NASM 0.98.16
[nasm.git] / assemble.c
blobf560112e9ab8494904e84b117d57b1a0c9a8c459
1 /* assemble.c code generation for the Netwide Assembler
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 * the actual codes (C syntax, i.e. octal):
9 * \0 - terminates the code. (Unless it's a literal of course.)
10 * \1, \2, \3 - that many literal bytes follow in the code stream
11 * \4, \6 - the POP/PUSH (respectively) codes for CS, DS, ES, SS
12 * (POP is never used for CS) depending on operand 0
13 * \5, \7 - the second byte of POP/PUSH codes for FS, GS, depending
14 * on operand 0
15 * \10, \11, \12 - a literal byte follows in the code stream, to be added
16 * to the register value of operand 0, 1 or 2
17 * \17 - encodes the literal byte 0. (Some compilers don't take
18 * kindly to a zero byte in the _middle_ of a compile time
19 * string constant, so I had to put this hack in.)
20 * \14, \15, \16 - a signed byte immediate operand, from operand 0, 1 or 2
21 * \20, \21, \22 - a byte immediate operand, from operand 0, 1 or 2
22 * \24, \25, \26 - an unsigned byte immediate operand, from operand 0, 1 or 2
23 * \30, \31, \32 - a word immediate operand, from operand 0, 1 or 2
24 * \34, \35, \36 - select between \3[012] and \4[012] depending on 16/32 bit
25 * assembly mode or the address-size override on the operand
26 * \37 - a word constant, from the _segment_ part of operand 0
27 * \40, \41, \42 - a long immediate operand, from operand 0, 1 or 2
28 * \50, \51, \52 - a byte relative operand, from operand 0, 1 or 2
29 * \60, \61, \62 - a word relative operand, from operand 0, 1 or 2
30 * \64, \65, \66 - select between \6[012] and \7[012] depending on 16/32 bit
31 * assembly mode or the address-size override on the operand
32 * \70, \71, \72 - a long relative operand, from operand 0, 1 or 2
33 * \1ab - a ModRM, calculated on EA in operand a, with the spare
34 * field the register value of operand b.
35 * \130,\131,\132 - an immediate word or signed byte for operand 0, 1, or 2
36 * \133,\134,\135 - or 2 (s-field) into next opcode byte if operand 0, 1, or 2
37 * is a signed byte rather than a word.
38 * \140,\141,\142 - an immediate dword or signed byte for operand 0, 1, or 2
39 * \143,\144,\145 - or 2 (s-field) into next opcode byte if operand 0, 1, or 2
40 * is a signed byte rather than a dword.
41 * \2ab - a ModRM, calculated on EA in operand a, with the spare
42 * field equal to digit b.
43 * \30x - might be an 0x67 byte, depending on the address size of
44 * the memory reference in operand x.
45 * \310 - indicates fixed 16-bit address size, i.e. optional 0x67.
46 * \311 - indicates fixed 32-bit address size, i.e. optional 0x67.
47 * \312 - (disassembler only) marker on LOOP, LOOPxx instructions.
48 * \320 - indicates fixed 16-bit operand size, i.e. optional 0x66.
49 * \321 - indicates fixed 32-bit operand size, i.e. optional 0x66.
50 * \322 - indicates that this instruction is only valid when the
51 * operand size is the default (instruction to disassembler,
52 * generates no code in the assembler)
53 * \330 - a literal byte follows in the code stream, to be added
54 * to the condition code value of the instruction.
55 * \331 - instruction not valid with REP prefix. Hint for
56 * disassembler only; for SSE instructions.
57 * \332 - disassemble a rep (0xF3 byte) prefix as repe not rep.
58 * \333 - REP prefix (0xF3 byte); for SSE instructions. Not encoded
59 * as a literal byte in order to aid the disassembler.
60 * \340 - reserve <operand 0> bytes of uninitialised storage.
61 * Operand 0 had better be a segmentless constant.
62 * \370,\371,\372 - match only if operand 0 meets byte jump criteria.
63 * 370 is used for Jcc, 371 is used for JMP.
64 * \373 - assemble 0x03 if bits==16, 0x05 if bits==32;
65 * used for conditional jump over longer jump
68 #include <stdio.h>
69 #include <string.h>
71 #include "nasm.h"
72 #include "nasmlib.h"
73 #include "assemble.h"
74 #include "insns.h"
76 extern struct itemplate *nasm_instructions[];
78 typedef struct {
79 int sib_present; /* is a SIB byte necessary? */
80 int bytes; /* # of bytes of offset needed */
81 int size; /* lazy - this is sib+bytes+1 */
82 unsigned char modrm, sib; /* the bytes themselves */
83 } ea;
85 static unsigned long cpu; /* cpu level received from nasm.c */
86 static efunc errfunc;
87 static struct ofmt *outfmt;
88 static ListGen *list;
90 static long calcsize (long, long, int, insn *, char *);
91 static void gencode (long, long, int, insn *, char *, long);
92 static int regval (operand *o);
93 static int matches (struct itemplate *, insn *);
94 static ea * process_ea (operand *, ea *, int, int, int);
95 static int chsize (operand *, int);
98 * This routine wrappers the real output format's output routine,
99 * in order to pass a copy of the data off to the listing file
100 * generator at the same time.
102 static void out (long offset, long segto, void *data, unsigned long type,
103 long segment, long wrt)
105 long lineno;
106 char *lnfname = NULL;
108 if ((type & OUT_TYPMASK) == OUT_ADDRESS) {
109 if (segment != NO_SEG || wrt != NO_SEG) {
111 * This address is relocated. We must write it as
112 * OUT_ADDRESS, so there's no work to be done here.
114 list->output (offset, data, type);
116 else {
117 unsigned char p[4], *q = p;
119 * This is a non-relocated address, and we're going to
120 * convert it into RAWDATA format.
