1 //===- OutputSections.cpp -------------------------------------------------===//
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
7 //===----------------------------------------------------------------------===//
9 #include "OutputSections.h"
11 #include "LinkerScript.h"
12 #include "SymbolTable.h"
13 #include "SyntheticSections.h"
15 #include "lld/Common/Memory.h"
16 #include "lld/Common/Strings.h"
17 #include "llvm/BinaryFormat/Dwarf.h"
18 #include "llvm/Support/Compression.h"
19 #include "llvm/Support/MD5.h"
20 #include "llvm/Support/MathExtras.h"
21 #include "llvm/Support/Parallel.h"
22 #include "llvm/Support/SHA1.h"
26 using namespace llvm::dwarf;
27 using namespace llvm::object;
28 using namespace llvm::support::endian;
29 using namespace llvm::ELF;
31 using namespace lld::elf;
33 uint8_t *Out::bufferStart;
35 PhdrEntry *Out::tlsPhdr;
36 OutputSection *Out::elfHeader;
37 OutputSection *Out::programHeaders;
38 OutputSection *Out::preinitArray;
39 OutputSection *Out::initArray;
40 OutputSection *Out::finiArray;
42 std::vector<OutputSection *> elf::outputSections;
44 uint32_t OutputSection::getPhdrFlags() const {
46 if (config->emachine != EM_ARM || !(flags & SHF_ARM_PURECODE))
48 if (flags & SHF_WRITE)
50 if (flags & SHF_EXECINSTR)
56 void OutputSection::writeHeaderTo(typename ELFT::Shdr *shdr) {
57 shdr->sh_entsize = entsize;
58 shdr->sh_addralign = alignment;
60 shdr->sh_offset = offset;
61 shdr->sh_flags = flags;
66 shdr->sh_name = shName;
69 OutputSection::OutputSection(StringRef name, uint32_t type, uint64_t flags)
70 : BaseCommand(OutputSectionKind),
71 SectionBase(Output, name, flags, /*Entsize*/ 0, /*Alignment*/ 1, type,
72 /*Info*/ 0, /*Link*/ 0) {}
74 // We allow sections of types listed below to merged into a
75 // single progbits section. This is typically done by linker
76 // scripts. Merging nobits and progbits will force disk space
77 // to be allocated for nobits sections. Other ones don't require
78 // any special treatment on top of progbits, so there doesn't
79 // seem to be a harm in merging them.
80 static bool canMergeToProgbits(unsigned type) {
81 return type == SHT_NOBITS || type == SHT_PROGBITS || type == SHT_INIT_ARRAY ||
82 type == SHT_PREINIT_ARRAY || type == SHT_FINI_ARRAY ||
86 // Record that isec will be placed in the OutputSection. isec does not become
87 // permanent until finalizeInputSections() is called. The function should not be
88 // used after finalizeInputSections() is called. If you need to add an
89 // InputSection post finalizeInputSections(), then you must do the following:
91 // 1. Find or create an InputSectionDescription to hold InputSection.
92 // 2. Add the InputSection to the InputSectionDescription::sections.
93 // 3. Call commitSection(isec).
94 void OutputSection::recordSection(InputSectionBase *isec) {
95 partition = isec->partition;
97 if (sectionCommands.empty() ||
98 !isa<InputSectionDescription>(sectionCommands.back()))
99 sectionCommands.push_back(make<InputSectionDescription>(""));
100 auto *isd = cast<InputSectionDescription>(sectionCommands.back());
101 isd->sectionBases.push_back(isec);
104 // Update fields (type, flags, alignment, etc) according to the InputSection
105 // isec. Also check whether the InputSection flags and type are consistent with
106 // other InputSections.
107 void OutputSection::commitSection(InputSection *isec) {
108 if (!hasInputSections) {
109 // If IS is the first section to be added to this section,
110 // initialize type, entsize and flags from isec.
