blob: ca4c6fa835bf10e0cb7bee2d91509433e4088d2f [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
26SymbolTable::SymbolTable() {
Rui Ueyama8d3010a2015-06-30 19:35:21 +000027 addSymbol(new (Alloc) DefinedAbsolute("__ImageBase", Config->ImageBase));
Rui Ueyama411c63602015-05-28 19:09:30 +000028}
29
Rui Ueyama8d3010a2015-06-30 19:35:21 +000030void SymbolTable::addFile(std::unique_ptr<InputFile> FileP) {
31 InputFile *File = FileP.get();
32 Files.push_back(std::move(FileP));
33 if (auto *F = dyn_cast<ArchiveFile>(File)) {
34 ArchiveQueue.push_back(F);
35 return;
36 }
37 ObjectQueue.push_back(File);
38 if (auto *F = dyn_cast<ObjectFile>(File)) {
39 ObjectFiles.push_back(F);
40 } else if (auto *F = dyn_cast<BitcodeFile>(File)) {
41 BitcodeFiles.push_back(F);
42 } else {
43 ImportFiles.push_back(cast<ImportFile>(File));
44 }
Rui Ueyama411c63602015-05-28 19:09:30 +000045}
46
Rui Ueyama85225b02015-07-02 03:15:15 +000047std::error_code SymbolTable::step() {
48 if (queueEmpty())
49 return std::error_code();
50 if (auto EC = readObjects())
51 return EC;
52 if (auto EC = readArchives())
53 return EC;
54 return std::error_code();
55}
56
Rui Ueyama0d2e9992015-06-23 23:56:39 +000057std::error_code SymbolTable::run() {
Rui Ueyama85225b02015-07-02 03:15:15 +000058 while (!queueEmpty())
59 if (auto EC = step())
Rui Ueyama0d2e9992015-06-23 23:56:39 +000060 return EC;
Peter Collingbourneace2f092015-06-06 02:00:45 +000061 return std::error_code();
62}
63
Rui Ueyama8d3010a2015-06-30 19:35:21 +000064std::error_code SymbolTable::readArchives() {
65 if (ArchiveQueue.empty())
66 return std::error_code();
67
68 // Add lazy symbols to the symbol table. Lazy symbols that conflict
69 // with existing undefined symbols are accumulated in LazySyms.
70 std::vector<Symbol *> LazySyms;
71 for (ArchiveFile *File : ArchiveQueue) {
72 if (Config->Verbose)
73 llvm::outs() << "Reading " << File->getShortName() << "\n";
74 if (auto EC = File->parse())
75 return EC;
76 for (Lazy *Sym : File->getLazySymbols())
77 addLazy(Sym, &LazySyms);
78 }
79 ArchiveQueue.clear();
80
81 // Add archive member files to ObjectQueue that should resolve
82 // existing undefined symbols.
83 for (Symbol *Sym : LazySyms)
84 if (auto EC = addMemberFile(cast<Lazy>(Sym->Body)))
85 return EC;
86 return std::error_code();
87}
88
89std::error_code SymbolTable::readObjects() {
90 if (ObjectQueue.empty())
91 return std::error_code();
92
93 // Add defined and undefined symbols to the symbol table.
94 std::vector<StringRef> Directives;
95 for (size_t I = 0; I < ObjectQueue.size(); ++I) {
96 InputFile *File = ObjectQueue[I];
97 if (Config->Verbose)
98 llvm::outs() << "Reading " << File->getShortName() << "\n";
99 if (auto EC = File->parse())
100 return EC;
101 // Adding symbols may add more files to ObjectQueue
102 // (but not to ArchiveQueue).
103 for (SymbolBody *Sym : File->getSymbols())
104 if (Sym->isExternal())
105 if (auto EC = addSymbol(Sym))
106 return EC;
107 StringRef S = File->getDirectives();
108 if (!S.empty())
109 Directives.push_back(S);
110 }
111 ObjectQueue.clear();
112
113 // Parse directive sections. This may add files to
114 // ArchiveQueue and ObjectQueue.
