1//===- InputSection.h -------------------------------------------*- C++ -*-===//
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#ifndef LLD_ELF_INPUT_SECTION_H
10#define LLD_ELF_INPUT_SECTION_H
11
12#include "Config.h"
13#include "Relocations.h"
14#include "lld/Common/CommonLinkerContext.h"
15#include "lld/Common/LLVM.h"
16#include "lld/Common/Memory.h"
17#include "llvm/ADT/CachedHashString.h"
18#include "llvm/ADT/DenseSet.h"
19#include "llvm/ADT/StringExtras.h"
20#include "llvm/ADT/TinyPtrVector.h"
21#include "llvm/Object/ELF.h"
22#include "llvm/Support/Compiler.h"
23
24namespace lld {
25namespace elf {
26
27class InputFile;
28class Symbol;
29
30class Defined;
31struct Partition;
32class SyntheticSection;
33template <class ELFT> class ObjFile;
34class OutputSection;
35
36// Returned by InputSectionBase::relsOrRelas. At least two members are empty.
37template <class ELFT> struct RelsOrRelas {
38 Relocs<typename ELFT::Rel> rels;
39 Relocs<typename ELFT::Rela> relas;
40 Relocs<typename ELFT::Crel> crels;
41 bool areRelocsRel() const { return rels.size(); }
42 bool areRelocsCrel() const { return crels.size(); }
43};
44
45#define invokeOnRelocs(sec, f, ...) \
46 { \
47 const RelsOrRelas<ELFT> rs = (sec).template relsOrRelas<ELFT>(); \
48 if (rs.areRelocsCrel()) \
49 f(__VA_ARGS__, rs.crels); \
50 else if (rs.areRelocsRel()) \
51 f(__VA_ARGS__, rs.rels); \
52 else \
53 f(__VA_ARGS__, rs.relas); \
54 }
55
56// This is the base class of all sections that lld handles. Some are sections in
57// input files, some are sections in the produced output file and some exist
58// just as a convenience for implementing special ways of combining some
59// sections.
60class SectionBase {
61public:
62 enum Kind : uint8_t {
63 Regular,
64 Synthetic,
65 Spill,
66 EHFrame,
67 Merge,
68 Output,
69 Class,
70 };
71
72 Kind kind() const { return sectionKind; }
73
74 // The file which contains this section. For InputSectionBase, its dynamic
75 // type is usually ObjFile<ELFT>, but may be an InputFile of InternalKind
76 // (for a synthetic section).
77 InputFile *file;
78
79 StringRef name;
80
81 // The 1-indexed partition that this section is assigned to by the garbage
82 // collector, or 0 if this section is dead. Normally there is only one
83 // partition, so this will either be 0 or 1.
84 elf::Partition &getPartition(Ctx &) const;
85
86 // These corresponds to the fields in Elf_Shdr.
87 uint64_t flags;
88 uint32_t type;
89 uint32_t link;
90 uint32_t info;
91 uint32_t addralign;
92 uint32_t entsize;
93
94 Kind sectionKind;
95 uint8_t partition = 1;
96
97 // The next two bit fields are only used by InputSectionBase, but we
98 // put them here so the struct packs better.
99
100 Ctx &getCtx() const;
101 OutputSection *getOutputSection();
102 const OutputSection *getOutputSection() const {
103 return const_cast<SectionBase *>(this)->getOutputSection();
104 }
105
106 // Translate an offset in the input section to an offset in the output
107 // section.
108 uint64_t getOffset(uint64_t offset) const;
109
110 uint64_t getVA(uint64_t offset = 0) const;
111
112 bool isLive() const { return partition != 0; }
113 void markLive() { partition = 1; }
114 void markDead() { partition = 0; }
115
116protected:
117 constexpr SectionBase(Kind sectionKind, InputFile *file, StringRef name,
118 uint32_t type, uint64_t flags, uint32_t link,
119 uint32_t info, uint32_t addralign, uint32_t entsize)
120 : file(file), name(name), flags(flags), type(type), link(link),
121 info(info), addralign(addralign), entsize(entsize),
122 sectionKind(sectionKind) {}
123};
124
125struct SymbolAnchor {
126 uint64_t offset;
127 Defined *d;
128 bool end; // true for the anchor of st_value+st_size
129};
130
131struct RelaxAux {
132 // This records symbol start and end offsets which will be adjusted according
133 // to the nearest relocDeltas element.
