xref: /freebsd/contrib/llvm-project/llvm/lib/MC/WasmObjectWriter.cpp (revision 0fca6ea1d4eea4c934cfff25ac9ee8ad6fe95583)
1 //===- lib/MC/WasmObjectWriter.cpp - Wasm File Writer ---------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This file implements Wasm object file writer information.
10 //
11 //===----------------------------------------------------------------------===//
12 
13 #include "llvm/ADT/STLExtras.h"
14 #include "llvm/BinaryFormat/Wasm.h"
15 #include "llvm/BinaryFormat/WasmTraits.h"
16 #include "llvm/Config/llvm-config.h"
17 #include "llvm/MC/MCAsmBackend.h"
18 #include "llvm/MC/MCAssembler.h"
19 #include "llvm/MC/MCContext.h"
20 #include "llvm/MC/MCExpr.h"
21 #include "llvm/MC/MCFixupKindInfo.h"
22 #include "llvm/MC/MCObjectWriter.h"
23 #include "llvm/MC/MCSectionWasm.h"
24 #include "llvm/MC/MCSymbolWasm.h"
25 #include "llvm/MC/MCValue.h"
26 #include "llvm/MC/MCWasmObjectWriter.h"
27 #include "llvm/Support/Casting.h"
28 #include "llvm/Support/Debug.h"
29 #include "llvm/Support/EndianStream.h"
30 #include "llvm/Support/ErrorHandling.h"
31 #include "llvm/Support/LEB128.h"
32 #include <vector>
33 
34 using namespace llvm;
35 
36 #define DEBUG_TYPE "mc"
37 
38 namespace {
39 
40 // When we create the indirect function table we start at 1, so that there is
41 // and empty slot at 0 and therefore calling a null function pointer will trap.
42 static const uint32_t InitialTableOffset = 1;
43 
44 // For patching purposes, we need to remember where each section starts, both
45 // for patching up the section size field, and for patching up references to
46 // locations within the section.
47 struct SectionBookkeeping {
48   // Where the size of the section is written.
49   uint64_t SizeOffset;
50   // Where the section header ends (without custom section name).
51   uint64_t PayloadOffset;
52   // Where the contents of the section starts.
53   uint64_t ContentsOffset;
54   uint32_t Index;
55 };
56 
57 // A wasm data segment.  A wasm binary contains only a single data section
58 // but that can contain many segments, each with their own virtual location
59 // in memory.  Each MCSection data created by llvm is modeled as its own
60 // wasm data segment.
61 struct WasmDataSegment {
62   MCSectionWasm *Section;
63   StringRef Name;
64   uint32_t InitFlags;
65   uint64_t Offset;
66   uint32_t Alignment;
67   uint32_t LinkingFlags;
68   SmallVector<char, 4> Data;
69 };
70 
71 // A wasm function to be written into the function section.
72 struct WasmFunction {
73   uint32_t SigIndex;
74   MCSection *Section;
75 };
76 
77 // A wasm global to be written into the global section.
78 struct WasmGlobal {
79   wasm::WasmGlobalType Type;
80   uint64_t InitialValue;
81 };
82 
83 // Information about a single item which is part of a COMDAT.  For each data
84 // segment or function which is in the COMDAT, there is a corresponding
85 // WasmComdatEntry.
86 struct WasmComdatEntry {
87   unsigned Kind;
88   uint32_t Index;
89 };
90 
91 // Information about a single relocation.
92 struct WasmRelocationEntry {
93   uint64_t Offset;                   // Where is the relocation.
94   const MCSymbolWasm *Symbol;        // The symbol to relocate with.
95   int64_t Addend;                    // A value to add to the symbol.
96   unsigned Type;                     // The type of the relocation.
97   const MCSectionWasm *FixupSection; // The section the relocation is targeting.
98 
WasmRelocationEntry__anon335a907b0111::WasmRelocationEntry99   WasmRelocationEntry(uint64_t Offset, const MCSymbolWasm *Symbol,
100                       int64_t Addend, unsigned Type,
101                       const MCSectionWasm *FixupSection)
102       : Offset(Offset), Symbol(Symbol), Addend(Addend), Type(Type),
103         FixupSection(FixupSection) {}
104 
hasAddend__anon335a907b0111::WasmRelocationEntry105   bool hasAddend() const { return wasm::relocTypeHasAddend(Type); }
106 
print__anon335a907b0111::WasmRelocationEntry107   void print(raw_ostream &Out) const {
108     Out << wasm::relocTypetoString(Type) << " Off=" << Offset
109         << ", Sym=" << *Symbol << ", Addend=" << Addend
110         << ", FixupSection=" << FixupSection->getName();
111   }
112 
113 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
dump__anon335a907b0111::WasmRelocationEntry114   LLVM_DUMP_METHOD void dump() const { print(dbgs()); }
115 #endif
116 };
117 
118 static const uint32_t InvalidIndex = -1;
119 
120 struct WasmCustomSection {
121 
122   StringRef Name;
123   MCSectionWasm *Section;
124 
125   uint32_t OutputContentsOffset = 0;
126   uint32_t OutputIndex = InvalidIndex;
127 
WasmCustomSection__anon335a907b0111::WasmCustomSection128   WasmCustomSection(StringRef Name, MCSectionWasm *Section)
129       : Name(Name), Section(Section) {}
130 };
131 
132 #if !defined(NDEBUG)
operator <<(raw_ostream & OS,const WasmRelocationEntry & Rel)133 raw_ostream &operator<<(raw_ostream &OS, const WasmRelocationEntry &Rel) {
134   Rel.print(OS);
135   return OS;
136 }
137 #endif
138 
139 // Write Value as an (unsigned) LEB value at offset Offset in Stream, padded
140 // to allow patching.
141 template <typename T, int W>
writePatchableULEB(raw_pwrite_stream & Stream,T Value,uint64_t Offset)142 void writePatchableULEB(raw_pwrite_stream &Stream, T Value, uint64_t Offset) {
143   uint8_t Buffer[W];
144   unsigned SizeLen = encodeULEB128(Value, Buffer, W);
145   assert(SizeLen == W);
146   Stream.pwrite((char *)Buffer, SizeLen, Offset);
147 }
148 
149 // Write Value as an signed LEB value at offset Offset in Stream, padded
150 // to allow patching.
151 template <typename T, int W>
writePatchableSLEB(raw_pwrite_stream & Stream,T Value,uint64_t Offset)152 void writePatchableSLEB(raw_pwrite_stream &Stream, T Value, uint64_t Offset) {
153   uint8_t Buffer[W];
154   unsigned SizeLen = encodeSLEB128(Value, Buffer, W);
155   assert(SizeLen == W);
156   Stream.pwrite((char *)Buffer, SizeLen, Offset);
157 }
158 
writePatchableU32(raw_pwrite_stream & Stream,uint32_t Value,uint64_t Offset)159 static void writePatchableU32(raw_pwrite_stream &Stream, uint32_t Value,
160                               uint64_t Offset) {
161   writePatchableULEB<uint32_t, 5>(Stream, Value, Offset);
162 }
163 
writePatchableS32(raw_pwrite_stream & Stream,int32_t Value,uint64_t Offset)164 static void writePatchableS32(raw_pwrite_stream &Stream, int32_t Value,
165                               uint64_t Offset) {
166   writePatchableSLEB<int32_t, 5>(Stream, Value, Offset);
167 }
168 
writePatchableU64(raw_pwrite_stream & Stream,uint64_t Value,uint64_t Offset)169 static void writePatchableU64(raw_pwrite_stream &Stream, uint64_t Value,
170                               uint64_t Offset) {
171   writePatchableSLEB<uint64_t, 10>(Stream, Value, Offset);
172 }
173 
writePatchableS64(raw_pwrite_stream & Stream,int64_t Value,uint64_t Offset)174 static void writePatchableS64(raw_pwrite_stream &Stream, int64_t Value,
175                               uint64_t Offset) {
176   writePatchableSLEB<int64_t, 10>(Stream, Value, Offset);
177 }
178 
179 // Write Value as a plain integer value at offset Offset in Stream.
patchI32(raw_pwrite_stream & Stream,uint32_t Value,uint64_t Offset)180 static void patchI32(raw_pwrite_stream &Stream, uint32_t Value,
181                      uint64_t Offset) {
182   uint8_t Buffer[4];
183   support::endian::write32le(Buffer, Value);
184   Stream.pwrite((char *)Buffer, sizeof(Buffer), Offset);
185 }
186 
patchI64(raw_pwrite_stream & Stream,uint64_t Value,uint64_t Offset)187 static void patchI64(raw_pwrite_stream &Stream, uint64_t Value,
188                      uint64_t Offset) {
189   uint8_t Buffer[8];
190   support::endian::write64le(Buffer, Value);
191   Stream.pwrite((char *)Buffer, sizeof(Buffer), Offset);
192 }
193 
isDwoSection(const MCSection & Sec)194 bool isDwoSection(const MCSection &Sec) {
195   return Sec.getName().ends_with(".dwo");
196 }
197 
198 class WasmObjectWriter : public MCObjectWriter {
199   support::endian::Writer *W = nullptr;
200 
201   /// The target specific Wasm writer instance.
202   std::unique_ptr<MCWasmObjectTargetWriter> TargetObjectWriter;
203 
204   // Relocations for fixing up references in the code section.
205   std::vector<WasmRelocationEntry> CodeRelocations;
206   // Relocations for fixing up references in the data section.
207   std::vector<WasmRelocationEntry> DataRelocations;
208 
209   // Index values to use for fixing up call_indirect type indices.
210   // Maps function symbols to the index of the type of the function
211   DenseMap<const MCSymbolWasm *, uint32_t> TypeIndices;
212   // Maps function symbols to the table element index space. Used
213   // for TABLE_INDEX relocation types (i.e. address taken functions).
214   DenseMap<const MCSymbolWasm *, uint32_t> TableIndices;
215   // Maps function/global/table symbols to the
216   // function/global/table/tag/section index space.
217   DenseMap<const MCSymbolWasm *, uint32_t> WasmIndices;
218   DenseMap<const MCSymbolWasm *, uint32_t> GOTIndices;
219   // Maps data symbols to the Wasm segment and offset/size with the segment.
220   DenseMap<const MCSymbolWasm *, wasm::WasmDataReference> DataLocations;
221 
222   // Stores output data (index, relocations, content offset) for custom
223   // section.
224   std::vector<WasmCustomSection> CustomSections;
225   std::unique_ptr<WasmCustomSection> ProducersSection;
226   std::unique_ptr<WasmCustomSection> TargetFeaturesSection;
227   // Relocations for fixing up references in the custom sections.
228   DenseMap<const MCSectionWasm *, std::vector<WasmRelocationEntry>>
229       CustomSectionsRelocations;
230 
231   // Map from section to defining function symbol.
