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
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 result
->forw_ref
= false;
179 stdscan_bufptr
= buffer
;
180 i
= stdscan(NULL
, &tokval
);
182 result
->label
= NULL
; /* Assume no label */
183 result
->eops
= NULL
; /* must do this, whatever happens */
184 result
->operands
= 0; /* must initialize this */
186 if (i
== 0) { /* blank line - ignore */
187 result
->opcode
= -1; /* and no instruction either */
190 if (i
!= TOKEN_ID
&& i
!= TOKEN_INSN
&& i
!= TOKEN_PREFIX
&&
191 (i
!= TOKEN_REG
|| (REG_SREG
& ~reg_flags
[tokval
.t_integer
]))) {
192 error(ERR_NONFATAL
, "label or instruction expected"
193 " at start of line");
198 if (i
== TOKEN_ID
) { /* there's a label here */
199 result
->label
= tokval
.t_charptr
;
200 i
= stdscan(NULL
, &tokval
);
201 if (i
== ':') { /* skip over the optional colon */
202 i
= stdscan(NULL
, &tokval
);
204 error(ERR_WARNING
| ERR_WARN_OL
| ERR_PASS1
,
205 "label alone on a line without a colon might be in error");
207 if (i
!= TOKEN_INSN
|| tokval
.t_integer
!= I_EQU
) {
209 * FIXME: location->segment could be NO_SEG, in which case
210 * it is possible we should be passing 'abs_seg'. Look into this.
211 * Work out whether that is *really* what we should be doing.
212 * Generally fix things. I think this is right as it is, but
213 * am still not certain.
215 ldef(result
->label
, in_abs_seg
? abs_seg
: location
->segment
,
216 location
->offset
, NULL
, true, false, outfmt
, errfunc
);
221 result
->opcode
= -1; /* this line contains just a label */
225 for (j
= 0; j
< MAXPREFIX
; j
++)
226 result
->prefixes
[j
] = P_none
;
229 while (i
== TOKEN_PREFIX
||
230 (i
== TOKEN_REG
&& !(REG_SREG
& ~reg_flags
[tokval
.t_integer
])))
233 * Handle special case: the TIMES prefix.
235 if (i
== TOKEN_PREFIX
&& tokval
.t_integer
== P_TIMES
) {
238 i
= stdscan(NULL
, &tokval
);
240 evaluate(stdscan
, NULL
, &tokval
, NULL
, pass0
, error
, NULL
);
242 if (!value
) { /* but, error in evaluator */
243 result
->opcode
= -1; /* unrecoverable parse error: */
244 return result
; /* ignore this instruction */
246 if (!is_simple(value
)) {
248 "non-constant argument supplied to TIMES");
251 result
->times
= value
->value
;
252 if (value
->value
< 0) {
253 error(ERR_NONFATAL
, "TIMES value %d is negative",
259 int slot
= prefix_slot(tokval
.t_integer
);
260 if (result
->prefixes
[slot
]) {
261 if (result
->prefixes
[slot
] == tokval
.t_integer
)
263 "instruction has redundant prefixes");
266 "instruction has conflicting prefixes");
268 result
->prefixes
[slot
] = tokval
.t_integer
;
269 i
= stdscan(NULL
, &tokval
);
273 if (i
!= TOKEN_INSN
) {
277 for (j
= 0; j
< MAXPREFIX
; j
++)
278 if ((pfx
= result
->prefixes
[j
]) != P_none
)
281 if (i
== 0 && pfx
!= P_none
) {
283 * Instruction prefixes are present, but no actual
284 * instruction. This is allowed: at this point we
285 * invent a notional instruction of RESB 0.