134 SmallVector<SymbolAnchor, 0> anchors;
135 // For relocations[i], the actual offset is
136 // r_offset - (i ? relocDeltas[i-1] : 0).
137 std::unique_ptr<uint32_t[]> relocDeltas;
138 // For relocations[i], the actual type is relocTypes[i].
139 std::unique_ptr<RelType[]> relocTypes;
140 SmallVector<uint32_t, 0> writes;
141};
142
143// This corresponds to a section of an input file.
144class InputSectionBase : public SectionBase {
145public:
146 struct ObjMsg {
147 const InputSectionBase *sec;
148 uint64_t offset;
149 };
150 struct SrcMsg {
151 const InputSectionBase &sec;
152 const Symbol &sym;
153 uint64_t offset;
154 };
155
156 template <class ELFT>
157 InputSectionBase(ObjFile<ELFT> &file, const typename ELFT::Shdr &header,
158 StringRef name, Kind sectionKind);
159
160 InputSectionBase(InputFile *file, StringRef name, uint32_t type,
161 uint64_t flags, uint32_t link, uint32_t info,
162 uint32_t addralign, uint32_t entsize, ArrayRef<uint8_t> data,
163 Kind sectionKind);
164
165 static bool classof(const SectionBase *s) {
166 return s->kind() != Output && s->kind() != Class;
167 }
168
169 LLVM_PREFERRED_TYPE(bool)
170 uint8_t bss : 1;
171
172 // Whether this section is SHT_CREL and has been decoded to RELA by
173 // relsOrRelas.
174 LLVM_PREFERRED_TYPE(bool)
175 uint8_t decodedCrel : 1;
176
177 // Set for sections that should not be folded by ICF.
178 LLVM_PREFERRED_TYPE(bool)
179 uint8_t keepUnique : 1;
180
181 // Whether the section needs to be padded with a NOP filler due to
182 // deleteFallThruJmpInsn.
183 LLVM_PREFERRED_TYPE(bool)
184 uint8_t nopFiller : 1;
185
186 mutable bool compressed = false;
187
188 // Input sections are part of an output section. Special sections
189 // like .eh_frame and merge sections are first combined into a
190 // synthetic section that is then added to an output section. In all
191 // cases this points one level up.
192 SectionBase *parent = nullptr;
193
194 // Section index of the relocation section if exists.
195 uint32_t relSecIdx = 0;
196
197 // Getter when the dynamic type is ObjFile<ELFT>.
198 template <class ELFT> ObjFile<ELFT> *getFile() const {
199 return cast<ObjFile<ELFT>>(file);
200 }
201
202 // Used by --optimize-bb-jumps and RISC-V linker relaxation temporarily to
203 // indicate the number of bytes which is not counted in the size. This should
204 // be reset to zero after uses.
205 uint32_t bytesDropped = 0;
206
207 void drop_back(unsigned num) {
208 assert(bytesDropped + num < 256);
209 bytesDropped += num;
210 }
211
212 void push_back(uint64_t num) {
213 assert(bytesDropped >= num);
214 bytesDropped -= num;
215 }
216
217 mutable const uint8_t *content_;
218 uint64_t size;
219
220 void trim() {
221 if (bytesDropped) {
222 size -= bytesDropped;
223 bytesDropped = 0;
224 }
225 }
226
227 ArrayRef<uint8_t> content() const {
228 return ArrayRef<uint8_t>(content_, size);
229 }
230 ArrayRef<uint8_t> contentMaybeDecompress() const {
231 if (compressed)
232 decompress();
233 return content();
234 }
235
236 // The next member in the section group if this section is in a group. This is
237 // used by --gc-sections.
