blob: 4a08c49757c65f7d9a004ff1952587fbc2c2c894 [file] [log] [blame]
Rui Ueyama411c63602015-05-28 19:09:30 +00001//===- SymbolTable.cpp ----------------------------------------------------===//
2//
3// The LLVM Linker
4//
5// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9
10#include "Config.h"
11#include "Driver.h"
Rui Ueyama8fd9fb92015-06-01 02:58:15 +000012#include "Error.h"
Rui Ueyama411c63602015-05-28 19:09:30 +000013#include "SymbolTable.h"
Chandler Carruthbe6e80b2015-06-29 18:50:11 +000014#include "Symbols.h"
Rui Ueyama411c63602015-05-28 19:09:30 +000015#include "llvm/ADT/STLExtras.h"
Peter Collingbourne60c16162015-06-01 20:10:10 +000016#include "llvm/LTO/LTOCodeGenerator.h"
Rui Ueyama411c63602015-05-28 19:09:30 +000017#include "llvm/Support/Debug.h"
18#include "llvm/Support/raw_ostream.h"
Rui Ueyamaf5313b32015-06-28 22:16:41 +000019#include <utility>
Rui Ueyama411c63602015-05-28 19:09:30 +000020
Rui Ueyamad68ff342015-05-31 03:57:30 +000021using namespace llvm;
22
Rui Ueyama411c63602015-05-28 19:09:30 +000023namespace lld {
24namespace coff {
25
Rui Ueyama8d3010a2015-06-30 19:35:21 +000026void SymbolTable::addFile(std::unique_ptr<InputFile> FileP) {
27 InputFile *File = FileP.get();
28 Files.push_back(std::move(FileP));
29 if (auto *F = dyn_cast<ArchiveFile>(File)) {
30 ArchiveQueue.push_back(F);
31 return;
32 }
33 ObjectQueue.push_back(File);
34 if (auto *F = dyn_cast<ObjectFile>(File)) {
35 ObjectFiles.push_back(F);
36 } else if (auto *F = dyn_cast<BitcodeFile>(File)) {
37 BitcodeFiles.push_back(F);
38 } else {
39 ImportFiles.push_back(cast<ImportFile>(File));
40 }
Rui Ueyama411c63602015-05-28 19:09:30 +000041}
42
Rui Ueyama85225b02015-07-02 03:15:15 +000043std::error_code SymbolTable::step() {
44 if (queueEmpty())
45 return std::error_code();
46 if (auto EC = readObjects())
47 return EC;
48 if (auto EC = readArchives())
49 return EC;
50 return std::error_code();
51}
52
Rui Ueyama0d2e9992015-06-23 23:56:39 +000053std::error_code SymbolTable::run() {
Rui Ueyama85225b02015-07-02 03:15:15 +000054 while (!queueEmpty())
55 if (auto EC = step())
Rui Ueyama0d2e9992015-06-23 23:56:39 +000056 return EC;
Peter Collingbourneace2f092015-06-06 02:00:45 +000057 return std::error_code();
58}
59
Rui Ueyama8d3010a2015-06-30 19:35:21 +000060std::error_code SymbolTable::readArchives() {
61 if (ArchiveQueue.empty())
62 return std::error_code();
63
64 // Add lazy symbols to the symbol table. Lazy symbols that conflict
65 // with existing undefined symbols are accumulated in LazySyms.
66 std::vector<Symbol *> LazySyms;
67 for (ArchiveFile *File : ArchiveQueue) {
68 if (Config->Verbose)
69 llvm::outs() << "Reading " << File->getShortName() << "\n";
70 if (auto EC = File->parse())
71 return EC;
72 for (Lazy *Sym : File->getLazySymbols())
73 addLazy(Sym, &LazySyms);
74 }
75 ArchiveQueue.clear();
76
77 // Add archive member files to ObjectQueue that should resolve
78 // existing undefined symbols.
79 for (Symbol *Sym : LazySyms)
80 if (auto EC = addMemberFile(cast<Lazy>(Sym->Body)))
81 return EC;
82 return std::error_code();
83}
84
85std::error_code SymbolTable::readObjects() {
86 if (ObjectQueue.empty())
87 return std::error_code();
88
89 // Add defined and undefined symbols to the symbol table.
90 std::vector<StringRef> Directives;
91 for (size_t I = 0; I < ObjectQueue.size(); ++I) {
92 InputFile *File = ObjectQueue[I];
93 if (Config->Verbose)
94 llvm::outs() << "Reading " << File->getShortName() << "\n";
95 if (auto EC = File->parse())
96 return EC;
97 // Adding symbols may add more files to ObjectQueue
98 // (but not to ArchiveQueue).
