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 void addPltEntries(const ObjectFile *Obj, 740 std::map<SectionRef, SectionSymbolsTy> &AllSymbols, 741 StringSaver &Saver) { 742 Optional<SectionRef> Plt = None; 743 for (const SectionRef &Section : Obj->sections()) { 744 Expected<StringRef> SecNameOrErr = Section.getName(); 745 if (!SecNameOrErr) { 746 consumeError(SecNameOrErr.takeError()); 747 continue; 748 } 749 if (*SecNameOrErr == ".plt") 750 Plt = Section; 751 } 752 if (!Plt) 753 return; 754 if (auto *ElfObj = dyn_cast<ELFObjectFileBase>(Obj)) { 755 for (auto PltEntry : ElfObj->getPltAddresses()) { 756 if (PltEntry.first) { 757 SymbolRef Symbol(*PltEntry.first, ElfObj); 758 uint8_t SymbolType = getElfSymbolType(Obj, Symbol); 759 if (Expected<StringRef> NameOrErr = Symbol.getName()) { 760 if (!NameOrErr->empty()) 761 AllSymbols[*Plt].emplace_back( 762 PltEntry.second, Saver.save((*NameOrErr + "@plt").str()), 763 SymbolType); 764 continue; 765 } else { 766 // The warning has been reported in disassembleObject(). 767 consumeError(NameOrErr.takeError()); 768 } 769 } 770 reportWarning("PLT entry at 0x" + Twine::utohexstr(PltEntry.second) + 771 " references an invalid symbol", 772 Obj->getFileName()); 773 } 774 } 775 } 776 777 // Normally the disassembly output will skip blocks of zeroes. This function 778 // returns the number of zero bytes that can be skipped when dumping the 779 // disassembly of the instructions in Buf. 780 static size_t countSkippableZeroBytes(ArrayRef<uint8_t> Buf) { 781 // Find the number of leading zeroes. 782 size_t N = 0; 783 while (N < Buf.size() && !Buf[N]) 784 ++N; 785 786 // We may want to skip blocks of zero bytes, but unless we see 787 // at least 8 of them in a row. 788 if (N < 8) 789 return 0; 790 791 // We skip zeroes in multiples of 4 because do not want to truncate an 792 // instruction if it starts with a zero byte. 793 return N & ~0x3; 794 } 795 796 // Returns a map from sections to their relocations. 797 static std::map<SectionRef, std::vector<RelocationRef>> 798 getRelocsMap(object::ObjectFile const &Obj) { 799 std::map<SectionRef, std::vector<RelocationRef>> Ret; 800 uint64_t I = (uint64_t)-1; 801 for (SectionRef Sec : Obj.sections()) { 802 ++I; 803 Expected<section_iterator> RelocatedOrErr = Sec.getRelocatedSection(); 804 if (!RelocatedOrErr) 805 reportError(Obj.getFileName(), 806 "section (" + Twine(I) + 807 "): failed to get a relocated section: " + 808 toString(RelocatedOrErr.takeError())); 809 810 section_iterator Relocated = *RelocatedOrErr; 811 if (Relocated == Obj.section_end() || !checkSectionFilter(*Relocated).Keep) 812 continue; 813 std::vector<RelocationRef> &V = Ret[*Relocated]; 814 append_range(V, Sec.relocations()); 815 // Sort relocations by address. 816 llvm::stable_sort(V, isRelocAddressLess); 817 } 818 return Ret; 819 } 820 821 // Used for --adjust-vma to check if address should be adjusted by the 822 // specified value for a given section. 823 // For ELF we do not adjust non-allocatable sections like debug ones, 824 // because they are not loadable. 825 // TODO: implement for other file formats. 826 static bool shouldAdjustVA(const SectionRef &Section) { 827 const ObjectFile *Obj = Section.getObject(); 828 if (Obj->isELF()) 829 return ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC; 830 return false; 831 } 832 833 834 typedef std::pair<uint64_t, char> MappingSymbolPair; 835 static char getMappingSymbolKind(ArrayRef<MappingSymbolPair> MappingSymbols, 836 uint64_t Address) { 837 auto It = 838 partition_point(MappingSymbols, [Address](const MappingSymbolPair &Val) { 839 return Val.first <= Address; 840 }); 841 // Return zero for any address before the first mapping symbol; this means 842 // we should use the default disassembly mode, depending on the target. 843 if (It == MappingSymbols.begin()) 844 return '\x00'; 845 return (It - 1)->second; 846 } 847 848 static uint64_t dumpARMELFData(uint64_t SectionAddr, uint64_t Index, 849 uint64_t End, const ObjectFile *Obj, 850 ArrayRef<uint8_t> Bytes, 851 ArrayRef<MappingSymbolPair> MappingSymbols, 852 raw_ostream &OS) { 853 support::endianness Endian = 854 Obj->isLittleEndian() ? support::little : support::big; 855 OS << format("%8" PRIx64 ":\t", SectionAddr + Index); 856 if (Index + 4 <= End) { 857 dumpBytes(Bytes.slice(Index, 4), OS); 858 OS << "\t.word\t" 859 << format_hex(support::endian::read32(Bytes.data() + Index, Endian), 860 10); 861 return 4; 862 } 863 if (Index + 2 <= End) { 864 dumpBytes(Bytes.slice(Index, 2), OS); 865 OS << "\t\t.short\t" 866 << format_hex(support::endian::read16(Bytes.data() + Index, Endian), 867 6); 868 return 2; 869 } 870 dumpBytes(Bytes.slice(Index, 1), OS); 871 OS << "\t\t.byte\t" << format_hex(Bytes[0], 4); 872 return 1; 873 } 874 875 static void dumpELFData(uint64_t SectionAddr, uint64_t Index, uint64_t End, 876 ArrayRef<uint8_t> Bytes) { 877 // print out data up to 8 bytes at a time in hex and ascii 878 uint8_t AsciiData[9] = {'\0'}; 879 uint8_t Byte; 880 int NumBytes = 0; 881 882 for (; Index < End; ++Index) { 883 if (NumBytes == 0) 884 outs() << format("%8" PRIx64 ":", SectionAddr + Index); 885 Byte = Bytes.slice(Index)[0]; 886 outs() << format(" %02x", Byte); 887 AsciiData[NumBytes] = isPrint(Byte) ? Byte : '.'; 888 889 uint8_t IndentOffset = 0; 890 NumBytes++; 891 if (Index == End - 1 || NumBytes > 8) { 892 // Indent the space for less than 8 bytes data. 893 // 2 spaces for byte and one for space between bytes 894 IndentOffset = 3 * (8 - NumBytes); 895 for (int Excess = NumBytes; Excess < 8; Excess++) 896 AsciiData[Excess] = '\0'; 897 NumBytes = 8; 898 } 899 if (NumBytes == 8) { 900 AsciiData[8] = '\0'; 901 outs() << std::string(IndentOffset, ' ') << " "; 902 outs() << reinterpret_cast<char *>(AsciiData); 903 outs() << '\n'; 904 NumBytes = 0; 905 } 906 } 907 } 908 909 SymbolInfoTy objdump::createSymbolInfo(const ObjectFile *Obj, 910 const SymbolRef &Symbol) { 911 const StringRef FileName = Obj->getFileName(); 912 const uint64_t Addr = unwrapOrError(Symbol.getAddress(), FileName); 913 const StringRef Name = unwrapOrError(Symbol.getName(), FileName); 914 915 if (Obj->isXCOFF() && SymbolDescription) { 916 const auto *XCOFFObj = cast<XCOFFObjectFile>(Obj); 917 DataRefImpl SymbolDRI = Symbol.getRawDataRefImpl(); 918 919 const uint32_t SymbolIndex = XCOFFObj->getSymbolIndex(SymbolDRI.p); 920 Optional<XCOFF::StorageMappingClass> Smc = 921 getXCOFFSymbolCsectSMC(XCOFFObj, Symbol); 922 return SymbolInfoTy(Addr, Name, Smc, SymbolIndex, 923 isLabel(XCOFFObj, Symbol)); 924 } else 925 return SymbolInfoTy(Addr, Name, 926 Obj->isELF() ? getElfSymbolType(Obj, Symbol) 927 : (uint8_t)ELF::STT_NOTYPE); 928 } 929 930 static SymbolInfoTy createDummySymbolInfo(const ObjectFile *Obj, 931 const uint64_t Addr, StringRef &Name, 932 uint8_t Type) { 933 if (Obj->isXCOFF() && SymbolDescription) 934 return SymbolInfoTy(Addr, Name, None, None, false); 935 else 936 return SymbolInfoTy(Addr, Name, Type); 937 } 938 939 static void 940 collectLocalBranchTargets(ArrayRef<uint8_t> Bytes, const MCInstrAnalysis *MIA, 941 MCDisassembler *DisAsm, MCInstPrinter *IP, 942 const MCSubtargetInfo *STI, uint64_t SectionAddr, 943 uint64_t Start, uint64_t End, 944 std::unordered_map<uint64_t, std::string> &Labels) { 945 // So far only supports X86. 946 if (!STI->getTargetTriple().isX86()) 947 return; 948 949 Labels.clear(); 950 unsigned LabelCount = 0; 951 Start += SectionAddr; 952 End += SectionAddr; 953 uint64_t Index = Start; 954 while (Index < End) { 955 // Disassemble a real instruction and record function-local branch labels. 956 MCInst Inst; 957 uint64_t Size; 958 bool Disassembled = DisAsm->getInstruction( 959 Inst, Size, Bytes.slice(Index - SectionAddr), Index, nulls()); 960 if (Size == 0) 961 Size = 1; 962 963 if (Disassembled && MIA) { 964 uint64_t Target; 965 bool TargetKnown = MIA->evaluateBranch(Inst, Index, Size, Target); 966 if (TargetKnown && (Target >= Start && Target < End) && 967 !Labels.count(Target)) 968 Labels[Target] = ("L" + Twine(LabelCount++)).str(); 969 } 970 971 Index += Size; 972 } 973 } 974 975 // Create an MCSymbolizer for the target and add it to the MCDisassembler. 976 // This is currently only used on AMDGPU, and assumes the format of the 977 // void * argument passed to AMDGPU's createMCSymbolizer. 