1 /* ----------------------------------------------------------------------- *
3 * Copyright 1996-2016 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 int has_prefix(insn
* ins
, enum prefix_pos pos
, int prefix
)
250 return ins
->prefixes
[pos
] == prefix
;
253 static void assert_no_prefix(insn
* ins
, enum prefix_pos pos
)
255 if (ins
->prefixes
[pos
])
256 nasm_error(ERR_NONFATAL
, "invalid %s prefix",
257 prefix_name(ins
->prefixes
[pos
]));
260 static const char *size_name(int size
)
284 static void warn_overflow(int pass
, int size
)
286 nasm_error(ERR_WARNING
| pass
| ERR_WARN_NOV
,
287 "%s data exceeds bounds", size_name(size
));
290 static void warn_overflow_const(int64_t data
, int size
)
292 if (overflow_general(data
, size
))
293 warn_overflow(ERR_PASS1
, size
);
296 static void warn_overflow_opd(const struct operand
*o
, int size
)
298 if (o
->wrt
== NO_SEG
&& o
->segment
== NO_SEG
) {
299 if (overflow_general(o
->offset
, size
))
300 warn_overflow(ERR_PASS2
, size
);
305 * This routine wrappers the real output format's output routine,
306 * in order to pass a copy of the data off to the listing file
307 * generator at the same time, flatten unnecessary relocations,
308 * and verify backend compatibility.
310 static void out(struct out_data
*data
)
312 static int32_t lineno
= 0; /* static!!! */
313 static const char *lnfname
= NULL
;
315 const int amax
= ofmt
->maxbits
>> 3; /* Maximum address size in bytes */
320 uint64_t size
= data
->size
;
323 return; /* Nothing to do */
325 switch (data
->type
) {
328 nasm_assert(asize
<= 8);
329 if (data
->tsegment
== NO_SEG
&& data
->twrt
== NO_SEG
) {
330 /* Convert to RAWDATA */
331 /* XXX: check for overflow */
332 uint8_t *q
= xdata
.b
;
334 WRITEADDR(q
, data
->toffset
, asize
);
335 data
->data
= xdata
.b
;
336 data
->type
= OUT_RAWDATA
;
337 asize
= 0; /* No longer an address */
343 nasm_assert(asize
<= 8);
344 if (data
->tsegment
== data
->segment
&& data
->twrt
== NO_SEG
) {
345 /* Convert to RAWDATA */
346 uint8_t *q
= xdata
.b
;
347 int64_t delta
= data
->toffset
- data
->offset
348 - (data
->inslen
- data
->insoffs
);
350 if (overflow_signed(delta
, asize
))
351 warn_overflow(ERR_PASS2
, asize
);
353 WRITEADDR(q
, delta
, asize
);
354 data
->data
= xdata
.b
;
355 data
->type
= OUT_RAWDATA
;
356 asize
= 0; /* No longer an address */
361 asize
= 0; /* Not an address */
368 * this call to src_get determines when we call the
369 * debug-format-specific "linenum" function
370 * it updates lineno and lnfname to the current values
371 * returning 0 if "same as last time", -2 if lnfname
372 * changed, and the amount by which lineno changed,
373 * if it did. thus, these variables must be static
376 if (src_get(&lineno
, &lnfname
))
377 dfmt
->linenum(lnfname
, lineno
, data
->segment
);
379 if (asize
&& asize
> amax
) {
380 if (data
->type
!= OUT_ADDRESS
|| data
->sign
== OUT_SIGNED
) {
381 nasm_error(ERR_NONFATAL
,
382 "%d-bit signed relocation unsupported by output format %s\n",
383 asize
<< 3, ofmt
->shortname
);
385 nasm_error(ERR_WARNING
| ERR_WARN_ZEXTRELOC
,
386 "%d-bit unsigned relocation zero-extended from %d bits\n",
387 asize
<< 3, ofmt
->maxbits
);
390 data
->insoffs
+= amax
;
391 data
->offset
+= amax
;
392 data
->size
= size
= asize
- amax
;
394 data
->data
= zero_buffer
;
395 data
->type
= OUT_RAWDATA
;
399 data
->offset
+= size
;
400 data
->insoffs
+= size
;
403 static inline void out_rawdata(struct out_data
*data
, const void *rawdata
,
406 data
->type
= OUT_RAWDATA
;
407 data
->data
= rawdata
;
412 static void out_rawbyte(struct out_data
*data
, uint8_t byte
)
414 data
->type
= OUT_RAWDATA
;
420 static inline void out_reserve(struct out_data
*data
, uint64_t size
)
422 data
->type
= OUT_RESERVE
;
427 static inline void out_imm(struct out_data
*data
, struct operand
*opx
,
428 int size
, enum out_sign sign
)
430 data
->type
= OUT_ADDRESS
;
433 data
->toffset
= opx
->offset
;
434 data
->tsegment
= opx
->segment
;
435 data
->twrt
= opx
->wrt
;
439 static inline void out_reladdr(struct out_data
*data
, struct operand
*opx
,
442 data
->type
= OUT_RELADDR
;
443 data
->sign
= OUT_SIGNED
;
445 data
->toffset
= opx
->offset
;
446 data
->tsegment
= opx
->segment
;
447 data
->twrt
= opx
->wrt
;
451 static inline void out_segment(struct out_data
*data
, struct operand
*opx
)
453 data
->type
= OUT_SEGMENT
;
454 data
->sign
= OUT_UNSIGNED
;
456 data
->toffset
= opx
->offset
;
457 data
->tsegment
= ofmt
->segbase(opx
->segment
+ 1);
458 data
->twrt
= opx
->wrt
;
462 static bool jmp_match(int32_t segment
, int64_t offset
, int bits
,
463 insn
* ins
, const struct itemplate
*temp
)
466 const uint8_t *code
= temp
->code
;
470 if (((c
& ~1) != 0370) || (ins
->oprs
[0].type
& STRICT
))
474 if (optimizing
< 0 && c
== 0371)
477 isize
= calcsize(segment
, offset
, bits
, ins
, temp
);
479 if (ins
->oprs
[0].opflags
& OPFLAG_UNKNOWN
)
480 /* Be optimistic in pass 1 */
483 if (ins
->oprs
[0].segment
!= segment
)
486 isize
= ins
->oprs
[0].offset
- offset
- isize
; /* isize is delta */
487 is_byte
= (isize
>= -128 && isize
<= 127); /* is it byte size? */
489 if (is_byte
&& c
== 0371 && ins
->prefixes
[PPS_REP
] == P_BND
) {
490 /* jmp short (opcode eb) cannot be used with bnd prefix. */
491 ins
->prefixes
[PPS_REP
] = P_none
;
492 nasm_error(ERR_WARNING
| ERR_WARN_BND
| ERR_PASS2
,
493 "jmp short does not init bnd regs - bnd prefix dropped.");
499 /* This is totally just a wild guess what is reasonable... */
500 #define INCBIN_MAX_BUF (ZERO_BUF_SIZE * 16)
502 int64_t assemble(int32_t segment
, int64_t start
, int bits
, iflag_t cp
,
505 struct out_data data
;
506 const struct itemplate
*temp
;
509 int64_t wsize
; /* size for DB etc. */
514 data
.segment
= segment
;
516 data
.sign
= OUT_WRAP
;
519 wsize
= idata_bytes(instruction
->opcode
);
525 int32_t t
= instruction
->times
;
527 nasm_panic(0, "instruction->times < 0 (%"PRId32
") in assemble()", t
);
529 while (t
--) { /* repeat TIMES times */
530 list_for_each(e
, instruction
->eops
) {
531 if (e
->type
== EOT_DB_NUMBER
) {
533 nasm_error(ERR_NONFATAL
,
534 "integer supplied to a DT, DO or DY"
538 data
.type
= OUT_ADDRESS
;
539 data
.inslen
= data
.size
= wsize
;
540 data
.toffset
= e
->offset
;
541 data
.tsegment
= e
->segment
;
545 } else if (e
->type
== EOT_DB_STRING
||
546 e
->type
== EOT_DB_STRING_FREE
) {
547 int align
= e
->stringlen
% wsize
;
549 align
= wsize
- align
;
552 data
.inslen
= e
->stringlen
+ align
;
554 out_rawdata(&data
, e
->stringval
, e
->stringlen
);
555 out_rawdata(&data
