92540ad625d15af39c895fe884f83b3053cbf329
[deliverable/binutils-gdb.git] / gas / config / tc-i386.c
1 /* i386.c -- Assemble code for the Intel 80386
2 Copyright 1989, 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
3 2000, 2001, 2002, 2003
4 Free Software Foundation, Inc.
5
6 This file is part of GAS, the GNU Assembler.
7
8 GAS is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2, or (at your option)
11 any later version.
12
13 GAS is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
17
18 You should have received a copy of the GNU General Public License
19 along with GAS; see the file COPYING. If not, write to the Free
20 Software Foundation, 59 Temple Place - Suite 330, Boston, MA
21 02111-1307, USA. */
22
23 /* Intel 80386 machine specific gas.
24 Written by Eliot Dresselhaus (eliot@mgm.mit.edu).
25 x86_64 support by Jan Hubicka (jh@suse.cz)
26 Bugs & suggestions are completely welcome. This is free software.
27 Please help us make it better. */
28
29 #include "as.h"
30 #include "safe-ctype.h"
31 #include "subsegs.h"
32 #include "dwarf2dbg.h"
33 #include "opcode/i386.h"
34
35 #ifndef REGISTER_WARNINGS
36 #define REGISTER_WARNINGS 1
37 #endif
38
39 #ifndef INFER_ADDR_PREFIX
40 #define INFER_ADDR_PREFIX 1
41 #endif
42
43 #ifndef SCALE1_WHEN_NO_INDEX
44 /* Specifying a scale factor besides 1 when there is no index is
45 futile. eg. `mov (%ebx,2),%al' does exactly the same as
46 `mov (%ebx),%al'. To slavishly follow what the programmer
47 specified, set SCALE1_WHEN_NO_INDEX to 0. */
48 #define SCALE1_WHEN_NO_INDEX 1
49 #endif
50
51 #ifdef BFD_ASSEMBLER
52 #define RELOC_ENUM enum bfd_reloc_code_real
53 #else
54 #define RELOC_ENUM int
55 #endif
56
57 #ifndef DEFAULT_ARCH
58 #define DEFAULT_ARCH "i386"
59 #endif
60
61 #ifndef INLINE
62 #if __GNUC__ >= 2
63 #define INLINE __inline__
64 #else
65 #define INLINE
66 #endif
67 #endif
68
69 static INLINE unsigned int mode_from_disp_size PARAMS ((unsigned int));
70 static INLINE int fits_in_signed_byte PARAMS ((offsetT));
71 static INLINE int fits_in_unsigned_byte PARAMS ((offsetT));
72 static INLINE int fits_in_unsigned_word PARAMS ((offsetT));
73 static INLINE int fits_in_signed_word PARAMS ((offsetT));
74 static INLINE int fits_in_unsigned_long PARAMS ((offsetT));
75 static INLINE int fits_in_signed_long PARAMS ((offsetT));
76 static int smallest_imm_type PARAMS ((offsetT));
77 static offsetT offset_in_range PARAMS ((offsetT, int));
78 static int add_prefix PARAMS ((unsigned int));
79 static void set_code_flag PARAMS ((int));
80 static void set_16bit_gcc_code_flag PARAMS ((int));
81 static void set_intel_syntax PARAMS ((int));
82 static void set_cpu_arch PARAMS ((int));
83 static char *output_invalid PARAMS ((int c));
84 static int i386_operand PARAMS ((char *operand_string));
85 static int i386_intel_operand PARAMS ((char *operand_string, int got_a_float));
86 static const reg_entry *parse_register PARAMS ((char *reg_string,
87 char **end_op));
88 static char *parse_insn PARAMS ((char *, char *));
89 static char *parse_operands PARAMS ((char *, const char *));
90 static void swap_operands PARAMS ((void));
91 static void optimize_imm PARAMS ((void));
92 static void optimize_disp PARAMS ((void));
93 static int match_template PARAMS ((void));
94 static int check_string PARAMS ((void));
95 static int process_suffix PARAMS ((void));
96 static int check_byte_reg PARAMS ((void));
97 static int check_long_reg PARAMS ((void));
98 static int check_qword_reg PARAMS ((void));
99 static int check_word_reg PARAMS ((void));
100 static int finalize_imm PARAMS ((void));
101 static int process_operands PARAMS ((void));
102 static const seg_entry *build_modrm_byte PARAMS ((void));
103 static void output_insn PARAMS ((void));
104 static void output_branch PARAMS ((void));
105 static void output_jump PARAMS ((void));
106 static void output_interseg_jump PARAMS ((void));
107 static void output_imm PARAMS ((fragS *insn_start_frag,
108 offsetT insn_start_off));
109 static void output_disp PARAMS ((fragS *insn_start_frag,
110 offsetT insn_start_off));
111 #ifndef I386COFF
112 static void s_bss PARAMS ((int));
113 #endif
114
115 static const char *default_arch = DEFAULT_ARCH;
116
117 /* 'md_assemble ()' gathers together information and puts it into a
118 i386_insn. */
119
120 union i386_op
121 {
122 expressionS *disps;
123 expressionS *imms;
124 const reg_entry *regs;
125 };
126
127 struct _i386_insn
128 {
129 /* TM holds the template for the insn were currently assembling. */
130 template tm;
131
132 /* SUFFIX holds the instruction mnemonic suffix if given.
133 (e.g. 'l' for 'movl') */
134 char suffix;
135
136 /* OPERANDS gives the number of given operands. */
137 unsigned int operands;
138
139 /* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number
140 of given register, displacement, memory operands and immediate
141 operands. */
142 unsigned int reg_operands, disp_operands, mem_operands, imm_operands;
143
144 /* TYPES [i] is the type (see above #defines) which tells us how to
145 use OP[i] for the corresponding operand. */
146 unsigned int types[MAX_OPERANDS];
147
148 /* Displacement expression, immediate expression, or register for each
149 operand. */
150 union i386_op op[MAX_OPERANDS];
151
152 /* Flags for operands. */
153 unsigned int flags[MAX_OPERANDS];
154 #define Operand_PCrel 1
155
156 /* Relocation type for operand */
157 RELOC_ENUM reloc[MAX_OPERANDS];
158
159 /* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode
160 the base index byte below. */
161 const reg_entry *base_reg;
162 const reg_entry *index_reg;
163 unsigned int log2_scale_factor;
164
165 /* SEG gives the seg_entries of this insn. They are zero unless
166 explicit segment overrides are given. */
167 const seg_entry *seg[2];
168
169 /* PREFIX holds all the given prefix opcodes (usually null).
170 PREFIXES is the number of prefix opcodes. */
171 unsigned int prefixes;
172 unsigned char prefix[MAX_PREFIXES];
173
174 /* RM and SIB are the modrm byte and the sib byte where the
175 addressing modes of this insn are encoded. */
176
177 modrm_byte rm;
178 rex_byte rex;
179 sib_byte sib;
180 };
181
182 typedef struct _i386_insn i386_insn;
183
184 /* List of chars besides those in app.c:symbol_chars that can start an
185 operand. Used to prevent the scrubber eating vital white-space. */
186 #ifdef LEX_AT
187 const char extra_symbol_chars[] = "*%-(@[";
188 #else
189 const char extra_symbol_chars[] = "*%-([";
190 #endif
191
192 #if (defined (TE_I386AIX) \
193 || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) \
194 && !defined (TE_LINUX) \
195 && !defined (TE_FreeBSD) \
196 && !defined (TE_NetBSD)))
197 /* This array holds the chars that always start a comment. If the
198 pre-processor is disabled, these aren't very useful. */
199 const char comment_chars[] = "#/";
200 #define PREFIX_SEPARATOR '\\'
201
202 /* This array holds the chars that only start a comment at the beginning of
203 a line. If the line seems to have the form '# 123 filename'
204 .line and .file directives will appear in the pre-processed output.
205 Note that input_file.c hand checks for '#' at the beginning of the
206 first line of the input file. This is because the compiler outputs
207 #NO_APP at the beginning of its output.
208 Also note that comments started like this one will always work if
209 '/' isn't otherwise defined. */
210 const char line_comment_chars[] = "#";
211
212 #else
213 /* Putting '/' here makes it impossible to use the divide operator.
214 However, we need it for compatibility with SVR4 systems. */
215 const char comment_chars[] = "#";
216 #define PREFIX_SEPARATOR '/'
217
218 const char line_comment_chars[] = "/#";
219 #endif
220
221 const char line_separator_chars[] = ";";
222
223 /* Chars that can be used to separate mant from exp in floating point
224 nums. */
225 const char EXP_CHARS[] = "eE";
226
227 /* Chars that mean this number is a floating point constant
228 As in 0f12.456
229 or 0d1.2345e12. */
230 const char FLT_CHARS[] = "fFdDxX";
231
232 /* Tables for lexical analysis. */
233 static char mnemonic_chars[256];
234 static char register_chars[256];
235 static char operand_chars[256];
236 static char identifier_chars[256];
237 static char digit_chars[256];
238
239 /* Lexical macros. */
240 #define is_mnemonic_char(x) (mnemonic_chars[(unsigned char) x])
241 #define is_operand_char(x) (operand_chars[(unsigned char) x])
242 #define is_register_char(x) (register_chars[(unsigned char) x])
243 #define is_space_char(x) ((x) == ' ')
244 #define is_identifier_char(x) (identifier_chars[(unsigned char) x])
245 #define is_digit_char(x) (digit_chars[(unsigned char) x])
246
247 /* All non-digit non-letter charcters that may occur in an operand. */
248 static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
249
250 /* md_assemble() always leaves the strings it's passed unaltered. To
251 effect this we maintain a stack of saved characters that we've smashed
252 with '\0's (indicating end of strings for various sub-fields of the
253 assembler instruction). */
254 static char save_stack[32];
255 static char *save_stack_p;
256 #define END_STRING_AND_SAVE(s) \
257 do { *save_stack_p++ = *(s); *(s) = '\0'; } while (0)
258 #define RESTORE_END_STRING(s) \
259 do { *(s) = *--save_stack_p; } while (0)
260
261 /* The instruction we're assembling. */
262 static i386_insn i;
263
264 /* Possible templates for current insn. */
265 static const templates *current_templates;
266
267 /* Per instruction expressionS buffers: 2 displacements & 2 immediate max. */
268 static expressionS disp_expressions[2], im_expressions[2];
269
270 /* Current operand we are working on. */
271 static int this_operand;
272
273 /* We support four different modes. FLAG_CODE variable is used to distinguish
274 these. */
275
276 enum flag_code {
277 CODE_32BIT,
278 CODE_16BIT,
279 CODE_64BIT };
280 #define NUM_FLAG_CODE ((int) CODE_64BIT + 1)
281
282 static enum flag_code flag_code;
283 static int use_rela_relocations = 0;
284
285 /* The names used to print error messages. */
286 static const char *flag_code_names[] =
287 {
288 "32",
289 "16",
290 "64"
291 };
292
293 /* 1 for intel syntax,
294 0 if att syntax. */
295 static int intel_syntax = 0;
296
297 /* 1 if register prefix % not required. */
298 static int allow_naked_reg = 0;
299
300 /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
301 leave, push, and pop instructions so that gcc has the same stack
302 frame as in 32 bit mode. */
303 static char stackop_size = '\0';
304
305 /* Non-zero to quieten some warnings. */
306 static int quiet_warnings = 0;
307
308 /* CPU name. */
309 static const char *cpu_arch_name = NULL;
310
311 /* CPU feature flags. */
312 static unsigned int cpu_arch_flags = CpuUnknownFlags | CpuNo64;
313
314 /* If set, conditional jumps are not automatically promoted to handle
315 larger than a byte offset. */
316 static unsigned int no_cond_jump_promotion = 0;
317
318 /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
319 symbolS *GOT_symbol;
320
321 /* Interface to relax_segment.
322 There are 3 major relax states for 386 jump insns because the
323 different types of jumps add different sizes to frags when we're
324 figuring out what sort of jump to choose to reach a given label. */
325
326 /* Types. */
327 #define UNCOND_JUMP 0
328 #define COND_JUMP 1
329 #define COND_JUMP86 2
330
331 /* Sizes. */
332 #define CODE16 1
333 #define SMALL 0
334 #define SMALL16 (SMALL | CODE16)
335 #define BIG 2
336 #define BIG16 (BIG | CODE16)
337
338 #ifndef INLINE
339 #ifdef __GNUC__
340 #define INLINE __inline__
341 #else
342 #define INLINE
343 #endif
344 #endif
345
346 #define ENCODE_RELAX_STATE(type, size) \
347 ((relax_substateT) (((type) << 2) | (size)))
348 #define TYPE_FROM_RELAX_STATE(s) \
349 ((s) >> 2)
350 #define DISP_SIZE_FROM_RELAX_STATE(s) \
351 ((((s) & 3) == BIG ? 4 : (((s) & 3) == BIG16 ? 2 : 1)))
352
353 /* This table is used by relax_frag to promote short jumps to long
354 ones where necessary. SMALL (short) jumps may be promoted to BIG
355 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
356 don't allow a short jump in a 32 bit code segment to be promoted to
357 a 16 bit offset jump because it's slower (requires data size
358 prefix), and doesn't work, unless the destination is in the bottom
359 64k of the code segment (The top 16 bits of eip are zeroed). */
360
361 const relax_typeS md_relax_table[] =
362 {
363 /* The fields are:
364 1) most positive reach of this state,
365 2) most negative reach of this state,
366 3) how many bytes this mode will have in the variable part of the frag
367 4) which index into the table to try if we can't fit into this one. */
368
369 /* UNCOND_JUMP states. */
370 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
371 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
372 /* dword jmp adds 4 bytes to frag:
373 0 extra opcode bytes, 4 displacement bytes. */
374 {0, 0, 4, 0},
375 /* word jmp adds 2 byte2 to frag:
376 0 extra opcode bytes, 2 displacement bytes. */
377 {0, 0, 2, 0},
378
379 /* COND_JUMP states. */
380 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
381 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
382 /* dword conditionals adds 5 bytes to frag:
383 1 extra opcode byte, 4 displacement bytes. */
384 {0, 0, 5, 0},
385 /* word conditionals add 3 bytes to frag:
386 1 extra opcode byte, 2 displacement bytes. */
387 {0, 0, 3, 0},
388
389 /* COND_JUMP86 states. */
390 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG)},
391 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG16)},
392 /* dword conditionals adds 5 bytes to frag:
393 1 extra opcode byte, 4 displacement bytes. */
394 {0, 0, 5, 0},
395 /* word conditionals add 4 bytes to frag:
396 1 displacement byte and a 3 byte long branch insn. */
397 {0, 0, 4, 0}
398 };
399
400 static const arch_entry cpu_arch[] = {
401 {"i8086", Cpu086 },
402 {"i186", Cpu086|Cpu186 },
403 {"i286", Cpu086|Cpu186|Cpu286 },
404 {"i386", Cpu086|Cpu186|Cpu286|Cpu386 },
405 {"i486", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486 },
406 {"i586", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
407 {"i686", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
408 {"pentium", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
409 {"pentiumpro",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
410 {"pentium4", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX|CpuSSE|CpuSSE2 },
411 {"k6", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow },
412 {"athlon", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuMMX|Cpu3dnow },
413 {"sledgehammer",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuSledgehammer|CpuMMX|Cpu3dnow|CpuSSE|CpuSSE2 },
414 {NULL, 0 }
415 };
416
417 const pseudo_typeS md_pseudo_table[] =
418 {
419 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
420 {"align", s_align_bytes, 0},
421 #else
422 {"align", s_align_ptwo, 0},
423 #endif
424 {"arch", set_cpu_arch, 0},
425 #ifndef I386COFF
426 {"bss", s_bss, 0},
427 #endif
428 {"ffloat", float_cons, 'f'},
429 {"dfloat", float_cons, 'd'},
430 {"tfloat", float_cons, 'x'},
431 {"value", cons, 2},
432 {"noopt", s_ignore, 0},
433 {"optim", s_ignore, 0},
434 {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
435 {"code16", set_code_flag, CODE_16BIT},
436 {"code32", set_code_flag, CODE_32BIT},
437 {"code64", set_code_flag, CODE_64BIT},
438 {"intel_syntax", set_intel_syntax, 1},
439 {"att_syntax", set_intel_syntax, 0},
440 {"file", (void (*) PARAMS ((int))) dwarf2_directive_file, 0},
441 {"loc", dwarf2_directive_loc, 0},
442 {0, 0, 0}
443 };
444
445 /* For interface with expression (). */
446 extern char *input_line_pointer;
447
448 /* Hash table for instruction mnemonic lookup. */
449 static struct hash_control *op_hash;
450
451 /* Hash table for register lookup. */
452 static struct hash_control *reg_hash;
453 \f
454 void
455 i386_align_code (fragP, count)
456 fragS *fragP;
457 int count;
458 {
459 /* Various efficient no-op patterns for aligning code labels.
460 Note: Don't try to assemble the instructions in the comments.
461 0L and 0w are not legal. */
462 static const char f32_1[] =
463 {0x90}; /* nop */
464 static const char f32_2[] =
465 {0x89,0xf6}; /* movl %esi,%esi */
466 static const char f32_3[] =
467 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
468 static const char f32_4[] =
469 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
470 static const char f32_5[] =
471 {0x90, /* nop */
472 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
473 static const char f32_6[] =
474 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
475 static const char f32_7[] =
476 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
477 static const char f32_8[] =
478 {0x90, /* nop */
479 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
480 static const char f32_9[] =
481 {0x89,0xf6, /* movl %esi,%esi */
482 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
483 static const char f32_10[] =
484 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
485 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
486 static const char f32_11[] =
487 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
488 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
489 static const char f32_12[] =
490 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
491 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
492 static const char f32_13[] =
493 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
494 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
495 static const char f32_14[] =
496 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
497 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
498 static const char f32_15[] =
499 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
500 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
501 static const char f16_3[] =
502 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
503 static const char f16_4[] =
504 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
505 static const char f16_5[] =
506 {0x90, /* nop */
507 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
508 static const char f16_6[] =
509 {0x89,0xf6, /* mov %si,%si */
510 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
511 static const char f16_7[] =
512 {0x8d,0x74,0x00, /* lea 0(%si),%si */
513 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
514 static const char f16_8[] =
515 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
516 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
517 static const char *const f32_patt[] = {
518 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
519 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
520 };
521 static const char *const f16_patt[] = {
522 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
523 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
524 };
525
526 if (count <= 0 || count > 15)
527 return;
528
529 /* The recommended way to pad 64bit code is to use NOPs preceded by
530 maximally four 0x66 prefixes. Balance the size of nops. */
531 if (flag_code == CODE_64BIT)
532 {
533 int i;
534 int nnops = (count + 3) / 4;
535 int len = count / nnops;
536 int remains = count - nnops * len;
537 int pos = 0;
538
539 for (i = 0; i < remains; i++)
540 {
541 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len);
542 fragP->fr_literal[fragP->fr_fix + pos + len] = 0x90;
543 pos += len + 1;
544 }
545 for (; i < nnops; i++)
546 {
547 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len - 1);
548 fragP->fr_literal[fragP->fr_fix + pos + len - 1] = 0x90;
549 pos += len;
550 }
551 }
552 else
553 if (flag_code == CODE_16BIT)
554 {
555 memcpy (fragP->fr_literal + fragP->fr_fix,
556 f16_patt[count - 1], count);
557 if (count > 8)
558 /* Adjust jump offset. */
559 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
560 }
561 else
562 memcpy (fragP->fr_literal + fragP->fr_fix,
563 f32_patt[count - 1], count);
564 fragP->fr_var = count;
565 }
566
567 static INLINE unsigned int
568 mode_from_disp_size (t)
569 unsigned int t;
570 {
571 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
572 }
573
574 static INLINE int
575 fits_in_signed_byte (num)
576 offsetT num;
577 {
578 return (num >= -128) && (num <= 127);
579 }
580
581 static INLINE int
582 fits_in_unsigned_byte (num)
583 offsetT num;
584 {
585 return (num & 0xff) == num;
586 }
587
588 static INLINE int
589 fits_in_unsigned_word (num)
590 offsetT num;
591 {
592 return (num & 0xffff) == num;
593 }
594
595 static INLINE int
596 fits_in_signed_word (num)
597 offsetT num;
598 {
599 return (-32768 <= num) && (num <= 32767);
600 }
601 static INLINE int
602 fits_in_signed_long (num)
603 offsetT num ATTRIBUTE_UNUSED;
604 {
605 #ifndef BFD64
606 return 1;
607 #else
608 return (!(((offsetT) -1 << 31) & num)
609 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
610 #endif
611 } /* fits_in_signed_long() */
612 static INLINE int
613 fits_in_unsigned_long (num)
614 offsetT num ATTRIBUTE_UNUSED;
615 {
616 #ifndef BFD64
617 return 1;
618 #else
619 return (num & (((offsetT) 2 << 31) - 1)) == num;
620 #endif
621 } /* fits_in_unsigned_long() */
622
623 static int
624 smallest_imm_type (num)
625 offsetT num;
626 {
627 if (cpu_arch_flags != (Cpu086 | Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64))
628 {
629 /* This code is disabled on the 486 because all the Imm1 forms
630 in the opcode table are slower on the i486. They're the
631 versions with the implicitly specified single-position
632 displacement, which has another syntax if you really want to
633 use that form. */
634 if (num == 1)
635 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
636 }
637 return (fits_in_signed_byte (num)
638 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
639 : fits_in_unsigned_byte (num)
640 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
641 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
642 ? (Imm16 | Imm32 | Imm32S | Imm64)
643 : fits_in_signed_long (num)
644 ? (Imm32 | Imm32S | Imm64)
645 : fits_in_unsigned_long (num)
646 ? (Imm32 | Imm64)
647 : Imm64);
648 }
649
650 static offsetT
651 offset_in_range (val, size)
652 offsetT val;
653 int size;
654 {
655 addressT mask;
656
657 switch (size)
658 {
659 case 1: mask = ((addressT) 1 << 8) - 1; break;
660 case 2: mask = ((addressT) 1 << 16) - 1; break;
661 case 4: mask = ((addressT) 2 << 31) - 1; break;
662 #ifdef BFD64
663 case 8: mask = ((addressT) 2 << 63) - 1; break;
664 #endif
665 default: abort ();
666 }
667
668 /* If BFD64, sign extend val. */
669 if (!use_rela_relocations)
670 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
671 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
672
673 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
674 {
675 char buf1[40], buf2[40];
676
677 sprint_value (buf1, val);
678 sprint_value (buf2, val & mask);
679 as_warn (_("%s shortened to %s"), buf1, buf2);
680 }
681 return val & mask;
682 }
683
684 /* Returns 0 if attempting to add a prefix where one from the same
