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"
24 #include "llvm/ADT/STLExtras.h"
25 #include "llvm/ADT/StringRef.h"
26 #include "llvm/Support/Casting.h"
27 #include "llvm/Support/ELF.h"
28 #include "llvm/Support/Endian.h"
29 #include "llvm/Support/ErrorHandling.h"
30 #include "llvm/Support/FileSystem.h"
31 #include "llvm/Support/Path.h"
42 using namespace llvm::ELF;
43 using namespace llvm::object;
44 using namespace llvm::support::endian;
46 using namespace lld::elf;
48 LinkerScript *elf::Script;
50 uint64_t ExprValue::getValue() const {
52 return Sec->getOffset(Val) + Sec->getOutputSection()->Addr;
56 uint64_t ExprValue::getSecAddr() const {
58 return Sec->getOffset(0) + Sec->getOutputSection()->Addr;
62 template <class ELFT> static SymbolBody *addRegular(SymbolAssignment *Cmd) {
64 uint8_t Visibility = Cmd->Hidden ? STV_HIDDEN : STV_DEFAULT;
65 std::tie(Sym, std::ignore) = Symtab<ELFT>::X->insert(
66 Cmd->Name, /*Type*/ 0, Visibility, /*CanOmitFromDynSym*/ false,
68 Sym->Binding = STB_GLOBAL;
69 ExprValue Value = Cmd->Expression();
70 SectionBase *Sec = Value.isAbsolute() ? nullptr : Value.Sec;
72 // We want to set symbol values early if we can. This allows us to use symbols
73 // as variables in linker scripts. Doing so allows us to write expressions
74 // like this: `alignment = 16; . = ALIGN(., alignment)`
75 uint64_t SymValue = Value.isAbsolute() ? Value.getValue() : 0;
76 replaceBody<DefinedRegular>(Sym, Cmd->Name, /*IsLocal=*/false, Visibility,
77 STT_NOTYPE, SymValue, 0, Sec, nullptr);
81 OutputSection *LinkerScript::getOutputSection(const Twine &Loc,
83 for (OutputSection *Sec : *OutputSections)
84 if (Sec->Name == Name)
87 static OutputSection Dummy("", 0, 0);
88 if (ErrorOnMissingSection)
89 error(Loc + ": undefined section " + Name);
93 // This function is essentially the same as getOutputSection(Name)->Size,
94 // but it won't print out an error message if a given section is not found.
96 // Linker script does not create an output section if its content is empty.
97 // We want to allow SIZEOF(.foo) where .foo is a section which happened to
98 // be empty. That is why this function is different from getOutputSection().
99 uint64_t LinkerScript::getOutputSectionSize(StringRef Name) {
100 for (OutputSection *Sec : *OutputSections)
101 if (Sec->Name == Name)
106 void LinkerScript::setDot(Expr E, const Twine &Loc, bool InSec) {
107 uint64_t Val = E().getValue();
110 error(Loc + ": unable to move location counter backward for: " +
113 error(Loc + ": unable to move location counter backward");
116 // Update to location counter means update to section size.
118 CurOutSec->Size = Dot - CurOutSec->Addr;
121 // Sets value of a symbol. Two kinds of symbols are processed: synthetic
122 // symbols, whose value is an offset from beginning of section and regular
123 // symbols whose value is absolute.
124 void LinkerScript::assignSymbol(SymbolAssignment *Cmd, bool InSec) {
125 if (Cmd->Name == ".") {
126 setDot(Cmd->Expression, Cmd->Location, InSec);
133 auto *Sym = cast<DefinedRegular>(Cmd->Sym);
134 ExprValue V = Cmd->Expression();
135 if (V.isAbsolute()) {
136 Sym->Value = V.getValue();
138 Sym->Section = V.Sec;
139 if (Sym->Section->Flags & SHF_ALLOC)
142 Sym->Value = V.getValue();
146 static SymbolBody *findSymbol(StringRef S) {
147 switch (Config->EKind) {
149 return Symtab<ELF32LE>::X->find(S);
151 return Symtab<ELF32BE>::X->find(S);
153 return Symtab<ELF64LE>::X->find(S);
155 return Symtab<ELF64BE>::X->find(S);
157 llvm_unreachable("unknown Config->EKind");
161 static SymbolBody *addRegularSymbol(SymbolAssignment *Cmd) {
162 switch (Config->EKind) {
164 return addRegular<ELF32LE>(Cmd);
166 return addRegular<ELF32BE>(Cmd);
168 return addRegular<ELF64LE>(Cmd);
170 return addRegular<ELF64BE>(Cmd);
172 llvm_unreachable("unknown Config->EKind");
176 void LinkerScript::addSymbol(SymbolAssignment *Cmd) {
177 if (Cmd->Name == ".")