238 InputSectionBase *nextInSectionGroup = nullptr;
239
240 template <class ELFT>
241 RelsOrRelas<ELFT> relsOrRelas(bool supportsCrel = true) const;
242
243 // InputSections that are dependent on us (reverse dependency for GC)
244 llvm::TinyPtrVector<InputSection *> dependentSections;
245
246 // Returns the size of this section (even if this is a common or BSS.)
247 size_t getSize() const;
248
249 InputSection *getLinkOrderDep() const;
250
251 // Get a symbol that encloses this offset from within the section. If type is
252 // not zero, return a symbol with the specified type.
253 Defined *getEnclosingSymbol(uint64_t offset, uint8_t type = 0) const;
254 Defined *getEnclosingFunction(uint64_t offset) const {
255 return getEnclosingSymbol(offset, type: llvm::ELF::STT_FUNC);
256 }
257
258 // Returns a source location string. Used to construct an error message.
259 std::string getLocation(uint64_t offset) const;
260 ObjMsg getObjMsg(uint64_t offset) const { return {.sec: this, .offset: offset}; }
261 SrcMsg getSrcMsg(const Symbol &sym, uint64_t offset) const {
262 return {.sec: *this, .sym: sym, .offset: offset};
263 }
264
265 // Each section knows how to relocate itself. These functions apply
266 // relocations, assuming that Buf points to this section's copy in
267 // the mmap'ed output buffer.
268 template <class ELFT> void relocate(Ctx &, uint8_t *buf, uint8_t *bufEnd);
269 uint64_t getRelocTargetVA(Ctx &, const Relocation &r, uint64_t p) const;
270
271 // The native ELF reloc data type is not very convenient to handle.
272 // So we convert ELF reloc records to our own records in Relocations.cpp.
273 // This vector contains such "cooked" relocations.
274 SmallVector<Relocation, 0> relocations;
275
276 void addReloc(const Relocation &r) { relocations.push_back(Elt: r); }
277 MutableArrayRef<Relocation> relocs() { return relocations; }
278 ArrayRef<Relocation> relocs() const { return relocations; }
279
280 union {
281 // These are modifiers to jump instructions that are necessary when basic
282 // block sections are enabled. Basic block sections creates opportunities
283 // to relax jump instructions at basic block boundaries after reordering the
284 // basic blocks.
285 JumpInstrMod *jumpInstrMod = nullptr;
286
287 // Auxiliary information for RISC-V and LoongArch linker relaxation.
288 // They do not use jumpInstrMod.
289 RelaxAux *relaxAux;
290
291 // The compressed content size when `compressed` is true.
292 size_t compressedSize;
293 };
294
295 // A function compiled with -fsplit-stack calling a function
296 // compiled without -fsplit-stack needs its prologue adjusted. Find
297 // such functions and adjust their prologues. This is very similar
298 // to relocation. See https://gcc.gnu.org/wiki/SplitStacks for more
299 // information.
300 template <typename ELFT>
301 void adjustSplitStackFunctionPrologues(Ctx &, uint8_t *buf, uint8_t *end);
302
303 template <typename T> llvm::ArrayRef<T> getDataAs() const {
304 size_t s = content().size();
305 assert(s % sizeof(T) == 0);
306 return llvm::ArrayRef<T>((const T *)content().data(), s / sizeof(T));
307 }
308
309protected:
310 template <typename ELFT> void parseCompressedHeader(Ctx &);
311 void decompress() const;
312};
313
314// SectionPiece represents a piece of splittable section contents.
315// We allocate a lot of these and binary search on them. This means that they
316// have to be as compact as possible, which is why we don't store the size (can
317// be found by looking at the next one).
318struct SectionPiece {
319 SectionPiece() = default;
320 SectionPiece(size_t off, uint32_t hash, bool live)
321 : inputOff(off), live(live), hash(hash >> 1) {}
322
323 uint32_t inputOff;
324 LLVM_PREFERRED_TYPE(bool)
325 uint32_t live : 1;
326 uint32_t hash : 31;
327 uint64_t outputOff = 0;
328};
329
330static_assert(sizeof(SectionPiece) == 16, "SectionPiece is too big");
331
332// This corresponds to a SHF_MERGE section of an input file.
