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
36LLVM_LIBRARY_VISIBILITY extern std::vector<Partition> partitions;
37
38// Returned by InputSectionBase::relsOrRelas. At most one member is empty.
39template <class ELFT> struct RelsOrRelas {
40 Relocs<typename ELFT::Rel> rels;
41 Relocs<typename ELFT::Rela> relas;
42 Relocs<typename ELFT::Crel> crels;
43 bool areRelocsRel() const { return rels.size(); }
44 bool areRelocsCrel() const { return crels.size(); }
45};
46
47#define invokeOnRelocs(sec, f, ...) \
48 { \
49 const RelsOrRelas<ELFT> rs = (sec).template relsOrRelas<ELFT>(); \
50 if (rs.areRelocsCrel()) \
51 f(__VA_ARGS__, rs.crels); \
52 else if (rs.areRelocsRel()) \
53 f(__VA_ARGS__, rs.rels); \
54 else \
55 f(__VA_ARGS__, rs.relas); \
56 }
57
58// This is the base class of all sections that lld handles. Some are sections in
59// input files, some are sections in the produced output file and some exist
60// just as a convenience for implementing special ways of combining some
61// sections.
62class SectionBase {
63public:
64 enum Kind { Regular, Synthetic, Spill, EHFrame, Merge, Output };
65
66 Kind kind() const { return (Kind)sectionKind; }
67
68 LLVM_PREFERRED_TYPE(Kind)
69 uint8_t sectionKind : 3;
70
71 // The next two bit fields are only used by InputSectionBase, but we
72 // put them here so the struct packs better.
73
74 LLVM_PREFERRED_TYPE(bool)
75 uint8_t bss : 1;
76
77 // Set for sections that should not be folded by ICF.
78 LLVM_PREFERRED_TYPE(bool)
79 uint8_t keepUnique : 1;
80
81 uint8_t partition = 1;
82 uint32_t type;
83 StringRef name;
84
85 // The 1-indexed partition that this section is assigned to by the garbage
86 // collector, or 0 if this section is dead. Normally there is only one
87 // partition, so this will either be 0 or 1.
88 elf::Partition &getPartition() const;
89
90 // These corresponds to the fields in Elf_Shdr.
91 uint64_t flags;
92 uint32_t addralign;
93 uint32_t entsize;
94 uint32_t link;
95 uint32_t info;
96
97 OutputSection *getOutputSection();
98 const OutputSection *getOutputSection() const {
99 return const_cast<SectionBase *>(this)->getOutputSection();
100 }
101
102 // Translate an offset in the input section to an offset in the output
103 // section.
104 uint64_t getOffset(uint64_t offset) const;
105
106 uint64_t getVA(uint64_t offset = 0) const;
107
108 bool isLive() const { return partition != 0; }
109 void markLive() { partition = 1; }
110 void markDead() { partition = 0; }
111
112protected:
113 constexpr SectionBase(Kind sectionKind, StringRef name, uint64_t flags,
114 uint32_t entsize, uint32_t addralign, uint32_t type,
115 uint32_t info, uint32_t link)
116 : sectionKind(sectionKind), bss(false), keepUnique(false), type(type),
117 name(name), flags(flags), addralign(addralign), entsize(entsize),
118 link(link), info(info) {}
119};
120
121struct SymbolAnchor {
122 uint64_t offset;
123 Defined *d;
124 bool end; // true for the anchor of st_value+st_size
125};
126
127struct RelaxAux {
128 // This records symbol start and end offsets which will be adjusted according
129 // to the nearest relocDeltas element.
130 SmallVector<SymbolAnchor, 0> anchors;
131 // For relocations[i], the actual offset is
132 // r_offset - (i ? relocDeltas[i-1] : 0).
133 std::unique_ptr<uint32_t[]> relocDeltas;
134 // For relocations[i], the actual type is relocTypes[i].
135 std::unique_ptr<RelType[]> relocTypes;
136 SmallVector<uint32_t, 0> writes;
137};
138
139// This corresponds to a section of an input file.
140class InputSectionBase : public SectionBase {
141public:
142 template <class ELFT>
143 InputSectionBase(ObjFile<ELFT> &file, const typename ELFT::Shdr &header,
144 StringRef name, Kind sectionKind);
145
146 InputSectionBase(InputFile *file, uint64_t flags, uint32_t type,
147 uint64_t entsize, uint32_t link, uint32_t info,
148 uint32_t addralign, ArrayRef<uint8_t> data, StringRef name,
149 Kind sectionKind);
150
151 static bool classof(const SectionBase *s) { return s->kind() != Output; }
152
153 // The file which contains this section. Its dynamic type is usually
154 // ObjFile<ELFT>, but may be an InputFile of InternalKind (for a synthetic
155 // section).