978 static void addSymbolizer( 979 MCContext &Ctx, const Target *Target, StringRef TripleName, 980 MCDisassembler *DisAsm, uint64_t SectionAddr, ArrayRef<uint8_t> Bytes, 981 SectionSymbolsTy &Symbols, 982 std::vector<std::unique_ptr<std::string>> &SynthesizedLabelNames) { 983 984 std::unique_ptr<MCRelocationInfo> RelInfo( 985 Target->createMCRelocationInfo(TripleName, Ctx)); 986 if (!RelInfo) 987 return; 988 std::unique_ptr<MCSymbolizer> Symbolizer(Target->createMCSymbolizer( 989 TripleName, nullptr, nullptr, &Symbols, &Ctx, std::move(RelInfo))); 990 MCSymbolizer *SymbolizerPtr = &*Symbolizer; 991 DisAsm->setSymbolizer(std::move(Symbolizer)); 992 993 if (!SymbolizeOperands) 994 return; 995 996 // Synthesize labels referenced by branch instructions by 997 // disassembling, discarding the output, and collecting the referenced 998 // addresses from the symbolizer. 999 for (size_t Index = 0; Index != Bytes.size();) { 1000 MCInst Inst; 1001 uint64_t Size; 1002 DisAsm->getInstruction(Inst, Size, Bytes.slice(Index), SectionAddr + Index, 1003 nulls()); 1004 if (Size == 0) 1005 Size = 1; 1006 Index += Size; 1007 } 1008 ArrayRef<uint64_t> LabelAddrsRef = SymbolizerPtr->getReferencedAddresses(); 1009 // Copy and sort to remove duplicates. 1010 std::vector<uint64_t> LabelAddrs; 1011 LabelAddrs.insert(LabelAddrs.end(), LabelAddrsRef.begin(), 1012 LabelAddrsRef.end()); 1013 llvm::sort(LabelAddrs); 1014 LabelAddrs.resize(std::unique(LabelAddrs.begin(), LabelAddrs.end()) - 1015 LabelAddrs.begin()); 1016 // Add the labels. 1017 for (unsigned LabelNum = 0; LabelNum != LabelAddrs.size(); ++LabelNum) { 1018 auto Name = std::make_unique<std::string>(); 1019 *Name = (Twine("L") + Twine(LabelNum)).str(); 1020 SynthesizedLabelNames.push_back(std::move(Name)); 1021 Symbols.push_back(SymbolInfoTy( 1022 LabelAddrs[LabelNum], *SynthesizedLabelNames.back(), ELF::STT_NOTYPE)); 1023 } 1024 llvm::stable_sort(Symbols); 1025 // Recreate the symbolizer with the new symbols list. 1026 RelInfo.reset(Target->createMCRelocationInfo(TripleName, Ctx)); 1027 Symbolizer.reset(Target->createMCSymbolizer( 1028 TripleName, nullptr, nullptr, &Symbols, &Ctx, std::move(RelInfo))); 1029 DisAsm->setSymbolizer(std::move(Symbolizer)); 1030 } 1031 1032 static StringRef getSegmentName(const MachOObjectFile *MachO, 1033 const SectionRef &Section) { 1034 if (MachO) { 1035 DataRefImpl DR = Section.getRawDataRefImpl(); 1036 StringRef SegmentName = MachO->getSectionFinalSegmentName(DR); 1037 return SegmentName; 1038 } 1039 return ""; 1040 } 1041 1042 static void emitPostInstructionInfo(formatted_raw_ostream &FOS, 1043 const MCAsmInfo &MAI, 1044 const MCSubtargetInfo &STI, 1045 StringRef Comments, 1046 LiveVariablePrinter &LVP) { 1047 do { 1048 if (!Comments.empty()) { 1049 // Emit a line of comments. 1050 StringRef Comment; 1051 std::tie(Comment, Comments) = Comments.split('\n'); 1052 // MAI.getCommentColumn() assumes that instructions are printed at the 1053 // position of 8, while getInstStartColumn() returns the actual position. 1054 unsigned CommentColumn = 1055 MAI.getCommentColumn() - 8 + getInstStartColumn(STI); 1056 FOS.PadToColumn(CommentColumn); 1057 FOS << MAI.getCommentString() << ' ' << Comment; 1058 } 1059 LVP.printAfterInst(FOS); 1060 FOS << '\n'; 1061 } while (!Comments.empty()); 1062 FOS.flush(); 1063 } 1064 1065 static void disassembleObject(const Target *TheTarget, const ObjectFile *Obj, 1066 MCContext &Ctx, MCDisassembler *PrimaryDisAsm, 1067 MCDisassembler *SecondaryDisAsm, 1068 const MCInstrAnalysis *MIA, MCInstPrinter *IP, 1069 const MCSubtargetInfo *PrimarySTI, 1070 const MCSubtargetInfo *SecondarySTI, 1071 PrettyPrinter &PIP, 1072 SourcePrinter &SP, bool InlineRelocs) { 1073 const MCSubtargetInfo *STI = PrimarySTI; 1074 MCDisassembler *DisAsm = PrimaryDisAsm; 1075 bool PrimaryIsThumb = false; 1076 if (isArmElf(Obj)) 1077 PrimaryIsThumb = STI->checkFeatures("+thumb-mode"); 1078 1079 std::map<SectionRef, std::vector<RelocationRef>> RelocMap; 1080 if (InlineRelocs) 1081 RelocMap = getRelocsMap(*Obj); 1082 bool Is64Bits = Obj->getBytesInAddress() > 4; 1083 1084 // Create a mapping from virtual address to symbol name. This is used to 1085 // pretty print the symbols while disassembling. 1086 std::map<SectionRef, SectionSymbolsTy> AllSymbols; 1087 SectionSymbolsTy AbsoluteSymbols; 1088 const StringRef FileName = Obj->getFileName(); 1089 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(Obj); 1090 for (const SymbolRef &Symbol : Obj->symbols()) { 1091 Expected<StringRef> NameOrErr = Symbol.getName(); 1092 if (!NameOrErr) { 1093 reportWarning(toString(NameOrErr.takeError()), FileName); 1094 continue; 1095 } 1096 if (NameOrErr->empty() && !(Obj->isXCOFF() && SymbolDescription)) 1097 continue; 1098 1099 if (Obj->isELF() && getElfSymbolType(Obj, Symbol) == ELF::STT_SECTION) 1100 continue; 1101 1102 if (MachO) { 1103 // __mh_(execute|dylib|dylinker|bundle|preload|object)_header are special 1104 // symbols that support MachO header introspection. They do not bind to 1105 // code locations and are irrelevant for disassembly. 1106 if (NameOrErr->startswith("__mh_") && NameOrErr->endswith("_header")) 1107 continue; 1108 // Don't ask a Mach-O STAB symbol for its section unless you know that 1109 // STAB symbol's section field refers to a valid section index. Otherwise 1110 // the symbol may error trying to load a section that does not exist. 1111 DataRefImpl SymDRI = Symbol.getRawDataRefImpl(); 1112 uint8_t NType = (MachO->is64Bit() ? 1113 MachO->getSymbol64TableEntry(SymDRI).n_type: 1114 MachO->getSymbolTableEntry(SymDRI).n_type); 1115 if (NType & MachO::N_STAB) 1116 continue; 1117 } 1118 1119 section_iterator SecI = unwrapOrError(Symbol.getSection(), FileName); 1120 if (SecI != Obj->section_end()) 1121 AllSymbols[*SecI].push_back(createSymbolInfo(Obj, Symbol)); 1122 else 1123 AbsoluteSymbols.push_back(createSymbolInfo(Obj, Symbol)); 1124 } 1125 1126 if (AllSymbols.empty() && Obj->isELF()) 1127 addDynamicElfSymbols(Obj, AllSymbols); 1128 1129 BumpPtrAllocator A; 1130 StringSaver Saver(A); 1131 addPltEntries(Obj, AllSymbols, Saver); 1132 1133 // Create a mapping from virtual address to section. An empty section can 1134 // cause more than one section at the same address. Sort such sections to be 1135 // before same-addressed non-empty sections so that symbol lookups prefer the 1136 // non-empty section. 1137 std::vector<std::pair<uint64_t, SectionRef>> SectionAddresses; 1138 for (SectionRef Sec : Obj->sections()) 1139 SectionAddresses.emplace_back(Sec.getAddress(), Sec); 1140 llvm::stable_sort(SectionAddresses, [](const auto &LHS, const auto &RHS) { 1141 if (LHS.first != RHS.first) 1142 return LHS.first < RHS.first; 1143 return LHS.second.getSize() < RHS.second.getSize(); 1144 }); 1145 1146 // Linked executables (.exe and .dll files) typically don't include a real 1147 // symbol table but they might contain an export table. 1148 if (const auto *COFFObj = dyn_cast<COFFObjectFile>(Obj)) { 1149 for (const auto &ExportEntry : COFFObj->export_directories()) { 1150 StringRef Name; 1151 if (Error E = ExportEntry.getSymbolName(Name)) 1152 reportError(std::move(E), Obj->getFileName()); 1153 if (Name.empty()) 1154 continue; 1155 1156 uint32_t RVA; 1157 if (Error E = ExportEntry.getExportRVA(RVA)) 1158 reportError(std::move(E), Obj->getFileName()); 1159 1160 uint64_t VA = COFFObj->getImageBase() + RVA; 1161 auto Sec = partition_point( 1162 SectionAddresses, [VA](const std::pair<uint64_t, SectionRef> &O) { 1163 return O.first <= VA; 1164 }); 1165 if (Sec != SectionAddresses.begin()) { 1166 --Sec; 1167 AllSymbols[Sec->second].emplace_back(VA, Name, ELF::STT_NOTYPE); 1168 } else 1169 AbsoluteSymbols.emplace_back(VA, Name, ELF::STT_NOTYPE); 1170 } 1171 } 1172 1173 // Sort all the symbols, this allows us to use a simple binary search to find 1174 // Multiple symbols can have the same address. Use a stable sort to stabilize 1175 // the output. 1176 StringSet<> FoundDisasmSymbolSet; 1177 for (std::pair<const SectionRef, SectionSymbolsTy> &SecSyms : AllSymbols) 1178 llvm::stable_sort(SecSyms.second); 1179 llvm::stable_sort(AbsoluteSymbols); 1180 1181 std::unique_ptr<DWARFContext> DICtx; 1182 LiveVariablePrinter LVP(*Ctx.getRegisterInfo(), *STI); 1183 1184 if (DbgVariables != DVDisabled) { 1185 DICtx = DWARFContext::create(*Obj); 1186 for (const std::unique_ptr<DWARFUnit> &CU : DICtx->compile_units()) 1187 LVP.addCompileUnit(CU->getUnitDIE(false)); 1188 } 1189 1190 LLVM_DEBUG(LVP.dump()); 1191 1192 for (const SectionRef &Section : ToolSectionFilter(*Obj)) { 1193 if (FilterSections.empty() && !DisassembleAll && 1194 (!Section.isText() || Section.isVirtual())) 1195 continue; 1196 1197 uint64_t SectionAddr = Section.getAddress(); 1198 uint64_t SectSize = Section.getSize(); 1199 if (!SectSize) 1200 continue; 1201 1202 // Get the list of all the symbols in this section. 