1 //===- InputSection.cpp ---------------------------------------------------===//
2 //
3 // The LLVM Linker
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9
10 #include "InputSection.h"
11 #include "Config.h"
12 #include "EhFrame.h"
13 #include "InputFiles.h"
14 #include "LinkerScript.h"
15 #include "OutputSections.h"
16 #include "Relocations.h"
17 #include "SymbolTable.h"
18 #include "Symbols.h"
19 #include "SyntheticSections.h"
20 #include "Target.h"
21 #include "Thunks.h"
22 #include "lld/Common/ErrorHandler.h"
23 #include "lld/Common/Memory.h"
24 #include "llvm/Support/Compiler.h"
25 #include "llvm/Support/Compression.h"
26 #include "llvm/Support/Endian.h"
27 #include "llvm/Support/Threading.h"
28 #include "llvm/Support/xxhash.h"
29 #include <algorithm>
30 #include <mutex>
31 #include <set>
32 #include <vector>
33
34 using namespace llvm;
35 using namespace llvm::ELF;
36 using namespace llvm::object;
37 using namespace llvm::support;
38 using namespace llvm::support::endian;
39 using namespace llvm::sys;
40
41 using namespace lld;
42 using namespace lld::elf;
43
44 std::vector<InputSectionBase *> elf::InputSections;
45
46 // Returns a string to construct an error message.
toString(const InputSectionBase * Sec)47 std::string lld::toString(const InputSectionBase *Sec) {
48 return (toString(Sec->File) + ":(" + Sec->Name + ")").str();
49 }
50
51 template <class ELFT>
getSectionContents(ObjFile<ELFT> & File,const typename ELFT::Shdr & Hdr)52 static ArrayRef<uint8_t> getSectionContents(ObjFile<ELFT> &File,
53 const typename ELFT::Shdr &Hdr) {
54 if (Hdr.sh_type == SHT_NOBITS)
55 return makeArrayRef<uint8_t>(nullptr, Hdr.sh_size);
56 return check(File.getObj().getSectionContents(&Hdr));
57 }
58
InputSectionBase(InputFile * File,uint64_t Flags,uint32_t Type,uint64_t Entsize,uint32_t Link,uint32_t Info,uint32_t Alignment,ArrayRef<uint8_t> Data,StringRef Name,Kind SectionKind)59 InputSectionBase::InputSectionBase(InputFile *File, uint64_t Flags,
60 uint32_t Type, uint64_t Entsize,
61 uint32_t Link, uint32_t Info,
62 uint32_t Alignment, ArrayRef<uint8_t> Data,
63 StringRef Name, Kind SectionKind)
64 : SectionBase(SectionKind, Name, Flags, Entsize, Alignment, Type, Info,
65 Link),
66 File(File), RawData(Data) {
67 // In order to reduce memory allocation, we assume that mergeable
68 // sections are smaller than 4 GiB, which is not an unreasonable
69 // assumption as of 2017.
70 if (SectionKind == SectionBase::Merge && RawData.size() > UINT32_MAX)
71 error(toString(this) + ": section too large");
72
73 NumRelocations = 0;
74 AreRelocsRela = false;
75
76 // The ELF spec states that a value of 0 means the section has
77 // no alignment constraits.
78 uint32_t V = std::max<uint64_t>(Alignment, 1);
79 if (!isPowerOf2_64(V))
80 fatal(toString(File) + ": section sh_addralign is not a power of 2");
81 this->Alignment = V;
82
83 // In ELF, each section can be compressed by zlib, and if compressed,
84 // section name may be mangled by appending "z" (e.g. ".zdebug_info").
85 // If that's the case, demangle section name so that we can handle a
86 // section as if it weren't compressed.
87 if ((Flags & SHF_COMPRESSED) || Name.startswith(".zdebug")) {
88 if (!zlib::isAvailable())
89 error(toString(File) + ": contains a compressed section, " +
90 "but zlib is not available");
91 parseCompressedHeader();
92 }
93 }
94
95 // Drop SHF_GROUP bit unless we are producing a re-linkable object file.
96 // SHF_GROUP is a marker that a section belongs to some comdat group.
97 // That flag doesn't make sense in an executable.
getFlags(uint64_t Flags)98 static uint64_t getFlags(uint64_t Flags) {
99 Flags &= ~(uint64_t)SHF_INFO_LINK;
100 if (!Config->Relocatable)
101 Flags &= ~(uint64_t)SHF_GROUP;
102 return Flags;
103 }
104
105 // GNU assembler 2.24 and LLVM 4.0.0's MC (the newest release as of
106 // March 2017) fail to infer section types for sections starting with
107 // ".init_array." or ".fini_array.". They set SHT_PROGBITS instead of
108 // SHF_INIT_ARRAY. As a result, the following assembler directive
109 // creates ".init_array.100" with SHT_PROGBITS, for example.
110 //
111 // .section .init_array.100, "aw"
112 //
113 // This function forces SHT_{INIT,FINI}_ARRAY so that we can handle
114 // incorrect inputs as if they were correct from the beginning.
getType(uint64_t Type,StringRef Name)115 static uint64_t getType(uint64_t Type, StringRef Name) {
116 if (Type == SHT_PROGBITS && Name.startswith(".init_array."))
117 return SHT_INIT_ARRAY;
118 if (Type == SHT_PROGBITS && Name.startswith(".fini_array."))
119 return SHT_FINI_ARRAY;
120 return Type;
121 }
122
123 template <class ELFT>
InputSectionBase(ObjFile<ELFT> & File,const typename ELFT::Shdr & Hdr,StringRef Name,Kind SectionKind)124 InputSectionBase::InputSectionBase(ObjFile<ELFT> &File,
125 const typename ELFT::Shdr &Hdr,
126 StringRef Name, Kind SectionKind)
127 : InputSectionBase(&File, getFlags(Hdr.sh_flags),
128 getType(Hdr.sh_type, Name), Hdr.sh_entsize, Hdr.sh_link,
129 Hdr.sh_info, Hdr.sh_addralign,
130 getSectionContents(File, Hdr), Name, SectionKind) {
131 // We reject object files having insanely large alignments even though
132 // they are allowed by the spec. I think 4GB is a reasonable limitation.
133 // We might want to relax this in the future.
134 if (Hdr.sh_addralign > UINT32_MAX)
135 fatal(toString(&File) + ": section sh_addralign is too large");
136 }
137
getSize() const138 size_t InputSectionBase::getSize() const {
139 if (auto *S = dyn_cast<SyntheticSection>(this))
140 return S->getSize();
141 if (UncompressedSize >= 0)
142 return UncompressedSize;
143 return RawData.size();
144 }
145
uncompress() const146 void InputSectionBase::uncompress() const {
147 size_t Size = UncompressedSize;
148 UncompressedBuf.reset(new char[Size]);
149
150 if (Error E =
151 zlib::uncompress(toStringRef(RawData), UncompressedBuf.get(), Size))
152 fatal(toString(this) +
153 ": uncompress failed: " + llvm::toString(std::move(E)));
154 RawData = makeArrayRef((uint8_t *)UncompressedBuf.get(), Size);
155 }
156
getOffsetInFile() const157 uint64_t InputSectionBase::getOffsetInFile() const {
158 const uint8_t *FileStart = (const uint8_t *)File->MB.getBufferStart();
159 const uint8_t *SecStart = data().begin();
160 return SecStart - FileStart;
161 }
162
getOffset(uint64_t Offset) const163 uint64_t SectionBase::getOffset(uint64_t Offset) const {
164 switch (kind()) {
165 case Output: {
166 auto *OS = cast<OutputSection>(this);
167 // For output sections we treat offset -1 as the end of the section.
