|
150
|
1 //===- InputFiles.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 "InputFiles.h"
|
|
|
10 #include "Config.h"
|
|
|
11 #include "InputChunks.h"
|
|
207
|
12 #include "InputElement.h"
|
|
|
13 #include "OutputSegment.h"
|
|
150
|
14 #include "SymbolTable.h"
|
|
|
15 #include "lld/Common/ErrorHandler.h"
|
|
|
16 #include "lld/Common/Memory.h"
|
|
|
17 #include "lld/Common/Reproduce.h"
|
|
|
18 #include "llvm/Object/Binary.h"
|
|
|
19 #include "llvm/Object/Wasm.h"
|
|
|
20 #include "llvm/Support/TarWriter.h"
|
|
|
21 #include "llvm/Support/raw_ostream.h"
|
|
|
22
|
|
|
23 #define DEBUG_TYPE "lld"
|
|
|
24
|
|
|
25 using namespace llvm;
|
|
|
26 using namespace llvm::object;
|
|
|
27 using namespace llvm::wasm;
|
|
|
28
|
|
|
29 namespace lld {
|
|
|
30
|
|
|
31 // Returns a string in the format of "foo.o" or "foo.a(bar.o)".
|
|
|
32 std::string toString(const wasm::InputFile *file) {
|
|
|
33 if (!file)
|
|
|
34 return "<internal>";
|
|
|
35
|
|
|
36 if (file->archiveName.empty())
|
|
|
37 return std::string(file->getName());
|
|
|
38
|
|
|
39 return (file->archiveName + "(" + file->getName() + ")").str();
|
|
|
40 }
|
|
|
41
|
|
|
42 namespace wasm {
|
|
207
|
43
|
|
|
44 void InputFile::checkArch(Triple::ArchType arch) const {
|
|
|
45 bool is64 = arch == Triple::wasm64;
|
|
|
46 if (is64 && !config->is64.hasValue()) {
|
|
|
47 fatal(toString(this) +
|
|
|
48 ": must specify -mwasm64 to process wasm64 object files");
|
|
|
49 } else if (config->is64.getValueOr(false) != is64) {
|
|
|
50 fatal(toString(this) +
|
|
|
51 ": wasm32 object file can't be linked in wasm64 mode");
|
|
|
52 }
|
|
|
53 }
|
|
|
54
|
|
150
|
55 std::unique_ptr<llvm::TarWriter> tar;
|
|
|
56
|
|
|
57 Optional<MemoryBufferRef> readFile(StringRef path) {
|
|
|
58 log("Loading: " + path);
|
|
|
59
|
|
|
60 auto mbOrErr = MemoryBuffer::getFile(path);
|
|
|
61 if (auto ec = mbOrErr.getError()) {
|
|
|
62 error("cannot open " + path + ": " + ec.message());
|
|
|
63 return None;
|
|
|
64 }
|
|
|
65 std::unique_ptr<MemoryBuffer> &mb = *mbOrErr;
|
|
|
66 MemoryBufferRef mbref = mb->getMemBufferRef();
|
|
|
67 make<std::unique_ptr<MemoryBuffer>>(std::move(mb)); // take MB ownership
|
|
|
68
|
|
|
69 if (tar)
|
|
|
70 tar->append(relativeToRoot(path), mbref.getBuffer());
|
|
|
71 return mbref;
|
|
|
72 }
|
|
|
73
|
|
207
|
74 InputFile *createObjectFile(MemoryBufferRef mb, StringRef archiveName) {
|
|
150
|
75 file_magic magic = identify_magic(mb.getBuffer());
|
|
|
76 if (magic == file_magic::wasm_object) {
|
|
|
77 std::unique_ptr<Binary> bin =
|
|
|
78 CHECK(createBinary(mb), mb.getBufferIdentifier());
|
|
|
79 auto *obj = cast<WasmObjectFile>(bin.get());
|
|
|
80 if (obj->isSharedObject())
|
|
|
81 return make<SharedFile>(mb);
|
|
|
82 return make<ObjFile>(mb, archiveName);
|
|
|
83 }
|
|
|
84
|
|
|
85 if (magic == file_magic::bitcode)
|
|
|
86 return make<BitcodeFile>(mb, archiveName);
|
|
|
87
|
|
|
88 fatal("unknown file type: " + mb.getBufferIdentifier());
|
|
|
89 }
|
|
|
90
|
|
|
91 void ObjFile::dumpInfo() const {
|
|
|
92 log("info for: " + toString(this) +
|
|
|
93 "\n Symbols : " + Twine(symbols.size()) +
|
|
|
94 "\n Function Imports : " + Twine(wasmObj->getNumImportedFunctions()) +
|
|
|
95 "\n Global Imports : " + Twine(wasmObj->getNumImportedGlobals()) +
|
|
207
|
96 "\n Event Imports : " + Twine(wasmObj->getNumImportedEvents()) +
|
|
|
97 "\n Table Imports : " + Twine(wasmObj->getNumImportedTables()));
|
|
150
|
98 }
|
|
|
99
|
|
|
100 // Relocations contain either symbol or type indices. This function takes a
|
|
|
101 // relocation and returns relocated index (i.e. translates from the input
|
|
|
102 // symbol/type space to the output symbol/type space).
|
|
|
103 uint32_t ObjFile::calcNewIndex(const WasmRelocation &reloc) const {
|
|
|
104 if (reloc.Type == R_WASM_TYPE_INDEX_LEB) {
|
|
|
105 assert(typeIsUsed[reloc.Index]);
|
|
|
106 return typeMap[reloc.Index];
|
|
|
107 }
|
|
|
108 const Symbol *sym = symbols[reloc.Index];
|
|
|
109 if (auto *ss = dyn_cast<SectionSymbol>(sym))
|
|
|
110 sym = ss->getOutputSectionSymbol();
|
|
|
111 return sym->getOutputSymbolIndex();
|
|
|
112 }
|
|
|
113
|
|
|
114 // Relocations can contain addend for combined sections. This function takes a
|
|
|
115 // relocation and returns updated addend by offset in the output section.
