2002-01-29 Daniel Jacobowitz <drow@mvista.com>
[deliverable/binutils-gdb.git] / gas / config / tc-i386.c
1 /* i386.c -- Assemble code for the Intel 80386
2 Copyright 1989, 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
3 2000, 2001
4 Free Software Foundation, Inc.
5
6 This file is part of GAS, the GNU Assembler.
7
8 GAS is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2, or (at your option)
11 any later version.
12
13 GAS is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
17
18 You should have received a copy of the GNU General Public License
19 along with GAS; see the file COPYING. If not, write to the Free
20 Software Foundation, 59 Temple Place - Suite 330, Boston, MA
21 02111-1307, USA. */
22
23 /* Intel 80386 machine specific gas.
24 Written by Eliot Dresselhaus (eliot@mgm.mit.edu).
25 x86_64 support by Jan Hubicka (jh@suse.cz)
26 Bugs & suggestions are completely welcome. This is free software.
27 Please help us make it better. */
28
29 #include "as.h"
30 #include "safe-ctype.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 #ifndef true
52 #define true 1
53 #endif
54 #ifndef false
55 #define false 0
56 #endif
57
58 static unsigned int mode_from_disp_size PARAMS ((unsigned int));
59 static int fits_in_signed_byte PARAMS ((offsetT));
60 static int fits_in_unsigned_byte PARAMS ((offsetT));
61 static int fits_in_unsigned_word PARAMS ((offsetT));
62 static int fits_in_signed_word PARAMS ((offsetT));
63 static int fits_in_unsigned_long PARAMS ((offsetT));
64 static int fits_in_signed_long PARAMS ((offsetT));
65 static int smallest_imm_type PARAMS ((offsetT));
66 static offsetT offset_in_range PARAMS ((offsetT, int));
67 static int add_prefix PARAMS ((unsigned int));
68 static void set_code_flag PARAMS ((int));
69 static void set_16bit_gcc_code_flag PARAMS ((int));
70 static void set_intel_syntax PARAMS ((int));
71 static void set_cpu_arch PARAMS ((int));
72
73 #ifdef BFD_ASSEMBLER
74 static bfd_reloc_code_real_type reloc
75 PARAMS ((int, int, int, bfd_reloc_code_real_type));
76 #define RELOC_ENUM enum bfd_reloc_code_real
77 #else
78 #define RELOC_ENUM int
79 #endif
80
81 #ifndef DEFAULT_ARCH
82 #define DEFAULT_ARCH "i386"
83 #endif
84 static const char *default_arch = DEFAULT_ARCH;
85
86 /* 'md_assemble ()' gathers together information and puts it into a
87 i386_insn. */
88
89 union i386_op
90 {
91 expressionS *disps;
92 expressionS *imms;
93 const reg_entry *regs;
94 };
95
96 struct _i386_insn
97 {
98 /* TM holds the template for the insn were currently assembling. */
99 template tm;
100
101 /* SUFFIX holds the instruction mnemonic suffix if given.
102 (e.g. 'l' for 'movl') */
103 char suffix;
104
105 /* OPERANDS gives the number of given operands. */
106 unsigned int operands;
107
108 /* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number
109 of given register, displacement, memory operands and immediate
110 operands. */
111 unsigned int reg_operands, disp_operands, mem_operands, imm_operands;
112
113 /* TYPES [i] is the type (see above #defines) which tells us how to
114 use OP[i] for the corresponding operand. */
115 unsigned int types[MAX_OPERANDS];
116
117 /* Displacement expression, immediate expression, or register for each
118 operand. */
119 union i386_op op[MAX_OPERANDS];
120
121 /* Flags for operands. */
122 unsigned int flags[MAX_OPERANDS];
123 #define Operand_PCrel 1
124
125 /* Relocation type for operand */
126 RELOC_ENUM reloc[MAX_OPERANDS];
127
128 /* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode
129 the base index byte below. */
130 const reg_entry *base_reg;
131 const reg_entry *index_reg;
132 unsigned int log2_scale_factor;
133
134 /* SEG gives the seg_entries of this insn. They are zero unless
135 explicit segment overrides are given. */
136 const seg_entry *seg[2];
137
138 /* PREFIX holds all the given prefix opcodes (usually null).
139 PREFIXES is the number of prefix opcodes. */
140 unsigned int prefixes;
141 unsigned char prefix[MAX_PREFIXES];
142
143 /* RM and SIB are the modrm byte and the sib byte where the
144 addressing modes of this insn are encoded. */
145
146 modrm_byte rm;
147 rex_byte rex;
148 sib_byte sib;
149 };
150
151 typedef struct _i386_insn i386_insn;
152
153 /* List of chars besides those in app.c:symbol_chars that can start an
154 operand. Used to prevent the scrubber eating vital white-space. */
155 #ifdef LEX_AT
156 const char extra_symbol_chars[] = "*%-(@";
157 #else
158 const char extra_symbol_chars[] = "*%-(";
159 #endif
160
161 /* This array holds the chars that always start a comment. If the
162 pre-processor is disabled, these aren't very useful. */
163 #if defined (TE_I386AIX) || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) && ! defined (TE_LINUX) && !defined(TE_FreeBSD) && !defined(TE_NetBSD))
164 /* Putting '/' here makes it impossible to use the divide operator.
165 However, we need it for compatibility with SVR4 systems. */
166 const char comment_chars[] = "#/";
167 #define PREFIX_SEPARATOR '\\'
168 #else
169 const char comment_chars[] = "#";
170 #define PREFIX_SEPARATOR '/'
171 #endif
172
173 /* This array holds the chars that only start a comment at the beginning of
174 a line. If the line seems to have the form '# 123 filename'
175 .line and .file directives will appear in the pre-processed output.
176 Note that input_file.c hand checks for '#' at the beginning of the
177 first line of the input file. This is because the compiler outputs
178 #NO_APP at the beginning of its output.
179 Also note that comments started like this one will always work if
180 '/' isn't otherwise defined. */
181 #if defined (TE_I386AIX) || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) && ! defined (TE_LINUX) && !defined(TE_FreeBSD) && !defined(TE_NetBSD))
182 const char line_comment_chars[] = "";
183 #else
184 const char line_comment_chars[] = "/";
185 #endif
186
187 const char line_separator_chars[] = ";";
188
189 /* Chars that can be used to separate mant from exp in floating point
190 nums. */
191 const char EXP_CHARS[] = "eE";
192
193 /* Chars that mean this number is a floating point constant
194 As in 0f12.456
195 or 0d1.2345e12. */
196 const char FLT_CHARS[] = "fFdDxX";
197
198 /* Tables for lexical analysis. */
199 static char mnemonic_chars[256];
200 static char register_chars[256];
201 static char operand_chars[256];
202 static char identifier_chars[256];
203 static char digit_chars[256];
204
205 /* Lexical macros. */
206 #define is_mnemonic_char(x) (mnemonic_chars[(unsigned char) x])
207 #define is_operand_char(x) (operand_chars[(unsigned char) x])
208 #define is_register_char(x) (register_chars[(unsigned char) x])
209 #define is_space_char(x) ((x) == ' ')
210 #define is_identifier_char(x) (identifier_chars[(unsigned char) x])
211 #define is_digit_char(x) (digit_chars[(unsigned char) x])
212
213 /* All non-digit non-letter charcters that may occur in an operand. */
214 static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
215
216 /* md_assemble() always leaves the strings it's passed unaltered. To
217 effect this we maintain a stack of saved characters that we've smashed
218 with '\0's (indicating end of strings for various sub-fields of the
219 assembler instruction). */
220 static char save_stack[32];
221 static char *save_stack_p;
222 #define END_STRING_AND_SAVE(s) \
223 do { *save_stack_p++ = *(s); *(s) = '\0'; } while (0)
224 #define RESTORE_END_STRING(s) \
225 do { *(s) = *--save_stack_p; } while (0)
226
227 /* The instruction we're assembling. */
228 static i386_insn i;
229
230 /* Possible templates for current insn. */
231 static const templates *current_templates;
232
233 /* Per instruction expressionS buffers: 2 displacements & 2 immediate max. */
234 static expressionS disp_expressions[2], im_expressions[2];
235
236 /* Current operand we are working on. */
237 static int this_operand;
238
239 /* We support four different modes. FLAG_CODE variable is used to distinguish
240 these. */
241
242 enum flag_code {
243 CODE_32BIT,
244 CODE_16BIT,
245 CODE_64BIT };
246 #define NUM_FLAG_CODE ((int) CODE_64BIT + 1)
247
248 static enum flag_code flag_code;
249 static int use_rela_relocations = 0;
250
251 /* The names used to print error messages. */
252 static const char *flag_code_names[] =
253 {
254 "32",
255 "16",
256 "64"
257 };
258
259 /* 1 for intel syntax,
260 0 if att syntax. */
261 static int intel_syntax = 0;
262
263 /* 1 if register prefix % not required. */
264 static int allow_naked_reg = 0;
265
266 /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
267 leave, push, and pop instructions so that gcc has the same stack
268 frame as in 32 bit mode. */
269 static char stackop_size = '\0';
270
271 /* Non-zero to quieten some warnings. */
272 static int quiet_warnings = 0;
273
274 /* CPU name. */
275 static const char *cpu_arch_name = NULL;
276
277 /* CPU feature flags. */
278 static unsigned int cpu_arch_flags = CpuUnknownFlags|CpuNo64;
279
280 /* If set, conditional jumps are not automatically promoted to handle
281 larger than a byte offset. */
282 static unsigned int no_cond_jump_promotion = 0;
283
284 /* Interface to relax_segment.
285 There are 3 major relax states for 386 jump insns because the
286 different types of jumps add different sizes to frags when we're
287 figuring out what sort of jump to choose to reach a given label. */
288
289 /* Types. */
290 #define UNCOND_JUMP 0
291 #define COND_JUMP 1
292 #define COND_JUMP86 2
293
294 /* Sizes. */
295 #define CODE16 1
296 #define SMALL 0
297 #define SMALL16 (SMALL|CODE16)
298 #define BIG 2
299 #define BIG16 (BIG|CODE16)
300
301 #ifndef INLINE
302 #ifdef __GNUC__
303 #define INLINE __inline__
304 #else
305 #define INLINE
306 #endif
307 #endif
308
309 #define ENCODE_RELAX_STATE(type, size) \
310 ((relax_substateT) (((type) << 2) | (size)))
311 #define TYPE_FROM_RELAX_STATE(s) \
312 ((s) >> 2)
313 #define DISP_SIZE_FROM_RELAX_STATE(s) \
314 ((((s) & 3) == BIG ? 4 : (((s) & 3) == BIG16 ? 2 : 1)))
315
316 /* This table is used by relax_frag to promote short jumps to long
317 ones where necessary. SMALL (short) jumps may be promoted to BIG
318 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
319 don't allow a short jump in a 32 bit code segment to be promoted to
320 a 16 bit offset jump because it's slower (requires data size
321 prefix), and doesn't work, unless the destination is in the bottom
322 64k of the code segment (The top 16 bits of eip are zeroed). */
323
324 const relax_typeS md_relax_table[] =
325 {
326 /* The fields are:
327 1) most positive reach of this state,
328 2) most negative reach of this state,
329 3) how many bytes this mode will have in the variable part of the frag
330 4) which index into the table to try if we can't fit into this one. */
331
332 /* UNCOND_JUMP states. */
333 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
334 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
335 /* dword jmp adds 4 bytes to frag:
336 0 extra opcode bytes, 4 displacement bytes. */
337 {0, 0, 4, 0},
338 /* word jmp adds 2 byte2 to frag:
339 0 extra opcode bytes, 2 displacement bytes. */
340 {0, 0, 2, 0},
341
342 /* COND_JUMP states. */
343 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
344 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
345 /* dword conditionals adds 5 bytes to frag:
346 1 extra opcode byte, 4 displacement bytes. */
347 {0, 0, 5, 0},
348 /* word conditionals add 3 bytes to frag:
349 1 extra opcode byte, 2 displacement bytes. */
350 {0, 0, 3, 0},
351
352 /* COND_JUMP86 states. */
353 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG)},
354 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG16)},
355 /* dword conditionals adds 5 bytes to frag:
356 1 extra opcode byte, 4 displacement bytes. */
357 {0, 0, 5, 0},
358 /* word conditionals add 4 bytes to frag:
359 1 displacement byte and a 3 byte long branch insn. */
360 {0, 0, 4, 0}
361 };
362
363 static const arch_entry cpu_arch[] = {
364 {"i8086", Cpu086 },
365 {"i186", Cpu086|Cpu186 },
366 {"i286", Cpu086|Cpu186|Cpu286 },
367 {"i386", Cpu086|Cpu186|Cpu286|Cpu386 },
368 {"i486", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486 },
369 {"i586", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
370 {"i686", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
371 {"pentium", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
372 {"pentiumpro",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
373 {"pentium4", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX|CpuSSE|CpuSSE2 },
374 {"k6", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow },
375 {"athlon", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuMMX|Cpu3dnow },
376 {"sledgehammer",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuSledgehammer|CpuMMX|Cpu3dnow|CpuSSE|CpuSSE2 },
377 {NULL, 0 }
378 };
379
380 void
381 i386_align_code (fragP, count)
382 fragS *fragP;
383 int count;
384 {
385 /* Various efficient no-op patterns for aligning code labels.
386 Note: Don't try to assemble the instructions in the comments.
387 0L and 0w are not legal. */
388 static const char f32_1[] =
389 {0x90}; /* nop */
390 static const char f32_2[] =
391 {0x89,0xf6}; /* movl %esi,%esi */
392 static const char f32_3[] =
393 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
394 static const char f32_4[] =
395 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
396 static const char f32_5[] =
397 {0x90, /* nop */
398 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
399 static const char f32_6[] =
400 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
401 static const char f32_7[] =
402 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
403 static const char f32_8[] =
404 {0x90, /* nop */
405 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
406 static const char f32_9[] =
407 {0x89,0xf6, /* movl %esi,%esi */
408 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
409 static const char f32_10[] =
410 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
411 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
412 static const char f32_11[] =
413 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
414 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
415 static const char f32_12[] =
416 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
417 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
418 static const char f32_13[] =
419 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
420 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
421 static const char f32_14[] =
422 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
423 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
424 static const char f32_15[] =
425 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
426 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
427 static const char f16_3[] =
428 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
429 static const char f16_4[] =
430 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
431 static const char f16_5[] =
432 {0x90, /* nop */
433 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
434 static const char f16_6[] =
435 {0x89,0xf6, /* mov %si,%si */
436 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
437 static const char f16_7[] =
438 {0x8d,0x74,0x00, /* lea 0(%si),%si */
439 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
440 static const char f16_8[] =
441 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
442 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
443 static const char *const f32_patt[] = {
444 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
445 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
446 };
447 static const char *const f16_patt[] = {
448 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
449 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
450 };
451
452 /* ??? We can't use these fillers for x86_64, since they often kills the
453 upper halves. Solve later. */
454 if (flag_code == CODE_64BIT)
455 count = 1;
456
457 if (count > 0 && count <= 15)
458 {
459 if (flag_code == CODE_16BIT)
460 {
461 memcpy (fragP->fr_literal + fragP->fr_fix,
462 f16_patt[count - 1], count);
463 if (count > 8)
464 /* Adjust jump offset. */
465 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
466 }
467 else
468 memcpy (fragP->fr_literal + fragP->fr_fix,
469 f32_patt[count - 1], count);
470 fragP->fr_var = count;
471 }
472 }
473
474 static char *output_invalid PARAMS ((int c));
475 static int i386_operand PARAMS ((char *operand_string));
476 static int i386_intel_operand PARAMS ((char *operand_string, int got_a_float));
477 static const reg_entry *parse_register PARAMS ((char *reg_string,
478 char **end_op));
479
480 #ifndef I386COFF
481 static void s_bss PARAMS ((int));
482 #endif
483
484 symbolS *GOT_symbol; /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
485
486 static INLINE unsigned int
487 mode_from_disp_size (t)
488 unsigned int t;
489 {
490 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
491 }
492
493 static INLINE int
494 fits_in_signed_byte (num)
495 offsetT num;
496 {
497 return (num >= -128) && (num <= 127);
498 }
499
500 static INLINE int
501 fits_in_unsigned_byte (num)
502 offsetT num;
503 {
504 return (num & 0xff) == num;
505 }
506
507 static INLINE int
508 fits_in_unsigned_word (num)
509 offsetT num;
510 {
511 return (num & 0xffff) == num;
512 }
513
514 static INLINE int
515 fits_in_signed_word (num)
516 offsetT num;
517 {
518 return (-32768 <= num) && (num <= 32767);
519 }
520 static INLINE int
521 fits_in_signed_long (num)
522 offsetT num ATTRIBUTE_UNUSED;
523 {
524 #ifndef BFD64
525 return 1;
526 #else
527 return (!(((offsetT) -1 << 31) & num)
528 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
529 #endif
530 } /* fits_in_signed_long() */
531 static INLINE int
532 fits_in_unsigned_long (num)
533 offsetT num ATTRIBUTE_UNUSED;
534 {
535 #ifndef BFD64
536 return 1;
537 #else
538 return (num & (((offsetT) 2 << 31) - 1)) == num;
539 #endif
540 } /* fits_in_unsigned_long() */
541
542 static int
543 smallest_imm_type (num)
544 offsetT num;
545 {
546 if (cpu_arch_flags != (Cpu086 | Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64))
547 {
548 /* This code is disabled on the 486 because all the Imm1 forms
549 in the opcode table are slower on the i486. They're the
550 versions with the implicitly specified single-position
551 displacement, which has another syntax if you really want to
552 use that form. */
553 if (num == 1)
554 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
555 }
556 return (fits_in_signed_byte (num)
557 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
558 : fits_in_unsigned_byte (num)
559 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
560 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
561 ? (Imm16 | Imm32 | Imm32S | Imm64)
562 : fits_in_signed_long (num)
563 ? (Imm32 | Imm32S | Imm64)
564 : fits_in_unsigned_long (num)
565 ? (Imm32 | Imm64)
566 : Imm64);
567 }
568
569 static offsetT
570 offset_in_range (val, size)
571 offsetT val;
572 int size;
573 {
574 addressT mask;
575
576 switch (size)
577 {
578 case 1: mask = ((addressT) 1 << 8) - 1; break;
579 case 2: mask = ((addressT) 1 << 16) - 1; break;
580 case 4: mask = ((addressT) 2 << 31) - 1; break;
581 #ifdef BFD64
582 case 8: mask = ((addressT) 2 << 63) - 1; break;
583 #endif
584 default: abort ();
585 }
586
587 /* If BFD64, sign extend val. */
588 if (!use_rela_relocations)
589 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
590 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
591
592 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
593 {
594 char buf1[40], buf2[40];
595
596 sprint_value (buf1, val);
597 sprint_value (buf2, val & mask);
598 as_warn (_("%s shortened to %s"), buf1, buf2);
599 }
600 return val & mask;
601 }
602
603 /* Returns 0 if attempting to add a prefix where one from the same
604 class already exists, 1 if non rep/repne added, 2 if rep/repne
605 added. */
606 static int
607 add_prefix (prefix)
608 unsigned int prefix;
609 {
610 int ret = 1;
611 int q;
612
613 if (prefix >= 0x40 && prefix < 0x50 && flag_code == CODE_64BIT)
614 q = REX_PREFIX;
615 else
616 switch (prefix)
617 {
618 default:
619 abort ();
620
621 case CS_PREFIX_OPCODE:
622 case DS_PREFIX_OPCODE:
623 case ES_PREFIX_OPCODE:
624 case FS_PREFIX_OPCODE:
625 case GS_PREFIX_OPCODE:
626 case SS_PREFIX_OPCODE:
627 q = SEG_PREFIX;
628 break;
629
630 case REPNE_PREFIX_OPCODE:
631 case REPE_PREFIX_OPCODE:
632 ret = 2;
633 /* fall thru */
634 case LOCK_PREFIX_OPCODE:
635 q = LOCKREP_PREFIX;
636 break;
637
638 case FWAIT_OPCODE:
639 q = WAIT_PREFIX;
640 break;
641
642 case ADDR_PREFIX_OPCODE:
643 q = ADDR_PREFIX;
644 break;
645
646 case DATA_PREFIX_OPCODE:
647 q = DATA_PREFIX;
648 break;
649 }
650
651 if (i.prefix[q])
652 {
653 as_bad (_("same type of prefix used twice"));
654 return 0;
655 }
656
657 i.prefixes += 1;
658 i.prefix[q] = prefix;
659 return ret;
660 }
661
662 static void
663 set_code_flag (value)
664 int value;
665 {
666 flag_code = value;
667 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
668 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
669 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
670 {
671 as_bad (_("64bit mode not supported on this CPU."));
672 }
673 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
674 {
675 as_bad (_("32bit mode not supported on this CPU."));
676 }
677 stackop_size = '\0';
678 }
679
680 static void
681 set_16bit_gcc_code_flag (new_code_flag)
682 int new_code_flag;
683 {
684 flag_code = new_code_flag;
685 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
686 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
687 stackop_size = 'l';
688 }
689
690 static void
