* config/tc-i386.c (md_apply_fix3): Replace S_IS_EXTERNAL,
[binutils-gdb.git] / gas / config / tc-i386.c
1 /* i386.c -- Assemble code for the Intel 80386
2 Copyright 1989, 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
3 2000, 2001, 2002
4 Free Software Foundation, Inc.
5
6 This file is part of GAS, the GNU Assembler.
7
8 GAS is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2, or (at your option)
11 any later version.
12
13 GAS is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
17
18 You should have received a copy of the GNU General Public License
19 along with GAS; see the file COPYING. If not, write to the Free
20 Software Foundation, 59 Temple Place - Suite 330, Boston, MA
21 02111-1307, USA. */
22
23 /* Intel 80386 machine specific gas.
24 Written by Eliot Dresselhaus (eliot@mgm.mit.edu).
25 x86_64 support by Jan Hubicka (jh@suse.cz)
26 Bugs & suggestions are completely welcome. This is free software.
27 Please help us make it better. */
28
29 #include "as.h"
30 #include "safe-ctype.h"
31 #include "subsegs.h"
32 #include "dwarf2dbg.h"
33 #include "opcode/i386.h"
34
35 #ifndef REGISTER_WARNINGS
36 #define REGISTER_WARNINGS 1
37 #endif
38
39 #ifndef INFER_ADDR_PREFIX
40 #define INFER_ADDR_PREFIX 1
41 #endif
42
43 #ifndef SCALE1_WHEN_NO_INDEX
44 /* Specifying a scale factor besides 1 when there is no index is
45 futile. eg. `mov (%ebx,2),%al' does exactly the same as
46 `mov (%ebx),%al'. To slavishly follow what the programmer
47 specified, set SCALE1_WHEN_NO_INDEX to 0. */
48 #define SCALE1_WHEN_NO_INDEX 1
49 #endif
50
51 #ifdef BFD_ASSEMBLER
52 #define RELOC_ENUM enum bfd_reloc_code_real
53 #else
54 #define RELOC_ENUM int
55 #endif
56
57 #ifndef DEFAULT_ARCH
58 #define DEFAULT_ARCH "i386"
59 #endif
60
61 #ifndef INLINE
62 #if __GNUC__ >= 2
63 #define INLINE __inline__
64 #else
65 #define INLINE
66 #endif
67 #endif
68
69 static INLINE unsigned int mode_from_disp_size PARAMS ((unsigned int));
70 static INLINE int fits_in_signed_byte PARAMS ((offsetT));
71 static INLINE int fits_in_unsigned_byte PARAMS ((offsetT));
72 static INLINE int fits_in_unsigned_word PARAMS ((offsetT));
73 static INLINE int fits_in_signed_word PARAMS ((offsetT));
74 static INLINE int fits_in_unsigned_long PARAMS ((offsetT));
75 static INLINE int fits_in_signed_long PARAMS ((offsetT));
76 static int smallest_imm_type PARAMS ((offsetT));
77 static offsetT offset_in_range PARAMS ((offsetT, int));
78 static int add_prefix PARAMS ((unsigned int));
79 static void set_code_flag PARAMS ((int));
80 static void set_16bit_gcc_code_flag PARAMS ((int));
81 static void set_intel_syntax PARAMS ((int));
82 static void set_cpu_arch PARAMS ((int));
83 static char *output_invalid PARAMS ((int c));
84 static int i386_operand PARAMS ((char *operand_string));
85 static int i386_intel_operand PARAMS ((char *operand_string, int got_a_float));
86 static const reg_entry *parse_register PARAMS ((char *reg_string,
87 char **end_op));
88 static char *parse_insn PARAMS ((char *, char *));
89 static char *parse_operands PARAMS ((char *, const char *));
90 static void swap_operands PARAMS ((void));
91 static void optimize_imm PARAMS ((void));
92 static void optimize_disp PARAMS ((void));
93 static int match_template PARAMS ((void));
94 static int check_string PARAMS ((void));
95 static int process_suffix PARAMS ((void));
96 static int check_byte_reg PARAMS ((void));
97 static int check_long_reg PARAMS ((void));
98 static int check_qword_reg PARAMS ((void));
99 static int check_word_reg PARAMS ((void));
100 static int finalize_imm PARAMS ((void));
101 static int process_operands PARAMS ((void));
102 static const seg_entry *build_modrm_byte PARAMS ((void));
103 static void output_insn PARAMS ((void));
104 static void output_branch PARAMS ((void));
105 static void output_jump PARAMS ((void));
106 static void output_interseg_jump PARAMS ((void));
107 static void output_imm PARAMS ((fragS *insn_start_frag,
108 offsetT insn_start_off));
109 static void output_disp PARAMS ((fragS *insn_start_frag,
110 offsetT insn_start_off));
111 #ifndef I386COFF
112 static void s_bss PARAMS ((int));
113 #endif
114
115 static const char *default_arch = DEFAULT_ARCH;
116
117 /* 'md_assemble ()' gathers together information and puts it into a
118 i386_insn. */
119
120 union i386_op
121 {
122 expressionS *disps;
123 expressionS *imms;
124 const reg_entry *regs;
125 };
126
127 struct _i386_insn
128 {
129 /* TM holds the template for the insn were currently assembling. */
130 template tm;
131
132 /* SUFFIX holds the instruction mnemonic suffix if given.
133 (e.g. 'l' for 'movl') */
134 char suffix;
135
136 /* OPERANDS gives the number of given operands. */
137 unsigned int operands;
138
139 /* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number
140 of given register, displacement, memory operands and immediate
141 operands. */
142 unsigned int reg_operands, disp_operands, mem_operands, imm_operands;
143
144 /* TYPES [i] is the type (see above #defines) which tells us how to
145 use OP[i] for the corresponding operand. */
146 unsigned int types[MAX_OPERANDS];
147
148 /* Displacement expression, immediate expression, or register for each
149 operand. */
150 union i386_op op[MAX_OPERANDS];
151
152 /* Flags for operands. */
153 unsigned int flags[MAX_OPERANDS];
154 #define Operand_PCrel 1
155
156 /* Relocation type for operand */
157 RELOC_ENUM reloc[MAX_OPERANDS];
158
159 /* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode
160 the base index byte below. */
161 const reg_entry *base_reg;
162 const reg_entry *index_reg;
163 unsigned int log2_scale_factor;
164
165 /* SEG gives the seg_entries of this insn. They are zero unless
166 explicit segment overrides are given. */
167 const seg_entry *seg[2];
168
169 /* PREFIX holds all the given prefix opcodes (usually null).
170 PREFIXES is the number of prefix opcodes. */
171 unsigned int prefixes;
172 unsigned char prefix[MAX_PREFIXES];
173
174 /* RM and SIB are the modrm byte and the sib byte where the
175 addressing modes of this insn are encoded. */
176
177 modrm_byte rm;
178 rex_byte rex;
179 sib_byte sib;
180 };
181
182 typedef struct _i386_insn i386_insn;
183
184 /* List of chars besides those in app.c:symbol_chars that can start an
185 operand. Used to prevent the scrubber eating vital white-space. */
186 #ifdef LEX_AT
187 const char extra_symbol_chars[] = "*%-(@[";
188 #else
189 const char extra_symbol_chars[] = "*%-([";
190 #endif
191
192 #if (defined (TE_I386AIX) \
193 || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) \
194 && !defined (TE_LINUX) \
195 && !defined (TE_FreeBSD) \
196 && !defined (TE_NetBSD)))
197 /* This array holds the chars that always start a comment. If the
198 pre-processor is disabled, these aren't very useful. */
199 const char comment_chars[] = "#/";
200 #define PREFIX_SEPARATOR '\\'
201
202 /* This array holds the chars that only start a comment at the beginning of
203 a line. If the line seems to have the form '# 123 filename'
204 .line and .file directives will appear in the pre-processed output.
205 Note that input_file.c hand checks for '#' at the beginning of the
206 first line of the input file. This is because the compiler outputs
207 #NO_APP at the beginning of its output.
208 Also note that comments started like this one will always work if
209 '/' isn't otherwise defined. */
210 const char line_comment_chars[] = "";
211
212 #else
213 /* Putting '/' here makes it impossible to use the divide operator.
214 However, we need it for compatibility with SVR4 systems. */
215 const char comment_chars[] = "#";
216 #define PREFIX_SEPARATOR '/'
217
218 const char line_comment_chars[] = "/";
219 #endif
220
221 const char line_separator_chars[] = ";";
222
223 /* Chars that can be used to separate mant from exp in floating point
224 nums. */
225 const char EXP_CHARS[] = "eE";
226
227 /* Chars that mean this number is a floating point constant
228 As in 0f12.456
229 or 0d1.2345e12. */
230 const char FLT_CHARS[] = "fFdDxX";
231
232 /* Tables for lexical analysis. */
233 static char mnemonic_chars[256];
234 static char register_chars[256];
235 static char operand_chars[256];
236 static char identifier_chars[256];
237 static char digit_chars[256];
238
239 /* Lexical macros. */
240 #define is_mnemonic_char(x) (mnemonic_chars[(unsigned char) x])
241 #define is_operand_char(x) (operand_chars[(unsigned char) x])
242 #define is_register_char(x) (register_chars[(unsigned char) x])
243 #define is_space_char(x) ((x) == ' ')
244 #define is_identifier_char(x) (identifier_chars[(unsigned char) x])
245 #define is_digit_char(x) (digit_chars[(unsigned char) x])
246
247 /* All non-digit non-letter charcters that may occur in an operand. */
248 static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
249
250 /* md_assemble() always leaves the strings it's passed unaltered. To
251 effect this we maintain a stack of saved characters that we've smashed
252 with '\0's (indicating end of strings for various sub-fields of the
253 assembler instruction). */
254 static char save_stack[32];
255 static char *save_stack_p;
256 #define END_STRING_AND_SAVE(s) \
257 do { *save_stack_p++ = *(s); *(s) = '\0'; } while (0)
258 #define RESTORE_END_STRING(s) \
259 do { *(s) = *--save_stack_p; } while (0)
260
261 /* The instruction we're assembling. */
262 static i386_insn i;
263
264 /* Possible templates for current insn. */
265 static const templates *current_templates;
266
267 /* Per instruction expressionS buffers: 2 displacements & 2 immediate max. */
268 static expressionS disp_expressions[2], im_expressions[2];
269
270 /* Current operand we are working on. */
271 static int this_operand;
272
273 /* We support four different modes. FLAG_CODE variable is used to distinguish
274 these. */
275
276 enum flag_code {
277 CODE_32BIT,
278 CODE_16BIT,
279 CODE_64BIT };
280 #define NUM_FLAG_CODE ((int) CODE_64BIT + 1)
281
282 static enum flag_code flag_code;
283 static int use_rela_relocations = 0;
284
285 /* The names used to print error messages. */
286 static const char *flag_code_names[] =
287 {
288 "32",
289 "16",
290 "64"
291 };
292
293 /* 1 for intel syntax,
294 0 if att syntax. */
295 static int intel_syntax = 0;
296
297 /* 1 if register prefix % not required. */
298 static int allow_naked_reg = 0;
299
300 /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
301 leave, push, and pop instructions so that gcc has the same stack
302 frame as in 32 bit mode. */
303 static char stackop_size = '\0';
304
305 /* Non-zero to quieten some warnings. */
306 static int quiet_warnings = 0;
307
308 /* CPU name. */
309 static const char *cpu_arch_name = NULL;
310
311 /* CPU feature flags. */
312 static unsigned int cpu_arch_flags = CpuUnknownFlags | CpuNo64;
313
314 /* If set, conditional jumps are not automatically promoted to handle
315 larger than a byte offset. */
316 static unsigned int no_cond_jump_promotion = 0;
317
318 /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
319 symbolS *GOT_symbol;
320
321 /* Interface to relax_segment.
322 There are 3 major relax states for 386 jump insns because the
323 different types of jumps add different sizes to frags when we're
324 figuring out what sort of jump to choose to reach a given label. */
325
326 /* Types. */
327 #define UNCOND_JUMP 0
328 #define COND_JUMP 1
329 #define COND_JUMP86 2
330
331 /* Sizes. */
332 #define CODE16 1
333 #define SMALL 0
334 #define SMALL16 (SMALL | CODE16)
335 #define BIG 2
336 #define BIG16 (BIG | CODE16)
337
338 #ifndef INLINE
339 #ifdef __GNUC__
340 #define INLINE __inline__
341 #else
342 #define INLINE
343 #endif
344 #endif
345
346 #define ENCODE_RELAX_STATE(type, size) \
347 ((relax_substateT) (((type) << 2) | (size)))
348 #define TYPE_FROM_RELAX_STATE(s) \
349 ((s) >> 2)
350 #define DISP_SIZE_FROM_RELAX_STATE(s) \
351 ((((s) & 3) == BIG ? 4 : (((s) & 3) == BIG16 ? 2 : 1)))
352
353 /* This table is used by relax_frag to promote short jumps to long
354 ones where necessary. SMALL (short) jumps may be promoted to BIG
355 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
356 don't allow a short jump in a 32 bit code segment to be promoted to
357 a 16 bit offset jump because it's slower (requires data size
358 prefix), and doesn't work, unless the destination is in the bottom
359 64k of the code segment (The top 16 bits of eip are zeroed). */
360
361 const relax_typeS md_relax_table[] =
362 {
363 /* The fields are:
364 1) most positive reach of this state,
365 2) most negative reach of this state,
366 3) how many bytes this mode will have in the variable part of the frag
367 4) which index into the table to try if we can't fit into this one. */
368
369 /* UNCOND_JUMP states. */
370 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
371 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
372 /* dword jmp adds 4 bytes to frag:
373 0 extra opcode bytes, 4 displacement bytes. */
374 {0, 0, 4, 0},
375 /* word jmp adds 2 byte2 to frag:
376 0 extra opcode bytes, 2 displacement bytes. */
377 {0, 0, 2, 0},
378
379 /* COND_JUMP states. */
380 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
381 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
382 /* dword conditionals adds 5 bytes to frag:
383 1 extra opcode byte, 4 displacement bytes. */
384 {0, 0, 5, 0},
385 /* word conditionals add 3 bytes to frag:
386 1 extra opcode byte, 2 displacement bytes. */
387 {0, 0, 3, 0},
388
389 /* COND_JUMP86 states. */
390 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG)},
391 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG16)},
392 /* dword conditionals adds 5 bytes to frag:
393 1 extra opcode byte, 4 displacement bytes. */
394 {0, 0, 5, 0},
395 /* word conditionals add 4 bytes to frag:
396 1 displacement byte and a 3 byte long branch insn. */
397 {0, 0, 4, 0}
398 };
399
400 static const arch_entry cpu_arch[] = {
401 {"i8086", Cpu086 },
402 {"i186", Cpu086|Cpu186 },
403 {"i286", Cpu086|Cpu186|Cpu286 },
404 {"i386", Cpu086|Cpu186|Cpu286|Cpu386 },
405 {"i486", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486 },
406 {"i586", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
407 {"i686", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
408 {"pentium", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
409 {"pentiumpro",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
410 {"pentium4", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX|CpuSSE|CpuSSE2 },
411 {"k6", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow },
412 {"athlon", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuMMX|Cpu3dnow },
413 {"sledgehammer",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuSledgehammer|CpuMMX|Cpu3dnow|CpuSSE|CpuSSE2 },
414 {NULL, 0 }
415 };
416
417 const pseudo_typeS md_pseudo_table[] =
418 {
419 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
420 {"align", s_align_bytes, 0},
421 #else
422 {"align", s_align_ptwo, 0},
423 #endif
424 {"arch", set_cpu_arch, 0},
425 #ifndef I386COFF
426 {"bss", s_bss, 0},
427 #endif
428 {"ffloat", float_cons, 'f'},
429 {"dfloat", float_cons, 'd'},
430 {"tfloat", float_cons, 'x'},
431 {"value", cons, 2},
432 {"noopt", s_ignore, 0},
433 {"optim", s_ignore, 0},
434 {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
435 {"code16", set_code_flag, CODE_16BIT},
436 {"code32", set_code_flag, CODE_32BIT},
437 {"code64", set_code_flag, CODE_64BIT},
438 {"intel_syntax", set_intel_syntax, 1},
439 {"att_syntax", set_intel_syntax, 0},
440 {"file", (void (*) PARAMS ((int))) dwarf2_directive_file, 0},
441 {"loc", dwarf2_directive_loc, 0},
442 {0, 0, 0}
443 };
444
445 /* For interface with expression (). */
446 extern char *input_line_pointer;
447
448 /* Hash table for instruction mnemonic lookup. */
449 static struct hash_control *op_hash;
450
451 /* Hash table for register lookup. */
452 static struct hash_control *reg_hash;
453 \f
454 void
455 i386_align_code (fragP, count)
456 fragS *fragP;
457 int count;
458 {
459 /* Various efficient no-op patterns for aligning code labels.
460 Note: Don't try to assemble the instructions in the comments.
461 0L and 0w are not legal. */
462 static const char f32_1[] =
463 {0x90}; /* nop */
464 static const char f32_2[] =
465 {0x89,0xf6}; /* movl %esi,%esi */
466 static const char f32_3[] =
467 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
468 static const char f32_4[] =
469 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
470 static const char f32_5[] =
471 {0x90, /* nop */
472 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
473 static const char f32_6[] =
474 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
475 static const char f32_7[] =
476 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
477 static const char f32_8[] =
478 {0x90, /* nop */
479 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
480 static const char f32_9[] =
481 {0x89,0xf6, /* movl %esi,%esi */
482 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
483 static const char f32_10[] =
484 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
485 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
486 static const char f32_11[] =
487 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
488 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
489 static const char f32_12[] =
490 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
491 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
492 static const char f32_13[] =
493 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
494 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
495 static const char f32_14[] =
496 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
497 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
498 static const char f32_15[] =
499 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
500 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
501 static const char f16_3[] =
502 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
503 static const char f16_4[] =
504 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
505 static const char f16_5[] =
506 {0x90, /* nop */
507 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
508 static const char f16_6[] =
509 {0x89,0xf6, /* mov %si,%si */
510 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
511 static const char f16_7[] =
512 {0x8d,0x74,0x00, /* lea 0(%si),%si */
513 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
514 static const char f16_8[] =
515 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
516 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
517 static const char *const f32_patt[] = {
518 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
519 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
520 };
521 static const char *const f16_patt[] = {
522 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
523 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
524 };
525
526 if (count <= 0 || count > 15)
527 return;
528
529 /* The recommended way to pad 64bit code is to use NOPs preceded by
530 maximally four 0x66 prefixes. Balance the size of nops. */
531 if (flag_code == CODE_64BIT)
532 {
533 int i;
534 int nnops = (count + 3) / 4;
535 int len = count / nnops;
536 int remains = count - nnops * len;
537 int pos = 0;
538
539 for (i = 0; i < remains; i++)
540 {
541 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len);
542 fragP->fr_literal[fragP->fr_fix + pos + len] = 0x90;
543 pos += len + 1;
544 }
545 for (; i < nnops; i++)
546 {
547 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len - 1);
548 fragP->fr_literal[fragP->fr_fix + pos + len - 1] = 0x90;
549 pos += len;
550 }
551 }
552 else
553 if (flag_code == CODE_16BIT)
554 {
555 memcpy (fragP->fr_literal + fragP->fr_fix,
556 f16_patt[count - 1], count);
557 if (count > 8)
558 /* Adjust jump offset. */
559 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
560 }
561 else
562 memcpy (fragP->fr_literal + fragP->fr_fix,
563 f32_patt[count - 1], count);
564 fragP->fr_var = count;
565 }
566
567 static INLINE unsigned int
568 mode_from_disp_size (t)
569 unsigned int t;
570 {
571 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
572 }
573
574 static INLINE int
575 fits_in_signed_byte (num)
576 offsetT num;
577 {
578 return (num >= -128) && (num <= 127);
579 }
580
581 static INLINE int
582 fits_in_unsigned_byte (num)
583 offsetT num;
584 {
585 return (num & 0xff) == num;
586 }
587
588 static INLINE int
589 fits_in_unsigned_word (num)
590 offsetT num;
591 {
592 return (num & 0xffff) == num;
593 }
594
595 static INLINE int
596 fits_in_signed_word (num)
597 offsetT num;
598 {
599 return (-32768 <= num) && (num <= 32767);
600 }
601 static INLINE int
602 fits_in_signed_long (num)
603 offsetT num ATTRIBUTE_UNUSED;
604 {
605 #ifndef BFD64
606 return 1;
607 #else
608 return (!(((offsetT) -1 << 31) & num)
609 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
610 #endif
611 } /* fits_in_signed_long() */
612 static INLINE int
613 fits_in_unsigned_long (num)
614 offsetT num ATTRIBUTE_UNUSED;
615 {
616 #ifndef BFD64
617 return 1;
618 #else
619 return (num & (((offsetT) 2 << 31) - 1)) == num;
620 #endif
621 } /* fits_in_unsigned_long() */
622
623 static int
624 smallest_imm_type (num)
625 offsetT num;
626 {
627 if (cpu_arch_flags != (Cpu086 | Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64))
628 {
629 /* This code is disabled on the 486 because all the Imm1 forms
630 in the opcode table are slower on the i486. They're the
631 versions with the implicitly specified single-position
632 displacement, which has another syntax if you really want to
633 use that form. */
634 if (num == 1)
635 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
636 }
637 return (fits_in_signed_byte (num)
638 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
639 : fits_in_unsigned_byte (num)
640 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
641 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
642 ? (Imm16 | Imm32 | Imm32S | Imm64)
643 : fits_in_signed_long (num)
644 ? (Imm32 | Imm32S | Imm64)
645 : fits_in_unsigned_long (num)
646 ? (Imm32 | Imm64)
647 : Imm64);
648 }
649
650 static offsetT
651 offset_in_range (val, size)
652 offsetT val;
653 int size;
654 {
655 addressT mask;
656
657 switch (size)
658 {
659 case 1: mask = ((addressT) 1 << 8) - 1; break;
660 case 2: mask = ((addressT) 1 << 16) - 1; break;
661 case 4: mask = ((addressT) 2 << 31) - 1; break;
662 #ifdef BFD64
663 case 8: mask = ((addressT) 2 << 63) - 1; break;
664 #endif
665 default: abort ();
666 }
667
668 /* If BFD64, sign extend val. */
669 if (!use_rela_relocations)
670 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
671 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
672
673 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
674 {
675 char buf1[40], buf2[40];
676
