1/* ----------------------------------------------------------------------- *
2 *
3 * Copyright 1996-2018 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 * nasm.h main header file for the Netwide Assembler: inter-module interface
36 */
37
38#ifndef NASM_NASM_H
39#define NASM_NASM_H
40
41#include "compiler.h"
42
43#include <stdio.h>
44#include <time.h>
45
46#include "nasmlib.h"
47#include "strlist.h"
48#include "preproc.h"
49#include "insnsi.h" /* For enum opcode */
50#include "directiv.h" /* For enum directive */
51#include "labels.h" /* For enum mangle_index, enum label_type */
52#include "opflags.h"
53#include "regs.h"
54
55/* Time stamp for the official start of compilation */
56struct compile_time {
57 time_t t;
58 bool have_local, have_gm, have_posix;
59 int64_t posix;
60 struct tm local;
61 struct tm gm;
62};
63extern struct compile_time official_compile_time;
64
65#define NO_SEG INT32_C(-1) /* null segment value */
66#define SEG_ABS 0x40000000L /* mask for far-absolute segments */
67
68#define IDLEN_MAX 4096
69#define DECOLEN_MAX 32
70
71/*
72 * Name pollution problems: <time.h> on Digital UNIX pulls in some
73 * strange hardware header file which sees fit to define R_SP. We
74 * undefine it here so as not to break the enum below.
75 */
76#ifdef R_SP
77#undef R_SP
78#endif
79
80/*
81 * We must declare the existence of this structure type up here,
82 * since we have to reference it before we define it...
83 */
84struct ofmt;
85
86/*
87 * Values for the `type' parameter to an output function.
88 */
89enum out_type {
90 OUT_RAWDATA, /* Plain bytes */
91 OUT_RESERVE, /* Reserved bytes (RESB et al) */
92 OUT_ZERODATA, /* Initialized data, but all zero */
93 OUT_ADDRESS, /* An address (symbol value) */
94 OUT_RELADDR, /* A relative address */
95 OUT_SEGMENT, /* A segment number */
96
97 /*
98 * These values are used by the legacy backend interface only;
99 * see output/legacy.c for more information. These should never
100 * be used otherwise. Once all backends have been migrated to the
101 * new interface they should be removed.
102 */
103 OUT_REL1ADR,
104 OUT_REL2ADR,
105 OUT_REL4ADR,
106 OUT_REL8ADR
107};
108
109enum out_sign {
110 OUT_WRAP, /* Undefined signedness (wraps) */
111 OUT_SIGNED, /* Value is signed */
112 OUT_UNSIGNED /* Value is unsigned */
113};
114
115/*
116 * The data we send down to the backend.
117 * XXX: We still want to push down the base address symbol if
118 * available, and replace the segment numbers with a structure.
119 */
120struct out_data {
121 int64_t offset; /* Offset within segment */
122 int32_t segment; /* Segment written to */
123 enum out_type type; /* See above */
124 enum out_sign sign; /* See above */
125 int inslen; /* Length of instruction */
126 int insoffs; /* Offset inside instruction */
127 int bits; /* Bits mode of compilation */
128 uint64_t size; /* Size of output */
129 const struct itemplate *itemp; /* Instruction template */
130 const void *data; /* Data for OUT_RAWDATA */
131 uint64_t toffset; /* Target address offset for relocation */
132 int32_t tsegment; /* Target segment for relocation */
133 int32_t twrt; /* Relocation with respect to */
134 int64_t relbase; /* Relative base for OUT_RELADDR */
135};
136
137/*
138 * And a label-definition function. The boolean parameter
139 * `is_norm' states whether the label is a `normal' label (which
140 * should affect the local-label system), or something odder like
141 * an EQU or a segment-base symbol, which shouldn't.
142 */
143typedef void (*ldfunc)(char *label, int32_t segment, int64_t offset,
144 char *special, bool is_norm);
145
146/*
147 * Token types returned by the scanner, in addition to ordinary
148 * ASCII character values, and zero for end-of-string.
149 */
150enum token_type { /* token types, other than chars */
151 TOKEN_INVALID = -1, /* a placeholder value */
152 TOKEN_EOS = 0, /* end of string */
153 TOKEN_EQ = '=',
154 TOKEN_GT = '>',
155 TOKEN_LT = '<', /* aliases */
156 TOKEN_ID = 256, /* identifier */
157 TOKEN_NUM, /* numeric constant */
158 TOKEN_ERRNUM, /* malformed numeric constant */
159 TOKEN_STR, /* string constant */
160 TOKEN_ERRSTR, /* unterminated string constant */
161 TOKEN_FLOAT, /* floating-point constant */
162 TOKEN_REG, /* register name */
163 TOKEN_INSN, /* instruction name */
164 TOKEN_HERE, /* $ */
165 TOKEN_BASE, /* $$ */
166 TOKEN_SPECIAL, /* BYTE, WORD, DWORD, QWORD, FAR, NEAR, etc */
167 TOKEN_PREFIX, /* A32, O16, LOCK, REPNZ, TIMES, etc */
168 TOKEN_SHL, /* << */
169 TOKEN_SHR, /* >> */
170 TOKEN_SDIV, /* // */
171 TOKEN_SMOD, /* %% */
172 TOKEN_GE, /* >= */
173 TOKEN_LE, /* <= */
174 TOKEN_NE, /* <> (!= is same as <>) */
175 TOKEN_DBL_AND, /* && */
176 TOKEN_DBL_OR, /* || */
177 TOKEN_DBL_XOR, /* ^^ */
178 TOKEN_SEG, /* SEG */
179 TOKEN_WRT, /* WRT */
180 TOKEN_FLOATIZE, /* __floatX__ */
181 TOKEN_STRFUNC, /* __utf16*__, __utf32*__ */
182 TOKEN_IFUNC, /* __ilog2*__ */
183 TOKEN_DECORATOR, /* decorators such as {...} */
184 TOKEN_OPMASK /* translated token for opmask registers */
185};
186
187enum floatize {
188 FLOAT_8,
189 FLOAT_16,
190 FLOAT_32,
191 FLOAT_64,
192 FLOAT_80M,
193 FLOAT_80E,
194 FLOAT_128L,
195 FLOAT_128H
196};
197
198/* Must match the list in string_transform(), in strfunc.c */
199enum strfunc {
200 STRFUNC_UTF16,
201 STRFUNC_UTF16LE,
202 STRFUNC_UTF16BE,
203 STRFUNC_UTF32,
204 STRFUNC_UTF32LE,
205 STRFUNC_UTF32BE
206};
207
208enum ifunc {
209 IFUNC_ILOG2E,
210 IFUNC_ILOG2W,
211 IFUNC_ILOG2F,
212 IFUNC_ILOG2C
213};
214
215size_t string_transform(char *, size_t, char **, enum strfunc);
216
217/*
218 * The expression evaluator must be passed a scanner function; a
219 * standard scanner is provided as part of nasmlib.c. The
220 * preprocessor will use a different one. Scanners, and the
221 * token-value structures they return, look like this.