|
|
207
|
116 uint64_t ObjFile::calcNewAddend(const WasmRelocation &reloc) const {
|
|
150
|
117 switch (reloc.Type) {
|
|
|
118 case R_WASM_MEMORY_ADDR_LEB:
|
|
207
|
119 case R_WASM_MEMORY_ADDR_LEB64:
|
|
|
120 case R_WASM_MEMORY_ADDR_SLEB64:
|
|
150
|
121 case R_WASM_MEMORY_ADDR_SLEB:
|
|
|
122 case R_WASM_MEMORY_ADDR_REL_SLEB:
|
|
207
|
123 case R_WASM_MEMORY_ADDR_REL_SLEB64:
|
|
150
|
124 case R_WASM_MEMORY_ADDR_I32:
|
|
207
|
125 case R_WASM_MEMORY_ADDR_I64:
|
|
|
126 case R_WASM_MEMORY_ADDR_TLS_SLEB:
|
|
150
|
127 case R_WASM_FUNCTION_OFFSET_I32:
|
|
207
|
128 case R_WASM_FUNCTION_OFFSET_I64:
|
|
|
129 case R_WASM_MEMORY_ADDR_LOCREL_I32:
|
|
150
|
130 return reloc.Addend;
|
|
|
131 case R_WASM_SECTION_OFFSET_I32:
|
|
207
|
132 return getSectionSymbol(reloc.Index)->section->getOffset(reloc.Addend);
|
|
150
|
133 default:
|
|
|
134 llvm_unreachable("unexpected relocation type");
|
|
|
135 }
|
|
|
136 }
|
|
|
137
|
|
|
138 // Translate from the relocation's index into the final linked output value.
|
|
207
|
139 uint64_t ObjFile::calcNewValue(const WasmRelocation &reloc, uint64_t tombstone,
|
|
|
140 const InputChunk *chunk) const {
|
|
150
|
141 const Symbol* sym = nullptr;
|
|
|
142 if (reloc.Type != R_WASM_TYPE_INDEX_LEB) {
|
|
|
143 sym = symbols[reloc.Index];
|
|
|
144
|
|
|
145 // We can end up with relocations against non-live symbols. For example
|
|
207
|
146 // in debug sections. We return a tombstone value in debug symbol sections
|
|
|
147 // so this will not produce a valid range conflicting with ranges of actual
|
|
|
148 // code. In other sections we return reloc.Addend.
|
|
|
149
|
|
|
150 if (!isa<SectionSymbol>(sym) && !sym->isLive())
|
|
|
151 return tombstone ? tombstone : reloc.Addend;
|
|
150
|
152 }
|
|
|
153
|
|
|
154 switch (reloc.Type) {
|
|
|
155 case R_WASM_TABLE_INDEX_I32:
|
|
207
|
156 case R_WASM_TABLE_INDEX_I64:
|
|
150
|
157 case R_WASM_TABLE_INDEX_SLEB:
|
|
207
|
158 case R_WASM_TABLE_INDEX_SLEB64:
|
|
|
159 case R_WASM_TABLE_INDEX_REL_SLEB:
|
|
|
160 case R_WASM_TABLE_INDEX_REL_SLEB64: {
|
|
150
|
161 if (!getFunctionSymbol(reloc.Index)->hasTableIndex())
|
|
|
162 return 0;
|
|
|
163 uint32_t index = getFunctionSymbol(reloc.Index)->getTableIndex();
|
|
207
|
164 if (reloc.Type == R_WASM_TABLE_INDEX_REL_SLEB ||
|
|
|
165 reloc.Type == R_WASM_TABLE_INDEX_REL_SLEB64)
|
|
150
|
166 index -= config->tableBase;
|
|
|
167 return index;
|
|
|
168 }
|
|
207
|
169 case R_WASM_MEMORY_ADDR_LEB:
|
|
|
170 case R_WASM_MEMORY_ADDR_LEB64:
|
|
150
|
171 case R_WASM_MEMORY_ADDR_SLEB:
|
|
207
|
172 case R_WASM_MEMORY_ADDR_SLEB64:
|
|
|
173 case R_WASM_MEMORY_ADDR_REL_SLEB:
|
|
|
174 case R_WASM_MEMORY_ADDR_REL_SLEB64:
|
|
150
|
175 case R_WASM_MEMORY_ADDR_I32:
|
|
207
|
176 case R_WASM_MEMORY_ADDR_I64:
|
|
|
177 case R_WASM_MEMORY_ADDR_LOCREL_I32: {
|
|
150
|
178 if (isa<UndefinedData>(sym) || sym->isUndefWeak())
|
|
|
179 return 0;
|
|
207
|
180 auto D = cast<DefinedData>(sym);
|
|
|
181 // Treat non-TLS relocation against symbols that live in the TLS segment
|
|
|
182 // like TLS relocations. This beaviour exists to support older object
|
|
|
183 // files created before we introduced TLS relocations.
|
|
|
184 // TODO(sbc): Remove this legacy behaviour one day. This will break
|
|
|
185 // backward compat with old object files built with `-fPIC`.
