1 /* disasm.c where all the _work_ gets done in the Netwide Disassembler
3 * The Netwide Assembler is copyright (C) 1996 Simon Tatham and
4 * Julian Hall. All rights reserved. The software is
5 * redistributable under the licence given in the file "Licence"
6 * distributed in the NASM archive.
8 * initial version 27/iii/95 by Simon Tatham
21 extern struct itemplate **itable[];
24 * Flags that go into the `segment' field of `insn' structures
27 #define SEG_RELATIVE 1
34 #define SEG_SIGNED 128
36 static int whichreg(long regflags, int regval)
40 if (!(REG_AL & ~regflags))
42 if (!(REG_AX & ~regflags))
44 if (!(REG_EAX & ~regflags))
46 if (!(REG_DL & ~regflags))
48 if (!(REG_DX & ~regflags))
50 if (!(REG_EDX & ~regflags))
52 if (!(REG_CL & ~regflags))
54 if (!(REG_CX & ~regflags))
56 if (!(REG_ECX & ~regflags))
58 if (!(FPU0 & ~regflags))
60 if (!(REG_CS & ~regflags))
61 return (regval == 1) ? R_CS : 0;
62 if (!(REG_DESS & ~regflags))
63 return (regval == 0 || regval == 2 || regval == 3 ? sreg[regval] : 0);
64 if (!(REG_FSGS & ~regflags))
65 return (regval == 4 || regval == 5 ? sreg[regval] : 0);
66 if (!(REG_SEG67 & ~regflags))
67 return (regval == 6 || regval == 7 ? sreg[regval] : 0);
69 /* All the entries below look up regval in an 8-entry array */
70 if (regval < 0 || regval > 7)
73 if (!((REGMEM|BITS8) & ~regflags))
75 if (!((REGMEM|BITS16) & ~regflags))
77 if (!((REGMEM|BITS32) & ~regflags))
79 if (!(REG_SREG & ~regflags))
81 if (!(REG_CREG & ~regflags))
83 if (!(REG_DREG & ~regflags))
85 if (!(REG_TREG & ~regflags))
87 if (!(FPUREG & ~regflags))
88 return fpureg[regval];
89 if (!(MMXREG & ~regflags))
90 return mmxreg[regval];
91 if (!(XMMREG & ~regflags))
92 return xmmreg[regval];
97 static const char *whichcond(int condval)
99 static int conds[] = {
100 C_O, C_NO, C_C, C_NC, C_Z, C_NZ, C_NA, C_A,
101 C_S, C_NS, C_PE, C_PO, C_L, C_NL, C_NG, C_G
103 return conditions[conds[condval]];
107 * Process an effective address (ModRM) specification.
109 static unsigned char *do_ea (unsigned char *data, int modrm, int asize,
110 int segsize, operand *op)
112 int mod, rm, scale, index, base;
114 mod = (modrm >> 6) & 03;
117 if (mod == 3) { /* pure register version */
119 op->segment |= SEG_RMREG;
127 * <mod> specifies the displacement size (none, byte or
128 * word), and <rm> specifies the register combination.
129 * Exception: mod=0,rm=6 does not specify [BP] as one might
130 * expect, but instead specifies [disp16].
132 op->indexreg = op->basereg = -1;
133 op->scale = 1; /* always, in 16 bits */
135 case 0: op->basereg = R_BX; op->indexreg = R_SI; break;
136 case 1: op->basereg = R_BX; op->indexreg = R_DI; break;
137 case 2: op->basereg = R_BP; op->indexreg = R_SI; break;
138 case 3: op->basereg = R_BP; op->indexreg = R_DI; break;
139 case 4: op->basereg = R_SI; break;
140 case 5: op->basereg = R_DI; break;
141 case 6: op->basereg = R_BP; break;
142 case 7: op->basereg = R_BX; break;
144 if (rm == 6 && mod == 0) { /* special case */
148 mod = 2; /* fake disp16 */
152 op->segment |= SEG_NODISP;
155 op->segment |= SEG_DISP8;
156 op->offset = (signed char) *data++;
159 op->segment |= SEG_DISP16;
160 op->offset = *data++;
161 op->offset |= ((unsigned) *data++) << 8;
167 * Once again, <mod> specifies displacement size (this time
168 * none, byte or *dword*), while <rm> specifies the base
169 * register. Again, [EBP] is missing, replaced by a pure
170 * disp32 (this time that's mod=0,rm=*5*). However, rm=4
171 * indicates not a single base register, but instead the
172 * presence of a SIB byte...