180 // If a symbol was in PROVIDE(), we need to define it only when
181 // it is a referenced undefined symbol.
182 SymbolBody *B = findSymbol(Cmd->Name);
183 if (Cmd->Provide && (!B || B->isDefined()))
186 Cmd->Sym = addRegularSymbol(Cmd);
189 bool SymbolAssignment::classof(const BaseCommand *C) {
190 return C->Kind == AssignmentKind;
193 bool OutputSectionCommand::classof(const BaseCommand *C) {
194 return C->Kind == OutputSectionKind;
197 bool InputSectionDescription::classof(const BaseCommand *C) {
198 return C->Kind == InputSectionKind;
201 bool AssertCommand::classof(const BaseCommand *C) {
202 return C->Kind == AssertKind;
205 bool BytesDataCommand::classof(const BaseCommand *C) {
206 return C->Kind == BytesDataKind;
209 static StringRef basename(InputSectionBase *S) {
211 return sys::path::filename(S->File->getName());
215 bool LinkerScript::shouldKeep(InputSectionBase *S) {
216 for (InputSectionDescription *ID : Opt.KeptSections)
217 if (ID->FilePat.match(basename(S)))
218 for (SectionPattern &P : ID->SectionPatterns)
219 if (P.SectionPat.match(S->Name))
224 // A helper function for the SORT() command.
225 static std::function<bool(InputSectionBase *, InputSectionBase *)>
226 getComparator(SortSectionPolicy K) {
228 case SortSectionPolicy::Alignment:
229 return [](InputSectionBase *A, InputSectionBase *B) {
230 // ">" is not a mistake. Sections with larger alignments are placed
231 // before sections with smaller alignments in order to reduce the
232 // amount of padding necessary. This is compatible with GNU.
233 return A->Alignment > B->Alignment;
235 case SortSectionPolicy::Name:
236 return [](InputSectionBase *A, InputSectionBase *B) {
237 return A->Name < B->Name;
239 case SortSectionPolicy::Priority:
240 return [](InputSectionBase *A, InputSectionBase *B) {
241 return getPriority(A->Name) < getPriority(B->Name);
244 llvm_unreachable("unknown sort policy");
248 // A helper function for the SORT() command.
249 static bool matchConstraints(ArrayRef<InputSectionBase *> Sections,
250 ConstraintKind Kind) {
251 if (Kind == ConstraintKind::NoConstraint)
254 bool IsRW = llvm::any_of(Sections, [](InputSectionBase *Sec) {
255 return static_cast<InputSectionBase *>(Sec)->Flags & SHF_WRITE;
258 return (IsRW && Kind == ConstraintKind::ReadWrite) ||
259 (!IsRW && Kind == ConstraintKind::ReadOnly);
262 static void sortSections(InputSectionBase **Begin, InputSectionBase **End,
263 SortSectionPolicy K) {
264 if (K != SortSectionPolicy::Default && K != SortSectionPolicy::None)
265 std::stable_sort(Begin, End, getComparator(K));
268 // Compute and remember which sections the InputSectionDescription matches.
269 std::vector<InputSectionBase *>
270 LinkerScript::computeInputSections(const InputSectionDescription *Cmd) {
271 std::vector<InputSectionBase *> Ret;
273 // Collects all sections that satisfy constraints of Cmd.
