NASM 2.11rc4
[nasm.git] / assemble.c
blobbdf9a10c7c443c75bf1a23e45134766b0ef04afd
1 /* ----------------------------------------------------------------------- *
3 * Copyright 1996-2013 The NASM Authors - All Rights Reserved
4 * See the file AUTHORS included with the NASM distribution for
5 * the specific copyright holders.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following
9 * conditions are met:
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 * the actual codes (C syntax, i.e. octal):
38 * \0 - terminates the code. (Unless it's a literal of course.)
39 * \1..\4 - that many literal bytes follow in the code stream
40 * \5 - add 4 to the primary operand number (b, low octdigit)
41 * \6 - add 4 to the secondary operand number (a, middle octdigit)
42 * \7 - add 4 to both the primary and the secondary operand number
43 * \10..\13 - a literal byte follows in the code stream, to be added
44 * to the register value of operand 0..3
45 * \14..\17 - the position of index register operand in MIB (BND insns)
46 * \20..\23 - a byte immediate operand, from operand 0..3
47 * \24..\27 - a zero-extended byte immediate operand, from operand 0..3
48 * \30..\33 - a word immediate operand, from operand 0..3
49 * \34..\37 - select between \3[0-3] and \4[0-3] depending on 16/32 bit
50 * assembly mode or the operand-size override on the operand
51 * \40..\43 - a long immediate operand, from operand 0..3
52 * \44..\47 - select between \3[0-3], \4[0-3] and \5[4-7]
53 * depending on the address size of the instruction.
54 * \50..\53 - a byte relative operand, from operand 0..3
55 * \54..\57 - a qword immediate operand, from operand 0..3
56 * \60..\63 - a word relative operand, from operand 0..3
57 * \64..\67 - select between \6[0-3] and \7[0-3] depending on 16/32 bit
58 * assembly mode or the operand-size override on the operand
59 * \70..\73 - a long relative operand, from operand 0..3
60 * \74..\77 - a word constant, from the _segment_ part of operand 0..3
61 * \1ab - a ModRM, calculated on EA in operand a, with the spare
62 * field the register value of operand b.
63 * \172\ab - the register number from operand a in bits 7..4, with
64 * the 4-bit immediate from operand b in bits 3..0.
65 * \173\xab - the register number from operand a in bits 7..4, with
66 * the value b in bits 3..0.
67 * \174..\177 - the register number from operand 0..3 in bits 7..4, and
68 * an arbitrary value in bits 3..0 (assembled as zero.)
69 * \2ab - a ModRM, calculated on EA in operand a, with the spare
70 * field equal to digit b.
72 * \240..\243 - this instruction uses EVEX rather than REX or VEX/XOP, with the
73 * V field taken from operand 0..3.
74 * \250 - this instruction uses EVEX rather than REX or VEX/XOP, with the
75 * V field set to 1111b.
76 * EVEX prefixes are followed by the sequence:
77 * \cm\wlp\tup where cm is:
78 * cc 000 0mm
79 * c = 2 for EVEX and m is the legacy escape (0f, 0f38, 0f3a)
80 * and wlp is:
81 * 00 wwl lpp
82 * [l0] ll = 0 (.128, .lz)
83 * [l1] ll = 1 (.256)
84 * [l2] ll = 2 (.512)
85 * [lig] ll = 3 for EVEX.L'L don't care (always assembled as 0)
87 * [w0] ww = 0 for W = 0
88 * [w1] ww = 1 for W = 1
89 * [wig] ww = 2 for W don't care (always assembled as 0)
90 * [ww] ww = 3 for W used as REX.W
92 * [p0] pp = 0 for no prefix
93 * [60] pp = 1 for legacy prefix 60
94 * [f3] pp = 2
95 * [f2] pp = 3
97 * tup is tuple type for Disp8*N from %tuple_codes in insns.pl
98 * (compressed displacement encoding)
100 * \254..\257 - a signed 32-bit operand to be extended to 64 bits.
101 * \260..\263 - this instruction uses VEX/XOP rather than REX, with the
102 * V field taken from operand 0..3.
103 * \270 - this instruction uses VEX/XOP rather than REX, with the
104 * V field set to 1111b.
106 * VEX/XOP prefixes are followed by the sequence:
107 * \tmm\wlp where mm is the M field; and wlp is:
108 * 00 wwl lpp
109 * [l0] ll = 0 for L = 0 (.128, .lz)
110 * [l1] ll = 1 for L = 1 (.256)
111 * [lig] ll = 2 for L don't care (always assembled as 0)
113 * [w0] ww = 0 for W = 0
114 * [w1 ] ww = 1 for W = 1
115 * [wig] ww = 2 for W don't care (always assembled as 0)
116 * [ww] ww = 3 for W used as REX.W
118 * t = 0 for VEX (C4/C5), t = 1 for XOP (8F).
120 * \271 - instruction takes XRELEASE (F3) with or without lock
121 * \272 - instruction takes XACQUIRE/XRELEASE with or without lock
122 * \273 - instruction takes XACQUIRE/XRELEASE with lock only
123 * \274..\277 - a byte immediate operand, from operand 0..3, sign-extended
124 * to the operand size (if o16/o32/o64 present) or the bit size
125 * \310 - indicates fixed 16-bit address size, i.e. optional 0x67.
126 * \311 - indicates fixed 32-bit address size, i.e. optional 0x67.
127 * \312 - (disassembler only) invalid with non-default address size.
128 * \313 - indicates fixed 64-bit address size, 0x67 invalid.
129 * \314 - (disassembler only) invalid with REX.B
130 * \315 - (disassembler only) invalid with REX.X
131 * \316 - (disassembler only) invalid with REX.R
132 * \317 - (disassembler only) invalid with REX.W
133 * \320 - indicates fixed 16-bit operand size, i.e. optional 0x66.
134 * \321 - indicates fixed 32-bit operand size, i.e. optional 0x66.
135 * \322 - indicates that this instruction is only valid when the
136 * operand size is the default (instruction to disassembler,
137 * generates no code in the assembler)
138 * \323 - indicates fixed 64-bit operand size, REX on extensions only.
139 * \324 - indicates 64-bit operand size requiring REX prefix.
140 * \325 - instruction which always uses spl/bpl/sil/dil
141 * \326 - instruction not valid with 0xF3 REP prefix. Hint for
142 disassembler only; for SSE instructions.
143 * \330 - a literal byte follows in the code stream, to be added
144 * to the condition code value of the instruction.
145 * \331 - instruction not valid with REP prefix. Hint for
146 * disassembler only; for SSE instructions.
147 * \332 - REP prefix (0xF2 byte) used as opcode extension.
148 * \333 - REP prefix (0xF3 byte) used as opcode extension.
149 * \334 - LOCK prefix used as REX.R (used in non-64-bit mode)
150 * \335 - disassemble a rep (0xF3 byte) prefix as repe not rep.
151 * \336 - force a REP(E) prefix (0xF3) even if not specified.
152 * \337 - force a REPNE prefix (0xF2) even if not specified.
153 * \336-\337 are still listed as prefixes in the disassembler.
154 * \340 - reserve <operand 0> bytes of uninitialized storage.
155 * Operand 0 had better be a segmentless constant.
156 * \341 - this instruction needs a WAIT "prefix"
157 * \360 - no SSE prefix (== \364\331)
158 * \361 - 66 SSE prefix (== \366\331)
159 * \364 - operand-size prefix (0x66) not permitted
160 * \365 - address-size prefix (0x67) not permitted
161 * \366 - operand-size prefix (0x66) used as opcode extension
162 * \367 - address-size prefix (0x67) used as opcode extension
163 * \370,\371 - match only if operand 0 meets byte jump criteria.
164 * 370 is used for Jcc, 371 is used for JMP.
165 * \373 - assemble 0x03 if bits==16, 0x05 if bits==32;
166 * used for conditional jump over longer jump
167 * \374 - this instruction takes an XMM VSIB memory EA
168 * \375 - this instruction takes an YMM VSIB memory EA
169 * \376 - this instruction takes an ZMM VSIB memory EA
172 #include "compiler.h"
174 #include <stdio.h>
175 #include <string.h>
176 #include <inttypes.h>
178 #include "nasm.h"
179 #include "nasmlib.h"
180 #include "assemble.h"
181 #include "insns.h"
182 #include "tables.h"
183 #include "disp8.h"
185 enum match_result {
187 * Matching errors. These should be sorted so that more specific
188 * errors come later in the sequence.
190 MERR_INVALOP,
191 MERR_OPSIZEMISSING,
192 MERR_OPSIZEMISMATCH,
193 MERR_BRNUMMISMATCH,
194 MERR_BADCPU,
195 MERR_BADMODE,
196 MERR_BADHLE,
197 MERR_ENCMISMATCH,
198 MERR_BADBND,
199 MERR_BADREPNE,
201 * Matching success; the conditional ones first
203 MOK_JUMP, /* Matching OK but needs jmp_match() */
204 MOK_GOOD /* Matching unconditionally OK */
207 typedef struct {
208 enum ea_type type; /* what kind of EA is this? */
209 int sib_present; /* is a SIB byte necessary? */
210 int bytes; /* # of bytes of offset needed */
211 int size; /* lazy - this is sib+bytes+1 */
212 uint8_t modrm, sib, rex, rip; /* the bytes themselves */
213 int8_t disp8; /* compressed displacement for EVEX */
214 } ea;
216 #define GEN_SIB(scale, index, base) \
217 (((scale) << 6) | ((index) << 3) | ((base)))
219 #define GEN_MODRM(mod, reg, rm) \
220 (((mod) << 6) | (((reg) & 7) << 3) | ((rm) & 7))
222 static iflag_t cpu; /* cpu level received from nasm.c */
223 static efunc errfunc;
224 static struct ofmt *outfmt;
225 static ListGen *list;
227 static int64_t calcsize(int32_t, int64_t, int, insn *,
228 const struct itemplate *);
229 static void gencode(int32_t segment, int64_t offset, int bits,
230 insn * ins, const struct itemplate *temp,
231 int64_t insn_end);
232 static enum match_result find_match(const struct itemplate **tempp,
233 insn *instruction,
234 int32_t segment, int64_t offset, int bits);
235 static enum match_result matches(const struct itemplate *, insn *, int bits);
236 static opflags_t regflag(const operand *);
237 static int32_t regval(const operand *);
238 static int rexflags(int, opflags_t, int);
239 static int op_rexflags(const operand *, int);
240 static int op_evexflags(const operand *, int, uint8_t);
241 static void add_asp(insn *, int);
243 static enum ea_type process_ea(operand *, ea *, int, int, opflags_t, insn *);
245 static int has_prefix(insn * ins, enum prefix_pos pos, int prefix)
247 return ins->prefixes[pos] == prefix;
250 static void assert_no_prefix(insn * ins, enum prefix_pos pos)
252 if (ins->prefixes[pos])
253 errfunc(ERR_NONFATAL, "invalid %s prefix",
254 prefix_name(ins->prefixes[pos]));
257 static const char *size_name(int size)
259 switch (size) {
260 case 1:
261 return "byte";
262 case 2:
263 return "word";
264 case 4:
265 return "dword";
266 case 8:
267 return "qword";
268 case 10:
269 return "tword";
270 case 16:
271 return "oword";
272 case 32:
273 return "yword";
274 case 64:
275 return "zword";
276 default:
277 return "???";
281 static void warn_overflow(int pass, int size)
283 errfunc(ERR_WARNING | pass | ERR_WARN_NOV,
284 "%s data exceeds bounds", size_name(size));
287 static void warn_overflow_const(int64_t data, int size)
289 if (overflow_general(data, size))
290 warn_overflow(ERR_PASS1, size);
293 static void warn_overflow_opd(const struct operand *o, int size)
295 if (o->wrt == NO_SEG && o->segment == NO_SEG) {
296 if (overflow_general(o->offset, size))
297 warn_overflow(ERR_PASS2, size);
302 * This routine wrappers the real output format's output routine,
303 * in order to pass a copy of the data off to the listing file
304 * generator at the same time.
