1 //===- Writer.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 "Writer.h" 10 #include "Config.h" 11 #include "InputChunks.h" 12 #include "InputElement.h" 13 #include "MapFile.h" 14 #include "OutputSections.h" 15 #include "OutputSegment.h" 16 #include "Relocations.h" 17 #include "SymbolTable.h" 18 #include "SyntheticSections.h" 19 #include "WriterUtils.h" 20 #include "lld/Common/ErrorHandler.h" 21 #include "lld/Common/Memory.h" 22 #include "lld/Common/Strings.h" 23 #include "llvm/ADT/DenseSet.h" 24 #include "llvm/ADT/SmallSet.h" 25 #include "llvm/ADT/SmallVector.h" 26 #include "llvm/ADT/StringMap.h" 27 #include "llvm/BinaryFormat/Wasm.h" 28 #include "llvm/BinaryFormat/WasmTraits.h" 29 #include "llvm/Support/FileOutputBuffer.h" 30 #include "llvm/Support/Format.h" 31 #include "llvm/Support/FormatVariadic.h" 32 #include "llvm/Support/LEB128.h" 33 #include "llvm/Support/Parallel.h" 34 35 #include <cstdarg> 36 #include <map> 37 38 #define DEBUG_TYPE "lld" 39 40 using namespace llvm; 41 using namespace llvm::wasm; 42 43 namespace lld { 44 namespace wasm { 45 static constexpr int stackAlignment = 16; 46 static constexpr int heapAlignment = 16; 47 48 namespace { 49 50 // The writer writes a SymbolTable result to a file. 51 class Writer { 52 public: 53 void run(); 54 55 private: 56 void openFile(); 57 58 bool needsPassiveInitialization(const OutputSegment *segment); 59 bool hasPassiveInitializedSegments(); 60 61 void createSyntheticInitFunctions(); 62 void createInitMemoryFunction(); 63 void createStartFunction(); 64 void createApplyDataRelocationsFunction(); 65 void createApplyGlobalRelocationsFunction(); 66 void createCallCtorsFunction(); 67 void createInitTLSFunction(); 68 void createCommandExportWrappers(); 69 void createCommandExportWrapper(uint32_t functionIndex, DefinedFunction *f); 70 71 void assignIndexes(); 72 void populateSymtab(); 73 void populateProducers(); 74 void populateTargetFeatures(); 75 void calculateInitFunctions(); 76 void calculateImports(); 77 void calculateExports(); 78 void calculateCustomSections(); 79 void calculateTypes(); 80 void createOutputSegments(); 81 OutputSegment *createOutputSegment(StringRef name); 82 void combineOutputSegments(); 83 void layoutMemory(); 84 void createHeader(); 85 86 void addSection(OutputSection *sec); 87 88 void addSections(); 89 90 void createCustomSections(); 91 void createSyntheticSections(); 92 void createSyntheticSectionsPostLayout(); 93 void finalizeSections(); 94 95 // Custom sections 96 void createRelocSections(); 97 98 void writeHeader(); 99 void writeSections(); 100 101 uint64_t fileSize = 0; 102 103 std::vector<WasmInitEntry> initFunctions; 104 llvm::StringMap<std::vector<InputChunk *>> customSectionMapping; 105 106 // Stable storage for command export wrapper function name strings. 107 std::list<std::string> commandExportWrapperNames; 108 109 // Elements that are used to construct the final output 110 std::string header; 111 std::vector<OutputSection *> outputSections; 112 113 std::unique_ptr<FileOutputBuffer> buffer; 114 115 std::vector<OutputSegment *> segments; 116 llvm::SmallDenseMap<StringRef, OutputSegment *> segmentMap; 117 }; 118 119 } // anonymous namespace 120 121 void Writer::calculateCustomSections() { 122 log("calculateCustomSections"); 123 bool stripDebug = config->stripDebug || config->stripAll; 124 for (ObjFile *file : symtab->objectFiles) { 125 for (InputChunk *section : file->customSections) { 126 // Exclude COMDAT sections that are not selected for inclusion 127 if (section->discarded) 128 continue; 129 StringRef name = section->getName(); 130 // These custom sections are known the linker and synthesized rather than 131 // blindly copied. 132 if (name == "linking" || name == "name" || name == "producers" || 133 name == "target_features" || name.startswith("reloc.")) 134 continue; 135 // These custom sections are generated by `clang -fembed-bitcode`. 136 // These are used by the rust toolchain to ship LTO data along with 137 // compiled object code, but they don't want this included in the linker 138 // output. 139 if (name == ".llvmbc" || name == ".llvmcmd") 140 continue; 141 // Strip debug section in that option was specified. 142 if (stripDebug && name.startswith(".debug_")) 143 continue; 144 // Otherwise include custom sections by default and concatenate their 145 // contents. 146 customSectionMapping[name].push_back(section); 147 } 148 } 149 } 150 151 void Writer::createCustomSections() { 152 log("createCustomSections"); 153 for (auto &pair : customSectionMapping) { 154 StringRef name = pair.first(); 155 LLVM_DEBUG(dbgs() << "createCustomSection: " << name << "\n"); 156 157 OutputSection *sec = make<CustomSection>(std::string(name), pair.second); 158 if (config->relocatable || config->emitRelocs) { 159 auto *sym = make<OutputSectionSymbol>(sec); 160 out.linkingSec->addToSymtab(sym); 161 sec->sectionSym = sym; 162 } 163 addSection(sec); 164 } 165 } 166 167 // Create relocations sections in the final output. 168 // These are only created when relocatable output is requested. 169 void Writer::createRelocSections() { 170 log("createRelocSections"); 171 // Don't use iterator here since we are adding to OutputSection 172 size_t origSize = outputSections.size(); 173 for (size_t i = 0; i < origSize; i++) { 174 LLVM_DEBUG(dbgs() << "check section " << i << "\n"); 175 OutputSection *sec = outputSections[i]; 176 177 // Count the number of needed sections. 178 uint32_t count = sec->getNumRelocations(); 179 if (!count) 180 continue; 181 182 StringRef name; 183 if (sec->type == WASM_SEC_DATA) 184 name = "reloc.DATA"; 185 else if (sec->type == WASM_SEC_CODE) 186 name = "reloc.CODE"; 187 else if (sec->type == WASM_SEC_CUSTOM) 188 name = saver.save("reloc." + sec->name); 189 else 190 llvm_unreachable( 191 "relocations only supported for code, data, or custom sections"); 192 193 addSection(make<RelocSection>(name, sec)); 194 } 195 } 196 197 void Writer::populateProducers() { 198 for (ObjFile *file : symtab->objectFiles) { 199 const WasmProducerInfo &info = file->getWasmObj()->getProducerInfo(); 200 out.producersSec->addInfo(info); 201 } 202 } 203 204 void Writer::writeHeader() { 205 memcpy(buffer->getBufferStart(), header.data(), header.size()); 206 } 207 208 void Writer::writeSections() { 209 uint8_t *buf = buffer->getBufferStart(); 210 parallelForEach(outputSections, [buf](OutputSection *s) { 211 assert(s->isNeeded()); 212 s->writeTo(buf); 213 }); 214 } 215 216 static void setGlobalPtr(DefinedGlobal *g, uint64_t memoryPtr) { 217 g->global->setPointerValue(memoryPtr); 218 } 219 220 // Fix the memory layout of the output binary. This assigns memory offsets 221 // to each of the input data sections as well as the explicit stack region. 