xref: /freebsd/contrib/llvm-project/llvm/tools/llvm-objdump/llvm-objdump.cpp (revision e64bea71c21eb42e97aa615188ba91f6cce0d36d)
1 //===-- llvm-objdump.cpp - Object file dumping utility for llvm -----------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This program is a utility that works like binutils "objdump", that is, it
10 // dumps out a plethora of information about an object file depending on the
11 // flags.
12 //
13 // The flags and output of this program should be near identical to those of
14 // binutils objdump.
15 //
16 //===----------------------------------------------------------------------===//
17 
18 #include "llvm-objdump.h"
19 #include "COFFDump.h"
20 #include "ELFDump.h"
21 #include "MachODump.h"
22 #include "ObjdumpOptID.h"
23 #include "OffloadDump.h"
24 #include "SourcePrinter.h"
25 #include "WasmDump.h"
26 #include "XCOFFDump.h"
27 #include "llvm/ADT/STLExtras.h"
28 #include "llvm/ADT/SetOperations.h"
29 #include "llvm/ADT/StringExtras.h"
30 #include "llvm/ADT/Twine.h"
31 #include "llvm/BinaryFormat/Wasm.h"
32 #include "llvm/DebugInfo/BTF/BTFParser.h"
33 #include "llvm/DebugInfo/DWARF/DWARFContext.h"
34 #include "llvm/DebugInfo/Symbolize/Symbolize.h"
35 #include "llvm/Debuginfod/BuildIDFetcher.h"
36 #include "llvm/Debuginfod/Debuginfod.h"
37 #include "llvm/Debuginfod/HTTPClient.h"
38 #include "llvm/Demangle/Demangle.h"
39 #include "llvm/MC/MCAsmInfo.h"
40 #include "llvm/MC/MCContext.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/MCTargetOptions.h"
49 #include "llvm/MC/TargetRegistry.h"
50 #include "llvm/Object/BuildID.h"
51 #include "llvm/Object/COFF.h"
52 #include "llvm/Object/COFFImportFile.h"
53 #include "llvm/Object/ELFObjectFile.h"
54 #include "llvm/Object/ELFTypes.h"
55 #include "llvm/Object/FaultMapParser.h"
56 #include "llvm/Object/MachO.h"
57 #include "llvm/Object/MachOUniversal.h"
58 #include "llvm/Object/OffloadBinary.h"
59 #include "llvm/Object/Wasm.h"
60 #include "llvm/Option/Arg.h"
61 #include "llvm/Option/ArgList.h"
62 #include "llvm/Option/Option.h"
63 #include "llvm/Support/Casting.h"
64 #include "llvm/Support/Debug.h"
65 #include "llvm/Support/Errc.h"
66 #include "llvm/Support/FileSystem.h"
67 #include "llvm/Support/Format.h"
68 #include "llvm/Support/LLVMDriver.h"
69 #include "llvm/Support/MemoryBuffer.h"
70 #include "llvm/Support/SourceMgr.h"
71 #include "llvm/Support/StringSaver.h"
72 #include "llvm/Support/TargetSelect.h"
73 #include "llvm/Support/WithColor.h"
74 #include "llvm/Support/raw_ostream.h"
75 #include "llvm/TargetParser/Host.h"
76 #include "llvm/TargetParser/Triple.h"
77 #include <algorithm>
78 #include <cctype>
79 #include <cstring>
80 #include <optional>
81 #include <set>
82 #include <system_error>
83 #include <unordered_map>
84 #include <utility>
85 
86 using namespace llvm;
87 using namespace llvm::object;
88 using namespace llvm::objdump;
89 using namespace llvm::opt;
90 
91 namespace {
92 
93 class CommonOptTable : public opt::GenericOptTable {
94 public:
CommonOptTable(const StringTable & StrTable,ArrayRef<StringTable::Offset> PrefixesTable,ArrayRef<Info> OptionInfos,const char * Usage,const char * Description)95   CommonOptTable(const StringTable &StrTable,
96                  ArrayRef<StringTable::Offset> PrefixesTable,
97                  ArrayRef<Info> OptionInfos, const char *Usage,
98                  const char *Description)
99       : opt::GenericOptTable(StrTable, PrefixesTable, OptionInfos),
100         Usage(Usage), Description(Description) {
101     setGroupedShortOptions(true);
102   }
103 
printHelp(StringRef Argv0,bool ShowHidden=false) const104   void printHelp(StringRef Argv0, bool ShowHidden = false) const {
105     Argv0 = sys::path::filename(Argv0);
106     opt::GenericOptTable::printHelp(outs(), (Argv0 + Usage).str().c_str(),
107                                     Description, ShowHidden, ShowHidden);
108     // TODO Replace this with OptTable API once it adds extrahelp support.
109     outs() << "\nPass @FILE as argument to read options from FILE.\n";
110   }
111 
112 private:
113   const char *Usage;
114   const char *Description;
115 };
116 
117 // ObjdumpOptID is in ObjdumpOptID.h
118 namespace objdump_opt {
119 #define OPTTABLE_STR_TABLE_CODE
120 #include "ObjdumpOpts.inc"
121 #undef OPTTABLE_STR_TABLE_CODE
122 
123 #define OPTTABLE_PREFIXES_TABLE_CODE
124 #include "ObjdumpOpts.inc"
125 #undef OPTTABLE_PREFIXES_TABLE_CODE
126 
127 static constexpr opt::OptTable::Info ObjdumpInfoTable[] = {
128 #define OPTION(...)                                                            \
129   LLVM_CONSTRUCT_OPT_INFO_WITH_ID_PREFIX(OBJDUMP_, __VA_ARGS__),
130 #include "ObjdumpOpts.inc"
131 #undef OPTION
132 };
133 } // namespace objdump_opt
134 
135 class ObjdumpOptTable : public CommonOptTable {
136 public:
ObjdumpOptTable()137   ObjdumpOptTable()
138       : CommonOptTable(
139             objdump_opt::OptionStrTable, objdump_opt::OptionPrefixesTable,
140             objdump_opt::ObjdumpInfoTable, " [options] <input object files>",
141             "llvm object file dumper") {}
142 };
143 
144 enum OtoolOptID {
145   OTOOL_INVALID = 0, // This is not an option ID.
146 #define OPTION(...) LLVM_MAKE_OPT_ID_WITH_ID_PREFIX(OTOOL_, __VA_ARGS__),
147 #include "OtoolOpts.inc"
148 #undef OPTION
149 };
150 
151 namespace otool {
152 #define OPTTABLE_STR_TABLE_CODE
153 #include "OtoolOpts.inc"
154 #undef OPTTABLE_STR_TABLE_CODE
155 
156 #define OPTTABLE_PREFIXES_TABLE_CODE
157 #include "OtoolOpts.inc"
158 #undef OPTTABLE_PREFIXES_TABLE_CODE
159 
160 static constexpr opt::OptTable::Info OtoolInfoTable[] = {
161 #define OPTION(...) LLVM_CONSTRUCT_OPT_INFO_WITH_ID_PREFIX(OTOOL_, __VA_ARGS__),
162 #include "OtoolOpts.inc"
163 #undef OPTION
164 };
165 } // namespace otool
166 
167 class OtoolOptTable : public CommonOptTable {
168 public:
OtoolOptTable()169   OtoolOptTable()
170       : CommonOptTable(otool::OptionStrTable, otool::OptionPrefixesTable,
171                        otool::OtoolInfoTable, " [option...] [file...]",
172                        "Mach-O object file displaying tool") {}
173 };
174 
175 struct BBAddrMapLabel {
176   std::string BlockLabel;
177   std::string PGOAnalysis;
178 };
179 
180 // This class represents the BBAddrMap and PGOMap associated with a single
181 // function.
182 class BBAddrMapFunctionEntry {
183 public:
BBAddrMapFunctionEntry(BBAddrMap AddrMap,PGOAnalysisMap PGOMap)184   BBAddrMapFunctionEntry(BBAddrMap AddrMap, PGOAnalysisMap PGOMap)
185       : AddrMap(std::move(AddrMap)), PGOMap(std::move(PGOMap)) {}
186 
getAddrMap() const187   const BBAddrMap &getAddrMap() const { return AddrMap; }
188 
189   // Returns the PGO string associated with the entry of index `PGOBBEntryIndex`
190   // in `PGOMap`. If PrettyPGOAnalysis is true, prints BFI as relative frequency
191   // and BPI as percentage. Otherwise raw values are displayed.
constructPGOLabelString(size_t PGOBBEntryIndex,bool PrettyPGOAnalysis) const192   std::string constructPGOLabelString(size_t PGOBBEntryIndex,
193                                       bool PrettyPGOAnalysis) const {
194     if (!PGOMap.FeatEnable.hasPGOAnalysis())
195       return "";
196     std::string PGOString;
197     raw_string_ostream PGOSS(PGOString);
198 
199     PGOSS << " (";
200     if (PGOMap.FeatEnable.FuncEntryCount && PGOBBEntryIndex == 0) {
201       PGOSS << "Entry count: " << Twine(PGOMap.FuncEntryCount);
202       if (PGOMap.FeatEnable.hasPGOAnalysisBBData()) {
203         PGOSS << ", ";
204       }
205     }
206 
207     if (PGOMap.FeatEnable.hasPGOAnalysisBBData()) {
208 
209       assert(PGOBBEntryIndex < PGOMap.BBEntries.size() &&
210              "Expected PGOAnalysisMap and BBAddrMap to have the same entries");
211       const PGOAnalysisMap::PGOBBEntry &PGOBBEntry =
212           PGOMap.BBEntries[PGOBBEntryIndex];
213 
214       if (PGOMap.FeatEnable.BBFreq) {
215         PGOSS << "Frequency: ";
216         if (PrettyPGOAnalysis)
217           printRelativeBlockFreq(PGOSS, PGOMap.BBEntries.front().BlockFreq,
218                                  PGOBBEntry.BlockFreq);
219         else
220           PGOSS << Twine(PGOBBEntry.BlockFreq.getFrequency());
221         if (PGOMap.FeatEnable.BrProb && PGOBBEntry.Successors.size() > 0) {
222           PGOSS << ", ";
223         }
224       }
225       if (PGOMap.FeatEnable.BrProb && PGOBBEntry.Successors.size() > 0) {
226         PGOSS << "Successors: ";
227         interleaveComma(
228             PGOBBEntry.Successors, PGOSS,
229             [&](const PGOAnalysisMap::PGOBBEntry::SuccessorEntry &SE) {
230               PGOSS << "BB" << SE.ID << ":";
231               if (PrettyPGOAnalysis)
232                 PGOSS << "[" << SE.Prob << "]";
233               else
234                 PGOSS.write_hex(SE.Prob.getNumerator());
235             });
236       }
237     }
238     PGOSS << ")";
239 
240     return PGOString;
241   }
242 
243 private:
244   const BBAddrMap AddrMap;
245   const PGOAnalysisMap PGOMap;
246 };
247 
248 // This class represents the BBAddrMap and PGOMap of potentially multiple
249 // functions in a section.
250 class BBAddrMapInfo {
251 public:
clear()252   void clear() {
253     FunctionAddrToMap.clear();
254     RangeBaseAddrToFunctionAddr.clear();
255   }
256 
empty() const257   bool empty() const { return FunctionAddrToMap.empty(); }
258 
AddFunctionEntry(BBAddrMap AddrMap,PGOAnalysisMap PGOMap)259   void AddFunctionEntry(BBAddrMap AddrMap, PGOAnalysisMap PGOMap) {
260     uint64_t FunctionAddr = AddrMap.getFunctionAddress();
261     for (size_t I = 1; I < AddrMap.BBRanges.size(); ++I)
262       RangeBaseAddrToFunctionAddr.emplace(AddrMap.BBRanges[I].BaseAddress,
263                                           FunctionAddr);
264     [[maybe_unused]] auto R = FunctionAddrToMap.try_emplace(
265         FunctionAddr, std::move(AddrMap), std::move(PGOMap));
266     assert(R.second && "duplicate function address");
267   }
268 
269   // Returns the BBAddrMap entry for the function associated with `BaseAddress`.
270   // `BaseAddress` could be the function address or the address of a range
271   // associated with that function. Returns `nullptr` if `BaseAddress` is not
272   // mapped to any entry.
getEntryForAddress(uint64_t BaseAddress) const273   const BBAddrMapFunctionEntry *getEntryForAddress(uint64_t BaseAddress) const {
274     uint64_t FunctionAddr = BaseAddress;
275     auto S = RangeBaseAddrToFunctionAddr.find(BaseAddress);
276     if (S != RangeBaseAddrToFunctionAddr.end())
277       FunctionAddr = S->second;
278     auto R = FunctionAddrToMap.find(FunctionAddr);
279     if (R == FunctionAddrToMap.end())
280       return nullptr;
281     return &R->second;
282   }
283 
284 private:
285   std::unordered_map<uint64_t, BBAddrMapFunctionEntry> FunctionAddrToMap;
286   std::unordered_map<uint64_t, uint64_t> RangeBaseAddrToFunctionAddr;
287 };
288 
289 } // namespace
290 
291 #define DEBUG_TYPE "objdump"
292 
293 enum class ColorOutput {
294   Auto,
295   Enable,
296   Disable,
297   Invalid,
298 };
299 
300 static uint64_t AdjustVMA;
301 static bool AllHeaders;
302 static std::string ArchName;
303 bool objdump::ArchiveHeaders;
304 bool objdump::Demangle;
305 bool objdump::Disassemble;
306 bool objdump::DisassembleAll;
307 std::vector<std::string> objdump::DisassemblerOptions;
308 bool objdump::SymbolDescription;
309 bool objdump::TracebackTable;
310 static std::vector<std::string> DisassembleSymbols;
311 static bool DisassembleZeroes;
312 static ColorOutput DisassemblyColor;
313 DIDumpType objdump::DwarfDumpType;
314 static bool DynamicRelocations;
315 static bool FaultMapSection;
316 static bool FileHeaders;
317 bool objdump::SectionContents;
318 static std::vector<std::string> InputFilenames;
319 bool objdump::PrintLines;
320 static bool MachOOpt;
321 std::string objdump::MCPU;
322 std::vector<std::string> objdump::MAttrs;
323 bool objdump::ShowRawInsn;
324 bool objdump::LeadingAddr;
325 static bool Offloading;
326 static bool RawClangAST;
327 bool objdump::Relocations;
328 bool objdump::PrintImmHex;
329 bool objdump::PrivateHeaders;
330 std::vector<std::string> objdump::FilterSections;
331 bool objdump::SectionHeaders;
332 static bool ShowAllSymbols;
333 static bool ShowLMA;
334 bool objdump::PrintSource;
335 
336 static uint64_t StartAddress;
337 static bool HasStartAddressFlag;
338 static uint64_t StopAddress = UINT64_MAX;
339 static bool HasStopAddressFlag;
340 
341 bool objdump::SymbolTable;
342 static bool SymbolizeOperands;
343 static bool PrettyPGOAnalysisMap;
344 static bool DynamicSymbolTable;
345 std::string objdump::TripleName;
346 bool objdump::UnwindInfo;
347 static bool Wide;
348 std::string objdump::Prefix;
349 uint32_t objdump::PrefixStrip;
350 
351 DebugVarsFormat objdump::DbgVariables = DVDisabled;
352 
353 int objdump::DbgIndent = 52;
354 
355 static StringSet<> DisasmSymbolSet;
356 StringSet<> objdump::FoundSectionSet;
357 static StringRef ToolName;
358 
359 std::unique_ptr<BuildIDFetcher> BIDFetcher;
360 
Dumper(const object::ObjectFile & O)361 Dumper::Dumper(const object::ObjectFile &O) : O(O), OS(outs()) {
362   WarningHandler = [this](const Twine &Msg) {
363     if (Warnings.insert(Msg.str()).second)
364       reportWarning(Msg, this->O.getFileName());
365     return Error::success();
366   };
367 }
368 
reportUniqueWarning(Error Err)369 void Dumper::reportUniqueWarning(Error Err) {
370   reportUniqueWarning(toString(std::move(Err)));
371 }
372 
reportUniqueWarning(const Twine & Msg)373 void Dumper::reportUniqueWarning(const Twine &Msg) {
374   cantFail(WarningHandler(Msg));
375 }
376 
createDumper(const ObjectFile & Obj)377 static Expected<std::unique_ptr<Dumper>> createDumper(const ObjectFile &Obj) {
378   if (const auto *O = dyn_cast<COFFObjectFile>(&Obj))
379     return createCOFFDumper(*O);
380   if (const auto *O = dyn_cast<ELFObjectFileBase>(&Obj))
381     return createELFDumper(*O);
382   if (const auto *O = dyn_cast<MachOObjectFile>(&Obj))
383     return createMachODumper(*O);
384   if (const auto *O = dyn_cast<WasmObjectFile>(&Obj))
385     return createWasmDumper(*O);
386   if (const auto *O = dyn_cast<XCOFFObjectFile>(&Obj))
387     return createXCOFFDumper(*O);
388 
389   return createStringError(errc::invalid_argument,
390                            "unsupported object file format");
391 }
392 
393 namespace {
394 struct FilterResult {
395   // True if the section should not be skipped.
396   bool Keep;
397 
398   // True if the index counter should be incremented, even if the section should
399   // be skipped. For example, sections may be skipped if they are not included
400   // in the --section flag, but we still want those to count toward the section
401   // count.
402   bool IncrementIndex;
403 };
404 } // namespace
405 
checkSectionFilter(object::SectionRef S)406 static FilterResult checkSectionFilter(object::SectionRef S) {
407   if (FilterSections.empty())
408     return {/*Keep=*/true, /*IncrementIndex=*/true};
409 
410   Expected<StringRef> SecNameOrErr = S.getName();
411   if (!SecNameOrErr) {
412     consumeError(SecNameOrErr.takeError());
413     return {/*Keep=*/false, /*IncrementIndex=*/false};
414   }
415   StringRef SecName = *SecNameOrErr;
416 
417   // StringSet does not allow empty key so avoid adding sections with
418   // no name (such as the section with index 0) here.
419   if (!SecName.empty())
420     FoundSectionSet.insert(SecName);
421 
422   // Only show the section if it's in the FilterSections list, but always
423   // increment so the indexing is stable.
424   return {/*Keep=*/is_contained(FilterSections, SecName),
425           /*IncrementIndex=*/true};
426 }
427 
ToolSectionFilter(object::ObjectFile const & O,uint64_t * Idx)428 SectionFilter objdump::ToolSectionFilter(object::ObjectFile const &O,
429                                          uint64_t *Idx) {
430   // Start at UINT64_MAX so that the first index returned after an increment is
431   // zero (after the unsigned wrap).
432   if (Idx)
433     *Idx = UINT64_MAX;
434   return SectionFilter(
435       [Idx](object::SectionRef S) {
436         FilterResult Result = checkSectionFilter(S);
437         if (Idx != nullptr && Result.IncrementIndex)
438           *Idx += 1;
439         return Result.Keep;
440       },
441       O);
442 }
443 
getFileNameForError(const object::Archive::Child & C,unsigned Index)444 std::string objdump::getFileNameForError(const object::Archive::Child &C,
445                                          unsigned Index) {
446   Expected<StringRef> NameOrErr = C.getName();
447   if (NameOrErr)
448     return std::string(NameOrErr.get());
449   // If we have an error getting the name then we print the index of the archive
450   // member. Since we are already in an error state, we just ignore this error.
451   consumeError(NameOrErr.takeError());
452   return "<file index: " + std::to_string(Index) + ">";
453 }
454 
reportWarning(const Twine & Message,StringRef File)455 void objdump::reportWarning(const Twine &Message, StringRef File) {
456   // Output order between errs() and outs() matters especially for archive
457   // files where the output is per member object.
458   outs().flush();
459   WithColor::warning(errs(), ToolName)
460       << "'" << File << "': " << Message << "\n";
461 }
462 
reportError(StringRef File,const Twine & Message)463 [[noreturn]] void objdump::reportError(StringRef File, const Twine &Message) {
464   outs().flush();
465   WithColor::error(errs(), ToolName) << "'" << File << "': " << Message << "\n";
466   exit(1);
467 }
468 
reportError(Error E,StringRef FileName,StringRef ArchiveName,StringRef ArchitectureName)469 [[noreturn]] void objdump::reportError(Error E, StringRef FileName,
470                                        StringRef ArchiveName,
471                                        StringRef ArchitectureName) {
472   assert(E);
473   outs().flush();
474   WithColor::error(errs(), ToolName);
475   if (ArchiveName != "")
476     errs() << ArchiveName << "(" << FileName << ")";
477   else
478     errs() << "'" << FileName << "'";
479   if (!ArchitectureName.empty())
480     errs() << " (for architecture " << ArchitectureName << ")";
481   errs() << ": ";
482   logAllUnhandledErrors(std::move(E), errs());
483   exit(1);
484 }
485 
reportCmdLineWarning(const Twine & Message)486 static void reportCmdLineWarning(const Twine &Message) {
487   WithColor::warning(errs(), ToolName) << Message << "\n";
488 }
489 
reportCmdLineError(const Twine & Message)490 [[noreturn]] static void reportCmdLineError(const Twine &Message) {
491   WithColor::error(errs(), ToolName) << Message << "\n";
492   exit(1);
493 }
494 
warnOnNoMatchForSections()495 static void warnOnNoMatchForSections() {
496   SetVector<StringRef> MissingSections;
497   for (StringRef S : FilterSections) {
498     if (FoundSectionSet.count(S))
499       return;
500     // User may specify a unnamed section. Don't warn for it.
501     if (!S.empty())
502       MissingSections.insert(S);
503   }
504 
505   // Warn only if no section in FilterSections is matched.
506   for (StringRef S : MissingSections)
507     reportCmdLineWarning("section '" + S +
508                          "' mentioned in a -j/--section option, but not "
509                          "found in any input file");
510 }
511 
getTarget(const ObjectFile * Obj)512 static const Target *getTarget(const ObjectFile *Obj) {
513   // Figure out the target triple.
514   Triple TheTriple("unknown-unknown-unknown");
515   if (TripleName.empty()) {
516     TheTriple = Obj->makeTriple();
517   } else {
518     TheTriple.setTriple(Triple::normalize(TripleName));
519     auto Arch = Obj->getArch();
520     if (Arch == Triple::arm || Arch == Triple::armeb)
521       Obj->setARMSubArch(TheTriple);
522   }
523 
524   // Get the target specific parser.
525   std::string Error;
526   const Target *TheTarget = TargetRegistry::lookupTarget(ArchName, TheTriple,
527                                                          Error);
528   if (!TheTarget)
529     reportError(Obj->getFileName(), "can't find target: " + Error);
530 
531   // Update the triple name and return the found target.
