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,
119 // Only expand the LMARegion if it is different from MemRegion.
120 if (Ctx->LMARegion && Ctx->MemRegion != Ctx->LMARegion)
121 expandMemoryRegion(Ctx->LMARegion, Size, Ctx->LMARegion->Name,
125 void LinkerScript::expandOutputSection(uint64_t Size) {
126 Ctx->OutSec->Size += Size;
127 expandMemoryRegions(Size);
130 void LinkerScript::setDot(Expr E, const Twine &Loc, bool InSec) {
131 uint64_t Val = E().getValue();
132 if (Val < Dot && InSec)
133 error(Loc + ": unable to move location counter backward for: " +
136 // Update to location counter means update to section size.
138 expandOutputSection(Val - Dot);
140 expandMemoryRegions(Val - Dot);
145 // Used for handling linker symbol assignments, for both finalizing
146 // their values and doing early declarations. Returns true if symbol
147 // should be defined from linker script.
148 static bool shouldDefineSym(SymbolAssignment *Cmd) {
149 if (Cmd->Name == ".")
155 // If a symbol was in PROVIDE(), we need to define it only
156 // when it is a referenced undefined symbol.
157 Symbol *B = Symtab->find(Cmd->Name);
158 if (B && !B->isDefined())
163 // This function is called from processSectionCommands,
164 // while we are fixing the output section layout.
165 void LinkerScript::addSymbol(SymbolAssignment *Cmd) {
166 if (!shouldDefineSym(Cmd))
171 uint8_t Visibility = Cmd->Hidden ? STV_HIDDEN : STV_DEFAULT;
172 std::tie(Sym, std::ignore) = Symtab->insert(Cmd->Name, /*Type*/ 0, Visibility,
173 /*CanOmitFromDynSym*/ false,
175 ExprValue Value = Cmd->Expression();
176 SectionBase *Sec = Value.isAbsolute() ? nullptr : Value.Sec;
178 // When this function is called, section addresses have not been
179 // fixed yet. So, we may or may not know the value of the RHS
182 // For example, if an expression is `x = 42`, we know x is always 42.
183 // However, if an expression is `x = .`, there's no way to know its
184 // value at the moment.
186 // We want to set symbol values early if we can. This allows us to
187 // use symbols as variables in linker scripts. Doing so allows us to
188 // write expressions like this: `alignment = 16; . = ALIGN(., alignment)`.
189 uint64_t SymValue = Value.Sec ? 0 : Value.getValue();
191 replaceSymbol<Defined>(Sym, nullptr, Cmd->Name, STB_GLOBAL, Visibility,
192 STT_NOTYPE, SymValue, 0, Sec);
193 Cmd->Sym = cast<Defined>(Sym);
196 // This function is called from LinkerScript::declareSymbols.
197 // It creates a placeholder symbol if needed.
198 static void declareSymbol(SymbolAssignment *Cmd) {
199 if (!shouldDefineSym(Cmd))
202 // We can't calculate final value right now.
204 uint8_t Visibility = Cmd->Hidden ? STV_HIDDEN : STV_DEFAULT;
205 std::tie(Sym, std::ignore) = Symtab->insert(Cmd->Name, /*Type*/ 0, Visibility,
206 /*CanOmitFromDynSym*/ false,
208 replaceSymbol<Defined>(Sym, nullptr, Cmd->Name, STB_GLOBAL, Visibility,
209 STT_NOTYPE, 0, 0, nullptr);
210 Cmd->Sym = cast<Defined>(Sym);
211 Cmd->Provide = false;
214 // This method is used to handle INSERT AFTER statement. Here we rebuild
215 // the list of script commands to mix sections inserted into.
216 void LinkerScript::processInsertCommands() {
217 std::vector<BaseCommand *> V;
218 auto Insert = [&](std::vector<BaseCommand *> &From) {
219 V.insert(V.end(), From.begin(), From.end());
223 for (BaseCommand *Base : SectionCommands) {
224 if (auto *OS = dyn_cast<OutputSection>(Base)) {
225 Insert(InsertBeforeCommands[OS->Name]);
227 Insert(InsertAfterCommands[OS->Name]);
233 for (auto &Cmds : {InsertBeforeCommands, InsertAfterCommands})
234 for (const std::pair<StringRef, std::vector<BaseCommand *>> &P : Cmds)
235 if (!P.second.empty())
236 error("unable to INSERT AFTER/BEFORE " + P.first +
237 ": section not defined");
239 SectionCommands = std::move(V);
242 // Symbols defined in script should not be inlined by LTO. At the same time
243 // we don't know their final values until late stages of link. Here we scan
244 // over symbol assignment commands and create placeholder symbols if needed.