, zero_buffer
, align
);
558 if (t
> 0 && t
== instruction
->times
- 1) {
559 lfmt
->set_offset(data
.offset
);
560 lfmt
->uplevel(LIST_TIMES
);
563 if (instruction
->times
> 1)
564 lfmt
->downlevel(LIST_TIMES
);
565 } else if (instruction
->opcode
== I_INCBIN
) {
566 const char *fname
= instruction
->eops
->stringval
;
568 size_t t
= instruction
->times
;
571 const void *map
= NULL
;
573 size_t blk
= 0; /* Buffered I/O block size */
574 size_t m
= 0; /* Bytes last read */
576 fp
= nasm_open_read(fname
, NF_BINARY
|NF_FORMAP
);
578 nasm_error(ERR_NONFATAL
, "`incbin': unable to open file `%s'",
583 len
= nasm_file_size(fp
);
585 if (len
== (off_t
)-1) {
586 nasm_error(ERR_NONFATAL
, "`incbin': unable to get length of file `%s'",
591 if (instruction
->eops
->next
) {
592 base
= instruction
->eops
->next
->offset
;
597 if (instruction
->eops
->next
->next
&&
598 len
> (off_t
)instruction
->eops
->next
->next
->offset
)
599 len
= (off_t
)instruction
->eops
->next
->next
->offset
;
603 lfmt
->set_offset(data
.offset
);
604 lfmt
->uplevel(LIST_INCBIN
);
609 /* Try to map file data */
610 map
= nasm_map_file(fp
, base
, len
);
612 blk
= len
< (off_t
)INCBIN_MAX_BUF
? (size_t)len
: INCBIN_MAX_BUF
;
613 buf
= nasm_malloc(blk
);
618 * Consider these irrelevant for INCBIN, since it is fully
619 * possible that these might be (way) bigger than an int
620 * can hold; there is, however, no reason to widen these
621 * types just for INCBIN. data.inslen == 0 signals to the
622 * backend that these fields are meaningless, if at all
629 out_rawdata(&data
, map
, len
);
630 } else if ((off_t
)m
== len
) {
631 out_rawdata(&data
, buf
, len
);
635 if (fseeko(fp
, base
, SEEK_SET
) < 0 || ferror(fp
)) {
636 nasm_error(ERR_NONFATAL
,
637 "`incbin': unable to seek on file `%s'",
642 m
= fread(buf
, 1, l
< (off_t
)blk
? (size_t)l
: blk
, fp
);
643 if (!m
|| feof(fp
)) {
645 * This shouldn't happen unless the file
646 * actually changes while we are reading
649 nasm_error(ERR_NONFATAL
,
650 "`incbin': unexpected EOF while"
651 " reading file `%s'", fname
);
654 out_rawdata(&data
, buf
, m
);
660 lfmt
->downlevel(LIST_INCBIN
);
661 if (instruction
->times
> 1) {
662 lfmt
->set_offset(data
.offset
);
663 lfmt
->uplevel(LIST_TIMES
);
664 lfmt
->downlevel(LIST_TIMES
);
667 nasm_error(ERR_NONFATAL
,
668 "`incbin': error while"
669 " reading file `%s'", fname
);
675 nasm_unmap_file(map
, len
);
680 /* "Real" instruction */
682 /* Check to see if we need an address-size prefix */
683 add_asp(instruction
, bits
);
685 m
= find_match(&temp
, instruction
, data
.segment
, data
.offset
, bits
);
689 int64_t insn_size
= calcsize(data
.segment
, data
.offset
,
690 bits
, instruction
, temp
);
691 itimes
= instruction
->times
;
692 if (insn_size
< 0) /* shouldn't be, on pass two */
693 nasm_panic(0, "errors made it through from pass one");
700 data
.inslen
= insn_size
;
702 gencode(&data
, instruction
);
703 nasm_assert(data
.insoffs
== insn_size
);
705 if (itimes
> 0 && itimes
== instruction
->times
- 1) {
706 lfmt
->set_offset(data
.offset
);
707 lfmt
->uplevel(LIST_TIMES
);
710 if (instruction
->times
> 1)
711 lfmt
->downlevel(LIST_TIMES
);
715 case MERR_OPSIZEMISSING
:
716 nasm_error(ERR_NONFATAL
, "operation size not specified");
718 case MERR_OPSIZEMISMATCH
:
719 nasm_error(ERR_NONFATAL
, "mismatch in operand sizes");
721 case MERR_BRNUMMISMATCH
:
722 nasm_error(ERR_NONFATAL
,
723 "mismatch in the number of broadcasting elements");
726 nasm_error(ERR_NONFATAL
, "no instruction for this cpu level");
729 nasm_error(ERR_NONFATAL
, "instruction not supported in %d-bit mode",
732 case MERR_ENCMISMATCH
:
733 nasm_error(ERR_NONFATAL
, "specific encoding scheme not available");
736 nasm_error(ERR_NONFATAL
, "bnd prefix is not allowed");
739 nasm_error(ERR_NONFATAL
, "%s prefix is not allowed",
740 (has_prefix(instruction
, PPS_REP
, P_REPNE
) ?
744 nasm_error(ERR_NONFATAL
,
745 "invalid combination of opcode and operands");
750 return data
.offset
- start
;
753 int64_t insn_size(int32_t segment
, int64_t offset
, int bits
, iflag_t cp
,
756 const struct itemplate
*temp
;
761 if (instruction
->opcode
== I_none
)
764 if (instruction
->opcode
== I_DB
|| instruction
->opcode
== I_DW
||
765 instruction
->opcode
== I_DD
|| instruction
->opcode
== I_DQ
||
766 instruction
->opcode
== I_DT
|| instruction
->opcode
== I_DO
||
767 instruction
->opcode
== I_DY
) {
769 int32_t isize
, osize
, wsize
;
772 wsize
= idata_bytes(instruction
->opcode
);
774 list_for_each(e
, instruction
->eops
) {
778 if (e
->type
== EOT_DB_NUMBER
) {
780 warn_overflow_const(e
->offset
, wsize
);
781 } else if (e
->type
== EOT_DB_STRING
||
782 e
->type
== EOT_DB_STRING_FREE
)
783 osize
= e
->stringlen
;
785 align
= (-osize
) % wsize
;
788 isize
+= osize
+ align
;
793 if (instruction
->opcode
== I_INCBIN
) {
794 const char *fname
= instruction
->eops
->stringval
;
797 len
= nasm_file_size_by_path(fname
);
798 if (len
== (off_t
)-1) {
799 nasm_error(ERR_NONFATAL
, "`incbin': unable to get length of file `%s'",
804 if (instruction
->eops
->next
) {
805 if (len
<= (off_t
)instruction
->eops
->next
->offset
) {
808 len
-= instruction
->eops
->next
->offset
;
809 if (instruction
->eops
->next
->next
&&
810 len
> (off_t
)instruction
->eops
->next
->next
->offset
) {
811 len
= (off_t
)instruction
->eops
->next
->next
->offset
;
819 /* Check to see if we need an address-size prefix */
820 add_asp(instruction
, bits
);
822 m
= find_match(&temp
, instruction
, segment
, offset
, bits
);
824 /* we've matched an instruction. */
825 return calcsize(segment
, offset
, bits
, instruction
, temp
);
827 return -1; /* didn't match any instruction */
831 static void bad_hle_warn(const insn
* ins
, uint8_t hleok
)
833 enum prefixes rep_pfx
= ins
->prefixes
[PPS_REP
];
834 enum whatwarn
{ w_none
, w_lock
, w_inval
} ww
;
835 static const enum whatwarn warn
[2][4] =
837 { w_inval
, w_inval
, w_none
, w_lock
}, /* XACQUIRE */
838 { w_inval
, w_none
, w_none
, w_lock
}, /* XRELEASE */
842 n
= (unsigned int)rep_pfx
- P_XACQUIRE
;
844 return; /* Not XACQUIRE/XRELEASE */
847 if (!is_class(MEMORY
, ins
->oprs
[0].type
))
848 ww
= w_inval
; /* HLE requires operand 0 to be memory */
855 if (ins
->prefixes
[PPS_LOCK
] != P_LOCK
) {
856 nasm_error(ERR_WARNING
| ERR_WARN_HLE
| ERR_PASS2
,
857 "%s with this instruction requires lock",
858 prefix_name(rep_pfx
));
863 nasm_error(ERR_WARNING
| ERR_WARN_HLE
| ERR_PASS2
,
864 "%s invalid with this instruction",
865 prefix_name(rep_pfx
));
870 /* Common construct */