532   TripleName = TheTriple.getTriple();
533   return TheTarget;
534 }
535 
isRelocAddressLess(RelocationRef A,RelocationRef B)536 bool objdump::isRelocAddressLess(RelocationRef A, RelocationRef B) {
537   return A.getOffset() < B.getOffset();
538 }
539 
getRelocationValueString(const RelocationRef & Rel,bool SymbolDescription,SmallVectorImpl<char> & Result)540 static Error getRelocationValueString(const RelocationRef &Rel,
541                                       bool SymbolDescription,
542                                       SmallVectorImpl<char> &Result) {
543   const ObjectFile *Obj = Rel.getObject();
544   if (auto *ELF = dyn_cast<ELFObjectFileBase>(Obj))
545     return getELFRelocationValueString(ELF, Rel, Result);
546   if (auto *COFF = dyn_cast<COFFObjectFile>(Obj))
547     return getCOFFRelocationValueString(COFF, Rel, Result);
548   if (auto *Wasm = dyn_cast<WasmObjectFile>(Obj))
549     return getWasmRelocationValueString(Wasm, Rel, Result);
550   if (auto *MachO = dyn_cast<MachOObjectFile>(Obj))
551     return getMachORelocationValueString(MachO, Rel, Result);
552   if (auto *XCOFF = dyn_cast<XCOFFObjectFile>(Obj))
553     return getXCOFFRelocationValueString(*XCOFF, Rel, SymbolDescription,
554                                          Result);
555   llvm_unreachable("unknown object file format");
556 }
557 
558 /// Indicates whether this relocation should hidden when listing
559 /// relocations, usually because it is the trailing part of a multipart
560 /// relocation that will be printed as part of the leading relocation.
getHidden(RelocationRef RelRef)561 static bool getHidden(RelocationRef RelRef) {
562   auto *MachO = dyn_cast<MachOObjectFile>(RelRef.getObject());
563   if (!MachO)
564     return false;
565 
566   unsigned Arch = MachO->getArch();
567   DataRefImpl Rel = RelRef.getRawDataRefImpl();
568   uint64_t Type = MachO->getRelocationType(Rel);
569 
570   // On arches that use the generic relocations, GENERIC_RELOC_PAIR
571   // is always hidden.
572   if (Arch == Triple::x86 || Arch == Triple::arm || Arch == Triple::ppc)
573     return Type == MachO::GENERIC_RELOC_PAIR;
574 
575   if (Arch == Triple::x86_64) {
576     // On x86_64, X86_64_RELOC_UNSIGNED is hidden only when it follows
577     // an X86_64_RELOC_SUBTRACTOR.
578     if (Type == MachO::X86_64_RELOC_UNSIGNED && Rel.d.a > 0) {
579       DataRefImpl RelPrev = Rel;
580       RelPrev.d.a--;
581       uint64_t PrevType = MachO->getRelocationType(RelPrev);
582       if (PrevType == MachO::X86_64_RELOC_SUBTRACTOR)
583         return true;
584     }
585   }
586 
587   return false;
588 }
589 
590 /// Get the column at which we want to start printing the instruction
591 /// disassembly, taking into account anything which appears to the left of it.
getInstStartColumn(const MCSubtargetInfo & STI)592 unsigned objdump::getInstStartColumn(const MCSubtargetInfo &STI) {
593   return !ShowRawInsn ? 16 : STI.getTargetTriple().isX86() ? 40 : 24;
594 }
595 
AlignToInstStartColumn(size_t Start,const MCSubtargetInfo & STI,raw_ostream & OS)596 static void AlignToInstStartColumn(size_t Start, const MCSubtargetInfo &STI,
597                                    raw_ostream &OS) {
598   // The output of printInst starts with a tab. Print some spaces so that
599   // the tab has 1 column and advances to the target tab stop.
600   unsigned TabStop = getInstStartColumn(STI);
601   unsigned Column = OS.tell() - Start;
602   OS.indent(Column < TabStop - 1 ? TabStop - 1 - Column : 7 - Column % 8);
603 }
604 
printRawData(ArrayRef<uint8_t> Bytes,uint64_t Address,formatted_raw_ostream & OS,MCSubtargetInfo const & STI)605 void objdump::printRawData(ArrayRef<uint8_t> Bytes, uint64_t Address,
606                            formatted_raw_ostream &OS,
607                            MCSubtargetInfo const &STI) {
608   size_t Start = OS.tell();
609   if (LeadingAddr)
610     OS << format("%8" PRIx64 ":", Address);
611   if (ShowRawInsn) {
612     OS << ' ';
613     dumpBytes(Bytes, OS);
614   }
615   AlignToInstStartColumn(Start, STI, OS);
616 }
617 
618 namespace {
619 
isAArch64Elf(const ObjectFile & Obj)620 static bool isAArch64Elf(const ObjectFile &Obj) {
621   const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj);
622   return Elf && Elf->getEMachine() == ELF::EM_AARCH64;
623 }
624 
isArmElf(const ObjectFile & Obj)625 static bool isArmElf(const ObjectFile &Obj) {
626   const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj);
627   return Elf && Elf->getEMachine() == ELF::EM_ARM;
628 }
629 
isCSKYElf(const ObjectFile & Obj)630 static bool isCSKYElf(const ObjectFile &Obj) {
631   const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj);
632   return Elf && Elf->getEMachine() == ELF::EM_CSKY;
633 }
634 
hasMappingSymbols(const ObjectFile & Obj)635 static bool hasMappingSymbols(const ObjectFile &Obj) {
636   return isArmElf(Obj) || isAArch64Elf(Obj) || isCSKYElf(Obj) ;
637 }
638 
printRelocation(formatted_raw_ostream & OS,StringRef FileName,const RelocationRef & Rel,uint64_t Address,bool Is64Bits)639 static void printRelocation(formatted_raw_ostream &OS, StringRef FileName,
640                             const RelocationRef &Rel, uint64_t Address,
641                             bool Is64Bits) {
642   StringRef Fmt = Is64Bits ? "%016" PRIx64 ":  " : "%08" PRIx64 ":  ";
643   SmallString<16> Name;
644   SmallString<32> Val;
645   Rel.getTypeName(Name);
646   if (Error E = getRelocationValueString(Rel, SymbolDescription, Val))
647     reportError(std::move(E), FileName);
648   OS << (Is64Bits || !LeadingAddr ? "\t\t" : "\t\t\t");
649   if (LeadingAddr)
650     OS << format(Fmt.data(), Address);
651   OS << Name << "\t" << Val;
652 }
653 
printBTFRelocation(formatted_raw_ostream & FOS,llvm::BTFParser & BTF,object::SectionedAddress Address,LiveVariablePrinter & LVP)654 static void printBTFRelocation(formatted_raw_ostream &FOS, llvm::BTFParser &BTF,
655                                object::SectionedAddress Address,
656                                LiveVariablePrinter &LVP) {
657   const llvm::BTF::BPFFieldReloc *Reloc = BTF.findFieldReloc(Address);
658   if (!Reloc)
659     return;
660 
661   SmallString<64> Val;
662   BTF.symbolize(Reloc, Val);
663   FOS << "\t\t";
664   if (LeadingAddr)
665     FOS << format("%016" PRIx64 ":  ", Address.Address + AdjustVMA);
666   FOS << "CO-RE " << Val;
667   LVP.printAfterOtherLine(FOS, true);
668 }
669 
670 class PrettyPrinter {
671 public:
672   virtual ~PrettyPrinter() = default;
673   virtual void
printInst(MCInstPrinter & IP,const MCInst * MI,ArrayRef<uint8_t> Bytes,object::SectionedAddress Address,formatted_raw_ostream & OS,StringRef Annot,MCSubtargetInfo const & STI,SourcePrinter * SP,StringRef ObjectFilename,std::vector<RelocationRef> * Rels,LiveVariablePrinter & LVP)674   printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes,
675             object::SectionedAddress Address, formatted_raw_ostream &OS,
676             StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP,
677             StringRef ObjectFilename, std::vector<RelocationRef> *Rels,
678             LiveVariablePrinter &LVP) {
679     if (SP && (PrintSource || PrintLines))
680       SP->printSourceLine(OS, Address, ObjectFilename, LVP);
681     LVP.printBetweenInsts(OS, false);
682 
683     printRawData(Bytes, Address.Address, OS, STI);
684 
685     if (MI) {
686       // See MCInstPrinter::printInst. On targets where a PC relative immediate
687       // is relative to the next instruction and the length of a MCInst is
688       // difficult to measure (x86), this is the address of the next
689       // instruction.
690       uint64_t Addr =
691           Address.Address + (STI.getTargetTriple().isX86() ? Bytes.size() : 0);
692       IP.printInst(MI, Addr, "", STI, OS);
693     } else
694       OS << "\t<unknown>";
695   }
696 
emitPostInstructionInfo(formatted_raw_ostream & FOS,const MCAsmInfo & MAI,const MCSubtargetInfo & STI,StringRef Comments,LiveVariablePrinter & LVP)697   virtual void emitPostInstructionInfo(formatted_raw_ostream &FOS,
698                                        const MCAsmInfo &MAI,
699                                        const MCSubtargetInfo &STI,
700                                        StringRef Comments,
701                                        LiveVariablePrinter &LVP) {
702     do {
703       if (!Comments.empty()) {
704         // Emit a line of comments.
705         StringRef Comment;
706         std::tie(Comment, Comments) = Comments.split('\n');
707         // MAI.getCommentColumn() assumes that instructions are printed at the
708         // position of 8, while getInstStartColumn() returns the actual
709         // position.
710         unsigned CommentColumn =
711             MAI.getCommentColumn() - 8 + getInstStartColumn(STI);
712         FOS.PadToColumn(CommentColumn);
713         FOS << MAI.getCommentString() << ' ' << Comment;
714       }
715       LVP.printAfterInst(FOS);
716       FOS << "\n";
717     } while (!Comments.empty());
718     FOS.flush();
719   }
720 
721   // Hook invoked when starting to disassemble a symbol at the current position.
722   // Default is no-op.
onSymbolStart()723   virtual void onSymbolStart() {}
724 };
725 PrettyPrinter PrettyPrinterInst;
726 
727 class HexagonPrettyPrinter : public PrettyPrinter {
728 public:
onSymbolStart()729   void onSymbolStart() override { reset(); }
730 
printLead(ArrayRef<uint8_t> Bytes,uint64_t Address,formatted_raw_ostream & OS)731   void printLead(ArrayRef<uint8_t> Bytes, uint64_t Address,
732                  formatted_raw_ostream &OS) {
733     if (LeadingAddr)
734       OS << format("%8" PRIx64 ":", Address);
735     if (ShowRawInsn) {
736       OS << "\t";
737       if (Bytes.size() >= 4) {
738         dumpBytes(Bytes.slice(0, 4), OS);
739         uint32_t opcode =
740             (Bytes[3] << 24) | (Bytes[2] << 16) | (Bytes[1] << 8) | Bytes[0];
741         OS << format("\t%08" PRIx32, opcode);
742       } else {
743         dumpBytes(Bytes, OS);
744       }
745     }
746   }
747 
getInstructionSeparator() const748   std::string getInstructionSeparator() const {
749     SmallString<40> Separator;
750     raw_svector_ostream OS(Separator);
751     if (ShouldClosePacket) {
752       OS << " }";
753       if (IsLoop0 || IsLoop1)
754         OS << "  ";
755       if (IsLoop0)
756         OS << (IsLoop1 ? ":endloop01" : ":endloop0");
757       else if (IsLoop1)
758         OS << ":endloop1";
759     }
760     OS << '\n';
761     return OS.str().str();
762   }
763 
emitPostInstructionInfo(formatted_raw_ostream & FOS,const MCAsmInfo & MAI,const MCSubtargetInfo & STI,StringRef Comments,LiveVariablePrinter & LVP)764   void emitPostInstructionInfo(formatted_raw_ostream &FOS, const MCAsmInfo &MAI,
765                                const MCSubtargetInfo &STI, StringRef Comments,
766                                LiveVariablePrinter &LVP) override {
767     // Hexagon does not write anything to the comment stream, so we can just
768     // print the separator.
769     LVP.printAfterInst(FOS);
770     FOS << getInstructionSeparator();
771     FOS.flush();
772     if (ShouldClosePacket)
773       reset();
774   }
775 
printInst(MCInstPrinter & IP,const MCInst * MI,ArrayRef<uint8_t> Bytes,object::SectionedAddress Address,formatted_raw_ostream & OS,StringRef Annot,MCSubtargetInfo const & STI,SourcePrinter * SP,StringRef ObjectFilename,std::vector<RelocationRef> * Rels,LiveVariablePrinter & LVP)776   void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes,
777                  object::SectionedAddress Address, formatted_raw_ostream &OS,
778                  StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP,
779                  StringRef ObjectFilename, std::vector<RelocationRef> *Rels,
780                  LiveVariablePrinter &LVP) override {
781     if (SP && (PrintSource || PrintLines))
782       SP->printSourceLine(OS, Address, ObjectFilename, LVP, "");
783     if (!MI) {
784       printLead(Bytes, Address.Address, OS);
785       OS << " <unknown>";
786       reset();
787       return;
788     }
789 
790     StringRef Preamble = IsStartOfBundle ? " { " : "   ";
791 
792     if (SP && (PrintSource || PrintLines))
793       SP->printSourceLine(OS, Address, ObjectFilename, LVP, "");
794     printLead(Bytes, Address.Address, OS);
795     OS << Preamble;
796     std::string Buf;
797     {
798       raw_string_ostream TempStream(Buf);
799       IP.printInst(MI, Address.Address, "", STI, TempStream);
800     }
801     StringRef Contents(Buf);
802 
803     auto Duplex = Contents.split('\v');
804     bool HasDuplex = !Duplex.second.empty();
805     if (HasDuplex) {
806       OS << Duplex.first;
807       OS << "; ";
808       OS << Duplex.second;
809     } else {
810       OS << Duplex.first;
811     }
812 
813     uint32_t Instruction = support::endian::read32le(Bytes.data());
814 
815     uint32_t ParseMask = 0x0000c000;
816     uint32_t PacketEndMask = 0x0000c000;
817     uint32_t LoopEndMask = 0x00008000;
818     uint32_t ParseBits = Instruction & ParseMask;
819 
820     if (ParseBits == LoopEndMask) {
821       if (IsStartOfBundle)
822         IsLoop0 = true;
823       else
824         IsLoop1 = true;
825     }
826 
827     IsStartOfBundle = false;
828 
829     if (ParseBits == PacketEndMask || HasDuplex)
830       ShouldClosePacket = true;
831   }
832 
833 private:
834   bool IsStartOfBundle = true;
835   bool IsLoop0 = false;
836   bool IsLoop1 = false;
837   bool ShouldClosePacket = false;
838 
reset()839   void reset() {
840     IsStartOfBundle = true;
841     IsLoop0 = false;
842     IsLoop1 = false;
843     ShouldClosePacket = false;
844   }
845 };
846 HexagonPrettyPrinter HexagonPrettyPrinterInst;
847 
848 class AMDGCNPrettyPrinter : public PrettyPrinter {
849 public:
printInst(MCInstPrinter & IP,const MCInst * MI,ArrayRef<uint8_t> Bytes,object::SectionedAddress Address,formatted_raw_ostream & OS,StringRef Annot,MCSubtargetInfo const & STI,SourcePrinter * SP,StringRef ObjectFilename,std::vector<RelocationRef> * Rels,LiveVariablePrinter & LVP)850   void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes,
851                  object::SectionedAddress Address, formatted_raw_ostream &OS,
852                  StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP,
853                  StringRef ObjectFilename, std::vector<RelocationRef> *Rels,
854                  LiveVariablePrinter &LVP) override {
855     if (SP && (PrintSource || PrintLines))
856       SP->printSourceLine(OS, Address, ObjectFilename, LVP);
857 
858     if (MI) {
859       SmallString<40> InstStr;
860       raw_svector_ostream IS(InstStr);
861 
862       IP.printInst(MI, Address.Address, "", STI, IS);
863 
864       OS << left_justify(IS.str(), 60);
865     } else {
866       // an unrecognized encoding - this is probably data so represent it
867       // using the .long directive, or .byte directive if fewer than 4 bytes
868       // remaining
869       if (Bytes.size() >= 4) {
870         OS << format(
871             "\t.long 0x%08" PRIx32 " ",
872             support::endian::read32<llvm::endianness::little>(Bytes.data()));
873         OS.indent(42);
874       } else {
875           OS << format("\t.byte 0x%02" PRIx8, Bytes[0]);
876           for (unsigned int i = 1; i < Bytes.size(); i++)
877             OS << format(", 0x%02" PRIx8, Bytes[i]);
878           OS.indent(55 - (6 * Bytes.size()));
879       }
880     }
881 
882     OS << format("// %012" PRIX64 ":", Address.Address);
883     if (Bytes.size() >= 4) {
884       // D should be casted to uint32_t here as it is passed by format to
885       // snprintf as vararg.
886       for (uint32_t D :
887            ArrayRef(reinterpret_cast<const support::little32_t *>(Bytes.data()),
888                     Bytes.size() / 4))
889           OS << format(" %08" PRIX32, D);
890     } else {
891       for (unsigned char B : Bytes)
892         OS << format(" %02" PRIX8, B);
893     }
894 
895     if (!Annot.empty())
896       OS << " // " << Annot;
897   }
898 };
899 AMDGCNPrettyPrinter AMDGCNPrettyPrinterInst;
900 
901 class BPFPrettyPrinter : public PrettyPrinter {
902 public:
printInst(MCInstPrinter & IP,const MCInst * MI,ArrayRef<uint8_t> Bytes,object::SectionedAddress Address,formatted_raw_ostream & OS,StringRef Annot,MCSubtargetInfo const & STI,SourcePrinter * SP,StringRef ObjectFilename,std::vector<RelocationRef> * Rels,LiveVariablePrinter & LVP)903   void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes,
904                  object::SectionedAddress Address, formatted_raw_ostream &OS,
905                  StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP,
906                  StringRef ObjectFilename, std::vector<RelocationRef> *Rels,
907                  LiveVariablePrinter &LVP) override {
908     if (SP && (PrintSource || PrintLines))
909       SP->printSourceLine(OS, Address, ObjectFilename, LVP);
910     if (LeadingAddr)
911       OS << format("%8" PRId64 ":", Address.Address / 8);
912     if (ShowRawInsn) {
913       OS << "\t";
914       dumpBytes(Bytes, OS);
915     }
916     if (MI)
917       IP.printInst(MI, Address.Address, "", STI, OS);
918     else
919       OS << "\t<unknown>";
920   }
921 };
922 BPFPrettyPrinter BPFPrettyPrinterInst;
923 
924 class ARMPrettyPrinter : public PrettyPrinter {
925 public:
printInst(MCInstPrinter & IP,const MCInst * MI,ArrayRef<uint8_t> Bytes,object::SectionedAddress Address,formatted_raw_ostream & OS,StringRef Annot,MCSubtargetInfo const & STI,SourcePrinter * SP,StringRef ObjectFilename,std::vector<RelocationRef> * Rels,LiveVariablePrinter & LVP)926   void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes,
927                  object::SectionedAddress Address, formatted_raw_ostream &OS,
928                  StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP,
929                  StringRef ObjectFilename, std::vector<RelocationRef> *Rels,
930                  LiveVariablePrinter &LVP) override {
931     if (SP && (PrintSource || PrintLines))
932       SP->printSourceLine(OS, Address, ObjectFilename, LVP);
933     LVP.printBetweenInsts(OS, false);
934 
935     size_t Start = OS.tell();
936     if (LeadingAddr)
937       OS << format("%8" PRIx64 ":", Address.Address);
938     if (ShowRawInsn) {
939       size_t Pos = 0, End = Bytes.size();
940       if (STI.checkFeatures("+thumb-mode")) {
941         for (; Pos + 2 <= End; Pos += 2)
942           OS << ' '
943              << format_hex_no_prefix(
944                     llvm::support::endian::read<uint16_t>(
945                         Bytes.data() + Pos, InstructionEndianness),
946                     4);
947       } else {
948         for (; Pos + 4 <= End; Pos += 4)
949           OS << ' '
950              << format_hex_no_prefix(
951                     llvm::support::endian::read<uint32_t>(
952                         Bytes.data() + Pos, InstructionEndianness),
953                     8);
954       }
955       if (Pos < End) {
956         OS << ' ';
957         dumpBytes(Bytes.slice(Pos), OS);
958       }
959     }
960 
961     AlignToInstStartColumn(Start, STI, OS);
962 
963     if (MI) {
964       IP.printInst(MI, Address.Address, "", STI, OS);
965     } else
966       OS << "\t<unknown>";
967   }
968 
setInstructionEndianness(llvm::endianness Endianness)969   void setInstructionEndianness(llvm::endianness Endianness) {
970     InstructionEndianness = Endianness;
971   }
972 
973 private:
974   llvm::endianness InstructionEndianness = llvm::endianness::little;
975 };
976 ARMPrettyPrinter ARMPrettyPrinterInst;
977 
978 class AArch64PrettyPrinter : public PrettyPrinter {
979 public:
printInst(MCInstPrinter & IP,const MCInst * MI,ArrayRef<uint8_t> Bytes,object::SectionedAddress Address,formatted_raw_ostream & OS,StringRef Annot,MCSubtargetInfo const & STI,SourcePrinter * SP,StringRef ObjectFilename,std::vector<RelocationRef> * Rels,LiveVariablePrinter & LVP)980   void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes,
981                  object::SectionedAddress Address, formatted_raw_ostream &OS,
982                  StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP,
983                  StringRef ObjectFilename, std::vector<RelocationRef> *Rels,
984                  LiveVariablePrinter &LVP) override {
985     if (SP && (PrintSource || PrintLines))
986       SP->printSourceLine(OS, Address, ObjectFilename, LVP);
987     LVP.printBetweenInsts(OS, false);
988 
989     size_t Start = OS.tell();
990     if (LeadingAddr)
991       OS << format("%8" PRIx64 ":", Address.Address);
992     if (ShowRawInsn) {
993       size_t Pos = 0, End = Bytes.size();
994       for (; Pos + 4 <= End; Pos += 4)
995         OS << ' '
996            << format_hex_no_prefix(
997                   llvm::support::endian::read<uint32_t>(
998                       Bytes.data() + Pos, llvm::endianness::little),
999                   8);
1000       if (Pos < End) {
1001         OS << ' ';
1002         dumpBytes(Bytes.slice(Pos), OS);
1003       }
1004     }
1005 
1006     AlignToInstStartColumn(Start, STI, OS);
1007 
1008     if (MI) {
1009       IP.printInst(MI, Address.Address, "", STI, OS);
1010     } else
1011       OS << "\t<unknown>";
1012   }
1013 };
1014 AArch64PrettyPrinter AArch64PrettyPrinterInst;
1015 
1016 class RISCVPrettyPrinter : public PrettyPrinter {
1017 public:
printInst(MCInstPrinter & IP,const MCInst * MI,ArrayRef<uint8_t> Bytes,object::SectionedAddress Address,formatted_raw_ostream & OS,StringRef Annot,MCSubtargetInfo const & STI,SourcePrinter * SP,StringRef ObjectFilename,std::vector<RelocationRef> * Rels,LiveVariablePrinter & LVP)1018   void printInst(MCInstPrinter &IP, const MCInst *MI, ArrayRef<uint8_t> Bytes,
1019                  object::SectionedAddress Address, formatted_raw_ostream &OS,
1020                  StringRef Annot, MCSubtargetInfo const &STI, SourcePrinter *SP,
1021                  StringRef ObjectFilename, std::vector<RelocationRef> *Rels,
1022                  LiveVariablePrinter &LVP) override {
1023     if (SP && (PrintSource || PrintLines))
1024       SP->printSourceLine(OS, Address, ObjectFilename, LVP);
1025     LVP.printBetweenInsts(OS, false);
1026 
1027     size_t Start = OS.tell();
1028     if (LeadingAddr)
1029       OS << format("%8" PRIx64 ":", Address.Address);
1030     if (ShowRawInsn) {
1031       size_t Pos = 0, End = Bytes.size();
1032       if (End % 4 == 0) {
1033         // 32-bit and 64-bit instructions.
