1//===- LinkerScript.cpp ---------------------------------------------------===//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8//
9// This file contains the parser/evaluator of the linker script.
10//
11//===----------------------------------------------------------------------===//
12
13#include "LinkerScript.h"
14#include "Config.h"
15#include "InputFiles.h"
16#include "InputSection.h"
17#include "OutputSections.h"
18#include "SymbolTable.h"
19#include "Symbols.h"
20#include "SyntheticSections.h"
21#include "Target.h"
22#include "Writer.h"
23#include "lld/Common/CommonLinkerContext.h"
24#include "lld/Common/Strings.h"
25#include "llvm/ADT/STLExtras.h"
26#include "llvm/ADT/StringRef.h"
27#include "llvm/BinaryFormat/ELF.h"
28#include "llvm/Support/Casting.h"
29#include "llvm/Support/ErrorHandling.h"
30#include "llvm/Support/TimeProfiler.h"
31#include <algorithm>
32#include <cassert>
33#include <cstddef>
34#include <cstdint>
35#include <limits>
36#include <string>
37
38using namespace llvm;
39using namespace llvm::ELF;
40using namespace llvm::object;
41using namespace llvm::support::endian;
42using namespace lld;
43using namespace lld::elf;
44
45static bool isSectionPrefix(StringRef prefix, StringRef name) {
46 return name.consume_front(Prefix: prefix) && (name.empty() || name[0] == '.');
47}
48
49StringRef LinkerScript::getOutputSectionName(const InputSectionBase *s) const {
50 // This is for --emit-relocs and -r. If .text.foo is emitted as .text.bar, we
51 // want to emit .rela.text.foo as .rela.text.bar for consistency (this is not
52 // technically required, but not doing it is odd). This code guarantees that.
53 if (auto *isec = dyn_cast<InputSection>(Val: s)) {
54 if (InputSectionBase *rel = isec->getRelocatedSection()) {
55 OutputSection *out = rel->getOutputSection();
56 if (!out) {
57 assert(ctx.arg.relocatable && (rel->flags & SHF_LINK_ORDER));
58 return s->name;
59 }
60 StringSaver &ss = ctx.saver;
61 if (s->type == SHT_CREL)
62 return ss.save(S: ".crel" + out->name);
63 if (s->type == SHT_RELA)
64 return ss.save(S: ".rela" + out->name);
65 return ss.save(S: ".rel" + out->name);
66 }
67 }
68
69 if (ctx.arg.relocatable)
70 return s->name;
71
72 // A BssSection created for a common symbol is identified as "COMMON" in
73 // linker scripts. It should go to .bss section.
74 if (s->name == "COMMON")
75 return ".bss";
76
77 if (hasSectionsCommand)
78 return s->name;
79
80 // When no SECTIONS is specified, emulate GNU ld's internal linker scripts
81 // by grouping sections with certain prefixes.
82
83 // GNU ld places text sections with prefix ".text.hot.", ".text.unknown.",
84 // ".text.unlikely.", ".text.startup." or ".text.exit." before others.
85 // We provide an option -z keep-text-section-prefix to group such sections
86 // into separate output sections. This is more flexible. See also
87 // sortISDBySectionOrder().
88 // ".text.unknown" means the hotness of the section is unknown. When
89 // SampleFDO is used, if a function doesn't have sample, it could be very
90 // cold or it could be a new function never being sampled. Those functions
91 // will be kept in the ".text.unknown" section.
92 // ".text.split." holds symbols which are split out from functions in other
93 // input sections. For example, with -fsplit-machine-functions, placing the
94 // cold parts in .text.split instead of .text.unlikely mitigates against poor
95 // profile inaccuracy. Techniques such as hugepage remapping can make
96 // conservative decisions at the section granularity.
97 if (isSectionPrefix(prefix: ".text", name: s->name)) {
98 if (ctx.arg.zKeepTextSectionPrefix)
99 for (StringRef v : {".text.hot", ".text.unknown", ".text.unlikely",
100 ".text.startup", ".text.exit", ".text.split"})
101 if (isSectionPrefix(prefix: v.substr(Start: 5), name: s->name.substr(Start: 5)))
102 return v;
103 return ".text";
104 }
105
106 for (StringRef v : {".data.rel.ro", ".data", ".rodata",
107 ".bss.rel.ro", ".bss", ".ldata",
108 ".lrodata", ".lbss", ".gcc_except_table",
109 ".init_array", ".fini_array", ".tbss",
110 ".tdata", ".ARM.exidx", ".ARM.extab",
111 ".ctors", ".dtors", ".sbss",
112 ".sdata", ".srodata"})
113 if (isSectionPrefix(prefix: v, name: s->name))
114 return v;
115
116 return s->name;
117}
118
119uint64_t ExprValue::getValue() const {
120 if (sec)
121 return alignToPowerOf2(Value: sec->getOutputSection()->addr + sec->getOffset(offset: val),
122 Align: alignment);
123 return alignToPowerOf2(Value: val, Align: alignment);
124}
125
126uint64_t ExprValue::getSecAddr() const {
127 return sec ? sec->getOutputSection()->addr + sec->getOffset(offset: 0) : 0;
128}
129
130uint64_t ExprValue::getSectionOffset() const {
131 return getValue() - getSecAddr();
132}
133
134// std::unique_ptr<OutputSection> may be incomplete type.
135LinkerScript::LinkerScript(Ctx &ctx) : ctx(ctx) {}
136LinkerScript::~LinkerScript() {}
137
138OutputDesc *LinkerScript::createOutputSection(StringRef name,
139 StringRef location) {
140 OutputDesc *&secRef = nameToOutputSection[CachedHashStringRef(name)];
141 OutputDesc *sec;
142 if (secRef && secRef->osec.location.empty()) {
143 // There was a forward reference.
144 sec = secRef;
145 } else {
146 descPool.emplace_back(
147 Args: std::make_unique<OutputDesc>(args&: ctx, args&: name, args: SHT_PROGBITS, args: 0));
148 sec = descPool.back().get();
149 if (!secRef)
150 secRef = sec;
151 }
152 sec->osec.location = std::string(location);
153 return sec;
154}
155
156OutputDesc *LinkerScript::getOrCreateOutputSection(StringRef name) {
157 auto &secRef = nameToOutputSection[CachedHashStringRef(name)];
158 if (!secRef) {
159 secRef = descPool
160 .emplace_back(
161 Args: std::make_unique<OutputDesc>(args&: ctx, args&: name, args: SHT_PROGBITS, args: 0))
162 .get();
163 }
164 return secRef;
165}
166
167// Expands the memory region by the specified size.
168static void expandMemoryRegion(MemoryRegion *memRegion, uint64_t size,
169 StringRef secName) {
170 memRegion->curPos += size;
171}
172
173void LinkerScript::expandMemoryRegions(uint64_t size) {
174 if (state->memRegion)
175 expandMemoryRegion(memRegion: state->memRegion, size, secName: state->outSec->name);
176 // Only expand the LMARegion if it is different from memRegion.
177 if (state->lmaRegion && state->memRegion != state->lmaRegion)
178 expandMemoryRegion(memRegion: state->lmaRegion, size, secName: state->outSec->name);
179}
180
181void LinkerScript::expandOutputSection(uint64_t size) {
182 state->outSec->size += size;
183 size_t regionSize = size;
184 if (state->outSec->inOverlay) {
185 // Expand the overlay if necessary, and expand the region by the
186 // corresponding amount.
187 if (state->outSec->size > state->overlaySize) {
188 regionSize = state->outSec->size - state->overlaySize;
189 state->overlaySize = state->outSec->size;
190 } else {
191 regionSize = 0;
192 }
193 }
194 expandMemoryRegions(size: regionSize);
195}
196
197void LinkerScript::setDot(Expr e, const Twine &loc, bool inSec) {
198 uint64_t val = e().getValue();
199 // If val is smaller and we are in an output section, record the error and
200 // report it if this is the last assignAddresses iteration. dot may be smaller
201 // if there is another assignAddresses iteration.
202 if (val < dot && inSec) {
203 recordError(msg: loc + ": unable to move location counter (0x" +
204 Twine::utohexstr(Val: dot) + ") backward to 0x" +
205 Twine::utohexstr(Val: val) + " for section '" + state->outSec->name +
206 "'");
207 }
208
209 // Update to location counter means update to section size.
210 if (inSec)
211 expandOutputSection(size: val - dot);
212
213 dot = val;
214}
215
216// Used for handling linker symbol assignments, for both finalizing
217// their values and doing early declarations. Returns true if symbol
218// should be defined from linker script.
219static bool shouldDefineSym(Ctx &ctx, SymbolAssignment *cmd) {
220 if (cmd->name == ".")
221 return false;
222
223 return !cmd->provide || ctx.script->shouldAddProvideSym(symName: cmd->name);
224}
225
226// Called by processSymbolAssignments() to assign definitions to
227// linker-script-defined symbols.
228void LinkerScript::addSymbol(SymbolAssignment *cmd) {
229 if (!shouldDefineSym(ctx, cmd))
230 return;
231
232 // Define a symbol.
233 ExprValue value = cmd->expression();
234 SectionBase *sec = value.isAbsolute() ? nullptr : value.sec;
235 uint8_t visibility = cmd->hidden ? STV_HIDDEN : STV_DEFAULT;
236
237 // When this function is called, section addresses have not been
238 // fixed yet. So, we may or may not know the value of the RHS
239 // expression.
240 //
241 // For example, if an expression is `x = 42`, we know x is always 42.
242 // However, if an expression is `x = .`, there's no way to know its
243 // value at the moment.
244 //
245 // We want to set symbol values early if we can. This allows us to
246 // use symbols as variables in linker scripts. Doing so allows us to
247 // write expressions like this: `alignment = 16; . = ALIGN(., alignment)`.
248 uint64_t symValue = value.sec ? 0 : value.getValue();
249
250 Defined newSym(ctx, createInternalFile(ctx, name: cmd->location), cmd->name,
251 STB_GLOBAL, visibility, value.type, symValue, 0, sec);
252
253 Symbol *sym = ctx.symtab->insert(name: cmd->name);
254 sym->mergeProperties(other: newSym);
255 newSym.overwrite(sym&: *sym);
256 sym->isUsedInRegularObj = true;
257 cmd->sym = cast<Defined>(Val: sym);
258}
259
260// This function is called from LinkerScript::declareSymbols.