122 if ((type & OUT_SIZMASK) == 4) {
123 WRITELONG (q, * (long *) data);
124 list->output (offset, p, OUT_RAWDATA+4);
126 else {
127 WRITESHORT (q, * (long *) data);
128 list->output (offset, p, OUT_RAWDATA+2);
132 else if ((type & OUT_TYPMASK) == OUT_RAWDATA) {
133 list->output (offset, data, type);
135 else if ((type & OUT_TYPMASK) == OUT_RESERVE) {
136 list->output (offset, NULL, type);
138 else if ((type & OUT_TYPMASK) == OUT_REL2ADR ||
139 (type & OUT_TYPMASK) == OUT_REL4ADR) {
140 list->output (offset, data, type);
143 if (src_get(&lineno,&lnfname))
145 outfmt->current_dfmt->linenum(lnfname,lineno,segto);
146 if (lnfname) nasm_free(lnfname);
149 outfmt->output (segto, data, type, segment, wrt);
152 static int jmp_match (long segment, long offset, int bits,
153 insn *ins, char *code)
154 { long isize;
155 unsigned char c = code[0];
158 if (c != 0370 && c != 0371) return 0;
159 if (ins->oprs[0].opflags & OPFLAG_FORWARD) {
160 if (optimizing<0 && c==0370) return 1;
161 else return (pass0==0); /* match a forward reference */
163 isize = calcsize (segment, offset, bits, ins, code);
164 if (ins->oprs[0].segment != segment) return 0;
165 isize = ins->oprs[0].offset - offset - isize; /* isize is now the delta */
166 if (isize >= -128L && isize <= 127L) return 1; /* it is byte size */
168 return 0;
172 long assemble (long segment, long offset, int bits, unsigned long cp,
173 insn *instruction, struct ofmt *output, efunc error,
174 ListGen *listgen)
176 struct itemplate *temp;
177 int j;
178 int size_prob;
179 long insn_end;
180 long itimes;
181 long start = offset;
182 long wsize = 0; /* size for DB etc. */
184 errfunc = error; /* to pass to other functions */
185 cpu = cp;
186 outfmt = output; /* likewise */
187 list = listgen; /* and again */
189 switch (instruction->opcode)
191 case -1: return 0;
192 case I_DB: wsize = 1; break;
193 case I_DW: wsize = 2; break;
194 case I_DD: wsize = 4; break;
195 case I_DQ: wsize = 8; break;
196 case I_DT: wsize = 10; break;
199 if (wsize) {
200 extop * e;
201 long t = instruction->times;
202 if (t < 0)
203 errfunc(ERR_PANIC, "instruction->times < 0 (%ld) in assemble()",t);
205 while (t--) /* repeat TIMES times */
207 for (e = instruction->eops; e; e = e->next)
209 if (e->type == EOT_DB_NUMBER)
211 if (wsize == 1) {
212 if (e->segment != NO_SEG)
213 errfunc (ERR_NONFATAL,
214 "one-byte relocation attempted");
215 else {
216 unsigned char out_byte = e->offset;
217 out (offset, segment, &out_byte, OUT_RAWDATA+1,
218 NO_SEG, NO_SEG);
221 else if (wsize > 5) {
222 errfunc (ERR_NONFATAL, "integer supplied to a D%c"
223 " instruction", wsize==8 ? 'Q' : 'T');
225 else
226 out (offset, segment, &e->offset,
227 OUT_ADDRESS+wsize, e->segment,
228 e->wrt);
229 offset += wsize;
231 else if (e->type == EOT_DB_STRING)
233 int align;
235 out (offset, segment, e->stringval,
236 OUT_RAWDATA+e->stringlen, NO_SEG, NO_SEG);
237 align = e->stringlen % wsize;
239 if (align) {
240 align = wsize - align;
241 out (offset, segment, "\0\0\0\0\0\0\0\0",
242 OUT_RAWDATA+align, NO_SEG, NO_SEG);
244 offset += e->stringlen + align;
247 if (t > 0 && t == instruction->times-1)
250 * Dummy call to list->output to give the offset to the
251 * listing module.
253 list->output (offset, NULL, OUT_RAWDATA);
254 list->uplevel (LIST_TIMES);
257 if (instruction->times > 1)
258 list->downlevel (LIST_TIMES);
259 return offset - start;
262 if (instruction->opcode == I_INCBIN)
264 static char fname[FILENAME_MAX];
265 FILE * fp;
266 long len;
268 len = FILENAME_MAX-1;
269 if (len > instruction->eops->stringlen)
270 len = instruction->eops->stringlen;
271 strncpy (fname, instruction->eops->stringval, len);
272 fname[len] = '\0';
274 if ( (fp = fopen(fname, "rb")) == NULL)
275 error (ERR_NONFATAL, "`incbin': unable to open file `%s'", fname);
276 else if (fseek(fp, 0L, SEEK_END) < 0)
277 error (ERR_NONFATAL, "`incbin': unable to seek on file `%s'",
278 fname);
279 else
281 static char buf[2048];
282 long t = instruction->times;
283 long base = 0;
285 len = ftell (fp);
286 if (instruction->eops->next) {
287 base = instruction->eops->next->offset;
288 len -= base;
289 if (instruction->eops->next->next &&
290 len > instruction->eops->next->next->offset)
291 len = instruction->eops->next->next->offset;
294 * Dummy call to list->output to give the offset to the
295 * listing module.
297 list->output (offset, NULL, OUT_RAWDATA);
298 list->uplevel(LIST_INCBIN);
299 while (t--)
301 long l;
303 fseek (fp, base, SEEK_SET);
304 l = len;
305 while (l > 0) {
306 long m = fread (buf, 1, (l>sizeof(buf)?sizeof(buf):l),
307 fp);
308 if (!m) {
310 * This shouldn't happen unless the file
311 * actually changes while we are reading
312 * it.
314 error (ERR_NONFATAL, "`incbin': unexpected EOF while"
315 " reading file `%s'", fname);
316 t=0; /* Try to exit cleanly */
317 break;
319 out (offset, segment, buf, OUT_RAWDATA+m,
320 NO_SEG, NO_SEG);
321 l -= m;
324 list->downlevel(LIST_INCBIN);
325 if (instruction->times > 1) {
327 * Dummy call to list->output to give the offset to the
328 * listing module.
330 list->output (offset, NULL, OUT_RAWDATA);
331 list->uplevel(LIST_TIMES);
332 list->downlevel(LIST_TIMES);
334 fclose (fp);
335 return instruction->times * len;
337 return 0; /* if we're here, there's an error */
340 size_prob = FALSE;
341 temp = nasm_instructions[instruction->opcode];
342 while (temp->opcode != -1) {
343 int m = matches (temp, instruction);
344 if (m == 99)
345 m += jmp_match(segment, offset, bits, instruction, temp->code);
347 if (m == 100) /* matches! */
349 char *codes = temp->code;
350 long insn_size = calcsize(segment, offset, bits,
351 instruction, codes);
352 itimes = instruction->times;
353 if (insn_size < 0) /* shouldn't be, on pass two */
354 error (ERR_PANIC, "errors made it through from pass one");
355 else while (itimes--) {
356 insn_end = offset + insn_size;
357 for (j=0; j<instruction->nprefix; j++) {
358 unsigned char c=0;
359 switch (instruction->prefixes[j]) {
360 case P_LOCK:
361 c = 0xF0; break;
362 case P_REPNE: case P_REPNZ:
363 c = 0xF2; break;
364 case P_REPE: case P_REPZ: case P_REP:
365 c = 0xF3; break;
366 case R_CS: c = 0x2E; break;
367 case R_DS: c = 0x3E; break;
368 case R_ES: c = 0x26; break;
369 case R_FS: c = 0x64; break;
370 case R_GS: c = 0x65; break;
371 case R_SS: c = 0x36; break;
372 case P_A16:
373 if (bits != 16)
374 c = 0x67;
375 break;
376 case P_A32:
377 if (bits != 32)
378 c = 0x67;
379 break;
380 case P_O16:
381 if (bits != 16)
382 c = 0x66;
383 break;
384 case P_O32:
385 if (bits != 32)
386 c = 0x66;
387 break;
388 default:
389 error (ERR_PANIC,
390 "invalid instruction prefix");
392 if (c != 0) {
393 out (offset, segment, &c, OUT_RAWDATA+1,
394 NO_SEG, NO_SEG);
395 offset++;
398 gencode (segment, offset, bits, instruction, codes, insn_end);
399 offset += insn_size;
400 if (itimes > 0 && itimes == instruction->times-1) {
402 * Dummy call to list->output to give the offset to the
403 * listing module.
