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