1 //===-- LLVMSymbolize.cpp -------------------------------------------------===//
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 // Implementation for LLVM symbolization library.
10 //
11 //===----------------------------------------------------------------------===//
12
13 #include "llvm/DebugInfo/Symbolize/Symbolize.h"
14
15 #include "SymbolizableObjectFile.h"
16
17 #include "llvm/ADT/STLExtras.h"
18 #include "llvm/BinaryFormat/COFF.h"
19 #include "llvm/DebugInfo/DWARF/DWARFContext.h"
20 #include "llvm/DebugInfo/PDB/PDB.h"
21 #include "llvm/DebugInfo/PDB/PDBContext.h"
22 #include "llvm/Demangle/Demangle.h"
23 #include "llvm/Object/COFF.h"
24 #include "llvm/Object/MachO.h"
25 #include "llvm/Object/MachOUniversal.h"
26 #include "llvm/Support/CRC.h"
27 #include "llvm/Support/Casting.h"
28 #include "llvm/Support/Compression.h"
29 #include "llvm/Support/DataExtractor.h"
30 #include "llvm/Support/Errc.h"
31 #include "llvm/Support/FileSystem.h"
32 #include "llvm/Support/MemoryBuffer.h"
33 #include "llvm/Support/Path.h"
34 #include <algorithm>
35 #include <cassert>
36 #include <cstring>
37
38 namespace llvm {
39 namespace symbolize {
40
41 Expected<DILineInfo>
symbolizeCodeCommon(SymbolizableModule * Info,object::SectionedAddress ModuleOffset)42 LLVMSymbolizer::symbolizeCodeCommon(SymbolizableModule *Info,
43 object::SectionedAddress ModuleOffset) {
44 // A null module means an error has already been reported. Return an empty
45 // result.
46 if (!Info)
47 return DILineInfo();
48
49 // If the user is giving us relative addresses, add the preferred base of the
50 // object to the offset before we do the query. It's what DIContext expects.
51 if (Opts.RelativeAddresses)
52 ModuleOffset.Address += Info->getModulePreferredBase();
53
54 DILineInfo LineInfo = Info->symbolizeCode(ModuleOffset, Opts.PrintFunctions,
55 Opts.UseSymbolTable);
56 if (Opts.Demangle)
57 LineInfo.FunctionName = DemangleName(LineInfo.FunctionName, Info);
58 return LineInfo;
59 }
60
61 Expected<DILineInfo>
symbolizeCode(const ObjectFile & Obj,object::SectionedAddress ModuleOffset)62 LLVMSymbolizer::symbolizeCode(const ObjectFile &Obj,
63 object::SectionedAddress ModuleOffset) {
64 StringRef ModuleName = Obj.getFileName();
65 auto I = Modules.find(ModuleName);
66 if (I != Modules.end())
67 return symbolizeCodeCommon(I->second.get(), ModuleOffset);
68
69 std::unique_ptr<DIContext> Context =
70 DWARFContext::create(Obj, nullptr, DWARFContext::defaultErrorHandler);
71 Expected<SymbolizableModule *> InfoOrErr =
72 createModuleInfo(&Obj, std::move(Context), ModuleName);
73 if (!InfoOrErr)
74 return InfoOrErr.takeError();
75 return symbolizeCodeCommon(*InfoOrErr, ModuleOffset);
76 }
77
78 Expected<DILineInfo>
symbolizeCode(const std::string & ModuleName,object::SectionedAddress ModuleOffset)79 LLVMSymbolizer::symbolizeCode(const std::string &ModuleName,
80 object::SectionedAddress ModuleOffset) {
81 Expected<SymbolizableModule *> InfoOrErr = getOrCreateModuleInfo(ModuleName);
82 if (!InfoOrErr)
83 return InfoOrErr.takeError();
84 return symbolizeCodeCommon(*InfoOrErr, ModuleOffset);
85 }
86
87 Expected<DIInliningInfo>
symbolizeInlinedCode(const std::string & ModuleName,object::SectionedAddress ModuleOffset)88 LLVMSymbolizer::symbolizeInlinedCode(const std::string &ModuleName,
89 object::SectionedAddress ModuleOffset) {
90 SymbolizableModule *Info;
91 if (auto InfoOrErr = getOrCreateModuleInfo(ModuleName))
92 Info = InfoOrErr.get();
93 else
94 return InfoOrErr.takeError();
95
96 // A null module means an error has already been reported. Return an empty
97 // result.
