?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
preproc: Add tokenization tracing
[nasm.git] / disasm.c
blob5943966601c10d77d580daee563a89509e43fe2b
1 /* ----------------------------------------------------------------------- *
2 *
3 * Copyright 1996-2010 The NASM Authors - All Rights Reserved
4 * See the file AUTHORS included with the NASM distribution for
5 * the specific copyright holders.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following
9 * conditions are met:
10 *
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above
14 * copyright notice, this list of conditions and the following
15 * disclaimer in the documentation and/or other materials provided
16 * with the distribution.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND
19 * CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES,
20 * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
21 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
22 * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
23 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
24 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
25 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
26 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
29 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
30 * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
31 *
32 * ----------------------------------------------------------------------- */
33
34 /*
35 * disasm.c where all the _work_ gets done in the Netwide Disassembler
36 */
37
38 #include "compiler.h"
39
40 #include <stdio.h>
41 #include <string.h>
42 #include <limits.h>
43 #include <inttypes.h>
44
45 #include "nasm.h"
46 #include "disasm.h"
47 #include "sync.h"
48 #include "insns.h"
49 #include "tables.h"
50 #include "regdis.h"
51
52 /*
53 * Flags that go into the `segment' field of `insn' structures
54 * during disassembly.
55 */
56 #define SEG_RELATIVE 1
57 #define SEG_32BIT 2
58 #define SEG_RMREG 4
59 #define SEG_DISP8 8
60 #define SEG_DISP16 16
61 #define SEG_DISP32 32
62 #define SEG_NODISP 64
63 #define SEG_SIGNED 128
64 #define SEG_64BIT 256
65
66 /*
67 * Prefix information
68 */
69 struct prefix_info {
70 uint8_t osize; /* Operand size */
71 uint8_t asize; /* Address size */
72 uint8_t osp; /* Operand size prefix present */
73 uint8_t asp; /* Address size prefix present */
74 uint8_t rep; /* Rep prefix present */
75 uint8_t seg; /* Segment override prefix present */
76 uint8_t wait; /* WAIT "prefix" present */
77 uint8_t lock; /* Lock prefix present */
78 uint8_t vex[3]; /* VEX prefix present */
79 uint8_t vex_c; /* VEX "class" (VEX, XOP, ...) */
80 uint8_t vex_m; /* VEX.M field */
81 uint8_t vex_v;
82 uint8_t vex_lp; /* VEX.LP fields */
83 uint32_t rex; /* REX prefix present */
84 };
85
86 #define getu8(x) (*(uint8_t *)(x))
87 #if X86_MEMORY
88 /* Littleendian CPU which can handle unaligned references */
89 #define getu16(x) (*(uint16_t *)(x))
90 #define getu32(x) (*(uint32_t *)(x))
91 #define getu64(x) (*(uint64_t *)(x))
92 #else
93 static uint16_t getu16(uint8_t *data)
94 {
95 return (uint16_t)data[0] + ((uint16_t)data[1] << 8);
96 }
97 static uint32_t getu32(uint8_t *data)
98 {
99 return (uint32_t)getu16(data) + ((uint32_t)getu16(data+2) << 16);
100 }
101 static uint64_t getu64(uint8_t *data)
102 {
103 return (uint64_t)getu32(data) + ((uint64_t)getu32(data+4) << 32);
104 }
105 #endif
106
107 #define gets8(x) ((int8_t)getu8(x))
108 #define gets16(x) ((int16_t)getu16(x))
109 #define gets32(x) ((int32_t)getu32(x))
110 #define gets64(x) ((int64_t)getu64(x))
111
112 /* Important: regval must already have been adjusted for rex extensions */
113 static enum reg_enum whichreg(opflags_t regflags, int regval, int rex)
114 {
115 if (!(regflags & (REGISTER|REGMEM)))
116 return 0; /* Registers not permissible?! */
117
118 regflags |= REGISTER;
119
120 if (!(REG_AL & ~regflags))
121 return R_AL;
122 if (!(REG_AX & ~regflags))
123 return R_AX;
124 if (!(REG_EAX & ~regflags))
125 return R_EAX;
126 if (!(REG_RAX & ~regflags))
127 return R_RAX;
128 if (!(REG_DL & ~regflags))
129 return R_DL;
130 if (!(REG_DX & ~regflags))
131 return R_DX;
132 if (!(REG_EDX & ~regflags))
133 return R_EDX;
134 if (!(REG_RDX & ~regflags))
135 return R_RDX;
136 if (!(REG_CL & ~regflags))
137 return R_CL;
138 if (!(REG_CX & ~regflags))
139 return R_CX;
140 if (!(REG_ECX & ~regflags))
141 return R_ECX;
142 if (!(REG_RCX & ~regflags))
143 return R_RCX;
144 if (!(FPU0 & ~regflags))
145 return R_ST0;
146 if (!(XMM0 & ~regflags))
147 return R_XMM0;
148 if (!(YMM0 & ~regflags))
149 return R_YMM0;
150 if (!(REG_CS & ~regflags))
151 return (regval == 1) ? R_CS : 0;
152 if (!(REG_DESS & ~regflags))
153 return (regval == 0 || regval == 2
154 || regval == 3 ? nasm_rd_sreg[regval] : 0);
155 if (!(REG_FSGS & ~regflags))
156 return (regval == 4 || regval == 5 ? nasm_rd_sreg[regval] : 0);
157 if (!(REG_SEG67 & ~regflags))
158 return (regval == 6 || regval == 7 ? nasm_rd_sreg[regval] : 0);
159
160 /* All the entries below look up regval in an 16-entry array */
161 if (regval < 0 || regval > 15)
162 return 0;
163
164 if (!(REG8 & ~regflags)) {
165 if (rex & (REX_P|REX_NH))
166 return nasm_rd_reg8_rex[regval];
167 else
168 return nasm_rd_reg8[regval];
169 }
170 if (!(REG16 & ~regflags))
171 return nasm_rd_reg16[regval];
172 if (!(REG32 & ~regflags))
173 return nasm_rd_reg32[regval];
174 if (!(REG64 & ~regflags))
175 return nasm_rd_reg64[regval];
176 if (!(REG_SREG & ~regflags))
177 return nasm_rd_sreg[regval & 7]; /* Ignore REX */
178 if (!(REG_CREG & ~regflags))
179 return nasm_rd_creg[regval];
180 if (!(REG_DREG & ~regflags))
181 return nasm_rd_dreg[regval];
182 if (!(REG_TREG & ~regflags)) {
183 if (regval > 7)
184 return 0; /* TR registers are ill-defined with rex */
185 return nasm_rd_treg[regval];
186 }
187 if (!(FPUREG & ~regflags))
188 return nasm_rd_fpureg[regval & 7]; /* Ignore REX */
189 if (!(MMXREG & ~regflags))
190 return nasm_rd_mmxreg[regval & 7]; /* Ignore REX */
191 if (!(XMMREG & ~regflags))
192 return nasm_rd_xmmreg[regval];
193 if (!(YMMREG & ~regflags))
194 return nasm_rd_ymmreg[regval];
195
196 return 0;
197 }
198
199 /*
200 * Process a DREX suffix
201 */
202 static uint8_t *do_drex(uint8_t *data, insn *ins)
203 {
204 uint8_t drex = *data++;
205 operand *dst = &ins->oprs[ins->drexdst];
206
207 if ((drex & 8) != ((ins->rex & REX_OC) ? 8 : 0))
208 return NULL; /* OC0 mismatch */
209 ins->rex = (ins->rex & ~7) | (drex & 7);
210
211 dst->segment = SEG_RMREG;
212 dst->basereg = drex >> 4;
213 return data;
214 }
215
216
217 /*
218 * Process an effective address (ModRM) specification.