232   DenseMap<const MCSection *, const MCSymbol *> SectionFunctions;
233 
234   DenseMap<wasm::WasmSignature, uint32_t> SignatureIndices;
235   SmallVector<wasm::WasmSignature, 4> Signatures;
236   SmallVector<WasmDataSegment, 4> DataSegments;
237   unsigned NumFunctionImports = 0;
238   unsigned NumGlobalImports = 0;
239   unsigned NumTableImports = 0;
240   unsigned NumTagImports = 0;
241   uint32_t SectionCount = 0;
242 
243   enum class DwoMode {
244     AllSections,
245     NonDwoOnly,
246     DwoOnly,
247   };
248   bool IsSplitDwarf = false;
249   raw_pwrite_stream *OS = nullptr;
250   raw_pwrite_stream *DwoOS = nullptr;
251 
252   // TargetObjectWriter wranppers.
is64Bit() const253   bool is64Bit() const { return TargetObjectWriter->is64Bit(); }
isEmscripten() const254   bool isEmscripten() const { return TargetObjectWriter->isEmscripten(); }
255 
256   void startSection(SectionBookkeeping &Section, unsigned SectionId);
257   void startCustomSection(SectionBookkeeping &Section, StringRef Name);
258   void endSection(SectionBookkeeping &Section);
259 
260 public:
WasmObjectWriter(std::unique_ptr<MCWasmObjectTargetWriter> MOTW,raw_pwrite_stream & OS_)261   WasmObjectWriter(std::unique_ptr<MCWasmObjectTargetWriter> MOTW,
262                    raw_pwrite_stream &OS_)
263       : TargetObjectWriter(std::move(MOTW)), OS(&OS_) {}
264 
WasmObjectWriter(std::unique_ptr<MCWasmObjectTargetWriter> MOTW,raw_pwrite_stream & OS_,raw_pwrite_stream & DwoOS_)265   WasmObjectWriter(std::unique_ptr<MCWasmObjectTargetWriter> MOTW,
266                    raw_pwrite_stream &OS_, raw_pwrite_stream &DwoOS_)
267       : TargetObjectWriter(std::move(MOTW)), IsSplitDwarf(true), OS(&OS_),
268         DwoOS(&DwoOS_) {}
269 
270 private:
reset()271   void reset() override {
272     CodeRelocations.clear();
273     DataRelocations.clear();
274     TypeIndices.clear();
275     WasmIndices.clear();
276     GOTIndices.clear();
277     TableIndices.clear();
278     DataLocations.clear();
279     CustomSections.clear();
280     ProducersSection.reset();
281     TargetFeaturesSection.reset();
282     CustomSectionsRelocations.clear();
283     SignatureIndices.clear();
284     Signatures.clear();
285     DataSegments.clear();
286     SectionFunctions.clear();
287     NumFunctionImports = 0;
288     NumGlobalImports = 0;
289     NumTableImports = 0;
290     MCObjectWriter::reset();
291   }
292 
293   void writeHeader(const MCAssembler &Asm);
294 
295   void recordRelocation(MCAssembler &Asm, const MCFragment *Fragment,
296                         const MCFixup &Fixup, MCValue Target,
297                         uint64_t &FixedValue) override;
298 
299   void executePostLayoutBinding(MCAssembler &Asm) override;
300   void prepareImports(SmallVectorImpl<wasm::WasmImport> &Imports,
301                       MCAssembler &Asm);
302   uint64_t writeObject(MCAssembler &Asm) override;
303 
304   uint64_t writeOneObject(MCAssembler &Asm, DwoMode Mode);
305 
writeString(const StringRef Str)306   void writeString(const StringRef Str) {
307     encodeULEB128(Str.size(), W->OS);
308     W->OS << Str;
309   }
310 
311   void writeStringWithAlignment(const StringRef Str, unsigned Alignment);
312 
writeI32(int32_t val)313   void writeI32(int32_t val) {
314     char Buffer[4];
315     support::endian::write32le(Buffer, val);
316     W->OS.write(Buffer, sizeof(Buffer));
317   }
318 
writeI64(int64_t val)319   void writeI64(int64_t val) {
320     char Buffer[8];
321     support::endian::write64le(Buffer, val);
322     W->OS.write(Buffer, sizeof(Buffer));
323   }
324 
writeValueType(wasm::ValType Ty)325   void writeValueType(wasm::ValType Ty) { W->OS << static_cast<char>(Ty); }
326 
327   void writeTypeSection(ArrayRef<wasm::WasmSignature> Signatures);
328   void writeImportSection(ArrayRef<wasm::WasmImport> Imports, uint64_t DataSize,
329                           uint32_t NumElements);
330   void writeFunctionSection(ArrayRef<WasmFunction> Functions);
331   void writeExportSection(ArrayRef<wasm::WasmExport> Exports);
332   void writeElemSection(const MCSymbolWasm *IndirectFunctionTable,
333                         ArrayRef<uint32_t> TableElems);
334   void writeDataCountSection();
335   uint32_t writeCodeSection(const MCAssembler &Asm,
336                             ArrayRef<WasmFunction> Functions);
337   uint32_t writeDataSection(const MCAssembler &Asm);
338   void writeTagSection(ArrayRef<uint32_t> TagTypes);
339   void writeGlobalSection(ArrayRef<wasm::WasmGlobal> Globals);
340   void writeTableSection(ArrayRef<wasm::WasmTable> Tables);
341   void writeRelocSection(uint32_t SectionIndex, StringRef Name,
342                          std::vector<WasmRelocationEntry> &Relocations);
343   void writeLinkingMetaDataSection(
344       ArrayRef<wasm::WasmSymbolInfo> SymbolInfos,
345       ArrayRef<std::pair<uint16_t, uint32_t>> InitFuncs,
346       const std::map<StringRef, std::vector<WasmComdatEntry>> &Comdats);
347   void writeCustomSection(WasmCustomSection &CustomSection,
348                           const MCAssembler &Asm);
349   void writeCustomRelocSections();
350 
351   uint64_t getProvisionalValue(const MCAssembler &Asm,
352                                const WasmRelocationEntry &RelEntry);
353   void applyRelocations(ArrayRef<WasmRelocationEntry> Relocations,
354                         uint64_t ContentsOffset, const MCAssembler &Asm);
355 
356   uint32_t getRelocationIndexValue(const WasmRelocationEntry &RelEntry);
357   uint32_t getFunctionType(const MCSymbolWasm &Symbol);
358   uint32_t getTagType(const MCSymbolWasm &Symbol);
359   void registerFunctionType(const MCSymbolWasm &Symbol);
360   void registerTagType(const MCSymbolWasm &Symbol);
361 };
362 
363 } // end anonymous namespace
364 
365 // Write out a section header and a patchable section size field.
startSection(SectionBookkeeping & Section,unsigned SectionId)366 void WasmObjectWriter::startSection(SectionBookkeeping &Section,
367                                     unsigned SectionId) {
368   LLVM_DEBUG(dbgs() << "startSection " << SectionId << "\n");
369   W->OS << char(SectionId);
370 
371   Section.SizeOffset = W->OS.tell();
372 
373   // The section size. We don't know the size yet, so reserve enough space
374   // for any 32-bit value; we'll patch it later.
375   encodeULEB128(0, W->OS, 5);
376 
377   // The position where the section starts, for measuring its size.
378   Section.ContentsOffset = W->OS.tell();
379   Section.PayloadOffset = W->OS.tell();
380   Section.Index = SectionCount++;
381 }
382 
383 // Write a string with extra paddings for trailing alignment
384 // TODO: support alignment at asm and llvm level?
writeStringWithAlignment(const StringRef Str,unsigned Alignment)385 void WasmObjectWriter::writeStringWithAlignment(const StringRef Str,
386                                                 unsigned Alignment) {
387 
388   // Calculate the encoded size of str length and add pads based on it and
389   // alignment.
390   raw_null_ostream NullOS;
391   uint64_t StrSizeLength = encodeULEB128(Str.size(), NullOS);
392   uint64_t Offset = W->OS.tell() + StrSizeLength + Str.size();
393   uint64_t Paddings = offsetToAlignment(Offset, Align(Alignment));
394   Offset += Paddings;
395 
396   // LEB128 greater than 5 bytes is invalid
397   assert((StrSizeLength + Paddings) <= 5 && "too long string to align");
398 
399   encodeSLEB128(Str.size(), W->OS, StrSizeLength + Paddings);
400   W->OS << Str;
401 
402   assert(W->OS.tell() == Offset && "invalid padding");
403 }
404 
startCustomSection(SectionBookkeeping & Section,StringRef Name)405 void WasmObjectWriter::startCustomSection(SectionBookkeeping &Section,
406                                           StringRef Name) {
407   LLVM_DEBUG(dbgs() << "startCustomSection " << Name << "\n");
408   startSection(Section, wasm::WASM_SEC_CUSTOM);
409 
410   // The position where the section header ends, for measuring its size.
411   Section.PayloadOffset = W->OS.tell();
412 
413   // Custom sections in wasm also have a string identifier.
414   if (Name != "__clangast") {
415     writeString(Name);
416   } else {
417     // The on-disk hashtable in clangast needs to be aligned by 4 bytes.
418     writeStringWithAlignment(Name, 4);
419   }
420 
421   // The position where the custom section starts.
422   Section.ContentsOffset = W->OS.tell();
423 }
424 
425 // Now that the section is complete and we know how big it is, patch up the
426 // section size field at the start of the section.
endSection(SectionBookkeeping & Section)427 void WasmObjectWriter::endSection(SectionBookkeeping &Section) {
428   uint64_t Size = W->OS.tell();
429   // /dev/null doesn't support seek/tell and can report offset of 0.
430   // Simply skip this patching in that case.
431   if (!Size)
432     return;
433 
434   Size -= Section.PayloadOffset;
435   if (uint32_t(Size) != Size)
436     report_fatal_error("section size does not fit in a uint32_t");
437 
438   LLVM_DEBUG(dbgs() << "endSection size=" << Size << "\n");
439 
440   // Write the final section size to the payload_len field, which follows
441   // the section id byte.
442   writePatchableU32(static_cast<raw_pwrite_stream &>(W->OS), Size,
443                     Section.SizeOffset);
444 }
445 
446 // Emit the Wasm header.
writeHeader(const MCAssembler & Asm)447 void WasmObjectWriter::writeHeader(const MCAssembler &Asm) {
448   W->OS.write(wasm::WasmMagic, sizeof(wasm::WasmMagic));
449   W->write<uint32_t>(wasm::WasmVersion);
450 }
451 
executePostLayoutBinding(MCAssembler & Asm)452 void WasmObjectWriter::executePostLayoutBinding(MCAssembler &Asm) {
453   // Some compilation units require the indirect function table to be present
454   // but don't explicitly reference it.  This is the case for call_indirect
455   // without the reference-types feature, and also function bitcasts in all
456   // cases.  In those cases the __indirect_function_table has the
457   // WASM_SYMBOL_NO_STRIP attribute.  Here we make sure this symbol makes it to
458   // the assembler, if needed.
459   if (auto *Sym = Asm.getContext().lookupSymbol("__indirect_function_table")) {
460     const auto *WasmSym = static_cast<const MCSymbolWasm *>(Sym);
461     if (WasmSym->isNoStrip())
462       Asm.registerSymbol(*Sym);
463   }
464 
465   // Build a map of sections to the function that defines them, for use
466   // in recordRelocation.
467   for (const MCSymbol &S : Asm.symbols()) {
468     const auto &WS = static_cast<const MCSymbolWasm &>(S);
469     if (WS.isDefined() && WS.isFunction() && !WS.isVariable()) {
470       const auto &Sec = static_cast<const MCSectionWasm &>(S.getSection());
471       auto Pair = SectionFunctions.insert(std::make_pair(&Sec, &S));
472       if (!Pair.second)
473         report_fatal_error("section already has a defining function: " +
474                            Sec.getName());
475     }
476   }
477 }
478 
recordRelocation(MCAssembler & Asm,const MCFragment * Fragment,const MCFixup & Fixup,MCValue Target,uint64_t & FixedValue)479 void WasmObjectWriter::recordRelocation(MCAssembler &Asm,
480                                         const MCFragment *Fragment,
481                                         const MCFixup &Fixup, MCValue Target,
482                                         uint64_t &FixedValue) {
483   // The WebAssembly backend should never generate FKF_IsPCRel fixups
484   assert(!(Asm.getBackend().getFixupKindInfo(Fixup.getKind()).Flags &
485            MCFixupKindInfo::FKF_IsPCRel));
486 
487   const auto &FixupSection = cast<MCSectionWasm>(*Fragment->getParent());
488   uint64_t C = Target.getConstant();
489   uint64_t FixupOffset = Asm.getFragmentOffset(*Fragment) + Fixup.getOffset();
490   MCContext &Ctx = Asm.getContext();
491   bool IsLocRel = false;
492 
493   if (const MCSymbolRefExpr *RefB = Target.getSymB()) {
494 
495     const auto &SymB = cast<MCSymbolWasm>(RefB->getSymbol());
496 
497     if (FixupSection.isText()) {
498       Ctx.reportError(Fixup.getLoc(),
499                       Twine("symbol '") + SymB.getName() +
500                           "' unsupported subtraction expression used in "
501                           "relocation in code section.");
502       return;
503     }
504 
505     if (SymB.isUndefined()) {
506       Ctx.reportError(Fixup.getLoc(),
507                       Twine("symbol '") + SymB.getName() +
508                           "' can not be undefined in a subtraction expression");
509       return;
510     }
511     const MCSection &SecB = SymB.getSection();
512     if (&SecB != &FixupSection) {
513       Ctx.reportError(Fixup.getLoc(),
514                       Twine("symbol '") + SymB.getName() +
515                           "' can not be placed in a different section");
516       return;
517     }
518     IsLocRel = true;
519     C += FixupOffset - Asm.getSymbolOffset(SymB);
520   }
521 
522   // We either rejected the fixup or folded B into C at this point.
