2000-10-25 Diego Novillo <dnovillo@cygnus.com>
[binutils-gdb.git] / gas / config / tc-i386.c
1 /* i386.c -- Assemble code for the Intel 80386
2 Copyright (C) 1989, 91, 92, 93, 94, 95, 96, 97, 98, 99, 2000
3 Free Software Foundation, Inc.
4
5 This file is part of GAS, the GNU Assembler.
6
7 GAS is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 2, or (at your option)
10 any later version.
11
12 GAS is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
16
17 You should have received a copy of the GNU General Public License
18 along with GAS; see the file COPYING. If not, write to the Free
19 Software Foundation, 59 Temple Place - Suite 330, Boston, MA
20 02111-1307, USA. */
21
22 /* Intel 80386 machine specific gas.
23 Written by Eliot Dresselhaus (eliot@mgm.mit.edu).
24 Bugs & suggestions are completely welcome. This is free software.
25 Please help us make it better. */
26
27 #include <ctype.h>
28
29 #include "as.h"
30 #include "subsegs.h"
31 #include "opcode/i386.h"
32
33 #ifndef REGISTER_WARNINGS
34 #define REGISTER_WARNINGS 1
35 #endif
36
37 #ifndef INFER_ADDR_PREFIX
38 #define INFER_ADDR_PREFIX 1
39 #endif
40
41 #ifndef SCALE1_WHEN_NO_INDEX
42 /* Specifying a scale factor besides 1 when there is no index is
43 futile. eg. `mov (%ebx,2),%al' does exactly the same as
44 `mov (%ebx),%al'. To slavishly follow what the programmer
45 specified, set SCALE1_WHEN_NO_INDEX to 0. */
46 #define SCALE1_WHEN_NO_INDEX 1
47 #endif
48
49 #define true 1
50 #define false 0
51
52 static unsigned int mode_from_disp_size PARAMS ((unsigned int));
53 static int fits_in_signed_byte PARAMS ((offsetT));
54 static int fits_in_unsigned_byte PARAMS ((offsetT));
55 static int fits_in_unsigned_word PARAMS ((offsetT));
56 static int fits_in_signed_word PARAMS ((offsetT));
57 static int smallest_imm_type PARAMS ((offsetT));
58 static offsetT offset_in_range PARAMS ((offsetT, int));
59 static int add_prefix PARAMS ((unsigned int));
60 static void set_16bit_code_flag PARAMS ((int));
61 static void set_16bit_gcc_code_flag PARAMS ((int));
62 static void set_intel_syntax PARAMS ((int));
63 static void set_cpu_arch PARAMS ((int));
64
65 #ifdef BFD_ASSEMBLER
66 static bfd_reloc_code_real_type reloc
67 PARAMS ((int, int, bfd_reloc_code_real_type));
68 #endif
69
70 /* 'md_assemble ()' gathers together information and puts it into a
71 i386_insn. */
72
73 union i386_op
74 {
75 expressionS *disps;
76 expressionS *imms;
77 const reg_entry *regs;
78 };
79
80 struct _i386_insn
81 {
82 /* TM holds the template for the insn were currently assembling. */
83 template tm;
84
85 /* SUFFIX holds the instruction mnemonic suffix if given.
86 (e.g. 'l' for 'movl') */
87 char suffix;
88
89 /* OPERANDS gives the number of given operands. */
90 unsigned int operands;
91
92 /* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number
93 of given register, displacement, memory operands and immediate
94 operands. */
95 unsigned int reg_operands, disp_operands, mem_operands, imm_operands;
96
97 /* TYPES [i] is the type (see above #defines) which tells us how to
98 use OP[i] for the corresponding operand. */
99 unsigned int types[MAX_OPERANDS];
100
101 /* Displacement expression, immediate expression, or register for each
102 operand. */
103 union i386_op op[MAX_OPERANDS];
104
105 /* Relocation type for operand */
106 #ifdef BFD_ASSEMBLER
107 enum bfd_reloc_code_real disp_reloc[MAX_OPERANDS];
108 #else
109 int disp_reloc[MAX_OPERANDS];
110 #endif
111
112 /* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode
113 the base index byte below. */
114 const reg_entry *base_reg;
115 const reg_entry *index_reg;
116 unsigned int log2_scale_factor;
117
118 /* SEG gives the seg_entries of this insn. They are zero unless
119 explicit segment overrides are given. */
120 const seg_entry *seg[2];
121
122 /* PREFIX holds all the given prefix opcodes (usually null).
123 PREFIXES is the number of prefix opcodes. */
124 unsigned int prefixes;
125 unsigned char prefix[MAX_PREFIXES];
126
127 /* RM and SIB are the modrm byte and the sib byte where the
128 addressing modes of this insn are encoded. */
129
130 modrm_byte rm;
131 sib_byte sib;
132 };
133
134 typedef struct _i386_insn i386_insn;
135
136 /* List of chars besides those in app.c:symbol_chars that can start an
137 operand. Used to prevent the scrubber eating vital white-space. */
138 #ifdef LEX_AT
139 const char extra_symbol_chars[] = "*%-(@";
140 #else
141 const char extra_symbol_chars[] = "*%-(";
142 #endif
143
144 /* This array holds the chars that always start a comment. If the
145 pre-processor is disabled, these aren't very useful. */
146 #if defined (TE_I386AIX) || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) && ! defined (TE_LINUX) && !defined(TE_FreeBSD))
147 /* Putting '/' here makes it impossible to use the divide operator.
148 However, we need it for compatibility with SVR4 systems. */
149 const char comment_chars[] = "#/";
150 #define PREFIX_SEPARATOR '\\'
151 #else
152 const char comment_chars[] = "#";
153 #define PREFIX_SEPARATOR '/'
154 #endif
155
156 /* This array holds the chars that only start a comment at the beginning of
157 a line. If the line seems to have the form '# 123 filename'
158 .line and .file directives will appear in the pre-processed output.
159 Note that input_file.c hand checks for '#' at the beginning of the
160 first line of the input file. This is because the compiler outputs
161 #NO_APP at the beginning of its output.
162 Also note that comments started like this one will always work if
163 '/' isn't otherwise defined. */
164 #if defined (TE_I386AIX) || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) && ! defined (TE_LINUX) && !defined(TE_FreeBSD))
165 const char line_comment_chars[] = "";
166 #else
167 const char line_comment_chars[] = "/";
168 #endif
169
170 const char line_separator_chars[] = ";";
171
172 /* Chars that can be used to separate mant from exp in floating point
173 nums. */
174 const char EXP_CHARS[] = "eE";
175
176 /* Chars that mean this number is a floating point constant
177 As in 0f12.456
178 or 0d1.2345e12. */
179 const char FLT_CHARS[] = "fFdDxX";
180
181 /* Tables for lexical analysis. */
182 static char mnemonic_chars[256];
183 static char register_chars[256];
184 static char operand_chars[256];
185 static char identifier_chars[256];
186 static char digit_chars[256];
187
188 /* Lexical macros. */
189 #define is_mnemonic_char(x) (mnemonic_chars[(unsigned char) x])
190 #define is_operand_char(x) (operand_chars[(unsigned char) x])
191 #define is_register_char(x) (register_chars[(unsigned char) x])
192 #define is_space_char(x) ((x) == ' ')
193 #define is_identifier_char(x) (identifier_chars[(unsigned char) x])
194 #define is_digit_char(x) (digit_chars[(unsigned char) x])
195
196 /* All non-digit non-letter charcters that may occur in an operand. */
197 static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
198
199 /* md_assemble() always leaves the strings it's passed unaltered. To
200 effect this we maintain a stack of saved characters that we've smashed
201 with '\0's (indicating end of strings for various sub-fields of the
202 assembler instruction). */
203 static char save_stack[32];
204 static char *save_stack_p;
205 #define END_STRING_AND_SAVE(s) \
206 do { *save_stack_p++ = *(s); *(s) = '\0'; } while (0)
207 #define RESTORE_END_STRING(s) \
208 do { *(s) = *--save_stack_p; } while (0)
209
210 /* The instruction we're assembling. */
211 static i386_insn i;
212
213 /* Possible templates for current insn. */
214 static const templates *current_templates;
215
216 /* Per instruction expressionS buffers: 2 displacements & 2 immediate max. */
217 static expressionS disp_expressions[2], im_expressions[2];
218
219 /* Current operand we are working on. */
220 static int this_operand;
221
222 /* 1 if we're writing 16-bit code,
223 0 if 32-bit. */
224 static int flag_16bit_code;
225
226 /* 1 for intel syntax,
227 0 if att syntax. */
228 static int intel_syntax = 0;
229
230 /* 1 if register prefix % not required. */
231 static int allow_naked_reg = 0;
232
233 /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
234 leave, push, and pop instructions so that gcc has the same stack
235 frame as in 32 bit mode. */
236 static char stackop_size = '\0';
237
238 /* Non-zero to quieten some warnings. */
239 static int quiet_warnings = 0;
240
241 /* CPU name. */
242 static const char *cpu_arch_name = NULL;
243
244 /* CPU feature flags. */
245 static unsigned int cpu_arch_flags = 0;
246
247 /* Interface to relax_segment.
248 There are 2 relax states for 386 jump insns: one for conditional &
249 one for unconditional jumps. This is because these two types of
250 jumps add different sizes to frags when we're figuring out what
251 sort of jump to choose to reach a given label. */
252
253 /* Types. */
254 #define COND_JUMP 1
255 #define UNCOND_JUMP 2
256 /* Sizes. */
257 #define CODE16 1
258 #define SMALL 0
259 #define SMALL16 (SMALL|CODE16)
260 #define BIG 2
261 #define BIG16 (BIG|CODE16)
262
263 #ifndef INLINE
264 #ifdef __GNUC__
265 #define INLINE __inline__
266 #else
267 #define INLINE
268 #endif
269 #endif
270
271 #define ENCODE_RELAX_STATE(type,size) \
272 ((relax_substateT)((type<<2) | (size)))
273 #define SIZE_FROM_RELAX_STATE(s) \
274 ( (((s) & 0x3) == BIG ? 4 : (((s) & 0x3) == BIG16 ? 2 : 1)) )
275
276 /* This table is used by relax_frag to promote short jumps to long
277 ones where necessary. SMALL (short) jumps may be promoted to BIG
278 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
279 don't allow a short jump in a 32 bit code segment to be promoted to
280 a 16 bit offset jump because it's slower (requires data size
281 prefix), and doesn't work, unless the destination is in the bottom
282 64k of the code segment (The top 16 bits of eip are zeroed). */
283
284 const relax_typeS md_relax_table[] =
285 {
286 /* The fields are:
287 1) most positive reach of this state,
288 2) most negative reach of this state,
289 3) how many bytes this mode will add to the size of the current frag
290 4) which index into the table to try if we can't fit into this one. */
291 {1, 1, 0, 0},
292 {1, 1, 0, 0},
293 {1, 1, 0, 0},
294 {1, 1, 0, 0},
295
296 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
297 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
298 /* dword conditionals adds 4 bytes to frag:
299 1 extra opcode byte, 3 extra displacement bytes. */
300 {0, 0, 4, 0},
301 /* word conditionals add 2 bytes to frag:
302 1 extra opcode byte, 1 extra displacement byte. */
303 {0, 0, 2, 0},
304
305 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
306 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
307 /* dword jmp adds 3 bytes to frag:
308 0 extra opcode bytes, 3 extra displacement bytes. */
309 {0, 0, 3, 0},
310 /* word jmp adds 1 byte to frag:
311 0 extra opcode bytes, 1 extra displacement byte. */
312 {0, 0, 1, 0}
313
314 };
315
316 static const arch_entry cpu_arch[] = {
317 {"i8086", Cpu086 },
318 {"i186", Cpu086|Cpu186 },
319 {"i286", Cpu086|Cpu186|Cpu286 },
320 {"i386", Cpu086|Cpu186|Cpu286|Cpu386 },
321 {"i486", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486 },
322 {"i586", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
323 {"i686", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
324 {"pentium", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
325 {"pentiumpro",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
326 {"k6", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX|Cpu3dnow },
327 {"athlon", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|Cpu3dnow },
328 {NULL, 0 }
329 };
330
331 void
332 i386_align_code (fragP, count)
333 fragS *fragP;
334 int count;
335 {
336 /* Various efficient no-op patterns for aligning code labels.
337 Note: Don't try to assemble the instructions in the comments.
338 0L and 0w are not legal. */
339 static const char f32_1[] =
340 {0x90}; /* nop */
341 static const char f32_2[] =
342 {0x89,0xf6}; /* movl %esi,%esi */
343 static const char f32_3[] =
344 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
345 static const char f32_4[] =
346 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
347 static const char f32_5[] =
348 {0x90, /* nop */
349 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
350 static const char f32_6[] =
351 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
352 static const char f32_7[] =
353 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
354 static const char f32_8[] =
355 {0x90, /* nop */
356 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
357 static const char f32_9[] =
358 {0x89,0xf6, /* movl %esi,%esi */
359 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
360 static const char f32_10[] =
361 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
362 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
363 static const char f32_11[] =
364 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
365 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
366 static const char f32_12[] =
367 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
368 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
369 static const char f32_13[] =
370 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
371 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
372 static const char f32_14[] =
373 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
374 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
375 static const char f32_15[] =
376 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
377 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
378 static const char f16_3[] =
379 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
380 static const char f16_4[] =
381 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
382 static const char f16_5[] =
383 {0x90, /* nop */
384 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
385 static const char f16_6[] =
386 {0x89,0xf6, /* mov %si,%si */
387 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
388 static const char f16_7[] =
389 {0x8d,0x74,0x00, /* lea 0(%si),%si */
390 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
391 static const char f16_8[] =
392 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
393 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
394 static const char *const f32_patt[] = {
395 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
396 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
397 };
398 static const char *const f16_patt[] = {
399 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
400 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
401 };
402
403 if (count > 0 && count <= 15)
404 {
405 if (flag_16bit_code)
406 {
407 memcpy (fragP->fr_literal + fragP->fr_fix,
408 f16_patt[count - 1], count);
409 if (count > 8)
410 /* Adjust jump offset. */
411 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
412 }
413 else
414 memcpy (fragP->fr_literal + fragP->fr_fix,
415 f32_patt[count - 1], count);
416 fragP->fr_var = count;
417 }
418 }
419
420 static char *output_invalid PARAMS ((int c));
421 static int i386_operand PARAMS ((char *operand_string));
422 static int i386_intel_operand PARAMS ((char *operand_string, int got_a_float));
423 static const reg_entry *parse_register PARAMS ((char *reg_string,
424 char **end_op));
425
426 #ifndef I386COFF
427 static void s_bss PARAMS ((int));
428 #endif
429
430 symbolS *GOT_symbol; /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
431
432 static INLINE unsigned int
433 mode_from_disp_size (t)
434 unsigned int t;
435 {
436 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32)) ? 2 : 0;
437 }
438
439 static INLINE int
440 fits_in_signed_byte (num)
441 offsetT num;
442 {
443 return (num >= -128) && (num <= 127);
444 }
445
446 static INLINE int
447 fits_in_unsigned_byte (num)
448 offsetT num;
449 {
450 return (num & 0xff) == num;
451 }
452
453 static INLINE int
454 fits_in_unsigned_word (num)
455 offsetT num;
456 {
457 return (num & 0xffff) == num;
458 }
459
460 static INLINE int
461 fits_in_signed_word (num)
462 offsetT num;
463 {
464 return (-32768 <= num) && (num <= 32767);
465 }
466
467 static int
468 smallest_imm_type (num)
469 offsetT num;
470 {
471 if (cpu_arch_flags != 0
472 && cpu_arch_flags != (Cpu086 | Cpu186 | Cpu286 | Cpu386 | Cpu486))
473 {
474 /* This code is disabled on the 486 because all the Imm1 forms
475 in the opcode table are slower on the i486. They're the
476 versions with the implicitly specified single-position
477 displacement, which has another syntax if you really want to
478 use that form. */
479 if (num == 1)
480 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32;
481 }
482 return (fits_in_signed_byte (num)
483 ? (Imm8S | Imm8 | Imm16 | Imm32)
484 : fits_in_unsigned_byte (num)
485 ? (Imm8 | Imm16 | Imm32)
486 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
487 ? (Imm16 | Imm32)
488 : (Imm32));
489 }
490
491 static offsetT
492 offset_in_range (val, size)
493 offsetT val;
494 int size;
495 {
496 addressT mask;
497
498 switch (size)
499 {
500 case 1: mask = ((addressT) 1 << 8) - 1; break;
501 case 2: mask = ((addressT) 1 << 16) - 1; break;
502 case 4: mask = ((addressT) 2 << 31) - 1; break;
503 default: abort ();
504 }
505
506 /* If BFD64, sign extend val. */
507 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
508 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
509
510 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
511 {
512 char buf1[40], buf2[40];
513
514 sprint_value (buf1, val);
515 sprint_value (buf2, val & mask);
516 as_warn (_("%s shortened to %s"), buf1, buf2);
517 }
518 return val & mask;
519 }
520
521 /* Returns 0 if attempting to add a prefix where one from the same
522 class already exists, 1 if non rep/repne added, 2 if rep/repne
523 added. */
524 static int
525 add_prefix (prefix)
526 unsigned int prefix;
527 {
528 int ret = 1;
529 int q;
530
531 switch (prefix)
532 {
533 default:
534 abort ();
535
536 case CS_PREFIX_OPCODE:
537 case DS_PREFIX_OPCODE:
538 case ES_PREFIX_OPCODE:
539 case FS_PREFIX_OPCODE:
540 case GS_PREFIX_OPCODE:
541 case SS_PREFIX_OPCODE:
542 q = SEG_PREFIX;
543 break;
544
545 case REPNE_PREFIX_OPCODE:
546 case REPE_PREFIX_OPCODE:
547 ret = 2;
548 /* fall thru */
549 case LOCK_PREFIX_OPCODE:
550 q = LOCKREP_PREFIX;
551 break;
552
553 case FWAIT_OPCODE:
554 q = WAIT_PREFIX;
555 break;
556
557 case ADDR_PREFIX_OPCODE:
558 q = ADDR_PREFIX;
559 break;
560
561 case DATA_PREFIX_OPCODE:
562 q = DATA_PREFIX;
563 break;
564 }
565
566 if (i.prefix[q])
567 {
568 as_bad (_("same type of prefix used twice"));
569 return 0;
570 }
571
572 i.prefixes += 1;
573 i.prefix[q] = prefix;
574 return ret;
575 }
576
577 static void
578 set_16bit_code_flag (new_16bit_code_flag)
579 int new_16bit_code_flag;
580 {
581 flag_16bit_code = new_16bit_code_flag;
582 stackop_size = '\0';
583 }
584
585 static void
586 set_16bit_gcc_code_flag (new_16bit_code_flag)