261// It creates a placeholder symbol if needed.
262void LinkerScript::declareSymbol(SymbolAssignment *cmd) {
263 if (!shouldDefineSym(ctx, cmd))
264 return;
265
266 uint8_t visibility = cmd->hidden ? STV_HIDDEN : STV_DEFAULT;
267 Defined newSym(ctx, ctx.internalFile, cmd->name, STB_GLOBAL, visibility,
268 STT_NOTYPE, 0, 0, nullptr);
269
270 // If the symbol is already defined, its order is 0 (with absence indicating
271 // 0); otherwise it's assigned the order of the SymbolAssignment.
272 Symbol *sym = ctx.symtab->insert(name: cmd->name);
273 if (!sym->isDefined())
274 ctx.scriptSymOrder.insert(KV: {sym, cmd->symOrder});
275
276 // We can't calculate final value right now.
277 sym->mergeProperties(other: newSym);
278 newSym.overwrite(sym&: *sym);
279
280 cmd->sym = cast<Defined>(Val: sym);
281 cmd->provide = false;
282 sym->isUsedInRegularObj = true;
283 sym->scriptDefined = true;
284}
285
286using SymbolAssignmentMap =
287 DenseMap<const Defined *, std::pair<SectionBase *, uint64_t>>;
288
289// Collect section/value pairs of linker-script-defined symbols. This is used to
290// check whether symbol values converge.
291static SymbolAssignmentMap
292getSymbolAssignmentValues(ArrayRef<SectionCommand *> sectionCommands) {
293 SymbolAssignmentMap ret;
294 for (SectionCommand *cmd : sectionCommands) {
295 if (auto *assign = dyn_cast<SymbolAssignment>(Val: cmd)) {
296 if (assign->sym) // sym is nullptr for dot.
297 ret.try_emplace(Key: assign->sym, Args: std::make_pair(x&: assign->sym->section,
298 y&: assign->sym->value));
299 continue;
300 }
301 if (isa<SectionClassDesc>(Val: cmd))
302 continue;
303 for (SectionCommand *subCmd : cast<OutputDesc>(Val: cmd)->osec.commands)
304 if (auto *assign = dyn_cast<SymbolAssignment>(Val: subCmd))
305 if (assign->sym)
306 ret.try_emplace(Key: assign->sym, Args: std::make_pair(x&: assign->sym->section,
307 y&: assign->sym->value));
308 }
309 return ret;
310}
311
312// Returns the lexicographical smallest (for determinism) Defined whose
313// section/value has changed.
314static const Defined *
315getChangedSymbolAssignment(const SymbolAssignmentMap &oldValues) {
316 const Defined *changed = nullptr;
317 for (auto &it : oldValues) {
318 const Defined *sym = it.first;
319 if (std::make_pair(x: sym->section, y: sym->value) != it.second &&
320 (!changed || sym->getName() < changed->getName()))
321 changed = sym;
322 }
323 return changed;
324}
325
326// Process INSERT [AFTER|BEFORE] commands. For each command, we move the
327// specified output section to the designated place.
328void LinkerScript::processInsertCommands() {
329 SmallVector<OutputDesc *, 0> moves;
330 for (const InsertCommand &cmd : insertCommands) {
331 if (ctx.arg.enableNonContiguousRegions)
332 ErrAlways(ctx)
333 << "INSERT cannot be used with --enable-non-contiguous-regions";
334
335 for (StringRef name : cmd.names) {
336 // If base is empty, it may have been discarded by
337 // adjustOutputSections(). We do not handle such output sections.
338 auto from = llvm::find_if(Range&: sectionCommands, P: [&](SectionCommand *subCmd) {
339 return isa<OutputDesc>(Val: subCmd) &&
340 cast<OutputDesc>(Val: subCmd)->osec.name == name;
341 });
342 if (from == sectionCommands.end())
343 continue;
344 moves.push_back(Elt: cast<OutputDesc>(Val: *from));
345 sectionCommands.erase(CI: from);
346 }
347
348 auto insertPos =
349 llvm::find_if(Range&: sectionCommands, P: [&cmd](SectionCommand *subCmd) {
350 auto *to = dyn_cast<OutputDesc>(Val: subCmd);
351 return to != nullptr && to->osec.name == cmd.where;
352 });
353 if (insertPos == sectionCommands.end()) {
354 ErrAlways(ctx) << "unable to insert " << cmd.names[0]
355 << (cmd.isAfter ? " after " : " before ") << cmd.where;
356 } else {
357 if (cmd.isAfter)
358 ++insertPos;
359 sectionCommands.insert(I: insertPos, From: moves.begin(), To: moves.end());
360 }
361 moves.clear();
362 }
363}
364
365// Symbols defined in script should not be inlined by LTO. At the same time
366// we don't know their final values until late stages of link. Here we scan
367// over symbol assignment commands and create placeholder symbols if needed.
368void LinkerScript::declareSymbols() {
369 assert(!state);
370 for (SectionCommand *cmd : sectionCommands) {
371 if (auto *assign = dyn_cast<SymbolAssignment>(Val: cmd)) {
372 declareSymbol(cmd: assign);
373 continue;
374 }
375 if (isa<SectionClassDesc>(Val: cmd))
376 continue;
377
378 // If the output section directive has constraints,
379 // we can't say for sure if it is going to be included or not.
380 // Skip such sections for now. Improve the checks if we ever
381 // need symbols from that sections to be declared early.
382 const OutputSection &sec = cast<OutputDesc>(Val: cmd)->osec;
383 if (sec.constraint != ConstraintKind::NoConstraint)
384 continue;
385 for (SectionCommand *cmd : sec.commands)
386 if (auto *assign = dyn_cast<SymbolAssignment>(Val: cmd))
387 declareSymbol(cmd: assign);
388 }
389}
390
391// This function is called from assignAddresses, while we are
392// fixing the output section addresses. This function is supposed
393// to set the final value for a given symbol assignment.
394void LinkerScript::assignSymbol(SymbolAssignment *cmd, bool inSec) {
395 if (cmd->name == ".") {
396 setDot(e: cmd->expression, loc: cmd->location, inSec);
397 return;
398 }
399
400 if (!cmd->sym)
401 return;
402
403 ExprValue v = cmd->expression();
404 if (v.isAbsolute()) {
405 cmd->sym->section = nullptr;
406 cmd->sym->value = v.getValue();
407 } else {
408 cmd->sym->section = v.sec;
409 cmd->sym->value = v.getSectionOffset();
410 }
411 cmd->sym->type = v.type;
412}
413
414bool InputSectionDescription::matchesFile(const InputFile &file) const {
415 if (filePat.isTrivialMatchAll())
416 return true;
417
418 if (!matchesFileCache || matchesFileCache->first != &file) {
419 if (matchType == MatchType::WholeArchive) {
420 matchesFileCache.emplace(args: &file, args: filePat.match(s: file.archiveName));
421 } else {
422 if (matchType == MatchType::ArchivesExcluded && !file.archiveName.empty())
423 matchesFileCache.emplace(args: &file, args: false);
424 else
425 matchesFileCache.emplace(args: &file, args: filePat.match(s: file.getNameForScript()));
426 }
427 }
428
429 return matchesFileCache->second;
430}
431
432bool SectionPattern::excludesFile(const InputFile &file) const {
433 if (excludedFilePat.empty())
434 return false;
435
436 if (!excludesFileCache || excludesFileCache->first != &file)
437 excludesFileCache.emplace(args: &file,
438 args: excludedFilePat.match(s: file.getNameForScript()));
439
440 return excludesFileCache->second;
441}
442
443bool LinkerScript::shouldKeep(InputSectionBase *s) {
444 for (InputSectionDescription *id : keptSections)
445 if (id->matchesFile(file: *s->file))
446 for (SectionPattern &p : id->sectionPatterns)
447 if (p.sectionPat.match(s: s->name) &&
448 (s->flags & id->withFlags) == id->withFlags &&
449 (s->flags & id->withoutFlags) == 0)
450 return true;
451 return false;
452}
453
454// A helper function for the SORT() command.
455static bool matchConstraints(ArrayRef<InputSectionBase *> sections,
456 ConstraintKind kind) {
457 if (kind == ConstraintKind::NoConstraint)
458 return true;
459
460 bool isRW = llvm::any_of(
461 Range&: sections, P: [](InputSectionBase *sec) { return sec->flags & SHF_WRITE; });
462
463 return (isRW && kind == ConstraintKind::ReadWrite) ||
464 (!isRW && kind == ConstraintKind::ReadOnly);
465}
466
467static void sortSections(MutableArrayRef<InputSectionBase *> vec,
468 SortSectionPolicy k) {
469 auto alignmentComparator = [](InputSectionBase *a, InputSectionBase *b) {
470 // ">" is not a mistake. Sections with larger alignments are placed
471 // before sections with smaller alignments in order to reduce the
472 // amount of padding necessary. This is compatible with GNU.
473 return a->addralign > b->addralign;
474 };
475 auto nameComparator = [](InputSectionBase *a, InputSectionBase *b) {
476 return a->name < b->name;
477 };
478 auto priorityComparator = [](InputSectionBase *a, InputSectionBase *b) {
479 return getPriority(s: a->name) < getPriority(s: b->name);
480 };
481
482 switch (k) {
483 case SortSectionPolicy::Default:
484 case SortSectionPolicy::None:
485 return;
486 case SortSectionPolicy::Alignment:
487 return llvm::stable_sort(Range&: vec, C: alignmentComparator);
488 case SortSectionPolicy::Name:
489 return llvm::stable_sort(Range&: vec, C: nameComparator);
490 case SortSectionPolicy::Priority:
491 return llvm::stable_sort(Range&: vec, C: priorityComparator);
492 case SortSectionPolicy::Reverse:
493 return std::reverse(first: vec.begin(), last: vec.end());
494 }
495}
496
497// Sort sections as instructed by SORT-family commands and --sort-section
498// option. Because SORT-family commands can be nested at most two depth
499// (e.g. SORT_BY_NAME(SORT_BY_ALIGNMENT(.text.*))) and because the command
500// line option is respected even if a SORT command is given, the exact
501// behavior we have here is a bit complicated. Here are the rules.
502//
503// 1. If two SORT commands are given, --sort-section is ignored.
