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
32namespace lld {
33namespace wasm {
34
35class ObjFile;
36class OutputSegment;
37class OutputSection;
38
39class InputChunk {
40public:
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 size_t getNumLiveRelocations() const;
81 void writeRelocations(llvm::raw_ostream &os) const;
82 bool generateRelocationCode(raw_ostream &os) const;
83
84 bool isTLS() const { return flags & llvm::wasm::WASM_SEG_FLAG_TLS; }
85 bool isRetained() const { return flags & llvm::wasm::WASM_SEG_FLAG_RETAIN; }
86
87 ObjFile *file;
88 OutputSection *outputSec = nullptr;
89 uint32_t comdat = UINT32_MAX;
90 uint32_t inputSectionOffset = 0;
91 uint32_t alignment;
92 uint32_t flags;
93
94 // Only applies to data segments.
95 uint32_t outputSegmentOffset = 0;
96 const OutputSegment *outputSeg = nullptr;
97
98 // After assignAddresses is called, this represents the offset from
99 // the beginning of the output section this chunk was assigned to.
100 int32_t outSecOff = 0;
101
102 uint8_t sectionKind : 3;
103
104 // Signals that the section is part of the output. The garbage collector,
105 // and COMDAT handling can set a sections' Live bit.
106 // If GC is disabled, all sections start out as live by default.
107 unsigned live : 1;
108
109 // Signals the chunk was discarded by COMDAT handling.
110 unsigned discarded : 1;
111
112protected:
113 InputChunk(ObjFile *f, Kind k, StringRef name, uint32_t alignment = 0,
114 uint32_t flags = 0)
115 : name(name), file(f), alignment(alignment), flags(flags), sectionKind(k),
116 live(!ctx.arg.gcSections), discarded(false) {}
117 ArrayRef<uint8_t> data() const { return rawData; }
118 uint64_t getTombstone() const;
119
120 ArrayRef<WasmRelocation> relocations;
121 ArrayRef<uint8_t> rawData;
122};
123
124// Represents a WebAssembly data segment which can be included as part of
125// an output data segments. Note that in WebAssembly, unlike ELF and other
126// formats, used the term "data segment" to refer to the continuous regions of
127// memory that make on the data section. See:
128// https://webassembly.github.io/spec/syntax/modules.html#syntax-data
129//
130// For example, by default, clang will produce a separate data section for
131// each global variable.
132class InputSegment : public InputChunk {
133public:
134 InputSegment(const WasmSegment &seg, ObjFile *f)
135 : InputChunk(f, InputChunk::DataSegment, seg.Data.Name,
136 seg.Data.Alignment, seg.Data.LinkingFlags),
137 segment(seg) {
138 rawData = segment.Data.Content;
139 comdat = segment.Data.Comdat;
140 inputSectionOffset = segment.SectionOffset;
141 }
142
143 static bool classof(const InputChunk *c) { return c->kind() == DataSegment; }
144
145protected:
146 const WasmSegment &segment;
147};
148
149class SyntheticMergedChunk;
150
151// Merge segment handling copied from lld/ELF/InputSection.h. Keep in sync
152// where possible.
153
154// SectionPiece represents a piece of splittable segment contents.
155// We allocate a lot of these and binary search on them. This means that they
156// have to be as compact as possible, which is why we don't store the size (can
157// be found by looking at the next one).
158struct SectionPiece {
159 SectionPiece(size_t off, uint32_t hash, bool live)
160 : inputOff(off), live(live || !ctx.arg.gcSections), hash(hash >> 1) {}
161
162 uint32_t inputOff;
163 uint32_t live : 1;
164 uint32_t hash : 31;
165 uint64_t outputOff = 0;
166};
167
168static_assert(sizeof(SectionPiece) == 16, "SectionPiece is too big");
169
170// This corresponds segments marked as WASM_SEG_FLAG_STRINGS.