287 result
->opcode
= I_RESB
;
288 result
->operands
= 1;
289 result
->oprs
[0].type
= IMMEDIATE
;
290 result
->oprs
[0].offset
= 0L;
291 result
->oprs
[0].segment
= result
->oprs
[0].wrt
= NO_SEG
;
294 error(ERR_NONFATAL
, "parser: instruction expected");
300 result
->opcode
= tokval
.t_integer
;
301 result
->condition
= tokval
.t_inttwo
;
304 * RESB, RESW and RESD cannot be satisfied with incorrectly
305 * evaluated operands, since the correct values _must_ be known
306 * on the first pass. Hence, even in pass one, we set the
307 * `critical' flag on calling evaluate(), so that it will bomb
308 * out on undefined symbols. Nasty, but there's nothing we can
311 * For the moment, EQU has the same difficulty, so we'll
314 if (result
->opcode
== I_RESB
|| result
->opcode
== I_RESW
||
315 result
->opcode
== I_RESD
|| result
->opcode
== I_RESQ
||
316 result
->opcode
== I_REST
|| result
->opcode
== I_RESO
||
317 result
->opcode
== I_EQU
|| result
->opcode
== I_INCBIN
) {
320 critical
= (pass
== 2 ? 2 : 0);
322 if (result
->opcode
== I_DB
|| result
->opcode
== I_DW
||
323 result
->opcode
== I_DD
|| result
->opcode
== I_DQ
||
324 result
->opcode
== I_DT
|| result
->opcode
== I_DO
||
325 result
->opcode
== I_INCBIN
) {
326 extop
*eop
, **tail
= &result
->eops
, **fixptr
;
329 result
->eops_float
= false;
332 * Begin to read the DB/DW/DD/DQ/DT/DO/INCBIN operands.
335 i
= stdscan(NULL
, &tokval
);
339 eop
= *tail
= nasm_malloc(sizeof(extop
));
342 eop
->type
= EOT_NOTHING
;
345 if (i
== TOKEN_NUM
&& tokval
.t_charptr
&& is_comma_next()) {
346 eop
->type
= EOT_DB_STRING
;
347 eop
->stringval
= tokval
.t_charptr
;
348 eop
->stringlen
= tokval
.t_inttwo
;
349 i
= stdscan(NULL
, &tokval
); /* eat the comma */
353 if ((i
== TOKEN_FLOAT
&& is_comma_next())
354 || i
== '-' || i
== '+') {
357 if (i
== '+' || i
== '-') {
358 char *save
= stdscan_bufptr
;
360 sign
= (i
== '-') ? -1 : 1;
361 i
= stdscan(NULL
, &tokval
);
362 if (i
!= TOKEN_FLOAT
|| !is_comma_next()) {
363 stdscan_bufptr
= save
;
364 i
= tokval
.t_type
= token
;
368 if (i
== TOKEN_FLOAT
) {
369 eop
->type
= EOT_DB_STRING
;
370 result
->eops_float
= true;
371 switch (result
->opcode
) {
391 error(ERR_NONFATAL
, "floating-point constant"
392 " encountered in unknown instruction");
394 * fix suggested by Pedro Gimeno... original line
396 * eop->type = EOT_NOTHING;
401 eop
= nasm_realloc(eop
, sizeof(extop
) + eop
->stringlen
);
404 eop
->stringval
= (char *)eop
+ sizeof(extop
);
405 if (!eop
->stringlen
||
406 !float_const(tokval
.t_charptr
, sign
,
407 (uint8_t *)eop
->stringval
,
408 eop
->stringlen
, error
))
409 eop
->type
= EOT_NOTHING
;
410 i
= stdscan(NULL
, &tokval
); /* eat the comma */
418 value
= evaluate(stdscan
, NULL
, &tokval
, NULL
,
419 critical
, error
, NULL
);
421 if (!value
) { /* error in evaluator */
422 result
->opcode
= -1; /* unrecoverable parse error: */
423 return result
; /* ignore this instruction */
425 if (is_unknown(value
)) {
426 eop
->type
= EOT_DB_NUMBER
;
427 eop
->offset
= 0; /* doesn't matter what we put */
428 eop
->segment
= eop
->wrt
= NO_SEG
; /* likewise */
429 } else if (is_reloc(value
)) {
430 eop
->type
= EOT_DB_NUMBER
;
431 eop
->offset
= reloc_value(value
);
432 eop
->segment
= reloc_seg(value
);
433 eop
->wrt
= reloc_wrt(value
);
436 "operand %d: expression is not simple"
437 " or relocatable", oper_num
);
442 * We're about to call stdscan(), which will eat the
443 * comma that we're currently sitting on between
444 * arguments. However, we'd better check first that it
447 if (i
== 0) /* also could be EOL */
450 error(ERR_NONFATAL
, "comma expected after operand %d",
452 result
->opcode
= -1; /* unrecoverable parse error: */
453 return result
; /* ignore this instruction */
457 if (result
->opcode
== I_INCBIN
) {
459 * Correct syntax for INCBIN is that there should be
460 * one string operand, followed by one or two numeric
463 if (!result
->eops
|| result
->eops
->type
!= EOT_DB_STRING
)
464 error(ERR_NONFATAL
, "`incbin' expects a file name");
465 else if (result
->eops
->next
&&
466 result
->eops
->next
->type
!= EOT_DB_NUMBER
)
467 error(ERR_NONFATAL
, "`incbin': second parameter is",
469 else if (result
->eops
->next
&& result
->eops
->next
->next
&&
470 result
->eops
->next
->next
->type
!= EOT_DB_NUMBER
)
471 error(ERR_NONFATAL
, "`incbin': third parameter is",
473 else if (result
->eops
->next
&& result
->eops
->next
->next
&&
474 result
->eops
->next
->next
->next
)
476 "`incbin': more than three parameters");
480 * If we reach here, one of the above errors happened.