523   const MCSymbolRefExpr *RefA = Target.getSymA();
524   const auto *SymA = cast<MCSymbolWasm>(&RefA->getSymbol());
525 
526   // The .init_array isn't translated as data, so don't do relocations in it.
527   if (FixupSection.getName().starts_with(".init_array")) {
528     SymA->setUsedInInitArray();
529     return;
530   }
531 
532   if (SymA->isVariable()) {
533     const MCExpr *Expr = SymA->getVariableValue();
534     if (const auto *Inner = dyn_cast<MCSymbolRefExpr>(Expr))
535       if (Inner->getKind() == MCSymbolRefExpr::VK_WEAKREF)
536         llvm_unreachable("weakref used in reloc not yet implemented");
537   }
538 
539   // Put any constant offset in an addend. Offsets can be negative, and
540   // LLVM expects wrapping, in contrast to wasm's immediates which can't
541   // be negative and don't wrap.
542   FixedValue = 0;
543 
544   unsigned Type =
545       TargetObjectWriter->getRelocType(Target, Fixup, FixupSection, IsLocRel);
546 
547   // Absolute offset within a section or a function.
548   // Currently only supported for metadata sections.
549   // See: test/MC/WebAssembly/blockaddress.ll
550   if ((Type == wasm::R_WASM_FUNCTION_OFFSET_I32 ||
551        Type == wasm::R_WASM_FUNCTION_OFFSET_I64 ||
552        Type == wasm::R_WASM_SECTION_OFFSET_I32) &&
553       SymA->isDefined()) {
554     // SymA can be a temp data symbol that represents a function (in which case
555     // it needs to be replaced by the section symbol), [XXX and it apparently
556     // later gets changed again to a func symbol?] or it can be a real
557     // function symbol, in which case it can be left as-is.
558 
559     if (!FixupSection.isMetadata())
560       report_fatal_error("relocations for function or section offsets are "
561                          "only supported in metadata sections");
562 
563     const MCSymbol *SectionSymbol = nullptr;
564     const MCSection &SecA = SymA->getSection();
565     if (SecA.isText()) {
566       auto SecSymIt = SectionFunctions.find(&SecA);
567       if (SecSymIt == SectionFunctions.end())
568         report_fatal_error("section doesn\'t have defining symbol");
569       SectionSymbol = SecSymIt->second;
570     } else {
571       SectionSymbol = SecA.getBeginSymbol();
572     }
573     if (!SectionSymbol)
574       report_fatal_error("section symbol is required for relocation");
575 
576     C += Asm.getSymbolOffset(*SymA);
577     SymA = cast<MCSymbolWasm>(SectionSymbol);
578   }
579 
580   if (Type == wasm::R_WASM_TABLE_INDEX_REL_SLEB ||
581       Type == wasm::R_WASM_TABLE_INDEX_REL_SLEB64 ||
582       Type == wasm::R_WASM_TABLE_INDEX_SLEB ||
583       Type == wasm::R_WASM_TABLE_INDEX_SLEB64 ||
584       Type == wasm::R_WASM_TABLE_INDEX_I32 ||
585       Type == wasm::R_WASM_TABLE_INDEX_I64) {
586     // TABLE_INDEX relocs implicitly use the default indirect function table.
587     // We require the function table to have already been defined.
588     auto TableName = "__indirect_function_table";
589     MCSymbolWasm *Sym = cast_or_null<MCSymbolWasm>(Ctx.lookupSymbol(TableName));
590     if (!Sym) {
591       report_fatal_error("missing indirect function table symbol");
592     } else {
593       if (!Sym->isFunctionTable())
594         report_fatal_error("__indirect_function_table symbol has wrong type");
595       // Ensure that __indirect_function_table reaches the output.
596       Sym->setNoStrip();
597       Asm.registerSymbol(*Sym);
598     }
599   }
600 
601   // Relocation other than R_WASM_TYPE_INDEX_LEB are required to be
602   // against a named symbol.
603   if (Type != wasm::R_WASM_TYPE_INDEX_LEB) {
604     if (SymA->getName().empty())
605       report_fatal_error("relocations against un-named temporaries are not yet "
606                          "supported by wasm");
607 
608     SymA->setUsedInReloc();
609   }
610 
611   switch (RefA->getKind()) {
612   case MCSymbolRefExpr::VK_GOT:
613   case MCSymbolRefExpr::VK_WASM_GOT_TLS:
614     SymA->setUsedInGOT();
615     break;
616   default:
617     break;
618   }
619 
620   WasmRelocationEntry Rec(FixupOffset, SymA, C, Type, &FixupSection);
621   LLVM_DEBUG(dbgs() << "WasmReloc: " << Rec << "\n");
622 
623   if (FixupSection.isWasmData()) {
624     DataRelocations.push_back(Rec);
625   } else if (FixupSection.isText()) {
626     CodeRelocations.push_back(Rec);
627   } else if (FixupSection.isMetadata()) {
628     CustomSectionsRelocations[&FixupSection].push_back(Rec);
629   } else {
630     llvm_unreachable("unexpected section type");
631   }
632 }
633 
634 // Compute a value to write into the code at the location covered
635 // by RelEntry. This value isn't used by the static linker; it just serves
636 // to make the object format more readable and more likely to be directly
637 // useable.
638 uint64_t
getProvisionalValue(const MCAssembler & Asm,const WasmRelocationEntry & RelEntry)639 WasmObjectWriter::getProvisionalValue(const MCAssembler &Asm,
640                                       const WasmRelocationEntry &RelEntry) {
641   if ((RelEntry.Type == wasm::R_WASM_GLOBAL_INDEX_LEB ||
642        RelEntry.Type == wasm::R_WASM_GLOBAL_INDEX_I32) &&
643       !RelEntry.Symbol->isGlobal()) {
644     assert(GOTIndices.count(RelEntry.Symbol) > 0 && "symbol not found in GOT index space");
645     return GOTIndices[RelEntry.Symbol];
646   }
647 
648   switch (RelEntry.Type) {
649   case wasm::R_WASM_TABLE_INDEX_REL_SLEB:
650   case wasm::R_WASM_TABLE_INDEX_REL_SLEB64:
651   case wasm::R_WASM_TABLE_INDEX_SLEB:
652   case wasm::R_WASM_TABLE_INDEX_SLEB64:
653   case wasm::R_WASM_TABLE_INDEX_I32:
654   case wasm::R_WASM_TABLE_INDEX_I64: {
655     // Provisional value is table address of the resolved symbol itself
656     const MCSymbolWasm *Base =
657         cast<MCSymbolWasm>(Asm.getBaseSymbol(*RelEntry.Symbol));
658     assert(Base->isFunction());
659     if (RelEntry.Type == wasm::R_WASM_TABLE_INDEX_REL_SLEB ||
660         RelEntry.Type == wasm::R_WASM_TABLE_INDEX_REL_SLEB64)
661       return TableIndices[Base] - InitialTableOffset;
662     else
663       return TableIndices[Base];
664   }
665   case wasm::R_WASM_TYPE_INDEX_LEB:
666     // Provisional value is same as the index
667     return getRelocationIndexValue(RelEntry);
668   case wasm::R_WASM_FUNCTION_INDEX_LEB:
669   case wasm::R_WASM_FUNCTION_INDEX_I32:
670   case wasm::R_WASM_GLOBAL_INDEX_LEB:
671   case wasm::R_WASM_GLOBAL_INDEX_I32:
672   case wasm::R_WASM_TAG_INDEX_LEB:
673   case wasm::R_WASM_TABLE_NUMBER_LEB:
674     // Provisional value is function/global/tag Wasm index
675     assert(WasmIndices.count(RelEntry.Symbol) > 0 && "symbol not found in wasm index space");
676     return WasmIndices[RelEntry.Symbol];
677   case wasm::R_WASM_FUNCTION_OFFSET_I32:
678   case wasm::R_WASM_FUNCTION_OFFSET_I64:
679   case wasm::R_WASM_SECTION_OFFSET_I32: {
680     if (!RelEntry.Symbol->isDefined())
681       return 0;
682     const auto &Section =
683         static_cast<const MCSectionWasm &>(RelEntry.Symbol->getSection());
684     return Section.getSectionOffset() + RelEntry.Addend;
685   }
686   case wasm::R_WASM_MEMORY_ADDR_LEB:
687   case wasm::R_WASM_MEMORY_ADDR_LEB64:
688   case wasm::R_WASM_MEMORY_ADDR_SLEB:
689   case wasm::R_WASM_MEMORY_ADDR_SLEB64:
690   case wasm::R_WASM_MEMORY_ADDR_REL_SLEB:
691   case wasm::R_WASM_MEMORY_ADDR_REL_SLEB64:
692   case wasm::R_WASM_MEMORY_ADDR_I32:
693   case wasm::R_WASM_MEMORY_ADDR_I64:
694   case wasm::R_WASM_MEMORY_ADDR_TLS_SLEB:
695   case wasm::R_WASM_MEMORY_ADDR_TLS_SLEB64:
696   case wasm::R_WASM_MEMORY_ADDR_LOCREL_I32: {
697     // Provisional value is address of the global plus the offset
698     // For undefined symbols, use zero
699     if (!RelEntry.Symbol->isDefined())
700       return 0;
701     const wasm::WasmDataReference &SymRef = DataLocations[RelEntry.Symbol];
702     const WasmDataSegment &Segment = DataSegments[SymRef.Segment];
703     // Ignore overflow. LLVM allows address arithmetic to silently wrap.
704     return Segment.Offset + SymRef.Offset + RelEntry.Addend;
705   }
706   default:
707     llvm_unreachable("invalid relocation type");
708   }
709 }
710 
addData(SmallVectorImpl<char> & DataBytes,MCSectionWasm & DataSection)711 static void addData(SmallVectorImpl<char> &DataBytes,
712                     MCSectionWasm &DataSection) {
713   LLVM_DEBUG(errs() << "addData: " << DataSection.getName() << "\n");
714 
715   DataBytes.resize(alignTo(DataBytes.size(), DataSection.getAlign()));
716 
717   for (const MCFragment &Frag : DataSection) {
718     if (Frag.hasInstructions())
719       report_fatal_error("only data supported in data sections");
720 
721     if (auto *Align = dyn_cast<MCAlignFragment>(&Frag)) {
722       if (Align->getValueSize() != 1)
723         report_fatal_error("only byte values supported for alignment");
724       // If nops are requested, use zeros, as this is the data section.
