1 //===- Driver.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 // The driver drives the entire linking process. It is responsible for 10 // parsing command line options and doing whatever it is instructed to do. 11 // 12 // One notable thing in the LLD's driver when compared to other linkers is 13 // that the LLD's driver is agnostic on the host operating system. 14 // Other linkers usually have implicit default values (such as a dynamic 15 // linker path or library paths) for each host OS. 16 // 17 // I don't think implicit default values are useful because they are 18 // usually explicitly specified by the compiler driver. They can even 19 // be harmful when you are doing cross-linking. Therefore, in LLD, we 20 // simply trust the compiler driver to pass all required options and 21 // don't try to make effort on our side. 22 // 23 //===----------------------------------------------------------------------===// 24 25 #include "Driver.h" 26 #include "Config.h" 27 #include "ICF.h" 28 #include "InputFiles.h" 29 #include "InputSection.h" 30 #include "LinkerScript.h" 31 #include "MarkLive.h" 32 #include "OutputSections.h" 33 #include "ScriptParser.h" 34 #include "SymbolTable.h" 35 #include "Symbols.h" 36 #include "SyntheticSections.h" 37 #include "Target.h" 38 #include "Writer.h" 39 #include "lld/Common/Args.h" 40 #include "lld/Common/Driver.h" 41 #include "lld/Common/ErrorHandler.h" 42 #include "lld/Common/Filesystem.h" 43 #include "lld/Common/Memory.h" 44 #include "lld/Common/Strings.h" 45 #include "lld/Common/TargetOptionsCommandFlags.h" 46 #include "lld/Common/Threads.h" 47 #include "lld/Common/Version.h" 48 #include "llvm/ADT/SetVector.h" 49 #include "llvm/ADT/StringExtras.h" 50 #include "llvm/ADT/StringSwitch.h" 51 #include "llvm/Support/CommandLine.h" 52 #include "llvm/Support/Compression.h" 53 #include "llvm/Support/GlobPattern.h" 54 #include "llvm/Support/LEB128.h" 55 #include "llvm/Support/Path.h" 56 #include "llvm/Support/TarWriter.h" 57 #include "llvm/Support/TargetSelect.h" 58 #include "llvm/Support/raw_ostream.h" 59 #include <cstdlib> 60 #include <utility> 61 62 using namespace llvm; 63 using namespace llvm::ELF; 64 using namespace llvm::object; 65 using namespace llvm::sys; 66 using namespace llvm::support; 67 68 using namespace lld; 69 using namespace lld::elf; 70 71 Configuration *elf::config; 72 LinkerDriver *elf::driver; 73 74 static void setConfigs(opt::InputArgList &args); 75 static void readConfigs(opt::InputArgList &args); 76 77 bool elf::link(ArrayRef<const char *> args, bool canExitEarly, 78 raw_ostream &error) { 79 errorHandler().logName = args::getFilenameWithoutExe(args[0]); 80 errorHandler().errorLimitExceededMsg = 81 "too many errors emitted, stopping now (use " 82 "-error-limit=0 to see all errors)"; 83 errorHandler().errorOS = &error; 84 errorHandler().exitEarly = canExitEarly; 85 errorHandler().colorDiagnostics = error.has_colors(); 86 87 inputSections.clear(); 88 outputSections.clear(); 89 binaryFiles.clear(); 90 bitcodeFiles.clear(); 91 objectFiles.clear(); 92 sharedFiles.clear(); 93 94 config = make<Configuration>(); 95 driver = make<LinkerDriver>(); 96 script = make<LinkerScript>(); 97 symtab = make<SymbolTable>(); 98 99 tar = nullptr; 100 memset(&in, 0, sizeof(in)); 101 102 partitions = {Partition()}; 103 104 SharedFile::vernauxNum = 0; 105 106 config->progName = args[0]; 107 108 driver->main(args); 109 110 // Exit immediately if we don't need to return to the caller. 111 // This saves time because the overhead of calling destructors 112 // for all globally-allocated objects is not negligible. 113 if (canExitEarly) 114 exitLld(errorCount() ? 1 : 0); 115 116 freeArena(); 117 return !errorCount(); 118 } 119 120 // Parses a linker -m option. 121 static std::tuple<ELFKind, uint16_t, uint8_t> parseEmulation(StringRef emul) { 122 uint8_t osabi = 0; 123 StringRef s = emul; 124 if (s.endswith("_fbsd")) { 125 s = s.drop_back(5); 126 osabi = ELFOSABI_FREEBSD; 127 } 128 129 std::pair<ELFKind, uint16_t> ret = 130 StringSwitch<std::pair<ELFKind, uint16_t>>(s) 131 .Cases("aarch64elf", "aarch64linux", "aarch64_elf64_le_vec", 132 {ELF64LEKind, EM_AARCH64}) 133 .Cases("armelf", "armelf_linux_eabi", {ELF32LEKind, EM_ARM}) 134 .Case("elf32_x86_64", {ELF32LEKind, EM_X86_64}) 135 .Cases("elf32btsmip", "elf32btsmipn32", {ELF32BEKind, EM_MIPS}) 136 .Cases("elf32ltsmip", "elf32ltsmipn32", {ELF32LEKind, EM_MIPS}) 137 .Case("elf32lriscv", {ELF32LEKind, EM_RISCV}) 138 .Cases("elf32ppc", "elf32ppclinux", {ELF32BEKind, EM_PPC}) 139 .Case("elf64btsmip", {ELF64BEKind, EM_MIPS}) 140 .Case("elf64ltsmip", {ELF64LEKind, EM_MIPS}) 141 .Case("elf64lriscv", {ELF64LEKind, EM_RISCV}) 142 .Case("elf64ppc", {ELF64BEKind, EM_PPC64}) 143 .Case("elf64lppc", {ELF64LEKind, EM_PPC64}) 144 .Cases("elf_amd64", "elf_x86_64", {ELF64LEKind, EM_X86_64}) 145 .Case("elf_i386", {ELF32LEKind, EM_386}) 146 .Case("elf_iamcu", {ELF32LEKind, EM_IAMCU}) 147 .Default({ELFNoneKind, EM_NONE}); 148 149 if (ret.first == ELFNoneKind) 150 error("unknown emulation: " + emul); 151 return std::make_tuple(ret.first, ret.second, osabi); 152 } 153 154 // Returns slices of MB by parsing MB as an archive file. 155 // Each slice consists of a member file in the archive. 156 std::vector<std::pair<MemoryBufferRef, uint64_t>> static getArchiveMembers( 157 MemoryBufferRef mb) { 158 std::unique_ptr<Archive> file = 159 CHECK(Archive::create(mb), 160 mb.getBufferIdentifier() + ": failed to parse archive"); 161 162 std::vector<std::pair<MemoryBufferRef, uint64_t>> v; 163 Error err = Error::success(); 164 bool addToTar = file->isThin() && tar; 165 for (const ErrorOr<Archive::Child> &cOrErr : file->children(err)) { 166 Archive::Child c = 167 CHECK(cOrErr, mb.getBufferIdentifier() + 168 ": could not get the child of the archive"); 169 MemoryBufferRef mbref = 170 CHECK(c.getMemoryBufferRef(), 171 mb.getBufferIdentifier() + 172 ": could not get the buffer for a child of the archive"); 173 if (addToTar) 174 tar->append(relativeToRoot(check(c.getFullName())), mbref.getBuffer()); 175 v.push_back(std::make_pair(mbref, c.getChildOffset())); 176 } 177 if (err) 178 fatal(mb.getBufferIdentifier() + ": Archive::children failed: " + 179 toString(std::move(err))); 180 181 // Take ownership of memory buffers created for members of thin archives. 182 for (std::unique_ptr<MemoryBuffer> &mb : file->takeThinBuffers()) 183 make<std::unique_ptr<MemoryBuffer>>(std::move(mb)); 184 185 return v; 186 } 187 188 // Opens a file and create a file object. Path has to be resolved already. 189 void LinkerDriver::addFile(StringRef path, bool withLOption) { 190 using namespace sys::fs; 191 192 Optional<MemoryBufferRef> buffer = readFile(path); 193 if (!buffer.hasValue()) 194 return; 195 MemoryBufferRef mbref = *buffer; 196 197 if (config->formatBinary) { 198 files.push_back(make<BinaryFile>(mbref)); 199 return; 200 } 201 202 switch (identify_magic(mbref.getBuffer())) { 203 case file_magic::unknown: 204 readLinkerScript(mbref); 205 return; 206 case file_magic::archive: { 207 // Handle -whole-archive. 208 if (inWholeArchive) { 209 for (const auto &p : getArchiveMembers(mbref)) 210 files.push_back(createObjectFile(p.first, path, p.second)); 211 return; 212 } 213 214 std::unique_ptr<Archive> file = 215 CHECK(Archive::create(mbref), path + ": failed to parse archive"); 216 217 // If an archive file has no symbol table, it is likely that a user 218 // is attempting LTO and using a default ar command that doesn't 219 // understand the LLVM bitcode file. It is a pretty common error, so 220 // we'll handle it as if it had a symbol table. 221 if (!file->isEmpty() && !file->hasSymbolTable()) { 222 // Check if all members are bitcode files. If not, ignore, which is the 223 // default action without the LTO hack described above. 224 for (const std::pair<MemoryBufferRef, uint64_t> &p : 225 getArchiveMembers(mbref)) 226 if (identify_magic(p.first.getBuffer()) != file_magic::bitcode) { 227 error(path + ": archive has no index; run ranlib to add one"); 228 return; 229 } 230 231 for (const std::pair<MemoryBufferRef, uint64_t> &p : 232 getArchiveMembers(mbref)) 233 files.push_back(make<LazyObjFile>(p.first, path, p.second)); 234 return; 235 } 236 237 // Handle the regular case. 238 files.push_back(make<ArchiveFile>(std::move(file))); 239 return; 240 } 241 case file_magic::elf_shared_object: 242 if (config->isStatic || config->relocatable) { 243 error("attempted static link of dynamic object " + path); 244 return; 245 } 246 247 // DSOs usually have DT_SONAME tags in their ELF headers, and the 248 // sonames are used to identify DSOs. But if they are missing, 249 // they are identified by filenames. We don't know whether the new 250 // file has a DT_SONAME or not because we haven't parsed it yet. 251 // Here, we set the default soname for the file because we might 252 // need it later. 253 // 254 // If a file was specified by -lfoo, the directory part is not 255 // significant, as a user did not specify it. This behavior is 256 // compatible with GNU. 257 files.push_back( 258 make<SharedFile>(mbref, withLOption ? path::filename(path) : path)); 259 return; 260 case file_magic::bitcode: 261 case file_magic::elf_relocatable: 262 if (inLib) 263 files.push_back(make<LazyObjFile>(mbref, "", 0)); 264 else 265 files.push_back(createObjectFile(mbref)); 266 break; 267 default: 268 error(path + ": unknown file type"); 269 } 270 } 271 272 // Add a given library by searching it from input search paths. 