1203 SectionSymbolsTy &Symbols = AllSymbols[Section]; 1204 std::vector<MappingSymbolPair> MappingSymbols; 1205 if (hasMappingSymbols(Obj)) { 1206 for (const auto &Symb : Symbols) { 1207 uint64_t Address = Symb.Addr; 1208 StringRef Name = Symb.Name; 1209 if (Name.startswith("$d")) 1210 MappingSymbols.emplace_back(Address - SectionAddr, 'd'); 1211 if (Name.startswith("$x")) 1212 MappingSymbols.emplace_back(Address - SectionAddr, 'x'); 1213 if (Name.startswith("$a")) 1214 MappingSymbols.emplace_back(Address - SectionAddr, 'a'); 1215 if (Name.startswith("$t")) 1216 MappingSymbols.emplace_back(Address - SectionAddr, 't'); 1217 } 1218 } 1219 1220 llvm::sort(MappingSymbols); 1221 1222 ArrayRef<uint8_t> Bytes = arrayRefFromStringRef( 1223 unwrapOrError(Section.getContents(), Obj->getFileName())); 1224 1225 std::vector<std::unique_ptr<std::string>> SynthesizedLabelNames; 1226 if (Obj->isELF() && Obj->getArch() == Triple::amdgcn) { 1227 // AMDGPU disassembler uses symbolizer for printing labels 1228 addSymbolizer(Ctx, TheTarget, TripleName, DisAsm, SectionAddr, Bytes, 1229 Symbols, SynthesizedLabelNames); 1230 } 1231 1232 StringRef SegmentName = getSegmentName(MachO, Section); 1233 StringRef SectionName = unwrapOrError(Section.getName(), Obj->getFileName()); 1234 // If the section has no symbol at the start, just insert a dummy one. 1235 if (Symbols.empty() || Symbols[0].Addr != 0) { 1236 Symbols.insert(Symbols.begin(), 1237 createDummySymbolInfo(Obj, SectionAddr, SectionName, 1238 Section.isText() ? ELF::STT_FUNC 1239 : ELF::STT_OBJECT)); 1240 } 1241 1242 SmallString<40> Comments; 1243 raw_svector_ostream CommentStream(Comments); 1244 1245 uint64_t VMAAdjustment = 0; 1246 if (shouldAdjustVA(Section)) 1247 VMAAdjustment = AdjustVMA; 1248 1249 uint64_t Size; 1250 uint64_t Index; 1251 bool PrintedSection = false; 1252 std::vector<RelocationRef> Rels = RelocMap[Section]; 1253 std::vector<RelocationRef>::const_iterator RelCur = Rels.begin(); 1254 std::vector<RelocationRef>::const_iterator RelEnd = Rels.end(); 1255 // Disassemble symbol by symbol. 1256 for (unsigned SI = 0, SE = Symbols.size(); SI != SE; ++SI) { 1257 std::string SymbolName = Symbols[SI].Name.str(); 1258 if (Demangle) 1259 SymbolName = demangle(SymbolName); 1260 1261 // Skip if --disassemble-symbols is not empty and the symbol is not in 1262 // the list. 1263 if (!DisasmSymbolSet.empty() && !DisasmSymbolSet.count(SymbolName)) 1264 continue; 1265 1266 uint64_t Start = Symbols[SI].Addr; 1267 if (Start < SectionAddr || StopAddress <= Start) 1268 continue; 1269 else 1270 FoundDisasmSymbolSet.insert(SymbolName); 1271 1272 // The end is the section end, the beginning of the next symbol, or 1273 // --stop-address. 1274 uint64_t End = std::min<uint64_t>(SectionAddr + SectSize, StopAddress); 1275 if (SI + 1 < SE) 1276 End = std::min(End, Symbols[SI + 1].Addr); 1277 if (Start >= End || End <= StartAddress) 1278 continue; 1279 Start -= SectionAddr; 1280 End -= SectionAddr; 1281 1282 if (!PrintedSection) { 1283 PrintedSection = true; 1284 outs() << "\nDisassembly of section "; 1285 if (!SegmentName.empty()) 1286 outs() << SegmentName << ","; 1287 outs() << SectionName << ":\n"; 1288 } 1289 1290 outs() << '\n'; 1291 if (LeadingAddr) 1292 outs() << format(Is64Bits ? "%016" PRIx64 " " : "%08" PRIx64 " ", 1293 SectionAddr + Start + VMAAdjustment); 1294 if (Obj->isXCOFF() && SymbolDescription) { 1295 outs() << getXCOFFSymbolDescription(Symbols[SI], SymbolName) << ":\n"; 1296 } else 1297 outs() << '<' << SymbolName << ">:\n"; 1298 1299 // Don't print raw contents of a virtual section. A virtual section 1300 // doesn't have any contents in the file. 1301 if (Section.isVirtual()) { 1302 outs() << "...\n"; 1303 continue; 1304 } 1305 1306 auto Status = DisAsm->onSymbolStart(Symbols[SI], Size, 1307 Bytes.slice(Start, End - Start), 1308 SectionAddr + Start, CommentStream); 1309 // To have round trippable disassembly, we fall back to decoding the 1310 // remaining bytes as instructions. 1311 // 1312 // If there is a failure, we disassemble the failed region as bytes before 1313 // falling back. The target is expected to print nothing in this case. 1314 // 1315 // If there is Success or SoftFail i.e no 'real' failure, we go ahead by 1316 // Size bytes before falling back. 1317 // So if the entire symbol is 'eaten' by the target: 1318 // Start += Size // Now Start = End and we will never decode as 1319 // // instructions 1320 // 1321 // Right now, most targets return None i.e ignore to treat a symbol 1322 // separately. But WebAssembly decodes preludes for some symbols. 1323 // 1324 if (Status.hasValue()) { 1325 if (Status.getValue() == MCDisassembler::Fail) { 1326 outs() << "// Error in decoding " << SymbolName 1327 << " : Decoding failed region as bytes.\n"; 1328 for (uint64_t I = 0; I < Size; ++I) { 1329 outs() << "\t.byte\t " << format_hex(Bytes[I], 1, /*Upper=*/true) 1330 << "\n"; 1331 } 1332 } 1333 } else { 1334 Size = 0; 1335 } 1336 1337 Start += Size; 1338 1339 Index = Start; 1340 if (SectionAddr < StartAddress) 1341 Index = std::max<uint64_t>(Index, StartAddress - SectionAddr); 1342 1343 // If there is a data/common symbol inside an ELF text section and we are 1344 // only disassembling text (applicable all architectures), we are in a 1345 // situation where we must print the data and not disassemble it. 1346 if (Obj->isELF() && !DisassembleAll && Section.isText()) { 1347 uint8_t SymTy = Symbols[SI].Type; 1348 if (SymTy == ELF::STT_OBJECT || SymTy == ELF::STT_COMMON) { 1349 dumpELFData(SectionAddr, Index, End, Bytes); 1350 Index = End; 1351 } 1352 } 1353 1354 bool CheckARMELFData = hasMappingSymbols(Obj) && 1355 Symbols[SI].Type != ELF::STT_OBJECT && 1356 !DisassembleAll; 1357 bool DumpARMELFData = false; 1358 formatted_raw_ostream FOS(outs()); 1359 1360 std::unordered_map<uint64_t, std::string> AllLabels; 1361 if (SymbolizeOperands) 1362 collectLocalBranchTargets(Bytes, MIA, DisAsm, IP, PrimarySTI, 1363 SectionAddr, Index, End, AllLabels); 1364 1365 while (Index < End) { 1366 // ARM and AArch64 ELF binaries can interleave data and text in the 1367 // same section. We rely on the markers introduced to understand what 1368 // we need to dump. If the data marker is within a function, it is 1369 // denoted as a word/short etc. 1370 if (CheckARMELFData) { 1371 char Kind = getMappingSymbolKind(MappingSymbols, Index); 1372 DumpARMELFData = Kind == 'd'; 1373 if (SecondarySTI) { 1374 if (Kind == 'a') { 1375 STI = PrimaryIsThumb ? SecondarySTI : PrimarySTI; 1376 DisAsm = PrimaryIsThumb ? SecondaryDisAsm : PrimaryDisAsm; 1377 } else if (Kind == 't') { 1378 STI = PrimaryIsThumb ? PrimarySTI : SecondarySTI; 1379 DisAsm = PrimaryIsThumb ? PrimaryDisAsm : SecondaryDisAsm; 1380 } 1381 } 1382 } 1383 1384 if (DumpARMELFData) { 1385 Size = dumpARMELFData(SectionAddr, Index, End, Obj, Bytes, 1386 MappingSymbols, FOS); 1387 } else { 1388 // When -z or --disassemble-zeroes are given we always dissasemble 1389 // them. Otherwise we might want to skip zero bytes we see. 1390 if (!DisassembleZeroes) { 1391 uint64_t MaxOffset = End - Index; 1392 // For --reloc: print zero blocks patched by relocations, so that 1393 // relocations can be shown in the dump. 1394 if (RelCur != RelEnd) 1395 MaxOffset = RelCur->getOffset() - Index; 1396 1397 if (size_t N = 1398 countSkippableZeroBytes(Bytes.slice(Index, MaxOffset))) { 1399 FOS << "\t\t..." << '\n'; 1400 Index += N; 1401 continue; 1402 } 1403 } 1404 1405 // Print local label if there's any. 1406 auto Iter = AllLabels.find(SectionAddr + Index); 1407 if (Iter != AllLabels.end()) 1408 FOS << "<" << Iter->second << ">:\n"; 1409 1410 // Disassemble a real instruction or a data when disassemble all is 1411 // provided 1412 MCInst Inst; 1413 bool Disassembled = 1414 DisAsm->getInstruction(Inst, Size, Bytes.slice(Index), 1415 SectionAddr + Index, CommentStream); 1416 if (Size == 0) 1417 Size = 1; 1418 1419 LVP.update({Index, Section.getIndex()}, 1420 {Index + Size, Section.getIndex()}, Index + Size != End); 1421 1422 IP->setCommentStream(CommentStream); 1423 1424 PIP.printInst( 1425 *IP, Disassembled ? &Inst : nullptr, Bytes.slice(Index, Size), 1426 {SectionAddr + Index + VMAAdjustment, Section.getIndex()}, FOS, 1427 "", *STI, &SP, Obj->getFileName(), &Rels, LVP); 1428 1429 IP->setCommentStream(llvm::nulls()); 1430 1431 // If disassembly has failed, avoid analysing invalid/incomplete 1432 // instruction information. Otherwise, try to resolve the target 1433 // address (jump target or memory operand address) and print it on the 1434 // right of the instruction. 1435 if (Disassembled && MIA) { 1436 // Branch targets are printed just after the instructions. 1437 llvm::raw_ostream *TargetOS = &FOS; 1438 uint64_t Target; 1439 bool PrintTarget = 1440 MIA->evaluateBranch(Inst, SectionAddr + Index, Size, Target); 1441 if (!PrintTarget) 1442 if (Optional<uint64_t> MaybeTarget = 1443 MIA->evaluateMemoryOperandAddress( 1444 Inst, SectionAddr + Index, Size)) { 1445 Target = *MaybeTarget; 1446 PrintTarget = true; 1447 // Do not print real address when symbolizing. 