245 void LinkerScript::declareSymbols() {
247 for (BaseCommand *Base : SectionCommands) {
248 if (auto *Cmd = dyn_cast<SymbolAssignment>(Base)) {
253 // If the output section directive has constraints,
254 // we can't say for sure if it is going to be included or not.
255 // Skip such sections for now. Improve the checks if we ever
256 // need symbols from that sections to be declared early.
257 auto *Sec = cast<OutputSection>(Base);
258 if (Sec->Constraint != ConstraintKind::NoConstraint)
260 for (BaseCommand *Base2 : Sec->SectionCommands)
261 if (auto *Cmd = dyn_cast<SymbolAssignment>(Base2))
266 // This function is called from assignAddresses, while we are
267 // fixing the output section addresses. This function is supposed
268 // to set the final value for a given symbol assignment.
269 void LinkerScript::assignSymbol(SymbolAssignment *Cmd, bool InSec) {
270 if (Cmd->Name == ".") {
271 setDot(Cmd->Expression, Cmd->Location, InSec);
278 ExprValue V = Cmd->Expression();
279 if (V.isAbsolute()) {
280 Cmd->Sym->Section = nullptr;
281 Cmd->Sym->Value = V.getValue();
283 Cmd->Sym->Section = V.Sec;
284 Cmd->Sym->Value = V.getSectionOffset();
288 static std::string getFilename(InputFile *File) {
291 if (File->ArchiveName.empty())
292 return File->getName();
293 return (File->ArchiveName + "(" + File->getName() + ")").str();
296 bool LinkerScript::shouldKeep(InputSectionBase *S) {
297 if (KeptSections.empty())
299 std::string Filename = getFilename(S->File);
300 for (InputSectionDescription *ID : KeptSections)
301 if (ID->FilePat.match(Filename))
302 for (SectionPattern &P : ID->SectionPatterns)
303 if (P.SectionPat.match(S->Name))
308 // A helper function for the SORT() command.
309 static std::function<bool(InputSectionBase *, InputSectionBase *)>
310 getComparator(SortSectionPolicy K) {
312 case SortSectionPolicy::Alignment:
313 return [](InputSectionBase *A, InputSectionBase *B) {
314 // ">" is not a mistake. Sections with larger alignments are placed
315 // before sections with smaller alignments in order to reduce the
316 // amount of padding necessary. This is compatible with GNU.
317 return A->Alignment > B->Alignment;
319 case SortSectionPolicy::Name:
320 return [](InputSectionBase *A, InputSectionBase *B) {
321 return A->Name < B->Name;
323 case SortSectionPolicy::Priority:
324 return [](InputSectionBase *A, InputSectionBase *B) {
325 return getPriority(A->Name) < getPriority(B->Name);
328 llvm_unreachable("unknown sort policy");
332 // A helper function for the SORT() command.
333 static bool matchConstraints(ArrayRef<InputSection *> Sections,
334 ConstraintKind Kind) {
335 if (Kind == ConstraintKind::NoConstraint)
338 bool IsRW = llvm::any_of(
339 Sections, [](InputSection *Sec) { return Sec->Flags & SHF_WRITE; });
341 return (IsRW && Kind == ConstraintKind::ReadWrite) ||
342 (!IsRW && Kind == ConstraintKind::ReadOnly);
345 static void sortSections(MutableArrayRef<InputSection *> Vec,
346 SortSectionPolicy K) {
347 if (K != SortSectionPolicy::Default && K != SortSectionPolicy::None)
348 std::stable_sort(Vec.begin(), Vec.end(), getComparator(K));
351 // Sort sections as instructed by SORT-family commands and --sort-section
352 // option. Because SORT-family commands can be nested at most two depth
353 // (e.g. SORT_BY_NAME(SORT_BY_ALIGNMENT(.text.*))) and because the command
354 // line option is respected even if a SORT command is given, the exact
355 // behavior we have here is a bit complicated. Here are the rules.
357 // 1. If two SORT commands are given, --sort-section is ignored.
358 // 2. If one SORT command is given, and if it is not SORT_NONE,
359 // --sort-section is handled as an inner SORT command.
360 // 3. If one SORT command is given, and if it is SORT_NONE, don't sort.
361 // 4. If no SORT command is given, sort according to --sort-section.
362 static void sortInputSections(MutableArrayRef<InputSection *> Vec,
363 const SectionPattern &Pat) {
364 if (Pat.SortOuter == SortSectionPolicy::None)
367 if (Pat.SortInner == SortSectionPolicy::Default)
368 sortSections(Vec, Config->SortSection);
370 sortSections(Vec, Pat.SortInner);
371 sortSections(Vec, Pat.SortOuter);
374 // Compute and remember which sections the InputSectionDescription matches.
375 std::vector<InputSection *>
376 LinkerScript::computeInputSections(const InputSectionDescription *Cmd) {
377 std::vector<InputSection *> Ret;
379 // Collects all sections that satisfy constraints of Cmd.
