2001-02-28 Andreas Jaeger <aj@suse.de>, Bo Thorsen <bo@suse.de>
[binutils-gdb.git] / gas / config / tc-i386.c
1 /* i386.c -- Assemble code for the Intel 80386
2 Copyright (C) 1989, 91, 92, 93, 94, 95, 96, 97, 98, 99, 2000, 2001
3 Free Software Foundation, Inc.
4
5 This file is part of GAS, the GNU Assembler.
6
7 GAS is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 2, or (at your option)
10 any later version.
11
12 GAS is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
16
17 You should have received a copy of the GNU General Public License
18 along with GAS; see the file COPYING. If not, write to the Free
19 Software Foundation, 59 Temple Place - Suite 330, Boston, MA
20 02111-1307, USA. */
21
22 /* Intel 80386 machine specific gas.
23 Written by Eliot Dresselhaus (eliot@mgm.mit.edu).
24 x86_64 support by Jan Hubicka (jh@suse.cz)
25 Bugs & suggestions are completely welcome. This is free software.
26 Please help us make it better. */
27
28 #include <ctype.h>
29
30 #include "as.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 #define true 1
52 #define false 0
53
54 static unsigned int mode_from_disp_size PARAMS ((unsigned int));
55 static int fits_in_signed_byte PARAMS ((offsetT));
56 static int fits_in_unsigned_byte PARAMS ((offsetT));
57 static int fits_in_unsigned_word PARAMS ((offsetT));
58 static int fits_in_signed_word PARAMS ((offsetT));
59 static int fits_in_unsigned_long PARAMS ((offsetT));
60 static int fits_in_signed_long PARAMS ((offsetT));
61 static int smallest_imm_type PARAMS ((offsetT));
62 static offsetT offset_in_range PARAMS ((offsetT, int));
63 static int add_prefix PARAMS ((unsigned int));
64 static void set_code_flag PARAMS ((int));
65 static void set_16bit_gcc_code_flag PARAMS ((int));
66 static void set_intel_syntax PARAMS ((int));
67 static void set_cpu_arch PARAMS ((int));
68
69 #ifdef BFD_ASSEMBLER
70 static bfd_reloc_code_real_type reloc
71 PARAMS ((int, int, int, bfd_reloc_code_real_type));
72 #endif
73
74 #ifndef DEFAULT_ARCH
75 #define DEFAULT_ARCH "i386"
76 #endif
77 static char *default_arch = DEFAULT_ARCH;
78
79 /* 'md_assemble ()' gathers together information and puts it into a
80 i386_insn. */
81
82 union i386_op
83 {
84 expressionS *disps;
85 expressionS *imms;
86 const reg_entry *regs;
87 };
88
89 struct _i386_insn
90 {
91 /* TM holds the template for the insn were currently assembling. */
92 template tm;
93
94 /* SUFFIX holds the instruction mnemonic suffix if given.
95 (e.g. 'l' for 'movl') */
96 char suffix;
97
98 /* OPERANDS gives the number of given operands. */
99 unsigned int operands;
100
101 /* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number
102 of given register, displacement, memory operands and immediate
103 operands. */
104 unsigned int reg_operands, disp_operands, mem_operands, imm_operands;
105
106 /* TYPES [i] is the type (see above #defines) which tells us how to
107 use OP[i] for the corresponding operand. */
108 unsigned int types[MAX_OPERANDS];
109
110 /* Displacement expression, immediate expression, or register for each
111 operand. */
112 union i386_op op[MAX_OPERANDS];
113
114 /* Flags for operands. */
115 unsigned int flags[MAX_OPERANDS];
116 #define Operand_PCrel 1
117
118 /* Relocation type for operand */
119 #ifdef BFD_ASSEMBLER
120 enum bfd_reloc_code_real disp_reloc[MAX_OPERANDS];
121 #else
122 int disp_reloc[MAX_OPERANDS];
123 #endif
124
125 /* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode
126 the base index byte below. */
127 const reg_entry *base_reg;
128 const reg_entry *index_reg;
129 unsigned int log2_scale_factor;
130
131 /* SEG gives the seg_entries of this insn. They are zero unless
132 explicit segment overrides are given. */
133 const seg_entry *seg[2];
134
135 /* PREFIX holds all the given prefix opcodes (usually null).
136 PREFIXES is the number of prefix opcodes. */
137 unsigned int prefixes;
138 unsigned char prefix[MAX_PREFIXES];
139
140 /* RM and SIB are the modrm byte and the sib byte where the
141 addressing modes of this insn are encoded. */
142
143 modrm_byte rm;
144 rex_byte rex;
145 sib_byte sib;
146 };
147
148 typedef struct _i386_insn i386_insn;
149
150 /* List of chars besides those in app.c:symbol_chars that can start an
151 operand. Used to prevent the scrubber eating vital white-space. */
152 #ifdef LEX_AT
153 const char extra_symbol_chars[] = "*%-(@";
154 #else
155 const char extra_symbol_chars[] = "*%-(";
156 #endif
157
158 /* This array holds the chars that always start a comment. If the
159 pre-processor is disabled, these aren't very useful. */
160 #if defined (TE_I386AIX) || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) && ! defined (TE_LINUX) && !defined(TE_FreeBSD))
161 /* Putting '/' here makes it impossible to use the divide operator.
162 However, we need it for compatibility with SVR4 systems. */
163 const char comment_chars[] = "#/";
164 #define PREFIX_SEPARATOR '\\'
165 #else
166 const char comment_chars[] = "#";
167 #define PREFIX_SEPARATOR '/'
168 #endif
169
170 /* This array holds the chars that only start a comment at the beginning of
171 a line. If the line seems to have the form '# 123 filename'
172 .line and .file directives will appear in the pre-processed output.
173 Note that input_file.c hand checks for '#' at the beginning of the
174 first line of the input file. This is because the compiler outputs
175 #NO_APP at the beginning of its output.
176 Also note that comments started like this one will always work if
177 '/' isn't otherwise defined. */
178 #if defined (TE_I386AIX) || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) && ! defined (TE_LINUX) && !defined(TE_FreeBSD))
179 const char line_comment_chars[] = "";
180 #else
181 const char line_comment_chars[] = "/";
182 #endif
183
184 const char line_separator_chars[] = ";";
185
186 /* Chars that can be used to separate mant from exp in floating point
187 nums. */
188 const char EXP_CHARS[] = "eE";
189
190 /* Chars that mean this number is a floating point constant
191 As in 0f12.456
192 or 0d1.2345e12. */
193 const char FLT_CHARS[] = "fFdDxX";
194
195 /* Tables for lexical analysis. */
196 static char mnemonic_chars[256];
197 static char register_chars[256];
198 static char operand_chars[256];
199 static char identifier_chars[256];
200 static char digit_chars[256];
201
202 /* Lexical macros. */
203 #define is_mnemonic_char(x) (mnemonic_chars[(unsigned char) x])
204 #define is_operand_char(x) (operand_chars[(unsigned char) x])
205 #define is_register_char(x) (register_chars[(unsigned char) x])
206 #define is_space_char(x) ((x) == ' ')
207 #define is_identifier_char(x) (identifier_chars[(unsigned char) x])
208 #define is_digit_char(x) (digit_chars[(unsigned char) x])
209
210 /* All non-digit non-letter charcters that may occur in an operand. */
211 static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
212
213 /* md_assemble() always leaves the strings it's passed unaltered. To
214 effect this we maintain a stack of saved characters that we've smashed
215 with '\0's (indicating end of strings for various sub-fields of the
216 assembler instruction). */
217 static char save_stack[32];
218 static char *save_stack_p;
219 #define END_STRING_AND_SAVE(s) \
220 do { *save_stack_p++ = *(s); *(s) = '\0'; } while (0)
221 #define RESTORE_END_STRING(s) \
222 do { *(s) = *--save_stack_p; } while (0)
223
224 /* The instruction we're assembling. */
225 static i386_insn i;
226
227 /* Possible templates for current insn. */
228 static const templates *current_templates;
229
230 /* Per instruction expressionS buffers: 2 displacements & 2 immediate max. */
231 static expressionS disp_expressions[2], im_expressions[2];
232
233 /* Current operand we are working on. */
234 static int this_operand;
235
236 /* We support four different modes. FLAG_CODE variable is used to distinguish
237 these. */
238
239 enum flag_code {
240 CODE_32BIT,
241 CODE_16BIT,
242 CODE_64BIT };
243
244 static enum flag_code flag_code;
245 static int use_rela_relocations = 0;
246
247 /* The names used to print error messages. */
248 static const char *flag_code_names[] =
249 {
250 "32",
251 "16",
252 "64"
253 };
254
255 /* 1 for intel syntax,
256 0 if att syntax. */
257 static int intel_syntax = 0;
258
259 /* 1 if register prefix % not required. */
260 static int allow_naked_reg = 0;
261
262 /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
263 leave, push, and pop instructions so that gcc has the same stack
264 frame as in 32 bit mode. */
265 static char stackop_size = '\0';
266
267 /* Non-zero to quieten some warnings. */
268 static int quiet_warnings = 0;
269
270 /* CPU name. */
271 static const char *cpu_arch_name = NULL;
272
273 /* CPU feature flags. */
274 static unsigned int cpu_arch_flags = CpuUnknownFlags|CpuNo64;
275
276 /* If set, conditional jumps are not automatically promoted to handle
277 larger than a byte offset. */
278 static unsigned int no_cond_jump_promotion = 0;
279
280 /* Interface to relax_segment.
281 There are 3 major relax states for 386 jump insns because the
282 different types of jumps add different sizes to frags when we're
283 figuring out what sort of jump to choose to reach a given label. */
284
285 /* Types. */
286 #define UNCOND_JUMP 1
287 #define COND_JUMP 2
288 #define COND_JUMP86 3
289
290 /* Sizes. */
291 #define CODE16 1
292 #define SMALL 0
293 #define SMALL16 (SMALL|CODE16)
294 #define BIG 2
295 #define BIG16 (BIG|CODE16)
296
297 #ifndef INLINE
298 #ifdef __GNUC__
299 #define INLINE __inline__
300 #else
301 #define INLINE
302 #endif
303 #endif
304
305 #define ENCODE_RELAX_STATE(type, size) \
306 ((relax_substateT) (((type) << 2) | (size)))
307 #define TYPE_FROM_RELAX_STATE(s) \
308 ((s) >> 2)
309 #define DISP_SIZE_FROM_RELAX_STATE(s) \
310 ((((s) & 3) == BIG ? 4 : (((s) & 3) == BIG16 ? 2 : 1)))
311
312 /* This table is used by relax_frag to promote short jumps to long
313 ones where necessary. SMALL (short) jumps may be promoted to BIG
314 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
315 don't allow a short jump in a 32 bit code segment to be promoted to
316 a 16 bit offset jump because it's slower (requires data size
317 prefix), and doesn't work, unless the destination is in the bottom
318 64k of the code segment (The top 16 bits of eip are zeroed). */
319
320 const relax_typeS md_relax_table[] =
321 {
322 /* The fields are:
323 1) most positive reach of this state,
324 2) most negative reach of this state,
325 3) how many bytes this mode will add to the size of the current frag
326 4) which index into the table to try if we can't fit into this one. */
327 {1, 1, 0, 0},
328 {1, 1, 0, 0},
329 {1, 1, 0, 0},
330 {1, 1, 0, 0},
331
332 /* UNCOND_JUMP states. */
333 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
334 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
335 /* dword jmp adds 3 bytes to frag:
336 0 extra opcode bytes, 3 extra displacement bytes. */
337 {0, 0, 3, 0},
338 /* word jmp adds 1 byte to frag:
339 0 extra opcode bytes, 1 extra displacement byte. */
340 {0, 0, 1, 0},
341
342 /* COND_JUMP states. */
343 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
344 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
345 /* dword conditionals adds 4 bytes to frag:
346 1 extra opcode byte, 3 extra displacement bytes. */
347 {0, 0, 4, 0},
348 /* word conditionals add 2 bytes to frag:
349 1 extra opcode byte, 1 extra displacement byte. */
350 {0, 0, 2, 0},
351
352 /* COND_JUMP86 states. */
353 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (COND_JUMP86, BIG)},
354 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (COND_JUMP86, BIG16)},
355 /* dword conditionals adds 4 bytes to frag:
356 1 extra opcode byte, 3 extra displacement bytes. */
357 {0, 0, 4, 0},
358 /* word conditionals add 3 bytes to frag:
359 1 extra opcode byte, 2 extra displacement bytes. */
360 {0, 0, 3, 0}
361 };
362
363 static const arch_entry cpu_arch[] = {
364 {"i8086", Cpu086 },
365 {"i186", Cpu086|Cpu186 },
366 {"i286", Cpu086|Cpu186|Cpu286 },
367 {"i386", Cpu086|Cpu186|Cpu286|Cpu386 },
368 {"i486", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486 },
369 {"i586", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
370 {"i686", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
371 {"pentium", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
372 {"pentiumpro",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
373 {"pentium4", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX|CpuSSE|CpuSSE2 },
374 {"k6", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow },
375 {"athlon", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuMMX|Cpu3dnow },
376 {"sledgehammer",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuSledgehammer|CpuMMX|Cpu3dnow|CpuSSE|CpuSSE2 },
377 {NULL, 0 }
378 };
379
380 void
381 i386_align_code (fragP, count)
382 fragS *fragP;
383 int count;
384 {
385 /* Various efficient no-op patterns for aligning code labels.
386 Note: Don't try to assemble the instructions in the comments.
387 0L and 0w are not legal. */
388 static const char f32_1[] =
389 {0x90}; /* nop */
390 static const char f32_2[] =
391 {0x89,0xf6}; /* movl %esi,%esi */
392 static const char f32_3[] =
393 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
394 static const char f32_4[] =
395 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
396 static const char f32_5[] =
397 {0x90, /* nop */
398 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
399 static const char f32_6[] =
400 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
401 static const char f32_7[] =
402 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
403 static const char f32_8[] =
404 {0x90, /* nop */
405 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
406 static const char f32_9[] =
407 {0x89,0xf6, /* movl %esi,%esi */
408 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
409 static const char f32_10[] =
410 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
411 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
412 static const char f32_11[] =
413 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
414 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
415 static const char f32_12[] =
416 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
417 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
418 static const char f32_13[] =
419 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
420 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
421 static const char f32_14[] =
422 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
423 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
424 static const char f32_15[] =
425 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
426 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
427 static const char f16_3[] =
428 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
429 static const char f16_4[] =
430 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
431 static const char f16_5[] =
432 {0x90, /* nop */
433 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
434 static const char f16_6[] =
435 {0x89,0xf6, /* mov %si,%si */
436 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
437 static const char f16_7[] =
438 {0x8d,0x74,0x00, /* lea 0(%si),%si */
439 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
440 static const char f16_8[] =
441 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
442 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
443 static const char *const f32_patt[] = {
444 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
445 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
446 };
447 static const char *const f16_patt[] = {
448 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
449 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
450 };
451
452 /* ??? We can't use these fillers for x86_64, since they often kills the
453 upper halves. Solve later. */
454 if (flag_code == CODE_64BIT)
455 count = 1;
456
457 if (count > 0 && count <= 15)
458 {
459 if (flag_code == CODE_16BIT)
460 {
461 memcpy (fragP->fr_literal + fragP->fr_fix,
462 f16_patt[count - 1], count);
463 if (count > 8)
464 /* Adjust jump offset. */
465 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
466 }
467 else
468 memcpy (fragP->fr_literal + fragP->fr_fix,
469 f32_patt[count - 1], count);
470 fragP->fr_var = count;
471 }
472 }
473
474 static char *output_invalid PARAMS ((int c));
475 static int i386_operand PARAMS ((char *operand_string));
476 static int i386_intel_operand PARAMS ((char *operand_string, int got_a_float));
477 static const reg_entry *parse_register PARAMS ((char *reg_string,
478 char **end_op));
479
480 #ifndef I386COFF
481 static void s_bss PARAMS ((int));
482 #endif
483
484 symbolS *GOT_symbol; /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
485
486 static INLINE unsigned int
487 mode_from_disp_size (t)
488 unsigned int t;
489 {
490 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
491 }
492
493 static INLINE int
494 fits_in_signed_byte (num)
495 offsetT num;
496 {
497 return (num >= -128) && (num <= 127);
498 }
499
500 static INLINE int
501 fits_in_unsigned_byte (num)
502 offsetT num;
503 {
504 return (num & 0xff) == num;
505 }
506
507 static INLINE int
508 fits_in_unsigned_word (num)
509 offsetT num;
510 {
511 return (num & 0xffff) == num;
512 }
513
514 static INLINE int
515 fits_in_signed_word (num)
516 offsetT num;
517 {
518 return (-32768 <= num) && (num <= 32767);
519 }
520 static INLINE int
521 fits_in_signed_long (num)
522 offsetT num ATTRIBUTE_UNUSED;
523 {
524 #ifndef BFD64
525 return 1;
526 #else
527 return (!(((offsetT) -1 << 31) & num)
528 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
529 #endif
530 } /* fits_in_signed_long() */
531 static INLINE int
532 fits_in_unsigned_long (num)
533 offsetT num ATTRIBUTE_UNUSED;
534 {
535 #ifndef BFD64
536 return 1;
537 #else
538 return (num & (((offsetT) 2 << 31) - 1)) == num;
539 #endif
540 } /* fits_in_unsigned_long() */
541
542 static int
543 smallest_imm_type (num)
544 offsetT num;
545 {
546 if (cpu_arch_flags != (Cpu086 | Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64)
547 && !(cpu_arch_flags & (CpuUnknown)))
548 {
549 /* This code is disabled on the 486 because all the Imm1 forms
550 in the opcode table are slower on the i486. They're the
551 versions with the implicitly specified single-position
552 displacement, which has another syntax if you really want to
553 use that form. */
554 if (num == 1)
555 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
556 }
557 return (fits_in_signed_byte (num)
558 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
559 : fits_in_unsigned_byte (num)
560 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
561 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
562 ? (Imm16 | Imm32 | Imm32S | Imm64)
563 : fits_in_signed_long (num)
564 ? (Imm32 | Imm32S | Imm64)
565 : fits_in_unsigned_long (num)
566 ? (Imm32 | Imm64)
567 : Imm64);
568 }
569
570 static offsetT
571 offset_in_range (val, size)
572 offsetT val;
573 int size;
574 {
575 addressT mask;
576
577 switch (size)
578 {
579 case 1: mask = ((addressT) 1 << 8) - 1; break;
580 case 2: mask = ((addressT) 1 << 16) - 1; break;
581 case 4: mask = ((addressT) 2 << 31) - 1; break;
582 #ifdef BFD64
583 case 8: mask = ((addressT) 2 << 63) - 1; break;
584 #endif
585 default: abort ();
586 }
587
588 /* If BFD64, sign extend val. */
589 if (!use_rela_relocations)
590 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
591 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
592
593 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
594 {
595 char buf1[40], buf2[40];
596
597 sprint_value (buf1, val);
598 sprint_value (buf2, val & mask);
599 as_warn (_("%s shortened to %s"), buf1, buf2);
600 }
601 return val & mask;
602 }
603
604 /* Returns 0 if attempting to add a prefix where one from the same
605 class already exists, 1 if non rep/repne added, 2 if rep/repne
606 added. */
607 static int
608 add_prefix (prefix)
609 unsigned int prefix;
610 {
611 int ret = 1;
612 int q;
613
614 if (prefix >= 0x40 && prefix < 0x50 && flag_code == CODE_64BIT)
615 q = REX_PREFIX;
616 else
617 switch (prefix)
618 {
619 default:
620 abort ();
621
622 case CS_PREFIX_OPCODE:
623 case DS_PREFIX_OPCODE:
624 case ES_PREFIX_OPCODE:
625 case FS_PREFIX_OPCODE:
626 case GS_PREFIX_OPCODE:
627 case SS_PREFIX_OPCODE:
628 q = SEG_PREFIX;
629 break;
630
631 case REPNE_PREFIX_OPCODE:
632 case REPE_PREFIX_OPCODE:
633 ret = 2;
634 /* fall thru */
635 case LOCK_PREFIX_OPCODE:
636 q = LOCKREP_PREFIX;
637 break;
638
639 case FWAIT_OPCODE:
640 q = WAIT_PREFIX;
641 break;
642
643 case ADDR_PREFIX_OPCODE:
644 q = ADDR_PREFIX;
645 break;
646
647 case DATA_PREFIX_OPCODE:
648 q = DATA_PREFIX;
649 break;
650 }
651
652 if (i.prefix[q])
653 {
654 as_bad (_("same type of prefix used twice"));
655 return 0;
656 }
657
658 i.prefixes += 1;
659 i.prefix[q] = prefix;
660 return ret;
661 }
662
663 static void
664 set_code_flag (value)
665 int value;
666 {
667 flag_code = value;
668 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
669 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
670 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
671 {
672 as_bad (_("64bit mode not supported on this CPU."));
673 }
674 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
675 {
676 as_bad (_("32bit mode not supported on this CPU."));
677 }
