NASM 0.98p7
[nasm.git] / parser.c
blob1a9c9e4e8c64dd92cddc8f7245d4e828fa3b000b
1 /* parser.c source line parser 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 * initial version 27/iii/95 by Simon Tatham
9 */
11 #include <stdio.h>
12 #include <stdlib.h>
13 #include <stddef.h>
14 #include <string.h>
15 #include <ctype.h>
17 #include "nasm.h"
18 #include "nasmlib.h"
19 #include "parser.h"
20 #include "float.h"
22 static long reg_flags[] = { /* sizes and special flags */
23 0, REG8, REG_AL, REG_AX, REG8, REG8, REG16, REG16, REG8, REG_CL,
24 REG_CREG, REG_CREG, REG_CREG, REG_CR4, REG_CS, REG_CX, REG8,
25 REG16, REG8, REG_DREG, REG_DREG, REG_DREG, REG_DREG, REG_DREG,
26 REG_DREG, REG_DESS, REG_DX, REG_EAX, REG32, REG32, REG_ECX,
27 REG32, REG32, REG_DESS, REG32, REG32, REG_FSGS, REG_FSGS,
28 MMXREG, MMXREG, MMXREG, MMXREG, MMXREG, MMXREG, MMXREG, MMXREG,
29 REG16, REG16, REG_DESS, FPU0, FPUREG, FPUREG, FPUREG, FPUREG,
30 FPUREG, FPUREG, FPUREG, REG_TREG, REG_TREG, REG_TREG, REG_TREG,
31 REG_TREG,
32 XMMREG, XMMREG, XMMREG, XMMREG, XMMREG, XMMREG, XMMREG, XMMREG
35 enum { /* special tokens */
36 S_BYTE, S_DWORD, S_FAR, S_LONG, S_NEAR, S_NOSPLIT, S_QWORD,
37 S_SHORT, S_TO, S_TWORD, S_WORD
40 static int is_comma_next (void);
42 static int i;
43 static struct tokenval tokval;
44 static efunc error;
45 static struct ofmt *outfmt; /* Structure of addresses of output routines */
46 static loc_t *location; /* Pointer to current line's segment,offset */
48 void parser_global_info (struct ofmt *output, loc_t *locp)
50 outfmt = output;
51 location = locp;
54 insn *parse_line (int pass, char *buffer, insn *result,
55 efunc errfunc, evalfunc evaluate, ldfunc ldef)
57 int operand;
58 int critical;
59 struct eval_hints hints;
61 result->forw_ref = FALSE;
62 error = errfunc;
64 stdscan_reset();
65 stdscan_bufptr = buffer;
66 i = stdscan(NULL, &tokval);
68 result->label = NULL; /* Assume no label */
69 result->eops = NULL; /* must do this, whatever happens */
70 result->operands = 0; /* must initialise this */
72 if (i==0) { /* blank line - ignore */
73 result->opcode = -1; /* and no instruction either */
74 return result;
76 if (i != TOKEN_ID && i != TOKEN_INSN && i != TOKEN_PREFIX &&
77 (i!=TOKEN_REG || (REG_SREG & ~reg_flags[tokval.t_integer]))) {
78 error (ERR_NONFATAL, "label or instruction expected"
79 " at start of line");
80 result->opcode = -1;
81 return result;
84 if (i == TOKEN_ID) { /* there's a label here */
85 result->label = tokval.t_charptr;
86 i = stdscan(NULL, &tokval);
87 if (i == ':') { /* skip over the optional colon */
88 i = stdscan(NULL, &tokval);
89 } else if (i == 0) {
90 error (ERR_WARNING|ERR_WARN_OL|ERR_PASS1,
91 "label alone on a line without a colon might be in error");
93 if (i != TOKEN_INSN || tokval.t_integer != I_EQU)
96 * FIXME: location->segment could be NO_SEG, in which case
97 * it is possible we should be passing 'abs_seg'. Look into this.
98 * Work out whether that is *really* what we should be doing.
99 * Generally fix things. I think this is right as it is, but
100 * am still not certain.
