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 "llvm/ADT/Optional.h" 20 #include "llvm/ADT/STLExtras.h" 21 #include "llvm/ADT/SetOperations.h" 22 #include "llvm/ADT/StringExtras.h" 23 #include "llvm/ADT/StringSet.h" 24 #include "llvm/ADT/Triple.h" 25 #include "llvm/CodeGen/FaultMaps.h" 26 #include "llvm/DebugInfo/DWARF/DWARFContext.h" 27 #include "llvm/DebugInfo/Symbolize/Symbolize.h" 28 #include "llvm/Demangle/Demangle.h" 29 #include "llvm/MC/MCAsmInfo.h" 30 #include "llvm/MC/MCContext.h" 31 #include "llvm/MC/MCDisassembler/MCDisassembler.h" 32 #include "llvm/MC/MCDisassembler/MCRelocationInfo.h" 33 #include "llvm/MC/MCInst.h" 34 #include "llvm/MC/MCInstPrinter.h" 35 #include "llvm/MC/MCInstrAnalysis.h" 36 #include "llvm/MC/MCInstrInfo.h" 37 #include "llvm/MC/MCObjectFileInfo.h" 38 #include "llvm/MC/MCRegisterInfo.h" 39 #include "llvm/MC/MCSubtargetInfo.h" 40 #include "llvm/Object/Archive.h" 41 #include "llvm/Object/COFF.h" 42 #include "llvm/Object/COFFImportFile.h" 43 #include "llvm/Object/ELFObjectFile.h" 44 #include "llvm/Object/MachO.h" 45 #include "llvm/Object/MachOUniversal.h" 46 #include "llvm/Object/ObjectFile.h" 47 #include "llvm/Object/Wasm.h" 48 #include "llvm/Support/Casting.h" 49 #include "llvm/Support/CommandLine.h" 50 #include "llvm/Support/Debug.h" 51 #include "llvm/Support/Errc.h" 52 #include "llvm/Support/FileSystem.h" 53 #include "llvm/Support/Format.h" 54 #include "llvm/Support/GraphWriter.h" 55 #include "llvm/Support/Host.h" 56 #include "llvm/Support/InitLLVM.h" 57 #include "llvm/Support/MemoryBuffer.h" 58 #include "llvm/Support/SourceMgr.h" 59 #include "llvm/Support/StringSaver.h" 60 #include "llvm/Support/TargetRegistry.h" 61 #include "llvm/Support/TargetSelect.h" 62 #include "llvm/Support/WithColor.h" 63 #include "llvm/Support/raw_ostream.h" 64 #include <algorithm> 65 #include <cctype> 66 #include <cstring> 67 #include <system_error> 68 #include <unordered_map> 69 #include <utility> 70 71 using namespace llvm::object; 72 73 namespace llvm { 74 75 cl::OptionCategory ObjdumpCat("llvm-objdump Options"); 76 77 // MachO specific 78 extern cl::OptionCategory MachOCat; 79 extern cl::opt<bool> Bind; 80 extern cl::opt<bool> DataInCode; 81 extern cl::opt<bool> DylibsUsed; 82 extern cl::opt<bool> DylibId; 83 extern cl::opt<bool> ExportsTrie; 84 extern cl::opt<bool> FirstPrivateHeader; 85 extern cl::opt<bool> IndirectSymbols; 86 extern cl::opt<bool> InfoPlist; 87 extern cl::opt<bool> LazyBind; 88 extern cl::opt<bool> LinkOptHints; 89 extern cl::opt<bool> ObjcMetaData; 90 extern cl::opt<bool> Rebase; 91 extern cl::opt<bool> UniversalHeaders; 92 extern cl::opt<bool> WeakBind; 93 94 static cl::opt<uint64_t> AdjustVMA( 95 "adjust-vma", 96 cl::desc("Increase the displayed address by the specified offset"), 97 cl::value_desc("offset"), cl::init(0), cl::cat(ObjdumpCat)); 98 99 static cl::opt<bool> 100 AllHeaders("all-headers", 101 cl::desc("Display all available header information"), 102 cl::cat(ObjdumpCat)); 103 static cl::alias AllHeadersShort("x", cl::desc("Alias for --all-headers"), 104 cl::NotHidden, cl::Grouping, 105 cl::aliasopt(AllHeaders)); 106 107 static cl::opt<std::string> 108 ArchName("arch-name", 109 cl::desc("Target arch to disassemble for, " 110 "see -version for available targets"), 111 cl::cat(ObjdumpCat)); 112 113 cl::opt<bool> ArchiveHeaders("archive-headers", 114 cl::desc("Display archive header information"), 115 cl::cat(ObjdumpCat)); 116 static cl::alias ArchiveHeadersShort("a", 117 cl::desc("Alias for --archive-headers"), 118 cl::NotHidden, cl::Grouping, 119 cl::aliasopt(ArchiveHeaders)); 120 121 cl::opt<bool> Demangle("demangle", cl::desc("Demangle symbols names"), 122 cl::init(false), cl::cat(ObjdumpCat)); 123 static cl::alias DemangleShort("C", cl::desc("Alias for --demangle"), 124 cl::NotHidden, cl::Grouping, 125 cl::aliasopt(Demangle)); 126 127 cl::opt<bool> Disassemble( 128 "disassemble", 129 cl::desc("Display assembler mnemonics for the machine instructions"), 130 cl::cat(ObjdumpCat)); 131 static cl::alias DisassembleShort("d", cl::desc("Alias for --disassemble"), 132 cl::NotHidden, cl::Grouping, 133 cl::aliasopt(Disassemble)); 134 135 cl::opt<bool> DisassembleAll( 136 "disassemble-all", 137 cl::desc("Display assembler mnemonics for the machine instructions"), 138 cl::cat(ObjdumpCat)); 139 static cl::alias DisassembleAllShort("D", 140 cl::desc("Alias for --disassemble-all"), 141 cl::NotHidden, cl::Grouping, 142 cl::aliasopt(DisassembleAll)); 143 144 static cl::list<std::string> 145 DisassembleFunctions("disassemble-functions", cl::CommaSeparated, 146 cl::desc("List of functions to disassemble. " 147 "Accept demangled names when --demangle is " 148 "specified, otherwise accept mangled names"), 149 cl::cat(ObjdumpCat)); 150 151 static cl::opt<bool> DisassembleZeroes( 152 "disassemble-zeroes", 153 cl::desc("Do not skip blocks of zeroes when disassembling"), 154 cl::cat(ObjdumpCat)); 155 static cl::alias 156 DisassembleZeroesShort("z", cl::desc("Alias for --disassemble-zeroes"), 157 cl::NotHidden, cl::Grouping, 158 cl::aliasopt(DisassembleZeroes)); 159 160 static cl::list<std::string> 161 DisassemblerOptions("disassembler-options", 162 cl::desc("Pass target specific disassembler options"), 163 cl::value_desc("options"), cl::CommaSeparated, 164 cl::cat(ObjdumpCat)); 165 static cl::alias 166 DisassemblerOptionsShort("M", cl::desc("Alias for --disassembler-options"), 167 cl::NotHidden, cl::Grouping, cl::Prefix, 168 cl::CommaSeparated, 169 cl::aliasopt(DisassemblerOptions)); 170 171 cl::opt<DIDumpType> DwarfDumpType( 172 "dwarf", cl::init(DIDT_Null), cl::desc("Dump of dwarf debug sections:"), 173 cl::values(clEnumValN(DIDT_DebugFrame, "frames", ".debug_frame")), 174 cl::cat(ObjdumpCat)); 175 176 static cl::opt<bool> DynamicRelocations( 177 "dynamic-reloc", 178 cl::desc("Display the dynamic relocation entries in the file"), 179 cl::cat(ObjdumpCat)); 180 static cl::alias DynamicRelocationShort("R", 181 cl::desc("Alias for --dynamic-reloc"), 182 cl::NotHidden, cl::Grouping, 183 cl::aliasopt(DynamicRelocations)); 184 185 static cl::opt<bool> 186 FaultMapSection("fault-map-section", 187 cl::desc("Display contents of faultmap section"), 188 cl::cat(ObjdumpCat)); 189 190 static cl::opt<bool> 191 FileHeaders("file-headers", 192 cl::desc("Display the contents of the overall file header"), 193 cl::cat(ObjdumpCat)); 194 static cl::alias FileHeadersShort("f", cl::desc("Alias for --file-headers"), 195 cl::NotHidden, cl::Grouping, 196 cl::aliasopt(FileHeaders)); 197 198 cl::opt<bool> SectionContents("full-contents", 199 cl::desc("Display the content of each section"), 200 cl::cat(ObjdumpCat)); 201 static cl::alias SectionContentsShort("s", 202 cl::desc("Alias for --full-contents"), 203 cl::NotHidden, cl::Grouping, 204 cl::aliasopt(SectionContents)); 205 206 static cl::list<std::string> InputFilenames(cl::Positional, 207 cl::desc("<input object files>"), 208 cl::ZeroOrMore, 209 cl::cat(ObjdumpCat)); 210 211 static cl::opt<bool> 212 PrintLines("line-numbers", 213 cl::desc("Display source line numbers with " 214 "disassembly. Implies disassemble object"), 215 cl::cat(ObjdumpCat)); 216 static cl::alias PrintLinesShort("l", cl::desc("Alias for --line-numbers"), 217 cl::NotHidden, cl::Grouping, 218 cl::aliasopt(PrintLines)); 219 220 static cl::opt<bool> MachOOpt("macho", 221 cl::desc("Use MachO specific object file parser"), 222 cl::cat(ObjdumpCat)); 223 static cl::alias MachOm("m", cl::desc("Alias for --macho"), cl::NotHidden, 224 cl::Grouping, cl::aliasopt(MachOOpt)); 225 226 cl::opt<std::string> 227 MCPU("mcpu", 228 cl::desc("Target a specific cpu type (-mcpu=help for details)"), 229 cl::value_desc("cpu-name"), cl::init(""), cl::cat(ObjdumpCat)); 230 231 cl::list<std::string> MAttrs("mattr", cl::CommaSeparated, 232 cl::desc("Target specific attributes"), 233 cl::value_desc("a1,+a2,-a3,..."), 234 cl::cat(ObjdumpCat)); 235 236 cl::opt<bool> NoShowRawInsn("no-show-raw-insn", 237 cl::desc("When disassembling " 238 "instructions, do not print " 239 "the instruction bytes."), 240 cl::cat(ObjdumpCat)); 241 cl::opt<bool> NoLeadingAddr("no-leading-addr", 242 cl::desc("Print no leading address"), 243 cl::cat(ObjdumpCat)); 244 245 static cl::opt<bool> RawClangAST( 246 "raw-clang-ast", 247 cl::desc("Dump the raw binary contents of the clang AST section"), 248 cl::cat(ObjdumpCat)); 249 250 cl::opt<bool> 251 Relocations("reloc", cl::desc("Display the relocation entries in the file"), 252 cl::cat(ObjdumpCat)); 253 static cl::alias RelocationsShort("r", cl::desc("Alias for --reloc"), 254 cl::NotHidden, cl::Grouping, 255 cl::aliasopt(Relocations)); 256 257 cl::opt<bool> PrintImmHex("print-imm-hex", 258 cl::desc("Use hex format for immediate values"), 259 cl::cat(ObjdumpCat)); 260 261 cl::opt<bool> PrivateHeaders("private-headers", 262 cl::desc("Display format specific file headers"), 263 cl::cat(ObjdumpCat)); 264 static cl::alias PrivateHeadersShort("p", 265 cl::desc("Alias for --private-headers"), 266 cl::NotHidden, cl::Grouping, 267 cl::aliasopt(PrivateHeaders)); 268 269 cl::list<std::string> 270 FilterSections("section", 271 cl::desc("Operate on the specified sections only. " 272 "With -macho dump segment,section"), 273 cl::cat(ObjdumpCat)); 274 static cl::alias FilterSectionsj("j", cl::desc("Alias for --section"), 275 cl::NotHidden, cl::Grouping, cl::Prefix, 