99 for (SymbolBody *Sym : File->getSymbols())
100 if (Sym->isExternal())
101 if (auto EC = addSymbol(Sym))
102 return EC;
103 StringRef S = File->getDirectives();
104 if (!S.empty())
105 Directives.push_back(S);
106 }
107 ObjectQueue.clear();
108
109 // Parse directive sections. This may add files to
110 // ArchiveQueue and ObjectQueue.
111 for (StringRef S : Directives)
112 if (auto EC = Driver->parseDirectives(S))
113 return EC;
114 return std::error_code();
115}
116
Rui Ueyama3d4c69c2015-07-02 02:38:59 +0000117bool SymbolTable::queueEmpty() {
118 return ArchiveQueue.empty() && ObjectQueue.empty();
119}
120
Rui Ueyama411c63602015-05-28 19:09:30 +0000121bool SymbolTable::reportRemainingUndefines() {
122 bool Ret = false;
123 for (auto &I : Symtab) {
124 Symbol *Sym = I.second;
125 auto *Undef = dyn_cast<Undefined>(Sym->Body);
126 if (!Undef)
127 continue;
Rui Ueyamad7666532015-06-25 02:21:44 +0000128 StringRef Name = Undef->getName();
Rui Ueyama3d4c69c2015-07-02 02:38:59 +0000129 // A weak alias may have been resovled, so check for that. A weak alias
130 // may be an weak alias to other symbol, so check recursively.
131 for (Undefined *U = Undef->WeakAlias; U; U = U->WeakAlias) {
132 if (auto *D = dyn_cast<Defined>(U->repl())) {
Rui Ueyama6bf638e2015-07-02 00:04:14 +0000133 Sym->Body = D;
Rui Ueyama3d4c69c2015-07-02 02:38:59 +0000134 goto next;
Rui Ueyama6bf638e2015-07-02 00:04:14 +0000135 }
Rui Ueyama411c63602015-05-28 19:09:30 +0000136 }
Rui Ueyamad7666532015-06-25 02:21:44 +0000137 // If we can resolve a symbol by removing __imp_ prefix, do that.
138 // This odd rule is for compatibility with MSVC linker.
139 if (Name.startswith("__imp_")) {
Rui Ueyama458d7442015-07-02 03:59:04 +0000140 Symbol *Imp = find(Name.substr(strlen("__imp_")));
141 if (Imp && isa<Defined>(Imp->Body)) {
142 auto *D = cast<Defined>(Imp->Body);
143 auto *S = new (Alloc) DefinedLocalImport(Name, D);
Rui Ueyama88e0f922015-06-25 03:31:47 +0000144 LocalImportChunks.push_back(S->getChunk());
145 Sym->Body = S;
Rui Ueyamad7666532015-06-25 02:21:44 +0000146 continue;
147 }
148 }
149 llvm::errs() << "undefined symbol: " << Name << "\n";
Rui Ueyama95925fd2015-06-28 19:35:15 +0000150 // Remaining undefined symbols are not fatal if /force is specified.
151 // They are replaced with dummy defined symbols.
152 if (Config->Force) {
153 Sym->Body = new (Alloc) DefinedAbsolute(Name, 0);
154 continue;
155 }
Rui Ueyama411c63602015-05-28 19:09:30 +0000156 Ret = true;
Rui Ueyama3d4c69c2015-07-02 02:38:59 +0000157 next:;
Rui Ueyama411c63602015-05-28 19:09:30 +0000158 }
159 return Ret;
160}
161
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000162void SymbolTable::addLazy(Lazy *New, std::vector<Symbol *> *Accum) {
Rui Ueyama6be90992015-07-02 22:52:33 +0000163 Symbol *Sym = insert(New);
164 if (Sym->Body == New)
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000165 return;
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000166 SymbolBody *Existing = Sym->Body;
167 if (!isa<Undefined>(Existing))
168 return;
169 Sym->Body = New;
170 New->setBackref(Sym);
171 Accum->push_back(Sym);
172}
173
174std::error_code SymbolTable::addSymbol(SymbolBody *New) {
175 // Find an existing symbol or create and insert a new one.
176 assert(isa<Defined>(New) || isa<Undefined>(New));
Rui Ueyama6be90992015-07-02 22:52:33 +0000177 Symbol *Sym = insert(New);
178 if (Sym->Body == New)
Rui Ueyama411c63602015-05-28 19:09:30 +0000179 return std::error_code();
Rui Ueyama6be90992015-07-02 22:52:33 +0000180 SymbolBody *Existing = Sym->Body;
Rui Ueyama411c63602015-05-28 19:09:30 +0000181
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000182 // If we have an undefined symbol and a lazy symbol,
183 // let the lazy symbol to read a member file.
