InputSection.cpp revision 360660
1//===- InputSection.cpp ---------------------------------------------------===//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8
9#include "InputSection.h"
10#include "Config.h"
11#include "EhFrame.h"
12#include "InputFiles.h"
13#include "LinkerScript.h"
14#include "OutputSections.h"
15#include "Relocations.h"
16#include "SymbolTable.h"
17#include "Symbols.h"
18#include "SyntheticSections.h"
19#include "Target.h"
20#include "Thunks.h"
21#include "lld/Common/ErrorHandler.h"
22#include "lld/Common/Memory.h"
23#include "llvm/Support/Compiler.h"
24#include "llvm/Support/Compression.h"
25#include "llvm/Support/Endian.h"
26#include "llvm/Support/Threading.h"
27#include "llvm/Support/xxhash.h"
28#include <algorithm>
29#include <mutex>
30#include <set>
31#include <vector>
32
33using namespace llvm;
34using namespace llvm::ELF;
35using namespace llvm::object;
36using namespace llvm::support;
37using namespace llvm::support::endian;
38using namespace llvm::sys;
39
40using namespace lld;
41using namespace lld::elf;
42
43std::vector<InputSectionBase *> elf::inputSections;
44
45// Returns a string to construct an error message.
46std::string lld::toString(const InputSectionBase *sec) {
47  return (toString(sec->file) + ":(" + sec->name + ")").str();
48}
49
50template <class ELFT>
51static ArrayRef<uint8_t> getSectionContents(ObjFile<ELFT> &file,
52                                            const typename ELFT::Shdr &hdr) {
53  if (hdr.sh_type == SHT_NOBITS)
54    return makeArrayRef<uint8_t>(nullptr, hdr.sh_size);
55  return check(file.getObj().getSectionContents(&hdr));
56}
57
58InputSectionBase::InputSectionBase(InputFile *file, uint64_t flags,
59                                   uint32_t type, uint64_t entsize,
60                                   uint32_t link, uint32_t info,
61                                   uint32_t alignment, ArrayRef<uint8_t> data,
62                                   StringRef name, Kind sectionKind)
63    : SectionBase(sectionKind, name, flags, entsize, alignment, type, info,
64                  link),
65      file(file), rawData(data) {
66  // In order to reduce memory allocation, we assume that mergeable
67  // sections are smaller than 4 GiB, which is not an unreasonable
68  // assumption as of 2017.
69  if (sectionKind == SectionBase::Merge && rawData.size() > UINT32_MAX)
70    error(toString(this) + ": section too large");
71
72  numRelocations = 0;
73  areRelocsRela = false;
74
75  // The ELF spec states that a value of 0 means the section has
76  // no alignment constraits.
77  uint32_t v = std::max<uint32_t>(alignment, 1);
78  if (!isPowerOf2_64(v))
79    fatal(toString(this) + ": sh_addralign is not a power of 2");
80  this->alignment = v;
81
82  // In ELF, each section can be compressed by zlib, and if compressed,
83  // section name may be mangled by appending "z" (e.g. ".zdebug_info").
84  // If that's the case, demangle section name so that we can handle a
85  // section as if it weren't compressed.
86  if ((flags & SHF_COMPRESSED) || name.startswith(".zdebug")) {
87    if (!zlib::isAvailable())
88      error(toString(file) + ": contains a compressed section, " +
89            "but zlib is not available");
90    parseCompressedHeader();
91  }
92}
93
94// Drop SHF_GROUP bit unless we are producing a re-linkable object file.
95// SHF_GROUP is a marker that a section belongs to some comdat group.
96// That flag doesn't make sense in an executable.
97static uint64_t getFlags(uint64_t flags) {
98  flags &= ~(uint64_t)SHF_INFO_LINK;
99  if (!config->relocatable)
100    flags &= ~(uint64_t)SHF_GROUP;
101  return flags;
102}
103
104// GNU assembler 2.24 and LLVM 4.0.0's MC (the newest release as of
105// March 2017) fail to infer section types for sections starting with
106// ".init_array." or ".fini_array.". They set SHT_PROGBITS instead of
107// SHF_INIT_ARRAY. As a result, the following assembler directive
108// creates ".init_array.100" with SHT_PROGBITS, for example.
109//
110//   .section .init_array.100, "aw"
111//
112// This function forces SHT_{INIT,FINI}_ARRAY so that we can handle
113// incorrect inputs as if they were correct from the beginning.
114static uint64_t getType(uint64_t type, StringRef name) {
115  if (type == SHT_PROGBITS && name.startswith(".init_array."))
116    return SHT_INIT_ARRAY;
117  if (type == SHT_PROGBITS && name.startswith(".fini_array."))
118    return SHT_FINI_ARRAY;
119  return type;
120}
121
122template <class ELFT>
123InputSectionBase::InputSectionBase(ObjFile<ELFT> &file,
124                                   const typename ELFT::Shdr &hdr,
125                                   StringRef name, Kind sectionKind)
126    : InputSectionBase(&file, getFlags(hdr.sh_flags),
127                       getType(hdr.sh_type, name), hdr.sh_entsize, hdr.sh_link,
128                       hdr.sh_info, hdr.sh_addralign,
129                       getSectionContents(file, hdr), name, sectionKind) {
130  // We reject object files having insanely large alignments even though
131  // they are allowed by the spec. I think 4GB is a reasonable limitation.
132  // We might want to relax this in the future.
133  if (hdr.sh_addralign > UINT32_MAX)
134    fatal(toString(&file) + ": section sh_addralign is too large");
135}
136
137size_t InputSectionBase::getSize() const {
138  if (auto *s = dyn_cast<SyntheticSection>(this))
139    return s->getSize();
140  if (uncompressedSize >= 0)
141    return uncompressedSize;
142  return rawData.size();
143}
144
145void InputSectionBase::uncompress() const {
146  size_t size = uncompressedSize;
147  char *uncompressedBuf;
148  {
149    static std::mutex mu;
150    std::lock_guard<std::mutex> lock(mu);
151    uncompressedBuf = bAlloc.Allocate<char>(size);
152  }
153
154  if (Error e = zlib::uncompress(toStringRef(rawData), uncompressedBuf, size))
155    fatal(toString(this) +
156          ": uncompress failed: " + llvm::toString(std::move(e)));
157  rawData = makeArrayRef((uint8_t *)uncompressedBuf, size);
158  uncompressedSize = -1;
159}
160
161uint64_t InputSectionBase::getOffsetInFile() const {
162  const uint8_t *fileStart = (const uint8_t *)file->mb.getBufferStart();
163  const uint8_t *secStart = data().begin();
164  return secStart - fileStart;
165}
166
167uint64_t SectionBase::getOffset(uint64_t offset) const {
168  switch (kind()) {
169  case Output: {
170    auto *os = cast<OutputSection>(this);
171    // For output sections we treat offset -1 as the end of the section.
172    return offset == uint64_t(-1) ? os->size : offset;
173  }
174  case Regular:
175  case Synthetic:
176    return cast<InputSection>(this)->getOffset(offset);
177  case EHFrame:
178    // The file crtbeginT.o has relocations pointing to the start of an empty
179    // .eh_frame that is known to be the first in the link. It does that to
180    // identify the start of the output .eh_frame.
