* as.c (parse_args <OPTION_VERSION>): Use VERSION is
[deliverable/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 ((void));
108 static void output_disp PARAMS ((void));
109 #ifndef I386COFF
110 static void s_bss PARAMS ((int));
111 #endif
112
113 static const char *default_arch = DEFAULT_ARCH;
114
115 /* 'md_assemble ()' gathers together information and puts it into a
116 i386_insn. */
117
118 union i386_op
119 {
120 expressionS *disps;
121 expressionS *imms;
122 const reg_entry *regs;
123 };
124
125 struct _i386_insn
126 {
127 /* TM holds the template for the insn were currently assembling. */
128 template tm;
129
130 /* SUFFIX holds the instruction mnemonic suffix if given.
131 (e.g. 'l' for 'movl') */
132 char suffix;
133
134 /* OPERANDS gives the number of given operands. */
135 unsigned int operands;
136
137 /* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number
138 of given register, displacement, memory operands and immediate
139 operands. */
140 unsigned int reg_operands, disp_operands, mem_operands, imm_operands;
141
142 /* TYPES [i] is the type (see above #defines) which tells us how to
143 use OP[i] for the corresponding operand. */
144 unsigned int types[MAX_OPERANDS];
145
146 /* Displacement expression, immediate expression, or register for each
147 operand. */
148 union i386_op op[MAX_OPERANDS];
149
150 /* Flags for operands. */
151 unsigned int flags[MAX_OPERANDS];
152 #define Operand_PCrel 1
153
154 /* Relocation type for operand */
155 RELOC_ENUM reloc[MAX_OPERANDS];
156
157 /* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode
158 the base index byte below. */
159 const reg_entry *base_reg;
160 const reg_entry *index_reg;
161 unsigned int log2_scale_factor;
162
163 /* SEG gives the seg_entries of this insn. They are zero unless
164 explicit segment overrides are given. */
165 const seg_entry *seg[2];
166
167 /* PREFIX holds all the given prefix opcodes (usually null).
168 PREFIXES is the number of prefix opcodes. */
169 unsigned int prefixes;
170 unsigned char prefix[MAX_PREFIXES];
171
172 /* RM and SIB are the modrm byte and the sib byte where the
173 addressing modes of this insn are encoded. */
174
175 modrm_byte rm;
176 rex_byte rex;
177 sib_byte sib;
178 };
179
180 typedef struct _i386_insn i386_insn;
181
182 /* List of chars besides those in app.c:symbol_chars that can start an
183 operand. Used to prevent the scrubber eating vital white-space. */
184 #ifdef LEX_AT
185 const char extra_symbol_chars[] = "*%-(@";
186 #else
187 const char extra_symbol_chars[] = "*%-(";
188 #endif
189
190 #if (defined (TE_I386AIX) \
191 || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) \
192 && !defined (TE_LINUX) \
193 && !defined (TE_FreeBSD) \
194 && !defined (TE_NetBSD)))
195 /* This array holds the chars that always start a comment. If the
196 pre-processor is disabled, these aren't very useful. */
197 const char comment_chars[] = "#/";
198 #define PREFIX_SEPARATOR '\\'
199
200 /* This array holds the chars that only start a comment at the beginning of
201 a line. If the line seems to have the form '# 123 filename'
202 .line and .file directives will appear in the pre-processed output.
203 Note that input_file.c hand checks for '#' at the beginning of the
204 first line of the input file. This is because the compiler outputs
205 #NO_APP at the beginning of its output.
206 Also note that comments started like this one will always work if
207 '/' isn't otherwise defined. */
208 const char line_comment_chars[] = "";
209
210 #else
211 /* Putting '/' here makes it impossible to use the divide operator.
212 However, we need it for compatibility with SVR4 systems. */
213 const char comment_chars[] = "#";
214 #define PREFIX_SEPARATOR '/'
215
216 const char line_comment_chars[] = "/";
217 #endif
218
219 const char line_separator_chars[] = ";";
220
221 /* Chars that can be used to separate mant from exp in floating point
222 nums. */
223 const char EXP_CHARS[] = "eE";
224
225 /* Chars that mean this number is a floating point constant
226 As in 0f12.456
227 or 0d1.2345e12. */
228 const char FLT_CHARS[] = "fFdDxX";
229
230 /* Tables for lexical analysis. */
231 static char mnemonic_chars[256];
232 static char register_chars[256];
233 static char operand_chars[256];
234 static char identifier_chars[256];
235 static char digit_chars[256];
236
237 /* Lexical macros. */
238 #define is_mnemonic_char(x) (mnemonic_chars[(unsigned char) x])
239 #define is_operand_char(x) (operand_chars[(unsigned char) x])
240 #define is_register_char(x) (register_chars[(unsigned char) x])
241 #define is_space_char(x) ((x) == ' ')
242 #define is_identifier_char(x) (identifier_chars[(unsigned char) x])
243 #define is_digit_char(x) (digit_chars[(unsigned char) x])
244
245 /* All non-digit non-letter charcters that may occur in an operand. */
246 static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
247
248 /* md_assemble() always leaves the strings it's passed unaltered. To
249 effect this we maintain a stack of saved characters that we've smashed
250 with '\0's (indicating end of strings for various sub-fields of the
251 assembler instruction). */
252 static char save_stack[32];
253 static char *save_stack_p;
254 #define END_STRING_AND_SAVE(s) \
255 do { *save_stack_p++ = *(s); *(s) = '\0'; } while (0)
256 #define RESTORE_END_STRING(s) \
257 do { *(s) = *--save_stack_p; } while (0)
258
259 /* The instruction we're assembling. */
260 static i386_insn i;
261
262 /* Possible templates for current insn. */
263 static const templates *current_templates;
264
265 /* Per instruction expressionS buffers: 2 displacements & 2 immediate max. */
266 static expressionS disp_expressions[2], im_expressions[2];
267
268 /* Current operand we are working on. */
269 static int this_operand;
270
271 /* We support four different modes. FLAG_CODE variable is used to distinguish
272 these. */
273
274 enum flag_code {
275 CODE_32BIT,
276 CODE_16BIT,
277 CODE_64BIT };
278 #define NUM_FLAG_CODE ((int) CODE_64BIT + 1)
279
280 static enum flag_code flag_code;
281 static int use_rela_relocations = 0;
282
283 /* The names used to print error messages. */
284 static const char *flag_code_names[] =
285 {
286 "32",
287 "16",
288 "64"
289 };
290
291 /* 1 for intel syntax,
292 0 if att syntax. */
293 static int intel_syntax = 0;
294
295 /* 1 if register prefix % not required. */
296 static int allow_naked_reg = 0;
297
298 /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
299 leave, push, and pop instructions so that gcc has the same stack
300 frame as in 32 bit mode. */
301 static char stackop_size = '\0';
302
303 /* Non-zero to quieten some warnings. */
304 static int quiet_warnings = 0;
305
306 /* CPU name. */
307 static const char *cpu_arch_name = NULL;
308
309 /* CPU feature flags. */
310 static unsigned int cpu_arch_flags = CpuUnknownFlags | CpuNo64;
311
312 /* If set, conditional jumps are not automatically promoted to handle
313 larger than a byte offset. */
314 static unsigned int no_cond_jump_promotion = 0;
315
316 /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
317 symbolS *GOT_symbol;
318
319 /* Interface to relax_segment.
320 There are 3 major relax states for 386 jump insns because the
321 different types of jumps add different sizes to frags when we're
322 figuring out what sort of jump to choose to reach a given label. */
323
324 /* Types. */
325 #define UNCOND_JUMP 0
326 #define COND_JUMP 1
327 #define COND_JUMP86 2
328
329 /* Sizes. */
330 #define CODE16 1
331 #define SMALL 0
332 #define SMALL16 (SMALL | CODE16)
333 #define BIG 2
334 #define BIG16 (BIG | CODE16)
335
336 #ifndef INLINE
337 #ifdef __GNUC__
338 #define INLINE __inline__
339 #else
340 #define INLINE
341 #endif
342 #endif
343
344 #define ENCODE_RELAX_STATE(type, size) \
345 ((relax_substateT) (((type) << 2) | (size)))
346 #define TYPE_FROM_RELAX_STATE(s) \
347 ((s) >> 2)
348 #define DISP_SIZE_FROM_RELAX_STATE(s) \
349 ((((s) & 3) == BIG ? 4 : (((s) & 3) == BIG16 ? 2 : 1)))
350
351 /* This table is used by relax_frag to promote short jumps to long
352 ones where necessary. SMALL (short) jumps may be promoted to BIG
353 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
354 don't allow a short jump in a 32 bit code segment to be promoted to
355 a 16 bit offset jump because it's slower (requires data size
356 prefix), and doesn't work, unless the destination is in the bottom
357 64k of the code segment (The top 16 bits of eip are zeroed). */
358
359 const relax_typeS md_relax_table[] =
360 {
361 /* The fields are:
362 1) most positive reach of this state,
363 2) most negative reach of this state,
364 3) how many bytes this mode will have in the variable part of the frag
365 4) which index into the table to try if we can't fit into this one. */
366
367 /* UNCOND_JUMP states. */
368 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
369 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
370 /* dword jmp adds 4 bytes to frag:
371 0 extra opcode bytes, 4 displacement bytes. */
372 {0, 0, 4, 0},
373 /* word jmp adds 2 byte2 to frag:
374 0 extra opcode bytes, 2 displacement bytes. */
375 {0, 0, 2, 0},
376
377 /* COND_JUMP states. */
378 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
379 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
380 /* dword conditionals adds 5 bytes to frag:
381 1 extra opcode byte, 4 displacement bytes. */
382 {0, 0, 5, 0},
383 /* word conditionals add 3 bytes to frag:
384 1 extra opcode byte, 2 displacement bytes. */
385 {0, 0, 3, 0},
386
387 /* COND_JUMP86 states. */
388 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG)},
389 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG16)},
390 /* dword conditionals adds 5 bytes to frag:
391 1 extra opcode byte, 4 displacement bytes. */
392 {0, 0, 5, 0},
393 /* word conditionals add 4 bytes to frag:
394 1 displacement byte and a 3 byte long branch insn. */
395 {0, 0, 4, 0}
396 };
397
398 static const arch_entry cpu_arch[] = {
399 {"i8086", Cpu086 },
400 {"i186", Cpu086|Cpu186 },
401 {"i286", Cpu086|Cpu186|Cpu286 },
402 {"i386", Cpu086|Cpu186|Cpu286|Cpu386 },
403 {"i486", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486 },
404 {"i586", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
405 {"i686", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
406 {"pentium", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
407 {"pentiumpro",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
408 {"pentium4", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX|CpuSSE|CpuSSE2 },
409 {"k6", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow },
410 {"athlon", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuMMX|Cpu3dnow },
411 {"sledgehammer",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuSledgehammer|CpuMMX|Cpu3dnow|CpuSSE|CpuSSE2 },
412 {NULL, 0 }
413 };
414
415 const pseudo_typeS md_pseudo_table[] =
416 {
417 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
418 {"align", s_align_bytes, 0},
419 #else
420 {"align", s_align_ptwo, 0},
421 #endif
422 {"arch", set_cpu_arch, 0},
423 #ifndef I386COFF
424 {"bss", s_bss, 0},
425 #endif
426 {"ffloat", float_cons, 'f'},
427 {"dfloat", float_cons, 'd'},
428 {"tfloat", float_cons, 'x'},
429 {"value", cons, 2},
430 {"noopt", s_ignore, 0},
431 {"optim", s_ignore, 0},
432 {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
433 {"code16", set_code_flag, CODE_16BIT},
434 {"code32", set_code_flag, CODE_32BIT},
435 {"code64", set_code_flag, CODE_64BIT},
436 {"intel_syntax", set_intel_syntax, 1},
437 {"att_syntax", set_intel_syntax, 0},
438 {"file", dwarf2_directive_file, 0},
439 {"loc", dwarf2_directive_loc, 0},
440 {0, 0, 0}
441 };
442
443 /* For interface with expression (). */
444 extern char *input_line_pointer;
445
446 /* Hash table for instruction mnemonic lookup. */
447 static struct hash_control *op_hash;
448
449 /* Hash table for register lookup. */
450 static struct hash_control *reg_hash;
451 \f
452 void
453 i386_align_code (fragP, count)
454 fragS *fragP;
455 int count;
456 {
457 /* Various efficient no-op patterns for aligning code labels.
458 Note: Don't try to assemble the instructions in the comments.
459 0L and 0w are not legal. */
460 static const char f32_1[] =
461 {0x90}; /* nop */
462 static const char f32_2[] =
463 {0x89,0xf6}; /* movl %esi,%esi */
464 static const char f32_3[] =
465 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
466 static const char f32_4[] =
467 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
468 static const char f32_5[] =
469 {0x90, /* nop */
470 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
471 static const char f32_6[] =
472 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
473 static const char f32_7[] =
474 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
475 static const char f32_8[] =
476 {0x90, /* nop */
477 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
478 static const char f32_9[] =
479 {0x89,0xf6, /* movl %esi,%esi */
480 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
481 static const char f32_10[] =
482 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
483 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
484 static const char f32_11[] =
485 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
486 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
487 static const char f32_12[] =
488 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
489 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
490 static const char f32_13[] =
491 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
492 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
493 static const char f32_14[] =
494 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
495 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
496 static const char f32_15[] =
497 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
498 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
499 static const char f16_3[] =
500 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
501 static const char f16_4[] =
502 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
503 static const char f16_5[] =
504 {0x90, /* nop */
505 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
506 static const char f16_6[] =
507 {0x89,0xf6, /* mov %si,%si */
508 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
509 static const char f16_7[] =
510 {0x8d,0x74,0x00, /* lea 0(%si),%si */
511 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
512 static const char f16_8[] =
513 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
514 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
515 static const char *const f32_patt[] = {
516 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
517 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
518 };
519 static const char *const f16_patt[] = {
520 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
521 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
522 };
523
524 /* ??? We can't use these fillers for x86_64, since they often kills the
525 upper halves. Solve later. */
526 if (flag_code == CODE_64BIT)
527 count = 1;
528
529 if (count > 0 && count <= 15)
530 {
531 if (flag_code == CODE_16BIT)
532 {
533 memcpy (fragP->fr_literal + fragP->fr_fix,
534 f16_patt[count - 1], count);
535 if (count > 8)
536 /* Adjust jump offset. */
537 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
538 }
539 else
540 memcpy (fragP->fr_literal + fragP->fr_fix,
541 f32_patt[count - 1], count);
542 fragP->fr_var = count;
543 }
544 }
545
546 static INLINE unsigned int
547 mode_from_disp_size (t)
548 unsigned int t;
549 {
550 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
551 }
552
553 static INLINE int
554 fits_in_signed_byte (num)
555 offsetT num;
556 {
557 return (num >= -128) && (num <= 127);
558 }
559
560 static INLINE int
561 fits_in_unsigned_byte (num)
562 offsetT num;
563 {
564 return (num & 0xff) == num;
565 }
566
567 static INLINE int
568 fits_in_unsigned_word (num)
569 offsetT num;
570 {
571 return (num & 0xffff) == num;
572 }
573
574 static INLINE int
575 fits_in_signed_word (num)
576 offsetT num;
577 {
578 return (-32768 <= num) && (num <= 32767);
579 }
580 static INLINE int
581 fits_in_signed_long (num)
582 offsetT num ATTRIBUTE_UNUSED;
583 {
584 #ifndef BFD64
585 return 1;
586 #else
587 return (!(((offsetT) -1 << 31) & num)
588 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
589 #endif
590 } /* fits_in_signed_long() */
591 static INLINE int
592 fits_in_unsigned_long (num)
593 offsetT num ATTRIBUTE_UNUSED;
594 {
595 #ifndef BFD64
596 return 1;
597 #else
598 return (num & (((offsetT) 2 << 31) - 1)) == num;
599 #endif
600 } /* fits_in_unsigned_long() */
601
602 static int
603 smallest_imm_type (num)
604 offsetT num;
605 {
606 if (cpu_arch_flags != (Cpu086 | Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64))
607 {
608 /* This code is disabled on the 486 because all the Imm1 forms
609 in the opcode table are slower on the i486. They're the
610 versions with the implicitly specified single-position
611 displacement, which has another syntax if you really want to
612 use that form. */
613 if (num == 1)
614 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
615 }
616 return (fits_in_signed_byte (num)
617 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
618 : fits_in_unsigned_byte (num)
619 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
620 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
621 ? (Imm16 | Imm32 | Imm32S | Imm64)
622 : fits_in_signed_long (num)
623 ? (Imm32 | Imm32S | Imm64)
624 : fits_in_unsigned_long (num)
625 ? (Imm32 | Imm64)
626 : Imm64);
627 }
628
629 static offsetT
630 offset_in_range (val, size)
631 offsetT val;
632 int size;
633 {
634 addressT mask;
635
636 switch (size)
637 {
638 case 1: mask = ((addressT) 1 << 8) - 1; break;
639 case 2: mask = ((addressT) 1 << 16) - 1; break;
640 case 4: mask = ((addressT) 2 << 31) - 1; break;
641 #ifdef BFD64
642 case 8: mask = ((addressT) 2 << 63) - 1; break;
643 #endif
644 default: abort ();
645 }
646
647 /* If BFD64, sign extend val. */
648 if (!use_rela_relocations)
649 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
650 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
651
652 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
653 {
654 char buf1[40], buf2[40];
655
656 sprint_value (buf1, val);
657 sprint_value (buf2, val & mask);
658 as_warn (_("%s shortened to %s"), buf1, buf2);
659 }
660 return val & mask;
661 }
662
663 /* Returns 0 if attempting to add a prefix where one from the same
664 class already exists, 1 if non rep/repne added, 2 if rep/repne
665 added. */
666 static int
667 add_prefix (prefix)
668 unsigned int prefix;
669 {
670 int ret = 1;
671 int q;
672
673 if (prefix >= REX_OPCODE && prefix < REX_OPCODE + 16
674 && flag_code == CODE_64BIT)
675 q = REX_PREFIX;
676 else
677 switch (prefix)
678 {
679 default:
680 abort ();
681
682 case CS_PREFIX_OPCODE:
683 case DS_PREFIX_OPCODE:
684 case ES_PREFIX_OPCODE:
685 case FS_PREFIX_OPCODE:
686 case GS_PREFIX_OPCODE:
687 case SS_PREFIX_OPCODE:
688 q = SEG_PREFIX;
689 break;
690
691 case REPNE_PREFIX_OPCODE:
692 case REPE_PREFIX_OPCODE:
693 ret = 2;
694 /* fall thru */
695 case LOCK_PREFIX_OPCODE:
696 q = LOCKREP_PREFIX;
697 break;
698
699 case FWAIT_OPCODE:
700 q = WAIT_PREFIX;
701 break;
702
703 case ADDR_PREFIX_OPCODE:
704 q = ADDR_PREFIX;
705 break;
706
707 case DATA_PREFIX_OPCODE:
708 q = DATA_PREFIX;
709 break;
710 }
711
712 if (i.prefix[q] != 0)
713 {
714 as_bad (_("same type of prefix used twice"));
715 return 0;
716 }
717
718 i.prefixes += 1;
719 i.prefix[q] = prefix;
720 return ret;
721 }
722
723 static void
724 set_code_flag (value)
725 int value;
726 {
727 flag_code = value;
728 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
729 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
730 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
731 {
732 as_bad (_("64bit mode not supported on this CPU."));
733 }
734 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
735 {
736 as_bad (_("32bit mode not supported on this CPU."));
737 }
738 stackop_size = '\0';
739 }
740
741 static void
742 set_16bit_gcc_code_flag (new_code_flag)
743 int new_code_flag;
744 {
745 flag_code = new_code_flag;
746 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
747 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
748 stackop_size = 'l';
749 }
750
751 static void
752 set_intel_syntax (syntax_flag)
753 int syntax_flag;
754 {
755 /* Find out if register prefixing is specified. */
756 int ask_naked_reg = 0;
757
758 SKIP_WHITESPACE ();
759 if (!is_end_of_line[(unsigned char) *input_line_pointer])
760 {
761 char *string = input_line_pointer;
762 int e = get_symbol_end ();
763
764 if (strcmp (string, "prefix") == 0)
765 ask_naked_reg = 1;
766 else if (strcmp (string, "noprefix") == 0)
767 ask_naked_reg = -1;
768 else
769 as_bad (_("bad argument to syntax directive."));
770 *input_line_pointer = e;
771 }
772 demand_empty_rest_of_line ();
773
774 intel_syntax = syntax_flag;
775
776 if (ask_naked_reg == 0)
777 {
778 #ifdef BFD_ASSEMBLER
779 allow_naked_reg = (intel_syntax
