?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Add copyright notice to insns.dat
[nasm.git] / disasm.c
blob4cf25ef2ba712c4b89e05371a5e490ed8383ff3e
1 /* disasm.c where all the _work_ gets done in the Netwide Disassembler
2 *
3 * The Netwide Assembler is copyright (C) 1996 Simon Tatham and
4 * Julian Hall. All rights reserved. The software is
5 * redistributable under the license given in the file "LICENSE"
6 * distributed in the NASM archive.
7 *
8 * initial version 27/iii/95 by Simon Tatham
9 */
10
11 #include "compiler.h"
12
13 #include <stdio.h>
14 #include <string.h>
15 #include <limits.h>
16 #include <inttypes.h>
17
18 #include "nasm.h"
19 #include "disasm.h"
20 #include "sync.h"
21 #include "insns.h"
22 #include "tables.h"
23 #include "regdis.h"
24
25 /*
26 * Flags that go into the `segment' field of `insn' structures
27 * during disassembly.
28 */
29 #define SEG_RELATIVE 1
30 #define SEG_32BIT 2
31 #define SEG_RMREG 4
32 #define SEG_DISP8 8
33 #define SEG_DISP16 16
34 #define SEG_DISP32 32
35 #define SEG_NODISP 64
36 #define SEG_SIGNED 128
37 #define SEG_64BIT 256
38
39 /*
40 * Prefix information
41 */
42 struct prefix_info {
43 uint8_t osize; /* Operand size */
44 uint8_t asize; /* Address size */
45 uint8_t osp; /* Operand size prefix present */
46 uint8_t asp; /* Address size prefix present */
47 uint8_t rep; /* Rep prefix present */
48 uint8_t seg; /* Segment override prefix present */
49 uint8_t wait; /* WAIT "prefix" present */
50 uint8_t lock; /* Lock prefix present */
51 uint8_t vex[3]; /* VEX prefix present */
52 uint8_t vex_c; /* VEX "class" (VEX, XOP, ...) */
53 uint8_t vex_m; /* VEX.M field */
54 uint8_t vex_v;
55 uint8_t vex_lp; /* VEX.LP fields */
56 uint32_t rex; /* REX prefix present */
57 };
58
59 #define getu8(x) (*(uint8_t *)(x))
60 #if X86_MEMORY
61 /* Littleendian CPU which can handle unaligned references */
62 #define getu16(x) (*(uint16_t *)(x))
63 #define getu32(x) (*(uint32_t *)(x))
64 #define getu64(x) (*(uint64_t *)(x))
65 #else
66 static uint16_t getu16(uint8_t *data)
67 {
68 return (uint16_t)data[0] + ((uint16_t)data[1] << 8);
69 }
70 static uint32_t getu32(uint8_t *data)
71 {
72 return (uint32_t)getu16(data) + ((uint32_t)getu16(data+2) << 16);
73 }
74 static uint64_t getu64(uint8_t *data)
75 {
76 return (uint64_t)getu32(data) + ((uint64_t)getu32(data+4) << 32);
77 }
78 #endif
79
80 #define gets8(x) ((int8_t)getu8(x))
81 #define gets16(x) ((int16_t)getu16(x))
82 #define gets32(x) ((int32_t)getu32(x))
83 #define gets64(x) ((int64_t)getu64(x))
84
85 /* Important: regval must already have been adjusted for rex extensions */
86 static enum reg_enum whichreg(int32_t regflags, int regval, int rex)
87 {
88 if (!(regflags & (REGISTER|REGMEM)))
89 return 0; /* Registers not permissible?! */
90
91 regflags |= REGISTER;
92
93 if (!(REG_AL & ~regflags))
94 return R_AL;
95 if (!(REG_AX & ~regflags))
96 return R_AX;
97 if (!(REG_EAX & ~regflags))
98 return R_EAX;
99 if (!(REG_RAX & ~regflags))
100 return R_RAX;
101 if (!(REG_DL & ~regflags))
102 return R_DL;
103 if (!(REG_DX & ~regflags))
104 return R_DX;
105 if (!(REG_EDX & ~regflags))
106 return R_EDX;
107 if (!(REG_RDX & ~regflags))
108 return R_RDX;
109 if (!(REG_CL & ~regflags))
110 return R_CL;
111 if (!(REG_CX & ~regflags))
112 return R_CX;
113 if (!(REG_ECX & ~regflags))
114 return R_ECX;
115 if (!(REG_RCX & ~regflags))
116 return R_RCX;
117 if (!(FPU0 & ~regflags))
118 return R_ST0;
119 if (!(XMM0 & ~regflags))
120 return R_XMM0;
121 if (!(YMM0 & ~regflags))
122 return R_YMM0;
123 if (!(REG_CS & ~regflags))
124 return (regval == 1) ? R_CS : 0;
125 if (!(REG_DESS & ~regflags))
126 return (regval == 0 || regval == 2
127 || regval == 3 ? nasm_rd_sreg[regval] : 0);
128 if (!(REG_FSGS & ~regflags))
129 return (regval == 4 || regval == 5 ? nasm_rd_sreg[regval] : 0);
130 if (!(REG_SEG67 & ~regflags))
131 return (regval == 6 || regval == 7 ? nasm_rd_sreg[regval] : 0);
132
133 /* All the entries below look up regval in an 16-entry array */
134 if (regval < 0 || regval > 15)
135 return 0;
136
137 if (!(REG8 & ~regflags)) {
138 if (rex & (REX_P|REX_NH))
139 return nasm_rd_reg8_rex[regval];
140 else
141 return nasm_rd_reg8[regval];
142 }
143 if (!(REG16 & ~regflags))
144 return nasm_rd_reg16[regval];
145 if (!(REG32 & ~regflags))
146 return nasm_rd_reg32[regval];
147 if (!(REG64 & ~regflags))
148 return nasm_rd_reg64[regval];
149 if (!(REG_SREG & ~regflags))
150 return nasm_rd_sreg[regval & 7]; /* Ignore REX */
151 if (!(REG_CREG & ~regflags))
152 return nasm_rd_creg[regval];
153 if (!(REG_DREG & ~regflags))
154 return nasm_rd_dreg[regval];
155 if (!(REG_TREG & ~regflags)) {
156 if (regval > 7)
157 return 0; /* TR registers are ill-defined with rex */
158 return nasm_rd_treg[regval];
159 }
160 if (!(FPUREG & ~regflags))
161 return nasm_rd_fpureg[regval & 7]; /* Ignore REX */
162 if (!(MMXREG & ~regflags))
163 return nasm_rd_mmxreg[regval & 7]; /* Ignore REX */
164 if (!(XMMREG & ~regflags))
165 return nasm_rd_xmmreg[regval];
166 if (!(YMMREG & ~regflags))
167 return nasm_rd_ymmreg[regval];
168
169 return 0;
170 }
171
172 /*
173 * Process a DREX suffix
174 */
175 static uint8_t *do_drex(uint8_t *data, insn *ins)
176 {
177 uint8_t drex = *data++;
178 operand *dst = &ins->oprs[ins->drexdst];
179
180 if ((drex & 8) != ((ins->rex & REX_OC) ? 8 : 0))
181 return NULL; /* OC0 mismatch */
182 ins->rex = (ins->rex & ~7) | (drex & 7);
183
184 dst->segment = SEG_RMREG;
185 dst->basereg = drex >> 4;
186 return data;
187 }
188
189
190 /*
191 * Process an effective address (ModRM) specification.
