // returns the number of zero bytes that can be skipped when dumping the 816 // disassembly of the instructions in Buf. 817 static size_t countSkippableZeroBytes(ArrayRef<uint8_t> Buf) { 818 // Find the number of leading zeroes. 819 size_t N = 0; 820 while (N < Buf.size() && !Buf[N]) 821 ++N; 822 823 // We may want to skip blocks of zero bytes, but unless we see 824 // at least 8 of them in a row. 825 if (N < 8) 826 return 0; 827 828 // We skip zeroes in multiples of 4 because do not want to truncate an 829 // instruction if it starts with a zero byte. 830 return N & ~0x3; 831 } 832 833 // Returns a map from sections to their relocations. 834 static std::map<SectionRef, std::vector<RelocationRef>> 835 getRelocsMap(object::ObjectFile const &Obj) { 836 std::map<SectionRef, std::vector<RelocationRef>> Ret; 837 uint64_t I = (uint64_t)-1; 838 for (SectionRef Sec : Obj.sections()) { 839 ++I; 840 Expected<section_iterator> RelocatedOrErr = Sec.getRelocatedSection(); 841 if (!RelocatedOrErr) 842 reportError(Obj.getFileName(), 843 "section (" + Twine(I) + 844 "): failed to get a relocated section: " + 845 toString(RelocatedOrErr.takeError())); 846 847 section_iterator Relocated = *RelocatedOrErr; 848 if (Relocated == Obj.section_end() || !checkSectionFilter(*Relocated).Keep) 849 continue; 850 std::vector<RelocationRef> &V = Ret[*Relocated]; 851 append_range(V, Sec.relocations()); 852 // Sort relocations by address. 853 llvm::stable_sort(V, isRelocAddressLess); 854 } 855 return Ret; 856 } 857 858 // Used for --adjust-vma to check if address should be adjusted by the 859 // specified value for a given section. 860 // For ELF we do not adjust non-allocatable sections like debug ones, 861 // because they are not loadable. 862 // TODO: implement for other file formats. 863 static bool shouldAdjustVA(const SectionRef &Section) { 864 const ObjectFile *Obj = Section.getObject(); 865 if (Obj->isELF()) 866 return ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC; 867 return false; 868 } 869 870 871 typedef std::pair<uint64_t, char> MappingSymbolPair; 872 static char getMappingSymbolKind(ArrayRef<MappingSymbolPair> MappingSymbols, 873 uint64_t Address) { 874 auto It = 875 partition_point(MappingSymbols, [Address](const MappingSymbolPair &Val) { 876 return Val.first <= Address; 877 }); 878 // Return zero for any address before the first mapping symbol; this means 879 // we should use the default disassembly mode, depending on the target. 880 if (It == MappingSymbols.begin()) 881 return '\x00'; 882 return (It - 1)->second; 883 } 884 885 static uint64_t dumpARMELFData(uint64_t SectionAddr, uint64_t Index, 886 uint64_t End, const ObjectFile *Obj, 887 ArrayRef<uint8_t> Bytes, 888 ArrayRef<MappingSymbolPair> MappingSymbols, 889 raw_ostream &OS) { 890 support::endianness Endian = 891 Obj->isLittleEndian() ? support::little : support::big; 892 OS << format("%8" PRIx64 ":\t", SectionAddr + Index); 893 if (Index + 4 <= End) { 894 dumpBytes(Bytes.slice(Index, 4), OS); 895 OS << "\t.word\t" 896 << format_hex(support::endian::read32(Bytes.data() + Index, Endian), 897 10); 898 return 4; 899 } 900 if (Index + 2 <= End) { 901 dumpBytes(Bytes.slice(Index, 2), OS); 902 OS << "\t\t.short\t" 903 << format_hex(support::endian::read16(Bytes.data() + Index, Endian), 904 6); 905 return 2; 906 } 907 dumpBytes(Bytes.slice(Index, 1), OS); 908 OS << "\t\t.byte\t" << format_hex(Bytes[0], 4); 909 return 1; 910 } 911 912 static void dumpELFData(uint64_t SectionAddr, uint64_t Index, uint64_t End, 913 ArrayRef<uint8_t> Bytes) { 914 // print out data up to 8 bytes at a time in hex and ascii 915 uint8_t AsciiData[9] = {'\0'}; 916 uint8_t Byte; 917 int NumBytes = 0; 918 919 for (; Index < End; ++Index) { 920 if (NumBytes == 0) 921 outs() << format("%8" PRIx64 ":", SectionAddr + Index); 922 Byte = Bytes.slice(Index)[0]; 923 outs() << format(" %02x", Byte); 924 AsciiData[NumBytes] = isPrint(Byte) ? Byte : '.'; 925 926 uint8_t IndentOffset = 0; 927 NumBytes++; 928 if (Index == End - 1 || NumBytes > 8) { 929 // Indent the space for less than 8 bytes data. 930 // 2 spaces for byte and one for space between bytes 931 IndentOffset = 3 * (8 - NumBytes); 932 for (int Excess = NumBytes; Excess < 8; Excess++) 933 AsciiData[Excess] = '\0'; 934 NumBytes = 8; 935 } 936 if (NumBytes == 8) { 937 AsciiData[8] = '\0'; 938 outs() << std::string(IndentOffset, ' ') << " "; 939 outs() << reinterpret_cast<char *>(AsciiData); 940 outs() << '\n'; 941 NumBytes = 0; 942 } 943 } 944 } 945 946 SymbolInfoTy objdump::createSymbolInfo(const ObjectFile *Obj, 947 const SymbolRef &Symbol) { 948 const StringRef FileName = Obj->getFileName(); 949 const uint64_t Addr = unwrapOrError(Symbol.getAddress(), FileName); 950 const StringRef Name = unwrapOrError(Symbol.getName(), FileName); 951 952 if (Obj->isXCOFF() && SymbolDescription) { 953 const auto *XCOFFObj = cast<XCOFFObjectFile>(Obj); 954 DataRefImpl SymbolDRI = Symbol.getRawDataRefImpl(); 955 956 const uint32_t SymbolIndex = XCOFFObj->getSymbolIndex(SymbolDRI.p); 957 Optional<XCOFF::StorageMappingClass> Smc = 958 getXCOFFSymbolCsectSMC(XCOFFObj, Symbol); 959 return SymbolInfoTy(Addr, Name, Smc, SymbolIndex, 960 isLabel(XCOFFObj, Symbol)); 961 } else 962 return SymbolInfoTy(Addr, Name, 963 Obj->isELF() ? getElfSymbolType(Obj, Symbol) 964 : (uint8_t)ELF::STT_NOTYPE); 965 } 966 967 static SymbolInfoTy createDummySymbolInfo(const ObjectFile *Obj, 968 const uint64_t Addr, StringRef &Name, 969 uint8_t Type) { 970 if (Obj->isXCOFF() && SymbolDescription) 971 return SymbolInfoTy(Addr, Name, None, None, false); 972 else 973 return SymbolInfoTy(Addr, Name, Type); 974 } 975 976 static void 977 collectLocalBranchTargets(ArrayRef<uint8_t> Bytes, const MCInstrAnalysis *MIA, 978 MCDisassembler *DisAsm, MCInstPrinter *IP, 979 const MCSubtargetInfo *STI, uint64_t SectionAddr, 980 uint64_t Start, uint64_t End, 981 std::unordered_map<uint64_t, std::string> &Labels) { 982 // So far only supports X86. 983 if (!STI->getTargetTriple().isX86()) 984 return; 985 986 Labels.clear(); 987 unsigned LabelCount = 0; 988 Start += SectionAddr; 989 End += SectionAddr; 990 uint64_t Index = Start; 991 while (Index < End) { 992 // Disassemble a real instruction and record function-local branch labels. 993 MCInst Inst; 994 uint64_t Size; 995 bool Disassembled = DisAsm->getInstruction( 996 Inst, Size, Bytes.slice(Index - SectionAddr), Index, nulls()); 997 if (Size == 0) 998 Size = 1; 999 1000 if (Disassembled && MIA) { 1001 uint64_t Target; 1002 bool TargetKnown = MIA->evaluateBranch(Inst, Index, Size, Target); 1003 if (TargetKnown && (Target >= Start && Target < End) && 1004 !Labels.count(Target)) 1005 Labels[Target] = ("L" + Twine(LabelCount++)).str(); 1006 } 1007 1008 Index += Size; 1009 } 1010 } 1011 1012 // Create an MCSymbolizer for the target and add it to the MCDisassembler. 1013 // This is currently only used on AMDGPU, and assumes the format of the 1014 // void * argument passed to AMDGPU's createMCSymbolizer. 1015 static void addSymbolizer( 1016 MCContext &Ctx, const Target *Target, StringRef TripleName, 1017 MCDisassembler *DisAsm, uint64_t SectionAddr, ArrayRef<uint8_t> Bytes, 1018 SectionSymbolsTy &Symbols, 1019 std::vector<std::unique_ptr<std::string>> &SynthesizedLabelNames) { 1020 1021 std::unique_ptr<MCRelocationInfo> RelInfo( 1022 Target->createMCRelocationInfo(TripleName, Ctx)); 1023 if (!RelInfo) 1024 return; 1025 std::unique_ptr<MCSymbolizer> Symbolizer(Target->createMCSymbolizer( 1026 TripleName, nullptr, nullptr, &Symbols, &Ctx, std::move(RelInfo))); 1027 MCSymbolizer *SymbolizerPtr = &*Symbolizer; 1028 DisAsm->setSymbolizer(std::move(Symbolizer)); 1029 1030 if (!SymbolizeOperands) 1031 return; 1032 1033 // Synthesize labels referenced by branch instructions by 1034 // disassembling, discarding the output, and collecting the referenced 1035 // addresses from the symbolizer. 1036 for (size_t Index = 0; Index != Bytes.size();) { 1037 MCInst Inst; 1038 uint64_t Size; 1039 DisAsm->getInstruction(Inst, Size, Bytes.slice(Index), SectionAddr + Index, 1040 nulls()); 1041 if (Size == 0) 1042 Size = 1; 1043 Index += Size; 1044 } 1045 ArrayRef<uint64_t> LabelAddrsRef = SymbolizerPtr->getReferencedAddresses(); 1046 // Copy and sort to remove duplicates. 1047 std::vector<uint64_t> LabelAddrs; 1048 LabelAddrs.insert(LabelAddrs.end(), LabelAddrsRef.begin(), 1049 LabelAddrsRef.end()); 1050 llvm::sort(LabelAddrs); 1051 LabelAddrs.resize(std::unique(LabelAddrs.begin(), LabelAddrs.end()) - 1052 LabelAddrs.begin()); 1053 // Add the labels. 