1 //===- DWARFEmitter - Convert YAML to DWARF binary data -------------------===// 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 /// \file 10 /// The DWARF component of yaml2obj. Provided as library code for tests. 11 /// 12 //===----------------------------------------------------------------------===// 13 14 #include "llvm/ObjectYAML/DWARFEmitter.h" 15 #include "llvm/ADT/ArrayRef.h" 16 #include "llvm/ADT/StringMap.h" 17 #include "llvm/ADT/StringRef.h" 18 #include "llvm/ADT/StringSwitch.h" 19 #include "llvm/BinaryFormat/Dwarf.h" 20 #include "llvm/ObjectYAML/DWARFYAML.h" 21 #include "llvm/Support/Errc.h" 22 #include "llvm/Support/Error.h" 23 #include "llvm/Support/Host.h" 24 #include "llvm/Support/LEB128.h" 25 #include "llvm/Support/MathExtras.h" 26 #include "llvm/Support/MemoryBuffer.h" 27 #include "llvm/Support/SourceMgr.h" 28 #include "llvm/Support/SwapByteOrder.h" 29 #include "llvm/Support/YAMLTraits.h" 30 #include "llvm/Support/raw_ostream.h" 31 #include <algorithm> 32 #include <cassert> 33 #include <cstddef> 34 #include <cstdint> 35 #include <memory> 36 #include <optional> 37 #include <string> 38 #include <vector> 39 40 using namespace llvm; 41 42 template <typename T> 43 static void writeInteger(T Integer, raw_ostream &OS, bool IsLittleEndian) { 44 if (IsLittleEndian != sys::IsLittleEndianHost) 45 sys::swapByteOrder(Integer); 46 OS.write(reinterpret_cast<char *>(&Integer), sizeof(T)); 47 } 48 49 static Error writeVariableSizedInteger(uint64_t Integer, size_t Size, 50 raw_ostream &OS, bool IsLittleEndian) { 51 if (8 == Size) 52 writeInteger((uint64_t)Integer, OS, IsLittleEndian); 53 else if (4 == Size) 54 writeInteger((uint32_t)Integer, OS, IsLittleEndian); 55 else if (2 == Size) 56 writeInteger((uint16_t)Integer, OS, IsLittleEndian); 57 else if (1 == Size) 58 writeInteger((uint8_t)Integer, OS, IsLittleEndian); 59 else 60 return createStringError(errc::not_supported, 61 "invalid integer write size: %zu", Size); 62 63 return Error::success(); 64 } 65 66 static void ZeroFillBytes(raw_ostream &OS, size_t Size) { 67 std::vector<uint8_t> FillData(Size, 0); 68 OS.write(reinterpret_cast<char *>(FillData.data()), Size); 69 } 70 71 static void writeInitialLength(const dwarf::DwarfFormat Format, 72 const uint64_t Length, raw_ostream &OS, 73 bool IsLittleEndian) { 74 bool IsDWARF64 = Format == dwarf::DWARF64; 75 if (IsDWARF64) 76 cantFail(writeVariableSizedInteger(dwarf::DW_LENGTH_DWARF64, 4, OS, 77 IsLittleEndian)); 78 cantFail( 79 writeVariableSizedInteger(Length, IsDWARF64 ? 8 : 4, OS, IsLittleEndian)); 80 } 81 82 static void writeDWARFOffset(uint64_t Offset, dwarf::DwarfFormat Format, 83 raw_ostream &OS, bool IsLittleEndian) { 84 cantFail(writeVariableSizedInteger(Offset, Format == dwarf::DWARF64 ? 8 : 4, 85 OS, IsLittleEndian)); 86 } 87 88 Error DWARFYAML::emitDebugStr(raw_ostream &OS, const DWARFYAML::Data &DI) { 89 for (StringRef Str : *DI.DebugStrings) { 90 OS.write(Str.data(), Str.size()); 91 OS.write('\0'); 92 } 93 94 return Error::success(); 95 } 96 97 StringRef DWARFYAML::Data::getAbbrevTableContentByIndex(uint64_t Index) const { 98 assert(Index < DebugAbbrev.size() && 99 "Index should be less than the size of DebugAbbrev array"); 100 auto It = AbbrevTableContents.find(Index); 101 if (It != AbbrevTableContents.cend()) 102 return It->second; 103 104 std::string AbbrevTableBuffer; 105 raw_string_ostream OS(AbbrevTableBuffer); 106 107 uint64_t AbbrevCode = 0; 108 for (const DWARFYAML::Abbrev &AbbrevDecl : DebugAbbrev[Index].Table) { 109 AbbrevCode = AbbrevDecl.Code ? (uint64_t)*AbbrevDecl.Code : AbbrevCode + 1; 110 encodeULEB128(AbbrevCode, OS); 111 encodeULEB128(AbbrevDecl.Tag, OS); 112 OS.write(AbbrevDecl.Children); 113 for (const auto &Attr : AbbrevDecl.Attributes) { 114 encodeULEB128(Attr.Attribute, OS); 115 encodeULEB128(Attr.Form, OS); 116 if (Attr.Form == dwarf::DW_FORM_implicit_const) 117 encodeSLEB128(Attr.Value, OS); 118 } 119 encodeULEB128(0, OS); 120 encodeULEB128(0, OS); 121 } 122 123 // The abbreviations for a given compilation unit end with an entry 124 // consisting of a 0 byte for the abbreviation code. 125 OS.write_zeros(1); 126 127 AbbrevTableContents.insert({Index, AbbrevTableBuffer}); 128 129 return AbbrevTableContents[Index]; 130 } 131 132 Error DWARFYAML::emitDebugAbbrev(raw_ostream &OS, const DWARFYAML::Data &DI) { 133 for (uint64_t I = 0; I < DI.DebugAbbrev.size(); ++I) { 134 StringRef AbbrevTableContent = DI.getAbbrevTableContentByIndex(I); 135 OS.write(AbbrevTableContent.data(), AbbrevTableContent.size()); 136 } 137 138 return Error::success(); 139 } 140 141 Error DWARFYAML::emitDebugAranges(raw_ostream &OS, const DWARFYAML::Data &DI) { 142 assert(DI.DebugAranges && "unexpected emitDebugAranges() call"); 143 for (const auto &Range : *DI.DebugAranges) { 144 uint8_t AddrSize; 145 if (Range.AddrSize) 146 AddrSize = *Range.AddrSize; 147 else 148 AddrSize = DI.Is64BitAddrSize ? 