780 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
781 #else
782 /* Conservative default. */
783 allow_naked_reg = 0;
784 #endif
785 }
786 else
787 allow_naked_reg = (ask_naked_reg < 0);
788 }
789
790 static void
791 set_cpu_arch (dummy)
792 int dummy ATTRIBUTE_UNUSED;
793 {
794 SKIP_WHITESPACE ();
795
796 if (!is_end_of_line[(unsigned char) *input_line_pointer])
797 {
798 char *string = input_line_pointer;
799 int e = get_symbol_end ();
800 int i;
801
802 for (i = 0; cpu_arch[i].name; i++)
803 {
804 if (strcmp (string, cpu_arch[i].name) == 0)
805 {
806 cpu_arch_name = cpu_arch[i].name;
807 cpu_arch_flags = (cpu_arch[i].flags
808 | (flag_code == CODE_64BIT ? Cpu64 : CpuNo64));
809 break;
810 }
811 }
812 if (!cpu_arch[i].name)
813 as_bad (_("no such architecture: `%s'"), string);
814
815 *input_line_pointer = e;
816 }
817 else
818 as_bad (_("missing cpu architecture"));
819
820 no_cond_jump_promotion = 0;
821 if (*input_line_pointer == ','
822 && !is_end_of_line[(unsigned char) input_line_pointer[1]])
823 {
824 char *string = ++input_line_pointer;
825 int e = get_symbol_end ();
826
827 if (strcmp (string, "nojumps") == 0)
828 no_cond_jump_promotion = 1;
829 else if (strcmp (string, "jumps") == 0)
830 ;
831 else
832 as_bad (_("no such architecture modifier: `%s'"), string);
833
834 *input_line_pointer = e;
835 }
836
837 demand_empty_rest_of_line ();
838 }
839
840 #ifdef BFD_ASSEMBLER
841 unsigned long
842 i386_mach ()
843 {
844 if (!strcmp (default_arch, "x86_64"))
845 return bfd_mach_x86_64;
846 else if (!strcmp (default_arch, "i386"))
847 return bfd_mach_i386_i386;
848 else
849 as_fatal (_("Unknown architecture"));
850 }
851 #endif
852 \f
853 void
854 md_begin ()
855 {
856 const char *hash_err;
857
858 /* Initialize op_hash hash table. */
859 op_hash = hash_new ();
860
861 {
862 const template *optab;
863 templates *core_optab;
864
865 /* Setup for loop. */
866 optab = i386_optab;
867 core_optab = (templates *) xmalloc (sizeof (templates));
868 core_optab->start = optab;
869
870 while (1)
871 {
872 ++optab;
873 if (optab->name == NULL
874 || strcmp (optab->name, (optab - 1)->name) != 0)
875 {
876 /* different name --> ship out current template list;
877 add to hash table; & begin anew. */
878 core_optab->end = optab;
879 hash_err = hash_insert (op_hash,
880 (optab - 1)->name,
881 (PTR) core_optab);
882 if (hash_err)
883 {
884 as_fatal (_("Internal Error: Can't hash %s: %s"),
885 (optab - 1)->name,
886 hash_err);
887 }
888 if (optab->name == NULL)
889 break;
890 core_optab = (templates *) xmalloc (sizeof (templates));
891 core_optab->start = optab;
892 }
893 }
894 }
895
896 /* Initialize reg_hash hash table. */
897 reg_hash = hash_new ();
898 {
899 const reg_entry *regtab;
900
901 for (regtab = i386_regtab;
902 regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
903 regtab++)
904 {
905 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
906 if (hash_err)
907 as_fatal (_("Internal Error: Can't hash %s: %s"),
908 regtab->reg_name,
909 hash_err);
910 }
911 }
912
913 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
914 {
915 int c;
916 char *p;
917
918 for (c = 0; c < 256; c++)
919 {
920 if (ISDIGIT (c))
921 {
922 digit_chars[c] = c;
923 mnemonic_chars[c] = c;
924 register_chars[c] = c;
925 operand_chars[c] = c;
926 }
927 else if (ISLOWER (c))
928 {
929 mnemonic_chars[c] = c;
930 register_chars[c] = c;
931 operand_chars[c] = c;
932 }
933 else if (ISUPPER (c))
934 {
935 mnemonic_chars[c] = TOLOWER (c);
936 register_chars[c] = mnemonic_chars[c];
937 operand_chars[c] = c;
938 }
939
940 if (ISALPHA (c) || ISDIGIT (c))
941 identifier_chars[c] = c;
942 else if (c >= 128)
943 {
944 identifier_chars[c] = c;
945 operand_chars[c] = c;
946 }
947 }
948
949 #ifdef LEX_AT
950 identifier_chars['@'] = '@';
951 #endif
952 digit_chars['-'] = '-';
953 identifier_chars['_'] = '_';
954 identifier_chars['.'] = '.';
955
956 for (p = operand_special_chars; *p != '\0'; p++)
957 operand_chars[(unsigned char) *p] = *p;
958 }
959
960 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
961 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
962 {
963 record_alignment (text_section, 2);
964 record_alignment (data_section, 2);
965 record_alignment (bss_section, 2);
966 }
967 #endif
968 }
969
970 void
971 i386_print_statistics (file)
972 FILE *file;
973 {
974 hash_print_statistics (file, "i386 opcode", op_hash);
975 hash_print_statistics (file, "i386 register", reg_hash);
976 }
977 \f
978 #ifdef DEBUG386
979
980 /* Debugging routines for md_assemble. */
981 static void pi PARAMS ((char *, i386_insn *));
982 static void pte PARAMS ((template *));
983 static void pt PARAMS ((unsigned int));
984 static void pe PARAMS ((expressionS *));
985 static void ps PARAMS ((symbolS *));
986
987 static void
988 pi (line, x)
989 char *line;
990 i386_insn *x;
991 {
992 unsigned int i;
993
994 fprintf (stdout, "%s: template ", line);
995 pte (&x->tm);
996 fprintf (stdout, " address: base %s index %s scale %x\n",
997 x->base_reg ? x->base_reg->reg_name : "none",
998 x->index_reg ? x->index_reg->reg_name : "none",
999 x->log2_scale_factor);
1000 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
1001 x->rm.mode, x->rm.reg, x->rm.regmem);
1002 fprintf (stdout, " sib: base %x index %x scale %x\n",
1003 x->sib.base, x->sib.index, x->sib.scale);
1004 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
1005 (x->rex & REX_MODE64) != 0,
1006 (x->rex & REX_EXTX) != 0,
1007 (x->rex & REX_EXTY) != 0,
1008 (x->rex & REX_EXTZ) != 0);
1009 for (i = 0; i < x->operands; i++)
1010 {
1011 fprintf (stdout, " #%d: ", i + 1);
1012 pt (x->types[i]);
1013 fprintf (stdout, "\n");
1014 if (x->types[i]
1015 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
1016 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
1017 if (x->types[i] & Imm)
1018 pe (x->op[i].imms);
1019 if (x->types[i] & Disp)
1020 pe (x->op[i].disps);
1021 }
1022 }
1023
1024 static void
1025 pte (t)
1026 template *t;
1027 {
1028 unsigned int i;
1029 fprintf (stdout, " %d operands ", t->operands);
1030 fprintf (stdout, "opcode %x ", t->base_opcode);
1031 if (t->extension_opcode != None)
1032 fprintf (stdout, "ext %x ", t->extension_opcode);
1033 if (t->opcode_modifier & D)
1034 fprintf (stdout, "D");
1035 if (t->opcode_modifier & W)
1036 fprintf (stdout, "W");
1037 fprintf (stdout, "\n");
1038 for (i = 0; i < t->operands; i++)
1039 {
1040 fprintf (stdout, " #%d type ", i + 1);
1041 pt (t->operand_types[i]);
1042 fprintf (stdout, "\n");
1043 }
1044 }
1045
1046 static void
1047 pe (e)
1048 expressionS *e;
1049 {
1050 fprintf (stdout, " operation %d\n", e->X_op);
1051 fprintf (stdout, " add_number %ld (%lx)\n",
1052 (long) e->X_add_number, (long) e->X_add_number);
1053 if (e->X_add_symbol)
1054 {
1055 fprintf (stdout, " add_symbol ");
1056 ps (e->X_add_symbol);
1057 fprintf (stdout, "\n");
1058 }
1059 if (e->X_op_symbol)
1060 {
1061 fprintf (stdout, " op_symbol ");
1062 ps (e->X_op_symbol);
1063 fprintf (stdout, "\n");
1064 }
1065 }
1066
1067 static void
1068 ps (s)
1069 symbolS *s;
1070 {
1071 fprintf (stdout, "%s type %s%s",
1072 S_GET_NAME (s),
1073 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
1074 segment_name (S_GET_SEGMENT (s)));
1075 }
1076
1077 struct type_name
1078 {
1079 unsigned int mask;
1080 char *tname;
1081 }
1082
1083 static const type_names[] =
1084 {
1085 { Reg8, "r8" },
1086 { Reg16, "r16" },
1087 { Reg32, "r32" },
1088 { Reg64, "r64" },
1089 { Imm8, "i8" },
1090 { Imm8S, "i8s" },
1091 { Imm16, "i16" },
1092 { Imm32, "i32" },
1093 { Imm32S, "i32s" },
1094 { Imm64, "i64" },
1095 { Imm1, "i1" },
1096 { BaseIndex, "BaseIndex" },
1097 { Disp8, "d8" },
1098 { Disp16, "d16" },
1099 { Disp32, "d32" },
1100 { Disp32S, "d32s" },
1101 { Disp64, "d64" },
1102 { InOutPortReg, "InOutPortReg" },
1103 { ShiftCount, "ShiftCount" },
1104 { Control, "control reg" },
1105 { Test, "test reg" },
1106 { Debug, "debug reg" },
1107 { FloatReg, "FReg" },
1108 { FloatAcc, "FAcc" },
1109 { SReg2, "SReg2" },
1110 { SReg3, "SReg3" },
1111 { Acc, "Acc" },
1112 { JumpAbsolute, "Jump Absolute" },
1113 { RegMMX, "rMMX" },
1114 { RegXMM, "rXMM" },
1115 { EsSeg, "es" },
1116 { 0, "" }
1117 };
1118
1119 static void
1120 pt (t)
1121 unsigned int t;
1122 {
1123 const struct type_name *ty;
1124
1125 for (ty = type_names; ty->mask; ty++)
1126 if (t & ty->mask)
1127 fprintf (stdout, "%s, ", ty->tname);
1128 fflush (stdout);
1129 }
1130
1131 #endif /* DEBUG386 */
1132 \f
1133 int
1134 tc_i386_force_relocation (fixp)
1135 struct fix *fixp;
1136 {
1137 #ifdef BFD_ASSEMBLER
1138 if (fixp->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1139 || fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1140 return 1;
1141 return 0;
1142 #else
1143 /* For COFF. */
1144 return fixp->fx_r_type == 7;
1145 #endif
1146 }
1147
1148 #ifdef BFD_ASSEMBLER
1149 static bfd_reloc_code_real_type reloc
1150 PARAMS ((int, int, int, bfd_reloc_code_real_type));
1151
1152 static bfd_reloc_code_real_type
1153 reloc (size, pcrel, sign, other)
1154 int size;
1155 int pcrel;
1156 int sign;
1157 bfd_reloc_code_real_type other;
1158 {
1159 if (other != NO_RELOC)
1160 return other;
1161
1162 if (pcrel)
1163 {
1164 if (!sign)
1165 as_bad (_("There are no unsigned pc-relative relocations"));
1166 switch (size)
1167 {
1168 case 1: return BFD_RELOC_8_PCREL;
1169 case 2: return BFD_RELOC_16_PCREL;
1170 case 4: return BFD_RELOC_32_PCREL;
1171 }
1172 as_bad (_("can not do %d byte pc-relative relocation"), size);
1173 }
1174 else
1175 {
1176 if (sign)
1177 switch (size)
1178 {
1179 case 4: return BFD_RELOC_X86_64_32S;
1180 }
1181 else
1182 switch (size)
1183 {
1184 case 1: return BFD_RELOC_8;
1185 case 2: return BFD_RELOC_16;
1186 case 4: return BFD_RELOC_32;
1187 case 8: return BFD_RELOC_64;
1188 }
1189 as_bad (_("can not do %s %d byte relocation"),
1190 sign ? "signed" : "unsigned", size);
1191 }
1192
1193 abort ();
1194 return BFD_RELOC_NONE;
1195 }
1196
1197 /* Here we decide which fixups can be adjusted to make them relative to
1198 the beginning of the section instead of the symbol. Basically we need
1199 to make sure that the dynamic relocations are done correctly, so in
1200 some cases we force the original symbol to be used. */
1201
1202 int
1203 tc_i386_fix_adjustable (fixP)
1204 fixS *fixP;
1205 {
1206 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1207 /* Prevent all adjustments to global symbols, or else dynamic
1208 linking will not work correctly. */
1209 if (S_IS_EXTERNAL (fixP->fx_addsy)
1210 || S_IS_WEAK (fixP->fx_addsy)
1211 /* Don't adjust pc-relative references to merge sections in 64-bit
1212 mode. */
1213 || (use_rela_relocations
1214 && (S_GET_SEGMENT (fixP->fx_addsy)->flags & SEC_MERGE) != 0
1215 && fixP->fx_pcrel))
1216 return 0;
1217 #endif
1218 /* adjust_reloc_syms doesn't know about the GOT. */
1219 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1220 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1221 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1222 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1223 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1224 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
1225 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1226 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1227 return 0;
1228 return 1;
1229 }
1230 #else
1231 #define reloc(SIZE,PCREL,SIGN,OTHER) 0
1232 #define BFD_RELOC_16 0
1233 #define BFD_RELOC_32 0
1234 #define BFD_RELOC_16_PCREL 0
1235 #define BFD_RELOC_32_PCREL 0
1236 #define BFD_RELOC_386_PLT32 0
1237 #define BFD_RELOC_386_GOT32 0
1238 #define BFD_RELOC_386_GOTOFF 0
1239 #define BFD_RELOC_X86_64_PLT32 0
1240 #define BFD_RELOC_X86_64_GOT32 0
1241 #define BFD_RELOC_X86_64_GOTPCREL 0
1242 #endif
1243
1244 static int intel_float_operand PARAMS ((const char *mnemonic));
1245
1246 static int
1247 intel_float_operand (mnemonic)
1248 const char *mnemonic;
1249 {
1250 if (mnemonic[0] == 'f' && mnemonic[1] == 'i')
1251 return 2;
1252
1253 if (mnemonic[0] == 'f')
1254 return 1;
1255
1256 return 0;
1257 }
1258
1259 /* This is the guts of the machine-dependent assembler. LINE points to a
1260 machine dependent instruction. This function is supposed to emit
1261 the frags/bytes it assembles to. */
1262
1263 void
1264 md_assemble (line)
1265 char *line;
1266 {
1267 int j;
1268 char mnemonic[MAX_MNEM_SIZE];
1269
1270 /* Initialize globals. */
1271 memset (&i, '\0', sizeof (i));
1272 for (j = 0; j < MAX_OPERANDS; j++)
1273 i.reloc[j] = NO_RELOC;
1274 memset (disp_expressions, '\0', sizeof (disp_expressions));
1275 memset (im_expressions, '\0', sizeof (im_expressions));
1276 save_stack_p = save_stack;
1277
1278 /* First parse an instruction mnemonic & call i386_operand for the operands.
1279 We assume that the scrubber has arranged it so that line[0] is the valid
1280 start of a (possibly prefixed) mnemonic. */
1281
1282 line = parse_insn (line, mnemonic);
1283 if (line == NULL)
1284 return;
1285
1286 line = parse_operands (line, mnemonic);
1287 if (line == NULL)
1288 return;
1289
1290 /* Now we've parsed the mnemonic into a set of templates, and have the
1291 operands at hand. */
1292
1293 /* All intel opcodes have reversed operands except for "bound" and
1294 "enter". We also don't reverse intersegment "jmp" and "call"
1295 instructions with 2 immediate operands so that the immediate segment
1296 precedes the offset, as it does when in AT&T mode. "enter" and the
1297 intersegment "jmp" and "call" instructions are the only ones that
1298 have two immediate operands. */
1299 if (intel_syntax && i.operands > 1
1300 && (strcmp (mnemonic, "bound") != 0)
1301 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1302 swap_operands ();
1303
1304 if (i.imm_operands)
1305 optimize_imm ();
1306
1307 if (i.disp_operands)
1308 optimize_disp ();
1309
1310 /* Next, we find a template that matches the given insn,
1311 making sure the overlap of the given operands types is consistent
1312 with the template operand types. */
1313
1314 if (!match_template ())
1315 return;
1316
1317 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1318 if (SYSV386_COMPAT
1319 && intel_syntax
1320 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1321 i.tm.base_opcode ^= FloatR;
1322
1323 if (i.tm.opcode_modifier & FWait)
1324 if (!add_prefix (FWAIT_OPCODE))
1325 return;
1326
1327 /* Check string instruction segment overrides. */
1328 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1329 {
1330 if (!check_string ())
1331 return;
1332 }
1333
1334 if (!process_suffix ())
1335 return;
1336
1337 /* Make still unresolved immediate matches conform to size of immediate
1338 given in i.suffix. */
1339 if (!finalize_imm ())
1340 return;
1341
1342 if (i.types[0] & Imm1)
1343 i.imm_operands = 0; /* kludge for shift insns. */
1344 if (i.types[0] & ImplicitRegister)
1345 i.reg_operands--;
1346 if (i.types[1] & ImplicitRegister)
1347 i.reg_operands--;
1348 if (i.types[2] & ImplicitRegister)
1349 i.reg_operands--;
1350
1351 if (i.tm.opcode_modifier & ImmExt)
1352 {
1353 /* These AMD 3DNow! and Intel Katmai New Instructions have an
1354 opcode suffix which is coded in the same place as an 8-bit
1355 immediate field would be. Here we fake an 8-bit immediate
1356 operand from the opcode suffix stored in tm.extension_opcode. */
1357
1358 expressionS *exp;
1359
1360 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
1361
1362 exp = &im_expressions[i.imm_operands++];
1363 i.op[i.operands].imms = exp;
1364 i.types[i.operands++] = Imm8;
1365 exp->X_op = O_constant;
1366 exp->X_add_number = i.tm.extension_opcode;
1367 i.tm.extension_opcode = None;
1368 }
1369
1370 /* For insns with operands there are more diddles to do to the opcode. */
1371 if (i.operands)
1372 {
1373 if (!process_operands ())
1374 return;
1375 }
1376 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
1377 {
1378 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
1379 as_warn (_("translating to `%sp'"), i.tm.name);
1380 }
1381
1382 /* Handle conversion of 'int $3' --> special int3 insn. */
1383 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
1384 {
1385 i.tm.base_opcode = INT3_OPCODE;
1386 i.imm_operands = 0;
1387 }
1388
1389 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
1390 && i.op[0].disps->X_op == O_constant)
1391 {
1392 /* Convert "jmp constant" (and "call constant") to a jump (call) to
1393 the absolute address given by the constant. Since ix86 jumps and
1394 calls are pc relative, we need to generate a reloc. */
1395 i.op[0].disps->X_add_symbol = &abs_symbol;
1396 i.op[0].disps->X_op = O_symbol;
1397 }
1398
1399 if ((i.tm.opcode_modifier & Rex64) != 0)
1400 i.rex |= REX_MODE64;
1401
1402 /* For 8 bit registers we need an empty rex prefix. Also if the
1403 instruction already has a prefix, we need to convert old
1404 registers to new ones. */
1405
1406 if (((i.types[0] & Reg8) != 0
1407 && (i.op[0].regs->reg_flags & RegRex64) != 0)
1408 || ((i.types[1] & Reg8) != 0
1409 && (i.op[1].regs->reg_flags & RegRex64) != 0)
1410 || (((i.types[0] & Reg8) != 0 || (i.types[1] & Reg8) != 0)
1411 && i.rex != 0))
1412 {
1413 int x;
1414
1415 i.rex |= REX_OPCODE;
1416 for (x = 0; x < 2; x++)
1417 {
1418 /* Look for 8 bit operand that uses old registers. */
1419 if ((i.types[x] & Reg8) != 0
1420 && (i.op[x].regs->reg_flags & RegRex64) == 0)
1421 {
1422 /* In case it is "hi" register, give up. */
1423 if (i.op[x].regs->reg_num > 3)
1424 as_bad (_("can't encode register '%%%s' in an instruction requiring REX prefix.\n"),
1425 i.op[x].regs->reg_name);
1426
1427 /* Otherwise it is equivalent to the extended register.
1428 Since the encoding doesn't change this is merely
1429 cosmetic cleanup for debug output. */
1430
1431 i.op[x].regs = i.op[x].regs + 8;
1432 }
1433 }
1434 }
1435
1436 if (i.rex != 0)
1437 add_prefix (REX_OPCODE | i.rex);
1438
1439 /* We are ready to output the insn. */
1440 output_insn ();
1441 }
1442
1443 static char *
1444 parse_insn (line, mnemonic)
1445 char *line;
1446 char *mnemonic;
1447 {
1448 char *l = line;
1449 char *token_start = l;
1450 char *mnem_p;
1451
1452 /* Non-zero if we found a prefix only acceptable with string insns. */
1453 const char *expecting_string_instruction = NULL;
1454
1455 while (1)
1456 {
1457 mnem_p = mnemonic;
1458 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1459 {
1460 mnem_p++;
1461 if (mnem_p >= mnemonic + MAX_MNEM_SIZE)
1462 {
1463 as_bad (_("no such instruction: `%s'"), token_start);
1464 return NULL;
1465 }
1466 l++;
1467 }
1468 if (!is_space_char (*l)
1469 && *l != END_OF_INSN
1470 && *l != PREFIX_SEPARATOR
1471 && *l != ',')
1472 {
1473 as_bad (_("invalid character %s in mnemonic"),
1474 output_invalid (*l));
1475 return NULL;
1476 }
1477 if (token_start == l)
1478 {
1479 if (*l == PREFIX_SEPARATOR)
1480 as_bad (_("expecting prefix; got nothing"));
1481 else
1482 as_bad (_("expecting mnemonic; got nothing"));
1483 return NULL;
1484 }
1485
1486 /* Look up instruction (or prefix) via hash table. */
1487 current_templates = hash_find (op_hash, mnemonic);
1488
1489 if (*l != END_OF_INSN
1490 && (!is_space_char (*l) || l[1] != END_OF_INSN)
1491 && current_templates
1492 && (current_templates->start->opcode_modifier & IsPrefix))
1493 {
1494 /* If we are in 16-bit mode, do not allow addr16 or data16.
1495 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1496 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1497 && flag_code != CODE_64BIT
1498 && (((current_templates->start->opcode_modifier & Size32) != 0)
1499 ^ (flag_code == CODE_16BIT)))
1500 {
1501 as_bad (_("redundant %s prefix"),
1502 current_templates->start->name);
1503 return NULL;
1504 }
1505 /* Add prefix, checking for repeated prefixes. */
1506 switch (add_prefix (current_templates->start->base_opcode))
1507 {
1508 case 0:
1509 return NULL;
1510 case 2:
1511 expecting_string_instruction = current_templates->start->name;
1512 break;
1513 }
1514 /* Skip past PREFIX_SEPARATOR and reset token_start. */
1515 token_start = ++l;
1516 }
1517 else
1518 break;
1519 }
1520
1521 if (!current_templates)
1522 {
1523 /* See if we can get a match by trimming off a suffix. */
1524 switch (mnem_p[-1])
1525 {
1526 case WORD_MNEM_SUFFIX:
1527 case BYTE_MNEM_SUFFIX:
1528 case QWORD_MNEM_SUFFIX:
1529 i.suffix = mnem_p[-1];
1530 mnem_p[-1] = '\0';
1531 current_templates = hash_find (op_hash, mnemonic);
1532 break;
1533 case SHORT_MNEM_SUFFIX:
1534 case LONG_MNEM_SUFFIX:
1535 if (!intel_syntax)
1536 {
1537 i.suffix = mnem_p[-1];
1538 mnem_p[-1] = '\0';
1539 current_templates = hash_find (op_hash, mnemonic);
1540 }
1541 break;
1542
1543 /* Intel Syntax. */
1544 case 'd':
1545 if (intel_syntax)
1546 {
1547 if (intel_float_operand (mnemonic))
1548 i.suffix = SHORT_MNEM_SUFFIX;
1549 else
1550 i.suffix = LONG_MNEM_SUFFIX;
1551 mnem_p[-1] = '\0';
1552 current_templates = hash_find (op_hash, mnemonic);
1553 }
1554 break;
1555 }
1556 if (!current_templates)
1557 {
1558 as_bad (_("no such instruction: `%s'"), token_start);
1559 return NULL;
1560 }
1561 }
1562
1563 if (current_templates->start->opcode_modifier & (Jump | JumpByte))
1564 {
1565 /* Check for a branch hint. We allow ",pt" and ",pn" for
1566 predict taken and predict not taken respectively.
