| 1 | //===- InputChunks.cpp ----------------------------------------------------===// |
| 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 | #include "InputChunks.h" |
| 10 | #include "Config.h" |
| 11 | #include "OutputSegment.h" |
| 12 | #include "WriterUtils.h" |
| 13 | #include "lld/Common/ErrorHandler.h" |
| 14 | #include "lld/Common/LLVM.h" |
| 15 | #include "llvm/Support/LEB128.h" |
| 16 | #include "llvm/Support/xxhash.h" |
| 17 | #include <algorithm> |
| 18 | |
| 19 | #define DEBUG_TYPE "lld" |
| 20 | |
| 21 | using namespace llvm; |
| 22 | using namespace llvm::wasm; |
| 23 | using namespace llvm::support::endian; |
| 24 | |
| 25 | namespace lld { |
| 26 | StringRef relocTypeToString(uint8_t relocType) { |
| 27 | switch (relocType) { |
| 28 | #define WASM_RELOC(NAME, REL) \ |
| 29 | case REL: \ |
| 30 | return #NAME; |
| 31 | #include "llvm/BinaryFormat/WasmRelocs.def" |
| 32 | #undef WASM_RELOC |
| 33 | } |
| 34 | llvm_unreachable("unknown reloc type" ); |
| 35 | } |
| 36 | |
| 37 | bool relocIs64(uint8_t relocType) { |
| 38 | switch (relocType) { |
| 39 | case R_WASM_MEMORY_ADDR_LEB64: |
| 40 | case R_WASM_MEMORY_ADDR_SLEB64: |
| 41 | case R_WASM_MEMORY_ADDR_REL_SLEB64: |
| 42 | case R_WASM_MEMORY_ADDR_I64: |
| 43 | case R_WASM_TABLE_INDEX_SLEB64: |
| 44 | case R_WASM_TABLE_INDEX_I64: |
| 45 | case R_WASM_FUNCTION_OFFSET_I64: |
| 46 | case R_WASM_TABLE_INDEX_REL_SLEB64: |
| 47 | case R_WASM_MEMORY_ADDR_TLS_SLEB64: |
| 48 | return true; |
| 49 | default: |
| 50 | return false; |
| 51 | } |
| 52 | } |
| 53 | |
| 54 | std::string toString(const wasm::InputChunk *c) { |
| 55 | return (toString(file: c->file) + ":(" + c->name + ")" ).str(); |
| 56 | } |
| 57 | |
| 58 | namespace wasm { |
| 59 | StringRef InputChunk::getComdatName() const { |
| 60 | uint32_t index = getComdat(); |
| 61 | if (index == UINT32_MAX) |
| 62 | return StringRef(); |
| 63 | return file->getWasmObj()->linkingData().Comdats[index]; |
| 64 | } |
| 65 | |
| 66 | uint32_t InputChunk::getSize() const { |
| 67 | if (const auto *ms = dyn_cast<SyntheticMergedChunk>(Val: this)) |
| 68 | return ms->builder.getSize(); |
| 69 | |
| 70 | if (const auto *f = dyn_cast<InputFunction>(Val: this)) { |
| 71 | if (ctx.arg.compressRelocations && f->file) { |
| 72 | return f->getCompressedSize(); |
| 73 | } |
| 74 | } |
| 75 | |
| 76 | return data().size(); |
| 77 | } |
| 78 | |
| 79 | uint32_t InputChunk::getInputSize() const { |
| 80 | if (const auto *f = dyn_cast<InputFunction>(Val: this)) |
| 81 | return f->function->Size; |
| 82 | return getSize(); |
| 83 | } |
| 84 | |
| 85 | // Copy this input chunk to an mmap'ed output file and apply relocations. |
| 86 | void InputChunk::writeTo(uint8_t *buf) const { |
| 87 | if (const auto *f = dyn_cast<InputFunction>(Val: this)) { |
| 88 | if (file && ctx.arg.compressRelocations) |
| 89 | return f->writeCompressed(buf); |
| 90 | } else if (const auto *ms = dyn_cast<SyntheticMergedChunk>(Val: this)) { |
| 91 | ms->builder.write(Buf: buf + outSecOff); |
| 92 | // Apply relocations |
| 93 | ms->relocate(buf: buf + outSecOff); |
| 94 | return; |
| 95 | } |
| 96 | |
| 97 | // Copy contents |
| 98 | memcpy(dest: buf + outSecOff, src: data().data(), n: data().size()); |