678 stackop_size = '\0';
679 }
680
681 static void
682 set_16bit_gcc_code_flag (new_code_flag)
683 int new_code_flag;
684 {
685 flag_code = new_code_flag;
686 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
687 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
688 stackop_size = 'l';
689 }
690
691 static void
692 set_intel_syntax (syntax_flag)
693 int syntax_flag;
694 {
695 /* Find out if register prefixing is specified. */
696 int ask_naked_reg = 0;
697
698 SKIP_WHITESPACE ();
699 if (! is_end_of_line[(unsigned char) *input_line_pointer])
700 {
701 char *string = input_line_pointer;
702 int e = get_symbol_end ();
703
704 if (strcmp (string, "prefix") == 0)
705 ask_naked_reg = 1;
706 else if (strcmp (string, "noprefix") == 0)
707 ask_naked_reg = -1;
708 else
709 as_bad (_("bad argument to syntax directive."));
710 *input_line_pointer = e;
711 }
712 demand_empty_rest_of_line ();
713
714 intel_syntax = syntax_flag;
715
716 if (ask_naked_reg == 0)
717 {
718 #ifdef BFD_ASSEMBLER
719 allow_naked_reg = (intel_syntax
720 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
721 #else
722 /* Conservative default. */
723 allow_naked_reg = 0;
724 #endif
725 }
726 else
727 allow_naked_reg = (ask_naked_reg < 0);
728 }
729
730 static void
731 set_cpu_arch (dummy)
732 int dummy ATTRIBUTE_UNUSED;
733 {
734 SKIP_WHITESPACE ();
735
736 if (! is_end_of_line[(unsigned char) *input_line_pointer])
737 {
738 char *string = input_line_pointer;
739 int e = get_symbol_end ();
740 int i;
741
742 for (i = 0; cpu_arch[i].name; i++)
743 {
744 if (strcmp (string, cpu_arch[i].name) == 0)
745 {
746 cpu_arch_name = cpu_arch[i].name;
747 cpu_arch_flags = (cpu_arch[i].flags
748 | (flag_code == CODE_64BIT ? Cpu64 : CpuNo64));
749 break;
750 }
751 }
752 if (!cpu_arch[i].name)
753 as_bad (_("no such architecture: `%s'"), string);
754
755 *input_line_pointer = e;
756 }
757 else
758 as_bad (_("missing cpu architecture"));
759
760 no_cond_jump_promotion = 0;
761 if (*input_line_pointer == ','
762 && ! is_end_of_line[(unsigned char) input_line_pointer[1]])
763 {
764 char *string = ++input_line_pointer;
765 int e = get_symbol_end ();
766
767 if (strcmp (string, "nojumps") == 0)
768 no_cond_jump_promotion = 1;
769 else if (strcmp (string, "jumps") == 0)
770 ;
771 else
772 as_bad (_("no such architecture modifier: `%s'"), string);
773
774 *input_line_pointer = e;
775 }
776
777 demand_empty_rest_of_line ();
778 }
779
780 const pseudo_typeS md_pseudo_table[] =
781 {
782 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
783 {"align", s_align_bytes, 0},
784 #else
785 {"align", s_align_ptwo, 0},
786 #endif
787 {"arch", set_cpu_arch, 0},
788 #ifndef I386COFF
789 {"bss", s_bss, 0},
790 #endif
791 {"ffloat", float_cons, 'f'},
792 {"dfloat", float_cons, 'd'},
793 {"tfloat", float_cons, 'x'},
794 {"value", cons, 2},
795 {"noopt", s_ignore, 0},
796 {"optim", s_ignore, 0},
797 {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
798 {"code16", set_code_flag, CODE_16BIT},
799 {"code32", set_code_flag, CODE_32BIT},
800 {"code64", set_code_flag, CODE_64BIT},
801 {"intel_syntax", set_intel_syntax, 1},
802 {"att_syntax", set_intel_syntax, 0},
803 {"file", dwarf2_directive_file, 0},
804 {"loc", dwarf2_directive_loc, 0},
805 {0, 0, 0}
806 };
807
808 /* For interface with expression (). */
809 extern char *input_line_pointer;
810
811 /* Hash table for instruction mnemonic lookup. */
812 static struct hash_control *op_hash;
813
814 /* Hash table for register lookup. */
815 static struct hash_control *reg_hash;
816 \f
817 #ifdef BFD_ASSEMBLER
818 unsigned long
819 i386_mach ()
820 {
821 if (!strcmp (default_arch, "x86_64"))
822 return bfd_mach_x86_64;
823 else if (!strcmp (default_arch, "i386"))
824 return bfd_mach_i386_i386;
825 else
826 as_fatal (_("Unknown architecture"));
827 }
828 #endif
829 \f
830 void
831 md_begin ()
832 {
833 const char *hash_err;
834
835 /* Initialize op_hash hash table. */
836 op_hash = hash_new ();
837
838 {
839 register const template *optab;
840 register templates *core_optab;
841
842 /* Setup for loop. */
843 optab = i386_optab;
844 core_optab = (templates *) xmalloc (sizeof (templates));
845 core_optab->start = optab;
846
847 while (1)
848 {
849 ++optab;
850 if (optab->name == NULL
851 || strcmp (optab->name, (optab - 1)->name) != 0)
852 {
853 /* different name --> ship out current template list;
854 add to hash table; & begin anew. */
855 core_optab->end = optab;
856 hash_err = hash_insert (op_hash,
857 (optab - 1)->name,
858 (PTR) core_optab);
859 if (hash_err)
860 {
861 as_fatal (_("Internal Error: Can't hash %s: %s"),
862 (optab - 1)->name,
863 hash_err);
864 }
865 if (optab->name == NULL)
866 break;
867 core_optab = (templates *) xmalloc (sizeof (templates));
868 core_optab->start = optab;
869 }
870 }
871 }
872
873 /* Initialize reg_hash hash table. */
874 reg_hash = hash_new ();
875 {
876 register const reg_entry *regtab;
877
878 for (regtab = i386_regtab;
879 regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
880 regtab++)
881 {
882 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
883 if (hash_err)
884 as_fatal (_("Internal Error: Can't hash %s: %s"),
885 regtab->reg_name,
886 hash_err);
887 }
888 }
889
890 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
891 {
892 register int c;
893 register char *p;
894
895 for (c = 0; c < 256; c++)
896 {
897 if (isdigit (c))
898 {
899 digit_chars[c] = c;
900 mnemonic_chars[c] = c;
901 register_chars[c] = c;
902 operand_chars[c] = c;
903 }
904 else if (islower (c))
905 {
906 mnemonic_chars[c] = c;
907 register_chars[c] = c;
908 operand_chars[c] = c;
909 }
910 else if (isupper (c))
911 {
912 mnemonic_chars[c] = tolower (c);
913 register_chars[c] = mnemonic_chars[c];
914 operand_chars[c] = c;
915 }
916
917 if (isalpha (c) || isdigit (c))
918 identifier_chars[c] = c;
919 else if (c >= 128)
920 {
921 identifier_chars[c] = c;
922 operand_chars[c] = c;
923 }
924 }
925
926 #ifdef LEX_AT
927 identifier_chars['@'] = '@';
928 #endif
929 digit_chars['-'] = '-';
930 identifier_chars['_'] = '_';
931 identifier_chars['.'] = '.';
932
933 for (p = operand_special_chars; *p != '\0'; p++)
934 operand_chars[(unsigned char) *p] = *p;
935 }
936
937 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
938 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
939 {
940 record_alignment (text_section, 2);
941 record_alignment (data_section, 2);
942 record_alignment (bss_section, 2);
943 }
944 #endif
945 }
946
947 void
948 i386_print_statistics (file)
949 FILE *file;
950 {
951 hash_print_statistics (file, "i386 opcode", op_hash);
952 hash_print_statistics (file, "i386 register", reg_hash);
953 }
954 \f
955 #ifdef DEBUG386
956
957 /* Debugging routines for md_assemble. */
958 static void pi PARAMS ((char *, i386_insn *));
959 static void pte PARAMS ((template *));
960 static void pt PARAMS ((unsigned int));
961 static void pe PARAMS ((expressionS *));
962 static void ps PARAMS ((symbolS *));
963
964 static void
965 pi (line, x)
966 char *line;
967 i386_insn *x;
968 {
969 unsigned int i;
970
971 fprintf (stdout, "%s: template ", line);
972 pte (&x->tm);
973 fprintf (stdout, " address: base %s index %s scale %x\n",
974 x->base_reg ? x->base_reg->reg_name : "none",
975 x->index_reg ? x->index_reg->reg_name : "none",
976 x->log2_scale_factor);
977 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
978 x->rm.mode, x->rm.reg, x->rm.regmem);
979 fprintf (stdout, " sib: base %x index %x scale %x\n",
980 x->sib.base, x->sib.index, x->sib.scale);
981 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
982 x->rex.mode64, x->rex.extX, x->rex.extY, x->rex.extZ);
983 for (i = 0; i < x->operands; i++)
984 {
985 fprintf (stdout, " #%d: ", i + 1);
986 pt (x->types[i]);
987 fprintf (stdout, "\n");
988 if (x->types[i]
989 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
990 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
991 if (x->types[i] & Imm)
992 pe (x->op[i].imms);
993 if (x->types[i] & Disp)
994 pe (x->op[i].disps);
995 }
996 }
997
998 static void
999 pte (t)
1000 template *t;
1001 {
1002 unsigned int i;
1003 fprintf (stdout, " %d operands ", t->operands);
1004 fprintf (stdout, "opcode %x ", t->base_opcode);
1005 if (t->extension_opcode != None)
1006 fprintf (stdout, "ext %x ", t->extension_opcode);
1007 if (t->opcode_modifier & D)
1008 fprintf (stdout, "D");
1009 if (t->opcode_modifier & W)
1010 fprintf (stdout, "W");
1011 fprintf (stdout, "\n");
1012 for (i = 0; i < t->operands; i++)
1013 {
1014 fprintf (stdout, " #%d type ", i + 1);
1015 pt (t->operand_types[i]);
1016 fprintf (stdout, "\n");
1017 }
1018 }
1019
1020 static void
1021 pe (e)
1022 expressionS *e;
1023 {
1024 fprintf (stdout, " operation %d\n", e->X_op);
1025 fprintf (stdout, " add_number %ld (%lx)\n",
1026 (long) e->X_add_number, (long) e->X_add_number);
1027 if (e->X_add_symbol)
1028 {
1029 fprintf (stdout, " add_symbol ");
1030 ps (e->X_add_symbol);
1031 fprintf (stdout, "\n");
1032 }
1033 if (e->X_op_symbol)
1034 {
1035 fprintf (stdout, " op_symbol ");
1036 ps (e->X_op_symbol);
1037 fprintf (stdout, "\n");
1038 }
1039 }
1040
1041 static void
1042 ps (s)
1043 symbolS *s;
1044 {
1045 fprintf (stdout, "%s type %s%s",
1046 S_GET_NAME (s),
1047 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
1048 segment_name (S_GET_SEGMENT (s)));
1049 }
1050
1051 struct type_name
1052 {
1053 unsigned int mask;
1054 char *tname;
1055 }
1056
1057 type_names[] =
1058 {
1059 { Reg8, "r8" },
1060 { Reg16, "r16" },
1061 { Reg32, "r32" },
1062 { Reg64, "r64" },
1063 { Imm8, "i8" },
1064 { Imm8S, "i8s" },
1065 { Imm16, "i16" },
1066 { Imm32, "i32" },
1067 { Imm32S, "i32s" },
1068 { Imm64, "i64" },
1069 { Imm1, "i1" },
1070 { BaseIndex, "BaseIndex" },
1071 { Disp8, "d8" },
1072 { Disp16, "d16" },
1073 { Disp32, "d32" },
1074 { Disp32S, "d32s" },
1075 { Disp64, "d64" },
1076 { InOutPortReg, "InOutPortReg" },
1077 { ShiftCount, "ShiftCount" },
1078 { Control, "control reg" },
1079 { Test, "test reg" },
1080 { Debug, "debug reg" },
1081 { FloatReg, "FReg" },
1082 { FloatAcc, "FAcc" },
1083 { SReg2, "SReg2" },
1084 { SReg3, "SReg3" },
1085 { Acc, "Acc" },
1086 { JumpAbsolute, "Jump Absolute" },
1087 { RegMMX, "rMMX" },
1088 { RegXMM, "rXMM" },
1089 { EsSeg, "es" },
1090 { 0, "" }
1091 };
1092
1093 static void
1094 pt (t)
1095 unsigned int t;
1096 {
1097 register struct type_name *ty;
1098
1099 for (ty = type_names; ty->mask; ty++)
1100 if (t & ty->mask)
1101 fprintf (stdout, "%s, ", ty->tname);
1102 fflush (stdout);
1103 }
1104
1105 #endif /* DEBUG386 */
1106 \f
1107 int
1108 tc_i386_force_relocation (fixp)
1109 struct fix *fixp;
1110 {
1111 #ifdef BFD_ASSEMBLER
1112 if (fixp->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1113 || fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1114 return 1;
1115 return 0;
1116 #else
1117 /* For COFF. */
1118 return fixp->fx_r_type == 7;
1119 #endif
1120 }
1121
1122 #ifdef BFD_ASSEMBLER
1123
1124 static bfd_reloc_code_real_type
1125 reloc (size, pcrel, sign, other)
1126 int size;
1127 int pcrel;
1128 int sign;
1129 bfd_reloc_code_real_type other;
1130 {
1131 if (other != NO_RELOC)
1132 return other;
1133
1134 if (pcrel)
1135 {
1136 if (!sign)
1137 as_bad (_("There are no unsigned pc-relative relocations"));
1138 switch (size)
1139 {
1140 case 1: return BFD_RELOC_8_PCREL;
1141 case 2: return BFD_RELOC_16_PCREL;
1142 case 4: return BFD_RELOC_32_PCREL;
1143 }
1144 as_bad (_("can not do %d byte pc-relative relocation"), size);
1145 }
1146 else
1147 {
1148 if (sign)
1149 switch (size)
1150 {
1151 case 4: return BFD_RELOC_X86_64_32S;
1152 }
1153 else
1154 switch (size)
1155 {
1156 case 1: return BFD_RELOC_8;
1157 case 2: return BFD_RELOC_16;
1158 case 4: return BFD_RELOC_32;
1159 case 8: return BFD_RELOC_64;
1160 }
1161 as_bad (_("can not do %s %d byte relocation"),
1162 sign ? "signed" : "unsigned", size);
1163 }
1164
1165 abort ();
1166 return BFD_RELOC_NONE;
1167 }
1168
1169 /* Here we decide which fixups can be adjusted to make them relative to
1170 the beginning of the section instead of the symbol. Basically we need
1171 to make sure that the dynamic relocations are done correctly, so in
1172 some cases we force the original symbol to be used. */
1173
1174 int
1175 tc_i386_fix_adjustable (fixP)
1176 fixS *fixP;
1177 {
1178 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1179 /* Prevent all adjustments to global symbols, or else dynamic
1180 linking will not work correctly. */
1181 if (S_IS_EXTERNAL (fixP->fx_addsy)
1182 || S_IS_WEAK (fixP->fx_addsy))
1183 return 0;
1184 #endif
1185 /* adjust_reloc_syms doesn't know about the GOT. */
1186 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1187 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1188 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1189 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1190 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1191 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
1192 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1193 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1194 return 0;
1195 return 1;
1196 }
1197 #else
1198 #define reloc(SIZE,PCREL,SIGN,OTHER) 0
1199 #define BFD_RELOC_16 0
1200 #define BFD_RELOC_32 0
1201 #define BFD_RELOC_16_PCREL 0
1202 #define BFD_RELOC_32_PCREL 0
1203 #define BFD_RELOC_386_PLT32 0
1204 #define BFD_RELOC_386_GOT32 0
1205 #define BFD_RELOC_386_GOTOFF 0
1206 #define BFD_RELOC_X86_64_PLT32 0
1207 #define BFD_RELOC_X86_64_GOT32 0
1208 #define BFD_RELOC_X86_64_GOTPCREL 0
1209 #endif
1210
1211 static int intel_float_operand PARAMS ((char *mnemonic));
1212
1213 static int
1214 intel_float_operand (mnemonic)
1215 char *mnemonic;
1216 {
1217 if (mnemonic[0] == 'f' && mnemonic[1] == 'i')
1218 return 2;
1219
1220 if (mnemonic[0] == 'f')
1221 return 1;
1222
1223 return 0;
1224 }
1225
1226 /* This is the guts of the machine-dependent assembler. LINE points to a
1227 machine dependent instruction. This function is supposed to emit
1228 the frags/bytes it assembles to. */
1229
1230 void
1231 md_assemble (line)
1232 char *line;
1233 {
1234 /* Points to template once we've found it. */
1235 const template *t;
1236
1237 /* Count the size of the instruction generated. Does not include
1238 variable part of jump insns before relax. */
1239 int insn_size = 0;
1240
1241 int j;
1242
1243 char mnemonic[MAX_MNEM_SIZE];
1244
1245 /* Initialize globals. */
1246 memset (&i, '\0', sizeof (i));
1247 for (j = 0; j < MAX_OPERANDS; j++)
1248 i.disp_reloc[j] = NO_RELOC;
1249 memset (disp_expressions, '\0', sizeof (disp_expressions));
1250 memset (im_expressions, '\0', sizeof (im_expressions));
1251 save_stack_p = save_stack;
1252
1253 /* First parse an instruction mnemonic & call i386_operand for the operands.
1254 We assume that the scrubber has arranged it so that line[0] is the valid
1255 start of a (possibly prefixed) mnemonic. */
1256 {
1257 char *l = line;
1258 char *token_start = l;
1259 char *mnem_p;
1260
1261 /* Non-zero if we found a prefix only acceptable with string insns. */
1262 const char *expecting_string_instruction = NULL;
1263
1264 while (1)
1265 {
1266 mnem_p = mnemonic;
1267 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1268 {
1269 mnem_p++;
1270 if (mnem_p >= mnemonic + sizeof (mnemonic))
1271 {
1272 as_bad (_("no such instruction: `%s'"), token_start);
1273 return;
1274 }
1275 l++;
1276 }
1277 if (!is_space_char (*l)
1278 && *l != END_OF_INSN
1279 && *l != PREFIX_SEPARATOR)
1280 {
1281 as_bad (_("invalid character %s in mnemonic"),
1282 output_invalid (*l));
1283 return;
1284 }
1285 if (token_start == l)
1286 {
1287 if (*l == PREFIX_SEPARATOR)
1288 as_bad (_("expecting prefix; got nothing"));
1289 else
1290 as_bad (_("expecting mnemonic; got nothing"));
1291 return;
1292 }
1293
1294 /* Look up instruction (or prefix) via hash table. */
1295 current_templates = hash_find (op_hash, mnemonic);
1296
1297 if (*l != END_OF_INSN
1298 && (! is_space_char (*l) || l[1] != END_OF_INSN)
1299 && current_templates
1300 && (current_templates->start->opcode_modifier & IsPrefix))
1301 {
1302 /* If we are in 16-bit mode, do not allow addr16 or data16.
1303 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1304 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1305 && (((current_templates->start->opcode_modifier & Size32) != 0)
1306 ^ (flag_code == CODE_16BIT)))
1307 {
1308 as_bad (_("redundant %s prefix"),
1309 current_templates->start->name);
1310 return;
1311 }
1312 /* Add prefix, checking for repeated prefixes. */
1313 switch (add_prefix (current_templates->start->base_opcode))
1314 {
1315 case 0:
1316 return;
1317 case 2:
1318 expecting_string_instruction = current_templates->start->name;
1319 break;
1320 }
1321 /* Skip past PREFIX_SEPARATOR and reset token_start. */
1322 token_start = ++l;
1323 }
1324 else
1325 break;
1326 }
1327
1328 if (!current_templates)
1329 {
1330 /* See if we can get a match by trimming off a suffix. */
1331 switch (mnem_p[-1])
1332 {
1333 case WORD_MNEM_SUFFIX:
1334 case BYTE_MNEM_SUFFIX:
1335 case QWORD_MNEM_SUFFIX:
1336 i.suffix = mnem_p[-1];
1337 mnem_p[-1] = '\0';
1338 current_templates = hash_find (op_hash, mnemonic);
1339 break;
1340 case SHORT_MNEM_SUFFIX:
1341 case LONG_MNEM_SUFFIX:
1342 if (!intel_syntax)
1343 {
1344 i.suffix = mnem_p[-1];
1345 mnem_p[-1] = '\0';
1346 current_templates = hash_find (op_hash, mnemonic);
1347 }
1348 break;
1349
1350 /* Intel Syntax. */
1351 case 'd':
1352 if (intel_syntax)
1353 {
1354 if (intel_float_operand (mnemonic))
1355 i.suffix = SHORT_MNEM_SUFFIX;
1356 else
1357 i.suffix = LONG_MNEM_SUFFIX;
1358 mnem_p[-1] = '\0';
1359 current_templates = hash_find (op_hash, mnemonic);
1360 }
1361 break;
1362 }
1363 if (!current_templates)
1364 {
1365 as_bad (_("no such instruction: `%s'"), token_start);
1366 return;
1367 }
1368 }
1369
1370 /* Check if instruction is supported on specified architecture. */
1371 if (cpu_arch_flags != 0)
1372 {
1373 if ((current_templates->start->cpu_flags & ~(Cpu64 | CpuNo64))
1374 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64)))
1375 {
1376 as_warn (_("`%s' is not supported on `%s'"),
1377 current_templates->start->name, cpu_arch_name);
1378 }
1379 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
1380 {
1381 as_warn (_("use .code16 to ensure correct addressing mode"));
1382 }
1383 }
1384
1385 /* Check for rep/repne without a string instruction. */
1386 if (expecting_string_instruction
1387 && !(current_templates->start->opcode_modifier & IsString))
1388 {
1389 as_bad (_("expecting string instruction after `%s'"),
1390 expecting_string_instruction);
1391 return;
1392 }
1393
1394 /* There may be operands to parse. */
1395 if (*l != END_OF_INSN)
1396 {
1397 /* 1 if operand is pending after ','. */
1398 unsigned int expecting_operand = 0;
1399
1400 /* Non-zero if operand parens not balanced. */
1401 unsigned int paren_not_balanced;
1402
1403 do
1404 {
1405 /* Skip optional white space before operand. */
1406 if (is_space_char (*l))
1407 ++l;
1408 if (!is_operand_char (*l) && *l != END_OF_INSN)
1409 {
1410 as_bad (_("invalid character %s before operand %d"),
1411 output_invalid (*l),
1412 i.operands + 1);
1413 return;
1414 }
1415 token_start = l; /* after white space */
1416 paren_not_balanced = 0;
1417 while (paren_not_balanced || *l != ',')
1418 {
1419 if (*l == END_OF_INSN)
1420 {
1421 if (paren_not_balanced)
1422 {
1423 if (!intel_syntax)
1424 as_bad (_("unbalanced parenthesis in operand %d."),
1425 i.operands + 1);
1426 else
1427 as_bad (_("unbalanced brackets in operand %d."),
1428 i.operands + 1);
1429 return;
1430 }
1431 else
1432 break; /* we are done */
1433 }
1434 else if (!is_operand_char (*l) && !is_space_char (*l))
1435 {
1436 as_bad (_("invalid character %s in operand %d"),
1437 output_invalid (*l),
1438 i.operands + 1);
1439 return;
1440 }
1441 if (!intel_syntax)
1442 {
1443 if (*l == '(')
1444 ++paren_not_balanced;
1445 if (*l == ')')
1446 --paren_not_balanced;
1447 }
1448 else
1449 {
1450 if (*l == '[')
1451 ++paren_not_balanced;
1452 if (*l == ']')
1453 --paren_not_balanced;
1454 }
1455 l++;
1456 }
1457 if (l != token_start)
1458 { /* Yes, we've read in another operand. */
1459 unsigned int operand_ok;
1460 this_operand = i.operands++;
1461 if (i.operands > MAX_OPERANDS)
1462 {
1463 as_bad (_("spurious operands; (%d operands/instruction max)"),
1464 MAX_OPERANDS);
1465 return;
1466 }
1467 /* Now parse operand adding info to 'i' as we go along. */
1468 END_STRING_AND_SAVE (l);
1469
1470 if (intel_syntax)
1471 operand_ok =
1472 i386_intel_operand (token_start,
1473 intel_float_operand (mnemonic));
1474 else
1475 operand_ok = i386_operand (token_start);
1476
1477 RESTORE_END_STRING (l);
1478 if (!operand_ok)
1479 return;
1480 }
1481 else
1482 {
1483 if (expecting_operand)
1484 {
1485 expecting_operand_after_comma:
1486 as_bad (_("expecting operand after ','; got nothing"));
1487 return;
1488 }
1489 if (*l == ',')
1490 {
1491 as_bad (_("expecting operand before ','; got nothing"));
1492 return;
1493 }
1494 }
1495
1496 /* Now *l must be either ',' or END_OF_INSN. */
1497 if (*l == ',')
1498 {
1499 if (*++l == END_OF_INSN)
1500 {
1501 /* Just skip it, if it's \n complain. */
1502 goto expecting_operand_after_comma;
1503 }
1504 expecting_operand = 1;
1505 }
1506 }
1507 while (*l != END_OF_INSN);
1508 }
1509 }
1510
1511 /* Now we've parsed the mnemonic into a set of templates, and have the
1512 operands at hand.