691 set_intel_syntax (syntax_flag)
692 int syntax_flag;
693 {
694 /* Find out if register prefixing is specified. */
695 int ask_naked_reg = 0;
696
697 SKIP_WHITESPACE ();
698 if (! is_end_of_line[(unsigned char) *input_line_pointer])
699 {
700 char *string = input_line_pointer;
701 int e = get_symbol_end ();
702
703 if (strcmp (string, "prefix") == 0)
704 ask_naked_reg = 1;
705 else if (strcmp (string, "noprefix") == 0)
706 ask_naked_reg = -1;
707 else
708 as_bad (_("bad argument to syntax directive."));
709 *input_line_pointer = e;
710 }
711 demand_empty_rest_of_line ();
712
713 intel_syntax = syntax_flag;
714
715 if (ask_naked_reg == 0)
716 {
717 #ifdef BFD_ASSEMBLER
718 allow_naked_reg = (intel_syntax
719 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
720 #else
721 /* Conservative default. */
722 allow_naked_reg = 0;
723 #endif
724 }
725 else
726 allow_naked_reg = (ask_naked_reg < 0);
727 }
728
729 static void
730 set_cpu_arch (dummy)
731 int dummy ATTRIBUTE_UNUSED;
732 {
733 SKIP_WHITESPACE ();
734
735 if (! is_end_of_line[(unsigned char) *input_line_pointer])
736 {
737 char *string = input_line_pointer;
738 int e = get_symbol_end ();
739 int i;
740
741 for (i = 0; cpu_arch[i].name; i++)
742 {
743 if (strcmp (string, cpu_arch[i].name) == 0)
744 {
745 cpu_arch_name = cpu_arch[i].name;
746 cpu_arch_flags = (cpu_arch[i].flags
747 | (flag_code == CODE_64BIT ? Cpu64 : CpuNo64));
748 break;
749 }
750 }
751 if (!cpu_arch[i].name)
752 as_bad (_("no such architecture: `%s'"), string);
753
754 *input_line_pointer = e;
755 }
756 else
757 as_bad (_("missing cpu architecture"));
758
759 no_cond_jump_promotion = 0;
760 if (*input_line_pointer == ','
761 && ! is_end_of_line[(unsigned char) input_line_pointer[1]])
762 {
763 char *string = ++input_line_pointer;
764 int e = get_symbol_end ();
765
766 if (strcmp (string, "nojumps") == 0)
767 no_cond_jump_promotion = 1;
768 else if (strcmp (string, "jumps") == 0)
769 ;
770 else
771 as_bad (_("no such architecture modifier: `%s'"), string);
772
773 *input_line_pointer = e;
774 }
775
776 demand_empty_rest_of_line ();
777 }
778
779 const pseudo_typeS md_pseudo_table[] =
780 {
781 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
782 {"align", s_align_bytes, 0},
783 #else
784 {"align", s_align_ptwo, 0},
785 #endif
786 {"arch", set_cpu_arch, 0},
787 #ifndef I386COFF
788 {"bss", s_bss, 0},
789 #endif
790 {"ffloat", float_cons, 'f'},
791 {"dfloat", float_cons, 'd'},
792 {"tfloat", float_cons, 'x'},
793 {"value", cons, 2},
794 {"noopt", s_ignore, 0},
795 {"optim", s_ignore, 0},
796 {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
797 {"code16", set_code_flag, CODE_16BIT},
798 {"code32", set_code_flag, CODE_32BIT},
799 {"code64", set_code_flag, CODE_64BIT},
800 {"intel_syntax", set_intel_syntax, 1},
801 {"att_syntax", set_intel_syntax, 0},
802 {"file", dwarf2_directive_file, 0},
803 {"loc", dwarf2_directive_loc, 0},
804 {0, 0, 0}
805 };
806
807 /* For interface with expression (). */
808 extern char *input_line_pointer;
809
810 /* Hash table for instruction mnemonic lookup. */
811 static struct hash_control *op_hash;
812
813 /* Hash table for register lookup. */
814 static struct hash_control *reg_hash;
815 \f
816 #ifdef BFD_ASSEMBLER
817 unsigned long
818 i386_mach ()
819 {
820 if (!strcmp (default_arch, "x86_64"))
821 return bfd_mach_x86_64;
822 else if (!strcmp (default_arch, "i386"))
823 return bfd_mach_i386_i386;
824 else
825 as_fatal (_("Unknown architecture"));
826 }
827 #endif
828 \f
829 void
830 md_begin ()
831 {
832 const char *hash_err;
833
834 /* Initialize op_hash hash table. */
835 op_hash = hash_new ();
836
837 {
838 register const template *optab;
839 register templates *core_optab;
840
841 /* Setup for loop. */
842 optab = i386_optab;
843 core_optab = (templates *) xmalloc (sizeof (templates));
844 core_optab->start = optab;
845
846 while (1)
847 {
848 ++optab;
849 if (optab->name == NULL
850 || strcmp (optab->name, (optab - 1)->name) != 0)
851 {
852 /* different name --> ship out current template list;
853 add to hash table; & begin anew. */
854 core_optab->end = optab;
855 hash_err = hash_insert (op_hash,
856 (optab - 1)->name,
857 (PTR) core_optab);
858 if (hash_err)
859 {
860 as_fatal (_("Internal Error: Can't hash %s: %s"),
861 (optab - 1)->name,
862 hash_err);
863 }
864 if (optab->name == NULL)
865 break;
866 core_optab = (templates *) xmalloc (sizeof (templates));
867 core_optab->start = optab;
868 }
869 }
870 }
871
872 /* Initialize reg_hash hash table. */
873 reg_hash = hash_new ();
874 {
875 register const reg_entry *regtab;
876
877 for (regtab = i386_regtab;
878 regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
879 regtab++)
880 {
881 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
882 if (hash_err)
883 as_fatal (_("Internal Error: Can't hash %s: %s"),
884 regtab->reg_name,
885 hash_err);
886 }
887 }
888
889 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
890 {
891 register int c;
892 register char *p;
893
894 for (c = 0; c < 256; c++)
895 {
896 if (ISDIGIT (c))
897 {
898 digit_chars[c] = c;
899 mnemonic_chars[c] = c;
900 register_chars[c] = c;
901 operand_chars[c] = c;
902 }
903 else if (ISLOWER (c))
904 {
905 mnemonic_chars[c] = c;
906 register_chars[c] = c;
907 operand_chars[c] = c;
908 }
909 else if (ISUPPER (c))
910 {
911 mnemonic_chars[c] = TOLOWER (c);
912 register_chars[c] = mnemonic_chars[c];
913 operand_chars[c] = c;
914 }
915
916 if (ISALPHA (c) || ISDIGIT (c))
917 identifier_chars[c] = c;
918 else if (c >= 128)
919 {
920 identifier_chars[c] = c;
921 operand_chars[c] = c;
922 }
923 }
924
925 #ifdef LEX_AT
926 identifier_chars['@'] = '@';
927 #endif
928 digit_chars['-'] = '-';
929 identifier_chars['_'] = '_';
930 identifier_chars['.'] = '.';
931
932 for (p = operand_special_chars; *p != '\0'; p++)
933 operand_chars[(unsigned char) *p] = *p;
934 }
935
936 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
937 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
938 {
939 record_alignment (text_section, 2);
940 record_alignment (data_section, 2);
941 record_alignment (bss_section, 2);
942 }
943 #endif
944 }
945
946 void
947 i386_print_statistics (file)
948 FILE *file;
949 {
950 hash_print_statistics (file, "i386 opcode", op_hash);
951 hash_print_statistics (file, "i386 register", reg_hash);
952 }
953 \f
954 #ifdef DEBUG386
955
956 /* Debugging routines for md_assemble. */
957 static void pi PARAMS ((char *, i386_insn *));
958 static void pte PARAMS ((template *));
959 static void pt PARAMS ((unsigned int));
960 static void pe PARAMS ((expressionS *));
961 static void ps PARAMS ((symbolS *));
962
963 static void
964 pi (line, x)
965 char *line;
966 i386_insn *x;
967 {
968 unsigned int i;
969
970 fprintf (stdout, "%s: template ", line);
971 pte (&x->tm);
972 fprintf (stdout, " address: base %s index %s scale %x\n",
973 x->base_reg ? x->base_reg->reg_name : "none",
974 x->index_reg ? x->index_reg->reg_name : "none",
975 x->log2_scale_factor);
976 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
977 x->rm.mode, x->rm.reg, x->rm.regmem);
978 fprintf (stdout, " sib: base %x index %x scale %x\n",
979 x->sib.base, x->sib.index, x->sib.scale);
980 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
981 x->rex.mode64, x->rex.extX, x->rex.extY, x->rex.extZ);
982 for (i = 0; i < x->operands; i++)
983 {
984 fprintf (stdout, " #%d: ", i + 1);
985 pt (x->types[i]);
986 fprintf (stdout, "\n");
987 if (x->types[i]
988 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
989 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
990 if (x->types[i] & Imm)
991 pe (x->op[i].imms);
992 if (x->types[i] & Disp)
993 pe (x->op[i].disps);
994 }
995 }
996
997 static void
998 pte (t)
999 template *t;
1000 {
1001 unsigned int i;
1002 fprintf (stdout, " %d operands ", t->operands);
1003 fprintf (stdout, "opcode %x ", t->base_opcode);
1004 if (t->extension_opcode != None)
1005 fprintf (stdout, "ext %x ", t->extension_opcode);
1006 if (t->opcode_modifier & D)
1007 fprintf (stdout, "D");
1008 if (t->opcode_modifier & W)
1009 fprintf (stdout, "W");
1010 fprintf (stdout, "\n");
1011 for (i = 0; i < t->operands; i++)
1012 {
1013 fprintf (stdout, " #%d type ", i + 1);
1014 pt (t->operand_types[i]);
1015 fprintf (stdout, "\n");
1016 }
1017 }
1018
1019 static void
1020 pe (e)
1021 expressionS *e;
1022 {
1023 fprintf (stdout, " operation %d\n", e->X_op);
1024 fprintf (stdout, " add_number %ld (%lx)\n",
1025 (long) e->X_add_number, (long) e->X_add_number);
1026 if (e->X_add_symbol)
1027 {
1028 fprintf (stdout, " add_symbol ");
1029 ps (e->X_add_symbol);
1030 fprintf (stdout, "\n");
1031 }
1032 if (e->X_op_symbol)
1033 {
1034 fprintf (stdout, " op_symbol ");
1035 ps (e->X_op_symbol);
1036 fprintf (stdout, "\n");
1037 }
1038 }
1039
1040 static void
1041 ps (s)
1042 symbolS *s;
1043 {
1044 fprintf (stdout, "%s type %s%s",
1045 S_GET_NAME (s),
1046 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
1047 segment_name (S_GET_SEGMENT (s)));
1048 }
1049
1050 struct type_name
1051 {
1052 unsigned int mask;
1053 char *tname;
1054 }
1055
1056 type_names[] =
1057 {
1058 { Reg8, "r8" },
1059 { Reg16, "r16" },
1060 { Reg32, "r32" },
1061 { Reg64, "r64" },
1062 { Imm8, "i8" },
1063 { Imm8S, "i8s" },
1064 { Imm16, "i16" },
1065 { Imm32, "i32" },
1066 { Imm32S, "i32s" },
1067 { Imm64, "i64" },
1068 { Imm1, "i1" },
1069 { BaseIndex, "BaseIndex" },
1070 { Disp8, "d8" },
1071 { Disp16, "d16" },
1072 { Disp32, "d32" },
1073 { Disp32S, "d32s" },
1074 { Disp64, "d64" },
1075 { InOutPortReg, "InOutPortReg" },
1076 { ShiftCount, "ShiftCount" },
1077 { Control, "control reg" },
1078 { Test, "test reg" },
1079 { Debug, "debug reg" },
1080 { FloatReg, "FReg" },
1081 { FloatAcc, "FAcc" },
1082 { SReg2, "SReg2" },
1083 { SReg3, "SReg3" },
1084 { Acc, "Acc" },
1085 { JumpAbsolute, "Jump Absolute" },
1086 { RegMMX, "rMMX" },
1087 { RegXMM, "rXMM" },
1088 { EsSeg, "es" },
1089 { 0, "" }
1090 };
1091
1092 static void
1093 pt (t)
1094 unsigned int t;
1095 {
1096 register struct type_name *ty;
1097
1098 for (ty = type_names; ty->mask; ty++)
1099 if (t & ty->mask)
1100 fprintf (stdout, "%s, ", ty->tname);
1101 fflush (stdout);
1102 }
1103
1104 #endif /* DEBUG386 */
1105 \f
1106 int
1107 tc_i386_force_relocation (fixp)
1108 struct fix *fixp;
1109 {
1110 #ifdef BFD_ASSEMBLER
1111 if (fixp->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1112 || fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1113 return 1;
1114 return 0;
1115 #else
1116 /* For COFF. */
1117 return fixp->fx_r_type == 7;
1118 #endif
1119 }
1120
1121 #ifdef BFD_ASSEMBLER
1122
1123 static bfd_reloc_code_real_type
1124 reloc (size, pcrel, sign, other)
1125 int size;
1126 int pcrel;
1127 int sign;
1128 bfd_reloc_code_real_type other;
1129 {
1130 if (other != NO_RELOC)
1131 return other;
1132
1133 if (pcrel)
1134 {
1135 if (!sign)
1136 as_bad (_("There are no unsigned pc-relative relocations"));
1137 switch (size)
1138 {
1139 case 1: return BFD_RELOC_8_PCREL;
1140 case 2: return BFD_RELOC_16_PCREL;
1141 case 4: return BFD_RELOC_32_PCREL;
1142 }
1143 as_bad (_("can not do %d byte pc-relative relocation"), size);
1144 }
1145 else
1146 {
1147 if (sign)
1148 switch (size)
1149 {
1150 case 4: return BFD_RELOC_X86_64_32S;
1151 }
1152 else
1153 switch (size)
1154 {
1155 case 1: return BFD_RELOC_8;
1156 case 2: return BFD_RELOC_16;
1157 case 4: return BFD_RELOC_32;
1158 case 8: return BFD_RELOC_64;
1159 }
1160 as_bad (_("can not do %s %d byte relocation"),
1161 sign ? "signed" : "unsigned", size);
1162 }
1163
1164 abort ();
1165 return BFD_RELOC_NONE;
1166 }
1167
1168 /* Here we decide which fixups can be adjusted to make them relative to
1169 the beginning of the section instead of the symbol. Basically we need
1170 to make sure that the dynamic relocations are done correctly, so in
1171 some cases we force the original symbol to be used. */
1172
1173 int
1174 tc_i386_fix_adjustable (fixP)
1175 fixS *fixP;
1176 {
1177 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1178 /* Prevent all adjustments to global symbols, or else dynamic
1179 linking will not work correctly. */
1180 if (S_IS_EXTERNAL (fixP->fx_addsy)
1181 || S_IS_WEAK (fixP->fx_addsy))
1182 return 0;
1183 #endif
1184 /* adjust_reloc_syms doesn't know about the GOT. */
1185 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1186 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1187 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1188 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1189 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1190 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
1191 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1192 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1193 return 0;
1194 return 1;
1195 }
1196 #else
1197 #define reloc(SIZE,PCREL,SIGN,OTHER) 0
1198 #define BFD_RELOC_16 0
1199 #define BFD_RELOC_32 0
1200 #define BFD_RELOC_16_PCREL 0
1201 #define BFD_RELOC_32_PCREL 0
1202 #define BFD_RELOC_386_PLT32 0
1203 #define BFD_RELOC_386_GOT32 0
1204 #define BFD_RELOC_386_GOTOFF 0
1205 #define BFD_RELOC_X86_64_PLT32 0
1206 #define BFD_RELOC_X86_64_GOT32 0
1207 #define BFD_RELOC_X86_64_GOTPCREL 0
1208 #endif
1209
1210 static int intel_float_operand PARAMS ((char *mnemonic));
1211
1212 static int
1213 intel_float_operand (mnemonic)
1214 char *mnemonic;
1215 {
1216 if (mnemonic[0] == 'f' && mnemonic[1] == 'i')
1217 return 2;
1218
1219 if (mnemonic[0] == 'f')
1220 return 1;
1221
1222 return 0;
1223 }
1224
1225 /* This is the guts of the machine-dependent assembler. LINE points to a
1226 machine dependent instruction. This function is supposed to emit
1227 the frags/bytes it assembles to. */
1228
1229 void
1230 md_assemble (line)
1231 char *line;
1232 {
1233 /* Points to template once we've found it. */
1234 const template *t;
1235
1236 int j;
1237
1238 char mnemonic[MAX_MNEM_SIZE];
1239
1240 /* Initialize globals. */
1241 memset (&i, '\0', sizeof (i));
1242 for (j = 0; j < MAX_OPERANDS; j++)
1243 i.reloc[j] = NO_RELOC;
1244 memset (disp_expressions, '\0', sizeof (disp_expressions));
1245 memset (im_expressions, '\0', sizeof (im_expressions));
1246 save_stack_p = save_stack;
1247
1248 /* First parse an instruction mnemonic & call i386_operand for the operands.
1249 We assume that the scrubber has arranged it so that line[0] is the valid
1250 start of a (possibly prefixed) mnemonic. */
1251 {
1252 char *l = line;
1253 char *token_start = l;
1254 char *mnem_p;
1255
1256 /* Non-zero if we found a prefix only acceptable with string insns. */
1257 const char *expecting_string_instruction = NULL;
1258
1259 while (1)
1260 {
1261 mnem_p = mnemonic;
1262 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1263 {
1264 mnem_p++;
1265 if (mnem_p >= mnemonic + sizeof (mnemonic))
1266 {
1267 as_bad (_("no such instruction: `%s'"), token_start);
1268 return;
1269 }
1270 l++;
1271 }
1272 if (!is_space_char (*l)
1273 && *l != END_OF_INSN
1274 && *l != PREFIX_SEPARATOR
1275 && *l != ',')
1276 {
1277 as_bad (_("invalid character %s in mnemonic"),
1278 output_invalid (*l));
1279 return;
1280 }
1281 if (token_start == l)
1282 {
1283 if (*l == PREFIX_SEPARATOR)
1284 as_bad (_("expecting prefix; got nothing"));
1285 else
1286 as_bad (_("expecting mnemonic; got nothing"));
1287 return;
1288 }
1289
1290 /* Look up instruction (or prefix) via hash table. */
1291 current_templates = hash_find (op_hash, mnemonic);
1292
1293 if (*l != END_OF_INSN
1294 && (! is_space_char (*l) || l[1] != END_OF_INSN)
1295 && current_templates
1296 && (current_templates->start->opcode_modifier & IsPrefix))
1297 {
1298 /* If we are in 16-bit mode, do not allow addr16 or data16.
1299 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1300 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1301 && (((current_templates->start->opcode_modifier & Size32) != 0)
1302 ^ (flag_code == CODE_16BIT)))
1303 {
1304 as_bad (_("redundant %s prefix"),
1305 current_templates->start->name);
1306 return;
1307 }
1308 /* Add prefix, checking for repeated prefixes. */
1309 switch (add_prefix (current_templates->start->base_opcode))
1310 {
1311 case 0:
1312 return;
1313 case 2:
1314 expecting_string_instruction = current_templates->start->name;
1315 break;
1316 }
1317 /* Skip past PREFIX_SEPARATOR and reset token_start. */
1318 token_start = ++l;
1319 }
1320 else
1321 break;
1322 }
1323
1324 if (!current_templates)
1325 {
1326 /* See if we can get a match by trimming off a suffix. */
1327 switch (mnem_p[-1])
1328 {
1329 case WORD_MNEM_SUFFIX:
1330 case BYTE_MNEM_SUFFIX:
1331 case QWORD_MNEM_SUFFIX:
1332 i.suffix = mnem_p[-1];
1333 mnem_p[-1] = '\0';
1334 current_templates = hash_find (op_hash, mnemonic);
1335 break;
1336 case SHORT_MNEM_SUFFIX:
1337 case LONG_MNEM_SUFFIX:
1338 if (!intel_syntax)
1339 {
1340 i.suffix = mnem_p[-1];
1341 mnem_p[-1] = '\0';
1342 current_templates = hash_find (op_hash, mnemonic);
1343 }
1344 break;
1345
1346 /* Intel Syntax. */
1347 case 'd':
1348 if (intel_syntax)
1349 {
1350 if (intel_float_operand (mnemonic))
1351 i.suffix = SHORT_MNEM_SUFFIX;
1352 else
1353 i.suffix = LONG_MNEM_SUFFIX;
1354 mnem_p[-1] = '\0';
1355 current_templates = hash_find (op_hash, mnemonic);
1356 }
1357 break;
1358 }
1359 if (!current_templates)
1360 {
1361 as_bad (_("no such instruction: `%s'"), token_start);
1362 return;
1363 }
1364 }
1365
1366 if (current_templates->start->opcode_modifier & (Jump | JumpByte))
1367 {
1368 /* Check for a branch hint. We allow ",pt" and ",pn" for
1369 predict taken and predict not taken respectively.
1370 I'm not sure that branch hints actually do anything on loop
1371 and jcxz insns (JumpByte) for current Pentium4 chips. They
1372 may work in the future and it doesn't hurt to accept them
1373 now. */
1374 if (l[0] == ',' && l[1] == 'p')
1375 {
1376 if (l[2] == 't')
1377 {
1378 if (! add_prefix (DS_PREFIX_OPCODE))
1379 return;
1380 l += 3;
1381 }
1382 else if (l[2] == 'n')
1383 {
1384 if (! add_prefix (CS_PREFIX_OPCODE))
1385 return;
1386 l += 3;
1387 }
1388 }
1389 }
1390 /* Any other comma loses. */
1391 if (*l == ',')
1392 {
1393 as_bad (_("invalid character %s in mnemonic"),
1394 output_invalid (*l));
1395 return;
1396 }
1397
1398 /* Check if instruction is supported on specified architecture. */
1399 if ((current_templates->start->cpu_flags & ~(Cpu64 | CpuNo64))
1400 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64)))
1401 {
1402 as_warn (_("`%s' is not supported on `%s'"),
1403 current_templates->start->name, cpu_arch_name);
1404 }
1405 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
1406 {
1407 as_warn (_("use .code16 to ensure correct addressing mode"));
1408 }
1409
1410 /* Check for rep/repne without a string instruction. */
1411 if (expecting_string_instruction
1412 && !(current_templates->start->opcode_modifier & IsString))
1413 {
1414 as_bad (_("expecting string instruction after `%s'"),
1415 expecting_string_instruction);
1416 return;
1417 }
1418
1419 /* There may be operands to parse. */
1420 if (*l != END_OF_INSN)
1421 {
1422 /* 1 if operand is pending after ','. */
1423 unsigned int expecting_operand = 0;
1424
1425 /* Non-zero if operand parens not balanced. */
1426 unsigned int paren_not_balanced;
1427
1428 do
1429 {
1430 /* Skip optional white space before operand. */
1431 if (is_space_char (*l))
1432 ++l;
1433 if (!is_operand_char (*l) && *l != END_OF_INSN)
1434 {
1435 as_bad (_("invalid character %s before operand %d"),
1436 output_invalid (*l),
1437 i.operands + 1);
1438 return;
1439 }
1440 token_start = l; /* after white space */
1441 paren_not_balanced = 0;
1442 while (paren_not_balanced || *l != ',')
1443 {
1444 if (*l == END_OF_INSN)
1445 {
1446 if (paren_not_balanced)
1447 {
1448 if (!intel_syntax)
1449 as_bad (_("unbalanced parenthesis in operand %d."),
1450 i.operands + 1);
1451 else
1452 as_bad (_("unbalanced brackets in operand %d."),
1453 i.operands + 1);
1454 return;
1455 }
1456 else
1457 break; /* we are done */
1458 }
1459 else if (!is_operand_char (*l) && !is_space_char (*l))
1460 {
1461 as_bad (_("invalid character %s in operand %d"),
1462 output_invalid (*l),
1463 i.operands + 1);
1464 return;
1465 }
1466 if (!intel_syntax)
1467 {
1468 if (*l == '(')
1469 ++paren_not_balanced;
1470 if (*l == ')')
1471 --paren_not_balanced;
1472 }
1473 else
1474 {
1475 if (*l == '[')
1476 ++paren_not_balanced;
1477 if (*l == ']')
1478 --paren_not_balanced;
1479 }
1480 l++;
1481 }
1482 if (l != token_start)
1483 { /* Yes, we've read in another operand. */
1484 unsigned int operand_ok;
1485 this_operand = i.operands++;
1486 if (i.operands > MAX_OPERANDS)
1487 {
1488 as_bad (_("spurious operands; (%d operands/instruction max)"),
1489 MAX_OPERANDS);
1490 return;
1491 }
1492 /* Now parse operand adding info to 'i' as we go along. */
1493 END_STRING_AND_SAVE (l);
1494
1495 if (intel_syntax)
1496 operand_ok =
1497 i386_intel_operand (token_start,
1498 intel_float_operand (mnemonic));
1499 else
1500 operand_ok = i386_operand (token_start);
1501
1502 RESTORE_END_STRING (l);
1503 if (!operand_ok)
1504 return;
1505 }
1506 else
1507 {
1508 if (expecting_operand)
1509 {
1510 expecting_operand_after_comma:
1511 as_bad (_("expecting operand after ','; got nothing"));
1512 return;
1513 }
1514 if (*l == ',')
1515 {
1516 as_bad (_("expecting operand before ','; got nothing"));
1517 return;
1518 }
1519 }
1520
1521 /* Now *l must be either ',' or END_OF_INSN. */
1522 if (*l == ',')
1523 {
1524 if (*++l == END_OF_INSN)
1525 {
1526 /* Just skip it, if it's \n complain. */
1527 goto expecting_operand_after_comma;
1528 }
1529 expecting_operand = 1;
1530 }
1531 }
1532 while (*l != END_OF_INSN);
1533 }
1534 }
1535
1536 /* Now we've parsed the mnemonic into a set of templates, and have the
1537 operands at hand.