222 *
223 * The return value from the scanner is always a copy of the
224 * `t_type' field in the structure.
225 */
226struct tokenval {
227 char *t_charptr;
228 int64_t t_integer;
229 int64_t t_inttwo;
230 enum token_type t_type;
231 int8_t t_flag;
232};
233typedef int (*scanner)(void *private_data, struct tokenval *tv);
234
235struct location {
236 int64_t offset;
237 int32_t segment;
238 int known;
239};
240extern struct location location;
241
242/*
243 * Expression-evaluator datatype. Expressions, within the
244 * evaluator, are stored as an array of these beasts, terminated by
245 * a record with type==0. Mostly, it's a vector type: each type
246 * denotes some kind of a component, and the value denotes the
247 * multiple of that component present in the expression. The
248 * exception is the WRT type, whose `value' field denotes the
249 * segment to which the expression is relative. These segments will
250 * be segment-base types, i.e. either odd segment values or SEG_ABS
251 * types. So it is still valid to assume that anything with a
252 * `value' field of zero is insignificant.
253 */
254typedef struct {
255 int32_t type; /* a register, or EXPR_xxx */
256 int64_t value; /* must be >= 32 bits */
257} expr;
258
259/*
260 * Library routines to manipulate expression data types.
261 */
262bool is_reloc(const expr *vect);
263bool is_simple(const expr *vect);
264bool is_really_simple(const expr *vect);
265bool is_unknown(const expr *vect);
266bool is_just_unknown(const expr *vect);
267int64_t reloc_value(const expr *vect);
268int32_t reloc_seg(const expr *vect);
269int32_t reloc_wrt(const expr *vect);
270bool is_self_relative(const expr *vect);
271void dump_expr(const expr *vect);
272
273/*
274 * The evaluator can also return hints about which of two registers
275 * used in an expression should be the base register. See also the
276 * `operand' structure.
277 */
278struct eval_hints {
279 int64_t base;
280 int type;
281};
282
283/*
284 * The actual expression evaluator function looks like this. When
285 * called, it expects the first token of its expression to already
286 * be in `*tv'; if it is not, set tv->t_type to TOKEN_INVALID and
287 * it will start by calling the scanner.
288 *
289 * If a forward reference happens during evaluation, the evaluator
290 * must set `*fwref' to true if `fwref' is non-NULL.
291 *
292 * `critical' is non-zero if the expression may not contain forward
293 * references. The evaluator will report its own error if this
294 * occurs; if `critical' is 1, the error will be "symbol not
295 * defined before use", whereas if `critical' is 2, the error will
296 * be "symbol undefined".
297 *
298 * If `critical' has bit 8 set (in addition to its main value: 0x101
299 * and 0x102 correspond to 1 and 2) then an extended expression
300 * syntax is recognised, in which relational operators such as =, <
301 * and >= are accepted, as well as low-precedence logical operators
302 * &&, ^^ and ||.
303 *
304 * If `hints' is non-NULL, it gets filled in with some hints as to
305 * the base register in complex effective addresses.
306 */
307#define CRITICAL 0x100
308typedef expr *(*evalfunc)(scanner sc, void *scprivate,
309 struct tokenval *tv, int *fwref, int critical,
310 struct eval_hints *hints);
311
312/*
313 * Special values for expr->type.
314 * These come after EXPR_REG_END as defined in regs.h.
315 * Expr types : 0 ~ EXPR_REG_END, EXPR_UNKNOWN, EXPR_...., EXPR_RDSAE,
316 * EXPR_SEGBASE ~ EXPR_SEGBASE + SEG_ABS, ...
317 */
318#define EXPR_UNKNOWN (EXPR_REG_END+1) /* forward references */
319#define EXPR_SIMPLE (EXPR_REG_END+2)
320#define EXPR_WRT (EXPR_REG_END+3)
321#define EXPR_RDSAE (EXPR_REG_END+4)
322#define EXPR_SEGBASE (EXPR_REG_END+5)
323
324/*
325 * preprocessors ought to look like this:
326 */
327struct preproc_ops {
328 /*
329 * Called once at the very start of assembly.
330 */
331 void (*init)(void);
332
333 /*
334 * Called at the start of a pass; given a file name, the number
335 * of the pass, an error reporting function, an evaluator
336 * function, and a listing generator to talk to.
337 */
338 void (*reset)(const char *file, int pass, StrList **deplist);
339
340 /*
341 * Called to fetch a line of preprocessed source. The line
342 * returned has been malloc'ed, and so should be freed after
343 * use.