1513
1514 Next, we find a template that matches the given insn,
1515 making sure the overlap of the given operands types is consistent
1516 with the template operand types. */
1517
1518 #define MATCH(overlap, given, template) \
1519 ((overlap & ~JumpAbsolute) \
1520 && ((given) & (BaseIndex|JumpAbsolute)) == ((overlap) & (BaseIndex|JumpAbsolute)))
1521
1522 /* If given types r0 and r1 are registers they must be of the same type
1523 unless the expected operand type register overlap is null.
1524 Note that Acc in a template matches every size of reg. */
1525 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
1526 ( ((g0) & Reg) == 0 || ((g1) & Reg) == 0 || \
1527 ((g0) & Reg) == ((g1) & Reg) || \
1528 ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
1529
1530 {
1531 register unsigned int overlap0, overlap1;
1532 unsigned int overlap2;
1533 unsigned int found_reverse_match;
1534 int suffix_check;
1535
1536 /* All intel opcodes have reversed operands except for "bound" and
1537 "enter". We also don't reverse intersegment "jmp" and "call"
1538 instructions with 2 immediate operands so that the immediate segment
1539 precedes the offset, as it does when in AT&T mode. "enter" and the
1540 intersegment "jmp" and "call" instructions are the only ones that
1541 have two immediate operands. */
1542 if (intel_syntax && i.operands > 1
1543 && (strcmp (mnemonic, "bound") != 0)
1544 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1545 {
1546 union i386_op temp_op;
1547 unsigned int temp_type;
1548 #ifdef BFD_ASSEMBLER
1549 enum bfd_reloc_code_real temp_reloc;
1550 #else
1551 int temp_reloc;
1552 #endif
1553 int xchg1 = 0;
1554 int xchg2 = 0;
1555
1556 if (i.operands == 2)
1557 {
1558 xchg1 = 0;
1559 xchg2 = 1;
1560 }
1561 else if (i.operands == 3)
1562 {
1563 xchg1 = 0;
1564 xchg2 = 2;
1565 }
1566 temp_type = i.types[xchg2];
1567 i.types[xchg2] = i.types[xchg1];
1568 i.types[xchg1] = temp_type;
1569 temp_op = i.op[xchg2];
1570 i.op[xchg2] = i.op[xchg1];
1571 i.op[xchg1] = temp_op;
1572 temp_reloc = i.disp_reloc[xchg2];
1573 i.disp_reloc[xchg2] = i.disp_reloc[xchg1];
1574 i.disp_reloc[xchg1] = temp_reloc;
1575
1576 if (i.mem_operands == 2)
1577 {
1578 const seg_entry *temp_seg;
1579 temp_seg = i.seg[0];
1580 i.seg[0] = i.seg[1];
1581 i.seg[1] = temp_seg;
1582 }
1583 }
1584
1585 if (i.imm_operands)
1586 {
1587 /* Try to ensure constant immediates are represented in the smallest
1588 opcode possible. */
1589 char guess_suffix = 0;
1590 int op;
1591
1592 if (i.suffix)
1593 guess_suffix = i.suffix;
1594 else if (i.reg_operands)
1595 {
1596 /* Figure out a suffix from the last register operand specified.
1597 We can't do this properly yet, ie. excluding InOutPortReg,
1598 but the following works for instructions with immediates.
1599 In any case, we can't set i.suffix yet. */
1600 for (op = i.operands; --op >= 0;)
1601 if (i.types[op] & Reg)
1602 {
1603 if (i.types[op] & Reg8)
1604 guess_suffix = BYTE_MNEM_SUFFIX;
1605 else if (i.types[op] & Reg16)
1606 guess_suffix = WORD_MNEM_SUFFIX;
1607 else if (i.types[op] & Reg32)
1608 guess_suffix = LONG_MNEM_SUFFIX;
1609 else if (i.types[op] & Reg64)
1610 guess_suffix = QWORD_MNEM_SUFFIX;
1611 break;
1612 }
1613 }
1614 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
1615 guess_suffix = WORD_MNEM_SUFFIX;
1616
1617 for (op = i.operands; --op >= 0;)
1618 if (i.types[op] & Imm)
1619 {
1620 switch (i.op[op].imms->X_op)
1621 {
1622 case O_constant:
1623 /* If a suffix is given, this operand may be shortened. */
1624 switch (guess_suffix)
1625 {
1626 case LONG_MNEM_SUFFIX:
1627 i.types[op] |= Imm32 | Imm64;
1628 break;
1629 case WORD_MNEM_SUFFIX:
1630 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
1631 break;
1632 case BYTE_MNEM_SUFFIX:
1633 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
1634 break;
1635 }
1636
1637 /* If this operand is at most 16 bits, convert it
1638 to a signed 16 bit number before trying to see
1639 whether it will fit in an even smaller size.
1640 This allows a 16-bit operand such as $0xffe0 to
1641 be recognised as within Imm8S range. */
1642 if ((i.types[op] & Imm16)
1643 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
1644 {
1645 i.op[op].imms->X_add_number =
1646 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
1647 }
1648 if ((i.types[op] & Imm32)
1649 && (i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1)) == 0)
1650 {
1651 i.op[op].imms->X_add_number =
1652 (i.op[op].imms->X_add_number ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1653 }
1654 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
1655 /* We must avoid matching of Imm32 templates when 64bit only immediate is available. */
1656 if (guess_suffix == QWORD_MNEM_SUFFIX)
1657 i.types[op] &= ~Imm32;
1658 break;
1659 case O_absent:
1660 case O_register:
1661 abort ();
1662 /* Symbols and expressions. */
1663 default:
1664 /* Convert symbolic operand to proper sizes for matching. */
1665 switch (guess_suffix)
1666 {
1667 case QWORD_MNEM_SUFFIX:
1668 i.types[op] = Imm64 | Imm32S;
1669 break;
1670 case LONG_MNEM_SUFFIX:
1671 i.types[op] = Imm32 | Imm64;
1672 break;
1673 case WORD_MNEM_SUFFIX:
1674 i.types[op] = Imm16 | Imm32 | Imm64;
1675 break;
1676 break;
1677 case BYTE_MNEM_SUFFIX:
1678 i.types[op] = Imm8 | Imm8S | Imm16 | Imm32S | Imm32;
1679 break;
1680 break;
1681 }
1682 break;
1683 }
1684 }
1685 }
1686
1687 if (i.disp_operands)
1688 {
1689 /* Try to use the smallest displacement type too. */
1690 int op;
1691
1692 for (op = i.operands; --op >= 0;)
1693 if ((i.types[op] & Disp)
1694 && i.op[op].imms->X_op == O_constant)
1695 {
1696 offsetT disp = i.op[op].disps->X_add_number;
1697
1698 if (i.types[op] & Disp16)
1699 {
1700 /* We know this operand is at most 16 bits, so
1701 convert to a signed 16 bit number before trying
1702 to see whether it will fit in an even smaller
1703 size. */
1704
1705 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
1706 }
1707 else if (i.types[op] & Disp32)
1708 {
1709 /* We know this operand is at most 32 bits, so convert to a
1710 signed 32 bit number before trying to see whether it will
1711 fit in an even smaller size. */
1712 disp &= (((offsetT) 2 << 31) - 1);
1713 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1714 }
1715 if (flag_code == CODE_64BIT)
1716 {
1717 if (fits_in_signed_long (disp))
1718 i.types[op] |= Disp32S;
1719 if (fits_in_unsigned_long (disp))
1720 i.types[op] |= Disp32;
1721 }
1722 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
1723 && fits_in_signed_byte (disp))
1724 i.types[op] |= Disp8;
1725 }
1726 }
1727
1728 overlap0 = 0;
1729 overlap1 = 0;
1730 overlap2 = 0;
1731 found_reverse_match = 0;
1732 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
1733 ? No_bSuf
1734 : (i.suffix == WORD_MNEM_SUFFIX
1735 ? No_wSuf
1736 : (i.suffix == SHORT_MNEM_SUFFIX
1737 ? No_sSuf
1738 : (i.suffix == LONG_MNEM_SUFFIX
1739 ? No_lSuf
1740 : (i.suffix == QWORD_MNEM_SUFFIX
1741 ? No_qSuf
1742 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX ? No_xSuf : 0))))));
1743
1744 for (t = current_templates->start;
1745 t < current_templates->end;
1746 t++)
1747 {
1748 /* Must have right number of operands. */
1749 if (i.operands != t->operands)
1750 continue;
1751
1752 /* Check the suffix, except for some instructions in intel mode. */
1753 if ((t->opcode_modifier & suffix_check)
1754 && !(intel_syntax
1755 && (t->opcode_modifier & IgnoreSize))
1756 && !(intel_syntax
1757 && t->base_opcode == 0xd9
1758 && (t->extension_opcode == 5 /* 0xd9,5 "fldcw" */
1759 || t->extension_opcode == 7))) /* 0xd9,7 "f{n}stcw" */
1760 continue;
1761
1762 /* Do not verify operands when there are none. */
1763 else if (!t->operands)
1764 {
1765 if (t->cpu_flags & ~cpu_arch_flags)
1766 continue;
1767 /* We've found a match; break out of loop. */
1768 break;
1769 }
1770
1771 overlap0 = i.types[0] & t->operand_types[0];
1772 switch (t->operands)
1773 {
1774 case 1:
1775 if (!MATCH (overlap0, i.types[0], t->operand_types[0]))
1776 continue;
1777 break;
1778 case 2:
1779 case 3:
1780 overlap1 = i.types[1] & t->operand_types[1];
1781 if (!MATCH (overlap0, i.types[0], t->operand_types[0])
1782 || !MATCH (overlap1, i.types[1], t->operand_types[1])
1783 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
1784 t->operand_types[0],
1785 overlap1, i.types[1],
1786 t->operand_types[1]))
1787 {
1788 /* Check if other direction is valid ... */
1789 if ((t->opcode_modifier & (D|FloatD)) == 0)
1790 continue;
1791
1792 /* Try reversing direction of operands. */
1793 overlap0 = i.types[0] & t->operand_types[1];
1794 overlap1 = i.types[1] & t->operand_types[0];
1795 if (!MATCH (overlap0, i.types[0], t->operand_types[1])
1796 || !MATCH (overlap1, i.types[1], t->operand_types[0])
1797 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
1798 t->operand_types[1],
1799 overlap1, i.types[1],
1800 t->operand_types[0]))
1801 {
1802 /* Does not match either direction. */
1803 continue;
1804 }
1805 /* found_reverse_match holds which of D or FloatDR
1806 we've found. */
1807 found_reverse_match = t->opcode_modifier & (D|FloatDR);
1808 }
1809 /* Found a forward 2 operand match here. */
1810 else if (t->operands == 3)
1811 {
1812 /* Here we make use of the fact that there are no
1813 reverse match 3 operand instructions, and all 3
1814 operand instructions only need to be checked for
1815 register consistency between operands 2 and 3. */
1816 overlap2 = i.types[2] & t->operand_types[2];
1817 if (!MATCH (overlap2, i.types[2], t->operand_types[2])
1818 || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1],
1819 t->operand_types[1],
1820 overlap2, i.types[2],
1821 t->operand_types[2]))
1822
1823 continue;
1824 }
1825 /* Found either forward/reverse 2 or 3 operand match here:
1826 slip through to break. */
1827 }
1828 if (t->cpu_flags & ~cpu_arch_flags)
1829 {
1830 found_reverse_match = 0;
1831 continue;
1832 }
1833 /* We've found a match; break out of loop. */
1834 break;
1835 }
1836 if (t == current_templates->end)
1837 {
1838 /* We found no match. */
1839 as_bad (_("suffix or operands invalid for `%s'"),
1840 current_templates->start->name);
1841 return;
1842 }
1843
1844 if (!quiet_warnings)
1845 {
1846 if (!intel_syntax
1847 && ((i.types[0] & JumpAbsolute)
1848 != (t->operand_types[0] & JumpAbsolute)))
1849 {
1850 as_warn (_("indirect %s without `*'"), t->name);
1851 }
1852
1853 if ((t->opcode_modifier & (IsPrefix|IgnoreSize))
1854 == (IsPrefix|IgnoreSize))
1855 {
1856 /* Warn them that a data or address size prefix doesn't
1857 affect assembly of the next line of code. */
1858 as_warn (_("stand-alone `%s' prefix"), t->name);
1859 }
1860 }
1861
1862 /* Copy the template we found. */
1863 i.tm = *t;
1864 if (found_reverse_match)
1865 {
1866 /* If we found a reverse match we must alter the opcode
1867 direction bit. found_reverse_match holds bits to change
1868 (different for int & float insns). */
1869
1870 i.tm.base_opcode ^= found_reverse_match;
1871
1872 i.tm.operand_types[0] = t->operand_types[1];
1873 i.tm.operand_types[1] = t->operand_types[0];
1874 }
1875
1876 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1877 if (SYSV386_COMPAT
1878 && intel_syntax
1879 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1880 i.tm.base_opcode ^= FloatR;
1881
1882 if (i.tm.opcode_modifier & FWait)
1883 if (! add_prefix (FWAIT_OPCODE))
1884 return;
1885
1886 /* Check string instruction segment overrides. */
1887 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1888 {
1889 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
1890 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
1891 {
1892 if (i.seg[0] != NULL && i.seg[0] != &es)
1893 {
1894 as_bad (_("`%s' operand %d must use `%%es' segment"),
1895 i.tm.name,
1896 mem_op + 1);
1897 return;
1898 }
1899 /* There's only ever one segment override allowed per instruction.
1900 This instruction possibly has a legal segment override on the
1901 second operand, so copy the segment to where non-string
1902 instructions store it, allowing common code. */
1903 i.seg[0] = i.seg[1];
1904 }
1905 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
1906 {
1907 if (i.seg[1] != NULL && i.seg[1] != &es)
1908 {
1909 as_bad (_("`%s' operand %d must use `%%es' segment"),
1910 i.tm.name,
1911 mem_op + 2);
1912 return;
1913 }
1914 }
1915 }
1916
1917 if (i.reg_operands && flag_code < CODE_64BIT)
1918 {
1919 int op;
1920 for (op = i.operands; --op >= 0;)
1921 if ((i.types[op] & Reg)
1922 && (i.op[op].regs->reg_flags & (RegRex64|RegRex)))
1923 {
1924 as_bad (_("Extended register `%%%s' available only in 64bit mode."),
1925 i.op[op].regs->reg_name);
1926 return;
1927 }
1928 }
1929
1930 /* If matched instruction specifies an explicit instruction mnemonic
1931 suffix, use it. */
1932 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
1933 {
1934 if (i.tm.opcode_modifier & Size16)
1935 i.suffix = WORD_MNEM_SUFFIX;
1936 else if (i.tm.opcode_modifier & Size64)
1937 i.suffix = QWORD_MNEM_SUFFIX;
1938 else
1939 i.suffix = LONG_MNEM_SUFFIX;
1940 }
1941 else if (i.reg_operands)
1942 {
1943 /* If there's no instruction mnemonic suffix we try to invent one
1944 based on register operands. */
1945 if (!i.suffix)
1946 {
1947 /* We take i.suffix from the last register operand specified,
1948 Destination register type is more significant than source
1949 register type. */
1950 int op;
1951 for (op = i.operands; --op >= 0;)
1952 if ((i.types[op] & Reg)
1953 && !(i.tm.operand_types[op] & InOutPortReg))
1954 {
1955 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
1956 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
1957 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
1958 LONG_MNEM_SUFFIX);
1959 break;
1960 }
1961 }
1962 else if (i.suffix == BYTE_MNEM_SUFFIX)
1963 {
1964 int op;
1965 for (op = i.operands; --op >= 0;)
1966 {
1967 /* If this is an eight bit register, it's OK. If it's
1968 the 16 or 32 bit version of an eight bit register,
1969 we will just use the low portion, and that's OK too. */
1970 if (i.types[op] & Reg8)
1971 continue;
1972
1973 /* movzx and movsx should not generate this warning. */
1974 if (intel_syntax
1975 && (i.tm.base_opcode == 0xfb7
1976 || i.tm.base_opcode == 0xfb6
1977 || i.tm.base_opcode == 0x63
1978 || i.tm.base_opcode == 0xfbe
1979 || i.tm.base_opcode == 0xfbf))
1980 continue;
1981
1982 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4
1983 #if 0
1984 /* Check that the template allows eight bit regs
1985 This kills insns such as `orb $1,%edx', which
1986 maybe should be allowed. */
1987 && (i.tm.operand_types[op] & (Reg8|InOutPortReg))
1988 #endif
1989 )
1990 {
1991 /* Prohibit these changes in the 64bit mode, since
1992 the lowering is more complicated. */
1993 if (flag_code == CODE_64BIT
1994 && (i.tm.operand_types[op] & InOutPortReg) == 0)
1995 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
1996 i.op[op].regs->reg_name,
1997 i.suffix);
1998 #if REGISTER_WARNINGS
1999 if (!quiet_warnings
2000 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2001 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2002 (i.op[op].regs - (i.types[op] & Reg16 ? 8 : 16))->reg_name,
2003 i.op[op].regs->reg_name,
2004 i.suffix);
2005 #endif
2006 continue;
2007 }
2008 /* Any other register is bad. */
2009 if (i.types[op] & (Reg | RegMMX | RegXMM
2010 | SReg2 | SReg3
2011 | Control | Debug | Test
2012 | FloatReg | FloatAcc))
2013 {
2014 as_bad (_("`%%%s' not allowed with `%s%c'"),
2015 i.op[op].regs->reg_name,
2016 i.tm.name,
2017 i.suffix);
2018 return;
2019 }
2020 }
2021 }
2022 else if (i.suffix == LONG_MNEM_SUFFIX)
2023 {
2024 int op;
2025
2026 for (op = i.operands; --op >= 0;)
2027 /* Reject eight bit registers, except where the template
2028 requires them. (eg. movzb) */
2029 if ((i.types[op] & Reg8) != 0
2030 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2031 {
2032 as_bad (_("`%%%s' not allowed with `%s%c'"),
2033 i.op[op].regs->reg_name,
2034 i.tm.name,
2035 i.suffix);
2036 return;
2037 }
2038 /* Warn if the e prefix on a general reg is missing. */
2039 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2040 && (i.types[op] & Reg16) != 0
2041 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
2042 {
2043 /* Prohibit these changes in the 64bit mode, since
2044 the lowering is more complicated. */
2045 if (flag_code == CODE_64BIT)
2046 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2047 i.op[op].regs->reg_name,
2048 i.suffix);
2049 #if REGISTER_WARNINGS
2050 else
2051 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2052 (i.op[op].regs + 8)->reg_name,
2053 i.op[op].regs->reg_name,
2054 i.suffix);
2055 #endif
2056 }
2057 /* Warn if the r prefix on a general reg is missing. */
2058 else if ((i.types[op] & Reg64) != 0
2059 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
2060 {
2061 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2062 i.op[op].regs->reg_name,
2063 i.suffix);
2064 }
2065 }
2066 else if (i.suffix == QWORD_MNEM_SUFFIX)
2067 {
2068 int op;
2069
2070 for (op = i.operands; --op >= 0; )
2071 /* Reject eight bit registers, except where the template
2072 requires them. (eg. movzb) */
2073 if ((i.types[op] & Reg8) != 0
2074 && (i.tm.operand_types[op] & (Reg16|Reg32|Acc)) != 0)
2075 {
2076 as_bad (_("`%%%s' not allowed with `%s%c'"),
2077 i.op[op].regs->reg_name,
2078 i.tm.name,
2079 i.suffix);
2080 return;
2081 }
2082 /* Warn if the e prefix on a general reg is missing. */
2083 else if (((i.types[op] & Reg16) != 0
2084 || (i.types[op] & Reg32) != 0)
2085 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
2086 {
2087 /* Prohibit these changes in the 64bit mode, since
2088 the lowering is more complicated. */
2089 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2090 i.op[op].regs->reg_name,
2091 i.suffix);
2092 }
2093 }
2094 else if (i.suffix == WORD_MNEM_SUFFIX)
2095 {
2096 int op;
2097 for (op = i.operands; --op >= 0;)
2098 /* Reject eight bit registers, except where the template
2099 requires them. (eg. movzb) */
2100 if ((i.types[op] & Reg8) != 0
2101 && (i.tm.operand_types[op] & (Reg16|Reg32|Acc)) != 0)
2102 {
2103 as_bad (_("`%%%s' not allowed with `%s%c'"),
2104 i.op[op].regs->reg_name,
2105 i.tm.name,
2106 i.suffix);
2107 return;
2108 }
2109 /* Warn if the e prefix on a general reg is present. */
2110 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2111 && (i.types[op] & Reg32) != 0
2112 && (i.tm.operand_types[op] & (Reg16|Acc)) != 0)
2113 {
2114 /* Prohibit these changes in the 64bit mode, since
2115 the lowering is more complicated. */
2116 if (flag_code == CODE_64BIT)
2117 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2118 i.op[op].regs->reg_name,
2119 i.suffix);
2120 else
2121 #if REGISTER_WARNINGS
2122 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2123 (i.op[op].regs - 8)->reg_name,
2124 i.op[op].regs->reg_name,
2125 i.suffix);
2126 #endif
2127 }
2128 }
2129 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2130 /* Do nothing if the instruction is going to ignore the prefix. */
2131 ;
2132 else
2133 abort ();
2134 }
2135 else if ((i.tm.opcode_modifier & DefaultSize) && !i.suffix)
2136 {
2137 i.suffix = stackop_size;
2138 }
2139 /* Make still unresolved immediate matches conform to size of immediate
2140 given in i.suffix. Note: overlap2 cannot be an immediate! */
2141 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S))
2142 && overlap0 != Imm8 && overlap0 != Imm8S
2143 && overlap0 != Imm16 && overlap0 != Imm32S
2144 && overlap0 != Imm32 && overlap0 != Imm64)
2145 {
2146 if (i.suffix)
2147 {
2148 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
2149 (i.suffix == WORD_MNEM_SUFFIX ? Imm16 :
2150 (i.suffix == QWORD_MNEM_SUFFIX ? Imm64 | Imm32S : Imm32)));
2151 }
2152 else if (overlap0 == (Imm16 | Imm32S | Imm32)
2153 || overlap0 == (Imm16 | Imm32)
2154 || overlap0 == (Imm16 | Imm32S))
2155 {
2156 overlap0 =
2157 ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32S;
2158 }
2159 if (overlap0 != Imm8 && overlap0 != Imm8S
2160 && overlap0 != Imm16 && overlap0 != Imm32S
2161 && overlap0 != Imm32 && overlap0 != Imm64)
2162 {
2163 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
2164 return;
2165 }
2166 }
2167 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32))
2168 && overlap1 != Imm8 && overlap1 != Imm8S
2169 && overlap1 != Imm16 && overlap1 != Imm32S
2170 && overlap1 != Imm32 && overlap1 != Imm64)
2171 {
2172 if (i.suffix)
2173 {
2174 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
2175 (i.suffix == WORD_MNEM_SUFFIX ? Imm16 :
2176 (i.suffix == QWORD_MNEM_SUFFIX ? Imm64 | Imm32S : Imm32)));
2177 }
2178 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
2179 || overlap1 == (Imm16 | Imm32)
2180 || overlap1 == (Imm16 | Imm32S))
2181 {
2182 overlap1 =
2183 ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32S;
2184 }
2185 if (overlap1 != Imm8 && overlap1 != Imm8S
2186 && overlap1 != Imm16 && overlap1 != Imm32S
2187 && overlap1 != Imm32 && overlap1 != Imm64)
2188 {
2189 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
2190 return;
2191 }
2192 }
2193 assert ((overlap2 & Imm) == 0);
2194
2195 i.types[0] = overlap0;
2196 if (overlap0 & ImplicitRegister)
2197 i.reg_operands--;
2198 if (overlap0 & Imm1)
2199 i.imm_operands = 0; /* kludge for shift insns. */
2200
2201 i.types[1] = overlap1;
2202 if (overlap1 & ImplicitRegister)
2203 i.reg_operands--;
2204
2205 i.types[2] = overlap2;
2206 if (overlap2 & ImplicitRegister)
2207 i.reg_operands--;
2208
2209 /* Finalize opcode. First, we change the opcode based on the operand
2210 size given by i.suffix: We need not change things for byte insns. */
2211
2212 if (!i.suffix && (i.tm.opcode_modifier & W))
2213 {
2214 as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
2215 return;
2216 }
2217
2218 /* For movzx and movsx, need to check the register type. */
2219 if (intel_syntax
2220 && (i.tm.base_opcode == 0xfb6 || i.tm.base_opcode == 0xfbe))
2221 if (i.suffix && i.suffix == BYTE_MNEM_SUFFIX)
2222 {
2223 unsigned int prefix = DATA_PREFIX_OPCODE;
2224
2225 if ((i.op[1].regs->reg_type & Reg16) != 0)
2226 if (!add_prefix (prefix))
2227 return;
2228 }
2229
2230 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2231 {
2232 /* It's not a byte, select word/dword operation. */
2233 if (i.tm.opcode_modifier & W)
2234 {
2235 if (i.tm.opcode_modifier & ShortForm)
2236 i.tm.base_opcode |= 8;
2237 else
2238 i.tm.base_opcode |= 1;
2239 }
2240 /* Now select between word & dword operations via the operand
2241 size prefix, except for instructions that will ignore this
2242 prefix anyway. */
2243 if (i.suffix != QWORD_MNEM_SUFFIX
2244 && (i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2245 && !(i.tm.opcode_modifier & IgnoreSize))
2246 {
2247 unsigned int prefix = DATA_PREFIX_OPCODE;
2248 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2249 prefix = ADDR_PREFIX_OPCODE;
2250
2251 if (! add_prefix (prefix))
2252 return;
2253 }
2254
2255 /* Set mode64 for an operand. */
2256 if (i.suffix == QWORD_MNEM_SUFFIX
2257 && !(i.tm.opcode_modifier & NoRex64))
2258 {
2259 i.rex.mode64 = 1;
2260 if (flag_code < CODE_64BIT)
2261 {
2262 as_bad (_("64bit operations available only in 64bit modes."));
2263 return;
2264 }
2265 }
2266
2267 /* Size floating point instruction. */
2268 if (i.suffix == LONG_MNEM_SUFFIX)
2269 {
2270 if (i.tm.opcode_modifier & FloatMF)
2271 i.tm.base_opcode ^= 4;
2272 }
2273 }
2274
2275 if (i.tm.opcode_modifier & ImmExt)
2276 {
2277 /* These AMD 3DNow! and Intel Katmai New Instructions have an
2278 opcode suffix which is coded in the same place as an 8-bit
2279 immediate field would be. Here we fake an 8-bit immediate
2280 operand from the opcode suffix stored in tm.extension_opcode. */
2281
2282 expressionS *exp;
2283
2284 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
2285
2286 exp = &im_expressions[i.imm_operands++];
2287 i.op[i.operands].imms = exp;
2288 i.types[i.operands++] = Imm8;
2289 exp->X_op = O_constant;
2290 exp->X_add_number = i.tm.extension_opcode;
2291 i.tm.extension_opcode = None;
2292 }
2293
2294 /* For insns with operands there are more diddles to do to the opcode. */
2295 if (i.operands)
2296 {
2297 /* Default segment register this instruction will use
2298 for memory accesses. 0 means unknown.
