asm/parser.c

   1 /* ----------------------------------------------------------------------- *
   2  *
   3  *   Copyright 1996-2017 The NASM Authors - All Rights Reserved
   4  *   See the file AUTHORS included with the NASM distribution for
   5  *   the specific copyright holders.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following
   9  *   conditions are met:
  10  *
  11  *   * Redistributions of source code must retain the above copyright
  12  *     notice, this list of conditions and the following disclaimer.
  13  *   * Redistributions in binary form must reproduce the above
  14  *     copyright notice, this list of conditions and the following
  15  *     disclaimer in the documentation and/or other materials provided
  16  *     with the distribution.
  17  *
  18  *     THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND
  19  *     CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES,
  20  *     INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  21  *     MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  22  *     DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
  23  *     CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  24  *     SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  25  *     NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  26  *     LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  27  *     HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
  28  *     CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
  29  *     OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
  30  *     EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  31  *
  32  * ----------------------------------------------------------------------- */
  33
  34 /*
  35  * parser.c   source line parser for the Netwide Assembler
  36  */
  37
  38 #include "compiler.h"
  39
  40 #include <stdio.h>
  41 #include <stdlib.h>
  42 #include <stddef.h>
  43 #include <string.h>
  44 #include <ctype.h>
  45
  46 #include "nasm.h"
  47 #include "insns.h"
  48 #include "nasmlib.h"
  49 #include "error.h"
  50 #include "stdscan.h"
  51 #include "eval.h"
  52 #include "parser.h"
  53 #include "float.h"
  54 #include "assemble.h"
  55 #include "tables.h"
  56
  57
  58 static int is_comma_next(void);
  59
  60 static struct tokenval tokval;
  61
  62 static int prefix_slot(int prefix)
  63 {
  64     switch (prefix) {
  65     case P_WAIT:
  66         return PPS_WAIT;
  67     case R_CS:
  68     case R_DS:
  69     case R_SS:
  70     case R_ES:
  71     case R_FS:
  72     case R_GS:
  73         return PPS_SEG;
  74     case P_LOCK:
  75         return PPS_LOCK;
  76     case P_REP:
  77     case P_REPE:
  78     case P_REPZ:
  79     case P_REPNE:
  80     case P_REPNZ:
  81     case P_XACQUIRE:
  82     case P_XRELEASE:
  83     case P_BND:
  84     case P_NOBND:
  85         return PPS_REP;
  86     case P_O16:
  87     case P_O32:
  88     case P_O64:
  89     case P_OSP:
  90         return PPS_OSIZE;
  91     case P_A16:
  92     case P_A32:
  93     case P_A64:
  94     case P_ASP:
  95         return PPS_ASIZE;
  96     case P_EVEX:
  97     case P_VEX3:
  98     case P_VEX2:
  99         return PPS_VEX;
 100     default:
 101         nasm_panic(0, "Invalid value %d passed to prefix_slot()", prefix);
 102         return -1;
 103     }
 104 }
 105
 106 static void process_size_override(insn *result, operand *op)
 107 {
 108     if (tasm_compatible_mode) {
 109         switch (tokval.t_integer) {
 110             /* For TASM compatibility a size override inside the
 111              * brackets changes the size of the operand, not the
 112              * address type of the operand as it does in standard
 113              * NASM syntax. Hence:
 114              *
 115              *  mov     eax,[DWORD val]
 116              *
 117              * is valid syntax in TASM compatibility mode. Note that
 118              * you lose the ability to override the default address
 119              * type for the instruction, but we never use anything
 120              * but 32-bit flat model addressing in our code.
 121              */
 122         case S_BYTE:
 123             op->type |= BITS8;
 124             break;
 125         case S_WORD:
 126             op->type |= BITS16;
 127             break;
 128         case S_DWORD:
 129         case S_LONG:
 130             op->type |= BITS32;
 131             break;
 132         case S_QWORD:
 133             op->type |= BITS64;
 134             break;
 135         case S_TWORD:
 136             op->type |= BITS80;
 137             break;
 138         case S_OWORD:
 139             op->type |= BITS128;
 140             break;
 141         default:
 142             nasm_error(ERR_NONFATAL,
 143                        "invalid operand size specification");
 144             break;
 145         }
 146     } else {
 147         /* Standard NASM compatible syntax */
 148         switch (tokval.t_integer) {
 149         case S_NOSPLIT:
 150             op->eaflags |= EAF_TIMESTWO;
 151             break;
 152         case S_REL:
 153             op->eaflags |= EAF_REL;
 154             break;
 155         case S_ABS:
 156             op->eaflags |= EAF_ABS;
 157             break;
 158         case S_BYTE:
 159             op->disp_size = 8;
 160             op->eaflags |= EAF_BYTEOFFS;
 161             break;
 162         case P_A16:
 163         case P_A32:
 164         case P_A64:
 165             if (result->prefixes[PPS_ASIZE] &&
 166                 result->prefixes[PPS_ASIZE] != tokval.t_integer)
 167                 nasm_error(ERR_NONFATAL,
 168                            "conflicting address size specifications");
 169             else
 170                 result->prefixes[PPS_ASIZE] = tokval.t_integer;
 171             break;
 172         case S_WORD:
 173             op->disp_size = 16;
 174             op->eaflags |= EAF_WORDOFFS;
 175             break;
 176         case S_DWORD:
 177         case S_LONG:
 178             op->disp_size = 32;
 179             op->eaflags |= EAF_WORDOFFS;
 180             break;
 181         case S_QWORD:
 182             op->disp_size = 64;
 183             op->eaflags |= EAF_WORDOFFS;
 184             break;
 185         default:
 186             nasm_error(ERR_NONFATAL, "invalid size specification in"
 187                        " effective address");
 188             break;
 189         }
 190     }
 191 }
 192
 193 /*
 194  * Brace decorators are are parsed here.  opmask and zeroing
 195  * decorators can be placed in any order.  e.g. zmm1 {k2}{z} or zmm2
 196  * {z}{k3} decorator(s) are placed at the end of an operand.