504// 2. If one SORT command is given, and if it is not SORT_NONE,
505// --sort-section is handled as an inner SORT command.
506// 3. If one SORT command is given, and if it is SORT_NONE, don't sort.
507// 4. If no SORT command is given, sort according to --sort-section.
508static void sortInputSections(Ctx &ctx, MutableArrayRef<InputSectionBase *> vec,
509 SortSectionPolicy outer,
510 SortSectionPolicy inner) {
511 if (outer == SortSectionPolicy::None)
512 return;
513
514 if (inner == SortSectionPolicy::Default)
515 sortSections(vec, k: ctx.arg.sortSection);
516 else
517 sortSections(vec, k: inner);
518 sortSections(vec, k: outer);
519}
520
521// Compute and remember which sections the InputSectionDescription matches.
522SmallVector<InputSectionBase *, 0>
523LinkerScript::computeInputSections(const InputSectionDescription *cmd,
524 ArrayRef<InputSectionBase *> sections,
525 const SectionBase &outCmd) {
526 SmallVector<InputSectionBase *, 0> ret;
527 DenseSet<InputSectionBase *> spills;
528
529 // Returns whether an input section's flags match the input section
530 // description's specifiers.
531 auto flagsMatch = [cmd](InputSectionBase *sec) {
532 return (sec->flags & cmd->withFlags) == cmd->withFlags &&
533 (sec->flags & cmd->withoutFlags) == 0;
534 };
535
536 // Collects all sections that satisfy constraints of Cmd.
537 if (cmd->classRef.empty()) {
538 DenseSet<size_t> seen;
539 size_t sizeAfterPrevSort = 0;
540 SmallVector<size_t, 0> indexes;
541 auto sortByPositionThenCommandLine = [&](size_t begin, size_t end) {
542 llvm::sort(C: MutableArrayRef<size_t>(indexes).slice(N: begin, M: end - begin));
543 for (size_t i = begin; i != end; ++i)
544 ret[i] = sections[indexes[i]];
545 sortInputSections(
546 ctx,
547 vec: MutableArrayRef<InputSectionBase *>(ret).slice(N: begin, M: end - begin),
548 outer: ctx.arg.sortSection, inner: SortSectionPolicy::None);
549 };
550
551 for (const SectionPattern &pat : cmd->sectionPatterns) {
552 size_t sizeBeforeCurrPat = ret.size();
553
554 for (size_t i = 0, e = sections.size(); i != e; ++i) {
555 // Skip if the section is dead or has been matched by a previous pattern
556 // in this input section description.
557 InputSectionBase *sec = sections[i];
558 if (!sec->isLive() || seen.contains(V: i))
559 continue;
560
561 // For --emit-relocs we have to ignore entries like
562 // .rela.dyn : { *(.rela.data) }
563 // which are common because they are in the default bfd script.
564 // We do not ignore SHT_REL[A] linker-synthesized sections here because
565 // want to support scripts that do custom layout for them.
566 if (isa<InputSection>(Val: sec) &&
567 cast<InputSection>(Val: sec)->getRelocatedSection())
568 continue;
569
570 // Check the name early to improve performance in the common case.
571 if (!pat.sectionPat.match(s: sec->name))
572 continue;
573
574 if (!cmd->matchesFile(file: *sec->file) || pat.excludesFile(file: *sec->file) ||
575 !flagsMatch(sec))
576 continue;
577
578 if (sec->parent) {
579 // Skip if not allowing multiple matches.
580 if (!ctx.arg.enableNonContiguousRegions)
581 continue;
582
583 // Disallow spilling into /DISCARD/; special handling would be needed
584 // for this in address assignment, and the semantics are nebulous.
585 if (outCmd.name == "/DISCARD/")
586 continue;
587
588 // Class definitions cannot contain spills, nor can a class definition
589 // generate a spill in a subsequent match. Those behaviors belong to
590 // class references and additional matches.
591 if (!isa<SectionClass>(Val: outCmd) && !isa<SectionClass>(Val: sec->parent))
592 spills.insert(V: sec);
593 }
594
595 ret.push_back(Elt: sec);
596 indexes.push_back(Elt: i);
597 seen.insert(V: i);
598 }
599
600 if (pat.sortOuter == SortSectionPolicy::Default)
601 continue;
602
603 // Matched sections are ordered by radix sort with the keys being (SORT*,
604 // --sort-section, input order), where SORT* (if present) is most
605 // significant.
606 //
607 // Matched sections between the previous SORT* and this SORT* are sorted
608 // by (--sort-alignment, input order).
609 sortByPositionThenCommandLine(sizeAfterPrevSort, sizeBeforeCurrPat);
610 // Matched sections by this SORT* pattern are sorted using all 3 keys.
611 // ret[sizeBeforeCurrPat,ret.size()) are already in the input order, so we
612 // just sort by sortOuter and sortInner.
613 sortInputSections(
614 ctx,
615 vec: MutableArrayRef<InputSectionBase *>(ret).slice(N: sizeBeforeCurrPat),
616 outer: pat.sortOuter, inner: pat.sortInner);
617 sizeAfterPrevSort = ret.size();
618 }
619
620 // Matched sections after the last SORT* are sorted by (--sort-alignment,
621 // input order).
622 sortByPositionThenCommandLine(sizeAfterPrevSort, ret.size());
623 } else {
624 SectionClassDesc *scd =
625 sectionClasses.lookup(Val: CachedHashStringRef(cmd->classRef));
626 if (!scd) {
627 Err(ctx) << "undefined section class '" << cmd->classRef << "'";
628 return ret;
629 }
630 if (!scd->sc.assigned) {
631 Err(ctx) << "section class '" << cmd->classRef << "' referenced by '"
632 << outCmd.name << "' before class definition";
633 return ret;
634 }
635
636 for (InputSectionDescription *isd : scd->sc.commands) {
637 for (InputSectionBase *sec : isd->sectionBases) {
638 if (!flagsMatch(sec))
639 continue;
640 bool isSpill = sec->parent && isa<OutputSection>(Val: sec->parent);
641 if (!sec->parent || (isSpill && outCmd.name == "/DISCARD/")) {
642 Err(ctx) << "section '" << sec->name
643 << "' cannot spill from/to /DISCARD/";
644 continue;
645 }
646 if (isSpill)
647 spills.insert(V: sec);
648 ret.push_back(Elt: sec);
649 }
650 }
651 }
652
653 // The flag --enable-non-contiguous-regions or the section CLASS syntax may
654 // cause sections to match an InputSectionDescription in more than one
655 // OutputSection. Matches after the first were collected in the spills set, so
656 // replace these with potential spill sections.
657 if (!spills.empty()) {
658 for (InputSectionBase *&sec : ret) {
659 if (!spills.contains(V: sec))
660 continue;
661
662 // Append the spill input section to the list for the input section,
663 // creating it if necessary.
664 PotentialSpillSection *pss = make<PotentialSpillSection>(
665 args&: *sec, args&: const_cast<InputSectionDescription &>(*cmd));
666 auto [it, inserted] =
667 potentialSpillLists.try_emplace(Key: sec, Args: PotentialSpillList{.head: pss, .tail: pss});
668 if (!inserted) {
669 PotentialSpillSection *&tail = it->second.tail;
670 tail = tail->next = pss;
671 }
672 sec = pss;
673 }
674 }
675
676 return ret;
677}
678
679void LinkerScript::discard(InputSectionBase &s) {
680 if (&s == ctx.in.shStrTab.get())
681 ErrAlways(ctx) << "discarding " << s.name << " section is not allowed";
682
683 s.markDead();
684 s.parent = nullptr;
685 for (InputSection *sec : s.dependentSections)
686 discard(s&: *sec);
687}
688
689void LinkerScript::discardSynthetic(OutputSection &outCmd) {
690 for (Partition &part : ctx.partitions) {
691 if (!part.armExidx || !part.armExidx->isLive())
692 continue;
693 SmallVector<InputSectionBase *, 0> secs(
694 part.armExidx->exidxSections.begin(),
695 part.armExidx->exidxSections.end());
696 for (SectionCommand *cmd : outCmd.commands)
697 if (auto *isd = dyn_cast<InputSectionDescription>(Val: cmd))
698 for (InputSectionBase *s : computeInputSections(cmd: isd, sections: secs, outCmd))
699 discard(s&: *s);
700 }
701}
702
703SmallVector<InputSectionBase *, 0>
704LinkerScript::createInputSectionList(OutputSection &outCmd) {
705 SmallVector<InputSectionBase *, 0> ret;
706
707 for (SectionCommand *cmd : outCmd.commands) {
708 if (auto *isd = dyn_cast<InputSectionDescription>(Val: cmd)) {
709 isd->sectionBases = computeInputSections(cmd: isd, sections: ctx.inputSections, outCmd);
710 for (InputSectionBase *s : isd->sectionBases)
711 s->parent = &outCmd;
712 ret.insert(I: ret.end(), From: isd->sectionBases.begin(), To: isd->sectionBases.end());
713 }
714 }
715 return ret;
716}
717
718// Create output sections described by SECTIONS commands.
719void LinkerScript::processSectionCommands() {
720 auto process = [this](OutputSection *osec) {
721 SmallVector<InputSectionBase *, 0> v = createInputSectionList(outCmd&: *osec);
722
723 // The output section name `/DISCARD/' is special.
724 // Any input section assigned to it is discarded.
725 if (osec->name == "/DISCARD/") {
726 for (InputSectionBase *s : v)
727 discard(s&: *s);
728 discardSynthetic(outCmd&: *osec);
729 osec->commands.clear();
730 return false;
731 }
732
733 // This is for ONLY_IF_RO and ONLY_IF_RW. An output section directive
734 // ".foo : ONLY_IF_R[OW] { ... }" is handled only if all member input
735 // sections satisfy a given constraint. If not, a directive is handled
736 // as if it wasn't present from the beginning.
737 //
738 // Because we'll iterate over SectionCommands many more times, the easy
739 // way to "make it as if it wasn't present" is to make it empty.
740 if (!matchConstraints(sections: v, kind: osec->constraint)) {
741 for (InputSectionBase *s : v)
742 s->parent = nullptr;
743 osec->commands.clear();
744 return false;
745 }
746
747 // Handle subalign (e.g. ".foo : SUBALIGN(32) { ... }"). If subalign
748 // is given, input sections are aligned to that value, whether the
749 // given value is larger or smaller than the original section alignment.
