1 //===- LinkerScript.cpp ---------------------------------------------------===//
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file contains the parser/evaluator of the linker script.
12 //===----------------------------------------------------------------------===//
14 #include "LinkerScript.h"
16 #include "InputSection.h"
18 #include "OutputSections.h"
20 #include "SymbolTable.h"
22 #include "SyntheticSections.h"
26 #include "llvm/ADT/STLExtras.h"
27 #include "llvm/ADT/StringRef.h"
28 #include "llvm/Support/Casting.h"
29 #include "llvm/Support/ELF.h"
30 #include "llvm/Support/Endian.h"
31 #include "llvm/Support/ErrorHandling.h"
32 #include "llvm/Support/FileSystem.h"
33 #include "llvm/Support/Path.h"
44 using namespace llvm::ELF;
45 using namespace llvm::object;
46 using namespace llvm::support::endian;
48 using namespace lld::elf;
50 LinkerScript *elf::Script;
52 uint64_t ExprValue::getValue() const {
54 if (Sec->getOutputSection())
55 return Sec->getOffset(Val) + Sec->getOutputSection()->Addr;
56 error("unable to evaluate expression: input section " + Sec->Name +
57 " has no output section assigned");
62 uint64_t ExprValue::getSecAddr() const {
64 return Sec->getOffset(0) + Sec->getOutputSection()->Addr;
68 template <class ELFT> static SymbolBody *addRegular(SymbolAssignment *Cmd) {
70 uint8_t Visibility = Cmd->Hidden ? STV_HIDDEN : STV_DEFAULT;
71 std::tie(Sym, std::ignore) = Symtab<ELFT>::X->insert(
72 Cmd->Name, /*Type*/ 0, Visibility, /*CanOmitFromDynSym*/ false,
74 Sym->Binding = STB_GLOBAL;
75 ExprValue Value = Cmd->Expression();
76 SectionBase *Sec = Value.isAbsolute() ? nullptr : Value.Sec;
78 // We want to set symbol values early if we can. This allows us to use symbols
79 // as variables in linker scripts. Doing so allows us to write expressions
80 // like this: `alignment = 16; . = ALIGN(., alignment)`
81 uint64_t SymValue = Value.isAbsolute() ? Value.getValue() : 0;
82 replaceBody<DefinedRegular>(Sym, Cmd->Name, /*IsLocal=*/false, Visibility,
83 STT_NOTYPE, SymValue, 0, Sec, nullptr);
87 OutputSection *LinkerScript::getOutputSection(const Twine &Loc,
89 for (OutputSection *Sec : *OutputSections)
90 if (Sec->Name == Name)
93 static OutputSection Dummy("", 0, 0);
94 if (ErrorOnMissingSection)
95 error(Loc + ": undefined section " + Name);
99 // This function is essentially the same as getOutputSection(Name)->Size,
100 // but it won't print out an error message if a given section is not found.
102 // Linker script does not create an output section if its content is empty.
103 // We want to allow SIZEOF(.foo) where .foo is a section which happened to
104 // be empty. That is why this function is different from getOutputSection().
105 uint64_t LinkerScript::getOutputSectionSize(StringRef Name) {
106 for (OutputSection *Sec : *OutputSections)
107 if (Sec->Name == Name)
112 void LinkerScript::setDot(Expr E, const Twine &Loc, bool InSec) {
113 uint64_t Val = E().getValue();
116 error(Loc + ": unable to move location counter backward for: " +
119 error(Loc + ": unable to move location counter backward");
122 // Update to location counter means update to section size.
124 CurOutSec->Size = Dot - CurOutSec->Addr;
127 // Sets value of a symbol. Two kinds of symbols are processed: synthetic
128 // symbols, whose value is an offset from beginning of section and regular
129 // symbols whose value is absolute.
130 void LinkerScript::assignSymbol(SymbolAssignment *Cmd, bool InSec) {
131 if (Cmd->Name == ".") {
132 setDot(Cmd->Expression, Cmd->Location, InSec);
139 auto *Sym = cast<DefinedRegular>(Cmd->Sym);
140 ExprValue V = Cmd->Expression();
141 if (V.isAbsolute()) {
142 Sym->Value = V.getValue();
144 Sym->Section = V.Sec;
145 if (Sym->Section->Flags & SHF_ALLOC)
148 Sym->Value = V.getValue();
152 static SymbolBody *findSymbol(StringRef S) {
153 switch (Config->EKind) {
155 return Symtab<ELF32LE>::X->find(S);
157 return Symtab<ELF32BE>::X->find(S);
159 return Symtab<ELF64LE>::X->find(S);
161 return Symtab<ELF64BE>::X->find(S);
163 llvm_unreachable("unknown Config->EKind");
167 static SymbolBody *addRegularSymbol(SymbolAssignment *Cmd) {
168 switch (Config->EKind) {
170 return addRegular<ELF32LE>(Cmd);
172 return addRegular<ELF32BE>(Cmd);
174 return addRegular<ELF64LE>(Cmd);
176 return addRegular<ELF64BE>(Cmd);
178 llvm_unreachable("unknown Config->EKind");
182 void LinkerScript::addSymbol(SymbolAssignment *Cmd) {
183 if (Cmd->Name == ".")
