|
150
|
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 "InputSection.h"
|
|
|
16 #include "OutputSections.h"
|
|
|
17 #include "SymbolTable.h"
|
|
|
18 #include "Symbols.h"
|
|
|
19 #include "SyntheticSections.h"
|
|
|
20 #include "Target.h"
|
|
|
21 #include "Writer.h"
|
|
|
22 #include "lld/Common/Memory.h"
|
|
|
23 #include "lld/Common/Strings.h"
|
|
|
24 #include "llvm/ADT/STLExtras.h"
|
|
|
25 #include "llvm/ADT/StringRef.h"
|
|
|
26 #include "llvm/BinaryFormat/ELF.h"
|
|
|
27 #include "llvm/Support/Casting.h"
|
|
|
28 #include "llvm/Support/Endian.h"
|
|
|
29 #include "llvm/Support/ErrorHandling.h"
|
|
|
30 #include "llvm/Support/FileSystem.h"
|
|
173
|
31 #include "llvm/Support/Parallel.h"
|
|
150
|
32 #include "llvm/Support/Path.h"
|
|
|
33 #include <algorithm>
|
|
|
34 #include <cassert>
|
|
|
35 #include <cstddef>
|
|
|
36 #include <cstdint>
|
|
|
37 #include <iterator>
|
|
|
38 #include <limits>
|
|
|
39 #include <string>
|
|
|
40 #include <vector>
|
|
|
41
|
|
|
42 using namespace llvm;
|
|
|
43 using namespace llvm::ELF;
|
|
|
44 using namespace llvm::object;
|
|
|
45 using namespace llvm::support::endian;
|
|
173
|
46 using namespace lld;
|
|
|
47 using namespace lld::elf;
|
|
150
|
48
|
|
173
|
49 LinkerScript *elf::script;
|
|
150
|
50
|
|
|
51 static uint64_t getOutputSectionVA(SectionBase *sec) {
|
|
|
52 OutputSection *os = sec->getOutputSection();
|
|
|
53 assert(os && "input section has no output section assigned");
|
|
|
54 return os ? os->addr : 0;
|
|
|
55 }
|
|
|
56
|
|
|
57 uint64_t ExprValue::getValue() const {
|
|
|
58 if (sec)
|
|
|
59 return alignTo(sec->getOffset(val) + getOutputSectionVA(sec),
|
|
|
60 alignment);
|
|
|
61 return alignTo(val, alignment);
|
|
|
62 }
|
|
|
63
|
|
|
64 uint64_t ExprValue::getSecAddr() const {
|
|
|
65 if (sec)
|
|
|
66 return sec->getOffset(0) + getOutputSectionVA(sec);
|
|
|
67 return 0;
|
|
|
68 }
|
|
|
69
|
|
|
70 uint64_t ExprValue::getSectionOffset() const {
|
|
|
71 // If the alignment is trivial, we don't have to compute the full
|
|
|
72 // value to know the offset. This allows this function to succeed in
|
|
|
73 // cases where the output section is not yet known.
|
|
|
74 if (alignment == 1 && !sec)
|
|
|
75 return val;
|
|
|
76 return getValue() - getSecAddr();
|
|
|
77 }
|
|
|
78
|
|
|
79 OutputSection *LinkerScript::createOutputSection(StringRef name,
|
|
|
80 StringRef location) {
|
|
|
81 OutputSection *&secRef = nameToOutputSection[name];
|
|
|
82 OutputSection *sec;
|
|
|
83 if (secRef && secRef->location.empty()) {
|
|
|
84 // There was a forward reference.
|
|
|
85 sec = secRef;
|
|
|
86 } else {
|
|
|
87 sec = make<OutputSection>(name, SHT_PROGBITS, 0);
|
|
|
88 if (!secRef)
|
|
|
89 secRef = sec;
|
|
|
90 }
|
|
|
91 sec->location = std::string(location);
|
|
|
92 return sec;
|
|
|
93 }
|
|
|
94
|
|
|
95 OutputSection *LinkerScript::getOrCreateOutputSection(StringRef name) {
|
|
|
96 OutputSection *&cmdRef = nameToOutputSection[name];
|
|
|
97 if (!cmdRef)
|
|
|
98 cmdRef = make<OutputSection>(name, SHT_PROGBITS, 0);
|
|
|
99 return cmdRef;
|
|
|
100 }
|
|
|
101
|
|
|
102 // Expands the memory region by the specified size.
|
|
|
103 static void expandMemoryRegion(MemoryRegion *memRegion, uint64_t size,
|
|
|
104 StringRef regionName, StringRef secName) {
|
|
|
105 memRegion->curPos += size;
|
|
173
|
106 uint64_t newSize = memRegion->curPos - (memRegion->origin)().getValue();
|
|
|
107 uint64_t length = (memRegion->length)().getValue();
|
|
|
108 if (newSize > length)
|
|
150
|
109 error("section '" + secName + "' will not fit in region '" + regionName +
|
|
173
|
110 "': overflowed by " + Twine(newSize - length) + " bytes");
|
|
150
|
111 }
|
|
|
112
|
|
|
113 void LinkerScript::expandMemoryRegions(uint64_t size) {
|
|
|
114 if (ctx->memRegion)
|
|
|
115 expandMemoryRegion(ctx->memRegion, size, ctx->memRegion->name,
|
|
|
116 ctx->outSec->name);
|
|
|
117 // Only expand the LMARegion if it is different from memRegion.
|
|
|
118 if (ctx->lmaRegion && ctx->memRegion != ctx->lmaRegion)
|
|
|
119 expandMemoryRegion(ctx->lmaRegion, size, ctx->lmaRegion->name,
|
|
|
120 ctx->outSec->name);
|
|
|
121 }
|
|
|
122
|
|
|
123 void LinkerScript::expandOutputSection(uint64_t size) {
|
|
|
124 ctx->outSec->size += size;
|
|
|
125 expandMemoryRegions(size);
|
|
|
126 }
|
|
|
127
|
|
|
128 void LinkerScript::setDot(Expr e, const Twine &loc, bool inSec) {
|
|
|
129 uint64_t val = e().getValue();
|
|
|
130 if (val < dot && inSec)
|
|
|
131 error(loc + ": unable to move location counter backward for: " +
|
|
|
132 ctx->outSec->name);
|
|
|
133
|
|
|
134 // Update to location counter means update to section size.
|
|
|
135 if (inSec)
|
|
|
136 expandOutputSection(val - dot);
|
|
|
137
|
|
|
138 dot = val;
|
|
|
139 }
|
|
|
140
|
|
|
141 // Used for handling linker symbol assignments, for both finalizing
|
|
|
142 // their values and doing early declarations. Returns true if symbol
|
|
|
143 // should be defined from linker script.
|
|
|
144 static bool shouldDefineSym(SymbolAssignment *cmd) {
|
|
|
145 if (cmd->name == ".")
|
|
|
146 return false;
|
|
|
147
|
|
|
148 if (!cmd->provide)
|
|
|
149 return true;
|
|
|
150
|
|
|
151 // If a symbol was in PROVIDE(), we need to define it only
|
|
|
152 // when it is a referenced undefined symbol.
|
|
|
153 Symbol *b = symtab->find(cmd->name);
|
|
|
154 if (b && !b->isDefined())
|
|
|
155 return true;
|
|
|
156 return false;
|
|
|
157 }
|
|
|
158
|
|
|
159 // Called by processSymbolAssignments() to assign definitions to
|
|
|
160 // linker-script-defined symbols.
|
|
|
161 void LinkerScript::addSymbol(SymbolAssignment *cmd) {
|
|
|
162 if (!shouldDefineSym(cmd))
|
|
|
163 return;
|
|
|
164
|
|
|
165 // Define a symbol.
|
|
|
166 ExprValue value = cmd->expression();
|
|
|
167 SectionBase *sec = value.isAbsolute() ? nullptr : value.sec;
|
|
|
168 uint8_t visibility = cmd->hidden ? STV_HIDDEN : STV_DEFAULT;
|
|
|
169
|
|
|
170 // When this function is called, section addresses have not been
|
|
|
171 // fixed yet. So, we may or may not know the value of the RHS
|
|
|
172 // expression.
|
|
|
173 //
|
|
|
174 // For example, if an expression is `x = 42`, we know x is always 42.
|
|
|
175 // However, if an expression is `x = .`, there's no way to know its
|
|
|
176 // value at the moment.
|
|
|
177 //
|
|
|
178 // We want to set symbol values early if we can. This allows us to
|
|
|
179 // use symbols as variables in linker scripts. Doing so allows us to
|
|
|
180 // write expressions like this: `alignment = 16; . = ALIGN(., alignment)`.