380 for (const SectionPattern &Pat : Cmd->SectionPatterns) {
381 size_t SizeBefore = Ret.size();
383 for (InputSectionBase *Sec : InputSections) {
384 if (!Sec->Live || Sec->Assigned)
387 // For -emit-relocs we have to ignore entries like
388 // .rela.dyn : { *(.rela.data) }
389 // which are common because they are in the default bfd script.
390 // We do not ignore SHT_REL[A] linker-synthesized sections here because
391 // want to support scripts that do custom layout for them.
392 if (auto *IS = dyn_cast<InputSection>(Sec))
393 if (IS->getRelocatedSection())
396 std::string Filename = getFilename(Sec->File);
397 if (!Cmd->FilePat.match(Filename) ||
398 Pat.ExcludedFilePat.match(Filename) ||
399 !Pat.SectionPat.match(Sec->Name))
402 // It is safe to assume that Sec is an InputSection
403 // because mergeable or EH input sections have already been
404 // handled and eliminated.
405 Ret.push_back(cast<InputSection>(Sec));
406 Sec->Assigned = true;
409 sortInputSections(MutableArrayRef<InputSection *>(Ret).slice(SizeBefore),
415 void LinkerScript::discard(ArrayRef<InputSection *> V) {
416 for (InputSection *S : V) {
417 if (S == InX::ShStrTab || S == InX::Dynamic || S == InX::DynSymTab ||
418 S == InX::DynStrTab || S == InX::RelaPlt || S == InX::RelaDyn ||
420 error("discarding " + S->Name + " section is not allowed");
422 // You can discard .hash and .gnu.hash sections by linker scripts. Since
423 // they are synthesized sections, we need to handle them differently than
424 // other regular sections.
425 if (S == InX::GnuHashTab)
426 InX::GnuHashTab = nullptr;
427 if (S == InX::HashTab)
428 InX::HashTab = nullptr;
432 discard(S->DependentSections);
436 std::vector<InputSection *>
437 LinkerScript::createInputSectionList(OutputSection &OutCmd) {
438 std::vector<InputSection *> Ret;
440 for (BaseCommand *Base : OutCmd.SectionCommands) {
441 if (auto *Cmd = dyn_cast<InputSectionDescription>(Base)) {
442 Cmd->Sections = computeInputSections(Cmd);
443 Ret.insert(Ret.end(), Cmd->Sections.begin(), Cmd->Sections.end());
449 void LinkerScript::processSectionCommands() {
450 // A symbol can be assigned before any section is mentioned in the linker
451 // script. In an DSO, the symbol values are addresses, so the only important
452 // section values are:
455 // * Any value meaning a regular section.
456 // To handle that, create a dummy aether section that fills the void before
457 // the linker scripts switches to another section. It has an index of one
458 // which will map to whatever the first actual section is.
459 Aether = make<OutputSection>("", 0, SHF_ALLOC);
460 Aether->SectionIndex = 1;
462 // Ctx captures the local AddressState and makes it accessible deliberately.
463 // This is needed as there are some cases where we cannot just
464 // thread the current state through to a lambda function created by the
466 auto Deleter = make_unique<AddressState>();
468 Ctx->OutSec = Aether;
471 // Add input sections to output sections.
472 for (BaseCommand *Base : SectionCommands) {
473 // Handle symbol assignments outside of any output section.
474 if (auto *Cmd = dyn_cast<SymbolAssignment>(Base)) {
479 if (auto *Sec = dyn_cast<OutputSection>(Base)) {
480 std::vector<InputSection *> V = createInputSectionList(*Sec);
482 // The output section name `/DISCARD/' is special.
483 // Any input section assigned to it is discarded.
484 if (Sec->Name == "/DISCARD/") {
486 Sec->SectionCommands.clear();
490 // This is for ONLY_IF_RO and ONLY_IF_RW. An output section directive
491 // ".foo : ONLY_IF_R[OW] { ... }" is handled only if all member input
492 // sections satisfy a given constraint. If not, a directive is handled
493 // as if it wasn't present from the beginning.
495 // Because we'll iterate over SectionCommands many more times, the easy
496 // way to "make it as if it wasn't present" is to make it empty.
497 if (!matchConstraints(V, Sec->Constraint)) {
498 for (InputSectionBase *S : V)
500 Sec->SectionCommands.clear();
504 // A directive may contain symbol definitions like this:
505 // ".foo : { ...; bar = .; }". Handle them.
506 for (BaseCommand *Base : Sec->SectionCommands)
507 if (auto *OutCmd = dyn_cast<SymbolAssignment>(Base))
510 // Handle subalign (e.g. ".foo : SUBALIGN(32) { ... }"). If subalign
511 // is given, input sections are aligned to that value, whether the
512 // given value is larger or smaller than the original section alignment.
