1 //===- InputSection.cpp ---------------------------------------------------===// 2 // 3 // The LLVM Linker 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 10 #include "InputSection.h" 11 #include "Config.h" 12 #include "EhFrame.h" 13 #include "InputFiles.h" 14 #include "LinkerScript.h" 15 #include "OutputSections.h" 16 #include "Relocations.h" 17 #include "SymbolTable.h" 18 #include "Symbols.h" 19 #include "SyntheticSections.h" 20 #include "Target.h" 21 #include "Thunks.h" 22 #include "lld/Common/ErrorHandler.h" 23 #include "lld/Common/Memory.h" 24 #include "llvm/Support/Compiler.h" 25 #include "llvm/Support/Compression.h" 26 #include "llvm/Support/Endian.h" 27 #include "llvm/Support/Threading.h" 28 #include "llvm/Support/xxhash.h" 29 #include <algorithm> 30 #include <mutex> 31 #include <set> 32 #include <vector> 33 34 using namespace llvm; 35 using namespace llvm::ELF; 36 using namespace llvm::object; 37 using namespace llvm::support; 38 using namespace llvm::support::endian; 39 using namespace llvm::sys; 40 41 using namespace lld; 42 using namespace lld::elf; 43 44 std::vector<InputSectionBase *> elf::InputSections; 45 46 // Returns a string to construct an error message. 47 std::string lld::toString(const InputSectionBase *Sec) { 48 return (toString(Sec->File) + ":(" + Sec->Name + ")").str(); 49 } 50 51 template <class ELFT> 52 static ArrayRef<uint8_t> getSectionContents(ObjFile<ELFT> &File, 53 const typename ELFT::Shdr &Hdr) { 54 if (Hdr.sh_type == SHT_NOBITS) 55 return makeArrayRef<uint8_t>(nullptr, Hdr.sh_size); 56 return check(File.getObj().getSectionContents(&Hdr)); 57 } 58 59 InputSectionBase::InputSectionBase(InputFile *File, uint64_t Flags, 60 uint32_t Type, uint64_t Entsize, 61 uint32_t Link, uint32_t Info, 62 uint32_t Alignment, ArrayRef<uint8_t> Data, 63 StringRef Name, Kind SectionKind) 64 : SectionBase(SectionKind, Name, Flags, Entsize, Alignment, Type, Info, 65 Link), 66 File(File), RawData(Data) { 67 // In order to reduce memory allocation, we assume that mergeable 68 // sections are smaller than 4 GiB, which is not an unreasonable 69 // assumption as of 2017. 70 if (SectionKind == SectionBase::Merge && RawData.size() > UINT32_MAX) 71 error(toString(this) + ": section too large"); 72 73 NumRelocations = 0; 74 AreRelocsRela = false; 75 76 // The ELF spec states that a value of 0 means the section has 77 // no alignment constraits. 78 uint32_t V = std::max<uint64_t>(Alignment, 1); 79 if (!isPowerOf2_64(V)) 80 fatal(toString(File) + ": section sh_addralign is not a power of 2"); 81 this->Alignment = V; 82 83 // In ELF, each section can be compressed by zlib, and if compressed, 84 // section name may be mangled by appending "z" (e.g. ".zdebug_info"). 85 // If that's the case, demangle section name so that we can handle a 86 // section as if it weren't compressed. 87 if ((Flags & SHF_COMPRESSED) || Name.startswith(".zdebug")) { 88 if (!zlib::isAvailable()) 89 error(toString(File) + ": contains a compressed section, " + 90 "but zlib is not available"); 91 parseCompressedHeader(); 92 } 93 } 94 95 // Drop SHF_GROUP bit unless we are producing a re-linkable object file. 96 // SHF_GROUP is a marker that a section belongs to some comdat group. 97 // That flag doesn't make sense in an executable. 98 static uint64_t getFlags(uint64_t Flags) { 99 Flags &= ~(uint64_t)SHF_INFO_LINK; 100 if (!Config->Relocatable) 101 Flags &= ~(uint64_t)SHF_GROUP; 102 return Flags; 103 } 104 105 // GNU assembler 2.24 and LLVM 4.0.0's MC (the newest release as of 106 // March 2017) fail to infer section types for sections starting with 107 // ".init_array." or ".fini_array.". They set SHT_PROGBITS instead of 108 // SHF_INIT_ARRAY. As a result, the following assembler directive 109 // creates ".init_array.100" with SHT_PROGBITS, for example. 110 // 111 // .section .init_array.100, "aw" 112 // 113 // This function forces SHT_{INIT,FINI}_ARRAY so that we can handle 114 // incorrect inputs as if they were correct from the beginning. 115 static uint64_t getType(uint64_t Type, StringRef Name) { 116 if (Type == SHT_PROGBITS && Name.startswith(".init_array.")) 117 return SHT_INIT_ARRAY; 118 if (Type == SHT_PROGBITS && Name.startswith(".fini_array.")) 119 return SHT_FINI_ARRAY; 120 return Type; 121 } 122 123 template <class ELFT> 124 InputSectionBase::InputSectionBase(ObjFile<ELFT> &File, 125 const typename ELFT::Shdr &Hdr, 126 StringRef Name, Kind SectionKind) 127 : InputSectionBase(&File, getFlags(Hdr.sh_flags), 128 getType(Hdr.sh_type, Name), Hdr.sh_entsize, Hdr.sh_link, 129 Hdr.sh_info, Hdr.sh_addralign, 130 getSectionContents(File, Hdr), Name, SectionKind) { 131 // We reject object files having insanely large alignments even though 132 // they are allowed by the spec. I think 4GB is a reasonable limitation. 133 // We might want to relax this in the future. 134 if (Hdr.sh_addralign > UINT32_MAX) 135 fatal(toString(&File) + ": section sh_addralign is too large"); 136 } 137 138 size_t InputSectionBase::getSize() const { 139 if (auto *S = dyn_cast<SyntheticSection>(this)) 140 return S->getSize(); 141 if (UncompressedSize >= 0) 142 return UncompressedSize; 143 return RawData.size(); 144 } 145 146 void InputSectionBase::uncompress() const { 147 size_t Size = UncompressedSize; 148 UncompressedBuf.reset(new char[Size]); 149 150 if (Error E = 151 zlib::uncompress(toStringRef(RawData), UncompressedBuf.get(), Size)) 152 fatal(toString(this) + 153 ": uncompress failed: " + llvm::toString(std::move(E))); 154 RawData = makeArrayRef((uint8_t *)UncompressedBuf.get(), Size); 155 } 156 157 uint64_t InputSectionBase::getOffsetInFile() const { 158 const uint8_t *FileStart = (const uint8_t *)File->MB.getBufferStart(); 159 const uint8_t *SecStart = data().begin(); 160 return SecStart - FileStart; 161 } 162 163 uint64_t SectionBase::getOffset(uint64_t Offset) const { 164 switch (kind()) { 165 case Output: { 166 auto *OS = cast<OutputSection>(this); 167 // For output sections we treat offset -1 as the end of the section. 