1 //===-- X86DisassemblerDecoder.cpp - Disassembler decoder -----------------===// 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 the implementation of the instruction decoder. 12 // Documentation for the disassembler can be found in X86Disassembler.h. 13 // 14 //===----------------------------------------------------------------------===// 15 16 #include <cstdarg> /* for va_*() */ 17 #include <cstdio> /* for vsnprintf() */ 18 #include <cstdlib> /* for exit() */ 19 #include <cstring> /* for memset() */ 20 21 #include "X86DisassemblerDecoder.h" 22 23 using namespace llvm::X86Disassembler; 24 25 /// Specifies whether a ModR/M byte is needed and (if so) which 26 /// instruction each possible value of the ModR/M byte corresponds to. Once 27 /// this information is known, we have narrowed down to a single instruction. 28 struct ModRMDecision { 29 uint8_t modrm_type; 30 uint16_t instructionIDs; 31 }; 32 33 /// Specifies which set of ModR/M->instruction tables to look at 34 /// given a particular opcode. 35 struct OpcodeDecision { 36 ModRMDecision modRMDecisions[256]; 37 }; 38 39 /// Specifies which opcode->instruction tables to look at given 40 /// a particular context (set of attributes). Since there are many possible 41 /// contexts, the decoder first uses CONTEXTS_SYM to determine which context 42 /// applies given a specific set of attributes. Hence there are only IC_max 43 /// entries in this table, rather than 2^(ATTR_max). 44 struct ContextDecision { 45 OpcodeDecision opcodeDecisions[IC_max]; 46 }; 47 48 #include "X86GenDisassemblerTables.inc" 49 50 #ifndef NDEBUG 51 #define debug(s) do { Debug(__FILE__, __LINE__, s); } while (0) 52 #else 53 #define debug(s) do { } while (0) 54 #endif 55 56 57 /* 58 * contextForAttrs - Client for the instruction context table. Takes a set of 59 * attributes and returns the appropriate decode context. 60 * 61 * @param attrMask - Attributes, from the enumeration attributeBits. 62 * @return - The InstructionContext to use when looking up an 63 * an instruction with these attributes. 64 */ 65 static InstructionContext contextForAttrs(uint16_t attrMask) { 66 return static_cast<InstructionContext>(CONTEXTS_SYM[attrMask]); 67 } 68 69 /* 70 * modRMRequired - Reads the appropriate instruction table to determine whether 71 * the ModR/M byte is required to decode a particular instruction. 72 * 73 * @param type - The opcode type (i.e., how many bytes it has). 74 * @param insnContext - The context for the instruction, as returned by 75 * contextForAttrs. 76 * @param opcode - The last byte of the instruction's opcode, not counting 77 * ModR/M extensions and escapes. 78 * @return - true if the ModR/M byte is required, false otherwise. 79 */ 80 static int modRMRequired(OpcodeType type, 81 InstructionContext insnContext, 82 uint16_t opcode) { 83 const struct ContextDecision* decision = nullptr; 84 85 switch (type) { 86 case ONEBYTE: 87 decision = &ONEBYTE_SYM; 88 break; 89 case TWOBYTE: 90 decision = &TWOBYTE_SYM; 91 break; 92 case THREEBYTE_38: 93 decision = &THREEBYTE38_SYM; 94 break; 95 case THREEBYTE_3A: 96 decision = &THREEBYTE3A_SYM; 97 break; 98 case XOP8_MAP: 99 decision = &XOP8_MAP_SYM; 100 break; 101 case XOP9_MAP: 102 decision = &XOP9_MAP_SYM; 103 break; 104 case XOPA_MAP: 105 decision = &XOPA_MAP_SYM; 106 break; 107 } 108 109 return decision->opcodeDecisions[insnContext].modRMDecisions[opcode]. 110 modrm_type != MODRM_ONEENTRY; 111 } 112 113 /* 114 * decode - Reads the appropriate instruction table to obtain the unique ID of 115 * an instruction. 116 * 117 * @param type - See modRMRequired(). 118 * @param insnContext - See modRMRequired(). 119 * @param opcode - See modRMRequired(). 120 * @param modRM - The ModR/M byte if required, or any value if not. 121 * @return - The UID of the instruction, or 0 on failure. 122 */ 123 static InstrUID decode(OpcodeType type, 124 InstructionContext insnContext, 125 uint8_t opcode, 126 uint8_t modRM) { 127 const struct ModRMDecision* dec = nullptr; 128 129 switch (type) { 130 case ONEBYTE: 131 dec = &ONEBYTE_SYM.opcodeDecisions[insnContext].modRMDecisions[opcode]; 132 break; 133 case TWOBYTE: 134 dec = &TWOBYTE_SYM.opcodeDecisions[insnContext].modRMDecisions[opcode]; 135 break; 136 case THREEBYTE_38: 137 dec = &THREEBYTE38_SYM.opcodeDecisions[insnContext].modRMDecisions[opcode]; 138 break; 139 case THREEBYTE_3A: 140 dec = &THREEBYTE3A_SYM.opcodeDecisions[insnContext].modRMDecisions[opcode]; 141 break; 142 case XOP8_MAP: 143 dec = &XOP8_MAP_SYM.opcodeDecisions[insnContext].modRMDecisions[opcode]; 144 break; 145 case XOP9_MAP: 146 dec = &XOP9_MAP_SYM.opcodeDecisions[insnContext].modRMDecisions[opcode]; 147 break; 148 case XOPA_MAP: 149 dec = &XOPA_MAP_SYM.opcodeDecisions[insnContext].modRMDecisions[opcode]; 150 break; 151 } 152 153 switch (dec->modrm_type) { 154 default: 155 debug("Corrupt table! Unknown modrm_type"); 156 return 0; 157 case MODRM_ONEENTRY: 158 return modRMTable[dec->instructionIDs]; 159 case MODRM_SPLITRM: 160 if (modFromModRM(modRM) == 0x3) 161 return modRMTable[dec->instructionIDs+1]; 162 return modRMTable[dec->instructionIDs]; 163 case MODRM_SPLITREG: 164 if (modFromModRM(modRM) == 0x3) 165 return modRMTable[dec->instructionIDs+((modRM & 0x38) >> 3)+8]; 166 return modRMTable[dec->instructionIDs+((modRM & 0x38) >> 3)]; 167 case MODRM_SPLITMISC: 168 if (modFromModRM(modRM) == 0x3) 169 return modRMTable[dec->instructionIDs+(modRM & 0x3f)+8]; 170 return modRMTable[dec->instructionIDs+((modRM & 0x38) >> 3)]; 171 case MODRM_FULL: 172 return modRMTable[dec->instructionIDs+modRM]; 173 } 174 } 175 176 /* 177 * specifierForUID - Given a UID, returns the name and operand specification for 178 * that instruction. 179 * 180 * @param uid - The unique ID for the instruction. This should be returned by 181 * decode(); specifierForUID will not check bounds. 182 * @return - A pointer to the specification for that instruction. 183 */ 184 static const struct InstructionSpecifier *specifierForUID(InstrUID uid) { 185 return &INSTRUCTIONS_SYM[uid]; 186 } 187 188 /* 189 * consumeByte - Uses the reader function provided by the user to consume one 190 * byte from the instruction's memory and advance the cursor. 191 * 192 * @param insn - The instruction with the reader function to use. The cursor 193 * for this instruction is advanced. 194 * @param byte - A pointer to a pre-allocated memory buffer to be populated 195 * with the data read. 196 * @return - 0 if the read was successful; nonzero otherwise. 197 */ 198 static int consumeByte(struct InternalInstruction* insn, uint8_t* byte) { 199 int ret = insn->reader(insn->readerArg, byte, insn->readerCursor); 200 201 if (!ret) 202 ++(insn->readerCursor); 203 204 return ret; 205 } 206 207 /* 208 * lookAtByte - Like consumeByte, but does not advance the cursor. 209 * 210 * @param insn - See consumeByte(). 211 * @param byte - See consumeByte(). 212 * @return - See consumeByte(). 