102 ldef (result->label, location->segment,
103 location->offset, NULL, TRUE, FALSE, outfmt, errfunc);
107 if (i==0) {
108 result->opcode = -1; /* this line contains just a label */
109 return result;
112 result->nprefix = 0;
113 result->times = 1L;
115 while (i == TOKEN_PREFIX ||
116 (i==TOKEN_REG && !(REG_SREG & ~reg_flags[tokval.t_integer])))
119 * Handle special case: the TIMES prefix.
121 if (i == TOKEN_PREFIX && tokval.t_integer == P_TIMES) {
122 expr *value;
124 i = stdscan(NULL, &tokval);
125 value = evaluate (stdscan, NULL, &tokval, NULL, pass, error, NULL);
126 i = tokval.t_type;
127 if (!value) { /* but, error in evaluator */
128 result->opcode = -1; /* unrecoverable parse error: */
129 return result; /* ignore this instruction */
131 if (!is_simple (value)) {
132 error (ERR_NONFATAL,
133 "non-constant argument supplied to TIMES");
134 result->times = 1L;
135 } else {
136 result->times = value->value;
137 if (value->value < 0) {
138 error(ERR_NONFATAL, "TIMES value %d is negative",
139 value->value);
140 result->times = 0;
143 } else {
144 if (result->nprefix == MAXPREFIX)
145 error (ERR_NONFATAL,
146 "instruction has more than %d prefixes", MAXPREFIX);
147 else
148 result->prefixes[result->nprefix++] = tokval.t_integer;
149 i = stdscan(NULL, &tokval);
153 if (i != TOKEN_INSN) {
154 if (result->nprefix > 0 && i == 0) {
156 * Instruction prefixes are present, but no actual
157 * instruction. This is allowed: at this point we
158 * invent a notional instruction of RESB 0.
160 result->opcode = I_RESB;
161 result->operands = 1;
162 result->oprs[0].type = IMMEDIATE;
163 result->oprs[0].offset = 0L;
164 result->oprs[0].segment = result->oprs[0].wrt = NO_SEG;
165 return result;
166 } else {
167 error (ERR_NONFATAL, "parser: instruction expected");
168 result->opcode = -1;
169 return result;
173 result->opcode = tokval.t_integer;
174 result->condition = tokval.t_inttwo;
177 * RESB, RESW and RESD cannot be satisfied with incorrectly
178 * evaluated operands, since the correct values _must_ be known
179 * on the first pass. Hence, even in pass one, we set the
180 * `critical' flag on calling evaluate(), so that it will bomb
181 * out on undefined symbols. Nasty, but there's nothing we can
182 * do about it.
184 * For the moment, EQU has the same difficulty, so we'll
185 * include that.
187 if (result->opcode == I_RESB ||
188 result->opcode == I_RESW ||
189 result->opcode == I_RESD ||
190 result->opcode == I_RESQ ||
191 result->opcode == I_REST ||
192 result->opcode == I_EQU)
194 critical = pass;
196 else
197 critical = (pass==2 ? 2 : 0);
199 if (result->opcode == I_DB ||
200 result->opcode == I_DW ||
201 result->opcode == I_DD ||
202 result->opcode == I_DQ ||
203 result->opcode == I_DT ||
204 result->opcode == I_INCBIN)
206 extop *eop, **tail = &result->eops, **fixptr;
207 int oper_num = 0;
209 result->eops_float = FALSE;
212 * Begin to read the DB/DW/DD/DQ/DT/INCBIN operands.