306 static void out(int64_t offset, int32_t segto, const void *data,
307 enum out_type type, uint64_t size,
308 int32_t segment, int32_t wrt)
310 static int32_t lineno = 0; /* static!!! */
311 static char *lnfname = NULL;
312 uint8_t p[8];
314 if (type == OUT_ADDRESS && segment == NO_SEG && wrt == NO_SEG) {
316 * This is a non-relocated address, and we're going to
317 * convert it into RAWDATA format.
319 uint8_t *q = p;
321 if (size > 8) {
322 errfunc(ERR_PANIC, "OUT_ADDRESS with size > 8");
323 return;
326 WRITEADDR(q, *(int64_t *)data, size);
327 data = p;
328 type = OUT_RAWDATA;
331 list->output(offset, data, type, size);
334 * this call to src_get determines when we call the
335 * debug-format-specific "linenum" function
336 * it updates lineno and lnfname to the current values
337 * returning 0 if "same as last time", -2 if lnfname
338 * changed, and the amount by which lineno changed,
339 * if it did. thus, these variables must be static
342 if (src_get(&lineno, &lnfname))
343 outfmt->current_dfmt->linenum(lnfname, lineno, segto);
345 outfmt->output(segto, data, type, size, segment, wrt);
348 static void out_imm8(int64_t offset, int32_t segment, struct operand *opx)
350 if (opx->segment != NO_SEG) {
351 uint64_t data = opx->offset;
352 out(offset, segment, &data, OUT_ADDRESS, 1, opx->segment, opx->wrt);
353 } else {
354 uint8_t byte = opx->offset;
355 out(offset, segment, &byte, OUT_RAWDATA, 1, NO_SEG, NO_SEG);
359 static bool jmp_match(int32_t segment, int64_t offset, int bits,
360 insn * ins, const struct itemplate *temp)
362 int64_t isize;
363 const uint8_t *code = temp->code;
364 uint8_t c = code[0];
365 bool is_byte;
367 if (((c & ~1) != 0370) || (ins->oprs[0].type & STRICT))
368 return false;
369 if (!optimizing)
370 return false;
371 if (optimizing < 0 && c == 0371)
372 return false;
374 isize = calcsize(segment, offset, bits, ins, temp);
376 if (ins->oprs[0].opflags & OPFLAG_UNKNOWN)
377 /* Be optimistic in pass 1 */
378 return true;
380 if (ins->oprs[0].segment != segment)
381 return false;
383 isize = ins->oprs[0].offset - offset - isize; /* isize is delta */
384 is_byte = (isize >= -128 && isize <= 127); /* is it byte size? */
386 if (is_byte && c == 0371 && ins->prefixes[PPS_REP] == P_BND) {
387 /* jmp short (opcode eb) cannot be used with bnd prefix. */
388 ins->prefixes[PPS_REP] = P_none;
389 errfunc(ERR_WARNING | ERR_WARN_BND | ERR_PASS2 ,
390 "jmp short does not init bnd regs - bnd prefix dropped.");
393 return is_byte;
396 int64_t assemble(int32_t segment, int64_t offset, int bits, iflag_t cp,
397 insn * instruction, struct ofmt *output, efunc error,
398 ListGen * listgen)
400 const struct itemplate *temp;
401 int j;
402 enum match_result m;
403 int64_t insn_end;
404 int32_t itimes;
405 int64_t start = offset;
406 int64_t wsize; /* size for DB etc. */
408 errfunc = error; /* to pass to other functions */
409 cpu = cp;
410 outfmt = output; /* likewise */
411 list = listgen; /* and again */
413 wsize = idata_bytes(instruction->opcode);
414 if (wsize == -1)
415 return 0;
417 if (wsize) {
418 extop *e;
419 int32_t t = instruction->times;
420 if (t < 0)
421 errfunc(ERR_PANIC,
422 "instruction->times < 0 (%ld) in assemble()", t);
424 while (t--) { /* repeat TIMES times */
425 list_for_each(e, instruction->eops) {
426 if (e->type == EOT_DB_NUMBER) {
427 if (wsize > 8) {
428 errfunc(ERR_NONFATAL,
429 "integer supplied to a DT, DO or DY"
430 " instruction");
431 } else {
432 out(offset, segment, &e->offset,
433 OUT_ADDRESS, wsize, e->segment, e->wrt);
434 offset += wsize;
436 } else if (e->type == EOT_DB_STRING ||
437 e->type == EOT_DB_STRING_FREE) {
438 int align;
440 out(offset, segment, e->stringval,
441 OUT_RAWDATA, e->stringlen, NO_SEG, NO_SEG);
442 align = e->stringlen % wsize;
444 if (align) {
445 align = wsize - align;
446 out(offset, segment, zero_buffer,
447 OUT_RAWDATA, align, NO_SEG, NO_SEG);
449 offset += e->stringlen + align;
452 if (t > 0 && t == instruction->times - 1) {
454 * Dummy call to list->output to give the offset to the
455 * listing module.
457 list->output(offset, NULL, OUT_RAWDATA, 0);
458 list->uplevel(LIST_TIMES);
461 if (instruction->times > 1)
462 list->downlevel(LIST_TIMES);
463 return offset - start;
466 if (instruction->opcode == I_INCBIN) {
467 const char *fname = instruction->eops->stringval;
468 FILE *fp;
470 fp = fopen(fname, "rb");
471 if (!fp) {
472 error(ERR_NONFATAL, "`incbin': unable to open file `%s'",
473 fname);
474 } else if (fseek(fp, 0L, SEEK_END) < 0) {
475 error(ERR_NONFATAL, "`incbin': unable to seek on file `%s'",
476 fname);
477 fclose(fp);
478 } else {
479 static char buf[4096];
480 size_t t = instruction->times;
481 size_t base = 0;
482 size_t len;
484 len = ftell(fp);
485 if (instruction->eops->next) {
486 base = instruction->eops->next->offset;
487 len -= base;
488 if (instruction->eops->next->next &&
489 len > (size_t)instruction->eops->next->next->offset)
490 len = (size_t)instruction->eops->next->next->offset;
493 * Dummy call to list->output to give the offset to the
494 * listing module.
496 list->output(offset, NULL, OUT_RAWDATA, 0);
497 list->uplevel(LIST_INCBIN);
498 while (t--) {
499 size_t l;
501 fseek(fp, base, SEEK_SET);
502 l = len;
503 while (l > 0) {
504 int32_t m;
505 m = fread(buf, 1, l > sizeof(buf) ? sizeof(buf) : l, fp);
506 if (!m) {
508 * This shouldn't happen unless the file
509 * actually changes while we are reading
510 * it.
512 error(ERR_NONFATAL,
513 "`incbin': unexpected EOF while"
514 " reading file `%s'", fname);
515 t = 0; /* Try to exit cleanly */
516 break;
518 out(offset, segment, buf, OUT_RAWDATA, m,
519 NO_SEG, NO_SEG);
520 l -= m;
523 list->downlevel(LIST_INCBIN);
524 if (instruction->times > 1) {
526 * Dummy call to list->output to give the offset to the
527 * listing module.
529 list->output(offset, NULL, OUT_RAWDATA, 0);
530 list->uplevel(LIST_TIMES);
531 list->downlevel(LIST_TIMES);
533 fclose(fp);
534 return instruction->times * len;
536 return 0; /* if we're here, there's an error */
539 /* Check to see if we need an address-size prefix */
540 add_asp(instruction, bits);
542 m = find_match(&temp, instruction, segment, offset, bits);
544 if (m == MOK_GOOD) {
545 /* Matches! */
546 int64_t insn_size = calcsize(segment, offset, bits, instruction, temp);
547 itimes = instruction->times;
548 if (insn_size < 0) /* shouldn't be, on pass two */
549 error(ERR_PANIC, "errors made it through from pass one");
550 else
551 while (itimes--) {
552 for (j = 0; j < MAXPREFIX; j++) {
553 uint8_t c = 0;
554 switch (instruction->prefixes[j]) {
555 case P_WAIT:
556 c = 0x9B;
557 break;
558 case P_LOCK:
559 c = 0xF0;
560 break;
561 case P_REPNE:
562 case P_REPNZ:
563 case P_XACQUIRE:
564 case P_BND:
565 c = 0xF2;
566 break;
567 case P_REPE:
568 case P_REPZ:
569 case P_REP:
570 case P_XRELEASE:
571 c = 0xF3;
572 break;
573 case R_CS:
574 if (bits == 64) {
575 error(ERR_WARNING | ERR_PASS2,
576 "cs segment base generated, but will be ignored in 64-bit mode");
578 c = 0x2E;
579 break;
580 case R_DS:
581 if (bits == 64) {
582 error(ERR_WARNING | ERR_PASS2,
583 "ds segment base generated, but will be ignored in 64-bit mode");
585 c = 0x3E;
586 break;
587 case R_ES:
588 if (bits == 64) {
589 error(ERR_WARNING | ERR_PASS2,
590 "es segment base generated, but will be ignored in 64-bit mode");
592 c = 0x26;
593 break;
594 case R_FS:
595 c = 0x64;
596 break;
597 case R_GS:
598 c = 0x65;
599 break;
600 case R_SS:
601 if (bits == 64) {
602 error(ERR_WARNING | ERR_PASS2,
603 "ss segment base generated, but will be ignored in 64-bit mode");
605 c = 0x36;
606 break;
607 case R_SEGR6:
608 case R_SEGR7:
609 error(ERR_NONFATAL,
610 "segr6 and segr7 cannot be used as prefixes");
611 break;
612 case P_A16:
613 if (bits == 64) {
614 error(ERR_NONFATAL,
615 "16-bit addressing is not supported "
616 "in 64-bit mode");
617 } else if (bits != 16)
618 c = 0x67;
619 break;
620 case P_A32:
621 if (bits != 32)
622 c = 0x67;
623 break;
624 case P_A64:
625 if (bits != 64) {
626 error(ERR_NONFATAL,
627 "64-bit addressing is only supported "
628 "in 64-bit mode");
630 break;
631 case P_ASP:
632 c = 0x67;
633 break;
634 case P_O16:
635 if (bits != 16)
636 c = 0x66;
637 break;
638 case P_O32:
639 if (bits == 16)
640 c = 0x66;
641 break;
642 case P_O64:
643 /* REX.W */
644 break;
645 case P_OSP:
646 c = 0x66;
647 break;
648 case P_EVEX:
649 case P_VEX3:
650 case P_VEX2:
651 case P_NOBND:
652 case P_none:
653 break;
654 default:
655 error(ERR_PANIC, "invalid instruction prefix");
657 if (c != 0) {
658 out(offset, segment, &c, OUT_RAWDATA, 1,
659 NO_SEG, NO_SEG);
660 offset++;
663 insn_end = offset + insn_size;
664 gencode(segment, offset, bits, instruction,
665 temp, insn_end);
666 offset += insn_size;
667 if (itimes > 0 && itimes == instruction->times - 1) {
669 * Dummy call to list->output to give the offset to the
670 * listing module.