|
|
|
181 uint64_t symValue = value.sec ? 0 : value.getValue();
|
|
|
182
|
|
|
183 Defined newSym(nullptr, cmd->name, STB_GLOBAL, visibility, STT_NOTYPE,
|
|
|
184 symValue, 0, sec);
|
|
|
185
|
|
|
186 Symbol *sym = symtab->insert(cmd->name);
|
|
|
187 sym->mergeProperties(newSym);
|
|
|
188 sym->replace(newSym);
|
|
|
189 cmd->sym = cast<Defined>(sym);
|
|
|
190 }
|
|
|
191
|
|
|
192 // This function is called from LinkerScript::declareSymbols.
|
|
|
193 // It creates a placeholder symbol if needed.
|
|
|
194 static void declareSymbol(SymbolAssignment *cmd) {
|
|
|
195 if (!shouldDefineSym(cmd))
|
|
|
196 return;
|
|
|
197
|
|
|
198 uint8_t visibility = cmd->hidden ? STV_HIDDEN : STV_DEFAULT;
|
|
|
199 Defined newSym(nullptr, cmd->name, STB_GLOBAL, visibility, STT_NOTYPE, 0, 0,
|
|
|
200 nullptr);
|
|
|
201
|
|
|
202 // We can't calculate final value right now.
|
|
|
203 Symbol *sym = symtab->insert(cmd->name);
|
|
|
204 sym->mergeProperties(newSym);
|
|
|
205 sym->replace(newSym);
|
|
|
206
|
|
|
207 cmd->sym = cast<Defined>(sym);
|
|
|
208 cmd->provide = false;
|
|
|
209 sym->scriptDefined = true;
|
|
|
210 }
|
|
|
211
|
|
|
212 using SymbolAssignmentMap =
|
|
|
213 DenseMap<const Defined *, std::pair<SectionBase *, uint64_t>>;
|
|
|
214
|
|
|
215 // Collect section/value pairs of linker-script-defined symbols. This is used to
|
|
|
216 // check whether symbol values converge.
|
|
|
217 static SymbolAssignmentMap
|
|
|
218 getSymbolAssignmentValues(const std::vector<BaseCommand *> §ionCommands) {
|
|
|
219 SymbolAssignmentMap ret;
|
|
|
220 for (BaseCommand *base : sectionCommands) {
|
|
|
221 if (auto *cmd = dyn_cast<SymbolAssignment>(base)) {
|
|
|
222 if (cmd->sym) // sym is nullptr for dot.
|
|
|
223 ret.try_emplace(cmd->sym,
|
|
|
224 std::make_pair(cmd->sym->section, cmd->sym->value));
|
|
|
225 continue;
|
|
|
226 }
|
|
|
227 for (BaseCommand *sub_base : cast<OutputSection>(base)->sectionCommands)
|
|
|
228 if (auto *cmd = dyn_cast<SymbolAssignment>(sub_base))
|
|
|
229 if (cmd->sym)
|
|
|
230 ret.try_emplace(cmd->sym,
|
|
|
231 std::make_pair(cmd->sym->section, cmd->sym->value));
|
|
|
232 }
|
|
|
233 return ret;
|
|
|
234 }
|
|
|
235
|
|
|
236 // Returns the lexicographical smallest (for determinism) Defined whose
|
|
|
237 // section/value has changed.
|
|
|
238 static const Defined *
|
|
|
239 getChangedSymbolAssignment(const SymbolAssignmentMap &oldValues) {
|
|
|
240 const Defined *changed = nullptr;
|
|
|
241 for (auto &it : oldValues) {
|
|
|
242 const Defined *sym = it.first;
|
|
|
243 if (std::make_pair(sym->section, sym->value) != it.second &&
|
|
|
244 (!changed || sym->getName() < changed->getName()))
|
|
|
245 changed = sym;
|
|
|
246 }
|
|
|
247 return changed;
|
|
|
248 }
|
|
|
249
|
|
|
250 // Process INSERT [AFTER|BEFORE] commands. For each command, we move the
|
|
|
251 // specified output section to the designated place.
|
|
|
252 void LinkerScript::processInsertCommands() {
|
|
|
253 for (const InsertCommand &cmd : insertCommands) {
|
|
|
254 // If cmd.os is empty, it may have been discarded by
|
|
|
255 // adjustSectionsBeforeSorting(). We do not handle such output sections.
|
|
|
256 auto from = llvm::find(sectionCommands, cmd.os);
|
|
|
257 if (from == sectionCommands.end())
|
|
|
258 continue;
|
|
|
259 sectionCommands.erase(from);
|
|
|
260
|
|
|
261 auto insertPos = llvm::find_if(sectionCommands, [&cmd](BaseCommand *base) {
|
|
|
262 auto *to = dyn_cast<OutputSection>(base);
|
|
|
263 return to != nullptr && to->name == cmd.where;
|
|
|
264 });
|
|
|
265 if (insertPos == sectionCommands.end()) {
|
|
|
266 error("unable to insert " + cmd.os->name +
|
|
|
267 (cmd.isAfter ? " after " : " before ") + cmd.where);
|
|
|
268 } else {
|
|
|
269 if (cmd.isAfter)
|
|
|
270 ++insertPos;
|
|
|
271 sectionCommands.insert(insertPos, cmd.os);
|
|
|
272 }
|
|
|
273 }
|
|
|
274 }
|
|
|
275
|
|
|
276 // Symbols defined in script should not be inlined by LTO. At the same time
|
|
|
277 // we don't know their final values until late stages of link. Here we scan
|
|
|
278 // over symbol assignment commands and create placeholder symbols if needed.
|
|
|
279 void LinkerScript::declareSymbols() {
|
|
|
280 assert(!ctx);
|
|
|
281 for (BaseCommand *base : sectionCommands) {
|
|
|
282 if (auto *cmd = dyn_cast<SymbolAssignment>(base)) {
|
|
|
283 declareSymbol(cmd);
|
|
|
284 continue;
|
|
|
285 }
|
|
|
286
|
|
|
287 // If the output section directive has constraints,
|
|
|
288 // we can't say for sure if it is going to be included or not.
|
|
|
289 // Skip such sections for now. Improve the checks if we ever
|
|
|
290 // need symbols from that sections to be declared early.
|
|
|
291 auto *sec = cast<OutputSection>(base);
|
|
|
292 if (sec->constraint != ConstraintKind::NoConstraint)
|
|
|
293 continue;
|
|
|
294 for (BaseCommand *base2 : sec->sectionCommands)
|
|
|
295 if (auto *cmd = dyn_cast<SymbolAssignment>(base2))
|
|
|
296 declareSymbol(cmd);
|
|
|
297 }
|
|
|
298 }
|
|
|
299
|
|
|
300 // This function is called from assignAddresses, while we are
|
|
|
301 // fixing the output section addresses. This function is supposed
|
|
|
302 // to set the final value for a given symbol assignment.
|
|
|
303 void LinkerScript::assignSymbol(SymbolAssignment *cmd, bool inSec) {
|
|
|
304 if (cmd->name == ".") {
|
|
|
305 setDot(cmd->expression, cmd->location, inSec);
|
|
|
306 return;
|
|
|
307 }
|
|
|
308
|
|
|
309 if (!cmd->sym)
|
|
|
310 return;
|
|
|
311
|
|
|
312 ExprValue v = cmd->expression();
|
|
|
313 if (v.isAbsolute()) {
|
|
|
314 cmd->sym->section = nullptr;
|
|
|
315 cmd->sym->value = v.getValue();
|
|
|
316 } else {
|
|
|
317 cmd->sym->section = v.sec;
|
|
|
318 cmd->sym->value = v.getSectionOffset();
|
|
|
319 }
|
|
|
320 }
|
|
|
321
|
|
|
322 static std::string getFilename(InputFile *file) {
|
|
|
323 if (!file)
|
|
|
324 return "";
|
|
|
325 if (file->archiveName.empty())
|
|
|
326 return std::string(file->getName());
|
|
173
|
327 return (file->archiveName + ':' + file->getName()).str();
|
|
150
|
328 }
|
|
|
329
|
|
|
330 bool LinkerScript::shouldKeep(InputSectionBase *s) {
|
|
|
331 if (keptSections.empty())
|
|
|
332 return false;
|
|
|
333 std::string filename = getFilename(s->file);
|
|
|
334 for (InputSectionDescription *id : keptSections)
|
|
|
335 if (id->filePat.match(filename))
|
|
|
336 for (SectionPattern &p : id->sectionPatterns)
|
|
|
337 if (p.sectionPat.match(s->name) &&
|
|
|
338 (s->flags & id->withFlags) == id->withFlags &&
|
|
|
339 (s->flags & id->withoutFlags) == 0)
|
|
|
340 return true;
|
|
|
341 return false;
|
|
|
342 }
|
|
|
343
|
|
|
344 // A helper function for the SORT() command.