192 */
193 static uint8_t *do_ea(uint8_t *data, int modrm, int asize,
194 int segsize, operand * op, insn *ins)
195 {
196 int mod, rm, scale, index, base;
197 int rex;
198 uint8_t sib = 0;
199
200 mod = (modrm >> 6) & 03;
201 rm = modrm & 07;
202
203 if (mod != 3 && rm == 4 && asize != 16)
204 sib = *data++;
205
206 if (ins->rex & REX_D) {
207 data = do_drex(data, ins);
208 if (!data)
209 return NULL;
210 }
211 rex = ins->rex;
212
213 if (mod == 3) { /* pure register version */
214 op->basereg = rm+(rex & REX_B ? 8 : 0);
215 op->segment |= SEG_RMREG;
216 return data;
217 }
218
219 op->disp_size = 0;
220 op->eaflags = 0;
221
222 if (asize == 16) {
223 /*
224 * <mod> specifies the displacement size (none, byte or
225 * word), and <rm> specifies the register combination.
226 * Exception: mod=0,rm=6 does not specify [BP] as one might
227 * expect, but instead specifies [disp16].
228 */
229 op->indexreg = op->basereg = -1;
230 op->scale = 1; /* always, in 16 bits */
231 switch (rm) {
232 case 0:
233 op->basereg = R_BX;
234 op->indexreg = R_SI;
235 break;
236 case 1:
237 op->basereg = R_BX;
238 op->indexreg = R_DI;
239 break;
240 case 2:
241 op->basereg = R_BP;
242 op->indexreg = R_SI;
243 break;
244 case 3:
245 op->basereg = R_BP;
246 op->indexreg = R_DI;
247 break;
248 case 4:
249 op->basereg = R_SI;
250 break;
251 case 5:
252 op->basereg = R_DI;
253 break;
254 case 6:
255 op->basereg = R_BP;
256 break;
257 case 7:
258 op->basereg = R_BX;
259 break;
260 }
261 if (rm == 6 && mod == 0) { /* special case */
262 op->basereg = -1;
263 if (segsize != 16)
264 op->disp_size = 16;
265 mod = 2; /* fake disp16 */
266 }
267 switch (mod) {
268 case 0:
269 op->segment |= SEG_NODISP;
270 break;
271 case 1:
272 op->segment |= SEG_DISP8;
273 op->offset = (int8_t)*data++;
274 break;
275 case 2:
276 op->segment |= SEG_DISP16;
277 op->offset = *data++;
278 op->offset |= ((unsigned)*data++) << 8;
279 break;
280 }
281 return data;
282 } else {
283 /*
284 * Once again, <mod> specifies displacement size (this time
285 * none, byte or *dword*), while <rm> specifies the base
286 * register. Again, [EBP] is missing, replaced by a pure
287 * disp32 (this time that's mod=0,rm=*5*) in 32-bit mode,
288 * and RIP-relative addressing in 64-bit mode.
289 *
290 * However, rm=4
291 * indicates not a single base register, but instead the
292 * presence of a SIB byte...
293 */
294 int a64 = asize == 64;
295
296 op->indexreg = -1;
297
298 if (a64)
299 op->basereg = nasm_rd_reg64[rm | ((rex & REX_B) ? 8 : 0)];
300 else
301 op->basereg = nasm_rd_reg32[rm | ((rex & REX_B) ? 8 : 0)];
302
303 if (rm == 5 && mod == 0) {
304 if (segsize == 64) {
305 op->eaflags |= EAF_REL;
306 op->segment |= SEG_RELATIVE;
307 mod = 2; /* fake disp32 */
308 }
309
310 if (asize != 64)
311 op->disp_size = asize;
312
313 op->basereg = -1;
314 mod = 2; /* fake disp32 */
315 }
316
317 if (rm == 4) { /* process SIB */
318 scale = (sib >> 6) & 03;
319 index = (sib >> 3) & 07;
320 base = sib & 07;
321
322 op->scale = 1 << scale;
323
324 if (index == 4 && !(rex & REX_X))
325 op->indexreg = -1; /* ESP/RSP cannot be an index */
326 else if (a64)
327 op->indexreg = nasm_rd_reg64[index | ((rex & REX_X) ? 8 : 0)];
328 else
329 op->indexreg = nasm_rd_reg32[index | ((rex & REX_X) ? 8 : 0)];
330
331 if (base == 5 && mod == 0) {
332 op->basereg = -1;
333 mod = 2; /* Fake disp32 */
334 } else if (a64)
335 op->basereg = nasm_rd_reg64[base | ((rex & REX_B) ? 8 : 0)];
336 else
337 op->basereg = nasm_rd_reg32[base | ((rex & REX_B) ? 8 : 0)];
338
339 if (segsize == 16)
340 op->disp_size = 32;
341 }
342
343 switch (mod) {
344 case 0:
345 op->segment |= SEG_NODISP;
346 break;
347 case 1:
348 op->segment |= SEG_DISP8;
349 op->offset = gets8(data);
350 data++;
351 break;
352 case 2:
353 op->segment |= SEG_DISP32;
354 op->offset = gets32(data);
355 data += 4;
356 break;
357 }
358 return data;
359 }
360 }
361
362 /*
363 * Determine whether the instruction template in t corresponds to the data
364 * stream in data. Return the number of bytes matched if so.
