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 "Error.h"
14 #include "InputFiles.h"
15 #include "LinkerScript.h"
16 #include "Memory.h"
17 #include "OutputSections.h"
18 #include "Relocations.h"
19 #include "SyntheticSections.h"
20 #include "Target.h"
21 #include "Thunks.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"
26 #include <mutex>
27 
28 using namespace llvm;
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;
34 
35 using namespace lld;
36 using namespace lld::elf;
37 
38 std::vector<InputSectionBase *> elf::InputSections;
39 
40 // Returns a string to construct an error message.
41 std::string lld::toString(const InputSectionBase *Sec) {
42   // File can be absent if section is synthetic.
43   std::string FileName = Sec->File ? Sec->File->getName() : "<internal>";
44   return (FileName + ":(" + Sec->Name + ")").str();
45 }
46 
47 template <class ELFT>
48 static ArrayRef<uint8_t> getSectionContents(elf::ObjectFile<ELFT> *File,
49                                             const typename ELFT::Shdr *Hdr) {
50   if (!File || Hdr->sh_type == SHT_NOBITS)
51     return makeArrayRef<uint8_t>(nullptr, Hdr->sh_size);
52   return check(File->getObj().getSectionContents(Hdr));
53 }
54 
55 InputSectionBase::InputSectionBase(InputFile *File, uint64_t Flags,
56                                    uint32_t Type, uint64_t Entsize,
57                                    uint32_t Link, uint32_t Info,
58                                    uint32_t Alignment, ArrayRef<uint8_t> Data,
59                                    StringRef Name, Kind SectionKind)
60     : SectionBase(SectionKind, Name, Flags, Entsize, Alignment, Type, Info,
61                   Link),
62       File(File), Data(Data), Repl(this) {
63   Live = !Config->GcSections || !(Flags & SHF_ALLOC);
64   Assigned = false;
65   NumRelocations = 0;
66   AreRelocsRela = false;
67 
68   // The ELF spec states that a value of 0 means the section has
69   // no alignment constraits.
70   uint32_t V = std::max<uint64_t>(Alignment, 1);
71   if (!isPowerOf2_64(V))
72     fatal(toString(File) + ": section sh_addralign is not a power of 2");
73   this->Alignment = V;
74 }
75 
76 // GNU assembler 2.24 and LLVM 4.0.0's MC (the newest release as of
77 // March 2017) fail to infer section types for sections starting with
78 // ".init_array." or ".fini_array.". They set SHT_PROGBITS instead of
79 // SHF_INIT_ARRAY. As a result, the following assembler directive
80 // creates ".init_array.100" with SHT_PROGBITS, for example.
81 //
82 //   .section .init_array.100, "aw"
83 //
84 // This function forces SHT_{INIT,FINI}_ARRAY so that we can handle
85 // incorrect inputs as if they were correct from the beginning.
86 static uint64_t getType(uint64_t Type, StringRef Name) {
87   if (Type == SHT_PROGBITS && Name.startswith(".init_array."))
88     return SHT_INIT_ARRAY;
89   if (Type == SHT_PROGBITS && Name.startswith(".fini_array."))
90     return SHT_FINI_ARRAY;
91   return Type;
92 }
93 
94 template <class ELFT>
95 InputSectionBase::InputSectionBase(elf::ObjectFile<ELFT> *File,
96                                    const typename ELFT::Shdr *Hdr,
97                                    StringRef Name, Kind SectionKind)
98     : InputSectionBase(File, Hdr->sh_flags & ~SHF_INFO_LINK,
99                        getType(Hdr->sh_type, Name), Hdr->sh_entsize,
100                        Hdr->sh_link, Hdr->sh_info, Hdr->sh_addralign,
101                        getSectionContents(File, Hdr), Name, SectionKind) {
102   // We reject object files having insanely large alignments even though
103   // they are allowed by the spec. I think 4GB is a reasonable limitation.
104   // We might want to relax this in the future.
105   if (Hdr->sh_addralign > UINT32_MAX)
106     fatal(toString(File) + ": section sh_addralign is too large");
107 }
108 
109 size_t InputSectionBase::getSize() const {
110   if (auto *S = dyn_cast<SyntheticSection>(this))
111     return S->getSize();
112 
113   return Data.size();
114 }
115 
116 uint64_t InputSectionBase::getOffsetInFile() const {
117   const uint8_t *FileStart = (const uint8_t *)File->MB.getBufferStart();
118   const uint8_t *SecStart = Data.begin();
119   return SecStart - FileStart;
120 }
121 
122 uint64_t SectionBase::getOffset(uint64_t Offset) const {
123   switch (kind()) {
124   case Output: {
125     auto *OS = cast<OutputSection>(this);
126     // For output sections we treat offset -1 as the end of the section.
127     return Offset == uint64_t(-1) ? OS->Size : Offset;
128   }
129   case Regular:
130     return cast<InputSection>(this)->OutSecOff + Offset;
131   case Synthetic: {
132     auto *IS = cast<InputSection>(this);
133     // For synthetic sections we treat offset -1 as the end of the section.
134     return IS->OutSecOff + (Offset == uint64_t(-1) ? IS->getSize() : Offset);
135   }
136   case EHFrame:
137     // The file crtbeginT.o has relocations pointing to the start of an empty
138     // .eh_frame that is known to be the first in the link. It does that to
139     // identify the start of the output .eh_frame.