111 hasInputSections = true;
113 entsize = isec->entsize;
116 // Otherwise, check if new type or flags are compatible with existing ones.
117 if ((flags ^ isec->flags) & SHF_TLS)
118 error("incompatible section flags for " + name + "\n>>> " + toString(isec) +
119 ": 0x" + utohexstr(isec->flags) + "\n>>> output section " + name +
120 ": 0x" + utohexstr(flags));
122 if (type != isec->type) {
123 if (!canMergeToProgbits(type) || !canMergeToProgbits(isec->type))
124 error("section type mismatch for " + isec->name + "\n>>> " +
125 toString(isec) + ": " +
126 getELFSectionTypeName(config->emachine, isec->type) +
127 "\n>>> output section " + name + ": " +
128 getELFSectionTypeName(config->emachine, type));
137 config->emachine == EM_ARM ? (uint64_t)SHF_ARM_PURECODE : 0;
138 uint64_t orMask = ~andMask;
139 uint64_t andFlags = (flags & isec->flags) & andMask;
140 uint64_t orFlags = (flags | isec->flags) & orMask;
141 flags = andFlags | orFlags;
143 flags &= ~(uint64_t)SHF_ALLOC;
145 alignment = std::max(alignment, isec->alignment);
147 // If this section contains a table of fixed-size entries, sh_entsize
148 // holds the element size. If it contains elements of different size we
149 // set sh_entsize to 0.
150 if (entsize != isec->entsize)
154 // This function scans over the InputSectionBase list sectionBases to create
155 // InputSectionDescription::sections.
157 // It removes MergeInputSections from the input section array and adds
158 // new synthetic sections at the location of the first input section
159 // that it replaces. It then finalizes each synthetic section in order
160 // to compute an output offset for each piece of each input section.
161 void OutputSection::finalizeInputSections() {
162 std::vector<MergeSyntheticSection *> mergeSections;
163 for (BaseCommand *base : sectionCommands) {
164 auto *cmd = dyn_cast<InputSectionDescription>(base);
167 cmd->sections.reserve(cmd->sectionBases.size());
168 for (InputSectionBase *s : cmd->sectionBases) {
169 MergeInputSection *ms = dyn_cast<MergeInputSection>(s);
171 cmd->sections.push_back(cast<InputSection>(s));
175 // We do not want to handle sections that are not alive, so just remove
176 // them instead of trying to merge.
180 auto i = llvm::find_if(mergeSections, [=](MergeSyntheticSection *sec) {
181 // While we could create a single synthetic section for two different
182 // values of Entsize, it is better to take Entsize into consideration.
184 // With a single synthetic section no two pieces with different Entsize
185 // could be equal, so we may as well have two sections.
187 // Using Entsize in here also allows us to propagate it to the synthetic
190 // SHF_STRINGS section with different alignments should not be merged.
191 return sec->flags == ms->flags && sec->entsize == ms->entsize &&
192 (sec->alignment == ms->alignment || !(sec->flags & SHF_STRINGS));
194 if (i == mergeSections.end()) {
195 MergeSyntheticSection *syn =
196 createMergeSynthetic(name, ms->type, ms->flags, ms->alignment);
197 mergeSections.push_back(syn);
198 i = std::prev(mergeSections.end());
199 syn->entsize = ms->entsize;
200 cmd->sections.push_back(syn);
202 (*i)->addSection(ms);
205 // sectionBases should not be used from this point onwards. Clear it to
207 cmd->sectionBases.clear();
209 // Some input sections may be removed from the list after ICF.
