1 /* ----------------------------------------------------------------------- *
3 * Copyright 1996-2017 The NASM Authors - All Rights Reserved
4 * See the file AUTHORS included with the NASM distribution for
5 * the specific copyright holders.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following
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.
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.
32 * ----------------------------------------------------------------------- */
35 * assemble.c code generation for the Netwide Assembler
37 * Bytecode specification
38 * ----------------------
41 * Codes Mnemonic Explanation
43 * \0 terminates the code. (Unless it's a literal of course.)
44 * \1..\4 that many literal bytes follow in the code stream
45 * \5 add 4 to the primary operand number (b, low octdigit)
46 * \6 add 4 to the secondary operand number (a, middle octdigit)
47 * \7 add 4 to both the primary and the secondary operand number
48 * \10..\13 a literal byte follows in the code stream, to be added
49 * to the register value of operand 0..3
50 * \14..\17 the position of index register operand in MIB (BND insns)
51 * \20..\23 ib a byte immediate operand, from operand 0..3
52 * \24..\27 ib,u a zero-extended byte immediate operand, from operand 0..3
53 * \30..\33 iw a word immediate operand, from operand 0..3
54 * \34..\37 iwd select between \3[0-3] and \4[0-3] depending on 16/32 bit
55 * assembly mode or the operand-size override on the operand
56 * \40..\43 id a long immediate operand, from operand 0..3
57 * \44..\47 iwdq select between \3[0-3], \4[0-3] and \5[4-7]
58 * depending on the address size of the instruction.
59 * \50..\53 rel8 a byte relative operand, from operand 0..3
60 * \54..\57 iq a qword immediate operand, from operand 0..3
61 * \60..\63 rel16 a word relative operand, from operand 0..3
62 * \64..\67 rel select between \6[0-3] and \7[0-3] depending on 16/32 bit
63 * assembly mode or the operand-size override on the operand
64 * \70..\73 rel32 a long relative operand, from operand 0..3
65 * \74..\77 seg a word constant, from the _segment_ part of operand 0..3
66 * \1ab a ModRM, calculated on EA in operand a, with the spare
67 * field the register value of operand b.
68 * \172\ab the register number from operand a in bits 7..4, with
69 * the 4-bit immediate from operand b in bits 3..0.
70 * \173\xab the register number from operand a in bits 7..4, with
71 * the value b in bits 3..0.
72 * \174..\177 the register number from operand 0..3 in bits 7..4, and
73 * an arbitrary value in bits 3..0 (assembled as zero.)
74 * \2ab a ModRM, calculated on EA in operand a, with the spare
75 * field equal to digit b.
77 * \240..\243 this instruction uses EVEX rather than REX or VEX/XOP, with the
78 * V field taken from operand 0..3.
79 * \250 this instruction uses EVEX rather than REX or VEX/XOP, with the
80 * V field set to 1111b.
82 * EVEX prefixes are followed by the sequence:
83 * \cm\wlp\tup where cm is:
85 * c = 2 for EVEX and mmmm is the M field (EVEX.P0[3:0])
88 * [l0] ll = 0 (.128, .lz)
91 * [lig] ll = 3 for EVEX.L'L don't care (always assembled as 0)
93 * [w0] ww = 0 for W = 0
94 * [w1] ww = 1 for W = 1
95 * [wig] ww = 2 for W don't care (always assembled as 0)
96 * [ww] ww = 3 for W used as REX.W
98 * [p0] pp = 0 for no prefix
99 * [60] pp = 1 for legacy prefix 60
103 * tup is tuple type for Disp8*N from %tuple_codes in insns.pl
104 * (compressed displacement encoding)
106 * \254..\257 id,s a signed 32-bit operand to be extended to 64 bits.
107 * \260..\263 this instruction uses VEX/XOP rather than REX, with the
108 * V field taken from operand 0..3.
109 * \270 this instruction uses VEX/XOP rather than REX, with the
110 * V field set to 1111b.
112 * VEX/XOP prefixes are followed by the sequence:
113 * \tmm\wlp where mm is the M field; and wlp is:
115 * [l0] ll = 0 for L = 0 (.128, .lz)
116 * [l1] ll = 1 for L = 1 (.256)
117 * [lig] ll = 2 for L don't care (always assembled as 0)
119 * [w0] ww = 0 for W = 0
120 * [w1 ] ww = 1 for W = 1
121 * [wig] ww = 2 for W don't care (always assembled as 0)
122 * [ww] ww = 3 for W used as REX.W
124 * t = 0 for VEX (C4/C5), t = 1 for XOP (8F).
126 * \271 hlexr instruction takes XRELEASE (F3) with or without lock
127 * \272 hlenl instruction takes XACQUIRE/XRELEASE with or without lock
128 * \273 hle instruction takes XACQUIRE/XRELEASE with lock only
129 * \274..\277 ib,s a byte immediate operand, from operand 0..3, sign-extended
130 * to the operand size (if o16/o32/o64 present) or the bit size
131 * \310 a16 indicates fixed 16-bit address size, i.e. optional 0x67.
132 * \311 a32 indicates fixed 32-bit address size, i.e. optional 0x67.
133 * \312 adf (disassembler only) invalid with non-default address size.
134 * \313 a64 indicates fixed 64-bit address size, 0x67 invalid.
135 * \314 norexb (disassembler only) invalid with REX.B
136 * \315 norexx (disassembler only) invalid with REX.X
137 * \316 norexr (disassembler only) invalid with REX.R
138 * \317 norexw (disassembler only) invalid with REX.W
139 * \320 o16 indicates fixed 16-bit operand size, i.e. optional 0x66.
140 * \321 o32 indicates fixed 32-bit operand size, i.e. optional 0x66.
141 * \322 odf indicates that this instruction is only valid when the
142 * operand size is the default (instruction to disassembler,
143 * generates no code in the assembler)
144 * \323 o64nw indicates fixed 64-bit operand size, REX on extensions only.
145 * \324 o64 indicates 64-bit operand size requiring REX prefix.
146 * \325 nohi instruction which always uses spl/bpl/sil/dil
147 * \326 nof3 instruction not valid with 0xF3 REP prefix. Hint for
148 disassembler only; for SSE instructions.
149 * \330 a literal byte follows in the code stream, to be added
150 * to the condition code value of the instruction.
151 * \331 norep instruction not valid with REP prefix. Hint for
152 * disassembler only; for SSE instructions.
153 * \332 f2i REP prefix (0xF2 byte) used as opcode extension.
154 * \333 f3i REP prefix (0xF3 byte) used as opcode extension.
155 * \334 rex.l LOCK prefix used as REX.R (used in non-64-bit mode)
156 * \335 repe disassemble a rep (0xF3 byte) prefix as repe not rep.
157 * \336 mustrep force a REP(E) prefix (0xF3) even if not specified.
158 * \337 mustrepne force a REPNE prefix (0xF2) even if not specified.
159 * \336-\337 are still listed as prefixes in the disassembler.
160 * \340 resb reserve <operand 0> bytes of uninitialized storage.
161 * Operand 0 had better be a segmentless constant.
162 * \341 wait this instruction needs a WAIT "prefix"
163 * \360 np no SSE prefix (== \364\331)
164 * \361 66 SSE prefix (== \366\331)
165 * \364 !osp operand-size prefix (0x66) not permitted
166 * \365 !asp address-size prefix (0x67) not permitted
167 * \366 operand-size prefix (0x66) used as opcode extension
168 * \367 address-size prefix (0x67) used as opcode extension
169 * \370,\371 jcc8 match only if operand 0 meets byte jump criteria.
170 * jmp8 370 is used for Jcc, 371 is used for JMP.
171 * \373 jlen assemble 0x03 if bits==16, 0x05 if bits==32;
172 * used for conditional jump over longer jump
173 * \374 vsibx|vm32x|vm64x this instruction takes an XMM VSIB memory EA
174 * \375 vsiby|vm32y|vm64y this instruction takes an YMM VSIB memory EA
175 * \376 vsibz|vm32z|vm64z this instruction takes an ZMM VSIB memory EA
178 #include "compiler.h"
186 #include "assemble.h"
194 * Matching errors. These should be sorted so that more specific
195 * errors come later in the sequence.
