* tc-i386.c (md_assemble): Handle third byte of the opcode as prefix.
[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 x86_64 support by Jan Hubicka (jh@suse.cz)
25 Bugs & suggestions are completely welcome. This is free software.
26 Please help us make it better. */
27
28 #include <ctype.h>
29
30 #include "as.h"
31 #include "subsegs.h"
32 #include "dwarf2dbg.h"
33 #include "opcode/i386.h"
34
35 #ifndef REGISTER_WARNINGS
36 #define REGISTER_WARNINGS 1
37 #endif
38
39 #ifndef INFER_ADDR_PREFIX
40 #define INFER_ADDR_PREFIX 1
41 #endif
42
43 #ifndef SCALE1_WHEN_NO_INDEX
44 /* Specifying a scale factor besides 1 when there is no index is
45 futile. eg. `mov (%ebx,2),%al' does exactly the same as
46 `mov (%ebx),%al'. To slavishly follow what the programmer
47 specified, set SCALE1_WHEN_NO_INDEX to 0. */
48 #define SCALE1_WHEN_NO_INDEX 1
49 #endif
50
51 #define true 1
52 #define false 0
53
54 static unsigned int mode_from_disp_size PARAMS ((unsigned int));
55 static int fits_in_signed_byte PARAMS ((offsetT));
56 static int fits_in_unsigned_byte PARAMS ((offsetT));
57 static int fits_in_unsigned_word PARAMS ((offsetT));
58 static int fits_in_signed_word PARAMS ((offsetT));
59 static int fits_in_unsigned_long PARAMS ((offsetT));
60 static int fits_in_signed_long PARAMS ((offsetT));
61 static int smallest_imm_type PARAMS ((offsetT));
62 static offsetT offset_in_range PARAMS ((offsetT, int));
63 static int add_prefix PARAMS ((unsigned int));
64 static void set_code_flag PARAMS ((int));
65 static void set_16bit_gcc_code_flag PARAMS ((int));
66 static void set_intel_syntax PARAMS ((int));
67 static void set_cpu_arch PARAMS ((int));
68
69 #ifdef BFD_ASSEMBLER
70 static bfd_reloc_code_real_type reloc
71 PARAMS ((int, int, int, bfd_reloc_code_real_type));
72 #endif
73
74 #ifndef DEFAULT_ARCH
75 #define DEFAULT_ARCH "i386"
76 #endif
77 static char *default_arch = DEFAULT_ARCH;
78
79 /* 'md_assemble ()' gathers together information and puts it into a
80 i386_insn. */
81
82 union i386_op
83 {
84 expressionS *disps;
85 expressionS *imms;
86 const reg_entry *regs;
87 };
88
89 struct _i386_insn
90 {
91 /* TM holds the template for the insn were currently assembling. */
92 template tm;
93
94 /* SUFFIX holds the instruction mnemonic suffix if given.
95 (e.g. 'l' for 'movl') */
96 char suffix;
97
98 /* OPERANDS gives the number of given operands. */
99 unsigned int operands;
100
101 /* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number
102 of given register, displacement, memory operands and immediate
103 operands. */
104 unsigned int reg_operands, disp_operands, mem_operands, imm_operands;
105
106 /* TYPES [i] is the type (see above #defines) which tells us how to
107 use OP[i] for the corresponding operand. */
108 unsigned int types[MAX_OPERANDS];
109
110 /* Displacement expression, immediate expression, or register for each
111 operand. */
112 union i386_op op[MAX_OPERANDS];
113
114 /* Flags for operands. */
115 unsigned int flags[MAX_OPERANDS];
116 #define Operand_PCrel 1
117
118 /* Relocation type for operand */
119 #ifdef BFD_ASSEMBLER
120 enum bfd_reloc_code_real disp_reloc[MAX_OPERANDS];
121 #else
122 int disp_reloc[MAX_OPERANDS];
123 #endif
124
125 /* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode
126 the base index byte below. */
127 const reg_entry *base_reg;
128 const reg_entry *index_reg;
129 unsigned int log2_scale_factor;
130
131 /* SEG gives the seg_entries of this insn. They are zero unless
132 explicit segment overrides are given. */
133 const seg_entry *seg[2];
134
135 /* PREFIX holds all the given prefix opcodes (usually null).
136 PREFIXES is the number of prefix opcodes. */
137 unsigned int prefixes;
138 unsigned char prefix[MAX_PREFIXES];
139
140 /* RM and SIB are the modrm byte and the sib byte where the
141 addressing modes of this insn are encoded. */
142
143 modrm_byte rm;
144 rex_byte rex;
145 sib_byte sib;
146 };
147
148 typedef struct _i386_insn i386_insn;
149
150 /* List of chars besides those in app.c:symbol_chars that can start an
151 operand. Used to prevent the scrubber eating vital white-space. */
152 #ifdef LEX_AT
153 const char extra_symbol_chars[] = "*%-(@";
154 #else
155 const char extra_symbol_chars[] = "*%-(";
156 #endif
157
158 /* This array holds the chars that always start a comment. If the
159 pre-processor is disabled, these aren't very useful. */
160 #if defined (TE_I386AIX) || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) && ! defined (TE_LINUX) && !defined(TE_FreeBSD))
161 /* Putting '/' here makes it impossible to use the divide operator.
162 However, we need it for compatibility with SVR4 systems. */
163 const char comment_chars[] = "#/";
164 #define PREFIX_SEPARATOR '\\'
165 #else
166 const char comment_chars[] = "#";
167 #define PREFIX_SEPARATOR '/'
168 #endif
169
170 /* This array holds the chars that only start a comment at the beginning of
171 a line. If the line seems to have the form '# 123 filename'
172 .line and .file directives will appear in the pre-processed output.
173 Note that input_file.c hand checks for '#' at the beginning of the
174 first line of the input file. This is because the compiler outputs
175 #NO_APP at the beginning of its output.
176 Also note that comments started like this one will always work if
177 '/' isn't otherwise defined. */
178 #if defined (TE_I386AIX) || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) && ! defined (TE_LINUX) && !defined(TE_FreeBSD))
179 const char line_comment_chars[] = "";
180 #else
181 const char line_comment_chars[] = "/";
182 #endif
183
184 const char line_separator_chars[] = ";";
185
186 /* Chars that can be used to separate mant from exp in floating point
187 nums. */
188 const char EXP_CHARS[] = "eE";
189
190 /* Chars that mean this number is a floating point constant
191 As in 0f12.456
192 or 0d1.2345e12. */
193 const char FLT_CHARS[] = "fFdDxX";
194
195 /* Tables for lexical analysis. */
196 static char mnemonic_chars[256];
197 static char register_chars[256];
198 static char operand_chars[256];
199 static char identifier_chars[256];
200 static char digit_chars[256];
201
202 /* Lexical macros. */
203 #define is_mnemonic_char(x) (mnemonic_chars[(unsigned char) x])
204 #define is_operand_char(x) (operand_chars[(unsigned char) x])
205 #define is_register_char(x) (register_chars[(unsigned char) x])
206 #define is_space_char(x) ((x) == ' ')
207 #define is_identifier_char(x) (identifier_chars[(unsigned char) x])
208 #define is_digit_char(x) (digit_chars[(unsigned char) x])
209
210 /* All non-digit non-letter charcters that may occur in an operand. */
211 static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
212
213 /* md_assemble() always leaves the strings it's passed unaltered. To
214 effect this we maintain a stack of saved characters that we've smashed
215 with '\0's (indicating end of strings for various sub-fields of the
216 assembler instruction). */
217 static char save_stack[32];
218 static char *save_stack_p;
219 #define END_STRING_AND_SAVE(s) \
220 do { *save_stack_p++ = *(s); *(s) = '\0'; } while (0)
221 #define RESTORE_END_STRING(s) \
222 do { *(s) = *--save_stack_p; } while (0)
223
224 /* The instruction we're assembling. */
225 static i386_insn i;
226
227 /* Possible templates for current insn. */
228 static const templates *current_templates;
229
230 /* Per instruction expressionS buffers: 2 displacements & 2 immediate max. */
231 static expressionS disp_expressions[2], im_expressions[2];
232
233 /* Current operand we are working on. */
234 static int this_operand;
235
236 /* We support four different modes. FLAG_CODE variable is used to distinguish
237 these. */
238
239 enum flag_code {
240 CODE_32BIT,
241 CODE_16BIT,
242 CODE_64BIT };
243
244 static enum flag_code flag_code;
245 static int use_rela_relocations = 0;
246
247 /* The names used to print error messages. */
248 static const char *flag_code_names[] =
249 {
250 "32",
251 "16",
252 "64"
253 };
254
255 /* 1 for intel syntax,
256 0 if att syntax. */
257 static int intel_syntax = 0;
258
259 /* 1 if register prefix % not required. */
260 static int allow_naked_reg = 0;
261
262 /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
263 leave, push, and pop instructions so that gcc has the same stack
264 frame as in 32 bit mode. */
265 static char stackop_size = '\0';
266
267 /* Non-zero to quieten some warnings. */
268 static int quiet_warnings = 0;
269
270 /* CPU name. */
271 static const char *cpu_arch_name = NULL;
272
273 /* CPU feature flags. */
274 static unsigned int cpu_arch_flags = CpuUnknownFlags|CpuNo64;
275
276 /* Interface to relax_segment.
277 There are 2 relax states for 386 jump insns: one for conditional &
278 one for unconditional jumps. This is because these two types of
279 jumps add different sizes to frags when we're figuring out what
280 sort of jump to choose to reach a given label. */
281
282 /* Types. */
283 #define COND_JUMP 1
284 #define UNCOND_JUMP 2
285 /* Sizes. */
286 #define CODE16 1
287 #define SMALL 0
288 #define SMALL16 (SMALL|CODE16)
289 #define BIG 2
290 #define BIG16 (BIG|CODE16)
291
292 #ifndef INLINE
293 #ifdef __GNUC__
294 #define INLINE __inline__
295 #else
296 #define INLINE
297 #endif
298 #endif
299
300 #define ENCODE_RELAX_STATE(type,size) \
301 ((relax_substateT) ((type<<2) | (size)))
302 #define SIZE_FROM_RELAX_STATE(s) \
303 ( (((s) & 0x3) == BIG ? 4 : (((s) & 0x3) == BIG16 ? 2 : 1)) )
304
305 /* This table is used by relax_frag to promote short jumps to long
306 ones where necessary. SMALL (short) jumps may be promoted to BIG
307 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
308 don't allow a short jump in a 32 bit code segment to be promoted to
309 a 16 bit offset jump because it's slower (requires data size
310 prefix), and doesn't work, unless the destination is in the bottom
311 64k of the code segment (The top 16 bits of eip are zeroed). */
312
313 const relax_typeS md_relax_table[] =
314 {
315 /* The fields are:
316 1) most positive reach of this state,
317 2) most negative reach of this state,
318 3) how many bytes this mode will add to the size of the current frag
319 4) which index into the table to try if we can't fit into this one. */
320 {1, 1, 0, 0},
321 {1, 1, 0, 0},
322 {1, 1, 0, 0},
323 {1, 1, 0, 0},
324
325 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
326 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
327 /* dword conditionals adds 4 bytes to frag:
328 1 extra opcode byte, 3 extra displacement bytes. */
329 {0, 0, 4, 0},
330 /* word conditionals add 2 bytes to frag:
331 1 extra opcode byte, 1 extra displacement byte. */
332 {0, 0, 2, 0},
333
334 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
335 {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
336 /* dword jmp adds 3 bytes to frag:
337 0 extra opcode bytes, 3 extra displacement bytes. */
338 {0, 0, 3, 0},
339 /* word jmp adds 1 byte to frag:
340 0 extra opcode bytes, 1 extra displacement byte. */
341 {0, 0, 1, 0}
342
343 };
344
345 static const arch_entry cpu_arch[] = {
346 {"i8086", Cpu086 },
347 {"i186", Cpu086|Cpu186 },
348 {"i286", Cpu086|Cpu186|Cpu286 },
349 {"i386", Cpu086|Cpu186|Cpu286|Cpu386 },
350 {"i486", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486 },
351 {"i586", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
352 {"i686", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
353 {"pentium", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
354 {"pentiumpro",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
355 {"pentium4", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX|CpuSSE|CpuSSE2 },
356 {"k6", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow },
357 {"athlon", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuMMX|Cpu3dnow },
358 {"sledgehammer",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuSledgehammer|CpuMMX|Cpu3dnow|CpuSSE|CpuSSE2 },
359 {NULL, 0 }
360 };
361
362 void
363 i386_align_code (fragP, count)
364 fragS *fragP;
365 int count;
366 {
367 /* Various efficient no-op patterns for aligning code labels.
368 Note: Don't try to assemble the instructions in the comments.
369 0L and 0w are not legal. */
370 static const char f32_1[] =
371 {0x90}; /* nop */
372 static const char f32_2[] =
373 {0x89,0xf6}; /* movl %esi,%esi */
374 static const char f32_3[] =
375 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
376 static const char f32_4[] =
377 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
378 static const char f32_5[] =
379 {0x90, /* nop */
380 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
381 static const char f32_6[] =
382 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
383 static const char f32_7[] =
384 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
385 static const char f32_8[] =
386 {0x90, /* nop */
387 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
388 static const char f32_9[] =
389 {0x89,0xf6, /* movl %esi,%esi */
390 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
391 static const char f32_10[] =
392 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
393 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
394 static const char f32_11[] =
395 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
396 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
397 static const char f32_12[] =
398 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
399 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
400 static const char f32_13[] =
401 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
402 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
403 static const char f32_14[] =
404 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
405 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
406 static const char f32_15[] =
407 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
408 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
409 static const char f16_3[] =
410 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
411 static const char f16_4[] =
412 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
413 static const char f16_5[] =
414 {0x90, /* nop */
415 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
416 static const char f16_6[] =
417 {0x89,0xf6, /* mov %si,%si */
418 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
419 static const char f16_7[] =
420 {0x8d,0x74,0x00, /* lea 0(%si),%si */
421 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
422 static const char f16_8[] =
423 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
424 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
425 static const char *const f32_patt[] = {
426 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
427 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
428 };
429 static const char *const f16_patt[] = {
430 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
431 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
432 };
433
434 /* ??? We can't use these fillers for x86_64, since they often kills the
435 upper halves. Solve later. */
436 if (flag_code == CODE_64BIT)
437 count = 1;
438
439 if (count > 0 && count <= 15)
440 {
441 if (flag_code == CODE_16BIT)
442 {
443 memcpy (fragP->fr_literal + fragP->fr_fix,
444 f16_patt[count - 1], count);
445 if (count > 8)
446 /* Adjust jump offset. */
447 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
448 }
449 else
450 memcpy (fragP->fr_literal + fragP->fr_fix,
451 f32_patt[count - 1], count);
452 fragP->fr_var = count;
453 }
454 }
455
456 static char *output_invalid PARAMS ((int c));
457 static int i386_operand PARAMS ((char *operand_string));
458 static int i386_intel_operand PARAMS ((char *operand_string, int got_a_float));
459 static const reg_entry *parse_register PARAMS ((char *reg_string,
460 char **end_op));
461
462 #ifndef I386COFF
463 static void s_bss PARAMS ((int));
464 #endif
465
466 symbolS *GOT_symbol; /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
467
468 static INLINE unsigned int
469 mode_from_disp_size (t)
470 unsigned int t;
471 {
472 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
473 }
474
475 static INLINE int
476 fits_in_signed_byte (num)
477 offsetT num;
478 {
479 return (num >= -128) && (num <= 127);
480 }
481
482 static INLINE int
483 fits_in_unsigned_byte (num)
484 offsetT num;
485 {
486 return (num & 0xff) == num;
487 }
488
489 static INLINE int
490 fits_in_unsigned_word (num)
491 offsetT num;
492 {
493 return (num & 0xffff) == num;
494 }
495
496 static INLINE int
497 fits_in_signed_word (num)
498 offsetT num;
499 {
500 return (-32768 <= num) && (num <= 32767);
501 }
502 static INLINE int
503 fits_in_signed_long (num)
504 offsetT num ATTRIBUTE_UNUSED;
505 {
506 #ifndef BFD64
507 return 1;
508 #else
509 return (!(((offsetT) -1 << 31) & num)
510 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
511 #endif
512 } /* fits_in_signed_long() */
513 static INLINE int
514 fits_in_unsigned_long (num)
515 offsetT num ATTRIBUTE_UNUSED;
516 {
517 #ifndef BFD64
518 return 1;
519 #else
520 return (num & (((offsetT) 2 << 31) - 1)) == num;
521 #endif
522 } /* fits_in_unsigned_long() */
523
524 static int
525 smallest_imm_type (num)
526 offsetT num;
527 {
528 if (cpu_arch_flags != (Cpu086 | Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64)
529 && !(cpu_arch_flags & (CpuUnknown)))
530 {
531 /* This code is disabled on the 486 because all the Imm1 forms
532 in the opcode table are slower on the i486. They're the
533 versions with the implicitly specified single-position
534 displacement, which has another syntax if you really want to
535 use that form. */
536 if (num == 1)
537 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
538 }
539 return (fits_in_signed_byte (num)
540 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
541 : fits_in_unsigned_byte (num)
542 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
543 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
544 ? (Imm16 | Imm32 | Imm32S | Imm64)
545 : fits_in_signed_long (num)
546 ? (Imm32 | Imm32S | Imm64)
547 : fits_in_unsigned_long (num)
548 ? (Imm32 | Imm64)
549 : Imm64);
550 }
551
552 static offsetT
553 offset_in_range (val, size)
554 offsetT val;
555 int size;
556 {
557 addressT mask;
558
559 switch (size)
560 {
561 case 1: mask = ((addressT) 1 << 8) - 1; break;
562 case 2: mask = ((addressT) 1 << 16) - 1; break;
563 case 4: mask = ((addressT) 2 << 31) - 1; break;
564 #ifdef BFD64
565 case 8: mask = ((addressT) 2 << 63) - 1; break;
566 #endif
567 default: abort ();
568 }
569
570 /* If BFD64, sign extend val. */
571 if (!use_rela_relocations)
572 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
573 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
574
575 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
576 {
577 char buf1[40], buf2[40];
578
579 sprint_value (buf1, val);
580 sprint_value (buf2, val & mask);
581 as_warn (_("%s shortened to %s"), buf1, buf2);
582 }
583 return val & mask;
584 }
585
586 /* Returns 0 if attempting to add a prefix where one from the same
587 class already exists, 1 if non rep/repne added, 2 if rep/repne
588 added. */
589 static int
590 add_prefix (prefix)
591 unsigned int prefix;
592 {
593 int ret = 1;
594 int q;
595
596 if (prefix >= 0x40 && prefix < 0x50 && flag_code == CODE_64BIT)
597 q = REX_PREFIX;
598 else
599 switch (prefix)
600 {
601 default:
602 abort ();
603
604 case CS_PREFIX_OPCODE:
605 case DS_PREFIX_OPCODE:
606 case ES_PREFIX_OPCODE:
607 case FS_PREFIX_OPCODE:
608 case GS_PREFIX_OPCODE:
609 case SS_PREFIX_OPCODE:
610 q = SEG_PREFIX;
611 break;
612
613 case REPNE_PREFIX_OPCODE:
614 case REPE_PREFIX_OPCODE:
615 ret = 2;
616 /* fall thru */
617 case LOCK_PREFIX_OPCODE:
618 q = LOCKREP_PREFIX;
619 break;
620
621 case FWAIT_OPCODE:
622 q = WAIT_PREFIX;
623 break;
624
625 case ADDR_PREFIX_OPCODE:
626 q = ADDR_PREFIX;
627 break;
628
629 case DATA_PREFIX_OPCODE:
630 q = DATA_PREFIX;
631 break;
632 }
633
634 if (i.prefix[q])
635 {
636 as_bad (_("same type of prefix used twice"));
637 return 0;
638 }
639
640 i.prefixes += 1;
641 i.prefix[q] = prefix;
642 return ret;
643 }
644
645 static void
646 set_code_flag (value)
647 int value;
648 {
649 flag_code = value;
650 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
651 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
652 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
653 {
654 as_bad (_("64bit mode not supported on this CPU."));
655 }
656 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
657 {
658 as_bad (_("32bit mode not supported on this CPU."));
659 }
660 stackop_size = '\0';
661 }
662
663 static void
664 set_16bit_gcc_code_flag (new_code_flag)
665 int new_code_flag;
666 {
667 flag_code = new_code_flag;
668 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