98 if (!Info)
99 return DIInliningInfo();
100
101 // If the user is giving us relative addresses, add the preferred base of the
102 // object to the offset before we do the query. It's what DIContext expects.
103 if (Opts.RelativeAddresses)
104 ModuleOffset.Address += Info->getModulePreferredBase();
105
106 DIInliningInfo InlinedContext = Info->symbolizeInlinedCode(
107 ModuleOffset, Opts.PrintFunctions, Opts.UseSymbolTable);
108 if (Opts.Demangle) {
109 for (int i = 0, n = InlinedContext.getNumberOfFrames(); i < n; i++) {
110 auto *Frame = InlinedContext.getMutableFrame(i);
111 Frame->FunctionName = DemangleName(Frame->FunctionName, Info);
112 }
113 }
114 return InlinedContext;
115 }
116
117 Expected<DIGlobal>
symbolizeData(const std::string & ModuleName,object::SectionedAddress ModuleOffset)118 LLVMSymbolizer::symbolizeData(const std::string &ModuleName,
119 object::SectionedAddress ModuleOffset) {
120 SymbolizableModule *Info;
121 if (auto InfoOrErr = getOrCreateModuleInfo(ModuleName))
122 Info = InfoOrErr.get();
123 else
124 return InfoOrErr.takeError();
125
126 // A null module means an error has already been reported. Return an empty
127 // result.
128 if (!Info)
129 return DIGlobal();
130
131 // If the user is giving us relative addresses, add the preferred base of
132 // the object to the offset before we do the query. It's what DIContext
133 // expects.
134 if (Opts.RelativeAddresses)
135 ModuleOffset.Address += Info->getModulePreferredBase();
136
137 DIGlobal Global = Info->symbolizeData(ModuleOffset);
138 if (Opts.Demangle)
139 Global.Name = DemangleName(Global.Name, Info);
140 return Global;
141 }
142
143 Expected<std::vector<DILocal>>
symbolizeFrame(const std::string & ModuleName,object::SectionedAddress ModuleOffset)144 LLVMSymbolizer::symbolizeFrame(const std::string &ModuleName,
145 object::SectionedAddress ModuleOffset) {
146 SymbolizableModule *Info;
147 if (auto InfoOrErr = getOrCreateModuleInfo(ModuleName))
148 Info = InfoOrErr.get();
149 else
150 return InfoOrErr.takeError();
151
152 // A null module means an error has already been reported. Return an empty
153 // result.
154 if (!Info)
155 return std::vector<DILocal>();
156
157 // If the user is giving us relative addresses, add the preferred base of
158 // the object to the offset before we do the query. It's what DIContext
159 // expects.
160 if (Opts.RelativeAddresses)
161 ModuleOffset.Address += Info->getModulePreferredBase();
162
163 return Info->symbolizeFrame(ModuleOffset);
164 }
165
flush()166 void LLVMSymbolizer::flush() {
167 ObjectForUBPathAndArch.clear();
168 BinaryForPath.clear();
169 ObjectPairForPathArch.clear();
170 Modules.clear();
171 }
172
173 namespace {
174
175 // For Path="/path/to/foo" and Basename="foo" assume that debug info is in
176 // /path/to/foo.dSYM/Contents/Resources/DWARF/foo.
177 // For Path="/path/to/bar.dSYM" and Basename="foo" assume that debug info is in
178 // /path/to/bar.dSYM/Contents/Resources/DWARF/foo.
