1 //===- PDB.cpp ------------------------------------------------------------===//
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
14 #include "SymbolTable.h"
17 #include "lld/Common/ErrorHandler.h"
18 #include "lld/Common/Timer.h"
19 #include "llvm/DebugInfo/CodeView/DebugFrameDataSubsection.h"
20 #include "llvm/DebugInfo/CodeView/DebugSubsectionRecord.h"
21 #include "llvm/DebugInfo/CodeView/GlobalTypeTableBuilder.h"
22 #include "llvm/DebugInfo/CodeView/LazyRandomTypeCollection.h"
23 #include "llvm/DebugInfo/CodeView/MergingTypeTableBuilder.h"
24 #include "llvm/DebugInfo/CodeView/RecordName.h"
25 #include "llvm/DebugInfo/CodeView/SymbolDeserializer.h"
26 #include "llvm/DebugInfo/CodeView/SymbolRecordHelpers.h"
27 #include "llvm/DebugInfo/CodeView/SymbolSerializer.h"
28 #include "llvm/DebugInfo/CodeView/TypeDeserializer.h"
29 #include "llvm/DebugInfo/CodeView/TypeDumpVisitor.h"
30 #include "llvm/DebugInfo/CodeView/TypeIndexDiscovery.h"
31 #include "llvm/DebugInfo/CodeView/TypeStreamMerger.h"
32 #include "llvm/DebugInfo/MSF/MSFBuilder.h"
33 #include "llvm/DebugInfo/MSF/MSFCommon.h"
34 #include "llvm/DebugInfo/PDB/GenericError.h"
35 #include "llvm/DebugInfo/PDB/Native/DbiModuleDescriptorBuilder.h"
36 #include "llvm/DebugInfo/PDB/Native/DbiStream.h"
37 #include "llvm/DebugInfo/PDB/Native/DbiStreamBuilder.h"
38 #include "llvm/DebugInfo/PDB/Native/GSIStreamBuilder.h"
39 #include "llvm/DebugInfo/PDB/Native/InfoStream.h"
40 #include "llvm/DebugInfo/PDB/Native/InfoStreamBuilder.h"
41 #include "llvm/DebugInfo/PDB/Native/NativeSession.h"
42 #include "llvm/DebugInfo/PDB/Native/PDBFile.h"
43 #include "llvm/DebugInfo/PDB/Native/PDBFileBuilder.h"
44 #include "llvm/DebugInfo/PDB/Native/PDBStringTableBuilder.h"
45 #include "llvm/DebugInfo/PDB/Native/TpiHashing.h"
46 #include "llvm/DebugInfo/PDB/Native/TpiStream.h"
47 #include "llvm/DebugInfo/PDB/Native/TpiStreamBuilder.h"
48 #include "llvm/DebugInfo/PDB/PDB.h"
49 #include "llvm/Object/COFF.h"
50 #include "llvm/Object/CVDebugRecord.h"
51 #include "llvm/Support/BinaryByteStream.h"
52 #include "llvm/Support/Endian.h"
53 #include "llvm/Support/Errc.h"
54 #include "llvm/Support/FormatVariadic.h"
55 #include "llvm/Support/JamCRC.h"
56 #include "llvm/Support/Parallel.h"
57 #include "llvm/Support/Path.h"
58 #include "llvm/Support/ScopedPrinter.h"
62 using namespace lld::coff;
64 using namespace llvm::codeview;
66 using llvm::object::coff_section;
68 static ExitOnError ExitOnErr;
70 static Timer TotalPdbLinkTimer("PDB Emission (Cumulative)", Timer::root());
72 static Timer AddObjectsTimer("Add Objects", TotalPdbLinkTimer);
73 static Timer TypeMergingTimer("Type Merging", AddObjectsTimer);
74 static Timer SymbolMergingTimer("Symbol Merging", AddObjectsTimer);
75 static Timer GlobalsLayoutTimer("Globals Stream Layout", TotalPdbLinkTimer);
76 static Timer TpiStreamLayoutTimer("TPI Stream Layout", TotalPdbLinkTimer);
77 static Timer DiskCommitTimer("Commit to Disk", TotalPdbLinkTimer);
80 /// Map from type index and item index in a type server PDB to the
81 /// corresponding index in the destination PDB.
83 SmallVector<TypeIndex, 0> TPIMap;
84 SmallVector<TypeIndex, 0> IPIMap;
85 bool IsTypeServerMap = false;
86 bool IsPrecompiledTypeMap = false;
95 PDBLinker(SymbolTable *Symtab)
96 : Alloc(), Symtab(Symtab), Builder(Alloc), TypeTable(Alloc),
97 IDTable(Alloc), GlobalTypeTable(Alloc), GlobalIDTable(Alloc) {
98 // This isn't strictly necessary, but link.exe usually puts an empty string
99 // as the first "valid" string in the string table, so we do the same in
100 // order to maintain as much byte-for-byte compatibility as possible.
101 PDBStrTab.insert("");
104 /// Emit the basic PDB structure: initial streams, headers, etc.
105 void initialize(llvm::codeview::DebugInfo *BuildId);
107 /// Add natvis files specified on the command line.
108 void addNatvisFiles();
110 /// Link CodeView from each object file in the symbol table into the PDB.
111 void addObjectsToPDB();
113 /// Link CodeView from a single object file into the target (output) PDB.
114 /// When a precompiled headers object is linked, its TPI map might be provided
116 void addObjFile(ObjFile *File, CVIndexMap *ExternIndexMap = nullptr);
118 /// Produce a mapping from the type and item indices used in the object
119 /// file to those in the destination PDB.
121 /// If the object file uses a type server PDB (compiled with /Zi), merge TPI
122 /// and IPI from the type server PDB and return a map for it. Each unique type
123 /// server PDB is merged at most once, so this may return an existing index
126 /// If the object does not use a type server PDB (compiled with /Z7), we merge
127 /// all the type and item records from the .debug$S stream and fill in the
128 /// caller-provided ObjectIndexMap.
129 Expected<const CVIndexMap &> mergeDebugT(ObjFile *File,
130 CVIndexMap *ObjectIndexMap);
132 /// Reads and makes available a PDB.
133 Expected<const CVIndexMap &> maybeMergeTypeServerPDB(ObjFile *File,
134 const CVType &FirstType);
136 /// Merges a precompiled headers TPI map into the current TPI map. The
137 /// precompiled headers object will also be loaded and remapped in the
139 Expected<const CVIndexMap &>
140 mergeInPrecompHeaderObj(ObjFile *File, const CVType &FirstType,
141 CVIndexMap *ObjectIndexMap);
143 /// Reads and makes available a precompiled headers object.
145 /// This is a requirement for objects compiled with cl.exe /Yu. In that
146 /// case, the referenced object (which was compiled with /Yc) has to be loaded
147 /// first. This is mainly because the current object's TPI stream has external
148 /// references to the precompiled headers object.
150 /// If the precompiled headers object was already loaded, this function will
151 /// simply return its (remapped) TPI map.
152 Expected<const CVIndexMap &> aquirePrecompObj(ObjFile *File,
153 PrecompRecord Precomp);
155 /// Adds a precompiled headers object signature -> TPI mapping.
156 std::pair<CVIndexMap &, bool /*already there*/>
157 registerPrecompiledHeaders(uint32_t Signature);
159 void mergeSymbolRecords(ObjFile *File, const CVIndexMap &IndexMap,
160 std::vector<ulittle32_t *> &StringTableRefs,
161 BinaryStreamRef SymData);
163 /// Add the section map and section contributions to the PDB.
164 void addSections(ArrayRef<OutputSection *> OutputSections,
165 ArrayRef<uint8_t> SectionTable);
167 /// Get the type table or the global type table if /DEBUG:GHASH is enabled.
168 TypeCollection &getTypeTable() {
169 if (Config->DebugGHashes)
170 return GlobalTypeTable;
174 /// Get the ID table or the global ID table if /DEBUG:GHASH is enabled.
175 TypeCollection &getIDTable() {
176 if (Config->DebugGHashes)
177 return GlobalIDTable;
181 /// Write the PDB to disk and store the Guid generated for it in *Guid.
182 void commit(codeview::GUID *Guid);
185 BumpPtrAllocator Alloc;
189 pdb::PDBFileBuilder Builder;
191 /// Type records that will go into the PDB TPI stream.
192 MergingTypeTableBuilder TypeTable;
194 /// Item records that will go into the PDB IPI stream.
195 MergingTypeTableBuilder IDTable;
197 /// Type records that will go into the PDB TPI stream (for /DEBUG:GHASH)
198 GlobalTypeTableBuilder GlobalTypeTable;
200 /// Item records that will go into the PDB IPI stream (for /DEBUG:GHASH)
201 GlobalTypeTableBuilder GlobalIDTable;
203 /// PDBs use a single global string table for filenames in the file checksum
205 DebugStringTableSubsection PDBStrTab;
207 llvm::SmallString<128> NativePath;
209 /// A list of other PDBs which are loaded during the linking process and which
210 /// we need to keep around since the linking operation may reference pointers
211 /// inside of these PDBs.
212 llvm::SmallVector<std::unique_ptr<pdb::NativeSession>, 2> LoadedPDBs;
214 std::vector<pdb::SecMapEntry> SectionMap;
216 /// Type index mappings of type server PDBs that we've loaded so far.
