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
)
38 static int reg32
[] = {
39 R_EAX
, R_ECX
, R_EDX
, R_EBX
, R_ESP
, R_EBP
, R_ESI
, R_EDI
};
40 static int reg16
[] = {
41 R_AX
, R_CX
, R_DX
, R_BX
, R_SP
, R_BP
, R_SI
, R_DI
};
43 R_AL
, R_CL
, R_DL
, R_BL
, R_AH
, R_CH
, R_DH
, R_BH
};
45 R_ES
, R_CS
, R_SS
, R_DS
, R_FS
, R_GS
, 0, 0 };
47 R_CR0
, 0, R_CR2
, R_CR3
, R_CR4
, 0, 0, 0 };
49 R_DR0
, R_DR1
, R_DR2
, R_DR3
, 0, 0, R_DR6
, R_DR7
};
51 0, 0, 0, R_TR3
, R_TR4
, R_TR5
, R_TR6
, R_TR7
};
52 static int fpureg
[] = {
53 R_ST0
, R_ST1
, R_ST2
, R_ST3
, R_ST4
, R_ST5
, R_ST6
, R_ST7
};
54 static int mmxreg
[] = {
55 R_MM0
, R_MM1
, R_MM2
, R_MM3
, R_MM4
, R_MM5
, R_MM6
, R_MM7
};
57 if (!(REG_AL
& ~regflags
))
59 if (!(REG_AX
& ~regflags
))
61 if (!(REG_EAX
& ~regflags
))
63 if (!(REG_DX
& ~regflags
))
65 if (!(REG_CL
& ~regflags
))
67 if (!(REG_CX
& ~regflags
))
69 if (!(REG_ECX
& ~regflags
))
71 if (!(REG_CR4
& ~regflags
))
73 if (!(FPU0
& ~regflags
))
75 if (!(REG_CS
& ~regflags
))
77 if (!(REG_DESS
& ~regflags
))
78 return (regval
== 0 || regval
== 2 || regval
== 3 ? sreg
[regval
] : 0);
79 if (!(REG_FSGS
& ~regflags
))
80 return (regval
== 4 || regval
== 5 ? sreg
[regval
] : 0);
81 if (!((REGMEM
|BITS8
) & ~regflags
))
83 if (!((REGMEM
|BITS16
) & ~regflags
))
85 if (!((REGMEM
|BITS32
) & ~regflags
))
87 if (!(REG_SREG
& ~regflags
))
89 if (!(REG_CREG
& ~regflags
))
91 if (!(REG_DREG
& ~regflags
))
93 if (!(REG_TREG
& ~regflags
))
95 if (!(FPUREG
& ~regflags
))
96 return fpureg
[regval
];
97 if (!(MMXREG
& ~regflags
))
98 return mmxreg
[regval
];
102 static char *whichcond(int condval
)
104 static int conds
[] = {
105 C_O
, C_NO
, C_C
, C_NC
, C_Z
, C_NZ
, C_NA
, C_A
,
106 C_S
, C_NS
, C_PE
, C_PO
, C_L
, C_NL
, C_NG
, C_G
108 return conditions
[conds
[condval
]];
112 * Process an effective address (ModRM) specification.
114 static unsigned char *do_ea (unsigned char *data
, int modrm
, int asize
,
115 int segsize
, operand
*op
)
117 int mod
, rm
, scale
, index
, base
;
119 mod
= (modrm
>> 6) & 03;
122 if (mod
== 3) { /* pure register version */
124 op
->segment
|= SEG_RMREG
;
132 * <mod> specifies the displacement size (none, byte or
133 * word), and <rm> specifies the register combination.
134 * Exception: mod=0,rm=6 does not specify [BP] as one might
135 * expect, but instead specifies [disp16].