672 list->output(offset, NULL, OUT_RAWDATA, 0);
673 list->uplevel(LIST_TIMES);
676 if (instruction->times > 1)
677 list->downlevel(LIST_TIMES);
678 return offset - start;
679 } else {
680 /* No match */
681 switch (m) {
682 case MERR_OPSIZEMISSING:
683 error(ERR_NONFATAL, "operation size not specified");
684 break;
685 case MERR_OPSIZEMISMATCH:
686 error(ERR_NONFATAL, "mismatch in operand sizes");
687 break;
688 case MERR_BRNUMMISMATCH:
689 error(ERR_NONFATAL,
690 "mismatch in the number of broadcasting elements");
691 break;
692 case MERR_BADCPU:
693 error(ERR_NONFATAL, "no instruction for this cpu level");
694 break;
695 case MERR_BADMODE:
696 error(ERR_NONFATAL, "instruction not supported in %d-bit mode",
697 bits);
698 break;
699 case MERR_ENCMISMATCH:
700 error(ERR_NONFATAL, "specific encoding scheme not available");
701 break;
702 case MERR_BADBND:
703 error(ERR_NONFATAL, "bnd prefix is not allowed");
704 break;
705 case MERR_BADREPNE:
706 error(ERR_NONFATAL, "%s prefix is not allowed",
707 (has_prefix(instruction, PPS_REP, P_REPNE) ?
708 "repne" : "repnz"));
709 break;
710 default:
711 error(ERR_NONFATAL,
712 "invalid combination of opcode and operands");
713 break;
716 return 0;
719 int64_t insn_size(int32_t segment, int64_t offset, int bits, iflag_t cp,
720 insn * instruction, efunc error)
722 const struct itemplate *temp;
723 enum match_result m;
725 errfunc = error; /* to pass to other functions */
726 cpu = cp;
728 if (instruction->opcode == I_none)
729 return 0;
731 if (instruction->opcode == I_DB || instruction->opcode == I_DW ||
732 instruction->opcode == I_DD || instruction->opcode == I_DQ ||
733 instruction->opcode == I_DT || instruction->opcode == I_DO ||
734 instruction->opcode == I_DY) {
735 extop *e;
736 int32_t isize, osize, wsize;
738 isize = 0;
739 wsize = idata_bytes(instruction->opcode);
741 list_for_each(e, instruction->eops) {
742 int32_t align;
744 osize = 0;
745 if (e->type == EOT_DB_NUMBER) {
746 osize = 1;
747 warn_overflow_const(e->offset, wsize);
748 } else if (e->type == EOT_DB_STRING ||
749 e->type == EOT_DB_STRING_FREE)
750 osize = e->stringlen;
752 align = (-osize) % wsize;
753 if (align < 0)
754 align += wsize;
755 isize += osize + align;
757 return isize * instruction->times;
760 if (instruction->opcode == I_INCBIN) {
761 const char *fname = instruction->eops->stringval;
762 FILE *fp;
763 int64_t val = 0;
764 size_t len;
766 fp = fopen(fname, "rb");
767 if (!fp)
768 error(ERR_NONFATAL, "`incbin': unable to open file `%s'",
769 fname);
770 else if (fseek(fp, 0L, SEEK_END) < 0)
771 error(ERR_NONFATAL, "`incbin': unable to seek on file `%s'",
772 fname);
773 else {
774 len = ftell(fp);
775 if (instruction->eops->next) {
776 len -= instruction->eops->next->offset;
777 if (instruction->eops->next->next &&
778 len > (size_t)instruction->eops->next->next->offset) {
779 len = (size_t)instruction->eops->next->next->offset;
782 val = instruction->times * len;
784 if (fp)
785 fclose(fp);
786 return val;
789 /* Check to see if we need an address-size prefix */
790 add_asp(instruction, bits);
792 m = find_match(&temp, instruction, segment, offset, bits);
793 if (m == MOK_GOOD) {
794 /* we've matched an instruction. */
795 int64_t isize;
796 int j;
798 isize = calcsize(segment, offset, bits, instruction, temp);
799 if (isize < 0)
800 return -1;
801 for (j = 0; j < MAXPREFIX; j++) {
802 switch (instruction->prefixes[j]) {
803 case P_A16:
804 if (bits != 16)
805 isize++;
806 break;
807 case P_A32:
808 if (bits != 32)
809 isize++;
810 break;
811 case P_O16:
812 if (bits != 16)
813 isize++;
814 break;
815 case P_O32:
816 if (bits == 16)
817 isize++;
818 break;
819 case P_A64:
820 case P_O64:
821 case P_EVEX:
822 case P_VEX3:
823 case P_VEX2:
824 case P_NOBND:
825 case P_none:
826 break;
827 default:
828 isize++;
829 break;
832 return isize * instruction->times;
833 } else {
834 return -1; /* didn't match any instruction */
838 static void bad_hle_warn(const insn * ins, uint8_t hleok)
840 enum prefixes rep_pfx = ins->prefixes[PPS_REP];
841 enum whatwarn { w_none, w_lock, w_inval } ww;
842 static const enum whatwarn warn[2][4] =
844 { w_inval, w_inval, w_none, w_lock }, /* XACQUIRE */
845 { w_inval, w_none, w_none, w_lock }, /* XRELEASE */
847 unsigned int n;
849 n = (unsigned int)rep_pfx - P_XACQUIRE;
850 if (n > 1)
851 return; /* Not XACQUIRE/XRELEASE */
853 ww = warn[n][hleok];
854 if (!is_class(MEMORY, ins->oprs[0].type))
855 ww = w_inval; /* HLE requires operand 0 to be memory */
857 switch (ww) {
858 case w_none:
859 break;
861 case w_lock:
862 if (ins->prefixes[PPS_LOCK] != P_LOCK) {
863 errfunc(ERR_WARNING | ERR_WARN_HLE | ERR_PASS2,
864 "%s with this instruction requires lock",
865 prefix_name(rep_pfx));
867 break;
869 case w_inval:
870 errfunc(ERR_WARNING | ERR_WARN_HLE | ERR_PASS2,
871 "%s invalid with this instruction",
872 prefix_name(rep_pfx));
873 break;
877 /* Common construct */
878 #define case3(x) case (x): case (x)+1: case (x)+2
879 #define case4(x) case3(x): case (x)+3
881 static int64_t calcsize(int32_t segment, int64_t offset, int bits,
882 insn * ins, const struct itemplate *temp)
884 const uint8_t *codes = temp->code;
885 int64_t length = 0;
886 uint8_t c;
887 int rex_mask = ~0;
888 int op1, op2;
889 struct operand *opx;
890 uint8_t opex = 0;
891 enum ea_type eat;
892 uint8_t hleok = 0;
893 bool lockcheck = true;
894 enum reg_enum mib_index = R_none; /* For a separate index MIB reg form */
896 ins->rex = 0; /* Ensure REX is reset */
897 eat = EA_SCALAR; /* Expect a scalar EA */
898 memset(ins->evex_p, 0, 3); /* Ensure EVEX is reset */
900 if (ins->prefixes[PPS_OSIZE] == P_O64)
901 ins->rex |= REX_W;
903 (void)segment; /* Don't warn that this parameter is unused */
904 (void)offset; /* Don't warn that this parameter is unused */
906 while (*codes) {
907 c = *codes++;
908 op1 = (c & 3) + ((opex & 1) << 2);
909 op2 = ((c >> 3) & 3) + ((opex & 2) << 1);
910 opx = &ins->oprs[op1];
911 opex = 0; /* For the next iteration */
913 switch (c) {
914 case4(01):
915 codes += c, length += c;
916 break;
918 case3(05):
919 opex = c;
920 break;
922 case4(010):
923 ins->rex |=
924 op_rexflags(opx, REX_B|REX_H|REX_P|REX_W);
925 codes++, length++;
926 break;
928 case4(014):
929 /* this is an index reg of MIB operand */
930 mib_index = opx->basereg;
931 break;
933 case4(020):
934 case4(024):
935 length++;
936 break;
938 case4(030):
939 length += 2;
940 break;
942 case4(034):
943 if (opx->type & (BITS16 | BITS32 | BITS64))
944 length += (opx->type & BITS16) ? 2 : 4;
945 else
946 length += (bits == 16) ? 2 : 4;
947 break;
949 case4(040):
950 length += 4;
951 break;
953 case4(044):
954 length += ins->addr_size >> 3;
955 break;
957 case4(050):
958 length++;
959 break;
961 case4(054):
962 length += 8; /* MOV reg64/imm */
963 break;
965 case4(060):
966 length += 2;
967 break;
969 case4(064):
970 if (opx->type & (BITS16 | BITS32 | BITS64))
971 length += (opx->type & BITS16) ? 2 : 4;
972 else
973 length += (bits == 16) ? 2 : 4;
974 break;
976 case4(070):
977 length += 4;
978 break;
980 case4(074):
981 length += 2;
982 break;
984 case 0172:
985 case 0173:
986 codes++;
987 length++;
988 break;
990 case4(0174):
991 length++;
992 break;
994 case4(0240):
995 ins->rex |= REX_EV;
996 ins->vexreg = regval(opx);
997 ins->evex_p[2] |= op_evexflags(opx, EVEX_P2VP, 2); /* High-16 NDS */
998 ins->vex_cm = *codes++;
999 ins->vex_wlp = *codes++;
1000 ins->evex_tuple = (*codes++ - 0300);
1001 break;
1003 case 0250:
1004 ins->rex |= REX_EV;
1005 ins->vexreg = 0;
1006 ins->vex_cm = *codes++;
1007 ins->vex_wlp = *codes++;
1008 ins->evex_tuple = (*codes++ - 0300);
1009 break;
1011 case4(0254):
1012 length += 4;
1013 break;
1015 case4(0260):
1016 ins->rex |= REX_V;
1017 ins->vexreg = regval(opx);
1018 ins->vex_cm = *codes++;
1019 ins->vex_wlp = *codes++;
1020 break;
1022 case 0270:
1023 ins->rex |= REX_V;
1024 ins->vexreg = 0;
1025 ins->vex_cm = *codes++;
1026 ins->vex_wlp = *codes++;
1027 break;
1029 case3(0271):
1030 hleok = c & 3;
1031 break;
1033 case4(0274):
1034 length++;
1035 break;
1037 case4(0300):
1038 break;
1040 case 0310:
1041 if (bits == 64)
1042 return -1;
1043 length += (bits != 16) && !has_prefix(ins, PPS_ASIZE, P_A16);
1044 break;
1046 case 0311:
1047 length += (bits != 32) && !has_prefix(ins, PPS_ASIZE, P_A32);
1048 break;
1050 case 0312:
1051 break;
1053 case 0313:
1054 if (bits != 64 || has_prefix(ins, PPS_ASIZE, P_A16) ||
1055 has_prefix(ins, PPS_ASIZE, P_A32))
1056 return -1;
1057 break;
1059 case4(0314):
1060 break;
1062 case 0320:
1064 enum prefixes pfx = ins->prefixes[PPS_OSIZE];
1065 if (pfx == P_O16)
1066 break;
1067 if (pfx != P_none)
1068 errfunc(ERR_WARNING | ERR_PASS2, "invalid operand size prefix");
1069 else
1070 ins->prefixes[PPS_OSIZE] = P_O16;
1071 break;
1074 case 0321:
1076 enum prefixes pfx = ins->prefixes[PPS_OSIZE];
1077 if (pfx == P_O32)
1078 break;
1079 if (pfx != P_none)
1080 errfunc(ERR_WARNING | ERR_PASS2, "invalid operand size prefix");
1081 else
1082 ins->prefixes[PPS_OSIZE] = P_O32;
1083 break;
1086 case 0322:
1087 break;
1089 case 0323:
1090 rex_mask &= ~REX_W;
1091 break;
1093 case 0324:
1094 ins->rex |= REX_W;
1095 break;
1097 case 0325:
1098 ins->rex |= REX_NH;
1099 break;
1101 case 0326:
1102 break;
1104 case 0330:
1105 codes++, length++;
1106 break;
1108 case 0331:
1109 break;
1111 case 0332:
1112 case 0333:
1113 length++;
1114 break;
1116 case 0334:
1117 ins->rex |= REX_L;
1118 break;
1120 case 0335:
1121 break;
1123 case 0336:
1124 if (!ins->prefixes[PPS_REP])
1125 ins->prefixes[PPS_REP] = P_REP;
1126 break;
1128 case 0337:
1129 if (!ins->prefixes[PPS_REP])
1130 ins->prefixes[PPS_REP] = P_REPNE;
1131 break;
1133 case 0340:
1134 if (ins->oprs[0].segment != NO_SEG)
1135 errfunc(ERR_NONFATAL, "attempt to reserve non-constant"
1136 " quantity of BSS space");
1137 else
1138 length += ins->oprs[0].offset;
1139 break;
1141 case 0341:
1142 if (!ins->prefixes[PPS_WAIT])
1143 ins->prefixes[PPS_WAIT] = P_WAIT;
1144 break;
1146 case 0360:
1147 break;
1149 case 0361:
1150 length++;
1151 break;
1153 case 0364:
1154 case 0365:
1155 break;
1157 case 0366:
1158 case 0367:
1159 length++;
1160 break;
1162 case 0370:
1163 case 0371:
1164 break;
1166 case 0373:
1167 length++;
1168 break;
1170 case 0374:
1171 eat = EA_XMMVSIB;
1172 break;
1174 case 0375:
1175 eat = EA_YMMVSIB;
1176 break;
1178 case 0376:
1179 eat = EA_ZMMVSIB;
1180 break;
1182 case4(0100):
1183 case4(0110):
1184 case4(0120):
1185 case4(0130):
1186 case4(0200):
1187 case4(0204):
1188 case4(0210):
1189 case4(0214):
1190 case4(0220):
1191 case4(0224):
1192 case4(0230):
1193 case4(0234):
1195 ea ea_data;
1196 int rfield;
1197 opflags_t rflags;
1198 struct operand *opy = &ins->oprs[op2];
1199 struct operand *op_er_sae;
1201 ea_data.rex = 0; /* Ensure ea.REX is initially 0 */
1203 if (c <= 0177) {
1204 /* pick rfield from operand b (opx) */
1205 rflags = regflag(opx);
1206 rfield = nasm_regvals[opx->basereg];
1207 } else {
1208 rflags = 0;
1209 rfield = c & 7;
1212 /* EVEX.b1 : evex_brerop contains the operand position */
1213 op_er_sae = (ins->evex_brerop >= 0 ?