186 // If a symbol was in PROVIDE(), we need to define it only when
187 // it is a referenced undefined symbol.
188 SymbolBody *B = findSymbol(Cmd->Name);
189 if (Cmd->Provide && (!B || B->isDefined()))
192 Cmd->Sym = addRegularSymbol(Cmd);
195 bool SymbolAssignment::classof(const BaseCommand *C) {
196 return C->Kind == AssignmentKind;
199 bool OutputSectionCommand::classof(const BaseCommand *C) {
200 return C->Kind == OutputSectionKind;
203 // Fill [Buf, Buf + Size) with Filler.
204 // This is used for linker script "=fillexp" command.
205 static void fill(uint8_t *Buf, size_t Size, uint32_t Filler) {
207 for (; I + 4 < Size; I += 4)
208 memcpy(Buf + I, &Filler, 4);
209 memcpy(Buf + I, &Filler, Size - I);
212 bool InputSectionDescription::classof(const BaseCommand *C) {
213 return C->Kind == InputSectionKind;
216 bool AssertCommand::classof(const BaseCommand *C) {
217 return C->Kind == AssertKind;
220 bool BytesDataCommand::classof(const BaseCommand *C) {
221 return C->Kind == BytesDataKind;
224 static StringRef basename(InputSectionBase *S) {
226 return sys::path::filename(S->File->getName());
230 bool LinkerScript::shouldKeep(InputSectionBase *S) {
231 for (InputSectionDescription *ID : Opt.KeptSections)
232 if (ID->FilePat.match(basename(S)))
233 for (SectionPattern &P : ID->SectionPatterns)
234 if (P.SectionPat.match(S->Name))
239 // A helper function for the SORT() command.
240 static std::function<bool(InputSectionBase *, InputSectionBase *)>
241 getComparator(SortSectionPolicy K) {
243 case SortSectionPolicy::Alignment:
244 return [](InputSectionBase *A, InputSectionBase *B) {
245 // ">" is not a mistake. Sections with larger alignments are placed
246 // before sections with smaller alignments in order to reduce the
247 // amount of padding necessary. This is compatible with GNU.
248 return A->Alignment > B->Alignment;
250 case SortSectionPolicy::Name:
251 return [](InputSectionBase *A, InputSectionBase *B) {
252 return A->Name < B->Name;
254 case SortSectionPolicy::Priority:
255 return [](InputSectionBase *A, InputSectionBase *B) {
256 return getPriority(A->Name) < getPriority(B->Name);
259 llvm_unreachable("unknown sort policy");
263 // A helper function for the SORT() command.
264 static bool matchConstraints(ArrayRef<InputSectionBase *> Sections,
265 ConstraintKind Kind) {
266 if (Kind == ConstraintKind::NoConstraint)
269 bool IsRW = llvm::any_of(Sections, [](InputSectionBase *Sec) {
270 return static_cast<InputSectionBase *>(Sec)->Flags & SHF_WRITE;
273 return (IsRW && Kind == ConstraintKind::ReadWrite) ||
274 (!IsRW && Kind == ConstraintKind::ReadOnly);
277 static void sortSections(InputSection **Begin, InputSection **End,
278 SortSectionPolicy K) {
279 if (K != SortSectionPolicy::Default && K != SortSectionPolicy::None)
280 std::stable_sort(Begin, End, getComparator(K));
283 // Compute and remember which sections the InputSectionDescription matches.
284 std::vector<InputSection *>
285 LinkerScript::computeInputSections(const InputSectionDescription *Cmd) {
286 std::vector<InputSection *> Ret;
288 // Collects all sections that satisfy constraints of Cmd.
289 for (const SectionPattern &Pat : Cmd->SectionPatterns) {
290 size_t SizeBefore = Ret.size();
292 for (InputSectionBase *Sec : InputSections) {
296 // For -emit-relocs we have to ignore entries like
297 // .rela.dyn : { *(.rela.data) }
298 // which are common because they are in the default bfd script.
299 if (Sec->Type == SHT_REL || Sec->Type == SHT_RELA)
302 StringRef Filename = basename(Sec);
303 if (!Cmd->FilePat.match(Filename) ||
304 Pat.ExcludedFilePat.match(Filename) ||
305 !Pat.SectionPat.match(Sec->Name))
308 Ret.push_back(cast<InputSection>(Sec));
309 Sec->Assigned = true;
312 // Sort sections as instructed by SORT-family commands and --sort-section
313 // option. Because SORT-family commands can be nested at most two depth
314 // (e.g. SORT_BY_NAME(SORT_BY_ALIGNMENT(.text.*))) and because the command
315 // line option is respected even if a SORT command is given, the exact
316 // behavior we have here is a bit complicated. Here are the rules.
318 // 1. If two SORT commands are given, --sort-section is ignored.