|
|
|
345 static bool matchConstraints(ArrayRef<InputSectionBase *> sections,
|
|
|
346 ConstraintKind kind) {
|
|
|
347 if (kind == ConstraintKind::NoConstraint)
|
|
|
348 return true;
|
|
|
349
|
|
|
350 bool isRW = llvm::any_of(
|
|
|
351 sections, [](InputSectionBase *sec) { return sec->flags & SHF_WRITE; });
|
|
|
352
|
|
|
353 return (isRW && kind == ConstraintKind::ReadWrite) ||
|
|
|
354 (!isRW && kind == ConstraintKind::ReadOnly);
|
|
|
355 }
|
|
|
356
|
|
|
357 static void sortSections(MutableArrayRef<InputSectionBase *> vec,
|
|
|
358 SortSectionPolicy k) {
|
|
|
359 auto alignmentComparator = [](InputSectionBase *a, InputSectionBase *b) {
|
|
|
360 // ">" is not a mistake. Sections with larger alignments are placed
|
|
|
361 // before sections with smaller alignments in order to reduce the
|
|
|
362 // amount of padding necessary. This is compatible with GNU.
|
|
|
363 return a->alignment > b->alignment;
|
|
|
364 };
|
|
|
365 auto nameComparator = [](InputSectionBase *a, InputSectionBase *b) {
|
|
|
366 return a->name < b->name;
|
|
|
367 };
|
|
|
368 auto priorityComparator = [](InputSectionBase *a, InputSectionBase *b) {
|
|
|
369 return getPriority(a->name) < getPriority(b->name);
|
|
|
370 };
|
|
|
371
|
|
|
372 switch (k) {
|
|
|
373 case SortSectionPolicy::Default:
|
|
|
374 case SortSectionPolicy::None:
|
|
|
375 return;
|
|
|
376 case SortSectionPolicy::Alignment:
|
|
|
377 return llvm::stable_sort(vec, alignmentComparator);
|
|
|
378 case SortSectionPolicy::Name:
|
|
|
379 return llvm::stable_sort(vec, nameComparator);
|
|
|
380 case SortSectionPolicy::Priority:
|
|
|
381 return llvm::stable_sort(vec, priorityComparator);
|
|
|
382 }
|
|
|
383 }
|
|
|
384
|
|
|
385 // Sort sections as instructed by SORT-family commands and --sort-section
|
|
|
386 // option. Because SORT-family commands can be nested at most two depth
|
|
|
387 // (e.g. SORT_BY_NAME(SORT_BY_ALIGNMENT(.text.*))) and because the command
|
|
|
388 // line option is respected even if a SORT command is given, the exact
|
|
|
389 // behavior we have here is a bit complicated. Here are the rules.
|
|
|
390 //
|
|
|
391 // 1. If two SORT commands are given, --sort-section is ignored.
|
|
|
392 // 2. If one SORT command is given, and if it is not SORT_NONE,
|
|
|
393 // --sort-section is handled as an inner SORT command.
|
|
|
394 // 3. If one SORT command is given, and if it is SORT_NONE, don't sort.
|
|
|
395 // 4. If no SORT command is given, sort according to --sort-section.
|
|
|
396 static void sortInputSections(MutableArrayRef<InputSectionBase *> vec,
|
|
|
397 const SectionPattern &pat) {
|
|
|
398 if (pat.sortOuter == SortSectionPolicy::None)
|
|
|
399 return;
|
|
|
400
|
|
|
401 if (pat.sortInner == SortSectionPolicy::Default)
|
|
|
402 sortSections(vec, config->sortSection);
|
|
|
403 else
|
|
|
404 sortSections(vec, pat.sortInner);
|
|
|
405 sortSections(vec, pat.sortOuter);
|
|
|
406 }
|
|
|
407
|
|
|
408 // Compute and remember which sections the InputSectionDescription matches.
|
|
|
409 std::vector<InputSectionBase *>
|
|
173
|
410 LinkerScript::computeInputSections(const InputSectionDescription *cmd,
|
|
|
411 ArrayRef<InputSectionBase *> sections) {
|
|
150
|
412 std::vector<InputSectionBase *> ret;
|
|
|
413
|
|
|
414 // Collects all sections that satisfy constraints of Cmd.
|
|
|
415 for (const SectionPattern &pat : cmd->sectionPatterns) {
|
|
|
416 size_t sizeBefore = ret.size();
|
|
|
417
|
|
173
|
418 for (InputSectionBase *sec : sections) {
|
|
150
|
419 if (!sec->isLive() || sec->parent)
|
|
|
420 continue;
|
|
|
421
|
|
|
422 // For -emit-relocs we have to ignore entries like
|
|
|
423 // .rela.dyn : { *(.rela.data) }
|
|
|
424 // which are common because they are in the default bfd script.
|
|
|
425 // We do not ignore SHT_REL[A] linker-synthesized sections here because
|
|
|
426 // want to support scripts that do custom layout for them.
|
|
|
427 if (isa<InputSection>(sec) &&
|
|
|
428 cast<InputSection>(sec)->getRelocatedSection())
|
|
|
429 continue;
|
|
|
430
|
|
|
431 // Check the name early to improve performance in the common case.
|
|
|
432 if (!pat.sectionPat.match(sec->name))
|
|
|
433 continue;
|
|
|
434
|
|
|
435 std::string filename = getFilename(sec->file);
|
|
|
436 if (!cmd->filePat.match(filename) ||
|
|
|
437 pat.excludedFilePat.match(filename) ||
|
|
|
438 (sec->flags & cmd->withFlags) != cmd->withFlags ||
|
|
|
439 (sec->flags & cmd->withoutFlags) != 0)
|
|
|
440 continue;
|
|
|
441
|
|
|
442 ret.push_back(sec);
|
|
|
443 }
|
|
|
444
|
|
|
445 sortInputSections(
|
|
|
446 MutableArrayRef<InputSectionBase *>(ret).slice(sizeBefore), pat);
|
|
|
447 }
|
|
|
448 return ret;
|
|
|
449 }
|
|
|
450
|
|
|
451 void LinkerScript::discard(InputSectionBase *s) {
|
|
|
452 if (s == in.shStrTab || s == mainPart->relrDyn)
|
|
|
453 error("discarding " + s->name + " section is not allowed");
|
|
|
454
|
|
|
455 // You can discard .hash and .gnu.hash sections by linker scripts. Since
|
|
|
456 // they are synthesized sections, we need to handle them differently than
|
|
|
457 // other regular sections.
|
|
|
458 if (s == mainPart->gnuHashTab)
|
|
|
459 mainPart->gnuHashTab = nullptr;
|
|
|
460 if (s == mainPart->hashTab)
|
|
|
461 mainPart->hashTab = nullptr;
|
|
|
462
|
|
|
463 s->markDead();
|
|
|
464 s->parent = nullptr;
|
|
|
465 for (InputSection *ds : s->dependentSections)
|
|
|
466 discard(ds);
|
|
|
467 }
|
|
|
468
|
|
173
|
469 void LinkerScript::discardSynthetic(OutputSection &outCmd) {
|
|
|
470 for (Partition &part : partitions) {
|
|
|
471 if (!part.armExidx || !part.armExidx->isLive())
|
|
|
472 continue;
|
|
|
473 std::vector<InputSectionBase *> secs(part.armExidx->exidxSections.begin(),
|
|
|
474 part.armExidx->exidxSections.end());
|
|
|
475 for (BaseCommand *base : outCmd.sectionCommands)
|
|
|
476 if (auto *cmd = dyn_cast<InputSectionDescription>(base)) {
|
|
|
477 std::vector<InputSectionBase *> matches =
|
|
|
478 computeInputSections(cmd, secs);
|
|
|
479 for (InputSectionBase *s : matches)
|
|
|
480 discard(s);
|
|
|
481 }
|
|
|
482 }
|
|
|
483 }
|
|
|
484
|
|
150
|
485 std::vector<InputSectionBase *>
|
|
|
486 LinkerScript::createInputSectionList(OutputSection &outCmd) {
|
|
|
487 std::vector<InputSectionBase *> ret;
|
|
|
488
|
|
|
489 for (BaseCommand *base : outCmd.sectionCommands) {
|
|
|
490 if (auto *cmd = dyn_cast<InputSectionDescription>(base)) {
|
|
173
|
491 cmd->sectionBases = computeInputSections(cmd, inputSections);
|
|
150
|
492 for (InputSectionBase *s : cmd->sectionBases)
|
|
|
493 s->parent = &outCmd;
|
|
|
494 ret.insert(ret.end(), cmd->sectionBases.begin(), cmd->sectionBases.end());
|
|
|
495 }
|
|
|
496 }
|
|
|
497 return ret;
|
|
|
498 }
|
|
|
499
|
|
|
500 // Create output sections described by SECTIONS commands.