210 for (InputSection *s : cmd->sections)
213 for (auto *ms : mergeSections)
214 ms->finalizeContents();
217 static void sortByOrder(MutableArrayRef<InputSection *> in,
218 llvm::function_ref<int(InputSectionBase *s)> order) {
219 std::vector<std::pair<int, InputSection *>> v;
220 for (InputSection *s : in)
221 v.push_back({order(s), s});
222 llvm::stable_sort(v, less_first());
224 for (size_t i = 0; i < v.size(); ++i)
228 uint64_t elf::getHeaderSize() {
229 if (config->oFormatBinary)
231 return Out::elfHeader->size + Out::programHeaders->size;
234 bool OutputSection::classof(const BaseCommand *c) {
235 return c->kind == OutputSectionKind;
238 void OutputSection::sort(llvm::function_ref<int(InputSectionBase *s)> order) {
240 for (BaseCommand *b : sectionCommands)
241 if (auto *isd = dyn_cast<InputSectionDescription>(b))
242 sortByOrder(isd->sections, order);
245 static void nopInstrFill(uint8_t *buf, size_t size) {
251 std::vector<std::vector<uint8_t>> nopFiller = *target->nopInstrs;
252 unsigned num = size / nopFiller.back().size();
253 for (unsigned c = 0; c < num; ++c) {
254 memcpy(buf + i, nopFiller.back().data(), nopFiller.back().size());
255 i += nopFiller.back().size();
257 unsigned remaining = size - i;
260 assert(nopFiller[remaining - 1].size() == remaining);
261 memcpy(buf + i, nopFiller[remaining - 1].data(), remaining);
264 // Fill [Buf, Buf + Size) with Filler.
265 // This is used for linker script "=fillexp" command.
266 static void fill(uint8_t *buf, size_t size,
267 const std::array<uint8_t, 4> &filler) {
269 for (; i + 4 < size; i += 4)
270 memcpy(buf + i, filler.data(), 4);
271 memcpy(buf + i, filler.data(), size - i);
274 // Compress section contents if this section contains debug info.
275 template <class ELFT> void OutputSection::maybeCompress() {
276 using Elf_Chdr = typename ELFT::Chdr;
278 // Compress only DWARF debug sections.
279 if (!config->compressDebugSections || (flags & SHF_ALLOC) ||
280 !name.startswith(".debug_"))
283 // Create a section header.
284 zDebugHeader.resize(sizeof(Elf_Chdr));
285 auto *hdr = reinterpret_cast<Elf_Chdr *>(zDebugHeader.data());
286 hdr->ch_type = ELFCOMPRESS_ZLIB;
288 hdr->ch_addralign = alignment;
290 // Write section contents to a temporary buffer and compress it.
291 std::vector<uint8_t> buf(size);
292 writeTo<ELFT>(buf.data());
293 // We chose 1 as the default compression level because it is the fastest. If
294 // -O2 is given, we use level 6 to compress debug info more by ~15%. We found
295 // that level 7 to 9 doesn't make much difference (~1% more compression) while
296 // they take significant amount of time (~2x), so level 6 seems enough.
297 if (Error e = zlib::compress(toStringRef(buf), compressedData,
298 config->optimize >= 2 ? 6 : 1))
299 fatal("compress failed: " + llvm::toString(std::move(e)));
301 // Update section headers.
302 size = sizeof(Elf_Chdr) + compressedData.size();
303 flags |= SHF_COMPRESSED;
306 static void writeInt(uint8_t *buf, uint64_t data, uint64_t size) {
316 llvm_unreachable("unsupported Size argument");
319 template <class ELFT> void OutputSection::writeTo(uint8_t *buf) {
320 if (type == SHT_NOBITS)
323 // If -compress-debug-section is specified and if this is a debug section,
324 // we've already compressed section contents. If that's the case,
325 // just write it down.
326 if (!compressedData.empty()) {
327 memcpy(buf, zDebugHeader.data(), zDebugHeader.size());
328 memcpy(buf + zDebugHeader.size(), compressedData.data(),
329 compressedData.size());
333 // Write leading padding.
334 std::vector<InputSection *> sections = getInputSections(this);
335 std::array<uint8_t, 4> filler = getFiller();
336 bool nonZeroFiller = read32(filler.data()) != 0;
338 fill(buf, sections.empty() ? size : sections[0]->outSecOff, filler);
340 parallelForEachN(0, sections.size(), [&](size_t i) {
341 InputSection *isec = sections[i];
342 isec->writeTo<ELFT>(buf);
344 // Fill gaps between sections.