168 return Offset == uint64_t(-1) ? OS->Size : Offset;
169 }
170 case Regular:
171 case Synthetic:
172 return cast<InputSection>(this)->getOffset(Offset);
173 case EHFrame:
174 // The file crtbeginT.o has relocations pointing to the start of an empty
175 // .eh_frame that is known to be the first in the link. It does that to
176 // identify the start of the output .eh_frame.
177 return Offset;
178 case Merge:
179 const MergeInputSection *MS = cast<MergeInputSection>(this);
180 if (InputSection *IS = MS->getParent())
181 return IS->getOffset(MS->getParentOffset(Offset));
182 return MS->getParentOffset(Offset);
183 }
184 llvm_unreachable("invalid section kind");
185 }
186
getVA(uint64_t Offset) const187 uint64_t SectionBase::getVA(uint64_t Offset) const {
188 const OutputSection *Out = getOutputSection();
189 return (Out ? Out->Addr : 0) + getOffset(Offset);
190 }
191
getOutputSection()192 OutputSection *SectionBase::getOutputSection() {
193 InputSection *Sec;
194 if (auto *IS = dyn_cast<InputSection>(this))
195 Sec = IS;
196 else if (auto *MS = dyn_cast<MergeInputSection>(this))
197 Sec = MS->getParent();
198 else if (auto *EH = dyn_cast<EhInputSection>(this))
199 Sec = EH->getParent();
200 else
201 return cast<OutputSection>(this);
202 return Sec ? Sec->getParent() : nullptr;
203 }
204
205 // When a section is compressed, `RawData` consists with a header followed
206 // by zlib-compressed data. This function parses a header to initialize
207 // `UncompressedSize` member and remove the header from `RawData`.
parseCompressedHeader()208 void InputSectionBase::parseCompressedHeader() {
209 typedef typename ELF64LE::Chdr Chdr64;
210 typedef typename ELF32LE::Chdr Chdr32;
211
212 // Old-style header
213 if (Name.startswith(".zdebug")) {
214 if (!toStringRef(RawData).startswith("ZLIB")) {
215 error(toString(this) + ": corrupted compressed section header");
216 return;
217 }
218 RawData = RawData.slice(4);
219
220 if (RawData.size() < 8) {
221 error(toString(this) + ": corrupted compressed section header");
222 return;
223 }
224
225 UncompressedSize = read64be(RawData.data());
226 RawData = RawData.slice(8);
227
228 // Restore the original section name.
229 // (e.g. ".zdebug_info" -> ".debug_info")
230 Name = Saver.save("." + Name.substr(2));
231 return;
232 }
233
234 assert(Flags & SHF_COMPRESSED);
235 Flags &= ~(uint64_t)SHF_COMPRESSED;
236
237 // New-style 64-bit header
238 if (Config->Is64) {
239 if (RawData.size() < sizeof(Chdr64)) {
240 error(toString(this) + ": corrupted compressed section");
241 return;
242 }
243
244 auto *Hdr = reinterpret_cast<const Chdr64 *>(RawData.data());
245 if (Hdr->ch_type != ELFCOMPRESS_ZLIB) {
246 error(toString(this) + ": unsupported compression type");
247 return;
248 }
249
250 UncompressedSize = Hdr->ch_size;
251 Alignment = std::max<uint64_t>(Hdr->ch_addralign, 1);
252 RawData = RawData.slice(sizeof(*Hdr));
253 return;
254 }
255
256 // New-style 32-bit header
257 if (RawData.size() < sizeof(Chdr32)) {
258 error(toString(this) + ": corrupted compressed section");
259 return;
260 }
261
262 auto *Hdr = reinterpret_cast<const Chdr32 *>(RawData.data());
263 if (Hdr->ch_type != ELFCOMPRESS_ZLIB) {
264 error(toString(this) + ": unsupported compression type");
265 return;
266 }
267
268 UncompressedSize = Hdr->ch_size;
269 Alignment = std::max<uint64_t>(Hdr->ch_addralign, 1);
270 RawData = RawData.slice(sizeof(*Hdr));
271 }
272
getLinkOrderDep() const273 InputSection *InputSectionBase::getLinkOrderDep() const {
274 assert(Link);
275 assert(Flags & SHF_LINK_ORDER);
276 return cast<InputSection>(File->getSections()[Link]);
277 }
278
279 // Find a function symbol that encloses a given location.
280 template <class ELFT>
getEnclosingFunction(uint64_t Offset)281 Defined *InputSectionBase::getEnclosingFunction(uint64_t Offset) {
282 for (Symbol *B : File->getSymbols())
283 if (Defined *D = dyn_cast<Defined>(B))
284 if (D->Section == this && D->Type == STT_FUNC && D->Value <= Offset &&
285 Offset < D->Value + D->Size)
286 return D;
287 return nullptr;
288 }
289
290 // Returns a source location string. Used to construct an error message.
291 template <class ELFT>
getLocation(uint64_t Offset)292 std::string InputSectionBase::getLocation(uint64_t Offset) {
293 std::string SecAndOffset = (Name + "+0x" + utohexstr(Offset)).str();
294
295 // We don't have file for synthetic sections.
296 if (getFile<ELFT>() == nullptr)
297 return (Config->OutputFile + ":(" + SecAndOffset + ")")
298 .str();
299
300 // First check if we can get desired values from debugging information.
301 if (Optional<DILineInfo> Info = getFile<ELFT>()->getDILineInfo(this, Offset))
302 return Info->FileName + ":" + std::to_string(Info->Line) + ":(" +
303 SecAndOffset + ")";
304
305 // File->SourceFile contains STT_FILE symbol that contains a
306 // source file name. If it's missing, we use an object file name.
307 std::string SrcFile = getFile<ELFT>()->SourceFile;
308 if (SrcFile.empty())
309 SrcFile = toString(File);
310
311 if (Defined *D = getEnclosingFunction<ELFT>(Offset))
312 return SrcFile + ":(function " + toString(*D) + ": " + SecAndOffset + ")";
313
314 // If there's no symbol, print out the offset in the section.
315 return (SrcFile + ":(" + SecAndOffset + ")");
316 }
317
318 // This function is intended to be used for constructing an error message.
319 // The returned message looks like this:
320 //
321 // foo.c:42 (/home/alice/possibly/very/long/path/foo.c:42)
322 //
323 // Returns an empty string if there's no way to get line info.
getSrcMsg(const Symbol & Sym,uint64_t Offset)324 std::string InputSectionBase::getSrcMsg(const Symbol &Sym, uint64_t Offset) {
325 return File->getSrcMsg(Sym, *this, Offset);
326 }
327
328 // Returns a filename string along with an optional section name. This
329 // function is intended to be used for constructing an error
330 // message. The returned message looks like this:
331 //
332 // path/to/foo.o:(function bar)
333 //
334 // or
335 //
336 // path/to/foo.o:(function bar) in archive path/to/bar.a
getObjMsg(uint64_t Off)337 std::string InputSectionBase::getObjMsg(uint64_t Off) {
338 std::string Filename = File->getName();
339
340 std::string Archive;
341 if (!File->ArchiveName.empty())
342 Archive = " in archive " + File->ArchiveName;
343
344 // Find a symbol that encloses a given location.
345 for (Symbol *B : File->getSymbols())
346 if (auto *D = dyn_cast<Defined>(B))
347 if (D->Section == this && D->Value <= Off && Off < D->Value + D->Size)
348 return Filename + ":(" + toString(*D) + ")" + Archive;
349
350 // If there's no symbol, print out the offset in the section.