274 for (const SectionPattern &Pat : Cmd->SectionPatterns) {
275 size_t SizeBefore = Ret.size();
277 for (InputSectionBase *Sec : InputSections) {
281 // For -emit-relocs we have to ignore entries like
282 // .rela.dyn : { *(.rela.data) }
283 // which are common because they are in the default bfd script.
284 if (Sec->Type == SHT_REL || Sec->Type == SHT_RELA)
287 StringRef Filename = basename(Sec);
288 if (!Cmd->FilePat.match(Filename) ||
289 Pat.ExcludedFilePat.match(Filename) ||
290 !Pat.SectionPat.match(Sec->Name))
294 Sec->Assigned = true;
297 // Sort sections as instructed by SORT-family commands and --sort-section
298 // option. Because SORT-family commands can be nested at most two depth
299 // (e.g. SORT_BY_NAME(SORT_BY_ALIGNMENT(.text.*))) and because the command
300 // line option is respected even if a SORT command is given, the exact
301 // behavior we have here is a bit complicated. Here are the rules.
303 // 1. If two SORT commands are given, --sort-section is ignored.
304 // 2. If one SORT command is given, and if it is not SORT_NONE,
305 // --sort-section is handled as an inner SORT command.
306 // 3. If one SORT command is given, and if it is SORT_NONE, don't sort.
307 // 4. If no SORT command is given, sort according to --sort-section.
308 InputSectionBase **Begin = Ret.data() + SizeBefore;
309 InputSectionBase **End = Ret.data() + Ret.size();
310 if (Pat.SortOuter != SortSectionPolicy::None) {
311 if (Pat.SortInner == SortSectionPolicy::Default)
312 sortSections(Begin, End, Config->SortSection);
314 sortSections(Begin, End, Pat.SortInner);
315 sortSections(Begin, End, Pat.SortOuter);
321 void LinkerScript::discard(ArrayRef<InputSectionBase *> V) {
322 for (InputSectionBase *S : V) {
324 if (S == InX::ShStrTab)
325 error("discarding .shstrtab section is not allowed");
326 discard(S->DependentSections);
330 std::vector<InputSectionBase *>
331 LinkerScript::createInputSectionList(OutputSectionCommand &OutCmd) {
332 std::vector<InputSectionBase *> Ret;
334 for (BaseCommand *Base : OutCmd.Commands) {
335 auto *Cmd = dyn_cast<InputSectionDescription>(Base);
339 Cmd->Sections = computeInputSections(Cmd);
340 Ret.insert(Ret.end(), Cmd->Sections.begin(), Cmd->Sections.end());
346 void LinkerScript::processCommands(OutputSectionFactory &Factory) {
347 // A symbol can be assigned before any section is mentioned in the linker
348 // script. In an DSO, the symbol values are addresses, so the only important
349 // section values are:
352 // * Any value meaning a regular section.
353 // To handle that, create a dummy aether section that fills the void before
354 // the linker scripts switches to another section. It has an index of one
355 // which will map to whatever the first actual section is.
356 Aether = make<OutputSection>("", 0, SHF_ALLOC);
357 Aether->SectionIndex = 1;
361 for (size_t I = 0; I < Opt.Commands.size(); ++I) {
362 // Handle symbol assignments outside of any output section.
363 if (auto *Cmd = dyn_cast<SymbolAssignment>(Opt.Commands[I])) {
368 if (auto *Cmd = dyn_cast<OutputSectionCommand>(Opt.Commands[I])) {
369 std::vector<InputSectionBase *> V = createInputSectionList(*Cmd);
371 // The output section name `/DISCARD/' is special.
372 // Any input section assigned to it is discarded.
373 if (Cmd->Name == "/DISCARD/") {
378 // This is for ONLY_IF_RO and ONLY_IF_RW. An output section directive
379 // ".foo : ONLY_IF_R[OW] { ... }" is handled only if all member input
380 // sections satisfy a given constraint. If not, a directive is handled
381 // as if it wasn't present from the beginning.
383 // Because we'll iterate over Commands many more times, the easiest
384 // way to "make it as if it wasn't present" is to just remove it.