181    return offset;
182  case Merge:
183    const MergeInputSection *ms = cast<MergeInputSection>(this);
184    if (InputSection *isec = ms->getParent())
185      return isec->getOffset(ms->getParentOffset(offset));
186    return ms->getParentOffset(offset);
187  }
188  llvm_unreachable("invalid section kind");
189}
190
191uint64_t SectionBase::getVA(uint64_t offset) const {
192  const OutputSection *out = getOutputSection();
193  return (out ? out->addr : 0) + getOffset(offset);
194}
195
196OutputSection *SectionBase::getOutputSection() {
197  InputSection *sec;
198  if (auto *isec = dyn_cast<InputSection>(this))
199    sec = isec;
200  else if (auto *ms = dyn_cast<MergeInputSection>(this))
201    sec = ms->getParent();
202  else if (auto *eh = dyn_cast<EhInputSection>(this))
203    sec = eh->getParent();
204  else
205    return cast<OutputSection>(this);
206  return sec ? sec->getParent() : nullptr;
207}
208
209// When a section is compressed, `rawData` consists with a header followed
210// by zlib-compressed data. This function parses a header to initialize
211// `uncompressedSize` member and remove the header from `rawData`.
212void InputSectionBase::parseCompressedHeader() {
213  using Chdr64 = typename ELF64LE::Chdr;
214  using Chdr32 = typename ELF32LE::Chdr;
215
216  // Old-style header
217  if (name.startswith(".zdebug")) {
218    if (!toStringRef(rawData).startswith("ZLIB")) {
219      error(toString(this) + ": corrupted compressed section header");
220      return;
221    }
222    rawData = rawData.slice(4);
223
224    if (rawData.size() < 8) {
225      error(toString(this) + ": corrupted compressed section header");
226      return;
227    }
228
229    uncompressedSize = read64be(rawData.data());
230    rawData = rawData.slice(8);
231
232    // Restore the original section name.
233    // (e.g. ".zdebug_info" -> ".debug_info")
234    name = saver.save("." + name.substr(2));
235    return;
236  }
237
238  assert(flags & SHF_COMPRESSED);
239  flags &= ~(uint64_t)SHF_COMPRESSED;
240
241  // New-style 64-bit header
242  if (config->is64) {
243    if (rawData.size() < sizeof(Chdr64)) {
244      error(toString(this) + ": corrupted compressed section");
245      return;
246    }
247
248    auto *hdr = reinterpret_cast<const Chdr64 *>(rawData.data());
249    if (hdr->ch_type != ELFCOMPRESS_ZLIB) {
250      error(toString(this) + ": unsupported compression type");
251      return;
252    }
253
254    uncompressedSize = hdr->ch_size;
255    alignment = std::max<uint32_t>(hdr->ch_addralign, 1);
256    rawData = rawData.slice(sizeof(*hdr));
257    return;
258  }
259
260  // New-style 32-bit header
261  if (rawData.size() < sizeof(Chdr32)) {
262    error(toString(this) + ": corrupted compressed section");
263    return;
264  }
265
266  auto *hdr = reinterpret_cast<const Chdr32 *>(rawData.data());
267  if (hdr->ch_type != ELFCOMPRESS_ZLIB) {
268    error(toString(this) + ": unsupported compression type");
269    return;
270  }
271
272  uncompressedSize = hdr->ch_size;
273  alignment = std::max<uint32_t>(hdr->ch_addralign, 1);
274  rawData = rawData.slice(sizeof(*hdr));
275}
276
277InputSection *InputSectionBase::getLinkOrderDep() const {
278  assert(link);
279  assert(flags & SHF_LINK_ORDER);
280  return cast<InputSection>(file->getSections()[link]);
281}
282
283// Find a function symbol that encloses a given location.
284template <class ELFT>
285Defined *InputSectionBase::getEnclosingFunction(uint64_t offset) {
286  for (Symbol *b : file->getSymbols())
287    if (Defined *d = dyn_cast<Defined>(b))
288      if (d->section == this && d->type == STT_FUNC && d->value <= offset &&
289          offset < d->value + d->size)
290        return d;
291  return nullptr;
292}
293
294// Returns a source location string. Used to construct an error message.
295template <class ELFT>
296std::string InputSectionBase::getLocation(uint64_t offset) {
297  std::string secAndOffset = (name + "+0x" + utohexstr(offset)).str();
298
299  // We don't have file for synthetic sections.
300  if (getFile<ELFT>() == nullptr)
301    return (config->outputFile + ":(" + secAndOffset + ")")
302        .str();
303
304  // First check if we can get desired values from debugging information.
305  if (Optional<DILineInfo> info = getFile<ELFT>()->getDILineInfo(this, offset))
306    return info->FileName + ":" + std::to_string(info->Line) + ":(" +
307           secAndOffset + ")";
308
309  // File->sourceFile contains STT_FILE symbol that contains a
310  // source file name. If it's missing, we use an object file name.
311  std::string srcFile = getFile<ELFT>()->sourceFile;
312  if (srcFile.empty())
313    srcFile = toString(file);
314
315  if (Defined *d = getEnclosingFunction<ELFT>(offset))
316    return srcFile + ":(function " + toString(*d) + ": " + secAndOffset + ")";
317
318  // If there's no symbol, print out the offset in the section.
319  return (srcFile + ":(" + secAndOffset + ")");
320}
321
322// This function is intended to be used for constructing an error message.
323// The returned message looks like this:
324//
325//   foo.c:42 (/home/alice/possibly/very/long/path/foo.c:42)
326//
327//  Returns an empty string if there's no way to get line info.
328std::string InputSectionBase::getSrcMsg(const Symbol &sym, uint64_t offset) {
329  return file->getSrcMsg(sym, *this, offset);
330}
331
332// Returns a filename string along with an optional section name. This
333// function is intended to be used for constructing an error
334// message. The returned message looks like this:
335//
336//   path/to/foo.o:(function bar)
337//
338// or
339//
340//   path/to/foo.o:(function bar) in archive path/to/bar.a
341std::string InputSectionBase::getObjMsg(uint64_t off) {
342  std::string filename = file->getName();
343
344  std::string archive;
345  if (!file->archiveName.empty())
346    archive = " in archive " + file->archiveName;
347
348  // Find a symbol that encloses a given location.
349  for (Symbol *b : file->getSymbols())
350    if (auto *d = dyn_cast<Defined>(b))
351      if (d->section == this && d->value <= off && off < d->value + d->size)
352        return filename + ":(" + toString(*d) + ")" + archive;
353
354  // If there's no symbol, print out the offset in the section.
355  return (filename + ":(" + name + "+0x" + utohexstr(off) + ")" + archive)
356      .str();
357}
358
359InputSection InputSection::discarded(nullptr, 0, 0, 0, ArrayRef<uint8_t>(), "");
360
361InputSection::InputSection(InputFile *f, uint64_t flags, uint32_t type,
362                           uint32_t alignment, ArrayRef<uint8_t> data,
363                           StringRef name, Kind k)
364    : InputSectionBase(f, flags, type,
365                       /*Entsize*/ 0, /*Link*/ 0, /*Info*/ 0, alignment, data,
366                       name, k) {}
367
368template <class ELFT>
369InputSection::InputSection(ObjFile<ELFT> &f, const typename ELFT::Shdr &header,
370                           StringRef name)
371    : InputSectionBase(f, header, name, InputSectionBase::Regular) {}
372
373bool InputSection::classof(const SectionBase *s) {
374  return s->kind() == SectionBase::Regular ||
375         s->kind() == SectionBase::Synthetic;
376}
377
378OutputSection *InputSection::getParent() const {
379  return cast_or_null<OutputSection>(parent);
380}
381
382// Copy SHT_GROUP section contents. Used only for the -r option.