276 cl::aliasopt(FilterSections)); 277 278 cl::opt<bool> SectionHeaders("section-headers", 279 cl::desc("Display summaries of the " 280 "headers for each section."), 281 cl::cat(ObjdumpCat)); 282 static cl::alias SectionHeadersShort("headers", 283 cl::desc("Alias for --section-headers"), 284 cl::NotHidden, 285 cl::aliasopt(SectionHeaders)); 286 static cl::alias SectionHeadersShorter("h", 287 cl::desc("Alias for --section-headers"), 288 cl::NotHidden, cl::Grouping, 289 cl::aliasopt(SectionHeaders)); 290 291 static cl::opt<bool> 292 ShowLMA("show-lma", 293 cl::desc("Display LMA column when dumping ELF section headers"), 294 cl::cat(ObjdumpCat)); 295 296 static cl::opt<bool> PrintSource( 297 "source", 298 cl::desc( 299 "Display source inlined with disassembly. Implies disassemble object"), 300 cl::cat(ObjdumpCat)); 301 static cl::alias PrintSourceShort("S", cl::desc("Alias for -source"), 302 cl::NotHidden, cl::Grouping, 303 cl::aliasopt(PrintSource)); 304 305 static cl::opt<uint64_t> 306 StartAddress("start-address", cl::desc("Disassemble beginning at address"), 307 cl::value_desc("address"), cl::init(0), cl::cat(ObjdumpCat)); 308 static cl::opt<uint64_t> StopAddress("stop-address", 309 cl::desc("Stop disassembly at address"), 310 cl::value_desc("address"), 311 cl::init(UINT64_MAX), cl::cat(ObjdumpCat)); 312 313 cl::opt<bool> SymbolTable("syms", cl::desc("Display the symbol table"), 314 cl::cat(ObjdumpCat)); 315 static cl::alias SymbolTableShort("t", cl::desc("Alias for --syms"), 316 cl::NotHidden, cl::Grouping, 317 cl::aliasopt(SymbolTable)); 318 319 cl::opt<std::string> TripleName("triple", 320 cl::desc("Target triple to disassemble for, " 321 "see -version for available targets"), 322 cl::cat(ObjdumpCat)); 323 324 cl::opt<bool> UnwindInfo("unwind-info", cl::desc("Display unwind information"), 325 cl::cat(ObjdumpCat)); 326 static cl::alias UnwindInfoShort("u", cl::desc("Alias for --unwind-info"), 327 cl::NotHidden, cl::Grouping, 328 cl::aliasopt(UnwindInfo)); 329 330 static cl::opt<bool> 331 Wide("wide", cl::desc("Ignored for compatibility with GNU objdump"), 332 cl::cat(ObjdumpCat)); 333 static cl::alias WideShort("w", cl::Grouping, cl::aliasopt(Wide)); 334 335 static cl::extrahelp 336 HelpResponse("\nPass @FILE as argument to read options from FILE.\n"); 337 338 static StringSet<> DisasmFuncsSet; 339 static StringRef ToolName; 340 341 typedef std::vector<std::tuple<uint64_t, StringRef, uint8_t>> SectionSymbolsTy; 342 343 static bool shouldKeep(object::SectionRef S) { 344 if (FilterSections.empty()) 345 return true; 346 StringRef String; 347 std::error_code error = S.getName(String); 348 if (error) 349 return false; 350 return is_contained(FilterSections, String); 351 } 352 353 SectionFilter ToolSectionFilter(object::ObjectFile const &O) { 354 return SectionFilter([](object::SectionRef S) { return shouldKeep(S); }, O); 355 } 356 357 void error(std::error_code EC) { 358 if (!EC) 359 return; 360 WithColor::error(errs(), ToolName) 361 << "reading file: " << EC.message() << ".\n"; 362 errs().flush(); 363 exit(1); 364 } 365 366 void error(Error E) { 367 if (!E) 368 return; 369 WithColor::error(errs(), ToolName) << toString(std::move(E)); 370 exit(1); 371 } 372 373 LLVM_ATTRIBUTE_NORETURN void error(Twine Message) { 374 WithColor::error(errs(), ToolName) << Message << ".\n"; 375 errs().flush(); 376 exit(1); 377 } 378 379 void warn(StringRef Message) { 380 WithColor::warning(errs(), ToolName) << Message << ".\n"; 381 errs().flush(); 382 } 383 384 void warn(Twine Message) { 385 WithColor::warning(errs(), ToolName) << Message << "\n"; 386 } 387 388 LLVM_ATTRIBUTE_NORETURN void report_error(StringRef File, Twine Message) { 389 WithColor::error(errs(), ToolName) 390 << "'" << File << "': " << Message << ".\n"; 391 exit(1); 392 } 393 394 LLVM_ATTRIBUTE_NORETURN void report_error(Error E, StringRef File) { 395 assert(E); 396 std::string Buf; 397 raw_string_ostream OS(Buf); 398 logAllUnhandledErrors(std::move(E), OS); 399 OS.flush(); 400 WithColor::error(errs(), ToolName) << "'" << File << "': " << Buf; 401 exit(1); 402 } 403 404 LLVM_ATTRIBUTE_NORETURN void report_error(Error E, StringRef ArchiveName, 405 StringRef FileName, 406 StringRef ArchitectureName) { 407 assert(E); 408 WithColor::error(errs(), ToolName); 409 if (ArchiveName != "") 410 errs() << ArchiveName << "(" << FileName << ")"; 411 else 412 errs() << "'" << FileName << "'"; 413 if (!ArchitectureName.empty()) 414 errs() << " (for architecture " << ArchitectureName << ")"; 415 std::string Buf; 416 raw_string_ostream OS(Buf); 417 logAllUnhandledErrors(std::move(E), OS); 418 OS.flush(); 419 errs() << ": " << Buf; 420 exit(1); 421 } 422 423 LLVM_ATTRIBUTE_NORETURN void report_error(Error E, StringRef ArchiveName, 424 const object::Archive::Child &C, 425 StringRef ArchitectureName) { 426 Expected<StringRef> NameOrErr = C.getName(); 427 // TODO: if we have a error getting the name then it would be nice to print 428 // the index of which archive member this is and or its offset in the 429 // archive instead of "???" as the name. 430 if (!NameOrErr) { 431 consumeError(NameOrErr.takeError()); 432 report_error(std::move(E), ArchiveName, "???", ArchitectureName); 433 } else 434 report_error(std::move(E), ArchiveName, NameOrErr.get(), ArchitectureName); 435 } 436 437 static const Target *getTarget(const ObjectFile *Obj = nullptr) { 438 // Figure out the target triple. 439 Triple TheTriple("unknown-unknown-unknown"); 440 if (TripleName.empty()) { 441 if (Obj) 442 TheTriple = Obj->makeTriple(); 443 } else { 444 TheTriple.setTriple(Triple::normalize(TripleName)); 445 446 // Use the triple, but also try to combine with ARM build attributes. 447 if (Obj) { 448 auto Arch = Obj->getArch(); 449 if (Arch == Triple::arm || Arch == Triple::armeb) 450 Obj->setARMSubArch(TheTriple); 451 } 452 } 453 454 // Get the target specific parser. 455 std::string Error; 456 const Target *TheTarget = TargetRegistry::lookupTarget(ArchName, TheTriple, 457 Error); 458 if (!TheTarget) { 459 if (Obj) 460 report_error(Obj->getFileName(), "can't find target: " + Error); 461 else 462 error("can't find target: " + Error); 463 } 464 465 // Update the triple name and return the found target. 466 TripleName = TheTriple.getTriple(); 467 return TheTarget; 468 } 469 470 bool isRelocAddressLess(RelocationRef A, RelocationRef B) { 471 return A.getOffset() < B.getOffset(); 472 } 473 474 static Error getRelocationValueString(const RelocationRef &Rel, 475 SmallVectorImpl<char> &Result) { 476 const ObjectFile *Obj = Rel.getObject(); 477 if (auto *ELF = dyn_cast<ELFObjectFileBase>(Obj)) 478 return getELFRelocationValueString(ELF, Rel, Result); 479 if (auto *COFF = dyn_cast<COFFObjectFile>(Obj)) 480 return getCOFFRelocationValueString(COFF, Rel, Result); 481 if (auto *Wasm = dyn_cast<WasmObjectFile>(Obj)) 482 return getWasmRelocationValueString(Wasm, Rel, Result); 483 if (auto *MachO = dyn_cast<MachOObjectFile>(Obj)) 484 return getMachORelocationValueString(MachO, Rel, Result); 485 llvm_unreachable("unknown object file format"); 486 } 487 488 /// Indicates whether this relocation should hidden when listing 489 /// relocations, usually because it is the trailing part of a multipart 490 /// relocation that will be printed as part of the leading relocation. 491 static bool getHidden(RelocationRef RelRef) { 492 auto *MachO = dyn_cast<MachOObjectFile>(RelRef.getObject()); 493 if (!MachO) 494 return false; 495 496 unsigned Arch = MachO->getArch(); 497 DataRefImpl Rel = RelRef.getRawDataRefImpl(); 498 uint64_t Type = MachO->getRelocationType(Rel); 499 500 // On arches that use the generic relocations, GENERIC_RELOC_PAIR 501 // is always hidden. 502 if (Arch == Triple::x86 || Arch == Triple::arm || Arch == Triple::ppc) 503 return Type == MachO::GENERIC_RELOC_PAIR; 504 505 if (Arch == Triple::x86_64) { 506 // On x86_64, X86_64_RELOC_UNSIGNED is hidden only when it follows 507 // an X86_64_RELOC_SUBTRACTOR. 