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"
17 #include "OutputSections.h"
18 #include "SymbolTable.h"
20 #include "SyntheticSections.h"
23 #include "lld/Common/Memory.h"
24 #include "lld/Common/Strings.h"
25 #include "lld/Common/Threads.h"
26 #include "llvm/ADT/STLExtras.h"
27 #include "llvm/ADT/StringRef.h"
28 #include "llvm/BinaryFormat/ELF.h"
29 #include "llvm/Support/Casting.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 static uint64_t getOutputSectionVA(SectionBase *InputSec, StringRef Loc) {
53 if (OutputSection *OS = InputSec->getOutputSection())
55 error(Loc + ": unable to evaluate expression: input section " +
56 InputSec->Name + " has no output section assigned");
60 uint64_t ExprValue::getValue() const {
62 return alignTo(Sec->getOffset(Val) + getOutputSectionVA(Sec, Loc),
64 return alignTo(Val, Alignment);
67 uint64_t ExprValue::getSecAddr() const {
69 return Sec->getOffset(0) + getOutputSectionVA(Sec, Loc);
73 uint64_t ExprValue::getSectionOffset() const {
74 // If the alignment is trivial, we don't have to compute the full
75 // value to know the offset. This allows this function to succeed in
76 // cases where the output section is not yet known.
77 if (Alignment == 1 && (!Sec || !Sec->getOutputSection()))
79 return getValue() - getSecAddr();
82 OutputSection *LinkerScript::createOutputSection(StringRef Name,
84 OutputSection *&SecRef = NameToOutputSection[Name];
86 if (SecRef && SecRef->Location.empty()) {
87 // There was a forward reference.
90 Sec = make<OutputSection>(Name, SHT_NOBITS, 0);
94 Sec->Location = Location;
98 OutputSection *LinkerScript::getOrCreateOutputSection(StringRef Name) {
99 OutputSection *&CmdRef = NameToOutputSection[Name];
101 CmdRef = make<OutputSection>(Name, SHT_PROGBITS, 0);
105 // Expands the memory region by the specified size.
106 static void expandMemoryRegion(MemoryRegion *MemRegion, uint64_t Size,
107 StringRef RegionName, StringRef SecName) {
108 MemRegion->CurPos += Size;
109 uint64_t NewSize = MemRegion->CurPos - MemRegion->Origin;
110 if (NewSize > MemRegion->Length)
111 error("section '" + SecName + "' will not fit in region '" + RegionName +
112 "': overflowed by " + Twine(NewSize - MemRegion->Length) + " bytes");
115 void LinkerScript::expandMemoryRegions(uint64_t Size) {
117 expandMemoryRegion(Ctx->MemRegion, Size, Ctx->MemRegion->Name,
120 expandMemoryRegion(Ctx->LMARegion, Size, Ctx->LMARegion->Name,
124 void LinkerScript::expandOutputSection(uint64_t Size) {
125 Ctx->OutSec->Size += Size;
126 expandMemoryRegions(Size);
129 void LinkerScript::setDot(Expr E, const Twine &Loc, bool InSec) {
130 uint64_t Val = E().getValue();
131 if (Val < Dot && InSec)
132 error(Loc + ": unable to move location counter backward for: " +
135 // Update to location counter means update to section size.
137 expandOutputSection(Val - Dot);
139 expandMemoryRegions(Val - Dot);
144 // Used for handling linker symbol assignments, for both finalizing
145 // their values and doing early declarations. Returns true if symbol
146 // should be defined from linker script.
147 static bool shouldDefineSym(SymbolAssignment *Cmd) {
148 if (Cmd->Name == ".")
154 // If a symbol was in PROVIDE(), we need to define it only
155 // when it is a referenced undefined symbol.
156 Symbol *B = Symtab->find(Cmd->Name);
157 if (B && !B->isDefined())
162 // This function is called from processSectionCommands,
163 // while we are fixing the output section layout.
164 void LinkerScript::addSymbol(SymbolAssignment *Cmd) {
165 if (!shouldDefineSym(Cmd))
170 uint8_t Visibility = Cmd->Hidden ? STV_HIDDEN : STV_DEFAULT;
171 std::tie(Sym, std::ignore) = Symtab->insert(Cmd->Name, /*Type*/ 0, Visibility,
172 /*CanOmitFromDynSym*/ false,
174 ExprValue Value = Cmd->Expression();
175 SectionBase *Sec = Value.isAbsolute() ? nullptr : Value.Sec;
177 // When this function is called, section addresses have not been
178 // fixed yet. So, we may or may not know the value of the RHS
181 // For example, if an expression is `x = 42`, we know x is always 42.
182 // However, if an expression is `x = .`, there's no way to know its
183 // value at the moment.
185 // We want to set symbol values early if we can. This allows us to
186 // use symbols as variables in linker scripts. Doing so allows us to
187 // write expressions like this: `alignment = 16; . = ALIGN(., alignment)`.