2299 This is only for optimizing out unnecessary segment overrides. */
2300 const seg_entry *default_seg = 0;
2301
2302 /* The imul $imm, %reg instruction is converted into
2303 imul $imm, %reg, %reg, and the clr %reg instruction
2304 is converted into xor %reg, %reg. */
2305 if (i.tm.opcode_modifier & regKludge)
2306 {
2307 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
2308 /* Pretend we saw the extra register operand. */
2309 assert (i.op[first_reg_op + 1].regs == 0);
2310 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
2311 i.types[first_reg_op + 1] = i.types[first_reg_op];
2312 i.reg_operands = 2;
2313 }
2314
2315 if (i.tm.opcode_modifier & ShortForm)
2316 {
2317 /* The register or float register operand is in operand 0 or 1. */
2318 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
2319 /* Register goes in low 3 bits of opcode. */
2320 i.tm.base_opcode |= i.op[op].regs->reg_num;
2321 if (i.op[op].regs->reg_flags & RegRex)
2322 i.rex.extZ = 1;
2323 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2324 {
2325 /* Warn about some common errors, but press on regardless.
2326 The first case can be generated by gcc (<= 2.8.1). */
2327 if (i.operands == 2)
2328 {
2329 /* Reversed arguments on faddp, fsubp, etc. */
2330 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
2331 i.op[1].regs->reg_name,
2332 i.op[0].regs->reg_name);
2333 }
2334 else
2335 {
2336 /* Extraneous `l' suffix on fp insn. */
2337 as_warn (_("translating to `%s %%%s'"), i.tm.name,
2338 i.op[0].regs->reg_name);
2339 }
2340 }
2341 }
2342 else if (i.tm.opcode_modifier & Modrm)
2343 {
2344 /* The opcode is completed (modulo i.tm.extension_opcode which
2345 must be put into the modrm byte).
2346 Now, we make the modrm & index base bytes based on all the
2347 info we've collected. */
2348
2349 /* i.reg_operands MUST be the number of real register operands;
2350 implicit registers do not count. */
2351 if (i.reg_operands == 2)
2352 {
2353 unsigned int source, dest;
2354 source = ((i.types[0]
2355 & (Reg | RegMMX | RegXMM
2356 | SReg2 | SReg3
2357 | Control | Debug | Test))
2358 ? 0 : 1);
2359 dest = source + 1;
2360
2361 i.rm.mode = 3;
2362 /* One of the register operands will be encoded in the
2363 i.tm.reg field, the other in the combined i.tm.mode
2364 and i.tm.regmem fields. If no form of this
2365 instruction supports a memory destination operand,
2366 then we assume the source operand may sometimes be
2367 a memory operand and so we need to store the
2368 destination in the i.rm.reg field. */
2369 if ((i.tm.operand_types[dest] & AnyMem) == 0)
2370 {
2371 i.rm.reg = i.op[dest].regs->reg_num;
2372 i.rm.regmem = i.op[source].regs->reg_num;
2373 if (i.op[dest].regs->reg_flags & RegRex)
2374 i.rex.extX = 1;
2375 if (i.op[source].regs->reg_flags & RegRex)
2376 i.rex.extZ = 1;
2377 }
2378 else
2379 {
2380 i.rm.reg = i.op[source].regs->reg_num;
2381 i.rm.regmem = i.op[dest].regs->reg_num;
2382 if (i.op[dest].regs->reg_flags & RegRex)
2383 i.rex.extZ = 1;
2384 if (i.op[source].regs->reg_flags & RegRex)
2385 i.rex.extX = 1;
2386 }
2387 }
2388 else
2389 { /* If it's not 2 reg operands... */
2390 if (i.mem_operands)
2391 {
2392 unsigned int fake_zero_displacement = 0;
2393 unsigned int op = ((i.types[0] & AnyMem)
2394 ? 0
2395 : (i.types[1] & AnyMem) ? 1 : 2);
2396
2397 default_seg = &ds;
2398
2399 if (! i.base_reg)
2400 {
2401 i.rm.mode = 0;
2402 if (! i.disp_operands)
2403 fake_zero_displacement = 1;
2404 if (! i.index_reg)
2405 {
2406 /* Operand is just <disp> */
2407 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
2408 {
2409 i.rm.regmem = NO_BASE_REGISTER_16;
2410 i.types[op] &= ~Disp;
2411 i.types[op] |= Disp16;
2412 }
2413 else if (flag_code != CODE_64BIT)
2414 {
2415 i.rm.regmem = NO_BASE_REGISTER;
2416 i.types[op] &= ~Disp;
2417 i.types[op] |= Disp32;
2418 }
2419 else
2420 {
2421 /* 64bit mode overwrites the 32bit
2422 absolute addressing by RIP relative
2423 addressing and absolute addressing
2424 is encoded by one of the redundant
2425 SIB forms. */
2426
2427 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2428 i.sib.base = NO_BASE_REGISTER;
2429 i.sib.index = NO_INDEX_REGISTER;
2430 i.types[op] &= ~Disp;
2431 i.types[op] |= Disp32S;
2432 }
2433 }
2434 else /* ! i.base_reg && i.index_reg */
2435 {
2436 i.sib.index = i.index_reg->reg_num;
2437 i.sib.base = NO_BASE_REGISTER;
2438 i.sib.scale = i.log2_scale_factor;
2439 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2440 i.types[op] &= ~Disp;
2441 if (flag_code != CODE_64BIT)
2442 i.types[op] |= Disp32; /* Must be 32 bit */
2443 else
2444 i.types[op] |= Disp32S;
2445 if (i.index_reg->reg_flags & RegRex)
2446 i.rex.extY = 1;
2447 }
2448 }
2449 /* RIP addressing for 64bit mode. */
2450 else if (i.base_reg->reg_type == BaseIndex)
2451 {
2452 i.rm.regmem = NO_BASE_REGISTER;
2453 i.types[op] &= ~Disp;
2454 i.types[op] |= Disp32S;
2455 i.flags[op] = Operand_PCrel;
2456 }
2457 else if (i.base_reg->reg_type & Reg16)
2458 {
2459 switch (i.base_reg->reg_num)
2460 {
2461 case 3: /* (%bx) */
2462 if (! i.index_reg)
2463 i.rm.regmem = 7;
2464 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
2465 i.rm.regmem = i.index_reg->reg_num - 6;
2466 break;
2467 case 5: /* (%bp) */
2468 default_seg = &ss;
2469 if (! i.index_reg)
2470 {
2471 i.rm.regmem = 6;
2472 if ((i.types[op] & Disp) == 0)
2473 {
2474 /* fake (%bp) into 0(%bp) */
2475 i.types[op] |= Disp8;
2476 fake_zero_displacement = 1;
2477 }
2478 }
2479 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
2480 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
2481 break;
2482 default: /* (%si) -> 4 or (%di) -> 5 */
2483 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
2484 }
2485 i.rm.mode = mode_from_disp_size (i.types[op]);
2486 }
2487 else /* i.base_reg and 32/64 bit mode */
2488 {
2489 if (flag_code == CODE_64BIT
2490 && (i.types[op] & Disp))
2491 {
2492 if (i.types[op] & Disp8)
2493 i.types[op] = Disp8 | Disp32S;
2494 else
2495 i.types[op] = Disp32S;
2496 }
2497 i.rm.regmem = i.base_reg->reg_num;
2498 if (i.base_reg->reg_flags & RegRex)
2499 i.rex.extZ = 1;
2500 i.sib.base = i.base_reg->reg_num;
2501 /* x86-64 ignores REX prefix bit here to avoid
2502 decoder complications. */
2503 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
2504 {
2505 default_seg = &ss;
2506 if (i.disp_operands == 0)
2507 {
2508 fake_zero_displacement = 1;
2509 i.types[op] |= Disp8;
2510 }
2511 }
2512 else if (i.base_reg->reg_num == ESP_REG_NUM)
2513 {
2514 default_seg = &ss;
2515 }
2516 i.sib.scale = i.log2_scale_factor;
2517 if (! i.index_reg)
2518 {
2519 /* <disp>(%esp) becomes two byte modrm
2520 with no index register. We've already
2521 stored the code for esp in i.rm.regmem
2522 ie. ESCAPE_TO_TWO_BYTE_ADDRESSING. Any
2523 base register besides %esp will not use
2524 the extra modrm byte. */
2525 i.sib.index = NO_INDEX_REGISTER;
2526 #if ! SCALE1_WHEN_NO_INDEX
2527 /* Another case where we force the second
2528 modrm byte. */
2529 if (i.log2_scale_factor)
2530 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2531 #endif
2532 }
2533 else
2534 {
2535 i.sib.index = i.index_reg->reg_num;
2536 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2537 if (i.index_reg->reg_flags & RegRex)
2538 i.rex.extY = 1;
2539 }
2540 i.rm.mode = mode_from_disp_size (i.types[op]);
2541 }
2542
2543 if (fake_zero_displacement)
2544 {
2545 /* Fakes a zero displacement assuming that i.types[op]
2546 holds the correct displacement size. */
2547 expressionS *exp;
2548
2549 assert (i.op[op].disps == 0);
2550 exp = &disp_expressions[i.disp_operands++];
2551 i.op[op].disps = exp;
2552 exp->X_op = O_constant;
2553 exp->X_add_number = 0;
2554 exp->X_add_symbol = (symbolS *) 0;
2555 exp->X_op_symbol = (symbolS *) 0;
2556 }
2557 }
2558
2559 /* Fill in i.rm.reg or i.rm.regmem field with register
2560 operand (if any) based on i.tm.extension_opcode.
2561 Again, we must be careful to make sure that
2562 segment/control/debug/test/MMX registers are coded
2563 into the i.rm.reg field. */
2564 if (i.reg_operands)
2565 {
2566 unsigned int op =
2567 ((i.types[0]
2568 & (Reg | RegMMX | RegXMM
2569 | SReg2 | SReg3
2570 | Control | Debug | Test))
2571 ? 0
2572 : ((i.types[1]
2573 & (Reg | RegMMX | RegXMM
2574 | SReg2 | SReg3
2575 | Control | Debug | Test))
2576 ? 1
2577 : 2));
2578 /* If there is an extension opcode to put here, the
2579 register number must be put into the regmem field. */
2580 if (i.tm.extension_opcode != None)
2581 {
2582 i.rm.regmem = i.op[op].regs->reg_num;
2583 if (i.op[op].regs->reg_flags & RegRex)
2584 i.rex.extZ = 1;
2585 }
2586 else
2587 {
2588 i.rm.reg = i.op[op].regs->reg_num;
2589 if (i.op[op].regs->reg_flags & RegRex)
2590 i.rex.extX = 1;
2591 }
2592
2593 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2
2594 we must set it to 3 to indicate this is a register
2595 operand in the regmem field. */
2596 if (!i.mem_operands)
2597 i.rm.mode = 3;
2598 }
2599
2600 /* Fill in i.rm.reg field with extension opcode (if any). */
2601 if (i.tm.extension_opcode != None)
2602 i.rm.reg = i.tm.extension_opcode;
2603 }
2604 }
2605 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
2606 {
2607 if (i.tm.base_opcode == POP_SEG_SHORT
2608 && i.op[0].regs->reg_num == 1)
2609 {
2610 as_bad (_("you can't `pop %%cs'"));
2611 return;
2612 }
2613 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
2614 if (i.op[0].regs->reg_flags & RegRex)
2615 i.rex.extZ = 1;
2616 }
2617 else if ((i.tm.base_opcode & ~(D|W)) == MOV_AX_DISP32)
2618 {
2619 default_seg = &ds;
2620 }
2621 else if ((i.tm.opcode_modifier & IsString) != 0)
2622 {
2623 /* For the string instructions that allow a segment override
2624 on one of their operands, the default segment is ds. */
2625 default_seg = &ds;
2626 }
2627
2628 /* If a segment was explicitly specified,
2629 and the specified segment is not the default,
2630 use an opcode prefix to select it.
2631 If we never figured out what the default segment is,
2632 then default_seg will be zero at this point,
2633 and the specified segment prefix will always be used. */
2634 if ((i.seg[0]) && (i.seg[0] != default_seg))
2635 {
2636 if (! add_prefix (i.seg[0]->seg_prefix))
2637 return;
2638 }
2639 }
2640 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2641 {
2642 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
2643 as_warn (_("translating to `%sp'"), i.tm.name);
2644 }
2645 }
2646
2647 /* Handle conversion of 'int $3' --> special int3 insn. */
2648 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
2649 {
2650 i.tm.base_opcode = INT3_OPCODE;
2651 i.imm_operands = 0;
2652 }
2653
2654 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
2655 && i.op[0].disps->X_op == O_constant)
2656 {
2657 /* Convert "jmp constant" (and "call constant") to a jump (call) to
2658 the absolute address given by the constant. Since ix86 jumps and
2659 calls are pc relative, we need to generate a reloc. */
2660 i.op[0].disps->X_add_symbol = &abs_symbol;
2661 i.op[0].disps->X_op = O_symbol;
2662 }
2663
2664 if (i.tm.opcode_modifier & Rex64)
2665 i.rex.mode64 = 1;
2666
2667 /* For 8bit registers we would need an empty rex prefix.
2668 Also in the case instruction is already having prefix,
2669 we need to convert old registers to new ones. */
2670
2671 if (((i.types[0] & Reg8) && (i.op[0].regs->reg_flags & RegRex64))
2672 || ((i.types[1] & Reg8) && (i.op[1].regs->reg_flags & RegRex64))
2673 || ((i.rex.mode64 || i.rex.extX || i.rex.extY || i.rex.extZ || i.rex.empty)
2674 && ((i.types[0] & Reg8) || (i.types[1] & Reg8))))
2675 {
2676 int x;
2677 i.rex.empty = 1;
2678 for (x = 0; x < 2; x++)
2679 {
2680 /* Look for 8bit operand that does use old registers. */
2681 if (i.types[x] & Reg8
2682 && !(i.op[x].regs->reg_flags & RegRex64))
2683 {
2684 /* In case it is "hi" register, give up. */
2685 if (i.op[x].regs->reg_num > 3)
2686 as_bad (_("Can't encode registers '%%%s' in the instruction requiring REX prefix.\n"),
2687 i.op[x].regs->reg_name);
2688
2689 /* Otherwise it is equivalent to the extended register.
2690 Since the encoding don't change this is merely cosmetical
2691 cleanup for debug output. */
2692
2693 i.op[x].regs = i.op[x].regs + 8;
2694 }
2695 }
2696 }
2697
2698 if (i.rex.mode64 || i.rex.extX || i.rex.extY || i.rex.extZ || i.rex.empty)
2699 add_prefix (0x40
2700 | (i.rex.mode64 ? 8 : 0)
2701 | (i.rex.extX ? 4 : 0)
2702 | (i.rex.extY ? 2 : 0)
2703 | (i.rex.extZ ? 1 : 0));
2704
2705 /* We are ready to output the insn. */
2706 {
2707 register char *p;
2708
2709 /* Output jumps. */
2710 if (i.tm.opcode_modifier & Jump)
2711 {
2712 int code16;
2713 int prefix;
2714
2715 code16 = 0;
2716 if (flag_code == CODE_16BIT)
2717 code16 = CODE16;
2718
2719 prefix = 0;
2720 if (i.prefix[DATA_PREFIX])
2721 {
2722 prefix = 1;
2723 i.prefixes -= 1;
2724 code16 ^= CODE16;
2725 }
2726 if (i.prefix[REX_PREFIX])
2727 {
2728 prefix++;
2729 i.prefixes--;
2730 }
2731
2732 if (i.prefixes != 0 && !intel_syntax)
2733 as_warn (_("skipping prefixes on this instruction"));
2734
2735 /* It's always a symbol; End frag & setup for relax.
2736 Make sure there is enough room in this frag for the largest
2737 instruction we may generate in md_convert_frag. This is 2
2738 bytes for the opcode and room for the prefix and largest
2739 displacement. */
2740 frag_grow (prefix + 2 + 4);
2741 insn_size += prefix + 1;
2742 /* Prefix and 1 opcode byte go in fr_fix. */
2743 p = frag_more (prefix + 1);
2744 if (i.prefix[DATA_PREFIX])
2745 *p++ = DATA_PREFIX_OPCODE;
2746 if (i.prefix[REX_PREFIX])
2747 *p++ = i.prefix[REX_PREFIX];
2748 *p = i.tm.base_opcode;
2749 /* 1 possible extra opcode + displacement go in var part.