871 #define case3(x) case (x): case (x)+1: case (x)+2
872 #define case4(x) case3(x): case (x)+3
874 static int64_t calcsize(int32_t segment
, int64_t offset
, int bits
,
875 insn
* ins
, const struct itemplate
*temp
)
877 const uint8_t *codes
= temp
->code
;
886 bool lockcheck
= true;
887 enum reg_enum mib_index
= R_none
; /* For a separate index MIB reg form */
889 ins
->rex
= 0; /* Ensure REX is reset */
890 eat
= EA_SCALAR
; /* Expect a scalar EA */
891 memset(ins
->evex_p
, 0, 3); /* Ensure EVEX is reset */
893 if (ins
->prefixes
[PPS_OSIZE
] == P_O64
)
896 (void)segment
; /* Don't warn that this parameter is unused */
897 (void)offset
; /* Don't warn that this parameter is unused */
901 op1
= (c
& 3) + ((opex
& 1) << 2);
902 op2
= ((c
>> 3) & 3) + ((opex
& 2) << 1);
903 opx
= &ins
->oprs
[op1
];
904 opex
= 0; /* For the next iteration */
908 codes
+= c
, length
+= c
;
917 op_rexflags(opx
, REX_B
|REX_H
|REX_P
|REX_W
);
922 /* this is an index reg of MIB operand */
923 mib_index
= opx
->basereg
;
936 if (opx
->type
& (BITS16
| BITS32
| BITS64
))
937 length
+= (opx
->type
& BITS16
) ? 2 : 4;
939 length
+= (bits
== 16) ? 2 : 4;
947 length
+= ins
->addr_size
>> 3;
955 length
+= 8; /* MOV reg64/imm */
963 if (opx
->type
& (BITS16
| BITS32
| BITS64
))
964 length
+= (opx
->type
& BITS16
) ? 2 : 4;
966 length
+= (bits
== 16) ? 2 : 4;
989 ins
->vexreg
= regval(opx
);
990 ins
->evex_p
[2] |= op_evexflags(opx
, EVEX_P2VP
, 2); /* High-16 NDS */
991 ins
->vex_cm
= *codes
++;
992 ins
->vex_wlp
= *codes
++;
993 ins
->evex_tuple
= (*codes
++ - 0300);
999 ins
->vex_cm
= *codes
++;
1000 ins
->vex_wlp
= *codes
++;
1001 ins
->evex_tuple
= (*codes
++ - 0300);
1010 ins
->vexreg
= regval(opx
);
1011 ins
->vex_cm
= *codes
++;
1012 ins
->vex_wlp
= *codes
++;
1018 ins
->vex_cm
= *codes
++;
1019 ins
->vex_wlp
= *codes
++;
1036 length
+= (bits
!= 16) && !has_prefix(ins
, PPS_ASIZE
, P_A16
);
1040 length
+= (bits
!= 32) && !has_prefix(ins
, PPS_ASIZE
, P_A32
);
1047 if (bits
!= 64 || has_prefix(ins
, PPS_ASIZE
, P_A16
) ||
1048 has_prefix(ins
, PPS_ASIZE
, P_A32
))
1057 enum prefixes pfx
= ins
->prefixes
[PPS_OSIZE
];
1061 nasm_error(ERR_WARNING
| ERR_PASS2
, "invalid operand size prefix");
1063 ins
->prefixes
[PPS_OSIZE
] = P_O16
;
1069 enum prefixes pfx
= ins
->prefixes
[PPS_OSIZE
];
1073 nasm_error(ERR_WARNING
| ERR_PASS2
, "invalid operand size prefix");
1075 ins
->prefixes
[PPS_OSIZE
] = P_O32
;
1117 if (!ins
->prefixes
[PPS_REP
])
1118 ins
->prefixes
[PPS_REP
] = P_REP
;
1122 if (!ins
->prefixes
[PPS_REP
])
1123 ins
->prefixes
[PPS_REP
] = P_REPNE
;
1127 if (ins
->oprs
[0].segment
!= NO_SEG
)
1128 nasm_error(ERR_NONFATAL
, "attempt to reserve non-constant"
1129 " quantity of BSS space");
1130 else if (ins
->oprs
[0].opflags
& OPFLAG_FORWARD
)
1131 nasm_error(ERR_WARNING
| ERR_PASS1
,
1132 "forward reference in RESx can have unpredictable results");
1134 length
+= ins
->oprs
[0].offset
;
1138 if (!ins
->prefixes
[PPS_WAIT
])
1139 ins
->prefixes
[PPS_WAIT
] = P_WAIT
;
1194 struct operand
*opy
= &ins
->oprs
[op2
];
1195 struct operand
*op_er_sae
;
1197 ea_data
.rex
= 0; /* Ensure ea.REX is initially 0 */
1200 /* pick rfield from operand b (opx) */
1201 rflags
= regflag(opx
);
1202 rfield
= nasm_regvals
[opx
->basereg
];
1208 /* EVEX.b1 : evex_brerop contains the operand position */
1209 op_er_sae
= (ins
->evex_brerop
>= 0 ?
1210 &ins
->oprs
[ins
->evex_brerop
] : NULL
);
1212 if (op_er_sae
&& (op_er_sae
->decoflags
& (ER
| SAE
))) {
1214 ins
->evex_p
[2] |= EVEX_P2B
;
1215 if (op_er_sae
->decoflags
& ER
) {
1216 /* set EVEX.RC (rounding control) */
1217 ins
->evex_p
[2] |= ((ins
->evex_rm
- BRC_RN
) << 5)
1221 /* set EVEX.L'L (vector length) */
1222 ins
->evex_p
[2] |= ((ins
->vex_wlp
<< (5 - 2)) & EVEX_P2LL
);
1223 ins
->evex_p
[1] |= ((ins
->vex_wlp
<< (7 - 4)) & EVEX_P1W
);
1224 if (opy
->decoflags
& BRDCAST_MASK
) {
1226 ins
->evex_p
[2] |= EVEX_P2B
;
1230 if (itemp_has(temp
, IF_MIB
)) {
1231 opy
->eaflags
|= EAF_MIB
;
1233 * if a separate form of MIB (ICC style) is used,
1234 * the index reg info is merged into mem operand
1236 if (mib_index
!= R_none
) {
1237 opy
->indexreg
= mib_index
;
1239 opy
->hintbase
= mib_index
;
1240 opy
->hinttype
= EAH_NOTBASE
;
1244 if (process_ea(opy
, &ea_data
, bits
,
1245 rfield
, rflags
, ins
) != eat
) {
1246 nasm_error(ERR_NONFATAL
, "invalid effective address");
1249 ins
->rex
|= ea_data
.rex
;
1250 length
+= ea_data
.size
;
1256 nasm_panic(0, "internal instruction table corrupt"
1257 ": instruction code \\%o (0x%02X) given", c
, c
);
1262 ins
->rex
&= rex_mask
;
1264 if (ins
->rex
& REX_NH
) {
1265 if (ins
->rex
& REX_H
) {
1266 nasm_error(ERR_NONFATAL
, "instruction cannot use high registers");
1269 ins
->rex
&= ~REX_P
; /* Don't force REX prefix due to high reg */
1272 switch (ins
->prefixes
[PPS_VEX
]) {
1274 if (!(ins
->rex
& REX_EV
))
1279 if (!(ins
->rex
& REX_V
))
1286 if (ins
->rex
& (REX_V
| REX_EV
)) {
1287 int bad32
= REX_R
|REX_W
|REX_X
|REX_B
;
1289 if (ins
->rex
& REX_H
) {
1290 nasm_error(ERR_NONFATAL
, "cannot use high register in AVX instruction");
1293 switch (ins
->vex_wlp
& 060) {
1307 if (bits
!= 64 && ((ins
->rex
& bad32
) || ins
->vexreg
> 7)) {
1308 nasm_error(ERR_NONFATAL
, "invalid operands in non-64-bit mode");
1310 } else if (!(ins
->rex
& REX_EV
) &&
1311 ((ins
->vexreg
> 15) || (ins
->evex_p
[0] & 0xf0))) {
1312 nasm_error(ERR_NONFATAL
, "invalid high-16 register in non-AVX-512");
1315 if (ins
->rex
& REX_EV
)
1317 else if (ins
->vex_cm
!= 1 || (ins
->rex
& (REX_W
|REX_X
|REX_B
)) ||
1318 ins
->prefixes
[PPS_VEX
] == P_VEX3
)
1322 } else if (ins
->rex
& REX_MASK
) {
1323 if (ins
->rex
& REX_H
) {
1324 nasm_error(ERR_NONFATAL
, "cannot use high register in rex instruction");
1326 } else if (bits
== 64) {
1328 } else if ((ins
->rex
& REX_L
) &&
1329 !(ins
->rex
& (REX_P
|REX_W
|REX_X
|REX_B
)) &&
1330 iflag_ffs(&cpu
) >= IF_X86_64
) {
1332 assert_no_prefix(ins
, PPS_LOCK
);
1333 lockcheck
= false; /* Already errored, no need for warning */
1336 nasm_error(ERR_NONFATAL
, "invalid operands in non-64-bit mode");
1341 if (has_prefix(ins
, PPS_LOCK
, P_LOCK
) && lockcheck
&&
1342 (!itemp_has(temp
,IF_LOCK
) || !is_class(MEMORY
, ins
->oprs
[0].type
))) {
1343 nasm_error(ERR_WARNING
| ERR_WARN_LOCK
| ERR_PASS2
,
1344 "instruction is not lockable");
1347 bad_hle_warn(ins
, hleok
);
1350 * when BND prefix is set by DEFAULT directive,
1351 * BND prefix is added to every appropriate instruction line
1352 * unless it is overridden by NOBND prefix.