1034         for (; Pos + 4 <= End; Pos += 4)
1035           OS << ' '
1036              << format_hex_no_prefix(
1037                     llvm::support::endian::read<uint32_t>(
1038                         Bytes.data() + Pos, llvm::endianness::little),
1039                     8);
1040       } else if (End % 2 == 0) {
1041         // 16-bit and 48-bits instructions.
1042         for (; Pos + 2 <= End; Pos += 2)
1043           OS << ' '
1044              << format_hex_no_prefix(
1045                     llvm::support::endian::read<uint16_t>(
1046                         Bytes.data() + Pos, llvm::endianness::little),
1047                     4);
1048       }
1049       if (Pos < End) {
1050         OS << ' ';
1051         dumpBytes(Bytes.slice(Pos), OS);
1052       }
1053     }
1054 
1055     AlignToInstStartColumn(Start, STI, OS);
1056 
1057     if (MI) {
1058       IP.printInst(MI, Address.Address, "", STI, OS);
1059     } else
1060       OS << "\t<unknown>";
1061   }
1062 };
1063 RISCVPrettyPrinter RISCVPrettyPrinterInst;
1064 
selectPrettyPrinter(Triple const & Triple)1065 PrettyPrinter &selectPrettyPrinter(Triple const &Triple) {
1066   switch(Triple.getArch()) {
1067   default:
1068     return PrettyPrinterInst;
1069   case Triple::hexagon:
1070     return HexagonPrettyPrinterInst;
1071   case Triple::amdgcn:
1072     return AMDGCNPrettyPrinterInst;
1073   case Triple::bpfel:
1074   case Triple::bpfeb:
1075     return BPFPrettyPrinterInst;
1076   case Triple::arm:
1077   case Triple::armeb:
1078   case Triple::thumb:
1079   case Triple::thumbeb:
1080     return ARMPrettyPrinterInst;
1081   case Triple::aarch64:
1082   case Triple::aarch64_be:
1083   case Triple::aarch64_32:
1084     return AArch64PrettyPrinterInst;
1085   case Triple::riscv32:
1086   case Triple::riscv64:
1087     return RISCVPrettyPrinterInst;
1088   }
1089 }
1090 
1091 class DisassemblerTarget {
1092 public:
1093   const Target *TheTarget;
1094   std::unique_ptr<const MCSubtargetInfo> SubtargetInfo;
1095   std::shared_ptr<MCContext> Context;
1096   std::unique_ptr<MCDisassembler> DisAsm;
1097   std::shared_ptr<MCInstrAnalysis> InstrAnalysis;
1098   std::shared_ptr<MCInstPrinter> InstPrinter;
1099   PrettyPrinter *Printer;
1100 
1101   DisassemblerTarget(const Target *TheTarget, ObjectFile &Obj,
1102                      StringRef TripleName, StringRef MCPU,
1103                      SubtargetFeatures &Features);
1104   DisassemblerTarget(DisassemblerTarget &Other, SubtargetFeatures &Features);
1105 
1106 private:
1107   MCTargetOptions Options;
1108   std::shared_ptr<const MCRegisterInfo> RegisterInfo;
1109   std::shared_ptr<const MCAsmInfo> AsmInfo;
1110   std::shared_ptr<const MCInstrInfo> InstrInfo;
1111   std::shared_ptr<MCObjectFileInfo> ObjectFileInfo;
1112 };
1113 
DisassemblerTarget(const Target * TheTarget,ObjectFile & Obj,StringRef TripleName,StringRef MCPU,SubtargetFeatures & Features)1114 DisassemblerTarget::DisassemblerTarget(const Target *TheTarget, ObjectFile &Obj,
1115                                        StringRef TripleName, StringRef MCPU,
1116                                        SubtargetFeatures &Features)
1117     : TheTarget(TheTarget),
1118       Printer(&selectPrettyPrinter(Triple(TripleName))),
1119       RegisterInfo(TheTarget->createMCRegInfo(TripleName)) {
1120   if (!RegisterInfo)
1121     reportError(Obj.getFileName(), "no register info for target " + TripleName);
1122 
1123   // Set up disassembler.
1124   AsmInfo.reset(TheTarget->createMCAsmInfo(*RegisterInfo, TripleName, Options));
1125   if (!AsmInfo)
1126     reportError(Obj.getFileName(), "no assembly info for target " + TripleName);
1127 
1128   SubtargetInfo.reset(
1129       TheTarget->createMCSubtargetInfo(TripleName, MCPU, Features.getString()));
1130   if (!SubtargetInfo)
1131     reportError(Obj.getFileName(),
1132                 "no subtarget info for target " + TripleName);
1133   InstrInfo.reset(TheTarget->createMCInstrInfo());
1134   if (!InstrInfo)
1135     reportError(Obj.getFileName(),
1136                 "no instruction info for target " + TripleName);
1137   Context =
1138       std::make_shared<MCContext>(Triple(TripleName), AsmInfo.get(),
1139                                   RegisterInfo.get(), SubtargetInfo.get());
1140 
1141   // FIXME: for now initialize MCObjectFileInfo with default values
1142   ObjectFileInfo.reset(
1143       TheTarget->createMCObjectFileInfo(*Context, /*PIC=*/false));
1144   Context->setObjectFileInfo(ObjectFileInfo.get());
1145 
1146   DisAsm.reset(TheTarget->createMCDisassembler(*SubtargetInfo, *Context));
1147   if (!DisAsm)
1148     reportError(Obj.getFileName(), "no disassembler for target " + TripleName);
1149 
1150   if (auto *ELFObj = dyn_cast<ELFObjectFileBase>(&Obj))
1151     DisAsm->setABIVersion(ELFObj->getEIdentABIVersion());
1152 
1153   InstrAnalysis.reset(TheTarget->createMCInstrAnalysis(InstrInfo.get()));
1154 
1155   int AsmPrinterVariant = AsmInfo->getAssemblerDialect();
1156   InstPrinter.reset(TheTarget->createMCInstPrinter(Triple(TripleName),
1157                                                    AsmPrinterVariant, *AsmInfo,
1158                                                    *InstrInfo, *RegisterInfo));
1159   if (!InstPrinter)
1160     reportError(Obj.getFileName(),
1161                 "no instruction printer for target " + TripleName);
1162   InstPrinter->setPrintImmHex(PrintImmHex);
1163   InstPrinter->setPrintBranchImmAsAddress(true);
1164   InstPrinter->setSymbolizeOperands(SymbolizeOperands);
1165   InstPrinter->setMCInstrAnalysis(InstrAnalysis.get());
1166 
1167   switch (DisassemblyColor) {
1168   case ColorOutput::Enable:
1169     InstPrinter->setUseColor(true);
1170     break;
1171   case ColorOutput::Auto:
1172     InstPrinter->setUseColor(outs().has_colors());
1173     break;
1174   case ColorOutput::Disable:
1175   case ColorOutput::Invalid:
1176     InstPrinter->setUseColor(false);
1177     break;
1178   };
1179 }
1180 
DisassemblerTarget(DisassemblerTarget & Other,SubtargetFeatures & Features)1181 DisassemblerTarget::DisassemblerTarget(DisassemblerTarget &Other,
1182                                        SubtargetFeatures &Features)
1183     : TheTarget(Other.TheTarget),
1184       SubtargetInfo(TheTarget->createMCSubtargetInfo(TripleName, MCPU,
1185                                                      Features.getString())),
1186       Context(Other.Context),
1187       DisAsm(TheTarget->createMCDisassembler(*SubtargetInfo, *Context)),
1188       InstrAnalysis(Other.InstrAnalysis), InstPrinter(Other.InstPrinter),
1189       Printer(Other.Printer), RegisterInfo(Other.RegisterInfo),
1190       AsmInfo(Other.AsmInfo), InstrInfo(Other.InstrInfo),
1191       ObjectFileInfo(Other.ObjectFileInfo) {}
1192 } // namespace
1193 
getElfSymbolType(const ObjectFile & Obj,const SymbolRef & Sym)1194 static uint8_t getElfSymbolType(const ObjectFile &Obj, const SymbolRef &Sym) {
1195   assert(Obj.isELF());
1196   if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj))
1197     return unwrapOrError(Elf32LEObj->getSymbol(Sym.getRawDataRefImpl()),
1198                          Obj.getFileName())
1199         ->getType();
1200   if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj))
1201     return unwrapOrError(Elf64LEObj->getSymbol(Sym.getRawDataRefImpl()),
1202                          Obj.getFileName())
1203         ->getType();
1204   if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj))
1205     return unwrapOrError(Elf32BEObj->getSymbol(Sym.getRawDataRefImpl()),
1206                          Obj.getFileName())
1207         ->getType();
1208   if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj))
1209     return unwrapOrError(Elf64BEObj->getSymbol(Sym.getRawDataRefImpl()),
1210                          Obj.getFileName())
1211         ->getType();
1212   llvm_unreachable("Unsupported binary format");
1213 }
1214 
1215 template <class ELFT>
1216 static void
addDynamicElfSymbols(const ELFObjectFile<ELFT> & Obj,std::map<SectionRef,SectionSymbolsTy> & AllSymbols)1217 addDynamicElfSymbols(const ELFObjectFile<ELFT> &Obj,
1218                      std::map<SectionRef, SectionSymbolsTy> &AllSymbols) {
1219   for (auto Symbol : Obj.getDynamicSymbolIterators()) {
1220     uint8_t SymbolType = Symbol.getELFType();
1221     if (SymbolType == ELF::STT_SECTION)
1222       continue;
1223 
1224     uint64_t Address = unwrapOrError(Symbol.getAddress(), Obj.getFileName());
1225     // ELFSymbolRef::getAddress() returns size instead of value for common
1226     // symbols which is not desirable for disassembly output. Overriding.
1227     if (SymbolType == ELF::STT_COMMON)
1228       Address = unwrapOrError(Obj.getSymbol(Symbol.getRawDataRefImpl()),
1229                               Obj.getFileName())
1230                     ->st_value;
1231 
1232     StringRef Name = unwrapOrError(Symbol.getName(), Obj.getFileName());
1233     if (Name.empty())
1234       continue;
1235 
1236     section_iterator SecI =
1237         unwrapOrError(Symbol.getSection(), Obj.getFileName());
1238     if (SecI == Obj.section_end())
1239       continue;
1240 
1241     AllSymbols[*SecI].emplace_back(Address, Name, SymbolType);
1242   }
1243 }
1244 
1245 static void
addDynamicElfSymbols(const ELFObjectFileBase & Obj,std::map<SectionRef,SectionSymbolsTy> & AllSymbols)1246 addDynamicElfSymbols(const ELFObjectFileBase &Obj,
1247                      std::map<SectionRef, SectionSymbolsTy> &AllSymbols) {
1248   if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj))
1249     addDynamicElfSymbols(*Elf32LEObj, AllSymbols);
1250   else if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj))
1251     addDynamicElfSymbols(*Elf64LEObj, AllSymbols);
1252   else if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj))
1253     addDynamicElfSymbols(*Elf32BEObj, AllSymbols);
1254   else if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj))
1255     addDynamicElfSymbols(*Elf64BEObj, AllSymbols);
1256   else
1257     llvm_unreachable("Unsupported binary format");
1258 }
1259 
getWasmCodeSection(const WasmObjectFile & Obj)1260 static std::optional<SectionRef> getWasmCodeSection(const WasmObjectFile &Obj) {
1261   for (auto SecI : Obj.sections()) {
1262     const WasmSection &Section = Obj.getWasmSection(SecI);
1263     if (Section.Type == wasm::WASM_SEC_CODE)
1264       return SecI;
1265   }
1266   return std::nullopt;
1267 }
1268 
1269 static void
addMissingWasmCodeSymbols(const WasmObjectFile & Obj,std::map<SectionRef,SectionSymbolsTy> & AllSymbols)1270 addMissingWasmCodeSymbols(const WasmObjectFile &Obj,
1271                           std::map<SectionRef, SectionSymbolsTy> &AllSymbols) {
1272   std::optional<SectionRef> Section = getWasmCodeSection(Obj);
1273   if (!Section)
1274     return;
1275   SectionSymbolsTy &Symbols = AllSymbols[*Section];
1276 
1277   std::set<uint64_t> SymbolAddresses;
1278   for (const auto &Sym : Symbols)
1279     SymbolAddresses.insert(Sym.Addr);
1280 
1281   for (const wasm::WasmFunction &Function : Obj.functions()) {
1282     // This adjustment mirrors the one in WasmObjectFile::getSymbolAddress.
1283     uint32_t Adjustment = Obj.isRelocatableObject() || Obj.isSharedObject()
1284                               ? 0
1285                               : Section->getAddress();
1286     uint64_t Address = Function.CodeSectionOffset + Adjustment;
1287     // Only add fallback symbols for functions not already present in the symbol
1288     // table.
1289     if (SymbolAddresses.count(Address))
1290       continue;
1291     // This function has no symbol, so it should have no SymbolName.
1292     assert(Function.SymbolName.empty());
1293     // We use DebugName for the name, though it may be empty if there is no
1294     // "name" custom section, or that section is missing a name for this
1295     // function.
1296     StringRef Name = Function.DebugName;
1297     Symbols.emplace_back(Address, Name, ELF::STT_NOTYPE);
1298   }
1299 }
1300 
getSectionNames(const ObjectFile & Obj)1301 static DenseMap<StringRef, SectionRef> getSectionNames(const ObjectFile &Obj) {
1302   DenseMap<StringRef, SectionRef> Sections;
1303   for (SectionRef Section : Obj.sections()) {
1304     Expected<StringRef> SecNameOrErr = Section.getName();
1305     if (!SecNameOrErr) {
1306       consumeError(SecNameOrErr.takeError());
1307       continue;
1308     }
1309     Sections[*SecNameOrErr] = Section;
1310   }
1311   return Sections;
1312 }
1313 
addPltEntries(const MCSubtargetInfo & STI,const ObjectFile & Obj,DenseMap<StringRef,SectionRef> & SectionNames,std::map<SectionRef,SectionSymbolsTy> & AllSymbols,StringSaver & Saver)1314 static void addPltEntries(const MCSubtargetInfo &STI, const ObjectFile &Obj,
1315                           DenseMap<StringRef, SectionRef> &SectionNames,
1316                           std::map<SectionRef, SectionSymbolsTy> &AllSymbols,
1317                           StringSaver &Saver) {
1318   auto *ElfObj = dyn_cast<ELFObjectFileBase>(&Obj);
1319   if (!ElfObj)
1320     return;
1321   for (auto Plt : ElfObj->getPltEntries(STI)) {
1322     if (Plt.Symbol) {
1323       SymbolRef Symbol(*Plt.Symbol, ElfObj);
1324       uint8_t SymbolType = getElfSymbolType(Obj, Symbol);
1325       if (Expected<StringRef> NameOrErr = Symbol.getName()) {
1326         if (!NameOrErr->empty())
1327           AllSymbols[SectionNames[Plt.Section]].emplace_back(
1328               Plt.Address, Saver.save((*NameOrErr + "@plt").str()), SymbolType);
1329         continue;
1330       } else {
1331         // The warning has been reported in disassembleObject().
1332         consumeError(NameOrErr.takeError());
1333       }
1334     }
1335     reportWarning("PLT entry at 0x" + Twine::utohexstr(Plt.Address) +
1336                       " references an invalid symbol",
1337                   Obj.getFileName());
1338   }
1339 }
1340 
1341 // Normally the disassembly output will skip blocks of zeroes. This function
1342 // returns the number of zero bytes that can be skipped when dumping the
1343 // disassembly of the instructions in Buf.
countSkippableZeroBytes(ArrayRef<uint8_t> Buf)1344 static size_t countSkippableZeroBytes(ArrayRef<uint8_t> Buf) {
1345   // Find the number of leading zeroes.
1346   size_t N = 0;
1347   while (N < Buf.size() && !Buf[N])
1348     ++N;
1349 
1350   // We may want to skip blocks of zero bytes, but unless we see
1351   // at least 8 of them in a row.
1352   if (N < 8)
1353     return 0;
1354 
1355   // We skip zeroes in multiples of 4 because do not want to truncate an
1356   // instruction if it starts with a zero byte.
1357   return N & ~0x3;
1358 }
1359 
1360 // Returns a map from sections to their relocations.
1361 static std::map<SectionRef, std::vector<RelocationRef>>
getRelocsMap(object::ObjectFile const & Obj)1362 getRelocsMap(object::ObjectFile const &Obj) {
1363   std::map<SectionRef, std::vector<RelocationRef>> Ret;
1364   uint64_t I = (uint64_t)-1;
1365   for (SectionRef Sec : Obj.sections()) {
1366     ++I;
1367     Expected<section_iterator> RelocatedOrErr = Sec.getRelocatedSection();
1368     if (!RelocatedOrErr)
1369       reportError(Obj.getFileName(),
1370                   "section (" + Twine(I) +
1371                       "): failed to get a relocated section: " +
1372                       toString(RelocatedOrErr.takeError()));
1373 
1374     section_iterator Relocated = *RelocatedOrErr;
1375     if (Relocated == Obj.section_end() || !checkSectionFilter(*Relocated).Keep)
1376       continue;
1377     std::vector<RelocationRef> &V = Ret[*Relocated];
1378     append_range(V, Sec.relocations());
1379     // Sort relocations by address.
1380     llvm::stable_sort(V, isRelocAddressLess);
1381   }
1382   return Ret;
1383 }
1384 
1385 // Used for --adjust-vma to check if address should be adjusted by the
1386 // specified value for a given section.
1387 // For ELF we do not adjust non-allocatable sections like debug ones,
1388 // because they are not loadable.
1389 // TODO: implement for other file formats.
shouldAdjustVA(const SectionRef & Section)1390 static bool shouldAdjustVA(const SectionRef &Section) {
1391   const ObjectFile *Obj = Section.getObject();
1392   if (Obj->isELF())
1393     return ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC;
1394   return false;
1395 }
1396 
1397 
1398 typedef std::pair<uint64_t, char> MappingSymbolPair;
getMappingSymbolKind(ArrayRef<MappingSymbolPair> MappingSymbols,uint64_t Address)1399 static char getMappingSymbolKind(ArrayRef<MappingSymbolPair> MappingSymbols,
1400                                  uint64_t Address) {
1401   auto It =
1402       partition_point(MappingSymbols, [Address](const MappingSymbolPair &Val) {
1403         return Val.first <= Address;
1404       });
1405   // Return zero for any address before the first mapping symbol; this means
1406   // we should use the default disassembly mode, depending on the target.
1407   if (It == MappingSymbols.begin())
1408     return '\x00';
1409   return (It - 1)->second;
1410 }
1411 
dumpARMELFData(uint64_t SectionAddr,uint64_t Index,uint64_t End,const ObjectFile & Obj,ArrayRef<uint8_t> Bytes,ArrayRef<MappingSymbolPair> MappingSymbols,const MCSubtargetInfo & STI,raw_ostream & OS)1412 static uint64_t dumpARMELFData(uint64_t SectionAddr, uint64_t Index,
1413                                uint64_t End, const ObjectFile &Obj,
1414                                ArrayRef<uint8_t> Bytes,
1415                                ArrayRef<MappingSymbolPair> MappingSymbols,
1416                                const MCSubtargetInfo &STI, raw_ostream &OS) {
1417   llvm::endianness Endian =
1418       Obj.isLittleEndian() ? llvm::endianness::little : llvm::endianness::big;
1419   size_t Start = OS.tell();
1420   OS << format("%8" PRIx64 ": ", SectionAddr + Index);
1421   if (Index + 4 <= End) {
1422     dumpBytes(Bytes.slice(Index, 4), OS);
1423     AlignToInstStartColumn(Start, STI, OS);
1424     OS << "\t.word\t"
1425            << format_hex(support::endian::read32(Bytes.data() + Index, Endian),
1426                          10);
1427     return 4;
1428   }
1429   if (Index + 2 <= End) {
1430     dumpBytes(Bytes.slice(Index, 2), OS);
1431     AlignToInstStartColumn(Start, STI, OS);
1432     OS << "\t.short\t"
1433        << format_hex(support::endian::read16(Bytes.data() + Index, Endian), 6);
1434     return 2;
1435   }
1436   dumpBytes(Bytes.slice(Index, 1), OS);
1437   AlignToInstStartColumn(Start, STI, OS);
1438   OS << "\t.byte\t" << format_hex(Bytes[Index], 4);
1439   return 1;
1440 }
1441 
dumpELFData(uint64_t SectionAddr,uint64_t Index,uint64_t End,ArrayRef<uint8_t> Bytes,raw_ostream & OS)1442 static void dumpELFData(uint64_t SectionAddr, uint64_t Index, uint64_t End,
1443                         ArrayRef<uint8_t> Bytes, raw_ostream &OS) {
1444   // print out data up to 8 bytes at a time in hex and ascii
1445   uint8_t AsciiData[9] = {'\0'};
1446   uint8_t Byte;
1447   int NumBytes = 0;
1448 
1449   for (; Index < End; ++Index) {
1450     if (NumBytes == 0)
1451       OS << format("%8" PRIx64 ":", SectionAddr + Index);
1452     Byte = Bytes.slice(Index)[0];
1453     OS << format(" %02x", Byte);
1454     AsciiData[NumBytes] = isPrint(Byte) ? Byte : '.';
1455 
1456     uint8_t IndentOffset = 0;
1457     NumBytes++;
1458     if (Index == End - 1 || NumBytes > 8) {
1459       // Indent the space for less than 8 bytes data.