1054 for (unsigned LabelNum = 0; LabelNum != LabelAddrs.size(); ++LabelNum) { 1055 auto Name = std::make_unique<std::string>(); 1056 *Name = (Twine("L") + Twine(LabelNum)).str(); 1057 SynthesizedLabelNames.push_back(std::move(Name)); 1058 Symbols.push_back(SymbolInfoTy( 1059 LabelAddrs[LabelNum], *SynthesizedLabelNames.back(), ELF::STT_NOTYPE)); 1060 } 1061 llvm::stable_sort(Symbols); 1062 // Recreate the symbolizer with the new symbols list. 1063 RelInfo.reset(Target->createMCRelocationInfo(TripleName, Ctx)); 1064 Symbolizer.reset(Target->createMCSymbolizer( 1065 TripleName, nullptr, nullptr, &Symbols, &Ctx, std::move(RelInfo))); 1066 DisAsm->setSymbolizer(std::move(Symbolizer)); 1067 } 1068 1069 static StringRef getSegmentName(const MachOObjectFile *MachO, 1070 const SectionRef &Section) { 1071 if (MachO) { 1072 DataRefImpl DR = Section.getRawDataRefImpl(); 1073 StringRef SegmentName = MachO->getSectionFinalSegmentName(DR); 1074 return SegmentName; 1075 } 1076 return ""; 1077 } 1078 1079 static void emitPostInstructionInfo(formatted_raw_ostream &FOS, 1080 const MCAsmInfo &MAI, 1081 const MCSubtargetInfo &STI, 1082 StringRef Comments, 1083 LiveVariablePrinter &LVP) { 1084 do { 1085 if (!Comments.empty()) { 1086 // Emit a line of comments. 1087 StringRef Comment; 1088 std::tie(Comment, Comments) = Comments.split('\n'); 1089 // MAI.getCommentColumn() assumes that instructions are printed at the 1090 // position of 8, while getInstStartColumn() returns the actual position. 1091 unsigned CommentColumn = 1092 MAI.getCommentColumn() - 8 + getInstStartColumn(STI); 1093 FOS.PadToColumn(CommentColumn); 1094 FOS << MAI.getCommentString() << ' ' << Comment; 1095 } 1096 LVP.printAfterInst(FOS); 1097 FOS << '\n'; 1098 } while (!Comments.empty()); 1099 FOS.flush(); 1100 } 1101 1102 static void disassembleObject(const Target *TheTarget, const ObjectFile *Obj, 1103 MCContext &Ctx, MCDisassembler *PrimaryDisAsm, 1104 MCDisassembler *SecondaryDisAsm, 1105 const MCInstrAnalysis *MIA, MCInstPrinter *IP, 1106 const MCSubtargetInfo *PrimarySTI, 1107 const MCSubtargetInfo *SecondarySTI, 1108 PrettyPrinter &PIP, 1109 SourcePrinter &SP, bool InlineRelocs) { 1110 const MCSubtargetInfo *STI = PrimarySTI; 1111 MCDisassembler *DisAsm = PrimaryDisAsm; 1112 bool PrimaryIsThumb = false; 1113 if (isArmElf(Obj)) 1114 PrimaryIsThumb = STI->checkFeatures("+thumb-mode"); 1115 1116 std::map<SectionRef, std::vector<RelocationRef>> RelocMap; 1117 if (InlineRelocs) 1118 RelocMap = getRelocsMap(*Obj); 1119 bool Is64Bits = Obj->getBytesInAddress() > 4; 1120 1121 // Create a mapping from virtual address to symbol name. This is used to 1122 // pretty print the symbols while disassembling. 1123 std::map<SectionRef, SectionSymbolsTy> AllSymbols; 1124 SectionSymbolsTy AbsoluteSymbols; 1125 const StringRef FileName = Obj->getFileName(); 1126 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(Obj); 1127 for (const SymbolRef &Symbol : Obj->symbols()) { 1128 Expected<StringRef> NameOrErr = Symbol.getName(); 1129 if (!NameOrErr) { 1130 reportWarning(toString(NameOrErr.takeError()), FileName); 1131 continue; 1132 } 1133 if (NameOrErr->empty() && !(Obj->isXCOFF() && SymbolDescription)) 1134 continue; 1135 1136 if (Obj->isELF() && getElfSymbolType(Obj, Symbol) == ELF::STT_SECTION) 1137 continue; 1138 1139 if (MachO) { 1140 // __mh_(execute|dylib|dylinker|bundle|preload|object)_header are special 1141 // symbols that support MachO header introspection. They do not bind to 1142 // code locations and are irrelevant for disassembly. 1143 if (NameOrErr->startswith("__mh_") && NameOrErr->endswith("_header")) 1144 continue; 1145 // Don't ask a Mach-O STAB symbol for its section unless you know that 1146 // STAB symbol's section field refers to a valid section index. Otherwise 1147 // the symbol may error trying to load a section that does not exist. 1148 DataRefImpl SymDRI = Symbol.getRawDataRefImpl(); 1149 uint8_t NType = (MachO->is64Bit() ? 1150 MachO->getSymbol64TableEntry(SymDRI).n_type: 1151 MachO->getSymbolTableEntry(SymDRI).n_type); 1152 if (NType & MachO::N_STAB) 1153 continue; 1154 } 1155 1156 section_iterator SecI = unwrapOrError(Symbol.getSection(), FileName); 1157 if (SecI != Obj->section_end()) 1158 AllSymbols[*SecI].push_back(createSymbolInfo(Obj, Symbol)); 1159 else 1160 AbsoluteSymbols.push_back(createSymbolInfo(Obj, Symbol)); 1161 } 1162 1163 if (AllSymbols.empty() && Obj->isELF()) 1164 addDynamicElfSymbols(Obj, AllSymbols); 1165 1166 if (Obj->isWasm()) 1167 addMissingWasmCodeSymbols(cast<WasmObjectFile>(Obj), AllSymbols); 1168 1169 BumpPtrAllocator A; 1170 StringSaver Saver(A); 1171 addPltEntries(Obj, AllSymbols, Saver); 1172 1173 // Create a mapping from virtual address to section. An empty section can 1174 // cause more than one section at the same address. Sort such sections to be 1175 // before same-addressed non-empty sections so that symbol lookups prefer the 1176 // non-empty section. 1177 std::vector<std::pair<uint64_t, SectionRef>> SectionAddresses; 1178 for (SectionRef Sec : Obj->sections()) 1179 SectionAddresses.emplace_back(Sec.getAddress(), Sec); 1180 llvm::stable_sort(SectionAddresses, [](const auto &LHS, const auto &RHS) { 1181 if (LHS.first != RHS.first) 1182 return LHS.first < RHS.first; 1183 return LHS.second.getSize() < RHS.second.getSize(); 1184 }); 1185 1186 // Linked executables (.exe and .dll files) typically don't include a real 1187 // symbol table but they might contain an export table. 1188 if (const auto *COFFObj = dyn_cast<COFFObjectFile>(Obj)) { 1189 for (const auto &ExportEntry : COFFObj->export_directories()) { 1190 StringRef Name; 1191 if (Error E = ExportEntry.getSymbolName(Name)) 1192 reportError(std::move(E), Obj->getFileName()); 1193 if (Name.empty()) 1194 continue; 1195 1196 uint32_t RVA; 1197 if (Error E = ExportEntry.getExportRVA(RVA)) 1198 reportError(std::move(E), Obj->getFileName()); 1199 1200 uint64_t VA = COFFObj->getImageBase() + RVA; 1201 auto Sec = partition_point( 1202 SectionAddresses, [VA](const std::pair<uint64_t, SectionRef> &O) { 1203 return O.first <= VA; 1204 }); 1205 if (Sec != SectionAddresses.begin()) { 1206 --Sec; 1207 AllSymbols[Sec->second].emplace_back(VA, Name, ELF::STT_NOTYPE); 1208 } else 1209 AbsoluteSymbols.emplace_back(VA, Name, ELF::STT_NOTYPE); 1210 } 1211 } 1212 1213 // Sort all the symbols, this allows us to use a simple binary search to find 1214 // Multiple symbols can have the same address. Use a stable sort to stabilize 1215 // the output. 1216 StringSet<> FoundDisasmSymbolSet; 1217 for (std::pair<const SectionRef, SectionSymbolsTy> &SecSyms : AllSymbols) 1218 llvm::stable_sort(SecSyms.second); 1219 llvm::stable_sort(AbsoluteSymbols); 1220 1221 std::unique_ptr<DWARFContext> DICtx; 1222 LiveVariablePrinter LVP(*Ctx.getRegisterInfo(), *STI); 1223 1224 if (DbgVariables != DVDisabled) { 1225 DICtx = DWARFContext::create(*Obj); 1226 for (const std::unique_ptr<DWARFUnit> &CU : DICtx->compile_units()) 1227 LVP.addCompileUnit(CU->getUnitDIE(false)); 1228 } 1229 1230 LLVM_DEBUG(LVP.dump()); 1231 1232 for (const SectionRef &Section : ToolSectionFilter(*Obj)) { 1233 if (FilterSections.empty() && !DisassembleAll && 1234 (!Section.isText() || Section.isVirtual())) 1235 continue; 1236 1237 uint64_t SectionAddr = Section.getAddress(); 1238 uint64_t SectSize = Section.getSize(); 1239 if (!SectSize) 1240 continue; 1241 1242 // Get the list of all the symbols in this section. 1243 SectionSymbolsTy &Symbols = AllSymbols[Section]; 1244 std::vector<MappingSymbolPair> MappingSymbols; 1245 if (hasMappingSymbols(Obj)) { 1246 for (const auto &Symb : Symbols) { 1247 uint64_t Address = Symb.Addr; 1248 StringRef Name = Symb.Name; 1249 if (Name.startswith("$d")) 1250 MappingSymbols.emplace_back(Address - SectionAddr, 'd'); 1251 if (Name.startswith("$x")) 1252 MappingSymbols.emplace_back(Address - SectionAddr, 'x'); 1253 if (Name.startswith("$a")) 1254 MappingSymbols.emplace_back(Address - SectionAddr, 'a'); 1255 if (Name.startswith("$t")) 1256 MappingSymbols.emplace_back(Address - SectionAddr, 't'); 1257 } 1258 } 1259 1260 llvm::sort(MappingSymbols); 1261 1262 ArrayRef<uint8_t> Bytes = arrayRefFromStringRef( 1263 unwrapOrError(Section.getContents(), Obj->getFileName())); 1264 1265 std::vector<std::unique_ptr<std::string>> SynthesizedLabelNames; 1266 if (Obj->isELF() && Obj->getArch() == Triple::amdgcn) { 1267 // AMDGPU disassembler uses symbolizer for printing labels 1268 addSymbolizer(Ctx, TheTarget, TripleName, DisAsm, SectionAddr, Bytes, 1269 Symbols, SynthesizedLabelNames); 1270 } 1271 1272 StringRef SegmentName = getSegmentName(MachO, Section); 1273 StringRef SectionName = unwrapOrError(Section.getName(), Obj->getFileName()); 1274 // If the section has no symbol at the start, just insert a dummy one. 1275 if (Symbols.empty() || Symbols[0].Addr != 0) { 1276 Symbols.insert(Symbols.begin(), 1277 createDummySymbolInfo(Obj, SectionAddr, SectionName, 1278 Section.isText() ? ELF::STT_FUNC 1279 : ELF::STT_OBJECT)); 1280 } 1281 1282 SmallString<40> Comments; 1283 raw_svector_ostream CommentStream(Comments); 1284 1285 uint64_t VMAAdjustment = 0; 1286 if (shouldAdjustVA(Section)) 1287 VMAAdjustment = AdjustVMA; 1288 1289 uint64_t Size; 1290 uint64_t Index; 1291 bool PrintedSection = false; 1292 std::vector<RelocationRef> Rels = RelocMap[Section]; 1293 std::vector<RelocationRef>::const_iterator RelCur = Rels.begin(); 1294 std::vector<RelocationRef>::const_iterator RelEnd = Rels.end(); 1295 // Disassemble symbol by symbol. 