137 op
->indexreg
= op
->basereg
= -1;
138 op
->scale
= 1; /* always, in 16 bits */
140 case 0: op
->basereg
= R_BX
; op
->indexreg
= R_SI
; break;
141 case 1: op
->basereg
= R_BX
; op
->indexreg
= R_DI
; break;
142 case 2: op
->basereg
= R_BP
; op
->indexreg
= R_SI
; break;
143 case 3: op
->basereg
= R_BP
; op
->indexreg
= R_DI
; break;
144 case 4: op
->basereg
= R_SI
; break;
145 case 5: op
->basereg
= R_DI
; break;
146 case 6: op
->basereg
= R_BP
; break;
147 case 7: op
->basereg
= R_BX
; break;
149 if (rm
== 6 && mod
== 0) { /* special case */
153 mod
= 2; /* fake disp16 */
157 op
->segment
|= SEG_NODISP
;
160 op
->segment
|= SEG_DISP8
;
161 op
->offset
= (signed char) *data
++;
164 op
->segment
|= SEG_DISP16
;
165 op
->offset
= *data
++;
166 op
->offset
|= (*data
++) << 8;
172 * Once again, <mod> specifies displacement size (this time
173 * none, byte or *dword*), while <rm> specifies the base
174 * register. Again, [EBP] is missing, replaced by a pure
175 * disp32 (this time that's mod=0,rm=*5*). However, rm=4
176 * indicates not a single base register, but instead the
177 * presence of a SIB byte...
181 case 0: op
->basereg
= R_EAX
; break;
182 case 1: op
->basereg
= R_ECX
; break;
183 case 2: op
->basereg
= R_EDX
; break;
184 case 3: op
->basereg
= R_EBX
; break;
185 case 5: op
->basereg
= R_EBP
; break;
186 case 6: op
->basereg
= R_ESI
; break;
187 case 7: op
->basereg
= R_EDI
; break;
189 if (rm
== 5 && mod
== 0) {
193 mod
= 2; /* fake disp32 */
195 if (rm
== 4) { /* process SIB */
196 scale
= (*data
>> 6) & 03;
197 index
= (*data
>> 3) & 07;
201 op
->scale
= 1 << scale
;
203 case 0: op
->indexreg
= R_EAX
; break;
204 case 1: op
->indexreg
= R_ECX
; break;
205 case 2: op
->indexreg
= R_EDX
; break;
206 case 3: op
->indexreg
= R_EBX
; break;
207 case 4: op
->indexreg
= -1; break;
208 case 5: op
->indexreg
= R_EBP
; break;
209 case 6: op
->indexreg
= R_ESI
; break;
210 case 7: op
->indexreg
= R_EDI
; break;
214 case 0: op
->basereg
= R_EAX
; break;
215 case 1: op
->basereg
= R_ECX
; break;
216 case 2: op
->basereg
= R_EDX
; break;
217 case 3: op
->basereg
= R_EBX
; break;
218 case 4: op
->basereg
= R_ESP
; break;
219 case 6: op
->basereg
= R_ESI
; break;
220 case 7: op
->basereg
= R_EDI
; break;
232 op
->segment
|= SEG_NODISP
;
235 op
->segment
|= SEG_DISP8
;
236 op
->offset
= (signed char) *data
++;
239 op
->segment
|= SEG_DISP32
;
240 op
->offset
= *data
++;
241 op
->offset
|= (*data
++) << 8;
242 op
->offset
|= ((long) *data
++) << 16;
243 op
->offset
|= ((long) *data
++) << 24;
251 * Determine whether the code string in r corresponds to the data
252 * stream in data. Return the number of bytes matched if so.