176 case 0: op->basereg = R_EAX; break;
177 case 1: op->basereg = R_ECX; break;
178 case 2: op->basereg = R_EDX; break;
179 case 3: op->basereg = R_EBX; break;
180 case 5: op->basereg = R_EBP; break;
181 case 6: op->basereg = R_ESI; break;
182 case 7: op->basereg = R_EDI; break;
184 if (rm == 5 && mod == 0) {
188 mod = 2; /* fake disp32 */
190 if (rm == 4) { /* process SIB */
191 scale = (*data >> 6) & 03;
192 index = (*data >> 3) & 07;
196 op->scale = 1 << scale;
198 case 0: op->indexreg = R_EAX; break;
199 case 1: op->indexreg = R_ECX; break;
200 case 2: op->indexreg = R_EDX; break;
201 case 3: op->indexreg = R_EBX; break;
202 case 4: op->indexreg = -1; break;
203 case 5: op->indexreg = R_EBP; break;
204 case 6: op->indexreg = R_ESI; break;
205 case 7: op->indexreg = R_EDI; break;
209 case 0: op->basereg = R_EAX; break;
210 case 1: op->basereg = R_ECX; break;
211 case 2: op->basereg = R_EDX; break;
212 case 3: op->basereg = R_EBX; break;
213 case 4: op->basereg = R_ESP; break;
214 case 6: op->basereg = R_ESI; break;
215 case 7: op->basereg = R_EDI; break;
227 op->segment |= SEG_NODISP;
230 op->segment |= SEG_DISP8;
231 op->offset = (signed char) *data++;
234 op->segment |= SEG_DISP32;
235 op->offset = *data++;
236 op->offset |= ((unsigned) *data++) << 8;
237 op->offset |= ((long) *data++) << 16;
238 op->offset |= ((long) *data++) << 24;
246 * Determine whether the instruction template in t corresponds to the data
247 * stream in data. Return the number of bytes matched if so.
249 static int matches (struct itemplate *t, unsigned char *data, int asize,
250 int osize, int segsize, int rep, insn *ins)
252 unsigned char * r = (unsigned char *)(t->code);
253 unsigned char * origdata = data;
254 int a_used = FALSE, o_used = FALSE;
259 else if ( rep == 0xF3 )
265 if (c >= 01 && c <= 03) {
272 case 0x07: ins->oprs[0].basereg = 0; break;
273 case 0x17: ins->oprs[0].basereg = 2; break;
274 case 0x1F: ins->oprs[0].basereg = 3; break;
275 default: return FALSE;
280 case 0xA1: ins->oprs[0].basereg = 4; break;
281 case 0xA9: ins->oprs[0].basereg = 5; break;
282 default: return FALSE;
287 case 0x06: ins->oprs[0].basereg = 0; break;
288 case 0x0E: ins->oprs[0].basereg = 1; break;
289 case 0x16: ins->oprs[0].basereg = 2; break;
290 case 0x1E: ins->oprs[0].basereg = 3; break;
291 default: return FALSE;
296 case 0xA0: ins->oprs[0].basereg = 4; break;
297 case 0xA8: ins->oprs[0].basereg = 5; break;
298 default: return FALSE;
301 if (c >= 010 && c <= 012) {
302 int t = *r++, d = *data++;
303 if (d < t || d > t+7)
306 ins->oprs[c-010].basereg = d-t;
307 ins->oprs[c-010].segment |= SEG_RMREG;
313 if (c >= 014 && c <= 016) {
314 ins->oprs[c-014].offset = (signed char) *data++;
315 ins->oprs[c-014].segment |= SEG_SIGNED;
317 if (c >= 020 && c <= 022)
318 ins->oprs[c-020].offset = *data++;
319 if (c >= 024 && c <= 026)
320 ins->oprs[c-024].offset = *data++;
321 if (c >= 030 && c <= 032) {
322 ins->oprs[c-030].offset = *data++;
323 ins->oprs[c-030].offset |= (((unsigned) *data++) << 8);
325 if (c >= 034 && c <= 036) {
326 ins->oprs[c-034].offset = *data++;