214 while (1) {
215 i = stdscan(NULL, &tokval);
216 if (i == 0)
217 break;
218 fixptr = tail;
219 eop = *tail = nasm_malloc(sizeof(extop));
220 tail = &eop->next;
221 eop->next = NULL;
222 eop->type = EOT_NOTHING;
223 oper_num++;
225 if (i == TOKEN_NUM && tokval.t_charptr && is_comma_next()) {
226 eop->type = EOT_DB_STRING;
227 eop->stringval = tokval.t_charptr;
228 eop->stringlen = tokval.t_inttwo;
229 i = stdscan(NULL, &tokval); /* eat the comma */
230 continue;
233 if ((i == TOKEN_FLOAT && is_comma_next()) || i == '-') {
234 long sign = +1L;
236 if (i == '-') {
237 char *save = stdscan_bufptr;
238 i = stdscan(NULL, &tokval);
239 sign = -1L;
240 if (i != TOKEN_FLOAT || !is_comma_next()) {
241 stdscan_bufptr = save;
242 i = tokval.t_type = '-';
246 if (i == TOKEN_FLOAT) {
247 eop->type = EOT_DB_STRING;
248 result->eops_float = TRUE;
249 if (result->opcode == I_DD)
250 eop->stringlen = 4;
251 else if (result->opcode == I_DQ)
252 eop->stringlen = 8;
253 else if (result->opcode == I_DT)
254 eop->stringlen = 10;
255 else {
256 error(ERR_NONFATAL, "floating-point constant"
257 " encountered in `D%c' instruction",
258 result->opcode == I_DW ? 'W' : 'B');
260 * fix suggested by Pedro Gimeno... original line
261 * was:
262 * eop->type = EOT_NOTHING;
264 eop->stringlen = 0;
266 eop = nasm_realloc(eop, sizeof(extop)+eop->stringlen);
267 tail = &eop->next;
268 *fixptr = eop;
269 eop->stringval = (char *)eop + sizeof(extop);
270 if (eop->stringlen < 4 ||
271 !float_const (tokval.t_charptr, sign,
272 (unsigned char *)eop->stringval,
273 eop->stringlen, error))
274 eop->type = EOT_NOTHING;
275 i = stdscan(NULL, &tokval); /* eat the comma */
276 continue;
280 /* anything else */
282 expr *value;
283 value = evaluate (stdscan, NULL, &tokval, NULL,
284 critical, error, NULL);
285 i = tokval.t_type;
286 if (!value) { /* error in evaluator */
287 result->opcode = -1;/* unrecoverable parse error: */
288 return result; /* ignore this instruction */
290 if (is_unknown(value)) {
291 eop->type = EOT_DB_NUMBER;
292 eop->offset = 0; /* doesn't matter what we put */
293 eop->segment = eop->wrt = NO_SEG; /* likewise */
294 } else if (is_reloc(value)) {
295 eop->type = EOT_DB_NUMBER;
296 eop->offset = reloc_value(value);
297 eop->segment = reloc_seg(value);
298 eop->wrt = reloc_wrt(value);
299 } else {
300 error (ERR_NONFATAL,
301 "operand %d: expression is not simple"
302 " or relocatable", oper_num);
307 * We're about to call stdscan(), which will eat the
308 * comma that we're currently sitting on between
309 * arguments. However, we'd better check first that it
310 * _is_ a comma.
312 if (i == 0) /* also could be EOL */
313 break;
314 if (i != ',') {
315 error (ERR_NONFATAL, "comma expected after operand %d",
316 oper_num);
317 result->opcode = -1;/* unrecoverable parse error: */
318 return result; /* ignore this instruction */
322 if (result->opcode == I_INCBIN) {
324 * Correct syntax for INCBIN is that there should be
325 * one string operand, followed by one or two numeric
326 * operands.
328 if (!result->eops || result->eops->type != EOT_DB_STRING)
329 error (ERR_NONFATAL, "`incbin' expects a file name");
330 else if (result->eops->next &&
331 result->eops->next->type != EOT_DB_NUMBER)
332 error (ERR_NONFATAL, "`incbin': second parameter is",
333 " non-numeric");
334 else if (result->eops->next && result->eops->next->next &&
335 result->eops->next->next->type != EOT_DB_NUMBER)
336 error (ERR_NONFATAL, "`incbin': third parameter is",
337 " non-numeric");
338 else if (result->eops->next && result->eops->next->next &&
339 result->eops->next->next->next)
340 error (ERR_NONFATAL, "`incbin': more than three parameters");
341 else
342 return result;
344 * If we reach here, one of the above errors happened.
345 * Throw the instruction away.