1296 for (unsigned SI = 0, SE = Symbols.size(); SI != SE; ++SI) { 1297 std::string SymbolName = Symbols[SI].Name.str(); 1298 if (Demangle) 1299 SymbolName = demangle(SymbolName); 1300 1301 // Skip if --disassemble-symbols is not empty and the symbol is not in 1302 // the list. 1303 if (!DisasmSymbolSet.empty() && !DisasmSymbolSet.count(SymbolName)) 1304 continue; 1305 1306 uint64_t Start = Symbols[SI].Addr; 1307 if (Start < SectionAddr || StopAddress <= Start) 1308 continue; 1309 else 1310 FoundDisasmSymbolSet.insert(SymbolName); 1311 1312 // The end is the section end, the beginning of the next symbol, or 1313 // --stop-address. 1314 uint64_t End = std::min<uint64_t>(SectionAddr + SectSize, StopAddress); 1315 if (SI + 1 < SE) 1316 End = std::min(End, Symbols[SI + 1].Addr); 1317 if (Start >= End || End <= StartAddress) 1318 continue; 1319 Start -= SectionAddr; 1320 End -= SectionAddr; 1321 1322 if (!PrintedSection) { 1323 PrintedSection = true; 1324 outs() << "\nDisassembly of section "; 1325 if (!SegmentName.empty()) 1326 outs() << SegmentName << ","; 1327 outs() << SectionName << ":\n"; 1328 } 1329 1330 outs() << '\n'; 1331 if (LeadingAddr) 1332 outs() << format(Is64Bits ? "%016" PRIx64 " " : "%08" PRIx64 " ", 1333 SectionAddr + Start + VMAAdjustment); 1334 if (Obj->isXCOFF() && SymbolDescription) { 1335 outs() << getXCOFFSymbolDescription(Symbols[SI], SymbolName) << ":\n"; 1336 } else 1337 outs() << '<' << SymbolName << ">:\n"; 1338 1339 // Don't print raw contents of a virtual section. A virtual section 1340 // doesn't have any contents in the file. 1341 if (Section.isVirtual()) { 1342 outs() << "...\n"; 1343 continue; 1344 } 1345 1346 auto Status = DisAsm->onSymbolStart(Symbols[SI], Size, 1347 Bytes.slice(Start, End - Start), 1348 SectionAddr + Start, CommentStream); 1349 // To have round trippable disassembly, we fall back to decoding the 1350 // remaining bytes as instructions. 1351 // 1352 // If there is a failure, we disassemble the failed region as bytes before 1353 // falling back. The target is expected to print nothing in this case. 1354 // 1355 // If there is Success or SoftFail i.e no 'real' failure, we go ahead by 1356 // Size bytes before falling back. 1357 // So if the entire symbol is 'eaten' by the target: 1358 // Start += Size // Now Start = End and we will never decode as 1359 // // instructions 1360 // 1361 // Right now, most targets return None i.e ignore to treat a symbol 1362 // separately. But WebAssembly decodes preludes for some symbols. 1363 // 1364 if (Status.hasValue()) { 1365 if (Status.getValue() == MCDisassembler::Fail) { 1366 outs() << "// Error in decoding " << SymbolName 1367 << " : Decoding failed region as bytes.\n"; 1368 for (uint64_t I = 0; I < Size; ++I) { 1369 outs() << "\t.byte\t " << format_hex(Bytes[I], 1, /*Upper=*/true) 1370 << "\n"; 1371 } 1372 } 1373 } else { 1374 Size = 0; 1375 } 1376 1377 Start += Size; 1378 1379 Index = Start; 1380 if (SectionAddr < StartAddress) 1381 Index = std::max<uint64_t>(Index, StartAddress - SectionAddr); 1382 1383 // If there is a data/common symbol inside an ELF text section and we are 1384 // only disassembling text (applicable all architectures), we are in a 1385 // situation where we must print the data and not disassemble it. 1386 if (Obj->isELF() && !DisassembleAll && Section.isText()) { 1387 uint8_t SymTy = Symbols[SI].Type; 1388 if (SymTy == ELF::STT_OBJECT || SymTy == ELF::STT_COMMON) { 1389 dumpELFData(SectionAddr, Index, End, Bytes); 1390 Index = End; 1391 } 1392 } 1393 1394 bool CheckARMELFData = hasMappingSymbols(Obj) && 1395 Symbols[SI].Type != ELF::STT_OBJECT && 1396 !DisassembleAll; 1397 bool DumpARMELFData = false; 1398 formatted_raw_ostream FOS(outs()); 1399 1400 std::unordered_map<uint64_t, std::string> AllLabels; 1401 if (SymbolizeOperands) 1402 collectLocalBranchTargets(Bytes, MIA, DisAsm, IP, PrimarySTI, 1403 SectionAddr, Index, End, AllLabels); 1404 1405 while (Index < End) { 1406 // ARM and AArch64 ELF binaries can interleave data and text in the 1407 // same section. We rely on the markers introduced to understand what 1408 // we need to dump. If the data marker is within a function, it is 1409 // denoted as a word/short etc. 1410 if (CheckARMELFData) { 1411 char Kind = getMappingSymbolKind(MappingSymbols, Index); 1412 DumpARMELFData = Kind == 'd'; 1413 if (SecondarySTI) { 1414 if (Kind == 'a') { 1415 STI = PrimaryIsThumb ? SecondarySTI : PrimarySTI; 1416 DisAsm = PrimaryIsThumb ? SecondaryDisAsm : PrimaryDisAsm; 1417 } else if (Kind == 't') { 1418 STI = PrimaryIsThumb ? PrimarySTI : SecondarySTI; 1419 DisAsm = PrimaryIsThumb ? PrimaryDisAsm : SecondaryDisAsm; 1420 } 1421 } 1422 } 1423 1424 if (DumpARMELFData) { 1425 Size = dumpARMELFData(SectionAddr, Index, End, Obj, Bytes, 1426 MappingSymbols, FOS); 1427 } else { 1428 // When -z or --disassemble-zeroes are given we always dissasemble 1429 // them. Otherwise we might want to skip zero bytes we see. 1430 if (!DisassembleZeroes) { 1431 uint64_t MaxOffset = End - Index; 1432 // For --reloc: print zero blocks patched by relocations, so that 1433 // relocations can be shown in the dump. 1434 if (RelCur != RelEnd) 1435 MaxOffset = RelCur->getOffset() - Index; 1436 1437 if (size_t N = 1438 countSkippableZeroBytes(Bytes.slice(Index, MaxOffset))) { 1439 FOS << "\t\t..." << '\n'; 1440 Index += N; 1441 continue; 1442 } 1443 } 1444 1445 // Print local label if there's any. 1446 auto Iter = AllLabels.find(SectionAddr + Index); 1447 if (Iter != AllLabels.end()) 1448 FOS << "<" << Iter->second << ">:\n"; 1449 1450 // Disassemble a real instruction or a data when disassemble all is 1451 // provided 1452 MCInst Inst; 1453 bool Disassembled = 1454 DisAsm->getInstruction(Inst, Size, Bytes.slice(Index), 1455 SectionAddr + Index, CommentStream); 1456 if (Size == 0) 1457 Size = 1; 1458 1459 LVP.update({Index, Section.getIndex()}, 1460 {Index + Size, Section.getIndex()}, Index + Size != End); 1461 1462 IP->setCommentStream(CommentStream); 1463 1464 PIP.printInst( 1465 *IP, Disassembled ? &Inst : nullptr, Bytes.slice(Index, Size), 1466 {SectionAddr + Index + VMAAdjustment, Section.getIndex()}, FOS, 1467 "", *STI, &SP, Obj->getFileName(), &Rels, LVP); 1468 1469 IP->setCommentStream(llvm::nulls()); 1470 1471 // If disassembly has failed, avoid analysing invalid/incomplete 1472 // instruction information. Otherwise, try to resolve the target 1473 // address (jump target or memory operand address) and print it on the 1474 // right of the instruction. 1475 if (Disassembled && MIA) { 1476 // Branch targets are printed just after the instructions. 1477 llvm::raw_ostream *TargetOS = &FOS; 1478 uint64_t Target; 1479 bool PrintTarget = 1480 MIA->evaluateBranch(Inst, SectionAddr + Index, Size, Target); 1481 if (!PrintTarget) 1482 if (Optional<uint64_t> MaybeTarget = 1483 MIA->evaluateMemoryOperandAddress( 1484 Inst, SectionAddr + Index, Size)) { 1485 Target = *MaybeTarget; 1486 PrintTarget = true; 1487 // Do not print real address when symbolizing. 1488 if (!SymbolizeOperands) { 1489 // Memory operand addresses are printed as comments. 1490 TargetOS = &CommentStream; 1491 *TargetOS << "0x" << Twine::utohexstr(Target); 1492 } 1493 } 1494 if (PrintTarget) { 1495 // In a relocatable object, the target's section must reside in 1496 // the same section as the call instruction or it is accessed 1497 // through a relocation. 1498 // 1499 // In a non-relocatable object, the target may be in any section. 1500 // In that case, locate the section(s) containing the target 1501 // address and find the symbol in one of those, if possible. 