273 void LinkerDriver::addLibrary(StringRef name) { 274 if (Optional<std::string> path = searchLibrary(name)) 275 addFile(*path, /*withLOption=*/true); 276 else 277 error("unable to find library -l" + name); 278 } 279 280 // This function is called on startup. We need this for LTO since 281 // LTO calls LLVM functions to compile bitcode files to native code. 282 // Technically this can be delayed until we read bitcode files, but 283 // we don't bother to do lazily because the initialization is fast. 284 static void initLLVM() { 285 InitializeAllTargets(); 286 InitializeAllTargetMCs(); 287 InitializeAllAsmPrinters(); 288 InitializeAllAsmParsers(); 289 } 290 291 // Some command line options or some combinations of them are not allowed. 292 // This function checks for such errors. 293 static void checkOptions() { 294 // The MIPS ABI as of 2016 does not support the GNU-style symbol lookup 295 // table which is a relatively new feature. 296 if (config->emachine == EM_MIPS && config->gnuHash) 297 error("the .gnu.hash section is not compatible with the MIPS target"); 298 299 if (config->fixCortexA53Errata843419 && config->emachine != EM_AARCH64) 300 error("--fix-cortex-a53-843419 is only supported on AArch64 targets"); 301 302 if (config->tocOptimize && config->emachine != EM_PPC64) 303 error("--toc-optimize is only supported on the PowerPC64 target"); 304 305 if (config->pie && config->shared) 306 error("-shared and -pie may not be used together"); 307 308 if (!config->shared && !config->filterList.empty()) 309 error("-F may not be used without -shared"); 310 311 if (!config->shared && !config->auxiliaryList.empty()) 312 error("-f may not be used without -shared"); 313 314 if (!config->relocatable && !config->defineCommon) 315 error("-no-define-common not supported in non relocatable output"); 316 317 if (config->zText && config->zIfuncNoplt) 318 error("-z text and -z ifunc-noplt may not be used together"); 319 320 if (config->relocatable) { 321 if (config->shared) 322 error("-r and -shared may not be used together"); 323 if (config->gcSections) 324 error("-r and --gc-sections may not be used together"); 325 if (config->gdbIndex) 326 error("-r and --gdb-index may not be used together"); 327 if (config->icf != ICFLevel::None) 328 error("-r and --icf may not be used together"); 329 if (config->pie) 330 error("-r and -pie may not be used together"); 331 } 332 333 if (config->executeOnly) { 334 if (config->emachine != EM_AARCH64) 335 error("-execute-only is only supported on AArch64 targets"); 336 337 if (config->singleRoRx && !script->hasSectionsCommand) 338 error("-execute-only and -no-rosegment cannot be used together"); 339 } 340 341 if (config->zRetpolineplt && config->requireCET) 342 error("--require-cet may not be used with -z retpolineplt"); 343 344 if (config->emachine != EM_AARCH64) { 345 if (config->pacPlt) 346 error("--pac-plt only supported on AArch64"); 347 if (config->forceBTI) 348 error("--force-bti only supported on AArch64"); 349 } 350 } 351 352 static const char *getReproduceOption(opt::InputArgList &args) { 353 if (auto *arg = args.getLastArg(OPT_reproduce)) 354 return arg->getValue(); 355 return getenv("LLD_REPRODUCE"); 356 } 357 358 static bool hasZOption(opt::InputArgList &args, StringRef key) { 359 for (auto *arg : args.filtered(OPT_z)) 360 if (key == arg->getValue()) 361 return true; 362 return false; 363 } 364 365 static bool getZFlag(opt::InputArgList &args, StringRef k1, StringRef k2, 366 bool Default) { 367 for (auto *arg : args.filtered_reverse(OPT_z)) { 368 if (k1 == arg->getValue()) 369 return true; 370 if (k2 == arg->getValue()) 371 return false; 372 } 373 return Default; 374 } 375 376 static bool isKnownZFlag(StringRef s) { 377 return s == "combreloc" || s == "copyreloc" || s == "defs" || 378 s == "execstack" || s == "global" || s == "hazardplt" || 379 s == "ifunc-noplt" || s == "initfirst" || s == "interpose" || 380 s == "keep-text-section-prefix" || s == "lazy" || s == "muldefs" || 381 s == "nocombreloc" || s == "nocopyreloc" || s == "nodefaultlib" || 382 s == "nodelete" || s == "nodlopen" || s == "noexecstack" || 383 s == "nokeep-text-section-prefix" || s == "norelro" || s == "notext" || 384 s == "now" || s == "origin" || s == "relro" || s == "retpolineplt" || 385 s == "rodynamic" || s == "text" || s == "wxneeded" || 386 s.startswith("common-page-size") || s.startswith("max-page-size=") || 387 s.startswith("stack-size="); 388 } 389 390 // Report an error for an unknown -z option. 391 static void checkZOptions(opt::InputArgList &args) { 392 for (auto *arg : args.filtered(OPT_z)) 393 if (!isKnownZFlag(arg->getValue())) 394 error("unknown -z value: " + StringRef(arg->getValue())); 395 } 396 397 void LinkerDriver::main(ArrayRef<const char *> argsArr) { 398 ELFOptTable parser; 399 opt::InputArgList args = parser.parse(argsArr.slice(1)); 400 401 // Interpret this flag early because error() depends on them. 402 errorHandler().errorLimit = args::getInteger(args, OPT_error_limit, 20); 403 checkZOptions(args); 404 405 // Handle -help 406 if (args.hasArg(OPT_help)) { 407 printHelp(); 408 return; 409 } 410 411 // Handle -v or -version. 412 // 413 // A note about "compatible with GNU linkers" message: this is a hack for 414 // scripts generated by GNU Libtool 2.4.6 (released in February 2014 and 415 // still the newest version in March 2017) or earlier to recognize LLD as 416 // a GNU compatible linker. As long as an output for the -v option 417 // contains "GNU" or "with BFD", they recognize us as GNU-compatible. 418 // 419 // This is somewhat ugly hack, but in reality, we had no choice other 420 // than doing this. Considering the very long release cycle of Libtool, 421 // it is not easy to improve it to recognize LLD as a GNU compatible 422 // linker in a timely manner. Even if we can make it, there are still a 423 // lot of "configure" scripts out there that are generated by old version 424 // of Libtool. We cannot convince every software developer to migrate to 425 // the latest version and re-generate scripts. So we have this hack. 426 if (args.hasArg(OPT_v) || args.hasArg(OPT_version)) 427 message(getLLDVersion() + " (compatible with GNU linkers)"); 428 429 if (const char *path = getReproduceOption(args)) { 430 // Note that --reproduce is a debug option so you can ignore it 431 // if you are trying to understand the whole picture of the code. 432 Expected<std::unique_ptr<TarWriter>> errOrWriter = 433 TarWriter::create(path, path::stem(path)); 434 if (errOrWriter) { 435 tar = std::move(*errOrWriter); 436 tar->append("response.txt", createResponseFile(args)); 437 tar->append("version.txt", getLLDVersion() + "\n"); 438 } else { 439 error("--reproduce: " + toString(errOrWriter.takeError())); 440 } 441 } 442 443 readConfigs(args); 444 445 // The behavior of -v or --version is a bit strange, but this is 446 // needed for compatibility with GNU linkers. 447 if (args.hasArg(OPT_v) && !args.hasArg(OPT_INPUT)) 448 return; 449 if (args.hasArg(OPT_version)) 450 return; 451 452 initLLVM(); 453 createFiles(args); 454 if (errorCount()) 455 return; 456 457 inferMachineType(); 458 setConfigs(args); 459 checkOptions(); 460 if (errorCount()) 461 return; 462 463 // The Target instance handles target-specific stuff, such as applying 464 // relocations or writing a PLT section. It also contains target-dependent 465 // values such as a default image base address. 466 target = getTarget(); 467 468 switch (config->ekind) { 469 case ELF32LEKind: 470 link<ELF32LE>(args); 471 return; 472 case ELF32BEKind: 473 link<ELF32BE>(args); 474 return; 475 case ELF64LEKind: 476 link<ELF64LE>(args); 477 return; 478 case ELF64BEKind: 479 link<ELF64BE>(args); 480 return; 481 default: 482 llvm_unreachable("unknown Config->EKind"); 483 } 484 } 485 486 static std::string getRpath(opt::InputArgList &args) { 487 std::vector<StringRef> v = args::getStrings(args, OPT_rpath); 488 return llvm::join(v.begin(), v.end(), ":"); 489 } 490 491 // Determines what we should do if there are remaining unresolved 492 // symbols after the name resolution. 