 197  */
 198 static bool parse_braces(decoflags_t *decoflags)
 199 {
 200     int i, j;
 201
 202     i = tokval.t_type;
 203
 204     while (true) {
 205         switch (i) {
 206         case TOKEN_OPMASK:
 207             if (*decoflags & OPMASK_MASK) {
 208                 nasm_error(ERR_NONFATAL,
 209                            "opmask k%"PRIu64" is already set",
 210                            *decoflags & OPMASK_MASK);
 211                 *decoflags &= ~OPMASK_MASK;
 212             }
 213             *decoflags |= VAL_OPMASK(nasm_regvals[tokval.t_integer]);
 214             break;
 215         case TOKEN_DECORATOR:
 216             j = tokval.t_integer;
 217             switch (j) {
 218             case BRC_Z:
 219                 *decoflags |= Z_MASK;
 220                 break;
 221             case BRC_1TO2:
 222             case BRC_1TO4:
 223             case BRC_1TO8:
 224             case BRC_1TO16:
 225                 *decoflags |= BRDCAST_MASK | VAL_BRNUM(j - BRC_1TO2);
 226                 break;
 227             default:
 228                 nasm_error(ERR_NONFATAL,
 229                            "{%s} is not an expected decorator",
 230                            tokval.t_charptr);
 231                 break;
 232             }
 233             break;
 234         case ',':
 235         case TOKEN_EOS:
 236             return false;
 237         default:
 238             nasm_error(ERR_NONFATAL,
 239                        "only a series of valid decorators expected");
 240             return true;
 241         }
 242         i = stdscan(NULL, &tokval);
 243     }
 244 }
 245
 246 static int parse_mref(operand *op, const expr *e)
 247 {
 248     int b, i, s;        /* basereg, indexreg, scale */
 249     int64_t o;          /* offset */
 250
 251     b = i = -1;
 252     o = s = 0;
 253     op->segment = op->wrt = NO_SEG;
 254
 255     if (e->type && e->type <= EXPR_REG_END) {   /* this bit's a register */
 256         bool is_gpr = is_class(REG_GPR,nasm_reg_flags[e->type]);
 257
 258         if (is_gpr && e->value == 1)
 259             b = e->type;        /* It can be basereg */
 260         else                    /* No, it has to be indexreg */
 261             i = e->type, s = e->value;
 262         e++;
 263     }
 264     if (e->type && e->type <= EXPR_REG_END) {   /* it's a 2nd register */
 265         bool is_gpr = is_class(REG_GPR,nasm_reg_flags[e->type]);
 266
 267         if (b != -1)    /* If the first was the base, ... */
 268             i = e->type, s = e->value;  /* second has to be indexreg */
 269
 270         else if (!is_gpr || e->value != 1) {
 271             /* If both want to be index */
 272             nasm_error(ERR_NONFATAL,
 273                        "invalid effective address: two index registers");
 274             return -1;
 275         } else
 276             b = e->type;
 277         e++;
 278     }
 279
 280     if (e->type) {                     /* is there an offset? */
 281         if (e->type <= EXPR_REG_END) {  /* in fact, is there an error? */
 282             nasm_error(ERR_NONFATAL,
 283                        "invalid effective address: impossible register");
 284             return -1;
 285         } else {
 286             if (e->type == EXPR_UNKNOWN) {
 287                 op->opflags |= OPFLAG_UNKNOWN;
 288                 o = 0;  /* doesn't matter what */
 289                 while (e->type)
 290                     e++;        /* go to the end of the line */
 291             } else {
 292                 if (e->type == EXPR_SIMPLE) {
 293                     o = e->value;
 294                     e++;
 295                 }
 296                 if (e->type == EXPR_WRT) {
 297                     op->wrt = e->value;
 298                     e++;
 299                 }
 300                 /*
 301                  * Look for a segment base type.