1538
1539 Next, we find a template that matches the given insn,
1540 making sure the overlap of the given operands types is consistent
1541 with the template operand types. */
1542
1543 #define MATCH(overlap, given, template) \
1544 ((overlap & ~JumpAbsolute) \
1545 && ((given) & (BaseIndex|JumpAbsolute)) == ((overlap) & (BaseIndex|JumpAbsolute)))
1546
1547 /* If given types r0 and r1 are registers they must be of the same type
1548 unless the expected operand type register overlap is null.
1549 Note that Acc in a template matches every size of reg. */
1550 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
1551 ( ((g0) & Reg) == 0 || ((g1) & Reg) == 0 || \
1552 ((g0) & Reg) == ((g1) & Reg) || \
1553 ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
1554
1555 {
1556 register unsigned int overlap0, overlap1;
1557 unsigned int overlap2;
1558 unsigned int found_reverse_match;
1559 int suffix_check;
1560
1561 /* All intel opcodes have reversed operands except for "bound" and
1562 "enter". We also don't reverse intersegment "jmp" and "call"
1563 instructions with 2 immediate operands so that the immediate segment
1564 precedes the offset, as it does when in AT&T mode. "enter" and the
1565 intersegment "jmp" and "call" instructions are the only ones that
1566 have two immediate operands. */
1567 if (intel_syntax && i.operands > 1
1568 && (strcmp (mnemonic, "bound") != 0)
1569 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1570 {
1571 union i386_op temp_op;
1572 unsigned int temp_type;
1573 RELOC_ENUM temp_reloc;
1574 int xchg1 = 0;
1575 int xchg2 = 0;
1576
1577 if (i.operands == 2)
1578 {
1579 xchg1 = 0;
1580 xchg2 = 1;
1581 }
1582 else if (i.operands == 3)
1583 {
1584 xchg1 = 0;
1585 xchg2 = 2;
1586 }
1587 temp_type = i.types[xchg2];
1588 i.types[xchg2] = i.types[xchg1];
1589 i.types[xchg1] = temp_type;
1590 temp_op = i.op[xchg2];
1591 i.op[xchg2] = i.op[xchg1];
1592 i.op[xchg1] = temp_op;
1593 temp_reloc = i.reloc[xchg2];
1594 i.reloc[xchg2] = i.reloc[xchg1];
1595 i.reloc[xchg1] = temp_reloc;
1596
1597 if (i.mem_operands == 2)
1598 {
1599 const seg_entry *temp_seg;
1600 temp_seg = i.seg[0];
1601 i.seg[0] = i.seg[1];
1602 i.seg[1] = temp_seg;
1603 }
1604 }
1605
1606 if (i.imm_operands)
1607 {
1608 /* Try to ensure constant immediates are represented in the smallest
1609 opcode possible. */
1610 char guess_suffix = 0;
1611 int op;
1612
1613 if (i.suffix)
1614 guess_suffix = i.suffix;
1615 else if (i.reg_operands)
1616 {
1617 /* Figure out a suffix from the last register operand specified.
1618 We can't do this properly yet, ie. excluding InOutPortReg,
1619 but the following works for instructions with immediates.
1620 In any case, we can't set i.suffix yet. */
1621 for (op = i.operands; --op >= 0;)
1622 if (i.types[op] & Reg)
1623 {
1624 if (i.types[op] & Reg8)
1625 guess_suffix = BYTE_MNEM_SUFFIX;
1626 else if (i.types[op] & Reg16)
1627 guess_suffix = WORD_MNEM_SUFFIX;
1628 else if (i.types[op] & Reg32)
1629 guess_suffix = LONG_MNEM_SUFFIX;
1630 else if (i.types[op] & Reg64)
1631 guess_suffix = QWORD_MNEM_SUFFIX;
1632 break;
1633 }
1634 }
1635 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
1636 guess_suffix = WORD_MNEM_SUFFIX;
1637
1638 for (op = i.operands; --op >= 0;)
1639 if (i.types[op] & Imm)
1640 {
1641 switch (i.op[op].imms->X_op)
1642 {
1643 case O_constant:
1644 /* If a suffix is given, this operand may be shortened. */
1645 switch (guess_suffix)
1646 {
1647 case LONG_MNEM_SUFFIX:
1648 i.types[op] |= Imm32 | Imm64;
1649 break;
1650 case WORD_MNEM_SUFFIX:
1651 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
1652 break;
1653 case BYTE_MNEM_SUFFIX:
1654 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
1655 break;
1656 }
1657
1658 /* If this operand is at most 16 bits, convert it
1659 to a signed 16 bit number before trying to see
1660 whether it will fit in an even smaller size.
1661 This allows a 16-bit operand such as $0xffe0 to
1662 be recognised as within Imm8S range. */
1663 if ((i.types[op] & Imm16)
1664 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
1665 {
1666 i.op[op].imms->X_add_number =
1667 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
1668 }
1669 if ((i.types[op] & Imm32)
1670 && (i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1)) == 0)
1671 {
1672 i.op[op].imms->X_add_number =
1673 (i.op[op].imms->X_add_number ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1674 }
1675 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
1676 /* We must avoid matching of Imm32 templates when 64bit only immediate is available. */
1677 if (guess_suffix == QWORD_MNEM_SUFFIX)
1678 i.types[op] &= ~Imm32;
1679 break;
1680 case O_absent:
1681 case O_register:
1682 abort ();
1683 /* Symbols and expressions. */
1684 default:
1685 /* Convert symbolic operand to proper sizes for matching. */
1686 switch (guess_suffix)
1687 {
1688 case QWORD_MNEM_SUFFIX:
1689 i.types[op] = Imm64 | Imm32S;
1690 break;
1691 case LONG_MNEM_SUFFIX:
1692 i.types[op] = Imm32 | Imm64;
1693 break;
1694 case WORD_MNEM_SUFFIX:
1695 i.types[op] = Imm16 | Imm32 | Imm64;
1696 break;
1697 break;
1698 case BYTE_MNEM_SUFFIX:
1699 i.types[op] = Imm8 | Imm8S | Imm16 | Imm32S | Imm32;
1700 break;
1701 break;
1702 }
1703 break;
1704 }
1705 }
1706 }
1707
1708 if (i.disp_operands)
1709 {
1710 /* Try to use the smallest displacement type too. */
1711 int op;
1712
1713 for (op = i.operands; --op >= 0;)
1714 if ((i.types[op] & Disp)
1715 && i.op[op].disps->X_op == O_constant)
1716 {
1717 offsetT disp = i.op[op].disps->X_add_number;
1718
1719 if (i.types[op] & Disp16)
1720 {
1721 /* We know this operand is at most 16 bits, so
1722 convert to a signed 16 bit number before trying
1723 to see whether it will fit in an even smaller
1724 size. */
1725
1726 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
1727 }
1728 else if (i.types[op] & Disp32)
1729 {
1730 /* We know this operand is at most 32 bits, so convert to a
1731 signed 32 bit number before trying to see whether it will
1732 fit in an even smaller size. */
1733 disp &= (((offsetT) 2 << 31) - 1);
1734 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1735 }
1736 if (flag_code == CODE_64BIT)
1737 {
1738 if (fits_in_signed_long (disp))
1739 i.types[op] |= Disp32S;
1740 if (fits_in_unsigned_long (disp))
1741 i.types[op] |= Disp32;
1742 }
1743 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
1744 && fits_in_signed_byte (disp))
1745 i.types[op] |= Disp8;
1746 }
1747 }
1748
1749 overlap0 = 0;
1750 overlap1 = 0;
1751 overlap2 = 0;
1752 found_reverse_match = 0;
1753 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
1754 ? No_bSuf
1755 : (i.suffix == WORD_MNEM_SUFFIX
1756 ? No_wSuf
1757 : (i.suffix == SHORT_MNEM_SUFFIX
1758 ? No_sSuf
1759 : (i.suffix == LONG_MNEM_SUFFIX
1760 ? No_lSuf
1761 : (i.suffix == QWORD_MNEM_SUFFIX
1762 ? No_qSuf
1763 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX ? No_xSuf : 0))))));
1764
1765 for (t = current_templates->start;
1766 t < current_templates->end;
1767 t++)
1768 {
1769 /* Must have right number of operands. */
1770 if (i.operands != t->operands)
1771 continue;
1772
1773 /* Check the suffix, except for some instructions in intel mode. */
1774 if ((t->opcode_modifier & suffix_check)
1775 && !(intel_syntax
1776 && (t->opcode_modifier & IgnoreSize))
1777 && !(intel_syntax
1778 && t->base_opcode == 0xd9
1779 && (t->extension_opcode == 5 /* 0xd9,5 "fldcw" */
1780 || t->extension_opcode == 7))) /* 0xd9,7 "f{n}stcw" */
1781 continue;
1782
1783 /* Do not verify operands when there are none. */
1784 else if (!t->operands)
1785 {
1786 if (t->cpu_flags & ~cpu_arch_flags)
1787 continue;
1788 /* We've found a match; break out of loop. */
1789 break;
1790 }
1791
1792 overlap0 = i.types[0] & t->operand_types[0];
1793 switch (t->operands)
1794 {
1795 case 1:
1796 if (!MATCH (overlap0, i.types[0], t->operand_types[0]))
1797 continue;
1798 break;
1799 case 2:
1800 case 3:
1801 overlap1 = i.types[1] & t->operand_types[1];
1802 if (!MATCH (overlap0, i.types[0], t->operand_types[0])
1803 || !MATCH (overlap1, i.types[1], t->operand_types[1])
1804 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
1805 t->operand_types[0],
1806 overlap1, i.types[1],
1807 t->operand_types[1]))
1808 {
1809 /* Check if other direction is valid ... */
1810 if ((t->opcode_modifier & (D|FloatD)) == 0)
1811 continue;
1812
1813 /* Try reversing direction of operands. */
1814 overlap0 = i.types[0] & t->operand_types[1];
1815 overlap1 = i.types[1] & t->operand_types[0];
1816 if (!MATCH (overlap0, i.types[0], t->operand_types[1])
1817 || !MATCH (overlap1, i.types[1], t->operand_types[0])
1818 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
1819 t->operand_types[1],
1820 overlap1, i.types[1],
1821 t->operand_types[0]))
1822 {
1823 /* Does not match either direction. */
1824 continue;
1825 }
1826 /* found_reverse_match holds which of D or FloatDR
1827 we've found. */
1828 found_reverse_match = t->opcode_modifier & (D|FloatDR);
1829 }
1830 /* Found a forward 2 operand match here. */
1831 else if (t->operands == 3)
1832 {
1833 /* Here we make use of the fact that there are no
1834 reverse match 3 operand instructions, and all 3
1835 operand instructions only need to be checked for
1836 register consistency between operands 2 and 3. */
1837 overlap2 = i.types[2] & t->operand_types[2];
1838 if (!MATCH (overlap2, i.types[2], t->operand_types[2])
1839 || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1],
1840 t->operand_types[1],
1841 overlap2, i.types[2],
1842 t->operand_types[2]))
1843
1844 continue;
1845 }
1846 /* Found either forward/reverse 2 or 3 operand match here:
1847 slip through to break. */
1848 }
1849 if (t->cpu_flags & ~cpu_arch_flags)
1850 {
1851 found_reverse_match = 0;
1852 continue;
1853 }
1854 /* We've found a match; break out of loop. */
1855 break;
1856 }
1857 if (t == current_templates->end)
1858 {
1859 /* We found no match. */
1860 as_bad (_("suffix or operands invalid for `%s'"),
1861 current_templates->start->name);
1862 return;
1863 }
1864
1865 if (!quiet_warnings)
1866 {
1867 if (!intel_syntax
1868 && ((i.types[0] & JumpAbsolute)
1869 != (t->operand_types[0] & JumpAbsolute)))
1870 {
1871 as_warn (_("indirect %s without `*'"), t->name);
1872 }
1873
1874 if ((t->opcode_modifier & (IsPrefix|IgnoreSize))
1875 == (IsPrefix|IgnoreSize))
1876 {
1877 /* Warn them that a data or address size prefix doesn't
1878 affect assembly of the next line of code. */
1879 as_warn (_("stand-alone `%s' prefix"), t->name);
1880 }
1881 }
1882
1883 /* Copy the template we found. */
1884 i.tm = *t;
1885 if (found_reverse_match)
1886 {
1887 /* If we found a reverse match we must alter the opcode
1888 direction bit. found_reverse_match holds bits to change
1889 (different for int & float insns). */
1890
1891 i.tm.base_opcode ^= found_reverse_match;
1892
1893 i.tm.operand_types[0] = t->operand_types[1];
1894 i.tm.operand_types[1] = t->operand_types[0];
1895 }
1896
1897 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1898 if (SYSV386_COMPAT
1899 && intel_syntax
1900 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1901 i.tm.base_opcode ^= FloatR;
1902
1903 if (i.tm.opcode_modifier & FWait)
1904 if (! add_prefix (FWAIT_OPCODE))
1905 return;
1906
1907 /* Check string instruction segment overrides. */
1908 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1909 {
1910 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
1911 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
1912 {
1913 if (i.seg[0] != NULL && i.seg[0] != &es)
1914 {
1915 as_bad (_("`%s' operand %d must use `%%es' segment"),
1916 i.tm.name,
1917 mem_op + 1);
1918 return;
1919 }
1920 /* There's only ever one segment override allowed per instruction.
1921 This instruction possibly has a legal segment override on the
1922 second operand, so copy the segment to where non-string
1923 instructions store it, allowing common code. */
1924 i.seg[0] = i.seg[1];
1925 }
1926 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
1927 {
1928 if (i.seg[1] != NULL && i.seg[1] != &es)
1929 {
1930 as_bad (_("`%s' operand %d must use `%%es' segment"),
1931 i.tm.name,
1932 mem_op + 2);
1933 return;
1934 }
1935 }
1936 }
1937
1938 /* If matched instruction specifies an explicit instruction mnemonic
1939 suffix, use it. */
1940 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
1941 {
1942 if (i.tm.opcode_modifier & Size16)
1943 i.suffix = WORD_MNEM_SUFFIX;
1944 else if (i.tm.opcode_modifier & Size64)
1945 i.suffix = QWORD_MNEM_SUFFIX;
1946 else
1947 i.suffix = LONG_MNEM_SUFFIX;
1948 }
1949 else if (i.reg_operands)
1950 {
1951 /* If there's no instruction mnemonic suffix we try to invent one
1952 based on register operands. */
1953 if (!i.suffix)
1954 {
1955 /* We take i.suffix from the last register operand specified,
1956 Destination register type is more significant than source
1957 register type. */
1958 int op;
1959 for (op = i.operands; --op >= 0;)
1960 if ((i.types[op] & Reg)
1961 && !(i.tm.operand_types[op] & InOutPortReg))
1962 {
1963 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
1964 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
1965 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
1966 LONG_MNEM_SUFFIX);
1967 break;
1968 }
1969 }
1970 else if (i.suffix == BYTE_MNEM_SUFFIX)
1971 {
1972 int op;
1973 for (op = i.operands; --op >= 0;)
1974 {
1975 /* If this is an eight bit register, it's OK. If it's
1976 the 16 or 32 bit version of an eight bit register,
1977 we will just use the low portion, and that's OK too. */
1978 if (i.types[op] & Reg8)
1979 continue;
1980
1981 /* movzx and movsx should not generate this warning. */
1982 if (intel_syntax
1983 && (i.tm.base_opcode == 0xfb7
1984 || i.tm.base_opcode == 0xfb6
1985 || i.tm.base_opcode == 0x63
1986 || i.tm.base_opcode == 0xfbe
1987 || i.tm.base_opcode == 0xfbf))
1988 continue;
1989
1990 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4
1991 #if 0
1992 /* Check that the template allows eight bit regs
1993 This kills insns such as `orb $1,%edx', which
1994 maybe should be allowed. */
1995 && (i.tm.operand_types[op] & (Reg8|InOutPortReg))
1996 #endif
1997 )
1998 {
1999 /* Prohibit these changes in the 64bit mode, since
2000 the lowering is more complicated. */
2001 if (flag_code == CODE_64BIT
2002 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2003 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2004 i.op[op].regs->reg_name,
2005 i.suffix);
2006 #if REGISTER_WARNINGS
2007 if (!quiet_warnings
2008 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2009 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2010 (i.op[op].regs
2011 + (i.types[op] & Reg16
2012 ? REGNAM_AL - REGNAM_AX
2013 : REGNAM_AL - REGNAM_EAX))->reg_name,
2014 i.op[op].regs->reg_name,
2015 i.suffix);
2016 #endif
2017 continue;
2018 }
2019 /* Any other register is bad. */
2020 if (i.types[op] & (Reg | RegMMX | RegXMM
2021 | SReg2 | SReg3
2022 | Control | Debug | Test
2023 | FloatReg | FloatAcc))
2024 {
2025 as_bad (_("`%%%s' not allowed with `%s%c'"),
2026 i.op[op].regs->reg_name,
2027 i.tm.name,
2028 i.suffix);
2029 return;
2030 }
2031 }
2032 }
2033 else if (i.suffix == LONG_MNEM_SUFFIX)
2034 {
2035 int op;
2036
2037 for (op = i.operands; --op >= 0;)
2038 /* Reject eight bit registers, except where the template
2039 requires them. (eg. movzb) */
2040 if ((i.types[op] & Reg8) != 0
2041 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2042 {
2043 as_bad (_("`%%%s' not allowed with `%s%c'"),
2044 i.op[op].regs->reg_name,
2045 i.tm.name,
2046 i.suffix);
2047 return;
2048 }
2049 /* Warn if the e prefix on a general reg is missing. */
2050 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2051 && (i.types[op] & Reg16) != 0
2052 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
2053 {
2054 /* Prohibit these changes in the 64bit mode, since
2055 the lowering is more complicated. */
2056 if (flag_code == CODE_64BIT)
2057 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2058 i.op[op].regs->reg_name,
2059 i.suffix);
2060 #if REGISTER_WARNINGS
2061 else
2062 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2063 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
2064 i.op[op].regs->reg_name,
2065 i.suffix);
2066 #endif
2067 }
2068 /* Warn if the r prefix on a general reg is missing. */
2069 else if ((i.types[op] & Reg64) != 0
2070 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
2071 {
2072 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2073 i.op[op].regs->reg_name,
2074 i.suffix);
2075 }
2076 }
2077 else if (i.suffix == QWORD_MNEM_SUFFIX)
2078 {
2079 int op;
2080
2081 for (op = i.operands; --op >= 0; )
2082 /* Reject eight bit registers, except where the template
2083 requires them. (eg. movzb) */
2084 if ((i.types[op] & Reg8) != 0
2085 && (i.tm.operand_types[op] & (Reg16|Reg32|Acc)) != 0)
2086 {
2087 as_bad (_("`%%%s' not allowed with `%s%c'"),
2088 i.op[op].regs->reg_name,
2089 i.tm.name,
2090 i.suffix);
2091 return;
2092 }
2093 /* Warn if the e prefix on a general reg is missing. */
2094 else if (((i.types[op] & Reg16) != 0
2095 || (i.types[op] & Reg32) != 0)
2096 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
2097 {
2098 /* Prohibit these changes in the 64bit mode, since
2099 the lowering is more complicated. */
2100 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2101 i.op[op].regs->reg_name,
2102 i.suffix);
2103 }
2104 }
2105 else if (i.suffix == WORD_MNEM_SUFFIX)
2106 {
2107 int op;
2108 for (op = i.operands; --op >= 0;)
2109 /* Reject eight bit registers, except where the template
2110 requires them. (eg. movzb) */
2111 if ((i.types[op] & Reg8) != 0
2112 && (i.tm.operand_types[op] & (Reg16|Reg32|Acc)) != 0)
2113 {
2114 as_bad (_("`%%%s' not allowed with `%s%c'"),
2115 i.op[op].regs->reg_name,
2116 i.tm.name,
2117 i.suffix);
2118 return;
2119 }
2120 /* Warn if the e prefix on a general reg is present. */
2121 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2122 && (i.types[op] & Reg32) != 0
2123 && (i.tm.operand_types[op] & (Reg16|Acc)) != 0)
2124 {
2125 /* Prohibit these changes in the 64bit mode, since
2126 the lowering is more complicated. */
2127 if (flag_code == CODE_64BIT)
2128 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2129 i.op[op].regs->reg_name,
2130 i.suffix);
2131 else
2132 #if REGISTER_WARNINGS
2133 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2134 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
2135 i.op[op].regs->reg_name,
2136 i.suffix);
2137 #endif
2138 }
2139 }
2140 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2141 /* Do nothing if the instruction is going to ignore the prefix. */
2142 ;
2143 else
2144 abort ();
2145 }
2146 else if ((i.tm.opcode_modifier & DefaultSize) && !i.suffix)
2147 {
2148 i.suffix = stackop_size;
2149 }
2150 /* Make still unresolved immediate matches conform to size of immediate
2151 given in i.suffix. Note: overlap2 cannot be an immediate! */
2152 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S))
2153 && overlap0 != Imm8 && overlap0 != Imm8S
2154 && overlap0 != Imm16 && overlap0 != Imm32S
2155 && overlap0 != Imm32 && overlap0 != Imm64)
2156 {
2157 if (i.suffix)
2158 {
2159 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
2160 (i.suffix == WORD_MNEM_SUFFIX ? Imm16 :
2161 (i.suffix == QWORD_MNEM_SUFFIX ? Imm64 | Imm32S : Imm32)));
2162 }
2163 else if (overlap0 == (Imm16 | Imm32S | Imm32)
2164 || overlap0 == (Imm16 | Imm32)
2165 || overlap0 == (Imm16 | Imm32S))
2166 {
2167 overlap0 =
2168 ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32S;
2169 }
2170 if (overlap0 != Imm8 && overlap0 != Imm8S
2171 && overlap0 != Imm16 && overlap0 != Imm32S
2172 && overlap0 != Imm32 && overlap0 != Imm64)
2173 {
2174 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
2175 return;
2176 }
2177 }
2178 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32))
2179 && overlap1 != Imm8 && overlap1 != Imm8S
2180 && overlap1 != Imm16 && overlap1 != Imm32S
2181 && overlap1 != Imm32 && overlap1 != Imm64)
2182 {
2183 if (i.suffix)
2184 {
2185 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
2186 (i.suffix == WORD_MNEM_SUFFIX ? Imm16 :
2187 (i.suffix == QWORD_MNEM_SUFFIX ? Imm64 | Imm32S : Imm32)));
2188 }
2189 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
2190 || overlap1 == (Imm16 | Imm32)
2191 || overlap1 == (Imm16 | Imm32S))
2192 {
2193 overlap1 =
2194 ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32S;
2195 }
2196 if (overlap1 != Imm8 && overlap1 != Imm8S
2197 && overlap1 != Imm16 && overlap1 != Imm32S
2198 && overlap1 != Imm32 && overlap1 != Imm64)
2199 {
2200 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
2201 return;
2202 }
2203 }
2204 assert ((overlap2 & Imm) == 0);
2205
2206 i.types[0] = overlap0;
2207 if (overlap0 & ImplicitRegister)
2208 i.reg_operands--;
2209 if (overlap0 & Imm1)
2210 i.imm_operands = 0; /* kludge for shift insns. */
2211
2212 i.types[1] = overlap1;
2213 if (overlap1 & ImplicitRegister)
2214 i.reg_operands--;
2215
2216 i.types[2] = overlap2;
2217 if (overlap2 & ImplicitRegister)
2218 i.reg_operands--;
2219
2220 /* Finalize opcode. First, we change the opcode based on the operand
2221 size given by i.suffix: We need not change things for byte insns. */
2222
2223 if (!i.suffix && (i.tm.opcode_modifier & W))
2224 {
2225 as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
2226 return;
2227 }
2228
2229 /* For movzx and movsx, need to check the register type. */
2230 if (intel_syntax
2231 && (i.tm.base_opcode == 0xfb6 || i.tm.base_opcode == 0xfbe))
2232 if (i.suffix && i.suffix == BYTE_MNEM_SUFFIX)
2233 {
2234 unsigned int prefix = DATA_PREFIX_OPCODE;
2235
2236 if ((i.op[1].regs->reg_type & Reg16) != 0)
2237 if (!add_prefix (prefix))
2238 return;
2239 }
2240
2241 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2242 {
2243 /* It's not a byte, select word/dword operation. */
2244 if (i.tm.opcode_modifier & W)
2245 {
2246 if (i.tm.opcode_modifier & ShortForm)
2247 i.tm.base_opcode |= 8;
2248 else
2249 i.tm.base_opcode |= 1;
2250 }
2251 /* Now select between word & dword operations via the operand
2252 size prefix, except for instructions that will ignore this
2253 prefix anyway. */
2254 if (i.suffix != QWORD_MNEM_SUFFIX
2255 && (i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2256 && !(i.tm.opcode_modifier & IgnoreSize))
2257 {
2258 unsigned int prefix = DATA_PREFIX_OPCODE;
2259 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2260 prefix = ADDR_PREFIX_OPCODE;
2261
2262 if (! add_prefix (prefix))
2263 return;
2264 }
2265
2266 /* Set mode64 for an operand. */
2267 if (i.suffix == QWORD_MNEM_SUFFIX
2268 && !(i.tm.opcode_modifier & NoRex64))
2269 {
2270 i.rex.mode64 = 1;
2271 if (flag_code < CODE_64BIT)
2272 {
2273 as_bad (_("64bit operations available only in 64bit modes."));
2274 return;
2275 }
2276 }
2277
2278 /* Size floating point instruction. */
2279 if (i.suffix == LONG_MNEM_SUFFIX)
2280 {
2281 if (i.tm.opcode_modifier & FloatMF)
2282 i.tm.base_opcode ^= 4;
2283 }
2284 }
2285
2286 if (i.tm.opcode_modifier & ImmExt)
2287 {
2288 /* These AMD 3DNow! and Intel Katmai New Instructions have an
2289 opcode suffix which is coded in the same place as an 8-bit
2290 immediate field would be. Here we fake an 8-bit immediate
2291 operand from the opcode suffix stored in tm.extension_opcode. */
2292
2293 expressionS *exp;
2294
2295 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
2296
2297 exp = &im_expressions[i.imm_operands++];
2298 i.op[i.operands].imms = exp;
2299 i.types[i.operands++] = Imm8;
2300 exp->X_op = O_constant;
2301 exp->X_add_number = i.tm.extension_opcode;
2302 i.tm.extension_opcode = None;
2303 }
2304
2305 /* For insns with operands there are more diddles to do to the opcode. */
2306 if (i.operands)
2307 {
2308 /* Default segment register this instruction will use
2309 for memory accesses. 0 means unknown.