346 uint8_t *start = buf + isec->outSecOff + isec->getSize();
348 if (i + 1 == sections.size())
351 end = buf + sections[i + 1]->outSecOff;
352 if (isec->nopFiller) {
353 assert(target->nopInstrs);
354 nopInstrFill(start, end - start);
356 fill(start, end - start, filler);
360 // Linker scripts may have BYTE()-family commands with which you
361 // can write arbitrary bytes to the output. Process them if any.
362 for (BaseCommand *base : sectionCommands)
363 if (auto *data = dyn_cast<ByteCommand>(base))
364 writeInt(buf + data->offset, data->expression().getValue(), data->size);
367 static void finalizeShtGroup(OutputSection *os,
368 InputSection *section) {
369 assert(config->relocatable);
371 // sh_link field for SHT_GROUP sections should contain the section index of
373 os->link = in.symTab->getParent()->sectionIndex;
375 // sh_info then contain index of an entry in symbol table section which
376 // provides signature of the section group.
377 ArrayRef<Symbol *> symbols = section->file->getSymbols();
378 os->info = in.symTab->getSymbolIndex(symbols[section->info]);
381 void OutputSection::finalize() {
382 InputSection *first = getFirstInputSection(this);
384 if (flags & SHF_LINK_ORDER) {
385 // We must preserve the link order dependency of sections with the
386 // SHF_LINK_ORDER flag. The dependency is indicated by the sh_link field. We
387 // need to translate the InputSection sh_link to the OutputSection sh_link,
388 // all InputSections in the OutputSection have the same dependency.
389 if (auto *ex = dyn_cast<ARMExidxSyntheticSection>(first))
390 link = ex->getLinkOrderDep()->getParent()->sectionIndex;
391 else if (first->flags & SHF_LINK_ORDER)
392 if (auto *d = first->getLinkOrderDep())
393 link = d->getParent()->sectionIndex;
396 if (type == SHT_GROUP) {
397 finalizeShtGroup(this, first);
401 if (!config->copyRelocs || (type != SHT_RELA && type != SHT_REL))
404 if (isa<SyntheticSection>(first))
407 link = in.symTab->getParent()->sectionIndex;
408 // sh_info for SHT_REL[A] sections should contain the section header index of
409 // the section to which the relocation applies.
410 InputSectionBase *s = first->getRelocatedSection();
411 info = s->getOutputSection()->sectionIndex;
412 flags |= SHF_INFO_LINK;
415 // Returns true if S is in one of the many forms the compiler driver may pass
418 // Gcc uses any of crtbegin[<empty>|S|T].o.
419 // Clang uses Gcc's plus clang_rt.crtbegin[<empty>|S|T][-<arch>|<empty>].o.
421 static bool isCrtbegin(StringRef s) {
422 static std::regex re(R"((clang_rt\.)?crtbegin[ST]?(-.*)?\.o)");
423 s = sys::path::filename(s);
424 return std::regex_match(s.begin(), s.end(), re);
427 static bool isCrtend(StringRef s) {
428 static std::regex re(R"((clang_rt\.)?crtend[ST]?(-.*)?\.o)");
429 s = sys::path::filename(s);
430 return std::regex_match(s.begin(), s.end(), re);
433 // .ctors and .dtors are sorted by this priority from highest to lowest.
435 // 1. The section was contained in crtbegin (crtbegin contains
436 // some sentinel value in its .ctors and .dtors so that the runtime
437 // can find the beginning of the sections.)
439 // 2. The section has an optional priority value in the form of ".ctors.N"
440 // or ".dtors.N" where N is a number. Unlike .{init,fini}_array,
441 // they are compared as string rather than number.
443 // 3. The section is just ".ctors" or ".dtors".
