Fix register name printed in warning message.
[deliverable/binutils-gdb.git] / gas / config / tc-i386.c
1 /* i386.c -- Assemble code for the Intel 80386
2 Copyright 1989, 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
3 2000, 2001
4 Free Software Foundation, Inc.
5
6 This file is part of GAS, the GNU Assembler.
7
8 GAS is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2, or (at your option)
11 any later version.
12
13 GAS is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
17
18 You should have received a copy of the GNU General Public License
19 along with GAS; see the file COPYING. If not, write to the Free
20 Software Foundation, 59 Temple Place - Suite 330, Boston, MA
21 02111-1307, USA. */
22
23 /* Intel 80386 machine specific gas.
24 Written by Eliot Dresselhaus (eliot@mgm.mit.edu).
25 x86_64 support by Jan Hubicka (jh@suse.cz)
26 Bugs & suggestions are completely welcome. This is free software.
27 Please help us make it better. */
28
29 #include <ctype.h>
30
31 #include "as.h"
32 #include "subsegs.h"
33 #include "dwarf2dbg.h"
34 #include "opcode/i386.h"
35
36 #ifndef REGISTER_WARNINGS
37 #define REGISTER_WARNINGS 1
38 #endif
39
40 #ifndef INFER_ADDR_PREFIX
41 #define INFER_ADDR_PREFIX 1
42 #endif
43
44 #ifndef SCALE1_WHEN_NO_INDEX
45 /* Specifying a scale factor besides 1 when there is no index is
46 futile. eg. `mov (%ebx,2),%al' does exactly the same as
47 `mov (%ebx),%al'. To slavishly follow what the programmer
48 specified, set SCALE1_WHEN_NO_INDEX to 0. */
49 #define SCALE1_WHEN_NO_INDEX 1
50 #endif
51
52 #define true 1
53 #define false 0
54
55 static unsigned int mode_from_disp_size PARAMS ((unsigned int));
56 static int fits_in_signed_byte PARAMS ((offsetT));
57 static int fits_in_unsigned_byte PARAMS ((offsetT));
58 static int fits_in_unsigned_word PARAMS ((offsetT));
59 static int fits_in_signed_word PARAMS ((offsetT));
60 static int fits_in_unsigned_long PARAMS ((offsetT));
61 static int fits_in_signed_long PARAMS ((offsetT));
62 static int smallest_imm_type PARAMS ((offsetT));
63 static offsetT offset_in_range PARAMS ((offsetT, int));
64 static int add_prefix PARAMS ((unsigned int));
65 static void set_code_flag PARAMS ((int));
66 static void set_16bit_gcc_code_flag PARAMS ((int));
67 static void set_intel_syntax PARAMS ((int));
68 static void set_cpu_arch PARAMS ((int));
69
70 #ifdef BFD_ASSEMBLER
71 static bfd_reloc_code_real_type reloc
72 PARAMS ((int, int, int, bfd_reloc_code_real_type));
73 #define RELOC_ENUM enum bfd_reloc_code_real
74 #else
75 #define RELOC_ENUM int
76 #endif
77
78 #ifndef DEFAULT_ARCH
79 #define DEFAULT_ARCH "i386"
80 #endif
81 static char *default_arch = DEFAULT_ARCH;
82
83 /* 'md_assemble ()' gathers together information and puts it into a
84 i386_insn. */
85
86 union i386_op
87 {
88 expressionS *disps;
89 expressionS *imms;
90 const reg_entry *regs;
91 };
92
93 struct _i386_insn
94 {
95 /* TM holds the template for the insn were currently assembling. */
96 template tm;
97
98 /* SUFFIX holds the instruction mnemonic suffix if given.
99 (e.g. 'l' for 'movl') */
100 char suffix;
101
102 /* OPERANDS gives the number of given operands. */
103 unsigned int operands;
104
105 /* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number
106 of given register, displacement, memory operands and immediate
107 operands. */
108 unsigned int reg_operands, disp_operands, mem_operands, imm_operands;
109
110 /* TYPES [i] is the type (see above #defines) which tells us how to
111 use OP[i] for the corresponding operand. */
112 unsigned int types[MAX_OPERANDS];
113
114 /* Displacement expression, immediate expression, or register for each
115 operand. */
116 union i386_op op[MAX_OPERANDS];
117
118 /* Flags for operands. */
119 unsigned int flags[MAX_OPERANDS];
120 #define Operand_PCrel 1
121
122 /* Relocation type for operand */
123 RELOC_ENUM reloc[MAX_OPERANDS];
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 #define NUM_FLAG_CODE ((int) CODE_64BIT + 1)
244
245 static enum flag_code flag_code;
246 static int use_rela_relocations = 0;
247
248 /* The names used to print error messages. */
249 static const char *flag_code_names[] =
250 {
251 "32",
252 "16",
253 "64"
254 };
255
256 /* 1 for intel syntax,
257 0 if att syntax. */
258 static int intel_syntax = 0;
259
260 /* 1 if register prefix % not required. */
261 static int allow_naked_reg = 0;
262
263 /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
264 leave, push, and pop instructions so that gcc has the same stack
265 frame as in 32 bit mode. */
266 static char stackop_size = '\0';
267
268 /* Non-zero to quieten some warnings. */
269 static int quiet_warnings = 0;
270
271 /* CPU name. */
272 static const char *cpu_arch_name = NULL;
273
274 /* CPU feature flags. */
275 static unsigned int cpu_arch_flags = CpuUnknownFlags|CpuNo64;
276
277 /* If set, conditional jumps are not automatically promoted to handle
278 larger than a byte offset. */
279 static unsigned int no_cond_jump_promotion = 0;
280
281 /* Interface to relax_segment.
282 There are 3 major relax states for 386 jump insns because the
283 different types of jumps add different sizes to frags when we're
284 figuring out what sort of jump to choose to reach a given label. */
285
286 /* Types. */
287 #define UNCOND_JUMP 1
288 #define COND_JUMP 2
289 #define COND_JUMP86 3
290
291 /* Sizes. */
292 #define CODE16 1
293 #define SMALL 0
294 #define SMALL16 (SMALL|CODE16)
295 #define BIG 2
296 #define BIG16 (BIG|CODE16)
297
298 #ifndef INLINE
299 #ifdef __GNUC__
300 #define INLINE __inline__
301 #else
302 #define INLINE
303 #endif
304 #endif
305
306 #define ENCODE_RELAX_STATE(type, size) \
307 ((relax_substateT) (((type) << 2) | (size)))
308 #define TYPE_FROM_RELAX_STATE(s) \
309 ((s) >> 2)
310 #define DISP_SIZE_FROM_RELAX_STATE(s) \
311 ((((s) & 3) == BIG ? 4 : (((s) & 3) == BIG16 ? 2 : 1)))
312
313 /* This table is used by relax_frag to promote short jumps to long
314 ones where necessary. SMALL (short) jumps may be promoted to BIG
315 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
316 don't allow a short jump in a 32 bit code segment to be promoted to
317 a 16 bit offset jump because it's slower (requires data size
318 prefix), and doesn't work, unless the destination is in the bottom
319 64k of the code segment (The top 16 bits of eip are zeroed). */
320
321 const relax_typeS md_relax_table[] =
322 {
323 /* The fields are:
324 1) most positive reach of this state,
325 2) most negative reach of this state,
326 3) how many bytes this mode will add to the size of the current frag
327 4) which index into the table to try if we can't fit into this one. */
328 {1, 1, 0, 0},
329 {1, 1, 0, 0},
330 {1, 1, 0, 0},
331 {1, 1, 0, 0},
332
333 /* UNCOND_JUMP states. */
334 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
335 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
336 /* dword jmp adds 3 bytes to frag:
337 0 extra opcode bytes, 3 extra displacement bytes. */
338 {0, 0, 3, 0},
339 /* word jmp adds 1 byte to frag:
340 0 extra opcode bytes, 1 extra displacement byte. */
341 {0, 0, 1, 0},
342
343 /* COND_JUMP states. */
344 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
345 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
346 /* dword conditionals adds 4 bytes to frag:
347 1 extra opcode byte, 3 extra displacement bytes. */
348 {0, 0, 4, 0},
349 /* word conditionals add 2 bytes to frag:
350 1 extra opcode byte, 1 extra displacement byte. */
351 {0, 0, 2, 0},
352
353 /* COND_JUMP86 states. */
354 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (COND_JUMP86, BIG)},
355 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (COND_JUMP86, BIG16)},
356 /* dword conditionals adds 4 bytes to frag:
357 1 extra opcode byte, 3 extra displacement bytes. */
358 {0, 0, 4, 0},
359 /* word conditionals add 3 bytes to frag:
360 1 extra opcode byte, 2 extra displacement bytes. */
361 {0, 0, 3, 0}
362 };
363
364 static const arch_entry cpu_arch[] = {
365 {"i8086", Cpu086 },
366 {"i186", Cpu086|Cpu186 },
367 {"i286", Cpu086|Cpu186|Cpu286 },
368 {"i386", Cpu086|Cpu186|Cpu286|Cpu386 },
369 {"i486", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486 },
370 {"i586", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
371 {"i686", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
372 {"pentium", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
373 {"pentiumpro",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
374 {"pentium4", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX|CpuSSE|CpuSSE2 },
375 {"k6", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow },
376 {"athlon", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuMMX|Cpu3dnow },
377 {"sledgehammer",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuSledgehammer|CpuMMX|Cpu3dnow|CpuSSE|CpuSSE2 },
378 {NULL, 0 }
379 };
380
381 void
382 i386_align_code (fragP, count)
383 fragS *fragP;
384 int count;
385 {
386 /* Various efficient no-op patterns for aligning code labels.
387 Note: Don't try to assemble the instructions in the comments.
388 0L and 0w are not legal. */
389 static const char f32_1[] =
390 {0x90}; /* nop */
391 static const char f32_2[] =
392 {0x89,0xf6}; /* movl %esi,%esi */
393 static const char f32_3[] =
394 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
395 static const char f32_4[] =
396 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
397 static const char f32_5[] =
398 {0x90, /* nop */
399 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
400 static const char f32_6[] =
401 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
402 static const char f32_7[] =
403 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
404 static const char f32_8[] =
405 {0x90, /* nop */
406 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
407 static const char f32_9[] =
408 {0x89,0xf6, /* movl %esi,%esi */
409 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
410 static const char f32_10[] =
411 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
412 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
413 static const char f32_11[] =
414 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
415 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
416 static const char f32_12[] =
417 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
418 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
419 static const char f32_13[] =
420 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
421 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
422 static const char f32_14[] =
423 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
424 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
425 static const char f32_15[] =
426 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
427 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
428 static const char f16_3[] =
429 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
430 static const char f16_4[] =
431 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
432 static const char f16_5[] =
433 {0x90, /* nop */
434 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
435 static const char f16_6[] =
436 {0x89,0xf6, /* mov %si,%si */
437 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
438 static const char f16_7[] =
439 {0x8d,0x74,0x00, /* lea 0(%si),%si */
440 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
441 static const char f16_8[] =
442 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
443 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
444 static const char *const f32_patt[] = {
445 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
446 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
447 };
448 static const char *const f16_patt[] = {
449 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
450 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
451 };
452
453 /* ??? We can't use these fillers for x86_64, since they often kills the
454 upper halves. Solve later. */
455 if (flag_code == CODE_64BIT)
456 count = 1;
457
458 if (count > 0 && count <= 15)
459 {
460 if (flag_code == CODE_16BIT)
461 {
462 memcpy (fragP->fr_literal + fragP->fr_fix,
463 f16_patt[count - 1], count);
464 if (count > 8)
465 /* Adjust jump offset. */
466 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
467 }
468 else
469 memcpy (fragP->fr_literal + fragP->fr_fix,
470 f32_patt[count - 1], count);
471 fragP->fr_var = count;
472 }
473 }
474
475 static char *output_invalid PARAMS ((int c));
476 static int i386_operand PARAMS ((char *operand_string));
477 static int i386_intel_operand PARAMS ((char *operand_string, int got_a_float));
478 static const reg_entry *parse_register PARAMS ((char *reg_string,
479 char **end_op));
480
481 #ifndef I386COFF
482 static void s_bss PARAMS ((int));
483 #endif
484
485 symbolS *GOT_symbol; /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
486
487 static INLINE unsigned int
488 mode_from_disp_size (t)
489 unsigned int t;
490 {
491 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
492 }
493
494 static INLINE int
495 fits_in_signed_byte (num)
496 offsetT num;
497 {
498 return (num >= -128) && (num <= 127);
499 }
500
501 static INLINE int
502 fits_in_unsigned_byte (num)
503 offsetT num;
504 {
505 return (num & 0xff) == num;
506 }
507
508 static INLINE int
509 fits_in_unsigned_word (num)
510 offsetT num;
511 {
512 return (num & 0xffff) == num;
513 }
514
515 static INLINE int
516 fits_in_signed_word (num)
517 offsetT num;
518 {
519 return (-32768 <= num) && (num <= 32767);
520 }
521 static INLINE int
522 fits_in_signed_long (num)
523 offsetT num ATTRIBUTE_UNUSED;
524 {
525 #ifndef BFD64
526 return 1;
527 #else
528 return (!(((offsetT) -1 << 31) & num)
529 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
530 #endif
531 } /* fits_in_signed_long() */
532 static INLINE int
533 fits_in_unsigned_long (num)
534 offsetT num ATTRIBUTE_UNUSED;
535 {
536 #ifndef BFD64
537 return 1;
538 #else
539 return (num & (((offsetT) 2 << 31) - 1)) == num;
540 #endif
541 } /* fits_in_unsigned_long() */
542
543 static int
544 smallest_imm_type (num)
545 offsetT num;
546 {
547 if (cpu_arch_flags != (Cpu086 | Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64)
548 && !(cpu_arch_flags & (CpuUnknown)))
549 {
550 /* This code is disabled on the 486 because all the Imm1 forms
551 in the opcode table are slower on the i486. They're the
552 versions with the implicitly specified single-position
553 displacement, which has another syntax if you really want to
554 use that form. */
555 if (num == 1)
556 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
557 }
558 return (fits_in_signed_byte (num)
559 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
560 : fits_in_unsigned_byte (num)
561 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
562 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
563 ? (Imm16 | Imm32 | Imm32S | Imm64)
564 : fits_in_signed_long (num)
565 ? (Imm32 | Imm32S | Imm64)
566 : fits_in_unsigned_long (num)
567 ? (Imm32 | Imm64)
568 : Imm64);
569 }
570
571 static offsetT
572 offset_in_range (val, size)
573 offsetT val;
574 int size;
575 {
576 addressT mask;
577
578 switch (size)
579 {
580 case 1: mask = ((addressT) 1 << 8) - 1; break;
581 case 2: mask = ((addressT) 1 << 16) - 1; break;
582 case 4: mask = ((addressT) 2 << 31) - 1; break;
583 #ifdef BFD64
584 case 8: mask = ((addressT) 2 << 63) - 1; break;
585 #endif
586 default: abort ();
587 }
588
589 /* If BFD64, sign extend val. */
590 if (!use_rela_relocations)
591 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
592 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
593
594 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
595 {
596 char buf1[40], buf2[40];
597
598 sprint_value (buf1, val);
599 sprint_value (buf2, val & mask);
600 as_warn (_("%s shortened to %s"), buf1, buf2);
601 }
602 return val & mask;
603 }
604
605 /* Returns 0 if attempting to add a prefix where one from the same
606 class already exists, 1 if non rep/repne added, 2 if rep/repne
607 added. */
608 static int
609 add_prefix (prefix)
610 unsigned int prefix;
611 {
612 int ret = 1;
613 int q;
614
615 if (prefix >= 0x40 && prefix < 0x50 && flag_code == CODE_64BIT)
616 q = REX_PREFIX;
617 else
618 switch (prefix)
619 {
620 default:
621 abort ();
622
623 case CS_PREFIX_OPCODE:
624 case DS_PREFIX_OPCODE:
625 case ES_PREFIX_OPCODE:
626 case FS_PREFIX_OPCODE:
627 case GS_PREFIX_OPCODE:
628 case SS_PREFIX_OPCODE:
629 q = SEG_PREFIX;
630 break;
631
632 case REPNE_PREFIX_OPCODE:
633 case REPE_PREFIX_OPCODE:
634 ret = 2;
635 /* fall thru */
636 case LOCK_PREFIX_OPCODE:
637 q = LOCKREP_PREFIX;
638 break;
639
640 case FWAIT_OPCODE:
641 q = WAIT_PREFIX;
642 break;
643
644 case ADDR_PREFIX_OPCODE:
645 q = ADDR_PREFIX;
646 break;
647
648 case DATA_PREFIX_OPCODE:
649 q = DATA_PREFIX;
650 break;
651 }
652
653 if (i.prefix[q])
654 {
655 as_bad (_("same type of prefix used twice"));
656 return 0;
657 }
658
659 i.prefixes += 1;
660 i.prefix[q] = prefix;
661 return ret;
662 }
663
664 static void
665 set_code_flag (value)
666 int value;
667 {
668 flag_code = value;
669 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
670 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
671 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
672 {
673 as_bad (_("64bit mode not supported on this CPU."));
674 }
675 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
676 {
677 as_bad (_("32bit mode not supported on this CPU."));
678 }
679 stackop_size = '\0';
680 }
681
682 static void
683 set_16bit_gcc_code_flag (new_code_flag)
684 int new_code_flag;
685 {
686 flag_code = new_code_flag;
687 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
688 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
689 stackop_size = 'l';
690 }
691
692 static void
693 set_intel_syntax (syntax_flag)
694 int syntax_flag;
695 {
696 /* Find out if register prefixing is specified. */
697 int ask_naked_reg = 0;
698
699 SKIP_WHITESPACE ();
700 if (! is_end_of_line[(unsigned char) *input_line_pointer])
701 {
702 char *string = input_line_pointer;
703 int e = get_symbol_end ();
704
705 if (strcmp (string, "prefix") == 0)
706 ask_naked_reg = 1;
707 else if (strcmp (string, "noprefix") == 0)
708 ask_naked_reg = -1;
709 else
710 as_bad (_("bad argument to syntax directive."));
711 *input_line_pointer = e;
712 }
713 demand_empty_rest_of_line ();
714
715 intel_syntax = syntax_flag;
716
717 if (ask_naked_reg == 0)
718 {
719 #ifdef BFD_ASSEMBLER
720 allow_naked_reg = (intel_syntax
721 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
722 #else
723 /* Conservative default. */
724 allow_naked_reg = 0;
725 #endif
726 }
727 else
728 allow_naked_reg = (ask_naked_reg < 0);
729 }
730
731 static void
732 set_cpu_arch (dummy)
733 int dummy ATTRIBUTE_UNUSED;
734 {
735 SKIP_WHITESPACE ();
736
737 if (! is_end_of_line[(unsigned char) *input_line_pointer])
738 {
739 char *string = input_line_pointer;
740 int e = get_symbol_end ();
741 int i;
742
743 for (i = 0; cpu_arch[i].name; i++)
744 {
745 if (strcmp (string, cpu_arch[i].name) == 0)
746 {
747 cpu_arch_name = cpu_arch[i].name;
748 cpu_arch_flags = (cpu_arch[i].flags
749 | (flag_code == CODE_64BIT ? Cpu64 : CpuNo64));
750 break;
751 }
752 }
753 if (!cpu_arch[i].name)
754 as_bad (_("no such architecture: `%s'"), string);
755
756 *input_line_pointer = e;
757 }
758 else
759 as_bad (_("missing cpu architecture"));
760
761 no_cond_jump_promotion = 0;
762 if (*input_line_pointer == ','
763 && ! is_end_of_line[(unsigned char) input_line_pointer[1]])
764 {
765 char *string = ++input_line_pointer;
766 int e = get_symbol_end ();
767
768 if (strcmp (string, "nojumps") == 0)
769 no_cond_jump_promotion = 1;
770 else if (strcmp (string, "jumps") == 0)
771 ;
772 else
773 as_bad (_("no such architecture modifier: `%s'"), string);
774
775 *input_line_pointer = e;
776 }
777
778 demand_empty_rest_of_line ();
779 }
780
781 const pseudo_typeS md_pseudo_table[] =
782 {
783 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
784 {"align", s_align_bytes, 0},
785 #else
786 {"align", s_align_ptwo, 0},
787 #endif
788 {"arch", set_cpu_arch, 0},
789 #ifndef I386COFF
790 {"bss", s_bss, 0},
791 #endif
792 {"ffloat", float_cons, 'f'},
793 {"dfloat", float_cons, 'd'},
794 {"tfloat", float_cons, 'x'},
795 {"value", cons, 2},
796 {"noopt", s_ignore, 0},
797 {"optim", s_ignore, 0},
798 {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
799 {"code16", set_code_flag, CODE_16BIT},
800 {"code32", set_code_flag, CODE_32BIT},
801 {"code64", set_code_flag, CODE_64BIT},
802 {"intel_syntax", set_intel_syntax, 1},
803 {"att_syntax", set_intel_syntax, 0},
804 {"file", dwarf2_directive_file, 0},
805 {"loc", dwarf2_directive_loc, 0},
806 {0, 0, 0}
807 };
808
809 /* For interface with expression (). */
810 extern char *input_line_pointer;
811
812 /* Hash table for instruction mnemonic lookup. */
813 static struct hash_control *op_hash;
814
815 /* Hash table for register lookup. */
816 static struct hash_control *reg_hash;
817 \f
818 #ifdef BFD_ASSEMBLER
819 unsigned long
820 i386_mach ()
821 {
822 if (!strcmp (default_arch, "x86_64"))
823 return bfd_mach_x86_64;
824 else if (!strcmp (default_arch, "i386"))
825 return bfd_mach_i386_i386;
826 else
827 as_fatal (_("Unknown architecture"));
828 }
829 #endif
830 \f
831 void
832 md_begin ()
833 {
834 const char *hash_err;
835
836 /* Initialize op_hash hash table. */
837 op_hash = hash_new ();
838
839 {
840 register const template *optab;
841 register templates *core_optab;
842
843 /* Setup for loop. */
844 optab = i386_optab;
845 core_optab = (templates *) xmalloc (sizeof (templates));
846 core_optab->start = optab;
847
848 while (1)
849 {
850 ++optab;
851 if (optab->name == NULL
852 || strcmp (optab->name, (optab - 1)->name) != 0)
853 {
854 /* different name --> ship out current template list;
855 add to hash table; & begin anew. */
856 core_optab->end = optab;
857 hash_err = hash_insert (op_hash,
858 (optab - 1)->name,
859 (PTR) core_optab);
860 if (hash_err)
861 {
862 as_fatal (_("Internal Error: Can't hash %s: %s"),
863 (optab - 1)->name,
864 hash_err);
865 }
866 if (optab->name == NULL)
867 break;
868 core_optab = (templates *) xmalloc (sizeof (templates));
869 core_optab->start = optab;
870 }
871 }
872 }
873
874 /* Initialize reg_hash hash table. */
875 reg_hash = hash_new ();
876 {
877 register const reg_entry *regtab;
878
879 for (regtab = i386_regtab;
880 regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
881 regtab++)
882 {
883 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
884 if (hash_err)
885 as_fatal (_("Internal Error: Can't hash %s: %s"),
886 regtab->reg_name,
887 hash_err);
888 }
889 }
890
891 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
892 {
893 register int c;
894 register char *p;
895
896 for (c = 0; c < 256; c++)
897 {
898 if (isdigit (c))
899 {
900 digit_chars[c] = c;
901 mnemonic_chars[c] = c;
902 register_chars[c] = c;
903 operand_chars[c] = c;
904 }
905 else if (islower (c))
906 {
907 mnemonic_chars[c] = c;
908 register_chars[c] = c;
909 operand_chars[c] = c;
910 }
911 else if (isupper (c))
912 {
913 mnemonic_chars[c] = tolower (c);
914 register_chars[c] = mnemonic_chars[c];
915 operand_chars[c] = c;
916 }
917
918 if (isalpha (c) || isdigit (c))
919 identifier_chars[c] = c;
920 else if (c >= 128)
921 {
922 identifier_chars[c] = c;
923 operand_chars[c] = c;
924 }
925 }
926
927 #ifdef LEX_AT
928 identifier_chars['@'] = '@';
929 #endif
930 digit_chars['-'] = '-';
931 identifier_chars['_'] = '_';
932 identifier_chars['.'] = '.';
933
934 for (p = operand_special_chars; *p != '\0'; p++)
935 operand_chars[(unsigned char) *p] = *p;
936 }
937
938 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
939 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
940 {
941 record_alignment (text_section, 2);
942 record_alignment (data_section, 2);
943 record_alignment (bss_section, 2);
944 }
945 #endif
946 }
947
948 void
949 i386_print_statistics (file)
950 FILE *file;
951 {
952 hash_print_statistics (file, "i386 opcode", op_hash);
953 hash_print_statistics (file, "i386 register", reg_hash);
954 }
955 \f
956 #ifdef DEBUG386
957
958 /* Debugging routines for md_assemble. */
959 static void pi PARAMS ((char *, i386_insn *));
960 static void pte PARAMS ((template *));
961 static void pt PARAMS ((unsigned int));
962 static void pe PARAMS ((expressionS *));
963 static void ps PARAMS ((symbolS *));
964
965 static void
966 pi (line, x)
967 char *line;
968 i386_insn *x;
969 {
970 unsigned int i;
971
972 fprintf (stdout, "%s: template ", line);
973 pte (&x->tm);
974 fprintf (stdout, " address: base %s index %s scale %x\n",
975 x->base_reg ? x->base_reg->reg_name : "none",
976 x->index_reg ? x->index_reg->reg_name : "none",
977 x->log2_scale_factor);
978 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
979 x->rm.mode, x->rm.reg, x->rm.regmem);
980 fprintf (stdout, " sib: base %x index %x scale %x\n",
981 x->sib.base, x->sib.index, x->sib.scale);
982 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
983 x->rex.mode64, x->rex.extX, x->rex.extY, x->rex.extZ);
984 for (i = 0; i < x->operands; i++)
985 {
986 fprintf (stdout, " #%d: ", i + 1);
987 pt (x->types[i]);
988 fprintf (stdout, "\n");
989 if (x->types[i]
990 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
991 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
992 if (x->types[i] & Imm)
993 pe (x->op[i].imms);
994 if (x->types[i] & Disp)
995 pe (x->op[i].disps);
996 }
997 }
998
999 static void
1000 pte (t)
1001 template *t;
1002 {
1003 unsigned int i;
1004 fprintf (stdout, " %d operands ", t->operands);
1005 fprintf (stdout, "opcode %x ", t->base_opcode);
1006 if (t->extension_opcode != None)
1007 fprintf (stdout, "ext %x ", t->extension_opcode);
1008 if (t->opcode_modifier & D)
1009 fprintf (stdout, "D");
1010 if (t->opcode_modifier & W)
1011 fprintf (stdout, "W");
1012 fprintf (stdout, "\n");
1013 for (i = 0; i < t->operands; i++)
1014 {
1015 fprintf (stdout, " #%d type ", i + 1);
1016 pt (t->operand_types[i]);
1017 fprintf (stdout, "\n");
1018 }
1019 }
1020
1021 static void
1022 pe (e)
1023 expressionS *e;
1024 {
1025 fprintf (stdout, " operation %d\n", e->X_op);
1026 fprintf (stdout, " add_number %ld (%lx)\n",
1027 (long) e->X_add_number, (long) e->X_add_number);
1028 if (e->X_add_symbol)
1029 {
1030 fprintf (stdout, " add_symbol ");
1031 ps (e->X_add_symbol);
1032 fprintf (stdout, "\n");
1033 }
1034 if (e->X_op_symbol)
1035 {
1036 fprintf (stdout, " op_symbol ");
1037 ps (e->X_op_symbol);
1038 fprintf (stdout, "\n");
1039 }
1040 }
1041
1042 static void
1043 ps (s)
1044 symbolS *s;
1045 {
1046 fprintf (stdout, "%s type %s%s",
1047 S_GET_NAME (s),
1048 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
1049 segment_name (S_GET_SEGMENT (s)));
1050 }
1051
1052 struct type_name
1053 {
1054 unsigned int mask;
1055 char *tname;
1056 }
1057
1058 type_names[] =
1059 {
1060 { Reg8, "r8" },
1061 { Reg16, "r16" },
1062 { Reg32, "r32" },
1063 { Reg64, "r64" },
1064 { Imm8, "i8" },
1065 { Imm8S, "i8s" },
1066 { Imm16, "i16" },
1067 { Imm32, "i32" },
1068 { Imm32S, "i32s" },
1069 { Imm64, "i64" },
1070 { Imm1, "i1" },
1071 { BaseIndex, "BaseIndex" },
1072 { Disp8, "d8" },
1073 { Disp16, "d16" },
1074 { Disp32, "d32" },
1075 { Disp32S, "d32s" },
1076 { Disp64, "d64" },
1077 { InOutPortReg, "InOutPortReg" },
1078 { ShiftCount, "ShiftCount" },
1079 { Control, "control reg" },
1080 { Test, "test reg" },
1081 { Debug, "debug reg" },
1082 { FloatReg, "FReg" },
1083 { FloatAcc, "FAcc" },
1084 { SReg2, "SReg2" },
1085 { SReg3, "SReg3" },
1086 { Acc, "Acc" },
1087 { JumpAbsolute, "Jump Absolute" },
1088 { RegMMX, "rMMX" },
1089 { RegXMM, "rXMM" },
1090 { EsSeg, "es" },
1091 { 0, "" }
1092 };
1093
1094 static void
1095 pt (t)
1096 unsigned int t;
1097 {
1098 register struct type_name *ty;
1099
1100 for (ty = type_names; ty->mask; ty++)
1101 if (t & ty->mask)
1102 fprintf (stdout, "%s, ", ty->tname);
1103 fflush (stdout);
1104 }
1105
1106 #endif /* DEBUG386 */
1107 \f
1108 int
1109 tc_i386_force_relocation (fixp)
1110 struct fix *fixp;
1111 {
1112 #ifdef BFD_ASSEMBLER
1113 if (fixp->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1114 || fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1115 return 1;
1116 return 0;
1117 #else
1118 /* For COFF. */
1119 return fixp->fx_r_type == 7;
1120 #endif
1121 }
1122
1123 #ifdef BFD_ASSEMBLER
1124
1125 static bfd_reloc_code_real_type
1126 reloc (size, pcrel, sign, other)
1127 int size;
1128 int pcrel;
1129 int sign;
1130 bfd_reloc_code_real_type other;
1131 {
1132 if (other != NO_RELOC)
1133 return other;
1134
1135 if (pcrel)
1136 {
1137 if (!sign)
1138 as_bad (_("There are no unsigned pc-relative relocations"));
1139 switch (size)
1140 {
1141 case 1: return BFD_RELOC_8_PCREL;
1142 case 2: return BFD_RELOC_16_PCREL;
1143 case 4: return BFD_RELOC_32_PCREL;
1144 }
1145 as_bad (_("can not do %d byte pc-relative relocation"), size);
1146 }
1147 else
1148 {
1149 if (sign)
1150 switch (size)
1151 {
1152 case 4: return BFD_RELOC_X86_64_32S;
1153 }
1154 else
1155 switch (size)
1156 {
1157 case 1: return BFD_RELOC_8;
1158 case 2: return BFD_RELOC_16;
1159 case 4: return BFD_RELOC_32;
1160 case 8: return BFD_RELOC_64;
1161 }
1162 as_bad (_("can not do %s %d byte relocation"),
1163 sign ? "signed" : "unsigned", size);
1164 }
1165
1166 abort ();
1167 return BFD_RELOC_NONE;
1168 }
1169
1170 /* Here we decide which fixups can be adjusted to make them relative to
1171 the beginning of the section instead of the symbol. Basically we need
1172 to make sure that the dynamic relocations are done correctly, so in
1173 some cases we force the original symbol to be used. */
1174
1175 int
1176 tc_i386_fix_adjustable (fixP)
1177 fixS *fixP;
1178 {
1179 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1180 /* Prevent all adjustments to global symbols, or else dynamic
1181 linking will not work correctly. */
1182 if (S_IS_EXTERNAL (fixP->fx_addsy)
1183 || S_IS_WEAK (fixP->fx_addsy))
1184 return 0;
1185 #endif
1186 /* adjust_reloc_syms doesn't know about the GOT. */
1187 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1188 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1189 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1190 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1191 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1192 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
1193 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1194 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1195 return 0;
1196 return 1;
1197 }
1198 #else
1199 #define reloc(SIZE,PCREL,SIGN,OTHER) 0
1200 #define BFD_RELOC_16 0
1201 #define BFD_RELOC_32 0
1202 #define BFD_RELOC_16_PCREL 0
1203 #define BFD_RELOC_32_PCREL 0
1204 #define BFD_RELOC_386_PLT32 0
1205 #define BFD_RELOC_386_GOT32 0
1206 #define BFD_RELOC_386_GOTOFF 0
1207 #define BFD_RELOC_X86_64_PLT32 0
1208 #define BFD_RELOC_X86_64_GOT32 0
1209 #define BFD_RELOC_X86_64_GOTPCREL 0
1210 #endif
1211
1212 static int intel_float_operand PARAMS ((char *mnemonic));
1213
1214 static int
1215 intel_float_operand (mnemonic)
1216 char *mnemonic;
1217 {
1218 if (mnemonic[0] == 'f' && mnemonic[1] == 'i')
1219 return 2;
1220
1221 if (mnemonic[0] == 'f')
1222 return 1;
1223
1224 return 0;
1225 }
1226
1227 /* This is the guts of the machine-dependent assembler. LINE points to a
1228 machine dependent instruction. This function is supposed to emit
1229 the frags/bytes it assembles to. */
1230
1231 void
1232 md_assemble (line)
1233 char *line;
1234 {
1235 /* Points to template once we've found it. */
1236 const template *t;
1237
1238 /* Count the size of the instruction generated. Does not include
1239 variable part of jump insns before relax. */
1240 int insn_size = 0;
1241
1242 int j;
1243
1244 char mnemonic[MAX_MNEM_SIZE];
1245
1246 /* Initialize globals. */
1247 memset (&i, '\0', sizeof (i));
1248 for (j = 0; j < MAX_OPERANDS; j++)
1249 i.reloc[j] = NO_RELOC;
1250 memset (disp_expressions, '\0', sizeof (disp_expressions));
1251 memset (im_expressions, '\0', sizeof (im_expressions));
1252 save_stack_p = save_stack;
1253
1254 /* First parse an instruction mnemonic & call i386_operand for the operands.
1255 We assume that the scrubber has arranged it so that line[0] is the valid
1256 start of a (possibly prefixed) mnemonic. */
1257 {
1258 char *l = line;
1259 char *token_start = l;
1260 char *mnem_p;
1261
1262 /* Non-zero if we found a prefix only acceptable with string insns. */
1263 const char *expecting_string_instruction = NULL;
1264
1265 while (1)
1266 {
1267 mnem_p = mnemonic;
1268 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1269 {
1270 mnem_p++;
1271 if (mnem_p >= mnemonic + sizeof (mnemonic))
1272 {
1273 as_bad (_("no such instruction: `%s'"), token_start);
1274 return;
1275 }
1276 l++;
1277 }
1278 if (!is_space_char (*l)
1279 && *l != END_OF_INSN
1280 && *l != PREFIX_SEPARATOR)
1281 {
1282 as_bad (_("invalid character %s in mnemonic"),
1283 output_invalid (*l));
1284 return;
1285 }
1286 if (token_start == l)
1287 {
1288 if (*l == PREFIX_SEPARATOR)
1289 as_bad (_("expecting prefix; got nothing"));
1290 else
1291 as_bad (_("expecting mnemonic; got nothing"));
1292 return;
1293 }
1294
1295 /* Look up instruction (or prefix) via hash table. */
1296 current_templates = hash_find (op_hash, mnemonic);
1297
1298 if (*l != END_OF_INSN
1299 && (! is_space_char (*l) || l[1] != END_OF_INSN)
1300 && current_templates
1301 && (current_templates->start->opcode_modifier & IsPrefix))
1302 {
1303 /* If we are in 16-bit mode, do not allow addr16 or data16.
