* tc-i386.c (md_assemble): Refuse 's' and 'l' suffixes in the intel
[deliverable/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 "dwarf2dbg.h"
32 #include "opcode/i386.h"
33
34 #ifndef REGISTER_WARNINGS
35 #define REGISTER_WARNINGS 1
36 #endif
37
38 #ifndef INFER_ADDR_PREFIX
39 #define INFER_ADDR_PREFIX 1
40 #endif
41
42 #ifndef SCALE1_WHEN_NO_INDEX
43 /* Specifying a scale factor besides 1 when there is no index is
44 futile. eg. `mov (%ebx,2),%al' does exactly the same as
45 `mov (%ebx),%al'. To slavishly follow what the programmer
46 specified, set SCALE1_WHEN_NO_INDEX to 0. */
47 #define SCALE1_WHEN_NO_INDEX 1
48 #endif
49
50 #define true 1
51 #define false 0
52
53 static unsigned int mode_from_disp_size PARAMS ((unsigned int));
54 static int fits_in_signed_byte PARAMS ((offsetT));
55 static int fits_in_unsigned_byte PARAMS ((offsetT));
56 static int fits_in_unsigned_word PARAMS ((offsetT));
57 static int fits_in_signed_word PARAMS ((offsetT));
58 static int smallest_imm_type PARAMS ((offsetT));
59 static offsetT offset_in_range PARAMS ((offsetT, int));
60 static int add_prefix PARAMS ((unsigned int));
61 static void set_16bit_code_flag PARAMS ((int));
62 static void set_16bit_gcc_code_flag PARAMS ((int));
63 static void set_intel_syntax PARAMS ((int));
64 static void set_cpu_arch PARAMS ((int));
65
66 #ifdef BFD_ASSEMBLER
67 static bfd_reloc_code_real_type reloc
68 PARAMS ((int, int, bfd_reloc_code_real_type));
69 #endif
70
71 /* 'md_assemble ()' gathers together information and puts it into a
72 i386_insn. */
73
74 union i386_op
75 {
76 expressionS *disps;
77 expressionS *imms;
78 const reg_entry *regs;
79 };
80
81 struct _i386_insn
82 {
83 /* TM holds the template for the insn were currently assembling. */
84 template tm;
85
86 /* SUFFIX holds the instruction mnemonic suffix if given.
87 (e.g. 'l' for 'movl') */
88 char suffix;
89
90 /* OPERANDS gives the number of given operands. */
91 unsigned int operands;
92
93 /* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number
94 of given register, displacement, memory operands and immediate
95 operands. */
96 unsigned int reg_operands, disp_operands, mem_operands, imm_operands;
97
98 /* TYPES [i] is the type (see above #defines) which tells us how to
99 use OP[i] for the corresponding operand. */
100 unsigned int types[MAX_OPERANDS];
101
102 /* Displacement expression, immediate expression, or register for each
103 operand. */
104 union i386_op op[MAX_OPERANDS];
105
106 /* Relocation type for operand */
107 #ifdef BFD_ASSEMBLER
108 enum bfd_reloc_code_real disp_reloc[MAX_OPERANDS];
109 #else
110 int disp_reloc[MAX_OPERANDS];
111 #endif
112
113 /* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode
114 the base index byte below. */
115 const reg_entry *base_reg;
116 const reg_entry *index_reg;
117 unsigned int log2_scale_factor;
118
119 /* SEG gives the seg_entries of this insn. They are zero unless
120 explicit segment overrides are given. */
121 const seg_entry *seg[2];
122
123 /* PREFIX holds all the given prefix opcodes (usually null).
124 PREFIXES is the number of prefix opcodes. */
125 unsigned int prefixes;
126 unsigned char prefix[MAX_PREFIXES];
127
128 /* RM and SIB are the modrm byte and the sib byte where the
129 addressing modes of this insn are encoded. */
130
131 modrm_byte rm;
132 sib_byte sib;
133 };
134
135 typedef struct _i386_insn i386_insn;
136
137 /* List of chars besides those in app.c:symbol_chars that can start an
138 operand. Used to prevent the scrubber eating vital white-space. */
139 #ifdef LEX_AT
140 const char extra_symbol_chars[] = "*%-(@";
141 #else
142 const char extra_symbol_chars[] = "*%-(";
143 #endif
144
145 /* This array holds the chars that always start a comment. If the
146 pre-processor is disabled, these aren't very useful. */
147 #if defined (TE_I386AIX) || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) && ! defined (TE_LINUX) && !defined(TE_FreeBSD))
148 /* Putting '/' here makes it impossible to use the divide operator.
149 However, we need it for compatibility with SVR4 systems. */
150 const char comment_chars[] = "#/";
151 #define PREFIX_SEPARATOR '\\'
152 #else
153 const char comment_chars[] = "#";
154 #define PREFIX_SEPARATOR '/'
155 #endif
156
157 /* This array holds the chars that only start a comment at the beginning of
158 a line. If the line seems to have the form '# 123 filename'
159 .line and .file directives will appear in the pre-processed output.
160 Note that input_file.c hand checks for '#' at the beginning of the
161 first line of the input file. This is because the compiler outputs
162 #NO_APP at the beginning of its output.
163 Also note that comments started like this one will always work if
164 '/' isn't otherwise defined. */
165 #if defined (TE_I386AIX) || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) && ! defined (TE_LINUX) && !defined(TE_FreeBSD))
166 const char line_comment_chars[] = "";
167 #else
168 const char line_comment_chars[] = "/";
169 #endif
170
171 const char line_separator_chars[] = ";";
172
173 /* Chars that can be used to separate mant from exp in floating point
174 nums. */
175 const char EXP_CHARS[] = "eE";
176
177 /* Chars that mean this number is a floating point constant
178 As in 0f12.456
179 or 0d1.2345e12. */
180 const char FLT_CHARS[] = "fFdDxX";
181
182 /* Tables for lexical analysis. */
183 static char mnemonic_chars[256];
184 static char register_chars[256];
185 static char operand_chars[256];
186 static char identifier_chars[256];
187 static char digit_chars[256];
188
189 /* Lexical macros. */
190 #define is_mnemonic_char(x) (mnemonic_chars[(unsigned char) x])
191 #define is_operand_char(x) (operand_chars[(unsigned char) x])
192 #define is_register_char(x) (register_chars[(unsigned char) x])
193 #define is_space_char(x) ((x) == ' ')
194 #define is_identifier_char(x) (identifier_chars[(unsigned char) x])
195 #define is_digit_char(x) (digit_chars[(unsigned char) x])
196
197 /* All non-digit non-letter charcters that may occur in an operand. */
198 static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
199
200 /* md_assemble() always leaves the strings it's passed unaltered. To
201 effect this we maintain a stack of saved characters that we've smashed
202 with '\0's (indicating end of strings for various sub-fields of the
203 assembler instruction). */
204 static char save_stack[32];
205 static char *save_stack_p;
206 #define END_STRING_AND_SAVE(s) \
207 do { *save_stack_p++ = *(s); *(s) = '\0'; } while (0)
208 #define RESTORE_END_STRING(s) \
209 do { *(s) = *--save_stack_p; } while (0)
210
211 /* The instruction we're assembling. */
212 static i386_insn i;
213
214 /* Possible templates for current insn. */
215 static const templates *current_templates;
216
217 /* Per instruction expressionS buffers: 2 displacements & 2 immediate max. */
218 static expressionS disp_expressions[2], im_expressions[2];
219
220 /* Current operand we are working on. */
221 static int this_operand;
222
223 /* 1 if we're writing 16-bit code,
224 0 if 32-bit. */
225 static int flag_16bit_code;
226
227 /* 1 for intel syntax,
228 0 if att syntax. */
229 static int intel_syntax = 0;
230
231 /* 1 if register prefix % not required. */
232 static int allow_naked_reg = 0;
233
234 /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
235 leave, push, and pop instructions so that gcc has the same stack
236 frame as in 32 bit mode. */
237 static char stackop_size = '\0';
238
239 /* Non-zero to quieten some warnings. */
240 static int quiet_warnings = 0;
241
242 /* CPU name. */
243 static const char *cpu_arch_name = NULL;
244
245 /* CPU feature flags. */
246 static unsigned int cpu_arch_flags = 0;
247
248 /* Interface to relax_segment.
249 There are 2 relax states for 386 jump insns: one for conditional &
250 one for unconditional jumps. This is because these two types of
251 jumps add different sizes to frags when we're figuring out what
252 sort of jump to choose to reach a given label. */
253
254 /* Types. */
255 #define COND_JUMP 1
256 #define UNCOND_JUMP 2
257 /* Sizes. */
258 #define CODE16 1
259 #define SMALL 0
260 #define SMALL16 (SMALL|CODE16)
261 #define BIG 2
262 #define BIG16 (BIG|CODE16)
263
264 #ifndef INLINE
265 #ifdef __GNUC__
266 #define INLINE __inline__
267 #else
268 #define INLINE
269 #endif
270 #endif
271
272 #define ENCODE_RELAX_STATE(type,size) \
273 ((relax_substateT) ((type<<2) | (size)))
274 #define SIZE_FROM_RELAX_STATE(s) \
275 ( (((s) & 0x3) == BIG ? 4 : (((s) & 0x3) == BIG16 ? 2 : 1)) )
276
277 /* This table is used by relax_frag to promote short jumps to long
278 ones where necessary. SMALL (short) jumps may be promoted to BIG
279 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
280 don't allow a short jump in a 32 bit code segment to be promoted to
281 a 16 bit offset jump because it's slower (requires data size
282 prefix), and doesn't work, unless the destination is in the bottom
283 64k of the code segment (The top 16 bits of eip are zeroed). */
284
285 const relax_typeS md_relax_table[] =
286 {
287 /* The fields are:
288 1) most positive reach of this state,
289 2) most negative reach of this state,
290 3) how many bytes this mode will add to the size of the current frag
291 4) which index into the table to try if we can't fit into this one. */
292 {1, 1, 0, 0},
293 {1, 1, 0, 0},
294 {1, 1, 0, 0},
295 {1, 1, 0, 0},
296
297 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
298 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
299 /* dword conditionals adds 4 bytes to frag:
300 1 extra opcode byte, 3 extra displacement bytes. */
301 {0, 0, 4, 0},
302 /* word conditionals add 2 bytes to frag:
303 1 extra opcode byte, 1 extra displacement byte. */
304 {0, 0, 2, 0},
305
306 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
307 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
308 /* dword jmp adds 3 bytes to frag:
309 0 extra opcode bytes, 3 extra displacement bytes. */
310 {0, 0, 3, 0},
311 /* word jmp adds 1 byte to frag:
312 0 extra opcode bytes, 1 extra displacement byte. */
313 {0, 0, 1, 0}
314
315 };
316
317 static const arch_entry cpu_arch[] = {
318 {"i8086", Cpu086 },
319 {"i186", Cpu086|Cpu186 },
320 {"i286", Cpu086|Cpu186|Cpu286 },
321 {"i386", Cpu086|Cpu186|Cpu286|Cpu386 },
322 {"i486", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486 },
323 {"i586", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
324 {"i686", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
325 {"pentium", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
326 {"pentiumpro",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
327 {"k6", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX|Cpu3dnow },
328 {"athlon", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|Cpu3dnow },
329 {NULL, 0 }
330 };
331
332 void
333 i386_align_code (fragP, count)
334 fragS *fragP;
335 int count;
336 {
337 /* Various efficient no-op patterns for aligning code labels.
338 Note: Don't try to assemble the instructions in the comments.
339 0L and 0w are not legal. */
340 static const char f32_1[] =
341 {0x90}; /* nop */
342 static const char f32_2[] =
343 {0x89,0xf6}; /* movl %esi,%esi */
344 static const char f32_3[] =
345 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
346 static const char f32_4[] =
347 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
348 static const char f32_5[] =
349 {0x90, /* nop */
350 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
351 static const char f32_6[] =
352 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
353 static const char f32_7[] =
354 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
355 static const char f32_8[] =
356 {0x90, /* nop */
357 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
358 static const char f32_9[] =
359 {0x89,0xf6, /* movl %esi,%esi */
360 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
361 static const char f32_10[] =
362 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
363 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
364 static const char f32_11[] =
365 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
366 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
367 static const char f32_12[] =
368 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
369 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
370 static const char f32_13[] =
371 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
372 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
373 static const char f32_14[] =
374 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
375 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
376 static const char f32_15[] =
377 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
378 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
379 static const char f16_3[] =
380 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
381 static const char f16_4[] =
382 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
383 static const char f16_5[] =
384 {0x90, /* nop */
385 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
386 static const char f16_6[] =
387 {0x89,0xf6, /* mov %si,%si */
388 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
389 static const char f16_7[] =
390 {0x8d,0x74,0x00, /* lea 0(%si),%si */
391 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
392 static const char f16_8[] =
393 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
394 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
395 static const char *const f32_patt[] = {
396 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
397 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
398 };
399 static const char *const f16_patt[] = {
400 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
401 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
402 };
403
404 if (count > 0 && count <= 15)
405 {
406 if (flag_16bit_code)
407 {
408 memcpy (fragP->fr_literal + fragP->fr_fix,
409 f16_patt[count - 1], count);
410 if (count > 8)
411 /* Adjust jump offset. */
412 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
413 }
414 else
415 memcpy (fragP->fr_literal + fragP->fr_fix,
416 f32_patt[count - 1], count);
417 fragP->fr_var = count;
418 }
419 }
420
421 static char *output_invalid PARAMS ((int c));
422 static int i386_operand PARAMS ((char *operand_string));
423 static int i386_intel_operand PARAMS ((char *operand_string, int got_a_float));
424 static const reg_entry *parse_register PARAMS ((char *reg_string,
425 char **end_op));
426
427 #ifndef I386COFF
428 static void s_bss PARAMS ((int));
429 #endif
430
431 symbolS *GOT_symbol; /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
432
433 static INLINE unsigned int
434 mode_from_disp_size (t)
435 unsigned int t;
436 {
437 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32)) ? 2 : 0;
438 }
439
440 static INLINE int
441 fits_in_signed_byte (num)
442 offsetT num;
443 {
444 return (num >= -128) && (num <= 127);
445 }
446
447 static INLINE int
448 fits_in_unsigned_byte (num)
449 offsetT num;
450 {
451 return (num & 0xff) == num;
452 }
453
454 static INLINE int
455 fits_in_unsigned_word (num)
456 offsetT num;
457 {
458 return (num & 0xffff) == num;
459 }
460
461 static INLINE int
462 fits_in_signed_word (num)
463 offsetT num;
464 {
465 return (-32768 <= num) && (num <= 32767);
466 }
467
468 static int
469 smallest_imm_type (num)
470 offsetT num;
471 {
472 if (cpu_arch_flags != 0
473 && cpu_arch_flags != (Cpu086 | Cpu186 | Cpu286 | Cpu386 | Cpu486))
474 {
475 /* This code is disabled on the 486 because all the Imm1 forms
476 in the opcode table are slower on the i486. They're the
477 versions with the implicitly specified single-position
478 displacement, which has another syntax if you really want to
479 use that form. */
480 if (num == 1)
481 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32;
482 }
483 return (fits_in_signed_byte (num)
484 ? (Imm8S | Imm8 | Imm16 | Imm32)
485 : fits_in_unsigned_byte (num)
486 ? (Imm8 | Imm16 | Imm32)
487 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
488 ? (Imm16 | Imm32)
489 : (Imm32));
490 }
491
492 static offsetT
493 offset_in_range (val, size)
494 offsetT val;
495 int size;
496 {
497 addressT mask;
498
499 switch (size)
500 {
501 case 1: mask = ((addressT) 1 << 8) - 1; break;
502 case 2: mask = ((addressT) 1 << 16) - 1; break;
503 case 4: mask = ((addressT) 2 << 31) - 1; break;
504 default: abort ();
505 }
506
507 /* If BFD64, sign extend val. */
508 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
509 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
510
511 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
512 {
513 char buf1[40], buf2[40];
514
515 sprint_value (buf1, val);
516 sprint_value (buf2, val & mask);
517 as_warn (_("%s shortened to %s"), buf1, buf2);
518 }
519 return val & mask;
520 }
521
522 /* Returns 0 if attempting to add a prefix where one from the same
523 class already exists, 1 if non rep/repne added, 2 if rep/repne
524 added. */
525 static int
526 add_prefix (prefix)
527 unsigned int prefix;
528 {
529 int ret = 1;
530 int q;
531
532 switch (prefix)
533 {
534 default:
535 abort ();
536
537 case CS_PREFIX_OPCODE:
538 case DS_PREFIX_OPCODE:
539 case ES_PREFIX_OPCODE:
540 case FS_PREFIX_OPCODE:
541 case GS_PREFIX_OPCODE:
542 case SS_PREFIX_OPCODE:
543 q = SEG_PREFIX;
544 break;
545
546 case REPNE_PREFIX_OPCODE:
547 case REPE_PREFIX_OPCODE:
548 ret = 2;
549 /* fall thru */
550 case LOCK_PREFIX_OPCODE:
551 q = LOCKREP_PREFIX;
552 break;
553
554 case FWAIT_OPCODE:
555 q = WAIT_PREFIX;
556 break;
557
558 case ADDR_PREFIX_OPCODE:
559 q = ADDR_PREFIX;
560 break;
561
562 case DATA_PREFIX_OPCODE:
563 q = DATA_PREFIX;
564 break;
565 }
566
567 if (i.prefix[q])
568 {
569 as_bad (_("same type of prefix used twice"));
570 return 0;
571 }
572
573 i.prefixes += 1;
574 i.prefix[q] = prefix;
575 return ret;
576 }
577
578 static void
579 set_16bit_code_flag (new_16bit_code_flag)
580 int new_16bit_code_flag;
581 {
582 flag_16bit_code = new_16bit_code_flag;
583 stackop_size = '\0';
584 }
585
586 static void
587 set_16bit_gcc_code_flag (new_16bit_code_flag)
588 int new_16bit_code_flag;
589 {
590 flag_16bit_code = new_16bit_code_flag;
591 stackop_size = new_16bit_code_flag ? 'l' : '\0';
592 }
593
594 static void
595 set_intel_syntax (syntax_flag)
596 int syntax_flag;
597 {
598 /* Find out if register prefixing is specified. */
599 int ask_naked_reg = 0;
600
601 SKIP_WHITESPACE ();
602 if (! is_end_of_line[(unsigned char) *input_line_pointer])
603 {
604 char *string = input_line_pointer;
605 int e = get_symbol_end ();
606
