1 // SPDX-License-Identifier: GPL-2.0-or-later 1
2 /*
3 * x86 instruction analysis
4 *
5 * Copyright (C) IBM Corporation, 2002, 2004,
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 "../include/asm/inat.h" /* __ignore_s
15 #include "../include/asm/insn.h" /* __ignore_s
16 #include "../include/linux/unaligned.h" /* __i
17
18 #include <linux/errno.h>
19 #include <linux/kconfig.h>
20
21 #include "../include/asm/emulate_prefix.h" /*
22
23 #define leXX_to_cpu(t, r)
24 ({
25 __typeof__(t) v;
26 switch (sizeof(t)) {
27 case 4: v = le32_to_cpu(r); break;
28 case 2: v = le16_to_cpu(r); break;
29 case 1: v = r; break;
30 default:
31 BUILD_BUG(); break;
32 }
33 v;
34 })
35
36 /* Verify next sizeof(t) bytes can be on the s
37 #define validate_next(t, insn, n) \
38 ((insn)->next_byte + sizeof(t) + n <=
39
40 #define __get_next(t, insn) \
41 ({ t r = get_unaligned((t *)(insn)->ne
42
43 #define __peek_nbyte_next(t, insn, n) \
44 ({ t r = get_unaligned((t *)(insn)->ne
45
46 #define get_next(t, insn) \
47 ({ if (unlikely(!validate_next(t, insn
48
49 #define peek_nbyte_next(t, insn, n) \
50 ({ if (unlikely(!validate_next(t, insn
51
52 #define peek_next(t, insn) peek_nbyte_nex
53
54 /**
55 * insn_init() - initialize struct insn
56 * @insn: &struct insn to be initialized
57 * @kaddr: address (in kernel memory) of
58 * @buf_len: length of the insn buffer at @
59 * @x86_64: !0 for 64-bit kernel or 64-bit
60 */
61 void insn_init(struct insn *insn, const void *
62 {
63 /*
64 * Instructions longer than MAX_INSN_S
65 * even if the input buffer is long en
66 */
67 if (buf_len > MAX_INSN_SIZE)
68 buf_len = MAX_INSN_SIZE;
69
70 memset(insn, 0, sizeof(*insn));
71 insn->kaddr = kaddr;
72 insn->end_kaddr = kaddr + buf_len;
73 insn->next_byte = kaddr;
74 insn->x86_64 = x86_64;
75 insn->opnd_bytes = 4;
76 if (x86_64)
77 insn->addr_bytes = 8;
78 else
79 insn->addr_bytes = 4;
80 }
81
82 static const insn_byte_t xen_prefix[] = { __XE
83 static const insn_byte_t kvm_prefix[] = { __KV
84
85 static int __insn_get_emulate_prefix(struct in
86 const ins
87 {
88 size_t i;
89
90 for (i = 0; i < len; i++) {
91 if (peek_nbyte_next(insn_byte_
92 goto err_out;
93 }
94
95 insn->emulate_prefix_size = len;
96 insn->next_byte += len;
97
98 return 1;
99
100 err_out:
101 return 0;
102 }
103
104 static void insn_get_emulate_prefix(struct ins
105 {
106 if (__insn_get_emulate_prefix(insn, xe
107 return;
108
109 __insn_get_emulate_prefix(insn, kvm_pr
110 }
111
112 /**
113 * insn_get_prefixes - scan x86 instruction pr
114 * @insn: &struct insn containing instru
115 *
116 * Populates the @insn->prefixes bitmap, and u
117 * to point to the (first) opcode. No effect
118 * is already set.