347 result->opcode = -1;
348 return result;
349 } else /* DB ... */
350 if (oper_num == 0)
351 error (ERR_WARNING|ERR_PASS1,
352 "no operand for data declaration");
353 else
354 result->operands = oper_num;
356 return result;
359 /* right. Now we begin to parse the operands. There may be up to three
360 * of these, separated by commas, and terminated by a zero token. */
362 for (operand = 0; operand < 3; operand++) {
363 expr *value; /* used most of the time */
364 int mref; /* is this going to be a memory ref? */
365 int bracket; /* is it a [] mref, or a & mref? */
366 int setsize = 0;
368 result->oprs[operand].addr_size = 0;/* have to zero this whatever */
369 result->oprs[operand].eaflags = 0; /* and this */
370 result->oprs[operand].opflags = 0;
372 i = stdscan(NULL, &tokval);
373 if (i == 0) break; /* end of operands: get out of here */
374 result->oprs[operand].type = 0; /* so far, no override */
375 while (i == TOKEN_SPECIAL) {/* size specifiers */
376 switch ((int)tokval.t_integer) {
377 case S_BYTE:
378 if (!setsize) /* we want to use only the first */
379 result->oprs[operand].type |= BITS8;
380 setsize = 1;
381 break;
382 case S_WORD:
383 if (!setsize)
384 result->oprs[operand].type |= BITS16;
385 setsize = 1;
386 break;
387 case S_DWORD:
388 case S_LONG:
389 if (!setsize)
390 result->oprs[operand].type |= BITS32;
391 setsize = 1;
392 break;
393 case S_QWORD:
394 if (!setsize)
395 result->oprs[operand].type |= BITS64;
396 setsize = 1;
397 break;
398 case S_TWORD:
399 if (!setsize)
400 result->oprs[operand].type |= BITS80;
401 setsize = 1;
402 break;
403 case S_TO:
404 result->oprs[operand].type |= TO;
405 break;
406 case S_FAR:
407 result->oprs[operand].type |= FAR;
408 break;
409 case S_NEAR:
410 result->oprs[operand].type |= NEAR;
411 break;
412 case S_SHORT:
413 result->oprs[operand].type |= SHORT;
414 break;
415 default:
416 error (ERR_NONFATAL, "invalid operand size specification");
418 i = stdscan(NULL, &tokval);
421 if (i == '[' || i == '&') { /* memory reference */
422 mref = TRUE;
423 bracket = (i == '[');
424 i = stdscan(NULL, &tokval);
425 if (i == TOKEN_SPECIAL) { /* check for address size override */
426 switch ((int)tokval.t_integer) {
427 case S_NOSPLIT:
428 result->oprs[operand].eaflags |= EAF_TIMESTWO;
429 break;
430 case S_BYTE:
431 result->oprs[operand].eaflags |= EAF_BYTEOFFS;
432 break;
433 case S_WORD:
434 result->oprs[operand].addr_size = 16;
435 result->oprs[operand].eaflags |= EAF_WORDOFFS;
436 break;
437 case S_DWORD:
438 case S_LONG:
439 result->oprs[operand].addr_size = 32;
440 result->oprs[operand].eaflags |= EAF_WORDOFFS;
441 break;
442 default:
443 error (ERR_NONFATAL, "invalid size specification in"
444 " effective address");
446 i = stdscan(NULL, &tokval);
448 } else { /* immediate operand, or register */
449 mref = FALSE;
450 bracket = FALSE; /* placate optimisers */
453 value = evaluate (stdscan, NULL, &tokval,
454 &result->oprs[operand].opflags,
455 critical, error, &hints);
456 i = tokval.t_type;
457 if (result->oprs[operand].opflags & OPFLAG_FORWARD) {
458 result->forw_ref = TRUE;
460 if (!value) { /* error in evaluator */
461 result->opcode = -1; /* unrecoverable parse error: */
462 return result; /* ignore this instruction */
464 if (i == ':' && mref) { /* it was seg:offset */
466 * Process the segment override.