1304 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1305 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1306 && (((current_templates->start->opcode_modifier & Size32) != 0)
1307 ^ (flag_code == CODE_16BIT)))
1308 {
1309 as_bad (_("redundant %s prefix"),
1310 current_templates->start->name);
1311 return;
1312 }
1313 /* Add prefix, checking for repeated prefixes. */
1314 switch (add_prefix (current_templates->start->base_opcode))
1315 {
1316 case 0:
1317 return;
1318 case 2:
1319 expecting_string_instruction = current_templates->start->name;
1320 break;
1321 }
1322 /* Skip past PREFIX_SEPARATOR and reset token_start. */
1323 token_start = ++l;
1324 }
1325 else
1326 break;
1327 }
1328
1329 if (!current_templates)
1330 {
1331 /* See if we can get a match by trimming off a suffix. */
1332 switch (mnem_p[-1])
1333 {
1334 case WORD_MNEM_SUFFIX:
1335 case BYTE_MNEM_SUFFIX:
1336 case QWORD_MNEM_SUFFIX:
1337 i.suffix = mnem_p[-1];
1338 mnem_p[-1] = '\0';
1339 current_templates = hash_find (op_hash, mnemonic);
1340 break;
1341 case SHORT_MNEM_SUFFIX:
1342 case LONG_MNEM_SUFFIX:
1343 if (!intel_syntax)
1344 {
1345 i.suffix = mnem_p[-1];
1346 mnem_p[-1] = '\0';
1347 current_templates = hash_find (op_hash, mnemonic);
1348 }
1349 break;
1350
1351 /* Intel Syntax. */
1352 case 'd':
1353 if (intel_syntax)
1354 {
1355 if (intel_float_operand (mnemonic))
1356 i.suffix = SHORT_MNEM_SUFFIX;
1357 else
1358 i.suffix = LONG_MNEM_SUFFIX;
1359 mnem_p[-1] = '\0';
1360 current_templates = hash_find (op_hash, mnemonic);
1361 }
1362 break;
1363 }
1364 if (!current_templates)
1365 {
1366 as_bad (_("no such instruction: `%s'"), token_start);
1367 return;
1368 }
1369 }
1370
1371 /* Check if instruction is supported on specified architecture. */
1372 if (cpu_arch_flags != 0)
1373 {
1374 if ((current_templates->start->cpu_flags & ~(Cpu64 | CpuNo64))
1375 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64)))
1376 {
1377 as_warn (_("`%s' is not supported on `%s'"),
1378 current_templates->start->name, cpu_arch_name);
1379 }
1380 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
1381 {
1382 as_warn (_("use .code16 to ensure correct addressing mode"));
1383 }
1384 }
1385
1386 /* Check for rep/repne without a string instruction. */
1387 if (expecting_string_instruction
1388 && !(current_templates->start->opcode_modifier & IsString))
1389 {
1390 as_bad (_("expecting string instruction after `%s'"),
1391 expecting_string_instruction);
1392 return;
1393 }
1394
1395 /* There may be operands to parse. */
1396 if (*l != END_OF_INSN)
1397 {
1398 /* 1 if operand is pending after ','. */
1399 unsigned int expecting_operand = 0;
1400
1401 /* Non-zero if operand parens not balanced. */
1402 unsigned int paren_not_balanced;
1403
1404 do
1405 {
1406 /* Skip optional white space before operand. */
1407 if (is_space_char (*l))
1408 ++l;
1409 if (!is_operand_char (*l) && *l != END_OF_INSN)
1410 {
1411 as_bad (_("invalid character %s before operand %d"),
1412 output_invalid (*l),
1413 i.operands + 1);
1414 return;
1415 }
1416 token_start = l; /* after white space */
1417 paren_not_balanced = 0;
1418 while (paren_not_balanced || *l != ',')
1419 {
1420 if (*l == END_OF_INSN)
1421 {
1422 if (paren_not_balanced)
1423 {
1424 if (!intel_syntax)
1425 as_bad (_("unbalanced parenthesis in operand %d."),
1426 i.operands + 1);
1427 else
1428 as_bad (_("unbalanced brackets in operand %d."),
1429 i.operands + 1);
1430 return;
1431 }
1432 else
1433 break; /* we are done */
1434 }
1435 else if (!is_operand_char (*l) && !is_space_char (*l))
1436 {
1437 as_bad (_("invalid character %s in operand %d"),
1438 output_invalid (*l),
1439 i.operands + 1);
1440 return;
1441 }
1442 if (!intel_syntax)
1443 {
1444 if (*l == '(')
1445 ++paren_not_balanced;
1446 if (*l == ')')
1447 --paren_not_balanced;
1448 }
1449 else
1450 {
1451 if (*l == '[')
1452 ++paren_not_balanced;
1453 if (*l == ']')
1454 --paren_not_balanced;
1455 }
1456 l++;
1457 }
1458 if (l != token_start)
1459 { /* Yes, we've read in another operand. */
1460 unsigned int operand_ok;
1461 this_operand = i.operands++;
1462 if (i.operands > MAX_OPERANDS)
1463 {
1464 as_bad (_("spurious operands; (%d operands/instruction max)"),
1465 MAX_OPERANDS);
1466 return;
1467 }
1468 /* Now parse operand adding info to 'i' as we go along. */
1469 END_STRING_AND_SAVE (l);
1470
1471 if (intel_syntax)
1472 operand_ok =
1473 i386_intel_operand (token_start,
1474 intel_float_operand (mnemonic));
1475 else
1476 operand_ok = i386_operand (token_start);
1477
1478 RESTORE_END_STRING (l);
1479 if (!operand_ok)
1480 return;
1481 }
1482 else
1483 {
1484 if (expecting_operand)
1485 {
1486 expecting_operand_after_comma:
1487 as_bad (_("expecting operand after ','; got nothing"));
1488 return;
1489 }
1490 if (*l == ',')
1491 {
1492 as_bad (_("expecting operand before ','; got nothing"));
1493 return;
1494 }
1495 }
1496
1497 /* Now *l must be either ',' or END_OF_INSN. */
1498 if (*l == ',')
1499 {
1500 if (*++l == END_OF_INSN)
1501 {
1502 /* Just skip it, if it's \n complain. */
1503 goto expecting_operand_after_comma;
1504 }
1505 expecting_operand = 1;
1506 }
1507 }
1508 while (*l != END_OF_INSN);
1509 }
1510 }
1511
1512 /* Now we've parsed the mnemonic into a set of templates, and have the
1513 operands at hand.
1514
1515 Next, we find a template that matches the given insn,
1516 making sure the overlap of the given operands types is consistent
1517 with the template operand types. */
1518
1519 #define MATCH(overlap, given, template) \
1520 ((overlap & ~JumpAbsolute) \
1521 && ((given) & (BaseIndex|JumpAbsolute)) == ((overlap) & (BaseIndex|JumpAbsolute)))
1522
1523 /* If given types r0 and r1 are registers they must be of the same type
1524 unless the expected operand type register overlap is null.
1525 Note that Acc in a template matches every size of reg. */
1526 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
1527 ( ((g0) & Reg) == 0 || ((g1) & Reg) == 0 || \
1528 ((g0) & Reg) == ((g1) & Reg) || \
1529 ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
1530
1531 {
1532 register unsigned int overlap0, overlap1;
1533 unsigned int overlap2;
1534 unsigned int found_reverse_match;
1535 int suffix_check;
1536
1537 /* All intel opcodes have reversed operands except for "bound" and
1538 "enter". We also don't reverse intersegment "jmp" and "call"
1539 instructions with 2 immediate operands so that the immediate segment
1540 precedes the offset, as it does when in AT&T mode. "enter" and the
1541 intersegment "jmp" and "call" instructions are the only ones that
1542 have two immediate operands. */
1543 if (intel_syntax && i.operands > 1
1544 && (strcmp (mnemonic, "bound") != 0)
1545 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1546 {
1547 union i386_op temp_op;
1548 unsigned int temp_type;
1549 RELOC_ENUM temp_reloc;
1550 int xchg1 = 0;
1551 int xchg2 = 0;
1552
1553 if (i.operands == 2)
1554 {
1555 xchg1 = 0;
1556 xchg2 = 1;
1557 }
1558 else if (i.operands == 3)
1559 {
1560 xchg1 = 0;
1561 xchg2 = 2;
1562 }
1563 temp_type = i.types[xchg2];
1564 i.types[xchg2] = i.types[xchg1];
1565 i.types[xchg1] = temp_type;
1566 temp_op = i.op[xchg2];
1567 i.op[xchg2] = i.op[xchg1];
1568 i.op[xchg1] = temp_op;
1569 temp_reloc = i.reloc[xchg2];
1570 i.reloc[xchg2] = i.reloc[xchg1];
1571 i.reloc[xchg1] = temp_reloc;
1572
1573 if (i.mem_operands == 2)
1574 {
1575 const seg_entry *temp_seg;
1576 temp_seg = i.seg[0];
1577 i.seg[0] = i.seg[1];
1578 i.seg[1] = temp_seg;
1579 }
1580 }
1581
1582 if (i.imm_operands)
1583 {
1584 /* Try to ensure constant immediates are represented in the smallest
1585 opcode possible. */
1586 char guess_suffix = 0;
1587 int op;
1588
1589 if (i.suffix)
1590 guess_suffix = i.suffix;
1591 else if (i.reg_operands)
1592 {
1593 /* Figure out a suffix from the last register operand specified.
1594 We can't do this properly yet, ie. excluding InOutPortReg,
1595 but the following works for instructions with immediates.
1596 In any case, we can't set i.suffix yet. */
1597 for (op = i.operands; --op >= 0;)
1598 if (i.types[op] & Reg)
1599 {
1600 if (i.types[op] & Reg8)
1601 guess_suffix = BYTE_MNEM_SUFFIX;
1602 else if (i.types[op] & Reg16)
1603 guess_suffix = WORD_MNEM_SUFFIX;
1604 else if (i.types[op] & Reg32)
1605 guess_suffix = LONG_MNEM_SUFFIX;
1606 else if (i.types[op] & Reg64)
1607 guess_suffix = QWORD_MNEM_SUFFIX;
1608 break;
1609 }
1610 }
1611 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
1612 guess_suffix = WORD_MNEM_SUFFIX;
1613
1614 for (op = i.operands; --op >= 0;)
1615 if (i.types[op] & Imm)
1616 {
1617 switch (i.op[op].imms->X_op)
1618 {
1619 case O_constant:
1620 /* If a suffix is given, this operand may be shortened. */
1621 switch (guess_suffix)
1622 {
1623 case LONG_MNEM_SUFFIX:
1624 i.types[op] |= Imm32 | Imm64;
1625 break;
1626 case WORD_MNEM_SUFFIX:
1627 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
1628 break;
1629 case BYTE_MNEM_SUFFIX:
1630 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
1631 break;
1632 }
1633
1634 /* If this operand is at most 16 bits, convert it
1635 to a signed 16 bit number before trying to see
1636 whether it will fit in an even smaller size.
1637 This allows a 16-bit operand such as $0xffe0 to
1638 be recognised as within Imm8S range. */
1639 if ((i.types[op] & Imm16)
1640 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
1641 {
1642 i.op[op].imms->X_add_number =
1643 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
1644 }
1645 if ((i.types[op] & Imm32)
1646 && (i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1)) == 0)
1647 {
1648 i.op[op].imms->X_add_number =
1649 (i.op[op].imms->X_add_number ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1650 }
1651 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
1652 /* We must avoid matching of Imm32 templates when 64bit only immediate is available. */
1653 if (guess_suffix == QWORD_MNEM_SUFFIX)
1654 i.types[op] &= ~Imm32;
1655 break;
1656 case O_absent:
1657 case O_register:
1658 abort ();
1659 /* Symbols and expressions. */
1660 default:
1661 /* Convert symbolic operand to proper sizes for matching. */
1662 switch (guess_suffix)
1663 {
1664 case QWORD_MNEM_SUFFIX:
1665 i.types[op] = Imm64 | Imm32S;
1666 break;
1667 case LONG_MNEM_SUFFIX:
1668 i.types[op] = Imm32 | Imm64;
1669 break;
1670 case WORD_MNEM_SUFFIX:
1671 i.types[op] = Imm16 | Imm32 | Imm64;
1672 break;
1673 break;
1674 case BYTE_MNEM_SUFFIX:
1675 i.types[op] = Imm8 | Imm8S | Imm16 | Imm32S | Imm32;
1676 break;
1677 break;
1678 }
1679 break;
1680 }
1681 }
1682 }
1683
1684 if (i.disp_operands)
1685 {
1686 /* Try to use the smallest displacement type too. */
1687 int op;
1688
1689 for (op = i.operands; --op >= 0;)
1690 if ((i.types[op] & Disp)
1691 && i.op[op].disps->X_op == O_constant)
1692 {
1693 offsetT disp = i.op[op].disps->X_add_number;
1694
1695 if (i.types[op] & Disp16)
1696 {
1697 /* We know this operand is at most 16 bits, so
1698 convert to a signed 16 bit number before trying
1699 to see whether it will fit in an even smaller
1700 size. */
1701
1702 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
1703 }
1704 else if (i.types[op] & Disp32)
1705 {
1706 /* We know this operand is at most 32 bits, so convert to a
1707 signed 32 bit number before trying to see whether it will
1708 fit in an even smaller size. */
1709 disp &= (((offsetT) 2 << 31) - 1);
1710 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1711 }
1712 if (flag_code == CODE_64BIT)
1713 {
1714 if (fits_in_signed_long (disp))
1715 i.types[op] |= Disp32S;
1716 if (fits_in_unsigned_long (disp))
1717 i.types[op] |= Disp32;
1718 }
1719 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
1720 && fits_in_signed_byte (disp))
1721 i.types[op] |= Disp8;
1722 }
1723 }
1724
1725 overlap0 = 0;
1726 overlap1 = 0;
1727 overlap2 = 0;
1728 found_reverse_match = 0;
1729 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
1730 ? No_bSuf
1731 : (i.suffix == WORD_MNEM_SUFFIX
1732 ? No_wSuf
1733 : (i.suffix == SHORT_MNEM_SUFFIX
1734 ? No_sSuf
1735 : (i.suffix == LONG_MNEM_SUFFIX
1736 ? No_lSuf
1737 : (i.suffix == QWORD_MNEM_SUFFIX
1738 ? No_qSuf
1739 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX ? No_xSuf : 0))))));
1740
1741 for (t = current_templates->start;
1742 t < current_templates->end;
1743 t++)
1744 {
1745 /* Must have right number of operands. */
1746 if (i.operands != t->operands)
1747 continue;
1748
1749 /* Check the suffix, except for some instructions in intel mode. */
1750 if ((t->opcode_modifier & suffix_check)
1751 && !(intel_syntax
1752 && (t->opcode_modifier & IgnoreSize))
1753 && !(intel_syntax
1754 && t->base_opcode == 0xd9
1755 && (t->extension_opcode == 5 /* 0xd9,5 "fldcw" */
1756 || t->extension_opcode == 7))) /* 0xd9,7 "f{n}stcw" */
1757 continue;
1758
1759 /* Do not verify operands when there are none. */
1760 else if (!t->operands)
1761 {
1762 if (t->cpu_flags & ~cpu_arch_flags)
1763 continue;
1764 /* We've found a match; break out of loop. */
1765 break;
1766 }
1767
1768 overlap0 = i.types[0] & t->operand_types[0];
1769 switch (t->operands)
1770 {
1771 case 1:
1772 if (!MATCH (overlap0, i.types[0], t->operand_types[0]))
1773 continue;
1774 break;
1775 case 2:
1776 case 3:
1777 overlap1 = i.types[1] & t->operand_types[1];
1778 if (!MATCH (overlap0, i.types[0], t->operand_types[0])
1779 || !MATCH (overlap1, i.types[1], t->operand_types[1])
1780 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
1781 t->operand_types[0],
1782 overlap1, i.types[1],
1783 t->operand_types[1]))
1784 {
1785 /* Check if other direction is valid ... */
1786 if ((t->opcode_modifier & (D|FloatD)) == 0)
1787 continue;
1788
1789 /* Try reversing direction of operands. */
1790 overlap0 = i.types[0] & t->operand_types[1];
1791 overlap1 = i.types[1] & t->operand_types[0];
1792 if (!MATCH (overlap0, i.types[0], t->operand_types[1])
1793 || !MATCH (overlap1, i.types[1], t->operand_types[0])
1794 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
1795 t->operand_types[1],
1796 overlap1, i.types[1],
1797 t->operand_types[0]))
1798 {
1799 /* Does not match either direction. */
1800 continue;
1801 }
1802 /* found_reverse_match holds which of D or FloatDR
1803 we've found. */
1804 found_reverse_match = t->opcode_modifier & (D|FloatDR);
1805 }
1806 /* Found a forward 2 operand match here. */
1807 else if (t->operands == 3)
1808 {
1809 /* Here we make use of the fact that there are no
1810 reverse match 3 operand instructions, and all 3
1811 operand instructions only need to be checked for
1812 register consistency between operands 2 and 3. */
1813 overlap2 = i.types[2] & t->operand_types[2];
1814 if (!MATCH (overlap2, i.types[2], t->operand_types[2])
1815 || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1],
1816 t->operand_types[1],
1817 overlap2, i.types[2],
1818 t->operand_types[2]))
1819
1820 continue;
1821 }
1822 /* Found either forward/reverse 2 or 3 operand match here:
1823 slip through to break. */
1824 }
1825 if (t->cpu_flags & ~cpu_arch_flags)
1826 {
1827 found_reverse_match = 0;
1828 continue;
1829 }
1830 /* We've found a match; break out of loop. */
1831 break;
1832 }
1833 if (t == current_templates->end)
1834 {
1835 /* We found no match. */
1836 as_bad (_("suffix or operands invalid for `%s'"),
1837 current_templates->start->name);
1838 return;
1839 }
1840
1841 if (!quiet_warnings)
1842 {
1843 if (!intel_syntax
1844 && ((i.types[0] & JumpAbsolute)
1845 != (t->operand_types[0] & JumpAbsolute)))
1846 {
1847 as_warn (_("indirect %s without `*'"), t->name);
1848 }
1849
1850 if ((t->opcode_modifier & (IsPrefix|IgnoreSize))
1851 == (IsPrefix|IgnoreSize))
1852 {
1853 /* Warn them that a data or address size prefix doesn't
1854 affect assembly of the next line of code. */
1855 as_warn (_("stand-alone `%s' prefix"), t->name);
1856 }
1857 }
1858
1859 /* Copy the template we found. */
1860 i.tm = *t;
1861 if (found_reverse_match)
1862 {
1863 /* If we found a reverse match we must alter the opcode
1864 direction bit. found_reverse_match holds bits to change
1865 (different for int & float insns). */
1866
1867 i.tm.base_opcode ^= found_reverse_match;
1868
1869 i.tm.operand_types[0] = t->operand_types[1];
1870 i.tm.operand_types[1] = t->operand_types[0];
1871 }
1872
1873 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1874 if (SYSV386_COMPAT
1875 && intel_syntax
1876 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1877 i.tm.base_opcode ^= FloatR;
1878
1879 if (i.tm.opcode_modifier & FWait)
1880 if (! add_prefix (FWAIT_OPCODE))
1881 return;
1882
1883 /* Check string instruction segment overrides. */
1884 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1885 {
1886 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
1887 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
1888 {
1889 if (i.seg[0] != NULL && i.seg[0] != &es)
1890 {
1891 as_bad (_("`%s' operand %d must use `%%es' segment"),
1892 i.tm.name,
1893 mem_op + 1);
1894 return;
1895 }
1896 /* There's only ever one segment override allowed per instruction.