607 if (strcmp (string, "prefix") == 0)
608 ask_naked_reg = 1;
609 else if (strcmp (string, "noprefix") == 0)
610 ask_naked_reg = -1;
611 else
612 as_bad (_("bad argument to syntax directive."));
613 *input_line_pointer = e;
614 }
615 demand_empty_rest_of_line ();
616
617 intel_syntax = syntax_flag;
618
619 if (ask_naked_reg == 0)
620 {
621 #ifdef BFD_ASSEMBLER
622 allow_naked_reg = (intel_syntax
623 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
624 #else
625 /* Conservative default. */
626 allow_naked_reg = 0;
627 #endif
628 }
629 else
630 allow_naked_reg = (ask_naked_reg < 0);
631 }
632
633 static void
634 set_cpu_arch (dummy)
635 int dummy ATTRIBUTE_UNUSED;
636 {
637 SKIP_WHITESPACE ();
638
639 if (! is_end_of_line[(unsigned char) *input_line_pointer])
640 {
641 char *string = input_line_pointer;
642 int e = get_symbol_end ();
643 int i;
644
645 for (i = 0; cpu_arch[i].name; i++)
646 {
647 if (strcmp (string, cpu_arch[i].name) == 0)
648 {
649 cpu_arch_name = cpu_arch[i].name;
650 cpu_arch_flags = cpu_arch[i].flags;
651 break;
652 }
653 }
654 if (!cpu_arch[i].name)
655 as_bad (_("no such architecture: `%s'"), string);
656
657 *input_line_pointer = e;
658 }
659 else
660 as_bad (_("missing cpu architecture"));
661
662 demand_empty_rest_of_line ();
663 }
664
665 const pseudo_typeS md_pseudo_table[] =
666 {
667 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
668 {"align", s_align_bytes, 0},
669 #else
670 {"align", s_align_ptwo, 0},
671 #endif
672 {"arch", set_cpu_arch, 0},
673 #ifndef I386COFF
674 {"bss", s_bss, 0},
675 #endif
676 {"ffloat", float_cons, 'f'},
677 {"dfloat", float_cons, 'd'},
678 {"tfloat", float_cons, 'x'},
679 {"value", cons, 2},
680 {"noopt", s_ignore, 0},
681 {"optim", s_ignore, 0},
682 {"code16gcc", set_16bit_gcc_code_flag, 1},
683 {"code16", set_16bit_code_flag, 1},
684 {"code32", set_16bit_code_flag, 0},
685 {"intel_syntax", set_intel_syntax, 1},
686 {"att_syntax", set_intel_syntax, 0},
687 {"file", dwarf2_directive_file, 0},
688 {"loc", dwarf2_directive_loc, 0},
689 {0, 0, 0}
690 };
691
692 /* For interface with expression (). */
693 extern char *input_line_pointer;
694
695 /* Hash table for instruction mnemonic lookup. */
696 static struct hash_control *op_hash;
697
698 /* Hash table for register lookup. */
699 static struct hash_control *reg_hash;
700 \f
701 void
702 md_begin ()
703 {
704 const char *hash_err;
705
706 /* Initialize op_hash hash table. */
707 op_hash = hash_new ();
708
709 {
710 register const template *optab;
711 register templates *core_optab;
712
713 /* Setup for loop. */
714 optab = i386_optab;
715 core_optab = (templates *) xmalloc (sizeof (templates));
716 core_optab->start = optab;
717
718 while (1)
719 {
720 ++optab;
721 if (optab->name == NULL
722 || strcmp (optab->name, (optab - 1)->name) != 0)
723 {
724 /* different name --> ship out current template list;
725 add to hash table; & begin anew. */
726 core_optab->end = optab;
727 hash_err = hash_insert (op_hash,
728 (optab - 1)->name,
729 (PTR) core_optab);
730 if (hash_err)
731 {
732 hash_error:
733 as_fatal (_("Internal Error: Can't hash %s: %s"),
734 (optab - 1)->name,
735 hash_err);
736 }
737 if (optab->name == NULL)
738 break;
739 core_optab = (templates *) xmalloc (sizeof (templates));
740 core_optab->start = optab;
741 }
742 }
743 }
744
745 /* Initialize reg_hash hash table. */
746 reg_hash = hash_new ();
747 {
748 register const reg_entry *regtab;
749
750 for (regtab = i386_regtab;
751 regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
752 regtab++)
753 {
754 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
755 if (hash_err)
756 goto hash_error;
757 }
758 }
759
760 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
761 {
762 register int c;
763 register char *p;
764
765 for (c = 0; c < 256; c++)
766 {
767 if (isdigit (c))
768 {
769 digit_chars[c] = c;
770 mnemonic_chars[c] = c;
771 register_chars[c] = c;
772 operand_chars[c] = c;
773 }
774 else if (islower (c))
775 {
776 mnemonic_chars[c] = c;
777 register_chars[c] = c;
778 operand_chars[c] = c;
779 }
780 else if (isupper (c))
781 {
782 mnemonic_chars[c] = tolower (c);
783 register_chars[c] = mnemonic_chars[c];
784 operand_chars[c] = c;
785 }
786
787 if (isalpha (c) || isdigit (c))
788 identifier_chars[c] = c;
789 else if (c >= 128)
790 {
791 identifier_chars[c] = c;
792 operand_chars[c] = c;
793 }
794 }
795
796 #ifdef LEX_AT
797 identifier_chars['@'] = '@';
798 #endif
799 digit_chars['-'] = '-';
800 identifier_chars['_'] = '_';
801 identifier_chars['.'] = '.';
802
803 for (p = operand_special_chars; *p != '\0'; p++)
804 operand_chars[(unsigned char) *p] = *p;
805 }
806
807 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
808 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
809 {
810 record_alignment (text_section, 2);
811 record_alignment (data_section, 2);
812 record_alignment (bss_section, 2);
813 }
814 #endif
815 }
816
817 void
818 i386_print_statistics (file)
819 FILE *file;
820 {
821 hash_print_statistics (file, "i386 opcode", op_hash);
822 hash_print_statistics (file, "i386 register", reg_hash);
823 }
824 \f
825 #ifdef DEBUG386
826
827 /* Debugging routines for md_assemble. */
828 static void pi PARAMS ((char *, i386_insn *));
829 static void pte PARAMS ((template *));
830 static void pt PARAMS ((unsigned int));
831 static void pe PARAMS ((expressionS *));
832 static void ps PARAMS ((symbolS *));
833
834 static void
835 pi (line, x)
836 char *line;
837 i386_insn *x;
838 {
839 register template *p;
840 int i;
841
842 fprintf (stdout, "%s: template ", line);
843 pte (&x->tm);
844 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x",
845 x->rm.mode, x->rm.reg, x->rm.regmem);
846 fprintf (stdout, " base %x index %x scale %x\n",
847 x->bi.base, x->bi.index, x->bi.scale);
848 for (i = 0; i < x->operands; i++)
849 {
850 fprintf (stdout, " #%d: ", i + 1);
851 pt (x->types[i]);
852 fprintf (stdout, "\n");
853 if (x->types[i]
854 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
855 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
856 if (x->types[i] & Imm)
857 pe (x->op[i].imms);
858 if (x->types[i] & Disp)
859 pe (x->op[i].disps);
860 }
861 }
862
863 static void
864 pte (t)
865 template *t;
866 {
867 int i;
868 fprintf (stdout, " %d operands ", t->operands);
869 fprintf (stdout, "opcode %x ", t->base_opcode);
870 if (t->extension_opcode != None)
871 fprintf (stdout, "ext %x ", t->extension_opcode);
872 if (t->opcode_modifier & D)
873 fprintf (stdout, "D");
874 if (t->opcode_modifier & W)
875 fprintf (stdout, "W");
876 fprintf (stdout, "\n");
877 for (i = 0; i < t->operands; i++)
878 {
879 fprintf (stdout, " #%d type ", i + 1);
880 pt (t->operand_types[i]);
881 fprintf (stdout, "\n");
882 }
883 }
884
885 static void
886 pe (e)
887 expressionS *e;
888 {
889 fprintf (stdout, " operation %d\n", e->X_op);
890 fprintf (stdout, " add_number %ld (%lx)\n",
891 (long) e->X_add_number, (long) e->X_add_number);
892 if (e->X_add_symbol)
893 {
894 fprintf (stdout, " add_symbol ");
895 ps (e->X_add_symbol);
896 fprintf (stdout, "\n");
897 }
898 if (e->X_op_symbol)
899 {
900 fprintf (stdout, " op_symbol ");
901 ps (e->X_op_symbol);
902 fprintf (stdout, "\n");
903 }
904 }
905
906 static void
907 ps (s)
908 symbolS *s;
909 {
910 fprintf (stdout, "%s type %s%s",
911 S_GET_NAME (s),
912 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
913 segment_name (S_GET_SEGMENT (s)));
914 }
915
916 struct type_name
917 {
918 unsigned int mask;
919 char *tname;
920 }
921
922 type_names[] =
923 {
924 { Reg8, "r8" },
925 { Reg16, "r16" },
926 { Reg32, "r32" },
927 { Imm8, "i8" },
928 { Imm8S, "i8s" },
929 { Imm16, "i16" },
930 { Imm32, "i32" },
931 { Imm1, "i1" },
932 { BaseIndex, "BaseIndex" },
933 { Disp8, "d8" },
934 { Disp16, "d16" },
935 { Disp32, "d32" },
936 { InOutPortReg, "InOutPortReg" },
937 { ShiftCount, "ShiftCount" },
938 { Control, "control reg" },
939 { Test, "test reg" },
940 { Debug, "debug reg" },
941 { FloatReg, "FReg" },
942 { FloatAcc, "FAcc" },
943 { SReg2, "SReg2" },
944 { SReg3, "SReg3" },
945 { Acc, "Acc" },
946 { JumpAbsolute, "Jump Absolute" },
947 { RegMMX, "rMMX" },
948 { RegXMM, "rXMM" },
949 { EsSeg, "es" },
950 { 0, "" }
951 };
952
953 static void
954 pt (t)
955 unsigned int t;
956 {
957 register struct type_name *ty;
958
959 if (t == Unknown)
960 {
961 fprintf (stdout, _("Unknown"));
962 }
963 else
964 {
965 for (ty = type_names; ty->mask; ty++)
966 if (t & ty->mask)
967 fprintf (stdout, "%s, ", ty->tname);
968 }
969 fflush (stdout);
970 }
971
972 #endif /* DEBUG386 */
973 \f
974 int
975 tc_i386_force_relocation (fixp)
976 struct fix *fixp;
977 {
978 #ifdef BFD_ASSEMBLER
979 if (fixp->fx_r_type == BFD_RELOC_VTABLE_INHERIT
980 || fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
981 return 1;
982 return 0;
983 #else
984 /* For COFF. */
985 return fixp->fx_r_type == 7;
986 #endif
987 }
988
989 #ifdef BFD_ASSEMBLER
990 static bfd_reloc_code_real_type reloc
991 PARAMS ((int, int, bfd_reloc_code_real_type));
992
993 static bfd_reloc_code_real_type
994 reloc (size, pcrel, other)
995 int size;
996 int pcrel;
997 bfd_reloc_code_real_type other;
998 {
999 if (other != NO_RELOC)
1000 return other;
1001
1002 if (pcrel)
1003 {
1004 switch (size)
1005 {
1006 case 1: return BFD_RELOC_8_PCREL;
1007 case 2: return BFD_RELOC_16_PCREL;
1008 case 4: return BFD_RELOC_32_PCREL;
1009 }
1010 as_bad (_("can not do %d byte pc-relative relocation"), size);
1011 }
1012 else
1013 {
1014 switch (size)
1015 {
1016 case 1: return BFD_RELOC_8;
1017 case 2: return BFD_RELOC_16;
1018 case 4: return BFD_RELOC_32;
1019 }
1020 as_bad (_("can not do %d byte relocation"), size);
1021 }
1022
1023 return BFD_RELOC_NONE;
1024 }
1025
1026 /* Here we decide which fixups can be adjusted to make them relative to
1027 the beginning of the section instead of the symbol. Basically we need
1028 to make sure that the dynamic relocations are done correctly, so in
1029 some cases we force the original symbol to be used. */
1030
1031 int
1032 tc_i386_fix_adjustable (fixP)
1033 fixS *fixP;
1034 {
1035 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1036 /* Prevent all adjustments to global symbols, or else dynamic
1037 linking will not work correctly. */
1038 if (S_IS_EXTERNAL (fixP->fx_addsy)
1039 || S_IS_WEAK (fixP->fx_addsy))
1040 return 0;
1041 #endif
1042 /* adjust_reloc_syms doesn't know about the GOT. */
1043 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1044 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1045 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1046 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1047 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1048 return 0;
1049 return 1;
1050 }
1051 #else
1052 #define reloc(SIZE,PCREL,OTHER) 0
1053 #define BFD_RELOC_16 0
1054 #define BFD_RELOC_32 0
1055 #define BFD_RELOC_16_PCREL 0
1056 #define BFD_RELOC_32_PCREL 0
1057 #define BFD_RELOC_386_PLT32 0
1058 #define BFD_RELOC_386_GOT32 0
1059 #define BFD_RELOC_386_GOTOFF 0
1060 #endif
1061
1062 static int intel_float_operand PARAMS ((char *mnemonic));
1063
1064 static int
1065 intel_float_operand (mnemonic)
1066 char *mnemonic;
1067 {
1068 if (mnemonic[0] == 'f' && mnemonic[1] == 'i')
1069 return 2;
1070
1071 if (mnemonic[0] == 'f')
1072 return 1;
1073
1074 return 0;
1075 }
1076
1077 /* This is the guts of the machine-dependent assembler. LINE points to a
1078 machine dependent instruction. This function is supposed to emit
1079 the frags/bytes it assembles to. */
1080
1081 void
1082 md_assemble (line)
1083 char *line;
1084 {
1085 /* Points to template once we've found it. */
1086 const template *t;
1087
1088 /* Count the size of the instruction generated. */
1089 int insn_size = 0;
1090
1091 int j;
1092
1093 char mnemonic[MAX_MNEM_SIZE];
1094
1095 /* Initialize globals. */
1096 memset (&i, '\0', sizeof (i));
1097 for (j = 0; j < MAX_OPERANDS; j++)
1098 i.disp_reloc[j] = NO_RELOC;
1099 memset (disp_expressions, '\0', sizeof (disp_expressions));
1100 memset (im_expressions, '\0', sizeof (im_expressions));
1101 save_stack_p = save_stack;
1102
1103 /* First parse an instruction mnemonic & call i386_operand for the operands.
1104 We assume that the scrubber has arranged it so that line[0] is the valid
1105 start of a (possibly prefixed) mnemonic. */
1106 {
1107 char *l = line;
1108 char *token_start = l;
1109 char *mnem_p;
1110
1111 /* Non-zero if we found a prefix only acceptable with string insns. */
1112 const char *expecting_string_instruction = NULL;
1113
1114 while (1)
1115 {
1116 mnem_p = mnemonic;
1117 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1118 {
1119 mnem_p++;
1120 if (mnem_p >= mnemonic + sizeof (mnemonic))
1121 {
1122 as_bad (_("no such instruction: `%s'"), token_start);
1123 return;
1124 }
1125 l++;
1126 }
1127 if (!is_space_char (*l)
1128 && *l != END_OF_INSN
1129 && *l != PREFIX_SEPARATOR)
1130 {
1131 as_bad (_("invalid character %s in mnemonic"),
1132 output_invalid (*l));
1133 return;
1134 }
1135 if (token_start == l)
1136 {
1137 if (*l == PREFIX_SEPARATOR)
1138 as_bad (_("expecting prefix; got nothing"));
1139 else
1140 as_bad (_("expecting mnemonic; got nothing"));
1141 return;
1142 }
1143
1144 /* Look up instruction (or prefix) via hash table. */
1145 current_templates = hash_find (op_hash, mnemonic);
1146
1147 if (*l != END_OF_INSN
1148 && (! is_space_char (*l) || l[1] != END_OF_INSN)
1149 && current_templates
1150 && (current_templates->start->opcode_modifier & IsPrefix))
1151 {
1152 /* If we are in 16-bit mode, do not allow addr16 or data16.
1153 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1154 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1155 && (((current_templates->start->opcode_modifier & Size32) != 0)
1156 ^ flag_16bit_code))
1157 {
1158 as_bad (_("redundant %s prefix"),
1159 current_templates->start->name);
1160 return;
1161 }
1162 /* Add prefix, checking for repeated prefixes. */
1163 switch (add_prefix (current_templates->start->base_opcode))
1164 {
1165 case 0:
1166 return;
1167 case 2:
1168 expecting_string_instruction = current_templates->start->name;
1169 break;
1170 }
1171 /* Skip past PREFIX_SEPARATOR and reset token_start. */
1172 token_start = ++l;
1173 }
1174 else
1175 break;
1176 }
1177
1178 if (!current_templates)
1179 {
1180 /* See if we can get a match by trimming off a suffix. */
1181 switch (mnem_p[-1])
1182 {
1183 case WORD_MNEM_SUFFIX:
1184 case BYTE_MNEM_SUFFIX:
1185 i.suffix = mnem_p[-1];
1186 mnem_p[-1] = '\0';
1187 current_templates = hash_find (op_hash, mnemonic);
1188 break;
1189 case SHORT_MNEM_SUFFIX:
1190 case LONG_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 }
1197 break;
1198
1199 /* Intel Syntax. */
1200 case 'd':
1201 if (intel_syntax)
1202 {
1203 if (intel_float_operand (mnemonic))
1204 i.suffix = SHORT_MNEM_SUFFIX;
1205 else
1206 i.suffix = LONG_MNEM_SUFFIX;
1207 mnem_p[-1] = '\0';
1208 current_templates = hash_find (op_hash, mnemonic);
1209 }
1210 break;
1211 }
1212 if (!current_templates)
1213 {
1214 as_bad (_("no such instruction: `%s'"), token_start);
1215 return;
1216 }
1217 }
1218
1219 /* Check if instruction is supported on specified architecture. */
1220 if (cpu_arch_flags != 0)
1221 {
1222 if (current_templates->start->cpu_flags & ~cpu_arch_flags)
1223 {
1224 as_warn (_("`%s' is not supported on `%s'"),
1225 current_templates->start->name, cpu_arch_name);
1226 }
1227 else if ((Cpu386 & ~cpu_arch_flags) && !flag_16bit_code)
1228 {
1229 as_warn (_("use .code16 to ensure correct addressing mode"));
1230 }
1231 }
1232
1233 /* Check for rep/repne without a string instruction. */
1234 if (expecting_string_instruction
1235 && !(current_templates->start->opcode_modifier & IsString))
1236 {
1237 as_bad (_("expecting string instruction after `%s'"),
1238 expecting_string_instruction);
1239 return;
1240 }
1241
1242 /* There may be operands to parse. */
1243 if (*l != END_OF_INSN)
1244 {
1245 /* 1 if operand is pending after ','. */
1246 unsigned int expecting_operand = 0;
1247
1248 /* Non-zero if operand parens not balanced. */
1249 unsigned int paren_not_balanced;
1250
1251 do
1252 {
1253 /* Skip optional white space before operand. */
1254 if (is_space_char (*l))
1255 ++l;
1256 if (!is_operand_char (*l) && *l != END_OF_INSN)
1257 {
1258 as_bad (_("invalid character %s before operand %d"),
1259 output_invalid (*l),
1260 i.operands + 1);
1261 return;
1262 }
1263 token_start = l; /* after white space */
1264 paren_not_balanced = 0;
1265 while (paren_not_balanced || *l != ',')
1266 {
1267 if (*l == END_OF_INSN)
1268 {
1269 if (paren_not_balanced)
1270 {
1271 if (!intel_syntax)
1272 as_bad (_("unbalanced parenthesis in operand %d."),
1273 i.operands + 1);
1274 else
1275 as_bad (_("unbalanced brackets in operand %d."),
1276 i.operands + 1);
1277 return;
1278 }
1279 else
1280 break; /* we are done */
1281 }
1282 else if (!is_operand_char (*l) && !is_space_char (*l))
1283 {
1284 as_bad (_("invalid character %s in operand %d"),
1285 output_invalid (*l),
1286 i.operands + 1);
1287 return;
1288 }
1289 if (!intel_syntax)
1290 {
1291 if (*l == '(')
1292 ++paren_not_balanced;
1293 if (*l == ')')
1294 --paren_not_balanced;
1295 }
1296 else
1297 {
1298 if (*l == '[')
1299 ++paren_not_balanced;
1300 if (*l == ']')
1301 --paren_not_balanced;
1302 }
1303 l++;
1304 }
1305 if (l != token_start)
1306 { /* Yes, we've read in another operand. */
1307 unsigned int operand_ok;
1308 this_operand = i.operands++;
1309 if (i.operands > MAX_OPERANDS)
1310 {
1311 as_bad (_("spurious operands; (%d operands/instruction max)"),
1312 MAX_OPERANDS);
1313 return;
1314 }
1315 /* Now parse operand adding info to 'i' as we go along. */
1316 END_STRING_AND_SAVE (l);
1317
1318 if (intel_syntax)
1319 operand_ok =
1320 i386_intel_operand (token_start,
1321 intel_float_operand (mnemonic));
1322 else
1323 operand_ok = i386_operand (token_start);
1324
1325 RESTORE_END_STRING (l);
1326 if (!operand_ok)
1327 return;
1328 }
1329 else
1330 {
1331 if (expecting_operand)
1332 {
1333 expecting_operand_after_comma:
1334 as_bad (_("expecting operand after ','; got nothing"));
1335 return;
1336 }
1337 if (*l == ',')
1338 {
1339 as_bad (_("expecting operand before ','; got nothing"));
1340 return;
1341 }
1342 }
1343
1344 /* Now *l must be either ',' or END_OF_INSN. */
1345 if (*l == ',')
1346 {
1347 if (*++l == END_OF_INSN)
1348 {
1349 /* Just skip it, if it's \n complain. */
1350 goto expecting_operand_after_comma;
1351 }
1352 expecting_operand = 1;
1353 }
1354 }
1355 while (*l != END_OF_INSN);
1356 }
1357 }
1358
1359 /* Now we've parsed the mnemonic into a set of templates, and have the
1360 operands at hand.