351 return (Filename + ":(" + Name + "+0x" + utohexstr(Off) + ")" + Archive)
352 .str();
353 }
354
355 InputSection InputSection::Discarded(nullptr, 0, 0, 0, ArrayRef<uint8_t>(), "");
356
InputSection(InputFile * F,uint64_t Flags,uint32_t Type,uint32_t Alignment,ArrayRef<uint8_t> Data,StringRef Name,Kind K)357 InputSection::InputSection(InputFile *F, uint64_t Flags, uint32_t Type,
358 uint32_t Alignment, ArrayRef<uint8_t> Data,
359 StringRef Name, Kind K)
360 : InputSectionBase(F, Flags, Type,
361 /*Entsize*/ 0, /*Link*/ 0, /*Info*/ 0, Alignment, Data,
362 Name, K) {}
363
364 template <class ELFT>
InputSection(ObjFile<ELFT> & F,const typename ELFT::Shdr & Header,StringRef Name)365 InputSection::InputSection(ObjFile<ELFT> &F, const typename ELFT::Shdr &Header,
366 StringRef Name)
367 : InputSectionBase(F, Header, Name, InputSectionBase::Regular) {}
368
classof(const SectionBase * S)369 bool InputSection::classof(const SectionBase *S) {
370 return S->kind() == SectionBase::Regular ||
371 S->kind() == SectionBase::Synthetic;
372 }
373
getParent() const374 OutputSection *InputSection::getParent() const {
375 return cast_or_null<OutputSection>(Parent);
376 }
377
378 // Copy SHT_GROUP section contents. Used only for the -r option.
copyShtGroup(uint8_t * Buf)379 template <class ELFT> void InputSection::copyShtGroup(uint8_t *Buf) {
380 // ELFT::Word is the 32-bit integral type in the target endianness.
381 typedef typename ELFT::Word u32;
382 ArrayRef<u32> From = getDataAs<u32>();
383 auto *To = reinterpret_cast<u32 *>(Buf);
384
385 // The first entry is not a section number but a flag.
386 *To++ = From[0];
387
388 // Adjust section numbers because section numbers in an input object
389 // files are different in the output.
390 ArrayRef<InputSectionBase *> Sections = File->getSections();
391 for (uint32_t Idx : From.slice(1))
392 *To++ = Sections[Idx]->getOutputSection()->SectionIndex;
393 }
394
getRelocatedSection() const395 InputSectionBase *InputSection::getRelocatedSection() const {
396 if (!File || (Type != SHT_RELA && Type != SHT_REL))
397 return nullptr;
398 ArrayRef<InputSectionBase *> Sections = File->getSections();
399 return Sections[Info];
400 }
401
402 // This is used for -r and --emit-relocs. We can't use memcpy to copy
403 // relocations because we need to update symbol table offset and section index
404 // for each relocation. So we copy relocations one by one.
405 template <class ELFT, class RelTy>
copyRelocations(uint8_t * Buf,ArrayRef<RelTy> Rels)406 void InputSection::copyRelocations(uint8_t *Buf, ArrayRef<RelTy> Rels) {
407 InputSectionBase *Sec = getRelocatedSection();
408
409 for (const RelTy &Rel : Rels) {
410 RelType Type = Rel.getType(Config->IsMips64EL);
411 Symbol &Sym = getFile<ELFT>()->getRelocTargetSym(Rel);
412
413 auto *P = reinterpret_cast<typename ELFT::Rela *>(Buf);
414 Buf += sizeof(RelTy);
415
416 if (RelTy::IsRela)
417 P->r_addend = getAddend<ELFT>(Rel);
418
419 // Output section VA is zero for -r, so r_offset is an offset within the
420 // section, but for --emit-relocs it is an virtual address.
421 P->r_offset = Sec->getVA(Rel.r_offset);
422 P->setSymbolAndType(In.SymTab->getSymbolIndex(&Sym), Type,
423 Config->IsMips64EL);
424
425 if (Sym.Type == STT_SECTION) {
426 // We combine multiple section symbols into only one per
427 // section. This means we have to update the addend. That is
428 // trivial for Elf_Rela, but for Elf_Rel we have to write to the
429 // section data. We do that by adding to the Relocation vector.
430
431 // .eh_frame is horribly special and can reference discarded sections. To
432 // avoid having to parse and recreate .eh_frame, we just replace any
433 // relocation in it pointing to discarded sections with R_*_NONE, which
434 // hopefully creates a frame that is ignored at runtime.
435 auto *D = dyn_cast<Defined>(&Sym);
436 if (!D) {
437 error("STT_SECTION symbol should be defined");
438 continue;
439 }
440 SectionBase *Section = D->Section->Repl;
441 if (!Section->Live) {
442 P->setSymbolAndType(0, 0, false);
443 continue;
444 }
445
446 int64_t Addend = getAddend<ELFT>(Rel);
447 const uint8_t *BufLoc = Sec->data().begin() + Rel.r_offset;
448 if (!RelTy::IsRela)
449 Addend = Target->getImplicitAddend(BufLoc, Type);
450
451 if (Config->EMachine == EM_MIPS && Config->Relocatable &&
452 Target->getRelExpr(Type, Sym, BufLoc) == R_MIPS_GOTREL) {
453 // Some MIPS relocations depend on "gp" value. By default,
454 // this value has 0x7ff0 offset from a .got section. But
455 // relocatable files produced by a complier or a linker
456 // might redefine this default value and we must use it
457 // for a calculation of the relocation result. When we
458 // generate EXE or DSO it's trivial. Generating a relocatable
459 // output is more difficult case because the linker does
460 // not calculate relocations in this mode and loses
461 // individual "gp" values used by each input object file.
462 // As a workaround we add the "gp" value to the relocation
463 // addend and save it back to the file.
464 Addend += Sec->getFile<ELFT>()->MipsGp0;
465 }
466
467 if (RelTy::IsRela)
468 P->r_addend = Sym.getVA(Addend) - Section->getOutputSection()->Addr;
469 else if (Config->Relocatable)
470 Sec->Relocations.push_back({R_ABS, Type, Rel.r_offset, Addend, &Sym});
471 }
472 }
473 }
474
475 // The ARM and AArch64 ABI handle pc-relative relocations to undefined weak
476 // references specially. The general rule is that the value of the symbol in
477 // this context is the address of the place P. A further special case is that
478 // branch relocations to an undefined weak reference resolve to the next
479 // instruction.
getARMUndefinedRelativeWeakVA(RelType Type,uint32_t A,uint32_t P)480 static uint32_t getARMUndefinedRelativeWeakVA(RelType Type, uint32_t A,
481 uint32_t P) {
482 switch (Type) {
483 // Unresolved branch relocations to weak references resolve to next
484 // instruction, this will be either 2 or 4 bytes on from P.
485 case R_ARM_THM_JUMP11:
486 return P + 2 + A;
487 case R_ARM_CALL:
488 case R_ARM_JUMP24:
489 case R_ARM_PC24:
490 case R_ARM_PLT32:
491 case R_ARM_PREL31:
492 case R_ARM_THM_JUMP19:
493 case R_ARM_THM_JUMP24:
494 return P + 4 + A;
495 case R_ARM_THM_CALL:
496 // We don't want an interworking BLX to ARM
497 return P + 5 + A;
498 // Unresolved non branch pc-relative relocations
499 // R_ARM_TARGET2 which can be resolved relatively is not present as it never
500 // targets a weak-reference.
501 case R_ARM_MOVW_PREL_NC:
502 case R_ARM_MOVT_PREL:
503 case R_ARM_REL32:
504 case R_ARM_THM_MOVW_PREL_NC:
505 case R_ARM_THM_MOVT_PREL:
506 return P + A;
507 }
508 llvm_unreachable("ARM pc-relative relocation expected\n");
509 }
510
511 // The comment above getARMUndefinedRelativeWeakVA applies to this function.
getAArch64UndefinedRelativeWeakVA(uint64_t Type,uint64_t A,uint64_t P)512 static uint64_t getAArch64UndefinedRelativeWeakVA(uint64_t Type, uint64_t A,
513 uint64_t P) {
514 switch (Type) {
515 // Unresolved branch relocations to weak references resolve to next
516 // instruction, this is 4 bytes on from P.