getDarwinDWARFResourceForPath(const std::string & Path,const std::string & Basename)179 std::string getDarwinDWARFResourceForPath(
180 const std::string &Path, const std::string &Basename) {
181 SmallString<16> ResourceName = StringRef(Path);
182 if (sys::path::extension(Path) != ".dSYM") {
183 ResourceName += ".dSYM";
184 }
185 sys::path::append(ResourceName, "Contents", "Resources", "DWARF");
186 sys::path::append(ResourceName, Basename);
187 return ResourceName.str();
188 }
189
checkFileCRC(StringRef Path,uint32_t CRCHash)190 bool checkFileCRC(StringRef Path, uint32_t CRCHash) {
191 ErrorOr<std::unique_ptr<MemoryBuffer>> MB =
192 MemoryBuffer::getFileOrSTDIN(Path);
193 if (!MB)
194 return false;
195 return CRCHash == llvm::crc32(arrayRefFromStringRef(MB.get()->getBuffer()));
196 }
197
findDebugBinary(const std::string & OrigPath,const std::string & DebuglinkName,uint32_t CRCHash,const std::string & FallbackDebugPath,std::string & Result)198 bool findDebugBinary(const std::string &OrigPath,
199 const std::string &DebuglinkName, uint32_t CRCHash,
200 const std::string &FallbackDebugPath,
201 std::string &Result) {
202 SmallString<16> OrigDir(OrigPath);
203 llvm::sys::path::remove_filename(OrigDir);
204 SmallString<16> DebugPath = OrigDir;
205 // Try relative/path/to/original_binary/debuglink_name
206 llvm::sys::path::append(DebugPath, DebuglinkName);
207 if (checkFileCRC(DebugPath, CRCHash)) {
208 Result = DebugPath.str();
209 return true;
210 }
211 // Try relative/path/to/original_binary/.debug/debuglink_name
212 DebugPath = OrigDir;
213 llvm::sys::path::append(DebugPath, ".debug", DebuglinkName);
214 if (checkFileCRC(DebugPath, CRCHash)) {
215 Result = DebugPath.str();
216 return true;
217 }
218 // Make the path absolute so that lookups will go to
219 // "/usr/lib/debug/full/path/to/debug", not
220 // "/usr/lib/debug/to/debug"
221 llvm::sys::fs::make_absolute(OrigDir);
222 if (!FallbackDebugPath.empty()) {
223 // Try <FallbackDebugPath>/absolute/path/to/original_binary/debuglink_name
224 DebugPath = FallbackDebugPath;
225 } else {
226 #if defined(__NetBSD__)
227 // Try /usr/libdata/debug/absolute/path/to/original_binary/debuglink_name
228 DebugPath = "/usr/libdata/debug";
229 #else
230 // Try /usr/lib/debug/absolute/path/to/original_binary/debuglink_name
231 DebugPath = "/usr/lib/debug";
232 #endif
233 }
234 llvm::sys::path::append(DebugPath, llvm::sys::path::relative_path(OrigDir),
235 DebuglinkName);
236 if (checkFileCRC(DebugPath, CRCHash)) {
237 Result = DebugPath.str();
238 return true;
239 }
240 return false;
241 }
242
getGNUDebuglinkContents(const ObjectFile * Obj,std::string & DebugName,uint32_t & CRCHash)243 bool getGNUDebuglinkContents(const ObjectFile *Obj, std::string &DebugName,
244 uint32_t &CRCHash) {
245 if (!Obj)
246 return false;
247 for (const SectionRef &Section : Obj->sections()) {
248 StringRef Name;
249 if (Expected<StringRef> NameOrErr = Section.getName())
250 Name = *NameOrErr;
251 else
252 consumeError(NameOrErr.takeError());
253
254 Name = Name.substr(Name.find_first_not_of("._"));
255 if (Name == "gnu_debuglink") {
256 Expected<StringRef> ContentsOrErr = Section.getContents();
257 if (!ContentsOrErr) {
258 consumeError(ContentsOrErr.takeError());
259 return false;
260 }
261 DataExtractor DE(*ContentsOrErr, Obj->isLittleEndian(), 0);
262 uint64_t Offset = 0;
263 if (const char *DebugNameStr = DE.getCStr(&Offset)) {
264 // 4-byte align the offset.