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000184 if (auto *L = dyn_cast<Lazy>(Existing)) {
Rui Ueyama48975962015-07-01 22:32:23 +0000185 // Undefined symbols with weak aliases need not to be resolved,
186 // since they would be replaced with weak aliases if they remain
187 // undefined.
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000188 if (auto *U = dyn_cast<Undefined>(New))
Rui Ueyama48975962015-07-01 22:32:23 +0000189 if (!U->WeakAlias)
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000190 return addMemberFile(L);
191 Sym->Body = New;
192 return std::error_code();
193 }
194
Rui Ueyama411c63602015-05-28 19:09:30 +0000195 // compare() returns -1, 0, or 1 if the lhs symbol is less preferable,
196 // equivalent (conflicting), or more preferable, respectively.
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000197 int Comp = Existing->compare(New);
198 if (Comp == 0) {
Rui Ueyama68633f12015-06-25 23:22:00 +0000199 llvm::errs() << "duplicate symbol: " << Existing->getDebugName()
200 << " and " << New->getDebugName() << "\n";
Rui Ueyama8fd9fb92015-06-01 02:58:15 +0000201 return make_error_code(LLDError::DuplicateSymbols);
202 }
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000203 if (Comp < 0)
204 Sym->Body = New;
Rui Ueyama411c63602015-05-28 19:09:30 +0000205 return std::error_code();
206}
207
Rui Ueyama6be90992015-07-02 22:52:33 +0000208Symbol *SymbolTable::insert(SymbolBody *New) {
209 Symbol *&Sym = Symtab[New->getName()];
210 if (Sym) {
211 New->setBackref(Sym);
212 return Sym;
213 }
214 Sym = new (Alloc) Symbol(New);
215 New->setBackref(Sym);
216 return Sym;
217}
218
Rui Ueyama411c63602015-05-28 19:09:30 +0000219// Reads an archive member file pointed by a given symbol.
220std::error_code SymbolTable::addMemberFile(Lazy *Body) {
221 auto FileOrErr = Body->getMember();
222 if (auto EC = FileOrErr.getError())
223 return EC;
224 std::unique_ptr<InputFile> File = std::move(FileOrErr.get());
225
226 // getMember returns an empty buffer if the member was already
227 // read from the library.
228 if (!File)
229 return std::error_code();
230 if (Config->Verbose)
Rui Ueyama5b2588a2015-06-08 05:43:50 +0000231 llvm::outs() << "Loaded " << File->getShortName() << " for "
Rui Ueyama411c63602015-05-28 19:09:30 +0000232 << Body->getName() << "\n";
Rui Ueyama0d2e9992015-06-23 23:56:39 +0000233 addFile(std::move(File));
234 return std::error_code();
Rui Ueyama411c63602015-05-28 19:09:30 +0000235}
236
237std::vector<Chunk *> SymbolTable::getChunks() {
238 std::vector<Chunk *> Res;
Rui Ueyama0d2e9992015-06-23 23:56:39 +0000239 for (ObjectFile *File : ObjectFiles) {
Rui Ueyama411c63602015-05-28 19:09:30 +0000240 std::vector<Chunk *> &V = File->getChunks();
241 Res.insert(Res.end(), V.begin(), V.end());
242 }
243 return Res;
244}
245
Rui Ueyama458d7442015-07-02 03:59:04 +0000246Symbol *SymbolTable::find(StringRef Name) {
Rui Ueyama45044f42015-06-29 01:03:53 +0000247 auto It = Symtab.find(Name);
248 if (It == Symtab.end())
249 return nullptr;
Rui Ueyama4b669892015-06-30 23:46:52 +0000250 return It->second;
Rui Ueyama45044f42015-06-29 01:03:53 +0000251}
252
Rui Ueyama6bf638e2015-07-02 00:04:14 +0000253void SymbolTable::mangleMaybe(Undefined *U) {
254 if (U->WeakAlias)
255 return;
Rui Ueyama0744e872015-07-02 00:21:11 +0000256 if (!isa<Undefined>(U->repl()))
Rui Ueyama6bf638e2015-07-02 00:04:14 +0000257 return;
Rui Ueyamaf5313b32015-06-28 22:16:41 +0000258
259 // In Microsoft ABI, a non-member function name is mangled this way.
