X86DisassemblerTables.cpp revision 8783087f69989b2350c158890b26952977d8e437
1//===- X86DisassemblerTables.cpp - Disassembler tables ----------*- 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// This file is part of the X86 Disassembler Emitter. 11// It contains the implementation of the disassembler tables. 12// Documentation for the disassembler emitter in general can be found in 13// X86DisasemblerEmitter.h. 14// 15//===----------------------------------------------------------------------===// 16 17#include "X86DisassemblerShared.h" 18#include "X86DisassemblerTables.h" 19 20#include "TableGenBackend.h" 21#include "llvm/Support/ErrorHandling.h" 22#include "llvm/Support/Format.h" 23 24#include <string> 25 26using namespace llvm; 27using namespace X86Disassembler; 28 29/// inheritsFrom - Indicates whether all instructions in one class also belong 30/// to another class. 31/// 32/// @param child - The class that may be the subset 33/// @param parent - The class that may be the superset 34/// @return - True if child is a subset of parent, false otherwise. 35static inline bool inheritsFrom(InstructionContext child, 36 InstructionContext parent) { 37 if (child == parent) 38 return true; 39 40 switch (parent) { 41 case IC: 42 return true; 43 case IC_64BIT: 44 return(inheritsFrom(child, IC_64BIT_REXW) || 45 inheritsFrom(child, IC_64BIT_OPSIZE) || 46 inheritsFrom(child, IC_64BIT_XD) || 47 inheritsFrom(child, IC_64BIT_XS)); 48 case IC_OPSIZE: 49 return(inheritsFrom(child, IC_64BIT_OPSIZE)); 50 case IC_XD: 51 return(inheritsFrom(child, IC_64BIT_XD)); 52 case IC_XS: 53 return(inheritsFrom(child, IC_64BIT_XS)); 54 case IC_64BIT_REXW: 55 return(inheritsFrom(child, IC_64BIT_REXW_XS) || 56 inheritsFrom(child, IC_64BIT_REXW_XD) || 57 inheritsFrom(child, IC_64BIT_REXW_OPSIZE)); 58 case IC_64BIT_OPSIZE: 59 return(inheritsFrom(child, IC_64BIT_REXW_OPSIZE)); 60 case IC_64BIT_XD: 61 return(inheritsFrom(child, IC_64BIT_REXW_XD)); 62 case IC_64BIT_XS: 63 return(inheritsFrom(child, IC_64BIT_REXW_XS)); 64 case IC_64BIT_REXW_XD: 65 return false; 66 case IC_64BIT_REXW_XS: 67 return false; 68 case IC_64BIT_REXW_OPSIZE: 69 return false; 70 default: 71 return false; 72 } 73} 74 75/// outranks - Indicates whether, if an instruction has two different applicable 76/// classes, which class should be preferred when performing decode. This 77/// imposes a total ordering (ties are resolved toward "lower") 78/// 79/// @param upper - The class that may be preferable 80/// @param lower - The class that may be less preferable 81/// @return - True if upper is to be preferred, false otherwise. 82static inline bool outranks(InstructionContext upper, 83 InstructionContext lower) { 84 assert(upper < IC_max); 85 assert(lower < IC_max); 86 87#define ENUM_ENTRY(n, r, d) r, 88 static int ranks[IC_max] = { 89 INSTRUCTION_CONTEXTS 90 }; 91#undef ENUM_ENTRY 92 93 return (ranks[upper] > ranks[lower]); 94} 95 96/// stringForContext - Returns a string containing the name of a particular 97/// InstructionContext, usually for diagnostic purposes. 98/// 99/// @param insnContext - The instruction class to transform to a string. 100/// @return - A statically-allocated string constant that contains the 101/// name of the instruction class. 