119 *
120 * * Returns:
121 * 0: on success
122 * < 0: on error
123 */
124 int insn_get_prefixes(struct insn *insn)
125 {
126 struct insn_field *prefixes = &insn->p
127 insn_attr_t attr;
128 insn_byte_t b, lb;
129 int i, nb;
130
131 if (prefixes->got)
132 return 0;
133
134 insn_get_emulate_prefix(insn);
135
136 nb = 0;
137 lb = 0;
138 b = peek_next(insn_byte_t, insn);
139 attr = inat_get_opcode_attribute(b);
140 while (inat_is_legacy_prefix(attr)) {
141 /* Skip if same prefix */
142 for (i = 0; i < nb; i++)
143 if (prefixes->bytes[i]
144 goto found;
145 if (nb == 4)
146 /* Invalid instruction
147 break;
148 prefixes->bytes[nb++] = b;
149 if (inat_is_address_size_prefi
150 /* address size switch
151 if (insn->x86_64)
152 insn->addr_byt
153 else
154 insn->addr_byt
155 } else if (inat_is_operand_siz
156 /* oprand size switche
157 insn->opnd_bytes ^= 6;
158 }
159 found:
160 prefixes->nbytes++;
161 insn->next_byte++;
162 lb = b;
163 b = peek_next(insn_byte_t, ins
164 attr = inat_get_opcode_attribu
165 }
166 /* Set the last prefix */
167 if (lb && lb != insn->prefixes.bytes[3
168 if (unlikely(insn->prefixes.by
169 /* Swap the last prefi
170 b = insn->prefixes.byt
171 for (i = 0; i < nb; i+
172 if (prefixes->
173 insn_s
174 }
175 insn_set_byte(&insn->prefixes,
176 }
177
178 /* Decode REX prefix */
179 if (insn->x86_64) {
180 b = peek_next(insn_byte_t, ins
181 attr = inat_get_opcode_attribu
182 if (inat_is_rex_prefix(attr))
183 insn_field_set(&insn->
184 insn->next_byte++;
185 if (X86_REX_W(b))
186 /* REX.W overr
187 insn->opnd_byt
188 } else if (inat_is_rex2_prefix
189 insn_set_byte(&insn->r
190 b = peek_nbyte_next(in
191 insn_set_byte(&insn->r
192 insn->rex_prefix.nbyte
193 insn->next_byte += 2;
194 if (X86_REX_W(b))
195 /* REX.W overr
196 insn->opnd_byt
197 insn->rex_prefix.got =
198 goto vex_end;
199 }
200 }
201 insn->rex_prefix.got = 1;
202
203 /* Decode VEX prefix */
204 b = peek_next(insn_byte_t, insn);
205 attr = inat_get_opcode_attribute(b);
206 if (inat_is_vex_prefix(attr)) {
207 insn_byte_t b2 = peek_nbyte_ne
208 if (!insn->x86_64) {
209 /*
210 * In 32-bits mode, if
211 * ModRM) on the secon
212 * LDS or LES or BOUND
213 */
214 if (X86_MODRM_MOD(b2)
215 goto vex_end;
216 }
217 insn_set_byte(&insn->vex_prefi
218 insn_set_byte(&insn->vex_prefi
219 if (inat_is_evex_prefix(attr))
220 b2 = peek_nbyte_next(i
221 insn_set_byte(&insn->v
222 b2 = peek_nbyte_next(i
223 insn_set_byte(&insn->v
224 insn->vex_prefix.nbyte
225 insn->next_byte += 4;
226 if (insn->x86_64 && X8
227 /* VEX.W overr
228 insn->opnd_byt
229 } else if (inat_is_vex3_prefix
230 b2 = peek_nbyte_next(i
231 insn_set_byte(&insn->v
232 insn->vex_prefix.nbyte
233 insn->next_byte += 3;
234 if (insn->x86_64 && X8
235 /* VEX.W overr
236 insn->opnd_byt
237 } else {
238 /*
239 * For VEX2, fake VEX3
240 * Makes it easier to
241 * vex.L and vex.pp. M
242 */
243 insn_set_byte(&insn->v
244 insn->vex_prefix.nbyte
245 insn->next_byte += 2;
246 }
247 }
248 vex_end:
249 insn->vex_prefix.got = 1;
250
251 prefixes->got = 1;
252
253 return 0;
254
255 err_out:
256 return -ENODATA;
257 }
258
259 /**
260 * insn_get_opcode - collect opcode(s)
261 * @insn: &struct insn containing instru
262 *
263 * Populates @insn->opcode, updates @insn->nex
264 * opcode byte(s), and set @insn->attr (except
265 * If necessary, first collects any preceding
266 * Sets @insn->opcode.value = opcode1. No eff
267 * is already 1.