265 Offset = (Offset + 3) & ~0x3;
266 if (DE.isValidOffsetForDataOfSize(Offset, 4)) {
267 DebugName = DebugNameStr;
268 CRCHash = DE.getU32(&Offset);
269 return true;
270 }
271 }
272 break;
273 }
274 }
275 return false;
276 }
277
darwinDsymMatchesBinary(const MachOObjectFile * DbgObj,const MachOObjectFile * Obj)278 bool darwinDsymMatchesBinary(const MachOObjectFile *DbgObj,
279 const MachOObjectFile *Obj) {
280 ArrayRef<uint8_t> dbg_uuid = DbgObj->getUuid();
281 ArrayRef<uint8_t> bin_uuid = Obj->getUuid();
282 if (dbg_uuid.empty() || bin_uuid.empty())
283 return false;
284 return !memcmp(dbg_uuid.data(), bin_uuid.data(), dbg_uuid.size());
285 }
286
287 template <typename ELFT>
getBuildID(const ELFFile<ELFT> * Obj)288 Optional<ArrayRef<uint8_t>> getBuildID(const ELFFile<ELFT> *Obj) {
289 if (!Obj)
290 return {};
291 auto PhdrsOrErr = Obj->program_headers();
292 if (!PhdrsOrErr) {
293 consumeError(PhdrsOrErr.takeError());
294 return {};
295 }
296 for (const auto &P : *PhdrsOrErr) {
297 if (P.p_type != ELF::PT_NOTE)
298 continue;
299 Error Err = Error::success();
300 for (auto N : Obj->notes(P, Err))
301 if (N.getType() == ELF::NT_GNU_BUILD_ID && N.getName() == ELF::ELF_NOTE_GNU)
302 return N.getDesc();
303 }
304 return {};
305 }
306
getBuildID(const ELFObjectFileBase * Obj)307 Optional<ArrayRef<uint8_t>> getBuildID(const ELFObjectFileBase *Obj) {
308 Optional<ArrayRef<uint8_t>> BuildID;
309 if (auto *O = dyn_cast<ELFObjectFile<ELF32LE>>(Obj))
310 BuildID = getBuildID(O->getELFFile());
311 else if (auto *O = dyn_cast<ELFObjectFile<ELF32BE>>(Obj))
312 BuildID = getBuildID(O->getELFFile());
313 else if (auto *O = dyn_cast<ELFObjectFile<ELF64LE>>(Obj))
314 BuildID = getBuildID(O->getELFFile());
315 else if (auto *O = dyn_cast<ELFObjectFile<ELF64BE>>(Obj))
316 BuildID = getBuildID(O->getELFFile());
317 else
318 llvm_unreachable("unsupported file format");
319 return BuildID;
320 }
321
findDebugBinary(const std::vector<std::string> & DebugFileDirectory,const ArrayRef<uint8_t> BuildID,std::string & Result)322 bool findDebugBinary(const std::vector<std::string> &DebugFileDirectory,
323 const ArrayRef<uint8_t> BuildID,
324 std::string &Result) {
325 auto getDebugPath = [&](StringRef Directory) {
326 SmallString<128> Path{Directory};
327 sys::path::append(Path, ".build-id",
328 llvm::toHex(BuildID[0], /*LowerCase=*/true),
329 llvm::toHex(BuildID.slice(1), /*LowerCase=*/true));
330 Path += ".debug";
331 return Path;
332 };
333 if (DebugFileDirectory.empty()) {
334 SmallString<128> Path = getDebugPath(
335 #if defined(__NetBSD__)
336 // Try /usr/libdata/debug/.build-id/../...
337 "/usr/libdata/debug"
338 #else
339 // Try /usr/lib/debug/.build-id/../...
340 "/usr/lib/debug"
341 #endif
342 );
343 if (llvm::sys::fs::exists(Path)) {
344 Result = Path.str();
345 return true;
346 }
347 } else {
348 for (const auto &Directory : DebugFileDirectory) {
349 // Try <debug-file-directory>/.build-id/../...
350 SmallString<128> Path = getDebugPath(Directory);
351 if (llvm::sys::fs::exists(Path)) {
352 Result = Path.str();
353 return true;
354 }
355 }
356 }
357 return false;
358 }
359
360 } // end anonymous namespace
361
lookUpDsymFile(const std::string & ExePath,const MachOObjectFile * MachExeObj,const std::string & ArchName)362 ObjectFile *LLVMSymbolizer::lookUpDsymFile(const std::string &ExePath,
363 const MachOObjectFile *MachExeObj, const std::string &ArchName) {
364 // On Darwin we may find DWARF in separate object file in
365 // resource directory.