|
|
|
186 if (D->segment && D->segment->outputSeg->isTLS())
|
|
|
187 return D->getOutputSegmentOffset() + reloc.Addend;
|
|
|
188
|
|
|
189 uint64_t value = D->getVA() + reloc.Addend;
|
|
|
190 if (reloc.Type == R_WASM_MEMORY_ADDR_LOCREL_I32) {
|
|
|
191 const auto *segment = cast<InputSegment>(chunk);
|
|
|
192 uint64_t p = segment->outputSeg->startVA + segment->outputSegmentOffset +
|
|
|
193 reloc.Offset - segment->getInputSectionOffset();
|
|
|
194 value -= p;
|
|
|
195 }
|
|
|
196 return value;
|
|
|
197 }
|
|
|
198 case R_WASM_MEMORY_ADDR_TLS_SLEB:
|
|
|
199 if (isa<UndefinedData>(sym) || sym->isUndefWeak())
|
|
|
200 return 0;
|
|
|
201 // TLS relocations are relative to the start of the TLS output segment
|
|
|
202 return cast<DefinedData>(sym)->getOutputSegmentOffset() + reloc.Addend;
|
|
150
|
203 case R_WASM_TYPE_INDEX_LEB:
|
|
|
204 return typeMap[reloc.Index];
|
|
|
205 case R_WASM_FUNCTION_INDEX_LEB:
|
|
|
206 return getFunctionSymbol(reloc.Index)->getFunctionIndex();
|
|
|
207 case R_WASM_GLOBAL_INDEX_LEB:
|
|
173
|
208 case R_WASM_GLOBAL_INDEX_I32:
|
|
150
|
209 if (auto gs = dyn_cast<GlobalSymbol>(sym))
|
|
|
210 return gs->getGlobalIndex();
|
|
|
211 return sym->getGOTIndex();
|
|
|
212 case R_WASM_EVENT_INDEX_LEB:
|
|
|
213 return getEventSymbol(reloc.Index)->getEventIndex();
|
|
207
|
214 case R_WASM_FUNCTION_OFFSET_I32:
|
|
|
215 case R_WASM_FUNCTION_OFFSET_I64: {
|
|
150
|
216 auto *f = cast<DefinedFunction>(sym);
|
|
207
|
217 return f->function->getOffset(f->function->getFunctionCodeOffset() +
|
|
|
218 reloc.Addend);
|
|
150
|
219 }
|
|
|
220 case R_WASM_SECTION_OFFSET_I32:
|
|
207
|
221 return getSectionSymbol(reloc.Index)->section->getOffset(reloc.Addend);
|
|
|
222 case R_WASM_TABLE_NUMBER_LEB:
|
|
|
223 return getTableSymbol(reloc.Index)->getTableNumber();
|
|
150
|
224 default:
|
|
|
225 llvm_unreachable("unknown relocation type");
|
|
|
226 }
|
|
|
227 }
|
|
|
228
|
|
|
229 template <class T>
|
|
|
230 static void setRelocs(const std::vector<T *> &chunks,
|
|
|
231 const WasmSection *section) {
|
|
|
232 if (!section)
|
|
|
233 return;
|
|
|
234
|
|
|
235 ArrayRef<WasmRelocation> relocs = section->Relocations;
|
|
173
|
236 assert(llvm::is_sorted(
|
|
|
237 relocs, [](const WasmRelocation &r1, const WasmRelocation &r2) {
|
|
|
238 return r1.Offset < r2.Offset;
|
|
150
|
239 }));
|
|
173
|
240 assert(llvm::is_sorted(chunks, [](InputChunk *c1, InputChunk *c2) {
|
|
|
241 return c1->getInputSectionOffset() < c2->getInputSectionOffset();
|
|
|
242 }));
|
|
150
|
243
|
|
|
244 auto relocsNext = relocs.begin();
|
|
|
245 auto relocsEnd = relocs.end();
|
|
|
246 auto relocLess = [](const WasmRelocation &r, uint32_t val) {
|
|
|
247 return r.Offset < val;
|
|
|
248 };
|
|
|
249 for (InputChunk *c : chunks) {
|
|
|
250 auto relocsStart = std::lower_bound(relocsNext, relocsEnd,
|
|
|
251 c->getInputSectionOffset(), relocLess);
|
|
|
252 relocsNext = std::lower_bound(
|
|
|
253 relocsStart, relocsEnd, c->getInputSectionOffset() + c->getInputSize(),
|
|
|
254 relocLess);
|
|
|
255 c->setRelocations(ArrayRef<WasmRelocation>(relocsStart, relocsNext));
|
|
|
256 }
|
|
|
257 }
|
|
|
258
|
|
207
|
259 // An object file can have two approaches to tables. With the reference-types
|
|
|
260 // feature enabled, input files that define or use tables declare the tables
|
|
|
261 // using symbols, and record each use with a relocation. This way when the
|
|
|
262 // linker combines inputs, it can collate the tables used by the inputs,
|
|
|
263 // assigning them distinct table numbers, and renumber all the uses as
|
|
|
264 // appropriate. At the same time, the linker has special logic to build the
|
|
|
265 // indirect function table if it is needed.
|
|
|
266 //
|
|
|
267 // However, MVP object files (those that target WebAssembly 1.0, the "minimum
|
|
|
268 // viable product" version of WebAssembly) neither write table symbols nor
|
|
|
269 // record relocations. These files can have at most one table, the indirect
|
|
|
270 // function table used by call_indirect and which is the address space for
|
|
|
271 // function pointers. If this table is present, it is always an import. If we
|
|
|
272 // have a file with a table import but no table symbols, it is an MVP object
|
|
|
273 // file. synthesizeMVPIndirectFunctionTableSymbolIfNeeded serves as a shim when
|
|
|
274 // loading these input files, defining the missing symbol to allow the indirect
|
|
|
275 // function table to be built.
|
|
|
276 //
|
|
|
277 // As indirect function table table usage in MVP objects cannot be relocated,
|
|
|
278 // the linker must ensure that this table gets assigned index zero.