168 return Offset == uint64_t(-1) ? OS->Size : Offset; 169 } 170 case Regular: 171 case Synthetic: 172 return cast<InputSection>(this)->getOffset(Offset); 173 case EHFrame: 174 // The file crtbeginT.o has relocations pointing to the start of an empty 175 // .eh_frame that is known to be the first in the link. It does that to 176 // identify the start of the output .eh_frame. 177 return Offset; 178 case Merge: 179 const MergeInputSection *MS = cast<MergeInputSection>(this); 180 if (InputSection *IS = MS->getParent()) 181 return IS->getOffset(MS->getParentOffset(Offset)); 182 return MS->getParentOffset(Offset); 183 } 184 llvm_unreachable("invalid section kind"); 185 } 186 187 uint64_t SectionBase::getVA(uint64_t Offset) const { 188 const OutputSection *Out = getOutputSection(); 189 return (Out ? Out->Addr : 0) + getOffset(Offset); 190 } 191 192 OutputSection *SectionBase::getOutputSection() { 193 InputSection *Sec; 194 if (auto *IS = dyn_cast<InputSection>(this)) 195 Sec = IS; 196 else if (auto *MS = dyn_cast<MergeInputSection>(this)) 197 Sec = MS->getParent(); 198 else if (auto *EH = dyn_cast<EhInputSection>(this)) 199 Sec = EH->getParent(); 200 else 201 return cast<OutputSection>(this); 202 return Sec ? Sec->getParent() : nullptr; 203 } 204 205 // When a section is compressed, `RawData` consists with a header followed 206 // by zlib-compressed data. This function parses a header to initialize 207 // `UncompressedSize` member and remove the header from `RawData`. 208 void InputSectionBase::parseCompressedHeader() { 209 typedef typename ELF64LE::Chdr Chdr64; 210 typedef typename ELF32LE::Chdr Chdr32; 211 212 // Old-style header 213 if (Name.startswith(".zdebug")) { 214 if (!toStringRef(RawData).startswith("ZLIB")) { 215 error(toString(this) + ": corrupted compressed section header"); 216 return; 217 } 218 RawData = RawData.slice(4); 219 220 if (RawData.size() < 8) { 221 error(toString(this) + ": corrupted compressed section header"); 222 return; 223 } 224 225 UncompressedSize = read64be(RawData.data()); 226 RawData = RawData.slice(8); 227 228 // Restore the original section name. 229 // (e.g. ".zdebug_info" -> ".debug_info") 230 Name = Saver.save("." + Name.substr(2)); 231 return; 232 } 233 234 assert(Flags & SHF_COMPRESSED); 235 Flags &= ~(uint64_t)SHF_COMPRESSED; 236 237 // New-style 64-bit header 238 if (Config->Is64) { 239 if (RawData.size() < sizeof(Chdr64)) { 240 error(toString(this) + ": corrupted compressed section"); 241 return; 242 } 243 244 auto *Hdr = reinterpret_cast<const Chdr64 *>(RawData.data()); 245 if (Hdr->ch_type != ELFCOMPRESS_ZLIB) { 246 error(toString(this) + ": unsupported compression type"); 247 return; 248 } 249 250 UncompressedSize = Hdr->ch_size; 251 RawData = RawData.slice(sizeof(*Hdr)); 252 return; 253 } 254 255 // New-style 32-bit header 256 if (RawData.size() < sizeof(Chdr32)) { 257 error(toString(this) + ": corrupted compressed section"); 258 return; 259 } 260 261 auto *Hdr = reinterpret_cast<const Chdr32 *>(RawData.data()); 262 if (Hdr->ch_type != ELFCOMPRESS_ZLIB) { 263 error(toString(this) + ": unsupported compression type"); 264 return; 265 } 266 267 UncompressedSize = Hdr->ch_size; 268 RawData = RawData.slice(sizeof(*Hdr)); 269 } 270 271 InputSection *InputSectionBase::getLinkOrderDep() const { 272 assert(Link); 273 assert(Flags & SHF_LINK_ORDER); 274 return cast<InputSection>(File->getSections()[Link]); 275 } 276 277 // Find a function symbol that encloses a given location. 278 template <class ELFT> 279 Defined *InputSectionBase::getEnclosingFunction(uint64_t Offset) { 280 for (Symbol *B : File->getSymbols()) 281 if (Defined *D = dyn_cast<Defined>(B)) 282 if (D->Section == this && D->Type == STT_FUNC && D->Value <= Offset && 283 Offset < D->Value + D->Size) 284 return D; 285 return nullptr; 286 } 287 288 // Returns a source location string. Used to construct an error message. 289 template <class ELFT> 290 std::string InputSectionBase::getLocation(uint64_t Offset) { 291 // We don't have file for synthetic sections. 292 if (getFile<ELFT>() == nullptr) 293 return (Config->OutputFile + ":(" + Name + "+0x" + utohexstr(Offset) + ")") 294 .str(); 295 296 // First check if we can get desired values from debugging information. 297 if (Optional<DILineInfo> Info = getFile<ELFT>()->getDILineInfo(this, Offset)) 298 return Info->FileName + ":" + std::to_string(Info->Line); 299 300 // File->SourceFile contains STT_FILE symbol that contains a 301 // source file name. If it's missing, we use an object file name. 302 std::string SrcFile = getFile<ELFT>()->SourceFile; 303 if (SrcFile.empty()) 304 SrcFile = toString(File); 305 306 if (Defined *D = getEnclosingFunction<ELFT>(Offset)) 307 return SrcFile + ":(function " + toString(*D) + ")"; 308 309 // If there's no symbol, print out the offset in the section. 310 return (SrcFile + ":(" + Name + "+0x" + utohexstr(Offset) + ")").str(); 311 } 312 313 // This function is intended to be used for constructing an error message. 314 // The returned message looks like this: 315 // 316 // foo.c:42 (/home/alice/possibly/very/long/path/foo.c:42) 317 // 318 // Returns an empty string if there's no way to get line info. 319 std::string InputSectionBase::getSrcMsg(const Symbol &Sym, uint64_t Offset) { 320 return File->getSrcMsg(Sym, *this, Offset); 321 } 322 323 // Returns a filename string along with an optional section name. This 324 // function is intended to be used for constructing an error 325 // message. The returned message looks like this: 326 // 327 // path/to/foo.o:(function bar) 328 // 329 // or 330 // 331 // path/to/foo.o:(function bar) in archive path/to/bar.a 332 std::string InputSectionBase::getObjMsg(uint64_t Off) { 333 std::string Filename = File->getName(); 334 335 std::string Archive; 336 if (!File->ArchiveName.empty()) 337 Archive = " in archive " + File->ArchiveName; 338 339 // Find a symbol that encloses a given location. 340 for (Symbol *B : File->getSymbols()) 341 if (auto *D = dyn_cast<Defined>(B)) 342 if (D->Section == this && D->Value <= Off && Off < D->Value + D->Size) 343 return Filename + ":(" + toString(*D) + ")" + Archive; 344 345 // If there's no symbol, print out the offset in the section. 