188 uint64_t SymValue = Value.Sec ? 0 : Value.getValue();
190 replaceSymbol<Defined>(Sym, nullptr, Cmd->Name, STB_GLOBAL, Visibility,
191 STT_NOTYPE, SymValue, 0, Sec);
192 Cmd->Sym = cast<Defined>(Sym);
195 // This function is called from LinkerScript::declareSymbols.
196 // It creates a placeholder symbol if needed.
197 static void declareSymbol(SymbolAssignment *Cmd) {
198 if (!shouldDefineSym(Cmd))
201 // We can't calculate final value right now.
203 uint8_t Visibility = Cmd->Hidden ? STV_HIDDEN : STV_DEFAULT;
204 std::tie(Sym, std::ignore) = Symtab->insert(Cmd->Name, /*Type*/ 0, Visibility,
205 /*CanOmitFromDynSym*/ false,
207 replaceSymbol<Defined>(Sym, nullptr, Cmd->Name, STB_GLOBAL, Visibility,
208 STT_NOTYPE, 0, 0, nullptr);
209 Cmd->Sym = cast<Defined>(Sym);
210 Cmd->Provide = false;
213 // This method is used to handle INSERT AFTER statement. Here we rebuild
214 // the list of script commands to mix sections inserted into.
215 void LinkerScript::processInsertCommands() {
216 std::vector<BaseCommand *> V;
217 auto Insert = [&](std::vector<BaseCommand *> &From) {
218 V.insert(V.end(), From.begin(), From.end());
222 for (BaseCommand *Base : SectionCommands) {
223 if (auto *OS = dyn_cast<OutputSection>(Base)) {
224 Insert(InsertBeforeCommands[OS->Name]);
226 Insert(InsertAfterCommands[OS->Name]);
232 for (auto &Cmds : {InsertBeforeCommands, InsertAfterCommands})
233 for (const std::pair<StringRef, std::vector<BaseCommand *>> &P : Cmds)
234 if (!P.second.empty())
235 error("unable to INSERT AFTER/BEFORE " + P.first +
236 ": section not defined");
238 SectionCommands = std::move(V);
241 // Symbols defined in script should not be inlined by LTO. At the same time
242 // we don't know their final values until late stages of link. Here we scan
243 // over symbol assignment commands and create placeholder symbols if needed.
244 void LinkerScript::declareSymbols() {
246 for (BaseCommand *Base : SectionCommands) {
247 if (auto *Cmd = dyn_cast<SymbolAssignment>(Base)) {
252 // If the output section directive has constraints,
253 // we can't say for sure if it is going to be included or not.
254 // Skip such sections for now. Improve the checks if we ever
255 // need symbols from that sections to be declared early.
256 auto *Sec = cast<OutputSection>(Base);
257 if (Sec->Constraint != ConstraintKind::NoConstraint)
259 for (BaseCommand *Base2 : Sec->SectionCommands)
260 if (auto *Cmd = dyn_cast<SymbolAssignment>(Base2))
265 // This function is called from assignAddresses, while we are
266 // fixing the output section addresses. This function is supposed
267 // to set the final value for a given symbol assignment.
268 void LinkerScript::assignSymbol(SymbolAssignment *Cmd, bool InSec) {
269 if (Cmd->Name == ".") {
270 setDot(Cmd->Expression, Cmd->Location, InSec);
277 ExprValue V = Cmd->Expression();
278 if (V.isAbsolute()) {
279 Cmd->Sym->Section = nullptr;
280 Cmd->Sym->Value = V.getValue();
282 Cmd->Sym->Section = V.Sec;
283 Cmd->Sym->Value = V.getSectionOffset();
287 static std::string getFilename(InputFile *File) {
290 if (File->ArchiveName.empty())
291 return File->getName();
292 return (File->ArchiveName + "(" + File->getName() + ")").str();
295 bool LinkerScript::shouldKeep(InputSectionBase *S) {
296 if (KeptSections.empty())
298 std::string Filename = getFilename(S->File);
299 for (InputSectionDescription *ID : KeptSections)
300 if (ID->FilePat.match(Filename))
301 for (SectionPattern &P : ID->SectionPatterns)
302 if (P.SectionPat.match(S->Name))
307 // A helper function for the SORT() command.
308 static std::function<bool(InputSectionBase *, InputSectionBase *)>
309 getComparator(SortSectionPolicy K) {
311 case SortSectionPolicy::Alignment:
312 return [](InputSectionBase *A, InputSectionBase *B) {
313 // ">" is not a mistake. Sections with larger alignments are placed
314 // before sections with smaller alignments in order to reduce the
315 // amount of padding necessary. This is compatible with GNU.
316 return A->Alignment > B->Alignment;
318 case SortSectionPolicy::Name:
319 return [](InputSectionBase *A, InputSectionBase *B) {
320 return A->Name < B->Name;
322 case SortSectionPolicy::Priority:
323 return [](InputSectionBase *A, InputSectionBase *B) {
324 return getPriority(A->Name) < getPriority(B->Name);
327 llvm_unreachable("unknown sort policy");
331 // A helper function for the SORT() command.