508 if (Type == MachO::X86_64_RELOC_UNSIGNED && Rel.d.a > 0) { 509 DataRefImpl RelPrev = Rel; 510 RelPrev.d.a--; 511 uint64_t PrevType = MachO->getRelocationType(RelPrev); 512 if (PrevType == MachO::X86_64_RELOC_SUBTRACTOR) 513 return true; 514 } 515 } 516 517 return false; 518 } 519 520 namespace { 521 class SourcePrinter { 522 protected: 523 DILineInfo OldLineInfo; 524 const ObjectFile *Obj = nullptr; 525 std::unique_ptr<symbolize::LLVMSymbolizer> Symbolizer; 526 // File name to file contents of source 527 std::unordered_map<std::string, std::unique_ptr<MemoryBuffer>> SourceCache; 528 // Mark the line endings of the cached source 529 std::unordered_map<std::string, std::vector<StringRef>> LineCache; 530 531 private: 532 bool cacheSource(const DILineInfo& LineInfoFile); 533 534 public: 535 SourcePrinter() = default; 536 SourcePrinter(const ObjectFile *Obj, StringRef DefaultArch) : Obj(Obj) { 537 symbolize::LLVMSymbolizer::Options SymbolizerOpts; 538 SymbolizerOpts.PrintFunctions = DILineInfoSpecifier::FunctionNameKind::None; 539 SymbolizerOpts.Demangle = false; 540 SymbolizerOpts.DefaultArch = DefaultArch; 541 Symbolizer.reset(new symbolize::LLVMSymbolizer(SymbolizerOpts)); 542 } 543 virtual ~SourcePrinter() = default; 544 virtual void printSourceLine(raw_ostream &OS, 545 object::SectionedAddress Address, 546 StringRef Delimiter = "; "); 547 }; 548 549 bool SourcePrinter::cacheSource(const DILineInfo &LineInfo) { 550 std::unique_ptr<MemoryBuffer> Buffer; 551 if (LineInfo.Source) { 552 Buffer = MemoryBuffer::getMemBuffer(*LineInfo.Source); 553 } else { 554 auto BufferOrError = MemoryBuffer::getFile(LineInfo.FileName); 555 if (!BufferOrError) 556 return false; 557 Buffer = std::move(*BufferOrError); 558 } 559 // Chomp the file to get lines 560 const char *BufferStart = Buffer->getBufferStart(), 561 *BufferEnd = Buffer->getBufferEnd(); 562 std::vector<StringRef> &Lines = LineCache[LineInfo.FileName]; 563 const char *Start = BufferStart; 564 for (const char *I = BufferStart; I != BufferEnd; ++I) 565 if (*I == '\n') { 566 Lines.emplace_back(Start, I - Start - (BufferStart < I && I[-1] == '\r')); 567 Start = I + 1; 568 } 569 if (Start < BufferEnd) 570 Lines.emplace_back(Start, BufferEnd - Start); 571 SourceCache[LineInfo.FileName] = std::move(Buffer); 572 return true; 573 } 574 575 void SourcePrinter::printSourceLine(raw_ostream &OS, 576 object::SectionedAddress Address, 577 StringRef Delimiter) { 578 if (!Symbolizer) 579 return; 580 581 DILineInfo LineInfo = DILineInfo(); 582 auto ExpectedLineInfo = 583 Symbolizer->symbolizeCode(Obj->getFileName(), Address); 584 if (!ExpectedLineInfo) 585 consumeError(ExpectedLineInfo.takeError()); 586 else 587 LineInfo = *ExpectedLineInfo; 588 589 if ((LineInfo.FileName == "<invalid>") || LineInfo.Line == 0 || 590 ((OldLineInfo.Line == LineInfo.Line) && 591 (OldLineInfo.FileName == LineInfo.FileName))) 592 return; 593 594 if (PrintLines) 595 OS << Delimiter << LineInfo.FileName << ":" << LineInfo.Line << "\n"; 596 if (PrintSource) { 597 if (SourceCache.find(LineInfo.FileName) == SourceCache.end()) 598 if (!cacheSource(LineInfo)) 599 return; 600 auto LineBuffer = LineCache.find(LineInfo.FileName); 601 if (LineBuffer != LineCache.end()) { 602 if (LineInfo.Line > LineBuffer->second.size()) 603 return; 604 // Vector begins at 0, line numbers are non-zero 605 OS << Delimiter << LineBuffer->second[LineInfo.Line - 1] << '\n'; 606 } 607 } 608 OldLineInfo = LineInfo; 609 } 610 611 static bool isAArch64Elf(const ObjectFile *Obj) { 612 const auto *Elf = dyn_cast<ELFObjectFileBase>(Obj); 613 return Elf && Elf->getEMachine() == ELF::EM_AARCH64; 614 } 615 616 static bool isArmElf(const ObjectFile *Obj) { 617 const auto *Elf = dyn_cast<ELFObjectFileBase>(Obj); 618 return Elf && Elf->getEMachine() == ELF::EM_ARM; 619 } 620 621 static bool hasMappingSymbols(const ObjectFile *Obj) { 622 return isArmElf(Obj) || isAArch64Elf(Obj); 623 } 624 625 static void printRelocation(const RelocationRef &Rel, uint64_t Address, 626 bool Is64Bits) { 627 StringRef Fmt = Is64Bits ? "\t\t%016" PRIx64 ": " : "\t\t\t%08" PRIx64 ": "; 628 SmallString<16> Name; 629 SmallString<32> Val; 630 Rel.getTypeName(Name); 631 error(getRelocationValueString(Rel, Val)); 632 outs() << format(Fmt.data(), Address) << Name << "\t" << Val << "\n"; 633 } 634 635 class PrettyPrinter { 636 public: 637 virtual ~PrettyPrinter() = default; 638 virtual void printInst(MCInstPrinter &IP, const MCInst *MI, 639 ArrayRef<uint8_t> Bytes, 640 object::SectionedAddress Address, raw_ostream &OS, 641 StringRef Annot, MCSubtargetInfo const &STI, 642 SourcePrinter *SP, 643 std::vector<RelocationRef> *Rels = nullptr) { 644 if (SP && (PrintSource || PrintLines)) 645 SP->printSourceLine(OS, Address); 646 647 { 648 formatted_raw_ostream FOS(OS); 649 if (!NoLeadingAddr) 650 FOS << format("%8" PRIx64 ":", Address.Address); 651 if (!NoShowRawInsn) { 652 FOS << ' '; 653 dumpBytes(Bytes, FOS); 654 } 655 FOS.flush(); 656 // The output of printInst starts with a tab. Print some spaces so that 657 // the tab has 1 column and advances to the target tab stop. 658 unsigned TabStop = NoShowRawInsn ? 16 : 40; 659 unsigned Column = FOS.getColumn(); 660 FOS.indent(Column < TabStop - 1 ? TabStop - 1 - Column : 7 - Column % 8); 661 662 // The dtor calls flush() to ensure the indent comes before printInst(). 663 } 664 665 if (MI) 666 IP.printInst(MI, OS, "", STI); 667 else 668 OS << "\t<unknown>"; 669 } 670 }; 671 PrettyPrinter PrettyPrinterInst; 672 673 class HexagonPrettyPrinter : public PrettyPrinter { 674 public: 675 void printLead(ArrayRef<uint8_t> Bytes, uint64_t Address, 676 raw_ostream &OS) { 677 uint32_t opcode = 678 (Bytes[3] << 24) | (Bytes[2] << 16) | (Bytes[1] << 8) | Bytes[0]; 679 if (!NoLeadingAddr) 680 OS << format("%8" PRIx64 ":", Address); 681 if (!NoShowRawInsn) { 682 OS << "\t"; 683 dumpBytes(Bytes.slice(0, 4), OS); 684 OS << format("\t%08" PRIx32, opcode); 685 } 686 } 687 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 688 object::SectionedAddress Address, raw_ostream &OS, 689 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 690 std::vector<RelocationRef> *Rels) override { 691 if (SP && (PrintSource || PrintLines)) 692 SP->printSourceLine(OS, Address, ""); 693 if (!MI) { 694 printLead(Bytes, Address.Address, OS); 695 OS << " <unknown>"; 696 return; 697 } 698 std::string Buffer; 699 { 700 raw_string_ostream TempStream(Buffer); 701 IP.printInst(MI, TempStream, "", STI); 702 } 703 StringRef Contents(Buffer); 704 // Split off bundle attributes 705 auto PacketBundle = Contents.rsplit('\n'); 706 // Split off first instruction from the rest 707 auto HeadTail = PacketBundle.first.split('\n'); 708 auto Preamble = " { "; 709 auto Separator = ""; 710 711 // Hexagon's packets require relocations to be inline rather than 712 // clustered at the end of the packet. 713 std::vector<RelocationRef>::const_iterator RelCur = Rels->begin(); 714 std::vector<RelocationRef>::const_iterator RelEnd = Rels->end(); 715 auto PrintReloc = [&]() -> void { 716 while ((RelCur != RelEnd) && (RelCur->getOffset() <= Address.Address)) { 717 if (RelCur->getOffset() == Address.Address) { 718 printRelocation(*RelCur, Address.Address, false); 719 return; 720 } 721 ++RelCur; 722 } 723 }; 724 725 while (!HeadTail.first.empty()) { 726 OS << Separator; 727 Separator = "\n"; 728 if (SP && (PrintSource || PrintLines)) 729 SP->printSourceLine(OS, Address, ""); 730 printLead(Bytes, Address.Address, OS); 731 OS << Preamble; 732 Preamble = " "; 733 StringRef Inst; 734 auto Duplex = HeadTail.first.split('\v'); 735 if (!Duplex.second.empty()) { 736 OS << Duplex.first; 737 OS << "; "; 738 Inst = Duplex.second; 739 } 740 else 741 Inst = HeadTail.first; 742 OS << Inst; 743 HeadTail = HeadTail.second.split('\n'); 744 if (HeadTail.first.empty()) 745 OS << " } " << PacketBundle.second; 746 PrintReloc(); 747 Bytes = Bytes.slice(4); 748 Address.Address += 4; 749 } 750 } 751 }; 752 HexagonPrettyPrinter HexagonPrettyPrinterInst; 753 754 class AMDGCNPrettyPrinter : public PrettyPrinter { 755 public: 756 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 757 object::SectionedAddress Address, raw_ostream &OS, 758 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 759 std::vector<RelocationRef> *Rels) override { 760 if (SP && (PrintSource || PrintLines)) 761 SP->printSourceLine(OS, Address); 762 763 if (MI) { 764 SmallString<40> InstStr; 765 raw_svector_ostream IS(InstStr); 766 767 IP.printInst(MI, IS, "", STI); 768 769 OS << left_justify(IS.str(), 60); 770 } else { 771 // an unrecognized encoding - this is probably data so represent it 772 // using the .long directive, or .byte directive if fewer than 4 bytes 773 // remaining 774 if (Bytes.size() >= 4) { 775 OS << format("\t.long 0x%08" PRIx32 " ", 776 support::endian::read32<support::little>(Bytes.data())); 777 OS.indent(42); 778 } else { 779 OS << format("\t.byte 0x%02" PRIx8, Bytes[0]); 780 for (unsigned int i = 1; i < Bytes.size(); i++) 781 OS << format(", 0x%02" PRIx8, Bytes[i]); 782 OS.indent(55 - (6 * Bytes.size())); 783 } 784 } 785 786 OS << format("// %012" PRIX64 ":", Address.Address); 787 if (Bytes.size() >= 4) { 788 // D should be casted to uint32_t here as it is passed by format to 789 // snprintf as vararg. 790 for (uint32_t D : makeArrayRef( 791 reinterpret_cast<const support::little32_t *>(Bytes.data()), 792 Bytes.size() / 4)) 793 OS << format(" %08" PRIX32, D); 794 } else { 795 for (unsigned char B : Bytes) 796 OS << format(" %02" PRIX8, B); 797 } 798 799 if (!Annot.empty()) 800 OS << " // " << Annot; 801 } 802 }; 803 AMDGCNPrettyPrinter AMDGCNPrettyPrinterInst; 804 805 class BPFPrettyPrinter : public PrettyPrinter { 806 public: 807 void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes, 808 object::SectionedAddress Address, raw_ostream &OS, 809 StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP, 810 std::vector<RelocationRef> *Rels) override { 811 if (SP && (PrintSource || PrintLines)) 812 SP->printSourceLine(OS, Address); 813 if (!NoLeadingAddr) 814 OS << format("%8" PRId64 ":", Address.Address / 8); 815 if (!NoShowRawInsn) { 816 OS << "\t"; 817 dumpBytes(Bytes, OS); 818 } 819 if (MI) 820 IP.printInst(MI, OS, "", STI); 821 else 822 OS << "\t<unknown>"; 823 } 824 }; 825 BPFPrettyPrinter BPFPrettyPrinterInst; 826 827 PrettyPrinter &selectPrettyPrinter(Triple const &Triple) { 828 switch(Triple.getArch()) { 829 default: 830 return PrettyPrinterInst; 831 case Triple::hexagon: 832 return HexagonPrettyPrinterInst; 833 case Triple::amdgcn: 834 return AMDGCNPrettyPrinterInst; 835 case Triple::bpfel: 836 case Triple::bpfeb: 837 return BPFPrettyPrinterInst; 838 } 839 } 840 } 841 842 static uint8_t getElfSymbolType(const ObjectFile *Obj, const SymbolRef &Sym) { 843 assert(Obj->isELF()); 844 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(Obj)) 845 return Elf32LEObj->getSymbol(Sym.getRawDataRefImpl())->getType(); 846 if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(Obj)) 847 return Elf64LEObj->getSymbol(Sym.getRawDataRefImpl())->getType(); 848 if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(Obj)) 849 return Elf32BEObj->getSymbol(Sym.getRawDataRefImpl())->getType(); 850 if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(Obj)) 851 return Elf64BEObj->getSymbol(Sym.getRawDataRefImpl())->getType(); 852 llvm_unreachable("Unsupported binary format"); 853 } 854 855 template <class ELFT> static void 856 addDynamicElfSymbols(const ELFObjectFile<ELFT> *Obj, 857 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 858 for (auto Symbol : Obj->getDynamicSymbolIterators()) { 859 uint8_t SymbolType = Symbol.getELFType(); 860 if (SymbolType == ELF::STT_SECTION) 861 continue; 862 863 uint64_t Address = unwrapOrError(Symbol.getAddress(), Obj->getFileName()); 864 // ELFSymbolRef::getAddress() returns size instead of value for common 865 // symbols which is not desirable for disassembly output. Overriding. 