1567 I'm not sure that branch hints actually do anything on loop
1568 and jcxz insns (JumpByte) for current Pentium4 chips. They
1569 may work in the future and it doesn't hurt to accept them
1570 now. */
1571 if (l[0] == ',' && l[1] == 'p')
1572 {
1573 if (l[2] == 't')
1574 {
1575 if (!add_prefix (DS_PREFIX_OPCODE))
1576 return NULL;
1577 l += 3;
1578 }
1579 else if (l[2] == 'n')
1580 {
1581 if (!add_prefix (CS_PREFIX_OPCODE))
1582 return NULL;
1583 l += 3;
1584 }
1585 }
1586 }
1587 /* Any other comma loses. */
1588 if (*l == ',')
1589 {
1590 as_bad (_("invalid character %s in mnemonic"),
1591 output_invalid (*l));
1592 return NULL;
1593 }
1594
1595 /* Check if instruction is supported on specified architecture. */
1596 if ((current_templates->start->cpu_flags & ~(Cpu64 | CpuNo64))
1597 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64)))
1598 {
1599 as_warn (_("`%s' is not supported on `%s'"),
1600 current_templates->start->name, cpu_arch_name);
1601 }
1602 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
1603 {
1604 as_warn (_("use .code16 to ensure correct addressing mode"));
1605 }
1606
1607 /* Check for rep/repne without a string instruction. */
1608 if (expecting_string_instruction
1609 && !(current_templates->start->opcode_modifier & IsString))
1610 {
1611 as_bad (_("expecting string instruction after `%s'"),
1612 expecting_string_instruction);
1613 return NULL;
1614 }
1615
1616 return l;
1617 }
1618
1619 static char *
1620 parse_operands (l, mnemonic)
1621 char *l;
1622 const char *mnemonic;
1623 {
1624 char *token_start;
1625
1626 /* 1 if operand is pending after ','. */
1627 unsigned int expecting_operand = 0;
1628
1629 /* Non-zero if operand parens not balanced. */
1630 unsigned int paren_not_balanced;
1631
1632 while (*l != END_OF_INSN)
1633 {
1634 /* Skip optional white space before operand. */
1635 if (is_space_char (*l))
1636 ++l;
1637 if (!is_operand_char (*l) && *l != END_OF_INSN)
1638 {
1639 as_bad (_("invalid character %s before operand %d"),
1640 output_invalid (*l),
1641 i.operands + 1);
1642 return NULL;
1643 }
1644 token_start = l; /* after white space */
1645 paren_not_balanced = 0;
1646 while (paren_not_balanced || *l != ',')
1647 {
1648 if (*l == END_OF_INSN)
1649 {
1650 if (paren_not_balanced)
1651 {
1652 if (!intel_syntax)
1653 as_bad (_("unbalanced parenthesis in operand %d."),
1654 i.operands + 1);
1655 else
1656 as_bad (_("unbalanced brackets in operand %d."),
1657 i.operands + 1);
1658 return NULL;
1659 }
1660 else
1661 break; /* we are done */
1662 }
1663 else if (!is_operand_char (*l) && !is_space_char (*l))
1664 {
1665 as_bad (_("invalid character %s in operand %d"),
1666 output_invalid (*l),
1667 i.operands + 1);
1668 return NULL;
1669 }
1670 if (!intel_syntax)
1671 {
1672 if (*l == '(')
1673 ++paren_not_balanced;
1674 if (*l == ')')
1675 --paren_not_balanced;
1676 }
1677 else
1678 {
1679 if (*l == '[')
1680 ++paren_not_balanced;
1681 if (*l == ']')
1682 --paren_not_balanced;
1683 }
1684 l++;
1685 }
1686 if (l != token_start)
1687 { /* Yes, we've read in another operand. */
1688 unsigned int operand_ok;
1689 this_operand = i.operands++;
1690 if (i.operands > MAX_OPERANDS)
1691 {
1692 as_bad (_("spurious operands; (%d operands/instruction max)"),
1693 MAX_OPERANDS);
1694 return NULL;
1695 }
1696 /* Now parse operand adding info to 'i' as we go along. */
1697 END_STRING_AND_SAVE (l);
1698
1699 if (intel_syntax)
1700 operand_ok =
1701 i386_intel_operand (token_start,
1702 intel_float_operand (mnemonic));
1703 else
1704 operand_ok = i386_operand (token_start);
1705
1706 RESTORE_END_STRING (l);
1707 if (!operand_ok)
1708 return NULL;
1709 }
1710 else
1711 {
1712 if (expecting_operand)
1713 {
1714 expecting_operand_after_comma:
1715 as_bad (_("expecting operand after ','; got nothing"));
1716 return NULL;
1717 }
1718 if (*l == ',')
1719 {
1720 as_bad (_("expecting operand before ','; got nothing"));
1721 return NULL;
1722 }
1723 }
1724
1725 /* Now *l must be either ',' or END_OF_INSN. */
1726 if (*l == ',')
1727 {
1728 if (*++l == END_OF_INSN)
1729 {
1730 /* Just skip it, if it's \n complain. */
1731 goto expecting_operand_after_comma;
1732 }
1733 expecting_operand = 1;
1734 }
1735 }
1736 return l;
1737 }
1738
1739 static void
1740 swap_operands ()
1741 {
1742 union i386_op temp_op;
1743 unsigned int temp_type;
1744 RELOC_ENUM temp_reloc;
1745 int xchg1 = 0;
1746 int xchg2 = 0;
1747
1748 if (i.operands == 2)
1749 {
1750 xchg1 = 0;
1751 xchg2 = 1;
1752 }
1753 else if (i.operands == 3)
1754 {
1755 xchg1 = 0;
1756 xchg2 = 2;
1757 }
1758 temp_type = i.types[xchg2];
1759 i.types[xchg2] = i.types[xchg1];
1760 i.types[xchg1] = temp_type;
1761 temp_op = i.op[xchg2];
1762 i.op[xchg2] = i.op[xchg1];
1763 i.op[xchg1] = temp_op;
1764 temp_reloc = i.reloc[xchg2];
1765 i.reloc[xchg2] = i.reloc[xchg1];
1766 i.reloc[xchg1] = temp_reloc;
1767
1768 if (i.mem_operands == 2)
1769 {
1770 const seg_entry *temp_seg;
1771 temp_seg = i.seg[0];
1772 i.seg[0] = i.seg[1];
1773 i.seg[1] = temp_seg;
1774 }
1775 }
1776
1777 /* Try to ensure constant immediates are represented in the smallest
1778 opcode possible. */
1779 static void
1780 optimize_imm ()
1781 {
1782 char guess_suffix = 0;
1783 int op;
1784
1785 if (i.suffix)
1786 guess_suffix = i.suffix;
1787 else if (i.reg_operands)
1788 {
1789 /* Figure out a suffix from the last register operand specified.
1790 We can't do this properly yet, ie. excluding InOutPortReg,
1791 but the following works for instructions with immediates.
1792 In any case, we can't set i.suffix yet. */
1793 for (op = i.operands; --op >= 0;)
1794 if (i.types[op] & Reg)
1795 {
1796 if (i.types[op] & Reg8)
1797 guess_suffix = BYTE_MNEM_SUFFIX;
1798 else if (i.types[op] & Reg16)
1799 guess_suffix = WORD_MNEM_SUFFIX;
1800 else if (i.types[op] & Reg32)
1801 guess_suffix = LONG_MNEM_SUFFIX;
1802 else if (i.types[op] & Reg64)
1803 guess_suffix = QWORD_MNEM_SUFFIX;
1804 break;
1805 }
1806 }
1807 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
1808 guess_suffix = WORD_MNEM_SUFFIX;
1809
1810 for (op = i.operands; --op >= 0;)
1811 if (i.types[op] & Imm)
1812 {
1813 switch (i.op[op].imms->X_op)
1814 {
1815 case O_constant:
1816 /* If a suffix is given, this operand may be shortened. */
1817 switch (guess_suffix)
1818 {
1819 case LONG_MNEM_SUFFIX:
1820 i.types[op] |= Imm32 | Imm64;
1821 break;
1822 case WORD_MNEM_SUFFIX:
1823 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
1824 break;
1825 case BYTE_MNEM_SUFFIX:
1826 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
1827 break;
1828 }
1829
1830 /* If this operand is at most 16 bits, convert it
1831 to a signed 16 bit number before trying to see
1832 whether it will fit in an even smaller size.
1833 This allows a 16-bit operand such as $0xffe0 to
1834 be recognised as within Imm8S range. */
1835 if ((i.types[op] & Imm16)
1836 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
1837 {
1838 i.op[op].imms->X_add_number =
1839 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
1840 }
1841 if ((i.types[op] & Imm32)
1842 && ((i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1))
1843 == 0))
1844 {
1845 i.op[op].imms->X_add_number = ((i.op[op].imms->X_add_number
1846 ^ ((offsetT) 1 << 31))
1847 - ((offsetT) 1 << 31));
1848 }
1849 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
1850
1851 /* We must avoid matching of Imm32 templates when 64bit
1852 only immediate is available. */
1853 if (guess_suffix == QWORD_MNEM_SUFFIX)
1854 i.types[op] &= ~Imm32;
1855 break;
1856
1857 case O_absent:
1858 case O_register:
1859 abort ();
1860
1861 /* Symbols and expressions. */
1862 default:
1863 /* Convert symbolic operand to proper sizes for matching. */
1864 switch (guess_suffix)
1865 {
1866 case QWORD_MNEM_SUFFIX:
1867 i.types[op] = Imm64 | Imm32S;
1868 break;
1869 case LONG_MNEM_SUFFIX:
1870 i.types[op] = Imm32 | Imm64;
1871 break;
1872 case WORD_MNEM_SUFFIX:
1873 i.types[op] = Imm16 | Imm32 | Imm64;
1874 break;
1875 break;
1876 case BYTE_MNEM_SUFFIX:
1877 i.types[op] = Imm8 | Imm8S | Imm16 | Imm32S | Imm32;
1878 break;
1879 break;
1880 }
1881 break;
1882 }
1883 }
1884 }
1885
1886 /* Try to use the smallest displacement type too. */
1887 static void
1888 optimize_disp ()
1889 {
1890 int op;
1891
1892 for (op = i.operands; --op >= 0;)
1893 if ((i.types[op] & Disp) && i.op[op].disps->X_op == O_constant)
1894 {
1895 offsetT disp = i.op[op].disps->X_add_number;
1896
1897 if (i.types[op] & Disp16)
1898 {
1899 /* We know this operand is at most 16 bits, so
1900 convert to a signed 16 bit number before trying
1901 to see whether it will fit in an even smaller
1902 size. */
1903
1904 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
1905 }
1906 else if (i.types[op] & Disp32)
1907 {
1908 /* We know this operand is at most 32 bits, so convert to a
1909 signed 32 bit number before trying to see whether it will
1910 fit in an even smaller size. */
1911 disp &= (((offsetT) 2 << 31) - 1);
1912 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1913 }
1914 if (flag_code == CODE_64BIT)
1915 {
1916 if (fits_in_signed_long (disp))
1917 i.types[op] |= Disp32S;
1918 if (fits_in_unsigned_long (disp))
1919 i.types[op] |= Disp32;
1920 }
1921 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
1922 && fits_in_signed_byte (disp))
1923 i.types[op] |= Disp8;
1924 }
1925 }
1926
1927 static int
1928 match_template ()
1929 {
1930 /* Points to template once we've found it. */
1931 const template *t;
1932 unsigned int overlap0, overlap1, overlap2;
1933 unsigned int found_reverse_match;
1934 int suffix_check;
1935
1936 #define MATCH(overlap, given, template) \
1937 ((overlap & ~JumpAbsolute) \
1938 && (((given) & (BaseIndex | JumpAbsolute)) \
1939 == ((overlap) & (BaseIndex | JumpAbsolute))))
1940
1941 /* If given types r0 and r1 are registers they must be of the same type
1942 unless the expected operand type register overlap is null.
1943 Note that Acc in a template matches every size of reg. */
1944 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
1945 (((g0) & Reg) == 0 || ((g1) & Reg) == 0 \
1946 || ((g0) & Reg) == ((g1) & Reg) \
1947 || ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
1948
1949 overlap0 = 0;
1950 overlap1 = 0;
1951 overlap2 = 0;
1952 found_reverse_match = 0;
1953 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
1954 ? No_bSuf
1955 : (i.suffix == WORD_MNEM_SUFFIX
1956 ? No_wSuf
1957 : (i.suffix == SHORT_MNEM_SUFFIX
1958 ? No_sSuf
1959 : (i.suffix == LONG_MNEM_SUFFIX
1960 ? No_lSuf
1961 : (i.suffix == QWORD_MNEM_SUFFIX
1962 ? No_qSuf
1963 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX
1964 ? No_xSuf : 0))))));
1965
1966 for (t = current_templates->start;
1967 t < current_templates->end;
1968 t++)
1969 {
1970 /* Must have right number of operands. */
1971 if (i.operands != t->operands)
1972 continue;
1973
1974 /* Check the suffix, except for some instructions in intel mode. */
1975 if ((t->opcode_modifier & suffix_check)
1976 && !(intel_syntax
1977 && (t->opcode_modifier & IgnoreSize))
1978 && !(intel_syntax
1979 && t->base_opcode == 0xd9
1980 && (t->extension_opcode == 5 /* 0xd9,5 "fldcw" */
1981 || t->extension_opcode == 7))) /* 0xd9,7 "f{n}stcw" */
1982 continue;
1983
1984 /* Do not verify operands when there are none. */
1985 else if (!t->operands)
1986 {
1987 if (t->cpu_flags & ~cpu_arch_flags)
1988 continue;
1989 /* We've found a match; break out of loop. */
1990 break;
1991 }
1992
1993 overlap0 = i.types[0] & t->operand_types[0];
1994 switch (t->operands)
1995 {
1996 case 1:
1997 if (!MATCH (overlap0, i.types[0], t->operand_types[0]))
1998 continue;
1999 break;
2000 case 2:
2001 case 3:
2002 overlap1 = i.types[1] & t->operand_types[1];
2003 if (!MATCH (overlap0, i.types[0], t->operand_types[0])
2004 || !MATCH (overlap1, i.types[1], t->operand_types[1])
2005 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2006 t->operand_types[0],
2007 overlap1, i.types[1],
2008 t->operand_types[1]))
2009 {
2010 /* Check if other direction is valid ... */
2011 if ((t->opcode_modifier & (D | FloatD)) == 0)
2012 continue;
2013
2014 /* Try reversing direction of operands. */
2015 overlap0 = i.types[0] & t->operand_types[1];
2016 overlap1 = i.types[1] & t->operand_types[0];
2017 if (!MATCH (overlap0, i.types[0], t->operand_types[1])
2018 || !MATCH (overlap1, i.types[1], t->operand_types[0])
2019 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2020 t->operand_types[1],
2021 overlap1, i.types[1],
2022 t->operand_types[0]))
2023 {
2024 /* Does not match either direction. */
2025 continue;
2026 }
2027 /* found_reverse_match holds which of D or FloatDR
2028 we've found. */
2029 found_reverse_match = t->opcode_modifier & (D | FloatDR);
2030 }
2031 /* Found a forward 2 operand match here. */
2032 else if (t->operands == 3)
2033 {
2034 /* Here we make use of the fact that there are no
2035 reverse match 3 operand instructions, and all 3
2036 operand instructions only need to be checked for
2037 register consistency between operands 2 and 3. */
2038 overlap2 = i.types[2] & t->operand_types[2];
2039 if (!MATCH (overlap2, i.types[2], t->operand_types[2])
2040 || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1],
2041 t->operand_types[1],
2042 overlap2, i.types[2],
2043 t->operand_types[2]))
2044
2045 continue;
2046 }
2047 /* Found either forward/reverse 2 or 3 operand match here:
2048 slip through to break. */
2049 }
2050 if (t->cpu_flags & ~cpu_arch_flags)
2051 {
2052 found_reverse_match = 0;
2053 continue;
2054 }
2055 /* We've found a match; break out of loop. */
2056 break;
2057 }
2058
2059 if (t == current_templates->end)
2060 {
2061 /* We found no match. */
2062 as_bad (_("suffix or operands invalid for `%s'"),
2063 current_templates->start->name);
2064 return 0;
2065 }
2066
2067 if (!quiet_warnings)
2068 {
2069 if (!intel_syntax
2070 && ((i.types[0] & JumpAbsolute)
2071 != (t->operand_types[0] & JumpAbsolute)))
2072 {
2073 as_warn (_("indirect %s without `*'"), t->name);
2074 }
2075
2076 if ((t->opcode_modifier & (IsPrefix | IgnoreSize))
2077 == (IsPrefix | IgnoreSize))
2078 {
2079 /* Warn them that a data or address size prefix doesn't
2080 affect assembly of the next line of code. */
2081 as_warn (_("stand-alone `%s' prefix"), t->name);
2082 }
2083 }
2084
2085 /* Copy the template we found. */
2086 i.tm = *t;
2087 if (found_reverse_match)
2088 {
2089 /* If we found a reverse match we must alter the opcode
2090 direction bit. found_reverse_match holds bits to change
2091 (different for int & float insns). */
2092
2093 i.tm.base_opcode ^= found_reverse_match;
2094
2095 i.tm.operand_types[0] = t->operand_types[1];
2096 i.tm.operand_types[1] = t->operand_types[0];
2097 }
2098
2099 return 1;
2100 }
2101
2102 static int
2103 check_string ()
2104 {
2105 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
2106 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
2107 {
2108 if (i.seg[0] != NULL && i.seg[0] != &es)
2109 {
2110 as_bad (_("`%s' operand %d must use `%%es' segment"),
2111 i.tm.name,
2112 mem_op + 1);
2113 return 0;
2114 }
2115 /* There's only ever one segment override allowed per instruction.
2116 This instruction possibly has a legal segment override on the
2117 second operand, so copy the segment to where non-string
2118 instructions store it, allowing common code. */
2119 i.seg[0] = i.seg[1];
2120 }
2121 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
2122 {
2123 if (i.seg[1] != NULL && i.seg[1] != &es)
2124 {
2125 as_bad (_("`%s' operand %d must use `%%es' segment"),
2126 i.tm.name,
2127 mem_op + 2);
2128 return 0;
2129 }
2130 }
2131 return 1;
2132 }
2133
2134 static int
2135 process_suffix ()
2136 {
2137 /* If matched instruction specifies an explicit instruction mnemonic
2138 suffix, use it. */
2139 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
2140 {
2141 if (i.tm.opcode_modifier & Size16)
2142 i.suffix = WORD_MNEM_SUFFIX;
2143 else if (i.tm.opcode_modifier & Size64)
2144 i.suffix = QWORD_MNEM_SUFFIX;
2145 else
2146 i.suffix = LONG_MNEM_SUFFIX;
2147 }
2148 else if (i.reg_operands)
2149 {
2150 /* If there's no instruction mnemonic suffix we try to invent one
2151 based on register operands. */
2152 if (!i.suffix)
2153 {
2154 /* We take i.suffix from the last register operand specified,
2155 Destination register type is more significant than source
2156 register type. */
2157 int op;
2158 for (op = i.operands; --op >= 0;)
2159 if ((i.types[op] & Reg)
2160 && !(i.tm.operand_types[op] & InOutPortReg))
2161 {
2162 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
2163 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
2164 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
2165 LONG_MNEM_SUFFIX);
2166 break;
2167 }
2168 }
2169 else if (i.suffix == BYTE_MNEM_SUFFIX)
2170 {
2171 if (!check_byte_reg ())
2172 return 0;
2173 }
2174 else if (i.suffix == LONG_MNEM_SUFFIX)
2175 {
2176 if (!check_long_reg ())
2177 return 0;
2178 }
2179 else if (i.suffix == QWORD_MNEM_SUFFIX)
2180 {
2181 if (!check_qword_reg ())
2182 return 0;
2183 }
2184 else if (i.suffix == WORD_MNEM_SUFFIX)
2185 {
2186 if (!check_word_reg ())
2187 return 0;
2188 }
2189 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2190 /* Do nothing if the instruction is going to ignore the prefix. */
2191 ;
2192 else
2193 abort ();
2194 }
2195 else if ((i.tm.opcode_modifier & DefaultSize) && !i.suffix)
2196 {
2197 i.suffix = stackop_size;
2198 }
2199
2200 /* Change the opcode based on the operand size given by i.suffix;
2201 We need not change things for byte insns. */
2202
2203 if (!i.suffix && (i.tm.opcode_modifier & W))
2204 {
2205 as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
2206 return 0;
2207 }
2208
2209 /* For movzx and movsx, need to check the register type. */
2210 if (intel_syntax
2211 && (i.tm.base_opcode == 0xfb6 || i.tm.base_opcode == 0xfbe))
2212 if (i.suffix && i.suffix == BYTE_MNEM_SUFFIX)
2213 {
2214 unsigned int prefix = DATA_PREFIX_OPCODE;
2215
2216 if ((i.op[1].regs->reg_type & Reg16) != 0)
2217 if (!add_prefix (prefix))
2218 return 0;
2219 }
2220
2221 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2222 {
2223 /* It's not a byte, select word/dword operation. */
2224 if (i.tm.opcode_modifier & W)
2225 {
2226 if (i.tm.opcode_modifier & ShortForm)
2227 i.tm.base_opcode |= 8;
2228 else
2229 i.tm.base_opcode |= 1;
2230 }
2231 /* Now select between word & dword operations via the operand
2232 size prefix, except for instructions that will ignore this
2233 prefix anyway. */
2234 if (i.suffix != QWORD_MNEM_SUFFIX
2235 && (i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2236 && !(i.tm.opcode_modifier & IgnoreSize))
2237 {
2238 unsigned int prefix = DATA_PREFIX_OPCODE;
2239 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2240 prefix = ADDR_PREFIX_OPCODE;
2241
2242 if (!add_prefix (prefix))
2243 return 0;
2244 }
2245
2246 if (i.suffix != QWORD_MNEM_SUFFIX && (flag_code == CODE_64BIT)
2247 && !(i.tm.opcode_modifier & IgnoreSize)
2248 && (i.tm.opcode_modifier & JumpByte))
2249 {
2250 if (!add_prefix (ADDR_PREFIX_OPCODE))
2251 return 0;
2252 }
2253
2254 /* Set mode64 for an operand. */
2255 if (i.suffix == QWORD_MNEM_SUFFIX
2256 && (i.tm.opcode_modifier & NoRex64) == 0)
2257 {
2258 i.rex |= REX_MODE64;
2259 if (flag_code < CODE_64BIT)
2260 {
2261 as_bad (_("64bit operations available only in 64bit modes."));
2262 return 0;
2263 }
2264 }
2265
2266 /* Size floating point instruction. */
2267 if (i.suffix == LONG_MNEM_SUFFIX)
2268 {
2269 if (i.tm.opcode_modifier & FloatMF)
2270 i.tm.base_opcode ^= 4;
2271 }
2272 }
2273
2274 return 1;
2275 }
2276
2277 static int
2278 check_byte_reg ()
2279 {
2280 int op;
2281 for (op = i.operands; --op >= 0;)
2282 {
2283 /* If this is an eight bit register, it's OK. If it's the 16 or
2284 32 bit version of an eight bit register, we will just use the
2285 low portion, and that's OK too. */
2286 if (i.types[op] & Reg8)
2287 continue;
2288
2289 /* movzx and movsx should not generate this warning. */
2290 if (intel_syntax
2291 && (i.tm.base_opcode == 0xfb7
2292 || i.tm.base_opcode == 0xfb6
2293 || i.tm.base_opcode == 0x63
2294 || i.tm.base_opcode == 0xfbe
2295 || i.tm.base_opcode == 0xfbf))
2296 continue;
2297
2298 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4
2299 #if 0
2300 /* Check that the template allows eight bit regs. This
2301 kills insns such as `orb $1,%edx', which maybe should be
2302 allowed. */
2303 && (i.tm.operand_types[op] & (Reg8 | InOutPortReg))
2304 #endif
2305 )
2306 {
2307 /* Prohibit these changes in the 64bit mode, since the
2308 lowering is more complicated. */
2309 if (flag_code == CODE_64BIT
2310 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2311 {
2312 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2313 i.op[op].regs->reg_name,
2314 i.suffix);
2315 return 0;
2316 }
2317 #if REGISTER_WARNINGS
2318 if (!quiet_warnings
2319 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2320 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2321 (i.op[op].regs + (i.types[op] & Reg16
2322 ? REGNAM_AL - REGNAM_AX
2323 : REGNAM_AL - REGNAM_EAX))->reg_name,
2324 i.op[op].regs->reg_name,
2325 i.suffix);
2326 #endif
2327 continue;
2328 }
2329 /* Any other register is bad. */
2330 if (i.types[op] & (Reg | RegMMX | RegXMM
2331 | SReg2 | SReg3
2332 | Control | Debug | Test
2333 | FloatReg | FloatAcc))
2334 {
2335 as_bad (_("`%%%s' not allowed with `%s%c'"),
2336 i.op[op].regs->reg_name,
2337 i.tm.name,
2338 i.suffix);
2339 return 0;
2340 }
2341 }
2342 return 1;
2343 }
2344
2345 static int
2346 check_long_reg ()
2347 {
2348 int op;
2349
2350 for (op = i.operands; --op >= 0;)
2351 /* Reject eight bit registers, except where the template requires
2352 them. (eg. movzb) */
2353 if ((i.types[op] & Reg8) != 0
2354 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2355 {
2356 as_bad (_("`%%%s' not allowed with `%s%c'"),
2357 i.op[op].regs->reg_name,
2358 i.tm.name,
2359 i.suffix);
2360 return 0;
2361 }
2362 /* Warn if the e prefix on a general reg is missing. */
2363 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2364 && (i.types[op] & Reg16) != 0
2365 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2366 {
2367 /* Prohibit these changes in the 64bit mode, since the
2368 lowering is more complicated. */
2369 if (flag_code == CODE_64BIT)
2370 {
2371 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2372 i.op[op].regs->reg_name,
2373 i.suffix);
2374 return 0;
2375 }
2376 #if REGISTER_WARNINGS
2377 else
2378 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2379 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
2380 i.op[op].regs->reg_name,
2381 i.suffix);
2382 #endif
2383 }
2384 /* Warn if the r prefix on a general reg is missing. */
2385 else if ((i.types[op] & Reg64) != 0
2386 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2387 {
2388 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2389 i.op[op].regs->reg_name,
2390 i.suffix);
2391 return 0;
2392 }
2393 return 1;
2394 }
2395
2396 static int
2397 check_qword_reg ()
2398 {
2399 int op;
2400
2401 for (op = i.operands; --op >= 0; )
2402 /* Reject eight bit registers, except where the template requires
2403 them. (eg. movzb) */
2404 if ((i.types[op] & Reg8) != 0
2405 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2406 {
2407 as_bad (_("`%%%s' not allowed with `%s%c'"),
2408 i.op[op].regs->reg_name,
2409 i.tm.name,
2410 i.suffix);
2411 return 0;
2412 }
2413 /* Warn if the e prefix on a general reg is missing. */
2414 else if (((i.types[op] & Reg16) != 0
2415 || (i.types[op] & Reg32) != 0)
2416 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2417 {
2418 /* Prohibit these changes in the 64bit mode, since the
2419 lowering is more complicated. */
2420 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2421 i.op[op].regs->reg_name,
2422 i.suffix);
2423 return 0;
2424 }
2425 return 1;
2426 }
2427
2428 static int
2429 check_word_reg ()
2430 {
2431 int op;
2432 for (op = i.operands; --op >= 0;)
2433 /* Reject eight bit registers, except where the template requires
2434 them. (eg. movzb) */
2435 if ((i.types[op] & Reg8) != 0
2436 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2437 {
2438 as_bad (_("`%%%s' not allowed with `%s%c'"),
2439 i.op[op].regs->reg_name,
2440 i.tm.name,
2441 i.suffix);
2442 return 0;
2443 }
2444 /* Warn if the e prefix on a general reg is present. */
2445 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2446 && (i.types[op] & Reg32) != 0
2447 && (i.tm.operand_types[op] & (Reg16 | Acc)) != 0)
2448 {
2449 /* Prohibit these changes in the 64bit mode, since the
2450 lowering is more complicated. */
2451 if (flag_code == CODE_64BIT)
2452 {
2453 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2454 i.op[op].regs->reg_name,
2455 i.suffix);
2456 return 0;
2457 }
2458 else
2459 #if REGISTER_WARNINGS
2460 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2461 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
2462 i.op[op].regs->reg_name,
2463 i.suffix);
2464 #endif
2465 }
2466 return 1;
2467 }
2468
2469 static int
2470 finalize_imm ()
2471 {
2472 unsigned int overlap0, overlap1, overlap2;
2473
2474 overlap0 = i.types[0] & i.tm.operand_types[0];
2475 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S))
2476 && overlap0 != Imm8 && overlap0 != Imm8S
2477 && overlap0 != Imm16 && overlap0 != Imm32S
2478 && overlap0 != Imm32 && overlap0 != Imm64)
2479 {
2480 if (i.suffix)
2481 {
2482 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX
2483 ? Imm8 | Imm8S
2484 : (i.suffix == WORD_MNEM_SUFFIX
2485 ? Imm16
2486 : (i.suffix == QWORD_MNEM_SUFFIX
2487 ? Imm64 | Imm32S
2488 : Imm32)));
2489 }
2490 else if (overlap0 == (Imm16 | Imm32S | Imm32)
2491 || overlap0 == (Imm16 | Imm32)
2492 || overlap0 == (Imm16 | Imm32S))
2493 {
2494 overlap0 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
2495 ? Imm16 : Imm32S);
2496 }
2497 if (overlap0 != Imm8 && overlap0 != Imm8S
2498 && overlap0 != Imm16 && overlap0 != Imm32S
2499 && overlap0 != Imm32 && overlap0 != Imm64)
2500 {
2501 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
2502 return 0;
2503 }
2504 }
2505 i.types[0] = overlap0;
2506
2507 overlap1 = i.types[1] & i.tm.operand_types[1];
2508 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32))
2509 && overlap1 != Imm8 && overlap1 != Imm8S
2510 && overlap1 != Imm16 && overlap1 != Imm32S
2511 && overlap1 != Imm32 && overlap1 != Imm64)
2512 {
2513 if (i.suffix)
2514 {
2515 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX
2516 ? Imm8 | Imm8S
2517 : (i.suffix == WORD_MNEM_SUFFIX
2518 ? Imm16
2519 : (i.suffix == QWORD_MNEM_SUFFIX
2520 ? Imm64 | Imm32S
2521 : Imm32)));
2522 }
2523 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
2524 || overlap1 == (Imm16 | Imm32)
2525 || overlap1 == (Imm16 | Imm32S))
2526 {
2527 overlap1 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
2528 ? Imm16 : Imm32S);
2529 }
2530 if (overlap1 != Imm8 && overlap1 != Imm8S
2531 && overlap1 != Imm16 && overlap1 != Imm32S
2532 && overlap1 != Imm32 && overlap1 != Imm64)
2533 {
2534 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
2535 return 0;
2536 }
2537 }
2538 i.types[1] = overlap1;
2539
2540 overlap2 = i.types[2] & i.tm.operand_types[2];
2541 assert ((overlap2 & Imm) == 0);
2542 i.types[2] = overlap2;
2543
2544 return 1;
2545 }
2546
2547 static int
2548 process_operands ()
2549 {
2550 /* Default segment register this instruction will use for memory
2551 accesses. 0 means unknown. This is only for optimizing out
2552 unnecessary segment overrides. */
2553 const seg_entry *default_seg = 0;
2554
2555 /* The imul $imm, %reg instruction is converted into
2556 imul $imm, %reg, %reg, and the clr %reg instruction
2557 is converted into xor %reg, %reg. */
2558 if (i.tm.opcode_modifier & regKludge)
2559 {
2560 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
2561 /* Pretend we saw the extra register operand. */
2562 assert (i.op[first_reg_op + 1].regs == 0);
2563 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
2564 i.types[first_reg_op + 1] = i.types[first_reg_op];
2565 i.reg_operands = 2;
2566 }
2567
2568 if (i.tm.opcode_modifier & ShortForm)
2569 {
2570 /* The register or float register operand is in operand 0 or 1. */
2571 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
2572 /* Register goes in low 3 bits of opcode. */
2573 i.tm.base_opcode |= i.op[op].regs->reg_num;
2574 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2575 i.rex |= REX_EXTZ;
2576 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2577 {
2578 /* Warn about some common errors, but press on regardless.
