* config/tc-i386.c (md_estimate_size_before_relax) Don't lose
[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 && 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
2232 /* Now select between word & dword operations via the operand
2233 size prefix, except for instructions that will ignore this
2234 prefix anyway. */
2235 if (i.suffix != QWORD_MNEM_SUFFIX
2236 && (i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2237 && !(i.tm.opcode_modifier & IgnoreSize))
2238 {
2239 unsigned int prefix = DATA_PREFIX_OPCODE;
2240 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2241 prefix = ADDR_PREFIX_OPCODE;
2242
2243 if (!add_prefix (prefix))
2244 return 0;
2245 }
2246
2247 if (i.suffix != QWORD_MNEM_SUFFIX && (flag_code == CODE_64BIT)
2248 && !(i.tm.opcode_modifier & IgnoreSize)
2249 && (i.tm.opcode_modifier & JumpByte))
2250 {
2251 if (!add_prefix (ADDR_PREFIX_OPCODE))
2252 return 0;
2253 }
2254
2255 /* Set mode64 for an operand. */
2256 if (i.suffix == QWORD_MNEM_SUFFIX
2257 && (i.tm.opcode_modifier & NoRex64) == 0)
2258 {
2259 i.rex |= REX_MODE64;
2260 if (flag_code < CODE_64BIT)
2261 {
2262 as_bad (_("64bit operations available only in 64bit modes."));
2263 return 0;
2264 }
2265 }
2266
2267 /* Size floating point instruction. */
2268 if (i.suffix == LONG_MNEM_SUFFIX)
2269 {
2270 if (i.tm.opcode_modifier & FloatMF)
2271 i.tm.base_opcode ^= 4;
2272 }
2273 }
2274
2275 return 1;
2276 }
2277
2278 static int
2279 check_byte_reg ()
2280 {
2281 int op;
2282 for (op = i.operands; --op >= 0;)
2283 {
2284 /* If this is an eight bit register, it's OK. If it's the 16 or
2285 32 bit version of an eight bit register, we will just use the
2286 low portion, and that's OK too. */
2287 if (i.types[op] & Reg8)
2288 continue;
2289
2290 /* movzx and movsx should not generate this warning. */
2291 if (intel_syntax
2292 && (i.tm.base_opcode == 0xfb7
2293 || i.tm.base_opcode == 0xfb6
2294 || i.tm.base_opcode == 0x63
2295 || i.tm.base_opcode == 0xfbe
2296 || i.tm.base_opcode == 0xfbf))
2297 continue;
2298
2299 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4
2300 #if 0
2301 /* Check that the template allows eight bit regs. This
2302 kills insns such as `orb $1,%edx', which maybe should be
2303 allowed. */
2304 && (i.tm.operand_types[op] & (Reg8 | InOutPortReg))
2305 #endif
2306 )
2307 {
2308 /* Prohibit these changes in the 64bit mode, since the
2309 lowering is more complicated. */
2310 if (flag_code == CODE_64BIT
2311 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2312 {
2313 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2314 i.op[op].regs->reg_name,
2315 i.suffix);
2316 return 0;
2317 }
2318 #if REGISTER_WARNINGS
2319 if (!quiet_warnings
2320 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2321 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2322 (i.op[op].regs + (i.types[op] & Reg16
2323 ? REGNAM_AL - REGNAM_AX
2324 : REGNAM_AL - REGNAM_EAX))->reg_name,
2325 i.op[op].regs->reg_name,
2326 i.suffix);
2327 #endif
2328 continue;
2329 }
2330 /* Any other register is bad. */
2331 if (i.types[op] & (Reg | RegMMX | RegXMM
2332 | SReg2 | SReg3
2333 | Control | Debug | Test
2334 | FloatReg | FloatAcc))
2335 {
2336 as_bad (_("`%%%s' not allowed with `%s%c'"),
2337 i.op[op].regs->reg_name,
2338 i.tm.name,
2339 i.suffix);
2340 return 0;
2341 }
2342 }
2343 return 1;
2344 }
2345
2346 static int
2347 check_long_reg ()
2348 {
2349 int op;
2350
2351 for (op = i.operands; --op >= 0;)
2352 /* Reject eight bit registers, except where the template requires
2353 them. (eg. movzb) */
2354 if ((i.types[op] & Reg8) != 0
2355 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2356 {
2357 as_bad (_("`%%%s' not allowed with `%s%c'"),
2358 i.op[op].regs->reg_name,
2359 i.tm.name,
2360 i.suffix);
2361 return 0;
2362 }
2363 /* Warn if the e prefix on a general reg is missing. */
2364 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2365 && (i.types[op] & Reg16) != 0
2366 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2367 {
2368 /* Prohibit these changes in the 64bit mode, since the
2369 lowering is more complicated. */
2370 if (flag_code == CODE_64BIT)
2371 {
2372 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2373 i.op[op].regs->reg_name,
2374 i.suffix);
2375 return 0;
2376 }
2377 #if REGISTER_WARNINGS
2378 else
2379 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2380 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
2381 i.op[op].regs->reg_name,
2382 i.suffix);
2383 #endif
2384 }
2385 /* Warn if the r prefix on a general reg is missing. */
2386 else if ((i.types[op] & Reg64) != 0
2387 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2388 {
2389 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2390 i.op[op].regs->reg_name,
2391 i.suffix);
2392 return 0;
2393 }
2394 return 1;
2395 }
2396
2397 static int
2398 check_qword_reg ()
2399 {
2400 int op;
2401
2402 for (op = i.operands; --op >= 0; )
2403 /* Reject eight bit registers, except where the template requires
2404 them. (eg. movzb) */
2405 if ((i.types[op] & Reg8) != 0
2406 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2407 {
2408 as_bad (_("`%%%s' not allowed with `%s%c'"),
2409 i.op[op].regs->reg_name,
2410 i.tm.name,
2411 i.suffix);
2412 return 0;
2413 }
2414 /* Warn if the e prefix on a general reg is missing. */
2415 else if (((i.types[op] & Reg16) != 0
2416 || (i.types[op] & Reg32) != 0)
2417 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2418 {
2419 /* Prohibit these changes in the 64bit mode, since the
2420 lowering is more complicated. */
2421 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2422 i.op[op].regs->reg_name,
2423 i.suffix);
2424 return 0;
2425 }
2426 return 1;
2427 }
2428
2429 static int
2430 check_word_reg ()
2431 {
2432 int op;
2433 for (op = i.operands; --op >= 0;)
2434 /* Reject eight bit registers, except where the template requires
2435 them. (eg. movzb) */
2436 if ((i.types[op] & Reg8) != 0
2437 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2438 {
2439 as_bad (_("`%%%s' not allowed with `%s%c'"),
2440 i.op[op].regs->reg_name,
2441 i.tm.name,
2442 i.suffix);
2443 return 0;
2444 }
2445 /* Warn if the e prefix on a general reg is present. */
2446 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2447 && (i.types[op] & Reg32) != 0
2448 && (i.tm.operand_types[op] & (Reg16 | Acc)) != 0)
2449 {
2450 /* Prohibit these changes in the 64bit mode, since the
2451 lowering is more complicated. */
2452 if (flag_code == CODE_64BIT)
2453 {
2454 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2455 i.op[op].regs->reg_name,
2456 i.suffix);
2457 return 0;
2458 }
2459 else
2460 #if REGISTER_WARNINGS
2461 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2462 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
2463 i.op[op].regs->reg_name,
2464 i.suffix);
2465 #endif
2466 }
2467 return 1;
2468 }
2469
2470 static int
2471 finalize_imm ()
2472 {
2473 unsigned int overlap0, overlap1, overlap2;
2474
2475 overlap0 = i.types[0] & i.tm.operand_types[0];
2476 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S))
2477 && overlap0 != Imm8 && overlap0 != Imm8S
2478 && overlap0 != Imm16 && overlap0 != Imm32S
2479 && overlap0 != Imm32 && overlap0 != Imm64)
2480 {
2481 if (i.suffix)
2482 {
2483 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX
2484 ? Imm8 | Imm8S
2485 : (i.suffix == WORD_MNEM_SUFFIX
2486 ? Imm16
2487 : (i.suffix == QWORD_MNEM_SUFFIX
2488 ? Imm64 | Imm32S
2489 : Imm32)));
2490 }
2491 else if (overlap0 == (Imm16 | Imm32S | Imm32)
2492 || overlap0 == (Imm16 | Imm32)
2493 || overlap0 == (Imm16 | Imm32S))
2494 {
2495 overlap0 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
2496 ? Imm16 : Imm32S);
2497 }
2498 if (overlap0 != Imm8 && overlap0 != Imm8S
2499 && overlap0 != Imm16 && overlap0 != Imm32S
2500 && overlap0 != Imm32 && overlap0 != Imm64)
2501 {
2502 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
2503 return 0;
2504 }
2505 }
2506 i.types[0] = overlap0;
2507
2508 overlap1 = i.types[1] & i.tm.operand_types[1];
2509 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32))
2510 && overlap1 != Imm8 && overlap1 != Imm8S
2511 && overlap1 != Imm16 && overlap1 != Imm32S
2512 && overlap1 != Imm32 && overlap1 != Imm64)
2513 {
2514 if (i.suffix)
2515 {
2516 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX
2517 ? Imm8 | Imm8S
2518 : (i.suffix == WORD_MNEM_SUFFIX
2519 ? Imm16
2520 : (i.suffix == QWORD_MNEM_SUFFIX
2521 ? Imm64 | Imm32S
2522 : Imm32)));
2523 }
2524 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
2525 || overlap1 == (Imm16 | Imm32)
2526 || overlap1 == (Imm16 | Imm32S))
2527 {
2528 overlap1 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
2529 ? Imm16 : Imm32S);
2530 }
2531 if (overlap1 != Imm8 && overlap1 != Imm8S
2532 && overlap1 != Imm16 && overlap1 != Imm32S
2533 && overlap1 != Imm32 && overlap1 != Imm64)
2534 {
2535 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
2536 return 0;
2537 }
2538 }
2539 i.types[1] = overlap1;
2540
2541 overlap2 = i.types[2] & i.tm.operand_types[2];
2542 assert ((overlap2 & Imm) == 0);
2543 i.types[2] = overlap2;
2544
2545 return 1;
2546 }
2547
2548 static int
2549 process_operands ()
2550 {
2551 /* Default segment register this instruction will use for memory
2552 accesses. 0 means unknown. This is only for optimizing out
2553 unnecessary segment overrides. */
2554 const seg_entry *default_seg = 0;
2555
2556 /* The imul $imm, %reg instruction is converted into
2557 imul $imm, %reg, %reg, and the clr %reg instruction
2558 is converted into xor %reg, %reg. */
2559 if (i.tm.opcode_modifier & regKludge)
2560 {
2561 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
2562 /* Pretend we saw the extra register operand. */
2563 assert (i.op[first_reg_op + 1].regs == 0);
2564 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
2565 i.types[first_reg_op + 1] = i.types[first_reg_op];
2566 i.reg_operands = 2;
2567 }
2568
2569 if (i.tm.opcode_modifier & ShortForm)
2570 {
2571 /* The register or float register operand is in operand 0 or 1. */
2572 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
2573 /* Register goes in low 3 bits of opcode. */
2574 i.tm.base_opcode |= i.op[op].regs->reg_num;
2575 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2576 i.rex |= REX_EXTZ;
2577 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2578 {
2579 /* Warn about some common errors, but press on regardless.
2580 The first case can be generated by gcc (<= 2.8.1). */
2581 if (i.operands == 2)
2582 {
2583 /* Reversed arguments on faddp, fsubp, etc. */
2584 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
2585 i.op[1].regs->reg_name,
2586 i.op[0].regs->reg_name);
2587 }
2588 else
2589 {
2590 /* Extraneous `l' suffix on fp insn. */
2591 as_warn (_("translating to `%s %%%s'"), i.tm.name,
2592 i.op[0].regs->reg_name);
2593 }
2594 }
2595 }
2596 else if (i.tm.opcode_modifier & Modrm)
2597 {
2598 /* The opcode is completed (modulo i.tm.extension_opcode which
2599 must be put into the modrm byte).
2600 Now, we make the modrm & index base bytes based on all the
2601 info we've collected. */
2602
2603 default_seg = build_modrm_byte ();
2604 }
2605 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
2606 {
2607 if (i.tm.base_opcode == POP_SEG_SHORT
2608 && i.op[0].regs->reg_num == 1)
2609 {
2610 as_bad (_("you can't `pop %%cs'"));
2611 return 0;
2612 }
2613 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
2614 if ((i.op[0].regs->reg_flags & RegRex) != 0)
2615 i.rex |= REX_EXTZ;
2616 }
2617 else if ((i.tm.base_opcode & ~(D | W)) == MOV_AX_DISP32)
2618 {
2619 default_seg = &ds;
2620 }
2621 else if ((i.tm.opcode_modifier & IsString) != 0)
2622 {
2623 /* For the string instructions that allow a segment override
2624 on one of their operands, the default segment is ds. */
2625 default_seg = &ds;
2626 }
2627
2628 /* If a segment was explicitly specified,
2629 and the specified segment is not the default,
2630 use an opcode prefix to select it.