385 if (!matchConstraints(V, Cmd->Constraint)) {
386 for (InputSectionBase *S : V)
388 Opt.Commands.erase(Opt.Commands.begin() + I);
393 // A directive may contain symbol definitions like this:
394 // ".foo : { ...; bar = .; }". Handle them.
395 for (BaseCommand *Base : Cmd->Commands)
396 if (auto *OutCmd = dyn_cast<SymbolAssignment>(Base))
399 // Handle subalign (e.g. ".foo : SUBALIGN(32) { ... }"). If subalign
400 // is given, input sections are aligned to that value, whether the
401 // given value is larger or smaller than the original section alignment.
402 if (Cmd->SubalignExpr) {
403 uint32_t Subalign = Cmd->SubalignExpr().getValue();
404 for (InputSectionBase *S : V)
405 S->Alignment = Subalign;
408 // Add input sections to an output section.
409 for (InputSectionBase *S : V)
410 Factory.addInputSec(S, Cmd->Name);
416 // Add sections that didn't match any sections command.
417 void LinkerScript::addOrphanSections(OutputSectionFactory &Factory) {
418 for (InputSectionBase *S : InputSections)
419 if (S->Live && !S->OutSec)
420 Factory.addInputSec(S, getOutputSectionName(S->Name));
423 static bool isTbss(OutputSection *Sec) {
424 return (Sec->Flags & SHF_TLS) && Sec->Type == SHT_NOBITS;
427 void LinkerScript::output(InputSection *S) {
428 if (!AlreadyOutputIS.insert(S).second)
430 bool IsTbss = isTbss(CurOutSec);
432 uint64_t Pos = IsTbss ? Dot + ThreadBssOffset : Dot;
433 Pos = alignTo(Pos, S->Alignment);
434 S->OutSecOff = Pos - CurOutSec->Addr;
437 // Update output section size after adding each section. This is so that
438 // SIZEOF works correctly in the case below:
439 // .foo { *(.aaa) a = SIZEOF(.foo); *(.bbb) }
440 CurOutSec->Size = Pos - CurOutSec->Addr;
442 // If there is a memory region associated with this input section, then
443 // place the section in that region and update the region index.
445 CurMemRegion->Offset += CurOutSec->Size;
446 uint64_t CurSize = CurMemRegion->Offset - CurMemRegion->Origin;
447 if (CurSize > CurMemRegion->Length) {
448 uint64_t OverflowAmt = CurSize - CurMemRegion->Length;
449 error("section '" + CurOutSec->Name + "' will not fit in region '" +
450 CurMemRegion->Name + "': overflowed by " + Twine(OverflowAmt) +
456 ThreadBssOffset = Pos - Dot;
461 void LinkerScript::flush() {
463 if (!AlreadyOutputOS.insert(CurOutSec).second)
465 for (InputSection *I : CurOutSec->Sections)
469 void LinkerScript::switchTo(OutputSection *Sec) {
470 if (CurOutSec == Sec)
472 if (AlreadyOutputOS.count(Sec))
477 Dot = alignTo(Dot, CurOutSec->Alignment);
478 CurOutSec->Addr = isTbss(CurOutSec) ? Dot + ThreadBssOffset : Dot;
480 // If neither AT nor AT> is specified for an allocatable section, the linker
481 // will set the LMA such that the difference between VMA and LMA for the
482 // section is the same as the preceding output section in the same region
483 // https://sourceware.org/binutils/docs-2.20/ld/Output-Section-LMA.html
485 CurOutSec->LMAOffset = LMAOffset();
488 void LinkerScript::process(BaseCommand &Base) {
489 // This handles the assignments to symbol or to the dot.
490 if (auto *Cmd = dyn_cast<SymbolAssignment>(&Base)) {
491 assignSymbol(Cmd, true);
495 // Handle BYTE(), SHORT(), LONG(), or QUAD().
496 if (auto *Cmd = dyn_cast<BytesDataCommand>(&Base)) {
497 Cmd->Offset = Dot - CurOutSec->Addr;
499 CurOutSec->Size = Dot - CurOutSec->Addr;
504 if (auto *Cmd = dyn_cast<AssertCommand>(&Base)) {
509 // Handle a single input section description command.