1214 &ins->oprs[ins->evex_brerop] : NULL);
1216 if (op_er_sae && (op_er_sae->decoflags & (ER | SAE))) {
1217 /* set EVEX.b */
1218 ins->evex_p[2] |= EVEX_P2B;
1219 if (op_er_sae->decoflags & ER) {
1220 /* set EVEX.RC (rounding control) */
1221 ins->evex_p[2] |= ((ins->evex_rm - BRC_RN) << 5)
1222 & EVEX_P2RC;
1224 } else {
1225 /* set EVEX.L'L (vector length) */
1226 ins->evex_p[2] |= ((ins->vex_wlp << (5 - 2)) & EVEX_P2LL);
1227 ins->evex_p[1] |= ((ins->vex_wlp << (7 - 4)) & EVEX_P1W);
1228 if (opy->decoflags & BRDCAST_MASK) {
1229 /* set EVEX.b */
1230 ins->evex_p[2] |= EVEX_P2B;
1235 * if a separate form of MIB (ICC style) is used,
1236 * the index reg info is merged into mem operand
1238 if (mib_index != R_none) {
1239 opy->indexreg = mib_index;
1240 opy->scale = 1;
1241 opy->hintbase = mib_index;
1242 opy->hinttype = EAH_NOTBASE;
1246 * only for mib operands, make a single reg index [reg*1].
1247 * gas uses this form to explicitly denote index register.
1249 if (itemp_has(temp, IF_MIB) &&
1250 (opy->indexreg == -1 && opy->hintbase == opy->basereg &&
1251 opy->hinttype == EAH_NOTBASE)) {
1252 opy->indexreg = opy->basereg;
1253 opy->basereg = -1;
1254 opy->scale = 1;
1257 if (process_ea(opy, &ea_data, bits,
1258 rfield, rflags, ins) != eat) {
1259 errfunc(ERR_NONFATAL, "invalid effective address");
1260 return -1;
1261 } else {
1262 ins->rex |= ea_data.rex;
1263 length += ea_data.size;
1266 break;
1268 default:
1269 errfunc(ERR_PANIC, "internal instruction table corrupt"
1270 ": instruction code \\%o (0x%02X) given", c, c);
1271 break;
1275 ins->rex &= rex_mask;
1277 if (ins->rex & REX_NH) {
1278 if (ins->rex & REX_H) {
1279 errfunc(ERR_NONFATAL, "instruction cannot use high registers");
1280 return -1;
1282 ins->rex &= ~REX_P; /* Don't force REX prefix due to high reg */
1285 switch (ins->prefixes[PPS_VEX]) {
1286 case P_EVEX:
1287 if (!(ins->rex & REX_EV))
1288 return -1;
1289 break;
1290 case P_VEX3:
1291 case P_VEX2:
1292 if (!(ins->rex & REX_V))
1293 return -1;
1294 break;
1295 default:
1296 break;
1299 if (ins->rex & (REX_V | REX_EV)) {
1300 int bad32 = REX_R|REX_W|REX_X|REX_B;
1302 if (ins->rex & REX_H) {
1303 errfunc(ERR_NONFATAL, "cannot use high register in AVX instruction");
1304 return -1;
1306 switch (ins->vex_wlp & 060) {
1307 case 000:
1308 case 040:
1309 ins->rex &= ~REX_W;
1310 break;
1311 case 020:
1312 ins->rex |= REX_W;
1313 bad32 &= ~REX_W;
1314 break;
1315 case 060:
1316 /* Follow REX_W */
1317 break;
1320 if (bits != 64 && ((ins->rex & bad32) || ins->vexreg > 7)) {
1321 errfunc(ERR_NONFATAL, "invalid operands in non-64-bit mode");
1322 return -1;
1323 } else if (!(ins->rex & REX_EV) &&
1324 ((ins->vexreg > 15) || (ins->evex_p[0] & 0xf0))) {
1325 errfunc(ERR_NONFATAL, "invalid high-16 register in non-AVX-512");
1326 return -1;
1328 if (ins->rex & REX_EV)
1329 length += 4;
1330 else if (ins->vex_cm != 1 || (ins->rex & (REX_W|REX_X|REX_B)) ||
1331 ins->prefixes[PPS_VEX] == P_VEX3)
1332 length += 3;
1333 else
1334 length += 2;
1335 } else if (ins->rex & REX_REAL) {
1336 if (ins->rex & REX_H) {
1337 errfunc(ERR_NONFATAL, "cannot use high register in rex instruction");
1338 return -1;
1339 } else if (bits == 64) {
1340 length++;
1341 } else if ((ins->rex & REX_L) &&
1342 !(ins->rex & (REX_P|REX_W|REX_X|REX_B)) &&
1343 iflag_ffs(&cpu) >= IF_X86_64) {
1344 /* LOCK-as-REX.R */
1345 assert_no_prefix(ins, PPS_LOCK);
1346 lockcheck = false; /* Already errored, no need for warning */
1347 length++;
1348 } else {
1349 errfunc(ERR_NONFATAL, "invalid operands in non-64-bit mode");
1350 return -1;
1354 if (has_prefix(ins, PPS_LOCK, P_LOCK) && lockcheck &&
1355 (!itemp_has(temp,IF_LOCK) || !is_class(MEMORY, ins->oprs[0].type))) {
1356 errfunc(ERR_WARNING | ERR_WARN_LOCK | ERR_PASS2 ,
1357 "instruction is not lockable");
1360 bad_hle_warn(ins, hleok);
1363 * when BND prefix is set by DEFAULT directive,
1364 * BND prefix is added to every appropriate instruction line
1365 * unless it is overridden by NOBND prefix.
1367 if (globalbnd &&
1368 (itemp_has(temp, IF_BND) && !has_prefix(ins, PPS_REP, P_NOBND)))
1369 ins->prefixes[PPS_REP] = P_BND;
1371 return length;
1374 static inline unsigned int emit_rex(insn *ins, int32_t segment, int64_t offset, int bits)
1376 if (bits == 64) {
1377 if ((ins->rex & REX_REAL) && !(ins->rex & (REX_V | REX_EV))) {
1378 ins->rex = (ins->rex & REX_REAL) | REX_P;
1379 out(offset, segment, &ins->rex, OUT_RAWDATA, 1, NO_SEG, NO_SEG);
1380 ins->rex = 0;
1381 return 1;
1385 return 0;
1388 static void gencode(int32_t segment, int64_t offset, int bits,
1389 insn * ins, const struct itemplate *temp,
1390 int64_t insn_end)
1392 uint8_t c;
1393 uint8_t bytes[4];
1394 int64_t size;
1395 int64_t data;
1396 int op1, op2;
1397 struct operand *opx;
1398 const uint8_t *codes = temp->code;
1399 uint8_t opex = 0;
1400 enum ea_type eat = EA_SCALAR;
1402 while (*codes) {
1403 c = *codes++;
1404 op1 = (c & 3) + ((opex & 1) << 2);
1405 op2 = ((c >> 3) & 3) + ((opex & 2) << 1);
1406 opx = &ins->oprs[op1];
1407 opex = 0; /* For the next iteration */
1409 switch (c) {
1410 case 01:
1411 case 02:
1412 case 03:
1413 case 04:
1414 offset += emit_rex(ins, segment, offset, bits);
1415 out(offset, segment, codes, OUT_RAWDATA, c, NO_SEG, NO_SEG);
1416 codes += c;
1417 offset += c;
1418 break;
1420 case 05:
1421 case 06:
1422 case 07:
1423 opex = c;
1424 break;
1426 case4(010):
1427 offset += emit_rex(ins, segment, offset, bits);
1428 bytes[0] = *codes++ + (regval(opx) & 7);
1429 out(offset, segment, bytes, OUT_RAWDATA, 1, NO_SEG, NO_SEG);
1430 offset += 1;
1431 break;
1433 case4(014):
1434 break;
1436 case4(020):
1437 if (opx->offset < -256 || opx->offset > 255) {
1438 errfunc(ERR_WARNING | ERR_PASS2 | ERR_WARN_NOV,
1439 "byte value exceeds bounds");
1441 out_imm8(offset, segment, opx);
1442 offset += 1;
1443 break;
1445 case4(024):
1446 if (opx->offset < 0 || opx->offset > 255)
1447 errfunc(ERR_WARNING | ERR_PASS2 | ERR_WARN_NOV,
1448 "unsigned byte value exceeds bounds");
1449 out_imm8(offset, segment, opx);
1450 offset += 1;
1451 break;
1453 case4(030):
1454 warn_overflow_opd(opx, 2);
1455 data = opx->offset;
1456 out(offset, segment, &data, OUT_ADDRESS, 2,
1457 opx->segment, opx->wrt);
1458 offset += 2;
1459 break;
1461 case4(034):
1462 if (opx->type & (BITS16 | BITS32))
1463 size = (opx->type & BITS16) ? 2 : 4;
1464 else
1465 size = (bits == 16) ? 2 : 4;
1466 warn_overflow_opd(opx, size);
1467 data = opx->offset;
1468 out(offset, segment, &data, OUT_ADDRESS, size,
1469 opx->segment, opx->wrt);
1470 offset += size;
1471 break;
1473 case4(040):
1474 warn_overflow_opd(opx, 4);
1475 data = opx->offset;
1476 out(offset, segment, &data, OUT_ADDRESS, 4,
1477 opx->segment, opx->wrt);
1478 offset += 4;
1479 break;
1481 case4(044):
1482 data = opx->offset;
1483 size = ins->addr_size >> 3;
1484 warn_overflow_opd(opx, size);
1485 out(offset, segment, &data, OUT_ADDRESS, size,
1486 opx->segment, opx->wrt);
1487 offset += size;
1488 break;
1490 case4(050):
1491 if (opx->segment != segment) {
1492 data = opx->offset;
1493 out(offset, segment, &data,
1494 OUT_REL1ADR, insn_end - offset,
1495 opx->segment, opx->wrt);
1496 } else {
1497 data = opx->offset - insn_end;
1498 if (data > 127 || data < -128)
1499 errfunc(ERR_NONFATAL, "short jump is out of range");
1500 out(offset, segment, &data,
1501 OUT_ADDRESS, 1, NO_SEG, NO_SEG);
1503 offset += 1;
1504 break;
1506 case4(054):
1507 data = (int64_t)opx->offset;
1508 out(offset, segment, &data, OUT_ADDRESS, 8,
1509 opx->segment, opx->wrt);
1510 offset += 8;
1511 break;
1513 case4(060):
1514 if (opx->segment != segment) {
1515 data = opx->offset;
1516 out(offset, segment, &data,
1517 OUT_REL2ADR, insn_end - offset,
1518 opx->segment, opx->wrt);
1519 } else {
1520 data = opx->offset - insn_end;
1521 out(offset, segment, &data,
1522 OUT_ADDRESS, 2, NO_SEG, NO_SEG);
1524 offset += 2;
1525 break;
1527 case4(064):
1528 if (opx->type & (BITS16 | BITS32 | BITS64))
1529 size = (opx->type & BITS16) ? 2 : 4;
1530 else
1531 size = (bits == 16) ? 2 : 4;