383template <class ELFT> void InputSection::copyShtGroup(uint8_t *buf) {
384  // ELFT::Word is the 32-bit integral type in the target endianness.
385  using u32 = typename ELFT::Word;
386  ArrayRef<u32> from = getDataAs<u32>();
387  auto *to = reinterpret_cast<u32 *>(buf);
388
389  // The first entry is not a section number but a flag.
390  *to++ = from[0];
391
392  // Adjust section numbers because section numbers in an input object
393  // files are different in the output.
394  ArrayRef<InputSectionBase *> sections = file->getSections();
395  for (uint32_t idx : from.slice(1))
396    *to++ = sections[idx]->getOutputSection()->sectionIndex;
397}
398
399InputSectionBase *InputSection::getRelocatedSection() const {
400  if (!file || (type != SHT_RELA && type != SHT_REL))
401    return nullptr;
402  ArrayRef<InputSectionBase *> sections = file->getSections();
403  return sections[info];
404}
405
406// This is used for -r and --emit-relocs. We can't use memcpy to copy
407// relocations because we need to update symbol table offset and section index
408// for each relocation. So we copy relocations one by one.
409template <class ELFT, class RelTy>
410void InputSection::copyRelocations(uint8_t *buf, ArrayRef<RelTy> rels) {
411  InputSectionBase *sec = getRelocatedSection();
412
413  for (const RelTy &rel : rels) {
414    RelType type = rel.getType(config->isMips64EL);
415    const ObjFile<ELFT> *file = getFile<ELFT>();
416    Symbol &sym = file->getRelocTargetSym(rel);
417
418    auto *p = reinterpret_cast<typename ELFT::Rela *>(buf);
419    buf += sizeof(RelTy);
420
421    if (RelTy::IsRela)
422      p->r_addend = getAddend<ELFT>(rel);
423
424    // Output section VA is zero for -r, so r_offset is an offset within the
425    // section, but for --emit-relocs it is an virtual address.
426    p->r_offset = sec->getVA(rel.r_offset);
427    p->setSymbolAndType(in.symTab->getSymbolIndex(&sym), type,
428                        config->isMips64EL);
429
430    if (sym.type == STT_SECTION) {
431      // We combine multiple section symbols into only one per
432      // section. This means we have to update the addend. That is
433      // trivial for Elf_Rela, but for Elf_Rel we have to write to the
434      // section data. We do that by adding to the Relocation vector.
435
436      // .eh_frame is horribly special and can reference discarded sections. To
437      // avoid having to parse and recreate .eh_frame, we just replace any
438      // relocation in it pointing to discarded sections with R_*_NONE, which
439      // hopefully creates a frame that is ignored at runtime. Also, don't warn
440      // on .gcc_except_table and debug sections.
441      //
442      // See the comment in maybeReportUndefined for PPC64 .toc .
443      auto *d = dyn_cast<Defined>(&sym);
444      if (!d) {
445        if (!sec->name.startswith(".debug") &&
446            !sec->name.startswith(".zdebug") && sec->name != ".eh_frame" &&
447            sec->name != ".gcc_except_table" && sec->name != ".toc") {
448          uint32_t secIdx = cast<Undefined>(sym).discardedSecIdx;
449          Elf_Shdr_Impl<ELFT> sec =
450              CHECK(file->getObj().sections(), file)[secIdx];
451          warn("relocation refers to a discarded section: " +
452               CHECK(file->getObj().getSectionName(&sec), file) +
453               "\n>>> referenced by " + getObjMsg(p->r_offset));
454        }
455        p->setSymbolAndType(0, 0, false);
456        continue;
457      }
458      SectionBase *section = d->section->repl;
459      if (!section->isLive()) {
460        p->setSymbolAndType(0, 0, false);
461        continue;
462      }
463
464      int64_t addend = getAddend<ELFT>(rel);
465      const uint8_t *bufLoc = sec->data().begin() + rel.r_offset;
466      if (!RelTy::IsRela)
467        addend = target->getImplicitAddend(bufLoc, type);
468
469      if (config->emachine == EM_MIPS && config->relocatable &&
470          target->getRelExpr(type, sym, bufLoc) == R_MIPS_GOTREL) {
471        // Some MIPS relocations depend on "gp" value. By default,
472        // this value has 0x7ff0 offset from a .got section. But
473        // relocatable files produced by a complier or a linker
474        // might redefine this default value and we must use it
475        // for a calculation of the relocation result. When we
476        // generate EXE or DSO it's trivial. Generating a relocatable
477        // output is more difficult case because the linker does
478        // not calculate relocations in this mode and loses
479        // individual "gp" values used by each input object file.
480        // As a workaround we add the "gp" value to the relocation
481        // addend and save it back to the file.
482        addend += sec->getFile<ELFT>()->mipsGp0;
483      }
484
485      if (RelTy::IsRela)
486        p->r_addend = sym.getVA(addend) - section->getOutputSection()->addr;
487      else if (config->relocatable && type != target->noneRel)
488        sec->relocations.push_back({R_ABS, type, rel.r_offset, addend, &sym});
489    }
490  }
491}
492
493// The ARM and AArch64 ABI handle pc-relative relocations to undefined weak
494// references specially. The general rule is that the value of the symbol in
495// this context is the address of the place P. A further special case is that
496// branch relocations to an undefined weak reference resolve to the next
497// instruction.
498static uint32_t getARMUndefinedRelativeWeakVA(RelType type, uint32_t a,
499                                              uint32_t p) {
500  switch (type) {
501  // Unresolved branch relocations to weak references resolve to next
502  // instruction, this will be either 2 or 4 bytes on from P.
503  case R_ARM_THM_JUMP11:
504    return p + 2 + a;
505  case R_ARM_CALL:
506  case R_ARM_JUMP24:
507  case R_ARM_PC24:
508  case R_ARM_PLT32:
509  case R_ARM_PREL31:
510  case R_ARM_THM_JUMP19:
511  case R_ARM_THM_JUMP24:
512    return p + 4 + a;
513  case R_ARM_THM_CALL:
514    // We don't want an interworking BLX to ARM
515    return p + 5 + a;
516  // Unresolved non branch pc-relative relocations
517  // R_ARM_TARGET2 which can be resolved relatively is not present as it never
518  // targets a weak-reference.
519  case R_ARM_MOVW_PREL_NC:
520  case R_ARM_MOVT_PREL:
521  case R_ARM_REL32:
522  case R_ARM_THM_MOVW_PREL_NC:
523  case R_ARM_THM_MOVT_PREL:
524    return p + a;
525  }
526  llvm_unreachable("ARM pc-relative relocation expected\n");
527}
528
529// The comment above getARMUndefinedRelativeWeakVA applies to this function.
530static uint64_t getAArch64UndefinedRelativeWeakVA(uint64_t type, uint64_t a,
531                                                  uint64_t p) {
532  switch (type) {
533  // Unresolved branch relocations to weak references resolve to next
534  // instruction, this is 4 bytes on from P.