332 static bool matchConstraints(ArrayRef<InputSection *> Sections,
333 ConstraintKind Kind) {
334 if (Kind == ConstraintKind::NoConstraint)
337 bool IsRW = llvm::any_of(
338 Sections, [](InputSection *Sec) { return Sec->Flags & SHF_WRITE; });
340 return (IsRW && Kind == ConstraintKind::ReadWrite) ||
341 (!IsRW && Kind == ConstraintKind::ReadOnly);
344 static void sortSections(MutableArrayRef<InputSection *> Vec,
345 SortSectionPolicy K) {
346 if (K != SortSectionPolicy::Default && K != SortSectionPolicy::None)
347 std::stable_sort(Vec.begin(), Vec.end(), getComparator(K));
350 // Sort sections as instructed by SORT-family commands and --sort-section
351 // option. Because SORT-family commands can be nested at most two depth
352 // (e.g. SORT_BY_NAME(SORT_BY_ALIGNMENT(.text.*))) and because the command
353 // line option is respected even if a SORT command is given, the exact
354 // behavior we have here is a bit complicated. Here are the rules.
356 // 1. If two SORT commands are given, --sort-section is ignored.
357 // 2. If one SORT command is given, and if it is not SORT_NONE,
358 // --sort-section is handled as an inner SORT command.
359 // 3. If one SORT command is given, and if it is SORT_NONE, don't sort.
360 // 4. If no SORT command is given, sort according to --sort-section.
361 static void sortInputSections(MutableArrayRef<InputSection *> Vec,
362 const SectionPattern &Pat) {
363 if (Pat.SortOuter == SortSectionPolicy::None)
366 if (Pat.SortInner == SortSectionPolicy::Default)
367 sortSections(Vec, Config->SortSection);
369 sortSections(Vec, Pat.SortInner);
370 sortSections(Vec, Pat.SortOuter);
373 // Compute and remember which sections the InputSectionDescription matches.
374 std::vector<InputSection *>
375 LinkerScript::computeInputSections(const InputSectionDescription *Cmd) {
376 std::vector<InputSection *> Ret;
378 // Collects all sections that satisfy constraints of Cmd.
379 for (const SectionPattern &Pat : Cmd->SectionPatterns) {
380 size_t SizeBefore = Ret.size();
382 for (InputSectionBase *Sec : InputSections) {
383 if (!Sec->Live || Sec->Assigned)
386 // For -emit-relocs we have to ignore entries like
387 // .rela.dyn : { *(.rela.data) }
388 // which are common because they are in the default bfd script.
389 // We do not ignore SHT_REL[A] linker-synthesized sections here because
390 // want to support scripts that do custom layout for them.
391 if (auto *IS = dyn_cast<InputSection>(Sec))
392 if (IS->getRelocatedSection())
395 std::string Filename = getFilename(Sec->File);
396 if (!Cmd->FilePat.match(Filename) ||
397 Pat.ExcludedFilePat.match(Filename) ||
398 !Pat.SectionPat.match(Sec->Name))
401 // It is safe to assume that Sec is an InputSection
402 // because mergeable or EH input sections have already been
403 // handled and eliminated.
404 Ret.push_back(cast<InputSection>(Sec));
405 Sec->Assigned = true;
408 sortInputSections(MutableArrayRef<InputSection *>(Ret).slice(SizeBefore),
414 void LinkerScript::discard(ArrayRef<InputSection *> V) {
415 for (InputSection *S : V) {
416 if (S == InX::ShStrTab || S == InX::Dynamic || S == InX::DynSymTab ||
417 S == InX::DynStrTab || S == InX::RelaPlt || S == InX::RelaDyn ||
419 error("discarding " + S->Name + " section is not allowed");
421 // You can discard .hash and .gnu.hash sections by linker scripts. Since
422 // they are synthesized sections, we need to handle them differently than
423 // other regular sections.
424 if (S == InX::GnuHashTab)
425 InX::GnuHashTab = nullptr;
426 if (S == InX::HashTab)
427 InX::HashTab = nullptr;
431 discard(S->DependentSections);
435 std::vector<InputSection *>
436 LinkerScript::createInputSectionList(OutputSection &OutCmd) {
437 std::vector<InputSection *> Ret;
439 for (BaseCommand *Base : OutCmd.SectionCommands) {
440 if (auto *Cmd = dyn_cast<InputSectionDescription>(Base)) {
441 Cmd->Sections = computeInputSections(Cmd);
442 Ret.insert(Ret.end(), Cmd->Sections.begin(), Cmd->Sections.end());
448 void LinkerScript::processSectionCommands() {
449 // A symbol can be assigned before any section is mentioned in the linker
450 // script. In an DSO, the symbol values are addresses, so the only important
451 // section values are:
454 // * Any value meaning a regular section.
455 // To handle that, create a dummy aether section that fills the void before
456 // the linker scripts switches to another section. It has an index of one
457 // which will map to whatever the first actual section is.
458 Aether = make<OutputSection>("", 0, SHF_ALLOC);
459 Aether->SectionIndex = 1;
461 // Ctx captures the local AddressState and makes it accessible deliberately.
462 // This is needed as there are some cases where we cannot just
463 // thread the current state through to a lambda function created by the
465 auto Deleter = make_unique<AddressState>();
467 Ctx->OutSec = Aether;
470 // Add input sections to output sections.