669 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
670 stackop_size = 'l';
671 }
672
673 static void
674 set_intel_syntax (syntax_flag)
675 int syntax_flag;
676 {
677 /* Find out if register prefixing is specified. */
678 int ask_naked_reg = 0;
679
680 SKIP_WHITESPACE ();
681 if (! is_end_of_line[(unsigned char) *input_line_pointer])
682 {
683 char *string = input_line_pointer;
684 int e = get_symbol_end ();
685
686 if (strcmp (string, "prefix") == 0)
687 ask_naked_reg = 1;
688 else if (strcmp (string, "noprefix") == 0)
689 ask_naked_reg = -1;
690 else
691 as_bad (_("bad argument to syntax directive."));
692 *input_line_pointer = e;
693 }
694 demand_empty_rest_of_line ();
695
696 intel_syntax = syntax_flag;
697
698 if (ask_naked_reg == 0)
699 {
700 #ifdef BFD_ASSEMBLER
701 allow_naked_reg = (intel_syntax
702 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
703 #else
704 /* Conservative default. */
705 allow_naked_reg = 0;
706 #endif
707 }
708 else
709 allow_naked_reg = (ask_naked_reg < 0);
710 }
711
712 static void
713 set_cpu_arch (dummy)
714 int dummy ATTRIBUTE_UNUSED;
715 {
716 SKIP_WHITESPACE ();
717
718 if (! is_end_of_line[(unsigned char) *input_line_pointer])
719 {
720 char *string = input_line_pointer;
721 int e = get_symbol_end ();
722 int i;
723
724 for (i = 0; cpu_arch[i].name; i++)
725 {
726 if (strcmp (string, cpu_arch[i].name) == 0)
727 {
728 cpu_arch_name = cpu_arch[i].name;
729 cpu_arch_flags = cpu_arch[i].flags | (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
730 break;
731 }
732 }
733 if (!cpu_arch[i].name)
734 as_bad (_("no such architecture: `%s'"), string);
735
736 *input_line_pointer = e;
737 }
738 else
739 as_bad (_("missing cpu architecture"));
740
741 demand_empty_rest_of_line ();
742 }
743
744 const pseudo_typeS md_pseudo_table[] =
745 {
746 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
747 {"align", s_align_bytes, 0},
748 #else
749 {"align", s_align_ptwo, 0},
750 #endif
751 {"arch", set_cpu_arch, 0},
752 #ifndef I386COFF
753 {"bss", s_bss, 0},
754 #endif
755 {"ffloat", float_cons, 'f'},
756 {"dfloat", float_cons, 'd'},
757 {"tfloat", float_cons, 'x'},
758 {"value", cons, 2},
759 {"noopt", s_ignore, 0},
760 {"optim", s_ignore, 0},
761 {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
762 {"code16", set_code_flag, CODE_16BIT},
763 {"code32", set_code_flag, CODE_32BIT},
764 {"code64", set_code_flag, CODE_64BIT},
765 {"intel_syntax", set_intel_syntax, 1},
766 {"att_syntax", set_intel_syntax, 0},
767 {"file", dwarf2_directive_file, 0},
768 {"loc", dwarf2_directive_loc, 0},
769 {0, 0, 0}
770 };
771
772 /* For interface with expression (). */
773 extern char *input_line_pointer;
774
775 /* Hash table for instruction mnemonic lookup. */
776 static struct hash_control *op_hash;
777
778 /* Hash table for register lookup. */
779 static struct hash_control *reg_hash;
780 \f
781 void
782 md_begin ()
783 {
784 const char *hash_err;
785
786 /* Initialize op_hash hash table. */
787 op_hash = hash_new ();
788
789 {
790 register const template *optab;
791 register templates *core_optab;
792
793 /* Setup for loop. */
794 optab = i386_optab;
795 core_optab = (templates *) xmalloc (sizeof (templates));
796 core_optab->start = optab;
797
798 while (1)
799 {
800 ++optab;
801 if (optab->name == NULL
802 || strcmp (optab->name, (optab - 1)->name) != 0)
803 {
804 /* different name --> ship out current template list;
805 add to hash table; & begin anew. */
806 core_optab->end = optab;
807 hash_err = hash_insert (op_hash,
808 (optab - 1)->name,
809 (PTR) core_optab);
810 if (hash_err)
811 {
812 as_fatal (_("Internal Error: Can't hash %s: %s"),
813 (optab - 1)->name,
814 hash_err);
815 }
816 if (optab->name == NULL)
817 break;
818 core_optab = (templates *) xmalloc (sizeof (templates));
819 core_optab->start = optab;
820 }
821 }
822 }
823
824 /* Initialize reg_hash hash table. */
825 reg_hash = hash_new ();
826 {
827 register const reg_entry *regtab;
828
829 for (regtab = i386_regtab;
830 regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
831 regtab++)
832 {
833 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
834 if (hash_err)
835 as_fatal (_("Internal Error: Can't hash %s: %s"),
836 regtab->reg_name,
837 hash_err);
838 }
839 }
840
841 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
842 {
843 register int c;
844 register char *p;
845
846 for (c = 0; c < 256; c++)
847 {
848 if (isdigit (c))
849 {
850 digit_chars[c] = c;
851 mnemonic_chars[c] = c;
852 register_chars[c] = c;
853 operand_chars[c] = c;
854 }
855 else if (islower (c))
856 {
857 mnemonic_chars[c] = c;
858 register_chars[c] = c;
859 operand_chars[c] = c;
860 }
861 else if (isupper (c))
862 {
863 mnemonic_chars[c] = tolower (c);
864 register_chars[c] = mnemonic_chars[c];
865 operand_chars[c] = c;
866 }
867
868 if (isalpha (c) || isdigit (c))
869 identifier_chars[c] = c;
870 else if (c >= 128)
871 {
872 identifier_chars[c] = c;
873 operand_chars[c] = c;
874 }
875 }
876
877 #ifdef LEX_AT
878 identifier_chars['@'] = '@';
879 #endif
880 digit_chars['-'] = '-';
881 identifier_chars['_'] = '_';
882 identifier_chars['.'] = '.';
883
884 for (p = operand_special_chars; *p != '\0'; p++)
885 operand_chars[(unsigned char) *p] = *p;
886 }
887
888 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
889 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
890 {
891 record_alignment (text_section, 2);
892 record_alignment (data_section, 2);
893 record_alignment (bss_section, 2);
894 }
895 #endif
896 }
897
898 void
899 i386_print_statistics (file)
900 FILE *file;
901 {
902 hash_print_statistics (file, "i386 opcode", op_hash);
903 hash_print_statistics (file, "i386 register", reg_hash);
904 }
905 \f
906 #ifdef DEBUG386
907
908 /* Debugging routines for md_assemble. */
909 static void pi PARAMS ((char *, i386_insn *));
910 static void pte PARAMS ((template *));
911 static void pt PARAMS ((unsigned int));
912 static void pe PARAMS ((expressionS *));
913 static void ps PARAMS ((symbolS *));
914
915 static void
916 pi (line, x)
917 char *line;
918 i386_insn *x;
919 {
920 unsigned int i;
921
922 fprintf (stdout, "%s: template ", line);
923 pte (&x->tm);
924 fprintf (stdout, " address: base %s index %s scale %x\n",
925 x->base_reg ? x->base_reg->reg_name : "none",
926 x->index_reg ? x->index_reg->reg_name : "none",
927 x->log2_scale_factor);
928 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
929 x->rm.mode, x->rm.reg, x->rm.regmem);
930 fprintf (stdout, " sib: base %x index %x scale %x\n",
931 x->sib.base, x->sib.index, x->sib.scale);
932 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
933 x->rex.mode64, x->rex.extX, x->rex.extY, x->rex.extZ);
934 for (i = 0; i < x->operands; i++)
935 {
936 fprintf (stdout, " #%d: ", i + 1);
937 pt (x->types[i]);
938 fprintf (stdout, "\n");
939 if (x->types[i]
940 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
941 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
942 if (x->types[i] & Imm)
943 pe (x->op[i].imms);
944 if (x->types[i] & Disp)
945 pe (x->op[i].disps);
946 }
947 }
948
949 static void
950 pte (t)
951 template *t;
952 {
953 unsigned int i;
954 fprintf (stdout, " %d operands ", t->operands);
955 fprintf (stdout, "opcode %x ", t->base_opcode);
956 if (t->extension_opcode != None)
957 fprintf (stdout, "ext %x ", t->extension_opcode);
958 if (t->opcode_modifier & D)
959 fprintf (stdout, "D");
960 if (t->opcode_modifier & W)
961 fprintf (stdout, "W");
962 fprintf (stdout, "\n");
963 for (i = 0; i < t->operands; i++)
964 {
965 fprintf (stdout, " #%d type ", i + 1);
966 pt (t->operand_types[i]);
967 fprintf (stdout, "\n");
968 }
969 }
970
971 static void
972 pe (e)
973 expressionS *e;
974 {
975 fprintf (stdout, " operation %d\n", e->X_op);
976 fprintf (stdout, " add_number %ld (%lx)\n",
977 (long) e->X_add_number, (long) e->X_add_number);
978 if (e->X_add_symbol)
979 {
980 fprintf (stdout, " add_symbol ");
981 ps (e->X_add_symbol);
982 fprintf (stdout, "\n");
983 }
984 if (e->X_op_symbol)
985 {
986 fprintf (stdout, " op_symbol ");
987 ps (e->X_op_symbol);
988 fprintf (stdout, "\n");
989 }
990 }
991
992 static void
993 ps (s)
994 symbolS *s;
995 {
996 fprintf (stdout, "%s type %s%s",
997 S_GET_NAME (s),
998 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
999 segment_name (S_GET_SEGMENT (s)));
1000 }
1001
1002 struct type_name
1003 {
1004 unsigned int mask;
1005 char *tname;
1006 }
1007
1008 type_names[] =
1009 {
1010 { Reg8, "r8" },
1011 { Reg16, "r16" },
1012 { Reg32, "r32" },
1013 { Reg64, "r64" },
1014 { Imm8, "i8" },
1015 { Imm8S, "i8s" },
1016 { Imm16, "i16" },
1017 { Imm32, "i32" },
1018 { Imm32S, "i32s" },
1019 { Imm64, "i64" },
1020 { Imm1, "i1" },
1021 { BaseIndex, "BaseIndex" },
1022 { Disp8, "d8" },
1023 { Disp16, "d16" },
1024 { Disp32, "d32" },
1025 { Disp32S, "d32s" },
1026 { Disp64, "d64" },
1027 { InOutPortReg, "InOutPortReg" },
1028 { ShiftCount, "ShiftCount" },
1029 { Control, "control reg" },
1030 { Test, "test reg" },
1031 { Debug, "debug reg" },
1032 { FloatReg, "FReg" },
1033 { FloatAcc, "FAcc" },
1034 { SReg2, "SReg2" },
1035 { SReg3, "SReg3" },
1036 { Acc, "Acc" },
1037 { JumpAbsolute, "Jump Absolute" },
1038 { RegMMX, "rMMX" },
1039 { RegXMM, "rXMM" },
1040 { EsSeg, "es" },
1041 { 0, "" }
1042 };
1043
1044 static void
1045 pt (t)
1046 unsigned int t;
1047 {
1048 register struct type_name *ty;
1049
1050 for (ty = type_names; ty->mask; ty++)
1051 if (t & ty->mask)
1052 fprintf (stdout, "%s, ", ty->tname);
1053 fflush (stdout);
1054 }
1055
1056 #endif /* DEBUG386 */
1057 \f
1058 int
1059 tc_i386_force_relocation (fixp)
1060 struct fix *fixp;
1061 {
1062 #ifdef BFD_ASSEMBLER
1063 if (fixp->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1064 || fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1065 return 1;
1066 return 0;
1067 #else
1068 /* For COFF. */
1069 return fixp->fx_r_type == 7;
1070 #endif
1071 }
1072
1073 #ifdef BFD_ASSEMBLER
1074
1075 static bfd_reloc_code_real_type
1076 reloc (size, pcrel, sign, other)
1077 int size;
1078 int pcrel;
1079 int sign;
1080 bfd_reloc_code_real_type other;
1081 {
1082 if (other != NO_RELOC)
1083 return other;
1084
1085 if (pcrel)
1086 {
1087 if (!sign)
1088 as_bad(_("There are no unsigned pc-relative relocations"));
1089 switch (size)
1090 {
1091 case 1: return BFD_RELOC_8_PCREL;
1092 case 2: return BFD_RELOC_16_PCREL;
1093 case 4: return BFD_RELOC_32_PCREL;
1094 }
1095 as_bad (_("can not do %d byte pc-relative relocation"), size);
1096 }
1097 else
1098 {
1099 if (sign)
1100 switch (size)
1101 {
1102 case 4: return BFD_RELOC_X86_64_32S;
1103 }
1104 else
1105 switch (size)
1106 {
1107 case 1: return BFD_RELOC_8;
1108 case 2: return BFD_RELOC_16;
1109 case 4: return BFD_RELOC_32;
1110 case 8: return BFD_RELOC_64;
1111 }
1112 as_bad (_("can not do %s %d byte relocation"),
1113 sign ? "signed" : "unsigned", size);
1114 }
1115
1116 abort();
1117 return BFD_RELOC_NONE;
1118 }
1119
1120 /* Here we decide which fixups can be adjusted to make them relative to
1121 the beginning of the section instead of the symbol. Basically we need
1122 to make sure that the dynamic relocations are done correctly, so in
1123 some cases we force the original symbol to be used. */
1124
1125 int
1126 tc_i386_fix_adjustable (fixP)
1127 fixS *fixP;
1128 {
1129 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1130 /* Prevent all adjustments to global symbols, or else dynamic
1131 linking will not work correctly. */
1132 if (S_IS_EXTERNAL (fixP->fx_addsy)
1133 || S_IS_WEAK (fixP->fx_addsy))
1134 return 0;
1135 #endif
1136 /* adjust_reloc_syms doesn't know about the GOT. */
1137 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1138 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1139 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1140 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1141 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1142 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1143 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1144 return 0;
1145 return 1;
1146 }
1147 #else
1148 #define reloc(SIZE,PCREL,SIGN,OTHER) 0
1149 #define BFD_RELOC_16 0
1150 #define BFD_RELOC_32 0
1151 #define BFD_RELOC_16_PCREL 0
1152 #define BFD_RELOC_32_PCREL 0
1153 #define BFD_RELOC_386_PLT32 0
1154 #define BFD_RELOC_386_GOT32 0
1155 #define BFD_RELOC_386_GOTOFF 0
1156 #define BFD_RELOC_X86_64_PLT32 0
1157 #define BFD_RELOC_X86_64_GOT32 0
1158 #define BFD_RELOC_X86_64_GOTPCREL 0
1159 #endif
1160
1161 static int intel_float_operand PARAMS ((char *mnemonic));
1162
1163 static int
1164 intel_float_operand (mnemonic)
1165 char *mnemonic;
1166 {
1167 if (mnemonic[0] == 'f' && mnemonic[1] == 'i')
1168 return 2;
1169
1170 if (mnemonic[0] == 'f')
1171 return 1;
1172
1173 return 0;
1174 }
1175
1176 /* This is the guts of the machine-dependent assembler. LINE points to a
1177 machine dependent instruction. This function is supposed to emit
1178 the frags/bytes it assembles to. */
1179
1180 void
1181 md_assemble (line)
1182 char *line;
1183 {
1184 /* Points to template once we've found it. */
1185 const template *t;
1186
1187 /* Count the size of the instruction generated. */
1188 int insn_size = 0;
1189
1190 int j;
1191
1192 char mnemonic[MAX_MNEM_SIZE];
1193
1194 /* Initialize globals. */
1195 memset (&i, '\0', sizeof (i));
1196 for (j = 0; j < MAX_OPERANDS; j++)
1197 i.disp_reloc[j] = NO_RELOC;
1198 memset (disp_expressions, '\0', sizeof (disp_expressions));
1199 memset (im_expressions, '\0', sizeof (im_expressions));
1200 save_stack_p = save_stack;
1201
1202 /* First parse an instruction mnemonic & call i386_operand for the operands.
1203 We assume that the scrubber has arranged it so that line[0] is the valid
1204 start of a (possibly prefixed) mnemonic. */
1205 {
1206 char *l = line;
1207 char *token_start = l;
1208 char *mnem_p;
1209
1210 /* Non-zero if we found a prefix only acceptable with string insns. */
1211 const char *expecting_string_instruction = NULL;
1212
1213 while (1)
1214 {
1215 mnem_p = mnemonic;
1216 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1217 {
1218 mnem_p++;
1219 if (mnem_p >= mnemonic + sizeof (mnemonic))
1220 {
1221 as_bad (_("no such instruction: `%s'"), token_start);
1222 return;
1223 }
1224 l++;
1225 }
1226 if (!is_space_char (*l)
1227 && *l != END_OF_INSN
1228 && *l != PREFIX_SEPARATOR)
1229 {
1230 as_bad (_("invalid character %s in mnemonic"),
1231 output_invalid (*l));
1232 return;
1233 }
1234 if (token_start == l)
1235 {
1236 if (*l == PREFIX_SEPARATOR)
1237 as_bad (_("expecting prefix; got nothing"));
1238 else
1239 as_bad (_("expecting mnemonic; got nothing"));
1240 return;
1241 }
1242
1243 /* Look up instruction (or prefix) via hash table. */
1244 current_templates = hash_find (op_hash, mnemonic);
1245
1246 if (*l != END_OF_INSN
1247 && (! is_space_char (*l) || l[1] != END_OF_INSN)
1248 && current_templates
1249 && (current_templates->start->opcode_modifier & IsPrefix))
1250 {
1251 /* If we are in 16-bit mode, do not allow addr16 or data16.
1252 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1253 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1254 && (((current_templates->start->opcode_modifier & Size32) != 0)
1255 ^ (flag_code == CODE_16BIT)))
1256 {
1257 as_bad (_("redundant %s prefix"),
1258 current_templates->start->name);
1259 return;
1260 }
1261 /* Add prefix, checking for repeated prefixes. */
1262 switch (add_prefix (current_templates->start->base_opcode))
1263 {
1264 case 0:
1265 return;
1266 case 2:
1267 expecting_string_instruction = current_templates->start->name;
1268 break;
1269 }
1270 /* Skip past PREFIX_SEPARATOR and reset token_start. */
1271 token_start = ++l;
1272 }
1273 else
1274 break;
1275 }
1276
1277 if (!current_templates)
1278 {
1279 /* See if we can get a match by trimming off a suffix. */
1280 switch (mnem_p[-1])
1281 {
1282 case WORD_MNEM_SUFFIX:
1283 case BYTE_MNEM_SUFFIX:
1284 case QWORD_MNEM_SUFFIX:
1285 i.suffix = mnem_p[-1];
1286 mnem_p[-1] = '\0';
1287 current_templates = hash_find (op_hash, mnemonic);
1288 break;
1289 case SHORT_MNEM_SUFFIX:
1290 case LONG_MNEM_SUFFIX:
1291 if (!intel_syntax)
1292 {
1293 i.suffix = mnem_p[-1];
1294 mnem_p[-1] = '\0';
1295 current_templates = hash_find (op_hash, mnemonic);
1296 }
1297 break;
1298
1299 /* Intel Syntax. */
1300 case 'd':
1301 if (intel_syntax)
1302 {
1303 if (intel_float_operand (mnemonic))
1304 i.suffix = SHORT_MNEM_SUFFIX;
1305 else
1306 i.suffix = LONG_MNEM_SUFFIX;
1307 mnem_p[-1] = '\0';
1308 current_templates = hash_find (op_hash, mnemonic);
1309 }
1310 break;
1311 }
1312 if (!current_templates)
1313 {
1314 as_bad (_("no such instruction: `%s'"), token_start);
1315 return;
1316 }
1317 }
1318
1319 /* Check if instruction is supported on specified architecture. */
1320 if (cpu_arch_flags != 0)
1321 {
1322 if ((current_templates->start->cpu_flags & ~(Cpu64 | CpuNo64))
1323 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64)))
1324 {
1325 as_warn (_("`%s' is not supported on `%s'"),
1326 current_templates->start->name, cpu_arch_name);
1327 }
1328 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
1329 {
1330 as_warn (_("use .code16 to ensure correct addressing mode"));
1331 }
1332 }
1333
1334 /* Check for rep/repne without a string instruction. */
1335 if (expecting_string_instruction
1336 && !(current_templates->start->opcode_modifier & IsString))
1337 {
1338 as_bad (_("expecting string instruction after `%s'"),
1339 expecting_string_instruction);
1340 return;
1341 }
1342
1343 /* There may be operands to parse. */
1344 if (*l != END_OF_INSN)
1345 {
1346 /* 1 if operand is pending after ','. */
1347 unsigned int expecting_operand = 0;
1348
1349 /* Non-zero if operand parens not balanced. */
1350 unsigned int paren_not_balanced;
1351
1352 do
1353 {
1354 /* Skip optional white space before operand. */
1355 if (is_space_char (*l))
1356 ++l;
1357 if (!is_operand_char (*l) && *l != END_OF_INSN)
1358 {
1359 as_bad (_("invalid character %s before operand %d"),
1360 output_invalid (*l),
1361 i.operands + 1);
1362 return;
1363 }
1364 token_start = l; /* after white space */
1365 paren_not_balanced = 0;
1366 while (paren_not_balanced || *l != ',')
1367 {
1368 if (*l == END_OF_INSN)
1369 {
1370 if (paren_not_balanced)
1371 {
1372 if (!intel_syntax)
1373 as_bad (_("unbalanced parenthesis in operand %d."),
1374 i.operands + 1);
1375 else
1376 as_bad (_("unbalanced brackets in operand %d."),
1377 i.operands + 1);
1378 return;
1379 }
1380 else
1381 break; /* we are done */
1382 }
1383 else if (!is_operand_char (*l) && !is_space_char (*l))
1384 {
1385 as_bad (_("invalid character %s in operand %d"),
1386 output_invalid (*l),
1387 i.operands + 1);
1388 return;
1389 }
1390 if (!intel_syntax)
1391 {
1392 if (*l == '(')
1393 ++paren_not_balanced;
1394 if (*l == ')')
1395 --paren_not_balanced;
1396 }
1397 else
1398 {
1399 if (*l == '[')
1400 ++paren_not_balanced;
1401 if (*l == ']')
1402 --paren_not_balanced;
1403 }
1404 l++;
1405 }
1406 if (l != token_start)
1407 { /* Yes, we've read in another operand. */
1408 unsigned int operand_ok;
1409 this_operand = i.operands++;
1410 if (i.operands > MAX_OPERANDS)
1411 {
1412 as_bad (_("spurious operands; (%d operands/instruction max)"),
1413 MAX_OPERANDS);
1414 return;
1415 }
1416 /* Now parse operand adding info to 'i' as we go along. */
1417 END_STRING_AND_SAVE (l);
1418
1419 if (intel_syntax)
1420 operand_ok =
1421 i386_intel_operand (token_start,
1422 intel_float_operand (mnemonic));
1423 else
1424 operand_ok = i386_operand (token_start);
1425
1426 RESTORE_END_STRING (l);
1427 if (!operand_ok)
1428 return;
1429 }
1430 else
1431 {
1432 if (expecting_operand)
1433 {
1434 expecting_operand_after_comma:
1435 as_bad (_("expecting operand after ','; got nothing"));
1436 return;
1437 }
1438 if (*l == ',')
1439 {
1440 as_bad (_("expecting operand before ','; got nothing"));
1441 return;
1442 }
1443 }
1444
1445 /* Now *l must be either ',' or END_OF_INSN. */
1446 if (*l == ',')
1447 {
1448 if (*++l == END_OF_INSN)
1449 {
1450 /* Just skip it, if it's \n complain. */
1451 goto expecting_operand_after_comma;
1452 }
1453 expecting_operand = 1;
1454 }
1455 }
1456 while (*l != END_OF_INSN);
1457 }
1458 }
1459
1460 /* Now we've parsed the mnemonic into a set of templates, and have the
1461 operands at hand.
1462
1463 Next, we find a template that matches the given insn,
1464 making sure the overlap of the given operands types is consistent
1465 with the template operand types. */
1466
1467 #define MATCH(overlap, given, template) \
1468 ((overlap & ~JumpAbsolute) \
1469 && ((given) & (BaseIndex|JumpAbsolute)) == ((overlap) & (BaseIndex|JumpAbsolute)))
1470
1471 /* If given types r0 and r1 are registers they must be of the same type
1472 unless the expected operand type register overlap is null.
