1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * x86 instruction analysis 4 * 5 * Copyright (C) IBM Corporation, 2002, 2004, 2009 6 */ 7 8 #include <linux/kernel.h> 9 #ifdef __KERNEL__ 10 #include <linux/string.h> 11 #else 12 #include <string.h> 13 #endif 14 #include <asm/inat.h> /*__ignore_sync_check__ */ 15 #include <asm/insn.h> /* __ignore_sync_check__ */ 16 17 #include <linux/errno.h> 18 #include <linux/kconfig.h> 19 20 #include <asm/emulate_prefix.h> /* __ignore_sync_check__ */ 21 22 #define leXX_to_cpu(t, r) \ 23 ({ \ 24 __typeof__(t) v; \ 25 switch (sizeof(t)) { \ 26 case 4: v = le32_to_cpu(r); break; \ 27 case 2: v = le16_to_cpu(r); break; \ 28 case 1: v = r; break; \ 29 default: \ 30 BUILD_BUG(); break; \ 31 } \ 32 v; \ 33 }) 34 35 /* Verify next sizeof(t) bytes can be on the same instruction */ 36 #define validate_next(t, insn, n) \ 37 ((insn)->next_byte + sizeof(t) + n <= (insn)->end_kaddr) 38 39 #define __get_next(t, insn) \ 40 ({ t r = *(t*)insn->next_byte; insn->next_byte += sizeof(t); leXX_to_cpu(t, r); }) 41 42 #define __peek_nbyte_next(t, insn, n) \ 43 ({ t r = *(t*)((insn)->next_byte + n); leXX_to_cpu(t, r); }) 44 45 #define get_next(t, insn) \ 46 ({ if (unlikely(!validate_next(t, insn, 0))) goto err_out; __get_next(t, insn); }) 47 48 #define peek_nbyte_next(t, insn, n) \ 49 ({ if (unlikely(!validate_next(t, insn, n))) goto err_out; __peek_nbyte_next(t, insn, n); }) 50 51 #define peek_next(t, insn) peek_nbyte_next(t, insn, 0) 52 53 /** 54 * insn_init() - initialize struct insn 55 * @insn: &struct insn to be initialized 56 * @kaddr: address (in kernel memory) of instruction (or copy thereof) 57 * @buf_len: length of the insn buffer at @kaddr 58 * @x86_64: !0 for 64-bit kernel or 64-bit app 59 */ 60 void insn_init(struct insn *insn, const void *kaddr, int buf_len, int x86_64) 61 { 62 /* 63 * Instructions longer than MAX_INSN_SIZE (15 bytes) are invalid 64 * even if the input buffer is long enough to hold them. 65 */ 66 if (buf_len > MAX_INSN_SIZE) 67 buf_len = MAX_INSN_SIZE; 68 69 memset(insn, 0, sizeof(*insn)); 70 insn->kaddr = kaddr; 71 insn->end_kaddr = kaddr + buf_len; 72 insn->next_byte = kaddr; 73 insn->x86_64 = x86_64 ? 1 : 0; 74 insn->opnd_bytes = 4; 75 if (x86_64) 76 insn->addr_bytes = 8; 77 else 78 insn->addr_bytes = 4; 79 } 80 81 static const insn_byte_t xen_prefix[] = { __XEN_EMULATE_PREFIX }; 82 static const insn_byte_t kvm_prefix[] = { __KVM_EMULATE_PREFIX }; 83 84 static int __insn_get_emulate_prefix(struct insn *insn, 85 const insn_byte_t *prefix, size_t len) 86 { 87 size_t i; 88 89 for (i = 0; i < len; i++) { 90 if (peek_nbyte_next(insn_byte_t, insn, i) != prefix[i]) 91 goto err_out; 92 } 93 94 insn->emulate_prefix_size = len; 95 insn->next_byte += len; 96 97 return 1; 98 99 err_out: 100 return 0; 101 } 102 103 static void insn_get_emulate_prefix(struct insn *insn) 104 { 105 if (__insn_get_emulate_prefix(insn, xen_prefix, sizeof(xen_prefix))) 106 return; 107 108 __insn_get_emulate_prefix(insn, kvm_prefix, sizeof(kvm_prefix)); 109 } 110 111 /** 112 * insn_get_prefixes - scan x86 instruction prefix bytes 113 * @insn: &struct insn containing instruction 114 * 115 * Populates the @insn->prefixes bitmap, and updates @insn->next_byte 116 * to point to the (first) opcode. No effect if @insn->prefixes.got 117 * is already set. 118 * 119 * * Returns: 120 * 0: on success 121 * < 0: on error 122 */ 123 int insn_get_prefixes(struct insn *insn) 124 { 125 struct insn_field *prefixes = &insn->prefixes; 126 insn_attr_t attr; 127 insn_byte_t b, lb; 128 int i, nb; 129 130 if (prefixes->got) 131 return 0; 132 133 insn_get_emulate_prefix(insn); 134 135 nb = 0; 136 lb = 0; 137 b = peek_next(insn_byte_t, insn); 138 attr = inat_get_opcode_attribute(b); 139 while (inat_is_legacy_prefix(attr)) { 140 /* Skip if same prefix */ 141 for (i = 0; i < nb; i++) 142 if (prefixes->bytes[i] == b) 143 goto found; 144 if (nb == 4) 145 /* Invalid instruction */ 146 break; 147 prefixes->bytes[nb++] = b; 148 if (inat_is_address_size_prefix(attr)) { 149 /* address size switches 2/4 or 4/8 */ 150 if (insn->x86_64) 151 insn->addr_bytes ^= 12; 152 else 153 insn->addr_bytes ^= 6; 154 } else if (inat_is_operand_size_prefix(attr)) { 155 /* oprand size switches 2/4 */ 156 insn->opnd_bytes ^= 6; 157 } 158 found: 159 prefixes->nbytes++; 160 insn->next_byte++; 161 lb = b; 162 b = peek_next(insn_byte_t, insn); 163 attr = inat_get_opcode_attribute(b); 164 } 165 /* Set the last prefix */ 166 if (lb && lb != insn->prefixes.bytes[3]) { 167 if (unlikely(insn->prefixes.bytes[3])) { 168 /* Swap the last prefix */ 169 b = insn->prefixes.bytes[3]; 170 for (i = 0; i < nb; i++) 171 if (prefixes->bytes[i] == lb) 172 insn_set_byte(prefixes, i, b); 173 } 174 insn_set_byte(&insn->prefixes, 3, lb); 175 } 176 177 /* Decode REX prefix */ 178 if (insn->x86_64) { 179 b = peek_next(insn_byte_t, insn); 180 attr = inat_get_opcode_attribute(b); 181 if (inat_is_rex_prefix(attr)) { 182 insn_field_set(&insn->rex_prefix, b, 1); 183 insn->next_byte++; 184 if (X86_REX_W(b)) 185 /* REX.W overrides opnd_size */ 186 insn->opnd_bytes = 8; 187 } 188 } 189 insn->rex_prefix.got = 1; 190 191 /* Decode VEX prefix */ 192 b = peek_next(insn_byte_t, insn); 193 attr = inat_get_opcode_attribute(b); 194 if (inat_is_vex_prefix(attr)) { 195 insn_byte_t b2 = peek_nbyte_next(insn_byte_t, insn, 1); 196 if (!insn->x86_64) { 197 /* 198 * In 32-bits mode, if the [7:6] bits (mod bits of 199 * ModRM) on the second byte are not 11b, it is 200 * LDS or LES or BOUND. 201 */ 202 if (X86_MODRM_MOD(b2) != 3) 203 goto vex_end; 204 } 205 insn_set_byte(&insn->vex_prefix, 0, b); 206 insn_set_byte(&insn->vex_prefix, 1, b2); 207 if (inat_is_evex_prefix(attr)) { 208 b2 = peek_nbyte_next(insn_byte_t, insn, 2); 209 insn_set_byte(&insn->vex_prefix, 2, b2); 210 b2 = peek_nbyte_next(insn_byte_t, insn, 3); 211 insn_set_byte(&insn->vex_prefix, 3, b2); 212 insn->vex_prefix.nbytes = 4; 213 insn->next_byte += 4; 214 if (insn->x86_64 && X86_VEX_W(b2)) 215 /* VEX.W overrides opnd_size */ 216 insn->opnd_bytes = 8; 217 } else if (inat_is_vex3_prefix(attr)) { 218 b2 = peek_nbyte_next(insn_byte_t, insn, 2); 219 insn_set_byte(&insn->vex_prefix, 2, b2); 220 insn->vex_prefix.nbytes = 3; 221 insn->next_byte += 3; 222 if (insn->x86_64 && X86_VEX_W(b2)) 223 /* VEX.W overrides opnd_size */ 224 insn->opnd_bytes = 8; 225 } else { 226 /* 227 * For VEX2, fake VEX3-like byte#2. 228 * Makes it easier to decode vex.W, vex.vvvv, 229 * vex.L and vex.pp. Masking with 0x7f sets vex.W == 0. 