685 class already exists, 1 if non rep/repne added, 2 if rep/repne
686 added. */
687 static int
688 add_prefix (prefix)
689 unsigned int prefix;
690 {
691 int ret = 1;
692 int q;
693
694 if (prefix >= REX_OPCODE && prefix < REX_OPCODE + 16
695 && flag_code == CODE_64BIT)
696 q = REX_PREFIX;
697 else
698 switch (prefix)
699 {
700 default:
701 abort ();
702
703 case CS_PREFIX_OPCODE:
704 case DS_PREFIX_OPCODE:
705 case ES_PREFIX_OPCODE:
706 case FS_PREFIX_OPCODE:
707 case GS_PREFIX_OPCODE:
708 case SS_PREFIX_OPCODE:
709 q = SEG_PREFIX;
710 break;
711
712 case REPNE_PREFIX_OPCODE:
713 case REPE_PREFIX_OPCODE:
714 ret = 2;
715 /* fall thru */
716 case LOCK_PREFIX_OPCODE:
717 q = LOCKREP_PREFIX;
718 break;
719
720 case FWAIT_OPCODE:
721 q = WAIT_PREFIX;
722 break;
723
724 case ADDR_PREFIX_OPCODE:
725 q = ADDR_PREFIX;
726 break;
727
728 case DATA_PREFIX_OPCODE:
729 q = DATA_PREFIX;
730 break;
731 }
732
733 if (i.prefix[q] != 0)
734 {
735 as_bad (_("same type of prefix used twice"));
736 return 0;
737 }
738
739 i.prefixes += 1;
740 i.prefix[q] = prefix;
741 return ret;
742 }
743
744 static void
745 set_code_flag (value)
746 int value;
747 {
748 flag_code = value;
749 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
750 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
751 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
752 {
753 as_bad (_("64bit mode not supported on this CPU."));
754 }
755 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
756 {
757 as_bad (_("32bit mode not supported on this CPU."));
758 }
759 stackop_size = '\0';
760 }
761
762 static void
763 set_16bit_gcc_code_flag (new_code_flag)
764 int new_code_flag;
765 {
766 flag_code = new_code_flag;
767 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
768 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
769 stackop_size = 'l';
770 }
771
772 static void
773 set_intel_syntax (syntax_flag)
774 int syntax_flag;
775 {
776 /* Find out if register prefixing is specified. */
777 int ask_naked_reg = 0;
778
779 SKIP_WHITESPACE ();
780 if (!is_end_of_line[(unsigned char) *input_line_pointer])
781 {
782 char *string = input_line_pointer;
783 int e = get_symbol_end ();
784
785 if (strcmp (string, "prefix") == 0)
786 ask_naked_reg = 1;
787 else if (strcmp (string, "noprefix") == 0)
788 ask_naked_reg = -1;
789 else
790 as_bad (_("bad argument to syntax directive."));
791 *input_line_pointer = e;
792 }
793 demand_empty_rest_of_line ();
794
795 intel_syntax = syntax_flag;
796
797 if (ask_naked_reg == 0)
798 {
799 #ifdef BFD_ASSEMBLER
800 allow_naked_reg = (intel_syntax
801 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
802 #else
803 /* Conservative default. */
804 allow_naked_reg = 0;
805 #endif
806 }
807 else
808 allow_naked_reg = (ask_naked_reg < 0);
809 }
810
811 static void
812 set_cpu_arch (dummy)
813 int dummy ATTRIBUTE_UNUSED;
814 {
815 SKIP_WHITESPACE ();
816
817 if (!is_end_of_line[(unsigned char) *input_line_pointer])
818 {
819 char *string = input_line_pointer;
820 int e = get_symbol_end ();
821 int i;
822
823 for (i = 0; cpu_arch[i].name; i++)
824 {
825 if (strcmp (string, cpu_arch[i].name) == 0)
826 {
827 cpu_arch_name = cpu_arch[i].name;
828 cpu_arch_flags = (cpu_arch[i].flags
829 | (flag_code == CODE_64BIT ? Cpu64 : CpuNo64));
830 break;
831 }
832 }
833 if (!cpu_arch[i].name)
834 as_bad (_("no such architecture: `%s'"), string);
835
836 *input_line_pointer = e;
837 }
838 else
839 as_bad (_("missing cpu architecture"));
840
841 no_cond_jump_promotion = 0;
842 if (*input_line_pointer == ','
843 && !is_end_of_line[(unsigned char) input_line_pointer[1]])
844 {
845 char *string = ++input_line_pointer;
846 int e = get_symbol_end ();
847
848 if (strcmp (string, "nojumps") == 0)
849 no_cond_jump_promotion = 1;
850 else if (strcmp (string, "jumps") == 0)
851 ;
852 else
853 as_bad (_("no such architecture modifier: `%s'"), string);
854
855 *input_line_pointer = e;
856 }
857
858 demand_empty_rest_of_line ();
859 }
860
861 #ifdef BFD_ASSEMBLER
862 unsigned long
863 i386_mach ()
864 {
865 if (!strcmp (default_arch, "x86_64"))
866 return bfd_mach_x86_64;
867 else if (!strcmp (default_arch, "i386"))
868 return bfd_mach_i386_i386;
869 else
870 as_fatal (_("Unknown architecture"));
871 }
872 #endif
873 \f
874 void
875 md_begin ()
876 {
877 const char *hash_err;
878
879 /* Initialize op_hash hash table. */
880 op_hash = hash_new ();
881
882 {
883 const template *optab;
884 templates *core_optab;
885
886 /* Setup for loop. */
887 optab = i386_optab;
888 core_optab = (templates *) xmalloc (sizeof (templates));
889 core_optab->start = optab;
890
891 while (1)
892 {
893 ++optab;
894 if (optab->name == NULL
895 || strcmp (optab->name, (optab - 1)->name) != 0)
896 {
897 /* different name --> ship out current template list;
898 add to hash table; & begin anew. */
899 core_optab->end = optab;
900 hash_err = hash_insert (op_hash,
901 (optab - 1)->name,
902 (PTR) core_optab);
903 if (hash_err)
904 {
905 as_fatal (_("Internal Error: Can't hash %s: %s"),
906 (optab - 1)->name,
907 hash_err);
908 }
909 if (optab->name == NULL)
910 break;
911 core_optab = (templates *) xmalloc (sizeof (templates));
912 core_optab->start = optab;
913 }
914 }
915 }
916
917 /* Initialize reg_hash hash table. */
918 reg_hash = hash_new ();
919 {
920 const reg_entry *regtab;
921
922 for (regtab = i386_regtab;
923 regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
924 regtab++)
925 {
926 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
927 if (hash_err)
928 as_fatal (_("Internal Error: Can't hash %s: %s"),
929 regtab->reg_name,
930 hash_err);
931 }
932 }
933
934 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
935 {
936 int c;
937 char *p;
938
939 for (c = 0; c < 256; c++)
940 {
941 if (ISDIGIT (c))
942 {
943 digit_chars[c] = c;
944 mnemonic_chars[c] = c;
945 register_chars[c] = c;
946 operand_chars[c] = c;
947 }
948 else if (ISLOWER (c))
949 {
950 mnemonic_chars[c] = c;
951 register_chars[c] = c;
952 operand_chars[c] = c;
953 }
954 else if (ISUPPER (c))
955 {
956 mnemonic_chars[c] = TOLOWER (c);
957 register_chars[c] = mnemonic_chars[c];
958 operand_chars[c] = c;
959 }
960
961 if (ISALPHA (c) || ISDIGIT (c))
962 identifier_chars[c] = c;
963 else if (c >= 128)
964 {
965 identifier_chars[c] = c;
966 operand_chars[c] = c;
967 }
968 }
969
970 #ifdef LEX_AT
971 identifier_chars['@'] = '@';
972 #endif
973 digit_chars['-'] = '-';
974 identifier_chars['_'] = '_';
975 identifier_chars['.'] = '.';
976
977 for (p = operand_special_chars; *p != '\0'; p++)
978 operand_chars[(unsigned char) *p] = *p;
979 }
980
981 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
982 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
983 {
984 record_alignment (text_section, 2);
985 record_alignment (data_section, 2);
986 record_alignment (bss_section, 2);
987 }
988 #endif
989 }
990
991 void
992 i386_print_statistics (file)
993 FILE *file;
994 {
995 hash_print_statistics (file, "i386 opcode", op_hash);
996 hash_print_statistics (file, "i386 register", reg_hash);
997 }
998 \f
999 #ifdef DEBUG386
1000
1001 /* Debugging routines for md_assemble. */
1002 static void pi PARAMS ((char *, i386_insn *));
1003 static void pte PARAMS ((template *));
1004 static void pt PARAMS ((unsigned int));
1005 static void pe PARAMS ((expressionS *));
1006 static void ps PARAMS ((symbolS *));
1007
1008 static void
1009 pi (line, x)
1010 char *line;
1011 i386_insn *x;
1012 {
1013 unsigned int i;
1014
1015 fprintf (stdout, "%s: template ", line);
1016 pte (&x->tm);
1017 fprintf (stdout, " address: base %s index %s scale %x\n",
1018 x->base_reg ? x->base_reg->reg_name : "none",
1019 x->index_reg ? x->index_reg->reg_name : "none",
1020 x->log2_scale_factor);
1021 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
1022 x->rm.mode, x->rm.reg, x->rm.regmem);
1023 fprintf (stdout, " sib: base %x index %x scale %x\n",
1024 x->sib.base, x->sib.index, x->sib.scale);
1025 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
1026 (x->rex & REX_MODE64) != 0,
1027 (x->rex & REX_EXTX) != 0,
1028 (x->rex & REX_EXTY) != 0,
1029 (x->rex & REX_EXTZ) != 0);
1030 for (i = 0; i < x->operands; i++)
1031 {
1032 fprintf (stdout, " #%d: ", i + 1);
1033 pt (x->types[i]);
1034 fprintf (stdout, "\n");
1035 if (x->types[i]
1036 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
1037 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
1038 if (x->types[i] & Imm)
1039 pe (x->op[i].imms);
1040 if (x->types[i] & Disp)
1041 pe (x->op[i].disps);
1042 }
1043 }
1044
1045 static void
1046 pte (t)
1047 template *t;
1048 {
1049 unsigned int i;
1050 fprintf (stdout, " %d operands ", t->operands);
1051 fprintf (stdout, "opcode %x ", t->base_opcode);
1052 if (t->extension_opcode != None)
1053 fprintf (stdout, "ext %x ", t->extension_opcode);
1054 if (t->opcode_modifier & D)
1055 fprintf (stdout, "D");
1056 if (t->opcode_modifier & W)
1057 fprintf (stdout, "W");
1058 fprintf (stdout, "\n");
1059 for (i = 0; i < t->operands; i++)
1060 {
1061 fprintf (stdout, " #%d type ", i + 1);
1062 pt (t->operand_types[i]);
1063 fprintf (stdout, "\n");
1064 }
1065 }
1066
1067 static void
1068 pe (e)
1069 expressionS *e;
1070 {
1071 fprintf (stdout, " operation %d\n", e->X_op);
1072 fprintf (stdout, " add_number %ld (%lx)\n",
1073 (long) e->X_add_number, (long) e->X_add_number);
1074 if (e->X_add_symbol)
1075 {
1076 fprintf (stdout, " add_symbol ");
1077 ps (e->X_add_symbol);
1078 fprintf (stdout, "\n");
1079 }
1080 if (e->X_op_symbol)
1081 {
1082 fprintf (stdout, " op_symbol ");
1083 ps (e->X_op_symbol);
1084 fprintf (stdout, "\n");
1085 }
1086 }
1087
1088 static void
1089 ps (s)
1090 symbolS *s;
1091 {
1092 fprintf (stdout, "%s type %s%s",
1093 S_GET_NAME (s),
1094 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
1095 segment_name (S_GET_SEGMENT (s)));
1096 }
1097
1098 struct type_name
1099 {
1100 unsigned int mask;
1101 char *tname;
1102 }
1103
1104 static const type_names[] =
1105 {
1106 { Reg8, "r8" },
1107 { Reg16, "r16" },
1108 { Reg32, "r32" },
1109 { Reg64, "r64" },
1110 { Imm8, "i8" },
1111 { Imm8S, "i8s" },
1112 { Imm16, "i16" },
1113 { Imm32, "i32" },
1114 { Imm32S, "i32s" },
1115 { Imm64, "i64" },
1116 { Imm1, "i1" },
1117 { BaseIndex, "BaseIndex" },
1118 { Disp8, "d8" },
1119 { Disp16, "d16" },
1120 { Disp32, "d32" },
1121 { Disp32S, "d32s" },
1122 { Disp64, "d64" },
1123 { InOutPortReg, "InOutPortReg" },
1124 { ShiftCount, "ShiftCount" },
1125 { Control, "control reg" },
1126 { Test, "test reg" },
1127 { Debug, "debug reg" },
1128 { FloatReg, "FReg" },
1129 { FloatAcc, "FAcc" },
1130 { SReg2, "SReg2" },
1131 { SReg3, "SReg3" },
1132 { Acc, "Acc" },
1133 { JumpAbsolute, "Jump Absolute" },
1134 { RegMMX, "rMMX" },
1135 { RegXMM, "rXMM" },
1136 { EsSeg, "es" },
1137 { 0, "" }
1138 };
1139
1140 static void
1141 pt (t)
1142 unsigned int t;
1143 {
1144 const struct type_name *ty;
1145
1146 for (ty = type_names; ty->mask; ty++)
1147 if (t & ty->mask)
1148 fprintf (stdout, "%s, ", ty->tname);
1149 fflush (stdout);
1150 }
1151
1152 #endif /* DEBUG386 */
1153 \f
1154 #ifdef BFD_ASSEMBLER
1155 static bfd_reloc_code_real_type reloc
1156 PARAMS ((int, int, int, bfd_reloc_code_real_type));
1157
1158 static bfd_reloc_code_real_type
1159 reloc (size, pcrel, sign, other)
1160 int size;
1161 int pcrel;
1162 int sign;
1163 bfd_reloc_code_real_type other;
1164 {
1165 if (other != NO_RELOC)
1166 return other;
1167
1168 if (pcrel)
1169 {
1170 if (!sign)
1171 as_bad (_("There are no unsigned pc-relative relocations"));
1172 switch (size)
1173 {
1174 case 1: return BFD_RELOC_8_PCREL;
1175 case 2: return BFD_RELOC_16_PCREL;
1176 case 4: return BFD_RELOC_32_PCREL;
1177 }
1178 as_bad (_("can not do %d byte pc-relative relocation"), size);
1179 }
1180 else
1181 {
1182 if (sign)
1183 switch (size)
1184 {
1185 case 4: return BFD_RELOC_X86_64_32S;
1186 }
1187 else
1188 switch (size)
1189 {
1190 case 1: return BFD_RELOC_8;
1191 case 2: return BFD_RELOC_16;
1192 case 4: return BFD_RELOC_32;
1193 case 8: return BFD_RELOC_64;
1194 }
1195 as_bad (_("can not do %s %d byte relocation"),
1196 sign ? "signed" : "unsigned", size);
1197 }
1198
1199 abort ();
1200 return BFD_RELOC_NONE;
1201 }
1202
1203 /* Here we decide which fixups can be adjusted to make them relative to
1204 the beginning of the section instead of the symbol. Basically we need
1205 to make sure that the dynamic relocations are done correctly, so in
1206 some cases we force the original symbol to be used. */
1207
1208 int
1209 tc_i386_fix_adjustable (fixP)
1210 fixS *fixP ATTRIBUTE_UNUSED;
1211 {
1212 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1213 if (OUTPUT_FLAVOR != bfd_target_elf_flavour)
1214 return 1;
1215
1216 /* Don't adjust pc-relative references to merge sections in 64-bit
1217 mode. */
1218 if (use_rela_relocations
1219 && (S_GET_SEGMENT (fixP->fx_addsy)->flags & SEC_MERGE) != 0
1220 && fixP->fx_pcrel)
1221 return 0;
1222
1223 /* adjust_reloc_syms doesn't know about the GOT. */
1224 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1225 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1226 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1227 || fixP->fx_r_type == BFD_RELOC_386_TLS_GD
1228 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDM
1229 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDO_32
1230 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE_32
1231 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE
1232 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTIE
1233 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE_32
1234 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE
1235 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1236 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1237 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
1238 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSGD
1239 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSLD
1240 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF32
1241 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTTPOFF
1242 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF32
1243 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1244 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1245 return 0;
1246 #endif
1247 return 1;
1248 }
1249 #else
1250 #define reloc(SIZE,PCREL,SIGN,OTHER) 0
1251 #define BFD_RELOC_8 0
1252 #define BFD_RELOC_16 0
1253 #define BFD_RELOC_32 0
1254 #define BFD_RELOC_8_PCREL 0
1255 #define BFD_RELOC_16_PCREL 0
1256 #define BFD_RELOC_32_PCREL 0
1257 #define BFD_RELOC_386_PLT32 0
1258 #define BFD_RELOC_386_GOT32 0
1259 #define BFD_RELOC_386_GOTOFF 0
1260 #define BFD_RELOC_386_TLS_GD 0
1261 #define BFD_RELOC_386_TLS_LDM 0
1262 #define BFD_RELOC_386_TLS_LDO_32 0
1263 #define BFD_RELOC_386_TLS_IE_32 0
1264 #define BFD_RELOC_386_TLS_IE 0
1265 #define BFD_RELOC_386_TLS_GOTIE 0
1266 #define BFD_RELOC_386_TLS_LE_32 0
1267 #define BFD_RELOC_386_TLS_LE 0
1268 #define BFD_RELOC_X86_64_PLT32 0
1269 #define BFD_RELOC_X86_64_GOT32 0
1270 #define BFD_RELOC_X86_64_GOTPCREL 0
1271 #define BFD_RELOC_X86_64_TLSGD 0
1272 #define BFD_RELOC_X86_64_TLSLD 0
1273 #define BFD_RELOC_X86_64_DTPOFF32 0
1274 #define BFD_RELOC_X86_64_GOTTPOFF 0
1275 #define BFD_RELOC_X86_64_TPOFF32 0
1276 #endif
1277
1278 static int intel_float_operand PARAMS ((const char *mnemonic));
1279
1280 static int
1281 intel_float_operand (mnemonic)
1282 const char *mnemonic;
1283 {
1284 if (mnemonic[0] == 'f' && mnemonic[1] == 'i')
1285 return 2;
1286
1287 if (mnemonic[0] == 'f')
1288 return 1;
1289
1290 return 0;
1291 }
1292
1293 /* This is the guts of the machine-dependent assembler. LINE points to a
1294 machine dependent instruction. This function is supposed to emit
1295 the frags/bytes it assembles to. */
1296
1297 void
1298 md_assemble (line)
1299 char *line;
1300 {
1301 int j;
1302 char mnemonic[MAX_MNEM_SIZE];
1303
1304 /* Initialize globals. */
1305 memset (&i, '\0', sizeof (i));
1306 for (j = 0; j < MAX_OPERANDS; j++)
1307 i.reloc[j] = NO_RELOC;
1308 memset (disp_expressions, '\0', sizeof (disp_expressions));
1309 memset (im_expressions, '\0', sizeof (im_expressions));
1310 save_stack_p = save_stack;
1311
1312 /* First parse an instruction mnemonic & call i386_operand for the operands.
1313 We assume that the scrubber has arranged it so that line[0] is the valid
1314 start of a (possibly prefixed) mnemonic. */
1315
1316 line = parse_insn (line, mnemonic);
1317 if (line == NULL)
1318 return;
1319
1320 line = parse_operands (line, mnemonic);
1321 if (line == NULL)
1322 return;
1323
1324 /* Now we've parsed the mnemonic into a set of templates, and have the
1325 operands at hand. */
1326
1327 /* All intel opcodes have reversed operands except for "bound" and
1328 "enter". We also don't reverse intersegment "jmp" and "call"
1329 instructions with 2 immediate operands so that the immediate segment
1330 precedes the offset, as it does when in AT&T mode. "enter" and the
1331 intersegment "jmp" and "call" instructions are the only ones that
1332 have two immediate operands. */
1333 if (intel_syntax && i.operands > 1
1334 && (strcmp (mnemonic, "bound") != 0)
1335 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1336 swap_operands ();
1337
1338 if (i.imm_operands)
1339 optimize_imm ();
1340
1341 if (i.disp_operands)
1342 optimize_disp ();
1343
1344 /* Next, we find a template that matches the given insn,
1345 making sure the overlap of the given operands types is consistent
1346 with the template operand types. */
1347
1348 if (!match_template ())
1349 return;
1350
1351 if (intel_syntax)
1352 {
1353 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1354 if (SYSV386_COMPAT
1355 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1356 i.tm.base_opcode ^= FloatR;
1357
1358 /* Zap movzx and movsx suffix. The suffix may have been set from
1359 "word ptr" or "byte ptr" on the source operand, but we'll use
1360 the suffix later to choose the destination register. */
1361 if ((i.tm.base_opcode & ~9) == 0x0fb6)
1362 i.suffix = 0;
1363 }
1364
1365 if (i.tm.opcode_modifier & FWait)
1366 if (!add_prefix (FWAIT_OPCODE))
1367 return;
1368
1369 /* Check string instruction segment overrides. */
1370 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1371 {
1372 if (!check_string ())
1373 return;
1374 }
1375
1376 if (!process_suffix ())
1377 return;
1378
1379 /* Make still unresolved immediate matches conform to size of immediate
1380 given in i.suffix. */
1381 if (!finalize_imm ())
1382 return;
1383
1384 if (i.types[0] & Imm1)
1385 i.imm_operands = 0; /* kludge for shift insns. */
1386 if (i.types[0] & ImplicitRegister)
1387 i.reg_operands--;
1388 if (i.types[1] & ImplicitRegister)
1389 i.reg_operands--;
1390 if (i.types[2] & ImplicitRegister)
1391 i.reg_operands--;
1392
1393 if (i.tm.opcode_modifier & ImmExt)
1394 {
1395 /* These AMD 3DNow! and Intel Katmai New Instructions have an
1396 opcode suffix which is coded in the same place as an 8-bit
1397 immediate field would be. Here we fake an 8-bit immediate
1398 operand from the opcode suffix stored in tm.extension_opcode. */
1399
1400 expressionS *exp;
1401
1402 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
1403
1404 exp = &im_expressions[i.imm_operands++];
1405 i.op[i.operands].imms = exp;
1406 i.types[i.operands++] = Imm8;
1407 exp->X_op = O_constant;
1408 exp->X_add_number = i.tm.extension_opcode;
1409 i.tm.extension_opcode = None;
1410 }
1411
1412 /* For insns with operands there are more diddles to do to the opcode. */
1413 if (i.operands)
1414 {
1415 if (!process_operands ())
1416 return;
1417 }
1418 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
1419 {
1420 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
1421 as_warn (_("translating to `%sp'"), i.tm.name);
1422 }
1423
1424 /* Handle conversion of 'int $3' --> special int3 insn. */
1425 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
1426 {
1427 i.tm.base_opcode = INT3_OPCODE;
1428 i.imm_operands = 0;
1429 }
1430
1431 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
1432 && i.op[0].disps->X_op == O_constant)
1433 {
1434 /* Convert "jmp constant" (and "call constant") to a jump (call) to
1435 the absolute address given by the constant. Since ix86 jumps and
1436 calls are pc relative, we need to generate a reloc. */
1437 i.op[0].disps->X_add_symbol = &abs_symbol;
1438 i.op[0].disps->X_op = O_symbol;
1439 }
1440
1441 if ((i.tm.opcode_modifier & Rex64) != 0)
1442 i.rex |= REX_MODE64;
1443
1444 /* For 8 bit registers we need an empty rex prefix. Also if the
1445 instruction already has a prefix, we need to convert old
1446 registers to new ones. */
1447
1448 if (((i.types[0] & Reg8) != 0
1449 && (i.op[0].regs->reg_flags & RegRex64) != 0)
1450 || ((i.types[1] & Reg8) != 0
1451 && (i.op[1].regs->reg_flags & RegRex64) != 0)
1452 || (((i.types[0] & Reg8) != 0 || (i.types[1] & Reg8) != 0)
1453 && i.rex != 0))
1454 {
1455 int x;
1456
1457 i.rex |= REX_OPCODE;
1458 for (x = 0; x < 2; x++)
1459 {
1460 /* Look for 8 bit operand that uses old registers. */
1461 if ((i.types[x] & Reg8) != 0
1462 && (i.op[x].regs->reg_flags & RegRex64) == 0)
1463 {
1464 /* In case it is "hi" register, give up. */
1465 if (i.op[x].regs->reg_num > 3)
1466 as_bad (_("can't encode register '%%%s' in an instruction requiring REX prefix.\n"),
1467 i.op[x].regs->reg_name);
1468
1469 /* Otherwise it is equivalent to the extended register.
1470 Since the encoding doesn't change this is merely
1471 cosmetic cleanup for debug output. */
1472
1473 i.op[x].regs = i.op[x].regs + 8;
1474 }
1475 }
1476 }
1477
1478 if (i.rex != 0)
1479 add_prefix (REX_OPCODE | i.rex);
1480
1481 /* We are ready to output the insn. */
1482 output_insn ();
1483 }
1484
1485 static char *
1486 parse_insn (line, mnemonic)
1487 char *line;
1488 char *mnemonic;
1489 {
1490 char *l = line;
1491 char *token_start = l;
1492 char *mnem_p;
1493
1494 /* Non-zero if we found a prefix only acceptable with string insns. */
1495 const char *expecting_string_instruction = NULL;
1496
1497 while (1)
1498 {
1499 mnem_p = mnemonic;
1500 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1501 {
1502 mnem_p++;
1503 if (mnem_p >= mnemonic + MAX_MNEM_SIZE)
1504 {
1505 as_bad (_("no such instruction: `%s'"), token_start);
1506 return NULL;
1507 }
1508 l++;
1509 }
1510 if (!is_space_char (*l)
1511 && *l != END_OF_INSN
1512 && *l != PREFIX_SEPARATOR
1513 && *l != ',')
1514 {
1515 as_bad (_("invalid character %s in mnemonic"),
1516 output_invalid (*l));
1517 return NULL;
1518 }
1519 if (token_start == l)
1520 {
1521 if (*l == PREFIX_SEPARATOR)
1522 as_bad (_("expecting prefix; got nothing"));
1523 else
1524 as_bad (_("expecting mnemonic; got nothing"));
1525 return NULL;
1526 }
1527
1528 /* Look up instruction (or prefix) via hash table. */
1529 current_templates = hash_find (op_hash, mnemonic);
1530
1531 if (*l != END_OF_INSN
1532 && (!is_space_char (*l) || l[1] != END_OF_INSN)
1533 && current_templates
1534 && (current_templates->start->opcode_modifier & IsPrefix))
1535 {
1536 /* If we are in 16-bit mode, do not allow addr16 or data16.