725       uint8_t Value = Align->hasEmitNops() ? 0 : Align->getValue();
726       uint64_t Size =
727           std::min<uint64_t>(alignTo(DataBytes.size(), Align->getAlignment()),
728                              DataBytes.size() + Align->getMaxBytesToEmit());
729       DataBytes.resize(Size, Value);
730     } else if (auto *Fill = dyn_cast<MCFillFragment>(&Frag)) {
731       int64_t NumValues;
732       if (!Fill->getNumValues().evaluateAsAbsolute(NumValues))
733         llvm_unreachable("The fill should be an assembler constant");
734       DataBytes.insert(DataBytes.end(), Fill->getValueSize() * NumValues,
735                        Fill->getValue());
736     } else if (auto *LEB = dyn_cast<MCLEBFragment>(&Frag)) {
737       const SmallVectorImpl<char> &Contents = LEB->getContents();
738       llvm::append_range(DataBytes, Contents);
739     } else {
740       const auto &DataFrag = cast<MCDataFragment>(Frag);
741       const SmallVectorImpl<char> &Contents = DataFrag.getContents();
742       llvm::append_range(DataBytes, Contents);
743     }
744   }
745 
746   LLVM_DEBUG(dbgs() << "addData -> " << DataBytes.size() << "\n");
747 }
748 
749 uint32_t
getRelocationIndexValue(const WasmRelocationEntry & RelEntry)750 WasmObjectWriter::getRelocationIndexValue(const WasmRelocationEntry &RelEntry) {
751   if (RelEntry.Type == wasm::R_WASM_TYPE_INDEX_LEB) {
752     if (!TypeIndices.count(RelEntry.Symbol))
753       report_fatal_error("symbol not found in type index space: " +
754                          RelEntry.Symbol->getName());
755     return TypeIndices[RelEntry.Symbol];
756   }
757 
758   return RelEntry.Symbol->getIndex();
759 }
760 
761 // Apply the portions of the relocation records that we can handle ourselves
762 // directly.
applyRelocations(ArrayRef<WasmRelocationEntry> Relocations,uint64_t ContentsOffset,const MCAssembler & Asm)763 void WasmObjectWriter::applyRelocations(
764     ArrayRef<WasmRelocationEntry> Relocations, uint64_t ContentsOffset,
765     const MCAssembler &Asm) {
766   auto &Stream = static_cast<raw_pwrite_stream &>(W->OS);
767   for (const WasmRelocationEntry &RelEntry : Relocations) {
768     uint64_t Offset = ContentsOffset +
769                       RelEntry.FixupSection->getSectionOffset() +
770                       RelEntry.Offset;
771 
772     LLVM_DEBUG(dbgs() << "applyRelocation: " << RelEntry << "\n");
773     uint64_t Value = getProvisionalValue(Asm, RelEntry);
774 
775     switch (RelEntry.Type) {
776     case wasm::R_WASM_FUNCTION_INDEX_LEB:
777     case wasm::R_WASM_TYPE_INDEX_LEB:
778     case wasm::R_WASM_GLOBAL_INDEX_LEB:
779     case wasm::R_WASM_MEMORY_ADDR_LEB:
780     case wasm::R_WASM_TAG_INDEX_LEB:
781     case wasm::R_WASM_TABLE_NUMBER_LEB:
782       writePatchableU32(Stream, Value, Offset);
783       break;
784     case wasm::R_WASM_MEMORY_ADDR_LEB64:
785       writePatchableU64(Stream, Value, Offset);
786       break;
787     case wasm::R_WASM_TABLE_INDEX_I32:
788     case wasm::R_WASM_MEMORY_ADDR_I32:
789     case wasm::R_WASM_FUNCTION_OFFSET_I32:
790     case wasm::R_WASM_FUNCTION_INDEX_I32:
791     case wasm::R_WASM_SECTION_OFFSET_I32:
792     case wasm::R_WASM_GLOBAL_INDEX_I32:
793     case wasm::R_WASM_MEMORY_ADDR_LOCREL_I32:
794       patchI32(Stream, Value, Offset);
795       break;
796     case wasm::R_WASM_TABLE_INDEX_I64:
797     case wasm::R_WASM_MEMORY_ADDR_I64:
798     case wasm::R_WASM_FUNCTION_OFFSET_I64:
799       patchI64(Stream, Value, Offset);
800       break;
801     case wasm::R_WASM_TABLE_INDEX_SLEB:
802     case wasm::R_WASM_TABLE_INDEX_REL_SLEB:
803     case wasm::R_WASM_MEMORY_ADDR_SLEB:
804     case wasm::R_WASM_MEMORY_ADDR_REL_SLEB:
805     case wasm::R_WASM_MEMORY_ADDR_TLS_SLEB:
806       writePatchableS32(Stream, Value, Offset);
807       break;
808     case wasm::R_WASM_TABLE_INDEX_SLEB64:
809     case wasm::R_WASM_TABLE_INDEX_REL_SLEB64:
810     case wasm::R_WASM_MEMORY_ADDR_SLEB64:
811     case wasm::R_WASM_MEMORY_ADDR_REL_SLEB64:
812     case wasm::R_WASM_MEMORY_ADDR_TLS_SLEB64:
813       writePatchableS64(Stream, Value, Offset);
814       break;
815     default:
816       llvm_unreachable("invalid relocation type");
817     }
818   }
819 }
820 
writeTypeSection(ArrayRef<wasm::WasmSignature> Signatures)821 void WasmObjectWriter::writeTypeSection(
822     ArrayRef<wasm::WasmSignature> Signatures) {
823   if (Signatures.empty())
824     return;
825 
826   SectionBookkeeping Section;
827   startSection(Section, wasm::WASM_SEC_TYPE);
828 
829   encodeULEB128(Signatures.size(), W->OS);
830 
831   for (const wasm::WasmSignature &Sig : Signatures) {
832     W->OS << char(wasm::WASM_TYPE_FUNC);
833     encodeULEB128(Sig.Params.size(), W->OS);
834     for (wasm::ValType Ty : Sig.Params)
835       writeValueType(Ty);
836     encodeULEB128(Sig.Returns.size(), W->OS);
837     for (wasm::ValType Ty : Sig.Returns)
838       writeValueType(Ty);
839   }
840 
841   endSection(Section);
842 }
843 
writeImportSection(ArrayRef<wasm::WasmImport> Imports,uint64_t DataSize,uint32_t NumElements)844 void WasmObjectWriter::writeImportSection(ArrayRef<wasm::WasmImport> Imports,
845                                           uint64_t DataSize,
846                                           uint32_t NumElements) {
847   if (Imports.empty())
848     return;
849 
850   uint64_t NumPages = (DataSize + wasm::WasmPageSize - 1) / wasm::WasmPageSize;
851 
852   SectionBookkeeping Section;
853   startSection(Section, wasm::WASM_SEC_IMPORT);
854 
855   encodeULEB128(Imports.size(), W->OS);
856   for (const wasm::WasmImport &Import : Imports) {
857     writeString(Import.Module);
858     writeString(Import.Field);
859     W->OS << char(Import.Kind);
860 
861     switch (Import.Kind) {
862     case wasm::WASM_EXTERNAL_FUNCTION:
863       encodeULEB128(Import.SigIndex, W->OS);
864       break;
865     case wasm::WASM_EXTERNAL_GLOBAL:
866       W->OS << char(Import.Global.Type);
867       W->OS << char(Import.Global.Mutable ? 1 : 0);
868       break;
869     case wasm::WASM_EXTERNAL_MEMORY:
870       encodeULEB128(Import.Memory.Flags, W->OS);
871       encodeULEB128(NumPages, W->OS); // initial
872       break;
873     case wasm::WASM_EXTERNAL_TABLE:
874       W->OS << char(Import.Table.ElemType);
875       encodeULEB128(Import.Table.Limits.Flags, W->OS);
876       encodeULEB128(NumElements, W->OS); // initial
877       break;
878     case wasm::WASM_EXTERNAL_TAG:
879       W->OS << char(0); // Reserved 'attribute' field
880       encodeULEB128(Import.SigIndex, W->OS);
881       break;
882     default:
883       llvm_unreachable("unsupported import kind");
884     }
885   }
886 
887   endSection(Section);
888 }
889 
writeFunctionSection(ArrayRef<WasmFunction> Functions)890 void WasmObjectWriter::writeFunctionSection(ArrayRef<WasmFunction> Functions) {
891   if (Functions.empty())
892     return;
893 
894   SectionBookkeeping Section;
895   startSection(Section, wasm::WASM_SEC_FUNCTION);
896 
897   encodeULEB128(Functions.size(), W->OS);
898   for (const WasmFunction &Func : Functions)
899     encodeULEB128(Func.SigIndex, W->OS);
900 
901   endSection(Section);
902 }
903 
writeTagSection(ArrayRef<uint32_t> TagTypes)904 void WasmObjectWriter::writeTagSection(ArrayRef<uint32_t> TagTypes) {
905   if (TagTypes.empty())
906     return;
907 
908   SectionBookkeeping Section;
909   startSection(Section, wasm::WASM_SEC_TAG);
910 
911   encodeULEB128(TagTypes.size(), W->OS);
912   for (uint32_t Index : TagTypes) {
913     W->OS << char(0); // Reserved 'attribute' field
914     encodeULEB128(Index, W->OS);
915   }
916 
917   endSection(Section);
918 }
919 
writeGlobalSection(ArrayRef<wasm::WasmGlobal> Globals)920 void WasmObjectWriter::writeGlobalSection(ArrayRef<wasm::WasmGlobal> Globals) {
921   if (Globals.empty())
922     return;
923 
924   SectionBookkeeping Section;
925   startSection(Section, wasm::WASM_SEC_GLOBAL);
926 
927   encodeULEB128(Globals.size(), W->OS);
928   for (const wasm::WasmGlobal &Global : Globals) {
929     encodeULEB128(Global.Type.Type, W->OS);
930     W->OS << char(Global.Type.Mutable);
931     if (Global.InitExpr.Extended) {
932       llvm_unreachable("extected init expressions not supported");
933     } else {
934       W->OS << char(Global.InitExpr.Inst.Opcode);
935       switch (Global.Type.Type) {
936       case wasm::WASM_TYPE_I32:
937         encodeSLEB128(0, W->OS);
938         break;
939       case wasm::WASM_TYPE_I64:
940         encodeSLEB128(0, W->OS);
941         break;
942       case wasm::WASM_TYPE_F32:
943         writeI32(0);
944         break;
945       case wasm::WASM_TYPE_F64:
946         writeI64(0);
947         break;
948       case wasm::WASM_TYPE_EXTERNREF:
949         writeValueType(wasm::ValType::EXTERNREF);
950         break;
951       default:
952         llvm_unreachable("unexpected type");
953       }
954     }
955     W->OS << char(wasm::WASM_OPCODE_END);
956   }
957 
958   endSection(Section);
959 }
960 
writeTableSection(ArrayRef<wasm::WasmTable> Tables)961 void WasmObjectWriter::writeTableSection(ArrayRef<wasm::WasmTable> Tables) {
962   if (Tables.empty())
963     return;
964 
965   SectionBookkeeping Section;
966   startSection(Section, wasm::WASM_SEC_TABLE);
967 
968   encodeULEB128(Tables.size(), W->OS);
969   for (const wasm::WasmTable &Table : Tables) {
970     assert(Table.Type.ElemType != wasm::ValType::OTHERREF &&
971            "Cannot encode general ref-typed tables");
972     encodeULEB128((uint32_t)Table.Type.ElemType, W->OS);
973     encodeULEB128(Table.Type.Limits.Flags, W->OS);
974     encodeULEB128(Table.Type.Limits.Minimum, W->OS);
975     if (Table.Type.Limits.Flags & wasm::WASM_LIMITS_FLAG_HAS_MAX)
976       encodeULEB128(Table.Type.Limits.Maximum, W->OS);
977   }
978   endSection(Section);
979 }
980 
writeExportSection(ArrayRef<wasm::WasmExport> Exports)981 void WasmObjectWriter::writeExportSection(ArrayRef<wasm::WasmExport> Exports) {
982   if (Exports.empty())
983     return;
984 
985   SectionBookkeeping Section;
986   startSection(Section, wasm::WASM_SEC_EXPORT);
987 
988   encodeULEB128(Exports.size(), W->OS);
989   for (const wasm::WasmExport &Export : Exports) {
990     writeString(Export.Name);
991     W->OS << char(Export.Kind);
992     encodeULEB128(Export.Index, W->OS);
993   }
994 
995   endSection(Section);
996 }
997 
writeElemSection(const MCSymbolWasm * IndirectFunctionTable,ArrayRef<uint32_t> TableElems)998 void WasmObjectWriter::writeElemSection(
999     const MCSymbolWasm *IndirectFunctionTable, ArrayRef<uint32_t> TableElems) {
1000   if (TableElems.empty())
1001     return;
1002 
1003   assert(IndirectFunctionTable);
1004 
1005   SectionBookkeeping Section;
1006   startSection(Section, wasm::WASM_SEC_ELEM);
1007 
1008   encodeULEB128(1, W->OS); // number of "segments"
1009 
1010   assert(WasmIndices.count(IndirectFunctionTable));
1011   uint32_t TableNumber = WasmIndices.find(IndirectFunctionTable)->second;
1012   uint32_t Flags = 0;
1013   if (TableNumber)
1014     Flags |= wasm::WASM_ELEM_SEGMENT_HAS_TABLE_NUMBER;
1015   encodeULEB128(Flags, W->OS);
1016   if (Flags & wasm::WASM_ELEM_SEGMENT_HAS_TABLE_NUMBER)
1017     encodeULEB128(TableNumber, W->OS); // the table number
1018 
1019   // init expr for starting offset
1020   W->OS << char(is64Bit() ? wasm::WASM_OPCODE_I64_CONST
1021                           : wasm::WASM_OPCODE_I32_CONST);
1022   encodeSLEB128(InitialTableOffset, W->OS);
1023   W->OS << char(wasm::WASM_OPCODE_END);
1024 
1025   if (Flags & wasm::WASM_ELEM_SEGMENT_MASK_HAS_ELEM_KIND) {
1026     // We only write active function table initializers, for which the elem kind
1027     // is specified to be written as 0x00 and interpreted to mean "funcref".