208 * Matching success; the conditional ones first
210 MOK_JUMP
, /* Matching OK but needs jmp_match() */
211 MOK_GOOD
/* Matching unconditionally OK */
215 enum ea_type type
; /* what kind of EA is this? */
216 int sib_present
; /* is a SIB byte necessary? */
217 int bytes
; /* # of bytes of offset needed */
218 int size
; /* lazy - this is sib+bytes+1 */
219 uint8_t modrm
, sib
, rex
, rip
; /* the bytes themselves */
220 int8_t disp8
; /* compressed displacement for EVEX */
223 #define GEN_SIB(scale, index, base) \
224 (((scale) << 6) | ((index) << 3) | ((base)))
226 #define GEN_MODRM(mod, reg, rm) \
227 (((mod) << 6) | (((reg) & 7) << 3) | ((rm) & 7))
229 static iflag_t cpu
; /* cpu level received from nasm.c */
231 static int64_t calcsize(int32_t, int64_t, int, insn
*,
232 const struct itemplate
*);
233 static int emit_prefix(struct out_data
*data
, const int bits
, insn
*ins
);
234 static void gencode(struct out_data
*data
, insn
*ins
);
235 static enum match_result
find_match(const struct itemplate
**tempp
,
237 int32_t segment
, int64_t offset
, int bits
);
238 static enum match_result
matches(const struct itemplate
*, insn
*, int bits
);
239 static opflags_t
regflag(const operand
*);
240 static int32_t regval(const operand
*);
241 static int rexflags(int, opflags_t
, int);
242 static int op_rexflags(const operand
*, int);
243 static int op_evexflags(const operand
*, int, uint8_t);
244 static void add_asp(insn
*, int);
246 static enum ea_type
process_ea(operand
*, ea
*, int, int, opflags_t
, insn
*);
248 static inline bool absolute_op(const struct operand
*o
)
250 return o
->segment
== NO_SEG
&& o
->wrt
== NO_SEG
&&
251 !(o
->opflags
& OPFLAG_RELATIVE
);
254 static int has_prefix(insn
* ins
, enum prefix_pos pos
, int prefix
)
256 return ins
->prefixes
[pos
] == prefix
;
259 static void assert_no_prefix(insn
* ins
, enum prefix_pos pos
)
261 if (ins
->prefixes
[pos
])
262 nasm_error(ERR_NONFATAL
, "invalid %s prefix",
263 prefix_name(ins
->prefixes
[pos
]));
266 static const char *size_name(int size
)
290 static void warn_overflow(int pass
, int size
)
292 nasm_error(ERR_WARNING
| pass
| ERR_WARN_NOV
,
293 "%s data exceeds bounds", size_name(size
));
296 static void warn_overflow_const(int64_t data
, int size
)
298 if (overflow_general(data
, size
))
299 warn_overflow(ERR_PASS1
, size
);
302 static void warn_overflow_opd(const struct operand
*o
, int size
)
304 if (absolute_op(o
)) {
305 if (overflow_general(o
->offset
, size
))
306 warn_overflow(ERR_PASS2
, size
);
310 static void warn_overflow_out(int64_t data
, int size
, enum out_sign sign
)
316 err
= overflow_general(data
, size
);
319 err
= overflow_signed(data
, size
);
322 err
= overflow_unsigned(data
, size
);
330 warn_overflow(ERR_PASS2
, size
);
334 * This routine wrappers the real output format's output routine,
335 * in order to pass a copy of the data off to the listing file
336 * generator at the same time, flatten unnecessary relocations,
337 * and verify backend compatibility.
339 static void out(struct out_data
*data
)
341 static int32_t lineno
= 0; /* static!!! */
342 static const char *lnfname
= NULL
;
344 const int amax
= ofmt
->maxbits
>> 3; /* Maximum address size in bytes */
349 uint64_t size
= data
->size
;
351 int32_t fixseg
; /* Segment for which to produce fixed data */
354 return; /* Nothing to do */
357 * Convert addresses to RAWDATA if possible
358 * XXX: not all backends want this for global symbols!!!!
360 switch (data
->type
) {
362 addrval
= data
->toffset
;
363 fixseg
= NO_SEG
; /* Absolute address is fixed data */
367 addrval
= data
->toffset
- data
->relbase
;
368 fixseg
= data
->segment
; /* Our own segment is fixed data */
373 nasm_assert(asize
<= 8);
374 if (data
->tsegment
== fixseg
&& data
->twrt
== NO_SEG
) {
375 uint8_t *q
= xdata
.b
;
377 warn_overflow_out(addrval
, asize
, data
->sign
);
379 WRITEADDR(q
, addrval
, asize
);
380 data
->data
= xdata
.b
;
381 data
->type
= OUT_RAWDATA
;
382 asize
= 0; /* No longer an address */
387 asize
= 0; /* Not an address */
394 * this call to src_get determines when we call the
395 * debug-format-specific "linenum" function
396 * it updates lineno and lnfname to the current values
397 * returning 0 if "same as last time", -2 if lnfname
398 * changed, and the amount by which lineno changed,
399 * if it did. thus, these variables must be static
402 if (src_get(&lineno
, &lnfname
))
403 dfmt
->linenum(lnfname
, lineno
, data
->segment
);
405 if (asize
&& asize
> amax
) {
406 if (data
->type
!= OUT_ADDRESS
|| data
->sign
== OUT_SIGNED
) {
407 nasm_error(ERR_NONFATAL
,
408 "%d-bit signed relocation unsupported by output format %s\n",
409 asize
<< 3, ofmt
->shortname
);
411 nasm_error(ERR_WARNING
| ERR_WARN_ZEXTRELOC
,
412 "%d-bit unsigned relocation zero-extended from %d bits\n",
413 asize
<< 3, ofmt
->maxbits
);
416 data
->insoffs
+= amax
;
417 data
->offset
+= amax
;
418 data
->size
= size
= asize
- amax
;
420 data
->data
= zero_buffer
;
421 data
->type
= OUT_RAWDATA
;
425 data
->offset
+= size
;
426 data
->insoffs
+= size
;
429 static inline void out_rawdata(struct out_data
*data
, const void *rawdata
,
432 data
->type
= OUT_RAWDATA
;
433 data
->data
= rawdata
;
438 static void out_rawbyte(struct out_data
*data
, uint8_t byte
)
440 data
->type
= OUT_RAWDATA
;
446 static inline void out_reserve(struct out_data
*data
, uint64_t size
)
448 data
->type
= OUT_RESERVE
;
453 static inline void out_imm(struct out_data
*data
, const struct operand
*opx
,
454 int size
, enum out_sign sign
)
457 (opx
->opflags
& OPFLAG_RELATIVE
) ? OUT_RELADDR
: OUT_ADDRESS
;
460 data
->toffset
= opx
->offset
;
461 data
->tsegment
= opx
->segment
;
462 data
->twrt
= opx
->wrt
;
464 * XXX: improve this if at some point in the future we can
465 * distinguish the subtrahend in expressions like [foo - bar]
466 * where bar is a symbol in the current segment. However, at the
467 * current point, if OPFLAG_RELATIVE is set that subtraction has
474 static void out_reladdr(struct out_data
*data
, const struct operand
*opx
,
477 if (opx
->opflags
& OPFLAG_RELATIVE
)
478 nasm_error(ERR_NONFATAL
, "invalid use of self-relative expression");
480 data
->type
= OUT_RELADDR
;
481 data
->sign
= OUT_SIGNED
;
483 data
->toffset
= opx
->offset
;
484 data
->tsegment
= opx
->segment
;
485 data
->twrt
= opx
->wrt
;
486 data
->relbase
= data
->offset
+ (data
->inslen
- data
->insoffs
);
490 static inline void out_segment(struct out_data
*data
,
491 const struct operand
*opx
)
493 data
->type
= OUT_SEGMENT
;
494 data
->sign
= OUT_UNSIGNED
;
496 data
->toffset
= opx
->offset
;
497 data
->tsegment
= ofmt
->segbase(opx
->segment
+ 1);
498 data
->twrt
= opx
->wrt
;
502 static bool jmp_match(int32_t segment
, int64_t offset
, int bits
,
503 insn
* ins
, const struct itemplate
*temp
)
506 const uint8_t *code
= temp
->code
;
510 if (((c
& ~1) != 0370) || (ins
->oprs
[0].type
& STRICT
))
514 if (optimizing
< 0 && c
== 0371)
517 isize
= calcsize(segment
, offset
, bits
, ins
, temp
);
519 if (ins
->oprs
[0].opflags
& OPFLAG_UNKNOWN
)
520 /* Be optimistic in pass 1 */
523 if (ins
->oprs
[0].segment
!= segment
)
526 isize
= ins
->oprs
[0].offset
- offset
- isize
; /* isize is delta */
527 is_byte
= (isize
>= -128 && isize
<= 127); /* is it byte size? */