1537 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1538 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1539 && flag_code != CODE_64BIT
1540 && (((current_templates->start->opcode_modifier & Size32) != 0)
1541 ^ (flag_code == CODE_16BIT)))
1542 {
1543 as_bad (_("redundant %s prefix"),
1544 current_templates->start->name);
1545 return NULL;
1546 }
1547 /* Add prefix, checking for repeated prefixes. */
1548 switch (add_prefix (current_templates->start->base_opcode))
1549 {
1550 case 0:
1551 return NULL;
1552 case 2:
1553 expecting_string_instruction = current_templates->start->name;
1554 break;
1555 }
1556 /* Skip past PREFIX_SEPARATOR and reset token_start. */
1557 token_start = ++l;
1558 }
1559 else
1560 break;
1561 }
1562
1563 if (!current_templates)
1564 {
1565 /* See if we can get a match by trimming off a suffix. */
1566 switch (mnem_p[-1])
1567 {
1568 case WORD_MNEM_SUFFIX:
1569 case BYTE_MNEM_SUFFIX:
1570 case QWORD_MNEM_SUFFIX:
1571 i.suffix = mnem_p[-1];
1572 mnem_p[-1] = '\0';
1573 current_templates = hash_find (op_hash, mnemonic);
1574 break;
1575 case SHORT_MNEM_SUFFIX:
1576 case LONG_MNEM_SUFFIX:
1577 if (!intel_syntax)
1578 {
1579 i.suffix = mnem_p[-1];
1580 mnem_p[-1] = '\0';
1581 current_templates = hash_find (op_hash, mnemonic);
1582 }
1583 break;
1584
1585 /* Intel Syntax. */
1586 case 'd':
1587 if (intel_syntax)
1588 {
1589 if (intel_float_operand (mnemonic))
1590 i.suffix = SHORT_MNEM_SUFFIX;
1591 else
1592 i.suffix = LONG_MNEM_SUFFIX;
1593 mnem_p[-1] = '\0';
1594 current_templates = hash_find (op_hash, mnemonic);
1595 }
1596 break;
1597 }
1598 if (!current_templates)
1599 {
1600 as_bad (_("no such instruction: `%s'"), token_start);
1601 return NULL;
1602 }
1603 }
1604
1605 if (current_templates->start->opcode_modifier & (Jump | JumpByte))
1606 {
1607 /* Check for a branch hint. We allow ",pt" and ",pn" for
1608 predict taken and predict not taken respectively.
1609 I'm not sure that branch hints actually do anything on loop
1610 and jcxz insns (JumpByte) for current Pentium4 chips. They
1611 may work in the future and it doesn't hurt to accept them
1612 now. */
1613 if (l[0] == ',' && l[1] == 'p')
1614 {
1615 if (l[2] == 't')
1616 {
1617 if (!add_prefix (DS_PREFIX_OPCODE))
1618 return NULL;
1619 l += 3;
1620 }
1621 else if (l[2] == 'n')
1622 {
1623 if (!add_prefix (CS_PREFIX_OPCODE))
1624 return NULL;
1625 l += 3;
1626 }
1627 }
1628 }
1629 /* Any other comma loses. */
1630 if (*l == ',')
1631 {
1632 as_bad (_("invalid character %s in mnemonic"),
1633 output_invalid (*l));
1634 return NULL;
1635 }
1636
1637 /* Check if instruction is supported on specified architecture. */
1638 if ((current_templates->start->cpu_flags & ~(Cpu64 | CpuNo64))
1639 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64)))
1640 {
1641 as_warn (_("`%s' is not supported on `%s'"),
1642 current_templates->start->name, cpu_arch_name);
1643 }
1644 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
1645 {
1646 as_warn (_("use .code16 to ensure correct addressing mode"));
1647 }
1648
1649 /* Check for rep/repne without a string instruction. */
1650 if (expecting_string_instruction
1651 && !(current_templates->start->opcode_modifier & IsString))
1652 {
1653 as_bad (_("expecting string instruction after `%s'"),
1654 expecting_string_instruction);
1655 return NULL;
1656 }
1657
1658 return l;
1659 }
1660
1661 static char *
1662 parse_operands (l, mnemonic)
1663 char *l;
1664 const char *mnemonic;
1665 {
1666 char *token_start;
1667
1668 /* 1 if operand is pending after ','. */
1669 unsigned int expecting_operand = 0;
1670
1671 /* Non-zero if operand parens not balanced. */
1672 unsigned int paren_not_balanced;
1673
1674 while (*l != END_OF_INSN)
1675 {
1676 /* Skip optional white space before operand. */
1677 if (is_space_char (*l))
1678 ++l;
1679 if (!is_operand_char (*l) && *l != END_OF_INSN)
1680 {
1681 as_bad (_("invalid character %s before operand %d"),
1682 output_invalid (*l),
1683 i.operands + 1);
1684 return NULL;
1685 }
1686 token_start = l; /* after white space */
1687 paren_not_balanced = 0;
1688 while (paren_not_balanced || *l != ',')
1689 {
1690 if (*l == END_OF_INSN)
1691 {
1692 if (paren_not_balanced)
1693 {
1694 if (!intel_syntax)
1695 as_bad (_("unbalanced parenthesis in operand %d."),
1696 i.operands + 1);
1697 else
1698 as_bad (_("unbalanced brackets in operand %d."),
1699 i.operands + 1);
1700 return NULL;
1701 }
1702 else
1703 break; /* we are done */
1704 }
1705 else if (!is_operand_char (*l) && !is_space_char (*l))
1706 {
1707 as_bad (_("invalid character %s in operand %d"),
1708 output_invalid (*l),
1709 i.operands + 1);
1710 return NULL;
1711 }
1712 if (!intel_syntax)
1713 {
1714 if (*l == '(')
1715 ++paren_not_balanced;
1716 if (*l == ')')
1717 --paren_not_balanced;
1718 }
1719 else
1720 {
1721 if (*l == '[')
1722 ++paren_not_balanced;
1723 if (*l == ']')
1724 --paren_not_balanced;
1725 }
1726 l++;
1727 }
1728 if (l != token_start)
1729 { /* Yes, we've read in another operand. */
1730 unsigned int operand_ok;
1731 this_operand = i.operands++;
1732 if (i.operands > MAX_OPERANDS)
1733 {
1734 as_bad (_("spurious operands; (%d operands/instruction max)"),
1735 MAX_OPERANDS);
1736 return NULL;
1737 }
1738 /* Now parse operand adding info to 'i' as we go along. */
1739 END_STRING_AND_SAVE (l);
1740
1741 if (intel_syntax)
1742 operand_ok =
1743 i386_intel_operand (token_start,
1744 intel_float_operand (mnemonic));
1745 else
1746 operand_ok = i386_operand (token_start);
1747
1748 RESTORE_END_STRING (l);
1749 if (!operand_ok)
1750 return NULL;
1751 }
1752 else
1753 {
1754 if (expecting_operand)
1755 {
1756 expecting_operand_after_comma:
1757 as_bad (_("expecting operand after ','; got nothing"));
1758 return NULL;
1759 }
1760 if (*l == ',')
1761 {
1762 as_bad (_("expecting operand before ','; got nothing"));
1763 return NULL;
1764 }
1765 }
1766
1767 /* Now *l must be either ',' or END_OF_INSN. */
1768 if (*l == ',')
1769 {
1770 if (*++l == END_OF_INSN)
1771 {
1772 /* Just skip it, if it's \n complain. */
1773 goto expecting_operand_after_comma;
1774 }
1775 expecting_operand = 1;
1776 }
1777 }
1778 return l;
1779 }
1780
1781 static void
1782 swap_operands ()
1783 {
1784 union i386_op temp_op;
1785 unsigned int temp_type;
1786 RELOC_ENUM temp_reloc;
1787 int xchg1 = 0;
1788 int xchg2 = 0;
1789
1790 if (i.operands == 2)
1791 {
1792 xchg1 = 0;
1793 xchg2 = 1;
1794 }
1795 else if (i.operands == 3)
1796 {
1797 xchg1 = 0;
1798 xchg2 = 2;
1799 }
1800 temp_type = i.types[xchg2];
1801 i.types[xchg2] = i.types[xchg1];
1802 i.types[xchg1] = temp_type;
1803 temp_op = i.op[xchg2];
1804 i.op[xchg2] = i.op[xchg1];
1805 i.op[xchg1] = temp_op;
1806 temp_reloc = i.reloc[xchg2];
1807 i.reloc[xchg2] = i.reloc[xchg1];
1808 i.reloc[xchg1] = temp_reloc;
1809
1810 if (i.mem_operands == 2)
1811 {
1812 const seg_entry *temp_seg;
1813 temp_seg = i.seg[0];
1814 i.seg[0] = i.seg[1];
1815 i.seg[1] = temp_seg;
1816 }
1817 }
1818
1819 /* Try to ensure constant immediates are represented in the smallest
1820 opcode possible. */
1821 static void
1822 optimize_imm ()
1823 {
1824 char guess_suffix = 0;
1825 int op;
1826
1827 if (i.suffix)
1828 guess_suffix = i.suffix;
1829 else if (i.reg_operands)
1830 {
1831 /* Figure out a suffix from the last register operand specified.
1832 We can't do this properly yet, ie. excluding InOutPortReg,
1833 but the following works for instructions with immediates.
1834 In any case, we can't set i.suffix yet. */
1835 for (op = i.operands; --op >= 0;)
1836 if (i.types[op] & Reg)
1837 {
1838 if (i.types[op] & Reg8)
1839 guess_suffix = BYTE_MNEM_SUFFIX;
1840 else if (i.types[op] & Reg16)
1841 guess_suffix = WORD_MNEM_SUFFIX;
1842 else if (i.types[op] & Reg32)
1843 guess_suffix = LONG_MNEM_SUFFIX;
1844 else if (i.types[op] & Reg64)
1845 guess_suffix = QWORD_MNEM_SUFFIX;
1846 break;
1847 }
1848 }
1849 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
1850 guess_suffix = WORD_MNEM_SUFFIX;
1851
1852 for (op = i.operands; --op >= 0;)
1853 if (i.types[op] & Imm)
1854 {
1855 switch (i.op[op].imms->X_op)
1856 {
1857 case O_constant:
1858 /* If a suffix is given, this operand may be shortened. */
1859 switch (guess_suffix)
1860 {
1861 case LONG_MNEM_SUFFIX:
1862 i.types[op] |= Imm32 | Imm64;
1863 break;
1864 case WORD_MNEM_SUFFIX:
1865 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
1866 break;
1867 case BYTE_MNEM_SUFFIX:
1868 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
1869 break;
1870 }
1871
1872 /* If this operand is at most 16 bits, convert it
1873 to a signed 16 bit number before trying to see
1874 whether it will fit in an even smaller size.
1875 This allows a 16-bit operand such as $0xffe0 to
1876 be recognised as within Imm8S range. */
1877 if ((i.types[op] & Imm16)
1878 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
1879 {
1880 i.op[op].imms->X_add_number =
1881 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
1882 }
1883 if ((i.types[op] & Imm32)
1884 && ((i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1))
1885 == 0))
1886 {
1887 i.op[op].imms->X_add_number = ((i.op[op].imms->X_add_number
1888 ^ ((offsetT) 1 << 31))
1889 - ((offsetT) 1 << 31));
1890 }
1891 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
1892
1893 /* We must avoid matching of Imm32 templates when 64bit
1894 only immediate is available. */
1895 if (guess_suffix == QWORD_MNEM_SUFFIX)
1896 i.types[op] &= ~Imm32;
1897 break;
1898
1899 case O_absent:
1900 case O_register:
1901 abort ();
1902
1903 /* Symbols and expressions. */
1904 default:
1905 /* Convert symbolic operand to proper sizes for matching. */
1906 switch (guess_suffix)
1907 {
1908 case QWORD_MNEM_SUFFIX:
1909 i.types[op] = Imm64 | Imm32S;
1910 break;
1911 case LONG_MNEM_SUFFIX:
1912 i.types[op] = Imm32 | Imm64;
1913 break;
1914 case WORD_MNEM_SUFFIX:
1915 i.types[op] = Imm16 | Imm32 | Imm64;
1916 break;
1917 break;
1918 case BYTE_MNEM_SUFFIX:
1919 i.types[op] = Imm8 | Imm8S | Imm16 | Imm32S | Imm32;
1920 break;
1921 break;
1922 }
1923 break;
1924 }
1925 }
1926 }
1927
1928 /* Try to use the smallest displacement type too. */
1929 static void
1930 optimize_disp ()
1931 {
1932 int op;
1933
1934 for (op = i.operands; --op >= 0;)
1935 if ((i.types[op] & Disp) && i.op[op].disps->X_op == O_constant)
1936 {
1937 offsetT disp = i.op[op].disps->X_add_number;
1938
1939 if (i.types[op] & Disp16)
1940 {
1941 /* We know this operand is at most 16 bits, so
1942 convert to a signed 16 bit number before trying
1943 to see whether it will fit in an even smaller
1944 size. */
1945
1946 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
1947 }
1948 else if (i.types[op] & Disp32)
1949 {
1950 /* We know this operand is at most 32 bits, so convert to a
1951 signed 32 bit number before trying to see whether it will
1952 fit in an even smaller size. */
1953 disp &= (((offsetT) 2 << 31) - 1);
1954 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1955 }
1956 if (flag_code == CODE_64BIT)
1957 {
1958 if (fits_in_signed_long (disp))
1959 i.types[op] |= Disp32S;
1960 if (fits_in_unsigned_long (disp))
1961 i.types[op] |= Disp32;
1962 }
1963 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
1964 && fits_in_signed_byte (disp))
1965 i.types[op] |= Disp8;
1966 }
1967 }
1968
1969 static int
1970 match_template ()
1971 {
1972 /* Points to template once we've found it. */
1973 const template *t;
1974 unsigned int overlap0, overlap1, overlap2;
1975 unsigned int found_reverse_match;
1976 int suffix_check;
1977
1978 #define MATCH(overlap, given, template) \
1979 ((overlap & ~JumpAbsolute) \
1980 && (((given) & (BaseIndex | JumpAbsolute)) \
1981 == ((overlap) & (BaseIndex | JumpAbsolute))))
1982
1983 /* If given types r0 and r1 are registers they must be of the same type
1984 unless the expected operand type register overlap is null.
1985 Note that Acc in a template matches every size of reg. */
1986 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
1987 (((g0) & Reg) == 0 || ((g1) & Reg) == 0 \
1988 || ((g0) & Reg) == ((g1) & Reg) \
1989 || ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
1990
1991 overlap0 = 0;
1992 overlap1 = 0;
1993 overlap2 = 0;
1994 found_reverse_match = 0;
1995 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
1996 ? No_bSuf
1997 : (i.suffix == WORD_MNEM_SUFFIX
1998 ? No_wSuf
1999 : (i.suffix == SHORT_MNEM_SUFFIX
2000 ? No_sSuf
2001 : (i.suffix == LONG_MNEM_SUFFIX
2002 ? No_lSuf
2003 : (i.suffix == QWORD_MNEM_SUFFIX
2004 ? No_qSuf
2005 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX
2006 ? No_xSuf : 0))))));
2007
2008 for (t = current_templates->start;
2009 t < current_templates->end;
2010 t++)
2011 {
2012 /* Must have right number of operands. */
2013 if (i.operands != t->operands)
2014 continue;
2015
2016 /* Check the suffix, except for some instructions in intel mode. */
2017 if ((t->opcode_modifier & suffix_check)
2018 && !(intel_syntax
2019 && (t->opcode_modifier & IgnoreSize))
2020 && !(intel_syntax
2021 && t->base_opcode == 0xd9
2022 && (t->extension_opcode == 5 /* 0xd9,5 "fldcw" */
2023 || t->extension_opcode == 7))) /* 0xd9,7 "f{n}stcw" */
2024 continue;
2025
2026 /* Do not verify operands when there are none. */
2027 else if (!t->operands)
2028 {
2029 if (t->cpu_flags & ~cpu_arch_flags)
2030 continue;
2031 /* We've found a match; break out of loop. */
2032 break;
2033 }
2034
2035 overlap0 = i.types[0] & t->operand_types[0];
2036 switch (t->operands)
2037 {
2038 case 1:
2039 if (!MATCH (overlap0, i.types[0], t->operand_types[0]))
2040 continue;
2041 break;
2042 case 2:
2043 case 3:
2044 overlap1 = i.types[1] & t->operand_types[1];
2045 if (!MATCH (overlap0, i.types[0], t->operand_types[0])
2046 || !MATCH (overlap1, i.types[1], t->operand_types[1])
2047 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2048 t->operand_types[0],
2049 overlap1, i.types[1],
2050 t->operand_types[1]))
2051 {
2052 /* Check if other direction is valid ... */
2053 if ((t->opcode_modifier & (D | FloatD)) == 0)
2054 continue;
2055
2056 /* Try reversing direction of operands. */
2057 overlap0 = i.types[0] & t->operand_types[1];
2058 overlap1 = i.types[1] & t->operand_types[0];
2059 if (!MATCH (overlap0, i.types[0], t->operand_types[1])
2060 || !MATCH (overlap1, i.types[1], t->operand_types[0])
2061 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2062 t->operand_types[1],
2063 overlap1, i.types[1],
2064 t->operand_types[0]))
2065 {
2066 /* Does not match either direction. */
2067 continue;
2068 }
2069 /* found_reverse_match holds which of D or FloatDR
2070 we've found. */
2071 found_reverse_match = t->opcode_modifier & (D | FloatDR);
2072 }
2073 /* Found a forward 2 operand match here. */
2074 else if (t->operands == 3)
2075 {
2076 /* Here we make use of the fact that there are no
2077 reverse match 3 operand instructions, and all 3
2078 operand instructions only need to be checked for
2079 register consistency between operands 2 and 3. */
2080 overlap2 = i.types[2] & t->operand_types[2];
2081 if (!MATCH (overlap2, i.types[2], t->operand_types[2])
2082 || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1],
2083 t->operand_types[1],
2084 overlap2, i.types[2],
2085 t->operand_types[2]))
2086
2087 continue;
2088 }
2089 /* Found either forward/reverse 2 or 3 operand match here:
2090 slip through to break. */
2091 }
2092 if (t->cpu_flags & ~cpu_arch_flags)
2093 {
2094 found_reverse_match = 0;
2095 continue;
2096 }
2097 /* We've found a match; break out of loop. */
2098 break;
2099 }
2100
2101 if (t == current_templates->end)
2102 {
2103 /* We found no match. */
2104 as_bad (_("suffix or operands invalid for `%s'"),
2105 current_templates->start->name);
2106 return 0;
2107 }
2108
2109 if (!quiet_warnings)
2110 {
2111 if (!intel_syntax
2112 && ((i.types[0] & JumpAbsolute)
2113 != (t->operand_types[0] & JumpAbsolute)))
2114 {
2115 as_warn (_("indirect %s without `*'"), t->name);
2116 }
2117
2118 if ((t->opcode_modifier & (IsPrefix | IgnoreSize))
2119 == (IsPrefix | IgnoreSize))
2120 {
2121 /* Warn them that a data or address size prefix doesn't
2122 affect assembly of the next line of code. */
2123 as_warn (_("stand-alone `%s' prefix"), t->name);
2124 }
2125 }
2126
2127 /* Copy the template we found. */
2128 i.tm = *t;
2129 if (found_reverse_match)
2130 {
2131 /* If we found a reverse match we must alter the opcode
2132 direction bit. found_reverse_match holds bits to change
2133 (different for int & float insns). */
2134
2135 i.tm.base_opcode ^= found_reverse_match;
2136
2137 i.tm.operand_types[0] = t->operand_types[1];
2138 i.tm.operand_types[1] = t->operand_types[0];
2139 }
2140
2141 return 1;
2142 }
2143
2144 static int
2145 check_string ()
2146 {
2147 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
2148 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
2149 {
2150 if (i.seg[0] != NULL && i.seg[0] != &es)
2151 {
2152 as_bad (_("`%s' operand %d must use `%%es' segment"),
2153 i.tm.name,
2154 mem_op + 1);
2155 return 0;
2156 }
2157 /* There's only ever one segment override allowed per instruction.