140     return Offset;
141   case Merge:
142     const MergeInputSection *MS = cast<MergeInputSection>(this);
143     if (MS->MergeSec)
144       return MS->MergeSec->OutSecOff + MS->getOffset(Offset);
145     return MS->getOffset(Offset);
146   }
147   llvm_unreachable("invalid section kind");
148 }
149 
150 OutputSection *SectionBase::getOutputSection() {
151   if (auto *IS = dyn_cast<InputSection>(this))
152     return IS->OutSec;
153   if (auto *MS = dyn_cast<MergeInputSection>(this))
154     return MS->MergeSec ? MS->MergeSec->OutSec : nullptr;
155   if (auto *EH = dyn_cast<EhInputSection>(this))
156     return EH->EHSec->OutSec;
157   return cast<OutputSection>(this);
158 }
159 
160 // Uncompress section contents. Note that this function is called
161 // from parallel_for_each, so it must be thread-safe.
162 void InputSectionBase::uncompress() {
163   Decompressor Dec = check(Decompressor::create(Name, toStringRef(Data),
164                                                 Config->IsLE, Config->Is64));
165 
166   size_t Size = Dec.getDecompressedSize();
167   char *OutputBuf;
168   {
169     static std::mutex Mu;
170     std::lock_guard<std::mutex> Lock(Mu);
171     OutputBuf = BAlloc.Allocate<char>(Size);
172   }
173 
174   if (Error E = Dec.decompress({OutputBuf, Size}))
175     fatal(toString(this) +
176           ": decompress failed: " + llvm::toString(std::move(E)));
177   Data = ArrayRef<uint8_t>((uint8_t *)OutputBuf, Size);
178 }
179 
180 uint64_t SectionBase::getOffset(const DefinedRegular &Sym) const {
181   return getOffset(Sym.Value);
182 }
183 
184 InputSectionBase *InputSectionBase::getLinkOrderDep() const {
185   if ((Flags & SHF_LINK_ORDER) && Link != 0)
186     return File->getSections()[Link];
187   return nullptr;
188 }
189 
190 // Returns a source location string. Used to construct an error message.
191 template <class ELFT>
192 std::string InputSectionBase::getLocation(uint64_t Offset) {
193   // We don't have file for synthetic sections.
194   if (getFile<ELFT>() == nullptr)
195     return (Config->OutputFile + ":(" + Name + "+0x" + utohexstr(Offset) + ")")
196         .str();
197 
198   // First check if we can get desired values from debugging information.
199   std::string LineInfo = getFile<ELFT>()->getLineInfo(this, Offset);
200   if (!LineInfo.empty())
201     return LineInfo;
202 
203   // File->SourceFile contains STT_FILE symbol that contains a
204   // source file name. If it's missing, we use an object file name.
205   std::string SrcFile = getFile<ELFT>()->SourceFile;
206   if (SrcFile.empty())
207     SrcFile = toString(File);
208 
209   // Find a function symbol that encloses a given location.
210   for (SymbolBody *B : getFile<ELFT>()->getSymbols())
211     if (auto *D = dyn_cast<DefinedRegular>(B))
212       if (D->Section == this && D->Type == STT_FUNC)
213         if (D->Value <= Offset && Offset < D->Value + D->Size)
214           return SrcFile + ":(function " + toString(*D) + ")";
215 
216   // If there's no symbol, print out the offset in the section.
217   return (SrcFile + ":(" + Name + "+0x" + utohexstr(Offset) + ")").str();
218 }
219 
220 // Returns a source location string. This function is intended to be
221 // used for constructing an error message. The returned message looks
222 // like this:
223 //
224 //   foo.c:42 (/home/alice/possibly/very/long/path/foo.c:42)
225 //
226 // Returns an empty string if there's no way to get line info.
227 template <class ELFT> std::string InputSectionBase::getSrcMsg(uint64_t Offset) {
228   // Synthetic sections don't have input files.
229   elf::ObjectFile<ELFT> *File = getFile<ELFT>();
230   if (!File)
231     return "";
232 
233   Optional<DILineInfo> Info = File->getDILineInfo(this, Offset);
234 
235   // File->SourceFile contains STT_FILE symbol, and that is a last resort.
236   if (!Info)
237     return File->SourceFile;
238 
239   std::string Path = Info->FileName;
240   std::string Filename = path::filename(Path);
241   std::string Lineno = ":" + std::to_string(Info->Line);
242   if (Filename == Path)
243     return Filename + Lineno;
244   return Filename + Lineno + " (" + Path + Lineno + ")";
245 }
246 
247 // Returns a filename string along with an optional section name. This
248 // function is intended to be used for constructing an error
249 // message. The returned message looks like this:
250 //
251 //   path/to/foo.o:(function bar)
252 //
253 // or
254 //
255 //   path/to/foo.o:(function bar) in archive path/to/bar.a
256 template <class ELFT> std::string InputSectionBase::getObjMsg(uint64_t Off) {
257   // Synthetic sections don't have input files.
258   elf::ObjectFile<ELFT> *File = getFile<ELFT>();
259   std::string Filename = File ? File->getName() : "(internal)";
260 
261   std::string Archive;
262   if (!File->ArchiveName.empty())
263     Archive = (" in archive " + File->ArchiveName).str();
264 
265   // Find a symbol that encloses a given location.