750 if (osec->subalignExpr) {
751 uint32_t subalign = osec->subalignExpr().getValue();
752 for (InputSectionBase *s : v)
753 s->addralign = subalign;
754 }
755
756 // Set the partition field the same way OutputSection::recordSection()
757 // does. Partitions cannot be used with the SECTIONS command, so this is
758 // always 1.
759 osec->partition = 1;
760 return true;
761 };
762
763 // Process OVERWRITE_SECTIONS first so that it can overwrite the main script
764 // or orphans.
765 if (ctx.arg.enableNonContiguousRegions && !overwriteSections.empty())
766 ErrAlways(ctx) << "OVERWRITE_SECTIONS cannot be used with "
767 "--enable-non-contiguous-regions";
768 DenseMap<CachedHashStringRef, OutputDesc *> map;
769 size_t i = 0;
770 for (OutputDesc *osd : overwriteSections) {
771 OutputSection *osec = &osd->osec;
772 if (process(osec) &&
773 !map.try_emplace(Key: CachedHashStringRef(osec->name), Args&: osd).second)
774 Warn(ctx) << "OVERWRITE_SECTIONS specifies duplicate " << osec->name;
775 }
776 for (SectionCommand *&base : sectionCommands) {
777 if (auto *osd = dyn_cast<OutputDesc>(Val: base)) {
778 OutputSection *osec = &osd->osec;
779 if (OutputDesc *overwrite = map.lookup(Val: CachedHashStringRef(osec->name))) {
780 Log(ctx) << overwrite->osec.location << " overwrites " << osec->name;
781 overwrite->osec.sectionIndex = i++;
782 base = overwrite;
783 } else if (process(osec)) {
784 osec->sectionIndex = i++;
785 }
786 } else if (auto *sc = dyn_cast<SectionClassDesc>(Val: base)) {
787 for (InputSectionDescription *isd : sc->sc.commands) {
788 isd->sectionBases =
789 computeInputSections(cmd: isd, sections: ctx.inputSections, outCmd: sc->sc);
790 for (InputSectionBase *s : isd->sectionBases) {
791 // A section class containing a section with different parent isn't
792 // necessarily an error due to --enable-non-contiguous-regions. Such
793 // sections all become potential spills when the class is referenced.
794 if (!s->parent)
795 s->parent = &sc->sc;
796 }
797 }
798 sc->sc.assigned = true;
799 }
800 }
801
802 // Check that input sections cannot spill into or out of INSERT,
803 // since the semantics are nebulous. This is also true for OVERWRITE_SECTIONS,
804 // but no check is needed, since the order of processing ensures they cannot
805 // legally reference classes.
806 if (!potentialSpillLists.empty()) {
807 DenseSet<StringRef> insertNames;
808 for (InsertCommand &ic : insertCommands)
809 insertNames.insert_range(R&: ic.names);
810 for (SectionCommand *&base : sectionCommands) {
811 auto *osd = dyn_cast<OutputDesc>(Val: base);
812 if (!osd)
813 continue;
814 OutputSection *os = &osd->osec;
815 if (!insertNames.contains(V: os->name))
816 continue;
817 for (SectionCommand *sc : os->commands) {
818 auto *isd = dyn_cast<InputSectionDescription>(Val: sc);
819 if (!isd)
820 continue;
821 for (InputSectionBase *isec : isd->sectionBases)
822 if (isa<PotentialSpillSection>(Val: isec) ||
823 potentialSpillLists.contains(Val: isec))
824 Err(ctx) << "section '" << isec->name
825 << "' cannot spill from/to INSERT section '" << os->name
826 << "'";
827 }
828 }
829 }
830
831 // If an OVERWRITE_SECTIONS specified output section is not in
832 // sectionCommands, append it to the end. The section will be inserted by
833 // orphan placement.
834 for (OutputDesc *osd : overwriteSections)
835 if (osd->osec.partition == 1 && osd->osec.sectionIndex == UINT32_MAX)
836 sectionCommands.push_back(Elt: osd);
837
838 // Input sections cannot have a section class parent past this point; they
839 // must have been assigned to an output section.
840 for (const auto &[_, sc] : sectionClasses) {
841 for (InputSectionDescription *isd : sc->sc.commands) {
842 for (InputSectionBase *sec : isd->sectionBases) {
843 if (sec->parent && isa<SectionClass>(Val: sec->parent)) {
844 Err(ctx) << "section class '" << sec->parent->name
845 << "' is unreferenced";
846 goto nextClass;
847 }
848 }
849 }
850 nextClass:;
851 }
852}
853
854void LinkerScript::processSymbolAssignments() {
855 // Dot outside an output section still represents a relative address, whose
856 // sh_shndx should not be SHN_UNDEF or SHN_ABS. Create a dummy aether section
857 // that fills the void outside a section. It has an index of one, which is
858 // indistinguishable from any other regular section index.
859 aether = std::make_unique<OutputSection>(args&: ctx, args: "", args: 0, args: SHF_ALLOC);
860 aether->sectionIndex = 1;
861
862 // `st` captures the local AddressState and makes it accessible deliberately.
863 // This is needed as there are some cases where we cannot just thread the
864 // current state through to a lambda function created by the script parser.
865 AddressState st(*this);
866 state = &st;
867 st.outSec = aether.get();
868
869 for (SectionCommand *cmd : sectionCommands) {
870 if (auto *assign = dyn_cast<SymbolAssignment>(Val: cmd))
871 addSymbol(cmd: assign);
872 else if (auto *osd = dyn_cast<OutputDesc>(Val: cmd))
873 for (SectionCommand *subCmd : osd->osec.commands)
874 if (auto *assign = dyn_cast<SymbolAssignment>(Val: subCmd))
875 addSymbol(cmd: assign);
876 }
877
878 state = nullptr;
879}
880
881static OutputSection *findByName(ArrayRef<SectionCommand *> vec,
882 StringRef name) {
883 for (SectionCommand *cmd : vec)
884 if (auto *osd = dyn_cast<OutputDesc>(Val: cmd))
885 if (osd->osec.name == name)
886 return &osd->osec;
887 return nullptr;
888}
889
890static OutputDesc *createSection(Ctx &ctx, InputSectionBase *isec,
891 StringRef outsecName) {
892 OutputDesc *osd = ctx.script->createOutputSection(name: outsecName, location: "<internal>");
893 osd->osec.recordSection(isec);
894 return osd;
895}
896
897static OutputDesc *addInputSec(Ctx &ctx,
898 StringMap<TinyPtrVector<OutputSection *>> &map,
899 InputSectionBase *isec, StringRef outsecName) {
900 // Sections with SHT_GROUP or SHF_GROUP attributes reach here only when the -r
901 // option is given. A section with SHT_GROUP defines a "section group", and
902 // its members have SHF_GROUP attribute. Usually these flags have already been
903 // stripped by InputFiles.cpp as section groups are processed and uniquified.
904 // However, for the -r option, we want to pass through all section groups
905 // as-is because adding/removing members or merging them with other groups
906 // change their semantics.
907 if (isec->type == SHT_GROUP || (isec->flags & SHF_GROUP))
908 return createSection(ctx, isec, outsecName);
909
910 // Imagine .zed : { *(.foo) *(.bar) } script. Both foo and bar may have
911 // relocation sections .rela.foo and .rela.bar for example. Most tools do
912 // not allow multiple REL[A] sections for output section. Hence we
913 // should combine these relocation sections into single output.
914 // We skip synthetic sections because it can be .rela.dyn/.rela.plt or any
915 // other REL[A] sections created by linker itself.
916 if (!isa<SyntheticSection>(Val: isec) && isStaticRelSecType(type: isec->type)) {
917 auto *sec = cast<InputSection>(Val: isec);
918 OutputSection *out = sec->getRelocatedSection()->getOutputSection();
919
920 if (auto *relSec = out->relocationSection) {
921 relSec->recordSection(isec: sec);
922 return nullptr;
923 }
924
925 OutputDesc *osd = createSection(ctx, isec, outsecName);
926 out->relocationSection = &osd->osec;
927 return osd;
928 }
929
930 // The ELF spec just says
931 // ----------------------------------------------------------------
932 // In the first phase, input sections that match in name, type and
933 // attribute flags should be concatenated into single sections.
934 // ----------------------------------------------------------------
935 //
936 // However, it is clear that at least some flags have to be ignored for
937 // section merging. At the very least SHF_GROUP and SHF_COMPRESSED have to be
938 // ignored. We should not have two output .text sections just because one was
939 // in a group and another was not for example.
940 //
941 // It also seems that wording was a late addition and didn't get the
942 // necessary scrutiny.
943 //
944 // Merging sections with different flags is expected by some users. One
945 // reason is that if one file has
946 //
947 // int *const bar __attribute__((section(".foo"))) = (int *)0;
948 //
949 // gcc with -fPIC will produce a read only .foo section. But if another
950 // file has
951 //
952 // int zed;
953 // int *const bar __attribute__((section(".foo"))) = (int *)&zed;
954 //
955 // gcc with -fPIC will produce a read write section.
956 //
957 // Last but not least, when using linker script the merge rules are forced by
958 // the script. Unfortunately, linker scripts are name based. This means that
959 // expressions like *(.foo*) can refer to multiple input sections with
960 // different flags. We cannot put them in different output sections or we
961 // would produce wrong results for
962 //
963 // start = .; *(.foo.*) end = .; *(.bar)
964 //
965 // and a mapping of .foo1 and .bar1 to one section and .foo2 and .bar2 to
966 // another. The problem is that there is no way to layout those output
967 // sections such that the .foo sections are the only thing between the start
968 // and end symbols.
969 //
970 // Given the above issues, we instead merge sections by name and error on
971 // incompatible types and flags.
972 TinyPtrVector<OutputSection *> &v = map[outsecName];
973 for (OutputSection *sec : v) {
974 if (sec->partition != isec->partition)
975 continue;
976
977 if (ctx.arg.relocatable && (isec->flags & SHF_LINK_ORDER)) {
978 // Merging two SHF_LINK_ORDER sections with different sh_link fields will
979 // change their semantics, so we only merge them in -r links if they will
980 // end up being linked to the same output section. The casts are fine
981 // because everything in the map was created by the orphan placement code.