254 static int matches (unsigned char *r
, unsigned char *data
, int asize
,
255 int osize
, int segsize
, insn
*ins
)
257 unsigned char * origdata
= data
;
258 int a_used
= FALSE
, o_used
= FALSE
;
263 if (c
>= 01 && c
<= 03) {
270 case 0x07: ins
->oprs
[0].basereg
= 0; break;
271 case 0x17: ins
->oprs
[0].basereg
= 2; break;
272 case 0x1F: ins
->oprs
[0].basereg
= 3; break;
273 default: return FALSE
;
278 case 0xA1: ins
->oprs
[0].basereg
= 4; break;
279 case 0xA9: ins
->oprs
[0].basereg
= 5; break;
280 default: return FALSE
;
285 case 0x06: ins
->oprs
[0].basereg
= 0; break;
286 case 0x0E: ins
->oprs
[0].basereg
= 1; break;
287 case 0x16: ins
->oprs
[0].basereg
= 2; break;
288 case 0x1E: ins
->oprs
[0].basereg
= 3; break;
289 default: return FALSE
;
294 case 0xA0: ins
->oprs
[0].basereg
= 4; break;
295 case 0xA8: ins
->oprs
[0].basereg
= 5; break;
296 default: return FALSE
;
299 if (c
>= 010 && c
<= 012) {
300 int t
= *r
++, d
= *data
++;
301 if (d
< t
|| d
> t
+7)
304 ins
->oprs
[c
-010].basereg
= d
-t
;
305 ins
->oprs
[c
-010].segment
|= SEG_RMREG
;
311 if (c
>= 014 && c
<= 016) {
312 ins
->oprs
[c
-014].offset
= (signed char) *data
++;
313 ins
->oprs
[c
-014].segment
|= SEG_SIGNED
;
315 if (c
>= 020 && c
<= 022)
316 ins
->oprs
[c
-020].offset
= *data
++;
317 if (c
>= 024 && c
<= 026)
318 ins
->oprs
[c
-024].offset
= *data
++;
319 if (c
>= 030 && c
<= 032) {
320 ins
->oprs
[c
-030].offset
= *data
++;
321 ins
->oprs
[c
-030].offset
|= (*data
++ << 8);
323 if (c
>= 034 && c
<= 036) {
324 ins
->oprs
[c
-034].offset
= *data
++;
325 ins
->oprs
[c
-034].offset
|= (*data
++ << 8);
327 ins
->oprs
[c
-034].offset
|= (((long) *data
++) << 16);
328 ins
->oprs
[c
-034].offset
|= (((long) *data
++) << 24);
330 if (segsize
!= asize
)
331 ins
->oprs
[c
-034].addr_size
= asize
;
333 if (c
>= 040 && c
<= 042) {
334 ins
->oprs
[c
-040].offset
= *data
++;
335 ins
->oprs
[c
-040].offset
|= (*data
++ << 8);
336 ins
->oprs
[c
-040].offset
|= (((long) *data
++) << 16);
337 ins
->oprs
[c
-040].offset
|= (((long) *data
++) << 24);
339 if (c
>= 050 && c
<= 052) {
340 ins
->oprs
[c
-050].offset
= (signed char) *data
++;
341 ins
->oprs
[c
-050].segment
|= SEG_RELATIVE
;
343 if (c
>= 060 && c
<= 062) {
344 ins
->oprs
[c
-060].offset
= *data
++;
345 ins
->oprs
[c
-060].offset
|= (*data
++ << 8);
346 ins
->oprs
[c
-060].segment
|= SEG_RELATIVE
;
347 ins
->oprs
[c
-060].segment
&= ~SEG_32BIT
;
349 if (c
>= 064 && c
<= 066) {
350 ins
->oprs
[c
-064].offset
= *data
++;
351 ins
->oprs
[c
-064].offset
|= (*data
++ << 8);
353 ins
->oprs
[c
-064].offset
|= (((long) *data
++) << 16);
354 ins
->oprs
[c
-064].offset
|= (((long) *data
++) << 24);
355 ins
->oprs
[c
-064].segment
|= SEG_32BIT
;
357 ins
->oprs
[c
-064].segment
&= ~SEG_32BIT
;
358 ins
->oprs
[c
-064].segment
|= SEG_RELATIVE
;
359 if (segsize
!= asize
)
360 ins
->oprs
[c
-064].addr_size
= asize
;
362 if (c
>= 070 && c
<= 072) {
363 ins
->oprs
[c
-070].offset
= *data
++;
364 ins
->oprs
[c
-070].offset
|= (*data