327 ins->oprs[c-034].offset |= (((unsigned) *data++) << 8);
329 ins->oprs[c-034].offset |= (((long) *data++) << 16);
330 ins->oprs[c-034].offset |= (((long) *data++) << 24);
332 if (segsize != asize)
333 ins->oprs[c-034].addr_size = asize;
335 if (c >= 040 && c <= 042) {
336 ins->oprs[c-040].offset = *data++;
337 ins->oprs[c-040].offset |= (((unsigned) *data++) << 8);
338 ins->oprs[c-040].offset |= (((long) *data++) << 16);
339 ins->oprs[c-040].offset |= (((long) *data++) << 24);
341 if (c >= 044 && c <= 046) {
342 ins->oprs[c-044].offset = *data++;
343 ins->oprs[c-044].offset |= (((unsigned) *data++) << 8);
345 ins->oprs[c-044].offset |= (((long) *data++) << 16);
346 ins->oprs[c-044].offset |= (((long) *data++) << 24);
348 if (segsize != asize)
349 ins->oprs[c-044].addr_size = asize;
351 if (c >= 050 && c <= 052) {
352 ins->oprs[c-050].offset = (signed char) *data++;
353 ins->oprs[c-050].segment |= SEG_RELATIVE;
355 if (c >= 060 && c <= 062) {
356 ins->oprs[c-060].offset = *data++;
357 ins->oprs[c-060].offset |= (((unsigned) *data++) << 8);
358 ins->oprs[c-060].segment |= SEG_RELATIVE;
359 ins->oprs[c-060].segment &= ~SEG_32BIT;
361 if (c >= 064 && c <= 066) {
362 ins->oprs[c-064].offset = *data++;
363 ins->oprs[c-064].offset |= (((unsigned) *data++) << 8);
365 ins->oprs[c-064].offset |= (((long) *data++) << 16);
366 ins->oprs[c-064].offset |= (((long) *data++) << 24);
367 ins->oprs[c-064].segment |= SEG_32BIT;
369 ins->oprs[c-064].segment &= ~SEG_32BIT;
370 ins->oprs[c-064].segment |= SEG_RELATIVE;
371 if (segsize != osize) {
372 ins->oprs[c-064].type =
373 (ins->oprs[c-064].type & NON_SIZE)
374 | ((osize == 16) ? BITS16 : BITS32);
377 if (c >= 070 && c <= 072) {
378 ins->oprs[c-070].offset = *data++;
379 ins->oprs[c-070].offset |= (((unsigned) *data++) << 8);
380 ins->oprs[c-070].offset |= (((long) *data++) << 16);
381 ins->oprs[c-070].offset |= (((long) *data++) << 24);
382 ins->oprs[c-070].segment |= SEG_32BIT | SEG_RELATIVE;
384 if (c >= 0100 && c < 0130) {
386 ins->oprs[c & 07].basereg = (modrm >> 3) & 07;
387 ins->oprs[c & 07].segment |= SEG_RMREG;
388 data = do_ea (data, modrm, asize, segsize,
389 &ins->oprs[(c >> 3) & 07]);
391 if (c >= 0130 && c <= 0132) {
392 ins->oprs[c-0130].offset = *data++;
393 ins->oprs[c-0130].offset |= (((unsigned) *data++) << 8);
395 if (c >= 0140 && c <= 0142) {
396 ins->oprs[c-0140].offset = *data++;
397 ins->oprs[c-0140].offset |= (((unsigned) *data++) << 8);
398 ins->oprs[c-0140].offset |= (((long) *data++) << 16);
399 ins->oprs[c-0140].offset |= (((long) *data++) << 24);
401 if (c >= 0200 && c <= 0277) {
403 if (((modrm >> 3) & 07) != (c & 07))
404 return FALSE; /* spare field doesn't match up */
405 data = do_ea (data, modrm, asize, segsize,
406 &ins->oprs[(c >> 3) & 07]);
408 if (c >= 0300 && c <= 0302) {
410 ins->oprs[c-0300].segment |= SEG_32BIT;
412 ins->oprs[c-0300].segment &= ~SEG_32BIT;
428 if (asize != segsize)
446 if (osize != segsize)
452 int t = *r++, d = *data++;
453 if (d < t || d > t+15)
456 ins->condition = d - t;
474 * Check for unused rep or a/o prefixes.