493 static UnresolvedPolicy getUnresolvedSymbolPolicy(opt::InputArgList &args) { 494 UnresolvedPolicy errorOrWarn = args.hasFlag(OPT_error_unresolved_symbols, 495 OPT_warn_unresolved_symbols, true) 496 ? UnresolvedPolicy::ReportError 497 : UnresolvedPolicy::Warn; 498 499 // Process the last of -unresolved-symbols, -no-undefined or -z defs. 500 for (auto *arg : llvm::reverse(args)) { 501 switch (arg->getOption().getID()) { 502 case OPT_unresolved_symbols: { 503 StringRef s = arg->getValue(); 504 if (s == "ignore-all" || s == "ignore-in-object-files") 505 return UnresolvedPolicy::Ignore; 506 if (s == "ignore-in-shared-libs" || s == "report-all") 507 return errorOrWarn; 508 error("unknown --unresolved-symbols value: " + s); 509 continue; 510 } 511 case OPT_no_undefined: 512 return errorOrWarn; 513 case OPT_z: 514 if (StringRef(arg->getValue()) == "defs") 515 return errorOrWarn; 516 continue; 517 } 518 } 519 520 // -shared implies -unresolved-symbols=ignore-all because missing 521 // symbols are likely to be resolved at runtime using other DSOs. 522 if (config->shared) 523 return UnresolvedPolicy::Ignore; 524 return errorOrWarn; 525 } 526 527 static Target2Policy getTarget2(opt::InputArgList &args) { 528 StringRef s = args.getLastArgValue(OPT_target2, "got-rel"); 529 if (s == "rel") 530 return Target2Policy::Rel; 531 if (s == "abs") 532 return Target2Policy::Abs; 533 if (s == "got-rel") 534 return Target2Policy::GotRel; 535 error("unknown --target2 option: " + s); 536 return Target2Policy::GotRel; 537 } 538 539 static bool isOutputFormatBinary(opt::InputArgList &args) { 540 StringRef s = args.getLastArgValue(OPT_oformat, "elf"); 541 if (s == "binary") 542 return true; 543 if (!s.startswith("elf")) 544 error("unknown --oformat value: " + s); 545 return false; 546 } 547 548 static DiscardPolicy getDiscard(opt::InputArgList &args) { 549 if (args.hasArg(OPT_relocatable)) 550 return DiscardPolicy::None; 551 552 auto *arg = 553 args.getLastArg(OPT_discard_all, OPT_discard_locals, OPT_discard_none); 554 if (!arg) 555 return DiscardPolicy::Default; 556 if (arg->getOption().getID() == OPT_discard_all) 557 return DiscardPolicy::All; 558 if (arg->getOption().getID() == OPT_discard_locals) 559 return DiscardPolicy::Locals; 560 return DiscardPolicy::None; 561 } 562 563 static StringRef getDynamicLinker(opt::InputArgList &args) { 564 auto *arg = args.getLastArg(OPT_dynamic_linker, OPT_no_dynamic_linker); 565 if (!arg || arg->getOption().getID() == OPT_no_dynamic_linker) 566 return ""; 567 return arg->getValue(); 568 } 569 570 static ICFLevel getICF(opt::InputArgList &args) { 571 auto *arg = args.getLastArg(OPT_icf_none, OPT_icf_safe, OPT_icf_all); 572 if (!arg || arg->getOption().getID() == OPT_icf_none) 573 return ICFLevel::None; 574 if (arg->getOption().getID() == OPT_icf_safe) 575 return ICFLevel::Safe; 576 return ICFLevel::All; 577 } 578 579 static StripPolicy getStrip(opt::InputArgList &args) { 580 if (args.hasArg(OPT_relocatable)) 581 return StripPolicy::None; 582 583 auto *arg = args.getLastArg(OPT_strip_all, OPT_strip_debug); 584 if (!arg) 585 return StripPolicy::None; 586 if (arg->getOption().getID() == OPT_strip_all) 587 return StripPolicy::All; 588 return StripPolicy::Debug; 589 } 590 591 static uint64_t parseSectionAddress(StringRef s, opt::InputArgList &args, 592 const opt::Arg &arg) { 593 uint64_t va = 0; 594 if (s.startswith("0x")) 595 s = s.drop_front(2); 596 if (!to_integer(s, va, 16)) 597 error("invalid argument: " + arg.getAsString(args)); 598 return va; 599 } 600 601 static StringMap<uint64_t> getSectionStartMap(opt::InputArgList &args) { 602 StringMap<uint64_t> ret; 603 for (auto *arg : args.filtered(OPT_section_start)) { 604 StringRef name; 605 StringRef addr; 606 std::tie(name, addr) = StringRef(arg->getValue()).split('='); 607 ret[name] = parseSectionAddress(addr, args, *arg); 608 } 609 610 if (auto *arg = args.getLastArg(OPT_Ttext)) 611 ret[".text"] = parseSectionAddress(arg->getValue(), args, *arg); 612 if (auto *arg = args.getLastArg(OPT_Tdata)) 613 ret[".data"] = parseSectionAddress(arg->getValue(), args, *arg); 614 if (auto *arg = args.getLastArg(OPT_Tbss)) 615 ret[".bss"] = parseSectionAddress(arg->getValue(), args, *arg); 616 return ret; 617 } 618 619 static SortSectionPolicy getSortSection(opt::InputArgList &args) { 620 StringRef s = args.getLastArgValue(OPT_sort_section); 621 if (s == "alignment") 622 return SortSectionPolicy::Alignment; 623 if (s == "name") 624 return SortSectionPolicy::Name; 625 if (!s.empty()) 626 error("unknown --sort-section rule: " + s); 627 return SortSectionPolicy::Default; 628 } 629 630 static OrphanHandlingPolicy getOrphanHandling(opt::InputArgList &args) { 631 StringRef s = args.getLastArgValue(OPT_orphan_handling, "place"); 632 if (s == "warn") 633 return OrphanHandlingPolicy::Warn; 634 if (s == "error") 635 return OrphanHandlingPolicy::Error; 636 if (s != "place") 637 error("unknown --orphan-handling mode: " + s); 638 return OrphanHandlingPolicy::Place; 639 } 640 641 // Parse --build-id or --build-id=<style>. We handle "tree" as a 642 // synonym for "sha1" because all our hash functions including 643 // -build-id=sha1 are actually tree hashes for performance reasons. 644 static std::pair<BuildIdKind, std::vector<uint8_t>> 645 getBuildId(opt::InputArgList &args) { 646 auto *arg = args.getLastArg(OPT_build_id, OPT_build_id_eq); 647 if (!arg) 648 return {BuildIdKind::None, {}}; 649 650 if (arg->getOption().getID() == OPT_build_id) 651 return {BuildIdKind::Fast, {}}; 652 653 StringRef s = arg->getValue(); 654 if (s == "fast") 655 return {BuildIdKind::Fast, {}}; 656 if (s == "md5") 657 return {BuildIdKind::Md5, {}}; 658 if (s == "sha1" || s == "tree") 659 return {BuildIdKind::Sha1, {}}; 660 if (s == "uuid") 661 return {BuildIdKind::Uuid, {}}; 662 if (s.startswith("0x")) 663 return {BuildIdKind::Hexstring, parseHex(s.substr(2))}; 664 665 if (s != "none") 666 error("unknown --build-id style: " + s); 667 return {BuildIdKind::None, {}}; 668 } 669 670 static std::pair<bool, bool> getPackDynRelocs(opt::InputArgList &args) { 671 StringRef s = args.getLastArgValue(OPT_pack_dyn_relocs, "none"); 672 if (s == "android") 673 return {true, false}; 674 if (s == "relr") 675 return {false, true}; 676 if (s == "android+relr") 677 return {true, true}; 678 679 if (s != "none") 680 error("unknown -pack-dyn-relocs format: " + s); 681 return {false, false}; 682 } 683 684 static void readCallGraph(MemoryBufferRef mb) { 685 // Build a map from symbol name to section 686 DenseMap<StringRef, Symbol *> map; 687 for (InputFile *file : objectFiles) 688 for (Symbol *sym : file->getSymbols()) 689 map[sym->getName()] = sym; 690 691 auto findSection = [&](StringRef name) -> InputSectionBase * { 692 Symbol *sym = map.lookup(name); 693 if (!sym) { 694 if (config->warnSymbolOrdering) 695 warn(mb.getBufferIdentifier() + ": no such symbol: " + name); 696 return nullptr; 697 } 698 maybeWarnUnorderableSymbol(sym); 699 700 if (Defined *dr = dyn_cast_or_null<Defined>(sym)) 701 return dyn_cast_or_null<InputSectionBase>(dr->section); 702 return nullptr; 703 }; 704 705 for (StringRef line : args::getLines(mb)) { 706 SmallVector<StringRef, 3> fields; 707 line.split(fields, ' '); 708 uint64_t count; 709 710 if (fields.size() != 3 || !to_integer(fields[2], count)) { 711 error(mb.getBufferIdentifier() + ": parse error"); 712 return; 713 } 714 715 if (InputSectionBase *from = findSection(fields[0])) 716 if (InputSectionBase *to = findSection(fields[1])) 717 config->callGraphProfile[std::make_pair(from, to)] += count; 718 } 719 } 720 721 template <class ELFT> static void readCallGraphsFromObjectFiles() { 722 for (auto file : objectFiles) { 723 auto *obj = cast<ObjFile<ELFT>>(file); 724 725 for (const Elf_CGProfile_Impl<ELFT> &cgpe : obj->cgProfile) { 726 auto *fromSym = dyn_cast<Defined>(&obj->getSymbol(cgpe.cgp_from)); 727 auto *toSym = dyn_cast<Defined>(&obj->getSymbol(cgpe.cgp_to)); 728 if (!fromSym || !toSym) 729 continue; 730 731 auto *from = dyn_cast_or_null<InputSectionBase>(fromSym->section); 732 auto *to = dyn_cast_or_null<InputSectionBase>(toSym->section); 733 if (from && to) 734 config->callGraphProfile[{from, to}] += cgpe.cgp_weight; 735 } 736 } 737 } 738 739 static bool getCompressDebugSections(opt::InputArgList &args) { 740 StringRef s = args.getLastArgValue(OPT_compress_debug_sections, "none"); 741 if (s == "none") 742 return false; 743 if (s != "zlib") 744 error("unknown --compress-debug-sections value: " + s); 745 if (!zlib::isAvailable()) 746 error("--compress-debug-sections: zlib is not available"); 747 return true; 748 } 749 750 static std::pair<StringRef, StringRef> getOldNewOptions(opt::InputArgList &args, 751 unsigned id) { 752 auto *arg = args.getLastArg(id); 753 if (!arg) 754 return {"", ""}; 755 756 StringRef s = arg->getValue(); 757 std::pair<StringRef, StringRef> ret = s.split(';'); 758 if (ret.second.empty()) 759 error(arg->getSpelling() + " expects 'old;new' format, but got " + s); 760 return ret; 761 } 762 763 // Parse the symbol ordering file and warn for any duplicate entries. 764 static std::vector<StringRef> getSymbolOrderingFile(MemoryBufferRef mb) { 765 SetVector<StringRef> names; 766 for (StringRef s : args::getLines(mb)) 767 if (!names.insert(s) && config->warnSymbolOrdering) 768 warn(mb.getBufferIdentifier() + ": duplicate ordered symbol: " + s); 769 770 return names.takeVector(); 771 } 772 773 static void parseClangOption(StringRef opt, const Twine &msg) { 774 std::string err; 775 raw_string_ostream os(err); 776 777 const char *argv[] = {config->progName.data(), opt.data()}; 778 if (cl::ParseCommandLineOptions(2, argv, "", &os)) 779 return; 780 os.flush(); 781 error(msg + ": " + StringRef(err).trim()); 782 } 783 784 // Initializes Config members by the command line options. 