266   for (SymbolBody *B : getFile<ELFT>()->getSymbols())
267     if (auto *D = dyn_cast<DefinedRegular>(B))
268       if (D->Section == this && D->Value <= Off && Off < D->Value + D->Size)
269         return Filename + ":(" + toString(*D) + ")" + Archive;
270 
271   // If there's no symbol, print out the offset in the section.
272   return (Filename + ":(" + Name + "+0x" + utohexstr(Off) + ")" + Archive)
273       .str();
274 }
275 
276 InputSectionBase InputSectionBase::Discarded;
277 
278 InputSection::InputSection(uint64_t Flags, uint32_t Type, uint32_t Alignment,
279                            ArrayRef<uint8_t> Data, StringRef Name, Kind K)
280     : InputSectionBase(nullptr, Flags, Type,
281                        /*Entsize*/ 0, /*Link*/ 0, /*Info*/ 0, Alignment, Data,
282                        Name, K) {}
283 
284 template <class ELFT>
285 InputSection::InputSection(elf::ObjectFile<ELFT> *F,
286                            const typename ELFT::Shdr *Header, StringRef Name)
287     : InputSectionBase(F, Header, Name, InputSectionBase::Regular) {}
288 
289 bool InputSection::classof(const SectionBase *S) {
290   return S->kind() == SectionBase::Regular ||
291          S->kind() == SectionBase::Synthetic;
292 }
293 
294 bool InputSectionBase::classof(const SectionBase *S) {
295   return S->kind() != Output;
296 }
297 
298 InputSectionBase *InputSection::getRelocatedSection() {
299   assert(this->Type == SHT_RELA || this->Type == SHT_REL);
300   ArrayRef<InputSectionBase *> Sections = this->File->getSections();
301   return Sections[this->Info];
302 }
303 
304 // This is used for -r and --emit-relocs. We can't use memcpy to copy
305 // relocations because we need to update symbol table offset and section index
306 // for each relocation. So we copy relocations one by one.
307 template <class ELFT, class RelTy>
308 void InputSection::copyRelocations(uint8_t *Buf, ArrayRef<RelTy> Rels) {
309   InputSectionBase *RelocatedSection = getRelocatedSection();
310 
311   // Loop is slow and have complexity O(N*M), where N - amount of
312   // relocations and M - amount of symbols in symbol table.
313   // That happens because getSymbolIndex(...) call below performs
314   // simple linear search.
315   for (const RelTy &Rel : Rels) {
316     uint32_t Type = Rel.getType(Config->IsMips64EL);
317     SymbolBody &Body = this->getFile<ELFT>()->getRelocTargetSym(Rel);
318 
319     auto *P = reinterpret_cast<typename ELFT::Rela *>(Buf);
320     Buf += sizeof(RelTy);
321 
322     if (Config->IsRela)
323       P->r_addend = getAddend<ELFT>(Rel);
324 
325     // Output section VA is zero for -r, so r_offset is an offset within the
326     // section, but for --emit-relocs it is an virtual address.
327     P->r_offset = RelocatedSection->OutSec->Addr +
328                   RelocatedSection->getOffset(Rel.r_offset);
329     P->setSymbolAndType(In<ELFT>::SymTab->getSymbolIndex(&Body), Type,
330                         Config->IsMips64EL);
331 
332     if (Body.Type == STT_SECTION) {
333       // We combine multiple section symbols into only one per
334       // section. This means we have to update the addend. That is
335       // trivial for Elf_Rela, but for Elf_Rel we have to write to the
336       // section data. We do that by adding to the Relocation vector.
337 
338       // .eh_frame is horribly special and can reference discarded sections. To
339       // avoid having to parse and recreate .eh_frame, we just replace any
340       // relocation in it pointing to discarded sections with R_*_NONE, which
341       // hopefully creates a frame that is ignored at runtime.