346 return (Filename + ":(" + Name + "+0x" + utohexstr(Off) + ")" + Archive) 347 .str(); 348 } 349 350 InputSection InputSection::Discarded(nullptr, 0, 0, 0, ArrayRef<uint8_t>(), ""); 351 352 InputSection::InputSection(InputFile *F, uint64_t Flags, uint32_t Type, 353 uint32_t Alignment, ArrayRef<uint8_t> Data, 354 StringRef Name, Kind K) 355 : InputSectionBase(F, Flags, Type, 356 /*Entsize*/ 0, /*Link*/ 0, /*Info*/ 0, Alignment, Data, 357 Name, K) {} 358 359 template <class ELFT> 360 InputSection::InputSection(ObjFile<ELFT> &F, const typename ELFT::Shdr &Header, 361 StringRef Name) 362 : InputSectionBase(F, Header, Name, InputSectionBase::Regular) {} 363 364 bool InputSection::classof(const SectionBase *S) { 365 return S->kind() == SectionBase::Regular || 366 S->kind() == SectionBase::Synthetic; 367 } 368 369 OutputSection *InputSection::getParent() const { 370 return cast_or_null<OutputSection>(Parent); 371 } 372 373 // Copy SHT_GROUP section contents. Used only for the -r option. 374 template <class ELFT> void InputSection::copyShtGroup(uint8_t *Buf) { 375 // ELFT::Word is the 32-bit integral type in the target endianness. 376 typedef typename ELFT::Word u32; 377 ArrayRef<u32> From = getDataAs<u32>(); 378 auto *To = reinterpret_cast<u32 *>(Buf); 379 380 // The first entry is not a section number but a flag. 381 *To++ = From[0]; 382 383 // Adjust section numbers because section numbers in an input object 384 // files are different in the output. 385 ArrayRef<InputSectionBase *> Sections = File->getSections(); 386 for (uint32_t Idx : From.slice(1)) 387 *To++ = Sections[Idx]->getOutputSection()->SectionIndex; 388 } 389 390 InputSectionBase *InputSection::getRelocatedSection() const { 391 if (!File || (Type != SHT_RELA && Type != SHT_REL)) 392 return nullptr; 393 ArrayRef<InputSectionBase *> Sections = File->getSections(); 394 return Sections[Info]; 395 } 396 397 // This is used for -r and --emit-relocs. We can't use memcpy to copy 398 // relocations because we need to update symbol table offset and section index 399 // for each relocation. So we copy relocations one by one. 400 template <class ELFT, class RelTy> 401 void InputSection::copyRelocations(uint8_t *Buf, ArrayRef<RelTy> Rels) { 402 InputSectionBase *Sec = getRelocatedSection(); 403 404 for (const RelTy &Rel : Rels) { 405 RelType Type = Rel.getType(Config->IsMips64EL); 406 Symbol &Sym = getFile<ELFT>()->getRelocTargetSym(Rel); 407 408 auto *P = reinterpret_cast<typename ELFT::Rela *>(Buf); 409 Buf += sizeof(RelTy); 410 411 if (RelTy::IsRela) 412 P->r_addend = getAddend<ELFT>(Rel); 413 414 // Output section VA is zero for -r, so r_offset is an offset within the 415 // section, but for --emit-relocs it is an virtual address. 416 P->r_offset = Sec->getVA(Rel.r_offset); 417 P->setSymbolAndType(In.SymTab->getSymbolIndex(&Sym), Type, 418 Config->IsMips64EL); 419 420 if (Sym.Type == STT_SECTION) { 421 // We combine multiple section symbols into only one per 422 // section. This means we have to update the addend. That is 423 // trivial for Elf_Rela, but for Elf_Rel we have to write to the 424 // section data. We do that by adding to the Relocation vector. 425 426 // .eh_frame is horribly special and can reference discarded sections. To 427 // avoid having to parse and recreate .eh_frame, we just replace any 428 // relocation in it pointing to discarded sections with R_*_NONE, which 429 // hopefully creates a frame that is ignored at runtime. 430 auto *D = dyn_cast<Defined>(&Sym); 431 if (!D) { 432 error("STT_SECTION symbol should be defined"); 433 continue; 434 } 435 SectionBase *Section = D->Section->Repl; 436 if (!Section->Live) { 437 P->setSymbolAndType(0, 0, false); 438 continue; 439 } 440 441 int64_t Addend = getAddend<ELFT>(Rel); 442 const uint8_t *BufLoc = Sec->data().begin() + Rel.r_offset; 443 if (!RelTy::IsRela) 444 Addend = Target->getImplicitAddend(BufLoc, Type); 445 446 if (Config->EMachine == EM_MIPS && Config->Relocatable && 447 Target->getRelExpr(Type, Sym, BufLoc) == R_MIPS_GOTREL) { 448 // Some MIPS relocations depend on "gp" value. By default, 449 // this value has 0x7ff0 offset from a .got section. But 450 // relocatable files produced by a complier or a linker 451 // might redefine this default value and we must use it 452 // for a calculation of the relocation result. When we 453 // generate EXE or DSO it's trivial. Generating a relocatable 454 // output is more difficult case because the linker does 455 // not calculate relocations in this mode and loses 456 // individual "gp" values used by each input object file. 457 // As a workaround we add the "gp" value to the relocation 458 // addend and save it back to the file. 459 Addend += Sec->getFile<ELFT>()->MipsGp0; 460 } 461 462 if (RelTy::IsRela) 463 P->r_addend = Sym.getVA(Addend) - Section->getOutputSection()->Addr; 464 else if (Config->Relocatable) 465 Sec->Relocations.push_back({R_ABS, Type, Rel.r_offset, Addend, &Sym}); 466 } 467 } 468 } 469 470 // The ARM and AArch64 ABI handle pc-relative relocations to undefined weak 471 // references specially. The general rule is that the value of the symbol in 472 // this context is the address of the place P. A further special case is that 473 // branch relocations to an undefined weak reference resolve to the next 474 // instruction. 475 static uint32_t getARMUndefinedRelativeWeakVA(RelType Type, uint32_t A, 476 uint32_t P) { 477 switch (Type) { 478 // Unresolved branch relocations to weak references resolve to next 479 // instruction, this will be either 2 or 4 bytes on from P. 480 case R_ARM_THM_JUMP11: 481 return P + 2 + A; 482 case R_ARM_CALL: 483 case R_ARM_JUMP24: 484 case R_ARM_PC24: 485 case R_ARM_PLT32: 486 case R_ARM_PREL31: 487 case R_ARM_THM_JUMP19: 488 case R_ARM_THM_JUMP24: 489 return P + 4 + A; 490 case R_ARM_THM_CALL: 491 // We don't want an interworking BLX to ARM 492 return P + 5 + A; 493 // Unresolved non branch pc-relative relocations 494 // R_ARM_TARGET2 which can be resolved relatively is not present as it never 495 // targets a weak-reference. 