2631 If we never figured out what the default segment is,
2632 then default_seg will be zero at this point,
2633 and the specified segment prefix will always be used. */
2634 if ((i.seg[0]) && (i.seg[0] != default_seg))
2635 {
2636 if (!add_prefix (i.seg[0]->seg_prefix))
2637 return 0;
2638 }
2639 return 1;
2640 }
2641
2642 static const seg_entry *
2643 build_modrm_byte ()
2644 {
2645 const seg_entry *default_seg = 0;
2646
2647 /* i.reg_operands MUST be the number of real register operands;
2648 implicit registers do not count. */
2649 if (i.reg_operands == 2)
2650 {
2651 unsigned int source, dest;
2652 source = ((i.types[0]
2653 & (Reg | RegMMX | RegXMM
2654 | SReg2 | SReg3
2655 | Control | Debug | Test))
2656 ? 0 : 1);
2657 dest = source + 1;
2658
2659 i.rm.mode = 3;
2660 /* One of the register operands will be encoded in the i.tm.reg
2661 field, the other in the combined i.tm.mode and i.tm.regmem
2662 fields. If no form of this instruction supports a memory
2663 destination operand, then we assume the source operand may
2664 sometimes be a memory operand and so we need to store the
2665 destination in the i.rm.reg field. */
2666 if ((i.tm.operand_types[dest] & AnyMem) == 0)
2667 {
2668 i.rm.reg = i.op[dest].regs->reg_num;
2669 i.rm.regmem = i.op[source].regs->reg_num;
2670 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
2671 i.rex |= REX_EXTX;
2672 if ((i.op[source].regs->reg_flags & RegRex) != 0)
2673 i.rex |= REX_EXTZ;
2674 }
2675 else
2676 {
2677 i.rm.reg = i.op[source].regs->reg_num;
2678 i.rm.regmem = i.op[dest].regs->reg_num;
2679 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
2680 i.rex |= REX_EXTZ;
2681 if ((i.op[source].regs->reg_flags & RegRex) != 0)
2682 i.rex |= REX_EXTX;
2683 }
2684 }
2685 else
2686 { /* If it's not 2 reg operands... */
2687 if (i.mem_operands)
2688 {
2689 unsigned int fake_zero_displacement = 0;
2690 unsigned int op = ((i.types[0] & AnyMem)
2691 ? 0
2692 : (i.types[1] & AnyMem) ? 1 : 2);
2693
2694 default_seg = &ds;
2695
2696 if (i.base_reg == 0)
2697 {
2698 i.rm.mode = 0;
2699 if (!i.disp_operands)
2700 fake_zero_displacement = 1;
2701 if (i.index_reg == 0)
2702 {
2703 /* Operand is just <disp> */
2704 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)
2705 && (flag_code != CODE_64BIT))
2706 {
2707 i.rm.regmem = NO_BASE_REGISTER_16;
2708 i.types[op] &= ~Disp;
2709 i.types[op] |= Disp16;
2710 }
2711 else if (flag_code != CODE_64BIT
2712 || (i.prefix[ADDR_PREFIX] != 0))
2713 {
2714 i.rm.regmem = NO_BASE_REGISTER;
2715 i.types[op] &= ~Disp;
2716 i.types[op] |= Disp32;
2717 }
2718 else
2719 {
2720 /* 64bit mode overwrites the 32bit absolute
2721 addressing by RIP relative addressing and
2722 absolute addressing is encoded by one of the
2723 redundant SIB forms. */
2724 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2725 i.sib.base = NO_BASE_REGISTER;
2726 i.sib.index = NO_INDEX_REGISTER;
2727 i.types[op] &= ~Disp;
2728 i.types[op] |= Disp32S;
2729 }
2730 }
2731 else /* !i.base_reg && i.index_reg */
2732 {
2733 i.sib.index = i.index_reg->reg_num;
2734 i.sib.base = NO_BASE_REGISTER;
2735 i.sib.scale = i.log2_scale_factor;
2736 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2737 i.types[op] &= ~Disp;
2738 if (flag_code != CODE_64BIT)
2739 i.types[op] |= Disp32; /* Must be 32 bit */
2740 else
2741 i.types[op] |= Disp32S;
2742 if ((i.index_reg->reg_flags & RegRex) != 0)
2743 i.rex |= REX_EXTY;
2744 }
2745 }
2746 /* RIP addressing for 64bit mode. */
2747 else if (i.base_reg->reg_type == BaseIndex)
2748 {
2749 i.rm.regmem = NO_BASE_REGISTER;
2750 i.types[op] &= ~Disp;
2751 i.types[op] |= Disp32S;
2752 i.flags[op] = Operand_PCrel;
2753 }
2754 else if (i.base_reg->reg_type & Reg16)
2755 {
2756 switch (i.base_reg->reg_num)
2757 {
2758 case 3: /* (%bx) */
2759 if (i.index_reg == 0)
2760 i.rm.regmem = 7;
2761 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
2762 i.rm.regmem = i.index_reg->reg_num - 6;
2763 break;
2764 case 5: /* (%bp) */
2765 default_seg = &ss;
2766 if (i.index_reg == 0)
2767 {
2768 i.rm.regmem = 6;
2769 if ((i.types[op] & Disp) == 0)
2770 {
2771 /* fake (%bp) into 0(%bp) */
2772 i.types[op] |= Disp8;
2773 fake_zero_displacement = 1;
2774 }
2775 }
2776 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
2777 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
2778 break;
2779 default: /* (%si) -> 4 or (%di) -> 5 */
2780 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
2781 }
2782 i.rm.mode = mode_from_disp_size (i.types[op]);
2783 }
2784 else /* i.base_reg and 32/64 bit mode */
2785 {
2786 if (flag_code == CODE_64BIT
2787 && (i.types[op] & Disp))
2788 {
2789 if (i.types[op] & Disp8)
2790 i.types[op] = Disp8 | Disp32S;
2791 else
2792 i.types[op] = Disp32S;
2793 }
2794 i.rm.regmem = i.base_reg->reg_num;
2795 if ((i.base_reg->reg_flags & RegRex) != 0)
2796 i.rex |= REX_EXTZ;
2797 i.sib.base = i.base_reg->reg_num;
2798 /* x86-64 ignores REX prefix bit here to avoid decoder
2799 complications. */
2800 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
2801 {
2802 default_seg = &ss;
2803 if (i.disp_operands == 0)
2804 {
2805 fake_zero_displacement = 1;
2806 i.types[op] |= Disp8;
2807 }
2808 }
2809 else if (i.base_reg->reg_num == ESP_REG_NUM)
2810 {
2811 default_seg = &ss;
2812 }
2813 i.sib.scale = i.log2_scale_factor;
2814 if (i.index_reg == 0)
2815 {
2816 /* <disp>(%esp) becomes two byte modrm with no index
2817 register. We've already stored the code for esp
2818 in i.rm.regmem ie. ESCAPE_TO_TWO_BYTE_ADDRESSING.
2819 Any base register besides %esp will not use the
2820 extra modrm byte. */
2821 i.sib.index = NO_INDEX_REGISTER;
2822 #if !SCALE1_WHEN_NO_INDEX
2823 /* Another case where we force the second modrm byte. */
2824 if (i.log2_scale_factor)
2825 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2826 #endif
2827 }
2828 else
2829 {
2830 i.sib.index = i.index_reg->reg_num;
2831 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2832 if ((i.index_reg->reg_flags & RegRex) != 0)
2833 i.rex |= REX_EXTY;
2834 }
2835 i.rm.mode = mode_from_disp_size (i.types[op]);
2836 }
2837
2838 if (fake_zero_displacement)
2839 {
2840 /* Fakes a zero displacement assuming that i.types[op]
2841 holds the correct displacement size. */
2842 expressionS *exp;
2843
2844 assert (i.op[op].disps == 0);
2845 exp = &disp_expressions[i.disp_operands++];
2846 i.op[op].disps = exp;
2847 exp->X_op = O_constant;
2848 exp->X_add_number = 0;
2849 exp->X_add_symbol = (symbolS *) 0;
2850 exp->X_op_symbol = (symbolS *) 0;
2851 }
2852 }
2853
2854 /* Fill in i.rm.reg or i.rm.regmem field with register operand
2855 (if any) based on i.tm.extension_opcode. Again, we must be
2856 careful to make sure that segment/control/debug/test/MMX
2857 registers are coded into the i.rm.reg field. */
2858 if (i.reg_operands)
2859 {
2860 unsigned int op =
2861 ((i.types[0]
2862 & (Reg | RegMMX | RegXMM
2863 | SReg2 | SReg3
2864 | Control | Debug | Test))
2865 ? 0
2866 : ((i.types[1]
2867 & (Reg | RegMMX | RegXMM
2868 | SReg2 | SReg3
2869 | Control | Debug | Test))
2870 ? 1
2871 : 2));
2872 /* If there is an extension opcode to put here, the register
2873 number must be put into the regmem field. */
2874 if (i.tm.extension_opcode != None)
2875 {
2876 i.rm.regmem = i.op[op].regs->reg_num;
2877 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2878 i.rex |= REX_EXTZ;
2879 }
2880 else
2881 {
2882 i.rm.reg = i.op[op].regs->reg_num;
2883 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2884 i.rex |= REX_EXTX;
2885 }
2886
2887 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2 we
2888 must set it to 3 to indicate this is a register operand
2889 in the regmem field. */
2890 if (!i.mem_operands)
2891 i.rm.mode = 3;
2892 }
2893
2894 /* Fill in i.rm.reg field with extension opcode (if any). */
2895 if (i.tm.extension_opcode != None)
2896 i.rm.reg = i.tm.extension_opcode;
2897 }
2898 return default_seg;
2899 }
2900
2901 static void
2902 output_branch ()
2903 {
2904 char *p;
2905 int code16;
2906 int prefix;
2907 relax_substateT subtype;
2908 symbolS *sym;
2909 offsetT off;
2910
2911 code16 = 0;
2912 if (flag_code == CODE_16BIT)
2913 code16 = CODE16;
2914
2915 prefix = 0;
2916 if (i.prefix[DATA_PREFIX] != 0)
2917 {
2918 prefix = 1;
2919 i.prefixes -= 1;
2920 code16 ^= CODE16;
2921 }
2922 /* Pentium4 branch hints. */
2923 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
2924 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
2925 {
2926 prefix++;
2927 i.prefixes--;
2928 }
2929 if (i.prefix[REX_PREFIX] != 0)
2930 {
2931 prefix++;
2932 i.prefixes--;
2933 }
2934
2935 if (i.prefixes != 0 && !intel_syntax)
2936 as_warn (_("skipping prefixes on this instruction"));
2937
2938 /* It's always a symbol; End frag & setup for relax.
2939 Make sure there is enough room in this frag for the largest
2940 instruction we may generate in md_convert_frag. This is 2
2941 bytes for the opcode and room for the prefix and largest
2942 displacement. */
2943 frag_grow (prefix + 2 + 4);
2944 /* Prefix and 1 opcode byte go in fr_fix. */
2945 p = frag_more (prefix + 1);
2946 if (i.prefix[DATA_PREFIX] != 0)
2947 *p++ = DATA_PREFIX_OPCODE;
2948 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
2949 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
2950 *p++ = i.prefix[SEG_PREFIX];
2951 if (i.prefix[REX_PREFIX] != 0)
2952 *p++ = i.prefix[REX_PREFIX];
2953 *p = i.tm.base_opcode;
2954
2955 if ((unsigned char) *p == JUMP_PC_RELATIVE)
2956 subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
2957 else if ((cpu_arch_flags & Cpu386) != 0)
2958 subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
2959 else
2960 subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
2961 subtype |= code16;
2962
2963 sym = i.op[0].disps->X_add_symbol;
2964 off = i.op[0].disps->X_add_number;
2965
2966 if (i.op[0].disps->X_op != O_constant
2967 && i.op[0].disps->X_op != O_symbol)
2968 {
2969 /* Handle complex expressions. */
2970 sym = make_expr_symbol (i.op[0].disps);
2971 off = 0;
2972 }
2973
2974 /* 1 possible extra opcode + 4 byte displacement go in var part.
2975 Pass reloc in fr_var. */
2976 frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
2977 }
2978
2979 static void
2980 output_jump ()
2981 {
2982 char *p;
2983 int size;
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 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 }
3037
3038 static void
3039 output_interseg_jump ()
3040 {
3041 char *p;
3042 int size;
3043 int prefix;
3044 int code16;
3045
3046 code16 = 0;
3047 if (flag_code == CODE_16BIT)
3048 code16 = CODE16;
3049
3050 prefix = 0;
3051 if (i.prefix[DATA_PREFIX] != 0)
3052 {
3053 prefix = 1;
3054 i.prefixes -= 1;
3055 code16 ^= CODE16;
3056 }
3057 if (i.prefix[REX_PREFIX] != 0)
3058 {
3059 prefix++;
3060 i.prefixes -= 1;
3061 }
3062
3063 size = 4;
3064 if (code16)
3065 size = 2;
3066
3067 if (i.prefixes != 0 && !intel_syntax)
3068 as_warn (_("skipping prefixes on this instruction"));
3069
3070 /* 1 opcode; 2 segment; offset */
3071 p = frag_more (prefix + 1 + 2 + size);
3072
3073 if (i.prefix[DATA_PREFIX] != 0)
3074 *p++ = DATA_PREFIX_OPCODE;
3075
3076 if (i.prefix[REX_PREFIX] != 0)
3077 *p++ = i.prefix[REX_PREFIX];
3078
3079 *p++ = i.tm.base_opcode;
3080 if (i.op[1].imms->X_op == O_constant)
3081 {
3082 offsetT n = i.op[1].imms->X_add_number;
3083
3084 if (size == 2
3085 && !fits_in_unsigned_word (n)
3086 && !fits_in_signed_word (n))
3087 {
3088 as_bad (_("16-bit jump out of range"));
3089 return;
3090 }
3091 md_number_to_chars (p, n, size);
3092 }
3093 else
3094 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3095 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
3096 if (i.op[0].imms->X_op != O_constant)
3097 as_bad (_("can't handle non absolute segment in `%s'"),
3098 i.tm.name);
3099 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
3100 }
3101
3102 static void
3103 output_insn ()
3104 {
3105 /* Tie dwarf2 debug info to the address at the start of the insn.
3106 We can't do this after the insn has been output as the current
3107 frag may have been closed off. eg. by frag_var. */
3108 dwarf2_emit_insn (0);
3109
3110 /* Output jumps. */
3111 if (i.tm.opcode_modifier & Jump)
3112 output_branch ();
3113 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
3114 output_jump ();
3115 else if (i.tm.opcode_modifier & JumpInterSegment)
3116 output_interseg_jump ();
3117 else
3118 {
3119 /* Output normal instructions here. */
3120 char *p;
3121 unsigned char *q;
3122
3123 /* All opcodes on i386 have either 1 or 2 bytes. We may use third
3124 byte for the SSE instructions to specify a prefix they require. */
3125 if (i.tm.base_opcode & 0xff0000)
3126 add_prefix ((i.tm.base_opcode >> 16) & 0xff);
3127
3128 /* The prefix bytes. */
3129 for (q = i.prefix;
3130 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
3131 q++)
3132 {
3133 if (*q)
3134 {
3135 p = frag_more (1);
3136 md_number_to_chars (p, (valueT) *q, 1);
3137 }
3138 }
3139
3140 /* Now the opcode; be careful about word order here! */
3141 if (fits_in_unsigned_byte (i.tm.base_opcode))
3142 {
3143 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
3144 }
3145 else
3146 {
3147 p = frag_more (2);
3148 /* Put out high byte first: can't use md_number_to_chars! */
3149 *p++ = (i.tm.base_opcode >> 8) & 0xff;
3150 *p = i.tm.base_opcode & 0xff;
3151 }
3152
3153 /* Now the modrm byte and sib byte (if present). */
3154 if (i.tm.opcode_modifier & Modrm)
3155 {
3156 p = frag_more (1);
3157 md_number_to_chars (p,
3158 (valueT) (i.rm.regmem << 0
3159 | i.rm.reg << 3
3160 | i.rm.mode << 6),
3161 1);
3162 /* If i.rm.regmem == ESP (4)
3163 && i.rm.mode != (Register mode)
3164 && not 16 bit
3165 ==> need second modrm byte. */
3166 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
3167 && i.rm.mode != 3
3168 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
3169 {
3170 p = frag_more (1);
3171 md_number_to_chars (p,
3172 (valueT) (i.sib.base << 0
3173 | i.sib.index << 3
3174 | i.sib.scale << 6),
3175 1);
3176 }
3177 }
3178
3179 if (i.disp_operands)
3180 output_disp ();
3181
3182 if (i.imm_operands)
3183 output_imm ();
3184 }
3185
3186 #ifdef DEBUG386
3187 if (flag_debug)
3188 {
3189 pi (line, &i);
3190 }
3191 #endif /* DEBUG386 */
3192 }
3193
3194 static void
3195 output_disp ()
3196 {
3197 char *p;
3198 unsigned int n;
3199
3200 for (n = 0; n < i.operands; n++)
3201 {
3202 if (i.types[n] & Disp)
3203 {
3204 if (i.op[n].disps->X_op == O_constant)
3205 {
3206 int size;
3207 offsetT val;
3208
3209 size = 4;
3210 if (i.types[n] & (Disp8 | Disp16 | Disp64))
3211 {
3212 size = 2;
3213 if (i.types[n] & Disp8)
3214 size = 1;
3215 if (i.types[n] & Disp64)
3216 size = 8;
3217 }
3218 val = offset_in_range (i.op[n].disps->X_add_number,
3219 size);
3220 p = frag_more (size);
3221 md_number_to_chars (p, val, size);
3222 }
3223 else
3224 {
3225 int size = 4;
3226 int sign = 0;
3227 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
3228
3229 /* The PC relative address is computed relative
3230 to the instruction boundary, so in case immediate
3231 fields follows, we need to adjust the value. */
3232 if (pcrel && i.imm_operands)
3233 {
3234 int imm_size = 4;
3235 unsigned int n1;
3236
3237 for (n1 = 0; n1 < i.operands; n1++)
3238 if (i.types[n1] & Imm)
3239 {
3240 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
3241 {
3242 imm_size = 2;
3243 if (i.types[n1] & (Imm8 | Imm8S))
3244 imm_size = 1;
3245 if (i.types[n1] & Imm64)
3246 imm_size = 8;
3247 }
3248 break;
3249 }
3250 /* We should find the immediate. */
3251 if (n1 == i.operands)
3252 abort ();
3253 i.op[n].disps->X_add_number -= imm_size;
3254 }
3255
3256 if (i.types[n] & Disp32S)
3257 sign = 1;
3258
3259 if (i.types[n] & (Disp16 | Disp64))
3260 {
3261 size = 2;
3262 if (i.types[n] & Disp64)
3263 size = 8;
3264 }
3265
3266 p = frag_more (size);
3267 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3268 i.op[n].disps, pcrel,
3269 reloc (size, pcrel, sign, i.reloc[n]));
3270 }
3271 }
3272 }
3273 }
3274
3275 static void
3276 output_imm ()
3277 {
3278 char *p;
3279 unsigned int n;
3280
3281 for (n = 0; n < i.operands; n++)
3282 {
3283 if (i.types[n] & Imm)
3284 {
3285 if (i.op[n].imms->X_op == O_constant)
3286 {
3287 int size;
3288 offsetT val;
3289
3290 size = 4;
3291 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3292 {
3293 size = 2;
3294 if (i.types[n] & (Imm8 | Imm8S))
3295 size = 1;
3296 else if (i.types[n] & Imm64)
3297 size = 8;
3298 }
3299 val = offset_in_range (i.op[n].imms->X_add_number,
3300 size);
3301 p = frag_more (size);
3302 md_number_to_chars (p, val, size);
3303 }
3304 else
3305 {
3306 /* Not absolute_section.