| 99 | |
| 100 | // Apply relocations |
| 101 | relocate(buf: buf + outSecOff); |
| 102 | } |
| 103 | |
| 104 | void InputChunk::relocate(uint8_t *buf) const { |
| 105 | if (relocations.empty()) |
| 106 | return; |
| 107 | |
| 108 | LLVM_DEBUG(dbgs() << "applying relocations: " << toString(this) |
| 109 | << " count=" << relocations.size() << "\n" ); |
| 110 | int32_t inputSectionOffset = getInputSectionOffset(); |
| 111 | uint64_t tombstone = getTombstone(); |
| 112 | |
| 113 | for (const WasmRelocation &rel : relocations) { |
| 114 | uint8_t *loc = buf + rel.Offset - inputSectionOffset; |
| 115 | LLVM_DEBUG(dbgs() << "apply reloc: type=" << relocTypeToString(rel.Type)); |
| 116 | if (rel.Type != R_WASM_TYPE_INDEX_LEB) |
| 117 | LLVM_DEBUG(dbgs() << " sym=" << file->getSymbols()[rel.Index]->getName()); |
| 118 | LLVM_DEBUG(dbgs() << " addend=" << rel.Addend << " index=" << rel.Index |
| 119 | << " offset=" << rel.Offset << "\n" ); |
| 120 | // TODO(sbc): Check that the value is within the range of the |
| 121 | // relocation type below. Most likely we must error out here |
| 122 | // if its not with range. |
| 123 | uint64_t value = file->calcNewValue(reloc: rel, tombstone, chunk: this); |
| 124 | |
| 125 | switch (rel.Type) { |
| 126 | case R_WASM_TYPE_INDEX_LEB: |
| 127 | case R_WASM_FUNCTION_INDEX_LEB: |
| 128 | case R_WASM_GLOBAL_INDEX_LEB: |
| 129 | case R_WASM_TAG_INDEX_LEB: |
| 130 | case R_WASM_MEMORY_ADDR_LEB: |
| 131 | case R_WASM_TABLE_NUMBER_LEB: |
| 132 | encodeULEB128(Value: static_cast<uint32_t>(value), p: loc, PadTo: 5); |
| 133 | break; |
| 134 | case R_WASM_MEMORY_ADDR_LEB64: |
| 135 | encodeULEB128(Value: value, p: loc, PadTo: 10); |
| 136 | break; |
| 137 | case R_WASM_TABLE_INDEX_SLEB: |
| 138 | case R_WASM_TABLE_INDEX_REL_SLEB: |
| 139 | case R_WASM_MEMORY_ADDR_SLEB: |
| 140 | case R_WASM_MEMORY_ADDR_REL_SLEB: |
| 141 | case R_WASM_MEMORY_ADDR_TLS_SLEB: |
| 142 | encodeSLEB128(Value: static_cast<int32_t>(value), p: loc, PadTo: 5); |
| 143 | break; |
| 144 | case R_WASM_TABLE_INDEX_SLEB64: |
| 145 | case R_WASM_TABLE_INDEX_REL_SLEB64: |
| 146 | case R_WASM_MEMORY_ADDR_SLEB64: |
| 147 | case R_WASM_MEMORY_ADDR_REL_SLEB64: |
| 148 | case R_WASM_MEMORY_ADDR_TLS_SLEB64: |
| 149 | encodeSLEB128(Value: static_cast<int64_t>(value), p: loc, PadTo: 10); |
| 150 | break; |
| 151 | case R_WASM_TABLE_INDEX_I32: |
| 152 | case R_WASM_MEMORY_ADDR_I32: |
| 153 | case R_WASM_FUNCTION_OFFSET_I32: |
| 154 | case R_WASM_FUNCTION_INDEX_I32: |
| 155 | case R_WASM_SECTION_OFFSET_I32: |
| 156 | case R_WASM_GLOBAL_INDEX_I32: |
| 157 | case R_WASM_MEMORY_ADDR_LOCREL_I32: |
| 158 | write32le(P: loc, V: value); |
| 159 | break; |
| 160 | case R_WASM_TABLE_INDEX_I64: |
| 161 | case R_WASM_MEMORY_ADDR_I64: |
| 162 | case R_WASM_FUNCTION_OFFSET_I64: |
| 163 | write64le(P: loc, V: value); |
| 164 | break; |
| 165 | default: |
| 166 | llvm_unreachable("unknown relocation type" ); |
| 167 | } |
| 168 | } |
| 169 | } |
| 170 | |
| 171 | static bool relocIsLive(const WasmRelocation &rel, ObjFile *file) { |
| 172 | return rel.Type == R_WASM_TYPE_INDEX_LEB || |
| 173 | file->getSymbol(index: rel.Index)->isLive(); |
| 174 | } |
| 175 | |