213 */ 214 static int lookAtByte(struct InternalInstruction* insn, uint8_t* byte) { 215 return insn->reader(insn->readerArg, byte, insn->readerCursor); 216 } 217 218 static void unconsumeByte(struct InternalInstruction* insn) { 219 insn->readerCursor--; 220 } 221 222 #define CONSUME_FUNC(name, type) \ 223 static int name(struct InternalInstruction* insn, type* ptr) { \ 224 type combined = 0; \ 225 unsigned offset; \ 226 for (offset = 0; offset < sizeof(type); ++offset) { \ 227 uint8_t byte; \ 228 int ret = insn->reader(insn->readerArg, \ 229 &byte, \ 230 insn->readerCursor + offset); \ 231 if (ret) \ 232 return ret; \ 233 combined = combined | ((uint64_t)byte << (offset * 8)); \ 234 } \ 235 *ptr = combined; \ 236 insn->readerCursor += sizeof(type); \ 237 return 0; \ 238 } 239 240 /* 241 * consume* - Use the reader function provided by the user to consume data 242 * values of various sizes from the instruction's memory and advance the 243 * cursor appropriately. These readers perform endian conversion. 244 * 245 * @param insn - See consumeByte(). 246 * @param ptr - A pointer to a pre-allocated memory of appropriate size to 247 * be populated with the data read. 248 * @return - See consumeByte(). 249 */ 250 CONSUME_FUNC(consumeInt8, int8_t) 251 CONSUME_FUNC(consumeInt16, int16_t) 252 CONSUME_FUNC(consumeInt32, int32_t) 253 CONSUME_FUNC(consumeUInt16, uint16_t) 254 CONSUME_FUNC(consumeUInt32, uint32_t) 255 CONSUME_FUNC(consumeUInt64, uint64_t) 256 257 /* 258 * dbgprintf - Uses the logging function provided by the user to log a single 259 * message, typically without a carriage-return. 260 * 261 * @param insn - The instruction containing the logging function. 262 * @param format - See printf(). 263 * @param ... - See printf(). 264 */ 265 static void dbgprintf(struct InternalInstruction* insn, 266 const char* format, 267 ...) { 268 char buffer[256]; 269 va_list ap; 270 271 if (!insn->dlog) 272 return; 273 274 va_start(ap, format); 275 (void)vsnprintf(buffer, sizeof(buffer), format, ap); 276 va_end(ap); 277 278 insn->dlog(insn->dlogArg, buffer); 279 280 return; 281 } 282 283 /* 284 * setPrefixPresent - Marks that a particular prefix is present at a particular 285 * location. 286 * 287 * @param insn - The instruction to be marked as having the prefix. 288 * @param prefix - The prefix that is present. 289 * @param location - The location where the prefix is located (in the address 290 * space of the instruction's reader). 291 */ 292 static void setPrefixPresent(struct InternalInstruction* insn, 293 uint8_t prefix, 294 uint64_t location) 295 { 296 insn->prefixPresent[prefix] = 1; 297 insn->prefixLocations[prefix] = location; 298 } 299 300 /* 301 * isPrefixAtLocation - Queries an instruction to determine whether a prefix is 302 * present at a given location. 303 * 304 * @param insn - The instruction to be queried. 305 * @param prefix - The prefix. 306 * @param location - The location to query. 307 * @return - Whether the prefix is at that location. 308 */ 309 static bool isPrefixAtLocation(struct InternalInstruction* insn, 310 uint8_t prefix, 311 uint64_t location) 312 { 313 return insn->prefixPresent[prefix] == 1 && 314 insn->prefixLocations[prefix] == location; 315 } 316 317 /* 318 * readPrefixes - Consumes all of an instruction's prefix bytes, and marks the 319 * instruction as having them. Also sets the instruction's default operand, 320 * address, and other relevant data sizes to report operands correctly. 321 * 322 * @param insn - The instruction whose prefixes are to be read. 323 * @return - 0 if the instruction could be read until the end of the prefix 324 * bytes, and no prefixes conflicted; nonzero otherwise. 325 */ 326 static int readPrefixes(struct InternalInstruction* insn) { 327 bool isPrefix = true; 328 bool prefixGroups[4] = { false }; 329 uint64_t prefixLocation; 330 uint8_t byte = 0; 331 uint8_t nextByte; 332 333 bool hasAdSize = false; 334 bool hasOpSize = false; 335 336 dbgprintf(insn, "readPrefixes()"); 337 338 while (isPrefix) { 339 prefixLocation = insn->readerCursor; 340 341 /* If we fail reading prefixes, just stop here and let the opcode reader deal with it */ 342 if (consumeByte(insn, &byte)) 343 break; 344 345 /* 346 * If the byte is a LOCK/REP/REPNE prefix and not a part of the opcode, then 347 * break and let it be disassembled as a normal "instruction". 348 */ 349 if (insn->readerCursor - 1 == insn->startLocation && byte == 0xf0) 350 break; 351 352 if (insn->readerCursor - 1 == insn->startLocation 353 && (byte == 0xf2 || byte == 0xf3) 354 && !lookAtByte(insn, &nextByte)) 355 { 356 /* 357 * If the byte is 0xf2 or 0xf3, and any of the following conditions are 358 * met: 359 * - it is followed by a LOCK (0xf0) prefix 360 * - it is followed by an xchg instruction 361 * then it should be disassembled as a xacquire/xrelease not repne/rep. 362 */ 363 if ((byte == 0xf2 || byte == 0xf3) && 364 ((nextByte == 0xf0) || 365 ((nextByte & 0xfe) == 0x86 || (nextByte & 0xf8) == 0x90))) 366 insn->xAcquireRelease = true; 367 /* 368 * Also if the byte is 0xf3, and the following condition is met: 369 * - it is followed by a "mov mem, reg" (opcode 0x88/0x89) or 370 * "mov mem, imm" (opcode 0xc6/0xc7) instructions. 371 * then it should be disassembled as an xrelease not rep. 372 */ 373 if (byte == 0xf3 && 374 (nextByte == 0x88 || nextByte == 0x89 || 375 nextByte == 0xc6 || nextByte == 0xc7)) 376 insn->xAcquireRelease = true; 377 if (insn->mode == MODE_64BIT && (nextByte & 0xf0) == 0x40) { 378 if (consumeByte(insn, &nextByte)) 379 return -1; 380 if (lookAtByte(insn, &nextByte)) 381 return -1; 382 unconsumeByte(insn); 383 } 384 if (nextByte != 0x0f && nextByte != 0x90) 385 break; 386 } 387 388 switch (byte) { 389 case 0xf0: /* LOCK */ 390 case 0xf2: /* REPNE/REPNZ */ 391 case 0xf3: /* REP or REPE/REPZ */ 392 if (prefixGroups[0]) 393 dbgprintf(insn, "Redundant Group 1 prefix"); 394 prefixGroups[0] = true; 395 setPrefixPresent(insn, byte, prefixLocation); 396 break; 397 case 0x2e: /* CS segment override -OR- Branch not taken */ 398 case 0x36: /* SS segment override -OR- Branch taken */ 399 case 0x3e: /* DS segment override */ 400 case 0x26: /* ES segment override */ 401 case 0x64: /* FS segment override */ 402 case 0x65: /* GS segment override */ 403 switch (byte) { 404 case 0x2e: 405 insn->segmentOverride = SEG_OVERRIDE_CS; 406 break; 407 case 0x36: 408 insn->segmentOverride = SEG_OVERRIDE_SS; 409 break; 410 case 0x3e: 411 insn->segmentOverride = SEG_OVERRIDE_DS; 412 break; 413 case 0x26: 414 insn->segmentOverride = SEG_OVERRIDE_ES; 415 break; 416 case 0x64: 417 insn->segmentOverride = SEG_OVERRIDE_FS; 418 break; 419 case 0x65: 420 insn->segmentOverride = SEG_OVERRIDE_GS; 421 break; 422 default: 423 debug("Unhandled override"); 424 return -1; 425 } 426 if (prefixGroups[1]) 427 dbgprintf(insn, "Redundant Group 2 prefix"); 428 prefixGroups[1] = true; 429 setPrefixPresent(insn, byte, prefixLocation); 430 break; 431 case 0x66: /* Operand-size override */ 432 if (prefixGroups[2]) 433 dbgprintf(insn, "Redundant Group 3 prefix"); 434 prefixGroups[2] = true; 435 hasOpSize = true; 436 setPrefixPresent(insn, byte, prefixLocation); 437 break; 438 case 0x67: /* Address-size override */ 439 if (prefixGroups[3]) 440 dbgprintf(insn, "Redundant Group 4 prefix"); 441 prefixGroups[3] = true; 442 hasAdSize = true; 443 setPrefixPresent(insn, byte, prefixLocation); 444 break; 445 default: /* Not a prefix byte */ 446 isPrefix = false; 447 break; 448 } 449 450 if (isPrefix) 451 dbgprintf(insn, "Found prefix 0x%hhx", byte); 452 } 453 454 insn->vectorExtensionType = TYPE_NO_VEX_XOP; 455 456 if (byte == 0x62) { 457 uint8_t byte1, byte2; 458 459 if (consumeByte(insn, &byte1)) { 460 dbgprintf(insn, "Couldn't read second byte of EVEX prefix"); 461 return -1; 462 } 463 464 if (lookAtByte(insn, &byte2)) { 465 dbgprintf(insn, "Couldn't read third byte of EVEX prefix"); 466 return -1; 467 } 468 469 if ((insn->mode == MODE_64BIT || (byte1 & 0xc0) == 0xc0) && 470 ((~byte1 & 0xc) == 0xc) && ((byte2 & 0x4) == 0x4)) { 471 insn->vectorExtensionType = TYPE_EVEX; 472 } else { 473 unconsumeByte(insn); /* unconsume byte1 */ 474 unconsumeByte(insn); /* unconsume byte */ 475 insn->necessaryPrefixLocation = insn->readerCursor - 2; 476 } 477 478 if (insn->vectorExtensionType == TYPE_EVEX) { 479 insn->vectorExtensionPrefix[0] = byte; 480 insn->vectorExtensionPrefix[1] = byte1; 481 if (consumeByte(insn, &insn->vectorExtensionPrefix[2])) { 482 dbgprintf(insn, "Couldn't read third byte of EVEX prefix"); 483 return -1; 484 } 485 if (consumeByte(insn, &insn->vectorExtensionPrefix[3])) { 486 dbgprintf(insn, "Couldn't read fourth byte of EVEX prefix"); 487 return -1; 488 } 489 490 /* We simulate the REX prefix for simplicity's sake */ 491 if (insn->mode == MODE_64BIT) { 492 insn->rexPrefix = 0x40 493 | (wFromEVEX3of4(insn->vectorExtensionPrefix[2]) << 3) 494 | (rFromEVEX2of4(insn->vectorExtensionPrefix[1]) << 2) 495 | (xFromEVEX2of4(insn->vectorExtensionPrefix[1]) << 1) 496 | (bFromEVEX2of4(insn->vectorExtensionPrefix[1]) << 0); 497 } 498 499 dbgprintf(insn, "Found EVEX prefix 0x%hhx 0x%hhx 0x%hhx 0x%hhx", 500 insn->vectorExtensionPrefix[0], insn->vectorExtensionPrefix[1], 501 insn->vectorExtensionPrefix[2], insn->vectorExtensionPrefix[3]); 502 } 503 } else if (byte == 0xc4) { 504 uint8_t byte1; 505 506 if (lookAtByte(insn, &byte1)) { 507 dbgprintf(insn, "Couldn't read second byte of VEX"); 508 return -1; 509 } 510 511 if (insn->mode == MODE_64BIT || (byte1 & 0xc0) == 0xc0) { 512 insn->vectorExtensionType = TYPE_VEX_3B; 513 insn->necessaryPrefixLocation = insn->readerCursor - 1; 514 } else { 515 unconsumeByte(insn); 516 insn->necessaryPrefixLocation = insn->readerCursor - 1; 517 } 518 519 if (insn->vectorExtensionType == TYPE_VEX_3B) { 520 insn->vectorExtensionPrefix[0] = byte; 521 consumeByte(insn, &insn->vectorExtensionPrefix[1]); 522 consumeByte(insn, &insn->vectorExtensionPrefix[2]); 523 524 /* We simulate the REX prefix for simplicity's sake */ 525 526 if (insn->mode == MODE_64BIT) { 527 insn->rexPrefix = 0x40 528 | (wFromVEX3of3(insn->vectorExtensionPrefix[2]) << 3) 529 | (rFromVEX2of3(insn->vectorExtensionPrefix[1]) << 2) 530 | (xFromVEX2of3(insn->vectorExtensionPrefix[1]) << 1) 531 | (bFromVEX2of3(insn->vectorExtensionPrefix[1]) << 0); 532 } 533 534 dbgprintf(insn, "Found VEX prefix 0x%hhx 0x%hhx 0x%hhx", 535 insn->vectorExtensionPrefix[0], insn->vectorExtensionPrefix[1], 536 insn->vectorExtensionPrefix[2]); 537 } 538 } else if (byte == 0xc5) { 539 uint8_t byte1; 540 541 if (lookAtByte(insn, &byte1)) { 542 dbgprintf(insn, "Couldn't read second byte of VEX"); 543 return -1; 544 } 545 546 if (insn->mode == MODE_64BIT || (byte1 & 0xc0) == 0xc0) { 547 insn->vectorExtensionType = TYPE_VEX_2B; 548 } else { 549 unconsumeByte(insn); 550 } 551 552 if (insn->vectorExtensionType == TYPE_VEX_2B) { 553 insn->vectorExtensionPrefix[0] = byte; 554 consumeByte(insn, &insn->vectorExtensionPrefix[1]); 555 556 if (insn->mode == MODE_64BIT) { 557 insn->rexPrefix = 0x40 558 | (rFromVEX2of2(insn->vectorExtensionPrefix[1]) << 2); 559 } 560 561 switch (ppFromVEX2of2(insn->vectorExtensionPrefix[1])) { 562 default: 563 break; 564 case VEX_PREFIX_66: 565 hasOpSize = true; 566 break; 567 } 568 569 dbgprintf(insn, "Found VEX prefix 0x%hhx 0x%hhx", 570 insn->vectorExtensionPrefix[0], 571 insn->vectorExtensionPrefix[1]); 572 } 573 } else if (byte == 0x8f) { 574 uint8_t byte1; 575 576 if (lookAtByte(insn, &byte1)) { 577 dbgprintf(insn, "Couldn't read second byte of XOP"); 578 return -1; 579 } 580 581 if ((byte1 & 0x38) != 0x0) { /* 0 in these 3 bits is a POP instruction. */ 582 insn->vectorExtensionType = TYPE_XOP; 583 insn->necessaryPrefixLocation = insn->readerCursor - 1; 584 } else { 585 unconsumeByte(insn); 586 insn->necessaryPrefixLocation = insn->readerCursor - 1; 587 } 588 589 if (insn->vectorExtensionType == TYPE_XOP) { 590 insn->vectorExtensionPrefix[0] = byte; 591 consumeByte(insn, &insn->vectorExtensionPrefix[1]); 592 consumeByte(insn, &insn->vectorExtensionPrefix[2]); 593 594 /* We simulate the REX prefix for simplicity's sake */ 595 596 if (insn->mode == MODE_64BIT) { 597 insn->rexPrefix = 0x40 598 | (wFromXOP3of3(insn->vectorExtensionPrefix[2]) << 3) 599 | (rFromXOP2of3(insn->vectorExtensionPrefix[1]) << 2) 600 | (xFromXOP2of3(insn->vectorExtensionPrefix[1]) << 1) 601 | (bFromXOP2of3(insn->vectorExtensionPrefix[1]) << 0); 602 } 603 604 switch (ppFromXOP3of3(insn->vectorExtensionPrefix[2])) { 605 default: 606 break; 607 case VEX_PREFIX_66: 608 hasOpSize = true; 609 break; 610 } 611 612 dbgprintf(insn, "Found XOP prefix 0x%hhx 0x%hhx 0x%hhx", 613 insn->vectorExtensionPrefix[0], insn->vectorExtensionPrefix[1], 614 insn->vectorExtensionPrefix[2]); 615 } 616 } else { 617 if (insn->mode == MODE_64BIT) { 618 if ((byte & 0xf0) == 0x40) { 619 uint8_t opcodeByte; 620 621 if (lookAtByte(insn, &opcodeByte) || ((opcodeByte & 0xf0) == 0x40)) { 622 dbgprintf(insn, "Redundant REX prefix"); 623 return -1; 624 } 625 626 insn->rexPrefix = byte; 627 insn->necessaryPrefixLocation = insn->readerCursor - 2; 628 629 dbgprintf(insn, "Found REX prefix 0x%hhx", byte); 630 } else { 631 unconsumeByte(insn); 632 insn->necessaryPrefixLocation = insn->readerCursor - 1; 633 } 634 } else { 635 unconsumeByte(insn); 636 insn->necessaryPrefixLocation = insn->readerCursor - 1; 637 } 638 } 639 640 if (insn->mode == MODE_16BIT) { 641 insn->registerSize = (hasOpSize ? 4 : 2); 642 insn->addressSize = (hasAdSize ? 4 : 2); 643 insn->displacementSize = (hasAdSize ? 4 : 2); 644 insn->immediateSize = (hasOpSize ? 4 : 2); 645 } else if (insn->mode == MODE_32BIT) { 646 insn->registerSize = (hasOpSize ? 2 : 4); 647 insn->addressSize = (hasAdSize ? 2 : 4); 648 insn->displacementSize = (hasAdSize ? 2 : 4); 649 insn->immediateSize = (hasOpSize ? 2 : 4); 650 } else if (insn->mode == MODE_64BIT) { 651 if (insn->rexPrefix && wFromREX(insn->rexPrefix)) { 652 insn->registerSize = 8; 653 insn->addressSize = (hasAdSize ? 4 : 8); 654 insn->displacementSize = 4; 655 insn->immediateSize = 4; 656 } else if (insn->rexPrefix) { 657 insn->registerSize = (hasOpSize ? 2 : 4); 658 insn->addressSize = (hasAdSize ? 4 : 8); 659 insn->displacementSize = (hasOpSize ? 2 : 4); 660 insn->immediateSize = (hasOpSize ? 2 : 4); 661 } else { 662 insn->registerSize = (hasOpSize ? 2 : 4); 663 insn->addressSize = (hasAdSize ? 4 : 8); 664 insn->displacementSize = (hasOpSize ? 2 : 4); 665 insn->immediateSize = (hasOpSize ? 2 : 4); 666 } 667 } 668 669 return 0; 670 } 671 672 /* 673 * readOpcode - Reads the opcode (excepting the ModR/M byte in the case of 674 * extended or escape opcodes). 675 * 676 * @param insn - The instruction whose opcode is to be read. 677 * @return - 0 if the opcode could be read successfully; nonzero otherwise. 