405 list->output (offset, NULL, OUT_RAWDATA);
406 list->uplevel (LIST_TIMES);
409 if (instruction->times > 1)
410 list->downlevel (LIST_TIMES);
411 return offset - start;
412 } else if (m > 0 && m > size_prob) {
413 size_prob = m;
415 temp++;
418 if (temp->opcode == -1) { /* didn't match any instruction */
419 if (size_prob == 1) /* would have matched, but for size */
420 error (ERR_NONFATAL, "operation size not specified");
421 else if (size_prob == 2)
422 error (ERR_NONFATAL, "mismatch in operand sizes");
423 else if (size_prob == 3)
424 error (ERR_NONFATAL, "no instruction for this cpu level");
425 else
426 error (ERR_NONFATAL,
427 "invalid combination of opcode and operands");
429 return 0;
432 long insn_size (long segment, long offset, int bits, unsigned long cp,
433 insn *instruction, efunc error)
435 struct itemplate *temp;
437 errfunc = error; /* to pass to other functions */
438 cpu = cp;
440 if (instruction->opcode == -1)
441 return 0;
443 if (instruction->opcode == I_DB ||
444 instruction->opcode == I_DW ||
445 instruction->opcode == I_DD ||
446 instruction->opcode == I_DQ ||
447 instruction->opcode == I_DT)
449 extop *e;
450 long isize, osize, wsize = 0; /* placate gcc */
452 isize = 0;
453 switch (instruction->opcode)
455 case I_DB: wsize = 1; break;
456 case I_DW: wsize = 2; break;
457 case I_DD: wsize = 4; break;
458 case I_DQ: wsize = 8; break;
459 case I_DT: wsize = 10; break;
462 for (e = instruction->eops; e; e = e->next)
464 long align;
466 osize = 0;
467 if (e->type == EOT_DB_NUMBER)
468 osize = 1;
469 else if (e->type == EOT_DB_STRING)
470 osize = e->stringlen;
472 align = (-osize) % wsize;
473 if (align < 0)
474 align += wsize;
475 isize += osize + align;
477 return isize * instruction->times;
480 if (instruction->opcode == I_INCBIN)
482 char fname[FILENAME_MAX];
483 FILE * fp;
484 long len;
486 len = FILENAME_MAX-1;
487 if (len > instruction->eops->stringlen)
488 len = instruction->eops->stringlen;
489 strncpy (fname, instruction->eops->stringval, len);
490 fname[len] = '\0';
491 if ( (fp = fopen(fname, "rb")) == NULL )
492 error (ERR_NONFATAL, "`incbin': unable to open file `%s'", fname);
493 else if (fseek(fp, 0L, SEEK_END) < 0)
494 error (ERR_NONFATAL, "`incbin': unable to seek on file `%s'",
495 fname);
496 else
498 len = ftell (fp);
499 fclose (fp);
500 if (instruction->eops->next)
502 len -= instruction->eops->next->offset;
503 if (instruction->eops->next->next &&
504 len > instruction->eops->next->next->offset)
506 len = instruction->eops->next->next->offset;
509 return instruction->times * len;
511 return 0; /* if we're here, there's an error */
514 temp = nasm_instructions[instruction->opcode];
515 while (temp->opcode != -1) {
516 int m = matches(temp, instruction);
517 if (m == 99)
518 m += jmp_match(segment, offset, bits, instruction, temp->code);
520 if (m == 100) {
521 /* we've matched an instruction. */
522 long isize;
523 char * codes = temp->code;
524 int j;
526 isize = calcsize(segment, offset, bits, instruction, codes);
527 if (isize < 0)
528 return -1;
529 for (j = 0; j < instruction->nprefix; j++)
531 if ((instruction->prefixes[j] != P_A16 &&
532 instruction->prefixes[j] != P_O16 && bits==16) ||
533 (instruction->prefixes[j] != P_A32 &&
534 instruction->prefixes[j] != P_O32 && bits==32))
536 isize++;
539 return isize * instruction->times;
541 temp++;
543 return -1; /* didn't match any instruction */
547 /* check that opn[op] is a signed byte of size 16 or 32,
548 and return the signed value*/
549 static int is_sbyte (insn *ins, int op, int size)
551 signed long v;
552 int ret;
554 ret = !(ins->forw_ref && ins->oprs[op].opflags ) && /* dead in the water on forward reference or External */
555 (optimizing>0 || !(ins->oprs[op].type & (BITS16|BITS32))) &&
556 ins->oprs[op].wrt==NO_SEG && ins->oprs[op].segment==NO_SEG;
558 v = ins->oprs[op].offset;
559 if (size==16) v = (signed short)v; /* sign extend if 16 bits */
561 return ret && v>=-128L && v<=127L;
564 static long calcsize (long segment, long offset, int bits,
565 insn *ins, char *codes)
567 long length = 0;
568 unsigned char c;
570 (void) segment; /* Don't warn that this parameter is unused */
571 (void) offset; /* Don't warn that this parameter is unused */
573 while (*codes) switch (c = *codes++) {
574 case 01: case 02: case 03:
575 codes += c, length += c; break;
576 case 04: case 05: case 06: case 07:
577 length++; break;
578 case 010: case 011: case 012:
579 codes++, length++; break;
580 case 017:
581 length++; break;
582 case 014: case 015: case 016:
583 length++; break;
584 case 020: case 021: case 022:
585 length++; break;
586 case 024: case 025: case 026:
587 length++; break;
588 case 030: case 031: case 032:
589 length += 2; break;
590 case 034: case 035: case 036:
591 length += ((ins->oprs[c-034].addr_size ?