268 *
269 * Returns:
270 * 0: on success
271 * < 0: on error
272 */
273 int insn_get_opcode(struct insn *insn)
274 {
275 struct insn_field *opcode = &insn->opc
276 int pfx_id, ret;
277 insn_byte_t op;
278
279 if (opcode->got)
280 return 0;
281
282 ret = insn_get_prefixes(insn);
283 if (ret)
284 return ret;
285
286 /* Get first opcode */
287 op = get_next(insn_byte_t, insn);
288 insn_set_byte(opcode, 0, op);
289 opcode->nbytes = 1;
290
291 /* Check if there is VEX prefix or not
292 if (insn_is_avx(insn)) {
293 insn_byte_t m, p;
294 m = insn_vex_m_bits(insn);
295 p = insn_vex_p_bits(insn);
296 insn->attr = inat_get_avx_attr
297 /* SCALABLE EVEX uses p bits t
298 if (inat_evex_scalable(insn->a
299 p == INAT_PFX_OPNDSZ)
300 insn->opnd_bytes = 2;
301 if ((inat_must_evex(insn->attr
302 (!inat_accept_vex(insn->at
303 !inat_is_group(insn->attr
304 /* This instruction is
305 insn->attr = 0;
306 return -EINVAL;
307 }
308 /* VEX has only 1 byte for opc
309 goto end;
310 }
311
312 /* Check if there is REX2 prefix or no
313 if (insn_is_rex2(insn)) {
314 if (insn_rex2_m_bit(insn)) {
315 /* map 1 is escape 0x0
316 insn_attr_t esc_attr =
317
318 pfx_id = insn_last_pre
319 insn->attr = inat_get_
320 } else {
321 insn->attr = inat_get_
322 }
323 goto end;
324 }
325
326 insn->attr = inat_get_opcode_attribute
327 while (inat_is_escape(insn->attr)) {
328 /* Get escaped opcode */
329 op = get_next(insn_byte_t, ins
330 opcode->bytes[opcode->nbytes++
331 pfx_id = insn_last_prefix_id(i
332 insn->attr = inat_get_escape_a
333 }
334
335 if (inat_must_vex(insn->attr)) {
336 /* This instruction is bad */
337 insn->attr = 0;
338 return -EINVAL;
339 }
340 end:
341 opcode->got = 1;
342 return 0;
343
344 err_out:
345 return -ENODATA;
346 }
347
348 /**
349 * insn_get_modrm - collect ModRM byte, if any
350 * @insn: &struct insn containing instru
351 *
352 * Populates @insn->modrm and updates @insn->n
353 * ModRM byte, if any. If necessary, first co
354 * (prefixes and opcode(s)). No effect if @in
355 *
356 * Returns:
357 * 0: on success
358 * < 0: on error
359 */
360 int insn_get_modrm(struct insn *insn)
361 {
362 struct insn_field *modrm = &insn->modr
363 insn_byte_t pfx_id, mod;
364 int ret;
365
366 if (modrm->got)
367 return 0;
368
369 ret = insn_get_opcode(insn);
370 if (ret)
371 return ret;
372
373 if (inat_has_modrm(insn->attr)) {
374 mod = get_next(insn_byte_t, in
375 insn_field_set(modrm, mod, 1);
376 if (inat_is_group(insn->attr))
377 pfx_id = insn_last_pre
378 insn->attr = inat_get_
379
380 if (insn_is_avx(insn)
381 /* Bad insn */
382 insn->attr = 0
383 return -EINVAL
384 }
385 }
386 }
387
388 if (insn->x86_64 && inat_is_force64(in
389 insn->opnd_bytes = 8;
390
391 modrm->got = 1;
392 return 0;
393
394 err_out:
395 return -ENODATA;
396 }
397
398
399 /**
400 * insn_rip_relative() - Does instruction use
401 * @insn: &struct insn containing instru
402 *
403 * If necessary, first collects the instructio
404 * ModRM byte. No effect if @insn->x86_64 is
405 */
406 int insn_rip_relative(struct insn *insn)
407 {
408 struct insn_field *modrm = &insn->modr
409 int ret;
410
411 if (!insn->x86_64)
412 return 0;
413
414 ret = insn_get_modrm(insn);
415 if (ret)
416 return 0;
417 /*
418 * For rip-relative instructions, the
419 * is zero and the r/m field (bottom 3
420 */
421 return (modrm->nbytes && (modrm->bytes
422 }
423
424 /**
425 * insn_get_sib() - Get the SIB byte of instru
426 * @insn: &struct insn containing instru
427 *
428 * If necessary, first collects the instructio
429 * ModRM byte.
430 *
431 * Returns:
432 * 0: if decoding succeeded
433 * < 0: otherwise.