510 // It calculates and assigns the offsets for each section and also
511 // updates the output section size.
512 auto &Cmd = cast<InputSectionDescription>(Base);
513 for (InputSectionBase *Sec : Cmd.Sections) {
514 // We tentatively added all synthetic sections at the beginning and removed
515 // empty ones afterwards (because there is no way to know whether they were
516 // going be empty or not other than actually running linker scripts.)
517 // We need to ignore remains of empty sections.
518 if (auto *S = dyn_cast<SyntheticSection>(Sec))
524 assert(CurOutSec == Sec->OutSec || AlreadyOutputOS.count(Sec->OutSec));
525 output(cast<InputSection>(Sec));
529 static OutputSection *
530 findSection(StringRef Name, const std::vector<OutputSection *> &Sections) {
531 for (OutputSection *Sec : Sections)
532 if (Sec->Name == Name)
537 // This function searches for a memory region to place the given output
538 // section in. If found, a pointer to the appropriate memory region is
539 // returned. Otherwise, a nullptr is returned.
540 MemoryRegion *LinkerScript::findMemoryRegion(OutputSectionCommand *Cmd) {
541 // If a memory region name was specified in the output section command,
542 // then try to find that region first.
543 if (!Cmd->MemoryRegionName.empty()) {
544 auto It = Opt.MemoryRegions.find(Cmd->MemoryRegionName);
545 if (It != Opt.MemoryRegions.end())
547 error("memory region '" + Cmd->MemoryRegionName + "' not declared");
551 // If at least one memory region is defined, all sections must
552 // belong to some memory region. Otherwise, we don't need to do
553 // anything for memory regions.
554 if (Opt.MemoryRegions.empty())
557 OutputSection *Sec = Cmd->Sec;
558 // See if a region can be found by matching section flags.
559 for (auto &Pair : Opt.MemoryRegions) {
560 MemoryRegion &M = Pair.second;
561 if ((M.Flags & Sec->Flags) && (M.NegFlags & Sec->Flags) == 0)
565 // Otherwise, no suitable region was found.
566 if (Sec->Flags & SHF_ALLOC)
567 error("no memory region specified for section '" + Sec->Name + "'");
571 // This function assigns offsets to input sections and an output section
572 // for a single sections command (e.g. ".text { *(.text); }").
573 void LinkerScript::assignOffsets(OutputSectionCommand *Cmd) {
574 OutputSection *Sec = Cmd->Sec;
578 if (Cmd->AddrExpr && (Sec->Flags & SHF_ALLOC))
579 setDot(Cmd->AddrExpr, Cmd->Location, false);
583 LMAOffset = [=] { return Cmd->LMAExpr().getValue() - D; };
586 CurMemRegion = Cmd->MemRegion;
588 Dot = CurMemRegion->Offset;
591 // flush() may add orphan sections, so the order of flush() and
592 // symbol assignments is important. We want to call flush() first so
593 // that symbols pointing the end of the current section points to
594 // the location after orphan sections.
596 std::find_if(Cmd->Commands.rbegin(), Cmd->Commands.rend(),
597 [](BaseCommand *Cmd) { return !isa<SymbolAssignment>(Cmd); })
599 for (auto I = Cmd->Commands.begin(); I != Mid; ++I)
602 for (auto I = Mid, E = Cmd->Commands.end(); I != E; ++I)
606 void LinkerScript::removeEmptyCommands() {
607 // It is common practice to use very generic linker scripts. So for any
608 // given run some of the output sections in the script will be empty.
609 // We could create corresponding empty output sections, but that would
610 // clutter the output.
611 // We instead remove trivially empty sections. The bfd linker seems even
612 // more aggressive at removing them.