592 ins->oprs[c-034].addr_size : bits) == 16 ? 2 : 4); break;
593 case 037:
594 length += 2; break;
595 case 040: case 041: case 042:
596 length += 4; break;
597 case 050: case 051: case 052:
598 length++; break;
599 case 060: case 061: case 062:
600 length += 2; break;
601 case 064: case 065: case 066:
602 length += ((ins->oprs[c-064].addr_size ?
603 ins->oprs[c-064].addr_size : bits) == 16 ? 2 : 4); break;
604 case 070: case 071: case 072:
605 length += 4; break;
606 case 0130: case 0131: case 0132:
607 length += is_sbyte(ins, c-0130, 16) ? 1 : 2; break;
608 case 0133: case 0134: case 0135:
609 codes+=2; length++; break;
610 case 0140: case 0141: case 0142:
611 length += is_sbyte(ins, c-0140, 32) ? 1 : 4; break;
612 case 0143: case 0144: case 0145:
613 codes+=2; length++; break;
614 case 0300: case 0301: case 0302:
615 length += chsize (&ins->oprs[c-0300], bits);
616 break;
617 case 0310:
618 length += (bits==32);
619 break;
620 case 0311:
621 length += (bits==16);
622 break;
623 case 0312:
624 break;
625 case 0320:
626 length += (bits==32);
627 break;
628 case 0321:
629 length += (bits==16);
630 break;
631 case 0322:
632 break;
633 case 0330:
634 codes++, length++; break;
635 case 0331:
636 case 0332:
637 break;
638 case 0333:
639 length++; break;
640 case 0340: case 0341: case 0342:
641 if (ins->oprs[0].segment != NO_SEG)
642 errfunc (ERR_NONFATAL, "attempt to reserve non-constant"
643 " quantity of BSS space");
644 else
645 length += ins->oprs[0].offset << (c-0340);
646 break;
647 case 0370: case 0371: case 0372:
648 break;
649 case 0373:
650 length++; break;
651 default: /* can't do it by 'case' statements */
652 if (c>=0100 && c<=0277) { /* it's an EA */
653 ea ea_data;
654 if (!process_ea (&ins->oprs[(c>>3)&7], &ea_data, bits, 0,
655 ins->forw_ref)) {
656 errfunc (ERR_NONFATAL, "invalid effective address");
657 return -1;
658 } else
659 length += ea_data.size;
660 } else
661 errfunc (ERR_PANIC, "internal instruction table corrupt"
662 ": instruction code 0x%02X given", c);
664 return length;
667 static void gencode (long segment, long offset, int bits,
668 insn *ins, char *codes, long insn_end)
670 static char condval[] = { /* conditional opcodes */
671 0x7, 0x3, 0x2, 0x6, 0x2, 0x4, 0xF, 0xD, 0xC, 0xE, 0x6, 0x2,
672 0x3, 0x7, 0x3, 0x5, 0xE, 0xC, 0xD, 0xF, 0x1, 0xB, 0x9, 0x5,
673 0x0, 0xA, 0xA, 0xB, 0x8, 0x4
675 unsigned char c;
676 unsigned char bytes[4];
677 long data, size;
679 while (*codes)
680 switch (c = *codes++)
682 case 01: case 02: case 03:
683 out (offset, segment, codes, OUT_RAWDATA+c, NO_SEG, NO_SEG);
684 codes += c;
685 offset += c;
686 break;
688 case 04: case 06:
689 switch (ins->oprs[0].basereg)
691 case R_CS:
692 bytes[0] = 0x0E + (c == 0x04 ? 1 : 0); break;
693 case R_DS:
694 bytes[0] = 0x1E + (c == 0x04 ? 1 : 0); break;
695 case R_ES:
696 bytes[0] = 0x06 + (c == 0x04 ? 1 : 0); break;
697 case R_SS:
698 bytes[0] = 0x16 + (c == 0x04 ? 1 : 0); break;
699 default:
700 errfunc (ERR_PANIC, "bizarre 8086 segment register received");
702 out (offset, segment, bytes, OUT_RAWDATA+1, NO_SEG, NO_SEG);
703 offset++;
704 break;
706 case 05: case 07:
707 switch (ins->oprs[0].basereg) {
708 case R_FS: bytes[0] = 0xA0 + (c == 0x05 ? 1 : 0); break;
709 case R_GS: bytes[0] = 0xA8 + (c == 0x05 ? 1 : 0); break;
710 default:
711 errfunc (ERR_PANIC, "bizarre 386 segment register received");
713 out (offset, segment, bytes, OUT_RAWDATA+1, NO_SEG, NO_SEG);
714 offset++;
715 break;
717 case 010: case 011: case 012:
718 bytes[0] = *codes++ + regval(&ins->oprs[c-010]);
719 out (offset, segment, bytes, OUT_RAWDATA+1, NO_SEG, NO_SEG);
720 offset += 1;
721 break;
723 case 017:
724 bytes[0] = 0;
725 out (offset, segment, bytes, OUT_RAWDATA+1, NO_SEG, NO_SEG);
726 offset += 1;
727 break;
729 case 014: case 015: case 016:
730 if (ins->oprs[c-014].offset < -128
731 || ins->oprs[c-014].offset > 127)
733 errfunc (ERR_WARNING, "signed byte value exceeds bounds");
736 if (ins->oprs[c-014].segment != NO_SEG)
738 data = ins->oprs[c-014].offset;
739 out (offset, segment, &data, OUT_ADDRESS+1,
740 ins->oprs[c-014].segment, ins->oprs[c-014].wrt);
742 else {
743 bytes[0] = ins->oprs[c-014].offset;
744 out (offset, segment, bytes, OUT_RAWDATA+1, NO_SEG, NO_SEG);
746 offset += 1;
747 break;
749 case 020: case 021: case 022:
750 if (ins->oprs[c-020].offset < -256
751 || ins->oprs[c-020].offset > 255)
753 errfunc (ERR_WARNING, "byte value exceeds bounds");
755 if (ins->oprs[c-020].segment != NO_SEG) {
756 data = ins->oprs[c-020].offset;
757 out (offset, segment, &data, OUT_ADDRESS+1,
758 ins->oprs[c-020].segment, ins->oprs[c-020].wrt);
760 else {
761 bytes[0] = ins->oprs[c-020].offset;
762 out (offset, segment, bytes, OUT_RAWDATA+1, NO_SEG, NO_SEG);
764 offset += 1;
765 break;
767 case 024: case 025: case 026:
768 if (ins->oprs[c-024].offset < 0 || ins->oprs[c-024].offset > 255)
769 errfunc (ERR_WARNING, "unsigned byte value exceeds bounds");
770 if (ins->oprs[c-024].segment != NO_SEG) {
771 data = ins->oprs[c-024].offset;
772 out (offset, segment, &data, OUT_ADDRESS+1,
773 ins->oprs[c-024].segment, ins->oprs[c-024].wrt);
775 else {
776 bytes[0] = ins->oprs[c-024].offset;
777 out (offset, segment, bytes, OUT_RAWDATA+1, NO_SEG, NO_SEG);
779 offset += 1;
780 break;
782 case 030: case 031: case 032:
783 if (ins->oprs[c-030].segment == NO_SEG &&
784 ins->oprs[c-030].wrt == NO_SEG &&
785 (ins->oprs[c-030].offset < -65536L ||
786 ins->oprs[c-030].offset > 65535L))
788 errfunc (ERR_WARNING, "word value exceeds bounds");
790 data = ins->oprs[c-030].offset;
791 out (offset, segment, &data, OUT_ADDRESS+2,
792 ins->oprs[c-030].segment, ins->oprs[c-030].wrt);
793 offset += 2;
794 break;
796 case 034: case 035: case 036:
797 data = ins->oprs[c-034].offset;
798 size = ((ins->oprs[c-034].addr_size ?