342       SectionBase *Section = cast<DefinedRegular>(Body).Section;
343       if (Section == &InputSection::Discarded) {
344         P->setSymbolAndType(0, 0, false);
345         continue;
346       }
347 
348       if (Config->IsRela) {
349         P->r_addend += Body.getVA() - Section->getOutputSection()->Addr;
350       } else if (Config->Relocatable) {
351         const uint8_t *BufLoc = RelocatedSection->Data.begin() + Rel.r_offset;
352         RelocatedSection->Relocations.push_back(
353             {R_ABS, Type, Rel.r_offset, Target->getImplicitAddend(BufLoc, Type),
354              &Body});
355       }
356     }
357 
358   }
359 }
360 
361 static uint32_t getARMUndefinedRelativeWeakVA(uint32_t Type, uint32_t A,
362                                               uint32_t P) {
363   switch (Type) {
364   case R_ARM_THM_JUMP11:
365     return P + 2;
366   case R_ARM_CALL:
367   case R_ARM_JUMP24:
368   case R_ARM_PC24:
369   case R_ARM_PLT32:
370   case R_ARM_PREL31:
371   case R_ARM_THM_JUMP19:
372   case R_ARM_THM_JUMP24:
373     return P + 4;
374   case R_ARM_THM_CALL:
375     // We don't want an interworking BLX to ARM
376     return P + 5;
377   default:
378     return A;
379   }
380 }
381 
382 static uint64_t getAArch64UndefinedRelativeWeakVA(uint64_t Type, uint64_t A,
383                                                   uint64_t P) {
384   switch (Type) {
385   case R_AARCH64_CALL26:
386   case R_AARCH64_CONDBR19:
387   case R_AARCH64_JUMP26:
388   case R_AARCH64_TSTBR14:
389     return P + 4;
390   default:
391     return A;
392   }
393 }
394 
395 template <class ELFT>
396 static typename ELFT::uint
397 getRelocTargetVA(uint32_t Type, int64_t A, typename ELFT::uint P,
398                  const SymbolBody &Body, RelExpr Expr) {
399   switch (Expr) {
400   case R_ABS:
401   case R_RELAX_GOT_PC_NOPIC:
402     return Body.getVA(A);
403   case R_GOT:
404   case R_RELAX_TLS_GD_TO_IE_ABS:
405     return Body.getGotVA<ELFT>() + A;
406   case R_GOTONLY_PC:
407     return In<ELFT>::Got->getVA() + A - P;
408   case R_GOTONLY_PC_FROM_END:
409     return In<ELFT>::Got->getVA() + A - P + In<ELFT>::Got->getSize();
410   case R_GOTREL:
411     return Body.getVA(A) - In<ELFT>::Got->getVA();
412   case R_GOTREL_FROM_END:
413     return Body.getVA(A) - In<ELFT>::Got->getVA() - In<ELFT>::Got->getSize();
414   case R_GOT_FROM_END:
415   case R_RELAX_TLS_GD_TO_IE_END:
416     return Body.getGotOffset() + A - In<ELFT>::Got->getSize();
417   case R_GOT_OFF:
418     return Body.getGotOffset() + A;
419   case R_GOT_PAGE_PC:
420   case R_RELAX_TLS_GD_TO_IE_PAGE_PC:
421     return getAArch64Page(Body.getGotVA<ELFT>() + A) - getAArch64Page(P);
422   case R_GOT_PC:
423   case R_RELAX_TLS_GD_TO_IE:
424     return Body.getGotVA<ELFT>() + A - P;
425   case R_HINT:
426   case R_NONE:
427   case R_TLSDESC_CALL:
428     llvm_unreachable("cannot relocate hint relocs");
429   case R_MIPS_GOTREL:
430     return Body.getVA(A) - In<ELFT>::MipsGot->getGp();
431   case R_MIPS_GOT_GP:
432     return In<ELFT>::MipsGot->getGp() + A;
433   case R_MIPS_GOT_GP_PC: {
434     // R_MIPS_LO16 expression has R_MIPS_GOT_GP_PC type iif the target
435     // is _gp_disp symbol. In that case we should use the following
436     // formula for calculation "AHL + GP - P + 4". For details see p. 4-19 at
437     // ftp://www.linux-mips.org/pub/linux/mips/doc/ABI/mipsabi.pdf
438     uint64_t V = In<ELFT>::MipsGot->getGp() + A - P;
439     if (Type == R_MIPS_LO16)
440       V += 4;
441     return V;
442   }
443   case R_MIPS_GOT_LOCAL_PAGE:
444     // If relocation against MIPS local symbol requires GOT entry, this entry
445     // should be initialized by 'page address'. This address is high 16-bits
446     // of sum the symbol's value and the addend.
447     return In<ELFT>::MipsGot->getVA() +
448            In<ELFT>::MipsGot->getPageEntryOffset(Body, A) -
449            In<ELFT>::MipsGot->getGp();
450   case R_MIPS_GOT_OFF:
451   case R_MIPS_GOT_OFF32:
452     // In case of MIPS if a GOT relocation has non-zero addend this addend
453     // should be applied to the GOT entry content not to the GOT entry offset.
454     // That is why we use separate expression type.
455     return In<ELFT>::MipsGot->getVA() +
456            In<ELFT>::MipsGot->getBodyEntryOffset(Body, A) -
457            In<ELFT>::MipsGot->getGp();
458   case R_MIPS_TLSGD:
459     return In<ELFT>::MipsGot->getVA() + In<ELFT>::MipsGot->getTlsOffset() +
460            In<ELFT>::MipsGot->getGlobalDynOffset(Body) -
461            In<ELFT>::MipsGot->getGp();
462   case R_MIPS_TLSLD:
463     return In<ELFT>::MipsGot->getVA() + In<ELFT>::MipsGot->getTlsOffset() +
464            In<ELFT>::MipsGot->getTlsIndexOff() - In<ELFT>::MipsGot->getGp();
465   case R_PAGE_PC:
466   case R_PLT_PAGE_PC:
467     if (Body.isUndefined() && !Body.isLocal() && Body.symbol()->isWeak())
468       return getAArch64Page(A);
469     return getAArch64Page(Body.getVA(A)) - getAArch64Page(P);
470   case R_PC:
471     if (Body.isUndefined() && !Body.isLocal() && Body.symbol()->isWeak()) {
472       // On ARM and AArch64 a branch to an undefined weak resolves to the
473       // next instruction, otherwise the place.
474       if (Config->EMachine == EM_ARM)
475         return getARMUndefinedRelativeWeakVA(Type, A, P);
476       if (Config->EMachine == EM_AARCH64)
477         return getAArch64UndefinedRelativeWeakVA(Type, A, P);
478     }
479     return Body.getVA(A) - P;
480   case R_PLT:
481     return Body.getPltVA() + A;
482   case R_PLT_PC:
483   case R_PPC_PLT_OPD:
484     return Body.getPltVA() + A - P;
485   case R_PPC_OPD: {
486     uint64_t SymVA = Body.getVA(A);
487     // If we have an undefined weak symbol, we might get here with a symbol
488     // address of zero. That could overflow, but the code must be unreachable,
489     // so don't bother doing anything at all.