613 auto Pos = std::remove_if(
614 Opt.Commands.begin(), Opt.Commands.end(), [&](BaseCommand *Base) {
615 if (auto *Cmd = dyn_cast<OutputSectionCommand>(Base))
619 Opt.Commands.erase(Pos, Opt.Commands.end());
622 static bool isAllSectionDescription(const OutputSectionCommand &Cmd) {
623 for (BaseCommand *Base : Cmd.Commands)
624 if (!isa<InputSectionDescription>(*Base))
629 void LinkerScript::adjustSectionsBeforeSorting() {
630 // If the output section contains only symbol assignments, create a
631 // corresponding output section. The bfd linker seems to only create them if
632 // '.' is assigned to, but creating these section should not have any bad
633 // consequeces and gives us a section to put the symbol in.
634 uint64_t Flags = SHF_ALLOC;
635 uint32_t Type = SHT_PROGBITS;
636 for (BaseCommand *Base : Opt.Commands) {
637 auto *Cmd = dyn_cast<OutputSectionCommand>(Base);
640 if (OutputSection *Sec = findSection(Cmd->Name, *OutputSections)) {
647 if (isAllSectionDescription(*Cmd))
650 auto *OutSec = make<OutputSection>(Cmd->Name, Type, Flags);
651 OutputSections->push_back(OutSec);
656 void LinkerScript::adjustSectionsAfterSorting() {
657 placeOrphanSections();
659 // Try and find an appropriate memory region to assign offsets in.
660 for (BaseCommand *Base : Opt.Commands) {
661 if (auto *Cmd = dyn_cast<OutputSectionCommand>(Base)) {
662 Cmd->MemRegion = findMemoryRegion(Cmd);
663 // Handle align (e.g. ".foo : ALIGN(16) { ... }").
665 Cmd->Sec->updateAlignment(Cmd->AlignExpr().getValue());
669 // If output section command doesn't specify any segments,
670 // and we haven't previously assigned any section to segment,
671 // then we simply assign section to the very first load segment.
672 // Below is an example of such linker script:
673 // PHDRS { seg PT_LOAD; }
674 // SECTIONS { .aaa : { *(.aaa) } }
675 std::vector<StringRef> DefPhdrs;
677 std::find_if(Opt.PhdrsCommands.begin(), Opt.PhdrsCommands.end(),
678 [](const PhdrsCommand &Cmd) { return Cmd.Type == PT_LOAD; });
679 if (FirstPtLoad != Opt.PhdrsCommands.end())
680 DefPhdrs.push_back(FirstPtLoad->Name);
682 // Walk the commands and propagate the program headers to commands that don't
683 // explicitly specify them.
684 for (BaseCommand *Base : Opt.Commands) {
685 auto *Cmd = dyn_cast<OutputSectionCommand>(Base);
689 if (Cmd->Phdrs.empty())
690 Cmd->Phdrs = DefPhdrs;
692 DefPhdrs = Cmd->Phdrs;
695 removeEmptyCommands();
698 // When placing orphan sections, we want to place them after symbol assignments
699 // so that an orphan after
703 // doesn't break the intended meaning of the begin/end symbols.
704 // We don't want to go over sections since Writer<ELFT>::sortSections is the
705 // one in charge of deciding the order of the sections.
706 // We don't want to go over alignments, since doing so in
707 // rx_sec : { *(rx_sec) }
708 // . = ALIGN(0x1000);
709 // /* The RW PT_LOAD starts here*/
710 // rw_sec : { *(rw_sec) }
711 // would mean that the RW PT_LOAD would become unaligned.
712 static bool shouldSkip(BaseCommand *Cmd) {
713 if (isa<OutputSectionCommand>(Cmd))
715 if (auto *Assign = dyn_cast<SymbolAssignment>(Cmd))
716 return Assign->Name != ".";
720 // Orphan sections are sections present in the input files which are
721 // not explicitly placed into the output file by the linker script.
723 // When the control reaches this function, Opt.Commands contains
724 // output section commands for non-orphan sections only. This function
725 // adds new elements for orphan sections so that all sections are
726 // explicitly handled by Opt.Commands.
728 // Writer<ELFT>::sortSections has already sorted output sections.
729 // What we need to do is to scan OutputSections vector and
730 // Opt.Commands in parallel to find orphan sections. If there is an
731 // output section that doesn't have a corresponding entry in
732 // Opt.Commands, we will insert a new entry to Opt.Commands.
