1 | //===- PDB.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 | #include "PDB.h" |
10 | #include "COFFLinkerContext.h" |
11 | #include "Chunks.h" |
12 | #include "Config.h" |
13 | #include "DebugTypes.h" |
14 | #include "Driver.h" |
15 | #include "SymbolTable.h" |
16 | #include "Symbols.h" |
17 | #include "TypeMerger.h" |
18 | #include "Writer.h" |
19 | #include "lld/Common/Timer.h" |
20 | #include "llvm/DebugInfo/CodeView/DebugFrameDataSubsection.h" |
21 | #include "llvm/DebugInfo/CodeView/DebugInlineeLinesSubsection.h" |
22 | #include "llvm/DebugInfo/CodeView/DebugLinesSubsection.h" |
23 | #include "llvm/DebugInfo/CodeView/DebugSubsectionRecord.h" |
24 | #include "llvm/DebugInfo/CodeView/GlobalTypeTableBuilder.h" |
25 | #include "llvm/DebugInfo/CodeView/LazyRandomTypeCollection.h" |
26 | #include "llvm/DebugInfo/CodeView/MergingTypeTableBuilder.h" |
27 | #include "llvm/DebugInfo/CodeView/RecordName.h" |
28 | #include "llvm/DebugInfo/CodeView/SymbolDeserializer.h" |
29 | #include "llvm/DebugInfo/CodeView/SymbolRecordHelpers.h" |
30 | #include "llvm/DebugInfo/CodeView/SymbolSerializer.h" |
31 | #include "llvm/DebugInfo/CodeView/TypeIndexDiscovery.h" |
32 | #include "llvm/DebugInfo/MSF/MSFBuilder.h" |
33 | #include "llvm/DebugInfo/MSF/MSFCommon.h" |
34 | #include "llvm/DebugInfo/MSF/MSFError.h" |
35 | #include "llvm/DebugInfo/PDB/GenericError.h" |
36 | #include "llvm/DebugInfo/PDB/Native/DbiModuleDescriptorBuilder.h" |
37 | #include "llvm/DebugInfo/PDB/Native/DbiStream.h" |
38 | #include "llvm/DebugInfo/PDB/Native/DbiStreamBuilder.h" |
39 | #include "llvm/DebugInfo/PDB/Native/GSIStreamBuilder.h" |
40 | #include "llvm/DebugInfo/PDB/Native/InfoStream.h" |
41 | #include "llvm/DebugInfo/PDB/Native/InfoStreamBuilder.h" |
42 | #include "llvm/DebugInfo/PDB/Native/NativeSession.h" |
43 | #include "llvm/DebugInfo/PDB/Native/PDBFile.h" |
44 | #include "llvm/DebugInfo/PDB/Native/PDBFileBuilder.h" |
45 | #include "llvm/DebugInfo/PDB/Native/PDBStringTableBuilder.h" |
46 | #include "llvm/DebugInfo/PDB/Native/TpiHashing.h" |
47 | #include "llvm/DebugInfo/PDB/Native/TpiStream.h" |
48 | #include "llvm/DebugInfo/PDB/Native/TpiStreamBuilder.h" |
49 | #include "llvm/DebugInfo/PDB/PDB.h" |
50 | #include "llvm/Object/COFF.h" |
51 | #include "llvm/Object/CVDebugRecord.h" |
52 | #include "llvm/Support/BinaryByteStream.h" |
53 | #include "llvm/Support/CRC.h" |
54 | #include "llvm/Support/Endian.h" |
55 | #include "llvm/Support/Errc.h" |
56 | #include "llvm/Support/FormatAdapters.h" |
57 | #include "llvm/Support/FormatVariadic.h" |
58 | #include "llvm/Support/Path.h" |
59 | #include "llvm/Support/ScopedPrinter.h" |
60 | #include "llvm/Support/TimeProfiler.h" |
61 | #include <memory> |
62 | #include <optional> |
63 | |
64 | using namespace llvm; |
65 | using namespace llvm::codeview; |
66 | using namespace lld; |
67 | using namespace lld::coff; |
68 | |
69 | using llvm::object::coff_section; |
70 | using llvm::pdb::StringTableFixup; |
71 | |
72 | namespace { |
73 | class DebugSHandler; |
74 | |
75 | class PDBLinker { |
76 | friend DebugSHandler; |
77 | |
78 | public: |
79 | PDBLinker(COFFLinkerContext &ctx) |
80 | : builder(bAlloc()), tMerger(ctx, bAlloc()), ctx(ctx) { |
81 | // This isn't strictly necessary, but link.exe usually puts an empty string |
82 | // as the first "valid" string in the string table, so we do the same in |
83 | // order to maintain as much byte-for-byte compatibility as possible. |
84 | pdbStrTab.insert(S: "" ); |
85 | } |
86 | |
87 | /// Emit the basic PDB structure: initial streams, headers, etc. |
88 | void initialize(llvm::codeview::DebugInfo *buildId); |
89 | |
90 | /// Add natvis files specified on the command line. |
91 | void addNatvisFiles(); |
92 | |
93 | /// Add named streams specified on the command line. |
94 | void addNamedStreams(); |
95 | |
96 | /// Link CodeView from each object file in the symbol table into the PDB. |
97 | void addObjectsToPDB(); |
98 | |
99 | /// Add every live, defined public symbol to the PDB. |
100 | void addPublicsToPDB(); |
101 | |
102 | /// Link info for each import file in the symbol table into the PDB. |
103 | void addImportFilesToPDB(); |
104 | |
105 | void createModuleDBI(ObjFile *file); |
106 | |
107 | /// Link CodeView from a single object file into the target (output) PDB. |
108 | /// When a precompiled headers object is linked, its TPI map might be provided |
109 | /// externally. |
110 | void addDebug(TpiSource *source); |
111 | |
112 | void addDebugSymbols(TpiSource *source); |
113 | |
114 | // Analyze the symbol records to separate module symbols from global symbols, |
115 | // find string references, and calculate how large the symbol stream will be |
116 | // in the PDB. |
117 | void analyzeSymbolSubsection(SectionChunk *debugChunk, |
118 | uint32_t &moduleSymOffset, |
119 | uint32_t &nextRelocIndex, |
120 | std::vector<StringTableFixup> &stringTableFixups, |
121 | BinaryStreamRef symData); |
122 | |
123 | // Write all module symbols from all live debug symbol subsections of the |
124 | // given object file into the given stream writer. |
125 | Error writeAllModuleSymbolRecords(ObjFile *file, BinaryStreamWriter &writer); |
126 | |
127 | // Callback to copy and relocate debug symbols during PDB file writing. |
128 | static Error commitSymbolsForObject(void *ctx, void *obj, |
129 | BinaryStreamWriter &writer); |
130 | |
131 | // Copy the symbol record, relocate it, and fix the alignment if necessary. |
132 | // Rewrite type indices in the record. Replace unrecognized symbol records |
133 | // with S_SKIP records. |
134 | void writeSymbolRecord(SectionChunk *debugChunk, |
135 | ArrayRef<uint8_t> sectionContents, CVSymbol sym, |
136 | size_t alignedSize, uint32_t &nextRelocIndex, |
137 | std::vector<uint8_t> &storage); |
138 | |
139 | /// Add the section map and section contributions to the PDB. |
140 | void addSections(ArrayRef<uint8_t> sectionTable); |
141 | |
142 | /// Write the PDB to disk and store the Guid generated for it in *Guid. |
143 | void commit(codeview::GUID *guid); |
144 | |
145 | // Print statistics regarding the final PDB |
146 | void printStats(); |
147 | |
148 | private: |
149 | void pdbMakeAbsolute(SmallVectorImpl<char> &fileName); |
150 | void translateIdSymbols(MutableArrayRef<uint8_t> &recordData, |
151 | TpiSource *source); |
152 | void addCommonLinkerModuleSymbols(StringRef path, |
153 | pdb::DbiModuleDescriptorBuilder &mod); |
154 | |
155 | pdb::PDBFileBuilder builder; |
156 | |
157 | TypeMerger tMerger; |
158 | |
159 | COFFLinkerContext &ctx; |
160 | |
161 | /// PDBs use a single global string table for filenames in the file checksum |
162 | /// table. |
163 | DebugStringTableSubsection pdbStrTab; |
164 | |
165 | llvm::SmallString<128> nativePath; |
166 | |
167 | // For statistics |
168 | uint64_t globalSymbols = 0; |
169 | uint64_t moduleSymbols = 0; |
170 | uint64_t publicSymbols = 0; |
171 | uint64_t nbTypeRecords = 0; |
172 | uint64_t nbTypeRecordsBytes = 0; |
173 | }; |
174 | |
175 | /// Represents an unrelocated DEBUG_S_FRAMEDATA subsection. |
176 | struct UnrelocatedFpoData { |
177 | SectionChunk *debugChunk = nullptr; |
178 | ArrayRef<uint8_t> subsecData; |
179 | uint32_t relocIndex = 0; |
180 | }; |
181 | |
182 | /// The size of the magic bytes at the beginning of a symbol section or stream. |
183 | enum : uint32_t { kSymbolStreamMagicSize = 4 }; |
184 | |
185 | class DebugSHandler { |
186 | PDBLinker &linker; |
187 | |
188 | /// The object file whose .debug$S sections we're processing. |
189 | ObjFile &file; |
190 | |
191 | /// The DEBUG_S_STRINGTABLE subsection. These strings are referred to by |
192 | /// index from other records in the .debug$S section. All of these strings |
193 | /// need to be added to the global PDB string table, and all references to |
194 | /// these strings need to have their indices re-written to refer to the |
195 | /// global PDB string table. |
196 | DebugStringTableSubsectionRef cvStrTab; |
197 | |
198 | /// The DEBUG_S_FILECHKSMS subsection. As above, these are referred to |
199 | /// by other records in the .debug$S section and need to be merged into the |
200 | /// PDB. |
201 | DebugChecksumsSubsectionRef checksums; |
202 | |
203 | /// The DEBUG_S_FRAMEDATA subsection(s). There can be more than one of |
204 | /// these and they need not appear in any specific order. However, they |
205 | /// contain string table references which need to be re-written, so we |
206 | /// collect them all here and re-write them after all subsections have been |
207 | /// discovered and processed. |
208 | std::vector<UnrelocatedFpoData> frameDataSubsecs; |
209 | |
210 | /// List of string table references in symbol records. Later they will be |
211 | /// applied to the symbols during PDB writing. |
212 | std::vector<StringTableFixup> stringTableFixups; |
213 | |
214 | /// Sum of the size of all module symbol records across all .debug$S sections. |
215 | /// Includes record realignment and the size of the symbol stream magic |
216 | /// prefix. |
217 | uint32_t moduleStreamSize = kSymbolStreamMagicSize; |
218 | |
219 | /// Next relocation index in the current .debug$S section. Resets every |
220 | /// handleDebugS call. |
221 | uint32_t nextRelocIndex = 0; |
222 | |
223 | void advanceRelocIndex(SectionChunk *debugChunk, ArrayRef<uint8_t> subsec); |
224 | |
225 | void addUnrelocatedSubsection(SectionChunk *debugChunk, |
226 | const DebugSubsectionRecord &ss); |
227 | |
228 | void addFrameDataSubsection(SectionChunk *debugChunk, |
229 | const DebugSubsectionRecord &ss); |
230 | |
231 | public: |
232 | DebugSHandler(PDBLinker &linker, ObjFile &file) |
233 | : linker(linker), file(file) {} |
234 | |
235 | void handleDebugS(SectionChunk *debugChunk); |
236 | |
237 | void finish(); |
238 | }; |
239 | } |
240 | |
241 | // Visual Studio's debugger requires absolute paths in various places in the |
242 | // PDB to work without additional configuration: |
243 | // https://docs.microsoft.com/en-us/visualstudio/debugger/debug-source-files-common-properties-solution-property-pages-dialog-box |
244 | void PDBLinker::pdbMakeAbsolute(SmallVectorImpl<char> &fileName) { |