785 static void readConfigs(opt::InputArgList &args) { 786 errorHandler().verbose = args.hasArg(OPT_verbose); 787 errorHandler().fatalWarnings = 788 args.hasFlag(OPT_fatal_warnings, OPT_no_fatal_warnings, false); 789 errorHandler().vsDiagnostics = 790 args.hasArg(OPT_visual_studio_diagnostics_format, false); 791 threadsEnabled = args.hasFlag(OPT_threads, OPT_no_threads, true); 792 793 config->allowMultipleDefinition = 794 args.hasFlag(OPT_allow_multiple_definition, 795 OPT_no_allow_multiple_definition, false) || 796 hasZOption(args, "muldefs"); 797 config->allowShlibUndefined = 798 args.hasFlag(OPT_allow_shlib_undefined, OPT_no_allow_shlib_undefined, 799 args.hasArg(OPT_shared)); 800 config->auxiliaryList = args::getStrings(args, OPT_auxiliary); 801 config->bsymbolic = args.hasArg(OPT_Bsymbolic); 802 config->bsymbolicFunctions = args.hasArg(OPT_Bsymbolic_functions); 803 config->checkSections = 804 args.hasFlag(OPT_check_sections, OPT_no_check_sections, true); 805 config->chroot = args.getLastArgValue(OPT_chroot); 806 config->compressDebugSections = getCompressDebugSections(args); 807 config->cref = args.hasFlag(OPT_cref, OPT_no_cref, false); 808 config->defineCommon = args.hasFlag(OPT_define_common, OPT_no_define_common, 809 !args.hasArg(OPT_relocatable)); 810 config->demangle = args.hasFlag(OPT_demangle, OPT_no_demangle, true); 811 config->dependentLibraries = args.hasFlag(OPT_dependent_libraries, OPT_no_dependent_libraries, true); 812 config->disableVerify = args.hasArg(OPT_disable_verify); 813 config->discard = getDiscard(args); 814 config->dwoDir = args.getLastArgValue(OPT_plugin_opt_dwo_dir_eq); 815 config->dynamicLinker = getDynamicLinker(args); 816 config->ehFrameHdr = 817 args.hasFlag(OPT_eh_frame_hdr, OPT_no_eh_frame_hdr, false); 818 config->emitLLVM = args.hasArg(OPT_plugin_opt_emit_llvm, false); 819 config->emitRelocs = args.hasArg(OPT_emit_relocs); 820 config->callGraphProfileSort = args.hasFlag( 821 OPT_call_graph_profile_sort, OPT_no_call_graph_profile_sort, true); 822 config->enableNewDtags = 823 args.hasFlag(OPT_enable_new_dtags, OPT_disable_new_dtags, true); 824 config->entry = args.getLastArgValue(OPT_entry); 825 config->executeOnly = 826 args.hasFlag(OPT_execute_only, OPT_no_execute_only, false); 827 config->exportDynamic = 828 args.hasFlag(OPT_export_dynamic, OPT_no_export_dynamic, false); 829 config->filterList = args::getStrings(args, OPT_filter); 830 config->fini = args.getLastArgValue(OPT_fini, "_fini"); 831 config->fixCortexA53Errata843419 = args.hasArg(OPT_fix_cortex_a53_843419); 832 config->forceBTI = args.hasArg(OPT_force_bti); 833 config->requireCET = args.hasArg(OPT_require_cet); 834 config->gcSections = args.hasFlag(OPT_gc_sections, OPT_no_gc_sections, false); 835 config->gnuUnique = args.hasFlag(OPT_gnu_unique, OPT_no_gnu_unique, true); 836 config->gdbIndex = args.hasFlag(OPT_gdb_index, OPT_no_gdb_index, false); 837 config->icf = getICF(args); 838 config->ignoreDataAddressEquality = 839 args.hasArg(OPT_ignore_data_address_equality); 840 config->ignoreFunctionAddressEquality = 841 args.hasArg(OPT_ignore_function_address_equality); 842 config->init = args.getLastArgValue(OPT_init, "_init"); 843 config->ltoAAPipeline = args.getLastArgValue(OPT_lto_aa_pipeline); 844 config->ltoCSProfileGenerate = args.hasArg(OPT_lto_cs_profile_generate); 845 config->ltoCSProfileFile = args.getLastArgValue(OPT_lto_cs_profile_file); 846 config->ltoDebugPassManager = args.hasArg(OPT_lto_debug_pass_manager); 847 config->ltoNewPassManager = args.hasArg(OPT_lto_new_pass_manager); 848 config->ltoNewPmPasses = args.getLastArgValue(OPT_lto_newpm_passes); 849 config->ltoo = args::getInteger(args, OPT_lto_O, 2); 850 config->ltoObjPath = args.getLastArgValue(OPT_plugin_opt_obj_path_eq); 851 config->ltoPartitions = args::getInteger(args, OPT_lto_partitions, 1); 852 config->ltoSampleProfile = args.getLastArgValue(OPT_lto_sample_profile); 853 config->mapFile = args.getLastArgValue(OPT_Map); 854 config->mipsGotSize = args::getInteger(args, OPT_mips_got_size, 0xfff0); 855 config->mergeArmExidx = 856 args.hasFlag(OPT_merge_exidx_entries, OPT_no_merge_exidx_entries, true); 857 config->nmagic = args.hasFlag(OPT_nmagic, OPT_no_nmagic, false); 858 config->noinhibitExec = args.hasArg(OPT_noinhibit_exec); 859 config->nostdlib = args.hasArg(OPT_nostdlib); 860 config->oFormatBinary = isOutputFormatBinary(args); 861 config->omagic = args.hasFlag(OPT_omagic, OPT_no_omagic, false); 862 config->optRemarksFilename = args.getLastArgValue(OPT_opt_remarks_filename); 863 config->optRemarksPasses = args.getLastArgValue(OPT_opt_remarks_passes); 864 config->optRemarksWithHotness = args.hasArg(OPT_opt_remarks_with_hotness); 865 config->optRemarksFormat = args.getLastArgValue(OPT_opt_remarks_format); 866 config->optimize = args::getInteger(args, OPT_O, 1); 867 config->orphanHandling = getOrphanHandling(args); 868 config->outputFile = args.getLastArgValue(OPT_o); 869 config->pacPlt = args.hasArg(OPT_pac_plt); 870 config->pie = args.hasFlag(OPT_pie, OPT_no_pie, false); 871 config->printIcfSections = 872 args.hasFlag(OPT_print_icf_sections, OPT_no_print_icf_sections, false); 873 config->printGcSections = 874 args.hasFlag(OPT_print_gc_sections, OPT_no_print_gc_sections, false); 875 config->printSymbolOrder = 876 args.getLastArgValue(OPT_print_symbol_order); 877 config->rpath = getRpath(args); 878 config->relocatable = args.hasArg(OPT_relocatable); 879 config->saveTemps = args.hasArg(OPT_save_temps); 880 config->searchPaths = args::getStrings(args, OPT_library_path); 881 config->sectionStartMap = getSectionStartMap(args); 882 config->shared = args.hasArg(OPT_shared); 883 config->singleRoRx = args.hasArg(OPT_no_rosegment); 884 config->soName = args.getLastArgValue(OPT_soname); 885 config->sortSection = getSortSection(args); 886 config->splitStackAdjustSize = args::getInteger(args, OPT_split_stack_adjust_size, 16384); 887 config->strip = getStrip(args); 888 config->sysroot = args.getLastArgValue(OPT_sysroot); 889 config->target1Rel = args.hasFlag(OPT_target1_rel, OPT_target1_abs, false); 890 config->target2 = getTarget2(args); 891 config->thinLTOCacheDir = args.getLastArgValue(OPT_thinlto_cache_dir); 892 config->thinLTOCachePolicy = CHECK( 893 parseCachePruningPolicy(args.getLastArgValue(OPT_thinlto_cache_policy)), 894 "--thinlto-cache-policy: invalid cache policy"); 895 config->thinLTOEmitImportsFiles = 896 args.hasArg(OPT_plugin_opt_thinlto_emit_imports_files); 897 config->thinLTOIndexOnly = args.hasArg(OPT_plugin_opt_thinlto_index_only) || 898 args.hasArg(OPT_plugin_opt_thinlto_index_only_eq); 899 config->thinLTOIndexOnlyArg = 900 args.getLastArgValue(OPT_plugin_opt_thinlto_index_only_eq); 901 config->thinLTOJobs = args::getInteger(args, OPT_thinlto_jobs, -1u); 902 config->thinLTOObjectSuffixReplace = 903 getOldNewOptions(args, OPT_plugin_opt_thinlto_object_suffix_replace_eq); 904 config->thinLTOPrefixReplace = 905 getOldNewOptions(args, OPT_plugin_opt_thinlto_prefix_replace_eq); 906 config->trace = args.hasArg(OPT_trace); 907 config->undefined = args::getStrings(args, OPT_undefined); 908 config->undefinedVersion = 909 args.hasFlag(OPT_undefined_version, OPT_no_undefined_version, true); 910 config->useAndroidRelrTags = args.hasFlag( 911 OPT_use_android_relr_tags, OPT_no_use_android_relr_tags, false); 912 config->unresolvedSymbols = getUnresolvedSymbolPolicy(args); 913 config->warnBackrefs = 914 args.hasFlag(OPT_warn_backrefs, OPT_no_warn_backrefs, false); 915 config->warnCommon = args.hasFlag(OPT_warn_common, OPT_no_warn_common, false); 916 config->warnIfuncTextrel = 917 args.hasFlag(OPT_warn_ifunc_textrel, OPT_no_warn_ifunc_textrel, false); 918 config->warnSymbolOrdering = 919 args.hasFlag(OPT_warn_symbol_ordering, OPT_no_warn_symbol_ordering, true); 920 config->zCombreloc = getZFlag(args, "combreloc", "nocombreloc", true); 921 config->zCopyreloc = getZFlag(args, "copyreloc", "nocopyreloc", true); 922 config->zExecstack = getZFlag(args, "execstack", "noexecstack", false); 923 config->zGlobal = hasZOption(args, "global"); 924 config->zHazardplt = hasZOption(args, "hazardplt"); 925 config->zIfuncNoplt = hasZOption(args, "ifunc-noplt"); 926 config->zInitfirst = hasZOption(args, "initfirst"); 927 config->zInterpose = hasZOption(args, "interpose"); 928 config->zKeepTextSectionPrefix = getZFlag( 929 args, "keep-text-section-prefix", "nokeep-text-section-prefix", false); 930 config->zNodefaultlib = hasZOption(args, "nodefaultlib"); 931 config->zNodelete = hasZOption(args, "nodelete"); 932 config->zNodlopen = hasZOption(args, "nodlopen"); 933 config->zNow = getZFlag(args, "now", "lazy", false); 934 config->zOrigin = hasZOption(args, "origin"); 935 config->zRelro = getZFlag(args, "relro", "norelro", true); 936 config->zRetpolineplt = hasZOption(args, "retpolineplt"); 937 config->zRodynamic = hasZOption(args, "rodynamic"); 938 config->zStackSize = args::getZOptionValue(args, OPT_z, "stack-size", 0); 939 config->zText = getZFlag(args, "text", "notext", true); 940 config->zWxneeded = hasZOption(args, "wxneeded"); 941 942 // Parse LTO options. 943 if (auto *arg = args.getLastArg(OPT_plugin_opt_mcpu_eq)) 944 parseClangOption(saver.save("-mcpu=" + StringRef(arg->getValue())), 945 arg->getSpelling()); 946 947 for (auto *arg : args.filtered(OPT_plugin_opt)) 948 parseClangOption(arg->getValue(), arg->getSpelling()); 949 950 // Parse -mllvm options. 