866 if (SymbolType == ELF::STT_COMMON) 867 Address = Obj->getSymbol(Symbol.getRawDataRefImpl())->st_value; 868 869 StringRef Name = unwrapOrError(Symbol.getName(), Obj->getFileName()); 870 if (Name.empty()) 871 continue; 872 873 section_iterator SecI = 874 unwrapOrError(Symbol.getSection(), Obj->getFileName()); 875 if (SecI == Obj->section_end()) 876 continue; 877 878 AllSymbols[*SecI].emplace_back(Address, Name, SymbolType); 879 } 880 } 881 882 static void 883 addDynamicElfSymbols(const ObjectFile *Obj, 884 std::map<SectionRef, SectionSymbolsTy> &AllSymbols) { 885 assert(Obj->isELF()); 886 if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(Obj)) 887 addDynamicElfSymbols(Elf32LEObj, AllSymbols); 888 else if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(Obj)) 889 addDynamicElfSymbols(Elf64LEObj, AllSymbols); 890 else if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(Obj)) 891 addDynamicElfSymbols(Elf32BEObj, AllSymbols); 892 else if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(Obj)) 893 addDynamicElfSymbols(Elf64BEObj, AllSymbols); 894 else 895 llvm_unreachable("Unsupported binary format"); 896 } 897 898 static void addPltEntries(const ObjectFile *Obj, 899 std::map<SectionRef, SectionSymbolsTy> &AllSymbols, 900 StringSaver &Saver) { 901 Optional<SectionRef> Plt = None; 902 for (const SectionRef &Section : Obj->sections()) { 903 StringRef Name; 904 if (Section.getName(Name)) 905 continue; 906 if (Name == ".plt") 907 Plt = Section; 908 } 909 if (!Plt) 910 return; 911 if (auto *ElfObj = dyn_cast<ELFObjectFileBase>(Obj)) { 912 for (auto PltEntry : ElfObj->getPltAddresses()) { 913 SymbolRef Symbol(PltEntry.first, ElfObj); 914 uint8_t SymbolType = getElfSymbolType(Obj, Symbol); 915 916 StringRef Name = unwrapOrError(Symbol.getName(), Obj->getFileName()); 917 if (!Name.empty()) 918 AllSymbols[*Plt].emplace_back( 919 PltEntry.second, Saver.save((Name + "@plt").str()), SymbolType); 920 } 921 } 922 } 923 924 // Normally the disassembly output will skip blocks of zeroes. This function 925 // returns the number of zero bytes that can be skipped when dumping the 926 // disassembly of the instructions in Buf. 927 static size_t countSkippableZeroBytes(ArrayRef<uint8_t> Buf) { 928 // Find the number of leading zeroes. 929 size_t N = 0; 930 while (N < Buf.size() && !Buf[N]) 931 ++N; 932 933 // We may want to skip blocks of zero bytes, but unless we see 934 // at least 8 of them in a row. 935 if (N < 8) 936 return 0; 937 938 // We skip zeroes in multiples of 4 because do not want to truncate an 939 // instruction if it starts with a zero byte. 940 return N & ~0x3; 941 } 942 943 // Returns a map from sections to their relocations. 944 static std::map<SectionRef, std::vector<RelocationRef>> 945 getRelocsMap(object::ObjectFile const &Obj) { 946 std::map<SectionRef, std::vector<RelocationRef>> Ret; 947 for (SectionRef Sec : Obj.sections()) { 948 section_iterator Relocated = Sec.getRelocatedSection(); 949 if (Relocated == Obj.section_end() || !shouldKeep(*Relocated)) 950 continue; 951 std::vector<RelocationRef> &V = Ret[*Relocated]; 952 for (const RelocationRef &R : Sec.relocations()) 953 V.push_back(R); 954 // Sort relocations by address. 955 llvm::stable_sort(V, isRelocAddressLess); 956 } 957 return Ret; 958 } 959 960 // Used for --adjust-vma to check if address should be adjusted by the 961 // specified value for a given section. 962 // For ELF we do not adjust non-allocatable sections like debug ones, 963 // because they are not loadable. 964 // TODO: implement for other file formats. 965 static bool shouldAdjustVA(const SectionRef &Section) { 966 const ObjectFile *Obj = Section.getObject(); 967 if (isa<object::ELFObjectFileBase>(Obj)) 968 return ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC; 969 return false; 970 } 971 972 973 typedef std::pair<uint64_t, char> MappingSymbolPair; 974 static char getMappingSymbolKind(ArrayRef<MappingSymbolPair> MappingSymbols, 975 uint64_t Address) { 976 auto Sym = bsearch(MappingSymbols, [Address](const MappingSymbolPair &Val) { 977 return Val.first > Address; 978 }); 979 // Return zero for any address before the first mapping symbol; this means 980 // we should use the default disassembly mode, depending on the target. 981 if (Sym == MappingSymbols.begin()) 982 return '\x00'; 983 return (Sym - 1)->second; 984 } 985 986 static uint64_t 987 dumpARMELFData(uint64_t SectionAddr, uint64_t Index, uint64_t End, 988 const ObjectFile *Obj, ArrayRef<uint8_t> Bytes, 989 ArrayRef<MappingSymbolPair> MappingSymbols) { 990 support::endianness Endian = 991 Obj->isLittleEndian() ? support::little : support::big; 992 while (Index < End) { 993 outs() << format("%8" PRIx64 ":", SectionAddr + Index); 994 outs() << "\t"; 995 if (Index + 4 <= End) { 996 dumpBytes(Bytes.slice(Index, 4), outs()); 997 outs() << "\t.word\t" 998 << format_hex( 999 support::endian::read32(Bytes.data() + Index, Endian), 10); 1000 Index += 4; 1001 } else if (Index + 2 <= End) { 1002 dumpBytes(Bytes.slice(Index, 2), outs()); 1003 outs() << "\t\t.short\t" 1004 << format_hex( 1005 support::endian::read16(Bytes.data() + Index, Endian), 6); 1006 Index += 2; 1007 } else { 1008 dumpBytes(Bytes.slice(Index, 1), outs()); 1009 outs() << "\t\t.byte\t" << format_hex(Bytes[0], 4); 1010 ++Index; 1011 } 1012 outs() << "\n"; 1013 if (getMappingSymbolKind(MappingSymbols, Index) != 'd') 1014 break; 1015 } 1016 return Index; 1017 } 1018 1019 static void dumpELFData(uint64_t SectionAddr, uint64_t Index, uint64_t End, 1020 ArrayRef<uint8_t> Bytes) { 1021 // print out data up to 8 bytes at a time in hex and ascii 1022 uint8_t AsciiData[9] = {'\0'}; 1023 uint8_t Byte; 1024 int NumBytes = 0; 1025 1026 for (; Index < End; ++Index) { 1027 if (NumBytes == 0) 1028 outs() << format("%8" PRIx64 ":", SectionAddr + Index); 1029 Byte = Bytes.slice(Index)[0]; 1030 outs() << format(" %02x", Byte); 1031 AsciiData[NumBytes] = isPrint(Byte) ? Byte : '.'; 1032 1033 uint8_t IndentOffset = 0; 1034 NumBytes++; 1035 if (Index == End - 1 || NumBytes > 8) { 1036 // Indent the space for less than 8 bytes data. 1037 // 2 spaces for byte and one for space between bytes 1038 IndentOffset = 3 * (8 - NumBytes); 1039 for (int Excess = NumBytes; Excess < 8; Excess++) 1040 AsciiData[Excess] = '\0'; 1041 NumBytes = 8; 1042 } 1043 if (NumBytes == 8) { 1044 AsciiData[8] = '\0'; 1045 outs() << std::string(IndentOffset, ' ') << " "; 1046 outs() << reinterpret_cast<char *>(AsciiData); 1047 outs() << '\n'; 1048 NumBytes = 0; 1049 } 1050 } 1051 } 1052 1053 static void disassembleObject(const Target *TheTarget, const ObjectFile *Obj, 1054 MCContext &Ctx, MCDisassembler *PrimaryDisAsm, 1055 MCDisassembler *SecondaryDisAsm, 1056 const MCInstrAnalysis *MIA, MCInstPrinter *IP, 1057 const MCSubtargetInfo *PrimarySTI, 1058 const MCSubtargetInfo *SecondarySTI, 1059 PrettyPrinter &PIP, 1060 SourcePrinter &SP, bool InlineRelocs) { 1061 const MCSubtargetInfo *STI = PrimarySTI; 1062 MCDisassembler *DisAsm = PrimaryDisAsm; 1063 bool PrimaryIsThumb = false; 1064 if (isArmElf(Obj)) 1065 PrimaryIsThumb = STI->checkFeatures("+thumb-mode"); 1066 1067 std::map<SectionRef, std::vector<RelocationRef>> RelocMap; 1068 if (InlineRelocs) 1069 RelocMap = getRelocsMap(*Obj); 1070 bool Is64Bits = Obj->getBytesInAddress() > 4; 1071 1072 // Create a mapping from virtual address to symbol name. This is used to 1073 // pretty print the symbols while disassembling. 1074 std::map<SectionRef, SectionSymbolsTy> AllSymbols; 1075 SectionSymbolsTy AbsoluteSymbols; 1076 const StringRef FileName = Obj->getFileName(); 1077 for (const SymbolRef &Symbol : Obj->symbols()) { 1078 uint64_t Address = unwrapOrError(Symbol.getAddress(), FileName); 1079 1080 StringRef Name = unwrapOrError(Symbol.getName(), FileName); 1081 if (Name.empty()) 1082 continue; 1083 1084 uint8_t SymbolType = ELF::STT_NOTYPE; 1085 if (Obj->isELF()) { 1086 SymbolType = getElfSymbolType(Obj, Symbol); 1087 if (SymbolType == ELF::STT_SECTION) 1088 continue; 1089 } 1090 1091 section_iterator SecI = unwrapOrError(Symbol.getSection(), FileName); 1092 if (SecI != Obj->section_end()) 1093 AllSymbols[*SecI].emplace_back(Address, Name, SymbolType); 1094 else 1095 AbsoluteSymbols.emplace_back(Address, Name, SymbolType); 1096 } 1097 if (AllSymbols.empty() && Obj->isELF()) 1098 addDynamicElfSymbols(Obj, AllSymbols); 1099 1100 BumpPtrAllocator A; 1101 StringSaver Saver(A); 1102 addPltEntries(Obj, AllSymbols, Saver); 1103 1104 // Create a mapping from virtual address to section. 