1460       // 2 spaces for byte and one for space between bytes
1461       IndentOffset = 3 * (8 - NumBytes);
1462       for (int Excess = NumBytes; Excess < 8; Excess++)
1463         AsciiData[Excess] = '\0';
1464       NumBytes = 8;
1465     }
1466     if (NumBytes == 8) {
1467       AsciiData[8] = '\0';
1468       OS << std::string(IndentOffset, ' ') << "         ";
1469       OS << reinterpret_cast<char *>(AsciiData);
1470       OS << '\n';
1471       NumBytes = 0;
1472     }
1473   }
1474 }
1475 
createSymbolInfo(const ObjectFile & Obj,const SymbolRef & Symbol,bool IsMappingSymbol)1476 SymbolInfoTy objdump::createSymbolInfo(const ObjectFile &Obj,
1477                                        const SymbolRef &Symbol,
1478                                        bool IsMappingSymbol) {
1479   const StringRef FileName = Obj.getFileName();
1480   const uint64_t Addr = unwrapOrError(Symbol.getAddress(), FileName);
1481   const StringRef Name = unwrapOrError(Symbol.getName(), FileName);
1482 
1483   if (Obj.isXCOFF() && (SymbolDescription || TracebackTable)) {
1484     const auto &XCOFFObj = cast<XCOFFObjectFile>(Obj);
1485     DataRefImpl SymbolDRI = Symbol.getRawDataRefImpl();
1486 
1487     const uint32_t SymbolIndex = XCOFFObj.getSymbolIndex(SymbolDRI.p);
1488     std::optional<XCOFF::StorageMappingClass> Smc =
1489         getXCOFFSymbolCsectSMC(XCOFFObj, Symbol);
1490     return SymbolInfoTy(Smc, Addr, Name, SymbolIndex,
1491                         isLabel(XCOFFObj, Symbol));
1492   } else if (Obj.isXCOFF()) {
1493     const SymbolRef::Type SymType = unwrapOrError(Symbol.getType(), FileName);
1494     return SymbolInfoTy(Addr, Name, SymType, /*IsMappingSymbol=*/false,
1495                         /*IsXCOFF=*/true);
1496   } else if (Obj.isWasm()) {
1497     uint8_t SymType =
1498         cast<WasmObjectFile>(&Obj)->getWasmSymbol(Symbol).Info.Kind;
1499     return SymbolInfoTy(Addr, Name, SymType, false);
1500   } else {
1501     uint8_t Type =
1502         Obj.isELF() ? getElfSymbolType(Obj, Symbol) : (uint8_t)ELF::STT_NOTYPE;
1503     return SymbolInfoTy(Addr, Name, Type, IsMappingSymbol);
1504   }
1505 }
1506 
createDummySymbolInfo(const ObjectFile & Obj,const uint64_t Addr,StringRef & Name,uint8_t Type)1507 static SymbolInfoTy createDummySymbolInfo(const ObjectFile &Obj,
1508                                           const uint64_t Addr, StringRef &Name,
1509                                           uint8_t Type) {
1510   if (Obj.isXCOFF() && (SymbolDescription || TracebackTable))
1511     return SymbolInfoTy(std::nullopt, Addr, Name, std::nullopt, false);
1512   if (Obj.isWasm())
1513     return SymbolInfoTy(Addr, Name, wasm::WASM_SYMBOL_TYPE_SECTION);
1514   return SymbolInfoTy(Addr, Name, Type);
1515 }
1516 
collectBBAddrMapLabels(const BBAddrMapInfo & FullAddrMap,uint64_t SectionAddr,uint64_t Start,uint64_t End,std::unordered_map<uint64_t,std::vector<BBAddrMapLabel>> & Labels)1517 static void collectBBAddrMapLabels(
1518     const BBAddrMapInfo &FullAddrMap, uint64_t SectionAddr, uint64_t Start,
1519     uint64_t End,
1520     std::unordered_map<uint64_t, std::vector<BBAddrMapLabel>> &Labels) {
1521   if (FullAddrMap.empty())
1522     return;
1523   Labels.clear();
1524   uint64_t StartAddress = SectionAddr + Start;
1525   uint64_t EndAddress = SectionAddr + End;
1526   const BBAddrMapFunctionEntry *FunctionMap =
1527       FullAddrMap.getEntryForAddress(StartAddress);
1528   if (!FunctionMap)
1529     return;
1530   std::optional<size_t> BBRangeIndex =
1531       FunctionMap->getAddrMap().getBBRangeIndexForBaseAddress(StartAddress);
1532   if (!BBRangeIndex)
1533     return;
1534   size_t NumBBEntriesBeforeRange = 0;
1535   for (size_t I = 0; I < *BBRangeIndex; ++I)
1536     NumBBEntriesBeforeRange +=
1537         FunctionMap->getAddrMap().BBRanges[I].BBEntries.size();
1538   const auto &BBRange = FunctionMap->getAddrMap().BBRanges[*BBRangeIndex];
1539   for (size_t I = 0; I < BBRange.BBEntries.size(); ++I) {
1540     const BBAddrMap::BBEntry &BBEntry = BBRange.BBEntries[I];
1541     uint64_t BBAddress = BBEntry.Offset + BBRange.BaseAddress;
1542     if (BBAddress >= EndAddress)
1543       continue;
1544 
1545     std::string LabelString = ("BB" + Twine(BBEntry.ID)).str();
1546     Labels[BBAddress].push_back(
1547         {LabelString, FunctionMap->constructPGOLabelString(
1548                           NumBBEntriesBeforeRange + I, PrettyPGOAnalysisMap)});
1549   }
1550 }
1551 
1552 static void
collectLocalBranchTargets(ArrayRef<uint8_t> Bytes,MCInstrAnalysis * MIA,MCDisassembler * DisAsm,MCInstPrinter * IP,const MCSubtargetInfo * STI,uint64_t SectionAddr,uint64_t Start,uint64_t End,std::unordered_map<uint64_t,std::string> & Labels)1553 collectLocalBranchTargets(ArrayRef<uint8_t> Bytes, MCInstrAnalysis *MIA,
1554                           MCDisassembler *DisAsm, MCInstPrinter *IP,
1555                           const MCSubtargetInfo *STI, uint64_t SectionAddr,
1556                           uint64_t Start, uint64_t End,
1557                           std::unordered_map<uint64_t, std::string> &Labels) {
1558   // Supported by certain targets.
1559   const bool isPPC = STI->getTargetTriple().isPPC();
1560   const bool isX86 = STI->getTargetTriple().isX86();
1561   const bool isAArch64 = STI->getTargetTriple().isAArch64();
1562   const bool isBPF = STI->getTargetTriple().isBPF();
1563   if (!isPPC && !isX86 && !isAArch64 && !isBPF)
1564     return;
1565 
1566   if (MIA)
1567     MIA->resetState();
1568 
1569   std::set<uint64_t> Targets;
1570   Start += SectionAddr;
1571   End += SectionAddr;
1572   const bool isXCOFF = STI->getTargetTriple().isOSBinFormatXCOFF();
1573   for (uint64_t Index = Start; Index < End;) {
1574     // Disassemble a real instruction and record function-local branch labels.
1575     MCInst Inst;
1576     uint64_t Size;
1577     ArrayRef<uint8_t> ThisBytes = Bytes.slice(Index - SectionAddr);
1578     bool Disassembled =
1579         DisAsm->getInstruction(Inst, Size, ThisBytes, Index, nulls());
1580     if (Size == 0)
1581       Size = std::min<uint64_t>(ThisBytes.size(),
1582                                 DisAsm->suggestBytesToSkip(ThisBytes, Index));
1583 
1584     if (MIA) {
1585       if (Disassembled) {
1586         uint64_t Target;
1587         bool TargetKnown = MIA->evaluateBranch(Inst, Index, Size, Target);
1588         if (TargetKnown && (Target >= Start && Target < End) &&
1589             !Targets.count(Target)) {
1590           // On PowerPC and AIX, a function call is encoded as a branch to 0.
1591           // On other PowerPC platforms (ELF), a function call is encoded as
1592           // a branch to self. Do not add a label for these cases.
1593           if (!(isPPC &&
1594                 ((Target == 0 && isXCOFF) || (Target == Index && !isXCOFF))))
1595             Targets.insert(Target);
1596         }
1597         MIA->updateState(Inst, Index);
1598       } else
1599         MIA->resetState();
1600     }
1601     Index += Size;
1602   }
1603 
1604   Labels.clear();
1605   for (auto [Idx, Target] : enumerate(Targets))
1606     Labels[Target] = ("L" + Twine(Idx)).str();
1607 }
1608 
1609 // Create an MCSymbolizer for the target and add it to the MCDisassembler.
1610 // This is currently only used on AMDGPU, and assumes the format of the
1611 // void * argument passed to AMDGPU's createMCSymbolizer.
addSymbolizer(MCContext & Ctx,const Target * Target,StringRef TripleName,MCDisassembler * DisAsm,uint64_t SectionAddr,ArrayRef<uint8_t> Bytes,SectionSymbolsTy & Symbols,std::vector<std::unique_ptr<std::string>> & SynthesizedLabelNames)1612 static void addSymbolizer(
1613     MCContext &Ctx, const Target *Target, StringRef TripleName,
1614     MCDisassembler *DisAsm, uint64_t SectionAddr, ArrayRef<uint8_t> Bytes,
1615     SectionSymbolsTy &Symbols,
1616     std::vector<std::unique_ptr<std::string>> &SynthesizedLabelNames) {
1617 
1618   std::unique_ptr<MCRelocationInfo> RelInfo(
1619       Target->createMCRelocationInfo(TripleName, Ctx));
1620   if (!RelInfo)
1621     return;
1622   std::unique_ptr<MCSymbolizer> Symbolizer(Target->createMCSymbolizer(
1623       TripleName, nullptr, nullptr, &Symbols, &Ctx, std::move(RelInfo)));
1624   MCSymbolizer *SymbolizerPtr = &*Symbolizer;
1625   DisAsm->setSymbolizer(std::move(Symbolizer));
1626 
1627   if (!SymbolizeOperands)
1628     return;
1629 
1630   // Synthesize labels referenced by branch instructions by
1631   // disassembling, discarding the output, and collecting the referenced
1632   // addresses from the symbolizer.
1633   for (size_t Index = 0; Index != Bytes.size();) {
1634     MCInst Inst;
1635     uint64_t Size;
1636     ArrayRef<uint8_t> ThisBytes = Bytes.slice(Index);
1637     const uint64_t ThisAddr = SectionAddr + Index;
1638     DisAsm->getInstruction(Inst, Size, ThisBytes, ThisAddr, nulls());
1639     if (Size == 0)
1640       Size = std::min<uint64_t>(ThisBytes.size(),
1641                                 DisAsm->suggestBytesToSkip(ThisBytes, Index));
1642     Index += Size;
1643   }
1644   ArrayRef<uint64_t> LabelAddrsRef = SymbolizerPtr->getReferencedAddresses();
1645   // Copy and sort to remove duplicates.
1646   std::vector<uint64_t> LabelAddrs;
1647   llvm::append_range(LabelAddrs, LabelAddrsRef);
1648   llvm::sort(LabelAddrs);
1649   LabelAddrs.resize(llvm::unique(LabelAddrs) - LabelAddrs.begin());
1650   // Add the labels.
1651   for (unsigned LabelNum = 0; LabelNum != LabelAddrs.size(); ++LabelNum) {
1652     auto Name = std::make_unique<std::string>();
1653     *Name = (Twine("L") + Twine(LabelNum)).str();
1654     SynthesizedLabelNames.push_back(std::move(Name));
1655     Symbols.push_back(SymbolInfoTy(
1656         LabelAddrs[LabelNum], *SynthesizedLabelNames.back(), ELF::STT_NOTYPE));
1657   }
1658   llvm::stable_sort(Symbols);
1659   // Recreate the symbolizer with the new symbols list.
1660   RelInfo.reset(Target->createMCRelocationInfo(TripleName, Ctx));
1661   Symbolizer.reset(Target->createMCSymbolizer(
1662       TripleName, nullptr, nullptr, &Symbols, &Ctx, std::move(RelInfo)));
1663   DisAsm->setSymbolizer(std::move(Symbolizer));
1664 }
1665 
getSegmentName(const MachOObjectFile * MachO,const SectionRef & Section)1666 static StringRef getSegmentName(const MachOObjectFile *MachO,
1667                                 const SectionRef &Section) {
1668   if (MachO) {
1669     DataRefImpl DR = Section.getRawDataRefImpl();
1670     StringRef SegmentName = MachO->getSectionFinalSegmentName(DR);
1671     return SegmentName;
1672   }
1673   return "";
1674 }
1675 
createFakeELFSections(ObjectFile & Obj)1676 static void createFakeELFSections(ObjectFile &Obj) {
1677   assert(Obj.isELF());
1678   if (auto *Elf32LEObj = dyn_cast<ELF32LEObjectFile>(&Obj))
1679     Elf32LEObj->createFakeSections();
1680   else if (auto *Elf64LEObj = dyn_cast<ELF64LEObjectFile>(&Obj))
1681     Elf64LEObj->createFakeSections();
1682   else if (auto *Elf32BEObj = dyn_cast<ELF32BEObjectFile>(&Obj))
1683     Elf32BEObj->createFakeSections();
1684   else if (auto *Elf64BEObj = cast<ELF64BEObjectFile>(&Obj))
1685     Elf64BEObj->createFakeSections();
1686   else
1687     llvm_unreachable("Unsupported binary format");
1688 }
1689 
1690 // Tries to fetch a more complete version of the given object file using its
1691 // Build ID. Returns std::nullopt if nothing was found.
1692 static std::optional<OwningBinary<Binary>>
fetchBinaryByBuildID(const ObjectFile & Obj)1693 fetchBinaryByBuildID(const ObjectFile &Obj) {
1694   object::BuildIDRef BuildID = getBuildID(&Obj);
1695   if (BuildID.empty())
1696     return std::nullopt;
1697   std::optional<std::string> Path = BIDFetcher->fetch(BuildID);
1698   if (!Path)
1699     return std::nullopt;
1700   Expected<OwningBinary<Binary>> DebugBinary = createBinary(*Path);
1701   if (!DebugBinary) {
1702     reportWarning(toString(DebugBinary.takeError()), *Path);
1703     return std::nullopt;
1704   }
1705   return std::move(*DebugBinary);
1706 }
1707 
1708 static void
disassembleObject(ObjectFile & Obj,const ObjectFile & DbgObj,DisassemblerTarget & PrimaryTarget,std::optional<DisassemblerTarget> & SecondaryTarget,SourcePrinter & SP,bool InlineRelocs,raw_ostream & OS)1709 disassembleObject(ObjectFile &Obj, const ObjectFile &DbgObj,
1710                   DisassemblerTarget &PrimaryTarget,
1711                   std::optional<DisassemblerTarget> &SecondaryTarget,
1712                   SourcePrinter &SP, bool InlineRelocs, raw_ostream &OS) {
1713   DisassemblerTarget *DT = &PrimaryTarget;
1714   bool PrimaryIsThumb = false;
1715   SmallVector<std::pair<uint64_t, uint64_t>, 0> CHPECodeMap;
1716 
1717   if (SecondaryTarget) {
1718     if (isArmElf(Obj)) {
1719       PrimaryIsThumb =
1720           PrimaryTarget.SubtargetInfo->checkFeatures("+thumb-mode");
1721     } else if (const auto *COFFObj = dyn_cast<COFFObjectFile>(&Obj)) {
1722       const chpe_metadata *CHPEMetadata = COFFObj->getCHPEMetadata();
1723       if (CHPEMetadata && CHPEMetadata->CodeMapCount) {
1724         uintptr_t CodeMapInt;
1725         cantFail(COFFObj->getRvaPtr(CHPEMetadata->CodeMap, CodeMapInt));
1726         auto CodeMap = reinterpret_cast<const chpe_range_entry *>(CodeMapInt);
1727 
1728         for (uint32_t i = 0; i < CHPEMetadata->CodeMapCount; ++i) {
1729           if (CodeMap[i].getType() == chpe_range_type::Amd64 &&
1730               CodeMap[i].Length) {
1731             // Store x86_64 CHPE code ranges.
1732             uint64_t Start = CodeMap[i].getStart() + COFFObj->getImageBase();
1733             CHPECodeMap.emplace_back(Start, Start + CodeMap[i].Length);
1734           }
1735         }
1736         llvm::sort(CHPECodeMap);
1737       }
1738     }
1739   }
1740 
1741   std::map<SectionRef, std::vector<RelocationRef>> RelocMap;
1742   if (InlineRelocs || Obj.isXCOFF())
1743     RelocMap = getRelocsMap(Obj);
1744   bool Is64Bits = Obj.getBytesInAddress() > 4;
1745 
1746   // Create a mapping from virtual address to symbol name.  This is used to
1747   // pretty print the symbols while disassembling.
1748   std::map<SectionRef, SectionSymbolsTy> AllSymbols;
1749   std::map<SectionRef, SmallVector<MappingSymbolPair, 0>> AllMappingSymbols;
1750   SectionSymbolsTy AbsoluteSymbols;
1751   const StringRef FileName = Obj.getFileName();
1752   const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(&Obj);
1753   for (const SymbolRef &Symbol : Obj.symbols()) {
1754     Expected<StringRef> NameOrErr = Symbol.getName();
1755     if (!NameOrErr) {
1756       reportWarning(toString(NameOrErr.takeError()), FileName);
1757       continue;
1758     }
1759     if (NameOrErr->empty() && !(Obj.isXCOFF() && SymbolDescription))
1760       continue;
1761 
1762     if (Obj.isELF() &&
1763         (cantFail(Symbol.getFlags()) & SymbolRef::SF_FormatSpecific)) {
1764       // Symbol is intended not to be displayed by default (STT_FILE,
1765       // STT_SECTION, or a mapping symbol). Ignore STT_SECTION symbols. We will
1766       // synthesize a section symbol if no symbol is defined at offset 0.
1767       //
1768       // For a mapping symbol, store it within both AllSymbols and
1769       // AllMappingSymbols. If --show-all-symbols is unspecified, its label will
1770       // not be printed in disassembly listing.
1771       if (getElfSymbolType(Obj, Symbol) != ELF::STT_SECTION &&
1772           hasMappingSymbols(Obj)) {
1773         section_iterator SecI = unwrapOrError(Symbol.getSection(), FileName);
1774         if (SecI != Obj.section_end()) {
1775           uint64_t SectionAddr = SecI->getAddress();
1776           uint64_t Address = cantFail(Symbol.getAddress());
1777           StringRef Name = *NameOrErr;
1778           if (Name.consume_front("$") && Name.size() &&
1779               strchr("adtx", Name[0])) {
1780             AllMappingSymbols[*SecI].emplace_back(Address - SectionAddr,
1781                                                   Name[0]);
1782             AllSymbols[*SecI].push_back(
1783                 createSymbolInfo(Obj, Symbol, /*MappingSymbol=*/true));
1784           }
1785         }
1786       }
1787       continue;
1788     }
1789 
1790     if (MachO) {
1791       // __mh_(execute|dylib|dylinker|bundle|preload|object)_header are special
1792       // symbols that support MachO header introspection. They do not bind to
1793       // code locations and are irrelevant for disassembly.
1794       if (NameOrErr->starts_with("__mh_") && NameOrErr->ends_with("_header"))
1795         continue;
1796       // Don't ask a Mach-O STAB symbol for its section unless you know that
1797       // STAB symbol's section field refers to a valid section index. Otherwise
1798       // the symbol may error trying to load a section that does not exist.
1799       DataRefImpl SymDRI = Symbol.getRawDataRefImpl();
1800       uint8_t NType = (MachO->is64Bit() ?
1801                        MachO->getSymbol64TableEntry(SymDRI).n_type:
1802                        MachO->getSymbolTableEntry(SymDRI).n_type);
1803       if (NType & MachO::N_STAB)
1804         continue;
1805     }
1806 
1807     section_iterator SecI = unwrapOrError(Symbol.getSection(), FileName);
1808     if (SecI != Obj.section_end())
1809       AllSymbols[*SecI].push_back(createSymbolInfo(Obj, Symbol));
1810     else
1811       AbsoluteSymbols.push_back(createSymbolInfo(Obj, Symbol));
1812   }
1813 
1814   if (AllSymbols.empty() && Obj.isELF())
1815     addDynamicElfSymbols(cast<ELFObjectFileBase>(Obj), AllSymbols);
1816 
1817   if (Obj.isWasm())
1818     addMissingWasmCodeSymbols(cast<WasmObjectFile>(Obj), AllSymbols);
1819 
1820   if (Obj.isELF() && Obj.sections().empty())
1821     createFakeELFSections(Obj);
1822 
1823   DisassemblerTarget *PltTarget = DT;
1824   auto SectionNames = getSectionNames(Obj);
1825   if (SecondaryTarget && isArmElf(Obj)) {
1826     auto PltSectionRef = SectionNames.find(".plt");
1827     if (PltSectionRef != SectionNames.end()) {
1828       bool PltIsThumb = false;
1829       for (auto [Addr, SymbolName] : AllMappingSymbols[PltSectionRef->second]) {
1830         if (Addr != 0)
1831           continue;
1832 
1833         if (SymbolName == 't') {
1834           PltIsThumb = true;
1835           break;
1836         }
1837         if (SymbolName == 'a')
1838           break;
1839       }
1840 
1841       if (PrimaryTarget.SubtargetInfo->checkFeatures("+thumb-mode"))
1842         PltTarget = PltIsThumb ? &PrimaryTarget : &*SecondaryTarget;
1843       else
1844         PltTarget = PltIsThumb ? &*SecondaryTarget : &PrimaryTarget;
1845     }
1846   }
1847   BumpPtrAllocator A;
1848   StringSaver Saver(A);
1849   addPltEntries(*PltTarget->SubtargetInfo, Obj, SectionNames, AllSymbols,
1850                 Saver);
1851 
1852   // Create a mapping from virtual address to section. An empty section can
1853   // cause more than one section at the same address. Sort such sections to be
1854   // before same-addressed non-empty sections so that symbol lookups prefer the
1855   // non-empty section.
1856   std::vector<std::pair<uint64_t, SectionRef>> SectionAddresses;
1857   for (SectionRef Sec : Obj.sections())
1858     SectionAddresses.emplace_back(Sec.getAddress(), Sec);
1859   llvm::stable_sort(SectionAddresses, [](const auto &LHS, const auto &RHS) {
1860     if (LHS.first != RHS.first)
1861       return LHS.first < RHS.first;
1862     return LHS.second.getSize() < RHS.second.getSize();
1863   });
1864 
1865   // Linked executables (.exe and .dll files) typically don't include a real
1866   // symbol table but they might contain an export table.
1867   if (const auto *COFFObj = dyn_cast<COFFObjectFile>(&Obj)) {
1868     for (const auto &ExportEntry : COFFObj->export_directories()) {
1869       StringRef Name;
1870       if (Error E = ExportEntry.getSymbolName(Name))
1871         reportError(std::move(E), Obj.getFileName());
1872       if (Name.empty())
1873         continue;
1874 
1875       uint32_t RVA;
1876       if (Error E = ExportEntry.getExportRVA(RVA))
1877         reportError(std::move(E), Obj.getFileName());
1878 
1879       uint64_t VA = COFFObj->getImageBase() + RVA;
1880       auto Sec = partition_point(
1881           SectionAddresses, [VA](const std::pair<uint64_t, SectionRef> &O) {
1882             return O.first <= VA;
1883           });
1884       if (Sec != SectionAddresses.begin()) {
1885         --Sec;
1886         AllSymbols[Sec->second].emplace_back(VA, Name, ELF::STT_NOTYPE);
1887       } else
1888         AbsoluteSymbols.emplace_back(VA, Name, ELF::STT_NOTYPE);
1889     }
1890   }
1891 
1892   // Sort all the symbols, this allows us to use a simple binary search to find
1893   // Multiple symbols can have the same address. Use a stable sort to stabilize
1894   // the output.