1355 (itemp_has(temp
, IF_BND
) && !has_prefix(ins
, PPS_REP
, P_NOBND
)))
1356 ins
->prefixes
[PPS_REP
] = P_BND
;
1359 * Add length of legacy prefixes
1361 length
+= emit_prefix(NULL
, bits
, ins
);
1366 static inline void emit_rex(struct out_data
*data
, insn
*ins
)
1368 if (data
->bits
== 64) {
1369 if ((ins
->rex
& REX_MASK
) &&
1370 !(ins
->rex
& (REX_V
| REX_EV
)) &&
1372 uint8_t rex
= (ins
->rex
& REX_MASK
) | REX_P
;
1373 out_rawbyte(data
, rex
);
1374 ins
->rex_done
= true;
1379 static int emit_prefix(struct out_data
*data
, const int bits
, insn
*ins
)
1384 for (j
= 0; j
< MAXPREFIX
; j
++) {
1386 switch (ins
->prefixes
[j
]) {
1407 nasm_error(ERR_WARNING
| ERR_PASS2
,
1408 "cs segment base generated, but will be ignored in 64-bit mode");
1414 nasm_error(ERR_WARNING
| ERR_PASS2
,
1415 "ds segment base generated, but will be ignored in 64-bit mode");
1421 nasm_error(ERR_WARNING
| ERR_PASS2
,
1422 "es segment base generated, but will be ignored in 64-bit mode");
1434 nasm_error(ERR_WARNING
| ERR_PASS2
,
1435 "ss segment base generated, but will be ignored in 64-bit mode");
1441 nasm_error(ERR_NONFATAL
,
1442 "segr6 and segr7 cannot be used as prefixes");
1446 nasm_error(ERR_NONFATAL
,
1447 "16-bit addressing is not supported "
1449 } else if (bits
!= 16)
1458 nasm_error(ERR_NONFATAL
,
1459 "64-bit addressing is only supported "
1487 nasm_panic(0, "invalid instruction prefix");
1491 out_rawbyte(data
, c
);
1498 static void gencode(struct out_data
*data
, insn
*ins
)
1504 struct operand
*opx
;
1505 const uint8_t *codes
= data
->itemp
->code
;
1507 enum ea_type eat
= EA_SCALAR
;
1509 const int bits
= data
->bits
;
1511 ins
->rex_done
= false;
1513 emit_prefix(data
, bits
, ins
);
1517 op1
= (c
& 3) + ((opex
& 1) << 2);
1518 op2
= ((c
>> 3) & 3) + ((opex
& 2) << 1);
1519 opx
= &ins
->oprs
[op1
];
1520 opex
= 0; /* For the next iteration */
1528 emit_rex(data
, ins
);
1529 out_rawdata(data
, codes
, c
);
1540 emit_rex(data
, ins
);
1541 out_rawbyte(data
, *codes
++ + (regval(opx
) & 7));
1548 if (opx
->offset
< -256 || opx
->offset
> 255)
1549 nasm_error(ERR_WARNING
| ERR_PASS2
| ERR_WARN_NOV
,
1550 "byte value exceeds bounds");
1551 out_imm(data
, opx
, 1, OUT_WRAP
);
1555 if (opx
->offset
< 0 || opx
->offset
> 255)
1556 nasm_error(ERR_WARNING
| ERR_PASS2
| ERR_WARN_NOV
,
1557 "unsigned byte value exceeds bounds");
1558 out_imm(data
, opx
, 1, OUT_UNSIGNED
);
1562 warn_overflow_opd(opx
, 2);
1563 out_imm(data
, opx
, 2, OUT_WRAP
);
1567 if (opx
->type
& (BITS16
| BITS32
))
1568 size
= (opx
->type
& BITS16
) ? 2 : 4;
1570 size
= (bits
== 16) ? 2 : 4;
1571 warn_overflow_opd(opx
, size
);
1572 out_imm(data
, opx
, size
, OUT_WRAP
);
1576 warn_overflow_opd(opx
, 4);
1577 out_imm(data
, opx
, 4, OUT_WRAP
);
1581 size
= ins
->addr_size
>> 3;
1582 warn_overflow_opd(opx
, size
);
1583 out_imm(data
, opx
, size
, OUT_WRAP
);
1587 if (opx
->segment
== data
->segment
) {
1588 int64_t delta
= opx
->offset
- data
->offset
1589 - (data
->inslen
- data
->insoffs
);
1590 if (delta
> 127 || delta
< -128)
1591 nasm_error(ERR_NONFATAL
, "short jump is out of range");
1593 out_reladdr(data
, opx
, 1);
1597 out_imm(data
, opx
, 8, OUT_WRAP
);
1601 out_reladdr(data
, opx
, 2);
1605 if (opx
->type
& (BITS16
| BITS32
| BITS64
))
1606 size
= (opx
->type
& BITS16
) ? 2 : 4;
1608 size
= (bits
== 16) ? 2 : 4;
1610 out_reladdr(data
, opx
, size
);
1614 out_reladdr(data
, opx
, 4);
1618 if (opx
->segment
== NO_SEG
)
1619 nasm_error(ERR_NONFATAL
, "value referenced by FAR is not"
1621 out_segment(data
, opx
);
1626 int mask
= ins
->prefixes
[PPS_VEX
] == P_EVEX
? 7 : 15;
1627 const struct operand
*opy
;
1630 opx
= &ins
->oprs
[c
>> 3];
1631 opy
= &ins
->oprs
[c
& 7];
1632 if (opy
->segment
!= NO_SEG
|| opy
->wrt
!= NO_SEG
) {
1633 nasm_error(ERR_NONFATAL
,
1634 "non-absolute expression not permitted as argument %d",
1636 } else if (opy
->offset
& ~mask
) {
1637 nasm_error(ERR_WARNING
| ERR_PASS2
| ERR_WARN_NOV
,
1638 "is4 argument exceeds bounds");
1640 c
= opy
->offset
& mask
;
1646 opx
= &ins
->oprs
[c
>> 4];
1653 r
= nasm_regvals
[opx
->basereg
];
1654 out_rawbyte(data
, (r
<< 4) | ((r
& 0x10) >> 1) | c
);
1658 if (opx
->wrt
== NO_SEG
&& opx
->segment
== NO_SEG
&&
1659 (int32_t)opx
->offset
!= (int64_t)opx
->offset
) {
1660 nasm_error(ERR_WARNING
| ERR_PASS2
| ERR_WARN_NOV
,
1661 "signed dword immediate exceeds bounds");
1663 out_imm(data
, opx
, 4, OUT_SIGNED
);
1669 ins
->evex_p
[2] |= op_evexflags(&ins
->oprs
[0],
1670 EVEX_P2Z
| EVEX_P2AAA
, 2);
1671 ins
->evex_p
[2] ^= EVEX_P2VP
; /* 1's complement */
1673 /* EVEX.X can be set by either REX or EVEX for different reasons */
1674 bytes
[1] = ((((ins
->rex
& 7) << 5) |
1675 (ins
->evex_p
[0] & (EVEX_P0X
| EVEX_P0RP
))) ^ 0xf0) |
1676 (ins
->vex_cm
& EVEX_P0MM
);
1677 bytes
[2] = ((ins
->rex
& REX_W
) << (7 - 3)) |
1678 ((~ins
->vexreg
& 15) << 3) |
1679 (1 << 2) | (ins
->vex_wlp
& 3);
1680 bytes
[3] = ins
->evex_p
[2];
1681 out_rawdata(data
, bytes
, 4);
1687 if (ins
->vex_cm
!= 1 || (ins
->rex
& (REX_W
|REX_X
|REX_B
)) ||
1688 ins
->prefixes
[PPS_VEX
] == P_VEX3
) {
1689 bytes
[0] = (ins
->vex_cm
>> 6) ? 0x8f : 0xc4;
1690 bytes
[1] = (ins
->vex_cm
& 31) | ((~ins
->rex
& 7) << 5);
1691 bytes
[2] = ((ins
->rex
& REX_W
) << (7-3)) |
1692 ((~ins
->vexreg
& 15)<< 3) | (ins
->vex_wlp
& 07);
1693 out_rawdata(data
, bytes
, 3);
1696 bytes
[1] = ((~ins
->rex
& REX_R
) << (7-2)) |
1697 ((~ins
->vexreg
& 15) << 3) | (ins
->vex_wlp
& 07);
1698 out_rawdata(data
, bytes
, 2);
1712 if (ins
->rex
& REX_W
)
1714 else if (ins
->prefixes
[PPS_OSIZE
] == P_O16
)
1716 else if (ins
->prefixes
[PPS_OSIZE
] == P_O32
)
1721 um
= (uint64_t)2 << (s
-1);
1724 if (uv
> 127 && uv
< (uint64_t)-128 &&
1725 (uv
< um
-128 || uv
> um
-1)) {
1726 /* If this wasn't explicitly byte-sized, warn as though we
1727 * had fallen through to the imm16/32/64 case.