344 */
345 char *(*getline)(void);
346
347 /* Called at the end of a pass */
348 void (*cleanup)(int pass);
349
350 /* Additional macros specific to output format */
351 void (*extra_stdmac)(macros_t *macros);
352
353 /* Early definitions and undefinitions for macros */
354 void (*pre_define)(char *definition);
355 void (*pre_undefine)(char *definition);
356
357 /* Include file from command line */
358 void (*pre_include)(char *fname);
359
360 /* Add a command from the command line */
361 void (*pre_command)(const char *what, char *str);
362
363 /* Include path from command line */
364 void (*include_path)(char *path);
365
366 /* Unwind the macro stack when printing an error message */
367 void (*error_list_macros)(int severity);
368};
369
370extern const struct preproc_ops nasmpp;
371extern const struct preproc_ops preproc_nop;
372
373/* List of dependency files */
374extern StrList *depend_list;
375
376/*
377 * Some lexical properties of the NASM source language, included
378 * here because they are shared between the parser and preprocessor.
379 */
380
381/*
382 * isidstart matches any character that may start an identifier, and isidchar
383 * matches any character that may appear at places other than the start of an
384 * identifier. E.g. a period may only appear at the start of an identifier
385 * (for local labels), whereas a number may appear anywhere *but* at the
386 * start.
387 * isbrcchar matches any character that may placed inside curly braces as a
388 * decorator. E.g. {rn-sae}, {1to8}, {k1}{z}
389 */
390
391#define isidstart(c) (nasm_isalpha(c) || \
392 (c) == '_' || \
393 (c) == '.' || \
394 (c) == '?' || \
395 (c) == '@')
396
397#define isidchar(c) (isidstart(c) || \
398 nasm_isdigit(c) || \
399 (c) == '$' || \
400 (c) == '#' || \
401 (c) == '~')
402
403#define isbrcchar(c) (isidchar(c) || \
404 (c) == '-')
405
406/* Ditto for numeric constants. */
407
408#define isnumstart(c) (nasm_isdigit(c) || (c) == '$')
409#define isnumchar(c) (nasm_isalnum(c) || (c) == '_')
410
411/*
412 * inline function to skip past an identifier; returns the first character past
413 * the identifier if valid, otherwise NULL.
414 */
415static inline char *nasm_skip_identifier(const char *str)
416{
417 const char *p = str;
418
419 if (!isidstart(*p++)) {
420 p = NULL;
421 } else {
422 while (isidchar(*p++))
423 ;
424 }
425 return (char *)p;
426}
427
428/*
429 * Data-type flags that get passed to listing-file routines.
430 */
431enum {
432 LIST_READ,
433 LIST_MACRO,
434 LIST_MACRO_NOLIST,
435 LIST_INCLUDE,
436 LIST_INCBIN,
437 LIST_TIMES
438};
439
440/*
441 * -----------------------------------------------------------
442 * Format of the `insn' structure returned from `parser.c' and
443 * passed into `assemble.c'
444 * -----------------------------------------------------------
445 */
446
447/* Verify value to be a valid register */
448static inline bool is_register(int reg)
449{
450 return reg >= EXPR_REG_START && reg < REG_ENUM_LIMIT;
451}
452
453enum ccode { /* condition code names */
454 C_A, C_AE, C_B, C_BE, C_C, C_E, C_G, C_GE, C_L, C_LE, C_NA, C_NAE,
455 C_NB, C_NBE, C_NC, C_NE, C_NG, C_NGE, C_NL, C_NLE, C_NO, C_NP,
456 C_NS, C_NZ, C_O, C_P, C_PE, C_PO, C_S, C_Z,
457 C_none = -1
458};
459
460/*
461 * token flags
462 */
463#define TFLAG_BRC (1 << 0) /* valid only with braces. {1to8}, {rd-sae}, ...*/
464#define TFLAG_BRC_OPT (1 << 1) /* may or may not have braces. opmasks {k1} */
465#define TFLAG_BRC_ANY (TFLAG_BRC | TFLAG_BRC_OPT)
466#define TFLAG_BRDCAST (1 << 2) /* broadcasting decorator */
467#define TFLAG_WARN (1 << 3) /* warning only, treat as ID */
468
469static inline uint8_t get_cond_opcode(enum ccode c)
470{
471 static const uint8_t ccode_opcodes[] = {
472 0x7, 0x3, 0x2, 0x6, 0x2, 0x4, 0xf, 0xd, 0xc, 0xe, 0x6, 0x2,
473 0x3, 0x7, 0x3, 0x5, 0xe, 0xc, 0xd, 0xf, 0x1, 0xb, 0x9, 0x5,
474 0x0, 0xa, 0xa, 0xb, 0x8, 0x4
475 };
476
477 return ccode_opcodes[(int)c];
478}
479
480/*
481 * REX flags
482 */
483#define REX_MASK 0x4f /* Actual REX prefix bits */
484#define REX_B 0x01 /* ModRM r/m extension */
485#define REX_X 0x02 /* SIB index extension */
486#define REX_R 0x04 /* ModRM reg extension */
487#define REX_W 0x08 /* 64-bit operand size */
488#define REX_L 0x20 /* Use LOCK prefix instead of REX.R */
489#define REX_P 0x40 /* REX prefix present/required */
490#define REX_H 0x80 /* High register present, REX forbidden */
491#define REX_V 0x0100 /* Instruction uses VEX/XOP instead of REX */
492#define REX_NH 0x0200 /* Instruction which doesn't use high regs */
493#define REX_EV 0x0400 /* Instruction uses EVEX instead of REX */
494
495/*
496 * EVEX bit field
497 */
498#define EVEX_P0MM 0x0f /* EVEX P[3:0] : Opcode map */
499#define EVEX_P0RP 0x10 /* EVEX P[4] : High-16 reg */
500#define EVEX_P0X 0x40 /* EVEX P[6] : High-16 rm */
501#define EVEX_P1PP 0x03 /* EVEX P[9:8] : Legacy prefix */
502#define EVEX_P1VVVV 0x78 /* EVEX P[14:11] : NDS register */
503#define EVEX_P1W 0x80 /* EVEX P[15] : Osize extension */
504#define EVEX_P2AAA 0x07 /* EVEX P[18:16] : Embedded opmask */
505#define EVEX_P2VP 0x08 /* EVEX P[19] : High-16 NDS reg */
506#define EVEX_P2B 0x10 /* EVEX P[20] : Broadcast / RC / SAE */
507#define EVEX_P2LL 0x60 /* EVEX P[22:21] : Vector length */
508#define EVEX_P2RC EVEX_P2LL /* EVEX P[22:21] : Rounding control */
509#define EVEX_P2Z 0x80 /* EVEX P[23] : Zeroing/Merging */
510
511/*
512 * REX_V "classes" (prefixes which behave like VEX)
513 */
514enum vex_class {
515 RV_VEX = 0, /* C4/C5 */
516 RV_XOP = 1, /* 8F */
517 RV_EVEX = 2 /* 62 */
518};
519
520/*
521 * Note that because segment registers may be used as instruction
522 * prefixes, we must ensure the enumerations for prefixes and
523 * register names do not overlap.