535  case R_AARCH64_CALL26:
536  case R_AARCH64_CONDBR19:
537  case R_AARCH64_JUMP26:
538  case R_AARCH64_TSTBR14:
539    return p + 4 + a;
540  // Unresolved non branch pc-relative relocations
541  case R_AARCH64_PREL16:
542  case R_AARCH64_PREL32:
543  case R_AARCH64_PREL64:
544  case R_AARCH64_ADR_PREL_LO21:
545  case R_AARCH64_LD_PREL_LO19:
546    return p + a;
547  }
548  llvm_unreachable("AArch64 pc-relative relocation expected\n");
549}
550
551// ARM SBREL relocations are of the form S + A - B where B is the static base
552// The ARM ABI defines base to be "addressing origin of the output segment
553// defining the symbol S". We defined the "addressing origin"/static base to be
554// the base of the PT_LOAD segment containing the Sym.
555// The procedure call standard only defines a Read Write Position Independent
556// RWPI variant so in practice we should expect the static base to be the base
557// of the RW segment.
558static uint64_t getARMStaticBase(const Symbol &sym) {
559  OutputSection *os = sym.getOutputSection();
560  if (!os || !os->ptLoad || !os->ptLoad->firstSec)
561    fatal("SBREL relocation to " + sym.getName() + " without static base");
562  return os->ptLoad->firstSec->addr;
563}
564
565// For R_RISCV_PC_INDIRECT (R_RISCV_PCREL_LO12_{I,S}), the symbol actually
566// points the corresponding R_RISCV_PCREL_HI20 relocation, and the target VA
567// is calculated using PCREL_HI20's symbol.
568//
569// This function returns the R_RISCV_PCREL_HI20 relocation from
570// R_RISCV_PCREL_LO12's symbol and addend.
571static Relocation *getRISCVPCRelHi20(const Symbol *sym, uint64_t addend) {
572  const Defined *d = cast<Defined>(sym);
573  if (!d->section) {
574    error("R_RISCV_PCREL_LO12 relocation points to an absolute symbol: " +
575          sym->getName());
576    return nullptr;
577  }
578  InputSection *isec = cast<InputSection>(d->section);
579
580  if (addend != 0)
581    warn("Non-zero addend in R_RISCV_PCREL_LO12 relocation to " +
582         isec->getObjMsg(d->value) + " is ignored");
583
584  // Relocations are sorted by offset, so we can use std::equal_range to do
585  // binary search.
586  Relocation r;
587  r.offset = d->value;
588  auto range =
589      std::equal_range(isec->relocations.begin(), isec->relocations.end(), r,
590                       [](const Relocation &lhs, const Relocation &rhs) {
591                         return lhs.offset < rhs.offset;
592                       });
593
594  for (auto it = range.first; it != range.second; ++it)
595    if (it->type == R_RISCV_PCREL_HI20 || it->type == R_RISCV_GOT_HI20 ||
596        it->type == R_RISCV_TLS_GD_HI20 || it->type == R_RISCV_TLS_GOT_HI20)
597      return &*it;
598
599  error("R_RISCV_PCREL_LO12 relocation points to " + isec->getObjMsg(d->value) +
600        " without an associated R_RISCV_PCREL_HI20 relocation");
601  return nullptr;
602}
603
604// A TLS symbol's virtual address is relative to the TLS segment. Add a
605// target-specific adjustment to produce a thread-pointer-relative offset.
606static int64_t getTlsTpOffset(const Symbol &s) {
607  // On targets that support TLSDESC, _TLS_MODULE_BASE_@tpoff = 0.
608  if (&s == ElfSym::tlsModuleBase)
609    return 0;
610
611  switch (config->emachine) {
612  case EM_ARM:
613  case EM_AARCH64:
614    // Variant 1. The thread pointer points to a TCB with a fixed 2-word size,
615    // followed by a variable amount of alignment padding, followed by the TLS
616    // segment.
617    return s.getVA(0) + alignTo(config->wordsize * 2, Out::tlsPhdr->p_align);
618  case EM_386:
619  case EM_X86_64:
620    // Variant 2. The TLS segment is located just before the thread pointer.
621    return s.getVA(0) - alignTo(Out::tlsPhdr->p_memsz, Out::tlsPhdr->p_align);
622  case EM_PPC:
623  case EM_PPC64:
624    // The thread pointer points to a fixed offset from the start of the
625    // executable's TLS segment. An offset of 0x7000 allows a signed 16-bit
626    // offset to reach 0x1000 of TCB/thread-library data and 0xf000 of the
627    // program's TLS segment.
628    return s.getVA(0) - 0x7000;
629  case EM_RISCV:
630    return s.getVA(0);
631  default:
632    llvm_unreachable("unhandled Config->EMachine");
633  }
634}
635
636static uint64_t getRelocTargetVA(const InputFile *file, RelType type, int64_t a,
637                                 uint64_t p, const Symbol &sym, RelExpr expr) {
638  switch (expr) {
639  case R_ABS:
640  case R_DTPREL:
641  case R_RELAX_TLS_LD_TO_LE_ABS:
642  case R_RELAX_GOT_PC_NOPIC:
643  case R_RISCV_ADD:
644    return sym.getVA(a);
645  case R_ADDEND:
646    return a;
647  case R_ARM_SBREL:
648    return sym.getVA(a) - getARMStaticBase(sym);
649  case R_GOT:
650  case R_RELAX_TLS_GD_TO_IE_ABS:
651    return sym.getGotVA() + a;
652  case R_GOTONLY_PC:
653    return in.got->getVA() + a - p;
654  case R_GOTPLTONLY_PC:
655    return in.gotPlt->getVA() + a - p;
656  case R_GOTREL:
657  case R_PPC64_RELAX_TOC:
658    return sym.getVA(a) - in.got->getVA();
659  case R_GOTPLTREL:
660    return sym.getVA(a) - in.gotPlt->getVA();
661  case R_GOTPLT:
662  case R_RELAX_TLS_GD_TO_IE_GOTPLT:
663    return sym.getGotVA() + a - in.gotPlt->getVA();
664  case R_TLSLD_GOT_OFF:
665  case R_GOT_OFF:
666  case R_RELAX_TLS_GD_TO_IE_GOT_OFF:
667    return sym.getGotOffset() + a;
668  case R_AARCH64_GOT_PAGE_PC:
669  case R_AARCH64_RELAX_TLS_GD_TO_IE_PAGE_PC:
670    return getAArch64Page(sym.getGotVA() + a) - getAArch64Page(p);
671  case R_GOT_PC:
672  case R_RELAX_TLS_GD_TO_IE:
673    return sym.getGotVA() + a - p;
674  case R_HEXAGON_GOT:
675    return sym.getGotVA() - in.gotPlt->getVA();
676  case R_MIPS_GOTREL:
677    return sym.getVA(a) - in.mipsGot->getGp(file);
678  case R_MIPS_GOT_GP:
679    return in.mipsGot->getGp(file) + a;
680  case R_MIPS_GOT_GP_PC: {
681    // R_MIPS_LO16 expression has R_MIPS_GOT_GP_PC type iif the target
682    // is _gp_disp symbol. In that case we should use the following
683    // formula for calculation "AHL + GP - P + 4". For details see p. 4-19 at
684    // ftp://www.linux-mips.org/pub/linux/mips/doc/ABI/mipsabi.pdf
685    // microMIPS variants of these relocations use slightly different
686    // expressions: AHL + GP - P + 3 for %lo() and AHL + GP - P - 1 for %hi()
687    // to correctly handle less-sugnificant bit of the microMIPS symbol.