2310 This is only for optimizing out unnecessary segment overrides. */
2311 const seg_entry *default_seg = 0;
2312
2313 /* The imul $imm, %reg instruction is converted into
2314 imul $imm, %reg, %reg, and the clr %reg instruction
2315 is converted into xor %reg, %reg. */
2316 if (i.tm.opcode_modifier & regKludge)
2317 {
2318 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
2319 /* Pretend we saw the extra register operand. */
2320 assert (i.op[first_reg_op + 1].regs == 0);
2321 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
2322 i.types[first_reg_op + 1] = i.types[first_reg_op];
2323 i.reg_operands = 2;
2324 }
2325
2326 if (i.tm.opcode_modifier & ShortForm)
2327 {
2328 /* The register or float register operand is in operand 0 or 1. */
2329 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
2330 /* Register goes in low 3 bits of opcode. */
2331 i.tm.base_opcode |= i.op[op].regs->reg_num;
2332 if (i.op[op].regs->reg_flags & RegRex)
2333 i.rex.extZ = 1;
2334 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2335 {
2336 /* Warn about some common errors, but press on regardless.
2337 The first case can be generated by gcc (<= 2.8.1). */
2338 if (i.operands == 2)
2339 {
2340 /* Reversed arguments on faddp, fsubp, etc. */
2341 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
2342 i.op[1].regs->reg_name,
2343 i.op[0].regs->reg_name);
2344 }
2345 else
2346 {
2347 /* Extraneous `l' suffix on fp insn. */
2348 as_warn (_("translating to `%s %%%s'"), i.tm.name,
2349 i.op[0].regs->reg_name);
2350 }
2351 }
2352 }
2353 else if (i.tm.opcode_modifier & Modrm)
2354 {
2355 /* The opcode is completed (modulo i.tm.extension_opcode which
2356 must be put into the modrm byte).
2357 Now, we make the modrm & index base bytes based on all the
2358 info we've collected. */
2359
2360 /* i.reg_operands MUST be the number of real register operands;
2361 implicit registers do not count. */
2362 if (i.reg_operands == 2)
2363 {
2364 unsigned int source, dest;
2365 source = ((i.types[0]
2366 & (Reg | RegMMX | RegXMM
2367 | SReg2 | SReg3
2368 | Control | Debug | Test))
2369 ? 0 : 1);
2370 dest = source + 1;
2371
2372 i.rm.mode = 3;
2373 /* One of the register operands will be encoded in the
2374 i.tm.reg field, the other in the combined i.tm.mode
2375 and i.tm.regmem fields. If no form of this
2376 instruction supports a memory destination operand,
2377 then we assume the source operand may sometimes be
2378 a memory operand and so we need to store the
2379 destination in the i.rm.reg field. */
2380 if ((i.tm.operand_types[dest] & AnyMem) == 0)
2381 {
2382 i.rm.reg = i.op[dest].regs->reg_num;
2383 i.rm.regmem = i.op[source].regs->reg_num;
2384 if (i.op[dest].regs->reg_flags & RegRex)
2385 i.rex.extX = 1;
2386 if (i.op[source].regs->reg_flags & RegRex)
2387 i.rex.extZ = 1;
2388 }
2389 else
2390 {
2391 i.rm.reg = i.op[source].regs->reg_num;
2392 i.rm.regmem = i.op[dest].regs->reg_num;
2393 if (i.op[dest].regs->reg_flags & RegRex)
2394 i.rex.extZ = 1;
2395 if (i.op[source].regs->reg_flags & RegRex)
2396 i.rex.extX = 1;
2397 }
2398 }
2399 else
2400 { /* If it's not 2 reg operands... */
2401 if (i.mem_operands)
2402 {
2403 unsigned int fake_zero_displacement = 0;
2404 unsigned int op = ((i.types[0] & AnyMem)
2405 ? 0
2406 : (i.types[1] & AnyMem) ? 1 : 2);
2407
2408 default_seg = &ds;
2409
2410 if (! i.base_reg)
2411 {
2412 i.rm.mode = 0;
2413 if (! i.disp_operands)
2414 fake_zero_displacement = 1;
2415 if (! i.index_reg)
2416 {
2417 /* Operand is just <disp> */
2418 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
2419 {
2420 i.rm.regmem = NO_BASE_REGISTER_16;
2421 i.types[op] &= ~Disp;
2422 i.types[op] |= Disp16;
2423 }
2424 else if (flag_code != CODE_64BIT)
2425 {
2426 i.rm.regmem = NO_BASE_REGISTER;
2427 i.types[op] &= ~Disp;
2428 i.types[op] |= Disp32;
2429 }
2430 else
2431 {
2432 /* 64bit mode overwrites the 32bit
2433 absolute addressing by RIP relative
2434 addressing and absolute addressing
2435 is encoded by one of the redundant
2436 SIB forms. */
2437
2438 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2439 i.sib.base = NO_BASE_REGISTER;
2440 i.sib.index = NO_INDEX_REGISTER;
2441 i.types[op] &= ~Disp;
2442 i.types[op] |= Disp32S;
2443 }
2444 }
2445 else /* ! i.base_reg && i.index_reg */
2446 {
2447 i.sib.index = i.index_reg->reg_num;
2448 i.sib.base = NO_BASE_REGISTER;
2449 i.sib.scale = i.log2_scale_factor;
2450 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2451 i.types[op] &= ~Disp;
2452 if (flag_code != CODE_64BIT)
2453 i.types[op] |= Disp32; /* Must be 32 bit */
2454 else
2455 i.types[op] |= Disp32S;
2456 if (i.index_reg->reg_flags & RegRex)
2457 i.rex.extY = 1;
2458 }
2459 }
2460 /* RIP addressing for 64bit mode. */
2461 else if (i.base_reg->reg_type == BaseIndex)
2462 {
2463 i.rm.regmem = NO_BASE_REGISTER;
2464 i.types[op] &= ~Disp;
2465 i.types[op] |= Disp32S;
2466 i.flags[op] = Operand_PCrel;
2467 }
2468 else if (i.base_reg->reg_type & Reg16)
2469 {
2470 switch (i.base_reg->reg_num)
2471 {
2472 case 3: /* (%bx) */
2473 if (! i.index_reg)
2474 i.rm.regmem = 7;
2475 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
2476 i.rm.regmem = i.index_reg->reg_num - 6;
2477 break;
2478 case 5: /* (%bp) */
2479 default_seg = &ss;
2480 if (! i.index_reg)
2481 {
2482 i.rm.regmem = 6;
2483 if ((i.types[op] & Disp) == 0)
2484 {
2485 /* fake (%bp) into 0(%bp) */
2486 i.types[op] |= Disp8;
2487 fake_zero_displacement = 1;
2488 }
2489 }
2490 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
2491 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
2492 break;
2493 default: /* (%si) -> 4 or (%di) -> 5 */
2494 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
2495 }
2496 i.rm.mode = mode_from_disp_size (i.types[op]);
2497 }
2498 else /* i.base_reg and 32/64 bit mode */
2499 {
2500 if (flag_code == CODE_64BIT
2501 && (i.types[op] & Disp))
2502 {
2503 if (i.types[op] & Disp8)
2504 i.types[op] = Disp8 | Disp32S;
2505 else
2506 i.types[op] = Disp32S;
2507 }
2508 i.rm.regmem = i.base_reg->reg_num;
2509 if (i.base_reg->reg_flags & RegRex)
2510 i.rex.extZ = 1;
2511 i.sib.base = i.base_reg->reg_num;
2512 /* x86-64 ignores REX prefix bit here to avoid
2513 decoder complications. */
2514 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
2515 {
2516 default_seg = &ss;
2517 if (i.disp_operands == 0)
2518 {
2519 fake_zero_displacement = 1;
2520 i.types[op] |= Disp8;
2521 }
2522 }
2523 else if (i.base_reg->reg_num == ESP_REG_NUM)
2524 {
2525 default_seg = &ss;
2526 }
2527 i.sib.scale = i.log2_scale_factor;
2528 if (! i.index_reg)
2529 {
2530 /* <disp>(%esp) becomes two byte modrm
2531 with no index register. We've already
2532 stored the code for esp in i.rm.regmem
2533 ie. ESCAPE_TO_TWO_BYTE_ADDRESSING. Any
2534 base register besides %esp will not use
2535 the extra modrm byte. */
2536 i.sib.index = NO_INDEX_REGISTER;
2537 #if ! SCALE1_WHEN_NO_INDEX
2538 /* Another case where we force the second
2539 modrm byte. */
2540 if (i.log2_scale_factor)
2541 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2542 #endif
2543 }
2544 else
2545 {
2546 i.sib.index = i.index_reg->reg_num;
2547 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2548 if (i.index_reg->reg_flags & RegRex)
2549 i.rex.extY = 1;
2550 }
2551 i.rm.mode = mode_from_disp_size (i.types[op]);
2552 }
2553
2554 if (fake_zero_displacement)
2555 {
2556 /* Fakes a zero displacement assuming that i.types[op]
2557 holds the correct displacement size. */
2558 expressionS *exp;
2559
2560 assert (i.op[op].disps == 0);
2561 exp = &disp_expressions[i.disp_operands++];
2562 i.op[op].disps = exp;
2563 exp->X_op = O_constant;
2564 exp->X_add_number = 0;
2565 exp->X_add_symbol = (symbolS *) 0;
2566 exp->X_op_symbol = (symbolS *) 0;
2567 }
2568 }
2569
2570 /* Fill in i.rm.reg or i.rm.regmem field with register
2571 operand (if any) based on i.tm.extension_opcode.
2572 Again, we must be careful to make sure that
2573 segment/control/debug/test/MMX registers are coded
2574 into the i.rm.reg field. */
2575 if (i.reg_operands)
2576 {
2577 unsigned int op =
2578 ((i.types[0]
2579 & (Reg | RegMMX | RegXMM
2580 | SReg2 | SReg3
2581 | Control | Debug | Test))
2582 ? 0
2583 : ((i.types[1]
2584 & (Reg | RegMMX | RegXMM
2585 | SReg2 | SReg3
2586 | Control | Debug | Test))
2587 ? 1
2588 : 2));
2589 /* If there is an extension opcode to put here, the
2590 register number must be put into the regmem field. */
2591 if (i.tm.extension_opcode != None)
2592 {
2593 i.rm.regmem = i.op[op].regs->reg_num;
2594 if (i.op[op].regs->reg_flags & RegRex)
2595 i.rex.extZ = 1;
2596 }
2597 else
2598 {
2599 i.rm.reg = i.op[op].regs->reg_num;
2600 if (i.op[op].regs->reg_flags & RegRex)
2601 i.rex.extX = 1;
2602 }
2603
2604 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2
2605 we must set it to 3 to indicate this is a register
2606 operand in the regmem field. */
2607 if (!i.mem_operands)
2608 i.rm.mode = 3;
2609 }
2610
2611 /* Fill in i.rm.reg field with extension opcode (if any). */
2612 if (i.tm.extension_opcode != None)
2613 i.rm.reg = i.tm.extension_opcode;
2614 }
2615 }
2616 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
2617 {
2618 if (i.tm.base_opcode == POP_SEG_SHORT
2619 && i.op[0].regs->reg_num == 1)
2620 {
2621 as_bad (_("you can't `pop %%cs'"));
2622 return;
2623 }
2624 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
2625 if (i.op[0].regs->reg_flags & RegRex)
2626 i.rex.extZ = 1;
2627 }
2628 else if ((i.tm.base_opcode & ~(D|W)) == MOV_AX_DISP32)
2629 {
2630 default_seg = &ds;
2631 }
2632 else if ((i.tm.opcode_modifier & IsString) != 0)
2633 {
2634 /* For the string instructions that allow a segment override
2635 on one of their operands, the default segment is ds. */
2636 default_seg = &ds;
2637 }
2638
2639 /* If a segment was explicitly specified,
2640 and the specified segment is not the default,
2641 use an opcode prefix to select it.
2642 If we never figured out what the default segment is,
2643 then default_seg will be zero at this point,
2644 and the specified segment prefix will always be used. */
2645 if ((i.seg[0]) && (i.seg[0] != default_seg))
2646 {
2647 if (! add_prefix (i.seg[0]->seg_prefix))
2648 return;
2649 }
2650 }
2651 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2652 {
2653 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
2654 as_warn (_("translating to `%sp'"), i.tm.name);
2655 }
2656 }
2657
2658 /* Handle conversion of 'int $3' --> special int3 insn. */
2659 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
2660 {
2661 i.tm.base_opcode = INT3_OPCODE;
2662 i.imm_operands = 0;
2663 }
2664
2665 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
2666 && i.op[0].disps->X_op == O_constant)
2667 {
2668 /* Convert "jmp constant" (and "call constant") to a jump (call) to
2669 the absolute address given by the constant. Since ix86 jumps and
2670 calls are pc relative, we need to generate a reloc. */
2671 i.op[0].disps->X_add_symbol = &abs_symbol;
2672 i.op[0].disps->X_op = O_symbol;
2673 }
2674
2675 if (i.tm.opcode_modifier & Rex64)
2676 i.rex.mode64 = 1;
2677
2678 /* For 8bit registers we would need an empty rex prefix.
2679 Also in the case instruction is already having prefix,
2680 we need to convert old registers to new ones. */
2681
2682 if (((i.types[0] & Reg8) && (i.op[0].regs->reg_flags & RegRex64))
2683 || ((i.types[1] & Reg8) && (i.op[1].regs->reg_flags & RegRex64))
2684 || ((i.rex.mode64 || i.rex.extX || i.rex.extY || i.rex.extZ || i.rex.empty)
2685 && ((i.types[0] & Reg8) || (i.types[1] & Reg8))))
2686 {
2687 int x;
2688 i.rex.empty = 1;
2689 for (x = 0; x < 2; x++)
2690 {
2691 /* Look for 8bit operand that does use old registers. */
2692 if (i.types[x] & Reg8
2693 && !(i.op[x].regs->reg_flags & RegRex64))
2694 {
2695 /* In case it is "hi" register, give up. */
2696 if (i.op[x].regs->reg_num > 3)
2697 as_bad (_("Can't encode registers '%%%s' in the instruction requiring REX prefix.\n"),
2698 i.op[x].regs->reg_name);
2699
2700 /* Otherwise it is equivalent to the extended register.
2701 Since the encoding don't change this is merely cosmetical
2702 cleanup for debug output. */
2703
2704 i.op[x].regs = i.op[x].regs + 8;
2705 }
2706 }
2707 }
2708
2709 if (i.rex.mode64 || i.rex.extX || i.rex.extY || i.rex.extZ || i.rex.empty)
2710 add_prefix (0x40
2711 | (i.rex.mode64 ? 8 : 0)
2712 | (i.rex.extX ? 4 : 0)
2713 | (i.rex.extY ? 2 : 0)
2714 | (i.rex.extZ ? 1 : 0));
2715
2716 /* We are ready to output the insn. */
2717 {
2718 register char *p;
2719
2720 /* Tie dwarf2 debug info to the address at the start of the insn.
2721 We can't do this after the insn has been output as the current
2722 frag may have been closed off. eg. by frag_var. */
2723 dwarf2_emit_insn (0);
2724
2725 /* Output jumps. */
2726 if (i.tm.opcode_modifier & Jump)
2727 {
2728 int code16;
2729 int prefix;
2730 relax_substateT subtype;
2731 symbolS *sym;
2732 offsetT off;
2733
2734 code16 = 0;
2735 if (flag_code == CODE_16BIT)
2736 code16 = CODE16;
2737
2738 prefix = 0;
2739 if (i.prefix[DATA_PREFIX])
2740 {
2741 prefix = 1;
2742 i.prefixes -= 1;
2743 code16 ^= CODE16;
2744 }
2745 /* Pentium4 branch hints. */
2746 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
2747 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
2748 {
2749 prefix++;
2750 i.prefixes--;
2751 }
2752 if (i.prefix[REX_PREFIX])
2753 {
2754 prefix++;
2755 i.prefixes--;
2756 }
2757
2758 if (i.prefixes != 0 && !intel_syntax)
2759 as_warn (_("skipping prefixes on this instruction"));
2760
2761 /* It's always a symbol; End frag & setup for relax.
2762 Make sure there is enough room in this frag for the largest
2763 instruction we may generate in md_convert_frag. This is 2
2764 bytes for the opcode and room for the prefix and largest
2765 displacement. */
2766 frag_grow (prefix + 2 + 4);
2767 /* Prefix and 1 opcode byte go in fr_fix. */
2768 p = frag_more (prefix + 1);
2769 if (i.prefix[DATA_PREFIX])
2770 *p++ = DATA_PREFIX_OPCODE;
2771 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
2772 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
2773 *p++ = i.prefix[SEG_PREFIX];
2774 if (i.prefix[REX_PREFIX])
2775 *p++ = i.prefix[REX_PREFIX];
2776 *p = i.tm.base_opcode;
2777
2778 if ((unsigned char) *p == JUMP_PC_RELATIVE)
2779 subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
2780 else if ((cpu_arch_flags & Cpu386) != 0)
2781 subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
2782 else
2783 subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
2784 subtype |= code16;
2785
2786 sym = i.op[0].disps->X_add_symbol;
2787 off = i.op[0].disps->X_add_number;
2788
2789 if (i.op[0].disps->X_op != O_constant
2790 && i.op[0].disps->X_op != O_symbol)
2791 {
2792 /* Handle complex expressions. */
2793 sym = make_expr_symbol (i.op[0].disps);
2794 off = 0;
2795 }
2796
2797 /* 1 possible extra opcode + 4 byte displacement go in var part.