587 int new_16bit_code_flag;
588 {
589 flag_16bit_code = new_16bit_code_flag;
590 stackop_size = new_16bit_code_flag ? 'l' : '\0';
591 }
592
593 static void
594 set_intel_syntax (syntax_flag)
595 int syntax_flag;
596 {
597 /* Find out if register prefixing is specified. */
598 int ask_naked_reg = 0;
599
600 SKIP_WHITESPACE ();
601 if (! is_end_of_line[(unsigned char) *input_line_pointer])
602 {
603 char *string = input_line_pointer;
604 int e = get_symbol_end ();
605
606 if (strcmp (string, "prefix") == 0)
607 ask_naked_reg = 1;
608 else if (strcmp (string, "noprefix") == 0)
609 ask_naked_reg = -1;
610 else
611 as_bad (_("bad argument to syntax directive."));
612 *input_line_pointer = e;
613 }
614 demand_empty_rest_of_line ();
615
616 intel_syntax = syntax_flag;
617
618 if (ask_naked_reg == 0)
619 {
620 #ifdef BFD_ASSEMBLER
621 allow_naked_reg = (intel_syntax
622 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
623 #else
624 /* Conservative default. */
625 allow_naked_reg = 0;
626 #endif
627 }
628 else
629 allow_naked_reg = (ask_naked_reg < 0);
630 }
631
632 static void
633 set_cpu_arch (dummy)
634 int dummy ATTRIBUTE_UNUSED;
635 {
636 SKIP_WHITESPACE ();
637
638 if (! is_end_of_line[(unsigned char) *input_line_pointer])
639 {
640 char *string = input_line_pointer;
641 int e = get_symbol_end ();
642 int i;
643
644 for (i = 0; cpu_arch[i].name; i++)
645 {
646 if (strcmp (string, cpu_arch[i].name) == 0)
647 {
648 cpu_arch_name = cpu_arch[i].name;
649 cpu_arch_flags = cpu_arch[i].flags;
650 break;
651 }
652 }
653 if (!cpu_arch[i].name)
654 as_bad (_("no such architecture: `%s'"), string);
655
656 *input_line_pointer = e;
657 }
658 else
659 as_bad (_("missing cpu architecture"));
660
661 demand_empty_rest_of_line ();
662 }
663
664 const pseudo_typeS md_pseudo_table[] =
665 {
666 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
667 {"align", s_align_bytes, 0},
668 #else
669 {"align", s_align_ptwo, 0},
670 #endif
671 {"arch", set_cpu_arch, 0},
672 #ifndef I386COFF
673 {"bss", s_bss, 0},
674 #endif
675 {"ffloat", float_cons, 'f'},
676 {"dfloat", float_cons, 'd'},
677 {"tfloat", float_cons, 'x'},
678 {"value", cons, 2},
679 {"noopt", s_ignore, 0},
680 {"optim", s_ignore, 0},
681 {"code16gcc", set_16bit_gcc_code_flag, 1},
682 {"code16", set_16bit_code_flag, 1},
683 {"code32", set_16bit_code_flag, 0},
684 {"intel_syntax", set_intel_syntax, 1},
685 {"att_syntax", set_intel_syntax, 0},
686 {0, 0, 0}
687 };
688
689 /* For interface with expression (). */
690 extern char *input_line_pointer;
691
692 /* Hash table for instruction mnemonic lookup. */
693 static struct hash_control *op_hash;
694
695 /* Hash table for register lookup. */
696 static struct hash_control *reg_hash;
697 \f
698 void
699 md_begin ()
700 {
701 const char *hash_err;
702
703 /* Initialize op_hash hash table. */
704 op_hash = hash_new ();
705
706 {
707 register const template *optab;
708 register templates *core_optab;
709
710 /* Setup for loop. */
711 optab = i386_optab;
712 core_optab = (templates *) xmalloc (sizeof (templates));
713 core_optab->start = optab;
714
715 while (1)
716 {
717 ++optab;
718 if (optab->name == NULL
719 || strcmp (optab->name, (optab - 1)->name) != 0)
720 {
721 /* different name --> ship out current template list;
722 add to hash table; & begin anew. */
723 core_optab->end = optab;
724 hash_err = hash_insert (op_hash,
725 (optab - 1)->name,
726 (PTR) core_optab);
727 if (hash_err)
728 {
729 hash_error:
730 as_fatal (_("Internal Error: Can't hash %s: %s"),
731 (optab - 1)->name,
732 hash_err);
733 }
734 if (optab->name == NULL)
735 break;
736 core_optab = (templates *) xmalloc (sizeof (templates));
737 core_optab->start = optab;
738 }
739 }
740 }
741
742 /* Initialize reg_hash hash table. */
743 reg_hash = hash_new ();
744 {
745 register const reg_entry *regtab;
746
747 for (regtab = i386_regtab;
748 regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
749 regtab++)
750 {
751 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
752 if (hash_err)
753 goto hash_error;
754 }
755 }
756
757 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
758 {
759 register int c;
760 register char *p;
761
762 for (c = 0; c < 256; c++)
763 {
764 if (isdigit (c))
765 {
766 digit_chars[c] = c;
767 mnemonic_chars[c] = c;
768 register_chars[c] = c;
769 operand_chars[c] = c;
770 }
771 else if (islower (c))
772 {
773 mnemonic_chars[c] = c;
774 register_chars[c] = c;
775 operand_chars[c] = c;
776 }
777 else if (isupper (c))
778 {
779 mnemonic_chars[c] = tolower (c);
780 register_chars[c] = mnemonic_chars[c];
781 operand_chars[c] = c;
782 }
783
784 if (isalpha (c) || isdigit (c))
785 identifier_chars[c] = c;
786 else if (c >= 128)
787 {
788 identifier_chars[c] = c;
789 operand_chars[c] = c;
790 }
791 }
792
793 #ifdef LEX_AT
794 identifier_chars['@'] = '@';
795 #endif
796 digit_chars['-'] = '-';
797 identifier_chars['_'] = '_';
798 identifier_chars['.'] = '.';
799
800 for (p = operand_special_chars; *p != '\0'; p++)
801 operand_chars[(unsigned char) *p] = *p;
802 }
803
804 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
805 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
806 {
807 record_alignment (text_section, 2);
808 record_alignment (data_section, 2);
809 record_alignment (bss_section, 2);
810 }
811 #endif
812 }
813
814 void
815 i386_print_statistics (file)
816 FILE *file;
817 {
818 hash_print_statistics (file, "i386 opcode", op_hash);
819 hash_print_statistics (file, "i386 register", reg_hash);
820 }
821 \f
822 #ifdef DEBUG386
823
824 /* Debugging routines for md_assemble. */
825 static void pi PARAMS ((char *, i386_insn *));
826 static void pte PARAMS ((template *));
827 static void pt PARAMS ((unsigned int));
828 static void pe PARAMS ((expressionS *));
829 static void ps PARAMS ((symbolS *));
830
831 static void
832 pi (line, x)
833 char *line;
834 i386_insn *x;
835 {
836 register template *p;
837 int i;
838
839 fprintf (stdout, "%s: template ", line);
840 pte (&x->tm);
841 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x",
842 x->rm.mode, x->rm.reg, x->rm.regmem);
843 fprintf (stdout, " base %x index %x scale %x\n",
844 x->bi.base, x->bi.index, x->bi.scale);
845 for (i = 0; i < x->operands; i++)
846 {
847 fprintf (stdout, " #%d: ", i + 1);
848 pt (x->types[i]);
849 fprintf (stdout, "\n");
850 if (x->types[i]
851 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
852 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
853 if (x->types[i] & Imm)
854 pe (x->op[i].imms);
855 if (x->types[i] & Disp)
856 pe (x->op[i].disps);
857 }
858 }
859
860 static void
861 pte (t)
862 template *t;
863 {
864 int i;
865 fprintf (stdout, " %d operands ", t->operands);
866 fprintf (stdout, "opcode %x ", t->base_opcode);
867 if (t->extension_opcode != None)
868 fprintf (stdout, "ext %x ", t->extension_opcode);
869 if (t->opcode_modifier & D)
870 fprintf (stdout, "D");
871 if (t->opcode_modifier & W)
872 fprintf (stdout, "W");
873 fprintf (stdout, "\n");
874 for (i = 0; i < t->operands; i++)
875 {
876 fprintf (stdout, " #%d type ", i + 1);
877 pt (t->operand_types[i]);
878 fprintf (stdout, "\n");
879 }
880 }
881
882 static void
883 pe (e)
884 expressionS *e;
885 {
886 fprintf (stdout, " operation %d\n", e->X_op);
887 fprintf (stdout, " add_number %ld (%lx)\n",
888 (long) e->X_add_number, (long) e->X_add_number);
889 if (e->X_add_symbol)
890 {
891 fprintf (stdout, " add_symbol ");
892 ps (e->X_add_symbol);
893 fprintf (stdout, "\n");
894 }
895 if (e->X_op_symbol)
896 {
897 fprintf (stdout, " op_symbol ");
898 ps (e->X_op_symbol);
899 fprintf (stdout, "\n");
900 }
901 }
902
903 static void
904 ps (s)
905 symbolS *s;
906 {
907 fprintf (stdout, "%s type %s%s",
908 S_GET_NAME (s),
909 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
910 segment_name (S_GET_SEGMENT (s)));
911 }
912
913 struct type_name
914 {
915 unsigned int mask;
916 char *tname;
917 }
918
919 type_names[] =
920 {
921 { Reg8, "r8" },
922 { Reg16, "r16" },
923 { Reg32, "r32" },
924 { Imm8, "i8" },
925 { Imm8S, "i8s" },
926 { Imm16, "i16" },
927 { Imm32, "i32" },
928 { Imm1, "i1" },
929 { BaseIndex, "BaseIndex" },
930 { Disp8, "d8" },
931 { Disp16, "d16" },
932 { Disp32, "d32" },
933 { InOutPortReg, "InOutPortReg" },
934 { ShiftCount, "ShiftCount" },
935 { Control, "control reg" },
936 { Test, "test reg" },
937 { Debug, "debug reg" },
938 { FloatReg, "FReg" },
939 { FloatAcc, "FAcc" },
940 { SReg2, "SReg2" },
941 { SReg3, "SReg3" },
942 { Acc, "Acc" },
943 { JumpAbsolute, "Jump Absolute" },
944 { RegMMX, "rMMX" },
945 { RegXMM, "rXMM" },
946 { EsSeg, "es" },
947 { 0, "" }
948 };
949
950 static void
951 pt (t)
952 unsigned int t;
953 {
954 register struct type_name *ty;
955
956 if (t == Unknown)
957 {
958 fprintf (stdout, _("Unknown"));
959 }
960 else
961 {
962 for (ty = type_names; ty->mask; ty++)
963 if (t & ty->mask)
964 fprintf (stdout, "%s, ", ty->tname);
965 }
966 fflush (stdout);
967 }
968
969 #endif /* DEBUG386 */
970 \f
971 int
972 tc_i386_force_relocation (fixp)
973 struct fix *fixp;
974 {
975 #ifdef BFD_ASSEMBLER
976 if (fixp->fx_r_type == BFD_RELOC_VTABLE_INHERIT
977 || fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
978 return 1;
979 return 0;
980 #else
981 /* For COFF. */
982 return fixp->fx_r_type == 7;
983 #endif
984 }
985
986 #ifdef BFD_ASSEMBLER
987 static bfd_reloc_code_real_type reloc
988 PARAMS ((int, int, bfd_reloc_code_real_type));
989
990 static bfd_reloc_code_real_type
991 reloc (size, pcrel, other)
992 int size;
993 int pcrel;
994 bfd_reloc_code_real_type other;
995 {
996 if (other != NO_RELOC)
997 return other;
998
999 if (pcrel)
1000 {
1001 switch (size)
1002 {
1003 case 1: return BFD_RELOC_8_PCREL;
1004 case 2: return BFD_RELOC_16_PCREL;
1005 case 4: return BFD_RELOC_32_PCREL;
1006 }
1007 as_bad (_("can not do %d byte pc-relative relocation"), size);
1008 }
1009 else
1010 {
1011 switch (size)
1012 {
1013 case 1: return BFD_RELOC_8;
1014 case 2: return BFD_RELOC_16;
1015 case 4: return BFD_RELOC_32;
1016 }
1017 as_bad (_("can not do %d byte relocation"), size);
1018 }
1019
1020 return BFD_RELOC_NONE;
1021 }
1022
1023 /* Here we decide which fixups can be adjusted to make them relative to
1024 the beginning of the section instead of the symbol. Basically we need
1025 to make sure that the dynamic relocations are done correctly, so in
1026 some cases we force the original symbol to be used. */
1027
1028 int
1029 tc_i386_fix_adjustable (fixP)
1030 fixS *fixP;
1031 {
1032 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1033 /* Prevent all adjustments to global symbols, or else dynamic
1034 linking will not work correctly. */
1035 if (S_IS_EXTERNAL (fixP->fx_addsy)
1036 || S_IS_WEAK (fixP->fx_addsy))
1037 return 0;
1038 #endif
1039 /* adjust_reloc_syms doesn't know about the GOT. */
1040 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1041 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1042 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1043 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1044 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1045 return 0;
1046 return 1;
1047 }
1048 #else
1049 #define reloc(SIZE,PCREL,OTHER) 0
1050 #define BFD_RELOC_16 0
1051 #define BFD_RELOC_32 0
1052 #define BFD_RELOC_16_PCREL 0
1053 #define BFD_RELOC_32_PCREL 0
1054 #define BFD_RELOC_386_PLT32 0
1055 #define BFD_RELOC_386_GOT32 0
1056 #define BFD_RELOC_386_GOTOFF 0
1057 #endif
1058
1059 static int intel_float_operand PARAMS ((char *mnemonic));
1060
1061 static int
1062 intel_float_operand (mnemonic)
1063 char *mnemonic;
1064 {
1065 if (mnemonic[0] == 'f' && mnemonic[1] == 'i')
1066 return 2;
1067
1068 if (mnemonic[0] == 'f')
1069 return 1;
1070
1071 return 0;
1072 }
1073
1074 /* This is the guts of the machine-dependent assembler. LINE points to a
1075 machine dependent instruction. This function is supposed to emit
1076 the frags/bytes it assembles to. */
1077
1078 void
1079 md_assemble (line)
1080 char *line;
1081 {
1082 /* Points to template once we've found it. */
1083 const template *t;
1084
1085 /* Count the size of the instruction generated. */
1086 int insn_size = 0;
1087
1088 int j;
1089
1090 char mnemonic[MAX_MNEM_SIZE];
1091
1092 /* Initialize globals. */
1093 memset (&i, '\0', sizeof (i));
1094 for (j = 0; j < MAX_OPERANDS; j++)
1095 i.disp_reloc[j] = NO_RELOC;
1096 memset (disp_expressions, '\0', sizeof (disp_expressions));
1097 memset (im_expressions, '\0', sizeof (im_expressions));
1098 save_stack_p = save_stack;
1099
1100 /* First parse an instruction mnemonic & call i386_operand for the operands.
1101 We assume that the scrubber has arranged it so that line[0] is the valid
1102 start of a (possibly prefixed) mnemonic. */
1103 {
1104 char *l = line;
1105 char *token_start = l;
1106 char *mnem_p;
1107
1108 /* Non-zero if we found a prefix only acceptable with string insns. */
1109 const char *expecting_string_instruction = NULL;
1110
1111 while (1)
1112 {
1113 mnem_p = mnemonic;
1114 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1115 {
1116 mnem_p++;
1117 if (mnem_p >= mnemonic + sizeof (mnemonic))
1118 {
1119 as_bad (_("no such instruction: `%s'"), token_start);
1120 return;
1121 }
1122 l++;
1123 }
1124 if (!is_space_char (*l)
1125 && *l != END_OF_INSN
1126 && *l != PREFIX_SEPARATOR)
1127 {
1128 as_bad (_("invalid character %s in mnemonic"),
1129 output_invalid (*l));
1130 return;
1131 }
1132 if (token_start == l)
1133 {
1134 if (*l == PREFIX_SEPARATOR)
1135 as_bad (_("expecting prefix; got nothing"));
1136 else
1137 as_bad (_("expecting mnemonic; got nothing"));
1138 return;
1139 }
1140
1141 /* Look up instruction (or prefix) via hash table. */
1142 current_templates = hash_find (op_hash, mnemonic);
1143
1144 if (*l != END_OF_INSN
1145 && (! is_space_char (*l) || l[1] != END_OF_INSN)
1146 && current_templates
1147 && (current_templates->start->opcode_modifier & IsPrefix))
1148 {
1149 /* If we are in 16-bit mode, do not allow addr16 or data16.
1150 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1151 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1152 && (((current_templates->start->opcode_modifier & Size32) != 0)
1153 ^ flag_16bit_code))
1154 {
1155 as_bad (_("redundant %s prefix"),
1156 current_templates->start->name);
1157 return;
1158 }
1159 /* Add prefix, checking for repeated prefixes. */
1160 switch (add_prefix (current_templates->start->base_opcode))
1161 {
1162 case 0:
1163 return;
1164 case 2:
1165 expecting_string_instruction = current_templates->start->name;
1166 break;
1167 }
1168 /* Skip past PREFIX_SEPARATOR and reset token_start. */
1169 token_start = ++l;
1170 }
1171 else
1172 break;
1173 }
1174
1175 if (!current_templates)
1176 {
1177 /* See if we can get a match by trimming off a suffix. */
1178 switch (mnem_p[-1])
1179 {
1180 case WORD_MNEM_SUFFIX:
1181 case BYTE_MNEM_SUFFIX:
1182 case SHORT_MNEM_SUFFIX:
1183 case LONG_MNEM_SUFFIX:
1184 i.suffix = mnem_p[-1];
1185 mnem_p[-1] = '\0';
1186 current_templates = hash_find (op_hash, mnemonic);
1187 break;
1188
1189 /* Intel Syntax. */
1190 case DWORD_MNEM_SUFFIX:
1191 if (intel_syntax)
1192 {
1193 i.suffix = mnem_p[-1];
1194 mnem_p[-1] = '\0';
1195 current_templates = hash_find (op_hash, mnemonic);
1196 break;
1197 }
1198 }
1199 if (!current_templates)
1200 {
1201 as_bad (_("no such instruction: `%s'"), token_start);
1202 return;
1203 }
1204 }
1205
1206 /* Check if instruction is supported on specified architecture. */
1207 if (cpu_arch_flags != 0)
1208 {
1209 if (current_templates->start->cpu_flags & ~cpu_arch_flags)
1210 {
1211 as_warn (_("`%s' is not supported on `%s'"),
1212 current_templates->start->name, cpu_arch_name);
1213 }
1214 else if ((Cpu386 & ~cpu_arch_flags) && !flag_16bit_code)
1215 {
1216 as_warn (_("use .code16 to ensure correct addressing mode"));
1217 }
1218 }
1219
1220 /* Check for rep/repne without a string instruction. */
1221 if (expecting_string_instruction
1222 && !(current_templates->start->opcode_modifier & IsString))
1223 {
1224 as_bad (_("expecting string instruction after `%s'"),
1225 expecting_string_instruction);
1226 return;
1227 }
1228
1229 /* There may be operands to parse. */
1230 if (*l != END_OF_INSN)
1231 {
1232 /* 1 if operand is pending after ','. */
1233 unsigned int expecting_operand = 0;
1234
1235 /* Non-zero if operand parens not balanced. */
1236 unsigned int paren_not_balanced;
1237
1238 do
1239 {
1240 /* Skip optional white space before operand. */
1241 if (is_space_char (*l))
1242 ++l;
1243 if (!is_operand_char (*l) && *l != END_OF_INSN)
1244 {
1245 as_bad (_("invalid character %s before operand %d"),
1246 output_invalid (*l),
1247 i.operands + 1);
1248 return;
1249 }
1250 token_start = l; /* after white space */
1251 paren_not_balanced = 0;
1252 while (paren_not_balanced || *l != ',')
1253 {
1254 if (*l == END_OF_INSN)
1255 {
1256 if (paren_not_balanced)
1257 {
1258 if (!intel_syntax)
1259 as_bad (_("unbalanced parenthesis in operand %d."),
1260 i.operands + 1);
1261 else
1262 as_bad (_("unbalanced brackets in operand %d."),
1263 i.operands + 1);
1264 return;
1265 }
1266 else
1267 break; /* we are done */
1268 }
1269 else if (!is_operand_char (*l) && !is_space_char (*l))
1270 {
1271 as_bad (_("invalid character %s in operand %d"),
1272 output_invalid (*l),
1273 i.operands + 1);
1274 return;
1275 }
1276 if (!intel_syntax)
1277 {
1278 if (*l == '(')
1279 ++paren_not_balanced;
1280 if (*l == ')')
1281 --paren_not_balanced;
1282 }
1283 else
1284 {
1285 if (*l == '[')
1286 ++paren_not_balanced;
1287 if (*l == ']')
1288 --paren_not_balanced;
1289 }
1290 l++;
1291 }
1292 if (l != token_start)
1293 { /* Yes, we've read in another operand. */
1294 unsigned int operand_ok;
1295 this_operand = i.operands++;
1296 if (i.operands > MAX_OPERANDS)
1297 {
1298 as_bad (_("spurious operands; (%d operands/instruction max)"),
1299 MAX_OPERANDS);
1300 return;
1301 }
1302 /* Now parse operand adding info to 'i' as we go along. */
1303 END_STRING_AND_SAVE (l);
1304
1305 if (intel_syntax)
1306 operand_ok =
1307 i386_intel_operand (token_start,
1308 intel_float_operand (mnemonic));
1309 else
1310 operand_ok = i386_operand (token_start);
1311
1312 RESTORE_END_STRING (l);
1313 if (!operand_ok)
1314 return;
1315 }
1316 else
1317 {
1318 if (expecting_operand)
1319 {
1320 expecting_operand_after_comma:
1321 as_bad (_("expecting operand after ','; got nothing"));
1322 return;
1323 }
1324 if (*l == ',')
1325 {
1326 as_bad (_("expecting operand before ','; got nothing"));
1327 return;
1328 }
1329 }
1330
1331 /* Now *l must be either ',' or END_OF_INSN. */
1332 if (*l == ',')
1333 {
1334 if (*++l == END_OF_INSN)
1335 {
1336 /* Just skip it, if it's \n complain. */
1337 goto expecting_operand_after_comma;
1338 }
1339 expecting_operand = 1;
1340 }
1341 }
1342 while (*l != END_OF_INSN);
1343 }
1344 }
1345
1346 /* Now we've parsed the mnemonic into a set of templates, and have the
1347 operands at hand.