| 176 | size_t InputChunk::getNumLiveRelocations() const { |
| 177 | return llvm::count_if(Range: relocations, P: [this](const WasmRelocation &rel) { |
| 178 | return relocIsLive(rel, file); |
| 179 | }); |
| 180 | } |
| 181 | |
| 182 | // Copy relocation entries to a given output stream. |
| 183 | // This function is used only when a user passes "-r". For a regular link, |
| 184 | // we consume relocations instead of copying them to an output file. |
| 185 | void InputChunk::writeRelocations(raw_ostream &os) const { |
| 186 | if (relocations.empty()) |
| 187 | return; |
| 188 | |
| 189 | int32_t off = outSecOff - getInputSectionOffset(); |
| 190 | LLVM_DEBUG(dbgs() << "writeRelocations: " << file->getName() |
| 191 | << " offset=" << Twine(off) << "\n" ); |
| 192 | |
| 193 | for (const WasmRelocation &rel : relocations) { |
| 194 | if (!relocIsLive(rel, file)) |
| 195 | continue; |
| 196 | writeUleb128(os, number: rel.Type, msg: "reloc type" ); |
| 197 | writeUleb128(os, number: rel.Offset + off, msg: "reloc offset" ); |
| 198 | writeUleb128(os, number: file->calcNewIndex(reloc: rel), msg: "reloc index" ); |
| 199 | |
| 200 | if (relocTypeHasAddend(type: rel.Type)) |
| 201 | writeSleb128(os, number: file->calcNewAddend(reloc: rel), msg: "reloc addend" ); |
| 202 | } |
| 203 | } |
| 204 | |
| 205 | uint64_t InputChunk::getTombstone() const { |
| 206 | if (const auto *s = dyn_cast<InputSection>(Val: this)) { |
| 207 | return s->tombstoneValue; |
| 208 | } |
| 209 | |
| 210 | return 0; |
| 211 | } |
| 212 | |
| 213 | void InputFunction::setFunctionIndex(uint32_t index) { |
| 214 | LLVM_DEBUG(dbgs() << "InputFunction::setFunctionIndex: " << name << " -> " |
| 215 | << index << "\n" ); |
| 216 | assert(!hasFunctionIndex()); |
| 217 | functionIndex = index; |
| 218 | } |
| 219 | |
| 220 | void InputFunction::setTableIndex(uint32_t index) { |
| 221 | LLVM_DEBUG(dbgs() << "InputFunction::setTableIndex: " << name << " -> " |
| 222 | << index << "\n" ); |
| 223 | assert(!hasTableIndex()); |
| 224 | tableIndex = index; |
| 225 | } |
| 226 | |
| 227 | // Write a relocation value without padding and return the number of bytes |
| 228 | // witten. |
| 229 | static unsigned writeCompressedReloc(uint8_t *buf, const WasmRelocation &rel, |
| 230 | uint64_t value) { |
| 231 | switch (rel.getType()) { |
| 232 | case R_WASM_TYPE_INDEX_LEB: |
| 233 | case R_WASM_FUNCTION_INDEX_LEB: |
| 234 | case R_WASM_GLOBAL_INDEX_LEB: |
| 235 | case R_WASM_TAG_INDEX_LEB: |
| 236 | case R_WASM_MEMORY_ADDR_LEB: |
| 237 | case R_WASM_MEMORY_ADDR_LEB64: |
| 238 | case R_WASM_TABLE_NUMBER_LEB: |
| 239 | return encodeULEB128(Value: value, p: buf); |
| 240 | case R_WASM_TABLE_INDEX_SLEB: |
| 241 | case R_WASM_TABLE_INDEX_SLEB64: |
| 242 | case R_WASM_TABLE_INDEX_REL_SLEB64: |
| 243 | case R_WASM_MEMORY_ADDR_SLEB: |
| 244 | case R_WASM_MEMORY_ADDR_SLEB64: |
| 245 | case R_WASM_MEMORY_ADDR_REL_SLEB: |
| 246 | case R_WASM_MEMORY_ADDR_REL_SLEB64: |
| 247 | case R_WASM_MEMORY_ADDR_TLS_SLEB: |
| 248 | case R_WASM_MEMORY_ADDR_TLS_SLEB64: |
| 249 | case R_WASM_TABLE_INDEX_REL_SLEB: |
| 250 | return encodeSLEB128(Value: static_cast<int64_t>(value), p: buf); |
| 251 | case R_WASM_TABLE_INDEX_I32: |
| 252 | case R_WASM_MEMORY_ADDR_I32: |