478 ins->prefixes[ins->nprefix++] = drep;
479 if (!a_used && asize != segsize)
480 ins->prefixes[ins->nprefix++] = (asize == 16 ? P_A16 : P_A32);
481 if (!o_used && osize != segsize)
482 ins->prefixes[ins->nprefix++] = (osize == 16 ? P_O16 : P_O32);
484 return data - origdata;
487 long disasm (unsigned char *data, char *output, int segsize, long offset,
488 int autosync, unsigned long prefer)
490 struct itemplate **p, **best_p;
491 int length, best_length = 0;
493 int rep, lock, asize, osize, i, slen, colon;
494 unsigned char *origdata;
497 unsigned long goodness, best;
502 asize = osize = segsize;
507 if (*data == 0xF3 || *data == 0xF2)
509 else if (*data == 0xF0)
511 else if (*data == 0x2E || *data == 0x36 || *data == 0x3E ||
512 *data == 0x26 || *data == 0x64 || *data == 0x65) {
514 case 0x2E: segover = "cs"; break;
515 case 0x36: segover = "ss"; break;
516 case 0x3E: segover = "ds"; break;
517 case 0x26: segover = "es"; break;
518 case 0x64: segover = "fs"; break;
519 case 0x65: segover = "gs"; break;
521 } else if (*data == 0x66)
522 osize = 48 - segsize, data++;
523 else if (*data == 0x67)
524 asize = 48 - segsize, data++;
529 tmp_ins.oprs[0].segment = tmp_ins.oprs[1].segment =
530 tmp_ins.oprs[2].segment =
531 tmp_ins.oprs[0].addr_size = tmp_ins.oprs[1].addr_size =
532 tmp_ins.oprs[2].addr_size = (segsize == 16 ? 0 : SEG_32BIT);
533 tmp_ins.condition = -1;
534 best = ~0UL; /* Worst possible */
536 for (p = itable[*data]; *p; p++) {
537 if ( (length = matches(*p, data, asize, osize,
538 segsize, rep, &tmp_ins)) ) {
541 * Final check to make sure the types of r/m match up.
543 for (i = 0; i < (*p)->operands; i++) {
545 /* If it's a mem-only EA but we have a register, die. */
546 ((tmp_ins.oprs[i].segment & SEG_RMREG) &&
547 !(MEMORY & ~(*p)->opd[i])) ||
549 /* If it's a reg-only EA but we have a memory ref, die. */
550 (!(tmp_ins.oprs[i].segment & SEG_RMREG) &&
551 !(REGNORM & ~(*p)->opd[i]) &&
552 !((*p)->opd[i] & REG_SMASK)) ||
554 /* Register type mismatch (eg FS vs REG_DESS): die. */
555 ((((*p)->opd[i] & (REGISTER | FPUREG)) ||
556 (tmp_ins.oprs[i].segment & SEG_RMREG)) &&
557 !whichreg ((*p)->opd[i], tmp_ins.oprs[i].basereg))) {
564 goodness = ((*p)->flags & IF_PFMASK) ^ prefer;
565 if ( goodness < best ) {
566 /* This is the best one found so far */
569 best_length = length;
577 return 0; /* no instruction was matched */
579 /* Pick the best match */
581 length = best_length;
586 slen += sprintf(output+slen, "lock ");
587 for (i = 0; i < ins.nprefix; i++)
588 switch (ins.prefixes[i]) {
589 case P_REP: slen += sprintf(output+slen, "rep "); break;
590 case P_REPE: slen += sprintf(output+slen, "repe "); break;
591 case P_REPNE: slen += sprintf(output+slen, "repne "); break;
592 case P_A16: slen += sprintf(output+slen, "a16 "); break;
593 case P_A32: slen += sprintf(output+slen, "a32 "); break;
594 case P_O16: slen += sprintf(output+slen, "o16 "); break;
595 case P_O32: slen += sprintf(output+slen, "o32 "); break;
598 for (i = 0; i < elements(ico); i++)
599 if ((*p)->opcode == ico[i]) {
600 slen += sprintf(output+slen, "%s%s", icn[i],
601 whichcond(ins.condition));
604 if (i >= elements(ico))
605 slen += sprintf(output+slen, "%s", insn_names[(*p)->opcode]);
607 length += data - origdata; /* fix up for prefixes */
608 for (i=0; i<(*p)->operands; i++) {
609 output[slen++] = (colon ? ':' : i==0 ? ' ' : ',');
611 if (ins.oprs[i].segment & SEG_RELATIVE) {
612 ins.oprs[i].offset += offset + length;