3307 Need a 32-bit fixup (don't support 8bit
3308 non-absolute imms). Try to support other
3309 sizes ... */
3310 RELOC_ENUM reloc_type;
3311 int size = 4;
3312 int sign = 0;
3313
3314 if ((i.types[n] & (Imm32S))
3315 && i.suffix == QWORD_MNEM_SUFFIX)
3316 sign = 1;
3317 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3318 {
3319 size = 2;
3320 if (i.types[n] & (Imm8 | Imm8S))
3321 size = 1;
3322 if (i.types[n] & Imm64)
3323 size = 8;
3324 }
3325
3326 p = frag_more (size);
3327 reloc_type = reloc (size, 0, sign, i.reloc[n]);
3328 #ifdef BFD_ASSEMBLER
3329 if (reloc_type == BFD_RELOC_32
3330 && GOT_symbol
3331 && GOT_symbol == i.op[n].imms->X_add_symbol
3332 && (i.op[n].imms->X_op == O_symbol
3333 || (i.op[n].imms->X_op == O_add
3334 && ((symbol_get_value_expression
3335 (i.op[n].imms->X_op_symbol)->X_op)
3336 == O_subtract))))
3337 {
3338 /* We don't support dynamic linking on x86-64 yet. */
3339 if (flag_code == CODE_64BIT)
3340 abort ();
3341 reloc_type = BFD_RELOC_386_GOTPC;
3342 i.op[n].imms->X_add_number += 3;
3343 }
3344 #endif
3345 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3346 i.op[n].imms, 0, reloc_type);
3347 }
3348 }
3349 }
3350 }
3351 \f
3352 #ifndef LEX_AT
3353 static char *lex_got PARAMS ((RELOC_ENUM *, int *));
3354
3355 /* Parse operands of the form
3356 <symbol>@GOTOFF+<nnn>
3357 and similar .plt or .got references.
3358
3359 If we find one, set up the correct relocation in RELOC and copy the
3360 input string, minus the `@GOTOFF' into a malloc'd buffer for
3361 parsing by the calling routine. Return this buffer, and if ADJUST
3362 is non-null set it to the length of the string we removed from the
3363 input line. Otherwise return NULL. */
3364 static char *
3365 lex_got (reloc, adjust)
3366 RELOC_ENUM *reloc;
3367 int *adjust;
3368 {
3369 static const char * const mode_name[NUM_FLAG_CODE] = { "32", "16", "64" };
3370 static const struct {
3371 const char *str;
3372 const RELOC_ENUM rel[NUM_FLAG_CODE];
3373 } gotrel[] = {
3374 { "PLT", { BFD_RELOC_386_PLT32, 0, BFD_RELOC_X86_64_PLT32 } },
3375 { "GOTOFF", { BFD_RELOC_386_GOTOFF, 0, 0 } },
3376 { "GOTPCREL", { 0, 0, BFD_RELOC_X86_64_GOTPCREL } },
3377 { "GOT", { BFD_RELOC_386_GOT32, 0, BFD_RELOC_X86_64_GOT32 } }
3378 };
3379 char *cp;
3380 unsigned int j;
3381
3382 for (cp = input_line_pointer; *cp != '@'; cp++)
3383 if (is_end_of_line[(unsigned char) *cp])
3384 return NULL;
3385
3386 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
3387 {
3388 int len;
3389
3390 len = strlen (gotrel[j].str);
3391 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
3392 {
3393 if (gotrel[j].rel[(unsigned int) flag_code] != 0)
3394 {
3395 int first, second;
3396 char *tmpbuf, *past_reloc;
3397
3398 *reloc = gotrel[j].rel[(unsigned int) flag_code];
3399 if (adjust)
3400 *adjust = len;
3401
3402 if (GOT_symbol == NULL)
3403 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3404
3405 /* Replace the relocation token with ' ', so that
3406 errors like foo@GOTOFF1 will be detected. */
3407
3408 /* The length of the first part of our input line. */
3409 first = cp - input_line_pointer;
3410
3411 /* The second part goes from after the reloc token until
3412 (and including) an end_of_line char. Don't use strlen
3413 here as the end_of_line char may not be a NUL. */
3414 past_reloc = cp + 1 + len;
3415 for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
3416 ;
3417 second = cp - past_reloc;
3418
3419 /* Allocate and copy string. The trailing NUL shouldn't
3420 be necessary, but be safe. */
3421 tmpbuf = xmalloc (first + second + 2);
3422 memcpy (tmpbuf, input_line_pointer, first);
3423 tmpbuf[first] = ' ';
3424 memcpy (tmpbuf + first + 1, past_reloc, second);
3425 tmpbuf[first + second + 1] = '\0';
3426 return tmpbuf;
3427 }
3428
3429 as_bad (_("@%s reloc is not supported in %s bit mode"),
3430 gotrel[j].str, mode_name[(unsigned int) flag_code]);
3431 return NULL;
3432 }
3433 }
3434
3435 /* Might be a symbol version string. Don't as_bad here. */
3436 return NULL;
3437 }
3438
3439 /* x86_cons_fix_new is called via the expression parsing code when a
3440 reloc is needed. We use this hook to get the correct .got reloc. */
3441 static RELOC_ENUM got_reloc = NO_RELOC;
3442
3443 void
3444 x86_cons_fix_new (frag, off, len, exp)
3445 fragS *frag;
3446 unsigned int off;
3447 unsigned int len;
3448 expressionS *exp;
3449 {
3450 RELOC_ENUM r = reloc (len, 0, 0, got_reloc);
3451 got_reloc = NO_RELOC;
3452 fix_new_exp (frag, off, len, exp, 0, r);
3453 }
3454
3455 void
3456 x86_cons (exp, size)
3457 expressionS *exp;
3458 int size;
3459 {
3460 if (size == 4)
3461 {
3462 /* Handle @GOTOFF and the like in an expression. */
3463 char *save;
3464 char *gotfree_input_line;
3465 int adjust;
3466
3467 save = input_line_pointer;
3468 gotfree_input_line = lex_got (&got_reloc, &adjust);
3469 if (gotfree_input_line)
3470 input_line_pointer = gotfree_input_line;
3471
3472 expression (exp);
3473
3474 if (gotfree_input_line)
3475 {
3476 /* expression () has merrily parsed up to the end of line,
3477 or a comma - in the wrong buffer. Transfer how far
3478 input_line_pointer has moved to the right buffer. */
3479 input_line_pointer = (save
3480 + (input_line_pointer - gotfree_input_line)
3481 + adjust);
3482 free (gotfree_input_line);
3483 }
3484 }
3485 else
3486 expression (exp);
3487 }
3488 #endif
3489
3490 static int i386_immediate PARAMS ((char *));
3491
3492 static int
3493 i386_immediate (imm_start)
3494 char *imm_start;
3495 {
3496 char *save_input_line_pointer;
3497 #ifndef LEX_AT
3498 char *gotfree_input_line;
3499 #endif
3500 segT exp_seg = 0;
3501 expressionS *exp;
3502
3503 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
3504 {
3505 as_bad (_("only 1 or 2 immediate operands are allowed"));
3506 return 0;
3507 }
3508
3509 exp = &im_expressions[i.imm_operands++];
3510 i.op[this_operand].imms = exp;
3511
3512 if (is_space_char (*imm_start))
3513 ++imm_start;
3514
3515 save_input_line_pointer = input_line_pointer;
3516 input_line_pointer = imm_start;
3517
3518 #ifndef LEX_AT
3519 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3520 if (gotfree_input_line)
3521 input_line_pointer = gotfree_input_line;
3522 #endif
3523
3524 exp_seg = expression (exp);
3525
3526 SKIP_WHITESPACE ();
3527 if (*input_line_pointer)
3528 as_bad (_("junk `%s' after expression"), input_line_pointer);
3529
3530 input_line_pointer = save_input_line_pointer;
3531 #ifndef LEX_AT
3532 if (gotfree_input_line)
3533 free (gotfree_input_line);
3534 #endif
3535
3536 if (exp->X_op == O_absent || exp->X_op == O_big)
3537 {
3538 /* Missing or bad expr becomes absolute 0. */
3539 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
3540 imm_start);
3541 exp->X_op = O_constant;
3542 exp->X_add_number = 0;
3543 exp->X_add_symbol = (symbolS *) 0;
3544 exp->X_op_symbol = (symbolS *) 0;
3545 }
3546 else if (exp->X_op == O_constant)
3547 {
3548 /* Size it properly later. */
3549 i.types[this_operand] |= Imm64;
3550 /* If BFD64, sign extend val. */
3551 if (!use_rela_relocations)
3552 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
3553 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
3554 }
3555 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3556 else if (1
3557 #ifdef BFD_ASSEMBLER
3558 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3559 #endif
3560 && exp_seg != text_section
3561 && exp_seg != data_section
3562 && exp_seg != bss_section
3563 && exp_seg != undefined_section
3564 #ifdef BFD_ASSEMBLER
3565 && !bfd_is_com_section (exp_seg)
3566 #endif
3567 )
3568 {
3569 #ifdef BFD_ASSEMBLER
3570 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3571 #else
3572 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3573 #endif
3574 return 0;
3575 }
3576 #endif
3577 else
3578 {
3579 /* This is an address. The size of the address will be
3580 determined later, depending on destination register,
3581 suffix, or the default for the section. */
3582 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
3583 }
3584
3585 return 1;
3586 }
3587
3588 static char *i386_scale PARAMS ((char *));
3589
3590 static char *
3591 i386_scale (scale)
3592 char *scale;
3593 {
3594 offsetT val;
3595 char *save = input_line_pointer;
3596
3597 input_line_pointer = scale;
3598 val = get_absolute_expression ();
3599
3600 switch (val)
3601 {
3602 case 0:
3603 case 1:
3604 i.log2_scale_factor = 0;
3605 break;
3606 case 2:
3607 i.log2_scale_factor = 1;
3608 break;
3609 case 4:
3610 i.log2_scale_factor = 2;
3611 break;
3612 case 8:
3613 i.log2_scale_factor = 3;
3614 break;
3615 default:
3616 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
3617 scale);
3618 input_line_pointer = save;
3619 return NULL;
3620 }
3621 if (i.log2_scale_factor != 0 && i.index_reg == 0)
3622 {
3623 as_warn (_("scale factor of %d without an index register"),
3624 1 << i.log2_scale_factor);
3625 #if SCALE1_WHEN_NO_INDEX
3626 i.log2_scale_factor = 0;
3627 #endif
3628 }
3629 scale = input_line_pointer;
3630 input_line_pointer = save;
3631 return scale;
3632 }
3633
3634 static int i386_displacement PARAMS ((char *, char *));
3635
3636 static int
3637 i386_displacement (disp_start, disp_end)
3638 char *disp_start;
3639 char *disp_end;
3640 {
3641 expressionS *exp;
3642 segT exp_seg = 0;
3643 char *save_input_line_pointer;
3644 #ifndef LEX_AT
3645 char *gotfree_input_line;
3646 #endif
3647 int bigdisp = Disp32;
3648
3649 if (flag_code == CODE_64BIT)
3650 {
3651 if (i.prefix[ADDR_PREFIX] == 0)
3652 bigdisp = Disp64;
3653 }
3654 else if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3655 bigdisp = Disp16;
3656 i.types[this_operand] |= bigdisp;
3657
3658 exp = &disp_expressions[i.disp_operands];
3659 i.op[this_operand].disps = exp;
3660 i.disp_operands++;
3661 save_input_line_pointer = input_line_pointer;
3662 input_line_pointer = disp_start;
3663 END_STRING_AND_SAVE (disp_end);
3664
3665 #ifndef GCC_ASM_O_HACK
3666 #define GCC_ASM_O_HACK 0
3667 #endif
3668 #if GCC_ASM_O_HACK
3669 END_STRING_AND_SAVE (disp_end + 1);
3670 if ((i.types[this_operand] & BaseIndex) != 0
3671 && displacement_string_end[-1] == '+')
3672 {
3673 /* This hack is to avoid a warning when using the "o"
3674 constraint within gcc asm statements.