529 if (is_byte
&& c
== 0371 && ins
->prefixes
[PPS_REP
] == P_BND
) {
530 /* jmp short (opcode eb) cannot be used with bnd prefix. */
531 ins
->prefixes
[PPS_REP
] = P_none
;
532 nasm_error(ERR_WARNING
| ERR_WARN_BND
| ERR_PASS2
,
533 "jmp short does not init bnd regs - bnd prefix dropped.");
539 /* This is totally just a wild guess what is reasonable... */
540 #define INCBIN_MAX_BUF (ZERO_BUF_SIZE * 16)
542 int64_t assemble(int32_t segment
, int64_t start
, int bits
, iflag_t cp
,
545 struct out_data data
;
546 const struct itemplate
*temp
;
549 int64_t wsize
; /* size for DB etc. */
555 data
.segment
= segment
;
557 data
.sign
= OUT_WRAP
;
560 wsize
= idata_bytes(instruction
->opcode
);
566 int32_t t
= instruction
->times
;
568 nasm_panic(0, "instruction->times < 0 (%"PRId32
") in assemble()", t
);
570 while (t
--) { /* repeat TIMES times */
571 list_for_each(e
, instruction
->eops
) {
572 if (e
->type
== EOT_DB_NUMBER
) {
574 nasm_error(ERR_NONFATAL
,
575 "integer supplied to a DT, DO or DY"
579 data
.type
= e
->relative
? OUT_RELADDR
: OUT_ADDRESS
;
580 data
.inslen
= data
.size
= wsize
;
581 data
.toffset
= e
->offset
;
582 data
.tsegment
= e
->segment
;
587 } else if (e
->type
== EOT_DB_STRING
||
588 e
->type
== EOT_DB_STRING_FREE
) {
589 int align
= e
->stringlen
% wsize
;
591 align
= wsize
- align
;
594 data
.inslen
= e
->stringlen
+ align
;
596 out_rawdata(&data
, e
->stringval
, e
->stringlen
);
597 out_rawdata(&data
, zero_buffer
, align
);
600 if (t
> 0 && t
== instruction
->times
- 1) {
601 lfmt
->set_offset(data
.offset
);
602 lfmt
->uplevel(LIST_TIMES
);
605 if (instruction
->times
> 1)
606 lfmt
->downlevel(LIST_TIMES
);
607 } else if (instruction
->opcode
== I_INCBIN
) {
608 const char *fname
= instruction
->eops
->stringval
;
610 size_t t
= instruction
->times
;
613 const void *map
= NULL
;
615 size_t blk
= 0; /* Buffered I/O block size */
616 size_t m
= 0; /* Bytes last read */
618 fp
= nasm_open_read(fname
, NF_BINARY
|NF_FORMAP
);
620 nasm_error(ERR_NONFATAL
, "`incbin': unable to open file `%s'",
625 len
= nasm_file_size(fp
);
627 if (len
== (off_t
)-1) {
628 nasm_error(ERR_NONFATAL
, "`incbin': unable to get length of file `%s'",
633 if (instruction
->eops
->next
) {
634 base
= instruction
->eops
->next
->offset
;
639 if (instruction
->eops
->next
->next
&&
640 len
> (off_t
)instruction
->eops
->next
->next
->offset
)
641 len
= (off_t
)instruction
->eops
->next
->next
->offset
;
645 lfmt
->set_offset(data
.offset
);
646 lfmt
->uplevel(LIST_INCBIN
);
651 /* Try to map file data */
652 map
= nasm_map_file(fp
, base
, len
);
654 blk
= len
< (off_t
)INCBIN_MAX_BUF
? (size_t)len
: INCBIN_MAX_BUF
;
655 buf
= nasm_malloc(blk
);
660 * Consider these irrelevant for INCBIN, since it is fully
661 * possible that these might be (way) bigger than an int
662 * can hold; there is, however, no reason to widen these
663 * types just for INCBIN. data.inslen == 0 signals to the
664 * backend that these fields are meaningless, if at all
671 out_rawdata(&data
, map
, len
);
672 } else if ((off_t
)m
== len
) {
673 out_rawdata(&data
, buf
, len
);
677 if (fseeko(fp
, base
, SEEK_SET
) < 0 || ferror(fp
)) {
678 nasm_error(ERR_NONFATAL
,
679 "`incbin': unable to seek on file `%s'",
684 m
= fread(buf
, 1, l
< (off_t
)blk
? (size_t)l
: blk
, fp
);
685 if (!m
|| feof(fp
)) {
687 * This shouldn't happen unless the file
688 * actually changes while we are reading
691 nasm_error(ERR_NONFATAL
,
692 "`incbin': unexpected EOF while"
693 " reading file `%s'", fname
);
696 out_rawdata(&data
, buf
, m
);
702 lfmt
->downlevel(LIST_INCBIN
);
703 if (instruction
->times
> 1) {
704 lfmt
->set_offset(data
.offset
);
705 lfmt
->uplevel(LIST_TIMES
);
706 lfmt
->downlevel(LIST_TIMES
);
709 nasm_error(ERR_NONFATAL
,
710 "`incbin': error while"
711 " reading file `%s'", fname
);
717 nasm_unmap_file(map
, len
);
722 /* "Real" instruction */
724 /* Check to see if we need an address-size prefix */
725 add_asp(instruction
, bits
);
727 m
= find_match(&temp
, instruction
, data
.segment
, data
.offset
, bits
);
731 int64_t insn_size
= calcsize(data
.segment
, data
.offset
,
732 bits
, instruction
, temp
);
733 itimes
= instruction
->times
;
734 if (insn_size
< 0) /* shouldn't be, on pass two */
735 nasm_panic(0, "errors made it through from pass one");
742 data
.inslen
= insn_size
;
744 gencode(&data
, instruction
);
745 nasm_assert(data
.insoffs
== insn_size
);
747 if (itimes
> 0 && itimes
== instruction
->times
- 1) {
748 lfmt
->set_offset(data
.offset
);
749 lfmt
->uplevel(LIST_TIMES
);
752 if (instruction
->times
> 1)
753 lfmt
->downlevel(LIST_TIMES
);
757 case MERR_OPSIZEMISSING
:
758 nasm_error(ERR_NONFATAL
, "operation size not specified");
760 case MERR_OPSIZEMISMATCH
:
761 nasm_error(ERR_NONFATAL
, "mismatch in operand sizes");
763 case MERR_BRNUMMISMATCH
:
764 nasm_error(ERR_NONFATAL
,
765 "mismatch in the number of broadcasting elements");
768 nasm_error(ERR_NONFATAL
, "no instruction for this cpu level");
771 nasm_error(ERR_NONFATAL
, "instruction not supported in %d-bit mode",
774 case MERR_ENCMISMATCH
:
775 nasm_error(ERR_NONFATAL
, "specific encoding scheme not available");
778 nasm_error(ERR_NONFATAL
, "bnd prefix is not allowed");
781 nasm_error(ERR_NONFATAL
, "%s prefix is not allowed",
782 (has_prefix(instruction
, PPS_REP
, P_REPNE
) ?
786 nasm_error(ERR_NONFATAL
,
787 "invalid combination of opcode and operands");
792 return data
.offset
- start
;
795 int64_t insn_size(int32_t segment
, int64_t offset
, int bits
, iflag_t cp
,
798 const struct itemplate
*temp
;
803 if (instruction
->opcode
== I_none
)
806 if (instruction
->opcode
== I_DB
|| instruction
->opcode
== I_DW
||
807 instruction
->opcode
== I_DD
|| instruction
->opcode
== I_DQ
||
808 instruction
->opcode
== I_DT
|| instruction
->opcode
== I_DO
||
809 instruction
->opcode
== I_DY
) {
811 int32_t isize
, osize
, wsize
;
814 wsize
= idata_bytes(instruction
->opcode
);
816 list_for_each(e
, instruction
->eops
) {
820 if (e
->type
== EOT_DB_NUMBER
) {
822 warn_overflow_const(e
->offset
, wsize
);
823 } else if (e
->type
== EOT_DB_STRING
||
824 e
->type
== EOT_DB_STRING_FREE
)
825 osize
= e
->stringlen
;
827 align
= (-osize
) % wsize
;
830 isize
+= osize
+ align
;
835 if (instruction
->opcode
== I_INCBIN
) {
836 const char *fname
= instruction
->eops
->stringval
;
839 len
= nasm_file_size_by_path(fname
);
840 if (len
== (off_t
)-1) {
841 nasm_error(ERR_NONFATAL
, "`incbin': unable to get length of file `%s'",
846 if (instruction
->eops
->next
) {
847 if (len
<= (off_t
)instruction
->eops
->next
->offset
) {
850 len
-= instruction
->eops
->next
->offset
;
851 if (instruction
->eops
->next
->next
&&
852 len
> (off_t
)instruction
->eops
->next
->next
->offset
) {
853 len
= (off_t
)instruction
->eops
->next
->next
->offset
;
861 /* Check to see if we need an address-size prefix */
862 add_asp(instruction
, bits
);
864 m
= find_match(&temp
, instruction
, segment
, offset
, bits
);
866 /* we've matched an instruction. */
867 return calcsize(segment
, offset
, bits
, instruction
, temp
);
869 return -1; /* didn't match any instruction */
873 static void bad_hle_warn(const insn
* ins
, uint8_t hleok
)
875 enum prefixes rep_pfx
= ins
->prefixes
[PPS_REP
];