982 auto *firstIsec = cast<InputSectionBase>(
983 Val: cast<InputSectionDescription>(Val: sec->commands[0])->sectionBases[0]);
984 OutputSection *firstIsecOut =
985 (firstIsec->flags & SHF_LINK_ORDER)
986 ? firstIsec->getLinkOrderDep()->getOutputSection()
987 : nullptr;
988 if (firstIsecOut != isec->getLinkOrderDep()->getOutputSection())
989 continue;
990 }
991
992 sec->recordSection(isec);
993 return nullptr;
994 }
995
996 OutputDesc *osd = createSection(ctx, isec, outsecName);
997 v.push_back(NewVal: &osd->osec);
998 return osd;
999}
1000
1001// Add sections that didn't match any sections command.
1002void LinkerScript::addOrphanSections() {
1003 StringMap<TinyPtrVector<OutputSection *>> map;
1004 SmallVector<OutputDesc *, 0> v;
1005
1006 auto add = [&](InputSectionBase *s) {
1007 if (s->isLive() && !s->parent) {
1008 orphanSections.push_back(Elt: s);
1009
1010 StringRef name = getOutputSectionName(s);
1011 if (ctx.arg.unique) {
1012 v.push_back(Elt: createSection(ctx, isec: s, outsecName: name));
1013 } else if (OutputSection *sec = findByName(vec: sectionCommands, name)) {
1014 sec->recordSection(isec: s);
1015 } else {
1016 if (OutputDesc *osd = addInputSec(ctx, map, isec: s, outsecName: name))
1017 v.push_back(Elt: osd);
1018 assert(isa<MergeInputSection>(s) ||
1019 s->getOutputSection()->sectionIndex == UINT32_MAX);
1020 }
1021 }
1022 };
1023
1024 // For further --emit-reloc handling code we need target output section
1025 // to be created before we create relocation output section, so we want
1026 // to create target sections first. We do not want priority handling
1027 // for synthetic sections because them are special.
1028 size_t n = 0;
1029 for (InputSectionBase *isec : ctx.inputSections) {
1030 // Process InputSection and MergeInputSection.
1031 if (LLVM_LIKELY(isa<InputSection>(isec)))
1032 ctx.inputSections[n++] = isec;
1033
1034 // In -r links, SHF_LINK_ORDER sections are added while adding their parent
1035 // sections because we need to know the parent's output section before we
1036 // can select an output section for the SHF_LINK_ORDER section.
1037 if (ctx.arg.relocatable && (isec->flags & SHF_LINK_ORDER))
1038 continue;
1039
1040 if (auto *sec = dyn_cast<InputSection>(Val: isec))
1041 if (InputSectionBase *rel = sec->getRelocatedSection())
1042 if (auto *relIS = dyn_cast_or_null<InputSectionBase>(Val: rel->parent))
1043 add(relIS);
1044 add(isec);
1045 if (ctx.arg.relocatable)
1046 for (InputSectionBase *depSec : isec->dependentSections)
1047 if (depSec->flags & SHF_LINK_ORDER)
1048 add(depSec);
1049 }
1050 // Keep just InputSection.
1051 ctx.inputSections.resize(N: n);
1052
1053 // If no SECTIONS command was given, we should insert sections commands
1054 // before others, so that we can handle scripts which refers them,
1055 // for example: "foo = ABSOLUTE(ADDR(.text)));".
1056 // When SECTIONS command is present we just add all orphans to the end.
1057 if (hasSectionsCommand)
1058 sectionCommands.insert(I: sectionCommands.end(), From: v.begin(), To: v.end());
1059 else
1060 sectionCommands.insert(I: sectionCommands.begin(), From: v.begin(), To: v.end());
1061}
1062
1063void LinkerScript::diagnoseOrphanHandling() const {
1064 llvm::TimeTraceScope timeScope("Diagnose orphan sections");
1065 if (ctx.arg.orphanHandling == OrphanHandlingPolicy::Place ||
1066 !hasSectionsCommand)
1067 return;
1068 for (const InputSectionBase *sec : orphanSections) {
1069 // .relro_padding is inserted before DATA_SEGMENT_RELRO_END, if present,
1070 // automatically. The section is not supposed to be specified by scripts.
1071 if (sec == ctx.in.relroPadding.get())
1072 continue;
1073 // Input SHT_REL[A] retained by --emit-relocs are ignored by
1074 // computeInputSections(). Don't warn/error.
1075 if (isa<InputSection>(Val: sec) &&
1076 cast<InputSection>(Val: sec)->getRelocatedSection())
1077 continue;
1078
1079 StringRef name = getOutputSectionName(s: sec);
1080 if (ctx.arg.orphanHandling == OrphanHandlingPolicy::Error)
1081 ErrAlways(ctx) << sec << " is being placed in '" << name << "'";
1082 else
1083 Warn(ctx) << sec << " is being placed in '" << name << "'";
1084 }
1085}
1086
1087void LinkerScript::diagnoseMissingSGSectionAddress() const {
1088 if (!ctx.arg.cmseImplib || !ctx.in.armCmseSGSection->isNeeded())
1089 return;
1090
1091 OutputSection *sec = findByName(vec: sectionCommands, name: ".gnu.sgstubs");
1092 if (sec && !sec->addrExpr && !ctx.arg.sectionStartMap.count(Key: ".gnu.sgstubs"))
1093 ErrAlways(ctx) << "no address assigned to the veneers output section "
1094 << sec->name;
1095}
1096
1097// This function searches for a memory region to place the given output
1098// section in. If found, a pointer to the appropriate memory region is
1099// returned in the first member of the pair. Otherwise, a nullptr is returned.
1100// The second member of the pair is a hint that should be passed to the
1101// subsequent call of this method.
1102std::pair<MemoryRegion *, MemoryRegion *>
1103LinkerScript::findMemoryRegion(OutputSection *sec, MemoryRegion *hint) {
1104 // Non-allocatable sections are not part of the process image.
1105 if (!(sec->flags & SHF_ALLOC)) {
1106 bool hasInputOrByteCommand =
1107 sec->hasInputSections ||
1108 llvm::any_of(Range&: sec->commands, P: [](SectionCommand *comm) {
1109 return ByteCommand::classof(c: comm);
1110 });
1111 if (!sec->memoryRegionName.empty() && hasInputOrByteCommand)
1112 Warn(ctx)
1113 << "ignoring memory region assignment for non-allocatable section '"
1114 << sec->name << "'";
1115 return {nullptr, nullptr};
1116 }
1117
1118 // If a memory region name was specified in the output section command,
1119 // then try to find that region first.
1120 if (!sec->memoryRegionName.empty()) {
1121 if (MemoryRegion *m = memoryRegions.lookup(Key: sec->memoryRegionName))
1122 return {m, m};
1123 ErrAlways(ctx) << "memory region '" << sec->memoryRegionName
1124 << "' not declared";
1125 return {nullptr, nullptr};
1126 }
1127
1128 // If at least one memory region is defined, all sections must
1129 // belong to some memory region. Otherwise, we don't need to do
1130 // anything for memory regions.
1131 if (memoryRegions.empty())
1132 return {nullptr, nullptr};
1133
1134 // An orphan section should continue the previous memory region.
1135 if (sec->sectionIndex == UINT32_MAX && hint)
1136 return {hint, hint};
1137
1138 // See if a region can be found by matching section flags.
1139 for (auto &pair : memoryRegions) {
1140 MemoryRegion *m = pair.second;
1141 if (m->compatibleWith(secFlags: sec->flags))
1142 return {m, nullptr};
1143 }
1144
1145 // Otherwise, no suitable region was found.
1146 ErrAlways(ctx) << "no memory region specified for section '" << sec->name
1147 << "'";
1148 return {nullptr, nullptr};
1149}
1150
1151static OutputSection *findFirstSection(Ctx &ctx, PhdrEntry *load) {
1152 for (OutputSection *sec : ctx.outputSections)
1153 if (sec->ptLoad == load)
1154 return sec;
1155 return nullptr;
1156}
1157
1158// Assign addresses to an output section and offsets to its input sections and
1159// symbol assignments. Return true if the output section's address has changed.
1160bool LinkerScript::assignOffsets(OutputSection *sec) {
1161 const bool isTbss = (sec->flags & SHF_TLS) && sec->type == SHT_NOBITS;
1162 const bool sameMemRegion = state->memRegion == sec->memRegion;
1163 const bool prevLMARegionIsDefault = state->lmaRegion == nullptr;
1164 const uint64_t savedDot = dot;
1165 bool addressChanged = false;
1166 state->memRegion = sec->memRegion;
1167 state->lmaRegion = sec->lmaRegion;
1168
1169 if (!(sec->flags & SHF_ALLOC)) {
1170 // Non-SHF_ALLOC sections have zero addresses.
1171 dot = 0;
1172 } else if (isTbss) {
1173 // Allow consecutive SHF_TLS SHT_NOBITS output sections. The address range
1174 // starts from the end address of the previous tbss section.
1175 if (state->tbssAddr == 0)
1176 state->tbssAddr = dot;
1177 else
1178 dot = state->tbssAddr;
1179 } else {
1180 if (state->memRegion)
1181 dot = state->memRegion->curPos;
1182 if (sec->addrExpr)
1183 setDot(e: sec->addrExpr, loc: sec->location, inSec: false);
1184
1185 // If the address of the section has been moved forward by an explicit
1186 // expression so that it now starts past the current curPos of the enclosing
1187 // region, we need to expand the current region to account for the space
1188 // between the previous section, if any, and the start of this section.
1189 if (state->memRegion && state->memRegion->curPos < dot)
1190 expandMemoryRegion(memRegion: state->memRegion, size: dot - state->memRegion->curPos,
1191 secName: sec->name);
1192 }
1193
1194 state->outSec = sec;
1195 if (!(sec->addrExpr && hasSectionsCommand)) {
1196 // ALIGN is respected. sec->alignment is the max of ALIGN and the maximum of
1197 // input section alignments.