|
|
|
501 void LinkerScript::processSectionCommands() {
|
|
|
502 size_t i = 0;
|
|
|
503 for (BaseCommand *base : sectionCommands) {
|
|
|
504 if (auto *sec = dyn_cast<OutputSection>(base)) {
|
|
|
505 std::vector<InputSectionBase *> v = createInputSectionList(*sec);
|
|
|
506
|
|
|
507 // The output section name `/DISCARD/' is special.
|
|
|
508 // Any input section assigned to it is discarded.
|
|
|
509 if (sec->name == "/DISCARD/") {
|
|
|
510 for (InputSectionBase *s : v)
|
|
|
511 discard(s);
|
|
173
|
512 discardSynthetic(*sec);
|
|
150
|
513 sec->sectionCommands.clear();
|
|
|
514 continue;
|
|
|
515 }
|
|
|
516
|
|
|
517 // This is for ONLY_IF_RO and ONLY_IF_RW. An output section directive
|
|
|
518 // ".foo : ONLY_IF_R[OW] { ... }" is handled only if all member input
|
|
|
519 // sections satisfy a given constraint. If not, a directive is handled
|
|
|
520 // as if it wasn't present from the beginning.
|
|
|
521 //
|
|
|
522 // Because we'll iterate over SectionCommands many more times, the easy
|
|
|
523 // way to "make it as if it wasn't present" is to make it empty.
|
|
|
524 if (!matchConstraints(v, sec->constraint)) {
|
|
|
525 for (InputSectionBase *s : v)
|
|
|
526 s->parent = nullptr;
|
|
|
527 sec->sectionCommands.clear();
|
|
|
528 continue;
|
|
|
529 }
|
|
|
530
|
|
|
531 // Handle subalign (e.g. ".foo : SUBALIGN(32) { ... }"). If subalign
|
|
|
532 // is given, input sections are aligned to that value, whether the
|
|
|
533 // given value is larger or smaller than the original section alignment.
|
|
|
534 if (sec->subalignExpr) {
|
|
|
535 uint32_t subalign = sec->subalignExpr().getValue();
|
|
|
536 for (InputSectionBase *s : v)
|
|
|
537 s->alignment = subalign;
|
|
|
538 }
|
|
|
539
|
|
|
540 // Set the partition field the same way OutputSection::recordSection()
|
|
|
541 // does. Partitions cannot be used with the SECTIONS command, so this is
|
|
|
542 // always 1.
|
|
|
543 sec->partition = 1;
|
|
|
544
|
|
|
545 sec->sectionIndex = i++;
|
|
|
546 }
|
|
|
547 }
|
|
|
548 }
|
|
|
549
|
|
|
550 void LinkerScript::processSymbolAssignments() {
|
|
|
551 // Dot outside an output section still represents a relative address, whose
|
|
|
552 // sh_shndx should not be SHN_UNDEF or SHN_ABS. Create a dummy aether section
|
|
|
553 // that fills the void outside a section. It has an index of one, which is
|
|
|
554 // indistinguishable from any other regular section index.
|
|
|
555 aether = make<OutputSection>("", 0, SHF_ALLOC);
|
|
|
556 aether->sectionIndex = 1;
|
|
|
557
|
|
|
558 // ctx captures the local AddressState and makes it accessible deliberately.
|
|
|
559 // This is needed as there are some cases where we cannot just thread the
|
|
|
560 // current state through to a lambda function created by the script parser.
|
|
|
561 AddressState state;
|
|
|
562 ctx = &state;
|
|
|
563 ctx->outSec = aether;
|
|
|
564
|
|
|
565 for (BaseCommand *base : sectionCommands) {
|
|
|
566 if (auto *cmd = dyn_cast<SymbolAssignment>(base))
|
|
|
567 addSymbol(cmd);
|
|
|
568 else
|
|
|
569 for (BaseCommand *sub_base : cast<OutputSection>(base)->sectionCommands)
|
|
|
570 if (auto *cmd = dyn_cast<SymbolAssignment>(sub_base))
|
|
|
571 addSymbol(cmd);
|
|
|
572 }
|
|
|
573
|
|
|
574 ctx = nullptr;
|
|
|
575 }
|
|
|
576
|
|
|
577 static OutputSection *findByName(ArrayRef<BaseCommand *> vec,
|
|
|
578 StringRef name) {
|
|
|
579 for (BaseCommand *base : vec)
|
|
|
580 if (auto *sec = dyn_cast<OutputSection>(base))
|
|
|
581 if (sec->name == name)
|
|
|
582 return sec;
|
|
|
583 return nullptr;
|
|
|
584 }
|
|
|
585
|
|
|
586 static OutputSection *createSection(InputSectionBase *isec,
|
|
|
587 StringRef outsecName) {
|
|
|
588 OutputSection *sec = script->createOutputSection(outsecName, "<internal>");
|
|
|
589 sec->recordSection(isec);
|
|
|
590 return sec;
|
|
|
591 }
|
|
|
592
|
|
|
593 static OutputSection *
|
|
|
594 addInputSec(StringMap<TinyPtrVector<OutputSection *>> &map,
|
|
|
595 InputSectionBase *isec, StringRef outsecName) {
|
|
|
596 // Sections with SHT_GROUP or SHF_GROUP attributes reach here only when the -r
|
|
|
597 // option is given. A section with SHT_GROUP defines a "section group", and
|
|
|
598 // its members have SHF_GROUP attribute. Usually these flags have already been
|
|
|
599 // stripped by InputFiles.cpp as section groups are processed and uniquified.
|
|
|
600 // However, for the -r option, we want to pass through all section groups
|
|
|
601 // as-is because adding/removing members or merging them with other groups
|
|
|
602 // change their semantics.
|
|
|
603 if (isec->type == SHT_GROUP || (isec->flags & SHF_GROUP))
|
|
|
604 return createSection(isec, outsecName);
|
|
|
605
|
|
|
606 // Imagine .zed : { *(.foo) *(.bar) } script. Both foo and bar may have
|
|
|
607 // relocation sections .rela.foo and .rela.bar for example. Most tools do
|
|
|
608 // not allow multiple REL[A] sections for output section. Hence we
|
|
|
609 // should combine these relocation sections into single output.
|
|
|
610 // We skip synthetic sections because it can be .rela.dyn/.rela.plt or any
|
|
|
611 // other REL[A] sections created by linker itself.
|
|
|
612 if (!isa<SyntheticSection>(isec) &&
|
|
|
613 (isec->type == SHT_REL || isec->type == SHT_RELA)) {
|
|
|
614 auto *sec = cast<InputSection>(isec);
|
|
|
615 OutputSection *out = sec->getRelocatedSection()->getOutputSection();
|
|
|
616
|
|
|
617 if (out->relocationSection) {
|
|
|
618 out->relocationSection->recordSection(sec);
|
|
|
619 return nullptr;
|
|
|
620 }
|
|
|
621
|
|
|
622 out->relocationSection = createSection(isec, outsecName);
|
|
|
623 return out->relocationSection;
|
|
|
624 }
|
|
|
625
|
|
|
626 // The ELF spec just says
|
|
|
627 // ----------------------------------------------------------------
|
|
|
628 // In the first phase, input sections that match in name, type and
|
|
|
629 // attribute flags should be concatenated into single sections.
|
|
|
630 // ----------------------------------------------------------------
|
|
|
631 //
|
|
|
632 // However, it is clear that at least some flags have to be ignored for
|
|
|
633 // section merging. At the very least SHF_GROUP and SHF_COMPRESSED have to be
|
|
|
634 // ignored. We should not have two output .text sections just because one was
|
|
|
635 // in a group and another was not for example.
|
|
|
636 //
|
|
|
637 // It also seems that wording was a late addition and didn't get the
|
|
|
638 // necessary scrutiny.
