* tc-i386.c (cpu_arch): Add Pentium4 and modify sledgehammer entry.
[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 /* The prefix bytes. */
2830 for (q = i.prefix;
2831 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
2832 q++)
2833 {
2834 if (*q)
2835 {
2836 insn_size += 1;
2837 p = frag_more (1);
2838 md_number_to_chars (p, (valueT) *q, 1);
2839 }
2840 }
2841
2842 /* Now the opcode; be careful about word order here! */
2843 if (fits_in_unsigned_byte (i.tm.base_opcode))
2844 {
2845 insn_size += 1;
2846 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
2847 }
2848 else if (fits_in_unsigned_word (i.tm.base_opcode))
2849 {
2850 insn_size += 2;
2851 p = frag_more (2);
2852 /* Put out high byte first: can't use md_number_to_chars! */
2853 *p++ = (i.tm.base_opcode >> 8) & 0xff;
2854 *p = i.tm.base_opcode & 0xff;
2855 }
2856 else
2857 { /* Opcode is either 3 or 4 bytes. */
2858 if (i.tm.base_opcode & 0xff000000)
2859 {
2860 insn_size += 4;
2861 p = frag_more (4);
2862 *p++ = (i.tm.base_opcode >> 24) & 0xff;
2863 }
2864 else
2865 {
2866 insn_size += 3;
2867 p = frag_more (3);
2868 }
2869 *p++ = (i.tm.base_opcode >> 16) & 0xff;
2870 *p++ = (i.tm.base_opcode >> 8) & 0xff;
2871 *p = (i.tm.base_opcode) & 0xff;
2872 }
2873
2874 /* Now the modrm byte and sib byte (if present). */
2875 if (i.tm.opcode_modifier & Modrm)
2876 {
2877 insn_size += 1;
2878 p = frag_more (1);
2879 md_number_to_chars (p,
2880 (valueT) (i.rm.regmem << 0
2881 | i.rm.reg << 3
2882 | i.rm.mode << 6),
2883 1);
2884 /* If i.rm.regmem == ESP (4)
2885 && i.rm.mode != (Register mode)
2886 && not 16 bit
2887 ==> need second modrm byte. */
2888 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
2889 && i.rm.mode != 3
2890 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
2891 {
2892 insn_size += 1;
2893 p = frag_more (1);
2894 md_number_to_chars (p,
2895 (valueT) (i.sib.base << 0
2896 | i.sib.index << 3
2897 | i.sib.scale << 6),
2898 1);
2899 }
2900 }
2901
2902 if (i.disp_operands)
2903 {
2904 register unsigned int n;
2905
2906 for (n = 0; n < i.operands; n++)
2907 {
2908 if (i.types[n] & Disp)
2909 {
2910 if (i.op[n].disps->X_op == O_constant)
2911 {
2912 int size;
2913 offsetT val;
2914
2915 size = 4;
2916 if (i.types[n] & (Disp8 | Disp16 | Disp64))
2917 {
2918 size = 2;
2919 if (i.types[n] & Disp8)
2920 size = 1;
2921 if (i.types[n] & Disp64)
2922 size = 8;
2923 }
2924 val = offset_in_range (i.op[n].disps->X_add_number,
2925 size);
2926 insn_size += size;
2927 p = frag_more (size);
2928 md_number_to_chars (p, val, size);
2929 }
2930 else
2931 {
2932 int size = 4;
2933 int sign = 0;
2934 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
2935
2936 /* The PC relative address is computed relative
2937 to the instruction boundary, so in case immediate
2938 fields follows, we need to adjust the value. */
2939 if (pcrel && i.imm_operands)
2940 {
2941 int imm_size = 4;
2942 register unsigned int n1;
2943
2944 for (n1 = 0; n1 < i.operands; n1++)
2945 if (i.types[n1] & Imm)
2946 {
2947 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
2948 {
2949 imm_size = 2;
2950 if (i.types[n1] & (Imm8 | Imm8S))
2951 imm_size = 1;
2952 if (i.types[n1] & Imm64)
2953 imm_size = 8;
2954 }
2955 break;
2956 }
2957 /* We should find the immediate. */
2958 if (n1 == i.operands)
2959 abort();
2960 i.op[n].disps->X_add_number -= imm_size;
2961 }
2962
2963 if (i.types[n] & Disp32S)
2964 sign = 1;
2965
2966 if (i.types[n] & (Disp16 | Disp64))
2967 {
2968 size = 2;
2969 if (i.types[n] & Disp64)
2970 size = 8;
2971 }
2972
2973 insn_size += size;
2974 p = frag_more (size);
2975 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2976 i.op[n].disps, pcrel,
2977 reloc (size, pcrel, sign, i.disp_reloc[n]));
2978 }
2979 }
2980 }
2981 }
2982
2983 /* Output immediate. */
2984 if (i.imm_operands)
2985 {
2986 register unsigned int n;
2987
2988 for (n = 0; n < i.operands; n++)
2989 {
2990 if (i.types[n] & Imm)
2991 {
2992 if (i.op[n].imms->X_op == O_constant)
2993 {
2994 int size;
2995 offsetT val;
2996
2997 size = 4;
2998 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
2999 {
3000 size = 2;
3001 if (i.types[n] & (Imm8 | Imm8S))
3002 size = 1;
3003 else if (i.types[n] & Imm64)
3004 size = 8;
3005 }
3006 val = offset_in_range (i.op[n].imms->X_add_number,
3007 size);
3008 insn_size += size;
3009 p = frag_more (size);
3010 md_number_to_chars (p, val, size);
3011 }
3012 else
3013 {
3014 /* Not absolute_section.
3015 Need a 32-bit fixup (don't support 8bit
3016 non-absolute imms). Try to support other
3017 sizes ... */
3018 #ifdef BFD_ASSEMBLER
3019 enum bfd_reloc_code_real reloc_type;
3020 #else
3021 int reloc_type;
3022 #endif
3023 int size = 4;
3024 int sign = 0;
3025
3026 if ((i.types[n] & (Imm32S))
3027 && i.suffix == QWORD_MNEM_SUFFIX)
3028 sign = 1;
3029 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3030 {
3031 size = 2;
3032 if (i.types[n] & (Imm8 | Imm8S))
3033 size = 1;
3034 if (i.types[n] & Imm64)
3035 size = 8;
3036 }
3037
3038 insn_size += size;
3039 p = frag_more (size);
3040 reloc_type = reloc (size, 0, sign, i.disp_reloc[0]);
3041 #ifdef BFD_ASSEMBLER
3042 if (reloc_type == BFD_RELOC_32
3043 && GOT_symbol
3044 && GOT_symbol == i.op[n].imms->X_add_symbol
3045 && (i.op[n].imms->X_op == O_symbol
3046 || (i.op[n].imms->X_op == O_add
3047 && ((symbol_get_value_expression
3048 (i.op[n].imms->X_op_symbol)->X_op)
3049 == O_subtract))))
3050 {
3051 /* We don't support dynamic linking on x86-64 yet. */
3052 if (flag_code == CODE_64BIT)
3053 abort();
3054 reloc_type = BFD_RELOC_386_GOTPC;
3055 i.op[n].imms->X_add_number += 3;
3056 }
3057 #endif
3058 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3059 i.op[n].imms, 0, reloc_type);
3060 }
3061 }
3062 }
3063 }
3064 }
3065
3066 dwarf2_emit_insn (insn_size);
3067
3068 #ifdef DEBUG386
3069 if (flag_debug)
3070 {
3071 pi (line, &i);
3072 }
3073 #endif /* DEBUG386 */
3074 }
3075 }
3076 \f
3077 static int i386_immediate PARAMS ((char *));
3078
3079 static int
3080 i386_immediate (imm_start)
3081 char *imm_start;
3082 {
3083 char *save_input_line_pointer;
3084 segT exp_seg = 0;
3085 expressionS *exp;
3086
3087 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
3088 {
3089 as_bad (_("only 1 or 2 immediate operands are allowed"));
3090 return 0;
3091 }
3092
3093 exp = &im_expressions[i.imm_operands++];
3094 i.op[this_operand].imms = exp;
3095
3096 if (is_space_char (*imm_start))
3097 ++imm_start;
3098
3099 save_input_line_pointer = input_line_pointer;
3100 input_line_pointer = imm_start;
3101
3102 #ifndef LEX_AT
3103 {
3104 /* We can have operands of the form
3105 <symbol>@GOTOFF+<nnn>
3106 Take the easy way out here and copy everything
3107 into a temporary buffer... */
3108 register char *cp;
3109
3110 cp = strchr (input_line_pointer, '@');
3111 if (cp != NULL)
3112 {
3113 char *tmpbuf;
3114 int len = 0;
3115 int first;
3116
3117 /* GOT relocations are not supported in 16 bit mode. */
3118 if (flag_code == CODE_16BIT)
3119 as_bad (_("GOT relocations not supported in 16 bit mode"));
3120
3121 if (GOT_symbol == NULL)
3122 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3123
3124 if (strncmp (cp + 1, "PLT", 3) == 0)
3125 {
3126 if (flag_code == CODE_64BIT)
3127 i.disp_reloc[this_operand] = BFD_RELOC_X86_64_PLT32;
3128 else
3129 i.disp_reloc[this_operand] = BFD_RELOC_386_PLT32;
3130 len = 3;
3131 }
3132 else if (strncmp (cp + 1, "GOTOFF", 6) == 0)
3133 {
3134 if (flag_code == CODE_64BIT)
3135 as_bad ("GOTOFF relocations are unsupported in 64bit mode.");
3136 i.disp_reloc[this_operand] = BFD_RELOC_386_GOTOFF;
3137 len = 6;
3138 }
3139 else if (strncmp (cp + 1, "GOT", 3) == 0)
3140 {
3141 if (flag_code == CODE_64BIT)
3142 i.disp_reloc[this_operand] = BFD_RELOC_X86_64_GOT32;
3143 else
3144 i.disp_reloc[this_operand] = BFD_RELOC_386_GOT32;
3145 len = 3;
3146 }
3147 else if (strncmp (cp + 1, "GOTPCREL", 3) == 0)
3148 {
3149 if (flag_code == CODE_64BIT)
3150 i.disp_reloc[this_operand] = BFD_RELOC_X86_64_GOTPCREL;
3151 else
3152 as_bad ("GOTPCREL relocations are supported only in 64bit mode.");
3153 len = 3;
3154 }
3155 else
3156 as_bad (_("bad reloc specifier in expression"));
3157
3158 /* Replace the relocation token with ' ', so that errors like
3159 foo@GOTOFF1 will be detected. */
3160 first = cp - input_line_pointer;
3161 tmpbuf = (char *) alloca (strlen (input_line_pointer));
3162 memcpy (tmpbuf, input_line_pointer, first);
3163 tmpbuf[first] = ' ';
3164 strcpy (tmpbuf + first + 1, cp + 1 + len);
3165 input_line_pointer = tmpbuf;
3166 }
3167 }
3168 #endif
3169
3170 exp_seg = expression (exp);
3171
3172 SKIP_WHITESPACE ();
3173 if (*input_line_pointer)
3174 as_bad (_("ignoring junk `%s' after expression"), input_line_pointer);
3175
3176 input_line_pointer = save_input_line_pointer;
3177
3178 if (exp->X_op == O_absent || exp->X_op == O_big)
3179 {
3180 /* Missing or bad expr becomes absolute 0. */
3181 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
3182 imm_start);
3183 exp->X_op = O_constant;
3184 exp->X_add_number = 0;
3185 exp->X_add_symbol = (symbolS *) 0;
3186 exp->X_op_symbol = (symbolS *) 0;
3187 }
3188 else if (exp->X_op == O_constant)
3189 {
3190 /* Size it properly later. */
3191 i.types[this_operand] |= Imm64;
3192 /* If BFD64, sign extend val. */
3193 if (!use_rela_relocations)
3194 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
