1 //===- X86Disassembler.cpp - Disassembler for x86 and x86_64 ----*- 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. 11 // It contains code to translate the data produced by the decoder into 12 // MCInsts. 13 // Documentation for the disassembler can be found in X86Disassembler.h. 14 // 15 //===----------------------------------------------------------------------===// 16 17 #include "X86Disassembler.h" 18 #include "X86DisassemblerDecoder.h" 19 20 #include "llvm/MC/EDInstInfo.h" 21 #include "llvm/MC/MCDisassembler.h" 22 #include "llvm/MC/MCDisassembler.h" 23 #include "llvm/MC/MCInst.h" 24 #include "llvm/Target/TargetRegistry.h" 25 #include "llvm/Support/Debug.h" 26 #include "llvm/Support/MemoryObject.h" 27 #include "llvm/Support/raw_ostream.h" 28 29 #include "X86GenRegisterNames.inc" 30 #include "X86GenEDInfo.inc" 31 32 using namespace llvm; 33 using namespace llvm::X86Disassembler; 34 35 void x86DisassemblerDebug(const char *file, 36 unsigned line, 37 const char *s) { 38 dbgs() << file << ":" << line << ": " << s; 39 } 40 41 #define debug(s) DEBUG(x86DisassemblerDebug(__FILE__, __LINE__, s)); 42 43 namespace llvm { 44 45 // Fill-ins to make the compiler happy. These constants are never actually 46 // assigned; they are just filler to make an automatically-generated switch 47 // statement work. 48 namespace X86 { 49 enum { 50 BX_SI = 500, 51 BX_DI = 501, 52 BP_SI = 502, 53 BP_DI = 503, 54 sib = 504, 55 sib64 = 505 56 }; 57 } 58 59 extern Target TheX86_32Target, TheX86_64Target; 60 61 } 62 63 static bool translateInstruction(MCInst &target, 64 InternalInstruction &source); 65 66 X86GenericDisassembler::X86GenericDisassembler(DisassemblerMode mode) : 67 MCDisassembler(), 68 fMode(mode) { 69 } 70 71 X86GenericDisassembler::~X86GenericDisassembler() { 72 } 73 74 EDInstInfo *X86GenericDisassembler::getEDInfo() const { 75 return instInfoX86; 76 } 77 78 /// regionReader - a callback function that wraps the readByte method from 79 /// MemoryObject. 80 /// 81 /// @param arg - The generic callback parameter. In this case, this should 82 /// be a pointer to a MemoryObject. 83 /// @param byte - A pointer to the byte to be read. 84 /// @param address - The address to be read. 85 static int regionReader(void* arg, uint8_t* byte, uint64_t address) { 86 MemoryObject* region = static_cast<MemoryObject*>(arg); 87 return region->readByte(address, byte); 88 } 89 90 /// logger - a callback function that wraps the operator<< method from 91 /// raw_ostream. 92 /// 93 /// @param arg - The generic callback parameter. This should be a pointe 94 /// to a raw_ostream. 95 /// @param log - A string to be logged. logger() adds a newline. 96 static void logger(void* arg, const char* log) { 97 if (!arg) 98 return; 99 100 raw_ostream &vStream = *(static_cast<raw_ostream*>(arg)); 101 vStream << log << "\n"; 102 } 103 104 // 105 // Public interface for the disassembler 106 // 107 108 bool X86GenericDisassembler::getInstruction(MCInst &instr, 109 uint64_t &size, 110 const MemoryObject ®ion, 111 uint64_t address, 112 raw_ostream &vStream) const { 113 InternalInstruction internalInstr; 114 115 int ret = decodeInstruction(&internalInstr, 116 regionReader, 117 (void*)®ion, 118 logger, 119 (void*)&vStream, 120 address, 121 fMode); 122 123 if (ret) { 124 size = internalInstr.readerCursor - address; 125 return false; 126 } 127 else { 128 size = internalInstr.length; 129 return !translateInstruction(instr, internalInstr); 130 } 131 } 132 133 // 134 // Private code that translates from struct InternalInstructions to MCInsts. 135 // 136 137 /// translateRegister - Translates an internal register to the appropriate LLVM 138 /// register, and appends it as an operand to an MCInst. 139 /// 140 /// @param mcInst - The MCInst to append to. 141 /// @param reg - The Reg to append. 