lld/wasm/InputChunks.h

   1 //===- InputChunks.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 // An InputChunks represents an indivisible opaque region of a input wasm file.
  10 // i.e. a single wasm data segment or a single wasm function.
  11 //
  12 // They are written directly to the mmap'd output file after which relocations
  13 // are applied.  Because each Chunk is independent they can be written in
  14 // parallel.
  15 //
  16 // Chunks are also unit on which garbage collection (--gc-sections) operates.
  17 //
  18 //===----------------------------------------------------------------------===//
  19
  20 #ifndef LLD_WASM_INPUT_CHUNKS_H
  21 #define LLD_WASM_INPUT_CHUNKS_H
  22
  23 #include "Config.h"
  24 #include "InputFiles.h"
  25 #include "lld/Common/ErrorHandler.h"
  26 #include "lld/Common/LLVM.h"
  27 #include "llvm/ADT/CachedHashString.h"
  28 #include "llvm/MC/StringTableBuilder.h"
  29 #include "llvm/Object/Wasm.h"
  30 #include <optional>
  31
  32 namespace lld {
  33 namespace wasm {
  34
  35 class ObjFile;
  36 class OutputSegment;
  37 class OutputSection;
  38
  39 class InputChunk {
  40 public:
  41   enum Kind {
  42     DataSegment,
  43     Merge,
  44     MergedChunk,
  45     Function,
  46     SyntheticFunction,
  47     Section,
  48   };
  49
  50   StringRef name;
  51   StringRef debugName;
  52
  53   Kind kind() const { return (Kind)sectionKind; }
  54
  55   uint32_t getSize() const;
  56   uint32_t getInputSize() const;
  57
  58   void writeTo(uint8_t *buf) const;
  59   void relocate(uint8_t *buf) const;
  60
  61   ArrayRef<WasmRelocation> getRelocations() const { return relocations; }
  62   void setRelocations(ArrayRef<WasmRelocation> rs) { relocations = rs; }
  63
  64   // Translate an offset into the input chunk to an offset in the output
  65   // section.
  66   uint64_t getOffset(uint64_t offset) const;
  67   // Translate an offset into the input chunk into an offset into the output
  68   // chunk.  For data segments (InputSegment) this will return and offset into
  69   // the output segment.  For MergeInputChunk, this will return an offset into
  70   // the parent merged chunk.  For other chunk types this is no-op and we just
  71   // return unmodified offset.
  72   uint64_t getChunkOffset(uint64_t offset) const;
  73   uint64_t getVA(uint64_t offset = 0) const;
  74
  75   uint32_t getComdat() const { return comdat; }
  76   StringRef getComdatName() const;
  77   uint32_t getInputSectionOffset() const { return inputSectionOffset; }
  78
  79   size_t getNumRelocations() const { return relocations.size(); }
  80   void writeRelocations(llvm::raw_ostream &os) const;
  81   void generateRelocationCode(raw_ostream &os) const;
  82
  83   bool isTLS() const { return flags & llvm::wasm::WASM_SEG_FLAG_TLS; }
  84
  85   ObjFile *file;
  86   OutputSection *outputSec = nullptr;
  87   uint32_t comdat = UINT32_MAX;
  88   uint32_t inputSectionOffset = 0;
  89   uint32_t alignment;
  90   uint32_t flags;
  91
  92   // Only applies to data segments.
  93   uint32_t outputSegmentOffset = 0;
  94   const OutputSegment *outputSeg = nullptr;
  95
  96   // After assignAddresses is called, this represents the offset from
  97   // the beginning of the output section this chunk was assigned to.
  98   int32_t outSecOff = 0;
  99
 100   uint8_t sectionKind : 3;
 101
 102   // Signals that the section is part of the output.  The garbage collector,
 103   // and COMDAT handling can set a sections' Live bit.
 104   // If GC is disabled, all sections start out as live by default.
 105   unsigned live : 1;
 106
 107   // Signals the chunk was discarded by COMDAT handling.
 108   unsigned discarded : 1;
 109
 110 protected:
 111   InputChunk(ObjFile *f, Kind k, StringRef name, uint32_t alignment = 0,
 112              uint32_t flags = 0)
 113       : name(name), file(f), alignment(alignment), flags(flags), sectionKind(k),
 114         live(!config->gcSections), discarded(false) {}
 115   ArrayRef<uint8_t> data() const { return rawData; }
 116   uint64_t getTombstone() const;
 117
 118   ArrayRef<WasmRelocation> relocations;
 119   ArrayRef<uint8_t> rawData;
 120 };
 121
 122 // Represents a WebAssembly data segment which can be included as part of
 123 // an output data segments.  Note that in WebAssembly, unlike ELF and other
 124 // formats, used the term "data segment" to refer to the continuous regions of
 125 // memory that make on the data section. See:
 126 // https://webassembly.github.io/spec/syntax/modules.html#syntax-data
 127 //
 128 // For example, by default, clang will produce a separate data section for
 129 // each global variable.