678 */ 679 static int readOpcode(struct InternalInstruction* insn) { 680 /* Determine the length of the primary opcode */ 681 682 uint8_t current; 683 684 dbgprintf(insn, "readOpcode()"); 685 686 insn->opcodeType = ONEBYTE; 687 688 if (insn->vectorExtensionType == TYPE_EVEX) { 689 switch (mmFromEVEX2of4(insn->vectorExtensionPrefix[1])) { 690 default: 691 dbgprintf(insn, "Unhandled mm field for instruction (0x%hhx)", 692 mmFromEVEX2of4(insn->vectorExtensionPrefix[1])); 693 return -1; 694 case VEX_LOB_0F: 695 insn->opcodeType = TWOBYTE; 696 return consumeByte(insn, &insn->opcode); 697 case VEX_LOB_0F38: 698 insn->opcodeType = THREEBYTE_38; 699 return consumeByte(insn, &insn->opcode); 700 case VEX_LOB_0F3A: 701 insn->opcodeType = THREEBYTE_3A; 702 return consumeByte(insn, &insn->opcode); 703 } 704 } else if (insn->vectorExtensionType == TYPE_VEX_3B) { 705 switch (mmmmmFromVEX2of3(insn->vectorExtensionPrefix[1])) { 706 default: 707 dbgprintf(insn, "Unhandled m-mmmm field for instruction (0x%hhx)", 708 mmmmmFromVEX2of3(insn->vectorExtensionPrefix[1])); 709 return -1; 710 case VEX_LOB_0F: 711 insn->opcodeType = TWOBYTE; 712 return consumeByte(insn, &insn->opcode); 713 case VEX_LOB_0F38: 714 insn->opcodeType = THREEBYTE_38; 715 return consumeByte(insn, &insn->opcode); 716 case VEX_LOB_0F3A: 717 insn->opcodeType = THREEBYTE_3A; 718 return consumeByte(insn, &insn->opcode); 719 } 720 } else if (insn->vectorExtensionType == TYPE_VEX_2B) { 721 insn->opcodeType = TWOBYTE; 722 return consumeByte(insn, &insn->opcode); 723 } else if (insn->vectorExtensionType == TYPE_XOP) { 724 switch (mmmmmFromXOP2of3(insn->vectorExtensionPrefix[1])) { 725 default: 726 dbgprintf(insn, "Unhandled m-mmmm field for instruction (0x%hhx)", 727 mmmmmFromVEX2of3(insn->vectorExtensionPrefix[1])); 728 return -1; 729 case XOP_MAP_SELECT_8: 730 insn->opcodeType = XOP8_MAP; 731 return consumeByte(insn, &insn->opcode); 732 case XOP_MAP_SELECT_9: 733 insn->opcodeType = XOP9_MAP; 734 return consumeByte(insn, &insn->opcode); 735 case XOP_MAP_SELECT_A: 736 insn->opcodeType = XOPA_MAP; 737 return consumeByte(insn, &insn->opcode); 738 } 739 } 740 741 if (consumeByte(insn, ¤t)) 742 return -1; 743 744 if (current == 0x0f) { 745 dbgprintf(insn, "Found a two-byte escape prefix (0x%hhx)", current); 746 747 if (consumeByte(insn, ¤t)) 748 return -1; 749 750 if (current == 0x38) { 751 dbgprintf(insn, "Found a three-byte escape prefix (0x%hhx)", current); 752 753 if (consumeByte(insn, ¤t)) 754 return -1; 755 756 insn->opcodeType = THREEBYTE_38; 757 } else if (current == 0x3a) { 758 dbgprintf(insn, "Found a three-byte escape prefix (0x%hhx)", current); 759 760 if (consumeByte(insn, ¤t)) 761 return -1; 762 763 insn->opcodeType = THREEBYTE_3A; 764 } else { 765 dbgprintf(insn, "Didn't find a three-byte escape prefix"); 766 767 insn->opcodeType = TWOBYTE; 768 } 769 } 770 771 /* 772 * At this point we have consumed the full opcode. 773 * Anything we consume from here on must be unconsumed. 774 */ 775 776 insn->opcode = current; 777 778 return 0; 779 } 780 781 static int readModRM(struct InternalInstruction* insn); 782 783 /* 784 * getIDWithAttrMask - Determines the ID of an instruction, consuming 785 * the ModR/M byte as appropriate for extended and escape opcodes, 786 * and using a supplied attribute mask. 787 * 788 * @param instructionID - A pointer whose target is filled in with the ID of the 789 * instruction. 790 * @param insn - The instruction whose ID is to be determined. 791 * @param attrMask - The attribute mask to search. 792 * @return - 0 if the ModR/M could be read when needed or was not 793 * needed; nonzero otherwise. 794 */ 795 static int getIDWithAttrMask(uint16_t* instructionID, 796 struct InternalInstruction* insn, 797 uint16_t attrMask) { 798 bool hasModRMExtension; 799 800 InstructionContext instructionClass = contextForAttrs(attrMask); 801 802 hasModRMExtension = modRMRequired(insn->opcodeType, 803 instructionClass, 804 insn->opcode); 805 806 if (hasModRMExtension) { 807 if (readModRM(insn)) 808 return -1; 809 810 *instructionID = decode(insn->opcodeType, 811 instructionClass, 812 insn->opcode, 813 insn->modRM); 814 } else { 815 *instructionID = decode(insn->opcodeType, 816 instructionClass, 817 insn->opcode, 818 0); 819 } 820 821 return 0; 822 } 823 824 /* 825 * is16BitEquivalent - Determines whether two instruction names refer to 826 * equivalent instructions but one is 16-bit whereas the other is not. 827 * 828 * @param orig - The instruction that is not 16-bit 829 * @param equiv - The instruction that is 16-bit 830 */ 831 static bool is16BitEquivalent(const char* orig, const char* equiv) { 832 off_t i; 833 834 for (i = 0;; i++) { 835 if (orig[i] == '\0' && equiv[i] == '\0') 836 return true; 837 if (orig[i] == '\0' || equiv[i] == '\0') 838 return false; 839 if (orig[i] != equiv[i]) { 840 if ((orig[i] == 'Q' || orig[i] == 'L') && equiv[i] == 'W') 841 continue; 842 if ((orig[i] == '6' || orig[i] == '3') && equiv[i] == '1') 843 continue; 844 if ((orig[i] == '4' || orig[i] == '2') && equiv[i] == '6') 845 continue; 846 return false; 847 } 848 } 849 } 850 851 /* 852 * is64Bit - Determines whether this instruction is a 64-bit instruction. 853 * 854 * @param name - The instruction that is not 16-bit 855 */ 856 static bool is64Bit(const char* name) { 857 off_t i; 858 859 for (i = 0;; ++i) { 860 if (name[i] == '\0') 861 return false; 862 if (name[i] == '6' && name[i+1] == '4') 863 return true; 864 } 865 } 866 867 /* 868 * getID - Determines the ID of an instruction, consuming the ModR/M byte as 869 * appropriate for extended and escape opcodes. Determines the attributes and 870 * context for the instruction before doing so. 871 * 872 * @param insn - The instruction whose ID is to be determined. 873 * @return - 0 if the ModR/M could be read when needed or was not needed; 874 * nonzero otherwise. 875 */ 876 static int getID(struct InternalInstruction* insn, const void *miiArg) { 877 uint16_t attrMask; 878 uint16_t instructionID; 879 880 dbgprintf(insn, "getID()"); 881 882 attrMask = ATTR_NONE; 883 884 if (insn->mode == MODE_64BIT) 885 attrMask |= ATTR_64BIT; 886 887 if (insn->vectorExtensionType != TYPE_NO_VEX_XOP) { 888 attrMask |= (insn->vectorExtensionType == TYPE_EVEX) ? ATTR_EVEX : ATTR_VEX; 889 890 if (insn->vectorExtensionType == TYPE_EVEX) { 891 switch (ppFromEVEX3of4(insn->vectorExtensionPrefix[2])) { 892 case VEX_PREFIX_66: 893 attrMask |= ATTR_OPSIZE; 894 break; 895 case VEX_PREFIX_F3: 896 attrMask |= ATTR_XS; 897 break; 898 case VEX_PREFIX_F2: 899 attrMask |= ATTR_XD; 900 break; 901 } 902 903 if (zFromEVEX4of4(insn->vectorExtensionPrefix[3])) 904 attrMask |= ATTR_EVEXKZ; 905 if (bFromEVEX4of4(insn->vectorExtensionPrefix[3])) 906 attrMask |= ATTR_EVEXB; 907 if (aaaFromEVEX4of4(insn->vectorExtensionPrefix[3])) 908 attrMask |= ATTR_EVEXK; 909 if (lFromEVEX4of4(insn->vectorExtensionPrefix[3])) 910 attrMask |= ATTR_EVEXL; 911 if (l2FromEVEX4of4(insn->vectorExtensionPrefix[3])) 912 attrMask |= ATTR_EVEXL2; 913 } else if (insn->vectorExtensionType == TYPE_VEX_3B) { 914 switch (ppFromVEX3of3(insn->vectorExtensionPrefix[2])) { 915 case VEX_PREFIX_66: 916 attrMask |= ATTR_OPSIZE; 917 break; 918 case VEX_PREFIX_F3: 919 attrMask |= ATTR_XS; 920 break; 921 case VEX_PREFIX_F2: 922 attrMask |= ATTR_XD; 923 break; 924 } 925 926 if (lFromVEX3of3(insn->vectorExtensionPrefix[2])) 927 attrMask |= ATTR_VEXL; 928 } else if (insn->vectorExtensionType == TYPE_VEX_2B) { 929 switch (ppFromVEX2of2(insn->vectorExtensionPrefix[1])) { 