1448 if (!SymbolizeOperands) { 1449 // Memory operand addresses are printed as comments. 1450 TargetOS = &CommentStream; 1451 *TargetOS << "0x" << Twine::utohexstr(Target); 1452 } 1453 } 1454 if (PrintTarget) { 1455 // In a relocatable object, the target's section must reside in 1456 // the same section as the call instruction or it is accessed 1457 // through a relocation. 1458 // 1459 // In a non-relocatable object, the target may be in any section. 1460 // In that case, locate the section(s) containing the target 1461 // address and find the symbol in one of those, if possible. 1462 // 1463 // N.B. We don't walk the relocations in the relocatable case yet. 1464 std::vector<const SectionSymbolsTy *> TargetSectionSymbols; 1465 if (!Obj->isRelocatableObject()) { 1466 auto It = llvm::partition_point( 1467 SectionAddresses, 1468 [=](const std::pair<uint64_t, SectionRef> &O) { 1469 return O.first <= Target; 1470 }); 1471 uint64_t TargetSecAddr = 0; 1472 while (It != SectionAddresses.begin()) { 1473 --It; 1474 if (TargetSecAddr == 0) 1475 TargetSecAddr = It->first; 1476 if (It->first != TargetSecAddr) 1477 break; 1478 TargetSectionSymbols.push_back(&AllSymbols[It->second]); 1479 } 1480 } else { 1481 TargetSectionSymbols.push_back(&Symbols); 1482 } 1483 TargetSectionSymbols.push_back(&AbsoluteSymbols); 1484 1485 // Find the last symbol in the first candidate section whose 1486 // offset is less than or equal to the target. If there are no 1487 // such symbols, try in the next section and so on, before finally 1488 // using the nearest preceding absolute symbol (if any), if there 1489 // are no other valid symbols. 1490 const SymbolInfoTy *TargetSym = nullptr; 1491 for (const SectionSymbolsTy *TargetSymbols : 1492 TargetSectionSymbols) { 1493 auto It = llvm::partition_point( 1494 *TargetSymbols, 1495 [=](const SymbolInfoTy &O) { return O.Addr <= Target; }); 1496 if (It != TargetSymbols->begin()) { 1497 TargetSym = &*(It - 1); 1498 break; 1499 } 1500 } 1501 1502 // Print the labels corresponding to the target if there's any. 1503 bool LabelAvailable = AllLabels.count(Target); 1504 if (TargetSym != nullptr) { 1505 uint64_t TargetAddress = TargetSym->Addr; 1506 uint64_t Disp = Target - TargetAddress; 1507 std::string TargetName = TargetSym->Name.str(); 1508 if (Demangle) 1509 TargetName = demangle(TargetName); 1510 1511 *TargetOS << " <"; 1512 if (!Disp) { 1513 // Always Print the binary symbol precisely corresponding to 1514 // the target address. 1515 *TargetOS << TargetName; 1516 } else if (!LabelAvailable) { 1517 // Always Print the binary symbol plus an offset if there's no 1518 // local label corresponding to the target address. 1519 *TargetOS << TargetName << "+0x" << Twine::utohexstr(Disp); 1520 } else { 1521 *TargetOS << AllLabels[Target]; 1522 } 1523 *TargetOS << ">"; 1524 } else if (LabelAvailable) { 1525 *TargetOS << " <" << AllLabels[Target] << ">"; 1526 } 1527 // By convention, each record in the comment stream should be 1528 // terminated. 1529 if (TargetOS == &CommentStream) 1530 *TargetOS << "\n"; 1531 } 1532 } 1533 } 1534 1535 assert(Ctx.getAsmInfo()); 1536 emitPostInstructionInfo(FOS, *Ctx.getAsmInfo(), *STI, 1537 CommentStream.str(), LVP); 1538 Comments.clear(); 1539 1540 // Hexagon does this in pretty printer 1541 if (Obj->getArch() != Triple::hexagon) { 1542 // Print relocation for instruction and data. 1543 while (RelCur != RelEnd) { 1544 uint64_t Offset = RelCur->getOffset(); 1545 // If this relocation is hidden, skip it. 1546 if (getHidden(*RelCur) || SectionAddr + Offset < StartAddress) { 1547 ++RelCur; 1548 continue; 1549 } 1550 1551 // Stop when RelCur's offset is past the disassembled 1552 // instruction/data. Note that it's possible the disassembled data 1553 // is not the complete data: we might see the relocation printed in 1554 // the middle of the data, but this matches the binutils objdump 1555 // output. 1556 if (Offset >= Index + Size) 1557 break; 1558 1559 // When --adjust-vma is used, update the address printed. 1560 if (RelCur->getSymbol() != Obj->symbol_end()) { 1561 Expected<section_iterator> SymSI = 1562 RelCur->getSymbol()->getSection(); 1563 if (SymSI && *SymSI != Obj->section_end() && 1564 shouldAdjustVA(**SymSI)) 1565 Offset += AdjustVMA; 1566 } 1567 1568 printRelocation(FOS, Obj->getFileName(), *RelCur, 1569 SectionAddr + Offset, Is64Bits); 1570 LVP.printAfterOtherLine(FOS, true); 1571 ++RelCur; 1572 } 1573 } 1574 1575 Index += Size; 1576 } 1577 } 1578 } 1579 StringSet<> MissingDisasmSymbolSet = 1580 set_difference(DisasmSymbolSet, FoundDisasmSymbolSet); 1581 for (StringRef Sym : MissingDisasmSymbolSet.keys()) 1582 reportWarning("failed to disassemble missing symbol " + Sym, FileName); 1583 } 1584 1585 static void disassembleObject(const ObjectFile *Obj, bool InlineRelocs) { 1586 const Target *TheTarget = getTarget(Obj); 1587 1588 // Package up features to be passed to target/subtarget 1589 SubtargetFeatures Features = Obj->getFeatures(); 1590 if (!MAttrs.empty()) 1591 for (unsigned I = 0; I != MAttrs.size(); ++I) 1592 Features.AddFeature(MAttrs[I]); 1593 1594 std::unique_ptr<const MCRegisterInfo> MRI( 1595 TheTarget->createMCRegInfo(TripleName)); 1596 if (!MRI) 1597 reportError(Obj->getFileName(), 1598 "no register info for target " + TripleName); 1599 1600 // Set up disassembler. 1601 MCTargetOptions MCOptions; 1602 std::unique_ptr<const MCAsmInfo> AsmInfo( 1603 TheTarget->createMCAsmInfo(*MRI, TripleName, MCOptions)); 1604 if (!AsmInfo) 1605 reportError(Obj->getFileName(), 1606 "no assembly info for target " + TripleName); 1607 1608 if (MCPU.empty()) 1609 MCPU = Obj->tryGetCPUName().getValueOr("").str(); 1610 1611 std::unique_ptr<const MCSubtargetInfo> STI( 1612 TheTarget->createMCSubtargetInfo(TripleName, MCPU, Features.getString())); 1613 if (!STI) 1614 reportError(Obj->getFileName(), 1615 "no subtarget info for target " + TripleName); 1616 std::unique_ptr<const MCInstrInfo> MII(TheTarget->createMCInstrInfo()); 1617 if (!MII) 1618 reportError(Obj->getFileName(), 1619 "no instruction info for target " + TripleName); 1620 MCContext Ctx(Triple(TripleName), AsmInfo.get(), MRI.get(), STI.get()); 1621 // FIXME: for now initialize MCObjectFileInfo with default values 1622 std::unique_ptr<MCObjectFileInfo> MOFI( 1623 TheTarget->createMCObjectFileInfo(Ctx, /*PIC=*/false)); 1624 Ctx.setObjectFileInfo(MOFI.get()); 1625 1626 std::unique_ptr<MCDisassembler> DisAsm( 1627 TheTarget->createMCDisassembler(*STI, Ctx)); 1628 if (!DisAsm) 1629 reportError(Obj->getFileName(), "no disassembler for target " + TripleName); 1630 1631 // If we have an ARM object file, we need a second disassembler, because 1632 // ARM CPUs have two different instruction sets: ARM mode, and Thumb mode. 1633 // We use mapping symbols to switch between the two assemblers, where 1634 // appropriate. 1635 std::unique_ptr<MCDisassembler> SecondaryDisAsm; 1636 std::unique_ptr<const MCSubtargetInfo> SecondarySTI; 1637 if (isArmElf(Obj) && !STI->checkFeatures("+mclass")) { 1638 if (STI->checkFeatures("+thumb-mode")) 1639 Features.AddFeature("-thumb-mode"); 1640 else 1641 Features.AddFeature("+thumb-mode"); 1642 SecondarySTI.reset(TheTarget->createMCSubtargetInfo(TripleName, MCPU, 1643 Features.getString())); 1644 SecondaryDisAsm.reset(TheTarget->createMCDisassembler(*SecondarySTI, Ctx)); 1645 } 1646 1647 std::unique_ptr<const MCInstrAnalysis> MIA( 1648 TheTarget->createMCInstrAnalysis(MII.get())); 1649 1650 int AsmPrinterVariant = AsmInfo->getAssemblerDialect(); 1651 std::unique_ptr<MCInstPrinter> IP(TheTarget->createMCInstPrinter( 1652 Triple(TripleName), AsmPrinterVariant, *AsmInfo, *MII, *MRI)); 1653 if (!IP) 1654 reportError(Obj->getFileName(), 1655 "no instruction printer for target " + TripleName); 1656 IP->setPrintImmHex(PrintImmHex); 1657 IP->setPrintBranchImmAsAddress(true); 1658 IP->setSymbolizeOperands(SymbolizeOperands); 1659 IP->setMCInstrAnalysis(MIA.get()); 1660 1661 PrettyPrinter &PIP = selectPrettyPrinter(Triple(TripleName)); 1662 SourcePrinter SP(Obj, TheTarget->getName()); 1663 1664 for (StringRef Opt : DisassemblerOptions) 1665 if (!IP->applyTargetSpecificCLOption(Opt)) 1666 reportError(Obj->getFileName(), 1667 "Unrecognized disassembler option: " + Opt); 1668 1669 disassembleObject(TheTarget, Obj, Ctx, DisAsm.get(), SecondaryDisAsm.get(), 1670 MIA.get(), IP.get(), STI.get(), SecondarySTI.get(), PIP, 1671 SP, InlineRelocs); 1672 } 1673 1674 void objdump::printRelocations(const ObjectFile *Obj) { 1675 StringRef Fmt = Obj->getBytesInAddress() > 4 ? "%016" PRIx64 : 1676 "%08" PRIx64; 1677 // Regular objdump doesn't print relocations in non-relocatable object 1678 // files. 