 302                  */
 303                 for (; e->type; e++) {
 304                     if (!e->value)
 305                         continue;
 306
 307                     if (e->type <= EXPR_REG_END) {
 308                         nasm_error(ERR_NONFATAL,
 309                                    "invalid effective address: too many registers");
 310                         return -1;
 311                     } else if (e->type < EXPR_SEGBASE) {
 312                         nasm_error(ERR_NONFATAL,
 313                                    "invalid effective address: bad subexpression type");
 314                         return -1;
 315                     } else if (e->value == 1) {
 316                         if (op->segment != NO_SEG) {
 317                             nasm_error(ERR_NONFATAL,
 318                                        "invalid effective address: multiple base segments");
 319                             return -1;
 320                         }
 321                         op->segment = e->type - EXPR_SEGBASE;
 322                     } else if (e->value == -1 &&
 323                                e->type == location.segment + EXPR_SEGBASE &&
 324                                !(op->opflags & OPFLAG_RELATIVE)) {
 325                         op->opflags |= OPFLAG_RELATIVE;
 326                     } else {
 327                         nasm_error(ERR_NONFATAL,
 328                                    "invalid effective address: impossible segment base multiplier");
 329                         return -1;
 330                     }
 331                 }
 332             }
 333         }
 334     }
 335
 336     nasm_assert(!e->type);      /* We should be at the end */
 337
 338     op->basereg = b;
 339     op->indexreg = i;
 340     op->scale = s;
 341     op->offset = o;
 342     return 0;
 343 }
 344
 345 static void mref_set_optype(operand *op)
 346 {
 347     int b = op->basereg;
 348     int i = op->indexreg;
 349     int s = op->scale;
 350
 351     /* It is memory, but it can match any r/m operand */
 352     op->type |= MEMORY_ANY;
 353
 354     if (b == -1 && (i == -1 || s == 0)) {
 355         int is_rel = globalbits == 64 &&
 356             !(op->eaflags & EAF_ABS) &&
 357             ((globalrel &&
 358               !(op->eaflags & EAF_FSGS)) ||
 359              (op->eaflags & EAF_REL));
 360
 361         op->type |= is_rel ? IP_REL : MEM_OFFS;
 362     }
 363
 364     if (i != -1) {
 365         opflags_t iclass = nasm_reg_flags[i];
 366
 367         if (is_class(XMMREG,iclass))
 368             op->type |= XMEM;
 369         else if (is_class(YMMREG,iclass))
 370             op->type |= YMEM;
 371         else if (is_class(ZMMREG,iclass))
 372             op->type |= ZMEM;
 373     }
 374 }
 375
 376 /*
 377  * Convert an expression vector returned from evaluate() into an
 378  * extop structure.  Return zero on success.
 379  */
 380 static int value_to_extop(expr * vect, extop *eop, int32_t myseg)
 381 {
 382     eop->type = EOT_DB_NUMBER;
 383     eop->offset = 0;
 384     eop->segment = eop->wrt = NO_SEG;
 385     eop->relative = false;
 386
 387     for (; vect->type; vect++) {
 388         if (!vect->value)       /* zero term, safe to ignore */
 389             continue;
 390
 391         if (vect->type <= EXPR_REG_END) /* false if a register is present */
 392             return -1;
 393
 394         if (vect->type == EXPR_UNKNOWN) /* something we can't resolve yet */
 395             return 0;
 396
 397         if (vect->type == EXPR_SIMPLE) {
 398             /* Simple number expression */
 399             eop->offset += vect->value;
 400             continue;
 401         }
 402         if (eop->wrt == NO_SEG && !eop->relative && vect->type == EXPR_WRT) {
 403             /* WRT term */
 404             eop->wrt = vect->value;
 405             continue;
 406         }
 407
 408         if (!eop->relative &&
 409             vect->type == EXPR_SEGBASE + myseg && vect->value == -1) {
 410             /* Expression of the form: foo - $ */
 411             eop->relative = true;
 412             continue;
 413         }
 414
 415         if (eop->segment == NO_SEG && vect->type >= EXPR_SEGBASE &&
 416             vect->value == 1) {
 417             eop->segment = vect->type - EXPR_SEGBASE;
 418             continue;
 419         }
 420
 421         /* Otherwise, badness */
 422         return -1;
 423     }
 424
 425     /* We got to the end and it was all okay */
 426     return 0;
 427 }
 428
 429 insn *parse_line(int pass, char *buffer, insn *result, ldfunc ldef)
 430 {
 431     bool insn_is_label = false;
 432     struct eval_hints hints;
 433     int opnum;
 434     int critical;
 435     bool first;
 436     bool recover;
 437     int i;
 438
 439 restart_parse:
 440     first               = true;
 441     result->forw_ref    = false;
 442
 443     stdscan_reset();
 444     stdscan_set(buffer);
 445     i = stdscan(NULL, &tokval);
 446
 447     result->label       = NULL; /* Assume no label */
 448     result->eops        = NULL; /* must do this, whatever happens */
 449     result->operands    = 0;    /* must initialize this */
 450     result->evex_rm     = 0;    /* Ensure EVEX rounding mode is reset */
 451     result->evex_brerop = -1;   /* Reset EVEX broadcasting/ER op position */
 452
 453     /* Ignore blank lines */
 454     if (i == TOKEN_EOS)
 455         goto fail;
 456
 457     if (i != TOKEN_ID       &&
 458         i != TOKEN_INSN     &&
 459         i != TOKEN_PREFIX   &&
 460         (i != TOKEN_REG || !IS_SREG(tokval.t_integer))) {
 461         nasm_error(ERR_NONFATAL,
 462                    "label or instruction expected at start of line");
 463         goto fail;
 464     }
 465
 466     if (i == TOKEN_ID || (insn_is_label && i == TOKEN_INSN)) {
 467         /* there's a label here */
 468         first = false;
 469         result->label = tokval.t_charptr;
 470         i = stdscan(NULL, &tokval);
 471         if (i == ':') {         /* skip over the optional colon */
 472             i = stdscan(NULL, &tokval);
 473         } else if (i == 0) {
 474             nasm_error(ERR_WARNING | ERR_WARN_OL | ERR_PASS1,
 475                   "label alone on a line without a colon might be in error");
 476         }
 477         if (i != TOKEN_INSN || tokval.t_integer != I_EQU) {
 478             /*
 479              * FIXME: location.segment could be NO_SEG, in which case
 480              * it is possible we should be passing 'absolute.segment'. Look into this.