1729 nasm_error(ERR_WARNING
| ERR_PASS2
| ERR_WARN_NOV
,
1730 "%s value exceeds bounds",
1731 (opx
->type
& BITS8
) ? "signed byte" :
1736 out_imm(data
, opx
, 1, OUT_WRAP
); /* XXX: OUT_SIGNED? */
1744 if (bits
== 32 && !has_prefix(ins
, PPS_ASIZE
, P_A16
))
1745 out_rawbyte(data
, 0x67);
1749 if (bits
!= 32 && !has_prefix(ins
, PPS_ASIZE
, P_A32
))
1750 out_rawbyte(data
, 0x67);
1782 out_rawbyte(data
, *codes
++ ^ get_cond_opcode(ins
->condition
));
1790 out_rawbyte(data
, c
- 0332 + 0xF2);
1794 if (ins
->rex
& REX_R
)
1795 out_rawbyte(data
, 0xF0);
1796 ins
->rex
&= ~(REX_L
|REX_R
);
1807 if (ins
->oprs
[0].segment
!= NO_SEG
)
1808 nasm_panic(0, "non-constant BSS size in pass two");
1810 out_reserve(data
, ins
->oprs
[0].offset
);
1820 out_rawbyte(data
, 0x66);
1829 out_rawbyte(data
, c
- 0366 + 0x66);
1836 out_rawbyte(data
, bits
== 16 ? 3 : 5);
1868 struct operand
*opy
= &ins
->oprs
[op2
];
1871 /* pick rfield from operand b (opx) */
1872 rflags
= regflag(opx
);
1873 rfield
= nasm_regvals
[opx
->basereg
];
1875 /* rfield is constant */
1880 if (process_ea(opy
, &ea_data
, bits
,
1881 rfield
, rflags
, ins
) != eat
)
1882 nasm_error(ERR_NONFATAL
, "invalid effective address");
1885 *p
++ = ea_data
.modrm
;
1886 if (ea_data
.sib_present
)
1888 out_rawdata(data
, bytes
, p
- bytes
);
1891 * Make sure the address gets the right offset in case
1892 * the line breaks in the .lst file (BR 1197827)
1895 if (ea_data
.bytes
) {
1896 /* use compressed displacement, if available */
1897 if (ea_data
.disp8
) {
1898 out_rawbyte(data
, ea_data
.disp8
);
1899 } else if (ea_data
.rip
) {
1900 out_reladdr(data
, opy
, ea_data
.bytes
);
1902 int asize
= ins
->addr_size
>> 3;
1904 if (overflow_general(opy
->offset
, asize
) ||
1905 signed_bits(opy
->offset
, ins
->addr_size
) !=
1906 signed_bits(opy
->offset
, ea_data
.bytes
<< 3))
1907 warn_overflow(ERR_PASS2
, ea_data
.bytes
);
1909 out_imm(data
, opy
, ea_data
.bytes
,
1910 (asize
> ea_data
.bytes
) ? OUT_SIGNED
: OUT_UNSIGNED
);
1917 nasm_panic(0, "internal instruction table corrupt"
1918 ": instruction code \\%o (0x%02X) given", c
, c
);
1924 static opflags_t
regflag(const operand
* o
)
1926 if (!is_register(o
->basereg
))
1927 nasm_panic(0, "invalid operand passed to regflag()");
1928 return nasm_reg_flags
[o
->basereg
];
1931 static int32_t regval(const operand
* o
)
1933 if (!is_register(o
->basereg
))
1934 nasm_panic(0, "invalid operand passed to regval()");
1935 return nasm_regvals
[o
->basereg
];
1938 static int op_rexflags(const operand
* o
, int mask
)
1943 if (!is_register(o
->basereg
))
1944 nasm_panic(0, "invalid operand passed to op_rexflags()");
1946 flags
= nasm_reg_flags
[o
->basereg
];
1947 val
= nasm_regvals
[o
->basereg
];
1949 return rexflags(val
, flags
, mask
);
1952 static int rexflags(int val
, opflags_t flags
, int mask
)
1956 if (val
>= 0 && (val
& 8))
1957 rex
|= REX_B
|REX_X
|REX_R
;
1960 if (!(REG_HIGH
& ~flags
)) /* AH, CH, DH, BH */
1962 else if (!(REG8
& ~flags
) && val
>= 4) /* SPL, BPL, SIL, DIL */
1968 static int evexflags(int val
, decoflags_t deco
,
1969 int mask
, uint8_t byte
)
1975 if (val
>= 0 && (val
& 16))
1976 evex
|= (EVEX_P0RP
| EVEX_P0X
);
1979 if (val
>= 0 && (val
& 16))
1983 if (deco
& OPMASK_MASK
)
1984 evex
|= deco
& EVEX_P2AAA
;
1990 static int op_evexflags(const operand
* o
, int mask
, uint8_t byte
)
1994 val
= nasm_regvals
[o
->basereg
];
1996 return evexflags(val
, o
->decoflags
, mask
, byte
);
1999 static enum match_result
find_match(const struct itemplate
**tempp
,
2001 int32_t segment
, int64_t offset
, int bits
)
2003 const struct itemplate
*temp
;
2004 enum match_result m
, merr
;
2005 opflags_t xsizeflags
[MAX_OPERANDS
];
2006 bool opsizemissing
= false;
2007 int8_t broadcast
= instruction
->evex_brerop
;
2010 /* broadcasting uses a different data element size */
2011 for (i
= 0; i
< instruction
->operands
; i
++)
2013 xsizeflags
[i
] = instruction
->oprs
[i
].decoflags
& BRSIZE_MASK
;
2015 xsizeflags
[i
] = instruction
->oprs
[i
].type
& SIZE_MASK
;
2017 merr
= MERR_INVALOP
;
2019 for (temp
= nasm_instructions
[instruction
->opcode
];
2020 temp
->opcode
!= I_none
; temp
++) {
2021 m
= matches(temp
, instruction
, bits
);
2022 if (m
== MOK_JUMP
) {
2023 if (jmp_match(segment
, offset
, bits
, instruction
, temp
))
2027 } else if (m
== MERR_OPSIZEMISSING
&& !itemp_has(temp
, IF_SX
)) {
2029 * Missing operand size and a candidate for fuzzy matching...
2031 for (i
= 0; i
< temp
->operands
; i
++)
2033 xsizeflags
[i
] |= temp
->deco
[i
] & BRSIZE_MASK
;
2035 xsizeflags
[i
] |= temp
->opd
[i
] & SIZE_MASK
;
2036 opsizemissing
= true;
2040 if (merr
== MOK_GOOD
)
2044 /* No match, but see if we can get a fuzzy operand size match... */
2048 for (i
= 0; i
< instruction
->operands
; i
++) {
2050 * We ignore extrinsic operand sizes on registers, so we should
2051 * never try to fuzzy-match on them. This also resolves the case
2052 * when we have e.g. "xmmrm128" in two different positions.