1679 if (!Obj->isRelocatableObject()) 1680 return; 1681 1682 // Build a mapping from relocation target to a vector of relocation 1683 // sections. Usually, there is an only one relocation section for 1684 // each relocated section. 1685 MapVector<SectionRef, std::vector<SectionRef>> SecToRelSec; 1686 uint64_t Ndx; 1687 for (const SectionRef &Section : ToolSectionFilter(*Obj, &Ndx)) { 1688 if (Section.relocation_begin() == Section.relocation_end()) 1689 continue; 1690 Expected<section_iterator> SecOrErr = Section.getRelocatedSection(); 1691 if (!SecOrErr) 1692 reportError(Obj->getFileName(), 1693 "section (" + Twine(Ndx) + 1694 "): unable to get a relocation target: " + 1695 toString(SecOrErr.takeError())); 1696 SecToRelSec[**SecOrErr].push_back(Section); 1697 } 1698 1699 for (std::pair<SectionRef, std::vector<SectionRef>> &P : SecToRelSec) { 1700 StringRef SecName = unwrapOrError(P.first.getName(), Obj->getFileName()); 1701 outs() << "\nRELOCATION RECORDS FOR [" << SecName << "]:\n"; 1702 uint32_t OffsetPadding = (Obj->getBytesInAddress() > 4 ? 16 : 8); 1703 uint32_t TypePadding = 24; 1704 outs() << left_justify("OFFSET", OffsetPadding) << " " 1705 << left_justify("TYPE", TypePadding) << " " 1706 << "VALUE\n"; 1707 1708 for (SectionRef Section : P.second) { 1709 for (const RelocationRef &Reloc : Section.relocations()) { 1710 uint64_t Address = Reloc.getOffset(); 1711 SmallString<32> RelocName; 1712 SmallString<32> ValueStr; 1713 if (Address < StartAddress || Address > StopAddress || getHidden(Reloc)) 1714 continue; 1715 Reloc.getTypeName(RelocName); 1716 if (Error E = getRelocationValueString(Reloc, ValueStr)) 1717 reportError(std::move(E), Obj->getFileName()); 1718 1719 outs() << format(Fmt.data(), Address) << " " 1720 << left_justify(RelocName, TypePadding) << " " << ValueStr 1721 << "\n"; 1722 } 1723 } 1724 } 1725 } 1726 1727 void objdump::printDynamicRelocations(const ObjectFile *Obj) { 1728 // For the moment, this option is for ELF only 1729 if (!Obj->isELF()) 1730 return; 1731 1732 const auto *Elf = dyn_cast<ELFObjectFileBase>(Obj); 1733 if (!Elf || Elf->getEType() != ELF::ET_DYN) { 1734 reportError(Obj->getFileName(), "not a dynamic object"); 1735 return; 1736 } 1737 1738 std::vector<SectionRef> DynRelSec = Obj->dynamic_relocation_sections(); 1739 if (DynRelSec.empty()) 1740 return; 1741 1742 outs() << "DYNAMIC RELOCATION RECORDS\n"; 1743 StringRef Fmt = Obj->getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 1744 for (const SectionRef &Section : DynRelSec) 1745 for (const RelocationRef &Reloc : Section.relocations()) { 1746 uint64_t Address = Reloc.getOffset(); 1747 SmallString<32> RelocName; 1748 SmallString<32> ValueStr; 1749 Reloc.getTypeName(RelocName); 1750 if (Error E = getRelocationValueString(Reloc, ValueStr)) 1751 reportError(std::move(E), Obj->getFileName()); 1752 outs() << format(Fmt.data(), Address) << " " << RelocName << " " 1753 << ValueStr << "\n"; 1754 } 1755 } 1756 1757 // Returns true if we need to show LMA column when dumping section headers. We 1758 // show it only when the platform is ELF and either we have at least one section 1759 // whose VMA and LMA are different and/or when --show-lma flag is used. 1760 static bool shouldDisplayLMA(const ObjectFile *Obj) { 1761 if (!Obj->isELF()) 1762 return false; 1763 for (const SectionRef &S : ToolSectionFilter(*Obj)) 1764 if (S.getAddress() != getELFSectionLMA(S)) 1765 return true; 1766 return ShowLMA; 1767 } 1768 1769 static size_t getMaxSectionNameWidth(const ObjectFile *Obj) { 1770 // Default column width for names is 13 even if no names are that long. 1771 size_t MaxWidth = 13; 1772 for (const SectionRef &Section : ToolSectionFilter(*Obj)) { 1773 StringRef Name = unwrapOrError(Section.getName(), Obj->getFileName()); 1774 MaxWidth = std::max(MaxWidth, Name.size()); 1775 } 1776 return MaxWidth; 1777 } 1778 1779 void objdump::printSectionHeaders(const ObjectFile *Obj) { 1780 size_t NameWidth = getMaxSectionNameWidth(Obj); 1781 size_t AddressWidth = 2 * Obj->getBytesInAddress(); 1782 bool HasLMAColumn = shouldDisplayLMA(Obj); 1783 outs() << "\nSections:\n"; 1784 if (HasLMAColumn) 1785 outs() << "Idx " << left_justify("Name", NameWidth) << " Size " 1786 << left_justify("VMA", AddressWidth) << " " 1787 << left_justify("LMA", AddressWidth) << " Type\n"; 1788 else 1789 outs() << "Idx " << left_justify("Name", NameWidth) << " Size " 1790 << left_justify("VMA", AddressWidth) << " Type\n"; 1791 1792 uint64_t Idx; 1793 for (const SectionRef &Section : ToolSectionFilter(*Obj, &Idx)) { 1794 StringRef Name = unwrapOrError(Section.getName(), Obj->getFileName()); 1795 uint64_t VMA = Section.getAddress(); 1796 if (shouldAdjustVA(Section)) 1797 VMA += AdjustVMA; 1798 1799 uint64_t Size = Section.getSize(); 1800 1801 std::string Type = Section.isText() ? "TEXT" : ""; 1802 if (Section.isData()) 1803 Type += Type.empty() ? "DATA" : ", DATA"; 1804 if (Section.isBSS()) 1805 Type += Type.empty() ? "BSS" : ", BSS"; 1806 if (Section.isDebugSection()) 1807 Type += Type.empty() ? "DEBUG" : ", DEBUG"; 1808 1809 if (HasLMAColumn) 1810 outs() << format("%3" PRIu64 " %-*s %08" PRIx64 " ", Idx, NameWidth, 1811 Name.str().c_str(), Size) 1812 << format_hex_no_prefix(VMA, AddressWidth) << " " 1813 << format_hex_no_prefix(getELFSectionLMA(Section), AddressWidth) 1814 << " " << Type << "\n"; 1815 else 1816 outs() << format("%3" PRIu64 " %-*s %08" PRIx64 " ", Idx, NameWidth, 1817 Name.str().c_str(), Size) 1818 << format_hex_no_prefix(VMA, AddressWidth) << " " << Type << "\n"; 1819 } 1820 } 1821 1822 void objdump::printSectionContents(const ObjectFile *Obj) { 1823 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(Obj); 1824 1825 for (const SectionRef &Section : ToolSectionFilter(*Obj)) { 1826 StringRef Name = unwrapOrError(Section.getName(), Obj->getFileName()); 1827 uint64_t BaseAddr = Section.getAddress(); 1828 uint64_t Size = Section.getSize(); 1829 if (!Size) 1830 continue; 1831 1832 outs() << "Contents of section "; 1833 StringRef SegmentName = getSegmentName(MachO, Section); 1834 if (!SegmentName.empty()) 1835 outs() << SegmentName << ","; 1836 outs() << Name << ":\n"; 1837 if (Section.isBSS()) { 1838 outs() << format("<skipping contents of bss section at [%04" PRIx64 1839 ", %04" PRIx64 ")>\n", 1840 BaseAddr, BaseAddr + Size); 1841 continue; 1842 } 1843 1844 StringRef Contents = unwrapOrError(Section.getContents(), Obj->getFileName()); 1845 1846 // Dump out the content as hex and printable ascii characters. 1847 for (std::size_t Addr = 0, End = Contents.size(); Addr < End; Addr += 16) { 1848 outs() << format(" %04" PRIx64 " ", BaseAddr + Addr); 1849 // Dump line of hex. 1850 for (std::size_t I = 0; I < 16; ++I) { 1851 if (I != 0 && I % 4 == 0) 1852 outs() << ' '; 1853 if (Addr + I < End) 1854 outs() << hexdigit((Contents[Addr + I] >> 4) & 0xF, true) 1855 << hexdigit(Contents[Addr + I] & 0xF, true); 1856 else 1857 outs() << " "; 1858 } 1859 // Print ascii. 1860 outs() << " "; 1861 for (std::size_t I = 0; I < 16 && Addr + I < End; ++I) { 1862 if (isPrint(static_cast<unsigned char>(Contents[Addr + I]) & 0xFF)) 1863 outs() << Contents[Addr + I]; 1864 else 1865 outs() << "."; 1866 } 1867 outs() << "\n"; 1868 } 1869 } 1870 } 1871 1872 void objdump::printSymbolTable(const ObjectFile *O, StringRef ArchiveName, 1873 StringRef ArchitectureName, bool DumpDynamic) { 1874 if (O->isCOFF() && !DumpDynamic) { 1875 outs() << "\nSYMBOL TABLE:\n"; 1876 printCOFFSymbolTable(cast<const COFFObjectFile>(O)); 1877 return; 1878 } 1879 1880 const StringRef FileName = O->getFileName(); 1881 1882 if (!DumpDynamic) { 1883 outs() << "\nSYMBOL TABLE:\n"; 1884 for (auto I = O->symbol_begin(); I != O->symbol_end(); ++I) 1885 printSymbol(O, *I, FileName, ArchiveName, ArchitectureName, DumpDynamic); 1886 return; 1887 } 1888 1889 outs() << "\nDYNAMIC SYMBOL TABLE:\n"; 1890 if (!O->isELF()) { 1891 reportWarning( 1892 "this operation is not currently supported for this file format", 1893 FileName); 1894 return; 1895 } 1896 1897 const ELFObjectFileBase *ELF = cast<const ELFObjectFileBase>(O); 1898 for (auto I = ELF->getDynamicSymbolIterators().begin(); 1899 I != ELF->getDynamicSymbolIterators().end(); ++I) 1900 printSymbol(O, *I, FileName, ArchiveName, ArchitectureName, DumpDynamic); 1901 } 1902 1903 void objdump::printSymbol(const ObjectFile *O, const SymbolRef &Symbol, 1904 StringRef FileName, StringRef ArchiveName, 1905 StringRef ArchitectureName, bool DumpDynamic) { 1906 const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(O); 1907 uint64_t Address = unwrapOrError(Symbol.getAddress(), FileName, ArchiveName, 1908 ArchitectureName); 1909 if ((Address < StartAddress) || (Address > StopAddress)) 1910 return; 1911 SymbolRef::Type Type = 1912 unwrapOrError(Symbol.getType(), FileName, ArchiveName, ArchitectureName); 1913 uint32_t Flags = 1914 unwrapOrError(Symbol.getFlags(), FileName, ArchiveName, ArchitectureName); 1915 1916 // Don't ask a Mach-O STAB symbol for its section unless you know that 1917 // STAB symbol's section field refers to a valid section index. Otherwise 1918 // the symbol may error trying to load a section that does not exist. 