517 case R_AARCH64_CALL26:
518 case R_AARCH64_CONDBR19:
519 case R_AARCH64_JUMP26:
520 case R_AARCH64_TSTBR14:
521 return P + 4 + A;
522 // Unresolved non branch pc-relative relocations
523 case R_AARCH64_PREL16:
524 case R_AARCH64_PREL32:
525 case R_AARCH64_PREL64:
526 case R_AARCH64_ADR_PREL_LO21:
527 case R_AARCH64_LD_PREL_LO19:
528 return P + A;
529 }
530 llvm_unreachable("AArch64 pc-relative relocation expected\n");
531 }
532
533 // ARM SBREL relocations are of the form S + A - B where B is the static base
534 // The ARM ABI defines base to be "addressing origin of the output segment
535 // defining the symbol S". We defined the "addressing origin"/static base to be
536 // the base of the PT_LOAD segment containing the Sym.
537 // The procedure call standard only defines a Read Write Position Independent
538 // RWPI variant so in practice we should expect the static base to be the base
539 // of the RW segment.
getARMStaticBase(const Symbol & Sym)540 static uint64_t getARMStaticBase(const Symbol &Sym) {
541 OutputSection *OS = Sym.getOutputSection();
542 if (!OS || !OS->PtLoad || !OS->PtLoad->FirstSec)
543 fatal("SBREL relocation to " + Sym.getName() + " without static base");
544 return OS->PtLoad->FirstSec->Addr;
545 }
546
547 // For R_RISCV_PC_INDIRECT (R_RISCV_PCREL_LO12_{I,S}), the symbol actually
548 // points the corresponding R_RISCV_PCREL_HI20 relocation, and the target VA
549 // is calculated using PCREL_HI20's symbol.
550 //
551 // This function returns the R_RISCV_PCREL_HI20 relocation from
552 // R_RISCV_PCREL_LO12's symbol and addend.
getRISCVPCRelHi20(const Symbol * Sym,uint64_t Addend)553 static Relocation *getRISCVPCRelHi20(const Symbol *Sym, uint64_t Addend) {
554 const Defined *D = cast<Defined>(Sym);
555 InputSection *IS = cast<InputSection>(D->Section);
556
557 if (Addend != 0)
558 warn("Non-zero addend in R_RISCV_PCREL_LO12 relocation to " +
559 IS->getObjMsg(D->Value) + " is ignored");
560
561 // Relocations are sorted by offset, so we can use std::equal_range to do
562 // binary search.
563 auto Range = std::equal_range(IS->Relocations.begin(), IS->Relocations.end(),
564 D->Value, RelocationOffsetComparator{});
565 for (auto It = std::get<0>(Range); It != std::get<1>(Range); ++It)
566 if (isRelExprOneOf<R_PC>(It->Expr))
567 return &*It;
568
569 error("R_RISCV_PCREL_LO12 relocation points to " + IS->getObjMsg(D->Value) +
570 " without an associated R_RISCV_PCREL_HI20 relocation");
571 return nullptr;
572 }
573
574 // A TLS symbol's virtual address is relative to the TLS segment. Add a
575 // target-specific adjustment to produce a thread-pointer-relative offset.
getTlsTpOffset()576 static int64_t getTlsTpOffset() {
577 switch (Config->EMachine) {
578 case EM_ARM:
579 case EM_AARCH64:
580 // Variant 1. The thread pointer points to a TCB with a fixed 2-word size,
581 // followed by a variable amount of alignment padding, followed by the TLS
582 // segment.
583 return alignTo(Config->Wordsize * 2, Out::TlsPhdr->p_align);
584 case EM_386:
585 case EM_X86_64:
586 // Variant 2. The TLS segment is located just before the thread pointer.
587 return -Out::TlsPhdr->p_memsz;
588 case EM_PPC64:
589 // The thread pointer points to a fixed offset from the start of the
590 // executable's TLS segment. An offset of 0x7000 allows a signed 16-bit
591 // offset to reach 0x1000 of TCB/thread-library data and 0xf000 of the
592 // program's TLS segment.
593 return -0x7000;
594 default:
595 llvm_unreachable("unhandled Config->EMachine");
596 }
597 }
598
getRelocTargetVA(const InputFile * File,RelType Type,int64_t A,uint64_t P,const Symbol & Sym,RelExpr Expr)599 static uint64_t getRelocTargetVA(const InputFile *File, RelType Type, int64_t A,
600 uint64_t P, const Symbol &Sym, RelExpr Expr) {
601 switch (Expr) {
602 case R_INVALID:
603 return 0;
604 case R_ABS:
605 case R_RELAX_TLS_LD_TO_LE_ABS:
606 case R_RELAX_GOT_PC_NOPIC:
607 return Sym.getVA(A);
608 case R_ADDEND:
609 return A;
610 case R_ARM_SBREL:
611 return Sym.getVA(A) - getARMStaticBase(Sym);
612 case R_GOT:
613 case R_GOT_PLT:
614 case R_RELAX_TLS_GD_TO_IE_ABS:
615 return Sym.getGotVA() + A;
616 case R_GOTONLY_PC:
617 return In.Got->getVA() + A - P;
618 case R_GOTONLY_PC_FROM_END:
619 return In.Got->getVA() + A - P + In.Got->getSize();
620 case R_GOTREL:
621 return Sym.getVA(A) - In.Got->getVA();
622 case R_GOTREL_FROM_END:
623 return Sym.getVA(A) - In.Got->getVA() - In.Got->getSize();
624 case R_GOT_FROM_END:
625 case R_RELAX_TLS_GD_TO_IE_END:
626 return Sym.getGotOffset() + A - In.Got->getSize();
627 case R_TLSLD_GOT_OFF:
628 case R_GOT_OFF:
629 case R_RELAX_TLS_GD_TO_IE_GOT_OFF:
630 return Sym.getGotOffset() + A;
631 case R_AARCH64_GOT_PAGE_PC:
632 case R_AARCH64_GOT_PAGE_PC_PLT:
633 case R_AARCH64_RELAX_TLS_GD_TO_IE_PAGE_PC:
634 return getAArch64Page(Sym.getGotVA() + A) - getAArch64Page(P);
635 case R_GOT_PC:
636 case R_RELAX_TLS_GD_TO_IE:
637 return Sym.getGotVA() + A - P;
638 case R_HEXAGON_GOT:
639 return Sym.getGotVA() - In.GotPlt->getVA();
640 case R_MIPS_GOTREL:
641 return Sym.getVA(A) - In.MipsGot->getGp(File);
642 case R_MIPS_GOT_GP:
643 return In.MipsGot->getGp(File) + A;
644 case R_MIPS_GOT_GP_PC: {
645 // R_MIPS_LO16 expression has R_MIPS_GOT_GP_PC type iif the target
646 // is _gp_disp symbol. In that case we should use the following
647 // formula for calculation "AHL + GP - P + 4". For details see p. 4-19 at
648 // ftp://www.linux-mips.org/pub/linux/mips/doc/ABI/mipsabi.pdf
649 // microMIPS variants of these relocations use slightly different
650 // expressions: AHL + GP - P + 3 for %lo() and AHL + GP - P - 1 for %hi()
651 // to correctly handle less-sugnificant bit of the microMIPS symbol.