3195 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
3196 }
3197 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3198 else if (1
3199 #ifdef BFD_ASSEMBLER
3200 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3201 #endif
3202 && exp_seg != text_section
3203 && exp_seg != data_section
3204 && exp_seg != bss_section
3205 && exp_seg != undefined_section
3206 #ifdef BFD_ASSEMBLER
3207 && !bfd_is_com_section (exp_seg)
3208 #endif
3209 )
3210 {
3211 #ifdef BFD_ASSEMBLER
3212 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3213 #else
3214 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3215 #endif
3216 return 0;
3217 }
3218 #endif
3219 else
3220 {
3221 /* This is an address. The size of the address will be
3222 determined later, depending on destination register,
3223 suffix, or the default for the section. */
3224 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
3225 }
3226
3227 return 1;
3228 }
3229
3230 static int i386_scale PARAMS ((char *));
3231
3232 static int
3233 i386_scale (scale)
3234 char *scale;
3235 {
3236 if (!isdigit (*scale))
3237 goto bad_scale;
3238
3239 switch (*scale)
3240 {
3241 case '0':
3242 case '1':
3243 i.log2_scale_factor = 0;
3244 break;
3245 case '2':
3246 i.log2_scale_factor = 1;
3247 break;
3248 case '4':
3249 i.log2_scale_factor = 2;
3250 break;
3251 case '8':
3252 i.log2_scale_factor = 3;
3253 break;
3254 default:
3255 bad_scale:
3256 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
3257 scale);
3258 return 0;
3259 }
3260 if (i.log2_scale_factor != 0 && ! i.index_reg)
3261 {
3262 as_warn (_("scale factor of %d without an index register"),
3263 1 << i.log2_scale_factor);
3264 #if SCALE1_WHEN_NO_INDEX
3265 i.log2_scale_factor = 0;
3266 #endif
3267 }
3268 return 1;
3269 }
3270
3271 static int i386_displacement PARAMS ((char *, char *));
3272
3273 static int
3274 i386_displacement (disp_start, disp_end)
3275 char *disp_start;
3276 char *disp_end;
3277 {
3278 register expressionS *exp;
3279 segT exp_seg = 0;
3280 char *save_input_line_pointer;
3281 int bigdisp = Disp32;
3282
3283 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3284 bigdisp = Disp16;
3285 if (flag_code == CODE_64BIT)
3286 bigdisp = Disp64;
3287 i.types[this_operand] |= bigdisp;
3288
3289 exp = &disp_expressions[i.disp_operands];
3290 i.op[this_operand].disps = exp;
3291 i.disp_operands++;
3292 save_input_line_pointer = input_line_pointer;
3293 input_line_pointer = disp_start;
3294 END_STRING_AND_SAVE (disp_end);
3295
3296 #ifndef GCC_ASM_O_HACK
3297 #define GCC_ASM_O_HACK 0
3298 #endif
3299 #if GCC_ASM_O_HACK
3300 END_STRING_AND_SAVE (disp_end + 1);
3301 if ((i.types[this_operand] & BaseIndex) != 0
3302 && displacement_string_end[-1] == '+')
3303 {
3304 /* This hack is to avoid a warning when using the "o"
3305 constraint within gcc asm statements.
3306 For instance:
3307
3308 #define _set_tssldt_desc(n,addr,limit,type) \
3309 __asm__ __volatile__ ( \
3310 "movw %w2,%0\n\t" \
3311 "movw %w1,2+%0\n\t" \
3312 "rorl $16,%1\n\t" \
3313 "movb %b1,4+%0\n\t" \
3314 "movb %4,5+%0\n\t" \
3315 "movb $0,6+%0\n\t" \
3316 "movb %h1,7+%0\n\t" \
3317 "rorl $16,%1" \
3318 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
3319
3320 This works great except that the output assembler ends
3321 up looking a bit weird if it turns out that there is
3322 no offset. You end up producing code that looks like:
3323
3324 #APP
3325 movw $235,(%eax)
3326 movw %dx,2+(%eax)
3327 rorl $16,%edx
3328 movb %dl,4+(%eax)
3329 movb $137,5+(%eax)
3330 movb $0,6+(%eax)
3331 movb %dh,7+(%eax)
3332 rorl $16,%edx
3333 #NO_APP
3334
3335 So here we provide the missing zero. */
3336
3337 *displacement_string_end = '0';
3338 }
3339 #endif
3340 #ifndef LEX_AT
3341 {
3342 /* We can have operands of the form
3343 <symbol>@GOTOFF+<nnn>
3344 Take the easy way out here and copy everything
3345 into a temporary buffer... */
3346 register char *cp;
3347
3348 cp = strchr (input_line_pointer, '@');
3349 if (cp != NULL)
3350 {
3351 char *tmpbuf;
3352 int len = 0;
3353 int first;
3354
3355 /* GOT relocations are not supported in 16 bit mode. */
3356 if (flag_code == CODE_16BIT)
3357 as_bad (_("GOT relocations not supported in 16 bit mode"));
3358
3359 if (GOT_symbol == NULL)
3360 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3361
3362 if (strncmp (cp + 1, "PLT", 3) == 0)
3363 {
3364 if (flag_code == CODE_64BIT)
3365 i.disp_reloc[this_operand] = BFD_RELOC_X86_64_PLT32;
3366 else
3367 i.disp_reloc[this_operand] = BFD_RELOC_386_PLT32;
3368 len = 3;
3369 }
3370 else if (strncmp (cp + 1, "GOTOFF", 6) == 0)
3371 {
3372 if (flag_code == CODE_64BIT)
3373 as_bad ("GOTOFF relocation is not supported in 64bit mode.");
3374 i.disp_reloc[this_operand] = BFD_RELOC_386_GOTOFF;
3375 len = 6;
3376 }
3377 else if (strncmp (cp + 1, "GOT", 3) == 0)
3378 {
3379 if (flag_code == CODE_64BIT)
3380 i.disp_reloc[this_operand] = BFD_RELOC_X86_64_GOT32;
3381 else
3382 i.disp_reloc[this_operand] = BFD_RELOC_386_GOT32;
3383 len = 3;
3384 }
3385 else if (strncmp (cp + 1, "GOTPCREL", 3) == 0)
3386 {
3387 if (flag_code != CODE_64BIT)
3388 as_bad ("GOTPCREL relocation is supported only in 64bit mode.");
3389 i.disp_reloc[this_operand] = BFD_RELOC_X86_64_GOTPCREL;
3390 len = 3;
3391 }
3392 else
3393 as_bad (_("bad reloc specifier in expression"));
3394
3395 /* Replace the relocation token with ' ', so that errors like
3396 foo@GOTOFF1 will be detected. */
3397 first = cp - input_line_pointer;
3398 tmpbuf = (char *) alloca (strlen (input_line_pointer));
3399 memcpy (tmpbuf, input_line_pointer, first);
3400 tmpbuf[first] = ' ';
3401 strcpy (tmpbuf + first + 1, cp + 1 + len);
3402 input_line_pointer = tmpbuf;
3403 }
3404 }
3405 #endif
3406
3407 exp_seg = expression (exp);
3408
3409 #ifdef BFD_ASSEMBLER
3410 /* We do this to make sure that the section symbol is in
3411 the symbol table. We will ultimately change the relocation
3412 to be relative to the beginning of the section. */
3413 if (i.disp_reloc[this_operand] == BFD_RELOC_386_GOTOFF
3414 || i.disp_reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3415 {
3416 if (S_IS_LOCAL(exp->X_add_symbol)
3417 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
3418 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
3419 assert (exp->X_op == O_symbol);
3420 exp->X_op = O_subtract;
3421 exp->X_op_symbol = GOT_symbol;
3422 i.disp_reloc[this_operand] = BFD_RELOC_32;
3423 }
3424 #endif
3425
3426 SKIP_WHITESPACE ();
3427 if (*input_line_pointer)
3428 as_bad (_("ignoring junk `%s' after expression"),
3429 input_line_pointer);
3430 #if GCC_ASM_O_HACK
3431 RESTORE_END_STRING (disp_end + 1);
3432 #endif
3433 RESTORE_END_STRING (disp_end);
3434 input_line_pointer = save_input_line_pointer;
3435
3436 if (exp->X_op == O_absent || exp->X_op == O_big)
3437 {
3438 /* Missing or bad expr becomes absolute 0. */
3439 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
3440 disp_start);
3441 exp->X_op = O_constant;
3442 exp->X_add_number = 0;
3443 exp->X_add_symbol = (symbolS *) 0;
3444 exp->X_op_symbol = (symbolS *) 0;
3445 }
3446
3447 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3448 if (exp->X_op != O_constant
3449 #ifdef BFD_ASSEMBLER
3450 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3451 #endif
3452 && exp_seg != text_section
3453 && exp_seg != data_section
3454 && exp_seg != bss_section
3455 && exp_seg != undefined_section)
3456 {
3457 #ifdef BFD_ASSEMBLER
3458 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3459 #else
3460 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3461 #endif
3462 return 0;
3463 }
3464 #endif
3465 else if (flag_code == CODE_64BIT)
3466 i.types[this_operand] |= Disp32S | Disp32;
3467 return 1;
3468 }
3469
3470 static int i386_index_check PARAMS((const char *));
3471
3472 /* Make sure the memory operand we've been dealt is valid.
3473 Return 1 on success, 0 on a failure. */
3474
3475 static int
3476 i386_index_check (operand_string)
3477 const char *operand_string;
3478 {
3479 int ok;
3480 #if INFER_ADDR_PREFIX
3481 int fudged = 0;
3482
3483 tryprefix:
3484 #endif
3485 ok = 1;
3486 if (flag_code == CODE_64BIT)
3487 {
3488 /* 64bit checks. */
3489 if ((i.base_reg
3490 && ((i.base_reg->reg_type & Reg64) == 0)
3491 && (i.base_reg->reg_type != BaseIndex
3492 || i.index_reg))
3493 || (i.index_reg
3494 && ((i.index_reg->reg_type & (Reg64|BaseIndex))
3495 != (Reg64|BaseIndex))))
3496 ok = 0;
3497 }
3498 else
3499 {
3500 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3501 {
3502 /* 16bit checks. */
3503 if ((i.base_reg
3504 && ((i.base_reg->reg_type & (Reg16|BaseIndex|RegRex))
3505 != (Reg16|BaseIndex)))
3506 || (i.index_reg
3507 && (((i.index_reg->reg_type & (Reg16|BaseIndex))
3508 != (Reg16|BaseIndex))
3509 || ! (i.base_reg
3510 && i.base_reg->reg_num < 6
3511 && i.index_reg->reg_num >= 6
3512 && i.log2_scale_factor == 0))))
3513 ok = 0;
3514 }
3515 else
3516 {
3517 /* 32bit checks. */
3518 if ((i.base_reg
3519 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
3520 || (i.index_reg
3521 && ((i.index_reg->reg_type & (Reg32|BaseIndex|RegRex))
3522 != (Reg32|BaseIndex))))
3523 ok = 0;
3524 }
3525 }
3526 if (!ok)
3527 {
3528 #if INFER_ADDR_PREFIX
3529 if (flag_code != CODE_64BIT
3530 && i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0')
3531 {
3532 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
3533 i.prefixes += 1;