1028     const uint8_t ElemKind = 0;
1029     W->OS << ElemKind;
1030   }
1031 
1032   encodeULEB128(TableElems.size(), W->OS);
1033   for (uint32_t Elem : TableElems)
1034     encodeULEB128(Elem, W->OS);
1035 
1036   endSection(Section);
1037 }
1038 
writeDataCountSection()1039 void WasmObjectWriter::writeDataCountSection() {
1040   if (DataSegments.empty())
1041     return;
1042 
1043   SectionBookkeeping Section;
1044   startSection(Section, wasm::WASM_SEC_DATACOUNT);
1045   encodeULEB128(DataSegments.size(), W->OS);
1046   endSection(Section);
1047 }
1048 
writeCodeSection(const MCAssembler & Asm,ArrayRef<WasmFunction> Functions)1049 uint32_t WasmObjectWriter::writeCodeSection(const MCAssembler &Asm,
1050                                             ArrayRef<WasmFunction> Functions) {
1051   if (Functions.empty())
1052     return 0;
1053 
1054   SectionBookkeeping Section;
1055   startSection(Section, wasm::WASM_SEC_CODE);
1056 
1057   encodeULEB128(Functions.size(), W->OS);
1058 
1059   for (const WasmFunction &Func : Functions) {
1060     auto *FuncSection = static_cast<MCSectionWasm *>(Func.Section);
1061 
1062     int64_t Size = Asm.getSectionAddressSize(*FuncSection);
1063     encodeULEB128(Size, W->OS);
1064     FuncSection->setSectionOffset(W->OS.tell() - Section.ContentsOffset);
1065     Asm.writeSectionData(W->OS, FuncSection);
1066   }
1067 
1068   // Apply fixups.
1069   applyRelocations(CodeRelocations, Section.ContentsOffset, Asm);
1070 
1071   endSection(Section);
1072   return Section.Index;
1073 }
1074 
writeDataSection(const MCAssembler & Asm)1075 uint32_t WasmObjectWriter::writeDataSection(const MCAssembler &Asm) {
1076   if (DataSegments.empty())
1077     return 0;
1078 
1079   SectionBookkeeping Section;
1080   startSection(Section, wasm::WASM_SEC_DATA);
1081 
1082   encodeULEB128(DataSegments.size(), W->OS); // count
1083 
1084   for (const WasmDataSegment &Segment : DataSegments) {
1085     encodeULEB128(Segment.InitFlags, W->OS); // flags
1086     if (Segment.InitFlags & wasm::WASM_DATA_SEGMENT_HAS_MEMINDEX)
1087       encodeULEB128(0, W->OS); // memory index
1088     if ((Segment.InitFlags & wasm::WASM_DATA_SEGMENT_IS_PASSIVE) == 0) {
1089       W->OS << char(is64Bit() ? wasm::WASM_OPCODE_I64_CONST
1090                               : wasm::WASM_OPCODE_I32_CONST);
1091       encodeSLEB128(Segment.Offset, W->OS); // offset
1092       W->OS << char(wasm::WASM_OPCODE_END);
1093     }
1094     encodeULEB128(Segment.Data.size(), W->OS); // size
1095     Segment.Section->setSectionOffset(W->OS.tell() - Section.ContentsOffset);
1096     W->OS << Segment.Data; // data
1097   }
1098 
1099   // Apply fixups.
1100   applyRelocations(DataRelocations, Section.ContentsOffset, Asm);
1101 
1102   endSection(Section);
1103   return Section.Index;
1104 }
1105 
writeRelocSection(uint32_t SectionIndex,StringRef Name,std::vector<WasmRelocationEntry> & Relocs)1106 void WasmObjectWriter::writeRelocSection(
1107     uint32_t SectionIndex, StringRef Name,
1108     std::vector<WasmRelocationEntry> &Relocs) {
1109   // See: https://github.com/WebAssembly/tool-conventions/blob/main/Linking.md
1110   // for descriptions of the reloc sections.
1111 
1112   if (Relocs.empty())
1113     return;
1114 
1115   // First, ensure the relocations are sorted in offset order.  In general they
1116   // should already be sorted since `recordRelocation` is called in offset
1117   // order, but for the code section we combine many MC sections into single
1118   // wasm section, and this order is determined by the order of Asm.Symbols()
1119   // not the sections order.
1120   llvm::stable_sort(
1121       Relocs, [](const WasmRelocationEntry &A, const WasmRelocationEntry &B) {
1122         return (A.Offset + A.FixupSection->getSectionOffset()) <
1123                (B.Offset + B.FixupSection->getSectionOffset());
1124       });
1125 
1126   SectionBookkeeping Section;
1127   startCustomSection(Section, std::string("reloc.") + Name.str());
1128 
1129   encodeULEB128(SectionIndex, W->OS);
1130   encodeULEB128(Relocs.size(), W->OS);
1131   for (const WasmRelocationEntry &RelEntry : Relocs) {
1132     uint64_t Offset =
1133         RelEntry.Offset + RelEntry.FixupSection->getSectionOffset();
1134     uint32_t Index = getRelocationIndexValue(RelEntry);
1135 
1136     W->OS << char(RelEntry.Type);
1137     encodeULEB128(Offset, W->OS);
1138     encodeULEB128(Index, W->OS);
1139     if (RelEntry.hasAddend())
1140       encodeSLEB128(RelEntry.Addend, W->OS);
1141   }
1142 
1143   endSection(Section);
1144 }
1145 
writeCustomRelocSections()1146 void WasmObjectWriter::writeCustomRelocSections() {
1147   for (const auto &Sec : CustomSections) {
1148     auto &Relocations = CustomSectionsRelocations[Sec.Section];
1149     writeRelocSection(Sec.OutputIndex, Sec.Name, Relocations);
1150   }
1151 }
1152 
writeLinkingMetaDataSection(ArrayRef<wasm::WasmSymbolInfo> SymbolInfos,ArrayRef<std::pair<uint16_t,uint32_t>> InitFuncs,const std::map<StringRef,std::vector<WasmComdatEntry>> & Comdats)1153 void WasmObjectWriter::writeLinkingMetaDataSection(
1154     ArrayRef<wasm::WasmSymbolInfo> SymbolInfos,
1155     ArrayRef<std::pair<uint16_t, uint32_t>> InitFuncs,
1156     const std::map<StringRef, std::vector<WasmComdatEntry>> &Comdats) {
1157   SectionBookkeeping Section;
1158   startCustomSection(Section, "linking");
1159   encodeULEB128(wasm::WasmMetadataVersion, W->OS);
1160 
1161   SectionBookkeeping SubSection;
1162   if (SymbolInfos.size() != 0) {
1163     startSection(SubSection, wasm::WASM_SYMBOL_TABLE);
1164     encodeULEB128(SymbolInfos.size(), W->OS);
1165     for (const wasm::WasmSymbolInfo &Sym : SymbolInfos) {
1166       encodeULEB128(Sym.Kind, W->OS);
1167       encodeULEB128(Sym.Flags, W->OS);
1168       switch (Sym.Kind) {
1169       case wasm::WASM_SYMBOL_TYPE_FUNCTION:
1170       case wasm::WASM_SYMBOL_TYPE_GLOBAL:
1171       case wasm::WASM_SYMBOL_TYPE_TAG:
1172       case wasm::WASM_SYMBOL_TYPE_TABLE:
1173         encodeULEB128(Sym.ElementIndex, W->OS);
1174         if ((Sym.Flags & wasm::WASM_SYMBOL_UNDEFINED) == 0 ||
1175             (Sym.Flags & wasm::WASM_SYMBOL_EXPLICIT_NAME) != 0)
1176           writeString(Sym.Name);
1177         break;
1178       case wasm::WASM_SYMBOL_TYPE_DATA:
1179         writeString(Sym.Name);
1180         if ((Sym.Flags & wasm::WASM_SYMBOL_UNDEFINED) == 0) {
1181           encodeULEB128(Sym.DataRef.Segment, W->OS);
1182           encodeULEB128(Sym.DataRef.Offset, W->OS);
1183           encodeULEB128(Sym.DataRef.Size, W->OS);
1184         }
1185         break;
1186       case wasm::WASM_SYMBOL_TYPE_SECTION: {
1187         const uint32_t SectionIndex =
1188             CustomSections[Sym.ElementIndex].OutputIndex;
1189         encodeULEB128(SectionIndex, W->OS);
1190         break;
1191       }
1192       default:
1193         llvm_unreachable("unexpected kind");
1194       }
1195     }
1196     endSection(SubSection);
1197   }
1198 
1199   if (DataSegments.size()) {
1200     startSection(SubSection, wasm::WASM_SEGMENT_INFO);
1201     encodeULEB128(DataSegments.size(), W->OS);
1202     for (const WasmDataSegment &Segment : DataSegments) {
1203       writeString(Segment.Name);
1204       encodeULEB128(Segment.Alignment, W->OS);
1205       encodeULEB128(Segment.LinkingFlags, W->OS);
1206     }
1207     endSection(SubSection);
1208   }
1209 
1210   if (!InitFuncs.empty()) {
1211     startSection(SubSection, wasm::WASM_INIT_FUNCS);
1212     encodeULEB128(InitFuncs.size(), W->OS);
1213     for (auto &StartFunc : InitFuncs) {
1214       encodeULEB128(StartFunc.first, W->OS);  // priority
1215       encodeULEB128(StartFunc.second, W->OS); // function index
1216     }
1217     endSection(SubSection);
1218   }
1219 
1220   if (Comdats.size()) {
1221     startSection(SubSection, wasm::WASM_COMDAT_INFO);
1222     encodeULEB128(Comdats.size(), W->OS);
1223     for (const auto &C : Comdats) {
1224       writeString(C.first);
1225       encodeULEB128(0, W->OS); // flags for future use
1226       encodeULEB128(C.second.size(), W->OS);
1227       for (const WasmComdatEntry &Entry : C.second) {
1228         encodeULEB128(Entry.Kind, W->OS);
1229         encodeULEB128(Entry.Index, W->OS);
1230       }
1231     }
1232     endSection(SubSection);
1233   }
1234 
1235   endSection(Section);
1236 }
1237 
writeCustomSection(WasmCustomSection & CustomSection,const MCAssembler & Asm)1238 void WasmObjectWriter::writeCustomSection(WasmCustomSection &CustomSection,
1239                                           const MCAssembler &Asm) {
1240   SectionBookkeeping Section;
1241   auto *Sec = CustomSection.Section;
1242   startCustomSection(Section, CustomSection.Name);
1243 
1244   Sec->setSectionOffset(W->OS.tell() - Section.ContentsOffset);
1245   Asm.writeSectionData(W->OS, Sec);
1246 
1247   CustomSection.OutputContentsOffset = Section.ContentsOffset;
1248   CustomSection.OutputIndex = Section.Index;
1249 
1250   endSection(Section);
1251 
1252   // Apply fixups.
