?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
REX: Set REX bits in accordance with 32-register environment
[nasm.git] / parser.c
blob37a5e1cf26074baa02219a7556b150e55e32731a
1 /* ----------------------------------------------------------------------- *
2 *
3 * Copyright 1996-2013 The NASM Authors - All Rights Reserved
4 * See the file AUTHORS included with the NASM distribution for
5 * the specific copyright holders.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following
9 * conditions are met:
10 *
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above
14 * copyright notice, this list of conditions and the following
15 * disclaimer in the documentation and/or other materials provided
16 * with the distribution.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND
19 * CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES,
20 * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
21 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
22 * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
23 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
24 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
25 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
26 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
29 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
30 * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
31 *
32 * ----------------------------------------------------------------------- */
33
34 /*
35 * parser.c source line parser for the Netwide Assembler
36 */
37
38 #include "compiler.h"
39
40 #include <stdio.h>
41 #include <stdlib.h>
42 #include <stddef.h>
43 #include <string.h>
44 #include <ctype.h>
45 #include <inttypes.h>
46
47 #include "nasm.h"
48 #include "insns.h"
49 #include "nasmlib.h"
50 #include "stdscan.h"
51 #include "eval.h"
52 #include "parser.h"
53 #include "float.h"
54 #include "tables.h"
55
56 extern int in_abs_seg; /* ABSOLUTE segment flag */
57 extern int32_t abs_seg; /* ABSOLUTE segment */
58 extern int32_t abs_offset; /* ABSOLUTE segment offset */
59
60 static int is_comma_next(void);
61
62 static int i;
63 static struct tokenval tokval;
64 static struct location *location; /* Pointer to current line's segment,offset */
65
66 void parser_global_info(struct location * locp)
67 {
68 location = locp;
69 }
70
71 static int prefix_slot(int prefix)
72 {
73 switch (prefix) {
74 case P_WAIT:
75 return PPS_WAIT;
76 case R_CS:
77 case R_DS:
78 case R_SS:
79 case R_ES:
80 case R_FS:
81 case R_GS:
82 return PPS_SEG;
83 case P_LOCK:
84 return PPS_LOCK;
85 case P_REP:
86 case P_REPE:
87 case P_REPZ:
88 case P_REPNE:
89 case P_REPNZ:
90 case P_XACQUIRE:
91 case P_XRELEASE:
92 return PPS_REP;
93 case P_O16:
94 case P_O32:
95 case P_O64:
96 case P_OSP:
97 return PPS_OSIZE;
98 case P_A16:
99 case P_A32:
100 case P_A64:
101 case P_ASP:
102 return PPS_ASIZE;
103 default:
104 nasm_error(ERR_PANIC, "Invalid value %d passed to prefix_slot()", prefix);
105 return -1;
106 }
107 }
108
109 static void process_size_override(insn *result, int operand)
110 {
111 if (tasm_compatible_mode) {
112 switch ((int)tokval.t_integer) {
113 /* For TASM compatibility a size override inside the
114 * brackets changes the size of the operand, not the
115 * address type of the operand as it does in standard
116 * NASM syntax. Hence:
117 *
118 * mov eax,[DWORD val]
119 *
120 * is valid syntax in TASM compatibility mode. Note that
121 * you lose the ability to override the default address
122 * type for the instruction, but we never use anything
123 * but 32-bit flat model addressing in our code.
124 */
125 case S_BYTE:
126 result->oprs[operand].type |= BITS8;
127 break;
128 case S_WORD:
129 result->oprs[operand].type |= BITS16;
130 break;
131 case S_DWORD:
132 case S_LONG:
133 result->oprs[operand].type |= BITS32;
134 break;
135 case S_QWORD:
136 result->oprs[operand].type |= BITS64;
137 break;
138 case S_TWORD:
139 result->oprs[operand].type |= BITS80;
140 break;
141 case S_OWORD:
142 result->oprs[operand].type |= BITS128;
143 break;
144 default:
145 nasm_error(ERR_NONFATAL,
146 "invalid operand size specification");
147 break;
148 }
149 } else {
150 /* Standard NASM compatible syntax */
151 switch ((int)tokval.t_integer) {
152 case S_NOSPLIT:
153 result->oprs[operand].eaflags |= EAF_TIMESTWO;
154 break;
155 case S_REL:
156 result->oprs[operand].eaflags |= EAF_REL;
157 break;
158 case S_ABS:
159 result->oprs[operand].eaflags |= EAF_ABS;
160 break;
161 case S_BYTE:
162 result->oprs[operand].disp_size = 8;
163 result->oprs[operand].eaflags |= EAF_BYTEOFFS;
164 break;
165 case P_A16:
166 case P_A32:
167 case P_A64:
168 if (result->prefixes[PPS_ASIZE] &&
169 result->prefixes[PPS_ASIZE] != tokval.t_integer)
170 nasm_error(ERR_NONFATAL,
171 "conflicting address size specifications");
172 else
173 result->prefixes[PPS_ASIZE] = tokval.t_integer;
174 break;
175 case S_WORD:
176 result->oprs[operand].disp_size = 16;
177 result->oprs[operand].eaflags |= EAF_WORDOFFS;
178 break;
179 case S_DWORD:
180 case S_LONG:
181 result->oprs[operand].disp_size = 32;
182 result->oprs[operand].eaflags |= EAF_WORDOFFS;
183 break;
184 case S_QWORD:
185 result->oprs[operand].disp_size = 64;
186 result->oprs[operand].eaflags |= EAF_WORDOFFS;
187 break;
188 default:
189 nasm_error(ERR_NONFATAL, "invalid size specification in"
190 " effective address");
191 break;
192 }
193 }
194 }
195
196 /*
197 * when two or more decorators follow a register operand,
198 * consecutive decorators are parsed here.