|
|
|
279 void ObjFile::addLegacyIndirectFunctionTableIfNeeded(
|
|
|
280 uint32_t tableSymbolCount) {
|
|
|
281 uint32_t tableCount = wasmObj->getNumImportedTables() + tables.size();
|
|
|
282
|
|
|
283 // If there are symbols for all tables, then all is good.
|
|
|
284 if (tableCount == tableSymbolCount)
|
|
|
285 return;
|
|
|
286
|
|
|
287 // It's possible for an input to define tables and also use the indirect
|
|
|
288 // function table, but forget to compile with -mattr=+reference-types.
|
|
|
289 // For these newer files, we require symbols for all tables, and
|
|
|
290 // relocations for all of their uses.
|
|
|
291 if (tableSymbolCount != 0) {
|
|
|
292 error(toString(this) +
|
|
|
293 ": expected one symbol table entry for each of the " +
|
|
|
294 Twine(tableCount) + " table(s) present, but got " +
|
|
|
295 Twine(tableSymbolCount) + " symbol(s) instead.");
|
|
|
296 return;
|
|
|
297 }
|
|
|
298
|
|
|
299 // An MVP object file can have up to one table import, for the indirect
|
|
|
300 // function table, but will have no table definitions.
|
|
|
301 if (tables.size()) {
|
|
|
302 error(toString(this) +
|
|
|
303 ": unexpected table definition(s) without corresponding "
|
|
|
304 "symbol-table entries.");
|
|
|
305 return;
|
|
|
306 }
|
|
|
307
|
|
|
308 // An MVP object file can have only one table import.
|
|
|
309 if (tableCount != 1) {
|
|
|
310 error(toString(this) +
|
|
|
311 ": multiple table imports, but no corresponding symbol-table "
|
|
|
312 "entries.");
|
|
|
313 return;
|
|
|
314 }
|
|
|
315
|
|
|
316 const WasmImport *tableImport = nullptr;
|
|
|
317 for (const auto &import : wasmObj->imports()) {
|
|
|
318 if (import.Kind == WASM_EXTERNAL_TABLE) {
|
|
|
319 assert(!tableImport);
|
|
|
320 tableImport = &import;
|
|
|
321 }
|
|
|
322 }
|
|
|
323 assert(tableImport);
|
|
|
324
|
|
|
325 // We can only synthesize a symtab entry for the indirect function table; if
|
|
|
326 // it has an unexpected name or type, assume that it's not actually the
|
|
|
327 // indirect function table.
|
|
|
328 if (tableImport->Field != functionTableName ||
|
|
|
329 tableImport->Table.ElemType != uint8_t(ValType::FUNCREF)) {
|
|
|
330 error(toString(this) + ": table import " + Twine(tableImport->Field) +
|
|
|
331 " is missing a symbol table entry.");
|
|
|
332 return;
|
|
|
333 }
|
|
|
334
|
|
|
335 auto *info = make<WasmSymbolInfo>();
|
|
|
336 info->Name = tableImport->Field;
|
|
|
337 info->Kind = WASM_SYMBOL_TYPE_TABLE;
|
|
|
338 info->ImportModule = tableImport->Module;
|
|
|
339 info->ImportName = tableImport->Field;
|
|
|
340 info->Flags = WASM_SYMBOL_UNDEFINED;
|
|
|
341 info->Flags |= WASM_SYMBOL_NO_STRIP;
|
|
|
342 info->ElementIndex = 0;
|
|
|
343 LLVM_DEBUG(dbgs() << "Synthesizing symbol for table import: " << info->Name
|
|
|
344 << "\n");
|
|
|
345 const WasmGlobalType *globalType = nullptr;
|
|
|
346 const WasmEventType *eventType = nullptr;
|
|
|
347 const WasmSignature *signature = nullptr;
|
|
|
348 auto *wasmSym = make<WasmSymbol>(*info, globalType, &tableImport->Table,
|
|
|
349 eventType, signature);
|
|
|
350 Symbol *sym = createUndefined(*wasmSym, false);
|
|
|
351 // We're only sure it's a TableSymbol if the createUndefined succeeded.
|
|
|
352 if (errorCount())
|
|
|
353 return;
|
|
|
354 symbols.push_back(sym);
|
|
|
355 // Because there are no TABLE_NUMBER relocs, we can't compute accurate
|
|
|
356 // liveness info; instead, just mark the symbol as always live.
|
|
|
357 sym->markLive();
|
|
|
358
|
|
|
359 // We assume that this compilation unit has unrelocatable references to
|
|
|
360 // this table.
|
|
|
361 config->legacyFunctionTable = true;
|
|
|
362 }
|
|
|
363
|
|
|
364 static bool shouldMerge(const WasmSection &sec) {
|
|
|
365 if (config->optimize == 0)
|
|
|
366 return false;
|
|
|
367 // Sadly we don't have section attributes yet for custom sections, so we
|
|
|
368 // currently go by the name alone.
|
|
|
369 // TODO(sbc): Add ability for wasm sections to carry flags so we don't
|
|
|
370 // need to use names here.
|
|
|
371 // For now, keep in sync with uses of wasm::WASM_SEG_FLAG_STRINGS in
|
|
|
372 // MCObjectFileInfo::initWasmMCObjectFileInfo which creates these custom
|
|
|
373 // sections.
|
|
|
374 return sec.Name == ".debug_str" || sec.Name == ".debug_str.dwo" ||
|
|
|
375 sec.Name == ".debug_line_str";
|
|
|
376 }
|
|
|
377
|
|
|
378 static bool shouldMerge(const WasmSegment &seg) {
|
|
|
379 // As of now we only support merging strings, and only with single byte
|
|
|
380 // alignment (2^0).