2579 The first case can be generated by gcc (<= 2.8.1). */
2580 if (i.operands == 2)
2581 {
2582 /* Reversed arguments on faddp, fsubp, etc. */
2583 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
2584 i.op[1].regs->reg_name,
2585 i.op[0].regs->reg_name);
2586 }
2587 else
2588 {
2589 /* Extraneous `l' suffix on fp insn. */
2590 as_warn (_("translating to `%s %%%s'"), i.tm.name,
2591 i.op[0].regs->reg_name);
2592 }
2593 }
2594 }
2595 else if (i.tm.opcode_modifier & Modrm)
2596 {
2597 /* The opcode is completed (modulo i.tm.extension_opcode which
2598 must be put into the modrm byte).
2599 Now, we make the modrm & index base bytes based on all the
2600 info we've collected. */
2601
2602 default_seg = build_modrm_byte ();
2603 }
2604 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
2605 {
2606 if (i.tm.base_opcode == POP_SEG_SHORT
2607 && i.op[0].regs->reg_num == 1)
2608 {
2609 as_bad (_("you can't `pop %%cs'"));
2610 return 0;
2611 }
2612 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
2613 if ((i.op[0].regs->reg_flags & RegRex) != 0)
2614 i.rex |= REX_EXTZ;
2615 }
2616 else if ((i.tm.base_opcode & ~(D | W)) == MOV_AX_DISP32)
2617 {
2618 default_seg = &ds;
2619 }
2620 else if ((i.tm.opcode_modifier & IsString) != 0)
2621 {
2622 /* For the string instructions that allow a segment override
2623 on one of their operands, the default segment is ds. */
2624 default_seg = &ds;
2625 }
2626
2627 /* If a segment was explicitly specified,
2628 and the specified segment is not the default,
2629 use an opcode prefix to select it.
2630 If we never figured out what the default segment is,
2631 then default_seg will be zero at this point,
2632 and the specified segment prefix will always be used. */
2633 if ((i.seg[0]) && (i.seg[0] != default_seg))
2634 {
2635 if (!add_prefix (i.seg[0]->seg_prefix))
2636 return 0;
2637 }
2638 return 1;
2639 }
2640
2641 static const seg_entry *
2642 build_modrm_byte ()
2643 {
2644 const seg_entry *default_seg = 0;
2645
2646 /* i.reg_operands MUST be the number of real register operands;
2647 implicit registers do not count. */
2648 if (i.reg_operands == 2)
2649 {
2650 unsigned int source, dest;
2651 source = ((i.types[0]
2652 & (Reg | RegMMX | RegXMM
2653 | SReg2 | SReg3
2654 | Control | Debug | Test))
2655 ? 0 : 1);
2656 dest = source + 1;
2657
2658 i.rm.mode = 3;
2659 /* One of the register operands will be encoded in the i.tm.reg
2660 field, the other in the combined i.tm.mode and i.tm.regmem
2661 fields. If no form of this instruction supports a memory
2662 destination operand, then we assume the source operand may
2663 sometimes be a memory operand and so we need to store the
2664 destination in the i.rm.reg field. */
2665 if ((i.tm.operand_types[dest] & AnyMem) == 0)
2666 {
2667 i.rm.reg = i.op[dest].regs->reg_num;
2668 i.rm.regmem = i.op[source].regs->reg_num;
2669 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
2670 i.rex |= REX_EXTX;
2671 if ((i.op[source].regs->reg_flags & RegRex) != 0)
2672 i.rex |= REX_EXTZ;
2673 }
2674 else
2675 {
2676 i.rm.reg = i.op[source].regs->reg_num;
2677 i.rm.regmem = i.op[dest].regs->reg_num;
2678 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
2679 i.rex |= REX_EXTZ;
2680 if ((i.op[source].regs->reg_flags & RegRex) != 0)
2681 i.rex |= REX_EXTX;
2682 }
2683 }
2684 else
2685 { /* If it's not 2 reg operands... */
2686 if (i.mem_operands)
2687 {
2688 unsigned int fake_zero_displacement = 0;
2689 unsigned int op = ((i.types[0] & AnyMem)
2690 ? 0
2691 : (i.types[1] & AnyMem) ? 1 : 2);
2692
2693 default_seg = &ds;
2694
2695 if (i.base_reg == 0)
2696 {
2697 i.rm.mode = 0;
2698 if (!i.disp_operands)
2699 fake_zero_displacement = 1;
2700 if (i.index_reg == 0)
2701 {
2702 /* Operand is just <disp> */
2703 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)
2704 && (flag_code != CODE_64BIT))
2705 {
2706 i.rm.regmem = NO_BASE_REGISTER_16;
2707 i.types[op] &= ~Disp;
2708 i.types[op] |= Disp16;
2709 }
2710 else if (flag_code != CODE_64BIT
2711 || (i.prefix[ADDR_PREFIX] != 0))
2712 {
2713 i.rm.regmem = NO_BASE_REGISTER;
2714 i.types[op] &= ~Disp;
2715 i.types[op] |= Disp32;
2716 }
2717 else
2718 {
2719 /* 64bit mode overwrites the 32bit absolute
2720 addressing by RIP relative addressing and
2721 absolute addressing is encoded by one of the
2722 redundant SIB forms. */
2723 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2724 i.sib.base = NO_BASE_REGISTER;
2725 i.sib.index = NO_INDEX_REGISTER;
2726 i.types[op] &= ~Disp;
2727 i.types[op] |= Disp32S;
2728 }
2729 }
2730 else /* !i.base_reg && i.index_reg */
2731 {
2732 i.sib.index = i.index_reg->reg_num;
2733 i.sib.base = NO_BASE_REGISTER;
2734 i.sib.scale = i.log2_scale_factor;
2735 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2736 i.types[op] &= ~Disp;
2737 if (flag_code != CODE_64BIT)
2738 i.types[op] |= Disp32; /* Must be 32 bit */
2739 else
2740 i.types[op] |= Disp32S;
2741 if ((i.index_reg->reg_flags & RegRex) != 0)
2742 i.rex |= REX_EXTY;
2743 }
2744 }
2745 /* RIP addressing for 64bit mode. */
2746 else if (i.base_reg->reg_type == BaseIndex)
2747 {
2748 i.rm.regmem = NO_BASE_REGISTER;
2749 i.types[op] &= ~Disp;
2750 i.types[op] |= Disp32S;
2751 i.flags[op] = Operand_PCrel;
2752 }
2753 else if (i.base_reg->reg_type & Reg16)
2754 {
2755 switch (i.base_reg->reg_num)
2756 {
2757 case 3: /* (%bx) */
2758 if (i.index_reg == 0)
2759 i.rm.regmem = 7;
2760 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
2761 i.rm.regmem = i.index_reg->reg_num - 6;
2762 break;
2763 case 5: /* (%bp) */
2764 default_seg = &ss;
2765 if (i.index_reg == 0)
2766 {
2767 i.rm.regmem = 6;
2768 if ((i.types[op] & Disp) == 0)
2769 {
2770 /* fake (%bp) into 0(%bp) */
2771 i.types[op] |= Disp8;
2772 fake_zero_displacement = 1;
2773 }
2774 }
2775 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
2776 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
2777 break;
2778 default: /* (%si) -> 4 or (%di) -> 5 */
2779 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
2780 }
2781 i.rm.mode = mode_from_disp_size (i.types[op]);
2782 }
2783 else /* i.base_reg and 32/64 bit mode */
2784 {
2785 if (flag_code == CODE_64BIT
2786 && (i.types[op] & Disp))
2787 {
2788 if (i.types[op] & Disp8)
2789 i.types[op] = Disp8 | Disp32S;
2790 else
2791 i.types[op] = Disp32S;
2792 }
2793 i.rm.regmem = i.base_reg->reg_num;
2794 if ((i.base_reg->reg_flags & RegRex) != 0)
2795 i.rex |= REX_EXTZ;
2796 i.sib.base = i.base_reg->reg_num;
2797 /* x86-64 ignores REX prefix bit here to avoid decoder
2798 complications. */
2799 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
2800 {
2801 default_seg = &ss;
2802 if (i.disp_operands == 0)
2803 {
2804 fake_zero_displacement = 1;
2805 i.types[op] |= Disp8;
2806 }
2807 }
2808 else if (i.base_reg->reg_num == ESP_REG_NUM)
2809 {
2810 default_seg = &ss;
2811 }
2812 i.sib.scale = i.log2_scale_factor;
2813 if (i.index_reg == 0)
2814 {
2815 /* <disp>(%esp) becomes two byte modrm with no index
2816 register. We've already stored the code for esp
2817 in i.rm.regmem ie. ESCAPE_TO_TWO_BYTE_ADDRESSING.
2818 Any base register besides %esp will not use the
2819 extra modrm byte. */
2820 i.sib.index = NO_INDEX_REGISTER;
2821 #if !SCALE1_WHEN_NO_INDEX
2822 /* Another case where we force the second modrm byte. */
2823 if (i.log2_scale_factor)
2824 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2825 #endif
2826 }
2827 else
2828 {
2829 i.sib.index = i.index_reg->reg_num;
2830 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2831 if ((i.index_reg->reg_flags & RegRex) != 0)
2832 i.rex |= REX_EXTY;
2833 }
2834 i.rm.mode = mode_from_disp_size (i.types[op]);
2835 }
2836
2837 if (fake_zero_displacement)
2838 {
2839 /* Fakes a zero displacement assuming that i.types[op]
2840 holds the correct displacement size. */
2841 expressionS *exp;
2842
2843 assert (i.op[op].disps == 0);
2844 exp = &disp_expressions[i.disp_operands++];
2845 i.op[op].disps = exp;
2846 exp->X_op = O_constant;
2847 exp->X_add_number = 0;
2848 exp->X_add_symbol = (symbolS *) 0;
2849 exp->X_op_symbol = (symbolS *) 0;
2850 }
2851 }
2852
2853 /* Fill in i.rm.reg or i.rm.regmem field with register operand
2854 (if any) based on i.tm.extension_opcode. Again, we must be
2855 careful to make sure that segment/control/debug/test/MMX
2856 registers are coded into the i.rm.reg field. */
2857 if (i.reg_operands)
2858 {
2859 unsigned int op =
2860 ((i.types[0]
2861 & (Reg | RegMMX | RegXMM
2862 | SReg2 | SReg3
2863 | Control | Debug | Test))
2864 ? 0
2865 : ((i.types[1]
2866 & (Reg | RegMMX | RegXMM
2867 | SReg2 | SReg3
2868 | Control | Debug | Test))
2869 ? 1
2870 : 2));
2871 /* If there is an extension opcode to put here, the register
2872 number must be put into the regmem field. */
2873 if (i.tm.extension_opcode != None)
2874 {
2875 i.rm.regmem = i.op[op].regs->reg_num;
2876 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2877 i.rex |= REX_EXTZ;
2878 }
2879 else
2880 {
2881 i.rm.reg = i.op[op].regs->reg_num;
2882 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2883 i.rex |= REX_EXTX;
2884 }
2885
2886 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2 we
2887 must set it to 3 to indicate this is a register operand
2888 in the regmem field. */
2889 if (!i.mem_operands)
2890 i.rm.mode = 3;
2891 }
2892
2893 /* Fill in i.rm.reg field with extension opcode (if any). */
2894 if (i.tm.extension_opcode != None)
2895 i.rm.reg = i.tm.extension_opcode;
2896 }
2897 return default_seg;
2898 }
2899
2900 static void
2901 output_branch ()
2902 {
2903 char *p;
2904 int code16;
2905 int prefix;
2906 relax_substateT subtype;
2907 symbolS *sym;
2908 offsetT off;
2909
2910 code16 = 0;
2911 if (flag_code == CODE_16BIT)
2912 code16 = CODE16;
2913
2914 prefix = 0;
2915 if (i.prefix[DATA_PREFIX] != 0)
2916 {
2917 prefix = 1;
2918 i.prefixes -= 1;
2919 code16 ^= CODE16;
2920 }
2921 /* Pentium4 branch hints. */
2922 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
2923 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
2924 {
2925 prefix++;
2926 i.prefixes--;
2927 }
2928 if (i.prefix[REX_PREFIX] != 0)
2929 {
2930 prefix++;
2931 i.prefixes--;
2932 }
2933
2934 if (i.prefixes != 0 && !intel_syntax)
2935 as_warn (_("skipping prefixes on this instruction"));
2936
2937 /* It's always a symbol; End frag & setup for relax.
2938 Make sure there is enough room in this frag for the largest
2939 instruction we may generate in md_convert_frag. This is 2
2940 bytes for the opcode and room for the prefix and largest
2941 displacement. */
2942 frag_grow (prefix + 2 + 4);
2943 /* Prefix and 1 opcode byte go in fr_fix. */
2944 p = frag_more (prefix + 1);
2945 if (i.prefix[DATA_PREFIX] != 0)
2946 *p++ = DATA_PREFIX_OPCODE;
2947 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
2948 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
2949 *p++ = i.prefix[SEG_PREFIX];
2950 if (i.prefix[REX_PREFIX] != 0)
2951 *p++ = i.prefix[REX_PREFIX];
2952 *p = i.tm.base_opcode;
2953
2954 if ((unsigned char) *p == JUMP_PC_RELATIVE)
2955 subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
2956 else if ((cpu_arch_flags & Cpu386) != 0)
2957 subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
2958 else
2959 subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
2960 subtype |= code16;
2961
2962 sym = i.op[0].disps->X_add_symbol;
2963 off = i.op[0].disps->X_add_number;
2964
2965 if (i.op[0].disps->X_op != O_constant
2966 && i.op[0].disps->X_op != O_symbol)
2967 {
2968 /* Handle complex expressions. */
2969 sym = make_expr_symbol (i.op[0].disps);
2970 off = 0;
2971 }
2972
2973 /* 1 possible extra opcode + 4 byte displacement go in var part.
2974 Pass reloc in fr_var. */
2975 frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
2976 }
2977
2978 static void
2979 output_jump ()
2980 {
2981 char *p;
2982 int size;
2983 fixS *fixP;
2984
2985 if (i.tm.opcode_modifier & JumpByte)
2986 {
2987 /* This is a loop or jecxz type instruction. */
2988 size = 1;
2989 if (i.prefix[ADDR_PREFIX] != 0)
2990 {
2991 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
2992 i.prefixes -= 1;
2993 }
2994 /* Pentium4 branch hints. */
2995 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
2996 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
2997 {
2998 FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
2999 i.prefixes--;
3000 }
3001 }
3002 else
3003 {
3004 int code16;
3005
3006 code16 = 0;
3007 if (flag_code == CODE_16BIT)
3008 code16 = CODE16;
3009
3010 if (i.prefix[DATA_PREFIX] != 0)
3011 {
3012 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
3013 i.prefixes -= 1;
3014 code16 ^= CODE16;
3015 }
3016
3017 size = 4;
3018 if (code16)
3019 size = 2;
3020 }
3021
3022 if (i.prefix[REX_PREFIX] != 0)
3023 {
3024 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
3025 i.prefixes -= 1;
3026 }
3027
3028 if (i.prefixes != 0 && !intel_syntax)
3029 as_warn (_("skipping prefixes on this instruction"));
3030
3031 p = frag_more (1 + size);
3032 *p++ = i.tm.base_opcode;
3033
3034 fixP = fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3035 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
3036 fixP->fx_pcrel_adjust = size;
3037 }
3038
3039 static void
3040 output_interseg_jump ()
3041 {
3042 char *p;
3043 int size;
3044 int prefix;
3045 int code16;
3046
3047 code16 = 0;
3048 if (flag_code == CODE_16BIT)
3049 code16 = CODE16;
3050
3051 prefix = 0;
3052 if (i.prefix[DATA_PREFIX] != 0)
3053 {
3054 prefix = 1;
3055 i.prefixes -= 1;
3056 code16 ^= CODE16;
3057 }
3058 if (i.prefix[REX_PREFIX] != 0)
3059 {
3060 prefix++;
3061 i.prefixes -= 1;
3062 }
3063
3064 size = 4;
3065 if (code16)
3066 size = 2;
3067
3068 if (i.prefixes != 0 && !intel_syntax)
3069 as_warn (_("skipping prefixes on this instruction"));
3070
3071 /* 1 opcode; 2 segment; offset */
3072 p = frag_more (prefix + 1 + 2 + size);
3073
3074 if (i.prefix[DATA_PREFIX] != 0)
3075 *p++ = DATA_PREFIX_OPCODE;
3076
3077 if (i.prefix[REX_PREFIX] != 0)
3078 *p++ = i.prefix[REX_PREFIX];
3079
3080 *p++ = i.tm.base_opcode;
3081 if (i.op[1].imms->X_op == O_constant)
3082 {
3083 offsetT n = i.op[1].imms->X_add_number;
3084
3085 if (size == 2
3086 && !fits_in_unsigned_word (n)
3087 && !fits_in_signed_word (n))
3088 {
3089 as_bad (_("16-bit jump out of range"));
3090 return;
3091 }
3092 md_number_to_chars (p, n, size);
3093 }
3094 else
3095 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3096 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
3097 if (i.op[0].imms->X_op != O_constant)
3098 as_bad (_("can't handle non absolute segment in `%s'"),
3099 i.tm.name);
3100 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
3101 }
3102
3103 static void
3104 output_insn ()
3105 {
3106 /* Tie dwarf2 debug info to the address at the start of the insn.