1473 Note that Acc in a template matches every size of reg. */
1474 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
1475 ( ((g0) & Reg) == 0 || ((g1) & Reg) == 0 || \
1476 ((g0) & Reg) == ((g1) & Reg) || \
1477 ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
1478
1479 {
1480 register unsigned int overlap0, overlap1;
1481 unsigned int overlap2;
1482 unsigned int found_reverse_match;
1483 int suffix_check;
1484
1485 /* All intel opcodes have reversed operands except for "bound" and
1486 "enter". We also don't reverse intersegment "jmp" and "call"
1487 instructions with 2 immediate operands so that the immediate segment
1488 precedes the offset, as it does when in AT&T mode. "enter" and the
1489 intersegment "jmp" and "call" instructions are the only ones that
1490 have two immediate operands. */
1491 if (intel_syntax && i.operands > 1
1492 && (strcmp (mnemonic, "bound") != 0)
1493 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1494 {
1495 union i386_op temp_op;
1496 unsigned int temp_type;
1497 #ifdef BFD_ASSEMBLER
1498 enum bfd_reloc_code_real temp_reloc;
1499 #else
1500 int temp_reloc;
1501 #endif
1502 int xchg1 = 0;
1503 int xchg2 = 0;
1504
1505 if (i.operands == 2)
1506 {
1507 xchg1 = 0;
1508 xchg2 = 1;
1509 }
1510 else if (i.operands == 3)
1511 {
1512 xchg1 = 0;
1513 xchg2 = 2;
1514 }
1515 temp_type = i.types[xchg2];
1516 i.types[xchg2] = i.types[xchg1];
1517 i.types[xchg1] = temp_type;
1518 temp_op = i.op[xchg2];
1519 i.op[xchg2] = i.op[xchg1];
1520 i.op[xchg1] = temp_op;
1521 temp_reloc = i.disp_reloc[xchg2];
1522 i.disp_reloc[xchg2] = i.disp_reloc[xchg1];
1523 i.disp_reloc[xchg1] = temp_reloc;
1524
1525 if (i.mem_operands == 2)
1526 {
1527 const seg_entry *temp_seg;
1528 temp_seg = i.seg[0];
1529 i.seg[0] = i.seg[1];
1530 i.seg[1] = temp_seg;
1531 }
1532 }
1533
1534 if (i.imm_operands)
1535 {
1536 /* Try to ensure constant immediates are represented in the smallest
1537 opcode possible. */
1538 char guess_suffix = 0;
1539 int op;
1540
1541 if (i.suffix)
1542 guess_suffix = i.suffix;
1543 else if (i.reg_operands)
1544 {
1545 /* Figure out a suffix from the last register operand specified.
1546 We can't do this properly yet, ie. excluding InOutPortReg,
1547 but the following works for instructions with immediates.
1548 In any case, we can't set i.suffix yet. */
1549 for (op = i.operands; --op >= 0;)
1550 if (i.types[op] & Reg)
1551 {
1552 if (i.types[op] & Reg8)
1553 guess_suffix = BYTE_MNEM_SUFFIX;
1554 else if (i.types[op] & Reg16)
1555 guess_suffix = WORD_MNEM_SUFFIX;
1556 else if (i.types[op] & Reg32)
1557 guess_suffix = LONG_MNEM_SUFFIX;
1558 else if (i.types[op] & Reg64)
1559 guess_suffix = QWORD_MNEM_SUFFIX;
1560 break;
1561 }
1562 }
1563 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
1564 guess_suffix = WORD_MNEM_SUFFIX;
1565
1566 for (op = i.operands; --op >= 0;)
1567 if (i.types[op] & Imm)
1568 {
1569 switch (i.op[op].imms->X_op)
1570 {
1571 case O_constant:
1572 /* If a suffix is given, this operand may be shortened. */
1573 switch (guess_suffix)
1574 {
1575 case LONG_MNEM_SUFFIX:
1576 i.types[op] |= Imm32 | Imm64;
1577 break;
1578 case WORD_MNEM_SUFFIX:
1579 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
1580 break;
1581 case BYTE_MNEM_SUFFIX:
1582 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
1583 break;
1584 }
1585
1586 /* If this operand is at most 16 bits, convert it to a
1587 signed 16 bit number before trying to see whether it will
1588 fit in an even smaller size. This allows a 16-bit operand
1589 such as $0xffe0 to be recognised as within Imm8S range. */
1590 if ((i.types[op] & Imm16)
1591 && (i.op[op].imms->X_add_number & ~(offsetT)0xffff) == 0)
1592 {
1593 i.op[op].imms->X_add_number =
1594 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
1595 }
1596 if ((i.types[op] & Imm32)
1597 && (i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1)) == 0)
1598 {
1599 i.op[op].imms->X_add_number =
1600 (i.op[op].imms->X_add_number ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1601 }
1602 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
1603 /* We must avoid matching of Imm32 templates when 64bit only immediate is available. */
1604 if (guess_suffix == QWORD_MNEM_SUFFIX)
1605 i.types[op] &= ~Imm32;
1606 break;
1607 case O_absent:
1608 case O_register:
1609 abort();
1610 /* Symbols and expressions. */
1611 default:
1612 /* Convert symbolic operand to proper sizes for matching. */
1613 switch (guess_suffix)
1614 {
1615 case QWORD_MNEM_SUFFIX:
1616 i.types[op] = Imm64 | Imm32S;
1617 break;
1618 case LONG_MNEM_SUFFIX:
1619 i.types[op] = Imm32 | Imm64;
1620 break;
1621 case WORD_MNEM_SUFFIX:
1622 i.types[op] = Imm16 | Imm32 | Imm64;
1623 break;
1624 break;
1625 case BYTE_MNEM_SUFFIX:
1626 i.types[op] = Imm8 | Imm8S | Imm16 | Imm32S | Imm32;
1627 break;
1628 break;
1629 }
1630 break;
1631 }
1632 }
1633 }
1634
1635 if (i.disp_operands)
1636 {
1637 /* Try to use the smallest displacement type too. */
1638 int op;
1639
1640 for (op = i.operands; --op >= 0;)
1641 if ((i.types[op] & Disp)
1642 && i.op[op].imms->X_op == O_constant)
1643 {
1644 offsetT disp = i.op[op].disps->X_add_number;
1645
1646 if (i.types[op] & Disp16)
1647 {
1648 /* We know this operand is at most 16 bits, so
1649 convert to a signed 16 bit number before trying
1650 to see whether it will fit in an even smaller
1651 size. */
1652
1653 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
1654 }
1655 else if (i.types[op] & Disp32)
1656 {
1657 /* We know this operand is at most 32 bits, so convert to a
1658 signed 32 bit number before trying to see whether it will
1659 fit in an even smaller size. */
1660 disp &= (((offsetT) 2 << 31) - 1);
1661 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1662 }
1663 if (flag_code == CODE_64BIT)
1664 {
1665 if (fits_in_signed_long (disp))
1666 i.types[op] |= Disp32S;
1667 if (fits_in_unsigned_long (disp))
1668 i.types[op] |= Disp32;
1669 }
1670 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
1671 && fits_in_signed_byte (disp))
1672 i.types[op] |= Disp8;
1673 }
1674 }
1675
1676 overlap0 = 0;
1677 overlap1 = 0;
1678 overlap2 = 0;
1679 found_reverse_match = 0;
1680 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
1681 ? No_bSuf
1682 : (i.suffix == WORD_MNEM_SUFFIX
1683 ? No_wSuf
1684 : (i.suffix == SHORT_MNEM_SUFFIX
1685 ? No_sSuf
1686 : (i.suffix == LONG_MNEM_SUFFIX
1687 ? No_lSuf
1688 : (i.suffix == QWORD_MNEM_SUFFIX
1689 ? No_qSuf
1690 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX ? No_xSuf : 0))))));
1691
1692 for (t = current_templates->start;
1693 t < current_templates->end;
1694 t++)
1695 {
1696 /* Must have right number of operands. */
1697 if (i.operands != t->operands)
1698 continue;
1699
1700 /* Check the suffix, except for some instructions in intel mode. */
1701 if ((t->opcode_modifier & suffix_check)
1702 && !(intel_syntax
1703 && (t->opcode_modifier & IgnoreSize))
1704 && !(intel_syntax
1705 && t->base_opcode == 0xd9
1706 && (t->extension_opcode == 5 /* 0xd9,5 "fldcw" */
1707 || t->extension_opcode == 7))) /* 0xd9,7 "f{n}stcw" */
1708 continue;
1709
1710 else if (!t->operands)
1711 /* 0 operands always matches. */
1712 break;
1713
1714 overlap0 = i.types[0] & t->operand_types[0];
1715 switch (t->operands)
1716 {
1717 case 1:
1718 if (!MATCH (overlap0, i.types[0], t->operand_types[0]))
1719 continue;
1720 break;
1721 case 2:
1722 case 3:
1723 overlap1 = i.types[1] & t->operand_types[1];
1724 if (!MATCH (overlap0, i.types[0], t->operand_types[0])
1725 || !MATCH (overlap1, i.types[1], t->operand_types[1])
1726 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
1727 t->operand_types[0],
1728 overlap1, i.types[1],
1729 t->operand_types[1]))
1730 {
1731 /* Check if other direction is valid ... */
1732 if ((t->opcode_modifier & (D|FloatD)) == 0)
1733 continue;
1734
1735 /* Try reversing direction of operands. */
1736 overlap0 = i.types[0] & t->operand_types[1];
1737 overlap1 = i.types[1] & t->operand_types[0];
1738 if (!MATCH (overlap0, i.types[0], t->operand_types[1])
1739 || !MATCH (overlap1, i.types[1], t->operand_types[0])
1740 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
1741 t->operand_types[1],
1742 overlap1, i.types[1],
1743 t->operand_types[0]))
1744 {
1745 /* Does not match either direction. */
1746 continue;
1747 }
1748 /* found_reverse_match holds which of D or FloatDR
1749 we've found. */
1750 found_reverse_match = t->opcode_modifier & (D|FloatDR);
1751 }
1752 /* Found a forward 2 operand match here. */
1753 else if (t->operands == 3)
1754 {
1755 /* Here we make use of the fact that there are no
1756 reverse match 3 operand instructions, and all 3
1757 operand instructions only need to be checked for
1758 register consistency between operands 2 and 3. */
1759 overlap2 = i.types[2] & t->operand_types[2];
1760 if (!MATCH (overlap2, i.types[2], t->operand_types[2])
1761 || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1],
1762 t->operand_types[1],
1763 overlap2, i.types[2],
1764 t->operand_types[2]))
1765
1766 continue;
1767 }
1768 /* Found either forward/reverse 2 or 3 operand match here:
1769 slip through to break. */
1770 }
1771 if (t->cpu_flags & ~cpu_arch_flags)
1772 {
1773 found_reverse_match = 0;
1774 continue;
1775 }
1776 /* We've found a match; break out of loop. */
1777 break;
1778 }
1779 if (t == current_templates->end)
1780 {
1781 /* We found no match. */
1782 as_bad (_("suffix or operands invalid for `%s'"),
1783 current_templates->start->name);
1784 return;
1785 }
1786
1787 if (!quiet_warnings)
1788 {
1789 if (!intel_syntax
1790 && ((i.types[0] & JumpAbsolute)
1791 != (t->operand_types[0] & JumpAbsolute)))
1792 {
1793 as_warn (_("indirect %s without `*'"), t->name);
1794 }
1795
1796 if ((t->opcode_modifier & (IsPrefix|IgnoreSize))
1797 == (IsPrefix|IgnoreSize))
1798 {
1799 /* Warn them that a data or address size prefix doesn't
1800 affect assembly of the next line of code. */
1801 as_warn (_("stand-alone `%s' prefix"), t->name);
1802 }
1803 }
1804
1805 /* Copy the template we found. */
1806 i.tm = *t;
1807 if (found_reverse_match)
1808 {
1809 /* If we found a reverse match we must alter the opcode
1810 direction bit. found_reverse_match holds bits to change
1811 (different for int & float insns). */
1812
1813 i.tm.base_opcode ^= found_reverse_match;
1814
1815 i.tm.operand_types[0] = t->operand_types[1];
1816 i.tm.operand_types[1] = t->operand_types[0];
1817 }
1818
1819 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1820 if (SYSV386_COMPAT
1821 && intel_syntax
1822 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1823 i.tm.base_opcode ^= FloatR;
1824
1825 if (i.tm.opcode_modifier & FWait)
1826 if (! add_prefix (FWAIT_OPCODE))
1827 return;
1828
1829 /* Check string instruction segment overrides. */
1830 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1831 {
1832 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
1833 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
1834 {
1835 if (i.seg[0] != NULL && i.seg[0] != &es)
1836 {
1837 as_bad (_("`%s' operand %d must use `%%es' segment"),
1838 i.tm.name,
1839 mem_op + 1);
1840 return;
1841 }
1842 /* There's only ever one segment override allowed per instruction.
1843 This instruction possibly has a legal segment override on the
1844 second operand, so copy the segment to where non-string
1845 instructions store it, allowing common code. */
1846 i.seg[0] = i.seg[1];
1847 }
1848 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
1849 {
1850 if (i.seg[1] != NULL && i.seg[1] != &es)
1851 {
1852 as_bad (_("`%s' operand %d must use `%%es' segment"),
1853 i.tm.name,
1854 mem_op + 2);
1855 return;
1856 }
1857 }
1858 }
1859
1860 if (i.reg_operands && flag_code < CODE_64BIT)
1861 {
1862 int op;
1863 for (op = i.operands; --op >= 0; )
1864 if ((i.types[op] & Reg)
1865 && (i.op[op].regs->reg_flags & (RegRex64|RegRex)))
1866 as_bad (_("Extended register `%%%s' available only in 64bit mode."),
1867 i.op[op].regs->reg_name);
1868 }
1869
1870 /* If matched instruction specifies an explicit instruction mnemonic
1871 suffix, use it. */
1872 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
1873 {
1874 if (i.tm.opcode_modifier & Size16)
1875 i.suffix = WORD_MNEM_SUFFIX;
1876 else if (i.tm.opcode_modifier & Size64)
1877 i.suffix = QWORD_MNEM_SUFFIX;
1878 else
1879 i.suffix = LONG_MNEM_SUFFIX;
1880 }
1881 else if (i.reg_operands)
1882 {
1883 /* If there's no instruction mnemonic suffix we try to invent one
1884 based on register operands. */
1885 if (!i.suffix)
1886 {
1887 /* We take i.suffix from the last register operand specified,
1888 Destination register type is more significant than source
1889 register type. */
1890 int op;
1891 for (op = i.operands; --op >= 0;)
1892 if ((i.types[op] & Reg)
1893 && !(i.tm.operand_types[op] & InOutPortReg))
1894 {
1895 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
1896 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
1897 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
1898 LONG_MNEM_SUFFIX);
1899 break;
1900 }
1901 }
1902 else if (i.suffix == BYTE_MNEM_SUFFIX)
1903 {
1904 int op;
1905 for (op = i.operands; --op >= 0;)
1906 {
1907 /* If this is an eight bit register, it's OK. If it's
1908 the 16 or 32 bit version of an eight bit register,
1909 we will just use the low portion, and that's OK too. */
1910 if (i.types[op] & Reg8)
1911 continue;
1912
1913 /* movzx and movsx should not generate this warning. */
1914 if (intel_syntax
1915 && (i.tm.base_opcode == 0xfb7
1916 || i.tm.base_opcode == 0xfb6
1917 || i.tm.base_opcode == 0x63
1918 || i.tm.base_opcode == 0xfbe
1919 || i.tm.base_opcode == 0xfbf))
1920 continue;
1921
1922 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4
1923 #if 0
1924 /* Check that the template allows eight bit regs
1925 This kills insns such as `orb $1,%edx', which
1926 maybe should be allowed. */
1927 && (i.tm.operand_types[op] & (Reg8|InOutPortReg))
1928 #endif
1929 )
1930 {
1931 /* Prohibit these changes in the 64bit mode, since
1932 the lowering is more complicated. */
1933 if (flag_code == CODE_64BIT
1934 && (i.tm.operand_types[op] & InOutPortReg) == 0)
1935 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
1936 i.op[op].regs->reg_name,
1937 i.suffix);
1938 #if REGISTER_WARNINGS
1939 if (!quiet_warnings
1940 && (i.tm.operand_types[op] & InOutPortReg) == 0)
1941 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
1942 (i.op[op].regs - (i.types[op] & Reg16 ? 8 : 16))->reg_name,
1943 i.op[op].regs->reg_name,
1944 i.suffix);
1945 #endif
1946 continue;
1947 }
1948 /* Any other register is bad. */
1949 if (i.types[op] & (Reg | RegMMX | RegXMM
1950 | SReg2 | SReg3
1951 | Control | Debug | Test
1952 | FloatReg | FloatAcc))
1953 {
1954 as_bad (_("`%%%s' not allowed with `%s%c'"),
1955 i.op[op].regs->reg_name,
1956 i.tm.name,
1957 i.suffix);
1958 return;
1959 }
1960 }
1961 }
1962 else if (i.suffix == LONG_MNEM_SUFFIX)
1963 {
1964 int op;
1965
1966 for (op = i.operands; --op >= 0;)
1967 /* Reject eight bit registers, except where the template
1968 requires them. (eg. movzb) */
1969 if ((i.types[op] & Reg8) != 0
1970 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
1971 {
1972 as_bad (_("`%%%s' not allowed with `%s%c'"),
1973 i.op[op].regs->reg_name,
1974 i.tm.name,
1975 i.suffix);
1976 return;
1977 }
1978 /* Warn if the e prefix on a general reg is missing. */
1979 else if ((!quiet_warnings || flag_code == CODE_64BIT)
1980 && (i.types[op] & Reg16) != 0
1981 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
1982 {
1983 /* Prohibit these changes in the 64bit mode, since
1984 the lowering is more complicated. */
1985 if (flag_code == CODE_64BIT)
1986 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
1987 i.op[op].regs->reg_name,
1988 i.suffix);
1989 #if REGISTER_WARNINGS
1990 else
1991 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
1992 (i.op[op].regs + 8)->reg_name,
1993 i.op[op].regs->reg_name,
1994 i.suffix);
1995 #endif
1996 }
1997 /* Warn if the r prefix on a general reg is missing. */
1998 else if ((i.types[op] & Reg64) != 0
1999 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
2000 {
2001 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2002 i.op[op].regs->reg_name,
2003 i.suffix);
2004 }
2005 }
2006 else if (i.suffix == QWORD_MNEM_SUFFIX)
2007 {
2008 int op;
2009 if (flag_code < CODE_64BIT)
2010 as_bad (_("64bit operations available only in 64bit modes."));
2011
2012 for (op = i.operands; --op >= 0; )
2013 /* Reject eight bit registers, except where the template
2014 requires them. (eg. movzb) */
2015 if ((i.types[op] & Reg8) != 0
2016 && (i.tm.operand_types[op] & (Reg16|Reg32|Acc)) != 0)
2017 {
2018 as_bad (_("`%%%s' not allowed with `%s%c'"),
2019 i.op[op].regs->reg_name,
2020 i.tm.name,
2021 i.suffix);
2022 return;
2023 }
2024 /* Warn if the e prefix on a general reg is missing. */
2025 else if (((i.types[op] & Reg16) != 0
2026 || (i.types[op] & Reg32) != 0)
2027 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
2028 {
2029 /* Prohibit these changes in the 64bit mode, since
2030 the lowering is more complicated. */
2031 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2032 i.op[op].regs->reg_name,
2033 i.suffix);
2034 }
2035 }
2036 else if (i.suffix == WORD_MNEM_SUFFIX)
2037 {
2038 int op;
2039 for (op = i.operands; --op >= 0;)
2040 /* Reject eight bit registers, except where the template
2041 requires them. (eg. movzb) */
2042 if ((i.types[op] & Reg8) != 0
2043 && (i.tm.operand_types[op] & (Reg16|Reg32|Acc)) != 0)
2044 {
2045 as_bad (_("`%%%s' not allowed with `%s%c'"),
2046 i.op[op].regs->reg_name,
2047 i.tm.name,
2048 i.suffix);
2049 return;
2050 }
2051 /* Warn if the e prefix on a general reg is present. */
2052 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2053 && (i.types[op] & Reg32) != 0
2054 && (i.tm.operand_types[op] & (Reg16|Acc)) != 0)
2055 {
2056 /* Prohibit these changes in the 64bit mode, since
2057 the lowering is more complicated. */
2058 if (flag_code == CODE_64BIT)
2059 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2060 i.op[op].regs->reg_name,
2061 i.suffix);
2062 else
2063 #if REGISTER_WARNINGS
2064 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2065 (i.op[op].regs - 8)->reg_name,
2066 i.op[op].regs->reg_name,
2067 i.suffix);
2068 #endif
2069 }
2070 }
2071 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2072 /* Do nothing if the instruction is going to ignore the prefix. */
2073 ;
2074 else
2075 abort ();
2076 }
2077 else if ((i.tm.opcode_modifier & DefaultSize) && !i.suffix)
2078 {
2079 i.suffix = stackop_size;
2080 }
2081 /* Make still unresolved immediate matches conform to size of immediate
2082 given in i.suffix. Note: overlap2 cannot be an immediate! */
2083 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S))
2084 && overlap0 != Imm8 && overlap0 != Imm8S
2085 && overlap0 != Imm16 && overlap0 != Imm32S
2086 && overlap0 != Imm32 && overlap0 != Imm64)
2087 {
2088 if (i.suffix)
2089 {
2090 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
2091 (i.suffix == WORD_MNEM_SUFFIX ? Imm16 :
2092 (i.suffix == QWORD_MNEM_SUFFIX ? Imm64 | Imm32S : Imm32)));
2093 }
2094 else if (overlap0 == (Imm16 | Imm32S | Imm32)
2095 || overlap0 == (Imm16 | Imm32)
2096 || overlap0 == (Imm16 | Imm32S))
2097 {
2098 overlap0 =
2099 ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32S;
2100 }
2101 if (overlap0 != Imm8 && overlap0 != Imm8S
2102 && overlap0 != Imm16 && overlap0 != Imm32S
2103 && overlap0 != Imm32 && overlap0 != Imm64)
2104 {
2105 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
2106 return;
2107 }
2108 }
2109 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32))
2110 && overlap1 != Imm8 && overlap1 != Imm8S
2111 && overlap1 != Imm16 && overlap1 != Imm32S
2112 && overlap1 != Imm32 && overlap1 != Imm64)
2113 {
2114 if (i.suffix)
2115 {
2116 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
2117 (i.suffix == WORD_MNEM_SUFFIX ? Imm16 :
2118 (i.suffix == QWORD_MNEM_SUFFIX ? Imm64 | Imm32S : Imm32)));
2119 }
2120 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
2121 || overlap1 == (Imm16 | Imm32)
2122 || overlap1 == (Imm16 | Imm32S))
2123 {
2124 overlap1 =
2125 ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32S;
2126 }
2127 if (overlap1 != Imm8 && overlap1 != Imm8S
2128 && overlap1 != Imm16 && overlap1 != Imm32S
2129 && overlap1 != Imm32 && overlap1 != Imm64)
2130 {
2131 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
2132 return;
2133 }
2134 }
2135 assert ((overlap2 & Imm) == 0);
2136
2137 i.types[0] = overlap0;
2138 if (overlap0 & ImplicitRegister)
2139 i.reg_operands--;
2140 if (overlap0 & Imm1)
2141 i.imm_operands = 0; /* kludge for shift insns. */
2142
2143 i.types[1] = overlap1;
2144 if (overlap1 & ImplicitRegister)
2145 i.reg_operands--;
2146
2147 i.types[2] = overlap2;
2148 if (overlap2 & ImplicitRegister)
2149 i.reg_operands--;
2150
2151 /* Finalize opcode. First, we change the opcode based on the operand
2152 size given by i.suffix: We need not change things for byte insns. */
2153
2154 if (!i.suffix && (i.tm.opcode_modifier & W))
2155 {
2156 as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
2157 return;
2158 }
2159
2160 /* For movzx and movsx, need to check the register type. */
2161 if (intel_syntax
2162 && (i.tm.base_opcode == 0xfb6 || i.tm.base_opcode == 0xfbe))
2163 if (i.suffix && i.suffix == BYTE_MNEM_SUFFIX)
2164 {
2165 unsigned int prefix = DATA_PREFIX_OPCODE;
2166
2167 if ((i.op[1].regs->reg_type & Reg16) != 0)
2168 if (!add_prefix (prefix))
2169 return;
2170 }
2171
2172 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2173 {
2174 /* It's not a byte, select word/dword operation. */
2175 if (i.tm.opcode_modifier & W)
2176 {
2177 if (i.tm.opcode_modifier & ShortForm)
2178 i.tm.base_opcode |= 8;
2179 else
2180 i.tm.base_opcode |= 1;
2181 }
2182 /* Now select between word & dword operations via the operand
2183 size prefix, except for instructions that will ignore this
2184 prefix anyway. */
2185 if (i.suffix != QWORD_MNEM_SUFFIX
2186 && (i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2187 && !(i.tm.opcode_modifier & IgnoreSize))
2188 {
2189 unsigned int prefix = DATA_PREFIX_OPCODE;
2190 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2191 prefix = ADDR_PREFIX_OPCODE;
2192
2193 if (! add_prefix (prefix))
2194 return;
2195 }
2196
2197 /* Set mode64 for an operand. */
2198 if (i.suffix == QWORD_MNEM_SUFFIX
2199 && !(i.tm.opcode_modifier & NoRex64))
2200 i.rex.mode64 = 1;
2201
2202 /* Size floating point instruction. */
2203 if (i.suffix == LONG_MNEM_SUFFIX)
2204 {
2205 if (i.tm.opcode_modifier & FloatMF)
2206 i.tm.base_opcode ^= 4;
2207 }
2208 }
2209
2210 if (i.tm.opcode_modifier & ImmExt)
2211 {
2212 /* These AMD 3DNow! and Intel Katmai New Instructions have an
2213 opcode suffix which is coded in the same place as an 8-bit
2214 immediate field would be. Here we fake an 8-bit immediate
2215 operand from the opcode suffix stored in tm.extension_opcode. */
2216
2217 expressionS *exp;
2218
2219 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
2220
2221 exp = &im_expressions[i.imm_operands++];
2222 i.op[i.operands].imms = exp;
2223 i.types[i.operands++] = Imm8;
2224 exp->X_op = O_constant;
2225 exp->X_add_number = i.tm.extension_opcode;
2226 i.tm.extension_opcode = None;
2227 }
2228
2229 /* For insns with operands there are more diddles to do to the opcode. */
2230 if (i.operands)
2231 {
2232 /* Default segment register this instruction will use
2233 for memory accesses. 0 means unknown.