490     if (!SymVA)
491       return 0;
492     if (Out::Opd) {
493       // If this is a local call, and we currently have the address of a
494       // function-descriptor, get the underlying code address instead.
495       uint64_t OpdStart = Out::Opd->Addr;
496       uint64_t OpdEnd = OpdStart + Out::Opd->Size;
497       bool InOpd = OpdStart <= SymVA && SymVA < OpdEnd;
498       if (InOpd)
499         SymVA = read64be(&Out::OpdBuf[SymVA - OpdStart]);
500     }
501     return SymVA - P;
502   }
503   case R_PPC_TOC:
504     return getPPC64TocBase() + A;
505   case R_RELAX_GOT_PC:
506     return Body.getVA(A) - P;
507   case R_RELAX_TLS_GD_TO_LE:
508   case R_RELAX_TLS_IE_TO_LE:
509   case R_RELAX_TLS_LD_TO_LE:
510   case R_TLS:
511     // A weak undefined TLS symbol resolves to the base of the TLS
512     // block, i.e. gets a value of zero. If we pass --gc-sections to
513     // lld and .tbss is not referenced, it gets reclaimed and we don't
514     // create a TLS program header. Therefore, we resolve this
515     // statically to zero.
516     if (Body.isTls() && (Body.isLazy() || Body.isUndefined()) &&
517         Body.symbol()->isWeak())
518       return 0;
519     if (Target->TcbSize)
520       return Body.getVA(A) + alignTo(Target->TcbSize, Out::TlsPhdr->p_align);
521     return Body.getVA(A) - Out::TlsPhdr->p_memsz;
522   case R_RELAX_TLS_GD_TO_LE_NEG:
523   case R_NEG_TLS:
524     return Out::TlsPhdr->p_memsz - Body.getVA(A);
525   case R_SIZE:
526     return Body.getSize<ELFT>() + A;
527   case R_TLSDESC:
528     return In<ELFT>::Got->getGlobalDynAddr(Body) + A;
529   case R_TLSDESC_PAGE:
530     return getAArch64Page(In<ELFT>::Got->getGlobalDynAddr(Body) + A) -
531            getAArch64Page(P);
532   case R_TLSGD:
533     return In<ELFT>::Got->getGlobalDynOffset(Body) + A -
534            In<ELFT>::Got->getSize();
535   case R_TLSGD_PC:
536     return In<ELFT>::Got->getGlobalDynAddr(Body) + A - P;
537   case R_TLSLD:
538     return In<ELFT>::Got->getTlsIndexOff() + A - In<ELFT>::Got->getSize();
539   case R_TLSLD_PC:
540     return In<ELFT>::Got->getTlsIndexVA() + A - P;
541   }
542   llvm_unreachable("Invalid expression");
543 }
544 
545 // This function applies relocations to sections without SHF_ALLOC bit.
546 // Such sections are never mapped to memory at runtime. Debug sections are
547 // an example. Relocations in non-alloc sections are much easier to
548 // handle than in allocated sections because it will never need complex
549 // treatement such as GOT or PLT (because at runtime no one refers them).
550 // So, we handle relocations for non-alloc sections directly in this
551 // function as a performance optimization.
552 template <class ELFT, class RelTy>
553 void InputSection::relocateNonAlloc(uint8_t *Buf, ArrayRef<RelTy> Rels) {
554   typedef typename ELFT::uint uintX_t;
555   for (const RelTy &Rel : Rels) {
556     uint32_t Type = Rel.getType(Config->IsMips64EL);
557     uint64_t Offset = getOffset(Rel.r_offset);
558     uint8_t *BufLoc = Buf + Offset;
559     int64_t Addend = getAddend<ELFT>(Rel);
560     if (!RelTy::IsRela)
561       Addend += Target->getImplicitAddend(BufLoc, Type);
562 
563     SymbolBody &Sym = this->getFile<ELFT>()->getRelocTargetSym(Rel);
564     RelExpr Expr = Target->getRelExpr(Type, Sym, BufLoc);
565     if (Expr == R_NONE)
566       continue;
567     if (Expr != R_ABS) {
568       error(this->getLocation<ELFT>(Offset) + ": has non-ABS reloc");
569       return;
570     }
571 
572     uintX_t AddrLoc = this->OutSec->Addr + Offset;
573     uint64_t SymVA = 0;
574     if (!Sym.isTls() || Out::TlsPhdr)
575       SymVA = SignExtend64<sizeof(uintX_t) * 8>(
576           getRelocTargetVA<ELFT>(Type, Addend, AddrLoc, Sym, R_ABS));
577     Target->relocateOne(BufLoc, Type, SymVA);
578   }
579 }
580 
581 template <class ELFT> elf::ObjectFile<ELFT> *InputSectionBase::getFile() const {
582   return cast_or_null<elf::ObjectFile<ELFT>>(File);
583 }
584 
585 template <class ELFT>
586 void InputSectionBase::relocate(uint8_t *Buf, uint8_t *BufEnd) {
587   // scanReloc function in Writer.cpp constructs Relocations
588   // vector only for SHF_ALLOC'ed sections. For other sections,
589   // we handle relocations directly here.