471 for (BaseCommand *Base : SectionCommands) {
472 // Handle symbol assignments outside of any output section.
473 if (auto *Cmd = dyn_cast<SymbolAssignment>(Base)) {
478 if (auto *Sec = dyn_cast<OutputSection>(Base)) {
479 std::vector<InputSection *> V = createInputSectionList(*Sec);
481 // The output section name `/DISCARD/' is special.
482 // Any input section assigned to it is discarded.
483 if (Sec->Name == "/DISCARD/") {
485 Sec->SectionCommands.clear();
489 // This is for ONLY_IF_RO and ONLY_IF_RW. An output section directive
490 // ".foo : ONLY_IF_R[OW] { ... }" is handled only if all member input
491 // sections satisfy a given constraint. If not, a directive is handled
492 // as if it wasn't present from the beginning.
494 // Because we'll iterate over SectionCommands many more times, the easy
495 // way to "make it as if it wasn't present" is to make it empty.
496 if (!matchConstraints(V, Sec->Constraint)) {
497 for (InputSectionBase *S : V)
499 Sec->SectionCommands.clear();
503 // A directive may contain symbol definitions like this:
504 // ".foo : { ...; bar = .; }". Handle them.
505 for (BaseCommand *Base : Sec->SectionCommands)
506 if (auto *OutCmd = dyn_cast<SymbolAssignment>(Base))
509 // Handle subalign (e.g. ".foo : SUBALIGN(32) { ... }"). If subalign
510 // is given, input sections are aligned to that value, whether the
511 // given value is larger or smaller than the original section alignment.
512 if (Sec->SubalignExpr) {
513 uint32_t Subalign = Sec->SubalignExpr().getValue();
514 for (InputSectionBase *S : V)
515 S->Alignment = Subalign;
518 // Add input sections to an output section.
519 for (InputSection *S : V)
522 Sec->SectionIndex = I++;
524 Sec->Type = SHT_NOBITS;
526 Sec->Flags &= ~(uint64_t)SHF_ALLOC;
532 static OutputSection *findByName(ArrayRef<BaseCommand *> Vec,
534 for (BaseCommand *Base : Vec)
535 if (auto *Sec = dyn_cast<OutputSection>(Base))
536 if (Sec->Name == Name)
541 static OutputSection *createSection(InputSectionBase *IS,
542 StringRef OutsecName) {
543 OutputSection *Sec = Script->createOutputSection(OutsecName, "<internal>");
544 Sec->addSection(cast<InputSection>(IS));
548 static OutputSection *addInputSec(StringMap<OutputSection *> &Map,
549 InputSectionBase *IS, StringRef OutsecName) {
550 // Sections with SHT_GROUP or SHF_GROUP attributes reach here only when the -r
551 // option is given. A section with SHT_GROUP defines a "section group", and
552 // its members have SHF_GROUP attribute. Usually these flags have already been
553 // stripped by InputFiles.cpp as section groups are processed and uniquified.
554 // However, for the -r option, we want to pass through all section groups
555 // as-is because adding/removing members or merging them with other groups
556 // change their semantics.
557 if (IS->Type == SHT_GROUP || (IS->Flags & SHF_GROUP))
558 return createSection(IS, OutsecName);
560 // Imagine .zed : { *(.foo) *(.bar) } script. Both foo and bar may have
561 // relocation sections .rela.foo and .rela.bar for example. Most tools do
562 // not allow multiple REL[A] sections for output section. Hence we
563 // should combine these relocation sections into single output.
564 // We skip synthetic sections because it can be .rela.dyn/.rela.plt or any
565 // other REL[A] sections created by linker itself.
566 if (!isa<SyntheticSection>(IS) &&
567 (IS->Type == SHT_REL || IS->Type == SHT_RELA)) {
568 auto *Sec = cast<InputSection>(IS);
569 OutputSection *Out = Sec->getRelocatedSection()->getOutputSection();
571 if (Out->RelocationSection) {
572 Out->RelocationSection->addSection(Sec);
576 Out->RelocationSection = createSection(IS, OutsecName);
577 return Out->RelocationSection;
580 // When control reaches here, mergeable sections have already been merged into
581 // synthetic sections. For relocatable case we want to create one output
582 // section per syntetic section so that they have a valid sh_entsize.
583 if (Config->Relocatable && (IS->Flags & SHF_MERGE))
584 return createSection(IS, OutsecName);
586 // The ELF spec just says
587 // ----------------------------------------------------------------
588 // In the first phase, input sections that match in name, type and
589 // attribute flags should be concatenated into single sections.
590 // ----------------------------------------------------------------
592 // However, it is clear that at least some flags have to be ignored for
593 // section merging. At the very least SHF_GROUP and SHF_COMPRESSED have to be
594 // ignored. We should not have two output .text sections just because one was
595 // in a group and another was not for example.
597 // It also seems that wording was a late addition and didn't get the
598 // necessary scrutiny.