1895   StringSet<> FoundDisasmSymbolSet;
1896   for (std::pair<const SectionRef, SectionSymbolsTy> &SecSyms : AllSymbols)
1897     llvm::stable_sort(SecSyms.second);
1898   llvm::stable_sort(AbsoluteSymbols);
1899 
1900   std::unique_ptr<DWARFContext> DICtx;
1901   LiveVariablePrinter LVP(*DT->Context->getRegisterInfo(), *DT->SubtargetInfo);
1902 
1903   if (DbgVariables != DVDisabled) {
1904     DICtx = DWARFContext::create(DbgObj);
1905     for (const std::unique_ptr<DWARFUnit> &CU : DICtx->compile_units())
1906       LVP.addCompileUnit(CU->getUnitDIE(false));
1907   }
1908 
1909   LLVM_DEBUG(LVP.dump());
1910 
1911   BBAddrMapInfo FullAddrMap;
1912   auto ReadBBAddrMap = [&](std::optional<unsigned> SectionIndex =
1913                                std::nullopt) {
1914     FullAddrMap.clear();
1915     if (const auto *Elf = dyn_cast<ELFObjectFileBase>(&Obj)) {
1916       std::vector<PGOAnalysisMap> PGOAnalyses;
1917       auto BBAddrMapsOrErr = Elf->readBBAddrMap(SectionIndex, &PGOAnalyses);
1918       if (!BBAddrMapsOrErr) {
1919         reportWarning(toString(BBAddrMapsOrErr.takeError()), Obj.getFileName());
1920         return;
1921       }
1922       for (auto &&[FunctionBBAddrMap, FunctionPGOAnalysis] :
1923            zip_equal(*std::move(BBAddrMapsOrErr), std::move(PGOAnalyses))) {
1924         FullAddrMap.AddFunctionEntry(std::move(FunctionBBAddrMap),
1925                                      std::move(FunctionPGOAnalysis));
1926       }
1927     }
1928   };
1929 
1930   // For non-relocatable objects, Read all LLVM_BB_ADDR_MAP sections into a
1931   // single mapping, since they don't have any conflicts.
1932   if (SymbolizeOperands && !Obj.isRelocatableObject())
1933     ReadBBAddrMap();
1934 
1935   std::optional<llvm::BTFParser> BTF;
1936   if (InlineRelocs && BTFParser::hasBTFSections(Obj)) {
1937     BTF.emplace();
1938     BTFParser::ParseOptions Opts = {};
1939     Opts.LoadTypes = true;
1940     Opts.LoadRelocs = true;
1941     if (Error E = BTF->parse(Obj, Opts))
1942       WithColor::defaultErrorHandler(std::move(E));
1943   }
1944 
1945   for (const SectionRef &Section : ToolSectionFilter(Obj)) {
1946     if (FilterSections.empty() && !DisassembleAll &&
1947         (!Section.isText() || Section.isVirtual()))
1948       continue;
1949 
1950     uint64_t SectionAddr = Section.getAddress();
1951     uint64_t SectSize = Section.getSize();
1952     if (!SectSize)
1953       continue;
1954 
1955     // For relocatable object files, read the LLVM_BB_ADDR_MAP section
1956     // corresponding to this section, if present.
1957     if (SymbolizeOperands && Obj.isRelocatableObject())
1958       ReadBBAddrMap(Section.getIndex());
1959 
1960     // Get the list of all the symbols in this section.
1961     SectionSymbolsTy &Symbols = AllSymbols[Section];
1962     auto &MappingSymbols = AllMappingSymbols[Section];
1963     llvm::sort(MappingSymbols);
1964 
1965     ArrayRef<uint8_t> Bytes = arrayRefFromStringRef(
1966         unwrapOrError(Section.getContents(), Obj.getFileName()));
1967 
1968     std::vector<std::unique_ptr<std::string>> SynthesizedLabelNames;
1969     if (Obj.isELF() && Obj.getArch() == Triple::amdgcn) {
1970       // AMDGPU disassembler uses symbolizer for printing labels
1971       addSymbolizer(*DT->Context, DT->TheTarget, TripleName, DT->DisAsm.get(),
1972                     SectionAddr, Bytes, Symbols, SynthesizedLabelNames);
1973     }
1974 
1975     StringRef SegmentName = getSegmentName(MachO, Section);
1976     StringRef SectionName = unwrapOrError(Section.getName(), Obj.getFileName());
1977     // If the section has no symbol at the start, just insert a dummy one.
1978     // Without --show-all-symbols, also insert one if all symbols at the start
1979     // are mapping symbols.
1980     bool CreateDummy = Symbols.empty();
1981     if (!CreateDummy) {
1982       CreateDummy = true;
1983       for (auto &Sym : Symbols) {
1984         if (Sym.Addr != SectionAddr)
1985           break;
1986         if (!Sym.IsMappingSymbol || ShowAllSymbols)
1987           CreateDummy = false;
1988       }
1989     }
1990     if (CreateDummy) {
1991       SymbolInfoTy Sym = createDummySymbolInfo(
1992           Obj, SectionAddr, SectionName,
1993           Section.isText() ? ELF::STT_FUNC : ELF::STT_OBJECT);
1994       if (Obj.isXCOFF())
1995         Symbols.insert(Symbols.begin(), Sym);
1996       else
1997         Symbols.insert(llvm::lower_bound(Symbols, Sym), Sym);
1998     }
1999 
2000     SmallString<40> Comments;
2001     raw_svector_ostream CommentStream(Comments);
2002 
2003     uint64_t VMAAdjustment = 0;
2004     if (shouldAdjustVA(Section))
2005       VMAAdjustment = AdjustVMA;
2006 
2007     // In executable and shared objects, r_offset holds a virtual address.
2008     // Subtract SectionAddr from the r_offset field of a relocation to get
2009     // the section offset.
2010     uint64_t RelAdjustment = Obj.isRelocatableObject() ? 0 : SectionAddr;
2011     uint64_t Size;
2012     uint64_t Index;
2013     bool PrintedSection = false;
2014     std::vector<RelocationRef> Rels = RelocMap[Section];
2015     std::vector<RelocationRef>::const_iterator RelCur = Rels.begin();
2016     std::vector<RelocationRef>::const_iterator RelEnd = Rels.end();
2017 
2018     // Loop over each chunk of code between two points where at least
2019     // one symbol is defined.
2020     for (size_t SI = 0, SE = Symbols.size(); SI != SE;) {
2021       // Advance SI past all the symbols starting at the same address,
2022       // and make an ArrayRef of them.
2023       unsigned FirstSI = SI;
2024       uint64_t Start = Symbols[SI].Addr;
2025       ArrayRef<SymbolInfoTy> SymbolsHere;
2026       while (SI != SE && Symbols[SI].Addr == Start)
2027         ++SI;
2028       SymbolsHere = ArrayRef<SymbolInfoTy>(&Symbols[FirstSI], SI - FirstSI);
2029 
2030       // Get the demangled names of all those symbols. We end up with a vector
2031       // of StringRef that holds the names we're going to use, and a vector of
2032       // std::string that stores the new strings returned by demangle(), if
2033       // any. If we don't call demangle() then that vector can stay empty.
2034       std::vector<StringRef> SymNamesHere;
2035       std::vector<std::string> DemangledSymNamesHere;
2036       if (Demangle) {
2037         // Fetch the demangled names and store them locally.
2038         for (const SymbolInfoTy &Symbol : SymbolsHere)
2039           DemangledSymNamesHere.push_back(demangle(Symbol.Name));
2040         // Now we've finished modifying that vector, it's safe to make
2041         // a vector of StringRefs pointing into it.
2042         SymNamesHere.insert(SymNamesHere.begin(), DemangledSymNamesHere.begin(),
2043                             DemangledSymNamesHere.end());
2044       } else {
2045         for (const SymbolInfoTy &Symbol : SymbolsHere)
2046           SymNamesHere.push_back(Symbol.Name);
2047       }
2048 
2049       // Distinguish ELF data from code symbols, which will be used later on to
2050       // decide whether to 'disassemble' this chunk as a data declaration via
2051       // dumpELFData(), or whether to treat it as code.
2052       //
2053       // If data _and_ code symbols are defined at the same address, the code
2054       // takes priority, on the grounds that disassembling code is our main
2055       // purpose here, and it would be a worse failure to _not_ interpret
2056       // something that _was_ meaningful as code than vice versa.
2057       //
2058       // Any ELF symbol type that is not clearly data will be regarded as code.
2059       // In particular, one of the uses of STT_NOTYPE is for branch targets
2060       // inside functions, for which STT_FUNC would be inaccurate.
2061       //
2062       // So here, we spot whether there's any non-data symbol present at all,
2063       // and only set the DisassembleAsELFData flag if there isn't. Also, we use
2064       // this distinction to inform the decision of which symbol to print at
2065       // the head of the section, so that if we're printing code, we print a
2066       // code-related symbol name to go with it.
2067       bool DisassembleAsELFData = false;
2068       size_t DisplaySymIndex = SymbolsHere.size() - 1;
2069       if (Obj.isELF() && !DisassembleAll && Section.isText()) {
2070         DisassembleAsELFData = true; // unless we find a code symbol below
2071 
2072         for (size_t i = 0; i < SymbolsHere.size(); ++i) {
2073           uint8_t SymTy = SymbolsHere[i].Type;
2074           if (SymTy != ELF::STT_OBJECT && SymTy != ELF::STT_COMMON) {
2075             DisassembleAsELFData = false;
2076             DisplaySymIndex = i;
2077           }
2078         }
2079       }
2080 
2081       // Decide which symbol(s) from this collection we're going to print.
2082       std::vector<bool> SymsToPrint(SymbolsHere.size(), false);
2083       // If the user has given the --disassemble-symbols option, then we must
2084       // display every symbol in that set, and no others.
2085       if (!DisasmSymbolSet.empty()) {
2086         bool FoundAny = false;
2087         for (size_t i = 0; i < SymbolsHere.size(); ++i) {
2088           if (DisasmSymbolSet.count(SymNamesHere[i])) {
2089             SymsToPrint[i] = true;
2090             FoundAny = true;
2091           }
2092         }
2093 
2094         // And if none of the symbols here is one that the user asked for, skip
2095         // disassembling this entire chunk of code.
2096         if (!FoundAny)
2097           continue;
2098       } else if (!SymbolsHere[DisplaySymIndex].IsMappingSymbol) {
2099         // Otherwise, print whichever symbol at this location is last in the
2100         // Symbols array, because that array is pre-sorted in a way intended to
2101         // correlate with priority of which symbol to display.
2102         SymsToPrint[DisplaySymIndex] = true;
2103       }
2104 
2105       // Now that we know we're disassembling this section, override the choice
2106       // of which symbols to display by printing _all_ of them at this address
2107       // if the user asked for all symbols.
2108       //
2109       // That way, '--show-all-symbols --disassemble-symbol=foo' will print
2110       // only the chunk of code headed by 'foo', but also show any other
2111       // symbols defined at that address, such as aliases for 'foo', or the ARM
2112       // mapping symbol preceding its code.
2113       if (ShowAllSymbols) {
2114         for (size_t i = 0; i < SymbolsHere.size(); ++i)
2115           SymsToPrint[i] = true;
2116       }
2117 
2118       if (Start < SectionAddr || StopAddress <= Start)
2119         continue;
2120 
2121       FoundDisasmSymbolSet.insert_range(SymNamesHere);
2122 
2123       // The end is the section end, the beginning of the next symbol, or
2124       // --stop-address.
2125       uint64_t End = std::min<uint64_t>(SectionAddr + SectSize, StopAddress);
2126       if (SI < SE)
2127         End = std::min(End, Symbols[SI].Addr);
2128       if (Start >= End || End <= StartAddress)
2129         continue;
2130       Start -= SectionAddr;
2131       End -= SectionAddr;
2132 
2133       if (!PrintedSection) {
2134         PrintedSection = true;
2135         OS << "\nDisassembly of section ";
2136         if (!SegmentName.empty())
2137           OS << SegmentName << ",";
2138         OS << SectionName << ":\n";
2139       }
2140 
2141       bool PrintedLabel = false;
2142       for (size_t i = 0; i < SymbolsHere.size(); ++i) {
2143         if (!SymsToPrint[i])
2144           continue;
2145 
2146         const SymbolInfoTy &Symbol = SymbolsHere[i];
2147         const StringRef SymbolName = SymNamesHere[i];
2148 
2149         if (!PrintedLabel) {
2150           OS << '\n';
2151           PrintedLabel = true;
2152         }
2153         if (LeadingAddr)
2154           OS << format(Is64Bits ? "%016" PRIx64 " " : "%08" PRIx64 " ",
2155                        SectionAddr + Start + VMAAdjustment);
2156         if (Obj.isXCOFF() && SymbolDescription) {
2157           OS << getXCOFFSymbolDescription(Symbol, SymbolName) << ":\n";
2158         } else
2159           OS << '<' << SymbolName << ">:\n";
2160       }
2161 
2162       // Don't print raw contents of a virtual section. A virtual section
2163       // doesn't have any contents in the file.
2164       if (Section.isVirtual()) {
2165         OS << "...\n";
2166         continue;
2167       }
2168 
2169       // See if any of the symbols defined at this location triggers target-
2170       // specific disassembly behavior, e.g. of special descriptors or function
2171       // prelude information.
2172       //
2173       // We stop this loop at the first symbol that triggers some kind of
2174       // interesting behavior (if any), on the assumption that if two symbols
2175       // defined at the same address trigger two conflicting symbol handlers,
2176       // the object file is probably confused anyway, and it would make even
2177       // less sense to present the output of _both_ handlers, because that
2178       // would describe the same data twice.
2179       for (size_t SHI = 0; SHI < SymbolsHere.size(); ++SHI) {
2180         SymbolInfoTy Symbol = SymbolsHere[SHI];
2181 
2182         Expected<bool> RespondedOrErr = DT->DisAsm->onSymbolStart(
2183             Symbol, Size, Bytes.slice(Start, End - Start), SectionAddr + Start);
2184 
2185         if (RespondedOrErr && !*RespondedOrErr) {
2186           // This symbol didn't trigger any interesting handling. Try the other
2187           // symbols defined at this address.
2188           continue;
2189         }
2190 
2191         // If onSymbolStart returned an Error, that means it identified some
2192         // kind of special data at this address, but wasn't able to disassemble
2193         // it meaningfully. So we fall back to printing the error out and
2194         // disassembling the failed region as bytes, assuming that the target
2195         // detected the failure before printing anything.
2196         if (!RespondedOrErr) {
2197           std::string ErrMsgStr = toString(RespondedOrErr.takeError());
2198           StringRef ErrMsg = ErrMsgStr;
2199           do {
2200             StringRef Line;
2201             std::tie(Line, ErrMsg) = ErrMsg.split('\n');
2202             OS << DT->Context->getAsmInfo()->getCommentString()
2203                << " error decoding " << SymNamesHere[SHI] << ": " << Line
2204                << '\n';
2205           } while (!ErrMsg.empty());
2206 
2207           if (Size) {
2208             OS << DT->Context->getAsmInfo()->getCommentString()
2209                << " decoding failed region as bytes\n";
2210             for (uint64_t I = 0; I < Size; ++I)
2211               OS << "\t.byte\t " << format_hex(Bytes[I], 1, /*Upper=*/true)
2212                  << '\n';
2213           }
2214         }
2215 
2216         // Regardless of whether onSymbolStart returned an Error or true, 'Size'
2217         // will have been set to the amount of data covered by whatever prologue
2218         // the target identified. So we advance our own position to beyond that.
2219         // Sometimes that will be the entire distance to the next symbol, and
2220         // sometimes it will be just a prologue and we should start
2221         // disassembling instructions from where it left off.
2222         Start += Size;
2223         break;
2224       }
2225       // Allow targets to reset any per-symbol state.
2226       DT->Printer->onSymbolStart();
2227       formatted_raw_ostream FOS(OS);
2228       Index = Start;
2229       if (SectionAddr < StartAddress)
2230         Index = std::max<uint64_t>(Index, StartAddress - SectionAddr);
2231 
2232       if (DisassembleAsELFData) {
2233         dumpELFData(SectionAddr, Index, End, Bytes, FOS);
2234         Index = End;
2235         continue;
2236       }
2237 
2238       // Skip relocations from symbols that are not dumped.
2239       for (; RelCur != RelEnd; ++RelCur) {
2240         uint64_t Offset = RelCur->getOffset() - RelAdjustment;
2241         if (Index <= Offset)
2242           break;
2243       }
2244 
2245       bool DumpARMELFData = false;
2246       bool DumpTracebackTableForXCOFFFunction =
2247           Obj.isXCOFF() && Section.isText() && TracebackTable &&
2248           Symbols[SI - 1].XCOFFSymInfo.StorageMappingClass &&
2249           (*Symbols[SI - 1].XCOFFSymInfo.StorageMappingClass == XCOFF::XMC_PR);
2250 
2251       std::unordered_map<uint64_t, std::string> AllLabels;
2252       std::unordered_map<uint64_t, std::vector<BBAddrMapLabel>> BBAddrMapLabels;
2253       if (SymbolizeOperands) {
2254         collectLocalBranchTargets(Bytes, DT->InstrAnalysis.get(),
2255                                   DT->DisAsm.get(), DT->InstPrinter.get(),
2256                                   PrimaryTarget.SubtargetInfo.get(),
2257                                   SectionAddr, Index, End, AllLabels);
2258         collectBBAddrMapLabels(FullAddrMap, SectionAddr, Index, End,
2259                                BBAddrMapLabels);
2260       }
2261 
2262       if (DT->InstrAnalysis)
2263         DT->InstrAnalysis->resetState();
2264 
2265       while (Index < End) {
2266         uint64_t RelOffset;
2267 
2268         // ARM and AArch64 ELF binaries can interleave data and text in the
2269         // same section. We rely on the markers introduced to understand what
2270         // we need to dump. If the data marker is within a function, it is
2271         // denoted as a word/short etc.
2272         if (!MappingSymbols.empty()) {
2273           char Kind = getMappingSymbolKind(MappingSymbols, Index);
2274           DumpARMELFData = Kind == 'd';
2275           if (SecondaryTarget) {
2276             if (Kind == 'a') {
2277               DT = PrimaryIsThumb ? &*SecondaryTarget : &PrimaryTarget;
2278             } else if (Kind == 't') {
2279               DT = PrimaryIsThumb ? &PrimaryTarget : &*SecondaryTarget;
2280             }
2281           }
2282         } else if (!CHPECodeMap.empty()) {
2283           uint64_t Address = SectionAddr + Index;
2284           auto It = partition_point(
2285               CHPECodeMap,
2286               [Address](const std::pair<uint64_t, uint64_t> &Entry) {
2287                 return Entry.first <= Address;
2288               });
2289           if (It != CHPECodeMap.begin() && Address < (It - 1)->second) {
2290             DT = &*SecondaryTarget;
2291           } else {
2292             DT = &PrimaryTarget;
2293             // X64 disassembler range may have left Index unaligned, so
2294             // make sure that it's aligned when we switch back to ARM64
2295             // code.
2296             Index = llvm::alignTo(Index, 4);
2297             if (Index >= End)
2298               break;
2299           }
2300         }
2301 
2302         auto findRel = [&]() {
2303           while (RelCur != RelEnd) {
2304             RelOffset = RelCur->getOffset() - RelAdjustment;
2305             // If this relocation is hidden, skip it.
2306             if (getHidden(*RelCur) || SectionAddr + RelOffset < StartAddress) {
2307               ++RelCur;
2308               continue;
2309             }
2310 
2311             // Stop when RelCur's offset is past the disassembled
2312             // instruction/data.
2313             if (RelOffset >= Index + Size)
2314               return false;
2315             if (RelOffset >= Index)
2316               return true;
2317             ++RelCur;
2318           }
2319           return false;
2320         };
2321 
2322         // When -z or --disassemble-zeroes are given we always dissasemble
2323         // them. Otherwise we might want to skip zero bytes we see.
2324         if (!DisassembleZeroes) {
2325           uint64_t MaxOffset = End - Index;
2326           // For --reloc: print zero blocks patched by relocations, so that
2327           // relocations can be shown in the dump.
2328           if (InlineRelocs && RelCur != RelEnd)
2329             MaxOffset = std::min(RelCur->getOffset() - RelAdjustment - Index,
2330                                  MaxOffset);
2331 
2332           if (size_t N =
2333                   countSkippableZeroBytes(Bytes.slice(Index, MaxOffset))) {
2334             FOS << "\t\t..." << '\n';
2335             Index += N;
2336             continue;
2337           }
2338         }
2339 
2340         if (DumpARMELFData) {
2341           Size = dumpARMELFData(SectionAddr, Index, End, Obj, Bytes,
2342                                 MappingSymbols, *DT->SubtargetInfo, FOS);
2343         } else {
2344 
2345           if (DumpTracebackTableForXCOFFFunction &&
2346               doesXCOFFTracebackTableBegin(Bytes.slice(Index, 4))) {
2347             dumpTracebackTable(Bytes.slice(Index),
2348                                SectionAddr + Index + VMAAdjustment, FOS,
2349                                SectionAddr + End + VMAAdjustment,
2350                                *DT->SubtargetInfo, cast<XCOFFObjectFile>(&Obj));
2351             Index = End;
2352             continue;
2353           }
2354 
2355           // Print local label if there's any.
2356           auto Iter1 = BBAddrMapLabels.find(SectionAddr + Index);
2357           if (Iter1 != BBAddrMapLabels.end()) {
2358             for (const auto &BBLabel : Iter1->second)
2359               FOS << "<" << BBLabel.BlockLabel << ">" << BBLabel.PGOAnalysis
2360                   << ":\n";
2361           } else {
2362             auto Iter2 = AllLabels.find(SectionAddr + Index);
2363             if (Iter2 != AllLabels.end())
2364               FOS << "<" << Iter2->second << ">:\n";
2365           }
2366 
2367           // Disassemble a real instruction or a data when disassemble all is
2368           // provided
2369           MCInst Inst;
2370           ArrayRef<uint8_t> ThisBytes = Bytes.slice(Index);
2371           uint64_t ThisAddr = SectionAddr + Index + VMAAdjustment;
2372           bool Disassembled = DT->DisAsm->getInstruction(
2373               Inst, Size, ThisBytes, ThisAddr, CommentStream);
2374           if (Size == 0)
2375             Size = std::min<uint64_t>(
2376                 ThisBytes.size(),
2377                 DT->DisAsm->suggestBytesToSkip(ThisBytes, ThisAddr));
2378 
2379           LVP.update({Index, Section.getIndex()},
2380                      {Index + Size, Section.getIndex()}, Index + Size != End);
2381 
2382           DT->InstPrinter->setCommentStream(CommentStream);
2383 
2384           DT->Printer->printInst(
2385               *DT->InstPrinter, Disassembled ? &Inst : nullptr,
2386               Bytes.slice(Index, Size),
2387               {SectionAddr + Index + VMAAdjustment, Section.getIndex()}, FOS,
2388               "", *DT->SubtargetInfo, &SP, Obj.getFileName(), &Rels, LVP);
2389 
2390           DT->InstPrinter->setCommentStream(llvm::nulls());
2391 
2392           // If disassembly succeeds, we try to resolve the target address
2393           // (jump target or memory operand address) and print it to the
2394           // right of the instruction.