115 for (StringRef S : Directives)
116 if (auto EC = Driver->parseDirectives(S))
117 return EC;
118 return std::error_code();
119}
120
Rui Ueyama3d4c69c2015-07-02 02:38:59 +0000121bool SymbolTable::queueEmpty() {
122 return ArchiveQueue.empty() && ObjectQueue.empty();
123}
124
Rui Ueyama411c63602015-05-28 19:09:30 +0000125bool SymbolTable::reportRemainingUndefines() {
126 bool Ret = false;
127 for (auto &I : Symtab) {
128 Symbol *Sym = I.second;
129 auto *Undef = dyn_cast<Undefined>(Sym->Body);
130 if (!Undef)
131 continue;
Rui Ueyamad7666532015-06-25 02:21:44 +0000132 StringRef Name = Undef->getName();
Rui Ueyama3d4c69c2015-07-02 02:38:59 +0000133 // A weak alias may have been resovled, so check for that. A weak alias
134 // may be an weak alias to other symbol, so check recursively.
135 for (Undefined *U = Undef->WeakAlias; U; U = U->WeakAlias) {
136 if (auto *D = dyn_cast<Defined>(U->repl())) {
Rui Ueyama6bf638e2015-07-02 00:04:14 +0000137 Sym->Body = D;
Rui Ueyama3d4c69c2015-07-02 02:38:59 +0000138 goto next;
Rui Ueyama6bf638e2015-07-02 00:04:14 +0000139 }
Rui Ueyama411c63602015-05-28 19:09:30 +0000140 }
Rui Ueyamad7666532015-06-25 02:21:44 +0000141 // If we can resolve a symbol by removing __imp_ prefix, do that.
142 // This odd rule is for compatibility with MSVC linker.
143 if (Name.startswith("__imp_")) {
144 if (Defined *Imp = find(Name.substr(strlen("__imp_")))) {
Rui Ueyama88e0f922015-06-25 03:31:47 +0000145 auto *S = new (Alloc) DefinedLocalImport(Name, Imp);
146 LocalImportChunks.push_back(S->getChunk());
147 Sym->Body = S;
Rui Ueyamad7666532015-06-25 02:21:44 +0000148 continue;
149 }
150 }
151 llvm::errs() << "undefined symbol: " << Name << "\n";
Rui Ueyama95925fd2015-06-28 19:35:15 +0000152 // Remaining undefined symbols are not fatal if /force is specified.
153 // They are replaced with dummy defined symbols.
154 if (Config->Force) {
155 Sym->Body = new (Alloc) DefinedAbsolute(Name, 0);
156 continue;
157 }
Rui Ueyama411c63602015-05-28 19:09:30 +0000158 Ret = true;
Rui Ueyama3d4c69c2015-07-02 02:38:59 +0000159 next:;
Rui Ueyama411c63602015-05-28 19:09:30 +0000160 }
161 return Ret;
162}
163
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000164void SymbolTable::addLazy(Lazy *New, std::vector<Symbol *> *Accum) {
165 Symbol *&Sym = Symtab[New->getName()];
Rui Ueyama411c63602015-05-28 19:09:30 +0000166 if (!Sym) {
167 Sym = new (Alloc) Symbol(New);
168 New->setBackref(Sym);
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000169 return;
170 }
171 SymbolBody *Existing = Sym->Body;
172 if (!isa<Undefined>(Existing))
173 return;
174 Sym->Body = New;
175 New->setBackref(Sym);
176 Accum->push_back(Sym);
177}
178
179std::error_code SymbolTable::addSymbol(SymbolBody *New) {
180 // Find an existing symbol or create and insert a new one.
181 assert(isa<Defined>(New) || isa<Undefined>(New));
182 Symbol *&Sym = Symtab[New->getName()];
183 if (!Sym) {
184 Sym = new (Alloc) Symbol(New);
185 New->setBackref(Sym);
Rui Ueyama411c63602015-05-28 19:09:30 +0000186 return std::error_code();
187 }
188 New->setBackref(Sym);
189
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000190 // If we have an undefined symbol and a lazy symbol,
191 // let the lazy symbol to read a member file.
192 SymbolBody *Existing = Sym->Body;
193 if (auto *L = dyn_cast<Lazy>(Existing)) {
Rui Ueyama48975962015-07-01 22:32:23 +0000194 // Undefined symbols with weak aliases need not to be resolved,
195 // since they would be replaced with weak aliases if they remain
196 // undefined.