245 | // The default behavior is to produce paths that are valid within the context |
246 | // of the machine that you perform the link on. If the linker is running on |
247 | // a POSIX system, we will output absolute POSIX paths. If the linker is |
248 | // running on a Windows system, we will output absolute Windows paths. If the |
249 | // user desires any other kind of behavior, they should explicitly pass |
250 | // /pdbsourcepath, in which case we will treat the exact string the user |
251 | // passed in as the gospel and not normalize, canonicalize it. |
252 | if (sys::path::is_absolute(path: fileName, style: sys::path::Style::windows) || |
253 | sys::path::is_absolute(path: fileName, style: sys::path::Style::posix)) |
254 | return; |
255 | |
256 | // It's not absolute in any path syntax. Relative paths necessarily refer to |
257 | // the local file system, so we can make it native without ending up with a |
258 | // nonsensical path. |
259 | if (ctx.config.pdbSourcePath.empty()) { |
260 | sys::path::native(path&: fileName); |
261 | sys::fs::make_absolute(path&: fileName); |
262 | sys::path::remove_dots(path&: fileName, remove_dot_dot: true); |
263 | return; |
264 | } |
265 | |
266 | // Try to guess whether /PDBSOURCEPATH is a unix path or a windows path. |
267 | // Since PDB's are more of a Windows thing, we make this conservative and only |
268 | // decide that it's a unix path if we're fairly certain. Specifically, if |
269 | // it starts with a forward slash. |
270 | SmallString<128> absoluteFileName = ctx.config.pdbSourcePath; |
271 | sys::path::Style guessedStyle = absoluteFileName.starts_with(Prefix: "/" ) |
272 | ? sys::path::Style::posix |
273 | : sys::path::Style::windows; |
274 | sys::path::append(path&: absoluteFileName, style: guessedStyle, a: fileName); |
275 | sys::path::native(path&: absoluteFileName, style: guessedStyle); |
276 | sys::path::remove_dots(path&: absoluteFileName, remove_dot_dot: true, style: guessedStyle); |
277 | |
278 | fileName = std::move(absoluteFileName); |
279 | } |
280 | |
281 | static void addTypeInfo(pdb::TpiStreamBuilder &tpiBuilder, |
282 | TypeCollection &typeTable) { |
283 | // Start the TPI or IPI stream header. |
284 | tpiBuilder.setVersionHeader(pdb::PdbTpiV80); |
285 | |
286 | // Flatten the in memory type table and hash each type. |
287 | typeTable.ForEachRecord(Func: [&](TypeIndex ti, const CVType &type) { |
288 | auto hash = pdb::hashTypeRecord(Type: type); |
289 | if (auto e = hash.takeError()) |
290 | fatal(msg: "type hashing error" ); |
291 | tpiBuilder.addTypeRecord(Type: type.RecordData, Hash: *hash); |
292 | }); |
293 | } |
294 | |
295 | static void addGHashTypeInfo(COFFLinkerContext &ctx, |
296 | pdb::PDBFileBuilder &builder) { |
297 | // Start the TPI or IPI stream header. |
298 | builder.getTpiBuilder().setVersionHeader(pdb::PdbTpiV80); |
299 | builder.getIpiBuilder().setVersionHeader(pdb::PdbTpiV80); |
300 | for (TpiSource *source : ctx.tpiSourceList) { |
301 | builder.getTpiBuilder().addTypeRecords(Types: source->mergedTpi.recs, |
302 | Sizes: source->mergedTpi.recSizes, |
303 | Hashes: source->mergedTpi.recHashes); |
304 | builder.getIpiBuilder().addTypeRecords(Types: source->mergedIpi.recs, |
305 | Sizes: source->mergedIpi.recSizes, |
306 | Hashes: source->mergedIpi.recHashes); |
307 | } |
308 | } |
309 | |
310 | static void |
311 | recordStringTableReferences(CVSymbol sym, uint32_t symOffset, |
312 | std::vector<StringTableFixup> &stringTableFixups) { |
313 | // For now we only handle S_FILESTATIC, but we may need the same logic for |
314 | // S_DEFRANGE and S_DEFRANGE_SUBFIELD. However, I cannot seem to generate any |
315 | // PDBs that contain these types of records, so because of the uncertainty |
316 | // they are omitted here until we can prove that it's necessary. |
317 | switch (sym.kind()) { |
318 | case SymbolKind::S_FILESTATIC: { |
319 | // FileStaticSym::ModFileOffset |
320 | uint32_t ref = *reinterpret_cast<const ulittle32_t *>(&sym.data()[8]); |
321 | stringTableFixups.push_back(x: {.StrTabOffset: ref, .SymOffsetOfReference: symOffset + 8}); |
322 | break; |
323 | } |
324 | case SymbolKind::S_DEFRANGE: |
325 | case SymbolKind::S_DEFRANGE_SUBFIELD: |
326 | log(msg: "Not fixing up string table reference in S_DEFRANGE / " |
327 | "S_DEFRANGE_SUBFIELD record" ); |
328 | break; |
329 | default: |
330 | break; |
331 | } |
332 | } |
333 | |
334 | static SymbolKind symbolKind(ArrayRef<uint8_t> recordData) { |
335 | const RecordPrefix *prefix = |
336 | reinterpret_cast<const RecordPrefix *>(recordData.data()); |
337 | return static_cast<SymbolKind>(uint16_t(prefix->RecordKind)); |
338 | } |
339 | |
340 | /// MSVC translates S_PROC_ID_END to S_END, and S_[LG]PROC32_ID to S_[LG]PROC32 |
341 | void PDBLinker::translateIdSymbols(MutableArrayRef<uint8_t> &recordData, |
342 | TpiSource *source) { |
343 | RecordPrefix *prefix = reinterpret_cast<RecordPrefix *>(recordData.data()); |
344 | |
345 | SymbolKind kind = symbolKind(recordData); |
346 | |
347 | if (kind == SymbolKind::S_PROC_ID_END) { |
348 | prefix->RecordKind = SymbolKind::S_END; |
349 | return; |
350 | } |
351 | |
352 | // In an object file, GPROC32_ID has an embedded reference which refers to the |
353 | // single object file type index namespace. This has already been translated |
354 | // to the PDB file's ID stream index space, but we need to convert this to a |
355 | // symbol that refers to the type stream index space. So we remap again from |
356 | // ID index space to type index space. |
357 | if (kind == SymbolKind::S_GPROC32_ID || kind == SymbolKind::S_LPROC32_ID) { |
358 | SmallVector<TiReference, 1> refs; |
359 | auto content = recordData.drop_front(N: sizeof(RecordPrefix)); |
360 | CVSymbol sym(recordData); |
361 | discoverTypeIndicesInSymbol(Symbol: sym, Refs&: refs); |
362 | assert(refs.size() == 1); |
363 | assert(refs.front().Count == 1); |
364 | |
365 | TypeIndex *ti = |
366 | reinterpret_cast<TypeIndex *>(content.data() + refs[0].Offset); |
367 | // `ti` is the index of a FuncIdRecord or MemberFuncIdRecord which lives in |
368 | // the IPI stream, whose `FunctionType` member refers to the TPI stream. |
369 | // Note that LF_FUNC_ID and LF_MFUNC_ID have the same record layout, and |
370 | // in both cases we just need the second type index. |
371 | if (!ti->isSimple() && !ti->isNoneType()) { |
372 | TypeIndex newType = TypeIndex(SimpleTypeKind::NotTranslated); |
373 | if (ctx.config.debugGHashes) { |
374 | auto idToType = tMerger.funcIdToType.find(Val: *ti); |
375 | if (idToType != tMerger.funcIdToType.end()) |
376 | newType = idToType->second; |
377 | } else { |
378 | if (tMerger.getIDTable().contains(Index: *ti)) { |
379 | CVType funcIdData = tMerger.getIDTable().getType(Index: *ti); |
380 | if (funcIdData.length() >= 8 && (funcIdData.kind() == LF_FUNC_ID || |
381 | funcIdData.kind() == LF_MFUNC_ID)) { |
382 | newType = *reinterpret_cast<const TypeIndex *>(&funcIdData.data()[8]); |
383 | } |
384 | } |
385 | } |
386 | if (newType == TypeIndex(SimpleTypeKind::NotTranslated)) { |
387 | warn(msg: formatv(Fmt: "procedure symbol record for `{0}` in {1} refers to PDB " |
388 | "item index {2:X} which is not a valid function ID record" , |
389 | Vals: getSymbolName(Sym: CVSymbol(recordData)), |
390 | Vals: source->file->getName(), Vals: ti->getIndex())); |
391 | } |
392 | *ti = newType; |
393 | } |
394 | |
395 | kind = (kind == SymbolKind::S_GPROC32_ID) ? SymbolKind::S_GPROC32 |
396 | : SymbolKind::S_LPROC32; |
397 | prefix->RecordKind = uint16_t(kind); |
398 | } |
399 | } |
400 | |
401 | namespace { |
402 | struct ScopeRecord { |
403 | ulittle32_t ptrParent; |
404 | ulittle32_t ptrEnd; |
405 | }; |
406 | } // namespace |
407 | |
408 | /// Given a pointer to a symbol record that opens a scope, return a pointer to |
409 | /// the scope fields. |
410 | static ScopeRecord *getSymbolScopeFields(void *sym) { |
411 | return reinterpret_cast<ScopeRecord *>(reinterpret_cast<char *>(sym) + |
412 | sizeof(RecordPrefix)); |
413 | } |
414 | |
415 | // To open a scope, push the offset of the current symbol record onto the |
416 | // stack. |
417 | static void scopeStackOpen(SmallVectorImpl<uint32_t> &stack, |
418 | std::vector<uint8_t> &storage) { |
419 | stack.push_back(Elt: storage.size()); |
420 | } |
421 | |
422 | // To close a scope, update the record that opened the scope. |
423 | static void scopeStackClose(SmallVectorImpl<uint32_t> &stack, |
424 | std::vector<uint8_t> &storage, |
425 | uint32_t storageBaseOffset, ObjFile *file) { |
426 | if (stack.empty()) { |
427 | warn(msg: "symbol scopes are not balanced in " + file->getName()); |
428 | return; |
429 | } |
430 | |
431 | // Update ptrEnd of the record that opened the scope to point to the |
432 | // current record, if we are writing into the module symbol stream. |
433 | uint32_t offOpen = stack.pop_back_val(); |
434 | uint32_t offEnd = storageBaseOffset + storage.size(); |
435 | uint32_t offParent = stack.empty() ? 0 : (stack.back() + storageBaseOffset); |
436 | ScopeRecord *scopeRec = getSymbolScopeFields(sym: &(storage)[offOpen]); |
437 | scopeRec->ptrParent = offParent; |
438 | scopeRec->ptrEnd = offEnd; |
439 | } |
440 | |
441 | static bool symbolGoesInModuleStream(const CVSymbol &sym, |
442 | unsigned symbolScopeDepth) { |
443 | switch (sym.kind()) { |
444 | case SymbolKind::S_GDATA32: |
445 | case SymbolKind::S_GTHREAD32: |
446 | // We really should not be seeing S_PROCREF and S_LPROCREF in the first place |
447 | // since they are synthesized by the linker in response to S_GPROC32 and |
448 | // S_LPROC32, but if we do see them, don't put them in the module stream I |
449 | // guess. |
450 | case SymbolKind::S_PROCREF: |
451 | case SymbolKind::S_LPROCREF: |
452 | return false; |
453 | // S_UDT and S_CONSTANT records go in the module stream if it is not a global record. |
454 | case SymbolKind::S_UDT: |
455 | case SymbolKind::S_CONSTANT: |
456 | return symbolScopeDepth > 0; |
457 | // S_GDATA32 does not go in the module stream, but S_LDATA32 does. |
458 | case SymbolKind::S_LDATA32: |
459 | case SymbolKind::S_LTHREAD32: |
460 | default: |
461 | return true; |
462 | } |
463 | } |
464 | |
465 | static bool symbolGoesInGlobalsStream(const CVSymbol &sym, |