496 case R_ARM_MOVW_PREL_NC: 497 case R_ARM_MOVT_PREL: 498 case R_ARM_REL32: 499 case R_ARM_THM_MOVW_PREL_NC: 500 case R_ARM_THM_MOVT_PREL: 501 return P + A; 502 } 503 llvm_unreachable("ARM pc-relative relocation expected\n"); 504 } 505 506 // The comment above getARMUndefinedRelativeWeakVA applies to this function. 507 static uint64_t getAArch64UndefinedRelativeWeakVA(uint64_t Type, uint64_t A, 508 uint64_t P) { 509 switch (Type) { 510 // Unresolved branch relocations to weak references resolve to next 511 // instruction, this is 4 bytes on from P. 512 case R_AARCH64_CALL26: 513 case R_AARCH64_CONDBR19: 514 case R_AARCH64_JUMP26: 515 case R_AARCH64_TSTBR14: 516 return P + 4 + A; 517 // Unresolved non branch pc-relative relocations 518 case R_AARCH64_PREL16: 519 case R_AARCH64_PREL32: 520 case R_AARCH64_PREL64: 521 case R_AARCH64_ADR_PREL_LO21: 522 case R_AARCH64_LD_PREL_LO19: 523 return P + A; 524 } 525 llvm_unreachable("AArch64 pc-relative relocation expected\n"); 526 } 527 528 // ARM SBREL relocations are of the form S + A - B where B is the static base 529 // The ARM ABI defines base to be "addressing origin of the output segment 530 // defining the symbol S". We defined the "addressing origin"/static base to be 531 // the base of the PT_LOAD segment containing the Sym. 532 // The procedure call standard only defines a Read Write Position Independent 533 // RWPI variant so in practice we should expect the static base to be the base 534 // of the RW segment. 535 static uint64_t getARMStaticBase(const Symbol &Sym) { 536 OutputSection *OS = Sym.getOutputSection(); 537 if (!OS || !OS->PtLoad || !OS->PtLoad->FirstSec) 538 fatal("SBREL relocation to " + Sym.getName() + " without static base"); 539 return OS->PtLoad->FirstSec->Addr; 540 } 541 542 // For R_RISCV_PC_INDIRECT (R_RISCV_PCREL_LO12_{I,S}), the symbol actually 543 // points the corresponding R_RISCV_PCREL_HI20 relocation, and the target VA 544 // is calculated using PCREL_HI20's symbol. 545 // 546 // This function returns the R_RISCV_PCREL_HI20 relocation from 547 // R_RISCV_PCREL_LO12's symbol and addend. 548 static Relocation *getRISCVPCRelHi20(const Symbol *Sym, uint64_t Addend) { 549 const Defined *D = cast<Defined>(Sym); 550 InputSection *IS = cast<InputSection>(D->Section); 551 552 if (Addend != 0) 553 warn("Non-zero addend in R_RISCV_PCREL_LO12 relocation to " + 554 IS->getObjMsg(D->Value) + " is ignored"); 555 556 // Relocations are sorted by offset, so we can use std::equal_range to do 557 // binary search. 558 auto Range = std::equal_range(IS->Relocations.begin(), IS->Relocations.end(), 559 D->Value, RelocationOffsetComparator{}); 560 for (auto It = std::get<0>(Range); It != std::get<1>(Range); ++It) 561 if (isRelExprOneOf<R_PC>(It->Expr)) 562 return &*It; 563 564 error("R_RISCV_PCREL_LO12 relocation points to " + IS->getObjMsg(D->Value) + 565 " without an associated R_RISCV_PCREL_HI20 relocation"); 566 return nullptr; 567 } 568 569 // A TLS symbol's virtual address is relative to the TLS segment. Add a 570 // target-specific adjustment to produce a thread-pointer-relative offset. 571 static int64_t getTlsTpOffset() { 572 switch (Config->EMachine) { 573 case EM_ARM: 574 case EM_AARCH64: 575 // Variant 1. The thread pointer points to a TCB with a fixed 2-word size, 576 // followed by a variable amount of alignment padding, followed by the TLS 577 // segment. 578 // 579 // NB: While the ARM/AArch64 ABI formally has a 2-word TCB size, lld 580 // effectively increases the TCB size to 8 words for Android compatibility. 581 // It accomplishes this by increasing the segment's alignment. 582 return alignTo(Config->Wordsize * 2, Out::TlsPhdr->p_align); 583 case EM_386: 584 case EM_X86_64: 585 // Variant 2. The TLS segment is located just before the thread pointer. 586 return -Out::TlsPhdr->p_memsz; 587 case EM_PPC64: 588 // The thread pointer points to a fixed offset from the start of the 589 // executable's TLS segment. An offset of 0x7000 allows a signed 16-bit 590 // offset to reach 0x1000 of TCB/thread-library data and 0xf000 of the 591 // program's TLS segment. 592 return -0x7000; 593 default: 594 llvm_unreachable("unhandled Config->EMachine"); 595 } 596 } 597 598 static uint64_t getRelocTargetVA(const InputFile *File, RelType Type, int64_t A, 599 uint64_t P, const Symbol &Sym, RelExpr Expr) { 600 switch (Expr) { 601 case R_INVALID: 602 return 0; 603 case R_ABS: 604 case R_RELAX_TLS_LD_TO_LE_ABS: 605 case R_RELAX_GOT_PC_NOPIC: 606 return Sym.getVA(A); 607 case R_ADDEND: 608 return A; 609 case R_ARM_SBREL: 610 return Sym.getVA(A) - getARMStaticBase(Sym); 611 case R_GOT: 612 case R_GOT_PLT: 613 case R_RELAX_TLS_GD_TO_IE_ABS: 614 return Sym.getGotVA() + A; 615 case R_GOTONLY_PC: 616 return In.Got->getVA() + A - P; 617 case R_GOTONLY_PC_FROM_END: 618 return In.Got->getVA() + A - P + In.Got->getSize(); 619 case R_GOTREL: 620 return Sym.getVA(A) - In.Got->getVA(); 621 case R_GOTREL_FROM_END: 622 return Sym.getVA(A) - In.Got->getVA() - In.Got->getSize(); 623 case R_GOT_FROM_END: 624 case R_RELAX_TLS_GD_TO_IE_END: 625 return Sym.getGotOffset() + A - In.Got->getSize(); 626 case R_TLSLD_GOT_OFF: 627 case R_GOT_OFF: 628 case R_RELAX_TLS_GD_TO_IE_GOT_OFF: 629 return Sym.getGotOffset() + A; 630 case R_AARCH64_GOT_PAGE_PC: 631 case R_AARCH64_GOT_PAGE_PC_PLT: 632 case R_AARCH64_RELAX_TLS_GD_TO_IE_PAGE_PC: 633 return getAArch64Page(Sym.getGotVA() + A) - getAArch64Page(P); 634 case R_GOT_PC: 635 case R_RELAX_TLS_GD_TO_IE: 636 return Sym.getGotVA() + A - P; 637 case R_HEXAGON_GOT: 638 return Sym.getGotVA() - In.GotPlt->getVA(); 639 case R_MIPS_GOTREL: 640 return Sym.getVA(A) - In.MipsGot->getGp(File); 641 case R_MIPS_GOT_GP: 642 return In.MipsGot->getGp(File) + A; 643 case R_MIPS_GOT_GP_PC: { 644 // R_MIPS_LO16 expression has R_MIPS_GOT_GP_PC type iif the target 645 // is _gp_disp symbol. In that case we should use the following 646 // formula for calculation "AHL + GP - P + 4". For details see p. 4-19 at 647 // ftp://www.linux-mips.org/pub/linux/mips/doc/ABI/mipsabi.pdf 648 // microMIPS variants of these relocations use slightly different 649 // expressions: AHL + GP - P + 3 for %lo() and AHL + GP - P - 1 for %hi() 650 // to correctly handle less-sugnificant bit of the microMIPS symbol. 