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 license given in the file "LICENSE"
6 * distributed in the NASM archive.
8 * initial version 27/iii/95 by Simon Tatham
27 extern int in_abs_seg; /* ABSOLUTE segment flag */
28 extern int32_t abs_seg; /* ABSOLUTE segment */
29 extern int32_t abs_offset; /* ABSOLUTE segment offset */
31 #include "regflags.c" /* List of register flags */
33 static int is_comma_next(void);
36 static struct tokenval tokval;
38 static struct ofmt *outfmt; /* Structure of addresses of output routines */
39 static struct location *location; /* Pointer to current line's segment,offset */
41 void parser_global_info(struct ofmt *output, struct location * locp)
47 static int prefix_slot(enum prefixes prefix)
75 error(ERR_PANIC, "Invalid value %d passed to prefix_slot()", prefix);
80 static void process_size_override(insn * result, int operand)
82 if (tasm_compatible_mode) {
83 switch ((int)tokval.t_integer) {
84 /* For TASM compatibility a size override inside the
85 * brackets changes the size of the operand, not the
86 * address type of the operand as it does in standard
91 * is valid syntax in TASM compatibility mode. Note that
92 * you lose the ability to override the default address
93 * type for the instruction, but we never use anything
94 * but 32-bit flat model addressing in our code.
97 result->oprs[operand].type |= BITS8;
100 result->oprs[operand].type |= BITS16;
104 result->oprs[operand].type |= BITS32;
107 result->oprs[operand].type |= BITS64;
110 result->oprs[operand].type |= BITS80;
113 result->oprs[operand].type |= BITS128;
117 "invalid operand size specification");
121 /* Standard NASM compatible syntax */
122 switch ((int)tokval.t_integer) {
124 result->oprs[operand].eaflags |= EAF_TIMESTWO;
127 result->oprs[operand].eaflags |= EAF_REL;
130 result->oprs[operand].eaflags |= EAF_ABS;
133 result->oprs[operand].disp_size = 8;
134 result->oprs[operand].eaflags |= EAF_BYTEOFFS;
139 if (result->prefixes[PPS_ASIZE] &&
140 result->prefixes[PPS_ASIZE] != tokval.t_integer)
142 "conflicting address size specifications");
144 result->prefixes[PPS_ASIZE] = tokval.t_integer;
147 result->oprs[operand].disp_size = 16;
148 result->oprs[operand].eaflags |= EAF_WORDOFFS;
152 result->oprs[operand].disp_size = 32;
153 result->oprs[operand].eaflags |= EAF_WORDOFFS;
156 result->oprs[operand].disp_size = 64;
157 result->oprs[operand].eaflags |= EAF_WORDOFFS;
160 error(ERR_NONFATAL, "invalid size specification in"
161 " effective address");
167 insn *parse_line(int pass, char *buffer, insn * result,
168 efunc errfunc, evalfunc evaluate, ldfunc ldef)
172 struct eval_hints hints;
175 bool insn_is_label = false;
179 result->forw_ref = false;
183 stdscan_bufptr = buffer;
184 i = stdscan(NULL, &tokval);
186 result->label = NULL; /* Assume no label */
187 result->eops = NULL; /* must do this, whatever happens */
188 result->operands = 0; /* must initialize this */
190 if (i == 0) { /* blank line - ignore */
191 result->opcode = -1; /* and no instruction either */
194 if (i != TOKEN_ID && i != TOKEN_INSN && i != TOKEN_PREFIX &&
195 (i != TOKEN_REG || (REG_SREG & ~reg_flags[tokval.t_integer]))) {
196 error(ERR_NONFATAL, "label or instruction expected"
197 " at start of line");
202 if (i == TOKEN_ID || (insn_is_label && i == TOKEN_INSN)) {
203 /* there's a label here */
205 result->label = tokval.t_charptr;
206 i = stdscan(NULL, &tokval);
207 if (i == ':') { /* skip over the optional colon */
208 i = stdscan(NULL, &tokval);
210 error(ERR_WARNING | ERR_WARN_OL | ERR_PASS1,
211 "label alone on a line without a colon might be in error");
213 if (i != TOKEN_INSN || tokval.t_integer != I_EQU) {
215 * FIXME: location->segment could be NO_SEG, in which case
216 * it is possible we should be passing 'abs_seg'. Look into this.
217 * Work out whether that is *really* what we should be doing.
218 * Generally fix things. I think this is right as it is, but
219 * am still not certain.