217 std::map<codeview::GUID, CVIndexMap> TypeServerIndexMappings;
219 /// Type index mappings of precompiled objects type map that we've loaded so
221 std::map<uint32_t, CVIndexMap> PrecompTypeIndexMappings;
223 /// List of TypeServer PDBs which cannot be loaded.
224 /// Cached to prevent repeated load attempts.
225 std::map<codeview::GUID, std::string> MissingTypeServerPDBs;
228 class DebugSHandler {
231 /// The object file whose .debug$S sections we're processing.
234 /// The result of merging type indices.
235 const CVIndexMap &IndexMap;
237 /// The DEBUG_S_STRINGTABLE subsection. These strings are referred to by
238 /// index from other records in the .debug$S section. All of these strings
239 /// need to be added to the global PDB string table, and all references to
240 /// these strings need to have their indices re-written to refer to the
241 /// global PDB string table.
242 DebugStringTableSubsectionRef CVStrTab;
244 /// The DEBUG_S_FILECHKSMS subsection. As above, these are referred to
245 /// by other records in the .debug$S section and need to be merged into the
247 DebugChecksumsSubsectionRef Checksums;
249 /// The DEBUG_S_FRAMEDATA subsection(s). There can be more than one of
250 /// these and they need not appear in any specific order. However, they
251 /// contain string table references which need to be re-written, so we
252 /// collect them all here and re-write them after all subsections have been
253 /// discovered and processed.
254 std::vector<DebugFrameDataSubsectionRef> NewFpoFrames;
256 /// Pointers to raw memory that we determine have string table references
257 /// that need to be re-written. We first process all .debug$S subsections
258 /// to ensure that we can handle subsections written in any order, building
259 /// up this list as we go. At the end, we use the string table (which must
260 /// have been discovered by now else it is an error) to re-write these
262 std::vector<ulittle32_t *> StringTableReferences;
265 DebugSHandler(PDBLinker &Linker, ObjFile &File, const CVIndexMap &IndexMap)
266 : Linker(Linker), File(File), IndexMap(IndexMap) {}
268 void handleDebugS(lld::coff::SectionChunk &DebugS);
273 // Visual Studio's debugger requires absolute paths in various places in the
274 // PDB to work without additional configuration:
275 // https://docs.microsoft.com/en-us/visualstudio/debugger/debug-source-files-common-properties-solution-property-pages-dialog-box
276 static void pdbMakeAbsolute(SmallVectorImpl<char> &FileName) {
277 // The default behavior is to produce paths that are valid within the context
278 // of the machine that you perform the link on. If the linker is running on
279 // a POSIX system, we will output absolute POSIX paths. If the linker is
280 // running on a Windows system, we will output absolute Windows paths. If the
281 // user desires any other kind of behavior, they should explicitly pass
282 // /pdbsourcepath, in which case we will treat the exact string the user
283 // passed in as the gospel and not normalize, canonicalize it.
284 if (sys::path::is_absolute(FileName, sys::path::Style::windows) ||
285 sys::path::is_absolute(FileName, sys::path::Style::posix))
288 // It's not absolute in any path syntax. Relative paths necessarily refer to
289 // the local file system, so we can make it native without ending up with a
291 sys::path::native(FileName);
292 if (Config->PDBSourcePath.empty()) {
293 sys::fs::make_absolute(FileName);
296 // Only apply native and dot removal to the relative file path. We want to
297 // leave the path the user specified untouched since we assume they specified
299 sys::path::remove_dots(FileName, /*remove_dot_dots=*/true);
301 SmallString<128> AbsoluteFileName = Config->PDBSourcePath;
302 sys::path::append(AbsoluteFileName, FileName);
303 FileName = std::move(AbsoluteFileName);
306 static SectionChunk *findByName(ArrayRef<SectionChunk *> Sections,
308 for (SectionChunk *C : Sections)
309 if (C->getSectionName() == Name)
314 static ArrayRef<uint8_t> consumeDebugMagic(ArrayRef<uint8_t> Data,
316 // First 4 bytes are section magic.
318 fatal(SecName + " too short");
319 if (support::endian::read32le(Data.data()) != COFF::DEBUG_SECTION_MAGIC)
320 fatal(SecName + " has an invalid magic");
321 return Data.slice(4);
324 static ArrayRef<uint8_t> getDebugSection(ObjFile *File, StringRef SecName) {
325 if (SectionChunk *Sec = findByName(File->getDebugChunks(), SecName))
326 return consumeDebugMagic(Sec->getContents(), SecName);
330 // A COFF .debug$H section is currently a clang extension. This function checks
331 // if a .debug$H section is in a format that we expect / understand, so that we
332 // can ignore any sections which are coincidentally also named .debug$H but do
333 // not contain a format we recognize.
334 static bool canUseDebugH(ArrayRef<uint8_t> DebugH) {
335 if (DebugH.size() < sizeof(object::debug_h_header))
338 reinterpret_cast<const object::debug_h_header *>(DebugH.data());
339 DebugH = DebugH.drop_front(sizeof(object::debug_h_header));
340 return Header->Magic == COFF::DEBUG_HASHES_SECTION_MAGIC &&
341 Header->Version == 0 &&
342 Header->HashAlgorithm == uint16_t(GlobalTypeHashAlg::SHA1_8) &&
343 (DebugH.size() % 8 == 0);
346 static Optional<ArrayRef<uint8_t>> getDebugH(ObjFile *File) {
347 SectionChunk *Sec = findByName(File->getDebugChunks(), ".debug$H");
350 ArrayRef<uint8_t> Contents = Sec->getContents();
351 if (!canUseDebugH(Contents))
356 static ArrayRef<GloballyHashedType>
357 getHashesFromDebugH(ArrayRef<uint8_t> DebugH) {
358 assert(canUseDebugH(DebugH));
360 DebugH = DebugH.drop_front(sizeof(object::debug_h_header));
361 uint32_t Count = DebugH.size() / sizeof(GloballyHashedType);
362 return {reinterpret_cast<const GloballyHashedType *>(DebugH.data()), Count};
365 static void addTypeInfo(pdb::TpiStreamBuilder &TpiBuilder,
366 TypeCollection &TypeTable) {
367 // Start the TPI or IPI stream header.
368 TpiBuilder.setVersionHeader(pdb::PdbTpiV80);
370 // Flatten the in memory type table and hash each type.
371 TypeTable.ForEachRecord([&](TypeIndex TI, const CVType &Type) {
372 auto Hash = pdb::hashTypeRecord(Type);
373 if (auto E = Hash.takeError())
374 fatal("type hashing error");
375 TpiBuilder.addTypeRecord(Type.RecordData, *Hash);
379 // OBJs usually start their symbol stream with a S_OBJNAME record. This record
380 // also contains the signature/key of the current PCH session. The signature
381 // must be same for all objects which depend on the precompiled object.
382 // Recompiling the precompiled headers will generate a new PCH key and thus
383 // invalidate all the dependent objects.
384 static uint32_t extractPCHSignature(ObjFile *File) {
385 auto DbgIt = find_if(File->getDebugChunks(), [](SectionChunk *C) {
386 return C->getSectionName() == ".debug$S";
391 ArrayRef<uint8_t> Contents =
392 consumeDebugMagic((*DbgIt)->getContents(), ".debug$S");
393 DebugSubsectionArray Subsections;
394 BinaryStreamReader Reader(Contents, support::little);
395 ExitOnErr(Reader.readArray(Subsections, Contents.size()));
397 for (const DebugSubsectionRecord &SS : Subsections) {
398 if (SS.kind() != DebugSubsectionKind::Symbols)
401 // If it's there, the S_OBJNAME record shall come first in the stream.
402 Expected<CVSymbol> Sym = readSymbolFromStream(SS.getRecordData(), 0);
404 consumeError(Sym.takeError());
407 if (auto ObjName = SymbolDeserializer::deserializeAs<ObjNameSym>(Sym.get()))
408 return ObjName->Signature;
413 Expected<const CVIndexMap &>
414 PDBLinker::mergeDebugT(ObjFile *File, CVIndexMap *ObjectIndexMap) {
415 ScopedTimer T(TypeMergingTimer);
417 bool IsPrecompiledHeader = false;
419 ArrayRef<uint8_t> Data = getDebugSection(File, ".debug$T");
421 // Try again, Microsoft precompiled headers use .debug$P instead of
423 Data = getDebugSection(File, ".debug$P");
424 IsPrecompiledHeader = true;
427 return *ObjectIndexMap; // no debug info
429 // Precompiled headers objects need to save the index map for further
430 // reference by other objects which use the precompiled headers.
431 if (IsPrecompiledHeader) {
432 uint32_t PCHSignature = extractPCHSignature(File);
433 if (PCHSignature == 0)
434 fatal("No signature found for the precompiled headers OBJ (" +
435 File->getName() + ")");
437 // When a precompiled headers object comes first on the command-line, we
438 // update the mapping here. Otherwise, if an object referencing the
439 // precompiled headers object comes first, the mapping is created in
440 // aquirePrecompObj(), thus we would skip this block.