156 InputFile *file;
157
158 // Input sections are part of an output section. Special sections
159 // like .eh_frame and merge sections are first combined into a
160 // synthetic section that is then added to an output section. In all
161 // cases this points one level up.
162 SectionBase *parent = nullptr;
163
164 // Section index of the relocation section if exists.
165 uint32_t relSecIdx = 0;
166
167 // Getter when the dynamic type is ObjFile<ELFT>.
168 template <class ELFT> ObjFile<ELFT> *getFile() const {
169 return cast<ObjFile<ELFT>>(file);
170 }
171
172 // Used by --optimize-bb-jumps and RISC-V linker relaxation temporarily to
173 // indicate the number of bytes which is not counted in the size. This should
174 // be reset to zero after uses.
175 uint32_t bytesDropped = 0;
176
177 mutable bool compressed = false;
178
179 // Whether the section needs to be padded with a NOP filler due to
180 // deleteFallThruJmpInsn.
181 bool nopFiller = false;
182
183 void drop_back(unsigned num) {
184 assert(bytesDropped + num < 256);
185 bytesDropped += num;
186 }
187
188 void push_back(uint64_t num) {
189 assert(bytesDropped >= num);
190 bytesDropped -= num;
191 }
192
193 mutable const uint8_t *content_;
194 uint64_t size;
195
196 void trim() {
197 if (bytesDropped) {
198 size -= bytesDropped;
199 bytesDropped = 0;
200 }
201 }
202
203 ArrayRef<uint8_t> content() const {
204 return ArrayRef<uint8_t>(content_, size);
205 }
206 ArrayRef<uint8_t> contentMaybeDecompress() const {
207 if (compressed)
208 decompress();
209 return content();
210 }
211
212 // The next member in the section group if this section is in a group. This is
213 // used by --gc-sections.
214 InputSectionBase *nextInSectionGroup = nullptr;
215
216 template <class ELFT>
217 RelsOrRelas<ELFT> relsOrRelas(bool supportsCrel = true) const;
218
219 // InputSections that are dependent on us (reverse dependency for GC)
220 llvm::TinyPtrVector<InputSection *> dependentSections;
221
222 // Returns the size of this section (even if this is a common or BSS.)
223 size_t getSize() const;
224
225 InputSection *getLinkOrderDep() const;
226
227 // Get a symbol that encloses this offset from within the section. If type is
228 // not zero, return a symbol with the specified type.
229 Defined *getEnclosingSymbol(uint64_t offset, uint8_t type = 0) const;
230 Defined *getEnclosingFunction(uint64_t offset) const {
231 return getEnclosingSymbol(offset, type: llvm::ELF::STT_FUNC);
232 }
233
234 // Returns a source location string. Used to construct an error message.
235 std::string getLocation(uint64_t offset) const;
236 std::string getSrcMsg(const Symbol &sym, uint64_t offset) const;
237 std::string getObjMsg(uint64_t offset) const;
238
239 // Each section knows how to relocate itself. These functions apply
240 // relocations, assuming that Buf points to this section's copy in
241 // the mmap'ed output buffer.
242 template <class ELFT> void relocate(uint8_t *buf, uint8_t *bufEnd);
243 static uint64_t getRelocTargetVA(const InputFile *File, RelType Type,
244 int64_t A, uint64_t P, const Symbol &Sym,
245 RelExpr Expr);
246
247 // The native ELF reloc data type is not very convenient to handle.
248 // So we convert ELF reloc records to our own records in Relocations.cpp.
249 // This vector contains such "cooked" relocations.
250 SmallVector<Relocation, 0> relocations;
251
252 void addReloc(const Relocation &r) { relocations.push_back(Elt: r); }
253 MutableArrayRef<Relocation> relocs() { return relocations; }
254 ArrayRef<Relocation> relocs() const { return relocations; }
255
256 union {
257 // These are modifiers to jump instructions that are necessary when basic
258 // block sections are enabled. Basic block sections creates opportunities
259 // to relax jump instructions at basic block boundaries after reordering the
260 // basic blocks.
261 JumpInstrMod *jumpInstrMod = nullptr;
262
263 // Auxiliary information for RISC-V and LoongArch linker relaxation.
264 // They do not use jumpInstrMod.
265 RelaxAux *relaxAux;
266
267 // The compressed content size when `compressed` is true.
268 size_t compressedSize;
269 };
270
271 // A function compiled with -fsplit-stack calling a function
272 // compiled without -fsplit-stack needs its prologue adjusted. Find
273 // such functions and adjust their prologues. This is very similar
274 // to relocation. See https://gcc.gnu.org/wiki/SplitStacks for more
275 // information.
