1 //===- Chunks.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_COFF_CHUNKS_H
10 #define LLD_COFF_CHUNKS_H
11
12 #include "Config.h"
13 #include "InputFiles.h"
14 #include "lld/Common/LLVM.h"
15 #include "llvm/ADT/ArrayRef.h"
16 #include "llvm/ADT/PointerIntPair.h"
17 #include "llvm/ADT/iterator.h"
18 #include "llvm/ADT/iterator_range.h"
19 #include "llvm/MC/StringTableBuilder.h"
20 #include "llvm/Object/COFF.h"
21 #include "llvm/Object/WindowsMachineFlag.h"
22 #include <utility>
23 #include <vector>
24
25 namespace lld::coff {
26
27 using llvm::COFF::ImportDirectoryTableEntry;
28 using llvm::object::chpe_range_type;
29 using llvm::object::coff_relocation;
30 using llvm::object::coff_section;
31 using llvm::object::COFFSymbolRef;
32 using llvm::object::SectionRef;
33
34 class Baserel;
35 class Defined;
36 class DefinedImportData;
37 class DefinedRegular;
38 class ObjFile;
39 class OutputSection;
40 class RuntimePseudoReloc;
41 class Symbol;
42
43 // Mask for permissions (discardable, writable, readable, executable, etc).
44 const uint32_t permMask = 0xFE000000;
45
46 // Mask for section types (code, data, bss).
47 const uint32_t typeMask = 0x000000E0;
48
49 // The log base 2 of the largest section alignment, which is log2(8192), or 13.
50 enum : unsigned { Log2MaxSectionAlignment = 13 };
51
52 // A Chunk represents a chunk of data that will occupy space in the
53 // output (if the resolver chose that). It may or may not be backed by
54 // a section of an input file. It could be linker-created data, or
55 // doesn't even have actual data (if common or bss).
56 class Chunk {
57 public:
58 enum Kind : uint8_t {
59 SectionKind,
60 SectionECKind,
61 OtherKind,
62 ImportThunkKind
63 };
kind()64 Kind kind() const { return chunkKind; }
65
66 // Returns the size of this chunk (even if this is a common or BSS.)
67 size_t getSize() const;
68
69 // Returns chunk alignment in power of two form. Value values are powers of
70 // two from 1 to 8192.
getAlignment()71 uint32_t getAlignment() const { return 1U << p2Align; }
72
73 // Update the chunk section alignment measured in bytes. Internally alignment
74 // is stored in log2.
setAlignment(uint32_t align)75 void setAlignment(uint32_t align) {
76 // Treat zero byte alignment as 1 byte alignment.
77 align = align ? align : 1;
78 assert(llvm::isPowerOf2_32(align) && "alignment is not a power of 2");
79 p2Align = llvm::Log2_32(align);
80 assert(p2Align <= Log2MaxSectionAlignment &&
81 "impossible requested alignment");
82 }
83
84 // Write this chunk to a mmap'ed file, assuming Buf is pointing to
85 // beginning of the file. Because this function may use RVA values
86 // of other chunks for relocations, you need to set them properly
87 // before calling this function.
88 void writeTo(uint8_t *buf) const;
89
90 // The writer sets and uses the addresses. In practice, PE images cannot be
91 // larger than 2GB. Chunks are always laid as part of the image, so Chunk RVAs
92 // can be stored with 32 bits.
getRVA()93 uint32_t getRVA() const { return rva; }
setRVA(uint64_t v)94 void setRVA(uint64_t v) {
95 // This may truncate. The writer checks for overflow later.
96 rva = (uint32_t)v;
97 }
98
99 // Returns readable/writable/executable bits.
100 uint32_t getOutputCharacteristics() const;
101
102 // Returns the section name if this is a section chunk.
103 // It is illegal to call this function on non-section chunks.
104 StringRef getSectionName() const;
105
106 // An output section has pointers to chunks in the section, and each
107 // chunk has a back pointer to an output section.
setOutputSectionIdx(uint16_t o)108 void setOutputSectionIdx(uint16_t o) { osidx = o; }
getOutputSectionIdx()109 uint16_t getOutputSectionIdx() const { return osidx; }
110
111 // Windows-specific.
112 // Collect all locations that contain absolute addresses for base relocations.
113 void getBaserels(std::vector<Baserel> *res);
114
115 // Returns a human-readable name of this chunk. Chunks are unnamed chunks of
116 // bytes, so this is used only for logging or debugging.