1919 bool IsSTAB = false; 1920 if (MachO) { 1921 DataRefImpl SymDRI = Symbol.getRawDataRefImpl(); 1922 uint8_t NType = 1923 (MachO->is64Bit() ? MachO->getSymbol64TableEntry(SymDRI).n_type 1924 : MachO->getSymbolTableEntry(SymDRI).n_type); 1925 if (NType & MachO::N_STAB) 1926 IsSTAB = true; 1927 } 1928 section_iterator Section = IsSTAB 1929 ? O->section_end() 1930 : unwrapOrError(Symbol.getSection(), FileName, 1931 ArchiveName, ArchitectureName); 1932 1933 StringRef Name; 1934 if (Type == SymbolRef::ST_Debug && Section != O->section_end()) { 1935 if (Expected<StringRef> NameOrErr = Section->getName()) 1936 Name = *NameOrErr; 1937 else 1938 consumeError(NameOrErr.takeError()); 1939 1940 } else { 1941 Name = unwrapOrError(Symbol.getName(), FileName, ArchiveName, 1942 ArchitectureName); 1943 } 1944 1945 bool Global = Flags & SymbolRef::SF_Global; 1946 bool Weak = Flags & SymbolRef::SF_Weak; 1947 bool Absolute = Flags & SymbolRef::SF_Absolute; 1948 bool Common = Flags & SymbolRef::SF_Common; 1949 bool Hidden = Flags & SymbolRef::SF_Hidden; 1950 1951 char GlobLoc = ' '; 1952 if ((Section != O->section_end() || Absolute) && !Weak) 1953 GlobLoc = Global ? 'g' : 'l'; 1954 char IFunc = ' '; 1955 if (O->isELF()) { 1956 if (ELFSymbolRef(Symbol).getELFType() == ELF::STT_GNU_IFUNC) 1957 IFunc = 'i'; 1958 if (ELFSymbolRef(Symbol).getBinding() == ELF::STB_GNU_UNIQUE) 1959 GlobLoc = 'u'; 1960 } 1961 1962 char Debug = ' '; 1963 if (DumpDynamic) 1964 Debug = 'D'; 1965 else if (Type == SymbolRef::ST_Debug || Type == SymbolRef::ST_File) 1966 Debug = 'd'; 1967 1968 char FileFunc = ' '; 1969 if (Type == SymbolRef::ST_File) 1970 FileFunc = 'f'; 1971 else if (Type == SymbolRef::ST_Function) 1972 FileFunc = 'F'; 1973 else if (Type == SymbolRef::ST_Data) 1974 FileFunc = 'O'; 1975 1976 const char *Fmt = O->getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 1977 1978 outs() << format(Fmt, Address) << " " 1979 << GlobLoc // Local -> 'l', Global -> 'g', Neither -> ' ' 1980 << (Weak ? 'w' : ' ') // Weak? 1981 << ' ' // Constructor. Not supported yet. 1982 << ' ' // Warning. Not supported yet. 1983 << IFunc // Indirect reference to another symbol. 1984 << Debug // Debugging (d) or dynamic (D) symbol. 1985 << FileFunc // Name of function (F), file (f) or object (O). 1986 << ' '; 1987 if (Absolute) { 1988 outs() << "*ABS*"; 1989 } else if (Common) { 1990 outs() << "*COM*"; 1991 } else if (Section == O->section_end()) { 1992 outs() << "*UND*"; 1993 } else { 1994 StringRef SegmentName = getSegmentName(MachO, *Section); 1995 if (!SegmentName.empty()) 1996 outs() << SegmentName << ","; 1997 StringRef SectionName = unwrapOrError(Section->getName(), FileName); 1998 outs() << SectionName; 1999 } 2000 2001 if (Common || O->isELF()) { 2002 uint64_t Val = 2003 Common ? Symbol.getAlignment() : ELFSymbolRef(Symbol).getSize(); 2004 outs() << '\t' << format(Fmt, Val); 2005 } 2006 2007 if (O->isELF()) { 2008 uint8_t Other = ELFSymbolRef(Symbol).getOther(); 2009 switch (Other) { 2010 case ELF::STV_DEFAULT: 2011 break; 2012 case ELF::STV_INTERNAL: 2013 outs() << " .internal"; 2014 break; 2015 case ELF::STV_HIDDEN: 2016 outs() << " .hidden"; 2017 break; 2018 case ELF::STV_PROTECTED: 2019 outs() << " .protected"; 2020 break; 2021 default: 2022 outs() << format(" 0x%02x", Other); 2023 break; 2024 } 2025 } else if (Hidden) { 2026 outs() << " .hidden"; 2027 } 2028 2029 if (Demangle) 2030 outs() << ' ' << demangle(std::string(Name)) << '\n'; 2031 else 2032 outs() << ' ' << Name << '\n'; 2033 } 2034 2035 static void printUnwindInfo(const ObjectFile *O) { 2036 outs() << "Unwind info:\n\n"; 2037 2038 if (const COFFObjectFile *Coff = dyn_cast<COFFObjectFile>(O)) 2039 printCOFFUnwindInfo(Coff); 2040 else if (const MachOObjectFile *MachO = dyn_cast<MachOObjectFile>(O)) 2041 printMachOUnwindInfo(MachO); 2042 else 2043 // TODO: Extract DWARF dump tool to objdump. 2044 WithColor::error(errs(), ToolName) 2045 << "This operation is only currently supported " 2046 "for COFF and MachO object files.\n"; 2047 } 2048 2049 /// Dump the raw contents of the __clangast section so the output can be piped 2050 /// into llvm-bcanalyzer. 2051 static void printRawClangAST(const ObjectFile *Obj) { 2052 if (outs().is_displayed()) { 2053 WithColor::error(errs(), ToolName) 2054 << "The -raw-clang-ast option will dump the raw binary contents of " 2055 "the clang ast section.\n" 2056 "Please redirect the output to a file or another program such as " 2057 "llvm-bcanalyzer.\n"; 2058 return; 2059 } 2060 2061 StringRef ClangASTSectionName("__clangast"); 2062 if (Obj->isCOFF()) { 2063 ClangASTSectionName = "clangast"; 2064 } 2065 2066 Optional<object::SectionRef> ClangASTSection; 2067 for (auto Sec : ToolSectionFilter(*Obj)) { 2068 StringRef Name; 2069 if (Expected<StringRef> NameOrErr = Sec.getName()) 2070 Name = *NameOrErr; 2071 else 2072 consumeError(NameOrErr.takeError()); 2073 2074 if (Name == ClangASTSectionName) { 2075 ClangASTSection = Sec; 2076 break; 2077 } 2078 } 2079 if (!ClangASTSection) 2080 return; 2081 2082 StringRef ClangASTContents = unwrapOrError( 2083 ClangASTSection.getValue().getContents(), Obj->getFileName()); 2084 outs().write(ClangASTContents.data(), ClangASTContents.size()); 2085 } 2086 2087 static void printFaultMaps(const ObjectFile *Obj) { 2088 StringRef FaultMapSectionName; 2089 2090 if (Obj->isELF()) { 2091 FaultMapSectionName = ".llvm_faultmaps"; 2092 } else if (Obj->isMachO()) { 2093 FaultMapSectionName = "__llvm_faultmaps"; 2094 } else { 2095 WithColor::error(errs(), ToolName) 2096 << "This operation is only currently supported " 2097 "for ELF and Mach-O executable files.\n"; 2098 return; 2099 } 2100 2101 Optional<object::SectionRef> FaultMapSection; 2102 2103 for (auto Sec : ToolSectionFilter(*Obj)) { 2104 StringRef Name; 2105 if (Expected<StringRef> NameOrErr = Sec.getName()) 2106 Name = *NameOrErr; 2107 else 2108 consumeError(NameOrErr.takeError()); 2109 2110 if (Name == FaultMapSectionName) { 2111 FaultMapSection = Sec; 2112 break; 2113 } 2114 } 2115 2116 outs() << "FaultMap table:\n"; 2117 2118 if (!FaultMapSection.hasValue()) { 2119 outs() << "<not found>\n"; 2120 return; 2121 } 2122 2123 StringRef FaultMapContents = 2124 unwrapOrError(FaultMapSection.getValue().getContents(), Obj->getFileName()); 2125 FaultMapParser FMP(FaultMapContents.bytes_begin(), 2126 FaultMapContents.bytes_end()); 2127 2128 outs() << FMP; 2129 } 2130 2131 static void printPrivateFileHeaders(const ObjectFile *O, bool OnlyFirst) { 2132 if (O->isELF()) { 2133 printELFFileHeader(O); 2134 printELFDynamicSection(O); 2135 printELFSymbolVersionInfo(O); 2136 return; 2137 } 2138 if (O->isCOFF()) 2139 return printCOFFFileHeader(O); 2140 if (O->isWasm()) 2141 return printWasmFileHeader(O); 2142 if (O->isMachO()) { 2143 printMachOFileHeader(O); 2144 if (!OnlyFirst) 2145 printMachOLoadCommands(O); 2146 return; 2147 } 2148 reportError(O->getFileName(), "Invalid/Unsupported object file format"); 2149 } 2150 2151 static void printFileHeaders(const ObjectFile *O) { 2152 if (!O->isELF() && !O->isCOFF()) 2153 reportError(O->getFileName(), "Invalid/Unsupported object file format"); 2154 2155 Triple::ArchType AT = O->getArch(); 2156 outs() << "architecture: " << Triple::getArchTypeName(AT) << "\n"; 2157 uint64_t Address = unwrapOrError(O->getStartAddress(), O->getFileName()); 2158 2159 StringRef Fmt = O->getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 2160 outs() << "start address: " 2161 << "0x" << format(Fmt.data(), Address) << "\n"; 2162 } 2163 2164 static void printArchiveChild(StringRef Filename, const Archive::Child &C) { 2165 Expected<sys::fs::perms> ModeOrErr = C.getAccessMode(); 2166 if (!ModeOrErr) { 2167 WithColor::error(errs(), ToolName) << "ill-formed archive entry.\n"; 2168 consumeError(ModeOrErr.takeError()); 2169 return; 2170 } 2171 sys::fs::perms Mode = ModeOrErr.get(); 2172 outs() << ((Mode & sys::fs::owner_read) ? "r" : "-"); 2173 outs() << ((Mode & sys::fs::owner_write) ? "w" : "-"); 2174 outs() << ((Mode & sys::fs::owner_exe) ? "x" : "-"); 2175 outs() << ((Mode & sys::fs::group_read) ? "r" : "-"); 2176 outs() << ((Mode & sys::fs::group_write) ? "w" : "-"); 2177 outs() << ((Mode & sys::fs::group_exe) ? "x" : "-"); 2178 outs() << ((Mode & sys::fs::others_read) ? "r" : "-"); 2179 outs() << ((Mode & sys::fs::others_write) ? "w" : "-"); 2180 outs() << ((Mode & sys::fs::others_exe) ? "x" : "-"); 2181 2182 outs() << " "; 2183 2184 outs() << format("%d/%d %6" PRId64 " ", unwrapOrError(C.getUID(), Filename), 2185 unwrapOrError(C.getGID(), Filename), 2186 unwrapOrError(C.getRawSize(), Filename)); 2187 2188 StringRef RawLastModified = C.getRawLastModified(); 2189 unsigned Seconds; 2190 if (RawLastModified.getAsInteger(10, Seconds)) 2191 outs() << "(date: \"" << RawLastModified 2192 << "\" contains non-decimal chars) "; 2193 else { 2194 // Since ctime(3) returns a 26 character string of the form: 2195 // "Sun Sep 16 01:03:52 1973\n\0" 2196 // just print 24 characters. 