++ << 8);
365 ins
->oprs
[c
-070].offset
|= (((long) *data
++) << 16);
366 ins
->oprs
[c
-070].offset
|= (((long) *data
++) << 24);
367 ins
->oprs
[c
-070].segment
|= SEG_32BIT
| SEG_RELATIVE
;
369 if (c
>= 0100 && c
<= 0177) {
371 ins
->oprs
[c
& 07].basereg
= (modrm
>> 3) & 07;
372 ins
->oprs
[c
& 07].segment
|= SEG_RMREG
;
373 data
= do_ea (data
, modrm
, asize
, segsize
,
374 &ins
->oprs
[(c
>> 3) & 07]);
376 if (c
>= 0200 && c
<= 0277) {
378 if (((modrm
>> 3) & 07) != (c
& 07))
379 return FALSE
; /* spare field doesn't match up */
380 data
= do_ea (data
, modrm
, asize
, segsize
,
381 &ins
->oprs
[(c
>> 3) & 07]);
383 if (c
>= 0300 && c
<= 0302) {
385 ins
->oprs
[c
-0300].segment
|= SEG_32BIT
;
387 ins
->oprs
[c
-0300].segment
&= ~SEG_32BIT
;
403 if (asize
!= segsize
)
421 if (osize
!= segsize
)
427 int t
= *r
++, d
= *data
++;
428 if (d
< t
|| d
> t
+15)
431 ins
->condition
= d
- t
;
436 * Check for unused a/o prefixes.
439 if (!a_used
&& asize
!= segsize
)
440 ins
->prefixes
[ins
->nprefix
++] = (asize
== 16 ? P_A16
: P_A32
);
441 if (!o_used
&& osize
!= segsize
)
442 ins
->prefixes
[ins
->nprefix
++] = (osize
== 16 ? P_O16
: P_O32
);
444 return data
- origdata
;
447 long disasm (unsigned char *data
, char *output
, int segsize
, long offset
,
450 struct itemplate
**p
;
453 int rep
, lock
, asize
, osize
, i
, slen
, colon
;
454 unsigned char *origdata
;
461 asize
= osize
= segsize
;
466 if (*data
== 0xF3 || *data
== 0xF2)
468 else if (*data
== 0xF0)
470 else if (*data
== 0x2E || *data
== 0x36 || *data
== 0x3E ||
471 *data
== 0x26 || *data
== 0x64 || *data
== 0x65) {
473 case 0x2E: segover
= "cs"; break;
474 case 0x36: segover
= "ss"; break;
475 case 0x3E: segover
= "ds"; break;
476 case 0x26: segover
= "es"; break;
477 case 0x64: segover
= "fs"; break;
478 case 0x65: segover
= "gs"; break;
480 } else if (*data
== 0x66)
481 osize
= 48 - segsize
, data
++;
482 else if (*data
== 0x67)
483 asize
= 48 - segsize
, data
++;
488 ins
.oprs
[0].segment
= ins
.oprs
[1].segment
= ins
.oprs
[2].segment
=
489 ins
.oprs
[0].addr_size
= ins
.oprs
[1].addr_size
= ins
.oprs
[2].addr_size
=
490 (segsize
== 16 ? 0 : SEG_32BIT
);
493 for (p
= itable
[*data
]; *p
; p
++)
494 if ( (length
= matches((unsigned char *)((*p
)->code
), data
,
495 asize
, osize
, segsize
, &ins
)) )
499 * Final check to make sure the types of r/m match up.
501 for (i
= 0; i
< (*p
)->operands
; i
++)
504 /* If it's a mem-only EA but we have a register, die. */
505 ((ins
.oprs
[i
].segment
& SEG_RMREG
) &&
506 !(MEMORY
& ~(*p
)->opd
[i
])) ||
508 /* If it's a reg-only EA but we have a memory ref, die. */
509 (!(ins
.oprs
[i
].segment
& SEG_RMREG
) &&
510 !(REGNORM
& ~(*p
)->opd
[i
]) &&
511 !((*p
)->opd
[i
] & REG_SMASK
)) ||
513 /* Register type mismatch (eg FS vs REG_DESS): die. */
514 ((((*p
)->opd
[i
] & (REGISTER
| FPUREG
)) ||
515 (ins
.oprs
[i
].segment
& SEG_RMREG
)) &&
516 !whichreg ((*p
)->opd
[i
], ins
.oprs
[i
].basereg
)))
520 * FIXME: can we do a break here?
