1 //===- InputSection.cpp ---------------------------------------------------===//
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 #include "InputSection.h"
14 #include "InputFiles.h"
15 #include "LinkerScript.h"
17 #include "OutputSections.h"
18 #include "Relocations.h"
19 #include "SyntheticSections.h"
22 #include "llvm/Object/Decompressor.h"
23 #include "llvm/Support/Compression.h"
24 #include "llvm/Support/Endian.h"
25 #include "llvm/Support/Path.h"
29 using namespace llvm::ELF;
30 using namespace llvm::object;
31 using namespace llvm::support;
32 using namespace llvm::support::endian;
33 using namespace llvm::sys;
36 using namespace lld::elf;
38 std::vector<InputSectionBase *> elf::InputSections;
40 // Returns a string to construct an error message.
41 std::string lld::toString(const InputSectionBase *Sec) {
42 return (toString(Sec->File) + ":(" + Sec->Name + ")").str();
46 static ArrayRef<uint8_t> getSectionContents(elf::ObjectFile<ELFT> *File,
47 const typename ELFT::Shdr *Hdr) {
48 if (!File || Hdr->sh_type == SHT_NOBITS)
49 return makeArrayRef<uint8_t>(nullptr, Hdr->sh_size);
50 return check(File->getObj().getSectionContents(Hdr));
53 InputSectionBase::InputSectionBase(InputFile *File, uint64_t Flags,
54 uint32_t Type, uint64_t Entsize,
55 uint32_t Link, uint32_t Info,
56 uint32_t Alignment, ArrayRef<uint8_t> Data,
57 StringRef Name, Kind SectionKind)
58 : SectionBase(SectionKind, Name, Flags, Entsize, Alignment, Type, Info,
60 File(File), Data(Data), Repl(this) {
61 Live = !Config->GcSections || !(Flags & SHF_ALLOC);
64 AreRelocsRela = false;
66 // The ELF spec states that a value of 0 means the section has
67 // no alignment constraits.
68 uint32_t V = std::max<uint64_t>(Alignment, 1);
69 if (!isPowerOf2_64(V))
70 fatal(toString(File) + ": section sh_addralign is not a power of 2");
74 // GNU assembler 2.24 and LLVM 4.0.0's MC (the newest release as of
75 // March 2017) fail to infer section types for sections starting with
76 // ".init_array." or ".fini_array.". They set SHT_PROGBITS instead of
77 // SHF_INIT_ARRAY. As a result, the following assembler directive
78 // creates ".init_array.100" with SHT_PROGBITS, for example.
80 // .section .init_array.100, "aw"
82 // This function forces SHT_{INIT,FINI}_ARRAY so that we can handle
83 // incorrect inputs as if they were correct from the beginning.
84 static uint64_t getType(uint64_t Type, StringRef Name) {
85 if (Type == SHT_PROGBITS && Name.startswith(".init_array."))
86 return SHT_INIT_ARRAY;
87 if (Type == SHT_PROGBITS && Name.startswith(".fini_array."))
88 return SHT_FINI_ARRAY;
93 InputSectionBase::InputSectionBase(elf::ObjectFile<ELFT> *File,
94 const typename ELFT::Shdr *Hdr,
95 StringRef Name, Kind SectionKind)
96 : InputSectionBase(File, Hdr->sh_flags & ~SHF_INFO_LINK,
97 getType(Hdr->sh_type, Name), Hdr->sh_entsize,
98 Hdr->sh_link, Hdr->sh_info, Hdr->sh_addralign,
99 getSectionContents(File, Hdr), Name, SectionKind) {
100 // We reject object files having insanely large alignments even though
101 // they are allowed by the spec. I think 4GB is a reasonable limitation.
102 // We might want to relax this in the future.
103 if (Hdr->sh_addralign > UINT32_MAX)
104 fatal(toString(File) + ": section sh_addralign is too large");
107 size_t InputSectionBase::getSize() const {
108 if (auto *S = dyn_cast<SyntheticSection>(this))
114 uint64_t InputSectionBase::getOffsetInFile() const {
115 const uint8_t *FileStart = (const uint8_t *)File->MB.getBufferStart();
116 const uint8_t *SecStart = Data.begin();
117 return SecStart - FileStart;
120 uint64_t SectionBase::getOffset(uint64_t Offset) const {
123 auto *OS = cast<OutputSection>(this);
124 // For output sections we treat offset -1 as the end of the section.
125 return Offset == uint64_t(-1) ? OS->Size : Offset;
128 return cast<InputSection>(this)->OutSecOff + Offset;
130 auto *IS = cast<InputSection>(this);
131 // For synthetic sections we treat offset -1 as the end of the section.