142 static void translateRegister(MCInst &mcInst, Reg reg) { 143 #define ENTRY(x) X86::x, 144 uint8_t llvmRegnums[] = { 145 ALL_REGS 146 0 147 }; 148 #undef ENTRY 149 150 uint8_t llvmRegnum = llvmRegnums[reg]; 151 mcInst.addOperand(MCOperand::CreateReg(llvmRegnum)); 152 } 153 154 /// translateImmediate - Appends an immediate operand to an MCInst. 155 /// 156 /// @param mcInst - The MCInst to append to. 157 /// @param immediate - The immediate value to append. 158 /// @param operand - The operand, as stored in the descriptor table. 159 /// @param insn - The internal instruction. 160 static void translateImmediate(MCInst &mcInst, uint64_t immediate, 161 const OperandSpecifier &operand, 162 InternalInstruction &insn) { 163 // Sign-extend the immediate if necessary. 164 165 OperandType type = operand.type; 166 167 if (type == TYPE_RELv) { 168 switch (insn.displacementSize) { 169 default: 170 break; 171 case 8: 172 type = TYPE_MOFFS8; 173 break; 174 case 16: 175 type = TYPE_MOFFS16; 176 break; 177 case 32: 178 type = TYPE_MOFFS32; 179 break; 180 case 64: 181 type = TYPE_MOFFS64; 182 break; 183 } 184 } 185 186 switch (type) { 187 case TYPE_MOFFS8: 188 case TYPE_REL8: 189 if(immediate & 0x80) 190 immediate |= ~(0xffull); 191 break; 192 case TYPE_MOFFS16: 193 if(immediate & 0x8000) 194 immediate |= ~(0xffffull); 195 break; 196 case TYPE_MOFFS32: 197 case TYPE_REL32: 198 case TYPE_REL64: 199 if(immediate & 0x80000000) 200 immediate |= ~(0xffffffffull); 201 break; 202 case TYPE_MOFFS64: 203 default: 204 // operand is 64 bits wide. Do nothing. 205 break; 206 } 207 208 mcInst.addOperand(MCOperand::CreateImm(immediate)); 209 } 210 211 /// translateRMRegister - Translates a register stored in the R/M field of the 212 /// ModR/M byte to its LLVM equivalent and appends it to an MCInst. 213 /// @param mcInst - The MCInst to append to. 214 /// @param insn - The internal instruction to extract the R/M field 215 /// from. 216 /// @return - 0 on success; -1 otherwise 217 static bool translateRMRegister(MCInst &mcInst, 218 InternalInstruction &insn) { 219 if (insn.eaBase == EA_BASE_sib || insn.eaBase == EA_BASE_sib64) { 220 debug("A R/M register operand may not have a SIB byte"); 221 return true; 222 } 223 224 switch (insn.eaBase) { 225 default: 226 debug("Unexpected EA base register"); 227 return true; 228 case EA_BASE_NONE: 229 debug("EA_BASE_NONE for ModR/M base"); 230 return true; 231 #define ENTRY(x) case EA_BASE_##x: 232 ALL_EA_BASES 233 #undef ENTRY 234 debug("A R/M register operand may not have a base; " 235 "the operand must be a register."); 236 return true; 237 #define ENTRY(x) \ 238 case EA_REG_##x: \ 239 mcInst.addOperand(MCOperand::CreateReg(X86::x)); break; 240 ALL_REGS 241 #undef ENTRY 242 } 243 244 return false; 245 } 246 247 /// translateRMMemory - Translates a memory operand stored in the Mod and R/M 248 /// fields of an internal instruction (and possibly its SIB byte) to a memory 249 /// operand in LLVM's format, and appends it to an MCInst. 250 /// 251 /// @param mcInst - The MCInst to append to. 252 /// @param insn - The instruction to extract Mod, R/M, and SIB fields 253 /// from. 254 /// @return - 0 on success; nonzero otherwise 255 static bool translateRMMemory(MCInst &mcInst, InternalInstruction &insn) { 256 // Addresses in an MCInst are represented as five operands: 257 // 1. basereg (register) The R/M base, or (if there is a SIB) the 258 // SIB base 259 // 2. scaleamount (immediate) 1, or (if there is a SIB) the specified 260 // scale amount 261 // 3. indexreg (register) x86_registerNONE, or (if there is a SIB) 262 // the index (which is multiplied by the 263 // scale amount) 264 // 4. displacement (immediate) 0, or the displacement if there is one 265 // 5. segmentreg (register) x86_registerNONE for now, but could be set 266 // if we have segment overrides 267 268 MCOperand baseReg; 269 MCOperand scaleAmount; 270 MCOperand indexReg; 271 MCOperand displacement; 272 MCOperand segmentReg; 273 274 if (insn.eaBase == EA_BASE_sib || insn.eaBase == EA_BASE_sib64) { 275 if (insn.sibBase != SIB_BASE_NONE) { 276 switch (insn.sibBase) { 