xref: /freebsd/contrib/llvm-project/lld/COFF/Chunks.h (revision 5f757f3f)
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 <utility>
22 #include <vector>
23 
24 namespace lld::coff {
25 
26 using llvm::COFF::ImportDirectoryTableEntry;
27 using llvm::object::chpe_range_type;
28 using llvm::object::coff_relocation;
29 using llvm::object::coff_section;
30 using llvm::object::COFFSymbolRef;
31 using llvm::object::SectionRef;
32 
33 class Baserel;
34 class Defined;
35 class DefinedImportData;
36 class DefinedRegular;
37 class ObjFile;
38 class OutputSection;
39 class RuntimePseudoReloc;
40 class Symbol;
41 
42 // Mask for permissions (discardable, writable, readable, executable, etc).
43 const uint32_t permMask = 0xFE000000;
44 
45 // Mask for section types (code, data, bss).
46 const uint32_t typeMask = 0x000000E0;
47 
48 // The log base 2 of the largest section alignment, which is log2(8192), or 13.
49 enum : unsigned { Log2MaxSectionAlignment = 13 };
50 
51 // A Chunk represents a chunk of data that will occupy space in the
52 // output (if the resolver chose that). It may or may not be backed by
53 // a section of an input file. It could be linker-created data, or
54 // doesn't even have actual data (if common or bss).
55 class Chunk {
56 public:
57   enum Kind : uint8_t { SectionKind, OtherKind, ImportThunkKind };
kind()58   Kind kind() const { return chunkKind; }
59 
60   // Returns the size of this chunk (even if this is a common or BSS.)
61   size_t getSize() const;
62 
63   // Returns chunk alignment in power of two form. Value values are powers of
64   // two from 1 to 8192.
getAlignment()65   uint32_t getAlignment() const { return 1U << p2Align; }
66 
67   // Update the chunk section alignment measured in bytes. Internally alignment
68   // is stored in log2.
setAlignment(uint32_t align)69   void setAlignment(uint32_t align) {
70     // Treat zero byte alignment as 1 byte alignment.
71     align = align ? align : 1;
72     assert(llvm::isPowerOf2_32(align) && "alignment is not a power of 2");
73     p2Align = llvm::Log2_32(align);
74     assert(p2Align <= Log2MaxSectionAlignment &&
75            "impossible requested alignment");
76   }
77 
78   // Write this chunk to a mmap'ed file, assuming Buf is pointing to
79   // beginning of the file. Because this function may use RVA values
80   // of other chunks for relocations, you need to set them properly
81   // before calling this function.
82   void writeTo(uint8_t *buf) const;
83 
84   // The writer sets and uses the addresses. In practice, PE images cannot be
85   // larger than 2GB. Chunks are always laid as part of the image, so Chunk RVAs
86   // can be stored with 32 bits.
getRVA()87   uint32_t getRVA() const { return rva; }
setRVA(uint64_t v)88   void setRVA(uint64_t v) {
89     // This may truncate. The writer checks for overflow later.
90     rva = (uint32_t)v;
91   }
92 
93   // Returns readable/writable/executable bits.
94   uint32_t getOutputCharacteristics() const;
95 
96   // Returns the section name if this is a section chunk.
97   // It is illegal to call this function on non-section chunks.
98   StringRef getSectionName() const;
99 
100   // An output section has pointers to chunks in the section, and each
101   // chunk has a back pointer to an output section.
setOutputSectionIdx(uint16_t o)102   void setOutputSectionIdx(uint16_t o) { osidx = o; }
getOutputSectionIdx()103   uint16_t getOutputSectionIdx() const { return osidx; }
104 
105   // Windows-specific.
106   // Collect all locations that contain absolute addresses for base relocations.
107   void getBaserels(std::vector<Baserel> *res);
108 
109   // Returns a human-readable name of this chunk. Chunks are unnamed chunks of
110   // bytes, so this is used only for logging or debugging.
111   StringRef getDebugName() const;
112 
113   // Return true if this file has the hotpatch flag set to true in the
114   // S_COMPILE3 record in codeview debug info. Also returns true for some thunks
115   // synthesized by the linker.