3675 For instance:
3676
3677 #define _set_tssldt_desc(n,addr,limit,type) \
3678 __asm__ __volatile__ ( \
3679 "movw %w2,%0\n\t" \
3680 "movw %w1,2+%0\n\t" \
3681 "rorl $16,%1\n\t" \
3682 "movb %b1,4+%0\n\t" \
3683 "movb %4,5+%0\n\t" \
3684 "movb $0,6+%0\n\t" \
3685 "movb %h1,7+%0\n\t" \
3686 "rorl $16,%1" \
3687 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
3688
3689 This works great except that the output assembler ends
3690 up looking a bit weird if it turns out that there is
3691 no offset. You end up producing code that looks like:
3692
3693 #APP
3694 movw $235,(%eax)
3695 movw %dx,2+(%eax)
3696 rorl $16,%edx
3697 movb %dl,4+(%eax)
3698 movb $137,5+(%eax)
3699 movb $0,6+(%eax)
3700 movb %dh,7+(%eax)
3701 rorl $16,%edx
3702 #NO_APP
3703
3704 So here we provide the missing zero. */
3705
3706 *displacement_string_end = '0';
3707 }
3708 #endif
3709 #ifndef LEX_AT
3710 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3711 if (gotfree_input_line)
3712 input_line_pointer = gotfree_input_line;
3713 #endif
3714
3715 exp_seg = expression (exp);
3716
3717 SKIP_WHITESPACE ();
3718 if (*input_line_pointer)
3719 as_bad (_("junk `%s' after expression"), input_line_pointer);
3720 #if GCC_ASM_O_HACK
3721 RESTORE_END_STRING (disp_end + 1);
3722 #endif
3723 RESTORE_END_STRING (disp_end);
3724 input_line_pointer = save_input_line_pointer;
3725 #ifndef LEX_AT
3726 if (gotfree_input_line)
3727 free (gotfree_input_line);
3728 #endif
3729
3730 #ifdef BFD_ASSEMBLER
3731 /* We do this to make sure that the section symbol is in
3732 the symbol table. We will ultimately change the relocation
3733 to be relative to the beginning of the section. */
3734 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
3735 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3736 {
3737 if (exp->X_op != O_symbol)
3738 {
3739 as_bad (_("bad expression used with @%s"),
3740 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
3741 ? "GOTPCREL"
3742 : "GOTOFF"));
3743 return 0;
3744 }
3745
3746 if (S_IS_LOCAL (exp->X_add_symbol)
3747 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
3748 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
3749 exp->X_op = O_subtract;
3750 exp->X_op_symbol = GOT_symbol;
3751 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3752 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
3753 else
3754 i.reloc[this_operand] = BFD_RELOC_32;
3755 }
3756 #endif
3757
3758 if (exp->X_op == O_absent || exp->X_op == O_big)
3759 {
3760 /* Missing or bad expr becomes absolute 0. */
3761 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
3762 disp_start);
3763 exp->X_op = O_constant;
3764 exp->X_add_number = 0;
3765 exp->X_add_symbol = (symbolS *) 0;
3766 exp->X_op_symbol = (symbolS *) 0;
3767 }
3768
3769 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3770 if (exp->X_op != O_constant
3771 #ifdef BFD_ASSEMBLER
3772 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3773 #endif
3774 && exp_seg != text_section
3775 && exp_seg != data_section
3776 && exp_seg != bss_section
3777 && exp_seg != undefined_section)
3778 {
3779 #ifdef BFD_ASSEMBLER
3780 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3781 #else
3782 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3783 #endif
3784 return 0;
3785 }
3786 #endif
3787 else if (flag_code == CODE_64BIT)
3788 i.types[this_operand] |= Disp32S | Disp32;
3789 return 1;
3790 }
3791
3792 static int i386_index_check PARAMS ((const char *));
3793
3794 /* Make sure the memory operand we've been dealt is valid.
3795 Return 1 on success, 0 on a failure. */
3796
3797 static int
3798 i386_index_check (operand_string)
3799 const char *operand_string;
3800 {
3801 int ok;
3802 #if INFER_ADDR_PREFIX
3803 int fudged = 0;
3804
3805 tryprefix:
3806 #endif
3807 ok = 1;
3808 if (flag_code == CODE_64BIT)
3809 {
3810 if (i.prefix[ADDR_PREFIX] == 0)
3811 {
3812 /* 64bit checks. */
3813 if ((i.base_reg
3814 && ((i.base_reg->reg_type & Reg64) == 0)
3815 && (i.base_reg->reg_type != BaseIndex
3816 || i.index_reg))
3817 || (i.index_reg
3818 && ((i.index_reg->reg_type & (Reg64 | BaseIndex))
3819 != (Reg64 | BaseIndex))))
3820 ok = 0;
3821 }
3822 else
3823 {
3824 /* 32bit checks. */
3825 if ((i.base_reg
3826 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
3827 || (i.index_reg
3828 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
3829 != (Reg32 | BaseIndex))))
3830 ok = 0;
3831 }
3832 }
3833 else
3834 {
3835 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3836 {
3837 /* 16bit checks. */
3838 if ((i.base_reg
3839 && ((i.base_reg->reg_type & (Reg16 | BaseIndex | RegRex))
3840 != (Reg16 | BaseIndex)))
3841 || (i.index_reg
3842 && (((i.index_reg->reg_type & (Reg16 | BaseIndex))
3843 != (Reg16 | BaseIndex))
3844 || !(i.base_reg
3845 && i.base_reg->reg_num < 6
3846 && i.index_reg->reg_num >= 6
3847 && i.log2_scale_factor == 0))))
3848 ok = 0;
3849 }
3850 else
3851 {
3852 /* 32bit checks. */
3853 if ((i.base_reg
3854 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
3855 || (i.index_reg
3856 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
3857 != (Reg32 | BaseIndex))))
3858 ok = 0;
3859 }
3860 }
3861 if (!ok)
3862 {
3863 #if INFER_ADDR_PREFIX
3864 if (flag_code != CODE_64BIT
3865 && i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0')
3866 {
3867 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
3868 i.prefixes += 1;
3869 /* Change the size of any displacement too. At most one of
3870 Disp16 or Disp32 is set.
3871 FIXME. There doesn't seem to be any real need for separate
3872 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
3873 Removing them would probably clean up the code quite a lot. */
3874 if (i.types[this_operand] & (Disp16 | Disp32))
3875 i.types[this_operand] ^= (Disp16 | Disp32);
3876 fudged = 1;
3877 goto tryprefix;
3878 }
3879 if (fudged)
3880 as_bad (_("`%s' is not a valid base/index expression"),
3881 operand_string);
3882 else
3883 #endif
3884 as_bad (_("`%s' is not a valid %s bit base/index expression"),
3885 operand_string,
3886 flag_code_names[flag_code]);
3887 return 0;
3888 }
3889 return 1;
3890 }
3891
3892 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
3893 on error. */
3894
3895 static int
3896 i386_operand (operand_string)
3897 char *operand_string;
3898 {
3899 const reg_entry *r;
3900 char *end_op;
3901 char *op_string = operand_string;
3902
3903 if (is_space_char (*op_string))
3904 ++op_string;
3905
3906 /* We check for an absolute prefix (differentiating,
3907 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
3908 if (*op_string == ABSOLUTE_PREFIX)
3909 {
3910 ++op_string;
3911 if (is_space_char (*op_string))
3912 ++op_string;
3913 i.types[this_operand] |= JumpAbsolute;
3914 }
3915
3916 /* Check if operand is a register. */
3917 if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
3918 && (r = parse_register (op_string, &end_op)) != NULL)
3919 {
3920 /* Check for a segment override by searching for ':' after a
3921 segment register. */
3922 op_string = end_op;
3923 if (is_space_char (*op_string))
3924 ++op_string;
3925 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
3926 {
3927 switch (r->reg_num)
3928 {
3929 case 0:
3930 i.seg[i.mem_operands] = &es;
3931 break;
3932 case 1:
3933 i.seg[i.mem_operands] = &cs;
3934 break;
3935 case 2:
3936 i.seg[i.mem_operands] = &ss;
3937 break;
3938 case 3:
3939 i.seg[i.mem_operands] = &ds;
3940 break;
3941 case 4:
3942 i.seg[i.mem_operands] = &fs;
3943 break;
3944 case 5:
3945 i.seg[i.mem_operands] = &gs;
3946 break;
3947 }
3948
3949 /* Skip the ':' and whitespace. */
3950 ++op_string;
3951 if (is_space_char (*op_string))
3952 ++op_string;
3953
3954 if (!is_digit_char (*op_string)
3955 && !is_identifier_char (*op_string)
3956 && *op_string != '('
3957 && *op_string != ABSOLUTE_PREFIX)
3958 {
3959 as_bad (_("bad memory operand `%s'"), op_string);
3960 return 0;
3961 }
3962 /* Handle case of %es:*foo. */
3963 if (*op_string == ABSOLUTE_PREFIX)
3964 {
3965 ++op_string;
3966 if (is_space_char (*op_string))
3967 ++op_string;
3968 i.types[this_operand] |= JumpAbsolute;
3969 }
3970 goto do_memory_reference;
3971 }
3972 if (*op_string)
3973 {
3974 as_bad (_("junk `%s' after register"), op_string);
3975 return 0;
3976 }
3977 i.types[this_operand] |= r->reg_type & ~BaseIndex;
3978 i.op[this_operand].regs = r;
3979 i.reg_operands++;
3980 }
3981 else if (*op_string == REGISTER_PREFIX)
3982 {
3983 as_bad (_("bad register name `%s'"), op_string);
3984 return 0;
3985 }
3986 else if (*op_string == IMMEDIATE_PREFIX)
3987 {
3988 ++op_string;
3989 if (i.types[this_operand] & JumpAbsolute)
3990 {
3991 as_bad (_("immediate operand illegal with absolute jump"));
3992 return 0;
3993 }
3994 if (!i386_immediate (op_string))
3995 return 0;
3996 }
3997 else if (is_digit_char (*op_string)
3998 || is_identifier_char (*op_string)
3999 || *op_string == '(')
4000 {
4001 /* This is a memory reference of some sort. */
4002 char *base_string;
4003
4004 /* Start and end of displacement string expression (if found). */
4005 char *displacement_string_start;
4006 char *displacement_string_end;
4007
4008 do_memory_reference:
4009 if ((i.mem_operands == 1
4010 && (current_templates->start->opcode_modifier & IsString) == 0)
4011 || i.mem_operands == 2)
4012 {
4013 as_bad (_("too many memory references for `%s'"),
4014 current_templates->start->name);
4015 return 0;
4016 }
4017
4018 /* Check for base index form. We detect the base index form by
4019 looking for an ')' at the end of the operand, searching
4020 for the '(' matching it, and finding a REGISTER_PREFIX or ','
4021 after the '('. */
4022 base_string = op_string + strlen (op_string);
4023
4024 --base_string;
4025 if (is_space_char (*base_string))
4026 --base_string;
4027
4028 /* If we only have a displacement, set-up for it to be parsed later. */
4029 displacement_string_start = op_string;
4030 displacement_string_end = base_string + 1;
4031
4032 if (*base_string == ')')
4033 {
4034 char *temp_string;
4035 unsigned int parens_balanced = 1;
4036 /* We've already checked that the number of left & right ()'s are
4037 equal, so this loop will not be infinite. */
4038 do
4039 {
4040 base_string--;
4041 if (*base_string == ')')
4042 parens_balanced++;
4043 if (*base_string == '(')
4044 parens_balanced--;
4045 }
4046 while (parens_balanced);
4047
4048 temp_string = base_string;
4049
4050 /* Skip past '(' and whitespace. */
4051 ++base_string;
4052 if (is_space_char (*base_string))
4053 ++base_string;
4054
4055 if (*base_string == ','
4056 || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
4057 && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
4058 {
4059 displacement_string_end = temp_string;
4060
4061 i.types[this_operand] |= BaseIndex;
4062
4063 if (i.base_reg)
4064 {
4065 base_string = end_op;
4066 if (is_space_char (*base_string))
4067 ++base_string;
4068 }
4069
4070 /* There may be an index reg or scale factor here. */
4071 if (*base_string == ',')
4072 {
4073 ++base_string;
4074 if (is_space_char (*base_string))
4075 ++base_string;
4076
4077 if ((*base_string == REGISTER_PREFIX || allow_naked_reg)
4078 && (i.index_reg = parse_register (base_string, &end_op)) != NULL)
4079 {
4080 base_string = end_op;
4081 if (is_space_char (*base_string))
4082 ++base_string;
4083 if (*base_string == ',')
4084 {
4085 ++base_string;
4086 if (is_space_char (*base_string))
4087 ++base_string;
4088 }
4089 else if (*base_string != ')')
4090 {
4091 as_bad (_("expecting `,' or `)' after index register in `%s'"),
4092 operand_string);
4093 return 0;
4094 }
4095 }
4096 else if (*base_string == REGISTER_PREFIX)
4097 {
4098 as_bad (_("bad register name `%s'"), base_string);
4099 return 0;
4100 }
4101
4102 /* Check for scale factor. */
4103 if (*base_string != ')')
4104 {
4105 char *end_scale = i386_scale (base_string);
4106
4107 if (!end_scale)
4108 return 0;
4109
4110 base_string = end_scale;
4111 if (is_space_char (*base_string))
4112 ++base_string;
4113 if (*base_string != ')')
4114 {
4115 as_bad (_("expecting `)' after scale factor in `%s'"),
4116 operand_string);
4117 return 0;
4118 }
4119 }
4120 else if (!i.index_reg)
4121 {
4122 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
4123 *base_string);
4124 return 0;
4125 }
4126 }
4127 else if (*base_string != ')')
4128 {
4129 as_bad (_("expecting `,' or `)' after base register in `%s'"),
4130 operand_string);
4131 return 0;
4132 }
4133 }
4134 else if (*base_string == REGISTER_PREFIX)
4135 {
4136 as_bad (_("bad register name `%s'"), base_string);
4137 return 0;
4138 }
4139 }
4140
4141 /* If there's an expression beginning the operand, parse it,
4142 assuming displacement_string_start and
4143 displacement_string_end are meaningful. */
4144 if (displacement_string_start != displacement_string_end)
4145 {
4146 if (!i386_displacement (displacement_string_start,
4147 displacement_string_end))
4148 return 0;
4149 }
4150
4151 /* Special case for (%dx) while doing input/output op. */
4152 if (i.base_reg
4153 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
4154 && i.index_reg == 0
4155 && i.log2_scale_factor == 0
4156 && i.seg[i.mem_operands] == 0
4157 && (i.types[this_operand] & Disp) == 0)
4158 {
4159 i.types[this_operand] = InOutPortReg;
4160 return 1;
4161 }
4162
4163 if (i386_index_check (operand_string) == 0)
4164 return 0;
4165 i.mem_operands++;
4166 }
4167 else
4168 {
4169 /* It's not a memory operand; argh! */
4170 as_bad (_("invalid char %s beginning operand %d `%s'"),
4171 output_invalid (*op_string),
4172 this_operand + 1,
4173 op_string);
4174 return 0;
4175 }
4176 return 1; /* Normal return. */
4177 }
4178 \f
4179 /* md_estimate_size_before_relax()
4180
4181 Called just before relax() for rs_machine_dependent frags. The x86
4182 assembler uses these frags to handle variable size jump
4183 instructions.