222 // The default memory layout is as follows, from low to high. 223 // 224 // - initialized data (starting at Config->globalBase) 225 // - BSS data (not currently implemented in llvm) 226 // - explicit stack (Config->ZStackSize) 227 // - heap start / unallocated 228 // 229 // The --stack-first option means that stack is placed before any static data. 230 // This can be useful since it means that stack overflow traps immediately 231 // rather than overwriting global data, but also increases code size since all 232 // static data loads and stores requires larger offsets. 233 void Writer::layoutMemory() { 234 uint64_t memoryPtr = 0; 235 236 auto placeStack = [&]() { 237 if (config->relocatable || config->isPic) 238 return; 239 memoryPtr = alignTo(memoryPtr, stackAlignment); 240 if (config->zStackSize != alignTo(config->zStackSize, stackAlignment)) 241 error("stack size must be " + Twine(stackAlignment) + "-byte aligned"); 242 log("mem: stack size = " + Twine(config->zStackSize)); 243 log("mem: stack base = " + Twine(memoryPtr)); 244 memoryPtr += config->zStackSize; 245 setGlobalPtr(cast<DefinedGlobal>(WasmSym::stackPointer), memoryPtr); 246 log("mem: stack top = " + Twine(memoryPtr)); 247 }; 248 249 if (config->stackFirst) { 250 placeStack(); 251 } else { 252 memoryPtr = config->globalBase; 253 log("mem: global base = " + Twine(config->globalBase)); 254 } 255 256 if (WasmSym::globalBase) 257 WasmSym::globalBase->setVA(memoryPtr); 258 259 uint64_t dataStart = memoryPtr; 260 261 // Arbitrarily set __dso_handle handle to point to the start of the data 262 // segments. 263 if (WasmSym::dsoHandle) 264 WasmSym::dsoHandle->setVA(dataStart); 265 266 out.dylinkSec->memAlign = 0; 267 for (OutputSegment *seg : segments) { 268 out.dylinkSec->memAlign = std::max(out.dylinkSec->memAlign, seg->alignment); 269 memoryPtr = alignTo(memoryPtr, 1ULL << seg->alignment); 270 seg->startVA = memoryPtr; 271 log(formatv("mem: {0,-15} offset={1,-8} size={2,-8} align={3}", seg->name, 272 memoryPtr, seg->size, seg->alignment)); 273 274 if (!config->relocatable && seg->isTLS()) { 275 if (config->sharedMemory) { 276 auto *tlsSize = cast<DefinedGlobal>(WasmSym::tlsSize); 277 setGlobalPtr(tlsSize, seg->size); 278 279 auto *tlsAlign = cast<DefinedGlobal>(WasmSym::tlsAlign); 280 setGlobalPtr(tlsAlign, int64_t{1} << seg->alignment); 281 } else { 282 auto *tlsBase = cast<DefinedGlobal>(WasmSym::tlsBase); 283 setGlobalPtr(tlsBase, memoryPtr); 284 } 285 } 286 287 memoryPtr += seg->size; 288 } 289 290 // Make space for the memory initialization flag 291 if (config->sharedMemory && hasPassiveInitializedSegments()) { 292 memoryPtr = alignTo(memoryPtr, 4); 293 WasmSym::initMemoryFlag = symtab->addSyntheticDataSymbol( 294 "__wasm_init_memory_flag", WASM_SYMBOL_VISIBILITY_HIDDEN); 295 WasmSym::initMemoryFlag->markLive(); 296 WasmSym::initMemoryFlag->setVA(memoryPtr); 297 log(formatv("mem: {0,-15} offset={1,-8} size={2,-8} align={3}", 298 "__wasm_init_memory_flag", memoryPtr, 4, 4)); 299 memoryPtr += 4; 300 } 301 302 if (WasmSym::dataEnd) 303 WasmSym::dataEnd->setVA(memoryPtr); 304 305 uint64_t staticDataSize = memoryPtr - dataStart; 306 log("mem: static data = " + Twine(staticDataSize)); 307 if (config->isPic) 308 out.dylinkSec->memSize = staticDataSize; 309 310 if (!config->stackFirst) 311 placeStack(); 312 313 if (WasmSym::heapBase) { 314 // Set `__heap_base` to follow the end of the stack or global data. The 315 // fact that this comes last means that a malloc/brk implementation can 316 // grow the heap at runtime. 317 // We'll align the heap base here because memory allocators might expect 318 // __heap_base to be aligned already. 319 memoryPtr = alignTo(memoryPtr, heapAlignment); 320 log("mem: heap base = " + Twine(memoryPtr)); 321 WasmSym::heapBase->setVA(memoryPtr); 322 } 323 324 uint64_t maxMemorySetting = 1ULL 325 << (config->is64.getValueOr(false) ? 48 : 32); 326 327 if (config->initialMemory != 0) { 328 if (config->initialMemory != alignTo(config->initialMemory, WasmPageSize)) 329 error("initial memory must be " + Twine(WasmPageSize) + "-byte aligned"); 330 if (memoryPtr > config->initialMemory) 331 error("initial memory too small, " + Twine(memoryPtr) + " bytes needed"); 332 if (config->initialMemory > maxMemorySetting) 333 error("initial memory too large, cannot be greater than " + 334 Twine(maxMemorySetting)); 335 memoryPtr = config->initialMemory; 336 } 337 out.memorySec->numMemoryPages = 338 alignTo(memoryPtr, WasmPageSize) / WasmPageSize; 339 log("mem: total pages = " + Twine(out.memorySec->numMemoryPages)); 340 341 if (config->maxMemory != 0) { 342 if (config->maxMemory != alignTo(config->maxMemory, WasmPageSize)) 343 error("maximum memory must be " + Twine(WasmPageSize) + "-byte aligned"); 344 if (memoryPtr > config->maxMemory) 345 error("maximum memory too small, " + Twine(memoryPtr) + " bytes needed"); 346 if (config->maxMemory > maxMemorySetting) 347 error("maximum memory too large, cannot be greater than " + 348 Twine(maxMemorySetting)); 349 } 350 351 // Check max if explicitly supplied or required by shared memory 352 if (config->maxMemory != 0 || config->sharedMemory) { 353 uint64_t max = config->maxMemory; 354 if (max == 0) { 355 // If no maxMemory config was supplied but we are building with 356 // shared memory, we need to pick a sensible upper limit. 357 if (config->isPic) 358 max = maxMemorySetting; 359 else 360 max = alignTo(memoryPtr, WasmPageSize); 361 } 362 out.memorySec->maxMemoryPages = max / WasmPageSize; 363 log("mem: max pages = " + Twine(out.memorySec->maxMemoryPages)); 364 } 365 } 366 367 void Writer::addSection(OutputSection *sec) { 368 if (!sec->isNeeded()) 369 return; 370 log("addSection: " + toString(*sec)); 371 sec->sectionIndex = outputSections.size(); 372 outputSections.push_back(sec); 373 } 374 375 // If a section name is valid as a C identifier (which is rare because of 376 // the leading '.'), linkers are expected to define __start_<secname> and 377 // __stop_<secname> symbols. They are at beginning and end of the section, 378 // respectively. This is not requested by the ELF standard, but GNU ld and 379 // gold provide the feature, and used by many programs. 