524 */
525enum prefixes { /* instruction prefixes */
526 P_none = 0,
527 PREFIX_ENUM_START = REG_ENUM_LIMIT,
528 P_A16 = PREFIX_ENUM_START,
529 P_A32,
530 P_A64,
531 P_ASP,
532 P_LOCK,
533 P_O16,
534 P_O32,
535 P_O64,
536 P_OSP,
537 P_REP,
538 P_REPE,
539 P_REPNE,
540 P_REPNZ,
541 P_REPZ,
542 P_TIMES,
543 P_WAIT,
544 P_XACQUIRE,
545 P_XRELEASE,
546 P_BND,
547 P_NOBND,
548 P_EVEX,
549 P_VEX3,
550 P_VEX2,
551 PREFIX_ENUM_LIMIT
552};
553
554enum extop_type { /* extended operand types */
555 EOT_NOTHING,
556 EOT_DB_STRING, /* Byte string */
557 EOT_DB_STRING_FREE, /* Byte string which should be nasm_free'd*/
558 EOT_DB_NUMBER /* Integer */
559};
560
561enum ea_flags { /* special EA flags */
562 EAF_BYTEOFFS = 1, /* force offset part to byte size */
563 EAF_WORDOFFS = 2, /* force offset part to [d]word size */
564 EAF_TIMESTWO = 4, /* really do EAX*2 not EAX+EAX */
565 EAF_REL = 8, /* IP-relative addressing */
566 EAF_ABS = 16, /* non-IP-relative addressing */
567 EAF_FSGS = 32, /* fs/gs segment override present */
568 EAF_MIB = 64 /* mib operand */
569};
570
571enum eval_hint { /* values for `hinttype' */
572 EAH_NOHINT = 0, /* no hint at all - our discretion */
573 EAH_MAKEBASE = 1, /* try to make given reg the base */
574 EAH_NOTBASE = 2, /* try _not_ to make reg the base */
575 EAH_SUMMED = 3 /* base and index are summed into index */
576};
577
578typedef struct operand { /* operand to an instruction */
579 opflags_t type; /* type of operand */
580 int disp_size; /* 0 means default; 16; 32; 64 */
581 enum reg_enum basereg;
582 enum reg_enum indexreg; /* address registers */
583 int scale; /* index scale */
584 int hintbase;
585 enum eval_hint hinttype; /* hint as to real base register */
586 int32_t segment; /* immediate segment, if needed */
587 int64_t offset; /* any immediate number */
588 int32_t wrt; /* segment base it's relative to */
589 int eaflags; /* special EA flags */
590 int opflags; /* see OPFLAG_* defines below */
591 decoflags_t decoflags; /* decorator flags such as {...} */
592} operand;
593
594#define OPFLAG_FORWARD 1 /* operand is a forward reference */
595#define OPFLAG_EXTERN 2 /* operand is an external reference */
596#define OPFLAG_UNKNOWN 4 /* operand is an unknown reference
597 (always a forward reference also) */
598#define OPFLAG_RELATIVE 8 /* operand is self-relative, e.g. [foo - $]
599 where foo is not in the current segment */
600
601typedef struct extop { /* extended operand */
602 struct extop *next; /* linked list */
603 char *stringval; /* if it's a string, then here it is */
604 size_t stringlen; /* ... and here's how long it is */
605 int64_t offset; /* ... it's given here ... */
606 int32_t segment; /* if it's a number/address, then... */
607 int32_t wrt; /* ... and here */
608 bool relative; /* self-relative expression */
609 enum extop_type type; /* defined above */
610} extop;
611
612enum ea_type {
613 EA_INVALID, /* Not a valid EA at all */
614 EA_SCALAR, /* Scalar EA */
615 EA_XMMVSIB, /* XMM vector EA */
616 EA_YMMVSIB, /* YMM vector EA */
617 EA_ZMMVSIB /* ZMM vector EA */
618};
619
620/*
621 * Prefix positions: each type of prefix goes in a specific slot.
622 * This affects the final ordering of the assembled output, which
623 * shouldn't matter to the processor, but if you have stylistic
624 * preferences, you can change this. REX prefixes are handled
625 * differently for the time being.
626 *
627 * LOCK and REP used to be one slot; this is no longer the case since
628 * the introduction of HLE.