799 ins->oprs[c-034].addr_size : bits) == 16 ? 2 : 4);
800 if (size==2 && (data < -65536L || data > 65535L))
801 errfunc (ERR_WARNING, "word value exceeds bounds");
802 out (offset, segment, &data, OUT_ADDRESS+size,
803 ins->oprs[c-034].segment, ins->oprs[c-034].wrt);
804 offset += size;
805 break;
807 case 037:
808 if (ins->oprs[0].segment == NO_SEG)
809 errfunc (ERR_NONFATAL, "value referenced by FAR is not"
810 " relocatable");
811 data = 0L;
812 out (offset, segment, &data, OUT_ADDRESS+2,
813 outfmt->segbase(1+ins->oprs[0].segment),
814 ins->oprs[0].wrt);
815 offset += 2;
816 break;
818 case 040: case 041: case 042:
819 data = ins->oprs[c-040].offset;
820 out (offset, segment, &data, OUT_ADDRESS+4,
821 ins->oprs[c-040].segment, ins->oprs[c-040].wrt);
822 offset += 4;
823 break;
825 case 050: case 051: case 052:
826 if (ins->oprs[c-050].segment != segment)
827 errfunc (ERR_NONFATAL, "short relative jump outside segment");
828 data = ins->oprs[c-050].offset - insn_end;
829 if (data > 127 || data < -128)
830 errfunc (ERR_NONFATAL, "short jump is out of range");
831 bytes[0] = data;
832 out (offset, segment, bytes, OUT_RAWDATA+1, NO_SEG, NO_SEG);
833 offset += 1;
834 break;
836 case 060: case 061: case 062:
837 if (ins->oprs[c-060].segment != segment) {
838 data = ins->oprs[c-060].offset;
839 out (offset, segment, &data, OUT_REL2ADR+insn_end-offset,
840 ins->oprs[c-060].segment, ins->oprs[c-060].wrt);
841 } else {
842 data = ins->oprs[c-060].offset - insn_end;
843 out (offset, segment, &data,
844 OUT_ADDRESS+2, NO_SEG, NO_SEG);
846 offset += 2;
847 break;
849 case 064: case 065: case 066:
850 size = ((ins->oprs[c-064].addr_size ?
851 ins->oprs[c-064].addr_size : bits) == 16 ? 2 : 4);
852 if (ins->oprs[c-064].segment != segment) {
853 data = ins->oprs[c-064].offset;
854 size = (bits == 16 ? OUT_REL2ADR : OUT_REL4ADR);
855 out (offset, segment, &data, size+insn_end-offset,
856 ins->oprs[c-064].segment, ins->oprs[c-064].wrt);
857 size = (bits == 16 ? 2 : 4);
858 } else {
859 data = ins->oprs[c-064].offset - insn_end;
860 out (offset, segment, &data,
861 OUT_ADDRESS+size, NO_SEG, NO_SEG);
863 offset += size;
864 break;
866 case 070: case 071: case 072:
867 if (ins->oprs[c-070].segment != segment) {
868 data = ins->oprs[c-070].offset;
869 out (offset, segment, &data, OUT_REL4ADR+insn_end-offset,
870 ins->oprs[c-070].segment, ins->oprs[c-070].wrt);
871 } else {
872 data = ins->oprs[c-070].offset - insn_end;
873 out (offset, segment, &data,
874 OUT_ADDRESS+4, NO_SEG, NO_SEG);
876 offset += 4;
877 break;
879 case 0130: case 0131: case 0132:
880 data = ins->oprs[c-0130].offset;
881 if (is_sbyte(ins, c-0130, 16)) {
882 out (offset, segment, &data, OUT_RAWDATA+1, NO_SEG, NO_SEG);
883 offset++;
884 } else {
885 if (ins->oprs[c-0130].segment == NO_SEG &&
886 ins->oprs[c-0130].wrt == NO_SEG &&
887 (data < -65536L || data > 65535L)) {
888 errfunc (ERR_WARNING, "word value exceeds bounds");
890 out (offset, segment, &data, OUT_ADDRESS+2,
891 ins->oprs[c-0130].segment, ins->oprs[c-0130].wrt);
892 offset += 2;
894 break;
896 case 0133: case 0134: case 0135:
897 codes++;
898 bytes[0] = *codes++;
899 if (is_sbyte(ins, c-0133, 16)) bytes[0] |= 2; /* s-bit */
900 out (offset, segment, bytes, OUT_RAWDATA+1, NO_SEG, NO_SEG);
901 offset++;
902 break;
904 case 0140: case 0141: case 0142:
905 data = ins->oprs[c-0140].offset;
906 if (is_sbyte(ins, c-0140, 32)) {
907 out (offset, segment, &data, OUT_RAWDATA+1, NO_SEG, NO_SEG);
908 offset++;
909 } else {
910 out (offset, segment, &data, OUT_ADDRESS+4,
911 ins->oprs[c-0140].segment, ins->oprs[c-0140].wrt);
912 offset += 4;
914 break;
916 case 0143: case 0144: case 0145:
917 codes++;
918 bytes[0] = *codes++;
919 if (is_sbyte(ins, c-0143, 32)) bytes[0] |= 2; /* s-bit */
920 out (offset, segment, bytes, OUT_RAWDATA+1, NO_SEG, NO_SEG);
921 offset++;
922 break;
924 case 0300: case 0301: case 0302:
925 if (chsize (&ins->oprs[c-0300], bits)) {
926 *bytes = 0x67;
927 out (offset, segment, bytes,
928 OUT_RAWDATA+1, NO_SEG, NO_SEG);
929 offset += 1;
930 } else
931 offset += 0;
932 break;
934 case 0310:
935 if (bits==32) {
936 *bytes = 0x67;
937 out (offset, segment, bytes,
938 OUT_RAWDATA+1, NO_SEG, NO_SEG);
939 offset += 1;
940 } else
941 offset += 0;
942 break;
944 case 0311:
945 if (bits==16) {
946 *bytes = 0x67;
947 out (offset, segment, bytes,
948 OUT_RAWDATA+1, NO_SEG, NO_SEG);
949 offset += 1;
950 } else
951 offset += 0;
952 break;
954 case 0312:
955 break;
957 case 0320:
958 if (bits==32) {
959 *bytes = 0x66;
960 out (offset, segment, bytes,
961 OUT_RAWDATA+1, NO_SEG, NO_SEG);
962 offset += 1;
963 } else
964 offset += 0;