481 * Throw the instruction away.
485 } else /* DB ... */ if (oper_num
== 0)
486 error(ERR_WARNING
| ERR_PASS1
,
487 "no operand for data declaration");
489 result
->operands
= oper_num
;
494 /* right. Now we begin to parse the operands. There may be up to four
495 * of these, separated by commas, and terminated by a zero token. */
497 for (operand
= 0; operand
< MAX_OPERANDS
; operand
++) {
498 expr
*value
; /* used most of the time */
499 int mref
; /* is this going to be a memory ref? */
500 int bracket
; /* is it a [] mref, or a & mref? */
503 result
->oprs
[operand
].disp_size
= 0; /* have to zero this whatever */
504 result
->oprs
[operand
].eaflags
= 0; /* and this */
505 result
->oprs
[operand
].opflags
= 0;
507 i
= stdscan(NULL
, &tokval
);
509 break; /* end of operands: get out of here */
510 result
->oprs
[operand
].type
= 0; /* so far, no override */
511 while (i
== TOKEN_SPECIAL
) { /* size specifiers */
512 switch ((int)tokval
.t_integer
) {
514 if (!setsize
) /* we want to use only the first */
515 result
->oprs
[operand
].type
|= BITS8
;
520 result
->oprs
[operand
].type
|= BITS16
;
526 result
->oprs
[operand
].type
|= BITS32
;
531 result
->oprs
[operand
].type
|= BITS64
;
536 result
->oprs
[operand
].type
|= BITS80
;
541 result
->oprs
[operand
].type
|= BITS128
;
545 result
->oprs
[operand
].type
|= TO
;
548 result
->oprs
[operand
].type
|= STRICT
;
551 result
->oprs
[operand
].type
|= FAR
;
554 result
->oprs
[operand
].type
|= NEAR
;
557 result
->oprs
[operand
].type
|= SHORT
;
560 error(ERR_NONFATAL
, "invalid operand size specification");
562 i
= stdscan(NULL
, &tokval
);
565 if (i
== '[' || i
== '&') { /* memory reference */
567 bracket
= (i
== '[');
568 i
= stdscan(NULL
, &tokval
); /* then skip the colon */
569 while (i
== TOKEN_SPECIAL
|| i
== TOKEN_PREFIX
) {
570 process_size_override(result
, operand
);
571 i
= stdscan(NULL
, &tokval
);
573 } else { /* immediate operand, or register */
575 bracket
= false; /* placate optimisers */
578 if ((result
->oprs
[operand
].type
& FAR
) && !mref
&&
579 result
->opcode
!= I_JMP
&& result
->opcode
!= I_CALL
) {
580 error(ERR_NONFATAL
, "invalid use of FAR operand specifier");
583 value
= evaluate(stdscan
, NULL
, &tokval
,
584 &result
->oprs
[operand
].opflags
,
585 critical
, error
, &hints
);
587 if (result
->oprs
[operand
].opflags
& OPFLAG_FORWARD
) {
588 result
->forw_ref
= true;
590 if (!value
) { /* error in evaluator */
591 result
->opcode
= -1; /* unrecoverable parse error: */
592 return result
; /* ignore this instruction */
594 if (i
== ':' && mref
) { /* it was seg:offset */
596 * Process the segment override.