1897 This instruction possibly has a legal segment override on the
1898 second operand, so copy the segment to where non-string
1899 instructions store it, allowing common code. */
1900 i.seg[0] = i.seg[1];
1901 }
1902 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
1903 {
1904 if (i.seg[1] != NULL && i.seg[1] != &es)
1905 {
1906 as_bad (_("`%s' operand %d must use `%%es' segment"),
1907 i.tm.name,
1908 mem_op + 2);
1909 return;
1910 }
1911 }
1912 }
1913
1914 if (i.reg_operands && flag_code < CODE_64BIT)
1915 {
1916 int op;
1917 for (op = i.operands; --op >= 0;)
1918 if ((i.types[op] & Reg)
1919 && (i.op[op].regs->reg_flags & (RegRex64|RegRex)))
1920 {
1921 as_bad (_("Extended register `%%%s' available only in 64bit mode."),
1922 i.op[op].regs->reg_name);
1923 return;
1924 }
1925 }
1926
1927 /* If matched instruction specifies an explicit instruction mnemonic
1928 suffix, use it. */
1929 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
1930 {
1931 if (i.tm.opcode_modifier & Size16)
1932 i.suffix = WORD_MNEM_SUFFIX;
1933 else if (i.tm.opcode_modifier & Size64)
1934 i.suffix = QWORD_MNEM_SUFFIX;
1935 else
1936 i.suffix = LONG_MNEM_SUFFIX;
1937 }
1938 else if (i.reg_operands)
1939 {
1940 /* If there's no instruction mnemonic suffix we try to invent one
1941 based on register operands. */
1942 if (!i.suffix)
1943 {
1944 /* We take i.suffix from the last register operand specified,
1945 Destination register type is more significant than source
1946 register type. */
1947 int op;
1948 for (op = i.operands; --op >= 0;)
1949 if ((i.types[op] & Reg)
1950 && !(i.tm.operand_types[op] & InOutPortReg))
1951 {
1952 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
1953 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
1954 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
1955 LONG_MNEM_SUFFIX);
1956 break;
1957 }
1958 }
1959 else if (i.suffix == BYTE_MNEM_SUFFIX)
1960 {
1961 int op;
1962 for (op = i.operands; --op >= 0;)
1963 {
1964 /* If this is an eight bit register, it's OK. If it's
1965 the 16 or 32 bit version of an eight bit register,
1966 we will just use the low portion, and that's OK too. */
1967 if (i.types[op] & Reg8)
1968 continue;
1969
1970 /* movzx and movsx should not generate this warning. */
1971 if (intel_syntax
1972 && (i.tm.base_opcode == 0xfb7
1973 || i.tm.base_opcode == 0xfb6
1974 || i.tm.base_opcode == 0x63
1975 || i.tm.base_opcode == 0xfbe
1976 || i.tm.base_opcode == 0xfbf))
1977 continue;
1978
1979 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4
1980 #if 0
1981 /* Check that the template allows eight bit regs
1982 This kills insns such as `orb $1,%edx', which
1983 maybe should be allowed. */
1984 && (i.tm.operand_types[op] & (Reg8|InOutPortReg))
1985 #endif
1986 )
1987 {
1988 /* Prohibit these changes in the 64bit mode, since
1989 the lowering is more complicated. */
1990 if (flag_code == CODE_64BIT
1991 && (i.tm.operand_types[op] & InOutPortReg) == 0)
1992 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
1993 i.op[op].regs->reg_name,
1994 i.suffix);
1995 #if REGISTER_WARNINGS
1996 if (!quiet_warnings
1997 && (i.tm.operand_types[op] & InOutPortReg) == 0)
1998 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
1999 (i.op[op].regs
2000 + (i.types[op] & Reg16
2001 ? REGNAM_AL - REGNAM_AX
2002 : REGNAM_AL - REGNAM_EAX))->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 + REGNAM_EAX - REGNAM_AX)->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 + REGNAM_AX - REGNAM_EAX)->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.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.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.reloc[1]));
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.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 RELOC_ENUM reloc_type;
3071 int size = 4;
3072 int sign = 0;
3073
3074 if ((i.types[n] & (Imm32S))
3075 && i.suffix == QWORD_MNEM_SUFFIX)
3076 sign = 1;
3077 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3078 {
3079 size = 2;
3080 if (i.types[n] & (Imm8 | Imm8S))
3081 size = 1;
3082 if (i.types[n] & Imm64)
3083 size = 8;
3084 }
3085
3086 insn_size += size;
3087 p = frag_more (size);
3088 reloc_type = reloc (size, 0, sign, i.reloc[n]);
3089 #ifdef BFD_ASSEMBLER
3090 if (reloc_type == BFD_RELOC_32
3091 && GOT_symbol
3092 && GOT_symbol == i.op[n].imms->X_add_symbol
3093 && (i.op[n].imms->X_op == O_symbol
3094 || (i.op[n].imms->X_op == O_add
3095 && ((symbol_get_value_expression
3096 (i.op[n].imms->X_op_symbol)->X_op)
3097 == O_subtract))))
3098 {
3099 /* We don't support dynamic linking on x86-64 yet. */
3100 if (flag_code == CODE_64BIT)
3101 abort ();
3102 reloc_type = BFD_RELOC_386_GOTPC;
3103 i.op[n].imms->X_add_number += 3;
3104 }
3105 #endif
3106 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3107 i.op[n].imms, 0, reloc_type);
3108 }
3109 }
3110 }
3111 }
3112 }
3113
3114 dwarf2_emit_insn (insn_size);
3115
3116 #ifdef DEBUG386
3117 if (flag_debug)
3118 {
3119 pi (line, &i);
3120 }
3121 #endif /* DEBUG386 */
3122 }
3123 }
3124 \f
3125 #ifndef LEX_AT
3126 static char *lex_got PARAMS ((RELOC_ENUM *, int *));
3127
3128 /* Parse operands of the form
3129 <symbol>@GOTOFF+<nnn>
3130 and similar .plt or .got references.
3131
3132 If we find one, set up the correct relocation in RELOC and copy the
3133 input string, minus the `@GOTOFF' into a malloc'd buffer for
3134 parsing by the calling routine. Return this buffer, and if ADJUST
3135 is non-null set it to the length of the string we removed from the
3136 input line. Otherwise return NULL. */
3137 static char *
3138 lex_got (reloc, adjust)
3139 RELOC_ENUM *reloc;
3140 int *adjust;
3141 {
3142 static const char * const mode_name[NUM_FLAG_CODE] = { "32", "16", "64" };
3143 static const struct {
3144 const char *str;
3145 const RELOC_ENUM rel[NUM_FLAG_CODE];
3146 } gotrel[] = {
3147 { "PLT", { BFD_RELOC_386_PLT32, 0, BFD_RELOC_X86_64_PLT32 } },
3148 { "GOTOFF", { BFD_RELOC_386_GOTOFF, 0, 0 } },
3149 { "GOTPCREL", { 0, 0, BFD_RELOC_X86_64_GOTPCREL } },
3150 { "GOT", { BFD_RELOC_386_GOT32, 0, BFD_RELOC_X86_64_GOT32 } }
3151 };
3152 char *cp;
3153 unsigned int j;
3154
3155 for (cp = input_line_pointer; *cp != '@'; cp++)
3156 if (is_end_of_line[(unsigned char) *cp])
3157 return NULL;
3158
3159 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
3160 {
3161 int len;
3162
3163 len = strlen (gotrel[j].str);
3164 if (strncmp (cp + 1, gotrel[j].str, len) == 0)
3165 {
3166 if (gotrel[j].rel[(unsigned int) flag_code] != 0)
3167 {
3168 int first;
3169 char *tmpbuf;
3170
3171 *reloc = gotrel[j].rel[(unsigned int) flag_code];
3172
3173 if (GOT_symbol == NULL)
3174 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3175
3176 /* Replace the relocation token with ' ', so that
3177 errors like foo@GOTOFF1 will be detected. */
3178 first = cp - input_line_pointer;
3179 tmpbuf = xmalloc (strlen (input_line_pointer));
3180 memcpy (tmpbuf, input_line_pointer, first);
3181 tmpbuf[first] = ' ';
3182 strcpy (tmpbuf + first + 1, cp + 1 + len);
3183 if (adjust)
3184 *adjust = len;
3185 return tmpbuf;
3186 }
3187
3188 as_bad (_("@%s reloc is not supported in %s bit mode"),
3189 gotrel[j].str, mode_name[(unsigned int) flag_code]);
3190 return NULL;
3191 }
3192 }
3193
3194 /* Might be a symbol version string. Don't as_bad here. */
3195 return NULL;
3196 }
3197
3198 /* x86_cons_fix_new is called via the expression parsing code when a
3199 reloc is needed. We use this hook to get the correct .got reloc. */
3200 static RELOC_ENUM got_reloc = NO_RELOC;
3201
3202 void
3203 x86_cons_fix_new (frag, off, len, exp)
3204 fragS *frag;
3205 unsigned int off;
3206 unsigned int len;
3207 expressionS *exp;
3208 {
3209 RELOC_ENUM r = reloc (len, 0, 0, got_reloc);
3210 got_reloc = NO_RELOC;
3211 fix_new_exp (frag, off, len, exp, 0, r);
3212 }
3213
3214 void
3215 x86_cons (exp, size)
3216 expressionS *exp;
3217 int size;
3218 {
3219 if (size == 4)
3220 {
3221 /* Handle @GOTOFF and the like in an expression. */
3222 char *save;
3223 char *gotfree_input_line;
3224 int adjust;
3225
3226 save = input_line_pointer;
3227 gotfree_input_line = lex_got (&got_reloc, &adjust);
3228 if (gotfree_input_line)
3229 input_line_pointer = gotfree_input_line;
3230
3231 expression (exp);
3232
3233 if (gotfree_input_line)
3234 {
3235 /* expression () has merrily parsed up to the end of line,
3236 or a comma - in the wrong buffer. Transfer how far
3237 input_line_pointer has moved to the right buffer. */
3238 input_line_pointer = (save
3239 + (input_line_pointer - gotfree_input_line)
3240 + adjust);
3241 free (gotfree_input_line);
3242 }
3243 }
3244 else
3245 expression (exp);
3246 }
3247 #endif
3248
3249 static int i386_immediate PARAMS ((char *));
3250
3251 static int
3252 i386_immediate (imm_start)
3253 char *imm_start;
3254 {
3255 char *save_input_line_pointer;
3256 #ifndef LEX_AT
3257 char *gotfree_input_line;
3258 #endif
3259 segT exp_seg = 0;
3260 expressionS *exp;
3261
3262 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
3263 {
3264 as_bad (_("only 1 or 2 immediate operands are allowed"));
3265 return 0;
3266 }
3267
3268 exp = &im_expressions[i.imm_operands++];
3269 i.op[this_operand].imms = exp;
3270
3271 if (is_space_char (*imm_start))
3272 ++imm_start;
3273
3274 save_input_line_pointer = input_line_pointer;
3275 input_line_pointer = imm_start;
3276
3277 #ifndef LEX_AT
3278 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3279 if (gotfree_input_line)
3280 input_line_pointer = gotfree_input_line;
3281 #endif
3282
3283 exp_seg = expression (exp);
3284
3285 SKIP_WHITESPACE ();
3286 if (*input_line_pointer)
3287 as_bad (_("junk `%s' after expression"), input_line_pointer);
3288
3289 input_line_pointer = save_input_line_pointer;
3290 #ifndef LEX_AT
3291 if (gotfree_input_line)
3292 free (gotfree_input_line);
3293 #endif
3294
3295 if (exp->X_op == O_absent || exp->X_op == O_big)
3296 {
3297 /* Missing or bad expr becomes absolute 0. */
3298 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
3299 imm_start);
3300 exp->X_op = O_constant;
3301 exp->X_add_number = 0;
3302 exp->X_add_symbol = (symbolS *) 0;
3303 exp->X_op_symbol = (symbolS *) 0;
3304 }
3305 else if (exp->X_op == O_constant)
3306 {
3307 /* Size it properly later. */
3308 i.types[this_operand] |= Imm64;
3309 /* If BFD64, sign extend val. */
3310 if (!use_rela_relocations)
3311 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
3312 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
3313 }
3314 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3315 else if (1
3316 #ifdef BFD_ASSEMBLER
3317 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3318 #endif
3319 && exp_seg != text_section
3320 && exp_seg != data_section
3321 && exp_seg != bss_section
3322 && exp_seg != undefined_section
3323 #ifdef BFD_ASSEMBLER
3324 && !bfd_is_com_section (exp_seg)
3325 #endif
3326 )
3327 {
3328 #ifdef BFD_ASSEMBLER
3329 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3330 #else
3331 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3332 #endif
3333 return 0;
3334 }
3335 #endif
3336 else
3337 {
3338 /* This is an address. The size of the address will be
3339 determined later, depending on destination register,
3340 suffix, or the default for the section. */
3341 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
3342 }
3343
3344 return 1;
3345 }
3346
3347 static int i386_scale PARAMS ((char *));
3348
3349 static int
3350 i386_scale (scale)
3351 char *scale;
3352 {
3353 if (!isdigit (*scale))
3354 goto bad_scale;
3355
3356 switch (*scale)
3357 {
3358 case '0':
3359 case '1':
3360 i.log2_scale_factor = 0;
3361 break;
3362 case '2':
3363 i.log2_scale_factor = 1;
3364 break;
3365 case '4':
3366 i.log2_scale_factor = 2;
3367 break;
3368 case '8':
3369 i.log2_scale_factor = 3;
3370 break;
3371 default:
3372 bad_scale:
3373 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
3374 scale);
3375 return 0;
3376 }
3377 if (i.log2_scale_factor != 0 && ! i.index_reg)
3378 {
3379 as_warn (_("scale factor of %d without an index register"),
3380 1 << i.log2_scale_factor);
3381 #if SCALE1_WHEN_NO_INDEX
3382 i.log2_scale_factor = 0;
3383 #endif
3384 }
3385 return 1;
3386 }
3387
3388 static int i386_displacement PARAMS ((char *, char *));
3389
3390 static int
3391 i386_displacement (disp_start, disp_end)
3392 char *disp_start;
3393 char *disp_end;
3394 {
3395 register expressionS *exp;
3396 segT exp_seg = 0;
3397 char *save_input_line_pointer;
3398 #ifndef LEX_AT
3399 char *gotfree_input_line;
3400 #endif
3401 int bigdisp = Disp32;
3402
3403 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3404 bigdisp = Disp16;
3405 if (flag_code == CODE_64BIT)
3406 bigdisp = Disp64;
3407 i.types[this_operand] |= bigdisp;
3408
3409 exp = &disp_expressions[i.disp_operands];
3410 i.op[this_operand].disps = exp;
3411 i.disp_operands++;
3412 save_input_line_pointer = input_line_pointer;
3413 input_line_pointer = disp_start;
3414 END_STRING_AND_SAVE (disp_end);
3415
3416 #ifndef GCC_ASM_O_HACK
3417 #define GCC_ASM_O_HACK 0
3418 #endif
3419 #if GCC_ASM_O_HACK
3420 END_STRING_AND_SAVE (disp_end + 1);
3421 if ((i.types[this_operand] & BaseIndex) != 0
3422 && displacement_string_end[-1] == '+')
3423 {
3424 /* This hack is to avoid a warning when using the "o"
3425 constraint within gcc asm statements.