102static inline const char* stringForContext(InstructionContext insnContext) { 103 switch (insnContext) { 104 default: 105 llvm_unreachable("Unhandled instruction class"); 106#define ENUM_ENTRY(n, r, d) case n: return #n; break; 107 INSTRUCTION_CONTEXTS 108#undef ENUM_ENTRY 109 } 110} 111 112/// stringForOperandType - Like stringForContext, but for OperandTypes. 113static inline const char* stringForOperandType(OperandType type) { 114 switch (type) { 115 default: 116 llvm_unreachable("Unhandled type"); 117#define ENUM_ENTRY(i, d) case i: return #i; 118 TYPES 119#undef ENUM_ENTRY 120 } 121} 122 123/// stringForOperandEncoding - like stringForContext, but for 124/// OperandEncodings. 125static inline const char* stringForOperandEncoding(OperandEncoding encoding) { 126 switch (encoding) { 127 default: 128 llvm_unreachable("Unhandled encoding"); 129#define ENUM_ENTRY(i, d) case i: return #i; 130 ENCODINGS 131#undef ENUM_ENTRY 132 } 133} 134 135void DisassemblerTables::emitOneID(raw_ostream &o, 136 uint32_t &i, 137 InstrUID id, 138 bool addComma) const { 139 if (id) 140 o.indent(i * 2) << format("0x%hx", id); 141 else 142 o.indent(i * 2) << 0; 143 144 if (addComma) 145 o << ", "; 146 else 147 o << " "; 148 149 o << "/* "; 150 o << InstructionSpecifiers[id].name; 151 o << "*/"; 152 153 o << "\n"; 154} 155 156/// emitEmptyTable - Emits the modRMEmptyTable, which is used as a ID table by 157/// all ModR/M decisions for instructions that are invalid for all possible 158/// ModR/M byte values. 159/// 160/// @param o - The output stream on which to emit the table. 161/// @param i - The indentation level for that output stream. 162static void emitEmptyTable(raw_ostream &o, uint32_t &i) 163{ 164 o.indent(i * 2) << "InstrUID modRMEmptyTable[1] = { 0 };" << "\n"; 165 o << "\n"; 166} 167 168/// getDecisionType - Determines whether a ModRM decision with 255 entries can 169/// be compacted by eliminating redundant information. 170/// 171/// @param decision - The decision to be compacted. 172/// @return - The compactest available representation for the decision. 173static ModRMDecisionType getDecisionType(ModRMDecision &decision) 174{ 175 bool satisfiesOneEntry = true; 176 bool satisfiesSplitRM = true; 177 178 uint16_t index; 179 180 for (index = 0; index < 256; ++index) { 181 if (decision.instructionIDs[index] != decision.instructionIDs[0]) 182 satisfiesOneEntry = false; 183 184 if (((index & 0xc0) == 0xc0) && 185 (decision.instructionIDs[index] != decision.instructionIDs[0xc0])) 186 satisfiesSplitRM = false; 187 188 if (((index & 0xc0) != 0xc0) && 189 (decision.instructionIDs[index] != decision.instructionIDs[0x00])) 190 satisfiesSplitRM = false; 191 } 192 193 if (satisfiesOneEntry) 194 return MODRM_ONEENTRY; 195 196 if (satisfiesSplitRM) 197 return MODRM_SPLITRM; 198 199 return MODRM_FULL; 200} 201 202/// stringForDecisionType - Returns a statically-allocated string corresponding 203/// to a particular decision type. 204/// 205/// @param dt - The decision type. 206/// @return - A pointer to the statically-allocated string (e.g., 207/// "MODRM_ONEENTRY" for MODRM_ONEENTRY). 