 130 class InputSegment : public InputChunk {
 131 public:
 132   InputSegment(const WasmSegment &seg, ObjFile *f)
 133       : InputChunk(f, InputChunk::DataSegment, seg.Data.Name,
 134                    seg.Data.Alignment, seg.Data.LinkingFlags),
 135         segment(seg) {
 136     rawData = segment.Data.Content;
 137     comdat = segment.Data.Comdat;
 138     inputSectionOffset = segment.SectionOffset;
 139   }
 140
 141   static bool classof(const InputChunk *c) { return c->kind() == DataSegment; }
 142
 143 protected:
 144   const WasmSegment &segment;
 145 };
 146
 147 class SyntheticMergedChunk;
 148
 149 // Merge segment handling copied from lld/ELF/InputSection.h.  Keep in sync
 150 // where possible.
 151
 152 // SectionPiece represents a piece of splittable segment contents.
 153 // We allocate a lot of these and binary search on them. This means that they
 154 // have to be as compact as possible, which is why we don't store the size (can
 155 // be found by looking at the next one).
 156 struct SectionPiece {
 157   SectionPiece(size_t off, uint32_t hash, bool live)
 158       : inputOff(off), live(live || !config->gcSections), hash(hash >> 1) {}
 159
 160   uint32_t inputOff;
 161   uint32_t live : 1;
 162   uint32_t hash : 31;
 163   uint64_t outputOff = 0;
 164 };
 165
 166 static_assert(sizeof(SectionPiece) == 16, "SectionPiece is too big");
 167
 168 // This corresponds segments marked as WASM_SEG_FLAG_STRINGS.
 169 class MergeInputChunk : public InputChunk {
 170 public:
 171   MergeInputChunk(const WasmSegment &seg, ObjFile *f)
 172       : InputChunk(f, Merge, seg.Data.Name, seg.Data.Alignment,
 173                    seg.Data.LinkingFlags) {
 174     rawData = seg.Data.Content;
 175     comdat = seg.Data.Comdat;
 176     inputSectionOffset = seg.SectionOffset;
 177   }
 178
 179   MergeInputChunk(const WasmSection &s, ObjFile *f)
 180       : InputChunk(f, Merge, s.Name, 0, llvm::wasm::WASM_SEG_FLAG_STRINGS) {
 181     assert(s.Type == llvm::wasm::WASM_SEC_CUSTOM);
 182     comdat = s.Comdat;
 183     rawData = s.Content;
 184   }
 185
 186   static bool classof(const InputChunk *s) { return s->kind() == Merge; }
 187   void splitIntoPieces();
 188
 189   // Translate an offset in the input section to an offset in the parent
 190   // MergeSyntheticSection.
 191   uint64_t getParentOffset(uint64_t offset) const;
 192
 193   // Splittable sections are handled as a sequence of data
 194   // rather than a single large blob of data.
 195   std::vector<SectionPiece> pieces;
 196
 197   // Returns I'th piece's data. This function is very hot when
 198   // string merging is enabled, so we want to inline.
 199   LLVM_ATTRIBUTE_ALWAYS_INLINE
 200   llvm::CachedHashStringRef getData(size_t i) const {
 201     size_t begin = pieces[i].inputOff;
 202     size_t end =
 203         (pieces.size() - 1 == i) ? data().size() : pieces[i + 1].inputOff;
 204     return {toStringRef(data().slice(begin, end - begin)), pieces[i].hash};
 205   }
 206
 207   // Returns the SectionPiece at a given input section offset.
 208   SectionPiece *getSectionPiece(uint64_t offset);
 209   const SectionPiece *getSectionPiece(uint64_t offset) const {
 210     return const_cast<MergeInputChunk *>(this)->getSectionPiece(offset);
 211   }
 212
 213   SyntheticMergedChunk *parent = nullptr;
 214
 215 private:
 216   void splitStrings(ArrayRef<uint8_t> a);
 217 };
 218
 219 // SyntheticMergedChunk is a class that allows us to put mergeable
 220 // sections with different attributes in a single output sections. To do that we
 221 // put them into SyntheticMergedChunk synthetic input sections which are
 222 // attached to regular output sections.