468 if (value[1].type!=0 || value->value!=1 ||
469 REG_SREG & ~reg_flags[value->type])
470 error (ERR_NONFATAL, "invalid segment override");
471 else if (result->nprefix == MAXPREFIX)
472 error (ERR_NONFATAL,
473 "instruction has more than %d prefixes",
474 MAXPREFIX);
475 else
476 result->prefixes[result->nprefix++] = value->type;
478 i = stdscan(NULL, &tokval); /* then skip the colon */
479 if (i == TOKEN_SPECIAL) { /* another check for size override */
480 switch ((int)tokval.t_integer) {
481 case S_WORD:
482 result->oprs[operand].addr_size = 16;
483 break;
484 case S_DWORD:
485 case S_LONG:
486 result->oprs[operand].addr_size = 32;
487 break;
488 default:
489 error (ERR_NONFATAL, "invalid size specification in"
490 " effective address");
492 i = stdscan(NULL, &tokval);
494 value = evaluate (stdscan, NULL, &tokval,
495 &result->oprs[operand].opflags,
496 critical, error, &hints);
497 i = tokval.t_type;
498 if (result->oprs[operand].opflags & OPFLAG_FORWARD) {
499 result->forw_ref = TRUE;
501 /* and get the offset */
502 if (!value) { /* but, error in evaluator */
503 result->opcode = -1; /* unrecoverable parse error: */
504 return result; /* ignore this instruction */
507 if (mref && bracket) { /* find ] at the end */
508 if (i != ']') {
509 error (ERR_NONFATAL, "parser: expecting ]");
510 do { /* error recovery again */
511 i = stdscan(NULL, &tokval);
512 } while (i != 0 && i != ',');
513 } else /* we got the required ] */
514 i = stdscan(NULL, &tokval);
515 } else { /* immediate operand */
516 if (i != 0 && i != ',' && i != ':') {
517 error (ERR_NONFATAL, "comma or end of line expected");
518 do { /* error recovery */
519 i = stdscan(NULL, &tokval);
520 } while (i != 0 && i != ',');
521 } else if (i == ':') {
522 result->oprs[operand].type |= COLON;
526 /* now convert the exprs returned from evaluate() into operand
527 * descriptions... */
529 if (mref) { /* it's a memory reference */
530 expr *e = value;
531 int b, i, s; /* basereg, indexreg, scale */
532 long o; /* offset */
534 b = i = -1, o = s = 0;
535 result->oprs[operand].hintbase = hints.base;
536 result->oprs[operand].hinttype = hints.type;
538 if (e->type <= EXPR_REG_END) { /* this bit's a register */
539 if (e->value == 1) /* in fact it can be basereg */
540 b = e->type;
541 else /* no, it has to be indexreg */
542 i = e->type, s = e->value;
543 e++;
545 if (e->type && e->type <= EXPR_REG_END) /* it's a 2nd register */
547 if (b != -1) /* If the first was the base, ... */
548 i = e->type, s = e->value; /* second has to be indexreg */
550 else if (e->value != 1) /* If both want to be index */
552 error(ERR_NONFATAL, "invalid effective address");
553 result->opcode = -1;
554 return result;
556 else
557 b = e->type;
558 e++;
560 if (e->type != 0) { /* is there an offset? */
561 if (e->type <= EXPR_REG_END) /* in fact, is there an error? */
563 error (ERR_NONFATAL, "invalid effective address");
564 result->opcode = -1;
565 return result;
567 else
569 if (e->type == EXPR_UNKNOWN) {
570 o = 0; /* doesn't matter what */
571 result->oprs[operand].wrt = NO_SEG; /* nor this */
572 result->oprs[operand].segment = NO_SEG; /* or this */
573 while (e->type) e++; /* go to the end of the line */
575 else
577 if (e->type == EXPR_SIMPLE) {
578 o = e->value;
579 e++;
581 if (e->type == EXPR_WRT) {
582 result->oprs[operand].wrt = e->value;
583 e++;
584 } else
585 result->oprs[operand].wrt = NO_SEG;
587 * Look for a segment base type.