319 // 2. If one SORT command is given, and if it is not SORT_NONE,
320 // --sort-section is handled as an inner SORT command.
321 // 3. If one SORT command is given, and if it is SORT_NONE, don't sort.
322 // 4. If no SORT command is given, sort according to --sort-section.
323 InputSection **Begin = Ret.data() + SizeBefore;
324 InputSection **End = Ret.data() + Ret.size();
325 if (Pat.SortOuter != SortSectionPolicy::None) {
326 if (Pat.SortInner == SortSectionPolicy::Default)
327 sortSections(Begin, End, Config->SortSection);
329 sortSections(Begin, End, Pat.SortInner);
330 sortSections(Begin, End, Pat.SortOuter);
336 void LinkerScript::discard(ArrayRef<InputSectionBase *> V) {
337 for (InputSectionBase *S : V) {
339 if (S == InX::ShStrTab)
340 error("discarding .shstrtab section is not allowed");
341 discard(S->DependentSections);
345 std::vector<InputSectionBase *>
346 LinkerScript::createInputSectionList(OutputSectionCommand &OutCmd) {
347 std::vector<InputSectionBase *> Ret;
349 for (BaseCommand *Base : OutCmd.Commands) {
350 auto *Cmd = dyn_cast<InputSectionDescription>(Base);
354 Cmd->Sections = computeInputSections(Cmd);
355 Ret.insert(Ret.end(), Cmd->Sections.begin(), Cmd->Sections.end());
361 void LinkerScript::processCommands(OutputSectionFactory &Factory) {
362 // A symbol can be assigned before any section is mentioned in the linker
363 // script. In an DSO, the symbol values are addresses, so the only important
364 // section values are:
367 // * Any value meaning a regular section.
368 // To handle that, create a dummy aether section that fills the void before
369 // the linker scripts switches to another section. It has an index of one
370 // which will map to whatever the first actual section is.
371 Aether = make<OutputSection>("", 0, SHF_ALLOC);
372 Aether->SectionIndex = 1;
376 for (size_t I = 0; I < Opt.Commands.size(); ++I) {
377 // Handle symbol assignments outside of any output section.
378 if (auto *Cmd = dyn_cast<SymbolAssignment>(Opt.Commands[I])) {
383 if (auto *Cmd = dyn_cast<OutputSectionCommand>(Opt.Commands[I])) {
384 std::vector<InputSectionBase *> V = createInputSectionList(*Cmd);
386 // The output section name `/DISCARD/' is special.
387 // Any input section assigned to it is discarded.
388 if (Cmd->Name == "/DISCARD/") {
393 // This is for ONLY_IF_RO and ONLY_IF_RW. An output section directive
394 // ".foo : ONLY_IF_R[OW] { ... }" is handled only if all member input
395 // sections satisfy a given constraint. If not, a directive is handled
396 // as if it wasn't present from the beginning.
398 // Because we'll iterate over Commands many more times, the easiest
399 // way to "make it as if it wasn't present" is to just remove it.
400 if (!matchConstraints(V, Cmd->Constraint)) {
401 for (InputSectionBase *S : V)
403 Opt.Commands.erase(Opt.Commands.begin() + I);
408 // A directive may contain symbol definitions like this:
409 // ".foo : { ...; bar = .; }". Handle them.
410 for (BaseCommand *Base : Cmd->Commands)
411 if (auto *OutCmd = dyn_cast<SymbolAssignment>(Base))
414 // Handle subalign (e.g. ".foo : SUBALIGN(32) { ... }"). If subalign
415 // is given, input sections are aligned to that value, whether the
416 // given value is larger or smaller than the original section alignment.
417 if (Cmd->SubalignExpr) {
418 uint32_t Subalign = Cmd->SubalignExpr().getValue();
419 for (InputSectionBase *S : V)
420 S->Alignment = Subalign;
423 // Add input sections to an output section.
424 for (InputSectionBase *S : V)
425 Factory.addInputSec(S, Cmd->Name, Cmd->Sec);
426 if (OutputSection *Sec = Cmd->Sec) {
427 assert(Sec->SectionIndex == INT_MAX);
428 Sec->SectionIndex = I;
429 SecToCommand[Sec] = Cmd;
436 void LinkerScript::fabricateDefaultCommands() {
437 std::vector<BaseCommand *> Commands;
439 // Define start address
440 uint64_t StartAddr = Config->ImageBase + elf::getHeaderSize();
442 // The Sections with -T<section> have been sorted in order of ascending
443 // address. We must lower StartAddr if the lowest -T<section address> as
444 // calls to setDot() must be monotonically increasing.