2750 Pass reloc in fr_var. */
2751 frag_var (rs_machine_dependent,
2752 1 + 4,
2753 i.disp_reloc[0],
2754 ((unsigned char) *p == JUMP_PC_RELATIVE
2755 ? ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL) | code16
2756 : ((cpu_arch_flags & Cpu386) != 0
2757 ? ENCODE_RELAX_STATE (COND_JUMP, SMALL) | code16
2758 : ENCODE_RELAX_STATE (COND_JUMP86, SMALL) | code16)),
2759 i.op[0].disps->X_add_symbol,
2760 i.op[0].disps->X_add_number,
2761 p);
2762 }
2763 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
2764 {
2765 int size;
2766
2767 if (i.tm.opcode_modifier & JumpByte)
2768 {
2769 /* This is a loop or jecxz type instruction. */
2770 size = 1;
2771 if (i.prefix[ADDR_PREFIX])
2772 {
2773 insn_size += 1;
2774 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
2775 i.prefixes -= 1;
2776 }
2777 }
2778 else
2779 {
2780 int code16;
2781
2782 code16 = 0;
2783 if (flag_code == CODE_16BIT)
2784 code16 = CODE16;
2785
2786 if (i.prefix[DATA_PREFIX])
2787 {
2788 insn_size += 1;
2789 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
2790 i.prefixes -= 1;
2791 code16 ^= CODE16;
2792 }
2793
2794 size = 4;
2795 if (code16)
2796 size = 2;
2797 }
2798
2799 if (i.prefix[REX_PREFIX])
2800 {
2801 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
2802 insn_size++;
2803 i.prefixes -= 1;
2804 }
2805
2806 if (i.prefixes != 0 && !intel_syntax)
2807 as_warn (_("skipping prefixes on this instruction"));
2808
2809 if (fits_in_unsigned_byte (i.tm.base_opcode))
2810 {
2811 insn_size += 1 + size;
2812 p = frag_more (1 + size);
2813 }
2814 else
2815 {
2816 /* Opcode can be at most two bytes. */
2817 insn_size += 2 + size;
2818 p = frag_more (2 + size);
2819 *p++ = (i.tm.base_opcode >> 8) & 0xff;
2820 }
2821 *p++ = i.tm.base_opcode & 0xff;
2822
2823 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2824 i.op[0].disps, 1, reloc (size, 1, 1, i.disp_reloc[0]));
2825 }
2826 else if (i.tm.opcode_modifier & JumpInterSegment)
2827 {
2828 int size;
2829 int prefix;
2830 int code16;
2831
2832 code16 = 0;
2833 if (flag_code == CODE_16BIT)
2834 code16 = CODE16;
2835
2836 prefix = 0;
2837 if (i.prefix[DATA_PREFIX])
2838 {
2839 prefix = 1;
2840 i.prefixes -= 1;
2841 code16 ^= CODE16;
2842 }
2843 if (i.prefix[REX_PREFIX])
2844 {
2845 prefix++;
2846 i.prefixes -= 1;
2847 }
2848
2849 size = 4;
2850 if (code16)
2851 size = 2;
2852
2853 if (i.prefixes != 0 && !intel_syntax)
2854 as_warn (_("skipping prefixes on this instruction"));
2855
2856 /* 1 opcode; 2 segment; offset */
2857 insn_size += prefix + 1 + 2 + size;
2858 p = frag_more (prefix + 1 + 2 + size);
2859
2860 if (i.prefix[DATA_PREFIX])
2861 *p++ = DATA_PREFIX_OPCODE;
2862
2863 if (i.prefix[REX_PREFIX])
2864 *p++ = i.prefix[REX_PREFIX];
2865
2866 *p++ = i.tm.base_opcode;
2867 if (i.op[1].imms->X_op == O_constant)
2868 {
2869 offsetT n = i.op[1].imms->X_add_number;
2870
2871 if (size == 2
2872 && !fits_in_unsigned_word (n)
2873 && !fits_in_signed_word (n))
2874 {
2875 as_bad (_("16-bit jump out of range"));
2876 return;
2877 }
2878 md_number_to_chars (p, n, size);
2879 }
2880 else
2881 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2882 i.op[1].imms, 0, reloc (size, 0, 0, i.disp_reloc[0]));
2883 if (i.op[0].imms->X_op != O_constant)
2884 as_bad (_("can't handle non absolute segment in `%s'"),
2885 i.tm.name);
2886 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
2887 }
2888 else
2889 {
2890 /* Output normal instructions here. */
2891 unsigned char *q;
2892
2893 /* All opcodes on i386 have eighter 1 or 2 bytes. We may use third
2894 byte for the SSE instructions to specify prefix they require. */
2895 if (i.tm.base_opcode & 0xff0000)
2896 add_prefix ((i.tm.base_opcode >> 16) & 0xff);
2897
2898 /* The prefix bytes. */
2899 for (q = i.prefix;
2900 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
2901 q++)
2902 {
2903 if (*q)
2904 {
2905 insn_size += 1;
2906 p = frag_more (1);
2907 md_number_to_chars (p, (valueT) *q, 1);
2908 }
2909 }
2910
2911 /* Now the opcode; be careful about word order here! */
2912 if (fits_in_unsigned_byte (i.tm.base_opcode))
2913 {
2914 insn_size += 1;
2915 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
2916 }
2917 else
2918 {
2919 insn_size += 2;
2920 p = frag_more (2);
2921 /* Put out high byte first: can't use md_number_to_chars! */
2922 *p++ = (i.tm.base_opcode >> 8) & 0xff;
2923 *p = i.tm.base_opcode & 0xff;
2924 }
2925
2926 /* Now the modrm byte and sib byte (if present). */
2927 if (i.tm.opcode_modifier & Modrm)
2928 {
2929 insn_size += 1;
2930 p = frag_more (1);
2931 md_number_to_chars (p,
2932 (valueT) (i.rm.regmem << 0
2933 | i.rm.reg << 3
2934 | i.rm.mode << 6),
2935 1);
2936 /* If i.rm.regmem == ESP (4)
2937 && i.rm.mode != (Register mode)
2938 && not 16 bit
2939 ==> need second modrm byte. */
2940 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
2941 && i.rm.mode != 3
2942 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
2943 {
2944 insn_size += 1;
2945 p = frag_more (1);
2946 md_number_to_chars (p,
2947 (valueT) (i.sib.base << 0
2948 | i.sib.index << 3
2949 | i.sib.scale << 6),
2950 1);
2951 }
2952 }
2953
2954 if (i.disp_operands)
2955 {
2956 register unsigned int n;
2957
2958 for (n = 0; n < i.operands; n++)
2959 {
2960 if (i.types[n] & Disp)
2961 {
2962 if (i.op[n].disps->X_op == O_constant)
2963 {
2964 int size;
2965 offsetT val;
2966
2967 size = 4;
2968 if (i.types[n] & (Disp8 | Disp16 | Disp64))
2969 {
2970 size = 2;
2971 if (i.types[n] & Disp8)
2972 size = 1;
2973 if (i.types[n] & Disp64)
2974 size = 8;
2975 }
2976 val = offset_in_range (i.op[n].disps->X_add_number,
2977 size);
2978 insn_size += size;
2979 p = frag_more (size);
2980 md_number_to_chars (p, val, size);
2981 }
2982 else
2983 {
2984 int size = 4;
2985 int sign = 0;
2986 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
2987
2988 /* The PC relative address is computed relative
2989 to the instruction boundary, so in case immediate
2990 fields follows, we need to adjust the value. */
2991 if (pcrel && i.imm_operands)
2992 {
2993 int imm_size = 4;
2994 register unsigned int n1;
2995
2996 for (n1 = 0; n1 < i.operands; n1++)
2997 if (i.types[n1] & Imm)
2998 {
2999 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
3000 {
3001 imm_size = 2;
3002 if (i.types[n1] & (Imm8 | Imm8S))
3003 imm_size = 1;
3004 if (i.types[n1] & Imm64)
3005 imm_size = 8;
3006 }
3007 break;
3008 }
3009 /* We should find the immediate. */
3010 if (n1 == i.operands)
3011 abort ();
3012 i.op[n].disps->X_add_number -= imm_size;
3013 }
3014
3015 if (i.types[n] & Disp32S)
3016 sign = 1;
3017
3018 if (i.types[n] & (Disp16 | Disp64))
3019 {
3020 size = 2;
3021 if (i.types[n] & Disp64)
3022 size = 8;
3023 }
3024
3025 insn_size += size;
3026 p = frag_more (size);
3027 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3028 i.op[n].disps, pcrel,
3029 reloc (size, pcrel, sign, i.disp_reloc[n]));
3030 }
3031 }
3032 }
3033 }
3034
3035 /* Output immediate. */
3036 if (i.imm_operands)
3037 {
3038 register unsigned int n;
3039
3040 for (n = 0; n < i.operands; n++)
3041 {
3042 if (i.types[n] & Imm)
3043 {
3044 if (i.op[n].imms->X_op == O_constant)
3045 {
3046 int size;
3047 offsetT val;
3048
3049 size = 4;
3050 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3051 {
3052 size = 2;
3053 if (i.types[n] & (Imm8 | Imm8S))
3054 size = 1;
3055 else if (i.types[n] & Imm64)
3056 size = 8;
3057 }
3058 val = offset_in_range (i.op[n].imms->X_add_number,
3059 size);
3060 insn_size += size;
3061 p = frag_more (size);
3062 md_number_to_chars (p, val, size);
3063 }
3064 else
3065 {
3066 /* Not absolute_section.
3067 Need a 32-bit fixup (don't support 8bit
3068 non-absolute imms). Try to support other
3069 sizes ... */
3070 #ifdef BFD_ASSEMBLER
3071 enum bfd_reloc_code_real reloc_type;
3072 #else
3073 int reloc_type;
3074 #endif
3075 int size = 4;
3076 int sign = 0;
3077
3078 if ((i.types[n] & (Imm32S))
3079 && i.suffix == QWORD_MNEM_SUFFIX)
3080 sign = 1;
3081 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3082 {
3083 size = 2;
3084 if (i.types[n] & (Imm8 | Imm8S))
3085 size = 1;
3086 if (i.types[n] & Imm64)
3087 size = 8;
3088 }
3089
3090 insn_size += size;
3091 p = frag_more (size);
3092 reloc_type = reloc (size, 0, sign, i.disp_reloc[0]);
3093 #ifdef BFD_ASSEMBLER
3094 if (reloc_type == BFD_RELOC_32
3095 && GOT_symbol
3096 && GOT_symbol == i.op[n].imms->X_add_symbol
3097 && (i.op[n].imms->X_op == O_symbol
3098 || (i.op[n].imms->X_op == O_add
3099 && ((symbol_get_value_expression
3100 (i.op[n].imms->X_op_symbol)->X_op)
3101 == O_subtract))))
3102 {
3103 /* We don't support dynamic linking on x86-64 yet. */
3104 if (flag_code == CODE_64BIT)
3105 abort ();
3106 reloc_type = BFD_RELOC_386_GOTPC;
3107 i.op[n].imms->X_add_number += 3;
3108 }
3109 #endif
3110 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3111 i.op[n].imms, 0, reloc_type);
3112 }
3113 }
3114 }
3115 }
3116 }
3117
3118 dwarf2_emit_insn (insn_size);
3119
3120 #ifdef DEBUG386
3121 if (flag_debug)
3122 {
3123 pi (line, &i);
3124 }
3125 #endif /* DEBUG386 */
3126 }
3127 }
3128 \f
3129 static int i386_immediate PARAMS ((char *));
3130
3131 static int
3132 i386_immediate (imm_start)
3133 char *imm_start;
3134 {
3135 char *save_input_line_pointer;
3136 segT exp_seg = 0;
3137 expressionS *exp;
3138
3139 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
3140 {
3141 as_bad (_("only 1 or 2 immediate operands are allowed"));
3142 return 0;
3143 }
3144
3145 exp = &im_expressions[i.imm_operands++];
3146 i.op[this_operand].imms = exp;
3147
3148 if (is_space_char (*imm_start))
3149 ++imm_start;
3150
3151 save_input_line_pointer = input_line_pointer;
3152 input_line_pointer = imm_start;
3153
3154 #ifndef LEX_AT
3155 {
3156 /* We can have operands of the form
3157 <symbol>@GOTOFF+<nnn>
3158 Take the easy way out here and copy everything
3159 into a temporary buffer... */
3160 register char *cp;
3161
3162 cp = strchr (input_line_pointer, '@');
3163 if (cp != NULL)
3164 {
3165 char *tmpbuf;
3166 int len = 0;
3167 int first;
3168
3169 /* GOT relocations are not supported in 16 bit mode. */
3170 if (flag_code == CODE_16BIT)
3171 as_bad (_("GOT relocations not supported in 16 bit mode"));
3172
3173 if (GOT_symbol == NULL)
3174 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3175
3176 if (strncmp (cp + 1, "PLT", 3) == 0)
3177 {
3178 if (flag_code == CODE_64BIT)
3179 i.disp_reloc[this_operand] = BFD_RELOC_X86_64_PLT32;
3180 else
3181 i.disp_reloc[this_operand] = BFD_RELOC_386_PLT32;
3182 len = 3;
3183 }
3184 else if (strncmp (cp + 1, "GOTOFF", 6) == 0)
3185 {
3186 if (flag_code == CODE_64BIT)
3187 as_bad ("GOTOFF relocations are unsupported in 64bit mode.");
3188 i.disp_reloc[this_operand] = BFD_RELOC_386_GOTOFF;
3189 len = 6;
3190 }
3191 else if (strncmp (cp + 1, "GOTPCREL", 8) == 0)
3192 {
3193 if (flag_code == CODE_64BIT)
3194 i.disp_reloc[this_operand] = BFD_RELOC_X86_64_GOTPCREL;
3195 else
3196 as_bad ("GOTPCREL relocations are supported only in 64bit mode.");
3197 len = 8;
3198 }
3199 else if (strncmp (cp + 1, "GOT", 3) == 0)
3200 {
3201 if (flag_code == CODE_64BIT)
3202 i.disp_reloc[this_operand] = BFD_RELOC_X86_64_GOT32;
3203 else
3204 i.disp_reloc[this_operand] = BFD_RELOC_386_GOT32;
3205 len = 3;
3206 }
3207 else
3208 as_bad (_("bad reloc specifier in expression"));
3209
3210 /* Replace the relocation token with ' ', so that errors like
3211 foo@GOTOFF1 will be detected. */
3212 first = cp - input_line_pointer;
3213 tmpbuf = (char *) alloca (strlen (input_line_pointer));
3214 memcpy (tmpbuf, input_line_pointer, first);
3215 tmpbuf[first] = ' ';
3216 strcpy (tmpbuf + first + 1, cp + 1 + len);
3217 input_line_pointer = tmpbuf;
3218 }
3219 }
3220 #endif
3221
3222 exp_seg = expression (exp);
3223
3224 SKIP_WHITESPACE ();
3225 if (*input_line_pointer)
3226 as_bad (_("ignoring junk `%s' after expression"), input_line_pointer);
3227
3228 input_line_pointer = save_input_line_pointer;
3229
3230 if (exp->X_op == O_absent || exp->X_op == O_big)
3231 {
3232 /* Missing or bad expr becomes absolute 0. */
3233 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
3234 imm_start);
3235 exp->X_op = O_constant;
3236 exp->X_add_number = 0;
3237 exp->X_add_symbol = (symbolS *) 0;
3238 exp->X_op_symbol = (symbolS *) 0;
3239 }
3240 else if (exp->X_op == O_constant)
3241 {
3242 /* Size it properly later. */
3243 i.types[this_operand] |= Imm64;
3244 /* If BFD64, sign extend val. */
3245 if (!use_rela_relocations)
3246 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
3247 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
3248 }
3249 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3250 else if (1
3251 #ifdef BFD_ASSEMBLER
3252 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3253 #endif
3254 && exp_seg != text_section
3255 && exp_seg != data_section
3256 && exp_seg != bss_section
3257 && exp_seg != undefined_section
3258 #ifdef BFD_ASSEMBLER
3259 && !bfd_is_com_section (exp_seg)
3260 #endif
3261 )
3262 {
3263 #ifdef BFD_ASSEMBLER
3264 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3265 #else
3266 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3267 #endif
3268 return 0;
3269 }
3270 #endif
3271 else
3272 {
3273 /* This is an address. The size of the address will be
3274 determined later, depending on destination register,
3275 suffix, or the default for the section. */
3276 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
3277 }
3278
3279 return 1;
3280 }
3281
3282 static int i386_scale PARAMS ((char *));
3283
3284 static int
3285 i386_scale (scale)
3286 char *scale;
3287 {
3288 if (!isdigit (*scale))
3289 goto bad_scale;
3290
3291 switch (*scale)
3292 {
3293 case '0':
3294 case '1':
3295 i.log2_scale_factor = 0;
3296 break;
3297 case '2':
3298 i.log2_scale_factor = 1;
3299 break;
3300 case '4':
3301 i.log2_scale_factor = 2;
3302 break;
3303 case '8':
3304 i.log2_scale_factor = 3;
3305 break;
3306 default:
3307 bad_scale:
3308 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
3309 scale);
3310 return 0;
3311 }
3312 if (i.log2_scale_factor != 0 && ! i.index_reg)
3313 {
3314 as_warn (_("scale factor of %d without an index register"),
3315 1 << i.log2_scale_factor);
3316 #if SCALE1_WHEN_NO_INDEX
3317 i.log2_scale_factor = 0;
3318 #endif
3319 }
3320 return 1;
3321 }
3322
3323 static int i386_displacement PARAMS ((char *, char *));
3324
3325 static int
3326 i386_displacement (disp_start, disp_end)
3327 char *disp_start;
3328 char *disp_end;
3329 {
3330 register expressionS *exp;
3331 segT exp_seg = 0;
3332 char *save_input_line_pointer;
3333 int bigdisp = Disp32;
3334
3335 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3336 bigdisp = Disp16;
3337 if (flag_code == CODE_64BIT)
3338 bigdisp = Disp64;
3339 i.types[this_operand] |= bigdisp;
3340
3341 exp = &disp_expressions[i.disp_operands];
3342 i.op[this_operand].disps = exp;
3343 i.disp_operands++;
3344 save_input_line_pointer = input_line_pointer;
3345 input_line_pointer = disp_start;
3346 END_STRING_AND_SAVE (disp_end);
3347
3348 #ifndef GCC_ASM_O_HACK
3349 #define GCC_ASM_O_HACK 0
3350 #endif
3351 #if GCC_ASM_O_HACK
3352 END_STRING_AND_SAVE (disp_end + 1);
3353 if ((i.types[this_operand] & BaseIndex) != 0
3354 && displacement_string_end[-1] == '+')
3355 {
3356 /* This hack is to avoid a warning when using the "o"
3357 constraint within gcc asm statements.
3358 For instance:
3359
3360 #define _set_tssldt_desc(n,addr,limit,type) \
3361 __asm__ __volatile__ ( \
3362 "movw %w2,%0\n\t" \
3363 "movw %w1,2+%0\n\t" \
3364 "rorl $16,%1\n\t" \
3365 "movb %b1,4+%0\n\t" \
3366 "movb %4,5+%0\n\t" \
3367 "movb $0,6+%0\n\t" \
3368 "movb %h1,7+%0\n\t" \
3369 "rorl $16,%1" \
3370 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
3371
3372 This works great except that the output assembler ends
3373 up looking a bit weird if it turns out that there is
3374 no offset. You end up producing code that looks like:
3375
3376 #APP
3377 movw $235,(%eax)
3378 movw %dx,2+(%eax)
3379 rorl $16,%edx
3380 movb %dl,4+(%eax)
3381 movb $137,5+(%eax)
3382 movb $0,6+(%eax)
3383 movb %dh,7+(%eax)
3384 rorl $16,%edx
3385 #NO_APP
3386
3387 So here we provide the missing zero. */
3388
3389 *displacement_string_end = '0';
3390 }
3391 #endif
3392 #ifndef LEX_AT
3393 {
3394 /* We can have operands of the form
3395 <symbol>@GOTOFF+<nnn>
3396 Take the easy way out here and copy everything
3397 into a temporary buffer... */
3398 register char *cp;
3399
3400 cp = strchr (input_line_pointer, '@');
3401 if (cp != NULL)
3402 {
3403 char *tmpbuf;
3404 int len = 0;
3405 int first;
3406
3407 /* GOT relocations are not supported in 16 bit mode. */
3408 if (flag_code == CODE_16BIT)
3409 as_bad (_("GOT relocations not supported in 16 bit mode"));
3410
3411 if (GOT_symbol == NULL)
3412 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3413
3414 if (strncmp (cp + 1, "PLT", 3) == 0)
3415 {
3416 if (flag_code == CODE_64BIT)
3417 i.disp_reloc[this_operand] = BFD_RELOC_X86_64_PLT32;
3418 else
3419 i.disp_reloc[this_operand] = BFD_RELOC_386_PLT32;
3420 len = 3;
3421 }
3422 else if (strncmp (cp + 1, "GOTOFF", 6) == 0)
3423 {
3424 if (flag_code == CODE_64BIT)
3425 as_bad ("GOTOFF relocation is not supported in 64bit mode.");
3426 i.disp_reloc[this_operand] = BFD_RELOC_386_GOTOFF;
3427 len = 6;
3428 }
3429 else if (strncmp (cp + 1, "GOTPCREL", 8) == 0)
3430 {
3431 if (flag_code != CODE_64BIT)
3432 as_bad ("GOTPCREL relocation is supported only in 64bit mode.");
3433 i.disp_reloc[this_operand] = BFD_RELOC_X86_64_GOTPCREL;
3434 len = 8;
3435 }
3436 else if (strncmp (cp + 1, "GOT", 3) == 0)
3437 {
3438 if (flag_code == CODE_64BIT)
3439 i.disp_reloc[this_operand] = BFD_RELOC_X86_64_GOT32;
3440 else
3441 i.disp_reloc[this_operand] = BFD_RELOC_386_GOT32;
3442 len = 3;
3443 }
3444 else
3445 as_bad (_("bad reloc specifier in expression"));
3446
3447 /* Replace the relocation token with ' ', so that errors like
3448 foo@GOTOFF1 will be detected. */
3449 first = cp - input_line_pointer;
3450 tmpbuf = (char *) alloca (strlen (input_line_pointer));
3451 memcpy (tmpbuf, input_line_pointer, first);
3452 tmpbuf[first] = ' ';
3453 strcpy (tmpbuf + first + 1, cp + 1 + len);
3454 input_line_pointer = tmpbuf;
3455 }
3456 }
3457 #endif
3458
3459 exp_seg = expression (exp);
3460
3461 #ifdef BFD_ASSEMBLER
3462 /* We do this to make sure that the section symbol is in
3463 the symbol table. We will ultimately change the relocation
3464 to be relative to the beginning of the section. */
3465 if (i.disp_reloc[this_operand] == BFD_RELOC_386_GOTOFF
3466 || i.disp_reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3467 {
3468 if (S_IS_LOCAL (exp->X_add_symbol)
3469 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
3470 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
3471 assert (exp->X_op == O_symbol);
3472 exp->X_op = O_subtract;
3473 exp->X_op_symbol = GOT_symbol;
3474 if (i.disp_reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3475 i.disp_reloc[this_operand] = BFD_RELOC_32_PCREL;
3476 else
3477 i.disp_reloc[this_operand] = BFD_RELOC_32;
3478 }
3479 #endif
3480
3481 SKIP_WHITESPACE ();
3482 if (*input_line_pointer)
3483 as_bad (_("ignoring junk `%s' after expression"),
3484 input_line_pointer);
3485 #if GCC_ASM_O_HACK
3486 RESTORE_END_STRING (disp_end + 1);
3487 #endif
3488 RESTORE_END_STRING (disp_end);
3489 input_line_pointer = save_input_line_pointer;
3490
3491 if (exp->X_op == O_absent || exp->X_op == O_big)
3492 {
3493 /* Missing or bad expr becomes absolute 0. */
3494 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
3495 disp_start);
3496 exp->X_op = O_constant;
3497 exp->X_add_number = 0;
3498 exp->X_add_symbol = (symbolS *) 0;
3499 exp->X_op_symbol = (symbolS *) 0;
3500 }
3501
3502 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3503 if (exp->X_op != O_constant
3504 #ifdef BFD_ASSEMBLER
3505 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3506 #endif
3507 && exp_seg != text_section
3508 && exp_seg != data_section
3509 && exp_seg != bss_section
3510 && exp_seg != undefined_section)
3511 {
3512 #ifdef BFD_ASSEMBLER
3513 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3514 #else
3515 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3516 #endif
3517 return 0;
3518 }
3519 #endif
3520 else if (flag_code == CODE_64BIT)
3521 i.types[this_operand] |= Disp32S | Disp32;
3522 return 1;
3523 }
3524
3525 static int i386_index_check PARAMS ((const char *));
3526
3527 /* Make sure the memory operand we've been dealt is valid.