441 if (!ObjectIndexMap->IsPrecompiledTypeMap) {
442 auto R = registerPrecompiledHeaders(PCHSignature);
445 "A precompiled headers OBJ with the same signature was already "
447 File->getName() + ")");
449 ObjectIndexMap = &R.first;
453 BinaryByteStream Stream(Data, support::little);
455 BinaryStreamReader Reader(Stream);
456 if (auto EC = Reader.readArray(Types, Reader.getLength()))
457 fatal("Reader::readArray failed: " + toString(std::move(EC)));
459 auto FirstType = Types.begin();
460 if (FirstType == Types.end())
461 return *ObjectIndexMap;
463 if (FirstType->kind() == LF_TYPESERVER2) {
464 // Look through type servers. If we've already seen this type server,
465 // don't merge any type information.
466 return maybeMergeTypeServerPDB(File, *FirstType);
467 } else if (FirstType->kind() == LF_PRECOMP) {
468 // This object was compiled with /Yu, so process the corresponding
469 // precompiled headers object (/Yc) first. Some type indices in the current
470 // object are referencing data in the precompiled headers object, so we need
471 // both to be loaded.
472 auto E = mergeInPrecompHeaderObj(File, *FirstType, ObjectIndexMap);
474 return E.takeError();
476 // Drop LF_PRECOMP record from the input stream, as it needs to be replaced
477 // with the precompiled headers object type stream.
478 // Note that we can't just call Types.drop_front(), as we explicitly want to
479 // rebase the stream.
480 Types.setUnderlyingStream(
481 Types.getUnderlyingStream().drop_front(FirstType->RecordData.size()));
484 // Fill in the temporary, caller-provided ObjectIndexMap.
485 if (Config->DebugGHashes) {
486 ArrayRef<GloballyHashedType> Hashes;
487 std::vector<GloballyHashedType> OwnedHashes;
488 if (Optional<ArrayRef<uint8_t>> DebugH = getDebugH(File))
489 Hashes = getHashesFromDebugH(*DebugH);
491 OwnedHashes = GloballyHashedType::hashTypes(Types);
492 Hashes = OwnedHashes;
495 if (auto Err = mergeTypeAndIdRecords(GlobalIDTable, GlobalTypeTable,
496 ObjectIndexMap->TPIMap, Types, Hashes,
498 fatal("codeview::mergeTypeAndIdRecords failed: " +
499 toString(std::move(Err)));
502 mergeTypeAndIdRecords(IDTable, TypeTable, ObjectIndexMap->TPIMap,
503 Types, File->PCHSignature))
504 fatal("codeview::mergeTypeAndIdRecords failed: " +
505 toString(std::move(Err)));
507 return *ObjectIndexMap;
510 static Expected<std::unique_ptr<pdb::NativeSession>>
511 tryToLoadPDB(const codeview::GUID &GuidFromObj, StringRef TSPath) {
512 // Ensure the file exists before anything else. We want to return ENOENT,
513 // "file not found", even if the path points to a removable device (in which
514 // case the return message would be EAGAIN, "resource unavailable try again")
515 if (!llvm::sys::fs::exists(TSPath))
516 return errorCodeToError(std::error_code(ENOENT, std::generic_category()));
518 ErrorOr<std::unique_ptr<MemoryBuffer>> MBOrErr = MemoryBuffer::getFile(
519 TSPath, /*FileSize=*/-1, /*RequiresNullTerminator=*/false);
521 return errorCodeToError(MBOrErr.getError());
523 std::unique_ptr<pdb::IPDBSession> ThisSession;
524 if (auto EC = pdb::NativeSession::createFromPdb(
525 MemoryBuffer::getMemBuffer(Driver->takeBuffer(std::move(*MBOrErr)),
526 /*RequiresNullTerminator=*/false),
528 return std::move(EC);
530 std::unique_ptr<pdb::NativeSession> NS(
531 static_cast<pdb::NativeSession *>(ThisSession.release()));
532 pdb::PDBFile &File = NS->getPDBFile();
533 auto ExpectedInfo = File.getPDBInfoStream();
534 // All PDB Files should have an Info stream.
536 return ExpectedInfo.takeError();
538 // Just because a file with a matching name was found and it was an actual
539 // PDB file doesn't mean it matches. For it to match the InfoStream's GUID
540 // must match the GUID specified in the TypeServer2 record.
541 if (ExpectedInfo->getGuid() != GuidFromObj)
542 return make_error<pdb::PDBError>(pdb::pdb_error_code::signature_out_of_date);
544 return std::move(NS);
547 Expected<const CVIndexMap &>
548 PDBLinker::maybeMergeTypeServerPDB(ObjFile *File, const CVType &FirstType) {
549 TypeServer2Record TS;
551 TypeDeserializer::deserializeAs(const_cast<CVType &>(FirstType), TS))
552 fatal("error reading record: " + toString(std::move(EC)));
554 const codeview::GUID &TSId = TS.getGuid();
555 StringRef TSPath = TS.getName();
557 // First, check if the PDB has previously failed to load.
558 auto PrevErr = MissingTypeServerPDBs.find(TSId);
559 if (PrevErr != MissingTypeServerPDBs.end())
560 return createFileError(
562 make_error<StringError>(PrevErr->second, inconvertibleErrorCode()));
564 // Second, check if we already loaded a PDB with this GUID. Return the type
565 // index mapping if we have it.
566 auto Insertion = TypeServerIndexMappings.insert({TSId, CVIndexMap()});
567 CVIndexMap &IndexMap = Insertion.first->second;
568 if (!Insertion.second)
571 // Mark this map as a type server map.
572 IndexMap.IsTypeServerMap = true;
574 // Check for a PDB at:
575 // 1. The given file path
576 // 2. Next to the object file or archive file
577 auto ExpectedSession = handleExpected(
578 tryToLoadPDB(TSId, TSPath),
580 StringRef LocalPath =
581 !File->ParentName.empty() ? File->ParentName : File->getName();
582 SmallString<128> Path = sys::path::parent_path(LocalPath);
583 // Currently, type server PDBs are only created by cl, which only runs
584 // on Windows, so we can assume type server paths are Windows style.
586 Path, sys::path::filename(TSPath, sys::path::Style::windows));
587 return tryToLoadPDB(TSId, Path);
589 [&](std::unique_ptr<ECError> EC) -> Error {
590 auto SysErr = EC->convertToErrorCode();
591 // Only re-try loading if the previous error was "No such file or
593 if (SysErr.category() == std::generic_category() &&
594 SysErr.value() == ENOENT)
595 return Error::success();
596 return Error(std::move(EC));
599 if (auto E = ExpectedSession.takeError()) {
600 TypeServerIndexMappings.erase(TSId);
602 // Flatten the error to a string, for later display, if the error occurs
603 // again on the same PDB.
605 raw_string_ostream S(ErrMsg);
607 MissingTypeServerPDBs.emplace(TSId, S.str());
609 return createFileError(TSPath, std::move(E));
612 pdb::NativeSession *Session = ExpectedSession->get();
614 // Keep a strong reference to this PDB, so that it's safe to hold pointers
616 LoadedPDBs.push_back(std::move(*ExpectedSession));
618 auto ExpectedTpi = Session->getPDBFile().getPDBTpiStream();
619 if (auto E = ExpectedTpi.takeError())
620 fatal("Type server does not have TPI stream: " + toString(std::move(E)));
621 auto ExpectedIpi = Session->getPDBFile().getPDBIpiStream();
622 if (auto E = ExpectedIpi.takeError())
623 fatal("Type server does not have TPI stream: " + toString(std::move(E)));
625 if (Config->DebugGHashes) {
626 // PDBs do not actually store global hashes, so when merging a type server
627 // PDB we have to synthesize global hashes. To do this, we first synthesize
628 // global hashes for the TPI stream, since it is independent, then we
629 // synthesize hashes for the IPI stream, using the hashes for the TPI stream
631 auto TpiHashes = GloballyHashedType::hashTypes(ExpectedTpi->typeArray());
633 GloballyHashedType::hashIds(ExpectedIpi->typeArray(), TpiHashes);
635 Optional<uint32_t> EndPrecomp;
636 // Merge TPI first, because the IPI stream will reference type indices.
637 if (auto Err = mergeTypeRecords(GlobalTypeTable, IndexMap.TPIMap,
638 ExpectedTpi->typeArray(), TpiHashes, EndPrecomp))
639 fatal("codeview::mergeTypeRecords failed: " + toString(std::move(Err)));
643 mergeIdRecords(GlobalIDTable, IndexMap.TPIMap, IndexMap.IPIMap,
644 ExpectedIpi->typeArray(), IpiHashes))
645 fatal("codeview::mergeIdRecords failed: " + toString(std::move(Err)));
647 // Merge TPI first, because the IPI stream will reference type indices.
648 if (auto Err = mergeTypeRecords(TypeTable, IndexMap.TPIMap,
649 ExpectedTpi->typeArray()))
650 fatal("codeview::mergeTypeRecords failed: " + toString(std::move(Err)));
653 if (auto Err = mergeIdRecords(IDTable, IndexMap.TPIMap, IndexMap.IPIMap,
654 ExpectedIpi->typeArray()))
655 fatal("codeview::mergeIdRecords failed: " + toString(std::move(Err)));
661 Expected<const CVIndexMap &>
662 PDBLinker::mergeInPrecompHeaderObj(ObjFile *File, const CVType &FirstType,
663 CVIndexMap *ObjectIndexMap) {
664 PrecompRecord Precomp;
665 if (auto EC = TypeDeserializer::deserializeAs(const_cast<CVType &>(FirstType),
667 fatal("error reading record: " + toString(std::move(EC)));
669 auto E = aquirePrecompObj(File, Precomp);
671 return E.takeError();
673 const CVIndexMap &PrecompIndexMap = *E;
674 assert(PrecompIndexMap.IsPrecompiledTypeMap);
676 if (PrecompIndexMap.TPIMap.empty())
677 return PrecompIndexMap;
679 assert(Precomp.getStartTypeIndex() == TypeIndex::FirstNonSimpleIndex);
680 assert(Precomp.getTypesCount() <= PrecompIndexMap.TPIMap.size());
681 // Use the previously remapped index map from the precompiled headers.