445 for (auto& KV : Config->SectionStartMap)
446 StartAddr = std::min(StartAddr, KV.second);
449 make<SymbolAssignment>(".", [=] { return StartAddr; }, ""));
451 // For each OutputSection that needs a VA fabricate an OutputSectionCommand
452 // with an InputSectionDescription describing the InputSections
453 for (OutputSection *Sec : *OutputSections) {
454 auto *OSCmd = make<OutputSectionCommand>(Sec->Name);
456 SecToCommand[Sec] = OSCmd;
458 // Prefer user supplied address over additional alignment constraint
459 auto I = Config->SectionStartMap.find(Sec->Name);
460 if (I != Config->SectionStartMap.end())
462 make<SymbolAssignment>(".", [=] { return I->second; }, ""));
463 else if (Sec->PageAlign)
464 OSCmd->AddrExpr = [=] {
465 return alignTo(Script->getDot(), Config->MaxPageSize);
468 Commands.push_back(OSCmd);
469 if (Sec->Sections.size()) {
470 auto *ISD = make<InputSectionDescription>("");
471 OSCmd->Commands.push_back(ISD);
472 for (InputSection *ISec : Sec->Sections) {
473 ISD->Sections.push_back(ISec);
474 ISec->Assigned = true;
478 // SECTIONS commands run before other non SECTIONS commands
479 Commands.insert(Commands.end(), Opt.Commands.begin(), Opt.Commands.end());
480 Opt.Commands = std::move(Commands);
483 // Add sections that didn't match any sections command.
484 void LinkerScript::addOrphanSections(OutputSectionFactory &Factory) {
485 for (InputSectionBase *S : InputSections) {
486 if (!S->Live || S->OutSec)
488 StringRef Name = getOutputSectionName(S->Name);
489 auto I = std::find_if(
490 Opt.Commands.begin(), Opt.Commands.end(), [&](BaseCommand *Base) {
491 if (auto *Cmd = dyn_cast<OutputSectionCommand>(Base))
492 return Cmd->Name == Name;
495 if (I == Opt.Commands.end()) {
496 Factory.addInputSec(S, Name);
498 auto *Cmd = cast<OutputSectionCommand>(*I);
499 Factory.addInputSec(S, Name, Cmd->Sec);
500 if (OutputSection *Sec = Cmd->Sec) {
501 SecToCommand[Sec] = Cmd;
502 unsigned Index = std::distance(Opt.Commands.begin(), I);
503 assert(Sec->SectionIndex == INT_MAX || Sec->SectionIndex == Index);
504 Sec->SectionIndex = Index;
506 auto *ISD = make<InputSectionDescription>("");
507 ISD->Sections.push_back(cast<InputSection>(S));
508 Cmd->Commands.push_back(ISD);
513 uint64_t LinkerScript::advance(uint64_t Size, unsigned Align) {
514 bool IsTbss = (CurOutSec->Flags & SHF_TLS) && CurOutSec->Type == SHT_NOBITS;
515 uint64_t Start = IsTbss ? Dot + ThreadBssOffset : Dot;
516 Start = alignTo(Start, Align);
517 uint64_t End = Start + Size;
520 ThreadBssOffset = End - Dot;
526 void LinkerScript::output(InputSection *S) {
527 uint64_t Pos = advance(S->getSize(), S->Alignment);
528 S->OutSecOff = Pos - S->getSize() - CurOutSec->Addr;
530 // Update output section size after adding each section. This is so that
531 // SIZEOF works correctly in the case below:
532 // .foo { *(.aaa) a = SIZEOF(.foo); *(.bbb) }
533 CurOutSec->Size = Pos - CurOutSec->Addr;
535 // If there is a memory region associated with this input section, then
536 // place the section in that region and update the region index.
538 CurMemRegion->Offset += CurOutSec->Size;
539 uint64_t CurSize = CurMemRegion->Offset - CurMemRegion->Origin;
540 if (CurSize > CurMemRegion->Length) {
541 uint64_t OverflowAmt = CurSize - CurMemRegion->Length;
542 error("section '" + CurOutSec->Name + "' will not fit in region '" +
543 CurMemRegion->Name + "': overflowed by " + Twine(OverflowAmt) +
549 void LinkerScript::switchTo(OutputSection *Sec) {
550 if (CurOutSec == Sec)
554 CurOutSec->Addr = advance(0, CurOutSec->Alignment);
556 // If neither AT nor AT> is specified for an allocatable section, the linker
557 // will set the LMA such that the difference between VMA and LMA for the
558 // section is the same as the preceding output section in the same region
559 // https://sourceware.org/binutils/docs-2.20/ld/Output-Section-LMA.html
561 CurOutSec->LMAOffset = LMAOffset();
564 void LinkerScript::process(BaseCommand &Base) {
565 // This handles the assignments to symbol or to the dot.
566 if (auto *Cmd = dyn_cast<SymbolAssignment>(&Base)) {
567 assignSymbol(Cmd, true);
571 // Handle BYTE(), SHORT(), LONG(), or QUAD().
572 if (auto *Cmd = dyn_cast<BytesDataCommand>(&Base)) {
573 Cmd->Offset = Dot - CurOutSec->Addr;
575 CurOutSec->Size = Dot - CurOutSec->Addr;
580 if (auto *Cmd = dyn_cast<AssertCommand>(&Base)) {
585 // Handle a single input section description command.
586 // It calculates and assigns the offsets for each section and also
587 // updates the output section size.