230 */ 231 insn_set_byte(&insn->vex_prefix, 2, b2 & 0x7f); 232 insn->vex_prefix.nbytes = 2; 233 insn->next_byte += 2; 234 } 235 } 236 vex_end: 237 insn->vex_prefix.got = 1; 238 239 prefixes->got = 1; 240 241 return 0; 242 243 err_out: 244 return -ENODATA; 245 } 246 247 /** 248 * insn_get_opcode - collect opcode(s) 249 * @insn: &struct insn containing instruction 250 * 251 * Populates @insn->opcode, updates @insn->next_byte to point past the 252 * opcode byte(s), and set @insn->attr (except for groups). 253 * If necessary, first collects any preceding (prefix) bytes. 254 * Sets @insn->opcode.value = opcode1. No effect if @insn->opcode.got 255 * is already 1. 256 * 257 * Returns: 258 * 0: on success 259 * < 0: on error 260 */ 261 int insn_get_opcode(struct insn *insn) 262 { 263 struct insn_field *opcode = &insn->opcode; 264 int pfx_id, ret; 265 insn_byte_t op; 266 267 if (opcode->got) 268 return 0; 269 270 if (!insn->prefixes.got) { 271 ret = insn_get_prefixes(insn); 272 if (ret) 273 return ret; 274 } 275 276 /* Get first opcode */ 277 op = get_next(insn_byte_t, insn); 278 insn_set_byte(opcode, 0, op); 279 opcode->nbytes = 1; 280 281 /* Check if there is VEX prefix or not */ 282 if (insn_is_avx(insn)) { 283 insn_byte_t m, p; 284 m = insn_vex_m_bits(insn); 285 p = insn_vex_p_bits(insn); 286 insn->attr = inat_get_avx_attribute(op, m, p); 287 if ((inat_must_evex(insn->attr) && !insn_is_evex(insn)) || 288 (!inat_accept_vex(insn->attr) && 289 !inat_is_group(insn->attr))) { 290 /* This instruction is bad */ 291 insn->attr = 0; 292 return -EINVAL; 293 } 294 /* VEX has only 1 byte for opcode */ 295 goto end; 296 } 297 298 insn->attr = inat_get_opcode_attribute(op); 299 while (inat_is_escape(insn->attr)) { 300 /* Get escaped opcode */ 301 op = get_next(insn_byte_t, insn); 302 opcode->bytes[opcode->nbytes++] = op; 303 pfx_id = insn_last_prefix_id(insn); 304 insn->attr = inat_get_escape_attribute(op, pfx_id, insn->attr); 305 } 306 307 if (inat_must_vex(insn->attr)) { 308 /* This instruction is bad */ 309 insn->attr = 0; 310 return -EINVAL; 311 } 312 end: 313 opcode->got = 1; 314 return 0; 315 316 err_out: 317 return -ENODATA; 318 } 319 320 /** 321 * insn_get_modrm - collect ModRM byte, if any 322 * @insn: &struct insn containing instruction 323 * 324 * Populates @insn->modrm and updates @insn->next_byte to point past the 325 * ModRM byte, if any. If necessary, first collects the preceding bytes 326 * (prefixes and opcode(s)). No effect if @insn->modrm.got is already 1. 327 * 328 * Returns: 329 * 0: on success 330 * < 0: on error 331 */ 332 int insn_get_modrm(struct insn *insn) 333 { 334 struct insn_field *modrm = &insn->modrm; 335 insn_byte_t pfx_id, mod; 336 int ret; 337 338 if (modrm->got) 339 return 0; 340 341 if (!insn->opcode.got) { 342 ret = insn_get_opcode(insn); 343 if (ret) 344 return ret; 345 } 346 347 if (inat_has_modrm(insn->attr)) { 348 mod = get_next(insn_byte_t, insn); 349 insn_field_set(modrm, mod, 1); 350 if (inat_is_group(insn->attr)) { 351 pfx_id = insn_last_prefix_id(insn); 352 insn->attr = inat_get_group_attribute(mod, pfx_id, 353 insn->attr); 354 if (insn_is_avx(insn) && !inat_accept_vex(insn->attr)) { 355 /* Bad insn */ 356 insn->attr = 0; 357 return -EINVAL; 358 } 359 } 360 } 361 362 if (insn->x86_64 && inat_is_force64(insn->attr)) 363 insn->opnd_bytes = 8; 364 365 modrm->got = 1; 366 return 0; 367 368 err_out: 369 return -ENODATA; 370 } 371 372 373 /** 374 * insn_rip_relative() - Does instruction use RIP-relative addressing mode? 375 * @insn: &struct insn containing instruction 376 * 377 * If necessary, first collects the instruction up to and including the 378 * ModRM byte. No effect if @insn->x86_64 is 0. 