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000197 if (auto *U = dyn_cast<Undefined>(New))
Rui Ueyama48975962015-07-01 22:32:23 +0000198 if (!U->WeakAlias)
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000199 return addMemberFile(L);
200 Sym->Body = New;
201 return std::error_code();
202 }
203
Rui Ueyama411c63602015-05-28 19:09:30 +0000204 // compare() returns -1, 0, or 1 if the lhs symbol is less preferable,
205 // equivalent (conflicting), or more preferable, respectively.
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000206 int Comp = Existing->compare(New);
207 if (Comp == 0) {
Rui Ueyama68633f12015-06-25 23:22:00 +0000208 llvm::errs() << "duplicate symbol: " << Existing->getDebugName()
209 << " and " << New->getDebugName() << "\n";
Rui Ueyama8fd9fb92015-06-01 02:58:15 +0000210 return make_error_code(LLDError::DuplicateSymbols);
211 }
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000212 if (Comp < 0)
213 Sym->Body = New;
Rui Ueyama411c63602015-05-28 19:09:30 +0000214 return std::error_code();
215}
216
217// Reads an archive member file pointed by a given symbol.
218std::error_code SymbolTable::addMemberFile(Lazy *Body) {
219 auto FileOrErr = Body->getMember();
220 if (auto EC = FileOrErr.getError())
221 return EC;
222 std::unique_ptr<InputFile> File = std::move(FileOrErr.get());
223
224 // getMember returns an empty buffer if the member was already
225 // read from the library.
226 if (!File)
227 return std::error_code();
228 if (Config->Verbose)
Rui Ueyama5b2588a2015-06-08 05:43:50 +0000229 llvm::outs() << "Loaded " << File->getShortName() << " for "
Rui Ueyama411c63602015-05-28 19:09:30 +0000230 << Body->getName() << "\n";
Rui Ueyama0d2e9992015-06-23 23:56:39 +0000231 addFile(std::move(File));
232 return std::error_code();
Rui Ueyama411c63602015-05-28 19:09:30 +0000233}
234
235std::vector<Chunk *> SymbolTable::getChunks() {
236 std::vector<Chunk *> Res;
Rui Ueyama0d2e9992015-06-23 23:56:39 +0000237 for (ObjectFile *File : ObjectFiles) {
Rui Ueyama411c63602015-05-28 19:09:30 +0000238 std::vector<Chunk *> &V = File->getChunks();
239 Res.insert(Res.end(), V.begin(), V.end());
240 }
241 return Res;
242}
243
Rui Ueyama5cff6852015-05-31 03:34:08 +0000244Defined *SymbolTable::find(StringRef Name) {
Rui Ueyama411c63602015-05-28 19:09:30 +0000245 auto It = Symtab.find(Name);
246 if (It == Symtab.end())
247 return nullptr;
Rui Ueyama5cff6852015-05-31 03:34:08 +0000248 if (auto *Def = dyn_cast<Defined>(It->second->Body))
249 return Def;
250 return nullptr;
251}
252
Rui Ueyama4b669892015-06-30 23:46:52 +0000253Symbol *SymbolTable::findSymbol(StringRef Name) {
Rui Ueyama45044f42015-06-29 01:03:53 +0000254 auto It = Symtab.find(Name);
255 if (It == Symtab.end())
256 return nullptr;
Rui Ueyama4b669892015-06-30 23:46:52 +0000257 return It->second;
Rui Ueyama45044f42015-06-29 01:03:53 +0000258}
259
Rui Ueyama6bf638e2015-07-02 00:04:14 +0000260void SymbolTable::mangleMaybe(Undefined *U) {
261 if (U->WeakAlias)
262 return;
Rui Ueyama0744e872015-07-02 00:21:11 +0000263 if (!isa<Undefined>(U->repl()))
Rui Ueyama6bf638e2015-07-02 00:04:14 +0000264 return;
Rui Ueyamaf5313b32015-06-28 22:16:41 +0000265
266 // In Microsoft ABI, a non-member function name is mangled this way.