117 StringRef getDebugName() const;
118
119 // Return true if this file has the hotpatch flag set to true in the
120 // S_COMPILE3 record in codeview debug info. Also returns true for some thunks
121 // synthesized by the linker.
122 bool isHotPatchable() const;
123
124 MachineTypes getMachine() const;
125 llvm::Triple::ArchType getArch() const;
126 std::optional<chpe_range_type> getArm64ECRangeType() const;
127
128 // ARM64EC entry thunk associated with the chunk.
129 Defined *getEntryThunk() const;
130 void setEntryThunk(Defined *entryThunk);
131
132 protected:
chunkKind(k)133 Chunk(Kind k = OtherKind) : chunkKind(k), hasData(true), p2Align(0) {}
134
135 const Kind chunkKind;
136
137 public:
138 // Returns true if this has non-zero data. BSS chunks return
139 // false. If false is returned, the space occupied by this chunk
140 // will be filled with zeros. Corresponds to the
141 // IMAGE_SCN_CNT_UNINITIALIZED_DATA section characteristic bit.
142 uint8_t hasData : 1;
143
144 public:
145 // The alignment of this chunk, stored in log2 form. The writer uses the
146 // value.
147 uint8_t p2Align : 7;
148
149 // The output section index for this chunk. The first valid section number is
150 // one.
151 uint16_t osidx = 0;
152
153 // The RVA of this chunk in the output. The writer sets a value.
154 uint32_t rva = 0;
155 };
156
157 class NonSectionChunk : public Chunk {
158 public:
159 virtual ~NonSectionChunk() = default;
160
161 // Returns the size of this chunk (even if this is a common or BSS.)
162 virtual size_t getSize() const = 0;
163
getOutputCharacteristics()164 virtual uint32_t getOutputCharacteristics() const { return 0; }
165
166 // Write this chunk to a mmap'ed file, assuming Buf is pointing to
167 // beginning of the file. Because this function may use RVA values
168 // of other chunks for relocations, you need to set them properly
169 // before calling this function.
writeTo(uint8_t * buf)170 virtual void writeTo(uint8_t *buf) const {}
171
172 // Returns the section name if this is a section chunk.
173 // It is illegal to call this function on non-section chunks.
getSectionName()174 virtual StringRef getSectionName() const {
175 llvm_unreachable("unimplemented getSectionName");
176 }
177
178 // Windows-specific.
179 // Collect all locations that contain absolute addresses for base relocations.
getBaserels(std::vector<Baserel> * res)180 virtual void getBaserels(std::vector<Baserel> *res) {}
181
getMachine()182 virtual MachineTypes getMachine() const { return IMAGE_FILE_MACHINE_UNKNOWN; }
183
184 // Returns a human-readable name of this chunk. Chunks are unnamed chunks of
185 // bytes, so this is used only for logging or debugging.
getDebugName()186 virtual StringRef getDebugName() const { return ""; }
187
classof(const Chunk * c)188 static bool classof(const Chunk *c) { return c->kind() >= OtherKind; }
189
190 protected:
Chunk(k)191 NonSectionChunk(Kind k = OtherKind) : Chunk(k) {}
192 };
193
194 class NonSectionCodeChunk : public NonSectionChunk {
195 public:
getOutputCharacteristics()196 virtual uint32_t getOutputCharacteristics() const override {
197 return llvm::COFF::IMAGE_SCN_MEM_READ | llvm::COFF::IMAGE_SCN_MEM_EXECUTE;
198 }
199
200 protected:
NonSectionChunk(k)201 NonSectionCodeChunk(Kind k = OtherKind) : NonSectionChunk(k) {}
202 };
203
204 // MinGW specific; information about one individual location in the image
205 // that needs to be fixed up at runtime after loading. This represents
206 // one individual element in the PseudoRelocTableChunk table.
207 class RuntimePseudoReloc {
208 public:
RuntimePseudoReloc(Defined * sym,SectionChunk * target,uint32_t targetOffset,int flags)209 RuntimePseudoReloc(Defined *sym, SectionChunk *target, uint32_t targetOffset,
210 int flags)
211 : sym(sym), target(target), targetOffset(targetOffset), flags(flags) {}
212
213 Defined *sym;
214 SectionChunk *target;
215 uint32_t targetOffset;
216 // The Flags field contains the size of the relocation, in bits. No other
217 // flags are currently defined.
218 int flags;
219 };
220
221 // A chunk corresponding a section of an input file.
222 class SectionChunk : public Chunk {
223 // Identical COMDAT Folding feature accesses section internal data.