| 253 | case R_WASM_FUNCTION_OFFSET_I32: |
| 254 | case R_WASM_SECTION_OFFSET_I32: |
| 255 | case R_WASM_GLOBAL_INDEX_I32: |
| 256 | case R_WASM_MEMORY_ADDR_I64: |
| 257 | case R_WASM_TABLE_INDEX_I64: |
| 258 | case R_WASM_FUNCTION_OFFSET_I64: |
| 259 | case R_WASM_MEMORY_ADDR_LOCREL_I32: |
| 260 | case R_WASM_FUNCTION_INDEX_I32: |
| 261 | fatal(msg: "relocation compression not supported for " + |
| 262 | relocTypeToString(relocType: rel.Type)); |
| 263 | } |
| 264 | llvm_unreachable("unhandled relocation type" ); |
| 265 | } |
| 266 | |
| 267 | static unsigned getRelocWidthPadded(const WasmRelocation &rel) { |
| 268 | switch (rel.getType()) { |
| 269 | case R_WASM_TYPE_INDEX_LEB: |
| 270 | case R_WASM_FUNCTION_INDEX_LEB: |
| 271 | case R_WASM_GLOBAL_INDEX_LEB: |
| 272 | case R_WASM_TAG_INDEX_LEB: |
| 273 | case R_WASM_MEMORY_ADDR_LEB: |
| 274 | case R_WASM_TABLE_NUMBER_LEB: |
| 275 | case R_WASM_TABLE_INDEX_SLEB: |
| 276 | case R_WASM_TABLE_INDEX_REL_SLEB: |
| 277 | case R_WASM_MEMORY_ADDR_SLEB: |
| 278 | case R_WASM_MEMORY_ADDR_REL_SLEB: |
| 279 | case R_WASM_MEMORY_ADDR_TLS_SLEB: |
| 280 | return 5; |
| 281 | case R_WASM_TABLE_INDEX_SLEB64: |
| 282 | case R_WASM_TABLE_INDEX_REL_SLEB64: |
| 283 | case R_WASM_MEMORY_ADDR_LEB64: |
| 284 | case R_WASM_MEMORY_ADDR_SLEB64: |
| 285 | case R_WASM_MEMORY_ADDR_REL_SLEB64: |
| 286 | case R_WASM_MEMORY_ADDR_TLS_SLEB64: |
| 287 | return 10; |
| 288 | case R_WASM_TABLE_INDEX_I32: |
| 289 | case R_WASM_MEMORY_ADDR_I32: |
| 290 | case R_WASM_FUNCTION_OFFSET_I32: |
| 291 | case R_WASM_SECTION_OFFSET_I32: |
| 292 | case R_WASM_GLOBAL_INDEX_I32: |
| 293 | case R_WASM_MEMORY_ADDR_I64: |
| 294 | case R_WASM_TABLE_INDEX_I64: |
| 295 | case R_WASM_FUNCTION_OFFSET_I64: |
| 296 | case R_WASM_MEMORY_ADDR_LOCREL_I32: |
| 297 | case R_WASM_FUNCTION_INDEX_I32: |
| 298 | fatal(msg: "relocation compression not supported for " + |
| 299 | relocTypeToString(relocType: rel.Type)); |
| 300 | } |
| 301 | llvm_unreachable("unhandled relocation type" ); |
| 302 | } |
| 303 | |
| 304 | static unsigned getRelocWidth(const WasmRelocation &rel, uint64_t value) { |
| 305 | uint8_t buf[10]; |
| 306 | return writeCompressedReloc(buf, rel, value); |
| 307 | } |
| 308 | |
| 309 | // Relocations of type LEB and SLEB in the code section are padded to 5 bytes |
| 310 | // so that a fast linker can blindly overwrite them without needing to worry |
| 311 | // about the number of bytes needed to encode the values. |
| 312 | // However, for optimal output the code section can be compressed to remove |
| 313 | // the padding then outputting non-relocatable files. |
| 314 | // In this case we need to perform a size calculation based on the value at each |
| 315 | // relocation. At best we end up saving 4 bytes for each relocation entry. |
| 316 | // |
| 317 | // This function only computes the final output size. It must be called |
| 318 | // before getSize() is used to calculate of layout of the code section. |
| 319 | void InputFunction::calculateSize() { |
| 320 | if (!file || !ctx.arg.compressRelocations) |
| 321 | return; |
| 322 | |
| 323 | LLVM_DEBUG(dbgs() << "calculateSize: " << name << "\n" ); |
| 324 | |