221 ldef(result->label, in_abs_seg ? abs_seg : location->segment,
222 location->offset, NULL, true, false, outfmt, errfunc);
227 result->opcode = -1; /* this line contains just a label */
231 for (j = 0; j < MAXPREFIX; j++)
232 result->prefixes[j] = P_none;
235 while (i == TOKEN_PREFIX ||
236 (i == TOKEN_REG && !(REG_SREG & ~reg_flags[tokval.t_integer])))
241 * Handle special case: the TIMES prefix.
243 if (i == TOKEN_PREFIX && tokval.t_integer == P_TIMES) {
246 i = stdscan(NULL, &tokval);
248 evaluate(stdscan, NULL, &tokval, NULL, pass0, error, NULL);
250 if (!value) { /* but, error in evaluator */
251 result->opcode = -1; /* unrecoverable parse error: */
252 return result; /* ignore this instruction */
254 if (!is_simple(value)) {
256 "non-constant argument supplied to TIMES");
259 result->times = value->value;
260 if (value->value < 0) {
261 error(ERR_NONFATAL, "TIMES value %d is negative",
267 int slot = prefix_slot(tokval.t_integer);
268 if (result->prefixes[slot]) {
269 if (result->prefixes[slot] == tokval.t_integer)
271 "instruction has redundant prefixes");
274 "instruction has conflicting prefixes");
276 result->prefixes[slot] = tokval.t_integer;
277 i = stdscan(NULL, &tokval);
281 if (i != TOKEN_INSN) {
285 for (j = 0; j < MAXPREFIX; j++)
286 if ((pfx = result->prefixes[j]) != P_none)
289 if (i == 0 && pfx != P_none) {
291 * Instruction prefixes are present, but no actual
292 * instruction. This is allowed: at this point we
293 * invent a notional instruction of RESB 0.
295 result->opcode = I_RESB;
296 result->operands = 1;
297 result->oprs[0].type = IMMEDIATE;
298 result->oprs[0].offset = 0L;
299 result->oprs[0].segment = result->oprs[0].wrt = NO_SEG;
302 error(ERR_NONFATAL, "parser: instruction expected");
308 result->opcode = tokval.t_integer;
309 result->condition = tokval.t_inttwo;
312 * RESB, RESW and RESD cannot be satisfied with incorrectly
313 * evaluated operands, since the correct values _must_ be known
314 * on the first pass. Hence, even in pass one, we set the
315 * `critical' flag on calling evaluate(), so that it will bomb
316 * out on undefined symbols. Nasty, but there's nothing we can
319 * For the moment, EQU has the same difficulty, so we'll
322 if (result->opcode == I_RESB || result->opcode == I_RESW ||
323 result->opcode == I_RESD || result->opcode == I_RESQ ||
324 result->opcode == I_REST || result->opcode == I_RESO ||
325 result->opcode == I_EQU || result->opcode == I_INCBIN) {
328 critical = (pass == 2 ? 2 : 0);
330 if (result->opcode == I_DB || result->opcode == I_DW ||
331 result->opcode == I_DD || result->opcode == I_DQ ||
332 result->opcode == I_DT || result->opcode == I_DO ||
333 result->opcode == I_INCBIN) {
334 extop *eop, **tail = &result->eops, **fixptr;
337 result->eops_float = false;
340 * Begin to read the DB/DW/DD/DQ/DT/DO/INCBIN operands.