677 sprint_value (buf1, val);
678 sprint_value (buf2, val & mask);
679 as_warn (_("%s shortened to %s"), buf1, buf2);
680 }
681 return val & mask;
682 }
683
684 /* Returns 0 if attempting to add a prefix where one from the same
685 class already exists, 1 if non rep/repne added, 2 if rep/repne
686 added. */
687 static int
688 add_prefix (prefix)
689 unsigned int prefix;
690 {
691 int ret = 1;
692 int q;
693
694 if (prefix >= REX_OPCODE && prefix < REX_OPCODE + 16
695 && flag_code == CODE_64BIT)
696 q = REX_PREFIX;
697 else
698 switch (prefix)
699 {
700 default:
701 abort ();
702
703 case CS_PREFIX_OPCODE:
704 case DS_PREFIX_OPCODE:
705 case ES_PREFIX_OPCODE:
706 case FS_PREFIX_OPCODE:
707 case GS_PREFIX_OPCODE:
708 case SS_PREFIX_OPCODE:
709 q = SEG_PREFIX;
710 break;
711
712 case REPNE_PREFIX_OPCODE:
713 case REPE_PREFIX_OPCODE:
714 ret = 2;
715 /* fall thru */
716 case LOCK_PREFIX_OPCODE:
717 q = LOCKREP_PREFIX;
718 break;
719
720 case FWAIT_OPCODE:
721 q = WAIT_PREFIX;
722 break;
723
724 case ADDR_PREFIX_OPCODE:
725 q = ADDR_PREFIX;
726 break;
727
728 case DATA_PREFIX_OPCODE:
729 q = DATA_PREFIX;
730 break;
731 }
732
733 if (i.prefix[q] != 0)
734 {
735 as_bad (_("same type of prefix used twice"));
736 return 0;
737 }
738
739 i.prefixes += 1;
740 i.prefix[q] = prefix;
741 return ret;
742 }
743
744 static void
745 set_code_flag (value)
746 int value;
747 {
748 flag_code = value;
749 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
750 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
751 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
752 {
753 as_bad (_("64bit mode not supported on this CPU."));
754 }
755 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
756 {
757 as_bad (_("32bit mode not supported on this CPU."));
758 }
759 stackop_size = '\0';
760 }
761
762 static void
763 set_16bit_gcc_code_flag (new_code_flag)
764 int new_code_flag;
765 {
766 flag_code = new_code_flag;
767 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
768 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
769 stackop_size = 'l';
770 }
771
772 static void
773 set_intel_syntax (syntax_flag)
774 int syntax_flag;
775 {
776 /* Find out if register prefixing is specified. */
777 int ask_naked_reg = 0;
778
779 SKIP_WHITESPACE ();
780 if (!is_end_of_line[(unsigned char) *input_line_pointer])
781 {
782 char *string = input_line_pointer;
783 int e = get_symbol_end ();
784
785 if (strcmp (string, "prefix") == 0)
786 ask_naked_reg = 1;
787 else if (strcmp (string, "noprefix") == 0)
788 ask_naked_reg = -1;
789 else
790 as_bad (_("bad argument to syntax directive."));
791 *input_line_pointer = e;
792 }
793 demand_empty_rest_of_line ();
794
795 intel_syntax = syntax_flag;
796
797 if (ask_naked_reg == 0)
798 {
799 #ifdef BFD_ASSEMBLER
800 allow_naked_reg = (intel_syntax
801 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
802 #else
803 /* Conservative default. */
804 allow_naked_reg = 0;
805 #endif
806 }
807 else
808 allow_naked_reg = (ask_naked_reg < 0);
809 }
810
811 static void
812 set_cpu_arch (dummy)
813 int dummy ATTRIBUTE_UNUSED;
814 {
815 SKIP_WHITESPACE ();
816
817 if (!is_end_of_line[(unsigned char) *input_line_pointer])
818 {
819 char *string = input_line_pointer;
820 int e = get_symbol_end ();
821 int i;
822
823 for (i = 0; cpu_arch[i].name; i++)
824 {
825 if (strcmp (string, cpu_arch[i].name) == 0)
826 {
827 cpu_arch_name = cpu_arch[i].name;
828 cpu_arch_flags = (cpu_arch[i].flags
829 | (flag_code == CODE_64BIT ? Cpu64 : CpuNo64));
830 break;
831 }
832 }
833 if (!cpu_arch[i].name)
834 as_bad (_("no such architecture: `%s'"), string);
835
836 *input_line_pointer = e;
837 }
838 else
839 as_bad (_("missing cpu architecture"));
840
841 no_cond_jump_promotion = 0;
842 if (*input_line_pointer == ','
843 && !is_end_of_line[(unsigned char) input_line_pointer[1]])
844 {
845 char *string = ++input_line_pointer;
846 int e = get_symbol_end ();
847
848 if (strcmp (string, "nojumps") == 0)
849 no_cond_jump_promotion = 1;
850 else if (strcmp (string, "jumps") == 0)
851 ;
852 else
853 as_bad (_("no such architecture modifier: `%s'"), string);
854
855 *input_line_pointer = e;
856 }
857
858 demand_empty_rest_of_line ();
859 }
860
861 #ifdef BFD_ASSEMBLER
862 unsigned long
863 i386_mach ()
864 {
865 if (!strcmp (default_arch, "x86_64"))
866 return bfd_mach_x86_64;
867 else if (!strcmp (default_arch, "i386"))
868 return bfd_mach_i386_i386;
869 else
870 as_fatal (_("Unknown architecture"));
871 }
872 #endif
873 \f
874 void
875 md_begin ()
876 {
877 const char *hash_err;
878
879 /* Initialize op_hash hash table. */
880 op_hash = hash_new ();
881
882 {
883 const template *optab;
884 templates *core_optab;
885
886 /* Setup for loop. */
887 optab = i386_optab;
888 core_optab = (templates *) xmalloc (sizeof (templates));
889 core_optab->start = optab;
890
891 while (1)
892 {
893 ++optab;
894 if (optab->name == NULL
895 || strcmp (optab->name, (optab - 1)->name) != 0)
896 {
897 /* different name --> ship out current template list;
898 add to hash table; & begin anew. */
899 core_optab->end = optab;
900 hash_err = hash_insert (op_hash,
901 (optab - 1)->name,
902 (PTR) core_optab);
903 if (hash_err)
904 {
905 as_fatal (_("Internal Error: Can't hash %s: %s"),
906 (optab - 1)->name,
907 hash_err);
908 }
909 if (optab->name == NULL)
910 break;
911 core_optab = (templates *) xmalloc (sizeof (templates));
912 core_optab->start = optab;
913 }
914 }
915 }
916
917 /* Initialize reg_hash hash table. */
918 reg_hash = hash_new ();
919 {
920 const reg_entry *regtab;
921
922 for (regtab = i386_regtab;
923 regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
924 regtab++)
925 {
926 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
927 if (hash_err)
928 as_fatal (_("Internal Error: Can't hash %s: %s"),
929 regtab->reg_name,
930 hash_err);
931 }
932 }
933
934 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
935 {
936 int c;
937 char *p;
938
939 for (c = 0; c < 256; c++)
940 {
941 if (ISDIGIT (c))
942 {
943 digit_chars[c] = c;
944 mnemonic_chars[c] = c;
945 register_chars[c] = c;
946 operand_chars[c] = c;
947 }
948 else if (ISLOWER (c))
949 {
950 mnemonic_chars[c] = c;
951 register_chars[c] = c;
952 operand_chars[c] = c;
953 }
954 else if (ISUPPER (c))
955 {
956 mnemonic_chars[c] = TOLOWER (c);
957 register_chars[c] = mnemonic_chars[c];
958 operand_chars[c] = c;
959 }
960
961 if (ISALPHA (c) || ISDIGIT (c))
962 identifier_chars[c] = c;
963 else if (c >= 128)
964 {
965 identifier_chars[c] = c;
966 operand_chars[c] = c;
967 }
968 }
969
970 #ifdef LEX_AT
971 identifier_chars['@'] = '@';
972 #endif
973 digit_chars['-'] = '-';
974 identifier_chars['_'] = '_';
975 identifier_chars['.'] = '.';
976
977 for (p = operand_special_chars; *p != '\0'; p++)
978 operand_chars[(unsigned char) *p] = *p;
979 }
980
981 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
982 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
983 {
984 record_alignment (text_section, 2);
985 record_alignment (data_section, 2);
986 record_alignment (bss_section, 2);
987 }
988 #endif
989 }
990
991 void
992 i386_print_statistics (file)
993 FILE *file;
994 {
995 hash_print_statistics (file, "i386 opcode", op_hash);
996 hash_print_statistics (file, "i386 register", reg_hash);
997 }
998 \f
999 #ifdef DEBUG386
1000
1001 /* Debugging routines for md_assemble. */
1002 static void pi PARAMS ((char *, i386_insn *));
1003 static void pte PARAMS ((template *));
1004 static void pt PARAMS ((unsigned int));
1005 static void pe PARAMS ((expressionS *));
1006 static void ps PARAMS ((symbolS *));
1007
1008 static void
1009 pi (line, x)
1010 char *line;
1011 i386_insn *x;
1012 {
1013 unsigned int i;
1014
1015 fprintf (stdout, "%s: template ", line);
1016 pte (&x->tm);
1017 fprintf (stdout, " address: base %s index %s scale %x\n",
1018 x->base_reg ? x->base_reg->reg_name : "none",
1019 x->index_reg ? x->index_reg->reg_name : "none",
1020 x->log2_scale_factor);
1021 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
1022 x->rm.mode, x->rm.reg, x->rm.regmem);
1023 fprintf (stdout, " sib: base %x index %x scale %x\n",
1024 x->sib.base, x->sib.index, x->sib.scale);
1025 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
1026 (x->rex & REX_MODE64) != 0,
1027 (x->rex & REX_EXTX) != 0,
1028 (x->rex & REX_EXTY) != 0,
1029 (x->rex & REX_EXTZ) != 0);
1030 for (i = 0; i < x->operands; i++)
1031 {
1032 fprintf (stdout, " #%d: ", i + 1);
1033 pt (x->types[i]);
1034 fprintf (stdout, "\n");
1035 if (x->types[i]
1036 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
1037 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
1038 if (x->types[i] & Imm)
1039 pe (x->op[i].imms);
1040 if (x->types[i] & Disp)
1041 pe (x->op[i].disps);
1042 }
1043 }
1044
1045 static void
1046 pte (t)
1047 template *t;
1048 {
1049 unsigned int i;
1050 fprintf (stdout, " %d operands ", t->operands);
1051 fprintf (stdout, "opcode %x ", t->base_opcode);
1052 if (t->extension_opcode != None)
1053 fprintf (stdout, "ext %x ", t->extension_opcode);
1054 if (t->opcode_modifier & D)
1055 fprintf (stdout, "D");
1056 if (t->opcode_modifier & W)
1057 fprintf (stdout, "W");
1058 fprintf (stdout, "\n");
1059 for (i = 0; i < t->operands; i++)
1060 {
1061 fprintf (stdout, " #%d type ", i + 1);
1062 pt (t->operand_types[i]);
1063 fprintf (stdout, "\n");
1064 }
1065 }
1066
1067 static void
1068 pe (e)
1069 expressionS *e;
1070 {
1071 fprintf (stdout, " operation %d\n", e->X_op);
1072 fprintf (stdout, " add_number %ld (%lx)\n",
1073 (long) e->X_add_number, (long) e->X_add_number);
1074 if (e->X_add_symbol)
1075 {
1076 fprintf (stdout, " add_symbol ");
1077 ps (e->X_add_symbol);
1078 fprintf (stdout, "\n");
1079 }
1080 if (e->X_op_symbol)
1081 {
1082 fprintf (stdout, " op_symbol ");
1083 ps (e->X_op_symbol);
1084 fprintf (stdout, "\n");
1085 }
1086 }
1087
1088 static void
1089 ps (s)
1090 symbolS *s;
1091 {
1092 fprintf (stdout, "%s type %s%s",
1093 S_GET_NAME (s),
1094 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
1095 segment_name (S_GET_SEGMENT (s)));
1096 }
1097
1098 struct type_name
1099 {
1100 unsigned int mask;
1101 char *tname;
1102 }
1103
1104 static const type_names[] =
1105 {
1106 { Reg8, "r8" },
1107 { Reg16, "r16" },
1108 { Reg32, "r32" },
1109 { Reg64, "r64" },
1110 { Imm8, "i8" },
1111 { Imm8S, "i8s" },
1112 { Imm16, "i16" },
1113 { Imm32, "i32" },
1114 { Imm32S, "i32s" },
1115 { Imm64, "i64" },
1116 { Imm1, "i1" },
1117 { BaseIndex, "BaseIndex" },
1118 { Disp8, "d8" },
1119 { Disp16, "d16" },
1120 { Disp32, "d32" },
1121 { Disp32S, "d32s" },
1122 { Disp64, "d64" },
1123 { InOutPortReg, "InOutPortReg" },
1124 { ShiftCount, "ShiftCount" },
1125 { Control, "control reg" },
1126 { Test, "test reg" },
1127 { Debug, "debug reg" },
1128 { FloatReg, "FReg" },
1129 { FloatAcc, "FAcc" },
1130 { SReg2, "SReg2" },
1131 { SReg3, "SReg3" },
1132 { Acc, "Acc" },
1133 { JumpAbsolute, "Jump Absolute" },
1134 { RegMMX, "rMMX" },
1135 { RegXMM, "rXMM" },
1136 { EsSeg, "es" },
1137 { 0, "" }
1138 };
1139
1140 static void
1141 pt (t)
1142 unsigned int t;
1143 {
1144 const struct type_name *ty;
1145
1146 for (ty = type_names; ty->mask; ty++)
1147 if (t & ty->mask)
1148 fprintf (stdout, "%s, ", ty->tname);
1149 fflush (stdout);
1150 }
1151
1152 #endif /* DEBUG386 */
1153 \f
1154 #ifdef BFD_ASSEMBLER
1155 static bfd_reloc_code_real_type reloc
1156 PARAMS ((int, int, int, bfd_reloc_code_real_type));
1157
1158 static bfd_reloc_code_real_type
1159 reloc (size, pcrel, sign, other)
1160 int size;
1161 int pcrel;
1162 int sign;
1163 bfd_reloc_code_real_type other;
1164 {
1165 if (other != NO_RELOC)
1166 return other;
1167
1168 if (pcrel)
1169 {
1170 if (!sign)
1171 as_bad (_("There are no unsigned pc-relative relocations"));
1172 switch (size)
1173 {
1174 case 1: return BFD_RELOC_8_PCREL;
1175 case 2: return BFD_RELOC_16_PCREL;
1176 case 4: return BFD_RELOC_32_PCREL;
1177 }
1178 as_bad (_("can not do %d byte pc-relative relocation"), size);
1179 }
1180 else
1181 {
1182 if (sign)
1183 switch (size)
1184 {
1185 case 4: return BFD_RELOC_X86_64_32S;
1186 }
1187 else
1188 switch (size)
1189 {
1190 case 1: return BFD_RELOC_8;
1191 case 2: return BFD_RELOC_16;
1192 case 4: return BFD_RELOC_32;
1193 case 8: return BFD_RELOC_64;
1194 }
1195 as_bad (_("can not do %s %d byte relocation"),
1196 sign ? "signed" : "unsigned", size);
1197 }
1198
1199 abort ();
1200 return BFD_RELOC_NONE;
1201 }
1202
1203 /* Here we decide which fixups can be adjusted to make them relative to
1204 the beginning of the section instead of the symbol. Basically we need
1205 to make sure that the dynamic relocations are done correctly, so in
1206 some cases we force the original symbol to be used. */
1207
1208 int
1209 tc_i386_fix_adjustable (fixP)
1210 fixS *fixP ATTRIBUTE_UNUSED;
1211 {
1212 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1213 if (OUTPUT_FLAVOR != bfd_target_elf_flavour)
1214 return 1;
1215
1216 /* Don't adjust pc-relative references to merge sections in 64-bit
1217 mode. */
1218 if (use_rela_relocations
1219 && (S_GET_SEGMENT (fixP->fx_addsy)->flags & SEC_MERGE) != 0
1220 && fixP->fx_pcrel)
1221 return 0;
1222
1223 /* adjust_reloc_syms doesn't know about the GOT. */
1224 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1225 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1226 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1227 || fixP->fx_r_type == BFD_RELOC_386_TLS_GD
1228 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDM
1229 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDO_32
1230 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE_32
1231 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE
1232 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTIE
1233 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE_32
1234 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE
1235 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1236 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1237 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
1238 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1239 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1240 return 0;
1241 #endif
1242 return 1;
1243 }
1244 #else
1245 #define reloc(SIZE,PCREL,SIGN,OTHER) 0
1246 #define BFD_RELOC_8 0
1247 #define BFD_RELOC_16 0
1248 #define BFD_RELOC_32 0
1249 #define BFD_RELOC_8_PCREL 0
1250 #define BFD_RELOC_16_PCREL 0
1251 #define BFD_RELOC_32_PCREL 0
1252 #define BFD_RELOC_386_PLT32 0
1253 #define BFD_RELOC_386_GOT32 0
1254 #define BFD_RELOC_386_GOTOFF 0
1255 #define BFD_RELOC_386_TLS_GD 0
1256 #define BFD_RELOC_386_TLS_LDM 0
1257 #define BFD_RELOC_386_TLS_LDO_32 0
1258 #define BFD_RELOC_386_TLS_IE_32 0
1259 #define BFD_RELOC_386_TLS_IE 0
1260 #define BFD_RELOC_386_TLS_GOTIE 0
1261 #define BFD_RELOC_386_TLS_LE_32 0
1262 #define BFD_RELOC_386_TLS_LE 0
1263 #define BFD_RELOC_X86_64_PLT32 0
1264 #define BFD_RELOC_X86_64_GOT32 0
1265 #define BFD_RELOC_X86_64_GOTPCREL 0
1266 #endif
1267
1268 static int intel_float_operand PARAMS ((const char *mnemonic));
1269
1270 static int
1271 intel_float_operand (mnemonic)
1272 const char *mnemonic;
1273 {
1274 if (mnemonic[0] == 'f' && mnemonic[1] == 'i')
1275 return 2;
1276
1277 if (mnemonic[0] == 'f')
1278 return 1;
1279
1280 return 0;
1281 }
1282
1283 /* This is the guts of the machine-dependent assembler. LINE points to a
1284 machine dependent instruction. This function is supposed to emit
1285 the frags/bytes it assembles to. */
1286
1287 void
1288 md_assemble (line)
1289 char *line;
1290 {
1291 int j;
1292 char mnemonic[MAX_MNEM_SIZE];
1293
1294 /* Initialize globals. */
1295 memset (&i, '\0', sizeof (i));
1296 for (j = 0; j < MAX_OPERANDS; j++)
1297 i.reloc[j] = NO_RELOC;
1298 memset (disp_expressions, '\0', sizeof (disp_expressions));
1299 memset (im_expressions, '\0', sizeof (im_expressions));
1300 save_stack_p = save_stack;
1301
1302 /* First parse an instruction mnemonic & call i386_operand for the operands.
1303 We assume that the scrubber has arranged it so that line[0] is the valid
1304 start of a (possibly prefixed) mnemonic. */
1305
1306 line = parse_insn (line, mnemonic);
1307 if (line == NULL)
1308 return;
1309
1310 line = parse_operands (line, mnemonic);
1311 if (line == NULL)
1312 return;
1313
1314 /* Now we've parsed the mnemonic into a set of templates, and have the
1315 operands at hand. */
1316
1317 /* All intel opcodes have reversed operands except for "bound" and
1318 "enter". We also don't reverse intersegment "jmp" and "call"
1319 instructions with 2 immediate operands so that the immediate segment
1320 precedes the offset, as it does when in AT&T mode. "enter" and the
1321 intersegment "jmp" and "call" instructions are the only ones that
1322 have two immediate operands. */
1323 if (intel_syntax && i.operands > 1
1324 && (strcmp (mnemonic, "bound") != 0)
1325 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1326 swap_operands ();
1327
1328 if (i.imm_operands)
1329 optimize_imm ();
1330
1331 if (i.disp_operands)
1332 optimize_disp ();
1333
1334 /* Next, we find a template that matches the given insn,
1335 making sure the overlap of the given operands types is consistent
1336 with the template operand types. */
1337
1338 if (!match_template ())
1339 return;
1340
1341 if (intel_syntax)
1342 {
1343 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1344 if (SYSV386_COMPAT
1345 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1346 i.tm.base_opcode ^= FloatR;
1347
1348 /* Zap movzx and movsx suffix. The suffix may have been set from
1349 "word ptr" or "byte ptr" on the source operand, but we'll use
1350 the suffix later to choose the destination register. */
1351 if ((i.tm.base_opcode & ~9) == 0x0fb6)
1352 i.suffix = 0;
1353 }
1354
1355 if (i.tm.opcode_modifier & FWait)
1356 if (!add_prefix (FWAIT_OPCODE))
1357 return;
1358
1359 /* Check string instruction segment overrides. */
1360 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1361 {
1362 if (!check_string ())
1363 return;
1364 }
1365
1366 if (!process_suffix ())
1367 return;
1368
1369 /* Make still unresolved immediate matches conform to size of immediate
1370 given in i.suffix. */
1371 if (!finalize_imm ())
1372 return;
1373
1374 if (i.types[0] & Imm1)
1375 i.imm_operands = 0; /* kludge for shift insns. */
1376 if (i.types[0] & ImplicitRegister)
1377 i.reg_operands--;
1378 if (i.types[1] & ImplicitRegister)
1379 i.reg_operands--;
1380 if (i.types[2] & ImplicitRegister)
1381 i.reg_operands--;
1382
1383 if (i.tm.opcode_modifier & ImmExt)
1384 {
1385 /* These AMD 3DNow! and Intel Katmai New Instructions have an
1386 opcode suffix which is coded in the same place as an 8-bit
1387 immediate field would be. Here we fake an 8-bit immediate
1388 operand from the opcode suffix stored in tm.extension_opcode. */
1389
1390 expressionS *exp;
1391
1392 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
1393
1394 exp = &im_expressions[i.imm_operands++];
1395 i.op[i.operands].imms = exp;
1396 i.types[i.operands++] = Imm8;
1397 exp->X_op = O_constant;
1398 exp->X_add_number = i.tm.extension_opcode;
1399 i.tm.extension_opcode = None;
1400 }
1401
1402 /* For insns with operands there are more diddles to do to the opcode. */
1403 if (i.operands)
1404 {
1405 if (!process_operands ())
1406 return;
1407 }
1408 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
1409 {
1410 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
1411 as_warn (_("translating to `%sp'"), i.tm.name);
1412 }
1413
1414 /* Handle conversion of 'int $3' --> special int3 insn. */
1415 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
1416 {
1417 i.tm.base_opcode = INT3_OPCODE;
1418 i.imm_operands = 0;
1419 }
1420
1421 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
1422 && i.op[0].disps->X_op == O_constant)
1423 {
1424 /* Convert "jmp constant" (and "call constant") to a jump (call) to
1425 the absolute address given by the constant. Since ix86 jumps and
1426 calls are pc relative, we need to generate a reloc. */
1427 i.op[0].disps->X_add_symbol = &abs_symbol;
1428 i.op[0].disps->X_op = O_symbol;
1429 }
1430
1431 if ((i.tm.opcode_modifier & Rex64) != 0)
1432 i.rex |= REX_MODE64;
1433
1434 /* For 8 bit registers we need an empty rex prefix. Also if the
1435 instruction already has a prefix, we need to convert old
1436 registers to new ones. */
1437
1438 if (((i.types[0] & Reg8) != 0
1439 && (i.op[0].regs->reg_flags & RegRex64) != 0)
1440 || ((i.types[1] & Reg8) != 0
1441 && (i.op[1].regs->reg_flags & RegRex64) != 0)
1442 || (((i.types[0] & Reg8) != 0 || (i.types[1] & Reg8) != 0)
1443 && i.rex != 0))
1444 {
1445 int x;
1446
1447 i.rex |= REX_OPCODE;
1448 for (x = 0; x < 2; x++)
1449 {
1450 /* Look for 8 bit operand that uses old registers. */
1451 if ((i.types[x] & Reg8) != 0
1452 && (i.op[x].regs->reg_flags & RegRex64) == 0)
1453 {
1454 /* In case it is "hi" register, give up. */
1455 if (i.op[x].regs->reg_num > 3)
1456 as_bad (_("can't encode register '%%%s' in an instruction requiring REX prefix.\n"),
1457 i.op[x].regs->reg_name);
1458
1459 /* Otherwise it is equivalent to the extended register.
1460 Since the encoding doesn't change this is merely
1461 cosmetic cleanup for debug output. */
1462
1463 i.op[x].regs = i.op[x].regs + 8;
1464 }
1465 }
1466 }
1467
1468 if (i.rex != 0)
1469 add_prefix (REX_OPCODE | i.rex);
1470
1471 /* We are ready to output the insn. */
1472 output_insn ();
1473 }
1474
1475 static char *
1476 parse_insn (line, mnemonic)
1477 char *line;
1478 char *mnemonic;
1479 {
1480 char *l = line;
1481 char *token_start = l;
1482 char *mnem_p;
1483
1484 /* Non-zero if we found a prefix only acceptable with string insns. */
1485 const char *expecting_string_instruction = NULL;
1486
1487 while (1)
1488 {
1489 mnem_p = mnemonic;
1490 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1491 {
1492 mnem_p++;
1493 if (mnem_p >= mnemonic + MAX_MNEM_SIZE)
1494 {
1495 as_bad (_("no such instruction: `%s'"), token_start);
1496 return NULL;
1497 }
1498 l++;
1499 }
1500 if (!is_space_char (*l)
1501 && *l != END_OF_INSN
1502 && *l != PREFIX_SEPARATOR
1503 && *l != ',')
1504 {
1505 as_bad (_("invalid character %s in mnemonic"),
1506 output_invalid (*l));
1507 return NULL;
1508 }
1509 if (token_start == l)
1510 {
1511 if (*l == PREFIX_SEPARATOR)
1512 as_bad (_("expecting prefix; got nothing"));
1513 else
1514 as_bad (_("expecting mnemonic; got nothing"));
1515 return NULL;
1516 }
1517
1518 /* Look up instruction (or prefix) via hash table. */
1519 current_templates = hash_find (op_hash, mnemonic);
1520
1521 if (*l != END_OF_INSN
1522 && (!is_space_char (*l) || l[1] != END_OF_INSN)
1523 && current_templates
1524 && (current_templates->start->opcode_modifier & IsPrefix))
1525 {
1526 /* If we are in 16-bit mode, do not allow addr16 or data16.