682 ObjectIndexMap->TPIMap.append(PrecompIndexMap.TPIMap.begin(),
683 PrecompIndexMap.TPIMap.begin() +
684 Precomp.getTypesCount());
685 return *ObjectIndexMap;
688 static bool equals_path(StringRef path1, StringRef path2) {
690 return path1.equals_lower(path2);
692 return path1.equals(path2);
696 // Find by name an OBJ provided on the command line
697 static ObjFile *findObjByName(StringRef FileNameOnly) {
698 SmallString<128> CurrentPath;
700 for (ObjFile *F : ObjFile::Instances) {
701 StringRef CurrentFileName = sys::path::filename(F->getName());
703 // Compare based solely on the file name (link.exe behavior)
704 if (equals_path(CurrentFileName, FileNameOnly))
710 std::pair<CVIndexMap &, bool /*already there*/>
711 PDBLinker::registerPrecompiledHeaders(uint32_t Signature) {
712 auto Insertion = PrecompTypeIndexMappings.insert({Signature, CVIndexMap()});
713 CVIndexMap &IndexMap = Insertion.first->second;
714 if (!Insertion.second)
715 return {IndexMap, true};
716 // Mark this map as a precompiled types map.
717 IndexMap.IsPrecompiledTypeMap = true;
718 return {IndexMap, false};
721 Expected<const CVIndexMap &>
722 PDBLinker::aquirePrecompObj(ObjFile *File, PrecompRecord Precomp) {
723 // First, check if we already loaded the precompiled headers object with this
724 // signature. Return the type index mapping if we've already seen it.
725 auto R = registerPrecompiledHeaders(Precomp.getSignature());
729 CVIndexMap &IndexMap = R.first;
731 // Cross-compile warning: given that Clang doesn't generate LF_PRECOMP
732 // records, we assume the OBJ comes from a Windows build of cl.exe. Thusly,
733 // the paths embedded in the OBJs are in the Windows format.
734 SmallString<128> PrecompFileName = sys::path::filename(
735 Precomp.getPrecompFilePath(), sys::path::Style::windows);
737 // link.exe requires that a precompiled headers object must always be provided
738 // on the command-line, even if that's not necessary.
739 auto PrecompFile = findObjByName(PrecompFileName);
741 return createFileError(
742 PrecompFileName.str(),
743 make_error<pdb::PDBError>(pdb::pdb_error_code::external_cmdline_ref));
745 addObjFile(PrecompFile, &IndexMap);
747 if (!PrecompFile->PCHSignature)
748 fatal(PrecompFile->getName() + " is not a precompiled headers object");
750 if (Precomp.getSignature() != PrecompFile->PCHSignature.getValueOr(0))
751 return createFileError(
752 Precomp.getPrecompFilePath().str(),
753 make_error<pdb::PDBError>(pdb::pdb_error_code::signature_out_of_date));
758 static bool remapTypeIndex(TypeIndex &TI, ArrayRef<TypeIndex> TypeIndexMap) {
761 if (TI.toArrayIndex() >= TypeIndexMap.size())
763 TI = TypeIndexMap[TI.toArrayIndex()];
767 static void remapTypesInSymbolRecord(ObjFile *File, SymbolKind SymKind,
768 MutableArrayRef<uint8_t> RecordBytes,
769 const CVIndexMap &IndexMap,
770 ArrayRef<TiReference> TypeRefs) {
771 MutableArrayRef<uint8_t> Contents =
772 RecordBytes.drop_front(sizeof(RecordPrefix));
773 for (const TiReference &Ref : TypeRefs) {
774 unsigned ByteSize = Ref.Count * sizeof(TypeIndex);
775 if (Contents.size() < Ref.Offset + ByteSize)
776 fatal("symbol record too short");
778 // This can be an item index or a type index. Choose the appropriate map.
779 ArrayRef<TypeIndex> TypeOrItemMap = IndexMap.TPIMap;
780 bool IsItemIndex = Ref.Kind == TiRefKind::IndexRef;
781 if (IsItemIndex && IndexMap.IsTypeServerMap)
782 TypeOrItemMap = IndexMap.IPIMap;
784 MutableArrayRef<TypeIndex> TIs(
785 reinterpret_cast<TypeIndex *>(Contents.data() + Ref.Offset), Ref.Count);
786 for (TypeIndex &TI : TIs) {
787 if (!remapTypeIndex(TI, TypeOrItemMap)) {
788 log("ignoring symbol record of kind 0x" + utohexstr(SymKind) + " in " +
789 File->getName() + " with bad " + (IsItemIndex ? "item" : "type") +
790 " index 0x" + utohexstr(TI.getIndex()));
791 TI = TypeIndex(SimpleTypeKind::NotTranslated);
799 recordStringTableReferenceAtOffset(MutableArrayRef<uint8_t> Contents,
801 std::vector<ulittle32_t *> &StrTableRefs) {
803 Contents.drop_front(Offset).take_front(sizeof(support::ulittle32_t));
804 ulittle32_t *Index = reinterpret_cast<ulittle32_t *>(Contents.data());
805 StrTableRefs.push_back(Index);
809 recordStringTableReferences(SymbolKind Kind, MutableArrayRef<uint8_t> Contents,
810 std::vector<ulittle32_t *> &StrTableRefs) {
811 // For now we only handle S_FILESTATIC, but we may need the same logic for
812 // S_DEFRANGE and S_DEFRANGE_SUBFIELD. However, I cannot seem to generate any
813 // PDBs that contain these types of records, so because of the uncertainty
814 // they are omitted here until we can prove that it's necessary.
816 case SymbolKind::S_FILESTATIC:
817 // FileStaticSym::ModFileOffset
818 recordStringTableReferenceAtOffset(Contents, 8, StrTableRefs);
820 case SymbolKind::S_DEFRANGE:
821 case SymbolKind::S_DEFRANGE_SUBFIELD:
822 log("Not fixing up string table reference in S_DEFRANGE / "
823 "S_DEFRANGE_SUBFIELD record");
830 static SymbolKind symbolKind(ArrayRef<uint8_t> RecordData) {
831 const RecordPrefix *Prefix =
832 reinterpret_cast<const RecordPrefix *>(RecordData.data());
833 return static_cast<SymbolKind>(uint16_t(Prefix->RecordKind));
836 /// MSVC translates S_PROC_ID_END to S_END, and S_[LG]PROC32_ID to S_[LG]PROC32
837 static void translateIdSymbols(MutableArrayRef<uint8_t> &RecordData,
838 TypeCollection &IDTable) {
839 RecordPrefix *Prefix = reinterpret_cast<RecordPrefix *>(RecordData.data());
841 SymbolKind Kind = symbolKind(RecordData);
843 if (Kind == SymbolKind::S_PROC_ID_END) {
844 Prefix->RecordKind = SymbolKind::S_END;
848 // In an object file, GPROC32_ID has an embedded reference which refers to the
849 // single object file type index namespace. This has already been translated
850 // to the PDB file's ID stream index space, but we need to convert this to a
851 // symbol that refers to the type stream index space. So we remap again from
852 // ID index space to type index space.
853 if (Kind == SymbolKind::S_GPROC32_ID || Kind == SymbolKind::S_LPROC32_ID) {
854 SmallVector<TiReference, 1> Refs;
855 auto Content = RecordData.drop_front(sizeof(RecordPrefix));
856 CVSymbol Sym(Kind, RecordData);
857 discoverTypeIndicesInSymbol(Sym, Refs);
858 assert(Refs.size() == 1);
859 assert(Refs.front().Count == 1);
862 reinterpret_cast<TypeIndex *>(Content.data() + Refs[0].Offset);
863 // `TI` is the index of a FuncIdRecord or MemberFuncIdRecord which lives in
864 // the IPI stream, whose `FunctionType` member refers to the TPI stream.
865 // Note that LF_FUNC_ID and LF_MEMFUNC_ID have the same record layout, and
866 // in both cases we just need the second type index.
867 if (!TI->isSimple() && !TI->isNoneType()) {
868 CVType FuncIdData = IDTable.getType(*TI);
869 SmallVector<TypeIndex, 2> Indices;
870 discoverTypeIndices(FuncIdData, Indices);
871 assert(Indices.size() == 2);
875 Kind = (Kind == SymbolKind::S_GPROC32_ID) ? SymbolKind::S_GPROC32
876 : SymbolKind::S_LPROC32;
877 Prefix->RecordKind = uint16_t(Kind);
881 /// Copy the symbol record. In a PDB, symbol records must be 4 byte aligned.
882 /// The object file may not be aligned.