366 std::vector<std::string> DsymPaths;
367 StringRef Filename = sys::path::filename(ExePath);
368 DsymPaths.push_back(getDarwinDWARFResourceForPath(ExePath, Filename));
369 for (const auto &Path : Opts.DsymHints) {
370 DsymPaths.push_back(getDarwinDWARFResourceForPath(Path, Filename));
371 }
372 for (const auto &Path : DsymPaths) {
373 auto DbgObjOrErr = getOrCreateObject(Path, ArchName);
374 if (!DbgObjOrErr) {
375 // Ignore errors, the file might not exist.
376 consumeError(DbgObjOrErr.takeError());
377 continue;
378 }
379 ObjectFile *DbgObj = DbgObjOrErr.get();
380 if (!DbgObj)
381 continue;
382 const MachOObjectFile *MachDbgObj = dyn_cast<const MachOObjectFile>(DbgObj);
383 if (!MachDbgObj)
384 continue;
385 if (darwinDsymMatchesBinary(MachDbgObj, MachExeObj))
386 return DbgObj;
387 }
388 return nullptr;
389 }
390
lookUpDebuglinkObject(const std::string & Path,const ObjectFile * Obj,const std::string & ArchName)391 ObjectFile *LLVMSymbolizer::lookUpDebuglinkObject(const std::string &Path,
392 const ObjectFile *Obj,
393 const std::string &ArchName) {
394 std::string DebuglinkName;
395 uint32_t CRCHash;
396 std::string DebugBinaryPath;
397 if (!getGNUDebuglinkContents(Obj, DebuglinkName, CRCHash))
398 return nullptr;
399 if (!findDebugBinary(Path, DebuglinkName, CRCHash, Opts.FallbackDebugPath,
400 DebugBinaryPath))
401 return nullptr;
402 auto DbgObjOrErr = getOrCreateObject(DebugBinaryPath, ArchName);
403 if (!DbgObjOrErr) {
404 // Ignore errors, the file might not exist.
405 consumeError(DbgObjOrErr.takeError());
406 return nullptr;
407 }
408 return DbgObjOrErr.get();
409 }
410
lookUpBuildIDObject(const std::string & Path,const ELFObjectFileBase * Obj,const std::string & ArchName)411 ObjectFile *LLVMSymbolizer::lookUpBuildIDObject(const std::string &Path,
412 const ELFObjectFileBase *Obj,
413 const std::string &ArchName) {
414 auto BuildID = getBuildID(Obj);
415 if (!BuildID)
416 return nullptr;
417 if (BuildID->size() < 2)
418 return nullptr;
419 std::string DebugBinaryPath;
420 if (!findDebugBinary(Opts.DebugFileDirectory, *BuildID, DebugBinaryPath))
421 return nullptr;
422 auto DbgObjOrErr = getOrCreateObject(DebugBinaryPath, ArchName);
423 if (!DbgObjOrErr) {
424 consumeError(DbgObjOrErr.takeError());
425 return nullptr;
426 }
427 return DbgObjOrErr.get();
428 }
429
430 Expected<LLVMSymbolizer::ObjectPair>
getOrCreateObjectPair(const std::string & Path,const std::string & ArchName)431 LLVMSymbolizer::getOrCreateObjectPair(const std::string &Path,
432 const std::string &ArchName) {
433 auto I = ObjectPairForPathArch.find(std::make_pair(Path, ArchName));
434 if (I != ObjectPairForPathArch.end())
435 return I->second;
436
437 auto ObjOrErr = getOrCreateObject(Path, ArchName);
438 if (!ObjOrErr) {
439 ObjectPairForPathArch.emplace(std::make_pair(Path, ArchName),
440 ObjectPair(nullptr, nullptr));
441 return ObjOrErr.takeError();
442 }
443
444 ObjectFile *Obj = ObjOrErr.get();
445 assert(Obj != nullptr);
446 ObjectFile *DbgObj = nullptr;
447
448 if (auto MachObj = dyn_cast<const MachOObjectFile>(Obj))
449 DbgObj = lookUpDsymFile(Path, MachObj, ArchName);
450 else if (auto ELFObj = dyn_cast<const ELFObjectFileBase>(Obj))
451 DbgObj = lookUpBuildIDObject(Path, ELFObj, ArchName);
452 if (!DbgObj)