1105 std::vector<std::pair<uint64_t, SectionRef>> SectionAddresses; 1106 for (SectionRef Sec : Obj->sections()) 1107 SectionAddresses.emplace_back(Sec.getAddress(), Sec); 1108 array_pod_sort(SectionAddresses.begin(), SectionAddresses.end()); 1109 1110 // Linked executables (.exe and .dll files) typically don't include a real 1111 // symbol table but they might contain an export table. 1112 if (const auto *COFFObj = dyn_cast<COFFObjectFile>(Obj)) { 1113 for (const auto &ExportEntry : COFFObj->export_directories()) { 1114 StringRef Name; 1115 error(ExportEntry.getSymbolName(Name)); 1116 if (Name.empty()) 1117 continue; 1118 uint32_t RVA; 1119 error(ExportEntry.getExportRVA(RVA)); 1120 1121 uint64_t VA = COFFObj->getImageBase() + RVA; 1122 auto Sec = llvm::bsearch( 1123 SectionAddresses, [VA](const std::pair<uint64_t, SectionRef> &RHS) { 1124 return VA < RHS.first; 1125 }); 1126 if (Sec != SectionAddresses.begin()) { 1127 --Sec; 1128 AllSymbols[Sec->second].emplace_back(VA, Name, ELF::STT_NOTYPE); 1129 } else 1130 AbsoluteSymbols.emplace_back(VA, Name, ELF::STT_NOTYPE); 1131 } 1132 } 1133 1134 // Sort all the symbols, this allows us to use a simple binary search to find 1135 // a symbol near an address. 1136 StringSet<> FoundDisasmFuncsSet; 1137 for (std::pair<const SectionRef, SectionSymbolsTy> &SecSyms : AllSymbols) 1138 array_pod_sort(SecSyms.second.begin(), SecSyms.second.end()); 1139 array_pod_sort(AbsoluteSymbols.begin(), AbsoluteSymbols.end()); 1140 1141 for (const SectionRef &Section : ToolSectionFilter(*Obj)) { 1142 if (FilterSections.empty() && !DisassembleAll && 1143 (!Section.isText() || Section.isVirtual())) 1144 continue; 1145 1146 uint64_t SectionAddr = Section.getAddress(); 1147 uint64_t SectSize = Section.getSize(); 1148 if (!SectSize) 1149 continue; 1150 1151 // Get the list of all the symbols in this section. 1152 SectionSymbolsTy &Symbols = AllSymbols[Section]; 1153 std::vector<MappingSymbolPair> MappingSymbols; 1154 if (hasMappingSymbols(Obj)) { 1155 for (const auto &Symb : Symbols) { 1156 uint64_t Address = std::get<0>(Symb); 1157 StringRef Name = std::get<1>(Symb); 1158 if (Name.startswith("$d")) 1159 MappingSymbols.emplace_back(Address - SectionAddr, 'd'); 1160 if (Name.startswith("$x")) 1161 MappingSymbols.emplace_back(Address - SectionAddr, 'x'); 1162 if (Name.startswith("$a")) 1163 MappingSymbols.emplace_back(Address - SectionAddr, 'a'); 1164 if (Name.startswith("$t")) 1165 MappingSymbols.emplace_back(Address - SectionAddr, 't'); 1166 } 1167 } 1168 1169 llvm::sort(MappingSymbols); 1170 1171 if (Obj->isELF() && Obj->getArch() == Triple::amdgcn) { 1172 // AMDGPU disassembler uses symbolizer for printing labels 1173 std::unique_ptr<MCRelocationInfo> RelInfo( 1174 TheTarget->createMCRelocationInfo(TripleName, Ctx)); 1175 if (RelInfo) { 1176 std::unique_ptr<MCSymbolizer> Symbolizer( 1177 TheTarget->createMCSymbolizer( 1178 TripleName, nullptr, nullptr, &Symbols, &Ctx, std::move(RelInfo))); 1179 DisAsm->setSymbolizer(std::move(Symbolizer)); 1180 } 1181 } 1182 1183 StringRef SegmentName = ""; 1184 if (const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(Obj)) { 1185 DataRefImpl DR = Section.getRawDataRefImpl(); 1186 SegmentName = MachO->getSectionFinalSegmentName(DR); 1187 } 1188 StringRef SectionName; 1189 error(Section.getName(SectionName)); 1190 1191 // If the section has no symbol at the start, just insert a dummy one. 1192 if (Symbols.empty() || std::get<0>(Symbols[0]) != 0) { 1193 Symbols.insert( 1194 Symbols.begin(), 1195 std::make_tuple(SectionAddr, SectionName, 1196 Section.isText() ? ELF::STT_FUNC : ELF::STT_OBJECT)); 1197 } 1198 1199 SmallString<40> Comments; 1200 raw_svector_ostream CommentStream(Comments); 1201 1202 ArrayRef<uint8_t> Bytes = arrayRefFromStringRef( 1203 unwrapOrError(Section.getContents(), Obj->getFileName())); 1204 1205 uint64_t VMAAdjustment = 0; 1206 if (shouldAdjustVA(Section)) 1207 VMAAdjustment = AdjustVMA; 1208 1209 uint64_t Size; 1210 uint64_t Index; 1211 bool PrintedSection = false; 1212 std::vector<RelocationRef> Rels = RelocMap[Section]; 1213 std::vector<RelocationRef>::const_iterator RelCur = Rels.begin(); 1214 std::vector<RelocationRef>::const_iterator RelEnd = Rels.end(); 1215 // Disassemble symbol by symbol. 1216 for (unsigned SI = 0, SE = Symbols.size(); SI != SE; ++SI) { 1217 std::string SymbolName = std::get<1>(Symbols[SI]).str(); 1218 if (Demangle) 1219 SymbolName = demangle(SymbolName); 1220 1221 // Skip if --disassemble-functions is not empty and the symbol is not in 1222 // the list. 1223 if (!DisasmFuncsSet.empty() && !DisasmFuncsSet.count(SymbolName)) 1224 continue; 1225 1226 uint64_t Start = std::get<0>(Symbols[SI]); 1227 if (Start < SectionAddr || StopAddress <= Start) 1228 continue; 1229 else 1230 FoundDisasmFuncsSet.insert(SymbolName); 1231 1232 // The end is the section end, the beginning of the next symbol, or 1233 // --stop-address. 1234 uint64_t End = std::min<uint64_t>(SectionAddr + SectSize, StopAddress); 1235 if (SI + 1 < SE) 1236 End = std::min(End, std::get<0>(Symbols[SI + 1])); 1237 if (Start >= End || End <= StartAddress) 1238 continue; 1239 Start -= SectionAddr; 1240 End -= SectionAddr; 1241 1242 if (!PrintedSection) { 1243 PrintedSection = true; 1244 outs() << "\nDisassembly of section "; 1245 if (!SegmentName.empty()) 1246 outs() << SegmentName << ","; 1247 outs() << SectionName << ":\n"; 1248 } 1249 1250 if (Obj->isELF() && Obj->getArch() == Triple::amdgcn) { 1251 if (std::get<2>(Symbols[SI]) == ELF::STT_AMDGPU_HSA_KERNEL) { 1252 // skip amd_kernel_code_t at the begining of kernel symbol (256 bytes) 1253 Start += 256; 1254 } 1255 if (SI == SE - 1 || 1256 std::get<2>(Symbols[SI + 1]) == ELF::STT_AMDGPU_HSA_KERNEL) { 1257 // cut trailing zeroes at the end of kernel 1258 // cut up to 256 bytes 1259 const uint64_t EndAlign = 256; 1260 const auto Limit = End - (std::min)(EndAlign, End - Start); 1261 while (End > Limit && 1262 *reinterpret_cast<const support::ulittle32_t*>(&Bytes[End - 4]) == 0) 1263 End -= 4; 1264 } 1265 } 1266 1267 outs() << '\n'; 1268 if (!NoLeadingAddr) 1269 outs() << format(Is64Bits ? "%016" PRIx64 " " : "%08" PRIx64 " ", 1270 SectionAddr + Start + VMAAdjustment); 1271 1272 outs() << SymbolName << ":\n"; 1273 1274 // Don't print raw contents of a virtual section. A virtual section 1275 // doesn't have any contents in the file. 1276 if (Section.isVirtual()) { 1277 outs() << "...\n"; 1278 continue; 1279 } 1280 1281 #ifndef NDEBUG 1282 raw_ostream &DebugOut = DebugFlag ? dbgs() : nulls(); 1283 #else 1284 raw_ostream &DebugOut = nulls(); 1285 #endif 1286 1287 // Some targets (like WebAssembly) have a special prelude at the start 1288 // of each symbol. 1289 DisAsm->onSymbolStart(SymbolName, Size, Bytes.slice(Start, End - Start), 1290 SectionAddr + Start, DebugOut, CommentStream); 1291 Start += Size; 1292 1293 Index = Start; 1294 if (SectionAddr < StartAddress) 1295 Index = std::max<uint64_t>(Index, StartAddress - SectionAddr); 1296 1297 // If there is a data/common symbol inside an ELF text section and we are 1298 // only disassembling text (applicable all architectures), we are in a 1299 // situation where we must print the data and not disassemble it. 1300 if (Obj->isELF() && !DisassembleAll && Section.isText()) { 1301 uint8_t SymTy = std::get<2>(Symbols[SI]); 1302 if (SymTy == ELF::STT_OBJECT || SymTy == ELF::STT_COMMON) { 1303 dumpELFData(SectionAddr, Index, End, Bytes); 1304 Index = End; 1305 } 1306 } 1307 1308 bool CheckARMELFData = hasMappingSymbols(Obj) && 1309 std::get<2>(Symbols[SI]) != ELF::STT_OBJECT && 1310 !DisassembleAll; 1311 while (Index < End) { 1312 // ARM and AArch64 ELF binaries can interleave data and text in the 1313 // same section. We rely on the markers introduced to understand what 1314 // we need to dump. If the data marker is within a function, it is 1315 // denoted as a word/short etc. 1316 if (CheckARMELFData && 1317 getMappingSymbolKind(MappingSymbols, Index) == 'd') { 1318 Index = dumpARMELFData(SectionAddr, Index, End, Obj, Bytes, 1319 MappingSymbols); 1320 continue; 1321 } 1322 1323 // When -z or --disassemble-zeroes are given we always dissasemble 1324 // them. Otherwise we might want to skip zero bytes we see. 1325 if (!DisassembleZeroes) { 1326 uint64_t MaxOffset = End - Index; 1327 // For -reloc: print zero blocks patched by relocations, so that 1328 // relocations can be shown in the dump. 1329 if (RelCur != RelEnd) 1330 MaxOffset = RelCur->getOffset() - Index; 1331 1332 if (size_t N = 1333 countSkippableZeroBytes(Bytes.slice(Index, MaxOffset))) { 1334 outs() << "\t\t..." << '\n'; 1335 Index += N; 1336 continue; 1337 } 1338 } 1339 1340 if (SecondarySTI) { 1341 if (getMappingSymbolKind(MappingSymbols, Index) == 'a') { 1342 STI = PrimaryIsThumb ? SecondarySTI : PrimarySTI; 1343 DisAsm = PrimaryIsThumb ? SecondaryDisAsm : PrimaryDisAsm; 1344 } else if (getMappingSymbolKind(MappingSymbols, Index) == 't') { 1345 STI = PrimaryIsThumb ? PrimarySTI : SecondarySTI; 1346 DisAsm = PrimaryIsThumb ? PrimaryDisAsm : SecondaryDisAsm; 1347 } 1348 } 1349 1350 // Disassemble a real instruction or a data when disassemble all is 1351 // provided 1352 MCInst Inst; 1353 bool Disassembled = DisAsm->getInstruction( 1354 Inst, Size, Bytes.slice(Index), SectionAddr + Index, DebugOut, 1355 CommentStream); 1356 if (Size == 0) 1357 Size = 1; 1358 1359 PIP.printInst( 1360 *IP, Disassembled ? &Inst : nullptr, Bytes.slice(Index, Size), 1361 {SectionAddr + Index + VMAAdjustment, Section.getIndex()}, outs(), 1362 "", *STI, &SP, &Rels); 1363 outs() << CommentStream.str(); 1364 Comments.clear(); 1365 1366 // Try to resolve the target of a call, tail call, etc. to a specific 1367 // symbol. 