|
|
|
381 if (!(seg.Data.LinkingFlags & WASM_SEG_FLAG_STRINGS) ||
|
|
|
382 (seg.Data.Alignment != 0))
|
|
|
383 return false;
|
|
|
384
|
|
|
385 // On a regular link we don't merge sections if -O0 (default is -O1). This
|
|
|
386 // sometimes makes the linker significantly faster, although the output will
|
|
|
387 // be bigger.
|
|
|
388 if (config->optimize == 0)
|
|
|
389 return false;
|
|
|
390
|
|
|
391 // A mergeable section with size 0 is useless because they don't have
|
|
|
392 // any data to merge. A mergeable string section with size 0 can be
|
|
|
393 // argued as invalid because it doesn't end with a null character.
|
|
|
394 // We'll avoid a mess by handling them as if they were non-mergeable.
|
|
|
395 if (seg.Data.Content.size() == 0)
|
|
|
396 return false;
|
|
|
397
|
|
|
398 return true;
|
|
|
399 }
|
|
|
400
|
|
150
|
401 void ObjFile::parse(bool ignoreComdats) {
|
|
|
402 // Parse a memory buffer as a wasm file.
|
|
|
403 LLVM_DEBUG(dbgs() << "Parsing object: " << toString(this) << "\n");
|
|
|
404 std::unique_ptr<Binary> bin = CHECK(createBinary(mb), toString(this));
|
|
|
405
|
|
|
406 auto *obj = dyn_cast<WasmObjectFile>(bin.get());
|
|
|
407 if (!obj)
|
|
|
408 fatal(toString(this) + ": not a wasm file");
|
|
|
409 if (!obj->isRelocatableObject())
|
|
|
410 fatal(toString(this) + ": not a relocatable wasm file");
|
|
|
411
|
|
|
412 bin.release();
|
|
|
413 wasmObj.reset(obj);
|
|
|
414
|
|
207
|
415 checkArch(obj->getArch());
|
|
|
416
|
|
150
|
417 // Build up a map of function indices to table indices for use when
|
|
|
418 // verifying the existing table index relocations
|
|
|
419 uint32_t totalFunctions =
|
|
|
420 wasmObj->getNumImportedFunctions() + wasmObj->functions().size();
|
|
207
|
421 tableEntriesRel.resize(totalFunctions);
|
|
150
|
422 tableEntries.resize(totalFunctions);
|
|
|
423 for (const WasmElemSegment &seg : wasmObj->elements()) {
|
|
207
|
424 int64_t offset;
|
|
|
425 if (seg.Offset.Opcode == WASM_OPCODE_I32_CONST)
|
|
|
426 offset = seg.Offset.Value.Int32;
|
|
|
427 else if (seg.Offset.Opcode == WASM_OPCODE_I64_CONST)
|
|
|
428 offset = seg.Offset.Value.Int64;
|
|
|
429 else
|
|
150
|
430 fatal(toString(this) + ": invalid table elements");
|
|
207
|
431 for (size_t index = 0; index < seg.Functions.size(); index++) {
|
|
|
432 auto functionIndex = seg.Functions[index];
|
|
|
433 tableEntriesRel[functionIndex] = index;
|
|
150
|
434 tableEntries[functionIndex] = offset + index;
|
|
|
435 }
|
|
|
436 }
|
|
|
437
|
|
207
|
438 ArrayRef<StringRef> comdats = wasmObj->linkingData().Comdats;
|
|
|
439 for (StringRef comdat : comdats) {
|
|
|
440 bool isNew = ignoreComdats || symtab->addComdat(comdat);
|
|
|
441 keptComdats.push_back(isNew);
|
|
|
442 }
|
|
|
443
|
|
150
|
444 uint32_t sectionIndex = 0;
|
|
|
445
|
|
|
446 // Bool for each symbol, true if called directly. This allows us to implement
|
|
|
447 // a weaker form of signature checking where undefined functions that are not
|
|
|
448 // called directly (i.e. only address taken) don't have to match the defined
|
|
|
449 // function's signature. We cannot do this for directly called functions
|
|
|
450 // because those signatures are checked at validation times.
|
|
|
451 // See https://bugs.llvm.org/show_bug.cgi?id=40412
|
|
|
452 std::vector<bool> isCalledDirectly(wasmObj->getNumberOfSymbols(), false);
|
|
|
453 for (const SectionRef &sec : wasmObj->sections()) {
|
|
|
454 const WasmSection §ion = wasmObj->getWasmSection(sec);
|
|
|
455 // Wasm objects can have at most one code and one data section.
|
|
|
456 if (section.Type == WASM_SEC_CODE) {
|
|
|
457 assert(!codeSection);
|
|
|
458 codeSection = §ion;
|
|
|
459 } else if (section.Type == WASM_SEC_DATA) {
|
|
|
460 assert(!dataSection);
|
|
|
461 dataSection = §ion;
|
|
|
462 } else if (section.Type == WASM_SEC_CUSTOM) {
|
|
207
|
463 InputChunk *customSec;
|
|
|
464 if (shouldMerge(section))
|
|
|
465 customSec = make<MergeInputChunk>(section, this);
|
|
|
466 else
|
|
|
467 customSec = make<InputSection>(section, this);
|
|
|
468 customSec->discarded = isExcludedByComdat(customSec);
|
|
|
469 customSections.emplace_back(customSec);
|
|
150
|
470 customSections.back()->setRelocations(section.Relocations);
|
|
|
471 customSectionsByIndex[sectionIndex] = customSections.back();
|
|
|
472 }
|
|
|
473 sectionIndex++;
|
|
|
474 // Scans relocations to determine if a function symbol is called directly.