2054 if (is_class(REGISTER
, instruction
->oprs
[i
].type
))
2057 /* This tests if xsizeflags[i] has more than one bit set */
2058 if ((xsizeflags
[i
] & (xsizeflags
[i
]-1)))
2059 goto done
; /* No luck */
2061 if (i
== broadcast
) {
2062 instruction
->oprs
[i
].decoflags
|= xsizeflags
[i
];
2063 instruction
->oprs
[i
].type
|= (xsizeflags
[i
] == BR_BITS32
?
2066 instruction
->oprs
[i
].type
|= xsizeflags
[i
]; /* Set the size */
2070 /* Try matching again... */
2071 for (temp
= nasm_instructions
[instruction
->opcode
];
2072 temp
->opcode
!= I_none
; temp
++) {
2073 m
= matches(temp
, instruction
, bits
);
2074 if (m
== MOK_JUMP
) {
2075 if (jmp_match(segment
, offset
, bits
, instruction
, temp
))
2082 if (merr
== MOK_GOOD
)
2091 static uint8_t get_broadcast_num(opflags_t opflags
, opflags_t brsize
)
2093 opflags_t opsize
= opflags
& SIZE_MASK
;
2097 * Due to discontinuity between BITS64 and BITS128 (BITS80),
2098 * this cannot be a simple arithmetic calculation.
2100 if (brsize
> BITS64
)
2101 nasm_error(ERR_FATAL
,
2102 "size of broadcasting element is greater than 64 bits");
2106 brcast_num
= BITS64
/ brsize
;
2109 brcast_num
= (opsize
/ BITS128
) * (BITS64
/ brsize
) * 2;
2116 static enum match_result
matches(const struct itemplate
*itemp
,
2117 insn
*instruction
, int bits
)
2119 opflags_t size
[MAX_OPERANDS
], asize
;
2120 bool opsizemissing
= false;
2126 if (itemp
->opcode
!= instruction
->opcode
)
2127 return MERR_INVALOP
;
2130 * Count the operands
2132 if (itemp
->operands
!= instruction
->operands
)
2133 return MERR_INVALOP
;
2138 if (!(optimizing
> 0) && itemp_has(itemp
, IF_OPT
))
2139 return MERR_INVALOP
;
2144 switch (instruction
->prefixes
[PPS_VEX
]) {
2146 if (!itemp_has(itemp
, IF_EVEX
))
2147 return MERR_ENCMISMATCH
;
2151 if (!itemp_has(itemp
, IF_VEX
))
2152 return MERR_ENCMISMATCH
;
2159 * Check that no spurious colons or TOs are present
2161 for (i
= 0; i
< itemp
->operands
; i
++)
2162 if (instruction
->oprs
[i
].type
& ~itemp
->opd
[i
] & (COLON
| TO
))
2163 return MERR_INVALOP
;
2166 * Process size flags
2168 switch (itemp_smask(itemp
)) {
2169 case IF_GENBIT(IF_SB
):
2172 case IF_GENBIT(IF_SW
):
2175 case IF_GENBIT(IF_SD
):
2178 case IF_GENBIT(IF_SQ
):
2181 case IF_GENBIT(IF_SO
):
2184 case IF_GENBIT(IF_SY
):
2187 case IF_GENBIT(IF_SZ
):
2190 case IF_GENBIT(IF_SIZE
):
2211 if (itemp_armask(itemp
)) {
2212 /* S- flags only apply to a specific operand */
2213 i
= itemp_arg(itemp
);
2214 memset(size
, 0, sizeof size
);
2217 /* S- flags apply to all operands */
2218 for (i
= 0; i
< MAX_OPERANDS
; i
++)
2223 * Check that the operand flags all match up,
2224 * it's a bit tricky so lets be verbose:
2226 * 1) Find out the size of operand. If instruction
2227 * doesn't have one specified -- we're trying to
2228 * guess it either from template (IF_S* flag) or
2231 * 2) If template operand do not match the instruction OR
2232 * template has an operand size specified AND this size differ
2233 * from which instruction has (perhaps we got it from code bits)
2235 * a) Check that only size of instruction and operand is differ
2236 * other characteristics do match
2237 * b) Perhaps it's a register specified in instruction so
2238 * for such a case we just mark that operand as "size
2239 * missing" and this will turn on fuzzy operand size
2240 * logic facility (handled by a caller)
2242 for (i
= 0; i
< itemp
->operands
; i
++) {
2243 opflags_t type
= instruction
->oprs
[i
].type
;
2244 decoflags_t deco
= instruction
->oprs
[i
].decoflags
;
2245 bool is_broadcast
= deco
& BRDCAST_MASK
;
2246 uint8_t brcast_num
= 0;
2247 opflags_t template_opsize
, insn_opsize
;
2249 if (!(type
& SIZE_MASK
))
2252 insn_opsize
= type
& SIZE_MASK
;
2253 if (!is_broadcast
) {
2254 template_opsize
= itemp
->opd
[i
] & SIZE_MASK
;
2256 decoflags_t deco_brsize
= itemp
->deco
[i
] & BRSIZE_MASK
;
2258 * when broadcasting, the element size depends on
2259 * the instruction type. decorator flag should match.
2263 template_opsize
= (deco_brsize
== BR_BITS32
? BITS32
: BITS64
);
2264 /* calculate the proper number : {1to<brcast_num>} */
2265 brcast_num
= get_broadcast_num(itemp
->opd
[i
], template_opsize
);
2267 template_opsize
= 0;
2271 if ((itemp
->opd
[i
] & ~type
& ~SIZE_MASK
) ||
2272 (deco
& ~itemp
->deco
[i
] & ~BRNUM_MASK
)) {
2273 return MERR_INVALOP
;
2274 } else if (template_opsize
) {
2275 if (template_opsize
!= insn_opsize
) {
2277 return MERR_INVALOP
;
2278 } else if (!is_class(REGISTER
, type
)) {
2280 * Note: we don't honor extrinsic operand sizes for registers,
2281 * so "missing operand size" for a register should be
2282 * considered a wildcard match rather than an error.
2284 opsizemissing
= true;
2286 } else if (is_broadcast
&&
2288 (2U << ((deco
& BRNUM_MASK
) >> BRNUM_SHIFT
)))) {
2290 * broadcasting opsize matches but the number of repeated memory
2291 * element does not match.
2292 * if 64b double precision float is broadcasted to ymm (256b),
2293 * broadcasting decorator must be {1to4}.
2295 return MERR_BRNUMMISMATCH
;
2301 return MERR_OPSIZEMISSING
;
2304 * Check operand sizes
2306 if (itemp_has(itemp
, IF_SM
) || itemp_has(itemp
, IF_SM2
)) {
2307 oprs
= (itemp_has(itemp
, IF_SM2
) ? 2 : itemp
->operands
);
2308 for (i
= 0; i
< oprs
; i
++) {
2309 asize
= itemp
->opd
[i
] & SIZE_MASK
;
2311 for (i
= 0; i
< oprs
; i
++)
2317 oprs
= itemp
->operands
;
2320 for (i
= 0; i
< itemp
->operands
; i
++) {
2321 if (!(itemp
->opd
[i
] & SIZE_MASK
) &&
2322 (instruction
->oprs
[i
].type
& SIZE_MASK
& ~size
[i
]))
2323 return MERR_OPSIZEMISMATCH
;
2327 * Check template is okay at the set cpu level
2329 if (iflag_cmp_cpu_level(&insns_flags
[itemp
->iflag_idx
], &cpu
) > 0)
2333 * Verify the appropriate long mode flag.
2335 if (itemp_has(itemp
, (bits
== 64 ? IF_NOLONG
: IF_LONG
)))
2336 return MERR_BADMODE
;
2339 * If we have a HLE prefix, look for the NOHLE flag
2341 if (itemp_has(itemp
, IF_NOHLE
) &&
2342 (has_prefix(instruction
, PPS_REP
, P_XACQUIRE
) ||
2343 has_prefix(instruction
, PPS_REP
, P_XRELEASE
)))
2347 * Check if special handling needed for Jumps
2349 if ((itemp
->code
[0] & ~1) == 0370)
2353 * Check if BND prefix is allowed.
2354 * Other 0xF2 (REPNE/REPNZ) prefix is prohibited.