380 static void addStartStopSymbols(const OutputSegment *seg) { 381 StringRef name = seg->name; 382 if (!isValidCIdentifier(name)) 383 return; 384 LLVM_DEBUG(dbgs() << "addStartStopSymbols: " << name << "\n"); 385 uint64_t start = seg->startVA; 386 uint64_t stop = start + seg->size; 387 symtab->addOptionalDataSymbol(saver.save("__start_" + name), start); 388 symtab->addOptionalDataSymbol(saver.save("__stop_" + name), stop); 389 } 390 391 void Writer::addSections() { 392 addSection(out.dylinkSec); 393 addSection(out.typeSec); 394 addSection(out.importSec); 395 addSection(out.functionSec); 396 addSection(out.tableSec); 397 addSection(out.memorySec); 398 addSection(out.tagSec); 399 addSection(out.globalSec); 400 addSection(out.exportSec); 401 addSection(out.startSec); 402 addSection(out.elemSec); 403 addSection(out.dataCountSec); 404 405 addSection(make<CodeSection>(out.functionSec->inputFunctions)); 406 addSection(make<DataSection>(segments)); 407 408 createCustomSections(); 409 410 addSection(out.linkingSec); 411 if (config->emitRelocs || config->relocatable) { 412 createRelocSections(); 413 } 414 415 addSection(out.nameSec); 416 addSection(out.producersSec); 417 addSection(out.targetFeaturesSec); 418 } 419 420 void Writer::finalizeSections() { 421 for (OutputSection *s : outputSections) { 422 s->setOffset(fileSize); 423 s->finalizeContents(); 424 fileSize += s->getSize(); 425 } 426 } 427 428 void Writer::populateTargetFeatures() { 429 StringMap<std::string> used; 430 StringMap<std::string> required; 431 StringMap<std::string> disallowed; 432 SmallSet<std::string, 8> &allowed = out.targetFeaturesSec->features; 433 bool tlsUsed = false; 434 435 // Only infer used features if user did not specify features 436 bool inferFeatures = !config->features.hasValue(); 437 438 if (!inferFeatures) { 439 auto &explicitFeatures = config->features.getValue(); 440 allowed.insert(explicitFeatures.begin(), explicitFeatures.end()); 441 if (!config->checkFeatures) 442 return; 443 } 444 445 // Find the sets of used, required, and disallowed features 446 for (ObjFile *file : symtab->objectFiles) { 447 StringRef fileName(file->getName()); 448 for (auto &feature : file->getWasmObj()->getTargetFeatures()) { 449 switch (feature.Prefix) { 450 case WASM_FEATURE_PREFIX_USED: 451 used.insert({feature.Name, std::string(fileName)}); 452 break; 453 case WASM_FEATURE_PREFIX_REQUIRED: 454 used.insert({feature.Name, std::string(fileName)}); 455 required.insert({feature.Name, std::string(fileName)}); 456 break; 457 case WASM_FEATURE_PREFIX_DISALLOWED: 458 disallowed.insert({feature.Name, std::string(fileName)}); 459 break; 460 default: 461 error("Unrecognized feature policy prefix " + 462 std::to_string(feature.Prefix)); 463 } 464 } 465 466 // Find TLS data segments 467 auto isTLS = [](InputChunk *segment) { 468 return segment->live && segment->isTLS(); 469 }; 470 tlsUsed = tlsUsed || 471 std::any_of(file->segments.begin(), file->segments.end(), isTLS); 472 } 473 474 if (inferFeatures) 475 for (const auto &key : used.keys()) 476 allowed.insert(std::string(key)); 477 478 if (!config->checkFeatures) 479 return; 480 481 if (!config->relocatable && allowed.count("mutable-globals") == 0) { 482 for (const Symbol *sym : out.importSec->importedSymbols) { 483 if (auto *global = dyn_cast<GlobalSymbol>(sym)) { 484 if (global->getGlobalType()->Mutable) { 485 error(Twine("mutable global imported but 'mutable-globals' feature " 486 "not present in inputs: `") + 487 toString(*sym) + "`. Use --no-check-features to suppress."); 488 } 489 } 490 } 491 for (const Symbol *sym : out.exportSec->exportedSymbols) { 492 if (isa<GlobalSymbol>(sym)) { 493 error(Twine("mutable global exported but 'mutable-globals' feature " 494 "not present in inputs: `") + 495 toString(*sym) + "`. Use --no-check-features to suppress."); 496 } 497 } 498 } 499 500 if (config->sharedMemory) { 501 if (disallowed.count("shared-mem")) 502 error("--shared-memory is disallowed by " + disallowed["shared-mem"] + 503 " because it was not compiled with 'atomics' or 'bulk-memory' " 504 "features."); 505 506 for (auto feature : {"atomics", "bulk-memory"}) 507 if (!allowed.count(feature)) 508 error(StringRef("'") + feature + 509 "' feature must be used in order to use shared memory"); 510 } 511 512 if (tlsUsed) { 513 for (auto feature : {"atomics", "bulk-memory"}) 514 if (!allowed.count(feature)) 515 error(StringRef("'") + feature + 516 "' feature must be used in order to use thread-local storage"); 517 } 518 519 // Validate that used features are allowed in output 520 if (!inferFeatures) { 521 for (auto &feature : used.keys()) { 522 if (!allowed.count(std::string(feature))) 523 error(Twine("Target feature '") + feature + "' used by " + 524 used[feature] + " is not allowed."); 525 } 526 } 527 528 // Validate the required and disallowed constraints for each file 529 for (ObjFile *file : symtab->objectFiles) { 530 StringRef fileName(file->getName()); 531 SmallSet<std::string, 8> objectFeatures; 532 for (auto &feature : file->getWasmObj()->getTargetFeatures()) { 533 if (feature.Prefix == WASM_FEATURE_PREFIX_DISALLOWED) 534 continue; 535 objectFeatures.insert(feature.Name); 536 if (disallowed.count(feature.Name)) 537 error(Twine("Target feature '") + feature.Name + "' used in " + 538 fileName + " is disallowed by " + disallowed[feature.Name] + 539 ". Use --no-check-features to suppress."); 540 } 541 for (auto &feature : required.keys()) { 542 if (!objectFeatures.count(std::string(feature))) 543 error(Twine("Missing target feature '") + feature + "' in " + fileName + 544 ", required by " + required[feature] + 545 ". Use --no-check-features to suppress."); 546 } 547 } 548 } 549 550 static bool shouldImport(Symbol *sym) { 551 if (!sym->isUndefined()) 552 return false; 553 if (sym->isWeak() && !config->relocatable && !config->isPic) 554 return false; 555 if (!sym->isLive()) 556 return false; 557 if (!sym->isUsedInRegularObj) 558 return false; 559 560 // We don't generate imports for data symbols. They however can be imported 561 // as GOT entries. 562 if (isa<DataSymbol>(sym)) 563 return false; 564 565 if (config->isPic || config->relocatable || config->importUndefined) 566 return true; 567 if (config->allowUndefinedSymbols.count(sym->getName()) != 0) 568 return true; 569 570 return sym->importName.hasValue(); 571 } 572 573 void Writer::calculateImports() { 574 // Some inputs require that the indirect function table be assigned to table 575 // number 0, so if it is present and is an import, allocate it before any 576 // other tables. 577 if (WasmSym::indirectFunctionTable && 578 shouldImport(WasmSym::indirectFunctionTable)) 579 out.importSec->addImport(WasmSym::indirectFunctionTable); 580 581 for (Symbol *sym : symtab->getSymbols()) { 582 if (!shouldImport(sym)) 583 continue; 584 if (sym == WasmSym::indirectFunctionTable) 585 continue; 586 LLVM_DEBUG(dbgs() << "import: " << sym->getName() << "\n"); 587 out.importSec->addImport(sym); 588 } 589 } 590 591 void Writer::calculateExports() { 592 if (config->relocatable) 593 return; 594 595 if (!config->relocatable && !config->importMemory) 596 out.exportSec->exports.push_back( 597 WasmExport{"memory", WASM_EXTERNAL_MEMORY, 0}); 598 599 unsigned globalIndex = 600 out.importSec->getNumImportedGlobals() + out.globalSec->numGlobals(); 601 602 for (Symbol *sym : symtab->getSymbols()) { 603 if (!sym->isExported()) 604 continue; 605 if (!sym->isLive()) 606 continue; 607 608 StringRef name = sym->getName(); 609 WasmExport export_; 610 if (auto *f = dyn_cast<DefinedFunction>(sym)) { 611 if (Optional<StringRef> exportName = f->function->getExportName()) { 612 name = *exportName; 613 } 614 export_ = {name, WASM_EXTERNAL_FUNCTION, f->getFunctionIndex()}; 615 } else if (auto *g = dyn_cast<DefinedGlobal>(sym)) { 616 if (g->getGlobalType()->Mutable && !g->getFile() && !g->forceExport) { 617 // Avoid exporting mutable globals are linker synthesized (e.g. 618 // __stack_pointer or __tls_base) unless they are explicitly exported 619 // from the command line. 620 // Without this check `--export-all` would cause any program using the 621 // stack pointer to export a mutable global even if none of the input 622 // files were built with the `mutable-globals` feature. 