876 enum whatwarn
{ w_none
, w_lock
, w_inval
} ww
;
877 static const enum whatwarn warn
[2][4] =
879 { w_inval
, w_inval
, w_none
, w_lock
}, /* XACQUIRE */
880 { w_inval
, w_none
, w_none
, w_lock
}, /* XRELEASE */
884 n
= (unsigned int)rep_pfx
- P_XACQUIRE
;
886 return; /* Not XACQUIRE/XRELEASE */
889 if (!is_class(MEMORY
, ins
->oprs
[0].type
))
890 ww
= w_inval
; /* HLE requires operand 0 to be memory */
897 if (ins
->prefixes
[PPS_LOCK
] != P_LOCK
) {
898 nasm_error(ERR_WARNING
| ERR_WARN_HLE
| ERR_PASS2
,
899 "%s with this instruction requires lock",
900 prefix_name(rep_pfx
));
905 nasm_error(ERR_WARNING
| ERR_WARN_HLE
| ERR_PASS2
,
906 "%s invalid with this instruction",
907 prefix_name(rep_pfx
));
912 /* Common construct */
913 #define case3(x) case (x): case (x)+1: case (x)+2
914 #define case4(x) case3(x): case (x)+3
916 static int64_t calcsize(int32_t segment
, int64_t offset
, int bits
,
917 insn
* ins
, const struct itemplate
*temp
)
919 const uint8_t *codes
= temp
->code
;
928 bool lockcheck
= true;
929 enum reg_enum mib_index
= R_none
; /* For a separate index MIB reg form */
931 ins
->rex
= 0; /* Ensure REX is reset */
932 eat
= EA_SCALAR
; /* Expect a scalar EA */
933 memset(ins
->evex_p
, 0, 3); /* Ensure EVEX is reset */
935 if (ins
->prefixes
[PPS_OSIZE
] == P_O64
)
938 (void)segment
; /* Don't warn that this parameter is unused */
939 (void)offset
; /* Don't warn that this parameter is unused */
943 op1
= (c
& 3) + ((opex
& 1) << 2);
944 op2
= ((c
>> 3) & 3) + ((opex
& 2) << 1);
945 opx
= &ins
->oprs
[op1
];
946 opex
= 0; /* For the next iteration */
950 codes
+= c
, length
+= c
;
959 op_rexflags(opx
, REX_B
|REX_H
|REX_P
|REX_W
);
964 /* this is an index reg of MIB operand */
965 mib_index
= opx
->basereg
;
978 if (opx
->type
& (BITS16
| BITS32
| BITS64
))
979 length
+= (opx
->type
& BITS16
) ? 2 : 4;
981 length
+= (bits
== 16) ? 2 : 4;
989 length
+= ins
->addr_size
>> 3;
997 length
+= 8; /* MOV reg64/imm */
1005 if (opx
->type
& (BITS16
| BITS32
| BITS64
))
1006 length
+= (opx
->type
& BITS16
) ? 2 : 4;
1008 length
+= (bits
== 16) ? 2 : 4;
1031 ins
->vexreg
= regval(opx
);
1032 ins
->evex_p
[2] |= op_evexflags(opx
, EVEX_P2VP
, 2); /* High-16 NDS */
1033 ins
->vex_cm
= *codes
++;
1034 ins
->vex_wlp
= *codes
++;
1035 ins
->evex_tuple
= (*codes
++ - 0300);
1041 ins
->vex_cm
= *codes
++;
1042 ins
->vex_wlp
= *codes
++;
1043 ins
->evex_tuple
= (*codes
++ - 0300);
1052 ins
->vexreg
= regval(opx
);
1053 ins
->vex_cm
= *codes
++;
1054 ins
->vex_wlp
= *codes
++;
1060 ins
->vex_cm
= *codes
++;
1061 ins
->vex_wlp
= *codes
++;
1078 length
+= (bits
!= 16) && !has_prefix(ins
, PPS_ASIZE
, P_A16
);
1082 length
+= (bits
!= 32) && !has_prefix(ins
, PPS_ASIZE
, P_A32
);
1089 if (bits
!= 64 || has_prefix(ins
, PPS_ASIZE
, P_A16
) ||
1090 has_prefix(ins
, PPS_ASIZE
, P_A32
))
1099 enum prefixes pfx
= ins
->prefixes
[PPS_OSIZE
];
1103 nasm_error(ERR_WARNING
| ERR_PASS2
, "invalid operand size prefix");
1105 ins
->prefixes
[PPS_OSIZE
] = P_O16
;
1111 enum prefixes pfx
= ins
->prefixes
[PPS_OSIZE
];
1115 nasm_error(ERR_WARNING
| ERR_PASS2
, "invalid operand size prefix");
1117 ins
->prefixes
[PPS_OSIZE
] = P_O32
;
1159 if (!ins
->prefixes
[PPS_REP
])
1160 ins
->prefixes
[PPS_REP
] = P_REP
;
1164 if (!ins
->prefixes
[PPS_REP
])
1165 ins
->prefixes
[PPS_REP
] = P_REPNE
;
1169 if (!absolute_op(&ins
->oprs
[0]))
1170 nasm_error(ERR_NONFATAL
, "attempt to reserve non-constant"
1171 " quantity of BSS space");
1172 else if (ins
->oprs
[0].opflags
& OPFLAG_FORWARD
)
1173 nasm_error(ERR_WARNING
| ERR_PASS1
,
1174 "forward reference in RESx can have unpredictable results");
1176 length
+= ins
->oprs
[0].offset
;
1180 if (!ins
->prefixes
[PPS_WAIT
])
1181 ins
->prefixes
[PPS_WAIT
] = P_WAIT
;
1236 struct operand
*opy
= &ins
->oprs
[op2
];
1237 struct operand
*op_er_sae
;
1239 ea_data
.rex
= 0; /* Ensure ea.REX is initially 0 */
1242 /* pick rfield from operand b (opx) */
1243 rflags
= regflag(opx
);
1244 rfield
= nasm_regvals
[opx
->basereg
];
1250 /* EVEX.b1 : evex_brerop contains the operand position */
1251 op_er_sae
= (ins
->evex_brerop
>= 0 ?
1252 &ins
->oprs
[ins
->evex_brerop
] : NULL
);
1254 if (op_er_sae
&& (op_er_sae
->decoflags
& (ER
| SAE
))) {
1256 ins
->evex_p
[2] |= EVEX_P2B
;
1257 if (op_er_sae
->decoflags
& ER
) {
1258 /* set EVEX.RC (rounding control) */
1259 ins
->evex_p
[2] |= ((ins
->evex_rm
- BRC_RN
) << 5)
1263 /* set EVEX.L'L (vector length) */
1264 ins
->evex_p
[2] |= ((ins
->vex_wlp
<< (5 - 2)) & EVEX_P2LL
);
1265 ins
->evex_p
[1] |= ((ins
->vex_wlp
<< (7 - 4)) & EVEX_P1W
);
1266 if (opy
->decoflags
& BRDCAST_MASK
) {
1268 ins
->evex_p
[2] |= EVEX_P2B
;
1272 if (itemp_has(temp
, IF_MIB
)) {
1273 opy
->eaflags
|= EAF_MIB
;
1275 * if a separate form of MIB (ICC style) is used,
1276 * the index reg info is merged into mem operand
1278 if (mib_index
!= R_none
) {
1279 opy
->indexreg
= mib_index
;
1281 opy
->hintbase
= mib_index
;
1282 opy
->hinttype
= EAH_NOTBASE
;
1286 if (process_ea(opy
, &ea_data
, bits
,
1287 rfield
, rflags
, ins
) != eat
) {
1288 nasm_error(ERR_NONFATAL
, "invalid effective address");
1291 ins
->rex
|= ea_data
.rex
;
1292 length
+= ea_data
.size
;
1298 nasm_panic(0, "internal instruction table corrupt"
1299 ": instruction code \\%o (0x%02X) given", c
, c
);
1304 ins
->rex
&= rex_mask
;
1306 if (ins
->rex
& REX_NH
) {
1307 if (ins
->rex
& REX_H
) {
1308 nasm_error(ERR_NONFATAL
, "instruction cannot use high registers");
1311 ins
->rex
&= ~REX_P
; /* Don't force REX prefix due to high reg */
1314 switch (ins
->prefixes
[PPS_VEX
]) {
1316 if (!(ins
->rex
& REX_EV
))
1321 if (!(ins
->rex
& REX_V
))
1328 if (ins
->rex
& (REX_V
| REX_EV
)) {
1329 int bad32
= REX_R
|REX_W
|REX_X
|REX_B
;
1331 if (ins
->rex
& REX_H
) {
1332 nasm_error(ERR_NONFATAL
, "cannot use high register in AVX instruction");
1335 switch (ins
->vex_wlp
& 060) {
1349 if (bits
!= 64 && ((ins
->rex
& bad32
) || ins
->vexreg
> 7)) {
1350 nasm_error(ERR_NONFATAL
, "invalid operands in non-64-bit mode");
1352 } else if (!(ins
->rex
& REX_EV
) &&
1353 ((ins
->vexreg
> 15) || (ins
->evex_p
[0] & 0xf0))) {
1354 nasm_error(ERR_NONFATAL
, "invalid high-16 register in non-AVX-512");
1357 if (ins
->rex
& REX_EV
)
1359 else if (ins
->vex_cm
!= 1 || (ins
->rex
& (REX_W
|REX_X
|REX_B
)) ||
1360 ins
->prefixes
[PPS_VEX
] == P_VEX3
)
1364 } else if (ins
->rex
& REX_MASK
) {
1365 if (ins
->rex
& REX_H
) {
1366 nasm_error(ERR_NONFATAL
, "cannot use high register in rex instruction");
1368 } else if (bits
== 64) {
1370 } else if ((ins
->rex
& REX_L
) &&
1371 !(ins
->rex
& (REX_P
|REX_W
|REX_X
|REX_B
)) &&
1372 iflag_ffs(&cpu
) >= IF_X86_64
) {
1374 assert_no_prefix(ins
, PPS_LOCK
);
1375 lockcheck
= false; /* Already errored, no need for warning */
1378 nasm_error(ERR_NONFATAL
, "invalid operands in non-64-bit mode");
1383 if (has_prefix(ins
, PPS_LOCK
, P_LOCK
) && lockcheck
&&
1384 (!itemp_has(temp
,IF_LOCK
) || !is_class(MEMORY
, ins
->oprs
[0].type
))) {
1385 nasm_error(ERR_WARNING
| ERR_WARN_LOCK
| ERR_PASS2
,
1386 "instruction is not lockable");
1389 bad_hle_warn(ins
, hleok
);
1392 * when BND prefix is set by DEFAULT directive,
1393 * BND prefix is added to every appropriate instruction line
1394 * unless it is overridden by NOBND prefix.