224 friend class ICF;
225
226 public:
227 class symbol_iterator : public llvm::iterator_adaptor_base<
228 symbol_iterator, const coff_relocation *,
229 std::random_access_iterator_tag, Symbol *> {
230 friend SectionChunk;
231
232 ObjFile *file;
233
symbol_iterator(ObjFile * file,const coff_relocation * i)234 symbol_iterator(ObjFile *file, const coff_relocation *i)
235 : symbol_iterator::iterator_adaptor_base(i), file(file) {}
236
237 public:
238 symbol_iterator() = default;
239
240 Symbol *operator*() const { return file->getSymbol(I->SymbolTableIndex); }
241 };
242
243 SectionChunk(ObjFile *file, const coff_section *header, Kind k = SectionKind);
classof(const Chunk * c)244 static bool classof(const Chunk *c) { return c->kind() <= SectionECKind; }
getSize()245 size_t getSize() const { return header->SizeOfRawData; }
246 ArrayRef<uint8_t> getContents() const;
247 void writeTo(uint8_t *buf) const;
248
getMachine()249 MachineTypes getMachine() const { return file->getMachineType(); }
250
251 // Defend against unsorted relocations. This may be overly conservative.
252 void sortRelocations();
253
254 // Write and relocate a portion of the section. This is intended to be called
255 // in a loop. Relocations must be sorted first.
256 void writeAndRelocateSubsection(ArrayRef<uint8_t> sec,
257 ArrayRef<uint8_t> subsec,
258 uint32_t &nextRelocIndex, uint8_t *buf) const;
259
getOutputCharacteristics()260 uint32_t getOutputCharacteristics() const {
261 return header->Characteristics & (permMask | typeMask);
262 }
getSectionName()263 StringRef getSectionName() const {
264 return StringRef(sectionNameData, sectionNameSize);
265 }
266 void getBaserels(std::vector<Baserel> *res);
267 bool isCOMDAT() const;
268 void applyRelocation(uint8_t *off, const coff_relocation &rel) const;
269 void applyRelX64(uint8_t *off, uint16_t type, OutputSection *os, uint64_t s,
270 uint64_t p, uint64_t imageBase) const;
271 void applyRelX86(uint8_t *off, uint16_t type, OutputSection *os, uint64_t s,
272 uint64_t p, uint64_t imageBase) const;
273 void applyRelARM(uint8_t *off, uint16_t type, OutputSection *os, uint64_t s,
274 uint64_t p, uint64_t imageBase) const;
275 void applyRelARM64(uint8_t *off, uint16_t type, OutputSection *os, uint64_t s,
276 uint64_t p, uint64_t imageBase) const;
277
278 void getRuntimePseudoRelocs(std::vector<RuntimePseudoReloc> &res);
279
280 // Called if the garbage collector decides to not include this chunk
281 // in a final output. It's supposed to print out a log message to stdout.
282 void printDiscardedMessage() const;
283
284 // Adds COMDAT associative sections to this COMDAT section. A chunk
285 // and its children are treated as a group by the garbage collector.
286 void addAssociative(SectionChunk *child);
287
288 StringRef getDebugName() const;
289
290 // True if this is a codeview debug info chunk. These will not be laid out in
291 // the image. Instead they will end up in the PDB, if one is requested.
isCodeView()292 bool isCodeView() const {
293 return getSectionName() == ".debug" || getSectionName().starts_with(".debug$");
294 }
295
296 // True if this is a DWARF debug info or exception handling chunk.
isDWARF()297 bool isDWARF() const {
298 return getSectionName().starts_with(".debug_") || getSectionName() == ".eh_frame";
299 }
300
301 // Allow iteration over the bodies of this chunk's relocated symbols.
symbols()302 llvm::iterator_range<symbol_iterator> symbols() const {
303 return llvm::make_range(symbol_iterator(file, relocsData),
304 symbol_iterator(file, relocsData + relocsSize));
305 }
306
getRelocs()307 ArrayRef<coff_relocation> getRelocs() const {
308 return llvm::ArrayRef(relocsData, relocsSize);
309 }
310
311 // Reloc setter used by ARM range extension thunk insertion.
setRelocs(ArrayRef<coff_relocation> newRelocs)312 void setRelocs(ArrayRef<coff_relocation> newRelocs) {
313 relocsData = newRelocs.data();
314 relocsSize = newRelocs.size();
315 assert(relocsSize == newRelocs.size() && "reloc size truncation");
316 }
317
318 // Single linked list iterator for associated comdat children.