132 return IS->OutSecOff + (Offset == uint64_t(-1) ? IS->getSize() : Offset);
135 // The file crtbeginT.o has relocations pointing to the start of an empty
136 // .eh_frame that is known to be the first in the link. It does that to
137 // identify the start of the output .eh_frame.
140 const MergeInputSection *MS = cast<MergeInputSection>(this);
142 return MS->MergeSec->OutSecOff + MS->getOffset(Offset);
143 return MS->getOffset(Offset);
145 llvm_unreachable("invalid section kind");
148 OutputSection *SectionBase::getOutputSection() {
149 if (auto *IS = dyn_cast<InputSection>(this))
151 if (auto *MS = dyn_cast<MergeInputSection>(this))
152 return MS->MergeSec ? MS->MergeSec->OutSec : nullptr;
153 if (auto *EH = dyn_cast<EhInputSection>(this))
154 return EH->EHSec->OutSec;
155 return cast<OutputSection>(this);
158 // Uncompress section contents. Note that this function is called
159 // from parallel_for_each, so it must be thread-safe.
160 void InputSectionBase::uncompress() {
161 Decompressor Dec = check(Decompressor::create(Name, toStringRef(Data),
162 Config->IsLE, Config->Is64));
164 size_t Size = Dec.getDecompressedSize();
167 static std::mutex Mu;
168 std::lock_guard<std::mutex> Lock(Mu);
169 OutputBuf = BAlloc.Allocate<char>(Size);
172 if (Error E = Dec.decompress({OutputBuf, Size}))
173 fatal(toString(this) +
174 ": decompress failed: " + llvm::toString(std::move(E)));
175 Data = ArrayRef<uint8_t>((uint8_t *)OutputBuf, Size);
178 uint64_t SectionBase::getOffset(const DefinedRegular &Sym) const {
179 return getOffset(Sym.Value);
182 InputSectionBase *InputSectionBase::getLinkOrderDep() const {
183 if ((Flags & SHF_LINK_ORDER) && Link != 0)
184 return File->getSections()[Link];
188 // Returns a source location string. Used to construct an error message.
189 template <class ELFT>
190 std::string InputSectionBase::getLocation(uint64_t Offset) {
191 // We don't have file for synthetic sections.
192 if (getFile<ELFT>() == nullptr)
193 return (Config->OutputFile + ":(" + Name + "+0x" + utohexstr(Offset) + ")")
196 // First check if we can get desired values from debugging information.
197 std::string LineInfo = getFile<ELFT>()->getLineInfo(this, Offset);
198 if (!LineInfo.empty())
201 // File->SourceFile contains STT_FILE symbol that contains a
202 // source file name. If it's missing, we use an object file name.
203 std::string SrcFile = getFile<ELFT>()->SourceFile;
205 SrcFile = toString(File);
207 // Find a function symbol that encloses a given location.
208 for (SymbolBody *B : getFile<ELFT>()->getSymbols())
209 if (auto *D = dyn_cast<DefinedRegular>(B))
210 if (D->Section == this && D->Type == STT_FUNC)
211 if (D->Value <= Offset && Offset < D->Value + D->Size)
212 return SrcFile + ":(function " + toString(*D) + ")";
214 // If there's no symbol, print out the offset in the section.
215 return (SrcFile + ":(" + Name + "+0x" + utohexstr(Offset) + ")").str();
218 // Returns a source location string. This function is intended to be
219 // used for constructing an error message. The returned message looks
222 // foo.c:42 (/home/alice/possibly/very/long/path/foo.c:42)
224 // Returns an empty string if there's no way to get line info.
225 template <class ELFT> std::string InputSectionBase::getSrcMsg(uint64_t Offset) {
226 // Synthetic sections don't have input files.
227 elf::ObjectFile<ELFT> *File = getFile<ELFT>();
231 Optional<DILineInfo> Info = File->getDILineInfo(this, Offset);
233 // File->SourceFile contains STT_FILE symbol, and that is a last resort.
235 return File->SourceFile;
237 std::string Path = Info->FileName;
238 std::string Filename = path::filename(Path);
239 std::string Lineno = ":" + std::to_string(Info->Line);
240 if (Filename == Path)
241 return Filename + Lineno;
242 return Filename + Lineno + " (" + Path + Lineno + ")";
245 // Returns a filename string along with an optional section name. This
246 // function is intended to be used for constructing an error
247 // message. The returned message looks like this:
249 // path/to/foo.o:(function bar)
253 // path/to/foo.o:(function bar) in archive path/to/bar.a
254 template <class ELFT> std::string InputSectionBase::getObjMsg(uint64_t Off) {
255 // Synthetic sections don't have input files.