623 continue; 624 } 625 export_ = {name, WASM_EXTERNAL_GLOBAL, g->getGlobalIndex()}; 626 } else if (auto *t = dyn_cast<DefinedTag>(sym)) { 627 export_ = {name, WASM_EXTERNAL_TAG, t->getTagIndex()}; 628 } else if (auto *d = dyn_cast<DefinedData>(sym)) { 629 if (d->segment && d->segment->isTLS()) { 630 // We can't currenly export TLS data symbols. 631 if (sym->isExportedExplicit()) 632 error("TLS symbols cannot yet be exported: `" + toString(*sym) + "`"); 633 continue; 634 } 635 out.globalSec->dataAddressGlobals.push_back(d); 636 export_ = {name, WASM_EXTERNAL_GLOBAL, globalIndex++}; 637 } else { 638 auto *t = cast<DefinedTable>(sym); 639 export_ = {name, WASM_EXTERNAL_TABLE, t->getTableNumber()}; 640 } 641 642 LLVM_DEBUG(dbgs() << "Export: " << name << "\n"); 643 out.exportSec->exports.push_back(export_); 644 out.exportSec->exportedSymbols.push_back(sym); 645 } 646 } 647 648 void Writer::populateSymtab() { 649 if (!config->relocatable && !config->emitRelocs) 650 return; 651 652 for (Symbol *sym : symtab->getSymbols()) 653 if (sym->isUsedInRegularObj && sym->isLive()) 654 out.linkingSec->addToSymtab(sym); 655 656 for (ObjFile *file : symtab->objectFiles) { 657 LLVM_DEBUG(dbgs() << "Local symtab entries: " << file->getName() << "\n"); 658 for (Symbol *sym : file->getSymbols()) 659 if (sym->isLocal() && !isa<SectionSymbol>(sym) && sym->isLive()) 660 out.linkingSec->addToSymtab(sym); 661 } 662 } 663 664 void Writer::calculateTypes() { 665 // The output type section is the union of the following sets: 666 // 1. Any signature used in the TYPE relocation 667 // 2. The signatures of all imported functions 668 // 3. The signatures of all defined functions 669 // 4. The signatures of all imported tags 670 // 5. The signatures of all defined tags 671 672 for (ObjFile *file : symtab->objectFiles) { 673 ArrayRef<WasmSignature> types = file->getWasmObj()->types(); 674 for (uint32_t i = 0; i < types.size(); i++) 675 if (file->typeIsUsed[i]) 676 file->typeMap[i] = out.typeSec->registerType(types[i]); 677 } 678 679 for (const Symbol *sym : out.importSec->importedSymbols) { 680 if (auto *f = dyn_cast<FunctionSymbol>(sym)) 681 out.typeSec->registerType(*f->signature); 682 else if (auto *t = dyn_cast<TagSymbol>(sym)) 683 out.typeSec->registerType(*t->signature); 684 } 685 686 for (const InputFunction *f : out.functionSec->inputFunctions) 687 out.typeSec->registerType(f->signature); 688 689 for (const InputTag *t : out.tagSec->inputTags) 690 out.typeSec->registerType(t->signature); 691 } 692 693 // In a command-style link, create a wrapper for each exported symbol 694 // which calls the constructors and destructors. 695 void Writer::createCommandExportWrappers() { 696 // This logic doesn't currently support Emscripten-style PIC mode. 697 assert(!config->isPic); 698 699 // If there are no ctors and there's no libc `__wasm_call_dtors` to 700 // call, don't wrap the exports. 701 if (initFunctions.empty() && WasmSym::callDtors == NULL) 702 return; 703 704 std::vector<DefinedFunction *> toWrap; 705 706 for (Symbol *sym : symtab->getSymbols()) 707 if (sym->isExported()) 708 if (auto *f = dyn_cast<DefinedFunction>(sym)) 709 toWrap.push_back(f); 710 711 for (auto *f : toWrap) { 712 auto funcNameStr = (f->getName() + ".command_export").str(); 713 commandExportWrapperNames.push_back(funcNameStr); 714 const std::string &funcName = commandExportWrapperNames.back(); 715 716 auto func = make<SyntheticFunction>(*f->getSignature(), funcName); 717 if (f->function->getExportName().hasValue()) 718 func->setExportName(f->function->getExportName()->str()); 719 else 720 func->setExportName(f->getName().str()); 721 722 DefinedFunction *def = 723 symtab->addSyntheticFunction(funcName, f->flags, func); 724 def->markLive(); 725 726 def->flags |= WASM_SYMBOL_EXPORTED; 727 def->flags &= ~WASM_SYMBOL_VISIBILITY_HIDDEN; 728 def->forceExport = f->forceExport; 729 730 f->flags |= WASM_SYMBOL_VISIBILITY_HIDDEN; 731 f->flags &= ~WASM_SYMBOL_EXPORTED; 732 f->forceExport = false; 733 734 out.functionSec->addFunction(func); 735 736 createCommandExportWrapper(f->getFunctionIndex(), def); 737 } 738 } 739 740 static void finalizeIndirectFunctionTable() { 741 if (!WasmSym::indirectFunctionTable) 742 return; 743 744 if (shouldImport(WasmSym::indirectFunctionTable) && 745 !WasmSym::indirectFunctionTable->hasTableNumber()) { 746 // Processing -Bsymbolic relocations resulted in a late requirement that the 747 // indirect function table be present, and we are running in --import-table 748 // mode. Add the table now to the imports section. Otherwise it will be 749 // added to the tables section later in assignIndexes. 750 out.importSec->addImport(WasmSym::indirectFunctionTable); 751 } 752 753 uint32_t tableSize = config->tableBase + out.elemSec->numEntries(); 754 WasmLimits limits = {0, tableSize, 0}; 755 if (WasmSym::indirectFunctionTable->isDefined() && !config->growableTable) { 756 limits.Flags |= WASM_LIMITS_FLAG_HAS_MAX; 757 limits.Maximum = limits.Minimum; 758 } 759 WasmSym::indirectFunctionTable->setLimits(limits); 760 } 761 762 static void scanRelocations() { 763 for (ObjFile *file : symtab->objectFiles) { 764 LLVM_DEBUG(dbgs() << "scanRelocations: " << file->getName() << "\n"); 765 for (InputChunk *chunk : file->functions) 766 scanRelocations(chunk); 767 for (InputChunk *chunk : file->segments) 768 scanRelocations(chunk); 769 for (auto &p : file->customSections) 770 scanRelocations(p); 771 } 772 } 773 774 void Writer::assignIndexes() { 775 // Seal the import section, since other index spaces such as function and 776 // global are effected by the number of imports. 