652 uint64_t V = In.MipsGot->getGp(File) + A - P;
653 if (Type == R_MIPS_LO16 || Type == R_MICROMIPS_LO16)
654 V += 4;
655 if (Type == R_MICROMIPS_LO16 || Type == R_MICROMIPS_HI16)
656 V -= 1;
657 return V;
658 }
659 case R_MIPS_GOT_LOCAL_PAGE:
660 // If relocation against MIPS local symbol requires GOT entry, this entry
661 // should be initialized by 'page address'. This address is high 16-bits
662 // of sum the symbol's value and the addend.
663 return In.MipsGot->getVA() + In.MipsGot->getPageEntryOffset(File, Sym, A) -
664 In.MipsGot->getGp(File);
665 case R_MIPS_GOT_OFF:
666 case R_MIPS_GOT_OFF32:
667 // In case of MIPS if a GOT relocation has non-zero addend this addend
668 // should be applied to the GOT entry content not to the GOT entry offset.
669 // That is why we use separate expression type.
670 return In.MipsGot->getVA() + In.MipsGot->getSymEntryOffset(File, Sym, A) -
671 In.MipsGot->getGp(File);
672 case R_MIPS_TLSGD:
673 return In.MipsGot->getVA() + In.MipsGot->getGlobalDynOffset(File, Sym) -
674 In.MipsGot->getGp(File);
675 case R_MIPS_TLSLD:
676 return In.MipsGot->getVA() + In.MipsGot->getTlsIndexOffset(File) -
677 In.MipsGot->getGp(File);
678 case R_AARCH64_PAGE_PC: {
679 uint64_t Val = Sym.isUndefWeak() ? P + A : Sym.getVA(A);
680 return getAArch64Page(Val) - getAArch64Page(P);
681 }
682 case R_AARCH64_PLT_PAGE_PC: {
683 uint64_t Val = Sym.isUndefWeak() ? P + A : Sym.getPltVA() + A;
684 return getAArch64Page(Val) - getAArch64Page(P);
685 }
686 case R_RISCV_PC_INDIRECT: {
687 if (const Relocation *HiRel = getRISCVPCRelHi20(&Sym, A))
688 return getRelocTargetVA(File, HiRel->Type, HiRel->Addend, Sym.getVA(),
689 *HiRel->Sym, HiRel->Expr);
690 return 0;
691 }
692 case R_PC: {
693 uint64_t Dest;
694 if (Sym.isUndefWeak()) {
695 // On ARM and AArch64 a branch to an undefined weak resolves to the
696 // next instruction, otherwise the place.
697 if (Config->EMachine == EM_ARM)
698 Dest = getARMUndefinedRelativeWeakVA(Type, A, P);
699 else if (Config->EMachine == EM_AARCH64)
700 Dest = getAArch64UndefinedRelativeWeakVA(Type, A, P);
701 else
702 Dest = Sym.getVA(A);
703 } else {
704 Dest = Sym.getVA(A);
705 }
706 return Dest - P;
707 }
708 case R_PLT:
709 return Sym.getPltVA() + A;
710 case R_PLT_PC:
711 case R_PPC_CALL_PLT:
712 return Sym.getPltVA() + A - P;
713 case R_PPC_CALL: {
714 uint64_t SymVA = Sym.getVA(A);
715 // If we have an undefined weak symbol, we might get here with a symbol
716 // address of zero. That could overflow, but the code must be unreachable,
717 // so don't bother doing anything at all.
718 if (!SymVA)
719 return 0;
720
721 // PPC64 V2 ABI describes two entry points to a function. The global entry
722 // point is used for calls where the caller and callee (may) have different
723 // TOC base pointers and r2 needs to be modified to hold the TOC base for
724 // the callee. For local calls the caller and callee share the same
725 // TOC base and so the TOC pointer initialization code should be skipped by
726 // branching to the local entry point.
727 return SymVA - P + getPPC64GlobalEntryToLocalEntryOffset(Sym.StOther);
728 }
729 case R_PPC_TOC:
730 return getPPC64TocBase() + A;
731 case R_RELAX_GOT_PC:
732 return Sym.getVA(A) - P;
733 case R_RELAX_TLS_GD_TO_LE:
734 case R_RELAX_TLS_IE_TO_LE:
735 case R_RELAX_TLS_LD_TO_LE:
736 case R_TLS:
737 // A weak undefined TLS symbol resolves to the base of the TLS
738 // block, i.e. gets a value of zero. If we pass --gc-sections to
739 // lld and .tbss is not referenced, it gets reclaimed and we don't
740 // create a TLS program header. Therefore, we resolve this
741 // statically to zero.
742 if (Sym.isTls() && Sym.isUndefWeak())
743 return 0;
744 return Sym.getVA(A) + getTlsTpOffset();
745 case R_RELAX_TLS_GD_TO_LE_NEG:
746 case R_NEG_TLS:
747 return Out::TlsPhdr->p_memsz - Sym.getVA(A);
748 case R_SIZE:
749 return Sym.getSize() + A;
750 case R_TLSDESC:
751 return In.Got->getGlobalDynAddr(Sym) + A;
752 case R_AARCH64_TLSDESC_PAGE:
753 return getAArch64Page(In.Got->getGlobalDynAddr(Sym) + A) -
754 getAArch64Page(P);
755 case R_TLSGD_GOT:
756 return In.Got->getGlobalDynOffset(Sym) + A;
757 case R_TLSGD_GOT_FROM_END:
758 return In.Got->getGlobalDynOffset(Sym) + A - In.Got->getSize();
759 case R_TLSGD_PC:
760 return In.Got->getGlobalDynAddr(Sym) + A - P;
761 case R_TLSLD_GOT_FROM_END:
762 return In.Got->getTlsIndexOff() + A - In.Got->getSize();
763 case R_TLSLD_GOT:
764 return In.Got->getTlsIndexOff() + A;
765 case R_TLSLD_PC:
766 return In.Got->getTlsIndexVA() + A - P;
767 default:
768 llvm_unreachable("invalid expression");
769 }
770 }
771
772 // This function applies relocations to sections without SHF_ALLOC bit.
773 // Such sections are never mapped to memory at runtime. Debug sections are
774 // an example. Relocations in non-alloc sections are much easier to
775 // handle than in allocated sections because it will never need complex
776 // treatement such as GOT or PLT (because at runtime no one refers them).
777 // So, we handle relocations for non-alloc sections directly in this
778 // function as a performance optimization.
779 template <class ELFT, class RelTy>
relocateNonAlloc(uint8_t * Buf,ArrayRef<RelTy> Rels)780 void InputSection::relocateNonAlloc(uint8_t *Buf, ArrayRef<RelTy> Rels) {
781 const unsigned Bits = sizeof(typename ELFT::uint) * 8;
782
783 for (const RelTy &Rel : Rels) {
784 RelType Type = Rel.getType(Config->IsMips64EL);
785
786 // GCC 8.0 or earlier have a bug that they emit R_386_GOTPC relocations
787 // against _GLOBAL_OFFSET_TABLE_ for .debug_info. The bug has been fixed
788 // in 2017 (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82630), but we
789 // need to keep this bug-compatible code for a while.
790 if (Config->EMachine == EM_386 && Type == R_386_GOTPC)
791 continue;
792
793 uint64_t Offset = getOffset(Rel.r_offset);
794 uint8_t *BufLoc = Buf + Offset;
795 int64_t Addend = getAddend<ELFT>(Rel);
796 if (!RelTy::IsRela)
797 Addend += Target->getImplicitAddend(BufLoc, Type);
798
799 Symbol &Sym = getFile<ELFT>()->getRelocTargetSym(Rel);
800 RelExpr Expr = Target->getRelExpr(Type, Sym, BufLoc);
801 if (Expr == R_NONE)
802 continue;
803
804 if (Expr != R_ABS) {
805 std::string Msg = getLocation<ELFT>(Offset) +
806 ": has non-ABS relocation " + toString(Type) +
807 " against symbol '" + toString(Sym) + "'";
808 if (Expr != R_PC) {
809 error(Msg);
810 return;
811 }
812
813 // If the control reaches here, we found a PC-relative relocation in a
814 // non-ALLOC section. Since non-ALLOC section is not loaded into memory
815 // at runtime, the notion of PC-relative doesn't make sense here. So,
816 // this is a usage error. However, GNU linkers historically accept such
817 // relocations without any errors and relocate them as if they were at
818 // address 0. For bug-compatibilty, we accept them with warnings. We
819 // know Steel Bank Common Lisp as of 2018 have this bug.