513 if (Sec->SubalignExpr) {
514 uint32_t Subalign = Sec->SubalignExpr().getValue();
515 for (InputSectionBase *S : V)
516 S->Alignment = Subalign;
519 // Add input sections to an output section.
520 for (InputSection *S : V)
523 Sec->SectionIndex = I++;
525 Sec->Type = SHT_NOBITS;
527 Sec->Flags &= ~(uint64_t)SHF_ALLOC;
533 static OutputSection *findByName(ArrayRef<BaseCommand *> Vec,
535 for (BaseCommand *Base : Vec)
536 if (auto *Sec = dyn_cast<OutputSection>(Base))
537 if (Sec->Name == Name)
542 static OutputSection *createSection(InputSectionBase *IS,
543 StringRef OutsecName) {
544 OutputSection *Sec = Script->createOutputSection(OutsecName, "<internal>");
545 Sec->addSection(cast<InputSection>(IS));
549 static OutputSection *addInputSec(StringMap<OutputSection *> &Map,
550 InputSectionBase *IS, StringRef OutsecName) {
551 // Sections with SHT_GROUP or SHF_GROUP attributes reach here only when the -r
552 // option is given. A section with SHT_GROUP defines a "section group", and
553 // its members have SHF_GROUP attribute. Usually these flags have already been
554 // stripped by InputFiles.cpp as section groups are processed and uniquified.
555 // However, for the -r option, we want to pass through all section groups
556 // as-is because adding/removing members or merging them with other groups
557 // change their semantics.
558 if (IS->Type == SHT_GROUP || (IS->Flags & SHF_GROUP))
559 return createSection(IS, OutsecName);
561 // Imagine .zed : { *(.foo) *(.bar) } script. Both foo and bar may have
562 // relocation sections .rela.foo and .rela.bar for example. Most tools do
563 // not allow multiple REL[A] sections for output section. Hence we
564 // should combine these relocation sections into single output.
565 // We skip synthetic sections because it can be .rela.dyn/.rela.plt or any
566 // other REL[A] sections created by linker itself.
567 if (!isa<SyntheticSection>(IS) &&
568 (IS->Type == SHT_REL || IS->Type == SHT_RELA)) {
569 auto *Sec = cast<InputSection>(IS);
570 OutputSection *Out = Sec->getRelocatedSection()->getOutputSection();
572 if (Out->RelocationSection) {
573 Out->RelocationSection->addSection(Sec);
577 Out->RelocationSection = createSection(IS, OutsecName);
578 return Out->RelocationSection;
581 // When control reaches here, mergeable sections have already been merged into
582 // synthetic sections. For relocatable case we want to create one output
583 // section per syntetic section so that they have a valid sh_entsize.
584 if (Config->Relocatable && (IS->Flags & SHF_MERGE))
585 return createSection(IS, OutsecName);
587 // The ELF spec just says
588 // ----------------------------------------------------------------
589 // In the first phase, input sections that match in name, type and
590 // attribute flags should be concatenated into single sections.
591 // ----------------------------------------------------------------
593 // However, it is clear that at least some flags have to be ignored for
594 // section merging. At the very least SHF_GROUP and SHF_COMPRESSED have to be
595 // ignored. We should not have two output .text sections just because one was
596 // in a group and another was not for example.
598 // It also seems that wording was a late addition and didn't get the
599 // necessary scrutiny.
601 // Merging sections with different flags is expected by some users. One
602 // reason is that if one file has
604 // int *const bar __attribute__((section(".foo"))) = (int *)0;
606 // gcc with -fPIC will produce a read only .foo section. But if another
610 // int *const bar __attribute__((section(".foo"))) = (int *)&zed;
612 // gcc with -fPIC will produce a read write section.
614 // Last but not least, when using linker script the merge rules are forced by
615 // the script. Unfortunately, linker scripts are name based. This means that
616 // expressions like *(.foo*) can refer to multiple input sections with
617 // different flags. We cannot put them in different output sections or we
618 // would produce wrong results for
620 // start = .; *(.foo.*) end = .; *(.bar)
622 // and a mapping of .foo1 and .bar1 to one section and .foo2 and .bar2 to
623 // another. The problem is that there is no way to layout those output
624 // sections such that the .foo sections are the only thing between the start
627 // Given the above issues, we instead merge sections by name and error on
628 // incompatible types and flags.
629 OutputSection *&Sec = Map[OutsecName];
631 Sec->addSection(cast<InputSection>(IS));
635 Sec = createSection(IS, OutsecName);
639 // Add sections that didn't match any sections command.