651 uint64_t V = In.MipsGot->getGp(File) + A - P; 652 if (Type == R_MIPS_LO16 || Type == R_MICROMIPS_LO16) 653 V += 4; 654 if (Type == R_MICROMIPS_LO16 || Type == R_MICROMIPS_HI16) 655 V -= 1; 656 return V; 657 } 658 case R_MIPS_GOT_LOCAL_PAGE: 659 // If relocation against MIPS local symbol requires GOT entry, this entry 660 // should be initialized by 'page address'. This address is high 16-bits 661 // of sum the symbol's value and the addend. 662 return In.MipsGot->getVA() + In.MipsGot->getPageEntryOffset(File, Sym, A) - 663 In.MipsGot->getGp(File); 664 case R_MIPS_GOT_OFF: 665 case R_MIPS_GOT_OFF32: 666 // In case of MIPS if a GOT relocation has non-zero addend this addend 667 // should be applied to the GOT entry content not to the GOT entry offset. 668 // That is why we use separate expression type. 669 return In.MipsGot->getVA() + In.MipsGot->getSymEntryOffset(File, Sym, A) - 670 In.MipsGot->getGp(File); 671 case R_MIPS_TLSGD: 672 return In.MipsGot->getVA() + In.MipsGot->getGlobalDynOffset(File, Sym) - 673 In.MipsGot->getGp(File); 674 case R_MIPS_TLSLD: 675 return In.MipsGot->getVA() + In.MipsGot->getTlsIndexOffset(File) - 676 In.MipsGot->getGp(File); 677 case R_AARCH64_PAGE_PC: { 678 uint64_t Val = Sym.isUndefWeak() ? P + A : Sym.getVA(A); 679 return getAArch64Page(Val) - getAArch64Page(P); 680 } 681 case R_AARCH64_PLT_PAGE_PC: { 682 uint64_t Val = Sym.isUndefWeak() ? P + A : Sym.getPltVA() + A; 683 return getAArch64Page(Val) - getAArch64Page(P); 684 } 685 case R_RISCV_PC_INDIRECT: { 686 if (const Relocation *HiRel = getRISCVPCRelHi20(&Sym, A)) 687 return getRelocTargetVA(File, HiRel->Type, HiRel->Addend, Sym.getVA(), 688 *HiRel->Sym, HiRel->Expr); 689 return 0; 690 } 691 case R_PC: { 692 uint64_t Dest; 693 if (Sym.isUndefWeak()) { 694 // On ARM and AArch64 a branch to an undefined weak resolves to the 695 // next instruction, otherwise the place. 696 if (Config->EMachine == EM_ARM) 697 Dest = getARMUndefinedRelativeWeakVA(Type, A, P); 698 else if (Config->EMachine == EM_AARCH64) 699 Dest = getAArch64UndefinedRelativeWeakVA(Type, A, P); 700 else 701 Dest = Sym.getVA(A); 702 } else { 703 Dest = Sym.getVA(A); 704 } 705 return Dest - P; 706 } 707 case R_PLT: 708 return Sym.getPltVA() + A; 709 case R_PLT_PC: 710 case R_PPC_CALL_PLT: 711 return Sym.getPltVA() + A - P; 712 case R_PPC_CALL: { 713 uint64_t SymVA = Sym.getVA(A); 714 // If we have an undefined weak symbol, we might get here with a symbol 715 // address of zero. That could overflow, but the code must be unreachable, 716 // so don't bother doing anything at all. 717 if (!SymVA) 718 return 0; 719 720 // PPC64 V2 ABI describes two entry points to a function. The global entry 721 // point is used for calls where the caller and callee (may) have different 722 // TOC base pointers and r2 needs to be modified to hold the TOC base for 723 // the callee. For local calls the caller and callee share the same 724 // TOC base and so the TOC pointer initialization code should be skipped by 725 // branching to the local entry point. 726 return SymVA - P + getPPC64GlobalEntryToLocalEntryOffset(Sym.StOther); 727 } 728 case R_PPC_TOC: 729 return getPPC64TocBase() + A; 730 case R_RELAX_GOT_PC: 731 return Sym.getVA(A) - P; 732 case R_RELAX_TLS_GD_TO_LE: 733 case R_RELAX_TLS_IE_TO_LE: 734 case R_RELAX_TLS_LD_TO_LE: 735 case R_TLS: 736 // A weak undefined TLS symbol resolves to the base of the TLS 737 // block, i.e. gets a value of zero. If we pass --gc-sections to 738 // lld and .tbss is not referenced, it gets reclaimed and we don't 739 // create a TLS program header. Therefore, we resolve this 740 // statically to zero. 741 if (Sym.isTls() && Sym.isUndefWeak()) 742 return 0; 743 return Sym.getVA(A) + getTlsTpOffset(); 744 case R_RELAX_TLS_GD_TO_LE_NEG: 745 case R_NEG_TLS: 746 return Out::TlsPhdr->p_memsz - Sym.getVA(A); 747 case R_SIZE: 748 return Sym.getSize() + A; 749 case R_TLSDESC: 750 return In.Got->getGlobalDynAddr(Sym) + A; 751 case R_AARCH64_TLSDESC_PAGE: 752 return getAArch64Page(In.Got->getGlobalDynAddr(Sym) + A) - 753 getAArch64Page(P); 754 case R_TLSGD_GOT: 755 return In.Got->getGlobalDynOffset(Sym) + A; 756 case R_TLSGD_GOT_FROM_END: 757 return In.Got->getGlobalDynOffset(Sym) + A - In.Got->getSize(); 758 case R_TLSGD_PC: 759 return In.Got->getGlobalDynAddr(Sym) + A - P; 760 case R_TLSLD_GOT_FROM_END: 761 return In.Got->getTlsIndexOff() + A - In.Got->getSize(); 762 case R_TLSLD_GOT: 763 return In.Got->getTlsIndexOff() + A; 764 case R_TLSLD_PC: 765 return In.Got->getTlsIndexVA() + A - P; 766 default: 767 llvm_unreachable("invalid expression"); 768 } 769 } 770 771 // This function applies relocations to sections without SHF_ALLOC bit. 772 // Such sections are never mapped to memory at runtime. Debug sections are 773 // an example. Relocations in non-alloc sections are much easier to 774 // handle than in allocated sections because it will never need complex 775 // treatement such as GOT or PLT (because at runtime no one refers them). 776 // So, we handle relocations for non-alloc sections directly in this 777 // function as a performance optimization. 778 template <class ELFT, class RelTy> 779 void InputSection::relocateNonAlloc(uint8_t *Buf, ArrayRef<RelTy> Rels) { 780 const unsigned Bits = sizeof(typename ELFT::uint) * 8; 781 782 for (const RelTy &Rel : Rels) { 783 RelType Type = Rel.getType(Config->IsMips64EL); 784 785 // GCC 8.0 or earlier have a bug that they emit R_386_GOTPC relocations 786 // against _GLOBAL_OFFSET_TABLE_ for .debug_info. The bug has been fixed 787 // in 2017 (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82630), but we 788 // need to keep this bug-compatible code for a while. 