 223 class SyntheticMergedChunk : public InputChunk {
 224 public:
 225   SyntheticMergedChunk(StringRef name, uint32_t alignment, uint32_t flags)
 226       : InputChunk(nullptr, InputChunk::MergedChunk, name, alignment, flags),
 227         builder(llvm::StringTableBuilder::RAW, llvm::Align(1ULL << alignment)) {
 228   }
 229
 230   static bool classof(const InputChunk *c) {
 231     return c->kind() == InputChunk::MergedChunk;
 232   }
 233
 234   void addMergeChunk(MergeInputChunk *ms) {
 235     comdat = ms->getComdat();
 236     ms->parent = this;
 237     chunks.push_back(ms);
 238   }
 239
 240   void finalizeContents();
 241
 242   llvm::StringTableBuilder builder;
 243
 244 protected:
 245   std::vector<MergeInputChunk *> chunks;
 246 };
 247
 248 // Represents a single wasm function within and input file.  These are
 249 // combined to create the final output CODE section.
 250 class InputFunction : public InputChunk {
 251 public:
 252   InputFunction(const WasmSignature &s, const WasmFunction *func, ObjFile *f)
 253       : InputChunk(f, InputChunk::Function, func->SymbolName), signature(s),
 254         function(func),
 255         exportName(func && func->ExportName ? (*func->ExportName).str()
 256                                             : std::optional<std::string>()) {
 257     inputSectionOffset = function->CodeSectionOffset;
 258     rawData =
 259         file->codeSection->Content.slice(inputSectionOffset, function->Size);
 260     debugName = function->DebugName;
 261     comdat = function->Comdat;
 262   }
 263
 264   InputFunction(StringRef name, const WasmSignature &s)
 265       : InputChunk(nullptr, InputChunk::Function, name), signature(s) {}
 266
 267   static bool classof(const InputChunk *c) {
 268     return c->kind() == InputChunk::Function ||
 269            c->kind() == InputChunk::SyntheticFunction;
 270   }
 271
 272   std::optional<StringRef> getExportName() const {
 273     return exportName ? std::optional<StringRef>(*exportName)
 274                       : std::optional<StringRef>();
 275   }
 276   void setExportName(std::string exportName) { this->exportName = exportName; }
 277   uint32_t getFunctionInputOffset() const { return getInputSectionOffset(); }
 278   uint32_t getFunctionCodeOffset() const { return function->CodeOffset; }
 279   uint32_t getFunctionIndex() const { return *functionIndex; }
 280   bool hasFunctionIndex() const { return functionIndex.has_value(); }
 281   void setFunctionIndex(uint32_t index);
 282   uint32_t getTableIndex() const { return *tableIndex; }
 283   bool hasTableIndex() const { return tableIndex.has_value(); }
 284   void setTableIndex(uint32_t index);
 285   void writeCompressed(uint8_t *buf) const;
 286
 287   // The size of a given input function can depend on the values of the
 288   // LEB relocations within it.  This finalizeContents method is called after
 289   // all the symbol values have be calculated but before getSize() is ever
 290   // called.
 291   void calculateSize();
 292
 293   const WasmSignature &signature;
 294
 295   uint32_t getCompressedSize() const {
 296     assert(compressedSize);
 297     return compressedSize;
 298   }
 299
 300   const WasmFunction *function;
 301
 302 protected:
 303   std::optional<std::string> exportName;
 304   std::optional<uint32_t> functionIndex;
 305   std::optional<uint32_t> tableIndex;
 306   uint32_t compressedFuncSize = 0;
 307   uint32_t compressedSize = 0;
 308 };
 309
 310 class SyntheticFunction : public InputFunction {
 311 public:
 312   SyntheticFunction(const WasmSignature &s, StringRef name,
 313                     StringRef debugName = {})
 314       : InputFunction(name, s) {
 315     sectionKind = InputChunk::SyntheticFunction;
 316     this->debugName = debugName;
 317   }
 318
 319   static bool classof(const InputChunk *c) {
 320     return c->kind() == InputChunk::SyntheticFunction;
 321   }
 322
 323   void setBody(ArrayRef<uint8_t> body) { rawData = body; }
 324 };
 325
 326 // Represents a single Wasm Section within an input file.
 327 class InputSection : public InputChunk {
 328 public:
 329   InputSection(const WasmSection &s, ObjFile *f)
 330       : InputChunk(f, InputChunk::Section, s.Name),
 331         tombstoneValue(getTombstoneForSection(s.Name)), section(s) {
 332     assert(section.Type == llvm::wasm::WASM_SEC_CUSTOM);
 333     comdat = section.Comdat;
 334     rawData = section.Content;
 335   }
 336
 337   static bool classof(const InputChunk *c) {
 338     return c->kind() == InputChunk::Section;
 339   }
 340
 341   const uint64_t tombstoneValue;
 342
 343 protected:
 344   static uint64_t getTombstoneForSection(StringRef name);
 345   const WasmSection &section;
 346 };
 347
 348 } // namespace wasm
 349
 350 std::string toString(const wasm::InputChunk *);
 351 StringRef relocTypeToString(uint8_t relocType);
 352
 353 } // namespace lld
 354
 355 #endif // LLD_WASM_INPUT_CHUNKS_H