319 class AssociatedIterator
320 : public llvm::iterator_facade_base<
321 AssociatedIterator, std::forward_iterator_tag, SectionChunk> {
322 public:
323 AssociatedIterator() = default;
AssociatedIterator(SectionChunk * head)324 AssociatedIterator(SectionChunk *head) : cur(head) {}
325 bool operator==(const AssociatedIterator &r) const { return cur == r.cur; }
326 // FIXME: Wrong const-ness, but it makes filter ranges work.
327 SectionChunk &operator*() const { return *cur; }
328 SectionChunk &operator*() { return *cur; }
329 AssociatedIterator &operator++() {
330 cur = cur->assocChildren;
331 return *this;
332 }
333
334 private:
335 SectionChunk *cur = nullptr;
336 };
337
338 // Allow iteration over the associated child chunks for this section.
children()339 llvm::iterator_range<AssociatedIterator> children() const {
340 // Associated sections do not have children. The assocChildren field is
341 // part of the parent's list of children.
342 bool isAssoc = selection == llvm::COFF::IMAGE_COMDAT_SELECT_ASSOCIATIVE;
343 return llvm::make_range(
344 AssociatedIterator(isAssoc ? nullptr : assocChildren),
345 AssociatedIterator(nullptr));
346 }
347
348 // The section ID this chunk belongs to in its Obj.
349 uint32_t getSectionNumber() const;
350
351 ArrayRef<uint8_t> consumeDebugMagic();
352
353 static ArrayRef<uint8_t> consumeDebugMagic(ArrayRef<uint8_t> data,
354 StringRef sectionName);
355
356 static SectionChunk *findByName(ArrayRef<SectionChunk *> sections,
357 StringRef name);
358
359 // The file that this chunk was created from.
360 ObjFile *file;
361
362 // Pointer to the COFF section header in the input file.
363 const coff_section *header;
364
365 // The COMDAT leader symbol if this is a COMDAT chunk.
366 DefinedRegular *sym = nullptr;
367
368 // The CRC of the contents as described in the COFF spec 4.5.5.
369 // Auxiliary Format 5: Section Definitions. Used for ICF.
370 uint32_t checksum = 0;
371
372 // Used by the garbage collector.
373 bool live;
374
375 // Whether this section needs to be kept distinct from other sections during
376 // ICF. This is set by the driver using address-significance tables.
377 bool keepUnique = false;
378
379 // The COMDAT selection if this is a COMDAT chunk.
380 llvm::COFF::COMDATType selection = (llvm::COFF::COMDATType)0;
381
382 // A pointer pointing to a replacement for this chunk.
383 // Initially it points to "this" object. If this chunk is merged
384 // with other chunk by ICF, it points to another chunk,
385 // and this chunk is considered as dead.
386 SectionChunk *repl;
387
388 private:
389 SectionChunk *assocChildren = nullptr;
390
391 // Used for ICF (Identical COMDAT Folding)
392 void replace(SectionChunk *other);
393 uint32_t eqClass[2] = {0, 0};
394
395 // Relocations for this section. Size is stored below.
396 const coff_relocation *relocsData;
397
398 // Section name string. Size is stored below.
399 const char *sectionNameData;
400
401 uint32_t relocsSize = 0;
402 uint32_t sectionNameSize = 0;
403 };
404
405 // A section chunk corresponding a section of an EC input file.
406 class SectionChunkEC final : public SectionChunk {
407 public:
classof(const Chunk * c)408 static bool classof(const Chunk *c) { return c->kind() == SectionECKind; }
409
SectionChunkEC(ObjFile * file,const coff_section * header)410 SectionChunkEC(ObjFile *file, const coff_section *header)
411 : SectionChunk(file, header, SectionECKind) {}
412 Defined *entryThunk = nullptr;
413 };
414
415 // Inline methods to implement faux-virtual dispatch for SectionChunk.