199 * opmask and zeroing decorators can be placed in any order.
200 * e.g. zmm1 {k2}{z} or zmm2 {z,k3}
201 * decorator(s) are placed at the end of an operand.
202 */
203 static bool parse_braces(decoflags_t *decoflags)
204 {
205 int i;
206 bool recover = false;
207
208 i = tokval.t_type;
209 do {
210 if (i == TOKEN_OPMASK) {
211 if (*decoflags & OPMASK_MASK) {
212 nasm_error(ERR_NONFATAL, "opmask k%lu is already set",
213 *decoflags & OPMASK_MASK);
214 *decoflags &= ~OPMASK_MASK;
215 }
216 *decoflags |= VAL_OPMASK(nasm_regvals[tokval.t_integer]);
217 } else if (i == TOKEN_DECORATOR) {
218 switch (tokval.t_integer) {
219 case BRC_Z:
220 /*
221 * according to AVX512 spec, only zeroing/merging decorator
222 * is supported with opmask
223 */
224 *decoflags |= GEN_Z(0);
225 break;
226 default:
227 nasm_error(ERR_NONFATAL, "{%s} is not an expected decorator",
228 tokval.t_charptr);
229 break;
230 }
231 } else if (i == ',' || i == TOKEN_EOS){
232 break;
233 } else {
234 nasm_error(ERR_NONFATAL, "only a series of valid decorators"
235 " expected");
236 recover = true;
237 break;
238 }
239 i = stdscan(NULL, &tokval);
240 } while(1);
241
242 return recover;
243 }
244
245 insn *parse_line(int pass, char *buffer, insn *result, ldfunc ldef)
246 {
247 bool insn_is_label = false;
248 struct eval_hints hints;
249 int operand;
250 int critical;
251 bool first;
252 bool recover;
253
254 restart_parse:
255 first = true;
256 result->forw_ref = false;
257
258 stdscan_reset();
259 stdscan_set(buffer);
260 i = stdscan(NULL, &tokval);
261
262 result->label = NULL; /* Assume no label */
263 result->eops = NULL; /* must do this, whatever happens */
264 result->operands = 0; /* must initialize this */
265 result->evex_rm = 0; /* Ensure EVEX rounding mode is reset */
266 result->evex_brerop = -1; /* Reset EVEX broadcasting/ER op position */
267
268 /* Ignore blank lines */
269 if (i == TOKEN_EOS) {
270 result->opcode = I_none;
271 return result;
272 }
273
274 if (i != TOKEN_ID &&
275 i != TOKEN_INSN &&
276 i != TOKEN_PREFIX &&
277 (i != TOKEN_REG || !IS_SREG(tokval.t_integer))) {
278 nasm_error(ERR_NONFATAL,
279 "label or instruction expected at start of line");
280 result->opcode = I_none;
281 return result;
282 }
283
284 if (i == TOKEN_ID || (insn_is_label && i == TOKEN_INSN)) {
285 /* there's a label here */
286 first = false;
287 result->label = tokval.t_charptr;
288 i = stdscan(NULL, &tokval);
289 if (i == ':') { /* skip over the optional colon */
290 i = stdscan(NULL, &tokval);
291 } else if (i == 0) {
292 nasm_error(ERR_WARNING | ERR_WARN_OL | ERR_PASS1,
293 "label alone on a line without a colon might be in error");
294 }
295 if (i != TOKEN_INSN || tokval.t_integer != I_EQU) {
296 /*
297 * FIXME: location->segment could be NO_SEG, in which case
298 * it is possible we should be passing 'abs_seg'. Look into this.
299 * Work out whether that is *really* what we should be doing.
300 * Generally fix things. I think this is right as it is, but
301 * am still not certain.