171class MergeInputChunk : public InputChunk {
172public:
173 MergeInputChunk(const WasmSegment &seg, ObjFile *f)
174 : InputChunk(f, Merge, seg.Data.Name, seg.Data.Alignment,
175 seg.Data.LinkingFlags) {
176 rawData = seg.Data.Content;
177 comdat = seg.Data.Comdat;
178 inputSectionOffset = seg.SectionOffset;
179 }
180
181 MergeInputChunk(const WasmSection &s, ObjFile *f, uint32_t alignment)
182 : InputChunk(f, Merge, s.Name, alignment,
183 llvm::wasm::WASM_SEG_FLAG_STRINGS) {
184 assert(s.Type == llvm::wasm::WASM_SEC_CUSTOM);
185 comdat = s.Comdat;
186 rawData = s.Content;
187 }
188
189 static bool classof(const InputChunk *s) { return s->kind() == Merge; }
190 void splitIntoPieces();
191
192 // Translate an offset in the input section to an offset in the parent
193 // MergeSyntheticSection.
194 uint64_t getParentOffset(uint64_t offset) const;
195
196 // Splittable sections are handled as a sequence of data
197 // rather than a single large blob of data.
198 std::vector<SectionPiece> pieces;
199
200 // Returns I'th piece's data. This function is very hot when
201 // string merging is enabled, so we want to inline.
202 LLVM_ATTRIBUTE_ALWAYS_INLINE
203 llvm::CachedHashStringRef getData(size_t i) const {
204 size_t begin = pieces[i].inputOff;
205 size_t end =
206 (pieces.size() - 1 == i) ? data().size() : pieces[i + 1].inputOff;
207 return {toStringRef(Input: data().slice(N: begin, M: end - begin)), pieces[i].hash};
208 }
209
210 // Returns the SectionPiece at a given input section offset.
211 SectionPiece *getSectionPiece(uint64_t offset);
212 const SectionPiece *getSectionPiece(uint64_t offset) const {
213 return const_cast<MergeInputChunk *>(this)->getSectionPiece(offset);
214 }
215
216 SyntheticMergedChunk *parent = nullptr;
217
218private:
219 void splitStrings(ArrayRef<uint8_t> a);
220};
221
222// SyntheticMergedChunk is a class that allows us to put mergeable
223// sections with different attributes in a single output sections. To do that we
224// put them into SyntheticMergedChunk synthetic input sections which are
225// attached to regular output sections.
226class SyntheticMergedChunk : public InputChunk {
227public:
228 SyntheticMergedChunk(StringRef name, uint32_t alignment, uint32_t flags)
229 : InputChunk(nullptr, InputChunk::MergedChunk, name, alignment, flags),
230 builder(llvm::StringTableBuilder::RAW, llvm::Align(1ULL << alignment)) {
231 }
232
233 static bool classof(const InputChunk *c) {
234 return c->kind() == InputChunk::MergedChunk;
235 }
236
237 void addMergeChunk(MergeInputChunk *ms) {
238 comdat = ms->getComdat();
239 alignment = std::max(a: alignment, b: ms->alignment);
240 ms->parent = this;
241 chunks.push_back(x: ms);
242 }
243
244 void finalizeContents();
245
246 llvm::StringTableBuilder builder;
247
248protected:
249 std::vector<MergeInputChunk *> chunks;
250};
251
252// Represents a single wasm function within and input file. These are
253// combined to create the final output CODE section.