600 // Merging sections with different flags is expected by some users. One
601 // reason is that if one file has
603 // int *const bar __attribute__((section(".foo"))) = (int *)0;
605 // gcc with -fPIC will produce a read only .foo section. But if another
609 // int *const bar __attribute__((section(".foo"))) = (int *)&zed;
611 // gcc with -fPIC will produce a read write section.
613 // Last but not least, when using linker script the merge rules are forced by
614 // the script. Unfortunately, linker scripts are name based. This means that
615 // expressions like *(.foo*) can refer to multiple input sections with
616 // different flags. We cannot put them in different output sections or we
617 // would produce wrong results for
619 // start = .; *(.foo.*) end = .; *(.bar)
621 // and a mapping of .foo1 and .bar1 to one section and .foo2 and .bar2 to
622 // another. The problem is that there is no way to layout those output
623 // sections such that the .foo sections are the only thing between the start
626 // Given the above issues, we instead merge sections by name and error on
627 // incompatible types and flags.
628 OutputSection *&Sec = Map[OutsecName];
630 Sec->addSection(cast<InputSection>(IS));
634 Sec = createSection(IS, OutsecName);
638 // Add sections that didn't match any sections command.
639 void LinkerScript::addOrphanSections() {
640 unsigned End = SectionCommands.size();
641 StringMap<OutputSection *> Map;
642 std::vector<OutputSection *> V;
644 auto Add = [&](InputSectionBase *S) {
645 if (!S->Live || S->Parent)
648 StringRef Name = getOutputSectionName(S);
650 if (Config->OrphanHandling == OrphanHandlingPolicy::Error)
651 error(toString(S) + " is being placed in '" + Name + "'");
652 else if (Config->OrphanHandling == OrphanHandlingPolicy::Warn)
653 warn(toString(S) + " is being placed in '" + Name + "'");
655 if (OutputSection *Sec =
656 findByName(makeArrayRef(SectionCommands).slice(0, End), Name)) {
657 Sec->addSection(cast<InputSection>(S));
661 if (OutputSection *OS = addInputSec(Map, S, Name))
663 assert(S->getOutputSection()->SectionIndex == UINT32_MAX);
666 // For futher --emit-reloc handling code we need target output section
667 // to be created before we create relocation output section, so we want
668 // to create target sections first. We do not want priority handling
669 // for synthetic sections because them are special.
670 for (InputSectionBase *IS : InputSections) {
671 if (auto *Sec = dyn_cast<InputSection>(IS))
672 if (InputSectionBase *Rel = Sec->getRelocatedSection())
673 if (auto *RelIS = dyn_cast_or_null<InputSectionBase>(Rel->Parent))
678 // If no SECTIONS command was given, we should insert sections commands
679 // before others, so that we can handle scripts which refers them,
680 // for example: "foo = ABSOLUTE(ADDR(.text)));".
681 // When SECTIONS command is present we just add all orphans to the end.
682 if (HasSectionsCommand)
683 SectionCommands.insert(SectionCommands.end(), V.begin(), V.end());
685 SectionCommands.insert(SectionCommands.begin(), V.begin(), V.end());
688 uint64_t LinkerScript::advance(uint64_t Size, unsigned Alignment) {
690 (Ctx->OutSec->Flags & SHF_TLS) && Ctx->OutSec->Type == SHT_NOBITS;
691 uint64_t Start = IsTbss ? Dot + Ctx->ThreadBssOffset : Dot;
692 Start = alignTo(Start, Alignment);
693 uint64_t End = Start + Size;
696 Ctx->ThreadBssOffset = End - Dot;
702 void LinkerScript::output(InputSection *S) {
703 uint64_t Before = advance(0, 1);
704 uint64_t Pos = advance(S->getSize(), S->Alignment);
705 S->OutSecOff = Pos - S->getSize() - Ctx->OutSec->Addr;
707 // Update output section size after adding each section. This is so that
708 // SIZEOF works correctly in the case below:
709 // .foo { *(.aaa) a = SIZEOF(.foo); *(.bbb) }
710 expandOutputSection(Pos - Before);
713 void LinkerScript::switchTo(OutputSection *Sec) {
716 uint64_t Before = advance(0, 1);
717 Ctx->OutSec->Addr = advance(0, Ctx->OutSec->Alignment);
718 expandMemoryRegions(Ctx->OutSec->Addr - Before);
721 // This function searches for a memory region to place the given output
722 // section in. If found, a pointer to the appropriate memory region is
723 // returned. Otherwise, a nullptr is returned.
724 MemoryRegion *LinkerScript::findMemoryRegion(OutputSection *Sec) {
725 // If a memory region name was specified in the output section command,
726 // then try to find that region first.
727 if (!Sec->MemoryRegionName.empty()) {
728 if (MemoryRegion *M = MemoryRegions.lookup(Sec->MemoryRegionName))
730 error("memory region '" + Sec->MemoryRegionName + "' not declared");
734 // If at least one memory region is defined, all sections must
735 // belong to some memory region. Otherwise, we don't need to do
736 // anything for memory regions.
737 if (MemoryRegions.empty())
740 // See if a region can be found by matching section flags.