302 */
303 ldef(result->label, in_abs_seg ? abs_seg : location->segment,
304 location->offset, NULL, true, false);
305 }
306 }
307
308 /* Just a label here */
309 if (i == TOKEN_EOS) {
310 result->opcode = I_none;
311 return result;
312 }
313
314 nasm_build_assert(P_none != 0);
315 memset(result->prefixes, P_none, sizeof(result->prefixes));
316 result->times = 1L;
317
318 while (i == TOKEN_PREFIX ||
319 (i == TOKEN_REG && IS_SREG(tokval.t_integer))) {
320 first = false;
321
322 /*
323 * Handle special case: the TIMES prefix.
324 */
325 if (i == TOKEN_PREFIX && tokval.t_integer == P_TIMES) {
326 expr *value;
327
328 i = stdscan(NULL, &tokval);
329 value = evaluate(stdscan, NULL, &tokval, NULL, pass0, nasm_error, NULL);
330 i = tokval.t_type;
331 if (!value) { /* but, error in evaluator */
332 result->opcode = I_none; /* unrecoverable parse error: */
333 return result; /* ignore this instruction */
334 }
335 if (!is_simple(value)) {
336 nasm_error(ERR_NONFATAL,
337 "non-constant argument supplied to TIMES");
338 result->times = 1L;
339 } else {
340 result->times = value->value;
341 if (value->value < 0 && pass0 == 2) {
342 nasm_error(ERR_NONFATAL, "TIMES value %"PRId64" is negative",
343 value->value);
344 result->times = 0;
345 }
346 }
347 } else {
348 int slot = prefix_slot(tokval.t_integer);
349 if (result->prefixes[slot]) {
350 if (result->prefixes[slot] == tokval.t_integer)
351 nasm_error(ERR_WARNING | ERR_PASS1,
352 "instruction has redundant prefixes");
353 else
354 nasm_error(ERR_NONFATAL,
355 "instruction has conflicting prefixes");
356 }
357 result->prefixes[slot] = tokval.t_integer;
358 i = stdscan(NULL, &tokval);
359 }
360 }
361
362 if (i != TOKEN_INSN) {
363 int j;
364 enum prefixes pfx;
365
366 for (j = 0; j < MAXPREFIX; j++) {
367 if ((pfx = result->prefixes[j]) != P_none)
368 break;
369 }
370
371 if (i == 0 && pfx != P_none) {
372 /*
373 * Instruction prefixes are present, but no actual
374 * instruction. This is allowed: at this point we
375 * invent a notional instruction of RESB 0.
376 */
377 result->opcode = I_RESB;
378 result->operands = 1;
379 result->oprs[0].type = IMMEDIATE;
380 result->oprs[0].offset = 0L;
381 result->oprs[0].segment = result->oprs[0].wrt = NO_SEG;
382 return result;
383 } else {
384 nasm_error(ERR_NONFATAL, "parser: instruction expected");
385 result->opcode = I_none;
386 return result;
387 }
388 }
389
390 result->opcode = tokval.t_integer;
391 result->condition = tokval.t_inttwo;
392
393 /*
394 * INCBIN cannot be satisfied with incorrectly
395 * evaluated operands, since the correct values _must_ be known
396 * on the first pass. Hence, even in pass one, we set the
397 * `critical' flag on calling evaluate(), so that it will bomb
398 * out on undefined symbols.
399 */
400 if (result->opcode == I_INCBIN) {
401 critical = (pass0 < 2 ? 1 : 2);
402
403 } else
404 critical = (pass == 2 ? 2 : 0);
405
406 if (result->opcode == I_DB || result->opcode == I_DW ||
407 result->opcode == I_DD || result->opcode == I_DQ ||
408 result->opcode == I_DT || result->opcode == I_DO ||
409 result->opcode == I_DY || result->opcode == I_DZ ||
410 result->opcode == I_INCBIN) {
411 extop *eop, **tail = &result->eops, **fixptr;
412 int oper_num = 0;
413 int32_t sign;
414
415 result->eops_float = false;
416
417 /*
418 * Begin to read the DB/DW/DD/DQ/DT/DO/DY/DZ/INCBIN operands.
419 */
420 while (1) {
421 i = stdscan(NULL, &tokval);
422 if (i == TOKEN_EOS)
423 break;
424 else if (first && i == ':') {
425 insn_is_label = true;
426 goto restart_parse;
427 }
428 first = false;
429 fixptr = tail;
430 eop = *tail = nasm_malloc(sizeof(extop));
431 tail = &eop->next;
432 eop->next = NULL;
433 eop->type = EOT_NOTHING;
434 oper_num++;
435 sign = +1;
436
437 /*
438 * is_comma_next() here is to distinguish this from
439 * a string used as part of an expression...