116   bool isHotPatchable() const;
117 
118   MachineTypes getMachine() const;
119   std::optional<chpe_range_type> getArm64ECRangeType() const;
120 
121 protected:
chunkKind(k)122   Chunk(Kind k = OtherKind) : chunkKind(k), hasData(true), p2Align(0) {}
123 
124   const Kind chunkKind;
125 
126 public:
127   // Returns true if this has non-zero data. BSS chunks return
128   // false. If false is returned, the space occupied by this chunk
129   // will be filled with zeros. Corresponds to the
130   // IMAGE_SCN_CNT_UNINITIALIZED_DATA section characteristic bit.
131   uint8_t hasData : 1;
132 
133 public:
134   // The alignment of this chunk, stored in log2 form. The writer uses the
135   // value.
136   uint8_t p2Align : 7;
137 
138   // The output section index for this chunk. The first valid section number is
139   // one.
140   uint16_t osidx = 0;
141 
142   // The RVA of this chunk in the output. The writer sets a value.
143   uint32_t rva = 0;
144 };
145 
146 class NonSectionChunk : public Chunk {
147 public:
148   virtual ~NonSectionChunk() = default;
149 
150   // Returns the size of this chunk (even if this is a common or BSS.)
151   virtual size_t getSize() const = 0;
152 
getOutputCharacteristics()153   virtual uint32_t getOutputCharacteristics() const { return 0; }
154 
155   // Write this chunk to a mmap'ed file, assuming Buf is pointing to
156   // beginning of the file. Because this function may use RVA values
157   // of other chunks for relocations, you need to set them properly
158   // before calling this function.
writeTo(uint8_t * buf)159   virtual void writeTo(uint8_t *buf) const {}
160 
161   // Returns the section name if this is a section chunk.
162   // It is illegal to call this function on non-section chunks.
getSectionName()163   virtual StringRef getSectionName() const {
164     llvm_unreachable("unimplemented getSectionName");
165   }
166 
167   // Windows-specific.
168   // Collect all locations that contain absolute addresses for base relocations.
getBaserels(std::vector<Baserel> * res)169   virtual void getBaserels(std::vector<Baserel> *res) {}
170 
getMachine()171   virtual MachineTypes getMachine() const { return IMAGE_FILE_MACHINE_UNKNOWN; }
172 
173   // Returns a human-readable name of this chunk. Chunks are unnamed chunks of
174   // bytes, so this is used only for logging or debugging.
getDebugName()175   virtual StringRef getDebugName() const { return ""; }
176 
classof(const Chunk * c)177   static bool classof(const Chunk *c) { return c->kind() != SectionKind; }
178 
179 protected:
Chunk(k)180   NonSectionChunk(Kind k = OtherKind) : Chunk(k) {}
181 };
182 
183 class NonSectionCodeChunk : public NonSectionChunk {
184 public:
getOutputCharacteristics()185   virtual uint32_t getOutputCharacteristics() const override {
186     return llvm::COFF::IMAGE_SCN_MEM_READ | llvm::COFF::IMAGE_SCN_MEM_EXECUTE;
187   }
188 
189 protected:
NonSectionChunk(k)190   NonSectionCodeChunk(Kind k = OtherKind) : NonSectionChunk(k) {}
191 };
192 
193 // MinGW specific; information about one individual location in the image
194 // that needs to be fixed up at runtime after loading. This represents
195 // one individual element in the PseudoRelocTableChunk table.
196 class RuntimePseudoReloc {
197 public:
RuntimePseudoReloc(Defined * sym,SectionChunk * target,uint32_t targetOffset,int flags)198   RuntimePseudoReloc(Defined *sym, SectionChunk *target, uint32_t targetOffset,
199                      int flags)
200       : sym(sym), target(target), targetOffset(targetOffset), flags(flags) {}
201 
202   Defined *sym;
203   SectionChunk *target;
204   uint32_t targetOffset;
205   // The Flags field contains the size of the relocation, in bits. No other
206   // flags are currently defined.
207   int flags;
208 };
209 
210 // A chunk corresponding a section of an input file.
211 class SectionChunk final : public Chunk {
212   // Identical COMDAT Folding feature accesses section internal data.