1502 // 1503 // N.B. We don't walk the relocations in the relocatable case yet. 1504 std::vector<const SectionSymbolsTy *> TargetSectionSymbols; 1505 if (!Obj->isRelocatableObject()) { 1506 auto It = llvm::partition_point( 1507 SectionAddresses, 1508 [=](const std::pair<uint64_t, SectionRef> &O) { 1509 return O.first <= Target; 1510 }); 1511 uint64_t TargetSecAddr = 0; 1512 while (It != SectionAddresses.begin()) { 1513 --It; 1514 if (TargetSecAddr == 0) 1515 TargetSecAddr = It->first; 1516 if (It->first != TargetSecAddr) 1517 break; 1518 TargetSectionSymbols.push_back(&AllSymbols[It->second]); 1519 } 1520 } else { 1521 TargetSectionSymbols.push_back(&Symbols); 1522 } 1523 TargetSectionSymbols.push_back(&AbsoluteSymbols); 1524 1525 // Find the last symbol in the first candidate section whose 1526 // offset is less than or equal to the target. If there are no 1527 // such symbols, try in the next section and so on, before finally 1528 // using the nearest preceding absolute symbol (if any), if there 1529 // are no other valid symbols. 1530 const SymbolInfoTy *TargetSym = nullptr; 1531 for (const SectionSymbolsTy *TargetSymbols : 1532 TargetSectionSymbols) { 1533 auto It = llvm::partition_point( 1534 *TargetSymbols, 1535 [=](const SymbolInfoTy &O) { return O.Addr <= Target; }); 1536 if (It != TargetSymbols->begin()) { 1537 TargetSym = &*(It - 1); 1538 break; 1539 } 1540 } 1541 1542 // Print the labels corresponding to the target if there's any. 1543 bool LabelAvailable = AllLabels.count(Target); 1544 if (TargetSym != nullptr) { 1545 uint64_t TargetAddress = TargetSym->Addr; 1546 uint64_t Disp = Target - TargetAddress; 1547 std::string TargetName = TargetSym->Name.str(); 1548 if (Demangle) 1549 TargetName = demangle(TargetName); 1550 1551 *TargetOS << " <"; 1552 if (!Disp) { 1553 // Always Print the binary symbol precisely corresponding to 1554 // the target address. 1555 *TargetOS << TargetName; 1556 } else if (!LabelAvailable) { 1557 // Always Print the binary symbol plus an offset if there's no 1558 // local label corresponding to the target address. 1559 *TargetOS << TargetName << "+0x" << Twine::utohexstr(Disp); 1560 } else { 1561 *TargetOS << AllLabels[Target]; 1562 } 1563 *TargetOS << ">"; 1564 } else if (LabelAvailable) { 1565 *TargetOS << " <" << AllLabels[Target] << ">"; 1566 } 1567 // By convention, each record in the comment stream should be 1568 // terminated. 1569 if (TargetOS == &CommentStream) 1570 *TargetOS << "\n"; 1571 } 1572 } 1573 } 1574 1575 assert(Ctx.getAsmInfo()); 1576 emitPostInstructionInfo(FOS, *Ctx.getAsmInfo(), *STI, 1577 CommentStream.str(), LVP); 1578 Comments.clear(); 1579 1580 // Hexagon does this in pretty printer 1581 if (Obj->getArch() != Triple::hexagon) { 1582 // Print relocation for instruction and data. 1583 while (RelCur != RelEnd) { 1584 uint64_t Offset = RelCur->getOffset(); 1585 // If this relocation is hidden, skip it. 1586 if (getHidden(*RelCur) || SectionAddr + Offset < StartAddress) { 1587 ++RelCur; 1588 continue; 1589 } 1590 1591 // Stop when RelCur's offset is past the disassembled 1592 // instruction/data. Note that it's possible the disassembled data 1593 // is not the complete data: we might see the relocation printed in 1594 // the middle of the data, but this matches the binutils objdump 1595 // output. 1596 if (Offset >= Index + Size) 1597 break; 1598 1599 // When --adjust-vma is used, update the address printed. 1600 if (RelCur->getSymbol() != Obj->symbol_end()) { 1601 Expected<section_iterator> SymSI = 1602 RelCur->getSymbol()->getSection(); 1603 if (SymSI && *SymSI != Obj->section_end() && 1604 shouldAdjustVA(**SymSI)) 1605 Offset += AdjustVMA; 1606 } 1607 1608 printRelocation(FOS, Obj->getFileName(), *RelCur, 1609 SectionAddr + Offset, Is64Bits); 1610 LVP.printAfterOtherLine(FOS, true); 1611 ++RelCur; 1612 } 1613 } 1614 1615 Index += Size; 1616 } 1617 } 1618 } 1619 StringSet<> MissingDisasmSymbolSet = 1620 set_difference(DisasmSymbolSet, FoundDisasmSymbolSet); 1621 for (StringRef Sym : MissingDisasmSymbolSet.keys()) 1622 reportWarning("failed to disassemble missing symbol " + Sym, FileName); 1623 } 1624 1625 static void disassembleObject(const ObjectFile *Obj, bool InlineRelocs) { 1626 const Target *TheTarget = getTarget(Obj); 1627 1628 // Package up features to be passed to target/subtarget 1629 SubtargetFeatures Features = Obj->getFeatures(); 1630 if (!MAttrs.empty()) 1631 for (unsigned I = 0; I != MAttrs.size(); ++I) 1632 Features.AddFeature(MAttrs[I]); 1633 1634 std::unique_ptr<const MCRegisterInfo> MRI( 1635 TheTarget->createMCRegInfo(TripleName)); 1636 if (!MRI) 1637 reportError(Obj->getFileName(), 1638 "no register info for target " + TripleName); 1639 1640 // Set up disassembler. 1641 MCTargetOptions MCOptions; 1642 std::unique_ptr<const MCAsmInfo> AsmInfo( 1643 TheTarget->createMCAsmInfo(*MRI, TripleName, MCOptions)); 1644 if (!AsmInfo) 1645 reportError(Obj->getFileName(), 1646 "no assembly info for target " + TripleName); 1647 1648 if (MCPU.empty()) 1649 MCPU = Obj->tryGetCPUName().getValueOr("").str(); 1650 1651 std::unique_ptr<const MCSubtargetInfo> STI( 1652 TheTarget->createMCSubtargetInfo(TripleName, MCPU, Features.getString())); 1653 if (!STI) 1654 reportError(Obj->getFileName(), 1655 "no subtarget info for target " + TripleName); 1656 std::unique_ptr<const MCInstrInfo> MII(TheTarget->createMCInstrInfo()); 1657 if (!MII) 1658 reportError(Obj->getFileName(), 1659 "no instruction info for target " + TripleName); 1660 MCContext Ctx(Triple(TripleName), AsmInfo.get(), MRI.get(), STI.get()); 1661 // FIXME: for now initialize MCObjectFileInfo with default values 1662 std::unique_ptr<MCObjectFileInfo> MOFI( 1663 TheTarget->createMCObjectFileInfo(Ctx, /*PIC=*/false)); 1664 Ctx.setObjectFileInfo(MOFI.get()); 1665 1666 std::unique_ptr<MCDisassembler> DisAsm( 1667 TheTarget->createMCDisassembler(*STI, Ctx)); 1668 if (!DisAsm) 1669 reportError(Obj->getFileName(), "no disassembler for target " + TripleName); 1670 1671 // If we have an ARM object file, we need a second disassembler, because 1672 // ARM CPUs have two different instruction sets: ARM mode, and Thumb mode. 1673 // We use mapping symbols to switch between the two assemblers, where 1674 // appropriate. 1675 std::unique_ptr<MCDisassembler> SecondaryDisAsm; 1676 std::unique_ptr<const MCSubtargetInfo> SecondarySTI; 1677 if (isArmElf(Obj) && !STI->checkFeatures("+mclass")) { 1678 if (STI->checkFeatures("+thumb-mode")) 1679 Features.AddFeature("-thumb-mode"); 1680 else 1681 Features.AddFeature("+thumb-mode"); 1682 SecondarySTI.reset(TheTarget->createMCSubtargetInfo(TripleName, MCPU, 1683 Features.getString())); 1684 SecondaryDisAsm.reset(TheTarget->createMCDisassembler(*SecondarySTI, Ctx)); 1685 } 1686 1687 std::unique_ptr<const MCInstrAnalysis> MIA( 1688 TheTarget->createMCInstrAnalysis(MII.get())); 1689 1690 int AsmPrinterVariant = AsmInfo->getAssemblerDialect(); 1691 std::unique_ptr<MCInstPrinter> IP(TheTarget->createMCInstPrinter( 1692 Triple(TripleName), AsmPrinterVariant, *AsmInfo, *MII, *MRI)); 1693 if (!IP) 1694 reportError(Obj->getFileName(), 1695 "no instruction printer for target " + TripleName); 1696 IP->setPrintImmHex(PrintImmHex); 1697 IP->setPrintBranchImmAsAddress(true); 1698 IP->setSymbolizeOperands(SymbolizeOperands); 1699 IP->setMCInstrAnalysis(MIA.get()); 1700 1701 PrettyPrinter &PIP = selectPrettyPrinter(Triple(TripleName)); 1702 SourcePrinter SP(Obj, TheTarget->getName()); 1703 1704 for (StringRef Opt : DisassemblerOptions) 1705 if (!IP->applyTargetSpecificCLOption(Opt)) 1706 reportError(Obj->getFileName(), 1707 "Unrecognized disassembler option: " + Opt); 1708 1709 disassembleObject(TheTarget, Obj, Ctx, DisAsm.get(), SecondaryDisAsm.get(), 1710 MIA.get(), IP.get(), STI.get(), SecondarySTI.get(), PIP, 1711 SP, InlineRelocs); 1712 } 1713 1714 void objdump::printRelocations(const ObjectFile *Obj) { 1715 StringRef Fmt = Obj->getBytesInAddress() > 4 ? "%016" PRIx64 : 1716 "%08" PRIx64; 1717 // Regular objdump doesn't print relocations in non-relocatable object 1718 // files. 1719 if (!Obj->isRelocatableObject()) 1720 return; 1721 1722 // Build a mapping from relocation target to a vector of relocation 1723 // sections. Usually, there is an only one relocation section for 1724 // each relocated section. 