445 // 4. The section was contained in crtend, which contains an end marker.
447 // In an ideal world, we don't need this function because .init_array and
448 // .ctors are duplicate features (and .init_array is newer.) However, there
449 // are too many real-world use cases of .ctors, so we had no choice to
450 // support that with this rather ad-hoc semantics.
451 static bool compCtors(const InputSection *a, const InputSection *b) {
452 bool beginA = isCrtbegin(a->file->getName());
453 bool beginB = isCrtbegin(b->file->getName());
454 if (beginA != beginB)
456 bool endA = isCrtend(a->file->getName());
457 bool endB = isCrtend(b->file->getName());
460 StringRef x = a->name;
461 StringRef y = b->name;
462 assert(x.startswith(".ctors") || x.startswith(".dtors"));
463 assert(y.startswith(".ctors") || y.startswith(".dtors"));
469 // Sorts input sections by the special rules for .ctors and .dtors.
470 // Unfortunately, the rules are different from the one for .{init,fini}_array.
471 // Read the comment above.
472 void OutputSection::sortCtorsDtors() {
473 assert(sectionCommands.size() == 1);
474 auto *isd = cast<InputSectionDescription>(sectionCommands[0]);
475 llvm::stable_sort(isd->sections, compCtors);
478 // If an input string is in the form of "foo.N" where N is a number,
479 // return N. Otherwise, returns 65536, which is one greater than the
481 int elf::getPriority(StringRef s) {
482 size_t pos = s.rfind('.');
483 if (pos == StringRef::npos)
486 if (!to_integer(s.substr(pos + 1), v, 10))
491 InputSection *elf::getFirstInputSection(const OutputSection *os) {
492 for (BaseCommand *base : os->sectionCommands)
493 if (auto *isd = dyn_cast<InputSectionDescription>(base))
494 if (!isd->sections.empty())
495 return isd->sections[0];
499 std::vector<InputSection *> elf::getInputSections(const OutputSection *os) {
500 std::vector<InputSection *> ret;
501 for (BaseCommand *base : os->sectionCommands)
502 if (auto *isd = dyn_cast<InputSectionDescription>(base))
503 ret.insert(ret.end(), isd->sections.begin(), isd->sections.end());
507 // Sorts input sections by section name suffixes, so that .foo.N comes
508 // before .foo.M if N < M. Used to sort .{init,fini}_array.N sections.
509 // We want to keep the original order if the priorities are the same
510 // because the compiler keeps the original initialization order in a
511 // translation unit and we need to respect that.
512 // For more detail, read the section of the GCC's manual about init_priority.
513 void OutputSection::sortInitFini() {
514 // Sort sections by priority.
515 sort([](InputSectionBase *s) { return getPriority(s->name); });
518 std::array<uint8_t, 4> OutputSection::getFiller() {
521 if (flags & SHF_EXECINSTR)
522 return target->trapInstr;
526 template void OutputSection::writeHeaderTo<ELF32LE>(ELF32LE::Shdr *Shdr);
527 template void OutputSection::writeHeaderTo<ELF32BE>(ELF32BE::Shdr *Shdr);
528 template void OutputSection::writeHeaderTo<ELF64LE>(ELF64LE::Shdr *Shdr);
529 template void OutputSection::writeHeaderTo<ELF64BE>(ELF64BE::Shdr *Shdr);
531 template void OutputSection::writeTo<ELF32LE>(uint8_t *Buf);
532 template void OutputSection::writeTo<ELF32BE>(uint8_t *Buf);
533 template void OutputSection::writeTo<ELF64LE>(uint8_t *Buf);
534 template void OutputSection::writeTo<ELF64BE>(uint8_t *Buf);
536 template void OutputSection::maybeCompress<ELF32LE>();
537 template void OutputSection::maybeCompress<ELF32BE>();
538 template void OutputSection::maybeCompress<ELF64LE>();
539 template void OutputSection::maybeCompress<ELF64BE>();