3534 /* Change the size of any displacement too. At most one of
3535 Disp16 or Disp32 is set.
3536 FIXME. There doesn't seem to be any real need for separate
3537 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
3538 Removing them would probably clean up the code quite a lot. */
3539 if (i.types[this_operand] & (Disp16|Disp32))
3540 i.types[this_operand] ^= (Disp16|Disp32);
3541 fudged = 1;
3542 goto tryprefix;
3543 }
3544 if (fudged)
3545 as_bad (_("`%s' is not a valid base/index expression"),
3546 operand_string);
3547 else
3548 #endif
3549 as_bad (_("`%s' is not a valid %s bit base/index expression"),
3550 operand_string,
3551 flag_code_names[flag_code]);
3552 return 0;
3553 }
3554 return 1;
3555 }
3556
3557 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
3558 on error. */
3559
3560 static int
3561 i386_operand (operand_string)
3562 char *operand_string;
3563 {
3564 const reg_entry *r;
3565 char *end_op;
3566 char *op_string = operand_string;
3567
3568 if (is_space_char (*op_string))
3569 ++op_string;
3570
3571 /* We check for an absolute prefix (differentiating,
3572 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
3573 if (*op_string == ABSOLUTE_PREFIX)
3574 {
3575 ++op_string;
3576 if (is_space_char (*op_string))
3577 ++op_string;
3578 i.types[this_operand] |= JumpAbsolute;
3579 }
3580
3581 /* Check if operand is a register. */
3582 if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
3583 && (r = parse_register (op_string, &end_op)) != NULL)
3584 {
3585 /* Check for a segment override by searching for ':' after a
3586 segment register. */
3587 op_string = end_op;
3588 if (is_space_char (*op_string))
3589 ++op_string;
3590 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
3591 {
3592 switch (r->reg_num)
3593 {
3594 case 0:
3595 i.seg[i.mem_operands] = &es;
3596 break;
3597 case 1:
3598 i.seg[i.mem_operands] = &cs;
3599 break;
3600 case 2:
3601 i.seg[i.mem_operands] = &ss;
3602 break;
3603 case 3:
3604 i.seg[i.mem_operands] = &ds;
3605 break;
3606 case 4:
3607 i.seg[i.mem_operands] = &fs;
3608 break;
3609 case 5:
3610 i.seg[i.mem_operands] = &gs;
3611 break;
3612 }
3613
3614 /* Skip the ':' and whitespace. */
3615 ++op_string;
3616 if (is_space_char (*op_string))
3617 ++op_string;
3618
3619 if (!is_digit_char (*op_string)
3620 && !is_identifier_char (*op_string)
3621 && *op_string != '('
3622 && *op_string != ABSOLUTE_PREFIX)
3623 {
3624 as_bad (_("bad memory operand `%s'"), op_string);
3625 return 0;
3626 }
3627 /* Handle case of %es:*foo. */
3628 if (*op_string == ABSOLUTE_PREFIX)
3629 {
3630 ++op_string;
3631 if (is_space_char (*op_string))
3632 ++op_string;
3633 i.types[this_operand] |= JumpAbsolute;
3634 }
3635 goto do_memory_reference;
3636 }
3637 if (*op_string)
3638 {
3639 as_bad (_("junk `%s' after register"), op_string);
3640 return 0;
3641 }
3642 i.types[this_operand] |= r->reg_type & ~BaseIndex;
3643 i.op[this_operand].regs = r;
3644 i.reg_operands++;
3645 }
3646 else if (*op_string == REGISTER_PREFIX)
3647 {
3648 as_bad (_("bad register name `%s'"), op_string);
3649 return 0;
3650 }
3651 else if (*op_string == IMMEDIATE_PREFIX)
3652 {
3653 ++op_string;
3654 if (i.types[this_operand] & JumpAbsolute)
3655 {
3656 as_bad (_("immediate operand illegal with absolute jump"));
3657 return 0;
3658 }
3659 if (!i386_immediate (op_string))
3660 return 0;
3661 }
3662 else if (is_digit_char (*op_string)
3663 || is_identifier_char (*op_string)
3664 || *op_string == '(' )
3665 {
3666 /* This is a memory reference of some sort. */
3667 char *base_string;
3668
3669 /* Start and end of displacement string expression (if found). */
3670 char *displacement_string_start;
3671 char *displacement_string_end;
3672
3673 do_memory_reference:
3674 if ((i.mem_operands == 1
3675 && (current_templates->start->opcode_modifier & IsString) == 0)
3676 || i.mem_operands == 2)
3677 {
3678 as_bad (_("too many memory references for `%s'"),
3679 current_templates->start->name);
3680 return 0;
3681 }
3682
3683 /* Check for base index form. We detect the base index form by
3684 looking for an ')' at the end of the operand, searching
3685 for the '(' matching it, and finding a REGISTER_PREFIX or ','
3686 after the '('. */
3687 base_string = op_string + strlen (op_string);
3688
3689 --base_string;
3690 if (is_space_char (*base_string))
3691 --base_string;
3692
3693 /* If we only have a displacement, set-up for it to be parsed later. */
3694 displacement_string_start = op_string;
3695 displacement_string_end = base_string + 1;
3696
3697 if (*base_string == ')')
3698 {
3699 char *temp_string;
3700 unsigned int parens_balanced = 1;
3701 /* We've already checked that the number of left & right ()'s are
3702 equal, so this loop will not be infinite. */
3703 do
3704 {
3705 base_string--;
3706 if (*base_string == ')')
3707 parens_balanced++;
3708 if (*base_string == '(')
3709 parens_balanced--;
3710 }
3711 while (parens_balanced);
3712
3713 temp_string = base_string;
3714
3715 /* Skip past '(' and whitespace. */
3716 ++base_string;
3717 if (is_space_char (*base_string))
3718 ++base_string;
3719
3720 if (*base_string == ','
3721 || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3722 && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
3723 {
3724 displacement_string_end = temp_string;
3725
3726 i.types[this_operand] |= BaseIndex;
3727
3728 if (i.base_reg)
3729 {
3730 base_string = end_op;
3731 if (is_space_char (*base_string))
3732 ++base_string;
3733 }
3734
3735 /* There may be an index reg or scale factor here. */
3736 if (*base_string == ',')
3737 {
3738 ++base_string;
3739 if (is_space_char (*base_string))
3740 ++base_string;
3741
3742 if ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3743 && (i.index_reg = parse_register (base_string, &end_op)) != NULL)
3744 {
3745 base_string = end_op;
3746 if (is_space_char (*base_string))
3747 ++base_string;
3748 if (*base_string == ',')
3749 {
3750 ++base_string;
3751 if (is_space_char (*base_string))
3752 ++base_string;
3753 }
3754 else if (*base_string != ')' )
3755 {
3756 as_bad (_("expecting `,' or `)' after index register in `%s'"),
3757 operand_string);
3758 return 0;
3759 }
3760 }
3761 else if (*base_string == REGISTER_PREFIX)
3762 {
3763 as_bad (_("bad register name `%s'"), base_string);
3764 return 0;
3765 }
3766
3767 /* Check for scale factor. */
3768 if (isdigit ((unsigned char) *base_string))
3769 {
3770 if (!i386_scale (base_string))
3771 return 0;
3772
3773 ++base_string;
3774 if (is_space_char (*base_string))
3775 ++base_string;
3776 if (*base_string != ')')
3777 {
3778 as_bad (_("expecting `)' after scale factor in `%s'"),
3779 operand_string);
3780 return 0;
3781 }
3782 }
3783 else if (!i.index_reg)
3784 {
3785 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
3786 *base_string);
3787 return 0;
3788 }
3789 }
3790 else if (*base_string != ')')
3791 {
3792 as_bad (_("expecting `,' or `)' after base register in `%s'"),
3793 operand_string);
3794 return 0;
3795 }
3796 }
3797 else if (*base_string == REGISTER_PREFIX)
3798 {
3799 as_bad (_("bad register name `%s'"), base_string);
3800 return 0;
3801 }
3802 }
3803
3804 /* If there's an expression beginning the operand, parse it,
3805 assuming displacement_string_start and
3806 displacement_string_end are meaningful. */
3807 if (displacement_string_start != displacement_string_end)
3808 {
3809 if (!i386_displacement (displacement_string_start,
3810 displacement_string_end))
3811 return 0;
3812 }
3813
3814 /* Special case for (%dx) while doing input/output op. */
3815 if (i.base_reg
3816 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
3817 && i.index_reg == 0
3818 && i.log2_scale_factor == 0
3819 && i.seg[i.mem_operands] == 0
3820 && (i.types[this_operand] & Disp) == 0)
3821 {
3822 i.types[this_operand] = InOutPortReg;
3823 return 1;
3824 }
3825
3826 if (i386_index_check (operand_string) == 0)
3827 return 0;
3828 i.mem_operands++;
3829 }
3830 else
3831 {
3832 /* It's not a memory operand; argh! */
3833 as_bad (_("invalid char %s beginning operand %d `%s'"),
3834 output_invalid (*op_string),
3835 this_operand + 1,
3836 op_string);
3837 return 0;
3838 }
3839 return 1; /* Normal return. */
3840 }
3841 \f
3842 /* md_estimate_size_before_relax()
3843
3844 Called just before relax() for rs_machine_dependent frags. The x86
3845 assembler uses these frags to handle variable size jump
3846 instructions.
3847
3848 Any symbol that is now undefined will not become defined.
3849 Return the correct fr_subtype in the frag.
3850 Return the initial "guess for variable size of frag" to caller.
3851 The guess is actually the growth beyond the fixed part. Whatever
3852 we do to grow the fixed or variable part contributes to our
3853 returned value. */
3854
3855 int
3856 md_estimate_size_before_relax (fragP, segment)
3857 register fragS *fragP;
3858 register segT segment;
3859 {
3860 /* We've already got fragP->fr_subtype right; all we have to do is
3861 check for un-relaxable symbols. On an ELF system, we can't relax
3862 an externally visible symbol, because it may be overridden by a
3863 shared library. */
3864 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
3865 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
3866 || S_IS_EXTERNAL (fragP->fr_symbol)
3867 || S_IS_WEAK (fragP->fr_symbol)
3868 #endif
3869 )
3870 {
3871 /* Symbol is undefined in this segment, or we need to keep a
3872 reloc so that weak symbols can be overridden. */
3873 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
3874 #ifdef BFD_ASSEMBLER
3875 enum bfd_reloc_code_real reloc_type;
3876 #else
3877 int reloc_type;
3878 #endif
3879 unsigned char *opcode;
3880 int old_fr_fix;
3881
3882 if (fragP->fr_var != NO_RELOC)
3883 reloc_type = fragP->fr_var;
3884 else if (size == 2)
3885 reloc_type = BFD_RELOC_16_PCREL;
3886 else
3887 reloc_type = BFD_RELOC_32_PCREL;
3888
3889 old_fr_fix = fragP->fr_fix;
3890 opcode = (unsigned char *) fragP->fr_opcode;
3891
3892 switch (opcode[0])
3893 {
3894 case JUMP_PC_RELATIVE:
3895 /* Make jmp (0xeb) a dword displacement jump. */
3896 opcode[0] = 0xe9;
3897 fragP->fr_fix += size;
3898 fix_new (fragP, old_fr_fix, size,
3899 fragP->fr_symbol,
3900 fragP->fr_offset, 1,
3901 reloc_type);
3902 break;
3903
3904 default:
3905 /* This changes the byte-displacement jump 0x7N
3906 to the dword-displacement jump 0x0f,0x8N. */
3907 opcode[1] = opcode[0] + 0x10;
3908 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
3909 /* We've added an opcode byte. */
3910 fragP->fr_fix += 1 + size;
3911 fix_new (fragP, old_fr_fix + 1, size,
3912 fragP->fr_symbol,
3913 fragP->fr_offset, 1,
3914 reloc_type);
3915 break;
3916 }
3917 frag_wane (fragP);
3918 return fragP->fr_fix - old_fr_fix;
3919 }
3920 /* Guess a short jump. */
3921 return 1;
3922 }
3923
3924 /* Called after relax() is finished.