951 for (auto *arg : args.filtered(OPT_mllvm)) 952 parseClangOption(arg->getValue(), arg->getSpelling()); 953 954 if (config->ltoo > 3) 955 error("invalid optimization level for LTO: " + Twine(config->ltoo)); 956 if (config->ltoPartitions == 0) 957 error("--lto-partitions: number of threads must be > 0"); 958 if (config->thinLTOJobs == 0) 959 error("--thinlto-jobs: number of threads must be > 0"); 960 961 if (config->splitStackAdjustSize < 0) 962 error("--split-stack-adjust-size: size must be >= 0"); 963 964 // Parse ELF{32,64}{LE,BE} and CPU type. 965 if (auto *arg = args.getLastArg(OPT_m)) { 966 StringRef s = arg->getValue(); 967 std::tie(config->ekind, config->emachine, config->osabi) = 968 parseEmulation(s); 969 config->mipsN32Abi = 970 (s.startswith("elf32btsmipn32") || s.startswith("elf32ltsmipn32")); 971 config->emulation = s; 972 } 973 974 // Parse -hash-style={sysv,gnu,both}. 975 if (auto *arg = args.getLastArg(OPT_hash_style)) { 976 StringRef s = arg->getValue(); 977 if (s == "sysv") 978 config->sysvHash = true; 979 else if (s == "gnu") 980 config->gnuHash = true; 981 else if (s == "both") 982 config->sysvHash = config->gnuHash = true; 983 else 984 error("unknown -hash-style: " + s); 985 } 986 987 if (args.hasArg(OPT_print_map)) 988 config->mapFile = "-"; 989 990 // Page alignment can be disabled by the -n (--nmagic) and -N (--omagic). 991 // As PT_GNU_RELRO relies on Paging, do not create it when we have disabled 992 // it. 993 if (config->nmagic || config->omagic) 994 config->zRelro = false; 995 996 std::tie(config->buildId, config->buildIdVector) = getBuildId(args); 997 998 std::tie(config->androidPackDynRelocs, config->relrPackDynRelocs) = 999 getPackDynRelocs(args); 1000 1001 if (auto *arg = args.getLastArg(OPT_symbol_ordering_file)){ 1002 if (args.hasArg(OPT_call_graph_ordering_file)) 1003 error("--symbol-ordering-file and --call-graph-order-file " 1004 "may not be used together"); 1005 if (Optional<MemoryBufferRef> buffer = readFile(arg->getValue())){ 1006 config->symbolOrderingFile = getSymbolOrderingFile(*buffer); 1007 // Also need to disable CallGraphProfileSort to prevent 1008 // LLD order symbols with CGProfile 1009 config->callGraphProfileSort = false; 1010 } 1011 } 1012 1013 // If --retain-symbol-file is used, we'll keep only the symbols listed in 1014 // the file and discard all others. 1015 if (auto *arg = args.getLastArg(OPT_retain_symbols_file)) { 1016 config->defaultSymbolVersion = VER_NDX_LOCAL; 1017 if (Optional<MemoryBufferRef> buffer = readFile(arg->getValue())) 1018 for (StringRef s : args::getLines(*buffer)) 1019 config->versionScriptGlobals.push_back( 1020 {s, /*IsExternCpp*/ false, /*HasWildcard*/ false}); 1021 } 1022 1023 bool hasExportDynamic = 1024 args.hasFlag(OPT_export_dynamic, OPT_no_export_dynamic, false); 1025 1026 // Parses -dynamic-list and -export-dynamic-symbol. They make some 1027 // symbols private. Note that -export-dynamic takes precedence over them 1028 // as it says all symbols should be exported. 1029 if (!hasExportDynamic) { 1030 for (auto *arg : args.filtered(OPT_dynamic_list)) 1031 if (Optional<MemoryBufferRef> buffer = readFile(arg->getValue())) 1032 readDynamicList(*buffer); 1033 1034 for (auto *arg : args.filtered(OPT_export_dynamic_symbol)) 1035 config->dynamicList.push_back( 1036 {arg->getValue(), /*IsExternCpp*/ false, /*HasWildcard*/ false}); 1037 } 1038 1039 // If --export-dynamic-symbol=foo is given and symbol foo is defined in 1040 // an object file in an archive file, that object file should be pulled 1041 // out and linked. (It doesn't have to behave like that from technical 1042 // point of view, but this is needed for compatibility with GNU.) 1043 for (auto *arg : args.filtered(OPT_export_dynamic_symbol)) 1044 config->undefined.push_back(arg->getValue()); 1045 1046 for (auto *arg : args.filtered(OPT_version_script)) 1047 if (Optional<std::string> path = searchScript(arg->getValue())) { 1048 if (Optional<MemoryBufferRef> buffer = readFile(*path)) 1049 readVersionScript(*buffer); 1050 } else { 1051 error(Twine("cannot find version script ") + arg->getValue()); 1052 } 1053 } 1054 1055 // Some Config members do not directly correspond to any particular 1056 // command line options, but computed based on other Config values. 1057 // This function initialize such members. See Config.h for the details 1058 // of these values. 1059 static void setConfigs(opt::InputArgList &args) { 1060 ELFKind k = config->ekind; 1061 uint16_t m = config->emachine; 1062 1063 config->copyRelocs = (config->relocatable || config->emitRelocs); 1064 config->is64 = (k == ELF64LEKind || k == ELF64BEKind); 1065 config->isLE = (k == ELF32LEKind || k == ELF64LEKind); 1066 config->endianness = config->isLE ? endianness::little : endianness::big; 1067 config->isMips64EL = (k == ELF64LEKind && m == EM_MIPS); 1068 config->isPic = config->pie || config->shared; 1069 config->picThunk = args.hasArg(OPT_pic_veneer, config->isPic); 1070 config->wordsize = config->is64 ? 8 : 4; 1071 1072 // ELF defines two different ways to store relocation addends as shown below: 1073 // 1074 // Rel: Addends are stored to the location where relocations are applied. 1075 // Rela: Addends are stored as part of relocation entry. 1076 // 1077 // In other words, Rela makes it easy to read addends at the price of extra 1078 // 4 or 8 byte for each relocation entry. We don't know why ELF defined two 1079 // different mechanisms in the first place, but this is how the spec is 1080 // defined. 1081 // 1082 // You cannot choose which one, Rel or Rela, you want to use. Instead each 1083 // ABI defines which one you need to use. The following expression expresses 1084 // that. 1085 config->isRela = m == EM_AARCH64 || m == EM_AMDGPU || m == EM_HEXAGON || 1086 m == EM_PPC || m == EM_PPC64 || m == EM_RISCV || 1087 m == EM_X86_64; 1088 1089 // If the output uses REL relocations we must store the dynamic relocation 1090 // addends to the output sections. We also store addends for RELA relocations 1091 // if --apply-dynamic-relocs is used. 1092 // We default to not writing the addends when using RELA relocations since 1093 // any standard conforming tool can find it in r_addend. 1094 config->writeAddends = args.hasFlag(OPT_apply_dynamic_relocs, 1095 OPT_no_apply_dynamic_relocs, false) || 1096 !config->isRela; 1097 1098 config->tocOptimize = 1099 args.hasFlag(OPT_toc_optimize, OPT_no_toc_optimize, m == EM_PPC64); 1100 } 1101 1102 // Returns a value of "-format" option. 1103 static bool isFormatBinary(StringRef s) { 1104 if (s == "binary") 1105 return true; 1106 if (s == "elf" || s == "default") 1107 return false; 1108 error("unknown -format value: " + s + 1109 " (supported formats: elf, default, binary)"); 1110 return false; 1111 } 1112 1113 void LinkerDriver::createFiles(opt::InputArgList &args) { 1114 // For --{push,pop}-state. 1115 std::vector<std::tuple<bool, bool, bool>> stack; 1116 1117 // Iterate over argv to process input files and positional arguments. 1118 for (auto *arg : args) { 1119 switch (arg->getOption().getID()) { 1120 case OPT_library: 1121 addLibrary(arg->getValue()); 1122 break; 1123 case OPT_INPUT: 1124 addFile(arg->getValue(), /*withLOption=*/false); 1125 break; 1126 case OPT_defsym: { 1127 StringRef from; 1128 StringRef to; 1129 std::tie(from, to) = StringRef(arg->getValue()).split('='); 1130 if (from.empty() || to.empty()) 1131 error("-defsym: syntax error: " + StringRef(arg->getValue())); 1132 else 1133 readDefsym(from, MemoryBufferRef(to, "-defsym")); 1134 break; 1135 } 1136 case OPT_script: 1137 if (Optional<std::string> path = searchScript(arg->getValue())) { 1138 if (Optional<MemoryBufferRef> mb = readFile(*path)) 1139 readLinkerScript(*mb); 1140 break; 1141 } 1142 error(Twine("cannot find linker script ") + arg->getValue()); 1143 break; 1144 case OPT_as_needed: 1145 config->asNeeded = true; 1146 break; 1147 case OPT_format: 1148 config->formatBinary = isFormatBinary(arg->getValue()); 1149 break; 1150 case OPT_no_as_needed: 1151 config->asNeeded = false; 1152 break; 1153 case OPT_Bstatic: 1154 case OPT_omagic: 1155 case OPT_nmagic: 1156 config->isStatic = true; 1157 break; 1158 case OPT_Bdynamic: 1159 config->isStatic = false; 1160 break; 1161 case OPT_whole_archive: 1162 inWholeArchive = true; 1163 break; 1164 case OPT_no_whole_archive: 1165 inWholeArchive = false; 1166 break; 1167 case OPT_just_symbols: 1168 if (Optional<MemoryBufferRef> mb = readFile(arg->getValue())) { 1169 files.push_back(createObjectFile(*mb)); 1170 files.back()->justSymbols = true; 1171 } 1172 break; 1173 case OPT_start_group: 1174 if (InputFile::isInGroup) 1175 error("nested --start-group"); 1176 InputFile::isInGroup = true; 1177 break; 1178 case OPT_end_group: 1179 if (!InputFile::isInGroup) 1180 error("stray --end-group"); 1181 InputFile::isInGroup = false; 1182 ++InputFile::nextGroupId; 1183 break; 1184 case OPT_start_lib: 1185 if (inLib) 1186 error("nested --start-lib"); 1187 if (InputFile::isInGroup) 1188 error("may not nest --start-lib in --start-group"); 1189 inLib = true; 1190 InputFile::isInGroup = true; 1191 break; 1192 case OPT_end_lib: 1193 if (!inLib) 1194 error("stray --end-lib"); 1195 inLib = false; 1196 InputFile::isInGroup = false; 1197 ++InputFile::nextGroupId; 1198 break; 1199 case OPT_push_state: 1200 stack.emplace_back(config->asNeeded, config->isStatic, inWholeArchive); 1201 break; 1202 case OPT_pop_state: 1203 if (stack.empty()) { 1204 error("unbalanced --push-state/--pop-state"); 1205 break; 1206 } 1207 std::tie(config->asNeeded, config->isStatic, inWholeArchive) = stack.back(); 1208 stack.pop_back(); 1209 break; 1210 } 1211 } 1212 1213 if (files.empty() && errorCount() == 0) 1214 error("no input files"); 1215 } 1216 1217 // If -m <machine_type> was not given, infer it from object files. 