MCAssembler.h revision 81e400092f55c2eba157172bfc0dd0df8317638d
1//===- MCAssembler.h - Object File Generation -------------------*- C++ -*-===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9 10#ifndef LLVM_MC_MCASSEMBLER_H 11#define LLVM_MC_MCASSEMBLER_H 12 13#include "llvm/ADT/SmallString.h" 14#include "llvm/ADT/ilist.h" 15#include "llvm/ADT/ilist_node.h" 16#include "llvm/MC/MCValue.h" 17#include "llvm/Support/Casting.h" 18#include "llvm/Support/DataTypes.h" 19#include <vector> // FIXME: Shouldn't be needed. 20 21namespace llvm { 22class raw_ostream; 23class MCAssembler; 24class MCSection; 25class MCSectionData; 26 27class MCFragment : public ilist_node<MCFragment> { 28 MCFragment(const MCFragment&); // DO NOT IMPLEMENT 29 void operator=(const MCFragment&); // DO NOT IMPLEMENT 30 31public: 32 enum FragmentType { 33 FT_Data, 34 FT_Align, 35 FT_Fill, 36 FT_Org 37 }; 38 39private: 40 FragmentType Kind; 41 42 /// Parent - The data for the section this fragment is in. 43 MCSectionData *Parent; 44 45 /// @name Assembler Backend Data 46 /// @{ 47 // 48 // FIXME: This could all be kept private to the assembler implementation. 49 50 /// Offset - The offset of this fragment in its section. This is ~0 until 51 /// initialized. 52 uint64_t Offset; 53 54 /// FileSize - The file size of this section. This is ~0 until initialized. 55 uint64_t FileSize; 56 57 /// @} 58 59protected: 60 MCFragment(FragmentType _Kind, MCSectionData *_Parent = 0); 61 62public: 63 // Only for sentinel. 64 MCFragment(); 65 virtual ~MCFragment(); 66 67 FragmentType getKind() const { return Kind; } 68 69 MCSectionData *getParent() const { return Parent; } 70 void setParent(MCSectionData *Value) { Parent = Value; } 71 72 // FIXME: This should be abstract, fix sentinel. 73 virtual uint64_t getMaxFileSize() const { 74 assert(0 && "Invalid getMaxFileSize call!"); 75 return 0; 76 }; 77 78 /// @name Assembler Backend Support 79 /// @{ 80 // 81 // FIXME: This could all be kept private to the assembler implementation. 82 83 uint64_t getAddress() const; 84 85 uint64_t getFileSize() const { 86 assert(FileSize != ~UINT64_C(0) && "File size not set!"); 87 return FileSize; 88 } 89 void setFileSize(uint64_t Value) { 90 assert(Value <= getMaxFileSize() && "Invalid file size!"); 91 FileSize = Value; 92 } 93 94 uint64_t getOffset() const { 95 assert(Offset != ~UINT64_C(0) && "File offset not set!"); 96 return Offset; 97 } 98 void setOffset(uint64_t Value) { Offset = Value; } 99 100 /// @} 101 102 static bool classof(const MCFragment *O) { return true; } 103}; 104 105class MCDataFragment : public MCFragment { 106 SmallString<32> Contents; 107 108public: 109 MCDataFragment(MCSectionData *SD = 0) : MCFragment(FT_Data, SD) {} 110 111 /// @name Accessors 112 /// @{ 113 114 uint64_t getMaxFileSize() const { 115 return Contents.size(); 116 } 117 118 SmallString<32> &getContents() { return Contents; } 119 const SmallString<32> &getContents() const { return Contents; } 120 121 /// @} 122 123 static bool classof(const MCFragment *F) { 124 return F->getKind() == MCFragment::FT_Data; 125 } 126 static bool classof(const MCDataFragment *) { return true; } 127}; 128 129class MCAlignFragment : public MCFragment { 130 /// Alignment - The alignment to ensure, in bytes. 131 unsigned Alignment; 132 133 /// Value - Value to use for filling padding bytes. 134 int64_t Value; 135 136 /// ValueSize - The size of the integer (in bytes) of \arg Value. 