4184
4185 Any symbol that is now undefined will not become defined.
4186 Return the correct fr_subtype in the frag.
4187 Return the initial "guess for variable size of frag" to caller.
4188 The guess is actually the growth beyond the fixed part. Whatever
4189 we do to grow the fixed or variable part contributes to our
4190 returned value. */
4191
4192 int
4193 md_estimate_size_before_relax (fragP, segment)
4194 fragS *fragP;
4195 segT segment;
4196 {
4197 /* We've already got fragP->fr_subtype right; all we have to do is
4198 check for un-relaxable symbols. On an ELF system, we can't relax
4199 an externally visible symbol, because it may be overridden by a
4200 shared library. */
4201 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
4202 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4203 || S_IS_EXTERNAL (fragP->fr_symbol)
4204 || S_IS_WEAK (fragP->fr_symbol)
4205 #endif
4206 )
4207 {
4208 /* Symbol is undefined in this segment, or we need to keep a
4209 reloc so that weak symbols can be overridden. */
4210 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
4211 RELOC_ENUM reloc_type;
4212 unsigned char *opcode;
4213 int old_fr_fix;
4214
4215 if (fragP->fr_var != NO_RELOC)
4216 reloc_type = fragP->fr_var;
4217 else if (size == 2)
4218 reloc_type = BFD_RELOC_16_PCREL;
4219 else
4220 reloc_type = BFD_RELOC_32_PCREL;
4221
4222 old_fr_fix = fragP->fr_fix;
4223 opcode = (unsigned char *) fragP->fr_opcode;
4224
4225 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
4226 {
4227 case UNCOND_JUMP:
4228 /* Make jmp (0xeb) a (d)word displacement jump. */
4229 opcode[0] = 0xe9;
4230 fragP->fr_fix += size;
4231 fix_new (fragP, old_fr_fix, size,
4232 fragP->fr_symbol,
4233 fragP->fr_offset, 1,
4234 reloc_type);
4235 break;
4236
4237 case COND_JUMP86:
4238 if (size == 2
4239 && (!no_cond_jump_promotion || fragP->fr_var != NO_RELOC))
4240 {
4241 /* Negate the condition, and branch past an
4242 unconditional jump. */
4243 opcode[0] ^= 1;
4244 opcode[1] = 3;
4245 /* Insert an unconditional jump. */
4246 opcode[2] = 0xe9;
4247 /* We added two extra opcode bytes, and have a two byte
4248 offset. */
4249 fragP->fr_fix += 2 + 2;
4250 fix_new (fragP, old_fr_fix + 2, 2,
4251 fragP->fr_symbol,
4252 fragP->fr_offset, 1,
4253 reloc_type);
4254 break;
4255 }
4256 /* Fall through. */
4257
4258 case COND_JUMP:
4259 if (no_cond_jump_promotion && fragP->fr_var == NO_RELOC)
4260 {
4261 fragP->fr_fix += 1;
4262 fix_new (fragP, old_fr_fix, 1,
4263 fragP->fr_symbol,
4264 fragP->fr_offset, 1,
4265 BFD_RELOC_8_PCREL);
4266 break;
4267 }
4268
4269 /* This changes the byte-displacement jump 0x7N
4270 to the (d)word-displacement jump 0x0f,0x8N. */
4271 opcode[1] = opcode[0] + 0x10;
4272 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4273 /* We've added an opcode byte. */
4274 fragP->fr_fix += 1 + size;
4275 fix_new (fragP, old_fr_fix + 1, size,
4276 fragP->fr_symbol,
4277 fragP->fr_offset, 1,
4278 reloc_type);
4279 break;
4280
4281 default:
4282 BAD_CASE (fragP->fr_subtype);
4283 break;
4284 }
4285 frag_wane (fragP);
4286 return fragP->fr_fix - old_fr_fix;
4287 }
4288
4289 /* Guess size depending on current relax state. Initially the relax
4290 state will correspond to a short jump and we return 1, because
4291 the variable part of the frag (the branch offset) is one byte
4292 long. However, we can relax a section more than once and in that
4293 case we must either set fr_subtype back to the unrelaxed state,
4294 or return the value for the appropriate branch. */
4295 return md_relax_table[fragP->fr_subtype].rlx_length;
4296 }
4297
4298 /* Called after relax() is finished.
4299
4300 In: Address of frag.
4301 fr_type == rs_machine_dependent.
4302 fr_subtype is what the address relaxed to.
4303
4304 Out: Any fixSs and constants are set up.
4305 Caller will turn frag into a ".space 0". */
4306
4307 #ifndef BFD_ASSEMBLER
4308 void
4309 md_convert_frag (headers, sec, fragP)
4310 object_headers *headers ATTRIBUTE_UNUSED;
4311 segT sec ATTRIBUTE_UNUSED;
4312 fragS *fragP;
4313 #else
4314 void
4315 md_convert_frag (abfd, sec, fragP)
4316 bfd *abfd ATTRIBUTE_UNUSED;
4317 segT sec ATTRIBUTE_UNUSED;
4318 fragS *fragP;
4319 #endif
4320 {
4321 unsigned char *opcode;
4322 unsigned char *where_to_put_displacement = NULL;
4323 offsetT target_address;
4324 offsetT opcode_address;
4325 unsigned int extension = 0;
4326 offsetT displacement_from_opcode_start;
4327
4328 opcode = (unsigned char *) fragP->fr_opcode;
4329
4330 /* Address we want to reach in file space. */
4331 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
4332
4333 /* Address opcode resides at in file space. */
4334 opcode_address = fragP->fr_address + fragP->fr_fix;
4335
4336 /* Displacement from opcode start to fill into instruction. */
4337 displacement_from_opcode_start = target_address - opcode_address;
4338
4339 if ((fragP->fr_subtype & BIG) == 0)
4340 {
4341 /* Don't have to change opcode. */
4342 extension = 1; /* 1 opcode + 1 displacement */
4343 where_to_put_displacement = &opcode[1];
4344 }
4345 else
4346 {
4347 if (no_cond_jump_promotion
4348 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
4349 as_warn_where (fragP->fr_file, fragP->fr_line, _("long jump required"));
4350
4351 switch (fragP->fr_subtype)
4352 {
4353 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
4354 extension = 4; /* 1 opcode + 4 displacement */
4355 opcode[0] = 0xe9;
4356 where_to_put_displacement = &opcode[1];
4357 break;
4358
4359 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
4360 extension = 2; /* 1 opcode + 2 displacement */
4361 opcode[0] = 0xe9;
4362 where_to_put_displacement = &opcode[1];
4363 break;
4364
4365 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
4366 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
4367 extension = 5; /* 2 opcode + 4 displacement */
4368 opcode[1] = opcode[0] + 0x10;
4369 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4370 where_to_put_displacement = &opcode[2];
4371 break;
4372
4373 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
4374 extension = 3; /* 2 opcode + 2 displacement */
4375 opcode[1] = opcode[0] + 0x10;
4376 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4377 where_to_put_displacement = &opcode[2];
4378 break;
4379
4380 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
4381 extension = 4;
4382 opcode[0] ^= 1;
4383 opcode[1] = 3;
4384 opcode[2] = 0xe9;
4385 where_to_put_displacement = &opcode[3];
4386 break;
4387
4388 default:
4389 BAD_CASE (fragP->fr_subtype);
4390 break;
4391 }
4392 }
4393
4394 /* Now put displacement after opcode. */
4395 md_number_to_chars ((char *) where_to_put_displacement,
4396 (valueT) (displacement_from_opcode_start - extension),
4397 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
4398 fragP->fr_fix += extension;
4399 }
4400 \f
4401 /* Size of byte displacement jmp. */
4402 int md_short_jump_size = 2;
4403
4404 /* Size of dword displacement jmp. */
4405 int md_long_jump_size = 5;
4406
4407 /* Size of relocation record. */
4408 const int md_reloc_size = 8;
4409
4410 void
4411 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
4412 char *ptr;
4413 addressT from_addr, to_addr;
4414 fragS *frag ATTRIBUTE_UNUSED;
4415 symbolS *to_symbol ATTRIBUTE_UNUSED;
4416 {
4417 offsetT offset;
4418
4419 offset = to_addr - (from_addr + 2);
4420 /* Opcode for byte-disp jump. */
4421 md_number_to_chars (ptr, (valueT) 0xeb, 1);
4422 md_number_to_chars (ptr + 1, (valueT) offset, 1);
4423 }
4424
4425 void
4426 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
4427 char *ptr;
4428 addressT from_addr, to_addr;
4429 fragS *frag ATTRIBUTE_UNUSED;
4430 symbolS *to_symbol ATTRIBUTE_UNUSED;
4431 {
4432 offsetT offset;
4433
4434 offset = to_addr - (from_addr + 5);
4435 md_number_to_chars (ptr, (valueT) 0xe9, 1);
4436 md_number_to_chars (ptr + 1, (valueT) offset, 4);
4437 }
4438 \f
4439 /* Apply a fixup (fixS) to segment data, once it has been determined
4440 by our caller that we have all the info we need to fix it up.
4441
4442 On the 386, immediates, displacements, and data pointers are all in
4443 the same (little-endian) format, so we don't need to care about which
4444 we are handling. */
4445
4446 void
4447 md_apply_fix3 (fixP, valP, seg)
4448 /* The fix we're to put in. */
4449 fixS *fixP;
4450 /* Pointer to the value of the bits. */
4451 valueT * valP;
4452 /* Segment fix is from. */
4453 segT seg ATTRIBUTE_UNUSED;
4454 {
4455 char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
4456 valueT value = * valP;
4457
4458 #if defined (BFD_ASSEMBLER) && !defined (TE_Mach)
4459 if (fixP->fx_pcrel)
4460 {
4461 switch (fixP->fx_r_type)
4462 {
4463 default:
4464 break;
4465
4466 case BFD_RELOC_32:
4467 fixP->fx_r_type = BFD_RELOC_32_PCREL;
4468 break;
4469 case BFD_RELOC_16:
4470 fixP->fx_r_type = BFD_RELOC_16_PCREL;
4471 break;
4472 case BFD_RELOC_8:
4473 fixP->fx_r_type = BFD_RELOC_8_PCREL;
4474 break;
4475 }
4476 }
4477
4478 /* This is a hack. There should be a better way to handle this.
4479 This covers for the fact that bfd_install_relocation will
4480 subtract the current location (for partial_inplace, PC relative
4481 relocations); see more below. */
4482 if ((fixP->fx_r_type == BFD_RELOC_32_PCREL
4483 || fixP->fx_r_type == BFD_RELOC_16_PCREL
4484 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
4485 && fixP->fx_addsy && !use_rela_relocations)
4486 {
4487 #ifndef OBJ_AOUT
4488 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4489 #ifdef TE_PE
4490 || OUTPUT_FLAVOR == bfd_target_coff_flavour
4491 #endif
4492 )
4493 value += fixP->fx_where + fixP->fx_frag->fr_address;
4494 #endif
4495 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4496 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
4497 {
4498 segT fseg = S_GET_SEGMENT (fixP->fx_addsy);
4499
4500 if ((fseg == seg
4501 || (symbol_section_p (fixP->fx_addsy)
4502 && fseg != absolute_section))
4503 && !S_IS_EXTERNAL (fixP->fx_addsy)
4504 && !S_IS_WEAK (fixP->fx_addsy)
4505 && S_IS_DEFINED (fixP->fx_addsy)
4506 && !S_IS_COMMON (fixP->fx_addsy))
4507 {
4508 /* Yes, we add the values in twice. This is because
4509 bfd_perform_relocation subtracts them out again. I think
4510 bfd_perform_relocation is broken, but I don't dare change
4511 it. FIXME. */
4512 value += fixP->fx_where + fixP->fx_frag->fr_address;
4513 }
4514 }
4515 #endif
4516 #if defined (OBJ_COFF) && defined (TE_PE)
4517 /* For some reason, the PE format does not store a section
4518 address offset for a PC relative symbol. */
4519 if (S_GET_SEGMENT (fixP->fx_addsy) != seg)
4520 value += md_pcrel_from (fixP);
4521 #endif
4522 }
4523
4524 /* Fix a few things - the dynamic linker expects certain values here,
4525 and we must not dissappoint it. */
4526 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4527 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4528 && fixP->fx_addsy)
4529 switch (fixP->fx_r_type)
4530 {
4531 case BFD_RELOC_386_PLT32:
4532 case BFD_RELOC_X86_64_PLT32:
4533 /* Make the jump instruction point to the address of the operand. At
4534 runtime we merely add the offset to the actual PLT entry. */
4535 value = -4;
4536 break;
4537 case BFD_RELOC_386_GOTPC:
4538
4539 /* This is tough to explain. We end up with this one if we have
4540 * operands that look like "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal
4541 * here is to obtain the absolute address of the GOT, and it is strongly
4542 * preferable from a performance point of view to avoid using a runtime
4543 * relocation for this. The actual sequence of instructions often look
4544 * something like:
4545 *
4546 * call .L66
4547 * .L66:
4548 * popl %ebx
4549 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
4550 *
4551 * The call and pop essentially return the absolute address of
4552 * the label .L66 and store it in %ebx. The linker itself will
4553 * ultimately change the first operand of the addl so that %ebx points to
4554 * the GOT, but to keep things simple, the .o file must have this operand
4555 * set so that it generates not the absolute address of .L66, but the
4556 * absolute address of itself. This allows the linker itself simply
4557 * treat a GOTPC relocation as asking for a pcrel offset to the GOT to be
4558 * added in, and the addend of the relocation is stored in the operand
4559 * field for the instruction itself.
4560 *
4561 * Our job here is to fix the operand so that it would add the correct
4562 * offset so that %ebx would point to itself. The thing that is tricky is
4563 * that .-.L66 will point to the beginning of the instruction, so we need
4564 * to further modify the operand so that it will point to itself.
4565 * There are other cases where you have something like:
4566 *
4567 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
4568 *
4569 * and here no correction would be required. Internally in the assembler
4570 * we treat operands of this form as not being pcrel since the '.' is
4571 * explicitly mentioned, and I wonder whether it would simplify matters
4572 * to do it this way. Who knows. In earlier versions of the PIC patches,
4573 * the pcrel_adjust field was used to store the correction, but since the
4574 * expression is not pcrel, I felt it would be confusing to do it this
4575 * way. */
4576
4577 value -= 1;
4578 break;
4579 case BFD_RELOC_386_GOT32:
4580 case BFD_RELOC_X86_64_GOT32:
4581 value = 0; /* Fully resolved at runtime. No addend. */
4582 break;
4583 case BFD_RELOC_386_GOTOFF:
4584 case BFD_RELOC_X86_64_GOTPCREL:
4585 break;
4586
4587 case BFD_RELOC_VTABLE_INHERIT:
4588 case BFD_RELOC_VTABLE_ENTRY:
4589 fixP->fx_done = 0;
4590 return;
4591
4592 default:
4593 break;
4594 }
4595 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
4596 * valP = value;
4597 #endif /* defined (BFD_ASSEMBLER) && !defined (TE_Mach) */
4598
4599 /* Are we finished with this relocation now? */
4600 if (fixP->fx_addsy == NULL && fixP->fx_pcrel == 0)
4601 fixP->fx_done = 1;
4602 #ifdef BFD_ASSEMBLER
4603 else if (use_rela_relocations)
4604 {
4605 fixP->fx_no_overflow = 1;
4606 /* Remember value for tc_gen_reloc. */
4607 fixP->fx_addnumber = value;
4608 value = 0;
4609 }
4610 #endif
4611 md_number_to_chars (p, value, fixP->fx_size);
4612 }
4613 \f
4614 #define MAX_LITTLENUMS 6
4615
4616 /* Turn the string pointed to by litP into a floating point constant
4617 of type TYPE, and emit the appropriate bytes. The number of
4618 LITTLENUMS emitted is stored in *SIZEP. An error message is
4619 returned, or NULL on OK. */
4620
4621 char *
4622 md_atof (type, litP, sizeP)
4623 int type;
4624 char *litP;
4625 int *sizeP;
4626 {
4627 int prec;
4628 LITTLENUM_TYPE words[MAX_LITTLENUMS];
4629 LITTLENUM_TYPE *wordP;
4630 char *t;
4631
4632 switch (type)
4633 {
4634 case 'f':
4635 case 'F':
4636 prec = 2;
4637 break;
4638
4639 case 'd':
4640 case 'D':
4641 prec = 4;
4642 break;
4643
4644 case 'x':
4645 case 'X':
4646 prec = 5;
4647 break;
4648
4649 default:
4650 *sizeP = 0;
4651 return _("Bad call to md_atof ()");
4652 }
4653 t = atof_ieee (input_line_pointer, type, words);
4654 if (t)
4655 input_line_pointer = t;
4656
4657 *sizeP = prec * sizeof (LITTLENUM_TYPE);
4658 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
4659 the bigendian 386. */
4660 for (wordP = words + prec - 1; prec--;)
4661 {
4662 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
4663 litP += sizeof (LITTLENUM_TYPE);
4664 }
4665 return 0;
4666 }
4667 \f
4668 char output_invalid_buf[8];
4669
4670 static char *
4671 output_invalid (c)
4672 int c;
4673 {
4674 if (ISPRINT (c))
4675 sprintf (output_invalid_buf, "'%c'", c);
4676 else
4677 sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
4678 return output_invalid_buf;
4679 }
4680
4681 /* REG_STRING starts *before* REGISTER_PREFIX. */
4682
4683 static const reg_entry *
4684 parse_register (reg_string, end_op)
4685 char *reg_string;
4686 char **end_op;
4687 {
4688 char *s = reg_string;
4689 char *p;
4690 char reg_name_given[MAX_REG_NAME_SIZE + 1];
4691 const reg_entry *r;
4692
4693 /* Skip possible REGISTER_PREFIX and possible whitespace. */
4694 if (*s == REGISTER_PREFIX)
4695 ++s;
4696
4697 if (is_space_char (*s))
4698 ++s;
4699
4700 p = reg_name_given;
4701 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
4702 {
4703 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
4704 return (const reg_entry *) NULL;
4705 s++;
4706 }
4707
4708 /* For naked regs, make sure that we are not dealing with an identifier.