3426 For instance:
3427
3428 #define _set_tssldt_desc(n,addr,limit,type) \
3429 __asm__ __volatile__ ( \
3430 "movw %w2,%0\n\t" \
3431 "movw %w1,2+%0\n\t" \
3432 "rorl $16,%1\n\t" \
3433 "movb %b1,4+%0\n\t" \
3434 "movb %4,5+%0\n\t" \
3435 "movb $0,6+%0\n\t" \
3436 "movb %h1,7+%0\n\t" \
3437 "rorl $16,%1" \
3438 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
3439
3440 This works great except that the output assembler ends
3441 up looking a bit weird if it turns out that there is
3442 no offset. You end up producing code that looks like:
3443
3444 #APP
3445 movw $235,(%eax)
3446 movw %dx,2+(%eax)
3447 rorl $16,%edx
3448 movb %dl,4+(%eax)
3449 movb $137,5+(%eax)
3450 movb $0,6+(%eax)
3451 movb %dh,7+(%eax)
3452 rorl $16,%edx
3453 #NO_APP
3454
3455 So here we provide the missing zero. */
3456
3457 *displacement_string_end = '0';
3458 }
3459 #endif
3460 #ifndef LEX_AT
3461 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3462 if (gotfree_input_line)
3463 input_line_pointer = gotfree_input_line;
3464 #endif
3465
3466 exp_seg = expression (exp);
3467
3468 #ifdef BFD_ASSEMBLER
3469 /* We do this to make sure that the section symbol is in
3470 the symbol table. We will ultimately change the relocation
3471 to be relative to the beginning of the section. */
3472 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
3473 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3474 {
3475 if (S_IS_LOCAL (exp->X_add_symbol)
3476 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
3477 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
3478 assert (exp->X_op == O_symbol);
3479 exp->X_op = O_subtract;
3480 exp->X_op_symbol = GOT_symbol;
3481 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3482 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
3483 else
3484 i.reloc[this_operand] = BFD_RELOC_32;
3485 }
3486 #endif
3487
3488 SKIP_WHITESPACE ();
3489 if (*input_line_pointer)
3490 as_bad (_("junk `%s' after expression"), input_line_pointer);
3491 #if GCC_ASM_O_HACK
3492 RESTORE_END_STRING (disp_end + 1);
3493 #endif
3494 RESTORE_END_STRING (disp_end);
3495 input_line_pointer = save_input_line_pointer;
3496 #ifndef LEX_AT
3497 if (gotfree_input_line)
3498 free (gotfree_input_line);
3499 #endif
3500
3501 if (exp->X_op == O_absent || exp->X_op == O_big)
3502 {
3503 /* Missing or bad expr becomes absolute 0. */
3504 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
3505 disp_start);
3506 exp->X_op = O_constant;
3507 exp->X_add_number = 0;
3508 exp->X_add_symbol = (symbolS *) 0;
3509 exp->X_op_symbol = (symbolS *) 0;
3510 }
3511
3512 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3513 if (exp->X_op != O_constant
3514 #ifdef BFD_ASSEMBLER
3515 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3516 #endif
3517 && exp_seg != text_section
3518 && exp_seg != data_section
3519 && exp_seg != bss_section
3520 && exp_seg != undefined_section)
3521 {
3522 #ifdef BFD_ASSEMBLER
3523 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3524 #else
3525 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3526 #endif
3527 return 0;
3528 }
3529 #endif
3530 else if (flag_code == CODE_64BIT)
3531 i.types[this_operand] |= Disp32S | Disp32;
3532 return 1;
3533 }
3534
3535 static int i386_index_check PARAMS ((const char *));
3536
3537 /* Make sure the memory operand we've been dealt is valid.
3538 Return 1 on success, 0 on a failure. */
3539
3540 static int
3541 i386_index_check (operand_string)
3542 const char *operand_string;
3543 {
3544 int ok;
3545 #if INFER_ADDR_PREFIX
3546 int fudged = 0;
3547
3548 tryprefix:
3549 #endif
3550 ok = 1;
3551 if (flag_code == CODE_64BIT)
3552 {
3553 /* 64bit checks. */
3554 if ((i.base_reg
3555 && ((i.base_reg->reg_type & Reg64) == 0)
3556 && (i.base_reg->reg_type != BaseIndex
3557 || i.index_reg))
3558 || (i.index_reg
3559 && ((i.index_reg->reg_type & (Reg64|BaseIndex))
3560 != (Reg64|BaseIndex))))
3561 ok = 0;
3562 }
3563 else
3564 {
3565 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3566 {
3567 /* 16bit checks. */
3568 if ((i.base_reg
3569 && ((i.base_reg->reg_type & (Reg16|BaseIndex|RegRex))
3570 != (Reg16|BaseIndex)))
3571 || (i.index_reg
3572 && (((i.index_reg->reg_type & (Reg16|BaseIndex))
3573 != (Reg16|BaseIndex))
3574 || ! (i.base_reg
3575 && i.base_reg->reg_num < 6
3576 && i.index_reg->reg_num >= 6
3577 && i.log2_scale_factor == 0))))
3578 ok = 0;
3579 }
3580 else
3581 {
3582 /* 32bit checks. */
3583 if ((i.base_reg
3584 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
3585 || (i.index_reg
3586 && ((i.index_reg->reg_type & (Reg32|BaseIndex|RegRex))
3587 != (Reg32|BaseIndex))))
3588 ok = 0;
3589 }
3590 }
3591 if (!ok)
3592 {
3593 #if INFER_ADDR_PREFIX
3594 if (flag_code != CODE_64BIT
3595 && i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0')
3596 {
3597 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
3598 i.prefixes += 1;
3599 /* Change the size of any displacement too. At most one of
3600 Disp16 or Disp32 is set.
3601 FIXME. There doesn't seem to be any real need for separate
3602 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
3603 Removing them would probably clean up the code quite a lot. */
3604 if (i.types[this_operand] & (Disp16|Disp32))
3605 i.types[this_operand] ^= (Disp16|Disp32);
3606 fudged = 1;
3607 goto tryprefix;
3608 }
3609 if (fudged)
3610 as_bad (_("`%s' is not a valid base/index expression"),
3611 operand_string);
3612 else
3613 #endif
3614 as_bad (_("`%s' is not a valid %s bit base/index expression"),
3615 operand_string,
3616 flag_code_names[flag_code]);
3617 return 0;
3618 }
3619 return 1;
3620 }
3621
3622 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
3623 on error. */
3624
3625 static int
3626 i386_operand (operand_string)
3627 char *operand_string;
3628 {
3629 const reg_entry *r;
3630 char *end_op;
3631 char *op_string = operand_string;
3632
3633 if (is_space_char (*op_string))
3634 ++op_string;
3635
3636 /* We check for an absolute prefix (differentiating,
3637 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
3638 if (*op_string == ABSOLUTE_PREFIX)
3639 {
3640 ++op_string;
3641 if (is_space_char (*op_string))
3642 ++op_string;
3643 i.types[this_operand] |= JumpAbsolute;
3644 }
3645
3646 /* Check if operand is a register. */
3647 if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
3648 && (r = parse_register (op_string, &end_op)) != NULL)
3649 {
3650 /* Check for a segment override by searching for ':' after a
3651 segment register. */
3652 op_string = end_op;
3653 if (is_space_char (*op_string))
3654 ++op_string;
3655 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
3656 {
3657 switch (r->reg_num)
3658 {
3659 case 0:
3660 i.seg[i.mem_operands] = &es;
3661 break;
3662 case 1:
3663 i.seg[i.mem_operands] = &cs;
3664 break;
3665 case 2:
3666 i.seg[i.mem_operands] = &ss;
3667 break;
3668 case 3:
3669 i.seg[i.mem_operands] = &ds;
3670 break;
3671 case 4:
3672 i.seg[i.mem_operands] = &fs;
3673 break;
3674 case 5:
3675 i.seg[i.mem_operands] = &gs;
3676 break;
3677 }
3678
3679 /* Skip the ':' and whitespace. */
3680 ++op_string;
3681 if (is_space_char (*op_string))
3682 ++op_string;
3683
3684 if (!is_digit_char (*op_string)
3685 && !is_identifier_char (*op_string)
3686 && *op_string != '('
3687 && *op_string != ABSOLUTE_PREFIX)
3688 {
3689 as_bad (_("bad memory operand `%s'"), op_string);
3690 return 0;
3691 }
3692 /* Handle case of %es:*foo. */
3693 if (*op_string == ABSOLUTE_PREFIX)
3694 {
3695 ++op_string;
3696 if (is_space_char (*op_string))
3697 ++op_string;
3698 i.types[this_operand] |= JumpAbsolute;
3699 }
3700 goto do_memory_reference;
3701 }
3702 if (*op_string)
3703 {
3704 as_bad (_("junk `%s' after register"), op_string);
3705 return 0;
3706 }
3707 i.types[this_operand] |= r->reg_type & ~BaseIndex;
3708 i.op[this_operand].regs = r;
3709 i.reg_operands++;
3710 }
3711 else if (*op_string == REGISTER_PREFIX)
3712 {
3713 as_bad (_("bad register name `%s'"), op_string);
3714 return 0;
3715 }
3716 else if (*op_string == IMMEDIATE_PREFIX)
3717 {
3718 ++op_string;
3719 if (i.types[this_operand] & JumpAbsolute)
3720 {
3721 as_bad (_("immediate operand illegal with absolute jump"));
3722 return 0;
3723 }
3724 if (!i386_immediate (op_string))
3725 return 0;
3726 }
3727 else if (is_digit_char (*op_string)
3728 || is_identifier_char (*op_string)
3729 || *op_string == '(')
3730 {
3731 /* This is a memory reference of some sort. */
3732 char *base_string;
3733
3734 /* Start and end of displacement string expression (if found). */
3735 char *displacement_string_start;
3736 char *displacement_string_end;
3737
3738 do_memory_reference:
3739 if ((i.mem_operands == 1
3740 && (current_templates->start->opcode_modifier & IsString) == 0)
3741 || i.mem_operands == 2)
3742 {
3743 as_bad (_("too many memory references for `%s'"),
3744 current_templates->start->name);
3745 return 0;
3746 }
3747
3748 /* Check for base index form. We detect the base index form by
3749 looking for an ')' at the end of the operand, searching
3750 for the '(' matching it, and finding a REGISTER_PREFIX or ','
3751 after the '('. */
3752 base_string = op_string + strlen (op_string);
3753
3754 --base_string;
3755 if (is_space_char (*base_string))
3756 --base_string;
3757
3758 /* If we only have a displacement, set-up for it to be parsed later. */
3759 displacement_string_start = op_string;
3760 displacement_string_end = base_string + 1;
3761
3762 if (*base_string == ')')
3763 {
3764 char *temp_string;
3765 unsigned int parens_balanced = 1;
3766 /* We've already checked that the number of left & right ()'s are
3767 equal, so this loop will not be infinite. */
3768 do
3769 {
3770 base_string--;
3771 if (*base_string == ')')
3772 parens_balanced++;
3773 if (*base_string == '(')
3774 parens_balanced--;
3775 }
3776 while (parens_balanced);
3777
3778 temp_string = base_string;
3779
3780 /* Skip past '(' and whitespace. */
3781 ++base_string;
3782 if (is_space_char (*base_string))
3783 ++base_string;
3784
3785 if (*base_string == ','
3786 || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3787 && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
3788 {
3789 displacement_string_end = temp_string;
3790
3791 i.types[this_operand] |= BaseIndex;
3792
3793 if (i.base_reg)
3794 {
3795 base_string = end_op;
3796 if (is_space_char (*base_string))
3797 ++base_string;
3798 }
3799
3800 /* There may be an index reg or scale factor here. */
3801 if (*base_string == ',')
3802 {
3803 ++base_string;
3804 if (is_space_char (*base_string))
3805 ++base_string;
3806
3807 if ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3808 && (i.index_reg = parse_register (base_string, &end_op)) != NULL)
3809 {
3810 base_string = end_op;
3811 if (is_space_char (*base_string))
3812 ++base_string;
3813 if (*base_string == ',')
3814 {
3815 ++base_string;
3816 if (is_space_char (*base_string))
3817 ++base_string;
3818 }
3819 else if (*base_string != ')')
3820 {
3821 as_bad (_("expecting `,' or `)' after index register in `%s'"),
3822 operand_string);
3823 return 0;
3824 }
3825 }
3826 else if (*base_string == REGISTER_PREFIX)
3827 {
3828 as_bad (_("bad register name `%s'"), base_string);
3829 return 0;
3830 }
3831
3832 /* Check for scale factor. */
3833 if (isdigit ((unsigned char) *base_string))
3834 {
3835 if (!i386_scale (base_string))
3836 return 0;
3837
3838 ++base_string;
3839 if (is_space_char (*base_string))
3840 ++base_string;
3841 if (*base_string != ')')
3842 {
3843 as_bad (_("expecting `)' after scale factor in `%s'"),
3844 operand_string);
3845 return 0;
3846 }
3847 }
3848 else if (!i.index_reg)
3849 {
3850 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
3851 *base_string);
3852 return 0;
3853 }
3854 }
3855 else if (*base_string != ')')
3856 {
3857 as_bad (_("expecting `,' or `)' after base register in `%s'"),
3858 operand_string);
3859 return 0;
3860 }
3861 }
3862 else if (*base_string == REGISTER_PREFIX)
3863 {
3864 as_bad (_("bad register name `%s'"), base_string);
3865 return 0;
3866 }
3867 }
3868
3869 /* If there's an expression beginning the operand, parse it,
3870 assuming displacement_string_start and
3871 displacement_string_end are meaningful. */
3872 if (displacement_string_start != displacement_string_end)
3873 {
3874 if (!i386_displacement (displacement_string_start,
3875 displacement_string_end))
3876 return 0;
3877 }
3878
3879 /* Special case for (%dx) while doing input/output op. */
3880 if (i.base_reg
3881 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
3882 && i.index_reg == 0
3883 && i.log2_scale_factor == 0
3884 && i.seg[i.mem_operands] == 0
3885 && (i.types[this_operand] & Disp) == 0)
3886 {
3887 i.types[this_operand] = InOutPortReg;
3888 return 1;
3889 }
3890
3891 if (i386_index_check (operand_string) == 0)
3892 return 0;
3893 i.mem_operands++;
3894 }
3895 else
3896 {
3897 /* It's not a memory operand; argh! */
3898 as_bad (_("invalid char %s beginning operand %d `%s'"),
3899 output_invalid (*op_string),
3900 this_operand + 1,
3901 op_string);
3902 return 0;
3903 }
3904 return 1; /* Normal return. */
3905 }
3906 \f
3907 /* md_estimate_size_before_relax()
3908
3909 Called just before relax() for rs_machine_dependent frags. The x86
3910 assembler uses these frags to handle variable size jump
3911 instructions.
3912
3913 Any symbol that is now undefined will not become defined.
3914 Return the correct fr_subtype in the frag.
3915 Return the initial "guess for variable size of frag" to caller.
3916 The guess is actually the growth beyond the fixed part. Whatever
3917 we do to grow the fixed or variable part contributes to our
3918 returned value. */
3919
3920 int
3921 md_estimate_size_before_relax (fragP, segment)
3922 register fragS *fragP;
3923 register segT segment;
3924 {
3925 /* We've already got fragP->fr_subtype right; all we have to do is
3926 check for un-relaxable symbols. On an ELF system, we can't relax
3927 an externally visible symbol, because it may be overridden by a
3928 shared library. */
3929 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
3930 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
3931 || S_IS_EXTERNAL (fragP->fr_symbol)
3932 || S_IS_WEAK (fragP->fr_symbol)
3933 #endif
3934 )
3935 {
3936 /* Symbol is undefined in this segment, or we need to keep a
3937 reloc so that weak symbols can be overridden. */
3938 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
3939 RELOC_ENUM reloc_type;
3940 unsigned char *opcode;
3941 int old_fr_fix;
3942
3943 if (fragP->fr_var != NO_RELOC)
3944 reloc_type = fragP->fr_var;
3945 else if (size == 2)
3946 reloc_type = BFD_RELOC_16_PCREL;
3947 else
3948 reloc_type = BFD_RELOC_32_PCREL;
3949
3950 old_fr_fix = fragP->fr_fix;
3951 opcode = (unsigned char *) fragP->fr_opcode;
3952
3953 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
3954 {
3955 case UNCOND_JUMP:
3956 /* Make jmp (0xeb) a (d)word displacement jump. */
3957 opcode[0] = 0xe9;
3958 fragP->fr_fix += size;
3959 fix_new (fragP, old_fr_fix, size,
3960 fragP->fr_symbol,
3961 fragP->fr_offset, 1,
3962 reloc_type);
3963 break;
3964
3965 case COND_JUMP86:
3966 if (no_cond_jump_promotion)
3967 return 1;
3968 if (size == 2)
3969 {
3970 /* Negate the condition, and branch past an
3971 unconditional jump. */
3972 opcode[0] ^= 1;
3973 opcode[1] = 3;
3974 /* Insert an unconditional jump. */
3975 opcode[2] = 0xe9;
3976 /* We added two extra opcode bytes, and have a two byte
3977 offset. */
3978 fragP->fr_fix += 2 + 2;
3979 fix_new (fragP, old_fr_fix + 2, 2,
3980 fragP->fr_symbol,
3981 fragP->fr_offset, 1,
3982 reloc_type);
3983 break;
3984 }
3985 /* Fall through. */
3986
3987 case COND_JUMP:
3988 if (no_cond_jump_promotion)
3989 return 1;
3990 /* This changes the byte-displacement jump 0x7N
3991 to the (d)word-displacement jump 0x0f,0x8N. */
3992 opcode[1] = opcode[0] + 0x10;
3993 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
3994 /* We've added an opcode byte. */
3995 fragP->fr_fix += 1 + size;
3996 fix_new (fragP, old_fr_fix + 1, size,
3997 fragP->fr_symbol,
3998 fragP->fr_offset, 1,
3999 reloc_type);
4000 break;
4001
4002 default:
4003 BAD_CASE (fragP->fr_subtype);
4004 break;
4005 }
4006 frag_wane (fragP);
4007 return fragP->fr_fix - old_fr_fix;
4008 }
4009 /* Guess a short jump. */
4010 return 1;
4011 }
4012
4013 /* Called after relax() is finished.
4014
4015 In: Address of frag.
4016 fr_type == rs_machine_dependent.
4017 fr_subtype is what the address relaxed to.