137 unsigned ValueSize; 138 139 /// MaxBytesToEmit - The maximum number of bytes to emit; if the alignment 140 /// cannot be satisfied in this width then this fragment is ignored. 141 unsigned MaxBytesToEmit; 142 143public: 144 MCAlignFragment(unsigned _Alignment, int64_t _Value, unsigned _ValueSize, 145 unsigned _MaxBytesToEmit, MCSectionData *SD = 0) 146 : MCFragment(FT_Align, SD), Alignment(_Alignment), 147 Value(_Value),ValueSize(_ValueSize), 148 MaxBytesToEmit(_MaxBytesToEmit) {} 149 150 /// @name Accessors 151 /// @{ 152 153 uint64_t getMaxFileSize() const { 154 return std::max(Alignment - 1, MaxBytesToEmit); 155 } 156 157 unsigned getAlignment() const { return Alignment; } 158 159 int64_t getValue() const { return Value; } 160 161 unsigned getValueSize() const { return ValueSize; } 162 163 unsigned getMaxBytesToEmit() const { return MaxBytesToEmit; } 164 165 /// @} 166 167 static bool classof(const MCFragment *F) { 168 return F->getKind() == MCFragment::FT_Align; 169 } 170 static bool classof(const MCAlignFragment *) { return true; } 171}; 172 173class MCFillFragment : public MCFragment { 174 /// Value - Value to use for filling bytes. 175 MCValue Value; 176 177 /// ValueSize - The size (in bytes) of \arg Value to use when filling. 178 unsigned ValueSize; 179 180 /// Count - The number of copies of \arg Value to insert. 181 uint64_t Count; 182 183public: 184 MCFillFragment(MCValue _Value, unsigned _ValueSize, uint64_t _Count, 185 MCSectionData *SD = 0) 186 : MCFragment(FT_Fill, SD), 187 Value(_Value), ValueSize(_ValueSize), Count(_Count) {} 188 189 /// @name Accessors 190 /// @{ 191 192 uint64_t getMaxFileSize() const { 193 return ValueSize * Count; 194 } 195 196 MCValue getValue() const { return Value; } 197 198 unsigned getValueSize() const { return ValueSize; } 199 200 uint64_t getCount() const { return Count; } 201 202 /// @} 203 204 static bool classof(const MCFragment *F) { 205 return F->getKind() == MCFragment::FT_Fill; 206 } 207 static bool classof(const MCFillFragment *) { return true; } 208}; 209 210class MCOrgFragment : public MCFragment { 211 /// Offset - The offset this fragment should start at. 212 MCValue Offset; 213 214 /// Value - Value to use for filling bytes. 215 int8_t Value; 216 217public: 218 MCOrgFragment(MCValue _Offset, int8_t _Value, MCSectionData *SD = 0) 219 : MCFragment(FT_Org, SD), 220 Offset(_Offset), Value(_Value) {} 221 /// @name Accessors 222 /// @{ 223 224 uint64_t getMaxFileSize() const { 225 // FIXME: This doesn't make much sense. 226 return ~UINT64_C(0); 227 } 228 229 MCValue getOffset() const { return Offset; } 230 231 uint8_t getValue() const { return Value; } 232 233 /// @} 234 235 static bool classof(const MCFragment *F) { 236 return F->getKind() == MCFragment::FT_Org; 237 } 238 static bool classof(const MCOrgFragment *) { return true; } 239}; 240 241// FIXME: Should this be a separate class, or just merged into MCSection? Since 242// we anticipate the fast path being through an MCAssembler, the only reason to 243// keep it out is for API abstraction. 244class MCSectionData : public ilist_node<MCSectionData> { 245 MCSectionData(const MCSectionData&); // DO NOT IMPLEMENT 246 void operator=(const MCSectionData&); // DO NOT IMPLEMENT 247 248public: 249 /// Fixup - Represent a fixed size region of bytes inside some fragment which 250 /// needs to be rewritten. This region will either be rewritten by the 251 /// assembler or cause a relocation entry to be generated. 