219 */
220 static uint8_t *do_ea(uint8_t *data, int modrm, int asize,
221 int segsize, enum ea_type type,
222 operand *op, insn *ins)
223 {
224 int mod, rm, scale, index, base;
225 int rex;
226 uint8_t sib = 0;
227
228 mod = (modrm >> 6) & 03;
229 rm = modrm & 07;
230
231 if (mod != 3 && asize != 16 && rm == 4)
232 sib = *data++;
233
234 if (ins->rex & REX_D) {
235 data = do_drex(data, ins);
236 if (!data)
237 return NULL;
238 }
239 rex = ins->rex;
240
241 if (mod == 3) { /* pure register version */
242 op->basereg = rm+(rex & REX_B ? 8 : 0);
243 op->segment |= SEG_RMREG;
244 return data;
245 }
246
247 op->disp_size = 0;
248 op->eaflags = 0;
249
250 if (asize == 16) {
251 /*
252 * <mod> specifies the displacement size (none, byte or
253 * word), and <rm> specifies the register combination.
254 * Exception: mod=0,rm=6 does not specify [BP] as one might
255 * expect, but instead specifies [disp16].
256 */
257
258 if (type != EA_SCALAR)
259 return NULL;
260
261 op->indexreg = op->basereg = -1;
262 op->scale = 1; /* always, in 16 bits */
263 switch (rm) {
264 case 0:
265 op->basereg = R_BX;
266 op->indexreg = R_SI;
267 break;
268 case 1:
269 op->basereg = R_BX;
270 op->indexreg = R_DI;
271 break;
272 case 2:
273 op->basereg = R_BP;
274 op->indexreg = R_SI;
275 break;
276 case 3:
277 op->basereg = R_BP;
278 op->indexreg = R_DI;
279 break;
280 case 4:
281 op->basereg = R_SI;
282 break;
283 case 5:
284 op->basereg = R_DI;
285 break;
286 case 6:
287 op->basereg = R_BP;
288 break;
289 case 7:
290 op->basereg = R_BX;
291 break;
292 }
293 if (rm == 6 && mod == 0) { /* special case */
294 op->basereg = -1;
295 if (segsize != 16)
296 op->disp_size = 16;
297 mod = 2; /* fake disp16 */
298 }
299 switch (mod) {
300 case 0:
301 op->segment |= SEG_NODISP;
302 break;
303 case 1:
304 op->segment |= SEG_DISP8;
305 op->offset = (int8_t)*data++;
306 break;
307 case 2:
308 op->segment |= SEG_DISP16;
309 op->offset = *data++;
310 op->offset |= ((unsigned)*data++) << 8;
311 break;
312 }
313 return data;
314 } else {
315 /*
316 * Once again, <mod> specifies displacement size (this time
317 * none, byte or *dword*), while <rm> specifies the base
318 * register. Again, [EBP] is missing, replaced by a pure
319 * disp32 (this time that's mod=0,rm=*5*) in 32-bit mode,
320 * and RIP-relative addressing in 64-bit mode.
321 *
322 * However, rm=4
323 * indicates not a single base register, but instead the
324 * presence of a SIB byte...