1361
1362 Next, we find a template that matches the given insn,
1363 making sure the overlap of the given operands types is consistent
1364 with the template operand types. */
1365
1366 #define MATCH(overlap, given, template) \
1367 ((overlap & ~JumpAbsolute) \
1368 && ((given) & (BaseIndex|JumpAbsolute)) == ((overlap) & (BaseIndex|JumpAbsolute)))
1369
1370 /* If given types r0 and r1 are registers they must be of the same type
1371 unless the expected operand type register overlap is null.
1372 Note that Acc in a template matches every size of reg. */
1373 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
1374 ( ((g0) & Reg) == 0 || ((g1) & Reg) == 0 || \
1375 ((g0) & Reg) == ((g1) & Reg) || \
1376 ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
1377
1378 {
1379 register unsigned int overlap0, overlap1;
1380 unsigned int overlap2;
1381 unsigned int found_reverse_match;
1382 int suffix_check;
1383
1384 /* All intel opcodes have reversed operands except for "bound" and
1385 "enter". We also don't reverse intersegment "jmp" and "call"
1386 instructions with 2 immediate operands so that the immediate segment
1387 precedes the offset, as it does when in AT&T mode. "enter" and the
1388 intersegment "jmp" and "call" instructions are the only ones that
1389 have two immediate operands. */
1390 if (intel_syntax && i.operands > 1
1391 && (strcmp (mnemonic, "bound") != 0)
1392 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1393 {
1394 union i386_op temp_op;
1395 unsigned int temp_type;
1396 #ifdef BFD_ASSEMBLER
1397 enum bfd_reloc_code_real temp_disp_reloc;
1398 #else
1399 int temp_disp_reloc;
1400 #endif
1401 int xchg1 = 0;
1402 int xchg2 = 0;
1403
1404 if (i.operands == 2)
1405 {
1406 xchg1 = 0;
1407 xchg2 = 1;
1408 }
1409 else if (i.operands == 3)
1410 {
1411 xchg1 = 0;
1412 xchg2 = 2;
1413 }
1414 temp_type = i.types[xchg2];
1415 i.types[xchg2] = i.types[xchg1];
1416 i.types[xchg1] = temp_type;
1417 temp_op = i.op[xchg2];
1418 i.op[xchg2] = i.op[xchg1];
1419 i.op[xchg1] = temp_op;
1420 temp_disp_reloc = i.disp_reloc[xchg2];
1421 i.disp_reloc[xchg2] = i.disp_reloc[xchg1];
1422 i.disp_reloc[xchg1] = temp_disp_reloc;
1423
1424 if (i.mem_operands == 2)
1425 {
1426 const seg_entry *temp_seg;
1427 temp_seg = i.seg[0];
1428 i.seg[0] = i.seg[1];
1429 i.seg[1] = temp_seg;
1430 }
1431 }
1432
1433 if (i.imm_operands)
1434 {
1435 /* Try to ensure constant immediates are represented in the smallest
1436 opcode possible. */
1437 char guess_suffix = 0;
1438 int op;
1439
1440 if (i.suffix)
1441 guess_suffix = i.suffix;
1442 else if (i.reg_operands)
1443 {
1444 /* Figure out a suffix from the last register operand specified.
1445 We can't do this properly yet, ie. excluding InOutPortReg,
1446 but the following works for instructions with immediates.
1447 In any case, we can't set i.suffix yet. */
1448 for (op = i.operands; --op >= 0;)
1449 if (i.types[op] & Reg)
1450 {
1451 if (i.types[op] & Reg8)
1452 guess_suffix = BYTE_MNEM_SUFFIX;
1453 else if (i.types[op] & Reg16)
1454 guess_suffix = WORD_MNEM_SUFFIX;
1455 break;
1456 }
1457 }
1458 else if (flag_16bit_code ^ (i.prefix[DATA_PREFIX] != 0))
1459 guess_suffix = WORD_MNEM_SUFFIX;
1460
1461 for (op = i.operands; --op >= 0;)
1462 if ((i.types[op] & Imm)
1463 && i.op[op].imms->X_op == O_constant)
1464 {
1465 /* If a suffix is given, this operand may be shortened. */
1466 switch (guess_suffix)
1467 {
1468 case WORD_MNEM_SUFFIX:
1469 i.types[op] |= Imm16;
1470 break;
1471 case BYTE_MNEM_SUFFIX:
1472 i.types[op] |= Imm16 | Imm8 | Imm8S;
1473 break;
1474 }
1475
1476 /* If this operand is at most 16 bits, convert it to a
1477 signed 16 bit number before trying to see whether it will
1478 fit in an even smaller size. This allows a 16-bit operand
1479 such as $0xffe0 to be recognised as within Imm8S range. */
1480 if ((i.types[op] & Imm16)
1481 && (i.op[op].imms->X_add_number & ~(offsetT)0xffff) == 0)
1482 {
1483 i.op[op].imms->X_add_number =
1484 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
1485 }
1486 i.types[op] |= smallest_imm_type ((long) i.op[op].imms->X_add_number);
1487 }
1488 }
1489
1490 if (i.disp_operands)
1491 {
1492 /* Try to use the smallest displacement type too. */
1493 int op;
1494
1495 for (op = i.operands; --op >= 0;)
1496 if ((i.types[op] & Disp)
1497 && i.op[op].imms->X_op == O_constant)
1498 {
1499 offsetT disp = i.op[op].disps->X_add_number;
1500
1501 if (i.types[op] & Disp16)
1502 {
1503 /* We know this operand is at most 16 bits, so
1504 convert to a signed 16 bit number before trying
1505 to see whether it will fit in an even smaller
1506 size. */
1507
1508 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
1509 }
1510 if (fits_in_signed_byte (disp))
1511 i.types[op] |= Disp8;
1512 }
1513 }
1514
1515 overlap0 = 0;
1516 overlap1 = 0;
1517 overlap2 = 0;
1518 found_reverse_match = 0;
1519 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
1520 ? No_bSuf
1521 : (i.suffix == WORD_MNEM_SUFFIX
1522 ? No_wSuf
1523 : (i.suffix == SHORT_MNEM_SUFFIX
1524 ? No_sSuf
1525 : (i.suffix == LONG_MNEM_SUFFIX
1526 ? No_lSuf
1527 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX ? No_xSuf : 0)))));
1528
1529 for (t = current_templates->start;
1530 t < current_templates->end;
1531 t++)
1532 {
1533 /* Must have right number of operands. */
1534 if (i.operands != t->operands)
1535 continue;
1536
1537 /* Check the suffix, except for some instructions in intel mode. */
1538 if ((t->opcode_modifier & suffix_check)
1539 && !(intel_syntax
1540 && (t->opcode_modifier & IgnoreSize))
1541 && !(intel_syntax
1542 && t->base_opcode == 0xd9
1543 && (t->extension_opcode == 5 /* 0xd9,5 "fldcw" */
1544 || t->extension_opcode == 7))) /* 0xd9,7 "f{n}stcw" */
1545 continue;
1546
1547 else if (!t->operands)
1548 /* 0 operands always matches. */
1549 break;
1550
1551 overlap0 = i.types[0] & t->operand_types[0];
1552 switch (t->operands)
1553 {
1554 case 1:
1555 if (!MATCH (overlap0, i.types[0], t->operand_types[0]))
1556 continue;
1557 break;
1558 case 2:
1559 case 3:
1560 overlap1 = i.types[1] & t->operand_types[1];
1561 if (!MATCH (overlap0, i.types[0], t->operand_types[0])
1562 || !MATCH (overlap1, i.types[1], t->operand_types[1])
1563 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
1564 t->operand_types[0],
1565 overlap1, i.types[1],
1566 t->operand_types[1]))
1567 {
1568 /* Check if other direction is valid ... */
1569 if ((t->opcode_modifier & (D|FloatD)) == 0)
1570 continue;
1571
1572 /* Try reversing direction of operands. */
1573 overlap0 = i.types[0] & t->operand_types[1];
1574 overlap1 = i.types[1] & t->operand_types[0];
1575 if (!MATCH (overlap0, i.types[0], t->operand_types[1])
1576 || !MATCH (overlap1, i.types[1], t->operand_types[0])
1577 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
1578 t->operand_types[1],
1579 overlap1, i.types[1],
1580 t->operand_types[0]))
1581 {
1582 /* Does not match either direction. */
1583 continue;
1584 }
1585 /* found_reverse_match holds which of D or FloatDR
1586 we've found. */
1587 found_reverse_match = t->opcode_modifier & (D|FloatDR);
1588 break;
1589 }
1590 /* Found a forward 2 operand match here. */
1591 if (t->operands == 3)
1592 {
1593 /* Here we make use of the fact that there are no
1594 reverse match 3 operand instructions, and all 3
1595 operand instructions only need to be checked for
1596 register consistency between operands 2 and 3. */
1597 overlap2 = i.types[2] & t->operand_types[2];
1598 if (!MATCH (overlap2, i.types[2], t->operand_types[2])
1599 || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1],
1600 t->operand_types[1],
1601 overlap2, i.types[2],
1602 t->operand_types[2]))
1603
1604 continue;
1605 }
1606 /* Found either forward/reverse 2 or 3 operand match here:
1607 slip through to break. */
1608 }
1609 /* We've found a match; break out of loop. */
1610 break;
1611 }
1612 if (t == current_templates->end)
1613 {
1614 /* We found no match. */
1615 as_bad (_("suffix or operands invalid for `%s'"),
1616 current_templates->start->name);
1617 return;
1618 }
1619
1620 if (!quiet_warnings)
1621 {
1622 if (!intel_syntax
1623 && ((i.types[0] & JumpAbsolute)
1624 != (t->operand_types[0] & JumpAbsolute)))
1625 {
1626 as_warn (_("indirect %s without `*'"), t->name);
1627 }
1628
1629 if ((t->opcode_modifier & (IsPrefix|IgnoreSize))
1630 == (IsPrefix|IgnoreSize))
1631 {
1632 /* Warn them that a data or address size prefix doesn't
1633 affect assembly of the next line of code. */
1634 as_warn (_("stand-alone `%s' prefix"), t->name);
1635 }
1636 }
1637
1638 /* Copy the template we found. */
1639 i.tm = *t;
1640 if (found_reverse_match)
1641 {
1642 /* If we found a reverse match we must alter the opcode
1643 direction bit. found_reverse_match holds bits to change
1644 (different for int & float insns). */
1645
1646 i.tm.base_opcode ^= found_reverse_match;
1647
1648 i.tm.operand_types[0] = t->operand_types[1];
1649 i.tm.operand_types[1] = t->operand_types[0];
1650 }
1651
1652 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1653 if (SYSV386_COMPAT
1654 && intel_syntax
1655 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1656 i.tm.base_opcode ^= FloatR;
1657
1658 if (i.tm.opcode_modifier & FWait)
1659 if (! add_prefix (FWAIT_OPCODE))
1660 return;
1661
1662 /* Check string instruction segment overrides. */
1663 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1664 {
1665 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
1666 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
1667 {
1668 if (i.seg[0] != NULL && i.seg[0] != &es)
1669 {
1670 as_bad (_("`%s' operand %d must use `%%es' segment"),
1671 i.tm.name,
1672 mem_op + 1);
1673 return;
1674 }
1675 /* There's only ever one segment override allowed per instruction.
1676 This instruction possibly has a legal segment override on the
1677 second operand, so copy the segment to where non-string
1678 instructions store it, allowing common code. */
1679 i.seg[0] = i.seg[1];
1680 }
1681 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
1682 {
1683 if (i.seg[1] != NULL && i.seg[1] != &es)
1684 {
1685 as_bad (_("`%s' operand %d must use `%%es' segment"),
1686 i.tm.name,
1687 mem_op + 2);
1688 return;
1689 }
1690 }
1691 }
1692
1693 /* If matched instruction specifies an explicit instruction mnemonic
1694 suffix, use it. */
1695 if (i.tm.opcode_modifier & (Size16 | Size32))
1696 {
1697 if (i.tm.opcode_modifier & Size16)
1698 i.suffix = WORD_MNEM_SUFFIX;
1699 else
1700 i.suffix = LONG_MNEM_SUFFIX;
1701 }
1702 else if (i.reg_operands)
1703 {
1704 /* If there's no instruction mnemonic suffix we try to invent one
1705 based on register operands. */
1706 if (!i.suffix)
1707 {
1708 /* We take i.suffix from the last register operand specified,
1709 Destination register type is more significant than source
1710 register type. */
1711 int op;
1712 for (op = i.operands; --op >= 0;)
1713 if ((i.types[op] & Reg)
1714 && !(i.tm.operand_types[op] & InOutPortReg))
1715 {
1716 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
1717 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
1718 LONG_MNEM_SUFFIX);
1719 break;
1720 }
1721 }
1722 else if (i.suffix == BYTE_MNEM_SUFFIX)
1723 {
1724 int op;
1725 for (op = i.operands; --op >= 0;)
1726 {
1727 /* If this is an eight bit register, it's OK. If it's
1728 the 16 or 32 bit version of an eight bit register,
1729 we will just use the low portion, and that's OK too. */
1730 if (i.types[op] & Reg8)
1731 continue;
1732
1733 /* movzx and movsx should not generate this warning. */
1734 if (intel_syntax
1735 && (i.tm.base_opcode == 0xfb7
1736 || i.tm.base_opcode == 0xfb6
1737 || i.tm.base_opcode == 0xfbe
1738 || i.tm.base_opcode == 0xfbf))
1739 continue;
1740
1741 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4
1742 #if 0
1743 /* Check that the template allows eight bit regs
1744 This kills insns such as `orb $1,%edx', which
1745 maybe should be allowed. */
1746 && (i.tm.operand_types[op] & (Reg8|InOutPortReg))
1747 #endif
1748 )
1749 {
1750 #if REGISTER_WARNINGS
1751 if (!quiet_warnings
1752 && (i.tm.operand_types[op] & InOutPortReg) == 0)
1753 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
1754 (i.op[op].regs - (i.types[op] & Reg16 ? 8 : 16))->reg_name,
1755 i.op[op].regs->reg_name,
1756 i.suffix);
1757 #endif
1758 continue;
1759 }
1760 /* Any other register is bad. */
1761 if (i.types[op] & (Reg | RegMMX | RegXMM
1762 | SReg2 | SReg3
1763 | Control | Debug | Test
1764 | FloatReg | FloatAcc))
1765 {
1766 as_bad (_("`%%%s' not allowed with `%s%c'"),
1767 i.op[op].regs->reg_name,
1768 i.tm.name,
1769 i.suffix);
1770 return;
1771 }
1772 }
1773 }
1774 else if (i.suffix == LONG_MNEM_SUFFIX)
1775 {
1776 int op;
1777
1778 for (op = i.operands; --op >= 0;)
1779 /* Reject eight bit registers, except where the template
1780 requires them. (eg. movzb) */
1781 if ((i.types[op] & Reg8) != 0
1782 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
1783 {
1784 as_bad (_("`%%%s' not allowed with `%s%c'"),
1785 i.op[op].regs->reg_name,
1786 i.tm.name,
1787 i.suffix);
1788 return;
1789 }
1790 #if REGISTER_WARNINGS
1791 /* Warn if the e prefix on a general reg is missing. */
1792 else if (!quiet_warnings
1793 && (i.types[op] & Reg16) != 0
1794 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
1795 {
1796 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
1797 (i.op[op].regs + 8)->reg_name,
1798 i.op[op].regs->reg_name,
1799 i.suffix);
1800 }
1801 #endif
1802 }
1803 else if (i.suffix == WORD_MNEM_SUFFIX)
1804 {
1805 int op;
1806 for (op = i.operands; --op >= 0;)
1807 /* Reject eight bit registers, except where the template
1808 requires them. (eg. movzb) */
1809 if ((i.types[op] & Reg8) != 0
1810 && (i.tm.operand_types[op] & (Reg16|Reg32|Acc)) != 0)
1811 {
1812 as_bad (_("`%%%s' not allowed with `%s%c'"),
1813 i.op[op].regs->reg_name,
1814 i.tm.name,
1815 i.suffix);
1816 return;
1817 }
1818 #if REGISTER_WARNINGS
1819 /* Warn if the e prefix on a general reg is present. */
1820 else if (!quiet_warnings
1821 && (i.types[op] & Reg32) != 0
1822 && (i.tm.operand_types[op] & (Reg16|Acc)) != 0)
1823 {
1824 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
1825 (i.op[op].regs - 8)->reg_name,
1826 i.op[op].regs->reg_name,
1827 i.suffix);
1828 }
1829 #endif
1830 }
1831 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
1832 /* Do nothing if the instruction is going to ignore the prefix. */
1833 ;
1834 else
1835 abort ();
1836 }
1837 else if ((i.tm.opcode_modifier & DefaultSize) && !i.suffix)
1838 {
1839 i.suffix = stackop_size;
1840 }
1841
1842 /* Make still unresolved immediate matches conform to size of immediate
1843 given in i.suffix. Note: overlap2 cannot be an immediate! */
1844 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32))
1845 && overlap0 != Imm8 && overlap0 != Imm8S
1846 && overlap0 != Imm16 && overlap0 != Imm32)
1847 {
1848 if (i.suffix)
1849 {
1850 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
1851 (i.suffix == WORD_MNEM_SUFFIX ? Imm16 : Imm32));
1852 }
1853 else if (overlap0 == (Imm16 | Imm32))
1854 {
1855 overlap0 =
1856 (flag_16bit_code ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32;
1857 }
1858 else
1859 {
1860 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
1861 return;
1862 }
1863 }
1864 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32))
1865 && overlap1 != Imm8 && overlap1 != Imm8S
1866 && overlap1 != Imm16 && overlap1 != Imm32)
1867 {
1868 if (i.suffix)
1869 {
1870 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
1871 (i.suffix == WORD_MNEM_SUFFIX ? Imm16 : Imm32));
1872 }
1873 else if (overlap1 == (Imm16 | Imm32))
1874 {
1875 overlap1 =
1876 (flag_16bit_code ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32;
1877 }
1878 else
1879 {
1880 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
1881 return;
1882 }
1883 }
1884 assert ((overlap2 & Imm) == 0);
1885
1886 i.types[0] = overlap0;
1887 if (overlap0 & ImplicitRegister)
1888 i.reg_operands--;
1889 if (overlap0 & Imm1)
1890 i.imm_operands = 0; /* kludge for shift insns. */
1891
1892 i.types[1] = overlap1;
1893 if (overlap1 & ImplicitRegister)
1894 i.reg_operands--;
1895
1896 i.types[2] = overlap2;
1897 if (overlap2 & ImplicitRegister)
1898 i.reg_operands--;
1899
1900 /* Finalize opcode. First, we change the opcode based on the operand
1901 size given by i.suffix: We need not change things for byte insns. */
1902
1903 if (!i.suffix && (i.tm.opcode_modifier & W))
1904 {
1905 as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
1906 return;
1907 }
1908
1909 /* For movzx and movsx, need to check the register type. */
1910 if (intel_syntax
1911 && (i.tm.base_opcode == 0xfb6 || i.tm.base_opcode == 0xfbe))
1912 if (i.suffix && i.suffix == BYTE_MNEM_SUFFIX)
1913 {
1914 unsigned int prefix = DATA_PREFIX_OPCODE;
1915
1916 if ((i.op[1].regs->reg_type & Reg16) != 0)
1917 if (!add_prefix (prefix))
1918 return;
1919 }
1920
1921 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
1922 {
1923 /* It's not a byte, select word/dword operation. */
1924 if (i.tm.opcode_modifier & W)
1925 {
1926 if (i.tm.opcode_modifier & ShortForm)
1927 i.tm.base_opcode |= 8;
1928 else
1929 i.tm.base_opcode |= 1;
1930 }
1931 /* Now select between word & dword operations via the operand
1932 size prefix, except for instructions that will ignore this
1933 prefix anyway. */
1934 if ((i.suffix == LONG_MNEM_SUFFIX) == flag_16bit_code
1935 && !(i.tm.opcode_modifier & IgnoreSize))
1936 {
1937 unsigned int prefix = DATA_PREFIX_OPCODE;
1938 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
1939 prefix = ADDR_PREFIX_OPCODE;
1940
1941 if (! add_prefix (prefix))
1942 return;
1943 }
1944 /* Size floating point instruction. */
1945 if (i.suffix == LONG_MNEM_SUFFIX)
1946 {
1947 if (i.tm.opcode_modifier & FloatMF)
1948 i.tm.base_opcode ^= 4;
1949 }
1950 }
1951
1952 if (i.tm.opcode_modifier & ImmExt)
1953 {
1954 /* These AMD 3DNow! and Intel Katmai New Instructions have an
1955 opcode suffix which is coded in the same place as an 8-bit
1956 immediate field would be. Here we fake an 8-bit immediate
1957 operand from the opcode suffix stored in tm.extension_opcode. */
1958
1959 expressionS *exp;
1960
1961 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
1962
1963 exp = &im_expressions[i.imm_operands++];
1964 i.op[i.operands].imms = exp;
1965 i.types[i.operands++] = Imm8;
1966 exp->X_op = O_constant;
1967 exp->X_add_number = i.tm.extension_opcode;
1968 i.tm.extension_opcode = None;
1969 }
1970
1971 /* For insns with operands there are more diddles to do to the opcode. */
1972 if (i.operands)
1973 {
1974 /* Default segment register this instruction will use
1975 for memory accesses. 0 means unknown.