252 struct Fixup { 253 /// Fragment - The fragment containing the fixup. 254 MCFragment *Fragment; 255 256 /// Offset - The offset inside the fragment which needs to be rewritten. 257 uint64_t Offset; 258 259 /// Value - The expression to eventually write into the fragment. 260 // 261 // FIXME: We could probably get away with requiring the client to pass in an 262 // owned reference whose lifetime extends past that of the fixup. 263 MCValue Value; 264 265 /// Size - The fixup size. 266 unsigned Size; 267 268 /// FixedValue - The value to replace the fix up by. 269 // 270 // FIXME: This should not be here. 271 uint64_t FixedValue; 272 273 public: 274 Fixup(MCFragment &_Fragment, uint64_t _Offset, const MCValue &_Value, 275 unsigned _Size) 276 : Fragment(&_Fragment), Offset(_Offset), Value(_Value), Size(_Size), 277 FixedValue(0) {} 278 }; 279 280 typedef iplist<MCFragment> FragmentListType; 281 282 typedef FragmentListType::const_iterator const_iterator; 283 typedef FragmentListType::iterator iterator; 284 285 typedef std::vector<Fixup>::const_iterator const_fixup_iterator; 286 typedef std::vector<Fixup>::iterator fixup_iterator; 287 288private: 289 iplist<MCFragment> Fragments; 290 const MCSection *Section; 291 292 /// Alignment - The maximum alignment seen in this section. 293 unsigned Alignment; 294 295 /// @name Assembler Backend Data 296 /// @{ 297 // 298 // FIXME: This could all be kept private to the assembler implementation. 299 300 /// Address - The computed address of this section. This is ~0 until 301 /// initialized. 302 uint64_t Address; 303 304 /// Size - The content size of this section. This is ~0 until initialized. 305 uint64_t Size; 306 307 /// FileSize - The size of this section in the object file. This is ~0 until 308 /// initialized. 309 uint64_t FileSize; 310 311 /// LastFixupLookup - Cache for the last looked up fixup. 312 mutable unsigned LastFixupLookup; 313 314 /// Fixups - The list of fixups in this section. 315 std::vector<Fixup> Fixups; 316 317 /// @} 318 319public: 320 // Only for use as sentinel. 321 MCSectionData(); 322 MCSectionData(const MCSection &Section, MCAssembler *A = 0); 323 324 const MCSection &getSection() const { return *Section; } 325 326 unsigned getAlignment() const { return Alignment; } 327 void setAlignment(unsigned Value) { Alignment = Value; } 328 329 /// @name Fragment Access 330 /// @{ 331 332 const FragmentListType &getFragmentList() const { return Fragments; } 333 FragmentListType &getFragmentList() { return Fragments; } 334 335 iterator begin() { return Fragments.begin(); } 336 const_iterator begin() const { return Fragments.begin(); } 337 338 iterator end() { return Fragments.end(); } 339 const_iterator end() const { return Fragments.end(); } 340 341 size_t size() const { return Fragments.size(); } 342 343 bool empty() const { return Fragments.empty(); } 344 345 /// @} 346 /// @name Fixup Access 347 /// @{ 348 349 std::vector<Fixup> &getFixups() { 350 return Fixups; 351 } 352 353 fixup_iterator fixup_begin() { 354 return Fixups.begin(); 355 } 356 357 fixup_iterator fixup_end() { 358 return Fixups.end(); 359 } 360 361 size_t fixup_size() const { return Fixups.size(); } 362 363 /// @} 364 /// @name Assembler Backend Support 365 /// @{ 366 // 367 // FIXME: This could all be kept private to the assembler implementation. 368 369 /// LookupFixup - Look up the fixup for the given \arg Fragment and \arg 370 /// Offset. 