528 if (!length
|| !works
)
529 return 0; /* no instruction was matched */
534 slen
+= sprintf(output
+slen
, "rep%s ",
535 (rep
== 0xF2 ? "ne" :
536 (*p
)->opcode
== I_CMPSB
||
537 (*p
)->opcode
== I_CMPSW
||
538 (*p
)->opcode
== I_CMPSD
||
539 (*p
)->opcode
== I_SCASB
||
540 (*p
)->opcode
== I_SCASW
||
541 (*p
)->opcode
== I_SCASD
? "e" : ""));
544 slen
+= sprintf(output
+slen
, "lock ");
545 for (i
= 0; i
< ins
.nprefix
; i
++)
546 switch (ins
.prefixes
[i
]) {
547 case P_A16
: slen
+= sprintf(output
+slen
, "a16 "); break;
548 case P_A32
: slen
+= sprintf(output
+slen
, "a32 "); break;
549 case P_O16
: slen
+= sprintf(output
+slen
, "o16 "); break;
550 case P_O32
: slen
+= sprintf(output
+slen
, "o32 "); break;
553 for (i
= 0; i
< elements(ico
); i
++)
554 if ((*p
)->opcode
== ico
[i
]) {
555 slen
+= sprintf(output
+slen
, "%s%s", icn
[i
],
556 whichcond(ins
.condition
));
559 if (i
>= elements(ico
))
560 slen
+= sprintf(output
+slen
, "%s", insn_names
[(*p
)->opcode
]);
562 length
+= data
- origdata
; /* fix up for prefixes */
563 for (i
=0; i
<(*p
)->operands
; i
++) {
564 output
[slen
++] = (colon
? ':' : i
==0 ? ' ' : ',');
566 if (ins
.oprs
[i
].segment
& SEG_RELATIVE
) {
567 ins
.oprs
[i
].offset
+= offset
+ length
;
569 * sort out wraparound
571 if (!(ins
.oprs
[i
].segment
& SEG_32BIT
))
572 ins
.oprs
[i
].offset
&= 0xFFFF;
574 * add sync marker, if autosync is on
577 add_sync (ins
.oprs
[i
].offset
, 0L);
580 if ((*p
)->opd
[i
] & COLON
)
585 if (((*p
)->opd
[i
] & (REGISTER
| FPUREG
)) ||
586 (ins
.oprs
[i
].segment
& SEG_RMREG
))
588 ins
.oprs
[i
].basereg
= whichreg ((*p
)->opd
[i
],
589 ins
.oprs
[i
].basereg
);
590 if ( (*p
)->opd
[i
] & TO
)
591 slen
+= sprintf(output
+slen
, "to ");
592 slen
+= sprintf(output
+slen
, "%s",
593 reg_names
[ins
.oprs
[i
].basereg
-EXPR_REG_START
]);
594 } else if (!(UNITY
& ~(*p
)->opd
[i
])) {
595 output
[slen
++] = '1';
596 } else if ( (*p
)->opd
[i
] & IMMEDIATE
) {
597 if ( (*p
)->opd
[i
] & BITS8
) {
598 slen
+= sprintf(output
+slen
, "byte ");
599 if (ins
.oprs
[i
].segment
& SEG_SIGNED
) {
600 if (ins
.oprs
[i
].offset
< 0) {
601 ins
.oprs
[i
].offset
*= -1;
602 output
[slen
++] = '-';
604 output
[slen
++] = '+';
606 } else if ( (*p
)->opd
[i
] & BITS16
) {
607 slen
+= sprintf(output
+slen
, "word ");
608 } else if ( (*p
)->opd
[i
] & BITS32
) {
609 slen
+= sprintf(output
+slen
, "dword ");
610 } else if ( (*p
)->opd
[i
] & NEAR
) {
611 slen
+= sprintf(output
+slen
, "near ");