688    uint64_t v = in.mipsGot->getGp(file) + a - p;
689    if (type == R_MIPS_LO16 || type == R_MICROMIPS_LO16)
690      v += 4;
691    if (type == R_MICROMIPS_LO16 || type == R_MICROMIPS_HI16)
692      v -= 1;
693    return v;
694  }
695  case R_MIPS_GOT_LOCAL_PAGE:
696    // If relocation against MIPS local symbol requires GOT entry, this entry
697    // should be initialized by 'page address'. This address is high 16-bits
698    // of sum the symbol's value and the addend.
699    return in.mipsGot->getVA() + in.mipsGot->getPageEntryOffset(file, sym, a) -
700           in.mipsGot->getGp(file);
701  case R_MIPS_GOT_OFF:
702  case R_MIPS_GOT_OFF32:
703    // In case of MIPS if a GOT relocation has non-zero addend this addend
704    // should be applied to the GOT entry content not to the GOT entry offset.
705    // That is why we use separate expression type.
706    return in.mipsGot->getVA() + in.mipsGot->getSymEntryOffset(file, sym, a) -
707           in.mipsGot->getGp(file);
708  case R_MIPS_TLSGD:
709    return in.mipsGot->getVA() + in.mipsGot->getGlobalDynOffset(file, sym) -
710           in.mipsGot->getGp(file);
711  case R_MIPS_TLSLD:
712    return in.mipsGot->getVA() + in.mipsGot->getTlsIndexOffset(file) -
713           in.mipsGot->getGp(file);
714  case R_AARCH64_PAGE_PC: {
715    uint64_t val = sym.isUndefWeak() ? p + a : sym.getVA(a);
716    return getAArch64Page(val) - getAArch64Page(p);
717  }
718  case R_RISCV_PC_INDIRECT: {
719    if (const Relocation *hiRel = getRISCVPCRelHi20(&sym, a))
720      return getRelocTargetVA(file, hiRel->type, hiRel->addend, sym.getVA(),
721                              *hiRel->sym, hiRel->expr);
722    return 0;
723  }
724  case R_PC: {
725    uint64_t dest;
726    if (sym.isUndefWeak()) {
727      // On ARM and AArch64 a branch to an undefined weak resolves to the
728      // next instruction, otherwise the place.
729      if (config->emachine == EM_ARM)
730        dest = getARMUndefinedRelativeWeakVA(type, a, p);
731      else if (config->emachine == EM_AARCH64)
732        dest = getAArch64UndefinedRelativeWeakVA(type, a, p);
733      else if (config->emachine == EM_PPC)
734        dest = p;
735      else
736        dest = sym.getVA(a);
737    } else {
738      dest = sym.getVA(a);
739    }
740    return dest - p;
741  }
742  case R_PLT:
743    return sym.getPltVA() + a;
744  case R_PLT_PC:
745  case R_PPC64_CALL_PLT:
746    return sym.getPltVA() + a - p;
747  case R_PPC32_PLTREL:
748    // R_PPC_PLTREL24 uses the addend (usually 0 or 0x8000) to indicate r30
749    // stores _GLOBAL_OFFSET_TABLE_ or .got2+0x8000. The addend is ignored for
750    // target VA compuation.
751    return sym.getPltVA() - p;
752  case R_PPC64_CALL: {
753    uint64_t symVA = sym.getVA(a);
754    // If we have an undefined weak symbol, we might get here with a symbol
755    // address of zero. That could overflow, but the code must be unreachable,
756    // so don't bother doing anything at all.
757    if (!symVA)
758      return 0;
759
760    // PPC64 V2 ABI describes two entry points to a function. The global entry
761    // point is used for calls where the caller and callee (may) have different
762    // TOC base pointers and r2 needs to be modified to hold the TOC base for
763    // the callee. For local calls the caller and callee share the same
764    // TOC base and so the TOC pointer initialization code should be skipped by
765    // branching to the local entry point.
766    return symVA - p + getPPC64GlobalEntryToLocalEntryOffset(sym.stOther);
767  }
768  case R_PPC64_TOCBASE:
769    return getPPC64TocBase() + a;
770  case R_RELAX_GOT_PC:
771    return sym.getVA(a) - p;
772  case R_RELAX_TLS_GD_TO_LE:
773  case R_RELAX_TLS_IE_TO_LE:
774  case R_RELAX_TLS_LD_TO_LE:
775  case R_TLS:
776    // It is not very clear what to return if the symbol is undefined. With
777    // --noinhibit-exec, even a non-weak undefined reference may reach here.
778    // Just return A, which matches R_ABS, and the behavior of some dynamic
779    // loaders.
780    if (sym.isUndefined())
781      return a;
782    return getTlsTpOffset(sym) + a;
783  case R_RELAX_TLS_GD_TO_LE_NEG:
784  case R_NEG_TLS:
785    if (sym.isUndefined())
786      return a;
787    return -getTlsTpOffset(sym) + a;
788  case R_SIZE:
789    return sym.getSize() + a;
790  case R_TLSDESC:
791    return in.got->getGlobalDynAddr(sym) + a;
792  case R_TLSDESC_PC:
793    return in.got->getGlobalDynAddr(sym) + a - p;
794  case R_AARCH64_TLSDESC_PAGE:
795    return getAArch64Page(in.got->getGlobalDynAddr(sym) + a) -
796           getAArch64Page(p);
797  case R_TLSGD_GOT:
798    return in.got->getGlobalDynOffset(sym) + a;
799  case R_TLSGD_GOTPLT:
800    return in.got->getVA() + in.got->getGlobalDynOffset(sym) + a - in.gotPlt->getVA();
801  case R_TLSGD_PC:
802    return in.got->getGlobalDynAddr(sym) + a - p;
803  case R_TLSLD_GOTPLT:
804    return in.got->getVA() + in.got->getTlsIndexOff() + a - in.gotPlt->getVA();
805  case R_TLSLD_GOT:
806    return in.got->getTlsIndexOff() + a;
807  case R_TLSLD_PC:
808    return in.got->getTlsIndexVA() + a - p;
809  default:
810    llvm_unreachable("invalid expression");
811  }
812}
813
814// This function applies relocations to sections without SHF_ALLOC bit.
815// Such sections are never mapped to memory at runtime. Debug sections are
816// an example. Relocations in non-alloc sections are much easier to
817// handle than in allocated sections because it will never need complex
818// treatement such as GOT or PLT (because at runtime no one refers them).
819// So, we handle relocations for non-alloc sections directly in this
820// function as a performance optimization.
821template <class ELFT, class RelTy>
822void InputSection::relocateNonAlloc(uint8_t *buf, ArrayRef<RelTy> rels) {
823  const unsigned bits = sizeof(typename ELFT::uint) * 8;
824
825  for (const RelTy &rel : rels) {
826    RelType type = rel.getType(config->isMips64EL);
827
828    // GCC 8.0 or earlier have a bug that they emit R_386_GOTPC relocations
829    // against _GLOBAL_OFFSET_TABLE_ for .debug_info. The bug has been fixed
830    // in 2017 (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82630), but we
831    // need to keep this bug-compatible code for a while.