640 void LinkerScript::addOrphanSections() {
641 unsigned End = SectionCommands.size();
642 StringMap<OutputSection *> Map;
643 std::vector<OutputSection *> V;
645 auto Add = [&](InputSectionBase *S) {
646 if (!S->Live || S->Parent)
649 StringRef Name = getOutputSectionName(S);
651 if (Config->OrphanHandling == OrphanHandlingPolicy::Error)
652 error(toString(S) + " is being placed in '" + Name + "'");
653 else if (Config->OrphanHandling == OrphanHandlingPolicy::Warn)
654 warn(toString(S) + " is being placed in '" + Name + "'");
656 if (OutputSection *Sec =
657 findByName(makeArrayRef(SectionCommands).slice(0, End), Name)) {
658 Sec->addSection(cast<InputSection>(S));
662 if (OutputSection *OS = addInputSec(Map, S, Name))
664 assert(S->getOutputSection()->SectionIndex == UINT32_MAX);
667 // For futher --emit-reloc handling code we need target output section
668 // to be created before we create relocation output section, so we want
669 // to create target sections first. We do not want priority handling
670 // for synthetic sections because them are special.
671 for (InputSectionBase *IS : InputSections) {
672 if (auto *Sec = dyn_cast<InputSection>(IS))
673 if (InputSectionBase *Rel = Sec->getRelocatedSection())
674 if (auto *RelIS = dyn_cast_or_null<InputSectionBase>(Rel->Parent))
679 // If no SECTIONS command was given, we should insert sections commands
680 // before others, so that we can handle scripts which refers them,
681 // for example: "foo = ABSOLUTE(ADDR(.text)));".
682 // When SECTIONS command is present we just add all orphans to the end.
683 if (HasSectionsCommand)
684 SectionCommands.insert(SectionCommands.end(), V.begin(), V.end());
686 SectionCommands.insert(SectionCommands.begin(), V.begin(), V.end());
689 uint64_t LinkerScript::advance(uint64_t Size, unsigned Alignment) {
691 (Ctx->OutSec->Flags & SHF_TLS) && Ctx->OutSec->Type == SHT_NOBITS;
692 uint64_t Start = IsTbss ? Dot + Ctx->ThreadBssOffset : Dot;
693 Start = alignTo(Start, Alignment);
694 uint64_t End = Start + Size;
697 Ctx->ThreadBssOffset = End - Dot;
703 void LinkerScript::output(InputSection *S) {
704 uint64_t Before = advance(0, 1);
705 uint64_t Pos = advance(S->getSize(), S->Alignment);
706 S->OutSecOff = Pos - S->getSize() - Ctx->OutSec->Addr;
708 // Update output section size after adding each section. This is so that
709 // SIZEOF works correctly in the case below:
710 // .foo { *(.aaa) a = SIZEOF(.foo); *(.bbb) }
711 expandOutputSection(Pos - Before);
714 void LinkerScript::switchTo(OutputSection *Sec) {
717 uint64_t Before = advance(0, 1);
718 Ctx->OutSec->Addr = advance(0, Ctx->OutSec->Alignment);
719 expandMemoryRegions(Ctx->OutSec->Addr - Before);
722 // This function searches for a memory region to place the given output
723 // section in. If found, a pointer to the appropriate memory region is
724 // returned. Otherwise, a nullptr is returned.
725 MemoryRegion *LinkerScript::findMemoryRegion(OutputSection *Sec) {
726 // If a memory region name was specified in the output section command,
727 // then try to find that region first.
728 if (!Sec->MemoryRegionName.empty()) {
729 if (MemoryRegion *M = MemoryRegions.lookup(Sec->MemoryRegionName))
731 error("memory region '" + Sec->MemoryRegionName + "' not declared");
735 // If at least one memory region is defined, all sections must
736 // belong to some memory region. Otherwise, we don't need to do
737 // anything for memory regions.
738 if (MemoryRegions.empty())
741 // See if a region can be found by matching section flags.
742 for (auto &Pair : MemoryRegions) {
743 MemoryRegion *M = Pair.second;
744 if ((M->Flags & Sec->Flags) && (M->NegFlags & Sec->Flags) == 0)
748 // Otherwise, no suitable region was found.
749 if (Sec->Flags & SHF_ALLOC)
750 error("no memory region specified for section '" + Sec->Name + "'");
754 static OutputSection *findFirstSection(PhdrEntry *Load) {
755 for (OutputSection *Sec : OutputSections)
756 if (Sec->PtLoad == Load)
761 // This function assigns offsets to input sections and an output section
762 // for a single sections command (e.g. ".text { *(.text); }").