213   friend class ICF;
214 
215 public:
216   class symbol_iterator : public llvm::iterator_adaptor_base<
217                               symbol_iterator, const coff_relocation *,
218                               std::random_access_iterator_tag, Symbol *> {
219     friend SectionChunk;
220 
221     ObjFile *file;
222 
symbol_iterator(ObjFile * file,const coff_relocation * i)223     symbol_iterator(ObjFile *file, const coff_relocation *i)
224         : symbol_iterator::iterator_adaptor_base(i), file(file) {}
225 
226   public:
227     symbol_iterator() = default;
228 
229     Symbol *operator*() const { return file->getSymbol(I->SymbolTableIndex); }
230   };
231 
232   SectionChunk(ObjFile *file, const coff_section *header);
classof(const Chunk * c)233   static bool classof(const Chunk *c) { return c->kind() == SectionKind; }
getSize()234   size_t getSize() const { return header->SizeOfRawData; }
235   ArrayRef<uint8_t> getContents() const;
236   void writeTo(uint8_t *buf) const;
237 
getMachine()238   MachineTypes getMachine() const { return file->getMachineType(); }
239 
240   // Defend against unsorted relocations. This may be overly conservative.
241   void sortRelocations();
242 
243   // Write and relocate a portion of the section. This is intended to be called
244   // in a loop. Relocations must be sorted first.
245   void writeAndRelocateSubsection(ArrayRef<uint8_t> sec,
246                                   ArrayRef<uint8_t> subsec,
247                                   uint32_t &nextRelocIndex, uint8_t *buf) const;
248 
getOutputCharacteristics()249   uint32_t getOutputCharacteristics() const {
250     return header->Characteristics & (permMask | typeMask);
251   }
getSectionName()252   StringRef getSectionName() const {
253     return StringRef(sectionNameData, sectionNameSize);
254   }
255   void getBaserels(std::vector<Baserel> *res);
256   bool isCOMDAT() const;
257   void applyRelocation(uint8_t *off, const coff_relocation &rel) const;
258   void applyRelX64(uint8_t *off, uint16_t type, OutputSection *os, uint64_t s,
259                    uint64_t p, uint64_t imageBase) const;
260   void applyRelX86(uint8_t *off, uint16_t type, OutputSection *os, uint64_t s,
261                    uint64_t p, uint64_t imageBase) const;
262   void applyRelARM(uint8_t *off, uint16_t type, OutputSection *os, uint64_t s,
263                    uint64_t p, uint64_t imageBase) const;
264   void applyRelARM64(uint8_t *off, uint16_t type, OutputSection *os, uint64_t s,
265                      uint64_t p, uint64_t imageBase) const;
266 
267   void getRuntimePseudoRelocs(std::vector<RuntimePseudoReloc> &res);
268 
269   // Called if the garbage collector decides to not include this chunk
270   // in a final output. It's supposed to print out a log message to stdout.
271   void printDiscardedMessage() const;
272 
273   // Adds COMDAT associative sections to this COMDAT section. A chunk
274   // and its children are treated as a group by the garbage collector.
275   void addAssociative(SectionChunk *child);
276 
277   StringRef getDebugName() const;
278 
279   // True if this is a codeview debug info chunk. These will not be laid out in
280   // the image. Instead they will end up in the PDB, if one is requested.
isCodeView()281   bool isCodeView() const {
282     return getSectionName() == ".debug" || getSectionName().starts_with(".debug$");
283   }
284 
285   // True if this is a DWARF debug info or exception handling chunk.
isDWARF()286   bool isDWARF() const {
287     return getSectionName().starts_with(".debug_") || getSectionName() == ".eh_frame";
288   }
289 
290   // Allow iteration over the bodies of this chunk's relocated symbols.
symbols()291   llvm::iterator_range<symbol_iterator> symbols() const {
292     return llvm::make_range(symbol_iterator(file, relocsData),
293                             symbol_iterator(file, relocsData + relocsSize));
294   }
295 
getRelocs()296   ArrayRef<coff_relocation> getRelocs() const {
297     return llvm::ArrayRef(relocsData, relocsSize);
298   }
299 
300   // Reloc setter used by ARM range extension thunk insertion.
setRelocs(ArrayRef<coff_relocation> newRelocs)301   void setRelocs(ArrayRef<coff_relocation> newRelocs) {
302     relocsData = newRelocs.data();
303     relocsSize = newRelocs.size();
304     assert(relocsSize == newRelocs.size() && "reloc size truncation");
305   }
306 
307   // Single linked list iterator for associated comdat children.