440 */
441 if (i == TOKEN_STR && is_comma_next()) {
442 eop->type = EOT_DB_STRING;
443 eop->stringval = tokval.t_charptr;
444 eop->stringlen = tokval.t_inttwo;
445 i = stdscan(NULL, &tokval); /* eat the comma */
446 } else if (i == TOKEN_STRFUNC) {
447 bool parens = false;
448 const char *funcname = tokval.t_charptr;
449 enum strfunc func = tokval.t_integer;
450 i = stdscan(NULL, &tokval);
451 if (i == '(') {
452 parens = true;
453 i = stdscan(NULL, &tokval);
454 }
455 if (i != TOKEN_STR) {
456 nasm_error(ERR_NONFATAL,
457 "%s must be followed by a string constant",
458 funcname);
459 eop->type = EOT_NOTHING;
460 } else {
461 eop->type = EOT_DB_STRING_FREE;
462 eop->stringlen =
463 string_transform(tokval.t_charptr, tokval.t_inttwo,
464 &eop->stringval, func);
465 if (eop->stringlen == (size_t)-1) {
466 nasm_error(ERR_NONFATAL, "invalid string for transform");
467 eop->type = EOT_NOTHING;
468 }
469 }
470 if (parens && i && i != ')') {
471 i = stdscan(NULL, &tokval);
472 if (i != ')') {
473 nasm_error(ERR_NONFATAL, "unterminated %s function",
474 funcname);
475 }
476 }
477 if (i && i != ',')
478 i = stdscan(NULL, &tokval);
479 } else if (i == '-' || i == '+') {
480 char *save = stdscan_get();
481 int token = i;
482 sign = (i == '-') ? -1 : 1;
483 i = stdscan(NULL, &tokval);
484 if (i != TOKEN_FLOAT) {
485 stdscan_set(save);
486 i = tokval.t_type = token;
487 goto is_expression;
488 } else {
489 goto is_float;
490 }
491 } else if (i == TOKEN_FLOAT) {
492 is_float:
493 eop->type = EOT_DB_STRING;
494 result->eops_float = true;
495
496 eop->stringlen = idata_bytes(result->opcode);
497 if (eop->stringlen > 16) {
498 nasm_error(ERR_NONFATAL, "floating-point constant"
499 " encountered in DY or DZ instruction");
500 eop->stringlen = 0;
501 } else if (eop->stringlen < 1) {
502 nasm_error(ERR_NONFATAL, "floating-point constant"
503 " encountered in unknown instruction");
504 /*
505 * fix suggested by Pedro Gimeno... original line was:
506 * eop->type = EOT_NOTHING;
507 */
508 eop->stringlen = 0;
509 }
510
511 eop = nasm_realloc(eop, sizeof(extop) + eop->stringlen);
512 tail = &eop->next;
513 *fixptr = eop;
514 eop->stringval = (char *)eop + sizeof(extop);
515 if (!eop->stringlen ||
516 !float_const(tokval.t_charptr, sign,
517 (uint8_t *)eop->stringval,
518 eop->stringlen, nasm_error))
519 eop->type = EOT_NOTHING;
520 i = stdscan(NULL, &tokval); /* eat the comma */
521 } else {
522 /* anything else, assume it is an expression */
523 expr *value;
524
525 is_expression:
526 value = evaluate(stdscan, NULL, &tokval, NULL,
527 critical, nasm_error, NULL);
528 i = tokval.t_type;
529 if (!value) { /* error in evaluator */
530 result->opcode = I_none; /* unrecoverable parse error: */
531 return result; /* ignore this instruction */
532 }
533 if (is_unknown(value)) {
534 eop->type = EOT_DB_NUMBER;
535 eop->offset = 0; /* doesn't matter what we put */
536 eop->segment = eop->wrt = NO_SEG; /* likewise */
537 } else if (is_reloc(value)) {
538 eop->type = EOT_DB_NUMBER;
539 eop->offset = reloc_value(value);
540 eop->segment = reloc_seg(value);
541 eop->wrt = reloc_wrt(value);
542 } else {
543 nasm_error(ERR_NONFATAL,
544 "operand %d: expression is not simple"
545 " or relocatable", oper_num);
546 }
547 }
548
549 /*
550 * We're about to call stdscan(), which will eat the
551 * comma that we're currently sitting on between
552 * arguments. However, we'd better check first that it
553 * _is_ a comma.
554 */
555 if (i == TOKEN_EOS) /* also could be EOL */
556 break;
557 if (i != ',') {
558 nasm_error(ERR_NONFATAL, "comma expected after operand %d",
559 oper_num);
560 result->opcode = I_none;/* unrecoverable parse error: */
561 return result; /* ignore this instruction */
562 }
563 }
564
565 if (result->opcode == I_INCBIN) {
566 /*
567 * Correct syntax for INCBIN is that there should be
568 * one string operand, followed by one or two numeric
569 * operands.