365 */
366 #define case4(x) case (x): case (x)+1: case (x)+2: case (x)+3
367
368 static int matches(const struct itemplate *t, uint8_t *data,
369 const struct prefix_info *prefix, int segsize, insn *ins)
370 {
371 uint8_t *r = (uint8_t *)(t->code);
372 uint8_t *origdata = data;
373 bool a_used = false, o_used = false;
374 enum prefixes drep = 0;
375 enum prefixes dwait = 0;
376 uint8_t lock = prefix->lock;
377 int osize = prefix->osize;
378 int asize = prefix->asize;
379 int i, c;
380 int op1, op2;
381 struct operand *opx, *opy;
382 uint8_t opex = 0;
383 int s_field_for = -1; /* No 144/154 series code encountered */
384 bool vex_ok = false;
385 int regmask = (segsize == 64) ? 15 : 7;
386
387 for (i = 0; i < MAX_OPERANDS; i++) {
388 ins->oprs[i].segment = ins->oprs[i].disp_size =
389 (segsize == 64 ? SEG_64BIT : segsize == 32 ? SEG_32BIT : 0);
390 }
391 ins->condition = -1;
392 ins->rex = prefix->rex;
393 memset(ins->prefixes, 0, sizeof ins->prefixes);
394
395 if (t->flags & (segsize == 64 ? IF_NOLONG : IF_LONG))
396 return false;
397
398 if (prefix->rep == 0xF2)
399 drep = P_REPNE;
400 else if (prefix->rep == 0xF3)
401 drep = P_REP;
402
403 dwait = prefix->wait ? P_WAIT : 0;
404
405 while ((c = *r++) != 0) {
406 op1 = (c & 3) + ((opex & 1) << 2);
407 op2 = ((c >> 3) & 3) + ((opex & 2) << 1);
408 opx = &ins->oprs[op1];
409 opy = &ins->oprs[op2];
410 opex = 0;
411
412 switch (c) {
413 case 01:
414 case 02:
415 case 03:
416 case 04:
417 while (c--)
418 if (*r++ != *data++)
419 return false;
420 break;
421
422 case 05:
423 case 06:
424 case 07:
425 opex = c;
426 break;
427
428 case4(010):
429 {
430 int t = *r++, d = *data++;
431 if (d < t || d > t + 7)
432 return false;
433 else {
434 opx->basereg = (d-t)+
435 (ins->rex & REX_B ? 8 : 0);
436 opx->segment |= SEG_RMREG;
437 }
438 break;
439 }
440
441 case4(014):
442 case4(0274):
443 opx->offset = (int8_t)*data++;
444 opx->segment |= SEG_SIGNED;
445 break;
446
447 case4(020):
448 opx->offset = *data++;
449 break;
450
451 case4(024):
452 opx->offset = *data++;
453 break;
454
455 case4(030):
456 opx->offset = getu16(data);
457 data += 2;
458 break;
459
460 case4(034):
461 if (osize == 32) {
462 opx->offset = getu32(data);
463 data += 4;
464 } else {
465 opx->offset = getu16(data);
466 data += 2;
467 }
468 if (segsize != asize)
469 opx->disp_size = asize;
470 break;
471
472 case4(040):
473 case4(0254):
474 opx->offset = getu32(data);
475 data += 4;
476 break;
477
478 case4(044):
479 switch (asize) {
480 case 16:
481 opx->offset = getu16(data);
482 data += 2;
483 if (segsize != 16)
484 opx->disp_size = 16;
485 break;
486 case 32:
487 opx->offset = getu32(data);
488 data += 4;
489 if (segsize == 16)
490 opx->disp_size = 32;
491 break;
492 case 64:
493 opx->offset = getu64(data);
494 opx->disp_size = 64;
495 data += 8;
496 break;
497 }
498 break;
499
500 case4(050):
501 opx->offset = gets8(data++);
502 opx->segment |= SEG_RELATIVE;
503 break;
504
505 case4(054):
506 opx->offset = getu64(data);
507 data += 8;
508 break;
509
510 case4(060):
511 opx->offset = gets16(data);
512 data += 2;
513 opx->segment |= SEG_RELATIVE;
514 opx->segment &= ~SEG_32BIT;
515 break;
516
517 case4(064):
518 opx->segment |= SEG_RELATIVE;
519 if (osize == 16) {
520 opx->offset = gets16(data);
521 data += 2;
522 opx->segment &= ~(SEG_32BIT|SEG_64BIT);
523 } else if (osize == 32) {
524 opx->offset = gets32(data);
525 data += 4;
526 opx->segment &= ~SEG_64BIT;
527 opx->segment |= SEG_32BIT;
528 }
529 if (segsize != osize) {
530 opx->type =
531 (opx->type & ~SIZE_MASK)
532 | ((osize == 16) ? BITS16 : BITS32);
533 }
534 break;
535
536 case4(070):
537 opx->offset = gets32(data);
538 data += 4;
539 opx->segment |= SEG_32BIT | SEG_RELATIVE;
540 break;
541
542 case4(0100):
543 case4(0110):
544 case4(0120):
545 case4(0130):
546 {
547 int modrm = *data++;
548 opx->segment |= SEG_RMREG;
549 data = do_ea(data, modrm, asize, segsize, opy, ins);
550 if (!data)
551 return false;
552 opx->basereg = ((modrm >> 3) & 7) + (ins->rex & REX_R ? 8 : 0);