777 out.importSec->seal(); 778 779 for (InputFunction *func : symtab->syntheticFunctions) 780 out.functionSec->addFunction(func); 781 782 for (ObjFile *file : symtab->objectFiles) { 783 LLVM_DEBUG(dbgs() << "Functions: " << file->getName() << "\n"); 784 for (InputFunction *func : file->functions) 785 out.functionSec->addFunction(func); 786 } 787 788 for (InputGlobal *global : symtab->syntheticGlobals) 789 out.globalSec->addGlobal(global); 790 791 for (ObjFile *file : symtab->objectFiles) { 792 LLVM_DEBUG(dbgs() << "Globals: " << file->getName() << "\n"); 793 for (InputGlobal *global : file->globals) 794 out.globalSec->addGlobal(global); 795 } 796 797 for (ObjFile *file : symtab->objectFiles) { 798 LLVM_DEBUG(dbgs() << "Tags: " << file->getName() << "\n"); 799 for (InputTag *tag : file->tags) 800 out.tagSec->addTag(tag); 801 } 802 803 for (ObjFile *file : symtab->objectFiles) { 804 LLVM_DEBUG(dbgs() << "Tables: " << file->getName() << "\n"); 805 for (InputTable *table : file->tables) 806 out.tableSec->addTable(table); 807 } 808 809 for (InputTable *table : symtab->syntheticTables) 810 out.tableSec->addTable(table); 811 812 out.globalSec->assignIndexes(); 813 out.tableSec->assignIndexes(); 814 } 815 816 static StringRef getOutputDataSegmentName(const InputChunk &seg) { 817 // We always merge .tbss and .tdata into a single TLS segment so all TLS 818 // symbols are be relative to single __tls_base. 819 if (seg.isTLS()) 820 return ".tdata"; 821 StringRef name = seg.getName(); 822 if (!config->mergeDataSegments) 823 return name; 824 if (name.startswith(".text.")) 825 return ".text"; 826 if (name.startswith(".data.")) 827 return ".data"; 828 if (name.startswith(".bss.")) 829 return ".bss"; 830 if (name.startswith(".rodata.")) 831 return ".rodata"; 832 return name; 833 } 834 835 OutputSegment *Writer::createOutputSegment(StringRef name) { 836 LLVM_DEBUG(dbgs() << "new segment: " << name << "\n"); 837 OutputSegment *s = make<OutputSegment>(name); 838 if (config->sharedMemory) 839 s->initFlags = WASM_DATA_SEGMENT_IS_PASSIVE; 840 // Exported memories are guaranteed to be zero-initialized, so no need 841 // to emit data segments for bss sections. 842 // TODO: consider initializing bss sections with memory.fill 843 // instructions when memory is imported and bulk-memory is available. 844 if (!config->importMemory && !config->relocatable && name.startswith(".bss")) 845 s->isBss = true; 846 segments.push_back(s); 847 return s; 848 } 849 850 void Writer::createOutputSegments() { 851 for (ObjFile *file : symtab->objectFiles) { 852 for (InputChunk *segment : file->segments) { 853 if (!segment->live) 854 continue; 855 StringRef name = getOutputDataSegmentName(*segment); 856 OutputSegment *s = nullptr; 857 // When running in relocatable mode we can't merge segments that are part 858 // of comdat groups since the ultimate linker needs to be able exclude or 859 // include them individually. 860 if (config->relocatable && !segment->getComdatName().empty()) { 861 s = createOutputSegment(name); 862 } else { 863 if (segmentMap.count(name) == 0) 864 segmentMap[name] = createOutputSegment(name); 865 s = segmentMap[name]; 866 } 867 s->addInputSegment(segment); 868 } 869 } 870 871 // Sort segments by type, placing .bss last 872 std::stable_sort(segments.begin(), segments.end(), 873 [](const OutputSegment *a, const OutputSegment *b) { 874 auto order = [](StringRef name) { 875 return StringSwitch<int>(name) 876 .StartsWith(".tdata", 0) 877 .StartsWith(".rodata", 1) 878 .StartsWith(".data", 2) 879 .StartsWith(".bss", 4) 880 .Default(3); 881 }; 882 return order(a->name) < order(b->name); 883 }); 884 885 for (size_t i = 0; i < segments.size(); ++i) 886 segments[i]->index = i; 887 888 // Merge MergeInputSections into a single MergeSyntheticSection. 889 LLVM_DEBUG(dbgs() << "-- finalize input semgments\n"); 890 for (OutputSegment *seg : segments) 891 seg->finalizeInputSegments(); 892 } 893 894 void Writer::combineOutputSegments() { 895 // With PIC code we currently only support a single active data segment since 896 // we only have a single __memory_base to use as our base address. This pass 897 // combines all data segments into a single .data segment. 898 // This restructions can be relaxed once we have extended constant 899 // expressions available: 900 // https://github.com/WebAssembly/extended-const 901 assert(config->isPic && !config->sharedMemory); 902 if (segments.size() <= 1) 903 return; 904 OutputSegment *combined = make<OutputSegment>(".data"); 905 combined->startVA = segments[0]->startVA; 906 for (OutputSegment *s : segments) { 907 bool first = true; 908 for (InputChunk *inSeg : s->inputSegments) { 909 if (first) 910 inSeg->alignment = std::max(inSeg->alignment, s->alignment); 911 first = false; 912 #ifndef NDEBUG 913 uint64_t oldVA = inSeg->getVA(); 914 #endif 915 combined->addInputSegment(inSeg); 916 #ifndef NDEBUG 917 uint64_t newVA = inSeg->getVA(); 918 LLVM_DEBUG(dbgs() << "added input segment. name=" << inSeg->getName() 919 << " oldVA=" << oldVA << " newVA=" << newVA << "\n"); 920 assert(oldVA == newVA); 921 #endif 922 } 923 } 924 925 segments = {combined}; 926 } 927 928 static void createFunction(DefinedFunction *func, StringRef bodyContent) { 929 std::string functionBody; 930 { 931 raw_string_ostream os(functionBody); 932 writeUleb128(os, bodyContent.size(), "function size"); 933 os << bodyContent; 934 } 935 ArrayRef<uint8_t> body = arrayRefFromStringRef(saver.save(functionBody)); 936 cast<SyntheticFunction>(func->function)->setBody(body); 937 } 938 939 bool Writer::needsPassiveInitialization(const OutputSegment *segment) { 940 return segment->initFlags & WASM_DATA_SEGMENT_IS_PASSIVE && 941 !segment->isTLS() && !segment->isBss; 942 } 943 944 bool Writer::hasPassiveInitializedSegments() { 945 return std::find_if(segments.begin(), segments.end(), 946 [this](const OutputSegment *s) { 947 return this->needsPassiveInitialization(s); 948 }) != segments.end(); 949 } 950 951 void Writer::createSyntheticInitFunctions() { 952 if (config->relocatable) 953 return; 954 955 static WasmSignature nullSignature = {{}, {}}; 956 957 // Passive segments are used to avoid memory being reinitialized on each 958 // thread's instantiation. These passive segments are initialized and 959 // dropped in __wasm_init_memory, which is registered as the start function 960 if (config->sharedMemory && hasPassiveInitializedSegments()) { 961 WasmSym::initMemory = symtab->addSyntheticFunction( 962 "__wasm_init_memory", WASM_SYMBOL_VISIBILITY_HIDDEN, 963 make<SyntheticFunction>(nullSignature, "__wasm_init_memory")); 964 WasmSym::initMemory->markLive(); 965 } 966 967 if (config->isPic) { 968 // For PIC code we create synthetic functions that apply relocations. 969 // These get called from __wasm_call_ctors before the user-level 970 // constructors. 