1976 This is only for optimizing out unnecessary segment overrides. */
1977 const seg_entry *default_seg = 0;
1978
1979 /* The imul $imm, %reg instruction is converted into
1980 imul $imm, %reg, %reg, and the clr %reg instruction
1981 is converted into xor %reg, %reg. */
1982 if (i.tm.opcode_modifier & regKludge)
1983 {
1984 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
1985 /* Pretend we saw the extra register operand. */
1986 assert (i.op[first_reg_op + 1].regs == 0);
1987 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
1988 i.types[first_reg_op + 1] = i.types[first_reg_op];
1989 i.reg_operands = 2;
1990 }
1991
1992 if (i.tm.opcode_modifier & ShortForm)
1993 {
1994 /* The register or float register operand is in operand 0 or 1. */
1995 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
1996 /* Register goes in low 3 bits of opcode. */
1997 i.tm.base_opcode |= i.op[op].regs->reg_num;
1998 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
1999 {
2000 /* Warn about some common errors, but press on regardless.
2001 The first case can be generated by gcc (<= 2.8.1). */
2002 if (i.operands == 2)
2003 {
2004 /* Reversed arguments on faddp, fsubp, etc. */
2005 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
2006 i.op[1].regs->reg_name,
2007 i.op[0].regs->reg_name);
2008 }
2009 else
2010 {
2011 /* Extraneous `l' suffix on fp insn. */
2012 as_warn (_("translating to `%s %%%s'"), i.tm.name,
2013 i.op[0].regs->reg_name);
2014 }
2015 }
2016 }
2017 else if (i.tm.opcode_modifier & Modrm)
2018 {
2019 /* The opcode is completed (modulo i.tm.extension_opcode which
2020 must be put into the modrm byte).
2021 Now, we make the modrm & index base bytes based on all the
2022 info we've collected. */
2023
2024 /* i.reg_operands MUST be the number of real register operands;
2025 implicit registers do not count. */
2026 if (i.reg_operands == 2)
2027 {
2028 unsigned int source, dest;
2029 source = ((i.types[0]
2030 & (Reg | RegMMX | RegXMM
2031 | SReg2 | SReg3
2032 | Control | Debug | Test))
2033 ? 0 : 1);
2034 dest = source + 1;
2035
2036 i.rm.mode = 3;
2037 /* One of the register operands will be encoded in the
2038 i.tm.reg field, the other in the combined i.tm.mode
2039 and i.tm.regmem fields. If no form of this
2040 instruction supports a memory destination operand,
2041 then we assume the source operand may sometimes be
2042 a memory operand and so we need to store the
2043 destination in the i.rm.reg field. */
2044 if ((i.tm.operand_types[dest] & AnyMem) == 0)
2045 {
2046 i.rm.reg = i.op[dest].regs->reg_num;
2047 i.rm.regmem = i.op[source].regs->reg_num;
2048 }
2049 else
2050 {
2051 i.rm.reg = i.op[source].regs->reg_num;
2052 i.rm.regmem = i.op[dest].regs->reg_num;
2053 }
2054 }
2055 else
2056 { /* If it's not 2 reg operands... */
2057 if (i.mem_operands)
2058 {
2059 unsigned int fake_zero_displacement = 0;
2060 unsigned int op = ((i.types[0] & AnyMem)
2061 ? 0
2062 : (i.types[1] & AnyMem) ? 1 : 2);
2063
2064 default_seg = &ds;
2065
2066 if (! i.base_reg)
2067 {
2068 i.rm.mode = 0;
2069 if (! i.disp_operands)
2070 fake_zero_displacement = 1;
2071 if (! i.index_reg)
2072 {
2073 /* Operand is just <disp> */
2074 if (flag_16bit_code ^ (i.prefix[ADDR_PREFIX] != 0))
2075 {
2076 i.rm.regmem = NO_BASE_REGISTER_16;
2077 i.types[op] &= ~Disp;
2078 i.types[op] |= Disp16;
2079 }
2080 else
2081 {
2082 i.rm.regmem = NO_BASE_REGISTER;
2083 i.types[op] &= ~Disp;
2084 i.types[op] |= Disp32;
2085 }
2086 }
2087 else /* ! i.base_reg && i.index_reg */
2088 {
2089 i.sib.index = i.index_reg->reg_num;
2090 i.sib.base = NO_BASE_REGISTER;
2091 i.sib.scale = i.log2_scale_factor;
2092 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2093 i.types[op] &= ~Disp;
2094 i.types[op] |= Disp32; /* Must be 32 bit. */
2095 }
2096 }
2097 else if (i.base_reg->reg_type & Reg16)
2098 {
2099 switch (i.base_reg->reg_num)
2100 {
2101 case 3: /* (%bx) */
2102 if (! i.index_reg)
2103 i.rm.regmem = 7;
2104 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
2105 i.rm.regmem = i.index_reg->reg_num - 6;
2106 break;
2107 case 5: /* (%bp) */
2108 default_seg = &ss;
2109 if (! i.index_reg)
2110 {
2111 i.rm.regmem = 6;
2112 if ((i.types[op] & Disp) == 0)
2113 {
2114 /* fake (%bp) into 0(%bp) */
2115 i.types[op] |= Disp8;
2116 fake_zero_displacement = 1;
2117 }
2118 }
2119 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
2120 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
2121 break;
2122 default: /* (%si) -> 4 or (%di) -> 5 */
2123 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
2124 }
2125 i.rm.mode = mode_from_disp_size (i.types[op]);
2126 }
2127 else /* i.base_reg and 32 bit mode */
2128 {
2129 i.rm.regmem = i.base_reg->reg_num;
2130 i.sib.base = i.base_reg->reg_num;
2131 if (i.base_reg->reg_num == EBP_REG_NUM)
2132 {
2133 default_seg = &ss;
2134 if (i.disp_operands == 0)
2135 {
2136 fake_zero_displacement = 1;
2137 i.types[op] |= Disp8;
2138 }
2139 }
2140 else if (i.base_reg->reg_num == ESP_REG_NUM)
2141 {
2142 default_seg = &ss;
2143 }
2144 i.sib.scale = i.log2_scale_factor;
2145 if (! i.index_reg)
2146 {
2147 /* <disp>(%esp) becomes two byte modrm
2148 with no index register. We've already
2149 stored the code for esp in i.rm.regmem
2150 ie. ESCAPE_TO_TWO_BYTE_ADDRESSING. Any
2151 base register besides %esp will not use
2152 the extra modrm byte. */
2153 i.sib.index = NO_INDEX_REGISTER;
2154 #if ! SCALE1_WHEN_NO_INDEX
2155 /* Another case where we force the second
2156 modrm byte. */
2157 if (i.log2_scale_factor)
2158 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2159 #endif
2160 }
2161 else
2162 {
2163 i.sib.index = i.index_reg->reg_num;
2164 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2165 }
2166 i.rm.mode = mode_from_disp_size (i.types[op]);
2167 }
2168
2169 if (fake_zero_displacement)
2170 {
2171 /* Fakes a zero displacement assuming that i.types[op]
2172 holds the correct displacement size. */
2173 expressionS *exp;
2174
2175 assert (i.op[op].disps == 0);
2176 exp = &disp_expressions[i.disp_operands++];
2177 i.op[op].disps = exp;
2178 exp->X_op = O_constant;
2179 exp->X_add_number = 0;
2180 exp->X_add_symbol = (symbolS *) 0;
2181 exp->X_op_symbol = (symbolS *) 0;
2182 }
2183 }
2184
2185 /* Fill in i.rm.reg or i.rm.regmem field with register
2186 operand (if any) based on i.tm.extension_opcode.
2187 Again, we must be careful to make sure that
2188 segment/control/debug/test/MMX registers are coded
2189 into the i.rm.reg field. */
2190 if (i.reg_operands)
2191 {
2192 unsigned int op =
2193 ((i.types[0]
2194 & (Reg | RegMMX | RegXMM
2195 | SReg2 | SReg3
2196 | Control | Debug | Test))
2197 ? 0
2198 : ((i.types[1]
2199 & (Reg | RegMMX | RegXMM
2200 | SReg2 | SReg3
2201 | Control | Debug | Test))
2202 ? 1
2203 : 2));
2204 /* If there is an extension opcode to put here, the
2205 register number must be put into the regmem field. */
2206 if (i.tm.extension_opcode != None)
2207 i.rm.regmem = i.op[op].regs->reg_num;
2208 else
2209 i.rm.reg = i.op[op].regs->reg_num;
2210
2211 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2
2212 we must set it to 3 to indicate this is a register
2213 operand in the regmem field. */
2214 if (!i.mem_operands)
2215 i.rm.mode = 3;
2216 }
2217
2218 /* Fill in i.rm.reg field with extension opcode (if any). */
2219 if (i.tm.extension_opcode != None)
2220 i.rm.reg = i.tm.extension_opcode;
2221 }
2222 }
2223 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
2224 {
2225 if (i.tm.base_opcode == POP_SEG_SHORT
2226 && i.op[0].regs->reg_num == 1)
2227 {
2228 as_bad (_("you can't `pop %%cs'"));
2229 return;
2230 }
2231 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
2232 }
2233 else if ((i.tm.base_opcode & ~(D|W)) == MOV_AX_DISP32)
2234 {
2235 default_seg = &ds;
2236 }
2237 else if ((i.tm.opcode_modifier & IsString) != 0)
2238 {
2239 /* For the string instructions that allow a segment override
2240 on one of their operands, the default segment is ds. */
2241 default_seg = &ds;
2242 }
2243
2244 /* If a segment was explicitly specified,
2245 and the specified segment is not the default,
2246 use an opcode prefix to select it.
2247 If we never figured out what the default segment is,
2248 then default_seg will be zero at this point,
2249 and the specified segment prefix will always be used. */
2250 if ((i.seg[0]) && (i.seg[0] != default_seg))
2251 {
2252 if (! add_prefix (i.seg[0]->seg_prefix))
2253 return;
2254 }
2255 }
2256 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2257 {
2258 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
2259 as_warn (_("translating to `%sp'"), i.tm.name);
2260 }
2261 }
2262
2263 /* Handle conversion of 'int $3' --> special int3 insn. */
2264 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
2265 {
2266 i.tm.base_opcode = INT3_OPCODE;
2267 i.imm_operands = 0;
2268 }
2269
2270 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
2271 && i.op[0].disps->X_op == O_constant)
2272 {
2273 /* Convert "jmp constant" (and "call constant") to a jump (call) to
2274 the absolute address given by the constant. Since ix86 jumps and
2275 calls are pc relative, we need to generate a reloc. */
2276 i.op[0].disps->X_add_symbol = &abs_symbol;
2277 i.op[0].disps->X_op = O_symbol;
2278 }
2279
2280 /* We are ready to output the insn. */
2281 {
2282 register char *p;
2283
2284 /* Output jumps. */
2285 if (i.tm.opcode_modifier & Jump)
2286 {
2287 int size;
2288 int code16;
2289 int prefix;
2290
2291 code16 = 0;
2292 if (flag_16bit_code)
2293 code16 = CODE16;
2294
2295 prefix = 0;
2296 if (i.prefix[DATA_PREFIX])
2297 {
2298 prefix = 1;
2299 i.prefixes -= 1;
2300 code16 ^= CODE16;
2301 }
2302
2303 size = 4;
2304 if (code16)
2305 size = 2;
2306
2307 if (i.prefixes != 0 && !intel_syntax)
2308 as_warn (_("skipping prefixes on this instruction"));
2309
2310 /* It's always a symbol; End frag & setup for relax.
2311 Make sure there is enough room in this frag for the largest
2312 instruction we may generate in md_convert_frag. This is 2
2313 bytes for the opcode and room for the prefix and largest
2314 displacement. */
2315 frag_grow (prefix + 2 + size);
2316 insn_size += prefix + 1;
2317 /* Prefix and 1 opcode byte go in fr_fix. */
2318 p = frag_more (prefix + 1);
2319 if (prefix)
2320 *p++ = DATA_PREFIX_OPCODE;
2321 *p = i.tm.base_opcode;
2322 /* 1 possible extra opcode + displacement go in var part.
2323 Pass reloc in fr_var. */
2324 frag_var (rs_machine_dependent,
2325 1 + size,
2326 i.disp_reloc[0],
2327 ((unsigned char) *p == JUMP_PC_RELATIVE
2328 ? ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL) | code16
2329 : ENCODE_RELAX_STATE (COND_JUMP, SMALL) | code16),
2330 i.op[0].disps->X_add_symbol,
2331 i.op[0].disps->X_add_number,
2332 p);
2333 }
2334 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
2335 {
2336 int size;
2337
2338 if (i.tm.opcode_modifier & JumpByte)
2339 {
2340 /* This is a loop or jecxz type instruction. */
2341 size = 1;
2342 if (i.prefix[ADDR_PREFIX])
2343 {
2344 insn_size += 1;
2345 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
2346 i.prefixes -= 1;
2347 }
2348 }
2349 else
2350 {
2351 int code16;
2352
2353 code16 = 0;
2354 if (flag_16bit_code)
2355 code16 = CODE16;
2356
2357 if (i.prefix[DATA_PREFIX])
2358 {
2359 insn_size += 1;
2360 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
2361 i.prefixes -= 1;
2362 code16 ^= CODE16;
2363 }
2364
2365 size = 4;
2366 if (code16)
2367 size = 2;
2368 }
2369
2370 if (i.prefixes != 0 && !intel_syntax)
2371 as_warn (_("skipping prefixes on this instruction"));
2372
2373 if (fits_in_unsigned_byte (i.tm.base_opcode))
2374 {
2375 insn_size += 1 + size;
2376 p = frag_more (1 + size);
2377 }
2378 else
2379 {
2380 /* Opcode can be at most two bytes. */
2381 insn_size += 2 + size;
2382 p = frag_more (2 + size);
2383 *p++ = (i.tm.base_opcode >> 8) & 0xff;
2384 }
2385 *p++ = i.tm.base_opcode & 0xff;
2386
2387 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2388 i.op[0].disps, 1, reloc (size, 1, i.disp_reloc[0]));
2389 }
2390 else if (i.tm.opcode_modifier & JumpInterSegment)
2391 {
2392 int size;
2393 int prefix;
2394 int code16;
2395
2396 code16 = 0;
2397 if (flag_16bit_code)
2398 code16 = CODE16;
2399
2400 prefix = 0;
2401 if (i.prefix[DATA_PREFIX])
2402 {
2403 prefix = 1;
2404 i.prefixes -= 1;
2405 code16 ^= CODE16;
2406 }
2407
2408 size = 4;
2409 if (code16)
2410 size = 2;
2411
2412 if (i.prefixes != 0 && !intel_syntax)
2413 as_warn (_("skipping prefixes on this instruction"));
2414
2415 /* 1 opcode; 2 segment; offset */
2416 insn_size += prefix + 1 + 2 + size;
2417 p = frag_more (prefix + 1 + 2 + size);
2418 if (prefix)
2419 *p++ = DATA_PREFIX_OPCODE;
2420 *p++ = i.tm.base_opcode;
2421 if (i.op[1].imms->X_op == O_constant)
2422 {
2423 offsetT n = i.op[1].imms->X_add_number;
2424
2425 if (size == 2
2426 && !fits_in_unsigned_word (n)
2427 && !fits_in_signed_word (n))
2428 {
2429 as_bad (_("16-bit jump out of range"));
2430 return;
2431 }
2432 md_number_to_chars (p, n, size);
2433 }
2434 else
2435 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2436 i.op[1].imms, 0, reloc (size, 0, i.disp_reloc[0]));
2437 if (i.op[0].imms->X_op != O_constant)
2438 as_bad (_("can't handle non absolute segment in `%s'"),
2439 i.tm.name);
2440 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
2441 }
2442 else
2443 {
2444 /* Output normal instructions here. */
2445 unsigned char *q;
2446
2447 /* The prefix bytes. */
2448 for (q = i.prefix;
2449 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
2450 q++)
2451 {
2452 if (*q)
2453 {
2454 insn_size += 1;
2455 p = frag_more (1);
2456 md_number_to_chars (p, (valueT) *q, 1);
2457 }
2458 }
2459
2460 /* Now the opcode; be careful about word order here! */
2461 if (fits_in_unsigned_byte (i.tm.base_opcode))
2462 {
2463 insn_size += 1;
2464 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
2465 }
2466 else if (fits_in_unsigned_word (i.tm.base_opcode))
2467 {
2468 insn_size += 2;
2469 p = frag_more (2);
2470 /* Put out high byte first: can't use md_number_to_chars! */
2471 *p++ = (i.tm.base_opcode >> 8) & 0xff;
2472 *p = i.tm.base_opcode & 0xff;
2473 }
2474 else
2475 { /* Opcode is either 3 or 4 bytes. */
2476 if (i.tm.base_opcode & 0xff000000)
2477 {
2478 insn_size += 4;
2479 p = frag_more (4);
2480 *p++ = (i.tm.base_opcode >> 24) & 0xff;
2481 }
2482 else
2483 {
2484 insn_size += 3;
2485 p = frag_more (3);
2486 }
2487 *p++ = (i.tm.base_opcode >> 16) & 0xff;
2488 *p++ = (i.tm.base_opcode >> 8) & 0xff;
2489 *p = (i.tm.base_opcode) & 0xff;
2490 }
2491
2492 /* Now the modrm byte and sib byte (if present). */
2493 if (i.tm.opcode_modifier & Modrm)
2494 {
2495 insn_size += 1;
2496 p = frag_more (1);
2497 md_number_to_chars (p,
2498 (valueT) (i.rm.regmem << 0
2499 | i.rm.reg << 3
2500 | i.rm.mode << 6),
2501 1);
2502 /* If i.rm.regmem == ESP (4)
2503 && i.rm.mode != (Register mode)
2504 && not 16 bit
2505 ==> need second modrm byte. */
2506 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
2507 && i.rm.mode != 3
2508 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
2509 {
2510 insn_size += 1;
2511 p = frag_more (1);
2512 md_number_to_chars (p,
2513 (valueT) (i.sib.base << 0
2514 | i.sib.index << 3
2515 | i.sib.scale << 6),
2516 1);
2517 }
2518 }
2519
2520 if (i.disp_operands)
2521 {
2522 register unsigned int n;
2523
2524 for (n = 0; n < i.operands; n++)
2525 {
2526 if (i.types[n] & Disp)
2527 {
2528 if (i.op[n].disps->X_op == O_constant)
2529 {
2530 int size;
2531 offsetT val;
2532
2533 size = 4;
2534 if (i.types[n] & (Disp8 | Disp16))
2535 {
2536 size = 2;
2537 if (i.types[n] & Disp8)
2538 size = 1;
2539 }
2540 val = offset_in_range (i.op[n].disps->X_add_number,
2541 size);
2542 insn_size += size;
2543 p = frag_more (size);
2544 md_number_to_chars (p, val, size);
2545 }
2546 else
2547 {
2548 int size = 4;
2549
2550 if (i.types[n] & Disp16)
2551 size = 2;
2552
2553 insn_size += size;
2554 p = frag_more (size);
2555 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2556 i.op[n].disps, 0,
2557 reloc (size, 0, i.disp_reloc[n]));
2558 }
2559 }
2560 }
2561 }
2562
2563 /* Output immediate. */
2564 if (i.imm_operands)
2565 {
2566 register unsigned int n;
2567
2568 for (n = 0; n < i.operands; n++)
2569 {
2570 if (i.types[n] & Imm)
2571 {
2572 if (i.op[n].imms->X_op == O_constant)
2573 {
2574 int size;
2575 offsetT val;
2576
2577 size = 4;
2578 if (i.types[n] & (Imm8 | Imm8S | Imm16))
2579 {
2580 size = 2;
2581 if (i.types[n] & (Imm8 | Imm8S))
2582 size = 1;
2583 }
2584 val = offset_in_range (i.op[n].imms->X_add_number,
2585 size);
2586 insn_size += size;
2587 p = frag_more (size);
2588 md_number_to_chars (p, val, size);
2589 }
2590 else
2591 {
2592 /* Not absolute_section.