553 break;
554 }
555
556 case4(0140):
557 if (s_field_for == op1) {
558 opx->offset = gets8(data);
559 data++;
560 } else {
561 opx->offset = getu16(data);
562 data += 2;
563 }
564 break;
565
566 case4(0144):
567 case4(0154):
568 s_field_for = (*data & 0x02) ? op1 : -1;
569 if ((*data++ & ~0x02) != *r++)
570 return false;
571 break;
572
573 case4(0150):
574 if (s_field_for == op1) {
575 opx->offset = gets8(data);
576 data++;
577 } else {
578 opx->offset = getu32(data);
579 data += 4;
580 }
581 break;
582
583 case4(0160):
584 ins->rex |= REX_D;
585 ins->drexdst = op1;
586 break;
587
588 case4(0164):
589 ins->rex |= REX_D|REX_OC;
590 ins->drexdst = op1;
591 break;
592
593 case 0171:
594 data = do_drex(data, ins);
595 if (!data)
596 return false;
597 break;
598
599 case 0172:
600 {
601 uint8_t ximm = *data++;
602 c = *r++;
603 ins->oprs[c >> 3].basereg = (ximm >> 4) & regmask;
604 ins->oprs[c >> 3].segment |= SEG_RMREG;
605 ins->oprs[c & 7].offset = ximm & 15;
606 }
607 break;
608
609 case 0173:
610 {
611 uint8_t ximm = *data++;
612 c = *r++;
613
614 if ((c ^ ximm) & 15)
615 return false;
616
617 ins->oprs[c >> 4].basereg = (ximm >> 4) & regmask;
618 ins->oprs[c >> 4].segment |= SEG_RMREG;
619 }
620 break;
621
622 case 0174:
623 {
624 uint8_t ximm = *data++;
625 c = *r++;
626
627 ins->oprs[c].basereg = (ximm >> 4) & regmask;
628 ins->oprs[c].segment |= SEG_RMREG;
629 }
630 break;
631
632 case4(0200):
633 case4(0204):
634 case4(0210):
635 case4(0214):
636 case4(0220):
637 case4(0224):
638 case4(0230):
639 case4(0234):
640 {
641 int modrm = *data++;
642 if (((modrm >> 3) & 07) != (c & 07))
643 return false; /* spare field doesn't match up */
644 data = do_ea(data, modrm, asize, segsize, opy, ins);
645 if (!data)
646 return false;
647 break;
648 }
649
650 case4(0260):
651 {
652 int vexm = *r++;
653 int vexwlp = *r++;
654 ins->rex |= REX_V;
655 if ((prefix->rex & (REX_V|REX_D|REX_P)) != REX_V)
656 return false;
657
658 if ((vexm & 0x1f) != prefix->vex_m)
659 return false;
660
661 switch (vexwlp & 030) {
662 case 000:
663 if (prefix->rex & REX_W)
664 return false;
665 break;
666 case 010:
667 if (!(prefix->rex & REX_W))
668 return false;
669 ins->rex &= ~REX_W;
670 break;
671 case 020: /* VEX.W is a don't care */
672 ins->rex &= ~REX_W;
673 break;
674 case 030:
675 break;
676 }
677
678 if ((vexwlp & 007) != prefix->vex_lp)
679 return false;
680
681 opx->segment |= SEG_RMREG;
682 opx->basereg = prefix->vex_v;
683 vex_ok = true;
684 break;
685 }
686
687 case 0270:
688 {
689 int vexm = *r++;
690 int vexwlp = *r++;
691 ins->rex |= REX_V;
692 if ((prefix->rex & (REX_V|REX_D|REX_P)) != REX_V)
693 return false;
694
695 if ((vexm & 0x1f) != prefix->vex_m)
696 return false;
697
698 switch (vexwlp & 030) {
699 case 000:
700 if (ins->rex & REX_W)
701 return false;
702 break;
703 case 010:
704 if (!(ins->rex & REX_W))
705 return false;
706 break;
707 default:
708 break; /* Need to do anything special here? */
709 }
710
711 if ((vexwlp & 007) != prefix->vex_lp)
712 return false;
713
714 if (prefix->vex_v != 0)
715 return false;
716
717 vex_ok = true;
718 break;
719 }
720
721 case 0310:
722 if (asize != 16)
723 return false;
724 else
725 a_used = true;
726 break;
727
728 case 0311:
729 if (asize != 32)
730 return false;
731 else
732 a_used = true;
733 break;
734
735 case 0312:
736 if (asize != segsize)
737 return false;
738 else
739 a_used = true;
740 break;
741
742 case 0313:
743 if (asize != 64)
744 return false;
745 else
746 a_used = true;
747 break;
748
749 case 0314:
750 if (prefix->rex & REX_B)
751 return false;
752 break;
753
754 case 0315:
755 if (prefix->rex & REX_X)
756 return false;
757 break;
758
759 case 0316:
760 if (prefix->rex & REX_R)
761 return false;
762 break;
763
764 case 0317:
765 if (prefix->rex & REX_W)
766 return false;
767 break;
768
769 case 0320:
770 if (osize != 16)
771 return false;
772 else
773 o_used = true;
774 break;
775
776 case 0321:
777 if (osize != 32)
778 return false;
779 else
780 o_used = true;
781 break;
782