2158 This instruction possibly has a legal segment override on the
2159 second operand, so copy the segment to where non-string
2160 instructions store it, allowing common code. */
2161 i.seg[0] = i.seg[1];
2162 }
2163 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
2164 {
2165 if (i.seg[1] != NULL && i.seg[1] != &es)
2166 {
2167 as_bad (_("`%s' operand %d must use `%%es' segment"),
2168 i.tm.name,
2169 mem_op + 2);
2170 return 0;
2171 }
2172 }
2173 return 1;
2174 }
2175
2176 static int
2177 process_suffix ()
2178 {
2179 /* If matched instruction specifies an explicit instruction mnemonic
2180 suffix, use it. */
2181 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
2182 {
2183 if (i.tm.opcode_modifier & Size16)
2184 i.suffix = WORD_MNEM_SUFFIX;
2185 else if (i.tm.opcode_modifier & Size64)
2186 i.suffix = QWORD_MNEM_SUFFIX;
2187 else
2188 i.suffix = LONG_MNEM_SUFFIX;
2189 }
2190 else if (i.reg_operands)
2191 {
2192 /* If there's no instruction mnemonic suffix we try to invent one
2193 based on register operands. */
2194 if (!i.suffix)
2195 {
2196 /* We take i.suffix from the last register operand specified,
2197 Destination register type is more significant than source
2198 register type. */
2199 int op;
2200 for (op = i.operands; --op >= 0;)
2201 if ((i.types[op] & Reg)
2202 && !(i.tm.operand_types[op] & InOutPortReg))
2203 {
2204 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
2205 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
2206 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
2207 LONG_MNEM_SUFFIX);
2208 break;
2209 }
2210 }
2211 else if (i.suffix == BYTE_MNEM_SUFFIX)
2212 {
2213 if (!check_byte_reg ())
2214 return 0;
2215 }
2216 else if (i.suffix == LONG_MNEM_SUFFIX)
2217 {
2218 if (!check_long_reg ())
2219 return 0;
2220 }
2221 else if (i.suffix == QWORD_MNEM_SUFFIX)
2222 {
2223 if (!check_qword_reg ())
2224 return 0;
2225 }
2226 else if (i.suffix == WORD_MNEM_SUFFIX)
2227 {
2228 if (!check_word_reg ())
2229 return 0;
2230 }
2231 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2232 /* Do nothing if the instruction is going to ignore the prefix. */
2233 ;
2234 else
2235 abort ();
2236 }
2237 else if ((i.tm.opcode_modifier & DefaultSize) && !i.suffix)
2238 {
2239 i.suffix = stackop_size;
2240 }
2241
2242 /* Change the opcode based on the operand size given by i.suffix;
2243 We need not change things for byte insns. */
2244
2245 if (!i.suffix && (i.tm.opcode_modifier & W))
2246 {
2247 as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
2248 return 0;
2249 }
2250
2251 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2252 {
2253 /* It's not a byte, select word/dword operation. */
2254 if (i.tm.opcode_modifier & W)
2255 {
2256 if (i.tm.opcode_modifier & ShortForm)
2257 i.tm.base_opcode |= 8;
2258 else
2259 i.tm.base_opcode |= 1;
2260 }
2261
2262 /* Now select between word & dword operations via the operand
2263 size prefix, except for instructions that will ignore this
2264 prefix anyway. */
2265 if (i.suffix != QWORD_MNEM_SUFFIX
2266 && !(i.tm.opcode_modifier & IgnoreSize)
2267 && ((i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2268 || (flag_code == CODE_64BIT
2269 && (i.tm.opcode_modifier & JumpByte))))
2270 {
2271 unsigned int prefix = DATA_PREFIX_OPCODE;
2272 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2273 prefix = ADDR_PREFIX_OPCODE;
2274
2275 if (!add_prefix (prefix))
2276 return 0;
2277 }
2278
2279 /* Set mode64 for an operand. */
2280 if (i.suffix == QWORD_MNEM_SUFFIX
2281 && flag_code == CODE_64BIT
2282 && (i.tm.opcode_modifier & NoRex64) == 0)
2283 i.rex |= REX_MODE64;
2284
2285 /* Size floating point instruction. */
2286 if (i.suffix == LONG_MNEM_SUFFIX)
2287 {
2288 if (i.tm.opcode_modifier & FloatMF)
2289 i.tm.base_opcode ^= 4;
2290 }
2291 }
2292
2293 return 1;
2294 }
2295
2296 static int
2297 check_byte_reg ()
2298 {
2299 int op;
2300 for (op = i.operands; --op >= 0;)
2301 {
2302 /* If this is an eight bit register, it's OK. If it's the 16 or
2303 32 bit version of an eight bit register, we will just use the
2304 low portion, and that's OK too. */
2305 if (i.types[op] & Reg8)
2306 continue;
2307
2308 /* movzx and movsx should not generate this warning. */
2309 if (intel_syntax
2310 && (i.tm.base_opcode == 0xfb7
2311 || i.tm.base_opcode == 0xfb6
2312 || i.tm.base_opcode == 0x63
2313 || i.tm.base_opcode == 0xfbe
2314 || i.tm.base_opcode == 0xfbf))
2315 continue;
2316
2317 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4
2318 #if 0
2319 /* Check that the template allows eight bit regs. This
2320 kills insns such as `orb $1,%edx', which maybe should be
2321 allowed. */
2322 && (i.tm.operand_types[op] & (Reg8 | InOutPortReg))
2323 #endif
2324 )
2325 {
2326 /* Prohibit these changes in the 64bit mode, since the
2327 lowering is more complicated. */
2328 if (flag_code == CODE_64BIT
2329 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2330 {
2331 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2332 i.op[op].regs->reg_name,
2333 i.suffix);
2334 return 0;
2335 }
2336 #if REGISTER_WARNINGS
2337 if (!quiet_warnings
2338 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2339 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2340 (i.op[op].regs + (i.types[op] & Reg16
2341 ? REGNAM_AL - REGNAM_AX
2342 : REGNAM_AL - REGNAM_EAX))->reg_name,
2343 i.op[op].regs->reg_name,
2344 i.suffix);
2345 #endif
2346 continue;
2347 }
2348 /* Any other register is bad. */
2349 if (i.types[op] & (Reg | RegMMX | RegXMM
2350 | SReg2 | SReg3
2351 | Control | Debug | Test
2352 | FloatReg | FloatAcc))
2353 {
2354 as_bad (_("`%%%s' not allowed with `%s%c'"),
2355 i.op[op].regs->reg_name,
2356 i.tm.name,
2357 i.suffix);
2358 return 0;
2359 }
2360 }
2361 return 1;
2362 }
2363
2364 static int
2365 check_long_reg ()
2366 {
2367 int op;
2368
2369 for (op = i.operands; --op >= 0;)
2370 /* Reject eight bit registers, except where the template requires
2371 them. (eg. movzb) */
2372 if ((i.types[op] & Reg8) != 0
2373 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2374 {
2375 as_bad (_("`%%%s' not allowed with `%s%c'"),
2376 i.op[op].regs->reg_name,
2377 i.tm.name,
2378 i.suffix);
2379 return 0;
2380 }
2381 /* Warn if the e prefix on a general reg is missing. */
2382 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2383 && (i.types[op] & Reg16) != 0
2384 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2385 {
2386 /* Prohibit these changes in the 64bit mode, since the
2387 lowering is more complicated. */
2388 if (flag_code == CODE_64BIT)
2389 {
2390 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2391 i.op[op].regs->reg_name,
2392 i.suffix);
2393 return 0;
2394 }
2395 #if REGISTER_WARNINGS
2396 else
2397 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2398 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
2399 i.op[op].regs->reg_name,
2400 i.suffix);
2401 #endif
2402 }
2403 /* Warn if the r prefix on a general reg is missing. */
2404 else if ((i.types[op] & Reg64) != 0
2405 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2406 {
2407 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2408 i.op[op].regs->reg_name,
2409 i.suffix);
2410 return 0;
2411 }
2412 return 1;
2413 }
2414
2415 static int
2416 check_qword_reg ()
2417 {
2418 int op;
2419
2420 for (op = i.operands; --op >= 0; )
2421 /* Reject eight bit registers, except where the template requires
2422 them. (eg. movzb) */
2423 if ((i.types[op] & Reg8) != 0
2424 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2425 {
2426 as_bad (_("`%%%s' not allowed with `%s%c'"),
2427 i.op[op].regs->reg_name,
2428 i.tm.name,
2429 i.suffix);
2430 return 0;
2431 }
2432 /* Warn if the e prefix on a general reg is missing. */
2433 else if (((i.types[op] & Reg16) != 0
2434 || (i.types[op] & Reg32) != 0)
2435 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2436 {
2437 /* Prohibit these changes in the 64bit mode, since the
2438 lowering is more complicated. */
2439 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2440 i.op[op].regs->reg_name,
2441 i.suffix);
2442 return 0;
2443 }
2444 return 1;
2445 }
2446
2447 static int
2448 check_word_reg ()
2449 {
2450 int op;
2451 for (op = i.operands; --op >= 0;)
2452 /* Reject eight bit registers, except where the template requires
2453 them. (eg. movzb) */
2454 if ((i.types[op] & Reg8) != 0
2455 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2456 {
2457 as_bad (_("`%%%s' not allowed with `%s%c'"),
2458 i.op[op].regs->reg_name,
2459 i.tm.name,
2460 i.suffix);
2461 return 0;
2462 }
2463 /* Warn if the e prefix on a general reg is present. */
2464 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2465 && (i.types[op] & Reg32) != 0
2466 && (i.tm.operand_types[op] & (Reg16 | Acc)) != 0)
2467 {
2468 /* Prohibit these changes in the 64bit mode, since the
2469 lowering is more complicated. */
2470 if (flag_code == CODE_64BIT)
2471 {
2472 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2473 i.op[op].regs->reg_name,
2474 i.suffix);
2475 return 0;
2476 }
2477 else
2478 #if REGISTER_WARNINGS
2479 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2480 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
2481 i.op[op].regs->reg_name,
2482 i.suffix);
2483 #endif
2484 }
2485 return 1;
2486 }
2487
2488 static int
2489 finalize_imm ()
2490 {
2491 unsigned int overlap0, overlap1, overlap2;
2492
2493 overlap0 = i.types[0] & i.tm.operand_types[0];
2494 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S))
2495 && overlap0 != Imm8 && overlap0 != Imm8S
2496 && overlap0 != Imm16 && overlap0 != Imm32S
2497 && overlap0 != Imm32 && overlap0 != Imm64)
2498 {
2499 if (i.suffix)
2500 {
2501 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX
2502 ? Imm8 | Imm8S
2503 : (i.suffix == WORD_MNEM_SUFFIX
2504 ? Imm16
2505 : (i.suffix == QWORD_MNEM_SUFFIX
2506 ? Imm64 | Imm32S
2507 : Imm32)));
2508 }
2509 else if (overlap0 == (Imm16 | Imm32S | Imm32)
2510 || overlap0 == (Imm16 | Imm32)
2511 || overlap0 == (Imm16 | Imm32S))
2512 {
2513 overlap0 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
2514 ? Imm16 : Imm32S);
2515 }
2516 if (overlap0 != Imm8 && overlap0 != Imm8S
2517 && overlap0 != Imm16 && overlap0 != Imm32S
2518 && overlap0 != Imm32 && overlap0 != Imm64)
2519 {
2520 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
2521 return 0;
2522 }
2523 }
2524 i.types[0] = overlap0;
2525
2526 overlap1 = i.types[1] & i.tm.operand_types[1];
2527 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32))
2528 && overlap1 != Imm8 && overlap1 != Imm8S
2529 && overlap1 != Imm16 && overlap1 != Imm32S
2530 && overlap1 != Imm32 && overlap1 != Imm64)
2531 {
2532 if (i.suffix)
2533 {
2534 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX
2535 ? Imm8 | Imm8S
2536 : (i.suffix == WORD_MNEM_SUFFIX
2537 ? Imm16
2538 : (i.suffix == QWORD_MNEM_SUFFIX
2539 ? Imm64 | Imm32S
2540 : Imm32)));
2541 }
2542 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
2543 || overlap1 == (Imm16 | Imm32)
2544 || overlap1 == (Imm16 | Imm32S))
2545 {
2546 overlap1 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
2547 ? Imm16 : Imm32S);
2548 }
2549 if (overlap1 != Imm8 && overlap1 != Imm8S
2550 && overlap1 != Imm16 && overlap1 != Imm32S
2551 && overlap1 != Imm32 && overlap1 != Imm64)
2552 {
2553 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
2554 return 0;
2555 }
2556 }
2557 i.types[1] = overlap1;
2558
2559 overlap2 = i.types[2] & i.tm.operand_types[2];
2560 assert ((overlap2 & Imm) == 0);
2561 i.types[2] = overlap2;
2562
2563 return 1;
2564 }
2565
2566 static int
2567 process_operands ()
2568 {
2569 /* Default segment register this instruction will use for memory
2570 accesses. 0 means unknown. This is only for optimizing out
2571 unnecessary segment overrides. */
2572 const seg_entry *default_seg = 0;
2573
2574 /* The imul $imm, %reg instruction is converted into
2575 imul $imm, %reg, %reg, and the clr %reg instruction
2576 is converted into xor %reg, %reg. */
2577 if (i.tm.opcode_modifier & regKludge)
2578 {
2579 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
2580 /* Pretend we saw the extra register operand. */
2581 assert (i.op[first_reg_op + 1].regs == 0);
2582 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
2583 i.types[first_reg_op + 1] = i.types[first_reg_op];
2584 i.reg_operands = 2;
2585 }
2586
2587 if (i.tm.opcode_modifier & ShortForm)
2588 {
2589 /* The register or float register operand is in operand 0 or 1. */
2590 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
2591 /* Register goes in low 3 bits of opcode. */
2592 i.tm.base_opcode |= i.op[op].regs->reg_num;
2593 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2594 i.rex |= REX_EXTZ;
2595 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2596 {
2597 /* Warn about some common errors, but press on regardless.
2598 The first case can be generated by gcc (<= 2.8.1). */
2599 if (i.operands == 2)
2600 {
2601 /* Reversed arguments on faddp, fsubp, etc. */
2602 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
2603 i.op[1].regs->reg_name,
2604 i.op[0].regs->reg_name);
2605 }
2606 else
2607 {
2608 /* Extraneous `l' suffix on fp insn. */
2609 as_warn (_("translating to `%s %%%s'"), i.tm.name,
2610 i.op[0].regs->reg_name);
2611 }
2612 }
2613 }
2614 else if (i.tm.opcode_modifier & Modrm)
2615 {
2616 /* The opcode is completed (modulo i.tm.extension_opcode which
2617 must be put into the modrm byte). Now, we make the modrm and
2618 index base bytes based on all the info we've collected. */
2619
2620 default_seg = build_modrm_byte ();
2621 }
2622 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
2623 {
2624 if (i.tm.base_opcode == POP_SEG_SHORT
2625 && i.op[0].regs->reg_num == 1)
2626 {
2627 as_bad (_("you can't `pop %%cs'"));
2628 return 0;
2629 }
2630 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
2631 if ((i.op[0].regs->reg_flags & RegRex) != 0)
2632 i.rex |= REX_EXTZ;
2633 }
2634 else if ((i.tm.base_opcode & ~(D | W)) == MOV_AX_DISP32)
2635 {
2636 default_seg = &ds;
2637 }
2638 else if ((i.tm.opcode_modifier & IsString) != 0)
2639 {
2640 /* For the string instructions that allow a segment override
2641 on one of their operands, the default segment is ds. */
2642 default_seg = &ds;
2643 }
2644
2645 if (i.tm.base_opcode == 0x8d /* lea */ && i.seg[0] && !quiet_warnings)
2646 as_warn (_("segment override on `lea' is ineffectual"));
2647
2648 /* If a segment was explicitly specified, and the specified segment
2649 is not the default, use an opcode prefix to select it. If we
2650 never figured out what the default segment is, then default_seg
2651 will be zero at this point, and the specified segment prefix will
2652 always be used. */
2653 if ((i.seg[0]) && (i.seg[0] != default_seg))
2654 {
2655 if (!add_prefix (i.seg[0]->seg_prefix))
2656 return 0;
2657 }
2658 return 1;
2659 }
2660
2661 static const seg_entry *
2662 build_modrm_byte ()
2663 {
2664 const seg_entry *default_seg = 0;
2665
2666 /* i.reg_operands MUST be the number of real register operands;
2667 implicit registers do not count. */
2668 if (i.reg_operands == 2)
2669 {
2670 unsigned int source, dest;
2671 source = ((i.types[0]
2672 & (Reg | RegMMX | RegXMM
2673 | SReg2 | SReg3
2674 | Control | Debug | Test))
2675 ? 0 : 1);
2676 dest = source + 1;
2677
2678 i.rm.mode = 3;
2679 /* One of the register operands will be encoded in the i.tm.reg
2680 field, the other in the combined i.tm.mode and i.tm.regmem
2681 fields. If no form of this instruction supports a memory
2682 destination operand, then we assume the source operand may
2683 sometimes be a memory operand and so we need to store the
2684 destination in the i.rm.reg field. */
2685 if ((i.tm.operand_types[dest] & AnyMem) == 0)
2686 {
2687 i.rm.reg = i.op[dest].regs->reg_num;
2688 i.rm.regmem = i.op[source].regs->reg_num;
2689 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
2690 i.rex |= REX_EXTX;
2691 if ((i.op[source].regs->reg_flags & RegRex) != 0)
2692 i.rex |= REX_EXTZ;
2693 }
2694 else
2695 {
2696 i.rm.reg = i.op[source].regs->reg_num;
2697 i.rm.regmem = i.op[dest].regs->reg_num;
2698 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
2699 i.rex |= REX_EXTZ;
2700 if ((i.op[source].regs->reg_flags & RegRex) != 0)
2701 i.rex |= REX_EXTX;
2702 }
2703 }
2704 else
2705 { /* If it's not 2 reg operands... */
2706 if (i.mem_operands)
2707 {
2708 unsigned int fake_zero_displacement = 0;
2709 unsigned int op = ((i.types[0] & AnyMem)
2710 ? 0
2711 : (i.types[1] & AnyMem) ? 1 : 2);
2712
2713 default_seg = &ds;
2714
2715 if (i.base_reg == 0)
2716 {
2717 i.rm.mode = 0;
2718 if (!i.disp_operands)
2719 fake_zero_displacement = 1;
2720 if (i.index_reg == 0)
2721 {
2722 /* Operand is just <disp> */
2723 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)
2724 && (flag_code != CODE_64BIT))
2725 {
2726 i.rm.regmem = NO_BASE_REGISTER_16;
2727 i.types[op] &= ~Disp;
2728 i.types[op] |= Disp16;
2729 }
2730 else if (flag_code != CODE_64BIT
2731 || (i.prefix[ADDR_PREFIX] != 0))
2732 {
2733 i.rm.regmem = NO_BASE_REGISTER;
2734 i.types[op] &= ~Disp;
2735 i.types[op] |= Disp32;
2736 }
2737 else
2738 {
2739 /* 64bit mode overwrites the 32bit absolute
2740 addressing by RIP relative addressing and
2741 absolute addressing is encoded by one of the
2742 redundant SIB forms. */
2743 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2744 i.sib.base = NO_BASE_REGISTER;
2745 i.sib.index = NO_INDEX_REGISTER;
2746 i.types[op] &= ~Disp;
2747 i.types[op] |= Disp32S;
2748 }
2749 }
2750 else /* !i.base_reg && i.index_reg */
2751 {
2752 i.sib.index = i.index_reg->reg_num;
2753 i.sib.base = NO_BASE_REGISTER;
2754 i.sib.scale = i.log2_scale_factor;
2755 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2756 i.types[op] &= ~Disp;
2757 if (flag_code != CODE_64BIT)
2758 i.types[op] |= Disp32; /* Must be 32 bit */
2759 else
2760 i.types[op] |= Disp32S;
2761 if ((i.index_reg->reg_flags & RegRex) != 0)
2762 i.rex |= REX_EXTY;
2763 }
2764 }
2765 /* RIP addressing for 64bit mode. */
2766 else if (i.base_reg->reg_type == BaseIndex)
2767 {
2768 i.rm.regmem = NO_BASE_REGISTER;
2769 i.types[op] &= ~Disp;
2770 i.types[op] |= Disp32S;
2771 i.flags[op] = Operand_PCrel;
2772 }
2773 else if (i.base_reg->reg_type & Reg16)
2774 {
2775 switch (i.base_reg->reg_num)
2776 {
2777 case 3: /* (%bx) */
2778 if (i.index_reg == 0)
2779 i.rm.regmem = 7;
2780 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
2781 i.rm.regmem = i.index_reg->reg_num - 6;
2782 break;
2783 case 5: /* (%bp) */
2784 default_seg = &ss;
2785 if (i.index_reg == 0)
2786 {
2787 i.rm.regmem = 6;
2788 if ((i.types[op] & Disp) == 0)
2789 {
2790 /* fake (%bp) into 0(%bp) */
2791 i.types[op] |= Disp8;
2792 fake_zero_displacement = 1;
2793 }
2794 }
2795 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
2796 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
2797 break;
2798 default: /* (%si) -> 4 or (%di) -> 5 */
2799 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
2800 }
2801 i.rm.mode = mode_from_disp_size (i.types[op]);
2802 }
2803 else /* i.base_reg and 32/64 bit mode */
2804 {
2805 if (flag_code == CODE_64BIT
2806 && (i.types[op] & Disp))
2807 {
2808 if (i.types[op] & Disp8)
2809 i.types[op] = Disp8 | Disp32S;
2810 else
2811 i.types[op] = Disp32S;
2812 }
2813 i.rm.regmem = i.base_reg->reg_num;
2814 if ((i.base_reg->reg_flags & RegRex) != 0)
2815 i.rex |= REX_EXTZ;
2816 i.sib.base = i.base_reg->reg_num;
2817 /* x86-64 ignores REX prefix bit here to avoid decoder
2818 complications. */
2819 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
2820 {
2821 default_seg = &ss;
2822 if (i.disp_operands == 0)
2823 {
2824 fake_zero_displacement = 1;
2825 i.types[op] |= Disp8;
2826 }
2827 }
2828 else if (i.base_reg->reg_num == ESP_REG_NUM)
2829 {
2830 default_seg = &ss;
2831 }
2832 i.sib.scale = i.log2_scale_factor;
2833 if (i.index_reg == 0)
2834 {
2835 /* <disp>(%esp) becomes two byte modrm with no index
2836 register. We've already stored the code for esp
2837 in i.rm.regmem ie. ESCAPE_TO_TWO_BYTE_ADDRESSING.
2838 Any base register besides %esp will not use the
2839 extra modrm byte. */
2840 i.sib.index = NO_INDEX_REGISTER;
2841 #if !SCALE1_WHEN_NO_INDEX
2842 /* Another case where we force the second modrm byte. */
2843 if (i.log2_scale_factor)
2844 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2845 #endif
2846 }
2847 else
2848 {
2849 i.sib.index = i.index_reg->reg_num;
2850 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2851 if ((i.index_reg->reg_flags & RegRex) != 0)
2852 i.rex |= REX_EXTY;
2853 }
2854 i.rm.mode = mode_from_disp_size (i.types[op]);
2855 }
2856
2857 if (fake_zero_displacement)
2858 {
2859 /* Fakes a zero displacement assuming that i.types[op]
2860 holds the correct displacement size. */
2861 expressionS *exp;
2862
2863 assert (i.op[op].disps == 0);
2864 exp = &disp_expressions[i.disp_operands++];
2865 i.op[op].disps = exp;
2866 exp->X_op = O_constant;
2867 exp->X_add_number = 0;
2868 exp->X_add_symbol = (symbolS *) 0;
2869 exp->X_op_symbol = (symbolS *) 0;
2870 }
2871 }
2872
2873 /* Fill in i.rm.reg or i.rm.regmem field with register operand
2874 (if any) based on i.tm.extension_opcode. Again, we must be
2875 careful to make sure that segment/control/debug/test/MMX
2876 registers are coded into the i.rm.reg field. */
2877 if (i.reg_operands)
2878 {
2879 unsigned int op =
2880 ((i.types[0]
2881 & (Reg | RegMMX | RegXMM
2882 | SReg2 | SReg3
2883 | Control | Debug | Test))
2884 ? 0
2885 : ((i.types[1]
2886 & (Reg | RegMMX | RegXMM
2887 | SReg2 | SReg3
2888 | Control | Debug | Test))
2889 ? 1
2890 : 2));
2891 /* If there is an extension opcode to put here, the register
2892 number must be put into the regmem field. */
2893 if (i.tm.extension_opcode != None)
2894 {
2895 i.rm.regmem = i.op[op].regs->reg_num;
2896 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2897 i.rex |= REX_EXTZ;
2898 }
2899 else
2900 {
2901 i.rm.reg = i.op[op].regs->reg_num;
2902 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2903 i.rex |= REX_EXTX;
2904 }
2905
2906 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2 we
2907 must set it to 3 to indicate this is a register operand
2908 in the regmem field. */
2909 if (!i.mem_operands)
2910 i.rm.mode = 3;
2911 }
2912
2913 /* Fill in i.rm.reg field with extension opcode (if any). */
2914 if (i.tm.extension_opcode != None)
2915 i.rm.reg = i.tm.extension_opcode;
2916 }
2917 return default_seg;
2918 }
2919
2920 static void
2921 output_branch ()
2922 {
2923 char *p;
2924 int code16;
2925 int prefix;
2926 relax_substateT subtype;
2927 symbolS *sym;
2928 offsetT off;
2929
2930 code16 = 0;
2931 if (flag_code == CODE_16BIT)
2932 code16 = CODE16;
2933
2934 prefix = 0;
2935 if (i.prefix[DATA_PREFIX] != 0)
2936 {
2937 prefix = 1;
2938 i.prefixes -= 1;
2939 code16 ^= CODE16;
2940 }
2941 /* Pentium4 branch hints. */
2942 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
2943 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
2944 {
2945 prefix++;
2946 i.prefixes--;
2947 }
2948 if (i.prefix[REX_PREFIX] != 0)
2949 {
2950 prefix++;
2951 i.prefixes--;
2952 }
2953
2954 if (i.prefixes != 0 && !intel_syntax)
2955 as_warn (_("skipping prefixes on this instruction"));
2956
2957 /* It's always a symbol; End frag & setup for relax.
2958 Make sure there is enough room in this frag for the largest
2959 instruction we may generate in md_convert_frag. This is 2
2960 bytes for the opcode and room for the prefix and largest
2961 displacement. */
2962 frag_grow (prefix + 2 + 4);
2963 /* Prefix and 1 opcode byte go in fr_fix. */
2964 p = frag_more (prefix + 1);
2965 if (i.prefix[DATA_PREFIX] != 0)
2966 *p++ = DATA_PREFIX_OPCODE;
2967 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
2968 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
2969 *p++ = i.prefix[SEG_PREFIX];
2970 if (i.prefix[REX_PREFIX] != 0)
2971 *p++ = i.prefix[REX_PREFIX];
2972 *p = i.tm.base_opcode;
2973
2974 if ((unsigned char) *p == JUMP_PC_RELATIVE)
2975 subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
2976 else if ((cpu_arch_flags & Cpu386) != 0)
2977 subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
2978 else
2979 subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
2980 subtype |= code16;
2981
2982 sym = i.op[0].disps->X_add_symbol;
2983 off = i.op[0].disps->X_add_number;
2984
2985 if (i.op[0].disps->X_op != O_constant
2986 && i.op[0].disps->X_op != O_symbol)
2987 {
2988 /* Handle complex expressions. */
2989 sym = make_expr_symbol (i.op[0].disps);
2990 off = 0;
2991 }
2992
2993 /* 1 possible extra opcode + 4 byte displacement go in var part.
2994 Pass reloc in fr_var. */
2995 frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
2996 }
2997
2998 static void
2999 output_jump ()
3000 {
3001 char *p;
3002 int size;
3003 fixS *fixP;
3004
3005 if (i.tm.opcode_modifier & JumpByte)
3006 {
3007 /* This is a loop or jecxz type instruction. */
3008 size = 1;
3009 if (i.prefix[ADDR_PREFIX] != 0)
3010 {
3011 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
3012 i.prefixes -= 1;
3013 }
3014 /* Pentium4 branch hints. */
3015 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3016 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3017 {
3018 FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
3019 i.prefixes--;
3020 }
3021 }
3022 else
3023 {
3024 int code16;
3025
3026 code16 = 0;
3027 if (flag_code == CODE_16BIT)
3028 code16 = CODE16;
3029
3030 if (i.prefix[DATA_PREFIX] != 0)
3031 {
3032 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
3033 i.prefixes -= 1;
3034 code16 ^= CODE16;
3035 }
3036
3037 size = 4;
3038 if (code16)
3039 size = 2;
3040 }
3041
3042 if (i.prefix[REX_PREFIX] != 0)
3043 {
3044 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
3045 i.prefixes -= 1;
3046 }
3047
3048 if (i.prefixes != 0 && !intel_syntax)
3049 as_warn (_("skipping prefixes on this instruction"));
3050
3051 p = frag_more (1 + size);
3052 *p++ = i.tm.base_opcode;
3053
3054 fixP = fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3055 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
3056
3057 /* All jumps handled here are signed, but don't use a signed limit
3058 check for 32 and 16 bit jumps as we want to allow wrap around at
3059 4G and 64k respectively. */
3060 if (size == 1)
3061 fixP->fx_signed = 1;
3062 }
3063
3064 static void
3065 output_interseg_jump ()
3066 {
3067 char *p;
3068 int size;
3069 int prefix;
3070 int code16;
3071
3072 code16 = 0;
3073 if (flag_code == CODE_16BIT)
3074 code16 = CODE16;
3075
3076 prefix = 0;
3077 if (i.prefix[DATA_PREFIX] != 0)
3078 {
3079 prefix = 1;
3080 i.prefixes -= 1;
3081 code16 ^= CODE16;
3082 }
3083 if (i.prefix[REX_PREFIX] != 0)
3084 {
3085 prefix++;
3086 i.prefixes -= 1;
3087 }
3088
3089 size = 4;
3090 if (code16)
3091 size = 2;
3092
3093 if (i.prefixes != 0 && !intel_syntax)
3094 as_warn (_("skipping prefixes on this instruction"));
3095
3096 /* 1 opcode; 2 segment; offset */
3097 p = frag_more (prefix + 1 + 2 + size);
3098
3099 if (i.prefix[DATA_PREFIX] != 0)
3100 *p++ = DATA_PREFIX_OPCODE;
3101
3102 if (i.prefix[REX_PREFIX] != 0)
3103 *p++ = i.prefix[REX_PREFIX];
3104
3105 *p++ = i.tm.base_opcode;
3106 if (i.op[1].imms->X_op == O_constant)
3107 {
3108 offsetT n = i.op[1].imms->X_add_number;
3109
3110 if (size == 2
3111 && !fits_in_unsigned_word (n)
3112 && !fits_in_signed_word (n))
3113 {
3114 as_bad (_("16-bit jump out of range"));
3115 return;
3116 }
3117 md_number_to_chars (p, n, size);
3118 }
3119 else
3120 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3121 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
3122 if (i.op[0].imms->X_op != O_constant)
3123 as_bad (_("can't handle non absolute segment in `%s'"),
3124 i.tm.name);
3125 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
3126 }
3127
3128
3129 static void
3130 output_insn ()
3131 {
3132 fragS *insn_start_frag;
3133 offsetT insn_start_off;
3134
3135 /* Tie dwarf2 debug info to the address at the start of the insn.