3107 We can't do this after the insn has been output as the current
3108 frag may have been closed off. eg. by frag_var. */
3109 dwarf2_emit_insn (0);
3110
3111 /* Output jumps. */
3112 if (i.tm.opcode_modifier & Jump)
3113 output_branch ();
3114 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
3115 output_jump ();
3116 else if (i.tm.opcode_modifier & JumpInterSegment)
3117 output_interseg_jump ();
3118 else
3119 {
3120 /* Output normal instructions here. */
3121 char *p;
3122 unsigned char *q;
3123
3124 /* All opcodes on i386 have either 1 or 2 bytes. We may use third
3125 byte for the SSE instructions to specify a prefix they require. */
3126 if (i.tm.base_opcode & 0xff0000)
3127 add_prefix ((i.tm.base_opcode >> 16) & 0xff);
3128
3129 /* The prefix bytes. */
3130 for (q = i.prefix;
3131 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
3132 q++)
3133 {
3134 if (*q)
3135 {
3136 p = frag_more (1);
3137 md_number_to_chars (p, (valueT) *q, 1);
3138 }
3139 }
3140
3141 /* Now the opcode; be careful about word order here! */
3142 if (fits_in_unsigned_byte (i.tm.base_opcode))
3143 {
3144 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
3145 }
3146 else
3147 {
3148 p = frag_more (2);
3149 /* Put out high byte first: can't use md_number_to_chars! */
3150 *p++ = (i.tm.base_opcode >> 8) & 0xff;
3151 *p = i.tm.base_opcode & 0xff;
3152 }
3153
3154 /* Now the modrm byte and sib byte (if present). */
3155 if (i.tm.opcode_modifier & Modrm)
3156 {
3157 p = frag_more (1);
3158 md_number_to_chars (p,
3159 (valueT) (i.rm.regmem << 0
3160 | i.rm.reg << 3
3161 | i.rm.mode << 6),
3162 1);
3163 /* If i.rm.regmem == ESP (4)
3164 && i.rm.mode != (Register mode)
3165 && not 16 bit
3166 ==> need second modrm byte. */
3167 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
3168 && i.rm.mode != 3
3169 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
3170 {
3171 p = frag_more (1);
3172 md_number_to_chars (p,
3173 (valueT) (i.sib.base << 0
3174 | i.sib.index << 3
3175 | i.sib.scale << 6),
3176 1);
3177 }
3178 }
3179
3180 if (i.disp_operands)
3181 output_disp ();
3182
3183 if (i.imm_operands)
3184 output_imm ();
3185 }
3186
3187 #ifdef DEBUG386
3188 if (flag_debug)
3189 {
3190 pi (line, &i);
3191 }
3192 #endif /* DEBUG386 */
3193 }
3194
3195 static void
3196 output_disp ()
3197 {
3198 char *p;
3199 unsigned int n;
3200
3201 for (n = 0; n < i.operands; n++)
3202 {
3203 if (i.types[n] & Disp)
3204 {
3205 if (i.op[n].disps->X_op == O_constant)
3206 {
3207 int size;
3208 offsetT val;
3209
3210 size = 4;
3211 if (i.types[n] & (Disp8 | Disp16 | Disp64))
3212 {
3213 size = 2;
3214 if (i.types[n] & Disp8)
3215 size = 1;
3216 if (i.types[n] & Disp64)
3217 size = 8;
3218 }
3219 val = offset_in_range (i.op[n].disps->X_add_number,
3220 size);
3221 p = frag_more (size);
3222 md_number_to_chars (p, val, size);
3223 }
3224 else
3225 {
3226 int size = 4;
3227 int sign = 0;
3228 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
3229 fixS *fixP;
3230
3231 /* The PC relative address is computed relative
3232 to the instruction boundary, so in case immediate
3233 fields follows, we need to adjust the value. */
3234 if (pcrel && i.imm_operands)
3235 {
3236 int imm_size = 4;
3237 unsigned int n1;
3238
3239 for (n1 = 0; n1 < i.operands; n1++)
3240 if (i.types[n1] & Imm)
3241 {
3242 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
3243 {
3244 imm_size = 2;
3245 if (i.types[n1] & (Imm8 | Imm8S))
3246 imm_size = 1;
3247 if (i.types[n1] & Imm64)
3248 imm_size = 8;
3249 }
3250 break;
3251 }
3252 /* We should find the immediate. */
3253 if (n1 == i.operands)
3254 abort ();
3255 i.op[n].disps->X_add_number -= imm_size;
3256 }
3257
3258 if (i.types[n] & Disp32S)
3259 sign = 1;
3260
3261 if (i.types[n] & (Disp16 | Disp64))
3262 {
3263 size = 2;
3264 if (i.types[n] & Disp64)
3265 size = 8;
3266 }
3267
3268 p = frag_more (size);
3269 fixP = fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3270 i.op[n].disps, pcrel,
3271 reloc (size, pcrel, sign, i.reloc[n]));
3272 if (pcrel)
3273 fixP->fx_pcrel_adjust = size;
3274 }
3275 }
3276 }
3277 }
3278
3279 static void
3280 output_imm ()
3281 {
3282 char *p;
3283 unsigned int n;
3284
3285 for (n = 0; n < i.operands; n++)
3286 {
3287 if (i.types[n] & Imm)
3288 {
3289 if (i.op[n].imms->X_op == O_constant)
3290 {
3291 int size;
3292 offsetT val;
3293
3294 size = 4;
3295 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3296 {
3297 size = 2;
3298 if (i.types[n] & (Imm8 | Imm8S))
3299 size = 1;
3300 else if (i.types[n] & Imm64)
3301 size = 8;
3302 }
3303 val = offset_in_range (i.op[n].imms->X_add_number,
3304 size);
3305 p = frag_more (size);
3306 md_number_to_chars (p, val, size);
3307 }
3308 else
3309 {
3310 /* Not absolute_section.
3311 Need a 32-bit fixup (don't support 8bit
3312 non-absolute imms). Try to support other
3313 sizes ... */
3314 RELOC_ENUM reloc_type;
3315 int size = 4;
3316 int sign = 0;
3317
3318 if ((i.types[n] & (Imm32S))
3319 && i.suffix == QWORD_MNEM_SUFFIX)
3320 sign = 1;
3321 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3322 {
3323 size = 2;
3324 if (i.types[n] & (Imm8 | Imm8S))
3325 size = 1;
3326 if (i.types[n] & Imm64)
3327 size = 8;
3328 }
3329
3330 p = frag_more (size);
3331 reloc_type = reloc (size, 0, sign, i.reloc[n]);
3332 #ifdef BFD_ASSEMBLER
3333 if (reloc_type == BFD_RELOC_32
3334 && GOT_symbol
3335 && GOT_symbol == i.op[n].imms->X_add_symbol
3336 && (i.op[n].imms->X_op == O_symbol
3337 || (i.op[n].imms->X_op == O_add
3338 && ((symbol_get_value_expression
3339 (i.op[n].imms->X_op_symbol)->X_op)
3340 == O_subtract))))
3341 {
3342 /* We don't support dynamic linking on x86-64 yet. */
3343 if (flag_code == CODE_64BIT)
3344 abort ();
3345 reloc_type = BFD_RELOC_386_GOTPC;
3346 i.op[n].imms->X_add_number += 3;
3347 }
3348 #endif
3349 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3350 i.op[n].imms, 0, reloc_type);
3351 }
3352 }
3353 }
3354 }
3355 \f
3356 #ifndef LEX_AT
3357 static char *lex_got PARAMS ((RELOC_ENUM *, int *));
3358
3359 /* Parse operands of the form
3360 <symbol>@GOTOFF+<nnn>
3361 and similar .plt or .got references.
3362
3363 If we find one, set up the correct relocation in RELOC and copy the
3364 input string, minus the `@GOTOFF' into a malloc'd buffer for
3365 parsing by the calling routine. Return this buffer, and if ADJUST
3366 is non-null set it to the length of the string we removed from the
3367 input line. Otherwise return NULL. */
3368 static char *
3369 lex_got (reloc, adjust)
3370 RELOC_ENUM *reloc;
3371 int *adjust;
3372 {
3373 static const char * const mode_name[NUM_FLAG_CODE] = { "32", "16", "64" };
3374 static const struct {
3375 const char *str;
3376 const RELOC_ENUM rel[NUM_FLAG_CODE];
3377 } gotrel[] = {
3378 { "PLT", { BFD_RELOC_386_PLT32, 0, BFD_RELOC_X86_64_PLT32 } },
3379 { "GOTOFF", { BFD_RELOC_386_GOTOFF, 0, 0 } },
3380 { "GOTPCREL", { 0, 0, BFD_RELOC_X86_64_GOTPCREL } },
3381 { "GOT", { BFD_RELOC_386_GOT32, 0, BFD_RELOC_X86_64_GOT32 } }
3382 };
3383 char *cp;
3384 unsigned int j;
3385
3386 for (cp = input_line_pointer; *cp != '@'; cp++)
3387 if (is_end_of_line[(unsigned char) *cp])
3388 return NULL;
3389
3390 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
3391 {
3392 int len;
3393
3394 len = strlen (gotrel[j].str);
3395 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
3396 {
3397 if (gotrel[j].rel[(unsigned int) flag_code] != 0)
3398 {
3399 int first, second;
3400 char *tmpbuf, *past_reloc;
3401
3402 *reloc = gotrel[j].rel[(unsigned int) flag_code];
3403 if (adjust)
3404 *adjust = len;
3405
3406 if (GOT_symbol == NULL)
3407 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3408
3409 /* Replace the relocation token with ' ', so that
3410 errors like foo@GOTOFF1 will be detected. */
3411
3412 /* The length of the first part of our input line. */
3413 first = cp - input_line_pointer;
3414
3415 /* The second part goes from after the reloc token until
3416 (and including) an end_of_line char. Don't use strlen
3417 here as the end_of_line char may not be a NUL. */
3418 past_reloc = cp + 1 + len;
3419 for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
3420 ;
3421 second = cp - past_reloc;
3422
3423 /* Allocate and copy string. The trailing NUL shouldn't
3424 be necessary, but be safe. */
3425 tmpbuf = xmalloc (first + second + 2);
3426 memcpy (tmpbuf, input_line_pointer, first);
3427 tmpbuf[first] = ' ';
3428 memcpy (tmpbuf + first + 1, past_reloc, second);
3429 tmpbuf[first + second + 1] = '\0';
3430 return tmpbuf;
3431 }
3432
3433 as_bad (_("@%s reloc is not supported in %s bit mode"),
3434 gotrel[j].str, mode_name[(unsigned int) flag_code]);
3435 return NULL;
3436 }
3437 }
3438
3439 /* Might be a symbol version string. Don't as_bad here. */
3440 return NULL;
3441 }
3442
3443 /* x86_cons_fix_new is called via the expression parsing code when a
3444 reloc is needed. We use this hook to get the correct .got reloc. */
3445 static RELOC_ENUM got_reloc = NO_RELOC;
3446
3447 void
3448 x86_cons_fix_new (frag, off, len, exp)
3449 fragS *frag;
3450 unsigned int off;
3451 unsigned int len;
3452 expressionS *exp;
3453 {
3454 RELOC_ENUM r = reloc (len, 0, 0, got_reloc);
3455 got_reloc = NO_RELOC;
3456 fix_new_exp (frag, off, len, exp, 0, r);
3457 }
3458
3459 void
3460 x86_cons (exp, size)
3461 expressionS *exp;
3462 int size;
3463 {
3464 if (size == 4)
3465 {
3466 /* Handle @GOTOFF and the like in an expression. */
3467 char *save;
3468 char *gotfree_input_line;
3469 int adjust;
3470
3471 save = input_line_pointer;
3472 gotfree_input_line = lex_got (&got_reloc, &adjust);
3473 if (gotfree_input_line)
3474 input_line_pointer = gotfree_input_line;
3475
3476 expression (exp);
3477
3478 if (gotfree_input_line)
3479 {
3480 /* expression () has merrily parsed up to the end of line,
3481 or a comma - in the wrong buffer. Transfer how far
3482 input_line_pointer has moved to the right buffer. */
3483 input_line_pointer = (save
3484 + (input_line_pointer - gotfree_input_line)
3485 + adjust);
3486 free (gotfree_input_line);
3487 }
3488 }
3489 else
3490 expression (exp);
3491 }
3492 #endif
3493
3494 static int i386_immediate PARAMS ((char *));
3495
3496 static int
3497 i386_immediate (imm_start)
3498 char *imm_start;
3499 {
3500 char *save_input_line_pointer;
3501 #ifndef LEX_AT
3502 char *gotfree_input_line;
3503 #endif
3504 segT exp_seg = 0;
3505 expressionS *exp;
3506
3507 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
3508 {
3509 as_bad (_("only 1 or 2 immediate operands are allowed"));
3510 return 0;
3511 }
3512
3513 exp = &im_expressions[i.imm_operands++];
3514 i.op[this_operand].imms = exp;
3515
3516 if (is_space_char (*imm_start))
3517 ++imm_start;
3518
3519 save_input_line_pointer = input_line_pointer;
3520 input_line_pointer = imm_start;
3521
3522 #ifndef LEX_AT
3523 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3524 if (gotfree_input_line)
3525 input_line_pointer = gotfree_input_line;
3526 #endif
3527
3528 exp_seg = expression (exp);
3529
3530 SKIP_WHITESPACE ();
3531 if (*input_line_pointer)
3532 as_bad (_("junk `%s' after expression"), input_line_pointer);
3533
3534 input_line_pointer = save_input_line_pointer;
3535 #ifndef LEX_AT
3536 if (gotfree_input_line)
3537 free (gotfree_input_line);
3538 #endif
3539
3540 if (exp->X_op == O_absent || exp->X_op == O_big)
3541 {
3542 /* Missing or bad expr becomes absolute 0. */
3543 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
3544 imm_start);
3545 exp->X_op = O_constant;
3546 exp->X_add_number = 0;
3547 exp->X_add_symbol = (symbolS *) 0;
3548 exp->X_op_symbol = (symbolS *) 0;
3549 }
3550 else if (exp->X_op == O_constant)
3551 {
3552 /* Size it properly later. */
3553 i.types[this_operand] |= Imm64;
3554 /* If BFD64, sign extend val. */
3555 if (!use_rela_relocations)
3556 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
3557 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
3558 }
3559 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3560 else if (1
3561 #ifdef BFD_ASSEMBLER
3562 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3563 #endif
3564 && exp_seg != text_section
3565 && exp_seg != data_section
3566 && exp_seg != bss_section
3567 && exp_seg != undefined_section
3568 #ifdef BFD_ASSEMBLER
3569 && !bfd_is_com_section (exp_seg)
3570 #endif
3571 )
3572 {
3573 #ifdef BFD_ASSEMBLER
3574 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3575 #else
3576 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3577 #endif
3578 return 0;
3579 }
3580 #endif
3581 else
3582 {
3583 /* This is an address. The size of the address will be
3584 determined later, depending on destination register,
3585 suffix, or the default for the section. */
3586 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
3587 }
3588
3589 return 1;
3590 }
3591
3592 static char *i386_scale PARAMS ((char *));
3593
3594 static char *
3595 i386_scale (scale)
3596 char *scale;
3597 {
3598 offsetT val;
3599 char *save = input_line_pointer;
3600
3601 input_line_pointer = scale;
3602 val = get_absolute_expression ();
3603
3604 switch (val)
3605 {
3606 case 0:
3607 case 1:
3608 i.log2_scale_factor = 0;
3609 break;
3610 case 2:
3611 i.log2_scale_factor = 1;
3612 break;
3613 case 4:
3614 i.log2_scale_factor = 2;
3615 break;
3616 case 8:
3617 i.log2_scale_factor = 3;
3618 break;
3619 default:
3620 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
3621 scale);
3622 input_line_pointer = save;
3623 return NULL;
3624 }
3625 if (i.log2_scale_factor != 0 && i.index_reg == 0)
3626 {
3627 as_warn (_("scale factor of %d without an index register"),
3628 1 << i.log2_scale_factor);
3629 #if SCALE1_WHEN_NO_INDEX
3630 i.log2_scale_factor = 0;
3631 #endif
3632 }
3633 scale = input_line_pointer;
3634 input_line_pointer = save;
3635 return scale;
3636 }
3637
3638 static int i386_displacement PARAMS ((char *, char *));
3639
3640 static int
3641 i386_displacement (disp_start, disp_end)
3642 char *disp_start;
3643 char *disp_end;
3644 {
3645 expressionS *exp;
3646 segT exp_seg = 0;
3647 char *save_input_line_pointer;
3648 #ifndef LEX_AT
3649 char *gotfree_input_line;
3650 #endif
3651 int bigdisp = Disp32;
3652
3653 if (flag_code == CODE_64BIT)
3654 {
3655 if (i.prefix[ADDR_PREFIX] == 0)
3656 bigdisp = Disp64;
3657 }
3658 else if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3659 bigdisp = Disp16;
3660 i.types[this_operand] |= bigdisp;
3661
3662 exp = &disp_expressions[i.disp_operands];
3663 i.op[this_operand].disps = exp;
3664 i.disp_operands++;
3665 save_input_line_pointer = input_line_pointer;
3666 input_line_pointer = disp_start;
3667 END_STRING_AND_SAVE (disp_end);
3668
3669 #ifndef GCC_ASM_O_HACK
3670 #define GCC_ASM_O_HACK 0
3671 #endif
3672 #if GCC_ASM_O_HACK
3673 END_STRING_AND_SAVE (disp_end + 1);
3674 if ((i.types[this_operand] & BaseIndex) != 0
3675 && displacement_string_end[-1] == '+')
3676 {
3677 /* This hack is to avoid a warning when using the "o"
3678 constraint within gcc asm statements.
3679 For instance:
3680
3681 #define _set_tssldt_desc(n,addr,limit,type) \
3682 __asm__ __volatile__ ( \
3683 "movw %w2,%0\n\t" \
3684 "movw %w1,2+%0\n\t" \
3685 "rorl $16,%1\n\t" \
3686 "movb %b1,4+%0\n\t" \
3687 "movb %4,5+%0\n\t" \
3688 "movb $0,6+%0\n\t" \
3689 "movb %h1,7+%0\n\t" \
3690 "rorl $16,%1" \
3691 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
3692
3693 This works great except that the output assembler ends
3694 up looking a bit weird if it turns out that there is
3695 no offset. You end up producing code that looks like:
3696
3697 #APP
3698 movw $235,(%eax)
3699 movw %dx,2+(%eax)
3700 rorl $16,%edx
3701 movb %dl,4+(%eax)
3702 movb $137,5+(%eax)
3703 movb $0,6+(%eax)
3704 movb %dh,7+(%eax)
3705 rorl $16,%edx
3706 #NO_APP
3707
3708 So here we provide the missing zero. */
3709
3710 *displacement_string_end = '0';
3711 }
3712 #endif
3713 #ifndef LEX_AT
3714 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3715 if (gotfree_input_line)
3716 input_line_pointer = gotfree_input_line;
3717 #endif
3718
3719 exp_seg = expression (exp);
3720
3721 SKIP_WHITESPACE ();
3722 if (*input_line_pointer)
3723 as_bad (_("junk `%s' after expression"), input_line_pointer);
3724 #if GCC_ASM_O_HACK
3725 RESTORE_END_STRING (disp_end + 1);
3726 #endif
3727 RESTORE_END_STRING (disp_end);
3728 input_line_pointer = save_input_line_pointer;
3729 #ifndef LEX_AT
3730 if (gotfree_input_line)
3731 free (gotfree_input_line);
3732 #endif
3733
3734 #ifdef BFD_ASSEMBLER
3735 /* We do this to make sure that the section symbol is in
3736 the symbol table. We will ultimately change the relocation
3737 to be relative to the beginning of the section. */
3738 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
3739 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3740 {
3741 if (exp->X_op != O_symbol)
3742 {
3743 as_bad (_("bad expression used with @%s"),
3744 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
3745 ? "GOTPCREL"
3746 : "GOTOFF"));
3747 return 0;
3748 }
3749
3750 if (S_IS_LOCAL (exp->X_add_symbol)
3751 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
3752 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
3753 exp->X_op = O_subtract;
3754 exp->X_op_symbol = GOT_symbol;
3755 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3756 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
3757 else
3758 i.reloc[this_operand] = BFD_RELOC_32;
3759 }
3760 #endif
3761
3762 if (exp->X_op == O_absent || exp->X_op == O_big)
3763 {
3764 /* Missing or bad expr becomes absolute 0. */
3765 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
3766 disp_start);
3767 exp->X_op = O_constant;
3768 exp->X_add_number = 0;
3769 exp->X_add_symbol = (symbolS *) 0;
3770 exp->X_op_symbol = (symbolS *) 0;
3771 }
3772
3773 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3774 if (exp->X_op != O_constant
3775 #ifdef BFD_ASSEMBLER
3776 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3777 #endif
3778 && exp_seg != text_section
3779 && exp_seg != data_section
3780 && exp_seg != bss_section
3781 && exp_seg != undefined_section)
3782 {
3783 #ifdef BFD_ASSEMBLER
3784 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3785 #else
3786 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3787 #endif
3788 return 0;
3789 }
3790 #endif
3791 else if (flag_code == CODE_64BIT)
3792 i.types[this_operand] |= Disp32S | Disp32;
3793 return 1;
3794 }
3795
3796 static int i386_index_check PARAMS ((const char *));
3797
3798 /* Make sure the memory operand we've been dealt is valid.