379 */ 380 int insn_rip_relative(struct insn *insn) 381 { 382 struct insn_field *modrm = &insn->modrm; 383 int ret; 384 385 if (!insn->x86_64) 386 return 0; 387 388 if (!modrm->got) { 389 ret = insn_get_modrm(insn); 390 if (ret) 391 return 0; 392 } 393 /* 394 * For rip-relative instructions, the mod field (top 2 bits) 395 * is zero and the r/m field (bottom 3 bits) is 0x5. 396 */ 397 return (modrm->nbytes && (modrm->bytes[0] & 0xc7) == 0x5); 398 } 399 400 /** 401 * insn_get_sib() - Get the SIB byte of instruction 402 * @insn: &struct insn containing instruction 403 * 404 * If necessary, first collects the instruction up to and including the 405 * ModRM byte. 406 * 407 * Returns: 408 * 0: if decoding succeeded 409 * < 0: otherwise. 410 */ 411 int insn_get_sib(struct insn *insn) 412 { 413 insn_byte_t modrm; 414 int ret; 415 416 if (insn->sib.got) 417 return 0; 418 419 if (!insn->modrm.got) { 420 ret = insn_get_modrm(insn); 421 if (ret) 422 return ret; 423 } 424 425 if (insn->modrm.nbytes) { 426 modrm = insn->modrm.bytes[0]; 427 if (insn->addr_bytes != 2 && 428 X86_MODRM_MOD(modrm) != 3 && X86_MODRM_RM(modrm) == 4) { 429 insn_field_set(&insn->sib, 430 get_next(insn_byte_t, insn), 1); 431 } 432 } 433 insn->sib.got = 1; 434 435 return 0; 436 437 err_out: 438 return -ENODATA; 439 } 440 441 442 /** 443 * insn_get_displacement() - Get the displacement of instruction 444 * @insn: &struct insn containing instruction 445 * 446 * If necessary, first collects the instruction up to and including the 447 * SIB byte. 448 * Displacement value is sign-expanded. 449 * 450 * * Returns: 451 * 0: if decoding succeeded 452 * < 0: otherwise. 453 */ 454 int insn_get_displacement(struct insn *insn) 455 { 456 insn_byte_t mod, rm, base; 457 int ret; 458 459 if (insn->displacement.got) 460 return 0; 461 462 if (!insn->sib.got) { 463 ret = insn_get_sib(insn); 464 if (ret) 465 return ret; 466 } 467 468 if (insn->modrm.nbytes) { 469 /* 470 * Interpreting the modrm byte: 471 * mod = 00 - no displacement fields (exceptions below) 472 * mod = 01 - 1-byte displacement field 473 * mod = 10 - displacement field is 4 bytes, or 2 bytes if 474 * address size = 2 (0x67 prefix in 32-bit mode) 475 * mod = 11 - no memory operand 476 * 477 * If address size = 2... 478 * mod = 00, r/m = 110 - displacement field is 2 bytes 479 * 480 * If address size != 2... 