930 case VEX_PREFIX_66: 931 attrMask |= ATTR_OPSIZE; 932 break; 933 case VEX_PREFIX_F3: 934 attrMask |= ATTR_XS; 935 break; 936 case VEX_PREFIX_F2: 937 attrMask |= ATTR_XD; 938 break; 939 } 940 941 if (lFromVEX2of2(insn->vectorExtensionPrefix[1])) 942 attrMask |= ATTR_VEXL; 943 } else if (insn->vectorExtensionType == TYPE_XOP) { 944 switch (ppFromXOP3of3(insn->vectorExtensionPrefix[2])) { 945 case VEX_PREFIX_66: 946 attrMask |= ATTR_OPSIZE; 947 break; 948 case VEX_PREFIX_F3: 949 attrMask |= ATTR_XS; 950 break; 951 case VEX_PREFIX_F2: 952 attrMask |= ATTR_XD; 953 break; 954 } 955 956 if (lFromXOP3of3(insn->vectorExtensionPrefix[2])) 957 attrMask |= ATTR_VEXL; 958 } else { 959 return -1; 960 } 961 } else { 962 if (insn->mode != MODE_16BIT && isPrefixAtLocation(insn, 0x66, insn->necessaryPrefixLocation)) 963 attrMask |= ATTR_OPSIZE; 964 else if (isPrefixAtLocation(insn, 0x67, insn->necessaryPrefixLocation)) 965 attrMask |= ATTR_ADSIZE; 966 else if (isPrefixAtLocation(insn, 0xf3, insn->necessaryPrefixLocation)) 967 attrMask |= ATTR_XS; 968 else if (isPrefixAtLocation(insn, 0xf2, insn->necessaryPrefixLocation)) 969 attrMask |= ATTR_XD; 970 } 971 972 if (insn->rexPrefix & 0x08) 973 attrMask |= ATTR_REXW; 974 975 /* 976 * JCXZ/JECXZ need special handling for 16-bit mode because the meaning 977 * of the AdSize prefix is inverted w.r.t. 32-bit mode. 978 */ 979 if (insn->mode == MODE_16BIT && insn->opcodeType == ONEBYTE && 980 insn->opcode == 0xE3) 981 attrMask ^= ATTR_ADSIZE; 982 983 /* 984 * In 64-bit mode all f64 superscripted opcodes ignore opcode size prefix 985 * CALL/JMP/JCC instructions need to ignore 0x66 and consume 4 bytes 986 */ 987 988 if (insn->mode == MODE_64BIT && 989 isPrefixAtLocation(insn, 0x66, insn->necessaryPrefixLocation)) { 990 switch (insn->opcode) { 991 case 0xE8: 992 case 0xE9: 993 // Take care of psubsb and other mmx instructions. 994 if (insn->opcodeType == ONEBYTE) { 995 attrMask ^= ATTR_OPSIZE; 996 insn->immediateSize = 4; 997 insn->displacementSize = 4; 998 } 999 break; 1000 case 0x82: 1001 case 0x83: 1002 case 0x84: 1003 case 0x85: 1004 case 0x86: 1005 case 0x87: 1006 case 0x88: 1007 case 0x89: 1008 case 0x8A: 1009 case 0x8B: 1010 case 0x8C: 1011 case 0x8D: 1012 case 0x8E: 1013 case 0x8F: 1014 // Take care of lea and three byte ops. 1015 if (insn->opcodeType == TWOBYTE) { 1016 attrMask ^= ATTR_OPSIZE; 1017 insn->immediateSize = 4; 1018 insn->displacementSize = 4; 1019 } 1020 break; 1021 } 1022 } 1023 1024 if (getIDWithAttrMask(&instructionID, insn, attrMask)) 1025 return -1; 1026 1027 /* The following clauses compensate for limitations of the tables. */ 1028 1029 if (insn->mode != MODE_64BIT && 1030 insn->vectorExtensionType != TYPE_NO_VEX_XOP) { 1031 /* 1032 * The tables can't distinquish between cases where the W-bit is used to 1033 * select register size and cases where its a required part of the opcode. 1034 */ 1035 if ((insn->vectorExtensionType == TYPE_EVEX && 1036 wFromEVEX3of4(insn->vectorExtensionPrefix[2])) || 1037 (insn->vectorExtensionType == TYPE_VEX_3B && 1038 wFromVEX3of3(insn->vectorExtensionPrefix[2])) || 1039 (insn->vectorExtensionType == TYPE_XOP && 1040 wFromXOP3of3(insn->vectorExtensionPrefix[2]))) { 1041 1042 uint16_t instructionIDWithREXW; 1043 if (getIDWithAttrMask(&instructionIDWithREXW, 1044 insn, attrMask | ATTR_REXW)) { 1045 insn->instructionID = instructionID; 1046 insn->spec = specifierForUID(instructionID); 1047 return 0; 1048 } 1049 1050 const char *SpecName = GetInstrName(instructionIDWithREXW, miiArg); 1051 // If not a 64-bit instruction. Switch the opcode. 1052 if (!is64Bit(SpecName)) { 1053 insn->instructionID = instructionIDWithREXW; 1054 insn->spec = specifierForUID(instructionIDWithREXW); 1055 return 0; 1056 } 1057 } 1058 } 1059 1060 /* 1061 * Absolute moves need special handling. 1062 * -For 16-bit mode because the meaning of the AdSize and OpSize prefixes are 1063 * inverted w.r.t. 1064 * -For 32-bit mode we need to ensure the ADSIZE prefix is observed in 1065 * any position. 1066 */ 1067 if (insn->opcodeType == ONEBYTE && ((insn->opcode & 0xFC) == 0xA0)) { 1068 /* Make sure we observed the prefixes in any position. */ 1069 if (insn->prefixPresent[0x67]) 1070 attrMask |= ATTR_ADSIZE; 1071 if (insn->prefixPresent[0x66]) 1072 attrMask |= ATTR_OPSIZE; 1073 1074 /* In 16-bit, invert the attributes. */ 1075 if (insn->mode == MODE_16BIT) 1076 attrMask ^= ATTR_ADSIZE | ATTR_OPSIZE; 1077 1078 if (getIDWithAttrMask(&instructionID, insn, attrMask)) 1079 return -1; 1080 1081 insn->instructionID = instructionID; 1082 insn->spec = specifierForUID(instructionID); 1083 return 0; 1084 } 1085 1086 if ((insn->mode == MODE_16BIT || insn->prefixPresent[0x66]) && 1087 !(attrMask & ATTR_OPSIZE)) { 1088 /* 1089 * The instruction tables make no distinction between instructions that 1090 * allow OpSize anywhere (i.e., 16-bit operations) and that need it in a 1091 * particular spot (i.e., many MMX operations). In general we're 1092 * conservative, but in the specific case where OpSize is present but not 1093 * in the right place we check if there's a 16-bit operation. 1094 */ 1095 1096 const struct InstructionSpecifier *spec; 1097 uint16_t instructionIDWithOpsize; 1098 const char *specName, *specWithOpSizeName; 1099 1100 spec = specifierForUID(instructionID); 1101 1102 if (getIDWithAttrMask(&instructionIDWithOpsize, 1103 insn, 1104 attrMask | ATTR_OPSIZE)) { 1105 /* 1106 * ModRM required with OpSize but not present; give up and return version 1107 * without OpSize set 1108 */ 1109 1110 insn->instructionID = instructionID; 1111 insn->spec = spec; 1112 return 0; 1113 } 1114 1115 specName = GetInstrName(instructionID, miiArg); 1116 specWithOpSizeName = GetInstrName(instructionIDWithOpsize, miiArg); 1117 1118 if (is16BitEquivalent(specName, specWithOpSizeName) && 1119 (insn->mode == MODE_16BIT) ^ insn->prefixPresent[0x66]) { 1120 insn->instructionID = instructionIDWithOpsize; 1121 insn->spec = specifierForUID(instructionIDWithOpsize); 1122 } else { 1123 insn->instructionID = instructionID; 1124 insn->spec = spec; 1125 } 1126 return 0; 1127 } 1128 1129 if (insn->opcodeType == ONEBYTE && insn->opcode == 0x90 && 1130 insn->rexPrefix & 0x01) { 1131 /* 1132 * NOOP shouldn't decode as NOOP if REX.b is set. Instead 1133 * it should decode as XCHG %r8, %eax. 1134 */ 1135 1136 const struct InstructionSpecifier *spec; 1137 uint16_t instructionIDWithNewOpcode; 1138 const struct InstructionSpecifier *specWithNewOpcode; 1139 1140 spec = specifierForUID(instructionID); 1141 1142 /* Borrow opcode from one of the other XCHGar opcodes */ 1143 insn->opcode = 0x91; 1144 1145 if (getIDWithAttrMask(&instructionIDWithNewOpcode, 1146 insn, 1147 attrMask)) { 1148 insn->opcode = 0x90; 1149 1150 insn->instructionID = instructionID; 1151 insn->spec = spec; 1152 return 0; 1153 } 1154 1155 specWithNewOpcode = specifierForUID(instructionIDWithNewOpcode); 1156 1157 /* Change back */ 1158 insn->opcode = 0x90; 1159 1160 insn->instructionID = instructionIDWithNewOpcode; 1161 insn->spec = specWithNewOpcode; 1162 1163 return 0; 1164 } 1165 1166 insn->instructionID = instructionID; 1167 insn->spec = specifierForUID(insn->instructionID); 1168 1169 return 0; 1170 } 1171 1172 /* 1173 * readSIB - Consumes the SIB byte to determine addressing information for an 1174 * instruction. 