3528 Return 1 on success, 0 on a failure. */
3529
3530 static int
3531 i386_index_check (operand_string)
3532 const char *operand_string;
3533 {
3534 int ok;
3535 #if INFER_ADDR_PREFIX
3536 int fudged = 0;
3537
3538 tryprefix:
3539 #endif
3540 ok = 1;
3541 if (flag_code == CODE_64BIT)
3542 {
3543 /* 64bit checks. */
3544 if ((i.base_reg
3545 && ((i.base_reg->reg_type & Reg64) == 0)
3546 && (i.base_reg->reg_type != BaseIndex
3547 || i.index_reg))
3548 || (i.index_reg
3549 && ((i.index_reg->reg_type & (Reg64|BaseIndex))
3550 != (Reg64|BaseIndex))))
3551 ok = 0;
3552 }
3553 else
3554 {
3555 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3556 {
3557 /* 16bit checks. */
3558 if ((i.base_reg
3559 && ((i.base_reg->reg_type & (Reg16|BaseIndex|RegRex))
3560 != (Reg16|BaseIndex)))
3561 || (i.index_reg
3562 && (((i.index_reg->reg_type & (Reg16|BaseIndex))
3563 != (Reg16|BaseIndex))
3564 || ! (i.base_reg
3565 && i.base_reg->reg_num < 6
3566 && i.index_reg->reg_num >= 6
3567 && i.log2_scale_factor == 0))))
3568 ok = 0;
3569 }
3570 else
3571 {
3572 /* 32bit checks. */
3573 if ((i.base_reg
3574 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
3575 || (i.index_reg
3576 && ((i.index_reg->reg_type & (Reg32|BaseIndex|RegRex))
3577 != (Reg32|BaseIndex))))
3578 ok = 0;
3579 }
3580 }
3581 if (!ok)
3582 {
3583 #if INFER_ADDR_PREFIX
3584 if (flag_code != CODE_64BIT
3585 && i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0')
3586 {
3587 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
3588 i.prefixes += 1;
3589 /* Change the size of any displacement too. At most one of
3590 Disp16 or Disp32 is set.
3591 FIXME. There doesn't seem to be any real need for separate
3592 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
3593 Removing them would probably clean up the code quite a lot. */
3594 if (i.types[this_operand] & (Disp16|Disp32))
3595 i.types[this_operand] ^= (Disp16|Disp32);
3596 fudged = 1;
3597 goto tryprefix;
3598 }
3599 if (fudged)
3600 as_bad (_("`%s' is not a valid base/index expression"),
3601 operand_string);
3602 else
3603 #endif
3604 as_bad (_("`%s' is not a valid %s bit base/index expression"),
3605 operand_string,
3606 flag_code_names[flag_code]);
3607 return 0;
3608 }
3609 return 1;
3610 }
3611
3612 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
3613 on error. */
3614
3615 static int
3616 i386_operand (operand_string)
3617 char *operand_string;
3618 {
3619 const reg_entry *r;
3620 char *end_op;
3621 char *op_string = operand_string;
3622
3623 if (is_space_char (*op_string))
3624 ++op_string;
3625
3626 /* We check for an absolute prefix (differentiating,
3627 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
3628 if (*op_string == ABSOLUTE_PREFIX)
3629 {
3630 ++op_string;
3631 if (is_space_char (*op_string))
3632 ++op_string;
3633 i.types[this_operand] |= JumpAbsolute;
3634 }
3635
3636 /* Check if operand is a register. */
3637 if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
3638 && (r = parse_register (op_string, &end_op)) != NULL)
3639 {
3640 /* Check for a segment override by searching for ':' after a
3641 segment register. */
3642 op_string = end_op;
3643 if (is_space_char (*op_string))
3644 ++op_string;
3645 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
3646 {
3647 switch (r->reg_num)
3648 {
3649 case 0:
3650 i.seg[i.mem_operands] = &es;
3651 break;
3652 case 1:
3653 i.seg[i.mem_operands] = &cs;
3654 break;
3655 case 2:
3656 i.seg[i.mem_operands] = &ss;
3657 break;
3658 case 3:
3659 i.seg[i.mem_operands] = &ds;
3660 break;
3661 case 4:
3662 i.seg[i.mem_operands] = &fs;
3663 break;
3664 case 5:
3665 i.seg[i.mem_operands] = &gs;
3666 break;
3667 }
3668
3669 /* Skip the ':' and whitespace. */
3670 ++op_string;
3671 if (is_space_char (*op_string))
3672 ++op_string;
3673
3674 if (!is_digit_char (*op_string)
3675 && !is_identifier_char (*op_string)
3676 && *op_string != '('
3677 && *op_string != ABSOLUTE_PREFIX)
3678 {
3679 as_bad (_("bad memory operand `%s'"), op_string);
3680 return 0;
3681 }
3682 /* Handle case of %es:*foo. */
3683 if (*op_string == ABSOLUTE_PREFIX)
3684 {
3685 ++op_string;
3686 if (is_space_char (*op_string))
3687 ++op_string;
3688 i.types[this_operand] |= JumpAbsolute;
3689 }
3690 goto do_memory_reference;
3691 }
3692 if (*op_string)
3693 {
3694 as_bad (_("junk `%s' after register"), op_string);
3695 return 0;
3696 }
3697 i.types[this_operand] |= r->reg_type & ~BaseIndex;
3698 i.op[this_operand].regs = r;
3699 i.reg_operands++;
3700 }
3701 else if (*op_string == REGISTER_PREFIX)
3702 {
3703 as_bad (_("bad register name `%s'"), op_string);
3704 return 0;
3705 }
3706 else if (*op_string == IMMEDIATE_PREFIX)
3707 {
3708 ++op_string;
3709 if (i.types[this_operand] & JumpAbsolute)
3710 {
3711 as_bad (_("immediate operand illegal with absolute jump"));
3712 return 0;
3713 }
3714 if (!i386_immediate (op_string))
3715 return 0;
3716 }
3717 else if (is_digit_char (*op_string)
3718 || is_identifier_char (*op_string)
3719 || *op_string == '(')
3720 {
3721 /* This is a memory reference of some sort. */
3722 char *base_string;
3723
3724 /* Start and end of displacement string expression (if found). */
3725 char *displacement_string_start;
3726 char *displacement_string_end;
3727
3728 do_memory_reference:
3729 if ((i.mem_operands == 1
3730 && (current_templates->start->opcode_modifier & IsString) == 0)
3731 || i.mem_operands == 2)
3732 {
3733 as_bad (_("too many memory references for `%s'"),
3734 current_templates->start->name);
3735 return 0;
3736 }
3737
3738 /* Check for base index form. We detect the base index form by
3739 looking for an ')' at the end of the operand, searching
3740 for the '(' matching it, and finding a REGISTER_PREFIX or ','
3741 after the '('. */
3742 base_string = op_string + strlen (op_string);
3743
3744 --base_string;
3745 if (is_space_char (*base_string))
3746 --base_string;
3747
3748 /* If we only have a displacement, set-up for it to be parsed later. */
3749 displacement_string_start = op_string;
3750 displacement_string_end = base_string + 1;
3751
3752 if (*base_string == ')')
3753 {
3754 char *temp_string;
3755 unsigned int parens_balanced = 1;
3756 /* We've already checked that the number of left & right ()'s are
3757 equal, so this loop will not be infinite. */
3758 do
3759 {
3760 base_string--;
3761 if (*base_string == ')')
3762 parens_balanced++;
3763 if (*base_string == '(')
3764 parens_balanced--;
3765 }
3766 while (parens_balanced);
3767
3768 temp_string = base_string;
3769
3770 /* Skip past '(' and whitespace. */
3771 ++base_string;
3772 if (is_space_char (*base_string))
3773 ++base_string;
3774
3775 if (*base_string == ','
3776 || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3777 && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
3778 {
3779 displacement_string_end = temp_string;
3780
3781 i.types[this_operand] |= BaseIndex;
3782
3783 if (i.base_reg)
3784 {
3785 base_string = end_op;
3786 if (is_space_char (*base_string))
3787 ++base_string;
3788 }
3789
3790 /* There may be an index reg or scale factor here. */
3791 if (*base_string == ',')
3792 {
3793 ++base_string;
3794 if (is_space_char (*base_string))
3795 ++base_string;
3796
3797 if ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3798 && (i.index_reg = parse_register (base_string, &end_op)) != NULL)
3799 {
3800 base_string = end_op;
3801 if (is_space_char (*base_string))
3802 ++base_string;
3803 if (*base_string == ',')
3804 {
3805 ++base_string;
3806 if (is_space_char (*base_string))
3807 ++base_string;
3808 }
3809 else if (*base_string != ')')
3810 {
3811 as_bad (_("expecting `,' or `)' after index register in `%s'"),
3812 operand_string);
3813 return 0;
3814 }
3815 }
3816 else if (*base_string == REGISTER_PREFIX)
3817 {
3818 as_bad (_("bad register name `%s'"), base_string);
3819 return 0;
3820 }
3821
3822 /* Check for scale factor. */
3823 if (isdigit ((unsigned char) *base_string))
3824 {
3825 if (!i386_scale (base_string))
3826 return 0;
3827
3828 ++base_string;
3829 if (is_space_char (*base_string))
3830 ++base_string;
3831 if (*base_string != ')')
3832 {
3833 as_bad (_("expecting `)' after scale factor in `%s'"),
3834 operand_string);
3835 return 0;
3836 }
3837 }
3838 else if (!i.index_reg)
3839 {
3840 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
3841 *base_string);
3842 return 0;
3843 }
3844 }
3845 else if (*base_string != ')')
3846 {
3847 as_bad (_("expecting `,' or `)' after base register in `%s'"),
3848 operand_string);
3849 return 0;
3850 }
3851 }
3852 else if (*base_string == REGISTER_PREFIX)
3853 {
3854 as_bad (_("bad register name `%s'"), base_string);
3855 return 0;
3856 }
3857 }
3858
3859 /* If there's an expression beginning the operand, parse it,
3860 assuming displacement_string_start and
3861 displacement_string_end are meaningful. */
3862 if (displacement_string_start != displacement_string_end)
3863 {
3864 if (!i386_displacement (displacement_string_start,
3865 displacement_string_end))
3866 return 0;
3867 }
3868
3869 /* Special case for (%dx) while doing input/output op. */
3870 if (i.base_reg
3871 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
3872 && i.index_reg == 0
3873 && i.log2_scale_factor == 0
3874 && i.seg[i.mem_operands] == 0
3875 && (i.types[this_operand] & Disp) == 0)
3876 {
3877 i.types[this_operand] = InOutPortReg;
3878 return 1;
3879 }
3880
3881 if (i386_index_check (operand_string) == 0)
3882 return 0;
3883 i.mem_operands++;
3884 }
3885 else
3886 {
3887 /* It's not a memory operand; argh! */
3888 as_bad (_("invalid char %s beginning operand %d `%s'"),
3889 output_invalid (*op_string),
3890 this_operand + 1,
3891 op_string);
3892 return 0;
3893 }
3894 return 1; /* Normal return. */
3895 }
3896 \f
3897 /* md_estimate_size_before_relax()
3898
3899 Called just before relax() for rs_machine_dependent frags. The x86
3900 assembler uses these frags to handle variable size jump
3901 instructions.
3902
3903 Any symbol that is now undefined will not become defined.
3904 Return the correct fr_subtype in the frag.
3905 Return the initial "guess for variable size of frag" to caller.
3906 The guess is actually the growth beyond the fixed part. Whatever
3907 we do to grow the fixed or variable part contributes to our
3908 returned value. */
3909
3910 int
3911 md_estimate_size_before_relax (fragP, segment)
3912 register fragS *fragP;
3913 register segT segment;
3914 {
3915 /* We've already got fragP->fr_subtype right; all we have to do is
3916 check for un-relaxable symbols. On an ELF system, we can't relax
3917 an externally visible symbol, because it may be overridden by a
3918 shared library. */
3919 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
3920 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
3921 || S_IS_EXTERNAL (fragP->fr_symbol)
3922 || S_IS_WEAK (fragP->fr_symbol)
3923 #endif
3924 )
3925 {
3926 /* Symbol is undefined in this segment, or we need to keep a
3927 reloc so that weak symbols can be overridden. */
3928 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
3929 #ifdef BFD_ASSEMBLER
3930 enum bfd_reloc_code_real reloc_type;
3931 #else
3932 int reloc_type;
3933 #endif
3934 unsigned char *opcode;
3935 int old_fr_fix;
3936
3937 if (fragP->fr_var != NO_RELOC)
3938 reloc_type = fragP->fr_var;
3939 else if (size == 2)
3940 reloc_type = BFD_RELOC_16_PCREL;
3941 else
3942 reloc_type = BFD_RELOC_32_PCREL;
3943
3944 old_fr_fix = fragP->fr_fix;
3945 opcode = (unsigned char *) fragP->fr_opcode;
3946
3947 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
3948 {
3949 case UNCOND_JUMP:
3950 /* Make jmp (0xeb) a (d)word displacement jump. */
3951 opcode[0] = 0xe9;
3952 fragP->fr_fix += size;
3953 fix_new (fragP, old_fr_fix, size,
3954 fragP->fr_symbol,
3955 fragP->fr_offset, 1,
3956 reloc_type);
3957 break;
3958
3959 case COND_JUMP86:
3960 if (no_cond_jump_promotion)
3961 return 1;
3962 if (size == 2)
3963 {
3964 /* Negate the condition, and branch past an
3965 unconditional jump. */
3966 opcode[0] ^= 1;
3967 opcode[1] = 3;
3968 /* Insert an unconditional jump. */
3969 opcode[2] = 0xe9;
3970 /* We added two extra opcode bytes, and have a two byte
3971 offset. */
3972 fragP->fr_fix += 2 + 2;
3973 fix_new (fragP, old_fr_fix + 2, 2,
3974 fragP->fr_symbol,
3975 fragP->fr_offset, 1,
3976 reloc_type);
3977 break;
3978 }
3979 /* Fall through. */
3980
3981 case COND_JUMP:
3982 if (no_cond_jump_promotion)
3983 return 1;
3984 /* This changes the byte-displacement jump 0x7N
3985 to the (d)word-displacement jump 0x0f,0x8N. */
3986 opcode[1] = opcode[0] + 0x10;
3987 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
3988 /* We've added an opcode byte. */
3989 fragP->fr_fix += 1 + size;
3990 fix_new (fragP, old_fr_fix + 1, size,
3991 fragP->fr_symbol,
3992 fragP->fr_offset, 1,
3993 reloc_type);
3994 break;
3995
3996 default:
3997 BAD_CASE (fragP->fr_subtype);
3998 break;
3999 }
4000 frag_wane (fragP);
4001 return fragP->fr_fix - old_fr_fix;
4002 }
4003 /* Guess a short jump. */
4004 return 1;
4005 }
4006
4007 /* Called after relax() is finished.
4008
4009 In: Address of frag.
4010 fr_type == rs_machine_dependent.
4011 fr_subtype is what the address relaxed to.
4012
4013 Out: Any fixSs and constants are set up.
4014 Caller will turn frag into a ".space 0". */
4015
4016 #ifndef BFD_ASSEMBLER
4017 void
4018 md_convert_frag (headers, sec, fragP)
4019 object_headers *headers ATTRIBUTE_UNUSED;
4020 segT sec ATTRIBUTE_UNUSED;
4021 register fragS *fragP;
4022 #else
4023 void
4024 md_convert_frag (abfd, sec, fragP)
4025 bfd *abfd ATTRIBUTE_UNUSED;
4026 segT sec ATTRIBUTE_UNUSED;
4027 register fragS *fragP;
4028 #endif
4029 {
4030 register unsigned char *opcode;
4031 unsigned char *where_to_put_displacement = NULL;
4032 offsetT target_address;
4033 offsetT opcode_address;
4034 unsigned int extension = 0;
4035 offsetT displacement_from_opcode_start;
4036
4037 opcode = (unsigned char *) fragP->fr_opcode;
4038
4039 /* Address we want to reach in file space. */
4040 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
4041 #ifdef BFD_ASSEMBLER
4042 /* Not needed otherwise? */
4043 target_address += symbol_get_frag (fragP->fr_symbol)->fr_address;
4044 #endif
4045
4046 /* Address opcode resides at in file space. */
4047 opcode_address = fragP->fr_address + fragP->fr_fix;
4048
4049 /* Displacement from opcode start to fill into instruction. */
4050 displacement_from_opcode_start = target_address - opcode_address;
4051
4052 if ((fragP->fr_subtype & BIG) == 0)
4053 {
4054 /* Don't have to change opcode. */
4055 extension = 1; /* 1 opcode + 1 displacement */
4056 where_to_put_displacement = &opcode[1];
4057 }
4058 else
4059 {
4060 if (no_cond_jump_promotion
4061 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
4062 as_warn_where (fragP->fr_file, fragP->fr_line, _("long jump required"));
4063
4064 switch (fragP->fr_subtype)
4065 {
4066 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
4067 extension = 4; /* 1 opcode + 4 displacement */
4068 opcode[0] = 0xe9;
4069 where_to_put_displacement = &opcode[1];
4070 break;
4071
4072 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
4073 extension = 2; /* 1 opcode + 2 displacement */
4074 opcode[0] = 0xe9;
4075 where_to_put_displacement = &opcode[1];
4076 break;
4077
4078 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
4079 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
4080 extension = 5; /* 2 opcode + 4 displacement */
4081 opcode[1] = opcode[0] + 0x10;
4082 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4083 where_to_put_displacement = &opcode[2];
4084 break;
4085
4086 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
4087 extension = 3; /* 2 opcode + 2 displacement */
4088 opcode[1] = opcode[0] + 0x10;
4089 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4090 where_to_put_displacement = &opcode[2];
4091 break;
4092
4093 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
4094 extension = 4;
4095 opcode[0] ^= 1;
4096 opcode[1] = 3;
4097 opcode[2] = 0xe9;
4098 where_to_put_displacement = &opcode[3];
4099 break;
4100
4101 default:
4102 BAD_CASE (fragP->fr_subtype);
4103 break;
4104 }
4105 }
4106
4107 /* Now put displacement after opcode. */
4108 md_number_to_chars ((char *) where_to_put_displacement,
4109 (valueT) (displacement_from_opcode_start - extension),
4110 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
4111 fragP->fr_fix += extension;
4112 }
4113 \f
4114 /* Size of byte displacement jmp. */
4115 int md_short_jump_size = 2;
4116
4117 /* Size of dword displacement jmp. */
4118 int md_long_jump_size = 5;
4119
4120 /* Size of relocation record. */
4121 const int md_reloc_size = 8;
4122
4123 void
4124 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
4125 char *ptr;
4126 addressT from_addr, to_addr;
4127 fragS *frag ATTRIBUTE_UNUSED;
4128 symbolS *to_symbol ATTRIBUTE_UNUSED;
4129 {
4130 offsetT offset;
4131
4132 offset = to_addr - (from_addr + 2);
4133 /* Opcode for byte-disp jump. */
4134 md_number_to_chars (ptr, (valueT) 0xeb, 1);
4135 md_number_to_chars (ptr + 1, (valueT) offset, 1);
4136 }
4137
4138 void
4139 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
4140 char *ptr;
4141 addressT from_addr, to_addr;
4142 fragS *frag ATTRIBUTE_UNUSED;
4143 symbolS *to_symbol ATTRIBUTE_UNUSED;
4144 {
4145 offsetT offset;
4146
4147 offset = to_addr - (from_addr + 5);
4148 md_number_to_chars (ptr, (valueT) 0xe9, 1);
4149 md_number_to_chars (ptr + 1, (valueT) offset, 4);
4150 }
4151 \f
4152 /* Apply a fixup (fixS) to segment data, once it has been determined
4153 by our caller that we have all the info we need to fix it up.
4154
4155 On the 386, immediates, displacements, and data pointers are all in
4156 the same (little-endian) format, so we don't need to care about which
4157 we are handling. */
4158
4159 int
4160 md_apply_fix3 (fixP, valp, seg)
4161 /* The fix we're to put in. */
4162 fixS *fixP;
4163
4164 /* Pointer to the value of the bits. */
4165 valueT *valp;
4166
4167 /* Segment fix is from. */
4168 segT seg ATTRIBUTE_UNUSED;
4169 {
4170 register char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
4171 valueT value = *valp;
4172
4173 #if defined (BFD_ASSEMBLER) && !defined (TE_Mach)
4174 if (fixP->fx_pcrel)
4175 {
4176 switch (fixP->fx_r_type)
4177 {
4178 default:
4179 break;
4180
4181 case BFD_RELOC_32:
4182 fixP->fx_r_type = BFD_RELOC_32_PCREL;
4183 break;
4184 case BFD_RELOC_16:
4185 fixP->fx_r_type = BFD_RELOC_16_PCREL;
4186 break;
4187 case BFD_RELOC_8:
4188 fixP->fx_r_type = BFD_RELOC_8_PCREL;
4189 break;
4190 }
4191 }
4192
4193 /* This is a hack. There should be a better way to handle this.
4194 This covers for the fact that bfd_install_relocation will
4195 subtract the current location (for partial_inplace, PC relative
4196 relocations); see more below. */
4197 if ((fixP->fx_r_type == BFD_RELOC_32_PCREL
4198 || fixP->fx_r_type == BFD_RELOC_16_PCREL
4199 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
4200 && fixP->fx_addsy && !use_rela_relocations)
4201 {
4202 #ifndef OBJ_AOUT
4203 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4204 #ifdef TE_PE
4205 || OUTPUT_FLAVOR == bfd_target_coff_flavour
4206 #endif
4207 )
4208 value += fixP->fx_where + fixP->fx_frag->fr_address;
4209 #endif
4210 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4211 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
4212 {
4213 segT fseg = S_GET_SEGMENT (fixP->fx_addsy);
4214
4215 if ((fseg == seg
4216 || (symbol_section_p (fixP->fx_addsy)
4217 && fseg != absolute_section))
4218 && ! S_IS_EXTERNAL (fixP->fx_addsy)
4219 && ! S_IS_WEAK (fixP->fx_addsy)
4220 && S_IS_DEFINED (fixP->fx_addsy)
4221 && ! S_IS_COMMON (fixP->fx_addsy))
4222 {
4223 /* Yes, we add the values in twice. This is because
4224 bfd_perform_relocation subtracts them out again. I think
4225 bfd_perform_relocation is broken, but I don't dare change
4226 it. FIXME. */
4227 value += fixP->fx_where + fixP->fx_frag->fr_address;
4228 }
4229 }
4230 #endif
4231 #if defined (OBJ_COFF) && defined (TE_PE)
4232 /* For some reason, the PE format does not store a section
4233 address offset for a PC relative symbol. */
4234 if (S_GET_SEGMENT (fixP->fx_addsy) != seg)
4235 value += md_pcrel_from (fixP);
4236 #endif
4237 }
4238
4239 /* Fix a few things - the dynamic linker expects certain values here,
4240 and we must not dissappoint it. */
4241 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4242 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4243 && fixP->fx_addsy)
4244 switch (fixP->fx_r_type)
4245 {
4246 case BFD_RELOC_386_PLT32:
4247 case BFD_RELOC_X86_64_PLT32:
4248 /* Make the jump instruction point to the address of the operand. At
4249 runtime we merely add the offset to the actual PLT entry. */
4250 value = -4;
4251 break;
4252 case BFD_RELOC_386_GOTPC:
4253
4254 /* This is tough to explain. We end up with this one if we have
4255 * operands that look like "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal
4256 * here is to obtain the absolute address of the GOT, and it is strongly
4257 * preferable from a performance point of view to avoid using a runtime
4258 * relocation for this. The actual sequence of instructions often look
4259 * something like:
4260 *
4261 * call .L66
4262 * .L66:
4263 * popl %ebx
4264 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
4265 *
4266 * The call and pop essentially return the absolute address of
4267 * the label .L66 and store it in %ebx. The linker itself will
4268 * ultimately change the first operand of the addl so that %ebx points to
4269 * the GOT, but to keep things simple, the .o file must have this operand
4270 * set so that it generates not the absolute address of .L66, but the
4271 * absolute address of itself. This allows the linker itself simply
4272 * treat a GOTPC relocation as asking for a pcrel offset to the GOT to be
4273 * added in, and the addend of the relocation is stored in the operand
4274 * field for the instruction itself.