466 | unsigned symbolScopeDepth) { |
467 | switch (sym.kind()) { |
468 | case SymbolKind::S_GDATA32: |
469 | case SymbolKind::S_GTHREAD32: |
470 | case SymbolKind::S_GPROC32: |
471 | case SymbolKind::S_LPROC32: |
472 | case SymbolKind::S_GPROC32_ID: |
473 | case SymbolKind::S_LPROC32_ID: |
474 | // We really should not be seeing S_PROCREF and S_LPROCREF in the first place |
475 | // since they are synthesized by the linker in response to S_GPROC32 and |
476 | // S_LPROC32, but if we do see them, copy them straight through. |
477 | case SymbolKind::S_PROCREF: |
478 | case SymbolKind::S_LPROCREF: |
479 | return true; |
480 | // Records that go in the globals stream, unless they are function-local. |
481 | case SymbolKind::S_UDT: |
482 | case SymbolKind::S_LDATA32: |
483 | case SymbolKind::S_LTHREAD32: |
484 | case SymbolKind::S_CONSTANT: |
485 | return symbolScopeDepth == 0; |
486 | default: |
487 | return false; |
488 | } |
489 | } |
490 | |
491 | static void addGlobalSymbol(pdb::GSIStreamBuilder &builder, uint16_t modIndex, |
492 | unsigned symOffset, |
493 | std::vector<uint8_t> &symStorage) { |
494 | CVSymbol sym{ArrayRef(symStorage)}; |
495 | switch (sym.kind()) { |
496 | case SymbolKind::S_CONSTANT: |
497 | case SymbolKind::S_UDT: |
498 | case SymbolKind::S_GDATA32: |
499 | case SymbolKind::S_GTHREAD32: |
500 | case SymbolKind::S_LTHREAD32: |
501 | case SymbolKind::S_LDATA32: |
502 | case SymbolKind::S_PROCREF: |
503 | case SymbolKind::S_LPROCREF: { |
504 | // sym is a temporary object, so we have to copy and reallocate the record |
505 | // to stabilize it. |
506 | uint8_t *mem = bAlloc().Allocate<uint8_t>(Num: sym.length()); |
507 | memcpy(dest: mem, src: sym.data().data(), n: sym.length()); |
508 | builder.addGlobalSymbol(Sym: CVSymbol(ArrayRef(mem, sym.length()))); |
509 | break; |
510 | } |
511 | case SymbolKind::S_GPROC32: |
512 | case SymbolKind::S_LPROC32: { |
513 | SymbolRecordKind k = SymbolRecordKind::ProcRefSym; |
514 | if (sym.kind() == SymbolKind::S_LPROC32) |
515 | k = SymbolRecordKind::LocalProcRef; |
516 | ProcRefSym ps(k); |
517 | ps.Module = modIndex; |
518 | // For some reason, MSVC seems to add one to this value. |
519 | ++ps.Module; |
520 | ps.Name = getSymbolName(Sym: sym); |
521 | ps.SumName = 0; |
522 | ps.SymOffset = symOffset; |
523 | builder.addGlobalSymbol(Sym: ps); |
524 | break; |
525 | } |
526 | default: |
527 | llvm_unreachable("Invalid symbol kind!" ); |
528 | } |
529 | } |
530 | |
531 | // Check if the given symbol record was padded for alignment. If so, zero out |
532 | // the padding bytes and update the record prefix with the new size. |
533 | static void fixRecordAlignment(MutableArrayRef<uint8_t> recordBytes, |
534 | size_t oldSize) { |
535 | size_t alignedSize = recordBytes.size(); |
536 | if (oldSize == alignedSize) |
537 | return; |
538 | reinterpret_cast<RecordPrefix *>(recordBytes.data())->RecordLen = |
539 | alignedSize - 2; |
540 | memset(s: recordBytes.data() + oldSize, c: 0, n: alignedSize - oldSize); |
541 | } |
542 | |
543 | // Replace any record with a skip record of the same size. This is useful when |
544 | // we have reserved size for a symbol record, but type index remapping fails. |
545 | static void replaceWithSkipRecord(MutableArrayRef<uint8_t> recordBytes) { |
546 | memset(s: recordBytes.data(), c: 0, n: recordBytes.size()); |
547 | auto *prefix = reinterpret_cast<RecordPrefix *>(recordBytes.data()); |
548 | prefix->RecordKind = SymbolKind::S_SKIP; |
549 | prefix->RecordLen = recordBytes.size() - 2; |
550 | } |
551 | |
552 | // Copy the symbol record, relocate it, and fix the alignment if necessary. |
553 | // Rewrite type indices in the record. Replace unrecognized symbol records with |
554 | // S_SKIP records. |
555 | void PDBLinker::writeSymbolRecord(SectionChunk *debugChunk, |
556 | ArrayRef<uint8_t> sectionContents, |
557 | CVSymbol sym, size_t alignedSize, |
558 | uint32_t &nextRelocIndex, |
559 | std::vector<uint8_t> &storage) { |
560 | // Allocate space for the new record at the end of the storage. |
561 | storage.resize(new_size: storage.size() + alignedSize); |
562 | auto recordBytes = MutableArrayRef<uint8_t>(storage).take_back(N: alignedSize); |
563 | |
564 | // Copy the symbol record and relocate it. |
565 | debugChunk->writeAndRelocateSubsection(sec: sectionContents, subsec: sym.data(), |
566 | nextRelocIndex, buf: recordBytes.data()); |
567 | fixRecordAlignment(recordBytes, oldSize: sym.length()); |
568 | |
569 | // Re-map all the type index references. |
570 | TpiSource *source = debugChunk->file->debugTypesObj; |
571 | if (!source->remapTypesInSymbolRecord(rec: recordBytes)) { |
572 | log(msg: "ignoring unknown symbol record with kind 0x" + utohexstr(X: sym.kind())); |
573 | replaceWithSkipRecord(recordBytes); |
574 | } |
575 | |
576 | // An object file may have S_xxx_ID symbols, but these get converted to |
577 | // "real" symbols in a PDB. |
578 | translateIdSymbols(recordData&: recordBytes, source); |
579 | } |
580 | |
581 | void PDBLinker::analyzeSymbolSubsection( |
582 | SectionChunk *debugChunk, uint32_t &moduleSymOffset, |
583 | uint32_t &nextRelocIndex, std::vector<StringTableFixup> &stringTableFixups, |
584 | BinaryStreamRef symData) { |
585 | ObjFile *file = debugChunk->file; |
586 | uint32_t moduleSymStart = moduleSymOffset; |
587 | |
588 | uint32_t scopeLevel = 0; |
589 | std::vector<uint8_t> storage; |
590 | ArrayRef<uint8_t> sectionContents = debugChunk->getContents(); |
591 | |
592 | ArrayRef<uint8_t> symsBuffer; |
593 | cantFail(Err: symData.readBytes(Offset: 0, Size: symData.getLength(), Buffer&: symsBuffer)); |
594 | |
595 | if (symsBuffer.empty()) |
596 | warn(msg: "empty symbols subsection in " + file->getName()); |
597 | |
598 | Error ec = forEachCodeViewRecord<CVSymbol>( |
599 | StreamBuffer: symsBuffer, F: [&](CVSymbol sym) -> llvm::Error { |
600 | // Track the current scope. |
601 | if (symbolOpensScope(Kind: sym.kind())) |
602 | ++scopeLevel; |
603 | else if (symbolEndsScope(Kind: sym.kind())) |
604 | --scopeLevel; |
605 | |
606 | uint32_t alignedSize = |
607 | alignTo(Value: sym.length(), Align: alignOf(Container: CodeViewContainer::Pdb)); |
608 | |
609 | // Copy global records. Some global records (mainly procedures) |
610 | // reference the current offset into the module stream. |
611 | if (symbolGoesInGlobalsStream(sym, symbolScopeDepth: scopeLevel)) { |
612 | storage.clear(); |
613 | writeSymbolRecord(debugChunk, sectionContents, sym, alignedSize, |
614 | nextRelocIndex, storage); |
615 | addGlobalSymbol(builder&: builder.getGsiBuilder(), |
616 | modIndex: file->moduleDBI->getModuleIndex(), symOffset: moduleSymOffset, |
617 | symStorage&: storage); |
618 | ++globalSymbols; |
619 | } |
620 | |
621 | // Update the module stream offset and record any string table index |
622 | // references. There are very few of these and they will be rewritten |
623 | // later during PDB writing. |
624 | if (symbolGoesInModuleStream(sym, symbolScopeDepth: scopeLevel)) { |
625 | recordStringTableReferences(sym, symOffset: moduleSymOffset, stringTableFixups); |
626 | moduleSymOffset += alignedSize; |
627 | ++moduleSymbols; |
628 | } |
629 | |
630 | return Error::success(); |
631 | }); |
632 | |
633 | // If we encountered corrupt records, ignore the whole subsection. If we wrote |
634 | // any partial records, undo that. For globals, we just keep what we have and |
635 | // continue. |
636 | if (ec) { |
637 | warn(msg: "corrupt symbol records in " + file->getName()); |
638 | moduleSymOffset = moduleSymStart; |
639 | consumeError(Err: std::move(ec)); |
640 | } |
641 | } |
642 | |
643 | Error PDBLinker::writeAllModuleSymbolRecords(ObjFile *file, |
644 | BinaryStreamWriter &writer) { |
645 | ExitOnError exitOnErr; |
646 | std::vector<uint8_t> storage; |
647 | SmallVector<uint32_t, 4> scopes; |
648 | |
649 | // Visit all live .debug$S sections a second time, and write them to the PDB. |
650 | for (SectionChunk *debugChunk : file->getDebugChunks()) { |
651 | if (!debugChunk->live || debugChunk->getSize() == 0 || |
652 | debugChunk->getSectionName() != ".debug$S" ) |
653 | continue; |
654 | |
655 | ArrayRef<uint8_t> sectionContents = debugChunk->getContents(); |
656 | auto contents = |
657 | SectionChunk::consumeDebugMagic(data: sectionContents, sectionName: ".debug$S" ); |
658 | DebugSubsectionArray subsections; |
659 | BinaryStreamReader reader(contents, llvm::endianness::little); |
660 | exitOnErr(reader.readArray(Array&: subsections, Size: contents.size())); |
661 | |
662 | uint32_t nextRelocIndex = 0; |
663 | for (const DebugSubsectionRecord &ss : subsections) { |
664 | if (ss.kind() != DebugSubsectionKind::Symbols) |
665 | continue; |
666 | |
667 | uint32_t moduleSymStart = writer.getOffset(); |
668 | scopes.clear(); |
669 | storage.clear(); |
670 | ArrayRef<uint8_t> symsBuffer; |
671 | BinaryStreamRef sr = ss.getRecordData(); |
672 | cantFail(Err: sr.readBytes(Offset: 0, Size: sr.getLength(), Buffer&: symsBuffer)); |
673 | auto ec = forEachCodeViewRecord<CVSymbol>( |
674 | StreamBuffer: symsBuffer, F: [&](CVSymbol sym) -> llvm::Error { |
675 | // Track the current scope. Only update records in the postmerge |
676 | // pass. |
677 | if (symbolOpensScope(Kind: sym.kind())) |
678 | scopeStackOpen(stack&: scopes, storage); |
679 | else if (symbolEndsScope(Kind: sym.kind())) |
680 | scopeStackClose(stack&: scopes, storage, storageBaseOffset: moduleSymStart, file); |
681 | |
682 | // Copy, relocate, and rewrite each module symbol. |
683 | if (symbolGoesInModuleStream(sym, symbolScopeDepth: scopes.size())) { |
684 | uint32_t alignedSize = |
685 | alignTo(Value: sym.length(), Align: alignOf(Container: CodeViewContainer::Pdb)); |
686 | writeSymbolRecord(debugChunk, sectionContents, sym, alignedSize, |
687 | nextRelocIndex, storage); |
688 | } |
689 | return Error::success(); |
690 | }); |
691 | |
692 | // If we encounter corrupt records in the second pass, ignore them. We |
693 | // already warned about them in the first analysis pass. |
694 | if (ec) { |
695 | consumeError(Err: std::move(ec)); |
696 | storage.clear(); |
697 | } |
698 | |
699 | // Writing bytes has a very high overhead, so write the entire subsection |