Rui Ueyama6bf638e2015-07-02 00:04:14 +0000260 std::string Prefix = ("?" + U->getName() + "@@Y").str();
Rui Ueyama3d4c69c2015-07-02 02:38:59 +0000261 for (auto Pair : Symtab) {
262 StringRef Name = Pair.first;
Rui Ueyamaf5313b32015-06-28 22:16:41 +0000263 if (!Name.startswith(Prefix))
264 continue;
Rui Ueyama3d4c69c2015-07-02 02:38:59 +0000265 U->WeakAlias = addUndefined(Name);
Rui Ueyama6bf638e2015-07-02 00:04:14 +0000266 return;
Rui Ueyamaf5313b32015-06-28 22:16:41 +0000267 }
Rui Ueyamaf5313b32015-06-28 22:16:41 +0000268}
269
Rui Ueyama6bf638e2015-07-02 00:04:14 +0000270Undefined *SymbolTable::addUndefined(StringRef Name) {
Rui Ueyama3d4c69c2015-07-02 02:38:59 +0000271 auto *New = new (Alloc) Undefined(Name);
272 addSymbol(New);
273 if (auto *U = dyn_cast<Undefined>(New->repl()))
274 return U;
275 return New;
Rui Ueyama360bace2015-05-31 22:31:31 +0000276}
277
Rui Ueyama49d6cd32015-07-03 00:02:19 +0000278void SymbolTable::addAbsolute(StringRef Name, uint64_t VA) {
279 addSymbol(new (Alloc) DefinedAbsolute(Name, VA));
280}
281
Peter Collingbournebe549552015-06-26 18:58:24 +0000282void SymbolTable::printMap(llvm::raw_ostream &OS) {
283 for (ObjectFile *File : ObjectFiles) {
284 OS << File->getShortName() << ":\n";
285 for (SymbolBody *Body : File->getSymbols())
286 if (auto *R = dyn_cast<DefinedRegular>(Body))
287 if (R->isLive())
288 OS << Twine::utohexstr(Config->ImageBase + R->getRVA())
289 << " " << R->getName() << "\n";
290 }
291}
292
Peter Collingbourne60c16162015-06-01 20:10:10 +0000293std::error_code SymbolTable::addCombinedLTOObject() {
294 if (BitcodeFiles.empty())
295 return std::error_code();
296
Peter Collingbourne60c16162015-06-01 20:10:10 +0000297 // Create an object file and add it to the symbol table by replacing any
298 // DefinedBitcode symbols with the definitions in the object file.
Rui Ueyamaefba7812015-06-09 17:52:17 +0000299 LTOCodeGenerator CG;
300 auto FileOrErr = createLTOObject(&CG);
301 if (auto EC = FileOrErr.getError())
Peter Collingbourne60c16162015-06-01 20:10:10 +0000302 return EC;
Rui Ueyamaefba7812015-06-09 17:52:17 +0000303 ObjectFile *Obj = FileOrErr.get();
304
Peter Collingbourne60c16162015-06-01 20:10:10 +0000305 for (SymbolBody *Body : Obj->getSymbols()) {
306 if (!Body->isExternal())
307 continue;
Peter Collingbourned9e4e982015-06-09 02:53:09 +0000308 // Find an existing Symbol. We should not see any new undefined symbols at
309 // this point.
Peter Collingbourne60c16162015-06-01 20:10:10 +0000310 StringRef Name = Body->getName();
Rui Ueyama6be90992015-07-02 22:52:33 +0000311 Symbol *Sym = insert(Body);
312 if (Sym->Body == Body && !isa<Defined>(Body)) {
313 llvm::errs() << "LTO: undefined symbol: " << Name << '\n';
314 return make_error_code(LLDError::BrokenFile);
Peter Collingbourne60c16162015-06-01 20:10:10 +0000315 }
Peter Collingbourne60c16162015-06-01 20:10:10 +0000316
317 if (isa<DefinedBitcode>(Sym->Body)) {
318 // The symbol should now be defined.
319 if (!isa<Defined>(Body)) {
320 llvm::errs() << "LTO: undefined symbol: " << Name << '\n';
321 return make_error_code(LLDError::BrokenFile);
322 }
323 Sym->Body = Body;
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000324 continue;
Peter Collingbourne60c16162015-06-01 20:10:10 +0000325 }
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000326 if (auto *L = dyn_cast<Lazy>(Sym->Body)) {
327 // We may see new references to runtime library symbols such as __chkstk
328 // here. These symbols must be wholly defined in non-bitcode files.