416
getSize()417 inline size_t Chunk::getSize() const {
418 if (isa<SectionChunk>(this))
419 return static_cast<const SectionChunk *>(this)->getSize();
420 return static_cast<const NonSectionChunk *>(this)->getSize();
421 }
422
getOutputCharacteristics()423 inline uint32_t Chunk::getOutputCharacteristics() const {
424 if (isa<SectionChunk>(this))
425 return static_cast<const SectionChunk *>(this)->getOutputCharacteristics();
426 return static_cast<const NonSectionChunk *>(this)->getOutputCharacteristics();
427 }
428
writeTo(uint8_t * buf)429 inline void Chunk::writeTo(uint8_t *buf) const {
430 if (isa<SectionChunk>(this))
431 static_cast<const SectionChunk *>(this)->writeTo(buf);
432 else
433 static_cast<const NonSectionChunk *>(this)->writeTo(buf);
434 }
435
getSectionName()436 inline StringRef Chunk::getSectionName() const {
437 if (isa<SectionChunk>(this))
438 return static_cast<const SectionChunk *>(this)->getSectionName();
439 return static_cast<const NonSectionChunk *>(this)->getSectionName();
440 }
441
getBaserels(std::vector<Baserel> * res)442 inline void Chunk::getBaserels(std::vector<Baserel> *res) {
443 if (isa<SectionChunk>(this))
444 static_cast<SectionChunk *>(this)->getBaserels(res);
445 else
446 static_cast<NonSectionChunk *>(this)->getBaserels(res);
447 }
448
getDebugName()449 inline StringRef Chunk::getDebugName() const {
450 if (isa<SectionChunk>(this))
451 return static_cast<const SectionChunk *>(this)->getDebugName();
452 return static_cast<const NonSectionChunk *>(this)->getDebugName();
453 }
454
getMachine()455 inline MachineTypes Chunk::getMachine() const {
456 if (isa<SectionChunk>(this))
457 return static_cast<const SectionChunk *>(this)->getMachine();
458 return static_cast<const NonSectionChunk *>(this)->getMachine();
459 }
460
getArch()461 inline llvm::Triple::ArchType Chunk::getArch() const {
462 return llvm::getMachineArchType(getMachine());
463 }
464
getArm64ECRangeType()465 inline std::optional<chpe_range_type> Chunk::getArm64ECRangeType() const {
466 // Data sections don't need codemap entries.
467 if (!(getOutputCharacteristics() & llvm::COFF::IMAGE_SCN_MEM_EXECUTE))
468 return std::nullopt;
469
470 switch (getMachine()) {
471 case AMD64:
472 return chpe_range_type::Amd64;
473 case ARM64EC:
474 return chpe_range_type::Arm64EC;
475 default:
476 return chpe_range_type::Arm64;
477 }
478 }
479
480 // This class is used to implement an lld-specific feature (not implemented in
481 // MSVC) that minimizes the output size by finding string literals sharing tail
482 // parts and merging them.
483 //
484 // If string tail merging is enabled and a section is identified as containing a
485 // string literal, it is added to a MergeChunk with an appropriate alignment.
486 // The MergeChunk then tail merges the strings using the StringTableBuilder
487 // class and assigns RVAs and section offsets to each of the member chunks based
488 // on the offsets assigned by the StringTableBuilder.
489 class MergeChunk : public NonSectionChunk {
490 public:
491 MergeChunk(uint32_t alignment);
492 static void addSection(COFFLinkerContext &ctx, SectionChunk *c);
493 void finalizeContents();
494 void assignSubsectionRVAs();
495
496 uint32_t getOutputCharacteristics() const override;
getSectionName()497 StringRef getSectionName() const override { return ".rdata"; }
498 size_t getSize() const override;
499 void writeTo(uint8_t *buf) const override;
500
501 std::vector<SectionChunk *> sections;
502
503 private:
504 llvm::StringTableBuilder builder;
505 bool finalized = false;
506 };
507
508 // A chunk for common symbols. Common chunks don't have actual data.
509 class CommonChunk : public NonSectionChunk {
510 public:
511 CommonChunk(const COFFSymbolRef sym);
getSize()512 size_t getSize() const override { return sym.getValue(); }
513 uint32_t getOutputCharacteristics() const override;
getSectionName()514 StringRef getSectionName() const override { return ".bss"; }
515
516 private:
517 const COFFSymbolRef sym;
518 };
519
520 // A chunk for linker-created strings.
521 class StringChunk : public NonSectionChunk {
522 public:
StringChunk(StringRef s)523 explicit StringChunk(StringRef s) : str(s) {}
getSize()524 size_t getSize() const override { return str.size() + 1; }
525 void writeTo(uint8_t *buf) const override;
526
527 private:
528 StringRef str;
529 };
530
531 static const uint8_t importThunkX86[] = {
532 0xff, 0x25, 0x00, 0x00, 0x00, 0x00, // JMP *0x0
533 };
534
535 static const uint8_t importThunkARM[] = {
536 0x40, 0xf2, 0x00, 0x0c, // mov.w ip, #0
537 0xc0, 0xf2, 0x00, 0x0c, // mov.t ip, #0
538 0xdc, 0xf8, 0x00, 0xf0, // ldr.w pc, [ip]
539 };
540
541 static const uint8_t importThunkARM64[] = {
542 0x10, 0x00, 0x00, 0x90, // adrp x16, #0
543 0x10, 0x02, 0x40, 0xf9, // ldr x16, [x16]
544 0x00, 0x02, 0x1f, 0xd6, // br x16
545 };
546
547 // Windows-specific.