700 | // at once. |
701 | // TODO: Consider buffering symbols for the entire object file to reduce |
702 | // overhead even further. |
703 | if (Error e = writer.writeBytes(Buffer: storage)) |
704 | return e; |
705 | } |
706 | } |
707 | |
708 | return Error::success(); |
709 | } |
710 | |
711 | Error PDBLinker::commitSymbolsForObject(void *ctx, void *obj, |
712 | BinaryStreamWriter &writer) { |
713 | return static_cast<PDBLinker *>(ctx)->writeAllModuleSymbolRecords( |
714 | file: static_cast<ObjFile *>(obj), writer); |
715 | } |
716 | |
717 | static pdb::SectionContrib createSectionContrib(COFFLinkerContext &ctx, |
718 | const Chunk *c, uint32_t modi) { |
719 | OutputSection *os = c ? ctx.getOutputSection(c) : nullptr; |
720 | pdb::SectionContrib sc; |
721 | memset(s: &sc, c: 0, n: sizeof(sc)); |
722 | sc.ISect = os ? os->sectionIndex : llvm::pdb::kInvalidStreamIndex; |
723 | sc.Off = c && os ? c->getRVA() - os->getRVA() : 0; |
724 | sc.Size = c ? c->getSize() : -1; |
725 | if (auto *secChunk = dyn_cast_or_null<SectionChunk>(Val: c)) { |
726 | sc.Characteristics = secChunk->header->Characteristics; |
727 | sc.Imod = secChunk->file->moduleDBI->getModuleIndex(); |
728 | ArrayRef<uint8_t> contents = secChunk->getContents(); |
729 | JamCRC crc(0); |
730 | crc.update(Data: contents); |
731 | sc.DataCrc = crc.getCRC(); |
732 | } else { |
733 | sc.Characteristics = os ? os->header.Characteristics : 0; |
734 | sc.Imod = modi; |
735 | } |
736 | sc.RelocCrc = 0; // FIXME |
737 | |
738 | return sc; |
739 | } |
740 | |
741 | static uint32_t |
742 | translateStringTableIndex(uint32_t objIndex, |
743 | const DebugStringTableSubsectionRef &objStrTable, |
744 | DebugStringTableSubsection &pdbStrTable) { |
745 | auto expectedString = objStrTable.getString(Offset: objIndex); |
746 | if (!expectedString) { |
747 | warn(msg: "Invalid string table reference" ); |
748 | consumeError(Err: expectedString.takeError()); |
749 | return 0; |
750 | } |
751 | |
752 | return pdbStrTable.insert(S: *expectedString); |
753 | } |
754 | |
755 | void DebugSHandler::handleDebugS(SectionChunk *debugChunk) { |
756 | // Note that we are processing the *unrelocated* section contents. They will |
757 | // be relocated later during PDB writing. |
758 | ArrayRef<uint8_t> contents = debugChunk->getContents(); |
759 | contents = SectionChunk::consumeDebugMagic(data: contents, sectionName: ".debug$S" ); |
760 | DebugSubsectionArray subsections; |
761 | BinaryStreamReader reader(contents, llvm::endianness::little); |
762 | ExitOnError exitOnErr; |
763 | exitOnErr(reader.readArray(Array&: subsections, Size: contents.size())); |
764 | debugChunk->sortRelocations(); |
765 | |
766 | // Reset the relocation index, since this is a new section. |
767 | nextRelocIndex = 0; |
768 | |
769 | for (const DebugSubsectionRecord &ss : subsections) { |
770 | // Ignore subsections with the 'ignore' bit. Some versions of the Visual C++ |
771 | // runtime have subsections with this bit set. |
772 | if (uint32_t(ss.kind()) & codeview::SubsectionIgnoreFlag) |
773 | continue; |
774 | |
775 | switch (ss.kind()) { |
776 | case DebugSubsectionKind::StringTable: { |
777 | assert(!cvStrTab.valid() && |
778 | "Encountered multiple string table subsections!" ); |
779 | exitOnErr(cvStrTab.initialize(Contents: ss.getRecordData())); |
780 | break; |
781 | } |
782 | case DebugSubsectionKind::FileChecksums: |
783 | assert(!checksums.valid() && |
784 | "Encountered multiple checksum subsections!" ); |
785 | exitOnErr(checksums.initialize(Stream: ss.getRecordData())); |
786 | break; |
787 | case DebugSubsectionKind::Lines: |
788 | case DebugSubsectionKind::InlineeLines: |
789 | addUnrelocatedSubsection(debugChunk, ss); |
790 | break; |
791 | case DebugSubsectionKind::FrameData: |
792 | addFrameDataSubsection(debugChunk, ss); |
793 | break; |
794 | case DebugSubsectionKind::Symbols: |
795 | linker.analyzeSymbolSubsection(debugChunk, moduleSymOffset&: moduleStreamSize, |
796 | nextRelocIndex, stringTableFixups, |
797 | symData: ss.getRecordData()); |
798 | break; |
799 | |
800 | case DebugSubsectionKind::CrossScopeImports: |
801 | case DebugSubsectionKind::CrossScopeExports: |
802 | // These appear to relate to cross-module optimization, so we might use |
803 | // these for ThinLTO. |
804 | break; |
805 | |
806 | case DebugSubsectionKind::ILLines: |
807 | case DebugSubsectionKind::FuncMDTokenMap: |
808 | case DebugSubsectionKind::TypeMDTokenMap: |
809 | case DebugSubsectionKind::MergedAssemblyInput: |
810 | // These appear to relate to .Net assembly info. |
811 | break; |
812 | |
813 | case DebugSubsectionKind::CoffSymbolRVA: |
814 | // Unclear what this is for. |
815 | break; |
816 | |
817 | case DebugSubsectionKind::XfgHashType: |
818 | case DebugSubsectionKind::XfgHashVirtual: |
819 | break; |
820 | |
821 | default: |
822 | warn(msg: "ignoring unknown debug$S subsection kind 0x" + |
823 | utohexstr(X: uint32_t(ss.kind())) + " in file " + toString(file: &file)); |
824 | break; |
825 | } |
826 | } |
827 | } |
828 | |
829 | void DebugSHandler::advanceRelocIndex(SectionChunk *sc, |
830 | ArrayRef<uint8_t> subsec) { |
831 | ptrdiff_t vaBegin = subsec.data() - sc->getContents().data(); |
832 | assert(vaBegin > 0); |
833 | auto relocs = sc->getRelocs(); |
834 | for (; nextRelocIndex < relocs.size(); ++nextRelocIndex) { |
835 | if (relocs[nextRelocIndex].VirtualAddress >= (uint32_t)vaBegin) |
836 | break; |
837 | } |
838 | } |
839 | |
840 | namespace { |
841 | /// Wrapper class for unrelocated line and inlinee line subsections, which |
842 | /// require only relocation and type index remapping to add to the PDB. |
843 | class UnrelocatedDebugSubsection : public DebugSubsection { |
844 | public: |
845 | UnrelocatedDebugSubsection(DebugSubsectionKind k, SectionChunk *debugChunk, |
846 | ArrayRef<uint8_t> subsec, uint32_t relocIndex) |
847 | : DebugSubsection(k), debugChunk(debugChunk), subsec(subsec), |
848 | relocIndex(relocIndex) {} |
849 | |
850 | Error commit(BinaryStreamWriter &writer) const override; |
851 | uint32_t calculateSerializedSize() const override { return subsec.size(); } |
852 | |
853 | SectionChunk *debugChunk; |
854 | ArrayRef<uint8_t> subsec; |
855 | uint32_t relocIndex; |
856 | }; |
857 | } // namespace |
858 | |
859 | Error UnrelocatedDebugSubsection::commit(BinaryStreamWriter &writer) const { |
860 | std::vector<uint8_t> relocatedBytes(subsec.size()); |
861 | uint32_t tmpRelocIndex = relocIndex; |
862 | debugChunk->writeAndRelocateSubsection(sec: debugChunk->getContents(), subsec, |
863 | nextRelocIndex&: tmpRelocIndex, buf: relocatedBytes.data()); |
864 | |
865 | // Remap type indices in inlinee line records in place. Skip the remapping if |
866 | // there is no type source info. |
867 | if (kind() == DebugSubsectionKind::InlineeLines && |
868 | debugChunk->file->debugTypesObj) { |
869 | TpiSource *source = debugChunk->file->debugTypesObj; |
870 | DebugInlineeLinesSubsectionRef inlineeLines; |
871 | BinaryStreamReader storageReader(relocatedBytes, llvm::endianness::little); |
872 | ExitOnError exitOnErr; |
873 | exitOnErr(inlineeLines.initialize(Reader: storageReader)); |
874 | for (const InlineeSourceLine &line : inlineeLines) { |
875 | TypeIndex &inlinee = *const_cast<TypeIndex *>(&line.Header->Inlinee); |
876 | if (!source->remapTypeIndex(ti&: inlinee, refKind: TiRefKind::IndexRef)) { |
877 | log(msg: "bad inlinee line record in " + debugChunk->file->getName() + |
878 | " with bad inlinee index 0x" + utohexstr(X: inlinee.getIndex())); |
879 | } |
880 | } |
881 | } |
882 | |
883 | return writer.writeBytes(Buffer: relocatedBytes); |
884 | } |
885 | |
886 | void DebugSHandler::addUnrelocatedSubsection(SectionChunk *debugChunk, |
887 | const DebugSubsectionRecord &ss) { |
888 | ArrayRef<uint8_t> subsec; |
889 | BinaryStreamRef sr = ss.getRecordData(); |
890 | cantFail(Err: sr.readBytes(Offset: 0, Size: sr.getLength(), Buffer&: subsec)); |
891 | advanceRelocIndex(sc: debugChunk, subsec); |
892 | file.moduleDBI->addDebugSubsection( |
893 | Subsection: std::make_shared<UnrelocatedDebugSubsection>(args: ss.kind(), args&: debugChunk, |
894 | args&: subsec, args&: nextRelocIndex)); |
895 | } |
896 | |
897 | void DebugSHandler::addFrameDataSubsection(SectionChunk *debugChunk, |
898 | const DebugSubsectionRecord &ss) { |
899 | // We need to re-write string table indices here, so save off all |
900 | // frame data subsections until we've processed the entire list of |
901 | // subsections so that we can be sure we have the string table. |
902 | ArrayRef<uint8_t> subsec; |
903 | BinaryStreamRef sr = ss.getRecordData(); |
904 | cantFail(Err: sr.readBytes(Offset: 0, Size: sr.getLength(), Buffer&: subsec)); |
905 | advanceRelocIndex(sc: debugChunk, subsec); |
906 | frameDataSubsecs.push_back(x: {.debugChunk: debugChunk, .subsecData: subsec, .relocIndex: nextRelocIndex}); |
907 | } |
908 | |
909 | static Expected<StringRef> |
910 | getFileName(const DebugStringTableSubsectionRef &strings, |
911 | const DebugChecksumsSubsectionRef &checksums, uint32_t fileID) { |
912 | auto iter = checksums.getArray().at(Offset: fileID); |
913 | if (iter == checksums.getArray().end()) |
914 | return make_error<CodeViewError>(Args: cv_error_code::no_records); |
915 | uint32_t offset = iter->FileNameOffset; |
916 | return strings.getString(Offset: offset); |
917 | } |
918 | |
919 | void DebugSHandler::finish() { |
920 | pdb::DbiStreamBuilder &dbiBuilder = linker.builder.getDbiBuilder(); |
921 | |
922 | // If we found any symbol records for the module symbol stream, defer them. |
923 | if (moduleStreamSize > kSymbolStreamMagicSize) |
924 | file.moduleDBI->addUnmergedSymbols(SymSrc: &file, SymLength: moduleStreamSize - |
925 | kSymbolStreamMagicSize); |
926 | |
927 | // We should have seen all debug subsections across the entire object file now |
928 | // which means that if a StringTable subsection and Checksums subsection were |
929 | // present, now is the time to handle them. |
930 | if (!cvStrTab.valid()) { |
931 | if (checksums.valid()) |
932 | fatal(msg: ".debug$S sections with a checksums subsection must also contain a " |
933 | "string table subsection" ); |
934 | |
935 | if (!stringTableFixups.empty()) |