8 : 4; 149 150 uint64_t Length = 4; // sizeof(version) 2 + sizeof(address_size) 1 + 151 // sizeof(segment_selector_size) 1 152 Length += 153 Range.Format == dwarf::DWARF64 ? 8 : 4; // sizeof(debug_info_offset) 154 155 const uint64_t HeaderLength = 156 Length + (Range.Format == dwarf::DWARF64 157 ? 12 158 : 4); // sizeof(unit_header) = 12 (DWARF64) or 4 (DWARF32) 159 const uint64_t PaddedHeaderLength = alignTo(HeaderLength, AddrSize * 2); 160 161 if (Range.Length) { 162 Length = *Range.Length; 163 } else { 164 Length += PaddedHeaderLength - HeaderLength; 165 Length += AddrSize * 2 * (Range.Descriptors.size() + 1); 166 } 167 168 writeInitialLength(Range.Format, Length, OS, DI.IsLittleEndian); 169 writeInteger((uint16_t)Range.Version, OS, DI.IsLittleEndian); 170 writeDWARFOffset(Range.CuOffset, Range.Format, OS, DI.IsLittleEndian); 171 writeInteger((uint8_t)AddrSize, OS, DI.IsLittleEndian); 172 writeInteger((uint8_t)Range.SegSize, OS, DI.IsLittleEndian); 173 ZeroFillBytes(OS, PaddedHeaderLength - HeaderLength); 174 175 for (const auto &Descriptor : Range.Descriptors) { 176 if (Error Err = writeVariableSizedInteger(Descriptor.Address, AddrSize, 177 OS, DI.IsLittleEndian)) 178 return createStringError(errc::not_supported, 179 "unable to write debug_aranges address: %s", 180 toString(std::move(Err)).c_str()); 181 cantFail(writeVariableSizedInteger(Descriptor.Length, AddrSize, OS, 182 DI.IsLittleEndian)); 183 } 184 ZeroFillBytes(OS, AddrSize * 2); 185 } 186 187 return Error::success(); 188 } 189 190 Error DWARFYAML::emitDebugRanges(raw_ostream &OS, const DWARFYAML::Data &DI) { 191 const size_t RangesOffset = OS.tell(); 192 uint64_t EntryIndex = 0; 193 for (const auto &DebugRanges : *DI.DebugRanges) { 194 const size_t CurrOffset = OS.tell() - RangesOffset; 195 if (DebugRanges.Offset && (uint64_t)*DebugRanges.Offset < CurrOffset) 196 return createStringError(errc::invalid_argument, 197 "'Offset' for 'debug_ranges' with index " + 198 Twine(EntryIndex) + 199 " must be greater than or equal to the " 200 "number of bytes written already (0x" + 201 Twine::utohexstr(CurrOffset) + ")"); 202 if (DebugRanges.Offset) 203 ZeroFillBytes(OS, *DebugRanges.Offset - CurrOffset); 204 205 uint8_t AddrSize; 206 if (DebugRanges.AddrSize) 207 AddrSize = *DebugRanges.AddrSize; 208 else 209 AddrSize = DI.Is64BitAddrSize ? 8 : 4; 210 for (const auto &Entry : DebugRanges.Entries) { 211 if (Error Err = writeVariableSizedInteger(Entry.LowOffset, AddrSize, OS, 212 DI.IsLittleEndian)) 213 return createStringError( 214 errc::not_supported, 215 "unable to write debug_ranges address offset: %s", 216 toString(std::move(Err)).c_str()); 217 cantFail(writeVariableSizedInteger(Entry.HighOffset, AddrSize, OS, 218 DI.IsLittleEndian)); 219 } 220 ZeroFillBytes(OS, AddrSize * 2); 221 ++EntryIndex; 222 } 223 224 return Error::success(); 225 } 226 227 static Error emitPubSection(raw_ostream &OS, const DWARFYAML::PubSection &Sect, 228 bool IsLittleEndian, bool IsGNUPubSec = false) { 229 writeInitialLength(Sect.Format, Sect.Length, OS, IsLittleEndian); 230 writeInteger((uint16_t)Sect.Version, OS, IsLittleEndian); 231 writeInteger((uint32_t)Sect.UnitOffset, OS, IsLittleEndian); 232 writeInteger((uint32_t)Sect.UnitSize, OS, IsLittleEndian); 233 for (const auto &Entry : Sect.Entries) { 234 writeInteger((uint32_t)Entry.DieOffset, OS, IsLittleEndian); 235 if (IsGNUPubSec) 236 writeInteger((uint8_t)Entry.Descriptor, OS, IsLittleEndian); 237 OS.write(Entry.Name.data(), Entry.Name.size()); 238 OS.write('\0'); 239 } 240 return Error::success(); 241 } 242 243 Error DWARFYAML::emitDebugPubnames(raw_ostream &OS, const Data &DI) { 244 assert(DI.PubNames && "unexpected emitDebugPubnames() call"); 245 return emitPubSection(OS, *DI.PubNames, DI.IsLittleEndian); 246 } 247 248 Error DWARFYAML::emitDebugPubtypes(raw_ostream &OS, const Data &DI) { 249 assert(DI.PubTypes && "unexpected emitDebugPubtypes() call"); 250 return emitPubSection(OS, *DI.PubTypes, DI.IsLittleEndian); 251 } 252 253 Error DWARFYAML::emitDebugGNUPubnames(raw_ostream &OS, const Data &DI) { 254 assert(DI.GNUPubNames && "unexpected emitDebugGNUPubnames() call"); 255 return emitPubSection(OS, *DI.GNUPubNames, DI.IsLittleEndian, 256 /*IsGNUStyle=*/true); 257 } 258 259 Error DWARFYAML::emitDebugGNUPubtypes(raw_ostream &OS, const Data &DI) { 260 assert(DI.GNUPubTypes && "unexpected emitDebugGNUPubtypes() call"); 261 return emitPubSection(OS, *DI.GNUPubTypes, DI.IsLittleEndian, 262 /*IsGNUStyle=*/true); 263 } 264 265 static Expected<uint64_t> writeDIE(const DWARFYAML::Data &DI, uint64_t CUIndex, 266 uint64_t AbbrevTableID, 267 const dwarf::FormParams &Params, 268 const DWARFYAML::Entry &Entry, 269 raw_ostream &OS, bool IsLittleEndian) { 270 uint64_t EntryBegin = OS.tell(); 271 encodeULEB128(Entry.AbbrCode, OS); 272 uint32_t AbbrCode = Entry.AbbrCode; 273 if (AbbrCode == 0 || Entry.Values.empty()) 274 return OS.tell() - EntryBegin; 275 276 Expected<DWARFYAML::Data::AbbrevTableInfo> AbbrevTableInfoOrErr = 277 DI.getAbbrevTableInfoByID(AbbrevTableID); 278 if (!AbbrevTableInfoOrErr) 279 return createStringError(errc::invalid_argument, 280 toString(AbbrevTableInfoOrErr.takeError()) + 281 " for compilation unit with index " + 282 utostr(CUIndex)); 283 284 ArrayRef<DWARFYAML::Abbrev> AbbrevDecls( 285 DI.DebugAbbrev[AbbrevTableInfoOrErr->Index].Table); 286 287 if (AbbrCode > AbbrevDecls.size()) 288 return createStringError( 289 errc::invalid_argument, 290 "abbrev code must be less than or equal to the number of " 291 "entries in abbreviation table"); 292 const DWARFYAML::Abbrev &Abbrev = AbbrevDecls[AbbrCode - 1]; 293 auto FormVal = Entry.Values.begin(); 294 auto AbbrForm = Abbrev.Attributes.begin(); 295 for (; FormVal != Entry.Values.end() && AbbrForm != Abbrev.Attributes.end(); 296 ++FormVal, ++AbbrForm) { 297 dwarf::Form Form = AbbrForm->Form; 298 bool Indirect; 299 do { 300 Indirect = false; 301 switch (Form) { 302 case dwarf::DW_FORM_addr: 303 // TODO: Test this error. 304 if (Error Err = writeVariableSizedInteger( 305 FormVal->Value, Params.AddrSize, OS, IsLittleEndian)) 306 return std::move(Err); 307 break; 308 case dwarf::DW_FORM_ref_addr: 309 // TODO: Test this error. 310 if (Error Err = writeVariableSizedInteger(FormVal->Value, 311 Params.getRefAddrByteSize(), 312 OS, IsLittleEndian)) 313 return std::move(Err); 314 break; 315 case dwarf::DW_FORM_exprloc: 316 case dwarf::DW_FORM_block: 317 encodeULEB128(FormVal->BlockData.size(), OS); 318 OS.write((const char *)FormVal->BlockData.data(), 319 FormVal->BlockData.size()); 320 break; 321 case dwarf::DW_FORM_block1: { 322 writeInteger((uint8_t)FormVal->BlockData.size(), OS, IsLittleEndian); 323 OS.write((const char *)FormVal->BlockData.data(), 324 FormVal->BlockData.size()); 325 break; 326 } 327 case dwarf::DW_FORM_block2: { 328 writeInteger((uint16_t)FormVal->BlockData.size(), OS, IsLittleEndian); 329 OS.write((const char *)FormVal->BlockData.data(), 330 FormVal->BlockData.size()); 331 break; 332 } 333 case dwarf::DW_FORM_block4: { 334 writeInteger((uint32_t)FormVal->BlockData.size(), OS, IsLittleEndian); 335 OS.write((const char *)FormVal->BlockData.data(), 336 FormVal->BlockData.size()); 337 break; 338 } 339 case dwarf::DW_FORM_strx: 340 case dwarf::DW_FORM_addrx: 341 case dwarf::DW_FORM_rnglistx: 342 case dwarf::DW_FORM_loclistx: 343 case dwarf::DW_FORM_udata: 344 case dwarf::DW_FORM_ref_udata: 345 case dwarf::DW_FORM_GNU_addr_index: 346 case dwarf::DW_FORM_GNU_str_index: 347 encodeULEB128(FormVal->Value, OS); 348 break; 349 case dwarf::DW_FORM_data1: 350 case dwarf::DW_FORM_ref1: 351 case dwarf::DW_FORM_flag: 352 case dwarf::DW_FORM_strx1: 353 case dwarf::DW_FORM_addrx1: 354 writeInteger((uint8_t)FormVal->Value, OS, IsLittleEndian); 355 break; 356 case dwarf::DW_FORM_data2: 357 case dwarf::DW_FORM_ref2: 358 case dwarf::DW_FORM_strx2: 359 case dwarf::DW_FORM_addrx2: 360 writeInteger((uint16_t)FormVal->Value, OS, IsLittleEndian); 361 break; 362 case dwarf::DW_FORM_data4: 363 case dwarf::DW_FORM_ref4: 364 case dwarf::DW_FORM_ref_sup4: 365 case dwarf::DW_FORM_strx4: 366 case dwarf::DW_FORM_addrx4: 367 writeInteger((uint32_t)FormVal->Value, OS, IsLittleEndian); 368 break; 369 case dwarf::DW_FORM_data8: 370 case dwarf::DW_FORM_ref8: 371 case dwarf::DW_FORM_ref_sup8: 372 case dwarf::DW_FORM_ref_sig8: 373 writeInteger((uint64_t)FormVal->Value, OS, IsLittleEndian); 374 break; 375 case dwarf::DW_FORM_sdata: 376 encodeSLEB128(FormVal->Value, OS); 377 break; 378 case dwarf::DW_FORM_string: 379 OS.write(FormVal->CStr.data(), FormVal->CStr.size()); 380 OS.write('\0'); 381 break; 382 case dwarf::DW_FORM_indirect: 383 encodeULEB128(FormVal->Value, OS); 384 Indirect = true; 385 Form = static_cast<dwarf::Form>((uint64_t)FormVal->Value); 386 ++FormVal; 387 break; 388 case dwarf::DW_FORM_strp: 389 case dwarf::DW_FORM_sec_offset: 390 case dwarf::DW_FORM_GNU_ref_alt: 391 case dwarf::DW_FORM_GNU_strp_alt: 392 case dwarf::DW_FORM_line_strp: 393 case dwarf::DW_FORM_strp_sup: 394 cantFail(writeVariableSizedInteger(FormVal->Value, 395 Params.getDwarfOffsetByteSize(), OS, 396 IsLittleEndian)); 397 break; 398 default: 399 break; 400 } 401 } while (Indirect); 402 } 403 404 return OS.tell() - EntryBegin; 405 } 406 407 Error DWARFYAML::emitDebugInfo(raw_ostream &OS, const DWARFYAML::Data &DI) { 408 for (uint64_t I = 0; I < DI.CompileUnits.size(); ++I) { 409 const DWARFYAML::Unit &Unit = DI.CompileUnits[I]; 410 uint8_t AddrSize; 411 if (Unit.AddrSize) 412 AddrSize = *Unit.AddrSize; 413 else 414 AddrSize = DI.Is64BitAddrSize ? 