734 // There is some ambiguity as to where exactly a new entry should be
735 // inserted, because Opt.Commands contains not only output section
736 // commands but also other types of commands such as symbol assignment
737 // expressions. There's no correct answer here due to the lack of the
738 // formal specification of the linker script. We use heuristics to
739 // determine whether a new output command should be added before or
740 // after another commands. For the details, look at shouldSkip
742 void LinkerScript::placeOrphanSections() {
743 // The OutputSections are already in the correct order.
744 // This loops creates or moves commands as needed so that they are in the
748 // As a horrible special case, skip the first . assignment if it is before any
749 // section. We do this because it is common to set a load address by starting
750 // the script with ". = 0xabcd" and the expectation is that every section is
752 auto FirstSectionOrDotAssignment =
753 std::find_if(Opt.Commands.begin(), Opt.Commands.end(),
754 [](BaseCommand *Cmd) { return !shouldSkip(Cmd); });
755 if (FirstSectionOrDotAssignment != Opt.Commands.end()) {
756 CmdIndex = FirstSectionOrDotAssignment - Opt.Commands.begin();
757 if (isa<SymbolAssignment>(**FirstSectionOrDotAssignment))
761 for (OutputSection *Sec : *OutputSections) {
762 StringRef Name = Sec->Name;
764 // Find the last spot where we can insert a command and still get the
766 auto CmdIter = Opt.Commands.begin() + CmdIndex;
767 auto E = Opt.Commands.end();
768 while (CmdIter != E && shouldSkip(*CmdIter)) {
773 auto Pos = std::find_if(CmdIter, E, [&](BaseCommand *Base) {
774 auto *Cmd = dyn_cast<OutputSectionCommand>(Base);
775 return Cmd && Cmd->Name == Name;
778 auto *Cmd = make<OutputSectionCommand>(Name);
780 Opt.Commands.insert(CmdIter, Cmd);
785 // Continue from where we found it.
786 CmdIndex = (Pos - Opt.Commands.begin()) + 1;
790 void LinkerScript::processNonSectionCommands() {
791 for (BaseCommand *Base : Opt.Commands) {
792 if (auto *Cmd = dyn_cast<SymbolAssignment>(Base))
793 assignSymbol(Cmd, false);
794 else if (auto *Cmd = dyn_cast<AssertCommand>(Base))
799 void LinkerScript::assignAddresses(std::vector<PhdrEntry> &Phdrs) {
800 // Assign addresses as instructed by linker script SECTIONS sub-commands.
802 ErrorOnMissingSection = true;
805 for (BaseCommand *Base : Opt.Commands) {
806 if (auto *Cmd = dyn_cast<SymbolAssignment>(Base)) {
807 assignSymbol(Cmd, false);
811 if (auto *Cmd = dyn_cast<AssertCommand>(Base)) {
816 auto *Cmd = cast<OutputSectionCommand>(Base);
820 uint64_t MinVA = std::numeric_limits<uint64_t>::max();
821 for (OutputSection *Sec : *OutputSections) {
822 if (Sec->Flags & SHF_ALLOC)
823 MinVA = std::min<uint64_t>(MinVA, Sec->Addr);
828 allocateHeaders(Phdrs, *OutputSections, MinVA);
831 // Creates program headers as instructed by PHDRS linker script command.
832 std::vector<PhdrEntry> LinkerScript::createPhdrs() {
833 std::vector<PhdrEntry> Ret;
835 // Process PHDRS and FILEHDR keywords because they are not
836 // real output sections and cannot be added in the following loop.
837 for (const PhdrsCommand &Cmd : Opt.PhdrsCommands) {
838 Ret.emplace_back(Cmd.Type, Cmd.Flags == UINT_MAX ? PF_R : Cmd.Flags);
839 PhdrEntry &Phdr = Ret.back();
842 Phdr.add(Out::ElfHeader);
844 Phdr.add(Out::ProgramHeaders);
847 Phdr.p_paddr = Cmd.LMAExpr().getValue();
852 // Add output sections to program headers.