883 static MutableArrayRef<uint8_t>
884 copyAndAlignSymbol(const CVSymbol &Sym, MutableArrayRef<uint8_t> &AlignedMem) {
885 size_t Size = alignTo(Sym.length(), alignOf(CodeViewContainer::Pdb));
886 assert(Size >= 4 && "record too short");
887 assert(Size <= MaxRecordLength && "record too long");
888 assert(AlignedMem.size() >= Size && "didn't preallocate enough");
890 // Copy the symbol record and zero out any padding bytes.
891 MutableArrayRef<uint8_t> NewData = AlignedMem.take_front(Size);
892 AlignedMem = AlignedMem.drop_front(Size);
893 memcpy(NewData.data(), Sym.data().data(), Sym.length());
894 memset(NewData.data() + Sym.length(), 0, Size - Sym.length());
896 // Update the record prefix length. It should point to the beginning of the
898 auto *Prefix = reinterpret_cast<RecordPrefix *>(NewData.data());
899 Prefix->RecordLen = Size - 2;
904 ulittle32_t PtrParent;
909 ScopeRecord *OpeningRecord;
910 uint32_t ScopeOffset;
913 static void scopeStackOpen(SmallVectorImpl<SymbolScope> &Stack,
914 uint32_t CurOffset, CVSymbol &Sym) {
915 assert(symbolOpensScope(Sym.kind()));
917 S.ScopeOffset = CurOffset;
918 S.OpeningRecord = const_cast<ScopeRecord *>(
919 reinterpret_cast<const ScopeRecord *>(Sym.content().data()));
920 S.OpeningRecord->PtrParent = Stack.empty() ? 0 : Stack.back().ScopeOffset;
924 static void scopeStackClose(SmallVectorImpl<SymbolScope> &Stack,
925 uint32_t CurOffset, ObjFile *File) {
927 warn("symbol scopes are not balanced in " + File->getName());
930 SymbolScope S = Stack.pop_back_val();
931 S.OpeningRecord->PtrEnd = CurOffset;
934 static bool symbolGoesInModuleStream(const CVSymbol &Sym, bool IsGlobalScope) {
935 switch (Sym.kind()) {
936 case SymbolKind::S_GDATA32:
937 case SymbolKind::S_CONSTANT:
938 // We really should not be seeing S_PROCREF and S_LPROCREF in the first place
939 // since they are synthesized by the linker in response to S_GPROC32 and
940 // S_LPROC32, but if we do see them, don't put them in the module stream I
942 case SymbolKind::S_PROCREF:
943 case SymbolKind::S_LPROCREF:
945 // S_UDT records go in the module stream if it is not a global S_UDT.
946 case SymbolKind::S_UDT:
947 return !IsGlobalScope;
948 // S_GDATA32 does not go in the module stream, but S_LDATA32 does.
949 case SymbolKind::S_LDATA32:
955 static bool symbolGoesInGlobalsStream(const CVSymbol &Sym, bool IsGlobalScope) {
956 switch (Sym.kind()) {
957 case SymbolKind::S_CONSTANT:
958 case SymbolKind::S_GDATA32:
959 // S_LDATA32 goes in both the module stream and the globals stream.
960 case SymbolKind::S_LDATA32:
961 case SymbolKind::S_GPROC32:
962 case SymbolKind::S_LPROC32:
963 // We really should not be seeing S_PROCREF and S_LPROCREF in the first place
964 // since they are synthesized by the linker in response to S_GPROC32 and
965 // S_LPROC32, but if we do see them, copy them straight through.
966 case SymbolKind::S_PROCREF:
967 case SymbolKind::S_LPROCREF:
969 // S_UDT records go in the globals stream if it is a global S_UDT.
970 case SymbolKind::S_UDT:
971 return IsGlobalScope;
977 static void addGlobalSymbol(pdb::GSIStreamBuilder &Builder, uint16_t ModIndex,
978 unsigned SymOffset, const CVSymbol &Sym) {
979 switch (Sym.kind()) {
980 case SymbolKind::S_CONSTANT:
981 case SymbolKind::S_UDT:
982 case SymbolKind::S_GDATA32:
983 case SymbolKind::S_LDATA32:
984 case SymbolKind::S_PROCREF:
985 case SymbolKind::S_LPROCREF:
986 Builder.addGlobalSymbol(Sym);
988 case SymbolKind::S_GPROC32:
989 case SymbolKind::S_LPROC32: {
990 SymbolRecordKind K = SymbolRecordKind::ProcRefSym;
991 if (Sym.kind() == SymbolKind::S_LPROC32)
992 K = SymbolRecordKind::LocalProcRef;
994 PS.Module = ModIndex;
995 // For some reason, MSVC seems to add one to this value.
997 PS.Name = getSymbolName(Sym);
999 PS.SymOffset = SymOffset;
1000 Builder.addGlobalSymbol(PS);
1004 llvm_unreachable("Invalid symbol kind!");
1008 void PDBLinker::mergeSymbolRecords(ObjFile *File, const CVIndexMap &IndexMap,
1009 std::vector<ulittle32_t *> &StringTableRefs,
1010 BinaryStreamRef SymData) {
1011 ArrayRef<uint8_t> SymsBuffer;
1012 cantFail(SymData.readBytes(0, SymData.getLength(), SymsBuffer));
1013 SmallVector<SymbolScope, 4> Scopes;
1015 // Iterate every symbol to check if any need to be realigned, and if so, how
1016 // much space we need to allocate for them.
1017 bool NeedsRealignment = false;
1018 unsigned TotalRealignedSize = 0;
1019 auto EC = forEachCodeViewRecord<CVSymbol>(
1020 SymsBuffer, [&](CVSymbol Sym) -> llvm::Error {
1021 unsigned RealignedSize =
1022 alignTo(Sym.length(), alignOf(CodeViewContainer::Pdb));
1023 NeedsRealignment |= RealignedSize != Sym.length();
1024 TotalRealignedSize += RealignedSize;
1025 return Error::success();
1028 // If any of the symbol record lengths was corrupt, ignore them all, warn
1029 // about it, and move on.
1031 warn("corrupt symbol records in " + File->getName());
1032 consumeError(std::move(EC));
1036 // If any symbol needed realignment, allocate enough contiguous memory for
1037 // them all. Typically symbol subsections are small enough that this will not
1038 // cause fragmentation.
1039 MutableArrayRef<uint8_t> AlignedSymbolMem;
1040 if (NeedsRealignment) {
1042 Alloc.Allocate(TotalRealignedSize, alignOf(CodeViewContainer::Pdb));
1043 AlignedSymbolMem = makeMutableArrayRef(
1044 reinterpret_cast<uint8_t *>(AlignedData), TotalRealignedSize);
1047 // Iterate again, this time doing the real work.
1048 unsigned CurSymOffset = File->ModuleDBI->getNextSymbolOffset();
1049 ArrayRef<uint8_t> BulkSymbols;
1050 cantFail(forEachCodeViewRecord<CVSymbol>(
1051 SymsBuffer, [&](CVSymbol Sym) -> llvm::Error {
1052 // Align the record if required.
1053 MutableArrayRef<uint8_t> RecordBytes;
1054 if (NeedsRealignment) {
1055 RecordBytes = copyAndAlignSymbol(Sym, AlignedSymbolMem);
1056 Sym = CVSymbol(Sym.kind(), RecordBytes);
1058 // Otherwise, we can actually mutate the symbol directly, since we
1059 // copied it to apply relocations.
1060 RecordBytes = makeMutableArrayRef(
1061 const_cast<uint8_t *>(Sym.data().data()), Sym.length());
1064 // Discover type index references in the record. Skip it if we don't
1065 // know where they are.
1066 SmallVector<TiReference, 32> TypeRefs;
1067 if (!discoverTypeIndicesInSymbol(Sym, TypeRefs)) {
1068 log("ignoring unknown symbol record with kind 0x" +
1069 utohexstr(Sym.kind()));
1070 return Error::success();
1073 // Re-map all the type index references.
1074 remapTypesInSymbolRecord(File, Sym.kind(), RecordBytes, IndexMap,
1077 // An object file may have S_xxx_ID symbols, but these get converted to
1078 // "real" symbols in a PDB.
1079 translateIdSymbols(RecordBytes, getIDTable());
1080 Sym = CVSymbol(symbolKind(RecordBytes), RecordBytes);
1082 // If this record refers to an offset in the object file's string table,
1083 // add that item to the global PDB string table and re-write the index.
1084 recordStringTableReferences(Sym.kind(), RecordBytes, StringTableRefs);
1086 // Fill in "Parent" and "End" fields by maintaining a stack of scopes.
1087 if (symbolOpensScope(Sym.kind()))
1088 scopeStackOpen(Scopes, CurSymOffset, Sym);
1089 else if (symbolEndsScope(Sym.kind()))
1090 scopeStackClose(Scopes, CurSymOffset, File);
1092 // Add the symbol to the globals stream if necessary. Do this before
1093 // adding the symbol to the module since we may need to get the next
1094 // symbol offset, and writing to the module's symbol stream will update
1096 if (symbolGoesInGlobalsStream(Sym, Scopes.empty()))
1097 addGlobalSymbol(Builder.getGsiBuilder(),
1098 File->ModuleDBI->getModuleIndex(), CurSymOffset, Sym);
1100 if (symbolGoesInModuleStream(Sym, Scopes.empty())) {
1101 // Add symbols to the module in bulk. If this symbol is contiguous
1102 // with the previous run of symbols to add, combine the ranges. If
1103 // not, close the previous range of symbols and start a new one.