1348
1349 Next, we find a template that matches the given insn,
1350 making sure the overlap of the given operands types is consistent
1351 with the template operand types. */
1352
1353 #define MATCH(overlap, given, template) \
1354 ((overlap & ~JumpAbsolute) \
1355 && ((given) & (BaseIndex|JumpAbsolute)) == ((overlap) & (BaseIndex|JumpAbsolute)))
1356
1357 /* If given types r0 and r1 are registers they must be of the same type
1358 unless the expected operand type register overlap is null.
1359 Note that Acc in a template matches every size of reg. */
1360 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
1361 ( ((g0) & Reg) == 0 || ((g1) & Reg) == 0 || \
1362 ((g0) & Reg) == ((g1) & Reg) || \
1363 ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
1364
1365 {
1366 register unsigned int overlap0, overlap1;
1367 unsigned int overlap2;
1368 unsigned int found_reverse_match;
1369 int suffix_check;
1370
1371 /* All intel opcodes have reversed operands except for "bound" and
1372 "enter". We also don't reverse intersegment "jmp" and "call"
1373 instructions with 2 immediate operands so that the immediate segment
1374 precedes the offset, as it does when in AT&T mode. "enter" and the
1375 intersegment "jmp" and "call" instructions are the only ones that
1376 have two immediate operands. */
1377 if (intel_syntax && i.operands > 1
1378 && (strcmp (mnemonic, "bound") != 0)
1379 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1380 {
1381 union i386_op temp_op;
1382 unsigned int temp_type;
1383 int xchg1 = 0;
1384 int xchg2 = 0;
1385
1386 if (i.operands == 2)
1387 {
1388 xchg1 = 0;
1389 xchg2 = 1;
1390 }
1391 else if (i.operands == 3)
1392 {
1393 xchg1 = 0;
1394 xchg2 = 2;
1395 }
1396 temp_type = i.types[xchg2];
1397 i.types[xchg2] = i.types[xchg1];
1398 i.types[xchg1] = temp_type;
1399 temp_op = i.op[xchg2];
1400 i.op[xchg2] = i.op[xchg1];
1401 i.op[xchg1] = temp_op;
1402
1403 if (i.mem_operands == 2)
1404 {
1405 const seg_entry *temp_seg;
1406 temp_seg = i.seg[0];
1407 i.seg[0] = i.seg[1];
1408 i.seg[1] = temp_seg;
1409 }
1410 }
1411
1412 if (i.imm_operands)
1413 {
1414 /* Try to ensure constant immediates are represented in the smallest
1415 opcode possible. */
1416 char guess_suffix = 0;
1417 int op;
1418
1419 if (i.suffix)
1420 guess_suffix = i.suffix;
1421 else if (i.reg_operands)
1422 {
1423 /* Figure out a suffix from the last register operand specified.
1424 We can't do this properly yet, ie. excluding InOutPortReg,
1425 but the following works for instructions with immediates.
1426 In any case, we can't set i.suffix yet. */
1427 for (op = i.operands; --op >= 0;)
1428 if (i.types[op] & Reg)
1429 {
1430 if (i.types[op] & Reg8)
1431 guess_suffix = BYTE_MNEM_SUFFIX;
1432 else if (i.types[op] & Reg16)
1433 guess_suffix = WORD_MNEM_SUFFIX;
1434 break;
1435 }
1436 }
1437 else if (flag_16bit_code ^ (i.prefix[DATA_PREFIX] != 0))
1438 guess_suffix = WORD_MNEM_SUFFIX;
1439
1440 for (op = i.operands; --op >= 0;)
1441 if ((i.types[op] & Imm)
1442 && i.op[op].imms->X_op == O_constant)
1443 {
1444 /* If a suffix is given, this operand may be shortened. */
1445 switch (guess_suffix)
1446 {
1447 case WORD_MNEM_SUFFIX:
1448 i.types[op] |= Imm16;
1449 break;
1450 case BYTE_MNEM_SUFFIX:
1451 i.types[op] |= Imm16 | Imm8 | Imm8S;
1452 break;
1453 }
1454
1455 /* If this operand is at most 16 bits, convert it to a
1456 signed 16 bit number before trying to see whether it will
1457 fit in an even smaller size. This allows a 16-bit operand
1458 such as $0xffe0 to be recognised as within Imm8S range. */
1459 if ((i.types[op] & Imm16)
1460 && (i.op[op].imms->X_add_number & ~(offsetT)0xffff) == 0)
1461 {
1462 i.op[op].imms->X_add_number =
1463 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
1464 }
1465 i.types[op] |= smallest_imm_type ((long) i.op[op].imms->X_add_number);
1466 }
1467 }
1468
1469 if (i.disp_operands)
1470 {
1471 /* Try to use the smallest displacement type too. */
1472 int op;
1473
1474 for (op = i.operands; --op >= 0;)
1475 if ((i.types[op] & Disp)
1476 && i.op[op].imms->X_op == O_constant)
1477 {
1478 offsetT disp = i.op[op].disps->X_add_number;
1479
1480 if (i.types[op] & Disp16)
1481 {
1482 /* We know this operand is at most 16 bits, so
1483 convert to a signed 16 bit number before trying
1484 to see whether it will fit in an even smaller
1485 size. */
1486
1487 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
1488 }
1489 if (fits_in_signed_byte (disp))
1490 i.types[op] |= Disp8;
1491 }
1492 }
1493
1494 overlap0 = 0;
1495 overlap1 = 0;
1496 overlap2 = 0;
1497 found_reverse_match = 0;
1498 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
1499 ? No_bSuf
1500 : (i.suffix == WORD_MNEM_SUFFIX
1501 ? No_wSuf
1502 : (i.suffix == SHORT_MNEM_SUFFIX
1503 ? No_sSuf
1504 : (i.suffix == LONG_MNEM_SUFFIX
1505 ? No_lSuf
1506 : (i.suffix == DWORD_MNEM_SUFFIX
1507 ? No_dSuf
1508 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX ? No_xSuf : 0))))));
1509
1510 for (t = current_templates->start;
1511 t < current_templates->end;
1512 t++)
1513 {
1514 /* Must have right number of operands. */
1515 if (i.operands != t->operands)
1516 continue;
1517
1518 /* Check the suffix, except for some instructions in intel mode. */
1519 if ((t->opcode_modifier & suffix_check)
1520 && !(intel_syntax
1521 && (t->opcode_modifier & IgnoreSize))
1522 && !(intel_syntax
1523 && t->base_opcode == 0xd9
1524 && (t->extension_opcode == 5 /* 0xd9,5 "fldcw" */
1525 || t->extension_opcode == 7))) /* 0xd9,7 "f{n}stcw" */
1526 continue;
1527
1528 else if (!t->operands)
1529 /* 0 operands always matches. */
1530 break;
1531
1532 overlap0 = i.types[0] & t->operand_types[0];
1533 switch (t->operands)
1534 {
1535 case 1:
1536 if (!MATCH (overlap0, i.types[0], t->operand_types[0]))
1537 continue;
1538 break;
1539 case 2:
1540 case 3:
1541 overlap1 = i.types[1] & t->operand_types[1];
1542 if (!MATCH (overlap0, i.types[0], t->operand_types[0])
1543 || !MATCH (overlap1, i.types[1], t->operand_types[1])
1544 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
1545 t->operand_types[0],
1546 overlap1, i.types[1],
1547 t->operand_types[1]))
1548 {
1549 /* Check if other direction is valid ... */
1550 if ((t->opcode_modifier & (D|FloatD)) == 0)
1551 continue;
1552
1553 /* Try reversing direction of operands. */
1554 overlap0 = i.types[0] & t->operand_types[1];
1555 overlap1 = i.types[1] & t->operand_types[0];
1556 if (!MATCH (overlap0, i.types[0], t->operand_types[1])
1557 || !MATCH (overlap1, i.types[1], t->operand_types[0])
1558 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
1559 t->operand_types[1],
1560 overlap1, i.types[1],
1561 t->operand_types[0]))
1562 {
1563 /* Does not match either direction. */
1564 continue;
1565 }
1566 /* found_reverse_match holds which of D or FloatDR
1567 we've found. */
1568 found_reverse_match = t->opcode_modifier & (D|FloatDR);
1569 break;
1570 }
1571 /* Found a forward 2 operand match here. */
1572 if (t->operands == 3)
1573 {
1574 /* Here we make use of the fact that there are no
1575 reverse match 3 operand instructions, and all 3
1576 operand instructions only need to be checked for
1577 register consistency between operands 2 and 3. */
1578 overlap2 = i.types[2] & t->operand_types[2];
1579 if (!MATCH (overlap2, i.types[2], t->operand_types[2])
1580 || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1],
1581 t->operand_types[1],
1582 overlap2, i.types[2],
1583 t->operand_types[2]))
1584
1585 continue;
1586 }
1587 /* Found either forward/reverse 2 or 3 operand match here:
1588 slip through to break. */
1589 }
1590 /* We've found a match; break out of loop. */
1591 break;
1592 }
1593 if (t == current_templates->end)
1594 {
1595 /* We found no match. */
1596 as_bad (_("suffix or operands invalid for `%s'"),
1597 current_templates->start->name);
1598 return;
1599 }
1600
1601 if (!quiet_warnings)
1602 {
1603 if (!intel_syntax
1604 && ((i.types[0] & JumpAbsolute)
1605 != (t->operand_types[0] & JumpAbsolute)))
1606 {
1607 as_warn (_("indirect %s without `*'"), t->name);
1608 }
1609
1610 if ((t->opcode_modifier & (IsPrefix|IgnoreSize))
1611 == (IsPrefix|IgnoreSize))
1612 {
1613 /* Warn them that a data or address size prefix doesn't
1614 affect assembly of the next line of code. */
1615 as_warn (_("stand-alone `%s' prefix"), t->name);
1616 }
1617 }
1618
1619 /* Copy the template we found. */
1620 i.tm = *t;
1621 if (found_reverse_match)
1622 {
1623 /* If we found a reverse match we must alter the opcode
1624 direction bit. found_reverse_match holds bits to change
1625 (different for int & float insns). */
1626
1627 i.tm.base_opcode ^= found_reverse_match;
1628
1629 i.tm.operand_types[0] = t->operand_types[1];
1630 i.tm.operand_types[1] = t->operand_types[0];
1631 }
1632
1633 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1634 if (SYSV386_COMPAT
1635 && intel_syntax
1636 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1637 i.tm.base_opcode ^= FloatR;
1638
1639 if (i.tm.opcode_modifier & FWait)
1640 if (! add_prefix (FWAIT_OPCODE))
1641 return;
1642
1643 /* Check string instruction segment overrides. */
1644 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1645 {
1646 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
1647 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
1648 {
1649 if (i.seg[0] != NULL && i.seg[0] != &es)
1650 {
1651 as_bad (_("`%s' operand %d must use `%%es' segment"),
1652 i.tm.name,
1653 mem_op + 1);
1654 return;
1655 }
1656 /* There's only ever one segment override allowed per instruction.
1657 This instruction possibly has a legal segment override on the
1658 second operand, so copy the segment to where non-string
1659 instructions store it, allowing common code. */
1660 i.seg[0] = i.seg[1];
1661 }
1662 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
1663 {
1664 if (i.seg[1] != NULL && i.seg[1] != &es)
1665 {
1666 as_bad (_("`%s' operand %d must use `%%es' segment"),
1667 i.tm.name,
1668 mem_op + 2);
1669 return;
1670 }
1671 }
1672 }
1673
1674 /* If matched instruction specifies an explicit instruction mnemonic
1675 suffix, use it. */
1676 if (i.tm.opcode_modifier & (Size16 | Size32))
1677 {
1678 if (i.tm.opcode_modifier & Size16)
1679 i.suffix = WORD_MNEM_SUFFIX;
1680 else
1681 i.suffix = LONG_MNEM_SUFFIX;
1682 }
1683 else if (i.reg_operands)
1684 {
1685 /* If there's no instruction mnemonic suffix we try to invent one
1686 based on register operands. */
1687 if (!i.suffix)
1688 {
1689 /* We take i.suffix from the last register operand specified,
1690 Destination register type is more significant than source
1691 register type. */
1692 int op;
1693 for (op = i.operands; --op >= 0;)
1694 if ((i.types[op] & Reg)
1695 && !(i.tm.operand_types[op] & InOutPortReg))
1696 {
1697 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
1698 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
1699 LONG_MNEM_SUFFIX);
1700 break;
1701 }
1702 }
1703 else if (i.suffix == BYTE_MNEM_SUFFIX)
1704 {
1705 int op;
1706 for (op = i.operands; --op >= 0;)
1707 {
1708 /* If this is an eight bit register, it's OK. If it's
1709 the 16 or 32 bit version of an eight bit register,
1710 we will just use the low portion, and that's OK too. */
1711 if (i.types[op] & Reg8)
1712 continue;
1713
1714 /* movzx and movsx should not generate this warning. */
1715 if (intel_syntax
1716 && (i.tm.base_opcode == 0xfb7
1717 || i.tm.base_opcode == 0xfb6
1718 || i.tm.base_opcode == 0xfbe
1719 || i.tm.base_opcode == 0xfbf))
1720 continue;
1721
1722 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4
1723 #if 0
1724 /* Check that the template allows eight bit regs
1725 This kills insns such as `orb $1,%edx', which
1726 maybe should be allowed. */
1727 && (i.tm.operand_types[op] & (Reg8|InOutPortReg))
1728 #endif
1729 )
1730 {
1731 #if REGISTER_WARNINGS
1732 if (!quiet_warnings
1733 && (i.tm.operand_types[op] & InOutPortReg) == 0)
1734 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
1735 (i.op[op].regs - (i.types[op] & Reg16 ? 8 : 16))->reg_name,
1736 i.op[op].regs->reg_name,
1737 i.suffix);
1738 #endif
1739 continue;
1740 }
1741 /* Any other register is bad. */
1742 if (i.types[op] & (Reg | RegMMX | RegXMM
1743 | SReg2 | SReg3
1744 | Control | Debug | Test
1745 | FloatReg | FloatAcc))
1746 {
1747 as_bad (_("`%%%s' not allowed with `%s%c'"),
1748 i.op[op].regs->reg_name,
1749 i.tm.name,
1750 i.suffix);
1751 return;
1752 }
1753 }
1754 }
1755 else if (i.suffix == LONG_MNEM_SUFFIX)
1756 {
1757 int op;
1758
1759 for (op = i.operands; --op >= 0;)
1760 /* Reject eight bit registers, except where the template
1761 requires them. (eg. movzb) */
1762 if ((i.types[op] & Reg8) != 0
1763 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
1764 {
1765 as_bad (_("`%%%s' not allowed with `%s%c'"),
1766 i.op[op].regs->reg_name,
1767 i.tm.name,
1768 i.suffix);
1769 return;
1770 }
1771 #if REGISTER_WARNINGS
1772 /* Warn if the e prefix on a general reg is missing. */
1773 else if (!quiet_warnings
1774 && (i.types[op] & Reg16) != 0
1775 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
1776 {
1777 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
1778 (i.op[op].regs + 8)->reg_name,
1779 i.op[op].regs->reg_name,
1780 i.suffix);
1781 }
1782 #endif
1783 }
1784 else if (i.suffix == WORD_MNEM_SUFFIX)
1785 {
1786 int op;
1787 for (op = i.operands; --op >= 0;)
1788 /* Reject eight bit registers, except where the template
1789 requires them. (eg. movzb) */
1790 if ((i.types[op] & Reg8) != 0
1791 && (i.tm.operand_types[op] & (Reg16|Reg32|Acc)) != 0)
1792 {
1793 as_bad (_("`%%%s' not allowed with `%s%c'"),
1794 i.op[op].regs->reg_name,
1795 i.tm.name,
1796 i.suffix);
1797 return;
1798 }
1799 #if REGISTER_WARNINGS
1800 /* Warn if the e prefix on a general reg is present. */
1801 else if (!quiet_warnings
1802 && (i.types[op] & Reg32) != 0
1803 && (i.tm.operand_types[op] & (Reg16|Acc)) != 0)
1804 {
1805 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
1806 (i.op[op].regs - 8)->reg_name,
1807 i.op[op].regs->reg_name,
1808 i.suffix);
1809 }
1810 #endif
1811 }
1812 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
1813 /* Do nothing if the instruction is going to ignore the prefix. */
1814 ;
1815 else
1816 abort ();
1817 }
1818 else if ((i.tm.opcode_modifier & DefaultSize) && !i.suffix)
1819 {
1820 i.suffix = stackop_size;
1821 }
1822
1823 /* Make still unresolved immediate matches conform to size of immediate
1824 given in i.suffix. Note: overlap2 cannot be an immediate! */
1825 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32))
1826 && overlap0 != Imm8 && overlap0 != Imm8S
1827 && overlap0 != Imm16 && overlap0 != Imm32)
1828 {
1829 if (i.suffix)
1830 {
1831 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
1832 (i.suffix == WORD_MNEM_SUFFIX ? Imm16 : Imm32));
1833 }
1834 else if (overlap0 == (Imm16 | Imm32))
1835 {
1836 overlap0 =
1837 (flag_16bit_code ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32;
1838 }
1839 else
1840 {
1841 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
1842 return;
1843 }
1844 }
1845 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32))
1846 && overlap1 != Imm8 && overlap1 != Imm8S
1847 && overlap1 != Imm16 && overlap1 != Imm32)
1848 {
1849 if (i.suffix)
1850 {
1851 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
1852 (i.suffix == WORD_MNEM_SUFFIX ? Imm16 : Imm32));
1853 }
1854 else if (overlap1 == (Imm16 | Imm32))
1855 {
1856 overlap1 =
1857 (flag_16bit_code ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32;
1858 }
1859 else
1860 {
1861 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
1862 return;
1863 }
1864 }
1865 assert ((overlap2 & Imm) == 0);
1866
1867 i.types[0] = overlap0;
1868 if (overlap0 & ImplicitRegister)
1869 i.reg_operands--;
1870 if (overlap0 & Imm1)
1871 i.imm_operands = 0; /* kludge for shift insns. */
1872
1873 i.types[1] = overlap1;
1874 if (overlap1 & ImplicitRegister)
1875 i.reg_operands--;
1876
1877 i.types[2] = overlap2;
1878 if (overlap2 & ImplicitRegister)
1879 i.reg_operands--;
1880
1881 /* Finalize opcode. First, we change the opcode based on the operand
1882 size given by i.suffix: We need not change things for byte insns. */
1883
1884 if (!i.suffix && (i.tm.opcode_modifier & W))
1885 {
1886 as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
1887 return;
1888 }
1889
1890 /* For movzx and movsx, need to check the register type. */
1891 if (intel_syntax
1892 && (i.tm.base_opcode == 0xfb6 || i.tm.base_opcode == 0xfbe))
1893 if (i.suffix && i.suffix == BYTE_MNEM_SUFFIX)
1894 {
1895 unsigned int prefix = DATA_PREFIX_OPCODE;
1896
1897 if ((i.op[1].regs->reg_type & Reg16) != 0)
1898 if (!add_prefix (prefix))
1899 return;
1900 }
1901
1902 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
1903 {
1904 /* It's not a byte, select word/dword operation. */
1905 if (i.tm.opcode_modifier & W)
1906 {
1907 if (i.tm.opcode_modifier & ShortForm)
1908 i.tm.base_opcode |= 8;
1909 else
1910 i.tm.base_opcode |= 1;
1911 }
1912 /* Now select between word & dword operations via the operand
1913 size prefix, except for instructions that will ignore this
1914 prefix anyway. */
1915 if (((intel_syntax && (i.suffix == DWORD_MNEM_SUFFIX))
1916 || i.suffix == LONG_MNEM_SUFFIX) == flag_16bit_code
1917 && !(i.tm.opcode_modifier & IgnoreSize))
1918 {
1919 unsigned int prefix = DATA_PREFIX_OPCODE;
1920 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
1921 prefix = ADDR_PREFIX_OPCODE;
1922
1923 if (! add_prefix (prefix))
1924 return;
1925 }
1926 /* Size floating point instruction. */
1927 if (i.suffix == LONG_MNEM_SUFFIX
1928 || (intel_syntax && i.suffix == DWORD_MNEM_SUFFIX))
1929 {
1930 if (i.tm.opcode_modifier & FloatMF)
1931 i.tm.base_opcode ^= 4;
1932 }
1933 }
1934
1935 if (i.tm.opcode_modifier & ImmExt)
1936 {
1937 /* These AMD 3DNow! and Intel Katmai New Instructions have an
1938 opcode suffix which is coded in the same place as an 8-bit
1939 immediate field would be. Here we fake an 8-bit immediate
1940 operand from the opcode suffix stored in tm.extension_opcode. */
1941
1942 expressionS *exp;
1943
1944 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
1945
1946 exp = &im_expressions[i.imm_operands++];
1947 i.op[i.operands].imms = exp;
1948 i.types[i.operands++] = Imm8;
1949 exp->X_op = O_constant;
1950 exp->X_add_number = i.tm.extension_opcode;
1951 i.tm.extension_opcode = None;
1952 }
1953
1954 /* For insns with operands there are more diddles to do to the opcode. */
1955 if (i.operands)
1956 {
1957 /* Default segment register this instruction will use
1958 for memory accesses. 0 means unknown.