3925
3926 In: Address of frag.
3927 fr_type == rs_machine_dependent.
3928 fr_subtype is what the address relaxed to.
3929
3930 Out: Any fixSs and constants are set up.
3931 Caller will turn frag into a ".space 0". */
3932
3933 #ifndef BFD_ASSEMBLER
3934 void
3935 md_convert_frag (headers, sec, fragP)
3936 object_headers *headers ATTRIBUTE_UNUSED;
3937 segT sec ATTRIBUTE_UNUSED;
3938 register fragS *fragP;
3939 #else
3940 void
3941 md_convert_frag (abfd, sec, fragP)
3942 bfd *abfd ATTRIBUTE_UNUSED;
3943 segT sec ATTRIBUTE_UNUSED;
3944 register fragS *fragP;
3945 #endif
3946 {
3947 register unsigned char *opcode;
3948 unsigned char *where_to_put_displacement = NULL;
3949 offsetT target_address;
3950 offsetT opcode_address;
3951 unsigned int extension = 0;
3952 offsetT displacement_from_opcode_start;
3953
3954 opcode = (unsigned char *) fragP->fr_opcode;
3955
3956 /* Address we want to reach in file space. */
3957 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
3958 #ifdef BFD_ASSEMBLER
3959 /* Not needed otherwise? */
3960 target_address += symbol_get_frag (fragP->fr_symbol)->fr_address;
3961 #endif
3962
3963 /* Address opcode resides at in file space. */
3964 opcode_address = fragP->fr_address + fragP->fr_fix;
3965
3966 /* Displacement from opcode start to fill into instruction. */
3967 displacement_from_opcode_start = target_address - opcode_address;
3968
3969 switch (fragP->fr_subtype)
3970 {
3971 case ENCODE_RELAX_STATE (COND_JUMP, SMALL):
3972 case ENCODE_RELAX_STATE (COND_JUMP, SMALL16):
3973 case ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL):
3974 case ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL16):
3975 /* Don't have to change opcode. */
3976 extension = 1; /* 1 opcode + 1 displacement */
3977 where_to_put_displacement = &opcode[1];
3978 break;
3979
3980 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
3981 extension = 5; /* 2 opcode + 4 displacement */
3982 opcode[1] = opcode[0] + 0x10;
3983 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
3984 where_to_put_displacement = &opcode[2];
3985 break;
3986
3987 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
3988 extension = 4; /* 1 opcode + 4 displacement */
3989 opcode[0] = 0xe9;
3990 where_to_put_displacement = &opcode[1];
3991 break;
3992
3993 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
3994 extension = 3; /* 2 opcode + 2 displacement */
3995 opcode[1] = opcode[0] + 0x10;
3996 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
3997 where_to_put_displacement = &opcode[2];
3998 break;
3999
4000 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
4001 extension = 2; /* 1 opcode + 2 displacement */
4002 opcode[0] = 0xe9;
4003 where_to_put_displacement = &opcode[1];
4004 break;
4005
4006 default:
4007 BAD_CASE (fragP->fr_subtype);
4008 break;
4009 }
4010 /* Now put displacement after opcode. */
4011 md_number_to_chars ((char *) where_to_put_displacement,
4012 (valueT) (displacement_from_opcode_start - extension),
4013 SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
4014 fragP->fr_fix += extension;
4015 }
4016 \f
4017 /* Size of byte displacement jmp. */
4018 int md_short_jump_size = 2;
4019
4020 /* Size of dword displacement jmp. */
4021 int md_long_jump_size = 5;
4022
4023 /* Size of relocation record. */
4024 const int md_reloc_size = 8;
4025
4026 void
4027 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
4028 char *ptr;
4029 addressT from_addr, to_addr;
4030 fragS *frag ATTRIBUTE_UNUSED;
4031 symbolS *to_symbol ATTRIBUTE_UNUSED;
4032 {
4033 offsetT offset;
4034
4035 offset = to_addr - (from_addr + 2);
4036 /* Opcode for byte-disp jump. */
4037 md_number_to_chars (ptr, (valueT) 0xeb, 1);
4038 md_number_to_chars (ptr + 1, (valueT) offset, 1);
4039 }
4040
4041 void
4042 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
4043 char *ptr;
4044 addressT from_addr, to_addr;
4045 fragS *frag ATTRIBUTE_UNUSED;
4046 symbolS *to_symbol ATTRIBUTE_UNUSED;
4047 {
4048 offsetT offset;
4049
4050 offset = to_addr - (from_addr + 5);
4051 md_number_to_chars (ptr, (valueT) 0xe9, 1);
4052 md_number_to_chars (ptr + 1, (valueT) offset, 4);
4053 }
4054 \f
4055 /* Apply a fixup (fixS) to segment data, once it has been determined
4056 by our caller that we have all the info we need to fix it up.
4057
4058 On the 386, immediates, displacements, and data pointers are all in
4059 the same (little-endian) format, so we don't need to care about which
4060 we are handling. */
4061
4062 int
4063 md_apply_fix3 (fixP, valp, seg)
4064 /* The fix we're to put in. */
4065 fixS *fixP;
4066
4067 /* Pointer to the value of the bits. */
4068 valueT *valp;
4069
4070 /* Segment fix is from. */
4071 segT seg ATTRIBUTE_UNUSED;
4072 {
4073 register char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
4074 valueT value = *valp;
4075
4076 #if defined (BFD_ASSEMBLER) && !defined (TE_Mach)
4077 if (fixP->fx_pcrel)
4078 {
4079 switch (fixP->fx_r_type)
4080 {
4081 default:
4082 break;
4083
4084 case BFD_RELOC_32:
4085 fixP->fx_r_type = BFD_RELOC_32_PCREL;
4086 break;
4087 case BFD_RELOC_16:
4088 fixP->fx_r_type = BFD_RELOC_16_PCREL;
4089 break;
4090 case BFD_RELOC_8:
4091 fixP->fx_r_type = BFD_RELOC_8_PCREL;
4092 break;
4093 }
4094 }
4095
4096 /* This is a hack. There should be a better way to handle this.
4097 This covers for the fact that bfd_install_relocation will
4098 subtract the current location (for partial_inplace, PC relative
4099 relocations); see more below. */
4100 if ((fixP->fx_r_type == BFD_RELOC_32_PCREL
4101 || fixP->fx_r_type == BFD_RELOC_16_PCREL
4102 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
4103 && fixP->fx_addsy)
4104 {
4105 #ifndef OBJ_AOUT
4106 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4107 #ifdef TE_PE
4108 || OUTPUT_FLAVOR == bfd_target_coff_flavour
4109 #endif
4110 )
4111 value += fixP->fx_where + fixP->fx_frag->fr_address;
4112 #endif
4113 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4114 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
4115 {
4116 segT fseg = S_GET_SEGMENT (fixP->fx_addsy);
4117
4118 if ((fseg == seg
4119 || (symbol_section_p (fixP->fx_addsy)
4120 && fseg != absolute_section))
4121 && ! S_IS_EXTERNAL (fixP->fx_addsy)
4122 && ! S_IS_WEAK (fixP->fx_addsy)
4123 && S_IS_DEFINED (fixP->fx_addsy)
4124 && ! S_IS_COMMON (fixP->fx_addsy))
4125 {
4126 /* Yes, we add the values in twice. This is because
4127 bfd_perform_relocation subtracts them out again. I think
4128 bfd_perform_relocation is broken, but I don't dare change
4129 it. FIXME. */
4130 value += fixP->fx_where + fixP->fx_frag->fr_address;
4131 }
4132 }
4133 #endif
4134 #if defined (OBJ_COFF) && defined (TE_PE)
4135 /* For some reason, the PE format does not store a section
4136 address offset for a PC relative symbol. */
4137 if (S_GET_SEGMENT (fixP->fx_addsy) != seg)
4138 value += md_pcrel_from (fixP);
4139 #endif
4140 }
4141
4142 /* Fix a few things - the dynamic linker expects certain values here,
4143 and we must not dissappoint it. */
4144 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4145 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4146 && fixP->fx_addsy)
4147 switch (fixP->fx_r_type)
4148 {
4149 case BFD_RELOC_386_PLT32:
4150 case BFD_RELOC_X86_64_PLT32:
4151 /* Make the jump instruction point to the address of the operand. At
4152 runtime we merely add the offset to the actual PLT entry. */
4153 value = -4;
4154 break;
4155 case BFD_RELOC_386_GOTPC:
4156
4157 /* This is tough to explain. We end up with this one if we have
4158 * operands that look like "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal
4159 * here is to obtain the absolute address of the GOT, and it is strongly
4160 * preferable from a performance point of view to avoid using a runtime
4161 * relocation for this. The actual sequence of instructions often look
4162 * something like:
4163 *
4164 * call .L66
4165 * .L66:
4166 * popl %ebx
4167 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
4168 *
4169 * The call and pop essentially return the absolute address of
4170 * the label .L66 and store it in %ebx. The linker itself will
4171 * ultimately change the first operand of the addl so that %ebx points to
4172 * the GOT, but to keep things simple, the .o file must have this operand
4173 * set so that it generates not the absolute address of .L66, but the
4174 * absolute address of itself. This allows the linker itself simply
4175 * treat a GOTPC relocation as asking for a pcrel offset to the GOT to be
4176 * added in, and the addend of the relocation is stored in the operand
4177 * field for the instruction itself.
4178 *
4179 * Our job here is to fix the operand so that it would add the correct
4180 * offset so that %ebx would point to itself. The thing that is tricky is
4181 * that .-.L66 will point to the beginning of the instruction, so we need
4182 * to further modify the operand so that it will point to itself.