4275 *
4276 * Our job here is to fix the operand so that it would add the correct
4277 * offset so that %ebx would point to itself. The thing that is tricky is
4278 * that .-.L66 will point to the beginning of the instruction, so we need
4279 * to further modify the operand so that it will point to itself.
4280 * There are other cases where you have something like:
4281 *
4282 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
4283 *
4284 * and here no correction would be required. Internally in the assembler
4285 * we treat operands of this form as not being pcrel since the '.' is
4286 * explicitly mentioned, and I wonder whether it would simplify matters
4287 * to do it this way. Who knows. In earlier versions of the PIC patches,
4288 * the pcrel_adjust field was used to store the correction, but since the
4289 * expression is not pcrel, I felt it would be confusing to do it this
4290 * way. */
4291
4292 value -= 1;
4293 break;
4294 case BFD_RELOC_386_GOT32:
4295 case BFD_RELOC_X86_64_GOT32:
4296 value = 0; /* Fully resolved at runtime. No addend. */
4297 break;
4298 case BFD_RELOC_386_GOTOFF:
4299 case BFD_RELOC_X86_64_GOTPCREL:
4300 break;
4301
4302 case BFD_RELOC_VTABLE_INHERIT:
4303 case BFD_RELOC_VTABLE_ENTRY:
4304 fixP->fx_done = 0;
4305 return 1;
4306
4307 default:
4308 break;
4309 }
4310 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
4311 *valp = value;
4312 #endif /* defined (BFD_ASSEMBLER) && !defined (TE_Mach) */
4313
4314 #ifndef BFD_ASSEMBLER
4315 md_number_to_chars (p, value, fixP->fx_size);
4316 #else
4317 /* Are we finished with this relocation now? */
4318 if (fixP->fx_addsy == 0 && fixP->fx_pcrel == 0)
4319 fixP->fx_done = 1;
4320 else if (use_rela_relocations)
4321 {
4322 fixP->fx_no_overflow = 1;
4323 value = 0;
4324 }
4325 md_number_to_chars (p, value, fixP->fx_size);
4326 #endif
4327
4328 return 1;
4329 }
4330 \f
4331 #define MAX_LITTLENUMS 6
4332
4333 /* Turn the string pointed to by litP into a floating point constant
4334 of type TYPE, and emit the appropriate bytes. The number of
4335 LITTLENUMS emitted is stored in *SIZEP. An error message is
4336 returned, or NULL on OK. */
4337
4338 char *
4339 md_atof (type, litP, sizeP)
4340 int type;
4341 char *litP;
4342 int *sizeP;
4343 {
4344 int prec;
4345 LITTLENUM_TYPE words[MAX_LITTLENUMS];
4346 LITTLENUM_TYPE *wordP;
4347 char *t;
4348
4349 switch (type)
4350 {
4351 case 'f':
4352 case 'F':
4353 prec = 2;
4354 break;
4355
4356 case 'd':
4357 case 'D':
4358 prec = 4;
4359 break;
4360
4361 case 'x':
4362 case 'X':
4363 prec = 5;
4364 break;
4365
4366 default:
4367 *sizeP = 0;
4368 return _("Bad call to md_atof ()");
4369 }
4370 t = atof_ieee (input_line_pointer, type, words);
4371 if (t)
4372 input_line_pointer = t;
4373
4374 *sizeP = prec * sizeof (LITTLENUM_TYPE);
4375 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
4376 the bigendian 386. */
4377 for (wordP = words + prec - 1; prec--;)
4378 {
4379 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
4380 litP += sizeof (LITTLENUM_TYPE);
4381 }
4382 return 0;
4383 }
4384 \f
4385 char output_invalid_buf[8];
4386
4387 static char *
4388 output_invalid (c)
4389 int c;
4390 {
4391 if (isprint (c))
4392 sprintf (output_invalid_buf, "'%c'", c);
4393 else
4394 sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
4395 return output_invalid_buf;
4396 }
4397
4398 /* REG_STRING starts *before* REGISTER_PREFIX. */
4399
4400 static const reg_entry *
4401 parse_register (reg_string, end_op)
4402 char *reg_string;
4403 char **end_op;
4404 {
4405 char *s = reg_string;
4406 char *p;
4407 char reg_name_given[MAX_REG_NAME_SIZE + 1];
4408 const reg_entry *r;
4409
4410 /* Skip possible REGISTER_PREFIX and possible whitespace. */
4411 if (*s == REGISTER_PREFIX)
4412 ++s;
4413
4414 if (is_space_char (*s))
4415 ++s;
4416
4417 p = reg_name_given;
4418 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
4419 {
4420 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
4421 return (const reg_entry *) NULL;
4422 s++;
4423 }
4424
4425 /* For naked regs, make sure that we are not dealing with an identifier.
4426 This prevents confusing an identifier like `eax_var' with register
4427 `eax'. */
4428 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
4429 return (const reg_entry *) NULL;
4430
4431 *end_op = s;
4432
4433 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
4434
4435 /* Handle floating point regs, allowing spaces in the (i) part. */
4436 if (r == i386_regtab /* %st is first entry of table */)
4437 {
4438 if (is_space_char (*s))
4439 ++s;
4440 if (*s == '(')
4441 {
4442 ++s;
4443 if (is_space_char (*s))
4444 ++s;
4445 if (*s >= '0' && *s <= '7')
4446 {
4447 r = &i386_float_regtab[*s - '0'];
4448 ++s;
4449 if (is_space_char (*s))
4450 ++s;
4451 if (*s == ')')
4452 {
4453 *end_op = s + 1;
4454 return r;
4455 }
4456 }
4457 /* We have "%st(" then garbage. */
4458 return (const reg_entry *) NULL;
4459 }
4460 }
4461
4462 return r;
4463 }
4464 \f
4465 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4466 const char *md_shortopts = "kVQ:sq";
4467 #else
4468 const char *md_shortopts = "q";
4469 #endif
4470
4471 struct option md_longopts[] = {
4472 #define OPTION_32 (OPTION_MD_BASE + 0)
4473 {"32", no_argument, NULL, OPTION_32},
4474 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4475 #define OPTION_64 (OPTION_MD_BASE + 1)
4476 {"64", no_argument, NULL, OPTION_64},
4477 #endif
4478 {NULL, no_argument, NULL, 0}
4479 };
4480 size_t md_longopts_size = sizeof (md_longopts);
4481
4482 int
4483 md_parse_option (c, arg)
4484 int c;
4485 char *arg ATTRIBUTE_UNUSED;
4486 {
4487 switch (c)
4488 {
4489 case 'q':
4490 quiet_warnings = 1;
4491 break;
4492
4493 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4494 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
4495 should be emitted or not. FIXME: Not implemented. */
4496 case 'Q':
4497 break;
4498
4499 /* -V: SVR4 argument to print version ID. */
4500 case 'V':
4501 print_version_id ();
4502 break;
4503
4504 /* -k: Ignore for FreeBSD compatibility. */
4505 case 'k':
4506 break;
4507
4508 case 's':
4509 /* -s: On i386 Solaris, this tells the native assembler to use
4510 .stab instead of .stab.excl. We always use .stab anyhow. */
4511 break;
4512
4513 case OPTION_64:
4514 {
4515 const char **list, **l;
4516
4517 list = bfd_target_list ();
4518 for (l = list; *l != NULL; l++)
4519 if (strcmp (*l, "elf64-x86-64") == 0)
4520 {
4521 default_arch = "x86_64";
4522 break;
4523 }
4524 if (*l == NULL)
4525 as_fatal (_("No compiled in support for x86_64"));
4526 free (list);
4527 }
4528 break;
4529 #endif
4530
4531 case OPTION_32:
4532 default_arch = "i386";
4533 break;
4534
4535 default:
4536 return 0;
4537 }
4538 return 1;
4539 }
4540
4541 void
4542 md_show_usage (stream)
4543 FILE *stream;
4544 {
4545 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4546 fprintf (stream, _("\
4547 -Q ignored\n\
4548 -V print assembler version number\n\
4549 -k ignored\n\
4550 -q quieten some warnings\n\
4551 -s ignored\n"));
4552 #else
4553 fprintf (stream, _("\
4554 -q quieten some warnings\n"));
4555 #endif
4556 }
4557
4558 #ifdef BFD_ASSEMBLER
4559 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
4560 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
4561
4562 /* Pick the target format to use. */
4563
4564 const char *
4565 i386_target_format ()
4566 {
4567 if (!strcmp (default_arch, "x86_64"))
4568 set_code_flag (CODE_64BIT);
4569 else if (!strcmp (default_arch, "i386"))
4570 set_code_flag (CODE_32BIT);
4571 else
4572 as_fatal (_("Unknown architecture"));
4573 switch (OUTPUT_FLAVOR)
4574 {
4575 #ifdef OBJ_MAYBE_AOUT
4576 case bfd_target_aout_flavour:
4577 return AOUT_TARGET_FORMAT;
4578 #endif
4579 #ifdef OBJ_MAYBE_COFF
4580 case bfd_target_coff_flavour:
4581 return "coff-i386";
4582 #endif
4583 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
4584 case bfd_target_elf_flavour:
4585 {
4586 if (flag_code == CODE_64BIT)
4587 use_rela_relocations = 1;
4588 return flag_code == CODE_64BIT ? "elf64-x86-64" : "elf32-i386";
4589 }
4590 #endif
4591 default:
4592 abort ();
4593 return NULL;
4594 }
4595 }
4596
4597 #endif /* OBJ_MAYBE_ more than one */
4598 #endif /* BFD_ASSEMBLER */
4599 \f
4600 symbolS *
4601 md_undefined_symbol (name)
4602 char *name;
4603 {
4604 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
4605 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
4606 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
4607 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
4608 {
4609 if (!GOT_symbol)
4610 {
4611 if (symbol_find (name))
4612 as_bad (_("GOT already in symbol table"));
4613 GOT_symbol = symbol_new (name, undefined_section,
4614 (valueT) 0, &zero_address_frag);
4615 };
4616 return GOT_symbol;
4617 }
4618 return 0;
4619 }
4620
4621 /* Round up a section size to the appropriate boundary. */
4622
4623 valueT
4624 md_section_align (segment, size)
4625 segT segment ATTRIBUTE_UNUSED;
4626 valueT size;
4627 {
4628 #ifdef BFD_ASSEMBLER
4629 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4630 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
4631 {
4632 /* For a.out, force the section size to be aligned. If we don't do
4633 this, BFD will align it for us, but it will not write out the
4634 final bytes of the section. This may be a bug in BFD, but it is
4635 easier to fix it here since that is how the other a.out targets
4636 work. */
4637 int align;
4638
4639 align = bfd_get_section_alignment (stdoutput, segment);
4640 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
4641 }
4642 #endif
4643 #endif
4644
4645 return size;
4646 }
4647
4648 /* On the i386, PC-relative offsets are relative to the start of the
4649 next instruction. That is, the address of the offset, plus its
4650 size, since the offset is always the last part of the insn. */
4651
4652 long
4653 md_pcrel_from (fixP)
4654 fixS *fixP;
4655 {
4656 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
4657 }
4658
4659 #ifndef I386COFF
4660
4661 static void
4662 s_bss (ignore)
4663 int ignore ATTRIBUTE_UNUSED;
4664 {
4665 register int temp;
4666
4667 temp = get_absolute_expression ();
4668 subseg_set (bss_section, (subsegT) temp);
4669 demand_empty_rest_of_line ();
4670 }
4671
4672 #endif
4673
4674 #ifdef BFD_ASSEMBLER
4675
4676 void
4677 i386_validate_fix (fixp)
4678 fixS *fixp;
4679 {
4680 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
4681 {
4682 /* GOTOFF relocation are nonsense in 64bit mode. */
4683 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
4684 {
4685 if (flag_code != CODE_64BIT)
4686 abort ();
4687 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
4688 }
4689 else
4690 {
4691 if (flag_code == CODE_64BIT)
4692 abort ();
4693 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
4694 }
4695 fixp->fx_subsy = 0;
4696 }
4697 }
4698
4699 arelent *
4700 tc_gen_reloc (section, fixp)
4701 asection *section ATTRIBUTE_UNUSED;
4702 fixS *fixp;
4703 {
4704 arelent *rel;
4705 bfd_reloc_code_real_type code;
4706
4707 switch (fixp->fx_r_type)
4708 {
4709 case BFD_RELOC_X86_64_PLT32:
4710 case BFD_RELOC_X86_64_GOT32:
4711 case BFD_RELOC_X86_64_GOTPCREL:
4712 case BFD_RELOC_386_PLT32:
4713 case BFD_RELOC_386_GOT32:
4714 case BFD_RELOC_386_GOTOFF:
4715 case BFD_RELOC_386_GOTPC:
4716 case BFD_RELOC_X86_64_32S:
4717 case BFD_RELOC_RVA:
4718 case BFD_RELOC_VTABLE_ENTRY:
4719 case BFD_RELOC_VTABLE_INHERIT:
4720 code = fixp->fx_r_type;
4721 break;
4722 default:
4723 if (fixp->fx_pcrel)
4724 {
4725 switch (fixp->fx_size)
4726 {
4727 default:
4728 as_bad (_("can not do %d byte pc-relative relocation"),
4729 fixp->fx_size);
4730 code = BFD_RELOC_32_PCREL;
4731 break;
4732 case 1: code = BFD_RELOC_8_PCREL; break;
4733 case 2: code = BFD_RELOC_16_PCREL; break;
4734 case 4: code = BFD_RELOC_32_PCREL; break;
4735 }
4736 }
4737 else
4738 {
4739 switch (fixp->fx_size)
4740 {
4741 default:
4742 as_bad (_("can not do %d byte relocation"), fixp->fx_size);
4743 code = BFD_RELOC_32;
4744 break;
4745 case 1: code = BFD_RELOC_8; break;
4746 case 2: code = BFD_RELOC_16; break;
4747 case 4: code = BFD_RELOC_32; break;
4748 case 8: code = BFD_RELOC_64; break;
4749 }
4750 }
4751 break;
4752 }
4753
4754 if (code == BFD_RELOC_32
4755 && GOT_symbol
4756 && fixp->fx_addsy == GOT_symbol)
4757 {
4758 /* We don't support GOTPC on 64bit targets. */
4759 if (flag_code == CODE_64BIT)
4760 abort ();
4761 code = BFD_RELOC_386_GOTPC;
4762 }
4763
4764 rel = (arelent *) xmalloc (sizeof (arelent));
4765 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
4766 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
4767
4768 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
4769 if (!use_rela_relocations)
4770 {
4771 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
4772 vtable entry to be used in the relocation's section offset. */
4773 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
4774 rel->address = fixp->fx_offset;
4775
4776 if (fixp->fx_pcrel)
4777 rel->addend = fixp->fx_addnumber;
4778 else
4779 rel->addend = 0;
4780 }
4781 /* Use the rela in 64bit mode. */
4782 else
4783 {
4784 rel->addend = fixp->fx_offset;
4785 if (fixp->fx_pcrel)
4786 rel->addend -= fixp->fx_size;
4787 }
4788
4789 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
4790 if (rel->howto == NULL)
4791 {
4792 as_bad_where (fixp->fx_file, fixp->fx_line,
4793 _("cannot represent relocation type %s"),
4794 bfd_get_reloc_code_name (code));
4795 /* Set howto to a garbage value so that we can keep going. */
4796 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
4797 assert (rel->howto != NULL);
4798 }
4799
4800 return rel;
4801 }
4802
4803 #else /* ! BFD_ASSEMBLER */
4804
4805 #if (defined(OBJ_AOUT) | defined(OBJ_BOUT))
4806 void
4807 tc_aout_fix_to_chars (where, fixP, segment_address_in_file)
4808 char *where;
4809 fixS *fixP;
4810 relax_addressT segment_address_in_file;
4811 {
4812 /* In: length of relocation (or of address) in chars: 1, 2 or 4.
4813 Out: GNU LD relocation length code: 0, 1, or 2. */
4814
4815 static const unsigned char nbytes_r_length[] = { 42, 0, 1, 42, 2 };
4816 long r_symbolnum;
4817
4818 know (fixP->fx_addsy != NULL);
4819
4820 md_number_to_chars (where,
4821 (valueT) (fixP->fx_frag->fr_address
4822 + fixP->fx_where - segment_address_in_file),
4823 4);
4824
4825 r_symbolnum = (S_IS_DEFINED (fixP->fx_addsy)
4826 ? S_GET_TYPE (fixP->fx_addsy)
4827 : fixP->fx_addsy->sy_number);
4828
4829 where[6] = (r_symbolnum >> 16) & 0x0ff;
4830 where[5] = (r_symbolnum >> 8) & 0x0ff;
4831 where[4] = r_symbolnum & 0x0ff;
4832 where[7] = ((((!S_IS_DEFINED (fixP->fx_addsy)) << 3) & 0x08)
4833 | ((nbytes_r_length[fixP->fx_size] << 1) & 0x06)
4834 | (((fixP->fx_pcrel << 0) & 0x01) & 0x0f));
4835 }
4836
4837 #endif /* OBJ_AOUT or OBJ_BOUT. */
4838
4839 #if defined (I386COFF)
4840
4841 short
4842 tc_coff_fix2rtype (fixP)
4843 fixS *fixP;
4844 {
4845 if (fixP->fx_r_type == R_IMAGEBASE)
4846 return R_IMAGEBASE;
4847
4848 return (fixP->fx_pcrel ?
4849 (fixP->fx_size == 1 ? R_PCRBYTE :
4850 fixP->fx_size == 2 ? R_PCRWORD :
4851 R_PCRLONG) :
4852 (fixP->fx_size == 1 ? R_RELBYTE :
4853 fixP->fx_size == 2 ? R_RELWORD :
4854 R_DIR32));
4855 }
4856
4857 int
4858 tc_coff_sizemachdep (frag)
4859 fragS *frag;
4860 {
4861 if (frag->fr_next)
4862 return (frag->fr_next->fr_address - frag->fr_address);
4863 else
4864 return 0;
4865 }
4866
4867 #endif /* I386COFF */
4868
4869 #endif /* ! BFD_ASSEMBLER */
4870 \f
4871 /* Parse operands using Intel syntax. This implements a recursive descent
4872 parser based on the BNF grammar published in Appendix B of the MASM 6.1
4873 Programmer's Guide.
4874
4875 FIXME: We do not recognize the full operand grammar defined in the MASM
4876 documentation. In particular, all the structure/union and
4877 high-level macro operands are missing.
4878
4879 Uppercase words are terminals, lower case words are non-terminals.
4880 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
4881 bars '|' denote choices. Most grammar productions are implemented in
4882 functions called 'intel_<production>'.
4883
4884 Initial production is 'expr'.
4885
4886 addOp + | -
4887
4888 alpha [a-zA-Z]
4889
4890 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
4891
4892 constant digits [[ radixOverride ]]
4893
4894 dataType BYTE | WORD | DWORD | QWORD | XWORD
4895
4896 digits decdigit
4897 | digits decdigit
4898 | digits hexdigit
4899
4900 decdigit [0-9]
4901
4902 e05 e05 addOp e06
4903 | e06
4904
4905 e06 e06 mulOp e09
4906 | e09
4907
4908 e09 OFFSET e10
4909 | e09 PTR e10
4910 | e09 : e10
4911 | e10
4912
4913 e10 e10 [ expr ]
4914 | e11
4915
4916 e11 ( expr )
4917 | [ expr ]
4918 | constant
4919 | dataType
4920 | id
4921 | $
4922 | register
4923
4924 => expr SHORT e05
4925 | e05
4926
4927 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
4928 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
4929
4930 hexdigit a | b | c | d | e | f
4931 | A | B | C | D | E | F
4932
4933 id alpha
4934 | id alpha
4935 | id decdigit
4936
4937 mulOp * | / | MOD
4938
4939 quote " | '
4940
4941 register specialRegister
4942 | gpRegister
4943 | byteRegister
4944
4945 segmentRegister CS | DS | ES | FS | GS | SS
4946
4947 specialRegister CR0 | CR2 | CR3
4948 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
4949 | TR3 | TR4 | TR5 | TR6 | TR7
4950
4951 We simplify the grammar in obvious places (e.g., register parsing is
4952 done by calling parse_register) and eliminate immediate left recursion
4953 to implement a recursive-descent parser.
4954
4955 expr SHORT e05
4956 | e05
4957
4958 e05 e06 e05'
4959
4960 e05' addOp e06 e05'
4961 | Empty
4962
4963 e06 e09 e06'
4964
4965 e06' mulOp e09 e06'
4966 | Empty
4967
4968 e09 OFFSET e10 e09'
4969 | e10 e09'
4970
4971 e09' PTR e10 e09'
4972 | : e10 e09'
4973 | Empty
4974
4975 e10 e11 e10'
4976
4977 e10' [ expr ] e10'
4978 | Empty
4979
4980 e11 ( expr )
4981 | [ expr ]
4982 | BYTE
4983 | WORD
4984 | DWORD
4985 | QWORD
4986 | XWORD
4987 | .