629 */
630enum prefix_pos {
631 PPS_WAIT, /* WAIT (technically not a prefix!) */
632 PPS_REP, /* REP/HLE prefix */
633 PPS_LOCK, /* LOCK prefix */
634 PPS_SEG, /* Segment override prefix */
635 PPS_OSIZE, /* Operand size prefix */
636 PPS_ASIZE, /* Address size prefix */
637 PPS_VEX, /* VEX type */
638 MAXPREFIX /* Total number of prefix slots */
639};
640
641/*
642 * Tuple types that are used when determining Disp8*N eligibility
643 * The order must match with a hash %tuple_codes in insns.pl
644 */
645enum ttypes {
646 FV = 001,
647 HV = 002,
648 FVM = 003,
649 T1S8 = 004,
650 T1S16 = 005,
651 T1S = 006,
652 T1F32 = 007,
653 T1F64 = 010,
654 T2 = 011,
655 T4 = 012,
656 T8 = 013,
657 HVM = 014,
658 QVM = 015,
659 OVM = 016,
660 M128 = 017,
661 DUP = 020
662};
663
664/* EVEX.L'L : Vector length on vector insns */
665enum vectlens {
666 VL128 = 0,
667 VL256 = 1,
668 VL512 = 2,
669 VLMAX = 3
670};
671
672/* If you need to change this, also change it in insns.pl */
673#define MAX_OPERANDS 5
674
675typedef struct insn { /* an instruction itself */
676 char *label; /* the label defined, or NULL */
677 int prefixes[MAXPREFIX]; /* instruction prefixes, if any */
678 enum opcode opcode; /* the opcode - not just the string */
679 enum ccode condition; /* the condition code, if Jcc/SETcc */
680 int operands; /* how many operands? 0-3 (more if db et al) */
681 int addr_size; /* address size */
682 operand oprs[MAX_OPERANDS]; /* the operands, defined as above */
683 extop *eops; /* extended operands */
684 int eops_float; /* true if DD and floating */
685 int32_t times; /* repeat count (TIMES prefix) */
686 bool forw_ref; /* is there a forward reference? */
687 bool rex_done; /* REX prefix emitted? */
688 int rex; /* Special REX Prefix */
689 int vexreg; /* Register encoded in VEX prefix */
690 int vex_cm; /* Class and M field for VEX prefix */
691 int vex_wlp; /* W, P and L information for VEX prefix */
692 uint8_t evex_p[3]; /* EVEX.P0: [RXB,R',00,mm], P1: [W,vvvv,1,pp] */
693 /* EVEX.P2: [z,L'L,b,V',aaa] */
694 enum ttypes evex_tuple; /* Tuple type for compressed Disp8*N */
695 int evex_rm; /* static rounding mode for AVX512 (EVEX) */
696 int8_t evex_brerop; /* BR/ER/SAE operand position */
697} insn;
698
699/* Instruction flags type: IF_* flags are defined in insns.h */
700typedef uint64_t iflags_t;
701
702/*
703 * What to return from a directive- or pragma-handling function.
704 * Currently DIRR_OK and DIRR_ERROR are treated the same way;
705 * in both cases the backend is expected to produce the appropriate
706 * error message on its own.
707 *
708 * DIRR_BADPARAM causes a generic error message to be printed. Note
709 * that it is an error, not a warning, even in the case of pragmas;
710 * don't use it where forward compatiblity would be compromised
711 * (instead consider adding a DIRR_WARNPARAM.)
712 */
713enum directive_result {
714 DIRR_UNKNOWN, /* Directive not handled by backend */
715 DIRR_OK, /* Directive processed */
716 DIRR_ERROR, /* Directive processed unsuccessfully */
717 DIRR_BADPARAM /* Print bad argument error message */
718};
719
720/*
721 * A pragma facility: this structure is used to request passing a
722 * parsed pragma directive for a specific facility. If the handler is
723 * NULL then this pragma facility is recognized but ignored; pragma
724 * processing stops at that point.
725 *
726 * Note that the handler is passed a pointer to the facility structure
727 * as part of the struct pragma.
728 */
729struct pragma;
730typedef enum directive_result (*pragma_handler)(const struct pragma *);
731
732struct pragma_facility {
733 const char *name;
734 pragma_handler handler;
735};
736
737/*
738 * This structure defines how a pragma directive is passed to a
739 * facility. This structure may be augmented in the future.
740 *
741 * Any facility MAY, but is not required to, add its operations
742 * keywords or a subset thereof into asm/directiv.dat, in which case
743 * the "opcode" field will be set to the corresponding D_ constant
744 * from directiv.h; otherwise it will be D_unknown.
745 */
746struct pragma {
747 const struct pragma_facility *facility;
748 const char *facility_name; /* Facility name exactly as entered by user */
749 const char *opname; /* First word after the facility name */
750 const char *tail; /* Anything after the operation */
751 enum directive opcode; /* Operation as a D_ directives constant */
752};
753
754/*
755 * These are semi-arbitrary limits to keep the assembler from going
756 * into a black hole on certain kinds of bugs. They can be overridden
757 * by command-line options or %pragma.
758 */
759enum nasm_limit {
760 LIMIT_PASSES,
761 LIMIT_STALLED,
762 LIMIT_MACROS,
763 LIMIT_REP,
764 LIMIT_EVAL,
765 LIMIT_LINES
766};
767#define LIMIT_MAX LIMIT_LINES
768extern int64_t nasm_limit[LIMIT_MAX+1];
769extern enum directive_result nasm_set_limit(const char *, const char *);
770
771/*
772 * The data structure defining an output format driver, and the
773 * interfaces to the functions therein.
774 */
775struct ofmt {
776 /*
777 * This is a short (one-liner) description of the type of
778 * output generated by the driver.
779 */
780 const char *fullname;
781
782 /*
783 * This is a single keyword used to select the driver.
784 */
785 const char *shortname;
786
787 /*
788 * Default output filename extension, or a null string
789 */
790 const char *extension;
791
792 /*
793 * Output format flags.