588 auto &Cmd = cast<InputSectionDescription>(Base);
589 for (InputSectionBase *Sec : Cmd.Sections) {
590 // We tentatively added all synthetic sections at the beginning and removed
591 // empty ones afterwards (because there is no way to know whether they were
592 // going be empty or not other than actually running linker scripts.)
593 // We need to ignore remains of empty sections.
594 if (auto *S = dyn_cast<SyntheticSection>(Sec))
600 assert(CurOutSec == Sec->OutSec);
601 output(cast<InputSection>(Sec));
605 // This function searches for a memory region to place the given output
606 // section in. If found, a pointer to the appropriate memory region is
607 // returned. Otherwise, a nullptr is returned.
608 MemoryRegion *LinkerScript::findMemoryRegion(OutputSectionCommand *Cmd) {
609 // If a memory region name was specified in the output section command,
610 // then try to find that region first.
611 if (!Cmd->MemoryRegionName.empty()) {
612 auto It = Opt.MemoryRegions.find(Cmd->MemoryRegionName);
613 if (It != Opt.MemoryRegions.end())
615 error("memory region '" + Cmd->MemoryRegionName + "' not declared");
619 // If at least one memory region is defined, all sections must
620 // belong to some memory region. Otherwise, we don't need to do
621 // anything for memory regions.
622 if (Opt.MemoryRegions.empty())
625 OutputSection *Sec = Cmd->Sec;
626 // See if a region can be found by matching section flags.
627 for (auto &Pair : Opt.MemoryRegions) {
628 MemoryRegion &M = Pair.second;
629 if ((M.Flags & Sec->Flags) && (M.NegFlags & Sec->Flags) == 0)
633 // Otherwise, no suitable region was found.
634 if (Sec->Flags & SHF_ALLOC)
635 error("no memory region specified for section '" + Sec->Name + "'");
639 // This function assigns offsets to input sections and an output section
640 // for a single sections command (e.g. ".text { *(.text); }").
641 void LinkerScript::assignOffsets(OutputSectionCommand *Cmd) {
642 OutputSection *Sec = Cmd->Sec;
646 if (Cmd->AddrExpr && (Sec->Flags & SHF_ALLOC))
647 setDot(Cmd->AddrExpr, Cmd->Location, false);
651 LMAOffset = [=] { return Cmd->LMAExpr().getValue() - D; };
654 CurMemRegion = Cmd->MemRegion;
656 Dot = CurMemRegion->Offset;
659 // We do not support custom layout for compressed debug sectons.
660 // At this point we already know their size and have compressed content.
661 if (CurOutSec->Flags & SHF_COMPRESSED)
664 for (BaseCommand *C : Cmd->Commands)
668 void LinkerScript::removeEmptyCommands() {
669 // It is common practice to use very generic linker scripts. So for any
670 // given run some of the output sections in the script will be empty.
671 // We could create corresponding empty output sections, but that would
672 // clutter the output.
673 // We instead remove trivially empty sections. The bfd linker seems even
674 // more aggressive at removing them.
675 auto Pos = std::remove_if(
676 Opt.Commands.begin(), Opt.Commands.end(), [&](BaseCommand *Base) {
677 if (auto *Cmd = dyn_cast<OutputSectionCommand>(Base))
678 return std::find(OutputSections->begin(), OutputSections->end(),
679 Cmd->Sec) == OutputSections->end();
682 Opt.Commands.erase(Pos, Opt.Commands.end());
685 static bool isAllSectionDescription(const OutputSectionCommand &Cmd) {
686 for (BaseCommand *Base : Cmd.Commands)
687 if (!isa<InputSectionDescription>(*Base))
692 void LinkerScript::adjustSectionsBeforeSorting() {
693 // If the output section contains only symbol assignments, create a
694 // corresponding output section. The bfd linker seems to only create them if
695 // '.' is assigned to, but creating these section should not have any bad
696 // consequeces and gives us a section to put the symbol in.
697 uint64_t Flags = SHF_ALLOC;
699 for (int I = 0, E = Opt.Commands.size(); I != E; ++I) {
700 auto *Cmd = dyn_cast<OutputSectionCommand>(Opt.Commands[I]);
703 if (OutputSection *Sec = Cmd->Sec) {
708 if (isAllSectionDescription(*Cmd))
711 auto *OutSec = make<OutputSection>(Cmd->Name, SHT_PROGBITS, Flags);
712 OutSec->SectionIndex = I;
713 OutputSections->push_back(OutSec);
715 SecToCommand[OutSec] = Cmd;
719 void LinkerScript::adjustSectionsAfterSorting() {
720 placeOrphanSections();
722 // Try and find an appropriate memory region to assign offsets in.
723 for (BaseCommand *Base : Opt.Commands) {
724 if (auto *Cmd = dyn_cast<OutputSectionCommand>(Base)) {
725 Cmd->MemRegion = findMemoryRegion(Cmd);
726 // Handle align (e.g. ".foo : ALIGN(16) { ... }").