589 if (e->type && e->type < EXPR_SEGBASE) {
590 error (ERR_NONFATAL, "invalid effective address");
591 result->opcode = -1;
592 return result;
594 while (e->type && e->value == 0)
595 e++;
596 if (e->type && e->value != 1) {
597 error (ERR_NONFATAL, "invalid effective address");
598 result->opcode = -1;
599 return result;
601 if (e->type) {
602 result->oprs[operand].segment =
603 e->type - EXPR_SEGBASE;
604 e++;
605 } else
606 result->oprs[operand].segment = NO_SEG;
607 while (e->type && e->value == 0)
608 e++;
609 if (e->type) {
610 error (ERR_NONFATAL, "invalid effective address");
611 result->opcode = -1;
612 return result;
616 } else {
617 o = 0;
618 result->oprs[operand].wrt = NO_SEG;
619 result->oprs[operand].segment = NO_SEG;
622 if (e->type != 0) { /* there'd better be nothing left! */
623 error (ERR_NONFATAL, "invalid effective address");
624 result->opcode = -1;
625 return result;
628 result->oprs[operand].type |= MEMORY;
629 if (b==-1 && (i==-1 || s==0))
630 result->oprs[operand].type |= MEM_OFFS;
631 result->oprs[operand].basereg = b;
632 result->oprs[operand].indexreg = i;
633 result->oprs[operand].scale = s;
634 result->oprs[operand].offset = o;
636 else /* it's not a memory reference */
638 if (is_just_unknown(value)) { /* it's immediate but unknown */
639 result->oprs[operand].type |= IMMEDIATE;
640 result->oprs[operand].offset = 0; /* don't care */
641 result->oprs[operand].segment = NO_SEG; /* don't care again */
642 result->oprs[operand].wrt = NO_SEG;/* still don't care */
644 else if (is_reloc(value)) /* it's immediate */
646 result->oprs[operand].type |= IMMEDIATE;
647 result->oprs[operand].offset = reloc_value(value);
648 result->oprs[operand].segment = reloc_seg(value);
649 result->oprs[operand].wrt = reloc_wrt(value);
650 if (is_simple(value) && reloc_value(value)==1)
651 result->oprs[operand].type |= UNITY;
653 else /* it's a register */
655 int i;
657 if (value->type>=EXPR_SIMPLE || value->value!=1) {
658 error (ERR_NONFATAL, "invalid operand type");
659 result->opcode = -1;
660 return result;
664 * check that its only 1 register, not an expression...
666 for (i = 1; value[i].type; i++)
667 if (value[i].value) {
668 error (ERR_NONFATAL, "invalid operand type");
669 result->opcode = -1;
670 return result;
673 /* clear overrides, except TO which applies to FPU regs */
674 if (result->oprs[operand].type & ~TO) {
676 * we want to produce a warning iff the specified size
677 * is different from the register size
679 i = result->oprs[operand].type & SIZE_MASK;
681 else
682 i = 0;
684 result->oprs[operand].type &= TO;
685 result->oprs[operand].type |= REGISTER;
686 result->oprs[operand].type |= reg_flags[value->type];
687 result->oprs[operand].basereg = value->type;
689 if (i && (result->oprs[operand].type & SIZE_MASK) != i)
690 error (ERR_WARNING|ERR_PASS1,
691 "register size specification ignored");
696 result->operands = operand; /* set operand count */
698 while (operand<3) /* clear remaining operands */
699 result->oprs[operand++].type = 0;
702 * Transform RESW, RESD, RESQ, REST into RESB.
704 switch (result->opcode) {
705 case I_RESW: result->opcode=I_RESB; result->oprs[0].offset*=2; break;
706 case I_RESD: result->opcode=I_RESB; result->oprs[0].offset*=4; break;
707 case I_RESQ: result->opcode=I_RESB; result->oprs[0].offset*=8; break;
708 case I_REST: result->opcode=I_RESB; result->oprs[0].offset*=10; break;
711 return result;
714 static int is_comma_next (void)
716 char *p;
717 int i;
718 struct tokenval tv;
720 p = stdscan_bufptr;
721 i = stdscan (NULL, &tv);
722 stdscan_bufptr = p;
723 return (i == ',' || i == ';' || !i);
726 void cleanup_insn (insn *i)
728 extop *e;
730 while (i->eops) {
731 e = i->eops;
732 i->eops = i->eops->next;
733 nasm_free (e);