| 325 | const uint8_t *secStart = file->codeSection->Content.data(); |
| 326 | const uint8_t *funcStart = secStart + getInputSectionOffset(); |
| 327 | uint32_t functionSizeLength; |
| 328 | decodeULEB128(p: funcStart, n: &functionSizeLength); |
| 329 | |
| 330 | uint32_t start = getInputSectionOffset(); |
| 331 | uint32_t end = start + function->Size; |
| 332 | |
| 333 | uint64_t tombstone = getTombstone(); |
| 334 | |
| 335 | uint32_t lastRelocEnd = start + functionSizeLength; |
| 336 | for (const WasmRelocation &rel : relocations) { |
| 337 | LLVM_DEBUG(dbgs() << " region: " << (rel.Offset - lastRelocEnd) << "\n" ); |
| 338 | compressedFuncSize += rel.Offset - lastRelocEnd; |
| 339 | compressedFuncSize += |
| 340 | getRelocWidth(rel, value: file->calcNewValue(reloc: rel, tombstone, chunk: this)); |
| 341 | lastRelocEnd = rel.Offset + getRelocWidthPadded(rel); |
| 342 | } |
| 343 | LLVM_DEBUG(dbgs() << " final region: " << (end - lastRelocEnd) << "\n" ); |
| 344 | compressedFuncSize += end - lastRelocEnd; |
| 345 | |
| 346 | // Now we know how long the resulting function is we can add the encoding |
| 347 | // of its length |
| 348 | uint8_t buf[5]; |
| 349 | compressedSize = compressedFuncSize + encodeULEB128(Value: compressedFuncSize, p: buf); |
| 350 | |
| 351 | LLVM_DEBUG(dbgs() << " calculateSize orig: " << function->Size << "\n" ); |
| 352 | LLVM_DEBUG(dbgs() << " calculateSize new: " << compressedSize << "\n" ); |
| 353 | } |
| 354 | |
| 355 | // Override the default writeTo method so that we can (optionally) write the |
| 356 | // compressed version of the function. |
| 357 | void InputFunction::writeCompressed(uint8_t *buf) const { |
| 358 | buf += outSecOff; |
| 359 | uint8_t *orig = buf; |
| 360 | (void)orig; |
| 361 | |
| 362 | const uint8_t *secStart = file->codeSection->Content.data(); |
| 363 | const uint8_t *funcStart = secStart + getInputSectionOffset(); |
| 364 | const uint8_t *end = funcStart + function->Size; |
| 365 | uint64_t tombstone = getTombstone(); |
| 366 | uint32_t count; |
| 367 | decodeULEB128(p: funcStart, n: &count); |
| 368 | funcStart += count; |
| 369 | |
| 370 | LLVM_DEBUG(dbgs() << "write func: " << name << "\n" ); |
| 371 | buf += encodeULEB128(Value: compressedFuncSize, p: buf); |
| 372 | const uint8_t *lastRelocEnd = funcStart; |
| 373 | for (const WasmRelocation &rel : relocations) { |
| 374 | unsigned chunkSize = (secStart + rel.Offset) - lastRelocEnd; |
| 375 | LLVM_DEBUG(dbgs() << " write chunk: " << chunkSize << "\n" ); |
| 376 | memcpy(dest: buf, src: lastRelocEnd, n: chunkSize); |
| 377 | buf += chunkSize; |
| 378 | buf += writeCompressedReloc(buf, rel, |
| 379 | value: file->calcNewValue(reloc: rel, tombstone, chunk: this)); |
| 380 | lastRelocEnd = secStart + rel.Offset + getRelocWidthPadded(rel); |
| 381 | } |
| 382 | |
| 383 | unsigned chunkSize = end - lastRelocEnd; |
| 384 | LLVM_DEBUG(dbgs() << " write final chunk: " << chunkSize << "\n" ); |
| 385 | memcpy(dest: buf, src: lastRelocEnd, n: chunkSize); |
| 386 | LLVM_DEBUG(dbgs() << " total: " << (buf + chunkSize - orig) << "\n" ); |
| 387 | } |
| 388 | |
| 389 | uint64_t InputChunk::getChunkOffset(uint64_t offset) const { |
| 390 | if (const auto *ms = dyn_cast<MergeInputChunk>(Val: this)) { |