|
|
|
639 //
|
|
|
640 // Merging sections with different flags is expected by some users. One
|
|
|
641 // reason is that if one file has
|
|
|
642 //
|
|
|
643 // int *const bar __attribute__((section(".foo"))) = (int *)0;
|
|
|
644 //
|
|
|
645 // gcc with -fPIC will produce a read only .foo section. But if another
|
|
|
646 // file has
|
|
|
647 //
|
|
|
648 // int zed;
|
|
|
649 // int *const bar __attribute__((section(".foo"))) = (int *)&zed;
|
|
|
650 //
|
|
|
651 // gcc with -fPIC will produce a read write section.
|
|
|
652 //
|
|
|
653 // Last but not least, when using linker script the merge rules are forced by
|
|
|
654 // the script. Unfortunately, linker scripts are name based. This means that
|
|
|
655 // expressions like *(.foo*) can refer to multiple input sections with
|
|
|
656 // different flags. We cannot put them in different output sections or we
|
|
|
657 // would produce wrong results for
|
|
|
658 //
|
|
|
659 // start = .; *(.foo.*) end = .; *(.bar)
|
|
|
660 //
|
|
|
661 // and a mapping of .foo1 and .bar1 to one section and .foo2 and .bar2 to
|
|
|
662 // another. The problem is that there is no way to layout those output
|
|
|
663 // sections such that the .foo sections are the only thing between the start
|
|
|
664 // and end symbols.
|
|
|
665 //
|
|
|
666 // Given the above issues, we instead merge sections by name and error on
|
|
|
667 // incompatible types and flags.
|
|
|
668 TinyPtrVector<OutputSection *> &v = map[outsecName];
|
|
|
669 for (OutputSection *sec : v) {
|
|
|
670 if (sec->partition != isec->partition)
|
|
|
671 continue;
|
|
|
672
|
|
|
673 if (config->relocatable && (isec->flags & SHF_LINK_ORDER)) {
|
|
|
674 // Merging two SHF_LINK_ORDER sections with different sh_link fields will
|
|
|
675 // change their semantics, so we only merge them in -r links if they will
|
|
|
676 // end up being linked to the same output section. The casts are fine
|
|
|
677 // because everything in the map was created by the orphan placement code.
|
|
|
678 auto *firstIsec = cast<InputSectionBase>(
|
|
|
679 cast<InputSectionDescription>(sec->sectionCommands[0])
|
|
|
680 ->sectionBases[0]);
|
|
|
681 if (firstIsec->getLinkOrderDep()->getOutputSection() !=
|
|
|
682 isec->getLinkOrderDep()->getOutputSection())
|
|
|
683 continue;
|
|
|
684 }
|
|
|
685
|
|
|
686 sec->recordSection(isec);
|
|
|
687 return nullptr;
|
|
|
688 }
|
|
|
689
|
|
|
690 OutputSection *sec = createSection(isec, outsecName);
|
|
|
691 v.push_back(sec);
|
|
|
692 return sec;
|
|
|
693 }
|
|
|
694
|
|
|
695 // Add sections that didn't match any sections command.
|
|
|
696 void LinkerScript::addOrphanSections() {
|
|
|
697 StringMap<TinyPtrVector<OutputSection *>> map;
|
|
|
698 std::vector<OutputSection *> v;
|
|
|
699
|
|
|
700 std::function<void(InputSectionBase *)> add;
|
|
|
701 add = [&](InputSectionBase *s) {
|
|
|
702 if (s->isLive() && !s->parent) {
|
|
173
|
703 orphanSections.push_back(s);
|
|
150
|
704
|
|
173
|
705 StringRef name = getOutputSectionName(s);
|
|
|
706 if (config->unique) {
|
|
|
707 v.push_back(createSection(s, name));
|
|
|
708 } else if (OutputSection *sec = findByName(sectionCommands, name)) {
|
|
150
|
709 sec->recordSection(s);
|
|
|
710 } else {
|
|
|
711 if (OutputSection *os = addInputSec(map, s, name))
|
|
|
712 v.push_back(os);
|
|
|
713 assert(isa<MergeInputSection>(s) ||
|
|
|
714 s->getOutputSection()->sectionIndex == UINT32_MAX);
|
|
|
715 }
|
|
|
716 }
|
|
|
717
|
|
|
718 if (config->relocatable)
|
|
|
719 for (InputSectionBase *depSec : s->dependentSections)
|
|
|
720 if (depSec->flags & SHF_LINK_ORDER)
|
|
|
721 add(depSec);
|
|
|
722 };
|
|
|
723
|
|
|
724 // For futher --emit-reloc handling code we need target output section
|
|
|
725 // to be created before we create relocation output section, so we want
|
|
|
726 // to create target sections first. We do not want priority handling
|
|
|
727 // for synthetic sections because them are special.
|
|
|
728 for (InputSectionBase *isec : inputSections) {
|
|
|
729 // In -r links, SHF_LINK_ORDER sections are added while adding their parent
|
|
|
730 // sections because we need to know the parent's output section before we
|
|
|
731 // can select an output section for the SHF_LINK_ORDER section.
|
|
|
732 if (config->relocatable && (isec->flags & SHF_LINK_ORDER))
|
|
|
733 continue;
|
|
|
734
|
|
|
735 if (auto *sec = dyn_cast<InputSection>(isec))
|
|
|
736 if (InputSectionBase *rel = sec->getRelocatedSection())
|
|
|
737 if (auto *relIS = dyn_cast_or_null<InputSectionBase>(rel->parent))
|
|
|
738 add(relIS);
|
|
|
739 add(isec);
|
|
|
740 }
|
|
|
741
|
|
|
742 // If no SECTIONS command was given, we should insert sections commands
|
|
|
743 // before others, so that we can handle scripts which refers them,
|
|
|
744 // for example: "foo = ABSOLUTE(ADDR(.text)));".
|
|
|
745 // When SECTIONS command is present we just add all orphans to the end.
|
|
|
746 if (hasSectionsCommand)
|
|
|
747 sectionCommands.insert(sectionCommands.end(), v.begin(), v.end());
|
|
|
748 else
|
|
|
749 sectionCommands.insert(sectionCommands.begin(), v.begin(), v.end());
|
|
|
750 }
|
|
|
751
|
|
173
|
752 void LinkerScript::diagnoseOrphanHandling() const {
|
|
|
753 for (const InputSectionBase *sec : orphanSections) {
|
|
|
754 // Input SHT_REL[A] retained by --emit-relocs are ignored by
|
|
|
755 // computeInputSections(). Don't warn/error.
|
|
|
756 if (isa<InputSection>(sec) &&
|
|
|
757 cast<InputSection>(sec)->getRelocatedSection())
|
|
|
758 continue;
|
|
|
759
|
|
|
760 StringRef name = getOutputSectionName(sec);
|
|
|
761 if (config->orphanHandling == OrphanHandlingPolicy::Error)
|
|
|
762 error(toString(sec) + " is being placed in '" + name + "'");
|
|
|
763 else if (config->orphanHandling == OrphanHandlingPolicy::Warn)
|
|
|
764 warn(toString(sec) + " is being placed in '" + name + "'");
|
|
|
765 }
|
|
|
766 }
|
|
|
767
|
|
150
|
768 uint64_t LinkerScript::advance(uint64_t size, unsigned alignment) {
|
|
|
769 bool isTbss =
|
|
|
770 (ctx->outSec->flags & SHF_TLS) && ctx->outSec->type == SHT_NOBITS;
|
|
|
771 uint64_t start = isTbss ? dot + ctx->threadBssOffset : dot;
|
|
|
772 start = alignTo(start, alignment);
|
|
|
773 uint64_t end = start + size;
|
|
|
774
|
|
|
775 if (isTbss)
|
|
|
776 ctx->threadBssOffset = end - dot;
|
|
|
777 else
|
|
|
778 dot = end;
|
|
|
779 return end;
|
|
|
780 }
|
|
|
781
|
|
|
782 void LinkerScript::output(InputSection *s) {
|
|
|
783 assert(ctx->outSec == s->getParent());
|
|
|
784 uint64_t before = advance(0, 1);
|
|
|
785 uint64_t pos = advance(s->getSize(), s->alignment);
|
|
|
786 s->outSecOff = pos - s->getSize() - ctx->outSec->addr;
|
|
|
787
|
|
|
788 // Update output section size after adding each section. This is so that
|
|
|
789 // SIZEOF works correctly in the case below:
|
|
|
790 // .foo { *(.aaa) a = SIZEOF(.foo); *(.bbb) }
|
|
|
791 expandOutputSection(pos - before);
|
|
|
792 }
|
|
|
793
|
|
|
794 void LinkerScript::switchTo(OutputSection *sec) {
|
|
|
795 ctx->outSec = sec;
|
|
|
796
|
|
173
|
797 uint64_t pos = advance(0, 1);
|
|
|
798 if (sec->addrExpr && script->hasSectionsCommand) {
|
|
|
799 // The alignment is ignored.