1104 if (Sym.data().data() == BulkSymbols.end()) {
1105 BulkSymbols = makeArrayRef(BulkSymbols.data(),
1106 BulkSymbols.size() + Sym.length());
1108 File->ModuleDBI->addSymbolsInBulk(BulkSymbols);
1109 BulkSymbols = RecordBytes;
1111 CurSymOffset += Sym.length();
1113 return Error::success();
1116 // Add any remaining symbols we've accumulated.
1117 File->ModuleDBI->addSymbolsInBulk(BulkSymbols);
1120 // Allocate memory for a .debug$S / .debug$F section and relocate it.
1121 static ArrayRef<uint8_t> relocateDebugChunk(BumpPtrAllocator &Alloc,
1122 SectionChunk &DebugChunk) {
1123 uint8_t *Buffer = Alloc.Allocate<uint8_t>(DebugChunk.getSize());
1124 assert(DebugChunk.OutputSectionOff == 0 &&
1125 "debug sections should not be in output sections");
1126 DebugChunk.readRelocTargets();
1127 DebugChunk.writeTo(Buffer);
1128 return makeArrayRef(Buffer, DebugChunk.getSize());
1131 static pdb::SectionContrib createSectionContrib(const Chunk *C, uint32_t Modi) {
1132 OutputSection *OS = C->getOutputSection();
1133 pdb::SectionContrib SC;
1134 memset(&SC, 0, sizeof(SC));
1135 SC.ISect = OS->SectionIndex;
1136 SC.Off = C->getRVA() - OS->getRVA();
1137 SC.Size = C->getSize();
1138 if (auto *SecChunk = dyn_cast<SectionChunk>(C)) {
1139 SC.Characteristics = SecChunk->Header->Characteristics;
1140 SC.Imod = SecChunk->File->ModuleDBI->getModuleIndex();
1141 ArrayRef<uint8_t> Contents = SecChunk->getContents();
1143 ArrayRef<char> CharContents = makeArrayRef(
1144 reinterpret_cast<const char *>(Contents.data()), Contents.size());
1145 CRC.update(CharContents);
1146 SC.DataCrc = CRC.getCRC();
1148 SC.Characteristics = OS->Header.Characteristics;
1149 // FIXME: When we start creating DBI for import libraries, use those here.
1152 SC.RelocCrc = 0; // FIXME
1158 translateStringTableIndex(uint32_t ObjIndex,
1159 const DebugStringTableSubsectionRef &ObjStrTable,
1160 DebugStringTableSubsection &PdbStrTable) {
1161 auto ExpectedString = ObjStrTable.getString(ObjIndex);
1162 if (!ExpectedString) {
1163 warn("Invalid string table reference");
1164 consumeError(ExpectedString.takeError());
1168 return PdbStrTable.insert(*ExpectedString);
1171 void DebugSHandler::handleDebugS(lld::coff::SectionChunk &DebugS) {
1172 DebugSubsectionArray Subsections;
1174 ArrayRef<uint8_t> RelocatedDebugContents = consumeDebugMagic(
1175 relocateDebugChunk(Linker.Alloc, DebugS), DebugS.getSectionName());
1177 BinaryStreamReader Reader(RelocatedDebugContents, support::little);
1178 ExitOnErr(Reader.readArray(Subsections, RelocatedDebugContents.size()));
1180 for (const DebugSubsectionRecord &SS : Subsections) {
1181 switch (SS.kind()) {
1182 case DebugSubsectionKind::StringTable: {
1183 assert(!CVStrTab.valid() &&
1184 "Encountered multiple string table subsections!");
1185 ExitOnErr(CVStrTab.initialize(SS.getRecordData()));
1188 case DebugSubsectionKind::FileChecksums:
1189 assert(!Checksums.valid() &&
1190 "Encountered multiple checksum subsections!");
1191 ExitOnErr(Checksums.initialize(SS.getRecordData()));
1193 case DebugSubsectionKind::Lines:
1194 // We can add the relocated line table directly to the PDB without
1195 // modification because the file checksum offsets will stay the same.
1196 File.ModuleDBI->addDebugSubsection(SS);
1198 case DebugSubsectionKind::FrameData: {
1199 // We need to re-write string table indices here, so save off all
1200 // frame data subsections until we've processed the entire list of
1201 // subsections so that we can be sure we have the string table.
1202 DebugFrameDataSubsectionRef FDS;
1203 ExitOnErr(FDS.initialize(SS.getRecordData()));
1204 NewFpoFrames.push_back(std::move(FDS));
1207 case DebugSubsectionKind::Symbols: {
1208 Linker.mergeSymbolRecords(&File, IndexMap, StringTableReferences,
1209 SS.getRecordData());
1213 // FIXME: Process the rest of the subsections.
1219 void DebugSHandler::finish() {
1220 pdb::DbiStreamBuilder &DbiBuilder = Linker.Builder.getDbiBuilder();
1222 // We should have seen all debug subsections across the entire object file now
1223 // which means that if a StringTable subsection and Checksums subsection were
1224 // present, now is the time to handle them.
1225 if (!CVStrTab.valid()) {
1226 if (Checksums.valid())
1227 fatal(".debug$S sections with a checksums subsection must also contain a "
1228 "string table subsection");
1230 if (!StringTableReferences.empty())
1231 warn("No StringTable subsection was encountered, but there are string "
1232 "table references");
1236 // Rewrite string table indices in the Fpo Data and symbol records to refer to
1237 // the global PDB string table instead of the object file string table.
1238 for (DebugFrameDataSubsectionRef &FDS : NewFpoFrames) {
1239 const ulittle32_t *Reloc = FDS.getRelocPtr();
1240 for (codeview::FrameData FD : FDS) {
1241 FD.RvaStart += *Reloc;
1243 translateStringTableIndex(FD.FrameFunc, CVStrTab, Linker.PDBStrTab);
1244 DbiBuilder.addNewFpoData(FD);
1248 for (ulittle32_t *Ref : StringTableReferences)
1249 *Ref = translateStringTableIndex(*Ref, CVStrTab, Linker.PDBStrTab);
1251 // Make a new file checksum table that refers to offsets in the PDB-wide
1252 // string table. Generally the string table subsection appears after the
1253 // checksum table, so we have to do this after looping over all the
1255 auto NewChecksums = make_unique<DebugChecksumsSubsection>(Linker.PDBStrTab);
1256 for (FileChecksumEntry &FC : Checksums) {
1257 SmallString<128> FileName =
1258 ExitOnErr(CVStrTab.getString(FC.FileNameOffset));
1259 pdbMakeAbsolute(FileName);
1260 ExitOnErr(Linker.Builder.getDbiBuilder().addModuleSourceFile(
1261 *File.ModuleDBI, FileName));
1262 NewChecksums->addChecksum(FileName, FC.Kind, FC.Checksum);
1264 File.ModuleDBI->addDebugSubsection(std::move(NewChecksums));
1267 void PDBLinker::addObjFile(ObjFile *File, CVIndexMap *ExternIndexMap) {
1268 if (File->wasProcessedForPDB())
1270 // Add a module descriptor for every object file. We need to put an absolute
1271 // path to the object into the PDB. If this is a plain object, we make its
1272 // path absolute. If it's an object in an archive, we make the archive path
1274 bool InArchive = !File->ParentName.empty();
1275 SmallString<128> Path = InArchive ? File->ParentName : File->getName();
1276 pdbMakeAbsolute(Path);
1277 StringRef Name = InArchive ? File->getName() : StringRef(Path);
1279 pdb::DbiStreamBuilder &DbiBuilder = Builder.getDbiBuilder();
1280 File->ModuleDBI = &ExitOnErr(DbiBuilder.addModuleInfo(Name));
1281 File->ModuleDBI->setObjFileName(Path);
1283 auto Chunks = File->getChunks();
1284 uint32_t Modi = File->ModuleDBI->getModuleIndex();
1285 for (Chunk *C : Chunks) {
1286 auto *SecChunk = dyn_cast<SectionChunk>(C);
1287 if (!SecChunk || !SecChunk->Live)
1289 pdb::SectionContrib SC = createSectionContrib(SecChunk, Modi);
1290 File->ModuleDBI->setFirstSectionContrib(SC);
1294 // Before we can process symbol substreams from .debug$S, we need to process
1295 // type information, file checksums, and the string table. Add type info to
1296 // the PDB first, so that we can get the map from object file type and item
1297 // indices to PDB type and item indices.
1298 CVIndexMap ObjectIndexMap;
1299 auto IndexMapResult =
1300 mergeDebugT(File, ExternIndexMap ? ExternIndexMap : &ObjectIndexMap);
1302 // If the .debug$T sections fail to merge, assume there is no debug info.
1303 if (!IndexMapResult) {
1304 if (!Config->WarnDebugInfoUnusable) {
1305 consumeError(IndexMapResult.takeError());
1308 StringRef FileName = sys::path::filename(Path);
1309 warn("Cannot use debug info for '" + FileName + "' [LNK4099]\n" +
1310 ">>> failed to load reference " +
1311 StringRef(toString(IndexMapResult.takeError())));
1315 ScopedTimer T(SymbolMergingTimer);
1317 DebugSHandler DSH(*this, *File, *IndexMapResult);
1318 // Now do all live .debug$S and .debug$F sections.