794 */
795#define OFMT_TEXT 1 /* Text file format */
796#define OFMT_KEEP_ADDR 2 /* Keep addr; no conversion to data */
797
798 unsigned int flags;
799
800 int maxbits; /* Maximum segment bits supported */
801
802 /*
803 * this is a pointer to the first element of the debug information
804 */
805 const struct dfmt * const *debug_formats;
806
807 /*
808 * the default debugging format if -F is not specified
809 */
810 const struct dfmt *default_dfmt;
811
812 /*
813 * This, if non-NULL, is a NULL-terminated list of `char *'s
814 * pointing to extra standard macros supplied by the object
815 * format (e.g. a sensible initial default value of __SECT__,
816 * and user-level equivalents for any format-specific
817 * directives).
818 */
819 macros_t *stdmac;
820
821 /*
822 * This procedure is called at the start of an output session to set
823 * up internal parameters.
824 */
825 void (*init)(void);
826
827 /*
828 * This procedure is called at the start of each pass.
829 */
830 void (*reset)(void);
831
832 /*
833 * This is the modern output function, which gets passed
834 * a struct out_data with much more information. See the
835 * definition of struct out_data.
836 */
837 void (*output)(const struct out_data *data);
838
839 /*
840 * This procedure is called by assemble() to write actual
841 * generated code or data to the object file. Typically it
842 * doesn't have to actually _write_ it, just store it for
843 * later.
844 *
845 * The `type' argument specifies the type of output data, and
846 * usually the size as well: its contents are described below.
847 *
848 * This is used for backends which have not yet been ported to
849 * the new interface, and should be NULL on ported backends.
850 * To use this entry point, set the output pointer to
851 * nasm_do_legacy_output.
852 */
853 void (*legacy_output)(int32_t segto, const void *data,
854 enum out_type type, uint64_t size,
855 int32_t segment, int32_t wrt);
856
857 /*
858 * This procedure is called once for every symbol defined in
859 * the module being assembled. It gives the name and value of
860 * the symbol, in NASM's terms, and indicates whether it has
861 * been declared to be global. Note that the parameter "name",
862 * when passed, will point to a piece of static storage
863 * allocated inside the label manager - it's safe to keep using
864 * that pointer, because the label manager doesn't clean up
865 * until after the output driver has.
866 *
867 * Values of `is_global' are: 0 means the symbol is local; 1
868 * means the symbol is global; 2 means the symbol is common (in
869 * which case `offset' holds the _size_ of the variable).
870 * Anything else is available for the output driver to use
871 * internally.
872 *
873 * This routine explicitly _is_ allowed to call the label
874 * manager to define further symbols, if it wants to, even
875 * though it's been called _from_ the label manager. That much
876 * re-entrancy is guaranteed in the label manager. However, the
877 * label manager will in turn call this routine, so it should
878 * be prepared to be re-entrant itself.
879 *
880 * The `special' parameter contains special information passed
881 * through from the command that defined the label: it may have
882 * been an EXTERN, a COMMON or a GLOBAL. The distinction should
883 * be obvious to the output format from the other parameters.
884 */
885 void (*symdef)(char *name, int32_t segment, int64_t offset,
886 int is_global, char *special);
887
888 /*
889 * This procedure is called when the source code requests a
890 * segment change. It should return the corresponding segment
891 * _number_ for the name, or NO_SEG if the name is not a valid
892 * segment name.
893 *
894 * It may also be called with NULL, in which case it is to
895 * return the _default_ section number for starting assembly in.
896 *
897 * It is allowed to modify the string it is given a pointer to.
898 *
899 * It is also allowed to specify a default instruction size for
900 * the segment, by setting `*bits' to 16 or 32. Or, if it
901 * doesn't wish to define a default, it can leave `bits' alone.
902 */
903 int32_t (*section)(char *name, int pass, int *bits);
904
905 /*
906 * This function is called when a label is defined
907 * in the source code. It is allowed to change the section
908 * number as a result, but not the bits value.
909 * This is *only* called if the symbol defined is at the
910 * current offset, i.e. "foo:" or "foo equ $".
911 * The offset isn't passed; and may not be stable at this point.
912 * The subsection number is a field available for use by the
913 * backend. It is initialized to NO_SEG.
914 *
915 * If "copyoffset" is set by the backend then the offset is
916 * copied from the previous segment, otherwise the new segment
917 * is treated as a new segment the normal way.
918 */
919 int32_t (*herelabel)(const char *name, enum label_type type,
920 int32_t seg, int32_t *subsection,
921 bool *copyoffset);
922
923 /*
924 * This procedure is called to modify section alignment,
925 * note there is a trick, the alignment can only increase
926 */
927 void (*sectalign)(int32_t seg, unsigned int value);
928
929 /*
930 * This procedure is called to modify the segment base values
931 * returned from the SEG operator. It is given a segment base
932 * value (i.e. a segment value with the low bit set), and is
933 * required to produce in return a segment value which may be
934 * different. It can map segment bases to absolute numbers by
935 * means of returning SEG_ABS types.
936 *
937 * It should return NO_SEG if the segment base cannot be
938 * determined; the evaluator (which calls this routine) is
939 * responsible for throwing an error condition if that occurs
940 * in pass two or in a critical expression.
941 */
942 int32_t (*segbase)(int32_t segment);
943
944 /*
945 * This procedure is called to allow the output driver to
946 * process its own specific directives. When called, it has the
947 * directive word in `directive' and the parameter string in
948 * `value'. It is called in both assembly passes, and `pass'
949 * will be either 1 or 2.
950 *
951 * The following values are (currently) possible for
952 * directive_result:
953 *
954 * 0 - DIRR_UNKNOWN - directive not recognized by backend
955 * 1 - DIRR_OK - directive processed ok
956 * 2 - DIRR_ERROR - backend printed its own error message
957 * 3 - DIRR_BADPARAM - print the generic message
958 * "invalid parameter to [*] directive"
959 */
960 enum directive_result
961 (*directive)(enum directive directive, char *value, int pass);
962
963 /*
964 * This procedure is called after assembly finishes, to allow
965 * the output driver to clean itself up and free its memory.