8 : 4; 415 dwarf::FormParams Params = {Unit.Version, AddrSize, Unit.Format}; 416 uint64_t Length = 3; // sizeof(version) + sizeof(address_size) 417 Length += Unit.Version >= 5 ? 1 : 0; // sizeof(unit_type) 418 Length += Params.getDwarfOffsetByteSize(); // sizeof(debug_abbrev_offset) 419 420 // Since the length of the current compilation unit is undetermined yet, we 421 // firstly write the content of the compilation unit to a buffer to 422 // calculate it and then serialize the buffer content to the actual output 423 // stream. 424 std::string EntryBuffer; 425 raw_string_ostream EntryBufferOS(EntryBuffer); 426 427 uint64_t AbbrevTableID = Unit.AbbrevTableID.value_or(I); 428 for (const DWARFYAML::Entry &Entry : Unit.Entries) { 429 if (Expected<uint64_t> EntryLength = 430 writeDIE(DI, I, AbbrevTableID, Params, Entry, EntryBufferOS, 431 DI.IsLittleEndian)) 432 Length += *EntryLength; 433 else 434 return EntryLength.takeError(); 435 } 436 437 // If the length is specified in the YAML description, we use it instead of 438 // the actual length. 439 if (Unit.Length) 440 Length = *Unit.Length; 441 442 writeInitialLength(Unit.Format, Length, OS, DI.IsLittleEndian); 443 writeInteger((uint16_t)Unit.Version, OS, DI.IsLittleEndian); 444 445 uint64_t AbbrevTableOffset = 0; 446 if (Unit.AbbrOffset) { 447 AbbrevTableOffset = *Unit.AbbrOffset; 448 } else { 449 if (Expected<DWARFYAML::Data::AbbrevTableInfo> AbbrevTableInfoOrErr = 450 DI.getAbbrevTableInfoByID(AbbrevTableID)) { 451 AbbrevTableOffset = AbbrevTableInfoOrErr->Offset; 452 } else { 453 // The current compilation unit may not have DIEs and it will not be 454 // able to find the associated abbrev table. We consume the error and 455 // assign 0 to the debug_abbrev_offset in such circumstances. 456 consumeError(AbbrevTableInfoOrErr.takeError()); 457 } 458 } 459 460 if (Unit.Version >= 5) { 461 writeInteger((uint8_t)Unit.Type, OS, DI.IsLittleEndian); 462 writeInteger((uint8_t)AddrSize, OS, DI.IsLittleEndian); 463 writeDWARFOffset(AbbrevTableOffset, Unit.Format, OS, DI.IsLittleEndian); 464 } else { 465 writeDWARFOffset(AbbrevTableOffset, Unit.Format, OS, DI.IsLittleEndian); 466 writeInteger((uint8_t)AddrSize, OS, DI.IsLittleEndian); 467 } 468 469 OS.write(EntryBuffer.data(), EntryBuffer.size()); 470 } 471 472 return Error::success(); 473 } 474 475 static void emitFileEntry(raw_ostream &OS, const DWARFYAML::File &File) { 476 OS.write(File.Name.data(), File.Name.size()); 477 OS.write('\0'); 478 encodeULEB128(File.DirIdx, OS); 479 encodeULEB128(File.ModTime, OS); 480 encodeULEB128(File.Length, OS); 481 } 482 483 static void writeExtendedOpcode(const DWARFYAML::LineTableOpcode &Op, 484 uint8_t AddrSize, bool IsLittleEndian, 485 raw_ostream &OS) { 486 // The first byte of extended opcodes is a zero byte. The next bytes are an 487 // ULEB128 integer giving the number of bytes in the instruction itself (does 488 // not include the first zero byte or the size). We serialize the instruction 489 // itself into the OpBuffer and then write the size of the buffer and the 490 // buffer to the real output stream. 491 std::string OpBuffer; 492 raw_string_ostream OpBufferOS(OpBuffer); 493 writeInteger((uint8_t)Op.SubOpcode, OpBufferOS, IsLittleEndian); 494 switch (Op.SubOpcode) { 495 case dwarf::DW_LNE_set_address: 496 cantFail(writeVariableSizedInteger(Op.Data, AddrSize, OpBufferOS, 497 IsLittleEndian)); 498 break; 499 case dwarf::DW_LNE_define_file: 500 emitFileEntry(OpBufferOS, Op.FileEntry); 501 break; 502 case dwarf::DW_LNE_set_discriminator: 503 encodeULEB128(Op.Data, OpBufferOS); 504 break; 505 case dwarf::DW_LNE_end_sequence: 506 break; 507 default: 508 for (auto OpByte : Op.UnknownOpcodeData) 509 writeInteger((uint8_t)OpByte, OpBufferOS, IsLittleEndian); 510 } 511 uint64_t ExtLen = Op.ExtLen.value_or(OpBuffer.size()); 512 encodeULEB128(ExtLen, OS); 513 OS.write(OpBuffer.data(), OpBuffer.size()); 514 } 515 516 static void writeLineTableOpcode(const DWARFYAML::LineTableOpcode &Op, 517 uint8_t OpcodeBase, uint8_t AddrSize, 518 raw_ostream &OS, bool IsLittleEndian) { 519 writeInteger((uint8_t)Op.Opcode, OS, IsLittleEndian); 520 if (Op.Opcode == 0) { 521 writeExtendedOpcode(Op, AddrSize, IsLittleEndian, OS); 522 } else if (Op.Opcode < OpcodeBase) { 523 switch (Op.Opcode) { 524 case dwarf::DW_LNS_copy: 525 case dwarf::DW_LNS_negate_stmt: 526 case dwarf::DW_LNS_set_basic_block: 527 case dwarf::DW_LNS_const_add_pc: 528 case dwarf::DW_LNS_set_prologue_end: 529 case dwarf::DW_LNS_set_epilogue_begin: 530 break; 531 532 case dwarf::DW_LNS_advance_pc: 533 case dwarf::DW_LNS_set_file: 534 case dwarf::DW_LNS_set_column: 535 case dwarf::DW_LNS_set_isa: 536 encodeULEB128(Op.Data, OS); 537 break; 538 539 case dwarf::DW_LNS_advance_line: 540 encodeSLEB128(Op.SData, OS); 541 break; 542 543 case dwarf::DW_LNS_fixed_advance_pc: 544 writeInteger((uint16_t)Op.Data, OS, IsLittleEndian); 545 break; 546 547 default: 548 for (auto OpData : Op.StandardOpcodeData) { 549 encodeULEB128(OpData, OS); 550 } 551 } 552 } 553 } 554 555 static std::vector<uint8_t> 556 getStandardOpcodeLengths(uint16_t Version, std::optional<uint8_t> OpcodeBase) { 557 // If the opcode_base field isn't specified, we returns the 558 // standard_opcode_lengths array according to the version by default. 