308   class AssociatedIterator
309       : public llvm::iterator_facade_base<
310             AssociatedIterator, std::forward_iterator_tag, SectionChunk> {
311   public:
312     AssociatedIterator() = default;
AssociatedIterator(SectionChunk * head)313     AssociatedIterator(SectionChunk *head) : cur(head) {}
314     bool operator==(const AssociatedIterator &r) const { return cur == r.cur; }
315     // FIXME: Wrong const-ness, but it makes filter ranges work.
316     SectionChunk &operator*() const { return *cur; }
317     SectionChunk &operator*() { return *cur; }
318     AssociatedIterator &operator++() {
319       cur = cur->assocChildren;
320       return *this;
321     }
322 
323   private:
324     SectionChunk *cur = nullptr;
325   };
326 
327   // Allow iteration over the associated child chunks for this section.
children()328   llvm::iterator_range<AssociatedIterator> children() const {
329     // Associated sections do not have children. The assocChildren field is
330     // part of the parent's list of children.
331     bool isAssoc = selection == llvm::COFF::IMAGE_COMDAT_SELECT_ASSOCIATIVE;
332     return llvm::make_range(
333         AssociatedIterator(isAssoc ? nullptr : assocChildren),
334         AssociatedIterator(nullptr));
335   }
336 
337   // The section ID this chunk belongs to in its Obj.
338   uint32_t getSectionNumber() const;
339 
340   ArrayRef<uint8_t> consumeDebugMagic();
341 
342   static ArrayRef<uint8_t> consumeDebugMagic(ArrayRef<uint8_t> data,
343                                              StringRef sectionName);
344 
345   static SectionChunk *findByName(ArrayRef<SectionChunk *> sections,
346                                   StringRef name);
347 
348   // The file that this chunk was created from.
349   ObjFile *file;
350 
351   // Pointer to the COFF section header in the input file.
352   const coff_section *header;
353 
354   // The COMDAT leader symbol if this is a COMDAT chunk.
355   DefinedRegular *sym = nullptr;
356 
357   // The CRC of the contents as described in the COFF spec 4.5.5.
358   // Auxiliary Format 5: Section Definitions. Used for ICF.
359   uint32_t checksum = 0;
360 
361   // Used by the garbage collector.
362   bool live;
363 
364   // Whether this section needs to be kept distinct from other sections during
365   // ICF. This is set by the driver using address-significance tables.
366   bool keepUnique = false;
367 
368   // The COMDAT selection if this is a COMDAT chunk.
369   llvm::COFF::COMDATType selection = (llvm::COFF::COMDATType)0;
370 
371   // A pointer pointing to a replacement for this chunk.
372   // Initially it points to "this" object. If this chunk is merged
373   // with other chunk by ICF, it points to another chunk,
374   // and this chunk is considered as dead.
375   SectionChunk *repl;
376 
377 private:
378   SectionChunk *assocChildren = nullptr;
379 
380   // Used for ICF (Identical COMDAT Folding)
381   void replace(SectionChunk *other);
382   uint32_t eqClass[2] = {0, 0};
383 
384   // Relocations for this section. Size is stored below.
385   const coff_relocation *relocsData;
386 
387   // Section name string. Size is stored below.
388   const char *sectionNameData;
389 
390   uint32_t relocsSize = 0;
391   uint32_t sectionNameSize = 0;
392 };
393 
394 // Inline methods to implement faux-virtual dispatch for SectionChunk.