4018
4019 Out: Any fixSs and constants are set up.
4020 Caller will turn frag into a ".space 0". */
4021
4022 #ifndef BFD_ASSEMBLER
4023 void
4024 md_convert_frag (headers, sec, fragP)
4025 object_headers *headers ATTRIBUTE_UNUSED;
4026 segT sec ATTRIBUTE_UNUSED;
4027 register fragS *fragP;
4028 #else
4029 void
4030 md_convert_frag (abfd, sec, fragP)
4031 bfd *abfd ATTRIBUTE_UNUSED;
4032 segT sec ATTRIBUTE_UNUSED;
4033 register fragS *fragP;
4034 #endif
4035 {
4036 register unsigned char *opcode;
4037 unsigned char *where_to_put_displacement = NULL;
4038 offsetT target_address;
4039 offsetT opcode_address;
4040 unsigned int extension = 0;
4041 offsetT displacement_from_opcode_start;
4042
4043 opcode = (unsigned char *) fragP->fr_opcode;
4044
4045 /* Address we want to reach in file space. */
4046 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
4047 #ifdef BFD_ASSEMBLER
4048 /* Not needed otherwise? */
4049 target_address += symbol_get_frag (fragP->fr_symbol)->fr_address;
4050 #endif
4051
4052 /* Address opcode resides at in file space. */
4053 opcode_address = fragP->fr_address + fragP->fr_fix;
4054
4055 /* Displacement from opcode start to fill into instruction. */
4056 displacement_from_opcode_start = target_address - opcode_address;
4057
4058 if ((fragP->fr_subtype & BIG) == 0)
4059 {
4060 /* Don't have to change opcode. */
4061 extension = 1; /* 1 opcode + 1 displacement */
4062 where_to_put_displacement = &opcode[1];
4063 }
4064 else
4065 {
4066 if (no_cond_jump_promotion
4067 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
4068 as_warn_where (fragP->fr_file, fragP->fr_line, _("long jump required"));
4069
4070 switch (fragP->fr_subtype)
4071 {
4072 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
4073 extension = 4; /* 1 opcode + 4 displacement */
4074 opcode[0] = 0xe9;
4075 where_to_put_displacement = &opcode[1];
4076 break;
4077
4078 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
4079 extension = 2; /* 1 opcode + 2 displacement */
4080 opcode[0] = 0xe9;
4081 where_to_put_displacement = &opcode[1];
4082 break;
4083
4084 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
4085 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
4086 extension = 5; /* 2 opcode + 4 displacement */
4087 opcode[1] = opcode[0] + 0x10;
4088 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4089 where_to_put_displacement = &opcode[2];
4090 break;
4091
4092 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
4093 extension = 3; /* 2 opcode + 2 displacement */
4094 opcode[1] = opcode[0] + 0x10;
4095 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4096 where_to_put_displacement = &opcode[2];
4097 break;
4098
4099 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
4100 extension = 4;
4101 opcode[0] ^= 1;
4102 opcode[1] = 3;
4103 opcode[2] = 0xe9;
4104 where_to_put_displacement = &opcode[3];
4105 break;
4106
4107 default:
4108 BAD_CASE (fragP->fr_subtype);
4109 break;
4110 }
4111 }
4112
4113 /* Now put displacement after opcode. */
4114 md_number_to_chars ((char *) where_to_put_displacement,
4115 (valueT) (displacement_from_opcode_start - extension),
4116 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
4117 fragP->fr_fix += extension;
4118 }
4119 \f
4120 /* Size of byte displacement jmp. */
4121 int md_short_jump_size = 2;
4122
4123 /* Size of dword displacement jmp. */
4124 int md_long_jump_size = 5;
4125
4126 /* Size of relocation record. */
4127 const int md_reloc_size = 8;
4128
4129 void
4130 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
4131 char *ptr;
4132 addressT from_addr, to_addr;
4133 fragS *frag ATTRIBUTE_UNUSED;
4134 symbolS *to_symbol ATTRIBUTE_UNUSED;
4135 {
4136 offsetT offset;
4137
4138 offset = to_addr - (from_addr + 2);
4139 /* Opcode for byte-disp jump. */
4140 md_number_to_chars (ptr, (valueT) 0xeb, 1);
4141 md_number_to_chars (ptr + 1, (valueT) offset, 1);
4142 }
4143
4144 void
4145 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
4146 char *ptr;
4147 addressT from_addr, to_addr;
4148 fragS *frag ATTRIBUTE_UNUSED;
4149 symbolS *to_symbol ATTRIBUTE_UNUSED;
4150 {
4151 offsetT offset;
4152
4153 offset = to_addr - (from_addr + 5);
4154 md_number_to_chars (ptr, (valueT) 0xe9, 1);
4155 md_number_to_chars (ptr + 1, (valueT) offset, 4);
4156 }
4157 \f
4158 /* Apply a fixup (fixS) to segment data, once it has been determined
4159 by our caller that we have all the info we need to fix it up.
4160
4161 On the 386, immediates, displacements, and data pointers are all in
4162 the same (little-endian) format, so we don't need to care about which
4163 we are handling. */
4164
4165 int
4166 md_apply_fix3 (fixP, valp, seg)
4167 /* The fix we're to put in. */
4168 fixS *fixP;
4169
4170 /* Pointer to the value of the bits. */
4171 valueT *valp;
4172
4173 /* Segment fix is from. */
4174 segT seg ATTRIBUTE_UNUSED;
4175 {
4176 register char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
4177 valueT value = *valp;
4178
4179 #if defined (BFD_ASSEMBLER) && !defined (TE_Mach)
4180 if (fixP->fx_pcrel)
4181 {
4182 switch (fixP->fx_r_type)
4183 {
4184 default:
4185 break;
4186
4187 case BFD_RELOC_32:
4188 fixP->fx_r_type = BFD_RELOC_32_PCREL;
4189 break;
4190 case BFD_RELOC_16:
4191 fixP->fx_r_type = BFD_RELOC_16_PCREL;
4192 break;
4193 case BFD_RELOC_8:
4194 fixP->fx_r_type = BFD_RELOC_8_PCREL;
4195 break;
4196 }
4197 }
4198
4199 /* This is a hack. There should be a better way to handle this.
4200 This covers for the fact that bfd_install_relocation will
4201 subtract the current location (for partial_inplace, PC relative
4202 relocations); see more below. */
4203 if ((fixP->fx_r_type == BFD_RELOC_32_PCREL
4204 || fixP->fx_r_type == BFD_RELOC_16_PCREL
4205 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
4206 && fixP->fx_addsy && !use_rela_relocations)
4207 {
4208 #ifndef OBJ_AOUT
4209 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4210 #ifdef TE_PE
4211 || OUTPUT_FLAVOR == bfd_target_coff_flavour
4212 #endif
4213 )
4214 value += fixP->fx_where + fixP->fx_frag->fr_address;
4215 #endif
4216 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4217 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
4218 {
4219 segT fseg = S_GET_SEGMENT (fixP->fx_addsy);
4220
4221 if ((fseg == seg
4222 || (symbol_section_p (fixP->fx_addsy)
4223 && fseg != absolute_section))
4224 && ! S_IS_EXTERNAL (fixP->fx_addsy)
4225 && ! S_IS_WEAK (fixP->fx_addsy)
4226 && S_IS_DEFINED (fixP->fx_addsy)
4227 && ! S_IS_COMMON (fixP->fx_addsy))
4228 {
4229 /* Yes, we add the values in twice. This is because
4230 bfd_perform_relocation subtracts them out again. I think
4231 bfd_perform_relocation is broken, but I don't dare change
4232 it. FIXME. */
4233 value += fixP->fx_where + fixP->fx_frag->fr_address;
4234 }
4235 }
4236 #endif
4237 #if defined (OBJ_COFF) && defined (TE_PE)
4238 /* For some reason, the PE format does not store a section
4239 address offset for a PC relative symbol. */
4240 if (S_GET_SEGMENT (fixP->fx_addsy) != seg)
4241 value += md_pcrel_from (fixP);
4242 #endif
4243 }
4244
4245 /* Fix a few things - the dynamic linker expects certain values here,
4246 and we must not dissappoint it. */
4247 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4248 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4249 && fixP->fx_addsy)
4250 switch (fixP->fx_r_type)
4251 {
4252 case BFD_RELOC_386_PLT32:
4253 case BFD_RELOC_X86_64_PLT32:
4254 /* Make the jump instruction point to the address of the operand. At
4255 runtime we merely add the offset to the actual PLT entry. */
4256 value = -4;
4257 break;
4258 case BFD_RELOC_386_GOTPC:
4259
4260 /* This is tough to explain. We end up with this one if we have
4261 * operands that look like "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal
4262 * here is to obtain the absolute address of the GOT, and it is strongly
4263 * preferable from a performance point of view to avoid using a runtime
4264 * relocation for this. The actual sequence of instructions often look
4265 * something like:
4266 *
4267 * call .L66
4268 * .L66:
4269 * popl %ebx
4270 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
4271 *
4272 * The call and pop essentially return the absolute address of
4273 * the label .L66 and store it in %ebx. The linker itself will
4274 * ultimately change the first operand of the addl so that %ebx points to
4275 * the GOT, but to keep things simple, the .o file must have this operand
4276 * set so that it generates not the absolute address of .L66, but the
4277 * absolute address of itself. This allows the linker itself simply
4278 * treat a GOTPC relocation as asking for a pcrel offset to the GOT to be
4279 * added in, and the addend of the relocation is stored in the operand
4280 * field for the instruction itself.
4281 *
4282 * Our job here is to fix the operand so that it would add the correct
4283 * offset so that %ebx would point to itself. The thing that is tricky is
4284 * that .-.L66 will point to the beginning of the instruction, so we need
4285 * to further modify the operand so that it will point to itself.
4286 * There are other cases where you have something like:
4287 *
4288 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
4289 *
4290 * and here no correction would be required. Internally in the assembler
4291 * we treat operands of this form as not being pcrel since the '.' is
4292 * explicitly mentioned, and I wonder whether it would simplify matters
4293 * to do it this way. Who knows. In earlier versions of the PIC patches,
4294 * the pcrel_adjust field was used to store the correction, but since the
4295 * expression is not pcrel, I felt it would be confusing to do it this
4296 * way. */
4297
4298 value -= 1;
4299 break;
4300 case BFD_RELOC_386_GOT32:
4301 case BFD_RELOC_X86_64_GOT32:
4302 value = 0; /* Fully resolved at runtime. No addend. */
4303 break;
4304 case BFD_RELOC_386_GOTOFF:
4305 case BFD_RELOC_X86_64_GOTPCREL:
4306 break;
4307
4308 case BFD_RELOC_VTABLE_INHERIT:
4309 case BFD_RELOC_VTABLE_ENTRY:
4310 fixP->fx_done = 0;
4311 return 1;
4312
4313 default:
4314 break;
4315 }
4316 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
4317 *valp = value;
4318 #endif /* defined (BFD_ASSEMBLER) && !defined (TE_Mach) */
4319
4320 #ifndef BFD_ASSEMBLER
4321 md_number_to_chars (p, value, fixP->fx_size);
4322 #else
4323 /* Are we finished with this relocation now? */
4324 if (fixP->fx_addsy == 0 && fixP->fx_pcrel == 0)
4325 fixP->fx_done = 1;
4326 else if (use_rela_relocations)
4327 {
4328 fixP->fx_no_overflow = 1;
4329 value = 0;
4330 }
4331 md_number_to_chars (p, value, fixP->fx_size);
4332 #endif
4333
4334 return 1;
4335 }
4336 \f
4337 #define MAX_LITTLENUMS 6
4338
4339 /* Turn the string pointed to by litP into a floating point constant
4340 of type TYPE, and emit the appropriate bytes. The number of
4341 LITTLENUMS emitted is stored in *SIZEP. An error message is
4342 returned, or NULL on OK. */
4343
4344 char *
4345 md_atof (type, litP, sizeP)
4346 int type;
4347 char *litP;
4348 int *sizeP;
4349 {
4350 int prec;
4351 LITTLENUM_TYPE words[MAX_LITTLENUMS];
4352 LITTLENUM_TYPE *wordP;
4353 char *t;
4354
4355 switch (type)
4356 {
4357 case 'f':
4358 case 'F':
4359 prec = 2;
4360 break;
4361
4362 case 'd':
4363 case 'D':
4364 prec = 4;
4365 break;
4366
4367 case 'x':
4368 case 'X':
4369 prec = 5;
4370 break;
4371
4372 default:
4373 *sizeP = 0;
4374 return _("Bad call to md_atof ()");
4375 }
4376 t = atof_ieee (input_line_pointer, type, words);
4377 if (t)
4378 input_line_pointer = t;
4379
4380 *sizeP = prec * sizeof (LITTLENUM_TYPE);
4381 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
4382 the bigendian 386. */
4383 for (wordP = words + prec - 1; prec--;)
4384 {
4385 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
4386 litP += sizeof (LITTLENUM_TYPE);
4387 }
4388 return 0;
4389 }
4390 \f
4391 char output_invalid_buf[8];
4392
4393 static char *
4394 output_invalid (c)
4395 int c;
4396 {
4397 if (isprint (c))
4398 sprintf (output_invalid_buf, "'%c'", c);
4399 else
4400 sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
4401 return output_invalid_buf;
4402 }
4403
4404 /* REG_STRING starts *before* REGISTER_PREFIX. */
4405
4406 static const reg_entry *
4407 parse_register (reg_string, end_op)
4408 char *reg_string;
4409 char **end_op;
4410 {
4411 char *s = reg_string;
4412 char *p;
4413 char reg_name_given[MAX_REG_NAME_SIZE + 1];
4414 const reg_entry *r;
4415
4416 /* Skip possible REGISTER_PREFIX and possible whitespace. */
4417 if (*s == REGISTER_PREFIX)
4418 ++s;
4419
4420 if (is_space_char (*s))
4421 ++s;
4422
4423 p = reg_name_given;
4424 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
4425 {
4426 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
4427 return (const reg_entry *) NULL;
4428 s++;
4429 }
4430
4431 /* For naked regs, make sure that we are not dealing with an identifier.
4432 This prevents confusing an identifier like `eax_var' with register
4433 `eax'. */
4434 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
4435 return (const reg_entry *) NULL;
4436
4437 *end_op = s;
4438
4439 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
4440
4441 /* Handle floating point regs, allowing spaces in the (i) part. */
4442 if (r == i386_regtab /* %st is first entry of table */)
4443 {
4444 if (is_space_char (*s))
4445 ++s;
4446 if (*s == '(')
4447 {
4448 ++s;
4449 if (is_space_char (*s))
4450 ++s;
4451 if (*s >= '0' && *s <= '7')
4452 {
4453 r = &i386_float_regtab[*s - '0'];
4454 ++s;
4455 if (is_space_char (*s))
4456 ++s;
4457 if (*s == ')')
4458 {
4459 *end_op = s + 1;
4460 return r;
4461 }
4462 }
4463 /* We have "%st(" then garbage. */
4464 return (const reg_entry *) NULL;
4465 }
4466 }
4467
4468 return r;
4469 }
4470 \f
4471 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4472 const char *md_shortopts = "kVQ:sq";
4473 #else
4474 const char *md_shortopts = "q";
4475 #endif
4476
4477 struct option md_longopts[] = {
4478 #define OPTION_32 (OPTION_MD_BASE + 0)
4479 {"32", no_argument, NULL, OPTION_32},
4480 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4481 #define OPTION_64 (OPTION_MD_BASE + 1)
4482 {"64", no_argument, NULL, OPTION_64},
4483 #endif
4484 {NULL, no_argument, NULL, 0}
4485 };
4486 size_t md_longopts_size = sizeof (md_longopts);
4487
4488 int
4489 md_parse_option (c, arg)
4490 int c;
4491 char *arg ATTRIBUTE_UNUSED;
4492 {
4493 switch (c)
4494 {
4495 case 'q':
4496 quiet_warnings = 1;
4497 break;
4498
4499 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4500 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
4501 should be emitted or not. FIXME: Not implemented. */
4502 case 'Q':
4503 break;
4504
4505 /* -V: SVR4 argument to print version ID. */
4506 case 'V':
4507 print_version_id ();
4508 break;
4509
4510 /* -k: Ignore for FreeBSD compatibility. */
4511 case 'k':
4512 break;
4513
4514 case 's':
4515 /* -s: On i386 Solaris, this tells the native assembler to use
4516 .stab instead of .stab.excl. We always use .stab anyhow. */
4517 break;
4518
4519 case OPTION_64:
4520 {
4521 const char **list, **l;
4522
4523 list = bfd_target_list ();
4524 for (l = list; *l != NULL; l++)
4525 if (strcmp (*l, "elf64-x86-64") == 0)
4526 {
4527 default_arch = "x86_64";
4528 break;
4529 }
4530 if (*l == NULL)
4531 as_fatal (_("No compiled in support for x86_64"));
4532 free (list);
4533 }
4534 break;
4535 #endif
4536
4537 case OPTION_32:
4538 default_arch = "i386";
4539 break;
4540
4541 default:
4542 return 0;
4543 }
4544 return 1;
4545 }
4546
4547 void
4548 md_show_usage (stream)
4549 FILE *stream;
4550 {
4551 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4552 fprintf (stream, _("\
4553 -Q ignored\n\
4554 -V print assembler version number\n\
4555 -k ignored\n\
4556 -q quieten some warnings\n\
4557 -s ignored\n"));
4558 #else
4559 fprintf (stream, _("\
4560 -q quieten some warnings\n"));
4561 #endif
4562 }
4563
4564 #ifdef BFD_ASSEMBLER
4565 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
4566 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
4567
4568 /* Pick the target format to use. */
4569
4570 const char *
4571 i386_target_format ()
4572 {
4573 if (!strcmp (default_arch, "x86_64"))
4574 set_code_flag (CODE_64BIT);
4575 else if (!strcmp (default_arch, "i386"))
4576 set_code_flag (CODE_32BIT);
4577 else
4578 as_fatal (_("Unknown architecture"));
4579 switch (OUTPUT_FLAVOR)
4580 {
4581 #ifdef OBJ_MAYBE_AOUT
4582 case bfd_target_aout_flavour:
4583 return AOUT_TARGET_FORMAT;
4584 #endif
4585 #ifdef OBJ_MAYBE_COFF
4586 case bfd_target_coff_flavour:
4587 return "coff-i386";
4588 #endif
4589 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
4590 case bfd_target_elf_flavour:
4591 {
4592 if (flag_code == CODE_64BIT)
4593 use_rela_relocations = 1;
4594 return flag_code == CODE_64BIT ? "elf64-x86-64" : "elf32-i386";
4595 }
4596 #endif
4597 default:
4598 abort ();
4599 return NULL;
4600 }
4601 }
4602
4603 #endif /* OBJ_MAYBE_ more than one */
4604 #endif /* BFD_ASSEMBLER */
4605 \f
4606 symbolS *
4607 md_undefined_symbol (name)
4608 char *name;
4609 {
4610 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
4611 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
4612 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
4613 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
4614 {
4615 if (!GOT_symbol)
4616 {
4617 if (symbol_find (name))
4618 as_bad (_("GOT already in symbol table"));
4619 GOT_symbol = symbol_new (name, undefined_section,
4620 (valueT) 0, &zero_address_frag);
4621 };
4622 return GOT_symbol;
4623 }
4624 return 0;
4625 }
4626
4627 /* Round up a section size to the appropriate boundary. */