371 /// 372 /// If multiple fixups exist for the same fragment and offset it is undefined 373 /// which one is returned. 374 // 375 // FIXME: This isn't horribly slow in practice, but there are much nicer 376 // solutions to applying the fixups. 377 const Fixup *LookupFixup(const MCFragment *Fragment, uint64_t Offset) const; 378 379 uint64_t getAddress() const { 380 assert(Address != ~UINT64_C(0) && "Address not set!"); 381 return Address; 382 } 383 void setAddress(uint64_t Value) { Address = Value; } 384 385 uint64_t getSize() const { 386 assert(Size != ~UINT64_C(0) && "File size not set!"); 387 return Size; 388 } 389 void setSize(uint64_t Value) { Size = Value; } 390 391 uint64_t getFileSize() const { 392 assert(FileSize != ~UINT64_C(0) && "File size not set!"); 393 return FileSize; 394 } 395 void setFileSize(uint64_t Value) { FileSize = Value; } 396 397 /// @} 398}; 399 400// FIXME: Same concerns as with SectionData. 401class MCSymbolData : public ilist_node<MCSymbolData> { 402public: 403 MCSymbol &Symbol; 404 405 /// Fragment - The fragment this symbol's value is relative to, if any. 406 MCFragment *Fragment; 407 408 /// Offset - The offset to apply to the fragment address to form this symbol's 409 /// value. 410 uint64_t Offset; 411 412 /// IsExternal - True if this symbol is visible outside this translation 413 /// unit. 414 unsigned IsExternal : 1; 415 416 /// IsPrivateExtern - True if this symbol is private extern. 417 unsigned IsPrivateExtern : 1; 418 419 /// Flags - The Flags field is used by object file implementations to store 420 /// additional per symbol information which is not easily classified. 421 uint32_t Flags; 422 423 /// Index - Index field, for use by the object file implementation. 424 uint64_t Index; 425 426public: 427 // Only for use as sentinel. 428 MCSymbolData(); 429 MCSymbolData(MCSymbol &_Symbol, MCFragment *_Fragment, uint64_t _Offset, 430 MCAssembler *A = 0); 431 432 /// @name Accessors 433 /// @{ 434 435 MCSymbol &getSymbol() const { return Symbol; } 436 437 MCFragment *getFragment() const { return Fragment; } 438 void setFragment(MCFragment *Value) { Fragment = Value; } 439 440 uint64_t getOffset() const { return Offset; } 441 void setOffset(uint64_t Value) { Offset = Value; } 442 443 /// @} 444 /// @name Symbol Attributes 445 /// @{ 446 447 bool isExternal() const { return IsExternal; } 448 void setExternal(bool Value) { IsExternal = Value; } 449 450 bool isPrivateExtern() const { return IsPrivateExtern; } 451 void setPrivateExtern(bool Value) { IsPrivateExtern = Value; } 452 453 /// getFlags - Get the (implementation defined) symbol flags. 454 uint32_t getFlags() const { return Flags; } 455 456 /// setFlags - Set the (implementation defined) symbol flags. 457 void setFlags(uint32_t Value) { Flags = Value; } 458 459 /// getIndex - Get the (implementation defined) index. 460 uint64_t getIndex() const { return Index; } 461 462 /// setIndex - Set the (implementation defined) index. 463 void setIndex(uint64_t Value) { Index = Value; } 464 465 /// @} 466}; 467 468// FIXME: This really doesn't belong here. See comments below. 