1218 void LinkerDriver::inferMachineType() { 1219 if (config->ekind != ELFNoneKind) 1220 return; 1221 1222 for (InputFile *f : files) { 1223 if (f->ekind == ELFNoneKind) 1224 continue; 1225 config->ekind = f->ekind; 1226 config->emachine = f->emachine; 1227 config->osabi = f->osabi; 1228 config->mipsN32Abi = config->emachine == EM_MIPS && isMipsN32Abi(f); 1229 return; 1230 } 1231 error("target emulation unknown: -m or at least one .o file required"); 1232 } 1233 1234 // Parse -z max-page-size=<value>. The default value is defined by 1235 // each target. 1236 static uint64_t getMaxPageSize(opt::InputArgList &args) { 1237 uint64_t val = args::getZOptionValue(args, OPT_z, "max-page-size", 1238 target->defaultMaxPageSize); 1239 if (!isPowerOf2_64(val)) 1240 error("max-page-size: value isn't a power of 2"); 1241 if (config->nmagic || config->omagic) { 1242 if (val != target->defaultMaxPageSize) 1243 warn("-z max-page-size set, but paging disabled by omagic or nmagic"); 1244 return 1; 1245 } 1246 return val; 1247 } 1248 1249 // Parse -z common-page-size=<value>. The default value is defined by 1250 // each target. 1251 static uint64_t getCommonPageSize(opt::InputArgList &args) { 1252 uint64_t val = args::getZOptionValue(args, OPT_z, "common-page-size", 1253 target->defaultCommonPageSize); 1254 if (!isPowerOf2_64(val)) 1255 error("common-page-size: value isn't a power of 2"); 1256 if (config->nmagic || config->omagic) { 1257 if (val != target->defaultCommonPageSize) 1258 warn("-z common-page-size set, but paging disabled by omagic or nmagic"); 1259 return 1; 1260 } 1261 // commonPageSize can't be larger than maxPageSize. 1262 if (val > config->maxPageSize) 1263 val = config->maxPageSize; 1264 return val; 1265 } 1266 1267 // Parses -image-base option. 1268 static Optional<uint64_t> getImageBase(opt::InputArgList &args) { 1269 // Because we are using "Config->maxPageSize" here, this function has to be 1270 // called after the variable is initialized. 1271 auto *arg = args.getLastArg(OPT_image_base); 1272 if (!arg) 1273 return None; 1274 1275 StringRef s = arg->getValue(); 1276 uint64_t v; 1277 if (!to_integer(s, v)) { 1278 error("-image-base: number expected, but got " + s); 1279 return 0; 1280 } 1281 if ((v % config->maxPageSize) != 0) 1282 warn("-image-base: address isn't multiple of page size: " + s); 1283 return v; 1284 } 1285 1286 // Parses `--exclude-libs=lib,lib,...`. 1287 // The library names may be delimited by commas or colons. 1288 static DenseSet<StringRef> getExcludeLibs(opt::InputArgList &args) { 1289 DenseSet<StringRef> ret; 1290 for (auto *arg : args.filtered(OPT_exclude_libs)) { 1291 StringRef s = arg->getValue(); 1292 for (;;) { 1293 size_t pos = s.find_first_of(",:"); 1294 if (pos == StringRef::npos) 1295 break; 1296 ret.insert(s.substr(0, pos)); 1297 s = s.substr(pos + 1); 1298 } 1299 ret.insert(s); 1300 } 1301 return ret; 1302 } 1303 1304 // Handles the -exclude-libs option. If a static library file is specified 1305 // by the -exclude-libs option, all public symbols from the archive become 1306 // private unless otherwise specified by version scripts or something. 1307 // A special library name "ALL" means all archive files. 1308 // 1309 // This is not a popular option, but some programs such as bionic libc use it. 1310 static void excludeLibs(opt::InputArgList &args) { 1311 DenseSet<StringRef> libs = getExcludeLibs(args); 1312 bool all = libs.count("ALL"); 1313 1314 auto visit = [&](InputFile *file) { 1315 if (!file->archiveName.empty()) 1316 if (all || libs.count(path::filename(file->archiveName))) 1317 for (Symbol *sym : file->getSymbols()) 1318 if (!sym->isLocal() && sym->file == file) 1319 sym->versionId = VER_NDX_LOCAL; 1320 }; 1321 1322 for (InputFile *file : objectFiles) 1323 visit(file); 1324 1325 for (BitcodeFile *file : bitcodeFiles) 1326 visit(file); 1327 } 1328 1329 // Force Sym to be entered in the output. Used for -u or equivalent. 1330 static void handleUndefined(Symbol *sym) { 1331 // Since a symbol may not be used inside the program, LTO may 1332 // eliminate it. Mark the symbol as "used" to prevent it. 1333 sym->isUsedInRegularObj = true; 1334 1335 if (sym->isLazy()) 1336 sym->fetch(); 1337 } 1338 1339 // As an extention to GNU linkers, lld supports a variant of `-u` 1340 // which accepts wildcard patterns. All symbols that match a given 1341 // pattern are handled as if they were given by `-u`. 1342 static void handleUndefinedGlob(StringRef arg) { 1343 Expected<GlobPattern> pat = GlobPattern::create(arg); 1344 if (!pat) { 1345 error("--undefined-glob: " + toString(pat.takeError())); 1346 return; 1347 } 1348 1349 std::vector<Symbol *> syms; 1350 symtab->forEachSymbol([&](Symbol *sym) { 1351 // Calling Sym->fetch() from here is not safe because it may 1352 // add new symbols to the symbol table, invalidating the 1353 // current iterator. So we just keep a note. 1354 if (pat->match(sym->getName())) 1355 syms.push_back(sym); 1356 }); 1357 1358 for (Symbol *sym : syms) 1359 handleUndefined(sym); 1360 } 1361 1362 static void handleLibcall(StringRef name) { 1363 Symbol *sym = symtab->find(name); 1364 if (!sym || !sym->isLazy()) 1365 return; 1366 1367 MemoryBufferRef mb; 1368 if (auto *lo = dyn_cast<LazyObject>(sym)) 1369 mb = lo->file->mb; 1370 else 1371 mb = cast<LazyArchive>(sym)->getMemberBuffer(); 1372 1373 if (isBitcode(mb)) 1374 sym->fetch(); 1375 } 1376 1377 // Replaces common symbols with defined symbols reside in .bss sections. 1378 // This function is called after all symbol names are resolved. As a 1379 // result, the passes after the symbol resolution won't see any 1380 // symbols of type CommonSymbol. 1381 static void replaceCommonSymbols() { 1382 symtab->forEachSymbol([](Symbol *sym) { 1383 auto *s = dyn_cast<CommonSymbol>(sym); 1384 if (!s) 1385 return; 1386 1387 auto *bss = make<BssSection>("COMMON", s->size, s->alignment); 1388 bss->file = s->file; 1389 bss->markDead(); 1390 inputSections.push_back(bss); 1391 s->replace(Defined{s->file, s->getName(), s->binding, s->stOther, s->type, 1392 /*value=*/0, s->size, bss}); 1393 }); 1394 } 1395 1396 // If all references to a DSO happen to be weak, the DSO is not added 1397 // to DT_NEEDED. If that happens, we need to eliminate shared symbols 1398 // created from the DSO. Otherwise, they become dangling references 1399 // that point to a non-existent DSO. 1400 static void demoteSharedSymbols() { 1401 symtab->forEachSymbol([](Symbol *sym) { 1402 auto *s = dyn_cast<SharedSymbol>(sym); 1403 if (!s || s->getFile().isNeeded) 1404 return; 1405 1406 bool used = s->used; 1407 s->replace(Undefined{nullptr, s->getName(), STB_WEAK, s->stOther, s->type}); 1408 s->used = used; 1409 }); 1410 } 1411 1412 // The section referred to by `s` is considered address-significant. Set the 1413 // keepUnique flag on the section if appropriate. 1414 static void markAddrsig(Symbol *s) { 1415 if (auto *d = dyn_cast_or_null<Defined>(s)) 1416 if (d->section) 1417 // We don't need to keep text sections unique under --icf=all even if they 1418 // are address-significant. 1419 if (config->icf == ICFLevel::Safe || !(d->section->flags & SHF_EXECINSTR)) 1420 d->section->keepUnique = true; 1421 } 1422 1423 // Record sections that define symbols mentioned in --keep-unique <symbol> 1424 // and symbols referred to by address-significance tables. These sections are 1425 // ineligible for ICF. 1426 template <class ELFT> 1427 static void findKeepUniqueSections(opt::InputArgList &args) { 1428 for (auto *arg : args.filtered(OPT_keep_unique)) { 1429 StringRef name = arg->getValue(); 1430 auto *d = dyn_cast_or_null<Defined>(symtab->find(name)); 1431 if (!d || !d->section) { 1432 warn("could not find symbol " + name + " to keep unique"); 1433 continue; 1434 } 1435 d->section->keepUnique = true; 1436 } 1437 1438 // --icf=all --ignore-data-address-equality means that we can ignore 1439 // the dynsym and address-significance tables entirely. 1440 if (config->icf == ICFLevel::All && config->ignoreDataAddressEquality) 1441 return; 1442 1443 // Symbols in the dynsym could be address-significant in other executables 1444 // or DSOs, so we conservatively mark them as address-significant. 1445 symtab->forEachSymbol([&](Symbol *sym) { 1446 if (sym->includeInDynsym()) 1447 markAddrsig(sym); 1448 }); 1449 1450 // Visit the address-significance table in each object file and mark each 1451 // referenced symbol as address-significant. 1452 for (InputFile *f : objectFiles) { 1453 auto *obj = cast<ObjFile<ELFT>>(f); 1454 ArrayRef<Symbol *> syms = obj->getSymbols(); 1455 if (obj->addrsigSec) { 1456 ArrayRef<uint8_t> contents = 1457 check(obj->getObj().getSectionContents(obj->addrsigSec)); 1458 const uint8_t *cur = contents.begin(); 1459 while (cur != contents.end()) { 1460 unsigned size; 1461 const char *err; 1462 uint64_t symIndex = decodeULEB128(cur, &size, contents.end(), &err); 1463 if (err) 1464 fatal(toString(f) + ": could not decode addrsig section: " + err); 1465 markAddrsig(syms[symIndex]); 1466 cur += size; 1467 } 1468 } else { 1469 // If an object file does not have an address-significance table, 1470 // conservatively mark all of its symbols as address-significant. 