| 391 | LLVM_DEBUG(dbgs() << "getChunkOffset(merged): " << name << "\n" ); |
| 392 | LLVM_DEBUG(dbgs() << "offset: " << offset << "\n" ); |
| 393 | LLVM_DEBUG(dbgs() << "parentOffset: " << ms->getParentOffset(offset) |
| 394 | << "\n" ); |
| 395 | assert(ms->parent); |
| 396 | return ms->parent->getChunkOffset(offset: ms->getParentOffset(offset)); |
| 397 | } |
| 398 | return outputSegmentOffset + offset; |
| 399 | } |
| 400 | |
| 401 | uint64_t InputChunk::getOffset(uint64_t offset) const { |
| 402 | return outSecOff + getChunkOffset(offset); |
| 403 | } |
| 404 | |
| 405 | uint64_t InputChunk::getVA(uint64_t offset) const { |
| 406 | return (outputSeg ? outputSeg->startVA : 0) + getChunkOffset(offset); |
| 407 | } |
| 408 | |
| 409 | // Generate code to apply relocations to the data section at runtime. |
| 410 | // This is only called when generating shared libraries (PIC) where address are |
| 411 | // not known at static link time. |
| 412 | bool InputChunk::generateRelocationCode(raw_ostream &os) const { |
| 413 | LLVM_DEBUG(dbgs() << "generating runtime relocations: " << name |
| 414 | << " count=" << relocations.size() << "\n" ); |
| 415 | |
| 416 | bool is64 = ctx.arg.is64.value_or(u: false); |
| 417 | bool generated = false; |
| 418 | unsigned opcode_ptr_const = is64 ? WASM_OPCODE_I64_CONST |
| 419 | : WASM_OPCODE_I32_CONST; |
| 420 | unsigned opcode_ptr_add = is64 ? WASM_OPCODE_I64_ADD |
| 421 | : WASM_OPCODE_I32_ADD; |
| 422 | |
| 423 | uint64_t tombstone = getTombstone(); |
| 424 | // TODO(sbc): Encode the relocations in the data section and write a loop |
| 425 | // here to apply them. |
| 426 | for (const WasmRelocation &rel : relocations) { |
| 427 | uint64_t offset = getVA(offset: rel.Offset) - getInputSectionOffset(); |
| 428 | |
| 429 | Symbol *sym = file->getSymbol(reloc: rel); |
| 430 | // Runtime relocations are needed when we don't know the address of |
| 431 | // a symbol statically. |
| 432 | bool requiresRuntimeReloc = ctx.isPic || sym->hasGOTIndex(); |
| 433 | if (!requiresRuntimeReloc) |
| 434 | continue; |
| 435 | |
| 436 | LLVM_DEBUG(dbgs() << "gen reloc: type=" << relocTypeToString(rel.Type) |
| 437 | << " addend=" << rel.Addend << " index=" << rel.Index |
| 438 | << " output offset=" << offset << "\n" ); |
| 439 | |
| 440 | // Calculate the address at which to apply the relocation |
| 441 | writeU8(os, byte: opcode_ptr_const, msg: "CONST" ); |
| 442 | writeSleb128(os, number: offset, msg: "offset" ); |
| 443 | |
| 444 | // In PIC mode we need to add the __memory_base |
| 445 | if (ctx.isPic) { |
| 446 | writeU8(os, byte: WASM_OPCODE_GLOBAL_GET, msg: "GLOBAL_GET" ); |
| 447 | if (isTLS()) |
| 448 | writeUleb128(os, number: ctx.sym.tlsBase->getGlobalIndex(), msg: "tls_base" ); |
| 449 | else |
| 450 | writeUleb128(os, number: ctx.sym.memoryBase->getGlobalIndex(), msg: "memory_base" ); |
| 451 | writeU8(os, byte: opcode_ptr_add, msg: "ADD" ); |
| 452 | } |
| 453 | |
| 454 | // Now figure out what we want to store at this location |
| 455 | bool is64 = relocIs64(relocType: rel.Type); |
| 456 | unsigned opcode_reloc_const = |
| 457 | is64 ? WASM_OPCODE_I64_CONST : WASM_OPCODE_I32_CONST; |
| 458 | unsigned opcode_reloc_add = |