1253   auto &Relocations = CustomSectionsRelocations[CustomSection.Section];
1254   applyRelocations(Relocations, CustomSection.OutputContentsOffset, Asm);
1255 }
1256 
getFunctionType(const MCSymbolWasm & Symbol)1257 uint32_t WasmObjectWriter::getFunctionType(const MCSymbolWasm &Symbol) {
1258   assert(Symbol.isFunction());
1259   assert(TypeIndices.count(&Symbol));
1260   return TypeIndices[&Symbol];
1261 }
1262 
getTagType(const MCSymbolWasm & Symbol)1263 uint32_t WasmObjectWriter::getTagType(const MCSymbolWasm &Symbol) {
1264   assert(Symbol.isTag());
1265   assert(TypeIndices.count(&Symbol));
1266   return TypeIndices[&Symbol];
1267 }
1268 
registerFunctionType(const MCSymbolWasm & Symbol)1269 void WasmObjectWriter::registerFunctionType(const MCSymbolWasm &Symbol) {
1270   assert(Symbol.isFunction());
1271 
1272   wasm::WasmSignature S;
1273 
1274   if (auto *Sig = Symbol.getSignature()) {
1275     S.Returns = Sig->Returns;
1276     S.Params = Sig->Params;
1277   }
1278 
1279   auto Pair = SignatureIndices.insert(std::make_pair(S, Signatures.size()));
1280   if (Pair.second)
1281     Signatures.push_back(S);
1282   TypeIndices[&Symbol] = Pair.first->second;
1283 
1284   LLVM_DEBUG(dbgs() << "registerFunctionType: " << Symbol
1285                     << " new:" << Pair.second << "\n");
1286   LLVM_DEBUG(dbgs() << "  -> type index: " << Pair.first->second << "\n");
1287 }
1288 
registerTagType(const MCSymbolWasm & Symbol)1289 void WasmObjectWriter::registerTagType(const MCSymbolWasm &Symbol) {
1290   assert(Symbol.isTag());
1291 
1292   // TODO Currently we don't generate imported exceptions, but if we do, we
1293   // should have a way of infering types of imported exceptions.
1294   wasm::WasmSignature S;
1295   if (auto *Sig = Symbol.getSignature()) {
1296     S.Returns = Sig->Returns;
1297     S.Params = Sig->Params;
1298   }
1299 
1300   auto Pair = SignatureIndices.insert(std::make_pair(S, Signatures.size()));
1301   if (Pair.second)
1302     Signatures.push_back(S);
1303   TypeIndices[&Symbol] = Pair.first->second;
1304 
1305   LLVM_DEBUG(dbgs() << "registerTagType: " << Symbol << " new:" << Pair.second
1306                     << "\n");
1307   LLVM_DEBUG(dbgs() << "  -> type index: " << Pair.first->second << "\n");
1308 }
1309 
isInSymtab(const MCSymbolWasm & Sym)1310 static bool isInSymtab(const MCSymbolWasm &Sym) {
1311   if (Sym.isUsedInReloc() || Sym.isUsedInInitArray())
1312     return true;
1313 
1314   if (Sym.isComdat() && !Sym.isDefined())
1315     return false;
1316 
1317   if (Sym.isTemporary())
1318     return false;
1319 
1320   if (Sym.isSection())
1321     return false;
1322 
1323   if (Sym.omitFromLinkingSection())
1324     return false;
1325 
1326   return true;
1327 }
1328 
prepareImports(SmallVectorImpl<wasm::WasmImport> & Imports,MCAssembler & Asm)1329 void WasmObjectWriter::prepareImports(
1330     SmallVectorImpl<wasm::WasmImport> &Imports, MCAssembler &Asm) {
1331   // For now, always emit the memory import, since loads and stores are not
1332   // valid without it. In the future, we could perhaps be more clever and omit
1333   // it if there are no loads or stores.
1334   wasm::WasmImport MemImport;
1335   MemImport.Module = "env";
1336   MemImport.Field = "__linear_memory";
1337   MemImport.Kind = wasm::WASM_EXTERNAL_MEMORY;
1338   MemImport.Memory.Flags = is64Bit() ? wasm::WASM_LIMITS_FLAG_IS_64
1339                                      : wasm::WASM_LIMITS_FLAG_NONE;
1340   Imports.push_back(MemImport);
1341 
1342   // Populate SignatureIndices, and Imports and WasmIndices for undefined
1343   // symbols.  This must be done before populating WasmIndices for defined
1344   // symbols.
1345   for (const MCSymbol &S : Asm.symbols()) {
1346     const auto &WS = static_cast<const MCSymbolWasm &>(S);
1347 
1348     // Register types for all functions, including those with private linkage
1349     // (because wasm always needs a type signature).
1350     if (WS.isFunction()) {
1351       const auto *BS = Asm.getBaseSymbol(S);
1352       if (!BS)
1353         report_fatal_error(Twine(S.getName()) +
1354                            ": absolute addressing not supported!");
1355       registerFunctionType(*cast<MCSymbolWasm>(BS));
1356     }
1357 
1358     if (WS.isTag())
1359       registerTagType(WS);
1360 
1361     if (WS.isTemporary())
1362       continue;
1363 
1364     // If the symbol is not defined in this translation unit, import it.
1365     if (!WS.isDefined() && !WS.isComdat()) {
1366       if (WS.isFunction()) {
1367         wasm::WasmImport Import;
1368         Import.Module = WS.getImportModule();
1369         Import.Field = WS.getImportName();
1370         Import.Kind = wasm::WASM_EXTERNAL_FUNCTION;
1371         Import.SigIndex = getFunctionType(WS);
1372         Imports.push_back(Import);
1373         assert(WasmIndices.count(&WS) == 0);
1374         WasmIndices[&WS] = NumFunctionImports++;
1375       } else if (WS.isGlobal()) {
1376         if (WS.isWeak())
1377           report_fatal_error("undefined global symbol cannot be weak");
1378 
1379         wasm::WasmImport Import;
1380         Import.Field = WS.getImportName();
1381         Import.Kind = wasm::WASM_EXTERNAL_GLOBAL;
1382         Import.Module = WS.getImportModule();
1383         Import.Global = WS.getGlobalType();
1384         Imports.push_back(Import);
1385         assert(WasmIndices.count(&WS) == 0);
1386         WasmIndices[&WS] = NumGlobalImports++;
1387       } else if (WS.isTag()) {
1388         if (WS.isWeak())
1389           report_fatal_error("undefined tag symbol cannot be weak");
1390 
1391         wasm::WasmImport Import;
1392         Import.Module = WS.getImportModule();
1393         Import.Field = WS.getImportName();
1394         Import.Kind = wasm::WASM_EXTERNAL_TAG;
1395         Import.SigIndex = getTagType(WS);
1396         Imports.push_back(Import);
1397         assert(WasmIndices.count(&WS) == 0);
1398         WasmIndices[&WS] = NumTagImports++;
1399       } else if (WS.isTable()) {
1400         if (WS.isWeak())
1401           report_fatal_error("undefined table symbol cannot be weak");
1402 
1403         wasm::WasmImport Import;
1404         Import.Module = WS.getImportModule();
1405         Import.Field = WS.getImportName();
1406         Import.Kind = wasm::WASM_EXTERNAL_TABLE;
1407         Import.Table = WS.getTableType();
1408         Imports.push_back(Import);
1409         assert(WasmIndices.count(&WS) == 0);
1410         WasmIndices[&WS] = NumTableImports++;
1411       }
1412     }
1413   }
1414 
1415   // Add imports for GOT globals
1416   for (const MCSymbol &S : Asm.symbols()) {
1417     const auto &WS = static_cast<const MCSymbolWasm &>(S);
1418     if (WS.isUsedInGOT()) {
1419       wasm::WasmImport Import;
1420       if (WS.isFunction())
1421         Import.Module = "GOT.func";
1422       else
1423         Import.Module = "GOT.mem";
1424       Import.Field = WS.getName();
1425       Import.Kind = wasm::WASM_EXTERNAL_GLOBAL;
1426       Import.Global = {wasm::WASM_TYPE_I32, true};
1427       Imports.push_back(Import);
1428       assert(GOTIndices.count(&WS) == 0);
1429       GOTIndices[&WS] = NumGlobalImports++;
1430     }
1431   }
1432 }
1433 
writeObject(MCAssembler & Asm)1434 uint64_t WasmObjectWriter::writeObject(MCAssembler &Asm) {
1435   support::endian::Writer MainWriter(*OS, llvm::endianness::little);
1436   W = &MainWriter;
1437   if (IsSplitDwarf) {
1438     uint64_t TotalSize = writeOneObject(Asm, DwoMode::NonDwoOnly);
1439     assert(DwoOS);
1440     support::endian::Writer DwoWriter(*DwoOS, llvm::endianness::little);
1441     W = &DwoWriter;
1442     return TotalSize + writeOneObject(Asm, DwoMode::DwoOnly);
1443   } else {
1444     return writeOneObject(Asm, DwoMode::AllSections);
1445   }
1446 }
1447 
writeOneObject(MCAssembler & Asm,DwoMode Mode)1448 uint64_t WasmObjectWriter::writeOneObject(MCAssembler &Asm,
1449                                           DwoMode Mode) {
1450   uint64_t StartOffset = W->OS.tell();
1451   SectionCount = 0;
1452   CustomSections.clear();
1453 
1454   LLVM_DEBUG(dbgs() << "WasmObjectWriter::writeObject\n");
1455 
1456   // Collect information from the available symbols.
1457   SmallVector<WasmFunction, 4> Functions;
1458   SmallVector<uint32_t, 4> TableElems;
1459   SmallVector<wasm::WasmImport, 4> Imports;
1460   SmallVector<wasm::WasmExport, 4> Exports;
1461   SmallVector<uint32_t, 2> TagTypes;
1462   SmallVector<wasm::WasmGlobal, 1> Globals;
1463   SmallVector<wasm::WasmTable, 1> Tables;
1464   SmallVector<wasm::WasmSymbolInfo, 4> SymbolInfos;
1465   SmallVector<std::pair<uint16_t, uint32_t>, 2> InitFuncs;
1466   std::map<StringRef, std::vector<WasmComdatEntry>> Comdats;
1467   uint64_t DataSize = 0;
1468   if (Mode != DwoMode::DwoOnly)
1469     prepareImports(Imports, Asm);
1470 
1471   // Populate DataSegments and CustomSections, which must be done before
1472   // populating DataLocations.