434 */
435 int insn_get_sib(struct insn *insn)
436 {
437 insn_byte_t modrm;
438 int ret;
439
440 if (insn->sib.got)
441 return 0;
442
443 ret = insn_get_modrm(insn);
444 if (ret)
445 return ret;
446
447 if (insn->modrm.nbytes) {
448 modrm = insn->modrm.bytes[0];
449 if (insn->addr_bytes != 2 &&
450 X86_MODRM_MOD(modrm) != 3
451 insn_field_set(&insn->
452 get_nex
453 }
454 }
455 insn->sib.got = 1;
456
457 return 0;
458
459 err_out:
460 return -ENODATA;
461 }
462
463
464 /**
465 * insn_get_displacement() - Get the displacem
466 * @insn: &struct insn containing instru
467 *
468 * If necessary, first collects the instructio
469 * SIB byte.
470 * Displacement value is sign-expanded.
471 *
472 * * Returns:
473 * 0: if decoding succeeded
474 * < 0: otherwise.
475 */
476 int insn_get_displacement(struct insn *insn)
477 {
478 insn_byte_t mod, rm, base;
479 int ret;
480
481 if (insn->displacement.got)
482 return 0;
483
484 ret = insn_get_sib(insn);
485 if (ret)
486 return ret;
487
488 if (insn->modrm.nbytes) {
489 /*
490 * Interpreting the modrm byte
491 * mod = 00 - no displacement
492 * mod = 01 - 1-byte displacem
493 * mod = 10 - displacement fie
494 * address size = 2 (0x67
495 * mod = 11 - no memory operan
496 *
497 * If address size = 2...
498 * mod = 00, r/m = 110 - displ
499 *
500 * If address size != 2...
501 * mod != 11, r/m = 100 - SIB
502 * mod = 00, SIB base = 101 -
503 * mod = 00, r/m = 101 - rip-r
504 * field is 4 bytes
505 */
506 mod = X86_MODRM_MOD(insn->modr
507 rm = X86_MODRM_RM(insn->modrm.
508 base = X86_SIB_BASE(insn->sib.
509 if (mod == 3)
510 goto out;
511 if (mod == 1) {
512 insn_field_set(&insn->
513 get_nex
514 } else if (insn->addr_bytes ==
515 if ((mod == 0 && rm ==
516 insn_field_set
517
518 }
519 } else {
520 if ((mod == 0 && rm ==
521 (mod == 0 && base
522 insn_field_set
523
524 }
525 }
526 }
527 out:
528 insn->displacement.got = 1;
529 return 0;
530
531 err_out:
532 return -ENODATA;
533 }
534
535 /* Decode moffset16/32/64. Return 0 if failed
536 static int __get_moffset(struct insn *insn)
537 {
538 switch (insn->addr_bytes) {
539 case 2:
540 insn_field_set(&insn->moffset1
541 break;
542 case 4:
543 insn_field_set(&insn->moffset1
544 break;
545 case 8:
546 insn_field_set(&insn->moffset1
547 insn_field_set(&insn->moffset2
548 break;
549 default: /* opnd_bytes must be
550 goto err_out;
551 }
552 insn->moffset1.got = insn->moffset2.go
553
554 return 1;
555
556 err_out:
557 return 0;
558 }
559
560 /* Decode imm v32(Iz). Return 0 if failed */
561 static int __get_immv32(struct insn *insn)
562 {
563 switch (insn->opnd_bytes) {
564 case 2:
565 insn_field_set(&insn->immediat
566 break;
567 case 4:
568 case 8:
569 insn_field_set(&insn->immediat
570 break;
571 default: /* opnd_bytes must be
572 goto err_out;
573 }
574
575 return 1;
576
577 err_out:
578 return 0;
579 }
580
581 /* Decode imm v64(Iv/Ov), Return 0 if failed *
582 static int __get_immv(struct insn *insn)
583 {
584 switch (insn->opnd_bytes) {
585 case 2:
586 insn_field_set(&insn->immediat
587 break;
588 case 4:
589 insn_field_set(&insn->immediat
590 insn->immediate1.nbytes = 4;
591 break;
592 case 8:
593 insn_field_set(&insn->immediat
594 insn_field_set(&insn->immediat
595 break;
596 default: /* opnd_bytes must be
597 goto err_out;
598 }
599 insn->immediate1.got = insn->immediate
600
601 return 1;
602 err_out:
603 return 0;
604 }
605
606 /* Decode ptr16:16/32(Ap) */
607 static int __get_immptr(struct insn *insn)
608 {
609 switch (insn->opnd_bytes) {
610 case 2:
611 insn_field_set(&insn->immediat
612 break;
613 case 4:
614 insn_field_set(&insn->immediat
615 break;
616 case 8:
617 /* ptr16:64 is not exist (no s
618 return 0;
619 default: /* opnd_bytes must be
620 goto err_out;
621 }
622 insn_field_set(&insn->immediate2, get_
623 insn->immediate1.got = insn->immediate
624
625 return 1;
626 err_out:
627 return 0;
628 }
629
630 /**
631 * insn_get_immediate() - Get the immediate in
632 * @insn: &struct insn containing instru
633 *
634 * If necessary, first collects the instructio
635 * displacement bytes.