598 if (value
[1].type
!= 0 || value
->value
!= 1 ||
599 REG_SREG
& ~reg_flags
[value
->type
])
600 error(ERR_NONFATAL
, "invalid segment override");
601 else if (result
->prefixes
[PPS_SEG
])
603 "instruction has conflicting segment overrides");
605 result
->prefixes
[PPS_SEG
] = value
->type
;
606 if (!(REG_FSGS
& ~reg_flags
[value
->type
]))
607 result
->oprs
[operand
].eaflags
|= EAF_FSGS
;
610 i
= stdscan(NULL
, &tokval
); /* then skip the colon */
611 while (i
== TOKEN_SPECIAL
|| i
== TOKEN_PREFIX
) {
612 process_size_override(result
, operand
);
613 i
= stdscan(NULL
, &tokval
);
615 value
= evaluate(stdscan
, NULL
, &tokval
,
616 &result
->oprs
[operand
].opflags
,
617 critical
, error
, &hints
);
619 if (result
->oprs
[operand
].opflags
& OPFLAG_FORWARD
) {
620 result
->forw_ref
= true;
622 /* and get the offset */
623 if (!value
) { /* but, error in evaluator */
624 result
->opcode
= -1; /* unrecoverable parse error: */
625 return result
; /* ignore this instruction */
628 if (mref
&& bracket
) { /* find ] at the end */
630 error(ERR_NONFATAL
, "parser: expecting ]");
631 do { /* error recovery again */
632 i
= stdscan(NULL
, &tokval
);
633 } while (i
!= 0 && i
!= ',');
634 } else /* we got the required ] */
635 i
= stdscan(NULL
, &tokval
);
636 } else { /* immediate operand */
637 if (i
!= 0 && i
!= ',' && i
!= ':') {
638 error(ERR_NONFATAL
, "comma or end of line expected");
639 do { /* error recovery */
640 i
= stdscan(NULL
, &tokval
);
641 } while (i
!= 0 && i
!= ',');
642 } else if (i
== ':') {
643 result
->oprs
[operand
].type
|= COLON
;
647 /* now convert the exprs returned from evaluate() into operand
650 if (mref
) { /* it's a memory reference */
652 int b
, i
, s
; /* basereg, indexreg, scale */
653 int64_t o
; /* offset */
655 b
= i
= -1, o
= s
= 0;
656 result
->oprs
[operand
].hintbase
= hints
.base
;
657 result
->oprs
[operand
].hinttype
= hints
.type
;
659 if (e
->type
&& e
->type
<= EXPR_REG_END
) { /* this bit's a register */
660 if (e
->value
== 1) /* in fact it can be basereg */
662 else /* no, it has to be indexreg */
663 i
= e
->type
, s
= e
->value
;
666 if (e
->type
&& e
->type
<= EXPR_REG_END
) { /* it's a 2nd register */
667 if (b
!= -1) /* If the first was the base, ... */
668 i
= e
->type
, s
= e
->value
; /* second has to be indexreg */
670 else if (e
->value
!= 1) { /* If both want to be index */
672 "beroset-p-592-invalid effective address");
679 if (e
->type
!= 0) { /* is there an offset? */
680 if (e
->type
<= EXPR_REG_END
) { /* in fact, is there an error? */
682 "beroset-p-603-invalid effective address");
686 if (e
->type
== EXPR_UNKNOWN
) {
687 o
= 0; /* doesn't matter what */
688 result
->oprs
[operand
].wrt
= NO_SEG
; /* nor this */
689 result
->oprs
[operand
].segment
= NO_SEG
; /* or this */
691 e
++; /* go to the end of the line */
693 if (e
->type
== EXPR_SIMPLE
) {
697 if (e
->type
== EXPR_WRT
) {
698 result
->oprs
[operand
].wrt
= e
->value
;
701 result
->oprs
[operand
].wrt
= NO_SEG
;
703 * Look for a segment base type.