728 Cmd->Sec->updateAlignment(Cmd->AlignExpr().getValue());
732 // If output section command doesn't specify any segments,
733 // and we haven't previously assigned any section to segment,
734 // then we simply assign section to the very first load segment.
735 // Below is an example of such linker script:
736 // PHDRS { seg PT_LOAD; }
737 // SECTIONS { .aaa : { *(.aaa) } }
738 std::vector<StringRef> DefPhdrs;
740 std::find_if(Opt.PhdrsCommands.begin(), Opt.PhdrsCommands.end(),
741 [](const PhdrsCommand &Cmd) { return Cmd.Type == PT_LOAD; });
742 if (FirstPtLoad != Opt.PhdrsCommands.end())
743 DefPhdrs.push_back(FirstPtLoad->Name);
745 // Walk the commands and propagate the program headers to commands that don't
746 // explicitly specify them.
747 for (BaseCommand *Base : Opt.Commands) {
748 auto *Cmd = dyn_cast<OutputSectionCommand>(Base);
752 if (Cmd->Phdrs.empty())
753 Cmd->Phdrs = DefPhdrs;
755 DefPhdrs = Cmd->Phdrs;
758 removeEmptyCommands();
761 // When placing orphan sections, we want to place them after symbol assignments
762 // so that an orphan after
766 // doesn't break the intended meaning of the begin/end symbols.
767 // We don't want to go over sections since Writer<ELFT>::sortSections is the
768 // one in charge of deciding the order of the sections.
769 // We don't want to go over alignments, since doing so in
770 // rx_sec : { *(rx_sec) }
771 // . = ALIGN(0x1000);
772 // /* The RW PT_LOAD starts here*/
773 // rw_sec : { *(rw_sec) }
774 // would mean that the RW PT_LOAD would become unaligned.
775 static bool shouldSkip(BaseCommand *Cmd) {
776 if (isa<OutputSectionCommand>(Cmd))
778 if (auto *Assign = dyn_cast<SymbolAssignment>(Cmd))
779 return Assign->Name != ".";
783 // Orphan sections are sections present in the input files which are
784 // not explicitly placed into the output file by the linker script.
786 // When the control reaches this function, Opt.Commands contains
787 // output section commands for non-orphan sections only. This function
788 // adds new elements for orphan sections so that all sections are
789 // explicitly handled by Opt.Commands.
791 // Writer<ELFT>::sortSections has already sorted output sections.
792 // What we need to do is to scan OutputSections vector and
793 // Opt.Commands in parallel to find orphan sections. If there is an
794 // output section that doesn't have a corresponding entry in
795 // Opt.Commands, we will insert a new entry to Opt.Commands.
797 // There is some ambiguity as to where exactly a new entry should be
798 // inserted, because Opt.Commands contains not only output section
799 // commands but also other types of commands such as symbol assignment
800 // expressions. There's no correct answer here due to the lack of the
801 // formal specification of the linker script. We use heuristics to
802 // determine whether a new output command should be added before or
803 // after another commands. For the details, look at shouldSkip
805 void LinkerScript::placeOrphanSections() {
806 // The OutputSections are already in the correct order.
807 // This loops creates or moves commands as needed so that they are in the
811 // As a horrible special case, skip the first . assignment if it is before any
812 // section. We do this because it is common to set a load address by starting
813 // the script with ". = 0xabcd" and the expectation is that every section is
815 auto FirstSectionOrDotAssignment =
816 std::find_if(Opt.Commands.begin(), Opt.Commands.end(),
817 [](BaseCommand *Cmd) { return !shouldSkip(Cmd); });
818 if (FirstSectionOrDotAssignment != Opt.Commands.end()) {
819 CmdIndex = FirstSectionOrDotAssignment - Opt.Commands.begin();
820 if (isa<SymbolAssignment>(**FirstSectionOrDotAssignment))
824 for (OutputSection *Sec : *OutputSections) {
825 StringRef Name = Sec->Name;
827 // Find the last spot where we can insert a command and still get the
829 auto CmdIter = Opt.Commands.begin() + CmdIndex;
830 auto E = Opt.Commands.end();
831 while (CmdIter != E && shouldSkip(*CmdIter)) {
836 // If there is no command corresponding to this output section,
837 // create one and put a InputSectionDescription in it so that both
838 // representations agree on which input sections to use.
839 OutputSectionCommand *Cmd = getCmd(Sec);
841 Cmd = make<OutputSectionCommand>(Name);
842 Opt.Commands.insert(CmdIter, Cmd);
846 SecToCommand[Sec] = Cmd;
847 auto *ISD = make<InputSectionDescription>("");
848 for (InputSection *IS : Sec->Sections)
849 ISD->Sections.push_back(IS);
850 Cmd->Commands.push_back(ISD);
855 // Continue from where we found it.