783 case 0322:
784 if (osize != (segsize == 16) ? 16 : 32)
785 return false;
786 else
787 o_used = true;
788 break;
789
790 case 0323:
791 ins->rex |= REX_W; /* 64-bit only instruction */
792 osize = 64;
793 o_used = true;
794 break;
795
796 case 0324:
797 if (!(ins->rex & (REX_P|REX_W)) || osize != 64)
798 return false;
799 o_used = true;
800 break;
801
802 case 0325:
803 ins->rex |= REX_NH;
804 break;
805
806 case 0330:
807 {
808 int t = *r++, d = *data++;
809 if (d < t || d > t + 15)
810 return false;
811 else
812 ins->condition = d - t;
813 break;
814 }
815
816 case 0331:
817 if (prefix->rep)
818 return false;
819 break;
820
821 case 0332:
822 if (prefix->rep != 0xF2)
823 return false;
824 drep = 0;
825 break;
826
827 case 0333:
828 if (prefix->rep != 0xF3)
829 return false;
830 drep = 0;
831 break;
832
833 case 0334:
834 if (lock) {
835 ins->rex |= REX_R;
836 lock = 0;
837 }
838 break;
839
840 case 0335:
841 if (drep == P_REP)
842 drep = P_REPE;
843 break;
844
845 case 0336:
846 case 0337:
847 break;
848
849 case 0340:
850 return false;
851
852 case 0341:
853 if (prefix->wait != 0x9B)
854 return false;
855 dwait = 0;
856 break;
857
858 case4(0344):
859 ins->oprs[0].basereg = (*data++ >> 3) & 7;
860 break;
861
862 case 0360:
863 if (prefix->osp || prefix->rep)
864 return false;
865 break;
866
867 case 0361:
868 if (!prefix->osp || prefix->rep)
869 return false;
870 o_used = true;
871 break;
872
873 case 0362:
874 if (prefix->osp || prefix->rep != 0xf2)
875 return false;
876 drep = 0;
877 break;
878
879 case 0363:
880 if (prefix->osp || prefix->rep != 0xf3)
881 return false;
882 drep = 0;
883 break;
884
885 case 0364:
886 if (prefix->osp)
887 return false;
888 break;
889
890 case 0365:
891 if (prefix->asp)
892 return false;
893 break;
894
895 case 0366:
896 if (!prefix->osp)
897 return false;
898 o_used = true;
899 break;
900
901 case 0367:
902 if (!prefix->asp)
903 return false;
904 a_used = true;
905 break;
906
907 default:
908 return false; /* Unknown code */
909 }
910 }
911
912 if (!vex_ok && (ins->rex & REX_V))
913 return false;
914
915 /* REX cannot be combined with DREX or VEX */
916 if ((ins->rex & (REX_D|REX_V)) && (prefix->rex & REX_P))
917 return false;
918
919 /*
920 * Check for unused rep or a/o prefixes.
921 */
922 for (i = 0; i < t->operands; i++) {
923 if (ins->oprs[i].segment != SEG_RMREG)
924 a_used = true;
925 }
926
927 if (lock) {
928 if (ins->prefixes[PPS_LREP])
929 return false;
930 ins->prefixes[PPS_LREP] = P_LOCK;
931 }
932 if (drep) {
933 if (ins->prefixes[PPS_LREP])
934 return false;
935 ins->prefixes[PPS_LREP] = drep;
936 }
937 ins->prefixes[PPS_WAIT] = dwait;
938 if (!o_used) {
939 if (osize != ((segsize == 16) ? 16 : 32)) {
940 enum prefixes pfx = 0;
941
942 switch (osize) {
943 case 16:
944 pfx = P_O16;
945 break;
946 case 32:
947 pfx = P_O32;
948 break;
949 case 64:
950 pfx = P_O64;
951 break;
952 }
953
954 if (ins->prefixes[PPS_OSIZE])
955 return false;
956 ins->prefixes[PPS_OSIZE] = pfx;
957 }
958 }
959 if (!a_used && asize != segsize) {
960 if (ins->prefixes[PPS_ASIZE])
961 return false;
962 ins->prefixes[PPS_ASIZE] = asize == 16 ? P_A16 : P_A32;
963 }
964
965 /* Fix: check for redundant REX prefixes */
966
967 return data - origdata;
968 }
969
970 /* Condition names for disassembly, sorted by x86 code */
971 static const char * const condition_name[16] = {
972 "o", "no", "c", "nc", "z", "nz", "na", "a",
973 "s", "ns", "pe", "po", "l", "nl", "ng", "g"
974 };
975
976 int32_t disasm(uint8_t *data, char *output, int outbufsize, int segsize,
977 int32_t offset, int autosync, uint32_t prefer)
978 {
979 const struct itemplate * const *p, * const *best_p;
980 const struct disasm_index *ix;
981 uint8_t *dp;
982 int length, best_length = 0;
983 char *segover;
984 int i, slen, colon, n;
985 uint8_t *origdata;
986 int works;
987 insn tmp_ins, ins;
988 uint32_t goodness, best;
989 int best_pref;
990 struct prefix_info prefix;
991 bool end_prefix;
992
993 memset(&ins, 0, sizeof ins);
994
995 /*
996 * Scan for prefixes.