2234 This is only for optimizing out unnecessary segment overrides. */
2235 const seg_entry *default_seg = 0;
2236
2237 /* The imul $imm, %reg instruction is converted into
2238 imul $imm, %reg, %reg, and the clr %reg instruction
2239 is converted into xor %reg, %reg. */
2240 if (i.tm.opcode_modifier & regKludge)
2241 {
2242 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
2243 /* Pretend we saw the extra register operand. */
2244 assert (i.op[first_reg_op + 1].regs == 0);
2245 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
2246 i.types[first_reg_op + 1] = i.types[first_reg_op];
2247 i.reg_operands = 2;
2248 }
2249
2250 if (i.tm.opcode_modifier & ShortForm)
2251 {
2252 /* The register or float register operand is in operand 0 or 1. */
2253 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
2254 /* Register goes in low 3 bits of opcode. */
2255 i.tm.base_opcode |= i.op[op].regs->reg_num;
2256 if (i.op[op].regs->reg_flags & RegRex)
2257 i.rex.extZ=1;
2258 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2259 {
2260 /* Warn about some common errors, but press on regardless.
2261 The first case can be generated by gcc (<= 2.8.1). */
2262 if (i.operands == 2)
2263 {
2264 /* Reversed arguments on faddp, fsubp, etc. */
2265 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
2266 i.op[1].regs->reg_name,
2267 i.op[0].regs->reg_name);
2268 }
2269 else
2270 {
2271 /* Extraneous `l' suffix on fp insn. */
2272 as_warn (_("translating to `%s %%%s'"), i.tm.name,
2273 i.op[0].regs->reg_name);
2274 }
2275 }
2276 }
2277 else if (i.tm.opcode_modifier & Modrm)
2278 {
2279 /* The opcode is completed (modulo i.tm.extension_opcode which
2280 must be put into the modrm byte).
2281 Now, we make the modrm & index base bytes based on all the
2282 info we've collected. */
2283
2284 /* i.reg_operands MUST be the number of real register operands;
2285 implicit registers do not count. */
2286 if (i.reg_operands == 2)
2287 {
2288 unsigned int source, dest;
2289 source = ((i.types[0]
2290 & (Reg | RegMMX | RegXMM
2291 | SReg2 | SReg3
2292 | Control | Debug | Test))
2293 ? 0 : 1);
2294 dest = source + 1;
2295
2296 i.rm.mode = 3;
2297 /* One of the register operands will be encoded in the
2298 i.tm.reg field, the other in the combined i.tm.mode
2299 and i.tm.regmem fields. If no form of this
2300 instruction supports a memory destination operand,
2301 then we assume the source operand may sometimes be
2302 a memory operand and so we need to store the
2303 destination in the i.rm.reg field. */
2304 if ((i.tm.operand_types[dest] & AnyMem) == 0)
2305 {
2306 i.rm.reg = i.op[dest].regs->reg_num;
2307 i.rm.regmem = i.op[source].regs->reg_num;
2308 if (i.op[dest].regs->reg_flags & RegRex)
2309 i.rex.extX=1;
2310 if (i.op[source].regs->reg_flags & RegRex)
2311 i.rex.extZ=1;
2312 }
2313 else
2314 {
2315 i.rm.reg = i.op[source].regs->reg_num;
2316 i.rm.regmem = i.op[dest].regs->reg_num;
2317 if (i.op[dest].regs->reg_flags & RegRex)
2318 i.rex.extZ=1;
2319 if (i.op[source].regs->reg_flags & RegRex)
2320 i.rex.extX=1;
2321 }
2322 }
2323 else
2324 { /* If it's not 2 reg operands... */
2325 if (i.mem_operands)
2326 {
2327 unsigned int fake_zero_displacement = 0;
2328 unsigned int op = ((i.types[0] & AnyMem)
2329 ? 0
2330 : (i.types[1] & AnyMem) ? 1 : 2);
2331
2332 default_seg = &ds;
2333
2334 if (! i.base_reg)
2335 {
2336 i.rm.mode = 0;
2337 if (! i.disp_operands)
2338 fake_zero_displacement = 1;
2339 if (! i.index_reg)
2340 {
2341 /* Operand is just <disp> */
2342 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
2343 {
2344 i.rm.regmem = NO_BASE_REGISTER_16;
2345 i.types[op] &= ~Disp;
2346 i.types[op] |= Disp16;
2347 }
2348 else if (flag_code != CODE_64BIT)
2349 {
2350 i.rm.regmem = NO_BASE_REGISTER;
2351 i.types[op] &= ~Disp;
2352 i.types[op] |= Disp32;
2353 }
2354 else
2355 {
2356 /* 64bit mode overwrites the 32bit absolute addressing
2357 by RIP relative addressing and absolute addressing
2358 is encoded by one of the redundant SIB forms. */
2359
2360 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2361 i.sib.base = NO_BASE_REGISTER;
2362 i.sib.index = NO_INDEX_REGISTER;
2363 i.types[op] &= ~Disp;
2364 i.types[op] |= Disp32S;
2365 }
2366 }
2367 else /* ! i.base_reg && i.index_reg */
2368 {
2369 i.sib.index = i.index_reg->reg_num;
2370 i.sib.base = NO_BASE_REGISTER;
2371 i.sib.scale = i.log2_scale_factor;
2372 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2373 i.types[op] &= ~Disp;
2374 if (flag_code != CODE_64BIT)
2375 i.types[op] |= Disp32; /* Must be 32 bit */
2376 else
2377 i.types[op] |= Disp32S;
2378 if (i.index_reg->reg_flags & RegRex)
2379 i.rex.extY=1;
2380 }
2381 }
2382 /* RIP addressing for 64bit mode. */
2383 else if (i.base_reg->reg_type == BaseIndex)
2384 {
2385 i.rm.regmem = NO_BASE_REGISTER;
2386 i.types[op] &= ~Disp;
2387 i.types[op] |= Disp32S;
2388 i.flags[op] = Operand_PCrel;
2389 }
2390 else if (i.base_reg->reg_type & Reg16)
2391 {
2392 switch (i.base_reg->reg_num)
2393 {
2394 case 3: /* (%bx) */
2395 if (! i.index_reg)
2396 i.rm.regmem = 7;
2397 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
2398 i.rm.regmem = i.index_reg->reg_num - 6;
2399 break;
2400 case 5: /* (%bp) */
2401 default_seg = &ss;
2402 if (! i.index_reg)
2403 {
2404 i.rm.regmem = 6;
2405 if ((i.types[op] & Disp) == 0)
2406 {
2407 /* fake (%bp) into 0(%bp) */
2408 i.types[op] |= Disp8;
2409 fake_zero_displacement = 1;
2410 }
2411 }
2412 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
2413 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
2414 break;
2415 default: /* (%si) -> 4 or (%di) -> 5 */
2416 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
2417 }
2418 i.rm.mode = mode_from_disp_size (i.types[op]);
2419 }
2420 else /* i.base_reg and 32/64 bit mode */
2421 {
2422 if (flag_code == CODE_64BIT
2423 && (i.types[op] & Disp))
2424 {
2425 if (i.types[op] & Disp8)
2426 i.types[op] = Disp8 | Disp32S;
2427 else
2428 i.types[op] = Disp32S;
2429 }
2430 i.rm.regmem = i.base_reg->reg_num;
2431 if (i.base_reg->reg_flags & RegRex)
2432 i.rex.extZ=1;
2433 i.sib.base = i.base_reg->reg_num;
2434 /* x86-64 ignores REX prefix bit here to avoid
2435 decoder complications. */
2436 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
2437 {
2438 default_seg = &ss;
2439 if (i.disp_operands == 0)
2440 {
2441 fake_zero_displacement = 1;
2442 i.types[op] |= Disp8;
2443 }
2444 }
2445 else if (i.base_reg->reg_num == ESP_REG_NUM)
2446 {
2447 default_seg = &ss;
2448 }
2449 i.sib.scale = i.log2_scale_factor;
2450 if (! i.index_reg)
2451 {
2452 /* <disp>(%esp) becomes two byte modrm
2453 with no index register. We've already
2454 stored the code for esp in i.rm.regmem
2455 ie. ESCAPE_TO_TWO_BYTE_ADDRESSING. Any
2456 base register besides %esp will not use
2457 the extra modrm byte. */
2458 i.sib.index = NO_INDEX_REGISTER;
2459 #if ! SCALE1_WHEN_NO_INDEX
2460 /* Another case where we force the second
2461 modrm byte. */
2462 if (i.log2_scale_factor)
2463 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2464 #endif
2465 }
2466 else
2467 {
2468 i.sib.index = i.index_reg->reg_num;
2469 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2470 if (i.index_reg->reg_flags & RegRex)
2471 i.rex.extY=1;
2472 }
2473 i.rm.mode = mode_from_disp_size (i.types[op]);
2474 }
2475
2476 if (fake_zero_displacement)
2477 {
2478 /* Fakes a zero displacement assuming that i.types[op]
2479 holds the correct displacement size. */
2480 expressionS *exp;
2481
2482 assert (i.op[op].disps == 0);
2483 exp = &disp_expressions[i.disp_operands++];
2484 i.op[op].disps = exp;
2485 exp->X_op = O_constant;
2486 exp->X_add_number = 0;
2487 exp->X_add_symbol = (symbolS *) 0;
2488 exp->X_op_symbol = (symbolS *) 0;
2489 }
2490 }
2491
2492 /* Fill in i.rm.reg or i.rm.regmem field with register
2493 operand (if any) based on i.tm.extension_opcode.
2494 Again, we must be careful to make sure that
2495 segment/control/debug/test/MMX registers are coded
2496 into the i.rm.reg field. */
2497 if (i.reg_operands)
2498 {
2499 unsigned int op =
2500 ((i.types[0]
2501 & (Reg | RegMMX | RegXMM
2502 | SReg2 | SReg3
2503 | Control | Debug | Test))
2504 ? 0
2505 : ((i.types[1]
2506 & (Reg | RegMMX | RegXMM
2507 | SReg2 | SReg3
2508 | Control | Debug | Test))
2509 ? 1
2510 : 2));
2511 /* If there is an extension opcode to put here, the
2512 register number must be put into the regmem field. */
2513 if (i.tm.extension_opcode != None)
2514 {
2515 i.rm.regmem = i.op[op].regs->reg_num;
2516 if (i.op[op].regs->reg_flags & RegRex)
2517 i.rex.extZ=1;
2518 }
2519 else
2520 {
2521 i.rm.reg = i.op[op].regs->reg_num;
2522 if (i.op[op].regs->reg_flags & RegRex)
2523 i.rex.extX=1;
2524 }
2525
2526 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2
2527 we must set it to 3 to indicate this is a register
2528 operand in the regmem field. */
2529 if (!i.mem_operands)
2530 i.rm.mode = 3;
2531 }
2532
2533 /* Fill in i.rm.reg field with extension opcode (if any). */
2534 if (i.tm.extension_opcode != None)
2535 i.rm.reg = i.tm.extension_opcode;
2536 }
2537 }
2538 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
2539 {
2540 if (i.tm.base_opcode == POP_SEG_SHORT
2541 && i.op[0].regs->reg_num == 1)
2542 {
2543 as_bad (_("you can't `pop %%cs'"));
2544 return;
2545 }
2546 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
2547 if (i.op[0].regs->reg_flags & RegRex)
2548 i.rex.extZ = 1;
2549 }
2550 else if ((i.tm.base_opcode & ~(D|W)) == MOV_AX_DISP32)
2551 {
2552 default_seg = &ds;
2553 }
2554 else if ((i.tm.opcode_modifier & IsString) != 0)
2555 {
2556 /* For the string instructions that allow a segment override
2557 on one of their operands, the default segment is ds. */
2558 default_seg = &ds;
2559 }
2560
2561 /* If a segment was explicitly specified,
2562 and the specified segment is not the default,
2563 use an opcode prefix to select it.
2564 If we never figured out what the default segment is,
2565 then default_seg will be zero at this point,
2566 and the specified segment prefix will always be used. */
2567 if ((i.seg[0]) && (i.seg[0] != default_seg))
2568 {
2569 if (! add_prefix (i.seg[0]->seg_prefix))
2570 return;
2571 }
2572 }
2573 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2574 {
2575 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
2576 as_warn (_("translating to `%sp'"), i.tm.name);
2577 }
2578 }
2579
2580 /* Handle conversion of 'int $3' --> special int3 insn. */
2581 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
2582 {
2583 i.tm.base_opcode = INT3_OPCODE;
2584 i.imm_operands = 0;
2585 }
2586
2587 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
2588 && i.op[0].disps->X_op == O_constant)
2589 {
2590 /* Convert "jmp constant" (and "call constant") to a jump (call) to
2591 the absolute address given by the constant. Since ix86 jumps and
2592 calls are pc relative, we need to generate a reloc. */
2593 i.op[0].disps->X_add_symbol = &abs_symbol;
2594 i.op[0].disps->X_op = O_symbol;
2595 }
2596
2597 if (i.tm.opcode_modifier & Rex64)
2598 i.rex.mode64 = 1;
2599
2600 /* For 8bit registers we would need an empty rex prefix.
2601 Also in the case instruction is already having prefix,
2602 we need to convert old registers to new ones. */
2603
2604 if (((i.types[0] & Reg8) && (i.op[0].regs->reg_flags & RegRex64))
2605 || ((i.types[1] & Reg8) && (i.op[1].regs->reg_flags & RegRex64))
2606 || ((i.rex.mode64 || i.rex.extX || i.rex.extY || i.rex.extZ || i.rex.empty)
2607 && ((i.types[0] & Reg8) || (i.types[1] & Reg8))))
2608 {
2609 int x;
2610 i.rex.empty=1;
2611 for (x = 0; x < 2; x++)
2612 {
2613 /* Look for 8bit operand that does use old registers. */
2614 if (i.types[x] & Reg8
2615 && !(i.op[x].regs->reg_flags & RegRex64))
2616 {
2617 /* In case it is "hi" register, give up. */
2618 if (i.op[x].regs->reg_num > 3)
2619 as_bad (_("Can't encode registers '%%%s' in the instruction requiring REX prefix.\n"),
2620 i.op[x].regs->reg_name);
2621
2622 /* Otherwise it is equivalent to the extended register.
2623 Since the encoding don't change this is merely cosmetical
2624 cleanup for debug output. */
2625
2626 i.op[x].regs = i.op[x].regs + 8;
2627 }
2628 }
2629 }
2630
2631 if (i.rex.mode64 || i.rex.extX || i.rex.extY || i.rex.extZ || i.rex.empty)
2632 add_prefix (0x40
2633 | (i.rex.mode64 ? 8 : 0)
2634 | (i.rex.extX ? 4 : 0)
2635 | (i.rex.extY ? 2 : 0)
2636 | (i.rex.extZ ? 1 : 0));
2637
2638 /* We are ready to output the insn. */
2639 {
2640 register char *p;
2641
2642 /* Output jumps. */
2643 if (i.tm.opcode_modifier & Jump)
2644 {
2645 int size;
2646 int code16;
2647 int prefix;
2648
2649 code16 = 0;
2650 if (flag_code == CODE_16BIT)
2651 code16 = CODE16;
2652
2653 prefix = 0;
2654 if (i.prefix[DATA_PREFIX])
2655 {
2656 prefix = 1;
2657 i.prefixes -= 1;
2658 code16 ^= CODE16;
2659 }
2660 if (i.prefix[REX_PREFIX])
2661 {
2662 prefix++;
2663 i.prefixes --;
2664 }
2665
2666 size = 4;
2667 if (code16)
2668 size = 2;
2669
2670 if (i.prefixes != 0 && !intel_syntax)
2671 as_warn (_("skipping prefixes on this instruction"));
2672
2673 /* It's always a symbol; End frag & setup for relax.
2674 Make sure there is enough room in this frag for the largest
2675 instruction we may generate in md_convert_frag. This is 2
2676 bytes for the opcode and room for the prefix and largest
2677 displacement. */
2678 frag_grow (prefix + 2 + size);
2679 insn_size += prefix + 1;
2680 /* Prefix and 1 opcode byte go in fr_fix. */
2681 p = frag_more (prefix + 1);
2682 if (i.prefix[DATA_PREFIX])
2683 *p++ = DATA_PREFIX_OPCODE;
2684 if (i.prefix[REX_PREFIX])
2685 *p++ = i.prefix[REX_PREFIX];
2686 *p = i.tm.base_opcode;
2687 /* 1 possible extra opcode + displacement go in var part.