2395           //
2396           // Otherwise, we don't print anything else so that we avoid
2397           // analyzing invalid or incomplete instruction information.
2398           if (Disassembled && DT->InstrAnalysis) {
2399             llvm::raw_ostream *TargetOS = &FOS;
2400             uint64_t Target;
2401             bool PrintTarget = DT->InstrAnalysis->evaluateBranch(
2402                 Inst, SectionAddr + Index, Size, Target);
2403 
2404             if (!PrintTarget) {
2405               if (std::optional<uint64_t> MaybeTarget =
2406                       DT->InstrAnalysis->evaluateMemoryOperandAddress(
2407                           Inst, DT->SubtargetInfo.get(), SectionAddr + Index,
2408                           Size)) {
2409                 Target = *MaybeTarget;
2410                 PrintTarget = true;
2411                 // Do not print real address when symbolizing.
2412                 if (!SymbolizeOperands) {
2413                   // Memory operand addresses are printed as comments.
2414                   TargetOS = &CommentStream;
2415                   *TargetOS << "0x" << Twine::utohexstr(Target);
2416                 }
2417               }
2418             }
2419 
2420             if (PrintTarget) {
2421               // In a relocatable object, the target's section must reside in
2422               // the same section as the call instruction or it is accessed
2423               // through a relocation.
2424               //
2425               // In a non-relocatable object, the target may be in any section.
2426               // In that case, locate the section(s) containing the target
2427               // address and find the symbol in one of those, if possible.
2428               //
2429               // N.B. Except for XCOFF, we don't walk the relocations in the
2430               // relocatable case yet.
2431               std::vector<const SectionSymbolsTy *> TargetSectionSymbols;
2432               if (!Obj.isRelocatableObject()) {
2433                 auto It = llvm::partition_point(
2434                     SectionAddresses,
2435                     [=](const std::pair<uint64_t, SectionRef> &O) {
2436                       return O.first <= Target;
2437                     });
2438                 uint64_t TargetSecAddr = 0;
2439                 while (It != SectionAddresses.begin()) {
2440                   --It;
2441                   if (TargetSecAddr == 0)
2442                     TargetSecAddr = It->first;
2443                   if (It->first != TargetSecAddr)
2444                     break;
2445                   TargetSectionSymbols.push_back(&AllSymbols[It->second]);
2446                 }
2447               } else {
2448                 TargetSectionSymbols.push_back(&Symbols);
2449               }
2450               TargetSectionSymbols.push_back(&AbsoluteSymbols);
2451 
2452               // Find the last symbol in the first candidate section whose
2453               // offset is less than or equal to the target. If there are no
2454               // such symbols, try in the next section and so on, before finally
2455               // using the nearest preceding absolute symbol (if any), if there
2456               // are no other valid symbols.
2457               const SymbolInfoTy *TargetSym = nullptr;
2458               for (const SectionSymbolsTy *TargetSymbols :
2459                    TargetSectionSymbols) {
2460                 auto It = llvm::partition_point(
2461                     *TargetSymbols,
2462                     [=](const SymbolInfoTy &O) { return O.Addr <= Target; });
2463                 while (It != TargetSymbols->begin()) {
2464                   --It;
2465                   // Skip mapping symbols to avoid possible ambiguity as they
2466                   // do not allow uniquely identifying the target address.
2467                   if (!It->IsMappingSymbol) {
2468                     TargetSym = &*It;
2469                     break;
2470                   }
2471                 }
2472                 if (TargetSym)
2473                   break;
2474               }
2475 
2476               // Branch targets are printed just after the instructions.
2477               // Print the labels corresponding to the target if there's any.
2478               bool BBAddrMapLabelAvailable = BBAddrMapLabels.count(Target);
2479               bool LabelAvailable = AllLabels.count(Target);
2480 
2481               if (TargetSym != nullptr) {
2482                 uint64_t TargetAddress = TargetSym->Addr;
2483                 uint64_t Disp = Target - TargetAddress;
2484                 std::string TargetName = Demangle ? demangle(TargetSym->Name)
2485                                                   : TargetSym->Name.str();
2486                 bool RelFixedUp = false;
2487                 SmallString<32> Val;
2488 
2489                 *TargetOS << " <";
2490                 // On XCOFF, we use relocations, even without -r, so we
2491                 // can print the correct name for an extern function call.
2492                 if (Obj.isXCOFF() && findRel()) {
2493                   // Check for possible branch relocations and
2494                   // branches to fixup code.
2495                   bool BranchRelocationType = true;
2496                   XCOFF::RelocationType RelocType;
2497                   if (Obj.is64Bit()) {
2498                     const XCOFFRelocation64 *Reloc =
2499                         reinterpret_cast<XCOFFRelocation64 *>(
2500                             RelCur->getRawDataRefImpl().p);
2501                     RelFixedUp = Reloc->isFixupIndicated();
2502                     RelocType = Reloc->Type;
2503                   } else {
2504                     const XCOFFRelocation32 *Reloc =
2505                         reinterpret_cast<XCOFFRelocation32 *>(
2506                             RelCur->getRawDataRefImpl().p);
2507                     RelFixedUp = Reloc->isFixupIndicated();
2508                     RelocType = Reloc->Type;
2509                   }
2510                   BranchRelocationType =
2511                       RelocType == XCOFF::R_BA || RelocType == XCOFF::R_BR ||
2512                       RelocType == XCOFF::R_RBA || RelocType == XCOFF::R_RBR;
2513 
2514                   // If we have a valid relocation, try to print its
2515                   // corresponding symbol name. Multiple relocations on the
2516                   // same instruction are not handled.
2517                   // Branches to fixup code will have the RelFixedUp flag set in
2518                   // the RLD. For these instructions, we print the correct
2519                   // branch target, but print the referenced symbol as a
2520                   // comment.
2521                   if (Error E = getRelocationValueString(*RelCur, false, Val)) {
2522                     // If -r was used, this error will be printed later.
2523                     // Otherwise, we ignore the error and print what
2524                     // would have been printed without using relocations.
2525                     consumeError(std::move(E));
2526                     *TargetOS << TargetName;
2527                     RelFixedUp = false; // Suppress comment for RLD sym name
2528                   } else if (BranchRelocationType && !RelFixedUp)
2529                     *TargetOS << Val;
2530                   else
2531                     *TargetOS << TargetName;
2532                   if (Disp)
2533                     *TargetOS << "+0x" << Twine::utohexstr(Disp);
2534                 } else if (!Disp) {
2535                   *TargetOS << TargetName;
2536                 } else if (BBAddrMapLabelAvailable) {
2537                   *TargetOS << BBAddrMapLabels[Target].front().BlockLabel;
2538                 } else if (LabelAvailable) {
2539                   *TargetOS << AllLabels[Target];
2540                 } else {
2541                   // Always Print the binary symbol plus an offset if there's no
2542                   // local label corresponding to the target address.
2543                   *TargetOS << TargetName << "+0x" << Twine::utohexstr(Disp);
2544                 }
2545                 *TargetOS << ">";
2546                 if (RelFixedUp && !InlineRelocs) {
2547                   // We have fixup code for a relocation. We print the
2548                   // referenced symbol as a comment.
2549                   *TargetOS << "\t# " << Val;
2550                 }
2551 
2552               } else if (BBAddrMapLabelAvailable) {
2553                 *TargetOS << " <" << BBAddrMapLabels[Target].front().BlockLabel
2554                           << ">";
2555               } else if (LabelAvailable) {
2556                 *TargetOS << " <" << AllLabels[Target] << ">";
2557               }
2558               // By convention, each record in the comment stream should be
2559               // terminated.
2560               if (TargetOS == &CommentStream)
2561                 *TargetOS << "\n";
2562             }
2563 
2564             DT->InstrAnalysis->updateState(Inst, SectionAddr + Index);
2565           } else if (!Disassembled && DT->InstrAnalysis) {
2566             DT->InstrAnalysis->resetState();
2567           }
2568         }
2569 
2570         assert(DT->Context->getAsmInfo());
2571         DT->Printer->emitPostInstructionInfo(FOS, *DT->Context->getAsmInfo(),
2572                                              *DT->SubtargetInfo,
2573                                              CommentStream.str(), LVP);
2574         Comments.clear();
2575 
2576         if (BTF)
2577           printBTFRelocation(FOS, *BTF, {Index, Section.getIndex()}, LVP);
2578 
2579         if (InlineRelocs) {
2580           while (findRel()) {
2581             // When --adjust-vma is used, update the address printed.
2582             printRelocation(FOS, Obj.getFileName(), *RelCur,
2583                             SectionAddr + RelOffset + VMAAdjustment, Is64Bits);
2584             LVP.printAfterOtherLine(FOS, true);
2585             ++RelCur;
2586           }
2587         }
2588 
2589         Index += Size;
2590       }
2591     }
2592   }
2593   StringSet<> MissingDisasmSymbolSet =
2594       set_difference(DisasmSymbolSet, FoundDisasmSymbolSet);
2595   for (StringRef Sym : MissingDisasmSymbolSet.keys())
2596     reportWarning("failed to disassemble missing symbol " + Sym, FileName);
2597 }
2598 
disassembleObject(ObjectFile * Obj,bool InlineRelocs,raw_ostream & OS)2599 static void disassembleObject(ObjectFile *Obj, bool InlineRelocs,
2600                               raw_ostream &OS) {
2601   // If information useful for showing the disassembly is missing, try to find a
2602   // more complete binary and disassemble that instead.
2603   OwningBinary<Binary> FetchedBinary;
2604   if (Obj->symbols().empty()) {
2605     if (std::optional<OwningBinary<Binary>> FetchedBinaryOpt =
2606             fetchBinaryByBuildID(*Obj)) {
2607       if (auto *O = dyn_cast<ObjectFile>(FetchedBinaryOpt->getBinary())) {
2608         if (!O->symbols().empty() ||
2609             (!O->sections().empty() && Obj->sections().empty())) {
2610           FetchedBinary = std::move(*FetchedBinaryOpt);
2611           Obj = O;
2612         }
2613       }
2614     }
2615   }
2616 
2617   const Target *TheTarget = getTarget(Obj);
2618 
2619   // Package up features to be passed to target/subtarget
2620   Expected<SubtargetFeatures> FeaturesValue = Obj->getFeatures();
2621   if (!FeaturesValue)
2622     reportError(FeaturesValue.takeError(), Obj->getFileName());
2623   SubtargetFeatures Features = *FeaturesValue;
2624   if (!MAttrs.empty()) {
2625     for (unsigned I = 0; I != MAttrs.size(); ++I)
2626       Features.AddFeature(MAttrs[I]);
2627   } else if (MCPU.empty() && Obj->makeTriple().isAArch64()) {
2628     Features.AddFeature("+all");
2629   }
2630 
2631   if (MCPU.empty())
2632     MCPU = Obj->tryGetCPUName().value_or("").str();
2633 
2634   if (isArmElf(*Obj)) {
2635     // When disassembling big-endian Arm ELF, the instruction endianness is
2636     // determined in a complex way. In relocatable objects, AAELF32 mandates
2637     // that instruction endianness matches the ELF file endianness; in
2638     // executable images, that's true unless the file header has the EF_ARM_BE8
2639     // flag, in which case instructions are little-endian regardless of data
2640     // endianness.
2641     //
2642     // We must set the big-endian-instructions SubtargetFeature to make the
2643     // disassembler read the instructions the right way round, and also tell
2644     // our own prettyprinter to retrieve the encodings the same way to print in
2645     // hex.
2646     const auto *Elf32BE = dyn_cast<ELF32BEObjectFile>(Obj);
2647 
2648     if (Elf32BE && (Elf32BE->isRelocatableObject() ||
2649                     !(Elf32BE->getPlatformFlags() & ELF::EF_ARM_BE8))) {
2650       Features.AddFeature("+big-endian-instructions");
2651       ARMPrettyPrinterInst.setInstructionEndianness(llvm::endianness::big);
2652     } else {
2653       ARMPrettyPrinterInst.setInstructionEndianness(llvm::endianness::little);
2654     }
2655   }
2656 
2657   DisassemblerTarget PrimaryTarget(TheTarget, *Obj, TripleName, MCPU, Features);
2658 
2659   // If we have an ARM object file, we need a second disassembler, because
2660   // ARM CPUs have two different instruction sets: ARM mode, and Thumb mode.
2661   // We use mapping symbols to switch between the two assemblers, where
2662   // appropriate.
2663   std::optional<DisassemblerTarget> SecondaryTarget;
2664 
2665   if (isArmElf(*Obj)) {
2666     if (!PrimaryTarget.SubtargetInfo->checkFeatures("+mclass")) {
2667       if (PrimaryTarget.SubtargetInfo->checkFeatures("+thumb-mode"))
2668         Features.AddFeature("-thumb-mode");
2669       else
2670         Features.AddFeature("+thumb-mode");
2671       SecondaryTarget.emplace(PrimaryTarget, Features);
2672     }
2673   } else if (const auto *COFFObj = dyn_cast<COFFObjectFile>(Obj)) {
2674     const chpe_metadata *CHPEMetadata = COFFObj->getCHPEMetadata();
2675     if (CHPEMetadata && CHPEMetadata->CodeMapCount) {
2676       // Set up x86_64 disassembler for ARM64EC binaries.
2677       Triple X64Triple(TripleName);
2678       X64Triple.setArch(Triple::ArchType::x86_64);
2679 
2680       std::string Error;
2681       const Target *X64Target =
2682           TargetRegistry::lookupTarget("", X64Triple, Error);
2683       if (X64Target) {
2684         SubtargetFeatures X64Features;
2685         SecondaryTarget.emplace(X64Target, *Obj, X64Triple.getTriple(), "",
2686                                 X64Features);
2687       } else {
2688         reportWarning(Error, Obj->getFileName());
2689       }
2690     }
2691   }
2692 
2693   const ObjectFile *DbgObj = Obj;
2694   if (!FetchedBinary.getBinary() && !Obj->hasDebugInfo()) {
2695     if (std::optional<OwningBinary<Binary>> DebugBinaryOpt =
2696             fetchBinaryByBuildID(*Obj)) {
2697       if (auto *FetchedObj =
2698               dyn_cast<const ObjectFile>(DebugBinaryOpt->getBinary())) {
2699         if (FetchedObj->hasDebugInfo()) {
2700           FetchedBinary = std::move(*DebugBinaryOpt);
2701           DbgObj = FetchedObj;
2702         }
2703       }
2704     }
2705   }
2706 
2707   std::unique_ptr<object::Binary> DSYMBinary;
2708   std::unique_ptr<MemoryBuffer> DSYMBuf;
2709   if (!DbgObj->hasDebugInfo()) {
2710     if (const MachOObjectFile *MachOOF = dyn_cast<MachOObjectFile>(&*Obj)) {
2711       DbgObj = objdump::getMachODSymObject(MachOOF, Obj->getFileName(),
2712                                            DSYMBinary, DSYMBuf);
2713       if (!DbgObj)
2714         return;
2715     }
2716   }
2717 
2718   SourcePrinter SP(DbgObj, TheTarget->getName());
2719 
2720   for (StringRef Opt : DisassemblerOptions)
2721     if (!PrimaryTarget.InstPrinter->applyTargetSpecificCLOption(Opt))
2722       reportError(Obj->getFileName(),
2723                   "Unrecognized disassembler option: " + Opt);
2724 
2725   disassembleObject(*Obj, *DbgObj, PrimaryTarget, SecondaryTarget, SP,
2726                     InlineRelocs, OS);
2727 }
2728 
printRelocations()2729 void Dumper::printRelocations() {
2730   StringRef Fmt = O.getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64;
2731 
2732   // Build a mapping from relocation target to a vector of relocation
2733   // sections. Usually, there is an only one relocation section for
2734   // each relocated section.
2735   MapVector<SectionRef, std::vector<SectionRef>> SecToRelSec;
2736   uint64_t Ndx;
2737   for (const SectionRef &Section : ToolSectionFilter(O, &Ndx)) {
2738     if (O.isELF() && (ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC))
2739       continue;
2740     if (Section.relocation_begin() == Section.relocation_end())
2741       continue;
2742     Expected<section_iterator> SecOrErr = Section.getRelocatedSection();
2743     if (!SecOrErr)
2744       reportError(O.getFileName(),
2745                   "section (" + Twine(Ndx) +
2746                       "): unable to get a relocation target: " +
2747                       toString(SecOrErr.takeError()));
2748     SecToRelSec[**SecOrErr].push_back(Section);
2749   }
2750 
2751   for (std::pair<SectionRef, std::vector<SectionRef>> &P : SecToRelSec) {
2752     StringRef SecName = unwrapOrError(P.first.getName(), O.getFileName());
2753     outs() << "\nRELOCATION RECORDS FOR [" << SecName << "]:\n";
2754     uint32_t OffsetPadding = (O.getBytesInAddress() > 4 ? 16 : 8);
2755     uint32_t TypePadding = 24;
2756     outs() << left_justify("OFFSET", OffsetPadding) << " "
2757            << left_justify("TYPE", TypePadding) << " "
2758            << "VALUE\n";
2759 
2760     for (SectionRef Section : P.second) {
2761       // CREL sections require decoding, each section may have its own specific
2762       // decode problems.
2763       if (O.isELF() && ELFSectionRef(Section).getType() == ELF::SHT_CREL) {
2764         StringRef Err =
2765             cast<const ELFObjectFileBase>(O).getCrelDecodeProblem(Section);
2766         if (!Err.empty()) {
2767           reportUniqueWarning(Err);
2768           continue;
2769         }
2770       }
2771       for (const RelocationRef &Reloc : Section.relocations()) {
2772         uint64_t Address = Reloc.getOffset();
2773         SmallString<32> RelocName;
2774         SmallString<32> ValueStr;
2775         if (Address < StartAddress || Address > StopAddress || getHidden(Reloc))
2776           continue;
2777         Reloc.getTypeName(RelocName);
2778         if (Error E =
2779                 getRelocationValueString(Reloc, SymbolDescription, ValueStr))
2780           reportUniqueWarning(std::move(E));
2781 
2782         outs() << format(Fmt.data(), Address) << " "
2783                << left_justify(RelocName, TypePadding) << " " << ValueStr
2784                << "\n";
2785       }
2786     }
2787   }
2788 }
2789 
2790 // Returns true if we need to show LMA column when dumping section headers. We
2791 // show it only when the platform is ELF and either we have at least one section
2792 // whose VMA and LMA are different and/or when --show-lma flag is used.
shouldDisplayLMA(const ObjectFile & Obj)2793 static bool shouldDisplayLMA(const ObjectFile &Obj) {
2794   if (!Obj.isELF())
2795     return false;
2796   for (const SectionRef &S : ToolSectionFilter(Obj))
2797     if (S.getAddress() != getELFSectionLMA(S))
2798       return true;
2799   return ShowLMA;
2800 }
2801 
getMaxSectionNameWidth(const ObjectFile & Obj)2802 static size_t getMaxSectionNameWidth(const ObjectFile &Obj) {
2803   // Default column width for names is 13 even if no names are that long.
2804   size_t MaxWidth = 13;
2805   for (const SectionRef &Section : ToolSectionFilter(Obj)) {
2806     StringRef Name = unwrapOrError(Section.getName(), Obj.getFileName());
2807     MaxWidth = std::max(MaxWidth, Name.size());
2808   }
2809   return MaxWidth;
2810 }
2811 
printSectionHeaders(ObjectFile & Obj)2812 void objdump::printSectionHeaders(ObjectFile &Obj) {
2813   if (Obj.isELF() && Obj.sections().empty())
2814     createFakeELFSections(Obj);
2815 
2816   size_t NameWidth = getMaxSectionNameWidth(Obj);
2817   size_t AddressWidth = 2 * Obj.getBytesInAddress();
2818   bool HasLMAColumn = shouldDisplayLMA(Obj);
2819   outs() << "\nSections:\n";
2820   if (HasLMAColumn)
2821     outs() << "Idx " << left_justify("Name", NameWidth) << " Size     "
2822            << left_justify("VMA", AddressWidth) << " "
2823            << left_justify("LMA", AddressWidth) << " Type\n";
2824   else
2825     outs() << "Idx " << left_justify("Name", NameWidth) << " Size     "
2826            << left_justify("VMA", AddressWidth) << " Type\n";
2827 
2828   uint64_t Idx;
2829   for (const SectionRef &Section : ToolSectionFilter(Obj, &Idx)) {
2830     StringRef Name = unwrapOrError(Section.getName(), Obj.getFileName());
2831     uint64_t VMA = Section.getAddress();
2832     if (shouldAdjustVA(Section))
2833       VMA += AdjustVMA;
2834 
2835     uint64_t Size = Section.getSize();
2836 
2837     std::string Type = Section.isText() ? "TEXT" : "";
2838     if (Section.isData())
2839       Type += Type.empty() ? "DATA" : ", DATA";
2840     if (Section.isBSS())
2841       Type += Type.empty() ? "BSS" : ", BSS";
2842     if (Section.isDebugSection())
2843       Type += Type.empty() ? "DEBUG" : ", DEBUG";
2844 
2845     if (HasLMAColumn)
2846       outs() << format("%3" PRIu64 " %-*s %08" PRIx64 " ", Idx, NameWidth,
2847                        Name.str().c_str(), Size)
2848              << format_hex_no_prefix(VMA, AddressWidth) << " "
2849              << format_hex_no_prefix(getELFSectionLMA(Section), AddressWidth)
2850              << " " << Type << "\n";
2851     else
2852       outs() << format("%3" PRIu64 " %-*s %08" PRIx64 " ", Idx, NameWidth,
2853                        Name.str().c_str(), Size)
2854              << format_hex_no_prefix(VMA, AddressWidth) << " " << Type << "\n";
2855   }
2856 }
2857 
printSectionContents(const ObjectFile * Obj)2858 void objdump::printSectionContents(const ObjectFile *Obj) {
2859   const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(Obj);
2860 
2861   for (const SectionRef &Section : ToolSectionFilter(*Obj)) {
2862     StringRef Name = unwrapOrError(Section.getName(), Obj->getFileName());
2863     uint64_t BaseAddr = Section.getAddress();
2864     uint64_t Size = Section.getSize();
2865     if (!Size)
2866       continue;
2867 
2868     outs() << "Contents of section ";
2869     StringRef SegmentName = getSegmentName(MachO, Section);
2870     if (!SegmentName.empty())
2871       outs() << SegmentName << ",";
2872     outs() << Name << ":\n";
2873     if (Section.isBSS()) {
2874       outs() << format("<skipping contents of bss section at [%04" PRIx64
2875                        ", %04" PRIx64 ")>\n",
2876                        BaseAddr, BaseAddr + Size);
2877       continue;
2878     }
2879 
2880     StringRef Contents = unwrapOrError(Section.getContents(), Obj->getFileName());
2881 
2882     // Dump out the content as hex and printable ascii characters.