|
|
|
800 ctx->outSec->addr = pos;
|
|
|
801 } else {
|
|
|
802 // ctx->outSec->alignment is the max of ALIGN and the maximum of input
|
|
|
803 // section alignments.
|
|
|
804 ctx->outSec->addr = advance(0, ctx->outSec->alignment);
|
|
|
805 expandMemoryRegions(ctx->outSec->addr - pos);
|
|
|
806 }
|
|
150
|
807 }
|
|
|
808
|
|
|
809 // This function searches for a memory region to place the given output
|
|
|
810 // section in. If found, a pointer to the appropriate memory region is
|
|
|
811 // returned. Otherwise, a nullptr is returned.
|
|
|
812 MemoryRegion *LinkerScript::findMemoryRegion(OutputSection *sec) {
|
|
|
813 // If a memory region name was specified in the output section command,
|
|
|
814 // then try to find that region first.
|
|
|
815 if (!sec->memoryRegionName.empty()) {
|
|
|
816 if (MemoryRegion *m = memoryRegions.lookup(sec->memoryRegionName))
|
|
|
817 return m;
|
|
|
818 error("memory region '" + sec->memoryRegionName + "' not declared");
|
|
|
819 return nullptr;
|
|
|
820 }
|
|
|
821
|
|
|
822 // If at least one memory region is defined, all sections must
|
|
|
823 // belong to some memory region. Otherwise, we don't need to do
|
|
|
824 // anything for memory regions.
|
|
|
825 if (memoryRegions.empty())
|
|
|
826 return nullptr;
|
|
|
827
|
|
|
828 // See if a region can be found by matching section flags.
|
|
|
829 for (auto &pair : memoryRegions) {
|
|
|
830 MemoryRegion *m = pair.second;
|
|
|
831 if ((m->flags & sec->flags) && (m->negFlags & sec->flags) == 0)
|
|
|
832 return m;
|
|
|
833 }
|
|
|
834
|
|
|
835 // Otherwise, no suitable region was found.
|
|
|
836 if (sec->flags & SHF_ALLOC)
|
|
|
837 error("no memory region specified for section '" + sec->name + "'");
|
|
|
838 return nullptr;
|
|
|
839 }
|
|
|
840
|
|
|
841 static OutputSection *findFirstSection(PhdrEntry *load) {
|
|
|
842 for (OutputSection *sec : outputSections)
|
|
|
843 if (sec->ptLoad == load)
|
|
|
844 return sec;
|
|
|
845 return nullptr;
|
|
|
846 }
|
|
|
847
|
|
|
848 // This function assigns offsets to input sections and an output section
|
|
|
849 // for a single sections command (e.g. ".text { *(.text); }").
|
|
|
850 void LinkerScript::assignOffsets(OutputSection *sec) {
|
|
|
851 if (!(sec->flags & SHF_ALLOC))
|
|
|
852 dot = 0;
|
|
|
853
|
|
173
|
854 bool prevLMARegionIsDefault = ctx->lmaRegion == nullptr;
|
|
150
|
855 ctx->memRegion = sec->memRegion;
|
|
|
856 ctx->lmaRegion = sec->lmaRegion;
|
|
|
857 if (ctx->memRegion)
|
|
|
858 dot = ctx->memRegion->curPos;
|
|
|
859
|
|
|
860 if ((sec->flags & SHF_ALLOC) && sec->addrExpr)
|
|
|
861 setDot(sec->addrExpr, sec->location, false);
|
|
|
862
|
|
|
863 // If the address of the section has been moved forward by an explicit
|
|
|
864 // expression so that it now starts past the current curPos of the enclosing
|
|
|
865 // region, we need to expand the current region to account for the space
|
|
|
866 // between the previous section, if any, and the start of this section.
|
|
|
867 if (ctx->memRegion && ctx->memRegion->curPos < dot)
|
|
|
868 expandMemoryRegion(ctx->memRegion, dot - ctx->memRegion->curPos,
|
|
|
869 ctx->memRegion->name, sec->name);
|
|
|
870
|
|
|
871 switchTo(sec);
|
|
|
872
|
|
173
|
873 // ctx->lmaOffset is LMA minus VMA. If LMA is explicitly specified via AT() or
|
|
|
874 // AT>, recompute ctx->lmaOffset; otherwise, if both previous/current LMA
|
|
|
875 // region is the default, reuse previous lmaOffset; otherwise, reset lmaOffset
|
|
|
876 // to 0. This emulates heuristics described in
|
|
|
877 // https://sourceware.org/binutils/docs/ld/Output-Section-LMA.html
|
|
150
|
878 if (sec->lmaExpr)
|
|
|
879 ctx->lmaOffset = sec->lmaExpr().getValue() - dot;
|
|
173
|
880 else if (MemoryRegion *mr = sec->lmaRegion)
|
|
150
|
881 ctx->lmaOffset = alignTo(mr->curPos, sec->alignment) - dot;
|
|
173
|
882 else if (!prevLMARegionIsDefault)
|
|
|
883 ctx->lmaOffset = 0;
|
|
150
|
884
|
|
173
|
885 // Propagate ctx->lmaOffset to the first "non-header" section.
|
|
150
|
886 if (PhdrEntry *l = ctx->outSec->ptLoad)
|
|
|
887 if (sec == findFirstSection(l))
|
|
|
888 l->lmaOffset = ctx->lmaOffset;
|
|
|
889
|
|
|
890 // We can call this method multiple times during the creation of
|
|
|
891 // thunks and want to start over calculation each time.
|
|
|
892 sec->size = 0;
|
|
|
893
|
|
|
894 // We visited SectionsCommands from processSectionCommands to
|
|
|
895 // layout sections. Now, we visit SectionsCommands again to fix
|
|
|
896 // section offsets.
|
|
|
897 for (BaseCommand *base : sec->sectionCommands) {
|
|
|
898 // This handles the assignments to symbol or to the dot.
|
|
|
899 if (auto *cmd = dyn_cast<SymbolAssignment>(base)) {
|
|
|
900 cmd->addr = dot;
|
|
|
901 assignSymbol(cmd, true);
|
|
|
902 cmd->size = dot - cmd->addr;
|
|
|
903 continue;
|
|
|
904 }
|
|
|
905
|
|
|
906 // Handle BYTE(), SHORT(), LONG(), or QUAD().
|
|
|
907 if (auto *cmd = dyn_cast<ByteCommand>(base)) {
|
|
|
908 cmd->offset = dot - ctx->outSec->addr;
|
|
|
909 dot += cmd->size;
|
|
|
910 expandOutputSection(cmd->size);
|
|
|
911 continue;
|
|
|
912 }
|
|
|
913
|
|
|
914 // Handle a single input section description command.
|
|
|
915 // It calculates and assigns the offsets for each section and also
|
|
|
916 // updates the output section size.
|
|
|
917 for (InputSection *sec : cast<InputSectionDescription>(base)->sections)
|
|
|
918 output(sec);
|
|
|
919 }
|
|
|
920 }
|
|
|
921
|
|
|
922 static bool isDiscardable(OutputSection &sec) {
|
|
|
923 if (sec.name == "/DISCARD/")
|
|
|
924 return true;
|
|
|
925
|
|
|
926 // We do not remove empty sections that are explicitly
|
|
|
927 // assigned to any segment.
|
|
|
928 if (!sec.phdrs.empty())
|
|
|
929 return false;
|
|
|
930
|
|
|
931 // We do not want to remove OutputSections with expressions that reference
|
|
|
932 // symbols even if the OutputSection is empty. We want to ensure that the
|
|
|
933 // expressions can be evaluated and report an error if they cannot.
|
|
|
934 if (sec.expressionsUseSymbols)
|
|
|
935 return false;
|
|
|
936
|
|
|
937 // OutputSections may be referenced by name in ADDR and LOADADDR expressions,
|
|
|
938 // as an empty Section can has a valid VMA and LMA we keep the OutputSection
|
|
|
939 // to maintain the integrity of the other Expression.
|
|
|
940 if (sec.usedInExpression)
|
|
|
941 return false;
|
|
|
942
|
|
|
943 for (BaseCommand *base : sec.sectionCommands) {
|
|
|
944 if (auto cmd = dyn_cast<SymbolAssignment>(base))
|
|
|
945 // Don't create empty output sections just for unreferenced PROVIDE
|
|
|
946 // symbols.