325 */
326 int a64 = asize == 64;
327
328 op->indexreg = -1;
329
330 if (a64)
331 op->basereg = nasm_rd_reg64[rm | ((rex & REX_B) ? 8 : 0)];
332 else
333 op->basereg = nasm_rd_reg32[rm | ((rex & REX_B) ? 8 : 0)];
334
335 if (rm == 5 && mod == 0) {
336 if (segsize == 64) {
337 op->eaflags |= EAF_REL;
338 op->segment |= SEG_RELATIVE;
339 mod = 2; /* fake disp32 */
340 }
341
342 if (asize != 64)
343 op->disp_size = asize;
344
345 op->basereg = -1;
346 mod = 2; /* fake disp32 */
347 }
348
349
350 if (rm == 4) { /* process SIB */
351 scale = (sib >> 6) & 03;
352 index = (sib >> 3) & 07;
353 base = sib & 07;
354
355 op->scale = 1 << scale;
356
357 if (index == 4 && !(rex & REX_X))
358 op->indexreg = -1; /* ESP/RSP cannot be an index */
359 else if (type == EA_XMMVSIB)
360 op->indexreg = nasm_rd_xmmreg[index | ((rex & REX_X) ? 8 : 0)];
361 else if (type == EA_YMMVSIB)
362 op->indexreg = nasm_rd_ymmreg[index | ((rex & REX_X) ? 8 : 0)];
363 else if (a64)
364 op->indexreg = nasm_rd_reg64[index | ((rex & REX_X) ? 8 : 0)];
365 else
366 op->indexreg = nasm_rd_reg32[index | ((rex & REX_X) ? 8 : 0)];
367
368 if (base == 5 && mod == 0) {
369 op->basereg = -1;
370 mod = 2; /* Fake disp32 */
371 } else if (a64)
372 op->basereg = nasm_rd_reg64[base | ((rex & REX_B) ? 8 : 0)];
373 else
374 op->basereg = nasm_rd_reg32[base | ((rex & REX_B) ? 8 : 0)];
375
376 if (segsize == 16)
377 op->disp_size = 32;
378 } else if (type != EA_SCALAR) {
379 /* Can't have VSIB without SIB */
380 return NULL;
381 }
382
383 switch (mod) {
384 case 0:
385 op->segment |= SEG_NODISP;
386 break;
387 case 1:
388 op->segment |= SEG_DISP8;
389 op->offset = gets8(data);
390 data++;
391 break;
392 case 2:
393 op->segment |= SEG_DISP32;
394 op->offset = gets32(data);
395 data += 4;
396 break;
397 }
398 return data;
399 }
400 }
401
402 /*
403 * Determine whether the instruction template in t corresponds to the data
404 * stream in data. Return the number of bytes matched if so.
405 */
406 #define case4(x) case (x): case (x)+1: case (x)+2: case (x)+3
407
408 static int matches(const struct itemplate *t, uint8_t *data,
409 const struct prefix_info *prefix, int segsize, insn *ins)
410 {
411 uint8_t *r = (uint8_t *)(t->code);
412 uint8_t *origdata = data;
413 bool a_used = false, o_used = false;
414 enum prefixes drep = 0;
415 enum prefixes dwait = 0;
416 uint8_t lock = prefix->lock;
417 int osize = prefix->osize;
418 int asize = prefix->asize;
419 int i, c;
420 int op1, op2;
421 struct operand *opx, *opy;
422 uint8_t opex = 0;
423 int s_field_for = -1; /* No 144/154 series code encountered */
424 bool vex_ok = false;
425 int regmask = (segsize == 64) ? 15 : 7;
426 enum ea_type eat = EA_SCALAR;
427
428 for (i = 0; i < MAX_OPERANDS; i++) {
429 ins->oprs[i].segment = ins->oprs[i].disp_size =
430 (segsize == 64 ? SEG_64BIT : segsize == 32 ? SEG_32BIT : 0);
431 }
432 ins->condition = -1;
433 ins->rex = prefix->rex;
434 memset(ins->prefixes, 0, sizeof ins->prefixes);
435
436 if (t->flags & (segsize == 64 ? IF_NOLONG : IF_LONG))
437 return false;
438
439 if (prefix->rep == 0xF2)
440 drep = P_REPNE;
441 else if (prefix->rep == 0xF3)
442 drep = P_REP;
443
444 dwait = prefix->wait ? P_WAIT : 0;
445
446 while ((c = *r++) != 0) {
447 op1 = (c & 3) + ((opex & 1) << 2);
448 op2 = ((c >> 3) & 3) + ((opex & 2) << 1);
449 opx = &ins->oprs[op1];
450 opy = &ins->oprs[op2];
451 opex = 0;
452
453 switch (c) {
454 case 01:
455 case 02:
456 case 03:
457 case 04:
458 while (c--)
459 if (*r++ != *data++)
460 return false;
461 break;
462
463 case 05:
464 case 06:
465 case 07:
466 opex = c;
467 break;
468
469 case4(010):
470 {
471 int t = *r++, d = *data++;
472 if (d < t || d > t + 7)
473 return false;
474 else {
475 opx->basereg = (d-t)+
476 (ins->rex & REX_B ? 8 : 0);
477 opx->segment |= SEG_RMREG;
478 }
479 break;
480 }
481
482 case4(014):
483 case4(0274):
484 opx->offset = (int8_t)*data++;
485 opx->segment |= SEG_SIGNED;
486 break;
487
488 case4(020):
489 opx->offset = *data++;
490 break;
491
492 case4(024):
493 opx->offset = *data++;
494 break;
495
496 case4(030):
497 opx->offset = getu16(data);
498 data += 2;
499 break;
500
501 case4(034):
502 if (osize == 32) {
503 opx->offset = getu32(data);
504 data += 4;
505 } else {
506 opx->offset = getu16(data);
507 data += 2;
508 }
509 if (segsize != asize)
510 opx->disp_size = asize;
511 break;
512
513 case4(040):
514 case4(0254):
515 opx->offset = getu32(data);
516 data += 4;
517 break;
518
519 case4(044):
520 switch (asize) {
521 case 16:
522 opx->offset = getu16(data);
523 data += 2;
524 if (segsize != 16)
525 opx->disp_size = 16;
526 break;
527 case 32:
528 opx->offset = getu32(data);
529 data += 4;
530 if (segsize == 16)
531 opx->disp_size = 32;
532 break;
533 case 64:
534 opx->offset = getu64(data);
535 opx->disp_size = 64;
536 data += 8;
537 break;
538 }
539 break;
540
541 case4(050):
542 opx->offset = gets8(data++);
543 opx->segment |= SEG_RELATIVE;
544 break;
545
546 case4(054):
547 opx->offset = getu64(data);
548 data += 8;
549 break;
550
551 case4(060):
552 opx->offset = gets16(data);
553 data += 2;
554 opx->segment |= SEG_RELATIVE;
555 opx->segment &= ~SEG_32BIT;
556 break;
557
558 case4(064):
559 opx->segment |= SEG_RELATIVE;
560 if (osize == 16) {
561 opx->offset = gets16(data);
562 data += 2;
563 opx->segment &= ~(SEG_32BIT|SEG_64BIT);
564 } else if (osize == 32) {
565 opx->offset = gets32(data);
566 data += 4;
567 opx->segment &= ~SEG_64BIT;
568 opx->segment |= SEG_32BIT;
569 }
570 if (segsize != osize) {
571 opx->type =
572 (opx->type & ~SIZE_MASK)
573 | ((osize == 16) ? BITS16 : BITS32);
574 }
575 break;
576
577 case4(070):
578 opx->offset = gets32(data);
579 data += 4;
580 opx->segment |= SEG_32BIT | SEG_RELATIVE;
581 break;
582
583 case4(0100):
584 case4(0110):
585 case4(0120):
586 case4(0130):
587 {
588 int modrm = *data++;
589 opx->segment |= SEG_RMREG;
590 data = do_ea(data, modrm, asize, segsize, eat, opy, ins);
591 if (!data)
592 return false;
593 opx->basereg = ((modrm >> 3) & 7) + (ins->rex & REX_R ? 8 : 0);
594 break;
595 }
596
597 case4(0140):
598 if (s_field_for == op1) {
599 opx->offset = gets8(data);
600 data++;
601 } else {
602 opx->offset = getu16(data);
603 data += 2;
604 }
605 break;
606
607 case4(0144):
608 case4(0154):
609 s_field_for = (*data & 0x02) ? op1 : -1;
610 if ((*data++ & ~0x02) != *r++)
611 return false;
612 break;
613
614 case4(0150):
615 if (s_field_for == op1) {
616 opx->offset = gets8(data);
617 data++;
618 } else {
619 opx->offset = getu32(data);
620 data += 4;
621 }
622 break;
623
624 case4(0160):
625 ins->rex |= REX_D;
626 ins->drexdst = op1;
627 break;
628
629 case4(0164):
630 ins->rex |= REX_D|REX_OC;
631 ins->drexdst = op1;
632 break;
633
634 case 0171:
635 data = do_drex(data, ins);
636 if (!data)
637 return false;
638 break;
639
640 case 0172:
641 {
642 uint8_t ximm = *data++;
643 c = *r++;
644 ins->oprs[c >> 3].basereg = (ximm >> 4) & regmask;
645 ins->oprs[c >> 3].segment |= SEG_RMREG;
646 ins->oprs[c & 7].offset = ximm & 15;
647 }
648 break;
649
650 case 0173:
651 {
652 uint8_t ximm = *data++;
653 c = *r++;
654
655 if ((c ^ ximm) & 15)
656 return false;
657
658 ins->oprs[c >> 4].basereg = (ximm >> 4) & regmask;
659 ins->oprs[c >> 4].segment |= SEG_RMREG;
660 }
661 break;
662
663 case 0174:
664 {
665 uint8_t ximm = *data++;
666 c = *r++;
667
668 ins->oprs[c].basereg = (ximm >> 4) & regmask;
669 ins->oprs[c].segment |= SEG_RMREG;
670 }
671 break;
672
673 case4(0200):
674 case4(0204):
675 case4(0210):
676 case4(0214):
677 case4(0220):
678 case4(0224):
679 case4(0230):
680 case4(0234):
681 {
682 int modrm = *data++;
683 if (((modrm >> 3) & 07) != (c & 07))
684 return false; /* spare field doesn't match up */
685 data = do_ea(data, modrm, asize, segsize, eat, opy, ins);
686 if (!data)
687 return false;
688 break;
689 }
690
691 case4(0250):
692 if (s_field_for == op1) {
693 opx->offset = gets8(data);
694 data++;
695 } else {
696 opx->offset = gets32(data);
697 data += 4;
698 }
699 break;
700
701 case4(0260):
702 case 0270:
703 {
704 int vexm = *r++;
705 int vexwlp = *r++;
706
707 ins->rex |= REX_V;
708 if ((prefix->rex & (REX_V|REX_D|REX_P)) != REX_V)
709 return false;
710
711 if ((vexm & 0x1f) != prefix->vex_m)
712 return false;
713
714 switch (vexwlp & 060) {
715 case 000:
716 if (prefix->rex & REX_W)
717 return false;
718 break;
719 case 020:
720 if (!(prefix->rex & REX_W))
721 return false;
722 ins->rex &= ~REX_W;
723 break;
724 case 040: /* VEX.W is a don't care */
725 ins->rex &= ~REX_W;
726 break;
727 case 060:
728 break;
729 }
730
731 /* The 010 bit of vexwlp is set if VEX.L is ignored */
732 if ((vexwlp ^ prefix->vex_lp) & ((vexwlp & 010) ? 03 : 07))
733 return false;
734
735 if (c == 0270) {
736 if (prefix->vex_v != 0)
737 return false;
738 } else {
739 opx->segment |= SEG_RMREG;
740 opx->basereg = prefix->vex_v;
741 }
742 vex_ok = true;
743 break;
744 }
745
746 case 0310:
747 if (asize != 16)
748 return false;
749 else
750 a_used = true;
751 break;
752
753 case 0311:
754 if (asize != 32)
755 return false;
756 else
757 a_used = true;
758 break;
759
760 case 0312:
761 if (asize != segsize)
762 return false;
763 else
764 a_used = true;
765 break;
766
767 case 0313:
768 if (asize != 64)
769 return false;
770 else
771 a_used = true;
772 break;