763 void LinkerScript::assignOffsets(OutputSection *Sec) {
764 if (!(Sec->Flags & SHF_ALLOC))
766 else if (Sec->AddrExpr)
767 setDot(Sec->AddrExpr, Sec->Location, false);
769 Ctx->MemRegion = Sec->MemRegion;
770 Ctx->LMARegion = Sec->LMARegion;
772 Dot = Ctx->MemRegion->CurPos;
777 Ctx->LMAOffset = Sec->LMAExpr().getValue() - Dot;
779 if (MemoryRegion *MR = Sec->LMARegion)
780 Ctx->LMAOffset = MR->CurPos - Dot;
782 // If neither AT nor AT> is specified for an allocatable section, the linker
783 // will set the LMA such that the difference between VMA and LMA for the
784 // section is the same as the preceding output section in the same region
785 // https://sourceware.org/binutils/docs-2.20/ld/Output-Section-LMA.html
786 // This, however, should only be done by the first "non-header" section
788 if (PhdrEntry *L = Ctx->OutSec->PtLoad)
789 if (Sec == findFirstSection(L))
790 L->LMAOffset = Ctx->LMAOffset;
792 // We can call this method multiple times during the creation of
793 // thunks and want to start over calculation each time.
796 // We visited SectionsCommands from processSectionCommands to
797 // layout sections. Now, we visit SectionsCommands again to fix
799 for (BaseCommand *Base : Sec->SectionCommands) {
800 // This handles the assignments to symbol or to the dot.
801 if (auto *Cmd = dyn_cast<SymbolAssignment>(Base)) {
803 assignSymbol(Cmd, true);
804 Cmd->Size = Dot - Cmd->Addr;
808 // Handle BYTE(), SHORT(), LONG(), or QUAD().
809 if (auto *Cmd = dyn_cast<ByteCommand>(Base)) {
810 Cmd->Offset = Dot - Ctx->OutSec->Addr;
812 expandOutputSection(Cmd->Size);
816 // Handle a single input section description command.
817 // It calculates and assigns the offsets for each section and also
818 // updates the output section size.
819 auto *Cmd = cast<InputSectionDescription>(Base);
820 for (InputSection *Sec : Cmd->Sections) {
821 // We tentatively added all synthetic sections at the beginning and
822 // removed empty ones afterwards (because there is no way to know
823 // whether they were going be empty or not other than actually running
824 // linker scripts.) We need to ignore remains of empty sections.
825 if (auto *S = dyn_cast<SyntheticSection>(Sec))
831 assert(Ctx->OutSec == Sec->getParent());
837 static bool isDiscardable(OutputSection &Sec) {
838 // We do not remove empty sections that are explicitly
839 // assigned to any segment.
840 if (!Sec.Phdrs.empty())
843 // We do not want to remove sections that reference symbols in address and
844 // other expressions. We add script symbols as undefined, and want to ensure
845 // all of them are defined in the output, hence have to keep them.
846 if (Sec.ExpressionsUseSymbols)
849 for (BaseCommand *Base : Sec.SectionCommands) {
850 if (auto Cmd = dyn_cast<SymbolAssignment>(Base))
851 // Don't create empty output sections just for unreferenced PROVIDE
853 if (Cmd->Name != "." && !Cmd->Sym)
856 if (!isa<InputSectionDescription>(*Base))
862 void LinkerScript::adjustSectionsBeforeSorting() {
863 // If the output section contains only symbol assignments, create a
864 // corresponding output section. The issue is what to do with linker script
865 // like ".foo : { symbol = 42; }". One option would be to convert it to
866 // "symbol = 42;". That is, move the symbol out of the empty section
867 // description. That seems to be what bfd does for this simple case. The
868 // problem is that this is not completely general. bfd will give up and
869 // create a dummy section too if there is a ". = . + 1" inside the section
871 // Given that we want to create the section, we have to worry what impact
872 // it will have on the link. For example, if we just create a section with
873 // 0 for flags, it would change which PT_LOADs are created.
874 // We could remember that particular section is dummy and ignore it in
875 // other parts of the linker, but unfortunately there are quite a few places
876 // that would need to change:
877 // * The program header creation.
878 // * The orphan section placement.
879 // * The address assignment.
880 // The other option is to pick flags that minimize the impact the section
881 // will have on the rest of the linker. That is why we copy the flags from
882 // the previous sections. Only a few flags are needed to keep the impact low.
883 uint64_t Flags = SHF_ALLOC;
885 for (BaseCommand *&Cmd : SectionCommands) {
886 auto *Sec = dyn_cast<OutputSection>(Cmd);
890 // Handle align (e.g. ".foo : ALIGN(16) { ... }").
893 std::max<uint32_t>(Sec->Alignment, Sec->AlignExpr().getValue());
895 // A live output section means that some input section was added to it. It
896 // might have been removed (if it was empty synthetic section), but we at
897 // least know the flags.
901 // We do not want to keep any special flags for output section
902 // in case it is empty.
903 bool IsEmpty = getInputSections(Sec).empty();
905 Sec->Flags = Flags & (SHF_ALLOC | SHF_WRITE | SHF_EXECINSTR);
907 if (IsEmpty && isDiscardable(*Sec)) {
913 // It is common practice to use very generic linker scripts. So for any
914 // given run some of the output sections in the script will be empty.