741 for (auto &Pair : MemoryRegions) {
742 MemoryRegion *M = Pair.second;
743 if ((M->Flags & Sec->Flags) && (M->NegFlags & Sec->Flags) == 0)
747 // Otherwise, no suitable region was found.
748 if (Sec->Flags & SHF_ALLOC)
749 error("no memory region specified for section '" + Sec->Name + "'");
753 // This function assigns offsets to input sections and an output section
754 // for a single sections command (e.g. ".text { *(.text); }").
755 void LinkerScript::assignOffsets(OutputSection *Sec) {
756 if (!(Sec->Flags & SHF_ALLOC))
758 else if (Sec->AddrExpr)
759 setDot(Sec->AddrExpr, Sec->Location, false);
761 Ctx->MemRegion = Sec->MemRegion;
762 Ctx->LMARegion = Sec->LMARegion;
764 Dot = Ctx->MemRegion->CurPos;
769 Ctx->LMAOffset = Sec->LMAExpr().getValue() - Dot;
771 if (MemoryRegion *MR = Sec->LMARegion)
772 Ctx->LMAOffset = MR->CurPos - Dot;
774 // If neither AT nor AT> is specified for an allocatable section, the linker
775 // will set the LMA such that the difference between VMA and LMA for the
776 // section is the same as the preceding output section in the same region
777 // https://sourceware.org/binutils/docs-2.20/ld/Output-Section-LMA.html
778 if (PhdrEntry *L = Ctx->OutSec->PtLoad)
779 L->LMAOffset = Ctx->LMAOffset;
781 // We can call this method multiple times during the creation of
782 // thunks and want to start over calculation each time.
785 // We visited SectionsCommands from processSectionCommands to
786 // layout sections. Now, we visit SectionsCommands again to fix
788 for (BaseCommand *Base : Sec->SectionCommands) {
789 // This handles the assignments to symbol or to the dot.
790 if (auto *Cmd = dyn_cast<SymbolAssignment>(Base)) {
792 assignSymbol(Cmd, true);
793 Cmd->Size = Dot - Cmd->Addr;
797 // Handle BYTE(), SHORT(), LONG(), or QUAD().
798 if (auto *Cmd = dyn_cast<ByteCommand>(Base)) {
799 Cmd->Offset = Dot - Ctx->OutSec->Addr;
801 expandOutputSection(Cmd->Size);
805 // Handle a single input section description command.
806 // It calculates and assigns the offsets for each section and also
807 // updates the output section size.
808 auto *Cmd = cast<InputSectionDescription>(Base);
809 for (InputSection *Sec : Cmd->Sections) {
810 // We tentatively added all synthetic sections at the beginning and
811 // removed empty ones afterwards (because there is no way to know
812 // whether they were going be empty or not other than actually running
813 // linker scripts.) We need to ignore remains of empty sections.
814 if (auto *S = dyn_cast<SyntheticSection>(Sec))
820 assert(Ctx->OutSec == Sec->getParent());
826 static bool isDiscardable(OutputSection &Sec) {
827 // We do not remove empty sections that are explicitly
828 // assigned to any segment.
829 if (!Sec.Phdrs.empty())
832 // We do not want to remove sections that reference symbols in address and
833 // other expressions. We add script symbols as undefined, and want to ensure
834 // all of them are defined in the output, hence have to keep them.
835 if (Sec.ExpressionsUseSymbols)
838 for (BaseCommand *Base : Sec.SectionCommands) {
839 if (auto Cmd = dyn_cast<SymbolAssignment>(Base))
840 // Don't create empty output sections just for unreferenced PROVIDE
842 if (Cmd->Name != "." && !Cmd->Sym)
845 if (!isa<InputSectionDescription>(*Base))
851 void LinkerScript::adjustSectionsBeforeSorting() {
852 // If the output section contains only symbol assignments, create a
853 // corresponding output section. The issue is what to do with linker script
854 // like ".foo : { symbol = 42; }". One option would be to convert it to
855 // "symbol = 42;". That is, move the symbol out of the empty section
856 // description. That seems to be what bfd does for this simple case. The
857 // problem is that this is not completely general. bfd will give up and
858 // create a dummy section too if there is a ". = . + 1" inside the section
860 // Given that we want to create the section, we have to worry what impact
861 // it will have on the link. For example, if we just create a section with
862 // 0 for flags, it would change which PT_LOADs are created.
863 // We could remember that particular section is dummy and ignore it in
864 // other parts of the linker, but unfortunately there are quite a few places
865 // that would need to change:
866 // * The program header creation.
867 // * The orphan section placement.
868 // * The address assignment.
869 // The other option is to pick flags that minimize the impact the section
870 // will have on the rest of the linker. That is why we copy the flags from
871 // the previous sections. Only a few flags are needed to keep the impact low.
872 uint64_t Flags = SHF_ALLOC;
874 for (BaseCommand *&Cmd : SectionCommands) {
875 auto *Sec = dyn_cast<OutputSection>(Cmd);
879 // Handle align (e.g. ".foo : ALIGN(16) { ... }").