570 */
571 if (!result->eops || result->eops->type != EOT_DB_STRING)
572 nasm_error(ERR_NONFATAL, "`incbin' expects a file name");
573 else if (result->eops->next &&
574 result->eops->next->type != EOT_DB_NUMBER)
575 nasm_error(ERR_NONFATAL, "`incbin': second parameter is"
576 " non-numeric");
577 else if (result->eops->next && result->eops->next->next &&
578 result->eops->next->next->type != EOT_DB_NUMBER)
579 nasm_error(ERR_NONFATAL, "`incbin': third parameter is"
580 " non-numeric");
581 else if (result->eops->next && result->eops->next->next &&
582 result->eops->next->next->next)
583 nasm_error(ERR_NONFATAL,
584 "`incbin': more than three parameters");
585 else
586 return result;
587 /*
588 * If we reach here, one of the above errors happened.
589 * Throw the instruction away.
590 */
591 result->opcode = I_none;
592 return result;
593 } else /* DB ... */ if (oper_num == 0)
594 nasm_error(ERR_WARNING | ERR_PASS1,
595 "no operand for data declaration");
596 else
597 result->operands = oper_num;
598
599 return result;
600 }
601
602 /*
603 * Now we begin to parse the operands. There may be up to four
604 * of these, separated by commas, and terminated by a zero token.
605 */
606
607 for (operand = 0; operand < MAX_OPERANDS; operand++) {
608 expr *value; /* used most of the time */
609 int mref; /* is this going to be a memory ref? */
610 int bracket; /* is it a [] mref, or a & mref? */
611 int setsize = 0;
612 decoflags_t brace_flags = 0; /* flags for decorators in braces */
613
614 result->oprs[operand].disp_size = 0; /* have to zero this whatever */
615 result->oprs[operand].eaflags = 0; /* and this */
616 result->oprs[operand].opflags = 0;
617 result->oprs[operand].decoflags = 0;
618
619 i = stdscan(NULL, &tokval);
620 if (i == TOKEN_EOS)
621 break; /* end of operands: get out of here */
622 else if (first && i == ':') {
623 insn_is_label = true;
624 goto restart_parse;
625 }
626 first = false;
627 result->oprs[operand].type = 0; /* so far, no override */
628 while (i == TOKEN_SPECIAL) { /* size specifiers */
629 switch ((int)tokval.t_integer) {
630 case S_BYTE:
631 if (!setsize) /* we want to use only the first */
632 result->oprs[operand].type |= BITS8;
633 setsize = 1;
634 break;
635 case S_WORD:
636 if (!setsize)
637 result->oprs[operand].type |= BITS16;
638 setsize = 1;
639 break;
640 case S_DWORD:
641 case S_LONG:
642 if (!setsize)
643 result->oprs[operand].type |= BITS32;
644 setsize = 1;
645 break;
646 case S_QWORD:
647 if (!setsize)
648 result->oprs[operand].type |= BITS64;
649 setsize = 1;
650 break;
651 case S_TWORD:
652 if (!setsize)
653 result->oprs[operand].type |= BITS80;
654 setsize = 1;
655 break;
656 case S_OWORD:
657 if (!setsize)
658 result->oprs[operand].type |= BITS128;
659 setsize = 1;
660 break;
661 case S_YWORD:
662 if (!setsize)
663 result->oprs[operand].type |= BITS256;
664 setsize = 1;
665 break;
666 case S_ZWORD:
667 if (!setsize)
668 result->oprs[operand].type |= BITS512;
669 setsize = 1;
670 break;
671 case S_TO:
672 result->oprs[operand].type |= TO;
673 break;
674 case S_STRICT:
675 result->oprs[operand].type |= STRICT;
676 break;
677 case S_FAR:
678 result->oprs[operand].type |= FAR;
679 break;
680 case S_NEAR:
681 result->oprs[operand].type |= NEAR;
682 break;
683 case S_SHORT:
684 result->oprs[operand].type |= SHORT;
685 break;
686 default:
687 nasm_error(ERR_NONFATAL, "invalid operand size specification");
688 }
689 i = stdscan(NULL, &tokval);
690 }
691
692 if (i == '[' || i == '&') { /* memory reference */
693 mref = true;
694 bracket = (i == '[');
695 i = stdscan(NULL, &tokval); /* then skip the colon */
696 while (i == TOKEN_SPECIAL || i == TOKEN_PREFIX) {
697 process_size_override(result, operand);
698 i = stdscan(NULL, &tokval);
699 }
700 } else { /* immediate operand, or register */
701 mref = false;
702 bracket = false; /* placate optimisers */
703 }
704
705 if ((result->oprs[operand].type & FAR) && !mref &&
706 result->opcode != I_JMP && result->opcode != I_CALL) {
707 nasm_error(ERR_NONFATAL, "invalid use of FAR operand specifier");
708 }
709
710 value = evaluate(stdscan, NULL, &tokval,
711 &result->oprs[operand].opflags,
712 critical, nasm_error, &hints);
713 i = tokval.t_type;
714 if (result->oprs[operand].opflags & OPFLAG_FORWARD) {
715 result->forw_ref = true;
716 }
717 if (!value) { /* nasm_error in evaluator */
718 result->opcode = I_none; /* unrecoverable parse error: */
719 return result; /* ignore this instruction */
720 }
721 if (i == ':' && mref) { /* it was seg:offset */
722 /*
723 * Process the segment override.