1319 for (SectionChunk *DebugChunk : File->getDebugChunks()) {
1320 if (!DebugChunk->Live || DebugChunk->getSize() == 0)
1323 if (DebugChunk->getSectionName() == ".debug$S") {
1324 DSH.handleDebugS(*DebugChunk);
1328 if (DebugChunk->getSectionName() == ".debug$F") {
1329 ArrayRef<uint8_t> RelocatedDebugContents =
1330 relocateDebugChunk(Alloc, *DebugChunk);
1332 FixedStreamArray<object::FpoData> FpoRecords;
1333 BinaryStreamReader Reader(RelocatedDebugContents, support::little);
1334 uint32_t Count = RelocatedDebugContents.size() / sizeof(object::FpoData);
1335 ExitOnErr(Reader.readArray(FpoRecords, Count));
1337 // These are already relocated and don't refer to the string table, so we
1338 // can just copy it.
1339 for (const object::FpoData &FD : FpoRecords)
1340 DbiBuilder.addOldFpoData(FD);
1345 // Do any post-processing now that all .debug$S sections have been processed.
1349 static PublicSym32 createPublic(Defined *Def) {
1350 PublicSym32 Pub(SymbolKind::S_PUB32);
1351 Pub.Name = Def->getName();
1352 if (auto *D = dyn_cast<DefinedCOFF>(Def)) {
1353 if (D->getCOFFSymbol().isFunctionDefinition())
1354 Pub.Flags = PublicSymFlags::Function;
1355 } else if (isa<DefinedImportThunk>(Def)) {
1356 Pub.Flags = PublicSymFlags::Function;
1359 OutputSection *OS = Def->getChunk()->getOutputSection();
1360 assert(OS && "all publics should be in final image");
1361 Pub.Offset = Def->getRVA() - OS->getRVA();
1362 Pub.Segment = OS->SectionIndex;
1366 // Add all object files to the PDB. Merge .debug$T sections into IpiData and
1368 void PDBLinker::addObjectsToPDB() {
1369 ScopedTimer T1(AddObjectsTimer);
1370 for (ObjFile *File : ObjFile::Instances)
1373 Builder.getStringTableBuilder().setStrings(PDBStrTab);
1376 // Construct TPI and IPI stream contents.
1377 ScopedTimer T2(TpiStreamLayoutTimer);
1378 addTypeInfo(Builder.getTpiBuilder(), getTypeTable());
1379 addTypeInfo(Builder.getIpiBuilder(), getIDTable());
1382 ScopedTimer T3(GlobalsLayoutTimer);
1383 // Compute the public and global symbols.
1384 auto &GsiBuilder = Builder.getGsiBuilder();
1385 std::vector<PublicSym32> Publics;
1386 Symtab->forEachSymbol([&Publics](Symbol *S) {
1387 // Only emit defined, live symbols that have a chunk.
1388 auto *Def = dyn_cast<Defined>(S);
1389 if (Def && Def->isLive() && Def->getChunk())
1390 Publics.push_back(createPublic(Def));
1393 if (!Publics.empty()) {
1394 // Sort the public symbols and add them to the stream.
1395 sort(parallel::par, Publics.begin(), Publics.end(),
1396 [](const PublicSym32 &L, const PublicSym32 &R) {
1397 return L.Name < R.Name;
1399 for (const PublicSym32 &Pub : Publics)
1400 GsiBuilder.addPublicSymbol(Pub);
1404 void PDBLinker::addNatvisFiles() {
1405 for (StringRef File : Config->NatvisFiles) {
1406 ErrorOr<std::unique_ptr<MemoryBuffer>> DataOrErr =
1407 MemoryBuffer::getFile(File);
1409 warn("Cannot open input file: " + File);
1412 Builder.addInjectedSource(File, std::move(*DataOrErr));
1416 static codeview::CPUType toCodeViewMachine(COFF::MachineTypes Machine) {
1418 case COFF::IMAGE_FILE_MACHINE_AMD64:
1419 return codeview::CPUType::X64;
1420 case COFF::IMAGE_FILE_MACHINE_ARM:
1421 return codeview::CPUType::ARM7;
1422 case COFF::IMAGE_FILE_MACHINE_ARM64:
1423 return codeview::CPUType::ARM64;
1424 case COFF::IMAGE_FILE_MACHINE_ARMNT:
1425 return codeview::CPUType::ARMNT;
1426 case COFF::IMAGE_FILE_MACHINE_I386:
1427 return codeview::CPUType::Intel80386;
1429 llvm_unreachable("Unsupported CPU Type");
1433 // Mimic MSVC which surrounds arguments containing whitespace with quotes.
1434 // Double double-quotes are handled, so that the resulting string can be
1435 // executed again on the cmd-line.
1436 static std::string quote(ArrayRef<StringRef> Args) {
1439 for (StringRef A : Args) {
1442 bool HasWS = A.find(' ') != StringRef::npos;
1443 bool HasQ = A.find('"') != StringRef::npos;
1447 SmallVector<StringRef, 4> S;
1449 R.append(join(S, "\"\""));
1459 static void addCommonLinkerModuleSymbols(StringRef Path,
1460 pdb::DbiModuleDescriptorBuilder &Mod,
1461 BumpPtrAllocator &Allocator) {
1462 ObjNameSym ONS(SymbolRecordKind::ObjNameSym);
1463 Compile3Sym CS(SymbolRecordKind::Compile3Sym);
1464 EnvBlockSym EBS(SymbolRecordKind::EnvBlockSym);
1466 ONS.Name = "* Linker *";
1469 CS.Machine = toCodeViewMachine(Config->Machine);
1470 // Interestingly, if we set the string to 0.0.0.0, then when trying to view
1471 // local variables WinDbg emits an error that private symbols are not present.
1472 // By setting this to a valid MSVC linker version string, local variables are
1473 // displayed properly. As such, even though it is not representative of
1474 // LLVM's version information, we need this for compatibility.
1475 CS.Flags = CompileSym3Flags::None;
1476 CS.VersionBackendBuild = 25019;
1477 CS.VersionBackendMajor = 14;
1478 CS.VersionBackendMinor = 10;
1479 CS.VersionBackendQFE = 0;
1481 // MSVC also sets the frontend to 0.0.0.0 since this is specifically for the
1482 // linker module (which is by definition a backend), so we don't need to do
1483 // anything here. Also, it seems we can use "LLVM Linker" for the linker name
1484 // without any problems. Only the backend version has to be hardcoded to a
1486 CS.VersionFrontendBuild = 0;
1487 CS.VersionFrontendMajor = 0;
1488 CS.VersionFrontendMinor = 0;
1489 CS.VersionFrontendQFE = 0;
1490 CS.Version = "LLVM Linker";
1491 CS.setLanguage(SourceLanguage::Link);
1493 ArrayRef<StringRef> Args = makeArrayRef(Config->Argv).drop_front();
1494 std::string ArgStr = quote(Args);
1495 EBS.Fields.push_back("cwd");
1496 SmallString<64> cwd;
1497 if (Config->PDBSourcePath.empty())
1498 sys::fs::current_path(cwd);
1500 cwd = Config->PDBSourcePath;
1501 EBS.Fields.push_back(cwd);
1502 EBS.Fields.push_back("exe");
1503 SmallString<64> exe = Config->Argv[0];
1504 pdbMakeAbsolute(exe);
1505 EBS.Fields.push_back(exe);
1506 EBS.Fields.push_back("pdb");
1507 EBS.Fields.push_back(Path);
1508 EBS.Fields.push_back("cmd");
1509 EBS.Fields.push_back(ArgStr);
1510 Mod.addSymbol(codeview::SymbolSerializer::writeOneSymbol(
1511 ONS, Allocator, CodeViewContainer::Pdb));
1512 Mod.addSymbol(codeview::SymbolSerializer::writeOneSymbol(
1513 CS, Allocator, CodeViewContainer::Pdb));
1514 Mod.addSymbol(codeview::SymbolSerializer::writeOneSymbol(
1515 EBS, Allocator, CodeViewContainer::Pdb));
1518 static void addLinkerModuleSectionSymbol(pdb::DbiModuleDescriptorBuilder &Mod,
1520 BumpPtrAllocator &Allocator) {
1521 SectionSym Sym(SymbolRecordKind::SectionSym);
1522 Sym.Alignment = 12; // 2^12 = 4KB
1523 Sym.Characteristics = OS.Header.Characteristics;
1524 Sym.Length = OS.getVirtualSize();
1526 Sym.Rva = OS.getRVA();
1527 Sym.SectionNumber = OS.SectionIndex;
1528 Mod.addSymbol(codeview::SymbolSerializer::writeOneSymbol(
1529 Sym, Allocator, CodeViewContainer::Pdb));
1532 // Creates a PDB file.
1533 void coff::createPDB(SymbolTable *Symtab,
1534 ArrayRef<OutputSection *> OutputSections,
1535 ArrayRef<uint8_t> SectionTable,
1536 llvm::codeview::DebugInfo *BuildId) {
1537 ScopedTimer T1(TotalPdbLinkTimer);
1538 PDBLinker PDB(Symtab);
1540 PDB.initialize(BuildId);
1541 PDB.addObjectsToPDB();
1542 PDB.addSections(OutputSections, SectionTable);
1543 PDB.addNatvisFiles();
1545 ScopedTimer T2(DiskCommitTimer);
1546 codeview::GUID Guid;
1548 memcpy(&BuildId->PDB70.Signature, &Guid, 16);
1551 void PDBLinker::initialize(llvm::codeview::DebugInfo *BuildId) {
1552 ExitOnErr(Builder.initialize(4096)); // 4096 is blocksize
1554 BuildId->Signature.CVSignature = OMF::Signature::PDB70;
1555 // Signature is set to a hash of the PDB contents when the PDB is done.