559 std::vector<uint8_t> StandardOpcodeLengths{0, 1, 1, 1, 1, 0, 560 0, 0, 1, 0, 0, 1}; 561 if (Version == 2) { 562 // DWARF v2 uses the same first 9 standard opcodes as v3-5. 563 StandardOpcodeLengths.resize(9); 564 } else if (OpcodeBase) { 565 StandardOpcodeLengths.resize(*OpcodeBase > 0 ? *OpcodeBase - 1 : 0, 0); 566 } 567 return StandardOpcodeLengths; 568 } 569 570 Error DWARFYAML::emitDebugLine(raw_ostream &OS, const DWARFYAML::Data &DI) { 571 for (const DWARFYAML::LineTable &LineTable : DI.DebugLines) { 572 // Buffer holds the bytes following the header_length (or prologue_length in 573 // DWARFv2) field to the end of the line number program itself. 574 std::string Buffer; 575 raw_string_ostream BufferOS(Buffer); 576 577 writeInteger(LineTable.MinInstLength, BufferOS, DI.IsLittleEndian); 578 // TODO: Add support for emitting DWARFv5 line table. 579 if (LineTable.Version >= 4) 580 writeInteger(LineTable.MaxOpsPerInst, BufferOS, DI.IsLittleEndian); 581 writeInteger(LineTable.DefaultIsStmt, BufferOS, DI.IsLittleEndian); 582 writeInteger(LineTable.LineBase, BufferOS, DI.IsLittleEndian); 583 writeInteger(LineTable.LineRange, BufferOS, DI.IsLittleEndian); 584 585 std::vector<uint8_t> StandardOpcodeLengths = 586 LineTable.StandardOpcodeLengths.value_or( 587 getStandardOpcodeLengths(LineTable.Version, LineTable.OpcodeBase)); 588 uint8_t OpcodeBase = LineTable.OpcodeBase 589 ? *LineTable.OpcodeBase 590 : StandardOpcodeLengths.size() + 1; 591 writeInteger(OpcodeBase, BufferOS, DI.IsLittleEndian); 592 for (uint8_t OpcodeLength : StandardOpcodeLengths) 593 writeInteger(OpcodeLength, BufferOS, DI.IsLittleEndian); 594 595 for (StringRef IncludeDir : LineTable.IncludeDirs) { 596 BufferOS.write(IncludeDir.data(), IncludeDir.size()); 597 BufferOS.write('\0'); 598 } 599 BufferOS.write('\0'); 600 601 for (const DWARFYAML::File &File : LineTable.Files) 602 emitFileEntry(BufferOS, File); 603 BufferOS.write('\0'); 604 605 uint64_t HeaderLength = 606 LineTable.PrologueLength ? *LineTable.PrologueLength : Buffer.size(); 607 608 for (const DWARFYAML::LineTableOpcode &Op : LineTable.Opcodes) 609 writeLineTableOpcode(Op, OpcodeBase, DI.Is64BitAddrSize ? 8 : 4, BufferOS, 610 DI.IsLittleEndian); 611 612 uint64_t Length; 613 if (LineTable.Length) { 614 Length = *LineTable.Length; 615 } else { 616 Length = 2; // sizeof(version) 617 Length += 618 (LineTable.Format == dwarf::DWARF64 ? 8 : 4); // sizeof(header_length) 619 Length += Buffer.size(); 620 } 621 622 writeInitialLength(LineTable.Format, Length, OS, DI.IsLittleEndian); 623 writeInteger(LineTable.Version, OS, DI.IsLittleEndian); 624 writeDWARFOffset(HeaderLength, LineTable.Format, OS, DI.IsLittleEndian); 625 OS.write(Buffer.data(), Buffer.size()); 626 } 627 628 return Error::success(); 629 } 630 631 Error DWARFYAML::emitDebugAddr(raw_ostream &OS, const Data &DI) { 632 for (const AddrTableEntry &TableEntry : *DI.DebugAddr) { 633 uint8_t AddrSize; 634 if (TableEntry.AddrSize) 635 AddrSize = *TableEntry.AddrSize; 636 else 637 AddrSize = DI.Is64BitAddrSize ? 8 : 4; 638 639 uint64_t Length; 640 if (TableEntry.Length) 641 Length = (uint64_t)*TableEntry.Length; 642 else 643 // 2 (version) + 1 (address_size) + 1 (segment_selector_size) = 4 644 Length = 4 + (AddrSize + TableEntry.SegSelectorSize) * 645 TableEntry.SegAddrPairs.size(); 646 647 writeInitialLength(TableEntry.Format, Length, OS, DI.IsLittleEndian); 648 writeInteger((uint16_t)TableEntry.Version, OS, DI.IsLittleEndian); 649 writeInteger((uint8_t)AddrSize, OS, DI.IsLittleEndian); 650 writeInteger((uint8_t)TableEntry.SegSelectorSize, OS, DI.IsLittleEndian); 651 652 for (const SegAddrPair &Pair : TableEntry.SegAddrPairs) { 653 if (TableEntry.SegSelectorSize != yaml::Hex8{0}) 654 if (Error Err = writeVariableSizedInteger(Pair.Segment, 655 TableEntry.SegSelectorSize, 656 OS, DI.IsLittleEndian)) 657 return createStringError(errc::not_supported, 658 "unable to write debug_addr segment: %s", 659 toString(std::move(Err)).c_str()); 660 if (AddrSize != 0) 661 if (Error Err = writeVariableSizedInteger(Pair.Address, AddrSize, OS, 662 DI.IsLittleEndian)) 663 return createStringError(errc::not_supported, 664 "unable to write debug_addr address: %s", 665 toString(std::move(Err)).c_str()); 666 } 667 } 668 669 return Error::success(); 670 } 671 672 Error DWARFYAML::emitDebugStrOffsets(raw_ostream &OS, const Data &DI) { 673 assert(DI.DebugStrOffsets && "unexpected emitDebugStrOffsets() call"); 674 for (const DWARFYAML::StringOffsetsTable &Table : *DI.DebugStrOffsets) { 675 uint64_t Length; 676 if (Table.Length) 677 Length = *Table.Length; 678 else 679 // sizeof(version) + sizeof(padding) = 4 680 Length = 681 4 + Table.Offsets.size() * (Table.Format == dwarf::DWARF64 ? 