2798 Pass reloc in fr_var. */
2799 frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
2800 }
2801 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
2802 {
2803 int size;
2804
2805 if (i.tm.opcode_modifier & JumpByte)
2806 {
2807 /* This is a loop or jecxz type instruction. */
2808 size = 1;
2809 if (i.prefix[ADDR_PREFIX])
2810 {
2811 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
2812 i.prefixes -= 1;
2813 }
2814 /* Pentium4 branch hints. */
2815 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
2816 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
2817 {
2818 FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
2819 i.prefixes--;
2820 }
2821 }
2822 else
2823 {
2824 int code16;
2825
2826 code16 = 0;
2827 if (flag_code == CODE_16BIT)
2828 code16 = CODE16;
2829
2830 if (i.prefix[DATA_PREFIX])
2831 {
2832 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
2833 i.prefixes -= 1;
2834 code16 ^= CODE16;
2835 }
2836
2837 size = 4;
2838 if (code16)
2839 size = 2;
2840 }
2841
2842 if (i.prefix[REX_PREFIX])
2843 {
2844 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
2845 i.prefixes -= 1;
2846 }
2847
2848 if (i.prefixes != 0 && !intel_syntax)
2849 as_warn (_("skipping prefixes on this instruction"));
2850
2851 p = frag_more (1 + size);
2852 *p++ = i.tm.base_opcode;
2853
2854 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2855 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
2856 }
2857 else if (i.tm.opcode_modifier & JumpInterSegment)
2858 {
2859 int size;
2860 int prefix;
2861 int code16;
2862
2863 code16 = 0;
2864 if (flag_code == CODE_16BIT)
2865 code16 = CODE16;
2866
2867 prefix = 0;
2868 if (i.prefix[DATA_PREFIX])
2869 {
2870 prefix = 1;
2871 i.prefixes -= 1;
2872 code16 ^= CODE16;
2873 }
2874 if (i.prefix[REX_PREFIX])
2875 {
2876 prefix++;
2877 i.prefixes -= 1;
2878 }
2879
2880 size = 4;
2881 if (code16)
2882 size = 2;
2883
2884 if (i.prefixes != 0 && !intel_syntax)
2885 as_warn (_("skipping prefixes on this instruction"));
2886
2887 /* 1 opcode; 2 segment; offset */
2888 p = frag_more (prefix + 1 + 2 + size);
2889
2890 if (i.prefix[DATA_PREFIX])
2891 *p++ = DATA_PREFIX_OPCODE;
2892
2893 if (i.prefix[REX_PREFIX])
2894 *p++ = i.prefix[REX_PREFIX];
2895
2896 *p++ = i.tm.base_opcode;
2897 if (i.op[1].imms->X_op == O_constant)
2898 {
2899 offsetT n = i.op[1].imms->X_add_number;
2900
2901 if (size == 2
2902 && !fits_in_unsigned_word (n)
2903 && !fits_in_signed_word (n))
2904 {
2905 as_bad (_("16-bit jump out of range"));
2906 return;
2907 }
2908 md_number_to_chars (p, n, size);
2909 }
2910 else
2911 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2912 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
2913 if (i.op[0].imms->X_op != O_constant)
2914 as_bad (_("can't handle non absolute segment in `%s'"),
2915 i.tm.name);
2916 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
2917 }
2918 else
2919 {
2920 /* Output normal instructions here. */
2921 unsigned char *q;
2922
2923 /* All opcodes on i386 have eighter 1 or 2 bytes. We may use third
2924 byte for the SSE instructions to specify prefix they require. */
2925 if (i.tm.base_opcode & 0xff0000)
2926 add_prefix ((i.tm.base_opcode >> 16) & 0xff);
2927
2928 /* The prefix bytes. */
2929 for (q = i.prefix;
2930 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
2931 q++)
2932 {
2933 if (*q)
2934 {
2935 p = frag_more (1);
2936 md_number_to_chars (p, (valueT) *q, 1);
2937 }
2938 }
2939
2940 /* Now the opcode; be careful about word order here! */
2941 if (fits_in_unsigned_byte (i.tm.base_opcode))
2942 {
2943 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
2944 }
2945 else
2946 {
2947 p = frag_more (2);
2948 /* Put out high byte first: can't use md_number_to_chars! */
2949 *p++ = (i.tm.base_opcode >> 8) & 0xff;
2950 *p = i.tm.base_opcode & 0xff;
2951 }
2952
2953 /* Now the modrm byte and sib byte (if present). */
2954 if (i.tm.opcode_modifier & Modrm)
2955 {
2956 p = frag_more (1);
2957 md_number_to_chars (p,
2958 (valueT) (i.rm.regmem << 0
2959 | i.rm.reg << 3
2960 | i.rm.mode << 6),
2961 1);
2962 /* If i.rm.regmem == ESP (4)
2963 && i.rm.mode != (Register mode)
2964 && not 16 bit
2965 ==> need second modrm byte. */
2966 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
2967 && i.rm.mode != 3
2968 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
2969 {
2970 p = frag_more (1);
2971 md_number_to_chars (p,
2972 (valueT) (i.sib.base << 0
2973 | i.sib.index << 3
2974 | i.sib.scale << 6),
2975 1);
2976 }
2977 }
2978
2979 if (i.disp_operands)
2980 {
2981 register unsigned int n;
2982
2983 for (n = 0; n < i.operands; n++)
2984 {
2985 if (i.types[n] & Disp)
2986 {
2987 if (i.op[n].disps->X_op == O_constant)
2988 {
2989 int size;
2990 offsetT val;
2991
2992 size = 4;
2993 if (i.types[n] & (Disp8 | Disp16 | Disp64))
2994 {
2995 size = 2;
2996 if (i.types[n] & Disp8)
2997 size = 1;
2998 if (i.types[n] & Disp64)
2999 size = 8;
3000 }
3001 val = offset_in_range (i.op[n].disps->X_add_number,
3002 size);
3003 p = frag_more (size);
3004 md_number_to_chars (p, val, size);
3005 }
3006 else
3007 {
3008 int size = 4;
3009 int sign = 0;
3010 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
3011
3012 /* The PC relative address is computed relative
3013 to the instruction boundary, so in case immediate
3014 fields follows, we need to adjust the value. */
3015 if (pcrel && i.imm_operands)
3016 {
3017 int imm_size = 4;
3018 register unsigned int n1;
3019
3020 for (n1 = 0; n1 < i.operands; n1++)
3021 if (i.types[n1] & Imm)
3022 {
3023 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
3024 {
3025 imm_size = 2;
3026 if (i.types[n1] & (Imm8 | Imm8S))
3027 imm_size = 1;
3028 if (i.types[n1] & Imm64)
3029 imm_size = 8;
3030 }
3031 break;
3032 }
3033 /* We should find the immediate. */
3034 if (n1 == i.operands)
3035 abort ();
3036 i.op[n].disps->X_add_number -= imm_size;
3037 }
3038
3039 if (i.types[n] & Disp32S)
3040 sign = 1;
3041
3042 if (i.types[n] & (Disp16 | Disp64))
3043 {
3044 size = 2;
3045 if (i.types[n] & Disp64)
3046 size = 8;
3047 }
3048
3049 p = frag_more (size);
3050 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3051 i.op[n].disps, pcrel,
3052 reloc (size, pcrel, sign, i.reloc[n]));
3053 }
3054 }
3055 }
3056 }
3057
3058 /* Output immediate. */
3059 if (i.imm_operands)
3060 {
3061 register unsigned int n;
3062
3063 for (n = 0; n < i.operands; n++)
3064 {
3065 if (i.types[n] & Imm)
3066 {
3067 if (i.op[n].imms->X_op == O_constant)
3068 {
3069 int size;
3070 offsetT val;
3071
3072 size = 4;
3073 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3074 {
3075 size = 2;
3076 if (i.types[n] & (Imm8 | Imm8S))
3077 size = 1;
3078 else if (i.types[n] & Imm64)
3079 size = 8;
3080 }
3081 val = offset_in_range (i.op[n].imms->X_add_number,
3082 size);
3083 p = frag_more (size);
3084 md_number_to_chars (p, val, size);
3085 }
3086 else
3087 {
3088 /* Not absolute_section.
3089 Need a 32-bit fixup (don't support 8bit
3090 non-absolute imms). Try to support other
3091 sizes ... */
3092 RELOC_ENUM reloc_type;
3093 int size = 4;
3094 int sign = 0;
3095
3096 if ((i.types[n] & (Imm32S))
3097 && i.suffix == QWORD_MNEM_SUFFIX)
3098 sign = 1;
3099 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3100 {
3101 size = 2;
3102 if (i.types[n] & (Imm8 | Imm8S))
3103 size = 1;
3104 if (i.types[n] & Imm64)
3105 size = 8;
3106 }
3107
3108 p = frag_more (size);
3109 reloc_type = reloc (size, 0, sign, i.reloc[n]);
3110 #ifdef BFD_ASSEMBLER
3111 if (reloc_type == BFD_RELOC_32
3112 && GOT_symbol
3113 && GOT_symbol == i.op[n].imms->X_add_symbol
3114 && (i.op[n].imms->X_op == O_symbol
3115 || (i.op[n].imms->X_op == O_add
3116 && ((symbol_get_value_expression
3117 (i.op[n].imms->X_op_symbol)->X_op)
3118 == O_subtract))))
3119 {
3120 /* We don't support dynamic linking on x86-64 yet. */
3121 if (flag_code == CODE_64BIT)
3122 abort ();
3123 reloc_type = BFD_RELOC_386_GOTPC;
3124 i.op[n].imms->X_add_number += 3;
3125 }
3126 #endif
3127 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3128 i.op[n].imms, 0, reloc_type);
3129 }
3130 }
3131 }
3132 }
3133 }
3134
3135 #ifdef DEBUG386
3136 if (flag_debug)
3137 {
3138 pi (line, &i);
3139 }
3140 #endif /* DEBUG386 */
3141 }
3142 }
3143 \f
3144 #ifndef LEX_AT
3145 static char *lex_got PARAMS ((RELOC_ENUM *, int *));
3146
3147 /* Parse operands of the form
3148 <symbol>@GOTOFF+<nnn>
3149 and similar .plt or .got references.
3150
3151 If we find one, set up the correct relocation in RELOC and copy the
3152 input string, minus the `@GOTOFF' into a malloc'd buffer for
3153 parsing by the calling routine. Return this buffer, and if ADJUST
3154 is non-null set it to the length of the string we removed from the
3155 input line. Otherwise return NULL. */
3156 static char *
3157 lex_got (reloc, adjust)
3158 RELOC_ENUM *reloc;
3159 int *adjust;
3160 {
3161 static const char * const mode_name[NUM_FLAG_CODE] = { "32", "16", "64" };
3162 static const struct {
3163 const char *str;
3164 const RELOC_ENUM rel[NUM_FLAG_CODE];
3165 } gotrel[] = {
3166 { "PLT", { BFD_RELOC_386_PLT32, 0, BFD_RELOC_X86_64_PLT32 } },
3167 { "GOTOFF", { BFD_RELOC_386_GOTOFF, 0, 0 } },
3168 { "GOTPCREL", { 0, 0, BFD_RELOC_X86_64_GOTPCREL } },
3169 { "GOT", { BFD_RELOC_386_GOT32, 0, BFD_RELOC_X86_64_GOT32 } }
3170 };
3171 char *cp;
3172 unsigned int j;
3173
3174 for (cp = input_line_pointer; *cp != '@'; cp++)
3175 if (is_end_of_line[(unsigned char) *cp])
3176 return NULL;
3177
3178 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
3179 {
3180 int len;
3181
3182 len = strlen (gotrel[j].str);
3183 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
3184 {
3185 if (gotrel[j].rel[(unsigned int) flag_code] != 0)
3186 {
3187 int first, second;
3188 char *tmpbuf, *past_reloc;
3189
3190 *reloc = gotrel[j].rel[(unsigned int) flag_code];
3191 if (adjust)
3192 *adjust = len;
3193
3194 if (GOT_symbol == NULL)
3195 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3196
3197 /* Replace the relocation token with ' ', so that
3198 errors like foo@GOTOFF1 will be detected. */
3199
3200 /* The length of the first part of our input line. */
3201 first = cp - input_line_pointer;
3202
3203 /* The second part goes from after the reloc token until
3204 (and including) an end_of_line char. Don't use strlen
3205 here as the end_of_line char may not be a NUL. */
3206 past_reloc = cp + 1 + len;
3207 for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
3208 ;
3209 second = cp - past_reloc;
3210
3211 /* Allocate and copy string. The trailing NUL shouldn't
3212 be necessary, but be safe. */
3213 tmpbuf = xmalloc (first + second + 2);
3214 memcpy (tmpbuf, input_line_pointer, first);
3215 tmpbuf[first] = ' ';
3216 memcpy (tmpbuf + first + 1, past_reloc, second);
3217 tmpbuf[first + second + 1] = '\0';
3218 return tmpbuf;
3219 }
3220
3221 as_bad (_("@%s reloc is not supported in %s bit mode"),
3222 gotrel[j].str, mode_name[(unsigned int) flag_code]);
3223 return NULL;
3224 }
3225 }
3226
3227 /* Might be a symbol version string. Don't as_bad here. */
3228 return NULL;
3229 }
3230
3231 /* x86_cons_fix_new is called via the expression parsing code when a
3232 reloc is needed. We use this hook to get the correct .got reloc. */
3233 static RELOC_ENUM got_reloc = NO_RELOC;
3234
3235 void
3236 x86_cons_fix_new (frag, off, len, exp)
3237 fragS *frag;
3238 unsigned int off;
3239 unsigned int len;
3240 expressionS *exp;
3241 {
3242 RELOC_ENUM r = reloc (len, 0, 0, got_reloc);
3243 got_reloc = NO_RELOC;
3244 fix_new_exp (frag, off, len, exp, 0, r);
3245 }
3246
3247 void
3248 x86_cons (exp, size)
3249 expressionS *exp;
3250 int size;
3251 {
3252 if (size == 4)
3253 {
3254 /* Handle @GOTOFF and the like in an expression. */
3255 char *save;
3256 char *gotfree_input_line;
3257 int adjust;
3258
3259 save = input_line_pointer;
3260 gotfree_input_line = lex_got (&got_reloc, &adjust);
3261 if (gotfree_input_line)
3262 input_line_pointer = gotfree_input_line;
3263
3264 expression (exp);
3265
3266 if (gotfree_input_line)
3267 {
3268 /* expression () has merrily parsed up to the end of line,
3269 or a comma - in the wrong buffer. Transfer how far
3270 input_line_pointer has moved to the right buffer. */
3271 input_line_pointer = (save
3272 + (input_line_pointer - gotfree_input_line)
3273 + adjust);
3274 free (gotfree_input_line);
3275 }
3276 }
3277 else
3278 expression (exp);
3279 }
3280 #endif
3281
3282 static int i386_immediate PARAMS ((char *));
3283
3284 static int
3285 i386_immediate (imm_start)
3286 char *imm_start;
3287 {
3288 char *save_input_line_pointer;
3289 #ifndef LEX_AT
3290 char *gotfree_input_line;
3291 #endif
3292 segT exp_seg = 0;
3293 expressionS *exp;
3294
3295 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
3296 {
3297 as_bad (_("only 1 or 2 immediate operands are allowed"));
3298 return 0;
3299 }
3300
3301 exp = &im_expressions[i.imm_operands++];
3302 i.op[this_operand].imms = exp;
3303
3304 if (is_space_char (*imm_start))
3305 ++imm_start;
3306
3307 save_input_line_pointer = input_line_pointer;
3308 input_line_pointer = imm_start;
3309
3310 #ifndef LEX_AT
3311 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3312 if (gotfree_input_line)
3313 input_line_pointer = gotfree_input_line;
3314 #endif
3315
3316 exp_seg = expression (exp);
3317
3318 SKIP_WHITESPACE ();
3319 if (*input_line_pointer)
3320 as_bad (_("junk `%s' after expression"), input_line_pointer);
3321
3322 input_line_pointer = save_input_line_pointer;
3323 #ifndef LEX_AT
3324 if (gotfree_input_line)
3325 free (gotfree_input_line);
3326 #endif
3327
3328 if (exp->X_op == O_absent || exp->X_op == O_big)
3329 {
3330 /* Missing or bad expr becomes absolute 0. */
3331 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
3332 imm_start);
3333 exp->X_op = O_constant;
3334 exp->X_add_number = 0;
3335 exp->X_add_symbol = (symbolS *) 0;
3336 exp->X_op_symbol = (symbolS *) 0;
3337 }
3338 else if (exp->X_op == O_constant)
3339 {
3340 /* Size it properly later. */
3341 i.types[this_operand] |= Imm64;
3342 /* If BFD64, sign extend val. */
3343 if (!use_rela_relocations)
3344 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
3345 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
3346 }
3347 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3348 else if (1
3349 #ifdef BFD_ASSEMBLER
3350 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3351 #endif
3352 && exp_seg != text_section
3353 && exp_seg != data_section
3354 && exp_seg != bss_section
3355 && exp_seg != undefined_section
3356 #ifdef BFD_ASSEMBLER
3357 && !bfd_is_com_section (exp_seg)
3358 #endif
3359 )
3360 {
3361 #ifdef BFD_ASSEMBLER
3362 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3363 #else
3364 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3365 #endif
3366 return 0;
3367 }
3368 #endif
3369 else
3370 {
3371 /* This is an address. The size of the address will be
3372 determined later, depending on destination register,
3373 suffix, or the default for the section. */
3374 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
3375 }
3376
3377 return 1;
3378 }
3379
3380 static char *i386_scale PARAMS ((char *));
3381
3382 static char *
3383 i386_scale (scale)
3384 char *scale;
3385 {
3386 offsetT val;
3387 char *save = input_line_pointer;
3388
3389 input_line_pointer = scale;
3390 val = get_absolute_expression ();
3391
3392 switch (val)
3393 {
3394 case 0:
3395 case 1:
3396 i.log2_scale_factor = 0;
3397 break;
3398 case 2:
3399 i.log2_scale_factor = 1;
3400 break;
3401 case 4:
3402 i.log2_scale_factor = 2;
3403 break;
3404 case 8:
3405 i.log2_scale_factor = 3;
3406 break;
3407 default:
3408 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
3409 scale);
3410 input_line_pointer = save;
3411 return NULL;
3412 }
3413 if (i.log2_scale_factor != 0 && ! i.index_reg)
3414 {
3415 as_warn (_("scale factor of %d without an index register"),
3416 1 << i.log2_scale_factor);
3417 #if SCALE1_WHEN_NO_INDEX
3418 i.log2_scale_factor = 0;
3419 #endif
3420 }
3421 scale = input_line_pointer;
3422 input_line_pointer = save;
3423 return scale;
3424 }
3425
3426 static int i386_displacement PARAMS ((char *, char *));
3427
3428 static int
3429 i386_displacement (disp_start, disp_end)
3430 char *disp_start;
3431 char *disp_end;
3432 {
3433 register expressionS *exp;
3434 segT exp_seg = 0;
3435 char *save_input_line_pointer;
3436 #ifndef LEX_AT
3437 char *gotfree_input_line;
3438 #endif
3439 int bigdisp = Disp32;
3440
3441 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3442 bigdisp = Disp16;
3443 if (flag_code == CODE_64BIT)
3444 bigdisp = Disp64;
3445 i.types[this_operand] |= bigdisp;
3446
3447 exp = &disp_expressions[i.disp_operands];
3448 i.op[this_operand].disps = exp;
3449 i.disp_operands++;
3450 save_input_line_pointer = input_line_pointer;
3451 input_line_pointer = disp_start;
3452 END_STRING_AND_SAVE (disp_end);
3453
3454 #ifndef GCC_ASM_O_HACK
3455 #define GCC_ASM_O_HACK 0
3456 #endif
3457 #if GCC_ASM_O_HACK
3458 END_STRING_AND_SAVE (disp_end + 1);
3459 if ((i.types[this_operand] & BaseIndex) != 0
3460 && displacement_string_end[-1] == '+')
3461 {
3462 /* This hack is to avoid a warning when using the "o"
3463 constraint within gcc asm statements.
3464 For instance:
3465
3466 #define _set_tssldt_desc(n,addr,limit,type) \
3467 __asm__ __volatile__ ( \
3468 "movw %w2,%0\n\t" \
3469 "movw %w1,2+%0\n\t" \
3470 "rorl $16,%1\n\t" \
3471 "movb %b1,4+%0\n\t" \
3472 "movb %4,5+%0\n\t" \
3473 "movb $0,6+%0\n\t" \
3474 "movb %h1,7+%0\n\t" \
3475 "rorl $16,%1" \
3476 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
3477
3478 This works great except that the output assembler ends
3479 up looking a bit weird if it turns out that there is
3480 no offset. You end up producing code that looks like:
3481
3482 #APP
3483 movw $235,(%eax)
3484 movw %dx,2+(%eax)
3485 rorl $16,%edx
3486 movb %dl,4+(%eax)
3487 movb $137,5+(%eax)
3488 movb $0,6+(%eax)
3489 movb %dh,7+(%eax)
3490 rorl $16,%edx
3491 #NO_APP
3492
3493 So here we provide the missing zero. */
3494
3495 *displacement_string_end = '0';
3496 }
3497 #endif
3498 #ifndef LEX_AT
3499 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3500 if (gotfree_input_line)
3501 input_line_pointer = gotfree_input_line;
3502 #endif
3503
3504 exp_seg = expression (exp);
3505
3506 SKIP_WHITESPACE ();
3507 if (*input_line_pointer)
3508 as_bad (_("junk `%s' after expression"), input_line_pointer);
3509 #if GCC_ASM_O_HACK
3510 RESTORE_END_STRING (disp_end + 1);
3511 #endif
3512 RESTORE_END_STRING (disp_end);
3513 input_line_pointer = save_input_line_pointer;
3514 #ifndef LEX_AT
3515 if (gotfree_input_line)
3516 free (gotfree_input_line);
3517 #endif
3518
3519 #ifdef BFD_ASSEMBLER
3520 /* We do this to make sure that the section symbol is in
3521 the symbol table. We will ultimately change the relocation
3522 to be relative to the beginning of the section. */
3523 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
3524 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3525 {
3526 if (exp->X_op != O_symbol)
3527 {
3528 as_bad (_("bad expression used with @%s"),
3529 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
3530 ? "GOTPCREL"
3531 : "GOTOFF"));
3532 return 0;
3533 }
3534
3535 if (S_IS_LOCAL (exp->X_add_symbol)
3536 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
3537 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
3538 exp->X_op = O_subtract;
3539 exp->X_op_symbol = GOT_symbol;
3540 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3541 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
3542 else
3543 i.reloc[this_operand] = BFD_RELOC_32;
3544 }
3545 #endif
3546
3547 if (exp->X_op == O_absent || exp->X_op == O_big)
3548 {
3549 /* Missing or bad expr becomes absolute 0. */
3550 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
3551 disp_start);
3552 exp->X_op = O_constant;
3553 exp->X_add_number = 0;
3554 exp->X_add_symbol = (symbolS *) 0;
3555 exp->X_op_symbol = (symbolS *) 0;
3556 }
3557
3558 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3559 if (exp->X_op != O_constant
3560 #ifdef BFD_ASSEMBLER
3561 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3562 #endif
3563 && exp_seg != text_section
3564 && exp_seg != data_section
3565 && exp_seg != bss_section
3566 && exp_seg != undefined_section)
3567 {
3568 #ifdef BFD_ASSEMBLER
3569 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3570 #else
3571 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3572 #endif
3573 return 0;
3574 }
3575 #endif
3576 else if (flag_code == CODE_64BIT)
3577 i.types[this_operand] |= Disp32S | Disp32;
3578 return 1;
3579 }
3580
3581 static int i386_index_check PARAMS ((const char *));
3582
3583 /* Make sure the memory operand we've been dealt is valid.