469struct IndirectSymbolData { 470 MCSymbol *Symbol; 471 MCSectionData *SectionData; 472}; 473 474class MCAssembler { 475public: 476 typedef iplist<MCSectionData> SectionDataListType; 477 typedef iplist<MCSymbolData> SymbolDataListType; 478 479 typedef SectionDataListType::const_iterator const_iterator; 480 typedef SectionDataListType::iterator iterator; 481 482 typedef SymbolDataListType::const_iterator const_symbol_iterator; 483 typedef SymbolDataListType::iterator symbol_iterator; 484 485 typedef std::vector<IndirectSymbolData>::iterator indirect_symbol_iterator; 486 487private: 488 MCAssembler(const MCAssembler&); // DO NOT IMPLEMENT 489 void operator=(const MCAssembler&); // DO NOT IMPLEMENT 490 491 raw_ostream &OS; 492 493 iplist<MCSectionData> Sections; 494 495 iplist<MCSymbolData> Symbols; 496 497 std::vector<IndirectSymbolData> IndirectSymbols; 498 499 unsigned SubsectionsViaSymbols : 1; 500 501private: 502 /// LayoutSection - Assign offsets and sizes to the fragments in the section 503 /// \arg SD, and update the section size. The section file offset should 504 /// already have been computed. 505 /// 506 /// \param NextAlign - The alignment for the section end address, which may 507 /// add padding bytes to the section (these are included in the section "file" 508 /// size, but not its regular size). 509 void LayoutSection(MCSectionData &SD, unsigned NextAlign); 510 511public: 512 /// Construct a new assembler instance. 513 /// 514 /// \arg OS - The stream to output to. 515 // 516 // FIXME: How are we going to parameterize this? Two obvious options are stay 517 // concrete and require clients to pass in a target like object. The other 518 // option is to make this abstract, and have targets provide concrete 519 // implementations as we do with AsmParser. 520 MCAssembler(raw_ostream &OS); 521 ~MCAssembler(); 522 523 /// Finish - Do final processing and write the object to the output stream. 524 void Finish(); 525 526 // FIXME: This does not belong here. 527 bool getSubsectionsViaSymbols() const { 528 return SubsectionsViaSymbols; 529 } 530 void setSubsectionsViaSymbols(bool Value) { 531 SubsectionsViaSymbols = Value; 532 } 533 534 /// @name Section List Access 535 /// @{ 536 537 const SectionDataListType &getSectionList() const { return Sections; } 538 SectionDataListType &getSectionList() { return Sections; } 539 540 iterator begin() { return Sections.begin(); } 541 const_iterator begin() const { return Sections.begin(); } 542 543 iterator end() { return Sections.end(); } 544 const_iterator end() const { return Sections.end(); } 545 546 size_t size() const { return Sections.size(); } 547 548 /// @} 549 /// @name Symbol List Access 550 /// @{ 551 552 const SymbolDataListType &getSymbolList() const { return Symbols; } 553 SymbolDataListType &getSymbolList() { return Symbols; } 554 555 symbol_iterator symbol_begin() { return Symbols.begin(); } 556 const_symbol_iterator symbol_begin() const { return Symbols.begin(); } 557 558 symbol_iterator symbol_end() { return Symbols.end(); } 559 const_symbol_iterator symbol_end() const { return Symbols.end(); } 560 561 size_t symbol_size() const { return Symbols.size(); } 562 563 /// @} 564 /// @name Indirect Symbol List Access 565 /// @{ 566 567 // FIXME: This is a total hack, this should not be here. Once things are 568 // factored so that the streamer has direct access to the .o writer, it can 569 // disappear. 570 std::vector<IndirectSymbolData> &getIndirectSymbols() { 571 return IndirectSymbols; 572 } 573 574 indirect_symbol_iterator indirect_symbol_begin() { 575 return IndirectSymbols.begin(); 576 } 577 578 indirect_symbol_iterator indirect_symbol_end() { 579 return IndirectSymbols.end(); 580 } 581 582 size_t indirect_symbol_size() const { return IndirectSymbols.size(); } 583 584 /// @} 585}; 586 587} // end namespace llvm 588 589#endif 590