3799 Return 1 on success, 0 on a failure. */
3800
3801 static int
3802 i386_index_check (operand_string)
3803 const char *operand_string;
3804 {
3805 int ok;
3806 #if INFER_ADDR_PREFIX
3807 int fudged = 0;
3808
3809 tryprefix:
3810 #endif
3811 ok = 1;
3812 if (flag_code == CODE_64BIT)
3813 {
3814 if (i.prefix[ADDR_PREFIX] == 0)
3815 {
3816 /* 64bit checks. */
3817 if ((i.base_reg
3818 && ((i.base_reg->reg_type & Reg64) == 0)
3819 && (i.base_reg->reg_type != BaseIndex
3820 || i.index_reg))
3821 || (i.index_reg
3822 && ((i.index_reg->reg_type & (Reg64 | BaseIndex))
3823 != (Reg64 | BaseIndex))))
3824 ok = 0;
3825 }
3826 else
3827 {
3828 /* 32bit checks. */
3829 if ((i.base_reg
3830 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
3831 || (i.index_reg
3832 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
3833 != (Reg32 | BaseIndex))))
3834 ok = 0;
3835 }
3836 }
3837 else
3838 {
3839 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3840 {
3841 /* 16bit checks. */
3842 if ((i.base_reg
3843 && ((i.base_reg->reg_type & (Reg16 | BaseIndex | RegRex))
3844 != (Reg16 | BaseIndex)))
3845 || (i.index_reg
3846 && (((i.index_reg->reg_type & (Reg16 | BaseIndex))
3847 != (Reg16 | BaseIndex))
3848 || !(i.base_reg
3849 && i.base_reg->reg_num < 6
3850 && i.index_reg->reg_num >= 6
3851 && i.log2_scale_factor == 0))))
3852 ok = 0;
3853 }
3854 else
3855 {
3856 /* 32bit checks. */
3857 if ((i.base_reg
3858 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
3859 || (i.index_reg
3860 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
3861 != (Reg32 | BaseIndex))))
3862 ok = 0;
3863 }
3864 }
3865 if (!ok)
3866 {
3867 #if INFER_ADDR_PREFIX
3868 if (flag_code != CODE_64BIT
3869 && i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0')
3870 {
3871 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
3872 i.prefixes += 1;
3873 /* Change the size of any displacement too. At most one of
3874 Disp16 or Disp32 is set.
3875 FIXME. There doesn't seem to be any real need for separate
3876 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
3877 Removing them would probably clean up the code quite a lot. */
3878 if (i.types[this_operand] & (Disp16 | Disp32))
3879 i.types[this_operand] ^= (Disp16 | Disp32);
3880 fudged = 1;
3881 goto tryprefix;
3882 }
3883 if (fudged)
3884 as_bad (_("`%s' is not a valid base/index expression"),
3885 operand_string);
3886 else
3887 #endif
3888 as_bad (_("`%s' is not a valid %s bit base/index expression"),
3889 operand_string,
3890 flag_code_names[flag_code]);
3891 return 0;
3892 }
3893 return 1;
3894 }
3895
3896 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
3897 on error. */
3898
3899 static int
3900 i386_operand (operand_string)
3901 char *operand_string;
3902 {
3903 const reg_entry *r;
3904 char *end_op;
3905 char *op_string = operand_string;
3906
3907 if (is_space_char (*op_string))
3908 ++op_string;
3909
3910 /* We check for an absolute prefix (differentiating,
3911 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
3912 if (*op_string == ABSOLUTE_PREFIX)
3913 {
3914 ++op_string;
3915 if (is_space_char (*op_string))
3916 ++op_string;
3917 i.types[this_operand] |= JumpAbsolute;
3918 }
3919
3920 /* Check if operand is a register. */
3921 if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
3922 && (r = parse_register (op_string, &end_op)) != NULL)
3923 {
3924 /* Check for a segment override by searching for ':' after a
3925 segment register. */
3926 op_string = end_op;
3927 if (is_space_char (*op_string))
3928 ++op_string;
3929 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
3930 {
3931 switch (r->reg_num)
3932 {
3933 case 0:
3934 i.seg[i.mem_operands] = &es;
3935 break;
3936 case 1:
3937 i.seg[i.mem_operands] = &cs;
3938 break;
3939 case 2:
3940 i.seg[i.mem_operands] = &ss;
3941 break;
3942 case 3:
3943 i.seg[i.mem_operands] = &ds;
3944 break;
3945 case 4:
3946 i.seg[i.mem_operands] = &fs;
3947 break;
3948 case 5:
3949 i.seg[i.mem_operands] = &gs;
3950 break;
3951 }
3952
3953 /* Skip the ':' and whitespace. */
3954 ++op_string;
3955 if (is_space_char (*op_string))
3956 ++op_string;
3957
3958 if (!is_digit_char (*op_string)
3959 && !is_identifier_char (*op_string)
3960 && *op_string != '('
3961 && *op_string != ABSOLUTE_PREFIX)
3962 {
3963 as_bad (_("bad memory operand `%s'"), op_string);
3964 return 0;
3965 }
3966 /* Handle case of %es:*foo. */
3967 if (*op_string == ABSOLUTE_PREFIX)
3968 {
3969 ++op_string;
3970 if (is_space_char (*op_string))
3971 ++op_string;
3972 i.types[this_operand] |= JumpAbsolute;
3973 }
3974 goto do_memory_reference;
3975 }
3976 if (*op_string)
3977 {
3978 as_bad (_("junk `%s' after register"), op_string);
3979 return 0;
3980 }
3981 i.types[this_operand] |= r->reg_type & ~BaseIndex;
3982 i.op[this_operand].regs = r;
3983 i.reg_operands++;
3984 }
3985 else if (*op_string == REGISTER_PREFIX)
3986 {
3987 as_bad (_("bad register name `%s'"), op_string);
3988 return 0;
3989 }
3990 else if (*op_string == IMMEDIATE_PREFIX)
3991 {
3992 ++op_string;
3993 if (i.types[this_operand] & JumpAbsolute)
3994 {
3995 as_bad (_("immediate operand illegal with absolute jump"));
3996 return 0;
3997 }
3998 if (!i386_immediate (op_string))
3999 return 0;
4000 }
4001 else if (is_digit_char (*op_string)
4002 || is_identifier_char (*op_string)
4003 || *op_string == '(')
4004 {
4005 /* This is a memory reference of some sort. */
4006 char *base_string;
4007
4008 /* Start and end of displacement string expression (if found). */
4009 char *displacement_string_start;
4010 char *displacement_string_end;
4011
4012 do_memory_reference:
4013 if ((i.mem_operands == 1
4014 && (current_templates->start->opcode_modifier & IsString) == 0)
4015 || i.mem_operands == 2)
4016 {
4017 as_bad (_("too many memory references for `%s'"),
4018 current_templates->start->name);
4019 return 0;
4020 }
4021
4022 /* Check for base index form. We detect the base index form by
4023 looking for an ')' at the end of the operand, searching
4024 for the '(' matching it, and finding a REGISTER_PREFIX or ','
4025 after the '('. */
4026 base_string = op_string + strlen (op_string);
4027
4028 --base_string;
4029 if (is_space_char (*base_string))
4030 --base_string;
4031
4032 /* If we only have a displacement, set-up for it to be parsed later. */
4033 displacement_string_start = op_string;
4034 displacement_string_end = base_string + 1;
4035
4036 if (*base_string == ')')
4037 {
4038 char *temp_string;
4039 unsigned int parens_balanced = 1;
4040 /* We've already checked that the number of left & right ()'s are
4041 equal, so this loop will not be infinite. */
4042 do
4043 {
4044 base_string--;
4045 if (*base_string == ')')
4046 parens_balanced++;
4047 if (*base_string == '(')
4048 parens_balanced--;
4049 }
4050 while (parens_balanced);
4051
4052 temp_string = base_string;
4053
4054 /* Skip past '(' and whitespace. */
4055 ++base_string;
4056 if (is_space_char (*base_string))
4057 ++base_string;
4058
4059 if (*base_string == ','
4060 || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
4061 && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
4062 {
4063 displacement_string_end = temp_string;
4064
4065 i.types[this_operand] |= BaseIndex;
4066
4067 if (i.base_reg)
4068 {
4069 base_string = end_op;
4070 if (is_space_char (*base_string))
4071 ++base_string;
4072 }
4073
4074 /* There may be an index reg or scale factor here. */
4075 if (*base_string == ',')
4076 {
4077 ++base_string;
4078 if (is_space_char (*base_string))
4079 ++base_string;
4080
4081 if ((*base_string == REGISTER_PREFIX || allow_naked_reg)
4082 && (i.index_reg = parse_register (base_string, &end_op)) != NULL)
4083 {
4084 base_string = end_op;
4085 if (is_space_char (*base_string))
4086 ++base_string;
4087 if (*base_string == ',')
4088 {
4089 ++base_string;
4090 if (is_space_char (*base_string))
4091 ++base_string;
4092 }
4093 else if (*base_string != ')')
4094 {
4095 as_bad (_("expecting `,' or `)' after index register in `%s'"),
4096 operand_string);
4097 return 0;
4098 }
4099 }
4100 else if (*base_string == REGISTER_PREFIX)
4101 {
4102 as_bad (_("bad register name `%s'"), base_string);
4103 return 0;
4104 }
4105
4106 /* Check for scale factor. */
4107 if (*base_string != ')')
4108 {
4109 char *end_scale = i386_scale (base_string);
4110
4111 if (!end_scale)
4112 return 0;
4113
4114 base_string = end_scale;
4115 if (is_space_char (*base_string))
4116 ++base_string;
4117 if (*base_string != ')')
4118 {
4119 as_bad (_("expecting `)' after scale factor in `%s'"),
4120 operand_string);
4121 return 0;
4122 }
4123 }
4124 else if (!i.index_reg)
4125 {
4126 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
4127 *base_string);
4128 return 0;
4129 }
4130 }
4131 else if (*base_string != ')')
4132 {
4133 as_bad (_("expecting `,' or `)' after base register in `%s'"),
4134 operand_string);
4135 return 0;
4136 }
4137 }
4138 else if (*base_string == REGISTER_PREFIX)
4139 {
4140 as_bad (_("bad register name `%s'"), base_string);
4141 return 0;
4142 }
4143 }
4144
4145 /* If there's an expression beginning the operand, parse it,
4146 assuming displacement_string_start and
4147 displacement_string_end are meaningful. */
4148 if (displacement_string_start != displacement_string_end)
4149 {
4150 if (!i386_displacement (displacement_string_start,
4151 displacement_string_end))
4152 return 0;
4153 }
4154
4155 /* Special case for (%dx) while doing input/output op. */
4156 if (i.base_reg
4157 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
4158 && i.index_reg == 0
4159 && i.log2_scale_factor == 0
4160 && i.seg[i.mem_operands] == 0
4161 && (i.types[this_operand] & Disp) == 0)
4162 {
4163 i.types[this_operand] = InOutPortReg;
4164 return 1;
4165 }
4166
4167 if (i386_index_check (operand_string) == 0)
4168 return 0;
4169 i.mem_operands++;
4170 }
4171 else
4172 {
4173 /* It's not a memory operand; argh! */
4174 as_bad (_("invalid char %s beginning operand %d `%s'"),
4175 output_invalid (*op_string),
4176 this_operand + 1,
4177 op_string);
4178 return 0;
4179 }
4180 return 1; /* Normal return. */
4181 }
4182 \f
4183 /* md_estimate_size_before_relax()
4184
4185 Called just before relax() for rs_machine_dependent frags. The x86
4186 assembler uses these frags to handle variable size jump
4187 instructions.
4188
4189 Any symbol that is now undefined will not become defined.
4190 Return the correct fr_subtype in the frag.
4191 Return the initial "guess for variable size of frag" to caller.
4192 The guess is actually the growth beyond the fixed part. Whatever
4193 we do to grow the fixed or variable part contributes to our
4194 returned value. */
4195
4196 int
4197 md_estimate_size_before_relax (fragP, segment)
4198 fragS *fragP;
4199 segT segment;
4200 {
4201 /* We've already got fragP->fr_subtype right; all we have to do is
4202 check for un-relaxable symbols. On an ELF system, we can't relax
4203 an externally visible symbol, because it may be overridden by a
4204 shared library. */
4205 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
4206 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4207 || S_IS_EXTERNAL (fragP->fr_symbol)
4208 || S_IS_WEAK (fragP->fr_symbol)
4209 #endif
4210 )
4211 {
4212 /* Symbol is undefined in this segment, or we need to keep a
4213 reloc so that weak symbols can be overridden. */
4214 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
4215 RELOC_ENUM reloc_type;
4216 unsigned char *opcode;
4217 int old_fr_fix;
4218 fixS *fixP;
4219
4220 if (fragP->fr_var != NO_RELOC)
4221 reloc_type = fragP->fr_var;
4222 else if (size == 2)
4223 reloc_type = BFD_RELOC_16_PCREL;
4224 else
4225 reloc_type = BFD_RELOC_32_PCREL;
4226
4227 old_fr_fix = fragP->fr_fix;
4228 opcode = (unsigned char *) fragP->fr_opcode;
4229
4230 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
4231 {
4232 case UNCOND_JUMP:
4233 /* Make jmp (0xeb) a (d)word displacement jump. */
4234 opcode[0] = 0xe9;
4235 fragP->fr_fix += size;
4236 fixP = fix_new (fragP, old_fr_fix, size,
4237 fragP->fr_symbol,
4238 fragP->fr_offset, 1,
4239 reloc_type);
4240 fixP->fx_pcrel_adjust = size;
4241 break;
4242
4243 case COND_JUMP86:
4244 if (no_cond_jump_promotion)
4245 goto relax_guess;
4246
4247 if (size == 2)
4248 {
4249 /* Negate the condition, and branch past an
4250 unconditional jump. */
4251 opcode[0] ^= 1;
4252 opcode[1] = 3;
4253 /* Insert an unconditional jump. */
4254 opcode[2] = 0xe9;
4255 /* We added two extra opcode bytes, and have a two byte
4256 offset. */
4257 fragP->fr_fix += 2 + 2;
4258 fixP = fix_new (fragP, old_fr_fix + 2, 2,
4259 fragP->fr_symbol,
4260 fragP->fr_offset, 1,
4261 reloc_type);
4262 fixP->fx_pcrel_adjust = size;
4263 break;
4264 }
4265 /* Fall through. */
4266
4267 case COND_JUMP:
4268 if (no_cond_jump_promotion)
4269 goto relax_guess;
4270
4271 /* This changes the byte-displacement jump 0x7N
4272 to the (d)word-displacement jump 0x0f,0x8N. */
4273 opcode[1] = opcode[0] + 0x10;
4274 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4275 /* We've added an opcode byte. */
4276 fragP->fr_fix += 1 + size;
4277 fixP = fix_new (fragP, old_fr_fix + 1, size,
4278 fragP->fr_symbol,
4279 fragP->fr_offset, 1,
4280 reloc_type);
4281 fixP->fx_pcrel_adjust = size;
4282 break;
4283
4284 default:
4285 BAD_CASE (fragP->fr_subtype);
4286 break;
4287 }
4288 frag_wane (fragP);
4289 return fragP->fr_fix - old_fr_fix;
4290 }
4291
4292 relax_guess:
4293 /* Guess size depending on current relax state. Initially the relax
4294 state will correspond to a short jump and we return 1, because
4295 the variable part of the frag (the branch offset) is one byte
4296 long. However, we can relax a section more than once and in that
4297 case we must either set fr_subtype back to the unrelaxed state,
4298 or return the value for the appropriate branch. */
4299 return md_relax_table[fragP->fr_subtype].rlx_length;
4300 }
4301
4302 /* Called after relax() is finished.
4303
4304 In: Address of frag.
4305 fr_type == rs_machine_dependent.
4306 fr_subtype is what the address relaxed to.
4307
4308 Out: Any fixSs and constants are set up.
4309 Caller will turn frag into a ".space 0". */
4310
4311 #ifndef BFD_ASSEMBLER
4312 void
4313 md_convert_frag (headers, sec, fragP)
4314 object_headers *headers ATTRIBUTE_UNUSED;
4315 segT sec ATTRIBUTE_UNUSED;
4316 fragS *fragP;
4317 #else
4318 void
4319 md_convert_frag (abfd, sec, fragP)
4320 bfd *abfd ATTRIBUTE_UNUSED;
4321 segT sec ATTRIBUTE_UNUSED;
4322 fragS *fragP;
4323 #endif
4324 {
4325 unsigned char *opcode;
4326 unsigned char *where_to_put_displacement = NULL;
4327 offsetT target_address;
4328 offsetT opcode_address;
4329 unsigned int extension = 0;
4330 offsetT displacement_from_opcode_start;
4331
4332 opcode = (unsigned char *) fragP->fr_opcode;
4333
4334 /* Address we want to reach in file space. */
4335 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
4336
4337 /* Address opcode resides at in file space. */
4338 opcode_address = fragP->fr_address + fragP->fr_fix;
4339
4340 /* Displacement from opcode start to fill into instruction. */
4341 displacement_from_opcode_start = target_address - opcode_address;
4342
4343 if ((fragP->fr_subtype & BIG) == 0)
4344 {
4345 /* Don't have to change opcode. */
4346 extension = 1; /* 1 opcode + 1 displacement */
4347 where_to_put_displacement = &opcode[1];
4348 }
4349 else
4350 {
4351 if (no_cond_jump_promotion
4352 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
4353 as_warn_where (fragP->fr_file, fragP->fr_line, _("long jump required"));
4354
4355 switch (fragP->fr_subtype)
4356 {
4357 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
4358 extension = 4; /* 1 opcode + 4 displacement */
4359 opcode[0] = 0xe9;
4360 where_to_put_displacement = &opcode[1];
4361 break;
4362
4363 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
4364 extension = 2; /* 1 opcode + 2 displacement */
4365 opcode[0] = 0xe9;
4366 where_to_put_displacement = &opcode[1];
4367 break;
4368
4369 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
4370 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
4371 extension = 5; /* 2 opcode + 4 displacement */
4372 opcode[1] = opcode[0] + 0x10;
4373 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4374 where_to_put_displacement = &opcode[2];
4375 break;
4376
4377 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
4378 extension = 3; /* 2 opcode + 2 displacement */
4379 opcode[1] = opcode[0] + 0x10;
4380 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4381 where_to_put_displacement = &opcode[2];
4382 break;
4383
4384 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
4385 extension = 4;
4386 opcode[0] ^= 1;
4387 opcode[1] = 3;
4388 opcode[2] = 0xe9;
4389 where_to_put_displacement = &opcode[3];
4390 break;
4391
4392 default:
4393 BAD_CASE (fragP->fr_subtype);
4394 break;
4395 }
4396 }
4397
4398 /* Now put displacement after opcode. */
4399 md_number_to_chars ((char *) where_to_put_displacement,
4400 (valueT) (displacement_from_opcode_start - extension),
4401 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
4402 fragP->fr_fix += extension;
4403 }
4404 \f
4405 /* Size of byte displacement jmp. */
4406 int md_short_jump_size = 2;
4407
4408 /* Size of dword displacement jmp. */
4409 int md_long_jump_size = 5;
4410
4411 /* Size of relocation record. */
4412 const int md_reloc_size = 8;
4413
4414 void
4415 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
4416 char *ptr;
4417 addressT from_addr, to_addr;
4418 fragS *frag ATTRIBUTE_UNUSED;
4419 symbolS *to_symbol ATTRIBUTE_UNUSED;
4420 {
4421 offsetT offset;
4422
4423 offset = to_addr - (from_addr + 2);
4424 /* Opcode for byte-disp jump. */
4425 md_number_to_chars (ptr, (valueT) 0xeb, 1);
4426 md_number_to_chars (ptr + 1, (valueT) offset, 1);
4427 }
4428
4429 void
4430 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
4431 char *ptr;
4432 addressT from_addr, to_addr;
4433 fragS *frag ATTRIBUTE_UNUSED;
4434 symbolS *to_symbol ATTRIBUTE_UNUSED;
4435 {
4436 offsetT offset;
4437
4438 offset = to_addr - (from_addr + 5);
4439 md_number_to_chars (ptr, (valueT) 0xe9, 1);
4440 md_number_to_chars (ptr + 1, (valueT) offset, 4);
4441 }
4442 \f
4443 /* Apply a fixup (fixS) to segment data, once it has been determined
4444 by our caller that we have all the info we need to fix it up.
4445
4446 On the 386, immediates, displacements, and data pointers are all in
4447 the same (little-endian) format, so we don't need to care about which
4448 we are handling. */
4449
4450 void
4451 md_apply_fix3 (fixP, valP, seg)
4452 /* The fix we're to put in. */
4453 fixS *fixP;
4454 /* Pointer to the value of the bits. */
4455 valueT * valP;
4456 /* Segment fix is from. */
4457 segT seg ATTRIBUTE_UNUSED;
4458 {
4459 char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
4460 valueT value = * valP;
4461
4462 #if defined (BFD_ASSEMBLER) && !defined (TE_Mach)
4463 if (fixP->fx_pcrel)
4464 {
4465 switch (fixP->fx_r_type)
4466 {
4467 default:
4468 break;
4469
4470 case BFD_RELOC_32:
4471 fixP->fx_r_type = BFD_RELOC_32_PCREL;
4472 break;
4473 case BFD_RELOC_16:
4474 fixP->fx_r_type = BFD_RELOC_16_PCREL;
4475 break;
4476 case BFD_RELOC_8:
4477 fixP->fx_r_type = BFD_RELOC_8_PCREL;
4478 break;
4479 }
4480 }
4481
4482 /* This is a hack. There should be a better way to handle this.
4483 This covers for the fact that bfd_install_relocation will
4484 subtract the current location (for partial_inplace, PC relative
4485 relocations); see more below. */
4486 if ((fixP->fx_r_type == BFD_RELOC_32_PCREL
4487 || fixP->fx_r_type == BFD_RELOC_16_PCREL
4488 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
4489 && fixP->fx_addsy && !use_rela_relocations)
4490 {
4491 #ifndef OBJ_AOUT
4492 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4493 #ifdef TE_PE
4494 || OUTPUT_FLAVOR == bfd_target_coff_flavour
4495 #endif
4496 )
4497 value += fixP->fx_where + fixP->fx_frag->fr_address;
4498 #endif
4499 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4500 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
4501 {
4502 segT fseg = S_GET_SEGMENT (fixP->fx_addsy);
4503
4504 if ((fseg == seg
4505 || (symbol_section_p (fixP->fx_addsy)
4506 && fseg != absolute_section))
4507 && !S_IS_EXTERNAL (fixP->fx_addsy)
4508 && !S_IS_WEAK (fixP->fx_addsy)
4509 && S_IS_DEFINED (fixP->fx_addsy)
4510 && !S_IS_COMMON (fixP->fx_addsy))
4511 {
4512 /* Yes, we add the values in twice. This is because
4513 bfd_perform_relocation subtracts them out again. I think
4514 bfd_perform_relocation is broken, but I don't dare change
4515 it. FIXME. */
4516 value += fixP->fx_where + fixP->fx_frag->fr_address;
4517 }
4518 }
4519 #endif
4520 #if defined (OBJ_COFF) && defined (TE_PE)
4521 /* For some reason, the PE format does not store a section
4522 address offset for a PC relative symbol. */
4523 if (S_GET_SEGMENT (fixP->fx_addsy) != seg)
4524 value += md_pcrel_from (fixP);
4525 #endif
4526 }
4527
4528 /* Fix a few things - the dynamic linker expects certain values here,
4529 and we must not dissappoint it. */
4530 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4531 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4532 && fixP->fx_addsy)
4533 switch (fixP->fx_r_type)
4534 {
4535 case BFD_RELOC_386_PLT32:
4536 case BFD_RELOC_X86_64_PLT32:
4537 /* Make the jump instruction point to the address of the operand. At
4538 runtime we merely add the offset to the actual PLT entry. */
4539 value = -4;
4540 break;
4541 case BFD_RELOC_386_GOTPC:
4542
4543 /* This is tough to explain. We end up with this one if we have
4544 * operands that look like "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal
4545 * here is to obtain the absolute address of the GOT, and it is strongly
4546 * preferable from a performance point of view to avoid using a runtime
4547 * relocation for this. The actual sequence of instructions often look
4548 * something like:
4549 *
4550 * call .L66
4551 * .L66:
4552 * popl %ebx
4553 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
4554 *
4555 * The call and pop essentially return the absolute address of
4556 * the label .L66 and store it in %ebx. The linker itself will
4557 * ultimately change the first operand of the addl so that %ebx points to
4558 * the GOT, but to keep things simple, the .o file must have this operand
4559 * set so that it generates not the absolute address of .L66, but the
4560 * absolute address of itself. This allows the linker itself simply
4561 * treat a GOTPC relocation as asking for a pcrel offset to the GOT to be
4562 * added in, and the addend of the relocation is stored in the operand
4563 * field for the instruction itself.
4564 *
4565 * Our job here is to fix the operand so that it would add the correct
4566 * offset so that %ebx would point to itself. The thing that is tricky is
4567 * that .-.L66 will point to the beginning of the instruction, so we need
4568 * to further modify the operand so that it will point to itself.