3136 We can't do this after the insn has been output as the current
3137 frag may have been closed off. eg. by frag_var. */
3138 dwarf2_emit_insn (0);
3139
3140 insn_start_frag = frag_now;
3141 insn_start_off = frag_now_fix ();
3142
3143 /* Output jumps. */
3144 if (i.tm.opcode_modifier & Jump)
3145 output_branch ();
3146 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
3147 output_jump ();
3148 else if (i.tm.opcode_modifier & JumpInterSegment)
3149 output_interseg_jump ();
3150 else
3151 {
3152 /* Output normal instructions here. */
3153 char *p;
3154 unsigned char *q;
3155
3156 /* All opcodes on i386 have either 1 or 2 bytes. We may use third
3157 byte for the SSE instructions to specify a prefix they require. */
3158 if (i.tm.base_opcode & 0xff0000)
3159 add_prefix ((i.tm.base_opcode >> 16) & 0xff);
3160
3161 /* The prefix bytes. */
3162 for (q = i.prefix;
3163 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
3164 q++)
3165 {
3166 if (*q)
3167 {
3168 p = frag_more (1);
3169 md_number_to_chars (p, (valueT) *q, 1);
3170 }
3171 }
3172
3173 /* Now the opcode; be careful about word order here! */
3174 if (fits_in_unsigned_byte (i.tm.base_opcode))
3175 {
3176 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
3177 }
3178 else
3179 {
3180 p = frag_more (2);
3181 /* Put out high byte first: can't use md_number_to_chars! */
3182 *p++ = (i.tm.base_opcode >> 8) & 0xff;
3183 *p = i.tm.base_opcode & 0xff;
3184 }
3185
3186 /* Now the modrm byte and sib byte (if present). */
3187 if (i.tm.opcode_modifier & Modrm)
3188 {
3189 p = frag_more (1);
3190 md_number_to_chars (p,
3191 (valueT) (i.rm.regmem << 0
3192 | i.rm.reg << 3
3193 | i.rm.mode << 6),
3194 1);
3195 /* If i.rm.regmem == ESP (4)
3196 && i.rm.mode != (Register mode)
3197 && not 16 bit
3198 ==> need second modrm byte. */
3199 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
3200 && i.rm.mode != 3
3201 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
3202 {
3203 p = frag_more (1);
3204 md_number_to_chars (p,
3205 (valueT) (i.sib.base << 0
3206 | i.sib.index << 3
3207 | i.sib.scale << 6),
3208 1);
3209 }
3210 }
3211
3212 if (i.disp_operands)
3213 output_disp (insn_start_frag, insn_start_off);
3214
3215 if (i.imm_operands)
3216 output_imm (insn_start_frag, insn_start_off);
3217 }
3218
3219 #ifdef DEBUG386
3220 if (flag_debug)
3221 {
3222 pi (line, &i);
3223 }
3224 #endif /* DEBUG386 */
3225 }
3226
3227 static void
3228 output_disp (insn_start_frag, insn_start_off)
3229 fragS *insn_start_frag;
3230 offsetT insn_start_off;
3231 {
3232 char *p;
3233 unsigned int n;
3234
3235 for (n = 0; n < i.operands; n++)
3236 {
3237 if (i.types[n] & Disp)
3238 {
3239 if (i.op[n].disps->X_op == O_constant)
3240 {
3241 int size;
3242 offsetT val;
3243
3244 size = 4;
3245 if (i.types[n] & (Disp8 | Disp16 | Disp64))
3246 {
3247 size = 2;
3248 if (i.types[n] & Disp8)
3249 size = 1;
3250 if (i.types[n] & Disp64)
3251 size = 8;
3252 }
3253 val = offset_in_range (i.op[n].disps->X_add_number,
3254 size);
3255 p = frag_more (size);
3256 md_number_to_chars (p, val, size);
3257 }
3258 else
3259 {
3260 RELOC_ENUM reloc_type;
3261 int size = 4;
3262 int sign = 0;
3263 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
3264
3265 /* The PC relative address is computed relative
3266 to the instruction boundary, so in case immediate
3267 fields follows, we need to adjust the value. */
3268 if (pcrel && i.imm_operands)
3269 {
3270 int imm_size = 4;
3271 unsigned int n1;
3272
3273 for (n1 = 0; n1 < i.operands; n1++)
3274 if (i.types[n1] & Imm)
3275 {
3276 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
3277 {
3278 imm_size = 2;
3279 if (i.types[n1] & (Imm8 | Imm8S))
3280 imm_size = 1;
3281 if (i.types[n1] & Imm64)
3282 imm_size = 8;
3283 }
3284 break;
3285 }
3286 /* We should find the immediate. */
3287 if (n1 == i.operands)
3288 abort ();
3289 i.op[n].disps->X_add_number -= imm_size;
3290 }
3291
3292 if (i.types[n] & Disp32S)
3293 sign = 1;
3294
3295 if (i.types[n] & (Disp16 | Disp64))
3296 {
3297 size = 2;
3298 if (i.types[n] & Disp64)
3299 size = 8;
3300 }
3301
3302 p = frag_more (size);
3303 reloc_type = reloc (size, pcrel, sign, i.reloc[n]);
3304 #ifdef BFD_ASSEMBLER
3305 if (reloc_type == BFD_RELOC_32
3306 && GOT_symbol
3307 && GOT_symbol == i.op[n].disps->X_add_symbol
3308 && (i.op[n].disps->X_op == O_symbol
3309 || (i.op[n].disps->X_op == O_add
3310 && ((symbol_get_value_expression
3311 (i.op[n].disps->X_op_symbol)->X_op)
3312 == O_subtract))))
3313 {
3314 offsetT add;
3315
3316 if (insn_start_frag == frag_now)
3317 add = (p - frag_now->fr_literal) - insn_start_off;
3318 else
3319 {
3320 fragS *fr;
3321
3322 add = insn_start_frag->fr_fix - insn_start_off;
3323 for (fr = insn_start_frag->fr_next;
3324 fr && fr != frag_now; fr = fr->fr_next)
3325 add += fr->fr_fix;
3326 add += p - frag_now->fr_literal;
3327 }
3328
3329 /* We don't support dynamic linking on x86-64 yet. */
3330 if (flag_code == CODE_64BIT)
3331 abort ();
3332 reloc_type = BFD_RELOC_386_GOTPC;
3333 i.op[n].disps->X_add_number += add;
3334 }
3335 #endif
3336 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3337 i.op[n].disps, pcrel, reloc_type);
3338 }
3339 }
3340 }
3341 }
3342
3343 static void
3344 output_imm (insn_start_frag, insn_start_off)
3345 fragS *insn_start_frag;
3346 offsetT insn_start_off;
3347 {
3348 char *p;
3349 unsigned int n;
3350
3351 for (n = 0; n < i.operands; n++)
3352 {
3353 if (i.types[n] & Imm)
3354 {
3355 if (i.op[n].imms->X_op == O_constant)
3356 {
3357 int size;
3358 offsetT val;
3359
3360 size = 4;
3361 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3362 {
3363 size = 2;
3364 if (i.types[n] & (Imm8 | Imm8S))
3365 size = 1;
3366 else if (i.types[n] & Imm64)
3367 size = 8;
3368 }
3369 val = offset_in_range (i.op[n].imms->X_add_number,
3370 size);
3371 p = frag_more (size);
3372 md_number_to_chars (p, val, size);
3373 }
3374 else
3375 {
3376 /* Not absolute_section.
3377 Need a 32-bit fixup (don't support 8bit
3378 non-absolute imms). Try to support other
3379 sizes ... */
3380 RELOC_ENUM reloc_type;
3381 int size = 4;
3382 int sign = 0;
3383
3384 if ((i.types[n] & (Imm32S))
3385 && i.suffix == QWORD_MNEM_SUFFIX)
3386 sign = 1;
3387 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3388 {
3389 size = 2;
3390 if (i.types[n] & (Imm8 | Imm8S))
3391 size = 1;
3392 if (i.types[n] & Imm64)
3393 size = 8;
3394 }
3395
3396 p = frag_more (size);
3397 reloc_type = reloc (size, 0, sign, i.reloc[n]);
3398 #ifdef BFD_ASSEMBLER
3399 /* This is tough to explain. We end up with this one if we
3400 * have operands that look like
3401 * "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal here is to
3402 * obtain the absolute address of the GOT, and it is strongly
3403 * preferable from a performance point of view to avoid using
3404 * a runtime relocation for this. The actual sequence of
3405 * instructions often look something like:
3406 *
3407 * call .L66
3408 * .L66:
3409 * popl %ebx
3410 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
3411 *
3412 * The call and pop essentially return the absolute address
3413 * of the label .L66 and store it in %ebx. The linker itself
3414 * will ultimately change the first operand of the addl so
3415 * that %ebx points to the GOT, but to keep things simple, the
3416 * .o file must have this operand set so that it generates not
3417 * the absolute address of .L66, but the absolute address of
3418 * itself. This allows the linker itself simply treat a GOTPC
3419 * relocation as asking for a pcrel offset to the GOT to be
3420 * added in, and the addend of the relocation is stored in the
3421 * operand field for the instruction itself.
3422 *
3423 * Our job here is to fix the operand so that it would add
3424 * the correct offset so that %ebx would point to itself. The
3425 * thing that is tricky is that .-.L66 will point to the
3426 * beginning of the instruction, so we need to further modify
3427 * the operand so that it will point to itself. There are
3428 * other cases where you have something like:
3429 *
3430 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
3431 *
3432 * and here no correction would be required. Internally in
3433 * the assembler we treat operands of this form as not being
3434 * pcrel since the '.' is explicitly mentioned, and I wonder
3435 * whether it would simplify matters to do it this way. Who
3436 * knows. In earlier versions of the PIC patches, the
3437 * pcrel_adjust field was used to store the correction, but
3438 * since the expression is not pcrel, I felt it would be
3439 * confusing to do it this way. */
3440
3441 if (reloc_type == BFD_RELOC_32
3442 && GOT_symbol
3443 && GOT_symbol == i.op[n].imms->X_add_symbol
3444 && (i.op[n].imms->X_op == O_symbol
3445 || (i.op[n].imms->X_op == O_add
3446 && ((symbol_get_value_expression
3447 (i.op[n].imms->X_op_symbol)->X_op)
3448 == O_subtract))))
3449 {
3450 offsetT add;
3451
3452 if (insn_start_frag == frag_now)
3453 add = (p - frag_now->fr_literal) - insn_start_off;
3454 else
3455 {
3456 fragS *fr;
3457
3458 add = insn_start_frag->fr_fix - insn_start_off;
3459 for (fr = insn_start_frag->fr_next;
3460 fr && fr != frag_now; fr = fr->fr_next)
3461 add += fr->fr_fix;
3462 add += p - frag_now->fr_literal;
3463 }
3464
3465 /* We don't support dynamic linking on x86-64 yet. */
3466 if (flag_code == CODE_64BIT)
3467 abort ();
3468 reloc_type = BFD_RELOC_386_GOTPC;
3469 i.op[n].imms->X_add_number += add;
3470 }
3471 #endif
3472 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3473 i.op[n].imms, 0, reloc_type);
3474 }
3475 }
3476 }
3477 }
3478 \f
3479 #ifndef LEX_AT
3480 static char *lex_got PARAMS ((RELOC_ENUM *, int *));
3481
3482 /* Parse operands of the form
3483 <symbol>@GOTOFF+<nnn>
3484 and similar .plt or .got references.
3485
3486 If we find one, set up the correct relocation in RELOC and copy the
3487 input string, minus the `@GOTOFF' into a malloc'd buffer for
3488 parsing by the calling routine. Return this buffer, and if ADJUST
3489 is non-null set it to the length of the string we removed from the
3490 input line. Otherwise return NULL. */
3491 static char *
3492 lex_got (reloc, adjust)
3493 RELOC_ENUM *reloc;
3494 int *adjust;
3495 {
3496 static const char * const mode_name[NUM_FLAG_CODE] = { "32", "16", "64" };
3497 static const struct {
3498 const char *str;
3499 const RELOC_ENUM rel[NUM_FLAG_CODE];
3500 } gotrel[] = {
3501 { "PLT", { BFD_RELOC_386_PLT32, 0, BFD_RELOC_X86_64_PLT32 } },
3502 { "GOTOFF", { BFD_RELOC_386_GOTOFF, 0, 0 } },
3503 { "GOTPCREL", { 0, 0, BFD_RELOC_X86_64_GOTPCREL } },
3504 { "TLSGD", { BFD_RELOC_386_TLS_GD, 0, BFD_RELOC_X86_64_TLSGD } },
3505 { "TLSLDM", { BFD_RELOC_386_TLS_LDM, 0, 0 } },
3506 { "TLSLD", { 0, 0, BFD_RELOC_X86_64_TLSLD } },
3507 { "GOTTPOFF", { BFD_RELOC_386_TLS_IE_32, 0, BFD_RELOC_X86_64_GOTTPOFF } },
3508 { "TPOFF", { BFD_RELOC_386_TLS_LE_32, 0, BFD_RELOC_X86_64_TPOFF32 } },
3509 { "NTPOFF", { BFD_RELOC_386_TLS_LE, 0, 0 } },
3510 { "DTPOFF", { BFD_RELOC_386_TLS_LDO_32, 0, BFD_RELOC_X86_64_DTPOFF32 } },
3511 { "GOTNTPOFF",{ BFD_RELOC_386_TLS_GOTIE, 0, 0 } },
3512 { "INDNTPOFF",{ BFD_RELOC_386_TLS_IE, 0, 0 } },
3513 { "GOT", { BFD_RELOC_386_GOT32, 0, BFD_RELOC_X86_64_GOT32 } }
3514 };
3515 char *cp;
3516 unsigned int j;
3517
3518 for (cp = input_line_pointer; *cp != '@'; cp++)
3519 if (is_end_of_line[(unsigned char) *cp])
3520 return NULL;
3521
3522 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
3523 {
3524 int len;
3525
3526 len = strlen (gotrel[j].str);
3527 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
3528 {
3529 if (gotrel[j].rel[(unsigned int) flag_code] != 0)
3530 {
3531 int first, second;
3532 char *tmpbuf, *past_reloc;
3533
3534 *reloc = gotrel[j].rel[(unsigned int) flag_code];
3535 if (adjust)
3536 *adjust = len;
3537
3538 if (GOT_symbol == NULL)
3539 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3540
3541 /* Replace the relocation token with ' ', so that
3542 errors like foo@GOTOFF1 will be detected. */
3543
3544 /* The length of the first part of our input line. */
3545 first = cp - input_line_pointer;
3546
3547 /* The second part goes from after the reloc token until
3548 (and including) an end_of_line char. Don't use strlen
3549 here as the end_of_line char may not be a NUL. */
3550 past_reloc = cp + 1 + len;
3551 for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
3552 ;
3553 second = cp - past_reloc;
3554
3555 /* Allocate and copy string. The trailing NUL shouldn't
3556 be necessary, but be safe. */
3557 tmpbuf = xmalloc (first + second + 2);
3558 memcpy (tmpbuf, input_line_pointer, first);
3559 tmpbuf[first] = ' ';
3560 memcpy (tmpbuf + first + 1, past_reloc, second);
3561 tmpbuf[first + second + 1] = '\0';
3562 return tmpbuf;
3563 }
3564
3565 as_bad (_("@%s reloc is not supported in %s bit mode"),
3566 gotrel[j].str, mode_name[(unsigned int) flag_code]);
3567 return NULL;
3568 }
3569 }
3570
3571 /* Might be a symbol version string. Don't as_bad here. */
3572 return NULL;
3573 }
3574
3575 /* x86_cons_fix_new is called via the expression parsing code when a
3576 reloc is needed. We use this hook to get the correct .got reloc. */
3577 static RELOC_ENUM got_reloc = NO_RELOC;
3578
3579 void
3580 x86_cons_fix_new (frag, off, len, exp)
3581 fragS *frag;
3582 unsigned int off;
3583 unsigned int len;
3584 expressionS *exp;
3585 {
3586 RELOC_ENUM r = reloc (len, 0, 0, got_reloc);
3587 got_reloc = NO_RELOC;
3588 fix_new_exp (frag, off, len, exp, 0, r);
3589 }
3590
3591 void
3592 x86_cons (exp, size)
3593 expressionS *exp;
3594 int size;
3595 {
3596 if (size == 4)
3597 {
3598 /* Handle @GOTOFF and the like in an expression. */
3599 char *save;
3600 char *gotfree_input_line;
3601 int adjust;
3602
3603 save = input_line_pointer;
3604 gotfree_input_line = lex_got (&got_reloc, &adjust);
3605 if (gotfree_input_line)
3606 input_line_pointer = gotfree_input_line;
3607
3608 expression (exp);
3609
3610 if (gotfree_input_line)
3611 {
3612 /* expression () has merrily parsed up to the end of line,
3613 or a comma - in the wrong buffer. Transfer how far
3614 input_line_pointer has moved to the right buffer. */
3615 input_line_pointer = (save
3616 + (input_line_pointer - gotfree_input_line)
3617 + adjust);
3618 free (gotfree_input_line);
3619 }
3620 }
3621 else
3622 expression (exp);
3623 }
3624 #endif
3625
3626 static int i386_immediate PARAMS ((char *));
3627
3628 static int
3629 i386_immediate (imm_start)
3630 char *imm_start;
3631 {
3632 char *save_input_line_pointer;
3633 #ifndef LEX_AT
3634 char *gotfree_input_line;
3635 #endif
3636 segT exp_seg = 0;
3637 expressionS *exp;
3638
3639 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
3640 {
3641 as_bad (_("only 1 or 2 immediate operands are allowed"));
3642 return 0;
3643 }
3644
3645 exp = &im_expressions[i.imm_operands++];
3646 i.op[this_operand].imms = exp;
3647
3648 if (is_space_char (*imm_start))
3649 ++imm_start;
3650
3651 save_input_line_pointer = input_line_pointer;
3652 input_line_pointer = imm_start;
3653
3654 #ifndef LEX_AT
3655 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3656 if (gotfree_input_line)
3657 input_line_pointer = gotfree_input_line;
3658 #endif
3659
3660 exp_seg = expression (exp);
3661
3662 SKIP_WHITESPACE ();
3663 if (*input_line_pointer)
3664 as_bad (_("junk `%s' after expression"), input_line_pointer);
3665
3666 input_line_pointer = save_input_line_pointer;
3667 #ifndef LEX_AT
3668 if (gotfree_input_line)
3669 free (gotfree_input_line);
3670 #endif
3671
3672 if (exp->X_op == O_absent || exp->X_op == O_big)
3673 {
3674 /* Missing or bad expr becomes absolute 0. */
3675 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
3676 imm_start);
3677 exp->X_op = O_constant;
3678 exp->X_add_number = 0;
3679 exp->X_add_symbol = (symbolS *) 0;
3680 exp->X_op_symbol = (symbolS *) 0;
3681 }
3682 else if (exp->X_op == O_constant)
3683 {
3684 /* Size it properly later. */
3685 i.types[this_operand] |= Imm64;
3686 /* If BFD64, sign extend val. */
3687 if (!use_rela_relocations)
3688 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
3689 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
3690 }
3691 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3692 else if (1
3693 #ifdef BFD_ASSEMBLER
3694 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3695 #endif
3696 && exp_seg != absolute_section
3697 && exp_seg != text_section
3698 && exp_seg != data_section
3699 && exp_seg != bss_section
3700 && exp_seg != undefined_section
3701 #ifdef BFD_ASSEMBLER
3702 && !bfd_is_com_section (exp_seg)
3703 #endif
3704 )
3705 {
3706 #ifdef BFD_ASSEMBLER
3707 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3708 #else
3709 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3710 #endif
3711 return 0;
3712 }
3713 #endif
3714 else
3715 {
3716 /* This is an address. The size of the address will be
3717 determined later, depending on destination register,
3718 suffix, or the default for the section. */
3719 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
3720 }
3721
3722 return 1;
3723 }
3724
3725 static char *i386_scale PARAMS ((char *));
3726
3727 static char *
3728 i386_scale (scale)
3729 char *scale;
3730 {
3731 offsetT val;
3732 char *save = input_line_pointer;
3733
3734 input_line_pointer = scale;
3735 val = get_absolute_expression ();
3736
3737 switch (val)
3738 {
3739 case 0:
3740 case 1:
3741 i.log2_scale_factor = 0;
3742 break;
3743 case 2:
3744 i.log2_scale_factor = 1;
3745 break;
3746 case 4:
3747 i.log2_scale_factor = 2;
3748 break;
3749 case 8:
3750 i.log2_scale_factor = 3;
3751 break;
3752 default:
3753 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
3754 scale);
3755 input_line_pointer = save;
3756 return NULL;
3757 }
3758 if (i.log2_scale_factor != 0 && i.index_reg == 0)
3759 {
3760 as_warn (_("scale factor of %d without an index register"),
3761 1 << i.log2_scale_factor);
3762 #if SCALE1_WHEN_NO_INDEX
3763 i.log2_scale_factor = 0;
3764 #endif
3765 }
3766 scale = input_line_pointer;
3767 input_line_pointer = save;
3768 return scale;
3769 }
3770
3771 static int i386_displacement PARAMS ((char *, char *));
3772
3773 static int
3774 i386_displacement (disp_start, disp_end)
3775 char *disp_start;
3776 char *disp_end;
3777 {
3778 expressionS *exp;
3779 segT exp_seg = 0;
3780 char *save_input_line_pointer;
3781 #ifndef LEX_AT
3782 char *gotfree_input_line;
3783 #endif
3784 int bigdisp = Disp32;
3785
3786 if (flag_code == CODE_64BIT)
3787 {
3788 if (i.prefix[ADDR_PREFIX] == 0)
3789 bigdisp = Disp64;
3790 }
3791 else if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3792 bigdisp = Disp16;
3793 i.types[this_operand] |= bigdisp;
3794
3795 exp = &disp_expressions[i.disp_operands];
3796 i.op[this_operand].disps = exp;
3797 i.disp_operands++;
3798 save_input_line_pointer = input_line_pointer;
3799 input_line_pointer = disp_start;
3800 END_STRING_AND_SAVE (disp_end);
3801
3802 #ifndef GCC_ASM_O_HACK
3803 #define GCC_ASM_O_HACK 0
3804 #endif
3805 #if GCC_ASM_O_HACK
3806 END_STRING_AND_SAVE (disp_end + 1);
3807 if ((i.types[this_operand] & BaseIndex) != 0
3808 && displacement_string_end[-1] == '+')
3809 {
3810 /* This hack is to avoid a warning when using the "o"
3811 constraint within gcc asm statements.