614 * sort out wraparound
616 if (!(ins.oprs[i].segment & SEG_32BIT))
617 ins.oprs[i].offset &= 0xFFFF;
619 * add sync marker, if autosync is on
622 add_sync (ins.oprs[i].offset, 0L);
625 if ((*p)->opd[i] & COLON)
630 if (((*p)->opd[i] & (REGISTER | FPUREG)) ||
631 (ins.oprs[i].segment & SEG_RMREG))
633 ins.oprs[i].basereg = whichreg ((*p)->opd[i],
634 ins.oprs[i].basereg);
635 if ( (*p)->opd[i] & TO )
636 slen += sprintf(output+slen, "to ");
637 slen += sprintf(output+slen, "%s",
638 reg_names[ins.oprs[i].basereg-EXPR_REG_START]);
639 } else if (!(UNITY & ~(*p)->opd[i])) {
640 output[slen++] = '1';
641 } else if ( (*p)->opd[i] & IMMEDIATE ) {
642 if ( (*p)->opd[i] & BITS8 ) {
643 slen += sprintf(output+slen, "byte ");
644 if (ins.oprs[i].segment & SEG_SIGNED) {
645 if (ins.oprs[i].offset < 0) {
646 ins.oprs[i].offset *= -1;
647 output[slen++] = '-';
649 output[slen++] = '+';
651 } else if ( (*p)->opd[i] & BITS16 ) {
652 slen += sprintf(output+slen, "word ");
653 } else if ( (*p)->opd[i] & BITS32 ) {
654 slen += sprintf(output+slen, "dword ");
655 } else if ( (*p)->opd[i] & NEAR ) {
656 slen += sprintf(output+slen, "near ");
657 } else if ( (*p)->opd[i] & SHORT ) {
658 slen += sprintf(output+slen, "short ");
660 slen += sprintf(output+slen, "0x%lx", ins.oprs[i].offset);
661 } else if ( !(MEM_OFFS & ~(*p)->opd[i]) ) {
662 slen += sprintf(output+slen, "[%s%s%s0x%lx]",
663 (segover ? segover : ""),
664 (segover ? ":" : ""),
665 (ins.oprs[i].addr_size == 32 ? "dword " :
666 ins.oprs[i].addr_size == 16 ? "word " : ""),
669 } else if ( !(REGMEM & ~(*p)->opd[i]) ) {
671 if ( (*p)->opd[i] & BITS8 )
672 slen += sprintf(output+slen, "byte ");
673 if ( (*p)->opd[i] & BITS16 )
674 slen += sprintf(output+slen, "word ");
675 if ( (*p)->opd[i] & BITS32 )
676 slen += sprintf(output+slen, "dword ");
677 if ( (*p)->opd[i] & BITS64 )
678 slen += sprintf(output+slen, "qword ");
679 if ( (*p)->opd[i] & BITS80 )
680 slen += sprintf(output+slen, "tword ");
681 if ( (*p)->opd[i] & FAR )
682 slen += sprintf(output+slen, "far ");
683 if ( (*p)->opd[i] & NEAR )
684 slen += sprintf(output+slen, "near ");
685 output[slen++] = '[';
686 if (ins.oprs[i].addr_size)
687 slen += sprintf(output+slen, "%s",
688 (ins.oprs[i].addr_size == 32 ? "dword " :
689 ins.oprs[i].addr_size == 16 ? "word " : ""));
691 slen += sprintf(output+slen, "%s:", segover);
694 if (ins.oprs[i].basereg != -1) {
695 slen += sprintf(output+slen, "%s",
696 reg_names[(ins.oprs[i].basereg -
700 if (ins.oprs[i].indexreg != -1) {
702 output[slen++] = '+';
703 slen += sprintf(output+slen, "%s",
704 reg_names[(ins.oprs[i].indexreg -
706 if (ins.oprs[i].scale > 1)
707 slen += sprintf(output+slen, "*%d", ins.oprs[i].scale);
710 if (ins.oprs[i].segment & SEG_DISP8) {
712 if (ins.oprs[i].offset & 0x80) {
713 ins.oprs[i].offset = - (signed char) ins.oprs[i].offset;
716 slen += sprintf(output+slen, "%c0x%lx", sign,
718 } else if (ins.oprs[i].segment & SEG_DISP16) {
720 output[slen++] = '+';
721 slen += sprintf(output+slen, "0x%lx", ins.oprs[i].offset);
722 } else if (ins.oprs[i].segment & SEG_DISP32) {
724 output[slen++] = '+';
725 slen += sprintf(output+slen, "0x%lx", ins.oprs[i].offset);
727 output[slen++] = ']';
729 slen += sprintf(output+slen, "<operand%d>", i);
733 if (segover) { /* unused segment override */
737 p[count+3] = p[count];
738 strncpy (output, segover, 2);
744 long eatbyte (unsigned char *data, char *output)
746 sprintf(output, "db 0x%02X", *data);