1959 This is only for optimizing out unnecessary segment overrides. */
1960 const seg_entry *default_seg = 0;
1961
1962 /* The imul $imm, %reg instruction is converted into
1963 imul $imm, %reg, %reg, and the clr %reg instruction
1964 is converted into xor %reg, %reg. */
1965 if (i.tm.opcode_modifier & regKludge)
1966 {
1967 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
1968 /* Pretend we saw the extra register operand. */
1969 assert (i.op[first_reg_op + 1].regs == 0);
1970 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
1971 i.types[first_reg_op + 1] = i.types[first_reg_op];
1972 i.reg_operands = 2;
1973 }
1974
1975 if (i.tm.opcode_modifier & ShortForm)
1976 {
1977 /* The register or float register operand is in operand 0 or 1. */
1978 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
1979 /* Register goes in low 3 bits of opcode. */
1980 i.tm.base_opcode |= i.op[op].regs->reg_num;
1981 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
1982 {
1983 /* Warn about some common errors, but press on regardless.
1984 The first case can be generated by gcc (<= 2.8.1). */
1985 if (i.operands == 2)
1986 {
1987 /* Reversed arguments on faddp, fsubp, etc. */
1988 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
1989 i.op[1].regs->reg_name,
1990 i.op[0].regs->reg_name);
1991 }
1992 else
1993 {
1994 /* Extraneous `l' suffix on fp insn. */
1995 as_warn (_("translating to `%s %%%s'"), i.tm.name,
1996 i.op[0].regs->reg_name);
1997 }
1998 }
1999 }
2000 else if (i.tm.opcode_modifier & Modrm)
2001 {
2002 /* The opcode is completed (modulo i.tm.extension_opcode which
2003 must be put into the modrm byte).
2004 Now, we make the modrm & index base bytes based on all the
2005 info we've collected. */
2006
2007 /* i.reg_operands MUST be the number of real register operands;
2008 implicit registers do not count. */
2009 if (i.reg_operands == 2)
2010 {
2011 unsigned int source, dest;
2012 source = ((i.types[0]
2013 & (Reg | RegMMX | RegXMM
2014 | SReg2 | SReg3
2015 | Control | Debug | Test))
2016 ? 0 : 1);
2017 dest = source + 1;
2018
2019 i.rm.mode = 3;
2020 /* One of the register operands will be encoded in the
2021 i.tm.reg field, the other in the combined i.tm.mode
2022 and i.tm.regmem fields. If no form of this
2023 instruction supports a memory destination operand,
2024 then we assume the source operand may sometimes be
2025 a memory operand and so we need to store the
2026 destination in the i.rm.reg field. */
2027 if ((i.tm.operand_types[dest] & AnyMem) == 0)
2028 {
2029 i.rm.reg = i.op[dest].regs->reg_num;
2030 i.rm.regmem = i.op[source].regs->reg_num;
2031 }
2032 else
2033 {
2034 i.rm.reg = i.op[source].regs->reg_num;
2035 i.rm.regmem = i.op[dest].regs->reg_num;
2036 }
2037 }
2038 else
2039 { /* If it's not 2 reg operands... */
2040 if (i.mem_operands)
2041 {
2042 unsigned int fake_zero_displacement = 0;
2043 unsigned int op = ((i.types[0] & AnyMem)
2044 ? 0
2045 : (i.types[1] & AnyMem) ? 1 : 2);
2046
2047 default_seg = &ds;
2048
2049 if (! i.base_reg)
2050 {
2051 i.rm.mode = 0;
2052 if (! i.disp_operands)
2053 fake_zero_displacement = 1;
2054 if (! i.index_reg)
2055 {
2056 /* Operand is just <disp> */
2057 if (flag_16bit_code ^ (i.prefix[ADDR_PREFIX] != 0))
2058 {
2059 i.rm.regmem = NO_BASE_REGISTER_16;
2060 i.types[op] &= ~Disp;
2061 i.types[op] |= Disp16;
2062 }
2063 else
2064 {
2065 i.rm.regmem = NO_BASE_REGISTER;
2066 i.types[op] &= ~Disp;
2067 i.types[op] |= Disp32;
2068 }
2069 }
2070 else /* ! i.base_reg && i.index_reg */
2071 {
2072 i.sib.index = i.index_reg->reg_num;
2073 i.sib.base = NO_BASE_REGISTER;
2074 i.sib.scale = i.log2_scale_factor;
2075 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2076 i.types[op] &= ~Disp;
2077 i.types[op] |= Disp32; /* Must be 32 bit. */
2078 }
2079 }
2080 else if (i.base_reg->reg_type & Reg16)
2081 {
2082 switch (i.base_reg->reg_num)
2083 {
2084 case 3: /* (%bx) */
2085 if (! i.index_reg)
2086 i.rm.regmem = 7;
2087 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
2088 i.rm.regmem = i.index_reg->reg_num - 6;
2089 break;
2090 case 5: /* (%bp) */
2091 default_seg = &ss;
2092 if (! i.index_reg)
2093 {
2094 i.rm.regmem = 6;
2095 if ((i.types[op] & Disp) == 0)
2096 {
2097 /* fake (%bp) into 0(%bp) */
2098 i.types[op] |= Disp8;
2099 fake_zero_displacement = 1;
2100 }
2101 }
2102 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
2103 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
2104 break;
2105 default: /* (%si) -> 4 or (%di) -> 5 */
2106 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
2107 }
2108 i.rm.mode = mode_from_disp_size (i.types[op]);
2109 }
2110 else /* i.base_reg and 32 bit mode */
2111 {
2112 i.rm.regmem = i.base_reg->reg_num;
2113 i.sib.base = i.base_reg->reg_num;
2114 if (i.base_reg->reg_num == EBP_REG_NUM)
2115 {
2116 default_seg = &ss;
2117 if (i.disp_operands == 0)
2118 {
2119 fake_zero_displacement = 1;
2120 i.types[op] |= Disp8;
2121 }
2122 }
2123 else if (i.base_reg->reg_num == ESP_REG_NUM)
2124 {
2125 default_seg = &ss;
2126 }
2127 i.sib.scale = i.log2_scale_factor;
2128 if (! i.index_reg)
2129 {
2130 /* <disp>(%esp) becomes two byte modrm
2131 with no index register. We've already
2132 stored the code for esp in i.rm.regmem
2133 ie. ESCAPE_TO_TWO_BYTE_ADDRESSING. Any
2134 base register besides %esp will not use
2135 the extra modrm byte. */
2136 i.sib.index = NO_INDEX_REGISTER;
2137 #if ! SCALE1_WHEN_NO_INDEX
2138 /* Another case where we force the second
2139 modrm byte. */
2140 if (i.log2_scale_factor)
2141 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2142 #endif
2143 }
2144 else
2145 {
2146 i.sib.index = i.index_reg->reg_num;
2147 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2148 }
2149 i.rm.mode = mode_from_disp_size (i.types[op]);
2150 }
2151
2152 if (fake_zero_displacement)
2153 {
2154 /* Fakes a zero displacement assuming that i.types[op]
2155 holds the correct displacement size. */
2156 expressionS *exp;
2157
2158 assert (i.op[op].disps == 0);
2159 exp = &disp_expressions[i.disp_operands++];
2160 i.op[op].disps = exp;
2161 exp->X_op = O_constant;
2162 exp->X_add_number = 0;
2163 exp->X_add_symbol = (symbolS *) 0;
2164 exp->X_op_symbol = (symbolS *) 0;
2165 }
2166 }
2167
2168 /* Fill in i.rm.reg or i.rm.regmem field with register
2169 operand (if any) based on i.tm.extension_opcode.
2170 Again, we must be careful to make sure that
2171 segment/control/debug/test/MMX registers are coded
2172 into the i.rm.reg field. */
2173 if (i.reg_operands)
2174 {
2175 unsigned int op =
2176 ((i.types[0]
2177 & (Reg | RegMMX | RegXMM
2178 | SReg2 | SReg3
2179 | Control | Debug | Test))
2180 ? 0
2181 : ((i.types[1]
2182 & (Reg | RegMMX | RegXMM
2183 | SReg2 | SReg3
2184 | Control | Debug | Test))
2185 ? 1
2186 : 2));
2187 /* If there is an extension opcode to put here, the
2188 register number must be put into the regmem field. */
2189 if (i.tm.extension_opcode != None)
2190 i.rm.regmem = i.op[op].regs->reg_num;
2191 else
2192 i.rm.reg = i.op[op].regs->reg_num;
2193
2194 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2
2195 we must set it to 3 to indicate this is a register
2196 operand in the regmem field. */
2197 if (!i.mem_operands)
2198 i.rm.mode = 3;
2199 }
2200
2201 /* Fill in i.rm.reg field with extension opcode (if any). */
2202 if (i.tm.extension_opcode != None)
2203 i.rm.reg = i.tm.extension_opcode;
2204 }
2205 }
2206 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
2207 {
2208 if (i.tm.base_opcode == POP_SEG_SHORT
2209 && i.op[0].regs->reg_num == 1)
2210 {
2211 as_bad (_("you can't `pop %%cs'"));
2212 return;
2213 }
2214 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
2215 }
2216 else if ((i.tm.base_opcode & ~(D|W)) == MOV_AX_DISP32)
2217 {
2218 default_seg = &ds;
2219 }
2220 else if ((i.tm.opcode_modifier & IsString) != 0)
2221 {
2222 /* For the string instructions that allow a segment override
2223 on one of their operands, the default segment is ds. */
2224 default_seg = &ds;
2225 }
2226
2227 /* If a segment was explicitly specified,
2228 and the specified segment is not the default,
2229 use an opcode prefix to select it.
2230 If we never figured out what the default segment is,
2231 then default_seg will be zero at this point,
2232 and the specified segment prefix will always be used. */
2233 if ((i.seg[0]) && (i.seg[0] != default_seg))
2234 {
2235 if (! add_prefix (i.seg[0]->seg_prefix))
2236 return;
2237 }
2238 }
2239 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2240 {
2241 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
2242 as_warn (_("translating to `%sp'"), i.tm.name);
2243 }
2244 }
2245
2246 /* Handle conversion of 'int $3' --> special int3 insn. */
2247 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
2248 {
2249 i.tm.base_opcode = INT3_OPCODE;
2250 i.imm_operands = 0;
2251 }
2252
2253 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
2254 && i.op[0].disps->X_op == O_constant)
2255 {
2256 /* Convert "jmp constant" (and "call constant") to a jump (call) to
2257 the absolute address given by the constant. Since ix86 jumps and
2258 calls are pc relative, we need to generate a reloc. */
2259 i.op[0].disps->X_add_symbol = &abs_symbol;
2260 i.op[0].disps->X_op = O_symbol;
2261 }
2262
2263 /* We are ready to output the insn. */
2264 {
2265 register char *p;
2266
2267 /* Output jumps. */
2268 if (i.tm.opcode_modifier & Jump)
2269 {
2270 int size;
2271 int code16;
2272 int prefix;
2273
2274 code16 = 0;
2275 if (flag_16bit_code)
2276 code16 = CODE16;
2277
2278 prefix = 0;
2279 if (i.prefix[DATA_PREFIX])
2280 {
2281 prefix = 1;
2282 i.prefixes -= 1;
2283 code16 ^= CODE16;
2284 }
2285
2286 size = 4;
2287 if (code16)
2288 size = 2;
2289
2290 if (i.prefixes != 0 && !intel_syntax)
2291 as_warn (_("skipping prefixes on this instruction"));
2292
2293 /* It's always a symbol; End frag & setup for relax.
2294 Make sure there is enough room in this frag for the largest
2295 instruction we may generate in md_convert_frag. This is 2
2296 bytes for the opcode and room for the prefix and largest
2297 displacement. */
2298 frag_grow (prefix + 2 + size);
2299 insn_size += prefix + 1;
2300 /* Prefix and 1 opcode byte go in fr_fix. */
2301 p = frag_more (prefix + 1);
2302 if (prefix)
2303 *p++ = DATA_PREFIX_OPCODE;
2304 *p = i.tm.base_opcode;
2305 /* 1 possible extra opcode + displacement go in var part.
2306 Pass reloc in fr_var. */
2307 frag_var (rs_machine_dependent,
2308 1 + size,
2309 i.disp_reloc[0],
2310 ((unsigned char) *p == JUMP_PC_RELATIVE
2311 ? ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL) | code16
2312 : ENCODE_RELAX_STATE (COND_JUMP, SMALL) | code16),
2313 i.op[0].disps->X_add_symbol,
2314 i.op[0].disps->X_add_number,
2315 p);
2316 }
2317 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
2318 {
2319 int size;
2320
2321 if (i.tm.opcode_modifier & JumpByte)
2322 {
2323 /* This is a loop or jecxz type instruction. */
2324 size = 1;
2325 if (i.prefix[ADDR_PREFIX])
2326 {
2327 insn_size += 1;
2328 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
2329 i.prefixes -= 1;
2330 }
2331 }
2332 else
2333 {
2334 int code16;
2335
2336 code16 = 0;
2337 if (flag_16bit_code)
2338 code16 = CODE16;
2339
2340 if (i.prefix[DATA_PREFIX])
2341 {
2342 insn_size += 1;
2343 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
2344 i.prefixes -= 1;
2345 code16 ^= CODE16;
2346 }
2347
2348 size = 4;
2349 if (code16)
2350 size = 2;
2351 }
2352
2353 if (i.prefixes != 0 && !intel_syntax)
2354 as_warn (_("skipping prefixes on this instruction"));
2355
2356 if (fits_in_unsigned_byte (i.tm.base_opcode))
2357 {
2358 insn_size += 1 + size;
2359 p = frag_more (1 + size);
2360 }
2361 else
2362 {
2363 /* Opcode can be at most two bytes. */
2364 insn_size += 2 + size;
2365 p = frag_more (2 + size);
2366 *p++ = (i.tm.base_opcode >> 8) & 0xff;
2367 }
2368 *p++ = i.tm.base_opcode & 0xff;
2369
2370 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2371 i.op[0].disps, 1, reloc (size, 1, i.disp_reloc[0]));
2372 }
2373 else if (i.tm.opcode_modifier & JumpInterSegment)
2374 {
2375 int size;
2376 int prefix;
2377 int code16;
2378
2379 code16 = 0;
2380 if (flag_16bit_code)
2381 code16 = CODE16;
2382
2383 prefix = 0;
2384 if (i.prefix[DATA_PREFIX])
2385 {
2386 prefix = 1;
2387 i.prefixes -= 1;
2388 code16 ^= CODE16;
2389 }
2390
2391 size = 4;
2392 if (code16)
2393 size = 2;
2394
2395 if (i.prefixes != 0 && !intel_syntax)
2396 as_warn (_("skipping prefixes on this instruction"));
2397
2398 /* 1 opcode; 2 segment; offset */
2399 insn_size += prefix + 1 + 2 + size;
2400 p = frag_more (prefix + 1 + 2 + size);
2401 if (prefix)
2402 *p++ = DATA_PREFIX_OPCODE;
2403 *p++ = i.tm.base_opcode;
2404 if (i.op[1].imms->X_op == O_constant)
2405 {
2406 offsetT n = i.op[1].imms->X_add_number;
2407
2408 if (size == 2
2409 && !fits_in_unsigned_word (n)
2410 && !fits_in_signed_word (n))
2411 {
2412 as_bad (_("16-bit jump out of range"));
2413 return;
2414 }
2415 md_number_to_chars (p, n, size);
2416 }
2417 else
2418 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2419 i.op[1].imms, 0, reloc (size, 0, i.disp_reloc[0]));
2420 if (i.op[0].imms->X_op != O_constant)
2421 as_bad (_("can't handle non absolute segment in `%s'"),
2422 i.tm.name);
2423 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
2424 }
2425 else
2426 {
2427 /* Output normal instructions here. */
2428 unsigned char *q;
2429
2430 /* The prefix bytes. */
2431 for (q = i.prefix;
2432 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
2433 q++)
2434 {
2435 if (*q)
2436 {
2437 insn_size += 1;
2438 p = frag_more (1);
2439 md_number_to_chars (p, (valueT) *q, 1);
2440 }
2441 }
2442
2443 /* Now the opcode; be careful about word order here! */
2444 if (fits_in_unsigned_byte (i.tm.base_opcode))
2445 {
2446 insn_size += 1;
2447 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
2448 }
2449 else if (fits_in_unsigned_word (i.tm.base_opcode))
2450 {
2451 insn_size += 2;
2452 p = frag_more (2);
2453 /* Put out high byte first: can't use md_number_to_chars! */
2454 *p++ = (i.tm.base_opcode >> 8) & 0xff;
2455 *p = i.tm.base_opcode & 0xff;
2456 }
2457 else
2458 { /* Opcode is either 3 or 4 bytes. */
2459 if (i.tm.base_opcode & 0xff000000)
2460 {
2461 insn_size += 4;
2462 p = frag_more (4);
2463 *p++ = (i.tm.base_opcode >> 24) & 0xff;
2464 }
2465 else
2466 {
2467 insn_size += 3;
2468 p = frag_more (3);
2469 }
2470 *p++ = (i.tm.base_opcode >> 16) & 0xff;
2471 *p++ = (i.tm.base_opcode >> 8) & 0xff;
2472 *p = (i.tm.base_opcode) & 0xff;
2473 }
2474
2475 /* Now the modrm byte and sib byte (if present). */
2476 if (i.tm.opcode_modifier & Modrm)
2477 {
2478 insn_size += 1;
2479 p = frag_more (1);
2480 md_number_to_chars (p,
2481 (valueT) (i.rm.regmem << 0
2482 | i.rm.reg << 3
2483 | i.rm.mode << 6),
2484 1);
2485 /* If i.rm.regmem == ESP (4)
2486 && i.rm.mode != (Register mode)
2487 && not 16 bit
2488 ==> need second modrm byte. */
2489 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
2490 && i.rm.mode != 3
2491 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
2492 {
2493 insn_size += 1;
2494 p = frag_more (1);
2495 md_number_to_chars (p,
2496 (valueT) (i.sib.base << 0
2497 | i.sib.index << 3
2498 | i.sib.scale << 6),
2499 1);
2500 }
2501 }
2502
2503 if (i.disp_operands)
2504 {
2505 register unsigned int n;
2506
2507 for (n = 0; n < i.operands; n++)
2508 {
2509 if (i.types[n] & Disp)
2510 {
2511 if (i.op[n].disps->X_op == O_constant)
2512 {
2513 int size;
2514 offsetT val;
2515
2516 size = 4;
2517 if (i.types[n] & (Disp8 | Disp16))
2518 {
2519 size = 2;
2520 if (i.types[n] & Disp8)
2521 size = 1;
2522 }
2523 val = offset_in_range (i.op[n].disps->X_add_number,
2524 size);
2525 insn_size += size;
2526 p = frag_more (size);
2527 md_number_to_chars (p, val, size);
2528 }
2529 else
2530 {
2531 int size = 4;
2532
2533 if (i.types[n] & Disp16)
2534 size = 2;
2535
2536 insn_size += size;
2537 p = frag_more (size);
2538 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2539 i.op[n].disps, 0,
2540 reloc (size, 0, i.disp_reloc[n]));
2541 }
2542 }
2543 }
2544 }
2545
2546 /* Output immediate. */
2547 if (i.imm_operands)
2548 {
2549 register unsigned int n;
2550
2551 for (n = 0; n < i.operands; n++)
2552 {
2553 if (i.types[n] & Imm)
2554 {
2555 if (i.op[n].imms->X_op == O_constant)
2556 {
2557 int size;
2558 offsetT val;
2559
2560 size = 4;
2561 if (i.types[n] & (Imm8 | Imm8S | Imm16))
2562 {
2563 size = 2;
2564 if (i.types[n] & (Imm8 | Imm8S))
2565 size = 1;
2566 }
2567 val = offset_in_range (i.op[n].imms->X_add_number,
2568 size);
2569 insn_size += size;
2570 p = frag_more (size);
2571 md_number_to_chars (p, val, size);
2572 }
2573 else
2574 {
2575 /* Not absolute_section.