2883     for (std::size_t Addr = 0, End = Contents.size(); Addr < End; Addr += 16) {
2884       outs() << format(" %04" PRIx64 " ", BaseAddr + Addr);
2885       // Dump line of hex.
2886       for (std::size_t I = 0; I < 16; ++I) {
2887         if (I != 0 && I % 4 == 0)
2888           outs() << ' ';
2889         if (Addr + I < End)
2890           outs() << hexdigit((Contents[Addr + I] >> 4) & 0xF, true)
2891                  << hexdigit(Contents[Addr + I] & 0xF, true);
2892         else
2893           outs() << "  ";
2894       }
2895       // Print ascii.
2896       outs() << "  ";
2897       for (std::size_t I = 0; I < 16 && Addr + I < End; ++I) {
2898         if (isPrint(static_cast<unsigned char>(Contents[Addr + I]) & 0xFF))
2899           outs() << Contents[Addr + I];
2900         else
2901           outs() << ".";
2902       }
2903       outs() << "\n";
2904     }
2905   }
2906 }
2907 
printSymbolTable(StringRef ArchiveName,StringRef ArchitectureName,bool DumpDynamic)2908 void Dumper::printSymbolTable(StringRef ArchiveName, StringRef ArchitectureName,
2909                               bool DumpDynamic) {
2910   if (O.isCOFF() && !DumpDynamic) {
2911     outs() << "\nSYMBOL TABLE:\n";
2912     printCOFFSymbolTable(cast<const COFFObjectFile>(O));
2913     return;
2914   }
2915 
2916   const StringRef FileName = O.getFileName();
2917 
2918   if (!DumpDynamic) {
2919     outs() << "\nSYMBOL TABLE:\n";
2920     for (auto I = O.symbol_begin(); I != O.symbol_end(); ++I)
2921       printSymbol(*I, {}, FileName, ArchiveName, ArchitectureName, DumpDynamic);
2922     return;
2923   }
2924 
2925   outs() << "\nDYNAMIC SYMBOL TABLE:\n";
2926   if (!O.isELF()) {
2927     reportWarning(
2928         "this operation is not currently supported for this file format",
2929         FileName);
2930     return;
2931   }
2932 
2933   const ELFObjectFileBase *ELF = cast<const ELFObjectFileBase>(&O);
2934   auto Symbols = ELF->getDynamicSymbolIterators();
2935   Expected<std::vector<VersionEntry>> SymbolVersionsOrErr =
2936       ELF->readDynsymVersions();
2937   if (!SymbolVersionsOrErr) {
2938     reportWarning(toString(SymbolVersionsOrErr.takeError()), FileName);
2939     SymbolVersionsOrErr = std::vector<VersionEntry>();
2940     (void)!SymbolVersionsOrErr;
2941   }
2942   for (auto &Sym : Symbols)
2943     printSymbol(Sym, *SymbolVersionsOrErr, FileName, ArchiveName,
2944                 ArchitectureName, DumpDynamic);
2945 }
2946 
printSymbol(const SymbolRef & Symbol,ArrayRef<VersionEntry> SymbolVersions,StringRef FileName,StringRef ArchiveName,StringRef ArchitectureName,bool DumpDynamic)2947 void Dumper::printSymbol(const SymbolRef &Symbol,
2948                          ArrayRef<VersionEntry> SymbolVersions,
2949                          StringRef FileName, StringRef ArchiveName,
2950                          StringRef ArchitectureName, bool DumpDynamic) {
2951   const MachOObjectFile *MachO = dyn_cast<const MachOObjectFile>(&O);
2952   Expected<uint64_t> AddrOrErr = Symbol.getAddress();
2953   if (!AddrOrErr) {
2954     reportUniqueWarning(AddrOrErr.takeError());
2955     return;
2956   }
2957 
2958   // Don't ask a Mach-O STAB symbol for its section unless you know that
2959   // STAB symbol's section field refers to a valid section index. Otherwise
2960   // the symbol may error trying to load a section that does not exist.
2961   bool IsSTAB = false;
2962   if (MachO) {
2963     DataRefImpl SymDRI = Symbol.getRawDataRefImpl();
2964     uint8_t NType =
2965         (MachO->is64Bit() ? MachO->getSymbol64TableEntry(SymDRI).n_type
2966                           : MachO->getSymbolTableEntry(SymDRI).n_type);
2967     if (NType & MachO::N_STAB)
2968       IsSTAB = true;
2969   }
2970   section_iterator Section = IsSTAB
2971                                  ? O.section_end()
2972                                  : unwrapOrError(Symbol.getSection(), FileName,
2973                                                  ArchiveName, ArchitectureName);
2974 
2975   uint64_t Address = *AddrOrErr;
2976   if (Section != O.section_end() && shouldAdjustVA(*Section))
2977     Address += AdjustVMA;
2978   if ((Address < StartAddress) || (Address > StopAddress))
2979     return;
2980   SymbolRef::Type Type =
2981       unwrapOrError(Symbol.getType(), FileName, ArchiveName, ArchitectureName);
2982   uint32_t Flags =
2983       unwrapOrError(Symbol.getFlags(), FileName, ArchiveName, ArchitectureName);
2984 
2985   StringRef Name;
2986   if (Type == SymbolRef::ST_Debug && Section != O.section_end()) {
2987     if (Expected<StringRef> NameOrErr = Section->getName())
2988       Name = *NameOrErr;
2989     else
2990       consumeError(NameOrErr.takeError());
2991 
2992   } else {
2993     Name = unwrapOrError(Symbol.getName(), FileName, ArchiveName,
2994                          ArchitectureName);
2995   }
2996 
2997   bool Global = Flags & SymbolRef::SF_Global;
2998   bool Weak = Flags & SymbolRef::SF_Weak;
2999   bool Absolute = Flags & SymbolRef::SF_Absolute;
3000   bool Common = Flags & SymbolRef::SF_Common;
3001   bool Hidden = Flags & SymbolRef::SF_Hidden;
3002 
3003   char GlobLoc = ' ';
3004   if ((Section != O.section_end() || Absolute) && !Weak)
3005     GlobLoc = Global ? 'g' : 'l';
3006   char IFunc = ' ';
3007   if (O.isELF()) {
3008     if (ELFSymbolRef(Symbol).getELFType() == ELF::STT_GNU_IFUNC)
3009       IFunc = 'i';
3010     if (ELFSymbolRef(Symbol).getBinding() == ELF::STB_GNU_UNIQUE)
3011       GlobLoc = 'u';
3012   }
3013 
3014   char Debug = ' ';
3015   if (DumpDynamic)
3016     Debug = 'D';
3017   else if (Type == SymbolRef::ST_Debug || Type == SymbolRef::ST_File)
3018     Debug = 'd';
3019 
3020   char FileFunc = ' ';
3021   if (Type == SymbolRef::ST_File)
3022     FileFunc = 'f';
3023   else if (Type == SymbolRef::ST_Function)
3024     FileFunc = 'F';
3025   else if (Type == SymbolRef::ST_Data)
3026     FileFunc = 'O';
3027 
3028   const char *Fmt = O.getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64;
3029 
3030   outs() << format(Fmt, Address) << " "
3031          << GlobLoc            // Local -> 'l', Global -> 'g', Neither -> ' '
3032          << (Weak ? 'w' : ' ') // Weak?
3033          << ' '                // Constructor. Not supported yet.
3034          << ' '                // Warning. Not supported yet.
3035          << IFunc              // Indirect reference to another symbol.
3036          << Debug              // Debugging (d) or dynamic (D) symbol.
3037          << FileFunc           // Name of function (F), file (f) or object (O).
3038          << ' ';
3039   if (Absolute) {
3040     outs() << "*ABS*";
3041   } else if (Common) {
3042     outs() << "*COM*";
3043   } else if (Section == O.section_end()) {
3044     if (O.isXCOFF()) {
3045       XCOFFSymbolRef XCOFFSym = cast<const XCOFFObjectFile>(O).toSymbolRef(
3046           Symbol.getRawDataRefImpl());
3047       if (XCOFF::N_DEBUG == XCOFFSym.getSectionNumber())
3048         outs() << "*DEBUG*";
3049       else
3050         outs() << "*UND*";
3051     } else
3052       outs() << "*UND*";
3053   } else {
3054     StringRef SegmentName = getSegmentName(MachO, *Section);
3055     if (!SegmentName.empty())
3056       outs() << SegmentName << ",";
3057     StringRef SectionName = unwrapOrError(Section->getName(), FileName);
3058     outs() << SectionName;
3059     if (O.isXCOFF()) {
3060       std::optional<SymbolRef> SymRef =
3061           getXCOFFSymbolContainingSymbolRef(cast<XCOFFObjectFile>(O), Symbol);
3062       if (SymRef) {
3063 
3064         Expected<StringRef> NameOrErr = SymRef->getName();
3065 
3066         if (NameOrErr) {
3067           outs() << " (csect:";
3068           std::string SymName =
3069               Demangle ? demangle(*NameOrErr) : NameOrErr->str();
3070 
3071           if (SymbolDescription)
3072             SymName = getXCOFFSymbolDescription(createSymbolInfo(O, *SymRef),
3073                                                 SymName);
3074 
3075           outs() << ' ' << SymName;
3076           outs() << ") ";
3077         } else
3078           reportWarning(toString(NameOrErr.takeError()), FileName);
3079       }
3080     }
3081   }
3082 
3083   if (Common)
3084     outs() << '\t' << format(Fmt, static_cast<uint64_t>(Symbol.getAlignment()));
3085   else if (O.isXCOFF())
3086     outs() << '\t'
3087            << format(Fmt, cast<XCOFFObjectFile>(O).getSymbolSize(
3088                               Symbol.getRawDataRefImpl()));
3089   else if (O.isELF())
3090     outs() << '\t' << format(Fmt, ELFSymbolRef(Symbol).getSize());
3091   else if (O.isWasm())
3092     outs() << '\t'
3093            << format(Fmt, static_cast<uint64_t>(
3094                               cast<WasmObjectFile>(O).getSymbolSize(Symbol)));
3095 
3096   if (O.isELF()) {
3097     if (!SymbolVersions.empty()) {
3098       const VersionEntry &Ver =
3099           SymbolVersions[Symbol.getRawDataRefImpl().d.b - 1];
3100       std::string Str;
3101       if (!Ver.Name.empty())
3102         Str = Ver.IsVerDef ? ' ' + Ver.Name : '(' + Ver.Name + ')';
3103       outs() << ' ' << left_justify(Str, 12);
3104     }
3105 
3106     uint8_t Other = ELFSymbolRef(Symbol).getOther();
3107     switch (Other) {
3108     case ELF::STV_DEFAULT:
3109       break;
3110     case ELF::STV_INTERNAL:
3111       outs() << " .internal";
3112       break;
3113     case ELF::STV_HIDDEN:
3114       outs() << " .hidden";
3115       break;
3116     case ELF::STV_PROTECTED:
3117       outs() << " .protected";
3118       break;
3119     default:
3120       outs() << format(" 0x%02x", Other);
3121       break;
3122     }
3123   } else if (Hidden) {
3124     outs() << " .hidden";
3125   }
3126 
3127   std::string SymName = Demangle ? demangle(Name) : Name.str();
3128   if (O.isXCOFF() && SymbolDescription)
3129     SymName = getXCOFFSymbolDescription(createSymbolInfo(O, Symbol), SymName);
3130 
3131   outs() << ' ' << SymName << '\n';
3132 }
3133 
printUnwindInfo(const ObjectFile * O)3134 static void printUnwindInfo(const ObjectFile *O) {
3135   outs() << "Unwind info:\n\n";
3136 
3137   if (const COFFObjectFile *Coff = dyn_cast<COFFObjectFile>(O))
3138     printCOFFUnwindInfo(Coff);
3139   else if (const MachOObjectFile *MachO = dyn_cast<MachOObjectFile>(O))
3140     printMachOUnwindInfo(MachO);
3141   else
3142     // TODO: Extract DWARF dump tool to objdump.
3143     WithColor::error(errs(), ToolName)
3144         << "This operation is only currently supported "
3145            "for COFF and MachO object files.\n";
3146 }
3147 
3148 /// Dump the raw contents of the __clangast section so the output can be piped
3149 /// into llvm-bcanalyzer.
printRawClangAST(const ObjectFile * Obj)3150 static void printRawClangAST(const ObjectFile *Obj) {
3151   if (outs().is_displayed()) {
3152     WithColor::error(errs(), ToolName)
3153         << "The -raw-clang-ast option will dump the raw binary contents of "
3154            "the clang ast section.\n"
3155            "Please redirect the output to a file or another program such as "
3156            "llvm-bcanalyzer.\n";
3157     return;
3158   }
3159 
3160   StringRef ClangASTSectionName("__clangast");
3161   if (Obj->isCOFF()) {
3162     ClangASTSectionName = "clangast";
3163   }
3164 
3165   std::optional<object::SectionRef> ClangASTSection;
3166   for (auto Sec : ToolSectionFilter(*Obj)) {
3167     StringRef Name;
3168     if (Expected<StringRef> NameOrErr = Sec.getName())
3169       Name = *NameOrErr;
3170     else
3171       consumeError(NameOrErr.takeError());
3172 
3173     if (Name == ClangASTSectionName) {
3174       ClangASTSection = Sec;
3175       break;
3176     }
3177   }
3178   if (!ClangASTSection)
3179     return;
3180 
3181   StringRef ClangASTContents =
3182       unwrapOrError(ClangASTSection->getContents(), Obj->getFileName());
3183   outs().write(ClangASTContents.data(), ClangASTContents.size());
3184 }
3185 
printFaultMaps(const ObjectFile * Obj)3186 static void printFaultMaps(const ObjectFile *Obj) {
3187   StringRef FaultMapSectionName;
3188 
3189   if (Obj->isELF()) {
3190     FaultMapSectionName = ".llvm_faultmaps";
3191   } else if (Obj->isMachO()) {
3192     FaultMapSectionName = "__llvm_faultmaps";
3193   } else {
3194     WithColor::error(errs(), ToolName)
3195         << "This operation is only currently supported "
3196            "for ELF and Mach-O executable files.\n";
3197     return;
3198   }
3199 
3200   std::optional<object::SectionRef> FaultMapSection;
3201 
3202   for (auto Sec : ToolSectionFilter(*Obj)) {
3203     StringRef Name;
3204     if (Expected<StringRef> NameOrErr = Sec.getName())
3205       Name = *NameOrErr;
3206     else
3207       consumeError(NameOrErr.takeError());
3208 
3209     if (Name == FaultMapSectionName) {
3210       FaultMapSection = Sec;
3211       break;
3212     }
3213   }
3214 
3215   outs() << "FaultMap table:\n";
3216 
3217   if (!FaultMapSection) {
3218     outs() << "<not found>\n";
3219     return;
3220   }
3221 
3222   StringRef FaultMapContents =
3223       unwrapOrError(FaultMapSection->getContents(), Obj->getFileName());
3224   FaultMapParser FMP(FaultMapContents.bytes_begin(),
3225                      FaultMapContents.bytes_end());
3226 
3227   outs() << FMP;
3228 }
3229 
printPrivateHeaders()3230 void Dumper::printPrivateHeaders() {
3231   reportError(O.getFileName(), "Invalid/Unsupported object file format");
3232 }
3233 
printFileHeaders(const ObjectFile * O)3234 static void printFileHeaders(const ObjectFile *O) {
3235   if (!O->isELF() && !O->isCOFF() && !O->isXCOFF())
3236     reportError(O->getFileName(), "Invalid/Unsupported object file format");
3237 
3238   Triple::ArchType AT = O->getArch();
3239   outs() << "architecture: " << Triple::getArchTypeName(AT) << "\n";
3240   uint64_t Address = unwrapOrError(O->getStartAddress(), O->getFileName());
3241 
3242   StringRef Fmt = O->getBytesInAddress() > 4 ? "%016" PRIx64 : "%08" PRIx64;
3243   outs() << "start address: "
3244          << "0x" << format(Fmt.data(), Address) << "\n";
3245 }
3246 
printArchiveChild(StringRef Filename,const Archive::Child & C)3247 static void printArchiveChild(StringRef Filename, const Archive::Child &C) {
3248   Expected<sys::fs::perms> ModeOrErr = C.getAccessMode();
3249   if (!ModeOrErr) {
3250     WithColor::error(errs(), ToolName) << "ill-formed archive entry.\n";
3251     consumeError(ModeOrErr.takeError());
3252     return;
3253   }
3254   sys::fs::perms Mode = ModeOrErr.get();
3255   outs() << ((Mode & sys::fs::owner_read) ? "r" : "-");
3256   outs() << ((Mode & sys::fs::owner_write) ? "w" : "-");
3257   outs() << ((Mode & sys::fs::owner_exe) ? "x" : "-");
3258   outs() << ((Mode & sys::fs::group_read) ? "r" : "-");
3259   outs() << ((Mode & sys::fs::group_write) ? "w" : "-");
3260   outs() << ((Mode & sys::fs::group_exe) ? "x" : "-");
3261   outs() << ((Mode & sys::fs::others_read) ? "r" : "-");
3262   outs() << ((Mode & sys::fs::others_write) ? "w" : "-");
3263   outs() << ((Mode & sys::fs::others_exe) ? "x" : "-");
3264 
3265   outs() << " ";
3266 
3267   outs() << format("%d/%d %6" PRId64 " ", unwrapOrError(C.getUID(), Filename),
3268                    unwrapOrError(C.getGID(), Filename),
3269                    unwrapOrError(C.getRawSize(), Filename));
3270 
3271   StringRef RawLastModified = C.getRawLastModified();
3272   unsigned Seconds;
3273   if (RawLastModified.getAsInteger(10, Seconds))
3274     outs() << "(date: \"" << RawLastModified
3275            << "\" contains non-decimal chars) ";
3276   else {
3277     // Since ctime(3) returns a 26 character string of the form:
3278     // "Sun Sep 16 01:03:52 1973\n\0"
3279     // just print 24 characters.
3280     time_t t = Seconds;
3281     outs() << format("%.24s ", ctime(&t));
3282   }
3283 
3284   StringRef Name = "";
3285   Expected<StringRef> NameOrErr = C.getName();
3286   if (!NameOrErr) {
3287     consumeError(NameOrErr.takeError());
3288     Name = unwrapOrError(C.getRawName(), Filename);
3289   } else {
3290     Name = NameOrErr.get();
3291   }
3292   outs() << Name << "\n";
3293 }
3294 
3295 // For ELF only now.
shouldWarnForInvalidStartStopAddress(ObjectFile * Obj)3296 static bool shouldWarnForInvalidStartStopAddress(ObjectFile *Obj) {
3297   if (const auto *Elf = dyn_cast<ELFObjectFileBase>(Obj)) {
3298     if (Elf->getEType() != ELF::ET_REL)
3299       return true;
3300   }
3301   return false;
3302 }
3303 
checkForInvalidStartStopAddress(ObjectFile * Obj,uint64_t Start,uint64_t Stop)3304 static void checkForInvalidStartStopAddress(ObjectFile *Obj,
3305                                             uint64_t Start, uint64_t Stop) {
3306   if (!shouldWarnForInvalidStartStopAddress(Obj))
3307     return;
3308 
3309   for (const SectionRef &Section : Obj->sections())
3310     if (ELFSectionRef(Section).getFlags() & ELF::SHF_ALLOC) {
3311       uint64_t BaseAddr = Section.getAddress();
3312       uint64_t Size = Section.getSize();
3313       if ((Start < BaseAddr + Size) && Stop > BaseAddr)
3314         return;
3315     }
3316 
3317   if (!HasStartAddressFlag)
3318     reportWarning("no section has address less than 0x" +
3319                       Twine::utohexstr(Stop) + " specified by --stop-address",
3320                   Obj->getFileName());
3321   else if (!HasStopAddressFlag)
3322     reportWarning("no section has address greater than or equal to 0x" +
3323                       Twine::utohexstr(Start) + " specified by --start-address",
3324                   Obj->getFileName());
3325   else
3326     reportWarning("no section overlaps the range [0x" +
3327                       Twine::utohexstr(Start) + ",0x" + Twine::utohexstr(Stop) +
3328                       ") specified by --start-address/--stop-address",
3329                   Obj->getFileName());
3330 }
3331 
dumpObject(ObjectFile * O,const Archive * A=nullptr,const Archive::Child * C=nullptr)3332 static void dumpObject(ObjectFile *O, const Archive *A = nullptr,
3333                        const Archive::Child *C = nullptr) {
3334   Expected<std::unique_ptr<Dumper>> DumperOrErr = createDumper(*O);
3335   if (!DumperOrErr) {
3336     reportError(DumperOrErr.takeError(), O->getFileName(),
3337                 A ? A->getFileName() : "");
3338     return;
3339   }
3340   Dumper &D = **DumperOrErr;
3341 
3342   // Avoid other output when using a raw option.
3343   if (!RawClangAST) {
3344     outs() << '\n';
3345     if (A)
3346       outs() << A->getFileName() << "(" << O->getFileName() << ")";
3347     else
3348       outs() << O->getFileName();
3349     outs() << ":\tfile format " << O->getFileFormatName().lower() << "\n";
3350   }
3351 
3352   if (HasStartAddressFlag || HasStopAddressFlag)
3353     checkForInvalidStartStopAddress(O, StartAddress, StopAddress);
3354 
3355   // TODO: Change print* free functions to Dumper member functions to utilitize
3356   // stateful functions like reportUniqueWarning.
3357 
3358   // Note: the order here matches GNU objdump for compatability.
3359   StringRef ArchiveName = A ? A->getFileName() : "";
3360   if (ArchiveHeaders && !MachOOpt && C)
3361     printArchiveChild(ArchiveName, *C);
3362   if (FileHeaders)
3363     printFileHeaders(O);
3364   if (PrivateHeaders || FirstPrivateHeader)
3365     D.printPrivateHeaders();
3366   if (SectionHeaders)
3367     printSectionHeaders(*O);
3368   if (SymbolTable)
3369     D.printSymbolTable(ArchiveName);
3370   if (DynamicSymbolTable)
3371     D.printSymbolTable(ArchiveName, /*ArchitectureName=*/"",
3372                        /*DumpDynamic=*/true);
3373   if (DwarfDumpType != DIDT_Null) {
3374     std::unique_ptr<DIContext> DICtx = DWARFContext::create(*O);
3375     // Dump the complete DWARF structure.