8 : 4); 682 683 writeInitialLength(Table.Format, Length, OS, DI.IsLittleEndian); 684 writeInteger((uint16_t)Table.Version, OS, DI.IsLittleEndian); 685 writeInteger((uint16_t)Table.Padding, OS, DI.IsLittleEndian); 686 687 for (uint64_t Offset : Table.Offsets) 688 writeDWARFOffset(Offset, Table.Format, OS, DI.IsLittleEndian); 689 } 690 691 return Error::success(); 692 } 693 694 static Error checkOperandCount(StringRef EncodingString, 695 ArrayRef<yaml::Hex64> Values, 696 uint64_t ExpectedOperands) { 697 if (Values.size() != ExpectedOperands) 698 return createStringError( 699 errc::invalid_argument, 700 "invalid number (%zu) of operands for the operator: %s, %" PRIu64 701 " expected", 702 Values.size(), EncodingString.str().c_str(), ExpectedOperands); 703 704 return Error::success(); 705 } 706 707 static Error writeListEntryAddress(StringRef EncodingName, raw_ostream &OS, 708 uint64_t Addr, uint8_t AddrSize, 709 bool IsLittleEndian) { 710 if (Error Err = writeVariableSizedInteger(Addr, AddrSize, OS, IsLittleEndian)) 711 return createStringError(errc::invalid_argument, 712 "unable to write address for the operator %s: %s", 713 EncodingName.str().c_str(), 714 toString(std::move(Err)).c_str()); 715 716 return Error::success(); 717 } 718 719 static Expected<uint64_t> 720 writeDWARFExpression(raw_ostream &OS, 721 const DWARFYAML::DWARFOperation &Operation, 722 uint8_t AddrSize, bool IsLittleEndian) { 723 auto CheckOperands = [&](uint64_t ExpectedOperands) -> Error { 724 return checkOperandCount(dwarf::OperationEncodingString(Operation.Operator), 725 Operation.Values, ExpectedOperands); 726 }; 727 728 uint64_t ExpressionBegin = OS.tell(); 729 writeInteger((uint8_t)Operation.Operator, OS, IsLittleEndian); 730 switch (Operation.Operator) { 731 case dwarf::DW_OP_consts: 732 if (Error Err = CheckOperands(1)) 733 return std::move(Err); 734 encodeSLEB128(Operation.Values[0], OS); 735 break; 736 case dwarf::DW_OP_stack_value: 737 if (Error Err = CheckOperands(0)) 738 return std::move(Err); 739 break; 740 default: 741 StringRef EncodingStr = dwarf::OperationEncodingString(Operation.Operator); 742 return createStringError(errc::not_supported, 743 "DWARF expression: " + 744 (EncodingStr.empty() 745 ? "0x" + utohexstr(Operation.Operator) 746 : EncodingStr) + 747 " is not supported"); 748 } 749 return OS.tell() - ExpressionBegin; 750 } 751 752 static Expected<uint64_t> writeListEntry(raw_ostream &OS, 753 const DWARFYAML::RnglistEntry &Entry, 754 uint8_t AddrSize, 755 bool IsLittleEndian) { 756 uint64_t BeginOffset = OS.tell(); 757 writeInteger((uint8_t)Entry.Operator, OS, IsLittleEndian); 758 759 StringRef EncodingName = dwarf::RangeListEncodingString(Entry.Operator); 760 761 auto CheckOperands = [&](uint64_t ExpectedOperands) -> Error { 762 return checkOperandCount(EncodingName, Entry.Values, ExpectedOperands); 763 }; 764 765 auto WriteAddress = [&](uint64_t Addr) -> Error { 766 return writeListEntryAddress(EncodingName, OS, Addr, AddrSize, 767 IsLittleEndian); 768 }; 769 770 switch (Entry.Operator) { 771 case dwarf::DW_RLE_end_of_list: 772 if (Error Err = CheckOperands(0)) 773 return std::move(Err); 774 break; 775 case dwarf::DW_RLE_base_addressx: 776 if (Error Err = CheckOperands(1)) 777 return std::move(Err); 778 encodeULEB128(Entry.Values[0], OS); 779 break; 780 case dwarf::DW_RLE_startx_endx: 781 case dwarf::DW_RLE_startx_length: 782 case dwarf::DW_RLE_offset_pair: 783 if (Error Err = CheckOperands(2)) 784 return std::move(Err); 785 encodeULEB128(Entry.Values[0], OS); 786 encodeULEB128(Entry.Values[1], OS); 787 break; 788 case dwarf::DW_RLE_base_address: 789 if (Error Err = CheckOperands(1)) 790 return std::move(Err); 791 if (Error Err = WriteAddress(Entry.Values[0])) 792 return std::move(Err); 793 break; 794 case dwarf::DW_RLE_start_end: 795 if (Error Err = CheckOperands(2)) 796 return std::move(Err); 797 if (Error Err = WriteAddress(Entry.Values[0])) 798 return std::move(Err); 799 cantFail(WriteAddress(Entry.Values[1])); 800 break; 801 case dwarf::DW_RLE_start_length: 802 if (Error Err = CheckOperands(2)) 803 return std::move(Err); 804 if (Error Err = WriteAddress(Entry.Values[0])) 805 return std::move(Err); 806 encodeULEB128(Entry.Values[1], OS); 807 break; 808 } 809 810 return OS.tell() - BeginOffset; 811 } 812 813 static