2593 Need a 32-bit fixup (don't support 8bit
2594 non-absolute imms). Try to support other
2595 sizes ... */
2596 #ifdef BFD_ASSEMBLER
2597 enum bfd_reloc_code_real reloc_type;
2598 #else
2599 int reloc_type;
2600 #endif
2601 int size = 4;
2602
2603 if (i.types[n] & Imm16)
2604 size = 2;
2605 else if (i.types[n] & (Imm8 | Imm8S))
2606 size = 1;
2607
2608 insn_size += size;
2609 p = frag_more (size);
2610 reloc_type = reloc (size, 0, i.disp_reloc[0]);
2611 #ifdef BFD_ASSEMBLER
2612 if (reloc_type == BFD_RELOC_32
2613 && GOT_symbol
2614 && GOT_symbol == i.op[n].imms->X_add_symbol
2615 && (i.op[n].imms->X_op == O_symbol
2616 || (i.op[n].imms->X_op == O_add
2617 && ((symbol_get_value_expression
2618 (i.op[n].imms->X_op_symbol)->X_op)
2619 == O_subtract))))
2620 {
2621 reloc_type = BFD_RELOC_386_GOTPC;
2622 i.op[n].imms->X_add_number += 3;
2623 }
2624 #endif
2625 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2626 i.op[n].imms, 0, reloc_type);
2627 }
2628 }
2629 }
2630 }
2631 }
2632
2633 dwarf2_emit_insn (insn_size);
2634
2635 #ifdef DEBUG386
2636 if (flag_debug)
2637 {
2638 pi (line, &i);
2639 }
2640 #endif /* DEBUG386 */
2641 }
2642 }
2643 \f
2644 static int i386_immediate PARAMS ((char *));
2645
2646 static int
2647 i386_immediate (imm_start)
2648 char *imm_start;
2649 {
2650 char *save_input_line_pointer;
2651 segT exp_seg = 0;
2652 expressionS *exp;
2653
2654 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
2655 {
2656 as_bad (_("only 1 or 2 immediate operands are allowed"));
2657 return 0;
2658 }
2659
2660 exp = &im_expressions[i.imm_operands++];
2661 i.op[this_operand].imms = exp;
2662
2663 if (is_space_char (*imm_start))
2664 ++imm_start;
2665
2666 save_input_line_pointer = input_line_pointer;
2667 input_line_pointer = imm_start;
2668
2669 #ifndef LEX_AT
2670 {
2671 /* We can have operands of the form
2672 <symbol>@GOTOFF+<nnn>
2673 Take the easy way out here and copy everything
2674 into a temporary buffer... */
2675 register char *cp;
2676
2677 cp = strchr (input_line_pointer, '@');
2678 if (cp != NULL)
2679 {
2680 char *tmpbuf;
2681 int len = 0;
2682 int first;
2683
2684 /* GOT relocations are not supported in 16 bit mode. */
2685 if (flag_16bit_code)
2686 as_bad (_("GOT relocations not supported in 16 bit mode"));
2687
2688 if (GOT_symbol == NULL)
2689 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
2690
2691 if (strncmp (cp + 1, "PLT", 3) == 0)
2692 {
2693 i.disp_reloc[this_operand] = BFD_RELOC_386_PLT32;
2694 len = 3;
2695 }
2696 else if (strncmp (cp + 1, "GOTOFF", 6) == 0)
2697 {
2698 i.disp_reloc[this_operand] = BFD_RELOC_386_GOTOFF;
2699 len = 6;
2700 }
2701 else if (strncmp (cp + 1, "GOT", 3) == 0)
2702 {
2703 i.disp_reloc[this_operand] = BFD_RELOC_386_GOT32;
2704 len = 3;
2705 }
2706 else
2707 as_bad (_("bad reloc specifier in expression"));
2708
2709 /* Replace the relocation token with ' ', so that errors like
2710 foo@GOTOFF1 will be detected. */
2711 first = cp - input_line_pointer;
2712 tmpbuf = (char *) alloca (strlen (input_line_pointer));
2713 memcpy (tmpbuf, input_line_pointer, first);
2714 tmpbuf[first] = ' ';
2715 strcpy (tmpbuf + first + 1, cp + 1 + len);
2716 input_line_pointer = tmpbuf;
2717 }
2718 }
2719 #endif
2720
2721 exp_seg = expression (exp);
2722
2723 SKIP_WHITESPACE ();
2724 if (*input_line_pointer)
2725 as_bad (_("ignoring junk `%s' after expression"), input_line_pointer);
2726
2727 input_line_pointer = save_input_line_pointer;
2728
2729 if (exp->X_op == O_absent || exp->X_op == O_big)
2730 {
2731 /* Missing or bad expr becomes absolute 0. */
2732 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
2733 imm_start);
2734 exp->X_op = O_constant;
2735 exp->X_add_number = 0;
2736 exp->X_add_symbol = (symbolS *) 0;
2737 exp->X_op_symbol = (symbolS *) 0;
2738 }
2739
2740 if (exp->X_op == O_constant)
2741 {
2742 /* Size it properly later. */
2743 i.types[this_operand] |= Imm32;
2744 }
2745 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
2746 else if (1
2747 #ifdef BFD_ASSEMBLER
2748 && OUTPUT_FLAVOR == bfd_target_aout_flavour
2749 #endif
2750 && exp_seg != text_section
2751 && exp_seg != data_section
2752 && exp_seg != bss_section
2753 && exp_seg != undefined_section
2754 #ifdef BFD_ASSEMBLER
2755 && !bfd_is_com_section (exp_seg)
2756 #endif
2757 )
2758 {
2759 #ifdef BFD_ASSEMBLER
2760 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
2761 #else
2762 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
2763 #endif
2764 return 0;
2765 }
2766 #endif
2767 else
2768 {
2769 /* This is an address. The size of the address will be
2770 determined later, depending on destination register,
2771 suffix, or the default for the section. We exclude
2772 Imm8S here so that `push $foo' and other instructions
2773 with an Imm8S form will use Imm16 or Imm32. */
2774 i.types[this_operand] |= (Imm8 | Imm16 | Imm32);
2775 }
2776
2777 return 1;
2778 }
2779
2780 static int i386_scale PARAMS ((char *));
2781
2782 static int
2783 i386_scale (scale)
2784 char *scale;
2785 {
2786 if (!isdigit (*scale))
2787 goto bad_scale;
2788
2789 switch (*scale)
2790 {
2791 case '0':
2792 case '1':
2793 i.log2_scale_factor = 0;
2794 break;
2795 case '2':
2796 i.log2_scale_factor = 1;
2797 break;
2798 case '4':
2799 i.log2_scale_factor = 2;
2800 break;
2801 case '8':
2802 i.log2_scale_factor = 3;
2803 break;
2804 default:
2805 bad_scale:
2806 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
2807 scale);
2808 return 0;
2809 }
2810 if (i.log2_scale_factor != 0 && ! i.index_reg)
2811 {
2812 as_warn (_("scale factor of %d without an index register"),
2813 1 << i.log2_scale_factor);
2814 #if SCALE1_WHEN_NO_INDEX
2815 i.log2_scale_factor = 0;
2816 #endif
2817 }
2818 return 1;
2819 }
2820
2821 static int i386_displacement PARAMS ((char *, char *));
2822
2823 static int
2824 i386_displacement (disp_start, disp_end)
2825 char *disp_start;
2826 char *disp_end;
2827 {
2828 register expressionS *exp;
2829 segT exp_seg = 0;
2830 char *save_input_line_pointer;
2831 int bigdisp = Disp32;
2832
2833 if (flag_16bit_code ^ (i.prefix[ADDR_PREFIX] != 0))
2834 bigdisp = Disp16;
2835 i.types[this_operand] |= bigdisp;
2836
2837 exp = &disp_expressions[i.disp_operands];
2838 i.op[this_operand].disps = exp;
2839 i.disp_operands++;
2840 save_input_line_pointer = input_line_pointer;
2841 input_line_pointer = disp_start;
2842 END_STRING_AND_SAVE (disp_end);
2843
2844 #ifndef GCC_ASM_O_HACK
2845 #define GCC_ASM_O_HACK 0
2846 #endif
2847 #if GCC_ASM_O_HACK
2848 END_STRING_AND_SAVE (disp_end + 1);
2849 if ((i.types[this_operand] & BaseIndex) != 0
2850 && displacement_string_end[-1] == '+')
2851 {
2852 /* This hack is to avoid a warning when using the "o"
2853 constraint within gcc asm statements.
2854 For instance:
2855
2856 #define _set_tssldt_desc(n,addr,limit,type) \
2857 __asm__ __volatile__ ( \
2858 "movw %w2,%0\n\t" \
2859 "movw %w1,2+%0\n\t" \
2860 "rorl $16,%1\n\t" \
2861 "movb %b1,4+%0\n\t" \
2862 "movb %4,5+%0\n\t" \
2863 "movb $0,6+%0\n\t" \
2864 "movb %h1,7+%0\n\t" \
2865 "rorl $16,%1" \
2866 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
2867
2868 This works great except that the output assembler ends
2869 up looking a bit weird if it turns out that there is
2870 no offset. You end up producing code that looks like:
2871
2872 #APP
2873 movw $235,(%eax)
2874 movw %dx,2+(%eax)
2875 rorl $16,%edx
2876 movb %dl,4+(%eax)
2877 movb $137,5+(%eax)
2878 movb $0,6+(%eax)
2879 movb %dh,7+(%eax)
2880 rorl $16,%edx
2881 #NO_APP
2882
2883 So here we provide the missing zero. */
2884
2885 *displacement_string_end = '0';
2886 }
2887 #endif
2888 #ifndef LEX_AT
2889 {
2890 /* We can have operands of the form
2891 <symbol>@GOTOFF+<nnn>
2892 Take the easy way out here and copy everything
2893 into a temporary buffer... */
2894 register char *cp;
2895
2896 cp = strchr (input_line_pointer, '@');
2897 if (cp != NULL)
2898 {
2899 char *tmpbuf;
2900 int len = 0;
2901 int first;
2902
2903 /* GOT relocations are not supported in 16 bit mode. */
2904 if (flag_16bit_code)
2905 as_bad (_("GOT relocations not supported in 16 bit mode"));
2906
2907 if (GOT_symbol == NULL)
2908 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
2909
2910 if (strncmp (cp + 1, "PLT", 3) == 0)
2911 {
2912 i.disp_reloc[this_operand] = BFD_RELOC_386_PLT32;
2913 len = 3;
2914 }
2915 else if (strncmp (cp + 1, "GOTOFF", 6) == 0)
2916 {
2917 i.disp_reloc[this_operand] = BFD_RELOC_386_GOTOFF;
2918 len = 6;
2919 }
2920 else if (strncmp (cp + 1, "GOT", 3) == 0)
2921 {
2922 i.disp_reloc[this_operand] = BFD_RELOC_386_GOT32;
2923 len = 3;
2924 }
2925 else
2926 as_bad (_("bad reloc specifier in expression"));
2927
2928 /* Replace the relocation token with ' ', so that errors like
2929 foo@GOTOFF1 will be detected. */
2930 first = cp - input_line_pointer;
2931 tmpbuf = (char *) alloca (strlen (input_line_pointer));
2932 memcpy (tmpbuf, input_line_pointer, first);
2933 tmpbuf[first] = ' ';
2934 strcpy (tmpbuf + first + 1, cp + 1 + len);
2935 input_line_pointer = tmpbuf;
2936 }
2937 }
2938 #endif
2939
2940 exp_seg = expression (exp);
2941
2942 #ifdef BFD_ASSEMBLER
2943 /* We do this to make sure that the section symbol is in
2944 the symbol table. We will ultimately change the relocation
2945 to be relative to the beginning of the section. */
2946 if (i.disp_reloc[this_operand] == BFD_RELOC_386_GOTOFF)
2947 {
2948 if (S_IS_LOCAL(exp->X_add_symbol)
2949 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
2950 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
2951 assert (exp->X_op == O_symbol);
2952 exp->X_op = O_subtract;
2953 exp->X_op_symbol = GOT_symbol;
2954 i.disp_reloc[this_operand] = BFD_RELOC_32;
2955 }
2956 #endif
2957
2958 SKIP_WHITESPACE ();
2959 if (*input_line_pointer)
2960 as_bad (_("ignoring junk `%s' after expression"),
2961 input_line_pointer);
2962 #if GCC_ASM_O_HACK
2963 RESTORE_END_STRING (disp_end + 1);
2964 #endif
2965 RESTORE_END_STRING (disp_end);
2966 input_line_pointer = save_input_line_pointer;
2967
2968 if (exp->X_op == O_absent || exp->X_op == O_big)
2969 {
2970 /* Missing or bad expr becomes absolute 0. */
2971 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
2972 disp_start);
2973 exp->X_op = O_constant;
2974 exp->X_add_number = 0;
2975 exp->X_add_symbol = (symbolS *) 0;
2976 exp->X_op_symbol = (symbolS *) 0;
2977 }
2978
2979 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
2980 if (exp->X_op != O_constant
2981 #ifdef BFD_ASSEMBLER
2982 && OUTPUT_FLAVOR == bfd_target_aout_flavour
2983 #endif
2984 && exp_seg != text_section
2985 && exp_seg != data_section
2986 && exp_seg != bss_section
2987 && exp_seg != undefined_section)
2988 {
2989 #ifdef BFD_ASSEMBLER
2990 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
2991 #else
2992 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
2993 #endif
2994 return 0;
2995 }
2996 #endif
2997 return 1;
2998 }
2999
3000 static int i386_index_check PARAMS((const char *));
3001
3002 /* Make sure the memory operand we've been dealt is valid.
3003 Return 1 on success, 0 on a failure. */
3004
3005 static int
3006 i386_index_check (operand_string)
3007 const char *operand_string;
3008 {
3009 #if INFER_ADDR_PREFIX
3010 int fudged = 0;
3011
3012 tryprefix:
3013 #endif
3014 if (flag_16bit_code ^ (i.prefix[ADDR_PREFIX] != 0)
3015 /* 16 bit mode checks. */
3016 ? ((i.base_reg
3017 && ((i.base_reg->reg_type & (Reg16|BaseIndex))
3018 != (Reg16|BaseIndex)))
3019 || (i.index_reg
3020 && (((i.index_reg->reg_type & (Reg16|BaseIndex))
3021 != (Reg16|BaseIndex))
3022 || ! (i.base_reg
3023 && i.base_reg->reg_num < 6
3024 && i.index_reg->reg_num >= 6
3025 && i.log2_scale_factor == 0))))
3026 /* 32 bit mode checks. */
3027 : ((i.base_reg
3028 && (i.base_reg->reg_type & Reg32) == 0)
3029 || (i.index_reg
3030 && ((i.index_reg->reg_type & (Reg32|BaseIndex))
3031 != (Reg32|BaseIndex)))))
3032 {
3033 #if INFER_ADDR_PREFIX
3034 if (i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0')
3035 {
3036 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
3037 i.prefixes += 1;