936 | warn(msg: "No StringTable subsection was encountered, but there are string " |
937 | "table references" ); |
938 | return; |
939 | } |
940 | |
941 | ExitOnError exitOnErr; |
942 | |
943 | // Handle FPO data. Each subsection begins with a single image base |
944 | // relocation, which is then added to the RvaStart of each frame data record |
945 | // when it is added to the PDB. The string table indices for the FPO program |
946 | // must also be rewritten to use the PDB string table. |
947 | for (const UnrelocatedFpoData &subsec : frameDataSubsecs) { |
948 | // Relocate the first four bytes of the subection and reinterpret them as a |
949 | // 32 bit little-endian integer. |
950 | SectionChunk *debugChunk = subsec.debugChunk; |
951 | ArrayRef<uint8_t> subsecData = subsec.subsecData; |
952 | uint32_t relocIndex = subsec.relocIndex; |
953 | auto unrelocatedRvaStart = subsecData.take_front(N: sizeof(uint32_t)); |
954 | uint8_t relocatedRvaStart[sizeof(uint32_t)]; |
955 | debugChunk->writeAndRelocateSubsection(sec: debugChunk->getContents(), |
956 | subsec: unrelocatedRvaStart, nextRelocIndex&: relocIndex, |
957 | buf: &relocatedRvaStart[0]); |
958 | // Use of memcpy here avoids violating type-based aliasing rules. |
959 | support::ulittle32_t rvaStart; |
960 | memcpy(dest: &rvaStart, src: &relocatedRvaStart[0], n: sizeof(support::ulittle32_t)); |
961 | |
962 | // Copy each frame data record, add in rvaStart, translate string table |
963 | // indices, and add the record to the PDB. |
964 | DebugFrameDataSubsectionRef fds; |
965 | BinaryStreamReader reader(subsecData, llvm::endianness::little); |
966 | exitOnErr(fds.initialize(Reader: reader)); |
967 | for (codeview::FrameData fd : fds) { |
968 | fd.RvaStart += rvaStart; |
969 | fd.FrameFunc = |
970 | translateStringTableIndex(objIndex: fd.FrameFunc, objStrTable: cvStrTab, pdbStrTable&: linker.pdbStrTab); |
971 | dbiBuilder.addNewFpoData(FD: fd); |
972 | } |
973 | } |
974 | |
975 | // Translate the fixups and pass them off to the module builder so they will |
976 | // be applied during writing. |
977 | for (StringTableFixup &ref : stringTableFixups) { |
978 | ref.StrTabOffset = |
979 | translateStringTableIndex(objIndex: ref.StrTabOffset, objStrTable: cvStrTab, pdbStrTable&: linker.pdbStrTab); |
980 | } |
981 | file.moduleDBI->setStringTableFixups(std::move(stringTableFixups)); |
982 | |
983 | // Make a new file checksum table that refers to offsets in the PDB-wide |
984 | // string table. Generally the string table subsection appears after the |
985 | // checksum table, so we have to do this after looping over all the |
986 | // subsections. The new checksum table must have the exact same layout and |
987 | // size as the original. Otherwise, the file references in the line and |
988 | // inlinee line tables will be incorrect. |
989 | auto newChecksums = std::make_unique<DebugChecksumsSubsection>(args&: linker.pdbStrTab); |
990 | for (const FileChecksumEntry &fc : checksums) { |
991 | SmallString<128> filename = |
992 | exitOnErr(cvStrTab.getString(Offset: fc.FileNameOffset)); |
993 | linker.pdbMakeAbsolute(fileName&: filename); |
994 | exitOnErr(dbiBuilder.addModuleSourceFile(Module&: *file.moduleDBI, File: filename)); |
995 | newChecksums->addChecksum(FileName: filename, Kind: fc.Kind, Bytes: fc.Checksum); |
996 | } |
997 | assert(checksums.getArray().getUnderlyingStream().getLength() == |
998 | newChecksums->calculateSerializedSize() && |
999 | "file checksum table must have same layout" ); |
1000 | |
1001 | file.moduleDBI->addDebugSubsection(Subsection: std::move(newChecksums)); |
1002 | } |
1003 | |
1004 | static void warnUnusable(InputFile *f, Error e, bool shouldWarn) { |
1005 | if (!shouldWarn) { |
1006 | consumeError(Err: std::move(e)); |
1007 | return; |
1008 | } |
1009 | auto msg = "Cannot use debug info for '" + toString(file: f) + "' [LNK4099]" ; |
1010 | if (e) |
1011 | warn(msg: msg + "\n>>> failed to load reference " + toString(E: std::move(e))); |
1012 | else |
1013 | warn(msg); |
1014 | } |
1015 | |
1016 | // Allocate memory for a .debug$S / .debug$F section and relocate it. |
1017 | static ArrayRef<uint8_t> relocateDebugChunk(SectionChunk &debugChunk) { |
1018 | uint8_t *buffer = bAlloc().Allocate<uint8_t>(Num: debugChunk.getSize()); |
1019 | assert(debugChunk.getOutputSectionIdx() == 0 && |
1020 | "debug sections should not be in output sections" ); |
1021 | debugChunk.writeTo(buf: buffer); |
1022 | return ArrayRef(buffer, debugChunk.getSize()); |
1023 | } |
1024 | |
1025 | void PDBLinker::addDebugSymbols(TpiSource *source) { |
1026 | // If this TpiSource doesn't have an object file, it must be from a type |
1027 | // server PDB. Type server PDBs do not contain symbols, so stop here. |
1028 | if (!source->file) |
1029 | return; |
1030 | |
1031 | llvm::TimeTraceScope timeScope("Merge symbols" ); |
1032 | ScopedTimer t(ctx.symbolMergingTimer); |
1033 | ExitOnError exitOnErr; |
1034 | pdb::DbiStreamBuilder &dbiBuilder = builder.getDbiBuilder(); |
1035 | DebugSHandler dsh(*this, *source->file); |
1036 | // Now do all live .debug$S and .debug$F sections. |
1037 | for (SectionChunk *debugChunk : source->file->getDebugChunks()) { |
1038 | if (!debugChunk->live || debugChunk->getSize() == 0) |
1039 | continue; |
1040 | |
1041 | bool isDebugS = debugChunk->getSectionName() == ".debug$S" ; |
1042 | bool isDebugF = debugChunk->getSectionName() == ".debug$F" ; |
1043 | if (!isDebugS && !isDebugF) |
1044 | continue; |
1045 | |
1046 | if (isDebugS) { |
1047 | dsh.handleDebugS(debugChunk); |
1048 | } else if (isDebugF) { |
1049 | // Handle old FPO data .debug$F sections. These are relatively rare. |
1050 | ArrayRef<uint8_t> relocatedDebugContents = |
1051 | relocateDebugChunk(debugChunk&: *debugChunk); |
1052 | FixedStreamArray<object::FpoData> fpoRecords; |
1053 | BinaryStreamReader reader(relocatedDebugContents, |
1054 | llvm::endianness::little); |
1055 | uint32_t count = relocatedDebugContents.size() / sizeof(object::FpoData); |
1056 | exitOnErr(reader.readArray(Array&: fpoRecords, NumItems: count)); |
1057 | |
1058 | // These are already relocated and don't refer to the string table, so we |
1059 | // can just copy it. |
1060 | for (const object::FpoData &fd : fpoRecords) |
1061 | dbiBuilder.addOldFpoData(Fpo: fd); |
1062 | } |
1063 | } |
1064 | |
1065 | // Do any post-processing now that all .debug$S sections have been processed. |
1066 | dsh.finish(); |
1067 | } |
1068 | |
1069 | // Add a module descriptor for every object file. We need to put an absolute |
1070 | // path to the object into the PDB. If this is a plain object, we make its |
1071 | // path absolute. If it's an object in an archive, we make the archive path |
1072 | // absolute. |
1073 | void PDBLinker::createModuleDBI(ObjFile *file) { |
1074 | pdb::DbiStreamBuilder &dbiBuilder = builder.getDbiBuilder(); |
1075 | SmallString<128> objName; |
1076 | ExitOnError exitOnErr; |
1077 | |
1078 | bool inArchive = !file->parentName.empty(); |
1079 | objName = inArchive ? file->parentName : file->getName(); |
1080 | pdbMakeAbsolute(fileName&: objName); |
1081 | StringRef modName = inArchive ? file->getName() : objName.str(); |
1082 | |
1083 | file->moduleDBI = &exitOnErr(dbiBuilder.addModuleInfo(ModuleName: modName)); |
1084 | file->moduleDBI->setObjFileName(objName); |
1085 | file->moduleDBI->setMergeSymbolsCallback(Ctx: this, Callback: &commitSymbolsForObject); |
1086 | |
1087 | ArrayRef<Chunk *> chunks = file->getChunks(); |
1088 | uint32_t modi = file->moduleDBI->getModuleIndex(); |
1089 | |
1090 | for (Chunk *c : chunks) { |
1091 | auto *secChunk = dyn_cast<SectionChunk>(Val: c); |
1092 | if (!secChunk || !secChunk->live) |
1093 | continue; |
1094 | pdb::SectionContrib sc = createSectionContrib(ctx, c: secChunk, modi); |
1095 | file->moduleDBI->setFirstSectionContrib(sc); |
1096 | break; |
1097 | } |
1098 | } |
1099 | |
1100 | void PDBLinker::addDebug(TpiSource *source) { |
1101 | // Before we can process symbol substreams from .debug$S, we need to process |
1102 | // type information, file checksums, and the string table. Add type info to |
1103 | // the PDB first, so that we can get the map from object file type and item |
1104 | // indices to PDB type and item indices. If we are using ghashes, types have |
1105 | // already been merged. |
1106 | if (!ctx.config.debugGHashes) { |
1107 | llvm::TimeTraceScope timeScope("Merge types (Non-GHASH)" ); |
1108 | ScopedTimer t(ctx.typeMergingTimer); |
1109 | if (Error e = source->mergeDebugT(m: &tMerger)) { |
1110 | // If type merging failed, ignore the symbols. |
1111 | warnUnusable(f: source->file, e: std::move(e), |
1112 | shouldWarn: ctx.config.warnDebugInfoUnusable); |
1113 | return; |
1114 | } |
1115 | } |
1116 | |
1117 | // If type merging failed, ignore the symbols. |
1118 | Error typeError = std::move(source->typeMergingError); |
1119 | if (typeError) { |
1120 | warnUnusable(f: source->file, e: std::move(typeError), |
1121 | shouldWarn: ctx.config.warnDebugInfoUnusable); |
1122 | return; |
1123 | } |
1124 | |
1125 | addDebugSymbols(source); |
1126 | } |
1127 | |
1128 | static pdb::BulkPublic createPublic(COFFLinkerContext &ctx, Defined *def) { |
1129 | pdb::BulkPublic pub; |
1130 | pub.Name = def->getName().data(); |
1131 | pub.NameLen = def->getName().size(); |
1132 | |
1133 | PublicSymFlags flags = PublicSymFlags::None; |
1134 | if (auto *d = dyn_cast<DefinedCOFF>(Val: def)) { |
1135 | if (d->getCOFFSymbol().isFunctionDefinition()) |
1136 | flags = PublicSymFlags::Function; |
1137 | } else if (isa<DefinedImportThunk>(Val: def)) { |
1138 | flags = PublicSymFlags::Function; |
1139 | } |
1140 | pub.setFlags(flags); |
1141 | |
1142 | OutputSection *os = ctx.getOutputSection(c: def->getChunk()); |
1143 | assert(os && "all publics should be in final image" ); |
1144 | pub.Offset = def->getRVA() - os->getRVA(); |
1145 | pub.Segment = os->sectionIndex; |
1146 | return pub; |
1147 | } |
1148 | |
1149 | // Add all object files to the PDB. Merge .debug$T sections into IpiData and |
1150 | // TpiData. |
1151 | void PDBLinker::addObjectsToPDB() { |
1152 | { |
1153 | llvm::TimeTraceScope timeScope("Add objects to PDB" ); |