2356 if (!itemp_has(itemp
, IF_BND
) &&
2357 (has_prefix(instruction
, PPS_REP
, P_BND
) ||
2358 has_prefix(instruction
, PPS_REP
, P_NOBND
)))
2360 else if (itemp_has(itemp
, IF_BND
) &&
2361 (has_prefix(instruction
, PPS_REP
, P_REPNE
) ||
2362 has_prefix(instruction
, PPS_REP
, P_REPNZ
)))
2363 return MERR_BADREPNE
;
2369 * Check if ModR/M.mod should/can be 01.
2370 * - EAF_BYTEOFFS is set
2371 * - offset can fit in a byte when EVEX is not used
2372 * - offset can be compressed when EVEX is used
2374 #define IS_MOD_01() (input->eaflags & EAF_BYTEOFFS || \
2375 (o >= -128 && o <= 127 && \
2376 seg == NO_SEG && !forw_ref && \
2377 !(input->eaflags & EAF_WORDOFFS) && \
2378 !(ins->rex & REX_EV)) || \
2379 (ins->rex & REX_EV && \
2380 is_disp8n(input, ins, &output->disp8)))
2382 static enum ea_type
process_ea(operand
*input
, ea
*output
, int bits
,
2383 int rfield
, opflags_t rflags
, insn
*ins
)
2385 bool forw_ref
= !!(input
->opflags
& OPFLAG_UNKNOWN
);
2386 int addrbits
= ins
->addr_size
;
2387 int eaflags
= input
->eaflags
;
2389 output
->type
= EA_SCALAR
;
2390 output
->rip
= false;
2393 /* REX flags for the rfield operand */
2394 output
->rex
|= rexflags(rfield
, rflags
, REX_R
| REX_P
| REX_W
| REX_H
);
2395 /* EVEX.R' flag for the REG operand */
2396 ins
->evex_p
[0] |= evexflags(rfield
, 0, EVEX_P0RP
, 0);
2398 if (is_class(REGISTER
, input
->type
)) {
2400 * It's a direct register.
2402 if (!is_register(input
->basereg
))
2405 if (!is_reg_class(REG_EA
, input
->basereg
))
2408 /* broadcasting is not available with a direct register operand. */
2409 if (input
->decoflags
& BRDCAST_MASK
) {
2410 nasm_error(ERR_NONFATAL
, "Broadcasting not allowed from a register");
2414 output
->rex
|= op_rexflags(input
, REX_B
| REX_P
| REX_W
| REX_H
);
2415 ins
->evex_p
[0] |= op_evexflags(input
, EVEX_P0X
, 0);
2416 output
->sib_present
= false; /* no SIB necessary */
2417 output
->bytes
= 0; /* no offset necessary either */
2418 output
->modrm
= GEN_MODRM(3, rfield
, nasm_regvals
[input
->basereg
]);
2421 * It's a memory reference.
2424 /* Embedded rounding or SAE is not available with a mem ref operand. */
2425 if (input
->decoflags
& (ER
| SAE
)) {
2426 nasm_error(ERR_NONFATAL
,
2427 "Embedded rounding is available only with reg-reg op.");
2431 if (input
->basereg
== -1 &&
2432 (input
->indexreg
== -1 || input
->scale
== 0)) {
2434 * It's a pure offset.
2436 if (bits
== 64 && ((input
->type
& IP_REL
) == IP_REL
) &&
2437 input
->segment
== NO_SEG
) {
2438 nasm_error(ERR_WARNING
| ERR_PASS1
, "absolute address can not be RIP-relative");
2439 input
->type
&= ~IP_REL
;
2440 input
->type
|= MEMORY
;
2444 !(IP_REL
& ~input
->type
) && (eaflags
& EAF_MIB
)) {
2445 nasm_error(ERR_NONFATAL
, "RIP-relative addressing is prohibited for mib.");
2449 if (eaflags
& EAF_BYTEOFFS
||
2450 (eaflags
& EAF_WORDOFFS
&&
2451 input
->disp_size
!= (addrbits
!= 16 ? 32 : 16))) {
2452 nasm_error(ERR_WARNING
| ERR_PASS1
, "displacement size ignored on absolute address");
2455 if (bits
== 64 && (~input
->type
& IP_REL
)) {
2456 output
->sib_present
= true;
2457 output
->sib
= GEN_SIB(0, 4, 5);
2459 output
->modrm
= GEN_MODRM(0, rfield
, 4);
2460 output
->rip
= false;
2462 output
->sib_present
= false;
2463 output
->bytes
= (addrbits
!= 16 ? 4 : 2);
2464 output
->modrm
= GEN_MODRM(0, rfield
, (addrbits
!= 16 ? 5 : 6));
2465 output
->rip
= bits
== 64;
2469 * It's an indirection.
2471 int i
= input
->indexreg
, b
= input
->basereg
, s
= input
->scale
;
2472 int32_t seg
= input
->segment
;
2473 int hb
= input
->hintbase
, ht
= input
->hinttype
;
2474 int t
, it
, bt
; /* register numbers */
2475 opflags_t x
, ix
, bx
; /* register flags */
2478 i
= -1; /* make this easy, at least */
2480 if (is_register(i
)) {
2481 it
= nasm_regvals
[i
];
2482 ix
= nasm_reg_flags
[i
];
2488 if (is_register(b
)) {
2489 bt
= nasm_regvals
[b
];
2490 bx
= nasm_reg_flags
[b
];
2496 /* if either one are a vector register... */
2497 if ((ix
|bx
) & (XMMREG
|YMMREG
|ZMMREG
) & ~REG_EA
) {
2498 opflags_t sok
= BITS32
| BITS64
;
2499 int32_t o
= input
->offset
;
2500 int mod
, scale
, index
, base
;
2503 * For a vector SIB, one has to be a vector and the other,
2504 * if present, a GPR. The vector must be the index operand.
2506 if (it
== -1 || (bx
& (XMMREG
|YMMREG
|ZMMREG
) & ~REG_EA
)) {
2512 t
= bt
, bt
= it
, it
= t
;
2513 x
= bx
, bx
= ix
, ix
= x
;
2519 if (!(REG64
& ~bx
) || !(REG32
& ~bx
))
2526 * While we're here, ensure the user didn't specify
2529 if (input
->disp_size
== 16 || input
->disp_size
== 64)
2532 if (addrbits
== 16 ||
2533 (addrbits
== 32 && !(sok
& BITS32
)) ||
2534 (addrbits
== 64 && !(sok
& BITS64
)))
2537 output
->type
= ((ix
& ZMMREG
& ~REG_EA
) ? EA_ZMMVSIB
2538 : ((ix
& YMMREG
& ~REG_EA
)
2539 ? EA_YMMVSIB
: EA_XMMVSIB
));
2541 output
->rex
|= rexflags(it
, ix
, REX_X
);
2542 output
->rex
|= rexflags(bt
, bx
, REX_B
);
2543 ins
->evex_p
[2] |= evexflags(it
, 0, EVEX_P2VP
, 2);
2545 index
= it
& 7; /* it is known to be != -1 */
2560 default: /* then what the smeg is it? */
2561 goto err
; /* panic */
2569 if (base
!= REG_NUM_EBP
&& o
== 0 &&
2570 seg
== NO_SEG
&& !forw_ref
&&
2571 !(eaflags
& (EAF_BYTEOFFS
| EAF_WORDOFFS
)))
2573 else if (IS_MOD_01())
2579 output
->sib_present
= true;
2580 output
->bytes
= (bt
== -1 || mod
== 2 ? 4 : mod
);
2581 output
->modrm
= GEN_MODRM(mod
, rfield
, 4);
2582 output
->sib
= GEN_SIB(scale
, index
, base
);
2583 } else if ((ix
|bx
) & (BITS32
|BITS64
)) {
2585 * it must be a 32/64-bit memory reference. Firstly we have
2586 * to check that all registers involved are type E/Rxx.