832    if (config->emachine == EM_386 && type == R_386_GOTPC)
833      continue;
834
835    uint64_t offset = getOffset(rel.r_offset);
836    uint8_t *bufLoc = buf + offset;
837    int64_t addend = getAddend<ELFT>(rel);
838    if (!RelTy::IsRela)
839      addend += target->getImplicitAddend(bufLoc, type);
840
841    Symbol &sym = getFile<ELFT>()->getRelocTargetSym(rel);
842    RelExpr expr = target->getRelExpr(type, sym, bufLoc);
843    if (expr == R_NONE)
844      continue;
845
846    if (expr != R_ABS && expr != R_DTPREL && expr != R_RISCV_ADD) {
847      std::string msg = getLocation<ELFT>(offset) +
848                        ": has non-ABS relocation " + toString(type) +
849                        " against symbol '" + toString(sym) + "'";
850      if (expr != R_PC) {
851        error(msg);
852        return;
853      }
854
855      // If the control reaches here, we found a PC-relative relocation in a
856      // non-ALLOC section. Since non-ALLOC section is not loaded into memory
857      // at runtime, the notion of PC-relative doesn't make sense here. So,
858      // this is a usage error. However, GNU linkers historically accept such
859      // relocations without any errors and relocate them as if they were at
860      // address 0. For bug-compatibilty, we accept them with warnings. We
861      // know Steel Bank Common Lisp as of 2018 have this bug.
862      warn(msg);
863      target->relocateOne(bufLoc, type,
864                          SignExtend64<bits>(sym.getVA(addend - offset)));
865      continue;
866    }
867
868    if (sym.isTls() && !Out::tlsPhdr)
869      target->relocateOne(bufLoc, type, 0);
870    else
871      target->relocateOne(bufLoc, type, SignExtend64<bits>(sym.getVA(addend)));
872  }
873}
874
875// This is used when '-r' is given.
876// For REL targets, InputSection::copyRelocations() may store artificial
877// relocations aimed to update addends. They are handled in relocateAlloc()
878// for allocatable sections, and this function does the same for
879// non-allocatable sections, such as sections with debug information.
880static void relocateNonAllocForRelocatable(InputSection *sec, uint8_t *buf) {
881  const unsigned bits = config->is64 ? 64 : 32;
882
883  for (const Relocation &rel : sec->relocations) {
884    // InputSection::copyRelocations() adds only R_ABS relocations.
885    assert(rel.expr == R_ABS);
886    uint8_t *bufLoc = buf + rel.offset + sec->outSecOff;
887    uint64_t targetVA = SignExtend64(rel.sym->getVA(rel.addend), bits);
888    target->relocateOne(bufLoc, rel.type, targetVA);
889  }
890}
891
892template <class ELFT>
893void InputSectionBase::relocate(uint8_t *buf, uint8_t *bufEnd) {
894  if (flags & SHF_EXECINSTR)
895    adjustSplitStackFunctionPrologues<ELFT>(buf, bufEnd);
896
897  if (flags & SHF_ALLOC) {
898    relocateAlloc(buf, bufEnd);
899    return;
900  }
901
902  auto *sec = cast<InputSection>(this);
903  if (config->relocatable)
904    relocateNonAllocForRelocatable(sec, buf);
905  else if (sec->areRelocsRela)
906    sec->relocateNonAlloc<ELFT>(buf, sec->template relas<ELFT>());
907  else
908    sec->relocateNonAlloc<ELFT>(buf, sec->template rels<ELFT>());
909}
910
911void InputSectionBase::relocateAlloc(uint8_t *buf, uint8_t *bufEnd) {
912  assert(flags & SHF_ALLOC);
913  const unsigned bits = config->wordsize * 8;
914
915  for (const Relocation &rel : relocations) {
916    uint64_t offset = rel.offset;
917    if (auto *sec = dyn_cast<InputSection>(this))
918      offset += sec->outSecOff;
919    uint8_t *bufLoc = buf + offset;
920    RelType type = rel.type;
921
922    uint64_t addrLoc = getOutputSection()->addr + offset;
923    RelExpr expr = rel.expr;
924    uint64_t targetVA = SignExtend64(
925        getRelocTargetVA(file, type, rel.addend, addrLoc, *rel.sym, expr),
926        bits);
927
928    switch (expr) {
929    case R_RELAX_GOT_PC:
930    case R_RELAX_GOT_PC_NOPIC:
931      target->relaxGot(bufLoc, type, targetVA);
932      break;
933    case R_PPC64_RELAX_TOC:
934      if (!tryRelaxPPC64TocIndirection(type, rel, bufLoc))
935        target->relocateOne(bufLoc, type, targetVA);
936      break;
937    case R_RELAX_TLS_IE_TO_LE:
938      target->relaxTlsIeToLe(bufLoc, type, targetVA);
939      break;
940    case R_RELAX_TLS_LD_TO_LE:
941    case R_RELAX_TLS_LD_TO_LE_ABS:
942      target->relaxTlsLdToLe(bufLoc, type, targetVA);
943      break;
944    case R_RELAX_TLS_GD_TO_LE:
945    case R_RELAX_TLS_GD_TO_LE_NEG:
946      target->relaxTlsGdToLe(bufLoc, type, targetVA);
947      break;
948    case R_AARCH64_RELAX_TLS_GD_TO_IE_PAGE_PC:
949    case R_RELAX_TLS_GD_TO_IE:
950    case R_RELAX_TLS_GD_TO_IE_ABS:
951    case R_RELAX_TLS_GD_TO_IE_GOT_OFF:
952    case R_RELAX_TLS_GD_TO_IE_GOTPLT:
953      target->relaxTlsGdToIe(bufLoc, type, targetVA);
954      break;
955    case R_PPC64_CALL:
956      // If this is a call to __tls_get_addr, it may be part of a TLS
957      // sequence that has been relaxed and turned into a nop. In this
958      // case, we don't want to handle it as a call.
959      if (read32(bufLoc) == 0x60000000) // nop
960        break;
961
962      // Patch a nop (0x60000000) to a ld.
963      if (rel.sym->needsTocRestore) {
964        if (bufLoc + 8 > bufEnd || read32(bufLoc + 4) != 0x60000000) {
965          error(getErrorLocation(bufLoc) + "call lacks nop, can't restore toc");
966          break;
967        }
968        write32(bufLoc + 4, 0xe8410018); // ld %r2, 24(%r1)
969      }
970      target->relocateOne(bufLoc, type, targetVA);
971      break;
972    default:
973      target->relocateOne(bufLoc, type, targetVA);
974      break;
975    }
976  }
977}
978
979// For each function-defining prologue, find any calls to __morestack,
980// and replace them with calls to __morestack_non_split.
981static void switchMorestackCallsToMorestackNonSplit(
982    DenseSet<Defined *> &prologues, std::vector<Relocation *> &morestackCalls) {
983
984  // If the target adjusted a function's prologue, all calls to
985  // __morestack inside that function should be switched to
986  // __morestack_non_split.
987  Symbol *moreStackNonSplit = symtab->find("__morestack_non_split");
988  if (!moreStackNonSplit) {
989    error("Mixing split-stack objects requires a definition of "
990          "__morestack_non_split");
991    return;
992  }
993
994  // Sort both collections to compare addresses efficiently.