966 * Typically, it will also be the point at which the object
967 * file actually gets _written_.
968 *
969 * One thing the cleanup routine should always do is to close
970 * the output file pointer.
971 */
972 void (*cleanup)(void);
973
974 /*
975 * List of pragma facility names that apply to this backend.
976 */
977 const struct pragma_facility *pragmas;
978};
979
980/*
981 * Output format driver alias
982 */
983struct ofmt_alias {
984 const char *shortname;
985 const char *fullname;
986 const struct ofmt *ofmt;
987};
988
989extern const struct ofmt *ofmt;
990extern FILE *ofile;
991
992/*
993 * ------------------------------------------------------------
994 * The data structure defining a debug format driver, and the
995 * interfaces to the functions therein.
996 * ------------------------------------------------------------
997 */
998
999struct dfmt {
1000 /*
1001 * This is a short (one-liner) description of the type of
1002 * output generated by the driver.
1003 */
1004 const char *fullname;
1005
1006 /*
1007 * This is a single keyword used to select the driver.
1008 */
1009 const char *shortname;
1010
1011 /*
1012 * init - called initially to set up local pointer to object format.
1013 */
1014 void (*init)(void);
1015
1016 /*
1017 * linenum - called any time there is output with a change of
1018 * line number or file.
1019 */
1020 void (*linenum)(const char *filename, int32_t linenumber, int32_t segto);
1021
1022 /*
1023 * debug_deflabel - called whenever a label is defined. Parameters
1024 * are the same as to 'symdef()' in the output format. This function
1025 * is called after the output format version.
1026 */
1027
1028 void (*debug_deflabel)(char *name, int32_t segment, int64_t offset,
1029 int is_global, char *special);
1030 /*
1031 * debug_directive - called whenever a DEBUG directive other than 'LINE'
1032 * is encountered. 'directive' contains the first parameter to the
1033 * DEBUG directive, and params contains the rest. For example,
1034 * 'DEBUG VAR _somevar:int' would translate to a call to this
1035 * function with 'directive' equal to "VAR" and 'params' equal to
1036 * "_somevar:int".
1037 */
1038 void (*debug_directive)(const char *directive, const char *params);
1039
1040 /*
1041 * typevalue - called whenever the assembler wishes to register a type
1042 * for the last defined label. This routine MUST detect if a type was
1043 * already registered and not re-register it.
1044 */
1045 void (*debug_typevalue)(int32_t type);
1046
1047 /*
1048 * debug_output - called whenever output is required
1049 * 'type' is the type of info required, and this is format-specific
1050 */
1051 void (*debug_output)(int type, void *param);
1052
1053 /*
1054 * cleanup - called after processing of file is complete
1055 */
1056 void (*cleanup)(void);
1057
1058 /*
1059 * List of pragma facility names that apply to this backend.
1060 */
1061 const struct pragma_facility *pragmas;
1062};
1063
1064extern const struct dfmt *dfmt;
1065
1066/*
1067 * The type definition macros
1068 * for debugging
1069 *
1070 * low 3 bits: reserved
1071 * next 5 bits: type
1072 * next 24 bits: number of elements for arrays (0 for labels)
1073 */
1074
1075#define TY_UNKNOWN 0x00
1076#define TY_LABEL 0x08
1077#define TY_BYTE 0x10
1078#define TY_WORD 0x18
1079#define TY_DWORD 0x20
1080#define TY_FLOAT 0x28
1081#define TY_QWORD 0x30
1082#define TY_TBYTE 0x38
1083#define TY_OWORD 0x40
1084#define TY_YWORD 0x48
1085#define TY_ZWORD 0x50
1086#define TY_COMMON 0xE0
1087#define TY_SEG 0xE8
1088#define TY_EXTERN 0xF0
1089#define TY_EQU 0xF8
1090
1091#define TYM_TYPE(x) ((x) & 0xF8)
1092#define TYM_ELEMENTS(x) (((x) & 0xFFFFFF00) >> 8)
1093
1094#define TYS_ELEMENTS(x) ((x) << 8)
1095
1096enum special_tokens {
1097 SPECIAL_ENUM_START = PREFIX_ENUM_LIMIT,
1098 S_ABS = SPECIAL_ENUM_START,
1099 S_BYTE,
1100 S_DWORD,
1101 S_FAR,
1102 S_LONG,
1103 S_NEAR,
1104 S_NOSPLIT,
1105 S_OWORD,
1106 S_QWORD,
1107 S_REL,
1108 S_SHORT,
1109 S_STRICT,
1110 S_TO,
1111 S_TWORD,
1112 S_WORD,
1113 S_YWORD,
1114 S_ZWORD,
1115 SPECIAL_ENUM_LIMIT
1116};
1117
1118enum decorator_tokens {
1119 DECORATOR_ENUM_START = SPECIAL_ENUM_LIMIT,
1120 BRC_1TO2 = DECORATOR_ENUM_START,
1121 BRC_1TO4,
1122 BRC_1TO8,
1123 BRC_1TO16,
1124 BRC_RN,
1125 BRC_RD,
1126 BRC_RU,
1127 BRC_RZ,
1128 BRC_SAE,
1129 BRC_Z,
1130 DECORATOR_ENUM_LIMIT
1131};
1132
1133/*
1134 * AVX512 Decorator (decoflags_t) bits distribution (counted from 0)
1135 * 3 2 1