1397 (itemp_has(temp
, IF_BND
) && !has_prefix(ins
, PPS_REP
, P_NOBND
)))
1398 ins
->prefixes
[PPS_REP
] = P_BND
;
1401 * Add length of legacy prefixes
1403 length
+= emit_prefix(NULL
, bits
, ins
);
1408 static inline void emit_rex(struct out_data
*data
, insn
*ins
)
1410 if (data
->bits
== 64) {
1411 if ((ins
->rex
& REX_MASK
) &&
1412 !(ins
->rex
& (REX_V
| REX_EV
)) &&
1414 uint8_t rex
= (ins
->rex
& REX_MASK
) | REX_P
;
1415 out_rawbyte(data
, rex
);
1416 ins
->rex_done
= true;
1421 static int emit_prefix(struct out_data
*data
, const int bits
, insn
*ins
)
1426 for (j
= 0; j
< MAXPREFIX
; j
++) {
1428 switch (ins
->prefixes
[j
]) {
1449 nasm_error(ERR_WARNING
| ERR_PASS2
,
1450 "cs segment base generated, but will be ignored in 64-bit mode");
1456 nasm_error(ERR_WARNING
| ERR_PASS2
,
1457 "ds segment base generated, but will be ignored in 64-bit mode");
1463 nasm_error(ERR_WARNING
| ERR_PASS2
,
1464 "es segment base generated, but will be ignored in 64-bit mode");
1476 nasm_error(ERR_WARNING
| ERR_PASS2
,
1477 "ss segment base generated, but will be ignored in 64-bit mode");
1483 nasm_error(ERR_NONFATAL
,
1484 "segr6 and segr7 cannot be used as prefixes");
1488 nasm_error(ERR_NONFATAL
,
1489 "16-bit addressing is not supported "
1491 } else if (bits
!= 16)
1500 nasm_error(ERR_NONFATAL
,
1501 "64-bit addressing is only supported "
1529 nasm_panic(0, "invalid instruction prefix");
1533 out_rawbyte(data
, c
);
1540 static void gencode(struct out_data
*data
, insn
*ins
)
1546 struct operand
*opx
;
1547 const uint8_t *codes
= data
->itemp
->code
;
1549 enum ea_type eat
= EA_SCALAR
;
1551 const int bits
= data
->bits
;
1553 ins
->rex_done
= false;
1555 emit_prefix(data
, bits
, ins
);
1559 op1
= (c
& 3) + ((opex
& 1) << 2);
1560 op2
= ((c
>> 3) & 3) + ((opex
& 2) << 1);
1561 opx
= &ins
->oprs
[op1
];
1562 opex
= 0; /* For the next iteration */
1570 emit_rex(data
, ins
);
1571 out_rawdata(data
, codes
, c
);
1582 emit_rex(data
, ins
);
1583 out_rawbyte(data
, *codes
++ + (regval(opx
) & 7));
1590 if (opx
->offset
< -256 || opx
->offset
> 255)
1591 nasm_error(ERR_WARNING
| ERR_PASS2
| ERR_WARN_NOV
,
1592 "byte value exceeds bounds");
1593 out_imm(data
, opx
, 1, OUT_WRAP
);
1597 if (opx
->offset
< 0 || opx
->offset
> 255)
1598 nasm_error(ERR_WARNING
| ERR_PASS2
| ERR_WARN_NOV
,
1599 "unsigned byte value exceeds bounds");
1600 out_imm(data
, opx
, 1, OUT_UNSIGNED
);
1604 warn_overflow_opd(opx
, 2);
1605 out_imm(data
, opx
, 2, OUT_WRAP
);
1609 if (opx
->type
& (BITS16
| BITS32
))
1610 size
= (opx
->type
& BITS16
) ? 2 : 4;
1612 size
= (bits
== 16) ? 2 : 4;
1613 warn_overflow_opd(opx
, size
);
1614 out_imm(data
, opx
, size
, OUT_WRAP
);
1618 warn_overflow_opd(opx
, 4);
1619 out_imm(data
, opx
, 4, OUT_WRAP
);
1623 size
= ins
->addr_size
>> 3;
1624 warn_overflow_opd(opx
, size
);
1625 out_imm(data
, opx
, size
, OUT_WRAP
);
1629 if (opx
->segment
== data
->segment
) {
1630 int64_t delta
= opx
->offset
- data
->offset
1631 - (data
->inslen
- data
->insoffs
);
1632 if (delta
> 127 || delta
< -128)
1633 nasm_error(ERR_NONFATAL
, "short jump is out of range");
1635 out_reladdr(data
, opx
, 1);
1639 out_imm(data
, opx
, 8, OUT_WRAP
);
1643 out_reladdr(data
, opx
, 2);
1647 if (opx
->type
& (BITS16
| BITS32
| BITS64
))
1648 size
= (opx
->type
& BITS16
) ? 2 : 4;
1650 size
= (bits
== 16) ? 2 : 4;
1652 out_reladdr(data
, opx
, size
);
1656 out_reladdr(data
, opx
, 4);
1660 if (opx
->segment
== NO_SEG
)
1661 nasm_error(ERR_NONFATAL
, "value referenced by FAR is not"
1663 out_segment(data
, opx
);
1668 int mask
= ins
->prefixes
[PPS_VEX
] == P_EVEX
? 7 : 15;
1669 const struct operand
*opy
;
1672 opx
= &ins
->oprs
[c
>> 3];
1673 opy
= &ins
->oprs
[c
& 7];
1674 if (!absolute_op(opy
)) {
1675 nasm_error(ERR_NONFATAL
,
1676 "non-absolute expression not permitted as argument %d",
1678 } else if (opy
->offset
& ~mask
) {
1679 nasm_error(ERR_WARNING
| ERR_PASS2
| ERR_WARN_NOV
,
1680 "is4 argument exceeds bounds");
1682 c
= opy
->offset
& mask
;
1688 opx
= &ins
->oprs
[c
>> 4];
1695 r
= nasm_regvals
[opx
->basereg
];
1696 out_rawbyte(data
, (r
<< 4) | ((r
& 0x10) >> 1) | c
);
1700 if (absolute_op(opx
) &&
1701 (int32_t)opx
->offset
!= (int64_t)opx
->offset
) {
1702 nasm_error(ERR_WARNING
| ERR_PASS2
| ERR_WARN_NOV
,
1703 "signed dword immediate exceeds bounds");
1705 out_imm(data
, opx
, 4, OUT_SIGNED
);
1711 ins
->evex_p
[2] |= op_evexflags(&ins
->oprs
[0],
1712 EVEX_P2Z
| EVEX_P2AAA
, 2);
1713 ins
->evex_p
[2] ^= EVEX_P2VP
; /* 1's complement */
1715 /* EVEX.X can be set by either REX or EVEX for different reasons */
1716 bytes
[1] = ((((ins
->rex
& 7) << 5) |
1717 (ins
->evex_p
[0] & (EVEX_P0X
| EVEX_P0RP
))) ^ 0xf0) |
1718 (ins
->vex_cm
& EVEX_P0MM
);
1719 bytes
[2] = ((ins
->rex
& REX_W
) << (7 - 3)) |
1720 ((~ins
->vexreg
& 15) << 3) |
1721 (1 << 2) | (ins
->vex_wlp
& 3);
1722 bytes
[3] = ins
->evex_p
[2];
1723 out_rawdata(data
, bytes
, 4);
1729 if (ins
->vex_cm
!= 1 || (ins
->rex
& (REX_W
|REX_X
|REX_B
)) ||
1730 ins
->prefixes
[PPS_VEX
] == P_VEX3
) {
1731 bytes
[0] = (ins
->vex_cm
>> 6) ? 0x8f : 0xc4;
1732 bytes
[1] = (ins
->vex_cm
& 31) | ((~ins
->rex
& 7) << 5);
1733 bytes
[2] = ((ins
->rex
& REX_W
) << (7-3)) |
1734 ((~ins
->vexreg
& 15)<< 3) | (ins
->vex_wlp
& 07);
1735 out_rawdata(data
, bytes
, 3);
1738 bytes
[1] = ((~ins
->rex
& REX_R
) << (7-2)) |
1739 ((~ins
->vexreg
& 15) << 3) | (ins
->vex_wlp
& 07);
1740 out_rawdata(data
, bytes
, 2);
1754 if (ins
->rex
& REX_W
)
1756 else if (ins
->prefixes
[PPS_OSIZE
] == P_O16
)
1758 else if (ins
->prefixes
[PPS_OSIZE
] == P_O32
)
1763 um
= (uint64_t)2 << (s
-1);
1766 if (uv
> 127 && uv
< (uint64_t)-128 &&
1767 (uv
< um
-128 || uv
> um
-1)) {
1768 /* If this wasn't explicitly byte-sized, warn as though we
1769 * had fallen through to the imm16/32/64 case.