277 default: 278 debug("Unexpected sibBase"); 279 return true; 280 #define ENTRY(x) \ 281 case SIB_BASE_##x: \ 282 baseReg = MCOperand::CreateReg(X86::x); break; 283 ALL_SIB_BASES 284 #undef ENTRY 285 } 286 } else { 287 baseReg = MCOperand::CreateReg(0); 288 } 289 290 if (insn.sibIndex != SIB_INDEX_NONE) { 291 switch (insn.sibIndex) { 292 default: 293 debug("Unexpected sibIndex"); 294 return true; 295 #define ENTRY(x) \ 296 case SIB_INDEX_##x: \ 297 indexReg = MCOperand::CreateReg(X86::x); break; 298 EA_BASES_32BIT 299 EA_BASES_64BIT 300 #undef ENTRY 301 } 302 } else { 303 indexReg = MCOperand::CreateReg(0); 304 } 305 306 scaleAmount = MCOperand::CreateImm(insn.sibScale); 307 } else { 308 switch (insn.eaBase) { 309 case EA_BASE_NONE: 310 if (insn.eaDisplacement == EA_DISP_NONE) { 311 debug("EA_BASE_NONE and EA_DISP_NONE for ModR/M base"); 312 return true; 313 } 314 if (insn.mode == MODE_64BIT) 315 baseReg = MCOperand::CreateReg(X86::RIP); // Section 2.2.1.6 316 else 317 baseReg = MCOperand::CreateReg(0); 318 319 indexReg = MCOperand::CreateReg(0); 320 break; 321 case EA_BASE_BX_SI: 322 baseReg = MCOperand::CreateReg(X86::BX); 323 indexReg = MCOperand::CreateReg(X86::SI); 324 break; 325 case EA_BASE_BX_DI: 326 baseReg = MCOperand::CreateReg(X86::BX); 327 indexReg = MCOperand::CreateReg(X86::DI); 328 break; 329 case EA_BASE_BP_SI: 330 baseReg = MCOperand::CreateReg(X86::BP); 331 indexReg = MCOperand::CreateReg(X86::SI); 332 break; 333 case EA_BASE_BP_DI: 334 baseReg = MCOperand::CreateReg(X86::BP); 335 indexReg = MCOperand::CreateReg(X86::DI); 336 break; 337 default: 338 indexReg = MCOperand::CreateReg(0); 339 switch (insn.eaBase) { 340 default: 341 debug("Unexpected eaBase"); 342 return true; 343 // Here, we will use the fill-ins defined above. However, 344 // BX_SI, BX_DI, BP_SI, and BP_DI are all handled above and 345 // sib and sib64 were handled in the top-level if, so they're only 346 // placeholders to keep the compiler happy. 347 #define ENTRY(x) \ 348 case EA_BASE_##x: \ 349 baseReg = MCOperand::CreateReg(X86::x); break; 350 ALL_EA_BASES 351 #undef ENTRY 352 #define ENTRY(x) case EA_REG_##x: 353 ALL_REGS 354 #undef ENTRY 355 debug("A R/M memory operand may not be a register; " 356 "the base field must be a base."); 357 return true; 358 } 359 } 360 361 scaleAmount = MCOperand::CreateImm(1); 362 } 363 364 displacement = MCOperand::CreateImm(insn.displacement); 365 366 static const uint8_t segmentRegnums[SEG_OVERRIDE_max] = { 367 0, // SEG_OVERRIDE_NONE 368 X86::CS, 369 X86::SS, 370 X86::DS, 371 X86::ES, 372 X86::FS, 373 X86::GS 374 }; 375 376 segmentReg = MCOperand::CreateReg(segmentRegnums[insn.segmentOverride]); 377 378 mcInst.addOperand(baseReg); 379 mcInst.addOperand(scaleAmount); 380 mcInst.addOperand(indexReg); 381 mcInst.addOperand(displacement); 382 mcInst.addOperand(segmentReg); 383 return false; 384 } 385 386 /// translateRM - Translates an operand stored in the R/M (and possibly SIB) 387 /// byte of an instruction to LLVM form, and appends it to an MCInst. 388 /// 389 /// @param mcInst - The MCInst to append to. 390 /// @param operand - The operand, as stored in the descriptor table. 391 /// @param insn - The instruction to extract Mod, R/M, and SIB fields 392 /// from. 393 /// @return - 0 on success; nonzero otherwise 394 static bool translateRM(MCInst &mcInst, const OperandSpecifier &operand, 395 InternalInstruction &insn) { 396 switch (operand.type) { 397 default: 398 debug("Unexpected type for a R/M operand"); 399 return true; 400 case TYPE_R8: 401 case TYPE_R16: 402 case TYPE_R32: 403 case TYPE_R64: 404 case TYPE_Rv: 405 case TYPE_MM: 406 case TYPE_MM32: 407 case TYPE_MM64: 408 case TYPE_XMM: 409 case TYPE_XMM32: 410 case TYPE_XMM64: 411 case TYPE_XMM128: 412 case TYPE_DEBUGREG: 413 case TYPE_CONTROLREG: 414 return translateRMRegister(mcInst, insn); 415 case TYPE_M: 416 case TYPE_M8: 417 case TYPE_M16: 418 case TYPE_M32: 419 case TYPE_M64: 420 case TYPE_M128: 421 case TYPE_M512: 422 case TYPE_Mv: 423 case TYPE_M32FP: 424 case TYPE_M64FP: 425 case TYPE_M80FP: 426 case TYPE_M16INT: 427 case TYPE_M32INT: 428 case TYPE_M64INT: 429 case TYPE_M1616: 430 case TYPE_M1632: 431 case TYPE_M1664: 432 case TYPE_LEA: 433 return translateRMMemory(mcInst, insn); 434 } 435 } 436 437 /// translateFPRegister - Translates a stack position on the FPU stack to its 438 /// LLVM form, and appends it to an MCInst. 