3584 Return 1 on success, 0 on a failure. */
3585
3586 static int
3587 i386_index_check (operand_string)
3588 const char *operand_string;
3589 {
3590 int ok;
3591 #if INFER_ADDR_PREFIX
3592 int fudged = 0;
3593
3594 tryprefix:
3595 #endif
3596 ok = 1;
3597 if (flag_code == CODE_64BIT)
3598 {
3599 /* 64bit checks. */
3600 if ((i.base_reg
3601 && ((i.base_reg->reg_type & Reg64) == 0)
3602 && (i.base_reg->reg_type != BaseIndex
3603 || i.index_reg))
3604 || (i.index_reg
3605 && ((i.index_reg->reg_type & (Reg64|BaseIndex))
3606 != (Reg64|BaseIndex))))
3607 ok = 0;
3608 }
3609 else
3610 {
3611 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3612 {
3613 /* 16bit checks. */
3614 if ((i.base_reg
3615 && ((i.base_reg->reg_type & (Reg16|BaseIndex|RegRex))
3616 != (Reg16|BaseIndex)))
3617 || (i.index_reg
3618 && (((i.index_reg->reg_type & (Reg16|BaseIndex))
3619 != (Reg16|BaseIndex))
3620 || ! (i.base_reg
3621 && i.base_reg->reg_num < 6
3622 && i.index_reg->reg_num >= 6
3623 && i.log2_scale_factor == 0))))
3624 ok = 0;
3625 }
3626 else
3627 {
3628 /* 32bit checks. */
3629 if ((i.base_reg
3630 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
3631 || (i.index_reg
3632 && ((i.index_reg->reg_type & (Reg32|BaseIndex|RegRex))
3633 != (Reg32|BaseIndex))))
3634 ok = 0;
3635 }
3636 }
3637 if (!ok)
3638 {
3639 #if INFER_ADDR_PREFIX
3640 if (flag_code != CODE_64BIT
3641 && i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0')
3642 {
3643 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
3644 i.prefixes += 1;
3645 /* Change the size of any displacement too. At most one of
3646 Disp16 or Disp32 is set.
3647 FIXME. There doesn't seem to be any real need for separate
3648 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
3649 Removing them would probably clean up the code quite a lot. */
3650 if (i.types[this_operand] & (Disp16|Disp32))
3651 i.types[this_operand] ^= (Disp16|Disp32);
3652 fudged = 1;
3653 goto tryprefix;
3654 }
3655 if (fudged)
3656 as_bad (_("`%s' is not a valid base/index expression"),
3657 operand_string);
3658 else
3659 #endif
3660 as_bad (_("`%s' is not a valid %s bit base/index expression"),
3661 operand_string,
3662 flag_code_names[flag_code]);
3663 return 0;
3664 }
3665 return 1;
3666 }
3667
3668 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
3669 on error. */
3670
3671 static int
3672 i386_operand (operand_string)
3673 char *operand_string;
3674 {
3675 const reg_entry *r;
3676 char *end_op;
3677 char *op_string = operand_string;
3678
3679 if (is_space_char (*op_string))
3680 ++op_string;
3681
3682 /* We check for an absolute prefix (differentiating,
3683 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
3684 if (*op_string == ABSOLUTE_PREFIX)
3685 {
3686 ++op_string;
3687 if (is_space_char (*op_string))
3688 ++op_string;
3689 i.types[this_operand] |= JumpAbsolute;
3690 }
3691
3692 /* Check if operand is a register. */
3693 if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
3694 && (r = parse_register (op_string, &end_op)) != NULL)
3695 {
3696 /* Check for a segment override by searching for ':' after a
3697 segment register. */
3698 op_string = end_op;
3699 if (is_space_char (*op_string))
3700 ++op_string;
3701 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
3702 {
3703 switch (r->reg_num)
3704 {
3705 case 0:
3706 i.seg[i.mem_operands] = &es;
3707 break;
3708 case 1:
3709 i.seg[i.mem_operands] = &cs;
3710 break;
3711 case 2:
3712 i.seg[i.mem_operands] = &ss;
3713 break;
3714 case 3:
3715 i.seg[i.mem_operands] = &ds;
3716 break;
3717 case 4:
3718 i.seg[i.mem_operands] = &fs;
3719 break;
3720 case 5:
3721 i.seg[i.mem_operands] = &gs;
3722 break;
3723 }
3724
3725 /* Skip the ':' and whitespace. */
3726 ++op_string;
3727 if (is_space_char (*op_string))
3728 ++op_string;
3729
3730 if (!is_digit_char (*op_string)
3731 && !is_identifier_char (*op_string)
3732 && *op_string != '('
3733 && *op_string != ABSOLUTE_PREFIX)
3734 {
3735 as_bad (_("bad memory operand `%s'"), op_string);
3736 return 0;
3737 }
3738 /* Handle case of %es:*foo. */
3739 if (*op_string == ABSOLUTE_PREFIX)
3740 {
3741 ++op_string;
3742 if (is_space_char (*op_string))
3743 ++op_string;
3744 i.types[this_operand] |= JumpAbsolute;
3745 }
3746 goto do_memory_reference;
3747 }
3748 if (*op_string)
3749 {
3750 as_bad (_("junk `%s' after register"), op_string);
3751 return 0;
3752 }
3753 i.types[this_operand] |= r->reg_type & ~BaseIndex;
3754 i.op[this_operand].regs = r;
3755 i.reg_operands++;
3756 }
3757 else if (*op_string == REGISTER_PREFIX)
3758 {
3759 as_bad (_("bad register name `%s'"), op_string);
3760 return 0;
3761 }
3762 else if (*op_string == IMMEDIATE_PREFIX)
3763 {
3764 ++op_string;
3765 if (i.types[this_operand] & JumpAbsolute)
3766 {
3767 as_bad (_("immediate operand illegal with absolute jump"));
3768 return 0;
3769 }
3770 if (!i386_immediate (op_string))
3771 return 0;
3772 }
3773 else if (is_digit_char (*op_string)
3774 || is_identifier_char (*op_string)
3775 || *op_string == '(')
3776 {
3777 /* This is a memory reference of some sort. */
3778 char *base_string;
3779
3780 /* Start and end of displacement string expression (if found). */
3781 char *displacement_string_start;
3782 char *displacement_string_end;
3783
3784 do_memory_reference:
3785 if ((i.mem_operands == 1
3786 && (current_templates->start->opcode_modifier & IsString) == 0)
3787 || i.mem_operands == 2)
3788 {
3789 as_bad (_("too many memory references for `%s'"),
3790 current_templates->start->name);
3791 return 0;
3792 }
3793
3794 /* Check for base index form. We detect the base index form by
3795 looking for an ')' at the end of the operand, searching
3796 for the '(' matching it, and finding a REGISTER_PREFIX or ','
3797 after the '('. */
3798 base_string = op_string + strlen (op_string);
3799
3800 --base_string;
3801 if (is_space_char (*base_string))
3802 --base_string;
3803
3804 /* If we only have a displacement, set-up for it to be parsed later. */
3805 displacement_string_start = op_string;
3806 displacement_string_end = base_string + 1;
3807
3808 if (*base_string == ')')
3809 {
3810 char *temp_string;
3811 unsigned int parens_balanced = 1;
3812 /* We've already checked that the number of left & right ()'s are
3813 equal, so this loop will not be infinite. */
3814 do
3815 {
3816 base_string--;
3817 if (*base_string == ')')
3818 parens_balanced++;
3819 if (*base_string == '(')
3820 parens_balanced--;
3821 }
3822 while (parens_balanced);
3823
3824 temp_string = base_string;
3825
3826 /* Skip past '(' and whitespace. */
3827 ++base_string;
3828 if (is_space_char (*base_string))
3829 ++base_string;
3830
3831 if (*base_string == ','
3832 || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3833 && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
3834 {
3835 displacement_string_end = temp_string;
3836
3837 i.types[this_operand] |= BaseIndex;
3838
3839 if (i.base_reg)
3840 {
3841 base_string = end_op;
3842 if (is_space_char (*base_string))
3843 ++base_string;
3844 }
3845
3846 /* There may be an index reg or scale factor here. */
3847 if (*base_string == ',')
3848 {
3849 ++base_string;
3850 if (is_space_char (*base_string))
3851 ++base_string;
3852
3853 if ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3854 && (i.index_reg = parse_register (base_string, &end_op)) != NULL)
3855 {
3856 base_string = end_op;
3857 if (is_space_char (*base_string))
3858 ++base_string;
3859 if (*base_string == ',')
3860 {
3861 ++base_string;
3862 if (is_space_char (*base_string))
3863 ++base_string;
3864 }
3865 else if (*base_string != ')')
3866 {
3867 as_bad (_("expecting `,' or `)' after index register in `%s'"),
3868 operand_string);
3869 return 0;
3870 }
3871 }
3872 else if (*base_string == REGISTER_PREFIX)
3873 {
3874 as_bad (_("bad register name `%s'"), base_string);
3875 return 0;
3876 }
3877
3878 /* Check for scale factor. */
3879 if (*base_string != ')')
3880 {
3881 char *end_scale = i386_scale (base_string);
3882
3883 if (!end_scale)
3884 return 0;
3885
3886 base_string = end_scale;
3887 if (is_space_char (*base_string))
3888 ++base_string;
3889 if (*base_string != ')')
3890 {
3891 as_bad (_("expecting `)' after scale factor in `%s'"),
3892 operand_string);
3893 return 0;
3894 }
3895 }
3896 else if (!i.index_reg)
3897 {
3898 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
3899 *base_string);
3900 return 0;
3901 }
3902 }
3903 else if (*base_string != ')')
3904 {
3905 as_bad (_("expecting `,' or `)' after base register in `%s'"),
3906 operand_string);
3907 return 0;
3908 }
3909 }
3910 else if (*base_string == REGISTER_PREFIX)
3911 {
3912 as_bad (_("bad register name `%s'"), base_string);
3913 return 0;
3914 }
3915 }
3916
3917 /* If there's an expression beginning the operand, parse it,
3918 assuming displacement_string_start and
3919 displacement_string_end are meaningful. */
3920 if (displacement_string_start != displacement_string_end)
3921 {
3922 if (!i386_displacement (displacement_string_start,
3923 displacement_string_end))
3924 return 0;
3925 }
3926
3927 /* Special case for (%dx) while doing input/output op. */
3928 if (i.base_reg
3929 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
3930 && i.index_reg == 0
3931 && i.log2_scale_factor == 0
3932 && i.seg[i.mem_operands] == 0
3933 && (i.types[this_operand] & Disp) == 0)
3934 {
3935 i.types[this_operand] = InOutPortReg;
3936 return 1;
3937 }
3938
3939 if (i386_index_check (operand_string) == 0)
3940 return 0;
3941 i.mem_operands++;
3942 }
3943 else
3944 {
3945 /* It's not a memory operand; argh! */
3946 as_bad (_("invalid char %s beginning operand %d `%s'"),
3947 output_invalid (*op_string),
3948 this_operand + 1,
3949 op_string);
3950 return 0;
3951 }
3952 return 1; /* Normal return. */
3953 }
3954 \f
3955 /* md_estimate_size_before_relax()
3956
3957 Called just before relax() for rs_machine_dependent frags. The x86
3958 assembler uses these frags to handle variable size jump
3959 instructions.
3960
3961 Any symbol that is now undefined will not become defined.
3962 Return the correct fr_subtype in the frag.
3963 Return the initial "guess for variable size of frag" to caller.
3964 The guess is actually the growth beyond the fixed part. Whatever
3965 we do to grow the fixed or variable part contributes to our
3966 returned value. */
3967
3968 int
3969 md_estimate_size_before_relax (fragP, segment)
3970 register fragS *fragP;
3971 register segT segment;
3972 {
3973 /* We've already got fragP->fr_subtype right; all we have to do is
3974 check for un-relaxable symbols. On an ELF system, we can't relax
3975 an externally visible symbol, because it may be overridden by a
3976 shared library. */
3977 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
3978 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
3979 || S_IS_EXTERNAL (fragP->fr_symbol)
3980 || S_IS_WEAK (fragP->fr_symbol)
3981 #endif
3982 )
3983 {
3984 /* Symbol is undefined in this segment, or we need to keep a
3985 reloc so that weak symbols can be overridden. */
3986 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
3987 RELOC_ENUM reloc_type;
3988 unsigned char *opcode;
3989 int old_fr_fix;
3990
3991 if (fragP->fr_var != NO_RELOC)
3992 reloc_type = fragP->fr_var;
3993 else if (size == 2)
3994 reloc_type = BFD_RELOC_16_PCREL;
3995 else
3996 reloc_type = BFD_RELOC_32_PCREL;
3997
3998 old_fr_fix = fragP->fr_fix;
3999 opcode = (unsigned char *) fragP->fr_opcode;
4000
4001 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
4002 {
4003 case UNCOND_JUMP:
4004 /* Make jmp (0xeb) a (d)word displacement jump. */
4005 opcode[0] = 0xe9;
4006 fragP->fr_fix += size;
4007 fix_new (fragP, old_fr_fix, size,
4008 fragP->fr_symbol,
4009 fragP->fr_offset, 1,
4010 reloc_type);
4011 break;
4012
4013 case COND_JUMP86:
4014 if (no_cond_jump_promotion)
4015 goto relax_guess;
4016
4017 if (size == 2)
4018 {
4019 /* Negate the condition, and branch past an
4020 unconditional jump. */
4021 opcode[0] ^= 1;
4022 opcode[1] = 3;
4023 /* Insert an unconditional jump. */
4024 opcode[2] = 0xe9;
4025 /* We added two extra opcode bytes, and have a two byte
4026 offset. */
4027 fragP->fr_fix += 2 + 2;
4028 fix_new (fragP, old_fr_fix + 2, 2,
4029 fragP->fr_symbol,
4030 fragP->fr_offset, 1,
4031 reloc_type);
4032 break;
4033 }
4034 /* Fall through. */
4035
4036 case COND_JUMP:
4037 if (no_cond_jump_promotion)
4038 goto relax_guess;
4039
4040 /* This changes the byte-displacement jump 0x7N
4041 to the (d)word-displacement jump 0x0f,0x8N. */
4042 opcode[1] = opcode[0] + 0x10;
4043 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4044 /* We've added an opcode byte. */
4045 fragP->fr_fix += 1 + size;
4046 fix_new (fragP, old_fr_fix + 1, size,
4047 fragP->fr_symbol,
4048 fragP->fr_offset, 1,
4049 reloc_type);
4050 break;
4051
4052 default:
4053 BAD_CASE (fragP->fr_subtype);
4054 break;
4055 }
4056 frag_wane (fragP);
4057 return fragP->fr_fix - old_fr_fix;
4058 }
4059
4060 relax_guess:
4061 /* Guess size depending on current relax state. Initially the relax
4062 state will correspond to a short jump and we return 1, because
4063 the variable part of the frag (the branch offset) is one byte
4064 long. However, we can relax a section more than once and in that
4065 case we must either set fr_subtype back to the unrelaxed state,
4066 or return the value for the appropriate branch. */
4067 return md_relax_table[fragP->fr_subtype].rlx_length;
4068 }
4069
4070 /* Called after relax() is finished.
4071
4072 In: Address of frag.
4073 fr_type == rs_machine_dependent.
4074 fr_subtype is what the address relaxed to.
4075
4076 Out: Any fixSs and constants are set up.
4077 Caller will turn frag into a ".space 0". */
4078
4079 #ifndef BFD_ASSEMBLER
4080 void
4081 md_convert_frag (headers, sec, fragP)
4082 object_headers *headers ATTRIBUTE_UNUSED;
4083 segT sec ATTRIBUTE_UNUSED;
4084 register fragS *fragP;
4085 #else
4086 void
4087 md_convert_frag (abfd, sec, fragP)
4088 bfd *abfd ATTRIBUTE_UNUSED;
4089 segT sec ATTRIBUTE_UNUSED;
4090 register fragS *fragP;
4091 #endif
4092 {
4093 register unsigned char *opcode;
4094 unsigned char *where_to_put_displacement = NULL;
4095 offsetT target_address;
4096 offsetT opcode_address;
4097 unsigned int extension = 0;
4098 offsetT displacement_from_opcode_start;
4099
4100 opcode = (unsigned char *) fragP->fr_opcode;
4101
4102 /* Address we want to reach in file space. */
4103 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
4104
4105 /* Address opcode resides at in file space. */
4106 opcode_address = fragP->fr_address + fragP->fr_fix;
4107
4108 /* Displacement from opcode start to fill into instruction. */
4109 displacement_from_opcode_start = target_address - opcode_address;
4110
4111 if ((fragP->fr_subtype & BIG) == 0)
4112 {
4113 /* Don't have to change opcode. */
4114 extension = 1; /* 1 opcode + 1 displacement */
4115 where_to_put_displacement = &opcode[1];
4116 }
4117 else
4118 {
4119 if (no_cond_jump_promotion
4120 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
4121 as_warn_where (fragP->fr_file, fragP->fr_line, _("long jump required"));
4122
4123 switch (fragP->fr_subtype)
4124 {
4125 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
4126 extension = 4; /* 1 opcode + 4 displacement */
4127 opcode[0] = 0xe9;
4128 where_to_put_displacement = &opcode[1];
4129 break;
4130
4131 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
4132 extension = 2; /* 1 opcode + 2 displacement */
4133 opcode[0] = 0xe9;
4134 where_to_put_displacement = &opcode[1];
4135 break;
4136
4137 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
4138 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
4139 extension = 5; /* 2 opcode + 4 displacement */
4140 opcode[1] = opcode[0] + 0x10;
4141 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4142 where_to_put_displacement = &opcode[2];
4143 break;
4144
4145 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
4146 extension = 3; /* 2 opcode + 2 displacement */
4147 opcode[1] = opcode[0] + 0x10;
4148 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4149 where_to_put_displacement = &opcode[2];
4150 break;
4151
4152 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
4153 extension = 4;
4154 opcode[0] ^= 1;
4155 opcode[1] = 3;
4156 opcode[2] = 0xe9;
4157 where_to_put_displacement = &opcode[3];
4158 break;
4159
4160 default:
4161 BAD_CASE (fragP->fr_subtype);
4162 break;
4163 }
4164 }
4165
4166 /* Now put displacement after opcode. */
4167 md_number_to_chars ((char *) where_to_put_displacement,
4168 (valueT) (displacement_from_opcode_start - extension),
4169 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
4170 fragP->fr_fix += extension;
4171 }
4172 \f
4173 /* Size of byte displacement jmp. */
4174 int md_short_jump_size = 2;
4175
4176 /* Size of dword displacement jmp. */
4177 int md_long_jump_size = 5;
4178
4179 /* Size of relocation record. */
4180 const int md_reloc_size = 8;
4181
4182 void
4183 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
4184 char *ptr;
4185 addressT from_addr, to_addr;
4186 fragS *frag ATTRIBUTE_UNUSED;
4187 symbolS *to_symbol ATTRIBUTE_UNUSED;
4188 {
4189 offsetT offset;
4190
4191 offset = to_addr - (from_addr + 2);
4192 /* Opcode for byte-disp jump. */
4193 md_number_to_chars (ptr, (valueT) 0xeb, 1);
4194 md_number_to_chars (ptr + 1, (valueT) offset, 1);
4195 }
4196
4197 void
4198 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
4199 char *ptr;
4200 addressT from_addr, to_addr;
4201 fragS *frag ATTRIBUTE_UNUSED;
4202 symbolS *to_symbol ATTRIBUTE_UNUSED;
4203 {
4204 offsetT offset;
4205
4206 offset = to_addr - (from_addr + 5);
4207 md_number_to_chars (ptr, (valueT) 0xe9, 1);
4208 md_number_to_chars (ptr + 1, (valueT) offset, 4);
4209 }
4210 \f
4211 /* Apply a fixup (fixS) to segment data, once it has been determined
4212 by our caller that we have all the info we need to fix it up.
4213
4214 On the 386, immediates, displacements, and data pointers are all in
4215 the same (little-endian) format, so we don't need to care about which
4216 we are handling. */
4217
4218 void
4219 md_apply_fix3 (fixP, valP, seg)
4220 /* The fix we're to put in. */
4221 fixS *fixP;
4222 /* Pointer to the value of the bits. */
4223 valueT * valP;
4224 /* Segment fix is from. */
4225 segT seg ATTRIBUTE_UNUSED;
4226 {
4227 char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
4228 valueT value = * valP;
4229
4230 #if defined (BFD_ASSEMBLER) && !defined (TE_Mach)
4231 if (fixP->fx_pcrel)
4232 {
4233 switch (fixP->fx_r_type)
4234 {
4235 default:
4236 break;
4237
4238 case BFD_RELOC_32:
4239 fixP->fx_r_type = BFD_RELOC_32_PCREL;
4240 break;
4241 case BFD_RELOC_16:
4242 fixP->fx_r_type = BFD_RELOC_16_PCREL;
4243 break;
4244 case BFD_RELOC_8:
4245 fixP->fx_r_type = BFD_RELOC_8_PCREL;
4246 break;
4247 }
4248 }
4249
4250 /* This is a hack. There should be a better way to handle this.
4251 This covers for the fact that bfd_install_relocation will
4252 subtract the current location (for partial_inplace, PC relative
4253 relocations); see more below. */
4254 if ((fixP->fx_r_type == BFD_RELOC_32_PCREL
4255 || fixP->fx_r_type == BFD_RELOC_16_PCREL
4256 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
4257 && fixP->fx_addsy && !use_rela_relocations)
4258 {
4259 #ifndef OBJ_AOUT
4260 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4261 #ifdef TE_PE
4262 || OUTPUT_FLAVOR == bfd_target_coff_flavour
4263 #endif
4264 )
4265 value += fixP->fx_where + fixP->fx_frag->fr_address;
4266 #endif
4267 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4268 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
4269 {
4270 segT fseg = S_GET_SEGMENT (fixP->fx_addsy);
4271
4272 if ((fseg == seg
4273 || (symbol_section_p (fixP->fx_addsy)
4274 && fseg != absolute_section))
4275 && ! S_IS_EXTERNAL (fixP->fx_addsy)
4276 && ! S_IS_WEAK (fixP->fx_addsy)
4277 && S_IS_DEFINED (fixP->fx_addsy)
4278 && ! S_IS_COMMON (fixP->fx_addsy))
4279 {
4280 /* Yes, we add the values in twice. This is because
4281 bfd_perform_relocation subtracts them out again. I think
4282 bfd_perform_relocation is broken, but I don't dare change
4283 it. FIXME. */
4284 value += fixP->fx_where + fixP->fx_frag->fr_address;
4285 }
4286 }
4287 #endif
4288 #if defined (OBJ_COFF) && defined (TE_PE)
4289 /* For some reason, the PE format does not store a section
4290 address offset for a PC relative symbol. */
4291 if (S_GET_SEGMENT (fixP->fx_addsy) != seg)
4292 value += md_pcrel_from (fixP);
4293 #endif
4294 }
4295
4296 /* Fix a few things - the dynamic linker expects certain values here,
4297 and we must not dissappoint it. */
4298 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4299 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4300 && fixP->fx_addsy)
4301 switch (fixP->fx_r_type)
4302 {
4303 case BFD_RELOC_386_PLT32:
4304 case BFD_RELOC_X86_64_PLT32:
4305 /* Make the jump instruction point to the address of the operand. At
4306 runtime we merely add the offset to the actual PLT entry. */
4307 value = -4;
4308 break;
4309 case BFD_RELOC_386_GOTPC:
4310
4311 /* This is tough to explain. We end up with this one if we have
4312 * operands that look like "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal
4313 * here is to obtain the absolute address of the GOT, and it is strongly
4314 * preferable from a performance point of view to avoid using a runtime
4315 * relocation for this. The actual sequence of instructions often look
4316 * something like:
4317 *
4318 * call .L66
4319 * .L66:
4320 * popl %ebx
4321 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
4322 *
4323 * The call and pop essentially return the absolute address of
4324 * the label .L66 and store it in %ebx. The linker itself will
4325 * ultimately change the first operand of the addl so that %ebx points to
4326 * the GOT, but to keep things simple, the .o file must have this operand
4327 * set so that it generates not the absolute address of .L66, but the
4328 * absolute address of itself. This allows the linker itself simply
4329 * treat a GOTPC relocation as asking for a pcrel offset to the GOT to be
4330 * added in, and the addend of the relocation is stored in the operand
4331 * field for the instruction itself.
4332 *
4333 * Our job here is to fix the operand so that it would add the correct
4334 * offset so that %ebx would point to itself. The thing that is tricky is
4335 * that .-.L66 will point to the beginning of the instruction, so we need
4336 * to further modify the operand so that it will point to itself.