590   auto *IS = dyn_cast<InputSection>(this);
591   if (IS && !(IS->Flags & SHF_ALLOC)) {
592     if (IS->AreRelocsRela)
593       IS->relocateNonAlloc<ELFT>(Buf, IS->template relas<ELFT>());
594     else
595       IS->relocateNonAlloc<ELFT>(Buf, IS->template rels<ELFT>());
596     return;
597   }
598 
599   typedef typename ELFT::uint uintX_t;
600   const unsigned Bits = sizeof(uintX_t) * 8;
601   for (const Relocation &Rel : Relocations) {
602     uint64_t Offset = getOffset(Rel.Offset);
603     uint8_t *BufLoc = Buf + Offset;
604     uint32_t Type = Rel.Type;
605 
606     uintX_t AddrLoc = getOutputSection()->Addr + Offset;
607     RelExpr Expr = Rel.Expr;
608     uint64_t TargetVA = SignExtend64<Bits>(
609         getRelocTargetVA<ELFT>(Type, Rel.Addend, AddrLoc, *Rel.Sym, Expr));
610 
611     switch (Expr) {
612     case R_RELAX_GOT_PC:
613     case R_RELAX_GOT_PC_NOPIC:
614       Target->relaxGot(BufLoc, TargetVA);
615       break;
616     case R_RELAX_TLS_IE_TO_LE:
617       Target->relaxTlsIeToLe(BufLoc, Type, TargetVA);
618       break;
619     case R_RELAX_TLS_LD_TO_LE:
620       Target->relaxTlsLdToLe(BufLoc, Type, TargetVA);
621       break;
622     case R_RELAX_TLS_GD_TO_LE:
623     case R_RELAX_TLS_GD_TO_LE_NEG:
624       Target->relaxTlsGdToLe(BufLoc, Type, TargetVA);
625       break;
626     case R_RELAX_TLS_GD_TO_IE:
627     case R_RELAX_TLS_GD_TO_IE_ABS:
628     case R_RELAX_TLS_GD_TO_IE_PAGE_PC:
629     case R_RELAX_TLS_GD_TO_IE_END:
630       Target->relaxTlsGdToIe(BufLoc, Type, TargetVA);
631       break;
632     case R_PPC_PLT_OPD:
633       // Patch a nop (0x60000000) to a ld.
634       if (BufLoc + 8 <= BufEnd && read32be(BufLoc + 4) == 0x60000000)
635         write32be(BufLoc + 4, 0xe8410028); // ld %r2, 40(%r1)
636     // fallthrough
637     default:
638       Target->relocateOne(BufLoc, Type, TargetVA);
639       break;
640     }
641   }
642 }
643 
644 template <class ELFT> void InputSection::writeTo(uint8_t *Buf) {
645   if (this->Type == SHT_NOBITS)
646     return;
647 
648   if (auto *S = dyn_cast<SyntheticSection>(this)) {
649     S->writeTo(Buf + OutSecOff);
650     return;
651   }
652 
653   // If -r or --emit-relocs is given, then an InputSection
654   // may be a relocation section.
655   if (this->Type == SHT_RELA) {
656     copyRelocations<ELFT>(Buf + OutSecOff,
657                           this->template getDataAs<typename ELFT::Rela>());
658     return;
659   }
660   if (this->Type == SHT_REL) {
661     copyRelocations<ELFT>(Buf + OutSecOff,
662                           this->template getDataAs<typename ELFT::Rel>());
663     return;
664   }
665 
666   // Copy section contents from source object file to output file.
667   ArrayRef<uint8_t> Data = this->Data;
668   memcpy(Buf + OutSecOff, Data.data(), Data.size());
669 
670   // Iterate over all relocation sections that apply to this section.
671   uint8_t *BufEnd = Buf + OutSecOff + Data.size();
672   this->relocate<ELFT>(Buf, BufEnd);
673 }
674 
675 void InputSection::replace(InputSection *Other) {
676   this->Alignment = std::max(this->Alignment, Other->Alignment);
677   Other->Repl = this->Repl;
678   Other->Live = false;
679 }
680 
681 template <class ELFT>
682 EhInputSection::EhInputSection(elf::ObjectFile<ELFT> *F,
683                                const typename ELFT::Shdr *Header,
684                                StringRef Name)
685     : InputSectionBase(F, Header, Name, InputSectionBase::EHFrame) {
686   // Mark .eh_frame sections as live by default because there are
687   // usually no relocations that point to .eh_frames. Otherwise,
688   // the garbage collector would drop all .eh_frame sections.
689   this->Live = true;
690 }
691 
692 bool EhInputSection::classof(const SectionBase *S) {
693   return S->kind() == InputSectionBase::EHFrame;
694 }
695 
696 // Returns the index of the first relocation that points to a region between
697 // Begin and Begin+Size.
698 template <class IntTy, class RelTy>
699 static unsigned getReloc(IntTy Begin, IntTy Size, const ArrayRef<RelTy> &Rels,
700                          unsigned &RelocI) {
701   // Start search from RelocI for fast access. That works because the
702   // relocations are sorted in .eh_frame.
703   for (unsigned N = Rels.size(); RelocI < N; ++RelocI) {
704     const RelTy &Rel = Rels[RelocI];
705     if (Rel.r_offset < Begin)
706       continue;
707 
708     if (Rel.r_offset < Begin + Size)
709       return RelocI;
710     return -1;
711   }
712   return -1;
713 }
714 
715 // .eh_frame is a sequence of CIE or FDE records.