3376     DIDumpOptions DumpOpts;
3377     DumpOpts.DumpType = DwarfDumpType;
3378     DICtx->dump(outs(), DumpOpts);
3379   }
3380   if (Relocations && !Disassemble)
3381     D.printRelocations();
3382   if (DynamicRelocations)
3383     D.printDynamicRelocations();
3384   if (SectionContents)
3385     printSectionContents(O);
3386   if (Disassemble)
3387     disassembleObject(O, Relocations, outs());
3388   if (UnwindInfo)
3389     printUnwindInfo(O);
3390 
3391   // Mach-O specific options:
3392   if (ExportsTrie)
3393     printExportsTrie(O);
3394   if (Rebase)
3395     printRebaseTable(O);
3396   if (Bind)
3397     printBindTable(O);
3398   if (LazyBind)
3399     printLazyBindTable(O);
3400   if (WeakBind)
3401     printWeakBindTable(O);
3402 
3403   // Other special sections:
3404   if (RawClangAST)
3405     printRawClangAST(O);
3406   if (FaultMapSection)
3407     printFaultMaps(O);
3408   if (Offloading)
3409     dumpOffloadBinary(*O, StringRef(ArchName));
3410 }
3411 
dumpObject(const COFFImportFile * I,const Archive * A,const Archive::Child * C=nullptr)3412 static void dumpObject(const COFFImportFile *I, const Archive *A,
3413                        const Archive::Child *C = nullptr) {
3414   StringRef ArchiveName = A ? A->getFileName() : "";
3415 
3416   // Avoid other output when using a raw option.
3417   if (!RawClangAST)
3418     outs() << '\n'
3419            << ArchiveName << "(" << I->getFileName() << ")"
3420            << ":\tfile format COFF-import-file"
3421            << "\n\n";
3422 
3423   if (ArchiveHeaders && !MachOOpt && C)
3424     printArchiveChild(ArchiveName, *C);
3425   if (SymbolTable)
3426     printCOFFSymbolTable(*I);
3427 }
3428 
3429 /// Dump each object file in \a a;
dumpArchive(const Archive * A)3430 static void dumpArchive(const Archive *A) {
3431   Error Err = Error::success();
3432   unsigned I = -1;
3433   for (auto &C : A->children(Err)) {
3434     ++I;
3435     Expected<std::unique_ptr<Binary>> ChildOrErr = C.getAsBinary();
3436     if (!ChildOrErr) {
3437       if (auto E = isNotObjectErrorInvalidFileType(ChildOrErr.takeError()))
3438         reportError(std::move(E), getFileNameForError(C, I), A->getFileName());
3439       continue;
3440     }
3441     if (ObjectFile *O = dyn_cast<ObjectFile>(&*ChildOrErr.get()))
3442       dumpObject(O, A, &C);
3443     else if (COFFImportFile *I = dyn_cast<COFFImportFile>(&*ChildOrErr.get()))
3444       dumpObject(I, A, &C);
3445     else
3446       reportError(errorCodeToError(object_error::invalid_file_type),
3447                   A->getFileName());
3448   }
3449   if (Err)
3450     reportError(std::move(Err), A->getFileName());
3451 }
3452 
3453 /// Open file and figure out how to dump it.
dumpInput(StringRef file)3454 static void dumpInput(StringRef file) {
3455   // If we are using the Mach-O specific object file parser, then let it parse
3456   // the file and process the command line options.  So the -arch flags can
3457   // be used to select specific slices, etc.
3458   if (MachOOpt) {
3459     parseInputMachO(file);
3460     return;
3461   }
3462 
3463   // Attempt to open the binary.
3464   OwningBinary<Binary> OBinary = unwrapOrError(createBinary(file), file);
3465   Binary &Binary = *OBinary.getBinary();
3466 
3467   if (Archive *A = dyn_cast<Archive>(&Binary))
3468     dumpArchive(A);
3469   else if (ObjectFile *O = dyn_cast<ObjectFile>(&Binary))
3470     dumpObject(O);
3471   else if (MachOUniversalBinary *UB = dyn_cast<MachOUniversalBinary>(&Binary))
3472     parseInputMachO(UB);
3473   else if (OffloadBinary *OB = dyn_cast<OffloadBinary>(&Binary))
3474     dumpOffloadSections(*OB);
3475   else
3476     reportError(errorCodeToError(object_error::invalid_file_type), file);
3477 }
3478 
3479 template <typename T>
parseIntArg(const llvm::opt::InputArgList & InputArgs,int ID,T & Value)3480 static void parseIntArg(const llvm::opt::InputArgList &InputArgs, int ID,
3481                         T &Value) {
3482   if (const opt::Arg *A = InputArgs.getLastArg(ID)) {
3483     StringRef V(A->getValue());
3484     if (!llvm::to_integer(V, Value, 0)) {
3485       reportCmdLineError(A->getSpelling() +
3486                          ": expected a non-negative integer, but got '" + V +
3487                          "'");
3488     }
3489   }
3490 }
3491 
parseBuildIDArg(const opt::Arg * A)3492 static object::BuildID parseBuildIDArg(const opt::Arg *A) {
3493   StringRef V(A->getValue());
3494   object::BuildID BID = parseBuildID(V);
3495   if (BID.empty())
3496     reportCmdLineError(A->getSpelling() + ": expected a build ID, but got '" +
3497                        V + "'");
3498   return BID;
3499 }
3500 
invalidArgValue(const opt::Arg * A)3501 void objdump::invalidArgValue(const opt::Arg *A) {
3502   reportCmdLineError("'" + StringRef(A->getValue()) +
3503                      "' is not a valid value for '" + A->getSpelling() + "'");
3504 }
3505 
3506 static std::vector<std::string>
commaSeparatedValues(const llvm::opt::InputArgList & InputArgs,int ID)3507 commaSeparatedValues(const llvm::opt::InputArgList &InputArgs, int ID) {
3508   std::vector<std::string> Values;
3509   for (StringRef Value : InputArgs.getAllArgValues(ID)) {
3510     llvm::SmallVector<StringRef, 2> SplitValues;
3511     llvm::SplitString(Value, SplitValues, ",");
3512     for (StringRef SplitValue : SplitValues)
3513       Values.push_back(SplitValue.str());
3514   }
3515   return Values;
3516 }
3517 
parseOtoolOptions(const llvm::opt::InputArgList & InputArgs)3518 static void parseOtoolOptions(const llvm::opt::InputArgList &InputArgs) {
3519   MachOOpt = true;
3520   FullLeadingAddr = true;
3521   PrintImmHex = true;
3522 
3523   ArchName = InputArgs.getLastArgValue(OTOOL_arch).str();
3524   LinkOptHints = InputArgs.hasArg(OTOOL_C);
3525   if (InputArgs.hasArg(OTOOL_d))
3526     FilterSections.push_back("__DATA,__data");
3527   DylibId = InputArgs.hasArg(OTOOL_D);
3528   UniversalHeaders = InputArgs.hasArg(OTOOL_f);
3529   DataInCode = InputArgs.hasArg(OTOOL_G);
3530   FirstPrivateHeader = InputArgs.hasArg(OTOOL_h);
3531   IndirectSymbols = InputArgs.hasArg(OTOOL_I);
3532   ShowRawInsn = InputArgs.hasArg(OTOOL_j);
3533   PrivateHeaders = InputArgs.hasArg(OTOOL_l);
3534   DylibsUsed = InputArgs.hasArg(OTOOL_L);
3535   MCPU = InputArgs.getLastArgValue(OTOOL_mcpu_EQ).str();
3536   ObjcMetaData = InputArgs.hasArg(OTOOL_o);
3537   DisSymName = InputArgs.getLastArgValue(OTOOL_p).str();
3538   InfoPlist = InputArgs.hasArg(OTOOL_P);
3539   Relocations = InputArgs.hasArg(OTOOL_r);
3540   if (const Arg *A = InputArgs.getLastArg(OTOOL_s)) {
3541     auto Filter = (A->getValue(0) + StringRef(",") + A->getValue(1)).str();
3542     FilterSections.push_back(Filter);
3543   }
3544   if (InputArgs.hasArg(OTOOL_t))
3545     FilterSections.push_back("__TEXT,__text");
3546   Verbose = InputArgs.hasArg(OTOOL_v) || InputArgs.hasArg(OTOOL_V) ||
3547             InputArgs.hasArg(OTOOL_o);
3548   SymbolicOperands = InputArgs.hasArg(OTOOL_V);
3549   if (InputArgs.hasArg(OTOOL_x))
3550     FilterSections.push_back(",__text");
3551   LeadingAddr = LeadingHeaders = !InputArgs.hasArg(OTOOL_X);
3552 
3553   ChainedFixups = InputArgs.hasArg(OTOOL_chained_fixups);
3554   DyldInfo = InputArgs.hasArg(OTOOL_dyld_info);
3555 
3556   InputFilenames = InputArgs.getAllArgValues(OTOOL_INPUT);
3557   if (InputFilenames.empty())
3558     reportCmdLineError("no input file");
3559 
3560   for (const Arg *A : InputArgs) {
3561     const Option &O = A->getOption();
3562     if (O.getGroup().isValid() && O.getGroup().getID() == OTOOL_grp_obsolete) {
3563       reportCmdLineWarning(O.getPrefixedName() +
3564                            " is obsolete and not implemented");
3565     }
3566   }
3567 }
3568 
parseObjdumpOptions(const llvm::opt::InputArgList & InputArgs)3569 static void parseObjdumpOptions(const llvm::opt::InputArgList &InputArgs) {
3570   parseIntArg(InputArgs, OBJDUMP_adjust_vma_EQ, AdjustVMA);
3571   AllHeaders = InputArgs.hasArg(OBJDUMP_all_headers);
3572   ArchName = InputArgs.getLastArgValue(OBJDUMP_arch_name_EQ).str();
3573   ArchiveHeaders = InputArgs.hasArg(OBJDUMP_archive_headers);
3574   Demangle = InputArgs.hasArg(OBJDUMP_demangle);
3575   Disassemble = InputArgs.hasArg(OBJDUMP_disassemble);
3576   DisassembleAll = InputArgs.hasArg(OBJDUMP_disassemble_all);
3577   SymbolDescription = InputArgs.hasArg(OBJDUMP_symbol_description);
3578   TracebackTable = InputArgs.hasArg(OBJDUMP_traceback_table);
3579   DisassembleSymbols =
3580       commaSeparatedValues(InputArgs, OBJDUMP_disassemble_symbols_EQ);
3581   DisassembleZeroes = InputArgs.hasArg(OBJDUMP_disassemble_zeroes);
3582   if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_dwarf_EQ)) {
3583     DwarfDumpType = StringSwitch<DIDumpType>(A->getValue())
3584                         .Case("frames", DIDT_DebugFrame)
3585                         .Default(DIDT_Null);
3586     if (DwarfDumpType == DIDT_Null)
3587       invalidArgValue(A);
3588   }
3589   DynamicRelocations = InputArgs.hasArg(OBJDUMP_dynamic_reloc);
3590   FaultMapSection = InputArgs.hasArg(OBJDUMP_fault_map_section);
3591   Offloading = InputArgs.hasArg(OBJDUMP_offloading);
3592   FileHeaders = InputArgs.hasArg(OBJDUMP_file_headers);
3593   SectionContents = InputArgs.hasArg(OBJDUMP_full_contents);
3594   PrintLines = InputArgs.hasArg(OBJDUMP_line_numbers);
3595   InputFilenames = InputArgs.getAllArgValues(OBJDUMP_INPUT);
3596   MachOOpt = InputArgs.hasArg(OBJDUMP_macho);
3597   MCPU = InputArgs.getLastArgValue(OBJDUMP_mcpu_EQ).str();
3598   MAttrs = commaSeparatedValues(InputArgs, OBJDUMP_mattr_EQ);
3599   ShowRawInsn = !InputArgs.hasArg(OBJDUMP_no_show_raw_insn);
3600   LeadingAddr = !InputArgs.hasArg(OBJDUMP_no_leading_addr);
3601   RawClangAST = InputArgs.hasArg(OBJDUMP_raw_clang_ast);
3602   Relocations = InputArgs.hasArg(OBJDUMP_reloc);
3603   PrintImmHex =
3604       InputArgs.hasFlag(OBJDUMP_print_imm_hex, OBJDUMP_no_print_imm_hex, true);
3605   PrivateHeaders = InputArgs.hasArg(OBJDUMP_private_headers);
3606   FilterSections = InputArgs.getAllArgValues(OBJDUMP_section_EQ);
3607   SectionHeaders = InputArgs.hasArg(OBJDUMP_section_headers);
3608   ShowAllSymbols = InputArgs.hasArg(OBJDUMP_show_all_symbols);
3609   ShowLMA = InputArgs.hasArg(OBJDUMP_show_lma);
3610   PrintSource = InputArgs.hasArg(OBJDUMP_source);
3611   parseIntArg(InputArgs, OBJDUMP_start_address_EQ, StartAddress);
3612   HasStartAddressFlag = InputArgs.hasArg(OBJDUMP_start_address_EQ);
3613   parseIntArg(InputArgs, OBJDUMP_stop_address_EQ, StopAddress);
3614   HasStopAddressFlag = InputArgs.hasArg(OBJDUMP_stop_address_EQ);
3615   SymbolTable = InputArgs.hasArg(OBJDUMP_syms);
3616   SymbolizeOperands = InputArgs.hasArg(OBJDUMP_symbolize_operands);
3617   PrettyPGOAnalysisMap = InputArgs.hasArg(OBJDUMP_pretty_pgo_analysis_map);
3618   if (PrettyPGOAnalysisMap && !SymbolizeOperands)
3619     reportCmdLineWarning("--symbolize-operands must be enabled for "
3620                          "--pretty-pgo-analysis-map to have an effect");
3621   DynamicSymbolTable = InputArgs.hasArg(OBJDUMP_dynamic_syms);
3622   TripleName = InputArgs.getLastArgValue(OBJDUMP_triple_EQ).str();
3623   UnwindInfo = InputArgs.hasArg(OBJDUMP_unwind_info);
3624   Wide = InputArgs.hasArg(OBJDUMP_wide);
3625   Prefix = InputArgs.getLastArgValue(OBJDUMP_prefix).str();
3626   parseIntArg(InputArgs, OBJDUMP_prefix_strip, PrefixStrip);
3627   if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_debug_vars_EQ)) {
3628     DbgVariables = StringSwitch<DebugVarsFormat>(A->getValue())
3629                        .Case("ascii", DVASCII)
3630                        .Case("unicode", DVUnicode)
3631                        .Default(DVInvalid);
3632     if (DbgVariables == DVInvalid)
3633       invalidArgValue(A);
3634   }
3635   if (const opt::Arg *A = InputArgs.getLastArg(OBJDUMP_disassembler_color_EQ)) {
3636     DisassemblyColor = StringSwitch<ColorOutput>(A->getValue())
3637                            .Case("on", ColorOutput::Enable)
3638                            .Case("off", ColorOutput::Disable)
3639                            .Case("terminal", ColorOutput::Auto)
3640                            .Default(ColorOutput::Invalid);
3641     if (DisassemblyColor == ColorOutput::Invalid)
3642       invalidArgValue(A);
3643   }
3644 
3645   parseIntArg(InputArgs, OBJDUMP_debug_vars_indent_EQ, DbgIndent);
3646 
3647   parseMachOOptions(InputArgs);
3648 
3649   // Parse -M (--disassembler-options) and deprecated
3650   // --x86-asm-syntax={att,intel}.
3651   //
3652   // Note, for x86, the asm dialect (AssemblerDialect) is initialized when the
3653   // MCAsmInfo is constructed. MCInstPrinter::applyTargetSpecificCLOption is
3654   // called too late. For now we have to use the internal cl::opt option.
3655   const char *AsmSyntax = nullptr;
3656   for (const auto *A : InputArgs.filtered(OBJDUMP_disassembler_options_EQ,
3657                                           OBJDUMP_x86_asm_syntax_att,
3658                                           OBJDUMP_x86_asm_syntax_intel)) {
3659     switch (A->getOption().getID()) {
3660     case OBJDUMP_x86_asm_syntax_att:
3661       AsmSyntax = "--x86-asm-syntax=att";
3662       continue;
3663     case OBJDUMP_x86_asm_syntax_intel:
3664       AsmSyntax = "--x86-asm-syntax=intel";
3665       continue;
3666     }
3667 
3668     SmallVector<StringRef, 2> Values;
3669     llvm::SplitString(A->getValue(), Values, ",");
3670     for (StringRef V : Values) {
3671       if (V == "att")
3672         AsmSyntax = "--x86-asm-syntax=att";
3673       else if (V == "intel")
3674         AsmSyntax = "--x86-asm-syntax=intel";
3675       else
3676         DisassemblerOptions.push_back(V.str());
3677     }
3678   }
3679   SmallVector<const char *> Args = {"llvm-objdump"};
3680   for (const opt::Arg *A : InputArgs.filtered(OBJDUMP_mllvm))
3681     Args.push_back(A->getValue());
3682   if (AsmSyntax)
3683     Args.push_back(AsmSyntax);
3684   if (Args.size() > 1)
3685     llvm::cl::ParseCommandLineOptions(Args.size(), Args.data());
3686 
3687   // Look up any provided build IDs, then append them to the input filenames.
3688   for (const opt::Arg *A : InputArgs.filtered(OBJDUMP_build_id)) {
3689     object::BuildID BuildID = parseBuildIDArg(A);
3690     std::optional<std::string> Path = BIDFetcher->fetch(BuildID);
3691     if (!Path) {
3692       reportCmdLineError(A->getSpelling() + ": could not find build ID '" +
3693                          A->getValue() + "'");
3694     }
3695     InputFilenames.push_back(std::move(*Path));
3696   }
3697 
3698   // objdump defaults to a.out if no filenames specified.
3699   if (InputFilenames.empty())
3700     InputFilenames.push_back("a.out");
3701 }
3702 
llvm_objdump_main(int argc,char ** argv,const llvm::ToolContext &)3703 int llvm_objdump_main(int argc, char **argv, const llvm::ToolContext &) {
3704   using namespace llvm;
3705 
3706   ToolName = argv[0];
3707   std::unique_ptr<CommonOptTable> T;
3708   OptSpecifier Unknown, HelpFlag, HelpHiddenFlag, VersionFlag;
3709 
3710   StringRef Stem = sys::path::stem(ToolName);
3711   auto Is = [=](StringRef Tool) {
3712     // We need to recognize the following filenames:
3713     //
3714     // llvm-objdump -> objdump
3715     // llvm-otool-10.exe -> otool
3716     // powerpc64-unknown-freebsd13-objdump -> objdump
3717     auto I = Stem.rfind_insensitive(Tool);
3718     return I != StringRef::npos &&
3719            (I + Tool.size() == Stem.size() || !isAlnum(Stem[I + Tool.size()]));
3720   };
3721   if (Is("otool")) {
3722     T = std::make_unique<OtoolOptTable>();
3723     Unknown = OTOOL_UNKNOWN;
3724     HelpFlag = OTOOL_help;
3725     HelpHiddenFlag = OTOOL_help_hidden;
3726     VersionFlag = OTOOL_version;
3727   } else {
3728     T = std::make_unique<ObjdumpOptTable>();
3729     Unknown = OBJDUMP_UNKNOWN;
3730     HelpFlag = OBJDUMP_help;
3731     HelpHiddenFlag = OBJDUMP_help_hidden;
3732     VersionFlag = OBJDUMP_version;
3733   }
3734 
3735   BumpPtrAllocator A;
3736   StringSaver Saver(A);
3737   opt::InputArgList InputArgs =
3738       T->parseArgs(argc, argv, Unknown, Saver,
3739                    [&](StringRef Msg) { reportCmdLineError(Msg); });
3740 
3741   if (InputArgs.size() == 0 || InputArgs.hasArg(HelpFlag)) {
3742     T->printHelp(ToolName);
3743     return 0;
3744   }
3745   if (InputArgs.hasArg(HelpHiddenFlag)) {
3746     T->printHelp(ToolName, /*ShowHidden=*/true);
3747     return 0;
3748   }
3749 
3750   // Initialize targets and assembly printers/parsers.
3751   InitializeAllTargetInfos();
3752   InitializeAllTargetMCs();
3753   InitializeAllDisassemblers();
3754 
3755   if (InputArgs.hasArg(VersionFlag)) {
3756     cl::PrintVersionMessage();
3757     if (!Is("otool")) {
3758       outs() << '\n';
3759       TargetRegistry::printRegisteredTargetsForVersion(outs());
3760     }
3761     return 0;
3762   }
3763 
3764   // Initialize debuginfod.
3765   const bool ShouldUseDebuginfodByDefault =
3766       InputArgs.hasArg(OBJDUMP_build_id) || canUseDebuginfod();
3767   std::vector<std::string> DebugFileDirectories =
3768       InputArgs.getAllArgValues(OBJDUMP_debug_file_directory);
3769   if (InputArgs.hasFlag(OBJDUMP_debuginfod, OBJDUMP_no_debuginfod,
3770                         ShouldUseDebuginfodByDefault)) {
3771     HTTPClient::initialize();
3772     BIDFetcher =
3773         std::make_unique<DebuginfodFetcher>(std::move(DebugFileDirectories));
3774   } else {
3775     BIDFetcher =
3776         std::make_unique<BuildIDFetcher>(std::move(DebugFileDirectories));
3777   }
3778 
3779   if (Is("otool"))
3780     parseOtoolOptions(InputArgs);
3781   else
3782     parseObjdumpOptions(InputArgs);
3783 
3784   if (StartAddress >= StopAddress)
3785     reportCmdLineError("start address should be less than stop address");
3786 
3787   // Removes trailing separators from prefix.
3788   while (!Prefix.empty() && sys::path::is_separator(Prefix.back()))
3789     Prefix.pop_back();
3790 
3791   if (AllHeaders)
3792     ArchiveHeaders = FileHeaders = PrivateHeaders = Relocations =
3793         SectionHeaders = SymbolTable = true;
3794 
3795   if (DisassembleAll || PrintSource || PrintLines || TracebackTable ||
3796       !DisassembleSymbols.empty())
3797     Disassemble = true;
3798 
3799   if (!ArchiveHeaders && !Disassemble && DwarfDumpType == DIDT_Null &&
3800       !DynamicRelocations && !FileHeaders && !PrivateHeaders && !RawClangAST &&
3801       !Relocations && !SectionHeaders && !SectionContents && !SymbolTable &&
3802       !DynamicSymbolTable && !UnwindInfo && !FaultMapSection && !Offloading &&
3803       !(MachOOpt &&
3804         (Bind || DataInCode || ChainedFixups || DyldInfo || DylibId ||
3805          DylibsUsed || ExportsTrie || FirstPrivateHeader ||
3806          FunctionStartsType != FunctionStartsMode::None || IndirectSymbols ||
3807          InfoPlist || LazyBind || LinkOptHints || ObjcMetaData || Rebase ||
3808          Rpaths || UniversalHeaders || WeakBind || !FilterSections.empty()))) {
3809     T->printHelp(ToolName);
3810     return 2;
3811   }
3812 
3813   DisasmSymbolSet.insert_range(DisassembleSymbols);
3814 
3815   llvm::for_each(InputFilenames, dumpInput);
3816 
3817   warnOnNoMatchForSections();
3818 
3819   return EXIT_SUCCESS;
3820 }
3821