856 while (*CmdIter != Cmd) {
864 void LinkerScript::processNonSectionCommands() {
865 for (BaseCommand *Base : Opt.Commands) {
866 if (auto *Cmd = dyn_cast<SymbolAssignment>(Base))
867 assignSymbol(Cmd, false);
868 else if (auto *Cmd = dyn_cast<AssertCommand>(Base))
873 // Do a last effort at synchronizing the linker script "AST" and the section
874 // list. This is needed to account for last minute changes, like adding a
875 // .ARM.exidx terminator and sorting SHF_LINK_ORDER sections.
877 // FIXME: We should instead create the "AST" earlier and the above changes would
878 // be done directly in the "AST".
880 // This can only handle new sections being added and sections being reordered.
881 void LinkerScript::synchronize() {
882 for (BaseCommand *Base : Opt.Commands) {
883 auto *Cmd = dyn_cast<OutputSectionCommand>(Base);
886 ArrayRef<InputSection *> Sections = Cmd->Sec->Sections;
887 std::vector<InputSection **> ScriptSections;
888 DenseSet<InputSection *> ScriptSectionsSet;
889 for (BaseCommand *Base : Cmd->Commands) {
890 auto *ISD = dyn_cast<InputSectionDescription>(Base);
893 for (InputSection *&IS : ISD->Sections) {
895 ScriptSections.push_back(&IS);
896 ScriptSectionsSet.insert(IS);
900 std::vector<InputSection *> Missing;
901 for (InputSection *IS : Sections)
902 if (!ScriptSectionsSet.count(IS))
903 Missing.push_back(IS);
904 if (!Missing.empty()) {
905 auto ISD = make<InputSectionDescription>("");
906 ISD->Sections = Missing;
907 Cmd->Commands.push_back(ISD);
908 for (InputSection *&IS : ISD->Sections)
910 ScriptSections.push_back(&IS);
912 assert(ScriptSections.size() == Sections.size());
913 for (int I = 0, N = Sections.size(); I < N; ++I)
914 *ScriptSections[I] = Sections[I];
918 static bool allocateHeaders(std::vector<PhdrEntry> &Phdrs,
919 ArrayRef<OutputSection *> OutputSections,
922 std::find_if(Phdrs.begin(), Phdrs.end(),
923 [](const PhdrEntry &E) { return E.p_type == PT_LOAD; });
924 if (FirstPTLoad == Phdrs.end())
927 uint64_t HeaderSize = getHeaderSize();
928 if (HeaderSize <= Min || Script->hasPhdrsCommands()) {
929 Min = alignDown(Min - HeaderSize, Config->MaxPageSize);
930 Out::ElfHeader->Addr = Min;
931 Out::ProgramHeaders->Addr = Min + Out::ElfHeader->Size;
935 assert(FirstPTLoad->First == Out::ElfHeader);
936 OutputSection *ActualFirst = nullptr;
937 for (OutputSection *Sec : OutputSections) {
938 if (Sec->FirstInPtLoad == Out::ElfHeader) {
944 for (OutputSection *Sec : OutputSections)
945 if (Sec->FirstInPtLoad == Out::ElfHeader)
946 Sec->FirstInPtLoad = ActualFirst;
947 FirstPTLoad->First = ActualFirst;
949 Phdrs.erase(FirstPTLoad);
952 auto PhdrI = std::find_if(Phdrs.begin(), Phdrs.end(), [](const PhdrEntry &E) {
953 return E.p_type == PT_PHDR;
955 if (PhdrI != Phdrs.end())
960 void LinkerScript::assignAddresses(std::vector<PhdrEntry> &Phdrs) {
961 // Assign addresses as instructed by linker script SECTIONS sub-commands.
963 ErrorOnMissingSection = true;
966 for (BaseCommand *Base : Opt.Commands) {
967 if (auto *Cmd = dyn_cast<SymbolAssignment>(Base)) {
968 assignSymbol(Cmd, false);
972 if (auto *Cmd = dyn_cast<AssertCommand>(Base)) {
977 auto *Cmd = cast<OutputSectionCommand>(Base);
981 uint64_t MinVA = std::numeric_limits<uint64_t>::max();
982 for (OutputSection *Sec : *OutputSections) {
983 if (Sec->Flags & SHF_ALLOC)
984 MinVA = std::min<uint64_t>(MinVA, Sec->Addr);
989 allocateHeaders(Phdrs, *OutputSections, MinVA);
992 // Creates program headers as instructed by PHDRS linker script command.
993 std::vector<PhdrEntry> LinkerScript::createPhdrs() {
994 std::vector<PhdrEntry> Ret;
996 // Process PHDRS and FILEHDR keywords because they are not
997 // real output sections and cannot be added in the following loop.
998 for (const PhdrsCommand &Cmd : Opt.PhdrsCommands) {
999 Ret.emplace_back(Cmd.Type, Cmd.Flags == UINT_MAX ? PF_R : Cmd.Flags);
1000 PhdrEntry &Phdr = Ret.back();
1003 Phdr.add(Out::ElfHeader);
1005 Phdr.add(Out::ProgramHeaders);
1008 Phdr.p_paddr = Cmd.LMAExpr().getValue();
1013 // Add output sections to program headers.