| 459 | is64 ? WASM_OPCODE_I64_ADD : WASM_OPCODE_I32_ADD; |
| 460 | unsigned opcode_reloc_store = |
| 461 | is64 ? WASM_OPCODE_I64_STORE : WASM_OPCODE_I32_STORE; |
| 462 | |
| 463 | if (sym->hasGOTIndex()) { |
| 464 | writeU8(os, byte: WASM_OPCODE_GLOBAL_GET, msg: "GLOBAL_GET" ); |
| 465 | writeUleb128(os, number: sym->getGOTIndex(), msg: "global index" ); |
| 466 | if (rel.Addend) { |
| 467 | writeU8(os, byte: opcode_reloc_const, msg: "CONST" ); |
| 468 | writeSleb128(os, number: rel.Addend, msg: "addend" ); |
| 469 | writeU8(os, byte: opcode_reloc_add, msg: "ADD" ); |
| 470 | } |
| 471 | } else { |
| 472 | assert(ctx.isPic); |
| 473 | const GlobalSymbol *baseSymbol = ctx.sym.memoryBase; |
| 474 | if (rel.Type == R_WASM_TABLE_INDEX_I32 || |
| 475 | rel.Type == R_WASM_TABLE_INDEX_I64) |
| 476 | baseSymbol = ctx.sym.tableBase; |
| 477 | else if (sym->isTLS()) |
| 478 | baseSymbol = ctx.sym.tlsBase; |
| 479 | writeU8(os, byte: WASM_OPCODE_GLOBAL_GET, msg: "GLOBAL_GET" ); |
| 480 | writeUleb128(os, number: baseSymbol->getGlobalIndex(), msg: "base" ); |
| 481 | writeU8(os, byte: opcode_reloc_const, msg: "CONST" ); |
| 482 | writeSleb128(os, number: file->calcNewValue(reloc: rel, tombstone, chunk: this), msg: "offset" ); |
| 483 | writeU8(os, byte: opcode_reloc_add, msg: "ADD" ); |
| 484 | } |
| 485 | |
| 486 | // Store that value at the virtual address |
| 487 | writeU8(os, byte: opcode_reloc_store, msg: "I32_STORE" ); |
| 488 | writeUleb128(os, number: 2, msg: "align" ); |
| 489 | writeUleb128(os, number: 0, msg: "offset" ); |
| 490 | generated = true; |
| 491 | } |
| 492 | return generated; |
| 493 | } |
| 494 | |
| 495 | // Split WASM_SEG_FLAG_STRINGS section. Such a section is a sequence of |
| 496 | // null-terminated strings. |
| 497 | void MergeInputChunk::splitStrings(ArrayRef<uint8_t> data) { |
| 498 | LLVM_DEBUG(llvm::dbgs() << "splitStrings\n" ); |
| 499 | size_t off = 0; |
| 500 | StringRef s = toStringRef(Input: data); |
| 501 | |
| 502 | while (!s.empty()) { |
| 503 | size_t end = s.find(C: 0); |
| 504 | if (end == StringRef::npos) |
| 505 | fatal(msg: toString(c: this) + ": string is not null terminated" ); |
| 506 | size_t size = end + 1; |
| 507 | |
| 508 | pieces.emplace_back(args&: off, args: xxh3_64bits(data: s.substr(Start: 0, N: size)), args: true); |
| 509 | s = s.substr(Start: size); |
| 510 | off += size; |
| 511 | } |
| 512 | } |
| 513 | |
| 514 | // This function is called after we obtain a complete list of input sections |
| 515 | // that need to be linked. This is responsible to split section contents |
| 516 | // into small chunks for further processing. |
| 517 | // |
| 518 | // Note that this function is called from parallelForEach. This must be |
| 519 | // thread-safe (i.e. no memory allocation from the pools). |
| 520 | void MergeInputChunk::splitIntoPieces() { |
| 521 | assert(pieces.empty()); |
| 522 | // As of now we only support WASM_SEG_FLAG_STRINGS but in the future we |
| 523 | // could add other types of splitting (see ELF's splitIntoPieces). |
| 524 | assert(flags & WASM_SEG_FLAG_STRINGS); |
| 525 | splitStrings(data: data()); |
| 526 | } |
| 527 | |
| 528 | SectionPiece *MergeInputChunk::getSectionPiece(uint64_t offset) { |