1771 nasm_error(ERR_WARNING
| ERR_PASS2
| ERR_WARN_NOV
,
1772 "%s value exceeds bounds",
1773 (opx
->type
& BITS8
) ? "signed byte" :
1778 out_imm(data
, opx
, 1, OUT_WRAP
); /* XXX: OUT_SIGNED? */
1786 if (bits
== 32 && !has_prefix(ins
, PPS_ASIZE
, P_A16
))
1787 out_rawbyte(data
, 0x67);
1791 if (bits
!= 32 && !has_prefix(ins
, PPS_ASIZE
, P_A32
))
1792 out_rawbyte(data
, 0x67);
1824 out_rawbyte(data
, *codes
++ ^ get_cond_opcode(ins
->condition
));
1832 out_rawbyte(data
, c
- 0332 + 0xF2);
1836 if (ins
->rex
& REX_R
)
1837 out_rawbyte(data
, 0xF0);
1838 ins
->rex
&= ~(REX_L
|REX_R
);
1849 if (ins
->oprs
[0].segment
!= NO_SEG
)
1850 nasm_panic(0, "non-constant BSS size in pass two");
1852 out_reserve(data
, ins
->oprs
[0].offset
);
1862 out_rawbyte(data
, 0x66);
1871 out_rawbyte(data
, c
- 0366 + 0x66);
1878 out_rawbyte(data
, bits
== 16 ? 3 : 5);
1910 struct operand
*opy
= &ins
->oprs
[op2
];
1913 /* pick rfield from operand b (opx) */
1914 rflags
= regflag(opx
);
1915 rfield
= nasm_regvals
[opx
->basereg
];
1917 /* rfield is constant */
1922 if (process_ea(opy
, &ea_data
, bits
,
1923 rfield
, rflags
, ins
) != eat
)
1924 nasm_error(ERR_NONFATAL
, "invalid effective address");
1927 *p
++ = ea_data
.modrm
;
1928 if (ea_data
.sib_present
)
1930 out_rawdata(data
, bytes
, p
- bytes
);
1933 * Make sure the address gets the right offset in case
1934 * the line breaks in the .lst file (BR 1197827)
1937 if (ea_data
.bytes
) {
1938 /* use compressed displacement, if available */
1939 if (ea_data
.disp8
) {
1940 out_rawbyte(data
, ea_data
.disp8
);
1941 } else if (ea_data
.rip
) {
1942 out_reladdr(data
, opy
, ea_data
.bytes
);
1944 int asize
= ins
->addr_size
>> 3;
1946 if (overflow_general(opy
->offset
, asize
) ||
1947 signed_bits(opy
->offset
, ins
->addr_size
) !=
1948 signed_bits(opy
->offset
, ea_data
.bytes
<< 3))
1949 warn_overflow(ERR_PASS2
, ea_data
.bytes
);
1951 out_imm(data
, opy
, ea_data
.bytes
,
1952 (asize
> ea_data
.bytes
) ? OUT_SIGNED
: OUT_UNSIGNED
);
1959 nasm_panic(0, "internal instruction table corrupt"
1960 ": instruction code \\%o (0x%02X) given", c
, c
);
1966 static opflags_t
regflag(const operand
* o
)
1968 if (!is_register(o
->basereg
))
1969 nasm_panic(0, "invalid operand passed to regflag()");
1970 return nasm_reg_flags
[o
->basereg
];
1973 static int32_t regval(const operand
* o
)
1975 if (!is_register(o
->basereg
))
1976 nasm_panic(0, "invalid operand passed to regval()");
1977 return nasm_regvals
[o
->basereg
];
1980 static int op_rexflags(const operand
* o
, int mask
)
1985 if (!is_register(o
->basereg
))
1986 nasm_panic(0, "invalid operand passed to op_rexflags()");
1988 flags
= nasm_reg_flags
[o
->basereg
];
1989 val
= nasm_regvals
[o
->basereg
];
1991 return rexflags(val
, flags
, mask
);
1994 static int rexflags(int val
, opflags_t flags
, int mask
)
1998 if (val
>= 0 && (val
& 8))
1999 rex
|= REX_B
|REX_X
|REX_R
;
2002 if (!(REG_HIGH
& ~flags
)) /* AH, CH, DH, BH */
2004 else if (!(REG8
& ~flags
) && val
>= 4) /* SPL, BPL, SIL, DIL */
2010 static int evexflags(int val
, decoflags_t deco
,
2011 int mask
, uint8_t byte
)
2017 if (val
>= 0 && (val
& 16))
2018 evex
|= (EVEX_P0RP
| EVEX_P0X
);
2021 if (val
>= 0 && (val
& 16))
2025 if (deco
& OPMASK_MASK
)
2026 evex
|= deco
& EVEX_P2AAA
;
2032 static int op_evexflags(const operand
* o
, int mask
, uint8_t byte
)
2036 val
= nasm_regvals
[o
->basereg
];
2038 return evexflags(val
, o
->decoflags
, mask
, byte
);
2041 static enum match_result
find_match(const struct itemplate
**tempp
,
2043 int32_t segment
, int64_t offset
, int bits
)
2045 const struct itemplate
*temp
;
2046 enum match_result m
, merr
;
2047 opflags_t xsizeflags
[MAX_OPERANDS
];
2048 bool opsizemissing
= false;
2049 int8_t broadcast
= instruction
->evex_brerop
;
2052 /* broadcasting uses a different data element size */
2053 for (i
= 0; i
< instruction
->operands
; i
++)
2055 xsizeflags
[i
] = instruction
->oprs
[i
].decoflags
& BRSIZE_MASK
;
2057 xsizeflags
[i
] = instruction
->oprs
[i
].type
& SIZE_MASK
;
2059 merr
= MERR_INVALOP
;
2061 for (temp
= nasm_instructions
[instruction
->opcode
];
2062 temp
->opcode
!= I_none
; temp
++) {
2063 m
= matches(temp
, instruction
, bits
);
2064 if (m
== MOK_JUMP
) {
2065 if (jmp_match(segment
, offset
, bits
, instruction
, temp
))
2069 } else if (m
== MERR_OPSIZEMISSING
&& !itemp_has(temp
, IF_SX
)) {
2071 * Missing operand size and a candidate for fuzzy matching...
2073 for (i
= 0; i
< temp
->operands
; i
++)
2075 xsizeflags
[i
] |= temp
->deco
[i
] & BRSIZE_MASK
;
2077 xsizeflags
[i
] |= temp
->opd
[i
] & SIZE_MASK
;
2078 opsizemissing
= true;
2082 if (merr
== MOK_GOOD
)
2086 /* No match, but see if we can get a fuzzy operand size match... */
2090 for (i
= 0; i
< instruction
->operands
; i
++) {
2092 * We ignore extrinsic operand sizes on registers, so we should
2093 * never try to fuzzy-match on them. This also resolves the case
2094 * when we have e.g. "xmmrm128" in two different positions.
2096 if (is_class(REGISTER
, instruction
->oprs
[i
].type
))
2099 /* This tests if xsizeflags[i] has more than one bit set */
2100 if ((xsizeflags
[i
] & (xsizeflags
[i
]-1)))
2101 goto done
; /* No luck */
2103 if (i
== broadcast
) {
2104 instruction
->oprs
[i
].decoflags
|= xsizeflags
[i
];
2105 instruction
->oprs
[i
].type
|= (xsizeflags
[i
] == BR_BITS32
?
2108 instruction
->oprs
[i
].type
|= xsizeflags
[i
]; /* Set the size */
2112 /* Try matching again... */
2113 for (temp
= nasm_instructions
[instruction
->opcode
];
2114 temp
->opcode
!= I_none
; temp
++) {
2115 m
= matches(temp
, instruction
, bits
);
2116 if (m
== MOK_JUMP
) {
2117 if (jmp_match(segment
, offset
, bits
, instruction
, temp
))
2124 if (merr
== MOK_GOOD
)
2133 static uint8_t get_broadcast_num(opflags_t opflags
, opflags_t brsize
)
2135 unsigned int opsize
= (opflags
& SIZE_MASK
) >> SIZE_SHIFT
;
2138 if (brsize
> BITS64
)
2139 nasm_error(ERR_FATAL
,
2140 "size of broadcasting element is greater than 64 bits");
2143 * The shift term is to take care of the extra BITS80 inserted
2144 * between BITS64 and BITS128.
2146 brcast_num
= ((opsize
/ (BITS64
>> SIZE_SHIFT
)) * (BITS64
/ brsize
))
2147 >> (opsize
> (BITS64
>> SIZE_SHIFT
));
2152 static enum match_result
matches(const struct itemplate
*itemp
,
2153 insn
*instruction
, int bits
)
2155 opflags_t size
[MAX_OPERANDS
], asize
;
2156 bool opsizemissing
= false;
2162 if (itemp
->opcode
!= instruction
->opcode
)
2163 return MERR_INVALOP
;
2166 * Count the operands
2168 if (itemp
->operands
!= instruction
->operands
)
2169 return MERR_INVALOP
;
2174 if (!(optimizing
> 0) && itemp_has(itemp
, IF_OPT
))
2175 return MERR_INVALOP
;
2180 switch (instruction
->prefixes
[PPS_VEX
]) {
2182 if (!itemp_has(itemp
, IF_EVEX
))
2183 return MERR_ENCMISMATCH
;
2187 if (!itemp_has(itemp
, IF_VEX
))
2188 return MERR_ENCMISMATCH
;
2195 * Check that no spurious colons or TOs are present
2197 for (i
= 0; i
< itemp
->operands
; i
++)
2198 if (instruction
->oprs
[i
].type
& ~itemp
->opd
[i
] & (COLON
| TO
))
2199 return MERR_INVALOP
;
2202 * Process size flags
2204 switch (itemp_smask(itemp
)) {
2205 case IF_GENBIT(IF_SB
):
2208 case IF_GENBIT(IF_SW
):
2211 case IF_GENBIT(IF_SD
):
2214 case IF_GENBIT(IF_SQ
):
2217 case IF_GENBIT(IF_SO
):
2220 case IF_GENBIT(IF_SY
):
2223 case IF_GENBIT(IF_SZ
):
2226 case IF_GENBIT(IF_SIZE
):
2247 if (itemp_armask(itemp
)) {
2248 /* S- flags only apply to a specific operand */
2249 i
= itemp_arg(itemp
);
2250 memset(size
, 0, sizeof size
);
2253 /* S- flags apply to all operands */
2254 for (i
= 0; i
< MAX_OPERANDS
; i
++)
2259 * Check that the operand flags all match up,
2260 * it's a bit tricky so lets be verbose:
2262 * 1) Find out the size of operand. If instruction
2263 * doesn't have one specified -- we're trying to
2264 * guess it either from template (IF_S* flag) or
2267 * 2) If template operand do not match the instruction OR
2268 * template has an operand size specified AND this size differ
2269 * from which instruction has (perhaps we got it from code bits)
2271 * a) Check that only size of instruction and operand is differ
2272 * other characteristics do match
2273 * b) Perhaps it's a register specified in instruction so
2274 * for such a case we just mark that operand as "size
2275 * missing" and this will turn on fuzzy operand size
2276 * logic facility (handled by a caller)
2278 for (i
= 0; i
< itemp
->operands
; i
++) {
2279 opflags_t type
= instruction
->oprs
[i
].type
;
2280 decoflags_t deco
= instruction
->oprs
[i
].decoflags
;
2281 bool is_broadcast
= deco
& BRDCAST_MASK
;
2282 uint8_t brcast_num
= 0;
2283 opflags_t template_opsize
, insn_opsize
;
2285 if (!(type
& SIZE_MASK
))
2288 insn_opsize
= type
& SIZE_MASK
;
2289 if (!is_broadcast
) {
2290 template_opsize
= itemp
->opd
[i
] & SIZE_MASK
;
2292 decoflags_t deco_brsize
= itemp
->deco
[i
] & BRSIZE_MASK
;
2294 * when broadcasting, the element size depends on
2295 * the instruction type. decorator flag should match.
2299 template_opsize
= (deco_brsize
== BR_BITS32
? BITS32
: BITS64
);
2300 /* calculate the proper number : {1to<brcast_num>} */
2301 brcast_num
= get_broadcast_num(itemp
->opd
[i
], template_opsize
);
2303 template_opsize
= 0;
2307 if ((itemp
->opd
[i
] & ~type
& ~SIZE_MASK
) ||
2308 (deco
& ~itemp
->deco
[i
] & ~BRNUM_MASK
)) {
2309 return MERR_INVALOP
;
2310 } else if (template_opsize
) {
2311 if (template_opsize
!= insn_opsize
) {
2313 return MERR_INVALOP
;
2314 } else if (!is_class(REGISTER
, type
)) {
2316 * Note: we don't honor extrinsic operand sizes for registers,
2317 * so "missing operand size" for a register should be
2318 * considered a wildcard match rather than an error.