4569 * There are other cases where you have something like:
4570 *
4571 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
4572 *
4573 * and here no correction would be required. Internally in the assembler
4574 * we treat operands of this form as not being pcrel since the '.' is
4575 * explicitly mentioned, and I wonder whether it would simplify matters
4576 * to do it this way. Who knows. In earlier versions of the PIC patches,
4577 * the pcrel_adjust field was used to store the correction, but since the
4578 * expression is not pcrel, I felt it would be confusing to do it this
4579 * way. */
4580
4581 value -= 1;
4582 break;
4583 case BFD_RELOC_386_GOT32:
4584 case BFD_RELOC_X86_64_GOT32:
4585 value = 0; /* Fully resolved at runtime. No addend. */
4586 break;
4587 case BFD_RELOC_386_GOTOFF:
4588 case BFD_RELOC_X86_64_GOTPCREL:
4589 break;
4590
4591 case BFD_RELOC_VTABLE_INHERIT:
4592 case BFD_RELOC_VTABLE_ENTRY:
4593 fixP->fx_done = 0;
4594 return;
4595
4596 default:
4597 break;
4598 }
4599 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
4600 * valP = value;
4601 #endif /* defined (BFD_ASSEMBLER) && !defined (TE_Mach) */
4602
4603 /* Are we finished with this relocation now? */
4604 if (fixP->fx_addsy == NULL && fixP->fx_pcrel == 0)
4605 fixP->fx_done = 1;
4606 #ifdef BFD_ASSEMBLER
4607 else if (use_rela_relocations)
4608 {
4609 fixP->fx_no_overflow = 1;
4610 value = 0;
4611 }
4612 #endif
4613 md_number_to_chars (p, value, fixP->fx_size);
4614 }
4615 \f
4616 #define MAX_LITTLENUMS 6
4617
4618 /* Turn the string pointed to by litP into a floating point constant
4619 of type TYPE, and emit the appropriate bytes. The number of
4620 LITTLENUMS emitted is stored in *SIZEP. An error message is
4621 returned, or NULL on OK. */
4622
4623 char *
4624 md_atof (type, litP, sizeP)
4625 int type;
4626 char *litP;
4627 int *sizeP;
4628 {
4629 int prec;
4630 LITTLENUM_TYPE words[MAX_LITTLENUMS];
4631 LITTLENUM_TYPE *wordP;
4632 char *t;
4633
4634 switch (type)
4635 {
4636 case 'f':
4637 case 'F':
4638 prec = 2;
4639 break;
4640
4641 case 'd':
4642 case 'D':
4643 prec = 4;
4644 break;
4645
4646 case 'x':
4647 case 'X':
4648 prec = 5;
4649 break;
4650
4651 default:
4652 *sizeP = 0;
4653 return _("Bad call to md_atof ()");
4654 }
4655 t = atof_ieee (input_line_pointer, type, words);
4656 if (t)
4657 input_line_pointer = t;
4658
4659 *sizeP = prec * sizeof (LITTLENUM_TYPE);
4660 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
4661 the bigendian 386. */
4662 for (wordP = words + prec - 1; prec--;)
4663 {
4664 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
4665 litP += sizeof (LITTLENUM_TYPE);
4666 }
4667 return 0;
4668 }
4669 \f
4670 char output_invalid_buf[8];
4671
4672 static char *
4673 output_invalid (c)
4674 int c;
4675 {
4676 if (ISPRINT (c))
4677 sprintf (output_invalid_buf, "'%c'", c);
4678 else
4679 sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
4680 return output_invalid_buf;
4681 }
4682
4683 /* REG_STRING starts *before* REGISTER_PREFIX. */
4684
4685 static const reg_entry *
4686 parse_register (reg_string, end_op)
4687 char *reg_string;
4688 char **end_op;
4689 {
4690 char *s = reg_string;
4691 char *p;
4692 char reg_name_given[MAX_REG_NAME_SIZE + 1];
4693 const reg_entry *r;
4694
4695 /* Skip possible REGISTER_PREFIX and possible whitespace. */
4696 if (*s == REGISTER_PREFIX)
4697 ++s;
4698
4699 if (is_space_char (*s))
4700 ++s;
4701
4702 p = reg_name_given;
4703 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
4704 {
4705 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
4706 return (const reg_entry *) NULL;
4707 s++;
4708 }
4709
4710 /* For naked regs, make sure that we are not dealing with an identifier.
4711 This prevents confusing an identifier like `eax_var' with register
4712 `eax'. */
4713 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
4714 return (const reg_entry *) NULL;
4715
4716 *end_op = s;
4717
4718 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
4719
4720 /* Handle floating point regs, allowing spaces in the (i) part. */
4721 if (r == i386_regtab /* %st is first entry of table */)
4722 {
4723 if (is_space_char (*s))
4724 ++s;
4725 if (*s == '(')
4726 {
4727 ++s;
4728 if (is_space_char (*s))
4729 ++s;
4730 if (*s >= '0' && *s <= '7')
4731 {
4732 r = &i386_float_regtab[*s - '0'];
4733 ++s;
4734 if (is_space_char (*s))
4735 ++s;
4736 if (*s == ')')
4737 {
4738 *end_op = s + 1;
4739 return r;
4740 }
4741 }
4742 /* We have "%st(" then garbage. */
4743 return (const reg_entry *) NULL;
4744 }
4745 }
4746
4747 if (r != NULL
4748 && (r->reg_flags & (RegRex64 | RegRex)) != 0
4749 && flag_code != CODE_64BIT)
4750 {
4751 return (const reg_entry *) NULL;
4752 }
4753
4754 return r;
4755 }
4756 \f
4757 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4758 const char *md_shortopts = "kVQ:sq";
4759 #else
4760 const char *md_shortopts = "q";
4761 #endif
4762
4763 struct option md_longopts[] = {
4764 #define OPTION_32 (OPTION_MD_BASE + 0)
4765 {"32", no_argument, NULL, OPTION_32},
4766 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4767 #define OPTION_64 (OPTION_MD_BASE + 1)
4768 {"64", no_argument, NULL, OPTION_64},
4769 #endif
4770 {NULL, no_argument, NULL, 0}
4771 };
4772 size_t md_longopts_size = sizeof (md_longopts);
4773
4774 int
4775 md_parse_option (c, arg)
4776 int c;
4777 char *arg ATTRIBUTE_UNUSED;
4778 {
4779 switch (c)
4780 {
4781 case 'q':
4782 quiet_warnings = 1;
4783 break;
4784
4785 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4786 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
4787 should be emitted or not. FIXME: Not implemented. */
4788 case 'Q':
4789 break;
4790
4791 /* -V: SVR4 argument to print version ID. */
4792 case 'V':
4793 print_version_id ();
4794 break;
4795
4796 /* -k: Ignore for FreeBSD compatibility. */
4797 case 'k':
4798 break;
4799
4800 case 's':
4801 /* -s: On i386 Solaris, this tells the native assembler to use
4802 .stab instead of .stab.excl. We always use .stab anyhow. */
4803 break;
4804
4805 case OPTION_64:
4806 {
4807 const char **list, **l;
4808
4809 list = bfd_target_list ();
4810 for (l = list; *l != NULL; l++)
4811 if (strcmp (*l, "elf64-x86-64") == 0)
4812 {
4813 default_arch = "x86_64";
4814 break;
4815 }
4816 if (*l == NULL)
4817 as_fatal (_("No compiled in support for x86_64"));
4818 free (list);
4819 }
4820 break;
4821 #endif
4822
4823 case OPTION_32:
4824 default_arch = "i386";
4825 break;
4826
4827 default:
4828 return 0;
4829 }
4830 return 1;
4831 }
4832
4833 void
4834 md_show_usage (stream)
4835 FILE *stream;
4836 {
4837 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4838 fprintf (stream, _("\
4839 -Q ignored\n\
4840 -V print assembler version number\n\
4841 -k ignored\n\
4842 -q quieten some warnings\n\
4843 -s ignored\n"));
4844 #else
4845 fprintf (stream, _("\
4846 -q quieten some warnings\n"));
4847 #endif
4848 }
4849
4850 #ifdef BFD_ASSEMBLER
4851 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
4852 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
4853
4854 /* Pick the target format to use. */
4855
4856 const char *
4857 i386_target_format ()
4858 {
4859 if (!strcmp (default_arch, "x86_64"))
4860 set_code_flag (CODE_64BIT);
4861 else if (!strcmp (default_arch, "i386"))
4862 set_code_flag (CODE_32BIT);
4863 else
4864 as_fatal (_("Unknown architecture"));
4865 switch (OUTPUT_FLAVOR)
4866 {
4867 #ifdef OBJ_MAYBE_AOUT
4868 case bfd_target_aout_flavour:
4869 return AOUT_TARGET_FORMAT;
4870 #endif
4871 #ifdef OBJ_MAYBE_COFF
4872 case bfd_target_coff_flavour:
4873 return "coff-i386";
4874 #endif
4875 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
4876 case bfd_target_elf_flavour:
4877 {
4878 if (flag_code == CODE_64BIT)
4879 use_rela_relocations = 1;
4880 return flag_code == CODE_64BIT ? "elf64-x86-64" : "elf32-i386";
4881 }
4882 #endif
4883 default:
4884 abort ();
4885 return NULL;
4886 }
4887 }
4888
4889 #endif /* OBJ_MAYBE_ more than one */
4890
4891 #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
4892 void i386_elf_emit_arch_note ()
4893 {
4894 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4895 && cpu_arch_name != NULL)
4896 {
4897 char *p;
4898 asection *seg = now_seg;
4899 subsegT subseg = now_subseg;
4900 Elf_Internal_Note i_note;
4901 Elf_External_Note e_note;
4902 asection *note_secp;
4903 int len;
4904
4905 /* Create the .note section. */
4906 note_secp = subseg_new (".note", 0);
4907 bfd_set_section_flags (stdoutput,
4908 note_secp,
4909 SEC_HAS_CONTENTS | SEC_READONLY);
4910
4911 /* Process the arch string. */
4912 len = strlen (cpu_arch_name);
4913
4914 i_note.namesz = len + 1;
4915 i_note.descsz = 0;
4916 i_note.type = NT_ARCH;
4917 p = frag_more (sizeof (e_note.namesz));
4918 md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
4919 p = frag_more (sizeof (e_note.descsz));
4920 md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
4921 p = frag_more (sizeof (e_note.type));
4922 md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
4923 p = frag_more (len + 1);
4924 strcpy (p, cpu_arch_name);
4925
4926 frag_align (2, 0, 0);
4927
4928 subseg_set (seg, subseg);
4929 }
4930 }
4931 #endif
4932 #endif /* BFD_ASSEMBLER */
4933 \f
4934 symbolS *
4935 md_undefined_symbol (name)
4936 char *name;
4937 {
4938 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
4939 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
4940 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
4941 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
4942 {
4943 if (!GOT_symbol)
4944 {
4945 if (symbol_find (name))
4946 as_bad (_("GOT already in symbol table"));
4947 GOT_symbol = symbol_new (name, undefined_section,
4948 (valueT) 0, &zero_address_frag);
4949 };
4950 return GOT_symbol;
4951 }
4952 return 0;
4953 }
4954
4955 /* Round up a section size to the appropriate boundary. */
4956
4957 valueT
4958 md_section_align (segment, size)
4959 segT segment ATTRIBUTE_UNUSED;
4960 valueT size;
4961 {
4962 #ifdef BFD_ASSEMBLER
4963 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4964 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
4965 {
4966 /* For a.out, force the section size to be aligned. If we don't do
4967 this, BFD will align it for us, but it will not write out the
4968 final bytes of the section. This may be a bug in BFD, but it is
4969 easier to fix it here since that is how the other a.out targets
4970 work. */
4971 int align;
4972
4973 align = bfd_get_section_alignment (stdoutput, segment);
4974 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
4975 }
4976 #endif
4977 #endif
4978
4979 return size;
4980 }
4981
4982 /* On the i386, PC-relative offsets are relative to the start of the
4983 next instruction. That is, the address of the offset, plus its
4984 size, since the offset is always the last part of the insn. */
4985
4986 long
4987 md_pcrel_from (fixP)
4988 fixS *fixP;
4989 {
4990 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
4991 }
4992
4993 #ifndef I386COFF
4994
4995 static void
4996 s_bss (ignore)
4997 int ignore ATTRIBUTE_UNUSED;
4998 {
4999 int temp;
5000
5001 temp = get_absolute_expression ();
5002 subseg_set (bss_section, (subsegT) temp);
5003 demand_empty_rest_of_line ();
5004 }
5005
5006 #endif
5007
5008 #ifdef BFD_ASSEMBLER
5009
5010 void
5011 i386_validate_fix (fixp)
5012 fixS *fixp;
5013 {
5014 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
5015 {
5016 /* GOTOFF relocation are nonsense in 64bit mode. */
5017 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
5018 {
5019 if (flag_code != CODE_64BIT)
5020 abort ();
5021 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
5022 }
5023 else
5024 {
5025 if (flag_code == CODE_64BIT)
5026 abort ();
5027 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
5028 }
5029 fixp->fx_subsy = 0;
5030 }
5031 }
5032
5033 arelent *
5034 tc_gen_reloc (section, fixp)
5035 asection *section ATTRIBUTE_UNUSED;
5036 fixS *fixp;
5037 {
5038 arelent *rel;
5039 bfd_reloc_code_real_type code;
5040
5041 switch (fixp->fx_r_type)
5042 {
5043 case BFD_RELOC_X86_64_PLT32:
5044 case BFD_RELOC_X86_64_GOT32:
5045 case BFD_RELOC_X86_64_GOTPCREL:
5046 case BFD_RELOC_386_PLT32:
5047 case BFD_RELOC_386_GOT32:
5048 case BFD_RELOC_386_GOTOFF:
5049 case BFD_RELOC_386_GOTPC:
5050 case BFD_RELOC_X86_64_32S:
5051 case BFD_RELOC_RVA:
5052 case BFD_RELOC_VTABLE_ENTRY:
5053 case BFD_RELOC_VTABLE_INHERIT:
5054 code = fixp->fx_r_type;
5055 break;
5056 default:
5057 if (fixp->fx_pcrel)
5058 {
5059 switch (fixp->fx_size)
5060 {
5061 default:
5062 as_bad_where (fixp->fx_file, fixp->fx_line,
5063 _("can not do %d byte pc-relative relocation"),
5064 fixp->fx_size);
5065 code = BFD_RELOC_32_PCREL;
5066 break;
5067 case 1: code = BFD_RELOC_8_PCREL; break;
5068 case 2: code = BFD_RELOC_16_PCREL; break;
5069 case 4: code = BFD_RELOC_32_PCREL; break;
5070 }
5071 }
5072 else
5073 {
5074 switch (fixp->fx_size)
5075 {
5076 default:
5077 as_bad_where (fixp->fx_file, fixp->fx_line,
5078 _("can not do %d byte relocation"),
5079 fixp->fx_size);
5080 code = BFD_RELOC_32;
5081 break;
5082 case 1: code = BFD_RELOC_8; break;
5083 case 2: code = BFD_RELOC_16; break;
5084 case 4: code = BFD_RELOC_32; break;
5085 #ifdef BFD64
5086 case 8: code = BFD_RELOC_64; break;
5087 #endif
5088 }
5089 }
5090 break;
5091 }
5092
5093 if (code == BFD_RELOC_32
5094 && GOT_symbol
5095 && fixp->fx_addsy == GOT_symbol)
5096 {
5097 /* We don't support GOTPC on 64bit targets. */
5098 if (flag_code == CODE_64BIT)
5099 abort ();
5100 code = BFD_RELOC_386_GOTPC;
5101 }
5102
5103 rel = (arelent *) xmalloc (sizeof (arelent));
5104 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
5105 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
5106
5107 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
5108 if (!use_rela_relocations)
5109 {
5110 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
5111 vtable entry to be used in the relocation's section offset. */
5112 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
5113 rel->address = fixp->fx_offset;
5114
5115 if (fixp->fx_pcrel)
5116 rel->addend = fixp->fx_addnumber;
5117 else
5118 rel->addend = 0;
5119 }
5120 /* Use the rela in 64bit mode. */
5121 else
5122 {
5123 rel->addend = fixp->fx_offset;
5124 if (fixp->fx_pcrel)
5125 rel->addend -= fixp->fx_pcrel_adjust;
5126 }
5127
5128 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
5129 if (rel->howto == NULL)
5130 {
5131 as_bad_where (fixp->fx_file, fixp->fx_line,
5132 _("cannot represent relocation type %s"),
5133 bfd_get_reloc_code_name (code));
5134 /* Set howto to a garbage value so that we can keep going. */
5135 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
5136 assert (rel->howto != NULL);
5137 }
5138
5139 return rel;
5140 }
5141
5142 #else /* !BFD_ASSEMBLER */
5143
5144 #if (defined(OBJ_AOUT) | defined(OBJ_BOUT))
5145 void
5146 tc_aout_fix_to_chars (where, fixP, segment_address_in_file)
5147 char *where;
5148 fixS *fixP;
5149 relax_addressT segment_address_in_file;
5150 {
5151 /* In: length of relocation (or of address) in chars: 1, 2 or 4.
5152 Out: GNU LD relocation length code: 0, 1, or 2. */
5153
5154 static const unsigned char nbytes_r_length[] = { 42, 0, 1, 42, 2 };
5155 long r_symbolnum;
5156
5157 know (fixP->fx_addsy != NULL);
5158
5159 md_number_to_chars (where,
5160 (valueT) (fixP->fx_frag->fr_address
5161 + fixP->fx_where - segment_address_in_file),
5162 4);
5163
5164 r_symbolnum = (S_IS_DEFINED (fixP->fx_addsy)
5165 ? S_GET_TYPE (fixP->fx_addsy)
5166 : fixP->fx_addsy->sy_number);
5167
5168 where[6] = (r_symbolnum >> 16) & 0x0ff;
5169 where[5] = (r_symbolnum >> 8) & 0x0ff;
5170 where[4] = r_symbolnum & 0x0ff;
5171 where[7] = ((((!S_IS_DEFINED (fixP->fx_addsy)) << 3) & 0x08)
5172 | ((nbytes_r_length[fixP->fx_size] << 1) & 0x06)
5173 | (((fixP->fx_pcrel << 0) & 0x01) & 0x0f));
5174 }
5175
5176 #endif /* OBJ_AOUT or OBJ_BOUT. */
5177
5178 #if defined (I386COFF)
5179
5180 short
5181 tc_coff_fix2rtype (fixP)
5182 fixS *fixP;
5183 {
5184 if (fixP->fx_r_type == R_IMAGEBASE)
5185 return R_IMAGEBASE;
5186
5187 return (fixP->fx_pcrel ?
5188 (fixP->fx_size == 1 ? R_PCRBYTE :
5189 fixP->fx_size == 2 ? R_PCRWORD :
5190 R_PCRLONG) :
5191 (fixP->fx_size == 1 ? R_RELBYTE :
5192 fixP->fx_size == 2 ? R_RELWORD :
5193 R_DIR32));
5194 }
5195
5196 int
5197 tc_coff_sizemachdep (frag)
5198 fragS *frag;
5199 {
5200 if (frag->fr_next)
5201 return (frag->fr_next->fr_address - frag->fr_address);
5202 else
5203 return 0;
5204 }
5205
5206 #endif /* I386COFF */
5207
5208 #endif /* !BFD_ASSEMBLER */
5209 \f
5210 /* Parse operands using Intel syntax. This implements a recursive descent
5211 parser based on the BNF grammar published in Appendix B of the MASM 6.1
5212 Programmer's Guide.
5213
5214 FIXME: We do not recognize the full operand grammar defined in the MASM
5215 documentation. In particular, all the structure/union and
5216 high-level macro operands are missing.
5217
5218 Uppercase words are terminals, lower case words are non-terminals.
5219 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
5220 bars '|' denote choices. Most grammar productions are implemented in
5221 functions called 'intel_<production>'.
5222
5223 Initial production is 'expr'.
5224
5225 addOp + | -
5226
5227 alpha [a-zA-Z]
5228
5229 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
5230
5231 constant digits [[ radixOverride ]]
5232
5233 dataType BYTE | WORD | DWORD | QWORD | XWORD
5234
5235 digits decdigit
5236 | digits decdigit
5237 | digits hexdigit
5238
5239 decdigit [0-9]
5240
5241 e05 e05 addOp e06
5242 | e06
5243
5244 e06 e06 mulOp e09
5245 | e09
5246
5247 e09 OFFSET e10
5248 | e09 PTR e10
5249 | e09 : e10
5250 | e10
5251
5252 e10 e10 [ expr ]
5253 | e11
5254
5255 e11 ( expr )
5256 | [ expr ]
5257 | constant
5258 | dataType
5259 | id
5260 | $
5261 | register
5262
5263 => expr SHORT e05
5264 | e05
5265
5266 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
5267 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
5268
5269 hexdigit a | b | c | d | e | f
5270 | A | B | C | D | E | F
5271
5272 id alpha
5273 | id alpha
5274 | id decdigit
5275
5276 mulOp * | / | MOD
5277
5278 quote " | '
5279
5280 register specialRegister
5281 | gpRegister
5282 | byteRegister
5283
5284 segmentRegister CS | DS | ES | FS | GS | SS
5285
5286 specialRegister CR0 | CR2 | CR3
5287 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
5288 | TR3 | TR4 | TR5 | TR6 | TR7
5289
5290 We simplify the grammar in obvious places (e.g., register parsing is
5291 done by calling parse_register) and eliminate immediate left recursion
5292 to implement a recursive-descent parser.
5293
5294 expr SHORT e05
5295 | e05
5296
5297 e05 e06 e05'
5298
5299 e05' addOp e06 e05'
5300 | Empty
5301
5302 e06 e09 e06'
5303
5304 e06' mulOp e09 e06'
5305 | Empty
5306
5307 e09 OFFSET e10 e09'
5308 | e10 e09'
5309
5310 e09' PTR e10 e09'
5311 | : e10 e09'
5312 | Empty
5313
5314 e10 e11 e10'
5315
5316 e10' [ expr ] e10'
5317 | Empty
5318
5319 e11 ( expr )
5320 | [ expr ]
5321 | BYTE
5322 | WORD
5323 | DWORD
5324 | QWORD
5325 | XWORD
5326 | .