1368 if (MIA && (MIA->isCall(Inst) || MIA->isUnconditionalBranch(Inst) || 1369 MIA->isConditionalBranch(Inst))) { 1370 uint64_t Target; 1371 if (MIA->evaluateBranch(Inst, SectionAddr + Index, Size, Target)) { 1372 // In a relocatable object, the target's section must reside in 1373 // the same section as the call instruction or it is accessed 1374 // through a relocation. 1375 // 1376 // In a non-relocatable object, the target may be in any section. 1377 // 1378 // N.B. We don't walk the relocations in the relocatable case yet. 1379 auto *TargetSectionSymbols = &Symbols; 1380 if (!Obj->isRelocatableObject()) { 1381 auto It = llvm::bsearch( 1382 SectionAddresses, 1383 [=](const std::pair<uint64_t, SectionRef> &RHS) { 1384 return Target < RHS.first; 1385 }); 1386 if (It != SectionAddresses.begin()) { 1387 --It; 1388 TargetSectionSymbols = &AllSymbols[It->second]; 1389 } else { 1390 TargetSectionSymbols = &AbsoluteSymbols; 1391 } 1392 } 1393 1394 // Find the last symbol in the section whose offset is less than 1395 // or equal to the target. If there isn't a section that contains 1396 // the target, find the nearest preceding absolute symbol. 1397 auto TargetSym = llvm::bsearch( 1398 *TargetSectionSymbols, 1399 [=](const std::tuple<uint64_t, StringRef, uint8_t> &RHS) { 1400 return Target < std::get<0>(RHS); 1401 }); 1402 if (TargetSym == TargetSectionSymbols->begin()) { 1403 TargetSectionSymbols = &AbsoluteSymbols; 1404 TargetSym = llvm::bsearch( 1405 AbsoluteSymbols, 1406 [=](const std::tuple<uint64_t, StringRef, uint8_t> &RHS) { 1407 return Target < std::get<0>(RHS); 1408 }); 1409 } 1410 if (TargetSym != TargetSectionSymbols->begin()) { 1411 --TargetSym; 1412 uint64_t TargetAddress = std::get<0>(*TargetSym); 1413 StringRef TargetName = std::get<1>(*TargetSym); 1414 outs() << " <" << TargetName; 1415 uint64_t Disp = Target - TargetAddress; 1416 if (Disp) 1417 outs() << "+0x" << Twine::utohexstr(Disp); 1418 outs() << '>'; 1419 } 1420 } 1421 } 1422 outs() << "\n"; 1423 1424 // Hexagon does this in pretty printer 1425 if (Obj->getArch() != Triple::hexagon) { 1426 // Print relocation for instruction. 1427 while (RelCur != RelEnd) { 1428 uint64_t Offset = RelCur->getOffset(); 1429 // If this relocation is hidden, skip it. 1430 if (getHidden(*RelCur) || SectionAddr + Offset < StartAddress) { 1431 ++RelCur; 1432 continue; 1433 } 1434 1435 // Stop when RelCur's offset is past the current instruction. 1436 if (Offset >= Index + Size) 1437 break; 1438 1439 // When --adjust-vma is used, update the address printed. 1440 if (RelCur->getSymbol() != Obj->symbol_end()) { 1441 Expected<section_iterator> SymSI = 1442 RelCur->getSymbol()->getSection(); 1443 if (SymSI && *SymSI != Obj->section_end() && 1444 shouldAdjustVA(**SymSI)) 1445 Offset += AdjustVMA; 1446 } 1447 1448 printRelocation(*RelCur, SectionAddr + Offset, Is64Bits); 1449 ++RelCur; 1450 } 1451 } 1452 1453 Index += Size; 1454 } 1455 } 1456 } 1457 StringSet<> MissingDisasmFuncsSet = 1458 set_difference(DisasmFuncsSet, FoundDisasmFuncsSet); 1459 for (StringRef MissingDisasmFunc : MissingDisasmFuncsSet.keys()) 1460 warn("failed to disassemble missing function " + MissingDisasmFunc); 1461 } 1462 1463 static void disassembleObject(const ObjectFile *Obj, bool InlineRelocs) { 1464 const Target *TheTarget = getTarget(Obj); 1465 1466 // Package up features to be passed to target/subtarget 1467 SubtargetFeatures Features = Obj->getFeatures(); 1468 if (!MAttrs.empty()) 1469 for (unsigned I = 0; I != MAttrs.size(); ++I) 1470 Features.AddFeature(MAttrs[I]); 1471 1472 std::unique_ptr<const MCRegisterInfo> MRI( 1473 TheTarget->createMCRegInfo(TripleName)); 1474 if (!MRI) 1475 report_error(Obj->getFileName(), 1476 "no register info for target " + TripleName); 1477 1478 // Set up disassembler. 1479 std::unique_ptr<const MCAsmInfo> AsmInfo( 1480 TheTarget->createMCAsmInfo(*MRI, TripleName)); 1481 if (!AsmInfo) 1482 report_error(Obj->getFileName(), 1483 "no assembly info for target " + TripleName); 1484 std::unique_ptr<const MCSubtargetInfo> STI( 1485 TheTarget->createMCSubtargetInfo(TripleName, MCPU, Features.getString())); 1486 if (!STI) 1487 report_error(Obj->getFileName(), 1488 "no subtarget info for target " + TripleName); 1489 std::unique_ptr<const MCInstrInfo> MII(TheTarget->createMCInstrInfo()); 1490 if (!MII) 1491 report_error(Obj->getFileName(), 1492 "no instruction info for target " + TripleName); 1493 MCObjectFileInfo MOFI; 1494 MCContext Ctx(AsmInfo.get(), MRI.get(), &MOFI); 1495 // FIXME: for now initialize MCObjectFileInfo with default values 1496 MOFI.InitMCObjectFileInfo(Triple(TripleName), false, Ctx); 1497 1498 std::unique_ptr<MCDisassembler> DisAsm( 1499 TheTarget->createMCDisassembler(*STI, Ctx)); 1500 if (!DisAsm) 1501 report_error(Obj->getFileName(), 1502 "no disassembler for target " + TripleName); 1503 1504 // If we have an ARM object file, we need a second disassembler, because 1505 // ARM CPUs have two different instruction sets: ARM mode, and Thumb mode. 1506 // We use mapping symbols to switch between the two assemblers, where 1507 // appropriate. 1508 std::unique_ptr<MCDisassembler> SecondaryDisAsm; 1509 std::unique_ptr<const MCSubtargetInfo> SecondarySTI; 1510 if (isArmElf(Obj) && !STI->checkFeatures("+mclass")) { 1511 if (STI->checkFeatures("+thumb-mode")) 1512 Features.AddFeature("-thumb-mode"); 1513 else 1514 Features.AddFeature("+thumb-mode"); 1515 SecondarySTI.reset(TheTarget->createMCSubtargetInfo(TripleName, MCPU, 1516 Features.getString())); 1517 SecondaryDisAsm.reset(TheTarget->createMCDisassembler(*SecondarySTI, Ctx)); 1518 } 1519 1520 std::unique_ptr<const MCInstrAnalysis> MIA( 1521 TheTarget->createMCInstrAnalysis(MII.get())); 1522 1523 int AsmPrinterVariant = AsmInfo->getAssemblerDialect(); 1524 std::unique_ptr<MCInstPrinter> IP(TheTarget->createMCInstPrinter( 1525 Triple(TripleName), AsmPrinterVariant, *AsmInfo, *MII, *MRI)); 1526 if (!IP) 1527 report_error(Obj->getFileName(), 1528 "no instruction printer for target " + TripleName); 1529 IP->setPrintImmHex(PrintImmHex); 1530 1531 PrettyPrinter &PIP = selectPrettyPrinter(Triple(TripleName)); 1532 SourcePrinter SP(Obj, TheTarget->getName()); 1533 1534 for (StringRef Opt : DisassemblerOptions) 1535 if (!IP->applyTargetSpecificCLOption(Opt)) 1536 error("Unrecognized disassembler option: " + Opt); 1537 1538 disassembleObject(TheTarget, Obj, Ctx, DisAsm.get(), SecondaryDisAsm.get(), 1539 MIA.get(), IP.get(), STI.get(), SecondarySTI.get(), PIP, 1540 SP, InlineRelocs); 1541 } 1542 1543 void printRelocations(const ObjectFile *Obj) { 1544 StringRef Fmt = Obj->getBytesInAddress() > 4 ? "%016" PRIx64 : 1545 "%08" PRIx64; 1546 // Regular objdump doesn't print relocations in non-relocatable object 1547 // files. 1548 if (!Obj->isRelocatableObject()) 1549 return; 1550 1551 // Build a mapping from relocation target to a vector of relocation 1552 // sections. Usually, there is an only one relocation section for 1553 // each relocated section. 1554 MapVector<SectionRef, std::vector<SectionRef>> SecToRelSec; 1555 for (const SectionRef &Section : ToolSectionFilter(*Obj)) { 1556 if (Section.relocation_begin() == Section.relocation_end()) 1557 continue; 1558 const SectionRef TargetSec = *Section.getRelocatedSection(); 1559 SecToRelSec[TargetSec].push_back(Section); 1560 } 1561 1562 for (std::pair<SectionRef, std::vector<SectionRef>> &P : SecToRelSec) { 1563 StringRef SecName; 1564 error(P.first.getName(SecName)); 1565 outs() << "RELOCATION RECORDS FOR [" << SecName << "]:\n"; 1566 1567 for (SectionRef Section : P.second) { 1568 for (const RelocationRef &Reloc : Section.relocations()) { 1569 uint64_t Address = Reloc.getOffset(); 1570 SmallString<32> RelocName; 1571 SmallString<32> ValueStr; 1572 if (Address < StartAddress || Address > StopAddress || getHidden(Reloc)) 1573 continue; 1574 Reloc.getTypeName(RelocName); 1575 error(getRelocationValueString(Reloc, ValueStr)); 1576 outs() << format(Fmt.data(), Address) << " " << RelocName << " " 1577 << ValueStr << "\n"; 1578 } 1579 } 1580 outs() << "\n"; 1581 } 1582 } 1583 1584 void printDynamicRelocations(const ObjectFile *Obj) { 1585 // For the moment, this option is for ELF only 1586 if (!Obj->isELF()) 1587 return; 1588 1589 const auto *Elf = dyn_cast<ELFObjectFileBase>(Obj); 1590 if (!Elf || Elf->getEType() != ELF::ET_DYN) { 1591 error("not a dynamic object"); 1592 return; 1593 } 1594 1595 std::vector<SectionRef> DynRelSec = Obj->dynamic_relocation_sections(); 1596 if (DynRelSec.empty()) 1597 return; 1598 1599 outs() << "DYNAMIC RELOCATION RECORDS\n"; 1600 StringRef Fmt = Obj->getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 1601 for (const SectionRef &Section : DynRelSec) 1602 for (const RelocationRef &Reloc : Section.relocations()) { 1603 uint64_t Address = Reloc.getOffset(); 1604 SmallString<32> RelocName; 1605 SmallString<32> ValueStr; 1606 Reloc.getTypeName(RelocName); 1607 error(getRelocationValueString(Reloc, ValueStr)); 1608 outs() << format(Fmt.data(), Address) << " " << RelocName << " " 1609 << ValueStr << "\n"; 1610 } 1611 } 1612 1613 // Returns true if we need to show LMA column when dumping section headers. We 1614 // show it only when the platform is ELF and either we have at least one section 1615 // whose VMA and LMA are different and/or when --show-lma flag is used. 1616 static bool shouldDisplayLMA(const ObjectFile *Obj) { 1617 if (!Obj->isELF()) 1618 return false; 1619 for (const SectionRef &S : ToolSectionFilter(*Obj)) 1620 if (S.getAddress() != getELFSectionLMA(S)) 1621 return true; 1622 return ShowLMA; 1623 } 1624 1625 void printSectionHeaders(const ObjectFile *Obj) { 1626 bool HasLMAColumn = shouldDisplayLMA(Obj); 1627 if (HasLMAColumn) 1628 outs() << "Sections:\n" 1629 "Idx Name Size VMA LMA " 1630 "Type\n"; 1631 else 1632 outs() << "Sections:\n" 1633 "Idx Name Size VMA Type\n"; 1634 1635 for (const SectionRef &Section : ToolSectionFilter(*Obj)) { 1636 StringRef Name; 1637 error(Section.getName(Name)); 1638 uint64_t VMA = Section.getAddress(); 1639 if (shouldAdjustVA(Section)) 1640 VMA += AdjustVMA; 1641 1642 uint64_t Size = Section.getSize(); 1643 bool Text = Section.isText(); 1644 bool Data = Section.isData(); 1645 bool BSS = Section.isBSS(); 1646 std::string Type = (std::string(Text ? "TEXT " : "") + 1647 (Data ? "DATA " : "") + (BSS ? "BSS" : "")); 1648 1649 if (HasLMAColumn) 1650 outs() << format("%3d %-13s %08" PRIx64 " %016" PRIx64 " %016" PRIx64 1651 " %s\n", 1652 (unsigned)Section.getIndex(), Name.str().c_str(), Size, 1653 VMA, getELFSectionLMA(Section), Type.c_str()); 1654 else 1655 outs() << format("%3d %-13s %08" PRIx64 " %016" PRIx64 " %s\n", 1656 (unsigned)Section.getIndex(), Name.str().c_str(), Size, 1657 VMA, Type.c_str()); 1658 } 1659 outs() << "\n"; 1660 } 1661 1662 void printSectionContents(const ObjectFile *Obj) { 1663 for (const SectionRef &Section : ToolSectionFilter(*Obj)) { 1664 StringRef Name; 1665 error(Section.getName(Name)); 1666 uint64_t BaseAddr = Section.getAddress(); 1667 uint64_t Size = Section.getSize(); 1668 if (!Size) 1669 continue; 1670 1671 outs() << "Contents of section " << Name << ":\n"; 1672 if (Section.isBSS()) { 1673 outs() << format("<skipping contents of bss section at [%04" PRIx64 1674 ", %04" PRIx64 ")>\n", 1675 BaseAddr, BaseAddr + Size); 1676 continue; 1677 } 1678 1679 StringRef Contents = unwrapOrError(Section.getContents(), Obj->getFileName()); 1680 1681 // Dump out the content as hex and printable ascii characters. 1682 for (std::size_t Addr = 0, End = Contents.size(); Addr < End; Addr += 16) { 1683 outs() << format(" %04" PRIx64 " ", BaseAddr + Addr); 1684 // Dump line of hex. 1685 for (std::size_t I = 0; I < 16; ++I) { 1686 if (I != 0 && I % 4 == 0) 1687 outs() << ' '; 1688 if (Addr + I < End) 1689 outs() << hexdigit((Contents[Addr + I] >> 4) & 0xF, true) 1690 << hexdigit(Contents[Addr + I] & 0xF, true); 1691 else 1692 outs() << " "; 1693 } 1694 // Print ascii. 1695 outs() << " "; 1696 for (std::size_t I = 0; I < 16 && Addr + I < End; ++I) { 1697 if (isPrint(static_cast<unsigned char>(Contents[Addr + I]) & 0xFF)) 1698 outs() << Contents[Addr + I]; 1699 else 1700 outs() << "."; 1701 } 1702 outs() << "\n"; 1703 } 1704 } 1705 } 1706 1707 void printSymbolTable(const ObjectFile *O, StringRef ArchiveName, 1708 StringRef ArchitectureName) { 1709 outs() << "SYMBOL TABLE:\n"; 1710 1711 if (const COFFObjectFile *Coff = dyn_cast<const COFFObjectFile>(O)) { 1712 printCOFFSymbolTable(Coff); 1713 return; 1714 } 1715 1716 const StringRef FileName = O->getFileName(); 1717 for (auto I = O->symbol_begin(), E = O->symbol_end(); I != E; ++I) { 1718 const SymbolRef &Symbol = *I; 1719 uint64_t Address = unwrapOrError(Symbol.getAddress(), ArchiveName, FileName, 1720 ArchitectureName); 1721 if ((Address < StartAddress) || (Address > StopAddress)) 1722 continue; 1723 SymbolRef::Type Type = unwrapOrError(Symbol.getType(), ArchiveName, 1724 FileName, ArchitectureName); 1725 uint32_t Flags = Symbol.getFlags(); 1726 section_iterator Section = unwrapOrError(Symbol.getSection(), ArchiveName, 1727 FileName, ArchitectureName); 1728 StringRef Name; 1729 if (Type == SymbolRef::ST_Debug && Section != O->section_end()) 1730 Section->getName(Name); 1731 else 1732 Name = unwrapOrError(Symbol.getName(), ArchiveName, FileName, 1733 ArchitectureName); 1734 1735 bool Global = Flags & SymbolRef::SF_Global; 1736 bool Weak = Flags & SymbolRef::SF_Weak; 1737 bool Absolute = Flags & SymbolRef::SF_Absolute; 1738 bool Common = Flags & SymbolRef::SF_Common; 1739 bool Hidden = Flags & SymbolRef::SF_Hidden; 1740 1741 char GlobLoc = ' '; 1742 if (Type != SymbolRef::ST_Unknown) 1743 GlobLoc = Global ? 'g' : 'l'; 1744 char Debug = (Type == SymbolRef::ST_Debug || Type == SymbolRef::ST_File) 1745 ? 'd' : ' '; 1746 char FileFunc = ' '; 1747 if (Type == SymbolRef::ST_File) 1748 FileFunc = 'f'; 1749 else if (Type == SymbolRef::ST_Function) 1750 FileFunc = 'F'; 1751 else if (Type == SymbolRef::ST_Data) 1752 FileFunc = 'O'; 1753 1754 const char *Fmt = O->getBytesInAddress() > 4 ? "%016" PRIx64 : 1755 "%08" PRIx64; 1756 1757 outs() << format(Fmt, Address) << " " 1758 << GlobLoc // Local -> 'l', Global -> 'g', Neither -> ' ' 1759 << (Weak ? 'w' : ' ') // Weak? 1760 << ' ' // Constructor. Not supported yet. 1761 << ' ' // Warning. Not supported yet. 1762 << ' ' // Indirect reference to another symbol. 1763 << Debug // Debugging (d) or dynamic (D) symbol. 1764 << FileFunc // Name of function (F), file (f) or object (O). 1765 << ' '; 1766 if (Absolute) { 1767 outs() << "*ABS*"; 1768 } else if (Common) { 1769 outs() << "*COM*"; 1770 } else if (Section == O->section_end()) { 1771 outs() << "*UND*"; 1772 } else { 1773 if (const MachOObjectFile *MachO = 1774 dyn_cast<const MachOObjectFile>(O)) { 1775 DataRefImpl DR = Section->getRawDataRefImpl(); 1776 StringRef SegmentName = MachO->getSectionFinalSegmentName(DR); 1777 outs() << SegmentName << ","; 1778 } 1779 StringRef SectionName; 1780 error(Section->getName(SectionName)); 1781 outs() << SectionName; 1782 } 1783 1784 if (Common || isa<ELFObjectFileBase>(O)) { 1785 uint64_t Val = 1786 Common ? Symbol.getAlignment() : ELFSymbolRef(Symbol).getSize(); 1787 outs() << format("\t%08" PRIx64, Val); 1788 } 1789 1790 if (isa<ELFObjectFileBase>(O)) { 1791 uint8_t Other = ELFSymbolRef(Symbol).getOther(); 1792 switch (Other) { 1793 case ELF::STV_DEFAULT: 1794 break; 1795 case ELF::STV_INTERNAL: 1796 outs() << " .internal"; 1797 break; 1798 case ELF::STV_HIDDEN: 1799 outs() << " .hidden"; 1800 break; 1801 case ELF::STV_PROTECTED: 1802 outs() << " .protected"; 1803 break; 1804 default: 1805 outs() << format(" 0x%02x", Other); 1806 break; 1807 } 1808 } else if (Hidden) { 1809 outs() << " .hidden"; 1810 } 1811 1812 if (Demangle) 1813 outs() << ' ' << demangle(Name) << '\n'; 1814 else 1815 outs() << ' ' << Name << '\n'; 1816 } 1817 } 1818 1819 static void printUnwindInfo(const ObjectFile *O) { 1820 outs() << "Unwind info:\n\n"; 1821 1822 if (const COFFObjectFile *Coff = dyn_cast<COFFObjectFile>(O)) 1823 printCOFFUnwindInfo(Coff); 1824 else if (const MachOObjectFile *MachO = dyn_cast<MachOObjectFile>(O)) 1825 printMachOUnwindInfo(MachO); 1826 else 1827 // TODO: Extract DWARF dump tool to objdump. 1828 WithColor::error(errs(), ToolName) 1829 << "This operation is only currently supported " 1830 "for COFF and MachO object files.\n"; 1831 } 1832 1833 /// Dump the raw contents of the __clangast section so the output can be piped 1834 /// into llvm-bcanalyzer. 1835 void printRawClangAST(const ObjectFile *Obj) { 1836 if (outs().is_displayed()) { 1837 WithColor::error(errs(), ToolName) 1838 << "The -raw-clang-ast option will dump the raw binary contents of " 1839 "the clang ast section.\n" 1840 "Please redirect the output to a file or another program such as " 1841 "llvm-bcanalyzer.\n"; 1842 return; 1843 } 1844 1845 StringRef ClangASTSectionName("__clangast"); 1846 if (isa<COFFObjectFile>(Obj)) { 1847 ClangASTSectionName = "clangast"; 1848 } 1849 1850 Optional<object::SectionRef> ClangASTSection; 1851 for (auto Sec : ToolSectionFilter(*Obj)) { 1852 StringRef Name; 1853 Sec.getName(Name); 1854 if (Name == ClangASTSectionName) { 1855 ClangASTSection = Sec; 1856 break; 1857 } 1858 } 1859 if (!ClangASTSection) 1860 return; 1861 1862 StringRef ClangASTContents = unwrapOrError( 1863 ClangASTSection.getValue().getContents(), Obj->getFileName()); 1864 outs().write(ClangASTContents.data(), ClangASTContents.size()); 1865 } 1866 1867 static void printFaultMaps(const ObjectFile *Obj) { 1868 StringRef FaultMapSectionName; 1869 1870 if (isa<ELFObjectFileBase>(Obj)) { 1871 FaultMapSectionName = ".llvm_faultmaps"; 1872 } else if (isa<MachOObjectFile>(Obj)) { 1873 FaultMapSectionName = "__llvm_faultmaps"; 1874 } else { 1875 WithColor::error(errs(), ToolName) 1876 << "This operation is only currently supported " 1877 "for ELF and Mach-O executable files.