256 elf::ObjectFile<ELFT> *File = getFile<ELFT>();
257 std::string Filename = File ? File->getName() : "(internal)";
260 if (!File->ArchiveName.empty())
261 Archive = (" in archive " + File->ArchiveName).str();
263 // Find a symbol that encloses a given location.
264 for (SymbolBody *B : getFile<ELFT>()->getSymbols())
265 if (auto *D = dyn_cast<DefinedRegular>(B))
266 if (D->Section == this && D->Value <= Off && Off < D->Value + D->Size)
267 return Filename + ":(" + toString(*D) + ")" + Archive;
269 // If there's no symbol, print out the offset in the section.
270 return (Filename + ":(" + Name + "+0x" + utohexstr(Off) + ")" + Archive)
274 InputSectionBase InputSectionBase::Discarded;
276 InputSection::InputSection(uint64_t Flags, uint32_t Type, uint32_t Alignment,
277 ArrayRef<uint8_t> Data, StringRef Name, Kind K)
278 : InputSectionBase(nullptr, Flags, Type,
279 /*Entsize*/ 0, /*Link*/ 0, /*Info*/ 0, Alignment, Data,
282 template <class ELFT>
283 InputSection::InputSection(elf::ObjectFile<ELFT> *F,
284 const typename ELFT::Shdr *Header, StringRef Name)
285 : InputSectionBase(F, Header, Name, InputSectionBase::Regular) {}
287 bool InputSection::classof(const SectionBase *S) {
288 return S->kind() == SectionBase::Regular ||
289 S->kind() == SectionBase::Synthetic;
292 bool InputSectionBase::classof(const SectionBase *S) {
293 return S->kind() != Output;
296 InputSectionBase *InputSection::getRelocatedSection() {
297 assert(this->Type == SHT_RELA || this->Type == SHT_REL);
298 ArrayRef<InputSectionBase *> Sections = this->File->getSections();
299 return Sections[this->Info];
302 // This is used for -r and --emit-relocs. We can't use memcpy to copy
303 // relocations because we need to update symbol table offset and section index
304 // for each relocation. So we copy relocations one by one.
305 template <class ELFT, class RelTy>
306 void InputSection::copyRelocations(uint8_t *Buf, ArrayRef<RelTy> Rels) {
307 InputSectionBase *RelocatedSection = getRelocatedSection();
309 // Loop is slow and have complexity O(N*M), where N - amount of
310 // relocations and M - amount of symbols in symbol table.
311 // That happens because getSymbolIndex(...) call below performs
312 // simple linear search.
313 for (const RelTy &Rel : Rels) {
314 uint32_t Type = Rel.getType(Config->IsMips64EL);
315 SymbolBody &Body = this->getFile<ELFT>()->getRelocTargetSym(Rel);
317 auto *P = reinterpret_cast<typename ELFT::Rela *>(Buf);
318 Buf += sizeof(RelTy);
321 P->r_addend = getAddend<ELFT>(Rel);
323 // Output section VA is zero for -r, so r_offset is an offset within the
324 // section, but for --emit-relocs it is an virtual address.
325 P->r_offset = RelocatedSection->OutSec->Addr +
326 RelocatedSection->getOffset(Rel.r_offset);
327 P->setSymbolAndType(InX::SymTab->getSymbolIndex(&Body), Type,
330 if (Body.Type == STT_SECTION) {
331 // We combine multiple section symbols into only one per
332 // section. This means we have to update the addend. That is
333 // trivial for Elf_Rela, but for Elf_Rel we have to write to the
334 // section data. We do that by adding to the Relocation vector.
336 // .eh_frame is horribly special and can reference discarded sections. To
337 // avoid having to parse and recreate .eh_frame, we just replace any
338 // relocation in it pointing to discarded sections with R_*_NONE, which
339 // hopefully creates a frame that is ignored at runtime.