 481              * Work out whether that is *really* what we should be doing.
 482              * Generally fix things. I think this is right as it is, but
 483              * am still not certain.
 484              */
 485             ldef(result->label, in_absolute ? absolute.segment : location.segment,
 486                  location.offset, NULL, true, false);
 487         }
 488     }
 489
 490     /* Just a label here */
 491     if (i == TOKEN_EOS)
 492         goto fail;
 493
 494     nasm_static_assert(P_none == 0);
 495     memset(result->prefixes, P_none, sizeof(result->prefixes));
 496     result->times = 1L;
 497
 498     while (i == TOKEN_PREFIX ||
 499            (i == TOKEN_REG && IS_SREG(tokval.t_integer))) {
 500         first = false;
 501
 502         /*
 503          * Handle special case: the TIMES prefix.
 504          */
 505         if (i == TOKEN_PREFIX && tokval.t_integer == P_TIMES) {
 506             expr *value;
 507
 508             i = stdscan(NULL, &tokval);
 509             value = evaluate(stdscan, NULL, &tokval, NULL, pass0, NULL);
 510             i = tokval.t_type;
 511             if (!value)                  /* Error in evaluator */
 512                 goto fail;
 513             if (!is_simple(value)) {
 514                 nasm_error(ERR_NONFATAL,
 515                       "non-constant argument supplied to TIMES");
 516                 result->times = 1L;
 517             } else {
 518                 result->times = value->value;
 519                 if (value->value < 0 && pass0 == 2) {
 520                     nasm_error(ERR_NONFATAL, "TIMES value %"PRId64" is negative",
 521                           value->value);
 522                     result->times = 0;
 523                 }
 524             }
 525         } else {
 526             int slot = prefix_slot(tokval.t_integer);
 527             if (result->prefixes[slot]) {
 528                if (result->prefixes[slot] == tokval.t_integer)
 529                     nasm_error(ERR_WARNING | ERR_PASS1,
 530                                "instruction has redundant prefixes");
 531                else
 532                     nasm_error(ERR_NONFATAL,
 533                                "instruction has conflicting prefixes");
 534             }
 535             result->prefixes[slot] = tokval.t_integer;
 536             i = stdscan(NULL, &tokval);
 537         }
 538     }
 539
 540     if (i != TOKEN_INSN) {
 541         int j;
 542         enum prefixes pfx;
 543
 544         for (j = 0; j < MAXPREFIX; j++) {
 545             if ((pfx = result->prefixes[j]) != P_none)
 546                 break;
 547         }
 548
 549         if (i == 0 && pfx != P_none) {
 550             /*
 551              * Instruction prefixes are present, but no actual
 552              * instruction. This is allowed: at this point we
 553              * invent a notional instruction of RESB 0.
 554              */
 555             result->opcode          = I_RESB;
 556             result->operands        = 1;
 557             nasm_zero(result->oprs);
 558             result->oprs[0].type    = IMMEDIATE;
 559             result->oprs[0].offset  = 0L;
 560             result->oprs[0].segment = result->oprs[0].wrt = NO_SEG;
 561             return result;
 562         } else {
 563             nasm_error(ERR_NONFATAL, "parser: instruction expected");
 564             goto fail;
 565         }
 566     }
 567
 568     result->opcode = tokval.t_integer;
 569     result->condition = tokval.t_inttwo;
 570
 571     /*
 572      * INCBIN cannot be satisfied with incorrectly
 573      * evaluated operands, since the correct values _must_ be known
 574      * on the first pass. Hence, even in pass one, we set the
 575      * `critical' flag on calling evaluate(), so that it will bomb
 576      * out on undefined symbols.
 577      */
 578     if (result->opcode == I_INCBIN) {
 579         critical = (pass0 < 2 ? 1 : 2);
 580
 581     } else
 582         critical = (pass == 2 ? 2 : 0);
 583
 584     if (result->opcode == I_DB || result->opcode == I_DW ||
 585         result->opcode == I_DD || result->opcode == I_DQ ||
 586         result->opcode == I_DT || result->opcode == I_DO ||
 587         result->opcode == I_DY || result->opcode == I_DZ ||
 588         result->opcode == I_INCBIN) {
 589         extop *eop, **tail = &result->eops, **fixptr;
 590         int oper_num = 0;
 591         int32_t sign;
 592
 593         result->eops_float = false;
 594
 595         /*
 596          * Begin to read the DB/DW/DD/DQ/DT/DO/DY/DZ/INCBIN operands.
 597          */
 598         while (1) {
 599             i = stdscan(NULL, &tokval);
 600             if (i == TOKEN_EOS)
 601                 break;
 602             else if (first && i == ':') {
 603                 insn_is_label = true;
 604                 goto restart_parse;
 605             }
 606             first = false;
 607             fixptr = tail;
 608             eop = *tail = nasm_malloc(sizeof(extop));
 609             tail = &eop->next;
 610             eop->next = NULL;
 611             eop->type = EOT_NOTHING;
 612             oper_num++;
 613             sign = +1;
 614
 615             /*
 616              * is_comma_next() here is to distinguish this from
 617              * a string used as part of an expression...