4337 * There are other cases where you have something like:
4338 *
4339 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
4340 *
4341 * and here no correction would be required. Internally in the assembler
4342 * we treat operands of this form as not being pcrel since the '.' is
4343 * explicitly mentioned, and I wonder whether it would simplify matters
4344 * to do it this way. Who knows. In earlier versions of the PIC patches,
4345 * the pcrel_adjust field was used to store the correction, but since the
4346 * expression is not pcrel, I felt it would be confusing to do it this
4347 * way. */
4348
4349 value -= 1;
4350 break;
4351 case BFD_RELOC_386_GOT32:
4352 case BFD_RELOC_X86_64_GOT32:
4353 value = 0; /* Fully resolved at runtime. No addend. */
4354 break;
4355 case BFD_RELOC_386_GOTOFF:
4356 case BFD_RELOC_X86_64_GOTPCREL:
4357 break;
4358
4359 case BFD_RELOC_VTABLE_INHERIT:
4360 case BFD_RELOC_VTABLE_ENTRY:
4361 fixP->fx_done = 0;
4362 return;
4363
4364 default:
4365 break;
4366 }
4367 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
4368 * valP = value;
4369 #endif /* defined (BFD_ASSEMBLER) && !defined (TE_Mach) */
4370
4371 /* Are we finished with this relocation now? */
4372 if (fixP->fx_addsy == NULL && fixP->fx_pcrel == 0)
4373 fixP->fx_done = 1;
4374 #ifdef BFD_ASSEMBLER
4375 else if (use_rela_relocations)
4376 {
4377 fixP->fx_no_overflow = 1;
4378 value = 0;
4379 }
4380 #endif
4381 md_number_to_chars (p, value, fixP->fx_size);
4382 }
4383 \f
4384 #define MAX_LITTLENUMS 6
4385
4386 /* Turn the string pointed to by litP into a floating point constant
4387 of type TYPE, and emit the appropriate bytes. The number of
4388 LITTLENUMS emitted is stored in *SIZEP. An error message is
4389 returned, or NULL on OK. */
4390
4391 char *
4392 md_atof (type, litP, sizeP)
4393 int type;
4394 char *litP;
4395 int *sizeP;
4396 {
4397 int prec;
4398 LITTLENUM_TYPE words[MAX_LITTLENUMS];
4399 LITTLENUM_TYPE *wordP;
4400 char *t;
4401
4402 switch (type)
4403 {
4404 case 'f':
4405 case 'F':
4406 prec = 2;
4407 break;
4408
4409 case 'd':
4410 case 'D':
4411 prec = 4;
4412 break;
4413
4414 case 'x':
4415 case 'X':
4416 prec = 5;
4417 break;
4418
4419 default:
4420 *sizeP = 0;
4421 return _("Bad call to md_atof ()");
4422 }
4423 t = atof_ieee (input_line_pointer, type, words);
4424 if (t)
4425 input_line_pointer = t;
4426
4427 *sizeP = prec * sizeof (LITTLENUM_TYPE);
4428 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
4429 the bigendian 386. */
4430 for (wordP = words + prec - 1; prec--;)
4431 {
4432 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
4433 litP += sizeof (LITTLENUM_TYPE);
4434 }
4435 return 0;
4436 }
4437 \f
4438 char output_invalid_buf[8];
4439
4440 static char *
4441 output_invalid (c)
4442 int c;
4443 {
4444 if (ISPRINT (c))
4445 sprintf (output_invalid_buf, "'%c'", c);
4446 else
4447 sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
4448 return output_invalid_buf;
4449 }
4450
4451 /* REG_STRING starts *before* REGISTER_PREFIX. */
4452
4453 static const reg_entry *
4454 parse_register (reg_string, end_op)
4455 char *reg_string;
4456 char **end_op;
4457 {
4458 char *s = reg_string;
4459 char *p;
4460 char reg_name_given[MAX_REG_NAME_SIZE + 1];
4461 const reg_entry *r;
4462
4463 /* Skip possible REGISTER_PREFIX and possible whitespace. */
4464 if (*s == REGISTER_PREFIX)
4465 ++s;
4466
4467 if (is_space_char (*s))
4468 ++s;
4469
4470 p = reg_name_given;
4471 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
4472 {
4473 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
4474 return (const reg_entry *) NULL;
4475 s++;
4476 }
4477
4478 /* For naked regs, make sure that we are not dealing with an identifier.
4479 This prevents confusing an identifier like `eax_var' with register
4480 `eax'. */
4481 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
4482 return (const reg_entry *) NULL;
4483
4484 *end_op = s;
4485
4486 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
4487
4488 /* Handle floating point regs, allowing spaces in the (i) part. */
4489 if (r == i386_regtab /* %st is first entry of table */)
4490 {
4491 if (is_space_char (*s))
4492 ++s;
4493 if (*s == '(')
4494 {
4495 ++s;
4496 if (is_space_char (*s))
4497 ++s;
4498 if (*s >= '0' && *s <= '7')
4499 {
4500 r = &i386_float_regtab[*s - '0'];
4501 ++s;
4502 if (is_space_char (*s))
4503 ++s;
4504 if (*s == ')')
4505 {
4506 *end_op = s + 1;
4507 return r;
4508 }
4509 }
4510 /* We have "%st(" then garbage. */
4511 return (const reg_entry *) NULL;
4512 }
4513 }
4514
4515 if (r != NULL
4516 && r->reg_flags & (RegRex64|RegRex)
4517 && flag_code != CODE_64BIT)
4518 {
4519 return (const reg_entry *) NULL;
4520 }
4521
4522 return r;
4523 }
4524 \f
4525 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4526 const char *md_shortopts = "kVQ:sq";
4527 #else
4528 const char *md_shortopts = "q";
4529 #endif
4530
4531 struct option md_longopts[] = {
4532 #define OPTION_32 (OPTION_MD_BASE + 0)
4533 {"32", no_argument, NULL, OPTION_32},
4534 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4535 #define OPTION_64 (OPTION_MD_BASE + 1)
4536 {"64", no_argument, NULL, OPTION_64},
4537 #endif
4538 {NULL, no_argument, NULL, 0}
4539 };
4540 size_t md_longopts_size = sizeof (md_longopts);
4541
4542 int
4543 md_parse_option (c, arg)
4544 int c;
4545 char *arg ATTRIBUTE_UNUSED;
4546 {
4547 switch (c)
4548 {
4549 case 'q':
4550 quiet_warnings = 1;
4551 break;
4552
4553 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4554 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
4555 should be emitted or not. FIXME: Not implemented. */
4556 case 'Q':
4557 break;
4558
4559 /* -V: SVR4 argument to print version ID. */
4560 case 'V':
4561 print_version_id ();
4562 break;
4563
4564 /* -k: Ignore for FreeBSD compatibility. */
4565 case 'k':
4566 break;
4567
4568 case 's':
4569 /* -s: On i386 Solaris, this tells the native assembler to use
4570 .stab instead of .stab.excl. We always use .stab anyhow. */
4571 break;
4572
4573 case OPTION_64:
4574 {
4575 const char **list, **l;
4576
4577 list = bfd_target_list ();
4578 for (l = list; *l != NULL; l++)
4579 if (strcmp (*l, "elf64-x86-64") == 0)
4580 {
4581 default_arch = "x86_64";
4582 break;
4583 }
4584 if (*l == NULL)
4585 as_fatal (_("No compiled in support for x86_64"));
4586 free (list);
4587 }
4588 break;
4589 #endif
4590
4591 case OPTION_32:
4592 default_arch = "i386";
4593 break;
4594
4595 default:
4596 return 0;
4597 }
4598 return 1;
4599 }
4600
4601 void
4602 md_show_usage (stream)
4603 FILE *stream;
4604 {
4605 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4606 fprintf (stream, _("\
4607 -Q ignored\n\
4608 -V print assembler version number\n\
4609 -k ignored\n\
4610 -q quieten some warnings\n\
4611 -s ignored\n"));
4612 #else
4613 fprintf (stream, _("\
4614 -q quieten some warnings\n"));
4615 #endif
4616 }
4617
4618 #ifdef BFD_ASSEMBLER
4619 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
4620 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
4621
4622 /* Pick the target format to use. */
4623
4624 const char *
4625 i386_target_format ()
4626 {
4627 if (!strcmp (default_arch, "x86_64"))
4628 set_code_flag (CODE_64BIT);
4629 else if (!strcmp (default_arch, "i386"))
4630 set_code_flag (CODE_32BIT);
4631 else
4632 as_fatal (_("Unknown architecture"));
4633 switch (OUTPUT_FLAVOR)
4634 {
4635 #ifdef OBJ_MAYBE_AOUT
4636 case bfd_target_aout_flavour:
4637 return AOUT_TARGET_FORMAT;
4638 #endif
4639 #ifdef OBJ_MAYBE_COFF
4640 case bfd_target_coff_flavour:
4641 return "coff-i386";
4642 #endif
4643 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
4644 case bfd_target_elf_flavour:
4645 {
4646 if (flag_code == CODE_64BIT)
4647 use_rela_relocations = 1;
4648 return flag_code == CODE_64BIT ? "elf64-x86-64" : "elf32-i386";
4649 }
4650 #endif
4651 default:
4652 abort ();
4653 return NULL;
4654 }
4655 }
4656
4657 #endif /* OBJ_MAYBE_ more than one */
4658
4659 #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
4660 void i386_elf_emit_arch_note ()
4661 {
4662 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4663 && cpu_arch_name != NULL)
4664 {
4665 char *p;
4666 asection *seg = now_seg;
4667 subsegT subseg = now_subseg;
4668 Elf_Internal_Note i_note;
4669 Elf_External_Note e_note;
4670 asection *note_secp;
4671 int len;
4672
4673 /* Create the .note section. */
4674 note_secp = subseg_new (".note", 0);
4675 bfd_set_section_flags (stdoutput,
4676 note_secp,
4677 SEC_HAS_CONTENTS | SEC_READONLY);
4678
4679 /* Process the arch string. */
4680 len = strlen (cpu_arch_name);
4681
4682 i_note.namesz = len + 1;
4683 i_note.descsz = 0;
4684 i_note.type = NT_ARCH;
4685 p = frag_more (sizeof (e_note.namesz));
4686 md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
4687 p = frag_more (sizeof (e_note.descsz));
4688 md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
4689 p = frag_more (sizeof (e_note.type));
4690 md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
4691 p = frag_more (len + 1);
4692 strcpy (p, cpu_arch_name);
4693
4694 frag_align (2, 0, 0);
4695
4696 subseg_set (seg, subseg);
4697 }
4698 }
4699 #endif
4700 #endif /* BFD_ASSEMBLER */
4701 \f
4702 symbolS *
4703 md_undefined_symbol (name)
4704 char *name;
4705 {
4706 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
4707 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
4708 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
4709 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
4710 {
4711 if (!GOT_symbol)
4712 {
4713 if (symbol_find (name))
4714 as_bad (_("GOT already in symbol table"));
4715 GOT_symbol = symbol_new (name, undefined_section,
4716 (valueT) 0, &zero_address_frag);
4717 };
4718 return GOT_symbol;
4719 }
4720 return 0;
4721 }
4722
4723 /* Round up a section size to the appropriate boundary. */
4724
4725 valueT
4726 md_section_align (segment, size)
4727 segT segment ATTRIBUTE_UNUSED;
4728 valueT size;
4729 {
4730 #ifdef BFD_ASSEMBLER
4731 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4732 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
4733 {
4734 /* For a.out, force the section size to be aligned. If we don't do
4735 this, BFD will align it for us, but it will not write out the
4736 final bytes of the section. This may be a bug in BFD, but it is
4737 easier to fix it here since that is how the other a.out targets
4738 work. */
4739 int align;
4740
4741 align = bfd_get_section_alignment (stdoutput, segment);
4742 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
4743 }
4744 #endif
4745 #endif
4746
4747 return size;
4748 }
4749
4750 /* On the i386, PC-relative offsets are relative to the start of the
4751 next instruction. That is, the address of the offset, plus its
4752 size, since the offset is always the last part of the insn. */
4753
4754 long
4755 md_pcrel_from (fixP)
4756 fixS *fixP;
4757 {
4758 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
4759 }
4760
4761 #ifndef I386COFF
4762
4763 static void
4764 s_bss (ignore)
4765 int ignore ATTRIBUTE_UNUSED;
4766 {
4767 register int temp;
4768
4769 temp = get_absolute_expression ();
4770 subseg_set (bss_section, (subsegT) temp);
4771 demand_empty_rest_of_line ();
4772 }
4773
4774 #endif
4775
4776 #ifdef BFD_ASSEMBLER
4777
4778 void
4779 i386_validate_fix (fixp)
4780 fixS *fixp;
4781 {
4782 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
4783 {
4784 /* GOTOFF relocation are nonsense in 64bit mode. */
4785 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
4786 {
4787 if (flag_code != CODE_64BIT)
4788 abort ();
4789 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
4790 }
4791 else
4792 {
4793 if (flag_code == CODE_64BIT)
4794 abort ();
4795 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
4796 }
4797 fixp->fx_subsy = 0;
4798 }
4799 }
4800
4801 arelent *
4802 tc_gen_reloc (section, fixp)
4803 asection *section ATTRIBUTE_UNUSED;
4804 fixS *fixp;
4805 {
4806 arelent *rel;
4807 bfd_reloc_code_real_type code;
4808
4809 switch (fixp->fx_r_type)
4810 {
4811 case BFD_RELOC_X86_64_PLT32:
4812 case BFD_RELOC_X86_64_GOT32:
4813 case BFD_RELOC_X86_64_GOTPCREL:
4814 case BFD_RELOC_386_PLT32:
4815 case BFD_RELOC_386_GOT32:
4816 case BFD_RELOC_386_GOTOFF:
4817 case BFD_RELOC_386_GOTPC:
4818 case BFD_RELOC_X86_64_32S:
4819 case BFD_RELOC_RVA:
4820 case BFD_RELOC_VTABLE_ENTRY:
4821 case BFD_RELOC_VTABLE_INHERIT:
4822 code = fixp->fx_r_type;
4823 break;
4824 default:
4825 if (fixp->fx_pcrel)
4826 {
4827 switch (fixp->fx_size)
4828 {
4829 default:
4830 as_bad_where (fixp->fx_file, fixp->fx_line,
4831 _("can not do %d byte pc-relative relocation"),
4832 fixp->fx_size);
4833 code = BFD_RELOC_32_PCREL;
4834 break;
4835 case 1: code = BFD_RELOC_8_PCREL; break;
4836 case 2: code = BFD_RELOC_16_PCREL; break;
4837 case 4: code = BFD_RELOC_32_PCREL; break;
4838 }
4839 }
4840 else
4841 {
4842 switch (fixp->fx_size)
4843 {
4844 default:
4845 as_bad_where (fixp->fx_file, fixp->fx_line,
4846 _("can not do %d byte relocation"),
4847 fixp->fx_size);
4848 code = BFD_RELOC_32;
4849 break;
4850 case 1: code = BFD_RELOC_8; break;
4851 case 2: code = BFD_RELOC_16; break;
4852 case 4: code = BFD_RELOC_32; break;
4853 case 8: code = BFD_RELOC_64; break;
4854 }
4855 }
4856 break;
4857 }
4858
4859 if (code == BFD_RELOC_32
4860 && GOT_symbol
4861 && fixp->fx_addsy == GOT_symbol)
4862 {
4863 /* We don't support GOTPC on 64bit targets. */
4864 if (flag_code == CODE_64BIT)
4865 abort ();
4866 code = BFD_RELOC_386_GOTPC;
4867 }
4868
4869 rel = (arelent *) xmalloc (sizeof (arelent));
4870 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
4871 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
4872
4873 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
4874 if (!use_rela_relocations)
4875 {
4876 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
4877 vtable entry to be used in the relocation's section offset. */
4878 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
4879 rel->address = fixp->fx_offset;
4880
4881 if (fixp->fx_pcrel)
4882 rel->addend = fixp->fx_addnumber;
4883 else
4884 rel->addend = 0;
4885 }
4886 /* Use the rela in 64bit mode. */
4887 else
4888 {
4889 rel->addend = fixp->fx_offset;
4890 if (fixp->fx_pcrel)
4891 rel->addend -= fixp->fx_size;
4892 }
4893
4894 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
4895 if (rel->howto == NULL)
4896 {
4897 as_bad_where (fixp->fx_file, fixp->fx_line,
4898 _("cannot represent relocation type %s"),
4899 bfd_get_reloc_code_name (code));
4900 /* Set howto to a garbage value so that we can keep going. */
4901 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
4902 assert (rel->howto != NULL);
4903 }
4904
4905 return rel;
4906 }
4907
4908 #else /* ! BFD_ASSEMBLER */
4909
4910 #if (defined(OBJ_AOUT) | defined(OBJ_BOUT))
4911 void
4912 tc_aout_fix_to_chars (where, fixP, segment_address_in_file)
4913 char *where;
4914 fixS *fixP;
4915 relax_addressT segment_address_in_file;
4916 {
4917 /* In: length of relocation (or of address) in chars: 1, 2 or 4.
4918 Out: GNU LD relocation length code: 0, 1, or 2. */
4919
4920 static const unsigned char nbytes_r_length[] = { 42, 0, 1, 42, 2 };
4921 long r_symbolnum;
4922
4923 know (fixP->fx_addsy != NULL);
4924
4925 md_number_to_chars (where,
4926 (valueT) (fixP->fx_frag->fr_address
4927 + fixP->fx_where - segment_address_in_file),
4928 4);
4929
4930 r_symbolnum = (S_IS_DEFINED (fixP->fx_addsy)
4931 ? S_GET_TYPE (fixP->fx_addsy)
4932 : fixP->fx_addsy->sy_number);
4933
4934 where[6] = (r_symbolnum >> 16) & 0x0ff;
4935 where[5] = (r_symbolnum >> 8) & 0x0ff;
4936 where[4] = r_symbolnum & 0x0ff;
4937 where[7] = ((((!S_IS_DEFINED (fixP->fx_addsy)) << 3) & 0x08)
4938 | ((nbytes_r_length[fixP->fx_size] << 1) & 0x06)
4939 | (((fixP->fx_pcrel << 0) & 0x01) & 0x0f));
4940 }
4941
4942 #endif /* OBJ_AOUT or OBJ_BOUT. */
4943
4944 #if defined (I386COFF)
4945
4946 short
4947 tc_coff_fix2rtype (fixP)
4948 fixS *fixP;
4949 {
4950 if (fixP->fx_r_type == R_IMAGEBASE)
4951 return R_IMAGEBASE;
4952
4953 return (fixP->fx_pcrel ?
4954 (fixP->fx_size == 1 ? R_PCRBYTE :
4955 fixP->fx_size == 2 ? R_PCRWORD :
4956 R_PCRLONG) :
4957 (fixP->fx_size == 1 ? R_RELBYTE :
4958 fixP->fx_size == 2 ? R_RELWORD :
4959 R_DIR32));
4960 }
4961
4962 int
4963 tc_coff_sizemachdep (frag)
4964 fragS *frag;
4965 {
4966 if (frag->fr_next)
4967 return (frag->fr_next->fr_address - frag->fr_address);
4968 else
4969 return 0;
4970 }
4971
4972 #endif /* I386COFF */
4973
4974 #endif /* ! BFD_ASSEMBLER */
4975 \f
4976 /* Parse operands using Intel syntax. This implements a recursive descent
4977 parser based on the BNF grammar published in Appendix B of the MASM 6.1
4978 Programmer's Guide.
4979
4980 FIXME: We do not recognize the full operand grammar defined in the MASM
4981 documentation. In particular, all the structure/union and
4982 high-level macro operands are missing.
4983
4984 Uppercase words are terminals, lower case words are non-terminals.
4985 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
4986 bars '|' denote choices. Most grammar productions are implemented in
4987 functions called 'intel_<production>'.
4988
4989 Initial production is 'expr'.
4990
4991 addOp + | -
4992
4993 alpha [a-zA-Z]
4994
4995 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
4996
4997 constant digits [[ radixOverride ]]
4998
4999 dataType BYTE | WORD | DWORD | QWORD | XWORD
5000
5001 digits decdigit
5002 | digits decdigit
5003 | digits hexdigit
5004
5005 decdigit [0-9]
5006
5007 e05 e05 addOp e06
5008 | e06
5009
5010 e06 e06 mulOp e09
5011 | e09
5012
5013 e09 OFFSET e10
5014 | e09 PTR e10
5015 | e09 : e10
5016 | e10
5017
5018 e10 e10 [ expr ]
5019 | e11
5020
5021 e11 ( expr )
5022 | [ expr ]
5023 | constant
5024 | dataType
5025 | id
5026 | $
5027 | register
5028
5029 => expr SHORT e05
5030 | e05
5031
5032 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
5033 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
5034
5035 hexdigit a | b | c | d | e | f
5036 | A | B | C | D | E | F
5037
5038 id alpha
5039 | id alpha
5040 | id decdigit
5041
5042 mulOp * | / | MOD
5043
5044 quote " | '
5045
5046 register specialRegister
5047 | gpRegister
5048 | byteRegister
5049
5050 segmentRegister CS | DS | ES | FS | GS | SS
5051
5052 specialRegister CR0 | CR2 | CR3
5053 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
5054 | TR3 | TR4 | TR5 | TR6 | TR7
5055
5056 We simplify the grammar in obvious places (e.g., register parsing is
5057 done by calling parse_register) and eliminate immediate left recursion
5058 to implement a recursive-descent parser.
5059
5060 expr SHORT e05
5061 | e05
5062
5063 e05 e06 e05'
5064
5065 e05' addOp e06 e05'
5066 | Empty
5067
5068 e06 e09 e06'
5069
5070 e06' mulOp e09 e06'
5071 | Empty
5072
5073 e09 OFFSET e10 e09'
5074 | e10 e09'
5075
5076 e09' PTR e10 e09'
5077 | : e10 e09'
5078 | Empty
5079
5080 e10 e11 e10'
5081
5082 e10' [ expr ] e10'
5083 | Empty
5084
5085 e11 ( expr )
5086 | [ expr ]
5087 | BYTE
5088 | WORD
5089 | DWORD
5090 | QWORD
5091 | XWORD
5092 | .