1532 if (opx->segment != segment) {
1533 data = opx->offset;
1534 out(offset, segment, &data,
1535 size == 2 ? OUT_REL2ADR : OUT_REL4ADR,
1536 insn_end - offset, opx->segment, opx->wrt);
1537 } else {
1538 data = opx->offset - insn_end;
1539 out(offset, segment, &data,
1540 OUT_ADDRESS, size, NO_SEG, NO_SEG);
1542 offset += size;
1543 break;
1545 case4(070):
1546 if (opx->segment != segment) {
1547 data = opx->offset;
1548 out(offset, segment, &data,
1549 OUT_REL4ADR, insn_end - offset,
1550 opx->segment, opx->wrt);
1551 } else {
1552 data = opx->offset - insn_end;
1553 out(offset, segment, &data,
1554 OUT_ADDRESS, 4, NO_SEG, NO_SEG);
1556 offset += 4;
1557 break;
1559 case4(074):
1560 if (opx->segment == NO_SEG)
1561 errfunc(ERR_NONFATAL, "value referenced by FAR is not"
1562 " relocatable");
1563 data = 0;
1564 out(offset, segment, &data, OUT_ADDRESS, 2,
1565 outfmt->segbase(1 + opx->segment),
1566 opx->wrt);
1567 offset += 2;
1568 break;
1570 case 0172:
1571 c = *codes++;
1572 opx = &ins->oprs[c >> 3];
1573 bytes[0] = nasm_regvals[opx->basereg] << 4;
1574 opx = &ins->oprs[c & 7];
1575 if (opx->segment != NO_SEG || opx->wrt != NO_SEG) {
1576 errfunc(ERR_NONFATAL,
1577 "non-absolute expression not permitted as argument %d",
1578 c & 7);
1579 } else {
1580 if (opx->offset & ~15) {
1581 errfunc(ERR_WARNING | ERR_PASS2 | ERR_WARN_NOV,
1582 "four-bit argument exceeds bounds");
1584 bytes[0] |= opx->offset & 15;
1586 out(offset, segment, bytes, OUT_RAWDATA, 1, NO_SEG, NO_SEG);
1587 offset++;
1588 break;
1590 case 0173:
1591 c = *codes++;
1592 opx = &ins->oprs[c >> 4];
1593 bytes[0] = nasm_regvals[opx->basereg] << 4;
1594 bytes[0] |= c & 15;
1595 out(offset, segment, bytes, OUT_RAWDATA, 1, NO_SEG, NO_SEG);
1596 offset++;
1597 break;
1599 case4(0174):
1600 bytes[0] = nasm_regvals[opx->basereg] << 4;
1601 out(offset, segment, bytes, OUT_RAWDATA, 1, NO_SEG, NO_SEG);
1602 offset++;
1603 break;
1605 case4(0254):
1606 data = opx->offset;
1607 if (opx->wrt == NO_SEG && opx->segment == NO_SEG &&
1608 (int32_t)data != (int64_t)data) {
1609 errfunc(ERR_WARNING | ERR_PASS2 | ERR_WARN_NOV,
1610 "signed dword immediate exceeds bounds");
1612 out(offset, segment, &data, OUT_ADDRESS, 4,
1613 opx->segment, opx->wrt);
1614 offset += 4;
1615 break;
1617 case4(0240):
1618 case 0250:
1619 codes += 3;
1620 ins->evex_p[2] |= op_evexflags(&ins->oprs[0],
1621 EVEX_P2Z | EVEX_P2AAA, 2);
1622 ins->evex_p[2] ^= EVEX_P2VP; /* 1's complement */
1623 bytes[0] = 0x62;
1624 /* EVEX.X can be set by either REX or EVEX for different reasons */
1625 bytes[1] = ((((ins->rex & 7) << 5) |
1626 (ins->evex_p[0] & (EVEX_P0X | EVEX_P0RP))) ^ 0xf0) |
1627 (ins->vex_cm & 3);
1628 bytes[2] = ((ins->rex & REX_W) << (7 - 3)) |
1629 ((~ins->vexreg & 15) << 3) |
1630 (1 << 2) | (ins->vex_wlp & 3);
1631 bytes[3] = ins->evex_p[2];
1632 out(offset, segment, &bytes, OUT_RAWDATA, 4, NO_SEG, NO_SEG);
1633 offset += 4;
1634 break;
1636 case4(0260):
1637 case 0270:
1638 codes += 2;
1639 if (ins->vex_cm != 1 || (ins->rex & (REX_W|REX_X|REX_B)) ||
1640 ins->prefixes[PPS_VEX] == P_VEX3) {
1641 bytes[0] = (ins->vex_cm >> 6) ? 0x8f : 0xc4;
1642 bytes[1] = (ins->vex_cm & 31) | ((~ins->rex & 7) << 5);
1643 bytes[2] = ((ins->rex & REX_W) << (7-3)) |
1644 ((~ins->vexreg & 15)<< 3) | (ins->vex_wlp & 07);
1645 out(offset, segment, &bytes, OUT_RAWDATA, 3, NO_SEG, NO_SEG);
1646 offset += 3;
1647 } else {
1648 bytes[0] = 0xc5;
1649 bytes[1] = ((~ins->rex & REX_R) << (7-2)) |
1650 ((~ins->vexreg & 15) << 3) | (ins->vex_wlp & 07);
1651 out(offset, segment, &bytes, OUT_RAWDATA, 2, NO_SEG, NO_SEG);
1652 offset += 2;
1654 break;
1656 case 0271:
1657 case 0272:
1658 case 0273:
1659 break;
1661 case4(0274):
1663 uint64_t uv, um;
1664 int s;
1666 if (ins->rex & REX_W)
1667 s = 64;
1668 else if (ins->prefixes[PPS_OSIZE] == P_O16)
1669 s = 16;
1670 else if (ins->prefixes[PPS_OSIZE] == P_O32)
1671 s = 32;
1672 else
1673 s = bits;
1675 um = (uint64_t)2 << (s-1);
1676 uv = opx->offset;
1678 if (uv > 127 && uv < (uint64_t)-128 &&
1679 (uv < um-128 || uv > um-1)) {
1680 /* If this wasn't explicitly byte-sized, warn as though we
1681 * had fallen through to the imm16/32/64 case.
1683 errfunc(ERR_WARNING | ERR_PASS2 | ERR_WARN_NOV,
1684 "%s value exceeds bounds",
1685 (opx->type & BITS8) ? "signed byte" :
1686 s == 16 ? "word" :
1687 s == 32 ? "dword" :
1688 "signed dword");
1690 if (opx->segment != NO_SEG) {
1691 data = uv;
1692 out(offset, segment, &data, OUT_ADDRESS, 1,
1693 opx->segment, opx->wrt);
1694 } else {
1695 bytes[0] = uv;
1696 out(offset, segment, bytes, OUT_RAWDATA, 1, NO_SEG,
1697 NO_SEG);
1699 offset += 1;
1700 break;
1703 case4(0300):
1704 break;
1706 case 0310:
1707 if (bits == 32 && !has_prefix(ins, PPS_ASIZE, P_A16)) {
1708 *bytes = 0x67;
1709 out(offset, segment, bytes, OUT_RAWDATA, 1, NO_SEG, NO_SEG);
1710 offset += 1;
1711 } else
1712 offset += 0;
1713 break;
1715 case 0311:
1716 if (bits != 32 && !has_prefix(ins, PPS_ASIZE, P_A32)) {
1717 *bytes = 0x67;
1718 out(offset, segment, bytes, OUT_RAWDATA, 1, NO_SEG, NO_SEG);
1719 offset += 1;
1720 } else
1721 offset += 0;
1722 break;
1724 case 0312:
1725 break;
1727 case 0313:
1728 ins->rex = 0;
1729 break;
1731 case4(0314):
1732 break;
1734 case 0320:
1735 case 0321:
1736 break;
1738 case 0322:
1739 case 0323:
1740 break;
1742 case 0324:
1743 ins->rex |= REX_W;
1744 break;
1746 case 0325:
1747 break;
1749 case 0326:
1750 break;
1752 case 0330:
1753 *bytes = *codes++ ^ get_cond_opcode(ins->condition);
1754 out(offset, segment, bytes, OUT_RAWDATA, 1, NO_SEG, NO_SEG);
1755 offset += 1;
1756 break;
1758 case 0331:
1759 break;
1761 case 0332:
1762 case 0333:
1763 *bytes = c - 0332 + 0xF2;
1764 out(offset, segment, bytes, OUT_RAWDATA, 1, NO_SEG, NO_SEG);
1765 offset += 1;
1766 break;
1768 case 0334:
1769 if (ins->rex & REX_R) {
1770 *bytes = 0xF0;
1771 out(offset, segment, bytes, OUT_RAWDATA, 1, NO_SEG, NO_SEG);
1772 offset += 1;
1774 ins->rex &= ~(REX_L|REX_R);
1775 break;
1777 case 0335:
1778 break;
1780 case 0336:
1781 case 0337:
1782 break;
1784 case 0340:
1785 if (ins->oprs[0].segment != NO_SEG)
1786 errfunc(ERR_PANIC, "non-constant BSS size in pass two");
1787 else {
1788 int64_t size = ins->oprs[0].offset;
1789 if (size > 0)
1790 out(offset, segment, NULL,
1791 OUT_RESERVE, size, NO_SEG, NO_SEG);
1792 offset += size;
1794 break;
1796 case 0341:
1797 break;
1799 case 0360:
1800 break;
1802 case 0361:
1803 bytes[0] = 0x66;
1804 out(offset, segment, bytes, OUT_RAWDATA, 1, NO_SEG, NO_SEG);
1805 offset += 1;
1806 break;
1808 case 0364:
1809 case 0365:
1810 break;
1812 case 0366:
1813 case 0367:
1814 *bytes = c - 0366 + 0x66;
1815 out(offset, segment, bytes, OUT_RAWDATA, 1, NO_SEG, NO_SEG);
1816 offset += 1;
1817 break;
1819 case3(0370):
1820 break;
1822 case 0373:
1823 *bytes = bits == 16 ? 3 : 5;
1824 out(offset, segment, bytes, OUT_RAWDATA, 1, NO_SEG, NO_SEG);
1825 offset += 1;
1826 break;
1828 case 0374:
1829 eat = EA_XMMVSIB;
1830 break;
1832 case 0375:
1833 eat = EA_YMMVSIB;
1834 break;
1836 case 0376:
1837 eat = EA_ZMMVSIB;
1838 break;
1840 case4(0100):
1841 case4(0110):
1842 case4(0120):
1843 case4(0130):
1844 case4(0200):
1845 case4(0204):
1846 case4(0210):
1847 case4(0214):
1848 case4(0220):
1849 case4(0224):
1850 case4(0230):
1851 case4(0234):
1853 ea ea_data;
1854 int rfield;
1855 opflags_t rflags;
1856 uint8_t *p;
1857 int32_t s;
1858 struct operand *opy = &ins->oprs[op2];
1860 if (c <= 0177) {
1861 /* pick rfield from operand b (opx) */
1862 rflags = regflag(opx);
1863 rfield = nasm_regvals[opx->basereg];
1864 } else {
1865 /* rfield is constant */
1866 rflags = 0;
1867 rfield = c & 7;
1870 if (process_ea(opy, &ea_data, bits,
1871 rfield, rflags, ins) != eat)
1872 errfunc(ERR_NONFATAL, "invalid effective address");
1874 p = bytes;
1875 *p++ = ea_data.modrm;
1876 if (ea_data.sib_present)
1877 *p++ = ea_data.sib;
1879 s = p - bytes;
1880 out(offset, segment, bytes, OUT_RAWDATA, s, NO_SEG, NO_SEG);
1883 * Make sure the address gets the right offset in case
1884 * the line breaks in the .lst file (BR 1197827)