971 WasmSym::applyDataRelocs = symtab->addSyntheticFunction( 972 "__wasm_apply_data_relocs", WASM_SYMBOL_VISIBILITY_HIDDEN, 973 make<SyntheticFunction>(nullSignature, "__wasm_apply_data_relocs")); 974 WasmSym::applyDataRelocs->markLive(); 975 976 if (out.globalSec->needsRelocations()) { 977 WasmSym::applyGlobalRelocs = symtab->addSyntheticFunction( 978 "__wasm_apply_global_relocs", WASM_SYMBOL_VISIBILITY_HIDDEN, 979 make<SyntheticFunction>(nullSignature, "__wasm_apply_global_relocs")); 980 WasmSym::applyGlobalRelocs->markLive(); 981 } 982 } 983 984 if (WasmSym::applyGlobalRelocs && WasmSym::initMemory) { 985 WasmSym::startFunction = symtab->addSyntheticFunction( 986 "__wasm_start", WASM_SYMBOL_VISIBILITY_HIDDEN, 987 make<SyntheticFunction>(nullSignature, "__wasm_start")); 988 WasmSym::startFunction->markLive(); 989 } 990 } 991 992 void Writer::createInitMemoryFunction() { 993 LLVM_DEBUG(dbgs() << "createInitMemoryFunction\n"); 994 assert(WasmSym::initMemory); 995 assert(WasmSym::initMemoryFlag); 996 assert(hasPassiveInitializedSegments()); 997 uint64_t flagAddress = WasmSym::initMemoryFlag->getVA(); 998 bool is64 = config->is64.getValueOr(false); 999 std::string bodyContent; 1000 { 1001 raw_string_ostream os(bodyContent); 1002 // Initialize memory in a thread-safe manner. The thread that successfully 1003 // increments the flag from 0 to 1 is is responsible for performing the 1004 // memory initialization. Other threads go sleep on the flag until the 1005 // first thread finishing initializing memory, increments the flag to 2, 1006 // and wakes all the other threads. Once the flag has been set to 2, 1007 // subsequently started threads will skip the sleep. All threads 1008 // unconditionally drop their passive data segments once memory has been 1009 // initialized. The generated code is as follows: 1010 // 1011 // (func $__wasm_init_memory 1012 // (if 1013 // (i32.atomic.rmw.cmpxchg align=2 offset=0 1014 // (i32.const $__init_memory_flag) 1015 // (i32.const 0) 1016 // (i32.const 1) 1017 // ) 1018 // (then 1019 // (drop 1020 // (i32.atomic.wait align=2 offset=0 1021 // (i32.const $__init_memory_flag) 1022 // (i32.const 1) 1023 // (i32.const -1) 1024 // ) 1025 // ) 1026 // ) 1027 // (else 1028 // ( ... initialize data segments ... ) 1029 // (i32.atomic.store align=2 offset=0 1030 // (i32.const $__init_memory_flag) 1031 // (i32.const 2) 1032 // ) 1033 // (drop 1034 // (i32.atomic.notify align=2 offset=0 1035 // (i32.const $__init_memory_flag) 1036 // (i32.const -1u) 1037 // ) 1038 // ) 1039 // ) 1040 // ) 1041 // ( ... drop data segments ... ) 1042 // ) 1043 // 1044 // When we are building with PIC, calculate the flag location using: 1045 // 1046 // (global.get $__memory_base) 1047 // (i32.const $__init_memory_flag) 1048 // (i32.const 1) 1049 1050 // With PIC code we cache the flag address in local 0 1051 if (config->isPic) { 1052 writeUleb128(os, 1, "num local decls"); 1053 writeUleb128(os, 1, "local count"); 1054 writeU8(os, is64 ? WASM_TYPE_I64 : WASM_TYPE_I32, "address type"); 1055 writeU8(os, WASM_OPCODE_GLOBAL_GET, "GLOBAL_GET"); 1056 writeUleb128(os, WasmSym::memoryBase->getGlobalIndex(), "memory_base"); 1057 writePtrConst(os, flagAddress, is64, "flag address"); 1058 writeU8(os, is64 ? WASM_OPCODE_I64_ADD : WASM_OPCODE_I32_ADD, "add"); 1059 writeU8(os, WASM_OPCODE_LOCAL_SET, "local.set"); 1060 writeUleb128(os, 0, "local 0"); 1061 } else { 1062 writeUleb128(os, 0, "num locals"); 1063 } 1064 1065 auto writeGetFlagAddress = [&]() { 1066 if (config->isPic) { 1067 writeU8(os, WASM_OPCODE_LOCAL_GET, "local.get"); 1068 writeUleb128(os, 0, "local 0"); 1069 } else { 1070 writePtrConst(os, flagAddress, is64, "flag address"); 1071 } 1072 }; 1073 1074 // Atomically check whether this is the main thread. 1075 writeGetFlagAddress(); 1076 writeI32Const(os, 0, "expected flag value"); 1077 writeI32Const(os, 1, "flag value"); 1078 writeU8(os, WASM_OPCODE_ATOMICS_PREFIX, "atomics prefix"); 1079 writeUleb128(os, WASM_OPCODE_I32_RMW_CMPXCHG, "i32.atomic.rmw.cmpxchg"); 1080 writeMemArg(os, 2, 0); 1081 writeU8(os, WASM_OPCODE_IF, "IF"); 1082 writeU8(os, WASM_TYPE_NORESULT, "blocktype"); 1083 1084 // Did not increment 0, so wait for main thread to initialize memory 1085 writeGetFlagAddress(); 1086 writeI32Const(os, 1, "expected flag value"); 1087 writeI64Const(os, -1, "timeout"); 1088 1089 writeU8(os, WASM_OPCODE_ATOMICS_PREFIX, "atomics prefix"); 1090 writeUleb128(os, WASM_OPCODE_I32_ATOMIC_WAIT, "i32.atomic.wait"); 1091 writeMemArg(os, 2, 0); 1092 writeU8(os, WASM_OPCODE_DROP, "drop"); 1093 1094 writeU8(os, WASM_OPCODE_ELSE, "ELSE"); 1095 1096 // Did increment 0, so conditionally initialize passive data segments 1097 for (const OutputSegment *s : segments) { 1098 if (needsPassiveInitialization(s)) { 1099 // destination address 1100 writePtrConst(os, s->startVA, is64, "destination address"); 1101 if (config->isPic) { 1102 writeU8(os, WASM_OPCODE_GLOBAL_GET, "GLOBAL_GET"); 1103 writeUleb128(os, WasmSym::memoryBase->getGlobalIndex(), 1104 "memory_base"); 1105 writeU8(os, is64 ? WASM_OPCODE_I64_ADD : WASM_OPCODE_I32_ADD, 1106 "i32.add"); 1107 } 1108 // source segment offset 1109 writeI32Const(os, 0, "segment offset"); 1110 // memory region size 1111 writeI32Const(os, s->size, "memory region size"); 1112 // memory.init instruction 1113 writeU8(os, WASM_OPCODE_MISC_PREFIX, "bulk-memory prefix"); 1114 writeUleb128(os, WASM_OPCODE_MEMORY_INIT, "memory.init"); 1115 writeUleb128(os, s->index, "segment index immediate"); 1116 writeU8(os, 0, "memory index immediate"); 1117 } 1118 } 1119 1120 // Set flag to 2 to mark end of initialization 1121 writeGetFlagAddress(); 1122 writeI32Const(os, 2, "flag value"); 1123 writeU8(os, WASM_OPCODE_ATOMICS_PREFIX, "atomics prefix"); 1124 writeUleb128(os, WASM_OPCODE_I32_ATOMIC_STORE, "i32.atomic.store"); 1125 writeMemArg(os, 2, 0); 1126 1127 // Notify any waiters that memory initialization is complete 1128 writeGetFlagAddress(); 1129 writeI32Const(os, -1, "number of waiters"); 1130 writeU8(os, WASM_OPCODE_ATOMICS_PREFIX, "atomics prefix"); 1131 writeUleb128(os, WASM_OPCODE_ATOMIC_NOTIFY, "atomic.notify"); 1132 writeMemArg(os, 2, 0); 1133 writeU8(os, WASM_OPCODE_DROP, "drop"); 1134 1135 writeU8(os, WASM_OPCODE_END, "END"); 1136 1137 // Unconditionally drop passive data segments 1138 for (const OutputSegment *s : segments) { 1139 if (needsPassiveInitialization(s)) { 1140 // data.drop instruction 1141 writeU8(os, WASM_OPCODE_MISC_PREFIX, "bulk-memory prefix"); 1142 writeUleb128(os, WASM_OPCODE_DATA_DROP, "data.drop"); 1143 writeUleb128(os, s->index, "segment index immediate"); 1144 } 1145 } 1146 writeU8(os, WASM_OPCODE_END, "END"); 1147 } 1148 1149 createFunction(WasmSym::initMemory, bodyContent); 1150 } 1151 1152 void Writer::createStartFunction() { 1153 if (WasmSym::startFunction) { 1154 std::string bodyContent; 1155 { 1156 raw_string_ostream os(bodyContent); 1157 writeUleb128(os, 0, "num locals"); 1158 writeU8(os, WASM_OPCODE_CALL, "CALL"); 1159 writeUleb128(os, WasmSym::initMemory->getFunctionIndex(), 1160 "function index"); 1161 writeU8(os, WASM_OPCODE_CALL, "CALL"); 1162 writeUleb128(os, WasmSym::applyGlobalRelocs->getFunctionIndex(), 1163 "function index"); 1164 writeU8(os, WASM_OPCODE_END, "END"); 1165 } 1166 createFunction(WasmSym::startFunction, bodyContent); 1167 } else if (WasmSym::initMemory) { 1168 WasmSym::startFunction = WasmSym::initMemory; 1169 } else if (WasmSym::applyGlobalRelocs) { 1170 WasmSym::startFunction = WasmSym::applyGlobalRelocs; 1171 } 1172 } 1173 1174 // For -shared (PIC) output, we create create a synthetic function which will 1175 // apply any relocations to the data segments on startup. This function is 1176 // called `__wasm_apply_data_relocs` and is added at the beginning of 1177 // `__wasm_call_ctors` before any of the constructors run. 1178 void Writer::createApplyDataRelocationsFunction() { 1179 LLVM_DEBUG(dbgs() << "createApplyDataRelocationsFunction\n"); 1180 // First write the body's contents to a string. 