997 */
998 memset(&prefix, 0, sizeof prefix);
999 prefix.asize = segsize;
1000 prefix.osize = (segsize == 64) ? 32 : segsize;
1001 segover = NULL;
1002 origdata = data;
1003
1004 ix = itable;
1005
1006 end_prefix = false;
1007 while (!end_prefix) {
1008 switch (*data) {
1009 case 0xF2:
1010 case 0xF3:
1011 prefix.rep = *data++;
1012 break;
1013
1014 case 0x9B:
1015 prefix.wait = *data++;
1016 break;
1017
1018 case 0xF0:
1019 prefix.lock = *data++;
1020 break;
1021
1022 case 0x2E:
1023 segover = "cs", prefix.seg = *data++;
1024 break;
1025 case 0x36:
1026 segover = "ss", prefix.seg = *data++;
1027 break;
1028 case 0x3E:
1029 segover = "ds", prefix.seg = *data++;
1030 break;
1031 case 0x26:
1032 segover = "es", prefix.seg = *data++;
1033 break;
1034 case 0x64:
1035 segover = "fs", prefix.seg = *data++;
1036 break;
1037 case 0x65:
1038 segover = "gs", prefix.seg = *data++;
1039 break;
1040
1041 case 0x66:
1042 prefix.osize = (segsize == 16) ? 32 : 16;
1043 prefix.osp = *data++;
1044 break;
1045 case 0x67:
1046 prefix.asize = (segsize == 32) ? 16 : 32;
1047 prefix.asp = *data++;
1048 break;
1049
1050 case 0xC4:
1051 case 0xC5:
1052 if (segsize == 64 || (data[1] & 0xc0) == 0xc0) {
1053 prefix.vex[0] = *data++;
1054 prefix.vex[1] = *data++;
1055
1056 prefix.rex = REX_V;
1057 prefix.vex_c = RV_VEX;
1058
1059 if (prefix.vex[0] == 0xc4) {
1060 prefix.vex[2] = *data++;
1061 prefix.rex |= (~prefix.vex[1] >> 5) & 7; /* REX_RXB */
1062 prefix.rex |= (prefix.vex[2] >> (7-3)) & REX_W;
1063 prefix.vex_m = prefix.vex[1] & 0x1f;
1064 prefix.vex_v = (~prefix.vex[2] >> 3) & 15;
1065 prefix.vex_lp = prefix.vex[2] & 7;
1066 } else {
1067 prefix.rex |= (~prefix.vex[1] >> (7-2)) & REX_R;
1068 prefix.vex_m = 1;
1069 prefix.vex_v = (~prefix.vex[1] >> 3) & 15;
1070 prefix.vex_lp = prefix.vex[1] & 7;
1071 }
1072
1073 ix = itable_vex[RV_VEX][prefix.vex_m][prefix.vex_lp];
1074 }
1075 end_prefix = true;
1076 break;
1077
1078 case 0x8F:
1079 if ((data[1] & 030) != 0 &&
1080 (segsize == 64 || (data[1] & 0xc0) == 0xc0)) {
1081 prefix.vex[0] = *data++;
1082 prefix.vex[1] = *data++;
1083 prefix.vex[2] = *data++;
1084
1085 prefix.rex = REX_V;
1086 prefix.vex_c = RV_XOP;
1087
1088 prefix.rex |= (~prefix.vex[1] >> 5) & 7; /* REX_RXB */
1089 prefix.rex |= (prefix.vex[2] >> (7-3)) & REX_W;
1090 prefix.vex_m = prefix.vex[1] & 0x1f;
1091 prefix.vex_v = (~prefix.vex[2] >> 3) & 15;
1092 prefix.vex_lp = prefix.vex[2] & 7;
1093
1094 ix = itable_vex[RV_XOP][prefix.vex_m][prefix.vex_lp];
1095 }
1096 end_prefix = true;
1097 break;
1098
1099 case REX_P + 0x0:
1100 case REX_P + 0x1:
1101 case REX_P + 0x2:
1102 case REX_P + 0x3:
1103 case REX_P + 0x4:
1104 case REX_P + 0x5:
1105 case REX_P + 0x6:
1106 case REX_P + 0x7:
1107 case REX_P + 0x8:
1108 case REX_P + 0x9:
1109 case REX_P + 0xA:
1110 case REX_P + 0xB:
1111 case REX_P + 0xC:
1112 case REX_P + 0xD:
1113 case REX_P + 0xE:
1114 case REX_P + 0xF:
1115 if (segsize == 64) {
1116 prefix.rex = *data++;
1117 if (prefix.rex & REX_W)
1118 prefix.osize = 64;
1119 }
1120 end_prefix = true;
1121 break;
1122
1123 default:
1124 end_prefix = true;
1125 break;
1126 }
1127 }
1128
1129 best = -1; /* Worst possible */
1130 best_p = NULL;
1131 best_pref = INT_MAX;
1132
1133 if (!ix)
1134 return 0; /* No instruction table at all... */
1135
1136 dp = data;
1137 ix += *dp++;
1138 while (ix->n == -1) {
1139 ix = (const struct disasm_index *)ix->p + *dp++;
1140 }
1141
1142 p = (const struct itemplate * const *)ix->p;
1143 for (n = ix->n; n; n--, p++) {
1144 if ((length = matches(*p, data, &prefix, segsize, &tmp_ins))) {
1145 works = true;
1146 /*
1147 * Final check to make sure the types of r/m match up.
1148 * XXX: Need to make sure this is actually correct.
1149 */
1150 for (i = 0; i < (*p)->operands; i++) {
1151 if (!((*p)->opd[i] & SAME_AS) &&
1152 (
1153 /* If it's a mem-only EA but we have a
1154 register, die. */
1155 ((tmp_ins.oprs[i].segment & SEG_RMREG) &&
1156 !(MEMORY & ~(*p)->opd[i])) ||
1157 /* If it's a reg-only EA but we have a memory
1158 ref, die. */
1159 (!(tmp_ins.oprs[i].segment & SEG_RMREG) &&
1160 !(REG_EA & ~(*p)->opd[i]) &&
1161 !((*p)->opd[i] & REG_SMASK)) ||
1162 /* Register type mismatch (eg FS vs REG_DESS):
1163 die. */
1164 ((((*p)->opd[i] & (REGISTER | FPUREG)) ||
1165 (tmp_ins.oprs[i].segment & SEG_RMREG)) &&
1166 !whichreg((*p)->opd[i],
1167 tmp_ins.oprs[i].basereg, tmp_ins.rex))
1168 )) {
1169 works = false;
1170 break;
1171 }
1172 }
1173
1174 /*
1175 * Note: we always prefer instructions which incorporate
1176 * prefixes in the instructions themselves. This is to allow
1177 * e.g. PAUSE to be preferred to REP NOP, and deal with
1178 * MMX/SSE instructions where prefixes are used to select
1179 * between MMX and SSE register sets or outright opcode
1180 * selection.