1473   for (MCSection &Sec : Asm) {
1474     auto &Section = static_cast<MCSectionWasm &>(Sec);
1475     StringRef SectionName = Section.getName();
1476 
1477     if (Mode == DwoMode::NonDwoOnly && isDwoSection(Sec))
1478       continue;
1479     if (Mode == DwoMode::DwoOnly && !isDwoSection(Sec))
1480       continue;
1481 
1482     LLVM_DEBUG(dbgs() << "Processing Section " << SectionName << "  group "
1483                       << Section.getGroup() << "\n";);
1484 
1485     // .init_array sections are handled specially elsewhere.
1486     if (SectionName.starts_with(".init_array"))
1487       continue;
1488 
1489     // Code is handled separately
1490     if (Section.isText())
1491       continue;
1492 
1493     if (Section.isWasmData()) {
1494       uint32_t SegmentIndex = DataSegments.size();
1495       DataSize = alignTo(DataSize, Section.getAlign());
1496       DataSegments.emplace_back();
1497       WasmDataSegment &Segment = DataSegments.back();
1498       Segment.Name = SectionName;
1499       Segment.InitFlags = Section.getPassive()
1500                               ? (uint32_t)wasm::WASM_DATA_SEGMENT_IS_PASSIVE
1501                               : 0;
1502       Segment.Offset = DataSize;
1503       Segment.Section = &Section;
1504       addData(Segment.Data, Section);
1505       Segment.Alignment = Log2(Section.getAlign());
1506       Segment.LinkingFlags = Section.getSegmentFlags();
1507       DataSize += Segment.Data.size();
1508       Section.setSegmentIndex(SegmentIndex);
1509 
1510       if (const MCSymbolWasm *C = Section.getGroup()) {
1511         Comdats[C->getName()].emplace_back(
1512             WasmComdatEntry{wasm::WASM_COMDAT_DATA, SegmentIndex});
1513       }
1514     } else {
1515       // Create custom sections
1516       assert(Section.isMetadata());
1517 
1518       StringRef Name = SectionName;
1519 
1520       // For user-defined custom sections, strip the prefix
1521       Name.consume_front(".custom_section.");
1522 
1523       MCSymbol *Begin = Sec.getBeginSymbol();
1524       if (Begin) {
1525         assert(WasmIndices.count(cast<MCSymbolWasm>(Begin)) == 0);
1526         WasmIndices[cast<MCSymbolWasm>(Begin)] = CustomSections.size();
1527       }
1528 
1529       // Separate out the producers and target features sections
1530       if (Name == "producers") {
1531         ProducersSection = std::make_unique<WasmCustomSection>(Name, &Section);
1532         continue;
1533       }
1534       if (Name == "target_features") {
1535         TargetFeaturesSection =
1536             std::make_unique<WasmCustomSection>(Name, &Section);
1537         continue;
1538       }
1539 
1540       // Custom sections can also belong to COMDAT groups. In this case the
1541       // decriptor's "index" field is the section index (in the final object
1542       // file), but that is not known until after layout, so it must be fixed up
1543       // later
1544       if (const MCSymbolWasm *C = Section.getGroup()) {
1545         Comdats[C->getName()].emplace_back(
1546             WasmComdatEntry{wasm::WASM_COMDAT_SECTION,
1547                             static_cast<uint32_t>(CustomSections.size())});
1548       }
1549 
1550       CustomSections.emplace_back(Name, &Section);
1551     }
1552   }
1553 
1554   if (Mode != DwoMode::DwoOnly) {
1555     // Populate WasmIndices and DataLocations for defined symbols.
1556     for (const MCSymbol &S : Asm.symbols()) {
1557       // Ignore unnamed temporary symbols, which aren't ever exported, imported,
1558       // or used in relocations.
1559       if (S.isTemporary() && S.getName().empty())
1560         continue;
1561 
1562       const auto &WS = static_cast<const MCSymbolWasm &>(S);
1563       LLVM_DEBUG(
1564           dbgs() << "MCSymbol: "
1565                  << toString(WS.getType().value_or(wasm::WASM_SYMBOL_TYPE_DATA))
1566                  << " '" << S << "'"
1567                  << " isDefined=" << S.isDefined() << " isExternal="
1568                  << S.isExternal() << " isTemporary=" << S.isTemporary()
1569                  << " isWeak=" << WS.isWeak() << " isHidden=" << WS.isHidden()
1570                  << " isVariable=" << WS.isVariable() << "\n");
1571 
1572       if (WS.isVariable())
1573         continue;
1574       if (WS.isComdat() && !WS.isDefined())
1575         continue;
1576 
1577       if (WS.isFunction()) {
1578         unsigned Index;
1579         if (WS.isDefined()) {
1580           if (WS.getOffset() != 0)
1581             report_fatal_error(
1582                 "function sections must contain one function each");
1583 
1584           // A definition. Write out the function body.
1585           Index = NumFunctionImports + Functions.size();
1586           WasmFunction Func;
1587           Func.SigIndex = getFunctionType(WS);
1588           Func.Section = &WS.getSection();
1589           assert(WasmIndices.count(&WS) == 0);
1590           WasmIndices[&WS] = Index;
1591           Functions.push_back(Func);
1592 
1593           auto &Section = static_cast<MCSectionWasm &>(WS.getSection());
1594           if (const MCSymbolWasm *C = Section.getGroup()) {
1595             Comdats[C->getName()].emplace_back(
1596                 WasmComdatEntry{wasm::WASM_COMDAT_FUNCTION, Index});
1597           }
1598 
1599           if (WS.hasExportName()) {
1600             wasm::WasmExport Export;
1601             Export.Name = WS.getExportName();
1602             Export.Kind = wasm::WASM_EXTERNAL_FUNCTION;
1603             Export.Index = Index;
1604             Exports.push_back(Export);
1605           }
1606         } else {
1607           // An import; the index was assigned above.
1608           Index = WasmIndices.find(&WS)->second;
1609         }
1610 
1611         LLVM_DEBUG(dbgs() << "  -> function index: " << Index << "\n");
1612 
1613       } else if (WS.isData()) {
1614         if (!isInSymtab(WS))
1615           continue;
1616 
1617         if (!WS.isDefined()) {
1618           LLVM_DEBUG(dbgs() << "  -> segment index: -1"
1619                             << "\n");
1620           continue;
1621         }
1622 
1623         if (!WS.getSize())
1624           report_fatal_error("data symbols must have a size set with .size: " +
1625                              WS.getName());
1626 
1627         int64_t Size = 0;
1628         if (!WS.getSize()->evaluateAsAbsolute(Size, Asm))
1629           report_fatal_error(".size expression must be evaluatable");
1630 
1631         auto &DataSection = static_cast<MCSectionWasm &>(WS.getSection());
1632         if (!DataSection.isWasmData())
1633           report_fatal_error("data symbols must live in a data section: " +
1634                              WS.getName());
1635 
1636         // For each data symbol, export it in the symtab as a reference to the
1637         // corresponding Wasm data segment.
1638         wasm::WasmDataReference Ref = wasm::WasmDataReference{
1639             DataSection.getSegmentIndex(), Asm.getSymbolOffset(WS),
1640             static_cast<uint64_t>(Size)};
1641         assert(DataLocations.count(&WS) == 0);
1642         DataLocations[&WS] = Ref;
1643         LLVM_DEBUG(dbgs() << "  -> segment index: " << Ref.Segment << "\n");
1644 
1645       } else if (WS.isGlobal()) {
1646         // A "true" Wasm global (currently just __stack_pointer)
1647         if (WS.isDefined()) {
1648           wasm::WasmGlobal Global;
1649           Global.Type = WS.getGlobalType();
1650           Global.Index = NumGlobalImports + Globals.size();
1651           Global.InitExpr.Extended = false;
1652           switch (Global.Type.Type) {
1653           case wasm::WASM_TYPE_I32:
1654             Global.InitExpr.Inst.Opcode = wasm::WASM_OPCODE_I32_CONST;
1655             break;
1656           case wasm::WASM_TYPE_I64:
1657             Global.InitExpr.Inst.Opcode = wasm::WASM_OPCODE_I64_CONST;
1658             break;
1659           case wasm::WASM_TYPE_F32:
1660             Global.InitExpr.Inst.Opcode = wasm::WASM_OPCODE_F32_CONST;
1661             break;
1662           case wasm::WASM_TYPE_F64:
1663             Global.InitExpr.Inst.Opcode = wasm::WASM_OPCODE_F64_CONST;
1664             break;
1665           case wasm::WASM_TYPE_EXTERNREF:
1666             Global.InitExpr.Inst.Opcode = wasm::WASM_OPCODE_REF_NULL;
1667             break;
1668           default:
1669             llvm_unreachable("unexpected type");
1670           }
1671           assert(WasmIndices.count(&WS) == 0);
1672           WasmIndices[&WS] = Global.Index;
1673           Globals.push_back(Global);
1674         } else {
1675           // An import; the index was assigned above
1676           LLVM_DEBUG(dbgs() << "  -> global index: "
1677                             << WasmIndices.find(&WS)->second << "\n");
1678         }
1679       } else if (WS.isTable()) {
1680         if (WS.isDefined()) {
1681           wasm::WasmTable Table;
1682           Table.Index = NumTableImports + Tables.size();
1683           Table.Type = WS.getTableType();
1684           assert(WasmIndices.count(&WS) == 0);
1685           WasmIndices[&WS] = Table.Index;
1686           Tables.push_back(Table);
1687         }
1688         LLVM_DEBUG(dbgs() << " -> table index: "
1689                           << WasmIndices.find(&WS)->second << "\n");
1690       } else if (WS.isTag()) {
1691         // C++ exception symbol (__cpp_exception) or longjmp symbol
1692         // (__c_longjmp)
1693         unsigned Index;
1694         if (WS.isDefined()) {
1695           Index = NumTagImports + TagTypes.size();
1696           uint32_t SigIndex = getTagType(WS);
1697           assert(WasmIndices.count(&WS) == 0);
1698           WasmIndices[&WS] = Index;
1699           TagTypes.push_back(SigIndex);
1700         } else {
1701           // An import; the index was assigned above.
1702           assert(WasmIndices.count(&WS) > 0);
1703         }
1704         LLVM_DEBUG(dbgs() << "  -> tag index: " << WasmIndices.find(&WS)->second
1705                           << "\n");
1706 
1707       } else {
1708         assert(WS.isSection());
1709       }
1710     }
1711 
1712     // Populate WasmIndices and DataLocations for aliased symbols.  We need to
1713     // process these in a separate pass because we need to have processed the
1714     // target of the alias before the alias itself and the symbols are not
1715     // necessarily ordered in this way.
1716     for (const MCSymbol &S : Asm.symbols()) {
1717       if (!S.isVariable())
1718         continue;
1719 
1720       assert(S.isDefined());
1721 
1722       const auto *BS = Asm.getBaseSymbol(S);
1723       if (!BS)
1724         report_fatal_error(Twine(S.getName()) +
1725                            ": absolute addressing not supported!");
1726       const MCSymbolWasm *Base = cast<MCSymbolWasm>(BS);
1727 
1728       // Find the target symbol of this weak alias and export that index
1729       const auto &WS = static_cast<const MCSymbolWasm &>(S);
1730       LLVM_DEBUG(dbgs() << WS.getName() << ": weak alias of '" << *Base
1731                         << "'\n");
1732 
1733       if (Base->isFunction()) {
1734         assert(WasmIndices.count(Base) > 0);
1735         uint32_t WasmIndex = WasmIndices.find(Base)->second;
1736         assert(WasmIndices.count(&WS) == 0);
1737         WasmIndices[&WS] = WasmIndex;
1738         LLVM_DEBUG(dbgs() << "  -> index:" << WasmIndex << "\n");
1739       } else if (Base->isData()) {
1740         auto &DataSection = static_cast<MCSectionWasm &>(WS.getSection());
1741         uint64_t Offset = Asm.getSymbolOffset(S);
1742         int64_t Size = 0;
1743         // For data symbol alias we use the size of the base symbol as the
1744         // size of the alias.  When an offset from the base is involved this
1745         // can result in a offset + size goes past the end of the data section
1746         // which out object format doesn't support.  So we must clamp it.