1527 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1528 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1529 && flag_code != CODE_64BIT
1530 && (((current_templates->start->opcode_modifier & Size32) != 0)
1531 ^ (flag_code == CODE_16BIT)))
1532 {
1533 as_bad (_("redundant %s prefix"),
1534 current_templates->start->name);
1535 return NULL;
1536 }
1537 /* Add prefix, checking for repeated prefixes. */
1538 switch (add_prefix (current_templates->start->base_opcode))
1539 {
1540 case 0:
1541 return NULL;
1542 case 2:
1543 expecting_string_instruction = current_templates->start->name;
1544 break;
1545 }
1546 /* Skip past PREFIX_SEPARATOR and reset token_start. */
1547 token_start = ++l;
1548 }
1549 else
1550 break;
1551 }
1552
1553 if (!current_templates)
1554 {
1555 /* See if we can get a match by trimming off a suffix. */
1556 switch (mnem_p[-1])
1557 {
1558 case WORD_MNEM_SUFFIX:
1559 case BYTE_MNEM_SUFFIX:
1560 case QWORD_MNEM_SUFFIX:
1561 i.suffix = mnem_p[-1];
1562 mnem_p[-1] = '\0';
1563 current_templates = hash_find (op_hash, mnemonic);
1564 break;
1565 case SHORT_MNEM_SUFFIX:
1566 case LONG_MNEM_SUFFIX:
1567 if (!intel_syntax)
1568 {
1569 i.suffix = mnem_p[-1];
1570 mnem_p[-1] = '\0';
1571 current_templates = hash_find (op_hash, mnemonic);
1572 }
1573 break;
1574
1575 /* Intel Syntax. */
1576 case 'd':
1577 if (intel_syntax)
1578 {
1579 if (intel_float_operand (mnemonic))
1580 i.suffix = SHORT_MNEM_SUFFIX;
1581 else
1582 i.suffix = LONG_MNEM_SUFFIX;
1583 mnem_p[-1] = '\0';
1584 current_templates = hash_find (op_hash, mnemonic);
1585 }
1586 break;
1587 }
1588 if (!current_templates)
1589 {
1590 as_bad (_("no such instruction: `%s'"), token_start);
1591 return NULL;
1592 }
1593 }
1594
1595 if (current_templates->start->opcode_modifier & (Jump | JumpByte))
1596 {
1597 /* Check for a branch hint. We allow ",pt" and ",pn" for
1598 predict taken and predict not taken respectively.
1599 I'm not sure that branch hints actually do anything on loop
1600 and jcxz insns (JumpByte) for current Pentium4 chips. They
1601 may work in the future and it doesn't hurt to accept them
1602 now. */
1603 if (l[0] == ',' && l[1] == 'p')
1604 {
1605 if (l[2] == 't')
1606 {
1607 if (!add_prefix (DS_PREFIX_OPCODE))
1608 return NULL;
1609 l += 3;
1610 }
1611 else if (l[2] == 'n')
1612 {
1613 if (!add_prefix (CS_PREFIX_OPCODE))
1614 return NULL;
1615 l += 3;
1616 }
1617 }
1618 }
1619 /* Any other comma loses. */
1620 if (*l == ',')
1621 {
1622 as_bad (_("invalid character %s in mnemonic"),
1623 output_invalid (*l));
1624 return NULL;
1625 }
1626
1627 /* Check if instruction is supported on specified architecture. */
1628 if ((current_templates->start->cpu_flags & ~(Cpu64 | CpuNo64))
1629 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64)))
1630 {
1631 as_warn (_("`%s' is not supported on `%s'"),
1632 current_templates->start->name, cpu_arch_name);
1633 }
1634 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
1635 {
1636 as_warn (_("use .code16 to ensure correct addressing mode"));
1637 }
1638
1639 /* Check for rep/repne without a string instruction. */
1640 if (expecting_string_instruction
1641 && !(current_templates->start->opcode_modifier & IsString))
1642 {
1643 as_bad (_("expecting string instruction after `%s'"),
1644 expecting_string_instruction);
1645 return NULL;
1646 }
1647
1648 return l;
1649 }
1650
1651 static char *
1652 parse_operands (l, mnemonic)
1653 char *l;
1654 const char *mnemonic;
1655 {
1656 char *token_start;
1657
1658 /* 1 if operand is pending after ','. */
1659 unsigned int expecting_operand = 0;
1660
1661 /* Non-zero if operand parens not balanced. */
1662 unsigned int paren_not_balanced;
1663
1664 while (*l != END_OF_INSN)
1665 {
1666 /* Skip optional white space before operand. */
1667 if (is_space_char (*l))
1668 ++l;
1669 if (!is_operand_char (*l) && *l != END_OF_INSN)
1670 {
1671 as_bad (_("invalid character %s before operand %d"),
1672 output_invalid (*l),
1673 i.operands + 1);
1674 return NULL;
1675 }
1676 token_start = l; /* after white space */
1677 paren_not_balanced = 0;
1678 while (paren_not_balanced || *l != ',')
1679 {
1680 if (*l == END_OF_INSN)
1681 {
1682 if (paren_not_balanced)
1683 {
1684 if (!intel_syntax)
1685 as_bad (_("unbalanced parenthesis in operand %d."),
1686 i.operands + 1);
1687 else
1688 as_bad (_("unbalanced brackets in operand %d."),
1689 i.operands + 1);
1690 return NULL;
1691 }
1692 else
1693 break; /* we are done */
1694 }
1695 else if (!is_operand_char (*l) && !is_space_char (*l))
1696 {
1697 as_bad (_("invalid character %s in operand %d"),
1698 output_invalid (*l),
1699 i.operands + 1);
1700 return NULL;
1701 }
1702 if (!intel_syntax)
1703 {
1704 if (*l == '(')
1705 ++paren_not_balanced;
1706 if (*l == ')')
1707 --paren_not_balanced;
1708 }
1709 else
1710 {
1711 if (*l == '[')
1712 ++paren_not_balanced;
1713 if (*l == ']')
1714 --paren_not_balanced;
1715 }
1716 l++;
1717 }
1718 if (l != token_start)
1719 { /* Yes, we've read in another operand. */
1720 unsigned int operand_ok;
1721 this_operand = i.operands++;
1722 if (i.operands > MAX_OPERANDS)
1723 {
1724 as_bad (_("spurious operands; (%d operands/instruction max)"),
1725 MAX_OPERANDS);
1726 return NULL;
1727 }
1728 /* Now parse operand adding info to 'i' as we go along. */
1729 END_STRING_AND_SAVE (l);
1730
1731 if (intel_syntax)
1732 operand_ok =
1733 i386_intel_operand (token_start,
1734 intel_float_operand (mnemonic));
1735 else
1736 operand_ok = i386_operand (token_start);
1737
1738 RESTORE_END_STRING (l);
1739 if (!operand_ok)
1740 return NULL;
1741 }
1742 else
1743 {
1744 if (expecting_operand)
1745 {
1746 expecting_operand_after_comma:
1747 as_bad (_("expecting operand after ','; got nothing"));
1748 return NULL;
1749 }
1750 if (*l == ',')
1751 {
1752 as_bad (_("expecting operand before ','; got nothing"));
1753 return NULL;
1754 }
1755 }
1756
1757 /* Now *l must be either ',' or END_OF_INSN. */
1758 if (*l == ',')
1759 {
1760 if (*++l == END_OF_INSN)
1761 {
1762 /* Just skip it, if it's \n complain. */
1763 goto expecting_operand_after_comma;
1764 }
1765 expecting_operand = 1;
1766 }
1767 }
1768 return l;
1769 }
1770
1771 static void
1772 swap_operands ()
1773 {
1774 union i386_op temp_op;
1775 unsigned int temp_type;
1776 RELOC_ENUM temp_reloc;
1777 int xchg1 = 0;
1778 int xchg2 = 0;
1779
1780 if (i.operands == 2)
1781 {
1782 xchg1 = 0;
1783 xchg2 = 1;
1784 }
1785 else if (i.operands == 3)
1786 {
1787 xchg1 = 0;
1788 xchg2 = 2;
1789 }
1790 temp_type = i.types[xchg2];
1791 i.types[xchg2] = i.types[xchg1];
1792 i.types[xchg1] = temp_type;
1793 temp_op = i.op[xchg2];
1794 i.op[xchg2] = i.op[xchg1];
1795 i.op[xchg1] = temp_op;
1796 temp_reloc = i.reloc[xchg2];
1797 i.reloc[xchg2] = i.reloc[xchg1];
1798 i.reloc[xchg1] = temp_reloc;
1799
1800 if (i.mem_operands == 2)
1801 {
1802 const seg_entry *temp_seg;
1803 temp_seg = i.seg[0];
1804 i.seg[0] = i.seg[1];
1805 i.seg[1] = temp_seg;
1806 }
1807 }
1808
1809 /* Try to ensure constant immediates are represented in the smallest
1810 opcode possible. */
1811 static void
1812 optimize_imm ()
1813 {
1814 char guess_suffix = 0;
1815 int op;
1816
1817 if (i.suffix)
1818 guess_suffix = i.suffix;
1819 else if (i.reg_operands)
1820 {
1821 /* Figure out a suffix from the last register operand specified.
1822 We can't do this properly yet, ie. excluding InOutPortReg,
1823 but the following works for instructions with immediates.
1824 In any case, we can't set i.suffix yet. */
1825 for (op = i.operands; --op >= 0;)
1826 if (i.types[op] & Reg)
1827 {
1828 if (i.types[op] & Reg8)
1829 guess_suffix = BYTE_MNEM_SUFFIX;
1830 else if (i.types[op] & Reg16)
1831 guess_suffix = WORD_MNEM_SUFFIX;
1832 else if (i.types[op] & Reg32)
1833 guess_suffix = LONG_MNEM_SUFFIX;
1834 else if (i.types[op] & Reg64)
1835 guess_suffix = QWORD_MNEM_SUFFIX;
1836 break;
1837 }
1838 }
1839 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
1840 guess_suffix = WORD_MNEM_SUFFIX;
1841
1842 for (op = i.operands; --op >= 0;)
1843 if (i.types[op] & Imm)
1844 {
1845 switch (i.op[op].imms->X_op)
1846 {
1847 case O_constant:
1848 /* If a suffix is given, this operand may be shortened. */
1849 switch (guess_suffix)
1850 {
1851 case LONG_MNEM_SUFFIX:
1852 i.types[op] |= Imm32 | Imm64;
1853 break;
1854 case WORD_MNEM_SUFFIX:
1855 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
1856 break;
1857 case BYTE_MNEM_SUFFIX:
1858 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
1859 break;
1860 }
1861
1862 /* If this operand is at most 16 bits, convert it
1863 to a signed 16 bit number before trying to see
1864 whether it will fit in an even smaller size.
1865 This allows a 16-bit operand such as $0xffe0 to
1866 be recognised as within Imm8S range. */
1867 if ((i.types[op] & Imm16)
1868 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
1869 {
1870 i.op[op].imms->X_add_number =
1871 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
1872 }
1873 if ((i.types[op] & Imm32)
1874 && ((i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1))
1875 == 0))
1876 {
1877 i.op[op].imms->X_add_number = ((i.op[op].imms->X_add_number
1878 ^ ((offsetT) 1 << 31))
1879 - ((offsetT) 1 << 31));
1880 }
1881 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
1882
1883 /* We must avoid matching of Imm32 templates when 64bit
1884 only immediate is available. */
1885 if (guess_suffix == QWORD_MNEM_SUFFIX)
1886 i.types[op] &= ~Imm32;
1887 break;
1888
1889 case O_absent:
1890 case O_register:
1891 abort ();
1892
1893 /* Symbols and expressions. */
1894 default:
1895 /* Convert symbolic operand to proper sizes for matching. */
1896 switch (guess_suffix)
1897 {
1898 case QWORD_MNEM_SUFFIX:
1899 i.types[op] = Imm64 | Imm32S;
1900 break;
1901 case LONG_MNEM_SUFFIX:
1902 i.types[op] = Imm32 | Imm64;
1903 break;
1904 case WORD_MNEM_SUFFIX:
1905 i.types[op] = Imm16 | Imm32 | Imm64;
1906 break;
1907 break;
1908 case BYTE_MNEM_SUFFIX:
1909 i.types[op] = Imm8 | Imm8S | Imm16 | Imm32S | Imm32;
1910 break;
1911 break;
1912 }
1913 break;
1914 }
1915 }
1916 }
1917
1918 /* Try to use the smallest displacement type too. */
1919 static void
1920 optimize_disp ()
1921 {
1922 int op;
1923
1924 for (op = i.operands; --op >= 0;)
1925 if ((i.types[op] & Disp) && i.op[op].disps->X_op == O_constant)
1926 {
1927 offsetT disp = i.op[op].disps->X_add_number;
1928
1929 if (i.types[op] & Disp16)
1930 {
1931 /* We know this operand is at most 16 bits, so
1932 convert to a signed 16 bit number before trying
1933 to see whether it will fit in an even smaller
1934 size. */
1935
1936 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
1937 }
1938 else if (i.types[op] & Disp32)
1939 {
1940 /* We know this operand is at most 32 bits, so convert to a
1941 signed 32 bit number before trying to see whether it will
1942 fit in an even smaller size. */
1943 disp &= (((offsetT) 2 << 31) - 1);
1944 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1945 }
1946 if (flag_code == CODE_64BIT)
1947 {
1948 if (fits_in_signed_long (disp))
1949 i.types[op] |= Disp32S;
1950 if (fits_in_unsigned_long (disp))
1951 i.types[op] |= Disp32;
1952 }
1953 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
1954 && fits_in_signed_byte (disp))
1955 i.types[op] |= Disp8;
1956 }
1957 }
1958
1959 static int
1960 match_template ()
1961 {
1962 /* Points to template once we've found it. */
1963 const template *t;
1964 unsigned int overlap0, overlap1, overlap2;
1965 unsigned int found_reverse_match;
1966 int suffix_check;
1967
1968 #define MATCH(overlap, given, template) \
1969 ((overlap & ~JumpAbsolute) \
1970 && (((given) & (BaseIndex | JumpAbsolute)) \
1971 == ((overlap) & (BaseIndex | JumpAbsolute))))
1972
1973 /* If given types r0 and r1 are registers they must be of the same type
1974 unless the expected operand type register overlap is null.
1975 Note that Acc in a template matches every size of reg. */
1976 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
1977 (((g0) & Reg) == 0 || ((g1) & Reg) == 0 \
1978 || ((g0) & Reg) == ((g1) & Reg) \
1979 || ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
1980
1981 overlap0 = 0;
1982 overlap1 = 0;
1983 overlap2 = 0;
1984 found_reverse_match = 0;
1985 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
1986 ? No_bSuf
1987 : (i.suffix == WORD_MNEM_SUFFIX
1988 ? No_wSuf
1989 : (i.suffix == SHORT_MNEM_SUFFIX
1990 ? No_sSuf
1991 : (i.suffix == LONG_MNEM_SUFFIX
1992 ? No_lSuf
1993 : (i.suffix == QWORD_MNEM_SUFFIX
1994 ? No_qSuf
1995 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX
1996 ? No_xSuf : 0))))));
1997
1998 for (t = current_templates->start;
1999 t < current_templates->end;
2000 t++)
2001 {
2002 /* Must have right number of operands. */
2003 if (i.operands != t->operands)
2004 continue;
2005
2006 /* Check the suffix, except for some instructions in intel mode. */
2007 if ((t->opcode_modifier & suffix_check)
2008 && !(intel_syntax
2009 && (t->opcode_modifier & IgnoreSize))
2010 && !(intel_syntax
2011 && t->base_opcode == 0xd9
2012 && (t->extension_opcode == 5 /* 0xd9,5 "fldcw" */
2013 || t->extension_opcode == 7))) /* 0xd9,7 "f{n}stcw" */
2014 continue;
2015
2016 /* Do not verify operands when there are none. */
2017 else if (!t->operands)
2018 {
2019 if (t->cpu_flags & ~cpu_arch_flags)
2020 continue;
2021 /* We've found a match; break out of loop. */
2022 break;
2023 }
2024
2025 overlap0 = i.types[0] & t->operand_types[0];
2026 switch (t->operands)
2027 {
2028 case 1:
2029 if (!MATCH (overlap0, i.types[0], t->operand_types[0]))
2030 continue;
2031 break;
2032 case 2:
2033 case 3:
2034 overlap1 = i.types[1] & t->operand_types[1];
2035 if (!MATCH (overlap0, i.types[0], t->operand_types[0])
2036 || !MATCH (overlap1, i.types[1], t->operand_types[1])
2037 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2038 t->operand_types[0],
2039 overlap1, i.types[1],
2040 t->operand_types[1]))
2041 {
2042 /* Check if other direction is valid ... */
2043 if ((t->opcode_modifier & (D | FloatD)) == 0)
2044 continue;
2045
2046 /* Try reversing direction of operands. */
2047 overlap0 = i.types[0] & t->operand_types[1];
2048 overlap1 = i.types[1] & t->operand_types[0];
2049 if (!MATCH (overlap0, i.types[0], t->operand_types[1])
2050 || !MATCH (overlap1, i.types[1], t->operand_types[0])
2051 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2052 t->operand_types[1],
2053 overlap1, i.types[1],
2054 t->operand_types[0]))
2055 {
2056 /* Does not match either direction. */
2057 continue;
2058 }
2059 /* found_reverse_match holds which of D or FloatDR
2060 we've found. */
2061 found_reverse_match = t->opcode_modifier & (D | FloatDR);
2062 }
2063 /* Found a forward 2 operand match here. */
2064 else if (t->operands == 3)
2065 {
2066 /* Here we make use of the fact that there are no
2067 reverse match 3 operand instructions, and all 3
2068 operand instructions only need to be checked for
2069 register consistency between operands 2 and 3. */
2070 overlap2 = i.types[2] & t->operand_types[2];
2071 if (!MATCH (overlap2, i.types[2], t->operand_types[2])
2072 || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1],
2073 t->operand_types[1],
2074 overlap2, i.types[2],
2075 t->operand_types[2]))
2076
2077 continue;
2078 }
2079 /* Found either forward/reverse 2 or 3 operand match here:
2080 slip through to break. */
2081 }
2082 if (t->cpu_flags & ~cpu_arch_flags)
2083 {
2084 found_reverse_match = 0;
2085 continue;
2086 }
2087 /* We've found a match; break out of loop. */
2088 break;
2089 }
2090
2091 if (t == current_templates->end)
2092 {
2093 /* We found no match. */
2094 as_bad (_("suffix or operands invalid for `%s'"),
2095 current_templates->start->name);
2096 return 0;
2097 }
2098
2099 if (!quiet_warnings)
2100 {
2101 if (!intel_syntax
2102 && ((i.types[0] & JumpAbsolute)
2103 != (t->operand_types[0] & JumpAbsolute)))
2104 {
2105 as_warn (_("indirect %s without `*'"), t->name);
2106 }
2107
2108 if ((t->opcode_modifier & (IsPrefix | IgnoreSize))
2109 == (IsPrefix | IgnoreSize))
2110 {
2111 /* Warn them that a data or address size prefix doesn't
2112 affect assembly of the next line of code. */
2113 as_warn (_("stand-alone `%s' prefix"), t->name);
2114 }
2115 }
2116
2117 /* Copy the template we found. */
2118 i.tm = *t;
2119 if (found_reverse_match)
2120 {
2121 /* If we found a reverse match we must alter the opcode
2122 direction bit. found_reverse_match holds bits to change
2123 (different for int & float insns). */
2124
2125 i.tm.base_opcode ^= found_reverse_match;
2126
2127 i.tm.operand_types[0] = t->operand_types[1];
2128 i.tm.operand_types[1] = t->operand_types[0];
2129 }
2130
2131 return 1;
2132 }
2133
2134 static int
2135 check_string ()
2136 {
2137 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
2138 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
2139 {
2140 if (i.seg[0] != NULL && i.seg[0] != &es)
2141 {
2142 as_bad (_("`%s' operand %d must use `%%es' segment"),
2143 i.tm.name,
2144 mem_op + 1);
2145 return 0;
2146 }
2147 /* There's only ever one segment override allowed per instruction.