724 */
725 if (value[1].type != 0 ||
726 value->value != 1 ||
727 !IS_SREG(value->type))
728 nasm_error(ERR_NONFATAL, "invalid segment override");
729 else if (result->prefixes[PPS_SEG])
730 nasm_error(ERR_NONFATAL,
731 "instruction has conflicting segment overrides");
732 else {
733 result->prefixes[PPS_SEG] = value->type;
734 if (IS_FSGS(value->type))
735 result->oprs[operand].eaflags |= EAF_FSGS;
736 }
737
738 i = stdscan(NULL, &tokval); /* then skip the colon */
739 while (i == TOKEN_SPECIAL || i == TOKEN_PREFIX) {
740 process_size_override(result, operand);
741 i = stdscan(NULL, &tokval);
742 }
743 value = evaluate(stdscan, NULL, &tokval,
744 &result->oprs[operand].opflags,
745 critical, nasm_error, &hints);
746 i = tokval.t_type;
747 if (result->oprs[operand].opflags & OPFLAG_FORWARD) {
748 result->forw_ref = true;
749 }
750 /* and get the offset */
751 if (!value) { /* but, error in evaluator */
752 result->opcode = I_none; /* unrecoverable parse error: */
753 return result; /* ignore this instruction */
754 }
755 }
756
757 recover = false;
758 if (mref && bracket) { /* find ] at the end */
759 if (i != ']') {
760 nasm_error(ERR_NONFATAL, "parser: expecting ]");
761 recover = true;
762 } else { /* we got the required ] */
763 i = stdscan(NULL, &tokval);
764 if ((i == TOKEN_DECORATOR) || (i == TOKEN_OPMASK)) {
765 /*
766 * according to AVX512 spec, broacast or opmask decorator
767 * is expected for memory reference operands
768 */
769 if (tokval.t_flag & TFLAG_BRDCAST) {
770 brace_flags |= GEN_BRDCAST(0);
771 i = stdscan(NULL, &tokval);
772 } else if (i == TOKEN_OPMASK) {
773 brace_flags |= VAL_OPMASK(nasm_regvals[tokval.t_integer]);
774 i = stdscan(NULL, &tokval);
775 } else {
776 nasm_error(ERR_NONFATAL, "broadcast or opmask "
777 "decorator expected inside braces");
778 recover = true;
779 }
780 }
781
782 if (i != 0 && i != ',') {
783 nasm_error(ERR_NONFATAL, "comma or end of line expected");
784 recover = true;
785 }
786 }
787 } else { /* immediate operand */
788 if (i != 0 && i != ',' && i != ':' &&
789 i != TOKEN_DECORATOR && i != TOKEN_OPMASK) {
790 nasm_error(ERR_NONFATAL, "comma, colon, decorator or end of "
791 "line expected after operand");
792 recover = true;
793 } else if (i == ':') {
794 result->oprs[operand].type |= COLON;
795 } else if (i == TOKEN_DECORATOR || i == TOKEN_OPMASK) {
796 /* parse opmask (and zeroing) after an operand */
797 recover = parse_braces(&brace_flags);
798 }
799 }
800 if (recover) {
801 do { /* error recovery */
802 i = stdscan(NULL, &tokval);
803 } while (i != 0 && i != ',');
804 }
805
806 /*
807 * now convert the exprs returned from evaluate()
808 * into operand descriptions...
809 */
810
811 if (mref) { /* it's a memory reference */
812 expr *e = value;
813 int b, i, s; /* basereg, indexreg, scale */
814 int64_t o; /* offset */
815
816 b = i = -1, o = s = 0;
817 result->oprs[operand].hintbase = hints.base;
818 result->oprs[operand].hinttype = hints.type;
819
820 if (e->type && e->type <= EXPR_REG_END) { /* this bit's a register */
821 bool is_gpr = is_class(REG_GPR,nasm_reg_flags[e->type]);
822
823 if (is_gpr && e->value == 1)
824 b = e->type; /* It can be basereg */
825 else /* No, it has to be indexreg */
826 i = e->type, s = e->value;
827 e++;
828 }
829 if (e->type && e->type <= EXPR_REG_END) { /* it's a 2nd register */
830 bool is_gpr = is_class(REG_GPR,nasm_reg_flags[e->type]);
831
832 if (b != -1) /* If the first was the base, ... */
833 i = e->type, s = e->value; /* second has to be indexreg */
834
835 else if (!is_gpr || e->value != 1) {
836 /* If both want to be index */
837 nasm_error(ERR_NONFATAL,
838 "invalid effective address: two index registers");
839 result->opcode = I_none;
840 return result;
841 } else
842 b = e->type;
843 e++;
844 }
845 if (e->type != 0) { /* is there an offset? */
846 if (e->type <= EXPR_REG_END) { /* in fact, is there an error? */
847 nasm_error(ERR_NONFATAL,
848 "beroset-p-603-invalid effective address");
849 result->opcode = I_none;
850 return result;
851 } else {
852 if (e->type == EXPR_UNKNOWN) {
853 result->oprs[operand].opflags |= OPFLAG_UNKNOWN;
854 o = 0; /* doesn't matter what */
855 result->oprs[operand].wrt = NO_SEG; /* nor this */
856 result->oprs[operand].segment = NO_SEG; /* or this */
857 while (e->type)
858 e++; /* go to the end of the line */
859 } else {
860 if (e->type == EXPR_SIMPLE) {
861 o = e->value;
862 e++;
863 }
864 if (e->type == EXPR_WRT) {
865 result->oprs[operand].wrt = e->value;
866 e++;
867 } else
868 result->oprs[operand].wrt = NO_SEG;
869 /*
870 * Look for a segment base type.