789 if (Config->EMachine == EM_386 && Type == R_386_GOTPC) 790 continue; 791 792 uint64_t Offset = getOffset(Rel.r_offset); 793 uint8_t *BufLoc = Buf + Offset; 794 int64_t Addend = getAddend<ELFT>(Rel); 795 if (!RelTy::IsRela) 796 Addend += Target->getImplicitAddend(BufLoc, Type); 797 798 Symbol &Sym = getFile<ELFT>()->getRelocTargetSym(Rel); 799 RelExpr Expr = Target->getRelExpr(Type, Sym, BufLoc); 800 if (Expr == R_NONE) 801 continue; 802 803 if (Expr != R_ABS) { 804 std::string Msg = getLocation<ELFT>(Offset) + 805 ": has non-ABS relocation " + toString(Type) + 806 " against symbol '" + toString(Sym) + "'"; 807 if (Expr != R_PC) { 808 error(Msg); 809 return; 810 } 811 812 // If the control reaches here, we found a PC-relative relocation in a 813 // non-ALLOC section. Since non-ALLOC section is not loaded into memory 814 // at runtime, the notion of PC-relative doesn't make sense here. So, 815 // this is a usage error. However, GNU linkers historically accept such 816 // relocations without any errors and relocate them as if they were at 817 // address 0. For bug-compatibilty, we accept them with warnings. We 818 // know Steel Bank Common Lisp as of 2018 have this bug. 819 warn(Msg); 820 Target->relocateOne(BufLoc, Type, 821 SignExtend64<Bits>(Sym.getVA(Addend - Offset))); 822 continue; 823 } 824 825 if (Sym.isTls() && !Out::TlsPhdr) 826 Target->relocateOne(BufLoc, Type, 0); 827 else 828 Target->relocateOne(BufLoc, Type, SignExtend64<Bits>(Sym.getVA(Addend))); 829 } 830 } 831 832 // This is used when '-r' is given. 833 // For REL targets, InputSection::copyRelocations() may store artificial 834 // relocations aimed to update addends. They are handled in relocateAlloc() 835 // for allocatable sections, and this function does the same for 836 // non-allocatable sections, such as sections with debug information. 837 static void relocateNonAllocForRelocatable(InputSection *Sec, uint8_t *Buf) { 838 const unsigned Bits = Config->Is64 ? 64 : 32; 839 840 for (const Relocation &Rel : Sec->Relocations) { 841 // InputSection::copyRelocations() adds only R_ABS relocations. 842 assert(Rel.Expr == R_ABS); 843 uint8_t *BufLoc = Buf + Rel.Offset + Sec->OutSecOff; 844 uint64_t TargetVA = SignExtend64(Rel.Sym->getVA(Rel.Addend), Bits); 845 Target->relocateOne(BufLoc, Rel.Type, TargetVA); 846 } 847 } 848 849 template <class ELFT> 850 void InputSectionBase::relocate(uint8_t *Buf, uint8_t *BufEnd) { 851 if (Flags & SHF_EXECINSTR) 852 adjustSplitStackFunctionPrologues<ELFT>(Buf, BufEnd); 853 854 if (Flags & SHF_ALLOC) { 855 relocateAlloc(Buf, BufEnd); 856 return; 857 } 858 859 auto *Sec = cast<InputSection>(this); 860 if (Config->Relocatable) 861 relocateNonAllocForRelocatable(Sec, Buf); 862 else if (Sec->AreRelocsRela) 863 Sec->relocateNonAlloc<ELFT>(Buf, Sec->template relas<ELFT>()); 864 else 865 Sec->relocateNonAlloc<ELFT>(Buf, Sec->template rels<ELFT>()); 866 } 867 868 void InputSectionBase::relocateAlloc(uint8_t *Buf, uint8_t *BufEnd) { 869 assert(Flags & SHF_ALLOC); 870 const unsigned Bits = Config->Wordsize * 8; 871 872 for (const Relocation &Rel : Relocations) { 873 uint64_t Offset = Rel.Offset; 874 if (auto *Sec = dyn_cast<InputSection>(this)) 875 Offset += Sec->OutSecOff; 876 uint8_t *BufLoc = Buf + Offset; 877 RelType Type = Rel.Type; 878 879 uint64_t AddrLoc = getOutputSection()->Addr + Offset; 880 RelExpr Expr = Rel.Expr; 881 uint64_t TargetVA = SignExtend64( 882 getRelocTargetVA(File, Type, Rel.Addend, AddrLoc, *Rel.Sym, Expr), 883 Bits); 884 885 switch (Expr) { 886 case R_RELAX_GOT_PC: 887 case R_RELAX_GOT_PC_NOPIC: 888 Target->relaxGot(BufLoc, TargetVA); 889 break; 890 case R_RELAX_TLS_IE_TO_LE: 891 Target->relaxTlsIeToLe(BufLoc, Type, TargetVA); 892 break; 893 case R_RELAX_TLS_LD_TO_LE: 894 case R_RELAX_TLS_LD_TO_LE_ABS: 895 Target->relaxTlsLdToLe(BufLoc, Type, TargetVA); 896 break; 897 case R_RELAX_TLS_GD_TO_LE: 898 case R_RELAX_TLS_GD_TO_LE_NEG: 899 Target->relaxTlsGdToLe(BufLoc, Type, TargetVA); 900 break; 901 case R_AARCH64_RELAX_TLS_GD_TO_IE_PAGE_PC: 902 case R_RELAX_TLS_GD_TO_IE: 903 case R_RELAX_TLS_GD_TO_IE_ABS: 904 case R_RELAX_TLS_GD_TO_IE_GOT_OFF: 905 case R_RELAX_TLS_GD_TO_IE_END: 906 Target->relaxTlsGdToIe(BufLoc, Type, TargetVA); 907 break; 908 case R_PPC_CALL: 909 // If this is a call to __tls_get_addr, it may be part of a TLS 910 // sequence that has been relaxed and turned into a nop. In this 911 // case, we don't want to handle it as a call. 912 if (read32(BufLoc) == 0x60000000) // nop 913 break; 914 915 // Patch a nop (0x60000000) to a ld. 916 if (Rel.Sym->NeedsTocRestore) { 917 if (BufLoc + 8 > BufEnd || read32(BufLoc + 4) != 0x60000000) { 918 error(getErrorLocation(BufLoc) + "call lacks nop, can't restore toc"); 919 break; 920 } 921 write32(BufLoc + 4, 0xe8410018); // ld %r2, 24(%r1) 922 } 923 Target->relocateOne(BufLoc, Type, TargetVA); 924 break; 925 default: 926 Target->relocateOne(BufLoc, Type, TargetVA); 927 break; 928 } 929 } 930 } 931 932 // For each function-defining prologue, find any calls to __morestack, 933 // and replace them with calls to __morestack_non_split. 934 static void switchMorestackCallsToMorestackNonSplit( 935 DenseSet<Defined *> &Prologues, std::vector<Relocation *> &MorestackCalls) { 936 937 // If the target adjusted a function's prologue, all calls to 938 // __morestack inside that function should be switched to 939 // __morestack_non_split. 940 Symbol *MoreStackNonSplit = Symtab->find("__morestack_non_split"); 941 if (!MoreStackNonSplit) { 942 error("Mixing split-stack objects requires a definition of " 943 "__morestack_non_split"); 944 return; 945 } 946 947 // Sort both collections to compare addresses efficiently. 948 llvm::sort(MorestackCalls, [](const Relocation *L, const Relocation *R) { 949 return L->Offset < R->Offset; 950 }); 951 std::vector<Defined *> Functions(Prologues.begin(), Prologues.end()); 952 llvm::sort(Functions, [](const Defined *L, const Defined *R) { 953 return L->Value < R->Value; 954 }); 955 956 auto It = MorestackCalls.begin(); 957 for (Defined *F : Functions) { 958 // Find the first call to __morestack within the function. 