2148 This instruction possibly has a legal segment override on the
2149 second operand, so copy the segment to where non-string
2150 instructions store it, allowing common code. */
2151 i.seg[0] = i.seg[1];
2152 }
2153 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
2154 {
2155 if (i.seg[1] != NULL && i.seg[1] != &es)
2156 {
2157 as_bad (_("`%s' operand %d must use `%%es' segment"),
2158 i.tm.name,
2159 mem_op + 2);
2160 return 0;
2161 }
2162 }
2163 return 1;
2164 }
2165
2166 static int
2167 process_suffix ()
2168 {
2169 /* If matched instruction specifies an explicit instruction mnemonic
2170 suffix, use it. */
2171 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
2172 {
2173 if (i.tm.opcode_modifier & Size16)
2174 i.suffix = WORD_MNEM_SUFFIX;
2175 else if (i.tm.opcode_modifier & Size64)
2176 i.suffix = QWORD_MNEM_SUFFIX;
2177 else
2178 i.suffix = LONG_MNEM_SUFFIX;
2179 }
2180 else if (i.reg_operands)
2181 {
2182 /* If there's no instruction mnemonic suffix we try to invent one
2183 based on register operands. */
2184 if (!i.suffix)
2185 {
2186 /* We take i.suffix from the last register operand specified,
2187 Destination register type is more significant than source
2188 register type. */
2189 int op;
2190 for (op = i.operands; --op >= 0;)
2191 if ((i.types[op] & Reg)
2192 && !(i.tm.operand_types[op] & InOutPortReg))
2193 {
2194 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
2195 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
2196 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
2197 LONG_MNEM_SUFFIX);
2198 break;
2199 }
2200 }
2201 else if (i.suffix == BYTE_MNEM_SUFFIX)
2202 {
2203 if (!check_byte_reg ())
2204 return 0;
2205 }
2206 else if (i.suffix == LONG_MNEM_SUFFIX)
2207 {
2208 if (!check_long_reg ())
2209 return 0;
2210 }
2211 else if (i.suffix == QWORD_MNEM_SUFFIX)
2212 {
2213 if (!check_qword_reg ())
2214 return 0;
2215 }
2216 else if (i.suffix == WORD_MNEM_SUFFIX)
2217 {
2218 if (!check_word_reg ())
2219 return 0;
2220 }
2221 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2222 /* Do nothing if the instruction is going to ignore the prefix. */
2223 ;
2224 else
2225 abort ();
2226 }
2227 else if ((i.tm.opcode_modifier & DefaultSize) && !i.suffix)
2228 {
2229 i.suffix = stackop_size;
2230 }
2231
2232 /* Change the opcode based on the operand size given by i.suffix;
2233 We need not change things for byte insns. */
2234
2235 if (!i.suffix && (i.tm.opcode_modifier & W))
2236 {
2237 as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
2238 return 0;
2239 }
2240
2241 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2242 {
2243 /* It's not a byte, select word/dword operation. */
2244 if (i.tm.opcode_modifier & W)
2245 {
2246 if (i.tm.opcode_modifier & ShortForm)
2247 i.tm.base_opcode |= 8;
2248 else
2249 i.tm.base_opcode |= 1;
2250 }
2251
2252 /* Now select between word & dword operations via the operand
2253 size prefix, except for instructions that will ignore this
2254 prefix anyway. */
2255 if (i.suffix != QWORD_MNEM_SUFFIX
2256 && !(i.tm.opcode_modifier & IgnoreSize)
2257 && ((i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2258 || (flag_code == CODE_64BIT
2259 && (i.tm.opcode_modifier & JumpByte))))
2260 {
2261 unsigned int prefix = DATA_PREFIX_OPCODE;
2262 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2263 prefix = ADDR_PREFIX_OPCODE;
2264
2265 if (!add_prefix (prefix))
2266 return 0;
2267 }
2268
2269 /* Set mode64 for an operand. */
2270 if (i.suffix == QWORD_MNEM_SUFFIX
2271 && flag_code == CODE_64BIT
2272 && (i.tm.opcode_modifier & NoRex64) == 0)
2273 i.rex |= REX_MODE64;
2274
2275 /* Size floating point instruction. */
2276 if (i.suffix == LONG_MNEM_SUFFIX)
2277 {
2278 if (i.tm.opcode_modifier & FloatMF)
2279 i.tm.base_opcode ^= 4;
2280 }
2281 }
2282
2283 return 1;
2284 }
2285
2286 static int
2287 check_byte_reg ()
2288 {
2289 int op;
2290 for (op = i.operands; --op >= 0;)
2291 {
2292 /* If this is an eight bit register, it's OK. If it's the 16 or
2293 32 bit version of an eight bit register, we will just use the
2294 low portion, and that's OK too. */
2295 if (i.types[op] & Reg8)
2296 continue;
2297
2298 /* movzx and movsx should not generate this warning. */
2299 if (intel_syntax
2300 && (i.tm.base_opcode == 0xfb7
2301 || i.tm.base_opcode == 0xfb6
2302 || i.tm.base_opcode == 0x63
2303 || i.tm.base_opcode == 0xfbe
2304 || i.tm.base_opcode == 0xfbf))
2305 continue;
2306
2307 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4
2308 #if 0
2309 /* Check that the template allows eight bit regs. This
2310 kills insns such as `orb $1,%edx', which maybe should be
2311 allowed. */
2312 && (i.tm.operand_types[op] & (Reg8 | InOutPortReg))
2313 #endif
2314 )
2315 {
2316 /* Prohibit these changes in the 64bit mode, since the
2317 lowering is more complicated. */
2318 if (flag_code == CODE_64BIT
2319 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2320 {
2321 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2322 i.op[op].regs->reg_name,
2323 i.suffix);
2324 return 0;
2325 }
2326 #if REGISTER_WARNINGS
2327 if (!quiet_warnings
2328 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2329 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2330 (i.op[op].regs + (i.types[op] & Reg16
2331 ? REGNAM_AL - REGNAM_AX
2332 : REGNAM_AL - REGNAM_EAX))->reg_name,
2333 i.op[op].regs->reg_name,
2334 i.suffix);
2335 #endif
2336 continue;
2337 }
2338 /* Any other register is bad. */
2339 if (i.types[op] & (Reg | RegMMX | RegXMM
2340 | SReg2 | SReg3
2341 | Control | Debug | Test
2342 | FloatReg | FloatAcc))
2343 {
2344 as_bad (_("`%%%s' not allowed with `%s%c'"),
2345 i.op[op].regs->reg_name,
2346 i.tm.name,
2347 i.suffix);
2348 return 0;
2349 }
2350 }
2351 return 1;
2352 }
2353
2354 static int
2355 check_long_reg ()
2356 {
2357 int op;
2358
2359 for (op = i.operands; --op >= 0;)
2360 /* Reject eight bit registers, except where the template requires
2361 them. (eg. movzb) */
2362 if ((i.types[op] & Reg8) != 0
2363 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2364 {
2365 as_bad (_("`%%%s' not allowed with `%s%c'"),
2366 i.op[op].regs->reg_name,
2367 i.tm.name,
2368 i.suffix);
2369 return 0;
2370 }
2371 /* Warn if the e prefix on a general reg is missing. */
2372 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2373 && (i.types[op] & Reg16) != 0
2374 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2375 {
2376 /* Prohibit these changes in the 64bit mode, since the
2377 lowering is more complicated. */
2378 if (flag_code == CODE_64BIT)
2379 {
2380 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2381 i.op[op].regs->reg_name,
2382 i.suffix);
2383 return 0;
2384 }
2385 #if REGISTER_WARNINGS
2386 else
2387 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2388 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
2389 i.op[op].regs->reg_name,
2390 i.suffix);
2391 #endif
2392 }
2393 /* Warn if the r prefix on a general reg is missing. */
2394 else if ((i.types[op] & Reg64) != 0
2395 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2396 {
2397 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2398 i.op[op].regs->reg_name,
2399 i.suffix);
2400 return 0;
2401 }
2402 return 1;
2403 }
2404
2405 static int
2406 check_qword_reg ()
2407 {
2408 int op;
2409
2410 for (op = i.operands; --op >= 0; )
2411 /* Reject eight bit registers, except where the template requires
2412 them. (eg. movzb) */
2413 if ((i.types[op] & Reg8) != 0
2414 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2415 {
2416 as_bad (_("`%%%s' not allowed with `%s%c'"),
2417 i.op[op].regs->reg_name,
2418 i.tm.name,
2419 i.suffix);
2420 return 0;
2421 }
2422 /* Warn if the e prefix on a general reg is missing. */
2423 else if (((i.types[op] & Reg16) != 0
2424 || (i.types[op] & Reg32) != 0)
2425 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2426 {
2427 /* Prohibit these changes in the 64bit mode, since the
2428 lowering is more complicated. */
2429 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2430 i.op[op].regs->reg_name,
2431 i.suffix);
2432 return 0;
2433 }
2434 return 1;
2435 }
2436
2437 static int
2438 check_word_reg ()
2439 {
2440 int op;
2441 for (op = i.operands; --op >= 0;)
2442 /* Reject eight bit registers, except where the template requires
2443 them. (eg. movzb) */
2444 if ((i.types[op] & Reg8) != 0
2445 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2446 {
2447 as_bad (_("`%%%s' not allowed with `%s%c'"),
2448 i.op[op].regs->reg_name,
2449 i.tm.name,
2450 i.suffix);
2451 return 0;
2452 }
2453 /* Warn if the e prefix on a general reg is present. */
2454 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2455 && (i.types[op] & Reg32) != 0
2456 && (i.tm.operand_types[op] & (Reg16 | Acc)) != 0)
2457 {
2458 /* Prohibit these changes in the 64bit mode, since the
2459 lowering is more complicated. */
2460 if (flag_code == CODE_64BIT)
2461 {
2462 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2463 i.op[op].regs->reg_name,
2464 i.suffix);
2465 return 0;
2466 }
2467 else
2468 #if REGISTER_WARNINGS
2469 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2470 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
2471 i.op[op].regs->reg_name,
2472 i.suffix);
2473 #endif
2474 }
2475 return 1;
2476 }
2477
2478 static int
2479 finalize_imm ()
2480 {
2481 unsigned int overlap0, overlap1, overlap2;
2482
2483 overlap0 = i.types[0] & i.tm.operand_types[0];
2484 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S))
2485 && overlap0 != Imm8 && overlap0 != Imm8S
2486 && overlap0 != Imm16 && overlap0 != Imm32S
2487 && overlap0 != Imm32 && overlap0 != Imm64)
2488 {
2489 if (i.suffix)
2490 {
2491 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX
2492 ? Imm8 | Imm8S
2493 : (i.suffix == WORD_MNEM_SUFFIX
2494 ? Imm16
2495 : (i.suffix == QWORD_MNEM_SUFFIX
2496 ? Imm64 | Imm32S
2497 : Imm32)));
2498 }
2499 else if (overlap0 == (Imm16 | Imm32S | Imm32)
2500 || overlap0 == (Imm16 | Imm32)
2501 || overlap0 == (Imm16 | Imm32S))
2502 {
2503 overlap0 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
2504 ? Imm16 : Imm32S);
2505 }
2506 if (overlap0 != Imm8 && overlap0 != Imm8S
2507 && overlap0 != Imm16 && overlap0 != Imm32S
2508 && overlap0 != Imm32 && overlap0 != Imm64)
2509 {
2510 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
2511 return 0;
2512 }
2513 }
2514 i.types[0] = overlap0;
2515
2516 overlap1 = i.types[1] & i.tm.operand_types[1];
2517 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32))
2518 && overlap1 != Imm8 && overlap1 != Imm8S
2519 && overlap1 != Imm16 && overlap1 != Imm32S
2520 && overlap1 != Imm32 && overlap1 != Imm64)
2521 {
2522 if (i.suffix)
2523 {
2524 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX
2525 ? Imm8 | Imm8S
2526 : (i.suffix == WORD_MNEM_SUFFIX
2527 ? Imm16
2528 : (i.suffix == QWORD_MNEM_SUFFIX
2529 ? Imm64 | Imm32S
2530 : Imm32)));
2531 }
2532 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
2533 || overlap1 == (Imm16 | Imm32)
2534 || overlap1 == (Imm16 | Imm32S))
2535 {
2536 overlap1 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
2537 ? Imm16 : Imm32S);
2538 }
2539 if (overlap1 != Imm8 && overlap1 != Imm8S
2540 && overlap1 != Imm16 && overlap1 != Imm32S
2541 && overlap1 != Imm32 && overlap1 != Imm64)
2542 {
2543 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
2544 return 0;
2545 }
2546 }
2547 i.types[1] = overlap1;
2548
2549 overlap2 = i.types[2] & i.tm.operand_types[2];
2550 assert ((overlap2 & Imm) == 0);
2551 i.types[2] = overlap2;
2552
2553 return 1;
2554 }
2555
2556 static int
2557 process_operands ()
2558 {
2559 /* Default segment register this instruction will use for memory
2560 accesses. 0 means unknown. This is only for optimizing out
2561 unnecessary segment overrides. */
2562 const seg_entry *default_seg = 0;
2563
2564 /* The imul $imm, %reg instruction is converted into
2565 imul $imm, %reg, %reg, and the clr %reg instruction
2566 is converted into xor %reg, %reg. */
2567 if (i.tm.opcode_modifier & regKludge)
2568 {
2569 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
2570 /* Pretend we saw the extra register operand. */
2571 assert (i.op[first_reg_op + 1].regs == 0);
2572 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
2573 i.types[first_reg_op + 1] = i.types[first_reg_op];
2574 i.reg_operands = 2;
2575 }
2576
2577 if (i.tm.opcode_modifier & ShortForm)
2578 {
2579 /* The register or float register operand is in operand 0 or 1. */
2580 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
2581 /* Register goes in low 3 bits of opcode. */
2582 i.tm.base_opcode |= i.op[op].regs->reg_num;
2583 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2584 i.rex |= REX_EXTZ;
2585 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2586 {
2587 /* Warn about some common errors, but press on regardless.
2588 The first case can be generated by gcc (<= 2.8.1). */
2589 if (i.operands == 2)
2590 {
2591 /* Reversed arguments on faddp, fsubp, etc. */
2592 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
2593 i.op[1].regs->reg_name,
2594 i.op[0].regs->reg_name);
2595 }
2596 else
2597 {
2598 /* Extraneous `l' suffix on fp insn. */
2599 as_warn (_("translating to `%s %%%s'"), i.tm.name,
2600 i.op[0].regs->reg_name);
2601 }
2602 }
2603 }
2604 else if (i.tm.opcode_modifier & Modrm)
2605 {
2606 /* The opcode is completed (modulo i.tm.extension_opcode which
2607 must be put into the modrm byte).
2608 Now, we make the modrm & index base bytes based on all the
2609 info we've collected. */
2610
2611 default_seg = build_modrm_byte ();
2612 }
2613 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
2614 {
2615 if (i.tm.base_opcode == POP_SEG_SHORT
2616 && i.op[0].regs->reg_num == 1)
2617 {
2618 as_bad (_("you can't `pop %%cs'"));
2619 return 0;
2620 }
2621 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
2622 if ((i.op[0].regs->reg_flags & RegRex) != 0)
2623 i.rex |= REX_EXTZ;
2624 }
2625 else if ((i.tm.base_opcode & ~(D | W)) == MOV_AX_DISP32)
2626 {
2627 default_seg = &ds;
2628 }
2629 else if ((i.tm.opcode_modifier & IsString) != 0)
2630 {
2631 /* For the string instructions that allow a segment override
2632 on one of their operands, the default segment is ds. */
2633 default_seg = &ds;
2634 }
2635
2636 /* If a segment was explicitly specified,
2637 and the specified segment is not the default,
2638 use an opcode prefix to select it.
2639 If we never figured out what the default segment is,
2640 then default_seg will be zero at this point,
2641 and the specified segment prefix will always be used. */
2642 if ((i.seg[0]) && (i.seg[0] != default_seg))
2643 {
2644 if (!add_prefix (i.seg[0]->seg_prefix))
2645 return 0;
2646 }
2647 return 1;
2648 }
2649
2650 static const seg_entry *
2651 build_modrm_byte ()
2652 {
2653 const seg_entry *default_seg = 0;
2654
2655 /* i.reg_operands MUST be the number of real register operands;
2656 implicit registers do not count. */
2657 if (i.reg_operands == 2)
2658 {
2659 unsigned int source, dest;
2660 source = ((i.types[0]
2661 & (Reg | RegMMX | RegXMM
2662 | SReg2 | SReg3
2663 | Control | Debug | Test))
2664 ? 0 : 1);
2665 dest = source + 1;
2666
2667 i.rm.mode = 3;
2668 /* One of the register operands will be encoded in the i.tm.reg
2669 field, the other in the combined i.tm.mode and i.tm.regmem
2670 fields. If no form of this instruction supports a memory
2671 destination operand, then we assume the source operand may
2672 sometimes be a memory operand and so we need to store the
2673 destination in the i.rm.reg field. */
2674 if ((i.tm.operand_types[dest] & AnyMem) == 0)
2675 {
2676 i.rm.reg = i.op[dest].regs->reg_num;
2677 i.rm.regmem = i.op[source].regs->reg_num;
2678 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
2679 i.rex |= REX_EXTX;
2680 if ((i.op[source].regs->reg_flags & RegRex) != 0)
2681 i.rex |= REX_EXTZ;
2682 }
2683 else
2684 {
2685 i.rm.reg = i.op[source].regs->reg_num;
2686 i.rm.regmem = i.op[dest].regs->reg_num;
2687 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
2688 i.rex |= REX_EXTZ;
2689 if ((i.op[source].regs->reg_flags & RegRex) != 0)
2690 i.rex |= REX_EXTX;
2691 }
2692 }
2693 else
2694 { /* If it's not 2 reg operands... */
2695 if (i.mem_operands)
2696 {
2697 unsigned int fake_zero_displacement = 0;
2698 unsigned int op = ((i.types[0] & AnyMem)
2699 ? 0
2700 : (i.types[1] & AnyMem) ? 1 : 2);
2701
2702 default_seg = &ds;
2703
2704 if (i.base_reg == 0)
2705 {
2706 i.rm.mode = 0;
2707 if (!i.disp_operands)
2708 fake_zero_displacement = 1;
2709 if (i.index_reg == 0)
2710 {
2711 /* Operand is just <disp> */
2712 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)
2713 && (flag_code != CODE_64BIT))
2714 {
2715 i.rm.regmem = NO_BASE_REGISTER_16;
2716 i.types[op] &= ~Disp;
2717 i.types[op] |= Disp16;
2718 }
2719 else if (flag_code != CODE_64BIT
2720 || (i.prefix[ADDR_PREFIX] != 0))
2721 {
2722 i.rm.regmem = NO_BASE_REGISTER;
2723 i.types[op] &= ~Disp;
2724 i.types[op] |= Disp32;
2725 }
2726 else
2727 {
2728 /* 64bit mode overwrites the 32bit absolute
2729 addressing by RIP relative addressing and
2730 absolute addressing is encoded by one of the
2731 redundant SIB forms. */
2732 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2733 i.sib.base = NO_BASE_REGISTER;
2734 i.sib.index = NO_INDEX_REGISTER;
2735 i.types[op] &= ~Disp;
2736 i.types[op] |= Disp32S;
2737 }
2738 }
2739 else /* !i.base_reg && i.index_reg */
2740 {
2741 i.sib.index = i.index_reg->reg_num;
2742 i.sib.base = NO_BASE_REGISTER;
2743 i.sib.scale = i.log2_scale_factor;
2744 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2745 i.types[op] &= ~Disp;
2746 if (flag_code != CODE_64BIT)
2747 i.types[op] |= Disp32; /* Must be 32 bit */
2748 else
2749 i.types[op] |= Disp32S;
2750 if ((i.index_reg->reg_flags & RegRex) != 0)
2751 i.rex |= REX_EXTY;
2752 }
2753 }
2754 /* RIP addressing for 64bit mode. */
2755 else if (i.base_reg->reg_type == BaseIndex)
2756 {
2757 i.rm.regmem = NO_BASE_REGISTER;
2758 i.types[op] &= ~Disp;
2759 i.types[op] |= Disp32S;
2760 i.flags[op] = Operand_PCrel;
2761 }
2762 else if (i.base_reg->reg_type & Reg16)
2763 {
2764 switch (i.base_reg->reg_num)
2765 {
2766 case 3: /* (%bx) */
2767 if (i.index_reg == 0)
2768 i.rm.regmem = 7;
2769 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
2770 i.rm.regmem = i.index_reg->reg_num - 6;
2771 break;
2772 case 5: /* (%bp) */
2773 default_seg = &ss;
2774 if (i.index_reg == 0)
2775 {
2776 i.rm.regmem = 6;
2777 if ((i.types[op] & Disp) == 0)
2778 {
2779 /* fake (%bp) into 0(%bp) */
2780 i.types[op] |= Disp8;
2781 fake_zero_displacement = 1;
2782 }
2783 }
2784 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
2785 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
2786 break;
2787 default: /* (%si) -> 4 or (%di) -> 5 */
2788 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
2789 }
2790 i.rm.mode = mode_from_disp_size (i.types[op]);
2791 }
2792 else /* i.base_reg and 32/64 bit mode */
2793 {
2794 if (flag_code == CODE_64BIT
2795 && (i.types[op] & Disp))
2796 {
2797 if (i.types[op] & Disp8)
2798 i.types[op] = Disp8 | Disp32S;
2799 else
2800 i.types[op] = Disp32S;
2801 }
2802 i.rm.regmem = i.base_reg->reg_num;
2803 if ((i.base_reg->reg_flags & RegRex) != 0)
2804 i.rex |= REX_EXTZ;
2805 i.sib.base = i.base_reg->reg_num;
2806 /* x86-64 ignores REX prefix bit here to avoid decoder
2807 complications. */
2808 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
2809 {
2810 default_seg = &ss;
2811 if (i.disp_operands == 0)
2812 {
2813 fake_zero_displacement = 1;
2814 i.types[op] |= Disp8;
2815 }
2816 }
2817 else if (i.base_reg->reg_num == ESP_REG_NUM)
2818 {
2819 default_seg = &ss;
2820 }
2821 i.sib.scale = i.log2_scale_factor;
2822 if (i.index_reg == 0)
2823 {
2824 /* <disp>(%esp) becomes two byte modrm with no index
2825 register. We've already stored the code for esp
2826 in i.rm.regmem ie. ESCAPE_TO_TWO_BYTE_ADDRESSING.
2827 Any base register besides %esp will not use the
2828 extra modrm byte. */
2829 i.sib.index = NO_INDEX_REGISTER;
2830 #if !SCALE1_WHEN_NO_INDEX
2831 /* Another case where we force the second modrm byte. */
2832 if (i.log2_scale_factor)
2833 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2834 #endif
2835 }
2836 else
2837 {
2838 i.sib.index = i.index_reg->reg_num;
2839 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2840 if ((i.index_reg->reg_flags & RegRex) != 0)
2841 i.rex |= REX_EXTY;
2842 }
2843 i.rm.mode = mode_from_disp_size (i.types[op]);
2844 }
2845
2846 if (fake_zero_displacement)
2847 {
2848 /* Fakes a zero displacement assuming that i.types[op]
2849 holds the correct displacement size. */
2850 expressionS *exp;
2851
2852 assert (i.op[op].disps == 0);
2853 exp = &disp_expressions[i.disp_operands++];
2854 i.op[op].disps = exp;
2855 exp->X_op = O_constant;
2856 exp->X_add_number = 0;
2857 exp->X_add_symbol = (symbolS *) 0;
2858 exp->X_op_symbol = (symbolS *) 0;
2859 }
2860 }
2861
2862 /* Fill in i.rm.reg or i.rm.regmem field with register operand
2863 (if any) based on i.tm.extension_opcode. Again, we must be
2864 careful to make sure that segment/control/debug/test/MMX
2865 registers are coded into the i.rm.reg field. */
2866 if (i.reg_operands)
2867 {
2868 unsigned int op =
2869 ((i.types[0]
2870 & (Reg | RegMMX | RegXMM
2871 | SReg2 | SReg3
2872 | Control | Debug | Test))
2873 ? 0
2874 : ((i.types[1]
2875 & (Reg | RegMMX | RegXMM
2876 | SReg2 | SReg3
2877 | Control | Debug | Test))
2878 ? 1
2879 : 2));
2880 /* If there is an extension opcode to put here, the register
2881 number must be put into the regmem field. */
2882 if (i.tm.extension_opcode != None)
2883 {
2884 i.rm.regmem = i.op[op].regs->reg_num;
2885 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2886 i.rex |= REX_EXTZ;
2887 }
2888 else
2889 {
2890 i.rm.reg = i.op[op].regs->reg_num;
2891 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2892 i.rex |= REX_EXTX;
2893 }
2894
2895 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2 we
2896 must set it to 3 to indicate this is a register operand
2897 in the regmem field. */
2898 if (!i.mem_operands)
2899 i.rm.mode = 3;
2900 }
2901
2902 /* Fill in i.rm.reg field with extension opcode (if any). */
2903 if (i.tm.extension_opcode != None)
2904 i.rm.reg = i.tm.extension_opcode;
2905 }
2906 return default_seg;
2907 }
2908
2909 static void
2910 output_branch ()
2911 {
2912 char *p;
2913 int code16;
2914 int prefix;
2915 relax_substateT subtype;
2916 symbolS *sym;
2917 offsetT off;
2918
2919 code16 = 0;
2920 if (flag_code == CODE_16BIT)
2921 code16 = CODE16;
2922
2923 prefix = 0;
2924 if (i.prefix[DATA_PREFIX] != 0)
2925 {
2926 prefix = 1;
2927 i.prefixes -= 1;
2928 code16 ^= CODE16;
2929 }
2930 /* Pentium4 branch hints. */
2931 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
2932 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
2933 {
2934 prefix++;
2935 i.prefixes--;
2936 }
2937 if (i.prefix[REX_PREFIX] != 0)
2938 {
2939 prefix++;
2940 i.prefixes--;
2941 }
2942
2943 if (i.prefixes != 0 && !intel_syntax)
2944 as_warn (_("skipping prefixes on this instruction"));
2945
2946 /* It's always a symbol; End frag & setup for relax.
2947 Make sure there is enough room in this frag for the largest
2948 instruction we may generate in md_convert_frag. This is 2
2949 bytes for the opcode and room for the prefix and largest
2950 displacement. */
2951 frag_grow (prefix + 2 + 4);
2952 /* Prefix and 1 opcode byte go in fr_fix. */
2953 p = frag_more (prefix + 1);
2954 if (i.prefix[DATA_PREFIX] != 0)
2955 *p++ = DATA_PREFIX_OPCODE;
2956 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
2957 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
2958 *p++ = i.prefix[SEG_PREFIX];
2959 if (i.prefix[REX_PREFIX] != 0)
2960 *p++ = i.prefix[REX_PREFIX];
2961 *p = i.tm.base_opcode;
2962
2963 if ((unsigned char) *p == JUMP_PC_RELATIVE)
2964 subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
2965 else if ((cpu_arch_flags & Cpu386) != 0)
2966 subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
2967 else
2968 subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
2969 subtype |= code16;
2970
2971 sym = i.op[0].disps->X_add_symbol;
2972 off = i.op[0].disps->X_add_number;
2973
2974 if (i.op[0].disps->X_op != O_constant
2975 && i.op[0].disps->X_op != O_symbol)
2976 {
2977 /* Handle complex expressions. */
2978 sym = make_expr_symbol (i.op[0].disps);
2979 off = 0;
2980 }
2981
2982 /* 1 possible extra opcode + 4 byte displacement go in var part.
2983 Pass reloc in fr_var. */
2984 frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
2985 }
2986
2987 static void
2988 output_jump ()
2989 {
2990 char *p;
2991 int size;
2992
2993 if (i.tm.opcode_modifier & JumpByte)
2994 {
2995 /* This is a loop or jecxz type instruction. */
2996 size = 1;
2997 if (i.prefix[ADDR_PREFIX] != 0)
2998 {
2999 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
3000 i.prefixes -= 1;
3001 }
3002 /* Pentium4 branch hints. */
3003 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3004 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3005 {
3006 FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
3007 i.prefixes--;
3008 }
3009 }
3010 else
3011 {
3012 int code16;
3013
3014 code16 = 0;
3015 if (flag_code == CODE_16BIT)
3016 code16 = CODE16;
3017
3018 if (i.prefix[DATA_PREFIX] != 0)
3019 {
3020 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
3021 i.prefixes -= 1;
3022 code16 ^= CODE16;
3023 }
3024
3025 size = 4;
3026 if (code16)
3027 size = 2;
3028 }
3029
3030 if (i.prefix[REX_PREFIX] != 0)
3031 {
3032 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
3033 i.prefixes -= 1;
3034 }
3035
3036 if (i.prefixes != 0 && !intel_syntax)
3037 as_warn (_("skipping prefixes on this instruction"));
3038
3039 p = frag_more (1 + size);
3040 *p++ = i.tm.base_opcode;
3041
3042 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3043 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
3044 }
3045
3046 static void
3047 output_interseg_jump ()
3048 {
3049 char *p;
3050 int size;
3051 int prefix;
3052 int code16;
3053
3054 code16 = 0;
3055 if (flag_code == CODE_16BIT)
3056 code16 = CODE16;
3057
3058 prefix = 0;
3059 if (i.prefix[DATA_PREFIX] != 0)
3060 {
3061 prefix = 1;
3062 i.prefixes -= 1;
3063 code16 ^= CODE16;
3064 }
3065 if (i.prefix[REX_PREFIX] != 0)
3066 {
3067 prefix++;
3068 i.prefixes -= 1;
3069 }
3070
3071 size = 4;
3072 if (code16)
3073 size = 2;
3074
3075 if (i.prefixes != 0 && !intel_syntax)
3076 as_warn (_("skipping prefixes on this instruction"));
3077
3078 /* 1 opcode; 2 segment; offset */
3079 p = frag_more (prefix + 1 + 2 + size);
3080
3081 if (i.prefix[DATA_PREFIX] != 0)
3082 *p++ = DATA_PREFIX_OPCODE;
3083
3084 if (i.prefix[REX_PREFIX] != 0)
3085 *p++ = i.prefix[REX_PREFIX];
3086
3087 *p++ = i.tm.base_opcode;
3088 if (i.op[1].imms->X_op == O_constant)
3089 {
3090 offsetT n = i.op[1].imms->X_add_number;
3091
3092 if (size == 2
3093 && !fits_in_unsigned_word (n)
3094 && !fits_in_signed_word (n))
3095 {
3096 as_bad (_("16-bit jump out of range"));
3097 return;
3098 }
3099 md_number_to_chars (p, n, size);
3100 }
3101 else
3102 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3103 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
3104 if (i.op[0].imms->X_op != O_constant)
3105 as_bad (_("can't handle non absolute segment in `%s'"),
3106 i.tm.name);
3107 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
3108 }
3109
3110
3111 static void
3112 output_insn ()
3113 {
3114 fragS *insn_start_frag;
3115 offsetT insn_start_off;
3116
3117 /* Tie dwarf2 debug info to the address at the start of the insn.