340 SectionBase *Section = cast<DefinedRegular>(Body).Section;
341 if (Section == &InputSection::Discarded) {
342 P->setSymbolAndType(0, 0, false);
346 if (Config->IsRela) {
347 P->r_addend += Body.getVA() - Section->getOutputSection()->Addr;
348 } else if (Config->Relocatable) {
349 const uint8_t *BufLoc = RelocatedSection->Data.begin() + Rel.r_offset;
350 RelocatedSection->Relocations.push_back(
351 {R_ABS, Type, Rel.r_offset, Target->getImplicitAddend(BufLoc, Type),
359 static uint32_t getARMUndefinedRelativeWeakVA(uint32_t Type, uint32_t A,
362 case R_ARM_THM_JUMP11:
369 case R_ARM_THM_JUMP19:
370 case R_ARM_THM_JUMP24:
373 // We don't want an interworking BLX to ARM
380 static uint64_t getAArch64UndefinedRelativeWeakVA(uint64_t Type, uint64_t A,
383 case R_AARCH64_CALL26:
384 case R_AARCH64_CONDBR19:
385 case R_AARCH64_JUMP26:
386 case R_AARCH64_TSTBR14:
393 template <class ELFT>
394 static typename ELFT::uint
395 getRelocTargetVA(uint32_t Type, int64_t A, typename ELFT::uint P,
396 const SymbolBody &Body, RelExpr Expr) {
399 case R_RELAX_GOT_PC_NOPIC:
400 return Body.getVA(A);
402 case R_RELAX_TLS_GD_TO_IE_ABS:
403 return Body.getGotVA() + A;
405 return InX::Got->getVA() + A - P;
406 case R_GOTONLY_PC_FROM_END:
407 return InX::Got->getVA() + A - P + InX::Got->getSize();
409 return Body.getVA(A) - InX::Got->getVA();
410 case R_GOTREL_FROM_END:
411 return Body.getVA(A) - InX::Got->getVA() - InX::Got->getSize();
413 case R_RELAX_TLS_GD_TO_IE_END:
414 return Body.getGotOffset() + A - InX::Got->getSize();
416 return Body.getGotOffset() + A;
418 case R_RELAX_TLS_GD_TO_IE_PAGE_PC:
419 return getAArch64Page(Body.getGotVA() + A) - getAArch64Page(P);
421 case R_RELAX_TLS_GD_TO_IE:
422 return Body.getGotVA() + A - P;
426 llvm_unreachable("cannot relocate hint relocs");
428 return Body.getVA(A) - InX::MipsGot->getGp();
430 return InX::MipsGot->getGp() + A;
431 case R_MIPS_GOT_GP_PC: {
432 // R_MIPS_LO16 expression has R_MIPS_GOT_GP_PC type iif the target
433 // is _gp_disp symbol. In that case we should use the following
434 // formula for calculation "AHL + GP - P + 4". For details see p. 4-19 at
435 // ftp://www.linux-mips.org/pub/linux/mips/doc/ABI/mipsabi.pdf
436 uint64_t V = InX::MipsGot->getGp() + A - P;
437 if (Type == R_MIPS_LO16)
441 case R_MIPS_GOT_LOCAL_PAGE:
442 // If relocation against MIPS local symbol requires GOT entry, this entry
443 // should be initialized by 'page address'. This address is high 16-bits
444 // of sum the symbol's value and the addend.
445 return InX::MipsGot->getVA() + InX::MipsGot->getPageEntryOffset(Body, A) -
446 InX::MipsGot->getGp();
448 case R_MIPS_GOT_OFF32:
449 // In case of MIPS if a GOT relocation has non-zero addend this addend
450 // should be applied to the GOT entry content not to the GOT entry offset.
451 // That is why we use separate expression type.
452 return InX::MipsGot->getVA() + InX::MipsGot->getBodyEntryOffset(Body, A) -
453 InX::MipsGot->getGp();
455 return InX::MipsGot->getVA() + InX::MipsGot->getTlsOffset() +
456 InX::MipsGot->getGlobalDynOffset(Body) - InX::MipsGot->getGp();
458 return InX::MipsGot->getVA() + InX::MipsGot->getTlsOffset() +
459 InX::MipsGot->getTlsIndexOff() - InX::MipsGot->getGp();
462 if (Body.isUndefined() && !Body.isLocal() && Body.symbol()->isWeak())
463 return getAArch64Page(A);
464 return getAArch64Page(Body.getVA(A)) - getAArch64Page(P);
466 if (Body.isUndefined() && !Body.isLocal() && Body.symbol()->isWeak()) {
467 // On ARM and AArch64 a branch to an undefined weak resolves to the
468 // next instruction, otherwise the place.
469 if (Config->EMachine == EM_ARM)
470 return getARMUndefinedRelativeWeakVA(Type, A, P);
471 if (Config->EMachine == EM_AARCH64)
472 return getAArch64UndefinedRelativeWeakVA(Type, A, P);
474 return Body.getVA(A) - P;
476 return Body.getPltVA() + A;
479 return Body.getPltVA() + A - P;
481 uint64_t SymVA = Body.getVA(A);
482 // If we have an undefined weak symbol, we might get here with a symbol
483 // address of zero. That could overflow, but the code must be unreachable,
484 // so don't bother doing anything at all.