716 // This function splits an input section into records and returns them.
717 template <class ELFT> void EhInputSection::split() {
718   // Early exit if already split.
719   if (!this->Pieces.empty())
720     return;
721 
722   if (this->NumRelocations) {
723     if (this->AreRelocsRela)
724       split<ELFT>(this->relas<ELFT>());
725     else
726       split<ELFT>(this->rels<ELFT>());
727     return;
728   }
729   split<ELFT>(makeArrayRef<typename ELFT::Rela>(nullptr, nullptr));
730 }
731 
732 template <class ELFT, class RelTy>
733 void EhInputSection::split(ArrayRef<RelTy> Rels) {
734   ArrayRef<uint8_t> Data = this->Data;
735   unsigned RelI = 0;
736   for (size_t Off = 0, End = Data.size(); Off != End;) {
737     size_t Size = readEhRecordSize<ELFT>(this, Off);
738     this->Pieces.emplace_back(Off, this, Size, getReloc(Off, Size, Rels, RelI));
739     // The empty record is the end marker.
740     if (Size == 4)
741       break;
742     Off += Size;
743   }
744 }
745 
746 static size_t findNull(ArrayRef<uint8_t> A, size_t EntSize) {
747   // Optimize the common case.
748   StringRef S((const char *)A.data(), A.size());
749   if (EntSize == 1)
750     return S.find(0);
751 
752   for (unsigned I = 0, N = S.size(); I != N; I += EntSize) {
753     const char *B = S.begin() + I;
754     if (std::all_of(B, B + EntSize, [](char C) { return C == 0; }))
755       return I;
756   }
757   return StringRef::npos;
758 }
759 
760 // Split SHF_STRINGS section. Such section is a sequence of
761 // null-terminated strings.
762 void MergeInputSection::splitStrings(ArrayRef<uint8_t> Data, size_t EntSize) {
763   size_t Off = 0;
764   bool IsAlloc = this->Flags & SHF_ALLOC;
765   while (!Data.empty()) {
766     size_t End = findNull(Data, EntSize);
767     if (End == StringRef::npos)
768       fatal(toString(this) + ": string is not null terminated");
769     size_t Size = End + EntSize;
770     Pieces.emplace_back(Off, !IsAlloc);
771     Hashes.push_back(hash_value(toStringRef(Data.slice(0, Size))));
772     Data = Data.slice(Size);
773     Off += Size;
774   }
775 }
776 
777 // Split non-SHF_STRINGS section. Such section is a sequence of
778 // fixed size records.
779 void MergeInputSection::splitNonStrings(ArrayRef<uint8_t> Data,
780                                         size_t EntSize) {
781   size_t Size = Data.size();
782   assert((Size % EntSize) == 0);
783   bool IsAlloc = this->Flags & SHF_ALLOC;
784   for (unsigned I = 0, N = Size; I != N; I += EntSize) {
785     Hashes.push_back(hash_value(toStringRef(Data.slice(I, EntSize))));
786     Pieces.emplace_back(I, !IsAlloc);
787   }
788 }
789 
790 template <class ELFT>
791 MergeInputSection::MergeInputSection(elf::ObjectFile<ELFT> *F,
792                                      const typename ELFT::Shdr *Header,
793                                      StringRef Name)
794     : InputSectionBase(F, Header, Name, InputSectionBase::Merge) {}
795 
796 // This function is called after we obtain a complete list of input sections
797 // that need to be linked. This is responsible to split section contents
798 // into small chunks for further processing.
799 //
800 // Note that this function is called from parallel_for_each. This must be
801 // thread-safe (i.e. no memory allocation from the pools).
802 void MergeInputSection::splitIntoPieces() {
803   ArrayRef<uint8_t> Data = this->Data;
804   uint64_t EntSize = this->Entsize;
805   if (this->Flags & SHF_STRINGS)
806     splitStrings(Data, EntSize);
807   else
808     splitNonStrings(Data, EntSize);
809 
810   if (Config->GcSections && (this->Flags & SHF_ALLOC))
811     for (uint64_t Off : LiveOffsets)
812       this->getSectionPiece(Off)->Live = true;
813 }
814 
815 bool MergeInputSection::classof(const SectionBase *S) {
816   return S->kind() == InputSectionBase::Merge;
817 }
818 
819 // Do binary search to get a section piece at a given input offset.
820 SectionPiece *MergeInputSection::getSectionPiece(uint64_t Offset) {
821   auto *This = static_cast<const MergeInputSection *>(this);
822   return const_cast<SectionPiece *>(This->getSectionPiece(Offset));
823 }
824 
825 template <class It, class T, class Compare>
826 static It fastUpperBound(It First, It Last, const T &Value, Compare Comp) {
827   size_t Size = std::distance(First, Last);
828   assert(Size != 0);
829   while (Size != 1) {
830     size_t H = Size / 2;
831     const It MI = First + H;
832     Size -= H;
833     First = Comp(Value, *MI) ? First : First + H;
834   }
835   return Comp(Value, *First) ? First : First + 1;
836 }
837 
838 const SectionPiece *MergeInputSection::getSectionPiece(uint64_t Offset) const {
839   uint64_t Size = this->Data.size();
840   if (Offset >= Size)
841     fatal(toString(this) + ": entry is past the end of the section");
842 
843   // Find the element this offset points to.