Rui Ueyama6bf638e2015-07-02 00:04:14 +0000267 std::string Prefix = ("?" + U->getName() + "@@Y").str();
Rui Ueyama3d4c69c2015-07-02 02:38:59 +0000268 for (auto Pair : Symtab) {
269 StringRef Name = Pair.first;
Rui Ueyamaf5313b32015-06-28 22:16:41 +0000270 if (!Name.startswith(Prefix))
271 continue;
Rui Ueyama3d4c69c2015-07-02 02:38:59 +0000272 U->WeakAlias = addUndefined(Name);
Rui Ueyama6bf638e2015-07-02 00:04:14 +0000273 return;
Rui Ueyamaf5313b32015-06-28 22:16:41 +0000274 }
Rui Ueyamaf5313b32015-06-28 22:16:41 +0000275}
276
Rui Ueyama6bf638e2015-07-02 00:04:14 +0000277Undefined *SymbolTable::addUndefined(StringRef Name) {
Rui Ueyama3d4c69c2015-07-02 02:38:59 +0000278 auto *New = new (Alloc) Undefined(Name);
279 addSymbol(New);
280 if (auto *U = dyn_cast<Undefined>(New->repl()))
281 return U;
282 return New;
Rui Ueyama360bace2015-05-31 22:31:31 +0000283}
284
Peter Collingbournebe549552015-06-26 18:58:24 +0000285void SymbolTable::printMap(llvm::raw_ostream &OS) {
286 for (ObjectFile *File : ObjectFiles) {
287 OS << File->getShortName() << ":\n";
288 for (SymbolBody *Body : File->getSymbols())
289 if (auto *R = dyn_cast<DefinedRegular>(Body))
290 if (R->isLive())
291 OS << Twine::utohexstr(Config->ImageBase + R->getRVA())
292 << " " << R->getName() << "\n";
293 }
294}
295
Peter Collingbourne60c16162015-06-01 20:10:10 +0000296std::error_code SymbolTable::addCombinedLTOObject() {
297 if (BitcodeFiles.empty())
298 return std::error_code();
299
Peter Collingbourne60c16162015-06-01 20:10:10 +0000300 // Create an object file and add it to the symbol table by replacing any
301 // DefinedBitcode symbols with the definitions in the object file.
Rui Ueyamaefba7812015-06-09 17:52:17 +0000302 LTOCodeGenerator CG;
303 auto FileOrErr = createLTOObject(&CG);
304 if (auto EC = FileOrErr.getError())
Peter Collingbourne60c16162015-06-01 20:10:10 +0000305 return EC;
Rui Ueyamaefba7812015-06-09 17:52:17 +0000306 ObjectFile *Obj = FileOrErr.get();
307
Peter Collingbourne60c16162015-06-01 20:10:10 +0000308 for (SymbolBody *Body : Obj->getSymbols()) {
309 if (!Body->isExternal())
310 continue;
Peter Collingbourned9e4e982015-06-09 02:53:09 +0000311 // Find an existing Symbol. We should not see any new undefined symbols at
312 // this point.
Peter Collingbourne60c16162015-06-01 20:10:10 +0000313 StringRef Name = Body->getName();
Peter Collingbourned9e4e982015-06-09 02:53:09 +0000314 Symbol *&Sym = Symtab[Name];
Peter Collingbourne60c16162015-06-01 20:10:10 +0000315 if (!Sym) {
Peter Collingbourned9e4e982015-06-09 02:53:09 +0000316 if (!isa<Defined>(Body)) {
317 llvm::errs() << "LTO: undefined symbol: " << Name << '\n';
318 return make_error_code(LLDError::BrokenFile);
319 }
320 Sym = new (Alloc) Symbol(Body);
321 Body->setBackref(Sym);
322 continue;
Peter Collingbourne60c16162015-06-01 20:10:10 +0000323 }
324 Body->setBackref(Sym);
325
326 if (isa<DefinedBitcode>(Sym->Body)) {
327 // The symbol should now be defined.
328 if (!isa<Defined>(Body)) {
329 llvm::errs() << "LTO: undefined symbol: " << Name << '\n';
330 return make_error_code(LLDError::BrokenFile);
331 }
332 Sym->Body = Body;
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000333 continue;
Peter Collingbourne60c16162015-06-01 20:10:10 +0000334 }
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000335 if (auto *L = dyn_cast<Lazy>(Sym->Body)) {
336 // We may see new references to runtime library symbols such as __chkstk
337 // here. These symbols must be wholly defined in non-bitcode files.
