blob: 44d943bf78e6b83b5aa0cc14b732b37b103d8a53 [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();
Rui Ueyamaa3d463d2015-07-04 01:39:11 +0000104 if (!S.empty()) {
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000105 Directives.push_back(S);
Rui Ueyamaa3d463d2015-07-04 01:39:11 +0000106 if (Config->Verbose)
107 llvm::outs() << "Directives: " << File->getShortName()
108 << ": " << S << "\n";
109 }
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000110 }
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
Peter Collingbourne2612a322015-07-04 05:28:41 +0000125bool SymbolTable::reportRemainingUndefines(bool Resolve) {
Rui Ueyama411c63602015-05-28 19:09:30 +0000126 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();
Peter Collingbourne2612a322015-07-04 05:28:41 +0000133 // A weak alias may have been resolved, so check for that.
134 if (Defined *D = Undef->getWeakAlias()) {
135 if (Resolve)
Rui Ueyama6bf638e2015-07-02 00:04:14 +0000136 Sym->Body = D;
Peter Collingbourne2612a322015-07-04 05:28:41 +0000137 continue;
Rui Ueyama411c63602015-05-28 19:09:30 +0000138 }
Rui Ueyamad7666532015-06-25 02:21:44 +0000139 // If we can resolve a symbol by removing __imp_ prefix, do that.
140 // This odd rule is for compatibility with MSVC linker.
141 if (Name.startswith("__imp_")) {
Rui Ueyama458d7442015-07-02 03:59:04 +0000142 Symbol *Imp = find(Name.substr(strlen("__imp_")));
143 if (Imp && isa<Defined>(Imp->Body)) {
Peter Collingbourne2612a322015-07-04 05:28:41 +0000144 if (!Resolve)
145 continue;
Rui Ueyama458d7442015-07-02 03:59:04 +0000146 auto *D = cast<Defined>(Imp->Body);
147 auto *S = new (Alloc) DefinedLocalImport(Name, D);
Rui Ueyama88e0f922015-06-25 03:31:47 +0000148 LocalImportChunks.push_back(S->getChunk());
149 Sym->Body = S;
Rui Ueyamad7666532015-06-25 02:21:44 +0000150 continue;
151 }
152 }
153 llvm::errs() << "undefined symbol: " << Name << "\n";
Rui Ueyama95925fd2015-06-28 19:35:15 +0000154 // Remaining undefined symbols are not fatal if /force is specified.
155 // They are replaced with dummy defined symbols.
156 if (Config->Force) {
Peter Collingbourne2612a322015-07-04 05:28:41 +0000157 if (Resolve)
158 Sym->Body = new (Alloc) DefinedAbsolute(Name, 0);
Rui Ueyama95925fd2015-06-28 19:35:15 +0000159 continue;
160 }
Rui Ueyama411c63602015-05-28 19:09:30 +0000161 Ret = true;
162 }
163 return Ret;
164}
165
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000166void SymbolTable::addLazy(Lazy *New, std::vector<Symbol *> *Accum) {
Rui Ueyama6be90992015-07-02 22:52:33 +0000167 Symbol *Sym = insert(New);
168 if (Sym->Body == New)
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000169 return;
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000170 SymbolBody *Existing = Sym->Body;
171 if (!isa<Undefined>(Existing))
172 return;
173 Sym->Body = New;
174 New->setBackref(Sym);
175 Accum->push_back(Sym);
176}
177
178std::error_code SymbolTable::addSymbol(SymbolBody *New) {
179 // Find an existing symbol or create and insert a new one.
180 assert(isa<Defined>(New) || isa<Undefined>(New));
Rui Ueyama6be90992015-07-02 22:52:33 +0000181 Symbol *Sym = insert(New);
182 if (Sym->Body == New)
Rui Ueyama411c63602015-05-28 19:09:30 +0000183 return std::error_code();
Rui Ueyama6be90992015-07-02 22:52:33 +0000184 SymbolBody *Existing = Sym->Body;
Rui Ueyama411c63602015-05-28 19:09:30 +0000185
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000186 // If we have an undefined symbol and a lazy symbol,
187 // let the lazy symbol to read a member file.
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000188 if (auto *L = dyn_cast<Lazy>(Existing)) {
Rui Ueyama48975962015-07-01 22:32:23 +0000189 // Undefined symbols with weak aliases need not to be resolved,
190 // since they would be replaced with weak aliases if they remain
191 // undefined.