3118 We can't do this after the insn has been output as the current
3119 frag may have been closed off. eg. by frag_var. */
3120 dwarf2_emit_insn (0);
3121
3122 insn_start_frag = frag_now;
3123 insn_start_off = frag_now_fix ();
3124
3125 /* Output jumps. */
3126 if (i.tm.opcode_modifier & Jump)
3127 output_branch ();
3128 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
3129 output_jump ();
3130 else if (i.tm.opcode_modifier & JumpInterSegment)
3131 output_interseg_jump ();
3132 else
3133 {
3134 /* Output normal instructions here. */
3135 char *p;
3136 unsigned char *q;
3137
3138 /* All opcodes on i386 have either 1 or 2 bytes. We may use third
3139 byte for the SSE instructions to specify a prefix they require. */
3140 if (i.tm.base_opcode & 0xff0000)
3141 add_prefix ((i.tm.base_opcode >> 16) & 0xff);
3142
3143 /* The prefix bytes. */
3144 for (q = i.prefix;
3145 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
3146 q++)
3147 {
3148 if (*q)
3149 {
3150 p = frag_more (1);
3151 md_number_to_chars (p, (valueT) *q, 1);
3152 }
3153 }
3154
3155 /* Now the opcode; be careful about word order here! */
3156 if (fits_in_unsigned_byte (i.tm.base_opcode))
3157 {
3158 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
3159 }
3160 else
3161 {
3162 p = frag_more (2);
3163 /* Put out high byte first: can't use md_number_to_chars! */
3164 *p++ = (i.tm.base_opcode >> 8) & 0xff;
3165 *p = i.tm.base_opcode & 0xff;
3166 }
3167
3168 /* Now the modrm byte and sib byte (if present). */
3169 if (i.tm.opcode_modifier & Modrm)
3170 {
3171 p = frag_more (1);
3172 md_number_to_chars (p,
3173 (valueT) (i.rm.regmem << 0
3174 | i.rm.reg << 3
3175 | i.rm.mode << 6),
3176 1);
3177 /* If i.rm.regmem == ESP (4)
3178 && i.rm.mode != (Register mode)
3179 && not 16 bit
3180 ==> need second modrm byte. */
3181 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
3182 && i.rm.mode != 3
3183 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
3184 {
3185 p = frag_more (1);
3186 md_number_to_chars (p,
3187 (valueT) (i.sib.base << 0
3188 | i.sib.index << 3
3189 | i.sib.scale << 6),
3190 1);
3191 }
3192 }
3193
3194 if (i.disp_operands)
3195 output_disp (insn_start_frag, insn_start_off);
3196
3197 if (i.imm_operands)
3198 output_imm (insn_start_frag, insn_start_off);
3199 }
3200
3201 #ifdef DEBUG386
3202 if (flag_debug)
3203 {
3204 pi (line, &i);
3205 }
3206 #endif /* DEBUG386 */
3207 }
3208
3209 static void
3210 output_disp (insn_start_frag, insn_start_off)
3211 fragS *insn_start_frag;
3212 offsetT insn_start_off;
3213 {
3214 char *p;
3215 unsigned int n;
3216
3217 for (n = 0; n < i.operands; n++)
3218 {
3219 if (i.types[n] & Disp)
3220 {
3221 if (i.op[n].disps->X_op == O_constant)
3222 {
3223 int size;
3224 offsetT val;
3225
3226 size = 4;
3227 if (i.types[n] & (Disp8 | Disp16 | Disp64))
3228 {
3229 size = 2;
3230 if (i.types[n] & Disp8)
3231 size = 1;
3232 if (i.types[n] & Disp64)
3233 size = 8;
3234 }
3235 val = offset_in_range (i.op[n].disps->X_add_number,
3236 size);
3237 p = frag_more (size);
3238 md_number_to_chars (p, val, size);
3239 }
3240 else
3241 {
3242 RELOC_ENUM reloc_type;
3243 int size = 4;
3244 int sign = 0;
3245 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
3246
3247 /* The PC relative address is computed relative
3248 to the instruction boundary, so in case immediate
3249 fields follows, we need to adjust the value. */
3250 if (pcrel && i.imm_operands)
3251 {
3252 int imm_size = 4;
3253 unsigned int n1;
3254
3255 for (n1 = 0; n1 < i.operands; n1++)
3256 if (i.types[n1] & Imm)
3257 {
3258 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
3259 {
3260 imm_size = 2;
3261 if (i.types[n1] & (Imm8 | Imm8S))
3262 imm_size = 1;
3263 if (i.types[n1] & Imm64)
3264 imm_size = 8;
3265 }
3266 break;
3267 }
3268 /* We should find the immediate. */
3269 if (n1 == i.operands)
3270 abort ();
3271 i.op[n].disps->X_add_number -= imm_size;
3272 }
3273
3274 if (i.types[n] & Disp32S)
3275 sign = 1;
3276
3277 if (i.types[n] & (Disp16 | Disp64))
3278 {
3279 size = 2;
3280 if (i.types[n] & Disp64)
3281 size = 8;
3282 }
3283
3284 p = frag_more (size);
3285 reloc_type = reloc (size, pcrel, sign, i.reloc[n]);
3286 #ifdef BFD_ASSEMBLER
3287 if (reloc_type == BFD_RELOC_32
3288 && GOT_symbol
3289 && GOT_symbol == i.op[n].disps->X_add_symbol
3290 && (i.op[n].disps->X_op == O_symbol
3291 || (i.op[n].disps->X_op == O_add
3292 && ((symbol_get_value_expression
3293 (i.op[n].disps->X_op_symbol)->X_op)
3294 == O_subtract))))
3295 {
3296 offsetT add;
3297
3298 if (insn_start_frag == frag_now)
3299 add = (p - frag_now->fr_literal) - insn_start_off;
3300 else
3301 {
3302 fragS *fr;
3303
3304 add = insn_start_frag->fr_fix - insn_start_off;
3305 for (fr = insn_start_frag->fr_next;
3306 fr && fr != frag_now; fr = fr->fr_next)
3307 add += fr->fr_fix;
3308 add += p - frag_now->fr_literal;
3309 }
3310
3311 /* We don't support dynamic linking on x86-64 yet. */
3312 if (flag_code == CODE_64BIT)
3313 abort ();
3314 reloc_type = BFD_RELOC_386_GOTPC;
3315 i.op[n].disps->X_add_number += add;
3316 }
3317 #endif
3318 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3319 i.op[n].disps, pcrel, reloc_type);
3320 }
3321 }
3322 }
3323 }
3324
3325 static void
3326 output_imm (insn_start_frag, insn_start_off)
3327 fragS *insn_start_frag;
3328 offsetT insn_start_off;
3329 {
3330 char *p;
3331 unsigned int n;
3332
3333 for (n = 0; n < i.operands; n++)
3334 {
3335 if (i.types[n] & Imm)
3336 {
3337 if (i.op[n].imms->X_op == O_constant)
3338 {
3339 int size;
3340 offsetT val;
3341
3342 size = 4;
3343 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3344 {
3345 size = 2;
3346 if (i.types[n] & (Imm8 | Imm8S))
3347 size = 1;
3348 else if (i.types[n] & Imm64)
3349 size = 8;
3350 }
3351 val = offset_in_range (i.op[n].imms->X_add_number,
3352 size);
3353 p = frag_more (size);
3354 md_number_to_chars (p, val, size);
3355 }
3356 else
3357 {
3358 /* Not absolute_section.
3359 Need a 32-bit fixup (don't support 8bit
3360 non-absolute imms). Try to support other
3361 sizes ... */
3362 RELOC_ENUM reloc_type;
3363 int size = 4;
3364 int sign = 0;
3365
3366 if ((i.types[n] & (Imm32S))
3367 && i.suffix == QWORD_MNEM_SUFFIX)
3368 sign = 1;
3369 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3370 {
3371 size = 2;
3372 if (i.types[n] & (Imm8 | Imm8S))
3373 size = 1;
3374 if (i.types[n] & Imm64)
3375 size = 8;
3376 }
3377
3378 p = frag_more (size);
3379 reloc_type = reloc (size, 0, sign, i.reloc[n]);
3380 #ifdef BFD_ASSEMBLER
3381 /* This is tough to explain. We end up with this one if we
3382 * have operands that look like
3383 * "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal here is to
3384 * obtain the absolute address of the GOT, and it is strongly
3385 * preferable from a performance point of view to avoid using
3386 * a runtime relocation for this. The actual sequence of
3387 * instructions often look something like:
3388 *
3389 * call .L66
3390 * .L66:
3391 * popl %ebx
3392 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
3393 *
3394 * The call and pop essentially return the absolute address
3395 * of the label .L66 and store it in %ebx. The linker itself
3396 * will ultimately change the first operand of the addl so
3397 * that %ebx points to the GOT, but to keep things simple, the
3398 * .o file must have this operand set so that it generates not
3399 * the absolute address of .L66, but the absolute address of
3400 * itself. This allows the linker itself simply treat a GOTPC
3401 * relocation as asking for a pcrel offset to the GOT to be
3402 * added in, and the addend of the relocation is stored in the
3403 * operand field for the instruction itself.
3404 *
3405 * Our job here is to fix the operand so that it would add
3406 * the correct offset so that %ebx would point to itself. The
3407 * thing that is tricky is that .-.L66 will point to the
3408 * beginning of the instruction, so we need to further modify
3409 * the operand so that it will point to itself. There are
3410 * other cases where you have something like:
3411 *
3412 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
3413 *
3414 * and here no correction would be required. Internally in
3415 * the assembler we treat operands of this form as not being
3416 * pcrel since the '.' is explicitly mentioned, and I wonder
3417 * whether it would simplify matters to do it this way. Who
3418 * knows. In earlier versions of the PIC patches, the
3419 * pcrel_adjust field was used to store the correction, but
3420 * since the expression is not pcrel, I felt it would be
3421 * confusing to do it this way. */
3422
3423 if (reloc_type == BFD_RELOC_32
3424 && GOT_symbol
3425 && GOT_symbol == i.op[n].imms->X_add_symbol
3426 && (i.op[n].imms->X_op == O_symbol
3427 || (i.op[n].imms->X_op == O_add
3428 && ((symbol_get_value_expression
3429 (i.op[n].imms->X_op_symbol)->X_op)
3430 == O_subtract))))
3431 {
3432 offsetT add;
3433
3434 if (insn_start_frag == frag_now)
3435 add = (p - frag_now->fr_literal) - insn_start_off;
3436 else
3437 {
3438 fragS *fr;
3439
3440 add = insn_start_frag->fr_fix - insn_start_off;
3441 for (fr = insn_start_frag->fr_next;
3442 fr && fr != frag_now; fr = fr->fr_next)
3443 add += fr->fr_fix;
3444 add += p - frag_now->fr_literal;
3445 }
3446
3447 /* We don't support dynamic linking on x86-64 yet. */
3448 if (flag_code == CODE_64BIT)
3449 abort ();
3450 reloc_type = BFD_RELOC_386_GOTPC;
3451 i.op[n].imms->X_add_number += add;
3452 }
3453 #endif
3454 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3455 i.op[n].imms, 0, reloc_type);
3456 }
3457 }
3458 }
3459 }
3460 \f
3461 #ifndef LEX_AT
3462 static char *lex_got PARAMS ((RELOC_ENUM *, int *));
3463
3464 /* Parse operands of the form
3465 <symbol>@GOTOFF+<nnn>
3466 and similar .plt or .got references.
3467
3468 If we find one, set up the correct relocation in RELOC and copy the
3469 input string, minus the `@GOTOFF' into a malloc'd buffer for
3470 parsing by the calling routine. Return this buffer, and if ADJUST
3471 is non-null set it to the length of the string we removed from the
3472 input line. Otherwise return NULL. */
3473 static char *
3474 lex_got (reloc, adjust)
3475 RELOC_ENUM *reloc;
3476 int *adjust;
3477 {
3478 static const char * const mode_name[NUM_FLAG_CODE] = { "32", "16", "64" };
3479 static const struct {
3480 const char *str;
3481 const RELOC_ENUM rel[NUM_FLAG_CODE];
3482 } gotrel[] = {
3483 { "PLT", { BFD_RELOC_386_PLT32, 0, BFD_RELOC_X86_64_PLT32 } },
3484 { "GOTOFF", { BFD_RELOC_386_GOTOFF, 0, 0 } },
3485 { "GOTPCREL", { 0, 0, BFD_RELOC_X86_64_GOTPCREL } },
3486 { "TLSGD", { BFD_RELOC_386_TLS_GD, 0, 0 } },
3487 { "TLSLDM", { BFD_RELOC_386_TLS_LDM, 0, 0 } },
3488 { "GOTTPOFF", { BFD_RELOC_386_TLS_IE_32, 0, 0 } },
3489 { "TPOFF", { BFD_RELOC_386_TLS_LE_32, 0, 0 } },
3490 { "NTPOFF", { BFD_RELOC_386_TLS_LE, 0, 0 } },
3491 { "DTPOFF", { BFD_RELOC_386_TLS_LDO_32, 0, 0 } },
3492 { "GOTNTPOFF",{ BFD_RELOC_386_TLS_GOTIE, 0, 0 } },
3493 { "INDNTPOFF",{ BFD_RELOC_386_TLS_IE, 0, 0 } },
3494 { "GOT", { BFD_RELOC_386_GOT32, 0, BFD_RELOC_X86_64_GOT32 } }
3495 };
3496 char *cp;
3497 unsigned int j;
3498
3499 for (cp = input_line_pointer; *cp != '@'; cp++)
3500 if (is_end_of_line[(unsigned char) *cp])
3501 return NULL;
3502
3503 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
3504 {
3505 int len;
3506
3507 len = strlen (gotrel[j].str);
3508 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
3509 {
3510 if (gotrel[j].rel[(unsigned int) flag_code] != 0)
3511 {
3512 int first, second;
3513 char *tmpbuf, *past_reloc;
3514
3515 *reloc = gotrel[j].rel[(unsigned int) flag_code];
3516 if (adjust)
3517 *adjust = len;
3518
3519 if (GOT_symbol == NULL)
3520 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3521
3522 /* Replace the relocation token with ' ', so that
3523 errors like foo@GOTOFF1 will be detected. */
3524
3525 /* The length of the first part of our input line. */
3526 first = cp - input_line_pointer;
3527
3528 /* The second part goes from after the reloc token until
3529 (and including) an end_of_line char. Don't use strlen
3530 here as the end_of_line char may not be a NUL. */
3531 past_reloc = cp + 1 + len;
3532 for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
3533 ;
3534 second = cp - past_reloc;
3535
3536 /* Allocate and copy string. The trailing NUL shouldn't
3537 be necessary, but be safe. */
3538 tmpbuf = xmalloc (first + second + 2);
3539 memcpy (tmpbuf, input_line_pointer, first);
3540 tmpbuf[first] = ' ';
3541 memcpy (tmpbuf + first + 1, past_reloc, second);
3542 tmpbuf[first + second + 1] = '\0';
3543 return tmpbuf;
3544 }
3545
3546 as_bad (_("@%s reloc is not supported in %s bit mode"),
3547 gotrel[j].str, mode_name[(unsigned int) flag_code]);
3548 return NULL;
3549 }
3550 }
3551
3552 /* Might be a symbol version string. Don't as_bad here. */
3553 return NULL;
3554 }
3555
3556 /* x86_cons_fix_new is called via the expression parsing code when a
3557 reloc is needed. We use this hook to get the correct .got reloc. */
3558 static RELOC_ENUM got_reloc = NO_RELOC;
3559
3560 void
3561 x86_cons_fix_new (frag, off, len, exp)
3562 fragS *frag;
3563 unsigned int off;
3564 unsigned int len;
3565 expressionS *exp;
3566 {
3567 RELOC_ENUM r = reloc (len, 0, 0, got_reloc);
3568 got_reloc = NO_RELOC;
3569 fix_new_exp (frag, off, len, exp, 0, r);
3570 }
3571
3572 void
3573 x86_cons (exp, size)
3574 expressionS *exp;
3575 int size;
3576 {
3577 if (size == 4)
3578 {
3579 /* Handle @GOTOFF and the like in an expression. */
3580 char *save;
3581 char *gotfree_input_line;
3582 int adjust;
3583
3584 save = input_line_pointer;
3585 gotfree_input_line = lex_got (&got_reloc, &adjust);
3586 if (gotfree_input_line)
3587 input_line_pointer = gotfree_input_line;
3588
3589 expression (exp);
3590
3591 if (gotfree_input_line)
3592 {
3593 /* expression () has merrily parsed up to the end of line,
3594 or a comma - in the wrong buffer. Transfer how far
3595 input_line_pointer has moved to the right buffer. */
3596 input_line_pointer = (save
3597 + (input_line_pointer - gotfree_input_line)
3598 + adjust);
3599 free (gotfree_input_line);
3600 }
3601 }
3602 else
3603 expression (exp);
3604 }
3605 #endif
3606
3607 static int i386_immediate PARAMS ((char *));
3608
3609 static int
3610 i386_immediate (imm_start)
3611 char *imm_start;
3612 {
3613 char *save_input_line_pointer;
3614 #ifndef LEX_AT
3615 char *gotfree_input_line;
3616 #endif
3617 segT exp_seg = 0;
3618 expressionS *exp;
3619
3620 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
3621 {
3622 as_bad (_("only 1 or 2 immediate operands are allowed"));
3623 return 0;
3624 }
3625
3626 exp = &im_expressions[i.imm_operands++];
3627 i.op[this_operand].imms = exp;
3628
3629 if (is_space_char (*imm_start))
3630 ++imm_start;
3631
3632 save_input_line_pointer = input_line_pointer;
3633 input_line_pointer = imm_start;
3634
3635 #ifndef LEX_AT
3636 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3637 if (gotfree_input_line)
3638 input_line_pointer = gotfree_input_line;
3639 #endif
3640
3641 exp_seg = expression (exp);
3642
3643 SKIP_WHITESPACE ();
3644 if (*input_line_pointer)
3645 as_bad (_("junk `%s' after expression"), input_line_pointer);
3646
3647 input_line_pointer = save_input_line_pointer;
3648 #ifndef LEX_AT
3649 if (gotfree_input_line)
3650 free (gotfree_input_line);
3651 #endif
3652
3653 if (exp->X_op == O_absent || exp->X_op == O_big)
3654 {
3655 /* Missing or bad expr becomes absolute 0. */
3656 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
3657 imm_start);
3658 exp->X_op = O_constant;
3659 exp->X_add_number = 0;
3660 exp->X_add_symbol = (symbolS *) 0;
3661 exp->X_op_symbol = (symbolS *) 0;
3662 }
3663 else if (exp->X_op == O_constant)
3664 {
3665 /* Size it properly later. */
3666 i.types[this_operand] |= Imm64;
3667 /* If BFD64, sign extend val. */
3668 if (!use_rela_relocations)
3669 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
3670 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
3671 }
3672 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3673 else if (1
3674 #ifdef BFD_ASSEMBLER
3675 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3676 #endif
3677 && exp_seg != absolute_section
3678 && exp_seg != text_section
3679 && exp_seg != data_section
3680 && exp_seg != bss_section
3681 && exp_seg != undefined_section
3682 #ifdef BFD_ASSEMBLER
3683 && !bfd_is_com_section (exp_seg)
3684 #endif
3685 )
3686 {
3687 #ifdef BFD_ASSEMBLER
3688 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3689 #else
3690 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3691 #endif
3692 return 0;
3693 }
3694 #endif
3695 else
3696 {
3697 /* This is an address. The size of the address will be
3698 determined later, depending on destination register,
3699 suffix, or the default for the section. */
3700 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
3701 }
3702
3703 return 1;
3704 }
3705
3706 static char *i386_scale PARAMS ((char *));
3707
3708 static char *
3709 i386_scale (scale)
3710 char *scale;
3711 {
3712 offsetT val;
3713 char *save = input_line_pointer;
3714
3715 input_line_pointer = scale;
3716 val = get_absolute_expression ();
3717
3718 switch (val)
3719 {
3720 case 0:
3721 case 1:
3722 i.log2_scale_factor = 0;
3723 break;
3724 case 2:
3725 i.log2_scale_factor = 1;
3726 break;
3727 case 4:
3728 i.log2_scale_factor = 2;
3729 break;
3730 case 8:
3731 i.log2_scale_factor = 3;
3732 break;
3733 default:
3734 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
3735 scale);
3736 input_line_pointer = save;
3737 return NULL;
3738 }
3739 if (i.log2_scale_factor != 0 && i.index_reg == 0)
3740 {
3741 as_warn (_("scale factor of %d without an index register"),
3742 1 << i.log2_scale_factor);
3743 #if SCALE1_WHEN_NO_INDEX
3744 i.log2_scale_factor = 0;
3745 #endif
3746 }
3747 scale = input_line_pointer;
3748 input_line_pointer = save;
3749 return scale;
3750 }
3751
3752 static int i386_displacement PARAMS ((char *, char *));
3753
3754 static int
3755 i386_displacement (disp_start, disp_end)
3756 char *disp_start;
3757 char *disp_end;
3758 {
3759 expressionS *exp;
3760 segT exp_seg = 0;
3761 char *save_input_line_pointer;
3762 #ifndef LEX_AT
3763 char *gotfree_input_line;
3764 #endif
3765 int bigdisp = Disp32;
3766
3767 if (flag_code == CODE_64BIT)
3768 {
3769 if (i.prefix[ADDR_PREFIX] == 0)
3770 bigdisp = Disp64;
3771 }
3772 else if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3773 bigdisp = Disp16;
3774 i.types[this_operand] |= bigdisp;
3775
3776 exp = &disp_expressions[i.disp_operands];
3777 i.op[this_operand].disps = exp;
3778 i.disp_operands++;
3779 save_input_line_pointer = input_line_pointer;
3780 input_line_pointer = disp_start;
3781 END_STRING_AND_SAVE (disp_end);
3782
3783 #ifndef GCC_ASM_O_HACK
3784 #define GCC_ASM_O_HACK 0
3785 #endif
3786 #if GCC_ASM_O_HACK
3787 END_STRING_AND_SAVE (disp_end + 1);
3788 if ((i.types[this_operand] & BaseIndex) != 0
3789 && displacement_string_end[-1] == '+')
3790 {
3791 /* This hack is to avoid a warning when using the "o"
3792 constraint within gcc asm statements.