3038 /* Change the size of any displacement too. At most one of
3039 Disp16 or Disp32 is set.
3040 FIXME. There doesn't seem to be any real need for separate
3041 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
3042 Removing them would probably clean up the code quite a lot. */
3043 if (i.types[this_operand] & (Disp16|Disp32))
3044 i.types[this_operand] ^= (Disp16|Disp32);
3045 fudged = 1;
3046 goto tryprefix;
3047 }
3048 if (fudged)
3049 as_bad (_("`%s' is not a valid base/index expression"),
3050 operand_string);
3051 else
3052 #endif
3053 as_bad (_("`%s' is not a valid %s bit base/index expression"),
3054 operand_string,
3055 flag_16bit_code ^ (i.prefix[ADDR_PREFIX] != 0) ? "16" : "32");
3056 return 0;
3057 }
3058 return 1;
3059 }
3060
3061 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
3062 on error. */
3063
3064 static int
3065 i386_operand (operand_string)
3066 char *operand_string;
3067 {
3068 const reg_entry *r;
3069 char *end_op;
3070 char *op_string = operand_string;
3071
3072 if (is_space_char (*op_string))
3073 ++op_string;
3074
3075 /* We check for an absolute prefix (differentiating,
3076 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
3077 if (*op_string == ABSOLUTE_PREFIX)
3078 {
3079 ++op_string;
3080 if (is_space_char (*op_string))
3081 ++op_string;
3082 i.types[this_operand] |= JumpAbsolute;
3083 }
3084
3085 /* Check if operand is a register. */
3086 if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
3087 && (r = parse_register (op_string, &end_op)) != NULL)
3088 {
3089 /* Check for a segment override by searching for ':' after a
3090 segment register. */
3091 op_string = end_op;
3092 if (is_space_char (*op_string))
3093 ++op_string;
3094 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
3095 {
3096 switch (r->reg_num)
3097 {
3098 case 0:
3099 i.seg[i.mem_operands] = &es;
3100 break;
3101 case 1:
3102 i.seg[i.mem_operands] = &cs;
3103 break;
3104 case 2:
3105 i.seg[i.mem_operands] = &ss;
3106 break;
3107 case 3:
3108 i.seg[i.mem_operands] = &ds;
3109 break;
3110 case 4:
3111 i.seg[i.mem_operands] = &fs;
3112 break;
3113 case 5:
3114 i.seg[i.mem_operands] = &gs;
3115 break;
3116 }
3117
3118 /* Skip the ':' and whitespace. */
3119 ++op_string;
3120 if (is_space_char (*op_string))
3121 ++op_string;
3122
3123 if (!is_digit_char (*op_string)
3124 && !is_identifier_char (*op_string)
3125 && *op_string != '('
3126 && *op_string != ABSOLUTE_PREFIX)
3127 {
3128 as_bad (_("bad memory operand `%s'"), op_string);
3129 return 0;
3130 }
3131 /* Handle case of %es:*foo. */
3132 if (*op_string == ABSOLUTE_PREFIX)
3133 {
3134 ++op_string;
3135 if (is_space_char (*op_string))
3136 ++op_string;
3137 i.types[this_operand] |= JumpAbsolute;
3138 }
3139 goto do_memory_reference;
3140 }
3141 if (*op_string)
3142 {
3143 as_bad (_("junk `%s' after register"), op_string);
3144 return 0;
3145 }
3146 i.types[this_operand] |= r->reg_type & ~BaseIndex;
3147 i.op[this_operand].regs = r;
3148 i.reg_operands++;
3149 }
3150 else if (*op_string == REGISTER_PREFIX)
3151 {
3152 as_bad (_("bad register name `%s'"), op_string);
3153 return 0;
3154 }
3155 else if (*op_string == IMMEDIATE_PREFIX)
3156 {
3157 ++op_string;
3158 if (i.types[this_operand] & JumpAbsolute)
3159 {
3160 as_bad (_("immediate operand illegal with absolute jump"));
3161 return 0;
3162 }
3163 if (!i386_immediate (op_string))
3164 return 0;
3165 }
3166 else if (is_digit_char (*op_string)
3167 || is_identifier_char (*op_string)
3168 || *op_string == '(' )
3169 {
3170 /* This is a memory reference of some sort. */
3171 char *base_string;
3172
3173 /* Start and end of displacement string expression (if found). */
3174 char *displacement_string_start;
3175 char *displacement_string_end;
3176
3177 do_memory_reference:
3178 if ((i.mem_operands == 1
3179 && (current_templates->start->opcode_modifier & IsString) == 0)
3180 || i.mem_operands == 2)
3181 {
3182 as_bad (_("too many memory references for `%s'"),
3183 current_templates->start->name);
3184 return 0;
3185 }
3186
3187 /* Check for base index form. We detect the base index form by
3188 looking for an ')' at the end of the operand, searching
3189 for the '(' matching it, and finding a REGISTER_PREFIX or ','
3190 after the '('. */
3191 base_string = op_string + strlen (op_string);
3192
3193 --base_string;
3194 if (is_space_char (*base_string))
3195 --base_string;
3196
3197 /* If we only have a displacement, set-up for it to be parsed later. */
3198 displacement_string_start = op_string;
3199 displacement_string_end = base_string + 1;
3200
3201 if (*base_string == ')')
3202 {
3203 char *temp_string;
3204 unsigned int parens_balanced = 1;
3205 /* We've already checked that the number of left & right ()'s are
3206 equal, so this loop will not be infinite. */
3207 do
3208 {
3209 base_string--;
3210 if (*base_string == ')')
3211 parens_balanced++;
3212 if (*base_string == '(')
3213 parens_balanced--;
3214 }
3215 while (parens_balanced);
3216
3217 temp_string = base_string;
3218
3219 /* Skip past '(' and whitespace. */
3220 ++base_string;
3221 if (is_space_char (*base_string))
3222 ++base_string;
3223
3224 if (*base_string == ','
3225 || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3226 && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
3227 {
3228 displacement_string_end = temp_string;
3229
3230 i.types[this_operand] |= BaseIndex;
3231
3232 if (i.base_reg)
3233 {
3234 base_string = end_op;
3235 if (is_space_char (*base_string))
3236 ++base_string;
3237 }
3238
3239 /* There may be an index reg or scale factor here. */
3240 if (*base_string == ',')
3241 {
3242 ++base_string;
3243 if (is_space_char (*base_string))
3244 ++base_string;
3245
3246 if ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3247 && (i.index_reg = parse_register (base_string, &end_op)) != NULL)
3248 {
3249 base_string = end_op;
3250 if (is_space_char (*base_string))
3251 ++base_string;
3252 if (*base_string == ',')
3253 {
3254 ++base_string;
3255 if (is_space_char (*base_string))
3256 ++base_string;
3257 }
3258 else if (*base_string != ')' )
3259 {
3260 as_bad (_("expecting `,' or `)' after index register in `%s'"),
3261 operand_string);
3262 return 0;
3263 }
3264 }
3265 else if (*base_string == REGISTER_PREFIX)
3266 {
3267 as_bad (_("bad register name `%s'"), base_string);
3268 return 0;
3269 }
3270
3271 /* Check for scale factor. */
3272 if (isdigit ((unsigned char) *base_string))
3273 {
3274 if (!i386_scale (base_string))
3275 return 0;
3276
3277 ++base_string;
3278 if (is_space_char (*base_string))
3279 ++base_string;
3280 if (*base_string != ')')
3281 {
3282 as_bad (_("expecting `)' after scale factor in `%s'"),
3283 operand_string);
3284 return 0;
3285 }
3286 }
3287 else if (!i.index_reg)
3288 {
3289 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
3290 *base_string);
3291 return 0;
3292 }
3293 }
3294 else if (*base_string != ')')
3295 {
3296 as_bad (_("expecting `,' or `)' after base register in `%s'"),
3297 operand_string);
3298 return 0;
3299 }
3300 }
3301 else if (*base_string == REGISTER_PREFIX)
3302 {
3303 as_bad (_("bad register name `%s'"), base_string);
3304 return 0;
3305 }
3306 }
3307
3308 /* If there's an expression beginning the operand, parse it,
3309 assuming displacement_string_start and
3310 displacement_string_end are meaningful. */
3311 if (displacement_string_start != displacement_string_end)
3312 {
3313 if (!i386_displacement (displacement_string_start,
3314 displacement_string_end))
3315 return 0;
3316 }
3317
3318 /* Special case for (%dx) while doing input/output op. */
3319 if (i.base_reg
3320 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
3321 && i.index_reg == 0
3322 && i.log2_scale_factor == 0
3323 && i.seg[i.mem_operands] == 0
3324 && (i.types[this_operand] & Disp) == 0)
3325 {
3326 i.types[this_operand] = InOutPortReg;
3327 return 1;
3328 }
3329
3330 if (i386_index_check (operand_string) == 0)
3331 return 0;
3332 i.mem_operands++;
3333 }
3334 else
3335 {
3336 /* It's not a memory operand; argh! */
3337 as_bad (_("invalid char %s beginning operand %d `%s'"),
3338 output_invalid (*op_string),
3339 this_operand + 1,
3340 op_string);
3341 return 0;
3342 }
3343 return 1; /* Normal return. */
3344 }
3345 \f
3346 /* md_estimate_size_before_relax()
3347
3348 Called just before relax() for rs_machine_dependent frags. The x86
3349 assembler uses these frags to handle variable size jump
3350 instructions.
3351
3352 Any symbol that is now undefined will not become defined.
3353 Return the correct fr_subtype in the frag.
3354 Return the initial "guess for variable size of frag" to caller.
3355 The guess is actually the growth beyond the fixed part. Whatever
3356 we do to grow the fixed or variable part contributes to our
3357 returned value. */
3358
3359 int
3360 md_estimate_size_before_relax (fragP, segment)
3361 register fragS *fragP;
3362 register segT segment;
3363 {
3364 /* We've already got fragP->fr_subtype right; all we have to do is
3365 check for un-relaxable symbols. On an ELF system, we can't relax
3366 an externally visible symbol, because it may be overridden by a
3367 shared library. */
3368 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
3369 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
3370 || S_IS_EXTERNAL (fragP->fr_symbol)
3371 || S_IS_WEAK (fragP->fr_symbol)
3372 #endif
3373 )
3374 {
3375 /* Symbol is undefined in this segment, or we need to keep a
3376 reloc so that weak symbols can be overridden. */
3377 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
3378 #ifdef BFD_ASSEMBLER
3379 enum bfd_reloc_code_real reloc_type;
3380 #else
3381 int reloc_type;
3382 #endif
3383 unsigned char *opcode;
3384 int old_fr_fix;
3385
3386 if (fragP->fr_var != NO_RELOC)
3387 reloc_type = fragP->fr_var;
3388 else if (size == 2)
3389 reloc_type = BFD_RELOC_16_PCREL;
3390 else
3391 reloc_type = BFD_RELOC_32_PCREL;
3392
3393 old_fr_fix = fragP->fr_fix;
3394 opcode = (unsigned char *) fragP->fr_opcode;
3395
3396 switch (opcode[0])
3397 {
3398 case JUMP_PC_RELATIVE:
3399 /* Make jmp (0xeb) a dword displacement jump. */
3400 opcode[0] = 0xe9;
3401 fragP->fr_fix += size;
3402 fix_new (fragP, old_fr_fix, size,
3403 fragP->fr_symbol,
3404 fragP->fr_offset, 1,
3405 reloc_type);
3406 break;
3407
3408 default:
3409 /* This changes the byte-displacement jump 0x7N
3410 to the dword-displacement jump 0x0f,0x8N. */
3411 opcode[1] = opcode[0] + 0x10;
3412 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
3413 /* We've added an opcode byte. */
3414 fragP->fr_fix += 1 + size;
3415 fix_new (fragP, old_fr_fix + 1, size,
3416 fragP->fr_symbol,
3417 fragP->fr_offset, 1,
3418 reloc_type);
3419 break;
3420 }
3421 frag_wane (fragP);
3422 return fragP->fr_fix - old_fr_fix;
3423 }
3424 /* Guess a short jump. */
3425 return 1;
3426 }
3427
3428 /* Called after relax() is finished.
3429
3430 In: Address of frag.
3431 fr_type == rs_machine_dependent.
3432 fr_subtype is what the address relaxed to.
3433
3434 Out: Any fixSs and constants are set up.
3435 Caller will turn frag into a ".space 0". */
3436
3437 #ifndef BFD_ASSEMBLER
3438 void
3439 md_convert_frag (headers, sec, fragP)
3440 object_headers *headers ATTRIBUTE_UNUSED;
3441 segT sec ATTRIBUTE_UNUSED;
3442 register fragS *fragP;
3443 #else
3444 void
3445 md_convert_frag (abfd, sec, fragP)
3446 bfd *abfd ATTRIBUTE_UNUSED;
3447 segT sec ATTRIBUTE_UNUSED;
3448 register fragS *fragP;
3449 #endif
3450 {
3451 register unsigned char *opcode;
3452 unsigned char *where_to_put_displacement = NULL;
3453 offsetT target_address;
3454 offsetT opcode_address;
3455 unsigned int extension = 0;
3456 offsetT displacement_from_opcode_start;
3457
3458 opcode = (unsigned char *) fragP->fr_opcode;
3459
3460 /* Address we want to reach in file space. */
3461 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
3462 #ifdef BFD_ASSEMBLER
3463 /* Not needed otherwise? */
3464 target_address += symbol_get_frag (fragP->fr_symbol)->fr_address;
3465 #endif
3466
3467 /* Address opcode resides at in file space. */
3468 opcode_address = fragP->fr_address + fragP->fr_fix;
3469
3470 /* Displacement from opcode start to fill into instruction. */
3471 displacement_from_opcode_start = target_address - opcode_address;
3472
3473 switch (fragP->fr_subtype)
3474 {
3475 case ENCODE_RELAX_STATE (COND_JUMP, SMALL):
3476 case ENCODE_RELAX_STATE (COND_JUMP, SMALL16):
3477 case ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL):
3478 case ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL16):
3479 /* Don't have to change opcode. */
3480 extension = 1; /* 1 opcode + 1 displacement */
3481 where_to_put_displacement = &opcode[1];
3482 break;
3483
3484 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
3485 extension = 5; /* 2 opcode + 4 displacement */
3486 opcode[1] = opcode[0] + 0x10;
3487 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
3488 where_to_put_displacement = &opcode[2];
3489 break;
3490
3491 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
3492 extension = 4; /* 1 opcode + 4 displacement */
3493 opcode[0] = 0xe9;
3494 where_to_put_displacement = &opcode[1];
3495 break;
3496
3497 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
3498 extension = 3; /* 2 opcode + 2 displacement */
3499 opcode[1] = opcode[0] + 0x10;
3500 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
3501 where_to_put_displacement = &opcode[2];
3502 break;
3503
3504 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
3505 extension = 2; /* 1 opcode + 2 displacement */
3506 opcode[0] = 0xe9;
3507 where_to_put_displacement = &opcode[1];
3508 break;
3509
3510 default:
3511 BAD_CASE (fragP->fr_subtype);
3512 break;
3513 }
3514 /* Now put displacement after opcode. */
3515 md_number_to_chars ((char *) where_to_put_displacement,
3516 (valueT) (displacement_from_opcode_start - extension),
3517 SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
3518 fragP->fr_fix += extension;
3519 }
3520 \f
3521 /* Size of byte displacement jmp. */
3522 int md_short_jump_size = 2;
3523
3524 /* Size of dword displacement jmp. */
3525 int md_long_jump_size = 5;
3526
3527 /* Size of relocation record. */
3528 const int md_reloc_size = 8;
3529
3530 void
3531 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
3532 char *ptr;
3533 addressT from_addr, to_addr;
3534 fragS *frag ATTRIBUTE_UNUSED;
3535 symbolS *to_symbol ATTRIBUTE_UNUSED;
3536 {
3537 offsetT offset;
3538
3539 offset = to_addr - (from_addr + 2);
3540 /* Opcode for byte-disp jump. */
3541 md_number_to_chars (ptr, (valueT) 0xeb, 1);
3542 md_number_to_chars (ptr + 1, (valueT) offset, 1);
3543 }
3544
3545 void
3546 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
3547 char *ptr;
3548 addressT from_addr, to_addr;
3549 fragS *frag ATTRIBUTE_UNUSED;
3550 symbolS *to_symbol ATTRIBUTE_UNUSED;
3551 {
3552 offsetT offset;
3553
3554 offset = to_addr - (from_addr + 5);
3555 md_number_to_chars (ptr, (valueT) 0xe9, 1);
3556 md_number_to_chars (ptr + 1, (valueT) offset, 4);
3557 }
3558 \f
3559 /* Apply a fixup (fixS) to segment data, once it has been determined
3560 by our caller that we have all the info we need to fix it up.
3561
3562 On the 386, immediates, displacements, and data pointers are all in
3563 the same (little-endian) format, so we don't need to care about which
3564 we are handling. */
3565
3566 int
3567 md_apply_fix3 (fixP, valp, seg)
3568 /* The fix we're to put in. */
3569 fixS *fixP;
3570
3571 /* Pointer to the value of the bits. */
3572 valueT *valp;
3573
3574 /* Segment fix is from. */
3575 segT seg ATTRIBUTE_UNUSED;
3576 {
3577 register char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
3578 valueT value = *valp;
3579
3580 #if defined (BFD_ASSEMBLER) && !defined (TE_Mach)
3581 if (fixP->fx_pcrel)
3582 {
3583 switch (fixP->fx_r_type)
3584 {
3585 default:
3586 break;
3587
3588 case BFD_RELOC_32:
3589 fixP->fx_r_type = BFD_RELOC_32_PCREL;
3590 break;
3591 case BFD_RELOC_16:
3592 fixP->fx_r_type = BFD_RELOC_16_PCREL;
3593 break;
3594 case BFD_RELOC_8:
3595 fixP->fx_r_type = BFD_RELOC_8_PCREL;
3596 break;
3597 }
3598 }
3599
3600 /* This is a hack. There should be a better way to handle this.
3601 This covers for the fact that bfd_install_relocation will
3602 subtract the current location (for partial_inplace, PC relative
3603 relocations); see more below. */
3604 if ((fixP->fx_r_type == BFD_RELOC_32_PCREL
3605 || fixP->fx_r_type == BFD_RELOC_16_PCREL
3606 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
3607 && fixP->fx_addsy)
3608 {
3609 #ifndef OBJ_AOUT
3610 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
3611 #ifdef TE_PE
3612 || OUTPUT_FLAVOR == bfd_target_coff_flavour
3613 #endif
3614 )
3615 value += fixP->fx_where + fixP->fx_frag->fr_address;
3616 #endif
3617 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
3618 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
3619 {
3620 segT fseg = S_GET_SEGMENT (fixP->fx_addsy);
3621
3622 if ((fseg == seg
3623 || (symbol_section_p (fixP->fx_addsy)
3624 && fseg != absolute_section))
3625 && ! S_IS_EXTERNAL (fixP->fx_addsy)
3626 && ! S_IS_WEAK (fixP->fx_addsy)
3627 && S_IS_DEFINED (fixP->fx_addsy)
3628 && ! S_IS_COMMON (fixP->fx_addsy))
3629 {
3630 /* Yes, we add the values in twice. This is because
3631 bfd_perform_relocation subtracts them out again. I think
3632 bfd_perform_relocation is broken, but I don't dare change
3633 it. FIXME. */
3634 value += fixP->fx_where + fixP->fx_frag->fr_address;
3635 }
3636 }
3637 #endif
3638 #if defined (OBJ_COFF) && defined (TE_PE)
3639 /* For some reason, the PE format does not store a section
3640 address offset for a PC relative symbol. */
3641 if (S_GET_SEGMENT (fixP->fx_addsy) != seg)
3642 value += md_pcrel_from (fixP);
3643 #endif
3644 }
3645
3646 /* Fix a few things - the dynamic linker expects certain values here,
3647 and we must not dissappoint it. */
3648 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
3649 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
3650 && fixP->fx_addsy)
3651 switch (fixP->fx_r_type)
3652 {
3653 case BFD_RELOC_386_PLT32:
3654 /* Make the jump instruction point to the address of the operand. At
3655 runtime we merely add the offset to the actual PLT entry. */
3656 value = -4;
3657 break;
3658 case BFD_RELOC_386_GOTPC:
3659
3660 /* This is tough to explain. We end up with this one if we have
3661 * operands that look like "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal
3662 * here is to obtain the absolute address of the GOT, and it is strongly
3663 * preferable from a performance point of view to avoid using a runtime
3664 * relocation for this. The actual sequence of instructions often look
3665 * something like:
3666 *
3667 * call .L66
3668 * .L66:
3669 * popl %ebx
3670 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
3671 *
3672 * The call and pop essentially return the absolute address of
3673 * the label .L66 and store it in %ebx. The linker itself will
3674 * ultimately change the first operand of the addl so that %ebx points to
3675 * the GOT, but to keep things simple, the .o file must have this operand
3676 * set so that it generates not the absolute address of .L66, but the
3677 * absolute address of itself. This allows the linker itself simply
3678 * treat a GOTPC relocation as asking for a pcrel offset to the GOT to be
3679 * added in, and the addend of the relocation is stored in the operand
3680 * field for the instruction itself.
3681 *
3682 * Our job here is to fix the operand so that it would add the correct
3683 * offset so that %ebx would point to itself. The thing that is tricky is
3684 * that .-.L66 will point to the beginning of the instruction, so we need
3685 * to further modify the operand so that it will point to itself.
3686 * There are other cases where you have something like:
3687 *
3688 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
3689 *
3690 * and here no correction would be required. Internally in the assembler
3691 * we treat operands of this form as not being pcrel since the '.' is
3692 * explicitly mentioned, and I wonder whether it would simplify matters
3693 * to do it this way. Who knows. In earlier versions of the PIC patches,
3694 * the pcrel_adjust field was used to store the correction, but since the
3695 * expression is not pcrel, I felt it would be confusing to do it this
3696 * way. */
3697
3698 value -= 1;
3699 break;
3700 case BFD_RELOC_386_GOT32:
3701 value = 0; /* Fully resolved at runtime. No addend. */
3702 break;
3703 case BFD_RELOC_386_GOTOFF:
3704 break;
3705
3706 case BFD_RELOC_VTABLE_INHERIT:
3707 case BFD_RELOC_VTABLE_ENTRY:
3708 fixP->fx_done = 0;
3709 return 1;
3710
3711 default:
3712 break;
3713 }
3714 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
3715 *valp = value;
3716 #endif /* defined (BFD_ASSEMBLER) && !defined (TE_Mach) */
3717 md_number_to_chars (p, value, fixP->fx_size);
3718
3719 return 1;
3720 }
3721 \f
3722 #define MAX_LITTLENUMS 6
3723
3724 /* Turn the string pointed to by litP into a floating point constant
3725 of type TYPE, and emit the appropriate bytes. The number of
3726 LITTLENUMS emitted is stored in *SIZEP. An error message is
3727 returned, or NULL on OK. */
3728
3729 char *
3730 md_atof (type, litP, sizeP)
3731 int type;
3732 char *litP;
3733 int *sizeP;
3734 {
3735 int prec;
3736 LITTLENUM_TYPE words[MAX_LITTLENUMS];
3737 LITTLENUM_TYPE *wordP;
3738 char *t;
3739
3740 switch (type)
3741 {
3742 case 'f':
3743 case 'F':
3744 prec = 2;
3745 break;
3746
3747 case 'd':
3748 case 'D':
3749 prec = 4;
3750 break;
3751
3752 case 'x':
3753 case 'X':
3754 prec = 5;
3755 break;
3756
3757 default:
3758 *sizeP = 0;
3759 return _("Bad call to md_atof ()");
3760 }
3761 t = atof_ieee (input_line_pointer, type, words);
3762 if (t)
3763 input_line_pointer = t;
3764
3765 *sizeP = prec * sizeof (LITTLENUM_TYPE);
3766 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
3767 the bigendian 386. */
3768 for (wordP = words + prec - 1; prec--;)
3769 {
3770 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
3771 litP += sizeof (LITTLENUM_TYPE);
3772 }
3773 return 0;
3774 }
3775 \f
3776 char output_invalid_buf[8];
3777
3778 static char *
3779 output_invalid (c)
3780 int c;
3781 {
3782 if (isprint (c))
3783 sprintf (output_invalid_buf, "'%c'", c);
3784 else
3785 sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
3786 return output_invalid_buf;
3787 }
3788
3789 /* REG_STRING starts *before* REGISTER_PREFIX. */
3790
3791 static const reg_entry *
3792 parse_register (reg_string, end_op)
3793 char *reg_string;
3794 char **end_op;
3795 {
3796 char *s = reg_string;
3797 char *p;
3798 char reg_name_given[MAX_REG_NAME_SIZE + 1];
3799 const reg_entry *r;
3800
3801 /* Skip possible REGISTER_PREFIX and possible whitespace. */
3802 if (*s == REGISTER_PREFIX)
3803 ++s;
3804
3805 if (is_space_char (*s))
3806 ++s;
3807
3808 p = reg_name_given;
3809 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
3810 {
3811 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
3812 return (const reg_entry *) NULL;
3813 s++;
3814 }
3815
3816 /* For naked regs, make sure that we are not dealing with an identifier.