1154 | ScopedTimer t1(ctx.addObjectsTimer); |
1155 | |
1156 | // Create module descriptors |
1157 | for (ObjFile *obj : ctx.objFileInstances) |
1158 | createModuleDBI(file: obj); |
1159 | |
1160 | // Reorder dependency type sources to come first. |
1161 | tMerger.sortDependencies(); |
1162 | |
1163 | // Merge type information from input files using global type hashing. |
1164 | if (ctx.config.debugGHashes) |
1165 | tMerger.mergeTypesWithGHash(); |
1166 | |
1167 | // Merge dependencies and then regular objects. |
1168 | { |
1169 | llvm::TimeTraceScope timeScope("Merge debug info (dependencies)" ); |
1170 | for (TpiSource *source : tMerger.dependencySources) |
1171 | addDebug(source); |
1172 | } |
1173 | { |
1174 | llvm::TimeTraceScope timeScope("Merge debug info (objects)" ); |
1175 | for (TpiSource *source : tMerger.objectSources) |
1176 | addDebug(source); |
1177 | } |
1178 | |
1179 | builder.getStringTableBuilder().setStrings(pdbStrTab); |
1180 | } |
1181 | |
1182 | // Construct TPI and IPI stream contents. |
1183 | { |
1184 | llvm::TimeTraceScope timeScope("TPI/IPI stream layout" ); |
1185 | ScopedTimer t2(ctx.tpiStreamLayoutTimer); |
1186 | |
1187 | // Collect all the merged types. |
1188 | if (ctx.config.debugGHashes) { |
1189 | addGHashTypeInfo(ctx, builder); |
1190 | } else { |
1191 | addTypeInfo(tpiBuilder&: builder.getTpiBuilder(), typeTable&: tMerger.getTypeTable()); |
1192 | addTypeInfo(tpiBuilder&: builder.getIpiBuilder(), typeTable&: tMerger.getIDTable()); |
1193 | } |
1194 | } |
1195 | |
1196 | if (ctx.config.showSummary) { |
1197 | for (TpiSource *source : ctx.tpiSourceList) { |
1198 | nbTypeRecords += source->nbTypeRecords; |
1199 | nbTypeRecordsBytes += source->nbTypeRecordsBytes; |
1200 | } |
1201 | } |
1202 | } |
1203 | |
1204 | void PDBLinker::addPublicsToPDB() { |
1205 | llvm::TimeTraceScope timeScope("Publics layout" ); |
1206 | ScopedTimer t3(ctx.publicsLayoutTimer); |
1207 | // Compute the public symbols. |
1208 | auto &gsiBuilder = builder.getGsiBuilder(); |
1209 | std::vector<pdb::BulkPublic> publics; |
1210 | ctx.symtab.forEachSymbol(callback: [&publics, this](Symbol *s) { |
1211 | // Only emit external, defined, live symbols that have a chunk. Static, |
1212 | // non-external symbols do not appear in the symbol table. |
1213 | auto *def = dyn_cast<Defined>(Val: s); |
1214 | if (def && def->isLive() && def->getChunk()) { |
1215 | // Don't emit a public symbol for coverage data symbols. LLVM code |
1216 | // coverage (and PGO) create a __profd_ and __profc_ symbol for every |
1217 | // function. C++ mangled names are long, and tend to dominate symbol size. |
1218 | // Including these names triples the size of the public stream, which |
1219 | // results in bloated PDB files. These symbols generally are not helpful |
1220 | // for debugging, so suppress them. |
1221 | StringRef name = def->getName(); |
1222 | if (name.data()[0] == '_' && name.data()[1] == '_') { |
1223 | // Drop the '_' prefix for x86. |
1224 | if (ctx.config.machine == I386) |
1225 | name = name.drop_front(N: 1); |
1226 | if (name.starts_with(Prefix: "__profd_" ) || name.starts_with(Prefix: "__profc_" ) || |
1227 | name.starts_with(Prefix: "__covrec_" )) { |
1228 | return; |
1229 | } |
1230 | } |
1231 | publics.push_back(x: createPublic(ctx, def)); |
1232 | } |
1233 | }); |
1234 | |
1235 | if (!publics.empty()) { |
1236 | publicSymbols = publics.size(); |
1237 | gsiBuilder.addPublicSymbols(PublicsIn: std::move(publics)); |
1238 | } |
1239 | } |
1240 | |
1241 | void PDBLinker::printStats() { |
1242 | if (!ctx.config.showSummary) |
1243 | return; |
1244 | |
1245 | SmallString<256> buffer; |
1246 | raw_svector_ostream stream(buffer); |
1247 | |
1248 | stream << center_justify(Str: "Summary" , Width: 80) << '\n' |
1249 | << std::string(80, '-') << '\n'; |
1250 | |
1251 | auto print = [&](uint64_t v, StringRef s) { |
1252 | stream << format_decimal(N: v, Width: 15) << " " << s << '\n'; |
1253 | }; |
1254 | |
1255 | print(ctx.objFileInstances.size(), |
1256 | "Input OBJ files (expanded from all cmd-line inputs)" ); |
1257 | print(ctx.typeServerSourceMappings.size(), "PDB type server dependencies" ); |
1258 | print(ctx.precompSourceMappings.size(), "Precomp OBJ dependencies" ); |
1259 | print(nbTypeRecords, "Input type records" ); |
1260 | print(nbTypeRecordsBytes, "Input type records bytes" ); |
1261 | print(builder.getTpiBuilder().getRecordCount(), "Merged TPI records" ); |
1262 | print(builder.getIpiBuilder().getRecordCount(), "Merged IPI records" ); |
1263 | print(pdbStrTab.size(), "Output PDB strings" ); |
1264 | print(globalSymbols, "Global symbol records" ); |
1265 | print(moduleSymbols, "Module symbol records" ); |
1266 | print(publicSymbols, "Public symbol records" ); |
1267 | |
1268 | auto printLargeInputTypeRecs = [&](StringRef name, |
1269 | ArrayRef<uint32_t> recCounts, |
1270 | TypeCollection &records) { |
1271 | // Figure out which type indices were responsible for the most duplicate |
1272 | // bytes in the input files. These should be frequently emitted LF_CLASS and |
1273 | // LF_FIELDLIST records. |
1274 | struct TypeSizeInfo { |
1275 | uint32_t typeSize; |
1276 | uint32_t dupCount; |
1277 | TypeIndex typeIndex; |
1278 | uint64_t totalInputSize() const { return uint64_t(dupCount) * typeSize; } |
1279 | bool operator<(const TypeSizeInfo &rhs) const { |
1280 | if (totalInputSize() == rhs.totalInputSize()) |
1281 | return typeIndex < rhs.typeIndex; |
1282 | return totalInputSize() < rhs.totalInputSize(); |
1283 | } |
1284 | }; |
1285 | SmallVector<TypeSizeInfo, 0> tsis; |
1286 | for (auto e : enumerate(First&: recCounts)) { |
1287 | TypeIndex typeIndex = TypeIndex::fromArrayIndex(Index: e.index()); |
1288 | uint32_t typeSize = records.getType(Index: typeIndex).length(); |
1289 | uint32_t dupCount = e.value(); |
1290 | tsis.push_back(Elt: {.typeSize: typeSize, .dupCount: dupCount, .typeIndex: typeIndex}); |
1291 | } |
1292 | |
1293 | if (!tsis.empty()) { |
1294 | stream << "\nTop 10 types responsible for the most " << name |
1295 | << " input:\n" ; |
1296 | stream << " index total bytes count size\n" ; |
1297 | llvm::sort(C&: tsis); |
1298 | unsigned i = 0; |
1299 | for (const auto &tsi : reverse(C&: tsis)) { |
1300 | stream << formatv(Fmt: " {0,10:X}: {1,14:N} = {2,5:N} * {3,6:N}\n" , |
1301 | Vals: tsi.typeIndex.getIndex(), Vals: tsi.totalInputSize(), |
1302 | Vals: tsi.dupCount, Vals: tsi.typeSize); |
1303 | if (++i >= 10) |
1304 | break; |
1305 | } |
1306 | stream |
1307 | << "Run llvm-pdbutil to print details about a particular record:\n" ; |
1308 | stream << formatv(Fmt: "llvm-pdbutil dump -{0}s -{0}-index {1:X} {2}\n" , |
1309 | Vals: (name == "TPI" ? "type" : "id" ), |
1310 | Vals: tsis.back().typeIndex.getIndex(), Vals&: ctx.config.pdbPath); |
1311 | } |
1312 | }; |
1313 | |
1314 | if (!ctx.config.debugGHashes) { |
1315 | // FIXME: Reimplement for ghash. |
1316 | printLargeInputTypeRecs("TPI" , tMerger.tpiCounts, tMerger.getTypeTable()); |
1317 | printLargeInputTypeRecs("IPI" , tMerger.ipiCounts, tMerger.getIDTable()); |
1318 | } |
1319 | |
1320 | message(msg: buffer); |
1321 | } |
1322 | |
1323 | void PDBLinker::addNatvisFiles() { |
1324 | llvm::TimeTraceScope timeScope("Natvis files" ); |
1325 | for (StringRef file : ctx.config.natvisFiles) { |
1326 | ErrorOr<std::unique_ptr<MemoryBuffer>> dataOrErr = |
1327 | MemoryBuffer::getFile(Filename: file); |
1328 | if (!dataOrErr) { |
1329 | warn(msg: "Cannot open input file: " + file); |
1330 | continue; |
1331 | } |
1332 | std::unique_ptr<MemoryBuffer> data = std::move(*dataOrErr); |
1333 | |
1334 | // Can't use takeBuffer() here since addInjectedSource() takes ownership. |
1335 | if (ctx.driver.tar) |
1336 | ctx.driver.tar->append(Path: relativeToRoot(path: data->getBufferIdentifier()), |
1337 | Data: data->getBuffer()); |
1338 | |
1339 | builder.addInjectedSource(Name: file, Buffer: std::move(data)); |
1340 | } |
1341 | } |
1342 | |
1343 | void PDBLinker::addNamedStreams() { |
1344 | llvm::TimeTraceScope timeScope("Named streams" ); |
1345 | ExitOnError exitOnErr; |
1346 | for (const auto &streamFile : ctx.config.namedStreams) { |
1347 | const StringRef stream = streamFile.getKey(), file = streamFile.getValue(); |
1348 | ErrorOr<std::unique_ptr<MemoryBuffer>> dataOrErr = |
1349 | MemoryBuffer::getFile(Filename: file); |
1350 | if (!dataOrErr) { |
1351 | warn(msg: "Cannot open input file: " + file); |
1352 | continue; |
1353 | } |
1354 | std::unique_ptr<MemoryBuffer> data = std::move(*dataOrErr); |
1355 | exitOnErr(builder.addNamedStream(Name: stream, Data: data->getBuffer())); |
1356 | ctx.driver.takeBuffer(mb: std::move(data)); |
1357 | } |
1358 | } |
1359 | |
1360 | static codeview::CPUType toCodeViewMachine(COFF::MachineTypes machine) { |
1361 | switch (machine) { |
1362 | case COFF::IMAGE_FILE_MACHINE_AMD64: |
1363 | return codeview::CPUType::X64; |
1364 | case COFF::IMAGE_FILE_MACHINE_ARM: |
1365 | return codeview::CPUType::ARM7; |
1366 | case COFF::IMAGE_FILE_MACHINE_ARM64: |
1367 | return codeview::CPUType::ARM64; |
1368 | case COFF::IMAGE_FILE_MACHINE_ARMNT: |
1369 | return codeview::CPUType::ARMNT; |
1370 | case COFF::IMAGE_FILE_MACHINE_I386: |
1371 | return codeview::CPUType::Intel80386; |
1372 | default: |
1373 | llvm_unreachable("Unsupported CPU Type" ); |
1374 | } |
1375 | } |
1376 | |
1377 | // Mimic MSVC which surrounds arguments containing whitespace with quotes. |
1378 | // Double double-quotes are handled, so that the resulting string can be |
1379 | // executed again on the cmd-line. |
1380 | static std::string quote(ArrayRef<StringRef> args) { |
1381 | std::string r; |
1382 | r.reserve(res_arg: 256); |
1383 | for (StringRef a : args) { |
1384 | if (!r.empty()) |
1385 | r.push_back(c: ' '); |
1386 | bool hasWS = a.contains(C: ' '); |
1387 | bool hasQ = a.contains(C: '"'); |
1388 | if (hasWS || hasQ) |
1389 | r.push_back(c: '"'); |
1390 | if (hasQ) { |
1391 | SmallVector<StringRef, 4> s; |
1392 | a.split(A&: s, Separator: '"'); |
1393 | r.append(str: join(R&: s, Separator: "\"\"" )); |
1394 | } else { |
1395 | r.append(str: std::string(a)); |
1396 | } |
1397 | if (hasWS || hasQ) |
1398 | r.push_back(c: '"'); |
1399 | } |