965 break;
967 case 0321:
968 if (bits==16) {
969 *bytes = 0x66;
970 out (offset, segment, bytes,
971 OUT_RAWDATA+1, NO_SEG, NO_SEG);
972 offset += 1;
973 } else
974 offset += 0;
975 break;
977 case 0322:
978 break;
980 case 0330:
981 *bytes = *codes++ ^ condval[ins->condition];
982 out (offset, segment, bytes,
983 OUT_RAWDATA+1, NO_SEG, NO_SEG);
984 offset += 1;
985 break;
987 case 0331:
988 case 0332:
989 break;
991 case 0333:
992 *bytes = 0xF3;
993 out (offset, segment, bytes,
994 OUT_RAWDATA+1, NO_SEG, NO_SEG);
995 offset += 1;
996 break;
998 case 0340: case 0341: case 0342:
999 if (ins->oprs[0].segment != NO_SEG)
1000 errfunc (ERR_PANIC, "non-constant BSS size in pass two");
1001 else {
1002 long size = ins->oprs[0].offset << (c-0340);
1003 if (size > 0)
1004 out (offset, segment, NULL,
1005 OUT_RESERVE+size, NO_SEG, NO_SEG);
1006 offset += size;
1008 break;
1010 case 0370: case 0371: case 0372:
1011 break;
1013 case 0373:
1014 *bytes = bits==16 ? 3 : 5;
1015 out (offset, segment, bytes,
1016 OUT_RAWDATA+1, NO_SEG, NO_SEG);
1017 offset += 1;
1018 break;
1020 default: /* can't do it by 'case' statements */
1021 if (c>=0100 && c<=0277) { /* it's an EA */
1022 ea ea_data;
1023 int rfield;
1024 unsigned char *p;
1025 long s;
1027 if (c<=0177) /* pick rfield from operand b */
1028 rfield = regval (&ins->oprs[c&7]);
1029 else /* rfield is constant */
1030 rfield = c & 7;
1032 if (!process_ea (&ins->oprs[(c>>3)&7], &ea_data, bits, rfield,
1033 ins->forw_ref))
1035 errfunc (ERR_NONFATAL, "invalid effective address");
1038 p = bytes;
1039 *p++ = ea_data.modrm;
1040 if (ea_data.sib_present)
1041 *p++ = ea_data.sib;
1043 s = p-bytes;
1044 out (offset, segment, bytes, OUT_RAWDATA + s,
1045 NO_SEG, NO_SEG);
1047 switch (ea_data.bytes) {
1048 case 0:
1049 break;
1050 case 1:
1051 if (ins->oprs[(c>>3)&7].segment != NO_SEG) {
1052 data = ins->oprs[(c>>3)&7].offset;
1053 out (offset, segment, &data, OUT_ADDRESS+1,
1054 ins->oprs[(c>>3)&7].segment,
1055 ins->oprs[(c>>3)&7].wrt);
1056 } else {
1057 *bytes = ins->oprs[(c>>3)&7].offset;
1058 out (offset, segment, bytes, OUT_RAWDATA+1,
1059 NO_SEG, NO_SEG);
1061 s++;
1062 break;
1063 case 2:
1064 case 4:
1065 data = ins->oprs[(c>>3)&7].offset;
1066 out (offset, segment, &data,
1067 OUT_ADDRESS+ea_data.bytes,
1068 ins->oprs[(c>>3)&7].segment, ins->oprs[(c>>3)&7].wrt);
1069 s += ea_data.bytes;
1070 break;
1072 offset += s;
1073 } else
1074 errfunc (ERR_PANIC, "internal instruction table corrupt"
1075 ": instruction code 0x%02X given", c);
1079 static int regval (operand *o)
1081 switch (o->basereg) {
1082 case R_EAX: case R_AX: case R_AL: case R_ES: case R_CR0: case R_DR0:
1083 case R_ST0: case R_MM0: case R_XMM0:
1084 return 0;
1085 case R_ECX: case R_CX: case R_CL: case R_CS: case R_DR1: case R_ST1:
1086 case R_MM1: case R_XMM1:
1087 return 1;
1088 case R_EDX: case R_DX: case R_DL: case R_SS: case R_CR2: case R_DR2:
1089 case R_ST2: case R_MM2: case R_XMM2:
1090 return 2;
1091 case R_EBX: case R_BX: case R_BL: case R_DS: case R_CR3: case R_DR3:
1092 case R_TR3: case R_ST3: case R_MM3: case R_XMM3:
1093 return 3;
1094 case R_ESP: case R_SP: case R_AH: case R_FS: case R_CR4: case R_TR4:
1095 case R_ST4: case R_MM4: case R_XMM4:
1096 return 4;
1097 case R_EBP: case R_BP: case R_CH: case R_GS: case R_TR5: case R_ST5:
1098 case R_MM5: case R_XMM5:
1099 return 5;
1100 case R_ESI: case R_SI: case R_DH: case R_DR6: case R_TR6: case R_ST6:
1101 case R_MM6: case R_XMM6:
1102 return 6;
1103 case R_EDI: case R_DI: case R_BH: case R_DR7: case R_TR7: case R_ST7:
1104 case R_MM7: case R_XMM7:
1105 return 7;
1106 default: /* panic */
1107 errfunc (ERR_PANIC, "invalid register operand given to regval()");
1108 return 0;
1112 static int matches (struct itemplate *itemp, insn *instruction)
1114 int i, size[3], asize, oprs, ret;
1116 ret = 100;
1119 * Check the opcode
1121 if (itemp->opcode != instruction->opcode) return 0;
1124 * Count the operands
1126 if (itemp->operands != instruction->operands) return 0;
1129 * Check that no spurious colons or TOs are present
1131 for (i=0; i<itemp->operands; i++)
1132 if (instruction->oprs[i].type & ~itemp->opd[i] & (COLON|TO))
1133 return 0;
1136 * Check that the operand flags all match up
1138 for (i=0; i<itemp->operands; i++)
1139 if (itemp->opd[i] & ~instruction->oprs[i].type ||
1140 ((itemp->opd[i] & SIZE_MASK) &&
1141 ((itemp->opd[i] ^ instruction->oprs[i].type) & SIZE_MASK)))
1143 if ((itemp->opd[i] & ~instruction->oprs[i].type & NON_SIZE) ||
1144 (instruction->oprs[i].type & SIZE_MASK))
1145 return 0;
1146 else
1147 /* ret = 1; */
1148 return 1;
1152 * Check operand sizes