|
|
|
475 for (const WasmRelocation &reloc : section.Relocations)
|
|
|
476 if (reloc.Type == R_WASM_FUNCTION_INDEX_LEB)
|
|
|
477 isCalledDirectly[reloc.Index] = true;
|
|
|
478 }
|
|
|
479
|
|
|
480 typeMap.resize(getWasmObj()->types().size());
|
|
|
481 typeIsUsed.resize(getWasmObj()->types().size(), false);
|
|
|
482
|
|
|
483
|
|
|
484 // Populate `Segments`.
|
|
|
485 for (const WasmSegment &s : wasmObj->dataSegments()) {
|
|
207
|
486 InputChunk *seg;
|
|
|
487 if (shouldMerge(s)) {
|
|
|
488 seg = make<MergeInputChunk>(s, this);
|
|
|
489 } else
|
|
|
490 seg = make<InputSegment>(s, this);
|
|
150
|
491 seg->discarded = isExcludedByComdat(seg);
|
|
207
|
492
|
|
150
|
493 segments.emplace_back(seg);
|
|
|
494 }
|
|
|
495 setRelocs(segments, dataSection);
|
|
|
496
|
|
|
497 // Populate `Functions`.
|
|
|
498 ArrayRef<WasmFunction> funcs = wasmObj->functions();
|
|
|
499 ArrayRef<uint32_t> funcTypes = wasmObj->functionTypes();
|
|
|
500 ArrayRef<WasmSignature> types = wasmObj->types();
|
|
|
501 functions.reserve(funcs.size());
|
|
|
502
|
|
|
503 for (size_t i = 0, e = funcs.size(); i != e; ++i) {
|
|
|
504 auto* func = make<InputFunction>(types[funcTypes[i]], &funcs[i], this);
|
|
|
505 func->discarded = isExcludedByComdat(func);
|
|
|
506 functions.emplace_back(func);
|
|
|
507 }
|
|
|
508 setRelocs(functions, codeSection);
|
|
|
509
|
|
207
|
510 // Populate `Tables`.
|
|
|
511 for (const WasmTable &t : wasmObj->tables())
|
|
|
512 tables.emplace_back(make<InputTable>(t, this));
|
|
|
513
|
|
150
|
514 // Populate `Globals`.
|
|
|
515 for (const WasmGlobal &g : wasmObj->globals())
|
|
|
516 globals.emplace_back(make<InputGlobal>(g, this));
|
|
|
517
|
|
|
518 // Populate `Events`.
|
|
|
519 for (const WasmEvent &e : wasmObj->events())
|
|
|
520 events.emplace_back(make<InputEvent>(types[e.Type.SigIndex], e, this));
|
|
|
521
|
|
|
522 // Populate `Symbols` based on the symbols in the object.
|
|
|
523 symbols.reserve(wasmObj->getNumberOfSymbols());
|
|
207
|
524 uint32_t tableSymbolCount = 0;
|
|
150
|
525 for (const SymbolRef &sym : wasmObj->symbols()) {
|
|
|
526 const WasmSymbol &wasmSym = wasmObj->getWasmSymbol(sym.getRawDataRefImpl());
|
|
207
|
527 if (wasmSym.isTypeTable())
|
|
|
528 tableSymbolCount++;
|
|
150
|
529 if (wasmSym.isDefined()) {
|
|
|
530 // createDefined may fail if the symbol is comdat excluded in which case
|
|
|
531 // we fall back to creating an undefined symbol
|
|
|
532 if (Symbol *d = createDefined(wasmSym)) {
|
|
|
533 symbols.push_back(d);
|
|
|
534 continue;
|
|
|
535 }
|
|
|
536 }
|
|
|
537 size_t idx = symbols.size();
|
|
|
538 symbols.push_back(createUndefined(wasmSym, isCalledDirectly[idx]));
|
|
|
539 }
|
|
207
|
540
|
|
|
541 addLegacyIndirectFunctionTableIfNeeded(tableSymbolCount);
|
|
150
|
542 }
|
|
|
543
|
|
|
544 bool ObjFile::isExcludedByComdat(InputChunk *chunk) const {
|
|
|
545 uint32_t c = chunk->getComdat();
|
|
|
546 if (c == UINT32_MAX)
|
|
|
547 return false;
|
|
|
548 return !keptComdats[c];
|
|
|
549 }
|
|
|
550
|
|
|
551 FunctionSymbol *ObjFile::getFunctionSymbol(uint32_t index) const {
|
|
|
552 return cast<FunctionSymbol>(symbols[index]);
|
|
|
553 }
|
|
|
554
|
|
|
555 GlobalSymbol *ObjFile::getGlobalSymbol(uint32_t index) const {
|
|
|
556 return cast<GlobalSymbol>(symbols[index]);
|
|
|
557 }
|
|
|
558
|
|
|
559 EventSymbol *ObjFile::getEventSymbol(uint32_t index) const {
|
|
|
560 return cast<EventSymbol>(symbols[index]);
|
|
|
561 }
|
|
|
562
|
|
207
|
563 TableSymbol *ObjFile::getTableSymbol(uint32_t index) const {
|
|
|
564 return cast<TableSymbol>(symbols[index]);
|
|
|
565 }
|
|
|
566
|
|
150
|
567 SectionSymbol *ObjFile::getSectionSymbol(uint32_t index) const {
|
|
|
568 return cast<SectionSymbol>(symbols[index]);
|
|
|
569 }
|
|
|
570
|
|
|
571 DataSymbol *ObjFile::getDataSymbol(uint32_t index) const {
|
|
|
572 return cast<DataSymbol>(symbols[index]);
|
|
|
573 }
|
|
|
574
|
|
|
575 Symbol *ObjFile::createDefined(const WasmSymbol &sym) {
|
|
|
576 StringRef name = sym.Info.Name;
|
|
|
577 uint32_t flags = sym.Info.Flags;
|
|
|
578
|
|
|
579 switch (sym.Info.Kind) {
|
|
|
580 case WASM_SYMBOL_TYPE_FUNCTION: {