1886 offset += s;
1887 s = 0;
1889 switch (ea_data.bytes) {
1890 case 0:
1891 break;
1892 case 1:
1893 case 2:
1894 case 4:
1895 case 8:
1896 /* use compressed displacement, if available */
1897 data = ea_data.disp8 ? ea_data.disp8 : opy->offset;
1898 s += ea_data.bytes;
1899 if (ea_data.rip) {
1900 if (opy->segment == segment) {
1901 data -= insn_end;
1902 if (overflow_signed(data, ea_data.bytes))
1903 warn_overflow(ERR_PASS2, ea_data.bytes);
1904 out(offset, segment, &data, OUT_ADDRESS,
1905 ea_data.bytes, NO_SEG, NO_SEG);
1906 } else {
1907 /* overflow check in output/linker? */
1908 out(offset, segment, &data, OUT_REL4ADR,
1909 insn_end - offset, opy->segment, opy->wrt);
1911 } else {
1912 if (overflow_general(data, ins->addr_size >> 3) ||
1913 signed_bits(data, ins->addr_size) !=
1914 signed_bits(data, ea_data.bytes * 8))
1915 warn_overflow(ERR_PASS2, ea_data.bytes);
1917 out(offset, segment, &data, OUT_ADDRESS,
1918 ea_data.bytes, opy->segment, opy->wrt);
1920 break;
1921 default:
1922 /* Impossible! */
1923 errfunc(ERR_PANIC,
1924 "Invalid amount of bytes (%d) for offset?!",
1925 ea_data.bytes);
1926 break;
1928 offset += s;
1930 break;
1932 default:
1933 errfunc(ERR_PANIC, "internal instruction table corrupt"
1934 ": instruction code \\%o (0x%02X) given", c, c);
1935 break;
1940 static opflags_t regflag(const operand * o)
1942 if (!is_register(o->basereg))
1943 errfunc(ERR_PANIC, "invalid operand passed to regflag()");
1944 return nasm_reg_flags[o->basereg];
1947 static int32_t regval(const operand * o)
1949 if (!is_register(o->basereg))
1950 errfunc(ERR_PANIC, "invalid operand passed to regval()");
1951 return nasm_regvals[o->basereg];
1954 static int op_rexflags(const operand * o, int mask)
1956 opflags_t flags;
1957 int val;
1959 if (!is_register(o->basereg))
1960 errfunc(ERR_PANIC, "invalid operand passed to op_rexflags()");
1962 flags = nasm_reg_flags[o->basereg];
1963 val = nasm_regvals[o->basereg];
1965 return rexflags(val, flags, mask);
1968 static int rexflags(int val, opflags_t flags, int mask)
1970 int rex = 0;
1972 if (val >= 0 && (val & 8))
1973 rex |= REX_B|REX_X|REX_R;
1974 if (flags & BITS64)
1975 rex |= REX_W;
1976 if (!(REG_HIGH & ~flags)) /* AH, CH, DH, BH */
1977 rex |= REX_H;
1978 else if (!(REG8 & ~flags) && val >= 4) /* SPL, BPL, SIL, DIL */
1979 rex |= REX_P;
1981 return rex & mask;
1984 static int evexflags(int val, decoflags_t deco,
1985 int mask, uint8_t byte)
1987 int evex = 0;
1989 switch (byte) {
1990 case 0:
1991 if (val >= 0 && (val & 16))
1992 evex |= (EVEX_P0RP | EVEX_P0X);
1993 break;
1994 case 2:
1995 if (val >= 0 && (val & 16))
1996 evex |= EVEX_P2VP;
1997 if (deco & Z)
1998 evex |= EVEX_P2Z;
1999 if (deco & OPMASK_MASK)
2000 evex |= deco & EVEX_P2AAA;
2001 break;
2003 return evex & mask;
2006 static int op_evexflags(const operand * o, int mask, uint8_t byte)
2008 int val;
2010 if (!is_register(o->basereg))
2011 errfunc(ERR_PANIC, "invalid operand passed to op_evexflags()");
2013 val = nasm_regvals[o->basereg];
2015 return evexflags(val, o->decoflags, mask, byte);
2018 static enum match_result find_match(const struct itemplate **tempp,
2019 insn *instruction,
2020 int32_t segment, int64_t offset, int bits)
2022 const struct itemplate *temp;
2023 enum match_result m, merr;
2024 opflags_t xsizeflags[MAX_OPERANDS];
2025 bool opsizemissing = false;
2026 int8_t broadcast = instruction->evex_brerop;
2027 int i;
2029 /* broadcasting uses a different data element size */
2030 for (i = 0; i < instruction->operands; i++)
2031 if (i == broadcast)
2032 xsizeflags[i] = instruction->oprs[i].decoflags & BRSIZE_MASK;
2033 else
2034 xsizeflags[i] = instruction->oprs[i].type & SIZE_MASK;
2036 merr = MERR_INVALOP;
2038 for (temp = nasm_instructions[instruction->opcode];
2039 temp->opcode != I_none; temp++) {
2040 m = matches(temp, instruction, bits);
2041 if (m == MOK_JUMP) {
2042 if (jmp_match(segment, offset, bits, instruction, temp))
2043 m = MOK_GOOD;
2044 else
2045 m = MERR_INVALOP;
2046 } else if (m == MERR_OPSIZEMISSING && !itemp_has(temp, IF_SX)) {
2048 * Missing operand size and a candidate for fuzzy matching...
2050 for (i = 0; i < temp->operands; i++)
2051 if (i == broadcast)
2052 xsizeflags[i] |= temp->deco[i] & BRSIZE_MASK;
2053 else
2054 xsizeflags[i] |= temp->opd[i] & SIZE_MASK;
2055 opsizemissing = true;
2057 if (m > merr)
2058 merr = m;
2059 if (merr == MOK_GOOD)
2060 goto done;
2063 /* No match, but see if we can get a fuzzy operand size match... */
2064 if (!opsizemissing)
2065 goto done;
2067 for (i = 0; i < instruction->operands; i++) {
2069 * We ignore extrinsic operand sizes on registers, so we should
2070 * never try to fuzzy-match on them. This also resolves the case
2071 * when we have e.g. "xmmrm128" in two different positions.
2073 if (is_class(REGISTER, instruction->oprs[i].type))
2074 continue;
2076 /* This tests if xsizeflags[i] has more than one bit set */
2077 if ((xsizeflags[i] & (xsizeflags[i]-1)))
2078 goto done; /* No luck */
2080 if (i == broadcast) {
2081 instruction->oprs[i].decoflags |= xsizeflags[i];
2082 instruction->oprs[i].type |= (xsizeflags[i] == BR_BITS32 ?
2083 BITS32 : BITS64);
2084 } else {
2085 instruction->oprs[i].type |= xsizeflags[i]; /* Set the size */
2089 /* Try matching again... */
2090 for (temp = nasm_instructions[instruction->opcode];
2091 temp->opcode != I_none; temp++) {
2092 m = matches(temp, instruction, bits);
2093 if (m == MOK_JUMP) {
2094 if (jmp_match(segment, offset, bits, instruction, temp))
2095 m = MOK_GOOD;
2096 else
2097 m = MERR_INVALOP;
2099 if (m > merr)
2100 merr = m;
2101 if (merr == MOK_GOOD)
2102 goto done;
2105 done:
2106 *tempp = temp;
2107 return merr;
2110 static enum match_result matches(const struct itemplate *itemp,
2111 insn *instruction, int bits)
2113 opflags_t size[MAX_OPERANDS], asize;
2114 bool opsizemissing = false;
2115 int i, oprs;
2118 * Check the opcode
2120 if (itemp->opcode != instruction->opcode)
2121 return MERR_INVALOP;
2124 * Count the operands
2126 if (itemp->operands != instruction->operands)
2127 return MERR_INVALOP;
2130 * Is it legal?
2132 if (!(optimizing > 0) && itemp_has(itemp, IF_OPT))
2133 return MERR_INVALOP;
2136 * {evex} available?
2138 switch (instruction->prefixes[PPS_VEX]) {
2139 case P_EVEX:
2140 if (!itemp_has(itemp, IF_EVEX))
2141 return MERR_ENCMISMATCH;
2142 break;
2143 case P_VEX3:
2144 case P_VEX2:
2145 if (!itemp_has(itemp, IF_VEX))
2146 return MERR_ENCMISMATCH;
2147 break;
2148 default:
2149 break;
2153 * Check that no spurious colons or TOs are present
2155 for (i = 0; i < itemp->operands; i++)
2156 if (instruction->oprs[i].type & ~itemp->opd[i] & (COLON | TO))
2157 return MERR_INVALOP;
2160 * Process size flags
2162 switch (itemp_smask(itemp)) {
2163 case IF_GENBIT(IF_SB):
2164 asize = BITS8;
2165 break;
2166 case IF_GENBIT(IF_SW):
2167 asize = BITS16;
2168 break;
2169 case IF_GENBIT(IF_SD):
2170 asize = BITS32;
2171 break;
2172 case IF_GENBIT(IF_SQ):
2173 asize = BITS64;
2174 break;
2175 case IF_GENBIT(IF_SO):
2176 asize = BITS128;
2177 break;
2178 case IF_GENBIT(IF_SY):
2179 asize = BITS256;
2180 break;
2181 case IF_GENBIT(IF_SZ):
2182 asize = BITS512;
2183 break;
2184 case IF_GENBIT(IF_SIZE):
2185 switch (bits) {
2186 case 16:
2187 asize = BITS16;
2188 break;
2189 case 32:
2190 asize = BITS32;
2191 break;
2192 case 64:
2193 asize = BITS64;
2194 break;
2195 default:
2196 asize = 0;
2197 break;
2199 break;
2200 default:
2201 asize = 0;
2202 break;
2205 if (itemp_armask(itemp)) {
2206 /* S- flags only apply to a specific operand */
2207 i = itemp_arg(itemp);
2208 memset(size, 0, sizeof size);
2209 size[i] = asize;
2210 } else {
2211 /* S- flags apply to all operands */
2212 for (i = 0; i < MAX_OPERANDS; i++)
2213 size[i] = asize;
2217 * Check that the operand flags all match up,
2218 * it's a bit tricky so lets be verbose:
2220 * 1) Find out the size of operand. If instruction
2221 * doesn't have one specified -- we're trying to
2222 * guess it either from template (IF_S* flag) or
2223 * from code bits.