439 /// 440 /// @param mcInst - The MCInst to append to. 441 /// @param stackPos - The stack position to translate. 442 /// @return - 0 on success; nonzero otherwise. 443 static bool translateFPRegister(MCInst &mcInst, 444 uint8_t stackPos) { 445 if (stackPos >= 8) { 446 debug("Invalid FP stack position"); 447 return true; 448 } 449 450 mcInst.addOperand(MCOperand::CreateReg(X86::ST0 + stackPos)); 451 452 return false; 453 } 454 455 /// translateOperand - Translates an operand stored in an internal instruction 456 /// to LLVM's format and appends it to an MCInst. 457 /// 458 /// @param mcInst - The MCInst to append to. 459 /// @param operand - The operand, as stored in the descriptor table. 460 /// @param insn - The internal instruction. 461 /// @return - false on success; true otherwise. 462 static bool translateOperand(MCInst &mcInst, const OperandSpecifier &operand, 463 InternalInstruction &insn) { 464 switch (operand.encoding) { 465 default: 466 debug("Unhandled operand encoding during translation"); 467 return true; 468 case ENCODING_REG: 469 translateRegister(mcInst, insn.reg); 470 return false; 471 case ENCODING_RM: 472 return translateRM(mcInst, operand, insn); 473 case ENCODING_CB: 474 case ENCODING_CW: 475 case ENCODING_CD: 476 case ENCODING_CP: 477 case ENCODING_CO: 478 case ENCODING_CT: 479 debug("Translation of code offsets isn't supported."); 480 return true; 481 case ENCODING_IB: 482 case ENCODING_IW: 483 case ENCODING_ID: 484 case ENCODING_IO: 485 case ENCODING_Iv: 486 case ENCODING_Ia: 487 translateImmediate(mcInst, 488 insn.immediates[insn.numImmediatesTranslated++], 489 operand, 490 insn); 491 return false; 492 case ENCODING_RB: 493 case ENCODING_RW: 494 case ENCODING_RD: 495 case ENCODING_RO: 496 translateRegister(mcInst, insn.opcodeRegister); 497 return false; 498 case ENCODING_I: 499 return translateFPRegister(mcInst, insn.opcodeModifier); 500 case ENCODING_Rv: 501 translateRegister(mcInst, insn.opcodeRegister); 502 return false; 503 case ENCODING_DUP: 504 return translateOperand(mcInst, 505 insn.spec->operands[operand.type - TYPE_DUP0], 506 insn); 507 } 508 } 509 510 /// translateInstruction - Translates an internal instruction and all its 511 /// operands to an MCInst. 512 /// 513 /// @param mcInst - The MCInst to populate with the instruction's data. 514 /// @param insn - The internal instruction. 515 /// @return - false on success; true otherwise. 516 static bool translateInstruction(MCInst &mcInst, 517 InternalInstruction &insn) { 518 if (!insn.spec) { 519 debug("Instruction has no specification"); 520 return true; 521 } 522 523 mcInst.setOpcode(insn.instructionID); 524 525 int index; 526 527 insn.numImmediatesTranslated = 0; 528 529 for (index = 0; index < X86_MAX_OPERANDS; ++index) { 530 if (insn.spec->operands[index].encoding != ENCODING_NONE) { 531 if (translateOperand(mcInst, insn.spec->operands[index], insn)) { 532 return true; 533 } 534 } 535 } 536 537 return false; 538 } 539 540 static MCDisassembler *createX86_32Disassembler(const Target &T) { 541 return new X86Disassembler::X86_32Disassembler; 542 } 543 544 static MCDisassembler *createX86_64Disassembler(const Target &T) { 545 return new X86Disassembler::X86_64Disassembler; 546 } 547 548 extern "C" void LLVMInitializeX86Disassembler() { 549 // Register the disassembler. 550 TargetRegistry::RegisterMCDisassembler(TheX86_32Target, 551 createX86_32Disassembler); 552 TargetRegistry::RegisterMCDisassembler(TheX86_64Target, 553 createX86_64Disassembler); 554 } 555