1198 const uint64_t pos = dot;
1199 dot = alignToPowerOf2(Value: dot, Align: sec->addralign);
1200 expandMemoryRegions(size: dot - pos);
1201 }
1202 addressChanged = sec->addr != dot;
1203 sec->addr = dot;
1204
1205 // state->lmaOffset is LMA minus VMA. If LMA is explicitly specified via AT()
1206 // or AT>, recompute state->lmaOffset; otherwise, if both previous/current LMA
1207 // region is the default, and the two sections are in the same memory region,
1208 // reuse previous lmaOffset; otherwise, reset lmaOffset to 0. This emulates
1209 // heuristics described in
1210 // https://sourceware.org/binutils/docs/ld/Output-Section-LMA.html
1211 if (sec->lmaExpr) {
1212 state->lmaOffset = sec->lmaExpr().getValue() - dot;
1213 } else if (MemoryRegion *mr = sec->lmaRegion) {
1214 uint64_t lmaStart = alignToPowerOf2(Value: mr->curPos, Align: sec->addralign);
1215 if (mr->curPos < lmaStart)
1216 expandMemoryRegion(memRegion: mr, size: lmaStart - mr->curPos, secName: sec->name);
1217 state->lmaOffset = lmaStart - dot;
1218 } else if (!sameMemRegion || !prevLMARegionIsDefault) {
1219 state->lmaOffset = 0;
1220 }
1221
1222 // Propagate state->lmaOffset to the first "non-header" section.
1223 if (PhdrEntry *l = sec->ptLoad)
1224 if (sec == findFirstSection(ctx, load: l))
1225 l->lmaOffset = state->lmaOffset;
1226
1227 // We can call this method multiple times during the creation of
1228 // thunks and want to start over calculation each time.
1229 sec->size = 0;
1230 if (sec->firstInOverlay)
1231 state->overlaySize = 0;
1232
1233 // We visited SectionsCommands from processSectionCommands to
1234 // layout sections. Now, we visit SectionsCommands again to fix
1235 // section offsets.
1236 for (SectionCommand *cmd : sec->commands) {
1237 // This handles the assignments to symbol or to the dot.
1238 if (auto *assign = dyn_cast<SymbolAssignment>(Val: cmd)) {
1239 assign->addr = dot;
1240 assignSymbol(cmd: assign, inSec: true);
1241 assign->size = dot - assign->addr;
1242 continue;
1243 }
1244
1245 // Handle BYTE(), SHORT(), LONG(), or QUAD().
1246 if (auto *data = dyn_cast<ByteCommand>(Val: cmd)) {
1247 data->offset = dot - sec->addr;
1248 dot += data->size;
1249 expandOutputSection(size: data->size);
1250 continue;
1251 }
1252
1253 // Handle a single input section description command.
1254 // It calculates and assigns the offsets for each section and also
1255 // updates the output section size.
1256
1257 auto &sections = cast<InputSectionDescription>(Val: cmd)->sections;
1258 for (InputSection *isec : sections) {
1259 assert(isec->getParent() == sec);
1260 if (isa<PotentialSpillSection>(Val: isec))
1261 continue;
1262 const uint64_t pos = dot;
1263 dot = alignToPowerOf2(Value: dot, Align: isec->addralign);
1264 isec->outSecOff = dot - sec->addr;
1265 dot += isec->getSize();
1266
1267 // Update output section size after adding each section. This is so that
1268 // SIZEOF works correctly in the case below:
1269 // .foo { *(.aaa) a = SIZEOF(.foo); *(.bbb) }
1270 expandOutputSection(size: dot - pos);
1271 }
1272 }
1273
1274 // If .relro_padding is present, round up the end to a common-page-size
1275 // boundary to protect the last page.
1276 if (ctx.in.relroPadding && sec == ctx.in.relroPadding->getParent())
1277 expandOutputSection(size: alignToPowerOf2(Value: dot, Align: ctx.arg.commonPageSize) - dot);
1278
1279 // Non-SHF_ALLOC sections do not affect the addresses of other OutputSections
1280 // as they are not part of the process image.
1281 if (!(sec->flags & SHF_ALLOC)) {
1282 dot = savedDot;
1283 } else if (isTbss) {
1284 // NOBITS TLS sections are similar. Additionally save the end address.
1285 state->tbssAddr = dot;
1286 dot = savedDot;
1287 }
1288 return addressChanged;
1289}
1290
1291static bool isDiscardable(const OutputSection &sec) {
1292 if (sec.name == "/DISCARD/")
1293 return true;
1294
1295 // We do not want to remove OutputSections with expressions that reference
1296 // symbols even if the OutputSection is empty. We want to ensure that the
1297 // expressions can be evaluated and report an error if they cannot.
1298 if (sec.expressionsUseSymbols)
1299 return false;
1300
1301 // OutputSections may be referenced by name in ADDR and LOADADDR expressions,
1302 // as an empty Section can has a valid VMA and LMA we keep the OutputSection
1303 // to maintain the integrity of the other Expression.
1304 if (sec.usedInExpression)
1305 return false;
1306
1307 for (SectionCommand *cmd : sec.commands) {
1308 if (auto assign = dyn_cast<SymbolAssignment>(Val: cmd))
1309 // Don't create empty output sections just for unreferenced PROVIDE
1310 // symbols.
1311 if (assign->name != "." && !assign->sym)
1312 continue;
1313
1314 if (!isa<InputSectionDescription>(Val: *cmd))
1315 return false;
1316 }
1317 return true;
1318}
1319
1320static void maybePropagatePhdrs(OutputSection &sec,
1321 SmallVector<StringRef, 0> &phdrs) {
1322 if (sec.phdrs.empty()) {
1323 // To match the bfd linker script behaviour, only propagate program
1324 // headers to sections that are allocated.
1325 if (sec.flags & SHF_ALLOC)
1326 sec.phdrs = phdrs;
1327 } else {
1328 phdrs = sec.phdrs;
1329 }
1330}
1331
1332void LinkerScript::adjustOutputSections() {
1333 // If the output section contains only symbol assignments, create a
1334 // corresponding output section. The issue is what to do with linker script
1335 // like ".foo : { symbol = 42; }". One option would be to convert it to
1336 // "symbol = 42;". That is, move the symbol out of the empty section
1337 // description. That seems to be what bfd does for this simple case. The
1338 // problem is that this is not completely general. bfd will give up and
1339 // create a dummy section too if there is a ". = . + 1" inside the section
1340 // for example.
1341 // Given that we want to create the section, we have to worry what impact
1342 // it will have on the link. For example, if we just create a section with
1343 // 0 for flags, it would change which PT_LOADs are created.
1344 // We could remember that particular section is dummy and ignore it in
1345 // other parts of the linker, but unfortunately there are quite a few places
1346 // that would need to change:
1347 // * The program header creation.
1348 // * The orphan section placement.
1349 // * The address assignment.
1350 // The other option is to pick flags that minimize the impact the section
1351 // will have on the rest of the linker. That is why we copy the flags from
1352 // the previous sections. We copy just SHF_ALLOC and SHF_WRITE to keep the
1353 // impact low. We do not propagate SHF_EXECINSTR as in some cases this can
1354 // lead to executable writeable section.
1355 uint64_t flags = SHF_ALLOC;
1356
1357 SmallVector<StringRef, 0> defPhdrs;
1358 bool seenRelro = false;
1359 for (SectionCommand *&cmd : sectionCommands) {
1360 if (!isa<OutputDesc>(Val: cmd))
1361 continue;
1362 auto *sec = &cast<OutputDesc>(Val: cmd)->osec;
1363
1364 // Handle align (e.g. ".foo : ALIGN(16) { ... }").
1365 if (sec->alignExpr)
1366 sec->addralign =
1367 std::max<uint32_t>(a: sec->addralign, b: sec->alignExpr().getValue());
1368
1369 bool isEmpty = (getFirstInputSection(os: sec) == nullptr);
1370 bool discardable = isEmpty && isDiscardable(sec: *sec);
1371 // If sec has at least one input section and not discarded, remember its
1372 // flags to be inherited by subsequent output sections. (sec may contain
1373 // just one empty synthetic section.)
1374 if (sec->hasInputSections && !discardable)
1375 flags = sec->flags;
1376
1377 // We do not want to keep any special flags for output section
1378 // in case it is empty.
1379 if (isEmpty) {
1380 sec->flags =
1381 flags & ((sec->nonAlloc ? 0 : (uint64_t)SHF_ALLOC) | SHF_WRITE);
1382 sec->sortRank = getSectionRank(ctx, osec&: *sec);
1383 }
1384
1385 // The code below may remove empty output sections. We should save the
1386 // specified program headers (if exist) and propagate them to subsequent
1387 // sections which do not specify program headers.
1388 // An example of such a linker script is:
1389 // SECTIONS { .empty : { *(.empty) } :rw
1390 // .foo : { *(.foo) } }
1391 // Note: at this point the order of output sections has not been finalized,
1392 // because orphans have not been inserted into their expected positions. We
1393 // will handle them in adjustSectionsAfterSorting().
1394 if (sec->sectionIndex != UINT32_MAX)
1395 maybePropagatePhdrs(sec&: *sec, phdrs&: defPhdrs);
1396
1397 // Discard .relro_padding if we have not seen one RELRO section. Note: when
1398 // .tbss is the only RELRO section, there is no associated PT_LOAD segment
1399 // (needsPtLoad), so we don't append .relro_padding in the case.
1400 if (ctx.in.relroPadding && ctx.in.relroPadding->getParent() == sec &&
1401 !seenRelro)
1402 discardable = true;
1403 if (discardable) {
1404 sec->markDead();
1405 cmd = nullptr;
1406 } else {
1407 seenRelro |=
1408 sec->relro && !(sec->type == SHT_NOBITS && (sec->flags & SHF_TLS));
1409 }
1410 }
1411
1412 // It is common practice to use very generic linker scripts. So for any
1413 // given run some of the output sections in the script will be empty.
1414 // We could create corresponding empty output sections, but that would
1415 // clutter the output.
1416 // We instead remove trivially empty sections. The bfd linker seems even
1417 // more aggressive at removing them.
1418 llvm::erase_if(C&: sectionCommands, P: [&](SectionCommand *cmd) { return !cmd; });
1419}
1420
1421void LinkerScript::adjustSectionsAfterSorting() {
1422 // Try and find an appropriate memory region to assign offsets in.