2576 Need a 32-bit fixup (don't support 8bit
2577 non-absolute imms). Try to support other
2578 sizes ... */
2579 #ifdef BFD_ASSEMBLER
2580 enum bfd_reloc_code_real reloc_type;
2581 #else
2582 int reloc_type;
2583 #endif
2584 int size = 4;
2585
2586 if (i.types[n] & Imm16)
2587 size = 2;
2588 else if (i.types[n] & (Imm8 | Imm8S))
2589 size = 1;
2590
2591 insn_size += size;
2592 p = frag_more (size);
2593 reloc_type = reloc (size, 0, i.disp_reloc[0]);
2594 #ifdef BFD_ASSEMBLER
2595 if (reloc_type == BFD_RELOC_32
2596 && GOT_symbol
2597 && GOT_symbol == i.op[n].imms->X_add_symbol
2598 && (i.op[n].imms->X_op == O_symbol
2599 || (i.op[n].imms->X_op == O_add
2600 && ((symbol_get_value_expression
2601 (i.op[n].imms->X_op_symbol)->X_op)
2602 == O_subtract))))
2603 {
2604 reloc_type = BFD_RELOC_386_GOTPC;
2605 i.op[n].imms->X_add_number += 3;
2606 }
2607 #endif
2608 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2609 i.op[n].imms, 0, reloc_type);
2610 }
2611 }
2612 }
2613 }
2614 }
2615
2616 #ifdef DEBUG386
2617 if (flag_debug)
2618 {
2619 pi (line, &i);
2620 }
2621 #endif /* DEBUG386 */
2622 }
2623 }
2624 \f
2625 static int i386_immediate PARAMS ((char *));
2626
2627 static int
2628 i386_immediate (imm_start)
2629 char *imm_start;
2630 {
2631 char *save_input_line_pointer;
2632 segT exp_seg = 0;
2633 expressionS *exp;
2634
2635 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
2636 {
2637 as_bad (_("only 1 or 2 immediate operands are allowed"));
2638 return 0;
2639 }
2640
2641 exp = &im_expressions[i.imm_operands++];
2642 i.op[this_operand].imms = exp;
2643
2644 if (is_space_char (*imm_start))
2645 ++imm_start;
2646
2647 save_input_line_pointer = input_line_pointer;
2648 input_line_pointer = imm_start;
2649
2650 #ifndef LEX_AT
2651 {
2652 /* We can have operands of the form
2653 <symbol>@GOTOFF+<nnn>
2654 Take the easy way out here and copy everything
2655 into a temporary buffer... */
2656 register char *cp;
2657
2658 cp = strchr (input_line_pointer, '@');
2659 if (cp != NULL)
2660 {
2661 char *tmpbuf;
2662 int len = 0;
2663 int first;
2664
2665 /* GOT relocations are not supported in 16 bit mode. */
2666 if (flag_16bit_code)
2667 as_bad (_("GOT relocations not supported in 16 bit mode"));
2668
2669 if (GOT_symbol == NULL)
2670 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
2671
2672 if (strncmp (cp + 1, "PLT", 3) == 0)
2673 {
2674 i.disp_reloc[this_operand] = BFD_RELOC_386_PLT32;
2675 len = 3;
2676 }
2677 else if (strncmp (cp + 1, "GOTOFF", 6) == 0)
2678 {
2679 i.disp_reloc[this_operand] = BFD_RELOC_386_GOTOFF;
2680 len = 6;
2681 }
2682 else if (strncmp (cp + 1, "GOT", 3) == 0)
2683 {
2684 i.disp_reloc[this_operand] = BFD_RELOC_386_GOT32;
2685 len = 3;
2686 }
2687 else
2688 as_bad (_("bad reloc specifier in expression"));
2689
2690 /* Replace the relocation token with ' ', so that errors like
2691 foo@GOTOFF1 will be detected. */
2692 first = cp - input_line_pointer;
2693 tmpbuf = (char *) alloca (strlen (input_line_pointer));
2694 memcpy (tmpbuf, input_line_pointer, first);
2695 tmpbuf[first] = ' ';
2696 strcpy (tmpbuf + first + 1, cp + 1 + len);
2697 input_line_pointer = tmpbuf;
2698 }
2699 }
2700 #endif
2701
2702 exp_seg = expression (exp);
2703
2704 SKIP_WHITESPACE ();
2705 if (*input_line_pointer)
2706 as_bad (_("ignoring junk `%s' after expression"), input_line_pointer);
2707
2708 input_line_pointer = save_input_line_pointer;
2709
2710 if (exp->X_op == O_absent || exp->X_op == O_big)
2711 {
2712 /* Missing or bad expr becomes absolute 0. */
2713 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
2714 imm_start);
2715 exp->X_op = O_constant;
2716 exp->X_add_number = 0;
2717 exp->X_add_symbol = (symbolS *) 0;
2718 exp->X_op_symbol = (symbolS *) 0;
2719 }
2720
2721 if (exp->X_op == O_constant)
2722 {
2723 /* Size it properly later. */
2724 i.types[this_operand] |= Imm32;
2725 }
2726 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
2727 else if (1
2728 #ifdef BFD_ASSEMBLER
2729 && OUTPUT_FLAVOR == bfd_target_aout_flavour
2730 #endif
2731 && exp_seg != text_section
2732 && exp_seg != data_section
2733 && exp_seg != bss_section
2734 && exp_seg != undefined_section
2735 #ifdef BFD_ASSEMBLER
2736 && !bfd_is_com_section (exp_seg)
2737 #endif
2738 )
2739 {
2740 #ifdef BFD_ASSEMBLER
2741 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
2742 #else
2743 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
2744 #endif
2745 return 0;
2746 }
2747 #endif
2748 else
2749 {
2750 /* This is an address. The size of the address will be
2751 determined later, depending on destination register,
2752 suffix, or the default for the section. We exclude
2753 Imm8S here so that `push $foo' and other instructions
2754 with an Imm8S form will use Imm16 or Imm32. */
2755 i.types[this_operand] |= (Imm8 | Imm16 | Imm32);
2756 }
2757
2758 return 1;
2759 }
2760
2761 static int i386_scale PARAMS ((char *));
2762
2763 static int
2764 i386_scale (scale)
2765 char *scale;
2766 {
2767 if (!isdigit (*scale))
2768 goto bad_scale;
2769
2770 switch (*scale)
2771 {
2772 case '0':
2773 case '1':
2774 i.log2_scale_factor = 0;
2775 break;
2776 case '2':
2777 i.log2_scale_factor = 1;
2778 break;
2779 case '4':
2780 i.log2_scale_factor = 2;
2781 break;
2782 case '8':
2783 i.log2_scale_factor = 3;
2784 break;
2785 default:
2786 bad_scale:
2787 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
2788 scale);
2789 return 0;
2790 }
2791 if (i.log2_scale_factor != 0 && ! i.index_reg)
2792 {
2793 as_warn (_("scale factor of %d without an index register"),
2794 1 << i.log2_scale_factor);
2795 #if SCALE1_WHEN_NO_INDEX
2796 i.log2_scale_factor = 0;
2797 #endif
2798 }
2799 return 1;
2800 }
2801
2802 static int i386_displacement PARAMS ((char *, char *));
2803
2804 static int
2805 i386_displacement (disp_start, disp_end)
2806 char *disp_start;
2807 char *disp_end;
2808 {
2809 register expressionS *exp;
2810 segT exp_seg = 0;
2811 char *save_input_line_pointer;
2812 int bigdisp = Disp32;
2813
2814 if (flag_16bit_code ^ (i.prefix[ADDR_PREFIX] != 0))
2815 bigdisp = Disp16;
2816 i.types[this_operand] |= bigdisp;
2817
2818 exp = &disp_expressions[i.disp_operands];
2819 i.op[this_operand].disps = exp;
2820 i.disp_operands++;
2821 save_input_line_pointer = input_line_pointer;
2822 input_line_pointer = disp_start;
2823 END_STRING_AND_SAVE (disp_end);
2824
2825 #ifndef GCC_ASM_O_HACK
2826 #define GCC_ASM_O_HACK 0
2827 #endif
2828 #if GCC_ASM_O_HACK
2829 END_STRING_AND_SAVE (disp_end + 1);
2830 if ((i.types[this_operand] & BaseIndex) != 0
2831 && displacement_string_end[-1] == '+')
2832 {
2833 /* This hack is to avoid a warning when using the "o"
2834 constraint within gcc asm statements.
2835 For instance:
2836
2837 #define _set_tssldt_desc(n,addr,limit,type) \
2838 __asm__ __volatile__ ( \
2839 "movw %w2,%0\n\t" \
2840 "movw %w1,2+%0\n\t" \
2841 "rorl $16,%1\n\t" \
2842 "movb %b1,4+%0\n\t" \
2843 "movb %4,5+%0\n\t" \
2844 "movb $0,6+%0\n\t" \
2845 "movb %h1,7+%0\n\t" \
2846 "rorl $16,%1" \
2847 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
2848
2849 This works great except that the output assembler ends
2850 up looking a bit weird if it turns out that there is
2851 no offset. You end up producing code that looks like:
2852
2853 #APP
2854 movw $235,(%eax)
2855 movw %dx,2+(%eax)
2856 rorl $16,%edx
2857 movb %dl,4+(%eax)
2858 movb $137,5+(%eax)
2859 movb $0,6+(%eax)
2860 movb %dh,7+(%eax)
2861 rorl $16,%edx
2862 #NO_APP
2863
2864 So here we provide the missing zero. */
2865
2866 *displacement_string_end = '0';
2867 }
2868 #endif
2869 #ifndef LEX_AT
2870 {
2871 /* We can have operands of the form
2872 <symbol>@GOTOFF+<nnn>
2873 Take the easy way out here and copy everything
2874 into a temporary buffer... */
2875 register char *cp;
2876
2877 cp = strchr (input_line_pointer, '@');
2878 if (cp != NULL)
2879 {
2880 char *tmpbuf;
2881 int len = 0;
2882 int first;
2883
2884 /* GOT relocations are not supported in 16 bit mode. */
2885 if (flag_16bit_code)
2886 as_bad (_("GOT relocations not supported in 16 bit mode"));
2887
2888 if (GOT_symbol == NULL)
2889 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
2890
2891 if (strncmp (cp + 1, "PLT", 3) == 0)
2892 {
2893 i.disp_reloc[this_operand] = BFD_RELOC_386_PLT32;
2894 len = 3;
2895 }
2896 else if (strncmp (cp + 1, "GOTOFF", 6) == 0)
2897 {
2898 i.disp_reloc[this_operand] = BFD_RELOC_386_GOTOFF;
2899 len = 6;
2900 }
2901 else if (strncmp (cp + 1, "GOT", 3) == 0)
2902 {
2903 i.disp_reloc[this_operand] = BFD_RELOC_386_GOT32;
2904 len = 3;
2905 }
2906 else
2907 as_bad (_("bad reloc specifier in expression"));
2908
2909 /* Replace the relocation token with ' ', so that errors like
2910 foo@GOTOFF1 will be detected. */
2911 first = cp - input_line_pointer;
2912 tmpbuf = (char *) alloca (strlen (input_line_pointer));
2913 memcpy (tmpbuf, input_line_pointer, first);
2914 tmpbuf[first] = ' ';
2915 strcpy (tmpbuf + first + 1, cp + 1 + len);
2916 input_line_pointer = tmpbuf;
2917 }
2918 }
2919 #endif
2920
2921 exp_seg = expression (exp);
2922
2923 #ifdef BFD_ASSEMBLER
2924 /* We do this to make sure that the section symbol is in
2925 the symbol table. We will ultimately change the relocation
2926 to be relative to the beginning of the section. */
2927 if (i.disp_reloc[this_operand] == BFD_RELOC_386_GOTOFF)
2928 {
2929 if (S_IS_LOCAL(exp->X_add_symbol)
2930 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
2931 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
2932 assert (exp->X_op == O_symbol);
2933 exp->X_op = O_subtract;
2934 exp->X_op_symbol = GOT_symbol;
2935 i.disp_reloc[this_operand] = BFD_RELOC_32;
2936 }
2937 #endif
2938
2939 SKIP_WHITESPACE ();
2940 if (*input_line_pointer)
2941 as_bad (_("ignoring junk `%s' after expression"),
2942 input_line_pointer);
2943 #if GCC_ASM_O_HACK
2944 RESTORE_END_STRING (disp_end + 1);
2945 #endif
2946 RESTORE_END_STRING (disp_end);
2947 input_line_pointer = save_input_line_pointer;
2948
2949 if (exp->X_op == O_absent || exp->X_op == O_big)
2950 {
2951 /* Missing or bad expr becomes absolute 0. */
2952 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
2953 disp_start);
2954 exp->X_op = O_constant;
2955 exp->X_add_number = 0;
2956 exp->X_add_symbol = (symbolS *) 0;
2957 exp->X_op_symbol = (symbolS *) 0;
2958 }
2959
2960 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
2961 if (exp->X_op != O_constant
2962 #ifdef BFD_ASSEMBLER
2963 && OUTPUT_FLAVOR == bfd_target_aout_flavour
2964 #endif
2965 && exp_seg != text_section
2966 && exp_seg != data_section
2967 && exp_seg != bss_section
2968 && exp_seg != undefined_section)
2969 {
2970 #ifdef BFD_ASSEMBLER
2971 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
2972 #else
2973 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
2974 #endif
2975 return 0;
2976 }
2977 #endif
2978 return 1;
2979 }
2980
2981 static int i386_index_check PARAMS((const char *));
2982
2983 /* Make sure the memory operand we've been dealt is valid.
2984 Return 1 on success, 0 on a failure. */
2985
2986 static int
2987 i386_index_check (operand_string)
2988 const char *operand_string;
2989 {
2990 #if INFER_ADDR_PREFIX
2991 int fudged = 0;
2992
2993 tryprefix:
2994 #endif
2995 if (flag_16bit_code ^ (i.prefix[ADDR_PREFIX] != 0)
2996 /* 16 bit mode checks. */
2997 ? ((i.base_reg
2998 && ((i.base_reg->reg_type & (Reg16|BaseIndex))
2999 != (Reg16|BaseIndex)))
3000 || (i.index_reg
3001 && (((i.index_reg->reg_type & (Reg16|BaseIndex))
3002 != (Reg16|BaseIndex))
3003 || ! (i.base_reg
3004 && i.base_reg->reg_num < 6
3005 && i.index_reg->reg_num >= 6
3006 && i.log2_scale_factor == 0))))
3007 /* 32 bit mode checks. */
3008 : ((i.base_reg
3009 && (i.base_reg->reg_type & Reg32) == 0)
3010 || (i.index_reg
3011 && ((i.index_reg->reg_type & (Reg32|BaseIndex))
3012 != (Reg32|BaseIndex)))))
3013 {
3014 #if INFER_ADDR_PREFIX
3015 if (i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0')
3016 {
3017 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
3018 i.prefixes += 1;