276 template <typename ELFT>
277 void adjustSplitStackFunctionPrologues(uint8_t *buf, uint8_t *end);
278
279
280 template <typename T> llvm::ArrayRef<T> getDataAs() const {
281 size_t s = content().size();
282 assert(s % sizeof(T) == 0);
283 return llvm::ArrayRef<T>((const T *)content().data(), s / sizeof(T));
284 }
285
286protected:
287 template <typename ELFT>
288 void parseCompressedHeader();
289 void decompress() const;
290};
291
292// SectionPiece represents a piece of splittable section contents.
293// We allocate a lot of these and binary search on them. This means that they
294// have to be as compact as possible, which is why we don't store the size (can
295// be found by looking at the next one).
296struct SectionPiece {
297 SectionPiece() = default;
298 SectionPiece(size_t off, uint32_t hash, bool live)
299 : inputOff(off), live(live), hash(hash >> 1) {}
300
301 uint32_t inputOff;
302 LLVM_PREFERRED_TYPE(bool)
303 uint32_t live : 1;
304 uint32_t hash : 31;
305 uint64_t outputOff = 0;
306};
307
308static_assert(sizeof(SectionPiece) == 16, "SectionPiece is too big");
309
310// This corresponds to a SHF_MERGE section of an input file.
311class MergeInputSection : public InputSectionBase {
312public:
313 template <class ELFT>
314 MergeInputSection(ObjFile<ELFT> &f, const typename ELFT::Shdr &header,
315 StringRef name);
316 MergeInputSection(uint64_t flags, uint32_t type, uint64_t entsize,
317 ArrayRef<uint8_t> data, StringRef name);
318
319 static bool classof(const SectionBase *s) { return s->kind() == Merge; }
320 void splitIntoPieces();
321
322 // Translate an offset in the input section to an offset in the parent
323 // MergeSyntheticSection.
324 uint64_t getParentOffset(uint64_t offset) const;
325
326 // Splittable sections are handled as a sequence of data
327 // rather than a single large blob of data.
328 SmallVector<SectionPiece, 0> pieces;
329
330 // Returns I'th piece's data. This function is very hot when
331 // string merging is enabled, so we want to inline.
332 LLVM_ATTRIBUTE_ALWAYS_INLINE
333 llvm::CachedHashStringRef getData(size_t i) const {
334 size_t begin = pieces[i].inputOff;
335 size_t end =
336 (pieces.size() - 1 == i) ? content().size() : pieces[i + 1].inputOff;
337 return {toStringRef(Input: content().slice(N: begin, M: end - begin)), pieces[i].hash};
338 }
339
340 // Returns the SectionPiece at a given input section offset.
341 SectionPiece &getSectionPiece(uint64_t offset);
342 const SectionPiece &getSectionPiece(uint64_t offset) const {
343 return const_cast<MergeInputSection *>(this)->getSectionPiece(offset);
344 }
345
346 SyntheticSection *getParent() const {
347 return cast_or_null<SyntheticSection>(Val: parent);
348 }
349
350private:
351 void splitStrings(StringRef s, size_t size);
352 void splitNonStrings(ArrayRef<uint8_t> a, size_t size);
353};
354
355struct EhSectionPiece {
356 EhSectionPiece(size_t off, InputSectionBase *sec, uint32_t size,
357 unsigned firstRelocation)
358 : inputOff(off), sec(sec), size(size), firstRelocation(firstRelocation) {}
359
360 ArrayRef<uint8_t> data() const {
361 return {sec->content().data() + this->inputOff, size};
362 }
363
364 size_t inputOff;
365 ssize_t outputOff = -1;
366 InputSectionBase *sec;
367 uint32_t size;
368 unsigned firstRelocation;
369};
370
371// This corresponds to a .eh_frame section of an input file.
372class EhInputSection : public InputSectionBase {
373public:
374 template <class ELFT>
375 EhInputSection(ObjFile<ELFT> &f, const typename ELFT::Shdr &header,
376 StringRef name);
377 static bool classof(const SectionBase *s) { return s->kind() == EHFrame; }
378 template <class ELFT> void split();
379 template <class ELFT, class RelTy> void split(ArrayRef<RelTy> rels);
380
381 // Splittable sections are handled as a sequence of data
382 // rather than a single large blob of data.
383 SmallVector<EhSectionPiece, 0> cies, fdes;
384
385 SyntheticSection *getParent() const;
386 uint64_t getParentOffset(uint64_t offset) const;
387};
388
389// This is a section that is added directly to an output section
390// instead of needing special combination via a synthetic section. This
391// includes all input sections with the exceptions of SHF_MERGE and
392// .eh_frame. It also includes the synthetic sections themselves.