2197 time_t t = Seconds; 2198 outs() << format("%.24s ", ctime(&t)); 2199 } 2200 2201 StringRef Name = ""; 2202 Expected<StringRef> NameOrErr = C.getName(); 2203 if (!NameOrErr) { 2204 consumeError(NameOrErr.takeError()); 2205 Name = unwrapOrError(C.getRawName(), Filename); 2206 } else { 2207 Name = NameOrErr.get(); 2208 } 2209 outs() << Name << "\n"; 2210 } 2211 2212 // For ELF only now. 2213 static bool shouldWarnForInvalidStartStopAddress(ObjectFile *Obj) { 2214 if (const auto *Elf = dyn_cast<ELFObjectFileBase>(Obj)) { 2215 if (Elf->getEType() != ELF::ET_REL) 2216 return true; 2217 } 2218 return false; 2219 } 2220 2221 static void checkForInvalidStartStopAddress(ObjectFile *Obj, 2222 uint64_t Start, uint64_t Stop) { 2223 if (!shouldWarnForInvalidStartStopAddress(Obj)) 2224 return; 2225 2226 for (const SectionRef &Section : Obj->sections()) 2227 if (ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC) { 2228 uint64_t BaseAddr = Section.getAddress(); 2229 uint64_t Size = Section.getSize(); 2230 if ((Start < BaseAddr + Size) && Stop > BaseAddr) 2231 return; 2232 } 2233 2234 if (!HasStartAddressFlag) 2235 reportWarning("no section has address less than 0x" + 2236 Twine::utohexstr(Stop) + " specified by --stop-address", 2237 Obj->getFileName()); 2238 else if (!HasStopAddressFlag) 2239 reportWarning("no section has address greater than or equal to 0x" + 2240 Twine::utohexstr(Start) + " specified by --start-address", 2241 Obj->getFileName()); 2242 else 2243 reportWarning("no section overlaps the range [0x" + 2244 Twine::utohexstr(Start) + ",0x" + Twine::utohexstr(Stop) + 2245 ") specified by --start-address/--stop-address", 2246 Obj->getFileName()); 2247 } 2248 2249 static void dumpObject(ObjectFile *O, const Archive *A = nullptr, 2250 const Archive::Child *C = nullptr) { 2251 // Avoid other output when using a raw option. 2252 if (!RawClangAST) { 2253 outs() << '\n'; 2254 if (A) 2255 outs() << A->getFileName() << "(" << O->getFileName() << ")"; 2256 else 2257 outs() << O->getFileName(); 2258 outs() << ":\tfile format " << O->getFileFormatName().lower() << "\n"; 2259 } 2260 2261 if (HasStartAddressFlag || HasStopAddressFlag) 2262 checkForInvalidStartStopAddress(O, StartAddress, StopAddress); 2263 2264 // Note: the order here matches GNU objdump for compatability. 2265 StringRef ArchiveName = A ? A->getFileName() : ""; 2266 if (ArchiveHeaders && !MachOOpt && C) 2267 printArchiveChild(ArchiveName, *C); 2268 if (FileHeaders) 2269 printFileHeaders(O); 2270 if (PrivateHeaders || FirstPrivateHeader) 2271 printPrivateFileHeaders(O, FirstPrivateHeader); 2272 if (SectionHeaders) 2273 printSectionHeaders(O); 2274 if (SymbolTable) 2275 printSymbolTable(O, ArchiveName); 2276 if (DynamicSymbolTable) 2277 printSymbolTable(O, ArchiveName, /*ArchitectureName=*/"", 2278 /*DumpDynamic=*/true); 2279 if (DwarfDumpType != DIDT_Null) { 2280 std::unique_ptr<DIContext> DICtx = DWARFContext::create(*O); 2281 // Dump the complete DWARF structure. 2282 DIDumpOptions DumpOpts; 2283 DumpOpts.DumpType = DwarfDumpType; 2284 DICtx->dump(outs(), DumpOpts); 2285 } 2286 if (Relocations && !Disassemble) 2287 printRelocations(O); 2288 if (DynamicRelocations) 2289 printDynamicRelocations(O); 2290 if (SectionContents) 2291 printSectionContents(O); 2292 if (Disassemble) 2293 disassembleObject(O, Relocations); 2294 if (UnwindInfo) 2295 printUnwindInfo(O); 2296 2297 // Mach-O specific options: 2298 if (ExportsTrie) 2299 printExportsTrie(O); 2300 if (Rebase) 2301 printRebaseTable(O); 2302 if (Bind) 2303 printBindTable(O); 2304 if (LazyBind) 2305 printLazyBindTable(O); 2306 if (WeakBind) 2307 printWeakBindTable(O); 2308 2309 // Other special sections: 2310 if (RawClangAST) 2311 printRawClangAST(O); 2312 if (FaultMapSection) 2313 printFaultMaps(O); 2314 } 2315 2316 static void dumpObject(const COFFImportFile *I, const Archive *A, 2317 const Archive::Child *C = nullptr) { 2318 StringRef ArchiveName = A ? A->getFileName() : ""; 2319 2320 // Avoid other output when using a raw option. 2321 if (!RawClangAST) 2322 outs() << '\n' 2323 << ArchiveName << "(" << I->getFileName() << ")" 2324 << ":\tfile format COFF-import-file" 2325 << "\n\n"; 2326 2327 if (ArchiveHeaders && !MachOOpt && C) 2328 printArchiveChild(ArchiveName, *C); 2329 if (SymbolTable) 2330 printCOFFSymbolTable(I); 2331 } 2332 2333 /// Dump each object file in \a a; 2334 static void dumpArchive(const Archive *A) { 2335 Error Err = Error::success(); 2336 unsigned I = -1; 2337 for (auto &C : A->children(Err)) { 2338 ++I; 2339 Expected<std::unique_ptr<Binary>> ChildOrErr = C.getAsBinary(); 2340 if (!ChildOrErr) { 2341 if (auto E = isNotObjectErrorInvalidFileType(ChildOrErr.takeError())) 2342 reportError(std::move(E), getFileNameForError(C, I), A->getFileName()); 2343 continue; 2344 } 2345 if (ObjectFile *O = dyn_cast<ObjectFile>(&*ChildOrErr.get())) 2346 dumpObject(O, A, &C); 2347 else if (COFFImportFile *I = dyn_cast<COFFImportFile>(&*ChildOrErr.get())) 2348 dumpObject(I, A, &C); 2349 else 2350 reportError(errorCodeToError(object_error::invalid_file_type), 2351 A->getFileName()); 2352 } 2353 if (Err) 2354 reportError(std::move(Err), A->getFileName()); 2355 } 2356 2357 /// Open file and figure out how to dump it. 2358 static void dumpInput(StringRef file) { 2359 // If we are using the Mach-O specific object file parser, then let it parse 2360 // the file and process the command line options. So the -arch flags can 2361 // be used to select specific slices, etc. 2362 if (MachOOpt) { 2363 parseInputMachO(file); 2364 return; 2365 } 2366 2367 // Attempt to open the binary. 2368 OwningBinary<Binary> OBinary = unwrapOrError(createBinary(file), file); 2369 Binary &Binary = *OBinary.getBinary(); 2370 2371 if (Archive *A = dyn_cast<Archive>(&Binary)) 2372 dumpArchive(A); 2373 else if (ObjectFile *O = dyn_cast<ObjectFile>(&Binary)) 2374 dumpObject(O); 2375 else if (MachOUniversalBinary *UB = dyn_cast<MachOUniversalBinary>(&Binary)) 2376 parseInputMachO(UB); 2377 else 2378 reportError(errorCodeToError(object_error::invalid_file_type), file); 2379 } 2380 2381 template <typename T> 2382 static void parseIntArg(const llvm::opt::InputArgList &InputArgs, int ID, 2383 T &Value) { 2384 if (const opt::Arg *A = InputArgs.getLastArg(ID)) { 2385 StringRef V(A->getValue()); 2386 if (!llvm::to_integer(V, Value, 0)) { 2387 reportCmdLineError(A->getSpelling() + 2388 ": expected a non-negative integer, but got '" + V + 2389 "'"); 2390 } 2391 } 2392 } 2393 2394 static std::vector<std::string> 2395 commaSeparatedValues(const llvm::opt::InputArgList &InputArgs, int ID) { 2396 std::vector<std::string> Values; 2397 for (StringRef Value : InputArgs.getAllArgValues(ID)) { 2398 llvm::SmallVector<StringRef, 2> SplitValues; 2399 llvm::SplitString(Value, SplitValues, ","); 2400 for (StringRef SplitValue : SplitValues) 2401 Values.push_back(SplitValue.str()); 2402 } 2403 return Values; 2404 } 2405 2406 static void parseOtoolOptions(const llvm::opt::InputArgList &InputArgs) { 2407 MachOOpt = true; 2408 FullLeadingAddr = true; 2409 PrintImmHex = true; 2410 2411 ArchName = InputArgs.getLastArgValue(OTOOL_arch).str(); 2412 LinkOptHints = InputArgs.hasArg(OTOOL_C); 2413 if (InputArgs.hasArg(OTOOL_d)) 2414 FilterSections.push_back("__DATA,__data"); 2415 DylibId = InputArgs.hasArg(OTOOL_D); 2416 UniversalHeaders = InputArgs.hasArg(OTOOL_f); 2417 DataInCode = InputArgs.hasArg(OTOOL_G); 2418 FirstPrivateHeader = InputArgs.hasArg(OTOOL_h); 2419 IndirectSymbols = InputArgs.hasArg(OTOOL_I); 2420 ShowRawInsn = InputArgs.hasArg(OTOOL_j); 2421 PrivateHeaders = InputArgs.hasArg(OTOOL_l); 2422 DylibsUsed = InputArgs.hasArg(OTOOL_L); 2423 MCPU = InputArgs.getLastArgValue(OTOOL_mcpu_EQ).str(); 2424 ObjcMetaData = InputArgs.hasArg(OTOOL_o); 2425 DisSymName = InputArgs.getLastArgValue(OTOOL_p).str(); 2426 InfoPlist = InputArgs.hasArg(OTOOL_P); 2427 Relocations = InputArgs.hasArg(OTOOL_r); 2428 if (const Arg *A = InputArgs.getLastArg(OTOOL_s)) { 2429 auto Filter = (A->getValue(0) + StringRef(",") + A->getValue(1)).str(); 2430 FilterSections.push_back(Filter); 2431 } 2432 if (InputArgs.hasArg(OTOOL_t)) 2433 