2320 opsizemissing
= true;
2322 } else if (is_broadcast
&&
2324 (2U << ((deco
& BRNUM_MASK
) >> BRNUM_SHIFT
)))) {
2326 * broadcasting opsize matches but the number of repeated memory
2327 * element does not match.
2328 * if 64b double precision float is broadcasted to ymm (256b),
2329 * broadcasting decorator must be {1to4}.
2331 return MERR_BRNUMMISMATCH
;
2337 return MERR_OPSIZEMISSING
;
2340 * Check operand sizes
2342 if (itemp_has(itemp
, IF_SM
) || itemp_has(itemp
, IF_SM2
)) {
2343 oprs
= (itemp_has(itemp
, IF_SM2
) ? 2 : itemp
->operands
);
2344 for (i
= 0; i
< oprs
; i
++) {
2345 asize
= itemp
->opd
[i
] & SIZE_MASK
;
2347 for (i
= 0; i
< oprs
; i
++)
2353 oprs
= itemp
->operands
;
2356 for (i
= 0; i
< itemp
->operands
; i
++) {
2357 if (!(itemp
->opd
[i
] & SIZE_MASK
) &&
2358 (instruction
->oprs
[i
].type
& SIZE_MASK
& ~size
[i
]))
2359 return MERR_OPSIZEMISMATCH
;
2363 * Check template is okay at the set cpu level
2365 if (iflag_cmp_cpu_level(&insns_flags
[itemp
->iflag_idx
], &cpu
) > 0)
2369 * Verify the appropriate long mode flag.
2371 if (itemp_has(itemp
, (bits
== 64 ? IF_NOLONG
: IF_LONG
)))
2372 return MERR_BADMODE
;
2375 * If we have a HLE prefix, look for the NOHLE flag
2377 if (itemp_has(itemp
, IF_NOHLE
) &&
2378 (has_prefix(instruction
, PPS_REP
, P_XACQUIRE
) ||
2379 has_prefix(instruction
, PPS_REP
, P_XRELEASE
)))
2383 * Check if special handling needed for Jumps
2385 if ((itemp
->code
[0] & ~1) == 0370)
2389 * Check if BND prefix is allowed.
2390 * Other 0xF2 (REPNE/REPNZ) prefix is prohibited.
2392 if (!itemp_has(itemp
, IF_BND
) &&
2393 (has_prefix(instruction
, PPS_REP
, P_BND
) ||
2394 has_prefix(instruction
, PPS_REP
, P_NOBND
)))
2396 else if (itemp_has(itemp
, IF_BND
) &&
2397 (has_prefix(instruction
, PPS_REP
, P_REPNE
) ||
2398 has_prefix(instruction
, PPS_REP
, P_REPNZ
)))
2399 return MERR_BADREPNE
;
2405 * Check if ModR/M.mod should/can be 01.
2406 * - EAF_BYTEOFFS is set
2407 * - offset can fit in a byte when EVEX is not used
2408 * - offset can be compressed when EVEX is used
2410 #define IS_MOD_01() (input->eaflags & EAF_BYTEOFFS || \
2411 (o >= -128 && o <= 127 && \
2412 seg == NO_SEG && !forw_ref && \
2413 !(input->eaflags & EAF_WORDOFFS) && \
2414 !(ins->rex & REX_EV)) || \
2415 (ins->rex & REX_EV && \
2416 is_disp8n(input, ins, &output->disp8)))
2418 static enum ea_type
process_ea(operand
*input
, ea
*output
, int bits
,
2419 int rfield
, opflags_t rflags
, insn
*ins
)
2421 bool forw_ref
= !!(input
->opflags
& OPFLAG_UNKNOWN
);
2422 int addrbits
= ins
->addr_size
;
2423 int eaflags
= input
->eaflags
;
2425 output
->type
= EA_SCALAR
;
2426 output
->rip
= false;
2429 /* REX flags for the rfield operand */
2430 output
->rex
|= rexflags(rfield
, rflags
, REX_R
| REX_P
| REX_W
| REX_H
);
2431 /* EVEX.R' flag for the REG operand */
2432 ins
->evex_p
[0] |= evexflags(rfield
, 0, EVEX_P0RP
, 0);
2434 if (is_class(REGISTER
, input
->type
)) {
2436 * It's a direct register.
2438 if (!is_register(input
->basereg
))
2441 if (!is_reg_class(REG_EA
, input
->basereg
))
2444 /* broadcasting is not available with a direct register operand. */
2445 if (input
->decoflags
& BRDCAST_MASK
) {
2446 nasm_error(ERR_NONFATAL
, "Broadcasting not allowed from a register");
2450 output
->rex
|= op_rexflags(input
, REX_B
| REX_P
| REX_W
| REX_H
);
2451 ins
->evex_p
[0] |= op_evexflags(input
, EVEX_P0X
, 0);
2452 output
->sib_present
= false; /* no SIB necessary */
2453 output
->bytes
= 0; /* no offset necessary either */
2454 output
->modrm
= GEN_MODRM(3, rfield
, nasm_regvals
[input
->basereg
]);
2457 * It's a memory reference.
2460 /* Embedded rounding or SAE is not available with a mem ref operand. */
2461 if (input
->decoflags
& (ER
| SAE
)) {
2462 nasm_error(ERR_NONFATAL
,
2463 "Embedded rounding is available only with reg-reg op.");
2467 if (input
->basereg
== -1 &&
2468 (input
->indexreg
== -1 || input
->scale
== 0)) {
2470 * It's a pure offset.
2472 if (bits
== 64 && ((input
->type
& IP_REL
) == IP_REL
)) {
2473 if (input
->segment
== NO_SEG
|| (input
->opflags
& OPFLAG_RELATIVE
)) {
2474 nasm_error(ERR_WARNING
| ERR_PASS2
, "absolute address can not be RIP-relative");
2475 input
->type
&= ~IP_REL
;
2476 input
->type
|= MEMORY
;
2481 !(IP_REL
& ~input
->type
) && (eaflags
& EAF_MIB
)) {
2482 nasm_error(ERR_NONFATAL
, "RIP-relative addressing is prohibited for mib.");
2486 if (eaflags
& EAF_BYTEOFFS
||
2487 (eaflags
& EAF_WORDOFFS
&&
2488 input
->disp_size
!= (addrbits
!= 16 ? 32 : 16))) {
2489 nasm_error(ERR_WARNING
| ERR_PASS1
, "displacement size ignored on absolute address");
2492 if (bits
== 64 && (~input
->type
& IP_REL
)) {
2493 output
->sib_present
= true;
2494 output
->sib
= GEN_SIB(0, 4, 5);
2496 output
->modrm
= GEN_MODRM(0, rfield
, 4);
2497 output
->rip
= false;
2499 output
->sib_present
= false;
2500 output
->bytes
= (addrbits
!= 16 ? 4 : 2);
2501 output
->modrm
= GEN_MODRM(0, rfield
, (addrbits
!= 16 ? 5 : 6));
2502 output
->rip
= bits
== 64;
2506 * It's an indirection.
2508 int i
= input
->indexreg
, b
= input
->basereg
, s
= input
->scale
;
2509 int32_t seg
= input
->segment
;
2510 int hb
= input
->hintbase
, ht
= input
->hinttype
;
2511 int t
, it
, bt
; /* register numbers */
2512 opflags_t x
, ix
, bx
; /* register flags */
2515 i
= -1; /* make this easy, at least */
2517 if (is_register(i
)) {
2518 it
= nasm_regvals
[i
];
2519 ix
= nasm_reg_flags
[i
];
2525 if (is_register(b
)) {
2526 bt
= nasm_regvals
[b
];
2527 bx
= nasm_reg_flags
[b
];
2533 /* if either one are a vector register... */
2534 if ((ix
|bx
) & (XMMREG
|YMMREG
|ZMMREG
) & ~REG_EA
) {
2535 opflags_t sok
= BITS32
| BITS64
;
2536 int32_t o
= input
->offset
;
2537 int mod
, scale
, index
, base
;
2540 * For a vector SIB, one has to be a vector and the other,
2541 * if present, a GPR. The vector must be the index operand.
2543 if (it
== -1 || (bx
& (XMMREG
|YMMREG
|ZMMREG
) & ~REG_EA
)) {
2549 t
= bt
, bt
= it
, it
= t
;
2550 x
= bx
, bx
= ix
, ix
= x
;
2556 if (!(REG64
& ~bx
) || !(REG32
& ~bx
))
2563 * While we're here, ensure the user didn't specify
2566 if (input
->disp_size
== 16 || input
->disp_size
== 64)
2569 if (addrbits
== 16 ||
2570 (addrbits
== 32 && !(sok
& BITS32
)) ||
2571 (addrbits
== 64 && !(sok
& BITS64
)))
2574 output
->type
= ((ix
& ZMMREG
& ~REG_EA
) ? EA_ZMMVSIB
2575 : ((ix
& YMMREG
& ~REG_EA
)
2576 ? EA_YMMVSIB
: EA_XMMVSIB
));
2578 output
->rex
|= rexflags(it
, ix
, REX_X
);
2579 output
->rex
|= rexflags(bt
, bx
, REX_B
);
2580 ins
->evex_p
[2] |= evexflags(it
, 0, EVEX_P2VP
, 2);
2582 index
= it
& 7; /* it is known to be != -1 */
2597 default: /* then what the smeg is it? */
2598 goto err
; /* panic */
2606 if (base
!= REG_NUM_EBP
&& o
== 0 &&
2607 seg
== NO_SEG
&& !forw_ref
&&
2608 !(eaflags
& (EAF_BYTEOFFS
| EAF_WORDOFFS
)))
2610 else if (IS_MOD_01())
2616 output
->sib_present
= true;
2617 output
->bytes
= (bt
== -1 || mod
== 2 ? 4 : mod
);
2618 output
->modrm
= GEN_MODRM(mod
, rfield
, 4);
2619 output
->sib
= GEN_SIB(scale
, index
, base
);
2620 } else if ((ix
|bx
) & (BITS32
|BITS64
)) {
2622 * it must be a 32/64-bit memory reference. Firstly we have
2623 * to check that all registers involved are type E/Rxx.