Expected<uint64_t> writeListEntry(raw_ostream &OS, 814 const DWARFYAML::LoclistEntry &Entry, 815 uint8_t AddrSize, 816 bool IsLittleEndian) { 817 uint64_t BeginOffset = OS.tell(); 818 writeInteger((uint8_t)Entry.Operator, OS, IsLittleEndian); 819 820 StringRef EncodingName = dwarf::LocListEncodingString(Entry.Operator); 821 822 auto CheckOperands = [&](uint64_t ExpectedOperands) -> Error { 823 return checkOperandCount(EncodingName, Entry.Values, ExpectedOperands); 824 }; 825 826 auto WriteAddress = [&](uint64_t Addr) -> Error { 827 return writeListEntryAddress(EncodingName, OS, Addr, AddrSize, 828 IsLittleEndian); 829 }; 830 831 auto WriteDWARFOperations = [&]() -> Error { 832 std::string OpBuffer; 833 raw_string_ostream OpBufferOS(OpBuffer); 834 uint64_t DescriptionsLength = 0; 835 836 for (const DWARFYAML::DWARFOperation &Op : Entry.Descriptions) { 837 if (Expected<uint64_t> OpSize = 838 writeDWARFExpression(OpBufferOS, Op, AddrSize, IsLittleEndian)) 839 DescriptionsLength += *OpSize; 840 else 841 return OpSize.takeError(); 842 } 843 844 if (Entry.DescriptionsLength) 845 DescriptionsLength = *Entry.DescriptionsLength; 846 else 847 DescriptionsLength = OpBuffer.size(); 848 849 encodeULEB128(DescriptionsLength, OS); 850 OS.write(OpBuffer.data(), OpBuffer.size()); 851 852 return Error::success(); 853 }; 854 855 switch (Entry.Operator) { 856 case dwarf::DW_LLE_end_of_list: 857 if (Error Err = CheckOperands(0)) 858 return std::move(Err); 859 break; 860 case dwarf::DW_LLE_base_addressx: 861 if (Error Err = CheckOperands(1)) 862 return std::move(Err); 863 encodeULEB128(Entry.Values[0], OS); 864 break; 865 case dwarf::DW_LLE_startx_endx: 866 case dwarf::DW_LLE_startx_length: 867 case dwarf::DW_LLE_offset_pair: 868 if (Error Err = CheckOperands(2)) 869 return std::move(Err); 870 encodeULEB128(Entry.Values[0], OS); 871 encodeULEB128(Entry.Values[1], OS); 872 if (Error Err = WriteDWARFOperations()) 873 return std::move(Err); 874 break; 875 case dwarf::DW_LLE_default_location: 876 if (Error Err = CheckOperands(0)) 877 return std::move(Err); 878 if (Error Err = WriteDWARFOperations()) 879 return std::move(Err); 880 break; 881 case dwarf::DW_LLE_base_address: 882 if (Error Err = CheckOperands(1)) 883 return std::move(Err); 884 if (Error Err = WriteAddress(Entry.Values[0])) 885 return std::move(Err); 886 break; 887 case dwarf::DW_LLE_start_end: 888 if (Error Err = CheckOperands(2)) 889 return std::move(Err); 890 if (Error Err = WriteAddress(Entry.Values[0])) 891 return std::move(Err); 892 cantFail(WriteAddress(Entry.Values[1])); 893 if (Error Err = WriteDWARFOperations()) 894 return std::move(Err); 895 break; 896 case dwarf::DW_LLE_start_length: 897 if (Error Err = CheckOperands(2)) 898 return std::move(Err); 899 if (Error Err = WriteAddress(Entry.Values[0])) 900 return std::move(Err); 901 encodeULEB128(Entry.Values[1], OS); 902 if (Error Err = WriteDWARFOperations()) 903 return std::move(Err); 904 break; 905 } 906 907 return OS.tell() - BeginOffset; 908 } 909 910 template <typename EntryType> 911 static Error writeDWARFLists(raw_ostream &OS, 912 ArrayRef<DWARFYAML::ListTable<EntryType>> Tables, 913 bool IsLittleEndian, bool Is64BitAddrSize) { 914 for (const DWARFYAML::ListTable<EntryType> &Table : Tables) { 915 // sizeof(version) + sizeof(address_size) + sizeof(segment_selector_size) + 916 // sizeof(offset_entry_count) = 8 917 uint64_t Length = 8; 918 919 uint8_t AddrSize; 920 if (Table.AddrSize) 921 AddrSize = *Table.AddrSize; 922 else 923 AddrSize = Is64BitAddrSize ? 8 : 4; 924 925 // Since the length of the current range/location lists entry is 926 // undetermined yet, we firstly write the content of the range/location 927 // lists to a buffer to calculate the length and then serialize the buffer 928 // content to the actual output stream. 929 std::string ListBuffer; 930 raw_string_ostream ListBufferOS(ListBuffer); 931 932 // Offsets holds offsets for each range/location list. The i-th element is 933 // the offset from the beginning of the first range/location list to the 934 // location of the i-th range list. 935 std::vector<uint64_t> Offsets; 936 937 for (const DWARFYAML::ListEntries<EntryType> &List : Table.Lists) { 938 Offsets.push_back(ListBufferOS.tell()); 939 if (List.Content) { 940 List.Content->writeAsBinary(ListBufferOS, UINT64_MAX); 941 Length += List.Content->binary_size(); 942 } else if (List.Entries) { 943 for (const EntryType &Entry : *List.Entries) { 944 Expected<uint64_t> EntrySize = 945 writeListEntry(ListBufferOS, Entry, AddrSize, IsLittleEndian); 946 if (!EntrySize) 947 return EntrySize.takeError(); 948 Length += *EntrySize; 949 } 950 } 951 } 952 953 // If the offset_entry_count field isn't specified, yaml2obj will infer it 954 // from the 'Offsets' field in the YAML description. If the 'Offsets' field 955 // isn't specified either, yaml2obj will infer it from the auto-generated 956 // offsets. 