3812 For instance:
3813
3814 #define _set_tssldt_desc(n,addr,limit,type) \
3815 __asm__ __volatile__ ( \
3816 "movw %w2,%0\n\t" \
3817 "movw %w1,2+%0\n\t" \
3818 "rorl $16,%1\n\t" \
3819 "movb %b1,4+%0\n\t" \
3820 "movb %4,5+%0\n\t" \
3821 "movb $0,6+%0\n\t" \
3822 "movb %h1,7+%0\n\t" \
3823 "rorl $16,%1" \
3824 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
3825
3826 This works great except that the output assembler ends
3827 up looking a bit weird if it turns out that there is
3828 no offset. You end up producing code that looks like:
3829
3830 #APP
3831 movw $235,(%eax)
3832 movw %dx,2+(%eax)
3833 rorl $16,%edx
3834 movb %dl,4+(%eax)
3835 movb $137,5+(%eax)
3836 movb $0,6+(%eax)
3837 movb %dh,7+(%eax)
3838 rorl $16,%edx
3839 #NO_APP
3840
3841 So here we provide the missing zero. */
3842
3843 *displacement_string_end = '0';
3844 }
3845 #endif
3846 #ifndef LEX_AT
3847 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3848 if (gotfree_input_line)
3849 input_line_pointer = gotfree_input_line;
3850 #endif
3851
3852 exp_seg = expression (exp);
3853
3854 SKIP_WHITESPACE ();
3855 if (*input_line_pointer)
3856 as_bad (_("junk `%s' after expression"), input_line_pointer);
3857 #if GCC_ASM_O_HACK
3858 RESTORE_END_STRING (disp_end + 1);
3859 #endif
3860 RESTORE_END_STRING (disp_end);
3861 input_line_pointer = save_input_line_pointer;
3862 #ifndef LEX_AT
3863 if (gotfree_input_line)
3864 free (gotfree_input_line);
3865 #endif
3866
3867 #ifdef BFD_ASSEMBLER
3868 /* We do this to make sure that the section symbol is in
3869 the symbol table. We will ultimately change the relocation
3870 to be relative to the beginning of the section. */
3871 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
3872 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3873 {
3874 if (exp->X_op != O_symbol)
3875 {
3876 as_bad (_("bad expression used with @%s"),
3877 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
3878 ? "GOTPCREL"
3879 : "GOTOFF"));
3880 return 0;
3881 }
3882
3883 if (S_IS_LOCAL (exp->X_add_symbol)
3884 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
3885 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
3886 exp->X_op = O_subtract;
3887 exp->X_op_symbol = GOT_symbol;
3888 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3889 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
3890 else
3891 i.reloc[this_operand] = BFD_RELOC_32;
3892 }
3893 #endif
3894
3895 if (exp->X_op == O_absent || exp->X_op == O_big)
3896 {
3897 /* Missing or bad expr becomes absolute 0. */
3898 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
3899 disp_start);
3900 exp->X_op = O_constant;
3901 exp->X_add_number = 0;
3902 exp->X_add_symbol = (symbolS *) 0;
3903 exp->X_op_symbol = (symbolS *) 0;
3904 }
3905
3906 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3907 if (exp->X_op != O_constant
3908 #ifdef BFD_ASSEMBLER
3909 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3910 #endif
3911 && exp_seg != absolute_section
3912 && exp_seg != text_section
3913 && exp_seg != data_section
3914 && exp_seg != bss_section
3915 && exp_seg != undefined_section
3916 #ifdef BFD_ASSEMBLER
3917 && !bfd_is_com_section (exp_seg)
3918 #endif
3919 )
3920 {
3921 #ifdef BFD_ASSEMBLER
3922 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3923 #else
3924 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3925 #endif
3926 return 0;
3927 }
3928 #endif
3929 else if (flag_code == CODE_64BIT)
3930 i.types[this_operand] |= Disp32S | Disp32;
3931 return 1;
3932 }
3933
3934 static int i386_index_check PARAMS ((const char *));
3935
3936 /* Make sure the memory operand we've been dealt is valid.
3937 Return 1 on success, 0 on a failure. */
3938
3939 static int
3940 i386_index_check (operand_string)
3941 const char *operand_string;
3942 {
3943 int ok;
3944 #if INFER_ADDR_PREFIX
3945 int fudged = 0;
3946
3947 tryprefix:
3948 #endif
3949 ok = 1;
3950 if (flag_code == CODE_64BIT)
3951 {
3952 if (i.prefix[ADDR_PREFIX] == 0)
3953 {
3954 /* 64bit checks. */
3955 if ((i.base_reg
3956 && ((i.base_reg->reg_type & Reg64) == 0)
3957 && (i.base_reg->reg_type != BaseIndex
3958 || i.index_reg))
3959 || (i.index_reg
3960 && ((i.index_reg->reg_type & (Reg64 | BaseIndex))
3961 != (Reg64 | BaseIndex))))
3962 ok = 0;
3963 }
3964 else
3965 {
3966 /* 32bit checks. */
3967 if ((i.base_reg
3968 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
3969 || (i.index_reg
3970 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
3971 != (Reg32 | BaseIndex))))
3972 ok = 0;
3973 }
3974 }
3975 else
3976 {
3977 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3978 {
3979 /* 16bit checks. */
3980 if ((i.base_reg
3981 && ((i.base_reg->reg_type & (Reg16 | BaseIndex | RegRex))
3982 != (Reg16 | BaseIndex)))
3983 || (i.index_reg
3984 && (((i.index_reg->reg_type & (Reg16 | BaseIndex))
3985 != (Reg16 | BaseIndex))
3986 || !(i.base_reg
3987 && i.base_reg->reg_num < 6
3988 && i.index_reg->reg_num >= 6
3989 && i.log2_scale_factor == 0))))
3990 ok = 0;
3991 }
3992 else
3993 {
3994 /* 32bit checks. */
3995 if ((i.base_reg
3996 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
3997 || (i.index_reg
3998 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
3999 != (Reg32 | BaseIndex))))
4000 ok = 0;
4001 }
4002 }
4003 if (!ok)
4004 {
4005 #if INFER_ADDR_PREFIX
4006 if (flag_code != CODE_64BIT
4007 && i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0')
4008 {
4009 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
4010 i.prefixes += 1;
4011 /* Change the size of any displacement too. At most one of
4012 Disp16 or Disp32 is set.
4013 FIXME. There doesn't seem to be any real need for separate
4014 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
4015 Removing them would probably clean up the code quite a lot. */
4016 if (i.types[this_operand] & (Disp16 | Disp32))
4017 i.types[this_operand] ^= (Disp16 | Disp32);
4018 fudged = 1;
4019 goto tryprefix;
4020 }
4021 if (fudged)
4022 as_bad (_("`%s' is not a valid base/index expression"),
4023 operand_string);
4024 else
4025 #endif
4026 as_bad (_("`%s' is not a valid %s bit base/index expression"),
4027 operand_string,
4028 flag_code_names[flag_code]);
4029 return 0;
4030 }
4031 return 1;
4032 }
4033
4034 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
4035 on error. */
4036
4037 static int
4038 i386_operand (operand_string)
4039 char *operand_string;
4040 {
4041 const reg_entry *r;
4042 char *end_op;
4043 char *op_string = operand_string;
4044
4045 if (is_space_char (*op_string))
4046 ++op_string;
4047
4048 /* We check for an absolute prefix (differentiating,
4049 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
4050 if (*op_string == ABSOLUTE_PREFIX)
4051 {
4052 ++op_string;
4053 if (is_space_char (*op_string))
4054 ++op_string;
4055 i.types[this_operand] |= JumpAbsolute;
4056 }
4057
4058 /* Check if operand is a register. */
4059 if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
4060 && (r = parse_register (op_string, &end_op)) != NULL)
4061 {
4062 /* Check for a segment override by searching for ':' after a
4063 segment register. */
4064 op_string = end_op;
4065 if (is_space_char (*op_string))
4066 ++op_string;
4067 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
4068 {
4069 switch (r->reg_num)
4070 {
4071 case 0:
4072 i.seg[i.mem_operands] = &es;
4073 break;
4074 case 1:
4075 i.seg[i.mem_operands] = &cs;
4076 break;
4077 case 2:
4078 i.seg[i.mem_operands] = &ss;
4079 break;
4080 case 3:
4081 i.seg[i.mem_operands] = &ds;
4082 break;
4083 case 4:
4084 i.seg[i.mem_operands] = &fs;
4085 break;
4086 case 5:
4087 i.seg[i.mem_operands] = &gs;
4088 break;
4089 }
4090
4091 /* Skip the ':' and whitespace. */
4092 ++op_string;
4093 if (is_space_char (*op_string))
4094 ++op_string;
4095
4096 if (!is_digit_char (*op_string)
4097 && !is_identifier_char (*op_string)
4098 && *op_string != '('
4099 && *op_string != ABSOLUTE_PREFIX)
4100 {
4101 as_bad (_("bad memory operand `%s'"), op_string);
4102 return 0;
4103 }
4104 /* Handle case of %es:*foo. */
4105 if (*op_string == ABSOLUTE_PREFIX)
4106 {
4107 ++op_string;
4108 if (is_space_char (*op_string))
4109 ++op_string;
4110 i.types[this_operand] |= JumpAbsolute;
4111 }
4112 goto do_memory_reference;
4113 }
4114 if (*op_string)
4115 {
4116 as_bad (_("junk `%s' after register"), op_string);
4117 return 0;
4118 }
4119 i.types[this_operand] |= r->reg_type & ~BaseIndex;
4120 i.op[this_operand].regs = r;
4121 i.reg_operands++;
4122 }
4123 else if (*op_string == REGISTER_PREFIX)
4124 {
4125 as_bad (_("bad register name `%s'"), op_string);
4126 return 0;
4127 }
4128 else if (*op_string == IMMEDIATE_PREFIX)
4129 {
4130 ++op_string;
4131 if (i.types[this_operand] & JumpAbsolute)
4132 {
4133 as_bad (_("immediate operand illegal with absolute jump"));
4134 return 0;
4135 }
4136 if (!i386_immediate (op_string))
4137 return 0;
4138 }
4139 else if (is_digit_char (*op_string)
4140 || is_identifier_char (*op_string)
4141 || *op_string == '(')
4142 {
4143 /* This is a memory reference of some sort. */
4144 char *base_string;
4145
4146 /* Start and end of displacement string expression (if found). */
4147 char *displacement_string_start;
4148 char *displacement_string_end;
4149
4150 do_memory_reference:
4151 if ((i.mem_operands == 1
4152 && (current_templates->start->opcode_modifier & IsString) == 0)
4153 || i.mem_operands == 2)
4154 {
4155 as_bad (_("too many memory references for `%s'"),
4156 current_templates->start->name);
4157 return 0;
4158 }
4159
4160 /* Check for base index form. We detect the base index form by
4161 looking for an ')' at the end of the operand, searching
4162 for the '(' matching it, and finding a REGISTER_PREFIX or ','
4163 after the '('. */
4164 base_string = op_string + strlen (op_string);
4165
4166 --base_string;
4167 if (is_space_char (*base_string))
4168 --base_string;
4169
4170 /* If we only have a displacement, set-up for it to be parsed later. */
4171 displacement_string_start = op_string;
4172 displacement_string_end = base_string + 1;
4173
4174 if (*base_string == ')')
4175 {
4176 char *temp_string;
4177 unsigned int parens_balanced = 1;
4178 /* We've already checked that the number of left & right ()'s are
4179 equal, so this loop will not be infinite. */
4180 do
4181 {
4182 base_string--;
4183 if (*base_string == ')')
4184 parens_balanced++;
4185 if (*base_string == '(')
4186 parens_balanced--;
4187 }
4188 while (parens_balanced);
4189
4190 temp_string = base_string;
4191
4192 /* Skip past '(' and whitespace. */
4193 ++base_string;
4194 if (is_space_char (*base_string))
4195 ++base_string;
4196
4197 if (*base_string == ','
4198 || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
4199 && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
4200 {
4201 displacement_string_end = temp_string;
4202
4203 i.types[this_operand] |= BaseIndex;
4204
4205 if (i.base_reg)
4206 {
4207 base_string = end_op;
4208 if (is_space_char (*base_string))
4209 ++base_string;
4210 }
4211
4212 /* There may be an index reg or scale factor here. */
4213 if (*base_string == ',')
4214 {
4215 ++base_string;
4216 if (is_space_char (*base_string))
4217 ++base_string;
4218
4219 if ((*base_string == REGISTER_PREFIX || allow_naked_reg)
4220 && (i.index_reg = parse_register (base_string, &end_op)) != NULL)
4221 {
4222 base_string = end_op;
4223 if (is_space_char (*base_string))
4224 ++base_string;
4225 if (*base_string == ',')
4226 {
4227 ++base_string;
4228 if (is_space_char (*base_string))
4229 ++base_string;
4230 }
4231 else if (*base_string != ')')
4232 {
4233 as_bad (_("expecting `,' or `)' after index register in `%s'"),
4234 operand_string);
4235 return 0;
4236 }
4237 }
4238 else if (*base_string == REGISTER_PREFIX)
4239 {
4240 as_bad (_("bad register name `%s'"), base_string);
4241 return 0;
4242 }
4243
4244 /* Check for scale factor. */
4245 if (*base_string != ')')
4246 {
4247 char *end_scale = i386_scale (base_string);
4248
4249 if (!end_scale)
4250 return 0;
4251
4252 base_string = end_scale;
4253 if (is_space_char (*base_string))
4254 ++base_string;
4255 if (*base_string != ')')
4256 {
4257 as_bad (_("expecting `)' after scale factor in `%s'"),
4258 operand_string);
4259 return 0;
4260 }
4261 }
4262 else if (!i.index_reg)
4263 {
4264 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
4265 *base_string);
4266 return 0;
4267 }
4268 }
4269 else if (*base_string != ')')
4270 {
4271 as_bad (_("expecting `,' or `)' after base register in `%s'"),
4272 operand_string);
4273 return 0;
4274 }
4275 }
4276 else if (*base_string == REGISTER_PREFIX)
4277 {
4278 as_bad (_("bad register name `%s'"), base_string);
4279 return 0;
4280 }
4281 }
4282
4283 /* If there's an expression beginning the operand, parse it,
4284 assuming displacement_string_start and
4285 displacement_string_end are meaningful. */
4286 if (displacement_string_start != displacement_string_end)
4287 {
4288 if (!i386_displacement (displacement_string_start,
4289 displacement_string_end))
4290 return 0;
4291 }
4292
4293 /* Special case for (%dx) while doing input/output op. */
4294 if (i.base_reg
4295 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
4296 && i.index_reg == 0
4297 && i.log2_scale_factor == 0
4298 && i.seg[i.mem_operands] == 0
4299 && (i.types[this_operand] & Disp) == 0)
4300 {
4301 i.types[this_operand] = InOutPortReg;
4302 return 1;
4303 }
4304
4305 if (i386_index_check (operand_string) == 0)
4306 return 0;
4307 i.mem_operands++;
4308 }
4309 else
4310 {
4311 /* It's not a memory operand; argh! */
4312 as_bad (_("invalid char %s beginning operand %d `%s'"),
4313 output_invalid (*op_string),
4314 this_operand + 1,
4315 op_string);
4316 return 0;
4317 }
4318 return 1; /* Normal return. */
4319 }
4320 \f
4321 /* md_estimate_size_before_relax()
4322
4323 Called just before relax() for rs_machine_dependent frags. The x86
4324 assembler uses these frags to handle variable size jump
4325 instructions.
4326
4327 Any symbol that is now undefined will not become defined.
4328 Return the correct fr_subtype in the frag.
4329 Return the initial "guess for variable size of frag" to caller.
4330 The guess is actually the growth beyond the fixed part. Whatever
4331 we do to grow the fixed or variable part contributes to our
4332 returned value. */
4333
4334 int
4335 md_estimate_size_before_relax (fragP, segment)
4336 fragS *fragP;
4337 segT segment;
4338 {
4339 /* We've already got fragP->fr_subtype right; all we have to do is
4340 check for un-relaxable symbols. On an ELF system, we can't relax
4341 an externally visible symbol, because it may be overridden by a
4342 shared library. */
4343 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
4344 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4345 || (OUTPUT_FLAVOR == bfd_target_elf_flavour
4346 && (S_IS_EXTERNAL (fragP->fr_symbol)
4347 || S_IS_WEAK (fragP->fr_symbol)))
4348 #endif
4349 )
4350 {
4351 /* Symbol is undefined in this segment, or we need to keep a
4352 reloc so that weak symbols can be overridden. */
4353 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
4354 RELOC_ENUM reloc_type;
4355 unsigned char *opcode;
4356 int old_fr_fix;
4357
4358 if (fragP->fr_var != NO_RELOC)
4359 reloc_type = fragP->fr_var;
4360 else if (size == 2)
4361 reloc_type = BFD_RELOC_16_PCREL;
4362 else
4363 reloc_type = BFD_RELOC_32_PCREL;
4364
4365 old_fr_fix = fragP->fr_fix;
4366 opcode = (unsigned char *) fragP->fr_opcode;
4367
4368 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
4369 {
4370 case UNCOND_JUMP:
4371 /* Make jmp (0xeb) a (d)word displacement jump. */
4372 opcode[0] = 0xe9;
4373 fragP->fr_fix += size;
4374 fix_new (fragP, old_fr_fix, size,
4375 fragP->fr_symbol,
4376 fragP->fr_offset, 1,
4377 reloc_type);
4378 break;
4379
4380 case COND_JUMP86:
4381 if (size == 2
4382 && (!no_cond_jump_promotion || fragP->fr_var != NO_RELOC))
4383 {
4384 /* Negate the condition, and branch past an
4385 unconditional jump. */
4386 opcode[0] ^= 1;
4387 opcode[1] = 3;
4388 /* Insert an unconditional jump. */
4389 opcode[2] = 0xe9;
4390 /* We added two extra opcode bytes, and have a two byte
4391 offset. */
4392 fragP->fr_fix += 2 + 2;
4393 fix_new (fragP, old_fr_fix + 2, 2,
4394 fragP->fr_symbol,
4395 fragP->fr_offset, 1,
4396 reloc_type);
4397 break;
4398 }
4399 /* Fall through. */
4400
4401 case COND_JUMP:
4402 if (no_cond_jump_promotion && fragP->fr_var == NO_RELOC)
4403 {
4404 fixS *fixP;
4405
4406 fragP->fr_fix += 1;
4407 fixP = fix_new (fragP, old_fr_fix, 1,
4408 fragP->fr_symbol,
4409 fragP->fr_offset, 1,
4410 BFD_RELOC_8_PCREL);
4411 fixP->fx_signed = 1;
4412 break;
4413 }
4414
4415 /* This changes the byte-displacement jump 0x7N
4416 to the (d)word-displacement jump 0x0f,0x8N. */
4417 opcode[1] = opcode[0] + 0x10;
4418 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4419 /* We've added an opcode byte. */
4420 fragP->fr_fix += 1 + size;
4421 fix_new (fragP, old_fr_fix + 1, size,
4422 fragP->fr_symbol,
4423 fragP->fr_offset, 1,
4424 reloc_type);
4425 break;
4426
4427 default:
4428 BAD_CASE (fragP->fr_subtype);
4429 break;
4430 }
4431 frag_wane (fragP);
4432 return fragP->fr_fix - old_fr_fix;
4433 }
4434
4435 /* Guess size depending on current relax state. Initially the relax
4436 state will correspond to a short jump and we return 1, because
4437 the variable part of the frag (the branch offset) is one byte
4438 long. However, we can relax a section more than once and in that
4439 case we must either set fr_subtype back to the unrelaxed state,
4440 or return the value for the appropriate branch. */
4441 return md_relax_table[fragP->fr_subtype].rlx_length;
4442 }
4443
4444 /* Called after relax() is finished.
4445
4446 In: Address of frag.
4447 fr_type == rs_machine_dependent.
4448 fr_subtype is what the address relaxed to.
4449
4450 Out: Any fixSs and constants are set up.
4451 Caller will turn frag into a ".space 0". */
4452
4453 #ifndef BFD_ASSEMBLER
4454 void
4455 md_convert_frag (headers, sec, fragP)
4456 object_headers *headers ATTRIBUTE_UNUSED;
4457 segT sec ATTRIBUTE_UNUSED;
4458 fragS *fragP;
4459 #else
4460 void
4461 md_convert_frag (abfd, sec, fragP)
4462 bfd *abfd ATTRIBUTE_UNUSED;
4463 segT sec ATTRIBUTE_UNUSED;
4464 fragS *fragP;
4465 #endif
4466 {
4467 unsigned char *opcode;
4468 unsigned char *where_to_put_displacement = NULL;
4469 offsetT target_address;
4470 offsetT opcode_address;
4471 unsigned int extension = 0;
4472 offsetT displacement_from_opcode_start;
4473
4474 opcode = (unsigned char *) fragP->fr_opcode;
4475
4476 /* Address we want to reach in file space. */
4477 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
4478
4479 /* Address opcode resides at in file space. */
4480 opcode_address = fragP->fr_address + fragP->fr_fix;
4481
4482 /* Displacement from opcode start to fill into instruction. */
4483 displacement_from_opcode_start = target_address - opcode_address;
4484
4485 if ((fragP->fr_subtype & BIG) == 0)
4486 {
4487 /* Don't have to change opcode. */
4488 extension = 1; /* 1 opcode + 1 displacement */
4489 where_to_put_displacement = &opcode[1];
4490 }
4491 else
4492 {
4493 if (no_cond_jump_promotion
4494 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
4495 as_warn_where (fragP->fr_file, fragP->fr_line, _("long jump required"));
4496
4497 switch (fragP->fr_subtype)
4498 {
4499 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
4500 extension = 4; /* 1 opcode + 4 displacement */
4501 opcode[0] = 0xe9;
4502 where_to_put_displacement = &opcode[1];
4503 break;
4504
4505 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
4506 extension = 2; /* 1 opcode + 2 displacement */
4507 opcode[0] = 0xe9;
4508 where_to_put_displacement = &opcode[1];
4509 break;
4510
4511 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
4512 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
4513 extension = 5; /* 2 opcode + 4 displacement */
4514 opcode[1] = opcode[0] + 0x10;
4515 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4516 where_to_put_displacement = &opcode[2];
4517 break;
4518
4519 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
4520 extension = 3; /* 2 opcode + 2 displacement */
4521 opcode[1] = opcode[0] + 0x10;
4522 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4523 where_to_put_displacement = &opcode[2];
4524 break;
4525
4526 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
4527 extension = 4;
4528 opcode[0] ^= 1;
4529 opcode[1] = 3;
4530 opcode[2] = 0xe9;
4531 where_to_put_displacement = &opcode[3];
4532 break;
4533
4534 default:
4535 BAD_CASE (fragP->fr_subtype);
4536 break;
4537 }
4538 }
4539
4540 /* Now put displacement after opcode. */
4541 md_number_to_chars ((char *) where_to_put_displacement,
4542 (valueT) (displacement_from_opcode_start - extension),
4543 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
4544 fragP->fr_fix += extension;
4545 }
4546 \f
4547 /* Size of byte displacement jmp. */
4548 int md_short_jump_size = 2;
4549
4550 /* Size of dword displacement jmp. */
4551 int md_long_jump_size = 5;
4552
4553 /* Size of relocation record. */
4554 const int md_reloc_size = 8;
4555
4556 void
4557 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
4558 char *ptr;
4559 addressT from_addr, to_addr;
4560 fragS *frag ATTRIBUTE_UNUSED;
4561 symbolS *to_symbol ATTRIBUTE_UNUSED;
4562 {
4563 offsetT offset;
4564
4565 offset = to_addr - (from_addr + 2);
4566 /* Opcode for byte-disp jump. */
4567 md_number_to_chars (ptr, (valueT) 0xeb, 1);
4568 md_number_to_chars (ptr + 1, (valueT) offset, 1);
4569 }
4570
4571 void
4572 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
4573 char *ptr;
4574 addressT from_addr, to_addr;
4575 fragS *frag ATTRIBUTE_UNUSED;
4576 symbolS *to_symbol ATTRIBUTE_UNUSED;
4577 {
4578 offsetT offset;
4579
4580 offset = to_addr - (from_addr + 5);
4581 md_number_to_chars (ptr, (valueT) 0xe9, 1);
4582 md_number_to_chars (ptr + 1, (valueT) offset, 4);
4583 }
4584 \f
4585 /* Apply a fixup (fixS) to segment data, once it has been determined
4586 by our caller that we have all the info we need to fix it up.
4587
4588 On the 386, immediates, displacements, and data pointers are all in
4589 the same (little-endian) format, so we don't need to care about which
4590 we are handling. */
4591
4592 void
4593 md_apply_fix3 (fixP, valP, seg)
4594 /* The fix we're to put in. */
4595 fixS *fixP;
4596 /* Pointer to the value of the bits. */
4597 valueT *valP;
4598 /* Segment fix is from. */
4599 segT seg ATTRIBUTE_UNUSED;
4600 {
4601 char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
4602 valueT value = *valP;
4603
4604 #if defined (BFD_ASSEMBLER) && !defined (TE_Mach)
4605 if (fixP->fx_pcrel)
4606 {
4607 switch (fixP->fx_r_type)
4608 {
4609 default:
4610 break;
4611
4612 case BFD_RELOC_32:
4613 fixP->fx_r_type = BFD_RELOC_32_PCREL;
4614 break;
4615 case BFD_RELOC_16:
4616 fixP->fx_r_type = BFD_RELOC_16_PCREL;
4617 break;
4618 case BFD_RELOC_8:
4619 fixP->fx_r_type = BFD_RELOC_8_PCREL;
4620 break;
4621 }
4622 }
4623
4624 if (fixP->fx_addsy != NULL
4625 && (fixP->fx_r_type == BFD_RELOC_32_PCREL
4626 || fixP->fx_r_type == BFD_RELOC_16_PCREL
4627 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
4628 && !use_rela_relocations)
4629 {
4630 /* This is a hack. There should be a better way to handle this.