2688 Pass reloc in fr_var. */
2689 frag_var (rs_machine_dependent,
2690 1 + size,
2691 i.disp_reloc[0],
2692 ((unsigned char) *p == JUMP_PC_RELATIVE
2693 ? ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL) | code16
2694 : ENCODE_RELAX_STATE (COND_JUMP, SMALL) | code16),
2695 i.op[0].disps->X_add_symbol,
2696 i.op[0].disps->X_add_number,
2697 p);
2698 }
2699 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
2700 {
2701 int size;
2702
2703 if (i.tm.opcode_modifier & JumpByte)
2704 {
2705 /* This is a loop or jecxz type instruction. */
2706 size = 1;
2707 if (i.prefix[ADDR_PREFIX])
2708 {
2709 insn_size += 1;
2710 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
2711 i.prefixes -= 1;
2712 }
2713 }
2714 else
2715 {
2716 int code16;
2717
2718 code16 = 0;
2719 if (flag_code == CODE_16BIT)
2720 code16 = CODE16;
2721
2722 if (i.prefix[DATA_PREFIX])
2723 {
2724 insn_size += 1;
2725 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
2726 i.prefixes -= 1;
2727 code16 ^= CODE16;
2728 }
2729
2730 size = 4;
2731 if (code16)
2732 size = 2;
2733 }
2734
2735 if (i.prefix[REX_PREFIX])
2736 {
2737 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
2738 insn_size++;
2739 i.prefixes -= 1;
2740 }
2741
2742 if (i.prefixes != 0 && !intel_syntax)
2743 as_warn (_("skipping prefixes on this instruction"));
2744
2745 if (fits_in_unsigned_byte (i.tm.base_opcode))
2746 {
2747 insn_size += 1 + size;
2748 p = frag_more (1 + size);
2749 }
2750 else
2751 {
2752 /* Opcode can be at most two bytes. */
2753 insn_size += 2 + size;
2754 p = frag_more (2 + size);
2755 *p++ = (i.tm.base_opcode >> 8) & 0xff;
2756 }
2757 *p++ = i.tm.base_opcode & 0xff;
2758
2759 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2760 i.op[0].disps, 1, reloc (size, 1, 1, i.disp_reloc[0]));
2761 }
2762 else if (i.tm.opcode_modifier & JumpInterSegment)
2763 {
2764 int size;
2765 int prefix;
2766 int code16;
2767
2768 code16 = 0;
2769 if (flag_code == CODE_16BIT)
2770 code16 = CODE16;
2771
2772 prefix = 0;
2773 if (i.prefix[DATA_PREFIX])
2774 {
2775 prefix = 1;
2776 i.prefixes -= 1;
2777 code16 ^= CODE16;
2778 }
2779 if (i.prefix[REX_PREFIX])
2780 {
2781 prefix++;
2782 i.prefixes -= 1;
2783 }
2784
2785 size = 4;
2786 if (code16)
2787 size = 2;
2788
2789 if (i.prefixes != 0 && !intel_syntax)
2790 as_warn (_("skipping prefixes on this instruction"));
2791
2792 /* 1 opcode; 2 segment; offset */
2793 insn_size += prefix + 1 + 2 + size;
2794 p = frag_more (prefix + 1 + 2 + size);
2795
2796 if (i.prefix[DATA_PREFIX])
2797 *p++ = DATA_PREFIX_OPCODE;
2798
2799 if (i.prefix[REX_PREFIX])
2800 *p++ = i.prefix[REX_PREFIX];
2801
2802 *p++ = i.tm.base_opcode;
2803 if (i.op[1].imms->X_op == O_constant)
2804 {
2805 offsetT n = i.op[1].imms->X_add_number;
2806
2807 if (size == 2
2808 && !fits_in_unsigned_word (n)
2809 && !fits_in_signed_word (n))
2810 {
2811 as_bad (_("16-bit jump out of range"));
2812 return;
2813 }
2814 md_number_to_chars (p, n, size);
2815 }
2816 else
2817 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2818 i.op[1].imms, 0, reloc (size, 0, 0, i.disp_reloc[0]));
2819 if (i.op[0].imms->X_op != O_constant)
2820 as_bad (_("can't handle non absolute segment in `%s'"),
2821 i.tm.name);
2822 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
2823 }
2824 else
2825 {
2826 /* Output normal instructions here. */
2827 unsigned char *q;
2828
2829 /* All opcodes on i386 have eighter 1 or 2 bytes. We may use third
2830 byte for the SSE instructions to specify prefix they require. */
2831 if (i.tm.base_opcode & 0xff0000)
2832 add_prefix ((i.tm.base_opcode >> 16) & 0xff);
2833
2834 /* The prefix bytes. */
2835 for (q = i.prefix;
2836 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
2837 q++)
2838 {
2839 if (*q)
2840 {
2841 insn_size += 1;
2842 p = frag_more (1);
2843 md_number_to_chars (p, (valueT) *q, 1);
2844 }
2845 }
2846
2847 /* Now the opcode; be careful about word order here! */
2848 if (fits_in_unsigned_byte (i.tm.base_opcode))
2849 {
2850 insn_size += 1;
2851 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
2852 }
2853 else
2854 {
2855 insn_size += 2;
2856 p = frag_more (2);
2857 /* Put out high byte first: can't use md_number_to_chars! */
2858 *p++ = (i.tm.base_opcode >> 8) & 0xff;
2859 *p = i.tm.base_opcode & 0xff;
2860 }
2861
2862 /* Now the modrm byte and sib byte (if present). */
2863 if (i.tm.opcode_modifier & Modrm)
2864 {
2865 insn_size += 1;
2866 p = frag_more (1);
2867 md_number_to_chars (p,
2868 (valueT) (i.rm.regmem << 0
2869 | i.rm.reg << 3
2870 | i.rm.mode << 6),
2871 1);
2872 /* If i.rm.regmem == ESP (4)
2873 && i.rm.mode != (Register mode)
2874 && not 16 bit
2875 ==> need second modrm byte. */
2876 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
2877 && i.rm.mode != 3
2878 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
2879 {
2880 insn_size += 1;
2881 p = frag_more (1);
2882 md_number_to_chars (p,
2883 (valueT) (i.sib.base << 0
2884 | i.sib.index << 3
2885 | i.sib.scale << 6),
2886 1);
2887 }
2888 }
2889
2890 if (i.disp_operands)
2891 {
2892 register unsigned int n;
2893
2894 for (n = 0; n < i.operands; n++)
2895 {
2896 if (i.types[n] & Disp)
2897 {
2898 if (i.op[n].disps->X_op == O_constant)
2899 {
2900 int size;
2901 offsetT val;
2902
2903 size = 4;
2904 if (i.types[n] & (Disp8 | Disp16 | Disp64))
2905 {
2906 size = 2;
2907 if (i.types[n] & Disp8)
2908 size = 1;
2909 if (i.types[n] & Disp64)
2910 size = 8;
2911 }
2912 val = offset_in_range (i.op[n].disps->X_add_number,
2913 size);
2914 insn_size += size;
2915 p = frag_more (size);
2916 md_number_to_chars (p, val, size);
2917 }
2918 else
2919 {
2920 int size = 4;
2921 int sign = 0;
2922 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
2923
2924 /* The PC relative address is computed relative
2925 to the instruction boundary, so in case immediate
2926 fields follows, we need to adjust the value. */
2927 if (pcrel && i.imm_operands)
2928 {
2929 int imm_size = 4;
2930 register unsigned int n1;
2931
2932 for (n1 = 0; n1 < i.operands; n1++)
2933 if (i.types[n1] & Imm)
2934 {
2935 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
2936 {
2937 imm_size = 2;
2938 if (i.types[n1] & (Imm8 | Imm8S))
2939 imm_size = 1;
2940 if (i.types[n1] & Imm64)
2941 imm_size = 8;
2942 }
2943 break;
2944 }
2945 /* We should find the immediate. */
2946 if (n1 == i.operands)
2947 abort();
2948 i.op[n].disps->X_add_number -= imm_size;
2949 }
2950
2951 if (i.types[n] & Disp32S)
2952 sign = 1;
2953
2954 if (i.types[n] & (Disp16 | Disp64))
2955 {
2956 size = 2;
2957 if (i.types[n] & Disp64)
2958 size = 8;
2959 }
2960
2961 insn_size += size;
2962 p = frag_more (size);
2963 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2964 i.op[n].disps, pcrel,
2965 reloc (size, pcrel, sign, i.disp_reloc[n]));
2966 }
2967 }
2968 }
2969 }
2970
2971 /* Output immediate. */
2972 if (i.imm_operands)
2973 {
2974 register unsigned int n;
2975
2976 for (n = 0; n < i.operands; n++)
2977 {
2978 if (i.types[n] & Imm)
2979 {
2980 if (i.op[n].imms->X_op == O_constant)
2981 {
2982 int size;
2983 offsetT val;
2984
2985 size = 4;
2986 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
2987 {
2988 size = 2;
2989 if (i.types[n] & (Imm8 | Imm8S))
2990 size = 1;
2991 else if (i.types[n] & Imm64)
2992 size = 8;
2993 }
2994 val = offset_in_range (i.op[n].imms->X_add_number,
2995 size);
2996 insn_size += size;
2997 p = frag_more (size);
2998 md_number_to_chars (p, val, size);
2999 }
3000 else
3001 {
3002 /* Not absolute_section.
3003 Need a 32-bit fixup (don't support 8bit
3004 non-absolute imms). Try to support other
3005 sizes ... */
3006 #ifdef BFD_ASSEMBLER
3007 enum bfd_reloc_code_real reloc_type;
3008 #else
3009 int reloc_type;
3010 #endif
3011 int size = 4;
3012 int sign = 0;
3013
3014 if ((i.types[n] & (Imm32S))
3015 && i.suffix == QWORD_MNEM_SUFFIX)
3016 sign = 1;
3017 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3018 {
3019 size = 2;
3020 if (i.types[n] & (Imm8 | Imm8S))
3021 size = 1;
3022 if (i.types[n] & Imm64)
3023 size = 8;
3024 }
3025
3026 insn_size += size;
3027 p = frag_more (size);
3028 reloc_type = reloc (size, 0, sign, i.disp_reloc[0]);
3029 #ifdef BFD_ASSEMBLER
3030 if (reloc_type == BFD_RELOC_32
3031 && GOT_symbol
3032 && GOT_symbol == i.op[n].imms->X_add_symbol
3033 && (i.op[n].imms->X_op == O_symbol
3034 || (i.op[n].imms->X_op == O_add
3035 && ((symbol_get_value_expression
3036 (i.op[n].imms->X_op_symbol)->X_op)
3037 == O_subtract))))
3038 {
3039 /* We don't support dynamic linking on x86-64 yet. */
3040 if (flag_code == CODE_64BIT)
3041 abort();
3042 reloc_type = BFD_RELOC_386_GOTPC;
3043 i.op[n].imms->X_add_number += 3;
3044 }
3045 #endif
3046 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3047 i.op[n].imms, 0, reloc_type);
3048 }
3049 }
3050 }
3051 }
3052 }
3053
3054 dwarf2_emit_insn (insn_size);
3055
3056 #ifdef DEBUG386
3057 if (flag_debug)
3058 {
3059 pi (line, &i);
3060 }
3061 #endif /* DEBUG386 */
3062 }
3063 }
3064 \f
3065 static int i386_immediate PARAMS ((char *));
3066
3067 static int
3068 i386_immediate (imm_start)
3069 char *imm_start;
3070 {
3071 char *save_input_line_pointer;
3072 segT exp_seg = 0;
3073 expressionS *exp;
3074
3075 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
3076 {
3077 as_bad (_("only 1 or 2 immediate operands are allowed"));
3078 return 0;
3079 }
3080
3081 exp = &im_expressions[i.imm_operands++];
3082 i.op[this_operand].imms = exp;
3083
3084 if (is_space_char (*imm_start))
3085 ++imm_start;
3086
3087 save_input_line_pointer = input_line_pointer;
3088 input_line_pointer = imm_start;
3089
3090 #ifndef LEX_AT
3091 {
3092 /* We can have operands of the form
3093 <symbol>@GOTOFF+<nnn>
3094 Take the easy way out here and copy everything
3095 into a temporary buffer... */
3096 register char *cp;
3097
3098 cp = strchr (input_line_pointer, '@');
3099 if (cp != NULL)
3100 {
3101 char *tmpbuf;
3102 int len = 0;
3103 int first;
3104
3105 /* GOT relocations are not supported in 16 bit mode. */
3106 if (flag_code == CODE_16BIT)
3107 as_bad (_("GOT relocations not supported in 16 bit mode"));
3108
3109 if (GOT_symbol == NULL)
3110 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3111
3112 if (strncmp (cp + 1, "PLT", 3) == 0)
3113 {
3114 if (flag_code == CODE_64BIT)
3115 i.disp_reloc[this_operand] = BFD_RELOC_X86_64_PLT32;
3116 else
3117 i.disp_reloc[this_operand] = BFD_RELOC_386_PLT32;
3118 len = 3;
3119 }
3120 else if (strncmp (cp + 1, "GOTOFF", 6) == 0)
3121 {
3122 if (flag_code == CODE_64BIT)
3123 as_bad ("GOTOFF relocations are unsupported in 64bit mode.");
3124 i.disp_reloc[this_operand] = BFD_RELOC_386_GOTOFF;
3125 len = 6;
3126 }
3127 else if (strncmp (cp + 1, "GOT", 3) == 0)
3128 {
3129 if (flag_code == CODE_64BIT)
3130 i.disp_reloc[this_operand] = BFD_RELOC_X86_64_GOT32;
3131 else
3132 i.disp_reloc[this_operand] = BFD_RELOC_386_GOT32;
3133 len = 3;
3134 }
3135 else if (strncmp (cp + 1, "GOTPCREL", 3) == 0)
3136 {
3137 if (flag_code == CODE_64BIT)
3138 i.disp_reloc[this_operand] = BFD_RELOC_X86_64_GOTPCREL;
3139 else
3140 as_bad ("GOTPCREL relocations are supported only in 64bit mode.");
3141 len = 3;
3142 }
3143 else
3144 as_bad (_("bad reloc specifier in expression"));
3145
3146 /* Replace the relocation token with ' ', so that errors like
3147 foo@GOTOFF1 will be detected. */
3148 first = cp - input_line_pointer;
3149 tmpbuf = (char *) alloca (strlen (input_line_pointer));
3150 memcpy (tmpbuf, input_line_pointer, first);
3151 tmpbuf[first] = ' ';
3152 strcpy (tmpbuf + first + 1, cp + 1 + len);
3153 input_line_pointer = tmpbuf;
3154 }
3155 }
3156 #endif
3157
3158 exp_seg = expression (exp);
3159
3160 SKIP_WHITESPACE ();
3161 if (*input_line_pointer)
3162 as_bad (_("ignoring junk `%s' after expression"), input_line_pointer);
3163
3164 input_line_pointer = save_input_line_pointer;
3165
3166 if (exp->X_op == O_absent || exp->X_op == O_big)
3167 {
3168 /* Missing or bad expr becomes absolute 0. */
3169 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
3170 imm_start);
3171 exp->X_op = O_constant;
3172 exp->X_add_number = 0;
3173 exp->X_add_symbol = (symbolS *) 0;
3174 exp->X_op_symbol = (symbolS *) 0;
3175 }
3176 else if (exp->X_op == O_constant)
3177 {
3178 /* Size it properly later. */
3179 i.types[this_operand] |= Imm64;
3180 /* If BFD64, sign extend val. */
3181 if (!use_rela_relocations)
3182 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
3183 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
3184 }
3185 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3186 else if (1
3187 #ifdef BFD_ASSEMBLER
3188 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3189 #endif
3190 && exp_seg != text_section
3191 && exp_seg != data_section
3192 && exp_seg != bss_section
3193 && exp_seg != undefined_section
3194 #ifdef BFD_ASSEMBLER
3195 && !bfd_is_com_section (exp_seg)
3196 #endif
3197 )
3198 {
3199 #ifdef BFD_ASSEMBLER
3200 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3201 #else
3202 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3203 #endif
3204 return 0;
3205 }
3206 #endif
3207 else
3208 {
3209 /* This is an address. The size of the address will be
3210 determined later, depending on destination register,
3211 suffix, or the default for the section. */
3212 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
3213 }
3214
3215 return 1;
3216 }
3217
3218 static int i386_scale PARAMS ((char *));
3219
3220 static int
3221 i386_scale (scale)
3222 char *scale;
3223 {
3224 if (!isdigit (*scale))
3225 goto bad_scale;
3226
3227 switch (*scale)
3228 {
3229 case '0':
3230 case '1':
3231 i.log2_scale_factor = 0;
3232 break;
3233 case '2':
3234 i.log2_scale_factor = 1;
3235 break;
3236 case '4':
3237 i.log2_scale_factor = 2;
3238 break;
3239 case '8':
3240 i.log2_scale_factor = 3;
3241 break;
3242 default:
3243 bad_scale:
3244 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
3245 scale);
3246 return 0;
3247 }
3248 if (i.log2_scale_factor != 0 && ! i.index_reg)
3249 {
3250 as_warn (_("scale factor of %d without an index register"),
3251 1 << i.log2_scale_factor);
3252 #if SCALE1_WHEN_NO_INDEX
3253 i.log2_scale_factor = 0;
3254 #endif
3255 }
3256 return 1;
3257 }
3258
3259 static int i386_displacement PARAMS ((char *, char *));
3260
3261 static int
3262 i386_displacement (disp_start, disp_end)
3263 char *disp_start;
3264 char *disp_end;
3265 {
3266 register expressionS *exp;
3267 segT exp_seg = 0;
3268 char *save_input_line_pointer;
3269 int bigdisp = Disp32;
3270
3271 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3272 bigdisp = Disp16;
3273 if (flag_code == CODE_64BIT)
3274 bigdisp = Disp64;
3275 i.types[this_operand] |= bigdisp;
3276
3277 exp = &disp_expressions[i.disp_operands];
3278 i.op[this_operand].disps = exp;
3279 i.disp_operands++;
3280 save_input_line_pointer = input_line_pointer;
3281 input_line_pointer = disp_start;
3282 END_STRING_AND_SAVE (disp_end);
3283
3284 #ifndef GCC_ASM_O_HACK
3285 #define GCC_ASM_O_HACK 0
3286 #endif
3287 #if GCC_ASM_O_HACK
3288 END_STRING_AND_SAVE (disp_end + 1);
3289 if ((i.types[this_operand] & BaseIndex) != 0
3290 && displacement_string_end[-1] == '+')
3291 {
3292 /* This hack is to avoid a warning when using the "o"
3293 constraint within gcc asm statements.
3294 For instance:
3295
3296 #define _set_tssldt_desc(n,addr,limit,type) \
3297 __asm__ __volatile__ ( \
3298 "movw %w2,%0\n\t" \
3299 "movw %w1,2+%0\n\t" \
3300 "rorl $16,%1\n\t" \
3301 "movb %b1,4+%0\n\t" \
3302 "movb %4,5+%0\n\t" \
3303 "movb $0,6+%0\n\t" \
3304 "movb %h1,7+%0\n\t" \
3305 "rorl $16,%1" \
3306 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
3307
3308 This works great except that the output assembler ends
3309 up looking a bit weird if it turns out that there is
3310 no offset. You end up producing code that looks like:
3311
3312 #APP
3313 movw $235,(%eax)
3314 movw %dx,2+(%eax)
3315 rorl $16,%edx
3316 movb %dl,4+(%eax)
3317 movb $137,5+(%eax)
3318 movb $0,6+(%eax)
3319 movb %dh,7+(%eax)
3320 rorl $16,%edx
3321 #NO_APP
3322
3323 So here we provide the missing zero. */
3324
3325 *displacement_string_end = '0';
3326 }
3327 #endif
3328 #ifndef LEX_AT
3329 {
3330 /* We can have operands of the form
3331 <symbol>@GOTOFF+<nnn>
3332 Take the easy way out here and copy everything
3333 into a temporary buffer... */
3334 register char *cp;
3335
3336 cp = strchr (input_line_pointer, '@');
3337 if (cp != NULL)
3338 {
3339 char *tmpbuf;
3340 int len = 0;
3341 int first;
3342
3343 /* GOT relocations are not supported in 16 bit mode. */
3344 if (flag_code == CODE_16BIT)
3345 as_bad (_("GOT relocations not supported in 16 bit mode"));
3346
3347 if (GOT_symbol == NULL)
3348 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3349
3350 if (strncmp (cp + 1, "PLT", 3) == 0)
3351 {
3352 if (flag_code == CODE_64BIT)
3353 i.disp_reloc[this_operand] = BFD_RELOC_X86_64_PLT32;
3354 else
3355 i.disp_reloc[this_operand] = BFD_RELOC_386_PLT32;
3356 len = 3;
3357 }
3358 else if (strncmp (cp + 1, "GOTOFF", 6) == 0)
3359 {
3360 if (flag_code == CODE_64BIT)
3361 as_bad ("GOTOFF relocation is not supported in 64bit mode.");
3362 i.disp_reloc[this_operand] = BFD_RELOC_386_GOTOFF;
3363 len = 6;
3364 }
3365 else if (strncmp (cp + 1, "GOT", 3) == 0)
3366 {
3367 if (flag_code == CODE_64BIT)
3368 i.disp_reloc[this_operand] = BFD_RELOC_X86_64_GOT32;
3369 else
3370 i.disp_reloc[this_operand] = BFD_RELOC_386_GOT32;
3371 len = 3;
3372 }
3373 else if (strncmp (cp + 1, "GOTPCREL", 3) == 0)
3374 {
3375 if (flag_code != CODE_64BIT)
3376 as_bad ("GOTPCREL relocation is supported only in 64bit mode.");
3377 i.disp_reloc[this_operand] = BFD_RELOC_X86_64_GOTPCREL;
3378 len = 3;
3379 }
3380 else
3381 as_bad (_("bad reloc specifier in expression"));
3382
3383 /* Replace the relocation token with ' ', so that errors like
3384 foo@GOTOFF1 will be detected. */
3385 first = cp - input_line_pointer;
3386 tmpbuf = (char *) alloca (strlen (input_line_pointer));
3387 memcpy (tmpbuf, input_line_pointer, first);
3388 tmpbuf[first] = ' ';
3389 strcpy (tmpbuf + first + 1, cp + 1 + len);
3390 input_line_pointer = tmpbuf;
3391 }
3392 }
3393 #endif
3394
3395 exp_seg = expression (exp);
3396
3397 #ifdef BFD_ASSEMBLER
3398 /* We do this to make sure that the section symbol is in
3399 the symbol table. We will ultimately change the relocation
3400 to be relative to the beginning of the section. */
3401 if (i.disp_reloc[this_operand] == BFD_RELOC_386_GOTOFF
3402 || i.disp_reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3403 {
3404 if (S_IS_LOCAL(exp->X_add_symbol)
3405 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
3406 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
3407 assert (exp->X_op == O_symbol);
3408 exp->X_op = O_subtract;
3409 exp->X_op_symbol = GOT_symbol;
3410 i.disp_reloc[this_operand] = BFD_RELOC_32;
3411 }
3412 #endif
3413
3414 SKIP_WHITESPACE ();
3415 if (*input_line_pointer)
3416 as_bad (_("ignoring junk `%s' after expression"),
3417 input_line_pointer);
3418 #if GCC_ASM_O_HACK
3419 RESTORE_END_STRING (disp_end + 1);
3420 #endif
3421 RESTORE_END_STRING (disp_end);
3422 input_line_pointer = save_input_line_pointer;
3423
3424 if (exp->X_op == O_absent || exp->X_op == O_big)
3425 {
3426 /* Missing or bad expr becomes absolute 0. */
3427 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
3428 disp_start);
3429 exp->X_op = O_constant;
3430 exp->X_add_number = 0;
3431 exp->X_add_symbol = (symbolS *) 0;
3432 exp->X_op_symbol = (symbolS *) 0;
3433 }
3434
3435 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3436 if (exp->X_op != O_constant
3437 #ifdef BFD_ASSEMBLER
3438 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3439 #endif
3440 && exp_seg != text_section
3441 && exp_seg != data_section
3442 && exp_seg != bss_section
3443 && exp_seg != undefined_section)
3444 {
3445 #ifdef BFD_ASSEMBLER
3446 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3447 #else
3448 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3449 #endif
3450 return 0;
3451 }
3452 #endif
3453 else if (flag_code == CODE_64BIT)
3454 i.types[this_operand] |= Disp32S | Disp32;
3455 return 1;
3456 }
3457
3458 static int i386_index_check PARAMS((const char *));
3459
3460 /* Make sure the memory operand we've been dealt is valid.