548 // A chunk for DLL import jump table entry. In a final output, its
549 // contents will be a JMP instruction to some __imp_ symbol.
550 class ImportThunkChunk : public NonSectionCodeChunk {
551 public:
ImportThunkChunk(COFFLinkerContext & ctx,Defined * s)552 ImportThunkChunk(COFFLinkerContext &ctx, Defined *s)
553 : NonSectionCodeChunk(ImportThunkKind), impSymbol(s), ctx(ctx) {}
classof(const Chunk * c)554 static bool classof(const Chunk *c) { return c->kind() == ImportThunkKind; }
555
556 protected:
557 Defined *impSymbol;
558 COFFLinkerContext &ctx;
559 };
560
561 class ImportThunkChunkX64 : public ImportThunkChunk {
562 public:
563 explicit ImportThunkChunkX64(COFFLinkerContext &ctx, Defined *s);
getSize()564 size_t getSize() const override { return sizeof(importThunkX86); }
565 void writeTo(uint8_t *buf) const override;
getMachine()566 MachineTypes getMachine() const override { return AMD64; }
567 };
568
569 class ImportThunkChunkX86 : public ImportThunkChunk {
570 public:
ImportThunkChunkX86(COFFLinkerContext & ctx,Defined * s)571 explicit ImportThunkChunkX86(COFFLinkerContext &ctx, Defined *s)
572 : ImportThunkChunk(ctx, s) {}
getSize()573 size_t getSize() const override { return sizeof(importThunkX86); }
574 void getBaserels(std::vector<Baserel> *res) override;
575 void writeTo(uint8_t *buf) const override;
getMachine()576 MachineTypes getMachine() const override { return I386; }
577 };
578
579 class ImportThunkChunkARM : public ImportThunkChunk {
580 public:
ImportThunkChunkARM(COFFLinkerContext & ctx,Defined * s)581 explicit ImportThunkChunkARM(COFFLinkerContext &ctx, Defined *s)
582 : ImportThunkChunk(ctx, s) {
583 setAlignment(2);
584 }
getSize()585 size_t getSize() const override { return sizeof(importThunkARM); }
586 void getBaserels(std::vector<Baserel> *res) override;
587 void writeTo(uint8_t *buf) const override;
getMachine()588 MachineTypes getMachine() const override { return ARMNT; }
589 };
590
591 class ImportThunkChunkARM64 : public ImportThunkChunk {
592 public:
ImportThunkChunkARM64(COFFLinkerContext & ctx,Defined * s)593 explicit ImportThunkChunkARM64(COFFLinkerContext &ctx, Defined *s)
594 : ImportThunkChunk(ctx, s) {
595 setAlignment(4);
596 }
getSize()597 size_t getSize() const override { return sizeof(importThunkARM64); }
598 void writeTo(uint8_t *buf) const override;
getMachine()599 MachineTypes getMachine() const override { return ARM64; }
600 };
601
602 class RangeExtensionThunkARM : public NonSectionCodeChunk {
603 public:
RangeExtensionThunkARM(COFFLinkerContext & ctx,Defined * t)604 explicit RangeExtensionThunkARM(COFFLinkerContext &ctx, Defined *t)
605 : target(t), ctx(ctx) {
606 setAlignment(2);
607 }
608 size_t getSize() const override;
609 void writeTo(uint8_t *buf) const override;
getMachine()610 MachineTypes getMachine() const override { return ARMNT; }
611
612 Defined *target;
613
614 private:
615 COFFLinkerContext &ctx;
616 };
617
618 class RangeExtensionThunkARM64 : public NonSectionCodeChunk {
619 public:
RangeExtensionThunkARM64(COFFLinkerContext & ctx,Defined * t)620 explicit RangeExtensionThunkARM64(COFFLinkerContext &ctx, Defined *t)
621 : target(t), ctx(ctx) {
622 setAlignment(4);
623 }
624 size_t getSize() const override;
625 void writeTo(uint8_t *buf) const override;
getMachine()626 MachineTypes getMachine() const override { return ARM64; }
627
628 Defined *target;
629
630 private:
631 COFFLinkerContext &ctx;
632 };
633
634 // Windows-specific.
635 // See comments for DefinedLocalImport class.
