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