3793 For instance:
3794
3795 #define _set_tssldt_desc(n,addr,limit,type) \
3796 __asm__ __volatile__ ( \
3797 "movw %w2,%0\n\t" \
3798 "movw %w1,2+%0\n\t" \
3799 "rorl $16,%1\n\t" \
3800 "movb %b1,4+%0\n\t" \
3801 "movb %4,5+%0\n\t" \
3802 "movb $0,6+%0\n\t" \
3803 "movb %h1,7+%0\n\t" \
3804 "rorl $16,%1" \
3805 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
3806
3807 This works great except that the output assembler ends
3808 up looking a bit weird if it turns out that there is
3809 no offset. You end up producing code that looks like:
3810
3811 #APP
3812 movw $235,(%eax)
3813 movw %dx,2+(%eax)
3814 rorl $16,%edx
3815 movb %dl,4+(%eax)
3816 movb $137,5+(%eax)
3817 movb $0,6+(%eax)
3818 movb %dh,7+(%eax)
3819 rorl $16,%edx
3820 #NO_APP
3821
3822 So here we provide the missing zero. */
3823
3824 *displacement_string_end = '0';
3825 }
3826 #endif
3827 #ifndef LEX_AT
3828 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3829 if (gotfree_input_line)
3830 input_line_pointer = gotfree_input_line;
3831 #endif
3832
3833 exp_seg = expression (exp);
3834
3835 SKIP_WHITESPACE ();
3836 if (*input_line_pointer)
3837 as_bad (_("junk `%s' after expression"), input_line_pointer);
3838 #if GCC_ASM_O_HACK
3839 RESTORE_END_STRING (disp_end + 1);
3840 #endif
3841 RESTORE_END_STRING (disp_end);
3842 input_line_pointer = save_input_line_pointer;
3843 #ifndef LEX_AT
3844 if (gotfree_input_line)
3845 free (gotfree_input_line);
3846 #endif
3847
3848 #ifdef BFD_ASSEMBLER
3849 /* We do this to make sure that the section symbol is in
3850 the symbol table. We will ultimately change the relocation
3851 to be relative to the beginning of the section. */
3852 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
3853 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3854 {
3855 if (exp->X_op != O_symbol)
3856 {
3857 as_bad (_("bad expression used with @%s"),
3858 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
3859 ? "GOTPCREL"
3860 : "GOTOFF"));
3861 return 0;
3862 }
3863
3864 if (S_IS_LOCAL (exp->X_add_symbol)
3865 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
3866 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
3867 exp->X_op = O_subtract;
3868 exp->X_op_symbol = GOT_symbol;
3869 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3870 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
3871 else
3872 i.reloc[this_operand] = BFD_RELOC_32;
3873 }
3874 #endif
3875
3876 if (exp->X_op == O_absent || exp->X_op == O_big)
3877 {
3878 /* Missing or bad expr becomes absolute 0. */
3879 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
3880 disp_start);
3881 exp->X_op = O_constant;
3882 exp->X_add_number = 0;
3883 exp->X_add_symbol = (symbolS *) 0;
3884 exp->X_op_symbol = (symbolS *) 0;
3885 }
3886
3887 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3888 if (exp->X_op != O_constant
3889 #ifdef BFD_ASSEMBLER
3890 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3891 #endif
3892 && exp_seg != absolute_section
3893 && exp_seg != text_section
3894 && exp_seg != data_section
3895 && exp_seg != bss_section
3896 && exp_seg != undefined_section
3897 #ifdef BFD_ASSEMBLER
3898 && !bfd_is_com_section (exp_seg)
3899 #endif
3900 )
3901 {
3902 #ifdef BFD_ASSEMBLER
3903 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3904 #else
3905 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3906 #endif
3907 return 0;
3908 }
3909 #endif
3910 else if (flag_code == CODE_64BIT)
3911 i.types[this_operand] |= Disp32S | Disp32;
3912 return 1;
3913 }
3914
3915 static int i386_index_check PARAMS ((const char *));
3916
3917 /* Make sure the memory operand we've been dealt is valid.
3918 Return 1 on success, 0 on a failure. */
3919
3920 static int
3921 i386_index_check (operand_string)
3922 const char *operand_string;
3923 {
3924 int ok;
3925 #if INFER_ADDR_PREFIX
3926 int fudged = 0;
3927
3928 tryprefix:
3929 #endif
3930 ok = 1;
3931 if (flag_code == CODE_64BIT)
3932 {
3933 if (i.prefix[ADDR_PREFIX] == 0)
3934 {
3935 /* 64bit checks. */
3936 if ((i.base_reg
3937 && ((i.base_reg->reg_type & Reg64) == 0)
3938 && (i.base_reg->reg_type != BaseIndex
3939 || i.index_reg))
3940 || (i.index_reg
3941 && ((i.index_reg->reg_type & (Reg64 | BaseIndex))
3942 != (Reg64 | BaseIndex))))
3943 ok = 0;
3944 }
3945 else
3946 {
3947 /* 32bit checks. */
3948 if ((i.base_reg
3949 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
3950 || (i.index_reg
3951 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
3952 != (Reg32 | BaseIndex))))
3953 ok = 0;
3954 }
3955 }
3956 else
3957 {
3958 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3959 {
3960 /* 16bit checks. */
3961 if ((i.base_reg
3962 && ((i.base_reg->reg_type & (Reg16 | BaseIndex | RegRex))
3963 != (Reg16 | BaseIndex)))
3964 || (i.index_reg
3965 && (((i.index_reg->reg_type & (Reg16 | BaseIndex))
3966 != (Reg16 | BaseIndex))
3967 || !(i.base_reg
3968 && i.base_reg->reg_num < 6
3969 && i.index_reg->reg_num >= 6
3970 && i.log2_scale_factor == 0))))
3971 ok = 0;
3972 }
3973 else
3974 {
3975 /* 32bit checks. */
3976 if ((i.base_reg
3977 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
3978 || (i.index_reg
3979 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
3980 != (Reg32 | BaseIndex))))
3981 ok = 0;
3982 }
3983 }
3984 if (!ok)
3985 {
3986 #if INFER_ADDR_PREFIX
3987 if (flag_code != CODE_64BIT
3988 && i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0')
3989 {
3990 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
3991 i.prefixes += 1;
3992 /* Change the size of any displacement too. At most one of
3993 Disp16 or Disp32 is set.
3994 FIXME. There doesn't seem to be any real need for separate
3995 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
3996 Removing them would probably clean up the code quite a lot. */
3997 if (i.types[this_operand] & (Disp16 | Disp32))
3998 i.types[this_operand] ^= (Disp16 | Disp32);
3999 fudged = 1;
4000 goto tryprefix;
4001 }
4002 if (fudged)
4003 as_bad (_("`%s' is not a valid base/index expression"),
4004 operand_string);
4005 else
4006 #endif
4007 as_bad (_("`%s' is not a valid %s bit base/index expression"),
4008 operand_string,
4009 flag_code_names[flag_code]);
4010 return 0;
4011 }
4012 return 1;
4013 }
4014
4015 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
4016 on error. */
4017
4018 static int
4019 i386_operand (operand_string)
4020 char *operand_string;
4021 {
4022 const reg_entry *r;
4023 char *end_op;
4024 char *op_string = operand_string;
4025
4026 if (is_space_char (*op_string))
4027 ++op_string;
4028
4029 /* We check for an absolute prefix (differentiating,
4030 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
4031 if (*op_string == ABSOLUTE_PREFIX)
4032 {
4033 ++op_string;
4034 if (is_space_char (*op_string))
4035 ++op_string;
4036 i.types[this_operand] |= JumpAbsolute;
4037 }
4038
4039 /* Check if operand is a register. */
4040 if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
4041 && (r = parse_register (op_string, &end_op)) != NULL)
4042 {
4043 /* Check for a segment override by searching for ':' after a
4044 segment register. */
4045 op_string = end_op;
4046 if (is_space_char (*op_string))
4047 ++op_string;
4048 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
4049 {
4050 switch (r->reg_num)
4051 {
4052 case 0:
4053 i.seg[i.mem_operands] = &es;
4054 break;
4055 case 1:
4056 i.seg[i.mem_operands] = &cs;
4057 break;
4058 case 2:
4059 i.seg[i.mem_operands] = &ss;
4060 break;
4061 case 3:
4062 i.seg[i.mem_operands] = &ds;
4063 break;
4064 case 4:
4065 i.seg[i.mem_operands] = &fs;
4066 break;
4067 case 5:
4068 i.seg[i.mem_operands] = &gs;
4069 break;
4070 }
4071
4072 /* Skip the ':' and whitespace. */
4073 ++op_string;
4074 if (is_space_char (*op_string))
4075 ++op_string;
4076
4077 if (!is_digit_char (*op_string)
4078 && !is_identifier_char (*op_string)
4079 && *op_string != '('
4080 && *op_string != ABSOLUTE_PREFIX)
4081 {
4082 as_bad (_("bad memory operand `%s'"), op_string);
4083 return 0;
4084 }
4085 /* Handle case of %es:*foo. */
4086 if (*op_string == ABSOLUTE_PREFIX)
4087 {
4088 ++op_string;
4089 if (is_space_char (*op_string))
4090 ++op_string;
4091 i.types[this_operand] |= JumpAbsolute;
4092 }
4093 goto do_memory_reference;
4094 }
4095 if (*op_string)
4096 {
4097 as_bad (_("junk `%s' after register"), op_string);
4098 return 0;
4099 }
4100 i.types[this_operand] |= r->reg_type & ~BaseIndex;
4101 i.op[this_operand].regs = r;
4102 i.reg_operands++;
4103 }
4104 else if (*op_string == REGISTER_PREFIX)
4105 {
4106 as_bad (_("bad register name `%s'"), op_string);
4107 return 0;
4108 }
4109 else if (*op_string == IMMEDIATE_PREFIX)
4110 {
4111 ++op_string;
4112 if (i.types[this_operand] & JumpAbsolute)
4113 {
4114 as_bad (_("immediate operand illegal with absolute jump"));
4115 return 0;
4116 }
4117 if (!i386_immediate (op_string))
4118 return 0;
4119 }
4120 else if (is_digit_char (*op_string)
4121 || is_identifier_char (*op_string)
4122 || *op_string == '(')
4123 {
4124 /* This is a memory reference of some sort. */
4125 char *base_string;
4126
4127 /* Start and end of displacement string expression (if found). */
4128 char *displacement_string_start;
4129 char *displacement_string_end;
4130
4131 do_memory_reference:
4132 if ((i.mem_operands == 1
4133 && (current_templates->start->opcode_modifier & IsString) == 0)
4134 || i.mem_operands == 2)
4135 {
4136 as_bad (_("too many memory references for `%s'"),
4137 current_templates->start->name);
4138 return 0;
4139 }
4140
4141 /* Check for base index form. We detect the base index form by
4142 looking for an ')' at the end of the operand, searching
4143 for the '(' matching it, and finding a REGISTER_PREFIX or ','
4144 after the '('. */
4145 base_string = op_string + strlen (op_string);
4146
4147 --base_string;
4148 if (is_space_char (*base_string))
4149 --base_string;
4150
4151 /* If we only have a displacement, set-up for it to be parsed later. */
4152 displacement_string_start = op_string;
4153 displacement_string_end = base_string + 1;
4154
4155 if (*base_string == ')')
4156 {
4157 char *temp_string;
4158 unsigned int parens_balanced = 1;
4159 /* We've already checked that the number of left & right ()'s are
4160 equal, so this loop will not be infinite. */
4161 do
4162 {
4163 base_string--;
4164 if (*base_string == ')')
4165 parens_balanced++;
4166 if (*base_string == '(')
4167 parens_balanced--;
4168 }
4169 while (parens_balanced);
4170
4171 temp_string = base_string;
4172
4173 /* Skip past '(' and whitespace. */
4174 ++base_string;
4175 if (is_space_char (*base_string))
4176 ++base_string;
4177
4178 if (*base_string == ','
4179 || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
4180 && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
4181 {
4182 displacement_string_end = temp_string;
4183
4184 i.types[this_operand] |= BaseIndex;
4185
4186 if (i.base_reg)
4187 {
4188 base_string = end_op;
4189 if (is_space_char (*base_string))
4190 ++base_string;
4191 }
4192
4193 /* There may be an index reg or scale factor here. */
4194 if (*base_string == ',')
4195 {
4196 ++base_string;
4197 if (is_space_char (*base_string))
4198 ++base_string;
4199
4200 if ((*base_string == REGISTER_PREFIX || allow_naked_reg)
4201 && (i.index_reg = parse_register (base_string, &end_op)) != NULL)
4202 {
4203 base_string = end_op;
4204 if (is_space_char (*base_string))
4205 ++base_string;
4206 if (*base_string == ',')
4207 {
4208 ++base_string;
4209 if (is_space_char (*base_string))
4210 ++base_string;
4211 }
4212 else if (*base_string != ')')
4213 {
4214 as_bad (_("expecting `,' or `)' after index register in `%s'"),
4215 operand_string);
4216 return 0;
4217 }
4218 }
4219 else if (*base_string == REGISTER_PREFIX)
4220 {
4221 as_bad (_("bad register name `%s'"), base_string);
4222 return 0;
4223 }
4224
4225 /* Check for scale factor. */
4226 if (*base_string != ')')
4227 {
4228 char *end_scale = i386_scale (base_string);
4229
4230 if (!end_scale)
4231 return 0;
4232
4233 base_string = end_scale;
4234 if (is_space_char (*base_string))
4235 ++base_string;
4236 if (*base_string != ')')
4237 {
4238 as_bad (_("expecting `)' after scale factor in `%s'"),
4239 operand_string);
4240 return 0;
4241 }
4242 }
4243 else if (!i.index_reg)
4244 {
4245 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
4246 *base_string);
4247 return 0;
4248 }
4249 }
4250 else if (*base_string != ')')
4251 {
4252 as_bad (_("expecting `,' or `)' after base register in `%s'"),
4253 operand_string);
4254 return 0;
4255 }
4256 }
4257 else if (*base_string == REGISTER_PREFIX)
4258 {
4259 as_bad (_("bad register name `%s'"), base_string);
4260 return 0;
4261 }
4262 }
4263
4264 /* If there's an expression beginning the operand, parse it,
4265 assuming displacement_string_start and
4266 displacement_string_end are meaningful. */
4267 if (displacement_string_start != displacement_string_end)
4268 {
4269 if (!i386_displacement (displacement_string_start,
4270 displacement_string_end))
4271 return 0;
4272 }
4273
4274 /* Special case for (%dx) while doing input/output op. */
4275 if (i.base_reg
4276 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
4277 && i.index_reg == 0
4278 && i.log2_scale_factor == 0
4279 && i.seg[i.mem_operands] == 0
4280 && (i.types[this_operand] & Disp) == 0)
4281 {
4282 i.types[this_operand] = InOutPortReg;
4283 return 1;
4284 }
4285
4286 if (i386_index_check (operand_string) == 0)
4287 return 0;
4288 i.mem_operands++;
4289 }
4290 else
4291 {
4292 /* It's not a memory operand; argh! */
4293 as_bad (_("invalid char %s beginning operand %d `%s'"),
4294 output_invalid (*op_string),
4295 this_operand + 1,
4296 op_string);
4297 return 0;
4298 }
4299 return 1; /* Normal return. */
4300 }
4301 \f
4302 /* md_estimate_size_before_relax()
4303
4304 Called just before relax() for rs_machine_dependent frags. The x86
4305 assembler uses these frags to handle variable size jump
4306 instructions.
4307
4308 Any symbol that is now undefined will not become defined.
4309 Return the correct fr_subtype in the frag.
4310 Return the initial "guess for variable size of frag" to caller.
4311 The guess is actually the growth beyond the fixed part. Whatever
4312 we do to grow the fixed or variable part contributes to our
4313 returned value. */
4314
4315 int
4316 md_estimate_size_before_relax (fragP, segment)
4317 fragS *fragP;
4318 segT segment;
4319 {
4320 /* We've already got fragP->fr_subtype right; all we have to do is
4321 check for un-relaxable symbols. On an ELF system, we can't relax
4322 an externally visible symbol, because it may be overridden by a
4323 shared library. */
4324 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
4325 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4326 || (OUTPUT_FLAVOR == bfd_target_elf_flavour
4327 && (S_IS_EXTERNAL (fragP->fr_symbol)
4328 || S_IS_WEAK (fragP->fr_symbol)))
4329 #endif
4330 )
4331 {
4332 /* Symbol is undefined in this segment, or we need to keep a
4333 reloc so that weak symbols can be overridden. */
4334 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
4335 RELOC_ENUM reloc_type;
4336 unsigned char *opcode;
4337 int old_fr_fix;
4338
4339 if (fragP->fr_var != NO_RELOC)
4340 reloc_type = fragP->fr_var;
4341 else if (size == 2)
4342 reloc_type = BFD_RELOC_16_PCREL;
4343 else
4344 reloc_type = BFD_RELOC_32_PCREL;
4345
4346 old_fr_fix = fragP->fr_fix;
4347 opcode = (unsigned char *) fragP->fr_opcode;
4348
4349 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
4350 {
4351 case UNCOND_JUMP:
4352 /* Make jmp (0xeb) a (d)word displacement jump. */
4353 opcode[0] = 0xe9;
4354 fragP->fr_fix += size;
4355 fix_new (fragP, old_fr_fix, size,
4356 fragP->fr_symbol,
4357 fragP->fr_offset, 1,
4358 reloc_type);
4359 break;
4360
4361 case COND_JUMP86:
4362 if (size == 2
4363 && (!no_cond_jump_promotion || fragP->fr_var != NO_RELOC))
4364 {
4365 /* Negate the condition, and branch past an
4366 unconditional jump. */
4367 opcode[0] ^= 1;
4368 opcode[1] = 3;
4369 /* Insert an unconditional jump. */
4370 opcode[2] = 0xe9;
4371 /* We added two extra opcode bytes, and have a two byte
4372 offset. */
4373 fragP->fr_fix += 2 + 2;
4374 fix_new (fragP, old_fr_fix + 2, 2,
4375 fragP->fr_symbol,
4376 fragP->fr_offset, 1,
4377 reloc_type);
4378 break;
4379 }
4380 /* Fall through. */
4381
4382 case COND_JUMP:
4383 if (no_cond_jump_promotion && fragP->fr_var == NO_RELOC)
4384 {
4385 fragP->fr_fix += 1;
4386 fix_new (fragP, old_fr_fix, 1,
4387 fragP->fr_symbol,
4388 fragP->fr_offset, 1,
4389 BFD_RELOC_8_PCREL);
4390 break;
4391 }
4392
4393 /* This changes the byte-displacement jump 0x7N
4394 to the (d)word-displacement jump 0x0f,0x8N. */
4395 opcode[1] = opcode[0] + 0x10;
4396 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4397 /* We've added an opcode byte. */
4398 fragP->fr_fix += 1 + size;
4399 fix_new (fragP, old_fr_fix + 1, size,
4400 fragP->fr_symbol,
4401 fragP->fr_offset, 1,
4402 reloc_type);
4403 break;
4404
4405 default:
4406 BAD_CASE (fragP->fr_subtype);
4407 break;
4408 }
4409 frag_wane (fragP);
4410 return fragP->fr_fix - old_fr_fix;
4411 }
4412
4413 /* Guess size depending on current relax state. Initially the relax
4414 state will correspond to a short jump and we return 1, because
4415 the variable part of the frag (the branch offset) is one byte
4416 long. However, we can relax a section more than once and in that
4417 case we must either set fr_subtype back to the unrelaxed state,
4418 or return the value for the appropriate branch. */
4419 return md_relax_table[fragP->fr_subtype].rlx_length;
4420 }
4421
4422 /* Called after relax() is finished.
4423
4424 In: Address of frag.
4425 fr_type == rs_machine_dependent.
4426 fr_subtype is what the address relaxed to.
4427
4428 Out: Any fixSs and constants are set up.
4429 Caller will turn frag into a ".space 0". */
4430
4431 #ifndef BFD_ASSEMBLER
4432 void
4433 md_convert_frag (headers, sec, fragP)
4434 object_headers *headers ATTRIBUTE_UNUSED;
4435 segT sec ATTRIBUTE_UNUSED;
4436 fragS *fragP;
4437 #else
4438 void
4439 md_convert_frag (abfd, sec, fragP)
4440 bfd *abfd ATTRIBUTE_UNUSED;
4441 segT sec ATTRIBUTE_UNUSED;
4442 fragS *fragP;
4443 #endif
4444 {
4445 unsigned char *opcode;
4446 unsigned char *where_to_put_displacement = NULL;
4447 offsetT target_address;
4448 offsetT opcode_address;
4449 unsigned int extension = 0;
4450 offsetT displacement_from_opcode_start;
4451
4452 opcode = (unsigned char *) fragP->fr_opcode;
4453
4454 /* Address we want to reach in file space. */
4455 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
4456
4457 /* Address opcode resides at in file space. */
4458 opcode_address = fragP->fr_address + fragP->fr_fix;
4459
4460 /* Displacement from opcode start to fill into instruction. */
4461 displacement_from_opcode_start = target_address - opcode_address;
4462
4463 if ((fragP->fr_subtype & BIG) == 0)
4464 {
4465 /* Don't have to change opcode. */
4466 extension = 1; /* 1 opcode + 1 displacement */
4467 where_to_put_displacement = &opcode[1];
4468 }
4469 else
4470 {
4471 if (no_cond_jump_promotion
4472 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
4473 as_warn_where (fragP->fr_file, fragP->fr_line, _("long jump required"));
4474
4475 switch (fragP->fr_subtype)
4476 {
4477 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
4478 extension = 4; /* 1 opcode + 4 displacement */
4479 opcode[0] = 0xe9;
4480 where_to_put_displacement = &opcode[1];
4481 break;
4482
4483 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
4484 extension = 2; /* 1 opcode + 2 displacement */
4485 opcode[0] = 0xe9;
4486 where_to_put_displacement = &opcode[1];
4487 break;
4488
4489 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
4490 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
4491 extension = 5; /* 2 opcode + 4 displacement */
4492 opcode[1] = opcode[0] + 0x10;
4493 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4494 where_to_put_displacement = &opcode[2];
4495 break;
4496
4497 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
4498 extension = 3; /* 2 opcode + 2 displacement */
4499 opcode[1] = opcode[0] + 0x10;
4500 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4501 where_to_put_displacement = &opcode[2];
4502 break;
4503
4504 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
4505 extension = 4;
4506 opcode[0] ^= 1;
4507 opcode[1] = 3;
4508 opcode[2] = 0xe9;
4509 where_to_put_displacement = &opcode[3];
4510 break;
4511
4512 default:
4513 BAD_CASE (fragP->fr_subtype);
4514 break;
4515 }
4516 }
4517
4518 /* Now put displacement after opcode. */
4519 md_number_to_chars ((char *) where_to_put_displacement,
4520 (valueT) (displacement_from_opcode_start - extension),
4521 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
4522 fragP->fr_fix += extension;
4523 }
4524 \f
4525 /* Size of byte displacement jmp. */
4526 int md_short_jump_size = 2;
4527
4528 /* Size of dword displacement jmp. */
4529 int md_long_jump_size = 5;
4530
4531 /* Size of relocation record. */
4532 const int md_reloc_size = 8;
4533
4534 void
4535 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
4536 char *ptr;
4537 addressT from_addr, to_addr;
4538 fragS *frag ATTRIBUTE_UNUSED;
4539 symbolS *to_symbol ATTRIBUTE_UNUSED;
4540 {
4541 offsetT offset;
4542
4543 offset = to_addr - (from_addr + 2);
4544 /* Opcode for byte-disp jump. */
4545 md_number_to_chars (ptr, (valueT) 0xeb, 1);
4546 md_number_to_chars (ptr + 1, (valueT) offset, 1);
4547 }
4548
4549 void
4550 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
4551 char *ptr;
4552 addressT from_addr, to_addr;
4553 fragS *frag ATTRIBUTE_UNUSED;
4554 symbolS *to_symbol ATTRIBUTE_UNUSED;
4555 {
4556 offsetT offset;
4557
4558 offset = to_addr - (from_addr + 5);
4559 md_number_to_chars (ptr, (valueT) 0xe9, 1);
4560 md_number_to_chars (ptr + 1, (valueT) offset, 4);
4561 }
4562 \f
4563 /* Apply a fixup (fixS) to segment data, once it has been determined
4564 by our caller that we have all the info we need to fix it up.
4565
4566 On the 386, immediates, displacements, and data pointers are all in
4567 the same (little-endian) format, so we don't need to care about which
4568 we are handling. */
4569
4570 void
4571 md_apply_fix3 (fixP, valP, seg)
4572 /* The fix we're to put in. */
4573 fixS *fixP;
4574 /* Pointer to the value of the bits. */
4575 valueT *valP;
4576 /* Segment fix is from. */
4577 segT seg ATTRIBUTE_UNUSED;
4578 {
4579 char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
4580 valueT value = *valP;
4581
4582 #if defined (BFD_ASSEMBLER) && !defined (TE_Mach)
4583 if (fixP->fx_pcrel)
4584 {
4585 switch (fixP->fx_r_type)
4586 {
4587 default:
4588 break;
4589
4590 case BFD_RELOC_32:
4591 fixP->fx_r_type = BFD_RELOC_32_PCREL;
4592 break;
4593 case BFD_RELOC_16:
4594 fixP->fx_r_type = BFD_RELOC_16_PCREL;
4595 break;
4596 case BFD_RELOC_8:
4597 fixP->fx_r_type = BFD_RELOC_8_PCREL;
4598 break;
4599 }
4600 }
4601
4602 if (fixP->fx_addsy != NULL
4603 && (fixP->fx_r_type == BFD_RELOC_32_PCREL
4604 || fixP->fx_r_type == BFD_RELOC_16_PCREL
4605 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
4606 && !use_rela_relocations)
4607 {
4608 /* This is a hack. There should be a better way to handle this.