333class MergeInputSection : public InputSectionBase {
334public:
335 template <class ELFT>
336 MergeInputSection(ObjFile<ELFT> &f, const typename ELFT::Shdr &header,
337 StringRef name);
338 MergeInputSection(Ctx &, StringRef name, uint32_t type, uint64_t flags,
339 uint64_t entsize, ArrayRef<uint8_t> data);
340
341 static bool classof(const SectionBase *s) { return s->kind() == Merge; }
342 void splitIntoPieces();
343
344 // Translate an offset in the input section to an offset in the parent
345 // MergeSyntheticSection.
346 uint64_t getParentOffset(uint64_t offset) const;
347
348 // Splittable sections are handled as a sequence of data
349 // rather than a single large blob of data.
350 SmallVector<SectionPiece, 0> pieces;
351
352 // Returns I'th piece's data. This function is very hot when
353 // string merging is enabled, so we want to inline.
354 LLVM_ATTRIBUTE_ALWAYS_INLINE
355 llvm::CachedHashStringRef getData(size_t i) const {
356 size_t begin = pieces[i].inputOff;
357 size_t end =
358 (pieces.size() - 1 == i) ? content().size() : pieces[i + 1].inputOff;
359 return {toStringRef(Input: content().slice(N: begin, M: end - begin)), pieces[i].hash};
360 }
361
362 // Returns the SectionPiece at a given input section offset.
363 SectionPiece &getSectionPiece(uint64_t offset);
364 const SectionPiece &getSectionPiece(uint64_t offset) const {
365 return const_cast<MergeInputSection *>(this)->getSectionPiece(offset);
366 }
367
368 SyntheticSection *getParent() const {
369 return cast_or_null<SyntheticSection>(Val: parent);
370 }
371
372private:
373 void splitStrings(StringRef s, size_t size);
374 void splitNonStrings(ArrayRef<uint8_t> a, size_t size);
375};
376
377struct EhSectionPiece {
378 EhSectionPiece(size_t off, InputSectionBase *sec, uint32_t size,
379 unsigned firstRelocation)
380 : inputOff(off), sec(sec), size(size), firstRelocation(firstRelocation) {}
381
382 ArrayRef<uint8_t> data() const {
383 return {sec->content().data() + this->inputOff, size};
384 }
385
386 size_t inputOff;
387 ssize_t outputOff = -1;
388 InputSectionBase *sec;
389 uint32_t size;
390 unsigned firstRelocation;
391};
392
393// This corresponds to a .eh_frame section of an input file.
394class EhInputSection : public InputSectionBase {
395public:
396 template <class ELFT>
397 EhInputSection(ObjFile<ELFT> &f, const typename ELFT::Shdr &header,
398 StringRef name);
399 static bool classof(const SectionBase *s) { return s->kind() == EHFrame; }
400 template <class ELFT> void split();
401 template <class ELFT, class RelTy> void split(ArrayRef<RelTy> rels);
402
403 // Splittable sections are handled as a sequence of data
404 // rather than a single large blob of data.
405 SmallVector<EhSectionPiece, 0> cies, fdes;
406
407 SyntheticSection *getParent() const;
408 uint64_t getParentOffset(uint64_t offset) const;
409};
410
411// This is a section that is added directly to an output section
412// instead of needing special combination via a synthetic section. This
413// includes all input sections with the exceptions of SHF_MERGE and
414// .eh_frame. It also includes the synthetic sections themselves.
415class InputSection : public InputSectionBase {
416public:
417 InputSection(InputFile *f, StringRef name, uint32_t type, uint64_t flags,
418 uint32_t addralign, uint32_t entsize, ArrayRef<uint8_t> data,
419 Kind k = Regular);
420 template <class ELFT>
421 InputSection(ObjFile<ELFT> &f, const typename ELFT::Shdr &header,
422 StringRef name);
423
424 static bool classof(const SectionBase *s) {
425 return s->kind() == SectionBase::Regular ||
426 s->kind() == SectionBase::Synthetic ||
427 s->kind() == SectionBase::Spill;
428 }
429
430 // Write this section to a mmap'ed file, assuming Buf is pointing to
431 // beginning of the output section.