5093 | $
5094 | register
5095 | id
5096 | constant */
5097
5098 /* Parsing structure for the intel syntax parser. Used to implement the
5099 semantic actions for the operand grammar. */
5100 struct intel_parser_s
5101 {
5102 char *op_string; /* The string being parsed. */
5103 int got_a_float; /* Whether the operand is a float. */
5104 int op_modifier; /* Operand modifier. */
5105 int is_mem; /* 1 if operand is memory reference. */
5106 const reg_entry *reg; /* Last register reference found. */
5107 char *disp; /* Displacement string being built. */
5108 };
5109
5110 static struct intel_parser_s intel_parser;
5111
5112 /* Token structure for parsing intel syntax. */
5113 struct intel_token
5114 {
5115 int code; /* Token code. */
5116 const reg_entry *reg; /* Register entry for register tokens. */
5117 char *str; /* String representation. */
5118 };
5119
5120 static struct intel_token cur_token, prev_token;
5121
5122 /* Token codes for the intel parser. Since T_SHORT is already used
5123 by COFF, undefine it first to prevent a warning. */
5124 #define T_NIL -1
5125 #define T_CONST 1
5126 #define T_REG 2
5127 #define T_BYTE 3
5128 #define T_WORD 4
5129 #define T_DWORD 5
5130 #define T_QWORD 6
5131 #define T_XWORD 7
5132 #undef T_SHORT
5133 #define T_SHORT 8
5134 #define T_OFFSET 9
5135 #define T_PTR 10
5136 #define T_ID 11
5137
5138 /* Prototypes for intel parser functions. */
5139 static int intel_match_token PARAMS ((int code));
5140 static void intel_get_token PARAMS ((void));
5141 static void intel_putback_token PARAMS ((void));
5142 static int intel_expr PARAMS ((void));
5143 static int intel_e05 PARAMS ((void));
5144 static int intel_e05_1 PARAMS ((void));
5145 static int intel_e06 PARAMS ((void));
5146 static int intel_e06_1 PARAMS ((void));
5147 static int intel_e09 PARAMS ((void));
5148 static int intel_e09_1 PARAMS ((void));
5149 static int intel_e10 PARAMS ((void));
5150 static int intel_e10_1 PARAMS ((void));
5151 static int intel_e11 PARAMS ((void));
5152
5153 static int
5154 i386_intel_operand (operand_string, got_a_float)
5155 char *operand_string;
5156 int got_a_float;
5157 {
5158 int ret;
5159 char *p;
5160
5161 /* Initialize token holders. */
5162 cur_token.code = prev_token.code = T_NIL;
5163 cur_token.reg = prev_token.reg = NULL;
5164 cur_token.str = prev_token.str = NULL;
5165
5166 /* Initialize parser structure. */
5167 p = intel_parser.op_string = (char *) malloc (strlen (operand_string) + 1);
5168 if (p == NULL)
5169 abort ();
5170 strcpy (intel_parser.op_string, operand_string);
5171 intel_parser.got_a_float = got_a_float;
5172 intel_parser.op_modifier = -1;
5173 intel_parser.is_mem = 0;
5174 intel_parser.reg = NULL;
5175 intel_parser.disp = (char *) malloc (strlen (operand_string) + 1);
5176 if (intel_parser.disp == NULL)
5177 abort ();
5178 intel_parser.disp[0] = '\0';
5179
5180 /* Read the first token and start the parser. */
5181 intel_get_token ();
5182 ret = intel_expr ();
5183
5184 if (ret)
5185 {
5186 /* If we found a memory reference, hand it over to i386_displacement
5187 to fill in the rest of the operand fields. */
5188 if (intel_parser.is_mem)
5189 {
5190 if ((i.mem_operands == 1
5191 && (current_templates->start->opcode_modifier & IsString) == 0)
5192 || i.mem_operands == 2)
5193 {
5194 as_bad (_("too many memory references for '%s'"),
5195 current_templates->start->name);
5196 ret = 0;
5197 }
5198 else
5199 {
5200 char *s = intel_parser.disp;
5201 i.mem_operands++;
5202
5203 /* Add the displacement expression. */
5204 if (*s != '\0')
5205 ret = i386_displacement (s, s + strlen (s))
5206 && i386_index_check (s);
5207 }
5208 }
5209
5210 /* Constant and OFFSET expressions are handled by i386_immediate. */
5211 else if (intel_parser.op_modifier == OFFSET_FLAT
5212 || intel_parser.reg == NULL)
5213 ret = i386_immediate (intel_parser.disp);
5214 }
5215
5216 free (p);
5217 free (intel_parser.disp);
5218
5219 return ret;
5220 }
5221
5222 /* expr SHORT e05
5223 | e05 */
5224 static int
5225 intel_expr ()
5226 {
5227 /* expr SHORT e05 */
5228 if (cur_token.code == T_SHORT)
5229 {
5230 intel_parser.op_modifier = SHORT;
5231 intel_match_token (T_SHORT);
5232
5233 return (intel_e05 ());
5234 }
5235
5236 /* expr e05 */
5237 else
5238 return intel_e05 ();
5239 }
5240
5241 /* e05 e06 e05'
5242
5243 e05' addOp e06 e05'
5244 | Empty */
5245 static int
5246 intel_e05 ()
5247 {
5248 return (intel_e06 () && intel_e05_1 ());
5249 }
5250
5251 static int
5252 intel_e05_1 ()
5253 {
5254 /* e05' addOp e06 e05' */
5255 if (cur_token.code == '+' || cur_token.code == '-')
5256 {
5257 strcat (intel_parser.disp, cur_token.str);
5258 intel_match_token (cur_token.code);
5259
5260 return (intel_e06 () && intel_e05_1 ());
5261 }
5262
5263 /* e05' Empty */
5264 else
5265 return 1;
5266 }
5267
5268 /* e06 e09 e06'
5269
5270 e06' mulOp e09 e06'
5271 | Empty */
5272 static int
5273 intel_e06 ()
5274 {
5275 return (intel_e09 () && intel_e06_1 ());
5276 }
5277
5278 static int
5279 intel_e06_1 ()
5280 {
5281 /* e06' mulOp e09 e06' */
5282 if (cur_token.code == '*' || cur_token.code == '/')
5283 {
5284 strcat (intel_parser.disp, cur_token.str);
5285 intel_match_token (cur_token.code);
5286
5287 return (intel_e09 () && intel_e06_1 ());
5288 }
5289
5290 /* e06' Empty */
5291 else
5292 return 1;
5293 }
5294
5295 /* e09 OFFSET e10 e09'
5296 | e10 e09'
5297
5298 e09' PTR e10 e09'
5299 | : e10 e09'
5300 | Empty */
5301 static int
5302 intel_e09 ()
5303 {
5304 /* e09 OFFSET e10 e09' */
5305 if (cur_token.code == T_OFFSET)
5306 {
5307 intel_parser.is_mem = 0;
5308 intel_parser.op_modifier = OFFSET_FLAT;
5309 intel_match_token (T_OFFSET);
5310
5311 return (intel_e10 () && intel_e09_1 ());
5312 }
5313
5314 /* e09 e10 e09' */
5315 else
5316 return (intel_e10 () && intel_e09_1 ());
5317 }
5318
5319 static int
5320 intel_e09_1 ()
5321 {
5322 /* e09' PTR e10 e09' */
5323 if (cur_token.code == T_PTR)
5324 {
5325 if (prev_token.code == T_BYTE)
5326 i.suffix = BYTE_MNEM_SUFFIX;
5327
5328 else if (prev_token.code == T_WORD)
5329 {
5330 if (intel_parser.got_a_float == 2) /* "fi..." */
5331 i.suffix = SHORT_MNEM_SUFFIX;
5332 else
5333 i.suffix = WORD_MNEM_SUFFIX;
5334 }
5335
5336 else if (prev_token.code == T_DWORD)
5337 {
5338 if (intel_parser.got_a_float == 1) /* "f..." */
5339 i.suffix = SHORT_MNEM_SUFFIX;
5340 else
5341 i.suffix = LONG_MNEM_SUFFIX;
5342 }
5343
5344 else if (prev_token.code == T_QWORD)
5345 {
5346 if (intel_parser.got_a_float == 1) /* "f..." */
5347 i.suffix = LONG_MNEM_SUFFIX;
5348 else
5349 i.suffix = QWORD_MNEM_SUFFIX;
5350 }
5351
5352 else if (prev_token.code == T_XWORD)
5353 i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
5354
5355 else
5356 {
5357 as_bad (_("Unknown operand modifier `%s'\n"), prev_token.str);
5358 return 0;
5359 }
5360
5361 intel_match_token (T_PTR);
5362
5363 return (intel_e10 () && intel_e09_1 ());
5364 }
5365
5366 /* e09 : e10 e09' */
5367 else if (cur_token.code == ':')
5368 {
5369 /* Mark as a memory operand only if it's not already known to be an
5370 offset expression. */
5371 if (intel_parser.op_modifier != OFFSET_FLAT)
5372 intel_parser.is_mem = 1;
5373
5374 return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
5375 }
5376
5377 /* e09' Empty */
5378 else
5379 return 1;
5380 }
5381
5382 /* e10 e11 e10'
5383
5384 e10' [ expr ] e10'
5385 | Empty */
5386 static int
5387 intel_e10 ()
5388 {
5389 return (intel_e11 () && intel_e10_1 ());
5390 }
5391
5392 static int
5393 intel_e10_1 ()
5394 {
5395 /* e10' [ expr ] e10' */
5396 if (cur_token.code == '[')
5397 {
5398 intel_match_token ('[');
5399
5400 /* Mark as a memory operand only if it's not already known to be an
5401 offset expression. If it's an offset expression, we need to keep
5402 the brace in. */
5403 if (intel_parser.op_modifier != OFFSET_FLAT)
5404 intel_parser.is_mem = 1;
5405 else
5406 strcat (intel_parser.disp, "[");
5407
5408 /* Add a '+' to the displacement string if necessary. */
5409 if (*intel_parser.disp != '\0'
5410 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5411 strcat (intel_parser.disp, "+");
5412
5413 if (intel_expr () && intel_match_token (']'))
5414 {
5415 /* Preserve brackets when the operand is an offset expression. */
5416 if (intel_parser.op_modifier == OFFSET_FLAT)
5417 strcat (intel_parser.disp, "]");
5418
5419 return intel_e10_1 ();
5420 }
5421 else
5422 return 0;
5423 }
5424
5425 /* e10' Empty */
5426 else
5427 return 1;
5428 }
5429
5430 /* e11 ( expr )
5431 | [ expr ]
5432 | BYTE
5433 | WORD
5434 | DWORD
5435 | QWORD
5436 | XWORD
5437 | $
5438 | .
5439 | register
5440 | id
5441 | constant */
5442 static int
5443 intel_e11 ()
5444 {
5445 /* e11 ( expr ) */
5446 if (cur_token.code == '(')
5447 {
5448 intel_match_token ('(');
5449 strcat (intel_parser.disp, "(");
5450
5451 if (intel_expr () && intel_match_token (')'))
5452 {
5453 strcat (intel_parser.disp, ")");
5454 return 1;
5455 }
5456 else
5457 return 0;
5458 }
5459
5460 /* e11 [ expr ] */
5461 else if (cur_token.code == '[')
5462 {
5463 intel_match_token ('[');
5464
5465 /* Mark as a memory operand only if it's not already known to be an
5466 offset expression. If it's an offset expression, we need to keep
5467 the brace in. */
5468 if (intel_parser.op_modifier != OFFSET_FLAT)
5469 intel_parser.is_mem = 1;
5470 else
5471 strcat (intel_parser.disp, "[");
5472
5473 /* Operands for jump/call inside brackets denote absolute addresses. */
5474 if (current_templates->start->opcode_modifier & Jump
5475 || current_templates->start->opcode_modifier & JumpDword
5476 || current_templates->start->opcode_modifier & JumpByte
5477 || current_templates->start->opcode_modifier & JumpInterSegment)
5478 i.types[this_operand] |= JumpAbsolute;
5479
5480 /* Add a '+' to the displacement string if necessary. */
5481 if (*intel_parser.disp != '\0'
5482 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5483 strcat (intel_parser.disp, "+");
5484
5485 if (intel_expr () && intel_match_token (']'))
5486 {
5487 /* Preserve brackets when the operand is an offset expression. */
5488 if (intel_parser.op_modifier == OFFSET_FLAT)
5489 strcat (intel_parser.disp, "]");
5490
5491 return 1;
5492 }
5493 else
5494 return 0;
5495 }
5496
5497 /* e11 BYTE
5498 | WORD
5499 | DWORD
5500 | QWORD
5501 | XWORD */
5502 else if (cur_token.code == T_BYTE
5503 || cur_token.code == T_WORD
5504 || cur_token.code == T_DWORD
5505 || cur_token.code == T_QWORD
5506 || cur_token.code == T_XWORD)
5507 {
5508 intel_match_token (cur_token.code);
5509
5510 return 1;
5511 }
5512
5513 /* e11 $
5514 | . */
5515 else if (cur_token.code == '$' || cur_token.code == '.')
5516 {
5517 strcat (intel_parser.disp, cur_token.str);
5518 intel_match_token (cur_token.code);
5519
5520 /* Mark as a memory operand only if it's not already known to be an
5521 offset expression. */
5522 if (intel_parser.op_modifier != OFFSET_FLAT)
5523 intel_parser.is_mem = 1;
5524
5525 return 1;
5526 }
5527
5528 /* e11 register */
5529 else if (cur_token.code == T_REG)
5530 {
5531 const reg_entry *reg = intel_parser.reg = cur_token.reg;
5532
5533 intel_match_token (T_REG);
5534
5535 /* Check for segment change. */
5536 if (cur_token.code == ':')
5537 {
5538 if (reg->reg_type & (SReg2 | SReg3))
5539 {
5540 switch (reg->reg_num)
5541 {
5542 case 0:
5543 i.seg[i.mem_operands] = &es;
5544 break;
5545 case 1:
5546 i.seg[i.mem_operands] = &cs;
5547 break;
5548 case 2:
5549 i.seg[i.mem_operands] = &ss;
5550 break;
5551 case 3:
5552 i.seg[i.mem_operands] = &ds;
5553 break;
5554 case 4:
5555 i.seg[i.mem_operands] = &fs;
5556 break;
5557 case 5:
5558 i.seg[i.mem_operands] = &gs;
5559 break;
5560 }
5561 }
5562 else
5563 {
5564 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
5565 return 0;
5566 }
5567 }
5568
5569 /* Not a segment register. Check for register scaling. */
5570 else if (cur_token.code == '*')
5571 {
5572 if (!intel_parser.is_mem)
5573 {
5574 as_bad (_("Register scaling only allowed in memory operands."));
5575 return 0;
5576 }
5577
5578 /* What follows must be a valid scale. */
5579 if (intel_match_token ('*')
5580 && strchr ("01248", *cur_token.str))
5581 {
5582 i.index_reg = reg;
5583 i.types[this_operand] |= BaseIndex;
5584
5585 /* Set the scale after setting the register (otherwise,
5586 i386_scale will complain) */
5587 i386_scale (cur_token.str);
5588 intel_match_token (T_CONST);
5589 }
5590 else
5591 {
5592 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
5593 cur_token.str);
5594 return 0;
5595 }
5596 }
5597
5598 /* No scaling. If this is a memory operand, the register is either a
5599 base register (first occurrence) or an index register (second
5600 occurrence). */
5601 else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3)))
5602 {
5603 if (i.base_reg && i.index_reg)
5604 {
5605 as_bad (_("Too many register references in memory operand.\n"));
5606 return 0;
5607 }
5608
5609 if (i.base_reg == NULL)
5610 i.base_reg = reg;
5611 else
5612 i.index_reg = reg;
5613
5614 i.types[this_operand] |= BaseIndex;
5615 }
5616
5617 /* Offset modifier. Add the register to the displacement string to be
5618 parsed as an immediate expression after we're done. */
5619 else if (intel_parser.op_modifier == OFFSET_FLAT)
5620 strcat (intel_parser.disp, reg->reg_name);
5621
5622 /* It's neither base nor index nor offset. */
5623 else
5624 {
5625 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
5626 i.op[this_operand].regs = reg;
5627 i.reg_operands++;
5628 }
5629
5630 /* Since registers are not part of the displacement string (except
5631 when we're parsing offset operands), we may need to remove any
5632 preceding '+' from the displacement string. */
5633 if (*intel_parser.disp != '\0'
5634 && intel_parser.op_modifier != OFFSET_FLAT)
5635 {
5636 char *s = intel_parser.disp;
5637 s += strlen (s) - 1;
5638 if (*s == '+')
5639 *s = '\0';
5640 }
5641
5642 return 1;
5643 }
5644
5645 /* e11 id */
5646 else if (cur_token.code == T_ID)
5647 {
5648 /* Add the identifier to the displacement string. */
5649 strcat (intel_parser.disp, cur_token.str);
5650 intel_match_token (T_ID);
5651
5652 /* The identifier represents a memory reference only if it's not
5653 preceded by an offset modifier. */
5654 if (intel_parser.op_modifier != OFFSET_FLAT)
5655 intel_parser.is_mem = 1;
5656
5657 return 1;
5658 }
5659
5660 /* e11 constant */
5661 else if (cur_token.code == T_CONST
5662 || cur_token.code == '-'
5663 || cur_token.code == '+')
5664 {
5665 char *save_str;
5666
5667 /* Allow constants that start with `+' or `-'. */
5668 if (cur_token.code == '-' || cur_token.code == '+')
5669 {
5670 strcat (intel_parser.disp, cur_token.str);
5671 intel_match_token (cur_token.code);
5672 if (cur_token.code != T_CONST)
5673 {
5674 as_bad (_("Syntax error. Expecting a constant. Got `%s'.\n"),
5675 cur_token.str);
5676 return 0;
5677 }
5678 }
5679
5680 save_str = (char *) malloc (strlen (cur_token.str) + 1);
5681 if (save_str == NULL)
5682 abort ();
5683 strcpy (save_str, cur_token.str);
5684
5685 /* Get the next token to check for register scaling. */
5686 intel_match_token (cur_token.code);
5687
5688 /* Check if this constant is a scaling factor for an index register. */
5689 if (cur_token.code == '*')
5690 {
5691 if (intel_match_token ('*') && cur_token.code == T_REG)
5692 {
5693 if (!intel_parser.is_mem)
5694 {
5695 as_bad (_("Register scaling only allowed in memory operands."));
5696 return 0;
5697 }
5698
5699 /* The constant is followed by `* reg', so it must be
5700 a valid scale. */
5701 if (strchr ("01248", *save_str))
5702 {
5703 i.index_reg = cur_token.reg;
5704 i.types[this_operand] |= BaseIndex;
5705
5706 /* Set the scale after setting the register (otherwise,
5707 i386_scale will complain) */
5708 i386_scale (save_str);
5709 intel_match_token (T_REG);
5710
5711 /* Since registers are not part of the displacement
5712 string, we may need to remove any preceding '+' from
5713 the displacement string. */
5714 if (*intel_parser.disp != '\0')
5715 {
5716 char *s = intel_parser.disp;
5717 s += strlen (s) - 1;
5718 if (*s == '+')
5719 *s = '\0';
5720 }
5721
5722 free (save_str);
5723
5724 return 1;
5725 }
5726 else
5727 return 0;
5728 }
5729
5730 /* The constant was not used for register scaling. Since we have
5731 already consumed the token following `*' we now need to put it
5732 back in the stream. */
5733 else
5734 intel_putback_token ();
5735 }
5736
5737 /* Add the constant to the displacement string. */
5738 strcat (intel_parser.disp, save_str);
5739 free (save_str);
5740
5741 return 1;
5742 }
5743
5744 as_bad (_("Unrecognized token '%s'"), cur_token.str);
5745 return 0;
5746 }
5747
5748 /* Match the given token against cur_token. If they match, read the next
5749 token from the operand string. */
5750 static int
5751 intel_match_token (code)
5752 int code;
5753 {
5754 if (cur_token.code == code)
5755 {
5756 intel_get_token ();
5757 return 1;
5758 }
5759 else
5760 {
5761 as_bad (_("Unexpected token `%s'\n"), cur_token.str);
5762 return 0;
5763 }
5764 }
5765
5766 /* Read a new token from intel_parser.op_string and store it in cur_token. */
5767 static void
5768 intel_get_token ()
5769 {
5770 char *end_op;
5771 const reg_entry *reg;
5772 struct intel_token new_token;
5773
5774 new_token.code = T_NIL;
5775 new_token.reg = NULL;
5776 new_token.str = NULL;
5777
5778 /* Free the memory allocated to the previous token and move
5779 cur_token to prev_token. */
5780 if (prev_token.str)
5781 free (prev_token.str);
5782
5783 prev_token = cur_token;
5784
5785 /* Skip whitespace. */
5786 while (is_space_char (*intel_parser.op_string))
5787 intel_parser.op_string++;
5788
5789 /* Return an empty token if we find nothing else on the line. */
5790 if (*intel_parser.op_string == '\0')
5791 {
5792 cur_token = new_token;
5793 return;
5794 }
5795
5796 /* The new token cannot be larger than the remainder of the operand
5797 string. */
5798 new_token.str = (char *) malloc (strlen (intel_parser.op_string) + 1);
5799 if (new_token.str == NULL)
5800 abort ();
5801 new_token.str[0] = '\0';
5802
5803 if (strchr ("0123456789", *intel_parser.op_string))
5804 {
5805 char *p = new_token.str;
5806 char *q = intel_parser.op_string;
5807 new_token.code = T_CONST;
5808
5809 /* Allow any kind of identifier char to encompass floating point and
5810 hexadecimal numbers. */
5811 while (is_identifier_char (*q))
5812 *p++ = *q++;
5813 *p = '\0';
5814
5815 /* Recognize special symbol names [0-9][bf]. */
5816 if (strlen (intel_parser.op_string) == 2
5817 && (intel_parser.op_string[1] == 'b'
5818 || intel_parser.op_string[1] == 'f'))
5819 new_token.code = T_ID;
5820 }
5821
5822 else if (strchr ("+-/*:[]()", *intel_parser.op_string))
5823 {
5824 new_token.code = *intel_parser.op_string;
5825 new_token.str[0] = *intel_parser.op_string;
5826 new_token.str[1] = '\0';
5827 }
5828
5829 else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
5830 && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
5831 {
5832 new_token.code = T_REG;
5833 new_token.reg = reg;
5834
5835 if (*intel_parser.op_string == REGISTER_PREFIX)
5836 {
5837 new_token.str[0] = REGISTER_PREFIX;
5838 new_token.str[1] = '\0';
5839 }
5840
5841 strcat (new_token.str, reg->reg_name);
5842 }
5843
5844 else if (is_identifier_char (*intel_parser.op_string))
5845 {
5846 char *p = new_token.str;
5847 char *q = intel_parser.op_string;
5848
5849 /* A '.' or '$' followed by an identifier char is an identifier.
5850 Otherwise, it's operator '.' followed by an expression. */
5851 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
5852 {
5853 new_token.code = *q;
5854 new_token.str[0] = *q;
5855 new_token.str[1] = '\0';
5856 }
5857 else
5858 {
5859 while (is_identifier_char (*q) || *q == '@')
5860 *p++ = *q++;
5861 *p = '\0';
5862
5863 if (strcasecmp (new_token.str, "BYTE") == 0)
5864 new_token.code = T_BYTE;
5865
5866 else if (strcasecmp (new_token.str, "WORD") == 0)
5867 new_token.code = T_WORD;
5868
5869 else if (strcasecmp (new_token.str, "DWORD") == 0)
5870 new_token.code = T_DWORD;
5871
5872 else if (strcasecmp (new_token.str, "QWORD") == 0)
5873 new_token.code = T_QWORD;
5874
5875 else if (strcasecmp (new_token.str, "XWORD") == 0)
5876 new_token.code = T_XWORD;
5877
5878 else if (strcasecmp (new_token.str, "PTR") == 0)
5879 new_token.code = T_PTR;
5880
5881 else if (strcasecmp (new_token.str, "SHORT") == 0)
5882 new_token.code = T_SHORT;
5883
5884 else if (strcasecmp (new_token.str, "OFFSET") == 0)
5885 {
5886 new_token.code = T_OFFSET;
5887
5888 /* ??? This is not mentioned in the MASM grammar but gcc
5889 makes use of it with -mintel-syntax. OFFSET may be
5890 followed by FLAT: */
5891 if (strncasecmp (q, " FLAT:", 6) == 0)
5892 strcat (new_token.str, " FLAT:");
5893 }
5894
5895 /* ??? This is not mentioned in the MASM grammar. */
5896 else if (strcasecmp (new_token.str, "FLAT") == 0)
5897 new_token.code = T_OFFSET;
5898
5899 else
5900 new_token.code = T_ID;
5901 }
5902 }
5903
5904 else
5905 as_bad (_("Unrecognized token `%s'\n"), intel_parser.op_string);
5906
5907 intel_parser.op_string += strlen (new_token.str);
5908 cur_token = new_token;
5909 }
5910
5911 /* Put cur_token back into the token stream and make cur_token point to
5912 prev_token. */
5913 static void
5914 intel_putback_token ()
5915 {
5916 intel_parser.op_string -= strlen (cur_token.str);
5917 free (cur_token.str);
5918 cur_token = prev_token;
5919
5920 /* Forget prev_token. */
5921 prev_token.code = T_NIL;
5922 prev_token.reg = NULL;
5923 prev_token.str = NULL;
5924 }
This page took 0.155377 seconds and 5 git commands to generate.