1181 */
1182 if (works) {
1183 int i, nprefix;
1184 goodness = ((*p)->flags & IF_PFMASK) ^ prefer;
1185 nprefix = 0;
1186 for (i = 0; i < MAXPREFIX; i++)
1187 if (tmp_ins.prefixes[i])
1188 nprefix++;
1189 if (nprefix < best_pref ||
1190 (nprefix == best_pref && goodness < best)) {
1191 /* This is the best one found so far */
1192 best = goodness;
1193 best_p = p;
1194 best_pref = nprefix;
1195 best_length = length;
1196 ins = tmp_ins;
1197 }
1198 }
1199 }
1200 }
1201
1202 if (!best_p)
1203 return 0; /* no instruction was matched */
1204
1205 /* Pick the best match */
1206 p = best_p;
1207 length = best_length;
1208
1209 slen = 0;
1210
1211 /* TODO: snprintf returns the value that the string would have if
1212 * the buffer were long enough, and not the actual length of
1213 * the returned string, so each instance of using the return
1214 * value of snprintf should actually be checked to assure that
1215 * the return value is "sane." Maybe a macro wrapper could
1216 * be used for that purpose.
1217 */
1218 for (i = 0; i < MAXPREFIX; i++) {
1219 const char *prefix = prefix_name(ins.prefixes[i]);
1220 if (prefix)
1221 slen += snprintf(output+slen, outbufsize-slen, "%s ", prefix);
1222 }
1223
1224 i = (*p)->opcode;
1225 if (i >= FIRST_COND_OPCODE)
1226 slen += snprintf(output + slen, outbufsize - slen, "%s%s",
1227 nasm_insn_names[i], condition_name[ins.condition]);
1228 else
1229 slen += snprintf(output + slen, outbufsize - slen, "%s",
1230 nasm_insn_names[i]);
1231
1232 colon = false;
1233 length += data - origdata; /* fix up for prefixes */
1234 for (i = 0; i < (*p)->operands; i++) {
1235 opflags_t t = (*p)->opd[i];
1236 const operand *o = &ins.oprs[i];
1237 int64_t offs;
1238
1239 if (t & SAME_AS) {
1240 o = &ins.oprs[t & ~SAME_AS];
1241 t = (*p)->opd[t & ~SAME_AS];
1242 }
1243
1244 output[slen++] = (colon ? ':' : i == 0 ? ' ' : ',');
1245
1246 offs = o->offset;
1247 if (o->segment & SEG_RELATIVE) {
1248 offs += offset + length;
1249 /*
1250 * sort out wraparound
1251 */
1252 if (!(o->segment & (SEG_32BIT|SEG_64BIT)))
1253 offs &= 0xffff;
1254 else if (segsize != 64)
1255 offs &= 0xffffffff;
1256
1257 /*
1258 * add sync marker, if autosync is on
1259 */
1260 if (autosync)
1261 add_sync(offs, 0L);
1262 }
1263
1264 if (t & COLON)
1265 colon = true;
1266 else
1267 colon = false;
1268
1269 if ((t & (REGISTER | FPUREG)) ||
1270 (o->segment & SEG_RMREG)) {
1271 enum reg_enum reg;
1272 reg = whichreg(t, o->basereg, ins.rex);
1273 if (t & TO)
1274 slen += snprintf(output + slen, outbufsize - slen, "to ");
1275 slen += snprintf(output + slen, outbufsize - slen, "%s",
1276 nasm_reg_names[reg-EXPR_REG_START]);
1277 } else if (!(UNITY & ~t)) {
1278 output[slen++] = '1';
1279 } else if (t & IMMEDIATE) {
1280 if (t & BITS8) {
1281 slen +=
1282 snprintf(output + slen, outbufsize - slen, "byte ");
1283 if (o->segment & SEG_SIGNED) {
1284 if (offs < 0) {
1285 offs *= -1;
1286 output[slen++] = '-';
1287 } else
1288 output[slen++] = '+';
1289 }
1290 } else if (t & BITS16) {
1291 slen +=
1292 snprintf(output + slen, outbufsize - slen, "word ");
1293 } else if (t & BITS32) {
1294 slen +=
1295 snprintf(output + slen, outbufsize - slen, "dword ");
1296 } else if (t & BITS64) {
1297 slen +=
1298 snprintf(output + slen, outbufsize - slen, "qword ");
1299 } else if (t & NEAR) {
1300 slen +=
1301 snprintf(output + slen, outbufsize - slen, "near ");
1302 } else if (t & SHORT) {
1303 slen +=
1304 snprintf(output + slen, outbufsize - slen, "short ");
1305 }
1306 slen +=
1307 snprintf(output + slen, outbufsize - slen, "0x%"PRIx64"",
1308 offs);
1309 } else if (!(MEM_OFFS & ~t)) {
1310 slen +=
1311 snprintf(output + slen, outbufsize - slen,
1312 "[%s%s%s0x%"PRIx64"]",
1313 (segover ? segover : ""),
1314 (segover ? ":" : ""),
1315 (o->disp_size == 64 ? "qword " :
1316 o->disp_size == 32 ? "dword " :
1317 o->disp_size == 16 ? "word " : ""), offs);
1318 segover = NULL;
1319 } else if (!(REGMEM & ~t)) {
1320 int started = false;
1321 if (t & BITS8)
1322 slen +=
1323 snprintf(output + slen, outbufsize - slen, "byte ");
1324 if (t & BITS16)
1325 slen +=
1326 snprintf(output + slen, outbufsize - slen, "word ");
1327 if (t & BITS32)
1328 slen +=
1329 snprintf(output + slen, outbufsize - slen, "dword ");
1330 if (t & BITS64)
1331 slen +=
1332 snprintf(output + slen, outbufsize - slen, "qword ");
1333 if (t & BITS80)
1334 slen +=
1335 snprintf(output + slen, outbufsize - slen, "tword ");
1336 if (t & BITS128)
1337 slen +=
1338 snprintf(output + slen, outbufsize - slen, "oword ");
1339 if (t & BITS256)
1340 slen +=
1341 snprintf(output + slen, outbufsize - slen, "yword ");
1342 if (t & FAR)