488 // If this is a local call, and we currently have the address of a
489 // function-descriptor, get the underlying code address instead.
490 uint64_t OpdStart = Out::Opd->Addr;
491 uint64_t OpdEnd = OpdStart + Out::Opd->Size;
492 bool InOpd = OpdStart <= SymVA && SymVA < OpdEnd;
494 SymVA = read64be(&Out::OpdBuf[SymVA - OpdStart]);
499 return getPPC64TocBase() + A;
501 return Body.getVA(A) - P;
502 case R_RELAX_TLS_GD_TO_LE:
503 case R_RELAX_TLS_IE_TO_LE:
504 case R_RELAX_TLS_LD_TO_LE:
506 // A weak undefined TLS symbol resolves to the base of the TLS
507 // block, i.e. gets a value of zero. If we pass --gc-sections to
508 // lld and .tbss is not referenced, it gets reclaimed and we don't
509 // create a TLS program header. Therefore, we resolve this
510 // statically to zero.
511 if (Body.isTls() && (Body.isLazy() || Body.isUndefined()) &&
512 Body.symbol()->isWeak())
515 return Body.getVA(A) + alignTo(Target->TcbSize, Out::TlsPhdr->p_align);
516 return Body.getVA(A) - Out::TlsPhdr->p_memsz;
517 case R_RELAX_TLS_GD_TO_LE_NEG:
519 return Out::TlsPhdr->p_memsz - Body.getVA(A);
521 return Body.getSize<ELFT>() + A;
523 return InX::Got->getGlobalDynAddr(Body) + A;
525 return getAArch64Page(InX::Got->getGlobalDynAddr(Body) + A) -
528 return InX::Got->getGlobalDynOffset(Body) + A - InX::Got->getSize();
530 return InX::Got->getGlobalDynAddr(Body) + A - P;
532 return InX::Got->getTlsIndexOff() + A - InX::Got->getSize();
534 return InX::Got->getTlsIndexVA() + A - P;
536 llvm_unreachable("Invalid expression");
539 // This function applies relocations to sections without SHF_ALLOC bit.
540 // Such sections are never mapped to memory at runtime. Debug sections are
541 // an example. Relocations in non-alloc sections are much easier to
542 // handle than in allocated sections because it will never need complex
543 // treatement such as GOT or PLT (because at runtime no one refers them).
544 // So, we handle relocations for non-alloc sections directly in this
545 // function as a performance optimization.
546 template <class ELFT, class RelTy>
547 void InputSection::relocateNonAlloc(uint8_t *Buf, ArrayRef<RelTy> Rels) {
548 for (const RelTy &Rel : Rels) {
549 uint32_t Type = Rel.getType(Config->IsMips64EL);
550 uint64_t Offset = getOffset(Rel.r_offset);
551 uint8_t *BufLoc = Buf + Offset;
552 int64_t Addend = getAddend<ELFT>(Rel);
554 Addend += Target->getImplicitAddend(BufLoc, Type);
556 SymbolBody &Sym = this->getFile<ELFT>()->getRelocTargetSym(Rel);
557 RelExpr Expr = Target->getRelExpr(Type, Sym, BufLoc);
561 error(this->getLocation<ELFT>(Offset) + ": has non-ABS reloc");
565 uint64_t AddrLoc = this->OutSec->Addr + Offset;
567 if (!Sym.isTls() || Out::TlsPhdr)
568 SymVA = SignExtend64<sizeof(typename ELFT::uint) * 8>(
569 getRelocTargetVA<ELFT>(Type, Addend, AddrLoc, Sym, R_ABS));
570 Target->relocateOne(BufLoc, Type, SymVA);
574 template <class ELFT> elf::ObjectFile<ELFT> *InputSectionBase::getFile() const {
575 return cast_or_null<elf::ObjectFile<ELFT>>(File);
578 template <class ELFT>
579 void InputSectionBase::relocate(uint8_t *Buf, uint8_t *BufEnd) {
580 // scanReloc function in Writer.cpp constructs Relocations
581 // vector only for SHF_ALLOC'ed sections. For other sections,
582 // we handle relocations directly here.