959 while (It != MorestackCalls.end() && (*It)->Offset < F->Value) 960 ++It; 961 // Adjust all calls inside the function. 962 while (It != MorestackCalls.end() && (*It)->Offset < F->Value + F->Size) { 963 (*It)->Sym = MoreStackNonSplit; 964 ++It; 965 } 966 } 967 } 968 969 static bool enclosingPrologueAttempted(uint64_t Offset, 970 const DenseSet<Defined *> &Prologues) { 971 for (Defined *F : Prologues) 972 if (F->Value <= Offset && Offset < F->Value + F->Size) 973 return true; 974 return false; 975 } 976 977 // If a function compiled for split stack calls a function not 978 // compiled for split stack, then the caller needs its prologue 979 // adjusted to ensure that the called function will have enough stack 980 // available. Find those functions, and adjust their prologues. 981 template <class ELFT> 982 void InputSectionBase::adjustSplitStackFunctionPrologues(uint8_t *Buf, 983 uint8_t *End) { 984 if (!getFile<ELFT>()->SplitStack) 985 return; 986 DenseSet<Defined *> Prologues; 987 std::vector<Relocation *> MorestackCalls; 988 989 for (Relocation &Rel : Relocations) { 990 // Local symbols can't possibly be cross-calls, and should have been 991 // resolved long before this line. 992 if (Rel.Sym->isLocal()) 993 continue; 994 995 // Ignore calls into the split-stack api. 996 if (Rel.Sym->getName().startswith("__morestack")) { 997 if (Rel.Sym->getName().equals("__morestack")) 998 MorestackCalls.push_back(&Rel); 999 continue; 1000 } 1001 1002 // A relocation to non-function isn't relevant. Sometimes 1003 // __morestack is not marked as a function, so this check comes 1004 // after the name check. 1005 if (Rel.Sym->Type != STT_FUNC) 1006 continue; 1007 1008 // If the callee's-file was compiled with split stack, nothing to do. In 1009 // this context, a "Defined" symbol is one "defined by the binary currently 1010 // being produced". So an "undefined" symbol might be provided by a shared 1011 // library. It is not possible to tell how such symbols were compiled, so be 1012 // conservative. 1013 if (Defined *D = dyn_cast<Defined>(Rel.Sym)) 1014 if (InputSection *IS = cast_or_null<InputSection>(D->Section)) 1015 if (!IS || !IS->getFile<ELFT>() || IS->getFile<ELFT>()->SplitStack) 1016 continue; 1017 1018 if (enclosingPrologueAttempted(Rel.Offset, Prologues)) 1019 continue; 1020 1021 if (Defined *F = getEnclosingFunction<ELFT>(Rel.Offset)) { 1022 Prologues.insert(F); 1023 if (Target->adjustPrologueForCrossSplitStack(Buf + getOffset(F->Value), 1024 End, F->StOther)) 1025 continue; 1026 if (!getFile<ELFT>()->SomeNoSplitStack) 1027 error(lld::toString(this) + ": " + F->getName() + 1028 " (with -fsplit-stack) calls " + Rel.Sym->getName() + 1029 " (without -fsplit-stack), but couldn't adjust its prologue"); 1030 } 1031 } 1032 1033 if (Target->NeedsMoreStackNonSplit) 1034 switchMorestackCallsToMorestackNonSplit(Prologues, MorestackCalls); 1035 } 1036 1037 template <class ELFT> void InputSection::writeTo(uint8_t *Buf) { 1038 if (Type == SHT_NOBITS) 1039 return; 1040 1041 if (auto *S = dyn_cast<SyntheticSection>(this)) { 1042 S->writeTo(Buf + OutSecOff); 1043 return; 1044 } 1045 1046 // If -r or --emit-relocs is given, then an InputSection 1047 // may be a relocation section. 1048 if (Type == SHT_RELA) { 1049 copyRelocations<ELFT>(Buf + OutSecOff, getDataAs<typename ELFT::Rela>()); 1050 return; 1051 } 1052 if (Type == SHT_REL) { 1053 copyRelocations<ELFT>(Buf + OutSecOff, getDataAs<typename ELFT::Rel>()); 1054 return; 1055 } 1056 1057 // If -r is given, we may have a SHT_GROUP section. 1058 if (Type == SHT_GROUP) { 1059 copyShtGroup<ELFT>(Buf + OutSecOff); 1060 return; 1061 } 1062 1063 // If this is a compressed section, uncompress section contents directly 1064 // to the buffer. 1065 if (UncompressedSize >= 0 && !UncompressedBuf) { 1066 size_t Size = UncompressedSize; 1067 if (Error E = zlib::uncompress(toStringRef(RawData), 1068 (char *)(Buf + OutSecOff), Size)) 1069 fatal(toString(this) + 1070 ": uncompress failed: " + llvm::toString(std::move(E))); 1071 uint8_t *BufEnd = Buf + OutSecOff + Size; 1072 relocate<ELFT>(Buf, BufEnd); 1073 return; 1074 } 1075 1076 // Copy section contents from source object file to output file 1077 // and then apply relocations. 1078 memcpy(Buf + OutSecOff, data().data(), data().size()); 1079 uint8_t *BufEnd = Buf + OutSecOff + data().size(); 1080 relocate<ELFT>(Buf, BufEnd); 1081 } 1082 1083 void InputSection::replace(InputSection *Other) { 1084 Alignment = std::max(Alignment, Other->Alignment); 1085 Other->Repl = Repl; 1086 Other->Live = false; 1087 } 1088 1089 template <class ELFT> 1090 EhInputSection::EhInputSection(ObjFile<ELFT> &F, 1091 const typename ELFT::Shdr &Header, 1092 StringRef Name) 1093 : InputSectionBase(F, Header, Name, InputSectionBase::EHFrame) {} 1094 1095 SyntheticSection *EhInputSection::getParent() const { 1096 return cast_or_null<SyntheticSection>(Parent); 1097 } 1098 1099 // Returns the index of the first relocation that points to a region between 1100 // Begin and Begin+Size. 1101 template <class IntTy, class RelTy> 1102 static unsigned getReloc(IntTy Begin, IntTy Size, const ArrayRef<RelTy> &Rels, 1103 unsigned &RelocI) { 1104 // Start search from RelocI for fast access. That works because the 1105 // relocations are sorted in .eh_frame. 1106 for (unsigned N = Rels.size(); RelocI < N; ++RelocI) { 1107 const RelTy &Rel = Rels[RelocI]; 1108 if (Rel.r_offset < Begin) 1109 continue; 1110 1111 if (Rel.r_offset < Begin + Size) 1112 return RelocI; 1113 return -1; 1114 } 1115 return -1; 1116 } 1117 1118 // .eh_frame is a sequence of CIE or FDE records. 1119 // This function splits an input section into records and returns them. 1120 template <class ELFT> void EhInputSection::split() { 1121 if (AreRelocsRela) 1122 split<ELFT>(relas<ELFT>()); 1123 else 1124 split<ELFT>(rels<ELFT>()); 1125 } 1126 1127 template <class ELFT, class RelTy> 1128 void EhInputSection::split(ArrayRef<RelTy> Rels) { 1129 unsigned RelI = 0; 1130 for (size_t Off = 0, End = data().size(); Off != End;) { 1131 size_t Size = readEhRecordSize(this, Off); 1132 Pieces.emplace_back(Off, this, Size, getReloc(Off, Size, Rels, RelI)); 1133 // The empty record is the end marker. 