453 DbgObj = lookUpDebuglinkObject(Path, Obj, ArchName);
454 if (!DbgObj)
455 DbgObj = Obj;
456 ObjectPair Res = std::make_pair(Obj, DbgObj);
457 ObjectPairForPathArch.emplace(std::make_pair(Path, ArchName), Res);
458 return Res;
459 }
460
461 Expected<ObjectFile *>
getOrCreateObject(const std::string & Path,const std::string & ArchName)462 LLVMSymbolizer::getOrCreateObject(const std::string &Path,
463 const std::string &ArchName) {
464 Binary *Bin;
465 auto Pair = BinaryForPath.emplace(Path, OwningBinary<Binary>());
466 if (!Pair.second) {
467 Bin = Pair.first->second.getBinary();
468 } else {
469 Expected<OwningBinary<Binary>> BinOrErr = createBinary(Path);
470 if (!BinOrErr)
471 return BinOrErr.takeError();
472 Pair.first->second = std::move(BinOrErr.get());
473 Bin = Pair.first->second.getBinary();
474 }
475
476 if (!Bin)
477 return static_cast<ObjectFile *>(nullptr);
478
479 if (MachOUniversalBinary *UB = dyn_cast_or_null<MachOUniversalBinary>(Bin)) {
480 auto I = ObjectForUBPathAndArch.find(std::make_pair(Path, ArchName));
481 if (I != ObjectForUBPathAndArch.end())
482 return I->second.get();
483
484 Expected<std::unique_ptr<ObjectFile>> ObjOrErr =
485 UB->getMachOObjectForArch(ArchName);
486 if (!ObjOrErr) {
487 ObjectForUBPathAndArch.emplace(std::make_pair(Path, ArchName),
488 std::unique_ptr<ObjectFile>());
489 return ObjOrErr.takeError();
490 }
491 ObjectFile *Res = ObjOrErr->get();
492 ObjectForUBPathAndArch.emplace(std::make_pair(Path, ArchName),
493 std::move(ObjOrErr.get()));
494 return Res;
495 }
496 if (Bin->isObject()) {
497 return cast<ObjectFile>(Bin);
498 }
499 return errorCodeToError(object_error::arch_not_found);
500 }
501
502 Expected<SymbolizableModule *>
createModuleInfo(const ObjectFile * Obj,std::unique_ptr<DIContext> Context,StringRef ModuleName)503 LLVMSymbolizer::createModuleInfo(const ObjectFile *Obj,
504 std::unique_ptr<DIContext> Context,
505 StringRef ModuleName) {
506 auto InfoOrErr = SymbolizableObjectFile::create(Obj, std::move(Context),
507 Opts.UntagAddresses);
508 std::unique_ptr<SymbolizableModule> SymMod;
509 if (InfoOrErr)
510 SymMod = std::move(*InfoOrErr);
511 auto InsertResult =
512 Modules.insert(std::make_pair(ModuleName, std::move(SymMod)));
513 assert(InsertResult.second);
514 if (std::error_code EC = InfoOrErr.getError())
515 return errorCodeToError(EC);
516 return InsertResult.first->second.get();
517 }
518
519 Expected<SymbolizableModule *>
getOrCreateModuleInfo(const std::string & ModuleName)520 LLVMSymbolizer::getOrCreateModuleInfo(const std::string &ModuleName) {
521 auto I = Modules.find(ModuleName);
522 if (I != Modules.end())
523 return I->second.get();
524
525 std::string BinaryName = ModuleName;
526 std::string ArchName = Opts.DefaultArch;
527 size_t ColonPos = ModuleName.find_last_of(':');
528 // Verify that substring after colon form a valid arch name.
529 if (ColonPos != std::string::npos) {
530 std::string ArchStr = ModuleName.substr(ColonPos + 1);
531 if (Triple(ArchStr).getArch() != Triple::UnknownArch) {
532 BinaryName = ModuleName.substr(0, ColonPos);
533 ArchName = ArchStr;
534 }
535 }
536 auto ObjectsOrErr = getOrCreateObjectPair(BinaryName, ArchName);
537 if (!ObjectsOrErr) {
538 // Failed to find valid object file.