773
774 case 0314:
775 if (prefix->rex & REX_B)
776 return false;
777 break;
778
779 case 0315:
780 if (prefix->rex & REX_X)
781 return false;
782 break;
783
784 case 0316:
785 if (prefix->rex & REX_R)
786 return false;
787 break;
788
789 case 0317:
790 if (prefix->rex & REX_W)
791 return false;
792 break;
793
794 case 0320:
795 if (osize != 16)
796 return false;
797 else
798 o_used = true;
799 break;
800
801 case 0321:
802 if (osize != 32)
803 return false;
804 else
805 o_used = true;
806 break;
807
808 case 0322:
809 if (osize != (segsize == 16) ? 16 : 32)
810 return false;
811 else
812 o_used = true;
813 break;
814
815 case 0323:
816 ins->rex |= REX_W; /* 64-bit only instruction */
817 osize = 64;
818 o_used = true;
819 break;
820
821 case 0324:
822 if (osize != 64)
823 return false;
824 o_used = true;
825 break;
826
827 case 0325:
828 ins->rex |= REX_NH;
829 break;
830
831 case 0330:
832 {
833 int t = *r++, d = *data++;
834 if (d < t || d > t + 15)
835 return false;
836 else
837 ins->condition = d - t;
838 break;
839 }
840
841 case 0331:
842 if (prefix->rep)
843 return false;
844 break;
845
846 case 0332:
847 if (prefix->rep != 0xF2)
848 return false;
849 drep = 0;
850 break;
851
852 case 0333:
853 if (prefix->rep != 0xF3)
854 return false;
855 drep = 0;
856 break;
857
858 case 0334:
859 if (lock) {
860 ins->rex |= REX_R;
861 lock = 0;
862 }
863 break;
864
865 case 0335:
866 if (drep == P_REP)
867 drep = P_REPE;
868 break;
869
870 case 0336:
871 case 0337:
872 break;
873
874 case 0340:
875 return false;
876
877 case 0341:
878 if (prefix->wait != 0x9B)
879 return false;
880 dwait = 0;
881 break;
882
883 case4(0344):
884 ins->oprs[0].basereg = (*data++ >> 3) & 7;
885 break;
886
887 case 0360:
888 if (prefix->osp || prefix->rep)
889 return false;
890 break;
891
892 case 0361:
893 if (!prefix->osp || prefix->rep)
894 return false;
895 o_used = true;
896 break;
897
898 case 0362:
899 if (prefix->osp || prefix->rep != 0xf2)
900 return false;
901 drep = 0;
902 break;
903
904 case 0363:
905 if (prefix->osp || prefix->rep != 0xf3)
906 return false;
907 drep = 0;
908 break;
909
910 case 0364:
911 if (prefix->osp)
912 return false;
913 break;
914
915 case 0365:
916 if (prefix->asp)
917 return false;
918 break;
919
920 case 0366:
921 if (!prefix->osp)
922 return false;
923 o_used = true;
924 break;
925
926 case 0367:
927 if (!prefix->asp)
928 return false;
929 a_used = true;
930 break;
931
932 case 0374:
933 eat = EA_XMMVSIB;
934 break;
935
936 case 0375:
937 eat = EA_YMMVSIB;
938 break;
939
940 default:
941 return false; /* Unknown code */
942 }
943 }
944
945 if (!vex_ok && (ins->rex & REX_V))
946 return false;
947
948 /* REX cannot be combined with DREX or VEX */
949 if ((ins->rex & (REX_D|REX_V)) && (prefix->rex & REX_P))
950 return false;
951
952 /*
953 * Check for unused rep or a/o prefixes.
954 */
955 for (i = 0; i < t->operands; i++) {
956 if (ins->oprs[i].segment != SEG_RMREG)
957 a_used = true;
958 }
959
960 if (lock) {
961 if (ins->prefixes[PPS_LREP])
962 return false;
963 ins->prefixes[PPS_LREP] = P_LOCK;
964 }
965 if (drep) {
966 if (ins->prefixes[PPS_LREP])
967 return false;
968 ins->prefixes[PPS_LREP] = drep;
969 }
970 ins->prefixes[PPS_WAIT] = dwait;
971 if (!o_used) {
972 if (osize != ((segsize == 16) ? 16 : 32)) {
973 enum prefixes pfx = 0;
974
975 switch (osize) {
976 case 16:
977 pfx = P_O16;
978 break;
979 case 32:
980 pfx = P_O32;
981 break;
982 case 64:
983 pfx = P_O64;
984 break;
985 }
986
987 if (ins->prefixes[PPS_OSIZE])
988 return false;
989 ins->prefixes[PPS_OSIZE] = pfx;
990 }
991 }
992 if (!a_used && asize != segsize) {
993 if (ins->prefixes[PPS_ASIZE])
994 return false;
995 ins->prefixes[PPS_ASIZE] = asize == 16 ? P_A16 : P_A32;
996 }
997
998 /* Fix: check for redundant REX prefixes */
999
1000 return data - origdata;
1001 }
1002
1003 /* Condition names for disassembly, sorted by x86 code */
1004 static const char * const condition_name[16] = {
1005 "o", "no", "c", "nc", "z", "nz", "na", "a",
1006 "s", "ns", "pe", "po", "l", "nl", "ng", "g"
1007 };
1008
1009 int32_t disasm(uint8_t *data, char *output, int outbufsize, int segsize,
1010 int32_t offset, int autosync, uint32_t prefer)
1011 {
1012 const struct itemplate * const *p, * const *best_p;
1013 const struct disasm_index *ix;
1014 uint8_t *dp;
1015 int length, best_length = 0;
1016 char *segover;
1017 int i, slen, colon, n;
1018 uint8_t *origdata;
1019 int works;
1020 insn tmp_ins, ins;
1021 uint32_t goodness, best;
1022 int best_pref;
1023 struct prefix_info prefix;
1024 bool end_prefix;
1025
1026 memset(&ins, 0, sizeof ins);
1027
1028 /*
1029 * Scan for prefixes.