636 class LocalImportChunk : public NonSectionChunk {
637 public:
638 explicit LocalImportChunk(COFFLinkerContext &ctx, Defined *s);
639 size_t getSize() const override;
640 void getBaserels(std::vector<Baserel> *res) override;
641 void writeTo(uint8_t *buf) const override;
642
643 private:
644 Defined *sym;
645 COFFLinkerContext &ctx;
646 };
647
648 // Duplicate RVAs are not allowed in RVA tables, so unique symbols by chunk and
649 // offset into the chunk. Order does not matter as the RVA table will be sorted
650 // later.
651 struct ChunkAndOffset {
652 Chunk *inputChunk;
653 uint32_t offset;
654
655 struct DenseMapInfo {
getEmptyKeyChunkAndOffset::DenseMapInfo656 static ChunkAndOffset getEmptyKey() {
657 return {llvm::DenseMapInfo<Chunk *>::getEmptyKey(), 0};
658 }
getTombstoneKeyChunkAndOffset::DenseMapInfo659 static ChunkAndOffset getTombstoneKey() {
660 return {llvm::DenseMapInfo<Chunk *>::getTombstoneKey(), 0};
661 }
getHashValueChunkAndOffset::DenseMapInfo662 static unsigned getHashValue(const ChunkAndOffset &co) {
663 return llvm::DenseMapInfo<std::pair<Chunk *, uint32_t>>::getHashValue(
664 {co.inputChunk, co.offset});
665 }
isEqualChunkAndOffset::DenseMapInfo666 static bool isEqual(const ChunkAndOffset &lhs, const ChunkAndOffset &rhs) {
667 return lhs.inputChunk == rhs.inputChunk && lhs.offset == rhs.offset;
668 }
669 };
670 };
671
672 using SymbolRVASet = llvm::DenseSet<ChunkAndOffset>;
673
674 // Table which contains symbol RVAs. Used for /safeseh and /guard:cf.
675 class RVATableChunk : public NonSectionChunk {
676 public:
RVATableChunk(SymbolRVASet s)677 explicit RVATableChunk(SymbolRVASet s) : syms(std::move(s)) {}
getSize()678 size_t getSize() const override { return syms.size() * 4; }
679 void writeTo(uint8_t *buf) const override;
680
681 private:
682 SymbolRVASet syms;
683 };
684
685 // Table which contains symbol RVAs with flags. Used for /guard:ehcont.
686 class RVAFlagTableChunk : public NonSectionChunk {
687 public:
RVAFlagTableChunk(SymbolRVASet s)688 explicit RVAFlagTableChunk(SymbolRVASet s) : syms(std::move(s)) {}
getSize()689 size_t getSize() const override { return syms.size() * 5; }
690 void writeTo(uint8_t *buf) const override;
691
692 private:
693 SymbolRVASet syms;
694 };
695
696 // Windows-specific.
697 // This class represents a block in .reloc section.
698 // See the PE/COFF spec 5.6 for details.
699 class BaserelChunk : public NonSectionChunk {
700 public:
701 BaserelChunk(uint32_t page, Baserel *begin, Baserel *end);
getSize()702 size_t getSize() const override { return data.size(); }
703 void writeTo(uint8_t *buf) const override;
704
705 private:
706 std::vector<uint8_t> data;
707 };
708
709 class Baserel {
710 public:
Baserel(uint32_t v,uint8_t ty)711 Baserel(uint32_t v, uint8_t ty) : rva(v), type(ty) {}
Baserel(uint32_t v,llvm::COFF::MachineTypes machine)712 explicit Baserel(uint32_t v, llvm::COFF::MachineTypes machine)
713 : Baserel(v, getDefaultType(machine)) {}
714 uint8_t getDefaultType(llvm::COFF::MachineTypes machine);
715
716 uint32_t rva;
717 uint8_t type;
718 };
719
720 // This is a placeholder Chunk, to allow attaching a DefinedSynthetic to a
721 // specific place in a section, without any data. This is used for the MinGW
722 // specific symbol __RUNTIME_PSEUDO_RELOC_LIST_END__, even though the concept
723 // of an empty chunk isn't MinGW specific.
724 class EmptyChunk : public NonSectionChunk {
725 public:
EmptyChunk()726 EmptyChunk() {}
getSize()727 size_t getSize() const override { return 0; }
writeTo(uint8_t * buf)728 void writeTo(uint8_t *buf) const override {}
729 };
730
731 class ECCodeMapEntry {
732 public:
ECCodeMapEntry(Chunk * first,Chunk * last,chpe_range_type type)733 ECCodeMapEntry(Chunk *first, Chunk *last, chpe_range_type type)
734 : first(first), last(last), type(type) {}
735 Chunk *first;
736 Chunk *last;
737 chpe_range_type type;
738 };
739
740 // This is a chunk containing CHPE code map on EC targets. It's a table
741 // of address ranges and their types.