539 Modules.emplace(ModuleName, std::unique_ptr<SymbolizableModule>());
540 return ObjectsOrErr.takeError();
541 }
542 ObjectPair Objects = ObjectsOrErr.get();
543
544 std::unique_ptr<DIContext> Context;
545 // If this is a COFF object containing PDB info, use a PDBContext to
546 // symbolize. Otherwise, use DWARF.
547 if (auto CoffObject = dyn_cast<COFFObjectFile>(Objects.first)) {
548 const codeview::DebugInfo *DebugInfo;
549 StringRef PDBFileName;
550 auto EC = CoffObject->getDebugPDBInfo(DebugInfo, PDBFileName);
551 if (!EC && DebugInfo != nullptr && !PDBFileName.empty()) {
552 using namespace pdb;
553 std::unique_ptr<IPDBSession> Session;
554 if (auto Err = loadDataForEXE(PDB_ReaderType::DIA,
555 Objects.first->getFileName(), Session)) {
556 Modules.emplace(ModuleName, std::unique_ptr<SymbolizableModule>());
557 // Return along the PDB filename to provide more context
558 return createFileError(PDBFileName, std::move(Err));
559 }
560 Context.reset(new PDBContext(*CoffObject, std::move(Session)));
561 }
562 }
563 if (!Context)
564 Context =
565 DWARFContext::create(*Objects.second, nullptr,
566 DWARFContext::defaultErrorHandler, Opts.DWPName);
567 return createModuleInfo(Objects.first, std::move(Context), ModuleName);
568 }
569
570 namespace {
571
572 // Undo these various manglings for Win32 extern "C" functions:
573 // cdecl - _foo
574 // stdcall - _foo@12
575 // fastcall - @foo@12
576 // vectorcall - foo@@12
577 // These are all different linkage names for 'foo'.
demanglePE32ExternCFunc(StringRef SymbolName)578 StringRef demanglePE32ExternCFunc(StringRef SymbolName) {
579 // Remove any '_' or '@' prefix.
580 char Front = SymbolName.empty() ? '\0' : SymbolName[0];
581 if (Front == '_' || Front == '@')
582 SymbolName = SymbolName.drop_front();
583
584 // Remove any '@[0-9]+' suffix.
585 if (Front != '?') {
586 size_t AtPos = SymbolName.rfind('@');
587 if (AtPos != StringRef::npos &&
588 std::all_of(SymbolName.begin() + AtPos + 1, SymbolName.end(),
589 [](char C) { return C >= '0' && C <= '9'; })) {
590 SymbolName = SymbolName.substr(0, AtPos);
591 }
592 }
593
594 // Remove any ending '@' for vectorcall.
595 if (SymbolName.endswith("@"))
596 SymbolName = SymbolName.drop_back();
597
598 return SymbolName;
599 }
600
601 } // end anonymous namespace
602
603 std::string
DemangleName(const std::string & Name,const SymbolizableModule * DbiModuleDescriptor)604 LLVMSymbolizer::DemangleName(const std::string &Name,
605 const SymbolizableModule *DbiModuleDescriptor) {
606 // We can spoil names of symbols with C linkage, so use an heuristic
607 // approach to check if the name should be demangled.
608 if (Name.substr(0, 2) == "_Z") {
609 int status = 0;
610 char *DemangledName = itaniumDemangle(Name.c_str(), nullptr, nullptr, &status);
611 if (status != 0)
612 return Name;
613 std::string Result = DemangledName;
614 free(DemangledName);
615 return Result;
616 }
617
618 if (!Name.empty() && Name.front() == '?') {
619 // Only do MSVC C++ demangling on symbols starting with '?'.
620 int status = 0;
621 char *DemangledName = microsoftDemangle(
622 Name.c_str(), nullptr, nullptr, &status,
623 MSDemangleFlags(MSDF_NoAccessSpecifier | MSDF_NoCallingConvention |
624 MSDF_NoMemberType | MSDF_NoReturnType));
625 if (status != 0)
626 return Name;
627 std::string Result = DemangledName;
628 free(DemangledName);
629 return Result;
630 }
631
632 if (DbiModuleDescriptor && DbiModuleDescriptor->isWin32Module())
633 return std::string(demanglePE32ExternCFunc(Name));
634 return Name;
635 }
636
637 } // namespace symbolize
638 } // namespace llvm
639