3019 /* Change the size of any displacement too. At most one of
3020 Disp16 or Disp32 is set.
3021 FIXME. There doesn't seem to be any real need for separate
3022 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
3023 Removing them would probably clean up the code quite a lot. */
3024 if (i.types[this_operand] & (Disp16|Disp32))
3025 i.types[this_operand] ^= (Disp16|Disp32);
3026 fudged = 1;
3027 goto tryprefix;
3028 }
3029 if (fudged)
3030 as_bad (_("`%s' is not a valid base/index expression"),
3031 operand_string);
3032 else
3033 #endif
3034 as_bad (_("`%s' is not a valid %s bit base/index expression"),
3035 operand_string,
3036 flag_16bit_code ^ (i.prefix[ADDR_PREFIX] != 0) ? "16" : "32");
3037 return 0;
3038 }
3039 return 1;
3040 }
3041
3042 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
3043 on error. */
3044
3045 static int
3046 i386_operand (operand_string)
3047 char *operand_string;
3048 {
3049 const reg_entry *r;
3050 char *end_op;
3051 char *op_string = operand_string;
3052
3053 if (is_space_char (*op_string))
3054 ++op_string;
3055
3056 /* We check for an absolute prefix (differentiating,
3057 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
3058 if (*op_string == ABSOLUTE_PREFIX)
3059 {
3060 ++op_string;
3061 if (is_space_char (*op_string))
3062 ++op_string;
3063 i.types[this_operand] |= JumpAbsolute;
3064 }
3065
3066 /* Check if operand is a register. */
3067 if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
3068 && (r = parse_register (op_string, &end_op)) != NULL)
3069 {
3070 /* Check for a segment override by searching for ':' after a
3071 segment register. */
3072 op_string = end_op;
3073 if (is_space_char (*op_string))
3074 ++op_string;
3075 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
3076 {
3077 switch (r->reg_num)
3078 {
3079 case 0:
3080 i.seg[i.mem_operands] = &es;
3081 break;
3082 case 1:
3083 i.seg[i.mem_operands] = &cs;
3084 break;
3085 case 2:
3086 i.seg[i.mem_operands] = &ss;
3087 break;
3088 case 3:
3089 i.seg[i.mem_operands] = &ds;
3090 break;
3091 case 4:
3092 i.seg[i.mem_operands] = &fs;
3093 break;
3094 case 5:
3095 i.seg[i.mem_operands] = &gs;
3096 break;
3097 }
3098
3099 /* Skip the ':' and whitespace. */
3100 ++op_string;
3101 if (is_space_char (*op_string))
3102 ++op_string;
3103
3104 if (!is_digit_char (*op_string)
3105 && !is_identifier_char (*op_string)
3106 && *op_string != '('
3107 && *op_string != ABSOLUTE_PREFIX)
3108 {
3109 as_bad (_("bad memory operand `%s'"), op_string);
3110 return 0;
3111 }
3112 /* Handle case of %es:*foo. */
3113 if (*op_string == ABSOLUTE_PREFIX)
3114 {
3115 ++op_string;
3116 if (is_space_char (*op_string))
3117 ++op_string;
3118 i.types[this_operand] |= JumpAbsolute;
3119 }
3120 goto do_memory_reference;
3121 }
3122 if (*op_string)
3123 {
3124 as_bad (_("junk `%s' after register"), op_string);
3125 return 0;
3126 }
3127 i.types[this_operand] |= r->reg_type & ~BaseIndex;
3128 i.op[this_operand].regs = r;
3129 i.reg_operands++;
3130 }
3131 else if (*op_string == REGISTER_PREFIX)
3132 {
3133 as_bad (_("bad register name `%s'"), op_string);
3134 return 0;
3135 }
3136 else if (*op_string == IMMEDIATE_PREFIX)
3137 {
3138 ++op_string;
3139 if (i.types[this_operand] & JumpAbsolute)
3140 {
3141 as_bad (_("immediate operand illegal with absolute jump"));
3142 return 0;
3143 }
3144 if (!i386_immediate (op_string))
3145 return 0;
3146 }
3147 else if (is_digit_char (*op_string)
3148 || is_identifier_char (*op_string)
3149 || *op_string == '(' )
3150 {
3151 /* This is a memory reference of some sort. */
3152 char *base_string;
3153
3154 /* Start and end of displacement string expression (if found). */
3155 char *displacement_string_start;
3156 char *displacement_string_end;
3157
3158 do_memory_reference:
3159 if ((i.mem_operands == 1
3160 && (current_templates->start->opcode_modifier & IsString) == 0)
3161 || i.mem_operands == 2)
3162 {
3163 as_bad (_("too many memory references for `%s'"),
3164 current_templates->start->name);
3165 return 0;
3166 }
3167
3168 /* Check for base index form. We detect the base index form by
3169 looking for an ')' at the end of the operand, searching
3170 for the '(' matching it, and finding a REGISTER_PREFIX or ','
3171 after the '('. */
3172 base_string = op_string + strlen (op_string);
3173
3174 --base_string;
3175 if (is_space_char (*base_string))
3176 --base_string;
3177
3178 /* If we only have a displacement, set-up for it to be parsed later. */
3179 displacement_string_start = op_string;
3180 displacement_string_end = base_string + 1;
3181
3182 if (*base_string == ')')
3183 {
3184 char *temp_string;
3185 unsigned int parens_balanced = 1;
3186 /* We've already checked that the number of left & right ()'s are
3187 equal, so this loop will not be infinite. */
3188 do
3189 {
3190 base_string--;
3191 if (*base_string == ')')
3192 parens_balanced++;
3193 if (*base_string == '(')
3194 parens_balanced--;
3195 }
3196 while (parens_balanced);
3197
3198 temp_string = base_string;
3199
3200 /* Skip past '(' and whitespace. */
3201 ++base_string;
3202 if (is_space_char (*base_string))
3203 ++base_string;
3204
3205 if (*base_string == ','
3206 || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3207 && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
3208 {
3209 displacement_string_end = temp_string;
3210
3211 i.types[this_operand] |= BaseIndex;
3212
3213 if (i.base_reg)
3214 {
3215 base_string = end_op;
3216 if (is_space_char (*base_string))
3217 ++base_string;
3218 }
3219
3220 /* There may be an index reg or scale factor here. */
3221 if (*base_string == ',')
3222 {
3223 ++base_string;
3224 if (is_space_char (*base_string))
3225 ++base_string;
3226
3227 if ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3228 && (i.index_reg = parse_register (base_string, &end_op)) != NULL)
3229 {
3230 base_string = end_op;
3231 if (is_space_char (*base_string))
3232 ++base_string;
3233 if (*base_string == ',')
3234 {
3235 ++base_string;
3236 if (is_space_char (*base_string))
3237 ++base_string;
3238 }
3239 else if (*base_string != ')' )
3240 {
3241 as_bad (_("expecting `,' or `)' after index register in `%s'"),
3242 operand_string);
3243 return 0;
3244 }
3245 }
3246 else if (*base_string == REGISTER_PREFIX)
3247 {
3248 as_bad (_("bad register name `%s'"), base_string);
3249 return 0;
3250 }
3251
3252 /* Check for scale factor. */
3253 if (isdigit ((unsigned char) *base_string))
3254 {
3255 if (!i386_scale (base_string))
3256 return 0;
3257
3258 ++base_string;
3259 if (is_space_char (*base_string))
3260 ++base_string;
3261 if (*base_string != ')')
3262 {
3263 as_bad (_("expecting `)' after scale factor in `%s'"),
3264 operand_string);
3265 return 0;
3266 }
3267 }
3268 else if (!i.index_reg)
3269 {
3270 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
3271 *base_string);
3272 return 0;
3273 }
3274 }
3275 else if (*base_string != ')')
3276 {
3277 as_bad (_("expecting `,' or `)' after base register in `%s'"),
3278 operand_string);
3279 return 0;
3280 }
3281 }
3282 else if (*base_string == REGISTER_PREFIX)
3283 {
3284 as_bad (_("bad register name `%s'"), base_string);
3285 return 0;
3286 }
3287 }
3288
3289 /* If there's an expression beginning the operand, parse it,
3290 assuming displacement_string_start and
3291 displacement_string_end are meaningful. */
3292 if (displacement_string_start != displacement_string_end)
3293 {
3294 if (!i386_displacement (displacement_string_start,
3295 displacement_string_end))
3296 return 0;
3297 }
3298
3299 /* Special case for (%dx) while doing input/output op. */
3300 if (i.base_reg
3301 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
3302 && i.index_reg == 0
3303 && i.log2_scale_factor == 0
3304 && i.seg[i.mem_operands] == 0
3305 && (i.types[this_operand] & Disp) == 0)
3306 {
3307 i.types[this_operand] = InOutPortReg;
3308 return 1;
3309 }
3310
3311 if (i386_index_check (operand_string) == 0)
3312 return 0;
3313 i.mem_operands++;
3314 }
3315 else
3316 {
3317 /* It's not a memory operand; argh! */
3318 as_bad (_("invalid char %s beginning operand %d `%s'"),
3319 output_invalid (*op_string),
3320 this_operand + 1,
3321 op_string);
3322 return 0;
3323 }
3324 return 1; /* Normal return. */
3325 }
3326 \f
3327 /* md_estimate_size_before_relax()
3328
3329 Called just before relax() for rs_machine_dependent frags. The x86
3330 assembler uses these frags to handle variable size jump
3331 instructions.
3332
3333 Any symbol that is now undefined will not become defined.
3334 Return the correct fr_subtype in the frag.
3335 Return the initial "guess for variable size of frag" to caller.
3336 The guess is actually the growth beyond the fixed part. Whatever
3337 we do to grow the fixed or variable part contributes to our
3338 returned value. */
3339
3340 int
3341 md_estimate_size_before_relax (fragP, segment)
3342 register fragS *fragP;
3343 register segT segment;
3344 {
3345 /* We've already got fragP->fr_subtype right; all we have to do is
3346 check for un-relaxable symbols. On an ELF system, we can't relax
3347 an externally visible symbol, because it may be overridden by a
3348 shared library. */
3349 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
3350 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
3351 || S_IS_EXTERNAL (fragP->fr_symbol)
3352 || S_IS_WEAK (fragP->fr_symbol)
3353 #endif
3354 )
3355 {
3356 /* Symbol is undefined in this segment, or we need to keep a
3357 reloc so that weak symbols can be overridden. */
3358 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
3359 #ifdef BFD_ASSEMBLER
3360 enum bfd_reloc_code_real reloc_type;
3361 #else
3362 int reloc_type;
3363 #endif
3364 unsigned char *opcode;
3365 int old_fr_fix;
3366
3367 if (fragP->fr_var != NO_RELOC)
3368 reloc_type = fragP->fr_var;
3369 else if (size == 2)
3370 reloc_type = BFD_RELOC_16_PCREL;
3371 else
3372 reloc_type = BFD_RELOC_32_PCREL;
3373
3374 old_fr_fix = fragP->fr_fix;
3375 opcode = (unsigned char *) fragP->fr_opcode;
3376
3377 switch (opcode[0])
3378 {
3379 case JUMP_PC_RELATIVE:
3380 /* Make jmp (0xeb) a dword displacement jump. */
3381 opcode[0] = 0xe9;
3382 fragP->fr_fix += size;
3383 fix_new (fragP, old_fr_fix, size,
3384 fragP->fr_symbol,
3385 fragP->fr_offset, 1,
3386 reloc_type);
3387 break;
3388
3389 default:
3390 /* This changes the byte-displacement jump 0x7N
3391 to the dword-displacement jump 0x0f,0x8N. */
3392 opcode[1] = opcode[0] + 0x10;
3393 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
3394 /* We've added an opcode byte. */
3395 fragP->fr_fix += 1 + size;
3396 fix_new (fragP, old_fr_fix + 1, size,
3397 fragP->fr_symbol,
3398 fragP->fr_offset, 1,
3399 reloc_type);
3400 break;
3401 }
3402 frag_wane (fragP);
3403 return fragP->fr_fix - old_fr_fix;
3404 }
3405 /* Guess a short jump. */
3406 return 1;
3407 }
3408
3409 /* Called after relax() is finished.
3410
3411 In: Address of frag.
3412 fr_type == rs_machine_dependent.
3413 fr_subtype is what the address relaxed to.
3414
3415 Out: Any fixSs and constants are set up.
3416 Caller will turn frag into a ".space 0". */
3417
3418 #ifndef BFD_ASSEMBLER
3419 void
3420 md_convert_frag (headers, sec, fragP)
3421 object_headers *headers ATTRIBUTE_UNUSED;
3422 segT sec ATTRIBUTE_UNUSED;
3423 register fragS *fragP;
3424 #else
3425 void
3426 md_convert_frag (abfd, sec, fragP)
3427 bfd *abfd ATTRIBUTE_UNUSED;
3428 segT sec ATTRIBUTE_UNUSED;
3429 register fragS *fragP;
3430 #endif
3431 {
3432 register unsigned char *opcode;
3433 unsigned char *where_to_put_displacement = NULL;
3434 offsetT target_address;
3435 offsetT opcode_address;
3436 unsigned int extension = 0;
3437 offsetT displacement_from_opcode_start;
3438
3439 opcode = (unsigned char *) fragP->fr_opcode;
3440
3441 /* Address we want to reach in file space. */
3442 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
3443 #ifdef BFD_ASSEMBLER
3444 /* Not needed otherwise? */
3445 target_address += symbol_get_frag (fragP->fr_symbol)->fr_address;
3446 #endif
3447
3448 /* Address opcode resides at in file space. */
3449 opcode_address = fragP->fr_address + fragP->fr_fix;
3450
3451 /* Displacement from opcode start to fill into instruction. */
3452 displacement_from_opcode_start = target_address - opcode_address;
3453
3454 switch (fragP->fr_subtype)
3455 {
3456 case ENCODE_RELAX_STATE (COND_JUMP, SMALL):
3457 case ENCODE_RELAX_STATE (COND_JUMP, SMALL16):
3458 case ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL):
3459 case ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL16):
3460 /* Don't have to change opcode. */
3461 extension = 1; /* 1 opcode + 1 displacement */
3462 where_to_put_displacement = &opcode[1];
3463 break;
3464
3465 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
3466 extension = 5; /* 2 opcode + 4 displacement */
3467 opcode[1] = opcode[0] + 0x10;
3468 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
3469 where_to_put_displacement = &opcode[2];
3470 break;
3471
3472 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
3473 extension = 4; /* 1 opcode + 4 displacement */
3474 opcode[0] = 0xe9;
3475 where_to_put_displacement = &opcode[1];
3476 break;
3477
3478 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
3479 extension = 3; /* 2 opcode + 2 displacement */
3480 opcode[1] = opcode[0] + 0x10;
3481 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
3482 where_to_put_displacement = &opcode[2];
3483 break;
3484
3485 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
3486 extension = 2; /* 1 opcode + 2 displacement */
3487 opcode[0] = 0xe9;
3488 where_to_put_displacement = &opcode[1];
3489 break;
3490
3491 default:
3492 BAD_CASE (fragP->fr_subtype);
3493 break;
3494 }
3495 /* Now put displacement after opcode. */
3496 md_number_to_chars ((char *) where_to_put_displacement,
3497 (valueT) (displacement_from_opcode_start - extension),
3498 SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
3499 fragP->fr_fix += extension;
3500 }
3501 \f
3502 /* Size of byte displacement jmp. */
3503 int md_short_jump_size = 2;
3504
3505 /* Size of dword displacement jmp. */
3506 int md_long_jump_size = 5;
3507
3508 /* Size of relocation record. */
3509 const int md_reloc_size = 8;
3510
3511 void
3512 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
3513 char *ptr;
3514 addressT from_addr, to_addr;
3515 fragS *frag ATTRIBUTE_UNUSED;
3516 symbolS *to_symbol ATTRIBUTE_UNUSED;
3517 {
3518 offsetT offset;
3519
3520 offset = to_addr - (from_addr + 2);
3521 /* Opcode for byte-disp jump. */
3522 md_number_to_chars (ptr, (valueT) 0xeb, 1);
3523 md_number_to_chars (ptr + 1, (valueT) offset, 1);
3524 }
3525
3526 void
3527 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
3528 char *ptr;
3529 addressT from_addr, to_addr;
3530 fragS *frag ATTRIBUTE_UNUSED;
3531 symbolS *to_symbol ATTRIBUTE_UNUSED;
3532 {
3533 offsetT offset;
3534
3535 offset = to_addr - (from_addr + 5);
3536 md_number_to_chars (ptr, (valueT) 0xe9, 1);
3537 md_number_to_chars (ptr + 1, (valueT) offset, 4);
3538 }
3539 \f
3540 /* Apply a fixup (fixS) to segment data, once it has been determined
3541 by our caller that we have all the info we need to fix it up.
3542
3543 On the 386, immediates, displacements, and data pointers are all in
3544 the same (little-endian) format, so we don't need to care about which
3545 we are handling. */
3546
3547 int
3548 md_apply_fix3 (fixP, valp, seg)
3549 /* The fix we're to put in. */
3550 fixS *fixP;
3551
3552 /* Pointer to the value of the bits. */
3553 valueT *valp;
3554
3555 /* Segment fix is from. */
3556 segT seg ATTRIBUTE_UNUSED;
3557 {
3558 register char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
3559 valueT value = *valp;
3560
3561 #if defined (BFD_ASSEMBLER) && !defined (TE_Mach)
3562 if (fixP->fx_pcrel)
3563 {
3564 switch (fixP->fx_r_type)
3565 {
3566 default:
3567 break;
3568
3569 case BFD_RELOC_32:
3570 fixP->fx_r_type = BFD_RELOC_32_PCREL;
3571 break;
3572 case BFD_RELOC_16:
3573 fixP->fx_r_type = BFD_RELOC_16_PCREL;
3574 break;
3575 case BFD_RELOC_8:
3576 fixP->fx_r_type = BFD_RELOC_8_PCREL;
3577 break;
3578 }
3579 }
3580
3581 /* This is a hack. There should be a better way to handle this.
3582 This covers for the fact that bfd_install_relocation will
3583 subtract the current location (for partial_inplace, PC relative
3584 relocations); see more below. */
3585 if ((fixP->fx_r_type == BFD_RELOC_32_PCREL
3586 || fixP->fx_r_type == BFD_RELOC_16_PCREL
3587 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
3588 && fixP->fx_addsy)
3589 {
3590 #ifndef OBJ_AOUT
3591 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
3592 #ifdef TE_PE
3593 || OUTPUT_FLAVOR == bfd_target_coff_flavour
3594 #endif
3595 )
3596 value += fixP->fx_where + fixP->fx_frag->fr_address;
3597 #endif
3598 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
3599 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
3600 {
3601 segT fseg = S_GET_SEGMENT (fixP->fx_addsy);
3602
3603 if ((fseg == seg
3604 || (symbol_section_p (fixP->fx_addsy)
3605 && fseg != absolute_section))
3606 && ! S_IS_EXTERNAL (fixP->fx_addsy)
3607 && ! S_IS_WEAK (fixP->fx_addsy)
3608 && S_IS_DEFINED (fixP->fx_addsy)
3609 && ! S_IS_COMMON (fixP->fx_addsy))
3610 {
3611 /* Yes, we add the values in twice. This is because
3612 bfd_perform_relocation subtracts them out again. I think
3613 bfd_perform_relocation is broken, but I don't dare change
3614 it. FIXME. */
3615 value += fixP->fx_where + fixP->fx_frag->fr_address;
3616 }
3617 }
3618 #endif
3619 #if defined (OBJ_COFF) && defined (TE_PE)
3620 /* For some reason, the PE format does not store a section
3621 address offset for a PC relative symbol. */
3622 if (S_GET_SEGMENT (fixP->fx_addsy) != seg)
3623 value += md_pcrel_from (fixP);
3624 #endif
3625 }
3626
3627 /* Fix a few things - the dynamic linker expects certain values here,
3628 and we must not dissappoint it. */
3629 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
3630 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
3631 && fixP->fx_addsy)
3632 switch (fixP->fx_r_type)
3633 {
3634 case BFD_RELOC_386_PLT32:
3635 /* Make the jump instruction point to the address of the operand. At
3636 runtime we merely add the offset to the actual PLT entry. */
3637 value = -4;
3638 break;
3639 case BFD_RELOC_386_GOTPC:
3640
3641 /* This is tough to explain. We end up with this one if we have
3642 * operands that look like "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal
3643 * here is to obtain the absolute address of the GOT, and it is strongly
3644 * preferable from a performance point of view to avoid using a runtime
3645 * relocation for this. The actual sequence of instructions often look
3646 * something like:
3647 *
3648 * call .L66
3649 * .L66:
3650 * popl %ebx
3651 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
3652 *
3653 * The call and pop essentially return the absolute address of
3654 * the label .L66 and store it in %ebx. The linker itself will
3655 * ultimately change the first operand of the addl so that %ebx points to
3656 * the GOT, but to keep things simple, the .o file must have this operand
3657 * set so that it generates not the absolute address of .L66, but the
3658 * absolute address of itself. This allows the linker itself simply
3659 * treat a GOTPC relocation as asking for a pcrel offset to the GOT to be
3660 * added in, and the addend of the relocation is stored in the operand
3661 * field for the instruction itself.
3662 *
3663 * Our job here is to fix the operand so that it would add the correct
3664 * offset so that %ebx would point to itself. The thing that is tricky is
3665 * that .-.L66 will point to the beginning of the instruction, so we need
3666 * to further modify the operand so that it will point to itself.