4988 | $
4989 | register
4990 | id
4991 | constant */
4992
4993 /* Parsing structure for the intel syntax parser. Used to implement the
4994 semantic actions for the operand grammar. */
4995 struct intel_parser_s
4996 {
4997 char *op_string; /* The string being parsed. */
4998 int got_a_float; /* Whether the operand is a float. */
4999 int op_modifier; /* Operand modifier. */
5000 int is_mem; /* 1 if operand is memory reference. */
5001 const reg_entry *reg; /* Last register reference found. */
5002 char *disp; /* Displacement string being built. */
5003 };
5004
5005 static struct intel_parser_s intel_parser;
5006
5007 /* Token structure for parsing intel syntax. */
5008 struct intel_token
5009 {
5010 int code; /* Token code. */
5011 const reg_entry *reg; /* Register entry for register tokens. */
5012 char *str; /* String representation. */
5013 };
5014
5015 static struct intel_token cur_token, prev_token;
5016
5017 /* Token codes for the intel parser. Since T_SHORT is already used
5018 by COFF, undefine it first to prevent a warning. */
5019 #define T_NIL -1
5020 #define T_CONST 1
5021 #define T_REG 2
5022 #define T_BYTE 3
5023 #define T_WORD 4
5024 #define T_DWORD 5
5025 #define T_QWORD 6
5026 #define T_XWORD 7
5027 #undef T_SHORT
5028 #define T_SHORT 8
5029 #define T_OFFSET 9
5030 #define T_PTR 10
5031 #define T_ID 11
5032
5033 /* Prototypes for intel parser functions. */
5034 static int intel_match_token PARAMS ((int code));
5035 static void intel_get_token PARAMS ((void));
5036 static void intel_putback_token PARAMS ((void));
5037 static int intel_expr PARAMS ((void));
5038 static int intel_e05 PARAMS ((void));
5039 static int intel_e05_1 PARAMS ((void));
5040 static int intel_e06 PARAMS ((void));
5041 static int intel_e06_1 PARAMS ((void));
5042 static int intel_e09 PARAMS ((void));
5043 static int intel_e09_1 PARAMS ((void));
5044 static int intel_e10 PARAMS ((void));
5045 static int intel_e10_1 PARAMS ((void));
5046 static int intel_e11 PARAMS ((void));
5047
5048 static int
5049 i386_intel_operand (operand_string, got_a_float)
5050 char *operand_string;
5051 int got_a_float;
5052 {
5053 int ret;
5054 char *p;
5055
5056 /* Initialize token holders. */
5057 cur_token.code = prev_token.code = T_NIL;
5058 cur_token.reg = prev_token.reg = NULL;
5059 cur_token.str = prev_token.str = NULL;
5060
5061 /* Initialize parser structure. */
5062 p = intel_parser.op_string = (char *) malloc (strlen (operand_string) + 1);
5063 if (p == NULL)
5064 abort ();
5065 strcpy (intel_parser.op_string, operand_string);
5066 intel_parser.got_a_float = got_a_float;
5067 intel_parser.op_modifier = -1;
5068 intel_parser.is_mem = 0;
5069 intel_parser.reg = NULL;
5070 intel_parser.disp = (char *) malloc (strlen (operand_string) + 1);
5071 if (intel_parser.disp == NULL)
5072 abort ();
5073 intel_parser.disp[0] = '\0';
5074
5075 /* Read the first token and start the parser. */
5076 intel_get_token ();
5077 ret = intel_expr ();
5078
5079 if (ret)
5080 {
5081 /* If we found a memory reference, hand it over to i386_displacement
5082 to fill in the rest of the operand fields. */
5083 if (intel_parser.is_mem)
5084 {
5085 if ((i.mem_operands == 1
5086 && (current_templates->start->opcode_modifier & IsString) == 0)
5087 || i.mem_operands == 2)
5088 {
5089 as_bad (_("too many memory references for '%s'"),
5090 current_templates->start->name);
5091 ret = 0;
5092 }
5093 else
5094 {
5095 char *s = intel_parser.disp;
5096 i.mem_operands++;
5097
5098 /* Add the displacement expression. */
5099 if (*s != '\0')
5100 ret = i386_displacement (s, s + strlen (s))
5101 && i386_index_check (s);
5102 }
5103 }
5104
5105 /* Constant and OFFSET expressions are handled by i386_immediate. */
5106 else if (intel_parser.op_modifier == OFFSET_FLAT
5107 || intel_parser.reg == NULL)
5108 ret = i386_immediate (intel_parser.disp);
5109 }
5110
5111 free (p);
5112 free (intel_parser.disp);
5113
5114 return ret;
5115 }
5116
5117 /* expr SHORT e05
5118 | e05 */
5119 static int
5120 intel_expr ()
5121 {
5122 /* expr SHORT e05 */
5123 if (cur_token.code == T_SHORT)
5124 {
5125 intel_parser.op_modifier = SHORT;
5126 intel_match_token (T_SHORT);
5127
5128 return (intel_e05 ());
5129 }
5130
5131 /* expr e05 */
5132 else
5133 return intel_e05 ();
5134 }
5135
5136 /* e05 e06 e05'
5137
5138 e05' addOp e06 e05'
5139 | Empty */
5140 static int
5141 intel_e05 ()
5142 {
5143 return (intel_e06 () && intel_e05_1 ());
5144 }
5145
5146 static int
5147 intel_e05_1 ()
5148 {
5149 /* e05' addOp e06 e05' */
5150 if (cur_token.code == '+' || cur_token.code == '-')
5151 {
5152 strcat (intel_parser.disp, cur_token.str);
5153 intel_match_token (cur_token.code);
5154
5155 return (intel_e06 () && intel_e05_1 ());
5156 }
5157
5158 /* e05' Empty */
5159 else
5160 return 1;
5161 }
5162
5163 /* e06 e09 e06'
5164
5165 e06' mulOp e09 e06'
5166 | Empty */
5167 static int
5168 intel_e06 ()
5169 {
5170 return (intel_e09 () && intel_e06_1 ());
5171 }
5172
5173 static int
5174 intel_e06_1 ()
5175 {
5176 /* e06' mulOp e09 e06' */
5177 if (cur_token.code == '*' || cur_token.code == '/')
5178 {
5179 strcat (intel_parser.disp, cur_token.str);
5180 intel_match_token (cur_token.code);
5181
5182 return (intel_e09 () && intel_e06_1 ());
5183 }
5184
5185 /* e06' Empty */
5186 else
5187 return 1;
5188 }
5189
5190 /* e09 OFFSET e10 e09'
5191 | e10 e09'
5192
5193 e09' PTR e10 e09'
5194 | : e10 e09'
5195 | Empty */
5196 static int
5197 intel_e09 ()
5198 {
5199 /* e09 OFFSET e10 e09' */
5200 if (cur_token.code == T_OFFSET)
5201 {
5202 intel_parser.is_mem = 0;
5203 intel_parser.op_modifier = OFFSET_FLAT;
5204 intel_match_token (T_OFFSET);
5205
5206 return (intel_e10 () && intel_e09_1 ());
5207 }
5208
5209 /* e09 e10 e09' */
5210 else
5211 return (intel_e10 () && intel_e09_1 ());
5212 }
5213
5214 static int
5215 intel_e09_1 ()
5216 {
5217 /* e09' PTR e10 e09' */
5218 if (cur_token.code == T_PTR)
5219 {
5220 if (prev_token.code == T_BYTE)
5221 i.suffix = BYTE_MNEM_SUFFIX;
5222
5223 else if (prev_token.code == T_WORD)
5224 {
5225 if (intel_parser.got_a_float == 2) /* "fi..." */
5226 i.suffix = SHORT_MNEM_SUFFIX;
5227 else
5228 i.suffix = WORD_MNEM_SUFFIX;
5229 }
5230
5231 else if (prev_token.code == T_DWORD)
5232 {
5233 if (intel_parser.got_a_float == 1) /* "f..." */
5234 i.suffix = SHORT_MNEM_SUFFIX;
5235 else
5236 i.suffix = LONG_MNEM_SUFFIX;
5237 }
5238
5239 else if (prev_token.code == T_QWORD)
5240 {
5241 if (intel_parser.got_a_float == 1) /* "f..." */
5242 i.suffix = LONG_MNEM_SUFFIX;
5243 else
5244 i.suffix = QWORD_MNEM_SUFFIX;
5245 }
5246
5247 else if (prev_token.code == T_XWORD)
5248 i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
5249
5250 else
5251 {
5252 as_bad (_("Unknown operand modifier `%s'\n"), prev_token.str);
5253 return 0;
5254 }
5255
5256 intel_match_token (T_PTR);
5257
5258 return (intel_e10 () && intel_e09_1 ());
5259 }
5260
5261 /* e09 : e10 e09' */
5262 else if (cur_token.code == ':')
5263 {
5264 /* Mark as a memory operand only if it's not already known to be an
5265 offset expression. */
5266 if (intel_parser.op_modifier != OFFSET_FLAT)
5267 intel_parser.is_mem = 1;
5268
5269 return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
5270 }
5271
5272 /* e09' Empty */
5273 else
5274 return 1;
5275 }
5276
5277 /* e10 e11 e10'
5278
5279 e10' [ expr ] e10'
5280 | Empty */
5281 static int
5282 intel_e10 ()
5283 {
5284 return (intel_e11 () && intel_e10_1 ());
5285 }
5286
5287 static int
5288 intel_e10_1 ()
5289 {
5290 /* e10' [ expr ] e10' */
5291 if (cur_token.code == '[')
5292 {
5293 intel_match_token ('[');
5294
5295 /* Mark as a memory operand only if it's not already known to be an
5296 offset expression. If it's an offset expression, we need to keep
5297 the brace in. */
5298 if (intel_parser.op_modifier != OFFSET_FLAT)
5299 intel_parser.is_mem = 1;
5300 else
5301 strcat (intel_parser.disp, "[");
5302
5303 /* Add a '+' to the displacement string if necessary. */
5304 if (*intel_parser.disp != '\0'
5305 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5306 strcat (intel_parser.disp, "+");
5307
5308 if (intel_expr () && intel_match_token (']'))
5309 {
5310 /* Preserve brackets when the operand is an offset expression. */
5311 if (intel_parser.op_modifier == OFFSET_FLAT)
5312 strcat (intel_parser.disp, "]");
5313
5314 return intel_e10_1 ();
5315 }
5316 else
5317 return 0;
5318 }
5319
5320 /* e10' Empty */
5321 else
5322 return 1;
5323 }
5324
5325 /* e11 ( expr )
5326 | [ expr ]
5327 | BYTE
5328 | WORD
5329 | DWORD
5330 | QWORD
5331 | XWORD
5332 | $
5333 | .
5334 | register
5335 | id
5336 | constant */
5337 static int
5338 intel_e11 ()
5339 {
5340 /* e11 ( expr ) */
5341 if (cur_token.code == '(')
5342 {
5343 intel_match_token ('(');
5344 strcat (intel_parser.disp, "(");
5345
5346 if (intel_expr () && intel_match_token (')'))
5347 {
5348 strcat (intel_parser.disp, ")");
5349 return 1;
5350 }
5351 else
5352 return 0;
5353 }
5354
5355 /* e11 [ expr ] */
5356 else if (cur_token.code == '[')
5357 {
5358 intel_match_token ('[');
5359
5360 /* Mark as a memory operand only if it's not already known to be an
5361 offset expression. If it's an offset expression, we need to keep
5362 the brace in. */
5363 if (intel_parser.op_modifier != OFFSET_FLAT)
5364 intel_parser.is_mem = 1;
5365 else
5366 strcat (intel_parser.disp, "[");
5367
5368 /* Operands for jump/call inside brackets denote absolute addresses. */
5369 if (current_templates->start->opcode_modifier & Jump
5370 || current_templates->start->opcode_modifier & JumpDword
5371 || current_templates->start->opcode_modifier & JumpByte
5372 || current_templates->start->opcode_modifier & JumpInterSegment)
5373 i.types[this_operand] |= JumpAbsolute;
5374
5375 /* Add a '+' to the displacement string if necessary. */
5376 if (*intel_parser.disp != '\0'
5377 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5378 strcat (intel_parser.disp, "+");
5379
5380 if (intel_expr () && intel_match_token (']'))
5381 {
5382 /* Preserve brackets when the operand is an offset expression. */
5383 if (intel_parser.op_modifier == OFFSET_FLAT)
5384 strcat (intel_parser.disp, "]");
5385
5386 return 1;
5387 }
5388 else
5389 return 0;
5390 }
5391
5392 /* e11 BYTE
5393 | WORD
5394 | DWORD
5395 | QWORD
5396 | XWORD */
5397 else if (cur_token.code == T_BYTE
5398 || cur_token.code == T_WORD
5399 || cur_token.code == T_DWORD
5400 || cur_token.code == T_QWORD
5401 || cur_token.code == T_XWORD)
5402 {
5403 intel_match_token (cur_token.code);
5404
5405 return 1;
5406 }
5407
5408 /* e11 $
5409 | . */
5410 else if (cur_token.code == '$' || cur_token.code == '.')
5411 {
5412 strcat (intel_parser.disp, cur_token.str);
5413 intel_match_token (cur_token.code);
5414
5415 /* Mark as a memory operand only if it's not already known to be an
5416 offset expression. */
5417 if (intel_parser.op_modifier != OFFSET_FLAT)
5418 intel_parser.is_mem = 1;
5419
5420 return 1;
5421 }
5422
5423 /* e11 register */
5424 else if (cur_token.code == T_REG)
5425 {
5426 const reg_entry *reg = intel_parser.reg = cur_token.reg;
5427
5428 intel_match_token (T_REG);
5429
5430 /* Check for segment change. */
5431 if (cur_token.code == ':')
5432 {
5433 if (reg->reg_type & (SReg2 | SReg3))
5434 {
5435 switch (reg->reg_num)
5436 {
5437 case 0:
5438 i.seg[i.mem_operands] = &es;
5439 break;
5440 case 1:
5441 i.seg[i.mem_operands] = &cs;
5442 break;
5443 case 2:
5444 i.seg[i.mem_operands] = &ss;
5445 break;
5446 case 3:
5447 i.seg[i.mem_operands] = &ds;
5448 break;
5449 case 4:
5450 i.seg[i.mem_operands] = &fs;
5451 break;
5452 case 5:
5453 i.seg[i.mem_operands] = &gs;
5454 break;
5455 }
5456 }
5457 else
5458 {
5459 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
5460 return 0;
5461 }
5462 }
5463
5464 /* Not a segment register. Check for register scaling. */
5465 else if (cur_token.code == '*')
5466 {
5467 if (!intel_parser.is_mem)
5468 {
5469 as_bad (_("Register scaling only allowed in memory operands."));
5470 return 0;
5471 }
5472
5473 /* What follows must be a valid scale. */
5474 if (intel_match_token ('*')
5475 && strchr ("01248", *cur_token.str))
5476 {
5477 i.index_reg = reg;
5478 i.types[this_operand] |= BaseIndex;
5479
5480 /* Set the scale after setting the register (otherwise,
5481 i386_scale will complain) */
5482 i386_scale (cur_token.str);
5483 intel_match_token (T_CONST);
5484 }
5485 else
5486 {
5487 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
5488 cur_token.str);
5489 return 0;
5490 }
5491 }
5492
5493 /* No scaling. If this is a memory operand, the register is either a
5494 base register (first occurrence) or an index register (second
5495 occurrence). */
5496 else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3)))
5497 {
5498 if (i.base_reg && i.index_reg)
5499 {
5500 as_bad (_("Too many register references in memory operand.\n"));
5501 return 0;
5502 }
5503
5504 if (i.base_reg == NULL)
5505 i.base_reg = reg;
5506 else
5507 i.index_reg = reg;
5508
5509 i.types[this_operand] |= BaseIndex;
5510 }
5511
5512 /* Offset modifier. Add the register to the displacement string to be
5513 parsed as an immediate expression after we're done. */
5514 else if (intel_parser.op_modifier == OFFSET_FLAT)
5515 strcat (intel_parser.disp, reg->reg_name);
5516
5517 /* It's neither base nor index nor offset. */
5518 else
5519 {
5520 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
5521 i.op[this_operand].regs = reg;
5522 i.reg_operands++;
5523 }
5524
5525 /* Since registers are not part of the displacement string (except
5526 when we're parsing offset operands), we may need to remove any
5527 preceding '+' from the displacement string. */
5528 if (*intel_parser.disp != '\0'
5529 && intel_parser.op_modifier != OFFSET_FLAT)
5530 {
5531 char *s = intel_parser.disp;
5532 s += strlen (s) - 1;
5533 if (*s == '+')
5534 *s = '\0';
5535 }
5536
5537 return 1;
5538 }
5539
5540 /* e11 id */
5541 else if (cur_token.code == T_ID)
5542 {
5543 /* Add the identifier to the displacement string. */
5544 strcat (intel_parser.disp, cur_token.str);
5545 intel_match_token (T_ID);
5546
5547 /* The identifier represents a memory reference only if it's not
5548 preceded by an offset modifier. */
5549 if (intel_parser.op_modifier != OFFSET_FLAT)
5550 intel_parser.is_mem = 1;
5551
5552 return 1;
5553 }
5554
5555 /* e11 constant */
5556 else if (cur_token.code == T_CONST
5557 || cur_token.code == '-'
5558 || cur_token.code == '+')
5559 {
5560 char *save_str;
5561
5562 /* Allow constants that start with `+' or `-'. */
5563 if (cur_token.code == '-' || cur_token.code == '+')
5564 {
5565 strcat (intel_parser.disp, cur_token.str);
5566 intel_match_token (cur_token.code);
5567 if (cur_token.code != T_CONST)
5568 {
5569 as_bad (_("Syntax error. Expecting a constant. Got `%s'.\n"),
5570 cur_token.str);
5571 return 0;
5572 }
5573 }
5574
5575 save_str = (char *) malloc (strlen (cur_token.str) + 1);
5576 if (save_str == NULL)
5577 abort ();
5578 strcpy (save_str, cur_token.str);
5579
5580 /* Get the next token to check for register scaling. */
5581 intel_match_token (cur_token.code);
5582
5583 /* Check if this constant is a scaling factor for an index register. */
5584 if (cur_token.code == '*')
5585 {
5586 if (intel_match_token ('*') && cur_token.code == T_REG)
5587 {
5588 if (!intel_parser.is_mem)
5589 {
5590 as_bad (_("Register scaling only allowed in memory operands."));
5591 return 0;
5592 }
5593
5594 /* The constant is followed by `* reg', so it must be
5595 a valid scale. */
5596 if (strchr ("01248", *save_str))
5597 {
5598 i.index_reg = cur_token.reg;
5599 i.types[this_operand] |= BaseIndex;
5600
5601 /* Set the scale after setting the register (otherwise,
5602 i386_scale will complain) */
5603 i386_scale (save_str);
5604 intel_match_token (T_REG);
5605
5606 /* Since registers are not part of the displacement
5607 string, we may need to remove any preceding '+' from
5608 the displacement string. */
5609 if (*intel_parser.disp != '\0')
5610 {
5611 char *s = intel_parser.disp;
5612 s += strlen (s) - 1;
5613 if (*s == '+')
5614 *s = '\0';
5615 }
5616
5617 free (save_str);
5618
5619 return 1;
5620 }
5621 else
5622 return 0;
5623 }
5624
5625 /* The constant was not used for register scaling. Since we have
5626 already consumed the token following `*' we now need to put it
5627 back in the stream. */
5628 else
5629 intel_putback_token ();
5630 }
5631
5632 /* Add the constant to the displacement string. */
5633 strcat (intel_parser.disp, save_str);
5634 free (save_str);
5635
5636 return 1;
5637 }
5638
5639 as_bad (_("Unrecognized token '%s'"), cur_token.str);
5640 return 0;
5641 }
5642
5643 /* Match the given token against cur_token. If they match, read the next
5644 token from the operand string. */
5645 static int
5646 intel_match_token (code)
5647 int code;
5648 {
5649 if (cur_token.code == code)
5650 {
5651 intel_get_token ();
5652 return 1;
5653 }
5654 else
5655 {
5656 as_bad (_("Unexpected token `%s'\n"), cur_token.str);
5657 return 0;
5658 }
5659 }
5660
5661 /* Read a new token from intel_parser.op_string and store it in cur_token. */
5662 static void
5663 intel_get_token ()
5664 {
5665 char *end_op;
5666 const reg_entry *reg;
5667 struct intel_token new_token;
5668
5669 new_token.code = T_NIL;
5670 new_token.reg = NULL;
5671 new_token.str = NULL;
5672
5673 /* Free the memory allocated to the previous token and move
5674 cur_token to prev_token. */
5675 if (prev_token.str)
5676 free (prev_token.str);
5677
5678 prev_token = cur_token;
5679
5680 /* Skip whitespace. */
5681 while (is_space_char (*intel_parser.op_string))
5682 intel_parser.op_string++;
5683
5684 /* Return an empty token if we find nothing else on the line. */
5685 if (*intel_parser.op_string == '\0')
5686 {
5687 cur_token = new_token;
5688 return;
5689 }
5690
5691 /* The new token cannot be larger than the remainder of the operand
5692 string. */
5693 new_token.str = (char *) malloc (strlen (intel_parser.op_string) + 1);
5694 if (new_token.str == NULL)
5695 abort ();
5696 new_token.str[0] = '\0';
5697
5698 if (strchr ("0123456789", *intel_parser.op_string))
5699 {
5700 char *p = new_token.str;
5701 char *q = intel_parser.op_string;
5702 new_token.code = T_CONST;
5703
5704 /* Allow any kind of identifier char to encompass floating point and
5705 hexadecimal numbers. */
5706 while (is_identifier_char (*q))
5707 *p++ = *q++;
5708 *p = '\0';
5709
5710 /* Recognize special symbol names [0-9][bf]. */
5711 if (strlen (intel_parser.op_string) == 2
5712 && (intel_parser.op_string[1] == 'b'
5713 || intel_parser.op_string[1] == 'f'))
5714 new_token.code = T_ID;
5715 }
5716
5717 else if (strchr ("+-/*:[]()", *intel_parser.op_string))
5718 {
5719 new_token.code = *intel_parser.op_string;
5720 new_token.str[0] = *intel_parser.op_string;
5721 new_token.str[1] = '\0';
5722 }
5723
5724 else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
5725 && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
5726 {
5727 new_token.code = T_REG;
5728 new_token.reg = reg;
5729
5730 if (*intel_parser.op_string == REGISTER_PREFIX)
5731 {
5732 new_token.str[0] = REGISTER_PREFIX;
5733 new_token.str[1] = '\0';
5734 }
5735
5736 strcat (new_token.str, reg->reg_name);
5737 }
5738
5739 else if (is_identifier_char (*intel_parser.op_string))
5740 {
5741 char *p = new_token.str;
5742 char *q = intel_parser.op_string;
5743
5744 /* A '.' or '$' followed by an identifier char is an identifier.
5745 Otherwise, it's operator '.' followed by an expression. */
5746 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
5747 {
5748 new_token.code = *q;
5749 new_token.str[0] = *q;
5750 new_token.str[1] = '\0';
5751 }
5752 else
5753 {
5754 while (is_identifier_char (*q) || *q == '@')
5755 *p++ = *q++;
5756 *p = '\0';
5757
5758 if (strcasecmp (new_token.str, "BYTE") == 0)
5759 new_token.code = T_BYTE;
5760
5761 else if (strcasecmp (new_token.str, "WORD") == 0)
5762 new_token.code = T_WORD;
5763
5764 else if (strcasecmp (new_token.str, "DWORD") == 0)
5765 new_token.code = T_DWORD;
5766
5767 else if (strcasecmp (new_token.str, "QWORD") == 0)
5768 new_token.code = T_QWORD;
5769
5770 else if (strcasecmp (new_token.str, "XWORD") == 0)
5771 new_token.code = T_XWORD;
5772
5773 else if (strcasecmp (new_token.str, "PTR") == 0)
5774 new_token.code = T_PTR;
5775
5776 else if (strcasecmp (new_token.str, "SHORT") == 0)
5777 new_token.code = T_SHORT;
5778
5779 else if (strcasecmp (new_token.str, "OFFSET") == 0)
5780 {
5781 new_token.code = T_OFFSET;
5782
5783 /* ??? This is not mentioned in the MASM grammar but gcc
5784 makes use of it with -mintel-syntax. OFFSET may be
5785 followed by FLAT: */
5786 if (strncasecmp (q, " FLAT:", 6) == 0)
5787 strcat (new_token.str, " FLAT:");
5788 }
5789
5790 /* ??? This is not mentioned in the MASM grammar. */
5791 else if (strcasecmp (new_token.str, "FLAT") == 0)
5792 new_token.code = T_OFFSET;
5793
5794 else
5795 new_token.code = T_ID;
5796 }
5797 }
5798
5799 else
5800 as_bad (_("Unrecognized token `%s'\n"), intel_parser.op_string);
5801
5802 intel_parser.op_string += strlen (new_token.str);
5803 cur_token = new_token;
5804 }
5805
5806 /* Put cur_token back into the token stream and make cur_token point to
5807 prev_token. */
5808 static void
5809 intel_putback_token ()
5810 {
5811 intel_parser.op_string -= strlen (cur_token.str);
5812 free (cur_token.str);
5813 cur_token = prev_token;
5814
5815 /* Forget prev_token. */
5816 prev_token.code = T_NIL;
5817 prev_token.reg = NULL;
5818 prev_token.str = NULL;
5819 }