4609 This covers for the fact that bfd_install_relocation will
4610 subtract the current location (for partial_inplace, PC relative
4611 relocations); see more below. */
4612 #ifndef OBJ_AOUT
4613 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4614 #ifdef TE_PE
4615 || OUTPUT_FLAVOR == bfd_target_coff_flavour
4616 #endif
4617 )
4618 value += fixP->fx_where + fixP->fx_frag->fr_address;
4619 #endif
4620 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4621 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
4622 {
4623 segT sym_seg = S_GET_SEGMENT (fixP->fx_addsy);
4624
4625 if ((sym_seg == seg
4626 || (symbol_section_p (fixP->fx_addsy)
4627 && sym_seg != absolute_section))
4628 && !S_FORCE_RELOC (fixP->fx_addsy))
4629 {
4630 /* Yes, we add the values in twice. This is because
4631 bfd_install_relocation subtracts them out again. I think
4632 bfd_install_relocation is broken, but I don't dare change
4633 it. FIXME. */
4634 value += fixP->fx_where + fixP->fx_frag->fr_address;
4635 }
4636 }
4637 #endif
4638 #if defined (OBJ_COFF) && defined (TE_PE)
4639 /* For some reason, the PE format does not store a section
4640 address offset for a PC relative symbol. */
4641 if (S_GET_SEGMENT (fixP->fx_addsy) != seg)
4642 value += md_pcrel_from (fixP);
4643 #endif
4644 }
4645
4646 /* Fix a few things - the dynamic linker expects certain values here,
4647 and we must not dissappoint it. */
4648 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4649 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4650 && fixP->fx_addsy)
4651 switch (fixP->fx_r_type)
4652 {
4653 case BFD_RELOC_386_PLT32:
4654 case BFD_RELOC_X86_64_PLT32:
4655 /* Make the jump instruction point to the address of the operand. At
4656 runtime we merely add the offset to the actual PLT entry. */
4657 value = -4;
4658 break;
4659
4660 case BFD_RELOC_386_GOT32:
4661 case BFD_RELOC_386_TLS_GD:
4662 case BFD_RELOC_386_TLS_LDM:
4663 case BFD_RELOC_386_TLS_IE_32:
4664 case BFD_RELOC_386_TLS_IE:
4665 case BFD_RELOC_386_TLS_GOTIE:
4666 case BFD_RELOC_X86_64_GOT32:
4667 value = 0; /* Fully resolved at runtime. No addend. */
4668 break;
4669
4670 case BFD_RELOC_VTABLE_INHERIT:
4671 case BFD_RELOC_VTABLE_ENTRY:
4672 fixP->fx_done = 0;
4673 return;
4674
4675 default:
4676 break;
4677 }
4678 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
4679 *valP = value;
4680 #endif /* defined (BFD_ASSEMBLER) && !defined (TE_Mach) */
4681
4682 /* Are we finished with this relocation now? */
4683 if (fixP->fx_addsy == NULL)
4684 fixP->fx_done = 1;
4685 #ifdef BFD_ASSEMBLER
4686 else if (use_rela_relocations)
4687 {
4688 fixP->fx_no_overflow = 1;
4689 /* Remember value for tc_gen_reloc. */
4690 fixP->fx_addnumber = value;
4691 value = 0;
4692 }
4693 #endif
4694 md_number_to_chars (p, value, fixP->fx_size);
4695 }
4696 \f
4697 #define MAX_LITTLENUMS 6
4698
4699 /* Turn the string pointed to by litP into a floating point constant
4700 of type TYPE, and emit the appropriate bytes. The number of
4701 LITTLENUMS emitted is stored in *SIZEP. An error message is
4702 returned, or NULL on OK. */
4703
4704 char *
4705 md_atof (type, litP, sizeP)
4706 int type;
4707 char *litP;
4708 int *sizeP;
4709 {
4710 int prec;
4711 LITTLENUM_TYPE words[MAX_LITTLENUMS];
4712 LITTLENUM_TYPE *wordP;
4713 char *t;
4714
4715 switch (type)
4716 {
4717 case 'f':
4718 case 'F':
4719 prec = 2;
4720 break;
4721
4722 case 'd':
4723 case 'D':
4724 prec = 4;
4725 break;
4726
4727 case 'x':
4728 case 'X':
4729 prec = 5;
4730 break;
4731
4732 default:
4733 *sizeP = 0;
4734 return _("Bad call to md_atof ()");
4735 }
4736 t = atof_ieee (input_line_pointer, type, words);
4737 if (t)
4738 input_line_pointer = t;
4739
4740 *sizeP = prec * sizeof (LITTLENUM_TYPE);
4741 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
4742 the bigendian 386. */
4743 for (wordP = words + prec - 1; prec--;)
4744 {
4745 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
4746 litP += sizeof (LITTLENUM_TYPE);
4747 }
4748 return 0;
4749 }
4750 \f
4751 char output_invalid_buf[8];
4752
4753 static char *
4754 output_invalid (c)
4755 int c;
4756 {
4757 if (ISPRINT (c))
4758 sprintf (output_invalid_buf, "'%c'", c);
4759 else
4760 sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
4761 return output_invalid_buf;
4762 }
4763
4764 /* REG_STRING starts *before* REGISTER_PREFIX. */
4765
4766 static const reg_entry *
4767 parse_register (reg_string, end_op)
4768 char *reg_string;
4769 char **end_op;
4770 {
4771 char *s = reg_string;
4772 char *p;
4773 char reg_name_given[MAX_REG_NAME_SIZE + 1];
4774 const reg_entry *r;
4775
4776 /* Skip possible REGISTER_PREFIX and possible whitespace. */
4777 if (*s == REGISTER_PREFIX)
4778 ++s;
4779
4780 if (is_space_char (*s))
4781 ++s;
4782
4783 p = reg_name_given;
4784 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
4785 {
4786 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
4787 return (const reg_entry *) NULL;
4788 s++;
4789 }
4790
4791 /* For naked regs, make sure that we are not dealing with an identifier.
4792 This prevents confusing an identifier like `eax_var' with register
4793 `eax'. */
4794 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
4795 return (const reg_entry *) NULL;
4796
4797 *end_op = s;
4798
4799 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
4800
4801 /* Handle floating point regs, allowing spaces in the (i) part. */
4802 if (r == i386_regtab /* %st is first entry of table */)
4803 {
4804 if (is_space_char (*s))
4805 ++s;
4806 if (*s == '(')
4807 {
4808 ++s;
4809 if (is_space_char (*s))
4810 ++s;
4811 if (*s >= '0' && *s <= '7')
4812 {
4813 r = &i386_float_regtab[*s - '0'];
4814 ++s;
4815 if (is_space_char (*s))
4816 ++s;
4817 if (*s == ')')
4818 {
4819 *end_op = s + 1;
4820 return r;
4821 }
4822 }
4823 /* We have "%st(" then garbage. */
4824 return (const reg_entry *) NULL;
4825 }
4826 }
4827
4828 if (r != NULL
4829 && (r->reg_flags & (RegRex64 | RegRex)) != 0
4830 && flag_code != CODE_64BIT)
4831 {
4832 return (const reg_entry *) NULL;
4833 }
4834
4835 return r;
4836 }
4837 \f
4838 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4839 const char *md_shortopts = "kVQ:sq";
4840 #else
4841 const char *md_shortopts = "q";
4842 #endif
4843
4844 struct option md_longopts[] = {
4845 #define OPTION_32 (OPTION_MD_BASE + 0)
4846 {"32", no_argument, NULL, OPTION_32},
4847 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4848 #define OPTION_64 (OPTION_MD_BASE + 1)
4849 {"64", no_argument, NULL, OPTION_64},
4850 #endif
4851 {NULL, no_argument, NULL, 0}
4852 };
4853 size_t md_longopts_size = sizeof (md_longopts);
4854
4855 int
4856 md_parse_option (c, arg)
4857 int c;
4858 char *arg ATTRIBUTE_UNUSED;
4859 {
4860 switch (c)
4861 {
4862 case 'q':
4863 quiet_warnings = 1;
4864 break;
4865
4866 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4867 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
4868 should be emitted or not. FIXME: Not implemented. */
4869 case 'Q':
4870 break;
4871
4872 /* -V: SVR4 argument to print version ID. */
4873 case 'V':
4874 print_version_id ();
4875 break;
4876
4877 /* -k: Ignore for FreeBSD compatibility. */
4878 case 'k':
4879 break;
4880
4881 case 's':
4882 /* -s: On i386 Solaris, this tells the native assembler to use
4883 .stab instead of .stab.excl. We always use .stab anyhow. */
4884 break;
4885
4886 case OPTION_64:
4887 {
4888 const char **list, **l;
4889
4890 list = bfd_target_list ();
4891 for (l = list; *l != NULL; l++)
4892 if (strcmp (*l, "elf64-x86-64") == 0)
4893 {
4894 default_arch = "x86_64";
4895 break;
4896 }
4897 if (*l == NULL)
4898 as_fatal (_("No compiled in support for x86_64"));
4899 free (list);
4900 }
4901 break;
4902 #endif
4903
4904 case OPTION_32:
4905 default_arch = "i386";
4906 break;
4907
4908 default:
4909 return 0;
4910 }
4911 return 1;
4912 }
4913
4914 void
4915 md_show_usage (stream)
4916 FILE *stream;
4917 {
4918 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4919 fprintf (stream, _("\
4920 -Q ignored\n\
4921 -V print assembler version number\n\
4922 -k ignored\n\
4923 -q quieten some warnings\n\
4924 -s ignored\n"));
4925 #else
4926 fprintf (stream, _("\
4927 -q quieten some warnings\n"));
4928 #endif
4929 }
4930
4931 #ifdef BFD_ASSEMBLER
4932 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
4933 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
4934
4935 /* Pick the target format to use. */
4936
4937 const char *
4938 i386_target_format ()
4939 {
4940 if (!strcmp (default_arch, "x86_64"))
4941 set_code_flag (CODE_64BIT);
4942 else if (!strcmp (default_arch, "i386"))
4943 set_code_flag (CODE_32BIT);
4944 else
4945 as_fatal (_("Unknown architecture"));
4946 switch (OUTPUT_FLAVOR)
4947 {
4948 #ifdef OBJ_MAYBE_AOUT
4949 case bfd_target_aout_flavour:
4950 return AOUT_TARGET_FORMAT;
4951 #endif
4952 #ifdef OBJ_MAYBE_COFF
4953 case bfd_target_coff_flavour:
4954 return "coff-i386";
4955 #endif
4956 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
4957 case bfd_target_elf_flavour:
4958 {
4959 if (flag_code == CODE_64BIT)
4960 use_rela_relocations = 1;
4961 return flag_code == CODE_64BIT ? "elf64-x86-64" : ELF_TARGET_FORMAT;
4962 }
4963 #endif
4964 default:
4965 abort ();
4966 return NULL;
4967 }
4968 }
4969
4970 #endif /* OBJ_MAYBE_ more than one */
4971
4972 #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
4973 void i386_elf_emit_arch_note ()
4974 {
4975 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4976 && cpu_arch_name != NULL)
4977 {
4978 char *p;
4979 asection *seg = now_seg;
4980 subsegT subseg = now_subseg;
4981 Elf_Internal_Note i_note;
4982 Elf_External_Note e_note;
4983 asection *note_secp;
4984 int len;
4985
4986 /* Create the .note section. */
4987 note_secp = subseg_new (".note", 0);
4988 bfd_set_section_flags (stdoutput,
4989 note_secp,
4990 SEC_HAS_CONTENTS | SEC_READONLY);
4991
4992 /* Process the arch string. */
4993 len = strlen (cpu_arch_name);
4994
4995 i_note.namesz = len + 1;
4996 i_note.descsz = 0;
4997 i_note.type = NT_ARCH;
4998 p = frag_more (sizeof (e_note.namesz));
4999 md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
5000 p = frag_more (sizeof (e_note.descsz));
5001 md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
5002 p = frag_more (sizeof (e_note.type));
5003 md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
5004 p = frag_more (len + 1);
5005 strcpy (p, cpu_arch_name);
5006
5007 frag_align (2, 0, 0);
5008
5009 subseg_set (seg, subseg);
5010 }
5011 }
5012 #endif
5013 #endif /* BFD_ASSEMBLER */
5014 \f
5015 symbolS *
5016 md_undefined_symbol (name)
5017 char *name;
5018 {
5019 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
5020 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
5021 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
5022 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
5023 {
5024 if (!GOT_symbol)
5025 {
5026 if (symbol_find (name))
5027 as_bad (_("GOT already in symbol table"));
5028 GOT_symbol = symbol_new (name, undefined_section,
5029 (valueT) 0, &zero_address_frag);
5030 };
5031 return GOT_symbol;
5032 }
5033 return 0;
5034 }
5035
5036 /* Round up a section size to the appropriate boundary. */
5037
5038 valueT
5039 md_section_align (segment, size)
5040 segT segment ATTRIBUTE_UNUSED;
5041 valueT size;
5042 {
5043 #ifdef BFD_ASSEMBLER
5044 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
5045 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
5046 {
5047 /* For a.out, force the section size to be aligned. If we don't do
5048 this, BFD will align it for us, but it will not write out the
5049 final bytes of the section. This may be a bug in BFD, but it is
5050 easier to fix it here since that is how the other a.out targets
5051 work. */
5052 int align;
5053
5054 align = bfd_get_section_alignment (stdoutput, segment);
5055 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
5056 }
5057 #endif
5058 #endif
5059
5060 return size;
5061 }
5062
5063 /* On the i386, PC-relative offsets are relative to the start of the
5064 next instruction. That is, the address of the offset, plus its
5065 size, since the offset is always the last part of the insn. */
5066
5067 long
5068 md_pcrel_from (fixP)
5069 fixS *fixP;
5070 {
5071 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
5072 }
5073
5074 #ifndef I386COFF
5075
5076 static void
5077 s_bss (ignore)
5078 int ignore ATTRIBUTE_UNUSED;
5079 {
5080 int temp;
5081
5082 temp = get_absolute_expression ();
5083 subseg_set (bss_section, (subsegT) temp);
5084 demand_empty_rest_of_line ();
5085 }
5086
5087 #endif
5088
5089 #ifdef BFD_ASSEMBLER
5090
5091 void
5092 i386_validate_fix (fixp)
5093 fixS *fixp;
5094 {
5095 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
5096 {
5097 /* GOTOFF relocation are nonsense in 64bit mode. */
5098 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
5099 {
5100 if (flag_code != CODE_64BIT)
5101 abort ();
5102 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
5103 }
5104 else
5105 {
5106 if (flag_code == CODE_64BIT)
5107 abort ();
5108 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
5109 }
5110 fixp->fx_subsy = 0;
5111 }
5112 }
5113
5114 boolean
5115 i386_force_relocation (fixp)
5116 fixS *fixp;
5117 {
5118 if (fixp->fx_r_type == BFD_RELOC_VTABLE_INHERIT
5119 || fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
5120 return 1;
5121
5122 return S_FORCE_RELOC (fixp->fx_addsy);
5123 }
5124
5125 arelent *
5126 tc_gen_reloc (section, fixp)
5127 asection *section ATTRIBUTE_UNUSED;
5128 fixS *fixp;
5129 {
5130 arelent *rel;
5131 bfd_reloc_code_real_type code;
5132
5133 switch (fixp->fx_r_type)
5134 {
5135 case BFD_RELOC_X86_64_PLT32:
5136 case BFD_RELOC_X86_64_GOT32:
5137 case BFD_RELOC_X86_64_GOTPCREL:
5138 case BFD_RELOC_386_PLT32:
5139 case BFD_RELOC_386_GOT32:
5140 case BFD_RELOC_386_GOTOFF:
5141 case BFD_RELOC_386_GOTPC:
5142 case BFD_RELOC_386_TLS_GD:
5143 case BFD_RELOC_386_TLS_LDM:
5144 case BFD_RELOC_386_TLS_LDO_32:
5145 case BFD_RELOC_386_TLS_IE_32:
5146 case BFD_RELOC_386_TLS_IE:
5147 case BFD_RELOC_386_TLS_GOTIE:
5148 case BFD_RELOC_386_TLS_LE_32:
5149 case BFD_RELOC_386_TLS_LE:
5150 case BFD_RELOC_X86_64_32S:
5151 case BFD_RELOC_RVA:
5152 case BFD_RELOC_VTABLE_ENTRY:
5153 case BFD_RELOC_VTABLE_INHERIT:
5154 code = fixp->fx_r_type;
5155 break;
5156 default:
5157 if (fixp->fx_pcrel)
5158 {
5159 switch (fixp->fx_size)
5160 {
5161 default:
5162 as_bad_where (fixp->fx_file, fixp->fx_line,
5163 _("can not do %d byte pc-relative relocation"),
5164 fixp->fx_size);
5165 code = BFD_RELOC_32_PCREL;
5166 break;
5167 case 1: code = BFD_RELOC_8_PCREL; break;
5168 case 2: code = BFD_RELOC_16_PCREL; break;
5169 case 4: code = BFD_RELOC_32_PCREL; break;
5170 }
5171 }
5172 else
5173 {
5174 switch (fixp->fx_size)
5175 {
5176 default:
5177 as_bad_where (fixp->fx_file, fixp->fx_line,
5178 _("can not do %d byte relocation"),
5179 fixp->fx_size);
5180 code = BFD_RELOC_32;
5181 break;
5182 case 1: code = BFD_RELOC_8; break;
5183 case 2: code = BFD_RELOC_16; break;
5184 case 4: code = BFD_RELOC_32; break;
5185 #ifdef BFD64
5186 case 8: code = BFD_RELOC_64; break;
5187 #endif
5188 }
5189 }
5190 break;
5191 }
5192
5193 if (code == BFD_RELOC_32
5194 && GOT_symbol
5195 && fixp->fx_addsy == GOT_symbol)
5196 {
5197 /* We don't support GOTPC on 64bit targets. */
5198 if (flag_code == CODE_64BIT)
5199 abort ();
5200 code = BFD_RELOC_386_GOTPC;
5201 }
5202
5203 rel = (arelent *) xmalloc (sizeof (arelent));
5204 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
5205 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
5206
5207 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
5208 if (!use_rela_relocations)
5209 {
5210 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
5211 vtable entry to be used in the relocation's section offset. */
5212 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
5213 rel->address = fixp->fx_offset;
5214
5215 rel->addend = 0;
5216 }
5217 /* Use the rela in 64bit mode. */
5218 else
5219 {
5220 if (!fixp->fx_pcrel)
5221 rel->addend = fixp->fx_offset;
5222 else
5223 switch (code)
5224 {
5225 case BFD_RELOC_X86_64_PLT32:
5226 case BFD_RELOC_X86_64_GOT32:
5227 case BFD_RELOC_X86_64_GOTPCREL:
5228 rel->addend = fixp->fx_offset - fixp->fx_size;
5229 break;
5230 default:
5231 rel->addend = (section->vma
5232 - fixp->fx_size
5233 + fixp->fx_addnumber
5234 + md_pcrel_from (fixp));
5235 break;
5236 }
5237 }
5238
5239 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
5240 if (rel->howto == NULL)
5241 {
5242 as_bad_where (fixp->fx_file, fixp->fx_line,
5243 _("cannot represent relocation type %s"),
5244 bfd_get_reloc_code_name (code));
5245 /* Set howto to a garbage value so that we can keep going. */
5246 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
5247 assert (rel->howto != NULL);
5248 }
5249
5250 return rel;
5251 }
5252
5253 #else /* !BFD_ASSEMBLER */
5254
5255 #if (defined(OBJ_AOUT) | defined(OBJ_BOUT))
5256 void
5257 tc_aout_fix_to_chars (where, fixP, segment_address_in_file)
5258 char *where;
5259 fixS *fixP;
5260 relax_addressT segment_address_in_file;
5261 {
5262 /* In: length of relocation (or of address) in chars: 1, 2 or 4.
5263 Out: GNU LD relocation length code: 0, 1, or 2. */
5264
5265 static const unsigned char nbytes_r_length[] = { 42, 0, 1, 42, 2 };
5266 long r_symbolnum;
5267
5268 know (fixP->fx_addsy != NULL);
5269
5270 md_number_to_chars (where,
5271 (valueT) (fixP->fx_frag->fr_address
5272 + fixP->fx_where - segment_address_in_file),
5273 4);
5274
5275 r_symbolnum = (S_IS_DEFINED (fixP->fx_addsy)
5276 ? S_GET_TYPE (fixP->fx_addsy)
5277 : fixP->fx_addsy->sy_number);
5278
5279 where[6] = (r_symbolnum >> 16) & 0x0ff;
5280 where[5] = (r_symbolnum >> 8) & 0x0ff;
5281 where[4] = r_symbolnum & 0x0ff;
5282 where[7] = ((((!S_IS_DEFINED (fixP->fx_addsy)) << 3) & 0x08)
5283 | ((nbytes_r_length[fixP->fx_size] << 1) & 0x06)
5284 | (((fixP->fx_pcrel << 0) & 0x01) & 0x0f));
5285 }
5286
5287 #endif /* OBJ_AOUT or OBJ_BOUT. */
5288
5289 #if defined (I386COFF)
5290
5291 short
5292 tc_coff_fix2rtype (fixP)
5293 fixS *fixP;
5294 {
5295 if (fixP->fx_r_type == R_IMAGEBASE)
5296 return R_IMAGEBASE;
5297
5298 return (fixP->fx_pcrel ?
5299 (fixP->fx_size == 1 ? R_PCRBYTE :
5300 fixP->fx_size == 2 ? R_PCRWORD :
5301 R_PCRLONG) :
5302 (fixP->fx_size == 1 ? R_RELBYTE :
5303 fixP->fx_size == 2 ? R_RELWORD :
5304 R_DIR32));
5305 }
5306
5307 int
5308 tc_coff_sizemachdep (frag)
5309 fragS *frag;
5310 {
5311 if (frag->fr_next)
5312 return (frag->fr_next->fr_address - frag->fr_address);
5313 else
5314 return 0;
5315 }
5316
5317 #endif /* I386COFF */
5318
5319 #endif /* !BFD_ASSEMBLER */
5320 \f
5321 /* Parse operands using Intel syntax. This implements a recursive descent
5322 parser based on the BNF grammar published in Appendix B of the MASM 6.1
5323 Programmer's Guide.
5324
5325 FIXME: We do not recognize the full operand grammar defined in the MASM
5326 documentation. In particular, all the structure/union and
5327 high-level macro operands are missing.
5328
5329 Uppercase words are terminals, lower case words are non-terminals.
5330 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
5331 bars '|' denote choices. Most grammar productions are implemented in
5332 functions called 'intel_<production>'.
5333
5334 Initial production is 'expr'.
5335
5336 addOp + | -
5337
5338 alpha [a-zA-Z]
5339
5340 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
5341
5342 constant digits [[ radixOverride ]]
5343
5344 dataType BYTE | WORD | DWORD | QWORD | XWORD
5345
5346 digits decdigit
5347 | digits decdigit
5348 | digits hexdigit
5349
5350 decdigit [0-9]
5351
5352 e05 e05 addOp e06
5353 | e06
5354
5355 e06 e06 mulOp e09
5356 | e09
5357
5358 e09 OFFSET e10
5359 | e09 PTR e10
5360 | e09 : e10
5361 | e10
5362
5363 e10 e10 [ expr ]
5364 | e11
5365
5366 e11 ( expr )
5367 | [ expr ]
5368 | constant
5369 | dataType
5370 | id
5371 | $
5372 | register
5373
5374 => expr SHORT e05
5375 | e05
5376
5377 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
5378 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
5379
5380 hexdigit a | b | c | d | e | f
5381 | A | B | C | D | E | F
5382
5383 id alpha
5384 | id alpha
5385 | id decdigit
5386
5387 mulOp * | / | MOD
5388
5389 quote " | '
5390
5391 register specialRegister
5392 | gpRegister
5393 | byteRegister
5394
5395 segmentRegister CS | DS | ES | FS | GS | SS
5396
5397 specialRegister CR0 | CR2 | CR3
5398 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
5399 | TR3 | TR4 | TR5 | TR6 | TR7
5400
5401 We simplify the grammar in obvious places (e.g., register parsing is
5402 done by calling parse_register) and eliminate immediate left recursion
5403 to implement a recursive-descent parser.
5404
5405 expr SHORT e05
5406 | e05
5407
5408 e05 e06 e05'
5409
5410 e05' addOp e06 e05'
5411 | Empty
5412
5413 e06 e09 e06'
5414
5415 e06' mulOp e09 e06'
5416 | Empty
5417
5418 e09 OFFSET e10 e09'
5419 | e10 e09'
5420
5421 e09' PTR e10 e09'
5422 | : e10 e09'
5423 | Empty
5424
5425 e10 e11 e10'
5426
5427 e10' [ expr ] e10'
5428 | Empty
5429
5430 e11 ( expr )
5431 | [ expr ]
5432 | BYTE
5433 | WORD
5434 | DWORD
5435 | QWORD
5436 | XWORD
5437 | .