254class InputFunction : public InputChunk {
255public:
256 InputFunction(const WasmSignature &s, const WasmFunction *func, ObjFile *f)
257 : InputChunk(f, InputChunk::Function, func->SymbolName), signature(s),
258 function(func),
259 exportName(func && func->ExportName ? (*func->ExportName).str()
260 : std::optional<std::string>()) {
261 inputSectionOffset = function->CodeSectionOffset;
262 rawData =
263 file->codeSection->Content.slice(N: inputSectionOffset, M: function->Size);
264 debugName = function->DebugName;
265 comdat = function->Comdat;
266 assert(s.Kind != WasmSignature::Placeholder);
267 }
268
269 InputFunction(StringRef name, const WasmSignature &s)
270 : InputChunk(nullptr, InputChunk::Function, name), signature(s) {
271 assert(s.Kind == WasmSignature::Function);
272 }
273
274 static bool classof(const InputChunk *c) {
275 return c->kind() == InputChunk::Function ||
276 c->kind() == InputChunk::SyntheticFunction;
277 }
278
279 std::optional<StringRef> getExportName() const {
280 return exportName ? std::optional<StringRef>(*exportName)
281 : std::optional<StringRef>();
282 }
283 void setExportName(std::string exportName) { this->exportName = exportName; }
284 uint32_t getFunctionInputOffset() const { return getInputSectionOffset(); }
285 uint32_t getFunctionCodeOffset() const {
286 // For generated synthetic functions, such as unreachable stubs generated
287 // for signature mismatches, 'function' reference does not exist. This
288 // function is used to get function offsets for .debug_info section, and for
289 // those generated stubs function offsets are not meaningful anyway. So just
290 // return 0 in those cases.
291 return function ? function->CodeOffset : 0;
292 }
293 uint32_t getFunctionIndex() const { return *functionIndex; }
294 bool hasFunctionIndex() const { return functionIndex.has_value(); }
295 void setFunctionIndex(uint32_t index);
296 uint32_t getTableIndex() const { return *tableIndex; }
297 bool hasTableIndex() const { return tableIndex.has_value(); }
298 void setTableIndex(uint32_t index);
299 void writeCompressed(uint8_t *buf) const;
300
301 // The size of a given input function can depend on the values of the
302 // LEB relocations within it. This finalizeContents method is called after
303 // all the symbol values have be calculated but before getSize() is ever
304 // called.
305 void calculateSize();
306
307 const WasmSignature &signature;
308
309 uint32_t getCompressedSize() const {
310 assert(compressedSize);
311 return compressedSize;
312 }
313
314 const WasmFunction *function = nullptr;
315
316protected:
317 std::optional<std::string> exportName;
318 std::optional<uint32_t> functionIndex;
319 std::optional<uint32_t> tableIndex;
320 uint32_t compressedFuncSize = 0;
321 uint32_t compressedSize = 0;
322};
323
324class SyntheticFunction : public InputFunction {
325public:
326 SyntheticFunction(const WasmSignature &s, StringRef name,
327 StringRef debugName = {})
328 : InputFunction(name, s) {
329 sectionKind = InputChunk::SyntheticFunction;
330 this->debugName = debugName;
331 }
332
333 static bool classof(const InputChunk *c) {
334 return c->kind() == InputChunk::SyntheticFunction;
335 }
336
337 void setBody(ArrayRef<uint8_t> body) { rawData = body; }
338};
339
340// Represents a single Wasm Section within an input file.
341class InputSection : public InputChunk {
342public:
343 InputSection(const WasmSection &s, ObjFile *f, uint32_t alignment)
344 : InputChunk(f, InputChunk::Section, s.Name, alignment),
345 tombstoneValue(getTombstoneForSection(name: s.Name)), section(s) {
346 assert(section.Type == llvm::wasm::WASM_SEC_CUSTOM);
347 comdat = section.Comdat;
348 rawData = section.Content;
349 }
350
351 static bool classof(const InputChunk *c) {
352 return c->kind() == InputChunk::Section;
353 }
354
355 const uint64_t tombstoneValue;
356
357protected:
358 static uint64_t getTombstoneForSection(StringRef name);
359 const WasmSection &section;
360};
361
362} // namespace wasm
363
364std::string toString(const wasm::InputChunk *);
365StringRef relocTypeToString(uint8_t relocType);
366
367} // namespace lld
368
369#endif // LLD_WASM_INPUT_CHUNKS_H
370