2588 opflags_t sok
= BITS32
| BITS64
;
2589 int32_t o
= input
->offset
;
2592 if (!(REG64
& ~ix
) || !(REG32
& ~ix
))
2600 goto err
; /* Invalid register */
2601 if (~sok
& bx
& SIZE_MASK
)
2602 goto err
; /* Invalid size */
2607 * While we're here, ensure the user didn't specify
2610 if (input
->disp_size
== 16 || input
->disp_size
== 64)
2613 if (addrbits
== 16 ||
2614 (addrbits
== 32 && !(sok
& BITS32
)) ||
2615 (addrbits
== 64 && !(sok
& BITS64
)))
2618 /* now reorganize base/index */
2619 if (s
== 1 && bt
!= it
&& bt
!= -1 && it
!= -1 &&
2620 ((hb
== b
&& ht
== EAH_NOTBASE
) ||
2621 (hb
== i
&& ht
== EAH_MAKEBASE
))) {
2622 /* swap if hints say so */
2623 t
= bt
, bt
= it
, it
= t
;
2624 x
= bx
, bx
= ix
, ix
= x
;
2627 if (bt
== -1 && s
== 1 && !(hb
== i
&& ht
== EAH_NOTBASE
)) {
2628 /* make single reg base, unless hint */
2629 bt
= it
, bx
= ix
, it
= -1, ix
= 0;
2631 if (eaflags
& EAF_MIB
) {
2632 /* only for mib operands */
2633 if (it
== -1 && (hb
== b
&& ht
== EAH_NOTBASE
)) {
2635 * make a single reg index [reg*1].
2636 * gas uses this form for an explicit index register.
2638 it
= bt
, ix
= bx
, bt
= -1, bx
= 0, s
= 1;
2640 if ((ht
== EAH_SUMMED
) && bt
== -1) {
2641 /* separate once summed index into [base, index] */
2642 bt
= it
, bx
= ix
, s
--;
2645 if (((s
== 2 && it
!= REG_NUM_ESP
&&
2646 (!(eaflags
& EAF_TIMESTWO
) || (ht
== EAH_SUMMED
))) ||
2647 s
== 3 || s
== 5 || s
== 9) && bt
== -1) {
2648 /* convert 3*EAX to EAX+2*EAX */
2649 bt
= it
, bx
= ix
, s
--;
2651 if (it
== -1 && (bt
& 7) != REG_NUM_ESP
&&
2652 (eaflags
& EAF_TIMESTWO
) &&
2653 (hb
== b
&& ht
== EAH_NOTBASE
)) {
2655 * convert [NOSPLIT EAX*1]
2656 * to sib format with 0x0 displacement - [EAX*1+0].
2658 it
= bt
, ix
= bx
, bt
= -1, bx
= 0, s
= 1;
2661 if (s
== 1 && it
== REG_NUM_ESP
) {
2662 /* swap ESP into base if scale is 1 */
2663 t
= it
, it
= bt
, bt
= t
;
2664 x
= ix
, ix
= bx
, bx
= x
;
2666 if (it
== REG_NUM_ESP
||
2667 (s
!= 1 && s
!= 2 && s
!= 4 && s
!= 8 && it
!= -1))
2668 goto err
; /* wrong, for various reasons */
2670 output
->rex
|= rexflags(it
, ix
, REX_X
);
2671 output
->rex
|= rexflags(bt
, bx
, REX_B
);
2673 if (it
== -1 && (bt
& 7) != REG_NUM_ESP
) {
2682 if (rm
!= REG_NUM_EBP
&& o
== 0 &&
2683 seg
== NO_SEG
&& !forw_ref
&&
2684 !(eaflags
& (EAF_BYTEOFFS
| EAF_WORDOFFS
)))
2686 else if (IS_MOD_01())
2692 output
->sib_present
= false;
2693 output
->bytes
= (bt
== -1 || mod
== 2 ? 4 : mod
);
2694 output
->modrm
= GEN_MODRM(mod
, rfield
, rm
);
2697 int mod
, scale
, index
, base
;
2717 default: /* then what the smeg is it? */
2718 goto err
; /* panic */
2726 if (base
!= REG_NUM_EBP
&& o
== 0 &&
2727 seg
== NO_SEG
&& !forw_ref
&&
2728 !(eaflags
& (EAF_BYTEOFFS
| EAF_WORDOFFS
)))
2730 else if (IS_MOD_01())
2736 output
->sib_present
= true;
2737 output
->bytes
= (bt
== -1 || mod
== 2 ? 4 : mod
);
2738 output
->modrm
= GEN_MODRM(mod
, rfield
, 4);
2739 output
->sib
= GEN_SIB(scale
, index
, base
);
2741 } else { /* it's 16-bit */
2743 int16_t o
= input
->offset
;
2745 /* check for 64-bit long mode */
2749 /* check all registers are BX, BP, SI or DI */
2750 if ((b
!= -1 && b
!= R_BP
&& b
!= R_BX
&& b
!= R_SI
&& b
!= R_DI
) ||
2751 (i
!= -1 && i
!= R_BP
&& i
!= R_BX
&& i
!= R_SI
&& i
!= R_DI
))
2754 /* ensure the user didn't specify DWORD/QWORD */
2755 if (input
->disp_size
== 32 || input
->disp_size
== 64)
2758 if (s
!= 1 && i
!= -1)
2759 goto err
; /* no can do, in 16-bit EA */
2760 if (b
== -1 && i
!= -1) {
2765 if ((b
== R_SI
|| b
== R_DI
) && i
!= -1) {
2770 /* have BX/BP as base, SI/DI index */
2772 goto err
; /* shouldn't ever happen, in theory */
2773 if (i
!= -1 && b
!= -1 &&
2774 (i
== R_BP
|| i
== R_BX
|| b
== R_SI
|| b
== R_DI
))
2775 goto err
; /* invalid combinations */
2776 if (b
== -1) /* pure offset: handled above */
2777 goto err
; /* so if it gets to here, panic! */
2781 switch (i
* 256 + b
) {
2782 case R_SI
* 256 + R_BX
:
2785 case R_DI
* 256 + R_BX
:
2788 case R_SI
* 256 + R_BP
:
2791 case R_DI
* 256 + R_BP
:
2809 if (rm
== -1) /* can't happen, in theory */
2810 goto err
; /* so panic if it does */
2812 if (o
== 0 && seg
== NO_SEG
&& !forw_ref
&& rm
!= 6 &&
2813 !(eaflags
& (EAF_BYTEOFFS
| EAF_WORDOFFS
)))
2815 else if (IS_MOD_01())
2820 output
->sib_present
= false; /* no SIB - it's 16-bit */
2821 output
->bytes
= mod
; /* bytes of offset needed */
2822 output
->modrm
= GEN_MODRM(mod
, rfield
, rm
);
2827 output
->size
= 1 + output
->sib_present
+ output
->bytes
;
2828 return output
->type
;
2831 return output
->type
= EA_INVALID
;
2834 static void add_asp(insn
*ins
, int addrbits
)
2839 valid
= (addrbits
== 64) ? 64|32 : 32|16;
2841 switch (ins
->prefixes
[PPS_ASIZE
]) {
2852 valid
&= (addrbits
== 32) ? 16 : 32;
2858 for (j
= 0; j
< ins
->operands
; j
++) {
2859 if (is_class(MEMORY
, ins
->oprs
[j
].type
)) {
2862 /* Verify as Register */
2863 if (!is_register(ins
->oprs
[j
].indexreg
))
2866 i
= nasm_reg_flags
[ins
->oprs
[j
].indexreg
];
2868 /* Verify as Register */
2869 if (!is_register(ins
->oprs
[j
].basereg
))
2872 b
= nasm_reg_flags
[ins
->oprs
[j
].basereg
];
2874 if (ins
->oprs
[j
].scale
== 0)
2878 int ds
= ins
->oprs
[j
].disp_size
;
2879 if ((addrbits
!= 64 && ds
> 8) ||
2880 (addrbits
== 64 && ds
== 16))
2900 if (valid
& addrbits
) {
2901 ins
->addr_size
= addrbits
;
2902 } else if (valid
& ((addrbits
== 32) ? 16 : 32)) {
2903 /* Add an address size prefix */
2904 ins
->prefixes
[PPS_ASIZE
] = (addrbits
== 32) ? P_A16
: P_A32
;;
2905 ins
->addr_size
= (addrbits
== 32) ? 16 : 32;
2908 nasm_error(ERR_NONFATAL
, "impossible combination of address sizes");
2909 ins
->addr_size
= addrbits
; /* Error recovery */
2912 defdisp
= ins
->addr_size
== 16 ? 16 : 32;
2914 for (j
= 0; j
< ins
->operands
; j
++) {
2915 if (!(MEM_OFFS
& ~ins
->oprs
[j
].type
) &&
2916 (ins
->oprs
[j
].disp_size
? ins
->oprs
[j
].disp_size
: defdisp
) != ins
->addr_size
) {
2918 * mem_offs sizes must match the address size; if not,
2919 * strip the MEM_OFFS bit and match only EA instructions
2921 ins
->oprs
[j
].type
&= ~(MEM_OFFS
& ~MEMORY
);