\n"; 1878 return; 1879 } 1880 1881 Optional<object::SectionRef> FaultMapSection; 1882 1883 for (auto Sec : ToolSectionFilter(*Obj)) { 1884 StringRef Name; 1885 Sec.getName(Name); 1886 if (Name == FaultMapSectionName) { 1887 FaultMapSection = Sec; 1888 break; 1889 } 1890 } 1891 1892 outs() << "FaultMap table:\n"; 1893 1894 if (!FaultMapSection.hasValue()) { 1895 outs() << "<not found>\n"; 1896 return; 1897 } 1898 1899 StringRef FaultMapContents = 1900 unwrapOrError(FaultMapSection.getValue().getContents(), Obj->getFileName()); 1901 FaultMapParser FMP(FaultMapContents.bytes_begin(), 1902 FaultMapContents.bytes_end()); 1903 1904 outs() << FMP; 1905 } 1906 1907 static void printPrivateFileHeaders(const ObjectFile *O, bool OnlyFirst) { 1908 if (O->isELF()) { 1909 printELFFileHeader(O); 1910 printELFDynamicSection(O); 1911 printELFSymbolVersionInfo(O); 1912 return; 1913 } 1914 if (O->isCOFF()) 1915 return printCOFFFileHeader(O); 1916 if (O->isWasm()) 1917 return printWasmFileHeader(O); 1918 if (O->isMachO()) { 1919 printMachOFileHeader(O); 1920 if (!OnlyFirst) 1921 printMachOLoadCommands(O); 1922 return; 1923 } 1924 report_error(O->getFileName(), "Invalid/Unsupported object file format"); 1925 } 1926 1927 static void printFileHeaders(const ObjectFile *O) { 1928 if (!O->isELF() && !O->isCOFF()) 1929 report_error(O->getFileName(), "Invalid/Unsupported object file format"); 1930 1931 Triple::ArchType AT = O->getArch(); 1932 outs() << "architecture: " << Triple::getArchTypeName(AT) << "\n"; 1933 uint64_t Address = unwrapOrError(O->getStartAddress(), O->getFileName()); 1934 1935 StringRef Fmt = O->getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64; 1936 outs() << "start address: " 1937 << "0x" << format(Fmt.data(), Address) << "\n\n"; 1938 } 1939 1940 static void printArchiveChild(StringRef Filename, const Archive::Child &C) { 1941 Expected<sys::fs::perms> ModeOrErr = C.getAccessMode(); 1942 if (!ModeOrErr) { 1943 WithColor::error(errs(), ToolName) << "ill-formed archive entry.\n"; 1944 consumeError(ModeOrErr.takeError()); 1945 return; 1946 } 1947 sys::fs::perms Mode = ModeOrErr.get(); 1948 outs() << ((Mode & sys::fs::owner_read) ? "r" : "-"); 1949 outs() << ((Mode & sys::fs::owner_write) ? "w" : "-"); 1950 outs() << ((Mode & sys::fs::owner_exe) ? "x" : "-"); 1951 outs() << ((Mode & sys::fs::group_read) ? "r" : "-"); 1952 outs() << ((Mode & sys::fs::group_write) ? "w" : "-"); 1953 outs() << ((Mode & sys::fs::group_exe) ? "x" : "-"); 1954 outs() << ((Mode & sys::fs::others_read) ? "r" : "-"); 1955 outs() << ((Mode & sys::fs::others_write) ? "w" : "-"); 1956 outs() << ((Mode & sys::fs::others_exe) ? "x" : "-"); 1957 1958 outs() << " "; 1959 1960 outs() << format("%d/%d %6" PRId64 " ", unwrapOrError(C.getUID(), Filename), 1961 unwrapOrError(C.getGID(), Filename), 1962 unwrapOrError(C.getRawSize(), Filename)); 1963 1964 StringRef RawLastModified = C.getRawLastModified(); 1965 unsigned Seconds; 1966 if (RawLastModified.getAsInteger(10, Seconds)) 1967 outs() << "(date: \"" << RawLastModified 1968 << "\" contains non-decimal chars) "; 1969 else { 1970 // Since ctime(3) returns a 26 character string of the form: 1971 // "Sun Sep 16 01:03:52 1973\n\0" 1972 // just print 24 characters. 1973 time_t t = Seconds; 1974 outs() << format("%.24s ", ctime(&t)); 1975 } 1976 1977 StringRef Name = ""; 1978 Expected<StringRef> NameOrErr = C.getName(); 1979 if (!NameOrErr) { 1980 consumeError(NameOrErr.takeError()); 1981 Name = unwrapOrError(C.getRawName(), Filename); 1982 } else { 1983 Name = NameOrErr.get(); 1984 } 1985 outs() << Name << "\n"; 1986 } 1987 1988 static void dumpObject(ObjectFile *O, const Archive *A = nullptr, 1989 const Archive::Child *C = nullptr) { 1990 // Avoid other output when using a raw option. 1991 if (!RawClangAST) { 1992 outs() << '\n'; 1993 if (A) 1994 outs() << A->getFileName() << "(" << O->getFileName() << ")"; 1995 else 1996 outs() << O->getFileName(); 1997 outs() << ":\tfile format " << O->getFileFormatName() << "\n\n"; 1998 } 1999 2000 StringRef ArchiveName = A ? A->getFileName() : ""; 2001 if (FileHeaders) 2002 printFileHeaders(O); 2003 if (ArchiveHeaders && !MachOOpt && C) 2004 printArchiveChild(ArchiveName, *C); 2005 if (Disassemble) 2006 disassembleObject(O, Relocations); 2007 if (Relocations && !Disassemble) 2008 printRelocations(O); 2009 if (DynamicRelocations) 2010 printDynamicRelocations(O); 2011 if (SectionHeaders) 2012 printSectionHeaders(O); 2013 if (SectionContents) 2014 printSectionContents(O); 2015 if (SymbolTable) 2016 printSymbolTable(O, ArchiveName); 2017 if (UnwindInfo) 2018 printUnwindInfo(O); 2019 if (PrivateHeaders || FirstPrivateHeader) 2020 printPrivateFileHeaders(O, FirstPrivateHeader); 2021 if (ExportsTrie) 2022 printExportsTrie(O); 2023 if (Rebase) 2024 printRebaseTable(O); 2025 if (Bind) 2026 printBindTable(O); 2027 if (LazyBind) 2028 printLazyBindTable(O); 2029 if (WeakBind) 2030 printWeakBindTable(O); 2031 if (RawClangAST) 2032 printRawClangAST(O); 2033 if (FaultMapSection) 2034 printFaultMaps(O); 2035 if (DwarfDumpType != DIDT_Null) { 2036 std::unique_ptr<DIContext> DICtx = DWARFContext::create(*O); 2037 // Dump the complete DWARF structure. 2038 DIDumpOptions DumpOpts; 2039 DumpOpts.DumpType = DwarfDumpType; 2040 DICtx->dump(outs(), DumpOpts); 2041 } 2042 } 2043 2044 static void dumpObject(const COFFImportFile *I, const Archive *A, 2045 const Archive::Child *C = nullptr) { 2046 StringRef ArchiveName = A ? A->getFileName() : ""; 2047 2048 // Avoid other output when using a raw option. 2049 if (!RawClangAST) 2050 outs() << '\n' 2051 << ArchiveName << "(" << I->getFileName() << ")" 2052 << ":\tfile format COFF-import-file" 2053 << "\n\n"; 2054 2055 if (ArchiveHeaders && !MachOOpt && C) 2056 printArchiveChild(ArchiveName, *C); 2057 if (SymbolTable) 2058 printCOFFSymbolTable(I); 2059 } 2060 2061 /// Dump each object file in \a a; 2062 static void dumpArchive(const Archive *A) { 2063 Error Err = Error::success(); 2064 for (auto &C : A->children(Err)) { 2065 Expected<std::unique_ptr<Binary>> ChildOrErr = C.getAsBinary(); 2066 if (!ChildOrErr) { 2067 if (auto E = isNotObjectErrorInvalidFileType(ChildOrErr.takeError())) 2068 report_error(std::move(E), A->getFileName(), C); 2069 continue; 2070 } 2071 if (ObjectFile *O = dyn_cast<ObjectFile>(&*ChildOrErr.get())) 2072 dumpObject(O, A, &C); 2073 else if (COFFImportFile *I = dyn_cast<COFFImportFile>(&*ChildOrErr.get())) 2074 dumpObject(I, A, &C); 2075 else 2076 report_error(errorCodeToError(object_error::invalid_file_type), 2077 A->getFileName()); 2078 } 2079 if (Err) 2080 report_error(std::move(Err), A->getFileName()); 2081 } 2082 2083 /// Open file and figure out how to dump it. 2084 static void dumpInput(StringRef file) { 2085 // If we are using the Mach-O specific object file parser, then let it parse 2086 // the file and process the command line options. So the -arch flags can 2087 // be used to select specific slices, etc. 2088 if (MachOOpt) { 2089 parseInputMachO(file); 2090 return; 2091 } 2092 2093 // Attempt to open the binary. 2094 OwningBinary<Binary> OBinary = unwrapOrError(createBinary(file), file); 2095 Binary &Binary = *OBinary.getBinary(); 2096 2097 if (Archive *A = dyn_cast<Archive>(&Binary)) 2098 dumpArchive(A); 2099 else if (ObjectFile *O = dyn_cast<ObjectFile>(&Binary)) 2100 dumpObject(O); 2101 else if (MachOUniversalBinary *UB = dyn_cast<MachOUniversalBinary>(&Binary)) 2102 parseInputMachO(UB); 2103 else 2104 report_error(errorCodeToError(object_error::invalid_file_type), file); 2105 } 2106 } // namespace llvm 2107 2108 int main(int argc, char **argv) { 2109 using namespace llvm; 2110 InitLLVM X(argc, argv); 2111 const cl::OptionCategory *OptionFilters[] = {&ObjdumpCat, &MachOCat}; 2112 cl::HideUnrelatedOptions(OptionFilters); 2113 2114 // Initialize targets and assembly printers/parsers. 2115 InitializeAllTargetInfos(); 2116 InitializeAllTargetMCs(); 2117 InitializeAllDisassemblers(); 2118 2119 // Register the target printer for --version. 2120 cl::AddExtraVersionPrinter(TargetRegistry::printRegisteredTargetsForVersion); 2121 2122 cl::ParseCommandLineOptions(argc, argv, "llvm object file dumper\n"); 2123 2124 if (StartAddress >= StopAddress) 2125 error("start address should be less than stop address"); 2126 2127 ToolName = argv[0]; 2128 2129 // Defaults to a.out if no filenames specified. 2130 if (InputFilenames.empty()) 2131 InputFilenames.push_back("a.out"); 2132 2133 if (AllHeaders) 2134 ArchiveHeaders = FileHeaders = PrivateHeaders = Relocations = 2135 SectionHeaders = SymbolTable = true; 2136 2137 if (DisassembleAll || PrintSource || PrintLines || 2138 (!DisassembleFunctions.empty())) 2139 Disassemble = true; 2140 2141 if (!ArchiveHeaders && !Disassemble && DwarfDumpType == DIDT_Null && 2142 !DynamicRelocations && !FileHeaders && !PrivateHeaders && !RawClangAST && 2143 !Relocations && !SectionHeaders && !SectionContents && !SymbolTable && 2144 !UnwindInfo && !FaultMapSection && 2145 !(MachOOpt && 2146 (Bind || DataInCode || DylibId || DylibsUsed || ExportsTrie || 2147 FirstPrivateHeader || IndirectSymbols || InfoPlist || LazyBind || 2148 LinkOptHints || ObjcMetaData || Rebase || UniversalHeaders || 2149 WeakBind || !FilterSections.empty()))) { 2150 cl::PrintHelpMessage(); 2151 return 2; 2152 } 2153 2154 DisasmFuncsSet.insert(DisassembleFunctions.begin(), 2155 DisassembleFunctions.end()); 2156 2157 llvm::for_each(InputFilenames, dumpInput); 2158 2159 return EXIT_SUCCESS; 2160 } 2161