1725 MapVector<SectionRef, std::vector<SectionRef>> SecToRelSec; 1726 uint64_t Ndx; 1727 for (const SectionRef &Section : ToolSectionFilter(*Obj, &Ndx)) { 1728 if (Section.relocation_begin() == Section.relocation_end()) 1729 continue; 1730 Expected<section_iterator> SecOrErr = Section.getRelocatedSection(); 1731 if (!SecOrErr) 1732 reportError(Obj->getFileName(), 1733 "section (" + Twine(Ndx) + 1734 "): unable to get a relocation target: " + 1735 toString(SecOrErr.takeError())); 1736 SecToRelSec[**SecOrErr].push_back(Section); 1737 } 1738 1739 for (std::pair<SectionRef, std::vector<SectionRef>> &P : SecToRelSec) { 1740 StringRef SecName = unwrapOrError(P.first.getName(), Obj->getFileName()); 1741 outs() << "\nRELOCATION RECORDS FOR [" << SecName << "]:\n"; 1742 uint32_t OffsetPadding = (Obj->getBytesInAddress() > 4 ? 16 : 8); 1743 uint32_t TypePadding = 24; 1744 outs() << left_justify("OFFSET", OffsetPadding) << " " 1745 << left_justify("TYPE", TypePadding) << " " 1746 << "VALUE\n"; 1747 1748 for (SectionRef Section : P.second) { 1749 for (const RelocationRef &Reloc : Section.relocations()) { 1750 uint64_t Address = Reloc.getOffset(); 1751 SmallString<32> RelocName; 1752 SmallString<32> ValueStr; 1753 if (Address < StartAddress || Address > StopAddress || getHidden(Reloc)) 1754 continue; 1755 Reloc.getTypeName(RelocName); 1756 if (Error E = getRelocationValueString(Reloc, ValueStr)) 1757 reportError(std::move(E), Obj->getFileName()); 1758 1759 outs() << format(Fmt.data(), Address) << " " 1760 << left_justify(RelocName, TypePadding) << " " << ValueStr 1761 << "\n"; 1762 } 1763 } 1764 } 1765 } 1766 1767 void objdump::printDynamicRelocations(const ObjectFile *Obj) { 1768 // For the moment, this option is for ELF only 1769 if (!Obj->isELF()) 1770 return; 1771 1772 const auto *Elf = dyn_cast<ELFObjectFileBase>(Obj); 1773 if (!Elf || Elf->getEType() != ELF::ET_DYN) { 1774 reportError(Obj->getFileName(), "not a dynamic object"); 1775 return; 1776 } 1777 1778 std::vector<SectionRef> DynRelSec = Obj->dynamic_relocation_sections(); 1779 if (DynRelSec.empty()) 1780 return; 1781 1782 outs() << "DYNAMIC RELOCATION RECORDS\n"; 1783 StringRef Fmt = Obj->getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 1784 for (const SectionRef &Section : DynRelSec) 1785 for (const RelocationRef &Reloc : Section.relocations()) { 1786 uint64_t Address = Reloc.getOffset(); 1787 SmallString<32> RelocName; 1788 SmallString<32> ValueStr; 1789 Reloc.getTypeName(RelocName); 1790 if (Error E = getRelocationValueString(Reloc, ValueStr)) 1791 reportError(std::move(E), Obj->getFileName()); 1792 outs() << format(Fmt.data(), Address) << " " << RelocName << " " 1793 << ValueStr << "\n"; 1794 } 1795 } 1796 1797 // Returns true if we need to show LMA column when dumping section headers. We 1798 // show it only when the platform is ELF and either we have at least one section 1799 // whose VMA and LMA are different and/or when --show-lma flag is used. 1800 static bool shouldDisplayLMA(const ObjectFile *Obj) { 1801 if (!Obj->isELF()) 1802 return false; 1803 for (const SectionRef &S : ToolSectionFilter(*Obj)) 1804 if (S.getAddress() != getELFSectionLMA(S)) 1805 return true; 1806 return ShowLMA; 1807 } 1808 1809 static size_t getMaxSectionNameWidth(const ObjectFile *Obj) { 1810 // Default column width for names is 13 even if no names are that long. 1811 size_t MaxWidth = 13; 1812 for (const SectionRef &Section : ToolSectionFilter(*Obj)) { 1813 StringRef Name = unwrapOrError(Section.getName(), Obj->getFileName()); 1814 MaxWidth = std::max(MaxWidth, Name.size()); 1815 } 1816 return MaxWidth; 1817 } 1818 1819 void objdump::printSectionHeaders(const ObjectFile *Obj) { 1820 size_t NameWidth = getMaxSectionNameWidth(Obj); 1821 size_t AddressWidth = 2 * Obj->getBytesInAddress(); 1822 bool HasLMAColumn = shouldDisplayLMA(Obj); 1823 outs() << "\nSections:\n"; 1824 if (HasLMAColumn) 1825 outs() << "Idx " << left_justify("Name", NameWidth) << " Size " 1826 << left_justify("VMA", AddressWidth) << " " 1827 << left_justify("LMA", AddressWidth) << " Type\n"; 1828 else 1829 outs() << "Idx " << left_justify("Name", NameWidth) << " Size " 1830 << left_justify("VMA", AddressWidth) << " Type\n"; 1831 1832 uint64_t Idx; 1833 for (const SectionRef &Section : ToolSectionFilter(*Obj, &Idx)) { 1834 StringRef Name = unwrapOrError(Section.getName(), Obj->getFileName()); 1835 uint64_t VMA = Section.getAddress(); 1836 if (shouldAdjustVA(Section)) 1837 VMA += AdjustVMA; 1838 1839 uint64_t Size = Section.getSize(); 1840 1841 std::string Type = Section.isText() ? "TEXT" : ""; 1842 if (Section.isData()) 1843 Type += Type.empty() ? "DATA" : ", DATA"; 1844 if (Section.isBSS()) 1845 Type += Type.empty() ? "BSS" : ", BSS"; 1846 if (Section.isDebugSection()) 1847 Type += Type.empty() ? "DEBUG" : ", DEBUG"; 1848 1849 if (HasLMAColumn) 1850 outs() << format("%3" PRIu64 " %-*s %08" PRIx64 " ", Idx, NameWidth, 1851 Name.str().c_str(), Size) 1852 << format_hex_no_prefix(VMA, AddressWidth) << " " 1853 << format_hex_no_prefix(getELFSectionLMA(Section), AddressWidth) 1854 << " " << Type << "\n"; 1855 else 1856 outs() << format("%3" PRIu64 " %-*s %08" PRIx64 " ", Idx, NameWidth, 1857 Name.str().c_str(), Size) 1858 << format_hex_no_prefix(VMA, AddressWidth) << " " << Type << "\n"; 1859 } 1860 } 1861 1862 void objdump::printSectionContents(const ObjectFile *Obj) { 1863 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(Obj); 1864 1865 for (const SectionRef &Section : ToolSectionFilter(*Obj)) { 1866 StringRef Name = unwrapOrError(Section.getName(), Obj->getFileName()); 1867 uint64_t BaseAddr = Section.getAddress(); 1868 uint64_t Size = Section.getSize(); 1869 if (!Size) 1870 continue; 1871 1872 outs() << "Contents of section "; 1873 StringRef SegmentName = getSegmentName(MachO, Section); 1874 if (!SegmentName.empty()) 1875 outs() << SegmentName << ","; 1876 outs() << Name << ":\n"; 1877 if (Section.isBSS()) { 1878 outs() << format("<skipping contents of bss section at [%04" PRIx64 1879 ", %04" PRIx64 ")>\n", 1880 BaseAddr, BaseAddr + Size); 1881 continue; 1882 } 1883 1884 StringRef Contents = unwrapOrError(Section.getContents(), Obj->getFileName()); 1885 1886 // Dump out the content as hex and printable ascii characters. 1887 for (std::size_t Addr = 0, End = Contents.size(); Addr < End; Addr += 16) { 1888 outs() << format(" %04" PRIx64 " ", BaseAddr + Addr); 1889 // Dump line of hex. 1890 for (std::size_t I = 0; I < 16; ++I) { 1891 if (I != 0 && I % 4 == 0) 1892 outs() << ' '; 1893 if (Addr + I < End) 1894 outs() << hexdigit((Contents[Addr + I] >> 4) & 0xF, true) 1895 << hexdigit(Contents[Addr + I] & 0xF, true); 1896 else 1897 outs() << " "; 1898 } 1899 // Print ascii. 1900 outs() << " "; 1901 for (std::size_t I = 0; I < 16 && Addr + I < End; ++I) { 1902 if (isPrint(static_cast<unsigned char>(Contents[Addr + I]) & 0xFF)) 1903 outs() << Contents[Addr + I]; 1904 else 1905 outs() << "."; 1906 } 1907 outs() << "\n"; 1908 } 1909 } 1910 } 1911 1912 void objdump::printSymbolTable(const ObjectFile *O, StringRef ArchiveName, 1913 StringRef ArchitectureName, bool DumpDynamic) { 1914 if (O->isCOFF() && !DumpDynamic) { 1915 outs() << "\nSYMBOL TABLE:\n"; 1916 printCOFFSymbolTable(cast<const COFFObjectFile>(O)); 1917 return; 1918 } 1919 1920 const StringRef FileName = O->getFileName(); 1921 1922 if (!DumpDynamic) { 1923 outs() << "\nSYMBOL TABLE:\n"; 1924 for (auto I = O->symbol_begin(); I != O->symbol_end(); ++I) 1925 printSymbol(O, *I, FileName, ArchiveName, ArchitectureName, DumpDynamic); 1926 return; 1927 } 1928 1929 outs() << "\nDYNAMIC SYMBOL TABLE:\n"; 1930 if (!O->isELF()) { 1931 reportWarning( 1932 "this operation is not currently supported for this file format", 1933 FileName); 1934 return; 1935 } 1936 1937 const ELFObjectFileBase *ELF = cast<const ELFObjectFileBase>(O); 1938 for (auto I = ELF->getDynamicSymbolIterators().begin(); 1939 I != ELF->getDynamicSymbolIterators().end(); ++I) 1940 printSymbol(O, *I, FileName, ArchiveName, ArchitectureName, DumpDynamic); 1941 } 1942 1943 void objdump::printSymbol(const ObjectFile *O, const SymbolRef &Symbol, 1944 StringRef FileName, StringRef ArchiveName, 1945 StringRef ArchitectureName, bool DumpDynamic) { 1946 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(O); 1947 uint64_t Address = unwrapOrError(Symbol.getAddress(), FileName, ArchiveName, 1948 ArchitectureName); 1949 if ((Address < StartAddress) || (Address > StopAddress)) 1950 return; 1951 SymbolRef::Type Type = 1952 unwrapOrError(Symbol.getType(), FileName, ArchiveName, ArchitectureName); 1953 uint32_t Flags = 1954 unwrapOrError(Symbol.getFlags(), FileName, ArchiveName, ArchitectureName); 1955 1956 // Don't ask a Mach-O STAB symbol for its section unless you know that 1957 // STAB symbol's section field refers to a valid section index. Otherwise 1958 // the symbol may error trying to load a section that does not exist. 