957 uint32_t OffsetEntryCount; 958 if (Table.OffsetEntryCount) 959 OffsetEntryCount = *Table.OffsetEntryCount; 960 else 961 OffsetEntryCount = Table.Offsets ? Table.Offsets->size() : Offsets.size(); 962 uint64_t OffsetsSize = 963 OffsetEntryCount * (Table.Format == dwarf::DWARF64 ? 8 : 4); 964 Length += OffsetsSize; 965 966 // If the length is specified in the YAML description, we use it instead of 967 // the actual length. 968 if (Table.Length) 969 Length = *Table.Length; 970 971 writeInitialLength(Table.Format, Length, OS, IsLittleEndian); 972 writeInteger((uint16_t)Table.Version, OS, IsLittleEndian); 973 writeInteger((uint8_t)AddrSize, OS, IsLittleEndian); 974 writeInteger((uint8_t)Table.SegSelectorSize, OS, IsLittleEndian); 975 writeInteger((uint32_t)OffsetEntryCount, OS, IsLittleEndian); 976 977 auto EmitOffsets = [&](ArrayRef<uint64_t> Offsets, uint64_t OffsetsSize) { 978 for (uint64_t Offset : Offsets) 979 writeDWARFOffset(OffsetsSize + Offset, Table.Format, OS, 980 IsLittleEndian); 981 }; 982 983 if (Table.Offsets) 984 EmitOffsets(ArrayRef<uint64_t>((const uint64_t *)Table.Offsets->data(), 985 Table.Offsets->size()), 986 0); 987 else if (OffsetEntryCount != 0) 988 EmitOffsets(Offsets, OffsetsSize); 989 990 OS.write(ListBuffer.data(), ListBuffer.size()); 991 } 992 993 return Error::success(); 994 } 995 996 Error DWARFYAML::emitDebugRnglists(raw_ostream &OS, const Data &DI) { 997 assert(DI.DebugRnglists && "unexpected emitDebugRnglists() call"); 998 return writeDWARFLists<DWARFYAML::RnglistEntry>( 999 OS, *DI.DebugRnglists, DI.IsLittleEndian, DI.Is64BitAddrSize); 1000 } 1001 1002 Error DWARFYAML::emitDebugLoclists(raw_ostream &OS, const Data &DI) { 1003 assert(DI.DebugLoclists && "unexpected emitDebugRnglists() call"); 1004 return writeDWARFLists<DWARFYAML::LoclistEntry>( 1005 OS, *DI.DebugLoclists, DI.IsLittleEndian, DI.Is64BitAddrSize); 1006 } 1007 1008 std::function<Error(raw_ostream &, const DWARFYAML::Data &)> 1009 DWARFYAML::getDWARFEmitterByName(StringRef SecName) { 1010 auto EmitFunc = 1011 StringSwitch< 1012 std::function<Error(raw_ostream &, const DWARFYAML::Data &)>>(SecName) 1013 .Case("debug_abbrev", DWARFYAML::emitDebugAbbrev) 1014 .Case("debug_addr", DWARFYAML::emitDebugAddr) 1015 .Case("debug_aranges", DWARFYAML::emitDebugAranges) 1016 .Case("debug_gnu_pubnames", DWARFYAML::emitDebugGNUPubnames) 1017 .Case("debug_gnu_pubtypes", DWARFYAML::emitDebugGNUPubtypes) 1018 .Case("debug_info", DWARFYAML::emitDebugInfo) 1019 .Case("debug_line", DWARFYAML::emitDebugLine) 1020 .Case("debug_loclists", DWARFYAML::emitDebugLoclists) 1021 .Case("debug_pubnames", DWARFYAML::emitDebugPubnames) 1022 .Case("debug_pubtypes", DWARFYAML::emitDebugPubtypes) 1023 .Case("debug_ranges", DWARFYAML::emitDebugRanges) 1024 .Case("debug_rnglists", DWARFYAML::emitDebugRnglists) 1025 .Case("debug_str", DWARFYAML::emitDebugStr) 1026 .Case("debug_str_offsets", DWARFYAML::emitDebugStrOffsets) 1027 .Default([&](raw_ostream &, const DWARFYAML::Data &) { 1028 return createStringError(errc::not_supported, 1029 SecName + " is not supported"); 1030 }); 1031 1032 return EmitFunc; 1033 } 1034 1035 static Error 1036 emitDebugSectionImpl(const DWARFYAML::Data &DI, StringRef Sec, 1037 StringMap<std::unique_ptr<MemoryBuffer>> &OutputBuffers) { 1038 std::string Data; 1039 raw_string_ostream DebugInfoStream(Data); 1040 1041 auto EmitFunc = DWARFYAML::getDWARFEmitterByName(Sec); 1042 1043 if (Error Err = EmitFunc(DebugInfoStream, DI)) 1044 return Err; 1045 DebugInfoStream.flush(); 1046 if (!Data.empty()) 1047 OutputBuffers[Sec] = MemoryBuffer::getMemBufferCopy(Data); 1048 1049 return Error::success(); 1050 } 1051 1052 Expected<StringMap<std::unique_ptr<MemoryBuffer>>> 1053 DWARFYAML::emitDebugSections(StringRef YAMLString, bool IsLittleEndian, 1054 bool Is64BitAddrSize) { 1055 auto CollectDiagnostic = [](const SMDiagnostic &Diag, void *DiagContext) { 1056 *static_cast<SMDiagnostic *>(DiagContext) = Diag; 1057 }; 1058 1059 SMDiagnostic GeneratedDiag; 1060 yaml::Input YIn(YAMLString, /*Ctxt=*/nullptr, CollectDiagnostic, 1061 &GeneratedDiag); 1062 1063 DWARFYAML::Data DI; 1064 DI.IsLittleEndian = IsLittleEndian; 1065 DI.Is64BitAddrSize = Is64BitAddrSize; 1066 1067 YIn >> DI; 1068 if (YIn.error()) 1069 return createStringError(YIn.error(), GeneratedDiag.getMessage()); 1070 1071 StringMap<std::unique_ptr<MemoryBuffer>> DebugSections; 1072 Error Err = Error::success(); 1073 1074 for (StringRef SecName : DI.getNonEmptySectionNames()) 1075 Err = joinErrors(std::move(Err), 1076 emitDebugSectionImpl(DI, SecName, DebugSections)); 1077 1078 if (Err) 1079 return std::move(Err); 1080 return std::move(DebugSections); 1081 } 1082