3667 * There are other cases where you have something like:
3668 *
3669 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
3670 *
3671 * and here no correction would be required. Internally in the assembler
3672 * we treat operands of this form as not being pcrel since the '.' is
3673 * explicitly mentioned, and I wonder whether it would simplify matters
3674 * to do it this way. Who knows. In earlier versions of the PIC patches,
3675 * the pcrel_adjust field was used to store the correction, but since the
3676 * expression is not pcrel, I felt it would be confusing to do it this
3677 * way. */
3678
3679 value -= 1;
3680 break;
3681 case BFD_RELOC_386_GOT32:
3682 value = 0; /* Fully resolved at runtime. No addend. */
3683 break;
3684 case BFD_RELOC_386_GOTOFF:
3685 break;
3686
3687 case BFD_RELOC_VTABLE_INHERIT:
3688 case BFD_RELOC_VTABLE_ENTRY:
3689 fixP->fx_done = 0;
3690 return 1;
3691
3692 default:
3693 break;
3694 }
3695 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
3696 *valp = value;
3697 #endif /* defined (BFD_ASSEMBLER) && !defined (TE_Mach) */
3698 md_number_to_chars (p, value, fixP->fx_size);
3699
3700 return 1;
3701 }
3702 \f
3703 #define MAX_LITTLENUMS 6
3704
3705 /* Turn the string pointed to by litP into a floating point constant
3706 of type TYPE, and emit the appropriate bytes. The number of
3707 LITTLENUMS emitted is stored in *SIZEP. An error message is
3708 returned, or NULL on OK. */
3709
3710 char *
3711 md_atof (type, litP, sizeP)
3712 int type;
3713 char *litP;
3714 int *sizeP;
3715 {
3716 int prec;
3717 LITTLENUM_TYPE words[MAX_LITTLENUMS];
3718 LITTLENUM_TYPE *wordP;
3719 char *t;
3720
3721 switch (type)
3722 {
3723 case 'f':
3724 case 'F':
3725 prec = 2;
3726 break;
3727
3728 case 'd':
3729 case 'D':
3730 prec = 4;
3731 break;
3732
3733 case 'x':
3734 case 'X':
3735 prec = 5;
3736 break;
3737
3738 default:
3739 *sizeP = 0;
3740 return _("Bad call to md_atof ()");
3741 }
3742 t = atof_ieee (input_line_pointer, type, words);
3743 if (t)
3744 input_line_pointer = t;
3745
3746 *sizeP = prec * sizeof (LITTLENUM_TYPE);
3747 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
3748 the bigendian 386. */
3749 for (wordP = words + prec - 1; prec--;)
3750 {
3751 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
3752 litP += sizeof (LITTLENUM_TYPE);
3753 }
3754 return 0;
3755 }
3756 \f
3757 char output_invalid_buf[8];
3758
3759 static char *
3760 output_invalid (c)
3761 int c;
3762 {
3763 if (isprint (c))
3764 sprintf (output_invalid_buf, "'%c'", c);
3765 else
3766 sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
3767 return output_invalid_buf;
3768 }
3769
3770 /* REG_STRING starts *before* REGISTER_PREFIX. */
3771
3772 static const reg_entry *
3773 parse_register (reg_string, end_op)
3774 char *reg_string;
3775 char **end_op;
3776 {
3777 char *s = reg_string;
3778 char *p;
3779 char reg_name_given[MAX_REG_NAME_SIZE + 1];
3780 const reg_entry *r;
3781
3782 /* Skip possible REGISTER_PREFIX and possible whitespace. */
3783 if (*s == REGISTER_PREFIX)
3784 ++s;
3785
3786 if (is_space_char (*s))
3787 ++s;
3788
3789 p = reg_name_given;
3790 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
3791 {
3792 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
3793 return (const reg_entry *) NULL;
3794 s++;
3795 }
3796
3797 /* For naked regs, make sure that we are not dealing with an identifier.
3798 This prevents confusing an identifier like `eax_var' with register
3799 `eax'. */
3800 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
3801 return (const reg_entry *) NULL;
3802
3803 *end_op = s;
3804
3805 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
3806
3807 /* Handle floating point regs, allowing spaces in the (i) part. */
3808 if (r == i386_regtab /* %st is first entry of table */)
3809 {
3810 if (is_space_char (*s))
3811 ++s;
3812 if (*s == '(')
3813 {
3814 ++s;
3815 if (is_space_char (*s))
3816 ++s;
3817 if (*s >= '0' && *s <= '7')
3818 {
3819 r = &i386_float_regtab[*s - '0'];
3820 ++s;
3821 if (is_space_char (*s))
3822 ++s;
3823 if (*s == ')')
3824 {
3825 *end_op = s + 1;
3826 return r;
3827 }
3828 }
3829 /* We have "%st(" then garbage. */
3830 return (const reg_entry *) NULL;
3831 }
3832 }
3833
3834 return r;
3835 }
3836 \f
3837 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
3838 const char *md_shortopts = "kVQ:sq";
3839 #else
3840 const char *md_shortopts = "q";
3841 #endif
3842 struct option md_longopts[] = {
3843 {NULL, no_argument, NULL, 0}
3844 };
3845 size_t md_longopts_size = sizeof (md_longopts);
3846
3847 int
3848 md_parse_option (c, arg)
3849 int c;
3850 char *arg ATTRIBUTE_UNUSED;
3851 {
3852 switch (c)
3853 {
3854 case 'q':
3855 quiet_warnings = 1;
3856 break;
3857
3858 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
3859 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
3860 should be emitted or not. FIXME: Not implemented. */
3861 case 'Q':
3862 break;
3863
3864 /* -V: SVR4 argument to print version ID. */
3865 case 'V':
3866 print_version_id ();
3867 break;
3868
3869 /* -k: Ignore for FreeBSD compatibility. */
3870 case 'k':
3871 break;
3872
3873 case 's':
3874 /* -s: On i386 Solaris, this tells the native assembler to use
3875 .stab instead of .stab.excl. We always use .stab anyhow. */
3876 break;
3877 #endif
3878
3879 default:
3880 return 0;
3881 }
3882 return 1;
3883 }
3884
3885 void
3886 md_show_usage (stream)
3887 FILE *stream;
3888 {
3889 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
3890 fprintf (stream, _("\
3891 -Q ignored\n\
3892 -V print assembler version number\n\
3893 -k ignored\n\
3894 -q quieten some warnings\n\
3895 -s ignored\n"));
3896 #else
3897 fprintf (stream, _("\
3898 -q quieten some warnings\n"));
3899 #endif
3900 }
3901
3902 #ifdef BFD_ASSEMBLER
3903 #if ((defined (OBJ_MAYBE_ELF) && defined (OBJ_MAYBE_COFF)) \
3904 || (defined (OBJ_MAYBE_ELF) && defined (OBJ_MAYBE_AOUT)) \
3905 || (defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)))
3906
3907 /* Pick the target format to use. */
3908
3909 const char *
3910 i386_target_format ()
3911 {
3912 switch (OUTPUT_FLAVOR)
3913 {
3914 #ifdef OBJ_MAYBE_AOUT
3915 case bfd_target_aout_flavour:
3916 return AOUT_TARGET_FORMAT;
3917 #endif
3918 #ifdef OBJ_MAYBE_COFF
3919 case bfd_target_coff_flavour:
3920 return "coff-i386";
3921 #endif
3922 #ifdef OBJ_MAYBE_ELF
3923 case bfd_target_elf_flavour:
3924 return "elf32-i386";
3925 #endif
3926 default:
3927 abort ();
3928 return NULL;
3929 }
3930 }
3931
3932 #endif /* OBJ_MAYBE_ more than one */
3933 #endif /* BFD_ASSEMBLER */
3934 \f
3935 symbolS *
3936 md_undefined_symbol (name)
3937 char *name;
3938 {
3939 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
3940 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
3941 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
3942 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
3943 {
3944 if (!GOT_symbol)
3945 {
3946 if (symbol_find (name))
3947 as_bad (_("GOT already in symbol table"));
3948 GOT_symbol = symbol_new (name, undefined_section,
3949 (valueT) 0, &zero_address_frag);
3950 };
3951 return GOT_symbol;
3952 }
3953 return 0;
3954 }
3955
3956 /* Round up a section size to the appropriate boundary. */
3957
3958 valueT
3959 md_section_align (segment, size)
3960 segT segment ATTRIBUTE_UNUSED;
3961 valueT size;
3962 {
3963 #ifdef BFD_ASSEMBLER
3964 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3965 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
3966 {
3967 /* For a.out, force the section size to be aligned. If we don't do
3968 this, BFD will align it for us, but it will not write out the
3969 final bytes of the section. This may be a bug in BFD, but it is
3970 easier to fix it here since that is how the other a.out targets
3971 work. */
3972 int align;
3973
3974 align = bfd_get_section_alignment (stdoutput, segment);
3975 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
3976 }
3977 #endif
3978 #endif
3979
3980 return size;
3981 }
3982
3983 /* On the i386, PC-relative offsets are relative to the start of the
3984 next instruction. That is, the address of the offset, plus its
3985 size, since the offset is always the last part of the insn. */
3986
3987 long
3988 md_pcrel_from (fixP)
3989 fixS *fixP;
3990 {
3991 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
3992 }
3993
3994 #ifndef I386COFF
3995
3996 static void
3997 s_bss (ignore)
3998 int ignore ATTRIBUTE_UNUSED;
3999 {
4000 register int temp;
4001
4002 temp = get_absolute_expression ();
4003 subseg_set (bss_section, (subsegT) temp);
4004 demand_empty_rest_of_line ();
4005 }
4006
4007 #endif
4008
4009 #ifdef BFD_ASSEMBLER
4010
4011 void
4012 i386_validate_fix (fixp)
4013 fixS *fixp;
4014 {
4015 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
4016 {
4017 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
4018 fixp->fx_subsy = 0;
4019 }
4020 }
4021
4022 arelent *
4023 tc_gen_reloc (section, fixp)
4024 asection *section ATTRIBUTE_UNUSED;
4025 fixS *fixp;
4026 {
4027 arelent *rel;
4028 bfd_reloc_code_real_type code;
4029
4030 switch (fixp->fx_r_type)
4031 {
4032 case BFD_RELOC_386_PLT32:
4033 case BFD_RELOC_386_GOT32:
4034 case BFD_RELOC_386_GOTOFF:
4035 case BFD_RELOC_386_GOTPC:
4036 case BFD_RELOC_RVA:
4037 case BFD_RELOC_VTABLE_ENTRY:
4038 case BFD_RELOC_VTABLE_INHERIT:
4039 code = fixp->fx_r_type;
4040 break;
4041 default:
4042 if (fixp->fx_pcrel)
4043 {
4044 switch (fixp->fx_size)
4045 {
4046 default:
4047 as_bad (_("can not do %d byte pc-relative relocation"),
4048 fixp->fx_size);
4049 code = BFD_RELOC_32_PCREL;
4050 break;
4051 case 1: code = BFD_RELOC_8_PCREL; break;
4052 case 2: code = BFD_RELOC_16_PCREL; break;
4053 case 4: code = BFD_RELOC_32_PCREL; break;
4054 }
4055 }
4056 else
4057 {
4058 switch (fixp->fx_size)
4059 {
4060 default:
4061 as_bad (_("can not do %d byte relocation"), fixp->fx_size);
4062 code = BFD_RELOC_32;
4063 break;
4064 case 1: code = BFD_RELOC_8; break;
4065 case 2: code = BFD_RELOC_16; break;
4066 case 4: code = BFD_RELOC_32; break;
4067 }
4068 }
4069 break;
4070 }
4071
4072 if (code == BFD_RELOC_32
4073 && GOT_symbol
4074 && fixp->fx_addsy == GOT_symbol)
4075 code = BFD_RELOC_386_GOTPC;
4076
4077 rel = (arelent *) xmalloc (sizeof (arelent));
4078 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
4079 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
4080
4081 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
4082 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
4083 vtable entry to be used in the relocation's section offset. */
4084 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
4085 rel->address = fixp->fx_offset;
4086
4087 if (fixp->fx_pcrel)
4088 rel->addend = fixp->fx_addnumber;
4089 else
4090 rel->addend = 0;
4091
4092 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
4093 if (rel->howto == NULL)
4094 {
4095 as_bad_where (fixp->fx_file, fixp->fx_line,
4096 _("cannot represent relocation type %s"),
4097 bfd_get_reloc_code_name (code));
4098 /* Set howto to a garbage value so that we can keep going. */
4099 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
4100 assert (rel->howto != NULL);
4101 }
4102
4103 return rel;
4104 }
4105
4106 #else /* ! BFD_ASSEMBLER */
4107
4108 #if (defined(OBJ_AOUT) | defined(OBJ_BOUT))
4109 void
4110 tc_aout_fix_to_chars (where, fixP, segment_address_in_file)
4111 char *where;
4112 fixS *fixP;
4113 relax_addressT segment_address_in_file;
4114 {
4115 /* In: length of relocation (or of address) in chars: 1, 2 or 4.
4116 Out: GNU LD relocation length code: 0, 1, or 2. */
4117
4118 static const unsigned char nbytes_r_length[] = { 42, 0, 1, 42, 2 };
4119 long r_symbolnum;
4120
4121 know (fixP->fx_addsy != NULL);
4122
4123 md_number_to_chars (where,
4124 (valueT) (fixP->fx_frag->fr_address
4125 + fixP->fx_where - segment_address_in_file),
4126 4);
4127
4128 r_symbolnum = (S_IS_DEFINED (fixP->fx_addsy)
4129 ? S_GET_TYPE (fixP->fx_addsy)
4130 : fixP->fx_addsy->sy_number);
4131
4132 where[6] = (r_symbolnum >> 16) & 0x0ff;
4133 where[5] = (r_symbolnum >> 8) & 0x0ff;
4134 where[4] = r_symbolnum & 0x0ff;
4135 where[7] = ((((!S_IS_DEFINED (fixP->fx_addsy)) << 3) & 0x08)
4136 | ((nbytes_r_length[fixP->fx_size] << 1) & 0x06)
4137 | (((fixP->fx_pcrel << 0) & 0x01) & 0x0f));
4138 }
4139
4140 #endif /* OBJ_AOUT or OBJ_BOUT. */
4141
4142 #if defined (I386COFF)
4143
4144 short
4145 tc_coff_fix2rtype (fixP)
4146 fixS *fixP;
4147 {
4148 if (fixP->fx_r_type == R_IMAGEBASE)
4149 return R_IMAGEBASE;
4150
4151 return (fixP->fx_pcrel ?
4152 (fixP->fx_size == 1 ? R_PCRBYTE :
4153 fixP->fx_size == 2 ? R_PCRWORD :
4154 R_PCRLONG) :
4155 (fixP->fx_size == 1 ? R_RELBYTE :
4156 fixP->fx_size == 2 ? R_RELWORD :
4157 R_DIR32));
4158 }
4159
4160 int
4161 tc_coff_sizemachdep (frag)
4162 fragS *frag;
4163 {
4164 if (frag->fr_next)
4165 return (frag->fr_next->fr_address - frag->fr_address);
4166 else
4167 return 0;
4168 }
4169
4170 #endif /* I386COFF */
4171
4172 #endif /* ! BFD_ASSEMBLER */
4173 \f
4174 /* Parse operands using Intel syntax. This implements a recursive descent
4175 parser based on the BNF grammar published in Appendix B of the MASM 6.1
4176 Programmer's Guide.
4177
4178 FIXME: We do not recognize the full operand grammar defined in the MASM
4179 documentation. In particular, all the structure/union and
4180 high-level macro operands are missing.
4181
4182 Uppercase words are terminals, lower case words are non-terminals.
4183 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
4184 bars '|' denote choices. Most grammar productions are implemented in
4185 functions called 'intel_<production>'.
4186
4187 Initial production is 'expr'.
4188
4189
4190 addOp + | -
4191
4192 alpha [a-zA-Z]
4193
4194 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
4195
4196 constant digits [[ radixOverride ]]
4197
4198 dataType BYTE | WORD | DWORD | QWORD | XWORD
4199
4200 digits decdigit
4201 | digits decdigit
4202 | digits hexdigit
4203
4204 decdigit [0-9]
4205
4206 e05 e05 addOp e06
4207 | e06
4208
4209 e06 e06 mulOp e09
4210 | e09
4211
4212 e09 OFFSET e10
4213 | e09 PTR e10
4214 | e09 : e10
4215 | e10
4216
4217 e10 e10 [ expr ]
4218 | e11
4219
4220 e11 ( expr )
4221 | [ expr ]
4222 | constant
4223 | dataType
4224 | id
4225 | $
4226 | register
4227
4228 => expr SHORT e05
4229 | e05
4230
4231 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
4232 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
4233
4234 hexdigit a | b | c | d | e | f
4235 | A | B | C | D | E | F
4236
4237 id alpha
4238 | id alpha
4239 | id decdigit
4240
4241 mulOp * | / | MOD
4242
4243 quote " | '
4244
4245 register specialRegister
4246 | gpRegister
4247 | byteRegister
4248
4249 segmentRegister CS | DS | ES | FS | GS | SS
4250
4251 specialRegister CR0 | CR2 | CR3
4252 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
4253 | TR3 | TR4 | TR5 | TR6 | TR7
4254
4255
4256 We simplify the grammar in obvious places (e.g., register parsing is
4257 done by calling parse_register) and eliminate immediate left recursion
4258 to implement a recursive-descent parser.
4259
4260 expr SHORT e05
4261 | e05
4262
4263 e05 e06 e05'
4264
4265 e05' addOp e06 e05'
4266 | Empty
4267
4268 e06 e09 e06'
4269
4270 e06' mulOp e09 e06'
4271 | Empty
4272
4273 e09 OFFSET e10 e09'
4274 | e10 e09'
4275
4276 e09' PTR e10 e09'
4277 | : e10 e09'
4278 | Empty
4279
4280 e10 e11 e10'
4281
4282 e10' [ expr ] e10'
4283 | Empty
4284
4285 e11 ( expr )
4286 | [ expr ]
4287 | BYTE
4288 | WORD
4289 | DWORD
4290 | QWORD
4291 | XWORD
4292 | .