1959 bool IsSTAB = false; 1960 if (MachO) { 1961 DataRefImpl SymDRI = Symbol.getRawDataRefImpl(); 1962 uint8_t NType = 1963 (MachO->is64Bit() ? MachO->getSymbol64TableEntry(SymDRI).n_type 1964 : MachO->getSymbolTableEntry(SymDRI).n_type); 1965 if (NType & MachO::N_STAB) 1966 IsSTAB = true; 1967 } 1968 section_iterator Section = IsSTAB 1969 ? O->section_end() 1970 : unwrapOrError(Symbol.getSection(), FileName, 1971 ArchiveName, ArchitectureName); 1972 1973 StringRef Name; 1974 if (Type == SymbolRef::ST_Debug && Section != O->section_end()) { 1975 if (Expected<StringRef> NameOrErr = Section->getName()) 1976 Name = *NameOrErr; 1977 else 1978 consumeError(NameOrErr.takeError()); 1979 1980 } else { 1981 Name = unwrapOrError(Symbol.getName(), FileName, ArchiveName, 1982 ArchitectureName); 1983 } 1984 1985 bool Global = Flags & SymbolRef::SF_Global; 1986 bool Weak = Flags & SymbolRef::SF_Weak; 1987 bool Absolute = Flags & SymbolRef::SF_Absolute; 1988 bool Common = Flags & SymbolRef::SF_Common; 1989 bool Hidden = Flags & SymbolRef::SF_Hidden; 1990 1991 char GlobLoc = ' '; 1992 if ((Section != O->section_end() || Absolute) && !Weak) 1993 GlobLoc = Global ? 'g' : 'l'; 1994 char IFunc = ' '; 1995 if (O->isELF()) { 1996 if (ELFSymbolRef(Symbol).getELFType() == ELF::STT_GNU_IFUNC) 1997 IFunc = 'i'; 1998 if (ELFSymbolRef(Symbol).getBinding() == ELF::STB_GNU_UNIQUE) 1999 GlobLoc = 'u'; 2000 } 2001 2002 char Debug = ' '; 2003 if (DumpDynamic) 2004 Debug = 'D'; 2005 else if (Type == SymbolRef::ST_Debug || Type == SymbolRef::ST_File) 2006 Debug = 'd'; 2007 2008 char FileFunc = ' '; 2009 if (Type == SymbolRef::ST_File) 2010 FileFunc = 'f'; 2011 else if (Type == SymbolRef::ST_Function) 2012 FileFunc = 'F'; 2013 else if (Type == SymbolRef::ST_Data) 2014 FileFunc = 'O'; 2015 2016 const char *Fmt = O->getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 2017 2018 outs() << format(Fmt, Address) << " " 2019 << GlobLoc // Local -> 'l', Global -> 'g', Neither -> ' ' 2020 << (Weak ? 'w' : ' ') // Weak? 2021 << ' ' // Constructor. Not supported yet. 2022 << ' ' // Warning. Not supported yet. 2023 << IFunc // Indirect reference to another symbol. 2024 << Debug // Debugging (d) or dynamic (D) symbol. 2025 << FileFunc // Name of function (F), file (f) or object (O). 2026 << ' '; 2027 if (Absolute) { 2028 outs() << "*ABS*"; 2029 } else if (Common) { 2030 outs() << "*COM*"; 2031 } else if (Section == O->section_end()) { 2032 outs() << "*UND*"; 2033 } else { 2034 StringRef SegmentName = getSegmentName(MachO, *Section); 2035 if (!SegmentName.empty()) 2036 outs() << SegmentName << ","; 2037 StringRef SectionName = unwrapOrError(Section->getName(), FileName); 2038 outs() << SectionName; 2039 } 2040 2041 if (Common || O->isELF()) { 2042 uint64_t Val = 2043 Common ? Symbol.getAlignment() : ELFSymbolRef(Symbol).getSize(); 2044 outs() << '\t' << format(Fmt, Val); 2045 } 2046 2047 if (O->isELF()) { 2048 uint8_t Other = ELFSymbolRef(Symbol).getOther(); 2049 switch (Other) { 2050 case ELF::STV_DEFAULT: 2051 break; 2052 case ELF::STV_INTERNAL: 2053 outs() << " .internal"; 2054 break; 2055 case ELF::STV_HIDDEN: 2056 outs() << " .hidden"; 2057 break; 2058 case ELF::STV_PROTECTED: 2059 outs() << " .protected"; 2060 break; 2061 default: 2062 outs() << format(" 0x%02x", Other); 2063 break; 2064 } 2065 } else if (Hidden) { 2066 outs() << " .hidden"; 2067 } 2068 2069 if (Demangle) 2070 outs() << ' ' << demangle(std::string(Name)) << '\n'; 2071 else 2072 outs() << ' ' << Name << '\n'; 2073 } 2074 2075 static void printUnwindInfo(const ObjectFile *O) { 2076 outs() << "Unwind info:\n\n"; 2077 2078 if (const COFFObjectFile *Coff = dyn_cast<COFFObjectFile>(O)) 2079 printCOFFUnwindInfo(Coff); 2080 else if (const MachOObjectFile *MachO = dyn_cast<MachOObjectFile>(O)) 2081 printMachOUnwindInfo(MachO); 2082 else 2083 // TODO: Extract DWARF dump tool to objdump. 2084 WithColor::error(errs(), ToolName) 2085 << "This operation is only currently supported " 2086 "for COFF and MachO object files.\n"; 2087 } 2088 2089 /// Dump the raw contents of the __clangast section so the output can be piped 2090 /// into llvm-bcanalyzer. 2091 static void printRawClangAST(const ObjectFile *Obj) { 2092 if (outs().is_displayed()) { 2093 WithColor::error(errs(), ToolName) 2094 << "The -raw-clang-ast option will dump the raw binary contents of " 2095 "the clang ast section.\n" 2096 "Please redirect the output to a file or another program such as " 2097 "llvm-bcanalyzer.\n"; 2098 return; 2099 } 2100 2101 StringRef ClangASTSectionName("__clangast"); 2102 if (Obj->isCOFF()) { 2103 ClangASTSectionName = "clangast"; 2104 } 2105 2106 Optional<object::SectionRef> ClangASTSection; 2107 for (auto Sec : ToolSectionFilter(*Obj)) { 2108 StringRef Name; 2109 if (Expected<StringRef> NameOrErr = Sec.getName()) 2110 Name = *NameOrErr; 2111 else 2112 consumeError(NameOrErr.takeError()); 2113 2114 if (Name == ClangASTSectionName) { 2115 ClangASTSection = Sec; 2116 break; 2117 } 2118 } 2119 if (!ClangASTSection) 2120 return; 2121 2122 StringRef ClangASTContents = unwrapOrError( 2123 ClangASTSection.getValue().getContents(), Obj->getFileName()); 2124 outs().write(ClangASTContents.data(), ClangASTContents.size()); 2125 } 2126 2127 static void printFaultMaps(const ObjectFile *Obj) { 2128 StringRef FaultMapSectionName; 2129 2130 if (Obj->isELF()) { 2131 FaultMapSectionName = ".llvm_faultmaps"; 2132 } else if (Obj->isMachO()) { 2133 FaultMapSectionName = "__llvm_faultmaps"; 2134 } else { 2135 WithColor::error(errs(), ToolName) 2136 << "This operation is only currently supported " 2137 "for ELF and Mach-O executable files.\n"; 2138 return; 2139 } 2140 2141 Optional<object::SectionRef> FaultMapSection; 2142 2143 for (auto Sec : ToolSectionFilter(*Obj)) { 2144 StringRef Name; 2145 if (Expected<StringRef> NameOrErr = Sec.getName()) 2146 Name = *NameOrErr; 2147 else 2148 consumeError(NameOrErr.takeError()); 2149 2150 if (Name == FaultMapSectionName) { 2151 FaultMapSection = Sec; 2152 break; 2153 } 2154 } 2155 2156 outs() << "FaultMap table:\n"; 2157 2158 if (!FaultMapSection.hasValue()) { 2159 outs() << "<not found>\n"; 2160 return; 2161 } 2162 2163 StringRef FaultMapContents = 2164 unwrapOrError(FaultMapSection.getValue().getContents(), Obj->getFileName()); 2165 FaultMapParser FMP(FaultMapContents.bytes_begin(), 2166 FaultMapContents.bytes_end()); 2167 2168 outs() << FMP; 2169 } 2170 2171 static void printPrivateFileHeaders(const ObjectFile *O, bool OnlyFirst) { 2172 if (O->isELF()) { 2173 printELFFileHeader(O); 2174 printELFDynamicSection(O); 2175 printELFSymbolVersionInfo(O); 2176 return; 2177 } 2178 if (O->isCOFF()) 2179 return printCOFFFileHeader(O); 2180 if (O->isWasm()) 2181 return printWasmFileHeader(O); 2182 if (O->isMachO()) { 2183 printMachOFileHeader(O); 2184 if (!OnlyFirst) 2185 printMachOLoadCommands(O); 2186 return; 2187 } 2188 reportError(O->getFileName(), "Invalid/Unsupported object file format"); 2189 } 2190 2191 static void printFileHeaders(const ObjectFile *O) { 2192 if (!O->isELF() && !O->isCOFF()) 2193 reportError(O->getFileName(), "Invalid/Unsupported object file format"); 2194 2195 Triple::ArchType AT = O->getArch(); 2196 outs() << "architecture: " << Triple::getArchTypeName(AT) << "\n"; 2197 uint64_t Address = unwrapOrError(O->getStartAddress(), O->getFileName()); 2198 2199 StringRef Fmt = O->getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 2200 outs() << "start address: " 2201 << "0x" << format(Fmt.data(), Address) << "\n"; 2202 } 2203 2204 static void printArchiveChild(StringRef Filename, const Archive::Child &C) { 2205 Expected<sys::fs::perms> ModeOrErr = C.getAccessMode(); 2206 if (!ModeOrErr) { 2207 WithColor::error(errs(), ToolName) << "ill-formed archive entry.\n"; 2208 consumeError(ModeOrErr.takeError()); 2209 return; 2210 } 2211 sys::fs::perms Mode = ModeOrErr.get(); 2212 outs() << ((Mode & sys::fs::owner_read) ? "r" : "-"); 2213 outs() << ((Mode & sys::fs::owner_write) ? "w" : "-"); 2214 outs() << ((Mode & sys::fs::owner_exe) ? "x" : "-"); 2215 outs() << ((Mode & sys::fs::group_read) ? "r" : "-"); 2216 outs() << ((Mode & sys::fs::group_write) ? "w" : "-"); 2217 outs() << ((Mode & sys::fs::group_exe) ? "x" : "-"); 2218 outs() << ((Mode & sys::fs::others_read) ? "r" : "-"); 2219 outs() << ((Mode & sys::fs::others_write) ? "w" : "-"); 2220 outs() << ((Mode & sys::fs::others_exe) ? "x" : "-"); 2221 2222 outs() << " "; 2223 2224 outs() << format("%d/%d %6" PRId64 " ", unwrapOrError(C.getUID(), Filename), 2225 unwrapOrError(C.getGID(), Filename), 2226 unwrapOrError(C.getRawSize(), Filename)); 2227 2228 StringRef RawLastModified = C.getRawLastModified(); 2229 unsigned Seconds; 2230 if (RawLastModified.getAsInteger(10, Seconds)) 2231 outs() << "(date: \"" << RawLastModified 2232 << "\" contains non-decimal chars) "; 2233 else { 2234 // Since ctime(3) returns a 26 character string of the form: 2235 // "Sun Sep 16 01:03:52 1973\n\0" 2236 // just print 24 characters. 