|
|
|
947 if (cmd->name != "." && !cmd->sym)
|
|
|
948 continue;
|
|
|
949
|
|
|
950 if (!isa<InputSectionDescription>(*base))
|
|
|
951 return false;
|
|
|
952 }
|
|
|
953 return true;
|
|
|
954 }
|
|
|
955
|
|
|
956 void LinkerScript::adjustSectionsBeforeSorting() {
|
|
|
957 // If the output section contains only symbol assignments, create a
|
|
|
958 // corresponding output section. The issue is what to do with linker script
|
|
|
959 // like ".foo : { symbol = 42; }". One option would be to convert it to
|
|
|
960 // "symbol = 42;". That is, move the symbol out of the empty section
|
|
|
961 // description. That seems to be what bfd does for this simple case. The
|
|
|
962 // problem is that this is not completely general. bfd will give up and
|
|
|
963 // create a dummy section too if there is a ". = . + 1" inside the section
|
|
|
964 // for example.
|
|
|
965 // Given that we want to create the section, we have to worry what impact
|
|
|
966 // it will have on the link. For example, if we just create a section with
|
|
|
967 // 0 for flags, it would change which PT_LOADs are created.
|
|
|
968 // We could remember that particular section is dummy and ignore it in
|
|
|
969 // other parts of the linker, but unfortunately there are quite a few places
|
|
|
970 // that would need to change:
|
|
|
971 // * The program header creation.
|
|
|
972 // * The orphan section placement.
|
|
|
973 // * The address assignment.
|
|
|
974 // The other option is to pick flags that minimize the impact the section
|
|
|
975 // will have on the rest of the linker. That is why we copy the flags from
|
|
|
976 // the previous sections. Only a few flags are needed to keep the impact low.
|
|
|
977 uint64_t flags = SHF_ALLOC;
|
|
|
978
|
|
|
979 for (BaseCommand *&cmd : sectionCommands) {
|
|
|
980 auto *sec = dyn_cast<OutputSection>(cmd);
|
|
|
981 if (!sec)
|
|
|
982 continue;
|
|
|
983
|
|
|
984 // Handle align (e.g. ".foo : ALIGN(16) { ... }").
|
|
|
985 if (sec->alignExpr)
|
|
|
986 sec->alignment =
|
|
|
987 std::max<uint32_t>(sec->alignment, sec->alignExpr().getValue());
|
|
|
988
|
|
|
989 // The input section might have been removed (if it was an empty synthetic
|
|
|
990 // section), but we at least know the flags.
|
|
|
991 if (sec->hasInputSections)
|
|
|
992 flags = sec->flags;
|
|
|
993
|
|
|
994 // We do not want to keep any special flags for output section
|
|
|
995 // in case it is empty.
|
|
|
996 bool isEmpty = (getFirstInputSection(sec) == nullptr);
|
|
|
997 if (isEmpty)
|
|
|
998 sec->flags = flags & ((sec->nonAlloc ? 0 : (uint64_t)SHF_ALLOC) |
|
|
|
999 SHF_WRITE | SHF_EXECINSTR);
|
|
|
1000
|
|
|
1001 if (isEmpty && isDiscardable(*sec)) {
|
|
|
1002 sec->markDead();
|
|
|
1003 cmd = nullptr;
|
|
|
1004 }
|
|
|
1005 }
|
|
|
1006
|
|
|
1007 // It is common practice to use very generic linker scripts. So for any
|
|
|
1008 // given run some of the output sections in the script will be empty.
|
|
|
1009 // We could create corresponding empty output sections, but that would
|
|
|
1010 // clutter the output.
|
|
|
1011 // We instead remove trivially empty sections. The bfd linker seems even
|
|
|
1012 // more aggressive at removing them.
|
|
|
1013 llvm::erase_if(sectionCommands, [&](BaseCommand *base) { return !base; });
|
|
|
1014 }
|
|
|
1015
|
|
|
1016 void LinkerScript::adjustSectionsAfterSorting() {
|
|
|
1017 // Try and find an appropriate memory region to assign offsets in.
|
|
|
1018 for (BaseCommand *base : sectionCommands) {
|
|
|
1019 if (auto *sec = dyn_cast<OutputSection>(base)) {
|
|
|
1020 if (!sec->lmaRegionName.empty()) {
|
|
|
1021 if (MemoryRegion *m = memoryRegions.lookup(sec->lmaRegionName))
|
|
|
1022 sec->lmaRegion = m;
|
|
|
1023 else
|
|
|
1024 error("memory region '" + sec->lmaRegionName + "' not declared");
|
|
|
1025 }
|
|
|
1026 sec->memRegion = findMemoryRegion(sec);
|
|
|
1027 }
|
|
|
1028 }
|
|
|
1029
|
|
|
1030 // If output section command doesn't specify any segments,
|
|
|
1031 // and we haven't previously assigned any section to segment,
|
|
|
1032 // then we simply assign section to the very first load segment.
|
|
|
1033 // Below is an example of such linker script:
|
|
|
1034 // PHDRS { seg PT_LOAD; }
|
|
|
1035 // SECTIONS { .aaa : { *(.aaa) } }
|
|
|
1036 std::vector<StringRef> defPhdrs;
|
|
|
1037 auto firstPtLoad = llvm::find_if(phdrsCommands, [](const PhdrsCommand &cmd) {
|
|
|
1038 return cmd.type == PT_LOAD;
|
|
|
1039 });
|
|
|
1040 if (firstPtLoad != phdrsCommands.end())
|
|
|
1041 defPhdrs.push_back(firstPtLoad->name);
|
|
|
1042
|
|
|
1043 // Walk the commands and propagate the program headers to commands that don't
|
|
|
1044 // explicitly specify them.
|
|
|
1045 for (BaseCommand *base : sectionCommands) {
|
|
|
1046 auto *sec = dyn_cast<OutputSection>(base);
|
|
|
1047 if (!sec)
|
|
|
1048 continue;
|
|
|
1049
|
|
|
1050 if (sec->phdrs.empty()) {
|
|
|
1051 // To match the bfd linker script behaviour, only propagate program
|
|
|
1052 // headers to sections that are allocated.
|
|
|
1053 if (sec->flags & SHF_ALLOC)
|
|
|
1054 sec->phdrs = defPhdrs;
|
|
|
1055 } else {
|
|
|
1056 defPhdrs = sec->phdrs;
|
|
|
1057 }
|
|
|
1058 }
|
|
|
1059 }
|
|
|
1060
|
|
|
1061 static uint64_t computeBase(uint64_t min, bool allocateHeaders) {
|
|
|
1062 // If there is no SECTIONS or if the linkerscript is explicit about program
|
|
|
1063 // headers, do our best to allocate them.
|
|
|
1064 if (!script->hasSectionsCommand || allocateHeaders)
|
|
|
1065 return 0;
|
|
|
1066 // Otherwise only allocate program headers if that would not add a page.
|
|
|
1067 return alignDown(min, config->maxPageSize);
|
|
|
1068 }
|
|
|
1069
|
|
|
1070 // When the SECTIONS command is used, try to find an address for the file and
|
|
|
1071 // program headers output sections, which can be added to the first PT_LOAD
|
|
|
1072 // segment when program headers are created.
|
|
|
1073 //
|
|
|
1074 // We check if the headers fit below the first allocated section. If there isn't
|
|
|
1075 // enough space for these sections, we'll remove them from the PT_LOAD segment,
|
|
|
1076 // and we'll also remove the PT_PHDR segment.
|
|
|
1077 void LinkerScript::allocateHeaders(std::vector<PhdrEntry *> &phdrs) {
|
|
|
1078 uint64_t min = std::numeric_limits<uint64_t>::max();
|
|
|
1079 for (OutputSection *sec : outputSections)
|
|
|
1080 if (sec->flags & SHF_ALLOC)
|
|
|
1081 min = std::min<uint64_t>(min, sec->addr);
|
|
|
1082
|
|
|
1083 auto it = llvm::find_if(
|
|
|
1084 phdrs, [](const PhdrEntry *e) { return e->p_type == PT_LOAD; });
|
|
|
1085 if (it == phdrs.end())
|
|
|
1086 return;
|
|
|
1087 PhdrEntry *firstPTLoad = *it;
|
|
|
1088
|
|
|
1089 bool hasExplicitHeaders =
|
|
|
1090 llvm::any_of(phdrsCommands, [](const PhdrsCommand &cmd) {
|
|
|
1091 return cmd.hasPhdrs || cmd.hasFilehdr;
|
|
|
1092 });
|
|
|
1093 bool paged = !config->omagic && !config->nmagic;
|
|
|
1094 uint64_t headerSize = getHeaderSize();
|
|
|
1095 if ((paged || hasExplicitHeaders) &&
|
|
|
1096 headerSize <= min - computeBase(min, hasExplicitHeaders)) {
|
|
|
1097 min = alignDown(min - headerSize, config->maxPageSize);
|
|
|
1098 Out::elfHeader->addr = min;
|
|
|
1099 Out::programHeaders->addr = min + Out::elfHeader->size;
|
|
|
1100 return;
|
|
|
1101 }
|
|
|
1102
|
|
|
1103 // Error if we were explicitly asked to allocate headers.