995  llvm::sort(morestackCalls, [](const Relocation *l, const Relocation *r) {
996    return l->offset < r->offset;
997  });
998  std::vector<Defined *> functions(prologues.begin(), prologues.end());
999  llvm::sort(functions, [](const Defined *l, const Defined *r) {
1000    return l->value < r->value;
1001  });
1002
1003  auto it = morestackCalls.begin();
1004  for (Defined *f : functions) {
1005    // Find the first call to __morestack within the function.
1006    while (it != morestackCalls.end() && (*it)->offset < f->value)
1007      ++it;
1008    // Adjust all calls inside the function.
1009    while (it != morestackCalls.end() && (*it)->offset < f->value + f->size) {
1010      (*it)->sym = moreStackNonSplit;
1011      ++it;
1012    }
1013  }
1014}
1015
1016static bool enclosingPrologueAttempted(uint64_t offset,
1017                                       const DenseSet<Defined *> &prologues) {
1018  for (Defined *f : prologues)
1019    if (f->value <= offset && offset < f->value + f->size)
1020      return true;
1021  return false;
1022}
1023
1024// If a function compiled for split stack calls a function not
1025// compiled for split stack, then the caller needs its prologue
1026// adjusted to ensure that the called function will have enough stack
1027// available. Find those functions, and adjust their prologues.
1028template <class ELFT>
1029void InputSectionBase::adjustSplitStackFunctionPrologues(uint8_t *buf,
1030                                                         uint8_t *end) {
1031  if (!getFile<ELFT>()->splitStack)
1032    return;
1033  DenseSet<Defined *> prologues;
1034  std::vector<Relocation *> morestackCalls;
1035
1036  for (Relocation &rel : relocations) {
1037    // Local symbols can't possibly be cross-calls, and should have been
1038    // resolved long before this line.
1039    if (rel.sym->isLocal())
1040      continue;
1041
1042    // Ignore calls into the split-stack api.
1043    if (rel.sym->getName().startswith("__morestack")) {
1044      if (rel.sym->getName().equals("__morestack"))
1045        morestackCalls.push_back(&rel);
1046      continue;
1047    }
1048
1049    // A relocation to non-function isn't relevant. Sometimes
1050    // __morestack is not marked as a function, so this check comes
1051    // after the name check.
1052    if (rel.sym->type != STT_FUNC)
1053      continue;
1054
1055    // If the callee's-file was compiled with split stack, nothing to do.  In
1056    // this context, a "Defined" symbol is one "defined by the binary currently
1057    // being produced". So an "undefined" symbol might be provided by a shared
1058    // library. It is not possible to tell how such symbols were compiled, so be
1059    // conservative.
1060    if (Defined *d = dyn_cast<Defined>(rel.sym))
1061      if (InputSection *isec = cast_or_null<InputSection>(d->section))
1062        if (!isec || !isec->getFile<ELFT>() || isec->getFile<ELFT>()->splitStack)
1063          continue;
1064
1065    if (enclosingPrologueAttempted(rel.offset, prologues))
1066      continue;
1067
1068    if (Defined *f = getEnclosingFunction<ELFT>(rel.offset)) {
1069      prologues.insert(f);
1070      if (target->adjustPrologueForCrossSplitStack(buf + getOffset(f->value),
1071                                                   end, f->stOther))
1072        continue;
1073      if (!getFile<ELFT>()->someNoSplitStack)
1074        error(lld::toString(this) + ": " + f->getName() +
1075              " (with -fsplit-stack) calls " + rel.sym->getName() +
1076              " (without -fsplit-stack), but couldn't adjust its prologue");
1077    }
1078  }
1079
1080  if (target->needsMoreStackNonSplit)
1081    switchMorestackCallsToMorestackNonSplit(prologues, morestackCalls);
1082}
1083
1084template <class ELFT> void InputSection::writeTo(uint8_t *buf) {
1085  if (type == SHT_NOBITS)
1086    return;
1087
1088  if (auto *s = dyn_cast<SyntheticSection>(this)) {
1089    s->writeTo(buf + outSecOff);
1090    return;
1091  }
1092
1093  // If -r or --emit-relocs is given, then an InputSection
1094  // may be a relocation section.
1095  if (type == SHT_RELA) {
1096    copyRelocations<ELFT>(buf + outSecOff, getDataAs<typename ELFT::Rela>());
1097    return;
1098  }
1099  if (type == SHT_REL) {
1100    copyRelocations<ELFT>(buf + outSecOff, getDataAs<typename ELFT::Rel>());
1101    return;
1102  }
1103
1104  // If -r is given, we may have a SHT_GROUP section.
1105  if (type == SHT_GROUP) {
1106    copyShtGroup<ELFT>(buf + outSecOff);
1107    return;
1108  }
1109
1110  // If this is a compressed section, uncompress section contents directly
1111  // to the buffer.
1112  if (uncompressedSize >= 0) {
1113    size_t size = uncompressedSize;
1114    if (Error e = zlib::uncompress(toStringRef(rawData),
1115                                   (char *)(buf + outSecOff), size))
1116      fatal(toString(this) +
1117            ": uncompress failed: " + llvm::toString(std::move(e)));
1118    uint8_t *bufEnd = buf + outSecOff + size;
1119    relocate<ELFT>(buf, bufEnd);
1120    return;
1121  }
1122
1123  // Copy section contents from source object file to output file
1124  // and then apply relocations.
1125  memcpy(buf + outSecOff, data().data(), data().size());
1126  uint8_t *bufEnd = buf + outSecOff + data().size();
1127  relocate<ELFT>(buf, bufEnd);
1128}
1129
1130void InputSection::replace(InputSection *other) {
1131  alignment = std::max(alignment, other->alignment);
1132
1133  // When a section is replaced with another section that was allocated to
1134  // another partition, the replacement section (and its associated sections)
1135  // need to be placed in the main partition so that both partitions will be
1136  // able to access it.
1137  if (partition != other->partition) {
1138    partition = 1;
1139    for (InputSection *isec : dependentSections)
1140      isec->partition = 1;
1141  }
1142
1143  other->repl = repl;
1144  other->markDead();
1145}
1146
1147template <class ELFT>
1148EhInputSection::EhInputSection(ObjFile<ELFT> &f,
1149                               const typename ELFT::Shdr &header,
1150                               StringRef name)
1151    : InputSectionBase(f, header, name, InputSectionBase::EHFrame) {}
1152
1153SyntheticSection *EhInputSection::getParent() const {
1154  return cast_or_null<SyntheticSection>(parent);
1155}
1156
1157// Returns the index of the first relocation that points to a region between
1158// Begin and Begin+Size.
1159template <class IntTy, class RelTy>
1160static unsigned getReloc(IntTy begin, IntTy size, const ArrayRef<RelTy> &rels,
1161                         unsigned &relocI) {
1162  // Start search from RelocI for fast access. That works because the
1163  // relocations are sorted in .eh_frame.
1164  for (unsigned n = rels.size(); relocI < n; ++relocI) {
1165    const RelTy &rel = rels[relocI];
1166    if (rel.r_offset < begin)
1167      continue;
1168
1169    if (rel.r_offset < begin + size)
1170      return relocI;
1171    return -1;
1172  }
1173  return -1;
1174}
1175
1176// .eh_frame is a sequence of CIE or FDE records.
1177// This function splits an input section into records and returns them.