481 * mod != 11, r/m = 100 - SIB byte exists 482 * mod = 00, SIB base = 101 - displacement field is 4 bytes 483 * mod = 00, r/m = 101 - rip-relative addressing, displacement 484 * field is 4 bytes 485 */ 486 mod = X86_MODRM_MOD(insn->modrm.value); 487 rm = X86_MODRM_RM(insn->modrm.value); 488 base = X86_SIB_BASE(insn->sib.value); 489 if (mod == 3) 490 goto out; 491 if (mod == 1) { 492 insn_field_set(&insn->displacement, 493 get_next(signed char, insn), 1); 494 } else if (insn->addr_bytes == 2) { 495 if ((mod == 0 && rm == 6) || mod == 2) { 496 insn_field_set(&insn->displacement, 497 get_next(short, insn), 2); 498 } 499 } else { 500 if ((mod == 0 && rm == 5) || mod == 2 || 501 (mod == 0 && base == 5)) { 502 insn_field_set(&insn->displacement, 503 get_next(int, insn), 4); 504 } 505 } 506 } 507 out: 508 insn->displacement.got = 1; 509 return 0; 510 511 err_out: 512 return -ENODATA; 513 } 514 515 /* Decode moffset16/32/64. Return 0 if failed */ 516 static int __get_moffset(struct insn *insn) 517 { 518 switch (insn->addr_bytes) { 519 case 2: 520 insn_field_set(&insn->moffset1, get_next(short, insn), 2); 521 break; 522 case 4: 523 insn_field_set(&insn->moffset1, get_next(int, insn), 4); 524 break; 525 case 8: 526 insn_field_set(&insn->moffset1, get_next(int, insn), 4); 527 insn_field_set(&insn->moffset2, get_next(int, insn), 4); 528 break; 529 default: /* opnd_bytes must be modified manually */ 530 goto err_out; 531 } 532 insn->moffset1.got = insn->moffset2.got = 1; 533 534 return 1; 535 536 err_out: 537 return 0; 538 } 539 540 /* Decode imm v32(Iz). Return 0 if failed */ 541 static int __get_immv32(struct insn *insn) 542 { 543 switch (insn->opnd_bytes) { 544 case 2: 545 insn_field_set(&insn->immediate, get_next(short, insn), 2); 546 break; 547 case 4: 548 case 8: 549 insn_field_set(&insn->immediate, get_next(int, insn), 4); 550 break; 551 default: /* opnd_bytes must be modified manually */ 552 goto err_out; 553 } 554 555 return 1; 556 557 err_out: 558 return 0; 559 } 560 561 /* Decode imm v64(Iv/Ov), Return 0 if failed */ 562 static int __get_immv(struct insn *insn) 563 { 564 switch (insn->opnd_bytes) { 565 case 2: 566 insn_field_set(&insn->immediate1, get_next(short, insn), 2); 567 break; 568 case 4: 569 insn_field_set(&insn->immediate1, get_next(int, insn), 4); 570 insn->immediate1.nbytes = 4; 571 break; 572 case 8: 573 insn_field_set(&insn->immediate1, get_next(int, insn), 4); 574 insn_field_set(&insn->immediate2, get_next(int, insn), 4); 575 break; 576 default: /* opnd_bytes must be modified manually */ 577 goto err_out; 578 } 579 insn->immediate1.got = insn->immediate2.got = 1; 580 581 return 1; 582 err_out: 583 return 0; 584 } 585 586 /* Decode ptr16:16/32(Ap) */ 587 static int __get_immptr(struct insn *insn) 588 { 589 switch (insn->opnd_bytes) { 590 case 2: 591 insn_field_set(&insn->immediate1, get_next(short, insn), 2); 592 break; 593 case 4: 594 insn_field_set(&insn->immediate1, get_next(int, insn), 4); 595 break; 596 case 8: 597 /* ptr16:64 is not exist (no segment) */ 598 return 0; 599 default: /* opnd_bytes must be modified manually */ 600 goto err_out; 601 } 602 insn_field_set(&insn->immediate2, get_next(unsigned short, insn), 2); 603 insn->immediate1.got = insn->immediate2.got = 1; 604 605 return 1; 606 err_out: 607 return 0; 608 } 609 610 /** 611 * insn_get_immediate() - Get the immediate in an instruction 612 * @insn: &struct insn containing instruction 613 * 614 * If necessary, first collects the instruction up to and including the 615 * displacement bytes. 