4183 * There are other cases where you have something like:
4184 *
4185 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
4186 *
4187 * and here no correction would be required. Internally in the assembler
4188 * we treat operands of this form as not being pcrel since the '.' is
4189 * explicitly mentioned, and I wonder whether it would simplify matters
4190 * to do it this way. Who knows. In earlier versions of the PIC patches,
4191 * the pcrel_adjust field was used to store the correction, but since the
4192 * expression is not pcrel, I felt it would be confusing to do it this
4193 * way. */
4194
4195 value -= 1;
4196 break;
4197 case BFD_RELOC_386_GOT32:
4198 case BFD_RELOC_X86_64_GOT32:
4199 value = 0; /* Fully resolved at runtime. No addend. */
4200 break;
4201 case BFD_RELOC_386_GOTOFF:
4202 case BFD_RELOC_X86_64_GOTPCREL:
4203 break;
4204
4205 case BFD_RELOC_VTABLE_INHERIT:
4206 case BFD_RELOC_VTABLE_ENTRY:
4207 fixP->fx_done = 0;
4208 return 1;
4209
4210 default:
4211 break;
4212 }
4213 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
4214 *valp = value;
4215 #endif /* defined (BFD_ASSEMBLER) && !defined (TE_Mach) */
4216
4217 #ifndef BFD_ASSEMBLER
4218 md_number_to_chars (p, value, fixP->fx_size);
4219 #else
4220 /* Are we finished with this relocation now? */
4221 if (fixP->fx_addsy == 0 && fixP->fx_pcrel == 0)
4222 fixP->fx_done = 1;
4223 else if (use_rela_relocations)
4224 {
4225 fixP->fx_no_overflow = 1;
4226 value = 0;
4227 }
4228 md_number_to_chars (p, value, fixP->fx_size);
4229 #endif
4230
4231 return 1;
4232 }
4233 \f
4234 #define MAX_LITTLENUMS 6
4235
4236 /* Turn the string pointed to by litP into a floating point constant
4237 of type TYPE, and emit the appropriate bytes. The number of
4238 LITTLENUMS emitted is stored in *SIZEP. An error message is
4239 returned, or NULL on OK. */
4240
4241 char *
4242 md_atof (type, litP, sizeP)
4243 int type;
4244 char *litP;
4245 int *sizeP;
4246 {
4247 int prec;
4248 LITTLENUM_TYPE words[MAX_LITTLENUMS];
4249 LITTLENUM_TYPE *wordP;
4250 char *t;
4251
4252 switch (type)
4253 {
4254 case 'f':
4255 case 'F':
4256 prec = 2;
4257 break;
4258
4259 case 'd':
4260 case 'D':
4261 prec = 4;
4262 break;
4263
4264 case 'x':
4265 case 'X':
4266 prec = 5;
4267 break;
4268
4269 default:
4270 *sizeP = 0;
4271 return _("Bad call to md_atof ()");
4272 }
4273 t = atof_ieee (input_line_pointer, type, words);
4274 if (t)
4275 input_line_pointer = t;
4276
4277 *sizeP = prec * sizeof (LITTLENUM_TYPE);
4278 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
4279 the bigendian 386. */
4280 for (wordP = words + prec - 1; prec--;)
4281 {
4282 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
4283 litP += sizeof (LITTLENUM_TYPE);
4284 }
4285 return 0;
4286 }
4287 \f
4288 char output_invalid_buf[8];
4289
4290 static char *
4291 output_invalid (c)
4292 int c;
4293 {
4294 if (isprint (c))
4295 sprintf (output_invalid_buf, "'%c'", c);
4296 else
4297 sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
4298 return output_invalid_buf;
4299 }
4300
4301 /* REG_STRING starts *before* REGISTER_PREFIX. */
4302
4303 static const reg_entry *
4304 parse_register (reg_string, end_op)
4305 char *reg_string;
4306 char **end_op;
4307 {
4308 char *s = reg_string;
4309 char *p;
4310 char reg_name_given[MAX_REG_NAME_SIZE + 1];
4311 const reg_entry *r;
4312
4313 /* Skip possible REGISTER_PREFIX and possible whitespace. */
4314 if (*s == REGISTER_PREFIX)
4315 ++s;
4316
4317 if (is_space_char (*s))
4318 ++s;
4319
4320 p = reg_name_given;
4321 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
4322 {
4323 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
4324 return (const reg_entry *) NULL;
4325 s++;
4326 }
4327
4328 /* For naked regs, make sure that we are not dealing with an identifier.
4329 This prevents confusing an identifier like `eax_var' with register
4330 `eax'. */
4331 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
4332 return (const reg_entry *) NULL;
4333
4334 *end_op = s;
4335
4336 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
4337
4338 /* Handle floating point regs, allowing spaces in the (i) part. */
4339 if (r == i386_regtab /* %st is first entry of table */)
4340 {
4341 if (is_space_char (*s))
4342 ++s;
4343 if (*s == '(')
4344 {
4345 ++s;
4346 if (is_space_char (*s))
4347 ++s;
4348 if (*s >= '0' && *s <= '7')
4349 {
4350 r = &i386_float_regtab[*s - '0'];
4351 ++s;
4352 if (is_space_char (*s))
4353 ++s;
4354 if (*s == ')')
4355 {
4356 *end_op = s + 1;
4357 return r;
4358 }
4359 }
4360 /* We have "%st(" then garbage. */
4361 return (const reg_entry *) NULL;
4362 }
4363 }
4364
4365 return r;
4366 }
4367 \f
4368 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4369 const char *md_shortopts = "kVQ:sq";
4370 #else
4371 const char *md_shortopts = "q";
4372 #endif
4373 struct option md_longopts[] = {
4374 #define OPTION_32 (OPTION_MD_BASE + 0)
4375 {"32", no_argument, NULL, OPTION_32},
4376 #define OPTION_64 (OPTION_MD_BASE + 1)
4377 {"64", no_argument, NULL, OPTION_64},
4378 {NULL, no_argument, NULL, 0}
4379 };
4380 size_t md_longopts_size = sizeof (md_longopts);
4381
4382 int
4383 md_parse_option (c, arg)
4384 int c;
4385 char *arg ATTRIBUTE_UNUSED;
4386 {
4387 switch (c)
4388 {
4389 case 'q':
4390 quiet_warnings = 1;
4391 break;
4392
4393 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4394 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
4395 should be emitted or not. FIXME: Not implemented. */
4396 case 'Q':
4397 break;
4398
4399 /* -V: SVR4 argument to print version ID. */
4400 case 'V':
4401 print_version_id ();
4402 break;
4403
4404 /* -k: Ignore for FreeBSD compatibility. */
4405 case 'k':
4406 break;
4407
4408 case 's':
4409 /* -s: On i386 Solaris, this tells the native assembler to use
4410 .stab instead of .stab.excl. We always use .stab anyhow. */
4411 break;
4412 #endif
4413 #ifdef OBJ_ELF
4414 case OPTION_32:
4415 case OPTION_64:
4416 {
4417 const char **list, **l;
4418
4419 default_arch = c == OPTION_32 ? "i386" : "x86_64";
4420 list = bfd_target_list ();
4421 for (l = list; *l != NULL; l++)
4422 {
4423 if (c == OPTION_32)
4424 {
4425 if (strcmp (*l, "elf32-i386") == 0)
4426 break;
4427 }
4428 else
4429 {
4430 if (strcmp (*l, "elf64-x86-64") == 0)
4431 break;
4432 }
4433 }
4434 if (*l == NULL)
4435 as_fatal (_("No compiled in support for %d bit object file format"),
4436 c == OPTION_32 ? 32 : 64);
4437 free (list);
4438 }
4439 break;
4440 #endif
4441
4442 default:
4443 return 0;
4444 }
4445 return 1;
4446 }
4447
4448 void
4449 md_show_usage (stream)
4450 FILE *stream;
4451 {
4452 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4453 fprintf (stream, _("\
4454 -Q ignored\n\
4455 -V print assembler version number\n\
4456 -k ignored\n\
4457 -q quieten some warnings\n\
4458 -s ignored\n"));
4459 #else
4460 fprintf (stream, _("\
4461 -q quieten some warnings\n"));
4462 #endif
4463 }
4464
4465 #ifdef BFD_ASSEMBLER
4466 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
4467 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
4468
4469 /* Pick the target format to use. */
4470
4471 const char *
4472 i386_target_format ()
4473 {
4474 if (!strcmp (default_arch, "x86_64"))
4475 set_code_flag (CODE_64BIT);
4476 else if (!strcmp (default_arch, "i386"))
4477 set_code_flag (CODE_32BIT);
4478 else
4479 as_fatal (_("Unknown architecture"));
4480 switch (OUTPUT_FLAVOR)
4481 {
4482 #ifdef OBJ_MAYBE_AOUT
4483 case bfd_target_aout_flavour:
4484 return AOUT_TARGET_FORMAT;
4485 #endif
4486 #ifdef OBJ_MAYBE_COFF
4487 case bfd_target_coff_flavour:
4488 return "coff-i386";
4489 #endif
4490 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
4491 case bfd_target_elf_flavour:
4492 {
4493 if (flag_code == CODE_64BIT)
4494 use_rela_relocations = 1;
4495 return flag_code == CODE_64BIT ? "elf64-x86-64" : "elf32-i386";
4496 }
4497 #endif
4498 default:
4499 abort ();
4500 return NULL;
4501 }
4502 }
4503
4504 #endif /* OBJ_MAYBE_ more than one */
4505 #endif /* BFD_ASSEMBLER */
4506 \f
4507 symbolS *
4508 md_undefined_symbol (name)
4509 char *name;
4510 {
4511 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
4512 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
4513 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
4514 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
4515 {
4516 if (!GOT_symbol)
4517 {
4518 if (symbol_find (name))
4519 as_bad (_("GOT already in symbol table"));
4520 GOT_symbol = symbol_new (name, undefined_section,
4521 (valueT) 0, &zero_address_frag);
4522 };
4523 return GOT_symbol;
4524 }
4525 return 0;
4526 }
4527
4528 /* Round up a section size to the appropriate boundary. */
4529
4530 valueT
4531 md_section_align (segment, size)
4532 segT segment ATTRIBUTE_UNUSED;
4533 valueT size;
4534 {
4535 #ifdef BFD_ASSEMBLER
4536 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4537 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
4538 {
4539 /* For a.out, force the section size to be aligned. If we don't do