FilterSections.push_back("__TEXT,__text"); 2434 Verbose = InputArgs.hasArg(OTOOL_v) || InputArgs.hasArg(OTOOL_V) || 2435 InputArgs.hasArg(OTOOL_o); 2436 SymbolicOperands = InputArgs.hasArg(OTOOL_V); 2437 if (InputArgs.hasArg(OTOOL_x)) 2438 FilterSections.push_back(",__text"); 2439 LeadingAddr = LeadingHeaders = !InputArgs.hasArg(OTOOL_X); 2440 2441 InputFilenames = InputArgs.getAllArgValues(OTOOL_INPUT); 2442 if (InputFilenames.empty()) 2443 reportCmdLineError("no input file"); 2444 2445 for (const Arg *A : InputArgs) { 2446 const Option &O = A->getOption(); 2447 if (O.getGroup().isValid() && O.getGroup().getID() == OTOOL_grp_obsolete) { 2448 reportCmdLineWarning(O.getPrefixedName() + 2449 " is obsolete and not implemented"); 2450 } 2451 } 2452 } 2453 2454 static void parseObjdumpOptions(const llvm::opt::InputArgList &InputArgs) { 2455 parseIntArg(InputArgs, OBJDUMP_adjust_vma_EQ, AdjustVMA); 2456 AllHeaders = InputArgs.hasArg(OBJDUMP_all_headers); 2457 ArchName = InputArgs.getLastArgValue(OBJDUMP_arch_name_EQ).str(); 2458 ArchiveHeaders = InputArgs.hasArg(OBJDUMP_archive_headers); 2459 Demangle = InputArgs.hasArg(OBJDUMP_demangle); 2460 Disassemble = InputArgs.hasArg(OBJDUMP_disassemble); 2461 DisassembleAll = InputArgs.hasArg(OBJDUMP_disassemble_all); 2462 SymbolDescription = InputArgs.hasArg(OBJDUMP_symbol_description); 2463 DisassembleSymbols = 2464 commaSeparatedValues(InputArgs, OBJDUMP_disassemble_symbols_EQ); 2465 DisassembleZeroes = InputArgs.hasArg(OBJDUMP_disassemble_zeroes); 2466 if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_dwarf_EQ)) { 2467 DwarfDumpType = 2468 StringSwitch<DIDumpType>(A->getValue()).Case("frames", DIDT_DebugFrame); 2469 } 2470 DynamicRelocations = InputArgs.hasArg(OBJDUMP_dynamic_reloc); 2471 FaultMapSection = InputArgs.hasArg(OBJDUMP_fault_map_section); 2472 FileHeaders = InputArgs.hasArg(OBJDUMP_file_headers); 2473 SectionContents = InputArgs.hasArg(OBJDUMP_full_contents); 2474 PrintLines = InputArgs.hasArg(OBJDUMP_line_numbers); 2475 InputFilenames = InputArgs.getAllArgValues(OBJDUMP_INPUT); 2476 MachOOpt = InputArgs.hasArg(OBJDUMP_macho); 2477 MCPU = InputArgs.getLastArgValue(OBJDUMP_mcpu_EQ).str(); 2478 MAttrs = commaSeparatedValues(InputArgs, OBJDUMP_mattr_EQ); 2479 ShowRawInsn = !InputArgs.hasArg(OBJDUMP_no_show_raw_insn); 2480 LeadingAddr = !InputArgs.hasArg(OBJDUMP_no_leading_addr); 2481 RawClangAST = InputArgs.hasArg(OBJDUMP_raw_clang_ast); 2482 Relocations = InputArgs.hasArg(OBJDUMP_reloc); 2483 PrintImmHex = 2484 InputArgs.hasFlag(OBJDUMP_print_imm_hex, OBJDUMP_no_print_imm_hex, false); 2485 PrivateHeaders = InputArgs.hasArg(OBJDUMP_private_headers); 2486 FilterSections = InputArgs.getAllArgValues(OBJDUMP_section_EQ); 2487 SectionHeaders = InputArgs.hasArg(OBJDUMP_section_headers); 2488 ShowLMA = InputArgs.hasArg(OBJDUMP_show_lma); 2489 PrintSource = InputArgs.hasArg(OBJDUMP_source); 2490 parseIntArg(InputArgs, OBJDUMP_start_address_EQ, StartAddress); 2491 HasStartAddressFlag = InputArgs.hasArg(OBJDUMP_start_address_EQ); 2492 parseIntArg(InputArgs, OBJDUMP_stop_address_EQ, StopAddress); 2493 HasStopAddressFlag = InputArgs.hasArg(OBJDUMP_stop_address_EQ); 2494 SymbolTable = InputArgs.hasArg(OBJDUMP_syms); 2495 SymbolizeOperands = InputArgs.hasArg(OBJDUMP_symbolize_operands); 2496 DynamicSymbolTable = InputArgs.hasArg(OBJDUMP_dynamic_syms); 2497 TripleName = InputArgs.getLastArgValue(OBJDUMP_triple_EQ).str(); 2498 UnwindInfo = InputArgs.hasArg(OBJDUMP_unwind_info); 2499 Wide = InputArgs.hasArg(OBJDUMP_wide); 2500 Prefix = InputArgs.getLastArgValue(OBJDUMP_prefix).str(); 2501 parseIntArg(InputArgs, OBJDUMP_prefix_strip, PrefixStrip); 2502 if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_debug_vars_EQ)) { 2503 DbgVariables = StringSwitch<DebugVarsFormat>(A->getValue()) 2504 .Case("ascii", DVASCII) 2505 .Case("unicode", DVUnicode); 2506 } 2507 parseIntArg(InputArgs, OBJDUMP_debug_vars_indent_EQ, DbgIndent); 2508 2509 parseMachOOptions(InputArgs); 2510 2511 // Parse -M (--disassembler-options) and deprecated 2512 // --x86-asm-syntax={att,intel}. 2513 // 2514 // Note, for x86, the asm dialect (AssemblerDialect) is initialized when the 2515 // MCAsmInfo is constructed. MCInstPrinter::applyTargetSpecificCLOption is 2516 // called too late. For now we have to use the internal cl::opt option. 2517 const char *AsmSyntax = nullptr; 2518 for (const auto *A : InputArgs.filtered(OBJDUMP_disassembler_options_EQ, 2519 OBJDUMP_x86_asm_syntax_att, 2520 OBJDUMP_x86_asm_syntax_intel)) { 2521 switch (A->getOption().getID()) { 2522 case OBJDUMP_x86_asm_syntax_att: 2523 AsmSyntax = "--x86-asm-syntax=att"; 2524 continue; 2525 case OBJDUMP_x86_asm_syntax_intel: 2526 AsmSyntax = "--x86-asm-syntax=intel"; 2527 continue; 2528 } 2529 2530 SmallVector<StringRef, 2> Values; 2531 llvm::SplitString(A->getValue(), Values, ","); 2532 for (StringRef V : Values) { 2533 if (V == "att") 2534 AsmSyntax = "--x86-asm-syntax=att"; 2535 else if (V == "intel") 2536 AsmSyntax = "--x86-asm-syntax=intel"; 2537 else 2538 DisassemblerOptions.push_back(V.str()); 2539 } 2540 } 2541 if (AsmSyntax) { 2542 const char *Argv[] = {"llvm-objdump", AsmSyntax}; 2543 llvm::cl::ParseCommandLineOptions(2, Argv); 2544 } 2545 2546 // objdump defaults to a.out if no filenames specified. 2547 if (InputFilenames.empty()) 2548 InputFilenames.push_back("a.out"); 2549 } 2550 2551 int main(int argc, char **argv) { 2552 using namespace llvm; 2553 InitLLVM X(argc, argv); 2554 2555 ToolName = argv[0]; 2556 std::unique_ptr<CommonOptTable> T; 2557 OptSpecifier Unknown, HelpFlag, HelpHiddenFlag, VersionFlag; 2558 2559 StringRef Stem = sys::path::stem(ToolName); 2560 auto Is = [=](StringRef Tool) { 2561 // We need to recognize the following filenames: 2562 // 2563 // llvm-objdump -> objdump 2564 // llvm-otool-10.exe -> otool 2565 // powerpc64-unknown-freebsd13-objdump -> objdump 2566 auto I = Stem.rfind_insensitive(Tool); 2567 return I != StringRef::npos && 2568 (I + Tool.size() == Stem.size() || !isAlnum(Stem[I + Tool.size()])); 2569 }; 2570 if (Is("otool")) { 2571 T = std::make_unique<OtoolOptTable>(); 2572 Unknown = OTOOL_UNKNOWN; 2573 HelpFlag = OTOOL_help; 2574 HelpHiddenFlag = OTOOL_help_hidden; 2575 VersionFlag = OTOOL_version; 2576 } else { 2577 T = std::make_unique<ObjdumpOptTable>(); 2578 Unknown = OBJDUMP_UNKNOWN; 2579 HelpFlag = OBJDUMP_help; 2580 HelpHiddenFlag = OBJDUMP_help_hidden; 2581 VersionFlag = OBJDUMP_version; 2582 } 2583 2584 BumpPtrAllocator A; 2585 StringSaver Saver(A); 2586 opt::InputArgList InputArgs = 2587 T->parseArgs(argc, argv, Unknown, Saver, 2588 [&](StringRef Msg) { reportCmdLineError(Msg); }); 2589 2590 if (InputArgs.size() == 0 || InputArgs.hasArg(HelpFlag)) { 2591 T->printHelp(ToolName); 2592 return 0; 2593 } 2594 if (InputArgs.hasArg(HelpHiddenFlag)) { 2595 T->printHelp(ToolName, /*show_hidden=*/true); 2596 return 0; 2597 } 2598 2599 // Initialize targets and assembly printers/parsers. 2600 InitializeAllTargetInfos(); 2601 InitializeAllTargetMCs(); 2602 InitializeAllDisassemblers(); 2603 2604 if (InputArgs.hasArg(VersionFlag)) { 2605 cl::PrintVersionMessage(); 2606 if (!Is("otool")) { 2607 outs() << '\n'; 2608 TargetRegistry::printRegisteredTargetsForVersion(outs()); 2609 } 2610 return 0; 2611 } 2612 2613 if (Is("otool")) 2614 parseOtoolOptions(InputArgs); 2615 else 2616 parseObjdumpOptions(InputArgs); 2617 2618 if (StartAddress >= StopAddress) 2619 reportCmdLineError("start address should be less than stop address"); 2620 2621 // Removes trailing separators from prefix. 2622 while (!Prefix.empty() && sys::path::is_separator(Prefix.back())) 2623 Prefix.pop_back(); 2624 2625 if (AllHeaders) 2626 ArchiveHeaders = FileHeaders = PrivateHeaders = Relocations = 2627 SectionHeaders = SymbolTable = true; 2628 2629 if (DisassembleAll || PrintSource || PrintLines || 2630 !DisassembleSymbols.empty()) 2631 Disassemble = true; 2632 2633 if (!ArchiveHeaders && !Disassemble && DwarfDumpType == DIDT_Null && 2634 !DynamicRelocations && !FileHeaders && !PrivateHeaders && !RawClangAST && 2635 !Relocations && !SectionHeaders && !SectionContents && !SymbolTable && 2636 !DynamicSymbolTable && !UnwindInfo && !FaultMapSection && 2637 !(MachOOpt && 2638 (Bind || DataInCode || DylibId || DylibsUsed || ExportsTrie || 2639 FirstPrivateHeader || FunctionStarts || IndirectSymbols || InfoPlist || 2640 LazyBind || LinkOptHints || ObjcMetaData || Rebase || Rpaths || 2641 UniversalHeaders || WeakBind || !FilterSections.empty()))) { 2642 T->printHelp(ToolName); 2643 return 2; 2644 } 2645 2646 DisasmSymbolSet.insert(DisassembleSymbols.begin(), DisassembleSymbols.end()); 2647 2648 llvm::for_each(InputFilenames, dumpInput); 2649 2650 warnOnNoMatchForSections(); 2651 2652 return EXIT_SUCCESS; 2653 } 2654