1175 * 1176 * @param insn - The instruction whose SIB byte is to be read. 1177 * @return - 0 if the SIB byte was successfully read; nonzero otherwise. 1178 */ 1179 static int readSIB(struct InternalInstruction* insn) { 1180 SIBIndex sibIndexBase = SIB_INDEX_NONE; 1181 SIBBase sibBaseBase = SIB_BASE_NONE; 1182 uint8_t index, base; 1183 1184 dbgprintf(insn, "readSIB()"); 1185 1186 if (insn->consumedSIB) 1187 return 0; 1188 1189 insn->consumedSIB = true; 1190 1191 switch (insn->addressSize) { 1192 case 2: 1193 dbgprintf(insn, "SIB-based addressing doesn't work in 16-bit mode"); 1194 return -1; 1195 case 4: 1196 sibIndexBase = SIB_INDEX_EAX; 1197 sibBaseBase = SIB_BASE_EAX; 1198 break; 1199 case 8: 1200 sibIndexBase = SIB_INDEX_RAX; 1201 sibBaseBase = SIB_BASE_RAX; 1202 break; 1203 } 1204 1205 if (consumeByte(insn, &insn->sib)) 1206 return -1; 1207 1208 index = indexFromSIB(insn->sib) | (xFromREX(insn->rexPrefix) << 3); 1209 1210 // FIXME: The fifth bit (bit index 4) is only to be used for instructions 1211 // that understand VSIB indexing. ORing the bit in here is mildy dangerous 1212 // because performing math on an 'enum SIBIndex' can produce garbage. 1213 // Excluding the "none" value, it should cover 6 spaces of register names: 1214 // - 16 possibilities for 16-bit GPR starting at SIB_INDEX_BX_SI 1215 // - 16 possibilities for 32-bit GPR starting at SIB_INDEX_EAX 1216 // - 16 possibilities for 64-bit GPR starting at SIB_INDEX_RAX 1217 // - 32 possibilities for each of XMM, YMM, ZMM registers 1218 // When sibIndexBase gets assigned SIB_INDEX_RAX as it does in 64-bit mode, 1219 // summing in a fully decoded index between 0 and 31 can end up with a value 1220 // that looks like something in the low half of the XMM range. 1221 // translateRMMemory() tries to reverse the damage, with only partial success, 1222 // as evidenced by known bugs in "test/MC/Disassembler/X86/x86-64.txt" 1223 if (insn->vectorExtensionType == TYPE_EVEX) 1224 index |= v2FromEVEX4of4(insn->vectorExtensionPrefix[3]) << 4; 1225 1226 if (index == 0x4) { 1227 insn->sibIndex = SIB_INDEX_NONE; 1228 } else { 1229 insn->sibIndex = (SIBIndex)(sibIndexBase + index); 1230 } 1231 1232 insn->sibScale = 1 << scaleFromSIB(insn->sib); 1233 1234 base = baseFromSIB(insn->sib) | (bFromREX(insn->rexPrefix) << 3); 1235 1236 switch (base) { 1237 case 0x5: 1238 case 0xd: 1239 switch (modFromModRM(insn->modRM)) { 1240 case 0x0: 1241 insn->eaDisplacement = EA_DISP_32; 1242 insn->sibBase = SIB_BASE_NONE; 1243 break; 1244 case 0x1: 1245 insn->eaDisplacement = EA_DISP_8; 1246 insn->sibBase = (SIBBase)(sibBaseBase + base); 1247 break; 1248 case 0x2: 1249 insn->eaDisplacement = EA_DISP_32; 1250 insn->sibBase = (SIBBase)(sibBaseBase + base); 1251 break; 1252 case 0x3: 1253 debug("Cannot have Mod = 0b11 and a SIB byte"); 1254 return -1; 1255 } 1256 break; 1257 default: 1258 insn->sibBase = (SIBBase)(sibBaseBase + base); 1259 break; 1260 } 1261 1262 return 0; 1263 } 1264 1265 /* 1266 * readDisplacement - Consumes the displacement of an instruction. 1267 * 1268 * @param insn - The instruction whose displacement is to be read. 1269 * @return - 0 if the displacement byte was successfully read; nonzero 1270 * otherwise. 1271 */ 1272 static int readDisplacement(struct InternalInstruction* insn) { 1273 int8_t d8; 1274 int16_t d16; 1275 int32_t d32; 1276 1277 dbgprintf(insn, "readDisplacement()"); 1278 1279 if (insn->consumedDisplacement) 1280 return 0; 1281 1282 insn->consumedDisplacement = true; 1283 insn->displacementOffset = insn->readerCursor - insn->startLocation; 1284 1285 switch (insn->eaDisplacement) { 1286 case EA_DISP_NONE: 1287 insn->consumedDisplacement = false; 1288 break; 1289 case EA_DISP_8: 1290 if (consumeInt8(insn, &d8)) 1291 return -1; 1292 insn->displacement = d8; 1293 break; 1294 case EA_DISP_16: 1295 if (consumeInt16(insn, &d16)) 1296 return -1; 1297 insn->displacement = d16; 1298 break; 1299 case EA_DISP_32: 1300 if (consumeInt32(insn, &d32)) 1301 return -1; 1302 insn->displacement = d32; 1303 break; 1304 } 1305 1306 insn->consumedDisplacement = true; 1307 return 0; 1308 } 1309 1310 /* 1311 * readModRM - Consumes all addressing information (ModR/M byte, SIB byte, and 1312 * displacement) for an instruction and interprets it. 1313 * 1314 * @param insn - The instruction whose addressing information is to be read. 1315 * @return - 0 if the information was successfully read; nonzero otherwise. 1316 */ 1317 static int readModRM(struct InternalInstruction* insn) { 1318 uint8_t mod, rm, reg; 1319 1320 dbgprintf(insn, "readModRM()"); 1321 1322 if (insn->consumedModRM) 1323 return 0; 1324 1325 if (consumeByte(insn, &insn->modRM)) 1326 return -1; 1327 insn->consumedModRM = true; 1328 1329 mod = modFromModRM(insn->modRM); 1330 rm = rmFromModRM(insn->modRM); 1331 reg = regFromModRM(insn->modRM); 1332 1333 /* 1334 * This goes by insn->registerSize to pick the correct register, which messes 1335 * up if we're using (say) XMM or 8-bit register operands. That gets fixed in 1336 * fixupReg(). 1337 */ 1338 switch (insn->registerSize) { 1339 case 2: 1340 insn->regBase = MODRM_REG_AX; 1341 insn->eaRegBase = EA_REG_AX; 1342 break; 1343 case 4: 1344 insn->regBase = MODRM_REG_EAX; 1345 insn->eaRegBase = EA_REG_EAX; 1346 break; 1347 case 8: 1348 insn->regBase = MODRM_REG_RAX; 1349 insn->eaRegBase = EA_REG_RAX; 1350 break; 1351 } 1352 1353 reg |= rFromREX(insn->rexPrefix) << 3; 1354 rm |= bFromREX(insn->rexPrefix) << 3; 1355 if (insn->vectorExtensionType == TYPE_EVEX) { 1356 reg |= r2FromEVEX2of4(insn->vectorExtensionPrefix[1]) << 4; 1357 rm |= xFromEVEX2of4(insn->vectorExtensionPrefix[1]) << 4; 1358 } 1359 1360 insn->reg = (Reg)(insn->regBase + reg); 1361 1362 switch (insn->addressSize) { 1363 case 2: 1364 insn->eaBaseBase = EA_BASE_BX_SI; 1365 1366 switch (mod) { 1367 case 0x0: 1368 if (rm == 0x6) { 1369 insn->eaBase = EA_BASE_NONE; 1370 insn->eaDisplacement = EA_DISP_16; 1371 if (readDisplacement(insn)) 1372 return -1; 1373 } else { 1374 insn->eaBase = (EABase)(insn->eaBaseBase + rm); 1375 insn->eaDisplacement = EA_DISP_NONE; 1376 } 1377 break; 1378 case 0x1: 1379 insn->eaBase = (EABase)(insn->eaBaseBase + rm); 1380 insn->eaDisplacement = EA_DISP_8; 1381 insn->displacementSize = 1; 1382 if (readDisplacement(insn)) 1383 return -1; 1384 break; 1385 case 0x2: 1386 insn->eaBase = (EABase)(insn->eaBaseBase + rm); 1387 insn->eaDisplacement = EA_DISP_16; 1388 if (readDisplacement(insn)) 1389 return -1; 1390 break; 1391 case 0x3: 1392 insn->eaBase = (EABase)(insn->eaRegBase + rm); 1393 if (readDisplacement(insn)) 1394 return -1; 1395 break; 1396 } 1397 break; 1398 case 4: 1399 case 8: 1400 insn->eaBaseBase = (insn->addressSize == 4 ? EA_BASE_EAX : EA_BASE_RAX); 1401 1402 switch (mod) { 1403 case 0x0: 1404 insn->eaDisplacement = EA_DISP_NONE; /* readSIB may override this */ 1405 // In determining whether RIP-relative mode is used (rm=5), 1406 // or whether a SIB byte is present (rm=4), 1407 // the extension bits (REX.b and EVEX.x) are ignored. 1408 switch (rm & 7) { 1409 case 0x4: // SIB byte is present 1410 insn->eaBase = (insn->addressSize == 4 ? 