1423 MemoryRegion *hint = nullptr;
1424 for (SectionCommand *cmd : sectionCommands) {
1425 if (auto *osd = dyn_cast<OutputDesc>(Val: cmd)) {
1426 OutputSection *sec = &osd->osec;
1427 if (!sec->lmaRegionName.empty()) {
1428 if (MemoryRegion *m = memoryRegions.lookup(Key: sec->lmaRegionName))
1429 sec->lmaRegion = m;
1430 else
1431 ErrAlways(ctx) << "memory region '" << sec->lmaRegionName
1432 << "' not declared";
1433 }
1434 std::tie(args&: sec->memRegion, args&: hint) = findMemoryRegion(sec, hint);
1435 }
1436 }
1437
1438 // If output section command doesn't specify any segments,
1439 // and we haven't previously assigned any section to segment,
1440 // then we simply assign section to the very first load segment.
1441 // Below is an example of such linker script:
1442 // PHDRS { seg PT_LOAD; }
1443 // SECTIONS { .aaa : { *(.aaa) } }
1444 SmallVector<StringRef, 0> defPhdrs;
1445 auto firstPtLoad = llvm::find_if(Range&: phdrsCommands, P: [](const PhdrsCommand &cmd) {
1446 return cmd.type == PT_LOAD;
1447 });
1448 if (firstPtLoad != phdrsCommands.end())
1449 defPhdrs.push_back(Elt: firstPtLoad->name);
1450
1451 // Walk the commands and propagate the program headers to commands that don't
1452 // explicitly specify them.
1453 for (SectionCommand *cmd : sectionCommands)
1454 if (auto *osd = dyn_cast<OutputDesc>(Val: cmd))
1455 maybePropagatePhdrs(sec&: osd->osec, phdrs&: defPhdrs);
1456}
1457
1458// When the SECTIONS command is used, try to find an address for the file and
1459// program headers output sections, which can be added to the first PT_LOAD
1460// segment when program headers are created.
1461//
1462// We check if the headers fit below the first allocated section. If there isn't
1463// enough space for these sections, we'll remove them from the PT_LOAD segment,
1464// and we'll also remove the PT_PHDR segment.
1465void LinkerScript::allocateHeaders(
1466 SmallVector<std::unique_ptr<PhdrEntry>, 0> &phdrs) {
1467 uint64_t min = std::numeric_limits<uint64_t>::max();
1468 for (OutputSection *sec : ctx.outputSections)
1469 if (sec->flags & SHF_ALLOC)
1470 min = std::min<uint64_t>(a: min, b: sec->addr);
1471
1472 auto it = llvm::find_if(Range&: phdrs, P: [](auto &e) { return e->p_type == PT_LOAD; });
1473 if (it == phdrs.end())
1474 return;
1475 PhdrEntry *firstPTLoad = it->get();
1476
1477 bool hasExplicitHeaders =
1478 llvm::any_of(Range&: phdrsCommands, P: [](const PhdrsCommand &cmd) {
1479 return cmd.hasPhdrs || cmd.hasFilehdr;
1480 });
1481 bool paged = !ctx.arg.omagic && !ctx.arg.nmagic;
1482 uint64_t headerSize = getHeaderSize(ctx);
1483
1484 uint64_t base = 0;
1485 // If SECTIONS is present and the linkerscript is not explicit about program
1486 // headers, only allocate program headers if that would not add a page.
1487 if (hasSectionsCommand && !hasExplicitHeaders)
1488 base = alignDown(Value: min, Align: ctx.arg.maxPageSize);
1489 if ((paged || hasExplicitHeaders) && headerSize <= min - base) {
1490 min = alignDown(Value: min - headerSize, Align: ctx.arg.maxPageSize);
1491 ctx.out.elfHeader->addr = min;
1492 ctx.out.programHeaders->addr = min + ctx.out.elfHeader->size;
1493 return;
1494 }
1495
1496 // Error if we were explicitly asked to allocate headers.
1497 if (hasExplicitHeaders)
1498 ErrAlways(ctx) << "could not allocate headers";
1499
1500 ctx.out.elfHeader->ptLoad = nullptr;
1501 ctx.out.programHeaders->ptLoad = nullptr;
1502 firstPTLoad->firstSec = findFirstSection(ctx, load: firstPTLoad);
1503
1504 llvm::erase_if(C&: phdrs, P: [](auto &e) { return e->p_type == PT_PHDR; });
1505}
1506
1507LinkerScript::AddressState::AddressState(const LinkerScript &script) {
1508 for (auto &mri : script.memoryRegions) {
1509 MemoryRegion *mr = mri.second;
1510 mr->curPos = (mr->origin)().getValue();
1511 }
1512}
1513
1514// Here we assign addresses as instructed by linker script SECTIONS
1515// sub-commands. Doing that allows us to use final VA values, so here
1516// we also handle rest commands like symbol assignments and ASSERTs.
1517// Return an output section that has changed its address or null, and a symbol
1518// that has changed its section or value (or nullptr if no symbol has changed).
1519std::pair<const OutputSection *, const Defined *>
1520LinkerScript::assignAddresses() {
1521 if (hasSectionsCommand) {
1522 // With a linker script, assignment of addresses to headers is covered by
1523 // allocateHeaders().
1524 dot = ctx.arg.imageBase.value_or(u: 0);
1525 } else {
1526 // Assign addresses to headers right now.
1527 dot = ctx.target->getImageBase();
1528 ctx.out.elfHeader->addr = dot;
1529 ctx.out.programHeaders->addr = dot + ctx.out.elfHeader->size;
1530 dot += getHeaderSize(ctx);
1531 }
1532
1533 OutputSection *changedOsec = nullptr;
1534 AddressState st(*this);
1535 state = &st;
1536 errorOnMissingSection = true;
1537 st.outSec = aether.get();
1538 recordedErrors.clear();
1539
1540 SymbolAssignmentMap oldValues = getSymbolAssignmentValues(sectionCommands);
1541 for (SectionCommand *cmd : sectionCommands) {
1542 if (auto *assign = dyn_cast<SymbolAssignment>(Val: cmd)) {
1543 assign->addr = dot;
1544 assignSymbol(cmd: assign, inSec: false);
1545 assign->size = dot - assign->addr;
1546 continue;
1547 }
1548 if (isa<SectionClassDesc>(Val: cmd))
1549 continue;
1550 if (assignOffsets(sec: &cast<OutputDesc>(Val: cmd)->osec) && !changedOsec)
1551 changedOsec = &cast<OutputDesc>(Val: cmd)->osec;
1552 }
1553
1554 state = nullptr;
1555 return {changedOsec, getChangedSymbolAssignment(oldValues)};
1556}
1557
1558static bool hasRegionOverflowed(MemoryRegion *mr) {
1559 if (!mr)
1560 return false;
1561 return mr->curPos - mr->getOrigin() > mr->getLength();
1562}
1563
1564// Spill input sections in reverse order of address assignment to (potentially)
1565// bring memory regions out of overflow. The size savings of a spill can only be
1566// estimated, since general linker script arithmetic may occur afterwards.
1567// Under-estimates may cause unnecessary spills, but over-estimates can always
1568// be corrected on the next pass.
1569bool LinkerScript::spillSections() {
1570 if (potentialSpillLists.empty())
1571 return false;
1572
1573 DenseSet<PotentialSpillSection *> skippedSpills;
1574
1575 bool spilled = false;
1576 for (SectionCommand *cmd : reverse(C&: sectionCommands)) {
1577 auto *osd = dyn_cast<OutputDesc>(Val: cmd);
1578 if (!osd)
1579 continue;
1580 OutputSection *osec = &osd->osec;
1581 if (!osec->memRegion)
1582 continue;
1583
1584 // Input sections that have replaced a potential spill and should be removed
1585 // from their input section description.
1586 DenseSet<InputSection *> spilledInputSections;
1587
1588 for (SectionCommand *cmd : reverse(C&: osec->commands)) {
1589 if (!hasRegionOverflowed(mr: osec->memRegion) &&
1590 !hasRegionOverflowed(mr: osec->lmaRegion))
1591 break;
1592
1593 auto *isd = dyn_cast<InputSectionDescription>(Val: cmd);
1594 if (!isd)
1595 continue;
1596 for (InputSection *isec : reverse(C&: isd->sections)) {
1597 // Potential spill locations cannot be spilled.
1598 if (isa<PotentialSpillSection>(Val: isec))
1599 continue;
1600
1601 auto it = potentialSpillLists.find(Val: isec);
1602 if (it == potentialSpillLists.end())
1603 break;
1604
1605 // Consume spills until finding one that might help, then consume it.
1606 auto canSpillHelp = [&](PotentialSpillSection *spill) {
1607 // Spills to the same region that overflowed cannot help.
1608 if (hasRegionOverflowed(mr: osec->memRegion) &&
1609 spill->getParent()->memRegion == osec->memRegion)
1610 return false;
1611 if (hasRegionOverflowed(mr: osec->lmaRegion) &&
1612 spill->getParent()->lmaRegion == osec->lmaRegion)
1613 return false;
1614 return true;
1615 };
1616 PotentialSpillList &list = it->second;
1617 PotentialSpillSection *spill;
1618 for (spill = list.head; spill; spill = spill->next) {
1619 if (list.head->next)
1620 list.head = spill->next;
1621 else
1622 potentialSpillLists.erase(Val: isec);
1623 if (canSpillHelp(spill))
1624 break;
1625 skippedSpills.insert(V: spill);
1626 }
1627 if (!spill)
1628 continue;
1629
1630 // Replace the next spill location with the spilled section and adjust
1631 // its properties to match the new location. Note that the alignment of
1632 // the spill section may have diverged from the original due to e.g. a
1633 // SUBALIGN. Correct assignment requires the spill's alignment to be
1634 // used, not the original.
1635 spilledInputSections.insert(V: isec);
1636 *llvm::find(Range&: spill->isd->sections, Val: spill) = isec;
1637 isec->parent = spill->parent;
1638 isec->addralign = spill->addralign;
1639
1640 // Record the (potential) reduction in the region's end position.
1641 osec->memRegion->curPos -= isec->getSize();
1642 if (osec->lmaRegion)
1643 osec->lmaRegion->curPos -= isec->getSize();
1644
1645 // Spilling continues until the end position no longer overflows the
1646 // region. Then, another round of address assignment will either confirm
1647 // the spill's success or lead to yet more spilling.
1648 if (!hasRegionOverflowed(mr: osec->memRegion) &&
1649 !hasRegionOverflowed(mr: osec->lmaRegion))
1650 break;
1651 }
1652
1653 // Remove any spilled input sections to complete their move.