| 529 | if (this->data().size() <= offset) |
| 530 | fatal(msg: toString(c: this) + ": offset is outside the section" ); |
| 531 | |
| 532 | // If Offset is not at beginning of a section piece, it is not in the map. |
| 533 | // In that case we need to do a binary search of the original section piece |
| 534 | // vector. |
| 535 | auto it = partition_point( |
| 536 | Range&: pieces, P: [=](SectionPiece p) { return p.inputOff <= offset; }); |
| 537 | return &it[-1]; |
| 538 | } |
| 539 | |
| 540 | // Returns the offset in an output section for a given input offset. |
| 541 | // Because contents of a mergeable section is not contiguous in output, |
| 542 | // it is not just an addition to a base output offset. |
| 543 | uint64_t MergeInputChunk::getParentOffset(uint64_t offset) const { |
| 544 | // If Offset is not at beginning of a section piece, it is not in the map. |
| 545 | // In that case we need to search from the original section piece vector. |
| 546 | const SectionPiece *piece = getSectionPiece(offset); |
| 547 | uint64_t addend = offset - piece->inputOff; |
| 548 | return piece->outputOff + addend; |
| 549 | } |
| 550 | |
| 551 | void SyntheticMergedChunk::finalizeContents() { |
| 552 | // Add all string pieces to the string table builder to create section |
| 553 | // contents. |
| 554 | for (MergeInputChunk *sec : chunks) |
| 555 | for (size_t i = 0, e = sec->pieces.size(); i != e; ++i) |
| 556 | if (sec->pieces[i].live) |
| 557 | builder.add(S: sec->getData(i)); |
| 558 | |
| 559 | // Fix the string table content. After this, the contents will never change. |
| 560 | builder.finalize(); |
| 561 | |
| 562 | // finalize() fixed tail-optimized strings, so we can now get |
| 563 | // offsets of strings. Get an offset for each string and save it |
| 564 | // to a corresponding SectionPiece for easy access. |
| 565 | for (MergeInputChunk *sec : chunks) |
| 566 | for (size_t i = 0, e = sec->pieces.size(); i != e; ++i) |
| 567 | if (sec->pieces[i].live) |
| 568 | sec->pieces[i].outputOff = builder.getOffset(S: sec->getData(i)); |
| 569 | } |
| 570 | |
| 571 | uint64_t InputSection::getTombstoneForSection(StringRef name) { |
| 572 | // When a function is not live we need to update relocations referring to it. |
| 573 | // If they occur in DWARF debug symbols, we want to change the pc of the |
| 574 | // function to -1 to avoid overlapping with a valid range. However for the |
| 575 | // debug_ranges and debug_loc sections that would conflict with the existing |
| 576 | // meaning of -1 so we use -2. |
| 577 | if (name == ".debug_ranges" || name == ".debug_loc" ) |
| 578 | return UINT64_C(-2); |
| 579 | if (name.starts_with(Prefix: ".debug_" )) |
| 580 | return UINT64_C(-1); |
| 581 | // If the function occurs in an function attribute section change it to -1 since |
| 582 | // 0 is a valid function index. |
| 583 | if (name.starts_with(Prefix: "llvm.func_attr." )) |
| 584 | return UINT64_C(-1); |
| 585 | // Returning 0 means there is no tombstone value for this section, and relocation |
| 586 | // will just use the addend. |
| 587 | return 0; |
| 588 | } |
| 589 | |
| 590 | } // namespace wasm |
| 591 | } // namespace lld |
| 592 | |