208static const char* stringForDecisionType(ModRMDecisionType dt) 209{ 210#define ENUM_ENTRY(n) case n: return #n; 211 switch (dt) { 212 default: 213 llvm_unreachable("Unknown decision type"); 214 MODRMTYPES 215 }; 216#undef ENUM_ENTRY 217} 218 219/// stringForModifierType - Returns a statically-allocated string corresponding 220/// to an opcode modifier type. 221/// 222/// @param mt - The modifier type. 223/// @return - A pointer to the statically-allocated string (e.g., 224/// "MODIFIER_NONE" for MODIFIER_NONE). 225static const char* stringForModifierType(ModifierType mt) 226{ 227#define ENUM_ENTRY(n) case n: return #n; 228 switch(mt) { 229 default: 230 llvm_unreachable("Unknown modifier type"); 231 MODIFIER_TYPES 232 }; 233#undef ENUM_ENTRY 234} 235 236DisassemblerTables::DisassemblerTables() { 237 unsigned i; 238 239 for (i = 0; i < 4; i++) { 240 Tables[i] = new ContextDecision; 241 memset(Tables[i], 0, sizeof(ContextDecision)); 242 } 243 244 HasConflicts = false; 245} 246 247DisassemblerTables::~DisassemblerTables() { 248 unsigned i; 249 250 for (i = 0; i < 4; i++) 251 delete Tables[i]; 252} 253 254void DisassemblerTables::emitModRMDecision(raw_ostream &o1, 255 raw_ostream &o2, 256 uint32_t &i1, 257 uint32_t &i2, 258 ModRMDecision &decision) 259 const { 260 static uint64_t sTableNumber = 0; 261 uint64_t thisTableNumber = sTableNumber; 262 ModRMDecisionType dt = getDecisionType(decision); 263 uint16_t index; 264 265 if (dt == MODRM_ONEENTRY && decision.instructionIDs[0] == 0) 266 { 267 o2.indent(i2) << "{ /* ModRMDecision */" << "\n"; 268 i2++; 269 270 o2.indent(i2) << stringForDecisionType(dt) << "," << "\n"; 271 o2.indent(i2) << "modRMEmptyTable"; 272 273 i2--; 274 o2.indent(i2) << "}"; 275 return; 276 } 277 278 o1.indent(i1) << "InstrUID modRMTable" << thisTableNumber; 279 280 switch (dt) { 281 default: 282 llvm_unreachable("Unknown decision type"); 283 case MODRM_ONEENTRY: 284 o1 << "[1]"; 285 break; 286 case MODRM_SPLITRM: 287 o1 << "[2]"; 288 break; 289 case MODRM_FULL: 290 o1 << "[256]"; 291 break; 292 } 293 294 o1 << " = {" << "\n"; 295 i1++; 296 297 switch (dt) { 298 default: 299 llvm_unreachable("Unknown decision type"); 300 case MODRM_ONEENTRY: 301 emitOneID(o1, i1, decision.instructionIDs[0], false); 302 break; 303 case MODRM_SPLITRM: 304 emitOneID(o1, i1, decision.instructionIDs[0x00], true); // mod = 0b00 305 emitOneID(o1, i1, decision.instructionIDs[0xc0], false); // mod = 0b11 306 break; 307 case MODRM_FULL: 308 for (index = 0; index < 256; ++index) 309 emitOneID(o1, i1, decision.instructionIDs[index], index < 255); 310 break; 311 } 312 313 i1--; 314 o1.indent(i1) << "};" << "\n"; 315 o1 << "\n"; 316 317 o2.indent(i2) << "{ /* struct ModRMDecision */" << "\n"; 318 i2++; 319 320 o2.indent(i2) << stringForDecisionType(dt) << "," << "\n"; 321 o2.indent(i2) << "modRMTable" << sTableNumber << "\n"; 322 323 i2--; 324 o2.indent(i2) << "}"; 325 326 ++sTableNumber; 327} 328 329void DisassemblerTables::emitOpcodeDecision( 330 raw_ostream &o1, 331 raw_ostream &o2, 332 uint32_t &i1, 333 uint32_t &i2, 334 OpcodeDecision &decision) const { 335 uint16_t index; 336 337 o2.indent(i2) << "{ /* struct OpcodeDecision */" << "\n"; 338 i2++; 339 o2.indent(i2) << "{" << "\n"; 340 i2++; 341 342 for (index = 0; index < 256; ++index) { 343 o2.indent(i2); 344 345 o2 << "/* 0x" << format("%02hhx", index) << " */" << "\n"; 346 347 emitModRMDecision(o1, o2, i1, i2, decision.modRMDecisions[index]); 