1411 EA_BASE_sib : EA_BASE_sib64); 1412 if (readSIB(insn) || readDisplacement(insn)) 1413 return -1; 1414 break; 1415 case 0x5: // RIP-relative 1416 insn->eaBase = EA_BASE_NONE; 1417 insn->eaDisplacement = EA_DISP_32; 1418 if (readDisplacement(insn)) 1419 return -1; 1420 break; 1421 default: 1422 insn->eaBase = (EABase)(insn->eaBaseBase + rm); 1423 break; 1424 } 1425 break; 1426 case 0x1: 1427 insn->displacementSize = 1; 1428 /* FALLTHROUGH */ 1429 case 0x2: 1430 insn->eaDisplacement = (mod == 0x1 ? EA_DISP_8 : EA_DISP_32); 1431 switch (rm & 7) { 1432 case 0x4: // SIB byte is present 1433 insn->eaBase = EA_BASE_sib; 1434 if (readSIB(insn) || readDisplacement(insn)) 1435 return -1; 1436 break; 1437 default: 1438 insn->eaBase = (EABase)(insn->eaBaseBase + rm); 1439 if (readDisplacement(insn)) 1440 return -1; 1441 break; 1442 } 1443 break; 1444 case 0x3: 1445 insn->eaDisplacement = EA_DISP_NONE; 1446 insn->eaBase = (EABase)(insn->eaRegBase + rm); 1447 break; 1448 } 1449 break; 1450 } /* switch (insn->addressSize) */ 1451 1452 return 0; 1453 } 1454 1455 #define GENERIC_FIXUP_FUNC(name, base, prefix) \ 1456 static uint8_t name(struct InternalInstruction *insn, \ 1457 OperandType type, \ 1458 uint8_t index, \ 1459 uint8_t *valid) { \ 1460 *valid = 1; \ 1461 switch (type) { \ 1462 default: \ 1463 debug("Unhandled register type"); \ 1464 *valid = 0; \ 1465 return 0; \ 1466 case TYPE_Rv: \ 1467 return base + index; \ 1468 case TYPE_R8: \ 1469 if (insn->rexPrefix && \ 1470 index >= 4 && index <= 7) { \ 1471 return prefix##_SPL + (index - 4); \ 1472 } else { \ 1473 return prefix##_AL + index; \ 1474 } \ 1475 case TYPE_R16: \ 1476 return prefix##_AX + index; \ 1477 case TYPE_R32: \ 1478 return prefix##_EAX + index; \ 1479 case TYPE_R64: \ 1480 return prefix##_RAX + index; \ 1481 case TYPE_XMM512: \ 1482 return prefix##_ZMM0 + index; \ 1483 case TYPE_XMM256: \ 1484 return prefix##_YMM0 + index; \ 1485 case TYPE_XMM128: \ 1486 case TYPE_XMM64: \ 1487 case TYPE_XMM32: \ 1488 case TYPE_XMM: \ 1489 return prefix##_XMM0 + index; \ 1490 case TYPE_VK1: \ 1491 case TYPE_VK2: \ 1492 case TYPE_VK4: \ 1493 case TYPE_VK8: \ 1494 case TYPE_VK16: \ 1495 case TYPE_VK32: \ 1496 case TYPE_VK64: \ 1497 if (index > 7) \ 1498 *valid = 0; \ 1499 return prefix##_K0 + index; \ 1500 case TYPE_MM64: \ 1501 return prefix##_MM0 + (index & 0x7); \ 1502 case TYPE_SEGMENTREG: \ 1503 if (index > 5) \ 1504 *valid = 0; \ 1505 return prefix##_ES + index; \ 1506 case TYPE_DEBUGREG: \ 1507 return prefix##_DR0 + index; \ 1508 case TYPE_CONTROLREG: \ 1509 return prefix##_CR0 + index; \ 1510 } \ 1511 } 1512 1513 /* 1514 * fixup*Value - Consults an operand type to determine the meaning of the 1515 * reg or R/M field. If the operand is an XMM operand, for example, an 1516 * operand would be XMM0 instead of AX, which readModRM() would otherwise 1517 * misinterpret it as. 1518 * 1519 * @param insn - The instruction containing the operand. 1520 * @param type - The operand type. 1521 * @param index - The existing value of the field as reported by readModRM(). 1522 * @param valid - The address of a uint8_t. The target is set to 1 if the 1523 * field is valid for the register class; 0 if not. 1524 * @return - The proper value. 1525 */ 1526 GENERIC_FIXUP_FUNC(fixupRegValue, insn->regBase, MODRM_REG) 1527 GENERIC_FIXUP_FUNC(fixupRMValue, insn->eaRegBase, EA_REG) 1528 1529 /* 1530 * fixupReg - Consults an operand specifier to determine which of the 1531 * fixup*Value functions to use in correcting readModRM()'ss interpretation. 1532 * 1533 * @param insn - See fixup*Value(). 1534 * @param op - The operand specifier. 1535 * @return - 0 if fixup was successful; -1 if the register returned was 1536 * invalid for its class. 1537 */ 1538 static int fixupReg(struct InternalInstruction *insn, 1539 const struct OperandSpecifier *op) { 1540 uint8_t valid; 1541 1542 dbgprintf(insn, "fixupReg()"); 1543 1544 switch ((OperandEncoding)op->encoding) { 1545 default: 1546 debug("Expected a REG or R/M encoding in fixupReg"); 1547 return -1; 1548 case ENCODING_VVVV: 1549 insn->vvvv = (Reg)fixupRegValue(insn, 1550 (OperandType)op->type, 1551 insn->vvvv, 1552 &valid); 1553 if (!valid) 1554 return -1; 1555 break; 1556 case ENCODING_REG: 1557 insn->reg = (Reg)fixupRegValue(insn, 1558 (OperandType)op->type, 1559 insn->reg - insn->regBase, 1560 &valid); 1561 if (!valid) 1562 return -1; 1563 break; 1564 CASE_ENCODING_RM: 1565 if (insn->eaBase >= insn->eaRegBase) { 1566 insn->eaBase = (EABase)fixupRMValue(insn, 1567 (OperandType)op->type, 1568 insn->eaBase - insn->eaRegBase, 1569 &valid); 1570 if (!valid) 1571 return -1; 1572 } 1573 break; 1574 } 1575 1576 return 0; 1577 } 1578 1579 /* 1580 * readOpcodeRegister - Reads an operand from the opcode field of an 1581 * instruction and interprets it appropriately given the operand width. 1582 * Handles AddRegFrm instructions. 1583 * 1584 * @param insn - the instruction whose opcode field is to be read. 1585 * @param size - The width (in bytes) of the register being specified. 1586 * 1 means AL and friends, 2 means AX, 4 means EAX, and 8 means 1587 * RAX. 1588 * @return - 0 on success; nonzero otherwise. 1589 */ 1590 static int readOpcodeRegister(struct InternalInstruction* insn, uint8_t size) { 1591 dbgprintf(insn, "readOpcodeRegister()"); 1592 1593 if (size == 0) 1594 size = insn->registerSize; 1595 1596 switch (size) { 1597 case 1: 1598 insn->opcodeRegister = (Reg)(MODRM_REG_AL + ((bFromREX(insn->rexPrefix) << 3) 1599 | (insn->opcode & 7))); 1600 if (insn->rexPrefix && 1601 insn->opcodeRegister >= MODRM_REG_AL + 0x4 && 1602 insn->opcodeRegister < MODRM_REG_AL + 0x8) { 1603 insn->opcodeRegister = (Reg)(MODRM_REG_SPL 1604 + (insn->opcodeRegister - MODRM_REG_AL - 4)); 1605 } 1606 1607 break; 1608 case 2: 1609 insn->opcodeRegister = (Reg)(MODRM_REG_AX 1610 + ((bFromREX(insn->rexPrefix) << 3) 1611 | (insn->opcode & 7))); 1612 break; 1613 case 4: 1614 insn->opcodeRegister = (Reg)(MODRM_REG_EAX 1615 + ((bFromREX(insn->rexPrefix) << 3) 1616 | (insn->opcode & 7))); 1617 break; 1618 case 8: 1619 insn->opcodeRegister = (Reg)(MODRM_REG_RAX 1620 + ((bFromREX(insn->rexPrefix) << 3) 1621 | (insn->opcode & 7))); 1622 break; 1623 } 1624 1625 return 0; 1626 } 1627 1628 /* 1629 * readImmediate - Consumes an immediate operand from an instruction, given the 1630 * desired operand size. 1631 * 1632 * @param insn - The instruction whose operand is to be read. 1633 * @param size - The width (in bytes) of the operand. 1634 * @return - 0 if the immediate was successfully consumed; nonzero 1635 * otherwise. 1636 */ 1637 static int readImmediate(struct InternalInstruction* insn, uint8_t size) { 1638 uint8_t imm8; 1639 uint16_t imm16; 1640 uint32_t imm32; 1641 uint64_t imm64; 1642 1643 dbgprintf(insn, "readImmediate()"); 1644 1645 if (insn->numImmediatesConsumed == 2) { 1646 debug("Already consumed two immediates"); 1647 return -1; 1648 } 1649 1650 if (size == 0) 1651 size = insn->immediateSize; 1652 else 1653 insn->immediateSize = size; 1654 insn->immediateOffset = insn->readerCursor - insn->startLocation; 1655 1656 switch (size) { 1657 case 1: 1658 if (consumeByte(insn, &imm8)) 1659 return -1; 1660 insn->immediates[insn->numImmediatesConsumed] = imm8; 1661 break; 1662 case 2: 1663 if (consumeUInt16(insn, &imm16)) 1664 return -1; 1665 insn->immediates[insn->numImmediatesConsumed] = imm16; 1666 break; 1667 case 4: 1668 if (consumeUInt32(insn, &imm32)) 1669 return -1; 1670 insn->immediates[insn->numImmediatesConsumed] = imm32; 1671 break; 1672 case 8: 1673 if (consumeUInt64(insn, &imm64)) 1674 return -1; 1675 insn->immediates[insn->numImmediatesConsumed] = imm64; 1676 break; 1677 } 1678 1679 insn->numImmediatesConsumed++; 1680 1681 return 0; 1682 } 1683 1684 /* 1685 * readVVVV - Consumes vvvv from an instruction if it has a VEX prefix. 