2237 time_t t = Seconds; 2238 outs() << format("%.24s ", ctime(&t)); 2239 } 2240 2241 StringRef Name = ""; 2242 Expected<StringRef> NameOrErr = C.getName(); 2243 if (!NameOrErr) { 2244 consumeError(NameOrErr.takeError()); 2245 Name = unwrapOrError(C.getRawName(), Filename); 2246 } else { 2247 Name = NameOrErr.get(); 2248 } 2249 outs() << Name << "\n"; 2250 } 2251 2252 // For ELF only now. 2253 static bool shouldWarnForInvalidStartStopAddress(ObjectFile *Obj) { 2254 if (const auto *Elf = dyn_cast<ELFObjectFileBase>(Obj)) { 2255 if (Elf->getEType() != ELF::ET_REL) 2256 return true; 2257 } 2258 return false; 2259 } 2260 2261 static void checkForInvalidStartStopAddress(ObjectFile *Obj, 2262 uint64_t Start, uint64_t Stop) { 2263 if (!shouldWarnForInvalidStartStopAddress(Obj)) 2264 return; 2265 2266 for (const SectionRef &Section : Obj->sections()) 2267 if (ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC) { 2268 uint64_t BaseAddr = Section.getAddress(); 2269 uint64_t Size = Section.getSize(); 2270 if ((Start < BaseAddr + Size) && Stop > BaseAddr) 2271 return; 2272 } 2273 2274 if (!HasStartAddressFlag) 2275 reportWarning("no section has address less than 0x" + 2276 Twine::utohexstr(Stop) + " specified by --stop-address", 2277 Obj->getFileName()); 2278 else if (!HasStopAddressFlag) 2279 reportWarning("no section has address greater than or equal to 0x" + 2280 Twine::utohexstr(Start) + " specified by --start-address", 2281 Obj->getFileName()); 2282 else 2283 reportWarning("no section overlaps the range [0x" + 2284 Twine::utohexstr(Start) + ",0x" + Twine::utohexstr(Stop) + 2285 ") specified by --start-address/--stop-address", 2286 Obj->getFileName()); 2287 } 2288 2289 static void dumpObject(ObjectFile *O, const Archive *A = nullptr, 2290 const Archive::Child *C = nullptr) { 2291 // Avoid other output when using a raw option. 2292 if (!RawClangAST) { 2293 outs() << '\n'; 2294 if (A) 2295 outs() << A->getFileName() << "(" << O->getFileName() << ")"; 2296 else 2297 outs() << O->getFileName(); 2298 outs() << ":\tfile format " << O->getFileFormatName().lower() << "\n"; 2299 } 2300 2301 if (HasStartAddressFlag || HasStopAddressFlag) 2302 checkForInvalidStartStopAddress(O, StartAddress, StopAddress); 2303 2304 // Note: the order here matches GNU objdump for compatability. 2305 StringRef ArchiveName = A ? A->getFileName() : ""; 2306 if (ArchiveHeaders && !MachOOpt && C) 2307 printArchiveChild(ArchiveName, *C); 2308 if (FileHeaders) 2309 printFileHeaders(O); 2310 if (PrivateHeaders || FirstPrivateHeader) 2311 printPrivateFileHeaders(O, FirstPrivateHeader); 2312 if (SectionHeaders) 2313 printSectionHeaders(O); 2314 if (SymbolTable) 2315 printSymbolTable(O, ArchiveName); 2316 if (DynamicSymbolTable) 2317 printSymbolTable(O, ArchiveName, /*ArchitectureName=*/"", 2318 /*DumpDynamic=*/true); 2319 if (DwarfDumpType != DIDT_Null) { 2320 std::unique_ptr<DIContext> DICtx = DWARFContext::create(*O); 2321 // Dump the complete DWARF structure. 2322 DIDumpOptions DumpOpts; 2323 DumpOpts.DumpType = DwarfDumpType; 2324 DICtx->dump(outs(), DumpOpts); 2325 } 2326 if (Relocations && !Disassemble) 2327 printRelocations(O); 2328 if (DynamicRelocations) 2329 printDynamicRelocations(O); 2330 if (SectionContents) 2331 printSectionContents(O); 2332 if (Disassemble) 2333 disassembleObject(O, Relocations); 2334 if (UnwindInfo) 2335 printUnwindInfo(O); 2336 2337 // Mach-O specific options: 2338 if (ExportsTrie) 2339 printExportsTrie(O); 2340 if (Rebase) 2341 printRebaseTable(O); 2342 if (Bind) 2343 printBindTable(O); 2344 if (LazyBind) 2345 printLazyBindTable(O); 2346 if (WeakBind) 2347 printWeakBindTable(O); 2348 2349 // Other special sections: 2350 if (RawClangAST) 2351 printRawClangAST(O); 2352 if (FaultMapSection) 2353 printFaultMaps(O); 2354 } 2355 2356 static void dumpObject(const COFFImportFile *I, const Archive *A, 2357 const Archive::Child *C = nullptr) { 2358 StringRef ArchiveName = A ? A->getFileName() : ""; 2359 2360 // Avoid other output when using a raw option. 2361 if (!RawClangAST) 2362 outs() << '\n' 2363 << ArchiveName << "(" << I->getFileName() << ")" 2364 << ":\tfile format COFF-import-file" 2365 << "\n\n"; 2366 2367 if (ArchiveHeaders && !MachOOpt && C) 2368 printArchiveChild(ArchiveName, *C); 2369 if (SymbolTable) 2370 printCOFFSymbolTable(I); 2371 } 2372 2373 /// Dump each object file in \a a; 2374 static void dumpArchive(const Archive *A) { 2375 Error Err = Error::success(); 2376 unsigned I = -1; 2377 for (auto &C : A->children(Err)) { 2378 ++I; 2379 Expected<std::unique_ptr<Binary>> ChildOrErr = C.getAsBinary(); 2380 if (!ChildOrErr) { 2381 if (auto E = isNotObjectErrorInvalidFileType(ChildOrErr.takeError())) 2382 reportError(std::move(E), getFileNameForError(C, I), A->getFileName()); 2383 continue; 2384 } 2385 if (ObjectFile *O = dyn_cast<ObjectFile>(&*ChildOrErr.get())) 2386 dumpObject(O, A, &C); 2387 else if (COFFImportFile *I = dyn_cast<COFFImportFile>(&*ChildOrErr.get())) 2388 dumpObject(I, A, &C); 2389 else 2390 reportError(errorCodeToError(object_error::invalid_file_type), 2391 A->getFileName()); 2392 } 2393 if (Err) 2394 reportError(std::move(Err), A->getFileName()); 2395 } 2396 2397 /// Open file and figure out how to dump it. 2398 static void dumpInput(StringRef file) { 2399 // If we are using the Mach-O specific object file parser, then let it parse 2400 // the file and process the command line options. So the -arch flags can 2401 // be used to select specific slices, etc. 2402 if (MachOOpt) { 2403 parseInputMachO(file); 2404 return; 2405 } 2406 2407 // Attempt to open the binary. 2408 OwningBinary<Binary> OBinary = unwrapOrError(createBinary(file), file); 2409 Binary &Binary = *OBinary.getBinary(); 2410 2411 if (Archive *A = dyn_cast<Archive>(&Binary)) 2412 dumpArchive(A); 2413 else if (ObjectFile *O = dyn_cast<ObjectFile>(&Binary)) 2414 dumpObject(O); 2415 else if (MachOUniversalBinary *UB = dyn_cast<MachOUniversalBinary>(&Binary)) 2416 parseInputMachO(UB); 2417 else 2418 reportError(errorCodeToError(object_error::invalid_file_type), file); 2419 } 2420 2421 template <typename T> 2422 static void parseIntArg(const llvm::opt::InputArgList &InputArgs, int ID, 2423 T &Value) { 2424 if (const opt::Arg *A = InputArgs.getLastArg(ID)) { 2425 StringRef V(A->getValue()); 2426 if (!llvm::to_integer(V, Value, 0)) { 2427 reportCmdLineError(A->getSpelling() + 2428 ": expected a non-negative integer, but got '" + V + 2429 "'"); 2430 } 2431 } 2432 } 2433 2434 static std::vector<std::string> 2435 commaSeparatedValues(const llvm::opt::InputArgList &InputArgs, int ID) { 2436 std::vector<std::string> Values; 2437 for (StringRef Value : InputArgs.getAllArgValues(ID)) { 2438 llvm::SmallVector<StringRef, 2> SplitValues; 2439 llvm::SplitString(Value, SplitValues, ","); 2440 for (StringRef SplitValue : SplitValues) 2441 Values.push_back(SplitValue.str()); 2442 } 2443 return Values; 2444 } 2445 2446 static void parseOtoolOptions(const llvm::opt::InputArgList &InputArgs) { 2447 MachOOpt = true; 2448 FullLeadingAddr = true; 2449 PrintImmHex = true; 2450 2451 ArchName = InputArgs.getLastArgValue(OTOOL_arch).str(); 2452 LinkOptHints = InputArgs.hasArg(OTOOL_C); 2453 if (InputArgs.hasArg(OTOOL_d)) 2454 FilterSections.push_back("__DATA,__data"); 2455 DylibId = InputArgs.hasArg(OTOOL_D); 2456 UniversalHeaders = InputArgs.hasArg(OTOOL_f); 2457 DataInCode = InputArgs.hasArg(OTOOL_G); 2458 FirstPrivateHeader = InputArgs.hasArg(OTOOL_h); 2459 IndirectSymbols = InputArgs.hasArg(OTOOL_I); 2460 ShowRawInsn = InputArgs.hasArg(OTOOL_j); 2461 PrivateHeaders = InputArgs.hasArg(OTOOL_l); 2462 DylibsUsed = InputArgs.hasArg(OTOOL_L); 2463 MCPU = InputArgs.getLastArgValue(OTOOL_mcpu_EQ).str(); 2464 ObjcMetaData = InputArgs.hasArg(OTOOL_o); 2465 DisSymName = InputArgs.getLastArgValue(OTOOL_p).str(); 2466 InfoPlist = InputArgs.hasArg(OTOOL_P); 2467 Relocations = InputArgs.hasArg(OTOOL_r); 2468 if (const Arg *A = InputArgs.getLastArg(OTOOL_s)) { 2469 auto Filter = (A->getValue(0) + StringRef(",") + A->getValue(1)).str(); 2470 FilterSections.push_back(Filter); 2471 } 2472 if (InputArgs.hasArg(OTOOL_t)) 2473 FilterSections.push_back("__TEXT,__text"); 2474 Verbose = InputArgs.hasArg(OTOOL_v) || InputArgs.hasArg(OTOOL_V) || 2475 InputArgs.hasArg(OTOOL_o); 2476 SymbolicOperands = InputArgs.hasArg(OTOOL_V); 2477 if (InputArgs.hasArg(OTOOL_x)) 2478 FilterSections.push_back(",__text"); 2479 LeadingAddr = LeadingHeaders = !InputArgs.hasArg(OTOOL_X); 2480 2481 InputFilenames = InputArgs.getAllArgValues(OTOOL_INPUT); 2482 if (InputFilenames.empty()) 2483 reportCmdLineError("no input file"); 2484 2485 for (const Arg *A : InputArgs) { 2486 const Option &O = A->getOption(); 2487 if (O.getGroup().isValid() && O.getGroup().getID() == OTOOL_grp_obsolete) { 2488 reportCmdLineWarning(O.getPrefixedName() + 2489 " is obsolete and not implemented"); 2490 } 2491 } 2492 } 2493 2494 static void parseObjdumpOptions(const llvm::opt::InputArgList &InputArgs) { 2495 parseIntArg(InputArgs, OBJDUMP_adjust_vma_EQ, AdjustVMA); 2496 AllHeaders = InputArgs.hasArg(OBJDUMP_all_headers); 2497 ArchName = InputArgs.getLastArgValue(OBJDUMP_arch_name_EQ).str(); 2498 ArchiveHeaders = InputArgs.hasArg(OBJDUMP_archive_headers); 2499 Demangle = InputArgs.hasArg(OBJDUMP_demangle); 2500 Disassemble = InputArgs.hasArg(OBJDUMP_disassemble); 2501 DisassembleAll = InputArgs.hasArg(OBJDUMP_disassemble_all); 2502 SymbolDescription = InputArgs.hasArg(OBJDUMP_symbol_description); 2503 DisassembleSymbols = 2504 commaSeparatedValues(InputArgs, OBJDUMP_disassemble_symbols_EQ); 2505 DisassembleZeroes = InputArgs.hasArg(OBJDUMP_disassemble_zeroes); 2506 if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_dwarf_EQ)) { 2507 DwarfDumpType = 2508 StringSwitch<DIDumpType>(A->getValue()).Case("frames", DIDT_DebugFrame); 2509 } 2510 DynamicRelocations = InputArgs.hasArg(OBJDUMP_dynamic_reloc); 2511 FaultMapSection = InputArgs.hasArg(OBJDUMP_fault_map_section); 2512 FileHeaders = InputArgs.hasArg(OBJDUMP_file_headers); 2513 SectionContents = InputArgs.hasArg(OBJDUMP_full_contents); 2514 PrintLines = InputArgs.hasArg(OBJDUMP_line_numbers); 2515 InputFilenames = InputArgs.getAllArgValues(OBJDUMP_INPUT); 2516 MachOOpt = InputArgs.hasArg(OBJDUMP_macho); 2517 MCPU = InputArgs.getLastArgValue(OBJDUMP_mcpu_EQ).str(); 2518 MAttrs = commaSeparatedValues(InputArgs, OBJDUMP_mattr_EQ); 2519 ShowRawInsn = !InputArgs.hasArg(OBJDUMP_no_show_raw_insn); 2520 LeadingAddr = !InputArgs.hasArg(OBJDUMP_no_leading_addr); 2521 RawClangAST = InputArgs.hasArg(OBJDUMP_raw_clang_ast); 2522 Relocations = InputArgs.hasArg(OBJDUMP_reloc); 2523 PrintImmHex = 2524 InputArgs.hasFlag(OBJDUMP_print_imm_hex, OBJDUMP_no_print_imm_hex, false); 2525 PrivateHeaders = InputArgs.hasArg(OBJDUMP_private_headers); 2526 FilterSections = InputArgs.getAllArgValues(OBJDUMP_section_EQ); 2527 SectionHeaders = InputArgs.hasArg(OBJDUMP_section_headers); 2528 ShowLMA = InputArgs.hasArg(OBJDUMP_show_lma); 2529 PrintSource = InputArgs.hasArg(OBJDUMP_source); 2530 parseIntArg(InputArgs, OBJDUMP_start_address_EQ, StartAddress); 2531 HasStartAddressFlag = InputArgs.hasArg(OBJDUMP_start_address_EQ); 2532 parseIntArg(InputArgs, OBJDUMP_stop_address_EQ, StopAddress); 2533 HasStopAddressFlag = InputArgs.hasArg(OBJDUMP_stop_address_EQ); 2534 SymbolTable = InputArgs.hasArg(OBJDUMP_syms); 2535 SymbolizeOperands = InputArgs.hasArg(OBJDUMP_symbolize_operands); 2536 DynamicSymbolTable = InputArgs.hasArg(OBJDUMP_dynamic_syms); 2537 TripleName = InputArgs.getLastArgValue(OBJDUMP_triple_EQ).str(); 2538 UnwindInfo = InputArgs.hasArg(OBJDUMP_unwind_info); 2539 Wide = InputArgs.hasArg(OBJDUMP_wide); 2540 Prefix = InputArgs.getLastArgValue(OBJDUMP_prefix).str(); 2541 parseIntArg(InputArgs, OBJDUMP_prefix_strip, PrefixStrip); 2542 if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_debug_vars_EQ)) { 2543 DbgVariables = StringSwitch<DebugVarsFormat>(A->getValue()) 2544 .Case("ascii", DVASCII) 2545 .Case("unicode", DVUnicode); 2546 } 2547 parseIntArg(InputArgs, OBJDUMP_debug_vars_indent_EQ, DbgIndent); 2548 2549 parseMachOOptions(InputArgs); 2550 2551 // Parse -M (--disassembler-options) and deprecated 2552 // --x86-asm-syntax={att,intel}. 2553 // 2554 // Note, for x86, the asm dialect (AssemblerDialect) is initialized when the 2555 // MCAsmInfo is constructed. MCInstPrinter::applyTargetSpecificCLOption is 2556 // called too late. For now we have to use the internal cl::opt option. 2557 const char *AsmSyntax = nullptr; 2558 for (const auto *A : InputArgs.filtered(OBJDUMP_disassembler_options_EQ, 2559 OBJDUMP_x86_asm_syntax_att, 2560 OBJDUMP_x86_asm_syntax_intel)) { 2561 switch (A->getOption().getID()) { 2562 case OBJDUMP_x86_asm_syntax_att: 2563 AsmSyntax = "--x86-asm-syntax=att"; 2564 continue; 2565 case OBJDUMP_x86_asm_syntax_intel: 2566 AsmSyntax = "--x86-asm-syntax=intel"; 2567 continue; 2568 } 2569 2570 SmallVector<StringRef, 2> Values; 2571 llvm::SplitString(A->getValue(), Values, ","); 2572 for (StringRef V : Values) { 2573 if (V == "att") 2574 AsmSyntax = "--x86-asm-syntax=att"; 2575 else if (V == "intel") 2576 AsmSyntax = "--x86-asm-syntax=intel"; 2577 else 2578 DisassemblerOptions.push_back(V.str()); 2579 } 2580 } 2581 if (AsmSyntax) { 2582 const char *Argv[] = {"llvm-objdump", AsmSyntax}; 2583 llvm::cl::ParseCommandLineOptions(2, Argv); 2584 } 2585 2586 // objdump defaults to a.out if no filenames specified. 2587 if (InputFilenames.empty()) 2588 InputFilenames.push_back("a.out"); 2589 } 2590 2591 int main(int argc, char **argv) { 2592 using namespace llvm; 2593 InitLLVM X(argc, argv); 2594 2595 ToolName = argv[0]; 2596 std::unique_ptr<CommonOptTable> T; 2597 OptSpecifier Unknown, HelpFlag, HelpHiddenFlag, VersionFlag; 2598 2599 StringRef Stem = sys::path::stem(ToolName); 2600 auto Is = [=](StringRef Tool) { 2601 // We need to recognize the following filenames: 2602 // 2603 // llvm-objdump -> objdump 2604 // llvm-otool-10.exe -> otool 2605 // powerpc64-unknown-freebsd13-objdump -> objdump 2606 auto I = Stem.rfind_insensitive(Tool); 2607 return I != StringRef::npos && 2608 (I + Tool.size() == Stem.size() || !isAlnum(Stem[I + Tool.size()])); 2609 }; 2610 if (Is("otool")) { 2611 T = std::make_unique<OtoolOptTable>(); 2612 Unknown = OTOOL_UNKNOWN; 2613 HelpFlag = OTOOL_help; 2614 HelpHiddenFlag = OTOOL_help_hidden; 2615 VersionFlag = OTOOL_version; 2616 } else { 2617 T = std::make_unique<ObjdumpOptTable>(); 2618 Unknown = OBJDUMP_UNKNOWN; 2619 HelpFlag = OBJDUMP_help; 2620 HelpHiddenFlag = OBJDUMP_help_hidden; 2621 VersionFlag = OBJDUMP_version; 2622 } 2623 2624 BumpPtrAllocator A; 2625 StringSaver Saver(A); 2626 opt::InputArgList InputArgs = 2627 T->parseArgs(argc, argv, Unknown, Saver, 2628 [&](StringRef Msg) { reportCmdLineError(Msg); }); 2629 2630 if (InputArgs.size() == 0 || InputArgs.hasArg(HelpFlag)) { 2631 T->printHelp(ToolName); 2632 return 0; 2633 } 2634 if (InputArgs.hasArg(HelpHiddenFlag)) { 2635 T->printHelp(ToolName, /*show_hidden=*/true); 2636 return 0; 2637 } 2638 2639 // Initialize targets and assembly printers/parsers. 2640 InitializeAllTargetInfos(); 2641 InitializeAllTargetMCs(); 2642 InitializeAllDisassemblers(); 2643 2644 if (InputArgs.hasArg(VersionFlag)) { 2645 cl::PrintVersionMessage(); 2646 if (!Is("otool")) { 2647 outs() << '\n'; 2648 TargetRegistry::printRegisteredTargetsForVersion(outs()); 2649 } 2650 return 0; 2651 } 2652 2653 if (Is("otool")) 2654 parseOtoolOptions(InputArgs); 2655 else 2656 parseObjdumpOptions(InputArgs); 2657 2658 if (StartAddress >= StopAddress) 2659 reportCmdLineError("start address should be less than stop address"); 2660 2661 // Removes trailing separators from prefix. 2662 while (!Prefix.empty() && sys::path::is_separator(Prefix.back())) 2663 Prefix.pop_back(); 2664 2665 if (AllHeaders) 2666 ArchiveHeaders = FileHeaders = PrivateHeaders = Relocations = 2667 SectionHeaders = SymbolTable = true; 2668 2669 if (DisassembleAll || PrintSource || PrintLines || 2670 !DisassembleSymbols.empty()) 2671 Disassemble = true; 2672 2673 if (!ArchiveHeaders && !Disassemble && DwarfDumpType == DIDT_Null && 2674 !DynamicRelocations && !FileHeaders && !PrivateHeaders && !RawClangAST && 2675 !Relocations && !SectionHeaders && !SectionContents && !SymbolTable && 2676 !DynamicSymbolTable && !UnwindInfo && !FaultMapSection && 2677 !(MachOOpt && 2678 (Bind || DataInCode || DylibId || DylibsUsed || ExportsTrie || 2679 FirstPrivateHeader || FunctionStarts || IndirectSymbols || InfoPlist || 2680 LazyBind || LinkOptHints || ObjcMetaData || Rebase || Rpaths || 2681 UniversalHeaders || WeakBind || !FilterSections.empty()))) { 2682 T->printHelp(ToolName); 2683 return 2; 2684 } 2685 2686 DisasmSymbolSet.insert(DisassembleSymbols.begin(), DisassembleSymbols.end()); 2687 2688 llvm::for_each(InputFilenames, dumpInput); 2689 2690 warnOnNoMatchForSections(); 2691 2692 return EXIT_SUCCESS; 2693 } 2694