1178template <class ELFT> void EhInputSection::split() {
1179  if (areRelocsRela)
1180    split<ELFT>(relas<ELFT>());
1181  else
1182    split<ELFT>(rels<ELFT>());
1183}
1184
1185template <class ELFT, class RelTy>
1186void EhInputSection::split(ArrayRef<RelTy> rels) {
1187  unsigned relI = 0;
1188  for (size_t off = 0, end = data().size(); off != end;) {
1189    size_t size = readEhRecordSize(this, off);
1190    pieces.emplace_back(off, this, size, getReloc(off, size, rels, relI));
1191    // The empty record is the end marker.
1192    if (size == 4)
1193      break;
1194    off += size;
1195  }
1196}
1197
1198static size_t findNull(StringRef s, size_t entSize) {
1199  // Optimize the common case.
1200  if (entSize == 1)
1201    return s.find(0);
1202
1203  for (unsigned i = 0, n = s.size(); i != n; i += entSize) {
1204    const char *b = s.begin() + i;
1205    if (std::all_of(b, b + entSize, [](char c) { return c == 0; }))
1206      return i;
1207  }
1208  return StringRef::npos;
1209}
1210
1211SyntheticSection *MergeInputSection::getParent() const {
1212  return cast_or_null<SyntheticSection>(parent);
1213}
1214
1215// Split SHF_STRINGS section. Such section is a sequence of
1216// null-terminated strings.
1217void MergeInputSection::splitStrings(ArrayRef<uint8_t> data, size_t entSize) {
1218  size_t off = 0;
1219  bool isAlloc = flags & SHF_ALLOC;
1220  StringRef s = toStringRef(data);
1221
1222  while (!s.empty()) {
1223    size_t end = findNull(s, entSize);
1224    if (end == StringRef::npos)
1225      fatal(toString(this) + ": string is not null terminated");
1226    size_t size = end + entSize;
1227
1228    pieces.emplace_back(off, xxHash64(s.substr(0, size)), !isAlloc);
1229    s = s.substr(size);
1230    off += size;
1231  }
1232}
1233
1234// Split non-SHF_STRINGS section. Such section is a sequence of
1235// fixed size records.
1236void MergeInputSection::splitNonStrings(ArrayRef<uint8_t> data,
1237                                        size_t entSize) {
1238  size_t size = data.size();
1239  assert((size % entSize) == 0);
1240  bool isAlloc = flags & SHF_ALLOC;
1241
1242  for (size_t i = 0; i != size; i += entSize)
1243    pieces.emplace_back(i, xxHash64(data.slice(i, entSize)), !isAlloc);
1244}
1245
1246template <class ELFT>
1247MergeInputSection::MergeInputSection(ObjFile<ELFT> &f,
1248                                     const typename ELFT::Shdr &header,
1249                                     StringRef name)
1250    : InputSectionBase(f, header, name, InputSectionBase::Merge) {}
1251
1252MergeInputSection::MergeInputSection(uint64_t flags, uint32_t type,
1253                                     uint64_t entsize, ArrayRef<uint8_t> data,
1254                                     StringRef name)
1255    : InputSectionBase(nullptr, flags, type, entsize, /*Link*/ 0, /*Info*/ 0,
1256                       /*Alignment*/ entsize, data, name, SectionBase::Merge) {}
1257
1258// This function is called after we obtain a complete list of input sections
1259// that need to be linked. This is responsible to split section contents
1260// into small chunks for further processing.
1261//
1262// Note that this function is called from parallelForEach. This must be
1263// thread-safe (i.e. no memory allocation from the pools).
1264void MergeInputSection::splitIntoPieces() {
1265  assert(pieces.empty());
1266
1267  if (flags & SHF_STRINGS)
1268    splitStrings(data(), entsize);
1269  else
1270    splitNonStrings(data(), entsize);
1271}
1272
1273SectionPiece *MergeInputSection::getSectionPiece(uint64_t offset) {
1274  if (this->data().size() <= offset)
1275    fatal(toString(this) + ": offset is outside the section");
1276
1277  // If Offset is not at beginning of a section piece, it is not in the map.
1278  // In that case we need to  do a binary search of the original section piece vector.
1279  auto it = partition_point(
1280      pieces, [=](SectionPiece p) { return p.inputOff <= offset; });
1281  return &it[-1];
1282}
1283
1284// Returns the offset in an output section for a given input offset.
1285// Because contents of a mergeable section is not contiguous in output,
1286// it is not just an addition to a base output offset.
1287uint64_t MergeInputSection::getParentOffset(uint64_t offset) const {
1288  // If Offset is not at beginning of a section piece, it is not in the map.
1289  // In that case we need to search from the original section piece vector.
1290  const SectionPiece &piece =
1291      *(const_cast<MergeInputSection *>(this)->getSectionPiece (offset));
1292  uint64_t addend = offset - piece.inputOff;
1293  return piece.outputOff + addend;
1294}
1295
1296template InputSection::InputSection(ObjFile<ELF32LE> &, const ELF32LE::Shdr &,
1297                                    StringRef);
1298template InputSection::InputSection(ObjFile<ELF32BE> &, const ELF32BE::Shdr &,
1299                                    StringRef);
1300template InputSection::InputSection(ObjFile<ELF64LE> &, const ELF64LE::Shdr &,
1301                                    StringRef);
1302template InputSection::InputSection(ObjFile<ELF64BE> &, const ELF64BE::Shdr &,
1303                                    StringRef);
1304
1305template std::string InputSectionBase::getLocation<ELF32LE>(uint64_t);
1306template std::string InputSectionBase::getLocation<ELF32BE>(uint64_t);
1307template std::string InputSectionBase::getLocation<ELF64LE>(uint64_t);
1308template std::string InputSectionBase::getLocation<ELF64BE>(uint64_t);
1309
1310template void InputSection::writeTo<ELF32LE>(uint8_t *);
1311template void InputSection::writeTo<ELF32BE>(uint8_t *);
1312template void InputSection::writeTo<ELF64LE>(uint8_t *);
1313template void InputSection::writeTo<ELF64BE>(uint8_t *);
1314
1315template MergeInputSection::MergeInputSection(ObjFile<ELF32LE> &,
1316                                              const ELF32LE::Shdr &, StringRef);
1317template MergeInputSection::MergeInputSection(ObjFile<ELF32BE> &,
1318                                              const ELF32BE::Shdr &, StringRef);
1319template MergeInputSection::MergeInputSection(ObjFile<ELF64LE> &,
1320                                              const ELF64LE::Shdr &, StringRef);
1321template MergeInputSection::MergeInputSection(ObjFile<ELF64BE> &,
1322                                              const ELF64BE::Shdr &, StringRef);
1323
1324template EhInputSection::EhInputSection(ObjFile<ELF32LE> &,
1325                                        const ELF32LE::Shdr &, StringRef);
1326template EhInputSection::EhInputSection(ObjFile<ELF32BE> &,
1327                                        const ELF32BE::Shdr &, StringRef);
1328template EhInputSection::EhInputSection(ObjFile<ELF64LE> &,
1329                                        const ELF64LE::Shdr &, StringRef);
1330template EhInputSection::EhInputSection(ObjFile<ELF64BE> &,
1331                                        const ELF64BE::Shdr &, StringRef);
1332
1333template void EhInputSection::split<ELF32LE>();
1334template void EhInputSection::split<ELF32BE>();
1335template void EhInputSection::split<ELF64LE>();
1336template void EhInputSection::split<ELF64BE>();
1337