1471 for (Symbol *s : syms) 1472 markAddrsig(s); 1473 } 1474 } 1475 } 1476 1477 // This function reads a symbol partition specification section. These sections 1478 // are used to control which partition a symbol is allocated to. See 1479 // https://lld.llvm.org/Partitions.html for more details on partitions. 1480 template <typename ELFT> 1481 static void readSymbolPartitionSection(InputSectionBase *s) { 1482 // Read the relocation that refers to the partition's entry point symbol. 1483 Symbol *sym; 1484 if (s->areRelocsRela) 1485 sym = &s->getFile<ELFT>()->getRelocTargetSym(s->template relas<ELFT>()[0]); 1486 else 1487 sym = &s->getFile<ELFT>()->getRelocTargetSym(s->template rels<ELFT>()[0]); 1488 if (!isa<Defined>(sym) || !sym->includeInDynsym()) 1489 return; 1490 1491 StringRef partName = reinterpret_cast<const char *>(s->data().data()); 1492 for (Partition &part : partitions) { 1493 if (part.name == partName) { 1494 sym->partition = part.getNumber(); 1495 return; 1496 } 1497 } 1498 1499 // Forbid partitions from being used on incompatible targets, and forbid them 1500 // from being used together with various linker features that assume a single 1501 // set of output sections. 1502 if (script->hasSectionsCommand) 1503 error(toString(s->file) + 1504 ": partitions cannot be used with the SECTIONS command"); 1505 if (script->hasPhdrsCommands()) 1506 error(toString(s->file) + 1507 ": partitions cannot be used with the PHDRS command"); 1508 if (!config->sectionStartMap.empty()) 1509 error(toString(s->file) + ": partitions cannot be used with " 1510 "--section-start, -Ttext, -Tdata or -Tbss"); 1511 if (config->emachine == EM_MIPS) 1512 error(toString(s->file) + ": partitions cannot be used on this target"); 1513 1514 // Impose a limit of no more than 254 partitions. This limit comes from the 1515 // sizes of the Partition fields in InputSectionBase and Symbol, as well as 1516 // the amount of space devoted to the partition number in RankFlags. 1517 if (partitions.size() == 254) 1518 fatal("may not have more than 254 partitions"); 1519 1520 partitions.emplace_back(); 1521 Partition &newPart = partitions.back(); 1522 newPart.name = partName; 1523 sym->partition = newPart.getNumber(); 1524 } 1525 1526 static Symbol *addUndefined(StringRef name) { 1527 return symtab->addSymbol( 1528 Undefined{nullptr, name, STB_GLOBAL, STV_DEFAULT, 0}); 1529 } 1530 1531 // This function is where all the optimizations of link-time 1532 // optimization takes place. When LTO is in use, some input files are 1533 // not in native object file format but in the LLVM bitcode format. 1534 // This function compiles bitcode files into a few big native files 1535 // using LLVM functions and replaces bitcode symbols with the results. 1536 // Because all bitcode files that the program consists of are passed to 1537 // the compiler at once, it can do a whole-program optimization. 1538 template <class ELFT> void LinkerDriver::compileBitcodeFiles() { 1539 // Compile bitcode files and replace bitcode symbols. 1540 lto.reset(new BitcodeCompiler); 1541 for (BitcodeFile *file : bitcodeFiles) 1542 lto->add(*file); 1543 1544 for (InputFile *file : lto->compile()) { 1545 auto *obj = cast<ObjFile<ELFT>>(file); 1546 obj->parse(/*ignoreComdats=*/true); 1547 for (Symbol *sym : obj->getGlobalSymbols()) 1548 sym->parseSymbolVersion(); 1549 objectFiles.push_back(file); 1550 } 1551 } 1552 1553 // The --wrap option is a feature to rename symbols so that you can write 1554 // wrappers for existing functions. If you pass `-wrap=foo`, all 1555 // occurrences of symbol `foo` are resolved to `wrap_foo` (so, you are 1556 // expected to write `wrap_foo` function as a wrapper). The original 1557 // symbol becomes accessible as `real_foo`, so you can call that from your 1558 // wrapper. 1559 // 1560 // This data structure is instantiated for each -wrap option. 1561 struct WrappedSymbol { 1562 Symbol *sym; 1563 Symbol *real; 1564 Symbol *wrap; 1565 }; 1566 1567 // Handles -wrap option. 1568 // 1569 // This function instantiates wrapper symbols. At this point, they seem 1570 // like they are not being used at all, so we explicitly set some flags so 1571 // that LTO won't eliminate them. 1572 static std::vector<WrappedSymbol> addWrappedSymbols(opt::InputArgList &args) { 1573 std::vector<WrappedSymbol> v; 1574 DenseSet<StringRef> seen; 1575 1576 for (auto *arg : args.filtered(OPT_wrap)) { 1577 StringRef name = arg->getValue(); 1578 if (!seen.insert(name).second) 1579 continue; 1580 1581 Symbol *sym = symtab->find(name); 1582 if (!sym) 1583 continue; 1584 1585 Symbol *real = addUndefined(saver.save("__real_" + name)); 1586 Symbol *wrap = addUndefined(saver.save("__wrap_" + name)); 1587 v.push_back({sym, real, wrap}); 1588 1589 // We want to tell LTO not to inline symbols to be overwritten 1590 // because LTO doesn't know the final symbol contents after renaming. 1591 real->canInline = false; 1592 sym->canInline = false; 1593 1594 // Tell LTO not to eliminate these symbols. 1595 sym->isUsedInRegularObj = true; 1596 wrap->isUsedInRegularObj = true; 1597 } 1598 return v; 1599 } 1600 1601 // Do renaming for -wrap by updating pointers to symbols. 1602 // 1603 // When this function is executed, only InputFiles and symbol table 1604 // contain pointers to symbol objects. We visit them to replace pointers, 1605 // so that wrapped symbols are swapped as instructed by the command line. 1606 static void wrapSymbols(ArrayRef<WrappedSymbol> wrapped) { 1607 DenseMap<Symbol *, Symbol *> map; 1608 for (const WrappedSymbol &w : wrapped) { 1609 map[w.sym] = w.wrap; 1610 map[w.real] = w.sym; 1611 } 1612 1613 // Update pointers in input files. 1614 parallelForEach(objectFiles, [&](InputFile *file) { 1615 MutableArrayRef<Symbol *> syms = file->getMutableSymbols(); 1616 for (size_t i = 0, e = syms.size(); i != e; ++i) 1617 if (Symbol *s = map.lookup(syms[i])) 1618 syms[i] = s; 1619 }); 1620 1621 // Update pointers in the symbol table. 1622 for (const WrappedSymbol &w : wrapped) 1623 symtab->wrap(w.sym, w.real, w.wrap); 1624 } 1625 1626 // To enable CET (x86's hardware-assited control flow enforcement), each 1627 // source file must be compiled with -fcf-protection. Object files compiled 1628 // with the flag contain feature flags indicating that they are compatible 1629 // with CET. We enable the feature only when all object files are compatible 1630 // with CET. 1631 // 1632 // This function returns the merged feature flags. If 0, we cannot enable CET. 1633 // This is also the case with AARCH64's BTI and PAC which use the similar 1634 // GNU_PROPERTY_AARCH64_FEATURE_1_AND mechanism. 1635 // 1636 // Note that the CET-aware PLT is not implemented yet. We do error 1637 // check only. 1638 template <class ELFT> static uint32_t getAndFeatures() { 1639 if (config->emachine != EM_386 && config->emachine != EM_X86_64 && 1640 config->emachine != EM_AARCH64) 1641 return 0; 1642 1643 uint32_t ret = -1; 1644 for (InputFile *f : objectFiles) { 1645 uint32_t features = cast<ObjFile<ELFT>>(f)->andFeatures; 1646 if (config->forceBTI && !(features & GNU_PROPERTY_AARCH64_FEATURE_1_BTI)) { 1647 warn(toString(f) + ": --force-bti: file does not have BTI property"); 1648 features |= GNU_PROPERTY_AARCH64_FEATURE_1_BTI; 1649 } else if (!features && config->requireCET) 1650 error(toString(f) + ": --require-cet: file is not compatible with CET"); 1651 ret &= features; 1652 } 1653 1654 // Force enable pointer authentication Plt, we don't warn in this case as 1655 // this does not require support in the object for correctness. 1656 if (config->pacPlt) 1657 ret |= GNU_PROPERTY_AARCH64_FEATURE_1_PAC; 1658 1659 return ret; 1660 } 1661 1662 static const char *libcallRoutineNames[] = { 1663 #define HANDLE_LIBCALL(code, name) name, 1664 #include "llvm/IR/RuntimeLibcalls.def" 1665 #undef HANDLE_LIBCALL 1666 }; 1667 1668 // Do actual linking. Note that when this function is called, 1669 // all linker scripts have already been parsed. 1670 template <class ELFT> void LinkerDriver::link(opt::InputArgList &args) { 1671 // If a -hash-style option was not given, set to a default value, 1672 // which varies depending on the target. 1673 if (!args.hasArg(OPT_hash_style)) { 1674 if (config->emachine == EM_MIPS) 1675 config->sysvHash = true; 1676 else 1677 config->sysvHash = config->gnuHash = true; 1678 } 1679 1680 // Default output filename is "a.out" by the Unix tradition. 1681 if (config->outputFile.empty()) 1682 config->outputFile = "a.out"; 1683 1684 // Fail early if the output file or map file is not writable. If a user has a 1685 // long link, e.g. due to a large LTO link, they do not wish to run it and 1686 // find that it failed because there was a mistake in their command-line. 1687 if (auto e = tryCreateFile(config->outputFile)) 1688 error("cannot open output file " + config->outputFile + ": " + e.message()); 1689 if (auto e = tryCreateFile(config->mapFile)) 1690 error("cannot open map file " + config->mapFile + ": " + e.message()); 1691 if (errorCount()) 1692 return; 1693 1694 // Use default entry point name if no name was given via the command 1695 // line nor linker scripts. For some reason, MIPS entry point name is 1696 // different from others. 