 618              */
 619             if (i == TOKEN_STR && is_comma_next()) {
 620                 eop->type       = EOT_DB_STRING;
 621                 eop->stringval  = tokval.t_charptr;
 622                 eop->stringlen  = tokval.t_inttwo;
 623                 i = stdscan(NULL, &tokval);     /* eat the comma */
 624             } else if (i == TOKEN_STRFUNC) {
 625                 bool parens = false;
 626                 const char *funcname = tokval.t_charptr;
 627                 enum strfunc func = tokval.t_integer;
 628                 i = stdscan(NULL, &tokval);
 629                 if (i == '(') {
 630                     parens = true;
 631                     i = stdscan(NULL, &tokval);
 632                 }
 633                 if (i != TOKEN_STR) {
 634                     nasm_error(ERR_NONFATAL,
 635                                "%s must be followed by a string constant",
 636                                funcname);
 637                         eop->type = EOT_NOTHING;
 638                 } else {
 639                     eop->type = EOT_DB_STRING_FREE;
 640                     eop->stringlen =
 641                         string_transform(tokval.t_charptr, tokval.t_inttwo,
 642                                          &eop->stringval, func);
 643                     if (eop->stringlen == (size_t)-1) {
 644                         nasm_error(ERR_NONFATAL, "invalid string for transform");
 645                         eop->type = EOT_NOTHING;
 646                     }
 647                 }
 648                 if (parens && i && i != ')') {
 649                     i = stdscan(NULL, &tokval);
 650                     if (i != ')') {
 651                         nasm_error(ERR_NONFATAL, "unterminated %s function",
 652                                    funcname);
 653                     }
 654                 }
 655                 if (i && i != ',')
 656                     i = stdscan(NULL, &tokval);
 657             } else if (i == '-' || i == '+') {
 658                 char *save = stdscan_get();
 659                 int token = i;
 660                 sign = (i == '-') ? -1 : 1;
 661                 i = stdscan(NULL, &tokval);
 662                 if (i != TOKEN_FLOAT) {
 663                     stdscan_set(save);
 664                     i = tokval.t_type = token;
 665                     goto is_expression;
 666                 } else {
 667                     goto is_float;
 668                 }
 669             } else if (i == TOKEN_FLOAT) {
 670 is_float:
 671                 eop->type = EOT_DB_STRING;
 672                 result->eops_float = true;
 673
 674                 eop->stringlen = idata_bytes(result->opcode);
 675                 if (eop->stringlen > 16) {
 676                     nasm_error(ERR_NONFATAL, "floating-point constant"
 677                                " encountered in DY or DZ instruction");
 678                     eop->stringlen = 0;
 679                 } else if (eop->stringlen < 1) {
 680                     nasm_error(ERR_NONFATAL, "floating-point constant"
 681                                " encountered in unknown instruction");
 682                     /*
 683                      * fix suggested by Pedro Gimeno... original line was:
 684                      * eop->type = EOT_NOTHING;
 685                      */
 686                     eop->stringlen = 0;
 687                 }
 688
 689                 eop = nasm_realloc(eop, sizeof(extop) + eop->stringlen);
 690                 tail = &eop->next;
 691                 *fixptr = eop;
 692                 eop->stringval = (char *)eop + sizeof(extop);
 693                 if (!eop->stringlen ||
 694                     !float_const(tokval.t_charptr, sign,
 695                                  (uint8_t *)eop->stringval, eop->stringlen))
 696                     eop->type = EOT_NOTHING;
 697                 i = stdscan(NULL, &tokval); /* eat the comma */
 698             } else {
 699                 /* anything else, assume it is an expression */
 700                 expr *value;
 701
 702 is_expression:
 703                 value = evaluate(stdscan, NULL, &tokval, NULL,
 704                                  critical, NULL);
 705                 i = tokval.t_type;
 706                 if (!value)                  /* Error in evaluator */
 707                     goto fail;
 708                 if (value_to_extop(value, eop, location.segment)) {
 709                     nasm_error(ERR_NONFATAL,
 710                                "operand %d: expression is not simple or relocatable",
 711                                oper_num);
 712                 }
 713             }
 714
 715             /*
 716              * We're about to call stdscan(), which will eat the
 717              * comma that we're currently sitting on between
 718              * arguments. However, we'd better check first that it
 719              * _is_ a comma.
 720              */
 721             if (i == TOKEN_EOS) /* also could be EOL */
 722                 break;
 723             if (i != ',') {
 724                 nasm_error(ERR_NONFATAL, "comma expected after operand %d",
 725                            oper_num);
 726                 goto fail;
 727             }
 728         }
 729
 730         if (result->opcode == I_INCBIN) {
 731             /*
 732              * Correct syntax for INCBIN is that there should be
 733              * one string operand, followed by one or two numeric
 734              * operands.
 735              */
 736             if (!result->eops || result->eops->type != EOT_DB_STRING)
 737                 nasm_error(ERR_NONFATAL, "`incbin' expects a file name");
 738             else if (result->eops->next &&
 739                      result->eops->next->type != EOT_DB_NUMBER)
 740                 nasm_error(ERR_NONFATAL, "`incbin': second parameter is"
 741                            " non-numeric");
 742             else if (result->eops->next && result->eops->next->next &&
 743                      result->eops->next->next->type != EOT_DB_NUMBER)
 744                 nasm_error(ERR_NONFATAL, "`incbin': third parameter is"
 745                            " non-numeric");
 746             else if (result->eops->next && result->eops->next->next &&
 747                      result->eops->next->next->next)
 748                 nasm_error(ERR_NONFATAL,
 749                            "`incbin': more than three parameters");
 750             else
 751                 return result;
 752             /*
 753              * If we reach here, one of the above errors happened.