|
|
|
581 InputFunction *func =
|
|
|
582 functions[sym.Info.ElementIndex - wasmObj->getNumImportedFunctions()];
|
|
|
583 if (sym.isBindingLocal())
|
|
|
584 return make<DefinedFunction>(name, flags, this, func);
|
|
|
585 if (func->discarded)
|
|
|
586 return nullptr;
|
|
|
587 return symtab->addDefinedFunction(name, flags, this, func);
|
|
|
588 }
|
|
|
589 case WASM_SYMBOL_TYPE_DATA: {
|
|
207
|
590 InputChunk *seg = segments[sym.Info.DataRef.Segment];
|
|
|
591 auto offset = sym.Info.DataRef.Offset;
|
|
|
592 auto size = sym.Info.DataRef.Size;
|
|
150
|
593 if (sym.isBindingLocal())
|
|
|
594 return make<DefinedData>(name, flags, this, seg, offset, size);
|
|
|
595 if (seg->discarded)
|
|
|
596 return nullptr;
|
|
|
597 return symtab->addDefinedData(name, flags, this, seg, offset, size);
|
|
|
598 }
|
|
|
599 case WASM_SYMBOL_TYPE_GLOBAL: {
|
|
|
600 InputGlobal *global =
|
|
|
601 globals[sym.Info.ElementIndex - wasmObj->getNumImportedGlobals()];
|
|
|
602 if (sym.isBindingLocal())
|
|
|
603 return make<DefinedGlobal>(name, flags, this, global);
|
|
|
604 return symtab->addDefinedGlobal(name, flags, this, global);
|
|
|
605 }
|
|
|
606 case WASM_SYMBOL_TYPE_SECTION: {
|
|
207
|
607 InputChunk *section = customSectionsByIndex[sym.Info.ElementIndex];
|
|
150
|
608 assert(sym.isBindingLocal());
|
|
207
|
609 // Need to return null if discarded here? data and func only do that when
|
|
|
610 // binding is not local.
|
|
|
611 if (section->discarded)
|
|
|
612 return nullptr;
|
|
150
|
613 return make<SectionSymbol>(flags, section, this);
|
|
|
614 }
|
|
|
615 case WASM_SYMBOL_TYPE_EVENT: {
|
|
|
616 InputEvent *event =
|
|
|
617 events[sym.Info.ElementIndex - wasmObj->getNumImportedEvents()];
|
|
|
618 if (sym.isBindingLocal())
|
|
|
619 return make<DefinedEvent>(name, flags, this, event);
|
|
|
620 return symtab->addDefinedEvent(name, flags, this, event);
|
|
|
621 }
|
|
207
|
622 case WASM_SYMBOL_TYPE_TABLE: {
|
|
|
623 InputTable *table =
|
|
|
624 tables[sym.Info.ElementIndex - wasmObj->getNumImportedTables()];
|
|
|
625 if (sym.isBindingLocal())
|
|
|
626 return make<DefinedTable>(name, flags, this, table);
|
|
|
627 return symtab->addDefinedTable(name, flags, this, table);
|
|
|
628 }
|
|
150
|
629 }
|
|
|
630 llvm_unreachable("unknown symbol kind");
|
|
|
631 }
|
|
|
632
|
|
|
633 Symbol *ObjFile::createUndefined(const WasmSymbol &sym, bool isCalledDirectly) {
|
|
|
634 StringRef name = sym.Info.Name;
|
|
|
635 uint32_t flags = sym.Info.Flags | WASM_SYMBOL_UNDEFINED;
|
|
|
636
|
|
|
637 switch (sym.Info.Kind) {
|
|
|
638 case WASM_SYMBOL_TYPE_FUNCTION:
|
|
|
639 if (sym.isBindingLocal())
|
|
|
640 return make<UndefinedFunction>(name, sym.Info.ImportName,
|
|
|
641 sym.Info.ImportModule, flags, this,
|
|
|
642 sym.Signature, isCalledDirectly);
|
|
|
643 return symtab->addUndefinedFunction(name, sym.Info.ImportName,
|
|
|
644 sym.Info.ImportModule, flags, this,
|
|
|
645 sym.Signature, isCalledDirectly);
|
|
|
646 case WASM_SYMBOL_TYPE_DATA:
|
|
|
647 if (sym.isBindingLocal())
|
|
|
648 return make<UndefinedData>(name, flags, this);
|
|
|
649 return symtab->addUndefinedData(name, flags, this);
|
|
|
650 case WASM_SYMBOL_TYPE_GLOBAL:
|
|
|
651 if (sym.isBindingLocal())
|
|
|
652 return make<UndefinedGlobal>(name, sym.Info.ImportName,
|
|
|
653 sym.Info.ImportModule, flags, this,
|
|
|
654 sym.GlobalType);
|
|
|
655 return symtab->addUndefinedGlobal(name, sym.Info.ImportName,
|
|
|
656 sym.Info.ImportModule, flags, this,
|
|
|
657 sym.GlobalType);
|
|
207
|
658 case WASM_SYMBOL_TYPE_TABLE:
|
|
|
659 if (sym.isBindingLocal())
|
|
|
660 return make<UndefinedTable>(name, sym.Info.ImportName,
|
|
|
661 sym.Info.ImportModule, flags, this,
|
|
|
662 sym.TableType);
|
|
|
663 return symtab->addUndefinedTable(name, sym.Info.ImportName,
|
|
|
664 sym.Info.ImportModule, flags, this,
|
|
|
665 sym.TableType);
|
|
150
|
666 case WASM_SYMBOL_TYPE_SECTION:
|
|
|
667 llvm_unreachable("section symbols cannot be undefined");
|
|
|
668 }
|
|
|
669 llvm_unreachable("unknown symbol kind");
|
|
|
670 }
|
|
|
671
|
|
|
672 void ArchiveFile::parse() {
|
|
|
673 // Parse a MemoryBufferRef as an archive file.