705 if (e
->type
&& e
->type
< EXPR_SEGBASE
) {
707 "beroset-p-630-invalid effective address");
711 while (e
->type
&& e
->value
== 0)
713 if (e
->type
&& e
->value
!= 1) {
715 "beroset-p-637-invalid effective address");
720 result
->oprs
[operand
].segment
=
721 e
->type
- EXPR_SEGBASE
;
724 result
->oprs
[operand
].segment
= NO_SEG
;
725 while (e
->type
&& e
->value
== 0)
729 "beroset-p-650-invalid effective address");
737 result
->oprs
[operand
].wrt
= NO_SEG
;
738 result
->oprs
[operand
].segment
= NO_SEG
;
741 if (e
->type
!= 0) { /* there'd better be nothing left! */
743 "beroset-p-663-invalid effective address");
748 /* It is memory, but it can match any r/m operand */
749 result
->oprs
[operand
].type
|= MEMORY_ANY
;
751 if (b
== -1 && (i
== -1 || s
== 0)) {
752 int is_rel
= globalbits
== 64 &&
753 !(result
->oprs
[operand
].eaflags
& EAF_ABS
) &&
755 !(result
->oprs
[operand
].eaflags
& EAF_FSGS
)) ||
756 (result
->oprs
[operand
].eaflags
& EAF_REL
));
758 result
->oprs
[operand
].type
|= is_rel
? IP_REL
: MEM_OFFS
;
760 result
->oprs
[operand
].basereg
= b
;
761 result
->oprs
[operand
].indexreg
= i
;
762 result
->oprs
[operand
].scale
= s
;
763 result
->oprs
[operand
].offset
= o
;
764 } else { /* it's not a memory reference */
766 if (is_just_unknown(value
)) { /* it's immediate but unknown */
767 result
->oprs
[operand
].type
|= IMMEDIATE
;
768 result
->oprs
[operand
].offset
= 0; /* don't care */
769 result
->oprs
[operand
].segment
= NO_SEG
; /* don't care again */
770 result
->oprs
[operand
].wrt
= NO_SEG
; /* still don't care */
771 } else if (is_reloc(value
)) { /* it's immediate */
772 result
->oprs
[operand
].type
|= IMMEDIATE
;
773 result
->oprs
[operand
].offset
= reloc_value(value
);
774 result
->oprs
[operand
].segment
= reloc_seg(value
);
775 result
->oprs
[operand
].wrt
= reloc_wrt(value
);
776 if (is_simple(value
)) {
777 if (reloc_value(value
) == 1)
778 result
->oprs
[operand
].type
|= UNITY
;
779 if (optimizing
>= 0 &&
780 !(result
->oprs
[operand
].type
& STRICT
)) {
781 if (reloc_value(value
) >= -128 &&
782 reloc_value(value
) <= 127)
783 result
->oprs
[operand
].type
|= SBYTE
;
786 } else { /* it's a register */
788 if (value
->type
>= EXPR_SIMPLE
|| value
->value
!= 1) {
789 error(ERR_NONFATAL
, "invalid operand type");
795 * check that its only 1 register, not an expression...
797 for (i
= 1; value
[i
].type
; i
++)
798 if (value
[i
].value
) {
799 error(ERR_NONFATAL
, "invalid operand type");
804 /* clear overrides, except TO which applies to FPU regs */
805 if (result
->oprs
[operand
].type
& ~TO
) {
807 * we want to produce a warning iff the specified size
808 * is different from the register size
810 i
= result
->oprs
[operand
].type
& SIZE_MASK
;
814 result
->oprs
[operand
].type
&= TO
;
815 result
->oprs
[operand
].type
|= REGISTER
;
816 result
->oprs
[operand
].type
|= reg_flags
[value
->type
];
817 result
->oprs
[operand
].basereg
= value
->type
;
819 if (i
&& (result
->oprs
[operand
].type
& SIZE_MASK
) != i
)
820 error(ERR_WARNING
| ERR_PASS1
,
821 "register size specification ignored");
826 result
->operands
= operand
; /* set operand count */
828 /* clear remaining operands */
829 while (operand
< MAX_OPERANDS
)
830 result
->oprs
[operand
++].type
= 0;
833 * Transform RESW, RESD, RESQ, REST, RESO into RESB.
835 switch (result
->opcode
) {
837 result
->opcode
= I_RESB
;
838 result
->oprs
[0].offset
*= 2;
841 result
->opcode
= I_RESB
;
842 result
->oprs
[0].offset
*= 4;
845 result
->opcode
= I_RESB
;
846 result
->oprs
[0].offset
*= 8;
849 result
->opcode
= I_RESB
;
850 result
->oprs
[0].offset
*= 10;
853 result
->opcode
= I_RESB
;
854 result
->oprs
[0].offset
*= 16;
863 static int is_comma_next(void)
870 i
= stdscan(NULL
, &tv
);
872 return (i
== ',' || i
== ';' || !i
);
875 void cleanup_insn(insn
* i
)
881 i
->eops
= i
->eops
->next
;