871 */
872 if (e->type && e->type < EXPR_SEGBASE) {
873 nasm_error(ERR_NONFATAL,
874 "beroset-p-630-invalid effective address");
875 result->opcode = I_none;
876 return result;
877 }
878 while (e->type && e->value == 0)
879 e++;
880 if (e->type && e->value != 1) {
881 nasm_error(ERR_NONFATAL,
882 "beroset-p-637-invalid effective address");
883 result->opcode = I_none;
884 return result;
885 }
886 if (e->type) {
887 result->oprs[operand].segment =
888 e->type - EXPR_SEGBASE;
889 e++;
890 } else
891 result->oprs[operand].segment = NO_SEG;
892 while (e->type && e->value == 0)
893 e++;
894 if (e->type) {
895 nasm_error(ERR_NONFATAL,
896 "beroset-p-650-invalid effective address");
897 result->opcode = I_none;
898 return result;
899 }
900 }
901 }
902 } else {
903 o = 0;
904 result->oprs[operand].wrt = NO_SEG;
905 result->oprs[operand].segment = NO_SEG;
906 }
907
908 if (e->type != 0) { /* there'd better be nothing left! */
909 nasm_error(ERR_NONFATAL,
910 "beroset-p-663-invalid effective address");
911 result->opcode = I_none;
912 return result;
913 }
914
915 /* It is memory, but it can match any r/m operand */
916 result->oprs[operand].type |= MEMORY_ANY;
917
918 if (b == -1 && (i == -1 || s == 0)) {
919 int is_rel = globalbits == 64 &&
920 !(result->oprs[operand].eaflags & EAF_ABS) &&
921 ((globalrel &&
922 !(result->oprs[operand].eaflags & EAF_FSGS)) ||
923 (result->oprs[operand].eaflags & EAF_REL));
924
925 result->oprs[operand].type |= is_rel ? IP_REL : MEM_OFFS;
926 }
927
928 if (i != -1) {
929 opflags_t iclass = nasm_reg_flags[i];
930
931 if (is_class(XMMREG,iclass))
932 result->oprs[operand].type |= XMEM;
933 else if (is_class(YMMREG,iclass))
934 result->oprs[operand].type |= YMEM;
935 else if (is_class(ZMMREG,iclass))
936 result->oprs[operand].type |= ZMEM;
937 }
938
939 result->oprs[operand].basereg = b;
940 result->oprs[operand].indexreg = i;
941 result->oprs[operand].scale = s;
942 result->oprs[operand].offset = o;
943 result->oprs[operand].decoflags |= brace_flags;
944 } else { /* it's not a memory reference */
945 if (is_just_unknown(value)) { /* it's immediate but unknown */
946 result->oprs[operand].type |= IMMEDIATE;
947 result->oprs[operand].opflags |= OPFLAG_UNKNOWN;
948 result->oprs[operand].offset = 0; /* don't care */
949 result->oprs[operand].segment = NO_SEG; /* don't care again */
950 result->oprs[operand].wrt = NO_SEG; /* still don't care */
951
952 if(optimizing >= 0 && !(result->oprs[operand].type & STRICT)) {
953 /* Be optimistic */
954 result->oprs[operand].type |=
955 UNITY | SBYTEWORD | SBYTEDWORD | UDWORD | SDWORD;
956 }
957 } else if (is_reloc(value)) { /* it's immediate */
958 result->oprs[operand].type |= IMMEDIATE;
959 result->oprs[operand].offset = reloc_value(value);
960 result->oprs[operand].segment = reloc_seg(value);
961 result->oprs[operand].wrt = reloc_wrt(value);
962
963 if (is_simple(value)) {
964 uint64_t n = reloc_value(value);
965 if (n == 1)
966 result->oprs[operand].type |= UNITY;
967 if (optimizing >= 0 &&
968 !(result->oprs[operand].type & STRICT)) {
969 if ((uint32_t) (n + 128) <= 255)
970 result->oprs[operand].type |= SBYTEDWORD;
971 if ((uint16_t) (n + 128) <= 255)
972 result->oprs[operand].type |= SBYTEWORD;
973 if (n <= 0xFFFFFFFF)
974 result->oprs[operand].type |= UDWORD;
975 if (n + 0x80000000 <= 0xFFFFFFFF)
976 result->oprs[operand].type |= SDWORD;
977 }
978 }
979 } else if(value->type == EXPR_RDSAE) {
980 /*
981 * it's not an operand but a rounding or SAE decorator.