4293 | $
4294 | register
4295 | id
4296 | constant */
4297
4298 /* Parsing structure for the intel syntax parser. Used to implement the
4299 semantic actions for the operand grammar. */
4300 struct intel_parser_s
4301 {
4302 char *op_string; /* The string being parsed. */
4303 int got_a_float; /* Whether the operand is a float. */
4304 int op_modifier; /* Operand modifier. */
4305 int is_mem; /* 1 if operand is memory reference. */
4306 const reg_entry *reg; /* Last register reference found. */
4307 char *disp; /* Displacement string being built. */
4308 };
4309
4310 static struct intel_parser_s intel_parser;
4311
4312 /* Token structure for parsing intel syntax. */
4313 struct intel_token
4314 {
4315 int code; /* Token code. */
4316 const reg_entry *reg; /* Register entry for register tokens. */
4317 char *str; /* String representation. */
4318 };
4319
4320 static struct intel_token cur_token, prev_token;
4321
4322 /* Token codes for the intel parser. */
4323 #define T_NIL -1
4324 #define T_CONST 1
4325 #define T_REG 2
4326 #define T_BYTE 3
4327 #define T_WORD 4
4328 #define T_DWORD 5
4329 #define T_QWORD 6
4330 #define T_XWORD 7
4331 #define T_SHORT 8
4332 #define T_OFFSET 9
4333 #define T_PTR 10
4334 #define T_ID 11
4335
4336 /* Prototypes for intel parser functions. */
4337 static int intel_match_token PARAMS ((int code));
4338 static void intel_get_token PARAMS ((void));
4339 static void intel_putback_token PARAMS ((void));
4340 static int intel_expr PARAMS ((void));
4341 static int intel_e05 PARAMS ((void));
4342 static int intel_e05_1 PARAMS ((void));
4343 static int intel_e06 PARAMS ((void));
4344 static int intel_e06_1 PARAMS ((void));
4345 static int intel_e09 PARAMS ((void));
4346 static int intel_e09_1 PARAMS ((void));
4347 static int intel_e10 PARAMS ((void));
4348 static int intel_e10_1 PARAMS ((void));
4349 static int intel_e11 PARAMS ((void));
4350
4351
4352 static int
4353 i386_intel_operand (operand_string, got_a_float)
4354 char *operand_string;
4355 int got_a_float;
4356 {
4357 int ret;
4358 char *p;
4359
4360 /* Initialize token holders. */
4361 cur_token.code = prev_token.code = T_NIL;
4362 cur_token.reg = prev_token.reg = NULL;
4363 cur_token.str = prev_token.str = NULL;
4364
4365 /* Initialize parser structure. */
4366 p = intel_parser.op_string = (char *)malloc (strlen (operand_string) + 1);
4367 if (p == NULL)
4368 abort ();
4369 strcpy (intel_parser.op_string, operand_string);
4370 intel_parser.got_a_float = got_a_float;
4371 intel_parser.op_modifier = -1;
4372 intel_parser.is_mem = 0;
4373 intel_parser.reg = NULL;
4374 intel_parser.disp = (char *)malloc (strlen (operand_string) + 1);
4375 if (intel_parser.disp == NULL)
4376 abort ();
4377 intel_parser.disp[0] = '\0';
4378
4379 /* Read the first token and start the parser. */
4380 intel_get_token ();
4381 ret = intel_expr ();
4382
4383 if (ret)
4384 {
4385 /* If we found a memory reference, hand it over to i386_displacement
4386 to fill in the rest of the operand fields. */
4387 if (intel_parser.is_mem)
4388 {
4389 if ((i.mem_operands == 1
4390 && (current_templates->start->opcode_modifier & IsString) == 0)
4391 || i.mem_operands == 2)
4392 {
4393 as_bad (_("too many memory references for '%s'"),
4394 current_templates->start->name);
4395 ret = 0;
4396 }
4397 else
4398 {
4399 char *s = intel_parser.disp;
4400 i.mem_operands++;
4401
4402 /* Add the displacement expression. */
4403 if (*s != '\0')
4404 ret = i386_displacement (s, s + strlen (s))
4405 && i386_index_check (s);
4406 }
4407 }
4408
4409 /* Constant and OFFSET expressions are handled by i386_immediate. */
4410 else if (intel_parser.op_modifier == OFFSET_FLAT
4411 || intel_parser.reg == NULL)
4412 ret = i386_immediate (intel_parser.disp);
4413 }
4414
4415 free (p);
4416 free (intel_parser.disp);
4417
4418 return ret;
4419 }
4420
4421
4422 /* expr SHORT e05
4423 | e05 */
4424 static int
4425 intel_expr ()
4426 {
4427 /* expr SHORT e05 */
4428 if (cur_token.code == T_SHORT)
4429 {
4430 intel_parser.op_modifier = SHORT;
4431 intel_match_token (T_SHORT);
4432
4433 return (intel_e05 ());
4434 }
4435
4436 /* expr e05 */
4437 else
4438 return intel_e05 ();
4439 }
4440
4441
4442 /* e05 e06 e05'
4443
4444 e05' addOp e06 e05'
4445 | Empty */
4446 static int
4447 intel_e05 ()
4448 {
4449 return (intel_e06 () && intel_e05_1 ());
4450 }
4451
4452 static int
4453 intel_e05_1 ()
4454 {
4455 /* e05' addOp e06 e05' */
4456 if (cur_token.code == '+' || cur_token.code == '-')
4457 {
4458 strcat (intel_parser.disp, cur_token.str);
4459 intel_match_token (cur_token.code);
4460
4461 return (intel_e06 () && intel_e05_1 ());
4462 }
4463
4464 /* e05' Empty */
4465 else
4466 return 1;
4467 }
4468
4469
4470 /* e06 e09 e06'
4471
4472 e06' mulOp e09 e06'
4473 | Empty */
4474 static int
4475 intel_e06 ()
4476 {
4477 return (intel_e09 () && intel_e06_1 ());
4478 }
4479
4480 static int
4481 intel_e06_1 ()
4482 {
4483 /* e06' mulOp e09 e06' */
4484 if (cur_token.code == '*' || cur_token.code == '/')
4485 {
4486 strcat (intel_parser.disp, cur_token.str);
4487 intel_match_token (cur_token.code);
4488
4489 return (intel_e09 () && intel_e06_1 ());
4490 }
4491
4492 /* e06' Empty */
4493 else
4494 return 1;
4495 }
4496
4497
4498 /* e09 OFFSET e10 e09'
4499 | e10 e09'
4500
4501 e09' PTR e10 e09'
4502 | : e10 e09'
4503 | Empty */
4504 static int
4505 intel_e09 ()
4506 {
4507 /* e09 OFFSET e10 e09' */
4508 if (cur_token.code == T_OFFSET)
4509 {
4510 intel_parser.is_mem = 0;
4511 intel_parser.op_modifier = OFFSET_FLAT;
4512 intel_match_token (T_OFFSET);
4513
4514 return (intel_e10 () && intel_e09_1 ());
4515 }
4516
4517 /* e09 e10 e09' */
4518 else
4519 return (intel_e10 () && intel_e09_1 ());
4520 }
4521
4522 static int
4523 intel_e09_1 ()
4524 {
4525 /* e09' PTR e10 e09' */
4526 if (cur_token.code == T_PTR)
4527 {
4528 if (prev_token.code == T_BYTE)
4529 i.suffix = BYTE_MNEM_SUFFIX;
4530
4531 else if (prev_token.code == T_WORD)
4532 {
4533 if (intel_parser.got_a_float == 2) /* "fi..." */
4534 i.suffix = SHORT_MNEM_SUFFIX;
4535 else
4536 i.suffix = WORD_MNEM_SUFFIX;
4537 }
4538
4539 else if (prev_token.code == T_DWORD)
4540 {
4541 if (intel_parser.got_a_float == 1) /* "f..." */
4542 i.suffix = SHORT_MNEM_SUFFIX;
4543 else
4544 i.suffix = LONG_MNEM_SUFFIX;
4545 }
4546
4547 else if (prev_token.code == T_QWORD)
4548 i.suffix = DWORD_MNEM_SUFFIX;
4549
4550 else if (prev_token.code == T_XWORD)
4551 i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
4552
4553 else
4554 {
4555 as_bad (_("Unknown operand modifier `%s'\n"), prev_token.str);
4556 return 0;
4557 }
4558
4559 intel_match_token (T_PTR);
4560
4561 return (intel_e10 () && intel_e09_1 ());
4562 }
4563
4564 /* e09 : e10 e09' */
4565 else if (cur_token.code == ':')
4566 {
4567 intel_parser.is_mem = 1;
4568
4569 return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
4570 }
4571
4572 /* e09' Empty */
4573 else
4574 return 1;
4575 }
4576
4577 /* e10 e11 e10'
4578
4579 e10' [ expr ] e10'
4580 | Empty */
4581 static int
4582 intel_e10 ()
4583 {
4584 return (intel_e11 () && intel_e10_1 ());
4585 }
4586
4587 static int
4588 intel_e10_1 ()
4589 {
4590 /* e10' [ expr ] e10' */
4591 if (cur_token.code == '[')
4592 {
4593 intel_match_token ('[');
4594 intel_parser.is_mem = 1;
4595
4596 /* Add a '+' to the displacement string if necessary. */
4597 if (*intel_parser.disp != '\0')
4598 strcat (intel_parser.disp, "+");
4599
4600 return (intel_expr () && intel_match_token (']') && intel_e10_1 ());
4601 }
4602
4603 /* e10' Empty */
4604 else
4605 return 1;
4606 }
4607
4608
4609 /* e11 ( expr )
4610 | [ expr ]
4611 | BYTE
4612 | WORD
4613 | DWORD
4614 | QWORD
4615 | XWORD
4616 | $
4617 | .
4618 | register
4619 | id
4620 | constant */
4621 static int
4622 intel_e11 ()
4623 {
4624 /* e11 ( expr ) */
4625 if (cur_token.code == '(')
4626 {
4627 intel_match_token ('(');
4628 strcat (intel_parser.disp, "(");
4629
4630 if (intel_expr () && intel_match_token (')'))
4631 {
4632 strcat (intel_parser.disp, ")");
4633 return 1;
4634 }
4635 else
4636 return 0;
4637 }
4638
4639 /* e11 [ expr ] */
4640 else if (cur_token.code == '[')
4641 {
4642 intel_match_token ('[');
4643 intel_parser.is_mem = 1;
4644
4645 /* Operands for jump/call inside brackets denote absolute addresses. */
4646 if (current_templates->start->opcode_modifier & Jump
4647 || current_templates->start->opcode_modifier & JumpDword
4648 || current_templates->start->opcode_modifier & JumpByte
4649 || current_templates->start->opcode_modifier & JumpInterSegment)
4650 i.types[this_operand] |= JumpAbsolute;
4651
4652 /* Add a '+' to the displacement string if necessary. */
4653 if (*intel_parser.disp != '\0')
4654 strcat (intel_parser.disp, "+");
4655
4656 return (intel_expr () && intel_match_token (']'));
4657 }
4658
4659 /* e11 BYTE
4660 | WORD
4661 | DWORD
4662 | QWORD
4663 | XWORD */
4664 else if (cur_token.code == T_BYTE
4665 || cur_token.code == T_WORD
4666 || cur_token.code == T_DWORD
4667 || cur_token.code == T_QWORD
4668 || cur_token.code == T_XWORD)
4669 {
4670 intel_match_token (cur_token.code);
4671
4672 return 1;
4673 }
4674
4675 /* e11 $
4676 | . */
4677 else if (cur_token.code == '$' || cur_token.code == '.')
4678 {
4679 strcat (intel_parser.disp, cur_token.str);
4680 intel_match_token (cur_token.code);
4681 intel_parser.is_mem = 1;
4682
4683 return 1;
4684 }
4685
4686 /* e11 register */
4687 else if (cur_token.code == T_REG)
4688 {
4689 const reg_entry *reg = intel_parser.reg = cur_token.reg;
4690
4691 intel_match_token (T_REG);
4692
4693 /* Check for segment change. */
4694 if (cur_token.code == ':')
4695 {
4696 if (reg->reg_type & (SReg2 | SReg3))
4697 {
4698 switch (reg->reg_num)
4699 {
4700 case 0:
4701 i.seg[i.mem_operands] = &es;
4702 break;
4703 case 1:
4704 i.seg[i.mem_operands] = &cs;
4705 break;
4706 case 2:
4707 i.seg[i.mem_operands] = &ss;
4708 break;
4709 case 3:
4710 i.seg[i.mem_operands] = &ds;
4711 break;
4712 case 4:
4713 i.seg[i.mem_operands] = &fs;
4714 break;
4715 case 5:
4716 i.seg[i.mem_operands] = &gs;
4717 break;
4718 }
4719 }
4720 else
4721 {
4722 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
4723 return 0;
4724 }
4725 }
4726
4727 /* Not a segment register. Check for register scaling. */
4728 else if (cur_token.code == '*')
4729 {
4730 if (!intel_parser.is_mem)
4731 {
4732 as_bad (_("Register scaling only allowed in memory operands."));
4733 return 0;
4734 }
4735
4736 /* What follows must be a valid scale. */
4737 if (intel_match_token ('*')
4738 && strchr ("01248", *cur_token.str))
4739 {
4740 i.index_reg = reg;
4741 i.types[this_operand] |= BaseIndex;
4742
4743 /* Set the scale after setting the register (otherwise,
4744 i386_scale will complain) */
4745 i386_scale (cur_token.str);
4746 intel_match_token (T_CONST);
4747 }
4748 else
4749 {
4750 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
4751 cur_token.str);
4752 return 0;
4753 }
4754 }
4755
4756 /* No scaling. If this is a memory operand, the register is either a
4757 base register (first occurrence) or an index register (second
4758 occurrence). */
4759 else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3)))
4760 {
4761 if (i.base_reg && i.index_reg)
4762 {
4763 as_bad (_("Too many register references in memory operand.\n"));
4764 return 0;
4765 }
4766
4767 if (i.base_reg == NULL)
4768 i.base_reg = reg;
4769 else
4770 i.index_reg = reg;
4771
4772 i.types[this_operand] |= BaseIndex;
4773 }
4774
4775 /* Offset modifier. Add the register to the displacement string to be
4776 parsed as an immediate expression after we're done. */
4777 else if (intel_parser.op_modifier == OFFSET_FLAT)
4778 strcat (intel_parser.disp, reg->reg_name);
4779
4780 /* It's neither base nor index nor offset. */
4781 else
4782 {
4783 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
4784 i.op[this_operand].regs = reg;
4785 i.reg_operands++;
4786 }
4787
4788 /* Since registers are not part of the displacement string (except
4789 when we're parsing offset operands), we may need to remove any
4790 preceding '+' from the displacement string. */
4791 if (*intel_parser.disp != '\0'
4792 && intel_parser.op_modifier != OFFSET_FLAT)
4793 {
4794 char *s = intel_parser.disp;
4795 s += strlen (s) - 1;
4796 if (*s == '+')
4797 *s = '\0';
4798 }
4799
4800 return 1;
4801 }
4802
4803 /* e11 id */
4804 else if (cur_token.code == T_ID)
4805 {
4806 /* Add the identifier to the displacement string. */
4807 strcat (intel_parser.disp, cur_token.str);
4808 intel_match_token (T_ID);
4809
4810 /* The identifier represents a memory reference only if it's not
4811 preceded by an offset modifier. */
4812 if (intel_parser.op_modifier != OFFSET_FLAT
4813 && intel_parser.op_modifier != FLAT)
4814 intel_parser.is_mem = 1;
4815
4816 return 1;
4817 }
4818
4819 /* e11 constant */
4820 else if (cur_token.code == T_CONST
4821 || cur_token.code == '-'
4822 || cur_token.code == '+')
4823 {
4824 char *save_str;
4825
4826 /* Allow constants that start with `+' or `-'. */
4827 if (cur_token.code == '-' || cur_token.code == '+')
4828 {
4829 strcat (intel_parser.disp, cur_token.str);
4830 intel_match_token (cur_token.code);
4831 if (cur_token.code != T_CONST)
4832 {
4833 as_bad (_("Syntax error. Expecting a constant. Got `%s'.\n"),
4834 cur_token.str);
4835 return 0;
4836 }
4837 }
4838
4839 save_str = (char *)malloc (strlen (cur_token.str) + 1);
4840 if (save_str == NULL)
4841 abort();
4842 strcpy (save_str, cur_token.str);
4843
4844 /* Get the next token to check for register scaling. */
4845 intel_match_token (cur_token.code);
4846
4847 /* Check if this constant is a scaling factor for an index register. */
4848 if (cur_token.code == '*')
4849 {
4850 if (intel_match_token ('*') && cur_token.code == T_REG)
4851 {
4852 if (!intel_parser.is_mem)
4853 {
4854 as_bad (_("Register scaling only allowed in memory operands."));
4855 return 0;
4856 }
4857
4858 /* The constant is followed by `* reg', so it must be
4859 a valid scale. */
4860 if (strchr ("01248", *save_str))
4861 {
4862 i.index_reg = cur_token.reg;
4863 i.types[this_operand] |= BaseIndex;
4864
4865 /* Set the scale after setting the register (otherwise,
4866 i386_scale will complain) */
4867 i386_scale (save_str);
4868 intel_match_token (T_REG);
4869
4870 /* Since registers are not part of the displacement
4871 string, we may need to remove any preceding '+' from
4872 the displacement string. */
4873 if (*intel_parser.disp != '\0')
4874 {
4875 char *s = intel_parser.disp;
4876 s += strlen (s) - 1;
4877 if (*s == '+')
4878 *s = '\0';
4879 }
4880
4881 free (save_str);
4882
4883 return 1;
4884 }
4885 else
4886 return 0;
4887 }
4888
4889 /* The constant was not used for register scaling. Since we have
4890 already consumed the token following `*' we now need to put it
4891 back in the stream. */
4892 else
4893 intel_putback_token ();
4894 }
4895
4896 /* Add the constant to the displacement string. */
4897 strcat (intel_parser.disp, save_str);
4898 free (save_str);
4899
4900 return 1;
4901 }
4902
4903
4904 as_bad (_("Unrecognized token '%s'"), cur_token.str);
4905 return 0;
4906 }
4907
4908
4909 /* Match the given token against cur_token. If they match, read the next
4910 token from the operand string. */
4911 static int
4912 intel_match_token (code)
4913 int code;
4914 {
4915 if (cur_token.code == code)
4916 {
4917 intel_get_token ();
4918 return 1;
4919 }
4920 else
4921 {
4922 as_bad (_("Unexpected token `%s'\n"), cur_token.str);
4923 return 0;
4924 }
4925 }
4926
4927
4928 /* Read a new token from intel_parser.op_string and store it in cur_token. */
4929 static void
4930 intel_get_token ()
4931 {
4932 char *end_op;
4933 const reg_entry *reg;
4934 struct intel_token new_token;
4935
4936 new_token.code = T_NIL;
4937 new_token.reg = NULL;
4938 new_token.str = NULL;
4939
4940 /* Free the memory allocated to the previous token and move
4941 cur_token to prev_token. */
4942 if (prev_token.str)
4943 free (prev_token.str);
4944
4945 prev_token = cur_token;
4946
4947 /* Skip whitespace. */
4948 while (is_space_char (*intel_parser.op_string))
4949 intel_parser.op_string++;
4950
4951 /* Return an empty token if we find nothing else on the line. */
4952 if (*intel_parser.op_string == '\0')
4953 {
4954 cur_token = new_token;
4955 return;
4956 }
4957
4958 /* The new token cannot be larger than the remainder of the operand
4959 string. */
4960 new_token.str = (char *)malloc (strlen (intel_parser.op_string) + 1);
4961 if (new_token.str == NULL)
4962 abort();
4963 new_token.str[0] = '\0';
4964
4965 if (strchr ("0123456789", *intel_parser.op_string))
4966 {
4967 char *p = new_token.str;
4968 char *q = intel_parser.op_string;
4969 new_token.code = T_CONST;
4970
4971 /* Allow any kind of identifier char to encompass floating point and
4972 hexadecimal numbers. */
4973 while (is_identifier_char (*q))
4974 *p++ = *q++;
4975 *p = '\0';
4976
4977 /* Recognize special symbol names [0-9][bf]. */
4978 if (strlen (intel_parser.op_string) == 2
4979 && (intel_parser.op_string[1] == 'b'
4980 || intel_parser.op_string[1] == 'f'))
4981 new_token.code = T_ID;
4982 }
4983
4984 else if (strchr ("+-/*:[]()", *intel_parser.op_string))
4985 {
4986 new_token.code = *intel_parser.op_string;
4987 new_token.str[0] = *intel_parser.op_string;
4988 new_token.str[1] = '\0';
4989 }
4990
4991 else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
4992 && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
4993 {
4994 new_token.code = T_REG;
4995 new_token.reg = reg;
4996
4997 if (*intel_parser.op_string == REGISTER_PREFIX)
4998 {
4999 new_token.str[0] = REGISTER_PREFIX;
5000 new_token.str[1] = '\0';
5001 }
5002
5003 strcat (new_token.str, reg->reg_name);
5004 }
5005
5006 else if (is_identifier_char (*intel_parser.op_string))
5007 {
5008 char *p = new_token.str;
5009 char *q = intel_parser.op_string;
5010
5011 /* A '.' or '$' followed by an identifier char is an identifier.
5012 Otherwise, it's operator '.' followed by an expression. */
5013 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
5014 {
5015 new_token.code = *q;
5016 new_token.str[0] = *q;
5017 new_token.str[1] = '\0';
5018 }
5019 else
5020 {
5021 while (is_identifier_char (*q) || *q == '@')
5022 *p++ = *q++;
5023 *p = '\0';
5024
5025 if (strcasecmp (new_token.str, "BYTE") == 0)
5026 new_token.code = T_BYTE;
5027
5028 else if (strcasecmp (new_token.str, "WORD") == 0)
5029 new_token.code = T_WORD;
5030
5031 else if (strcasecmp (new_token.str, "DWORD") == 0)
5032 new_token.code = T_DWORD;
5033
5034 else if (strcasecmp (new_token.str, "QWORD") == 0)
5035 new_token.code = T_QWORD;
5036
5037 else if (strcasecmp (new_token.str, "XWORD") == 0)
5038 new_token.code = T_XWORD;
5039
5040 else if (strcasecmp (new_token.str, "PTR") == 0)
5041 new_token.code = T_PTR;
5042
5043 else if (strcasecmp (new_token.str, "SHORT") == 0)
5044 new_token.code = T_SHORT;
5045
5046 else if (strcasecmp (new_token.str, "OFFSET") == 0)
5047 {
5048 new_token.code = T_OFFSET;
5049
5050 /* ??? This is not mentioned in the MASM grammar but gcc
5051 makes use of it with -mintel-syntax. OFFSET may be
5052 followed by FLAT: */
5053 if (strncasecmp (q, " FLAT:", 6) == 0)
5054 strcat (new_token.str, " FLAT:");
5055 }
5056
5057 /* ??? This is not mentioned in the MASM grammar. */
5058 else if (strcasecmp (new_token.str, "FLAT") == 0)
5059 new_token.code = T_OFFSET;
5060
5061 else
5062 new_token.code = T_ID;
5063 }
5064 }
5065
5066 else
5067 as_bad (_("Unrecognized token `%s'\n"), intel_parser.op_string);
5068
5069 intel_parser.op_string += strlen (new_token.str);
5070 cur_token = new_token;
5071 }
5072
5073
5074 /* Put cur_token back into the token stream and make cur_token point to
5075 prev_token. */
5076 static void
5077 intel_putback_token ()
5078 {
5079 intel_parser.op_string -= strlen (cur_token.str);
5080 free (cur_token.str);
5081 cur_token = prev_token;
5082
5083 /* Forget prev_token. */
5084 prev_token.code = T_NIL;
5085 prev_token.reg = NULL;
5086 prev_token.str = NULL;
5087 }