4628
4629 valueT
4630 md_section_align (segment, size)
4631 segT segment ATTRIBUTE_UNUSED;
4632 valueT size;
4633 {
4634 #ifdef BFD_ASSEMBLER
4635 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4636 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
4637 {
4638 /* For a.out, force the section size to be aligned. If we don't do
4639 this, BFD will align it for us, but it will not write out the
4640 final bytes of the section. This may be a bug in BFD, but it is
4641 easier to fix it here since that is how the other a.out targets
4642 work. */
4643 int align;
4644
4645 align = bfd_get_section_alignment (stdoutput, segment);
4646 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
4647 }
4648 #endif
4649 #endif
4650
4651 return size;
4652 }
4653
4654 /* On the i386, PC-relative offsets are relative to the start of the
4655 next instruction. That is, the address of the offset, plus its
4656 size, since the offset is always the last part of the insn. */
4657
4658 long
4659 md_pcrel_from (fixP)
4660 fixS *fixP;
4661 {
4662 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
4663 }
4664
4665 #ifndef I386COFF
4666
4667 static void
4668 s_bss (ignore)
4669 int ignore ATTRIBUTE_UNUSED;
4670 {
4671 register int temp;
4672
4673 temp = get_absolute_expression ();
4674 subseg_set (bss_section, (subsegT) temp);
4675 demand_empty_rest_of_line ();
4676 }
4677
4678 #endif
4679
4680 #ifdef BFD_ASSEMBLER
4681
4682 void
4683 i386_validate_fix (fixp)
4684 fixS *fixp;
4685 {
4686 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
4687 {
4688 /* GOTOFF relocation are nonsense in 64bit mode. */
4689 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
4690 {
4691 if (flag_code != CODE_64BIT)
4692 abort ();
4693 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
4694 }
4695 else
4696 {
4697 if (flag_code == CODE_64BIT)
4698 abort ();
4699 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
4700 }
4701 fixp->fx_subsy = 0;
4702 }
4703 }
4704
4705 arelent *
4706 tc_gen_reloc (section, fixp)
4707 asection *section ATTRIBUTE_UNUSED;
4708 fixS *fixp;
4709 {
4710 arelent *rel;
4711 bfd_reloc_code_real_type code;
4712
4713 switch (fixp->fx_r_type)
4714 {
4715 case BFD_RELOC_X86_64_PLT32:
4716 case BFD_RELOC_X86_64_GOT32:
4717 case BFD_RELOC_X86_64_GOTPCREL:
4718 case BFD_RELOC_386_PLT32:
4719 case BFD_RELOC_386_GOT32:
4720 case BFD_RELOC_386_GOTOFF:
4721 case BFD_RELOC_386_GOTPC:
4722 case BFD_RELOC_X86_64_32S:
4723 case BFD_RELOC_RVA:
4724 case BFD_RELOC_VTABLE_ENTRY:
4725 case BFD_RELOC_VTABLE_INHERIT:
4726 code = fixp->fx_r_type;
4727 break;
4728 default:
4729 if (fixp->fx_pcrel)
4730 {
4731 switch (fixp->fx_size)
4732 {
4733 default:
4734 as_bad (_("can not do %d byte pc-relative relocation"),
4735 fixp->fx_size);
4736 code = BFD_RELOC_32_PCREL;
4737 break;
4738 case 1: code = BFD_RELOC_8_PCREL; break;
4739 case 2: code = BFD_RELOC_16_PCREL; break;
4740 case 4: code = BFD_RELOC_32_PCREL; break;
4741 }
4742 }
4743 else
4744 {
4745 switch (fixp->fx_size)
4746 {
4747 default:
4748 as_bad (_("can not do %d byte relocation"), fixp->fx_size);
4749 code = BFD_RELOC_32;
4750 break;
4751 case 1: code = BFD_RELOC_8; break;
4752 case 2: code = BFD_RELOC_16; break;
4753 case 4: code = BFD_RELOC_32; break;
4754 case 8: code = BFD_RELOC_64; break;
4755 }
4756 }
4757 break;
4758 }
4759
4760 if (code == BFD_RELOC_32
4761 && GOT_symbol
4762 && fixp->fx_addsy == GOT_symbol)
4763 {
4764 /* We don't support GOTPC on 64bit targets. */
4765 if (flag_code == CODE_64BIT)
4766 abort ();
4767 code = BFD_RELOC_386_GOTPC;
4768 }
4769
4770 rel = (arelent *) xmalloc (sizeof (arelent));
4771 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
4772 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
4773
4774 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
4775 if (!use_rela_relocations)
4776 {
4777 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
4778 vtable entry to be used in the relocation's section offset. */
4779 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
4780 rel->address = fixp->fx_offset;
4781
4782 if (fixp->fx_pcrel)
4783 rel->addend = fixp->fx_addnumber;
4784 else
4785 rel->addend = 0;
4786 }
4787 /* Use the rela in 64bit mode. */
4788 else
4789 {
4790 rel->addend = fixp->fx_offset;
4791 if (fixp->fx_pcrel)
4792 rel->addend -= fixp->fx_size;
4793 }
4794
4795 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
4796 if (rel->howto == NULL)
4797 {
4798 as_bad_where (fixp->fx_file, fixp->fx_line,
4799 _("cannot represent relocation type %s"),
4800 bfd_get_reloc_code_name (code));
4801 /* Set howto to a garbage value so that we can keep going. */
4802 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
4803 assert (rel->howto != NULL);
4804 }
4805
4806 return rel;
4807 }
4808
4809 #else /* ! BFD_ASSEMBLER */
4810
4811 #if (defined(OBJ_AOUT) | defined(OBJ_BOUT))
4812 void
4813 tc_aout_fix_to_chars (where, fixP, segment_address_in_file)
4814 char *where;
4815 fixS *fixP;
4816 relax_addressT segment_address_in_file;
4817 {
4818 /* In: length of relocation (or of address) in chars: 1, 2 or 4.
4819 Out: GNU LD relocation length code: 0, 1, or 2. */
4820
4821 static const unsigned char nbytes_r_length[] = { 42, 0, 1, 42, 2 };
4822 long r_symbolnum;
4823
4824 know (fixP->fx_addsy != NULL);
4825
4826 md_number_to_chars (where,
4827 (valueT) (fixP->fx_frag->fr_address
4828 + fixP->fx_where - segment_address_in_file),
4829 4);
4830
4831 r_symbolnum = (S_IS_DEFINED (fixP->fx_addsy)
4832 ? S_GET_TYPE (fixP->fx_addsy)
4833 : fixP->fx_addsy->sy_number);
4834
4835 where[6] = (r_symbolnum >> 16) & 0x0ff;
4836 where[5] = (r_symbolnum >> 8) & 0x0ff;
4837 where[4] = r_symbolnum & 0x0ff;
4838 where[7] = ((((!S_IS_DEFINED (fixP->fx_addsy)) << 3) & 0x08)
4839 | ((nbytes_r_length[fixP->fx_size] << 1) & 0x06)
4840 | (((fixP->fx_pcrel << 0) & 0x01) & 0x0f));
4841 }
4842
4843 #endif /* OBJ_AOUT or OBJ_BOUT. */
4844
4845 #if defined (I386COFF)
4846
4847 short
4848 tc_coff_fix2rtype (fixP)
4849 fixS *fixP;
4850 {
4851 if (fixP->fx_r_type == R_IMAGEBASE)
4852 return R_IMAGEBASE;
4853
4854 return (fixP->fx_pcrel ?
4855 (fixP->fx_size == 1 ? R_PCRBYTE :
4856 fixP->fx_size == 2 ? R_PCRWORD :
4857 R_PCRLONG) :
4858 (fixP->fx_size == 1 ? R_RELBYTE :
4859 fixP->fx_size == 2 ? R_RELWORD :
4860 R_DIR32));
4861 }
4862
4863 int
4864 tc_coff_sizemachdep (frag)
4865 fragS *frag;
4866 {
4867 if (frag->fr_next)
4868 return (frag->fr_next->fr_address - frag->fr_address);
4869 else
4870 return 0;
4871 }
4872
4873 #endif /* I386COFF */
4874
4875 #endif /* ! BFD_ASSEMBLER */
4876 \f
4877 /* Parse operands using Intel syntax. This implements a recursive descent
4878 parser based on the BNF grammar published in Appendix B of the MASM 6.1
4879 Programmer's Guide.
4880
4881 FIXME: We do not recognize the full operand grammar defined in the MASM
4882 documentation. In particular, all the structure/union and
4883 high-level macro operands are missing.
4884
4885 Uppercase words are terminals, lower case words are non-terminals.
4886 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
4887 bars '|' denote choices. Most grammar productions are implemented in
4888 functions called 'intel_<production>'.
4889
4890 Initial production is 'expr'.
4891
4892 addOp + | -
4893
4894 alpha [a-zA-Z]
4895
4896 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
4897
4898 constant digits [[ radixOverride ]]
4899
4900 dataType BYTE | WORD | DWORD | QWORD | XWORD
4901
4902 digits decdigit
4903 | digits decdigit
4904 | digits hexdigit
4905
4906 decdigit [0-9]
4907
4908 e05 e05 addOp e06
4909 | e06
4910
4911 e06 e06 mulOp e09
4912 | e09
4913
4914 e09 OFFSET e10
4915 | e09 PTR e10
4916 | e09 : e10
4917 | e10
4918
4919 e10 e10 [ expr ]
4920 | e11
4921
4922 e11 ( expr )
4923 | [ expr ]
4924 | constant
4925 | dataType
4926 | id
4927 | $
4928 | register
4929
4930 => expr SHORT e05
4931 | e05
4932
4933 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
4934 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
4935
4936 hexdigit a | b | c | d | e | f
4937 | A | B | C | D | E | F
4938
4939 id alpha
4940 | id alpha
4941 | id decdigit
4942
4943 mulOp * | / | MOD
4944
4945 quote " | '
4946
4947 register specialRegister
4948 | gpRegister
4949 | byteRegister
4950
4951 segmentRegister CS | DS | ES | FS | GS | SS
4952
4953 specialRegister CR0 | CR2 | CR3
4954 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
4955 | TR3 | TR4 | TR5 | TR6 | TR7
4956
4957 We simplify the grammar in obvious places (e.g., register parsing is
4958 done by calling parse_register) and eliminate immediate left recursion
4959 to implement a recursive-descent parser.
4960
4961 expr SHORT e05
4962 | e05
4963
4964 e05 e06 e05'
4965
4966 e05' addOp e06 e05'
4967 | Empty
4968
4969 e06 e09 e06'
4970
4971 e06' mulOp e09 e06'
4972 | Empty
4973
4974 e09 OFFSET e10 e09'
4975 | e10 e09'
4976
4977 e09' PTR e10 e09'
4978 | : e10 e09'
4979 | Empty
4980
4981 e10 e11 e10'
4982
4983 e10' [ expr ] e10'
4984 | Empty
4985
4986 e11 ( expr )
4987 | [ expr ]
4988 | BYTE
4989 | WORD
4990 | DWORD
4991 | QWORD
4992 | XWORD
4993 | .
4994 | $
4995 | register
4996 | id
4997 | constant */
4998
4999 /* Parsing structure for the intel syntax parser. Used to implement the
5000 semantic actions for the operand grammar. */
5001 struct intel_parser_s
5002 {
5003 char *op_string; /* The string being parsed. */
5004 int got_a_float; /* Whether the operand is a float. */
5005 int op_modifier; /* Operand modifier. */
5006 int is_mem; /* 1 if operand is memory reference. */
5007 const reg_entry *reg; /* Last register reference found. */
5008 char *disp; /* Displacement string being built. */
5009 };
5010
5011 static struct intel_parser_s intel_parser;
5012
5013 /* Token structure for parsing intel syntax. */
5014 struct intel_token
5015 {
5016 int code; /* Token code. */
5017 const reg_entry *reg; /* Register entry for register tokens. */
5018 char *str; /* String representation. */
5019 };
5020
5021 static struct intel_token cur_token, prev_token;
5022
5023 /* Token codes for the intel parser. Since T_SHORT is already used
5024 by COFF, undefine it first to prevent a warning. */
5025 #define T_NIL -1
5026 #define T_CONST 1
5027 #define T_REG 2
5028 #define T_BYTE 3
5029 #define T_WORD 4
5030 #define T_DWORD 5
5031 #define T_QWORD 6
5032 #define T_XWORD 7
5033 #undef T_SHORT
5034 #define T_SHORT 8
5035 #define T_OFFSET 9
5036 #define T_PTR 10
5037 #define T_ID 11
5038
5039 /* Prototypes for intel parser functions. */
5040 static int intel_match_token PARAMS ((int code));
5041 static void intel_get_token PARAMS ((void));
5042 static void intel_putback_token PARAMS ((void));
5043 static int intel_expr PARAMS ((void));
5044 static int intel_e05 PARAMS ((void));
5045 static int intel_e05_1 PARAMS ((void));
5046 static int intel_e06 PARAMS ((void));
5047 static int intel_e06_1 PARAMS ((void));
5048 static int intel_e09 PARAMS ((void));
5049 static int intel_e09_1 PARAMS ((void));
5050 static int intel_e10 PARAMS ((void));
5051 static int intel_e10_1 PARAMS ((void));
5052 static int intel_e11 PARAMS ((void));
5053
5054 static int
5055 i386_intel_operand (operand_string, got_a_float)
5056 char *operand_string;
5057 int got_a_float;
5058 {
5059 int ret;
5060 char *p;
5061
5062 /* Initialize token holders. */
5063 cur_token.code = prev_token.code = T_NIL;
5064 cur_token.reg = prev_token.reg = NULL;
5065 cur_token.str = prev_token.str = NULL;
5066
5067 /* Initialize parser structure. */
5068 p = intel_parser.op_string = (char *) malloc (strlen (operand_string) + 1);
5069 if (p == NULL)
5070 abort ();
5071 strcpy (intel_parser.op_string, operand_string);
5072 intel_parser.got_a_float = got_a_float;
5073 intel_parser.op_modifier = -1;
5074 intel_parser.is_mem = 0;
5075 intel_parser.reg = NULL;
5076 intel_parser.disp = (char *) malloc (strlen (operand_string) + 1);
5077 if (intel_parser.disp == NULL)
5078 abort ();
5079 intel_parser.disp[0] = '\0';
5080
5081 /* Read the first token and start the parser. */
5082 intel_get_token ();
5083 ret = intel_expr ();
5084
5085 if (ret)
5086 {
5087 /* If we found a memory reference, hand it over to i386_displacement
5088 to fill in the rest of the operand fields. */
5089 if (intel_parser.is_mem)
5090 {
5091 if ((i.mem_operands == 1
5092 && (current_templates->start->opcode_modifier & IsString) == 0)
5093 || i.mem_operands == 2)
5094 {
5095 as_bad (_("too many memory references for '%s'"),
5096 current_templates->start->name);
5097 ret = 0;
5098 }
5099 else
5100 {
5101 char *s = intel_parser.disp;
5102 i.mem_operands++;
5103
5104 /* Add the displacement expression. */
5105 if (*s != '\0')
5106 ret = i386_displacement (s, s + strlen (s))
5107 && i386_index_check (s);
5108 }
5109 }
5110
5111 /* Constant and OFFSET expressions are handled by i386_immediate. */
5112 else if (intel_parser.op_modifier == OFFSET_FLAT
5113 || intel_parser.reg == NULL)
5114 ret = i386_immediate (intel_parser.disp);
5115 }
5116
5117 free (p);
5118 free (intel_parser.disp);
5119
5120 return ret;
5121 }
5122
5123 /* expr SHORT e05
5124 | e05 */
5125 static int
5126 intel_expr ()
5127 {
5128 /* expr SHORT e05 */
5129 if (cur_token.code == T_SHORT)
5130 {
5131 intel_parser.op_modifier = SHORT;
5132 intel_match_token (T_SHORT);
5133
5134 return (intel_e05 ());
5135 }
5136
5137 /* expr e05 */
5138 else
5139 return intel_e05 ();
5140 }
5141
5142 /* e05 e06 e05'
5143
5144 e05' addOp e06 e05'
5145 | Empty */
5146 static int
5147 intel_e05 ()
5148 {
5149 return (intel_e06 () && intel_e05_1 ());
5150 }
5151
5152 static int
5153 intel_e05_1 ()
5154 {
5155 /* e05' addOp e06 e05' */
5156 if (cur_token.code == '+' || cur_token.code == '-')
5157 {
5158 strcat (intel_parser.disp, cur_token.str);
5159 intel_match_token (cur_token.code);
5160
5161 return (intel_e06 () && intel_e05_1 ());
5162 }
5163
5164 /* e05' Empty */
5165 else
5166 return 1;
5167 }
5168
5169 /* e06 e09 e06'
5170
5171 e06' mulOp e09 e06'
5172 | Empty */
5173 static int
5174 intel_e06 ()
5175 {
5176 return (intel_e09 () && intel_e06_1 ());
5177 }
5178
5179 static int
5180 intel_e06_1 ()
5181 {
5182 /* e06' mulOp e09 e06' */
5183 if (cur_token.code == '*' || cur_token.code == '/')
5184 {
5185 strcat (intel_parser.disp, cur_token.str);
5186 intel_match_token (cur_token.code);
5187
5188 return (intel_e09 () && intel_e06_1 ());
5189 }
5190
5191 /* e06' Empty */
5192 else
5193 return 1;
5194 }
5195
5196 /* e09 OFFSET e10 e09'
5197 | e10 e09'
5198
5199 e09' PTR e10 e09'
5200 | : e10 e09'
5201 | Empty */
5202 static int
5203 intel_e09 ()
5204 {
5205 /* e09 OFFSET e10 e09' */
5206 if (cur_token.code == T_OFFSET)
5207 {
5208 intel_parser.is_mem = 0;
5209 intel_parser.op_modifier = OFFSET_FLAT;
5210 intel_match_token (T_OFFSET);
5211
5212 return (intel_e10 () && intel_e09_1 ());
5213 }
5214
5215 /* e09 e10 e09' */
5216 else
5217 return (intel_e10 () && intel_e09_1 ());
5218 }
5219
5220 static int
5221 intel_e09_1 ()
5222 {
5223 /* e09' PTR e10 e09' */
5224 if (cur_token.code == T_PTR)
5225 {
5226 if (prev_token.code == T_BYTE)
5227 i.suffix = BYTE_MNEM_SUFFIX;
5228
5229 else if (prev_token.code == T_WORD)
5230 {
5231 if (intel_parser.got_a_float == 2) /* "fi..." */
5232 i.suffix = SHORT_MNEM_SUFFIX;
5233 else
5234 i.suffix = WORD_MNEM_SUFFIX;
5235 }
5236
5237 else if (prev_token.code == T_DWORD)
5238 {
5239 if (intel_parser.got_a_float == 1) /* "f..." */
5240 i.suffix = SHORT_MNEM_SUFFIX;
5241 else
5242 i.suffix = LONG_MNEM_SUFFIX;
5243 }
5244
5245 else if (prev_token.code == T_QWORD)
5246 {
5247 if (intel_parser.got_a_float == 1) /* "f..." */
5248 i.suffix = LONG_MNEM_SUFFIX;
5249 else
5250 i.suffix = QWORD_MNEM_SUFFIX;
5251 }
5252
5253 else if (prev_token.code == T_XWORD)
5254 i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
5255
5256 else
5257 {
5258 as_bad (_("Unknown operand modifier `%s'\n"), prev_token.str);
5259 return 0;
5260 }
5261
5262 intel_match_token (T_PTR);
5263
5264 return (intel_e10 () && intel_e09_1 ());
5265 }
5266
5267 /* e09 : e10 e09' */
5268 else if (cur_token.code == ':')
5269 {
5270 /* Mark as a memory operand only if it's not already known to be an
5271 offset expression. */
5272 if (intel_parser.op_modifier != OFFSET_FLAT)
5273 intel_parser.is_mem = 1;
5274
5275 return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
5276 }
5277
5278 /* e09' Empty */
5279 else
5280 return 1;
5281 }
5282
5283 /* e10 e11 e10'
5284
5285 e10' [ expr ] e10'
5286 | Empty */
5287 static int
5288 intel_e10 ()
5289 {
5290 return (intel_e11 () && intel_e10_1 ());
5291 }
5292
5293 static int
5294 intel_e10_1 ()
5295 {
5296 /* e10' [ expr ] e10' */
5297 if (cur_token.code == '[')
5298 {
5299 intel_match_token ('[');
5300
5301 /* Mark as a memory operand only if it's not already known to be an
5302 offset expression. If it's an offset expression, we need to keep
5303 the brace in. */
5304 if (intel_parser.op_modifier != OFFSET_FLAT)
5305 intel_parser.is_mem = 1;
5306 else
5307 strcat (intel_parser.disp, "[");
5308
5309 /* Add a '+' to the displacement string if necessary. */
5310 if (*intel_parser.disp != '\0'
5311 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5312 strcat (intel_parser.disp, "+");
5313
5314 if (intel_expr () && intel_match_token (']'))
5315 {
5316 /* Preserve brackets when the operand is an offset expression. */
5317 if (intel_parser.op_modifier == OFFSET_FLAT)
5318 strcat (intel_parser.disp, "]");
5319
5320 return intel_e10_1 ();
5321 }
5322 else
5323 return 0;
5324 }
5325
5326 /* e10' Empty */
5327 else
5328 return 1;
5329 }
5330
5331 /* e11 ( expr )
5332 | [ expr ]
5333 | BYTE
5334 | WORD
5335 | DWORD
5336 | QWORD
5337 | XWORD
5338 | $
5339 | .