1697 config->warnMissingEntry = 1698 (!config->entry.empty() || (!config->shared && !config->relocatable)); 1699 if (config->entry.empty() && !config->relocatable) 1700 config->entry = (config->emachine == EM_MIPS) ? "__start" : "_start"; 1701 1702 // Handle --trace-symbol. 1703 for (auto *arg : args.filtered(OPT_trace_symbol)) 1704 symtab->insert(arg->getValue())->traced = true; 1705 1706 // Add all files to the symbol table. This will add almost all 1707 // symbols that we need to the symbol table. This process might 1708 // add files to the link, via autolinking, these files are always 1709 // appended to the Files vector. 1710 for (size_t i = 0; i < files.size(); ++i) 1711 parseFile(files[i]); 1712 1713 // Now that we have every file, we can decide if we will need a 1714 // dynamic symbol table. 1715 // We need one if we were asked to export dynamic symbols or if we are 1716 // producing a shared library. 1717 // We also need one if any shared libraries are used and for pie executables 1718 // (probably because the dynamic linker needs it). 1719 config->hasDynSymTab = 1720 !sharedFiles.empty() || config->isPic || config->exportDynamic; 1721 1722 // Some symbols (such as __ehdr_start) are defined lazily only when there 1723 // are undefined symbols for them, so we add these to trigger that logic. 1724 for (StringRef name : script->referencedSymbols) 1725 addUndefined(name); 1726 1727 // Handle the `--undefined <sym>` options. 1728 for (StringRef arg : config->undefined) 1729 if (Symbol *sym = symtab->find(arg)) 1730 handleUndefined(sym); 1731 1732 // If an entry symbol is in a static archive, pull out that file now. 1733 if (Symbol *sym = symtab->find(config->entry)) 1734 handleUndefined(sym); 1735 1736 // Handle the `--undefined-glob <pattern>` options. 1737 for (StringRef pat : args::getStrings(args, OPT_undefined_glob)) 1738 handleUndefinedGlob(pat); 1739 1740 // If any of our inputs are bitcode files, the LTO code generator may create 1741 // references to certain library functions that might not be explicit in the 1742 // bitcode file's symbol table. If any of those library functions are defined 1743 // in a bitcode file in an archive member, we need to arrange to use LTO to 1744 // compile those archive members by adding them to the link beforehand. 1745 // 1746 // However, adding all libcall symbols to the link can have undesired 1747 // consequences. For example, the libgcc implementation of 1748 // __sync_val_compare_and_swap_8 on 32-bit ARM pulls in an .init_array entry 1749 // that aborts the program if the Linux kernel does not support 64-bit 1750 // atomics, which would prevent the program from running even if it does not 1751 // use 64-bit atomics. 1752 // 1753 // Therefore, we only add libcall symbols to the link before LTO if we have 1754 // to, i.e. if the symbol's definition is in bitcode. Any other required 1755 // libcall symbols will be added to the link after LTO when we add the LTO 1756 // object file to the link. 1757 if (!bitcodeFiles.empty()) 1758 for (const char *s : libcallRoutineNames) 1759 handleLibcall(s); 1760 1761 // Return if there were name resolution errors. 1762 if (errorCount()) 1763 return; 1764 1765 // Now when we read all script files, we want to finalize order of linker 1766 // script commands, which can be not yet final because of INSERT commands. 1767 script->processInsertCommands(); 1768 1769 // We want to declare linker script's symbols early, 1770 // so that we can version them. 1771 // They also might be exported if referenced by DSOs. 1772 script->declareSymbols(); 1773 1774 // Handle the -exclude-libs option. 1775 if (args.hasArg(OPT_exclude_libs)) 1776 excludeLibs(args); 1777 1778 // Create elfHeader early. We need a dummy section in 1779 // addReservedSymbols to mark the created symbols as not absolute. 1780 Out::elfHeader = make<OutputSection>("", 0, SHF_ALLOC); 1781 Out::elfHeader->size = sizeof(typename ELFT::Ehdr); 1782 1783 // Create wrapped symbols for -wrap option. 1784 std::vector<WrappedSymbol> wrapped = addWrappedSymbols(args); 1785 1786 // We need to create some reserved symbols such as _end. Create them. 1787 if (!config->relocatable) 1788 addReservedSymbols(); 1789 1790 // Apply version scripts. 1791 // 1792 // For a relocatable output, version scripts don't make sense, and 1793 // parsing a symbol version string (e.g. dropping "@ver1" from a symbol 1794 // name "foo@ver1") rather do harm, so we don't call this if -r is given. 1795 if (!config->relocatable) 1796 symtab->scanVersionScript(); 1797 1798 // Do link-time optimization if given files are LLVM bitcode files. 1799 // This compiles bitcode files into real object files. 1800 // 1801 // With this the symbol table should be complete. After this, no new names 1802 // except a few linker-synthesized ones will be added to the symbol table. 1803 compileBitcodeFiles<ELFT>(); 1804 if (errorCount()) 1805 return; 1806 1807 // If -thinlto-index-only is given, we should create only "index 1808 // files" and not object files. Index file creation is already done 1809 // in addCombinedLTOObject, so we are done if that's the case. 1810 if (config->thinLTOIndexOnly) 1811 return; 1812 1813 // Likewise, --plugin-opt=emit-llvm is an option to make LTO create 1814 // an output file in bitcode and exit, so that you can just get a 1815 // combined bitcode file. 1816 if (config->emitLLVM) 1817 return; 1818 1819 // Apply symbol renames for -wrap. 1820 if (!wrapped.empty()) 1821 wrapSymbols(wrapped); 1822 1823 // Now that we have a complete list of input files. 1824 // Beyond this point, no new files are added. 1825 // Aggregate all input sections into one place. 1826 for (InputFile *f : objectFiles) 1827 for (InputSectionBase *s : f->getSections()) 1828 if (s && s != &InputSection::discarded) 1829 inputSections.push_back(s); 1830 for (BinaryFile *f : binaryFiles) 1831 for (InputSectionBase *s : f->getSections()) 1832 inputSections.push_back(cast<InputSection>(s)); 1833 1834 llvm::erase_if(inputSections, [](InputSectionBase *s) { 1835 if (s->type == SHT_LLVM_SYMPART) { 1836 readSymbolPartitionSection<ELFT>(s); 1837 return true; 1838 } 1839 1840 // We do not want to emit debug sections if --strip-all 1841 // or -strip-debug are given. 1842 return config->strip != StripPolicy::None && 1843 (s->name.startswith(".debug") || s->name.startswith(".zdebug")); 1844 }); 1845 1846 // Now that the number of partitions is fixed, save a pointer to the main 1847 // partition. 1848 mainPart = &partitions[0]; 1849 1850 // Read .note.gnu.property sections from input object files which 1851 // contain a hint to tweak linker's and loader's behaviors. 1852 config->andFeatures = getAndFeatures<ELFT>(); 1853 1854 // The Target instance handles target-specific stuff, such as applying 1855 // relocations or writing a PLT section. It also contains target-dependent 1856 // values such as a default image base address. 1857 target = getTarget(); 1858 1859 config->eflags = target->calcEFlags(); 1860 // maxPageSize (sometimes called abi page size) is the maximum page size that 1861 // the output can be run on. For example if the OS can use 4k or 64k page 1862 // sizes then maxPageSize must be 64k for the output to be useable on both. 1863 // All important alignment decisions must use this value. 1864 config->maxPageSize = getMaxPageSize(args); 1865 // commonPageSize is the most common page size that the output will be run on. 1866 // For example if an OS can use 4k or 64k page sizes and 4k is more common 1867 // than 64k then commonPageSize is set to 4k. commonPageSize can be used for 1868 // optimizations such as DATA_SEGMENT_ALIGN in linker scripts. LLD's use of it 1869 // is limited to writing trap instructions on the last executable segment. 1870 config->commonPageSize = getCommonPageSize(args); 1871 1872 config->imageBase = getImageBase(args); 1873 1874 if (config->emachine == EM_ARM) { 1875 // FIXME: These warnings can be removed when lld only uses these features 1876 // when the input objects have been compiled with an architecture that 1877 // supports them. 1878 if (config->armHasBlx == false) 1879 warn("lld uses blx instruction, no object with architecture supporting " 1880 "feature detected"); 1881 } 1882 1883 // This adds a .comment section containing a version string. We have to add it 1884 // before mergeSections because the .comment section is a mergeable section. 1885 if (!config->relocatable) 1886 inputSections.push_back(createCommentSection()); 1887 1888 // Replace common symbols with regular symbols. 1889 replaceCommonSymbols(); 1890 1891 // Do size optimizations: garbage collection, merging of SHF_MERGE sections 1892 // and identical code folding. 1893 splitSections<ELFT>(); 1894 markLive<ELFT>(); 1895 demoteSharedSymbols(); 1896 mergeSections(); 1897 if (config->icf != ICFLevel::None) { 1898 findKeepUniqueSections<ELFT>(args); 1899 doIcf<ELFT>(); 1900 } 1901 1902 // Read the callgraph now that we know what was gced or icfed 1903 if (config->callGraphProfileSort) { 1904 if (auto *arg = args.getLastArg(OPT_call_graph_ordering_file)) 1905 if (Optional<MemoryBufferRef> buffer = readFile(arg->getValue())) 1906 readCallGraph(*buffer); 1907 readCallGraphsFromObjectFiles<ELFT>(); 1908 } 1909 1910 // Write the result to the file. 1911 writeResult<ELFT>(); 1912 } 1913