3817 This prevents confusing an identifier like `eax_var' with register
3818 `eax'. */
3819 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
3820 return (const reg_entry *) NULL;
3821
3822 *end_op = s;
3823
3824 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
3825
3826 /* Handle floating point regs, allowing spaces in the (i) part. */
3827 if (r == i386_regtab /* %st is first entry of table */)
3828 {
3829 if (is_space_char (*s))
3830 ++s;
3831 if (*s == '(')
3832 {
3833 ++s;
3834 if (is_space_char (*s))
3835 ++s;
3836 if (*s >= '0' && *s <= '7')
3837 {
3838 r = &i386_float_regtab[*s - '0'];
3839 ++s;
3840 if (is_space_char (*s))
3841 ++s;
3842 if (*s == ')')
3843 {
3844 *end_op = s + 1;
3845 return r;
3846 }
3847 }
3848 /* We have "%st(" then garbage. */
3849 return (const reg_entry *) NULL;
3850 }
3851 }
3852
3853 return r;
3854 }
3855 \f
3856 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
3857 const char *md_shortopts = "kVQ:sq";
3858 #else
3859 const char *md_shortopts = "q";
3860 #endif
3861 struct option md_longopts[] = {
3862 {NULL, no_argument, NULL, 0}
3863 };
3864 size_t md_longopts_size = sizeof (md_longopts);
3865
3866 int
3867 md_parse_option (c, arg)
3868 int c;
3869 char *arg ATTRIBUTE_UNUSED;
3870 {
3871 switch (c)
3872 {
3873 case 'q':
3874 quiet_warnings = 1;
3875 break;
3876
3877 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
3878 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
3879 should be emitted or not. FIXME: Not implemented. */
3880 case 'Q':
3881 break;
3882
3883 /* -V: SVR4 argument to print version ID. */
3884 case 'V':
3885 print_version_id ();
3886 break;
3887
3888 /* -k: Ignore for FreeBSD compatibility. */
3889 case 'k':
3890 break;
3891
3892 case 's':
3893 /* -s: On i386 Solaris, this tells the native assembler to use
3894 .stab instead of .stab.excl. We always use .stab anyhow. */
3895 break;
3896 #endif
3897
3898 default:
3899 return 0;
3900 }
3901 return 1;
3902 }
3903
3904 void
3905 md_show_usage (stream)
3906 FILE *stream;
3907 {
3908 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
3909 fprintf (stream, _("\
3910 -Q ignored\n\
3911 -V print assembler version number\n\
3912 -k ignored\n\
3913 -q quieten some warnings\n\
3914 -s ignored\n"));
3915 #else
3916 fprintf (stream, _("\
3917 -q quieten some warnings\n"));
3918 #endif
3919 }
3920
3921 #ifdef BFD_ASSEMBLER
3922 #if ((defined (OBJ_MAYBE_ELF) && defined (OBJ_MAYBE_COFF)) \
3923 || (defined (OBJ_MAYBE_ELF) && defined (OBJ_MAYBE_AOUT)) \
3924 || (defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)))
3925
3926 /* Pick the target format to use. */
3927
3928 const char *
3929 i386_target_format ()
3930 {
3931 switch (OUTPUT_FLAVOR)
3932 {
3933 #ifdef OBJ_MAYBE_AOUT
3934 case bfd_target_aout_flavour:
3935 return AOUT_TARGET_FORMAT;
3936 #endif
3937 #ifdef OBJ_MAYBE_COFF
3938 case bfd_target_coff_flavour:
3939 return "coff-i386";
3940 #endif
3941 #ifdef OBJ_MAYBE_ELF
3942 case bfd_target_elf_flavour:
3943 return "elf32-i386";
3944 #endif
3945 default:
3946 abort ();
3947 return NULL;
3948 }
3949 }
3950
3951 #endif /* OBJ_MAYBE_ more than one */
3952 #endif /* BFD_ASSEMBLER */
3953 \f
3954 symbolS *
3955 md_undefined_symbol (name)
3956 char *name;
3957 {
3958 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
3959 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
3960 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
3961 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
3962 {
3963 if (!GOT_symbol)
3964 {
3965 if (symbol_find (name))
3966 as_bad (_("GOT already in symbol table"));
3967 GOT_symbol = symbol_new (name, undefined_section,
3968 (valueT) 0, &zero_address_frag);
3969 };
3970 return GOT_symbol;
3971 }
3972 return 0;
3973 }
3974
3975 /* Round up a section size to the appropriate boundary. */
3976
3977 valueT
3978 md_section_align (segment, size)
3979 segT segment ATTRIBUTE_UNUSED;
3980 valueT size;
3981 {
3982 #ifdef BFD_ASSEMBLER
3983 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3984 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
3985 {
3986 /* For a.out, force the section size to be aligned. If we don't do
3987 this, BFD will align it for us, but it will not write out the
3988 final bytes of the section. This may be a bug in BFD, but it is
3989 easier to fix it here since that is how the other a.out targets
3990 work. */
3991 int align;
3992
3993 align = bfd_get_section_alignment (stdoutput, segment);
3994 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
3995 }
3996 #endif
3997 #endif
3998
3999 return size;
4000 }
4001
4002 /* On the i386, PC-relative offsets are relative to the start of the
4003 next instruction. That is, the address of the offset, plus its
4004 size, since the offset is always the last part of the insn. */
4005
4006 long
4007 md_pcrel_from (fixP)
4008 fixS *fixP;
4009 {
4010 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
4011 }
4012
4013 #ifndef I386COFF
4014
4015 static void
4016 s_bss (ignore)
4017 int ignore ATTRIBUTE_UNUSED;
4018 {
4019 register int temp;
4020
4021 temp = get_absolute_expression ();
4022 subseg_set (bss_section, (subsegT) temp);
4023 demand_empty_rest_of_line ();
4024 }
4025
4026 #endif
4027
4028 #ifdef BFD_ASSEMBLER
4029
4030 void
4031 i386_validate_fix (fixp)
4032 fixS *fixp;
4033 {
4034 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
4035 {
4036 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
4037 fixp->fx_subsy = 0;
4038 }
4039 }
4040
4041 arelent *
4042 tc_gen_reloc (section, fixp)
4043 asection *section ATTRIBUTE_UNUSED;
4044 fixS *fixp;
4045 {
4046 arelent *rel;
4047 bfd_reloc_code_real_type code;
4048
4049 switch (fixp->fx_r_type)
4050 {
4051 case BFD_RELOC_386_PLT32:
4052 case BFD_RELOC_386_GOT32:
4053 case BFD_RELOC_386_GOTOFF:
4054 case BFD_RELOC_386_GOTPC:
4055 case BFD_RELOC_RVA:
4056 case BFD_RELOC_VTABLE_ENTRY:
4057 case BFD_RELOC_VTABLE_INHERIT:
4058 code = fixp->fx_r_type;
4059 break;
4060 default:
4061 if (fixp->fx_pcrel)
4062 {
4063 switch (fixp->fx_size)
4064 {
4065 default:
4066 as_bad (_("can not do %d byte pc-relative relocation"),
4067 fixp->fx_size);
4068 code = BFD_RELOC_32_PCREL;
4069 break;
4070 case 1: code = BFD_RELOC_8_PCREL; break;
4071 case 2: code = BFD_RELOC_16_PCREL; break;
4072 case 4: code = BFD_RELOC_32_PCREL; break;
4073 }
4074 }
4075 else
4076 {
4077 switch (fixp->fx_size)
4078 {
4079 default:
4080 as_bad (_("can not do %d byte relocation"), fixp->fx_size);
4081 code = BFD_RELOC_32;
4082 break;
4083 case 1: code = BFD_RELOC_8; break;
4084 case 2: code = BFD_RELOC_16; break;
4085 case 4: code = BFD_RELOC_32; break;
4086 }
4087 }
4088 break;
4089 }
4090
4091 if (code == BFD_RELOC_32
4092 && GOT_symbol
4093 && fixp->fx_addsy == GOT_symbol)
4094 code = BFD_RELOC_386_GOTPC;
4095
4096 rel = (arelent *) xmalloc (sizeof (arelent));
4097 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
4098 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
4099
4100 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
4101 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
4102 vtable entry to be used in the relocation's section offset. */
4103 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
4104 rel->address = fixp->fx_offset;
4105
4106 if (fixp->fx_pcrel)
4107 rel->addend = fixp->fx_addnumber;
4108 else
4109 rel->addend = 0;
4110
4111 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
4112 if (rel->howto == NULL)
4113 {
4114 as_bad_where (fixp->fx_file, fixp->fx_line,
4115 _("cannot represent relocation type %s"),
4116 bfd_get_reloc_code_name (code));
4117 /* Set howto to a garbage value so that we can keep going. */
4118 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
4119 assert (rel->howto != NULL);
4120 }
4121
4122 return rel;
4123 }
4124
4125 #else /* ! BFD_ASSEMBLER */
4126
4127 #if (defined(OBJ_AOUT) | defined(OBJ_BOUT))
4128 void
4129 tc_aout_fix_to_chars (where, fixP, segment_address_in_file)
4130 char *where;
4131 fixS *fixP;
4132 relax_addressT segment_address_in_file;
4133 {
4134 /* In: length of relocation (or of address) in chars: 1, 2 or 4.
4135 Out: GNU LD relocation length code: 0, 1, or 2. */
4136
4137 static const unsigned char nbytes_r_length[] = { 42, 0, 1, 42, 2 };
4138 long r_symbolnum;
4139
4140 know (fixP->fx_addsy != NULL);
4141
4142 md_number_to_chars (where,
4143 (valueT) (fixP->fx_frag->fr_address
4144 + fixP->fx_where - segment_address_in_file),
4145 4);
4146
4147 r_symbolnum = (S_IS_DEFINED (fixP->fx_addsy)
4148 ? S_GET_TYPE (fixP->fx_addsy)
4149 : fixP->fx_addsy->sy_number);
4150
4151 where[6] = (r_symbolnum >> 16) & 0x0ff;
4152 where[5] = (r_symbolnum >> 8) & 0x0ff;
4153 where[4] = r_symbolnum & 0x0ff;
4154 where[7] = ((((!S_IS_DEFINED (fixP->fx_addsy)) << 3) & 0x08)
4155 | ((nbytes_r_length[fixP->fx_size] << 1) & 0x06)
4156 | (((fixP->fx_pcrel << 0) & 0x01) & 0x0f));
4157 }
4158
4159 #endif /* OBJ_AOUT or OBJ_BOUT. */
4160
4161 #if defined (I386COFF)
4162
4163 short
4164 tc_coff_fix2rtype (fixP)
4165 fixS *fixP;
4166 {
4167 if (fixP->fx_r_type == R_IMAGEBASE)
4168 return R_IMAGEBASE;
4169
4170 return (fixP->fx_pcrel ?
4171 (fixP->fx_size == 1 ? R_PCRBYTE :
4172 fixP->fx_size == 2 ? R_PCRWORD :
4173 R_PCRLONG) :
4174 (fixP->fx_size == 1 ? R_RELBYTE :
4175 fixP->fx_size == 2 ? R_RELWORD :
4176 R_DIR32));
4177 }
4178
4179 int
4180 tc_coff_sizemachdep (frag)
4181 fragS *frag;
4182 {
4183 if (frag->fr_next)
4184 return (frag->fr_next->fr_address - frag->fr_address);
4185 else
4186 return 0;
4187 }
4188
4189 #endif /* I386COFF */
4190
4191 #endif /* ! BFD_ASSEMBLER */
4192 \f
4193 /* Parse operands using Intel syntax. This implements a recursive descent
4194 parser based on the BNF grammar published in Appendix B of the MASM 6.1
4195 Programmer's Guide.
4196
4197 FIXME: We do not recognize the full operand grammar defined in the MASM
4198 documentation. In particular, all the structure/union and
4199 high-level macro operands are missing.
4200
4201 Uppercase words are terminals, lower case words are non-terminals.
4202 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
4203 bars '|' denote choices. Most grammar productions are implemented in
4204 functions called 'intel_<production>'.
4205
4206 Initial production is 'expr'.
4207
4208 addOp + | -
4209
4210 alpha [a-zA-Z]
4211
4212 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
4213
4214 constant digits [[ radixOverride ]]
4215
4216 dataType BYTE | WORD | DWORD | QWORD | XWORD
4217
4218 digits decdigit
4219 | digits decdigit
4220 | digits hexdigit
4221
4222 decdigit [0-9]
4223
4224 e05 e05 addOp e06
4225 | e06
4226
4227 e06 e06 mulOp e09
4228 | e09
4229
4230 e09 OFFSET e10
4231 | e09 PTR e10
4232 | e09 : e10
4233 | e10
4234
4235 e10 e10 [ expr ]
4236 | e11
4237
4238 e11 ( expr )
4239 | [ expr ]
4240 | constant
4241 | dataType
4242 | id
4243 | $
4244 | register
4245
4246 => expr SHORT e05
4247 | e05
4248
4249 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
4250 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
4251
4252 hexdigit a | b | c | d | e | f
4253 | A | B | C | D | E | F
4254
4255 id alpha
4256 | id alpha
4257 | id decdigit
4258
4259 mulOp * | / | MOD
4260
4261 quote " | '
4262
4263 register specialRegister
4264 | gpRegister
4265 | byteRegister
4266
4267 segmentRegister CS | DS | ES | FS | GS | SS
4268
4269 specialRegister CR0 | CR2 | CR3
4270 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
4271 | TR3 | TR4 | TR5 | TR6 | TR7
4272
4273 We simplify the grammar in obvious places (e.g., register parsing is
4274 done by calling parse_register) and eliminate immediate left recursion
4275 to implement a recursive-descent parser.
4276
4277 expr SHORT e05
4278 | e05
4279
4280 e05 e06 e05'
4281
4282 e05' addOp e06 e05'
4283 | Empty
4284
4285 e06 e09 e06'
4286
4287 e06' mulOp e09 e06'
4288 | Empty
4289
4290 e09 OFFSET e10 e09'
4291 | e10 e09'
4292
4293 e09' PTR e10 e09'
4294 | : e10 e09'
4295 | Empty
4296
4297 e10 e11 e10'
4298
4299 e10' [ expr ] e10'
4300 | Empty
4301
4302 e11 ( expr )
4303 | [ expr ]
4304 | BYTE
4305 | WORD
4306 | DWORD
4307 | QWORD
4308 | XWORD
4309 | .