1400 | return r; |
1401 | } |
1402 | |
1403 | static void fillLinkerVerRecord(Compile3Sym &cs, MachineTypes machine) { |
1404 | cs.Machine = toCodeViewMachine(machine); |
1405 | // Interestingly, if we set the string to 0.0.0.0, then when trying to view |
1406 | // local variables WinDbg emits an error that private symbols are not present. |
1407 | // By setting this to a valid MSVC linker version string, local variables are |
1408 | // displayed properly. As such, even though it is not representative of |
1409 | // LLVM's version information, we need this for compatibility. |
1410 | cs.Flags = CompileSym3Flags::None; |
1411 | cs.VersionBackendBuild = 25019; |
1412 | cs.VersionBackendMajor = 14; |
1413 | cs.VersionBackendMinor = 10; |
1414 | cs.VersionBackendQFE = 0; |
1415 | |
1416 | // MSVC also sets the frontend to 0.0.0.0 since this is specifically for the |
1417 | // linker module (which is by definition a backend), so we don't need to do |
1418 | // anything here. Also, it seems we can use "LLVM Linker" for the linker name |
1419 | // without any problems. Only the backend version has to be hardcoded to a |
1420 | // magic number. |
1421 | cs.VersionFrontendBuild = 0; |
1422 | cs.VersionFrontendMajor = 0; |
1423 | cs.VersionFrontendMinor = 0; |
1424 | cs.VersionFrontendQFE = 0; |
1425 | cs.Version = "LLVM Linker" ; |
1426 | cs.setLanguage(SourceLanguage::Link); |
1427 | } |
1428 | |
1429 | void PDBLinker::addCommonLinkerModuleSymbols( |
1430 | StringRef path, pdb::DbiModuleDescriptorBuilder &mod) { |
1431 | ObjNameSym ons(SymbolRecordKind::ObjNameSym); |
1432 | EnvBlockSym ebs(SymbolRecordKind::EnvBlockSym); |
1433 | Compile3Sym cs(SymbolRecordKind::Compile3Sym); |
1434 | fillLinkerVerRecord(cs, machine: ctx.config.machine); |
1435 | |
1436 | ons.Name = "* Linker *" ; |
1437 | ons.Signature = 0; |
1438 | |
1439 | ArrayRef<StringRef> args = ArrayRef(ctx.config.argv).drop_front(); |
1440 | std::string argStr = quote(args); |
1441 | ebs.Fields.push_back(x: "cwd" ); |
1442 | SmallString<64> cwd; |
1443 | if (ctx.config.pdbSourcePath.empty()) |
1444 | sys::fs::current_path(result&: cwd); |
1445 | else |
1446 | cwd = ctx.config.pdbSourcePath; |
1447 | ebs.Fields.push_back(x: cwd); |
1448 | ebs.Fields.push_back(x: "exe" ); |
1449 | SmallString<64> exe = ctx.config.argv[0]; |
1450 | pdbMakeAbsolute(fileName&: exe); |
1451 | ebs.Fields.push_back(x: exe); |
1452 | ebs.Fields.push_back(x: "pdb" ); |
1453 | ebs.Fields.push_back(x: path); |
1454 | ebs.Fields.push_back(x: "cmd" ); |
1455 | ebs.Fields.push_back(x: argStr); |
1456 | llvm::BumpPtrAllocator &bAlloc = lld::bAlloc(); |
1457 | mod.addSymbol(Symbol: codeview::SymbolSerializer::writeOneSymbol( |
1458 | Sym&: ons, Storage&: bAlloc, Container: CodeViewContainer::Pdb)); |
1459 | mod.addSymbol(Symbol: codeview::SymbolSerializer::writeOneSymbol( |
1460 | Sym&: cs, Storage&: bAlloc, Container: CodeViewContainer::Pdb)); |
1461 | mod.addSymbol(Symbol: codeview::SymbolSerializer::writeOneSymbol( |
1462 | Sym&: ebs, Storage&: bAlloc, Container: CodeViewContainer::Pdb)); |
1463 | } |
1464 | |
1465 | static void addLinkerModuleCoffGroup(PartialSection *sec, |
1466 | pdb::DbiModuleDescriptorBuilder &mod, |
1467 | OutputSection &os) { |
1468 | // If there's a section, there's at least one chunk |
1469 | assert(!sec->chunks.empty()); |
1470 | const Chunk *firstChunk = *sec->chunks.begin(); |
1471 | const Chunk *lastChunk = *sec->chunks.rbegin(); |
1472 | |
1473 | // Emit COFF group |
1474 | CoffGroupSym cgs(SymbolRecordKind::CoffGroupSym); |
1475 | cgs.Name = sec->name; |
1476 | cgs.Segment = os.sectionIndex; |
1477 | cgs.Offset = firstChunk->getRVA() - os.getRVA(); |
1478 | cgs.Size = lastChunk->getRVA() + lastChunk->getSize() - firstChunk->getRVA(); |
1479 | cgs.Characteristics = sec->characteristics; |
1480 | |
1481 | // Somehow .idata sections & sections groups in the debug symbol stream have |
1482 | // the "write" flag set. However the section header for the corresponding |
1483 | // .idata section doesn't have it. |
1484 | if (cgs.Name.starts_with(Prefix: ".idata" )) |
1485 | cgs.Characteristics |= llvm::COFF::IMAGE_SCN_MEM_WRITE; |
1486 | |
1487 | mod.addSymbol(Symbol: codeview::SymbolSerializer::writeOneSymbol( |
1488 | Sym&: cgs, Storage&: bAlloc(), Container: CodeViewContainer::Pdb)); |
1489 | } |
1490 | |
1491 | static void addLinkerModuleSectionSymbol(pdb::DbiModuleDescriptorBuilder &mod, |
1492 | OutputSection &os, bool isMinGW) { |
1493 | SectionSym sym(SymbolRecordKind::SectionSym); |
1494 | sym.Alignment = 12; // 2^12 = 4KB |
1495 | sym.Characteristics = os.header.Characteristics; |
1496 | sym.Length = os.getVirtualSize(); |
1497 | sym.Name = os.name; |
1498 | sym.Rva = os.getRVA(); |
1499 | sym.SectionNumber = os.sectionIndex; |
1500 | mod.addSymbol(Symbol: codeview::SymbolSerializer::writeOneSymbol( |
1501 | Sym&: sym, Storage&: bAlloc(), Container: CodeViewContainer::Pdb)); |
1502 | |
1503 | // Skip COFF groups in MinGW because it adds a significant footprint to the |
1504 | // PDB, due to each function being in its own section |
1505 | if (isMinGW) |
1506 | return; |
1507 | |
1508 | // Output COFF groups for individual chunks of this section. |
1509 | for (PartialSection *sec : os.contribSections) { |
1510 | addLinkerModuleCoffGroup(sec, mod, os); |
1511 | } |
1512 | } |
1513 | |
1514 | // Add all import files as modules to the PDB. |
1515 | void PDBLinker::addImportFilesToPDB() { |
1516 | if (ctx.importFileInstances.empty()) |
1517 | return; |
1518 | |
1519 | llvm::TimeTraceScope timeScope("Import files" ); |
1520 | ExitOnError exitOnErr; |
1521 | std::map<std::string, llvm::pdb::DbiModuleDescriptorBuilder *> dllToModuleDbi; |
1522 | |
1523 | for (ImportFile *file : ctx.importFileInstances) { |
1524 | if (!file->live) |
1525 | continue; |
1526 | |
1527 | if (!file->thunkSym) |
1528 | continue; |
1529 | |
1530 | if (!file->thunkLive) |
1531 | continue; |
1532 | |
1533 | std::string dll = StringRef(file->dllName).lower(); |
1534 | llvm::pdb::DbiModuleDescriptorBuilder *&mod = dllToModuleDbi[dll]; |
1535 | if (!mod) { |
1536 | pdb::DbiStreamBuilder &dbiBuilder = builder.getDbiBuilder(); |
1537 | SmallString<128> libPath = file->parentName; |
1538 | pdbMakeAbsolute(fileName&: libPath); |
1539 | sys::path::native(path&: libPath); |
1540 | |
1541 | // Name modules similar to MSVC's link.exe. |
1542 | // The first module is the simple dll filename |
1543 | llvm::pdb::DbiModuleDescriptorBuilder &firstMod = |
1544 | exitOnErr(dbiBuilder.addModuleInfo(ModuleName: file->dllName)); |
1545 | firstMod.setObjFileName(libPath); |
1546 | pdb::SectionContrib sc = |
1547 | createSectionContrib(ctx, c: nullptr, modi: llvm::pdb::kInvalidStreamIndex); |
1548 | firstMod.setFirstSectionContrib(sc); |
1549 | |
1550 | // The second module is where the import stream goes. |
1551 | mod = &exitOnErr(dbiBuilder.addModuleInfo(ModuleName: "Import:" + file->dllName)); |
1552 | mod->setObjFileName(libPath); |
1553 | } |
1554 | |
1555 | DefinedImportThunk *thunk = cast<DefinedImportThunk>(Val: file->thunkSym); |
1556 | Chunk *thunkChunk = thunk->getChunk(); |
1557 | OutputSection *thunkOS = ctx.getOutputSection(c: thunkChunk); |
1558 | |
1559 | ObjNameSym ons(SymbolRecordKind::ObjNameSym); |
1560 | Compile3Sym cs(SymbolRecordKind::Compile3Sym); |
1561 | Thunk32Sym ts(SymbolRecordKind::Thunk32Sym); |
1562 | ScopeEndSym es(SymbolRecordKind::ScopeEndSym); |
1563 | |
1564 | ons.Name = file->dllName; |
1565 | ons.Signature = 0; |
1566 | |
1567 | fillLinkerVerRecord(cs, machine: ctx.config.machine); |
1568 | |
1569 | ts.Name = thunk->getName(); |
1570 | ts.Parent = 0; |
1571 | ts.End = 0; |
1572 | ts.Next = 0; |
1573 | ts.Thunk = ThunkOrdinal::Standard; |
1574 | ts.Length = thunkChunk->getSize(); |
1575 | ts.Segment = thunkOS->sectionIndex; |
1576 | ts.Offset = thunkChunk->getRVA() - thunkOS->getRVA(); |
1577 | |
1578 | llvm::BumpPtrAllocator &bAlloc = lld::bAlloc(); |
1579 | mod->addSymbol(Symbol: codeview::SymbolSerializer::writeOneSymbol( |
1580 | Sym&: ons, Storage&: bAlloc, Container: CodeViewContainer::Pdb)); |
1581 | mod->addSymbol(Symbol: codeview::SymbolSerializer::writeOneSymbol( |
1582 | Sym&: cs, Storage&: bAlloc, Container: CodeViewContainer::Pdb)); |
1583 | |
1584 | CVSymbol newSym = codeview::SymbolSerializer::writeOneSymbol( |
1585 | Sym&: ts, Storage&: bAlloc, Container: CodeViewContainer::Pdb); |
1586 | |
1587 | // Write ptrEnd for the S_THUNK32. |
1588 | ScopeRecord *thunkSymScope = |
1589 | getSymbolScopeFields(sym: const_cast<uint8_t *>(newSym.data().data())); |
1590 | |
1591 | mod->addSymbol(Symbol: newSym); |
1592 | |
1593 | newSym = codeview::SymbolSerializer::writeOneSymbol(Sym&: es, Storage&: bAlloc, |
1594 | Container: CodeViewContainer::Pdb); |
1595 | thunkSymScope->ptrEnd = mod->getNextSymbolOffset(); |
1596 | |
1597 | mod->addSymbol(Symbol: newSym); |
1598 | |
1599 | pdb::SectionContrib sc = |
1600 | createSectionContrib(ctx, c: thunk->getChunk(), modi: mod->getModuleIndex()); |
1601 | mod->setFirstSectionContrib(sc); |
1602 | } |
1603 | } |
1604 | |
1605 | // Creates a PDB file. |
1606 | void lld::coff::createPDB(COFFLinkerContext &ctx, |
1607 | ArrayRef<uint8_t> sectionTable, |
1608 | llvm::codeview::DebugInfo *buildId) { |
1609 | llvm::TimeTraceScope timeScope("PDB file" ); |
1610 | ScopedTimer t1(ctx.totalPdbLinkTimer); |
1611 | { |
1612 | PDBLinker pdb(ctx); |
1613 | |
1614 | pdb.initialize(buildId); |
1615 | pdb.addObjectsToPDB(); |
1616 | pdb.addImportFilesToPDB(); |
1617 | pdb.addSections(sectionTable); |
1618 | pdb.addNatvisFiles(); |
1619 | pdb.addNamedStreams(); |
1620 | pdb.addPublicsToPDB(); |
1621 | |
1622 | { |
1623 | llvm::TimeTraceScope timeScope("Commit PDB file to disk" ); |
1624 | ScopedTimer t2(ctx.diskCommitTimer); |
1625 | codeview::GUID guid; |
1626 | pdb.commit(guid: &guid); |
1627 | memcpy(dest: &buildId->PDB70.Signature, src: &guid, n: 16); |
1628 | } |
1629 | |
1630 | t1.stop(); |
1631 | pdb.printStats(); |
1632 | |
1633 | // Manually start this profile point to measure ~PDBLinker(). |
1634 | if (getTimeTraceProfilerInstance() != nullptr) |
1635 | timeTraceProfilerBegin(Name: "PDBLinker destructor" , Detail: StringRef("" )); |