395 
getSize()396 inline size_t Chunk::getSize() const {
397   if (isa<SectionChunk>(this))
398     return static_cast<const SectionChunk *>(this)->getSize();
399   return static_cast<const NonSectionChunk *>(this)->getSize();
400 }
401 
getOutputCharacteristics()402 inline uint32_t Chunk::getOutputCharacteristics() const {
403   if (isa<SectionChunk>(this))
404     return static_cast<const SectionChunk *>(this)->getOutputCharacteristics();
405   return static_cast<const NonSectionChunk *>(this)->getOutputCharacteristics();
406 }
407 
writeTo(uint8_t * buf)408 inline void Chunk::writeTo(uint8_t *buf) const {
409   if (isa<SectionChunk>(this))
410     static_cast<const SectionChunk *>(this)->writeTo(buf);
411   else
412     static_cast<const NonSectionChunk *>(this)->writeTo(buf);
413 }
414 
getSectionName()415 inline StringRef Chunk::getSectionName() const {
416   if (isa<SectionChunk>(this))
417     return static_cast<const SectionChunk *>(this)->getSectionName();
418   return static_cast<const NonSectionChunk *>(this)->getSectionName();
419 }
420 
getBaserels(std::vector<Baserel> * res)421 inline void Chunk::getBaserels(std::vector<Baserel> *res) {
422   if (isa<SectionChunk>(this))
423     static_cast<SectionChunk *>(this)->getBaserels(res);
424   else
425     static_cast<NonSectionChunk *>(this)->getBaserels(res);
426 }
427 
getDebugName()428 inline StringRef Chunk::getDebugName() const {
429   if (isa<SectionChunk>(this))
430     return static_cast<const SectionChunk *>(this)->getDebugName();
431   return static_cast<const NonSectionChunk *>(this)->getDebugName();
432 }
433 
getMachine()434 inline MachineTypes Chunk::getMachine() const {
435   if (isa<SectionChunk>(this))
436     return static_cast<const SectionChunk *>(this)->getMachine();
437   return static_cast<const NonSectionChunk *>(this)->getMachine();
438 }
439 
getArm64ECRangeType()440 inline std::optional<chpe_range_type> Chunk::getArm64ECRangeType() const {
441   // Data sections don't need codemap entries.
442   if (!(getOutputCharacteristics() & llvm::COFF::IMAGE_SCN_MEM_EXECUTE))
443     return std::nullopt;
444 
445   switch (getMachine()) {
446   case AMD64:
447     return chpe_range_type::Amd64;
448   case ARM64EC:
449     return chpe_range_type::Arm64EC;
450   default:
451     return chpe_range_type::Arm64;
452   }
453 }
454 
455 // This class is used to implement an lld-specific feature (not implemented in
456 // MSVC) that minimizes the output size by finding string literals sharing tail
457 // parts and merging them.
458 //
459 // If string tail merging is enabled and a section is identified as containing a
460 // string literal, it is added to a MergeChunk with an appropriate alignment.
461 // The MergeChunk then tail merges the strings using the StringTableBuilder
462 // class and assigns RVAs and section offsets to each of the member chunks based
463 // on the offsets assigned by the StringTableBuilder.
464 class MergeChunk : public NonSectionChunk {
465 public:
466   MergeChunk(uint32_t alignment);
467   static void addSection(COFFLinkerContext &ctx, SectionChunk *c);
468   void finalizeContents();
469   void assignSubsectionRVAs();
470 
471   uint32_t getOutputCharacteristics() const override;
getSectionName()472   StringRef getSectionName() const override { return ".rdata"; }
473   size_t getSize() const override;
474   void writeTo(uint8_t *buf) const override;
475 
476   std::vector<SectionChunk *> sections;
477 
478 private:
479   llvm::StringTableBuilder builder;
480   bool finalized = false;
481 };
482 
483 // A chunk for common symbols. Common chunks don't have actual data.
484 class CommonChunk : public NonSectionChunk {
485 public:
486   CommonChunk(const COFFSymbolRef sym);
getSize()487   size_t getSize() const override { return sym.getValue(); }
488   uint32_t getOutputCharacteristics() const override;
getSectionName()489   StringRef getSectionName() const override { return ".bss"; }
490 
491 private:
492   const COFFSymbolRef sym;
493 };
494 
495 // A chunk for linker-created strings.
496 class StringChunk : public NonSectionChunk {
497 public:
StringChunk(StringRef s)498   explicit StringChunk(StringRef s) : str(s) {}
getSize()499   size_t getSize() const override { return str.size() + 1; }
500   void writeTo(uint8_t *buf) const override;
501 
502 private:
503   StringRef str;
504 };
505 
506 static const uint8_t importThunkX86[] = {
507     0xff, 0x25, 0x00, 0x00, 0x00, 0x00, // JMP *0x0
508 };
509 
510 static const uint8_t importThunkARM[] = {
511     0x40, 0xf2, 0x00, 0x0c, // mov.w ip, #0
512     0xc0, 0xf2, 0x00, 0x0c, // mov.t ip, #0
513     0xdc, 0xf8, 0x00, 0xf0, // ldr.w pc, [ip]
514 };
515 
516 static const uint8_t importThunkARM64[] = {
517     0x10, 0x00, 0x00, 0x90, // adrp x16, #0
518     0x10, 0x02, 0x40, 0xf9, // ldr  x16, [x16]
519     0x00, 0x02, 0x1f, 0xd6, // br   x16
520 };
521 
522 // Windows-specific.