5340 | register
5341 | id
5342 | constant */
5343 static int
5344 intel_e11 ()
5345 {
5346 /* e11 ( expr ) */
5347 if (cur_token.code == '(')
5348 {
5349 intel_match_token ('(');
5350 strcat (intel_parser.disp, "(");
5351
5352 if (intel_expr () && intel_match_token (')'))
5353 {
5354 strcat (intel_parser.disp, ")");
5355 return 1;
5356 }
5357 else
5358 return 0;
5359 }
5360
5361 /* e11 [ expr ] */
5362 else if (cur_token.code == '[')
5363 {
5364 intel_match_token ('[');
5365
5366 /* Mark as a memory operand only if it's not already known to be an
5367 offset expression. If it's an offset expression, we need to keep
5368 the brace in. */
5369 if (intel_parser.op_modifier != OFFSET_FLAT)
5370 intel_parser.is_mem = 1;
5371 else
5372 strcat (intel_parser.disp, "[");
5373
5374 /* Operands for jump/call inside brackets denote absolute addresses. */
5375 if (current_templates->start->opcode_modifier & Jump
5376 || current_templates->start->opcode_modifier & JumpDword
5377 || current_templates->start->opcode_modifier & JumpByte
5378 || current_templates->start->opcode_modifier & JumpInterSegment)
5379 i.types[this_operand] |= JumpAbsolute;
5380
5381 /* Add a '+' to the displacement string if necessary. */
5382 if (*intel_parser.disp != '\0'
5383 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5384 strcat (intel_parser.disp, "+");
5385
5386 if (intel_expr () && intel_match_token (']'))
5387 {
5388 /* Preserve brackets when the operand is an offset expression. */
5389 if (intel_parser.op_modifier == OFFSET_FLAT)
5390 strcat (intel_parser.disp, "]");
5391
5392 return 1;
5393 }
5394 else
5395 return 0;
5396 }
5397
5398 /* e11 BYTE
5399 | WORD
5400 | DWORD
5401 | QWORD
5402 | XWORD */
5403 else if (cur_token.code == T_BYTE
5404 || cur_token.code == T_WORD
5405 || cur_token.code == T_DWORD
5406 || cur_token.code == T_QWORD
5407 || cur_token.code == T_XWORD)
5408 {
5409 intel_match_token (cur_token.code);
5410
5411 return 1;
5412 }
5413
5414 /* e11 $
5415 | . */
5416 else if (cur_token.code == '$' || cur_token.code == '.')
5417 {
5418 strcat (intel_parser.disp, cur_token.str);
5419 intel_match_token (cur_token.code);
5420
5421 /* Mark as a memory operand only if it's not already known to be an
5422 offset expression. */
5423 if (intel_parser.op_modifier != OFFSET_FLAT)
5424 intel_parser.is_mem = 1;
5425
5426 return 1;
5427 }
5428
5429 /* e11 register */
5430 else if (cur_token.code == T_REG)
5431 {
5432 const reg_entry *reg = intel_parser.reg = cur_token.reg;
5433
5434 intel_match_token (T_REG);
5435
5436 /* Check for segment change. */
5437 if (cur_token.code == ':')
5438 {
5439 if (reg->reg_type & (SReg2 | SReg3))
5440 {
5441 switch (reg->reg_num)
5442 {
5443 case 0:
5444 i.seg[i.mem_operands] = &es;
5445 break;
5446 case 1:
5447 i.seg[i.mem_operands] = &cs;
5448 break;
5449 case 2:
5450 i.seg[i.mem_operands] = &ss;
5451 break;
5452 case 3:
5453 i.seg[i.mem_operands] = &ds;
5454 break;
5455 case 4:
5456 i.seg[i.mem_operands] = &fs;
5457 break;
5458 case 5:
5459 i.seg[i.mem_operands] = &gs;
5460 break;
5461 }
5462 }
5463 else
5464 {
5465 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
5466 return 0;
5467 }
5468 }
5469
5470 /* Not a segment register. Check for register scaling. */
5471 else if (cur_token.code == '*')
5472 {
5473 if (!intel_parser.is_mem)
5474 {
5475 as_bad (_("Register scaling only allowed in memory operands."));
5476 return 0;
5477 }
5478
5479 /* What follows must be a valid scale. */
5480 if (intel_match_token ('*')
5481 && strchr ("01248", *cur_token.str))
5482 {
5483 i.index_reg = reg;
5484 i.types[this_operand] |= BaseIndex;
5485
5486 /* Set the scale after setting the register (otherwise,
5487 i386_scale will complain) */
5488 i386_scale (cur_token.str);
5489 intel_match_token (T_CONST);
5490 }
5491 else
5492 {
5493 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
5494 cur_token.str);
5495 return 0;
5496 }
5497 }
5498
5499 /* No scaling. If this is a memory operand, the register is either a
5500 base register (first occurrence) or an index register (second
5501 occurrence). */
5502 else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3)))
5503 {
5504 if (i.base_reg && i.index_reg)
5505 {
5506 as_bad (_("Too many register references in memory operand.\n"));
5507 return 0;
5508 }
5509
5510 if (i.base_reg == NULL)
5511 i.base_reg = reg;
5512 else
5513 i.index_reg = reg;
5514
5515 i.types[this_operand] |= BaseIndex;
5516 }
5517
5518 /* Offset modifier. Add the register to the displacement string to be
5519 parsed as an immediate expression after we're done. */
5520 else if (intel_parser.op_modifier == OFFSET_FLAT)
5521 strcat (intel_parser.disp, reg->reg_name);
5522
5523 /* It's neither base nor index nor offset. */
5524 else
5525 {
5526 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
5527 i.op[this_operand].regs = reg;
5528 i.reg_operands++;
5529 }
5530
5531 /* Since registers are not part of the displacement string (except
5532 when we're parsing offset operands), we may need to remove any
5533 preceding '+' from the displacement string. */
5534 if (*intel_parser.disp != '\0'
5535 && intel_parser.op_modifier != OFFSET_FLAT)
5536 {
5537 char *s = intel_parser.disp;
5538 s += strlen (s) - 1;
5539 if (*s == '+')
5540 *s = '\0';
5541 }
5542
5543 return 1;
5544 }
5545
5546 /* e11 id */
5547 else if (cur_token.code == T_ID)
5548 {
5549 /* Add the identifier to the displacement string. */
5550 strcat (intel_parser.disp, cur_token.str);
5551 intel_match_token (T_ID);
5552
5553 /* The identifier represents a memory reference only if it's not
5554 preceded by an offset modifier. */
5555 if (intel_parser.op_modifier != OFFSET_FLAT)
5556 intel_parser.is_mem = 1;
5557
5558 return 1;
5559 }
5560
5561 /* e11 constant */
5562 else if (cur_token.code == T_CONST
5563 || cur_token.code == '-'
5564 || cur_token.code == '+')
5565 {
5566 char *save_str;
5567
5568 /* Allow constants that start with `+' or `-'. */
5569 if (cur_token.code == '-' || cur_token.code == '+')
5570 {
5571 strcat (intel_parser.disp, cur_token.str);
5572 intel_match_token (cur_token.code);
5573 if (cur_token.code != T_CONST)
5574 {
5575 as_bad (_("Syntax error. Expecting a constant. Got `%s'.\n"),
5576 cur_token.str);
5577 return 0;
5578 }
5579 }
5580
5581 save_str = (char *) malloc (strlen (cur_token.str) + 1);
5582 if (save_str == NULL)
5583 abort ();
5584 strcpy (save_str, cur_token.str);
5585
5586 /* Get the next token to check for register scaling. */
5587 intel_match_token (cur_token.code);
5588
5589 /* Check if this constant is a scaling factor for an index register. */
5590 if (cur_token.code == '*')
5591 {
5592 if (intel_match_token ('*') && cur_token.code == T_REG)
5593 {
5594 if (!intel_parser.is_mem)
5595 {
5596 as_bad (_("Register scaling only allowed in memory operands."));
5597 return 0;
5598 }
5599
5600 /* The constant is followed by `* reg', so it must be
5601 a valid scale. */
5602 if (strchr ("01248", *save_str))
5603 {
5604 i.index_reg = cur_token.reg;
5605 i.types[this_operand] |= BaseIndex;
5606
5607 /* Set the scale after setting the register (otherwise,
5608 i386_scale will complain) */
5609 i386_scale (save_str);
5610 intel_match_token (T_REG);
5611
5612 /* Since registers are not part of the displacement
5613 string, we may need to remove any preceding '+' from
5614 the displacement string. */
5615 if (*intel_parser.disp != '\0')
5616 {
5617 char *s = intel_parser.disp;
5618 s += strlen (s) - 1;
5619 if (*s == '+')
5620 *s = '\0';
5621 }
5622
5623 free (save_str);
5624
5625 return 1;
5626 }
5627 else
5628 return 0;
5629 }
5630
5631 /* The constant was not used for register scaling. Since we have
5632 already consumed the token following `*' we now need to put it
5633 back in the stream. */
5634 else
5635 intel_putback_token ();
5636 }
5637
5638 /* Add the constant to the displacement string. */
5639 strcat (intel_parser.disp, save_str);
5640 free (save_str);
5641
5642 return 1;
5643 }
5644
5645 as_bad (_("Unrecognized token '%s'"), cur_token.str);
5646 return 0;
5647 }
5648
5649 /* Match the given token against cur_token. If they match, read the next
5650 token from the operand string. */
5651 static int
5652 intel_match_token (code)
5653 int code;
5654 {
5655 if (cur_token.code == code)
5656 {
5657 intel_get_token ();
5658 return 1;
5659 }
5660 else
5661 {
5662 as_bad (_("Unexpected token `%s'\n"), cur_token.str);
5663 return 0;
5664 }
5665 }
5666
5667 /* Read a new token from intel_parser.op_string and store it in cur_token. */
5668 static void
5669 intel_get_token ()
5670 {
5671 char *end_op;
5672 const reg_entry *reg;
5673 struct intel_token new_token;
5674
5675 new_token.code = T_NIL;
5676 new_token.reg = NULL;
5677 new_token.str = NULL;
5678
5679 /* Free the memory allocated to the previous token and move
5680 cur_token to prev_token. */
5681 if (prev_token.str)
5682 free (prev_token.str);
5683
5684 prev_token = cur_token;
5685
5686 /* Skip whitespace. */
5687 while (is_space_char (*intel_parser.op_string))
5688 intel_parser.op_string++;
5689
5690 /* Return an empty token if we find nothing else on the line. */
5691 if (*intel_parser.op_string == '\0')
5692 {
5693 cur_token = new_token;
5694 return;
5695 }
5696
5697 /* The new token cannot be larger than the remainder of the operand
5698 string. */
5699 new_token.str = (char *) malloc (strlen (intel_parser.op_string) + 1);
5700 if (new_token.str == NULL)
5701 abort ();
5702 new_token.str[0] = '\0';
5703
5704 if (strchr ("0123456789", *intel_parser.op_string))
5705 {
5706 char *p = new_token.str;
5707 char *q = intel_parser.op_string;
5708 new_token.code = T_CONST;
5709
5710 /* Allow any kind of identifier char to encompass floating point and
5711 hexadecimal numbers. */
5712 while (is_identifier_char (*q))
5713 *p++ = *q++;
5714 *p = '\0';
5715
5716 /* Recognize special symbol names [0-9][bf]. */
5717 if (strlen (intel_parser.op_string) == 2
5718 && (intel_parser.op_string[1] == 'b'
5719 || intel_parser.op_string[1] == 'f'))
5720 new_token.code = T_ID;
5721 }
5722
5723 else if (strchr ("+-/*:[]()", *intel_parser.op_string))
5724 {
5725 new_token.code = *intel_parser.op_string;
5726 new_token.str[0] = *intel_parser.op_string;
5727 new_token.str[1] = '\0';
5728 }
5729
5730 else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
5731 && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
5732 {
5733 new_token.code = T_REG;
5734 new_token.reg = reg;
5735
5736 if (*intel_parser.op_string == REGISTER_PREFIX)
5737 {
5738 new_token.str[0] = REGISTER_PREFIX;
5739 new_token.str[1] = '\0';
5740 }
5741
5742 strcat (new_token.str, reg->reg_name);
5743 }
5744
5745 else if (is_identifier_char (*intel_parser.op_string))
5746 {
5747 char *p = new_token.str;
5748 char *q = intel_parser.op_string;
5749
5750 /* A '.' or '$' followed by an identifier char is an identifier.
5751 Otherwise, it's operator '.' followed by an expression. */
5752 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
5753 {
5754 new_token.code = *q;
5755 new_token.str[0] = *q;
5756 new_token.str[1] = '\0';
5757 }
5758 else
5759 {
5760 while (is_identifier_char (*q) || *q == '@')
5761 *p++ = *q++;
5762 *p = '\0';
5763
5764 if (strcasecmp (new_token.str, "BYTE") == 0)
5765 new_token.code = T_BYTE;
5766
5767 else if (strcasecmp (new_token.str, "WORD") == 0)
5768 new_token.code = T_WORD;
5769
5770 else if (strcasecmp (new_token.str, "DWORD") == 0)
5771 new_token.code = T_DWORD;
5772
5773 else if (strcasecmp (new_token.str, "QWORD") == 0)
5774 new_token.code = T_QWORD;
5775
5776 else if (strcasecmp (new_token.str, "XWORD") == 0)
5777 new_token.code = T_XWORD;
5778
5779 else if (strcasecmp (new_token.str, "PTR") == 0)
5780 new_token.code = T_PTR;
5781
5782 else if (strcasecmp (new_token.str, "SHORT") == 0)
5783 new_token.code = T_SHORT;
5784
5785 else if (strcasecmp (new_token.str, "OFFSET") == 0)
5786 {
5787 new_token.code = T_OFFSET;
5788
5789 /* ??? This is not mentioned in the MASM grammar but gcc
5790 makes use of it with -mintel-syntax. OFFSET may be
5791 followed by FLAT: */
5792 if (strncasecmp (q, " FLAT:", 6) == 0)
5793 strcat (new_token.str, " FLAT:");
5794 }
5795
5796 /* ??? This is not mentioned in the MASM grammar. */
5797 else if (strcasecmp (new_token.str, "FLAT") == 0)
5798 new_token.code = T_OFFSET;
5799
5800 else
5801 new_token.code = T_ID;
5802 }
5803 }
5804
5805 else
5806 as_bad (_("Unrecognized token `%s'\n"), intel_parser.op_string);
5807
5808 intel_parser.op_string += strlen (new_token.str);
5809 cur_token = new_token;
5810 }
5811
5812 /* Put cur_token back into the token stream and make cur_token point to
5813 prev_token. */
5814 static void
5815 intel_putback_token ()
5816 {
5817 intel_parser.op_string -= strlen (cur_token.str);
5818 free (cur_token.str);
5819 cur_token = prev_token;
5820
5821 /* Forget prev_token. */
5822 prev_token.code = T_NIL;
5823 prev_token.reg = NULL;
5824 prev_token.str = NULL;
5825 }
This page took 0.161091 seconds and 4 git commands to generate.