5327 | $
5328 | register
5329 | id
5330 | constant */
5331
5332 /* Parsing structure for the intel syntax parser. Used to implement the
5333 semantic actions for the operand grammar. */
5334 struct intel_parser_s
5335 {
5336 char *op_string; /* The string being parsed. */
5337 int got_a_float; /* Whether the operand is a float. */
5338 int op_modifier; /* Operand modifier. */
5339 int is_mem; /* 1 if operand is memory reference. */
5340 const reg_entry *reg; /* Last register reference found. */
5341 char *disp; /* Displacement string being built. */
5342 };
5343
5344 static struct intel_parser_s intel_parser;
5345
5346 /* Token structure for parsing intel syntax. */
5347 struct intel_token
5348 {
5349 int code; /* Token code. */
5350 const reg_entry *reg; /* Register entry for register tokens. */
5351 char *str; /* String representation. */
5352 };
5353
5354 static struct intel_token cur_token, prev_token;
5355
5356 /* Token codes for the intel parser. Since T_SHORT is already used
5357 by COFF, undefine it first to prevent a warning. */
5358 #define T_NIL -1
5359 #define T_CONST 1
5360 #define T_REG 2
5361 #define T_BYTE 3
5362 #define T_WORD 4
5363 #define T_DWORD 5
5364 #define T_QWORD 6
5365 #define T_XWORD 7
5366 #undef T_SHORT
5367 #define T_SHORT 8
5368 #define T_OFFSET 9
5369 #define T_PTR 10
5370 #define T_ID 11
5371
5372 /* Prototypes for intel parser functions. */
5373 static int intel_match_token PARAMS ((int code));
5374 static void intel_get_token PARAMS ((void));
5375 static void intel_putback_token PARAMS ((void));
5376 static int intel_expr PARAMS ((void));
5377 static int intel_e05 PARAMS ((void));
5378 static int intel_e05_1 PARAMS ((void));
5379 static int intel_e06 PARAMS ((void));
5380 static int intel_e06_1 PARAMS ((void));
5381 static int intel_e09 PARAMS ((void));
5382 static int intel_e09_1 PARAMS ((void));
5383 static int intel_e10 PARAMS ((void));
5384 static int intel_e10_1 PARAMS ((void));
5385 static int intel_e11 PARAMS ((void));
5386
5387 static int
5388 i386_intel_operand (operand_string, got_a_float)
5389 char *operand_string;
5390 int got_a_float;
5391 {
5392 int ret;
5393 char *p;
5394
5395 /* Initialize token holders. */
5396 cur_token.code = prev_token.code = T_NIL;
5397 cur_token.reg = prev_token.reg = NULL;
5398 cur_token.str = prev_token.str = NULL;
5399
5400 /* Initialize parser structure. */
5401 p = intel_parser.op_string = (char *) malloc (strlen (operand_string) + 1);
5402 if (p == NULL)
5403 abort ();
5404 strcpy (intel_parser.op_string, operand_string);
5405 intel_parser.got_a_float = got_a_float;
5406 intel_parser.op_modifier = -1;
5407 intel_parser.is_mem = 0;
5408 intel_parser.reg = NULL;
5409 intel_parser.disp = (char *) malloc (strlen (operand_string) + 1);
5410 if (intel_parser.disp == NULL)
5411 abort ();
5412 intel_parser.disp[0] = '\0';
5413
5414 /* Read the first token and start the parser. */
5415 intel_get_token ();
5416 ret = intel_expr ();
5417
5418 if (ret)
5419 {
5420 /* If we found a memory reference, hand it over to i386_displacement
5421 to fill in the rest of the operand fields. */
5422 if (intel_parser.is_mem)
5423 {
5424 if ((i.mem_operands == 1
5425 && (current_templates->start->opcode_modifier & IsString) == 0)
5426 || i.mem_operands == 2)
5427 {
5428 as_bad (_("too many memory references for '%s'"),
5429 current_templates->start->name);
5430 ret = 0;
5431 }
5432 else
5433 {
5434 char *s = intel_parser.disp;
5435 i.mem_operands++;
5436
5437 /* Add the displacement expression. */
5438 if (*s != '\0')
5439 ret = i386_displacement (s, s + strlen (s))
5440 && i386_index_check (s);
5441 }
5442 }
5443
5444 /* Constant and OFFSET expressions are handled by i386_immediate. */
5445 else if (intel_parser.op_modifier == OFFSET_FLAT
5446 || intel_parser.reg == NULL)
5447 ret = i386_immediate (intel_parser.disp);
5448 }
5449
5450 free (p);
5451 free (intel_parser.disp);
5452
5453 return ret;
5454 }
5455
5456 /* expr SHORT e05
5457 | e05 */
5458 static int
5459 intel_expr ()
5460 {
5461 /* expr SHORT e05 */
5462 if (cur_token.code == T_SHORT)
5463 {
5464 intel_parser.op_modifier = SHORT;
5465 intel_match_token (T_SHORT);
5466
5467 return (intel_e05 ());
5468 }
5469
5470 /* expr e05 */
5471 else
5472 return intel_e05 ();
5473 }
5474
5475 /* e05 e06 e05'
5476
5477 e05' addOp e06 e05'
5478 | Empty */
5479 static int
5480 intel_e05 ()
5481 {
5482 return (intel_e06 () && intel_e05_1 ());
5483 }
5484
5485 static int
5486 intel_e05_1 ()
5487 {
5488 /* e05' addOp e06 e05' */
5489 if (cur_token.code == '+' || cur_token.code == '-')
5490 {
5491 strcat (intel_parser.disp, cur_token.str);
5492 intel_match_token (cur_token.code);
5493
5494 return (intel_e06 () && intel_e05_1 ());
5495 }
5496
5497 /* e05' Empty */
5498 else
5499 return 1;
5500 }
5501
5502 /* e06 e09 e06'
5503
5504 e06' mulOp e09 e06'
5505 | Empty */
5506 static int
5507 intel_e06 ()
5508 {
5509 return (intel_e09 () && intel_e06_1 ());
5510 }
5511
5512 static int
5513 intel_e06_1 ()
5514 {
5515 /* e06' mulOp e09 e06' */
5516 if (cur_token.code == '*' || cur_token.code == '/')
5517 {
5518 strcat (intel_parser.disp, cur_token.str);
5519 intel_match_token (cur_token.code);
5520
5521 return (intel_e09 () && intel_e06_1 ());
5522 }
5523
5524 /* e06' Empty */
5525 else
5526 return 1;
5527 }
5528
5529 /* e09 OFFSET e10 e09'
5530 | e10 e09'
5531
5532 e09' PTR e10 e09'
5533 | : e10 e09'
5534 | Empty */
5535 static int
5536 intel_e09 ()
5537 {
5538 /* e09 OFFSET e10 e09' */
5539 if (cur_token.code == T_OFFSET)
5540 {
5541 intel_parser.is_mem = 0;
5542 intel_parser.op_modifier = OFFSET_FLAT;
5543 intel_match_token (T_OFFSET);
5544
5545 return (intel_e10 () && intel_e09_1 ());
5546 }
5547
5548 /* e09 e10 e09' */
5549 else
5550 return (intel_e10 () && intel_e09_1 ());
5551 }
5552
5553 static int
5554 intel_e09_1 ()
5555 {
5556 /* e09' PTR e10 e09' */
5557 if (cur_token.code == T_PTR)
5558 {
5559 if (prev_token.code == T_BYTE)
5560 i.suffix = BYTE_MNEM_SUFFIX;
5561
5562 else if (prev_token.code == T_WORD)
5563 {
5564 if (intel_parser.got_a_float == 2) /* "fi..." */
5565 i.suffix = SHORT_MNEM_SUFFIX;
5566 else
5567 i.suffix = WORD_MNEM_SUFFIX;
5568 }
5569
5570 else if (prev_token.code == T_DWORD)
5571 {
5572 if (intel_parser.got_a_float == 1) /* "f..." */
5573 i.suffix = SHORT_MNEM_SUFFIX;
5574 else
5575 i.suffix = LONG_MNEM_SUFFIX;
5576 }
5577
5578 else if (prev_token.code == T_QWORD)
5579 {
5580 if (intel_parser.got_a_float == 1) /* "f..." */
5581 i.suffix = LONG_MNEM_SUFFIX;
5582 else
5583 i.suffix = QWORD_MNEM_SUFFIX;
5584 }
5585
5586 else if (prev_token.code == T_XWORD)
5587 i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
5588
5589 else
5590 {
5591 as_bad (_("Unknown operand modifier `%s'\n"), prev_token.str);
5592 return 0;
5593 }
5594
5595 intel_match_token (T_PTR);
5596
5597 return (intel_e10 () && intel_e09_1 ());
5598 }
5599
5600 /* e09 : e10 e09' */
5601 else if (cur_token.code == ':')
5602 {
5603 /* Mark as a memory operand only if it's not already known to be an
5604 offset expression. */
5605 if (intel_parser.op_modifier != OFFSET_FLAT)
5606 intel_parser.is_mem = 1;
5607
5608 return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
5609 }
5610
5611 /* e09' Empty */
5612 else
5613 return 1;
5614 }
5615
5616 /* e10 e11 e10'
5617
5618 e10' [ expr ] e10'
5619 | Empty */
5620 static int
5621 intel_e10 ()
5622 {
5623 return (intel_e11 () && intel_e10_1 ());
5624 }
5625
5626 static int
5627 intel_e10_1 ()
5628 {
5629 /* e10' [ expr ] e10' */
5630 if (cur_token.code == '[')
5631 {
5632 intel_match_token ('[');
5633
5634 /* Mark as a memory operand only if it's not already known to be an
5635 offset expression. If it's an offset expression, we need to keep
5636 the brace in. */
5637 if (intel_parser.op_modifier != OFFSET_FLAT)
5638 intel_parser.is_mem = 1;
5639 else
5640 strcat (intel_parser.disp, "[");
5641
5642 /* Add a '+' to the displacement string if necessary. */
5643 if (*intel_parser.disp != '\0'
5644 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5645 strcat (intel_parser.disp, "+");
5646
5647 if (intel_expr () && intel_match_token (']'))
5648 {
5649 /* Preserve brackets when the operand is an offset expression. */
5650 if (intel_parser.op_modifier == OFFSET_FLAT)
5651 strcat (intel_parser.disp, "]");
5652
5653 return intel_e10_1 ();
5654 }
5655 else
5656 return 0;
5657 }
5658
5659 /* e10' Empty */
5660 else
5661 return 1;
5662 }
5663
5664 /* e11 ( expr )
5665 | [ expr ]
5666 | BYTE
5667 | WORD
5668 | DWORD
5669 | QWORD
5670 | XWORD
5671 | $
5672 | .
5673 | register
5674 | id
5675 | constant */
5676 static int
5677 intel_e11 ()
5678 {
5679 /* e11 ( expr ) */
5680 if (cur_token.code == '(')
5681 {
5682 intel_match_token ('(');
5683 strcat (intel_parser.disp, "(");
5684
5685 if (intel_expr () && intel_match_token (')'))
5686 {
5687 strcat (intel_parser.disp, ")");
5688 return 1;
5689 }
5690 else
5691 return 0;
5692 }
5693
5694 /* e11 [ expr ] */
5695 else if (cur_token.code == '[')
5696 {
5697 intel_match_token ('[');
5698
5699 /* Mark as a memory operand only if it's not already known to be an
5700 offset expression. If it's an offset expression, we need to keep
5701 the brace in. */
5702 if (intel_parser.op_modifier != OFFSET_FLAT)
5703 intel_parser.is_mem = 1;
5704 else
5705 strcat (intel_parser.disp, "[");
5706
5707 /* Operands for jump/call inside brackets denote absolute addresses. */
5708 if (current_templates->start->opcode_modifier & Jump
5709 || current_templates->start->opcode_modifier & JumpDword
5710 || current_templates->start->opcode_modifier & JumpByte
5711 || current_templates->start->opcode_modifier & JumpInterSegment)
5712 i.types[this_operand] |= JumpAbsolute;
5713
5714 /* Add a '+' to the displacement string if necessary. */
5715 if (*intel_parser.disp != '\0'
5716 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5717 strcat (intel_parser.disp, "+");
5718
5719 if (intel_expr () && intel_match_token (']'))
5720 {
5721 /* Preserve brackets when the operand is an offset expression. */
5722 if (intel_parser.op_modifier == OFFSET_FLAT)
5723 strcat (intel_parser.disp, "]");
5724
5725 return 1;
5726 }
5727 else
5728 return 0;
5729 }
5730
5731 /* e11 BYTE
5732 | WORD
5733 | DWORD
5734 | QWORD
5735 | XWORD */
5736 else if (cur_token.code == T_BYTE
5737 || cur_token.code == T_WORD
5738 || cur_token.code == T_DWORD
5739 || cur_token.code == T_QWORD
5740 || cur_token.code == T_XWORD)
5741 {
5742 intel_match_token (cur_token.code);
5743
5744 return 1;
5745 }
5746
5747 /* e11 $
5748 | . */
5749 else if (cur_token.code == '$' || cur_token.code == '.')
5750 {
5751 strcat (intel_parser.disp, cur_token.str);
5752 intel_match_token (cur_token.code);
5753
5754 /* Mark as a memory operand only if it's not already known to be an
5755 offset expression. */
5756 if (intel_parser.op_modifier != OFFSET_FLAT)
5757 intel_parser.is_mem = 1;
5758
5759 return 1;
5760 }
5761
5762 /* e11 register */
5763 else if (cur_token.code == T_REG)
5764 {
5765 const reg_entry *reg = intel_parser.reg = cur_token.reg;
5766
5767 intel_match_token (T_REG);
5768
5769 /* Check for segment change. */
5770 if (cur_token.code == ':')
5771 {
5772 if (reg->reg_type & (SReg2 | SReg3))
5773 {
5774 switch (reg->reg_num)
5775 {
5776 case 0:
5777 i.seg[i.mem_operands] = &es;
5778 break;
5779 case 1:
5780 i.seg[i.mem_operands] = &cs;
5781 break;
5782 case 2:
5783 i.seg[i.mem_operands] = &ss;
5784 break;
5785 case 3:
5786 i.seg[i.mem_operands] = &ds;
5787 break;
5788 case 4:
5789 i.seg[i.mem_operands] = &fs;
5790 break;
5791 case 5:
5792 i.seg[i.mem_operands] = &gs;
5793 break;
5794 }
5795 }
5796 else
5797 {
5798 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
5799 return 0;
5800 }
5801 }
5802
5803 /* Not a segment register. Check for register scaling. */
5804 else if (cur_token.code == '*')
5805 {
5806 if (!intel_parser.is_mem)
5807 {
5808 as_bad (_("Register scaling only allowed in memory operands."));
5809 return 0;
5810 }
5811
5812 /* What follows must be a valid scale. */
5813 if (intel_match_token ('*')
5814 && strchr ("01248", *cur_token.str))
5815 {
5816 i.index_reg = reg;
5817 i.types[this_operand] |= BaseIndex;
5818
5819 /* Set the scale after setting the register (otherwise,
5820 i386_scale will complain) */
5821 i386_scale (cur_token.str);
5822 intel_match_token (T_CONST);
5823 }
5824 else
5825 {
5826 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
5827 cur_token.str);
5828 return 0;
5829 }
5830 }
5831
5832 /* No scaling. If this is a memory operand, the register is either a
5833 base register (first occurrence) or an index register (second
5834 occurrence). */
5835 else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3)))
5836 {
5837 if (i.base_reg && i.index_reg)
5838 {
5839 as_bad (_("Too many register references in memory operand.\n"));
5840 return 0;
5841 }
5842
5843 if (i.base_reg == NULL)
5844 i.base_reg = reg;
5845 else
5846 i.index_reg = reg;
5847
5848 i.types[this_operand] |= BaseIndex;
5849 }
5850
5851 /* Offset modifier. Add the register to the displacement string to be
5852 parsed as an immediate expression after we're done. */
5853 else if (intel_parser.op_modifier == OFFSET_FLAT)
5854 strcat (intel_parser.disp, reg->reg_name);
5855
5856 /* It's neither base nor index nor offset. */
5857 else
5858 {
5859 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
5860 i.op[this_operand].regs = reg;
5861 i.reg_operands++;
5862 }
5863
5864 /* Since registers are not part of the displacement string (except
5865 when we're parsing offset operands), we may need to remove any
5866 preceding '+' from the displacement string. */
5867 if (*intel_parser.disp != '\0'
5868 && intel_parser.op_modifier != OFFSET_FLAT)
5869 {
5870 char *s = intel_parser.disp;
5871 s += strlen (s) - 1;
5872 if (*s == '+')
5873 *s = '\0';
5874 }
5875
5876 return 1;
5877 }
5878
5879 /* e11 id */
5880 else if (cur_token.code == T_ID)
5881 {
5882 /* Add the identifier to the displacement string. */
5883 strcat (intel_parser.disp, cur_token.str);
5884 intel_match_token (T_ID);
5885
5886 /* The identifier represents a memory reference only if it's not
5887 preceded by an offset modifier. */
5888 if (intel_parser.op_modifier != OFFSET_FLAT)
5889 intel_parser.is_mem = 1;
5890
5891 return 1;
5892 }
5893
5894 /* e11 constant */
5895 else if (cur_token.code == T_CONST
5896 || cur_token.code == '-'
5897 || cur_token.code == '+')
5898 {
5899 char *save_str;
5900
5901 /* Allow constants that start with `+' or `-'. */
5902 if (cur_token.code == '-' || cur_token.code == '+')
5903 {
5904 strcat (intel_parser.disp, cur_token.str);
5905 intel_match_token (cur_token.code);
5906 if (cur_token.code != T_CONST)
5907 {
5908 as_bad (_("Syntax error. Expecting a constant. Got `%s'.\n"),
5909 cur_token.str);
5910 return 0;
5911 }
5912 }
5913
5914 save_str = (char *) malloc (strlen (cur_token.str) + 1);
5915 if (save_str == NULL)
5916 abort ();
5917 strcpy (save_str, cur_token.str);
5918
5919 /* Get the next token to check for register scaling. */
5920 intel_match_token (cur_token.code);
5921
5922 /* Check if this constant is a scaling factor for an index register. */
5923 if (cur_token.code == '*')
5924 {
5925 if (intel_match_token ('*') && cur_token.code == T_REG)
5926 {
5927 if (!intel_parser.is_mem)
5928 {
5929 as_bad (_("Register scaling only allowed in memory operands."));
5930 return 0;
5931 }
5932
5933 /* The constant is followed by `* reg', so it must be
5934 a valid scale. */
5935 if (strchr ("01248", *save_str))
5936 {
5937 i.index_reg = cur_token.reg;
5938 i.types[this_operand] |= BaseIndex;
5939
5940 /* Set the scale after setting the register (otherwise,
5941 i386_scale will complain) */
5942 i386_scale (save_str);
5943 intel_match_token (T_REG);
5944
5945 /* Since registers are not part of the displacement
5946 string, we may need to remove any preceding '+' from
5947 the displacement string. */
5948 if (*intel_parser.disp != '\0')
5949 {
5950 char *s = intel_parser.disp;
5951 s += strlen (s) - 1;
5952 if (*s == '+')
5953 *s = '\0';
5954 }
5955
5956 free (save_str);
5957
5958 return 1;
5959 }
5960 else
5961 return 0;
5962 }
5963
5964 /* The constant was not used for register scaling. Since we have
5965 already consumed the token following `*' we now need to put it
5966 back in the stream. */
5967 else
5968 intel_putback_token ();
5969 }
5970
5971 /* Add the constant to the displacement string. */
5972 strcat (intel_parser.disp, save_str);
5973 free (save_str);
5974
5975 return 1;
5976 }
5977
5978 as_bad (_("Unrecognized token '%s'"), cur_token.str);
5979 return 0;
5980 }
5981
5982 /* Match the given token against cur_token. If they match, read the next
5983 token from the operand string. */
5984 static int
5985 intel_match_token (code)
5986 int code;
5987 {
5988 if (cur_token.code == code)
5989 {
5990 intel_get_token ();
5991 return 1;
5992 }
5993 else
5994 {
5995 as_bad (_("Unexpected token `%s'\n"), cur_token.str);
5996 return 0;
5997 }
5998 }
5999
6000 /* Read a new token from intel_parser.op_string and store it in cur_token. */
6001 static void
6002 intel_get_token ()
6003 {
6004 char *end_op;
6005 const reg_entry *reg;
6006 struct intel_token new_token;
6007
6008 new_token.code = T_NIL;
6009 new_token.reg = NULL;
6010 new_token.str = NULL;
6011
6012 /* Free the memory allocated to the previous token and move
6013 cur_token to prev_token. */
6014 if (prev_token.str)
6015 free (prev_token.str);
6016
6017 prev_token = cur_token;
6018
6019 /* Skip whitespace. */
6020 while (is_space_char (*intel_parser.op_string))
6021 intel_parser.op_string++;
6022
6023 /* Return an empty token if we find nothing else on the line. */
6024 if (*intel_parser.op_string == '\0')
6025 {
6026 cur_token = new_token;
6027 return;
6028 }
6029
6030 /* The new token cannot be larger than the remainder of the operand
6031 string. */
6032 new_token.str = (char *) malloc (strlen (intel_parser.op_string) + 1);
6033 if (new_token.str == NULL)
6034 abort ();
6035 new_token.str[0] = '\0';
6036
6037 if (strchr ("0123456789", *intel_parser.op_string))
6038 {
6039 char *p = new_token.str;
6040 char *q = intel_parser.op_string;
6041 new_token.code = T_CONST;
6042
6043 /* Allow any kind of identifier char to encompass floating point and
6044 hexadecimal numbers. */
6045 while (is_identifier_char (*q))
6046 *p++ = *q++;
6047 *p = '\0';
6048
6049 /* Recognize special symbol names [0-9][bf]. */
6050 if (strlen (intel_parser.op_string) == 2
6051 && (intel_parser.op_string[1] == 'b'
6052 || intel_parser.op_string[1] == 'f'))
6053 new_token.code = T_ID;
6054 }
6055
6056 else if (strchr ("+-/*:[]()", *intel_parser.op_string))
6057 {
6058 new_token.code = *intel_parser.op_string;
6059 new_token.str[0] = *intel_parser.op_string;
6060 new_token.str[1] = '\0';
6061 }
6062
6063 else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
6064 && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
6065 {
6066 new_token.code = T_REG;
6067 new_token.reg = reg;
6068
6069 if (*intel_parser.op_string == REGISTER_PREFIX)
6070 {
6071 new_token.str[0] = REGISTER_PREFIX;
6072 new_token.str[1] = '\0';
6073 }
6074
6075 strcat (new_token.str, reg->reg_name);
6076 }
6077
6078 else if (is_identifier_char (*intel_parser.op_string))
6079 {
6080 char *p = new_token.str;
6081 char *q = intel_parser.op_string;
6082
6083 /* A '.' or '$' followed by an identifier char is an identifier.
6084 Otherwise, it's operator '.' followed by an expression. */
6085 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
6086 {
6087 new_token.code = *q;
6088 new_token.str[0] = *q;
6089 new_token.str[1] = '\0';
6090 }
6091 else
6092 {
6093 while (is_identifier_char (*q) || *q == '@')
6094 *p++ = *q++;
6095 *p = '\0';
6096
6097 if (strcasecmp (new_token.str, "BYTE") == 0)
6098 new_token.code = T_BYTE;
6099
6100 else if (strcasecmp (new_token.str, "WORD") == 0)
6101 new_token.code = T_WORD;
6102
6103 else if (strcasecmp (new_token.str, "DWORD") == 0)
6104 new_token.code = T_DWORD;
6105
6106 else if (strcasecmp (new_token.str, "QWORD") == 0)
6107 new_token.code = T_QWORD;
6108
6109 else if (strcasecmp (new_token.str, "XWORD") == 0)
6110 new_token.code = T_XWORD;
6111
6112 else if (strcasecmp (new_token.str, "PTR") == 0)
6113 new_token.code = T_PTR;
6114
6115 else if (strcasecmp (new_token.str, "SHORT") == 0)
6116 new_token.code = T_SHORT;
6117
6118 else if (strcasecmp (new_token.str, "OFFSET") == 0)
6119 {
6120 new_token.code = T_OFFSET;
6121
6122 /* ??? This is not mentioned in the MASM grammar but gcc
6123 makes use of it with -mintel-syntax. OFFSET may be
6124 followed by FLAT: */
6125 if (strncasecmp (q, " FLAT:", 6) == 0)
6126 strcat (new_token.str, " FLAT:");
6127 }
6128
6129 /* ??? This is not mentioned in the MASM grammar. */
6130 else if (strcasecmp (new_token.str, "FLAT") == 0)
6131 new_token.code = T_OFFSET;
6132
6133 else
6134 new_token.code = T_ID;
6135 }
6136 }
6137
6138 else
6139 as_bad (_("Unrecognized token `%s'\n"), intel_parser.op_string);
6140
6141 intel_parser.op_string += strlen (new_token.str);
6142 cur_token = new_token;
6143 }
6144
6145 /* Put cur_token back into the token stream and make cur_token point to
6146 prev_token. */
6147 static void
6148 intel_putback_token ()
6149 {
6150 intel_parser.op_string -= strlen (cur_token.str);
6151 free (cur_token.str);
6152 cur_token = prev_token;
6153
6154 /* Forget prev_token. */
6155 prev_token.code = T_NIL;
6156 prev_token.reg = NULL;
6157 prev_token.str = NULL;
6158 }
This page took 0.176717 seconds and 4 git commands to generate.