844   auto I = fastUpperBound(
845       Pieces.begin(), Pieces.end(), Offset,
846       [](const uint64_t &A, const SectionPiece &B) { return A < B.InputOff; });
847   --I;
848   return &*I;
849 }
850 
851 // Returns the offset in an output section for a given input offset.
852 // Because contents of a mergeable section is not contiguous in output,
853 // it is not just an addition to a base output offset.
854 uint64_t MergeInputSection::getOffset(uint64_t Offset) const {
855   // Initialize OffsetMap lazily.
856   std::call_once(InitOffsetMap, [&] {
857     OffsetMap.reserve(Pieces.size());
858     for (const SectionPiece &Piece : Pieces)
859       OffsetMap[Piece.InputOff] = Piece.OutputOff;
860   });
861 
862   // Find a string starting at a given offset.
863   auto It = OffsetMap.find(Offset);
864   if (It != OffsetMap.end())
865     return It->second;
866 
867   if (!this->Live)
868     return 0;
869 
870   // If Offset is not at beginning of a section piece, it is not in the map.
871   // In that case we need to search from the original section piece vector.
872   const SectionPiece &Piece = *this->getSectionPiece(Offset);
873   if (!Piece.Live)
874     return 0;
875 
876   uint64_t Addend = Offset - Piece.InputOff;
877   return Piece.OutputOff + Addend;
878 }
879 
880 template InputSection::InputSection(elf::ObjectFile<ELF32LE> *,
881                                     const ELF32LE::Shdr *, StringRef);
882 template InputSection::InputSection(elf::ObjectFile<ELF32BE> *,
883                                     const ELF32BE::Shdr *, StringRef);
884 template InputSection::InputSection(elf::ObjectFile<ELF64LE> *,
885                                     const ELF64LE::Shdr *, StringRef);
886 template InputSection::InputSection(elf::ObjectFile<ELF64BE> *,
887                                     const ELF64BE::Shdr *, StringRef);
888 
889 template std::string InputSectionBase::getLocation<ELF32LE>(uint64_t);
890 template std::string InputSectionBase::getLocation<ELF32BE>(uint64_t);
891 template std::string InputSectionBase::getLocation<ELF64LE>(uint64_t);
892 template std::string InputSectionBase::getLocation<ELF64BE>(uint64_t);
893 
894 template std::string InputSectionBase::getSrcMsg<ELF32LE>(uint64_t);
895 template std::string InputSectionBase::getSrcMsg<ELF32BE>(uint64_t);
896 template std::string InputSectionBase::getSrcMsg<ELF64LE>(uint64_t);
897 template std::string InputSectionBase::getSrcMsg<ELF64BE>(uint64_t);
898 
899 template std::string InputSectionBase::getObjMsg<ELF32LE>(uint64_t);
900 template std::string InputSectionBase::getObjMsg<ELF32BE>(uint64_t);
901 template std::string InputSectionBase::getObjMsg<ELF64LE>(uint64_t);
902 template std::string InputSectionBase::getObjMsg<ELF64BE>(uint64_t);
903 
904 template void InputSection::writeTo<ELF32LE>(uint8_t *);
905 template void InputSection::writeTo<ELF32BE>(uint8_t *);
906 template void InputSection::writeTo<ELF64LE>(uint8_t *);
907 template void InputSection::writeTo<ELF64BE>(uint8_t *);
908 
909 template elf::ObjectFile<ELF32LE> *InputSectionBase::getFile<ELF32LE>() const;
910 template elf::ObjectFile<ELF32BE> *InputSectionBase::getFile<ELF32BE>() const;
911 template elf::ObjectFile<ELF64LE> *InputSectionBase::getFile<ELF64LE>() const;
912 template elf::ObjectFile<ELF64BE> *InputSectionBase::getFile<ELF64BE>() const;
913 
914 template MergeInputSection::MergeInputSection(elf::ObjectFile<ELF32LE> *,
915                                               const ELF32LE::Shdr *, StringRef);
916 template MergeInputSection::MergeInputSection(elf::ObjectFile<ELF32BE> *,
917                                               const ELF32BE::Shdr *, StringRef);
918 template MergeInputSection::MergeInputSection(elf::ObjectFile<ELF64LE> *,
919                                               const ELF64LE::Shdr *, StringRef);
920 template MergeInputSection::MergeInputSection(elf::ObjectFile<ELF64BE> *,
921                                               const ELF64BE::Shdr *, StringRef);
922 
923 template EhInputSection::EhInputSection(elf::ObjectFile<ELF32LE> *,
924                                         const ELF32LE::Shdr *, StringRef);
925 template EhInputSection::EhInputSection(elf::ObjectFile<ELF32BE> *,
926                                         const ELF32BE::Shdr *, StringRef);
927 template EhInputSection::EhInputSection(elf::ObjectFile<ELF64LE> *,
928                                         const ELF64LE::Shdr *, StringRef);
929 template EhInputSection::EhInputSection(elf::ObjectFile<ELF64BE> *,
930                                         const ELF64BE::Shdr *, StringRef);
931 
932 template void EhInputSection::split<ELF32LE>();
933 template void EhInputSection::split<ELF32BE>();
934 template void EhInputSection::split<ELF64LE>();
935 template void EhInputSection::split<ELF64BE>();
936