583 auto *IS = dyn_cast<InputSection>(this);
584 if (IS && !(IS->Flags & SHF_ALLOC)) {
585 if (IS->AreRelocsRela)
586 IS->relocateNonAlloc<ELFT>(Buf, IS->template relas<ELFT>());
588 IS->relocateNonAlloc<ELFT>(Buf, IS->template rels<ELFT>());
592 const unsigned Bits = sizeof(typename ELFT::uint) * 8;
593 for (const Relocation &Rel : Relocations) {
594 uint64_t Offset = getOffset(Rel.Offset);
595 uint8_t *BufLoc = Buf + Offset;
596 uint32_t Type = Rel.Type;
598 uint64_t AddrLoc = getOutputSection()->Addr + Offset;
599 RelExpr Expr = Rel.Expr;
600 uint64_t TargetVA = SignExtend64<Bits>(
601 getRelocTargetVA<ELFT>(Type, Rel.Addend, AddrLoc, *Rel.Sym, Expr));
605 case R_RELAX_GOT_PC_NOPIC:
606 Target->relaxGot(BufLoc, TargetVA);
608 case R_RELAX_TLS_IE_TO_LE:
609 Target->relaxTlsIeToLe(BufLoc, Type, TargetVA);
611 case R_RELAX_TLS_LD_TO_LE:
612 Target->relaxTlsLdToLe(BufLoc, Type, TargetVA);
614 case R_RELAX_TLS_GD_TO_LE:
615 case R_RELAX_TLS_GD_TO_LE_NEG:
616 Target->relaxTlsGdToLe(BufLoc, Type, TargetVA);
618 case R_RELAX_TLS_GD_TO_IE:
619 case R_RELAX_TLS_GD_TO_IE_ABS:
620 case R_RELAX_TLS_GD_TO_IE_PAGE_PC:
621 case R_RELAX_TLS_GD_TO_IE_END:
622 Target->relaxTlsGdToIe(BufLoc, Type, TargetVA);
625 // Patch a nop (0x60000000) to a ld.
626 if (BufLoc + 8 <= BufEnd && read32be(BufLoc + 4) == 0x60000000)
627 write32be(BufLoc + 4, 0xe8410028); // ld %r2, 40(%r1)
630 Target->relocateOne(BufLoc, Type, TargetVA);
636 template <class ELFT> void InputSection::writeTo(uint8_t *Buf) {
637 if (this->Type == SHT_NOBITS)
640 if (auto *S = dyn_cast<SyntheticSection>(this)) {
641 S->writeTo(Buf + OutSecOff);
645 // If -r or --emit-relocs is given, then an InputSection
646 // may be a relocation section.
647 if (this->Type == SHT_RELA) {
648 copyRelocations<ELFT>(Buf + OutSecOff,
649 this->template getDataAs<typename ELFT::Rela>());
652 if (this->Type == SHT_REL) {
653 copyRelocations<ELFT>(Buf + OutSecOff,
654 this->template getDataAs<typename ELFT::Rel>());
658 // Copy section contents from source object file to output file
659 // and then apply relocations.
660 memcpy(Buf + OutSecOff, Data.data(), Data.size());
661 uint8_t *BufEnd = Buf + OutSecOff + Data.size();
662 this->relocate<ELFT>(Buf, BufEnd);
665 void InputSection::replace(InputSection *Other) {
666 this->Alignment = std::max(this->Alignment, Other->Alignment);
667 Other->Repl = this->Repl;
671 template <class ELFT>
672 EhInputSection::EhInputSection(elf::ObjectFile<ELFT> *F,
673 const typename ELFT::Shdr *Header,
675 : InputSectionBase(F, Header, Name, InputSectionBase::EHFrame) {
676 // Mark .eh_frame sections as live by default because there are
677 // usually no relocations that point to .eh_frames. Otherwise,
678 // the garbage collector would drop all .eh_frame sections.
682 bool EhInputSection::classof(const SectionBase *S) {
683 return S->kind() == InputSectionBase::EHFrame;
686 // Returns the index of the first relocation that points to a region between
687 // Begin and Begin+Size.
688 template <class IntTy, class RelTy>
689 static unsigned getReloc(IntTy Begin, IntTy Size, const ArrayRef<RelTy> &Rels,
691 // Start search from RelocI for fast access. That works because the
692 // relocations are sorted in .eh_frame.
693 for (unsigned N = Rels.size(); RelocI < N; ++RelocI) {
694 const RelTy &Rel = Rels[RelocI];
695 if (Rel.r_offset < Begin)
698 if (Rel.r_offset < Begin + Size)
705 // .eh_frame is a sequence of CIE or FDE records.
706 // This function splits an input section into records and returns them.
707 template <class ELFT> void EhInputSection::split() {
708 // Early exit if already split.