1636 | } |
1637 | // Manually end this profile point to measure ~PDBLinker(). |
1638 | if (getTimeTraceProfilerInstance() != nullptr) |
1639 | timeTraceProfilerEnd(); |
1640 | } |
1641 | |
1642 | void PDBLinker::initialize(llvm::codeview::DebugInfo *buildId) { |
1643 | ExitOnError exitOnErr; |
1644 | exitOnErr(builder.initialize(BlockSize: ctx.config.pdbPageSize)); |
1645 | |
1646 | buildId->Signature.CVSignature = OMF::Signature::PDB70; |
1647 | // Signature is set to a hash of the PDB contents when the PDB is done. |
1648 | memset(s: buildId->PDB70.Signature, c: 0, n: 16); |
1649 | buildId->PDB70.Age = 1; |
1650 | |
1651 | // Create streams in MSF for predefined streams, namely |
1652 | // PDB, TPI, DBI and IPI. |
1653 | for (int i = 0; i < (int)pdb::kSpecialStreamCount; ++i) |
1654 | exitOnErr(builder.getMsfBuilder().addStream(Size: 0)); |
1655 | |
1656 | // Add an Info stream. |
1657 | auto &infoBuilder = builder.getInfoBuilder(); |
1658 | infoBuilder.setVersion(pdb::PdbRaw_ImplVer::PdbImplVC70); |
1659 | infoBuilder.setHashPDBContentsToGUID(true); |
1660 | |
1661 | // Add an empty DBI stream. |
1662 | pdb::DbiStreamBuilder &dbiBuilder = builder.getDbiBuilder(); |
1663 | dbiBuilder.setAge(buildId->PDB70.Age); |
1664 | dbiBuilder.setVersionHeader(pdb::PdbDbiV70); |
1665 | dbiBuilder.setMachineType(ctx.config.machine); |
1666 | // Technically we are not link.exe 14.11, but there are known cases where |
1667 | // debugging tools on Windows expect Microsoft-specific version numbers or |
1668 | // they fail to work at all. Since we know we produce PDBs that are |
1669 | // compatible with LINK 14.11, we set that version number here. |
1670 | dbiBuilder.setBuildNumber(Major: 14, Minor: 11); |
1671 | } |
1672 | |
1673 | void PDBLinker::addSections(ArrayRef<uint8_t> sectionTable) { |
1674 | llvm::TimeTraceScope timeScope("PDB output sections" ); |
1675 | ExitOnError exitOnErr; |
1676 | // It's not entirely clear what this is, but the * Linker * module uses it. |
1677 | pdb::DbiStreamBuilder &dbiBuilder = builder.getDbiBuilder(); |
1678 | nativePath = ctx.config.pdbPath; |
1679 | pdbMakeAbsolute(fileName&: nativePath); |
1680 | uint32_t pdbFilePathNI = dbiBuilder.addECName(Name: nativePath); |
1681 | auto &linkerModule = exitOnErr(dbiBuilder.addModuleInfo(ModuleName: "* Linker *" )); |
1682 | linkerModule.setPdbFilePathNI(pdbFilePathNI); |
1683 | addCommonLinkerModuleSymbols(path: nativePath, mod&: linkerModule); |
1684 | |
1685 | // Add section contributions. They must be ordered by ascending RVA. |
1686 | for (OutputSection *os : ctx.outputSections) { |
1687 | addLinkerModuleSectionSymbol(mod&: linkerModule, os&: *os, isMinGW: ctx.config.mingw); |
1688 | for (Chunk *c : os->chunks) { |
1689 | pdb::SectionContrib sc = |
1690 | createSectionContrib(ctx, c, modi: linkerModule.getModuleIndex()); |
1691 | builder.getDbiBuilder().addSectionContrib(SC: sc); |
1692 | } |
1693 | } |
1694 | |
1695 | // The * Linker * first section contrib is only used along with /INCREMENTAL, |
1696 | // to provide trampolines thunks for incremental function patching. Set this |
1697 | // as "unused" because LLD doesn't support /INCREMENTAL link. |
1698 | pdb::SectionContrib sc = |
1699 | createSectionContrib(ctx, c: nullptr, modi: llvm::pdb::kInvalidStreamIndex); |
1700 | linkerModule.setFirstSectionContrib(sc); |
1701 | |
1702 | // Add Section Map stream. |
1703 | ArrayRef<object::coff_section> sections = { |
1704 | (const object::coff_section *)sectionTable.data(), |
1705 | sectionTable.size() / sizeof(object::coff_section)}; |
1706 | dbiBuilder.createSectionMap(SecHdrs: sections); |
1707 | |
1708 | // Add COFF section header stream. |
1709 | exitOnErr( |
1710 | dbiBuilder.addDbgStream(Type: pdb::DbgHeaderType::SectionHdr, Data: sectionTable)); |
1711 | } |
1712 | |
1713 | void PDBLinker::commit(codeview::GUID *guid) { |
1714 | // Print an error and continue if PDB writing fails. This is done mainly so |
1715 | // the user can see the output of /time and /summary, which is very helpful |
1716 | // when trying to figure out why a PDB file is too large. |
1717 | if (Error e = builder.commit(Filename: ctx.config.pdbPath, Guid: guid)) { |
1718 | e = handleErrors(E: std::move(e), |
1719 | Hs: [](const llvm::msf::MSFError &me) { |
1720 | error(msg: me.message()); |
1721 | if (me.isPageOverflow()) |
1722 | error(msg: "try setting a larger /pdbpagesize" ); |
1723 | }); |
1724 | checkError(e: std::move(e)); |
1725 | error(msg: "failed to write PDB file " + Twine(ctx.config.pdbPath)); |
1726 | } |
1727 | } |
1728 | |
1729 | static uint32_t getSecrelReloc(Triple::ArchType arch) { |
1730 | switch (arch) { |
1731 | case Triple::x86_64: |
1732 | return COFF::IMAGE_REL_AMD64_SECREL; |
1733 | case Triple::x86: |
1734 | return COFF::IMAGE_REL_I386_SECREL; |
1735 | case Triple::thumb: |
1736 | return COFF::IMAGE_REL_ARM_SECREL; |
1737 | case Triple::aarch64: |
1738 | return COFF::IMAGE_REL_ARM64_SECREL; |
1739 | default: |
1740 | llvm_unreachable("unknown machine type" ); |
1741 | } |
1742 | } |
1743 | |
1744 | // Try to find a line table for the given offset Addr into the given chunk C. |
1745 | // If a line table was found, the line table, the string and checksum tables |
1746 | // that are used to interpret the line table, and the offset of Addr in the line |
1747 | // table are stored in the output arguments. Returns whether a line table was |
1748 | // found. |
1749 | static bool findLineTable(const SectionChunk *c, uint32_t addr, |
1750 | DebugStringTableSubsectionRef &cvStrTab, |
1751 | DebugChecksumsSubsectionRef &checksums, |
1752 | DebugLinesSubsectionRef &lines, |
1753 | uint32_t &offsetInLinetable) { |
1754 | ExitOnError exitOnErr; |
1755 | const uint32_t secrelReloc = getSecrelReloc(arch: c->getArch()); |
1756 | |
1757 | for (SectionChunk *dbgC : c->file->getDebugChunks()) { |
1758 | if (dbgC->getSectionName() != ".debug$S" ) |
1759 | continue; |
1760 | |
1761 | // Build a mapping of SECREL relocations in dbgC that refer to `c`. |
1762 | DenseMap<uint32_t, uint32_t> secrels; |
1763 | for (const coff_relocation &r : dbgC->getRelocs()) { |
1764 | if (r.Type != secrelReloc) |
1765 | continue; |
1766 | |
1767 | if (auto *s = dyn_cast_or_null<DefinedRegular>( |
1768 | Val: c->file->getSymbols()[r.SymbolTableIndex])) |
1769 | if (s->getChunk() == c) |
1770 | secrels[r.VirtualAddress] = s->getValue(); |
1771 | } |
1772 | |
1773 | ArrayRef<uint8_t> contents = |
1774 | SectionChunk::consumeDebugMagic(data: dbgC->getContents(), sectionName: ".debug$S" ); |
1775 | DebugSubsectionArray subsections; |
1776 | BinaryStreamReader reader(contents, llvm::endianness::little); |
1777 | exitOnErr(reader.readArray(Array&: subsections, Size: contents.size())); |
1778 | |
1779 | for (const DebugSubsectionRecord &ss : subsections) { |
1780 | switch (ss.kind()) { |
1781 | case DebugSubsectionKind::StringTable: { |
1782 | assert(!cvStrTab.valid() && |
1783 | "Encountered multiple string table subsections!" ); |
1784 | exitOnErr(cvStrTab.initialize(Contents: ss.getRecordData())); |
1785 | break; |
1786 | } |
1787 | case DebugSubsectionKind::FileChecksums: |
1788 | assert(!checksums.valid() && |
1789 | "Encountered multiple checksum subsections!" ); |
1790 | exitOnErr(checksums.initialize(Stream: ss.getRecordData())); |
1791 | break; |
1792 | case DebugSubsectionKind::Lines: { |
1793 | ArrayRef<uint8_t> bytes; |
1794 | auto ref = ss.getRecordData(); |
1795 | exitOnErr(ref.readLongestContiguousChunk(Offset: 0, Buffer&: bytes)); |
1796 | size_t offsetInDbgC = bytes.data() - dbgC->getContents().data(); |
1797 | |
1798 | // Check whether this line table refers to C. |
1799 | auto i = secrels.find(Val: offsetInDbgC); |
1800 | if (i == secrels.end()) |
1801 | break; |
1802 | |
1803 | // Check whether this line table covers Addr in C. |
1804 | DebugLinesSubsectionRef linesTmp; |
1805 | exitOnErr(linesTmp.initialize(Reader: BinaryStreamReader(ref))); |
1806 | uint32_t offsetInC = i->second + linesTmp.header()->RelocOffset; |
1807 | if (addr < offsetInC || addr >= offsetInC + linesTmp.header()->CodeSize) |
1808 | break; |
1809 | |
1810 | assert(!lines.header() && |
1811 | "Encountered multiple line tables for function!" ); |
1812 | exitOnErr(lines.initialize(Reader: BinaryStreamReader(ref))); |
1813 | offsetInLinetable = addr - offsetInC; |
1814 | break; |
1815 | } |
1816 | default: |
1817 | break; |
1818 | } |
1819 | |
1820 | if (cvStrTab.valid() && checksums.valid() && lines.header()) |
1821 | return true; |
1822 | } |
1823 | } |
1824 | |
1825 | return false; |
1826 | } |
1827 | |
1828 | // Use CodeView line tables to resolve a file and line number for the given |
1829 | // offset into the given chunk and return them, or std::nullopt if a line table |
1830 | // was not found. |
1831 | std::optional<std::pair<StringRef, uint32_t>> |
1832 | lld::coff::getFileLineCodeView(const SectionChunk *c, uint32_t addr) { |
1833 | ExitOnError exitOnErr; |
1834 | |
1835 | DebugStringTableSubsectionRef cvStrTab; |
1836 | DebugChecksumsSubsectionRef checksums; |
1837 | DebugLinesSubsectionRef lines; |
1838 | uint32_t offsetInLinetable; |
1839 | |
1840 | if (!findLineTable(c, addr, cvStrTab, checksums, lines, offsetInLinetable)) |
1841 | return std::nullopt; |
1842 | |
1843 | std::optional<uint32_t> nameIndex; |
1844 | std::optional<uint32_t> lineNumber; |
1845 | for (const LineColumnEntry &entry : lines) { |
1846 | for (const LineNumberEntry &ln : entry.LineNumbers) { |
1847 | LineInfo li(ln.Flags); |
1848 | if (ln.Offset > offsetInLinetable) { |
1849 | if (!nameIndex) { |
1850 | nameIndex = entry.NameIndex; |
1851 | lineNumber = li.getStartLine(); |
1852 | } |
1853 | StringRef filename = |
1854 | exitOnErr(getFileName(strings: cvStrTab, checksums, fileID: *nameIndex)); |
1855 | return std::make_pair(x&: filename, y&: *lineNumber); |
1856 | } |
1857 | nameIndex = entry.NameIndex; |
1858 | lineNumber = li.getStartLine(); |
1859 | } |
1860 | } |
1861 | if (!nameIndex) |
1862 | return std::nullopt; |
1863 | StringRef filename = exitOnErr(getFileName(strings: cvStrTab, checksums, fileID: *nameIndex)); |
1864 | return std::make_pair(x&: filename, y&: *lineNumber); |
1865 | } |
1866 | |