 754              * Throw the instruction away.
 755              */
 756             goto fail;
 757         } else /* DB ... */ if (oper_num == 0)
 758             nasm_error(ERR_WARNING | ERR_PASS1,
 759                   "no operand for data declaration");
 760         else
 761             result->operands = oper_num;
 762
 763         return result;
 764     }
 765
 766     /*
 767      * Now we begin to parse the operands. There may be up to four
 768      * of these, separated by commas, and terminated by a zero token.
 769      */
 770
 771     for (opnum = 0; opnum < MAX_OPERANDS; opnum++) {
 772         operand *op = &result->oprs[opnum];
 773         expr *value;            /* used most of the time */
 774         bool mref;              /* is this going to be a memory ref? */
 775         bool bracket;           /* is it a [] mref, or a & mref? */
 776         bool mib;               /* compound (mib) mref? */
 777         int setsize = 0;
 778         decoflags_t brace_flags = 0;    /* flags for decorators in braces */
 779
 780         op->disp_size = 0;    /* have to zero this whatever */
 781         op->eaflags   = 0;    /* and this */
 782         op->opflags   = 0;
 783         op->decoflags = 0;
 784
 785         i = stdscan(NULL, &tokval);
 786         if (i == TOKEN_EOS)
 787             break;              /* end of operands: get out of here */
 788         else if (first && i == ':') {
 789             insn_is_label = true;
 790             goto restart_parse;
 791         }
 792         first = false;
 793         op->type = 0; /* so far, no override */
 794         while (i == TOKEN_SPECIAL) {    /* size specifiers */
 795             switch (tokval.t_integer) {
 796             case S_BYTE:
 797                 if (!setsize)   /* we want to use only the first */
 798                     op->type |= BITS8;
 799                 setsize = 1;
 800                 break;
 801             case S_WORD:
 802                 if (!setsize)
 803                     op->type |= BITS16;
 804                 setsize = 1;
 805                 break;
 806             case S_DWORD:
 807             case S_LONG:
 808                 if (!setsize)
 809                     op->type |= BITS32;
 810                 setsize = 1;
 811                 break;
 812             case S_QWORD:
 813                 if (!setsize)
 814                     op->type |= BITS64;
 815                 setsize = 1;
 816                 break;
 817             case S_TWORD:
 818                 if (!setsize)
 819                     op->type |= BITS80;
 820                 setsize = 1;
 821                 break;
 822             case S_OWORD:
 823                 if (!setsize)
 824                     op->type |= BITS128;
 825                 setsize = 1;
 826                 break;
 827             case S_YWORD:
 828                 if (!setsize)
 829                     op->type |= BITS256;
 830                 setsize = 1;
 831                 break;
 832             case S_ZWORD:
 833                 if (!setsize)
 834                     op->type |= BITS512;
 835                 setsize = 1;
 836                 break;
 837             case S_TO:
 838                 op->type |= TO;
 839                 break;
 840             case S_STRICT:
 841                 op->type |= STRICT;
 842                 break;
 843             case S_FAR:
 844                 op->type |= FAR;
 845                 break;
 846             case S_NEAR:
 847                 op->type |= NEAR;
 848                 break;
 849             case S_SHORT:
 850                 op->type |= SHORT;
 851                 break;
 852             default:
 853                 nasm_error(ERR_NONFATAL, "invalid operand size specification");
 854             }
 855             i = stdscan(NULL, &tokval);
 856         }
 857
 858         if (i == '[' || i == '&') {     /* memory reference */
 859             mref = true;
 860             bracket = (i == '[');
 861             i = stdscan(NULL, &tokval); /* then skip the colon */
 862             while (i == TOKEN_SPECIAL || i == TOKEN_PREFIX) {
 863                 process_size_override(result, op);
 864                 i = stdscan(NULL, &tokval);
 865             }
 866             /* when a comma follows an opening bracket - [ , eax*4] */
 867             if (i == ',') {
 868                 /* treat as if there is a zero displacement virtually */
 869                 tokval.t_type = TOKEN_NUM;
 870                 tokval.t_integer = 0;
 871                 stdscan_set(stdscan_get() - 1);     /* rewind the comma */
 872             }
 873         } else {                /* immediate operand, or register */
 874             mref = false;
 875             bracket = false;    /* placate optimisers */
 876         }
 877
 878         if ((op->type & FAR) && !mref &&
 879             result->opcode != I_JMP && result->opcode != I_CALL) {
 880             nasm_error(ERR_NONFATAL, "invalid use of FAR operand specifier");
 881         }
 882
 883         value = evaluate(stdscan, NULL, &tokval,
 884                          &op->opflags, critical, &hints);
 885         i = tokval.t_type;
 886         if (op->opflags & OPFLAG_FORWARD) {
 887             result->forw_ref = true;
 888         }
 889         if (!value)                  /* Error in evaluator */
 890             goto fail;
 891         if (i == ':' && mref) { /* it was seg:offset */
 892             /*
 893              * Process the segment override.