3461 Return 1 on success, 0 on a failure. */
3462
3463 static int
3464 i386_index_check (operand_string)
3465 const char *operand_string;
3466 {
3467 int ok;
3468 #if INFER_ADDR_PREFIX
3469 int fudged = 0;
3470
3471 tryprefix:
3472 #endif
3473 ok = 1;
3474 if (flag_code == CODE_64BIT)
3475 {
3476 /* 64bit checks. */
3477 if ((i.base_reg
3478 && ((i.base_reg->reg_type & Reg64) == 0)
3479 && (i.base_reg->reg_type != BaseIndex
3480 || i.index_reg))
3481 || (i.index_reg
3482 && ((i.index_reg->reg_type & (Reg64|BaseIndex))
3483 != (Reg64|BaseIndex))))
3484 ok = 0;
3485 }
3486 else
3487 {
3488 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3489 {
3490 /* 16bit checks. */
3491 if ((i.base_reg
3492 && ((i.base_reg->reg_type & (Reg16|BaseIndex|RegRex))
3493 != (Reg16|BaseIndex)))
3494 || (i.index_reg
3495 && (((i.index_reg->reg_type & (Reg16|BaseIndex))
3496 != (Reg16|BaseIndex))
3497 || ! (i.base_reg
3498 && i.base_reg->reg_num < 6
3499 && i.index_reg->reg_num >= 6
3500 && i.log2_scale_factor == 0))))
3501 ok = 0;
3502 }
3503 else
3504 {
3505 /* 32bit checks. */
3506 if ((i.base_reg
3507 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
3508 || (i.index_reg
3509 && ((i.index_reg->reg_type & (Reg32|BaseIndex|RegRex))
3510 != (Reg32|BaseIndex))))
3511 ok = 0;
3512 }
3513 }
3514 if (!ok)
3515 {
3516 #if INFER_ADDR_PREFIX
3517 if (flag_code != CODE_64BIT
3518 && i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0')
3519 {
3520 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
3521 i.prefixes += 1;
3522 /* Change the size of any displacement too. At most one of
3523 Disp16 or Disp32 is set.
3524 FIXME. There doesn't seem to be any real need for separate
3525 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
3526 Removing them would probably clean up the code quite a lot. */
3527 if (i.types[this_operand] & (Disp16|Disp32))
3528 i.types[this_operand] ^= (Disp16|Disp32);
3529 fudged = 1;
3530 goto tryprefix;
3531 }
3532 if (fudged)
3533 as_bad (_("`%s' is not a valid base/index expression"),
3534 operand_string);
3535 else
3536 #endif
3537 as_bad (_("`%s' is not a valid %s bit base/index expression"),
3538 operand_string,
3539 flag_code_names[flag_code]);
3540 return 0;
3541 }
3542 return 1;
3543 }
3544
3545 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
3546 on error. */
3547
3548 static int
3549 i386_operand (operand_string)
3550 char *operand_string;
3551 {
3552 const reg_entry *r;
3553 char *end_op;
3554 char *op_string = operand_string;
3555
3556 if (is_space_char (*op_string))
3557 ++op_string;
3558
3559 /* We check for an absolute prefix (differentiating,
3560 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
3561 if (*op_string == ABSOLUTE_PREFIX)
3562 {
3563 ++op_string;
3564 if (is_space_char (*op_string))
3565 ++op_string;
3566 i.types[this_operand] |= JumpAbsolute;
3567 }
3568
3569 /* Check if operand is a register. */
3570 if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
3571 && (r = parse_register (op_string, &end_op)) != NULL)
3572 {
3573 /* Check for a segment override by searching for ':' after a
3574 segment register. */
3575 op_string = end_op;
3576 if (is_space_char (*op_string))
3577 ++op_string;
3578 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
3579 {
3580 switch (r->reg_num)
3581 {
3582 case 0:
3583 i.seg[i.mem_operands] = &es;
3584 break;
3585 case 1:
3586 i.seg[i.mem_operands] = &cs;
3587 break;
3588 case 2:
3589 i.seg[i.mem_operands] = &ss;
3590 break;
3591 case 3:
3592 i.seg[i.mem_operands] = &ds;
3593 break;
3594 case 4:
3595 i.seg[i.mem_operands] = &fs;
3596 break;
3597 case 5:
3598 i.seg[i.mem_operands] = &gs;
3599 break;
3600 }
3601
3602 /* Skip the ':' and whitespace. */
3603 ++op_string;
3604 if (is_space_char (*op_string))
3605 ++op_string;
3606
3607 if (!is_digit_char (*op_string)
3608 && !is_identifier_char (*op_string)
3609 && *op_string != '('
3610 && *op_string != ABSOLUTE_PREFIX)
3611 {
3612 as_bad (_("bad memory operand `%s'"), op_string);
3613 return 0;
3614 }
3615 /* Handle case of %es:*foo. */
3616 if (*op_string == ABSOLUTE_PREFIX)
3617 {
3618 ++op_string;
3619 if (is_space_char (*op_string))
3620 ++op_string;
3621 i.types[this_operand] |= JumpAbsolute;
3622 }
3623 goto do_memory_reference;
3624 }
3625 if (*op_string)
3626 {
3627 as_bad (_("junk `%s' after register"), op_string);
3628 return 0;
3629 }
3630 i.types[this_operand] |= r->reg_type & ~BaseIndex;
3631 i.op[this_operand].regs = r;
3632 i.reg_operands++;
3633 }
3634 else if (*op_string == REGISTER_PREFIX)
3635 {
3636 as_bad (_("bad register name `%s'"), op_string);
3637 return 0;
3638 }
3639 else if (*op_string == IMMEDIATE_PREFIX)
3640 {
3641 ++op_string;
3642 if (i.types[this_operand] & JumpAbsolute)
3643 {
3644 as_bad (_("immediate operand illegal with absolute jump"));
3645 return 0;
3646 }
3647 if (!i386_immediate (op_string))
3648 return 0;
3649 }
3650 else if (is_digit_char (*op_string)
3651 || is_identifier_char (*op_string)
3652 || *op_string == '(' )
3653 {
3654 /* This is a memory reference of some sort. */
3655 char *base_string;
3656
3657 /* Start and end of displacement string expression (if found). */
3658 char *displacement_string_start;
3659 char *displacement_string_end;
3660
3661 do_memory_reference:
3662 if ((i.mem_operands == 1
3663 && (current_templates->start->opcode_modifier & IsString) == 0)
3664 || i.mem_operands == 2)
3665 {
3666 as_bad (_("too many memory references for `%s'"),
3667 current_templates->start->name);
3668 return 0;
3669 }
3670
3671 /* Check for base index form. We detect the base index form by
3672 looking for an ')' at the end of the operand, searching
3673 for the '(' matching it, and finding a REGISTER_PREFIX or ','
3674 after the '('. */
3675 base_string = op_string + strlen (op_string);
3676
3677 --base_string;
3678 if (is_space_char (*base_string))
3679 --base_string;
3680
3681 /* If we only have a displacement, set-up for it to be parsed later. */
3682 displacement_string_start = op_string;
3683 displacement_string_end = base_string + 1;
3684
3685 if (*base_string == ')')
3686 {
3687 char *temp_string;
3688 unsigned int parens_balanced = 1;
3689 /* We've already checked that the number of left & right ()'s are
3690 equal, so this loop will not be infinite. */
3691 do
3692 {
3693 base_string--;
3694 if (*base_string == ')')
3695 parens_balanced++;
3696 if (*base_string == '(')
3697 parens_balanced--;
3698 }
3699 while (parens_balanced);
3700
3701 temp_string = base_string;
3702
3703 /* Skip past '(' and whitespace. */
3704 ++base_string;
3705 if (is_space_char (*base_string))
3706 ++base_string;
3707
3708 if (*base_string == ','
3709 || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3710 && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
3711 {
3712 displacement_string_end = temp_string;
3713
3714 i.types[this_operand] |= BaseIndex;
3715
3716 if (i.base_reg)
3717 {
3718 base_string = end_op;
3719 if (is_space_char (*base_string))
3720 ++base_string;
3721 }
3722
3723 /* There may be an index reg or scale factor here. */
3724 if (*base_string == ',')
3725 {
3726 ++base_string;
3727 if (is_space_char (*base_string))
3728 ++base_string;
3729
3730 if ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3731 && (i.index_reg = parse_register (base_string, &end_op)) != NULL)
3732 {
3733 base_string = end_op;
3734 if (is_space_char (*base_string))
3735 ++base_string;
3736 if (*base_string == ',')
3737 {
3738 ++base_string;
3739 if (is_space_char (*base_string))
3740 ++base_string;
3741 }
3742 else if (*base_string != ')' )
3743 {
3744 as_bad (_("expecting `,' or `)' after index register in `%s'"),
3745 operand_string);
3746 return 0;
3747 }
3748 }
3749 else if (*base_string == REGISTER_PREFIX)
3750 {
3751 as_bad (_("bad register name `%s'"), base_string);
3752 return 0;
3753 }
3754
3755 /* Check for scale factor. */
3756 if (isdigit ((unsigned char) *base_string))
3757 {
3758 if (!i386_scale (base_string))
3759 return 0;
3760
3761 ++base_string;
3762 if (is_space_char (*base_string))
3763 ++base_string;
3764 if (*base_string != ')')
3765 {
3766 as_bad (_("expecting `)' after scale factor in `%s'"),
3767 operand_string);
3768 return 0;
3769 }
3770 }
3771 else if (!i.index_reg)
3772 {
3773 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
3774 *base_string);
3775 return 0;
3776 }
3777 }
3778 else if (*base_string != ')')
3779 {
3780 as_bad (_("expecting `,' or `)' after base register in `%s'"),
3781 operand_string);
3782 return 0;
3783 }
3784 }
3785 else if (*base_string == REGISTER_PREFIX)
3786 {
3787 as_bad (_("bad register name `%s'"), base_string);
3788 return 0;
3789 }
3790 }
3791
3792 /* If there's an expression beginning the operand, parse it,
3793 assuming displacement_string_start and
3794 displacement_string_end are meaningful. */
3795 if (displacement_string_start != displacement_string_end)
3796 {
3797 if (!i386_displacement (displacement_string_start,
3798 displacement_string_end))
3799 return 0;
3800 }
3801
3802 /* Special case for (%dx) while doing input/output op. */
3803 if (i.base_reg
3804 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
3805 && i.index_reg == 0
3806 && i.log2_scale_factor == 0
3807 && i.seg[i.mem_operands] == 0
3808 && (i.types[this_operand] & Disp) == 0)
3809 {
3810 i.types[this_operand] = InOutPortReg;
3811 return 1;
3812 }
3813
3814 if (i386_index_check (operand_string) == 0)
3815 return 0;
3816 i.mem_operands++;
3817 }
3818 else
3819 {
3820 /* It's not a memory operand; argh! */
3821 as_bad (_("invalid char %s beginning operand %d `%s'"),
3822 output_invalid (*op_string),
3823 this_operand + 1,
3824 op_string);
3825 return 0;
3826 }
3827 return 1; /* Normal return. */
3828 }
3829 \f
3830 /* md_estimate_size_before_relax()
3831
3832 Called just before relax() for rs_machine_dependent frags. The x86
3833 assembler uses these frags to handle variable size jump
3834 instructions.
3835
3836 Any symbol that is now undefined will not become defined.
3837 Return the correct fr_subtype in the frag.
3838 Return the initial "guess for variable size of frag" to caller.
3839 The guess is actually the growth beyond the fixed part. Whatever
3840 we do to grow the fixed or variable part contributes to our
3841 returned value. */
3842
3843 int
3844 md_estimate_size_before_relax (fragP, segment)
3845 register fragS *fragP;
3846 register segT segment;
3847 {
3848 /* We've already got fragP->fr_subtype right; all we have to do is
3849 check for un-relaxable symbols. On an ELF system, we can't relax
3850 an externally visible symbol, because it may be overridden by a
3851 shared library. */
3852 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
3853 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
3854 || S_IS_EXTERNAL (fragP->fr_symbol)
3855 || S_IS_WEAK (fragP->fr_symbol)
3856 #endif
3857 )
3858 {
3859 /* Symbol is undefined in this segment, or we need to keep a
3860 reloc so that weak symbols can be overridden. */
3861 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
3862 #ifdef BFD_ASSEMBLER
3863 enum bfd_reloc_code_real reloc_type;
3864 #else
3865 int reloc_type;
3866 #endif
3867 unsigned char *opcode;
3868 int old_fr_fix;
3869
3870 if (fragP->fr_var != NO_RELOC)
3871 reloc_type = fragP->fr_var;
3872 else if (size == 2)
3873 reloc_type = BFD_RELOC_16_PCREL;
3874 else
3875 reloc_type = BFD_RELOC_32_PCREL;
3876
3877 old_fr_fix = fragP->fr_fix;
3878 opcode = (unsigned char *) fragP->fr_opcode;
3879
3880 switch (opcode[0])
3881 {
3882 case JUMP_PC_RELATIVE:
3883 /* Make jmp (0xeb) a dword displacement jump. */
3884 opcode[0] = 0xe9;
3885 fragP->fr_fix += size;
3886 fix_new (fragP, old_fr_fix, size,
3887 fragP->fr_symbol,
3888 fragP->fr_offset, 1,
3889 reloc_type);
3890 break;
3891
3892 default:
3893 /* This changes the byte-displacement jump 0x7N
3894 to the dword-displacement jump 0x0f,0x8N. */
3895 opcode[1] = opcode[0] + 0x10;
3896 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
3897 /* We've added an opcode byte. */
3898 fragP->fr_fix += 1 + size;
3899 fix_new (fragP, old_fr_fix + 1, size,
3900 fragP->fr_symbol,
3901 fragP->fr_offset, 1,
3902 reloc_type);
3903 break;
3904 }
3905 frag_wane (fragP);
3906 return fragP->fr_fix - old_fr_fix;
3907 }
3908 /* Guess a short jump. */
3909 return 1;
3910 }
3911
3912 /* Called after relax() is finished.
3913
3914 In: Address of frag.
3915 fr_type == rs_machine_dependent.
3916 fr_subtype is what the address relaxed to.
3917
3918 Out: Any fixSs and constants are set up.
3919 Caller will turn frag into a ".space 0". */
3920
3921 #ifndef BFD_ASSEMBLER
3922 void
3923 md_convert_frag (headers, sec, fragP)
3924 object_headers *headers ATTRIBUTE_UNUSED;
3925 segT sec ATTRIBUTE_UNUSED;
3926 register fragS *fragP;
3927 #else
3928 void
3929 md_convert_frag (abfd, sec, fragP)
3930 bfd *abfd ATTRIBUTE_UNUSED;
3931 segT sec ATTRIBUTE_UNUSED;
3932 register fragS *fragP;
3933 #endif
3934 {
3935 register unsigned char *opcode;
3936 unsigned char *where_to_put_displacement = NULL;
3937 offsetT target_address;
3938 offsetT opcode_address;
3939 unsigned int extension = 0;
3940 offsetT displacement_from_opcode_start;
3941
3942 opcode = (unsigned char *) fragP->fr_opcode;
3943
3944 /* Address we want to reach in file space. */
3945 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
3946 #ifdef BFD_ASSEMBLER
3947 /* Not needed otherwise? */
3948 target_address += symbol_get_frag (fragP->fr_symbol)->fr_address;
3949 #endif
3950
3951 /* Address opcode resides at in file space. */
3952 opcode_address = fragP->fr_address + fragP->fr_fix;
3953
3954 /* Displacement from opcode start to fill into instruction. */
3955 displacement_from_opcode_start = target_address - opcode_address;
3956
3957 switch (fragP->fr_subtype)
3958 {
3959 case ENCODE_RELAX_STATE (COND_JUMP, SMALL):
3960 case ENCODE_RELAX_STATE (COND_JUMP, SMALL16):
3961 case ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL):
3962 case ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL16):
3963 /* Don't have to change opcode. */
3964 extension = 1; /* 1 opcode + 1 displacement */
3965 where_to_put_displacement = &opcode[1];
3966 break;
3967
3968 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
3969 extension = 5; /* 2 opcode + 4 displacement */
3970 opcode[1] = opcode[0] + 0x10;
3971 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
3972 where_to_put_displacement = &opcode[2];
3973 break;
3974
3975 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
3976 extension = 4; /* 1 opcode + 4 displacement */
3977 opcode[0] = 0xe9;
3978 where_to_put_displacement = &opcode[1];
3979 break;
3980
3981 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
3982 extension = 3; /* 2 opcode + 2 displacement */
3983 opcode[1] = opcode[0] + 0x10;
3984 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
3985 where_to_put_displacement = &opcode[2];
3986 break;
3987
3988 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
3989 extension = 2; /* 1 opcode + 2 displacement */
3990 opcode[0] = 0xe9;
3991 where_to_put_displacement = &opcode[1];
3992 break;
3993
3994 default:
3995 BAD_CASE (fragP->fr_subtype);
3996 break;
3997 }
3998 /* Now put displacement after opcode. */
3999 md_number_to_chars ((char *) where_to_put_displacement,
4000 (valueT) (displacement_from_opcode_start - extension),
4001 SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
4002 fragP->fr_fix += extension;
4003 }
4004 \f
4005 /* Size of byte displacement jmp. */
4006 int md_short_jump_size = 2;
4007
4008 /* Size of dword displacement jmp. */
4009 int md_long_jump_size = 5;
4010
4011 /* Size of relocation record. */
4012 const int md_reloc_size = 8;
4013
4014 void
4015 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
4016 char *ptr;
4017 addressT from_addr, to_addr;
4018 fragS *frag ATTRIBUTE_UNUSED;
4019 symbolS *to_symbol ATTRIBUTE_UNUSED;
4020 {
4021 offsetT offset;
4022
4023 offset = to_addr - (from_addr + 2);
4024 /* Opcode for byte-disp jump. */
4025 md_number_to_chars (ptr, (valueT) 0xeb, 1);
4026 md_number_to_chars (ptr + 1, (valueT) offset, 1);
4027 }
4028
4029 void
4030 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
4031 char *ptr;
4032 addressT from_addr, to_addr;
4033 fragS *frag ATTRIBUTE_UNUSED;
4034 symbolS *to_symbol ATTRIBUTE_UNUSED;
4035 {
4036 offsetT offset;
4037
4038 offset = to_addr - (from_addr + 5);
4039 md_number_to_chars (ptr, (valueT) 0xe9, 1);
4040 md_number_to_chars (ptr + 1, (valueT) offset, 4);
4041 }
4042 \f
4043 /* Apply a fixup (fixS) to segment data, once it has been determined
4044 by our caller that we have all the info we need to fix it up.
4045
4046 On the 386, immediates, displacements, and data pointers are all in
4047 the same (little-endian) format, so we don't need to care about which
4048 we are handling. */
4049
4050 int
4051 md_apply_fix3 (fixP, valp, seg)
4052 /* The fix we're to put in. */
4053 fixS *fixP;
4054
4055 /* Pointer to the value of the bits. */
4056 valueT *valp;
4057
4058 /* Segment fix is from. */
4059 segT seg ATTRIBUTE_UNUSED;
4060 {
4061 register char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
4062 valueT value = *valp;
4063
4064 #if defined (BFD_ASSEMBLER) && !defined (TE_Mach)
4065 if (fixP->fx_pcrel)
4066 {
4067 switch (fixP->fx_r_type)
4068 {
4069 default:
4070 break;
4071
4072 case BFD_RELOC_32:
4073 fixP->fx_r_type = BFD_RELOC_32_PCREL;
4074 break;
4075 case BFD_RELOC_16:
4076 fixP->fx_r_type = BFD_RELOC_16_PCREL;
4077 break;
4078 case BFD_RELOC_8:
4079 fixP->fx_r_type = BFD_RELOC_8_PCREL;
4080 break;
4081 }
4082 }
4083
4084 /* This is a hack. There should be a better way to handle this.
4085 This covers for the fact that bfd_install_relocation will
4086 subtract the current location (for partial_inplace, PC relative
4087 relocations); see more below. */
4088 if ((fixP->fx_r_type == BFD_RELOC_32_PCREL
4089 || fixP->fx_r_type == BFD_RELOC_16_PCREL
4090 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
4091 && fixP->fx_addsy)
4092 {
4093 #ifndef OBJ_AOUT
4094 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4095 #ifdef TE_PE
4096 || OUTPUT_FLAVOR == bfd_target_coff_flavour
4097 #endif
4098 )
4099 value += fixP->fx_where + fixP->fx_frag->fr_address;
4100 #endif
4101 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4102 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
4103 {
4104 segT fseg = S_GET_SEGMENT (fixP->fx_addsy);
4105
4106 if ((fseg == seg
4107 || (symbol_section_p (fixP->fx_addsy)
4108 && fseg != absolute_section))
4109 && ! S_IS_EXTERNAL (fixP->fx_addsy)
4110 && ! S_IS_WEAK (fixP->fx_addsy)
4111 && S_IS_DEFINED (fixP->fx_addsy)
4112 && ! S_IS_COMMON (fixP->fx_addsy))
4113 {
4114 /* Yes, we add the values in twice. This is because
4115 bfd_perform_relocation subtracts them out again. I think
4116 bfd_perform_relocation is broken, but I don't dare change
4117 it. FIXME. */
4118 value += fixP->fx_where + fixP->fx_frag->fr_address;
4119 }
4120 }
4121 #endif
4122 #if defined (OBJ_COFF) && defined (TE_PE)
4123 /* For some reason, the PE format does not store a section
4124 address offset for a PC relative symbol. */
4125 if (S_GET_SEGMENT (fixP->fx_addsy) != seg)
4126 value += md_pcrel_from (fixP);
4127 #endif
4128 }
4129
4130 /* Fix a few things - the dynamic linker expects certain values here,
4131 and we must not dissappoint it. */
4132 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4133 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4134 && fixP->fx_addsy)
4135 switch (fixP->fx_r_type)
4136 {
4137 case BFD_RELOC_386_PLT32:
4138 case BFD_RELOC_X86_64_PLT32:
4139 /* Make the jump instruction point to the address of the operand. At
4140 runtime we merely add the offset to the actual PLT entry. */
4141 value = -4;
4142 break;
4143 case BFD_RELOC_386_GOTPC:
4144
4145 /* This is tough to explain. We end up with this one if we have
4146 * operands that look like "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal
4147 * here is to obtain the absolute address of the GOT, and it is strongly
4148 * preferable from a performance point of view to avoid using a runtime
4149 * relocation for this. The actual sequence of instructions often look
4150 * something like:
4151 *
4152 * call .L66
4153 * .L66:
4154 * popl %ebx
4155 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
4156 *
4157 * The call and pop essentially return the absolute address of
4158 * the label .L66 and store it in %ebx. The linker itself will
4159 * ultimately change the first operand of the addl so that %ebx points to
4160 * the GOT, but to keep things simple, the .o file must have this operand
4161 * set so that it generates not the absolute address of .L66, but the
4162 * absolute address of itself. This allows the linker itself simply
4163 * treat a GOTPC relocation as asking for a pcrel offset to the GOT to be
4164 * added in, and the addend of the relocation is stored in the operand
4165 * field for the instruction itself.
4166 *
4167 * Our job here is to fix the operand so that it would add the correct
4168 * offset so that %ebx would point to itself. The thing that is tricky is
4169 * that .-.L66 will point to the beginning of the instruction, so we need
4170 * to further modify the operand so that it will point to itself.