1030 */
1031 memset(&prefix, 0, sizeof prefix);
1032 prefix.asize = segsize;
1033 prefix.osize = (segsize == 64) ? 32 : segsize;
1034 segover = NULL;
1035 origdata = data;
1036
1037 ix = itable;
1038
1039 end_prefix = false;
1040 while (!end_prefix) {
1041 switch (*data) {
1042 case 0xF2:
1043 case 0xF3:
1044 prefix.rep = *data++;
1045 break;
1046
1047 case 0x9B:
1048 prefix.wait = *data++;
1049 break;
1050
1051 case 0xF0:
1052 prefix.lock = *data++;
1053 break;
1054
1055 case 0x2E:
1056 segover = "cs", prefix.seg = *data++;
1057 break;
1058 case 0x36:
1059 segover = "ss", prefix.seg = *data++;
1060 break;
1061 case 0x3E:
1062 segover = "ds", prefix.seg = *data++;
1063 break;
1064 case 0x26:
1065 segover = "es", prefix.seg = *data++;
1066 break;
1067 case 0x64:
1068 segover = "fs", prefix.seg = *data++;
1069 break;
1070 case 0x65:
1071 segover = "gs", prefix.seg = *data++;
1072 break;
1073
1074 case 0x66:
1075 prefix.osize = (segsize == 16) ? 32 : 16;
1076 prefix.osp = *data++;
1077 break;
1078 case 0x67:
1079 prefix.asize = (segsize == 32) ? 16 : 32;
1080 prefix.asp = *data++;
1081 break;
1082
1083 case 0xC4:
1084 case 0xC5:
1085 if (segsize == 64 || (data[1] & 0xc0) == 0xc0) {
1086 prefix.vex[0] = *data++;
1087 prefix.vex[1] = *data++;
1088
1089 prefix.rex = REX_V;
1090 prefix.vex_c = RV_VEX;
1091
1092 if (prefix.vex[0] == 0xc4) {
1093 prefix.vex[2] = *data++;
1094 prefix.rex |= (~prefix.vex[1] >> 5) & 7; /* REX_RXB */
1095 prefix.rex |= (prefix.vex[2] >> (7-3)) & REX_W;
1096 prefix.vex_m = prefix.vex[1] & 0x1f;
1097 prefix.vex_v = (~prefix.vex[2] >> 3) & 15;
1098 prefix.vex_lp = prefix.vex[2] & 7;
1099 } else {
1100 prefix.rex |= (~prefix.vex[1] >> (7-2)) & REX_R;
1101 prefix.vex_m = 1;
1102 prefix.vex_v = (~prefix.vex[1] >> 3) & 15;
1103 prefix.vex_lp = prefix.vex[1] & 7;
1104 }
1105
1106 ix = itable_vex[RV_VEX][prefix.vex_m][prefix.vex_lp & 3];
1107 }
1108 end_prefix = true;
1109 break;
1110
1111 case 0x8F:
1112 if ((data[1] & 030) != 0 &&
1113 (segsize == 64 || (data[1] & 0xc0) == 0xc0)) {
1114 prefix.vex[0] = *data++;
1115 prefix.vex[1] = *data++;
1116 prefix.vex[2] = *data++;
1117
1118 prefix.rex = REX_V;
1119 prefix.vex_c = RV_XOP;
1120
1121 prefix.rex |= (~prefix.vex[1] >> 5) & 7; /* REX_RXB */
1122 prefix.rex |= (prefix.vex[2] >> (7-3)) & REX_W;
1123 prefix.vex_m = prefix.vex[1] & 0x1f;
1124 prefix.vex_v = (~prefix.vex[2] >> 3) & 15;
1125 prefix.vex_lp = prefix.vex[2] & 7;
1126
1127 ix = itable_vex[RV_XOP][prefix.vex_m][prefix.vex_lp & 3];
1128 }
1129 end_prefix = true;
1130 break;
1131
1132 case REX_P + 0x0:
1133 case REX_P + 0x1:
1134 case REX_P + 0x2:
1135 case REX_P + 0x3:
1136 case REX_P + 0x4:
1137 case REX_P + 0x5:
1138 case REX_P + 0x6:
1139 case REX_P + 0x7:
1140 case REX_P + 0x8:
1141 case REX_P + 0x9:
1142 case REX_P + 0xA:
1143 case REX_P + 0xB:
1144 case REX_P + 0xC:
1145 case REX_P + 0xD:
1146 case REX_P + 0xE:
1147 case REX_P + 0xF:
1148 if (segsize == 64) {
1149 prefix.rex = *data++;
1150 if (prefix.rex & REX_W)
1151 prefix.osize = 64;
1152 }
1153 end_prefix = true;
1154 break;
1155
1156 default:
1157 end_prefix = true;
1158 break;
1159 }
1160 }
1161
1162 best = -1; /* Worst possible */
1163 best_p = NULL;
1164 best_pref = INT_MAX;
1165
1166 if (!ix)
1167 return 0; /* No instruction table at all... */
1168
1169 dp = data;
1170 ix += *dp++;
1171 while (ix->n == -1) {
1172 ix = (const struct disasm_index *)ix->p + *dp++;
1173 }
1174
1175 p = (const struct itemplate * const *)ix->p;
1176 for (n = ix->n; n; n--, p++) {
1177 if ((length = matches(*p, data, &prefix, segsize, &tmp_ins))) {
1178 works = true;
1179 /*
1180 * Final check to make sure the types of r/m match up.
1181 * XXX: Need to make sure this is actually correct.
1182 */
1183 for (i = 0; i < (*p)->operands; i++) {
1184 if (!((*p)->opd[i] & SAME_AS) &&
1185 (
1186 /* If it's a mem-only EA but we have a
1187 register, die. */
1188 ((tmp_ins.oprs[i].segment & SEG_RMREG) &&
1189 is_class(MEMORY, (*p)->opd[i])) ||
1190 /* If it's a reg-only EA but we have a memory
1191 ref, die. */
1192 (!(tmp_ins.oprs[i].segment & SEG_RMREG) &&
1193 !(REG_EA & ~(*p)->opd[i]) &&
1194 !((*p)->opd[i] & REG_SMASK)) ||
1195 /* Register type mismatch (eg FS vs REG_DESS):
1196 die. */
1197 ((((*p)->opd[i] & (REGISTER | FPUREG)) ||
1198 (tmp_ins.oprs[i].segment & SEG_RMREG)) &&
1199 !whichreg((*p)->opd[i],
1200 tmp_ins.oprs[i].basereg, tmp_ins.rex))
1201 )) {
1202 works = false;
1203 break;
1204 }
1205 }
1206
1207 /*
1208 * Note: we always prefer instructions which incorporate
1209 * prefixes in the instructions themselves. This is to allow
1210 * e.g. PAUSE to be preferred to REP NOP, and deal with
1211 * MMX/SSE instructions where prefixes are used to select
1212 * between MMX and SSE register sets or outright opcode
1213 * selection.