343 i = stdscan(NULL, &tokval);
346 else if (first && i == ':') {
347 insn_is_label = true;
352 eop = *tail = nasm_malloc(sizeof(extop));
355 eop->type = EOT_NOTHING;
358 if (i == TOKEN_NUM && tokval.t_charptr && is_comma_next()) {
359 eop->type = EOT_DB_STRING;
360 eop->stringval = tokval.t_charptr;
361 eop->stringlen = tokval.t_inttwo;
362 i = stdscan(NULL, &tokval); /* eat the comma */
366 if ((i == TOKEN_FLOAT && is_comma_next())
367 || i == '-' || i == '+') {
370 if (i == '+' || i == '-') {
371 char *save = stdscan_bufptr;
373 sign = (i == '-') ? -1 : 1;
374 i = stdscan(NULL, &tokval);
375 if (i != TOKEN_FLOAT || !is_comma_next()) {
376 stdscan_bufptr = save;
377 i = tokval.t_type = token;
381 if (i == TOKEN_FLOAT) {
382 eop->type = EOT_DB_STRING;
383 result->eops_float = true;
384 switch (result->opcode) {
404 error(ERR_NONFATAL, "floating-point constant"
405 " encountered in unknown instruction");
407 * fix suggested by Pedro Gimeno... original line
409 * eop->type = EOT_NOTHING;
414 eop = nasm_realloc(eop, sizeof(extop) + eop->stringlen);
417 eop->stringval = (char *)eop + sizeof(extop);
418 if (!eop->stringlen ||
419 !float_const(tokval.t_charptr, sign,
420 (uint8_t *)eop->stringval,
421 eop->stringlen, error))
422 eop->type = EOT_NOTHING;
423 i = stdscan(NULL, &tokval); /* eat the comma */
431 value = evaluate(stdscan, NULL, &tokval, NULL,
432 critical, error, NULL);
434 if (!value) { /* error in evaluator */
435 result->opcode = -1; /* unrecoverable parse error: */
436 return result; /* ignore this instruction */
438 if (is_unknown(value)) {
439 eop->type = EOT_DB_NUMBER;
440 eop->offset = 0; /* doesn't matter what we put */
441 eop->segment = eop->wrt = NO_SEG; /* likewise */
442 } else if (is_reloc(value)) {
443 eop->type = EOT_DB_NUMBER;
444 eop->offset = reloc_value(value);
445 eop->segment = reloc_seg(value);
446 eop->wrt = reloc_wrt(value);
449 "operand %d: expression is not simple"
450 " or relocatable", oper_num);
455 * We're about to call stdscan(), which will eat the
456 * comma that we're currently sitting on between
457 * arguments. However, we'd better check first that it
460 if (i == 0) /* also could be EOL */
463 error(ERR_NONFATAL, "comma expected after operand %d",
465 result->opcode = -1; /* unrecoverable parse error: */
466 return result; /* ignore this instruction */
470 if (result->opcode == I_INCBIN) {
472 * Correct syntax for INCBIN is that there should be
473 * one string operand, followed by one or two numeric
476 if (!result->eops || result->eops->type != EOT_DB_STRING)
477 error(ERR_NONFATAL, "`incbin' expects a file name");
478 else if (result->eops->next &&
479 result->eops->next->type != EOT_DB_NUMBER)
480 error(ERR_NONFATAL, "`incbin': second parameter is",
482 else if (result->eops->next && result->eops->next->next &&
483 result->eops->next->next->type != EOT_DB_NUMBER)
484 error(ERR_NONFATAL, "`incbin': third parameter is",
486 else if (result->eops->next && result->eops->next->next &&
487 result->eops->next->next->next)
489 "`incbin': more than three parameters");
493 * If we reach here, one of the above errors happened.
494 * Throw the instruction away.
498 } else /* DB ... */ if (oper_num == 0)
499 error(ERR_WARNING | ERR_PASS1,
500 "no operand for data declaration");
502 result->operands = oper_num;
507 /* right. Now we begin to parse the operands. There may be up to four
508 * of these, separated by commas, and terminated by a zero token. */
510 for (operand = 0; operand < MAX_OPERANDS; operand++) {
511 expr *value; /* used most of the time */
512 int mref; /* is this going to be a memory ref? */
513 int bracket; /* is it a [] mref, or a & mref? */
516 result->oprs[operand].disp_size = 0; /* have to zero this whatever */
517 result->oprs[operand].eaflags = 0; /* and this */
518 result->oprs[operand].opflags = 0;
520 i = stdscan(NULL, &tokval);
522 break; /* end of operands: get out of here */
523 else if (first && i == ':') {
524 insn_is_label = true;
528 result->oprs[operand].type = 0; /* so far, no override */
529 while (i == TOKEN_SPECIAL) { /* size specifiers */
530 switch ((int)tokval.t_integer) {
532 if (!setsize) /* we want to use only the first */
533 result->oprs[operand].type |= BITS8;
538 result->oprs[operand].type |= BITS16;
544 result->oprs[operand].type |= BITS32;
549 result->oprs[operand].type |= BITS64;
554 result->oprs[operand].type |= BITS80;
559 result->oprs[operand].type |= BITS128;
563 result->oprs[operand].type |= TO;
566 result->oprs[operand].type |= STRICT;
569 result->oprs[operand].type |= FAR;
572 result->oprs[operand].type |= NEAR;
575 result->oprs[operand].type |= SHORT;
578 error(ERR_NONFATAL, "invalid operand size specification");
580 i = stdscan(NULL, &tokval);
583 if (i == '[' || i == '&') { /* memory reference */
585 bracket = (i == '[');
586 i = stdscan(NULL, &tokval); /* then skip the colon */
587 while (i == TOKEN_SPECIAL || i == TOKEN_PREFIX) {
588 process_size_override(result, operand);
589 i = stdscan(NULL, &tokval);
591 } else { /* immediate operand, or register */
593 bracket = false; /* placate optimisers */
596 if ((result->oprs[operand].type & FAR) && !mref &&
597 result->opcode != I_JMP && result->opcode != I_CALL) {
598 error(ERR_NONFATAL, "invalid use of FAR operand specifier");
601 value = evaluate(stdscan, NULL, &tokval,
602 &result->oprs[operand].opflags,
603 critical, error, &hints);
605 if (result->oprs[operand].opflags & OPFLAG_FORWARD) {
606 result->forw_ref = true;
608 if (!value) { /* error in evaluator */
609 result->opcode = -1; /* unrecoverable parse error: */
610 return result; /* ignore this instruction */
612 if (i == ':' && mref) { /* it was seg:offset */
614 * Process the segment override.