329 if (auto EC = addMemberFile(L))
Peter Collingbourne2ed4c8f2015-06-24 00:12:34 +0000330 return EC;
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000331 continue;
332 }
333 SymbolBody *Existing = Sym->Body;
334 int Comp = Existing->compare(Body);
335 if (Comp == 0) {
336 llvm::errs() << "LTO: unexpected duplicate symbol: " << Name << "\n";
337 return make_error_code(LLDError::BrokenFile);
338 }
339 if (Comp < 0)
340 Sym->Body = Body;
Rui Ueyama0d2e9992015-06-23 23:56:39 +0000341 }
342
343 size_t NumBitcodeFiles = BitcodeFiles.size();
344 if (auto EC = run())
345 return EC;
346 if (BitcodeFiles.size() != NumBitcodeFiles) {
347 llvm::errs() << "LTO: late loaded symbol created new bitcode reference\n";
348 return make_error_code(LLDError::BrokenFile);
Peter Collingbourne60c16162015-06-01 20:10:10 +0000349 }
350
Peter Collingbourne73b75e32015-06-09 04:29:54 +0000351 // New runtime library symbol references may have created undefined references.
352 if (reportRemainingUndefines())
353 return make_error_code(LLDError::BrokenFile);
Peter Collingbourne60c16162015-06-01 20:10:10 +0000354 return std::error_code();
355}
356
Rui Ueyamaefba7812015-06-09 17:52:17 +0000357// Combine and compile bitcode files and then return the result
358// as a regular COFF object file.
359ErrorOr<ObjectFile *> SymbolTable::createLTOObject(LTOCodeGenerator *CG) {
360 // All symbols referenced by non-bitcode objects must be preserved.
Rui Ueyama0d2e9992015-06-23 23:56:39 +0000361 for (ObjectFile *File : ObjectFiles)
Rui Ueyamaefba7812015-06-09 17:52:17 +0000362 for (SymbolBody *Body : File->getSymbols())
Rui Ueyama0744e872015-07-02 00:21:11 +0000363 if (auto *S = dyn_cast<DefinedBitcode>(Body->repl()))
Rui Ueyamaefba7812015-06-09 17:52:17 +0000364 CG->addMustPreserveSymbol(S->getName());
365
Peter Collingbourne1b6fd1f2015-06-11 21:49:54 +0000366 // Likewise for bitcode symbols which we initially resolved to non-bitcode.
Rui Ueyama0d2e9992015-06-23 23:56:39 +0000367 for (BitcodeFile *File : BitcodeFiles)
Peter Collingbourne1b6fd1f2015-06-11 21:49:54 +0000368 for (SymbolBody *Body : File->getSymbols())
Rui Ueyama0744e872015-07-02 00:21:11 +0000369 if (isa<DefinedBitcode>(Body) && !isa<DefinedBitcode>(Body->repl()))
Peter Collingbourne1b6fd1f2015-06-11 21:49:54 +0000370 CG->addMustPreserveSymbol(Body->getName());
371
Rui Ueyamaefba7812015-06-09 17:52:17 +0000372 // Likewise for other symbols that must be preserved.
Rui Ueyama18f8d2c2015-07-02 00:21:08 +0000373 for (Undefined *U : Config->GCRoot)
Rui Ueyama0744e872015-07-02 00:21:11 +0000374 if (isa<DefinedBitcode>(U->repl()))
Rui Ueyama18f8d2c2015-07-02 00:21:08 +0000375 CG->addMustPreserveSymbol(U->getName());
Rui Ueyamaefba7812015-06-09 17:52:17 +0000376
377 CG->setModule(BitcodeFiles[0]->releaseModule());
378 for (unsigned I = 1, E = BitcodeFiles.size(); I != E; ++I)
379 CG->addModule(BitcodeFiles[I]->getModule());
380
381 std::string ErrMsg;
382 LTOMB = CG->compile(false, false, false, ErrMsg); // take MB ownership
383 if (!LTOMB) {
384 llvm::errs() << ErrMsg << '\n';
385 return make_error_code(LLDError::BrokenFile);
386 }
Rui Ueyama0d2e9992015-06-23 23:56:39 +0000387 auto *Obj = new ObjectFile(LTOMB->getMemBufferRef());
388 Files.emplace_back(Obj);
389 ObjectFiles.push_back(Obj);
Rui Ueyamaefba7812015-06-09 17:52:17 +0000390 if (auto EC = Obj->parse())
391 return EC;
392 return Obj;
393}
394
Rui Ueyama411c63602015-05-28 19:09:30 +0000395} // namespace coff
396} // namespace lld