1014 for (OutputSection *Sec : *OutputSections) {
1015 if (!(Sec->Flags & SHF_ALLOC))
1018 // Assign headers specified by linker script
1019 for (size_t Id : getPhdrIndices(Sec)) {
1021 if (Opt.PhdrsCommands[Id].Flags == UINT_MAX)
1022 Ret[Id].p_flags |= Sec->getPhdrFlags();
1028 bool LinkerScript::ignoreInterpSection() {
1029 // Ignore .interp section in case we have PHDRS specification
1030 // and PT_INTERP isn't listed.
1031 if (Opt.PhdrsCommands.empty())
1033 for (PhdrsCommand &Cmd : Opt.PhdrsCommands)
1034 if (Cmd.Type == PT_INTERP)
1039 OutputSectionCommand *LinkerScript::getCmd(OutputSection *Sec) const {
1040 auto I = SecToCommand.find(Sec);
1041 if (I == SecToCommand.end())
1046 uint32_t OutputSectionCommand::getFiller() {
1049 if (Sec->Flags & SHF_EXECINSTR)
1050 return Target->TrapInstr;
1054 static void writeInt(uint8_t *Buf, uint64_t Data, uint64_t Size) {
1058 write16(Buf, Data, Config->Endianness);
1060 write32(Buf, Data, Config->Endianness);
1062 write64(Buf, Data, Config->Endianness);
1064 llvm_unreachable("unsupported Size argument");
1067 template <class ELFT> void OutputSectionCommand::writeTo(uint8_t *Buf) {
1070 // We may have already rendered compressed content when using
1071 // -compress-debug-sections option. Write it together with header.
1072 if (!Sec->CompressedData.empty()) {
1073 memcpy(Buf, Sec->ZDebugHeader.data(), Sec->ZDebugHeader.size());
1074 memcpy(Buf + Sec->ZDebugHeader.size(), Sec->CompressedData.data(),
1075 Sec->CompressedData.size());
1079 // Write leading padding.
1080 ArrayRef<InputSection *> Sections = Sec->Sections;
1081 uint32_t Filler = getFiller();
1083 fill(Buf, Sections.empty() ? Sec->Size : Sections[0]->OutSecOff, Filler);
1085 parallelForEachN(0, Sections.size(), [=](size_t I) {
1086 InputSection *IS = Sections[I];
1087 IS->writeTo<ELFT>(Buf);
1089 // Fill gaps between sections.
1091 uint8_t *Start = Buf + IS->OutSecOff + IS->getSize();
1093 if (I + 1 == Sections.size())
1094 End = Buf + Sec->Size;
1096 End = Buf + Sections[I + 1]->OutSecOff;
1097 fill(Start, End - Start, Filler);
1101 // Linker scripts may have BYTE()-family commands with which you
1102 // can write arbitrary bytes to the output. Process them if any.
1103 for (BaseCommand *Base : Commands)
1104 if (auto *Data = dyn_cast<BytesDataCommand>(Base))
1105 writeInt(Buf + Data->Offset, Data->Expression().getValue(), Data->Size);
1108 bool LinkerScript::hasLMA(OutputSection *Sec) {
1109 if (OutputSectionCommand *Cmd = getCmd(Sec))
1115 ExprValue LinkerScript::getSymbolValue(const Twine &Loc, StringRef S) {
1117 return {CurOutSec, Dot - CurOutSec->Addr};
1118 if (SymbolBody *B = findSymbol(S)) {
1119 if (auto *D = dyn_cast<DefinedRegular>(B))
1120 return {D->Section, D->Value};
1121 if (auto *C = dyn_cast<DefinedCommon>(B))
1122 return {InX::Common, C->Offset};
1124 error(Loc + ": symbol not found: " + S);
1128 bool LinkerScript::isDefined(StringRef S) { return findSymbol(S) != nullptr; }
1130 // Returns indices of ELF headers containing specific section. Each index is a
1131 // zero based number of ELF header listed within PHDRS {} script block.
1132 std::vector<size_t> LinkerScript::getPhdrIndices(OutputSection *Sec) {
1133 if (OutputSectionCommand *Cmd = getCmd(Sec)) {
1134 std::vector<size_t> Ret;
1135 for (StringRef PhdrName : Cmd->Phdrs)
1136 Ret.push_back(getPhdrIndex(Cmd->Location, PhdrName));
1142 size_t LinkerScript::getPhdrIndex(const Twine &Loc, StringRef PhdrName) {
1144 for (PhdrsCommand &Cmd : Opt.PhdrsCommands) {
1145 if (Cmd.Name == PhdrName)
1149 error(Loc + ": section header '" + PhdrName + "' is not listed in PHDRS");
1153 template void OutputSectionCommand::writeTo<ELF32LE>(uint8_t *Buf);
1154 template void OutputSectionCommand::writeTo<ELF32BE>(uint8_t *Buf);
1155 template void OutputSectionCommand::writeTo<ELF64LE>(uint8_t *Buf);
1156 template void OutputSectionCommand::writeTo<ELF64BE>(uint8_t *Buf);