1654 if (!spilledInputSections.empty()) {
1655 spilled = true;
1656 llvm::erase_if(C&: isd->sections, P: [&](InputSection *isec) {
1657 return spilledInputSections.contains(V: isec);
1658 });
1659 }
1660 }
1661 }
1662
1663 // Clean up any skipped spills.
1664 DenseSet<InputSectionDescription *> isds;
1665 for (PotentialSpillSection *s : skippedSpills)
1666 isds.insert(V: s->isd);
1667 for (InputSectionDescription *isd : isds)
1668 llvm::erase_if(C&: isd->sections, P: [&](InputSection *s) {
1669 return skippedSpills.contains(V: dyn_cast<PotentialSpillSection>(Val: s));
1670 });
1671
1672 return spilled;
1673}
1674
1675// Erase any potential spill sections that were not used.
1676void LinkerScript::erasePotentialSpillSections() {
1677 if (potentialSpillLists.empty())
1678 return;
1679
1680 // Collect the set of input section descriptions that contain potential
1681 // spills.
1682 DenseSet<InputSectionDescription *> isds;
1683 for (const auto &[_, list] : potentialSpillLists)
1684 for (PotentialSpillSection *s = list.head; s; s = s->next)
1685 isds.insert(V: s->isd);
1686
1687 for (InputSectionDescription *isd : isds)
1688 llvm::erase_if(C&: isd->sections, P: [](InputSection *s) {
1689 return isa<PotentialSpillSection>(Val: s);
1690 });
1691
1692 potentialSpillLists.clear();
1693}
1694
1695// Creates program headers as instructed by PHDRS linker script command.
1696SmallVector<std::unique_ptr<PhdrEntry>, 0> LinkerScript::createPhdrs() {
1697 SmallVector<std::unique_ptr<PhdrEntry>, 0> ret;
1698
1699 // Process PHDRS and FILEHDR keywords because they are not
1700 // real output sections and cannot be added in the following loop.
1701 for (const PhdrsCommand &cmd : phdrsCommands) {
1702 auto phdr =
1703 std::make_unique<PhdrEntry>(args&: ctx, args: cmd.type, args: cmd.flags.value_or(u: PF_R));
1704
1705 if (cmd.hasFilehdr)
1706 phdr->add(sec: ctx.out.elfHeader.get());
1707 if (cmd.hasPhdrs)
1708 phdr->add(sec: ctx.out.programHeaders.get());
1709
1710 if (cmd.lmaExpr) {
1711 phdr->p_paddr = cmd.lmaExpr().getValue();
1712 phdr->hasLMA = true;
1713 }
1714 ret.push_back(Elt: std::move(phdr));
1715 }
1716
1717 // Add output sections to program headers.
1718 for (OutputSection *sec : ctx.outputSections) {
1719 // Assign headers specified by linker script
1720 for (size_t id : getPhdrIndices(sec)) {
1721 ret[id]->add(sec);
1722 if (!phdrsCommands[id].flags)
1723 ret[id]->p_flags |= sec->getPhdrFlags();
1724 }
1725 }
1726 return ret;
1727}
1728
1729// Returns true if we should emit an .interp section.
1730//
1731// We usually do. But if PHDRS commands are given, and
1732// no PT_INTERP is there, there's no place to emit an
1733// .interp, so we don't do that in that case.
1734bool LinkerScript::needsInterpSection() {
1735 if (phdrsCommands.empty())
1736 return true;
1737 for (PhdrsCommand &cmd : phdrsCommands)
1738 if (cmd.type == PT_INTERP)
1739 return true;
1740 return false;
1741}
1742
1743ExprValue LinkerScript::getSymbolValue(StringRef name, const Twine &loc) {
1744 if (name == ".") {
1745 if (state)
1746 return {state->outSec, false, dot - state->outSec->addr, loc};
1747 ErrAlways(ctx) << loc << ": unable to get location counter value";
1748 return 0;
1749 }
1750
1751 if (Symbol *sym = ctx.symtab->find(name)) {
1752 if (auto *ds = dyn_cast<Defined>(Val: sym)) {
1753 ExprValue v{ds->section, false, ds->value, loc};
1754 // Retain the original st_type, so that the alias will get the same
1755 // behavior in relocation processing. Any operation will reset st_type to
1756 // STT_NOTYPE.
1757 v.type = ds->type;
1758 return v;
1759 }
1760 if (isa<SharedSymbol>(Val: sym))
1761 if (!errorOnMissingSection)
1762 return {nullptr, false, 0, loc};
1763 }
1764
1765 ErrAlways(ctx) << loc << ": symbol not found: " << name;
1766 return 0;
1767}
1768
1769// Returns the index of the segment named Name.
1770static std::optional<size_t> getPhdrIndex(ArrayRef<PhdrsCommand> vec,
1771 StringRef name) {
1772 for (size_t i = 0; i < vec.size(); ++i)
1773 if (vec[i].name == name)
1774 return i;
1775 return std::nullopt;
1776}
1777
1778// Returns indices of ELF headers containing specific section. Each index is a
1779// zero based number of ELF header listed within PHDRS {} script block.
1780SmallVector<size_t, 0> LinkerScript::getPhdrIndices(OutputSection *cmd) {
1781 SmallVector<size_t, 0> ret;
1782
1783 for (StringRef s : cmd->phdrs) {
1784 if (std::optional<size_t> idx = getPhdrIndex(vec: phdrsCommands, name: s))
1785 ret.push_back(Elt: *idx);
1786 else if (s != "NONE")
1787 ErrAlways(ctx) << cmd->location << ": program header '" << s
1788 << "' is not listed in PHDRS";
1789 }
1790 return ret;
1791}
1792
1793void LinkerScript::printMemoryUsage(raw_ostream& os) {
1794 auto printSize = [&](uint64_t size) {
1795 if ((size & 0x3fffffff) == 0)
1796 os << format_decimal(N: size >> 30, Width: 10) << " GB";
1797 else if ((size & 0xfffff) == 0)
1798 os << format_decimal(N: size >> 20, Width: 10) << " MB";
1799 else if ((size & 0x3ff) == 0)
1800 os << format_decimal(N: size >> 10, Width: 10) << " KB";
1801 else
1802 os << " " << format_decimal(N: size, Width: 10) << " B";
1803 };
1804 os << "Memory region Used Size Region Size %age Used\n";
1805 for (auto &pair : memoryRegions) {
1806 MemoryRegion *m = pair.second;
1807 uint64_t usedLength = m->curPos - m->getOrigin();
1808 os << right_justify(Str: m->name, Width: 16) << ": ";
1809 printSize(usedLength);
1810 uint64_t length = m->getLength();
1811 if (length != 0) {
1812 printSize(length);
1813 double percent = usedLength * 100.0 / length;
1814 os << " " << format(Fmt: "%6.2f%%", Vals: percent);
1815 }
1816 os << '\n';
1817 }
1818}
1819
1820void LinkerScript::recordError(const Twine &msg) {
1821 auto &str = recordedErrors.emplace_back();
1822 msg.toVector(Out&: str);
1823}
1824
1825static void checkMemoryRegion(Ctx &ctx, const MemoryRegion *region,
1826 const OutputSection *osec, uint64_t addr) {
1827 uint64_t osecEnd = addr + osec->size;
1828 uint64_t regionEnd = region->getOrigin() + region->getLength();
1829 if (osecEnd > regionEnd) {
1830 ErrAlways(ctx) << "section '" << osec->name << "' will not fit in region '"
1831 << region->name << "': overflowed by "
1832 << (osecEnd - regionEnd) << " bytes";
1833 }
1834}
1835
1836void LinkerScript::checkFinalScriptConditions() const {
1837 for (StringRef err : recordedErrors)
1838 Err(ctx) << err;
1839 for (const OutputSection *sec : ctx.outputSections) {
1840 if (const MemoryRegion *memoryRegion = sec->memRegion)
1841 checkMemoryRegion(ctx, region: memoryRegion, osec: sec, addr: sec->addr);
1842 if (const MemoryRegion *lmaRegion = sec->lmaRegion)
1843 checkMemoryRegion(ctx, region: lmaRegion, osec: sec, addr: sec->getLMA());
1844 }
1845}
1846
1847void LinkerScript::addScriptReferencedSymbolsToSymTable() {
1848 // Some symbols (such as __ehdr_start) are defined lazily only when there
1849 // are undefined symbols for them, so we add these to trigger that logic.
1850 auto reference = [&ctx = ctx](StringRef name) {
1851 Symbol *sym = ctx.symtab->addUnusedUndefined(name);
1852 sym->isUsedInRegularObj = true;
1853 sym->referenced = true;
1854 };
1855 for (StringRef name : referencedSymbols)
1856 reference(name);
1857
1858 // Keeps track of references from which PROVIDE symbols have been added to the
1859 // symbol table.
1860 DenseSet<StringRef> added;
1861 SmallVector<const SmallVector<StringRef, 0> *, 0> symRefsVec;
1862 for (const auto &[name, symRefs] : provideMap)
1863 if (shouldAddProvideSym(symName: name) && added.insert(V: name).second)
1864 symRefsVec.push_back(Elt: &symRefs);
1865 while (symRefsVec.size()) {
1866 for (StringRef name : *symRefsVec.pop_back_val()) {
1867 reference(name);
1868 // Prevent the symbol from being discarded by --gc-sections.
1869 referencedSymbols.push_back(Elt: name);
1870 auto it = provideMap.find(Key: name);
1871 if (it != provideMap.end() && shouldAddProvideSym(symName: name) &&
1872 added.insert(V: name).second) {
1873 symRefsVec.push_back(Elt: &it->second);
1874 }
1875 }
1876 }
1877}
1878
1879bool LinkerScript::shouldAddProvideSym(StringRef symName) {
1880 // This function is called before and after garbage collection. To prevent
1881 // undefined references from the RHS, the result of this function for a
1882 // symbol must be the same for each call. We use unusedProvideSyms to not
1883 // change the return value of a demoted symbol.
1884 Symbol *sym = ctx.symtab->find(name: symName);
1885 if (!sym)
1886 return false;
1887 if (sym->isDefined() || sym->isCommon()) {
1888 unusedProvideSyms.insert(V: sym);
1889 return false;
1890 }
1891 return !unusedProvideSyms.count(V: sym);
1892}
1893