523 // A chunk for DLL import jump table entry. In a final output, its
524 // contents will be a JMP instruction to some __imp_ symbol.
525 class ImportThunkChunk : public NonSectionCodeChunk {
526 public:
ImportThunkChunk(COFFLinkerContext & ctx,Defined * s)527   ImportThunkChunk(COFFLinkerContext &ctx, Defined *s)
528       : NonSectionCodeChunk(ImportThunkKind), impSymbol(s), ctx(ctx) {}
classof(const Chunk * c)529   static bool classof(const Chunk *c) { return c->kind() == ImportThunkKind; }
530 
531 protected:
532   Defined *impSymbol;
533   COFFLinkerContext &ctx;
534 };
535 
536 class ImportThunkChunkX64 : public ImportThunkChunk {
537 public:
538   explicit ImportThunkChunkX64(COFFLinkerContext &ctx, Defined *s);
getSize()539   size_t getSize() const override { return sizeof(importThunkX86); }
540   void writeTo(uint8_t *buf) const override;
getMachine()541   MachineTypes getMachine() const override { return AMD64; }
542 };
543 
544 class ImportThunkChunkX86 : public ImportThunkChunk {
545 public:
ImportThunkChunkX86(COFFLinkerContext & ctx,Defined * s)546   explicit ImportThunkChunkX86(COFFLinkerContext &ctx, Defined *s)
547       : ImportThunkChunk(ctx, s) {}
getSize()548   size_t getSize() const override { return sizeof(importThunkX86); }
549   void getBaserels(std::vector<Baserel> *res) override;
550   void writeTo(uint8_t *buf) const override;
getMachine()551   MachineTypes getMachine() const override { return I386; }
552 };
553 
554 class ImportThunkChunkARM : public ImportThunkChunk {
555 public:
ImportThunkChunkARM(COFFLinkerContext & ctx,Defined * s)556   explicit ImportThunkChunkARM(COFFLinkerContext &ctx, Defined *s)
557       : ImportThunkChunk(ctx, s) {
558     setAlignment(2);
559   }
getSize()560   size_t getSize() const override { return sizeof(importThunkARM); }
561   void getBaserels(std::vector<Baserel> *res) override;
562   void writeTo(uint8_t *buf) const override;
getMachine()563   MachineTypes getMachine() const override { return ARMNT; }
564 };
565 
566 class ImportThunkChunkARM64 : public ImportThunkChunk {
567 public:
ImportThunkChunkARM64(COFFLinkerContext & ctx,Defined * s)568   explicit ImportThunkChunkARM64(COFFLinkerContext &ctx, Defined *s)
569       : ImportThunkChunk(ctx, s) {
570     setAlignment(4);
571   }
getSize()572   size_t getSize() const override { return sizeof(importThunkARM64); }
573   void writeTo(uint8_t *buf) const override;
getMachine()574   MachineTypes getMachine() const override { return ARM64; }
575 };
576 
577 class RangeExtensionThunkARM : public NonSectionCodeChunk {
578 public:
RangeExtensionThunkARM(COFFLinkerContext & ctx,Defined * t)579   explicit RangeExtensionThunkARM(COFFLinkerContext &ctx, Defined *t)
580       : target(t), ctx(ctx) {
581     setAlignment(2);
582   }
583   size_t getSize() const override;
584   void writeTo(uint8_t *buf) const override;
getMachine()585   MachineTypes getMachine() const override { return ARMNT; }
586 
587   Defined *target;
588 
589 private:
590   COFFLinkerContext &ctx;
591 };
592 
593 class RangeExtensionThunkARM64 : public NonSectionCodeChunk {
594 public:
RangeExtensionThunkARM64(COFFLinkerContext & ctx,Defined * t)595   explicit RangeExtensionThunkARM64(COFFLinkerContext &ctx, Defined *t)
596       : target(t), ctx(ctx) {
597     setAlignment(4);
598   }
599   size_t getSize() const override;
600   void writeTo(uint8_t *buf) const override;
getMachine()601   MachineTypes getMachine() const override { return ARM64; }
602 
603   Defined *target;
604 
605 private:
606   COFFLinkerContext &ctx;
607 };
608 
609 // Windows-specific.
610 // See comments for DefinedLocalImport class.