432 template <class ELFT> void writeTo(Ctx &, uint8_t *buf);
433
434 OutputSection *getParent() const {
435 return reinterpret_cast<OutputSection *>(parent);
436 }
437
438 // This variable has two usages. Initially, it represents an index in the
439 // OutputSection's InputSection list, and is used when ordering SHF_LINK_ORDER
440 // sections. After assignAddresses is called, it represents the offset from
441 // the beginning of the output section this section was assigned to.
442 uint64_t outSecOff = 0;
443
444 InputSectionBase *getRelocatedSection() const;
445
446 template <class ELFT, class RelTy>
447 void relocateNonAlloc(Ctx &, uint8_t *buf, Relocs<RelTy> rels);
448
449 // Points to the canonical section. If ICF folds two sections, repl pointer of
450 // one section points to the other.
451 InputSection *repl = this;
452
453 // Used by ICF.
454 uint32_t eqClass[2] = {0, 0};
455
456 // Called by ICF to merge two input sections.
457 void replace(InputSection *other);
458
459 static InputSection discarded;
460
461private:
462 template <class ELFT, class RelTy> void copyRelocations(Ctx &, uint8_t *buf);
463
464 template <class ELFT, class RelTy, class RelIt>
465 void copyRelocations(Ctx &, uint8_t *buf, llvm::iterator_range<RelIt> rels);
466
467 template <class ELFT> void copyShtGroup(uint8_t *buf);
468};
469
470// A marker for a potential spill location for another input section. This
471// broadly acts as if it were the original section until address assignment.
472// Then it is either replaced with the real input section or removed.
473class PotentialSpillSection : public InputSection {
474public:
475 // The containing input section description; used to quickly replace this stub
476 // with the actual section.
477 InputSectionDescription *isd;
478
479 // Next potential spill location for the same source input section.
480 PotentialSpillSection *next = nullptr;
481
482 PotentialSpillSection(const InputSectionBase &source,
483 InputSectionDescription &isd);
484
485 static bool classof(const SectionBase *sec) {
486 return sec->kind() == InputSectionBase::Spill;
487 }
488};
489
490#ifndef _WIN32
491static_assert(sizeof(InputSection) <= 152, "InputSection is too big");
492#endif
493
494class SyntheticSection : public InputSection {
495public:
496 Ctx &ctx;
497 SyntheticSection(Ctx &ctx, StringRef name, uint32_t type, uint64_t flags,
498 uint32_t addralign)
499 : InputSection(ctx.internalFile, name, type, flags, addralign,
500 /*entsize=*/0, {}, InputSectionBase::Synthetic),
501 ctx(ctx) {}
502
503 virtual ~SyntheticSection() = default;
504 virtual size_t getSize() const = 0;
505 virtual bool updateAllocSize(Ctx &) { return false; }
506 // If the section has the SHF_ALLOC flag and the size may be changed if
507 // thunks are added, update the section size.
508 virtual bool isNeeded() const { return true; }
509 virtual void finalizeContents() {}
510 virtual void writeTo(uint8_t *buf) = 0;
511
512 static bool classof(const SectionBase *sec) {
513 return sec->kind() == InputSectionBase::Synthetic;
514 }
515};
516
517inline bool isStaticRelSecType(uint32_t type) {
518 return type == llvm::ELF::SHT_RELA || type == llvm::ELF::SHT_CREL ||
519 type == llvm::ELF::SHT_REL;
520}
521
522inline bool isDebugSection(const InputSectionBase &sec) {
523 return (sec.flags & llvm::ELF::SHF_ALLOC) == 0 &&
524 sec.name.starts_with(Prefix: ".debug");
525}
526
527std::string toStr(elf::Ctx &, const elf::InputSectionBase *);
528const ELFSyncStream &operator<<(const ELFSyncStream &,
529 const InputSectionBase *);
530const ELFSyncStream &operator<<(const ELFSyncStream &,
531 InputSectionBase::ObjMsg &&);
532const ELFSyncStream &operator<<(const ELFSyncStream &,
533 InputSectionBase::SrcMsg &&);
534} // namespace elf
535} // namespace lld
536
537#endif
538