853 for (OutputSection *Sec : *OutputSections) {
854 if (!(Sec->Flags & SHF_ALLOC))
857 // Assign headers specified by linker script
858 for (size_t Id : getPhdrIndices(Sec->Name)) {
860 if (Opt.PhdrsCommands[Id].Flags == UINT_MAX)
861 Ret[Id].p_flags |= Sec->getPhdrFlags();
867 bool LinkerScript::ignoreInterpSection() {
868 // Ignore .interp section in case we have PHDRS specification
869 // and PT_INTERP isn't listed.
870 if (Opt.PhdrsCommands.empty())
872 for (PhdrsCommand &Cmd : Opt.PhdrsCommands)
873 if (Cmd.Type == PT_INTERP)
878 Optional<uint32_t> LinkerScript::getFiller(StringRef Name) {
879 for (BaseCommand *Base : Opt.Commands)
880 if (auto *Cmd = dyn_cast<OutputSectionCommand>(Base))
881 if (Cmd->Name == Name)
886 static void writeInt(uint8_t *Buf, uint64_t Data, uint64_t Size) {
890 write16(Buf, Data, Config->Endianness);
892 write32(Buf, Data, Config->Endianness);
894 write64(Buf, Data, Config->Endianness);
896 llvm_unreachable("unsupported Size argument");
899 void LinkerScript::writeDataBytes(StringRef Name, uint8_t *Buf) {
900 int I = getSectionIndex(Name);
904 auto *Cmd = dyn_cast<OutputSectionCommand>(Opt.Commands[I]);
905 for (BaseCommand *Base : Cmd->Commands)
906 if (auto *Data = dyn_cast<BytesDataCommand>(Base))
907 writeInt(Buf + Data->Offset, Data->Expression().getValue(), Data->Size);
910 bool LinkerScript::hasLMA(StringRef Name) {
911 for (BaseCommand *Base : Opt.Commands)
912 if (auto *Cmd = dyn_cast<OutputSectionCommand>(Base))
913 if (Cmd->LMAExpr && Cmd->Name == Name)
918 // Returns the index of the given section name in linker script
919 // SECTIONS commands. Sections are laid out as the same order as they
920 // were in the script. If a given name did not appear in the script,
921 // it returns INT_MAX, so that it will be laid out at end of file.
922 int LinkerScript::getSectionIndex(StringRef Name) {
923 for (int I = 0, E = Opt.Commands.size(); I != E; ++I)
924 if (auto *Cmd = dyn_cast<OutputSectionCommand>(Opt.Commands[I]))
925 if (Cmd->Name == Name)
930 ExprValue LinkerScript::getSymbolValue(const Twine &Loc, StringRef S) {
932 return {CurOutSec, Dot - CurOutSec->Addr};
933 if (SymbolBody *B = findSymbol(S)) {
934 if (auto *D = dyn_cast<DefinedRegular>(B))
935 return {D->Section, D->Value};
936 if (auto *C = dyn_cast<DefinedCommon>(B))
937 return {InX::Common, C->Offset};
939 error(Loc + ": symbol not found: " + S);
943 bool LinkerScript::isDefined(StringRef S) { return findSymbol(S) != nullptr; }
945 // Returns indices of ELF headers containing specific section, identified
946 // by Name. Each index is a zero based number of ELF header listed within
947 // PHDRS {} script block.
948 std::vector<size_t> LinkerScript::getPhdrIndices(StringRef SectionName) {
949 for (BaseCommand *Base : Opt.Commands) {
950 auto *Cmd = dyn_cast<OutputSectionCommand>(Base);
951 if (!Cmd || Cmd->Name != SectionName)
954 std::vector<size_t> Ret;
955 for (StringRef PhdrName : Cmd->Phdrs)
956 Ret.push_back(getPhdrIndex(Cmd->Location, PhdrName));
962 size_t LinkerScript::getPhdrIndex(const Twine &Loc, StringRef PhdrName) {
964 for (PhdrsCommand &Cmd : Opt.PhdrsCommands) {
965 if (Cmd.Name == PhdrName)
969 error(Loc + ": section header '" + PhdrName + "' is not listed in PHDRS");