709 if (!this->Pieces.empty())
712 if (this->NumRelocations) {
713 if (this->AreRelocsRela)
714 split<ELFT>(this->relas<ELFT>());
716 split<ELFT>(this->rels<ELFT>());
719 split<ELFT>(makeArrayRef<typename ELFT::Rela>(nullptr, nullptr));
722 template <class ELFT, class RelTy>
723 void EhInputSection::split(ArrayRef<RelTy> Rels) {
724 ArrayRef<uint8_t> Data = this->Data;
726 for (size_t Off = 0, End = Data.size(); Off != End;) {
727 size_t Size = readEhRecordSize<ELFT>(this, Off);
728 this->Pieces.emplace_back(Off, this, Size, getReloc(Off, Size, Rels, RelI));
729 // The empty record is the end marker.
736 static size_t findNull(ArrayRef<uint8_t> A, size_t EntSize) {
737 // Optimize the common case.
738 StringRef S((const char *)A.data(), A.size());
742 for (unsigned I = 0, N = S.size(); I != N; I += EntSize) {
743 const char *B = S.begin() + I;
744 if (std::all_of(B, B + EntSize, [](char C) { return C == 0; }))
747 return StringRef::npos;
750 // Split SHF_STRINGS section. Such section is a sequence of
751 // null-terminated strings.
752 void MergeInputSection::splitStrings(ArrayRef<uint8_t> Data, size_t EntSize) {
754 bool IsAlloc = this->Flags & SHF_ALLOC;
755 while (!Data.empty()) {
756 size_t End = findNull(Data, EntSize);
757 if (End == StringRef::npos)
758 fatal(toString(this) + ": string is not null terminated");
759 size_t Size = End + EntSize;
760 Pieces.emplace_back(Off, !IsAlloc);
761 Hashes.push_back(hash_value(toStringRef(Data.slice(0, Size))));
762 Data = Data.slice(Size);
767 // Split non-SHF_STRINGS section. Such section is a sequence of
768 // fixed size records.
769 void MergeInputSection::splitNonStrings(ArrayRef<uint8_t> Data,
771 size_t Size = Data.size();
772 assert((Size % EntSize) == 0);
773 bool IsAlloc = this->Flags & SHF_ALLOC;
774 for (unsigned I = 0, N = Size; I != N; I += EntSize) {
775 Hashes.push_back(hash_value(toStringRef(Data.slice(I, EntSize))));
776 Pieces.emplace_back(I, !IsAlloc);
780 template <class ELFT>
781 MergeInputSection::MergeInputSection(elf::ObjectFile<ELFT> *F,
782 const typename ELFT::Shdr *Header,
784 : InputSectionBase(F, Header, Name, InputSectionBase::Merge) {}
786 // This function is called after we obtain a complete list of input sections
787 // that need to be linked. This is responsible to split section contents
788 // into small chunks for further processing.
790 // Note that this function is called from parallel_for_each. This must be
791 // thread-safe (i.e. no memory allocation from the pools).
792 void MergeInputSection::splitIntoPieces() {
793 ArrayRef<uint8_t> Data = this->Data;
794 uint64_t EntSize = this->Entsize;
795 if (this->Flags & SHF_STRINGS)
796 splitStrings(Data, EntSize);
798 splitNonStrings(Data, EntSize);
800 if (Config->GcSections && (this->Flags & SHF_ALLOC))
801 for (uint64_t Off : LiveOffsets)
802 this->getSectionPiece(Off)->Live = true;
805 bool MergeInputSection::classof(const SectionBase *S) {
806 return S->kind() == InputSectionBase::Merge;
809 // Do binary search to get a section piece at a given input offset.
810 SectionPiece *MergeInputSection::getSectionPiece(uint64_t Offset) {
811 auto *This = static_cast<const MergeInputSection *>(this);
812 return const_cast<SectionPiece *>(This->getSectionPiece(Offset));
815 template <class It, class T, class Compare>
816 static It fastUpperBound(It First, It Last, const T &Value, Compare Comp) {
817 size_t Size = std::distance(First, Last);
821 const It MI = First + H;
823 First = Comp(Value, *MI) ? First : First + H;
825 return Comp(Value, *First) ? First : First + 1;
828 const SectionPiece *MergeInputSection::getSectionPiece(uint64_t Offset) const {
829 uint64_t Size = this->Data.size();
831 fatal(toString(this) + ": entry is past the end of the section");
833 // Find the element this offset points to.
834 auto I = fastUpperBound(
835 Pieces.begin(), Pieces.end(), Offset,
836 [](const uint64_t &A, const SectionPiece &B) { return A < B.InputOff; });
841 // Returns the offset in an output section for a given input offset.
842 // Because contents of a mergeable section is not contiguous in output,
843 // it is not just an addition to a base output offset.