1181 std::string bodyContent; 1182 { 1183 raw_string_ostream os(bodyContent); 1184 writeUleb128(os, 0, "num locals"); 1185 for (const OutputSegment *seg : segments) 1186 for (const InputChunk *inSeg : seg->inputSegments) 1187 inSeg->generateRelocationCode(os); 1188 1189 writeU8(os, WASM_OPCODE_END, "END"); 1190 } 1191 1192 createFunction(WasmSym::applyDataRelocs, bodyContent); 1193 } 1194 1195 // Similar to createApplyDataRelocationsFunction but generates relocation code 1196 // fro WebAssembly globals. Because these globals are not shared between threads 1197 // these relocation need to run on every thread. 1198 void Writer::createApplyGlobalRelocationsFunction() { 1199 // First write the body's contents to a string. 1200 std::string bodyContent; 1201 { 1202 raw_string_ostream os(bodyContent); 1203 writeUleb128(os, 0, "num locals"); 1204 out.globalSec->generateRelocationCode(os); 1205 writeU8(os, WASM_OPCODE_END, "END"); 1206 } 1207 1208 createFunction(WasmSym::applyGlobalRelocs, bodyContent); 1209 } 1210 1211 // Create synthetic "__wasm_call_ctors" function based on ctor functions 1212 // in input object. 1213 void Writer::createCallCtorsFunction() { 1214 // If __wasm_call_ctors isn't referenced, there aren't any ctors, and we 1215 // aren't calling `__wasm_apply_data_relocs` for Emscripten-style PIC, don't 1216 // define the `__wasm_call_ctors` function. 1217 if (!WasmSym::callCtors->isLive() && !WasmSym::applyDataRelocs && 1218 initFunctions.empty()) 1219 return; 1220 1221 // First write the body's contents to a string. 1222 std::string bodyContent; 1223 { 1224 raw_string_ostream os(bodyContent); 1225 writeUleb128(os, 0, "num locals"); 1226 1227 if (WasmSym::applyDataRelocs) { 1228 writeU8(os, WASM_OPCODE_CALL, "CALL"); 1229 writeUleb128(os, WasmSym::applyDataRelocs->getFunctionIndex(), 1230 "function index"); 1231 } 1232 1233 // Call constructors 1234 for (const WasmInitEntry &f : initFunctions) { 1235 writeU8(os, WASM_OPCODE_CALL, "CALL"); 1236 writeUleb128(os, f.sym->getFunctionIndex(), "function index"); 1237 for (size_t i = 0; i < f.sym->signature->Returns.size(); i++) { 1238 writeU8(os, WASM_OPCODE_DROP, "DROP"); 1239 } 1240 } 1241 1242 writeU8(os, WASM_OPCODE_END, "END"); 1243 } 1244 1245 createFunction(WasmSym::callCtors, bodyContent); 1246 } 1247 1248 // Create a wrapper around a function export which calls the 1249 // static constructors and destructors. 1250 void Writer::createCommandExportWrapper(uint32_t functionIndex, 1251 DefinedFunction *f) { 1252 // First write the body's contents to a string. 1253 std::string bodyContent; 1254 { 1255 raw_string_ostream os(bodyContent); 1256 writeUleb128(os, 0, "num locals"); 1257 1258 // Call `__wasm_call_ctors` which call static constructors (and 1259 // applies any runtime relocations in Emscripten-style PIC mode) 1260 if (WasmSym::callCtors->isLive()) { 1261 writeU8(os, WASM_OPCODE_CALL, "CALL"); 1262 writeUleb128(os, WasmSym::callCtors->getFunctionIndex(), 1263 "function index"); 1264 } 1265 1266 // Call the user's code, leaving any return values on the operand stack. 1267 for (size_t i = 0; i < f->signature->Params.size(); ++i) { 1268 writeU8(os, WASM_OPCODE_LOCAL_GET, "local.get"); 1269 writeUleb128(os, i, "local index"); 1270 } 1271 writeU8(os, WASM_OPCODE_CALL, "CALL"); 1272 writeUleb128(os, functionIndex, "function index"); 1273 1274 // Call the function that calls the destructors. 1275 if (DefinedFunction *callDtors = WasmSym::callDtors) { 1276 writeU8(os, WASM_OPCODE_CALL, "CALL"); 1277 writeUleb128(os, callDtors->getFunctionIndex(), "function index"); 1278 } 1279 1280 // End the function, returning the return values from the user's code. 1281 writeU8(os, WASM_OPCODE_END, "END"); 1282 } 1283 1284 createFunction(f, bodyContent); 1285 } 1286 1287 void Writer::createInitTLSFunction() { 1288 std::string bodyContent; 1289 { 1290 raw_string_ostream os(bodyContent); 1291 1292 OutputSegment *tlsSeg = nullptr; 1293 for (auto *seg : segments) { 1294 if (seg->name == ".tdata") { 1295 tlsSeg = seg; 1296 break; 1297 } 1298 } 1299 1300 writeUleb128(os, 0, "num locals"); 1301 if (tlsSeg) { 1302 writeU8(os, WASM_OPCODE_LOCAL_GET, "local.get"); 1303 writeUleb128(os, 0, "local index"); 1304 1305 writeU8(os, WASM_OPCODE_GLOBAL_SET, "global.set"); 1306 writeUleb128(os, WasmSym::tlsBase->getGlobalIndex(), "global index"); 1307 1308 // FIXME(wvo): this local needs to be I64 in wasm64, or we need an extend op. 1309 writeU8(os, WASM_OPCODE_LOCAL_GET, "local.get"); 1310 writeUleb128(os, 0, "local index"); 1311 1312 writeI32Const(os, 0, "segment offset"); 1313 1314 writeI32Const(os, tlsSeg->size, "memory region size"); 1315 1316 writeU8(os, WASM_OPCODE_MISC_PREFIX, "bulk-memory prefix"); 1317 writeUleb128(os, WASM_OPCODE_MEMORY_INIT, "MEMORY.INIT"); 1318 writeUleb128(os, tlsSeg->index, "segment index immediate"); 1319 writeU8(os, 0, "memory index immediate"); 1320 } 1321 writeU8(os, WASM_OPCODE_END, "end function"); 1322 } 1323 1324 createFunction(WasmSym::initTLS, bodyContent); 1325 } 1326 1327 // Populate InitFunctions vector with init functions from all input objects. 1328 // This is then used either when creating the output linking section or to 1329 // synthesize the "__wasm_call_ctors" function. 1330 void Writer::calculateInitFunctions() { 1331 if (!config->relocatable && !WasmSym::callCtors->isLive()) 1332 return; 1333 1334 for (ObjFile *file : symtab->objectFiles) { 1335 const WasmLinkingData &l = file->getWasmObj()->linkingData(); 1336 for (const WasmInitFunc &f : l.InitFunctions) { 1337 FunctionSymbol *sym = file->getFunctionSymbol(f.Symbol); 1338 // comdat exclusions can cause init functions be discarded. 1339 if (sym->isDiscarded() || !sym->isLive()) 1340 continue; 1341 if (sym->signature->Params.size() != 0) 1342 error("constructor functions cannot take arguments: " + toString(*sym)); 1343 LLVM_DEBUG(dbgs() << "initFunctions: " << toString(*sym) << "\n"); 1344 initFunctions.emplace_back(WasmInitEntry{sym, f.Priority}); 1345 } 1346 } 1347 1348 // Sort in order of priority (lowest first) so that they are called 1349 // in the correct order. 