4709 This prevents confusing an identifier like `eax_var' with register
4710 `eax'. */
4711 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
4712 return (const reg_entry *) NULL;
4713
4714 *end_op = s;
4715
4716 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
4717
4718 /* Handle floating point regs, allowing spaces in the (i) part. */
4719 if (r == i386_regtab /* %st is first entry of table */)
4720 {
4721 if (is_space_char (*s))
4722 ++s;
4723 if (*s == '(')
4724 {
4725 ++s;
4726 if (is_space_char (*s))
4727 ++s;
4728 if (*s >= '0' && *s <= '7')
4729 {
4730 r = &i386_float_regtab[*s - '0'];
4731 ++s;
4732 if (is_space_char (*s))
4733 ++s;
4734 if (*s == ')')
4735 {
4736 *end_op = s + 1;
4737 return r;
4738 }
4739 }
4740 /* We have "%st(" then garbage. */
4741 return (const reg_entry *) NULL;
4742 }
4743 }
4744
4745 if (r != NULL
4746 && (r->reg_flags & (RegRex64 | RegRex)) != 0
4747 && flag_code != CODE_64BIT)
4748 {
4749 return (const reg_entry *) NULL;
4750 }
4751
4752 return r;
4753 }
4754 \f
4755 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4756 const char *md_shortopts = "kVQ:sq";
4757 #else
4758 const char *md_shortopts = "q";
4759 #endif
4760
4761 struct option md_longopts[] = {
4762 #define OPTION_32 (OPTION_MD_BASE + 0)
4763 {"32", no_argument, NULL, OPTION_32},
4764 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4765 #define OPTION_64 (OPTION_MD_BASE + 1)
4766 {"64", no_argument, NULL, OPTION_64},
4767 #endif
4768 {NULL, no_argument, NULL, 0}
4769 };
4770 size_t md_longopts_size = sizeof (md_longopts);
4771
4772 int
4773 md_parse_option (c, arg)
4774 int c;
4775 char *arg ATTRIBUTE_UNUSED;
4776 {
4777 switch (c)
4778 {
4779 case 'q':
4780 quiet_warnings = 1;
4781 break;
4782
4783 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4784 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
4785 should be emitted or not. FIXME: Not implemented. */
4786 case 'Q':
4787 break;
4788
4789 /* -V: SVR4 argument to print version ID. */
4790 case 'V':
4791 print_version_id ();
4792 break;
4793
4794 /* -k: Ignore for FreeBSD compatibility. */
4795 case 'k':
4796 break;
4797
4798 case 's':
4799 /* -s: On i386 Solaris, this tells the native assembler to use
4800 .stab instead of .stab.excl. We always use .stab anyhow. */
4801 break;
4802
4803 case OPTION_64:
4804 {
4805 const char **list, **l;
4806
4807 list = bfd_target_list ();
4808 for (l = list; *l != NULL; l++)
4809 if (strcmp (*l, "elf64-x86-64") == 0)
4810 {
4811 default_arch = "x86_64";
4812 break;
4813 }
4814 if (*l == NULL)
4815 as_fatal (_("No compiled in support for x86_64"));
4816 free (list);
4817 }
4818 break;
4819 #endif
4820
4821 case OPTION_32:
4822 default_arch = "i386";
4823 break;
4824
4825 default:
4826 return 0;
4827 }
4828 return 1;
4829 }
4830
4831 void
4832 md_show_usage (stream)
4833 FILE *stream;
4834 {
4835 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4836 fprintf (stream, _("\
4837 -Q ignored\n\
4838 -V print assembler version number\n\
4839 -k ignored\n\
4840 -q quieten some warnings\n\
4841 -s ignored\n"));
4842 #else
4843 fprintf (stream, _("\
4844 -q quieten some warnings\n"));
4845 #endif
4846 }
4847
4848 #ifdef BFD_ASSEMBLER
4849 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
4850 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
4851
4852 /* Pick the target format to use. */
4853
4854 const char *
4855 i386_target_format ()
4856 {
4857 if (!strcmp (default_arch, "x86_64"))
4858 set_code_flag (CODE_64BIT);
4859 else if (!strcmp (default_arch, "i386"))
4860 set_code_flag (CODE_32BIT);
4861 else
4862 as_fatal (_("Unknown architecture"));
4863 switch (OUTPUT_FLAVOR)
4864 {
4865 #ifdef OBJ_MAYBE_AOUT
4866 case bfd_target_aout_flavour:
4867 return AOUT_TARGET_FORMAT;
4868 #endif
4869 #ifdef OBJ_MAYBE_COFF
4870 case bfd_target_coff_flavour:
4871 return "coff-i386";
4872 #endif
4873 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
4874 case bfd_target_elf_flavour:
4875 {
4876 if (flag_code == CODE_64BIT)
4877 use_rela_relocations = 1;
4878 return flag_code == CODE_64BIT ? "elf64-x86-64" : "elf32-i386";
4879 }
4880 #endif
4881 default:
4882 abort ();
4883 return NULL;
4884 }
4885 }
4886
4887 #endif /* OBJ_MAYBE_ more than one */
4888
4889 #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
4890 void i386_elf_emit_arch_note ()
4891 {
4892 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4893 && cpu_arch_name != NULL)
4894 {
4895 char *p;
4896 asection *seg = now_seg;
4897 subsegT subseg = now_subseg;
4898 Elf_Internal_Note i_note;
4899 Elf_External_Note e_note;
4900 asection *note_secp;
4901 int len;
4902
4903 /* Create the .note section. */
4904 note_secp = subseg_new (".note", 0);
4905 bfd_set_section_flags (stdoutput,
4906 note_secp,
4907 SEC_HAS_CONTENTS | SEC_READONLY);
4908
4909 /* Process the arch string. */
4910 len = strlen (cpu_arch_name);
4911
4912 i_note.namesz = len + 1;
4913 i_note.descsz = 0;
4914 i_note.type = NT_ARCH;
4915 p = frag_more (sizeof (e_note.namesz));
4916 md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
4917 p = frag_more (sizeof (e_note.descsz));
4918 md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
4919 p = frag_more (sizeof (e_note.type));
4920 md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
4921 p = frag_more (len + 1);
4922 strcpy (p, cpu_arch_name);
4923
4924 frag_align (2, 0, 0);
4925
4926 subseg_set (seg, subseg);
4927 }
4928 }
4929 #endif
4930 #endif /* BFD_ASSEMBLER */
4931 \f
4932 symbolS *
4933 md_undefined_symbol (name)
4934 char *name;
4935 {
4936 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
4937 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
4938 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
4939 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
4940 {
4941 if (!GOT_symbol)
4942 {
4943 if (symbol_find (name))
4944 as_bad (_("GOT already in symbol table"));
4945 GOT_symbol = symbol_new (name, undefined_section,
4946 (valueT) 0, &zero_address_frag);
4947 };
4948 return GOT_symbol;
4949 }
4950 return 0;
4951 }
4952
4953 /* Round up a section size to the appropriate boundary. */
4954
4955 valueT
4956 md_section_align (segment, size)
4957 segT segment ATTRIBUTE_UNUSED;
4958 valueT size;
4959 {
4960 #ifdef BFD_ASSEMBLER
4961 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4962 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
4963 {
4964 /* For a.out, force the section size to be aligned. If we don't do
4965 this, BFD will align it for us, but it will not write out the
4966 final bytes of the section. This may be a bug in BFD, but it is
4967 easier to fix it here since that is how the other a.out targets
4968 work. */
4969 int align;
4970
4971 align = bfd_get_section_alignment (stdoutput, segment);
4972 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
4973 }
4974 #endif
4975 #endif
4976
4977 return size;
4978 }
4979
4980 /* On the i386, PC-relative offsets are relative to the start of the
4981 next instruction. That is, the address of the offset, plus its
4982 size, since the offset is always the last part of the insn. */
4983
4984 long
4985 md_pcrel_from (fixP)
4986 fixS *fixP;
4987 {
4988 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
4989 }
4990
4991 #ifndef I386COFF
4992
4993 static void
4994 s_bss (ignore)
4995 int ignore ATTRIBUTE_UNUSED;
4996 {
4997 int temp;
4998
4999 temp = get_absolute_expression ();
5000 subseg_set (bss_section, (subsegT) temp);
5001 demand_empty_rest_of_line ();
5002 }
5003
5004 #endif
5005
5006 #ifdef BFD_ASSEMBLER
5007
5008 void
5009 i386_validate_fix (fixp)
5010 fixS *fixp;
5011 {
5012 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
5013 {
5014 /* GOTOFF relocation are nonsense in 64bit mode. */
5015 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
5016 {
5017 if (flag_code != CODE_64BIT)
5018 abort ();
5019 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
5020 }
5021 else
5022 {
5023 if (flag_code == CODE_64BIT)
5024 abort ();
5025 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
5026 }
5027 fixp->fx_subsy = 0;
5028 }
5029 }
5030
5031 arelent *
5032 tc_gen_reloc (section, fixp)
5033 asection *section ATTRIBUTE_UNUSED;
5034 fixS *fixp;
5035 {
5036 arelent *rel;
5037 bfd_reloc_code_real_type code;
5038
5039 switch (fixp->fx_r_type)
5040 {
5041 case BFD_RELOC_X86_64_PLT32:
5042 case BFD_RELOC_X86_64_GOT32:
5043 case BFD_RELOC_X86_64_GOTPCREL:
5044 case BFD_RELOC_386_PLT32:
5045 case BFD_RELOC_386_GOT32:
5046 case BFD_RELOC_386_GOTOFF:
5047 case BFD_RELOC_386_GOTPC:
5048 case BFD_RELOC_X86_64_32S:
5049 case BFD_RELOC_RVA:
5050 case BFD_RELOC_VTABLE_ENTRY:
5051 case BFD_RELOC_VTABLE_INHERIT:
5052 code = fixp->fx_r_type;
5053 break;
5054 default:
5055 if (fixp->fx_pcrel)
5056 {
5057 switch (fixp->fx_size)
5058 {
5059 default:
5060 as_bad_where (fixp->fx_file, fixp->fx_line,
5061 _("can not do %d byte pc-relative relocation"),
5062 fixp->fx_size);
5063 code = BFD_RELOC_32_PCREL;
5064 break;
5065 case 1: code = BFD_RELOC_8_PCREL; break;
5066 case 2: code = BFD_RELOC_16_PCREL; break;
5067 case 4: code = BFD_RELOC_32_PCREL; break;
5068 }
5069 }
5070 else
5071 {
5072 switch (fixp->fx_size)
5073 {
5074 default:
5075 as_bad_where (fixp->fx_file, fixp->fx_line,
5076 _("can not do %d byte relocation"),
5077 fixp->fx_size);
5078 code = BFD_RELOC_32;
5079 break;
5080 case 1: code = BFD_RELOC_8; break;
5081 case 2: code = BFD_RELOC_16; break;
5082 case 4: code = BFD_RELOC_32; break;
5083 #ifdef BFD64
5084 case 8: code = BFD_RELOC_64; break;
5085 #endif
5086 }
5087 }
5088 break;
5089 }
5090
5091 if (code == BFD_RELOC_32
5092 && GOT_symbol
5093 && fixp->fx_addsy == GOT_symbol)
5094 {
5095 /* We don't support GOTPC on 64bit targets. */
5096 if (flag_code == CODE_64BIT)
5097 abort ();
5098 code = BFD_RELOC_386_GOTPC;
5099 }
5100
5101 rel = (arelent *) xmalloc (sizeof (arelent));
5102 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
5103 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
5104
5105 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
5106 if (!use_rela_relocations)
5107 {
5108 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
5109 vtable entry to be used in the relocation's section offset. */
5110 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
5111 rel->address = fixp->fx_offset;
5112
5113 if (fixp->fx_pcrel)
5114 rel->addend = fixp->fx_addnumber;
5115 else
5116 rel->addend = 0;
5117 }
5118 /* Use the rela in 64bit mode. */
5119 else
5120 {
5121 if (!fixp->fx_pcrel)
5122 rel->addend = fixp->fx_offset;
5123 else
5124 switch (code)
5125 {
5126 case BFD_RELOC_X86_64_PLT32:
5127 case BFD_RELOC_X86_64_GOT32:
5128 case BFD_RELOC_X86_64_GOTPCREL:
5129 rel->addend = fixp->fx_offset - fixp->fx_size;
5130 break;
5131 default:
5132 rel->addend = (section->vma
5133 - fixp->fx_size
5134 + fixp->fx_addnumber
5135 + md_pcrel_from (fixp));
5136 break;
5137 }
5138 }
5139
5140 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
5141 if (rel->howto == NULL)
5142 {
5143 as_bad_where (fixp->fx_file, fixp->fx_line,
5144 _("cannot represent relocation type %s"),
5145 bfd_get_reloc_code_name (code));
5146 /* Set howto to a garbage value so that we can keep going. */
5147 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
5148 assert (rel->howto != NULL);
5149 }
5150
5151 return rel;
5152 }
5153
5154 #else /* !BFD_ASSEMBLER */
5155
5156 #if (defined(OBJ_AOUT) | defined(OBJ_BOUT))
5157 void
5158 tc_aout_fix_to_chars (where, fixP, segment_address_in_file)
5159 char *where;
5160 fixS *fixP;
5161 relax_addressT segment_address_in_file;
5162 {
5163 /* In: length of relocation (or of address) in chars: 1, 2 or 4.