2225 * 2) If template operand do not match the instruction OR
2226 * template has an operand size specified AND this size differ
2227 * from which instruction has (perhaps we got it from code bits)
2228 * we are:
2229 * a) Check that only size of instruction and operand is differ
2230 * other characteristics do match
2231 * b) Perhaps it's a register specified in instruction so
2232 * for such a case we just mark that operand as "size
2233 * missing" and this will turn on fuzzy operand size
2234 * logic facility (handled by a caller)
2236 for (i = 0; i < itemp->operands; i++) {
2237 opflags_t type = instruction->oprs[i].type;
2238 decoflags_t deco = instruction->oprs[i].decoflags;
2239 bool is_broadcast = deco & BRDCAST_MASK;
2240 uint8_t brcast_num = 0;
2241 opflags_t template_opsize, insn_opsize;
2243 if (!(type & SIZE_MASK))
2244 type |= size[i];
2246 insn_opsize = type & SIZE_MASK;
2247 if (!is_broadcast) {
2248 template_opsize = itemp->opd[i] & SIZE_MASK;
2249 } else {
2250 decoflags_t deco_brsize = itemp->deco[i] & BRSIZE_MASK;
2252 * when broadcasting, the element size depends on
2253 * the instruction type. decorator flag should match.
2256 if (deco_brsize) {
2257 template_opsize = (deco_brsize == BR_BITS32 ? BITS32 : BITS64);
2258 /* calculate the proper number : {1to<brcast_num>} */
2259 brcast_num = (itemp->opd[i] & SIZE_MASK) / BITS128 *
2260 BITS64 / template_opsize * 2;
2261 } else {
2262 template_opsize = 0;
2266 if ((itemp->opd[i] & ~type & ~SIZE_MASK) ||
2267 (deco & ~itemp->deco[i] & ~BRNUM_MASK)) {
2268 return MERR_INVALOP;
2269 } else if (template_opsize) {
2270 if (template_opsize != insn_opsize) {
2271 if (insn_opsize) {
2272 return MERR_INVALOP;
2273 } else if (!is_class(REGISTER, type)) {
2275 * Note: we don't honor extrinsic operand sizes for registers,
2276 * so "missing operand size" for a register should be
2277 * considered a wildcard match rather than an error.
2279 opsizemissing = true;
2281 } else if (is_broadcast &&
2282 (brcast_num !=
2283 (8U << ((deco & BRNUM_MASK) >> BRNUM_SHIFT)))) {
2285 * broadcasting opsize matches but the number of repeated memory
2286 * element does not match.
2287 * if 64b double precision float is broadcasted to zmm (512b),
2288 * broadcasting decorator must be {1to8}.
2290 return MERR_BRNUMMISMATCH;
2295 if (opsizemissing)
2296 return MERR_OPSIZEMISSING;
2299 * Check operand sizes
2301 if (itemp_has(itemp, IF_SM) || itemp_has(itemp, IF_SM2)) {
2302 oprs = (itemp_has(itemp, IF_SM2) ? 2 : itemp->operands);
2303 for (i = 0; i < oprs; i++) {
2304 asize = itemp->opd[i] & SIZE_MASK;
2305 if (asize) {
2306 for (i = 0; i < oprs; i++)
2307 size[i] = asize;
2308 break;
2311 } else {
2312 oprs = itemp->operands;
2315 for (i = 0; i < itemp->operands; i++) {
2316 if (!(itemp->opd[i] & SIZE_MASK) &&
2317 (instruction->oprs[i].type & SIZE_MASK & ~size[i]))
2318 return MERR_OPSIZEMISMATCH;
2322 * Check template is okay at the set cpu level
2324 if (iflag_cmp_cpu_level(&insns_flags[itemp->iflag_idx], &cpu) > 0)
2325 return MERR_BADCPU;
2328 * Verify the appropriate long mode flag.
2330 if (itemp_has(itemp, (bits == 64 ? IF_NOLONG : IF_LONG)))
2331 return MERR_BADMODE;
2334 * If we have a HLE prefix, look for the NOHLE flag
2336 if (itemp_has(itemp, IF_NOHLE) &&
2337 (has_prefix(instruction, PPS_REP, P_XACQUIRE) ||
2338 has_prefix(instruction, PPS_REP, P_XRELEASE)))
2339 return MERR_BADHLE;
2342 * Check if special handling needed for Jumps
2344 if ((itemp->code[0] & ~1) == 0370)
2345 return MOK_JUMP;
2348 * Check if BND prefix is allowed.
2349 * Other 0xF2 (REPNE/REPNZ) prefix is prohibited.
2351 if (!itemp_has(itemp, IF_BND) &&
2352 (has_prefix(instruction, PPS_REP, P_BND) ||
2353 has_prefix(instruction, PPS_REP, P_NOBND)))
2354 return MERR_BADBND;
2355 else if (itemp_has(itemp, IF_BND) &&
2356 (has_prefix(instruction, PPS_REP, P_REPNE) ||
2357 has_prefix(instruction, PPS_REP, P_REPNZ)))
2358 return MERR_BADREPNE;
2360 return MOK_GOOD;
2364 * Check if ModR/M.mod should/can be 01.
2365 * - EAF_BYTEOFFS is set
2366 * - offset can fit in a byte when EVEX is not used
2367 * - offset can be compressed when EVEX is used
2369 #define IS_MOD_01() (input->eaflags & EAF_BYTEOFFS || \
2370 (o >= -128 && o <= 127 && \
2371 seg == NO_SEG && !forw_ref && \
2372 !(input->eaflags & EAF_WORDOFFS) && \
2373 !(ins->rex & REX_EV)) || \
2374 (ins->rex & REX_EV && \
2375 is_disp8n(input, ins, &output->disp8)))
2377 static enum ea_type process_ea(operand *input, ea *output, int bits,
2378 int rfield, opflags_t rflags, insn *ins)
2380 bool forw_ref = !!(input->opflags & OPFLAG_UNKNOWN);
2381 int addrbits = ins->addr_size;
2383 output->type = EA_SCALAR;
2384 output->rip = false;
2385 output->disp8 = 0;
2387 /* REX flags for the rfield operand */
2388 output->rex |= rexflags(rfield, rflags, REX_R | REX_P | REX_W | REX_H);
2389 /* EVEX.R' flag for the REG operand */
2390 ins->evex_p[0] |= evexflags(rfield, 0, EVEX_P0RP, 0);
2392 if (is_class(REGISTER, input->type)) {
2394 * It's a direct register.
2396 if (!is_register(input->basereg))
2397 goto err;
2399 if (!is_reg_class(REG_EA, input->basereg))
2400 goto err;
2402 /* broadcasting is not available with a direct register operand. */
2403 if (input->decoflags & BRDCAST_MASK) {
2404 nasm_error(ERR_NONFATAL, "Broadcasting not allowed from a register");
2405 goto err;
2408 output->rex |= op_rexflags(input, REX_B | REX_P | REX_W | REX_H);
2409 ins->evex_p[0] |= op_evexflags(input, EVEX_P0X, 0);
2410 output->sib_present = false; /* no SIB necessary */
2411 output->bytes = 0; /* no offset necessary either */
2412 output->modrm = GEN_MODRM(3, rfield, nasm_regvals[input->basereg]);
2413 } else {
2415 * It's a memory reference.
2418 /* Embedded rounding or SAE is not available with a mem ref operand. */
2419 if (input->decoflags & (ER | SAE)) {
2420 nasm_error(ERR_NONFATAL,
2421 "Embedded rounding is available only with reg-reg op.");
2422 return -1;
2425 if (input->basereg == -1 &&
2426 (input->indexreg == -1 || input->scale == 0)) {
2428 * It's a pure offset.
2430 if (bits == 64 && ((input->type & IP_REL) == IP_REL) &&
2431 input->segment == NO_SEG) {
2432 nasm_error(ERR_WARNING | ERR_PASS1, "absolute address can not be RIP-relative");
2433 input->type &= ~IP_REL;
2434 input->type |= MEMORY;
2437 if (input->eaflags & EAF_BYTEOFFS ||
2438 (input->eaflags & EAF_WORDOFFS &&
2439 input->disp_size != (addrbits != 16 ? 32 : 16))) {
2440 nasm_error(ERR_WARNING | ERR_PASS1, "displacement size ignored on absolute address");
2443 if (bits == 64 && (~input->type & IP_REL)) {
2444 output->sib_present = true;
2445 output->sib = GEN_SIB(0, 4, 5);
2446 output->bytes = 4;
2447 output->modrm = GEN_MODRM(0, rfield, 4);
2448 output->rip = false;
2449 } else {
2450 output->sib_present = false;
2451 output->bytes = (addrbits != 16 ? 4 : 2);
2452 output->modrm = GEN_MODRM(0, rfield, (addrbits != 16 ? 5 : 6));
2453 output->rip = bits == 64;
2455 } else {
2457 * It's an indirection.
2459 int i = input->indexreg, b = input->basereg, s = input->scale;
2460 int32_t seg = input->segment;
2461 int hb = input->hintbase, ht = input->hinttype;
2462 int t, it, bt; /* register numbers */
2463 opflags_t x, ix, bx; /* register flags */
2465 if (s == 0)
2466 i = -1; /* make this easy, at least */
2468 if (is_register(i)) {
2469 it = nasm_regvals[i];
2470 ix = nasm_reg_flags[i];
2471 } else {
2472 it = -1;
2473 ix = 0;
2476 if (is_register(b)) {
2477 bt = nasm_regvals[b];
2478 bx = nasm_reg_flags[b];
2479 } else {
2480 bt = -1;
2481 bx = 0;
2484 /* if either one are a vector register... */
2485 if ((ix|bx) & (XMMREG|YMMREG|ZMMREG) & ~REG_EA) {
2486 opflags_t sok = BITS32 | BITS64;
2487 int32_t o = input->offset;
2488 int mod, scale, index, base;
2491 * For a vector SIB, one has to be a vector and the other,
2492 * if present, a GPR. The vector must be the index operand.
2494 if (it == -1 || (bx & (XMMREG|YMMREG|ZMMREG) & ~REG_EA)) {
2495 if (s == 0)
2496 s = 1;
2497 else if (s != 1)
2498 goto err;
2500 t = bt, bt = it, it = t;
2501 x = bx, bx = ix, ix = x;
2504 if (bt != -1) {
2505 if (REG_GPR & ~bx)
2506 goto err;
2507 if (!(REG64 & ~bx) || !(REG32 & ~bx))
2508 sok &= bx;
2509 else
2510 goto err;
2514 * While we're here, ensure the user didn't specify
2515 * WORD or QWORD
2517 if (input->disp_size == 16 || input->disp_size == 64)
2518 goto err;
2520 if (addrbits == 16 ||
2521 (addrbits == 32 && !(sok & BITS32)) ||
2522 (addrbits == 64 && !(sok & BITS64)))
2523 goto err;
2525 output->type = ((ix & ZMMREG & ~REG_EA) ? EA_ZMMVSIB
2526 : ((ix & YMMREG & ~REG_EA)
2527 ? EA_YMMVSIB : EA_XMMVSIB));
2529 output->rex |= rexflags(it, ix, REX_X);
2530 output->rex |= rexflags(bt, bx, REX_B);
2531 ins->evex_p[2] |= evexflags(it, 0, EVEX_P2VP, 2);
2533 index = it & 7; /* it is known to be != -1 */
2535 switch (s) {
2536 case 1:
2537 scale = 0;
2538 break;
2539 case 2:
2540 scale = 1;
2541 break;
2542 case 4:
2543 scale = 2;
2544 break;
2545 case 8:
2546 scale = 3;
2547 break;
2548 default: /* then what the smeg is it? */
2549 goto err; /* panic */
2552 if (bt == -1) {
2553 base = 5;
2554 mod = 0;
2555 } else {
2556 base = (bt & 7);
2557 if (base != REG_NUM_EBP && o == 0 &&
2558 seg == NO_SEG && !forw_ref &&
2559 !(input->eaflags & (EAF_BYTEOFFS | EAF_WORDOFFS)))
2560 mod = 0;
2561 else if (IS_MOD_01())
2562 mod = 1;
2563 else
2564 mod = 2;
2567 output->sib_present = true;
2568 output->bytes = (bt == -1 || mod == 2 ? 4 : mod);
2569 output->modrm = GEN_MODRM(mod, rfield, 4);
2570 output->sib = GEN_SIB(scale, index, base);
2571 } else if ((ix|bx) & (BITS32|BITS64)) {
2573 * it must be a 32/64-bit memory reference. Firstly we have
2574 * to check that all registers involved are type E/Rxx.