|
|
|
674 LLVM_DEBUG(dbgs() << "Parsing library: " << toString(this) << "\n");
|
|
|
675 file = CHECK(Archive::create(mb), toString(this));
|
|
|
676
|
|
|
677 // Read the symbol table to construct Lazy symbols.
|
|
|
678 int count = 0;
|
|
|
679 for (const Archive::Symbol &sym : file->symbols()) {
|
|
|
680 symtab->addLazy(this, &sym);
|
|
|
681 ++count;
|
|
|
682 }
|
|
|
683 LLVM_DEBUG(dbgs() << "Read " << count << " symbols\n");
|
|
|
684 }
|
|
|
685
|
|
|
686 void ArchiveFile::addMember(const Archive::Symbol *sym) {
|
|
|
687 const Archive::Child &c =
|
|
|
688 CHECK(sym->getMember(),
|
|
|
689 "could not get the member for symbol " + sym->getName());
|
|
|
690
|
|
|
691 // Don't try to load the same member twice (this can happen when members
|
|
|
692 // mutually reference each other).
|
|
|
693 if (!seen.insert(c.getChildOffset()).second)
|
|
|
694 return;
|
|
|
695
|
|
|
696 LLVM_DEBUG(dbgs() << "loading lazy: " << sym->getName() << "\n");
|
|
|
697 LLVM_DEBUG(dbgs() << "from archive: " << toString(this) << "\n");
|
|
|
698
|
|
|
699 MemoryBufferRef mb =
|
|
|
700 CHECK(c.getMemoryBufferRef(),
|
|
|
701 "could not get the buffer for the member defining symbol " +
|
|
|
702 sym->getName());
|
|
|
703
|
|
|
704 InputFile *obj = createObjectFile(mb, getName());
|
|
|
705 symtab->addFile(obj);
|
|
|
706 }
|
|
|
707
|
|
|
708 static uint8_t mapVisibility(GlobalValue::VisibilityTypes gvVisibility) {
|
|
|
709 switch (gvVisibility) {
|
|
|
710 case GlobalValue::DefaultVisibility:
|
|
|
711 return WASM_SYMBOL_VISIBILITY_DEFAULT;
|
|
|
712 case GlobalValue::HiddenVisibility:
|
|
|
713 case GlobalValue::ProtectedVisibility:
|
|
|
714 return WASM_SYMBOL_VISIBILITY_HIDDEN;
|
|
|
715 }
|
|
|
716 llvm_unreachable("unknown visibility");
|
|
|
717 }
|
|
|
718
|
|
|
719 static Symbol *createBitcodeSymbol(const std::vector<bool> &keptComdats,
|
|
|
720 const lto::InputFile::Symbol &objSym,
|
|
|
721 BitcodeFile &f) {
|
|
|
722 StringRef name = saver.save(objSym.getName());
|
|
|
723
|
|
|
724 uint32_t flags = objSym.isWeak() ? WASM_SYMBOL_BINDING_WEAK : 0;
|
|
|
725 flags |= mapVisibility(objSym.getVisibility());
|
|
|
726
|
|
|
727 int c = objSym.getComdatIndex();
|
|
|
728 bool excludedByComdat = c != -1 && !keptComdats[c];
|
|
|
729
|
|
|
730 if (objSym.isUndefined() || excludedByComdat) {
|
|
|
731 flags |= WASM_SYMBOL_UNDEFINED;
|
|
|
732 if (objSym.isExecutable())
|
|
173
|
733 return symtab->addUndefinedFunction(name, None, None, flags, &f, nullptr,
|
|
150
|
734 true);
|
|
|
735 return symtab->addUndefinedData(name, flags, &f);
|
|
|
736 }
|
|
|
737
|
|
|
738 if (objSym.isExecutable())
|
|
|
739 return symtab->addDefinedFunction(name, flags, &f, nullptr);
|
|
|
740 return symtab->addDefinedData(name, flags, &f, nullptr, 0, 0);
|
|
|
741 }
|
|
|
742
|
|
|
743 bool BitcodeFile::doneLTO = false;
|
|
|
744
|
|
|
745 void BitcodeFile::parse() {
|
|
|
746 if (doneLTO) {
|
|
207
|
747 error(toString(this) + ": attempt to add bitcode file after LTO.");
|
|
150
|
748 return;
|
|
|
749 }
|
|
|
750
|
|
|
751 obj = check(lto::InputFile::create(MemoryBufferRef(
|
|
|
752 mb.getBuffer(), saver.save(archiveName + mb.getBufferIdentifier()))));
|
|
|
753 Triple t(obj->getTargetTriple());
|
|
207
|
754 if (!t.isWasm()) {
|
|
|
755 error(toString(this) + ": machine type must be wasm32 or wasm64");
|
|
150
|
756 return;
|
|
|
757 }
|
|
207
|
758 checkArch(t.getArch());
|
|
150
|
759 std::vector<bool> keptComdats;
|
|
|
760 for (StringRef s : obj->getComdatTable())
|
|
|
761 keptComdats.push_back(symtab->addComdat(s));
|
|
|
762
|
|
|
763 for (const lto::InputFile::Symbol &objSym : obj->symbols())
|
|
|
764 symbols.push_back(createBitcodeSymbol(keptComdats, objSym, *this));
|
|
|
765 }
|
|
|
766
|
|
|
767 } // namespace wasm
|
|
|
768 } // namespace lld
|