636 * Basically, most of immediates are sign-expa
637 * computed by bit masking with ((1 << (nbytes
638 *
639 * Returns:
640 * 0: on success
641 * < 0: on error
642 */
643 int insn_get_immediate(struct insn *insn)
644 {
645 int ret;
646
647 if (insn->immediate.got)
648 return 0;
649
650 ret = insn_get_displacement(insn);
651 if (ret)
652 return ret;
653
654 if (inat_has_moffset(insn->attr)) {
655 if (!__get_moffset(insn))
656 goto err_out;
657 goto done;
658 }
659
660 if (!inat_has_immediate(insn->attr))
661 /* no immediates */
662 goto done;
663
664 switch (inat_immediate_size(insn->attr
665 case INAT_IMM_BYTE:
666 insn_field_set(&insn->immediat
667 break;
668 case INAT_IMM_WORD:
669 insn_field_set(&insn->immediat
670 break;
671 case INAT_IMM_DWORD:
672 insn_field_set(&insn->immediat
673 break;
674 case INAT_IMM_QWORD:
675 insn_field_set(&insn->immediat
676 insn_field_set(&insn->immediat
677 break;
678 case INAT_IMM_PTR:
679 if (!__get_immptr(insn))
680 goto err_out;
681 break;
682 case INAT_IMM_VWORD32:
683 if (!__get_immv32(insn))
684 goto err_out;
685 break;
686 case INAT_IMM_VWORD:
687 if (!__get_immv(insn))
688 goto err_out;
689 break;
690 default:
691 /* Here, insn must have an imm
692 goto err_out;
693 }
694 if (inat_has_second_immediate(insn->at
695 insn_field_set(&insn->immediat
696 }
697 done:
698 insn->immediate.got = 1;
699 return 0;
700
701 err_out:
702 return -ENODATA;
703 }
704
705 /**
706 * insn_get_length() - Get the length of instr
707 * @insn: &struct insn containing instru
708 *
709 * If necessary, first collects the instructio
710 * immediates bytes.
711 *
712 * Returns:
713 * - 0 on success
714 * - < 0 on error
715 */
716 int insn_get_length(struct insn *insn)
717 {
718 int ret;
719
720 if (insn->length)
721 return 0;
722
723 ret = insn_get_immediate(insn);
724 if (ret)
725 return ret;
726
727 insn->length = (unsigned char)((unsign
728 - (unsign
729
730 return 0;
731 }
732
733 /* Ensure this instruction is decoded complete
734 static inline int insn_complete(struct insn *i
735 {
736 return insn->opcode.got && insn->modrm
737 insn->displacement.got && insn
738 }
739
740 /**
741 * insn_decode() - Decode an x86 instruction
742 * @insn: &struct insn to be initialized
743 * @kaddr: address (in kernel memory) of
744 * @buf_len: length of the insn buffer at @
745 * @m: insn mode, see enum insn_mode
746 *
747 * Returns:
748 * 0: if decoding succeeded
749 * < 0: otherwise.
750 */
751 int insn_decode(struct insn *insn, const void
752 {
753 int ret;
754
755 #define INSN_MODE_KERN (enum insn_mode)-1 /* _
756
757 if (m == INSN_MODE_KERN)
758 insn_init(insn, kaddr, buf_len
759 else
760 insn_init(insn, kaddr, buf_len
761
762 ret = insn_get_length(insn);
763 if (ret)
764 return ret;
765
766 if (insn_complete(insn))
767 return 0;
768
769 return -EINVAL;
770 }
771
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.