1154 if (itemp->flags & IF_ARMASK) {
1155 size[0] = size[1] = size[2] = 0;
1157 switch (itemp->flags & IF_ARMASK) {
1158 case IF_AR0: i = 0; break;
1159 case IF_AR1: i = 1; break;
1160 case IF_AR2: i = 2; break;
1161 default: break; /* Shouldn't happen */
1163 if (itemp->flags & IF_SB) {
1164 size[i] = BITS8;
1165 } else if (itemp->flags & IF_SW) {
1166 size[i] = BITS16;
1167 } else if (itemp->flags & IF_SD) {
1168 size[i] = BITS32;
1170 } else {
1171 asize = 0;
1172 if (itemp->flags & IF_SB) {
1173 asize = BITS8;
1174 oprs = itemp->operands;
1175 } else if (itemp->flags & IF_SW) {
1176 asize = BITS16;
1177 oprs = itemp->operands;
1178 } else if (itemp->flags & IF_SD) {
1179 asize = BITS32;
1180 oprs = itemp->operands;
1182 size[0] = size[1] = size[2] = asize;
1185 if (itemp->flags & (IF_SM | IF_SM2)) {
1186 oprs = (itemp->flags & IF_SM2 ? 2 : itemp->operands);
1187 asize = 0;
1188 for (i=0; i<oprs; i++) {
1189 if ( (asize = itemp->opd[i] & SIZE_MASK) != 0) {
1190 int j;
1191 for (j=0; j<oprs; j++)
1192 size[j] = asize;
1193 break;
1196 } else {
1197 oprs = itemp->operands;
1200 for (i=0; i<itemp->operands; i++)
1201 if (!(itemp->opd[i] & SIZE_MASK) &&
1202 (instruction->oprs[i].type & SIZE_MASK & ~size[i]))
1203 /* ret = 2; */
1204 return 2;
1207 * Check template is okay at the set cpu level
1209 if ((itemp->flags & IF_PLEVEL) > cpu) return 3;
1212 * Check if special handling needed for Jumps
1214 if ((unsigned char)(itemp->code[0]) >= 0370) return 99;
1216 return ret;
1219 static ea *process_ea (operand *input, ea *output, int addrbits, int rfield,
1220 int forw_ref)
1222 if (!(REGISTER & ~input->type)) { /* it's a single register */
1223 static int regs[] = {
1224 R_AL, R_CL, R_DL, R_BL, R_AH, R_CH, R_DH, R_BH,
1225 R_AX, R_CX, R_DX, R_BX, R_SP, R_BP, R_SI, R_DI,
1226 R_EAX, R_ECX, R_EDX, R_EBX, R_ESP, R_EBP, R_ESI, R_EDI,
1227 R_MM0, R_MM1, R_MM2, R_MM3, R_MM4, R_MM5, R_MM6, R_MM7,
1228 R_XMM0, R_XMM1, R_XMM2, R_XMM3, R_XMM4, R_XMM5, R_XMM6, R_XMM7
1230 int i;
1232 for (i=0; i<elements(regs); i++)
1233 if (input->basereg == regs[i]) break;
1234 if (i<elements(regs)) {
1235 output->sib_present = FALSE;/* no SIB necessary */
1236 output->bytes = 0; /* no offset necessary either */
1237 output->modrm = 0xC0 | (rfield << 3) | (i & 7);
1239 else
1240 return NULL;
1241 } else { /* it's a memory reference */
1242 if (input->basereg==-1 && (input->indexreg==-1 || input->scale==0)) {
1243 /* it's a pure offset */
1244 if (input->addr_size)
1245 addrbits = input->addr_size;
1246 output->sib_present = FALSE;
1247 output->bytes = (addrbits==32 ? 4 : 2);
1248 output->modrm = (addrbits==32 ? 5 : 6) | (rfield << 3);
1250 else { /* it's an indirection */
1251 int i=input->indexreg, b=input->basereg, s=input->scale;
1252 long o=input->offset, seg=input->segment;
1253 int hb=input->hintbase, ht=input->hinttype;
1254 int t;
1256 if (s==0) i = -1; /* make this easy, at least */
1258 if (i==R_EAX || i==R_EBX || i==R_ECX || i==R_EDX
1259 || i==R_EBP || i==R_ESP || i==R_ESI || i==R_EDI
1260 || b==R_EAX || b==R_EBX || b==R_ECX || b==R_EDX
1261 || b==R_EBP || b==R_ESP || b==R_ESI || b==R_EDI) {
1262 /* it must be a 32-bit memory reference. Firstly we have
1263 * to check that all registers involved are type Exx. */
1264 if (i!=-1 && i!=R_EAX && i!=R_EBX && i!=R_ECX && i!=R_EDX
1265 && i!=R_EBP && i!=R_ESP && i!=R_ESI && i!=R_EDI)
1266 return NULL;
1267 if (b!=-1 && b!=R_EAX && b!=R_EBX && b!=R_ECX && b!=R_EDX
1268 && b!=R_EBP && b!=R_ESP && b!=R_ESI && b!=R_EDI)
1269 return NULL;
1271 /* While we're here, ensure the user didn't specify WORD. */
1272 if (input->addr_size == 16)
1273 return NULL;
1275 /* now reorganise base/index */
1276 if (s == 1 && b != i && b != -1 && i != -1 &&
1277 ((hb==b&&ht==EAH_NOTBASE) || (hb==i&&ht==EAH_MAKEBASE)))
1278 t = b, b = i, i = t; /* swap if hints say so */
1279 if (b==i) /* convert EAX+2*EAX to 3*EAX */
1280 b = -1, s++;
1281 if (b==-1 && s==1 && !(hb == i && ht == EAH_NOTBASE))
1282 b = i, i = -1; /* make single reg base, unless hint */
1283 if (((s==2 && i!=R_ESP && !(input->eaflags & EAF_TIMESTWO)) ||
1284 s==3 || s==5 || s==9) && b==-1)
1285 b = i, s--; /* convert 3*EAX to EAX+2*EAX */
1286 if (s==1 && i==R_ESP) /* swap ESP into base if scale is 1 */
1287 i = b, b = R_ESP;
1288 if (i==R_ESP || (s!=1 && s!=2 && s!=4 && s!=8 && i!=-1))
1289 return NULL; /* wrong, for various reasons */
1291 if (i==-1 && b!=R_ESP) {/* no SIB needed */
1292 int mod, rm;
1293 switch(b) {
1294 case R_EAX: rm = 0; break;
1295 case R_ECX: rm = 1; break;
1296 case R_EDX: rm = 2; break;
1297 case R_EBX: rm = 3; break;
1298 case R_EBP: rm = 5; break;