820 warn(Msg);
821 Target->relocateOne(BufLoc, Type,
822 SignExtend64<Bits>(Sym.getVA(Addend - Offset)));
823 continue;
824 }
825
826 if (Sym.isTls() && !Out::TlsPhdr)
827 Target->relocateOne(BufLoc, Type, 0);
828 else
829 Target->relocateOne(BufLoc, Type, SignExtend64<Bits>(Sym.getVA(Addend)));
830 }
831 }
832
833 // This is used when '-r' is given.
834 // For REL targets, InputSection::copyRelocations() may store artificial
835 // relocations aimed to update addends. They are handled in relocateAlloc()
836 // for allocatable sections, and this function does the same for
837 // non-allocatable sections, such as sections with debug information.
relocateNonAllocForRelocatable(InputSection * Sec,uint8_t * Buf)838 static void relocateNonAllocForRelocatable(InputSection *Sec, uint8_t *Buf) {
839 const unsigned Bits = Config->Is64 ? 64 : 32;
840
841 for (const Relocation &Rel : Sec->Relocations) {
842 // InputSection::copyRelocations() adds only R_ABS relocations.
843 assert(Rel.Expr == R_ABS);
844 uint8_t *BufLoc = Buf + Rel.Offset + Sec->OutSecOff;
845 uint64_t TargetVA = SignExtend64(Rel.Sym->getVA(Rel.Addend), Bits);
846 Target->relocateOne(BufLoc, Rel.Type, TargetVA);
847 }
848 }
849
850 template <class ELFT>
relocate(uint8_t * Buf,uint8_t * BufEnd)851 void InputSectionBase::relocate(uint8_t *Buf, uint8_t *BufEnd) {
852 if (Flags & SHF_EXECINSTR)
853 adjustSplitStackFunctionPrologues<ELFT>(Buf, BufEnd);
854
855 if (Flags & SHF_ALLOC) {
856 relocateAlloc(Buf, BufEnd);
857 return;
858 }
859
860 auto *Sec = cast<InputSection>(this);
861 if (Config->Relocatable)
862 relocateNonAllocForRelocatable(Sec, Buf);
863 else if (Sec->AreRelocsRela)
864 Sec->relocateNonAlloc<ELFT>(Buf, Sec->template relas<ELFT>());
865 else
866 Sec->relocateNonAlloc<ELFT>(Buf, Sec->template rels<ELFT>());
867 }
868
relocateAlloc(uint8_t * Buf,uint8_t * BufEnd)869 void InputSectionBase::relocateAlloc(uint8_t *Buf, uint8_t *BufEnd) {
870 assert(Flags & SHF_ALLOC);
871 const unsigned Bits = Config->Wordsize * 8;
872
873 for (const Relocation &Rel : Relocations) {
874 uint64_t Offset = Rel.Offset;
875 if (auto *Sec = dyn_cast<InputSection>(this))
876 Offset += Sec->OutSecOff;
877 uint8_t *BufLoc = Buf + Offset;
878 RelType Type = Rel.Type;
879
880 uint64_t AddrLoc = getOutputSection()->Addr + Offset;
881 RelExpr Expr = Rel.Expr;
882 uint64_t TargetVA = SignExtend64(
883 getRelocTargetVA(File, Type, Rel.Addend, AddrLoc, *Rel.Sym, Expr),
884 Bits);
885
886 switch (Expr) {
887 case R_RELAX_GOT_PC:
888 case R_RELAX_GOT_PC_NOPIC:
889 Target->relaxGot(BufLoc, TargetVA);
890 break;
891 case R_RELAX_TLS_IE_TO_LE:
892 Target->relaxTlsIeToLe(BufLoc, Type, TargetVA);
893 break;
894 case R_RELAX_TLS_LD_TO_LE:
895 case R_RELAX_TLS_LD_TO_LE_ABS:
896 Target->relaxTlsLdToLe(BufLoc, Type, TargetVA);
897 break;
898 case R_RELAX_TLS_GD_TO_LE:
899 case R_RELAX_TLS_GD_TO_LE_NEG:
900 Target->relaxTlsGdToLe(BufLoc, Type, TargetVA);
901 break;
902 case R_AARCH64_RELAX_TLS_GD_TO_IE_PAGE_PC:
903 case R_RELAX_TLS_GD_TO_IE:
904 case R_RELAX_TLS_GD_TO_IE_ABS:
905 case R_RELAX_TLS_GD_TO_IE_GOT_OFF:
906 case R_RELAX_TLS_GD_TO_IE_END:
907 Target->relaxTlsGdToIe(BufLoc, Type, TargetVA);
908 break;
909 case R_PPC_CALL:
910 // If this is a call to __tls_get_addr, it may be part of a TLS
911 // sequence that has been relaxed and turned into a nop. In this
912 // case, we don't want to handle it as a call.
913 if (read32(BufLoc) == 0x60000000) // nop
914 break;
915
916 // Patch a nop (0x60000000) to a ld.
917 if (Rel.Sym->NeedsTocRestore) {
918 if (BufLoc + 8 > BufEnd || read32(BufLoc + 4) != 0x60000000) {
919 error(getErrorLocation(BufLoc) + "call lacks nop, can't restore toc");
920 break;
921 }
922 write32(BufLoc + 4, 0xe8410018); // ld %r2, 24(%r1)
923 }
924 Target->relocateOne(BufLoc, Type, TargetVA);
925 break;
926 default:
927 Target->relocateOne(BufLoc, Type, TargetVA);
928 break;
929 }
930 }
931 }
932
933 // For each function-defining prologue, find any calls to __morestack,
934 // and replace them with calls to __morestack_non_split.
switchMorestackCallsToMorestackNonSplit(DenseSet<Defined * > & Prologues,std::vector<Relocation * > & MorestackCalls)935 static void switchMorestackCallsToMorestackNonSplit(
936 DenseSet<Defined *> &Prologues, std::vector<Relocation *> &MorestackCalls) {
937
938 // If the target adjusted a function's prologue, all calls to
939 // __morestack inside that function should be switched to
940 // __morestack_non_split.
941 Symbol *MoreStackNonSplit = Symtab->find("__morestack_non_split");
942 if (!MoreStackNonSplit) {
943 error("Mixing split-stack objects requires a definition of "
944 "__morestack_non_split");
945 return;
946 }
947
948 // Sort both collections to compare addresses efficiently.
949 llvm::sort(MorestackCalls, [](const Relocation *L, const Relocation *R) {
950 return L->Offset < R->Offset;
951 });
952 std::vector<Defined *> Functions(Prologues.begin(), Prologues.end());
953 llvm::sort(Functions, [](const Defined *L, const Defined *R) {
954 return L->Value < R->Value;
955 });
956
957 auto It = MorestackCalls.begin();
958 for (Defined *F : Functions) {
959 // Find the first call to __morestack within the function.
960 while (It != MorestackCalls.end() && (*It)->Offset < F->Value)
961 ++It;
962 // Adjust all calls inside the function.