1350 llvm::stable_sort(initFunctions, 1351 [](const WasmInitEntry &l, const WasmInitEntry &r) { 1352 return l.priority < r.priority; 1353 }); 1354 } 1355 1356 void Writer::createSyntheticSections() { 1357 out.dylinkSec = make<DylinkSection>(); 1358 out.typeSec = make<TypeSection>(); 1359 out.importSec = make<ImportSection>(); 1360 out.functionSec = make<FunctionSection>(); 1361 out.tableSec = make<TableSection>(); 1362 out.memorySec = make<MemorySection>(); 1363 out.tagSec = make<TagSection>(); 1364 out.globalSec = make<GlobalSection>(); 1365 out.exportSec = make<ExportSection>(); 1366 out.startSec = make<StartSection>(); 1367 out.elemSec = make<ElemSection>(); 1368 out.producersSec = make<ProducersSection>(); 1369 out.targetFeaturesSec = make<TargetFeaturesSection>(); 1370 } 1371 1372 void Writer::createSyntheticSectionsPostLayout() { 1373 out.dataCountSec = make<DataCountSection>(segments); 1374 out.linkingSec = make<LinkingSection>(initFunctions, segments); 1375 out.nameSec = make<NameSection>(segments); 1376 } 1377 1378 void Writer::run() { 1379 if (config->relocatable || config->isPic) 1380 config->globalBase = 0; 1381 1382 // For PIC code the table base is assigned dynamically by the loader. 1383 // For non-PIC, we start at 1 so that accessing table index 0 always traps. 1384 if (!config->isPic) { 1385 config->tableBase = 1; 1386 if (WasmSym::definedTableBase) 1387 WasmSym::definedTableBase->setVA(config->tableBase); 1388 if (WasmSym::definedTableBase32) 1389 WasmSym::definedTableBase32->setVA(config->tableBase); 1390 } 1391 1392 log("-- createOutputSegments"); 1393 createOutputSegments(); 1394 log("-- createSyntheticSections"); 1395 createSyntheticSections(); 1396 log("-- layoutMemory"); 1397 layoutMemory(); 1398 1399 if (!config->relocatable) { 1400 // Create linker synthesized __start_SECNAME/__stop_SECNAME symbols 1401 // This has to be done after memory layout is performed. 1402 for (const OutputSegment *seg : segments) { 1403 addStartStopSymbols(seg); 1404 } 1405 } 1406 1407 for (auto &pair : config->exportedSymbols) { 1408 Symbol *sym = symtab->find(pair.first()); 1409 if (sym && sym->isDefined()) 1410 sym->forceExport = true; 1411 } 1412 1413 // Delay reporting error about explict exports until after addStartStopSymbols 1414 // which can create optional symbols. 1415 for (auto &name : config->requiredExports) { 1416 Symbol *sym = symtab->find(name); 1417 if (!sym || !sym->isDefined()) { 1418 if (config->unresolvedSymbols == UnresolvedPolicy::ReportError) 1419 error(Twine("symbol exported via --export not found: ") + name); 1420 if (config->unresolvedSymbols == UnresolvedPolicy::Warn) 1421 warn(Twine("symbol exported via --export not found: ") + name); 1422 } 1423 } 1424 1425 if (config->isPic && !config->sharedMemory) { 1426 // In shared memory mode all data segments are passive and initilized 1427 // via __wasm_init_memory. 1428 log("-- combineOutputSegments"); 1429 combineOutputSegments(); 1430 } 1431 1432 log("-- createSyntheticSectionsPostLayout"); 1433 createSyntheticSectionsPostLayout(); 1434 log("-- populateProducers"); 1435 populateProducers(); 1436 log("-- calculateImports"); 1437 calculateImports(); 1438 log("-- scanRelocations"); 1439 scanRelocations(); 1440 log("-- finalizeIndirectFunctionTable"); 1441 finalizeIndirectFunctionTable(); 1442 log("-- createSyntheticInitFunctions"); 1443 createSyntheticInitFunctions(); 1444 log("-- assignIndexes"); 1445 assignIndexes(); 1446 log("-- calculateInitFunctions"); 1447 calculateInitFunctions(); 1448 1449 if (!config->relocatable) { 1450 // Create linker synthesized functions 1451 if (WasmSym::applyDataRelocs) 1452 createApplyDataRelocationsFunction(); 1453 if (WasmSym::applyGlobalRelocs) 1454 createApplyGlobalRelocationsFunction(); 1455 if (WasmSym::initMemory) 1456 createInitMemoryFunction(); 1457 createStartFunction(); 1458 1459 createCallCtorsFunction(); 1460 1461 // Create export wrappers for commands if needed. 1462 // 1463 // If the input contains a call to `__wasm_call_ctors`, either in one of 1464 // the input objects or an explicit export from the command-line, we 1465 // assume ctors and dtors are taken care of already. 1466 if (!config->relocatable && !config->isPic && 1467 !WasmSym::callCtors->isUsedInRegularObj && 1468 !WasmSym::callCtors->isExported()) { 1469 log("-- createCommandExportWrappers"); 1470 createCommandExportWrappers(); 1471 } 1472 } 1473 1474 if (WasmSym::initTLS && WasmSym::initTLS->isLive()) 1475 createInitTLSFunction(); 1476 1477 if (errorCount()) 1478 return; 1479 1480 log("-- calculateTypes"); 1481 calculateTypes(); 1482 log("-- calculateExports"); 1483 calculateExports(); 1484 log("-- calculateCustomSections"); 1485 calculateCustomSections(); 1486 log("-- populateSymtab"); 1487 populateSymtab(); 1488 log("-- populateTargetFeatures"); 1489 populateTargetFeatures(); 1490 log("-- addSections"); 1491 addSections(); 1492 1493 if (errorHandler().verbose) { 1494 log("Defined Functions: " + Twine(out.functionSec->inputFunctions.size())); 1495 log("Defined Globals : " + Twine(out.globalSec->numGlobals())); 1496 log("Defined Tags : " + Twine(out.tagSec->inputTags.size())); 1497 log("Defined Tables : " + Twine(out.tableSec->inputTables.size())); 1498 log("Function Imports : " + 1499 Twine(out.importSec->getNumImportedFunctions())); 1500 log("Global Imports : " + Twine(out.importSec->getNumImportedGlobals())); 1501 log("Tag Imports : " + Twine(out.importSec->getNumImportedTags())); 1502 log("Table Imports : " + Twine(out.importSec->getNumImportedTables())); 1503 for (ObjFile *file : symtab->objectFiles) 1504 file->dumpInfo(); 1505 } 1506 1507 createHeader(); 1508 log("-- finalizeSections"); 1509 finalizeSections(); 1510 1511 log("-- writeMapFile"); 1512 writeMapFile(outputSections); 1513 1514 log("-- openFile"); 1515 openFile(); 1516 if (errorCount()) 1517 return; 1518 1519 writeHeader(); 1520 1521 log("-- writeSections"); 1522 writeSections(); 1523 if (errorCount()) 1524 return; 1525 1526 if (Error e = buffer->commit()) 1527 fatal("failed to write the output file: " + toString(std::move(e))); 1528 } 1529 1530 // Open a result file. 1531 void Writer::openFile() { 1532 log("writing: " + config->outputFile); 1533 1534 Expected<std::unique_ptr<FileOutputBuffer>> bufferOrErr = 1535 FileOutputBuffer::create(config->outputFile, fileSize, 1536 FileOutputBuffer::F_executable); 1537 1538 if (!bufferOrErr) 1539 error("failed to open " + config->outputFile + ": " + 1540 toString(bufferOrErr.takeError())); 1541 else 1542 buffer = std::move(*bufferOrErr); 1543 } 1544 1545 void Writer::createHeader() { 1546 raw_string_ostream os(header); 1547 writeBytes(os, WasmMagic, sizeof(WasmMagic), "wasm magic"); 1548 writeU32(os, WasmVersion, "wasm version"); 1549 os.flush(); 1550 fileSize += header.size(); 1551 } 1552 1553 void writeResult() { Writer().run(); } 1554 1555 } // namespace wasm 1556 } // namespace lld 1557