1299 case R_ESI: rm = 6; break;
1300 case R_EDI: rm = 7; break;
1301 case -1: rm = 5; break;
1302 default: /* should never happen */
1303 return NULL;
1305 if (b==-1 || (b!=R_EBP && o==0 &&
1306 seg==NO_SEG && !forw_ref &&
1307 !(input->eaflags &
1308 (EAF_BYTEOFFS|EAF_WORDOFFS))))
1309 mod = 0;
1310 else if (input->eaflags & EAF_BYTEOFFS ||
1311 (o>=-128 && o<=127 && seg==NO_SEG && !forw_ref &&
1312 !(input->eaflags & EAF_WORDOFFS))) {
1313 mod = 1;
1315 else
1316 mod = 2;
1318 output->sib_present = FALSE;
1319 output->bytes = (b==-1 || mod==2 ? 4 : mod);
1320 output->modrm = (mod<<6) | (rfield<<3) | rm;
1322 else { /* we need a SIB */
1323 int mod, scale, index, base;
1325 switch (b) {
1326 case R_EAX: base = 0; break;
1327 case R_ECX: base = 1; break;
1328 case R_EDX: base = 2; break;
1329 case R_EBX: base = 3; break;
1330 case R_ESP: base = 4; break;
1331 case R_EBP: case -1: base = 5; break;
1332 case R_ESI: base = 6; break;
1333 case R_EDI: base = 7; break;
1334 default: /* then what the smeg is it? */
1335 return NULL; /* panic */
1338 switch (i) {
1339 case R_EAX: index = 0; break;
1340 case R_ECX: index = 1; break;
1341 case R_EDX: index = 2; break;
1342 case R_EBX: index = 3; break;
1343 case -1: index = 4; break;
1344 case R_EBP: index = 5; break;
1345 case R_ESI: index = 6; break;
1346 case R_EDI: index = 7; break;
1347 default: /* then what the smeg is it? */
1348 return NULL; /* panic */
1351 if (i==-1) s = 1;
1352 switch (s) {
1353 case 1: scale = 0; break;
1354 case 2: scale = 1; break;
1355 case 4: scale = 2; break;
1356 case 8: scale = 3; break;
1357 default: /* then what the smeg is it? */
1358 return NULL; /* panic */
1361 if (b==-1 || (b!=R_EBP && o==0 &&
1362 seg==NO_SEG && !forw_ref &&
1363 !(input->eaflags &
1364 (EAF_BYTEOFFS|EAF_WORDOFFS))))
1365 mod = 0;
1366 else if (input->eaflags & EAF_BYTEOFFS ||
1367 (o>=-128 && o<=127 && seg==NO_SEG && !forw_ref &&
1368 !(input->eaflags & EAF_WORDOFFS)))
1369 mod = 1;
1370 else
1371 mod = 2;
1373 output->sib_present = TRUE;
1374 output->bytes = (b==-1 || mod==2 ? 4 : mod);
1375 output->modrm = (mod<<6) | (rfield<<3) | 4;
1376 output->sib = (scale<<6) | (index<<3) | base;
1379 else { /* it's 16-bit */
1380 int mod, rm;
1382 /* check all registers are BX, BP, SI or DI */
1383 if ((b!=-1 && b!=R_BP && b!=R_BX && b!=R_SI && b!=R_DI) ||
1384 (i!=-1 && i!=R_BP && i!=R_BX && i!=R_SI && i!=R_DI))
1385 return NULL;
1387 /* ensure the user didn't specify DWORD */
1388 if (input->addr_size == 32)
1389 return NULL;
1391 if (s!=1 && i!=-1) return NULL;/* no can do, in 16-bit EA */
1392 if (b==-1 && i!=-1) b ^= i ^= b ^= i; /* swap them round */
1393 if ((b==R_SI || b==R_DI) && i!=-1)
1394 b ^= i ^= b ^= i; /* have BX/BP as base, SI/DI index */
1395 if (b==i) return NULL;/* shouldn't ever happen, in theory */
1396 if (i!=-1 && b!=-1 &&
1397 (i==R_BP || i==R_BX || b==R_SI || b==R_DI))
1398 return NULL; /* invalid combinations */
1399 if (b==-1) /* pure offset: handled above */
1400 return NULL; /* so if it gets to here, panic! */
1402 rm = -1;
1403 if (i!=-1)
1404 switch (i*256 + b) {
1405 case R_SI*256+R_BX: rm=0; break;
1406 case R_DI*256+R_BX: rm=1; break;
1407 case R_SI*256+R_BP: rm=2; break;
1408 case R_DI*256+R_BP: rm=3; break;
1410 else
1411 switch (b) {
1412 case R_SI: rm=4; break;
1413 case R_DI: rm=5; break;
1414 case R_BP: rm=6; break;
1415 case R_BX: rm=7; break;
1417 if (rm==-1) /* can't happen, in theory */
1418 return NULL; /* so panic if it does */
1420 if (o==0 && seg==NO_SEG && !forw_ref && rm!=6 &&
1421 !(input->eaflags & (EAF_BYTEOFFS|EAF_WORDOFFS)))
1422 mod = 0;
1423 else if (input->eaflags & EAF_BYTEOFFS ||
1424 (o>=-128 && o<=127 && seg==NO_SEG && !forw_ref &&
1425 !(input->eaflags & EAF_WORDOFFS)))
1426 mod = 1;
1427 else
1428 mod = 2;
1430 output->sib_present = FALSE; /* no SIB - it's 16-bit */
1431 output->bytes = mod; /* bytes of offset needed */
1432 output->modrm = (mod<<6) | (rfield<<3) | rm;
1436 output->size = 1 + output->sib_present + output->bytes;
1437 return output;
1440 static int chsize (operand *input, int addrbits)
1442 if (!(MEMORY & ~input->type)) {
1443 int i=input->indexreg, b=input->basereg;
1445 if (input->scale==0) i = -1;
1447 if (i == -1 && b == -1) /* pure offset */
1448 return (input->addr_size != 0 && input->addr_size != addrbits);
1450 if (i==R_EAX || i==R_EBX || i==R_ECX || i==R_EDX
1451 || i==R_EBP || i==R_ESP || i==R_ESI || i==R_EDI
1452 || b==R_EAX || b==R_EBX || b==R_ECX || b==R_EDX
1453 || b==R_EBP || b==R_ESP || b==R_ESI || b==R_EDI)
1454 return (addrbits==16);
1455 else
1456 return (addrbits==32);
1458 else
1459 return 0;