915 // We could create corresponding empty output sections, but that would
916 // clutter the output.
917 // We instead remove trivially empty sections. The bfd linker seems even
918 // more aggressive at removing them.
919 llvm::erase_if(SectionCommands, [&](BaseCommand *Base) { return !Base; });
922 void LinkerScript::adjustSectionsAfterSorting() {
923 // Try and find an appropriate memory region to assign offsets in.
924 for (BaseCommand *Base : SectionCommands) {
925 if (auto *Sec = dyn_cast<OutputSection>(Base)) {
926 if (!Sec->LMARegionName.empty()) {
927 if (MemoryRegion *M = MemoryRegions.lookup(Sec->LMARegionName))
930 error("memory region '" + Sec->LMARegionName + "' not declared");
932 Sec->MemRegion = findMemoryRegion(Sec);
936 // If output section command doesn't specify any segments,
937 // and we haven't previously assigned any section to segment,
938 // then we simply assign section to the very first load segment.
939 // Below is an example of such linker script:
940 // PHDRS { seg PT_LOAD; }
941 // SECTIONS { .aaa : { *(.aaa) } }
942 std::vector<StringRef> DefPhdrs;
943 auto FirstPtLoad = llvm::find_if(PhdrsCommands, [](const PhdrsCommand &Cmd) {
944 return Cmd.Type == PT_LOAD;
946 if (FirstPtLoad != PhdrsCommands.end())
947 DefPhdrs.push_back(FirstPtLoad->Name);
949 // Walk the commands and propagate the program headers to commands that don't
950 // explicitly specify them.
951 for (BaseCommand *Base : SectionCommands) {
952 auto *Sec = dyn_cast<OutputSection>(Base);
956 if (Sec->Phdrs.empty()) {
957 // To match the bfd linker script behaviour, only propagate program
958 // headers to sections that are allocated.
959 if (Sec->Flags & SHF_ALLOC)
960 Sec->Phdrs = DefPhdrs;
962 DefPhdrs = Sec->Phdrs;
967 static uint64_t computeBase(uint64_t Min, bool AllocateHeaders) {
968 // If there is no SECTIONS or if the linkerscript is explicit about program
969 // headers, do our best to allocate them.
970 if (!Script->HasSectionsCommand || AllocateHeaders)
972 // Otherwise only allocate program headers if that would not add a page.
973 return alignDown(Min, Config->MaxPageSize);
976 // Try to find an address for the file and program headers output sections,
977 // which were unconditionally added to the first PT_LOAD segment earlier.
979 // When using the default layout, we check if the headers fit below the first
980 // allocated section. When using a linker script, we also check if the headers
981 // are covered by the output section. This allows omitting the headers by not
982 // leaving enough space for them in the linker script; this pattern is common
983 // in embedded systems.
985 // If there isn't enough space for these sections, we'll remove them from the
986 // PT_LOAD segment, and we'll also remove the PT_PHDR segment.
987 void LinkerScript::allocateHeaders(std::vector<PhdrEntry *> &Phdrs) {
988 uint64_t Min = std::numeric_limits<uint64_t>::max();
989 for (OutputSection *Sec : OutputSections)
990 if (Sec->Flags & SHF_ALLOC)
991 Min = std::min<uint64_t>(Min, Sec->Addr);
993 auto It = llvm::find_if(
994 Phdrs, [](const PhdrEntry *E) { return E->p_type == PT_LOAD; });
995 if (It == Phdrs.end())
997 PhdrEntry *FirstPTLoad = *It;
999 bool HasExplicitHeaders =
1000 llvm::any_of(PhdrsCommands, [](const PhdrsCommand &Cmd) {
1001 return Cmd.HasPhdrs || Cmd.HasFilehdr;
1003 uint64_t HeaderSize = getHeaderSize();
1004 if (HeaderSize <= Min - computeBase(Min, HasExplicitHeaders)) {
1005 Min = alignDown(Min - HeaderSize, Config->MaxPageSize);
1006 Out::ElfHeader->Addr = Min;
1007 Out::ProgramHeaders->Addr = Min + Out::ElfHeader->Size;
1011 // Error if we were explicitly asked to allocate headers.
1012 if (HasExplicitHeaders)
1013 error("could not allocate headers");
1015 Out::ElfHeader->PtLoad = nullptr;
1016 Out::ProgramHeaders->PtLoad = nullptr;
1017 FirstPTLoad->FirstSec = findFirstSection(FirstPTLoad);
1019 llvm::erase_if(Phdrs,
1020 [](const PhdrEntry *E) { return E->p_type == PT_PHDR; });
1023 LinkerScript::AddressState::AddressState() {
1024 for (auto &MRI : Script->MemoryRegions) {
1025 MemoryRegion *MR = MRI.second;
1026 MR->CurPos = MR->Origin;
1030 static uint64_t getInitialDot() {
1031 // By default linker scripts use an initial value of 0 for '.',
1032 // but prefer -image-base if set.