1214 */
1215 if (works) {
1216 int i, nprefix;
1217 goodness = ((*p)->flags & IF_PFMASK) ^ prefer;
1218 nprefix = 0;
1219 for (i = 0; i < MAXPREFIX; i++)
1220 if (tmp_ins.prefixes[i])
1221 nprefix++;
1222 if (nprefix < best_pref ||
1223 (nprefix == best_pref && goodness < best)) {
1224 /* This is the best one found so far */
1225 best = goodness;
1226 best_p = p;
1227 best_pref = nprefix;
1228 best_length = length;
1229 ins = tmp_ins;
1230 }
1231 }
1232 }
1233 }
1234
1235 if (!best_p)
1236 return 0; /* no instruction was matched */
1237
1238 /* Pick the best match */
1239 p = best_p;
1240 length = best_length;
1241
1242 slen = 0;
1243
1244 /* TODO: snprintf returns the value that the string would have if
1245 * the buffer were long enough, and not the actual length of
1246 * the returned string, so each instance of using the return
1247 * value of snprintf should actually be checked to assure that
1248 * the return value is "sane." Maybe a macro wrapper could
1249 * be used for that purpose.
1250 */
1251 for (i = 0; i < MAXPREFIX; i++) {
1252 const char *prefix = prefix_name(ins.prefixes[i]);
1253 if (prefix)
1254 slen += snprintf(output+slen, outbufsize-slen, "%s ", prefix);
1255 }
1256
1257 i = (*p)->opcode;
1258 if (i >= FIRST_COND_OPCODE)
1259 slen += snprintf(output + slen, outbufsize - slen, "%s%s",
1260 nasm_insn_names[i], condition_name[ins.condition]);
1261 else
1262 slen += snprintf(output + slen, outbufsize - slen, "%s",
1263 nasm_insn_names[i]);
1264
1265 colon = false;
1266 length += data - origdata; /* fix up for prefixes */
1267 for (i = 0; i < (*p)->operands; i++) {
1268 opflags_t t = (*p)->opd[i];
1269 const operand *o = &ins.oprs[i];
1270 int64_t offs;
1271
1272 if (t & SAME_AS) {
1273 o = &ins.oprs[t & ~SAME_AS];
1274 t = (*p)->opd[t & ~SAME_AS];
1275 }
1276
1277 output[slen++] = (colon ? ':' : i == 0 ? ' ' : ',');
1278
1279 offs = o->offset;
1280 if (o->segment & SEG_RELATIVE) {
1281 offs += offset + length;
1282 /*
1283 * sort out wraparound
1284 */
1285 if (!(o->segment & (SEG_32BIT|SEG_64BIT)))
1286 offs &= 0xffff;
1287 else if (segsize != 64)
1288 offs &= 0xffffffff;
1289
1290 /*
1291 * add sync marker, if autosync is on
1292 */
1293 if (autosync)
1294 add_sync(offs, 0L);
1295 }
1296
1297 if (t & COLON)
1298 colon = true;
1299 else
1300 colon = false;
1301
1302 if ((t & (REGISTER | FPUREG)) ||
1303 (o->segment & SEG_RMREG)) {
1304 enum reg_enum reg;
1305 reg = whichreg(t, o->basereg, ins.rex);
1306 if (t & TO)
1307 slen += snprintf(output + slen, outbufsize - slen, "to ");
1308 slen += snprintf(output + slen, outbufsize - slen, "%s",
1309 nasm_reg_names[reg-EXPR_REG_START]);
1310 } else if (!(UNITY & ~t)) {
1311 output[slen++] = '1';
1312 } else if (t & IMMEDIATE) {
1313 if (t & BITS8) {
1314 slen +=
1315 snprintf(output + slen, outbufsize - slen, "byte ");
1316 if (o->segment & SEG_SIGNED) {
1317 if (offs < 0) {
1318 offs *= -1;
1319 output[slen++] = '-';
1320 } else
1321 output[slen++] = '+';
1322 }
1323 } else if (t & BITS16) {
1324 slen +=
1325 snprintf(output + slen, outbufsize - slen, "word ");
1326 } else if (t & BITS32) {
1327 slen +=
1328 snprintf(output + slen, outbufsize - slen, "dword ");
1329 } else if (t & BITS64) {
1330 slen +=
1331 snprintf(output + slen, outbufsize - slen, "qword ");
1332 } else if (t & NEAR) {
1333 slen +=
1334 snprintf(output + slen, outbufsize - slen, "near ");
1335 } else if (t & SHORT) {
1336 slen +=
1337 snprintf(output + slen, outbufsize - slen, "short ");
1338 }
1339 slen +=
1340 snprintf(output + slen, outbufsize - slen, "0x%"PRIx64"",
1341 offs);
1342 } else if (!(MEM_OFFS & ~t)) {
1343 slen +=
1344 snprintf(output + slen, outbufsize - slen,
1345 "[%s%s%s0x%"PRIx64"]",
1346 (segover ? segover : ""),
1347 (segover ? ":" : ""),
1348 (o->disp_size == 64 ? "qword " :
1349 o->disp_size == 32 ? "dword " :
1350 o->disp_size == 16 ? "word " : ""), offs);
1351 segover = NULL;
1352 } else if (is_class(REGMEM, t)) {
1353 int started = false;
1354 if (t & BITS8)
1355 slen +=
1356 snprintf(output + slen, outbufsize - slen, "byte ");
1357 if (t & BITS16)
1358 slen +=
1359 snprintf(output + slen, outbufsize - slen, "word ");
1360 if (t & BITS32)
1361 slen +=
1362 snprintf(output + slen, outbufsize - slen, "dword ");
1363 if (t & BITS64)
1364 slen +=
1365 snprintf(output + slen, outbufsize - slen, "qword ");
1366 if (t & BITS80)
1367 slen +=
1368 snprintf(output + slen, outbufsize - slen, "tword ");
1369 if (t & BITS128)
1370 slen +=
1371 snprintf(output + slen, outbufsize - slen, "oword ");
1372 if (t & BITS256)
1373 slen +=
1374 snprintf(output + slen, outbufsize - slen, "yword ");
1375 if (t & FAR)