338 if (auto EC = addMemberFile(L))
Peter Collingbourne2ed4c8f2015-06-24 00:12:34 +0000339 return EC;
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000340 continue;
341 }
342 SymbolBody *Existing = Sym->Body;
343 int Comp = Existing->compare(Body);
344 if (Comp == 0) {
345 llvm::errs() << "LTO: unexpected duplicate symbol: " << Name << "\n";
346 return make_error_code(LLDError::BrokenFile);
347 }
348 if (Comp < 0)
349 Sym->Body = Body;
Rui Ueyama0d2e9992015-06-23 23:56:39 +0000350 }
351
352 size_t NumBitcodeFiles = BitcodeFiles.size();
353 if (auto EC = run())
354 return EC;
355 if (BitcodeFiles.size() != NumBitcodeFiles) {
356 llvm::errs() << "LTO: late loaded symbol created new bitcode reference\n";
357 return make_error_code(LLDError::BrokenFile);
Peter Collingbourne60c16162015-06-01 20:10:10 +0000358 }
359
Peter Collingbourne73b75e32015-06-09 04:29:54 +0000360 // New runtime library symbol references may have created undefined references.
361 if (reportRemainingUndefines())
362 return make_error_code(LLDError::BrokenFile);
Peter Collingbourne60c16162015-06-01 20:10:10 +0000363 return std::error_code();
364}
365
Rui Ueyamaefba7812015-06-09 17:52:17 +0000366// Combine and compile bitcode files and then return the result
367// as a regular COFF object file.
368ErrorOr<ObjectFile *> SymbolTable::createLTOObject(LTOCodeGenerator *CG) {
369 // All symbols referenced by non-bitcode objects must be preserved.
Rui Ueyama0d2e9992015-06-23 23:56:39 +0000370 for (ObjectFile *File : ObjectFiles)
Rui Ueyamaefba7812015-06-09 17:52:17 +0000371 for (SymbolBody *Body : File->getSymbols())
Rui Ueyama0744e872015-07-02 00:21:11 +0000372 if (auto *S = dyn_cast<DefinedBitcode>(Body->repl()))
Rui Ueyamaefba7812015-06-09 17:52:17 +0000373 CG->addMustPreserveSymbol(S->getName());
374
Peter Collingbourne1b6fd1f2015-06-11 21:49:54 +0000375 // Likewise for bitcode symbols which we initially resolved to non-bitcode.
Rui Ueyama0d2e9992015-06-23 23:56:39 +0000376 for (BitcodeFile *File : BitcodeFiles)
Peter Collingbourne1b6fd1f2015-06-11 21:49:54 +0000377 for (SymbolBody *Body : File->getSymbols())
Rui Ueyama0744e872015-07-02 00:21:11 +0000378 if (isa<DefinedBitcode>(Body) && !isa<DefinedBitcode>(Body->repl()))
Peter Collingbourne1b6fd1f2015-06-11 21:49:54 +0000379 CG->addMustPreserveSymbol(Body->getName());
380
Rui Ueyamaefba7812015-06-09 17:52:17 +0000381 // Likewise for other symbols that must be preserved.
Rui Ueyama18f8d2c2015-07-02 00:21:08 +0000382 for (Undefined *U : Config->GCRoot)
Rui Ueyama0744e872015-07-02 00:21:11 +0000383 if (isa<DefinedBitcode>(U->repl()))
Rui Ueyama18f8d2c2015-07-02 00:21:08 +0000384 CG->addMustPreserveSymbol(U->getName());
Rui Ueyamaefba7812015-06-09 17:52:17 +0000385
386 CG->setModule(BitcodeFiles[0]->releaseModule());
387 for (unsigned I = 1, E = BitcodeFiles.size(); I != E; ++I)
388 CG->addModule(BitcodeFiles[I]->getModule());
389
390 std::string ErrMsg;
391 LTOMB = CG->compile(false, false, false, ErrMsg); // take MB ownership
392 if (!LTOMB) {
393 llvm::errs() << ErrMsg << '\n';
394 return make_error_code(LLDError::BrokenFile);
395 }
Rui Ueyama0d2e9992015-06-23 23:56:39 +0000396 auto *Obj = new ObjectFile(LTOMB->getMemBufferRef());
397 Files.emplace_back(Obj);
398 ObjectFiles.push_back(Obj);
Rui Ueyamaefba7812015-06-09 17:52:17 +0000399 if (auto EC = Obj->parse())
400 return EC;
401 return Obj;
402}
403
Rui Ueyama411c63602015-05-28 19:09:30 +0000404} // namespace coff
405} // namespace lld