882 std::max<uint32_t>(Sec->Alignment, Sec->AlignExpr().getValue());
884 // A live output section means that some input section was added to it. It
885 // might have been removed (if it was empty synthetic section), but we at
886 // least know the flags.
890 // We do not want to keep any special flags for output section
891 // in case it is empty.
892 bool IsEmpty = getInputSections(Sec).empty();
894 Sec->Flags = Flags & (SHF_ALLOC | SHF_WRITE | SHF_EXECINSTR);
896 if (IsEmpty && isDiscardable(*Sec)) {
902 // It is common practice to use very generic linker scripts. So for any
903 // given run some of the output sections in the script will be empty.
904 // We could create corresponding empty output sections, but that would
905 // clutter the output.
906 // We instead remove trivially empty sections. The bfd linker seems even
907 // more aggressive at removing them.
908 llvm::erase_if(SectionCommands, [&](BaseCommand *Base) { return !Base; });
911 void LinkerScript::adjustSectionsAfterSorting() {
912 // Try and find an appropriate memory region to assign offsets in.
913 for (BaseCommand *Base : SectionCommands) {
914 if (auto *Sec = dyn_cast<OutputSection>(Base)) {
915 if (!Sec->LMARegionName.empty()) {
916 if (MemoryRegion *M = MemoryRegions.lookup(Sec->LMARegionName))
919 error("memory region '" + Sec->LMARegionName + "' not declared");
921 Sec->MemRegion = findMemoryRegion(Sec);
925 // If output section command doesn't specify any segments,
926 // and we haven't previously assigned any section to segment,
927 // then we simply assign section to the very first load segment.
928 // Below is an example of such linker script:
929 // PHDRS { seg PT_LOAD; }
930 // SECTIONS { .aaa : { *(.aaa) } }
931 std::vector<StringRef> DefPhdrs;
932 auto FirstPtLoad = llvm::find_if(PhdrsCommands, [](const PhdrsCommand &Cmd) {
933 return Cmd.Type == PT_LOAD;
935 if (FirstPtLoad != PhdrsCommands.end())
936 DefPhdrs.push_back(FirstPtLoad->Name);
938 // Walk the commands and propagate the program headers to commands that don't
939 // explicitly specify them.
940 for (BaseCommand *Base : SectionCommands) {
941 auto *Sec = dyn_cast<OutputSection>(Base);
945 if (Sec->Phdrs.empty()) {
946 // To match the bfd linker script behaviour, only propagate program
947 // headers to sections that are allocated.
948 if (Sec->Flags & SHF_ALLOC)
949 Sec->Phdrs = DefPhdrs;
951 DefPhdrs = Sec->Phdrs;
956 static OutputSection *findFirstSection(PhdrEntry *Load) {
957 for (OutputSection *Sec : OutputSections)
958 if (Sec->PtLoad == Load)
963 static uint64_t computeBase(uint64_t Min, bool AllocateHeaders) {
964 // If there is no SECTIONS or if the linkerscript is explicit about program
965 // headers, do our best to allocate them.
966 if (!Script->HasSectionsCommand || AllocateHeaders)
968 // Otherwise only allocate program headers if that would not add a page.
969 return alignDown(Min, Config->MaxPageSize);
972 // Try to find an address for the file and program headers output sections,
973 // which were unconditionally added to the first PT_LOAD segment earlier.
975 // When using the default layout, we check if the headers fit below the first
976 // allocated section. When using a linker script, we also check if the headers
977 // are covered by the output section. This allows omitting the headers by not
978 // leaving enough space for them in the linker script; this pattern is common
979 // in embedded systems.
981 // If there isn't enough space for these sections, we'll remove them from the
982 // PT_LOAD segment, and we'll also remove the PT_PHDR segment.
983 void LinkerScript::allocateHeaders(std::vector<PhdrEntry *> &Phdrs) {
984 uint64_t Min = std::numeric_limits<uint64_t>::max();
985 for (OutputSection *Sec : OutputSections)
986 if (Sec->Flags & SHF_ALLOC)
987 Min = std::min<uint64_t>(Min, Sec->Addr);
989 auto It = llvm::find_if(
990 Phdrs, [](const PhdrEntry *E) { return E->p_type == PT_LOAD; });
991 if (It == Phdrs.end())
993 PhdrEntry *FirstPTLoad = *It;
995 bool HasExplicitHeaders =
996 llvm::any_of(PhdrsCommands, [](const PhdrsCommand &Cmd) {
997 return Cmd.HasPhdrs || Cmd.HasFilehdr;
999 uint64_t HeaderSize = getHeaderSize();
1000 if (HeaderSize <= Min - computeBase(Min, HasExplicitHeaders)) {
1001 Min = alignDown(Min - HeaderSize, Config->MaxPageSize);
1002 Out::ElfHeader->Addr = Min;
1003 Out::ProgramHeaders->Addr = Min + Out::ElfHeader->Size;
1007 // Error if we were explicitly asked to allocate headers.