963 while (It != MorestackCalls.end() && (*It)->Offset < F->Value + F->Size) {
964 (*It)->Sym = MoreStackNonSplit;
965 ++It;
966 }
967 }
968 }
969
enclosingPrologueAttempted(uint64_t Offset,const DenseSet<Defined * > & Prologues)970 static bool enclosingPrologueAttempted(uint64_t Offset,
971 const DenseSet<Defined *> &Prologues) {
972 for (Defined *F : Prologues)
973 if (F->Value <= Offset && Offset < F->Value + F->Size)
974 return true;
975 return false;
976 }
977
978 // If a function compiled for split stack calls a function not
979 // compiled for split stack, then the caller needs its prologue
980 // adjusted to ensure that the called function will have enough stack
981 // available. Find those functions, and adjust their prologues.
982 template <class ELFT>
adjustSplitStackFunctionPrologues(uint8_t * Buf,uint8_t * End)983 void InputSectionBase::adjustSplitStackFunctionPrologues(uint8_t *Buf,
984 uint8_t *End) {
985 if (!getFile<ELFT>()->SplitStack)
986 return;
987 DenseSet<Defined *> Prologues;
988 std::vector<Relocation *> MorestackCalls;
989
990 for (Relocation &Rel : Relocations) {
991 // Local symbols can't possibly be cross-calls, and should have been
992 // resolved long before this line.
993 if (Rel.Sym->isLocal())
994 continue;
995
996 // Ignore calls into the split-stack api.
997 if (Rel.Sym->getName().startswith("__morestack")) {
998 if (Rel.Sym->getName().equals("__morestack"))
999 MorestackCalls.push_back(&Rel);
1000 continue;
1001 }
1002
1003 // A relocation to non-function isn't relevant. Sometimes
1004 // __morestack is not marked as a function, so this check comes
1005 // after the name check.
1006 if (Rel.Sym->Type != STT_FUNC)
1007 continue;
1008
1009 // If the callee's-file was compiled with split stack, nothing to do. In
1010 // this context, a "Defined" symbol is one "defined by the binary currently
1011 // being produced". So an "undefined" symbol might be provided by a shared
1012 // library. It is not possible to tell how such symbols were compiled, so be
1013 // conservative.
1014 if (Defined *D = dyn_cast<Defined>(Rel.Sym))
1015 if (InputSection *IS = cast_or_null<InputSection>(D->Section))
1016 if (!IS || !IS->getFile<ELFT>() || IS->getFile<ELFT>()->SplitStack)
1017 continue;
1018
1019 if (enclosingPrologueAttempted(Rel.Offset, Prologues))
1020 continue;
1021
1022 if (Defined *F = getEnclosingFunction<ELFT>(Rel.Offset)) {
1023 Prologues.insert(F);
1024 if (Target->adjustPrologueForCrossSplitStack(Buf + getOffset(F->Value),
1025 End, F->StOther))
1026 continue;
1027 if (!getFile<ELFT>()->SomeNoSplitStack)
1028 error(lld::toString(this) + ": " + F->getName() +
1029 " (with -fsplit-stack) calls " + Rel.Sym->getName() +
1030 " (without -fsplit-stack), but couldn't adjust its prologue");
1031 }
1032 }
1033
1034 if (Target->NeedsMoreStackNonSplit)
1035 switchMorestackCallsToMorestackNonSplit(Prologues, MorestackCalls);
1036 }
1037
writeTo(uint8_t * Buf)1038 template <class ELFT> void InputSection::writeTo(uint8_t *Buf) {
1039 if (Type == SHT_NOBITS)
1040 return;
1041
1042 if (auto *S = dyn_cast<SyntheticSection>(this)) {
1043 S->writeTo(Buf + OutSecOff);
1044 return;
1045 }
1046
1047 // If -r or --emit-relocs is given, then an InputSection
1048 // may be a relocation section.
1049 if (Type == SHT_RELA) {
1050 copyRelocations<ELFT>(Buf + OutSecOff, getDataAs<typename ELFT::Rela>());
1051 return;
1052 }
1053 if (Type == SHT_REL) {
1054 copyRelocations<ELFT>(Buf + OutSecOff, getDataAs<typename ELFT::Rel>());
1055 return;
1056 }
1057
1058 // If -r is given, we may have a SHT_GROUP section.
1059 if (Type == SHT_GROUP) {
1060 copyShtGroup<ELFT>(Buf + OutSecOff);
1061 return;
1062 }
1063
1064 // If this is a compressed section, uncompress section contents directly
1065 // to the buffer.
1066 if (UncompressedSize >= 0 && !UncompressedBuf) {
1067 size_t Size = UncompressedSize;
1068 if (Error E = zlib::uncompress(toStringRef(RawData),
1069 (char *)(Buf + OutSecOff), Size))
1070 fatal(toString(this) +
1071 ": uncompress failed: " + llvm::toString(std::move(E)));
1072 uint8_t *BufEnd = Buf + OutSecOff + Size;
1073 relocate<ELFT>(Buf, BufEnd);
1074 return;
1075 }
1076
1077 // Copy section contents from source object file to output file
1078 // and then apply relocations.
1079 memcpy(Buf + OutSecOff, data().data(), data().size());
1080 uint8_t *BufEnd = Buf + OutSecOff + data().size();
1081 relocate<ELFT>(Buf, BufEnd);
1082 }
1083
replace(InputSection * Other)1084 void InputSection::replace(InputSection *Other) {
1085 Alignment = std::max(Alignment, Other->Alignment);
1086 Other->Repl = Repl;
1087 Other->Live = false;
1088 }
1089
1090 template <class ELFT>
EhInputSection(ObjFile<ELFT> & F,const typename ELFT::Shdr & Header,StringRef Name)1091 EhInputSection::EhInputSection(ObjFile<ELFT> &F,
1092 const typename ELFT::Shdr &Header,
1093 StringRef Name)
1094 : InputSectionBase(F, Header, Name, InputSectionBase::EHFrame) {}
1095
getParent() const1096 SyntheticSection *EhInputSection::getParent() const {
1097 return cast_or_null<SyntheticSection>(Parent);
1098 }
1099
1100 // Returns the index of the first relocation that points to a region between
1101 // Begin and Begin+Size.
1102 template <class IntTy, class RelTy>
getReloc(IntTy Begin,IntTy Size,const ArrayRef<RelTy> & Rels,unsigned & RelocI)1103 static unsigned getReloc(IntTy Begin, IntTy Size, const ArrayRef<RelTy> &Rels,
1104 unsigned &RelocI) {
1105 // Start search from RelocI for fast access. That works because the
1106 // relocations are sorted in .eh_frame.
1107 for (unsigned N = Rels.size(); RelocI < N; ++RelocI) {
1108 const RelTy &Rel = Rels[RelocI];
1109 if (Rel.r_offset < Begin)
1110 continue;
1111
1112 if (Rel.r_offset < Begin + Size)
1113 return RelocI;
1114 return -1;
1115 }
1116 return -1;
1117 }
1118
1119 // .eh_frame is a sequence of CIE or FDE records.
1120 // This function splits an input section into records and returns them.
split()1121 template <class ELFT> void EhInputSection::split() {
1122 if (AreRelocsRela)
1123 split<ELFT>(relas<ELFT>());
1124 else
1125 split<ELFT>(rels<ELFT>());
1126 }
1127
1128 template <class ELFT, class RelTy>
split(ArrayRef<RelTy> Rels)1129 void EhInputSection::split(ArrayRef<RelTy> Rels) {
1130 unsigned RelI = 0;
1131 for (size_t Off = 0, End = data().size(); Off != End;) {
1132 size_t Size = readEhRecordSize(this, Off);
1133 Pieces.emplace_back(Off, this, Size, getReloc(Off, Size, Rels, RelI));
1134 // The empty record is the end marker.