348 349 if (index < 255) 350 o2 << ","; 351 352 o2 << "\n"; 353 } 354 355 i2--; 356 o2.indent(i2) << "}" << "\n"; 357 i2--; 358 o2.indent(i2) << "}" << "\n"; 359} 360 361void DisassemblerTables::emitContextDecision( 362 raw_ostream &o1, 363 raw_ostream &o2, 364 uint32_t &i1, 365 uint32_t &i2, 366 ContextDecision &decision, 367 const char* name) const { 368 o2.indent(i2) << "struct ContextDecision " << name << " = {" << "\n"; 369 i2++; 370 o2.indent(i2) << "{ /* opcodeDecisions */" << "\n"; 371 i2++; 372 373 unsigned index; 374 375 for (index = 0; index < IC_max; ++index) { 376 o2.indent(i2) << "/* "; 377 o2 << stringForContext((InstructionContext)index); 378 o2 << " */"; 379 o2 << "\n"; 380 381 emitOpcodeDecision(o1, o2, i1, i2, decision.opcodeDecisions[index]); 382 383 if (index + 1 < IC_max) 384 o2 << ", "; 385 } 386 387 i2--; 388 o2.indent(i2) << "}" << "\n"; 389 i2--; 390 o2.indent(i2) << "};" << "\n"; 391} 392 393void DisassemblerTables::emitInstructionInfo(raw_ostream &o, uint32_t &i) 394 const { 395 o.indent(i * 2) << "struct InstructionSpecifier "; 396 o << INSTRUCTIONS_STR << "["; 397 o << InstructionSpecifiers.size(); 398 o << "] = {" << "\n"; 399 400 i++; 401 402 uint16_t numInstructions = InstructionSpecifiers.size(); 403 uint16_t index, operandIndex; 404 405 for (index = 0; index < numInstructions; ++index) { 406 o.indent(i * 2) << "{ /* " << index << " */" << "\n"; 407 i++; 408 409 o.indent(i * 2) << 410 stringForModifierType(InstructionSpecifiers[index].modifierType); 411 o << "," << "\n"; 412 413 o.indent(i * 2) << "0x"; 414 o << format("%02hhx", (uint16_t)InstructionSpecifiers[index].modifierBase); 415 o << "," << "\n"; 416 417 o.indent(i * 2) << "{" << "\n"; 418 i++; 419 420 for (operandIndex = 0; operandIndex < X86_MAX_OPERANDS; ++operandIndex) { 421 o.indent(i * 2) << "{ "; 422 o << stringForOperandEncoding(InstructionSpecifiers[index] 423 .operands[operandIndex] 424 .encoding); 425 o << ", "; 426 o << stringForOperandType(InstructionSpecifiers[index] 427 .operands[operandIndex] 428 .type); 429 o << " }"; 430 431 if (operandIndex < X86_MAX_OPERANDS - 1) 432 o << ","; 433 434 o << "\n"; 435 } 436 437 i--; 438 o.indent(i * 2) << "}," << "\n"; 439 440 o.indent(i * 2) << "\"" << InstructionSpecifiers[index].name << "\""; 441 o << "\n"; 442 443 i--; 444 o.indent(i * 2) << "}"; 445 446 if (index + 1 < numInstructions) 447 o << ","; 448 449 o << "\n"; 450 } 451 452 i--; 453 o.indent(i * 2) << "};" << "\n"; 454} 455 456void DisassemblerTables::emitContextTable(raw_ostream &o, uint32_t &i) const { 457 uint16_t index; 458 459 o.indent(i * 2) << "InstructionContext "; 460 o << CONTEXTS_STR << "[256] = {" << "\n"; 461 i++; 462 463 for (index = 0; index < 256; ++index) { 464 o.indent(i * 2); 465 466 if ((index & ATTR_64BIT) && (index & ATTR_REXW) && (index & ATTR_XS)) 467 o << "IC_64BIT_REXW_XS"; 468 else if ((index & ATTR_64BIT) && (index & ATTR_REXW) && (index & ATTR_XD)) 469 o << "IC_64BIT_REXW_XD"; 470 else if ((index & ATTR_64BIT) && (index & ATTR_REXW) && 471 (index & ATTR_OPSIZE)) 472 o << "IC_64BIT_REXW_OPSIZE"; 473 else if ((index & ATTR_64BIT) && (index & ATTR_XS)) 474 o << "IC_64BIT_XS"; 475 else if ((index & ATTR_64BIT) && (index & ATTR_XD)) 476 o << "IC_64BIT_XD"; 477 else if ((index & ATTR_64BIT) && (index & ATTR_OPSIZE)) 478 o << "IC_64BIT_OPSIZE"; 479 else if ((index & ATTR_64BIT) && (index & ATTR_REXW)) 480 o << "IC_64BIT_REXW"; 481 else if ((index & ATTR_64BIT)) 482 o << "IC_64BIT"; 483 else if (index & ATTR_XS) 484 o << "IC_XS"; 485 else if (index & ATTR_XD) 486 o << "IC_XD"; 487 else if (index & ATTR_OPSIZE) 488 o << "IC_OPSIZE"; 489 else 490 o << "IC"; 491 492 if (index < 255) 493 o << ","; 494 else 495 o << " "; 496 497 o << " /* " << index << " */"; 498 499 o << "\n"; 500 } 501 502 i--; 503 o.indent(i * 2) << "};" << "\n"; 504} 505 506void DisassemblerTables::emitContextDecisions(raw_ostream &o1, 507 raw_ostream &o2, 508 uint32_t &i1, 509 uint32_t &i2) 510 const { 511 emitContextDecision(o1, o2, i1, i2, *Tables[0], ONEBYTE_STR); 512 emitContextDecision(o1, o2, i1, i2, *Tables[1], TWOBYTE_STR); 513 emitContextDecision(o1, o2, i1, i2, *Tables[2], THREEBYTE38_STR); 514 emitContextDecision(o1, o2, i1, i2, *Tables[3], THREEBYTE3A_STR); 515} 516 517void DisassemblerTables::emit(raw_ostream &o) const { 518 uint32_t i1 = 0; 519 uint32_t i2 = 0; 520 521 std::string s1; 522 std::string s2; 523 524 raw_string_ostream o1(s1); 525 raw_string_ostream o2(s2); 526 527 emitInstructionInfo(o, i2); 528 o << "\n"; 529 530 emitContextTable(o, i2); 531 o << "\n"; 532 533 emitEmptyTable(o1, i1); 534 emitContextDecisions(o1, o2, i1, i2); 535 536 o << o1.str(); 537 o << "\n"; 538 o << o2.str(); 539 o << "\n"; 540 o << "\n"; 541} 542 543void DisassemblerTables::setTableFields(ModRMDecision &decision, 544 const ModRMFilter &filter, 545 InstrUID uid, 546 uint8_t opcode) { 547 unsigned index; 548 549 for (index = 0; index < 256; ++index) { 550 if (filter.accepts(index)) { 551 if (decision.instructionIDs[index] == uid) 552 continue; 553 554 if (decision.instructionIDs[index] != 0) { 555 InstructionSpecifier &newInfo = 556 InstructionSpecifiers[uid]; 557 InstructionSpecifier &previousInfo = 558 InstructionSpecifiers[decision.instructionIDs[index]]; 559 560 if(newInfo.filtered) 561 continue; // filtered instructions get lowest priority 562 563 if(previousInfo.name == "NOOP") 564 continue; // special case for XCHG32ar and NOOP 565 566 if (outranks(previousInfo.insnContext, newInfo.insnContext)) 567 continue; 568 569 if (previousInfo.insnContext == newInfo.insnContext && 570 !previousInfo.filtered) { 571 errs() << "Error: Primary decode conflict: "; 572 errs() << newInfo.name << " would overwrite " << previousInfo.name; 573 errs() << "\n"; 574 errs() << "ModRM " << index << "\n"; 575 errs() << "Opcode " << (uint16_t)opcode << "\n"; 576 errs() << "Context " << stringForContext(newInfo.insnContext) << "\n"; 577 HasConflicts = true; 578 } 579 } 580 581 decision.instructionIDs[index] = uid; 582 } 583 } 584} 585 586void DisassemblerTables::setTableFields(OpcodeType type, 587 InstructionContext insnContext, 588 uint8_t opcode, 589 const ModRMFilter &filter, 590 InstrUID uid) { 591 unsigned index; 592 593 ContextDecision &decision = *Tables[type]; 594 595 for (index = 0; index < IC_max; ++index) { 596 if (inheritsFrom((InstructionContext)index, 597 InstructionSpecifiers[uid].insnContext)) 598 setTableFields(decision.opcodeDecisions[index].modRMDecisions[opcode], 599 filter, 600 uid, 601 opcode); 602 } 603} 604