 894              */
 895             if (value[1].type   != 0    ||
 896                 value->value    != 1    ||
 897                 !IS_SREG(value->type))
 898                 nasm_error(ERR_NONFATAL, "invalid segment override");
 899             else if (result->prefixes[PPS_SEG])
 900                 nasm_error(ERR_NONFATAL,
 901                       "instruction has conflicting segment overrides");
 902             else {
 903                 result->prefixes[PPS_SEG] = value->type;
 904                 if (IS_FSGS(value->type))
 905                     op->eaflags |= EAF_FSGS;
 906             }
 907
 908             i = stdscan(NULL, &tokval); /* then skip the colon */
 909             while (i == TOKEN_SPECIAL || i == TOKEN_PREFIX) {
 910                 process_size_override(result, op);
 911                 i = stdscan(NULL, &tokval);
 912             }
 913             value = evaluate(stdscan, NULL, &tokval,
 914                              &op->opflags, critical, &hints);
 915             i = tokval.t_type;
 916             if (op->opflags & OPFLAG_FORWARD) {
 917                 result->forw_ref = true;
 918             }
 919             /* and get the offset */
 920             if (!value)                  /* Error in evaluator */
 921                 goto fail;
 922         }
 923
 924         mib = false;
 925         if (mref && bracket && i == ',') {
 926             /* [seg:base+offset,index*scale] syntax (mib) */
 927
 928             operand o1, o2;     /* Partial operands */
 929
 930             if (parse_mref(&o1, value))
 931                 goto fail;
 932
 933             i = stdscan(NULL, &tokval); /* Eat comma */
 934             value = evaluate(stdscan, NULL, &tokval, &op->opflags,
 935                              critical, &hints);
 936             i = tokval.t_type;
 937             if (!value)
 938                 goto fail;
 939
 940             if (parse_mref(&o2, value))
 941                 goto fail;
 942
 943             if (o2.basereg != -1 && o2.indexreg == -1) {
 944                 o2.indexreg = o2.basereg;
 945                 o2.scale = 1;
 946                 o2.basereg = -1;
 947             }
 948
 949             if (o1.indexreg != -1 || o2.basereg != -1 || o2.offset != 0 ||
 950                 o2.segment != NO_SEG || o2.wrt != NO_SEG) {
 951                 nasm_error(ERR_NONFATAL, "invalid mib expression");
 952                 goto fail;
 953             }
 954
 955             op->basereg = o1.basereg;
 956             op->indexreg = o2.indexreg;
 957             op->scale = o2.scale;
 958             op->offset = o1.offset;
 959             op->segment = o1.segment;
 960             op->wrt = o1.wrt;
 961
 962             if (op->basereg != -1) {
 963                 op->hintbase = op->basereg;
 964                 op->hinttype = EAH_MAKEBASE;
 965             } else if (op->indexreg != -1) {
 966                 op->hintbase = op->indexreg;
 967                 op->hinttype = EAH_NOTBASE;
 968             } else {
 969                 op->hintbase = -1;
 970                 op->hinttype = EAH_NOHINT;
 971             }
 972
 973             mib = true;
 974         }
 975
 976         recover = false;
 977         if (mref && bracket) {  /* find ] at the end */
 978             if (i != ']') {
 979                 nasm_error(ERR_NONFATAL, "parser: expecting ]");
 980                 recover = true;
 981             } else {            /* we got the required ] */
 982                 i = stdscan(NULL, &tokval);
 983                 if (i == TOKEN_DECORATOR || i == TOKEN_OPMASK) {
 984                     /* parse opmask (and zeroing) after an operand */
 985                     recover = parse_braces(&brace_flags);
 986                     i = tokval.t_type;
 987                 }
 988                 if (i != 0 && i != ',') {
 989                     nasm_error(ERR_NONFATAL, "comma or end of line expected");
 990                     recover = true;
 991                 }
 992             }
 993         } else {                /* immediate operand */
 994             if (i != 0 && i != ',' && i != ':' &&
 995                 i != TOKEN_DECORATOR && i != TOKEN_OPMASK) {
 996                 nasm_error(ERR_NONFATAL, "comma, colon, decorator or end of "
 997                                          "line expected after operand");
 998                 recover = true;
 999             } else if (i == ':') {
1000                 op->type |= COLON;
1001             } else if (i == TOKEN_DECORATOR || i == TOKEN_OPMASK) {
1002                 /* parse opmask (and zeroing) after an operand */
1003                 recover = parse_braces(&brace_flags);
1004             }
1005         }
1006         if (recover) {
1007             do {                /* error recovery */
1008                 i = stdscan(NULL, &tokval);
1009             } while (i != 0 && i != ',');
1010         }
1011
1012         /*
1013          * now convert the exprs returned from evaluate()
1014          * into operand descriptions...