1008 if (HasExplicitHeaders)
1009 error("could not allocate headers");
1011 Out::ElfHeader->PtLoad = nullptr;
1012 Out::ProgramHeaders->PtLoad = nullptr;
1013 FirstPTLoad->FirstSec = findFirstSection(FirstPTLoad);
1015 llvm::erase_if(Phdrs,
1016 [](const PhdrEntry *E) { return E->p_type == PT_PHDR; });
1019 LinkerScript::AddressState::AddressState() {
1020 for (auto &MRI : Script->MemoryRegions) {
1021 MemoryRegion *MR = MRI.second;
1022 MR->CurPos = MR->Origin;
1026 static uint64_t getInitialDot() {
1027 // By default linker scripts use an initial value of 0 for '.',
1028 // but prefer -image-base if set.
1029 if (Script->HasSectionsCommand)
1030 return Config->ImageBase ? *Config->ImageBase : 0;
1032 uint64_t StartAddr = UINT64_MAX;
1033 // The Sections with -T<section> have been sorted in order of ascending
1034 // address. We must lower StartAddr if the lowest -T<section address> as
1035 // calls to setDot() must be monotonically increasing.
1036 for (auto &KV : Config->SectionStartMap)
1037 StartAddr = std::min(StartAddr, KV.second);
1038 return std::min(StartAddr, Target->getImageBase() + elf::getHeaderSize());
1041 // Here we assign addresses as instructed by linker script SECTIONS
1042 // sub-commands. Doing that allows us to use final VA values, so here
1043 // we also handle rest commands like symbol assignments and ASSERTs.
1044 void LinkerScript::assignAddresses() {
1045 Dot = getInitialDot();
1047 auto Deleter = make_unique<AddressState>();
1048 Ctx = Deleter.get();
1049 ErrorOnMissingSection = true;
1052 for (BaseCommand *Base : SectionCommands) {
1053 if (auto *Cmd = dyn_cast<SymbolAssignment>(Base)) {
1055 assignSymbol(Cmd, false);
1056 Cmd->Size = Dot - Cmd->Addr;
1059 assignOffsets(cast<OutputSection>(Base));
1064 // Creates program headers as instructed by PHDRS linker script command.
1065 std::vector<PhdrEntry *> LinkerScript::createPhdrs() {
1066 std::vector<PhdrEntry *> Ret;
1068 // Process PHDRS and FILEHDR keywords because they are not
1069 // real output sections and cannot be added in the following loop.
1070 for (const PhdrsCommand &Cmd : PhdrsCommands) {
1071 PhdrEntry *Phdr = make<PhdrEntry>(Cmd.Type, Cmd.Flags ? *Cmd.Flags : PF_R);
1074 Phdr->add(Out::ElfHeader);
1076 Phdr->add(Out::ProgramHeaders);
1079 Phdr->p_paddr = Cmd.LMAExpr().getValue();
1080 Phdr->HasLMA = true;
1082 Ret.push_back(Phdr);
1085 // Add output sections to program headers.
1086 for (OutputSection *Sec : OutputSections) {
1087 // Assign headers specified by linker script
1088 for (size_t Id : getPhdrIndices(Sec)) {
1090 if (!PhdrsCommands[Id].Flags.hasValue())
1091 Ret[Id]->p_flags |= Sec->getPhdrFlags();
1097 // Returns true if we should emit an .interp section.
1099 // We usually do. But if PHDRS commands are given, and
1100 // no PT_INTERP is there, there's no place to emit an
1101 // .interp, so we don't do that in that case.
1102 bool LinkerScript::needsInterpSection() {
1103 if (PhdrsCommands.empty())
1105 for (PhdrsCommand &Cmd : PhdrsCommands)
1106 if (Cmd.Type == PT_INTERP)
1111 ExprValue LinkerScript::getSymbolValue(StringRef Name, const Twine &Loc) {
1114 return {Ctx->OutSec, false, Dot - Ctx->OutSec->Addr, Loc};
1115 error(Loc + ": unable to get location counter value");
1119 if (Symbol *Sym = Symtab->find(Name)) {
1120 if (auto *DS = dyn_cast<Defined>(Sym))
1121 return {DS->Section, false, DS->Value, Loc};
1122 if (isa<SharedSymbol>(Sym))
1123 if (!ErrorOnMissingSection)
1124 return {nullptr, false, 0, Loc};
1127 error(Loc + ": symbol not found: " + Name);
1131 // Returns the index of the segment named Name.
1132 static Optional<size_t> getPhdrIndex(ArrayRef<PhdrsCommand> Vec,
1134 for (size_t I = 0; I < Vec.size(); ++I)
1135 if (Vec[I].Name == Name)
1140 // Returns indices of ELF headers containing specific section. Each index is a
1141 // zero based number of ELF header listed within PHDRS {} script block.
1142 std::vector<size_t> LinkerScript::getPhdrIndices(OutputSection *Cmd) {
1143 std::vector<size_t> Ret;
1145 for (StringRef S : Cmd->Phdrs) {
1146 if (Optional<size_t> Idx = getPhdrIndex(PhdrsCommands, S))
1147 Ret.push_back(*Idx);
1148 else if (S != "NONE")
1149 error(Cmd->Location + ": section header '" + S +
1150 "' is not listed in PHDRS");