2625 opflags_t sok
= BITS32
| BITS64
;
2626 int32_t o
= input
->offset
;
2629 if (!(REG64
& ~ix
) || !(REG32
& ~ix
))
2637 goto err
; /* Invalid register */
2638 if (~sok
& bx
& SIZE_MASK
)
2639 goto err
; /* Invalid size */
2644 * While we're here, ensure the user didn't specify
2647 if (input
->disp_size
== 16 || input
->disp_size
== 64)
2650 if (addrbits
== 16 ||
2651 (addrbits
== 32 && !(sok
& BITS32
)) ||
2652 (addrbits
== 64 && !(sok
& BITS64
)))
2655 /* now reorganize base/index */
2656 if (s
== 1 && bt
!= it
&& bt
!= -1 && it
!= -1 &&
2657 ((hb
== b
&& ht
== EAH_NOTBASE
) ||
2658 (hb
== i
&& ht
== EAH_MAKEBASE
))) {
2659 /* swap if hints say so */
2660 t
= bt
, bt
= it
, it
= t
;
2661 x
= bx
, bx
= ix
, ix
= x
;
2664 if (bt
== -1 && s
== 1 && !(hb
== i
&& ht
== EAH_NOTBASE
)) {
2665 /* make single reg base, unless hint */
2666 bt
= it
, bx
= ix
, it
= -1, ix
= 0;
2668 if (eaflags
& EAF_MIB
) {
2669 /* only for mib operands */
2670 if (it
== -1 && (hb
== b
&& ht
== EAH_NOTBASE
)) {
2672 * make a single reg index [reg*1].
2673 * gas uses this form for an explicit index register.
2675 it
= bt
, ix
= bx
, bt
= -1, bx
= 0, s
= 1;
2677 if ((ht
== EAH_SUMMED
) && bt
== -1) {
2678 /* separate once summed index into [base, index] */
2679 bt
= it
, bx
= ix
, s
--;
2682 if (((s
== 2 && it
!= REG_NUM_ESP
&&
2683 (!(eaflags
& EAF_TIMESTWO
) || (ht
== EAH_SUMMED
))) ||
2684 s
== 3 || s
== 5 || s
== 9) && bt
== -1) {
2685 /* convert 3*EAX to EAX+2*EAX */
2686 bt
= it
, bx
= ix
, s
--;
2688 if (it
== -1 && (bt
& 7) != REG_NUM_ESP
&&
2689 (eaflags
& EAF_TIMESTWO
) &&
2690 (hb
== b
&& ht
== EAH_NOTBASE
)) {
2692 * convert [NOSPLIT EAX*1]
2693 * to sib format with 0x0 displacement - [EAX*1+0].
2695 it
= bt
, ix
= bx
, bt
= -1, bx
= 0, s
= 1;
2698 if (s
== 1 && it
== REG_NUM_ESP
) {
2699 /* swap ESP into base if scale is 1 */
2700 t
= it
, it
= bt
, bt
= t
;
2701 x
= ix
, ix
= bx
, bx
= x
;
2703 if (it
== REG_NUM_ESP
||
2704 (s
!= 1 && s
!= 2 && s
!= 4 && s
!= 8 && it
!= -1))
2705 goto err
; /* wrong, for various reasons */
2707 output
->rex
|= rexflags(it
, ix
, REX_X
);
2708 output
->rex
|= rexflags(bt
, bx
, REX_B
);
2710 if (it
== -1 && (bt
& 7) != REG_NUM_ESP
) {
2719 if (rm
!= REG_NUM_EBP
&& o
== 0 &&
2720 seg
== NO_SEG
&& !forw_ref
&&
2721 !(eaflags
& (EAF_BYTEOFFS
| EAF_WORDOFFS
)))
2723 else if (IS_MOD_01())
2729 output
->sib_present
= false;
2730 output
->bytes
= (bt
== -1 || mod
== 2 ? 4 : mod
);
2731 output
->modrm
= GEN_MODRM(mod
, rfield
, rm
);
2734 int mod
, scale
, index
, base
;
2754 default: /* then what the smeg is it? */
2755 goto err
; /* panic */
2763 if (base
!= REG_NUM_EBP
&& o
== 0 &&
2764 seg
== NO_SEG
&& !forw_ref
&&
2765 !(eaflags
& (EAF_BYTEOFFS
| EAF_WORDOFFS
)))
2767 else if (IS_MOD_01())
2773 output
->sib_present
= true;
2774 output
->bytes
= (bt
== -1 || mod
== 2 ? 4 : mod
);
2775 output
->modrm
= GEN_MODRM(mod
, rfield
, 4);
2776 output
->sib
= GEN_SIB(scale
, index
, base
);
2778 } else { /* it's 16-bit */
2780 int16_t o
= input
->offset
;
2782 /* check for 64-bit long mode */
2786 /* check all registers are BX, BP, SI or DI */
2787 if ((b
!= -1 && b
!= R_BP
&& b
!= R_BX
&& b
!= R_SI
&& b
!= R_DI
) ||
2788 (i
!= -1 && i
!= R_BP
&& i
!= R_BX
&& i
!= R_SI
&& i
!= R_DI
))
2791 /* ensure the user didn't specify DWORD/QWORD */
2792 if (input
->disp_size
== 32 || input
->disp_size
== 64)
2795 if (s
!= 1 && i
!= -1)
2796 goto err
; /* no can do, in 16-bit EA */
2797 if (b
== -1 && i
!= -1) {
2802 if ((b
== R_SI
|| b
== R_DI
) && i
!= -1) {
2807 /* have BX/BP as base, SI/DI index */
2809 goto err
; /* shouldn't ever happen, in theory */
2810 if (i
!= -1 && b
!= -1 &&
2811 (i
== R_BP
|| i
== R_BX
|| b
== R_SI
|| b
== R_DI
))
2812 goto err
; /* invalid combinations */
2813 if (b
== -1) /* pure offset: handled above */
2814 goto err
; /* so if it gets to here, panic! */
2818 switch (i
* 256 + b
) {
2819 case R_SI
* 256 + R_BX
:
2822 case R_DI
* 256 + R_BX
:
2825 case R_SI
* 256 + R_BP
:
2828 case R_DI
* 256 + R_BP
:
2846 if (rm
== -1) /* can't happen, in theory */
2847 goto err
; /* so panic if it does */
2849 if (o
== 0 && seg
== NO_SEG
&& !forw_ref
&& rm
!= 6 &&
2850 !(eaflags
& (EAF_BYTEOFFS
| EAF_WORDOFFS
)))
2852 else if (IS_MOD_01())
2857 output
->sib_present
= false; /* no SIB - it's 16-bit */
2858 output
->bytes
= mod
; /* bytes of offset needed */
2859 output
->modrm
= GEN_MODRM(mod
, rfield
, rm
);
2864 output
->size
= 1 + output
->sib_present
+ output
->bytes
;
2865 return output
->type
;
2868 return output
->type
= EA_INVALID
;
2871 static void add_asp(insn
*ins
, int addrbits
)
2876 valid
= (addrbits
== 64) ? 64|32 : 32|16;
2878 switch (ins
->prefixes
[PPS_ASIZE
]) {
2889 valid
&= (addrbits
== 32) ? 16 : 32;
2895 for (j
= 0; j
< ins
->operands
; j
++) {
2896 if (is_class(MEMORY
, ins
->oprs
[j
].type
)) {
2899 /* Verify as Register */
2900 if (!is_register(ins
->oprs
[j
].indexreg
))
2903 i
= nasm_reg_flags
[ins
->oprs
[j
].indexreg
];
2905 /* Verify as Register */
2906 if (!is_register(ins
->oprs
[j
].basereg
))
2909 b
= nasm_reg_flags
[ins
->oprs
[j
].basereg
];
2911 if (ins
->oprs
[j
].scale
== 0)
2915 int ds
= ins
->oprs
[j
].disp_size
;
2916 if ((addrbits
!= 64 && ds
> 8) ||
2917 (addrbits
== 64 && ds
== 16))
2937 if (valid
& addrbits
) {
2938 ins
->addr_size
= addrbits
;
2939 } else if (valid
& ((addrbits
== 32) ? 16 : 32)) {
2940 /* Add an address size prefix */
2941 ins
->prefixes
[PPS_ASIZE
] = (addrbits
== 32) ? P_A16
: P_A32
;;
2942 ins
->addr_size
= (addrbits
== 32) ? 16 : 32;
2945 nasm_error(ERR_NONFATAL
, "impossible combination of address sizes");
2946 ins
->addr_size
= addrbits
; /* Error recovery */
2949 defdisp
= ins
->addr_size
== 16 ? 16 : 32;
2951 for (j
= 0; j
< ins
->operands
; j
++) {
2952 if (!(MEM_OFFS
& ~ins
->oprs
[j
].type
) &&
2953 (ins
->oprs
[j
].disp_size
? ins
->oprs
[j
].disp_size
: defdisp
) != ins
->addr_size
) {
2955 * mem_offs sizes must match the address size; if not,
2956 * strip the MEM_OFFS bit and match only EA instructions
2958 ins
->oprs
[j
].type
&= ~(MEM_OFFS
& ~MEMORY
);