1343 slen += snprintf(output + slen, outbufsize - slen, "far ");
1344 if (t & NEAR)
1345 slen +=
1346 snprintf(output + slen, outbufsize - slen, "near ");
1347 output[slen++] = '[';
1348 if (o->disp_size)
1349 slen += snprintf(output + slen, outbufsize - slen, "%s",
1350 (o->disp_size == 64 ? "qword " :
1351 o->disp_size == 32 ? "dword " :
1352 o->disp_size == 16 ? "word " :
1353 ""));
1354 if (o->eaflags & EAF_REL)
1355 slen += snprintf(output + slen, outbufsize - slen, "rel ");
1356 if (segover) {
1357 slen +=
1358 snprintf(output + slen, outbufsize - slen, "%s:",
1359 segover);
1360 segover = NULL;
1361 }
1362 if (o->basereg != -1) {
1363 slen += snprintf(output + slen, outbufsize - slen, "%s",
1364 nasm_reg_names[(o->basereg-EXPR_REG_START)]);
1365 started = true;
1366 }
1367 if (o->indexreg != -1) {
1368 if (started)
1369 output[slen++] = '+';
1370 slen += snprintf(output + slen, outbufsize - slen, "%s",
1371 nasm_reg_names[(o->indexreg-EXPR_REG_START)]);
1372 if (o->scale > 1)
1373 slen +=
1374 snprintf(output + slen, outbufsize - slen, "*%d",
1375 o->scale);
1376 started = true;
1377 }
1378
1379
1380 if (o->segment & SEG_DISP8) {
1381 const char *prefix;
1382 uint8_t offset = offs;
1383 if ((int8_t)offset < 0) {
1384 prefix = "-";
1385 offset = -offset;
1386 } else {
1387 prefix = "+";
1388 }
1389 slen +=
1390 snprintf(output + slen, outbufsize - slen, "%s0x%"PRIx8"",
1391 prefix, offset);
1392 } else if (o->segment & SEG_DISP16) {
1393 const char *prefix;
1394 uint16_t offset = offs;
1395 if ((int16_t)offset < 0 && started) {
1396 offset = -offset;
1397 prefix = "-";
1398 } else {
1399 prefix = started ? "+" : "";
1400 }
1401 slen +=
1402 snprintf(output + slen, outbufsize - slen,
1403 "%s0x%"PRIx16"", prefix, offset);
1404 } else if (o->segment & SEG_DISP32) {
1405 if (prefix.asize == 64) {
1406 const char *prefix;
1407 uint64_t offset = (int64_t)(int32_t)offs;
1408 if ((int32_t)offs < 0 && started) {
1409 offset = -offset;
1410 prefix = "-";
1411 } else {
1412 prefix = started ? "+" : "";
1413 }
1414 slen +=
1415 snprintf(output + slen, outbufsize - slen,
1416 "%s0x%"PRIx64"", prefix, offset);
1417 } else {
1418 const char *prefix;
1419 uint32_t offset = offs;
1420 if ((int32_t) offset < 0 && started) {
1421 offset = -offset;
1422 prefix = "-";
1423 } else {
1424 prefix = started ? "+" : "";
1425 }
1426 slen +=
1427 snprintf(output + slen, outbufsize - slen,
1428 "%s0x%"PRIx32"", prefix, offset);
1429 }
1430 }
1431 output[slen++] = ']';
1432 } else {
1433 slen +=
1434 snprintf(output + slen, outbufsize - slen, "<operand%d>",
1435 i);
1436 }
1437 }
1438 output[slen] = '\0';
1439 if (segover) { /* unused segment override */
1440 char *p = output;
1441 int count = slen + 1;
1442 while (count--)
1443 p[count + 3] = p[count];
1444 strncpy(output, segover, 2);
1445 output[2] = ' ';
1446 }
1447 return length;
1448 }
1449
1450 /*
1451 * This is called when we don't have a complete instruction. If it
1452 * is a standalone *single-byte* prefix show it as such, otherwise
1453 * print it as a literal.
1454 */
1455 int32_t eatbyte(uint8_t *data, char *output, int outbufsize, int segsize)
1456 {
1457 uint8_t byte = *data;
1458 const char *str = NULL;
1459
1460 switch (byte) {
1461 case 0xF2:
1462 str = "repne";
1463 break;
1464 case 0xF3:
1465 str = "rep";
1466 break;
1467 case 0x9B:
1468 str = "wait";
1469 break;
1470 case 0xF0:
1471 str = "lock";
1472 break;
1473 case 0x2E:
1474 str = "cs";
1475 break;
1476 case 0x36:
1477 str = "ss";
1478 break;
1479 case 0x3E:
1480 str = "ss";
1481 break;
1482 case 0x26:
1483 str = "es";
1484 break;
1485 case 0x64:
1486 str = "fs";
1487 break;
1488 case 0x65:
1489 str = "gs";
1490 break;
1491 case 0x66:
1492 str = (segsize == 16) ? "o32" : "o16";
1493 break;
1494 case 0x67:
1495 str = (segsize == 32) ? "a16" : "a32";
1496 break;
1497 case REX_P + 0x0:
1498 case REX_P + 0x1:
1499 case REX_P + 0x2:
1500 case REX_P + 0x3:
1501 case REX_P + 0x4:
1502 case REX_P + 0x5:
1503 case REX_P + 0x6:
1504 case REX_P + 0x7:
1505 case REX_P + 0x8:
1506 case REX_P + 0x9:
1507 case REX_P + 0xA:
1508 case REX_P + 0xB:
1509 case REX_P + 0xC:
1510 case REX_P + 0xD:
1511 case REX_P + 0xE:
1512 case REX_P + 0xF:
1513 if (segsize == 64) {
1514 snprintf(output, outbufsize, "rex%s%s%s%s%s",
1515 (byte == REX_P) ? "" : ".",
1516 (byte & REX_W) ? "w" : "",
1517 (byte & REX_R) ? "r" : "",
1518 (byte & REX_X) ? "x" : "",
1519 (byte & REX_B) ? "b" : "");
1520 break;
1521 }
1522 /* else fall through */
1523 default:
1524 snprintf(output, outbufsize, "db 0x%02x", byte);
1525 break;
1526 }
1527
1528 if (str)
1529 strcpy(output, str);
1530
1531 return 1;
1532 }
The Netwide Assembler