1135 if (Size == 4)
1136 break;
1137 Off += Size;
1138 }
1139 }
1140
findNull(StringRef S,size_t EntSize)1141 static size_t findNull(StringRef S, size_t EntSize) {
1142 // Optimize the common case.
1143 if (EntSize == 1)
1144 return S.find(0);
1145
1146 for (unsigned I = 0, N = S.size(); I != N; I += EntSize) {
1147 const char *B = S.begin() + I;
1148 if (std::all_of(B, B + EntSize, [](char C) { return C == 0; }))
1149 return I;
1150 }
1151 return StringRef::npos;
1152 }
1153
getParent() const1154 SyntheticSection *MergeInputSection::getParent() const {
1155 return cast_or_null<SyntheticSection>(Parent);
1156 }
1157
1158 // Split SHF_STRINGS section. Such section is a sequence of
1159 // null-terminated strings.
splitStrings(ArrayRef<uint8_t> Data,size_t EntSize)1160 void MergeInputSection::splitStrings(ArrayRef<uint8_t> Data, size_t EntSize) {
1161 size_t Off = 0;
1162 bool IsAlloc = Flags & SHF_ALLOC;
1163 StringRef S = toStringRef(Data);
1164
1165 while (!S.empty()) {
1166 size_t End = findNull(S, EntSize);
1167 if (End == StringRef::npos)
1168 fatal(toString(this) + ": string is not null terminated");
1169 size_t Size = End + EntSize;
1170
1171 Pieces.emplace_back(Off, xxHash64(S.substr(0, Size)), !IsAlloc);
1172 S = S.substr(Size);
1173 Off += Size;
1174 }
1175 }
1176
1177 // Split non-SHF_STRINGS section. Such section is a sequence of
1178 // fixed size records.
splitNonStrings(ArrayRef<uint8_t> Data,size_t EntSize)1179 void MergeInputSection::splitNonStrings(ArrayRef<uint8_t> Data,
1180 size_t EntSize) {
1181 size_t Size = Data.size();
1182 assert((Size % EntSize) == 0);
1183 bool IsAlloc = Flags & SHF_ALLOC;
1184
1185 for (size_t I = 0; I != Size; I += EntSize)
1186 Pieces.emplace_back(I, xxHash64(Data.slice(I, EntSize)), !IsAlloc);
1187 }
1188
1189 template <class ELFT>
MergeInputSection(ObjFile<ELFT> & F,const typename ELFT::Shdr & Header,StringRef Name)1190 MergeInputSection::MergeInputSection(ObjFile<ELFT> &F,
1191 const typename ELFT::Shdr &Header,
1192 StringRef Name)
1193 : InputSectionBase(F, Header, Name, InputSectionBase::Merge) {}
1194
MergeInputSection(uint64_t Flags,uint32_t Type,uint64_t Entsize,ArrayRef<uint8_t> Data,StringRef Name)1195 MergeInputSection::MergeInputSection(uint64_t Flags, uint32_t Type,
1196 uint64_t Entsize, ArrayRef<uint8_t> Data,
1197 StringRef Name)
1198 : InputSectionBase(nullptr, Flags, Type, Entsize, /*Link*/ 0, /*Info*/ 0,
1199 /*Alignment*/ Entsize, Data, Name, SectionBase::Merge) {}
1200
1201 // This function is called after we obtain a complete list of input sections
1202 // that need to be linked. This is responsible to split section contents
1203 // into small chunks for further processing.
1204 //
1205 // Note that this function is called from parallelForEach. This must be
1206 // thread-safe (i.e. no memory allocation from the pools).
splitIntoPieces()1207 void MergeInputSection::splitIntoPieces() {
1208 assert(Pieces.empty());
1209
1210 if (Flags & SHF_STRINGS)
1211 splitStrings(data(), Entsize);
1212 else
1213 splitNonStrings(data(), Entsize);
1214 }
1215
getSectionPiece(uint64_t Offset)1216 SectionPiece *MergeInputSection::getSectionPiece(uint64_t Offset) {
1217 if (this->data().size() <= Offset)
1218 fatal(toString(this) + ": offset is outside the section");
1219
1220 // If Offset is not at beginning of a section piece, it is not in the map.
1221 // In that case we need to do a binary search of the original section piece vector.
1222 auto It2 =
1223 llvm::upper_bound(Pieces, Offset, [](uint64_t Offset, SectionPiece P) {
1224 return Offset < P.InputOff;
1225 });
1226 return &It2[-1];
1227 }
1228
1229 // Returns the offset in an output section for a given input offset.
1230 // Because contents of a mergeable section is not contiguous in output,
1231 // it is not just an addition to a base output offset.
getParentOffset(uint64_t Offset) const1232 uint64_t MergeInputSection::getParentOffset(uint64_t Offset) const {
1233 // If Offset is not at beginning of a section piece, it is not in the map.
1234 // In that case we need to search from the original section piece vector.
1235 const SectionPiece &Piece =
1236 *(const_cast<MergeInputSection *>(this)->getSectionPiece (Offset));
1237 uint64_t Addend = Offset - Piece.InputOff;
1238 return Piece.OutputOff + Addend;
1239 }
1240
1241 template InputSection::InputSection(ObjFile<ELF32LE> &, const ELF32LE::Shdr &,
1242 StringRef);
1243 template InputSection::InputSection(ObjFile<ELF32BE> &, const ELF32BE::Shdr &,
1244 StringRef);
1245 template InputSection::InputSection(ObjFile<ELF64LE> &, const ELF64LE::Shdr &,
1246 StringRef);
1247 template InputSection::InputSection(ObjFile<ELF64BE> &, const ELF64BE::Shdr &,
1248 StringRef);
1249
1250 template std::string InputSectionBase::getLocation<ELF32LE>(uint64_t);
1251 template std::string InputSectionBase::getLocation<ELF32BE>(uint64_t);
1252 template std::string InputSectionBase::getLocation<ELF64LE>(uint64_t);
1253 template std::string InputSectionBase::getLocation<ELF64BE>(uint64_t);
1254
1255 template void InputSection::writeTo<ELF32LE>(uint8_t *);
1256 template void InputSection::writeTo<ELF32BE>(uint8_t *);
1257 template void InputSection::writeTo<ELF64LE>(uint8_t *);
1258 template void InputSection::writeTo<ELF64BE>(uint8_t *);
1259
1260 template MergeInputSection::MergeInputSection(ObjFile<ELF32LE> &,
1261 const ELF32LE::Shdr &, StringRef);
1262 template MergeInputSection::MergeInputSection(ObjFile<ELF32BE> &,
1263 const ELF32BE::Shdr &, StringRef);
1264 template MergeInputSection::MergeInputSection(ObjFile<ELF64LE> &,
1265 const ELF64LE::Shdr &, StringRef);
1266 template MergeInputSection::MergeInputSection(ObjFile<ELF64BE> &,
1267 const ELF64BE::Shdr &, StringRef);
1268
1269 template EhInputSection::EhInputSection(ObjFile<ELF32LE> &,
1270 const ELF32LE::Shdr &, StringRef);
1271 template EhInputSection::EhInputSection(ObjFile<ELF32BE> &,
1272 const ELF32BE::Shdr &, StringRef);
1273 template EhInputSection::EhInputSection(ObjFile<ELF64LE> &,
1274 const ELF64LE::Shdr &, StringRef);
1275 template EhInputSection::EhInputSection(ObjFile<ELF64BE> &,
1276 const ELF64BE::Shdr &, StringRef);
1277
1278 template void EhInputSection::split<ELF32LE>();
1279 template void EhInputSection::split<ELF32BE>();
1280 template void EhInputSection::split<ELF64LE>();
1281 template void EhInputSection::split<ELF64BE>();
1282