4540 this, BFD will align it for us, but it will not write out the
4541 final bytes of the section. This may be a bug in BFD, but it is
4542 easier to fix it here since that is how the other a.out targets
4543 work. */
4544 int align;
4545
4546 align = bfd_get_section_alignment (stdoutput, segment);
4547 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
4548 }
4549 #endif
4550 #endif
4551
4552 return size;
4553 }
4554
4555 /* On the i386, PC-relative offsets are relative to the start of the
4556 next instruction. That is, the address of the offset, plus its
4557 size, since the offset is always the last part of the insn. */
4558
4559 long
4560 md_pcrel_from (fixP)
4561 fixS *fixP;
4562 {
4563 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
4564 }
4565
4566 #ifndef I386COFF
4567
4568 static void
4569 s_bss (ignore)
4570 int ignore ATTRIBUTE_UNUSED;
4571 {
4572 register int temp;
4573
4574 temp = get_absolute_expression ();
4575 subseg_set (bss_section, (subsegT) temp);
4576 demand_empty_rest_of_line ();
4577 }
4578
4579 #endif
4580
4581 #ifdef BFD_ASSEMBLER
4582
4583 void
4584 i386_validate_fix (fixp)
4585 fixS *fixp;
4586 {
4587 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
4588 {
4589 /* GOTOFF relocation are nonsense in 64bit mode. */
4590 if (flag_code == CODE_64BIT)
4591 abort();
4592 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
4593 fixp->fx_subsy = 0;
4594 }
4595 }
4596
4597 arelent *
4598 tc_gen_reloc (section, fixp)
4599 asection *section ATTRIBUTE_UNUSED;
4600 fixS *fixp;
4601 {
4602 arelent *rel;
4603 bfd_reloc_code_real_type code;
4604
4605 switch (fixp->fx_r_type)
4606 {
4607 case BFD_RELOC_X86_64_PLT32:
4608 case BFD_RELOC_X86_64_GOT32:
4609 case BFD_RELOC_X86_64_GOTPCREL:
4610 case BFD_RELOC_386_PLT32:
4611 case BFD_RELOC_386_GOT32:
4612 case BFD_RELOC_386_GOTOFF:
4613 case BFD_RELOC_386_GOTPC:
4614 case BFD_RELOC_X86_64_32S:
4615 case BFD_RELOC_RVA:
4616 case BFD_RELOC_VTABLE_ENTRY:
4617 case BFD_RELOC_VTABLE_INHERIT:
4618 code = fixp->fx_r_type;
4619 break;
4620 default:
4621 if (fixp->fx_pcrel)
4622 {
4623 switch (fixp->fx_size)
4624 {
4625 default:
4626 as_bad (_("can not do %d byte pc-relative relocation"),
4627 fixp->fx_size);
4628 code = BFD_RELOC_32_PCREL;
4629 break;
4630 case 1: code = BFD_RELOC_8_PCREL; break;
4631 case 2: code = BFD_RELOC_16_PCREL; break;
4632 case 4: code = BFD_RELOC_32_PCREL; break;
4633 }
4634 }
4635 else
4636 {
4637 switch (fixp->fx_size)
4638 {
4639 default:
4640 as_bad (_("can not do %d byte relocation"), fixp->fx_size);
4641 code = BFD_RELOC_32;
4642 break;
4643 case 1: code = BFD_RELOC_8; break;
4644 case 2: code = BFD_RELOC_16; break;
4645 case 4: code = BFD_RELOC_32; break;
4646 case 8: code = BFD_RELOC_64; break;
4647 }
4648 }
4649 break;
4650 }
4651
4652 if (code == BFD_RELOC_32
4653 && GOT_symbol
4654 && fixp->fx_addsy == GOT_symbol)
4655 {
4656 /* We don't support GOTPC on 64bit targets. */
4657 if (flag_code == CODE_64BIT)
4658 abort();
4659 code = BFD_RELOC_386_GOTPC;
4660 }
4661
4662 rel = (arelent *) xmalloc (sizeof (arelent));
4663 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
4664 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
4665
4666 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
4667 if (!use_rela_relocations)
4668 {
4669 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
4670 vtable entry to be used in the relocation's section offset. */
4671 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
4672 rel->address = fixp->fx_offset;
4673
4674 if (fixp->fx_pcrel)
4675 rel->addend = fixp->fx_addnumber;
4676 else
4677 rel->addend = 0;
4678 }
4679 /* Use the rela in 64bit mode. */
4680 else
4681 {
4682 rel->addend = fixp->fx_offset;
4683 #ifdef OBJ_ELF
4684 /* Ohhh, this is ugly. The problem is that if this is a local global
4685 symbol, the relocation will entirely be performed at link time, not
4686 at assembly time. bfd_perform_reloc doesn't know about this sort
4687 of thing, and as a result we need to fake it out here. */
4688 if ((S_IS_EXTERN (fixp->fx_addsy) || S_IS_WEAK (fixp->fx_addsy))
4689 && !S_IS_COMMON(fixp->fx_addsy))
4690 rel->addend -= symbol_get_bfdsym (fixp->fx_addsy)->value;
4691 #endif
4692 if (fixp->fx_pcrel)
4693 rel->addend -= fixp->fx_size;
4694 }
4695
4696
4697 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
4698 if (rel->howto == NULL)
4699 {
4700 as_bad_where (fixp->fx_file, fixp->fx_line,
4701 _("cannot represent relocation type %s"),
4702 bfd_get_reloc_code_name (code));
4703 /* Set howto to a garbage value so that we can keep going. */
4704 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
4705 assert (rel->howto != NULL);
4706 }
4707
4708 return rel;
4709 }
4710
4711 #else /* ! BFD_ASSEMBLER */
4712
4713 #if (defined(OBJ_AOUT) | defined(OBJ_BOUT))
4714 void
4715 tc_aout_fix_to_chars (where, fixP, segment_address_in_file)
4716 char *where;
4717 fixS *fixP;
4718 relax_addressT segment_address_in_file;
4719 {
4720 /* In: length of relocation (or of address) in chars: 1, 2 or 4.
4721 Out: GNU LD relocation length code: 0, 1, or 2. */
4722
4723 static const unsigned char nbytes_r_length[] = { 42, 0, 1, 42, 2 };
4724 long r_symbolnum;
4725
4726 know (fixP->fx_addsy != NULL);
4727
4728 md_number_to_chars (where,
4729 (valueT) (fixP->fx_frag->fr_address
4730 + fixP->fx_where - segment_address_in_file),
4731 4);
4732
4733 r_symbolnum = (S_IS_DEFINED (fixP->fx_addsy)
4734 ? S_GET_TYPE (fixP->fx_addsy)
4735 : fixP->fx_addsy->sy_number);
4736
4737 where[6] = (r_symbolnum >> 16) & 0x0ff;
4738 where[5] = (r_symbolnum >> 8) & 0x0ff;
4739 where[4] = r_symbolnum & 0x0ff;
4740 where[7] = ((((!S_IS_DEFINED (fixP->fx_addsy)) << 3) & 0x08)
4741 | ((nbytes_r_length[fixP->fx_size] << 1) & 0x06)
4742 | (((fixP->fx_pcrel << 0) & 0x01) & 0x0f));
4743 }
4744
4745 #endif /* OBJ_AOUT or OBJ_BOUT. */
4746
4747 #if defined (I386COFF)
4748
4749 short
4750 tc_coff_fix2rtype (fixP)
4751 fixS *fixP;
4752 {
4753 if (fixP->fx_r_type == R_IMAGEBASE)
4754 return R_IMAGEBASE;
4755
4756 return (fixP->fx_pcrel ?
4757 (fixP->fx_size == 1 ? R_PCRBYTE :
4758 fixP->fx_size == 2 ? R_PCRWORD :
4759 R_PCRLONG) :
4760 (fixP->fx_size == 1 ? R_RELBYTE :
4761 fixP->fx_size == 2 ? R_RELWORD :
4762 R_DIR32));
4763 }
4764
4765 int
4766 tc_coff_sizemachdep (frag)
4767 fragS *frag;
4768 {
4769 if (frag->fr_next)
4770 return (frag->fr_next->fr_address - frag->fr_address);
4771 else
4772 return 0;
4773 }
4774
4775 #endif /* I386COFF */
4776
4777 #endif /* ! BFD_ASSEMBLER */
4778 \f
4779 /* Parse operands using Intel syntax. This implements a recursive descent
4780 parser based on the BNF grammar published in Appendix B of the MASM 6.1
4781 Programmer's Guide.
4782
4783 FIXME: We do not recognize the full operand grammar defined in the MASM
4784 documentation. In particular, all the structure/union and
4785 high-level macro operands are missing.
4786
4787 Uppercase words are terminals, lower case words are non-terminals.
4788 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
4789 bars '|' denote choices. Most grammar productions are implemented in
4790 functions called 'intel_<production>'.
4791
4792 Initial production is 'expr'.
4793
4794 addOp + | -
4795
4796 alpha [a-zA-Z]
4797
4798 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
4799
4800 constant digits [[ radixOverride ]]
4801
4802 dataType BYTE | WORD | DWORD | QWORD | XWORD
4803
4804 digits decdigit
4805 | digits decdigit
4806 | digits hexdigit
4807
4808 decdigit [0-9]
4809
4810 e05 e05 addOp e06
4811 | e06
4812
4813 e06 e06 mulOp e09
4814 | e09
4815
4816 e09 OFFSET e10
4817 | e09 PTR e10
4818 | e09 : e10
4819 | e10
4820
4821 e10 e10 [ expr ]
4822 | e11
4823
4824 e11 ( expr )
4825 | [ expr ]
4826 | constant
4827 | dataType
4828 | id
4829 | $
4830 | register
4831
4832 => expr SHORT e05
4833 | e05
4834
4835 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
4836 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
4837
4838 hexdigit a | b | c | d | e | f
4839 | A | B | C | D | E | F
4840
4841 id alpha
4842 | id alpha
4843 | id decdigit
4844
4845 mulOp * | / | MOD
4846
4847 quote " | '
4848
4849 register specialRegister
4850 | gpRegister
4851 | byteRegister
4852
4853 segmentRegister CS | DS | ES | FS | GS | SS
4854
4855 specialRegister CR0 | CR2 | CR3
4856 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
4857 | TR3 | TR4 | TR5 | TR6 | TR7
4858
4859 We simplify the grammar in obvious places (e.g., register parsing is
4860 done by calling parse_register) and eliminate immediate left recursion
4861 to implement a recursive-descent parser.
4862
4863 expr SHORT e05
4864 | e05
4865
4866 e05 e06 e05'
4867
4868 e05' addOp e06 e05'
4869 | Empty
4870
4871 e06 e09 e06'
4872
4873 e06' mulOp e09 e06'
4874 | Empty
4875
4876 e09 OFFSET e10 e09'
4877 | e10 e09'
4878
4879 e09' PTR e10 e09'
4880 | : e10 e09'
4881 | Empty
4882
4883 e10 e11 e10'
4884
4885 e10' [ expr ] e10'
4886 | Empty
4887
4888 e11 ( expr )
4889 | [ expr ]
4890 | BYTE
4891 | WORD
4892 | DWORD
4893 | QWORD
4894 | XWORD
4895 | .