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000192 if (auto *U = dyn_cast<Undefined>(New))
Rui Ueyama48975962015-07-01 22:32:23 +0000193 if (!U->WeakAlias)
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000194 return addMemberFile(L);
195 Sym->Body = New;
196 return std::error_code();
197 }
198
Rui Ueyama411c63602015-05-28 19:09:30 +0000199 // compare() returns -1, 0, or 1 if the lhs symbol is less preferable,
200 // equivalent (conflicting), or more preferable, respectively.
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000201 int Comp = Existing->compare(New);
202 if (Comp == 0) {
Rui Ueyama68633f12015-06-25 23:22:00 +0000203 llvm::errs() << "duplicate symbol: " << Existing->getDebugName()
204 << " and " << New->getDebugName() << "\n";
Rui Ueyama8fd9fb92015-06-01 02:58:15 +0000205 return make_error_code(LLDError::DuplicateSymbols);
206 }
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000207 if (Comp < 0)
208 Sym->Body = New;
Rui Ueyama411c63602015-05-28 19:09:30 +0000209 return std::error_code();
210}
211
Rui Ueyama6be90992015-07-02 22:52:33 +0000212Symbol *SymbolTable::insert(SymbolBody *New) {
213 Symbol *&Sym = Symtab[New->getName()];
214 if (Sym) {
215 New->setBackref(Sym);
216 return Sym;
217 }
218 Sym = new (Alloc) Symbol(New);
219 New->setBackref(Sym);
220 return Sym;
221}
222
Rui Ueyama411c63602015-05-28 19:09:30 +0000223// Reads an archive member file pointed by a given symbol.
224std::error_code SymbolTable::addMemberFile(Lazy *Body) {
225 auto FileOrErr = Body->getMember();
226 if (auto EC = FileOrErr.getError())
227 return EC;
228 std::unique_ptr<InputFile> File = std::move(FileOrErr.get());
229
230 // getMember returns an empty buffer if the member was already
231 // read from the library.
232 if (!File)
233 return std::error_code();
234 if (Config->Verbose)
Rui Ueyama5b2588a2015-06-08 05:43:50 +0000235 llvm::outs() << "Loaded " << File->getShortName() << " for "
Rui Ueyama411c63602015-05-28 19:09:30 +0000236 << Body->getName() << "\n";
Rui Ueyama0d2e9992015-06-23 23:56:39 +0000237 addFile(std::move(File));
238 return std::error_code();
Rui Ueyama411c63602015-05-28 19:09:30 +0000239}
240
241std::vector<Chunk *> SymbolTable::getChunks() {
242 std::vector<Chunk *> Res;
Rui Ueyama0d2e9992015-06-23 23:56:39 +0000243 for (ObjectFile *File : ObjectFiles) {
Rui Ueyama411c63602015-05-28 19:09:30 +0000244 std::vector<Chunk *> &V = File->getChunks();
245 Res.insert(Res.end(), V.begin(), V.end());
246 }
247 return Res;
248}
249
Rui Ueyama458d7442015-07-02 03:59:04 +0000250Symbol *SymbolTable::find(StringRef Name) {
Rui Ueyama45044f42015-06-29 01:03:53 +0000251 auto It = Symtab.find(Name);
252 if (It == Symtab.end())
253 return nullptr;
Rui Ueyama4b669892015-06-30 23:46:52 +0000254 return It->second;
Rui Ueyama45044f42015-06-29 01:03:53 +0000255}
256
Rui Ueyama6bf638e2015-07-02 00:04:14 +0000257void SymbolTable::mangleMaybe(Undefined *U) {
258 if (U->WeakAlias)
259 return;
Rui Ueyama0744e872015-07-02 00:21:11 +0000260 if (!isa<Undefined>(U->repl()))
Rui Ueyama6bf638e2015-07-02 00:04:14 +0000261 return;
Rui Ueyamaf5313b32015-06-28 22:16:41 +0000262
263 // In Microsoft ABI, a non-member function name is mangled this way.