1015          */
1016         op->decoflags |= brace_flags;
1017
1018         if (mref) {             /* it's a memory reference */
1019             /* A mib reference was fully parsed already */
1020             if (!mib) {
1021                 if (parse_mref(op, value))
1022                     goto fail;
1023                 op->hintbase = hints.base;
1024                 op->hinttype = hints.type;
1025             }
1026             mref_set_optype(op);
1027         } else {                /* it's not a memory reference */
1028             if (is_just_unknown(value)) {       /* it's immediate but unknown */
1029                 op->type      |= IMMEDIATE;
1030                 op->opflags   |= OPFLAG_UNKNOWN;
1031                 op->offset    = 0;        /* don't care */
1032                 op->segment   = NO_SEG;   /* don't care again */
1033                 op->wrt       = NO_SEG;   /* still don't care */
1034
1035                 if(optimizing >= 0 && !(op->type & STRICT)) {
1036                     /* Be optimistic */
1037                     op->type |=
1038                         UNITY | SBYTEWORD | SBYTEDWORD | UDWORD | SDWORD;
1039                 }
1040             } else if (is_reloc(value)) {       /* it's immediate */
1041                 uint64_t n = reloc_value(value);
1042
1043                 op->type      |= IMMEDIATE;
1044                 op->offset    = n;
1045                 op->segment   = reloc_seg(value);
1046                 op->wrt       = reloc_wrt(value);
1047                 op->opflags   |= is_self_relative(value) ? OPFLAG_RELATIVE : 0;
1048
1049                 if (is_simple(value)) {
1050                     if (n == 1)
1051                         op->type |= UNITY;
1052                     if (optimizing >= 0 && !(op->type & STRICT)) {
1053                         if ((uint32_t) (n + 128) <= 255)
1054                             op->type |= SBYTEDWORD;
1055                         if ((uint16_t) (n + 128) <= 255)
1056                             op->type |= SBYTEWORD;
1057                         if (n <= UINT64_C(0xFFFFFFFF))
1058                             op->type |= UDWORD;
1059                         if (n + UINT64_C(0x80000000) <= UINT64_C(0xFFFFFFFF))
1060                             op->type |= SDWORD;
1061                     }
1062                 }
1063             } else if (value->type == EXPR_RDSAE) {
1064                 /*
1065                  * it's not an operand but a rounding or SAE decorator.
1066                  * put the decorator information in the (opflag_t) type field
1067                  * of previous operand.
1068                  */
1069                 opnum--; op--;
1070                 switch (value->value) {
1071                 case BRC_RN:
1072                 case BRC_RU:
1073                 case BRC_RD:
1074                 case BRC_RZ:
1075                 case BRC_SAE:
1076                     op->decoflags |= (value->value == BRC_SAE ? SAE : ER);
1077                     result->evex_rm = value->value;
1078                     break;
1079                 default:
1080                     nasm_error(ERR_NONFATAL, "invalid decorator");
1081                     break;
1082                 }
1083             } else {            /* it's a register */
1084                 opflags_t rs;
1085
1086                 if (value->type >= EXPR_SIMPLE || value->value != 1) {
1087                     nasm_error(ERR_NONFATAL, "invalid operand type");
1088                     goto fail;
1089                 }
1090
1091                 /*
1092                  * check that its only 1 register, not an expression...
1093                  */
1094                 for (i = 1; value[i].type; i++)
1095                     if (value[i].value) {
1096                         nasm_error(ERR_NONFATAL, "invalid operand type");
1097                         goto fail;
1098                     }
1099
1100                 /* clear overrides, except TO which applies to FPU regs */
1101                 if (op->type & ~TO) {
1102                     /*
1103                      * we want to produce a warning iff the specified size
1104                      * is different from the register size
1105                      */
1106                     rs = op->type & SIZE_MASK;
1107                 } else
1108                     rs = 0;
1109
1110                 op->type      &= TO;
1111                 op->type      |= REGISTER;
1112                 op->type      |= nasm_reg_flags[value->type];
1113                 op->decoflags |= brace_flags;
1114                 op->basereg   = value->type;
1115
1116                 if (rs && (op->type & SIZE_MASK) != rs)
1117                     nasm_error(ERR_WARNING | ERR_PASS1,
1118                           "register size specification ignored");
1119             }
1120         }
1121
1122         /* remember the position of operand having broadcasting/ER mode */
1123         if (op->decoflags & (BRDCAST_MASK | ER | SAE))
1124             result->evex_brerop = opnum;
1125     }
1126
1127     result->operands = opnum; /* set operand count */
1128
1129     /* clear remaining operands */
1130     while (opnum < MAX_OPERANDS)
1131         result->oprs[opnum++].type = 0;
1132
1133     /*
1134      * Transform RESW, RESD, RESQ, REST, RESO, RESY, RESZ into RESB.
1135      */
1136     switch (result->opcode) {
1137     case I_RESW:
1138         result->opcode = I_RESB;
1139         result->oprs[0].offset *= 2;
1140         break;
1141     case I_RESD:
1142         result->opcode = I_RESB;
1143         result->oprs[0].offset *= 4;
1144         break;
1145     case I_RESQ:
1146         result->opcode = I_RESB;
1147         result->oprs[0].offset *= 8;
1148         break;
1149     case I_REST:
1150         result->opcode = I_RESB;
1151         result->oprs[0].offset *= 10;
1152         break;
1153     case I_RESO:
1154         result->opcode = I_RESB;
1155         result->oprs[0].offset *= 16;
1156         break;
1157     case I_RESY:
1158         result->opcode = I_RESB;
1159         result->oprs[0].offset *= 32;
1160         break;
1161     case I_RESZ:
1162         result->opcode = I_RESB;
1163         result->oprs[0].offset *= 64;
1164         break;
1165     default:
1166         break;
1167     }
1168
1169     return result;
1170
1171 fail:
1172     result->opcode = I_none;
1173     return result;
1174 }
1175
1176 static int is_comma_next(void)
1177 {
1178     struct tokenval tv;
1179     char *p;
1180     int i;
1181
1182     p = stdscan_get();
1183     i = stdscan(NULL, &tv);
1184     stdscan_set(p);
1185
1186     return (i == ',' || i == ';' || !i);
1187 }
1188
1189 void cleanup_insn(insn * i)
1190 {
1191     extop *e;
1192
1193     while ((e = i->eops)) {
1194         i->eops = e->next;
1195         if (e->type == EOT_DB_STRING_FREE)
1196             nasm_free(e->stringval);
1197         nasm_free(e);
1198     }
1199 }