4631 This covers for the fact that bfd_install_relocation will
4632 subtract the current location (for partial_inplace, PC relative
4633 relocations); see more below. */
4634 #ifndef OBJ_AOUT
4635 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4636 #ifdef TE_PE
4637 || OUTPUT_FLAVOR == bfd_target_coff_flavour
4638 #endif
4639 )
4640 value += fixP->fx_where + fixP->fx_frag->fr_address;
4641 #endif
4642 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4643 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
4644 {
4645 segT sym_seg = S_GET_SEGMENT (fixP->fx_addsy);
4646
4647 if ((sym_seg == seg
4648 || (symbol_section_p (fixP->fx_addsy)
4649 && sym_seg != absolute_section))
4650 && !generic_force_reloc (fixP))
4651 {
4652 /* Yes, we add the values in twice. This is because
4653 bfd_install_relocation subtracts them out again. I think
4654 bfd_install_relocation is broken, but I don't dare change
4655 it. FIXME. */
4656 value += fixP->fx_where + fixP->fx_frag->fr_address;
4657 }
4658 }
4659 #endif
4660 #if defined (OBJ_COFF) && defined (TE_PE)
4661 /* For some reason, the PE format does not store a section
4662 address offset for a PC relative symbol. */
4663 if (S_GET_SEGMENT (fixP->fx_addsy) != seg)
4664 value += md_pcrel_from (fixP);
4665 #endif
4666 }
4667
4668 /* Fix a few things - the dynamic linker expects certain values here,
4669 and we must not dissappoint it. */
4670 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4671 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4672 && fixP->fx_addsy)
4673 switch (fixP->fx_r_type)
4674 {
4675 case BFD_RELOC_386_PLT32:
4676 case BFD_RELOC_X86_64_PLT32:
4677 /* Make the jump instruction point to the address of the operand. At
4678 runtime we merely add the offset to the actual PLT entry. */
4679 value = -4;
4680 break;
4681
4682 case BFD_RELOC_386_TLS_GD:
4683 case BFD_RELOC_386_TLS_LDM:
4684 case BFD_RELOC_386_TLS_IE_32:
4685 case BFD_RELOC_386_TLS_IE:
4686 case BFD_RELOC_386_TLS_GOTIE:
4687 case BFD_RELOC_X86_64_TLSGD:
4688 case BFD_RELOC_X86_64_TLSLD:
4689 case BFD_RELOC_X86_64_GOTTPOFF:
4690 value = 0; /* Fully resolved at runtime. No addend. */
4691 /* Fallthrough */
4692 case BFD_RELOC_386_TLS_LE:
4693 case BFD_RELOC_386_TLS_LDO_32:
4694 case BFD_RELOC_386_TLS_LE_32:
4695 case BFD_RELOC_X86_64_DTPOFF32:
4696 case BFD_RELOC_X86_64_TPOFF32:
4697 S_SET_THREAD_LOCAL (fixP->fx_addsy);
4698 break;
4699
4700 case BFD_RELOC_386_GOT32:
4701 case BFD_RELOC_X86_64_GOT32:
4702 value = 0; /* Fully resolved at runtime. No addend. */
4703 break;
4704
4705 case BFD_RELOC_VTABLE_INHERIT:
4706 case BFD_RELOC_VTABLE_ENTRY:
4707 fixP->fx_done = 0;
4708 return;
4709
4710 default:
4711 break;
4712 }
4713 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
4714 *valP = value;
4715 #endif /* defined (BFD_ASSEMBLER) && !defined (TE_Mach) */
4716
4717 /* Are we finished with this relocation now? */
4718 if (fixP->fx_addsy == NULL)
4719 fixP->fx_done = 1;
4720 #ifdef BFD_ASSEMBLER
4721 else if (use_rela_relocations)
4722 {
4723 fixP->fx_no_overflow = 1;
4724 /* Remember value for tc_gen_reloc. */
4725 fixP->fx_addnumber = value;
4726 value = 0;
4727 }
4728 #endif
4729 md_number_to_chars (p, value, fixP->fx_size);
4730 }
4731 \f
4732 #define MAX_LITTLENUMS 6
4733
4734 /* Turn the string pointed to by litP into a floating point constant
4735 of type TYPE, and emit the appropriate bytes. The number of
4736 LITTLENUMS emitted is stored in *SIZEP. An error message is
4737 returned, or NULL on OK. */
4738
4739 char *
4740 md_atof (type, litP, sizeP)
4741 int type;
4742 char *litP;
4743 int *sizeP;
4744 {
4745 int prec;
4746 LITTLENUM_TYPE words[MAX_LITTLENUMS];
4747 LITTLENUM_TYPE *wordP;
4748 char *t;
4749
4750 switch (type)
4751 {
4752 case 'f':
4753 case 'F':
4754 prec = 2;
4755 break;
4756
4757 case 'd':
4758 case 'D':
4759 prec = 4;
4760 break;
4761
4762 case 'x':
4763 case 'X':
4764 prec = 5;
4765 break;
4766
4767 default:
4768 *sizeP = 0;
4769 return _("Bad call to md_atof ()");
4770 }
4771 t = atof_ieee (input_line_pointer, type, words);
4772 if (t)
4773 input_line_pointer = t;
4774
4775 *sizeP = prec * sizeof (LITTLENUM_TYPE);
4776 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
4777 the bigendian 386. */
4778 for (wordP = words + prec - 1; prec--;)
4779 {
4780 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
4781 litP += sizeof (LITTLENUM_TYPE);
4782 }
4783 return 0;
4784 }
4785 \f
4786 char output_invalid_buf[8];
4787
4788 static char *
4789 output_invalid (c)
4790 int c;
4791 {
4792 if (ISPRINT (c))
4793 sprintf (output_invalid_buf, "'%c'", c);
4794 else
4795 sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
4796 return output_invalid_buf;
4797 }
4798
4799 /* REG_STRING starts *before* REGISTER_PREFIX. */
4800
4801 static const reg_entry *
4802 parse_register (reg_string, end_op)
4803 char *reg_string;
4804 char **end_op;
4805 {
4806 char *s = reg_string;
4807 char *p;
4808 char reg_name_given[MAX_REG_NAME_SIZE + 1];
4809 const reg_entry *r;
4810
4811 /* Skip possible REGISTER_PREFIX and possible whitespace. */
4812 if (*s == REGISTER_PREFIX)
4813 ++s;
4814
4815 if (is_space_char (*s))
4816 ++s;
4817
4818 p = reg_name_given;
4819 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
4820 {
4821 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
4822 return (const reg_entry *) NULL;
4823 s++;
4824 }
4825
4826 /* For naked regs, make sure that we are not dealing with an identifier.
4827 This prevents confusing an identifier like `eax_var' with register
4828 `eax'. */
4829 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
4830 return (const reg_entry *) NULL;
4831
4832 *end_op = s;
4833
4834 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
4835
4836 /* Handle floating point regs, allowing spaces in the (i) part. */
4837 if (r == i386_regtab /* %st is first entry of table */)
4838 {
4839 if (is_space_char (*s))
4840 ++s;
4841 if (*s == '(')
4842 {
4843 ++s;
4844 if (is_space_char (*s))
4845 ++s;
4846 if (*s >= '0' && *s <= '7')
4847 {
4848 r = &i386_float_regtab[*s - '0'];
4849 ++s;
4850 if (is_space_char (*s))
4851 ++s;
4852 if (*s == ')')
4853 {
4854 *end_op = s + 1;
4855 return r;
4856 }
4857 }
4858 /* We have "%st(" then garbage. */
4859 return (const reg_entry *) NULL;
4860 }
4861 }
4862
4863 if (r != NULL
4864 && (r->reg_flags & (RegRex64 | RegRex)) != 0
4865 && flag_code != CODE_64BIT)
4866 {
4867 return (const reg_entry *) NULL;
4868 }
4869
4870 return r;
4871 }
4872 \f
4873 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4874 const char *md_shortopts = "kVQ:sq";
4875 #else
4876 const char *md_shortopts = "q";
4877 #endif
4878
4879 struct option md_longopts[] = {
4880 #define OPTION_32 (OPTION_MD_BASE + 0)
4881 {"32", no_argument, NULL, OPTION_32},
4882 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4883 #define OPTION_64 (OPTION_MD_BASE + 1)
4884 {"64", no_argument, NULL, OPTION_64},
4885 #endif
4886 {NULL, no_argument, NULL, 0}
4887 };
4888 size_t md_longopts_size = sizeof (md_longopts);
4889
4890 int
4891 md_parse_option (c, arg)
4892 int c;
4893 char *arg ATTRIBUTE_UNUSED;
4894 {
4895 switch (c)
4896 {
4897 case 'q':
4898 quiet_warnings = 1;
4899 break;
4900
4901 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4902 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
4903 should be emitted or not. FIXME: Not implemented. */
4904 case 'Q':
4905 break;
4906
4907 /* -V: SVR4 argument to print version ID. */
4908 case 'V':
4909 print_version_id ();
4910 break;
4911
4912 /* -k: Ignore for FreeBSD compatibility. */
4913 case 'k':
4914 break;
4915
4916 case 's':
4917 /* -s: On i386 Solaris, this tells the native assembler to use
4918 .stab instead of .stab.excl. We always use .stab anyhow. */
4919 break;
4920
4921 case OPTION_64:
4922 {
4923 const char **list, **l;
4924
4925 list = bfd_target_list ();
4926 for (l = list; *l != NULL; l++)
4927 if (strcmp (*l, "elf64-x86-64") == 0)
4928 {
4929 default_arch = "x86_64";
4930 break;
4931 }
4932 if (*l == NULL)
4933 as_fatal (_("No compiled in support for x86_64"));
4934 free (list);
4935 }
4936 break;
4937 #endif
4938
4939 case OPTION_32:
4940 default_arch = "i386";
4941 break;
4942
4943 default:
4944 return 0;
4945 }
4946 return 1;
4947 }
4948
4949 void
4950 md_show_usage (stream)
4951 FILE *stream;
4952 {
4953 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4954 fprintf (stream, _("\
4955 -Q ignored\n\
4956 -V print assembler version number\n\
4957 -k ignored\n\
4958 -q quieten some warnings\n\
4959 -s ignored\n"));
4960 #else
4961 fprintf (stream, _("\
4962 -q quieten some warnings\n"));
4963 #endif
4964 }
4965
4966 #ifdef BFD_ASSEMBLER
4967 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
4968 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
4969
4970 /* Pick the target format to use. */
4971
4972 const char *
4973 i386_target_format ()
4974 {
4975 if (!strcmp (default_arch, "x86_64"))
4976 set_code_flag (CODE_64BIT);
4977 else if (!strcmp (default_arch, "i386"))
4978 set_code_flag (CODE_32BIT);
4979 else
4980 as_fatal (_("Unknown architecture"));
4981 switch (OUTPUT_FLAVOR)
4982 {
4983 #ifdef OBJ_MAYBE_AOUT
4984 case bfd_target_aout_flavour:
4985 return AOUT_TARGET_FORMAT;
4986 #endif
4987 #ifdef OBJ_MAYBE_COFF
4988 case bfd_target_coff_flavour:
4989 return "coff-i386";
4990 #endif
4991 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
4992 case bfd_target_elf_flavour:
4993 {
4994 if (flag_code == CODE_64BIT)
4995 use_rela_relocations = 1;
4996 return flag_code == CODE_64BIT ? "elf64-x86-64" : ELF_TARGET_FORMAT;
4997 }
4998 #endif
4999 default:
5000 abort ();
5001 return NULL;
5002 }
5003 }
5004
5005 #endif /* OBJ_MAYBE_ more than one */
5006
5007 #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
5008 void i386_elf_emit_arch_note ()
5009 {
5010 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
5011 && cpu_arch_name != NULL)
5012 {
5013 char *p;
5014 asection *seg = now_seg;
5015 subsegT subseg = now_subseg;
5016 Elf_Internal_Note i_note;
5017 Elf_External_Note e_note;
5018 asection *note_secp;
5019 int len;
5020
5021 /* Create the .note section. */
5022 note_secp = subseg_new (".note", 0);
5023 bfd_set_section_flags (stdoutput,
5024 note_secp,
5025 SEC_HAS_CONTENTS | SEC_READONLY);
5026
5027 /* Process the arch string. */
5028 len = strlen (cpu_arch_name);
5029
5030 i_note.namesz = len + 1;
5031 i_note.descsz = 0;
5032 i_note.type = NT_ARCH;
5033 p = frag_more (sizeof (e_note.namesz));
5034 md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
5035 p = frag_more (sizeof (e_note.descsz));
5036 md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
5037 p = frag_more (sizeof (e_note.type));
5038 md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
5039 p = frag_more (len + 1);
5040 strcpy (p, cpu_arch_name);
5041
5042 frag_align (2, 0, 0);
5043
5044 subseg_set (seg, subseg);
5045 }
5046 }
5047 #endif
5048 #endif /* BFD_ASSEMBLER */
5049 \f
5050 symbolS *
5051 md_undefined_symbol (name)
5052 char *name;
5053 {
5054 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
5055 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
5056 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
5057 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
5058 {
5059 if (!GOT_symbol)
5060 {
5061 if (symbol_find (name))
5062 as_bad (_("GOT already in symbol table"));
5063 GOT_symbol = symbol_new (name, undefined_section,
5064 (valueT) 0, &zero_address_frag);
5065 };
5066 return GOT_symbol;
5067 }
5068 return 0;
5069 }
5070
5071 /* Round up a section size to the appropriate boundary. */
5072
5073 valueT
5074 md_section_align (segment, size)
5075 segT segment ATTRIBUTE_UNUSED;
5076 valueT size;
5077 {
5078 #ifdef BFD_ASSEMBLER
5079 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
5080 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
5081 {
5082 /* For a.out, force the section size to be aligned. If we don't do
5083 this, BFD will align it for us, but it will not write out the
5084 final bytes of the section. This may be a bug in BFD, but it is
5085 easier to fix it here since that is how the other a.out targets
5086 work. */
5087 int align;
5088
5089 align = bfd_get_section_alignment (stdoutput, segment);
5090 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
5091 }
5092 #endif
5093 #endif
5094
5095 return size;
5096 }
5097
5098 /* On the i386, PC-relative offsets are relative to the start of the
5099 next instruction. That is, the address of the offset, plus its
5100 size, since the offset is always the last part of the insn. */
5101
5102 long
5103 md_pcrel_from (fixP)
5104 fixS *fixP;
5105 {
5106 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
5107 }
5108
5109 #ifndef I386COFF
5110
5111 static void
5112 s_bss (ignore)
5113 int ignore ATTRIBUTE_UNUSED;
5114 {
5115 int temp;
5116
5117 temp = get_absolute_expression ();
5118 subseg_set (bss_section, (subsegT) temp);
5119 demand_empty_rest_of_line ();
5120 }
5121
5122 #endif
5123
5124 #ifdef BFD_ASSEMBLER
5125
5126 void
5127 i386_validate_fix (fixp)
5128 fixS *fixp;
5129 {
5130 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
5131 {
5132 /* GOTOFF relocation are nonsense in 64bit mode. */
5133 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
5134 {
5135 if (flag_code != CODE_64BIT)
5136 abort ();
5137 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
5138 }
5139 else
5140 {
5141 if (flag_code == CODE_64BIT)
5142 abort ();
5143 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
5144 }
5145 fixp->fx_subsy = 0;
5146 }
5147 }
5148
5149 arelent *
5150 tc_gen_reloc (section, fixp)
5151 asection *section ATTRIBUTE_UNUSED;
5152 fixS *fixp;
5153 {
5154 arelent *rel;
5155 bfd_reloc_code_real_type code;
5156
5157 switch (fixp->fx_r_type)
5158 {
5159 case BFD_RELOC_X86_64_PLT32:
5160 case BFD_RELOC_X86_64_GOT32:
5161 case BFD_RELOC_X86_64_GOTPCREL:
5162 case BFD_RELOC_386_PLT32:
5163 case BFD_RELOC_386_GOT32:
5164 case BFD_RELOC_386_GOTOFF:
5165 case BFD_RELOC_386_GOTPC:
5166 case BFD_RELOC_386_TLS_GD:
5167 case BFD_RELOC_386_TLS_LDM:
5168 case BFD_RELOC_386_TLS_LDO_32:
5169 case BFD_RELOC_386_TLS_IE_32:
5170 case BFD_RELOC_386_TLS_IE:
5171 case BFD_RELOC_386_TLS_GOTIE:
5172 case BFD_RELOC_386_TLS_LE_32:
5173 case BFD_RELOC_386_TLS_LE:
5174 case BFD_RELOC_X86_64_32S:
5175 case BFD_RELOC_X86_64_TLSGD:
5176 case BFD_RELOC_X86_64_TLSLD:
5177 case BFD_RELOC_X86_64_DTPOFF32:
5178 case BFD_RELOC_X86_64_GOTTPOFF:
5179 case BFD_RELOC_X86_64_TPOFF32:
5180 case BFD_RELOC_RVA:
5181 case BFD_RELOC_VTABLE_ENTRY:
5182 case BFD_RELOC_VTABLE_INHERIT:
5183 code = fixp->fx_r_type;
5184 break;
5185 default:
5186 if (fixp->fx_pcrel)
5187 {
5188 switch (fixp->fx_size)
5189 {
5190 default:
5191 as_bad_where (fixp->fx_file, fixp->fx_line,
5192 _("can not do %d byte pc-relative relocation"),
5193 fixp->fx_size);
5194 code = BFD_RELOC_32_PCREL;
5195 break;
5196 case 1: code = BFD_RELOC_8_PCREL; break;
5197 case 2: code = BFD_RELOC_16_PCREL; break;
5198 case 4: code = BFD_RELOC_32_PCREL; break;
5199 }
5200 }
5201 else
5202 {
5203 switch (fixp->fx_size)
5204 {
5205 default:
5206 as_bad_where (fixp->fx_file, fixp->fx_line,
5207 _("can not do %d byte relocation"),
5208 fixp->fx_size);
5209 code = BFD_RELOC_32;
5210 break;
5211 case 1: code = BFD_RELOC_8; break;
5212 case 2: code = BFD_RELOC_16; break;
5213 case 4: code = BFD_RELOC_32; break;
5214 #ifdef BFD64
5215 case 8: code = BFD_RELOC_64; break;
5216 #endif
5217 }
5218 }
5219 break;
5220 }
5221
5222 if (code == BFD_RELOC_32
5223 && GOT_symbol
5224 && fixp->fx_addsy == GOT_symbol)
5225 {
5226 /* We don't support GOTPC on 64bit targets. */
5227 if (flag_code == CODE_64BIT)
5228 abort ();
5229 code = BFD_RELOC_386_GOTPC;
5230 }
5231
5232 rel = (arelent *) xmalloc (sizeof (arelent));
5233 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
5234 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
5235
5236 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
5237 if (!use_rela_relocations)
5238 {
5239 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
5240 vtable entry to be used in the relocation's section offset. */
5241 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
5242 rel->address = fixp->fx_offset;
5243
5244 rel->addend = 0;
5245 }
5246 /* Use the rela in 64bit mode. */
5247 else
5248 {
5249 if (!fixp->fx_pcrel)
5250 rel->addend = fixp->fx_offset;
5251 else
5252 switch (code)
5253 {
5254 case BFD_RELOC_X86_64_PLT32:
5255 case BFD_RELOC_X86_64_GOT32:
5256 case BFD_RELOC_X86_64_GOTPCREL:
5257 case BFD_RELOC_X86_64_TLSGD:
5258 case BFD_RELOC_X86_64_TLSLD:
5259 case BFD_RELOC_X86_64_GOTTPOFF:
5260 rel->addend = fixp->fx_offset - fixp->fx_size;
5261 break;
5262 default:
5263 rel->addend = (section->vma
5264 - fixp->fx_size
5265 + fixp->fx_addnumber
5266 + md_pcrel_from (fixp));
5267 break;
5268 }
5269 }
5270
5271 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
5272 if (rel->howto == NULL)
5273 {
5274 as_bad_where (fixp->fx_file, fixp->fx_line,
5275 _("cannot represent relocation type %s"),
5276 bfd_get_reloc_code_name (code));
5277 /* Set howto to a garbage value so that we can keep going. */
5278 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
5279 assert (rel->howto != NULL);
5280 }
5281
5282 return rel;
5283 }
5284
5285 #else /* !BFD_ASSEMBLER */
5286
5287 #if (defined(OBJ_AOUT) | defined(OBJ_BOUT))
5288 void
5289 tc_aout_fix_to_chars (where, fixP, segment_address_in_file)
5290 char *where;
5291 fixS *fixP;
5292 relax_addressT segment_address_in_file;
5293 {
5294 /* In: length of relocation (or of address) in chars: 1, 2 or 4.
5295 Out: GNU LD relocation length code: 0, 1, or 2. */
5296
5297 static const unsigned char nbytes_r_length[] = { 42, 0, 1, 42, 2 };
5298 long r_symbolnum;
5299
5300 know (fixP->fx_addsy != NULL);
5301
5302 md_number_to_chars (where,
5303 (valueT) (fixP->fx_frag->fr_address
5304 + fixP->fx_where - segment_address_in_file),
5305 4);
5306
5307 r_symbolnum = (S_IS_DEFINED (fixP->fx_addsy)
5308 ? S_GET_TYPE (fixP->fx_addsy)
5309 : fixP->fx_addsy->sy_number);
5310
5311 where[6] = (r_symbolnum >> 16) & 0x0ff;
5312 where[5] = (r_symbolnum >> 8) & 0x0ff;
5313 where[4] = r_symbolnum & 0x0ff;
5314 where[7] = ((((!S_IS_DEFINED (fixP->fx_addsy)) << 3) & 0x08)
5315 | ((nbytes_r_length[fixP->fx_size] << 1) & 0x06)
5316 | (((fixP->fx_pcrel << 0) & 0x01) & 0x0f));
5317 }
5318
5319 #endif /* OBJ_AOUT or OBJ_BOUT. */
5320
5321 #if defined (I386COFF)
5322
5323 short
5324 tc_coff_fix2rtype (fixP)
5325 fixS *fixP;
5326 {
5327 if (fixP->fx_r_type == R_IMAGEBASE)
5328 return R_IMAGEBASE;
5329
5330 return (fixP->fx_pcrel ?
5331 (fixP->fx_size == 1 ? R_PCRBYTE :
5332 fixP->fx_size == 2 ? R_PCRWORD :
5333 R_PCRLONG) :
5334 (fixP->fx_size == 1 ? R_RELBYTE :
5335 fixP->fx_size == 2 ? R_RELWORD :
5336 R_DIR32));
5337 }
5338
5339 int
5340 tc_coff_sizemachdep (frag)
5341 fragS *frag;
5342 {
5343 if (frag->fr_next)
5344 return (frag->fr_next->fr_address - frag->fr_address);
5345 else
5346 return 0;
5347 }
5348
5349 #endif /* I386COFF */
5350
5351 #endif /* !BFD_ASSEMBLER */
5352 \f
5353 /* Parse operands using Intel syntax. This implements a recursive descent
5354 parser based on the BNF grammar published in Appendix B of the MASM 6.1
5355 Programmer's Guide.
5356
5357 FIXME: We do not recognize the full operand grammar defined in the MASM
5358 documentation. In particular, all the structure/union and
5359 high-level macro operands are missing.
5360
5361 Uppercase words are terminals, lower case words are non-terminals.
5362 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
5363 bars '|' denote choices. Most grammar productions are implemented in
5364 functions called 'intel_<production>'.
5365
5366 Initial production is 'expr'.
5367
5368 addOp + | -
5369
5370 alpha [a-zA-Z]
5371
5372 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
5373
5374 constant digits [[ radixOverride ]]
5375
5376 dataType BYTE | WORD | DWORD | QWORD | XWORD
5377
5378 digits decdigit
5379 | digits decdigit
5380 | digits hexdigit
5381
5382 decdigit [0-9]
5383
5384 e05 e05 addOp e06
5385 | e06
5386
5387 e06 e06 mulOp e09
5388 | e09
5389
5390 e09 OFFSET e10
5391 | e09 PTR e10
5392 | e09 : e10
5393 | e10
5394
5395 e10 e10 [ expr ]
5396 | e11
5397
5398 e11 ( expr )
5399 | [ expr ]
5400 | constant
5401 | dataType
5402 | id
5403 | $
5404 | register
5405
5406 => expr SHORT e05
5407 | e05
5408
5409 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
5410 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
5411
5412 hexdigit a | b | c | d | e | f
5413 | A | B | C | D | E | F
5414
5415 id alpha
5416 | id alpha
5417 | id decdigit
5418
5419 mulOp * | / | MOD
5420
5421 quote " | '
5422
5423 register specialRegister
5424 | gpRegister
5425 | byteRegister
5426
5427 segmentRegister CS | DS | ES | FS | GS | SS
5428
5429 specialRegister CR0 | CR2 | CR3
5430 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
5431 | TR3 | TR4 | TR5 | TR6 | TR7
5432
5433 We simplify the grammar in obvious places (e.g., register parsing is
5434 done by calling parse_register) and eliminate immediate left recursion
5435 to implement a recursive-descent parser.
5436
5437 expr SHORT e05
5438 | e05
5439
5440 e05 e06 e05'
5441
5442 e05' addOp e06 e05'
5443 | Empty
5444
5445 e06 e09 e06'
5446
5447 e06' mulOp e09 e06'
5448 | Empty
5449
5450 e09 OFFSET e10 e09'
5451 | e10 e09'
5452
5453 e09' PTR e10 e09'
5454 | : e10 e09'
5455 | Empty
5456
5457 e10 e11 e10'
5458
5459 e10' [ expr ] e10'
5460 | Empty
5461
5462 e11 ( expr )
5463 | [ expr ]
5464 | BYTE
5465 | WORD
5466 | DWORD
5467 | QWORD
5468 | XWORD
5469 | .