Rui Ueyama6bf638e2015-07-02 00:04:14 +0000264 std::string Prefix = ("?" + U->getName() + "@@Y").str();
Rui Ueyama3d4c69c2015-07-02 02:38:59 +0000265 for (auto Pair : Symtab) {
266 StringRef Name = Pair.first;
Rui Ueyamaf5313b32015-06-28 22:16:41 +0000267 if (!Name.startswith(Prefix))
268 continue;
Rui Ueyama3d4c69c2015-07-02 02:38:59 +0000269 U->WeakAlias = addUndefined(Name);
Rui Ueyama6bf638e2015-07-02 00:04:14 +0000270 return;
Rui Ueyamaf5313b32015-06-28 22:16:41 +0000271 }
Rui Ueyamaf5313b32015-06-28 22:16:41 +0000272}
273
Rui Ueyama6bf638e2015-07-02 00:04:14 +0000274Undefined *SymbolTable::addUndefined(StringRef Name) {
Rui Ueyama3d4c69c2015-07-02 02:38:59 +0000275 auto *New = new (Alloc) Undefined(Name);
276 addSymbol(New);
277 if (auto *U = dyn_cast<Undefined>(New->repl()))
278 return U;
279 return New;
Rui Ueyama360bace2015-05-31 22:31:31 +0000280}
281
Rui Ueyama49d6cd32015-07-03 00:02:19 +0000282void SymbolTable::addAbsolute(StringRef Name, uint64_t VA) {
283 addSymbol(new (Alloc) DefinedAbsolute(Name, VA));
284}
285
Peter Collingbournebe549552015-06-26 18:58:24 +0000286void SymbolTable::printMap(llvm::raw_ostream &OS) {
287 for (ObjectFile *File : ObjectFiles) {
288 OS << File->getShortName() << ":\n";
289 for (SymbolBody *Body : File->getSymbols())
290 if (auto *R = dyn_cast<DefinedRegular>(Body))
291 if (R->isLive())
292 OS << Twine::utohexstr(Config->ImageBase + R->getRVA())
293 << " " << R->getName() << "\n";
294 }
295}
296
Peter Collingbourne60c16162015-06-01 20:10:10 +0000297std::error_code SymbolTable::addCombinedLTOObject() {
298 if (BitcodeFiles.empty())
299 return std::error_code();
300
Peter Collingbourne2612a322015-07-04 05:28:41 +0000301 // Diagnose any undefined symbols early, but do not resolve weak externals,
302 // as resolution breaks the invariant that each Symbol points to a unique
303 // SymbolBody, which we rely on to replace DefinedBitcode symbols correctly.
304 if (reportRemainingUndefines(/*Resolve=*/false))
305 return make_error_code(LLDError::BrokenFile);
306
Peter Collingbourne60c16162015-06-01 20:10:10 +0000307 // Create an object file and add it to the symbol table by replacing any
308 // DefinedBitcode symbols with the definitions in the object file.
Rui Ueyamaefba7812015-06-09 17:52:17 +0000309 LTOCodeGenerator CG;
310 auto FileOrErr = createLTOObject(&CG);
311 if (auto EC = FileOrErr.getError())
Peter Collingbourne60c16162015-06-01 20:10:10 +0000312 return EC;
Rui Ueyamaefba7812015-06-09 17:52:17 +0000313 ObjectFile *Obj = FileOrErr.get();
314
Peter Collingbourne60c16162015-06-01 20:10:10 +0000315 for (SymbolBody *Body : Obj->getSymbols()) {
316 if (!Body->isExternal())
317 continue;
Peter Collingbourne2612a322015-07-04 05:28:41 +0000318 // We should not see any new undefined symbols at this point, but we'll
319 // diagnose them later in reportRemainingUndefines().
Peter Collingbourne60c16162015-06-01 20:10:10 +0000320 StringRef Name = Body->getName();
Rui Ueyama6be90992015-07-02 22:52:33 +0000321 Symbol *Sym = insert(Body);
Peter Collingbourne60c16162015-06-01 20:10:10 +0000322
323 if (isa<DefinedBitcode>(Sym->Body)) {
Peter Collingbourne60c16162015-06-01 20:10:10 +0000324 Sym->Body = Body;
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000325 continue;
Peter Collingbourne60c16162015-06-01 20:10:10 +0000326 }
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000327 if (auto *L = dyn_cast<Lazy>(Sym->Body)) {
328 // We may see new references to runtime library symbols such as __chkstk
329 // here. These symbols must be wholly defined in non-bitcode files.