1033 if (Script->HasSectionsCommand)
1034 return Config->ImageBase ? *Config->ImageBase : 0;
1036 uint64_t StartAddr = UINT64_MAX;
1037 // The Sections with -T<section> have been sorted in order of ascending
1038 // address. We must lower StartAddr if the lowest -T<section address> as
1039 // calls to setDot() must be monotonically increasing.
1040 for (auto &KV : Config->SectionStartMap)
1041 StartAddr = std::min(StartAddr, KV.second);
1042 return std::min(StartAddr, Target->getImageBase() + elf::getHeaderSize());
1045 // Here we assign addresses as instructed by linker script SECTIONS
1046 // sub-commands. Doing that allows us to use final VA values, so here
1047 // we also handle rest commands like symbol assignments and ASSERTs.
1048 void LinkerScript::assignAddresses() {
1049 Dot = getInitialDot();
1051 auto Deleter = make_unique<AddressState>();
1052 Ctx = Deleter.get();
1053 ErrorOnMissingSection = true;
1056 for (BaseCommand *Base : SectionCommands) {
1057 if (auto *Cmd = dyn_cast<SymbolAssignment>(Base)) {
1059 assignSymbol(Cmd, false);
1060 Cmd->Size = Dot - Cmd->Addr;
1063 assignOffsets(cast<OutputSection>(Base));
1068 // Creates program headers as instructed by PHDRS linker script command.
1069 std::vector<PhdrEntry *> LinkerScript::createPhdrs() {
1070 std::vector<PhdrEntry *> Ret;
1072 // Process PHDRS and FILEHDR keywords because they are not
1073 // real output sections and cannot be added in the following loop.
1074 for (const PhdrsCommand &Cmd : PhdrsCommands) {
1075 PhdrEntry *Phdr = make<PhdrEntry>(Cmd.Type, Cmd.Flags ? *Cmd.Flags : PF_R);
1078 Phdr->add(Out::ElfHeader);
1080 Phdr->add(Out::ProgramHeaders);
1083 Phdr->p_paddr = Cmd.LMAExpr().getValue();
1084 Phdr->HasLMA = true;
1086 Ret.push_back(Phdr);
1089 // Add output sections to program headers.
1090 for (OutputSection *Sec : OutputSections) {
1091 // Assign headers specified by linker script
1092 for (size_t Id : getPhdrIndices(Sec)) {
1094 if (!PhdrsCommands[Id].Flags.hasValue())
1095 Ret[Id]->p_flags |= Sec->getPhdrFlags();
1101 // Returns true if we should emit an .interp section.
1103 // We usually do. But if PHDRS commands are given, and
1104 // no PT_INTERP is there, there's no place to emit an
1105 // .interp, so we don't do that in that case.
1106 bool LinkerScript::needsInterpSection() {
1107 if (PhdrsCommands.empty())
1109 for (PhdrsCommand &Cmd : PhdrsCommands)
1110 if (Cmd.Type == PT_INTERP)
1115 ExprValue LinkerScript::getSymbolValue(StringRef Name, const Twine &Loc) {
1118 return {Ctx->OutSec, false, Dot - Ctx->OutSec->Addr, Loc};
1119 error(Loc + ": unable to get location counter value");
1123 if (Symbol *Sym = Symtab->find(Name)) {
1124 if (auto *DS = dyn_cast<Defined>(Sym))
1125 return {DS->Section, false, DS->Value, Loc};
1126 if (isa<SharedSymbol>(Sym))
1127 if (!ErrorOnMissingSection)
1128 return {nullptr, false, 0, Loc};
1131 error(Loc + ": symbol not found: " + Name);
1135 // Returns the index of the segment named Name.
1136 static Optional<size_t> getPhdrIndex(ArrayRef<PhdrsCommand> Vec,
1138 for (size_t I = 0; I < Vec.size(); ++I)
1139 if (Vec[I].Name == Name)
1144 // Returns indices of ELF headers containing specific section. Each index is a
1145 // zero based number of ELF header listed within PHDRS {} script block.
1146 std::vector<size_t> LinkerScript::getPhdrIndices(OutputSection *Cmd) {
1147 std::vector<size_t> Ret;
1149 for (StringRef S : Cmd->Phdrs) {
1150 if (Optional<size_t> Idx = getPhdrIndex(PhdrsCommands, S))
1151 Ret.push_back(*Idx);
1152 else if (S != "NONE")
1153 error(Cmd->Location + ": section header '" + S +
1154 "' is not listed in PHDRS");