5164 Out: GNU LD relocation length code: 0, 1, or 2. */
5165
5166 static const unsigned char nbytes_r_length[] = { 42, 0, 1, 42, 2 };
5167 long r_symbolnum;
5168
5169 know (fixP->fx_addsy != NULL);
5170
5171 md_number_to_chars (where,
5172 (valueT) (fixP->fx_frag->fr_address
5173 + fixP->fx_where - segment_address_in_file),
5174 4);
5175
5176 r_symbolnum = (S_IS_DEFINED (fixP->fx_addsy)
5177 ? S_GET_TYPE (fixP->fx_addsy)
5178 : fixP->fx_addsy->sy_number);
5179
5180 where[6] = (r_symbolnum >> 16) & 0x0ff;
5181 where[5] = (r_symbolnum >> 8) & 0x0ff;
5182 where[4] = r_symbolnum & 0x0ff;
5183 where[7] = ((((!S_IS_DEFINED (fixP->fx_addsy)) << 3) & 0x08)
5184 | ((nbytes_r_length[fixP->fx_size] << 1) & 0x06)
5185 | (((fixP->fx_pcrel << 0) & 0x01) & 0x0f));
5186 }
5187
5188 #endif /* OBJ_AOUT or OBJ_BOUT. */
5189
5190 #if defined (I386COFF)
5191
5192 short
5193 tc_coff_fix2rtype (fixP)
5194 fixS *fixP;
5195 {
5196 if (fixP->fx_r_type == R_IMAGEBASE)
5197 return R_IMAGEBASE;
5198
5199 return (fixP->fx_pcrel ?
5200 (fixP->fx_size == 1 ? R_PCRBYTE :
5201 fixP->fx_size == 2 ? R_PCRWORD :
5202 R_PCRLONG) :
5203 (fixP->fx_size == 1 ? R_RELBYTE :
5204 fixP->fx_size == 2 ? R_RELWORD :
5205 R_DIR32));
5206 }
5207
5208 int
5209 tc_coff_sizemachdep (frag)
5210 fragS *frag;
5211 {
5212 if (frag->fr_next)
5213 return (frag->fr_next->fr_address - frag->fr_address);
5214 else
5215 return 0;
5216 }
5217
5218 #endif /* I386COFF */
5219
5220 #endif /* !BFD_ASSEMBLER */
5221 \f
5222 /* Parse operands using Intel syntax. This implements a recursive descent
5223 parser based on the BNF grammar published in Appendix B of the MASM 6.1
5224 Programmer's Guide.
5225
5226 FIXME: We do not recognize the full operand grammar defined in the MASM
5227 documentation. In particular, all the structure/union and
5228 high-level macro operands are missing.
5229
5230 Uppercase words are terminals, lower case words are non-terminals.
5231 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
5232 bars '|' denote choices. Most grammar productions are implemented in
5233 functions called 'intel_<production>'.
5234
5235 Initial production is 'expr'.
5236
5237 addOp + | -
5238
5239 alpha [a-zA-Z]
5240
5241 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
5242
5243 constant digits [[ radixOverride ]]
5244
5245 dataType BYTE | WORD | DWORD | QWORD | XWORD
5246
5247 digits decdigit
5248 | digits decdigit
5249 | digits hexdigit
5250
5251 decdigit [0-9]
5252
5253 e05 e05 addOp e06
5254 | e06
5255
5256 e06 e06 mulOp e09
5257 | e09
5258
5259 e09 OFFSET e10
5260 | e09 PTR e10
5261 | e09 : e10
5262 | e10
5263
5264 e10 e10 [ expr ]
5265 | e11
5266
5267 e11 ( expr )
5268 | [ expr ]
5269 | constant
5270 | dataType
5271 | id
5272 | $
5273 | register
5274
5275 => expr SHORT e05
5276 | e05
5277
5278 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
5279 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
5280
5281 hexdigit a | b | c | d | e | f
5282 | A | B | C | D | E | F
5283
5284 id alpha
5285 | id alpha
5286 | id decdigit
5287
5288 mulOp * | / | MOD
5289
5290 quote " | '
5291
5292 register specialRegister
5293 | gpRegister
5294 | byteRegister
5295
5296 segmentRegister CS | DS | ES | FS | GS | SS
5297
5298 specialRegister CR0 | CR2 | CR3
5299 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
5300 | TR3 | TR4 | TR5 | TR6 | TR7
5301
5302 We simplify the grammar in obvious places (e.g., register parsing is
5303 done by calling parse_register) and eliminate immediate left recursion
5304 to implement a recursive-descent parser.
5305
5306 expr SHORT e05
5307 | e05
5308
5309 e05 e06 e05'
5310
5311 e05' addOp e06 e05'
5312 | Empty
5313
5314 e06 e09 e06'
5315
5316 e06' mulOp e09 e06'
5317 | Empty
5318
5319 e09 OFFSET e10 e09'
5320 | e10 e09'
5321
5322 e09' PTR e10 e09'
5323 | : e10 e09'
5324 | Empty
5325
5326 e10 e11 e10'
5327
5328 e10' [ expr ] e10'
5329 | Empty
5330
5331 e11 ( expr )
5332 | [ expr ]
5333 | BYTE
5334 | WORD
5335 | DWORD
5336 | QWORD
5337 | XWORD
5338 | .
5339 | $
5340 | register
5341 | id
5342 | constant */
5343
5344 /* Parsing structure for the intel syntax parser. Used to implement the
5345 semantic actions for the operand grammar. */
5346 struct intel_parser_s
5347 {
5348 char *op_string; /* The string being parsed. */
5349 int got_a_float; /* Whether the operand is a float. */
5350 int op_modifier; /* Operand modifier. */
5351 int is_mem; /* 1 if operand is memory reference. */
5352 const reg_entry *reg; /* Last register reference found. */
5353 char *disp; /* Displacement string being built. */
5354 };
5355
5356 static struct intel_parser_s intel_parser;
5357
5358 /* Token structure for parsing intel syntax. */
5359 struct intel_token
5360 {
5361 int code; /* Token code. */
5362 const reg_entry *reg; /* Register entry for register tokens. */
5363 char *str; /* String representation. */
5364 };
5365
5366 static struct intel_token cur_token, prev_token;
5367
5368 /* Token codes for the intel parser. Since T_SHORT is already used
5369 by COFF, undefine it first to prevent a warning. */
5370 #define T_NIL -1
5371 #define T_CONST 1
5372 #define T_REG 2
5373 #define T_BYTE 3
5374 #define T_WORD 4
5375 #define T_DWORD 5
5376 #define T_QWORD 6
5377 #define T_XWORD 7
5378 #undef T_SHORT
5379 #define T_SHORT 8
5380 #define T_OFFSET 9
5381 #define T_PTR 10
5382 #define T_ID 11
5383
5384 /* Prototypes for intel parser functions. */
5385 static int intel_match_token PARAMS ((int code));
5386 static void intel_get_token PARAMS ((void));
5387 static void intel_putback_token PARAMS ((void));
5388 static int intel_expr PARAMS ((void));
5389 static int intel_e05 PARAMS ((void));
5390 static int intel_e05_1 PARAMS ((void));
5391 static int intel_e06 PARAMS ((void));
5392 static int intel_e06_1 PARAMS ((void));
5393 static int intel_e09 PARAMS ((void));
5394 static int intel_e09_1 PARAMS ((void));
5395 static int intel_e10 PARAMS ((void));
5396 static int intel_e10_1 PARAMS ((void));
5397 static int intel_e11 PARAMS ((void));
5398
5399 static int
5400 i386_intel_operand (operand_string, got_a_float)
5401 char *operand_string;
5402 int got_a_float;
5403 {
5404 int ret;
5405 char *p;
5406
5407 /* Initialize token holders. */
5408 cur_token.code = prev_token.code = T_NIL;
5409 cur_token.reg = prev_token.reg = NULL;
5410 cur_token.str = prev_token.str = NULL;
5411
5412 /* Initialize parser structure. */
5413 p = intel_parser.op_string = (char *) malloc (strlen (operand_string) + 1);
5414 if (p == NULL)
5415 abort ();
5416 strcpy (intel_parser.op_string, operand_string);
5417 intel_parser.got_a_float = got_a_float;
5418 intel_parser.op_modifier = -1;
5419 intel_parser.is_mem = 0;
5420 intel_parser.reg = NULL;
5421 intel_parser.disp = (char *) malloc (strlen (operand_string) + 1);
5422 if (intel_parser.disp == NULL)
5423 abort ();
5424 intel_parser.disp[0] = '\0';
5425
5426 /* Read the first token and start the parser. */
5427 intel_get_token ();
5428 ret = intel_expr ();
5429
5430 if (ret)
5431 {
5432 /* If we found a memory reference, hand it over to i386_displacement
5433 to fill in the rest of the operand fields. */
5434 if (intel_parser.is_mem)
5435 {
5436 if ((i.mem_operands == 1
5437 && (current_templates->start->opcode_modifier & IsString) == 0)
5438 || i.mem_operands == 2)
5439 {
5440 as_bad (_("too many memory references for '%s'"),
5441 current_templates->start->name);
5442 ret = 0;
5443 }
5444 else
5445 {
5446 char *s = intel_parser.disp;
5447 i.mem_operands++;
5448
5449 /* Add the displacement expression. */
5450 if (*s != '\0')
5451 ret = i386_displacement (s, s + strlen (s))
5452 && i386_index_check (s);
5453 }
5454 }
5455
5456 /* Constant and OFFSET expressions are handled by i386_immediate. */
5457 else if (intel_parser.op_modifier == OFFSET_FLAT
5458 || intel_parser.reg == NULL)
5459 ret = i386_immediate (intel_parser.disp);
5460 }
5461
5462 free (p);
5463 free (intel_parser.disp);
5464
5465 return ret;
5466 }
5467
5468 /* expr SHORT e05
5469 | e05 */
5470 static int
5471 intel_expr ()
5472 {
5473 /* expr SHORT e05 */
5474 if (cur_token.code == T_SHORT)
5475 {
5476 intel_parser.op_modifier = SHORT;
5477 intel_match_token (T_SHORT);
5478
5479 return (intel_e05 ());
5480 }
5481
5482 /* expr e05 */
5483 else
5484 return intel_e05 ();
5485 }
5486
5487 /* e05 e06 e05'
5488
5489 e05' addOp e06 e05'
5490 | Empty */
5491 static int
5492 intel_e05 ()
5493 {
5494 return (intel_e06 () && intel_e05_1 ());
5495 }
5496
5497 static int
5498 intel_e05_1 ()
5499 {
5500 /* e05' addOp e06 e05' */
5501 if (cur_token.code == '+' || cur_token.code == '-')
5502 {
5503 strcat (intel_parser.disp, cur_token.str);
5504 intel_match_token (cur_token.code);
5505
5506 return (intel_e06 () && intel_e05_1 ());
5507 }
5508
5509 /* e05' Empty */
5510 else
5511 return 1;
5512 }
5513
5514 /* e06 e09 e06'
5515
5516 e06' mulOp e09 e06'
5517 | Empty */
5518 static int
5519 intel_e06 ()
5520 {
5521 return (intel_e09 () && intel_e06_1 ());
5522 }
5523
5524 static int
5525 intel_e06_1 ()
5526 {
5527 /* e06' mulOp e09 e06' */
5528 if (cur_token.code == '*' || cur_token.code == '/')
5529 {
5530 strcat (intel_parser.disp, cur_token.str);
5531 intel_match_token (cur_token.code);
5532
5533 return (intel_e09 () && intel_e06_1 ());
5534 }
5535
5536 /* e06' Empty */
5537 else
5538 return 1;
5539 }
5540
5541 /* e09 OFFSET e10 e09'
5542 | e10 e09'
5543
5544 e09' PTR e10 e09'
5545 | : e10 e09'
5546 | Empty */
5547 static int
5548 intel_e09 ()
5549 {
5550 /* e09 OFFSET e10 e09' */
5551 if (cur_token.code == T_OFFSET)
5552 {
5553 intel_parser.is_mem = 0;
5554 intel_parser.op_modifier = OFFSET_FLAT;
5555 intel_match_token (T_OFFSET);
5556
5557 return (intel_e10 () && intel_e09_1 ());
5558 }
5559
5560 /* e09 e10 e09' */
5561 else
5562 return (intel_e10 () && intel_e09_1 ());
5563 }
5564
5565 static int
5566 intel_e09_1 ()
5567 {
5568 /* e09' PTR e10 e09' */
5569 if (cur_token.code == T_PTR)
5570 {
5571 if (prev_token.code == T_BYTE)
5572 i.suffix = BYTE_MNEM_SUFFIX;
5573
5574 else if (prev_token.code == T_WORD)
5575 {
5576 if (intel_parser.got_a_float == 2) /* "fi..." */
5577 i.suffix = SHORT_MNEM_SUFFIX;
5578 else
5579 i.suffix = WORD_MNEM_SUFFIX;
5580 }
5581
5582 else if (prev_token.code == T_DWORD)
5583 {
5584 if (intel_parser.got_a_float == 1) /* "f..." */
5585 i.suffix = SHORT_MNEM_SUFFIX;
5586 else
5587 i.suffix = LONG_MNEM_SUFFIX;
5588 }
5589
5590 else if (prev_token.code == T_QWORD)
5591 {
5592 if (intel_parser.got_a_float == 1) /* "f..." */
5593 i.suffix = LONG_MNEM_SUFFIX;
5594 else
5595 i.suffix = QWORD_MNEM_SUFFIX;
5596 }
5597
5598 else if (prev_token.code == T_XWORD)
5599 i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
5600
5601 else
5602 {
5603 as_bad (_("Unknown operand modifier `%s'\n"), prev_token.str);
5604 return 0;
5605 }
5606
5607 intel_match_token (T_PTR);
5608
5609 return (intel_e10 () && intel_e09_1 ());
5610 }
5611
5612 /* e09 : e10 e09' */
5613 else if (cur_token.code == ':')
5614 {
5615 /* Mark as a memory operand only if it's not already known to be an
5616 offset expression. */
5617 if (intel_parser.op_modifier != OFFSET_FLAT)
5618 intel_parser.is_mem = 1;
5619
5620 return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
5621 }
5622
5623 /* e09' Empty */
5624 else
5625 return 1;
5626 }
5627
5628 /* e10 e11 e10'
5629
5630 e10' [ expr ] e10'
5631 | Empty */
5632 static int
5633 intel_e10 ()
5634 {
5635 return (intel_e11 () && intel_e10_1 ());
5636 }
5637
5638 static int
5639 intel_e10_1 ()
5640 {
5641 /* e10' [ expr ] e10' */
5642 if (cur_token.code == '[')
5643 {
5644 intel_match_token ('[');
5645
5646 /* Mark as a memory operand only if it's not already known to be an
5647 offset expression. If it's an offset expression, we need to keep
5648 the brace in. */
5649 if (intel_parser.op_modifier != OFFSET_FLAT)
5650 intel_parser.is_mem = 1;
5651 else
5652 strcat (intel_parser.disp, "[");
5653
5654 /* Add a '+' to the displacement string if necessary. */
5655 if (*intel_parser.disp != '\0'
5656 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5657 strcat (intel_parser.disp, "+");
5658
5659 if (intel_expr () && intel_match_token (']'))
5660 {
5661 /* Preserve brackets when the operand is an offset expression. */
5662 if (intel_parser.op_modifier == OFFSET_FLAT)
5663 strcat (intel_parser.disp, "]");
5664
5665 return intel_e10_1 ();
5666 }
5667 else
5668 return 0;
5669 }
5670
5671 /* e10' Empty */
5672 else
5673 return 1;
5674 }
5675
5676 /* e11 ( expr )
5677 | [ expr ]
5678 | BYTE
5679 | WORD
5680 | DWORD
5681 | QWORD
5682 | XWORD
5683 | $
5684 | .