616 if (value[1].type != 0 || value->value != 1 ||
617 REG_SREG & ~reg_flags[value->type])
618 error(ERR_NONFATAL, "invalid segment override");
619 else if (result->prefixes[PPS_SEG])
621 "instruction has conflicting segment overrides");
623 result->prefixes[PPS_SEG] = value->type;
624 if (!(REG_FSGS & ~reg_flags[value->type]))
625 result->oprs[operand].eaflags |= EAF_FSGS;
628 i = stdscan(NULL, &tokval); /* then skip the colon */
629 while (i == TOKEN_SPECIAL || i == TOKEN_PREFIX) {
630 process_size_override(result, operand);
631 i = stdscan(NULL, &tokval);
633 value = evaluate(stdscan, NULL, &tokval,
634 &result->oprs[operand].opflags,
635 critical, error, &hints);
637 if (result->oprs[operand].opflags & OPFLAG_FORWARD) {
638 result->forw_ref = true;
640 /* and get the offset */
641 if (!value) { /* but, error in evaluator */
642 result->opcode = -1; /* unrecoverable parse error: */
643 return result; /* ignore this instruction */
646 if (mref && bracket) { /* find ] at the end */
648 error(ERR_NONFATAL, "parser: expecting ]");
649 do { /* error recovery again */
650 i = stdscan(NULL, &tokval);
651 } while (i != 0 && i != ',');
652 } else /* we got the required ] */
653 i = stdscan(NULL, &tokval);
654 } else { /* immediate operand */
655 if (i != 0 && i != ',' && i != ':') {
656 error(ERR_NONFATAL, "comma or end of line expected");
657 do { /* error recovery */
658 i = stdscan(NULL, &tokval);
659 } while (i != 0 && i != ',');
660 } else if (i == ':') {
661 result->oprs[operand].type |= COLON;
665 /* now convert the exprs returned from evaluate() into operand
668 if (mref) { /* it's a memory reference */
670 int b, i, s; /* basereg, indexreg, scale */
671 int64_t o; /* offset */
673 b = i = -1, o = s = 0;
674 result->oprs[operand].hintbase = hints.base;
675 result->oprs[operand].hinttype = hints.type;
677 if (e->type && e->type <= EXPR_REG_END) { /* this bit's a register */
678 if (e->value == 1) /* in fact it can be basereg */
680 else /* no, it has to be indexreg */
681 i = e->type, s = e->value;
684 if (e->type && e->type <= EXPR_REG_END) { /* it's a 2nd register */
685 if (b != -1) /* If the first was the base, ... */
686 i = e->type, s = e->value; /* second has to be indexreg */
688 else if (e->value != 1) { /* If both want to be index */
690 "beroset-p-592-invalid effective address");
697 if (e->type != 0) { /* is there an offset? */
698 if (e->type <= EXPR_REG_END) { /* in fact, is there an error? */
700 "beroset-p-603-invalid effective address");
704 if (e->type == EXPR_UNKNOWN) {
705 o = 0; /* doesn't matter what */
706 result->oprs[operand].wrt = NO_SEG; /* nor this */
707 result->oprs[operand].segment = NO_SEG; /* or this */
709 e++; /* go to the end of the line */
711 if (e->type == EXPR_SIMPLE) {
715 if (e->type == EXPR_WRT) {
716 result->oprs[operand].wrt = e->value;
719 result->oprs[operand].wrt = NO_SEG;
721 * Look for a segment base type.