330 if (auto EC = addMemberFile(L))
Peter Collingbourne2ed4c8f2015-06-24 00:12:34 +0000331 return EC;
Rui Ueyama8d3010a2015-06-30 19:35:21 +0000332 continue;
333 }
334 SymbolBody *Existing = Sym->Body;
335 int Comp = Existing->compare(Body);
336 if (Comp == 0) {
337 llvm::errs() << "LTO: unexpected duplicate symbol: " << Name << "\n";
338 return make_error_code(LLDError::BrokenFile);
339 }
340 if (Comp < 0)
341 Sym->Body = Body;
Rui Ueyama0d2e9992015-06-23 23:56:39 +0000342 }
343
344 size_t NumBitcodeFiles = BitcodeFiles.size();
345 if (auto EC = run())
346 return EC;
347 if (BitcodeFiles.size() != NumBitcodeFiles) {
348 llvm::errs() << "LTO: late loaded symbol created new bitcode reference\n";
349 return make_error_code(LLDError::BrokenFile);
Peter Collingbourne60c16162015-06-01 20:10:10 +0000350 }
351
352 return std::error_code();
353}
354
Rui Ueyamaefba7812015-06-09 17:52:17 +0000355// Combine and compile bitcode files and then return the result
356// as a regular COFF object file.
357ErrorOr<ObjectFile *> SymbolTable::createLTOObject(LTOCodeGenerator *CG) {
358 // All symbols referenced by non-bitcode objects must be preserved.
Rui Ueyama0d2e9992015-06-23 23:56:39 +0000359 for (ObjectFile *File : ObjectFiles)
Rui Ueyamaefba7812015-06-09 17:52:17 +0000360 for (SymbolBody *Body : File->getSymbols())
Rui Ueyama0744e872015-07-02 00:21:11 +0000361 if (auto *S = dyn_cast<DefinedBitcode>(Body->repl()))
Rui Ueyamaefba7812015-06-09 17:52:17 +0000362 CG->addMustPreserveSymbol(S->getName());
363
Peter Collingbourne1b6fd1f2015-06-11 21:49:54 +0000364 // Likewise for bitcode symbols which we initially resolved to non-bitcode.
Rui Ueyama0d2e9992015-06-23 23:56:39 +0000365 for (BitcodeFile *File : BitcodeFiles)
Peter Collingbourne1b6fd1f2015-06-11 21:49:54 +0000366 for (SymbolBody *Body : File->getSymbols())
Rui Ueyama0744e872015-07-02 00:21:11 +0000367 if (isa<DefinedBitcode>(Body) && !isa<DefinedBitcode>(Body->repl()))
Peter Collingbourne1b6fd1f2015-06-11 21:49:54 +0000368 CG->addMustPreserveSymbol(Body->getName());
369
Rui Ueyamaefba7812015-06-09 17:52:17 +0000370 // Likewise for other symbols that must be preserved.
Peter Collingbourne2612a322015-07-04 05:28:41 +0000371 for (Undefined *U : Config->GCRoot) {
372 if (auto *S = dyn_cast<DefinedBitcode>(U->repl()))
373 CG->addMustPreserveSymbol(S->getName());
374 else if (auto *S = dyn_cast_or_null<DefinedBitcode>(U->getWeakAlias()))
375 CG->addMustPreserveSymbol(S->getName());
376 }
Rui Ueyamaefba7812015-06-09 17:52:17 +0000377
378 CG->setModule(BitcodeFiles[0]->releaseModule());
379 for (unsigned I = 1, E = BitcodeFiles.size(); I != E; ++I)
380 CG->addModule(BitcodeFiles[I]->getModule());
381
382 std::string ErrMsg;
383 LTOMB = CG->compile(false, false, false, ErrMsg); // take MB ownership
384 if (!LTOMB) {
385 llvm::errs() << ErrMsg << '\n';
386 return make_error_code(LLDError::BrokenFile);
387 }
Rui Ueyama0d2e9992015-06-23 23:56:39 +0000388 auto *Obj = new ObjectFile(LTOMB->getMemBufferRef());
389 Files.emplace_back(Obj);
390 ObjectFiles.push_back(Obj);
Rui Ueyamaefba7812015-06-09 17:52:17 +0000391 if (auto EC = Obj->parse())
392 return EC;
393 return Obj;
394}
395
Rui Ueyama411c63602015-05-28 19:09:30 +0000396} // namespace coff
397} // namespace lld