616 * Basically, most of immediates are sign-expanded. Unsigned-value can be 617 * computed by bit masking with ((1 << (nbytes * 8)) - 1) 618 * 619 * Returns: 620 * 0: on success 621 * < 0: on error 622 */ 623 int insn_get_immediate(struct insn *insn) 624 { 625 int ret; 626 627 if (insn->immediate.got) 628 return 0; 629 630 if (!insn->displacement.got) { 631 ret = insn_get_displacement(insn); 632 if (ret) 633 return ret; 634 } 635 636 if (inat_has_moffset(insn->attr)) { 637 if (!__get_moffset(insn)) 638 goto err_out; 639 goto done; 640 } 641 642 if (!inat_has_immediate(insn->attr)) 643 /* no immediates */ 644 goto done; 645 646 switch (inat_immediate_size(insn->attr)) { 647 case INAT_IMM_BYTE: 648 insn_field_set(&insn->immediate, get_next(signed char, insn), 1); 649 break; 650 case INAT_IMM_WORD: 651 insn_field_set(&insn->immediate, get_next(short, insn), 2); 652 break; 653 case INAT_IMM_DWORD: 654 insn_field_set(&insn->immediate, get_next(int, insn), 4); 655 break; 656 case INAT_IMM_QWORD: 657 insn_field_set(&insn->immediate1, get_next(int, insn), 4); 658 insn_field_set(&insn->immediate2, get_next(int, insn), 4); 659 break; 660 case INAT_IMM_PTR: 661 if (!__get_immptr(insn)) 662 goto err_out; 663 break; 664 case INAT_IMM_VWORD32: 665 if (!__get_immv32(insn)) 666 goto err_out; 667 break; 668 case INAT_IMM_VWORD: 669 if (!__get_immv(insn)) 670 goto err_out; 671 break; 672 default: 673 /* Here, insn must have an immediate, but failed */ 674 goto err_out; 675 } 676 if (inat_has_second_immediate(insn->attr)) { 677 insn_field_set(&insn->immediate2, get_next(signed char, insn), 1); 678 } 679 done: 680 insn->immediate.got = 1; 681 return 0; 682 683 err_out: 684 return -ENODATA; 685 } 686 687 /** 688 * insn_get_length() - Get the length of instruction 689 * @insn: &struct insn containing instruction 690 * 691 * If necessary, first collects the instruction up to and including the 692 * immediates bytes. 693 * 694 * Returns: 695 * - 0 on success 696 * - < 0 on error 697 */ 698 int insn_get_length(struct insn *insn) 699 { 700 int ret; 701 702 if (insn->length) 703 return 0; 704 705 if (!insn->immediate.got) { 706 ret = insn_get_immediate(insn); 707 if (ret) 708 return ret; 709 } 710 711 insn->length = (unsigned char)((unsigned long)insn->next_byte 712 - (unsigned long)insn->kaddr); 713 714 return 0; 715 } 716 717 /* Ensure this instruction is decoded completely */ 718 static inline int insn_complete(struct insn *insn) 719 { 720 return insn->opcode.got && insn->modrm.got && insn->sib.got && 721 insn->displacement.got && insn->immediate.got; 722 } 723 724 /** 725 * insn_decode() - Decode an x86 instruction 726 * @insn: &struct insn to be initialized 727 * @kaddr: address (in kernel memory) of instruction (or copy thereof) 728 * @buf_len: length of the insn buffer at @kaddr 729 * @m: insn mode, see enum insn_mode 730 * 731 * Returns: 732 * 0: if decoding succeeded 733 * < 0: otherwise. 734 */ 735 int insn_decode(struct insn *insn, const void *kaddr, int buf_len, enum insn_mode m) 736 { 737 int ret; 738 739 /* #define INSN_MODE_KERN -1 __ignore_sync_check__ mode is only valid in the kernel */ 740 741 if (m == INSN_MODE_KERN) 742 insn_init(insn, kaddr, buf_len, IS_ENABLED(CONFIG_X86_64)); 743 else 744 insn_init(insn, kaddr, buf_len, m == INSN_MODE_64); 745 746 ret = insn_get_length(insn); 747 if (ret) 748 return ret; 749 750 if (insn_complete(insn)) 751 return 0; 752 753 return -EINVAL; 754 } 755