611 class LocalImportChunk : public NonSectionChunk {
612 public:
613   explicit LocalImportChunk(COFFLinkerContext &ctx, Defined *s);
614   size_t getSize() const override;
615   void getBaserels(std::vector<Baserel> *res) override;
616   void writeTo(uint8_t *buf) const override;
617 
618 private:
619   Defined *sym;
620   COFFLinkerContext &ctx;
621 };
622 
623 // Duplicate RVAs are not allowed in RVA tables, so unique symbols by chunk and
624 // offset into the chunk. Order does not matter as the RVA table will be sorted
625 // later.
626 struct ChunkAndOffset {
627   Chunk *inputChunk;
628   uint32_t offset;
629 
630   struct DenseMapInfo {
getEmptyKeyChunkAndOffset::DenseMapInfo631     static ChunkAndOffset getEmptyKey() {
632       return {llvm::DenseMapInfo<Chunk *>::getEmptyKey(), 0};
633     }
getTombstoneKeyChunkAndOffset::DenseMapInfo634     static ChunkAndOffset getTombstoneKey() {
635       return {llvm::DenseMapInfo<Chunk *>::getTombstoneKey(), 0};
636     }
getHashValueChunkAndOffset::DenseMapInfo637     static unsigned getHashValue(const ChunkAndOffset &co) {
638       return llvm::DenseMapInfo<std::pair<Chunk *, uint32_t>>::getHashValue(
639           {co.inputChunk, co.offset});
640     }
isEqualChunkAndOffset::DenseMapInfo641     static bool isEqual(const ChunkAndOffset &lhs, const ChunkAndOffset &rhs) {
642       return lhs.inputChunk == rhs.inputChunk && lhs.offset == rhs.offset;
643     }
644   };
645 };
646 
647 using SymbolRVASet = llvm::DenseSet<ChunkAndOffset>;
648 
649 // Table which contains symbol RVAs. Used for /safeseh and /guard:cf.
650 class RVATableChunk : public NonSectionChunk {
651 public:
RVATableChunk(SymbolRVASet s)652   explicit RVATableChunk(SymbolRVASet s) : syms(std::move(s)) {}
getSize()653   size_t getSize() const override { return syms.size() * 4; }
654   void writeTo(uint8_t *buf) const override;
655 
656 private:
657   SymbolRVASet syms;
658 };
659 
660 // Table which contains symbol RVAs with flags. Used for /guard:ehcont.
661 class RVAFlagTableChunk : public NonSectionChunk {
662 public:
RVAFlagTableChunk(SymbolRVASet s)663   explicit RVAFlagTableChunk(SymbolRVASet s) : syms(std::move(s)) {}
getSize()664   size_t getSize() const override { return syms.size() * 5; }
665   void writeTo(uint8_t *buf) const override;
666 
667 private:
668   SymbolRVASet syms;
669 };
670 
671 // Windows-specific.
672 // This class represents a block in .reloc section.
673 // See the PE/COFF spec 5.6 for details.
674 class BaserelChunk : public NonSectionChunk {
675 public:
676   BaserelChunk(uint32_t page, Baserel *begin, Baserel *end);
getSize()677   size_t getSize() const override { return data.size(); }
678   void writeTo(uint8_t *buf) const override;
679 
680 private:
681   std::vector<uint8_t> data;
682 };
683 
684 class Baserel {
685 public:
Baserel(uint32_t v,uint8_t ty)686   Baserel(uint32_t v, uint8_t ty) : rva(v), type(ty) {}
Baserel(uint32_t v,llvm::COFF::MachineTypes machine)687   explicit Baserel(uint32_t v, llvm::COFF::MachineTypes machine)
688       : Baserel(v, getDefaultType(machine)) {}
689   uint8_t getDefaultType(llvm::COFF::MachineTypes machine);
690 
691   uint32_t rva;
692   uint8_t type;
693 };
694 
695 // This is a placeholder Chunk, to allow attaching a DefinedSynthetic to a
696 // specific place in a section, without any data. This is used for the MinGW
697 // specific symbol __RUNTIME_PSEUDO_RELOC_LIST_END__, even though the concept
698 // of an empty chunk isn't MinGW specific.