393class InputSection : public InputSectionBase {
394public:
395 InputSection(InputFile *f, uint64_t flags, uint32_t type, uint32_t addralign,
396 ArrayRef<uint8_t> data, StringRef name, Kind k = Regular);
397 template <class ELFT>
398 InputSection(ObjFile<ELFT> &f, const typename ELFT::Shdr &header,
399 StringRef name);
400
401 static bool classof(const SectionBase *s) {
402 return s->kind() == SectionBase::Regular ||
403 s->kind() == SectionBase::Synthetic ||
404 s->kind() == SectionBase::Spill;
405 }
406
407 // Write this section to a mmap'ed file, assuming Buf is pointing to
408 // beginning of the output section.
409 template <class ELFT> void writeTo(uint8_t *buf);
410
411 OutputSection *getParent() const {
412 return reinterpret_cast<OutputSection *>(parent);
413 }
414
415 // This variable has two usages. Initially, it represents an index in the
416 // OutputSection's InputSection list, and is used when ordering SHF_LINK_ORDER
417 // sections. After assignAddresses is called, it represents the offset from
418 // the beginning of the output section this section was assigned to.
419 uint64_t outSecOff = 0;
420
421 InputSectionBase *getRelocatedSection() const;
422
423 template <class ELFT, class RelTy>
424 void relocateNonAlloc(uint8_t *buf, Relocs<RelTy> rels);
425
426 // Points to the canonical section. If ICF folds two sections, repl pointer of
427 // one section points to the other.
428 InputSection *repl = this;
429
430 // Used by ICF.
431 uint32_t eqClass[2] = {0, 0};
432
433 // Called by ICF to merge two input sections.
434 void replace(InputSection *other);
435
436 static InputSection discarded;
437
438private:
439 template <class ELFT, class RelTy> void copyRelocations(uint8_t *buf);
440
441 template <class ELFT, class RelTy, class RelIt>
442 void copyRelocations(uint8_t *buf, llvm::iterator_range<RelIt> rels);
443
444 template <class ELFT> void copyShtGroup(uint8_t *buf);
445};
446
447// A marker for a potential spill location for another input section. This
448// broadly acts as if it were the original section until address assignment.
449// Then it is either replaced with the real input section or removed.
450class PotentialSpillSection : public InputSection {
451public:
452 // The containing input section description; used to quickly replace this stub
453 // with the actual section.
454 InputSectionDescription *isd;
455
456 // Next potential spill location for the same source input section.
457 PotentialSpillSection *next = nullptr;
458
459 PotentialSpillSection(const InputSectionBase &source,
460 InputSectionDescription &isd);
461
462 static bool classof(const SectionBase *sec) {
463 return sec->kind() == InputSectionBase::Spill;
464 }
465};
466
467static_assert(sizeof(InputSection) <= 160, "InputSection is too big");
468
469class SyntheticSection : public InputSection {
470public:
471 SyntheticSection(uint64_t flags, uint32_t type, uint32_t addralign,
472 StringRef name)
473 : InputSection(ctx.internalFile, flags, type, addralign, {}, name,
474 InputSectionBase::Synthetic) {}
475
476 virtual ~SyntheticSection() = default;
477 virtual size_t getSize() const = 0;
478 virtual bool updateAllocSize() { return false; }
479 // If the section has the SHF_ALLOC flag and the size may be changed if
480 // thunks are added, update the section size.
481 virtual bool isNeeded() const { return true; }
482 virtual void finalizeContents() {}
483 virtual void writeTo(uint8_t *buf) = 0;
484
485 static bool classof(const SectionBase *sec) {
486 return sec->kind() == InputSectionBase::Synthetic;
487 }
488};
489
490inline bool isStaticRelSecType(uint32_t type) {
491 return type == llvm::ELF::SHT_RELA || type == llvm::ELF::SHT_CREL ||
492 type == llvm::ELF::SHT_REL;
493}
494
495inline bool isDebugSection(const InputSectionBase &sec) {
496 return (sec.flags & llvm::ELF::SHF_ALLOC) == 0 &&
497 sec.name.starts_with(Prefix: ".debug");
498}
499
500// The set of TOC entries (.toc + addend) for which we should not apply
501// toc-indirect to toc-relative relaxation. const Symbol * refers to the
502// STT_SECTION symbol associated to the .toc input section.
503extern llvm::DenseSet<std::pair<const Symbol *, uint64_t>> ppc64noTocRelax;
504
505} // namespace elf
506
507std::string toString(const elf::InputSectionBase *);
508} // namespace lld
509
510#endif
511