2576 opflags_t sok = BITS32 | BITS64;
2577 int32_t o = input->offset;
2579 if (it != -1) {
2580 if (!(REG64 & ~ix) || !(REG32 & ~ix))
2581 sok &= ix;
2582 else
2583 goto err;
2586 if (bt != -1) {
2587 if (REG_GPR & ~bx)
2588 goto err; /* Invalid register */
2589 if (~sok & bx & SIZE_MASK)
2590 goto err; /* Invalid size */
2591 sok &= bx;
2595 * While we're here, ensure the user didn't specify
2596 * WORD or QWORD
2598 if (input->disp_size == 16 || input->disp_size == 64)
2599 goto err;
2601 if (addrbits == 16 ||
2602 (addrbits == 32 && !(sok & BITS32)) ||
2603 (addrbits == 64 && !(sok & BITS64)))
2604 goto err;
2606 /* now reorganize base/index */
2607 if (s == 1 && bt != it && bt != -1 && it != -1 &&
2608 ((hb == b && ht == EAH_NOTBASE) ||
2609 (hb == i && ht == EAH_MAKEBASE))) {
2610 /* swap if hints say so */
2611 t = bt, bt = it, it = t;
2612 x = bx, bx = ix, ix = x;
2614 if (bt == it) /* convert EAX+2*EAX to 3*EAX */
2615 bt = -1, bx = 0, s++;
2616 if (bt == -1 && s == 1 && !(hb == i && ht == EAH_NOTBASE)) {
2617 /* make single reg base, unless hint */
2618 bt = it, bx = ix, it = -1, ix = 0;
2620 if (((s == 2 && it != REG_NUM_ESP && !(input->eaflags & EAF_TIMESTWO)) ||
2621 s == 3 || s == 5 || s == 9) && bt == -1)
2622 bt = it, bx = ix, s--; /* convert 3*EAX to EAX+2*EAX */
2623 if (it == -1 && (bt & 7) != REG_NUM_ESP &&
2624 (input->eaflags & EAF_TIMESTWO))
2625 it = bt, ix = bx, bt = -1, bx = 0, s = 1;
2626 /* convert [NOSPLIT EAX] to sib format with 0x0 displacement */
2627 if (s == 1 && it == REG_NUM_ESP) {
2628 /* swap ESP into base if scale is 1 */
2629 t = it, it = bt, bt = t;
2630 x = ix, ix = bx, bx = x;
2632 if (it == REG_NUM_ESP ||
2633 (s != 1 && s != 2 && s != 4 && s != 8 && it != -1))
2634 goto err; /* wrong, for various reasons */
2636 output->rex |= rexflags(it, ix, REX_X);
2637 output->rex |= rexflags(bt, bx, REX_B);
2639 if (it == -1 && (bt & 7) != REG_NUM_ESP) {
2640 /* no SIB needed */
2641 int mod, rm;
2643 if (bt == -1) {
2644 rm = 5;
2645 mod = 0;
2646 } else {
2647 rm = (bt & 7);
2648 if (rm != REG_NUM_EBP && o == 0 &&
2649 seg == NO_SEG && !forw_ref &&
2650 !(input->eaflags & (EAF_BYTEOFFS | EAF_WORDOFFS)))
2651 mod = 0;
2652 else if (IS_MOD_01())
2653 mod = 1;
2654 else
2655 mod = 2;
2658 output->sib_present = false;
2659 output->bytes = (bt == -1 || mod == 2 ? 4 : mod);
2660 output->modrm = GEN_MODRM(mod, rfield, rm);
2661 } else {
2662 /* we need a SIB */
2663 int mod, scale, index, base;
2665 if (it == -1)
2666 index = 4, s = 1;
2667 else
2668 index = (it & 7);
2670 switch (s) {
2671 case 1:
2672 scale = 0;
2673 break;
2674 case 2:
2675 scale = 1;
2676 break;
2677 case 4:
2678 scale = 2;
2679 break;
2680 case 8:
2681 scale = 3;
2682 break;
2683 default: /* then what the smeg is it? */
2684 goto err; /* panic */
2687 if (bt == -1) {
2688 base = 5;
2689 mod = 0;
2690 } else {
2691 base = (bt & 7);
2692 if (base != REG_NUM_EBP && o == 0 &&
2693 seg == NO_SEG && !forw_ref &&
2694 !(input->eaflags & (EAF_BYTEOFFS | EAF_WORDOFFS)))
2695 mod = 0;
2696 else if (IS_MOD_01())
2697 mod = 1;
2698 else
2699 mod = 2;
2702 output->sib_present = true;
2703 output->bytes = (bt == -1 || mod == 2 ? 4 : mod);
2704 output->modrm = GEN_MODRM(mod, rfield, 4);
2705 output->sib = GEN_SIB(scale, index, base);
2707 } else { /* it's 16-bit */
2708 int mod, rm;
2709 int16_t o = input->offset;
2711 /* check for 64-bit long mode */
2712 if (addrbits == 64)
2713 goto err;
2715 /* check all registers are BX, BP, SI or DI */
2716 if ((b != -1 && b != R_BP && b != R_BX && b != R_SI && b != R_DI) ||
2717 (i != -1 && i != R_BP && i != R_BX && i != R_SI && i != R_DI))
2718 goto err;
2720 /* ensure the user didn't specify DWORD/QWORD */
2721 if (input->disp_size == 32 || input->disp_size == 64)
2722 goto err;
2724 if (s != 1 && i != -1)
2725 goto err; /* no can do, in 16-bit EA */
2726 if (b == -1 && i != -1) {
2727 int tmp = b;
2728 b = i;
2729 i = tmp;
2730 } /* swap */
2731 if ((b == R_SI || b == R_DI) && i != -1) {
2732 int tmp = b;
2733 b = i;
2734 i = tmp;
2736 /* have BX/BP as base, SI/DI index */
2737 if (b == i)
2738 goto err; /* shouldn't ever happen, in theory */
2739 if (i != -1 && b != -1 &&
2740 (i == R_BP || i == R_BX || b == R_SI || b == R_DI))
2741 goto err; /* invalid combinations */
2742 if (b == -1) /* pure offset: handled above */
2743 goto err; /* so if it gets to here, panic! */
2745 rm = -1;
2746 if (i != -1)
2747 switch (i * 256 + b) {
2748 case R_SI * 256 + R_BX:
2749 rm = 0;
2750 break;
2751 case R_DI * 256 + R_BX:
2752 rm = 1;
2753 break;
2754 case R_SI * 256 + R_BP:
2755 rm = 2;
2756 break;
2757 case R_DI * 256 + R_BP:
2758 rm = 3;
2759 break;
2760 } else
2761 switch (b) {
2762 case R_SI:
2763 rm = 4;
2764 break;
2765 case R_DI:
2766 rm = 5;
2767 break;
2768 case R_BP:
2769 rm = 6;
2770 break;
2771 case R_BX:
2772 rm = 7;
2773 break;
2775 if (rm == -1) /* can't happen, in theory */
2776 goto err; /* so panic if it does */
2778 if (o == 0 && seg == NO_SEG && !forw_ref && rm != 6 &&
2779 !(input->eaflags & (EAF_BYTEOFFS | EAF_WORDOFFS)))
2780 mod = 0;
2781 else if (IS_MOD_01())
2782 mod = 1;
2783 else
2784 mod = 2;
2786 output->sib_present = false; /* no SIB - it's 16-bit */
2787 output->bytes = mod; /* bytes of offset needed */
2788 output->modrm = GEN_MODRM(mod, rfield, rm);
2793 output->size = 1 + output->sib_present + output->bytes;
2794 return output->type;
2796 err:
2797 return output->type = EA_INVALID;
2800 static void add_asp(insn *ins, int addrbits)
2802 int j, valid;
2803 int defdisp;
2805 valid = (addrbits == 64) ? 64|32 : 32|16;
2807 switch (ins->prefixes[PPS_ASIZE]) {
2808 case P_A16:
2809 valid &= 16;
2810 break;
2811 case P_A32:
2812 valid &= 32;
2813 break;
2814 case P_A64:
2815 valid &= 64;
2816 break;
2817 case P_ASP:
2818 valid &= (addrbits == 32) ? 16 : 32;
2819 break;
2820 default:
2821 break;
2824 for (j = 0; j < ins->operands; j++) {
2825 if (is_class(MEMORY, ins->oprs[j].type)) {
2826 opflags_t i, b;
2828 /* Verify as Register */
2829 if (!is_register(ins->oprs[j].indexreg))
2830 i = 0;
2831 else
2832 i = nasm_reg_flags[ins->oprs[j].indexreg];
2834 /* Verify as Register */
2835 if (!is_register(ins->oprs[j].basereg))
2836 b = 0;
2837 else
2838 b = nasm_reg_flags[ins->oprs[j].basereg];
2840 if (ins->oprs[j].scale == 0)
2841 i = 0;
2843 if (!i && !b) {
2844 int ds = ins->oprs[j].disp_size;
2845 if ((addrbits != 64 && ds > 8) ||
2846 (addrbits == 64 && ds == 16))
2847 valid &= ds;
2848 } else {
2849 if (!(REG16 & ~b))
2850 valid &= 16;
2851 if (!(REG32 & ~b))
2852 valid &= 32;
2853 if (!(REG64 & ~b))
2854 valid &= 64;
2856 if (!(REG16 & ~i))
2857 valid &= 16;
2858 if (!(REG32 & ~i))
2859 valid &= 32;
2860 if (!(REG64 & ~i))
2861 valid &= 64;
2866 if (valid & addrbits) {
2867 ins->addr_size = addrbits;
2868 } else if (valid & ((addrbits == 32) ? 16 : 32)) {
2869 /* Add an address size prefix */
2870 ins->prefixes[PPS_ASIZE] = (addrbits == 32) ? P_A16 : P_A32;;
2871 ins->addr_size = (addrbits == 32) ? 16 : 32;
2872 } else {
2873 /* Impossible... */
2874 errfunc(ERR_NONFATAL, "impossible combination of address sizes");
2875 ins->addr_size = addrbits; /* Error recovery */
2878 defdisp = ins->addr_size == 16 ? 16 : 32;
2880 for (j = 0; j < ins->operands; j++) {
2881 if (!(MEM_OFFS & ~ins->oprs[j].type) &&
2882 (ins->oprs[j].disp_size ? ins->oprs[j].disp_size : defdisp) != ins->addr_size) {
2884 * mem_offs sizes must match the address size; if not,
2885 * strip the MEM_OFFS bit and match only EA instructions
2887 ins->oprs[j].type &= ~(MEM_OFFS & ~MEMORY);