723 if (e->type && e->type < EXPR_SEGBASE) {
725 "beroset-p-630-invalid effective address");
729 while (e->type && e->value == 0)
731 if (e->type && e->value != 1) {
733 "beroset-p-637-invalid effective address");
738 result->oprs[operand].segment =
739 e->type - EXPR_SEGBASE;
742 result->oprs[operand].segment = NO_SEG;
743 while (e->type && e->value == 0)
747 "beroset-p-650-invalid effective address");
755 result->oprs[operand].wrt = NO_SEG;
756 result->oprs[operand].segment = NO_SEG;
759 if (e->type != 0) { /* there'd better be nothing left! */
761 "beroset-p-663-invalid effective address");
766 /* It is memory, but it can match any r/m operand */
767 result->oprs[operand].type |= MEMORY_ANY;
769 if (b == -1 && (i == -1 || s == 0)) {
770 int is_rel = globalbits == 64 &&
771 !(result->oprs[operand].eaflags & EAF_ABS) &&
773 !(result->oprs[operand].eaflags & EAF_FSGS)) ||
774 (result->oprs[operand].eaflags & EAF_REL));
776 result->oprs[operand].type |= is_rel ? IP_REL : MEM_OFFS;
778 result->oprs[operand].basereg = b;
779 result->oprs[operand].indexreg = i;
780 result->oprs[operand].scale = s;
781 result->oprs[operand].offset = o;
782 } else { /* it's not a memory reference */
784 if (is_just_unknown(value)) { /* it's immediate but unknown */
785 result->oprs[operand].type |= IMMEDIATE;
786 result->oprs[operand].offset = 0; /* don't care */
787 result->oprs[operand].segment = NO_SEG; /* don't care again */
788 result->oprs[operand].wrt = NO_SEG; /* still don't care */
789 } else if (is_reloc(value)) { /* it's immediate */
790 result->oprs[operand].type |= IMMEDIATE;
791 result->oprs[operand].offset = reloc_value(value);
792 result->oprs[operand].segment = reloc_seg(value);
793 result->oprs[operand].wrt = reloc_wrt(value);
794 if (is_simple(value)) {
795 if (reloc_value(value) == 1)
796 result->oprs[operand].type |= UNITY;
797 if (optimizing >= 0 &&
798 !(result->oprs[operand].type & STRICT)) {
799 if (reloc_value(value) >= -128 &&
800 reloc_value(value) <= 127)
801 result->oprs[operand].type |= SBYTE;
804 } else { /* it's a register */
807 if (value->type >= EXPR_SIMPLE || value->value != 1) {
808 error(ERR_NONFATAL, "invalid operand type");
814 * check that its only 1 register, not an expression...
816 for (i = 1; value[i].type; i++)
817 if (value[i].value) {
818 error(ERR_NONFATAL, "invalid operand type");
823 /* clear overrides, except TO which applies to FPU regs */
824 if (result->oprs[operand].type & ~TO) {
826 * we want to produce a warning iff the specified size
827 * is different from the register size
829 rs = result->oprs[operand].type & SIZE_MASK;
833 result->oprs[operand].type &= TO;
834 result->oprs[operand].type |= REGISTER;
835 result->oprs[operand].type |= reg_flags[value->type];
836 result->oprs[operand].basereg = value->type;
838 if (rs && (result->oprs[operand].type & SIZE_MASK) != rs)
839 error(ERR_WARNING | ERR_PASS1,
840 "register size specification ignored");
845 result->operands = operand; /* set operand count */
847 /* clear remaining operands */
848 while (operand < MAX_OPERANDS)
849 result->oprs[operand++].type = 0;
852 * Transform RESW, RESD, RESQ, REST, RESO into RESB.
854 switch (result->opcode) {
856 result->opcode = I_RESB;
857 result->oprs[0].offset *= 2;
860 result->opcode = I_RESB;
861 result->oprs[0].offset *= 4;
864 result->opcode = I_RESB;
865 result->oprs[0].offset *= 8;
868 result->opcode = I_RESB;
869 result->oprs[0].offset *= 10;
872 result->opcode = I_RESB;
873 result->oprs[0].offset *= 16;
882 static int is_comma_next(void)
889 i = stdscan(NULL, &tv);
891 return (i == ',' || i == ';' || !i);
894 void cleanup_insn(insn * i)
900 i->eops = i->eops->next;