5438 | $
5439 | register
5440 | id
5441 | constant */
5442
5443 /* Parsing structure for the intel syntax parser. Used to implement the
5444 semantic actions for the operand grammar. */
5445 struct intel_parser_s
5446 {
5447 char *op_string; /* The string being parsed. */
5448 int got_a_float; /* Whether the operand is a float. */
5449 int op_modifier; /* Operand modifier. */
5450 int is_mem; /* 1 if operand is memory reference. */
5451 const reg_entry *reg; /* Last register reference found. */
5452 char *disp; /* Displacement string being built. */
5453 };
5454
5455 static struct intel_parser_s intel_parser;
5456
5457 /* Token structure for parsing intel syntax. */
5458 struct intel_token
5459 {
5460 int code; /* Token code. */
5461 const reg_entry *reg; /* Register entry for register tokens. */
5462 char *str; /* String representation. */
5463 };
5464
5465 static struct intel_token cur_token, prev_token;
5466
5467 /* Token codes for the intel parser. Since T_SHORT is already used
5468 by COFF, undefine it first to prevent a warning. */
5469 #define T_NIL -1
5470 #define T_CONST 1
5471 #define T_REG 2
5472 #define T_BYTE 3
5473 #define T_WORD 4
5474 #define T_DWORD 5
5475 #define T_QWORD 6
5476 #define T_XWORD 7
5477 #undef T_SHORT
5478 #define T_SHORT 8
5479 #define T_OFFSET 9
5480 #define T_PTR 10
5481 #define T_ID 11
5482
5483 /* Prototypes for intel parser functions. */
5484 static int intel_match_token PARAMS ((int code));
5485 static void intel_get_token PARAMS ((void));
5486 static void intel_putback_token PARAMS ((void));
5487 static int intel_expr PARAMS ((void));
5488 static int intel_e05 PARAMS ((void));
5489 static int intel_e05_1 PARAMS ((void));
5490 static int intel_e06 PARAMS ((void));
5491 static int intel_e06_1 PARAMS ((void));
5492 static int intel_e09 PARAMS ((void));
5493 static int intel_e09_1 PARAMS ((void));
5494 static int intel_e10 PARAMS ((void));
5495 static int intel_e10_1 PARAMS ((void));
5496 static int intel_e11 PARAMS ((void));
5497
5498 static int
5499 i386_intel_operand (operand_string, got_a_float)
5500 char *operand_string;
5501 int got_a_float;
5502 {
5503 int ret;
5504 char *p;
5505
5506 /* Initialize token holders. */
5507 cur_token.code = prev_token.code = T_NIL;
5508 cur_token.reg = prev_token.reg = NULL;
5509 cur_token.str = prev_token.str = NULL;
5510
5511 /* Initialize parser structure. */
5512 p = intel_parser.op_string = (char *) malloc (strlen (operand_string) + 1);
5513 if (p == NULL)
5514 abort ();
5515 strcpy (intel_parser.op_string, operand_string);
5516 intel_parser.got_a_float = got_a_float;
5517 intel_parser.op_modifier = -1;
5518 intel_parser.is_mem = 0;
5519 intel_parser.reg = NULL;
5520 intel_parser.disp = (char *) malloc (strlen (operand_string) + 1);
5521 if (intel_parser.disp == NULL)
5522 abort ();
5523 intel_parser.disp[0] = '\0';
5524
5525 /* Read the first token and start the parser. */
5526 intel_get_token ();
5527 ret = intel_expr ();
5528
5529 if (ret)
5530 {
5531 /* If we found a memory reference, hand it over to i386_displacement
5532 to fill in the rest of the operand fields. */
5533 if (intel_parser.is_mem)
5534 {
5535 if ((i.mem_operands == 1
5536 && (current_templates->start->opcode_modifier & IsString) == 0)
5537 || i.mem_operands == 2)
5538 {
5539 as_bad (_("too many memory references for '%s'"),
5540 current_templates->start->name);
5541 ret = 0;
5542 }
5543 else
5544 {
5545 char *s = intel_parser.disp;
5546 i.mem_operands++;
5547
5548 /* Add the displacement expression. */
5549 if (*s != '\0')
5550 ret = i386_displacement (s, s + strlen (s))
5551 && i386_index_check (s);
5552 }
5553 }
5554
5555 /* Constant and OFFSET expressions are handled by i386_immediate. */
5556 else if (intel_parser.op_modifier == OFFSET_FLAT
5557 || intel_parser.reg == NULL)
5558 ret = i386_immediate (intel_parser.disp);
5559 }
5560
5561 free (p);
5562 free (intel_parser.disp);
5563
5564 return ret;
5565 }
5566
5567 /* expr SHORT e05
5568 | e05 */
5569 static int
5570 intel_expr ()
5571 {
5572 /* expr SHORT e05 */
5573 if (cur_token.code == T_SHORT)
5574 {
5575 intel_parser.op_modifier = SHORT;
5576 intel_match_token (T_SHORT);
5577
5578 return (intel_e05 ());
5579 }
5580
5581 /* expr e05 */
5582 else
5583 return intel_e05 ();
5584 }
5585
5586 /* e05 e06 e05'
5587
5588 e05' addOp e06 e05'
5589 | Empty */
5590 static int
5591 intel_e05 ()
5592 {
5593 return (intel_e06 () && intel_e05_1 ());
5594 }
5595
5596 static int
5597 intel_e05_1 ()
5598 {
5599 /* e05' addOp e06 e05' */
5600 if (cur_token.code == '+' || cur_token.code == '-')
5601 {
5602 strcat (intel_parser.disp, cur_token.str);
5603 intel_match_token (cur_token.code);
5604
5605 return (intel_e06 () && intel_e05_1 ());
5606 }
5607
5608 /* e05' Empty */
5609 else
5610 return 1;
5611 }
5612
5613 /* e06 e09 e06'
5614
5615 e06' mulOp e09 e06'
5616 | Empty */
5617 static int
5618 intel_e06 ()
5619 {
5620 return (intel_e09 () && intel_e06_1 ());
5621 }
5622
5623 static int
5624 intel_e06_1 ()
5625 {
5626 /* e06' mulOp e09 e06' */
5627 if (cur_token.code == '*' || cur_token.code == '/')
5628 {
5629 strcat (intel_parser.disp, cur_token.str);
5630 intel_match_token (cur_token.code);
5631
5632 return (intel_e09 () && intel_e06_1 ());
5633 }
5634
5635 /* e06' Empty */
5636 else
5637 return 1;
5638 }
5639
5640 /* e09 OFFSET e10 e09'
5641 | e10 e09'
5642
5643 e09' PTR e10 e09'
5644 | : e10 e09'
5645 | Empty */
5646 static int
5647 intel_e09 ()
5648 {
5649 /* e09 OFFSET e10 e09' */
5650 if (cur_token.code == T_OFFSET)
5651 {
5652 intel_parser.is_mem = 0;
5653 intel_parser.op_modifier = OFFSET_FLAT;
5654 intel_match_token (T_OFFSET);
5655
5656 return (intel_e10 () && intel_e09_1 ());
5657 }
5658
5659 /* e09 e10 e09' */
5660 else
5661 return (intel_e10 () && intel_e09_1 ());
5662 }
5663
5664 static int
5665 intel_e09_1 ()
5666 {
5667 /* e09' PTR e10 e09' */
5668 if (cur_token.code == T_PTR)
5669 {
5670 if (prev_token.code == T_BYTE)
5671 i.suffix = BYTE_MNEM_SUFFIX;
5672
5673 else if (prev_token.code == T_WORD)
5674 {
5675 if (intel_parser.got_a_float == 2) /* "fi..." */
5676 i.suffix = SHORT_MNEM_SUFFIX;
5677 else
5678 i.suffix = WORD_MNEM_SUFFIX;
5679 }
5680
5681 else if (prev_token.code == T_DWORD)
5682 {
5683 if (intel_parser.got_a_float == 1) /* "f..." */
5684 i.suffix = SHORT_MNEM_SUFFIX;
5685 else
5686 i.suffix = LONG_MNEM_SUFFIX;
5687 }
5688
5689 else if (prev_token.code == T_QWORD)
5690 {
5691 if (intel_parser.got_a_float == 1) /* "f..." */
5692 i.suffix = LONG_MNEM_SUFFIX;
5693 else
5694 i.suffix = QWORD_MNEM_SUFFIX;
5695 }
5696
5697 else if (prev_token.code == T_XWORD)
5698 i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
5699
5700 else
5701 {
5702 as_bad (_("Unknown operand modifier `%s'\n"), prev_token.str);
5703 return 0;
5704 }
5705
5706 intel_match_token (T_PTR);
5707
5708 return (intel_e10 () && intel_e09_1 ());
5709 }
5710
5711 /* e09 : e10 e09' */
5712 else if (cur_token.code == ':')
5713 {
5714 /* Mark as a memory operand only if it's not already known to be an
5715 offset expression. */
5716 if (intel_parser.op_modifier != OFFSET_FLAT)
5717 intel_parser.is_mem = 1;
5718
5719 return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
5720 }
5721
5722 /* e09' Empty */
5723 else
5724 return 1;
5725 }
5726
5727 /* e10 e11 e10'
5728
5729 e10' [ expr ] e10'
5730 | Empty */
5731 static int
5732 intel_e10 ()
5733 {
5734 return (intel_e11 () && intel_e10_1 ());
5735 }
5736
5737 static int
5738 intel_e10_1 ()
5739 {
5740 /* e10' [ expr ] e10' */
5741 if (cur_token.code == '[')
5742 {
5743 intel_match_token ('[');
5744
5745 /* Mark as a memory operand only if it's not already known to be an
5746 offset expression. If it's an offset expression, we need to keep
5747 the brace in. */
5748 if (intel_parser.op_modifier != OFFSET_FLAT)
5749 intel_parser.is_mem = 1;
5750 else
5751 strcat (intel_parser.disp, "[");
5752
5753 /* Add a '+' to the displacement string if necessary. */
5754 if (*intel_parser.disp != '\0'
5755 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5756 strcat (intel_parser.disp, "+");
5757
5758 if (intel_expr () && intel_match_token (']'))
5759 {
5760 /* Preserve brackets when the operand is an offset expression. */
5761 if (intel_parser.op_modifier == OFFSET_FLAT)
5762 strcat (intel_parser.disp, "]");
5763
5764 return intel_e10_1 ();
5765 }
5766 else
5767 return 0;
5768 }
5769
5770 /* e10' Empty */
5771 else
5772 return 1;
5773 }
5774
5775 /* e11 ( expr )
5776 | [ expr ]
5777 | BYTE
5778 | WORD
5779 | DWORD
5780 | QWORD
5781 | XWORD
5782 | $
5783 | .
5784 | register
5785 | id
5786 | constant */
5787 static int
5788 intel_e11 ()
5789 {
5790 /* e11 ( expr ) */
5791 if (cur_token.code == '(')
5792 {
5793 intel_match_token ('(');
5794 strcat (intel_parser.disp, "(");
5795
5796 if (intel_expr () && intel_match_token (')'))
5797 {
5798 strcat (intel_parser.disp, ")");
5799 return 1;
5800 }
5801 else
5802 return 0;
5803 }
5804
5805 /* e11 [ expr ] */
5806 else if (cur_token.code == '[')
5807 {
5808 intel_match_token ('[');
5809
5810 /* Mark as a memory operand only if it's not already known to be an
5811 offset expression. If it's an offset expression, we need to keep
5812 the brace in. */
5813 if (intel_parser.op_modifier != OFFSET_FLAT)
5814 intel_parser.is_mem = 1;
5815 else
5816 strcat (intel_parser.disp, "[");
5817
5818 /* Operands for jump/call inside brackets denote absolute addresses. */
5819 if (current_templates->start->opcode_modifier & Jump
5820 || current_templates->start->opcode_modifier & JumpDword
5821 || current_templates->start->opcode_modifier & JumpByte
5822 || current_templates->start->opcode_modifier & JumpInterSegment)
5823 i.types[this_operand] |= JumpAbsolute;
5824
5825 /* Add a '+' to the displacement string if necessary. */
5826 if (*intel_parser.disp != '\0'
5827 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5828 strcat (intel_parser.disp, "+");
5829
5830 if (intel_expr () && intel_match_token (']'))
5831 {
5832 /* Preserve brackets when the operand is an offset expression. */
5833 if (intel_parser.op_modifier == OFFSET_FLAT)
5834 strcat (intel_parser.disp, "]");
5835
5836 return 1;
5837 }
5838 else
5839 return 0;
5840 }
5841
5842 /* e11 BYTE
5843 | WORD
5844 | DWORD
5845 | QWORD
5846 | XWORD */
5847 else if (cur_token.code == T_BYTE
5848 || cur_token.code == T_WORD
5849 || cur_token.code == T_DWORD
5850 || cur_token.code == T_QWORD
5851 || cur_token.code == T_XWORD)
5852 {
5853 intel_match_token (cur_token.code);
5854
5855 return 1;
5856 }
5857
5858 /* e11 $
5859 | . */
5860 else if (cur_token.code == '$' || cur_token.code == '.')
5861 {
5862 strcat (intel_parser.disp, cur_token.str);
5863 intel_match_token (cur_token.code);
5864
5865 /* Mark as a memory operand only if it's not already known to be an
5866 offset expression. */
5867 if (intel_parser.op_modifier != OFFSET_FLAT)
5868 intel_parser.is_mem = 1;
5869
5870 return 1;
5871 }
5872
5873 /* e11 register */
5874 else if (cur_token.code == T_REG)
5875 {
5876 const reg_entry *reg = intel_parser.reg = cur_token.reg;
5877
5878 intel_match_token (T_REG);
5879
5880 /* Check for segment change. */
5881 if (cur_token.code == ':')
5882 {
5883 if (reg->reg_type & (SReg2 | SReg3))
5884 {
5885 switch (reg->reg_num)
5886 {
5887 case 0:
5888 i.seg[i.mem_operands] = &es;
5889 break;
5890 case 1:
5891 i.seg[i.mem_operands] = &cs;
5892 break;
5893 case 2:
5894 i.seg[i.mem_operands] = &ss;
5895 break;
5896 case 3:
5897 i.seg[i.mem_operands] = &ds;
5898 break;
5899 case 4:
5900 i.seg[i.mem_operands] = &fs;
5901 break;
5902 case 5:
5903 i.seg[i.mem_operands] = &gs;
5904 break;
5905 }
5906 }
5907 else
5908 {
5909 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
5910 return 0;
5911 }
5912 }
5913
5914 /* Not a segment register. Check for register scaling. */
5915 else if (cur_token.code == '*')
5916 {
5917 if (!intel_parser.is_mem)
5918 {
5919 as_bad (_("Register scaling only allowed in memory operands."));
5920 return 0;
5921 }
5922
5923 /* What follows must be a valid scale. */
5924 if (intel_match_token ('*')
5925 && strchr ("01248", *cur_token.str))
5926 {
5927 i.index_reg = reg;
5928 i.types[this_operand] |= BaseIndex;
5929
5930 /* Set the scale after setting the register (otherwise,
5931 i386_scale will complain) */
5932 i386_scale (cur_token.str);
5933 intel_match_token (T_CONST);
5934 }
5935 else
5936 {
5937 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
5938 cur_token.str);
5939 return 0;
5940 }
5941 }
5942
5943 /* No scaling. If this is a memory operand, the register is either a
5944 base register (first occurrence) or an index register (second
5945 occurrence). */
5946 else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3)))
5947 {
5948 if (i.base_reg && i.index_reg)
5949 {
5950 as_bad (_("Too many register references in memory operand.\n"));
5951 return 0;
5952 }
5953
5954 if (i.base_reg == NULL)
5955 i.base_reg = reg;
5956 else
5957 i.index_reg = reg;
5958
5959 i.types[this_operand] |= BaseIndex;
5960 }
5961
5962 /* Offset modifier. Add the register to the displacement string to be
5963 parsed as an immediate expression after we're done. */
5964 else if (intel_parser.op_modifier == OFFSET_FLAT)
5965 strcat (intel_parser.disp, reg->reg_name);
5966
5967 /* It's neither base nor index nor offset. */
5968 else
5969 {
5970 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
5971 i.op[this_operand].regs = reg;
5972 i.reg_operands++;
5973 }
5974
5975 /* Since registers are not part of the displacement string (except
5976 when we're parsing offset operands), we may need to remove any
5977 preceding '+' from the displacement string. */
5978 if (*intel_parser.disp != '\0'
5979 && intel_parser.op_modifier != OFFSET_FLAT)
5980 {
5981 char *s = intel_parser.disp;
5982 s += strlen (s) - 1;
5983 if (*s == '+')
5984 *s = '\0';
5985 }
5986
5987 return 1;
5988 }
5989
5990 /* e11 id */
5991 else if (cur_token.code == T_ID)
5992 {
5993 /* Add the identifier to the displacement string. */
5994 strcat (intel_parser.disp, cur_token.str);
5995 intel_match_token (T_ID);
5996
5997 /* The identifier represents a memory reference only if it's not
5998 preceded by an offset modifier. */
5999 if (intel_parser.op_modifier != OFFSET_FLAT)
6000 intel_parser.is_mem = 1;
6001
6002 return 1;
6003 }
6004
6005 /* e11 constant */
6006 else if (cur_token.code == T_CONST
6007 || cur_token.code == '-'
6008 || cur_token.code == '+')
6009 {
6010 char *save_str;
6011
6012 /* Allow constants that start with `+' or `-'. */
6013 if (cur_token.code == '-' || cur_token.code == '+')
6014 {
6015 strcat (intel_parser.disp, cur_token.str);
6016 intel_match_token (cur_token.code);
6017 if (cur_token.code != T_CONST)
6018 {
6019 as_bad (_("Syntax error. Expecting a constant. Got `%s'.\n"),
6020 cur_token.str);
6021 return 0;
6022 }
6023 }
6024
6025 save_str = (char *) malloc (strlen (cur_token.str) + 1);
6026 if (save_str == NULL)
6027 abort ();
6028 strcpy (save_str, cur_token.str);
6029
6030 /* Get the next token to check for register scaling. */
6031 intel_match_token (cur_token.code);
6032
6033 /* Check if this constant is a scaling factor for an index register. */
6034 if (cur_token.code == '*')
6035 {
6036 if (intel_match_token ('*') && cur_token.code == T_REG)
6037 {
6038 if (!intel_parser.is_mem)
6039 {
6040 as_bad (_("Register scaling only allowed in memory operands."));
6041 return 0;
6042 }
6043
6044 /* The constant is followed by `* reg', so it must be
6045 a valid scale. */
6046 if (strchr ("01248", *save_str))
6047 {
6048 i.index_reg = cur_token.reg;
6049 i.types[this_operand] |= BaseIndex;
6050
6051 /* Set the scale after setting the register (otherwise,
6052 i386_scale will complain) */
6053 i386_scale (save_str);
6054 intel_match_token (T_REG);
6055
6056 /* Since registers are not part of the displacement
6057 string, we may need to remove any preceding '+' from
6058 the displacement string. */
6059 if (*intel_parser.disp != '\0')
6060 {
6061 char *s = intel_parser.disp;
6062 s += strlen (s) - 1;
6063 if (*s == '+')
6064 *s = '\0';
6065 }
6066
6067 free (save_str);
6068
6069 return 1;
6070 }
6071 else
6072 return 0;
6073 }
6074
6075 /* The constant was not used for register scaling. Since we have
6076 already consumed the token following `*' we now need to put it
6077 back in the stream. */
6078 else
6079 intel_putback_token ();
6080 }
6081
6082 /* Add the constant to the displacement string. */
6083 strcat (intel_parser.disp, save_str);
6084 free (save_str);
6085
6086 return 1;
6087 }
6088
6089 as_bad (_("Unrecognized token '%s'"), cur_token.str);
6090 return 0;
6091 }
6092
6093 /* Match the given token against cur_token. If they match, read the next
6094 token from the operand string. */
6095 static int
6096 intel_match_token (code)
6097 int code;
6098 {
6099 if (cur_token.code == code)
6100 {
6101 intel_get_token ();
6102 return 1;
6103 }
6104 else
6105 {
6106 as_bad (_("Unexpected token `%s'\n"), cur_token.str);
6107 return 0;
6108 }
6109 }
6110
6111 /* Read a new token from intel_parser.op_string and store it in cur_token. */
6112 static void
6113 intel_get_token ()
6114 {
6115 char *end_op;
6116 const reg_entry *reg;
6117 struct intel_token new_token;
6118
6119 new_token.code = T_NIL;
6120 new_token.reg = NULL;
6121 new_token.str = NULL;
6122
6123 /* Free the memory allocated to the previous token and move
6124 cur_token to prev_token. */
6125 if (prev_token.str)
6126 free (prev_token.str);
6127
6128 prev_token = cur_token;
6129
6130 /* Skip whitespace. */
6131 while (is_space_char (*intel_parser.op_string))
6132 intel_parser.op_string++;
6133
6134 /* Return an empty token if we find nothing else on the line. */
6135 if (*intel_parser.op_string == '\0')
6136 {
6137 cur_token = new_token;
6138 return;
6139 }
6140
6141 /* The new token cannot be larger than the remainder of the operand
6142 string. */
6143 new_token.str = (char *) malloc (strlen (intel_parser.op_string) + 1);
6144 if (new_token.str == NULL)
6145 abort ();
6146 new_token.str[0] = '\0';
6147
6148 if (strchr ("0123456789", *intel_parser.op_string))
6149 {
6150 char *p = new_token.str;
6151 char *q = intel_parser.op_string;
6152 new_token.code = T_CONST;
6153
6154 /* Allow any kind of identifier char to encompass floating point and
6155 hexadecimal numbers. */
6156 while (is_identifier_char (*q))
6157 *p++ = *q++;
6158 *p = '\0';
6159
6160 /* Recognize special symbol names [0-9][bf]. */
6161 if (strlen (intel_parser.op_string) == 2
6162 && (intel_parser.op_string[1] == 'b'
6163 || intel_parser.op_string[1] == 'f'))
6164 new_token.code = T_ID;
6165 }
6166
6167 else if (strchr ("+-/*:[]()", *intel_parser.op_string))
6168 {
6169 new_token.code = *intel_parser.op_string;
6170 new_token.str[0] = *intel_parser.op_string;
6171 new_token.str[1] = '\0';
6172 }
6173
6174 else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
6175 && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
6176 {
6177 new_token.code = T_REG;
6178 new_token.reg = reg;
6179
6180 if (*intel_parser.op_string == REGISTER_PREFIX)
6181 {
6182 new_token.str[0] = REGISTER_PREFIX;
6183 new_token.str[1] = '\0';
6184 }
6185
6186 strcat (new_token.str, reg->reg_name);
6187 }
6188
6189 else if (is_identifier_char (*intel_parser.op_string))
6190 {
6191 char *p = new_token.str;
6192 char *q = intel_parser.op_string;
6193
6194 /* A '.' or '$' followed by an identifier char is an identifier.
6195 Otherwise, it's operator '.' followed by an expression. */
6196 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
6197 {
6198 new_token.code = *q;
6199 new_token.str[0] = *q;
6200 new_token.str[1] = '\0';
6201 }
6202 else
6203 {
6204 while (is_identifier_char (*q) || *q == '@')
6205 *p++ = *q++;
6206 *p = '\0';
6207
6208 if (strcasecmp (new_token.str, "BYTE") == 0)
6209 new_token.code = T_BYTE;
6210
6211 else if (strcasecmp (new_token.str, "WORD") == 0)
6212 new_token.code = T_WORD;
6213
6214 else if (strcasecmp (new_token.str, "DWORD") == 0)
6215 new_token.code = T_DWORD;
6216
6217 else if (strcasecmp (new_token.str, "QWORD") == 0)
6218 new_token.code = T_QWORD;
6219
6220 else if (strcasecmp (new_token.str, "XWORD") == 0)
6221 new_token.code = T_XWORD;
6222
6223 else if (strcasecmp (new_token.str, "PTR") == 0)
6224 new_token.code = T_PTR;
6225
6226 else if (strcasecmp (new_token.str, "SHORT") == 0)
6227 new_token.code = T_SHORT;
6228
6229 else if (strcasecmp (new_token.str, "OFFSET") == 0)
6230 {
6231 new_token.code = T_OFFSET;
6232
6233 /* ??? This is not mentioned in the MASM grammar but gcc
6234 makes use of it with -mintel-syntax. OFFSET may be
6235 followed by FLAT: */
6236 if (strncasecmp (q, " FLAT:", 6) == 0)
6237 strcat (new_token.str, " FLAT:");
6238 }
6239
6240 /* ??? This is not mentioned in the MASM grammar. */
6241 else if (strcasecmp (new_token.str, "FLAT") == 0)
6242 new_token.code = T_OFFSET;
6243
6244 else
6245 new_token.code = T_ID;
6246 }
6247 }
6248
6249 else
6250 as_bad (_("Unrecognized token `%s'\n"), intel_parser.op_string);
6251
6252 intel_parser.op_string += strlen (new_token.str);
6253 cur_token = new_token;
6254 }
6255
6256 /* Put cur_token back into the token stream and make cur_token point to
6257 prev_token. */
6258 static void
6259 intel_putback_token ()
6260 {
6261 intel_parser.op_string -= strlen (cur_token.str);
6262 free (cur_token.str);
6263 cur_token = prev_token;
6264
6265 /* Forget prev_token. */
6266 prev_token.code = T_NIL;
6267 prev_token.reg = NULL;
6268 prev_token.str = NULL;
6269 }