|
|
|
1104 if (hasExplicitHeaders)
|
|
|
1105 error("could not allocate headers");
|
|
|
1106
|
|
|
1107 Out::elfHeader->ptLoad = nullptr;
|
|
|
1108 Out::programHeaders->ptLoad = nullptr;
|
|
|
1109 firstPTLoad->firstSec = findFirstSection(firstPTLoad);
|
|
|
1110
|
|
|
1111 llvm::erase_if(phdrs,
|
|
|
1112 [](const PhdrEntry *e) { return e->p_type == PT_PHDR; });
|
|
|
1113 }
|
|
|
1114
|
|
|
1115 LinkerScript::AddressState::AddressState() {
|
|
|
1116 for (auto &mri : script->memoryRegions) {
|
|
|
1117 MemoryRegion *mr = mri.second;
|
|
173
|
1118 mr->curPos = (mr->origin)().getValue();
|
|
150
|
1119 }
|
|
|
1120 }
|
|
|
1121
|
|
|
1122 // Here we assign addresses as instructed by linker script SECTIONS
|
|
|
1123 // sub-commands. Doing that allows us to use final VA values, so here
|
|
|
1124 // we also handle rest commands like symbol assignments and ASSERTs.
|
|
|
1125 // Returns a symbol that has changed its section or value, or nullptr if no
|
|
|
1126 // symbol has changed.
|
|
|
1127 const Defined *LinkerScript::assignAddresses() {
|
|
|
1128 if (script->hasSectionsCommand) {
|
|
|
1129 // With a linker script, assignment of addresses to headers is covered by
|
|
|
1130 // allocateHeaders().
|
|
|
1131 dot = config->imageBase.getValueOr(0);
|
|
|
1132 } else {
|
|
|
1133 // Assign addresses to headers right now.
|
|
|
1134 dot = target->getImageBase();
|
|
|
1135 Out::elfHeader->addr = dot;
|
|
|
1136 Out::programHeaders->addr = dot + Out::elfHeader->size;
|
|
|
1137 dot += getHeaderSize();
|
|
|
1138 }
|
|
|
1139
|
|
|
1140 auto deleter = std::make_unique<AddressState>();
|
|
|
1141 ctx = deleter.get();
|
|
|
1142 errorOnMissingSection = true;
|
|
|
1143 switchTo(aether);
|
|
|
1144
|
|
|
1145 SymbolAssignmentMap oldValues = getSymbolAssignmentValues(sectionCommands);
|
|
|
1146 for (BaseCommand *base : sectionCommands) {
|
|
|
1147 if (auto *cmd = dyn_cast<SymbolAssignment>(base)) {
|
|
|
1148 cmd->addr = dot;
|
|
|
1149 assignSymbol(cmd, false);
|
|
|
1150 cmd->size = dot - cmd->addr;
|
|
|
1151 continue;
|
|
|
1152 }
|
|
|
1153 assignOffsets(cast<OutputSection>(base));
|
|
|
1154 }
|
|
|
1155
|
|
|
1156 ctx = nullptr;
|
|
|
1157 return getChangedSymbolAssignment(oldValues);
|
|
|
1158 }
|
|
|
1159
|
|
|
1160 // Creates program headers as instructed by PHDRS linker script command.
|
|
|
1161 std::vector<PhdrEntry *> LinkerScript::createPhdrs() {
|
|
|
1162 std::vector<PhdrEntry *> ret;
|
|
|
1163
|
|
|
1164 // Process PHDRS and FILEHDR keywords because they are not
|
|
|
1165 // real output sections and cannot be added in the following loop.
|
|
|
1166 for (const PhdrsCommand &cmd : phdrsCommands) {
|
|
|
1167 PhdrEntry *phdr = make<PhdrEntry>(cmd.type, cmd.flags ? *cmd.flags : PF_R);
|
|
|
1168
|
|
|
1169 if (cmd.hasFilehdr)
|
|
|
1170 phdr->add(Out::elfHeader);
|
|
|
1171 if (cmd.hasPhdrs)
|
|
|
1172 phdr->add(Out::programHeaders);
|
|
|
1173
|
|
|
1174 if (cmd.lmaExpr) {
|
|
|
1175 phdr->p_paddr = cmd.lmaExpr().getValue();
|
|
|
1176 phdr->hasLMA = true;
|
|
|
1177 }
|
|
|
1178 ret.push_back(phdr);
|
|
|
1179 }
|
|
|
1180
|
|
|
1181 // Add output sections to program headers.
|
|
|
1182 for (OutputSection *sec : outputSections) {
|
|
|
1183 // Assign headers specified by linker script
|
|
|
1184 for (size_t id : getPhdrIndices(sec)) {
|
|
|
1185 ret[id]->add(sec);
|
|
|
1186 if (!phdrsCommands[id].flags.hasValue())
|
|
|
1187 ret[id]->p_flags |= sec->getPhdrFlags();
|
|
|
1188 }
|
|
|
1189 }
|
|
|
1190 return ret;
|
|
|
1191 }
|
|
|
1192
|
|
|
1193 // Returns true if we should emit an .interp section.
|
|
|
1194 //
|
|
|
1195 // We usually do. But if PHDRS commands are given, and
|
|
|
1196 // no PT_INTERP is there, there's no place to emit an
|
|
|
1197 // .interp, so we don't do that in that case.
|
|
|
1198 bool LinkerScript::needsInterpSection() {
|
|
|
1199 if (phdrsCommands.empty())
|
|
|
1200 return true;
|
|
|
1201 for (PhdrsCommand &cmd : phdrsCommands)
|
|
|
1202 if (cmd.type == PT_INTERP)
|
|
|
1203 return true;
|
|
|
1204 return false;
|
|
|
1205 }
|
|
|
1206
|
|
|
1207 ExprValue LinkerScript::getSymbolValue(StringRef name, const Twine &loc) {
|
|
|
1208 if (name == ".") {
|
|
|
1209 if (ctx)
|
|
|
1210 return {ctx->outSec, false, dot - ctx->outSec->addr, loc};
|
|
|
1211 error(loc + ": unable to get location counter value");
|
|
|
1212 return 0;
|
|
|
1213 }
|
|
|
1214
|
|
|
1215 if (Symbol *sym = symtab->find(name)) {
|
|
|
1216 if (auto *ds = dyn_cast<Defined>(sym))
|
|
|
1217 return {ds->section, false, ds->value, loc};
|
|
|
1218 if (isa<SharedSymbol>(sym))
|
|
|
1219 if (!errorOnMissingSection)
|
|
|
1220 return {nullptr, false, 0, loc};
|
|
|
1221 }
|
|
|
1222
|
|
|
1223 error(loc + ": symbol not found: " + name);
|
|
|
1224 return 0;
|
|
|
1225 }
|
|
|
1226
|
|
|
1227 // Returns the index of the segment named Name.
|
|
|
1228 static Optional<size_t> getPhdrIndex(ArrayRef<PhdrsCommand> vec,
|
|
|
1229 StringRef name) {
|
|
|
1230 for (size_t i = 0; i < vec.size(); ++i)
|
|
|
1231 if (vec[i].name == name)
|
|
|
1232 return i;
|
|
|
1233 return None;
|
|
|
1234 }
|
|
|
1235
|
|
|
1236 // Returns indices of ELF headers containing specific section. Each index is a
|
|
|
1237 // zero based number of ELF header listed within PHDRS {} script block.
|
|
|
1238 std::vector<size_t> LinkerScript::getPhdrIndices(OutputSection *cmd) {
|
|
|
1239 std::vector<size_t> ret;
|
|
|
1240
|
|
|
1241 for (StringRef s : cmd->phdrs) {
|
|
|
1242 if (Optional<size_t> idx = getPhdrIndex(phdrsCommands, s))
|
|
|
1243 ret.push_back(*idx);
|
|
|
1244 else if (s != "NONE")
|
|
173
|
1245 error(cmd->location + ": program header '" + s +
|
|
150
|
1246 "' is not listed in PHDRS");
|
|
|
1247 }
|
|
|
1248 return ret;
|
|
|
1249 }
|