699 class EmptyChunk : public NonSectionChunk {
700 public:
EmptyChunk()701   EmptyChunk() {}
getSize()702   size_t getSize() const override { return 0; }
writeTo(uint8_t * buf)703   void writeTo(uint8_t *buf) const override {}
704 };
705 
706 class ECCodeMapEntry {
707 public:
ECCodeMapEntry(Chunk * first,Chunk * last,chpe_range_type type)708   ECCodeMapEntry(Chunk *first, Chunk *last, chpe_range_type type)
709       : first(first), last(last), type(type) {}
710   Chunk *first;
711   Chunk *last;
712   chpe_range_type type;
713 };
714 
715 // This is a chunk containing CHPE code map on EC targets. It's a table
716 // of address ranges and their types.
717 class ECCodeMapChunk : public NonSectionChunk {
718 public:
ECCodeMapChunk(std::vector<ECCodeMapEntry> & map)719   ECCodeMapChunk(std::vector<ECCodeMapEntry> &map) : map(map) {}
720   size_t getSize() const override;
721   void writeTo(uint8_t *buf) const override;
722 
723 private:
724   std::vector<ECCodeMapEntry> &map;
725 };
726 
727 // MinGW specific, for the "automatic import of variables from DLLs" feature.
728 // This provides the table of runtime pseudo relocations, for variable
729 // references that turned out to need to be imported from a DLL even though
730 // the reference didn't use the dllimport attribute. The MinGW runtime will
731 // process this table after loading, before handling control over to user
732 // code.
733 class PseudoRelocTableChunk : public NonSectionChunk {
734 public:
PseudoRelocTableChunk(std::vector<RuntimePseudoReloc> & relocs)735   PseudoRelocTableChunk(std::vector<RuntimePseudoReloc> &relocs)
736       : relocs(std::move(relocs)) {
737     setAlignment(4);
738   }
739   size_t getSize() const override;
740   void writeTo(uint8_t *buf) const override;
741 
742 private:
743   std::vector<RuntimePseudoReloc> relocs;
744 };
745 
746 // MinGW specific. A Chunk that contains one pointer-sized absolute value.
747 class AbsolutePointerChunk : public NonSectionChunk {
748 public:
AbsolutePointerChunk(COFFLinkerContext & ctx,uint64_t value)749   AbsolutePointerChunk(COFFLinkerContext &ctx, uint64_t value)
750       : value(value), ctx(ctx) {
751     setAlignment(getSize());
752   }
753   size_t getSize() const override;
754   void writeTo(uint8_t *buf) const override;
755 
756 private:
757   uint64_t value;
758   COFFLinkerContext &ctx;
759 };
760 
761 // Return true if this file has the hotpatch flag set to true in the S_COMPILE3
762 // record in codeview debug info. Also returns true for some thunks synthesized
763 // by the linker.
isHotPatchable()764 inline bool Chunk::isHotPatchable() const {
765   if (auto *sc = dyn_cast<SectionChunk>(this))
766     return sc->file->hotPatchable;
767   else if (isa<ImportThunkChunk>(this))
768     return true;
769   return false;
770 }
771 
772 void applyMOV32T(uint8_t *off, uint32_t v);
773 void applyBranch24T(uint8_t *off, int32_t v);
774 
775 void applyArm64Addr(uint8_t *off, uint64_t s, uint64_t p, int shift);
776 void applyArm64Imm(uint8_t *off, uint64_t imm, uint32_t rangeLimit);
777 void applyArm64Branch26(uint8_t *off, int64_t v);
778 
779 // Convenience class for initializing a coff_section with specific flags.
780 class FakeSection {
781 public:
FakeSection(int c)782   FakeSection(int c) { section.Characteristics = c; }
783 
784   coff_section section;
785 };
786 
787 // Convenience class for initializing a SectionChunk with specific flags.
788 class FakeSectionChunk {
789 public:
FakeSectionChunk(const coff_section * section)790   FakeSectionChunk(const coff_section *section) : chunk(nullptr, section) {
791     // Comdats from LTO files can't be fully treated as regular comdats
792     // at this point; we don't know what size or contents they are going to
793     // have, so we can't do proper checking of such aspects of them.
794     chunk.selection = llvm::COFF::IMAGE_COMDAT_SELECT_ANY;
795   }
796 
797   SectionChunk chunk;
798 };
799 
800 } // namespace lld::coff
801 
802 namespace llvm {
803 template <>
804 struct DenseMapInfo<lld::coff::ChunkAndOffset>
805     : lld::coff::ChunkAndOffset::DenseMapInfo {};
806 }
807 
808 #endif
809