1134 if (Size == 4) 1135 break; 1136 Off += Size; 1137 } 1138 } 1139 1140 static size_t findNull(StringRef S, size_t EntSize) { 1141 // Optimize the common case. 1142 if (EntSize == 1) 1143 return S.find(0); 1144 1145 for (unsigned I = 0, N = S.size(); I != N; I += EntSize) { 1146 const char *B = S.begin() + I; 1147 if (std::all_of(B, B + EntSize, [](char C) { return C == 0; })) 1148 return I; 1149 } 1150 return StringRef::npos; 1151 } 1152 1153 SyntheticSection *MergeInputSection::getParent() const { 1154 return cast_or_null<SyntheticSection>(Parent); 1155 } 1156 1157 // Split SHF_STRINGS section. Such section is a sequence of 1158 // null-terminated strings. 1159 void MergeInputSection::splitStrings(ArrayRef<uint8_t> Data, size_t EntSize) { 1160 size_t Off = 0; 1161 bool IsAlloc = Flags & SHF_ALLOC; 1162 StringRef S = toStringRef(Data); 1163 1164 while (!S.empty()) { 1165 size_t End = findNull(S, EntSize); 1166 if (End == StringRef::npos) 1167 fatal(toString(this) + ": string is not null terminated"); 1168 size_t Size = End + EntSize; 1169 1170 Pieces.emplace_back(Off, xxHash64(S.substr(0, Size)), !IsAlloc); 1171 S = S.substr(Size); 1172 Off += Size; 1173 } 1174 } 1175 1176 // Split non-SHF_STRINGS section. Such section is a sequence of 1177 // fixed size records. 1178 void MergeInputSection::splitNonStrings(ArrayRef<uint8_t> Data, 1179 size_t EntSize) { 1180 size_t Size = Data.size(); 1181 assert((Size % EntSize) == 0); 1182 bool IsAlloc = Flags & SHF_ALLOC; 1183 1184 for (size_t I = 0; I != Size; I += EntSize) 1185 Pieces.emplace_back(I, xxHash64(Data.slice(I, EntSize)), !IsAlloc); 1186 } 1187 1188 template <class ELFT> 1189 MergeInputSection::MergeInputSection(ObjFile<ELFT> &F, 1190 const typename ELFT::Shdr &Header, 1191 StringRef Name) 1192 : InputSectionBase(F, Header, Name, InputSectionBase::Merge) {} 1193 1194 MergeInputSection::MergeInputSection(uint64_t Flags, uint32_t Type, 1195 uint64_t Entsize, ArrayRef<uint8_t> Data, 1196 StringRef Name) 1197 : InputSectionBase(nullptr, Flags, Type, Entsize, /*Link*/ 0, /*Info*/ 0, 1198 /*Alignment*/ Entsize, Data, Name, SectionBase::Merge) {} 1199 1200 // This function is called after we obtain a complete list of input sections 1201 // that need to be linked. This is responsible to split section contents 1202 // into small chunks for further processing. 1203 // 1204 // Note that this function is called from parallelForEach. This must be 1205 // thread-safe (i.e. no memory allocation from the pools). 1206 void MergeInputSection::splitIntoPieces() { 1207 assert(Pieces.empty()); 1208 1209 if (Flags & SHF_STRINGS) 1210 splitStrings(data(), Entsize); 1211 else 1212 splitNonStrings(data(), Entsize); 1213 } 1214 1215 SectionPiece *MergeInputSection::getSectionPiece(uint64_t Offset) { 1216 if (this->data().size() <= Offset) 1217 fatal(toString(this) + ": offset is outside the section"); 1218 1219 // If Offset is not at beginning of a section piece, it is not in the map. 1220 // In that case we need to do a binary search of the original section piece vector. 1221 auto It2 = 1222 llvm::upper_bound(Pieces, Offset, [](uint64_t Offset, SectionPiece P) { 1223 return Offset < P.InputOff; 1224 }); 1225 return &It2[-1]; 1226 } 1227 1228 // Returns the offset in an output section for a given input offset. 1229 // Because contents of a mergeable section is not contiguous in output, 1230 // it is not just an addition to a base output offset. 1231 uint64_t MergeInputSection::getParentOffset(uint64_t Offset) const { 1232 // If Offset is not at beginning of a section piece, it is not in the map. 1233 // In that case we need to search from the original section piece vector. 1234 const SectionPiece &Piece = 1235 *(const_cast<MergeInputSection *>(this)->getSectionPiece (Offset)); 1236 uint64_t Addend = Offset - Piece.InputOff; 1237 return Piece.OutputOff + Addend; 1238 } 1239 1240 template InputSection::InputSection(ObjFile<ELF32LE> &, const ELF32LE::Shdr &, 1241 StringRef); 1242 template InputSection::InputSection(ObjFile<ELF32BE> &, const ELF32BE::Shdr &, 1243 StringRef); 1244 template InputSection::InputSection(ObjFile<ELF64LE> &, const ELF64LE::Shdr &, 1245 StringRef); 1246 template InputSection::InputSection(ObjFile<ELF64BE> &, const ELF64BE::Shdr &, 1247 StringRef); 1248 1249 template std::string InputSectionBase::getLocation<ELF32LE>(uint64_t); 1250 template std::string InputSectionBase::getLocation<ELF32BE>(uint64_t); 1251 template std::string InputSectionBase::getLocation<ELF64LE>(uint64_t); 1252 template std::string InputSectionBase::getLocation<ELF64BE>(uint64_t); 1253 1254 template void InputSection::writeTo<ELF32LE>(uint8_t *); 1255 template void InputSection::writeTo<ELF32BE>(uint8_t *); 1256 template void InputSection::writeTo<ELF64LE>(uint8_t *); 1257 template void InputSection::writeTo<ELF64BE>(uint8_t *); 1258 1259 template MergeInputSection::MergeInputSection(ObjFile<ELF32LE> &, 1260 const ELF32LE::Shdr &, StringRef); 1261 template MergeInputSection::MergeInputSection(ObjFile<ELF32BE> &, 1262 const ELF32BE::Shdr &, StringRef); 1263 template MergeInputSection::MergeInputSection(ObjFile<ELF64LE> &, 1264 const ELF64LE::Shdr &, StringRef); 1265 template MergeInputSection::MergeInputSection(ObjFile<ELF64BE> &, 1266 const ELF64BE::Shdr &, StringRef); 1267 1268 template EhInputSection::EhInputSection(ObjFile<ELF32LE> &, 1269 const ELF32LE::Shdr &, StringRef); 1270 template EhInputSection::EhInputSection(ObjFile<ELF32BE> &, 1271 const ELF32BE::Shdr &, StringRef); 1272 template EhInputSection::EhInputSection(ObjFile<ELF64LE> &, 1273 const ELF64LE::Shdr &, StringRef); 1274 template EhInputSection::EhInputSection(ObjFile<ELF64BE> &, 1275 const ELF64BE::Shdr &, StringRef); 1276 1277 template void EhInputSection::split<ELF32LE>(); 1278 template void EhInputSection::split<ELF32BE>(); 1279 template void EhInputSection::split<ELF64LE>(); 1280 template void EhInputSection::split<ELF64BE>(); 1281