742 class ECCodeMapChunk : public NonSectionChunk {
743 public:
ECCodeMapChunk(std::vector<ECCodeMapEntry> & map)744 ECCodeMapChunk(std::vector<ECCodeMapEntry> &map) : map(map) {}
745 size_t getSize() const override;
746 void writeTo(uint8_t *buf) const override;
747
748 private:
749 std::vector<ECCodeMapEntry> ↦
750 };
751
752 // MinGW specific, for the "automatic import of variables from DLLs" feature.
753 // This provides the table of runtime pseudo relocations, for variable
754 // references that turned out to need to be imported from a DLL even though
755 // the reference didn't use the dllimport attribute. The MinGW runtime will
756 // process this table after loading, before handling control over to user
757 // code.
758 class PseudoRelocTableChunk : public NonSectionChunk {
759 public:
PseudoRelocTableChunk(std::vector<RuntimePseudoReloc> & relocs)760 PseudoRelocTableChunk(std::vector<RuntimePseudoReloc> &relocs)
761 : relocs(std::move(relocs)) {
762 setAlignment(4);
763 }
764 size_t getSize() const override;
765 void writeTo(uint8_t *buf) const override;
766
767 private:
768 std::vector<RuntimePseudoReloc> relocs;
769 };
770
771 // MinGW specific. A Chunk that contains one pointer-sized absolute value.
772 class AbsolutePointerChunk : public NonSectionChunk {
773 public:
AbsolutePointerChunk(COFFLinkerContext & ctx,uint64_t value)774 AbsolutePointerChunk(COFFLinkerContext &ctx, uint64_t value)
775 : value(value), ctx(ctx) {
776 setAlignment(getSize());
777 }
778 size_t getSize() const override;
779 void writeTo(uint8_t *buf) const override;
780
781 private:
782 uint64_t value;
783 COFFLinkerContext &ctx;
784 };
785
786 // Return true if this file has the hotpatch flag set to true in the S_COMPILE3
787 // record in codeview debug info. Also returns true for some thunks synthesized
788 // by the linker.
isHotPatchable()789 inline bool Chunk::isHotPatchable() const {
790 if (auto *sc = dyn_cast<SectionChunk>(this))
791 return sc->file->hotPatchable;
792 else if (isa<ImportThunkChunk>(this))
793 return true;
794 return false;
795 }
796
getEntryThunk()797 inline Defined *Chunk::getEntryThunk() const {
798 if (auto *c = dyn_cast<const SectionChunkEC>(this))
799 return c->entryThunk;
800 return nullptr;
801 }
802
setEntryThunk(Defined * entryThunk)803 inline void Chunk::setEntryThunk(Defined *entryThunk) {
804 if (auto c = dyn_cast<SectionChunkEC>(this))
805 c->entryThunk = entryThunk;
806 }
807
808 void applyMOV32T(uint8_t *off, uint32_t v);
809 void applyBranch24T(uint8_t *off, int32_t v);
810
811 void applyArm64Addr(uint8_t *off, uint64_t s, uint64_t p, int shift);
812 void applyArm64Imm(uint8_t *off, uint64_t imm, uint32_t rangeLimit);
813 void applyArm64Branch26(uint8_t *off, int64_t v);
814
815 // Convenience class for initializing a coff_section with specific flags.
816 class FakeSection {
817 public:
FakeSection(int c)818 FakeSection(int c) { section.Characteristics = c; }
819
820 coff_section section;
821 };
822
823 // Convenience class for initializing a SectionChunk with specific flags.
824 class FakeSectionChunk {
825 public:
FakeSectionChunk(const coff_section * section)826 FakeSectionChunk(const coff_section *section) : chunk(nullptr, section) {
827 // Comdats from LTO files can't be fully treated as regular comdats
828 // at this point; we don't know what size or contents they are going to
829 // have, so we can't do proper checking of such aspects of them.
830 chunk.selection = llvm::COFF::IMAGE_COMDAT_SELECT_ANY;
831 }
832
833 SectionChunk chunk;
834 };
835
836 } // namespace lld::coff
837
838 namespace llvm {
839 template <>
840 struct DenseMapInfo<lld::coff::ChunkAndOffset>
841 : lld::coff::ChunkAndOffset::DenseMapInfo {};
842 }
843
844 #endif
845