4171 * There are other cases where you have something like:
4172 *
4173 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
4174 *
4175 * and here no correction would be required. Internally in the assembler
4176 * we treat operands of this form as not being pcrel since the '.' is
4177 * explicitly mentioned, and I wonder whether it would simplify matters
4178 * to do it this way. Who knows. In earlier versions of the PIC patches,
4179 * the pcrel_adjust field was used to store the correction, but since the
4180 * expression is not pcrel, I felt it would be confusing to do it this
4181 * way. */
4182
4183 value -= 1;
4184 break;
4185 case BFD_RELOC_386_GOT32:
4186 case BFD_RELOC_X86_64_GOT32:
4187 value = 0; /* Fully resolved at runtime. No addend. */
4188 break;
4189 case BFD_RELOC_386_GOTOFF:
4190 case BFD_RELOC_X86_64_GOTPCREL:
4191 break;
4192
4193 case BFD_RELOC_VTABLE_INHERIT:
4194 case BFD_RELOC_VTABLE_ENTRY:
4195 fixP->fx_done = 0;
4196 return 1;
4197
4198 default:
4199 break;
4200 }
4201 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
4202 *valp = value;
4203 #endif /* defined (BFD_ASSEMBLER) && !defined (TE_Mach) */
4204
4205 #ifndef BFD_ASSEMBLER
4206 md_number_to_chars (p, value, fixP->fx_size);
4207 #else
4208 /* Are we finished with this relocation now? */
4209 if (fixP->fx_addsy == 0 && fixP->fx_pcrel == 0)
4210 fixP->fx_done = 1;
4211 else if (use_rela_relocations)
4212 {
4213 fixP->fx_no_overflow = 1;
4214 value = 0;
4215 }
4216 md_number_to_chars (p, value, fixP->fx_size);
4217 #endif
4218
4219 return 1;
4220 }
4221 \f
4222 #define MAX_LITTLENUMS 6
4223
4224 /* Turn the string pointed to by litP into a floating point constant
4225 of type TYPE, and emit the appropriate bytes. The number of
4226 LITTLENUMS emitted is stored in *SIZEP. An error message is
4227 returned, or NULL on OK. */
4228
4229 char *
4230 md_atof (type, litP, sizeP)
4231 int type;
4232 char *litP;
4233 int *sizeP;
4234 {
4235 int prec;
4236 LITTLENUM_TYPE words[MAX_LITTLENUMS];
4237 LITTLENUM_TYPE *wordP;
4238 char *t;
4239
4240 switch (type)
4241 {
4242 case 'f':
4243 case 'F':
4244 prec = 2;
4245 break;
4246
4247 case 'd':
4248 case 'D':
4249 prec = 4;
4250 break;
4251
4252 case 'x':
4253 case 'X':
4254 prec = 5;
4255 break;
4256
4257 default:
4258 *sizeP = 0;
4259 return _("Bad call to md_atof ()");
4260 }
4261 t = atof_ieee (input_line_pointer, type, words);
4262 if (t)
4263 input_line_pointer = t;
4264
4265 *sizeP = prec * sizeof (LITTLENUM_TYPE);
4266 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
4267 the bigendian 386. */
4268 for (wordP = words + prec - 1; prec--;)
4269 {
4270 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
4271 litP += sizeof (LITTLENUM_TYPE);
4272 }
4273 return 0;
4274 }
4275 \f
4276 char output_invalid_buf[8];
4277
4278 static char *
4279 output_invalid (c)
4280 int c;
4281 {
4282 if (isprint (c))
4283 sprintf (output_invalid_buf, "'%c'", c);
4284 else
4285 sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
4286 return output_invalid_buf;
4287 }
4288
4289 /* REG_STRING starts *before* REGISTER_PREFIX. */
4290
4291 static const reg_entry *
4292 parse_register (reg_string, end_op)
4293 char *reg_string;
4294 char **end_op;
4295 {
4296 char *s = reg_string;
4297 char *p;
4298 char reg_name_given[MAX_REG_NAME_SIZE + 1];
4299 const reg_entry *r;
4300
4301 /* Skip possible REGISTER_PREFIX and possible whitespace. */
4302 if (*s == REGISTER_PREFIX)
4303 ++s;
4304
4305 if (is_space_char (*s))
4306 ++s;
4307
4308 p = reg_name_given;
4309 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
4310 {
4311 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
4312 return (const reg_entry *) NULL;
4313 s++;
4314 }
4315
4316 /* For naked regs, make sure that we are not dealing with an identifier.
4317 This prevents confusing an identifier like `eax_var' with register
4318 `eax'. */
4319 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
4320 return (const reg_entry *) NULL;
4321
4322 *end_op = s;
4323
4324 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
4325
4326 /* Handle floating point regs, allowing spaces in the (i) part. */
4327 if (r == i386_regtab /* %st is first entry of table */)
4328 {
4329 if (is_space_char (*s))
4330 ++s;
4331 if (*s == '(')
4332 {
4333 ++s;
4334 if (is_space_char (*s))
4335 ++s;
4336 if (*s >= '0' && *s <= '7')
4337 {
4338 r = &i386_float_regtab[*s - '0'];
4339 ++s;
4340 if (is_space_char (*s))
4341 ++s;
4342 if (*s == ')')
4343 {
4344 *end_op = s + 1;
4345 return r;
4346 }
4347 }
4348 /* We have "%st(" then garbage. */
4349 return (const reg_entry *) NULL;
4350 }
4351 }
4352
4353 return r;
4354 }
4355 \f
4356 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4357 const char *md_shortopts = "kVQ:sq";
4358 #else
4359 const char *md_shortopts = "q";
4360 #endif
4361 struct option md_longopts[] = {
4362 #define OPTION_32 (OPTION_MD_BASE + 0)
4363 {"32", no_argument, NULL, OPTION_32},
4364 #define OPTION_64 (OPTION_MD_BASE + 1)
4365 {"64", no_argument, NULL, OPTION_64},
4366 {NULL, no_argument, NULL, 0}
4367 };
4368 size_t md_longopts_size = sizeof (md_longopts);
4369
4370 int
4371 md_parse_option (c, arg)
4372 int c;
4373 char *arg ATTRIBUTE_UNUSED;
4374 {
4375 switch (c)
4376 {
4377 case 'q':
4378 quiet_warnings = 1;
4379 break;
4380
4381 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4382 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
4383 should be emitted or not. FIXME: Not implemented. */
4384 case 'Q':
4385 break;
4386
4387 /* -V: SVR4 argument to print version ID. */
4388 case 'V':
4389 print_version_id ();
4390 break;
4391
4392 /* -k: Ignore for FreeBSD compatibility. */
4393 case 'k':
4394 break;
4395
4396 case 's':
4397 /* -s: On i386 Solaris, this tells the native assembler to use
4398 .stab instead of .stab.excl. We always use .stab anyhow. */
4399 break;
4400 #endif
4401 #ifdef OBJ_ELF
4402 case OPTION_32:
4403 case OPTION_64:
4404 {
4405 const char **list, **l;
4406
4407 default_arch = c == OPTION_32 ? "i386" : "x86_64";
4408 list = bfd_target_list ();
4409 for (l = list; *l != NULL; l++)
4410 {
4411 if (c == OPTION_32)
4412 {
4413 if (strcmp (*l, "elf32-i386") == 0)
4414 break;
4415 }
4416 else
4417 {
4418 if (strcmp (*l, "elf64-x86-64") == 0)
4419 break;
4420 }
4421 }
4422 if (*l == NULL)
4423 as_fatal (_("No compiled in support for %d bit object file format"),
4424 c == OPTION_32 ? 32 : 64);
4425 free (list);
4426 }
4427 break;
4428 #endif
4429
4430 default:
4431 return 0;
4432 }
4433 return 1;
4434 }
4435
4436 void
4437 md_show_usage (stream)
4438 FILE *stream;
4439 {
4440 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4441 fprintf (stream, _("\
4442 -Q ignored\n\
4443 -V print assembler version number\n\
4444 -k ignored\n\
4445 -q quieten some warnings\n\
4446 -s ignored\n"));
4447 #else
4448 fprintf (stream, _("\
4449 -q quieten some warnings\n"));
4450 #endif
4451 }
4452
4453 #ifdef BFD_ASSEMBLER
4454 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
4455 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
4456
4457 /* Pick the target format to use. */
4458
4459 const char *
4460 i386_target_format ()
4461 {
4462 if (!strcmp (default_arch, "x86_64"))
4463 set_code_flag (CODE_64BIT);
4464 else if (!strcmp (default_arch, "i386"))
4465 set_code_flag (CODE_32BIT);
4466 else
4467 as_fatal (_("Unknown architecture"));
4468 switch (OUTPUT_FLAVOR)
4469 {
4470 #ifdef OBJ_MAYBE_AOUT
4471 case bfd_target_aout_flavour:
4472 return AOUT_TARGET_FORMAT;
4473 #endif
4474 #ifdef OBJ_MAYBE_COFF
4475 case bfd_target_coff_flavour:
4476 return "coff-i386";
4477 #endif
4478 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
4479 case bfd_target_elf_flavour:
4480 {
4481 if (flag_code == CODE_64BIT)
4482 use_rela_relocations = 1;
4483 return flag_code == CODE_64BIT ? "elf64-x86-64" : "elf32-i386";
4484 }
4485 #endif
4486 default:
4487 abort ();
4488 return NULL;
4489 }
4490 }
4491
4492 #endif /* OBJ_MAYBE_ more than one */
4493 #endif /* BFD_ASSEMBLER */
4494 \f
4495 symbolS *
4496 md_undefined_symbol (name)
4497 char *name;
4498 {
4499 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
4500 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
4501 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
4502 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
4503 {
4504 if (!GOT_symbol)
4505 {
4506 if (symbol_find (name))
4507 as_bad (_("GOT already in symbol table"));
4508 GOT_symbol = symbol_new (name, undefined_section,
4509 (valueT) 0, &zero_address_frag);
4510 };
4511 return GOT_symbol;
4512 }
4513 return 0;
4514 }
4515
4516 /* Round up a section size to the appropriate boundary. */
4517
4518 valueT
4519 md_section_align (segment, size)
4520 segT segment ATTRIBUTE_UNUSED;
4521 valueT size;
4522 {
4523 #ifdef BFD_ASSEMBLER
4524 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4525 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
4526 {
4527 /* For a.out, force the section size to be aligned. If we don't do
4528 this, BFD will align it for us, but it will not write out the
4529 final bytes of the section. This may be a bug in BFD, but it is
4530 easier to fix it here since that is how the other a.out targets
4531 work. */
4532 int align;
4533
4534 align = bfd_get_section_alignment (stdoutput, segment);
4535 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
4536 }
4537 #endif
4538 #endif
4539
4540 return size;
4541 }
4542
4543 /* On the i386, PC-relative offsets are relative to the start of the
4544 next instruction. That is, the address of the offset, plus its
4545 size, since the offset is always the last part of the insn. */
4546
4547 long
4548 md_pcrel_from (fixP)
4549 fixS *fixP;
4550 {
4551 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
4552 }
4553
4554 #ifndef I386COFF
4555
4556 static void
4557 s_bss (ignore)
4558 int ignore ATTRIBUTE_UNUSED;
4559 {
4560 register int temp;
4561
4562 temp = get_absolute_expression ();
4563 subseg_set (bss_section, (subsegT) temp);
4564 demand_empty_rest_of_line ();
4565 }
4566
4567 #endif
4568
4569 #ifdef BFD_ASSEMBLER
4570
4571 void
4572 i386_validate_fix (fixp)
4573 fixS *fixp;
4574 {
4575 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
4576 {
4577 /* GOTOFF relocation are nonsense in 64bit mode. */
4578 if (flag_code == CODE_64BIT)
4579 abort();
4580 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
4581 fixp->fx_subsy = 0;
4582 }
4583 }
4584
4585 arelent *
4586 tc_gen_reloc (section, fixp)
4587 asection *section ATTRIBUTE_UNUSED;
4588 fixS *fixp;
4589 {
4590 arelent *rel;
4591 bfd_reloc_code_real_type code;
4592
4593 switch (fixp->fx_r_type)
4594 {
4595 case BFD_RELOC_X86_64_PLT32:
4596 case BFD_RELOC_X86_64_GOT32:
4597 case BFD_RELOC_X86_64_GOTPCREL:
4598 case BFD_RELOC_386_PLT32:
4599 case BFD_RELOC_386_GOT32:
4600 case BFD_RELOC_386_GOTOFF:
4601 case BFD_RELOC_386_GOTPC:
4602 case BFD_RELOC_X86_64_32S:
4603 case BFD_RELOC_RVA:
4604 case BFD_RELOC_VTABLE_ENTRY:
4605 case BFD_RELOC_VTABLE_INHERIT:
4606 code = fixp->fx_r_type;
4607 break;
4608 default:
4609 if (fixp->fx_pcrel)
4610 {
4611 switch (fixp->fx_size)
4612 {
4613 default:
4614 as_bad (_("can not do %d byte pc-relative relocation"),
4615 fixp->fx_size);
4616 code = BFD_RELOC_32_PCREL;
4617 break;
4618 case 1: code = BFD_RELOC_8_PCREL; break;
4619 case 2: code = BFD_RELOC_16_PCREL; break;
4620 case 4: code = BFD_RELOC_32_PCREL; break;
4621 }
4622 }
4623 else
4624 {
4625 switch (fixp->fx_size)
4626 {
4627 default:
4628 as_bad (_("can not do %d byte relocation"), fixp->fx_size);
4629 code = BFD_RELOC_32;
4630 break;
4631 case 1: code = BFD_RELOC_8; break;
4632 case 2: code = BFD_RELOC_16; break;
4633 case 4: code = BFD_RELOC_32; break;
4634 case 8: code = BFD_RELOC_64; break;
4635 }
4636 }
4637 break;
4638 }
4639
4640 if (code == BFD_RELOC_32
4641 && GOT_symbol
4642 && fixp->fx_addsy == GOT_symbol)
4643 {
4644 /* We don't support GOTPC on 64bit targets. */
4645 if (flag_code == CODE_64BIT)
4646 abort();
4647 code = BFD_RELOC_386_GOTPC;
4648 }
4649
4650 rel = (arelent *) xmalloc (sizeof (arelent));
4651 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
4652 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
4653
4654 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
4655 if (!use_rela_relocations)
4656 {
4657 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
4658 vtable entry to be used in the relocation's section offset. */
4659 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
4660 rel->address = fixp->fx_offset;
4661
4662 if (fixp->fx_pcrel)
4663 rel->addend = fixp->fx_addnumber;
4664 else
4665 rel->addend = 0;
4666 }
4667 /* Use the rela in 64bit mode. */
4668 else
4669 {
4670 rel->addend = fixp->fx_offset;
4671 #ifdef OBJ_ELF
4672 /* Ohhh, this is ugly. The problem is that if this is a local global
4673 symbol, the relocation will entirely be performed at link time, not
4674 at assembly time. bfd_perform_reloc doesn't know about this sort
4675 of thing, and as a result we need to fake it out here. */
4676 if ((S_IS_EXTERN (fixp->fx_addsy) || S_IS_WEAK (fixp->fx_addsy))
4677 && !S_IS_COMMON(fixp->fx_addsy))
4678 rel->addend -= symbol_get_bfdsym (fixp->fx_addsy)->value;
4679 #endif
4680 if (fixp->fx_pcrel)
4681 rel->addend -= fixp->fx_size;
4682 }
4683
4684
4685 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
4686 if (rel->howto == NULL)
4687 {
4688 as_bad_where (fixp->fx_file, fixp->fx_line,
4689 _("cannot represent relocation type %s"),
4690 bfd_get_reloc_code_name (code));
4691 /* Set howto to a garbage value so that we can keep going. */
4692 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
4693 assert (rel->howto != NULL);
4694 }
4695
4696 return rel;
4697 }
4698
4699 #else /* ! BFD_ASSEMBLER */
4700
4701 #if (defined(OBJ_AOUT) | defined(OBJ_BOUT))
4702 void
4703 tc_aout_fix_to_chars (where, fixP, segment_address_in_file)
4704 char *where;
4705 fixS *fixP;
4706 relax_addressT segment_address_in_file;
4707 {
4708 /* In: length of relocation (or of address) in chars: 1, 2 or 4.
4709 Out: GNU LD relocation length code: 0, 1, or 2. */
4710
4711 static const unsigned char nbytes_r_length[] = { 42, 0, 1, 42, 2 };
4712 long r_symbolnum;
4713
4714 know (fixP->fx_addsy != NULL);
4715
4716 md_number_to_chars (where,
4717 (valueT) (fixP->fx_frag->fr_address
4718 + fixP->fx_where - segment_address_in_file),
4719 4);
4720
4721 r_symbolnum = (S_IS_DEFINED (fixP->fx_addsy)
4722 ? S_GET_TYPE (fixP->fx_addsy)
4723 : fixP->fx_addsy->sy_number);
4724
4725 where[6] = (r_symbolnum >> 16) & 0x0ff;
4726 where[5] = (r_symbolnum >> 8) & 0x0ff;
4727 where[4] = r_symbolnum & 0x0ff;
4728 where[7] = ((((!S_IS_DEFINED (fixP->fx_addsy)) << 3) & 0x08)
4729 | ((nbytes_r_length[fixP->fx_size] << 1) & 0x06)
4730 | (((fixP->fx_pcrel << 0) & 0x01) & 0x0f));
4731 }
4732
4733 #endif /* OBJ_AOUT or OBJ_BOUT. */
4734
4735 #if defined (I386COFF)
4736
4737 short
4738 tc_coff_fix2rtype (fixP)
4739 fixS *fixP;
4740 {
4741 if (fixP->fx_r_type == R_IMAGEBASE)
4742 return R_IMAGEBASE;
4743
4744 return (fixP->fx_pcrel ?
4745 (fixP->fx_size == 1 ? R_PCRBYTE :
4746 fixP->fx_size == 2 ? R_PCRWORD :
4747 R_PCRLONG) :
4748 (fixP->fx_size == 1 ? R_RELBYTE :
4749 fixP->fx_size == 2 ? R_RELWORD :
4750 R_DIR32));
4751 }
4752
4753 int
4754 tc_coff_sizemachdep (frag)
4755 fragS *frag;
4756 {
4757 if (frag->fr_next)
4758 return (frag->fr_next->fr_address - frag->fr_address);
4759 else
4760 return 0;
4761 }
4762
4763 #endif /* I386COFF */
4764
4765 #endif /* ! BFD_ASSEMBLER */
4766 \f
4767 /* Parse operands using Intel syntax. This implements a recursive descent
4768 parser based on the BNF grammar published in Appendix B of the MASM 6.1
4769 Programmer's Guide.
4770
4771 FIXME: We do not recognize the full operand grammar defined in the MASM
4772 documentation. In particular, all the structure/union and
4773 high-level macro operands are missing.
4774
4775 Uppercase words are terminals, lower case words are non-terminals.
4776 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
4777 bars '|' denote choices. Most grammar productions are implemented in
4778 functions called 'intel_<production>'.
4779
4780 Initial production is 'expr'.
4781
4782 addOp + | -
4783
4784 alpha [a-zA-Z]
4785
4786 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
4787
4788 constant digits [[ radixOverride ]]
4789
4790 dataType BYTE | WORD | DWORD | QWORD | XWORD
4791
4792 digits decdigit
4793 | digits decdigit
4794 | digits hexdigit
4795
4796 decdigit [0-9]
4797
4798 e05 e05 addOp e06
4799 | e06
4800
4801 e06 e06 mulOp e09
4802 | e09
4803
4804 e09 OFFSET e10
4805 | e09 PTR e10
4806 | e09 : e10
4807 | e10
4808
4809 e10 e10 [ expr ]
4810 | e11
4811
4812 e11 ( expr )
4813 | [ expr ]
4814 | constant
4815 | dataType
4816 | id
4817 | $
4818 | register
4819
4820 => expr SHORT e05
4821 | e05
4822
4823 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
4824 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
4825
4826 hexdigit a | b | c | d | e | f
4827 | A | B | C | D | E | F
4828
4829 id alpha
4830 | id alpha
4831 | id decdigit
4832
4833 mulOp * | / | MOD
4834
4835 quote " | '
4836
4837 register specialRegister
4838 | gpRegister
4839 | byteRegister
4840
4841 segmentRegister CS | DS | ES | FS | GS | SS
4842
4843 specialRegister CR0 | CR2 | CR3
4844 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
4845 | TR3 | TR4 | TR5 | TR6 | TR7
4846
4847 We simplify the grammar in obvious places (e.g., register parsing is
4848 done by calling parse_register) and eliminate immediate left recursion
4849 to implement a recursive-descent parser.
4850
4851 expr SHORT e05
4852 | e05
4853
4854 e05 e06 e05'
4855
4856 e05' addOp e06 e05'
4857 | Empty
4858
4859 e06 e09 e06'
4860
4861 e06' mulOp e09 e06'
4862 | Empty
4863
4864 e09 OFFSET e10 e09'
4865 | e10 e09'
4866
4867 e09' PTR e10 e09'
4868 | : e10 e09'
4869 | Empty
4870
4871 e10 e11 e10'
4872
4873 e10' [ expr ] e10'
4874 | Empty
4875
4876 e11 ( expr )
4877 | [ expr ]
4878 | BYTE
4879 | WORD
4880 | DWORD
4881 | QWORD
4882 | XWORD
4883 | .