612 } else if ( (*p
)->opd
[i
] & SHORT
) {
613 slen
+= sprintf(output
+slen
, "short ");
615 slen
+= sprintf(output
+slen
, "0x%lx", ins
.oprs
[i
].offset
);
616 } else if ( !(MEM_OFFS
& ~(*p
)->opd
[i
]) ) {
617 slen
+= sprintf(output
+slen
, "[%s%s%s0x%lx]",
618 (segover
? segover
: ""),
619 (segover
? ":" : ""),
620 (ins
.oprs
[i
].addr_size
== 32 ? "dword " :
621 ins
.oprs
[i
].addr_size
== 16 ? "word " : ""),
624 } else if ( !(REGMEM
& ~(*p
)->opd
[i
]) ) {
626 if ( (*p
)->opd
[i
] & BITS8
)
627 slen
+= sprintf(output
+slen
, "byte ");
628 if ( (*p
)->opd
[i
] & BITS16
)
629 slen
+= sprintf(output
+slen
, "word ");
630 if ( (*p
)->opd
[i
] & BITS32
)
631 slen
+= sprintf(output
+slen
, "dword ");
632 if ( (*p
)->opd
[i
] & BITS64
)
633 slen
+= sprintf(output
+slen
, "qword ");
634 if ( (*p
)->opd
[i
] & BITS80
)
635 slen
+= sprintf(output
+slen
, "tword ");
636 if ( (*p
)->opd
[i
] & FAR
)
637 slen
+= sprintf(output
+slen
, "far ");
638 if ( (*p
)->opd
[i
] & NEAR
)
639 slen
+= sprintf(output
+slen
, "near ");
640 output
[slen
++] = '[';
641 if (ins
.oprs
[i
].addr_size
)
642 slen
+= sprintf(output
+slen
, "%s",
643 (ins
.oprs
[i
].addr_size
== 32 ? "dword " :
644 ins
.oprs
[i
].addr_size
== 16 ? "word " : ""));
646 slen
+= sprintf(output
+slen
, "%s:", segover
);
649 if (ins
.oprs
[i
].basereg
!= -1) {
650 slen
+= sprintf(output
+slen
, "%s",
651 reg_names
[(ins
.oprs
[i
].basereg
-
655 if (ins
.oprs
[i
].indexreg
!= -1) {
657 output
[slen
++] = '+';
658 slen
+= sprintf(output
+slen
, "%s",
659 reg_names
[(ins
.oprs
[i
].indexreg
-
661 if (ins
.oprs
[i
].scale
> 1)
662 slen
+= sprintf(output
+slen
, "*%d", ins
.oprs
[i
].scale
);
665 if (ins
.oprs
[i
].segment
& SEG_DISP8
) {
667 if (ins
.oprs
[i
].offset
& 0x80) {
668 ins
.oprs
[i
].offset
= - (signed char) ins
.oprs
[i
].offset
;
671 slen
+= sprintf(output
+slen
, "%c0x%lx", sign
,
673 } else if (ins
.oprs
[i
].segment
& SEG_DISP16
) {
675 output
[slen
++] = '+';
676 slen
+= sprintf(output
+slen
, "0x%lx", ins
.oprs
[i
].offset
);
677 } else if (ins
.oprs
[i
].segment
& SEG_DISP32
) {
679 output
[slen
++] = '+';
680 slen
+= sprintf(output
+slen
, "0x%lx", ins
.oprs
[i
].offset
);
682 output
[slen
++] = ']';
684 slen
+= sprintf(output
+slen
, "<operand%d>", i
);
688 if (segover
) { /* unused segment override */
692 p
[count
+3] = p
[count
];
693 strncpy (output
, segover
, 2);
699 long eatbyte (unsigned char *data
, char *output
)
701 sprintf(output
, "db 0x%02X", *data
);