5685 | register
5686 | id
5687 | constant */
5688 static int
5689 intel_e11 ()
5690 {
5691 /* e11 ( expr ) */
5692 if (cur_token.code == '(')
5693 {
5694 intel_match_token ('(');
5695 strcat (intel_parser.disp, "(");
5696
5697 if (intel_expr () && intel_match_token (')'))
5698 {
5699 strcat (intel_parser.disp, ")");
5700 return 1;
5701 }
5702 else
5703 return 0;
5704 }
5705
5706 /* e11 [ expr ] */
5707 else if (cur_token.code == '[')
5708 {
5709 intel_match_token ('[');
5710
5711 /* Mark as a memory operand only if it's not already known to be an
5712 offset expression. If it's an offset expression, we need to keep
5713 the brace in. */
5714 if (intel_parser.op_modifier != OFFSET_FLAT)
5715 intel_parser.is_mem = 1;
5716 else
5717 strcat (intel_parser.disp, "[");
5718
5719 /* Operands for jump/call inside brackets denote absolute addresses. */
5720 if (current_templates->start->opcode_modifier & Jump
5721 || current_templates->start->opcode_modifier & JumpDword
5722 || current_templates->start->opcode_modifier & JumpByte
5723 || current_templates->start->opcode_modifier & JumpInterSegment)
5724 i.types[this_operand] |= JumpAbsolute;
5725
5726 /* Add a '+' to the displacement string if necessary. */
5727 if (*intel_parser.disp != '\0'
5728 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5729 strcat (intel_parser.disp, "+");
5730
5731 if (intel_expr () && intel_match_token (']'))
5732 {
5733 /* Preserve brackets when the operand is an offset expression. */
5734 if (intel_parser.op_modifier == OFFSET_FLAT)
5735 strcat (intel_parser.disp, "]");
5736
5737 return 1;
5738 }
5739 else
5740 return 0;
5741 }
5742
5743 /* e11 BYTE
5744 | WORD
5745 | DWORD
5746 | QWORD
5747 | XWORD */
5748 else if (cur_token.code == T_BYTE
5749 || cur_token.code == T_WORD
5750 || cur_token.code == T_DWORD
5751 || cur_token.code == T_QWORD
5752 || cur_token.code == T_XWORD)
5753 {
5754 intel_match_token (cur_token.code);
5755
5756 return 1;
5757 }
5758
5759 /* e11 $
5760 | . */
5761 else if (cur_token.code == '$' || cur_token.code == '.')
5762 {
5763 strcat (intel_parser.disp, cur_token.str);
5764 intel_match_token (cur_token.code);
5765
5766 /* Mark as a memory operand only if it's not already known to be an
5767 offset expression. */
5768 if (intel_parser.op_modifier != OFFSET_FLAT)
5769 intel_parser.is_mem = 1;
5770
5771 return 1;
5772 }
5773
5774 /* e11 register */
5775 else if (cur_token.code == T_REG)
5776 {
5777 const reg_entry *reg = intel_parser.reg = cur_token.reg;
5778
5779 intel_match_token (T_REG);
5780
5781 /* Check for segment change. */
5782 if (cur_token.code == ':')
5783 {
5784 if (reg->reg_type & (SReg2 | SReg3))
5785 {
5786 switch (reg->reg_num)
5787 {
5788 case 0:
5789 i.seg[i.mem_operands] = &es;
5790 break;
5791 case 1:
5792 i.seg[i.mem_operands] = &cs;
5793 break;
5794 case 2:
5795 i.seg[i.mem_operands] = &ss;
5796 break;
5797 case 3:
5798 i.seg[i.mem_operands] = &ds;
5799 break;
5800 case 4:
5801 i.seg[i.mem_operands] = &fs;
5802 break;
5803 case 5:
5804 i.seg[i.mem_operands] = &gs;
5805 break;
5806 }
5807 }
5808 else
5809 {
5810 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
5811 return 0;
5812 }
5813 }
5814
5815 /* Not a segment register. Check for register scaling. */
5816 else if (cur_token.code == '*')
5817 {
5818 if (!intel_parser.is_mem)
5819 {
5820 as_bad (_("Register scaling only allowed in memory operands."));
5821 return 0;
5822 }
5823
5824 /* What follows must be a valid scale. */
5825 if (intel_match_token ('*')
5826 && strchr ("01248", *cur_token.str))
5827 {
5828 i.index_reg = reg;
5829 i.types[this_operand] |= BaseIndex;
5830
5831 /* Set the scale after setting the register (otherwise,
5832 i386_scale will complain) */
5833 i386_scale (cur_token.str);
5834 intel_match_token (T_CONST);
5835 }
5836 else
5837 {
5838 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
5839 cur_token.str);
5840 return 0;
5841 }
5842 }
5843
5844 /* No scaling. If this is a memory operand, the register is either a
5845 base register (first occurrence) or an index register (second
5846 occurrence). */
5847 else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3)))
5848 {
5849 if (i.base_reg && i.index_reg)
5850 {
5851 as_bad (_("Too many register references in memory operand.\n"));
5852 return 0;
5853 }
5854
5855 if (i.base_reg == NULL)
5856 i.base_reg = reg;
5857 else
5858 i.index_reg = reg;
5859
5860 i.types[this_operand] |= BaseIndex;
5861 }
5862
5863 /* Offset modifier. Add the register to the displacement string to be
5864 parsed as an immediate expression after we're done. */
5865 else if (intel_parser.op_modifier == OFFSET_FLAT)
5866 strcat (intel_parser.disp, reg->reg_name);
5867
5868 /* It's neither base nor index nor offset. */
5869 else
5870 {
5871 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
5872 i.op[this_operand].regs = reg;
5873 i.reg_operands++;
5874 }
5875
5876 /* Since registers are not part of the displacement string (except
5877 when we're parsing offset operands), we may need to remove any
5878 preceding '+' from the displacement string. */
5879 if (*intel_parser.disp != '\0'
5880 && intel_parser.op_modifier != OFFSET_FLAT)
5881 {
5882 char *s = intel_parser.disp;
5883 s += strlen (s) - 1;
5884 if (*s == '+')
5885 *s = '\0';
5886 }
5887
5888 return 1;
5889 }
5890
5891 /* e11 id */
5892 else if (cur_token.code == T_ID)
5893 {
5894 /* Add the identifier to the displacement string. */
5895 strcat (intel_parser.disp, cur_token.str);
5896 intel_match_token (T_ID);
5897
5898 /* The identifier represents a memory reference only if it's not
5899 preceded by an offset modifier. */
5900 if (intel_parser.op_modifier != OFFSET_FLAT)
5901 intel_parser.is_mem = 1;
5902
5903 return 1;
5904 }
5905
5906 /* e11 constant */
5907 else if (cur_token.code == T_CONST
5908 || cur_token.code == '-'
5909 || cur_token.code == '+')
5910 {
5911 char *save_str;
5912
5913 /* Allow constants that start with `+' or `-'. */
5914 if (cur_token.code == '-' || cur_token.code == '+')
5915 {
5916 strcat (intel_parser.disp, cur_token.str);
5917 intel_match_token (cur_token.code);
5918 if (cur_token.code != T_CONST)
5919 {
5920 as_bad (_("Syntax error. Expecting a constant. Got `%s'.\n"),
5921 cur_token.str);
5922 return 0;
5923 }
5924 }
5925
5926 save_str = (char *) malloc (strlen (cur_token.str) + 1);
5927 if (save_str == NULL)
5928 abort ();
5929 strcpy (save_str, cur_token.str);
5930
5931 /* Get the next token to check for register scaling. */
5932 intel_match_token (cur_token.code);
5933
5934 /* Check if this constant is a scaling factor for an index register. */
5935 if (cur_token.code == '*')
5936 {
5937 if (intel_match_token ('*') && cur_token.code == T_REG)
5938 {
5939 if (!intel_parser.is_mem)
5940 {
5941 as_bad (_("Register scaling only allowed in memory operands."));
5942 return 0;
5943 }
5944
5945 /* The constant is followed by `* reg', so it must be
5946 a valid scale. */
5947 if (strchr ("01248", *save_str))
5948 {
5949 i.index_reg = cur_token.reg;
5950 i.types[this_operand] |= BaseIndex;
5951
5952 /* Set the scale after setting the register (otherwise,
5953 i386_scale will complain) */
5954 i386_scale (save_str);
5955 intel_match_token (T_REG);
5956
5957 /* Since registers are not part of the displacement
5958 string, we may need to remove any preceding '+' from
5959 the displacement string. */
5960 if (*intel_parser.disp != '\0')
5961 {
5962 char *s = intel_parser.disp;
5963 s += strlen (s) - 1;
5964 if (*s == '+')
5965 *s = '\0';
5966 }
5967
5968 free (save_str);
5969
5970 return 1;
5971 }
5972 else
5973 return 0;
5974 }
5975
5976 /* The constant was not used for register scaling. Since we have
5977 already consumed the token following `*' we now need to put it
5978 back in the stream. */
5979 else
5980 intel_putback_token ();
5981 }
5982
5983 /* Add the constant to the displacement string. */
5984 strcat (intel_parser.disp, save_str);
5985 free (save_str);
5986
5987 return 1;
5988 }
5989
5990 as_bad (_("Unrecognized token '%s'"), cur_token.str);
5991 return 0;
5992 }
5993
5994 /* Match the given token against cur_token. If they match, read the next
5995 token from the operand string. */
5996 static int
5997 intel_match_token (code)
5998 int code;
5999 {
6000 if (cur_token.code == code)
6001 {
6002 intel_get_token ();
6003 return 1;
6004 }
6005 else
6006 {
6007 as_bad (_("Unexpected token `%s'\n"), cur_token.str);
6008 return 0;
6009 }
6010 }
6011
6012 /* Read a new token from intel_parser.op_string and store it in cur_token. */
6013 static void
6014 intel_get_token ()
6015 {
6016 char *end_op;
6017 const reg_entry *reg;
6018 struct intel_token new_token;
6019
6020 new_token.code = T_NIL;
6021 new_token.reg = NULL;
6022 new_token.str = NULL;
6023
6024 /* Free the memory allocated to the previous token and move
6025 cur_token to prev_token. */
6026 if (prev_token.str)
6027 free (prev_token.str);
6028
6029 prev_token = cur_token;
6030
6031 /* Skip whitespace. */
6032 while (is_space_char (*intel_parser.op_string))
6033 intel_parser.op_string++;
6034
6035 /* Return an empty token if we find nothing else on the line. */
6036 if (*intel_parser.op_string == '\0')
6037 {
6038 cur_token = new_token;
6039 return;
6040 }
6041
6042 /* The new token cannot be larger than the remainder of the operand
6043 string. */
6044 new_token.str = (char *) malloc (strlen (intel_parser.op_string) + 1);
6045 if (new_token.str == NULL)
6046 abort ();
6047 new_token.str[0] = '\0';
6048
6049 if (strchr ("0123456789", *intel_parser.op_string))
6050 {
6051 char *p = new_token.str;
6052 char *q = intel_parser.op_string;
6053 new_token.code = T_CONST;
6054
6055 /* Allow any kind of identifier char to encompass floating point and
6056 hexadecimal numbers. */
6057 while (is_identifier_char (*q))
6058 *p++ = *q++;
6059 *p = '\0';
6060
6061 /* Recognize special symbol names [0-9][bf]. */
6062 if (strlen (intel_parser.op_string) == 2
6063 && (intel_parser.op_string[1] == 'b'
6064 || intel_parser.op_string[1] == 'f'))
6065 new_token.code = T_ID;
6066 }
6067
6068 else if (strchr ("+-/*:[]()", *intel_parser.op_string))
6069 {
6070 new_token.code = *intel_parser.op_string;
6071 new_token.str[0] = *intel_parser.op_string;
6072 new_token.str[1] = '\0';
6073 }
6074
6075 else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
6076 && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
6077 {
6078 new_token.code = T_REG;
6079 new_token.reg = reg;
6080
6081 if (*intel_parser.op_string == REGISTER_PREFIX)
6082 {
6083 new_token.str[0] = REGISTER_PREFIX;
6084 new_token.str[1] = '\0';
6085 }
6086
6087 strcat (new_token.str, reg->reg_name);
6088 }
6089
6090 else if (is_identifier_char (*intel_parser.op_string))
6091 {
6092 char *p = new_token.str;
6093 char *q = intel_parser.op_string;
6094
6095 /* A '.' or '$' followed by an identifier char is an identifier.
6096 Otherwise, it's operator '.' followed by an expression. */
6097 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
6098 {
6099 new_token.code = *q;
6100 new_token.str[0] = *q;
6101 new_token.str[1] = '\0';
6102 }
6103 else
6104 {
6105 while (is_identifier_char (*q) || *q == '@')
6106 *p++ = *q++;
6107 *p = '\0';
6108
6109 if (strcasecmp (new_token.str, "BYTE") == 0)
6110 new_token.code = T_BYTE;
6111
6112 else if (strcasecmp (new_token.str, "WORD") == 0)
6113 new_token.code = T_WORD;
6114
6115 else if (strcasecmp (new_token.str, "DWORD") == 0)
6116 new_token.code = T_DWORD;
6117
6118 else if (strcasecmp (new_token.str, "QWORD") == 0)
6119 new_token.code = T_QWORD;
6120
6121 else if (strcasecmp (new_token.str, "XWORD") == 0)
6122 new_token.code = T_XWORD;
6123
6124 else if (strcasecmp (new_token.str, "PTR") == 0)
6125 new_token.code = T_PTR;
6126
6127 else if (strcasecmp (new_token.str, "SHORT") == 0)
6128 new_token.code = T_SHORT;
6129
6130 else if (strcasecmp (new_token.str, "OFFSET") == 0)
6131 {
6132 new_token.code = T_OFFSET;
6133
6134 /* ??? This is not mentioned in the MASM grammar but gcc
6135 makes use of it with -mintel-syntax. OFFSET may be
6136 followed by FLAT: */
6137 if (strncasecmp (q, " FLAT:", 6) == 0)
6138 strcat (new_token.str, " FLAT:");
6139 }
6140
6141 /* ??? This is not mentioned in the MASM grammar. */
6142 else if (strcasecmp (new_token.str, "FLAT") == 0)
6143 new_token.code = T_OFFSET;
6144
6145 else
6146 new_token.code = T_ID;
6147 }
6148 }
6149
6150 else
6151 as_bad (_("Unrecognized token `%s'\n"), intel_parser.op_string);
6152
6153 intel_parser.op_string += strlen (new_token.str);
6154 cur_token = new_token;
6155 }
6156
6157 /* Put cur_token back into the token stream and make cur_token point to
6158 prev_token. */
6159 static void
6160 intel_putback_token ()
6161 {
6162 intel_parser.op_string -= strlen (cur_token.str);
6163 free (cur_token.str);
6164 cur_token = prev_token;
6165
6166 /* Forget prev_token. */
6167 prev_token.code = T_NIL;
6168 prev_token.reg = NULL;
6169 prev_token.str = NULL;
6170 }