4884 | $
4885 | register
4886 | id
4887 | constant */
4888
4889 /* Parsing structure for the intel syntax parser. Used to implement the
4890 semantic actions for the operand grammar. */
4891 struct intel_parser_s
4892 {
4893 char *op_string; /* The string being parsed. */
4894 int got_a_float; /* Whether the operand is a float. */
4895 int op_modifier; /* Operand modifier. */
4896 int is_mem; /* 1 if operand is memory reference. */
4897 const reg_entry *reg; /* Last register reference found. */
4898 char *disp; /* Displacement string being built. */
4899 };
4900
4901 static struct intel_parser_s intel_parser;
4902
4903 /* Token structure for parsing intel syntax. */
4904 struct intel_token
4905 {
4906 int code; /* Token code. */
4907 const reg_entry *reg; /* Register entry for register tokens. */
4908 char *str; /* String representation. */
4909 };
4910
4911 static struct intel_token cur_token, prev_token;
4912
4913
4914 /* Token codes for the intel parser. Since T_SHORT is already used
4915 by COFF, undefine it first to prevent a warning. */
4916 #define T_NIL -1
4917 #define T_CONST 1
4918 #define T_REG 2
4919 #define T_BYTE 3
4920 #define T_WORD 4
4921 #define T_DWORD 5
4922 #define T_QWORD 6
4923 #define T_XWORD 7
4924 #undef T_SHORT
4925 #define T_SHORT 8
4926 #define T_OFFSET 9
4927 #define T_PTR 10
4928 #define T_ID 11
4929
4930 /* Prototypes for intel parser functions. */
4931 static int intel_match_token PARAMS ((int code));
4932 static void intel_get_token PARAMS ((void));
4933 static void intel_putback_token PARAMS ((void));
4934 static int intel_expr PARAMS ((void));
4935 static int intel_e05 PARAMS ((void));
4936 static int intel_e05_1 PARAMS ((void));
4937 static int intel_e06 PARAMS ((void));
4938 static int intel_e06_1 PARAMS ((void));
4939 static int intel_e09 PARAMS ((void));
4940 static int intel_e09_1 PARAMS ((void));
4941 static int intel_e10 PARAMS ((void));
4942 static int intel_e10_1 PARAMS ((void));
4943 static int intel_e11 PARAMS ((void));
4944
4945 static int
4946 i386_intel_operand (operand_string, got_a_float)
4947 char *operand_string;
4948 int got_a_float;
4949 {
4950 int ret;
4951 char *p;
4952
4953 /* Initialize token holders. */
4954 cur_token.code = prev_token.code = T_NIL;
4955 cur_token.reg = prev_token.reg = NULL;
4956 cur_token.str = prev_token.str = NULL;
4957
4958 /* Initialize parser structure. */
4959 p = intel_parser.op_string = (char *)malloc (strlen (operand_string) + 1);
4960 if (p == NULL)
4961 abort ();
4962 strcpy (intel_parser.op_string, operand_string);
4963 intel_parser.got_a_float = got_a_float;
4964 intel_parser.op_modifier = -1;
4965 intel_parser.is_mem = 0;
4966 intel_parser.reg = NULL;
4967 intel_parser.disp = (char *)malloc (strlen (operand_string) + 1);
4968 if (intel_parser.disp == NULL)
4969 abort ();
4970 intel_parser.disp[0] = '\0';
4971
4972 /* Read the first token and start the parser. */
4973 intel_get_token ();
4974 ret = intel_expr ();
4975
4976 if (ret)
4977 {
4978 /* If we found a memory reference, hand it over to i386_displacement
4979 to fill in the rest of the operand fields. */
4980 if (intel_parser.is_mem)
4981 {
4982 if ((i.mem_operands == 1
4983 && (current_templates->start->opcode_modifier & IsString) == 0)
4984 || i.mem_operands == 2)
4985 {
4986 as_bad (_("too many memory references for '%s'"),
4987 current_templates->start->name);
4988 ret = 0;
4989 }
4990 else
4991 {
4992 char *s = intel_parser.disp;
4993 i.mem_operands++;
4994
4995 /* Add the displacement expression. */
4996 if (*s != '\0')
4997 ret = i386_displacement (s, s + strlen (s))
4998 && i386_index_check (s);
4999 }
5000 }
5001
5002 /* Constant and OFFSET expressions are handled by i386_immediate. */
5003 else if (intel_parser.op_modifier == OFFSET_FLAT
5004 || intel_parser.reg == NULL)
5005 ret = i386_immediate (intel_parser.disp);
5006 }
5007
5008 free (p);
5009 free (intel_parser.disp);
5010
5011 return ret;
5012 }
5013
5014 /* expr SHORT e05
5015 | e05 */
5016 static int
5017 intel_expr ()
5018 {
5019 /* expr SHORT e05 */
5020 if (cur_token.code == T_SHORT)
5021 {
5022 intel_parser.op_modifier = SHORT;
5023 intel_match_token (T_SHORT);
5024
5025 return (intel_e05 ());
5026 }
5027
5028 /* expr e05 */
5029 else
5030 return intel_e05 ();
5031 }
5032
5033 /* e05 e06 e05'
5034
5035 e05' addOp e06 e05'
5036 | Empty */
5037 static int
5038 intel_e05 ()
5039 {
5040 return (intel_e06 () && intel_e05_1 ());
5041 }
5042
5043 static int
5044 intel_e05_1 ()
5045 {
5046 /* e05' addOp e06 e05' */
5047 if (cur_token.code == '+' || cur_token.code == '-')
5048 {
5049 strcat (intel_parser.disp, cur_token.str);
5050 intel_match_token (cur_token.code);
5051
5052 return (intel_e06 () && intel_e05_1 ());
5053 }
5054
5055 /* e05' Empty */
5056 else
5057 return 1;
5058 }
5059
5060 /* e06 e09 e06'
5061
5062 e06' mulOp e09 e06'
5063 | Empty */
5064 static int
5065 intel_e06 ()
5066 {
5067 return (intel_e09 () && intel_e06_1 ());
5068 }
5069
5070 static int
5071 intel_e06_1 ()
5072 {
5073 /* e06' mulOp e09 e06' */
5074 if (cur_token.code == '*' || cur_token.code == '/')
5075 {
5076 strcat (intel_parser.disp, cur_token.str);
5077 intel_match_token (cur_token.code);
5078
5079 return (intel_e09 () && intel_e06_1 ());
5080 }
5081
5082 /* e06' Empty */
5083 else
5084 return 1;
5085 }
5086
5087 /* e09 OFFSET e10 e09'
5088 | e10 e09'
5089
5090 e09' PTR e10 e09'
5091 | : e10 e09'
5092 | Empty */
5093 static int
5094 intel_e09 ()
5095 {
5096 /* e09 OFFSET e10 e09' */
5097 if (cur_token.code == T_OFFSET)
5098 {
5099 intel_parser.is_mem = 0;
5100 intel_parser.op_modifier = OFFSET_FLAT;
5101 intel_match_token (T_OFFSET);
5102
5103 return (intel_e10 () && intel_e09_1 ());
5104 }
5105
5106 /* e09 e10 e09' */
5107 else
5108 return (intel_e10 () && intel_e09_1 ());
5109 }
5110
5111 static int
5112 intel_e09_1 ()
5113 {
5114 /* e09' PTR e10 e09' */
5115 if (cur_token.code == T_PTR)
5116 {
5117 if (prev_token.code == T_BYTE)
5118 i.suffix = BYTE_MNEM_SUFFIX;
5119
5120 else if (prev_token.code == T_WORD)
5121 {
5122 if (intel_parser.got_a_float == 2) /* "fi..." */
5123 i.suffix = SHORT_MNEM_SUFFIX;
5124 else
5125 i.suffix = WORD_MNEM_SUFFIX;
5126 }
5127
5128 else if (prev_token.code == T_DWORD)
5129 {
5130 if (intel_parser.got_a_float == 1) /* "f..." */
5131 i.suffix = SHORT_MNEM_SUFFIX;
5132 else
5133 i.suffix = LONG_MNEM_SUFFIX;
5134 }
5135
5136 else if (prev_token.code == T_QWORD)
5137 {
5138 if (intel_parser.got_a_float == 1) /* "f..." */
5139 i.suffix = LONG_MNEM_SUFFIX;
5140 else
5141 i.suffix = QWORD_MNEM_SUFFIX;
5142 }
5143
5144 else if (prev_token.code == T_XWORD)
5145 i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
5146
5147 else
5148 {
5149 as_bad (_("Unknown operand modifier `%s'\n"), prev_token.str);
5150 return 0;
5151 }
5152
5153 intel_match_token (T_PTR);
5154
5155 return (intel_e10 () && intel_e09_1 ());
5156 }
5157
5158 /* e09 : e10 e09' */
5159 else if (cur_token.code == ':')
5160 {
5161 /* Mark as a memory operand only if it's not already known to be an
5162 offset expression. */
5163 if (intel_parser.op_modifier != OFFSET_FLAT)
5164 intel_parser.is_mem = 1;
5165
5166 return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
5167 }
5168
5169 /* e09' Empty */
5170 else
5171 return 1;
5172 }
5173
5174 /* e10 e11 e10'
5175
5176 e10' [ expr ] e10'
5177 | Empty */
5178 static int
5179 intel_e10 ()
5180 {
5181 return (intel_e11 () && intel_e10_1 ());
5182 }
5183
5184 static int
5185 intel_e10_1 ()
5186 {
5187 /* e10' [ expr ] e10' */
5188 if (cur_token.code == '[')
5189 {
5190 intel_match_token ('[');
5191
5192 /* Mark as a memory operand only if it's not already known to be an
5193 offset expression. If it's an offset expression, we need to keep
5194 the brace in. */
5195 if (intel_parser.op_modifier != OFFSET_FLAT)
5196 intel_parser.is_mem = 1;
5197 else
5198 strcat (intel_parser.disp, "[");
5199
5200 /* Add a '+' to the displacement string if necessary. */
5201 if (*intel_parser.disp != '\0'
5202 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5203 strcat (intel_parser.disp, "+");
5204
5205 if (intel_expr () && intel_match_token (']'))
5206 {
5207 /* Preserve brackets when the operand is an offset expression. */
5208 if (intel_parser.op_modifier == OFFSET_FLAT)
5209 strcat (intel_parser.disp, "]");
5210
5211 return intel_e10_1 ();
5212 }
5213 else
5214 return 0;
5215 }
5216
5217 /* e10' Empty */
5218 else
5219 return 1;
5220 }
5221
5222 /* e11 ( expr )
5223 | [ expr ]
5224 | BYTE
5225 | WORD
5226 | DWORD
5227 | QWORD
5228 | XWORD
5229 | $
5230 | .
5231 | register
5232 | id
5233 | constant */
5234 static int
5235 intel_e11 ()
5236 {
5237 /* e11 ( expr ) */
5238 if (cur_token.code == '(')
5239 {
5240 intel_match_token ('(');
5241 strcat (intel_parser.disp, "(");
5242
5243 if (intel_expr () && intel_match_token (')'))
5244 {
5245 strcat (intel_parser.disp, ")");
5246 return 1;
5247 }
5248 else
5249 return 0;
5250 }
5251
5252 /* e11 [ expr ] */
5253 else if (cur_token.code == '[')
5254 {
5255 intel_match_token ('[');
5256
5257 /* Mark as a memory operand only if it's not already known to be an
5258 offset expression. If it's an offset expression, we need to keep
5259 the brace in. */
5260 if (intel_parser.op_modifier != OFFSET_FLAT)
5261 intel_parser.is_mem = 1;
5262 else
5263 strcat (intel_parser.disp, "[");
5264
5265 /* Operands for jump/call inside brackets denote absolute addresses. */
5266 if (current_templates->start->opcode_modifier & Jump
5267 || current_templates->start->opcode_modifier & JumpDword
5268 || current_templates->start->opcode_modifier & JumpByte
5269 || current_templates->start->opcode_modifier & JumpInterSegment)
5270 i.types[this_operand] |= JumpAbsolute;
5271
5272 /* Add a '+' to the displacement string if necessary. */
5273 if (*intel_parser.disp != '\0'
5274 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5275 strcat (intel_parser.disp, "+");
5276
5277 if (intel_expr () && intel_match_token (']'))
5278 {
5279 /* Preserve brackets when the operand is an offset expression. */
5280 if (intel_parser.op_modifier == OFFSET_FLAT)
5281 strcat (intel_parser.disp, "]");
5282
5283 return 1;
5284 }
5285 else
5286 return 0;
5287 }
5288
5289 /* e11 BYTE
5290 | WORD
5291 | DWORD
5292 | QWORD
5293 | XWORD */
5294 else if (cur_token.code == T_BYTE
5295 || cur_token.code == T_WORD
5296 || cur_token.code == T_DWORD
5297 || cur_token.code == T_QWORD
5298 || cur_token.code == T_XWORD)
5299 {
5300 intel_match_token (cur_token.code);
5301
5302 return 1;
5303 }
5304
5305 /* e11 $
5306 | . */
5307 else if (cur_token.code == '$' || cur_token.code == '.')
5308 {
5309 strcat (intel_parser.disp, cur_token.str);
5310 intel_match_token (cur_token.code);
5311
5312 /* Mark as a memory operand only if it's not already known to be an
5313 offset expression. */
5314 if (intel_parser.op_modifier != OFFSET_FLAT)
5315 intel_parser.is_mem = 1;
5316
5317 return 1;
5318 }
5319
5320 /* e11 register */
5321 else if (cur_token.code == T_REG)
5322 {
5323 const reg_entry *reg = intel_parser.reg = cur_token.reg;
5324
5325 intel_match_token (T_REG);
5326
5327 /* Check for segment change. */
5328 if (cur_token.code == ':')
5329 {
5330 if (reg->reg_type & (SReg2 | SReg3))
5331 {
5332 switch (reg->reg_num)
5333 {
5334 case 0:
5335 i.seg[i.mem_operands] = &es;
5336 break;
5337 case 1:
5338 i.seg[i.mem_operands] = &cs;
5339 break;
5340 case 2:
5341 i.seg[i.mem_operands] = &ss;
5342 break;
5343 case 3:
5344 i.seg[i.mem_operands] = &ds;
5345 break;
5346 case 4:
5347 i.seg[i.mem_operands] = &fs;
5348 break;
5349 case 5:
5350 i.seg[i.mem_operands] = &gs;
5351 break;
5352 }
5353 }
5354 else
5355 {
5356 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
5357 return 0;
5358 }
5359 }
5360
5361 /* Not a segment register. Check for register scaling. */
5362 else if (cur_token.code == '*')
5363 {
5364 if (!intel_parser.is_mem)
5365 {
5366 as_bad (_("Register scaling only allowed in memory operands."));
5367 return 0;
5368 }
5369
5370 /* What follows must be a valid scale. */
5371 if (intel_match_token ('*')
5372 && strchr ("01248", *cur_token.str))
5373 {
5374 i.index_reg = reg;
5375 i.types[this_operand] |= BaseIndex;
5376
5377 /* Set the scale after setting the register (otherwise,
5378 i386_scale will complain) */
5379 i386_scale (cur_token.str);
5380 intel_match_token (T_CONST);
5381 }
5382 else
5383 {
5384 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
5385 cur_token.str);
5386 return 0;
5387 }
5388 }
5389
5390 /* No scaling. If this is a memory operand, the register is either a
5391 base register (first occurrence) or an index register (second
5392 occurrence). */
5393 else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3)))
5394 {
5395 if (i.base_reg && i.index_reg)
5396 {
5397 as_bad (_("Too many register references in memory operand.\n"));
5398 return 0;
5399 }
5400
5401 if (i.base_reg == NULL)
5402 i.base_reg = reg;
5403 else
5404 i.index_reg = reg;
5405
5406 i.types[this_operand] |= BaseIndex;
5407 }
5408
5409 /* Offset modifier. Add the register to the displacement string to be
5410 parsed as an immediate expression after we're done. */
5411 else if (intel_parser.op_modifier == OFFSET_FLAT)
5412 strcat (intel_parser.disp, reg->reg_name);
5413
5414 /* It's neither base nor index nor offset. */
5415 else
5416 {
5417 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
5418 i.op[this_operand].regs = reg;
5419 i.reg_operands++;
5420 }
5421
5422 /* Since registers are not part of the displacement string (except
5423 when we're parsing offset operands), we may need to remove any
5424 preceding '+' from the displacement string. */
5425 if (*intel_parser.disp != '\0'
5426 && intel_parser.op_modifier != OFFSET_FLAT)
5427 {
5428 char *s = intel_parser.disp;
5429 s += strlen (s) - 1;
5430 if (*s == '+')
5431 *s = '\0';
5432 }
5433
5434 return 1;
5435 }
5436
5437 /* e11 id */
5438 else if (cur_token.code == T_ID)
5439 {
5440 /* Add the identifier to the displacement string. */
5441 strcat (intel_parser.disp, cur_token.str);
5442 intel_match_token (T_ID);
5443
5444 /* The identifier represents a memory reference only if it's not
5445 preceded by an offset modifier. */
5446 if (intel_parser.op_modifier != OFFSET_FLAT)
5447 intel_parser.is_mem = 1;
5448
5449 return 1;
5450 }
5451
5452 /* e11 constant */
5453 else if (cur_token.code == T_CONST
5454 || cur_token.code == '-'
5455 || cur_token.code == '+')
5456 {
5457 char *save_str;
5458
5459 /* Allow constants that start with `+' or `-'. */
5460 if (cur_token.code == '-' || cur_token.code == '+')
5461 {
5462 strcat (intel_parser.disp, cur_token.str);
5463 intel_match_token (cur_token.code);
5464 if (cur_token.code != T_CONST)
5465 {
5466 as_bad (_("Syntax error. Expecting a constant. Got `%s'.\n"),
5467 cur_token.str);
5468 return 0;
5469 }
5470 }
5471
5472 save_str = (char *)malloc (strlen (cur_token.str) + 1);
5473 if (save_str == NULL)
5474 abort ();
5475 strcpy (save_str, cur_token.str);
5476
5477 /* Get the next token to check for register scaling. */
5478 intel_match_token (cur_token.code);
5479
5480 /* Check if this constant is a scaling factor for an index register. */
5481 if (cur_token.code == '*')
5482 {
5483 if (intel_match_token ('*') && cur_token.code == T_REG)
5484 {
5485 if (!intel_parser.is_mem)
5486 {
5487 as_bad (_("Register scaling only allowed in memory operands."));
5488 return 0;
5489 }
5490
5491 /* The constant is followed by `* reg', so it must be
5492 a valid scale. */
5493 if (strchr ("01248", *save_str))
5494 {
5495 i.index_reg = cur_token.reg;
5496 i.types[this_operand] |= BaseIndex;
5497
5498 /* Set the scale after setting the register (otherwise,
5499 i386_scale will complain) */
5500 i386_scale (save_str);
5501 intel_match_token (T_REG);
5502
5503 /* Since registers are not part of the displacement
5504 string, we may need to remove any preceding '+' from
5505 the displacement string. */
5506 if (*intel_parser.disp != '\0')
5507 {
5508 char *s = intel_parser.disp;
5509 s += strlen (s) - 1;
5510 if (*s == '+')
5511 *s = '\0';
5512 }
5513
5514 free (save_str);
5515
5516 return 1;
5517 }
5518 else
5519 return 0;
5520 }
5521
5522 /* The constant was not used for register scaling. Since we have
5523 already consumed the token following `*' we now need to put it
5524 back in the stream. */
5525 else
5526 intel_putback_token ();
5527 }
5528
5529 /* Add the constant to the displacement string. */
5530 strcat (intel_parser.disp, save_str);
5531 free (save_str);
5532
5533 return 1;
5534 }
5535
5536 as_bad (_("Unrecognized token '%s'"), cur_token.str);
5537 return 0;
5538 }
5539
5540 /* Match the given token against cur_token. If they match, read the next
5541 token from the operand string. */
5542 static int
5543 intel_match_token (code)
5544 int code;
5545 {
5546 if (cur_token.code == code)
5547 {
5548 intel_get_token ();
5549 return 1;
5550 }
5551 else
5552 {
5553 as_bad (_("Unexpected token `%s'\n"), cur_token.str);
5554 return 0;
5555 }
5556 }
5557
5558 /* Read a new token from intel_parser.op_string and store it in cur_token. */
5559 static void
5560 intel_get_token ()
5561 {
5562 char *end_op;
5563 const reg_entry *reg;
5564 struct intel_token new_token;
5565
5566 new_token.code = T_NIL;
5567 new_token.reg = NULL;
5568 new_token.str = NULL;
5569
5570 /* Free the memory allocated to the previous token and move
5571 cur_token to prev_token. */
5572 if (prev_token.str)
5573 free (prev_token.str);
5574
5575 prev_token = cur_token;
5576
5577 /* Skip whitespace. */
5578 while (is_space_char (*intel_parser.op_string))
5579 intel_parser.op_string++;
5580
5581 /* Return an empty token if we find nothing else on the line. */
5582 if (*intel_parser.op_string == '\0')
5583 {
5584 cur_token = new_token;
5585 return;
5586 }
5587
5588 /* The new token cannot be larger than the remainder of the operand
5589 string. */
5590 new_token.str = (char *)malloc (strlen (intel_parser.op_string) + 1);
5591 if (new_token.str == NULL)
5592 abort ();
5593 new_token.str[0] = '\0';
5594
5595 if (strchr ("0123456789", *intel_parser.op_string))
5596 {
5597 char *p = new_token.str;
5598 char *q = intel_parser.op_string;
5599 new_token.code = T_CONST;
5600
5601 /* Allow any kind of identifier char to encompass floating point and
5602 hexadecimal numbers. */
5603 while (is_identifier_char (*q))
5604 *p++ = *q++;
5605 *p = '\0';
5606
5607 /* Recognize special symbol names [0-9][bf]. */
5608 if (strlen (intel_parser.op_string) == 2
5609 && (intel_parser.op_string[1] == 'b'
5610 || intel_parser.op_string[1] == 'f'))
5611 new_token.code = T_ID;
5612 }
5613
5614 else if (strchr ("+-/*:[]()", *intel_parser.op_string))
5615 {
5616 new_token.code = *intel_parser.op_string;
5617 new_token.str[0] = *intel_parser.op_string;
5618 new_token.str[1] = '\0';
5619 }
5620
5621 else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
5622 && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
5623 {
5624 new_token.code = T_REG;
5625 new_token.reg = reg;
5626
5627 if (*intel_parser.op_string == REGISTER_PREFIX)
5628 {
5629 new_token.str[0] = REGISTER_PREFIX;
5630 new_token.str[1] = '\0';
5631 }
5632
5633 strcat (new_token.str, reg->reg_name);
5634 }
5635
5636 else if (is_identifier_char (*intel_parser.op_string))
5637 {
5638 char *p = new_token.str;
5639 char *q = intel_parser.op_string;
5640
5641 /* A '.' or '$' followed by an identifier char is an identifier.
5642 Otherwise, it's operator '.' followed by an expression. */
5643 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
5644 {
5645 new_token.code = *q;
5646 new_token.str[0] = *q;
5647 new_token.str[1] = '\0';
5648 }
5649 else
5650 {
5651 while (is_identifier_char (*q) || *q == '@')
5652 *p++ = *q++;
5653 *p = '\0';
5654
5655 if (strcasecmp (new_token.str, "BYTE") == 0)
5656 new_token.code = T_BYTE;
5657
5658 else if (strcasecmp (new_token.str, "WORD") == 0)
5659 new_token.code = T_WORD;
5660
5661 else if (strcasecmp (new_token.str, "DWORD") == 0)
5662 new_token.code = T_DWORD;
5663
5664 else if (strcasecmp (new_token.str, "QWORD") == 0)
5665 new_token.code = T_QWORD;
5666
5667 else if (strcasecmp (new_token.str, "XWORD") == 0)
5668 new_token.code = T_XWORD;
5669
5670 else if (strcasecmp (new_token.str, "PTR") == 0)
5671 new_token.code = T_PTR;
5672
5673 else if (strcasecmp (new_token.str, "SHORT") == 0)
5674 new_token.code = T_SHORT;
5675
5676 else if (strcasecmp (new_token.str, "OFFSET") == 0)
5677 {
5678 new_token.code = T_OFFSET;
5679
5680 /* ??? This is not mentioned in the MASM grammar but gcc
5681 makes use of it with -mintel-syntax. OFFSET may be
5682 followed by FLAT: */
5683 if (strncasecmp (q, " FLAT:", 6) == 0)
5684 strcat (new_token.str, " FLAT:");
5685 }
5686
5687 /* ??? This is not mentioned in the MASM grammar. */
5688 else if (strcasecmp (new_token.str, "FLAT") == 0)
5689 new_token.code = T_OFFSET;
5690
5691 else
5692 new_token.code = T_ID;
5693 }
5694 }
5695
5696 else
5697 as_bad (_("Unrecognized token `%s'\n"), intel_parser.op_string);
5698
5699 intel_parser.op_string += strlen (new_token.str);
5700 cur_token = new_token;
5701 }
5702
5703 /* Put cur_token back into the token stream and make cur_token point to
5704 prev_token. */
5705 static void
5706 intel_putback_token ()
5707 {
5708 intel_parser.op_string -= strlen (cur_token.str);
5709 free (cur_token.str);
5710 cur_token = prev_token;
5711
5712 /* Forget prev_token. */
5713 prev_token.code = T_NIL;
5714 prev_token.reg = NULL;
5715 prev_token.str = NULL;
5716 }
This page took 0.159635 seconds and 4 git commands to generate.