844 uint64_t MergeInputSection::getOffset(uint64_t Offset) const {
845 // Initialize OffsetMap lazily.
846 std::call_once(InitOffsetMap, [&] {
847 OffsetMap.reserve(Pieces.size());
848 for (const SectionPiece &Piece : Pieces)
849 OffsetMap[Piece.InputOff] = Piece.OutputOff;
852 // Find a string starting at a given offset.
853 auto It = OffsetMap.find(Offset);
854 if (It != OffsetMap.end())
860 // If Offset is not at beginning of a section piece, it is not in the map.
861 // In that case we need to search from the original section piece vector.
862 const SectionPiece &Piece = *this->getSectionPiece(Offset);
866 uint64_t Addend = Offset - Piece.InputOff;
867 return Piece.OutputOff + Addend;
870 template InputSection::InputSection(elf::ObjectFile<ELF32LE> *,
871 const ELF32LE::Shdr *, StringRef);
872 template InputSection::InputSection(elf::ObjectFile<ELF32BE> *,
873 const ELF32BE::Shdr *, StringRef);
874 template InputSection::InputSection(elf::ObjectFile<ELF64LE> *,
875 const ELF64LE::Shdr *, StringRef);
876 template InputSection::InputSection(elf::ObjectFile<ELF64BE> *,
877 const ELF64BE::Shdr *, StringRef);
879 template std::string InputSectionBase::getLocation<ELF32LE>(uint64_t);
880 template std::string InputSectionBase::getLocation<ELF32BE>(uint64_t);
881 template std::string InputSectionBase::getLocation<ELF64LE>(uint64_t);
882 template std::string InputSectionBase::getLocation<ELF64BE>(uint64_t);
884 template std::string InputSectionBase::getSrcMsg<ELF32LE>(uint64_t);
885 template std::string InputSectionBase::getSrcMsg<ELF32BE>(uint64_t);
886 template std::string InputSectionBase::getSrcMsg<ELF64LE>(uint64_t);
887 template std::string InputSectionBase::getSrcMsg<ELF64BE>(uint64_t);
889 template std::string InputSectionBase::getObjMsg<ELF32LE>(uint64_t);
890 template std::string InputSectionBase::getObjMsg<ELF32BE>(uint64_t);
891 template std::string InputSectionBase::getObjMsg<ELF64LE>(uint64_t);
892 template std::string InputSectionBase::getObjMsg<ELF64BE>(uint64_t);
894 template void InputSection::writeTo<ELF32LE>(uint8_t *);
895 template void InputSection::writeTo<ELF32BE>(uint8_t *);
896 template void InputSection::writeTo<ELF64LE>(uint8_t *);
897 template void InputSection::writeTo<ELF64BE>(uint8_t *);
899 template elf::ObjectFile<ELF32LE> *InputSectionBase::getFile<ELF32LE>() const;
900 template elf::ObjectFile<ELF32BE> *InputSectionBase::getFile<ELF32BE>() const;
901 template elf::ObjectFile<ELF64LE> *InputSectionBase::getFile<ELF64LE>() const;
902 template elf::ObjectFile<ELF64BE> *InputSectionBase::getFile<ELF64BE>() const;
904 template MergeInputSection::MergeInputSection(elf::ObjectFile<ELF32LE> *,
905 const ELF32LE::Shdr *, StringRef);
906 template MergeInputSection::MergeInputSection(elf::ObjectFile<ELF32BE> *,
907 const ELF32BE::Shdr *, StringRef);
908 template MergeInputSection::MergeInputSection(elf::ObjectFile<ELF64LE> *,
909 const ELF64LE::Shdr *, StringRef);
910 template MergeInputSection::MergeInputSection(elf::ObjectFile<ELF64BE> *,
911 const ELF64BE::Shdr *, StringRef);
913 template EhInputSection::EhInputSection(elf::ObjectFile<ELF32LE> *,
914 const ELF32LE::Shdr *, StringRef);
915 template EhInputSection::EhInputSection(elf::ObjectFile<ELF32BE> *,
916 const ELF32BE::Shdr *, StringRef);
917 template EhInputSection::EhInputSection(elf::ObjectFile<ELF64LE> *,
918 const ELF64LE::Shdr *, StringRef);
919 template EhInputSection::EhInputSection(elf::ObjectFile<ELF64BE> *,
920 const ELF64BE::Shdr *, StringRef);
922 template void EhInputSection::split<ELF32LE>();
923 template void EhInputSection::split<ELF32BE>();
924 template void EhInputSection::split<ELF64LE>();
925 template void EhInputSection::split<ELF64BE>();