982 * put the decorator information in the (opflag_t) type field
983 * of previous operand.
984 */
985 operand --;
986 switch (value->value) {
987 case BRC_RN:
988 case BRC_RU:
989 case BRC_RD:
990 case BRC_RZ:
991 case BRC_SAE:
992 result->oprs[operand].decoflags |=
993 (value->value == BRC_SAE ? SAE : ER);
994 result->evex_rm = value->value;
995 break;
996 default:
997 nasm_error(ERR_NONFATAL, "invalid decorator");
998 break;
999 }
1000 } else { /* it's a register */
1001 opflags_t rs;
1002
1003 if (value->type >= EXPR_SIMPLE || value->value != 1) {
1004 nasm_error(ERR_NONFATAL, "invalid operand type");
1005 result->opcode = I_none;
1006 return result;
1007 }
1008
1009 /*
1010 * check that its only 1 register, not an expression...
1011 */
1012 for (i = 1; value[i].type; i++)
1013 if (value[i].value) {
1014 nasm_error(ERR_NONFATAL, "invalid operand type");
1015 result->opcode = I_none;
1016 return result;
1017 }
1018
1019 /* clear overrides, except TO which applies to FPU regs */
1020 if (result->oprs[operand].type & ~TO) {
1021 /*
1022 * we want to produce a warning iff the specified size
1023 * is different from the register size
1024 */
1025 rs = result->oprs[operand].type & SIZE_MASK;
1026 } else
1027 rs = 0;
1028
1029 result->oprs[operand].type &= TO;
1030 result->oprs[operand].type |= REGISTER;
1031 result->oprs[operand].type |= nasm_reg_flags[value->type];
1032 result->oprs[operand].decoflags |= brace_flags;
1033 result->oprs[operand].basereg = value->type;
1034
1035 if (rs && (result->oprs[operand].type & SIZE_MASK) != rs)
1036 nasm_error(ERR_WARNING | ERR_PASS1,
1037 "register size specification ignored");
1038 }
1039 }
1040
1041 /* remember the position of operand having broadcasting/ER mode */
1042 if (result->oprs[operand].decoflags & (BRDCAST_MASK | ER | SAE))
1043 result->evex_brerop = operand;
1044 }
1045
1046 result->operands = operand; /* set operand count */
1047
1048 /* clear remaining operands */
1049 while (operand < MAX_OPERANDS)
1050 result->oprs[operand++].type = 0;
1051
1052 /*
1053 * Transform RESW, RESD, RESQ, REST, RESO, RESY, RESZ into RESB.
1054 */
1055 switch (result->opcode) {
1056 case I_RESW:
1057 result->opcode = I_RESB;
1058 result->oprs[0].offset *= 2;
1059 break;
1060 case I_RESD:
1061 result->opcode = I_RESB;
1062 result->oprs[0].offset *= 4;
1063 break;
1064 case I_RESQ:
1065 result->opcode = I_RESB;
1066 result->oprs[0].offset *= 8;
1067 break;
1068 case I_REST:
1069 result->opcode = I_RESB;
1070 result->oprs[0].offset *= 10;
1071 break;
1072 case I_RESO:
1073 result->opcode = I_RESB;
1074 result->oprs[0].offset *= 16;
1075 break;
1076 case I_RESY:
1077 result->opcode = I_RESB;
1078 result->oprs[0].offset *= 32;
1079 break;
1080 case I_RESZ:
1081 result->opcode = I_RESB;
1082 result->oprs[0].offset *= 64;
1083 break;
1084 default:
1085 break;
1086 }
1087
1088 return result;
1089 }
1090
1091 static int is_comma_next(void)
1092 {
1093 struct tokenval tv;
1094 char *p;
1095 int i;
1096
1097 p = stdscan_get();
1098 i = stdscan(NULL, &tv);
1099 stdscan_set(p);
1100
1101 return (i == ',' || i == ';' || !i);
1102 }
1103
1104 void cleanup_insn(insn * i)
1105 {
1106 extop *e;
1107
1108 while ((e = i->eops)) {
1109 i->eops = e->next;
1110 if (e->type == EOT_DB_STRING_FREE)
1111 nasm_free(e->stringval);
1112 nasm_free(e);
1113 }
1114 }
The Netwide Assembler