4896 | $
4897 | register
4898 | id
4899 | constant */
4900
4901 /* Parsing structure for the intel syntax parser. Used to implement the
4902 semantic actions for the operand grammar. */
4903 struct intel_parser_s
4904 {
4905 char *op_string; /* The string being parsed. */
4906 int got_a_float; /* Whether the operand is a float. */
4907 int op_modifier; /* Operand modifier. */
4908 int is_mem; /* 1 if operand is memory reference. */
4909 const reg_entry *reg; /* Last register reference found. */
4910 char *disp; /* Displacement string being built. */
4911 };
4912
4913 static struct intel_parser_s intel_parser;
4914
4915 /* Token structure for parsing intel syntax. */
4916 struct intel_token
4917 {
4918 int code; /* Token code. */
4919 const reg_entry *reg; /* Register entry for register tokens. */
4920 char *str; /* String representation. */
4921 };
4922
4923 static struct intel_token cur_token, prev_token;
4924
4925
4926 /* Token codes for the intel parser. Since T_SHORT is already used
4927 by COFF, undefine it first to prevent a warning. */
4928 #define T_NIL -1
4929 #define T_CONST 1
4930 #define T_REG 2
4931 #define T_BYTE 3
4932 #define T_WORD 4
4933 #define T_DWORD 5
4934 #define T_QWORD 6
4935 #define T_XWORD 7
4936 #undef T_SHORT
4937 #define T_SHORT 8
4938 #define T_OFFSET 9
4939 #define T_PTR 10
4940 #define T_ID 11
4941
4942 /* Prototypes for intel parser functions. */
4943 static int intel_match_token PARAMS ((int code));
4944 static void intel_get_token PARAMS ((void));
4945 static void intel_putback_token PARAMS ((void));
4946 static int intel_expr PARAMS ((void));
4947 static int intel_e05 PARAMS ((void));
4948 static int intel_e05_1 PARAMS ((void));
4949 static int intel_e06 PARAMS ((void));
4950 static int intel_e06_1 PARAMS ((void));
4951 static int intel_e09 PARAMS ((void));
4952 static int intel_e09_1 PARAMS ((void));
4953 static int intel_e10 PARAMS ((void));
4954 static int intel_e10_1 PARAMS ((void));
4955 static int intel_e11 PARAMS ((void));
4956
4957 static int
4958 i386_intel_operand (operand_string, got_a_float)
4959 char *operand_string;
4960 int got_a_float;
4961 {
4962 int ret;
4963 char *p;
4964
4965 /* Initialize token holders. */
4966 cur_token.code = prev_token.code = T_NIL;
4967 cur_token.reg = prev_token.reg = NULL;
4968 cur_token.str = prev_token.str = NULL;
4969
4970 /* Initialize parser structure. */
4971 p = intel_parser.op_string = (char *)malloc (strlen (operand_string) + 1);
4972 if (p == NULL)
4973 abort ();
4974 strcpy (intel_parser.op_string, operand_string);
4975 intel_parser.got_a_float = got_a_float;
4976 intel_parser.op_modifier = -1;
4977 intel_parser.is_mem = 0;
4978 intel_parser.reg = NULL;
4979 intel_parser.disp = (char *)malloc (strlen (operand_string) + 1);
4980 if (intel_parser.disp == NULL)
4981 abort ();
4982 intel_parser.disp[0] = '\0';
4983
4984 /* Read the first token and start the parser. */
4985 intel_get_token ();
4986 ret = intel_expr ();
4987
4988 if (ret)
4989 {
4990 /* If we found a memory reference, hand it over to i386_displacement
4991 to fill in the rest of the operand fields. */
4992 if (intel_parser.is_mem)
4993 {
4994 if ((i.mem_operands == 1
4995 && (current_templates->start->opcode_modifier & IsString) == 0)
4996 || i.mem_operands == 2)
4997 {
4998 as_bad (_("too many memory references for '%s'"),
4999 current_templates->start->name);
5000 ret = 0;
5001 }
5002 else
5003 {
5004 char *s = intel_parser.disp;
5005 i.mem_operands++;
5006
5007 /* Add the displacement expression. */
5008 if (*s != '\0')
5009 ret = i386_displacement (s, s + strlen (s))
5010 && i386_index_check (s);
5011 }
5012 }
5013
5014 /* Constant and OFFSET expressions are handled by i386_immediate. */
5015 else if (intel_parser.op_modifier == OFFSET_FLAT
5016 || intel_parser.reg == NULL)
5017 ret = i386_immediate (intel_parser.disp);
5018 }
5019
5020 free (p);
5021 free (intel_parser.disp);
5022
5023 return ret;
5024 }
5025
5026 /* expr SHORT e05
5027 | e05 */
5028 static int
5029 intel_expr ()
5030 {
5031 /* expr SHORT e05 */
5032 if (cur_token.code == T_SHORT)
5033 {
5034 intel_parser.op_modifier = SHORT;
5035 intel_match_token (T_SHORT);
5036
5037 return (intel_e05 ());
5038 }
5039
5040 /* expr e05 */
5041 else
5042 return intel_e05 ();
5043 }
5044
5045 /* e05 e06 e05'
5046
5047 e05' addOp e06 e05'
5048 | Empty */
5049 static int
5050 intel_e05 ()
5051 {
5052 return (intel_e06 () && intel_e05_1 ());
5053 }
5054
5055 static int
5056 intel_e05_1 ()
5057 {
5058 /* e05' addOp e06 e05' */
5059 if (cur_token.code == '+' || cur_token.code == '-')
5060 {
5061 strcat (intel_parser.disp, cur_token.str);
5062 intel_match_token (cur_token.code);
5063
5064 return (intel_e06 () && intel_e05_1 ());
5065 }
5066
5067 /* e05' Empty */
5068 else
5069 return 1;
5070 }
5071
5072 /* e06 e09 e06'
5073
5074 e06' mulOp e09 e06'
5075 | Empty */
5076 static int
5077 intel_e06 ()
5078 {
5079 return (intel_e09 () && intel_e06_1 ());
5080 }
5081
5082 static int
5083 intel_e06_1 ()
5084 {
5085 /* e06' mulOp e09 e06' */
5086 if (cur_token.code == '*' || cur_token.code == '/')
5087 {
5088 strcat (intel_parser.disp, cur_token.str);
5089 intel_match_token (cur_token.code);
5090
5091 return (intel_e09 () && intel_e06_1 ());
5092 }
5093
5094 /* e06' Empty */
5095 else
5096 return 1;
5097 }
5098
5099 /* e09 OFFSET e10 e09'
5100 | e10 e09'
5101
5102 e09' PTR e10 e09'
5103 | : e10 e09'
5104 | Empty */
5105 static int
5106 intel_e09 ()
5107 {
5108 /* e09 OFFSET e10 e09' */
5109 if (cur_token.code == T_OFFSET)
5110 {
5111 intel_parser.is_mem = 0;
5112 intel_parser.op_modifier = OFFSET_FLAT;
5113 intel_match_token (T_OFFSET);
5114
5115 return (intel_e10 () && intel_e09_1 ());
5116 }
5117
5118 /* e09 e10 e09' */
5119 else
5120 return (intel_e10 () && intel_e09_1 ());
5121 }
5122
5123 static int
5124 intel_e09_1 ()
5125 {
5126 /* e09' PTR e10 e09' */
5127 if (cur_token.code == T_PTR)
5128 {
5129 if (prev_token.code == T_BYTE)
5130 i.suffix = BYTE_MNEM_SUFFIX;
5131
5132 else if (prev_token.code == T_WORD)
5133 {
5134 if (intel_parser.got_a_float == 2) /* "fi..." */
5135 i.suffix = SHORT_MNEM_SUFFIX;
5136 else
5137 i.suffix = WORD_MNEM_SUFFIX;
5138 }
5139
5140 else if (prev_token.code == T_DWORD)
5141 {
5142 if (intel_parser.got_a_float == 1) /* "f..." */
5143 i.suffix = SHORT_MNEM_SUFFIX;
5144 else
5145 i.suffix = LONG_MNEM_SUFFIX;
5146 }
5147
5148 else if (prev_token.code == T_QWORD)
5149 {
5150 if (intel_parser.got_a_float == 1) /* "f..." */
5151 i.suffix = LONG_MNEM_SUFFIX;
5152 else
5153 i.suffix = QWORD_MNEM_SUFFIX;
5154 }
5155
5156 else if (prev_token.code == T_XWORD)
5157 i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
5158
5159 else
5160 {
5161 as_bad (_("Unknown operand modifier `%s'\n"), prev_token.str);
5162 return 0;
5163 }
5164
5165 intel_match_token (T_PTR);
5166
5167 return (intel_e10 () && intel_e09_1 ());
5168 }
5169
5170 /* e09 : e10 e09' */
5171 else if (cur_token.code == ':')
5172 {
5173 /* Mark as a memory operand only if it's not already known to be an
5174 offset expression. */
5175 if (intel_parser.op_modifier != OFFSET_FLAT)
5176 intel_parser.is_mem = 1;
5177
5178 return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
5179 }
5180
5181 /* e09' Empty */
5182 else
5183 return 1;
5184 }
5185
5186 /* e10 e11 e10'
5187
5188 e10' [ expr ] e10'
5189 | Empty */
5190 static int
5191 intel_e10 ()
5192 {
5193 return (intel_e11 () && intel_e10_1 ());
5194 }
5195
5196 static int
5197 intel_e10_1 ()
5198 {
5199 /* e10' [ expr ] e10' */
5200 if (cur_token.code == '[')
5201 {
5202 intel_match_token ('[');
5203
5204 /* Mark as a memory operand only if it's not already known to be an
5205 offset expression. If it's an offset expression, we need to keep
5206 the brace in. */
5207 if (intel_parser.op_modifier != OFFSET_FLAT)
5208 intel_parser.is_mem = 1;
5209 else
5210 strcat (intel_parser.disp, "[");
5211
5212 /* Add a '+' to the displacement string if necessary. */
5213 if (*intel_parser.disp != '\0'
5214 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5215 strcat (intel_parser.disp, "+");
5216
5217 if (intel_expr () && intel_match_token (']'))
5218 {
5219 /* Preserve brackets when the operand is an offset expression. */
5220 if (intel_parser.op_modifier == OFFSET_FLAT)
5221 strcat (intel_parser.disp, "]");
5222
5223 return intel_e10_1 ();
5224 }
5225 else
5226 return 0;
5227 }
5228
5229 /* e10' Empty */
5230 else
5231 return 1;
5232 }
5233
5234 /* e11 ( expr )
5235 | [ expr ]
5236 | BYTE
5237 | WORD
5238 | DWORD
5239 | QWORD
5240 | XWORD
5241 | $
5242 | .
5243 | register
5244 | id
5245 | constant */
5246 static int
5247 intel_e11 ()
5248 {
5249 /* e11 ( expr ) */
5250 if (cur_token.code == '(')
5251 {
5252 intel_match_token ('(');
5253 strcat (intel_parser.disp, "(");
5254
5255 if (intel_expr () && intel_match_token (')'))
5256 {
5257 strcat (intel_parser.disp, ")");
5258 return 1;
5259 }
5260 else
5261 return 0;
5262 }
5263
5264 /* e11 [ expr ] */
5265 else if (cur_token.code == '[')
5266 {
5267 intel_match_token ('[');
5268
5269 /* Mark as a memory operand only if it's not already known to be an
5270 offset expression. If it's an offset expression, we need to keep
5271 the brace in. */
5272 if (intel_parser.op_modifier != OFFSET_FLAT)
5273 intel_parser.is_mem = 1;
5274 else
5275 strcat (intel_parser.disp, "[");
5276
5277 /* Operands for jump/call inside brackets denote absolute addresses. */
5278 if (current_templates->start->opcode_modifier & Jump
5279 || current_templates->start->opcode_modifier & JumpDword
5280 || current_templates->start->opcode_modifier & JumpByte
5281 || current_templates->start->opcode_modifier & JumpInterSegment)
5282 i.types[this_operand] |= JumpAbsolute;
5283
5284 /* Add a '+' to the displacement string if necessary. */
5285 if (*intel_parser.disp != '\0'
5286 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5287 strcat (intel_parser.disp, "+");
5288
5289 if (intel_expr () && intel_match_token (']'))
5290 {
5291 /* Preserve brackets when the operand is an offset expression. */
5292 if (intel_parser.op_modifier == OFFSET_FLAT)
5293 strcat (intel_parser.disp, "]");
5294
5295 return 1;
5296 }
5297 else
5298 return 0;
5299 }
5300
5301 /* e11 BYTE
5302 | WORD
5303 | DWORD
5304 | QWORD
5305 | XWORD */
5306 else if (cur_token.code == T_BYTE
5307 || cur_token.code == T_WORD
5308 || cur_token.code == T_DWORD
5309 || cur_token.code == T_QWORD
5310 || cur_token.code == T_XWORD)
5311 {
5312 intel_match_token (cur_token.code);
5313
5314 return 1;
5315 }
5316
5317 /* e11 $
5318 | . */
5319 else if (cur_token.code == '$' || cur_token.code == '.')