1136 * 10987654321098765432109876543210
1137 * |
1138 * | word boundary
1139 * ............................1111 opmask
1140 * ...........................1.... zeroing / merging
1141 * ..........................1..... broadcast
1142 * .........................1...... static rounding
1143 * ........................1....... SAE
1144 * ......................11........ broadcast element size
1145 * ....................11.......... number of broadcast elements
1146 */
1147#define OP_GENVAL(val, bits, shift) (((val) & ((UINT64_C(1) << (bits)) - 1)) << (shift))
1148
1149/*
1150 * Opmask register number
1151 * identical to EVEX.aaa
1152 *
1153 * Bits: 0 - 3
1154 */
1155#define OPMASK_SHIFT (0)
1156#define OPMASK_BITS (4)
1157#define OPMASK_MASK OP_GENMASK(OPMASK_BITS, OPMASK_SHIFT)
1158#define GEN_OPMASK(bit) OP_GENBIT(bit, OPMASK_SHIFT)
1159#define VAL_OPMASK(val) OP_GENVAL(val, OPMASK_BITS, OPMASK_SHIFT)
1160
1161/*
1162 * zeroing / merging control available
1163 * matching to EVEX.z
1164 *
1165 * Bits: 4
1166 */
1167#define Z_SHIFT (4)
1168#define Z_BITS (1)
1169#define Z_MASK OP_GENMASK(Z_BITS, Z_SHIFT)
1170#define GEN_Z(bit) OP_GENBIT(bit, Z_SHIFT)
1171
1172/*
1173 * broadcast - Whether this operand can be broadcasted
1174 *
1175 * Bits: 5
1176 */
1177#define BRDCAST_SHIFT (5)
1178#define BRDCAST_BITS (1)
1179#define BRDCAST_MASK OP_GENMASK(BRDCAST_BITS, BRDCAST_SHIFT)
1180#define GEN_BRDCAST(bit) OP_GENBIT(bit, BRDCAST_SHIFT)
1181
1182/*
1183 * Whether this instruction can have a static rounding mode.
1184 * It goes with the last simd operand because the static rounding mode
1185 * decorator is located between the last simd operand and imm8 (if any).
1186 *
1187 * Bits: 6
1188 */
1189#define STATICRND_SHIFT (6)
1190#define STATICRND_BITS (1)
1191#define STATICRND_MASK OP_GENMASK(STATICRND_BITS, STATICRND_SHIFT)
1192#define GEN_STATICRND(bit) OP_GENBIT(bit, STATICRND_SHIFT)
1193
1194/*
1195 * SAE(Suppress all exception) available
1196 *
1197 * Bits: 7
1198 */
1199#define SAE_SHIFT (7)
1200#define SAE_BITS (1)
1201#define SAE_MASK OP_GENMASK(SAE_BITS, SAE_SHIFT)
1202#define GEN_SAE(bit) OP_GENBIT(bit, SAE_SHIFT)
1203
1204/*
1205 * Broadcasting element size.
1206 *
1207 * Bits: 8 - 9
1208 */
1209#define BRSIZE_SHIFT (8)
1210#define BRSIZE_BITS (2)
1211#define BRSIZE_MASK OP_GENMASK(BRSIZE_BITS, BRSIZE_SHIFT)
1212#define GEN_BRSIZE(bit) OP_GENBIT(bit, BRSIZE_SHIFT)
1213
1214#define BR_BITS32 GEN_BRSIZE(0)
1215#define BR_BITS64 GEN_BRSIZE(1)
1216
1217/*
1218 * Number of broadcasting elements
1219 *
1220 * Bits: 10 - 11
1221 */
1222#define BRNUM_SHIFT (10)
1223#define BRNUM_BITS (2)
1224#define BRNUM_MASK OP_GENMASK(BRNUM_BITS, BRNUM_SHIFT)
1225#define VAL_BRNUM(val) OP_GENVAL(val, BRNUM_BITS, BRNUM_SHIFT)
1226
1227#define BR_1TO2 VAL_BRNUM(0)
1228#define BR_1TO4 VAL_BRNUM(1)
1229#define BR_1TO8 VAL_BRNUM(2)
1230#define BR_1TO16 VAL_BRNUM(3)
1231
1232#define MASK OPMASK_MASK /* Opmask (k1 ~ 7) can be used */
1233#define Z Z_MASK
1234#define B32 (BRDCAST_MASK|BR_BITS32) /* {1to16} : broadcast 32b * 16 to zmm(512b) */
1235#define B64 (BRDCAST_MASK|BR_BITS64) /* {1to8} : broadcast 64b * 8 to zmm(512b) */
1236#define ER STATICRND_MASK /* ER(Embedded Rounding) == Static rounding mode */
1237#define SAE SAE_MASK /* SAE(Suppress All Exception) */
1238
1239/*
1240 * Global modes
1241 */
1242
1243/*
1244 * This declaration passes the "pass" number to all other modules
1245 * "pass0" assumes the values: 0, 0, ..., 0, 1, 2
1246 * where 0 = optimizing pass
1247 * 1 = pass 1
1248 * 2 = pass 2
1249 */
1250
1251/*
1252 * flag to disable optimizations selectively
1253 * this is useful to turn-off certain optimizations
1254 */
1255enum optimization_disable_flag {
1256 OPTIM_ALL_ENABLED = 0,
1257 OPTIM_DISABLE_JMP_MATCH = 1
1258};
1259
1260struct optimization {
1261 int level;
1262 int flag;
1263};
1264
1265extern int pass0;
1266extern int64_t passn; /* Actual pass number */
1267
1268extern bool tasm_compatible_mode;
1269extern struct optimization optimizing;
1270extern int globalbits; /* 16, 32 or 64-bit mode */
1271extern int globalrel; /* default to relative addressing? */
1272extern int globalbnd; /* default to using bnd prefix? */
1273
1274extern const char *inname; /* primary input filename */
1275extern const char *outname; /* output filename */
1276
1277/*
1278 * Switch to a different segment and return the current offset
1279 */
1280int64_t switch_segment(int32_t segment);
1281
1282#endif
1283