1376 slen += snprintf(output + slen, outbufsize - slen, "far ");
1377 if (t & NEAR)
1378 slen +=
1379 snprintf(output + slen, outbufsize - slen, "near ");
1380 output[slen++] = '[';
1381 if (o->disp_size)
1382 slen += snprintf(output + slen, outbufsize - slen, "%s",
1383 (o->disp_size == 64 ? "qword " :
1384 o->disp_size == 32 ? "dword " :
1385 o->disp_size == 16 ? "word " :
1386 ""));
1387 if (o->eaflags & EAF_REL)
1388 slen += snprintf(output + slen, outbufsize - slen, "rel ");
1389 if (segover) {
1390 slen +=
1391 snprintf(output + slen, outbufsize - slen, "%s:",
1392 segover);
1393 segover = NULL;
1394 }
1395 if (o->basereg != -1) {
1396 slen += snprintf(output + slen, outbufsize - slen, "%s",
1397 nasm_reg_names[(o->basereg-EXPR_REG_START)]);
1398 started = true;
1399 }
1400 if (o->indexreg != -1) {
1401 if (started)
1402 output[slen++] = '+';
1403 slen += snprintf(output + slen, outbufsize - slen, "%s",
1404 nasm_reg_names[(o->indexreg-EXPR_REG_START)]);
1405 if (o->scale > 1)
1406 slen +=
1407 snprintf(output + slen, outbufsize - slen, "*%d",
1408 o->scale);
1409 started = true;
1410 }
1411
1412
1413 if (o->segment & SEG_DISP8) {
1414 const char *prefix;
1415 uint8_t offset = offs;
1416 if ((int8_t)offset < 0) {
1417 prefix = "-";
1418 offset = -offset;
1419 } else {
1420 prefix = "+";
1421 }
1422 slen +=
1423 snprintf(output + slen, outbufsize - slen, "%s0x%"PRIx8"",
1424 prefix, offset);
1425 } else if (o->segment & SEG_DISP16) {
1426 const char *prefix;
1427 uint16_t offset = offs;
1428 if ((int16_t)offset < 0 && started) {
1429 offset = -offset;
1430 prefix = "-";
1431 } else {
1432 prefix = started ? "+" : "";
1433 }
1434 slen +=
1435 snprintf(output + slen, outbufsize - slen,
1436 "%s0x%"PRIx16"", prefix, offset);
1437 } else if (o->segment & SEG_DISP32) {
1438 if (prefix.asize == 64) {
1439 const char *prefix;
1440 uint64_t offset = (int64_t)(int32_t)offs;
1441 if ((int32_t)offs < 0 && started) {
1442 offset = -offset;
1443 prefix = "-";
1444 } else {
1445 prefix = started ? "+" : "";
1446 }
1447 slen +=
1448 snprintf(output + slen, outbufsize - slen,
1449 "%s0x%"PRIx64"", prefix, offset);
1450 } else {
1451 const char *prefix;
1452 uint32_t offset = offs;
1453 if ((int32_t) offset < 0 && started) {
1454 offset = -offset;
1455 prefix = "-";
1456 } else {
1457 prefix = started ? "+" : "";
1458 }
1459 slen +=
1460 snprintf(output + slen, outbufsize - slen,
1461 "%s0x%"PRIx32"", prefix, offset);
1462 }
1463 }
1464 output[slen++] = ']';
1465 } else {
1466 slen +=
1467 snprintf(output + slen, outbufsize - slen, "<operand%d>",
1468 i);
1469 }
1470 }
1471 output[slen] = '\0';
1472 if (segover) { /* unused segment override */
1473 char *p = output;
1474 int count = slen + 1;
1475 while (count--)
1476 p[count + 3] = p[count];
1477 strncpy(output, segover, 2);
1478 output[2] = ' ';
1479 }
1480 return length;
1481 }
1482
1483 /*
1484 * This is called when we don't have a complete instruction. If it
1485 * is a standalone *single-byte* prefix show it as such, otherwise
1486 * print it as a literal.
1487 */
1488 int32_t eatbyte(uint8_t *data, char *output, int outbufsize, int segsize)
1489 {
1490 uint8_t byte = *data;
1491 const char *str = NULL;
1492
1493 switch (byte) {
1494 case 0xF2:
1495 str = "repne";
1496 break;
1497 case 0xF3:
1498 str = "rep";
1499 break;
1500 case 0x9B:
1501 str = "wait";
1502 break;
1503 case 0xF0:
1504 str = "lock";
1505 break;
1506 case 0x2E:
1507 str = "cs";
1508 break;
1509 case 0x36:
1510 str = "ss";
1511 break;
1512 case 0x3E:
1513 str = "ss";
1514 break;
1515 case 0x26:
1516 str = "es";
1517 break;
1518 case 0x64:
1519 str = "fs";
1520 break;
1521 case 0x65:
1522 str = "gs";
1523 break;
1524 case 0x66:
1525 str = (segsize == 16) ? "o32" : "o16";
1526 break;
1527 case 0x67:
1528 str = (segsize == 32) ? "a16" : "a32";
1529 break;
1530 case REX_P + 0x0:
1531 case REX_P + 0x1:
1532 case REX_P + 0x2:
1533 case REX_P + 0x3:
1534 case REX_P + 0x4:
1535 case REX_P + 0x5:
1536 case REX_P + 0x6:
1537 case REX_P + 0x7:
1538 case REX_P + 0x8:
1539 case REX_P + 0x9:
1540 case REX_P + 0xA:
1541 case REX_P + 0xB:
1542 case REX_P + 0xC:
1543 case REX_P + 0xD:
1544 case REX_P + 0xE:
1545 case REX_P + 0xF:
1546 if (segsize == 64) {
1547 snprintf(output, outbufsize, "rex%s%s%s%s%s",
1548 (byte == REX_P) ? "" : ".",
1549 (byte & REX_W) ? "w" : "",
1550 (byte & REX_R) ? "r" : "",
1551 (byte & REX_X) ? "x" : "",
1552 (byte & REX_B) ? "b" : "");
1553 break;
1554 }
1555 /* else fall through */
1556 default:
1557 snprintf(output, outbufsize, "db 0x%02x", byte);
1558 break;
1559 }
1560
1561 if (str)
1562 snprintf(output, outbufsize, "%s", str);
1563
1564 return 1;
1565 }
The Netwide Assembler