1686 * 1687 * @param insn - The instruction whose operand is to be read. 1688 * @return - 0 if the vvvv was successfully consumed; nonzero 1689 * otherwise. 1690 */ 1691 static int readVVVV(struct InternalInstruction* insn) { 1692 dbgprintf(insn, "readVVVV()"); 1693 1694 int vvvv; 1695 if (insn->vectorExtensionType == TYPE_EVEX) 1696 vvvv = (v2FromEVEX4of4(insn->vectorExtensionPrefix[3]) << 4 | 1697 vvvvFromEVEX3of4(insn->vectorExtensionPrefix[2])); 1698 else if (insn->vectorExtensionType == TYPE_VEX_3B) 1699 vvvv = vvvvFromVEX3of3(insn->vectorExtensionPrefix[2]); 1700 else if (insn->vectorExtensionType == TYPE_VEX_2B) 1701 vvvv = vvvvFromVEX2of2(insn->vectorExtensionPrefix[1]); 1702 else if (insn->vectorExtensionType == TYPE_XOP) 1703 vvvv = vvvvFromXOP3of3(insn->vectorExtensionPrefix[2]); 1704 else 1705 return -1; 1706 1707 if (insn->mode != MODE_64BIT) 1708 vvvv &= 0x7; 1709 1710 insn->vvvv = static_cast<Reg>(vvvv); 1711 return 0; 1712 } 1713 1714 /* 1715 * readMaskRegister - Reads an mask register from the opcode field of an 1716 * instruction. 1717 * 1718 * @param insn - The instruction whose opcode field is to be read. 1719 * @return - 0 on success; nonzero otherwise. 1720 */ 1721 static int readMaskRegister(struct InternalInstruction* insn) { 1722 dbgprintf(insn, "readMaskRegister()"); 1723 1724 if (insn->vectorExtensionType != TYPE_EVEX) 1725 return -1; 1726 1727 insn->writemask = 1728 static_cast<Reg>(aaaFromEVEX4of4(insn->vectorExtensionPrefix[3])); 1729 return 0; 1730 } 1731 1732 /* 1733 * readOperands - Consults the specifier for an instruction and consumes all 1734 * operands for that instruction, interpreting them as it goes. 1735 * 1736 * @param insn - The instruction whose operands are to be read and interpreted. 1737 * @return - 0 if all operands could be read; nonzero otherwise. 1738 */ 1739 static int readOperands(struct InternalInstruction* insn) { 1740 int hasVVVV, needVVVV; 1741 int sawRegImm = 0; 1742 1743 dbgprintf(insn, "readOperands()"); 1744 1745 /* If non-zero vvvv specified, need to make sure one of the operands 1746 uses it. */ 1747 hasVVVV = !readVVVV(insn); 1748 needVVVV = hasVVVV && (insn->vvvv != 0); 1749 1750 for (const auto &Op : x86OperandSets[insn->spec->operands]) { 1751 switch (Op.encoding) { 1752 case ENCODING_NONE: 1753 case ENCODING_SI: 1754 case ENCODING_DI: 1755 break; 1756 case ENCODING_REG: 1757 CASE_ENCODING_RM: 1758 if (readModRM(insn)) 1759 return -1; 1760 if (fixupReg(insn, &Op)) 1761 return -1; 1762 // Apply the AVX512 compressed displacement scaling factor. 1763 if (Op.encoding != ENCODING_REG && insn->eaDisplacement == EA_DISP_8) 1764 insn->displacement *= 1 << (Op.encoding - ENCODING_RM); 1765 break; 1766 case ENCODING_CB: 1767 case ENCODING_CW: 1768 case ENCODING_CD: 1769 case ENCODING_CP: 1770 case ENCODING_CO: 1771 case ENCODING_CT: 1772 dbgprintf(insn, "We currently don't hande code-offset encodings"); 1773 return -1; 1774 case ENCODING_IB: 1775 if (sawRegImm) { 1776 /* Saw a register immediate so don't read again and instead split the 1777 previous immediate. FIXME: This is a hack. */ 1778 insn->immediates[insn->numImmediatesConsumed] = 1779 insn->immediates[insn->numImmediatesConsumed - 1] & 0xf; 1780 ++insn->numImmediatesConsumed; 1781 break; 1782 } 1783 if (readImmediate(insn, 1)) 1784 return -1; 1785 if (Op.type == TYPE_XMM128 || 1786 Op.type == TYPE_XMM256) 1787 sawRegImm = 1; 1788 break; 1789 case ENCODING_IW: 1790 if (readImmediate(insn, 2)) 1791 return -1; 1792 break; 1793 case ENCODING_ID: 1794 if (readImmediate(insn, 4)) 1795 return -1; 1796 break; 1797 case ENCODING_IO: 1798 if (readImmediate(insn, 8)) 1799 return -1; 1800 break; 1801 case ENCODING_Iv: 1802 if (readImmediate(insn, insn->immediateSize)) 1803 return -1; 1804 break; 1805 case ENCODING_Ia: 1806 if (readImmediate(insn, insn->addressSize)) 1807 return -1; 1808 break; 1809 case ENCODING_RB: 1810 if (readOpcodeRegister(insn, 1)) 1811 return -1; 1812 break; 1813 case ENCODING_RW: 1814 if (readOpcodeRegister(insn, 2)) 1815 return -1; 1816 break; 1817 case ENCODING_RD: 1818 if (readOpcodeRegister(insn, 4)) 1819 return -1; 1820 break; 1821 case ENCODING_RO: 1822 if (readOpcodeRegister(insn, 8)) 1823 return -1; 1824 break; 1825 case ENCODING_Rv: 1826 if (readOpcodeRegister(insn, 0)) 1827 return -1; 1828 break; 1829 case ENCODING_FP: 1830 break; 1831 case ENCODING_VVVV: 1832 needVVVV = 0; /* Mark that we have found a VVVV operand. */ 1833 if (!hasVVVV) 1834 return -1; 1835 if (fixupReg(insn, &Op)) 1836 return -1; 1837 break; 1838 case ENCODING_WRITEMASK: 1839 if (readMaskRegister(insn)) 1840 return -1; 1841 break; 1842 case ENCODING_DUP: 1843 break; 1844 default: 1845 dbgprintf(insn, "Encountered an operand with an unknown encoding."); 1846 return -1; 1847 } 1848 } 1849 1850 /* If we didn't find ENCODING_VVVV operand, but non-zero vvvv present, fail */ 1851 if (needVVVV) return -1; 1852 1853 return 0; 1854 } 1855 1856 /* 1857 * decodeInstruction - Reads and interprets a full instruction provided by the 1858 * user. 1859 * 1860 * @param insn - A pointer to the instruction to be populated. Must be 1861 * pre-allocated. 1862 * @param reader - The function to be used to read the instruction's bytes. 1863 * @param readerArg - A generic argument to be passed to the reader to store 1864 * any internal state. 1865 * @param logger - If non-NULL, the function to be used to write log messages 1866 * and warnings. 1867 * @param loggerArg - A generic argument to be passed to the logger to store 1868 * any internal state. 1869 * @param startLoc - The address (in the reader's address space) of the first 1870 * byte in the instruction. 1871 * @param mode - The mode (real mode, IA-32e, or IA-32e in 64-bit mode) to 1872 * decode the instruction in. 1873 * @return - 0 if the instruction's memory could be read; nonzero if 1874 * not. 1875 */ 1876 int llvm::X86Disassembler::decodeInstruction( 1877 struct InternalInstruction *insn, byteReader_t reader, 1878 const void *readerArg, dlog_t logger, void *loggerArg, const void *miiArg, 1879 uint64_t startLoc, DisassemblerMode mode) { 1880 memset(insn, 0, sizeof(struct InternalInstruction)); 1881 1882 insn->reader = reader; 1883 insn->readerArg = readerArg; 1884 insn->dlog = logger; 1885 insn->dlogArg = loggerArg; 1886 insn->startLocation = startLoc; 1887 insn->readerCursor = startLoc; 1888 insn->mode = mode; 1889 insn->numImmediatesConsumed = 0; 1890 1891 if (readPrefixes(insn) || 1892 readOpcode(insn) || 1893 getID(insn, miiArg) || 1894 insn->instructionID == 0 || 1895 readOperands(insn)) 1896 return -1; 1897 1898 insn->operands = x86OperandSets[insn->spec->operands]; 1899 1900 insn->length = insn->readerCursor - insn->startLocation; 1901 1902 dbgprintf(insn, "Read from 0x%llx to 0x%llx: length %zu", 1903 startLoc, insn->readerCursor, insn->length); 1904 1905 if (insn->length > 15) 1906 dbgprintf(insn, "Instruction exceeds 15-byte limit"); 1907 1908 return 0; 1909 } 1910