1747         if (!Base->getSize()->evaluateAsAbsolute(Size, Asm))
1748           report_fatal_error(".size expression must be evaluatable");
1749         const WasmDataSegment &Segment =
1750             DataSegments[DataSection.getSegmentIndex()];
1751         Size =
1752             std::min(static_cast<uint64_t>(Size), Segment.Data.size() - Offset);
1753         wasm::WasmDataReference Ref = wasm::WasmDataReference{
1754             DataSection.getSegmentIndex(),
1755             static_cast<uint32_t>(Asm.getSymbolOffset(S)),
1756             static_cast<uint32_t>(Size)};
1757         DataLocations[&WS] = Ref;
1758         LLVM_DEBUG(dbgs() << "  -> index:" << Ref.Segment << "\n");
1759       } else {
1760         report_fatal_error("don't yet support global/tag aliases");
1761       }
1762     }
1763   }
1764 
1765   // Finally, populate the symbol table itself, in its "natural" order.
1766   for (const MCSymbol &S : Asm.symbols()) {
1767     const auto &WS = static_cast<const MCSymbolWasm &>(S);
1768     if (!isInSymtab(WS)) {
1769       WS.setIndex(InvalidIndex);
1770       continue;
1771     }
1772     LLVM_DEBUG(dbgs() << "adding to symtab: " << WS << "\n");
1773 
1774     uint32_t Flags = 0;
1775     if (WS.isWeak())
1776       Flags |= wasm::WASM_SYMBOL_BINDING_WEAK;
1777     if (WS.isHidden())
1778       Flags |= wasm::WASM_SYMBOL_VISIBILITY_HIDDEN;
1779     if (!WS.isExternal() && WS.isDefined())
1780       Flags |= wasm::WASM_SYMBOL_BINDING_LOCAL;
1781     if (WS.isUndefined())
1782       Flags |= wasm::WASM_SYMBOL_UNDEFINED;
1783     if (WS.isNoStrip()) {
1784       Flags |= wasm::WASM_SYMBOL_NO_STRIP;
1785       if (isEmscripten()) {
1786         Flags |= wasm::WASM_SYMBOL_EXPORTED;
1787       }
1788     }
1789     if (WS.hasImportName())
1790       Flags |= wasm::WASM_SYMBOL_EXPLICIT_NAME;
1791     if (WS.hasExportName())
1792       Flags |= wasm::WASM_SYMBOL_EXPORTED;
1793     if (WS.isTLS())
1794       Flags |= wasm::WASM_SYMBOL_TLS;
1795 
1796     wasm::WasmSymbolInfo Info;
1797     Info.Name = WS.getName();
1798     Info.Kind = WS.getType().value_or(wasm::WASM_SYMBOL_TYPE_DATA);
1799     Info.Flags = Flags;
1800     if (!WS.isData()) {
1801       assert(WasmIndices.count(&WS) > 0);
1802       Info.ElementIndex = WasmIndices.find(&WS)->second;
1803     } else if (WS.isDefined()) {
1804       assert(DataLocations.count(&WS) > 0);
1805       Info.DataRef = DataLocations.find(&WS)->second;
1806     }
1807     WS.setIndex(SymbolInfos.size());
1808     SymbolInfos.emplace_back(Info);
1809   }
1810 
1811   {
1812     auto HandleReloc = [&](const WasmRelocationEntry &Rel) {
1813       // Functions referenced by a relocation need to put in the table.  This is
1814       // purely to make the object file's provisional values readable, and is
1815       // ignored by the linker, which re-calculates the relocations itself.
1816       if (Rel.Type != wasm::R_WASM_TABLE_INDEX_I32 &&
1817           Rel.Type != wasm::R_WASM_TABLE_INDEX_I64 &&
1818           Rel.Type != wasm::R_WASM_TABLE_INDEX_SLEB &&
1819           Rel.Type != wasm::R_WASM_TABLE_INDEX_SLEB64 &&
1820           Rel.Type != wasm::R_WASM_TABLE_INDEX_REL_SLEB &&
1821           Rel.Type != wasm::R_WASM_TABLE_INDEX_REL_SLEB64)
1822         return;
1823       assert(Rel.Symbol->isFunction());
1824       const MCSymbolWasm *Base =
1825           cast<MCSymbolWasm>(Asm.getBaseSymbol(*Rel.Symbol));
1826       uint32_t FunctionIndex = WasmIndices.find(Base)->second;
1827       uint32_t TableIndex = TableElems.size() + InitialTableOffset;
1828       if (TableIndices.try_emplace(Base, TableIndex).second) {
1829         LLVM_DEBUG(dbgs() << "  -> adding " << Base->getName()
1830                           << " to table: " << TableIndex << "\n");
1831         TableElems.push_back(FunctionIndex);
1832         registerFunctionType(*Base);
1833       }
1834     };
1835 
1836     for (const WasmRelocationEntry &RelEntry : CodeRelocations)
1837       HandleReloc(RelEntry);
1838     for (const WasmRelocationEntry &RelEntry : DataRelocations)
1839       HandleReloc(RelEntry);
1840   }
1841 
1842   // Translate .init_array section contents into start functions.
1843   for (const MCSection &S : Asm) {
1844     const auto &WS = static_cast<const MCSectionWasm &>(S);
1845     if (WS.getName().starts_with(".fini_array"))
1846       report_fatal_error(".fini_array sections are unsupported");
1847     if (!WS.getName().starts_with(".init_array"))
1848       continue;
1849     auto IT = WS.begin();
1850     if (IT == WS.end())
1851       continue;
1852     const MCFragment &EmptyFrag = *IT;
1853     if (EmptyFrag.getKind() != MCFragment::FT_Data)
1854       report_fatal_error(".init_array section should be aligned");
1855 
1856     const MCFragment &AlignFrag = *EmptyFrag.getNext();
1857     if (AlignFrag.getKind() != MCFragment::FT_Align)
1858       report_fatal_error(".init_array section should be aligned");
1859     if (cast<MCAlignFragment>(AlignFrag).getAlignment() !=
1860         Align(is64Bit() ? 8 : 4))
1861       report_fatal_error(".init_array section should be aligned for pointers");
1862 
1863     const MCFragment &Frag = *AlignFrag.getNext();
1864     if (Frag.hasInstructions() || Frag.getKind() != MCFragment::FT_Data)
1865       report_fatal_error("only data supported in .init_array section");
1866 
1867     uint16_t Priority = UINT16_MAX;
1868     unsigned PrefixLength = strlen(".init_array");
1869     if (WS.getName().size() > PrefixLength) {
1870       if (WS.getName()[PrefixLength] != '.')
1871         report_fatal_error(
1872             ".init_array section priority should start with '.'");
1873       if (WS.getName().substr(PrefixLength + 1).getAsInteger(10, Priority))
1874         report_fatal_error("invalid .init_array section priority");
1875     }
1876     const auto &DataFrag = cast<MCDataFragment>(Frag);
1877     const SmallVectorImpl<char> &Contents = DataFrag.getContents();
1878     for (const uint8_t *
1879              P = (const uint8_t *)Contents.data(),
1880             *End = (const uint8_t *)Contents.data() + Contents.size();
1881          P != End; ++P) {
1882       if (*P != 0)
1883         report_fatal_error("non-symbolic data in .init_array section");
1884     }
1885     for (const MCFixup &Fixup : DataFrag.getFixups()) {
1886       assert(Fixup.getKind() ==
1887              MCFixup::getKindForSize(is64Bit() ? 8 : 4, false));
1888       const MCExpr *Expr = Fixup.getValue();
1889       auto *SymRef = dyn_cast<MCSymbolRefExpr>(Expr);
1890       if (!SymRef)
1891         report_fatal_error("fixups in .init_array should be symbol references");
1892       const auto &TargetSym = cast<const MCSymbolWasm>(SymRef->getSymbol());
1893       if (TargetSym.getIndex() == InvalidIndex)
1894         report_fatal_error("symbols in .init_array should exist in symtab");
1895       if (!TargetSym.isFunction())
1896         report_fatal_error("symbols in .init_array should be for functions");
1897       InitFuncs.push_back(
1898           std::make_pair(Priority, TargetSym.getIndex()));
1899     }
1900   }
1901 
1902   // Write out the Wasm header.
1903   writeHeader(Asm);
1904 
1905   uint32_t CodeSectionIndex, DataSectionIndex;
1906   if (Mode != DwoMode::DwoOnly) {
1907     writeTypeSection(Signatures);
1908     writeImportSection(Imports, DataSize, TableElems.size());
1909     writeFunctionSection(Functions);
1910     writeTableSection(Tables);
1911     // Skip the "memory" section; we import the memory instead.
1912     writeTagSection(TagTypes);
1913     writeGlobalSection(Globals);
1914     writeExportSection(Exports);
1915     const MCSymbol *IndirectFunctionTable =
1916         Asm.getContext().lookupSymbol("__indirect_function_table");
1917     writeElemSection(cast_or_null<const MCSymbolWasm>(IndirectFunctionTable),
1918                      TableElems);
1919     writeDataCountSection();
1920 
1921     CodeSectionIndex = writeCodeSection(Asm, Functions);
1922     DataSectionIndex = writeDataSection(Asm);
1923   }
1924 
1925   // The Sections in the COMDAT list have placeholder indices (their index among
1926   // custom sections, rather than among all sections). Fix them up here.
1927   for (auto &Group : Comdats) {
1928     for (auto &Entry : Group.second) {
1929       if (Entry.Kind == wasm::WASM_COMDAT_SECTION) {
1930         Entry.Index += SectionCount;
1931       }
1932     }
1933   }
1934   for (auto &CustomSection : CustomSections)
1935     writeCustomSection(CustomSection, Asm);
1936 
1937   if (Mode != DwoMode::DwoOnly) {
1938     writeLinkingMetaDataSection(SymbolInfos, InitFuncs, Comdats);
1939 
1940     writeRelocSection(CodeSectionIndex, "CODE", CodeRelocations);
1941     writeRelocSection(DataSectionIndex, "DATA", DataRelocations);
1942   }
1943   writeCustomRelocSections();
1944   if (ProducersSection)
1945     writeCustomSection(*ProducersSection, Asm);
1946   if (TargetFeaturesSection)
1947     writeCustomSection(*TargetFeaturesSection, Asm);
1948 
1949   // TODO: Translate the .comment section to the output.
1950   return W->OS.tell() - StartOffset;
1951 }
1952 
1953 std::unique_ptr<MCObjectWriter>
createWasmObjectWriter(std::unique_ptr<MCWasmObjectTargetWriter> MOTW,raw_pwrite_stream & OS)1954 llvm::createWasmObjectWriter(std::unique_ptr<MCWasmObjectTargetWriter> MOTW,
1955                              raw_pwrite_stream &OS) {
1956   return std::make_unique<WasmObjectWriter>(std::move(MOTW), OS);
1957 }
1958 
1959 std::unique_ptr<MCObjectWriter>
createWasmDwoObjectWriter(std::unique_ptr<MCWasmObjectTargetWriter> MOTW,raw_pwrite_stream & OS,raw_pwrite_stream & DwoOS)1960 llvm::createWasmDwoObjectWriter(std::unique_ptr<MCWasmObjectTargetWriter> MOTW,
1961                                 raw_pwrite_stream &OS,
1962                                 raw_pwrite_stream &DwoOS) {
1963   return std::make_unique<WasmObjectWriter>(std::move(MOTW), OS, DwoOS);
1964 }
1965