1556 memset(BuildId->PDB70.Signature, 0, 16);
1557 BuildId->PDB70.Age = 1;
1559 // Create streams in MSF for predefined streams, namely
1560 // PDB, TPI, DBI and IPI.
1561 for (int I = 0; I < (int)pdb::kSpecialStreamCount; ++I)
1562 ExitOnErr(Builder.getMsfBuilder().addStream(0));
1564 // Add an Info stream.
1565 auto &InfoBuilder = Builder.getInfoBuilder();
1566 InfoBuilder.setVersion(pdb::PdbRaw_ImplVer::PdbImplVC70);
1567 InfoBuilder.setHashPDBContentsToGUID(true);
1569 // Add an empty DBI stream.
1570 pdb::DbiStreamBuilder &DbiBuilder = Builder.getDbiBuilder();
1571 DbiBuilder.setAge(BuildId->PDB70.Age);
1572 DbiBuilder.setVersionHeader(pdb::PdbDbiV70);
1573 DbiBuilder.setMachineType(Config->Machine);
1574 // Technically we are not link.exe 14.11, but there are known cases where
1575 // debugging tools on Windows expect Microsoft-specific version numbers or
1576 // they fail to work at all. Since we know we produce PDBs that are
1577 // compatible with LINK 14.11, we set that version number here.
1578 DbiBuilder.setBuildNumber(14, 11);
1581 void PDBLinker::addSections(ArrayRef<OutputSection *> OutputSections,
1582 ArrayRef<uint8_t> SectionTable) {
1583 // It's not entirely clear what this is, but the * Linker * module uses it.
1584 pdb::DbiStreamBuilder &DbiBuilder = Builder.getDbiBuilder();
1585 NativePath = Config->PDBPath;
1586 pdbMakeAbsolute(NativePath);
1587 uint32_t PdbFilePathNI = DbiBuilder.addECName(NativePath);
1588 auto &LinkerModule = ExitOnErr(DbiBuilder.addModuleInfo("* Linker *"));
1589 LinkerModule.setPdbFilePathNI(PdbFilePathNI);
1590 addCommonLinkerModuleSymbols(NativePath, LinkerModule, Alloc);
1592 // Add section contributions. They must be ordered by ascending RVA.
1593 for (OutputSection *OS : OutputSections) {
1594 addLinkerModuleSectionSymbol(LinkerModule, *OS, Alloc);
1595 for (Chunk *C : OS->Chunks) {
1596 pdb::SectionContrib SC =
1597 createSectionContrib(C, LinkerModule.getModuleIndex());
1598 Builder.getDbiBuilder().addSectionContrib(SC);
1602 // Add Section Map stream.
1603 ArrayRef<object::coff_section> Sections = {
1604 (const object::coff_section *)SectionTable.data(),
1605 SectionTable.size() / sizeof(object::coff_section)};
1606 SectionMap = pdb::DbiStreamBuilder::createSectionMap(Sections);
1607 DbiBuilder.setSectionMap(SectionMap);
1609 // Add COFF section header stream.
1611 DbiBuilder.addDbgStream(pdb::DbgHeaderType::SectionHdr, SectionTable));
1614 void PDBLinker::commit(codeview::GUID *Guid) {
1616 ExitOnErr(Builder.commit(Config->PDBPath, Guid));
1619 static Expected<StringRef>
1620 getFileName(const DebugStringTableSubsectionRef &Strings,
1621 const DebugChecksumsSubsectionRef &Checksums, uint32_t FileID) {
1622 auto Iter = Checksums.getArray().at(FileID);
1623 if (Iter == Checksums.getArray().end())
1624 return make_error<CodeViewError>(cv_error_code::no_records);
1625 uint32_t Offset = Iter->FileNameOffset;
1626 return Strings.getString(Offset);
1629 static uint32_t getSecrelReloc() {
1630 switch (Config->Machine) {
1632 return COFF::IMAGE_REL_AMD64_SECREL;
1634 return COFF::IMAGE_REL_I386_SECREL;
1636 return COFF::IMAGE_REL_ARM_SECREL;
1638 return COFF::IMAGE_REL_ARM64_SECREL;
1640 llvm_unreachable("unknown machine type");
1644 // Try to find a line table for the given offset Addr into the given chunk C.
1645 // If a line table was found, the line table, the string and checksum tables
1646 // that are used to interpret the line table, and the offset of Addr in the line
1647 // table are stored in the output arguments. Returns whether a line table was
1649 static bool findLineTable(const SectionChunk *C, uint32_t Addr,
1650 DebugStringTableSubsectionRef &CVStrTab,
1651 DebugChecksumsSubsectionRef &Checksums,
1652 DebugLinesSubsectionRef &Lines,
1653 uint32_t &OffsetInLinetable) {
1654 ExitOnError ExitOnErr;
1655 uint32_t SecrelReloc = getSecrelReloc();
1657 for (SectionChunk *DbgC : C->File->getDebugChunks()) {
1658 if (DbgC->getSectionName() != ".debug$S")
1661 // Build a mapping of SECREL relocations in DbgC that refer to C.
1662 DenseMap<uint32_t, uint32_t> Secrels;
1663 for (const coff_relocation &R : DbgC->Relocs) {
1664 if (R.Type != SecrelReloc)
1667 if (auto *S = dyn_cast_or_null<DefinedRegular>(
1668 C->File->getSymbols()[R.SymbolTableIndex]))
1669 if (S->getChunk() == C)
1670 Secrels[R.VirtualAddress] = S->getValue();
1673 ArrayRef<uint8_t> Contents =
1674 consumeDebugMagic(DbgC->getContents(), ".debug$S");
1675 DebugSubsectionArray Subsections;
1676 BinaryStreamReader Reader(Contents, support::little);
1677 ExitOnErr(Reader.readArray(Subsections, Contents.size()));
1679 for (const DebugSubsectionRecord &SS : Subsections) {
1680 switch (SS.kind()) {
1681 case DebugSubsectionKind::StringTable: {
1682 assert(!CVStrTab.valid() &&
1683 "Encountered multiple string table subsections!");
1684 ExitOnErr(CVStrTab.initialize(SS.getRecordData()));
1687 case DebugSubsectionKind::FileChecksums:
1688 assert(!Checksums.valid() &&
1689 "Encountered multiple checksum subsections!");
1690 ExitOnErr(Checksums.initialize(SS.getRecordData()));
1692 case DebugSubsectionKind::Lines: {
1693 ArrayRef<uint8_t> Bytes;
1694 auto Ref = SS.getRecordData();
1695 ExitOnErr(Ref.readLongestContiguousChunk(0, Bytes));
1696 size_t OffsetInDbgC = Bytes.data() - DbgC->getContents().data();
1698 // Check whether this line table refers to C.
1699 auto I = Secrels.find(OffsetInDbgC);
1700 if (I == Secrels.end())
1703 // Check whether this line table covers Addr in C.
1704 DebugLinesSubsectionRef LinesTmp;
1705 ExitOnErr(LinesTmp.initialize(BinaryStreamReader(Ref)));
1706 uint32_t OffsetInC = I->second + LinesTmp.header()->RelocOffset;
1707 if (Addr < OffsetInC || Addr >= OffsetInC + LinesTmp.header()->CodeSize)
1710 assert(!Lines.header() &&
1711 "Encountered multiple line tables for function!");
1712 ExitOnErr(Lines.initialize(BinaryStreamReader(Ref)));
1713 OffsetInLinetable = Addr - OffsetInC;
1720 if (CVStrTab.valid() && Checksums.valid() && Lines.header())
1728 // Use CodeView line tables to resolve a file and line number for the given
1729 // offset into the given chunk and return them, or {"", 0} if a line table was
1731 std::pair<StringRef, uint32_t> coff::getFileLine(const SectionChunk *C,
1733 ExitOnError ExitOnErr;
1735 DebugStringTableSubsectionRef CVStrTab;
1736 DebugChecksumsSubsectionRef Checksums;
1737 DebugLinesSubsectionRef Lines;
1738 uint32_t OffsetInLinetable;
1740 if (!findLineTable(C, Addr, CVStrTab, Checksums, Lines, OffsetInLinetable))
1743 Optional<uint32_t> NameIndex;
1744 Optional<uint32_t> LineNumber;
1745 for (LineColumnEntry &Entry : Lines) {
1746 for (const LineNumberEntry &LN : Entry.LineNumbers) {
1747 LineInfo LI(LN.Flags);
1748 if (LN.Offset > OffsetInLinetable) {
1750 NameIndex = Entry.NameIndex;
1751 LineNumber = LI.getStartLine();
1753 StringRef Filename =
1754 ExitOnErr(getFileName(CVStrTab, Checksums, *NameIndex));
1755 return {Filename, *LineNumber};
1757 NameIndex = Entry.NameIndex;
1758 LineNumber = LI.getStartLine();
1763 StringRef Filename = ExitOnErr(getFileName(CVStrTab, Checksums, *NameIndex));
1764 return {Filename, *LineNumber};