5320 {
5321 strcat (intel_parser.disp, cur_token.str);
5322 intel_match_token (cur_token.code);
5323
5324 /* Mark as a memory operand only if it's not already known to be an
5325 offset expression. */
5326 if (intel_parser.op_modifier != OFFSET_FLAT)
5327 intel_parser.is_mem = 1;
5328
5329 return 1;
5330 }
5331
5332 /* e11 register */
5333 else if (cur_token.code == T_REG)
5334 {
5335 const reg_entry *reg = intel_parser.reg = cur_token.reg;
5336
5337 intel_match_token (T_REG);
5338
5339 /* Check for segment change. */
5340 if (cur_token.code == ':')
5341 {
5342 if (reg->reg_type & (SReg2 | SReg3))
5343 {
5344 switch (reg->reg_num)
5345 {
5346 case 0:
5347 i.seg[i.mem_operands] = &es;
5348 break;
5349 case 1:
5350 i.seg[i.mem_operands] = &cs;
5351 break;
5352 case 2:
5353 i.seg[i.mem_operands] = &ss;
5354 break;
5355 case 3:
5356 i.seg[i.mem_operands] = &ds;
5357 break;
5358 case 4:
5359 i.seg[i.mem_operands] = &fs;
5360 break;
5361 case 5:
5362 i.seg[i.mem_operands] = &gs;
5363 break;
5364 }
5365 }
5366 else
5367 {
5368 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
5369 return 0;
5370 }
5371 }
5372
5373 /* Not a segment register. Check for register scaling. */
5374 else if (cur_token.code == '*')
5375 {
5376 if (!intel_parser.is_mem)
5377 {
5378 as_bad (_("Register scaling only allowed in memory operands."));
5379 return 0;
5380 }
5381
5382 /* What follows must be a valid scale. */
5383 if (intel_match_token ('*')
5384 && strchr ("01248", *cur_token.str))
5385 {
5386 i.index_reg = reg;
5387 i.types[this_operand] |= BaseIndex;
5388
5389 /* Set the scale after setting the register (otherwise,
5390 i386_scale will complain) */
5391 i386_scale (cur_token.str);
5392 intel_match_token (T_CONST);
5393 }
5394 else
5395 {
5396 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
5397 cur_token.str);
5398 return 0;
5399 }
5400 }
5401
5402 /* No scaling. If this is a memory operand, the register is either a
5403 base register (first occurrence) or an index register (second
5404 occurrence). */
5405 else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3)))
5406 {
5407 if (i.base_reg && i.index_reg)
5408 {
5409 as_bad (_("Too many register references in memory operand.\n"));
5410 return 0;
5411 }
5412
5413 if (i.base_reg == NULL)
5414 i.base_reg = reg;
5415 else
5416 i.index_reg = reg;
5417
5418 i.types[this_operand] |= BaseIndex;
5419 }
5420
5421 /* Offset modifier. Add the register to the displacement string to be
5422 parsed as an immediate expression after we're done. */
5423 else if (intel_parser.op_modifier == OFFSET_FLAT)
5424 strcat (intel_parser.disp, reg->reg_name);
5425
5426 /* It's neither base nor index nor offset. */
5427 else
5428 {
5429 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
5430 i.op[this_operand].regs = reg;
5431 i.reg_operands++;
5432 }
5433
5434 /* Since registers are not part of the displacement string (except
5435 when we're parsing offset operands), we may need to remove any
5436 preceding '+' from the displacement string. */
5437 if (*intel_parser.disp != '\0'
5438 && intel_parser.op_modifier != OFFSET_FLAT)
5439 {
5440 char *s = intel_parser.disp;
5441 s += strlen (s) - 1;
5442 if (*s == '+')
5443 *s = '\0';
5444 }
5445
5446 return 1;
5447 }
5448
5449 /* e11 id */
5450 else if (cur_token.code == T_ID)
5451 {
5452 /* Add the identifier to the displacement string. */
5453 strcat (intel_parser.disp, cur_token.str);
5454 intel_match_token (T_ID);
5455
5456 /* The identifier represents a memory reference only if it's not
5457 preceded by an offset modifier. */
5458 if (intel_parser.op_modifier != OFFSET_FLAT)
5459 intel_parser.is_mem = 1;
5460
5461 return 1;
5462 }
5463
5464 /* e11 constant */
5465 else if (cur_token.code == T_CONST
5466 || cur_token.code == '-'
5467 || cur_token.code == '+')
5468 {
5469 char *save_str;
5470
5471 /* Allow constants that start with `+' or `-'. */
5472 if (cur_token.code == '-' || cur_token.code == '+')
5473 {
5474 strcat (intel_parser.disp, cur_token.str);
5475 intel_match_token (cur_token.code);
5476 if (cur_token.code != T_CONST)
5477 {
5478 as_bad (_("Syntax error. Expecting a constant. Got `%s'.\n"),
5479 cur_token.str);
5480 return 0;
5481 }
5482 }
5483
5484 save_str = (char *)malloc (strlen (cur_token.str) + 1);
5485 if (save_str == NULL)
5486 abort ();
5487 strcpy (save_str, cur_token.str);
5488
5489 /* Get the next token to check for register scaling. */
5490 intel_match_token (cur_token.code);
5491
5492 /* Check if this constant is a scaling factor for an index register. */
5493 if (cur_token.code == '*')
5494 {
5495 if (intel_match_token ('*') && cur_token.code == T_REG)
5496 {
5497 if (!intel_parser.is_mem)
5498 {
5499 as_bad (_("Register scaling only allowed in memory operands."));
5500 return 0;
5501 }
5502
5503 /* The constant is followed by `* reg', so it must be
5504 a valid scale. */
5505 if (strchr ("01248", *save_str))
5506 {
5507 i.index_reg = cur_token.reg;
5508 i.types[this_operand] |= BaseIndex;
5509
5510 /* Set the scale after setting the register (otherwise,
5511 i386_scale will complain) */
5512 i386_scale (save_str);
5513 intel_match_token (T_REG);
5514
5515 /* Since registers are not part of the displacement
5516 string, we may need to remove any preceding '+' from
5517 the displacement string. */
5518 if (*intel_parser.disp != '\0')
5519 {
5520 char *s = intel_parser.disp;
5521 s += strlen (s) - 1;
5522 if (*s == '+')
5523 *s = '\0';
5524 }
5525
5526 free (save_str);
5527
5528 return 1;
5529 }
5530 else
5531 return 0;
5532 }
5533
5534 /* The constant was not used for register scaling. Since we have
5535 already consumed the token following `*' we now need to put it
5536 back in the stream. */
5537 else
5538 intel_putback_token ();
5539 }
5540
5541 /* Add the constant to the displacement string. */
5542 strcat (intel_parser.disp, save_str);
5543 free (save_str);
5544
5545 return 1;
5546 }
5547
5548 as_bad (_("Unrecognized token '%s'"), cur_token.str);
5549 return 0;
5550 }
5551
5552 /* Match the given token against cur_token. If they match, read the next
5553 token from the operand string. */
5554 static int
5555 intel_match_token (code)
5556 int code;
5557 {
5558 if (cur_token.code == code)
5559 {
5560 intel_get_token ();
5561 return 1;
5562 }
5563 else
5564 {
5565 as_bad (_("Unexpected token `%s'\n"), cur_token.str);
5566 return 0;
5567 }
5568 }
5569
5570 /* Read a new token from intel_parser.op_string and store it in cur_token. */
5571 static void
5572 intel_get_token ()
5573 {
5574 char *end_op;
5575 const reg_entry *reg;
5576 struct intel_token new_token;
5577
5578 new_token.code = T_NIL;
5579 new_token.reg = NULL;
5580 new_token.str = NULL;
5581
5582 /* Free the memory allocated to the previous token and move
5583 cur_token to prev_token. */
5584 if (prev_token.str)
5585 free (prev_token.str);
5586
5587 prev_token = cur_token;
5588
5589 /* Skip whitespace. */
5590 while (is_space_char (*intel_parser.op_string))
5591 intel_parser.op_string++;
5592
5593 /* Return an empty token if we find nothing else on the line. */
5594 if (*intel_parser.op_string == '\0')
5595 {
5596 cur_token = new_token;
5597 return;
5598 }
5599
5600 /* The new token cannot be larger than the remainder of the operand
5601 string. */
5602 new_token.str = (char *)malloc (strlen (intel_parser.op_string) + 1);
5603 if (new_token.str == NULL)
5604 abort ();
5605 new_token.str[0] = '\0';
5606
5607 if (strchr ("0123456789", *intel_parser.op_string))
5608 {
5609 char *p = new_token.str;
5610 char *q = intel_parser.op_string;
5611 new_token.code = T_CONST;
5612
5613 /* Allow any kind of identifier char to encompass floating point and
5614 hexadecimal numbers. */
5615 while (is_identifier_char (*q))
5616 *p++ = *q++;
5617 *p = '\0';
5618
5619 /* Recognize special symbol names [0-9][bf]. */
5620 if (strlen (intel_parser.op_string) == 2
5621 && (intel_parser.op_string[1] == 'b'
5622 || intel_parser.op_string[1] == 'f'))
5623 new_token.code = T_ID;
5624 }
5625
5626 else if (strchr ("+-/*:[]()", *intel_parser.op_string))
5627 {
5628 new_token.code = *intel_parser.op_string;
5629 new_token.str[0] = *intel_parser.op_string;
5630 new_token.str[1] = '\0';
5631 }
5632
5633 else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
5634 && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
5635 {
5636 new_token.code = T_REG;
5637 new_token.reg = reg;
5638
5639 if (*intel_parser.op_string == REGISTER_PREFIX)
5640 {
5641 new_token.str[0] = REGISTER_PREFIX;
5642 new_token.str[1] = '\0';
5643 }
5644
5645 strcat (new_token.str, reg->reg_name);
5646 }
5647
5648 else if (is_identifier_char (*intel_parser.op_string))
5649 {
5650 char *p = new_token.str;
5651 char *q = intel_parser.op_string;
5652
5653 /* A '.' or '$' followed by an identifier char is an identifier.
5654 Otherwise, it's operator '.' followed by an expression. */
5655 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
5656 {
5657 new_token.code = *q;
5658 new_token.str[0] = *q;
5659 new_token.str[1] = '\0';
5660 }
5661 else
5662 {
5663 while (is_identifier_char (*q) || *q == '@')
5664 *p++ = *q++;
5665 *p = '\0';
5666
5667 if (strcasecmp (new_token.str, "BYTE") == 0)
5668 new_token.code = T_BYTE;
5669
5670 else if (strcasecmp (new_token.str, "WORD") == 0)
5671 new_token.code = T_WORD;
5672
5673 else if (strcasecmp (new_token.str, "DWORD") == 0)
5674 new_token.code = T_DWORD;
5675
5676 else if (strcasecmp (new_token.str, "QWORD") == 0)
5677 new_token.code = T_QWORD;
5678
5679 else if (strcasecmp (new_token.str, "XWORD") == 0)
5680 new_token.code = T_XWORD;
5681
5682 else if (strcasecmp (new_token.str, "PTR") == 0)
5683 new_token.code = T_PTR;
5684
5685 else if (strcasecmp (new_token.str, "SHORT") == 0)
5686 new_token.code = T_SHORT;
5687
5688 else if (strcasecmp (new_token.str, "OFFSET") == 0)
5689 {
5690 new_token.code = T_OFFSET;
5691
5692 /* ??? This is not mentioned in the MASM grammar but gcc
5693 makes use of it with -mintel-syntax. OFFSET may be
5694 followed by FLAT: */
5695 if (strncasecmp (q, " FLAT:", 6) == 0)
5696 strcat (new_token.str, " FLAT:");
5697 }
5698
5699 /* ??? This is not mentioned in the MASM grammar. */
5700 else if (strcasecmp (new_token.str, "FLAT") == 0)
5701 new_token.code = T_OFFSET;
5702
5703 else
5704 new_token.code = T_ID;
5705 }
5706 }
5707
5708 else
5709 as_bad (_("Unrecognized token `%s'\n"), intel_parser.op_string);
5710
5711 intel_parser.op_string += strlen (new_token.str);
5712 cur_token = new_token;
5713 }
5714
5715 /* Put cur_token back into the token stream and make cur_token point to
5716 prev_token. */
5717 static void
5718 intel_putback_token ()
5719 {
5720 intel_parser.op_string -= strlen (cur_token.str);
5721 free (cur_token.str);
5722 cur_token = prev_token;
5723
5724 /* Forget prev_token. */
5725 prev_token.code = T_NIL;
5726 prev_token.reg = NULL;
5727 prev_token.str = NULL;
5728 }