5470 | $
5471 | register
5472 | id
5473 | constant */
5474
5475 /* Parsing structure for the intel syntax parser. Used to implement the
5476 semantic actions for the operand grammar. */
5477 struct intel_parser_s
5478 {
5479 char *op_string; /* The string being parsed. */
5480 int got_a_float; /* Whether the operand is a float. */
5481 int op_modifier; /* Operand modifier. */
5482 int is_mem; /* 1 if operand is memory reference. */
5483 const reg_entry *reg; /* Last register reference found. */
5484 char *disp; /* Displacement string being built. */
5485 };
5486
5487 static struct intel_parser_s intel_parser;
5488
5489 /* Token structure for parsing intel syntax. */
5490 struct intel_token
5491 {
5492 int code; /* Token code. */
5493 const reg_entry *reg; /* Register entry for register tokens. */
5494 char *str; /* String representation. */
5495 };
5496
5497 static struct intel_token cur_token, prev_token;
5498
5499 /* Token codes for the intel parser. Since T_SHORT is already used
5500 by COFF, undefine it first to prevent a warning. */
5501 #define T_NIL -1
5502 #define T_CONST 1
5503 #define T_REG 2
5504 #define T_BYTE 3
5505 #define T_WORD 4
5506 #define T_DWORD 5
5507 #define T_QWORD 6
5508 #define T_XWORD 7
5509 #undef T_SHORT
5510 #define T_SHORT 8
5511 #define T_OFFSET 9
5512 #define T_PTR 10
5513 #define T_ID 11
5514
5515 /* Prototypes for intel parser functions. */
5516 static int intel_match_token PARAMS ((int code));
5517 static void intel_get_token PARAMS ((void));
5518 static void intel_putback_token PARAMS ((void));
5519 static int intel_expr PARAMS ((void));
5520 static int intel_e05 PARAMS ((void));
5521 static int intel_e05_1 PARAMS ((void));
5522 static int intel_e06 PARAMS ((void));
5523 static int intel_e06_1 PARAMS ((void));
5524 static int intel_e09 PARAMS ((void));
5525 static int intel_e09_1 PARAMS ((void));
5526 static int intel_e10 PARAMS ((void));
5527 static int intel_e10_1 PARAMS ((void));
5528 static int intel_e11 PARAMS ((void));
5529
5530 static int
5531 i386_intel_operand (operand_string, got_a_float)
5532 char *operand_string;
5533 int got_a_float;
5534 {
5535 int ret;
5536 char *p;
5537
5538 /* Initialize token holders. */
5539 cur_token.code = prev_token.code = T_NIL;
5540 cur_token.reg = prev_token.reg = NULL;
5541 cur_token.str = prev_token.str = NULL;
5542
5543 /* Initialize parser structure. */
5544 p = intel_parser.op_string = (char *) malloc (strlen (operand_string) + 1);
5545 if (p == NULL)
5546 abort ();
5547 strcpy (intel_parser.op_string, operand_string);
5548 intel_parser.got_a_float = got_a_float;
5549 intel_parser.op_modifier = -1;
5550 intel_parser.is_mem = 0;
5551 intel_parser.reg = NULL;
5552 intel_parser.disp = (char *) malloc (strlen (operand_string) + 1);
5553 if (intel_parser.disp == NULL)
5554 abort ();
5555 intel_parser.disp[0] = '\0';
5556
5557 /* Read the first token and start the parser. */
5558 intel_get_token ();
5559 ret = intel_expr ();
5560
5561 if (ret)
5562 {
5563 /* If we found a memory reference, hand it over to i386_displacement
5564 to fill in the rest of the operand fields. */
5565 if (intel_parser.is_mem)
5566 {
5567 if ((i.mem_operands == 1
5568 && (current_templates->start->opcode_modifier & IsString) == 0)
5569 || i.mem_operands == 2)
5570 {
5571 as_bad (_("too many memory references for '%s'"),
5572 current_templates->start->name);
5573 ret = 0;
5574 }
5575 else
5576 {
5577 char *s = intel_parser.disp;
5578 i.mem_operands++;
5579
5580 /* Add the displacement expression. */
5581 if (*s != '\0')
5582 ret = i386_displacement (s, s + strlen (s))
5583 && i386_index_check (s);
5584 }
5585 }
5586
5587 /* Constant and OFFSET expressions are handled by i386_immediate. */
5588 else if (intel_parser.op_modifier == OFFSET_FLAT
5589 || intel_parser.reg == NULL)
5590 ret = i386_immediate (intel_parser.disp);
5591 }
5592
5593 free (p);
5594 free (intel_parser.disp);
5595
5596 return ret;
5597 }
5598
5599 /* expr SHORT e05
5600 | e05 */
5601 static int
5602 intel_expr ()
5603 {
5604 /* expr SHORT e05 */
5605 if (cur_token.code == T_SHORT)
5606 {
5607 intel_parser.op_modifier = SHORT;
5608 intel_match_token (T_SHORT);
5609
5610 return (intel_e05 ());
5611 }
5612
5613 /* expr e05 */
5614 else
5615 return intel_e05 ();
5616 }
5617
5618 /* e05 e06 e05'
5619
5620 e05' addOp e06 e05'
5621 | Empty */
5622 static int
5623 intel_e05 ()
5624 {
5625 return (intel_e06 () && intel_e05_1 ());
5626 }
5627
5628 static int
5629 intel_e05_1 ()
5630 {
5631 /* e05' addOp e06 e05' */
5632 if (cur_token.code == '+' || cur_token.code == '-')
5633 {
5634 strcat (intel_parser.disp, cur_token.str);
5635 intel_match_token (cur_token.code);
5636
5637 return (intel_e06 () && intel_e05_1 ());
5638 }
5639
5640 /* e05' Empty */
5641 else
5642 return 1;
5643 }
5644
5645 /* e06 e09 e06'
5646
5647 e06' mulOp e09 e06'
5648 | Empty */
5649 static int
5650 intel_e06 ()
5651 {
5652 return (intel_e09 () && intel_e06_1 ());
5653 }
5654
5655 static int
5656 intel_e06_1 ()
5657 {
5658 /* e06' mulOp e09 e06' */
5659 if (cur_token.code == '*' || cur_token.code == '/')
5660 {
5661 strcat (intel_parser.disp, cur_token.str);
5662 intel_match_token (cur_token.code);
5663
5664 return (intel_e09 () && intel_e06_1 ());
5665 }
5666
5667 /* e06' Empty */
5668 else
5669 return 1;
5670 }
5671
5672 /* e09 OFFSET e10 e09'
5673 | e10 e09'
5674
5675 e09' PTR e10 e09'
5676 | : e10 e09'
5677 | Empty */
5678 static int
5679 intel_e09 ()
5680 {
5681 /* e09 OFFSET e10 e09' */
5682 if (cur_token.code == T_OFFSET)
5683 {
5684 intel_parser.is_mem = 0;
5685 intel_parser.op_modifier = OFFSET_FLAT;
5686 intel_match_token (T_OFFSET);
5687
5688 return (intel_e10 () && intel_e09_1 ());
5689 }
5690
5691 /* e09 e10 e09' */
5692 else
5693 return (intel_e10 () && intel_e09_1 ());
5694 }
5695
5696 static int
5697 intel_e09_1 ()
5698 {
5699 /* e09' PTR e10 e09' */
5700 if (cur_token.code == T_PTR)
5701 {
5702 if (prev_token.code == T_BYTE)
5703 i.suffix = BYTE_MNEM_SUFFIX;
5704
5705 else if (prev_token.code == T_WORD)
5706 {
5707 if (intel_parser.got_a_float == 2) /* "fi..." */
5708 i.suffix = SHORT_MNEM_SUFFIX;
5709 else
5710 i.suffix = WORD_MNEM_SUFFIX;
5711 }
5712
5713 else if (prev_token.code == T_DWORD)
5714 {
5715 if (intel_parser.got_a_float == 1) /* "f..." */
5716 i.suffix = SHORT_MNEM_SUFFIX;
5717 else
5718 i.suffix = LONG_MNEM_SUFFIX;
5719 }
5720
5721 else if (prev_token.code == T_QWORD)
5722 {
5723 if (intel_parser.got_a_float == 1) /* "f..." */
5724 i.suffix = LONG_MNEM_SUFFIX;
5725 else
5726 i.suffix = QWORD_MNEM_SUFFIX;
5727 }
5728
5729 else if (prev_token.code == T_XWORD)
5730 i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
5731
5732 else
5733 {
5734 as_bad (_("Unknown operand modifier `%s'\n"), prev_token.str);
5735 return 0;
5736 }
5737
5738 intel_match_token (T_PTR);
5739
5740 return (intel_e10 () && intel_e09_1 ());
5741 }
5742
5743 /* e09 : e10 e09' */
5744 else if (cur_token.code == ':')
5745 {
5746 /* Mark as a memory operand only if it's not already known to be an
5747 offset expression. */
5748 if (intel_parser.op_modifier != OFFSET_FLAT)
5749 intel_parser.is_mem = 1;
5750
5751 return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
5752 }
5753
5754 /* e09' Empty */
5755 else
5756 return 1;
5757 }
5758
5759 /* e10 e11 e10'
5760
5761 e10' [ expr ] e10'
5762 | Empty */
5763 static int
5764 intel_e10 ()
5765 {
5766 return (intel_e11 () && intel_e10_1 ());
5767 }
5768
5769 static int
5770 intel_e10_1 ()
5771 {
5772 /* e10' [ expr ] e10' */
5773 if (cur_token.code == '[')
5774 {
5775 intel_match_token ('[');
5776
5777 /* Mark as a memory operand only if it's not already known to be an
5778 offset expression. If it's an offset expression, we need to keep
5779 the brace in. */
5780 if (intel_parser.op_modifier != OFFSET_FLAT)
5781 intel_parser.is_mem = 1;
5782 else
5783 strcat (intel_parser.disp, "[");
5784
5785 /* Add a '+' to the displacement string if necessary. */
5786 if (*intel_parser.disp != '\0'
5787 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5788 strcat (intel_parser.disp, "+");
5789
5790 if (intel_expr () && intel_match_token (']'))
5791 {
5792 /* Preserve brackets when the operand is an offset expression. */
5793 if (intel_parser.op_modifier == OFFSET_FLAT)
5794 strcat (intel_parser.disp, "]");
5795
5796 return intel_e10_1 ();
5797 }
5798 else
5799 return 0;
5800 }
5801
5802 /* e10' Empty */
5803 else
5804 return 1;
5805 }
5806
5807 /* e11 ( expr )
5808 | [ expr ]
5809 | BYTE
5810 | WORD
5811 | DWORD
5812 | QWORD
5813 | XWORD
5814 | $
5815 | .
5816 | register
5817 | id
5818 | constant */
5819 static int
5820 intel_e11 ()
5821 {
5822 /* e11 ( expr ) */
5823 if (cur_token.code == '(')
5824 {
5825 intel_match_token ('(');
5826 strcat (intel_parser.disp, "(");
5827
5828 if (intel_expr () && intel_match_token (')'))
5829 {
5830 strcat (intel_parser.disp, ")");
5831 return 1;
5832 }
5833 else
5834 return 0;
5835 }
5836
5837 /* e11 [ expr ] */
5838 else if (cur_token.code == '[')
5839 {
5840 intel_match_token ('[');
5841
5842 /* Mark as a memory operand only if it's not already known to be an
5843 offset expression. If it's an offset expression, we need to keep
5844 the brace in. */
5845 if (intel_parser.op_modifier != OFFSET_FLAT)
5846 intel_parser.is_mem = 1;
5847 else
5848 strcat (intel_parser.disp, "[");
5849
5850 /* Operands for jump/call inside brackets denote absolute addresses. */
5851 if (current_templates->start->opcode_modifier & Jump
5852 || current_templates->start->opcode_modifier & JumpDword
5853 || current_templates->start->opcode_modifier & JumpByte
5854 || current_templates->start->opcode_modifier & JumpInterSegment)
5855 i.types[this_operand] |= JumpAbsolute;
5856
5857 /* Add a '+' to the displacement string if necessary. */
5858 if (*intel_parser.disp != '\0'
5859 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5860 strcat (intel_parser.disp, "+");
5861
5862 if (intel_expr () && intel_match_token (']'))
5863 {
5864 /* Preserve brackets when the operand is an offset expression. */
5865 if (intel_parser.op_modifier == OFFSET_FLAT)
5866 strcat (intel_parser.disp, "]");
5867
5868 return 1;
5869 }
5870 else
5871 return 0;
5872 }
5873
5874 /* e11 BYTE
5875 | WORD
5876 | DWORD
5877 | QWORD
5878 | XWORD */
5879 else if (cur_token.code == T_BYTE
5880 || cur_token.code == T_WORD
5881 || cur_token.code == T_DWORD
5882 || cur_token.code == T_QWORD
5883 || cur_token.code == T_XWORD)
5884 {
5885 intel_match_token (cur_token.code);
5886
5887 return 1;
5888 }
5889
5890 /* e11 $
5891 | . */
5892 else if (cur_token.code == '$' || cur_token.code == '.')
5893 {
5894 strcat (intel_parser.disp, cur_token.str);
5895 intel_match_token (cur_token.code);
5896
5897 /* Mark as a memory operand only if it's not already known to be an
5898 offset expression. */
5899 if (intel_parser.op_modifier != OFFSET_FLAT)
5900 intel_parser.is_mem = 1;
5901
5902 return 1;
5903 }
5904
5905 /* e11 register */
5906 else if (cur_token.code == T_REG)
5907 {
5908 const reg_entry *reg = intel_parser.reg = cur_token.reg;
5909
5910 intel_match_token (T_REG);
5911
5912 /* Check for segment change. */
5913 if (cur_token.code == ':')
5914 {
5915 if (reg->reg_type & (SReg2 | SReg3))
5916 {
5917 switch (reg->reg_num)
5918 {
5919 case 0:
5920 i.seg[i.mem_operands] = &es;
5921 break;
5922 case 1:
5923 i.seg[i.mem_operands] = &cs;
5924 break;
5925 case 2:
5926 i.seg[i.mem_operands] = &ss;
5927 break;
5928 case 3:
5929 i.seg[i.mem_operands] = &ds;
5930 break;
5931 case 4:
5932 i.seg[i.mem_operands] = &fs;
5933 break;
5934 case 5:
5935 i.seg[i.mem_operands] = &gs;
5936 break;
5937 }
5938 }
5939 else
5940 {
5941 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
5942 return 0;
5943 }
5944 }
5945
5946 /* Not a segment register. Check for register scaling. */
5947 else if (cur_token.code == '*')
5948 {
5949 if (!intel_parser.is_mem)
5950 {
5951 as_bad (_("Register scaling only allowed in memory operands."));
5952 return 0;
5953 }
5954
5955 /* What follows must be a valid scale. */
5956 if (intel_match_token ('*')
5957 && strchr ("01248", *cur_token.str))
5958 {
5959 i.index_reg = reg;
5960 i.types[this_operand] |= BaseIndex;
5961
5962 /* Set the scale after setting the register (otherwise,
5963 i386_scale will complain) */
5964 i386_scale (cur_token.str);
5965 intel_match_token (T_CONST);
5966 }
5967 else
5968 {
5969 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
5970 cur_token.str);
5971 return 0;
5972 }
5973 }
5974
5975 /* No scaling. If this is a memory operand, the register is either a
5976 base register (first occurrence) or an index register (second
5977 occurrence). */
5978 else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3)))
5979 {
5980 if (i.base_reg && i.index_reg)
5981 {
5982 as_bad (_("Too many register references in memory operand.\n"));
5983 return 0;
5984 }
5985
5986 if (i.base_reg == NULL)
5987 i.base_reg = reg;
5988 else
5989 i.index_reg = reg;
5990
5991 i.types[this_operand] |= BaseIndex;
5992 }
5993
5994 /* Offset modifier. Add the register to the displacement string to be
5995 parsed as an immediate expression after we're done. */
5996 else if (intel_parser.op_modifier == OFFSET_FLAT)
5997 strcat (intel_parser.disp, reg->reg_name);
5998
5999 /* It's neither base nor index nor offset. */
6000 else
6001 {
6002 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
6003 i.op[this_operand].regs = reg;
6004 i.reg_operands++;
6005 }
6006
6007 /* Since registers are not part of the displacement string (except
6008 when we're parsing offset operands), we may need to remove any
6009 preceding '+' from the displacement string. */
6010 if (*intel_parser.disp != '\0'
6011 && intel_parser.op_modifier != OFFSET_FLAT)
6012 {
6013 char *s = intel_parser.disp;
6014 s += strlen (s) - 1;
6015 if (*s == '+')
6016 *s = '\0';
6017 }
6018
6019 return 1;
6020 }
6021
6022 /* e11 id */
6023 else if (cur_token.code == T_ID)
6024 {
6025 /* Add the identifier to the displacement string. */
6026 strcat (intel_parser.disp, cur_token.str);
6027 intel_match_token (T_ID);
6028
6029 /* The identifier represents a memory reference only if it's not
6030 preceded by an offset modifier. */
6031 if (intel_parser.op_modifier != OFFSET_FLAT)
6032 intel_parser.is_mem = 1;
6033
6034 return 1;
6035 }
6036
6037 /* e11 constant */
6038 else if (cur_token.code == T_CONST
6039 || cur_token.code == '-'
6040 || cur_token.code == '+')
6041 {
6042 char *save_str;
6043
6044 /* Allow constants that start with `+' or `-'. */
6045 if (cur_token.code == '-' || cur_token.code == '+')
6046 {
6047 strcat (intel_parser.disp, cur_token.str);
6048 intel_match_token (cur_token.code);
6049 if (cur_token.code != T_CONST)
6050 {
6051 as_bad (_("Syntax error. Expecting a constant. Got `%s'.\n"),
6052 cur_token.str);
6053 return 0;
6054 }
6055 }
6056
6057 save_str = (char *) malloc (strlen (cur_token.str) + 1);
6058 if (save_str == NULL)
6059 abort ();
6060 strcpy (save_str, cur_token.str);
6061
6062 /* Get the next token to check for register scaling. */
6063 intel_match_token (cur_token.code);
6064
6065 /* Check if this constant is a scaling factor for an index register. */
6066 if (cur_token.code == '*')
6067 {
6068 if (intel_match_token ('*') && cur_token.code == T_REG)
6069 {
6070 if (!intel_parser.is_mem)
6071 {
6072 as_bad (_("Register scaling only allowed in memory operands."));
6073 return 0;
6074 }
6075
6076 /* The constant is followed by `* reg', so it must be
6077 a valid scale. */
6078 if (strchr ("01248", *save_str))
6079 {
6080 i.index_reg = cur_token.reg;
6081 i.types[this_operand] |= BaseIndex;
6082
6083 /* Set the scale after setting the register (otherwise,
6084 i386_scale will complain) */
6085 i386_scale (save_str);
6086 intel_match_token (T_REG);
6087
6088 /* Since registers are not part of the displacement
6089 string, we may need to remove any preceding '+' from
6090 the displacement string. */
6091 if (*intel_parser.disp != '\0')
6092 {
6093 char *s = intel_parser.disp;
6094 s += strlen (s) - 1;
6095 if (*s == '+')
6096 *s = '\0';
6097 }
6098
6099 free (save_str);
6100
6101 return 1;
6102 }
6103 else
6104 return 0;
6105 }
6106
6107 /* The constant was not used for register scaling. Since we have
6108 already consumed the token following `*' we now need to put it
6109 back in the stream. */
6110 else
6111 intel_putback_token ();
6112 }
6113
6114 /* Add the constant to the displacement string. */
6115 strcat (intel_parser.disp, save_str);
6116 free (save_str);
6117
6118 return 1;
6119 }
6120
6121 as_bad (_("Unrecognized token '%s'"), cur_token.str);
6122 return 0;
6123 }
6124
6125 /* Match the given token against cur_token. If they match, read the next
6126 token from the operand string. */
6127 static int
6128 intel_match_token (code)
6129 int code;
6130 {
6131 if (cur_token.code == code)
6132 {
6133 intel_get_token ();
6134 return 1;
6135 }
6136 else
6137 {
6138 as_bad (_("Unexpected token `%s'\n"), cur_token.str);
6139 return 0;
6140 }
6141 }
6142
6143 /* Read a new token from intel_parser.op_string and store it in cur_token. */
6144 static void
6145 intel_get_token ()
6146 {
6147 char *end_op;
6148 const reg_entry *reg;
6149 struct intel_token new_token;
6150
6151 new_token.code = T_NIL;
6152 new_token.reg = NULL;
6153 new_token.str = NULL;
6154
6155 /* Free the memory allocated to the previous token and move
6156 cur_token to prev_token. */
6157 if (prev_token.str)
6158 free (prev_token.str);
6159
6160 prev_token = cur_token;
6161
6162 /* Skip whitespace. */
6163 while (is_space_char (*intel_parser.op_string))
6164 intel_parser.op_string++;
6165
6166 /* Return an empty token if we find nothing else on the line. */
6167 if (*intel_parser.op_string == '\0')
6168 {
6169 cur_token = new_token;
6170 return;
6171 }
6172
6173 /* The new token cannot be larger than the remainder of the operand
6174 string. */
6175 new_token.str = (char *) malloc (strlen (intel_parser.op_string) + 1);
6176 if (new_token.str == NULL)
6177 abort ();
6178 new_token.str[0] = '\0';
6179
6180 if (strchr ("0123456789", *intel_parser.op_string))
6181 {
6182 char *p = new_token.str;
6183 char *q = intel_parser.op_string;
6184 new_token.code = T_CONST;
6185
6186 /* Allow any kind of identifier char to encompass floating point and
6187 hexadecimal numbers. */
6188 while (is_identifier_char (*q))
6189 *p++ = *q++;
6190 *p = '\0';
6191
6192 /* Recognize special symbol names [0-9][bf]. */
6193 if (strlen (intel_parser.op_string) == 2
6194 && (intel_parser.op_string[1] == 'b'
6195 || intel_parser.op_string[1] == 'f'))
6196 new_token.code = T_ID;
6197 }
6198
6199 else if (strchr ("+-/*:[]()", *intel_parser.op_string))
6200 {
6201 new_token.code = *intel_parser.op_string;
6202 new_token.str[0] = *intel_parser.op_string;
6203 new_token.str[1] = '\0';
6204 }
6205
6206 else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
6207 && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
6208 {
6209 new_token.code = T_REG;
6210 new_token.reg = reg;
6211
6212 if (*intel_parser.op_string == REGISTER_PREFIX)
6213 {
6214 new_token.str[0] = REGISTER_PREFIX;
6215 new_token.str[1] = '\0';
6216 }
6217
6218 strcat (new_token.str, reg->reg_name);
6219 }
6220
6221 else if (is_identifier_char (*intel_parser.op_string))
6222 {
6223 char *p = new_token.str;
6224 char *q = intel_parser.op_string;
6225
6226 /* A '.' or '$' followed by an identifier char is an identifier.
6227 Otherwise, it's operator '.' followed by an expression. */
6228 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
6229 {
6230 new_token.code = *q;
6231 new_token.str[0] = *q;
6232 new_token.str[1] = '\0';
6233 }
6234 else
6235 {
6236 while (is_identifier_char (*q) || *q == '@')
6237 *p++ = *q++;
6238 *p = '\0';
6239
6240 if (strcasecmp (new_token.str, "BYTE") == 0)
6241 new_token.code = T_BYTE;
6242
6243 else if (strcasecmp (new_token.str, "WORD") == 0)
6244 new_token.code = T_WORD;
6245
6246 else if (strcasecmp (new_token.str, "DWORD") == 0)
6247 new_token.code = T_DWORD;
6248
6249 else if (strcasecmp (new_token.str, "QWORD") == 0)
6250 new_token.code = T_QWORD;
6251
6252 else if (strcasecmp (new_token.str, "XWORD") == 0)
6253 new_token.code = T_XWORD;
6254
6255 else if (strcasecmp (new_token.str, "PTR") == 0)
6256 new_token.code = T_PTR;
6257
6258 else if (strcasecmp (new_token.str, "SHORT") == 0)
6259 new_token.code = T_SHORT;
6260
6261 else if (strcasecmp (new_token.str, "OFFSET") == 0)
6262 {
6263 new_token.code = T_OFFSET;
6264
6265 /* ??? This is not mentioned in the MASM grammar but gcc
6266 makes use of it with -mintel-syntax. OFFSET may be
6267 followed by FLAT: */
6268 if (strncasecmp (q, " FLAT:", 6) == 0)
6269 strcat (new_token.str, " FLAT:");
6270 }
6271
6272 /* ??? This is not mentioned in the MASM grammar. */
6273 else if (strcasecmp (new_token.str, "FLAT") == 0)
6274 new_token.code = T_OFFSET;
6275
6276 else
6277 new_token.code = T_ID;
6278 }
6279 }
6280
6281 else
6282 as_bad (_("Unrecognized token `%s'\n"), intel_parser.op_string);
6283
6284 intel_parser.op_string += strlen (new_token.str);
6285 cur_token = new_token;
6286 }
6287
6288 /* Put cur_token back into the token stream and make cur_token point to
6289 prev_token. */
6290 static void
6291 intel_putback_token ()
6292 {
6293 intel_parser.op_string -= strlen (cur_token.str);
6294 free (cur_token.str);
6295 cur_token = prev_token;
6296
6297 /* Forget prev_token. */
6298 prev_token.code = T_NIL;
6299 prev_token.reg = NULL;
6300 prev_token.str = NULL;
6301 }
This page took 0.165255 seconds and 3 git commands to generate.