4310 | $
4311 | register
4312 | id
4313 | constant */
4314
4315 /* Parsing structure for the intel syntax parser. Used to implement the
4316 semantic actions for the operand grammar. */
4317 struct intel_parser_s
4318 {
4319 char *op_string; /* The string being parsed. */
4320 int got_a_float; /* Whether the operand is a float. */
4321 int op_modifier; /* Operand modifier. */
4322 int is_mem; /* 1 if operand is memory reference. */
4323 const reg_entry *reg; /* Last register reference found. */
4324 char *disp; /* Displacement string being built. */
4325 };
4326
4327 static struct intel_parser_s intel_parser;
4328
4329 /* Token structure for parsing intel syntax. */
4330 struct intel_token
4331 {
4332 int code; /* Token code. */
4333 const reg_entry *reg; /* Register entry for register tokens. */
4334 char *str; /* String representation. */
4335 };
4336
4337 static struct intel_token cur_token, prev_token;
4338
4339
4340 /* Token codes for the intel parser. Since T_SHORT is already used
4341 by COFF, undefine it first to prevent a warning. */
4342 #define T_NIL -1
4343 #define T_CONST 1
4344 #define T_REG 2
4345 #define T_BYTE 3
4346 #define T_WORD 4
4347 #define T_DWORD 5
4348 #define T_QWORD 6
4349 #define T_XWORD 7
4350 #undef T_SHORT
4351 #define T_SHORT 8
4352 #define T_OFFSET 9
4353 #define T_PTR 10
4354 #define T_ID 11
4355
4356 /* Prototypes for intel parser functions. */
4357 static int intel_match_token PARAMS ((int code));
4358 static void intel_get_token PARAMS ((void));
4359 static void intel_putback_token PARAMS ((void));
4360 static int intel_expr PARAMS ((void));
4361 static int intel_e05 PARAMS ((void));
4362 static int intel_e05_1 PARAMS ((void));
4363 static int intel_e06 PARAMS ((void));
4364 static int intel_e06_1 PARAMS ((void));
4365 static int intel_e09 PARAMS ((void));
4366 static int intel_e09_1 PARAMS ((void));
4367 static int intel_e10 PARAMS ((void));
4368 static int intel_e10_1 PARAMS ((void));
4369 static int intel_e11 PARAMS ((void));
4370
4371 static int
4372 i386_intel_operand (operand_string, got_a_float)
4373 char *operand_string;
4374 int got_a_float;
4375 {
4376 int ret;
4377 char *p;
4378
4379 /* Initialize token holders. */
4380 cur_token.code = prev_token.code = T_NIL;
4381 cur_token.reg = prev_token.reg = NULL;
4382 cur_token.str = prev_token.str = NULL;
4383
4384 /* Initialize parser structure. */
4385 p = intel_parser.op_string = (char *)malloc (strlen (operand_string) + 1);
4386 if (p == NULL)
4387 abort ();
4388 strcpy (intel_parser.op_string, operand_string);
4389 intel_parser.got_a_float = got_a_float;
4390 intel_parser.op_modifier = -1;
4391 intel_parser.is_mem = 0;
4392 intel_parser.reg = NULL;
4393 intel_parser.disp = (char *)malloc (strlen (operand_string) + 1);
4394 if (intel_parser.disp == NULL)
4395 abort ();
4396 intel_parser.disp[0] = '\0';
4397
4398 /* Read the first token and start the parser. */
4399 intel_get_token ();
4400 ret = intel_expr ();
4401
4402 if (ret)
4403 {
4404 /* If we found a memory reference, hand it over to i386_displacement
4405 to fill in the rest of the operand fields. */
4406 if (intel_parser.is_mem)
4407 {
4408 if ((i.mem_operands == 1
4409 && (current_templates->start->opcode_modifier & IsString) == 0)
4410 || i.mem_operands == 2)
4411 {
4412 as_bad (_("too many memory references for '%s'"),
4413 current_templates->start->name);
4414 ret = 0;
4415 }
4416 else
4417 {
4418 char *s = intel_parser.disp;
4419 i.mem_operands++;
4420
4421 /* Add the displacement expression. */
4422 if (*s != '\0')
4423 ret = i386_displacement (s, s + strlen (s))
4424 && i386_index_check (s);
4425 }
4426 }
4427
4428 /* Constant and OFFSET expressions are handled by i386_immediate. */
4429 else if (intel_parser.op_modifier == OFFSET_FLAT
4430 || intel_parser.reg == NULL)
4431 ret = i386_immediate (intel_parser.disp);
4432 }
4433
4434 free (p);
4435 free (intel_parser.disp);
4436
4437 return ret;
4438 }
4439
4440 /* expr SHORT e05
4441 | e05 */
4442 static int
4443 intel_expr ()
4444 {
4445 /* expr SHORT e05 */
4446 if (cur_token.code == T_SHORT)
4447 {
4448 intel_parser.op_modifier = SHORT;
4449 intel_match_token (T_SHORT);
4450
4451 return (intel_e05 ());
4452 }
4453
4454 /* expr e05 */
4455 else
4456 return intel_e05 ();
4457 }
4458
4459 /* e05 e06 e05'
4460
4461 e05' addOp e06 e05'
4462 | Empty */
4463 static int
4464 intel_e05 ()
4465 {
4466 return (intel_e06 () && intel_e05_1 ());
4467 }
4468
4469 static int
4470 intel_e05_1 ()
4471 {
4472 /* e05' addOp e06 e05' */
4473 if (cur_token.code == '+' || cur_token.code == '-')
4474 {
4475 strcat (intel_parser.disp, cur_token.str);
4476 intel_match_token (cur_token.code);
4477
4478 return (intel_e06 () && intel_e05_1 ());
4479 }
4480
4481 /* e05' Empty */
4482 else
4483 return 1;
4484 }
4485
4486 /* e06 e09 e06'
4487
4488 e06' mulOp e09 e06'
4489 | Empty */
4490 static int
4491 intel_e06 ()
4492 {
4493 return (intel_e09 () && intel_e06_1 ());
4494 }
4495
4496 static int
4497 intel_e06_1 ()
4498 {
4499 /* e06' mulOp e09 e06' */
4500 if (cur_token.code == '*' || cur_token.code == '/')
4501 {
4502 strcat (intel_parser.disp, cur_token.str);
4503 intel_match_token (cur_token.code);
4504
4505 return (intel_e09 () && intel_e06_1 ());
4506 }
4507
4508 /* e06' Empty */
4509 else
4510 return 1;
4511 }
4512
4513 /* e09 OFFSET e10 e09'
4514 | e10 e09'
4515
4516 e09' PTR e10 e09'
4517 | : e10 e09'
4518 | Empty */
4519 static int
4520 intel_e09 ()
4521 {
4522 /* e09 OFFSET e10 e09' */
4523 if (cur_token.code == T_OFFSET)
4524 {
4525 intel_parser.is_mem = 0;
4526 intel_parser.op_modifier = OFFSET_FLAT;
4527 intel_match_token (T_OFFSET);
4528
4529 return (intel_e10 () && intel_e09_1 ());
4530 }
4531
4532 /* e09 e10 e09' */
4533 else
4534 return (intel_e10 () && intel_e09_1 ());
4535 }
4536
4537 static int
4538 intel_e09_1 ()
4539 {
4540 /* e09' PTR e10 e09' */
4541 if (cur_token.code == T_PTR)
4542 {
4543 if (prev_token.code == T_BYTE)
4544 i.suffix = BYTE_MNEM_SUFFIX;
4545
4546 else if (prev_token.code == T_WORD)
4547 {
4548 if (intel_parser.got_a_float == 2) /* "fi..." */
4549 i.suffix = SHORT_MNEM_SUFFIX;
4550 else
4551 i.suffix = WORD_MNEM_SUFFIX;
4552 }
4553
4554 else if (prev_token.code == T_DWORD)
4555 {
4556 if (intel_parser.got_a_float == 1) /* "f..." */
4557 i.suffix = SHORT_MNEM_SUFFIX;
4558 else
4559 i.suffix = LONG_MNEM_SUFFIX;
4560 }
4561
4562 else if (prev_token.code == T_QWORD)
4563 {
4564 if (intel_parser.got_a_float == 1) /* "f..." */
4565 i.suffix = LONG_MNEM_SUFFIX;
4566 else
4567 {
4568 as_bad (_("operand modifier `%s' supported only for i387 operations\n"),
4569 prev_token.str);
4570 return 0;
4571 }
4572 }
4573
4574 else if (prev_token.code == T_XWORD)
4575 i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
4576
4577 else
4578 {
4579 as_bad (_("Unknown operand modifier `%s'\n"), prev_token.str);
4580 return 0;
4581 }
4582
4583 intel_match_token (T_PTR);
4584
4585 return (intel_e10 () && intel_e09_1 ());
4586 }
4587
4588 /* e09 : e10 e09' */
4589 else if (cur_token.code == ':')
4590 {
4591 intel_parser.is_mem = 1;
4592
4593 return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
4594 }
4595
4596 /* e09' Empty */
4597 else
4598 return 1;
4599 }
4600
4601 /* e10 e11 e10'
4602
4603 e10' [ expr ] e10'
4604 | Empty */
4605 static int
4606 intel_e10 ()
4607 {
4608 return (intel_e11 () && intel_e10_1 ());
4609 }
4610
4611 static int
4612 intel_e10_1 ()
4613 {
4614 /* e10' [ expr ] e10' */
4615 if (cur_token.code == '[')
4616 {
4617 intel_match_token ('[');
4618 intel_parser.is_mem = 1;
4619
4620 /* Add a '+' to the displacement string if necessary. */
4621 if (*intel_parser.disp != '\0')
4622 strcat (intel_parser.disp, "+");
4623
4624 return (intel_expr () && intel_match_token (']') && intel_e10_1 ());
4625 }
4626
4627 /* e10' Empty */
4628 else
4629 return 1;
4630 }
4631
4632 /* e11 ( expr )
4633 | [ expr ]
4634 | BYTE
4635 | WORD
4636 | DWORD
4637 | QWORD
4638 | XWORD
4639 | $
4640 | .
4641 | register
4642 | id
4643 | constant */
4644 static int
4645 intel_e11 ()
4646 {
4647 /* e11 ( expr ) */
4648 if (cur_token.code == '(')
4649 {
4650 intel_match_token ('(');
4651 strcat (intel_parser.disp, "(");
4652
4653 if (intel_expr () && intel_match_token (')'))
4654 {
4655 strcat (intel_parser.disp, ")");
4656 return 1;
4657 }
4658 else
4659 return 0;
4660 }
4661
4662 /* e11 [ expr ] */
4663 else if (cur_token.code == '[')
4664 {
4665 intel_match_token ('[');
4666 intel_parser.is_mem = 1;
4667
4668 /* Operands for jump/call inside brackets denote absolute addresses. */
4669 if (current_templates->start->opcode_modifier & Jump
4670 || current_templates->start->opcode_modifier & JumpDword
4671 || current_templates->start->opcode_modifier & JumpByte
4672 || current_templates->start->opcode_modifier & JumpInterSegment)
4673 i.types[this_operand] |= JumpAbsolute;
4674
4675 /* Add a '+' to the displacement string if necessary. */
4676 if (*intel_parser.disp != '\0')
4677 strcat (intel_parser.disp, "+");
4678
4679 return (intel_expr () && intel_match_token (']'));
4680 }
4681
4682 /* e11 BYTE
4683 | WORD
4684 | DWORD
4685 | QWORD
4686 | XWORD */
4687 else if (cur_token.code == T_BYTE
4688 || cur_token.code == T_WORD
4689 || cur_token.code == T_DWORD
4690 || cur_token.code == T_QWORD
4691 || cur_token.code == T_XWORD)
4692 {
4693 intel_match_token (cur_token.code);
4694
4695 return 1;
4696 }
4697
4698 /* e11 $
4699 | . */
4700 else if (cur_token.code == '$' || cur_token.code == '.')
4701 {
4702 strcat (intel_parser.disp, cur_token.str);
4703 intel_match_token (cur_token.code);
4704 intel_parser.is_mem = 1;
4705
4706 return 1;
4707 }
4708
4709 /* e11 register */
4710 else if (cur_token.code == T_REG)
4711 {
4712 const reg_entry *reg = intel_parser.reg = cur_token.reg;
4713
4714 intel_match_token (T_REG);
4715
4716 /* Check for segment change. */
4717 if (cur_token.code == ':')
4718 {
4719 if (reg->reg_type & (SReg2 | SReg3))
4720 {
4721 switch (reg->reg_num)
4722 {
4723 case 0:
4724 i.seg[i.mem_operands] = &es;
4725 break;
4726 case 1:
4727 i.seg[i.mem_operands] = &cs;
4728 break;
4729 case 2:
4730 i.seg[i.mem_operands] = &ss;
4731 break;
4732 case 3:
4733 i.seg[i.mem_operands] = &ds;
4734 break;
4735 case 4:
4736 i.seg[i.mem_operands] = &fs;
4737 break;
4738 case 5:
4739 i.seg[i.mem_operands] = &gs;
4740 break;
4741 }
4742 }
4743 else
4744 {
4745 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
4746 return 0;
4747 }
4748 }
4749
4750 /* Not a segment register. Check for register scaling. */
4751 else if (cur_token.code == '*')
4752 {
4753 if (!intel_parser.is_mem)
4754 {
4755 as_bad (_("Register scaling only allowed in memory operands."));
4756 return 0;
4757 }
4758
4759 /* What follows must be a valid scale. */
4760 if (intel_match_token ('*')
4761 && strchr ("01248", *cur_token.str))
4762 {
4763 i.index_reg = reg;
4764 i.types[this_operand] |= BaseIndex;
4765
4766 /* Set the scale after setting the register (otherwise,
4767 i386_scale will complain) */
4768 i386_scale (cur_token.str);
4769 intel_match_token (T_CONST);
4770 }
4771 else
4772 {
4773 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
4774 cur_token.str);
4775 return 0;
4776 }
4777 }
4778
4779 /* No scaling. If this is a memory operand, the register is either a
4780 base register (first occurrence) or an index register (second
4781 occurrence). */
4782 else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3)))
4783 {
4784 if (i.base_reg && i.index_reg)
4785 {
4786 as_bad (_("Too many register references in memory operand.\n"));
4787 return 0;
4788 }
4789
4790 if (i.base_reg == NULL)
4791 i.base_reg = reg;
4792 else
4793 i.index_reg = reg;
4794
4795 i.types[this_operand] |= BaseIndex;
4796 }
4797
4798 /* Offset modifier. Add the register to the displacement string to be
4799 parsed as an immediate expression after we're done. */
4800 else if (intel_parser.op_modifier == OFFSET_FLAT)
4801 strcat (intel_parser.disp, reg->reg_name);
4802
4803 /* It's neither base nor index nor offset. */
4804 else
4805 {
4806 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
4807 i.op[this_operand].regs = reg;
4808 i.reg_operands++;
4809 }
4810
4811 /* Since registers are not part of the displacement string (except
4812 when we're parsing offset operands), we may need to remove any
4813 preceding '+' from the displacement string. */
4814 if (*intel_parser.disp != '\0'
4815 && intel_parser.op_modifier != OFFSET_FLAT)
4816 {
4817 char *s = intel_parser.disp;
4818 s += strlen (s) - 1;
4819 if (*s == '+')
4820 *s = '\0';
4821 }
4822
4823 return 1;
4824 }
4825
4826 /* e11 id */
4827 else if (cur_token.code == T_ID)
4828 {
4829 /* Add the identifier to the displacement string. */
4830 strcat (intel_parser.disp, cur_token.str);
4831 intel_match_token (T_ID);
4832
4833 /* The identifier represents a memory reference only if it's not
4834 preceded by an offset modifier. */
4835 if (intel_parser.op_modifier != OFFSET_FLAT
4836 && intel_parser.op_modifier != FLAT)
4837 intel_parser.is_mem = 1;
4838
4839 return 1;
4840 }
4841
4842 /* e11 constant */
4843 else if (cur_token.code == T_CONST
4844 || cur_token.code == '-'
4845 || cur_token.code == '+')
4846 {
4847 char *save_str;
4848
4849 /* Allow constants that start with `+' or `-'. */
4850 if (cur_token.code == '-' || cur_token.code == '+')
4851 {
4852 strcat (intel_parser.disp, cur_token.str);
4853 intel_match_token (cur_token.code);
4854 if (cur_token.code != T_CONST)
4855 {
4856 as_bad (_("Syntax error. Expecting a constant. Got `%s'.\n"),
4857 cur_token.str);
4858 return 0;
4859 }
4860 }
4861
4862 save_str = (char *)malloc (strlen (cur_token.str) + 1);
4863 if (save_str == NULL)
4864 abort ();
4865 strcpy (save_str, cur_token.str);
4866
4867 /* Get the next token to check for register scaling. */
4868 intel_match_token (cur_token.code);
4869
4870 /* Check if this constant is a scaling factor for an index register. */
4871 if (cur_token.code == '*')
4872 {
4873 if (intel_match_token ('*') && cur_token.code == T_REG)
4874 {
4875 if (!intel_parser.is_mem)
4876 {
4877 as_bad (_("Register scaling only allowed in memory operands."));
4878 return 0;
4879 }
4880
4881 /* The constant is followed by `* reg', so it must be
4882 a valid scale. */
4883 if (strchr ("01248", *save_str))
4884 {
4885 i.index_reg = cur_token.reg;
4886 i.types[this_operand] |= BaseIndex;
4887
4888 /* Set the scale after setting the register (otherwise,
4889 i386_scale will complain) */
4890 i386_scale (save_str);
4891 intel_match_token (T_REG);
4892
4893 /* Since registers are not part of the displacement
4894 string, we may need to remove any preceding '+' from
4895 the displacement string. */
4896 if (*intel_parser.disp != '\0')
4897 {
4898 char *s = intel_parser.disp;
4899 s += strlen (s) - 1;
4900 if (*s == '+')
4901 *s = '\0';
4902 }
4903
4904 free (save_str);
4905
4906 return 1;
4907 }
4908 else
4909 return 0;
4910 }
4911
4912 /* The constant was not used for register scaling. Since we have
4913 already consumed the token following `*' we now need to put it
4914 back in the stream. */
4915 else
4916 intel_putback_token ();
4917 }
4918
4919 /* Add the constant to the displacement string. */
4920 strcat (intel_parser.disp, save_str);
4921 free (save_str);
4922
4923 return 1;
4924 }
4925
4926 as_bad (_("Unrecognized token '%s'"), cur_token.str);
4927 return 0;
4928 }
4929
4930 /* Match the given token against cur_token. If they match, read the next
4931 token from the operand string. */
4932 static int
4933 intel_match_token (code)
4934 int code;
4935 {
4936 if (cur_token.code == code)
4937 {
4938 intel_get_token ();
4939 return 1;
4940 }
4941 else
4942 {
4943 as_bad (_("Unexpected token `%s'\n"), cur_token.str);
4944 return 0;
4945 }
4946 }
4947
4948 /* Read a new token from intel_parser.op_string and store it in cur_token. */
4949 static void
4950 intel_get_token ()
4951 {
4952 char *end_op;
4953 const reg_entry *reg;
4954 struct intel_token new_token;
4955
4956 new_token.code = T_NIL;
4957 new_token.reg = NULL;
4958 new_token.str = NULL;
4959
4960 /* Free the memory allocated to the previous token and move
4961 cur_token to prev_token. */
4962 if (prev_token.str)
4963 free (prev_token.str);
4964
4965 prev_token = cur_token;
4966
4967 /* Skip whitespace. */
4968 while (is_space_char (*intel_parser.op_string))
4969 intel_parser.op_string++;
4970
4971 /* Return an empty token if we find nothing else on the line. */
4972 if (*intel_parser.op_string == '\0')
4973 {
4974 cur_token = new_token;
4975 return;
4976 }
4977
4978 /* The new token cannot be larger than the remainder of the operand
4979 string. */
4980 new_token.str = (char *)malloc (strlen (intel_parser.op_string) + 1);
4981 if (new_token.str == NULL)
4982 abort ();
4983 new_token.str[0] = '\0';
4984
4985 if (strchr ("0123456789", *intel_parser.op_string))
4986 {
4987 char *p = new_token.str;
4988 char *q = intel_parser.op_string;
4989 new_token.code = T_CONST;
4990
4991 /* Allow any kind of identifier char to encompass floating point and
4992 hexadecimal numbers. */
4993 while (is_identifier_char (*q))
4994 *p++ = *q++;
4995 *p = '\0';
4996
4997 /* Recognize special symbol names [0-9][bf]. */
4998 if (strlen (intel_parser.op_string) == 2
4999 && (intel_parser.op_string[1] == 'b'
5000 || intel_parser.op_string[1] == 'f'))
5001 new_token.code = T_ID;
5002 }
5003
5004 else if (strchr ("+-/*:[]()", *intel_parser.op_string))
5005 {
5006 new_token.code = *intel_parser.op_string;
5007 new_token.str[0] = *intel_parser.op_string;
5008 new_token.str[1] = '\0';
5009 }
5010
5011 else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
5012 && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
5013 {
5014 new_token.code = T_REG;
5015 new_token.reg = reg;
5016
5017 if (*intel_parser.op_string == REGISTER_PREFIX)
5018 {
5019 new_token.str[0] = REGISTER_PREFIX;
5020 new_token.str[1] = '\0';
5021 }
5022
5023 strcat (new_token.str, reg->reg_name);
5024 }
5025
5026 else if (is_identifier_char (*intel_parser.op_string))
5027 {
5028 char *p = new_token.str;
5029 char *q = intel_parser.op_string;
5030
5031 /* A '.' or '$' followed by an identifier char is an identifier.
5032 Otherwise, it's operator '.' followed by an expression. */
5033 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
5034 {
5035 new_token.code = *q;
5036 new_token.str[0] = *q;
5037 new_token.str[1] = '\0';
5038 }
5039 else
5040 {
5041 while (is_identifier_char (*q) || *q == '@')
5042 *p++ = *q++;
5043 *p = '\0';
5044
5045 if (strcasecmp (new_token.str, "BYTE") == 0)
5046 new_token.code = T_BYTE;
5047
5048 else if (strcasecmp (new_token.str, "WORD") == 0)
5049 new_token.code = T_WORD;
5050
5051 else if (strcasecmp (new_token.str, "DWORD") == 0)
5052 new_token.code = T_DWORD;
5053
5054 else if (strcasecmp (new_token.str, "QWORD") == 0)
5055 new_token.code = T_QWORD;
5056
5057 else if (strcasecmp (new_token.str, "XWORD") == 0)
5058 new_token.code = T_XWORD;
5059
5060 else if (strcasecmp (new_token.str, "PTR") == 0)
5061 new_token.code = T_PTR;
5062
5063 else if (strcasecmp (new_token.str, "SHORT") == 0)
5064 new_token.code = T_SHORT;
5065
5066 else if (strcasecmp (new_token.str, "OFFSET") == 0)
5067 {
5068 new_token.code = T_OFFSET;
5069
5070 /* ??? This is not mentioned in the MASM grammar but gcc
5071 makes use of it with -mintel-syntax. OFFSET may be
5072 followed by FLAT: */
5073 if (strncasecmp (q, " FLAT:", 6) == 0)
5074 strcat (new_token.str, " FLAT:");
5075 }
5076
5077 /* ??? This is not mentioned in the MASM grammar. */
5078 else if (strcasecmp (new_token.str, "FLAT") == 0)
5079 new_token.code = T_OFFSET;
5080
5081 else
5082 new_token.code = T_ID;
5083 }
5084 }
5085
5086 else
5087 as_bad (_("Unrecognized token `%s'\n"), intel_parser.op_string);
5088
5089 intel_parser.op_string += strlen (new_token.str);
5090 cur_token = new_token;
5091 }
5092
5093 /* Put cur_token back into the token stream and make cur_token point to
5094 prev_token. */
5095 static void
5096 intel_putback_token ()
5097 {
5098 intel_parser.op_string -= strlen (cur_token.str);
5099 free (cur_token.str);
5100 cur_token = prev_token;
5101
5102 /* Forget prev_token. */
5103 prev_token.code = T_NIL;
5104 prev_token.reg = NULL;
5105 prev_token.str = NULL;
5106 }
This page took 0.167385 seconds and 4 git commands to generate.