cce25d8b3328c1ac260aaf1eaf3a3d3516d1882e
[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 && flag_code != CODE_64BIT
1302 && (((current_templates->start->opcode_modifier & Size32) != 0)
1303 ^ (flag_code == CODE_16BIT)))
1304 {
1305 as_bad (_("redundant %s prefix"),
1306 current_templates->start->name);
1307 return;
1308 }
1309 /* Add prefix, checking for repeated prefixes. */
1310 switch (add_prefix (current_templates->start->base_opcode))
1311 {
1312 case 0:
1313 return;
1314 case 2:
1315 expecting_string_instruction = current_templates->start->name;
1316 break;
1317 }
1318 /* Skip past PREFIX_SEPARATOR and reset token_start. */
1319 token_start = ++l;
1320 }
1321 else
1322 break;
1323 }
1324
1325 if (!current_templates)
1326 {
1327 /* See if we can get a match by trimming off a suffix. */
1328 switch (mnem_p[-1])
1329 {
1330 case WORD_MNEM_SUFFIX:
1331 case BYTE_MNEM_SUFFIX:
1332 case QWORD_MNEM_SUFFIX:
1333 i.suffix = mnem_p[-1];
1334 mnem_p[-1] = '\0';
1335 current_templates = hash_find (op_hash, mnemonic);
1336 break;
1337 case SHORT_MNEM_SUFFIX:
1338 case LONG_MNEM_SUFFIX:
1339 if (!intel_syntax)
1340 {
1341 i.suffix = mnem_p[-1];
1342 mnem_p[-1] = '\0';
1343 current_templates = hash_find (op_hash, mnemonic);
1344 }
1345 break;
1346
1347 /* Intel Syntax. */
1348 case 'd':
1349 if (intel_syntax)
1350 {
1351 if (intel_float_operand (mnemonic))
1352 i.suffix = SHORT_MNEM_SUFFIX;
1353 else
1354 i.suffix = LONG_MNEM_SUFFIX;
1355 mnem_p[-1] = '\0';
1356 current_templates = hash_find (op_hash, mnemonic);
1357 }
1358 break;
1359 }
1360 if (!current_templates)
1361 {
1362 as_bad (_("no such instruction: `%s'"), token_start);
1363 return;
1364 }
1365 }
1366
1367 if (current_templates->start->opcode_modifier & (Jump | JumpByte))
1368 {
1369 /* Check for a branch hint. We allow ",pt" and ",pn" for
1370 predict taken and predict not taken respectively.
1371 I'm not sure that branch hints actually do anything on loop
1372 and jcxz insns (JumpByte) for current Pentium4 chips. They
1373 may work in the future and it doesn't hurt to accept them
1374 now. */
1375 if (l[0] == ',' && l[1] == 'p')
1376 {
1377 if (l[2] == 't')
1378 {
1379 if (! add_prefix (DS_PREFIX_OPCODE))
1380 return;
1381 l += 3;
1382 }
1383 else if (l[2] == 'n')
1384 {
1385 if (! add_prefix (CS_PREFIX_OPCODE))
1386 return;
1387 l += 3;
1388 }
1389 }
1390 }
1391 /* Any other comma loses. */
1392 if (*l == ',')
1393 {
1394 as_bad (_("invalid character %s in mnemonic"),
1395 output_invalid (*l));
1396 return;
1397 }
1398
1399 /* Check if instruction is supported on specified architecture. */
1400 if ((current_templates->start->cpu_flags & ~(Cpu64 | CpuNo64))
1401 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64)))
1402 {
1403 as_warn (_("`%s' is not supported on `%s'"),
1404 current_templates->start->name, cpu_arch_name);
1405 }
1406 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
1407 {
1408 as_warn (_("use .code16 to ensure correct addressing mode"));
1409 }
1410
1411 /* Check for rep/repne without a string instruction. */
1412 if (expecting_string_instruction
1413 && !(current_templates->start->opcode_modifier & IsString))
1414 {
1415 as_bad (_("expecting string instruction after `%s'"),
1416 expecting_string_instruction);
1417 return;
1418 }
1419
1420 /* There may be operands to parse. */
1421 if (*l != END_OF_INSN)
1422 {
1423 /* 1 if operand is pending after ','. */
1424 unsigned int expecting_operand = 0;
1425
1426 /* Non-zero if operand parens not balanced. */
1427 unsigned int paren_not_balanced;
1428
1429 do
1430 {
1431 /* Skip optional white space before operand. */
1432 if (is_space_char (*l))
1433 ++l;
1434 if (!is_operand_char (*l) && *l != END_OF_INSN)
1435 {
1436 as_bad (_("invalid character %s before operand %d"),
1437 output_invalid (*l),
1438 i.operands + 1);
1439 return;
1440 }
1441 token_start = l; /* after white space */
1442 paren_not_balanced = 0;
1443 while (paren_not_balanced || *l != ',')
1444 {
1445 if (*l == END_OF_INSN)
1446 {
1447 if (paren_not_balanced)
1448 {
1449 if (!intel_syntax)
1450 as_bad (_("unbalanced parenthesis in operand %d."),
1451 i.operands + 1);
1452 else
1453 as_bad (_("unbalanced brackets in operand %d."),
1454 i.operands + 1);
1455 return;
1456 }
1457 else
1458 break; /* we are done */
1459 }
1460 else if (!is_operand_char (*l) && !is_space_char (*l))
1461 {
1462 as_bad (_("invalid character %s in operand %d"),
1463 output_invalid (*l),
1464 i.operands + 1);
1465 return;
1466 }
1467 if (!intel_syntax)
1468 {
1469 if (*l == '(')
1470 ++paren_not_balanced;
1471 if (*l == ')')
1472 --paren_not_balanced;
1473 }
1474 else
1475 {
1476 if (*l == '[')
1477 ++paren_not_balanced;
1478 if (*l == ']')
1479 --paren_not_balanced;
1480 }
1481 l++;
1482 }
1483 if (l != token_start)
1484 { /* Yes, we've read in another operand. */
1485 unsigned int operand_ok;
1486 this_operand = i.operands++;
1487 if (i.operands > MAX_OPERANDS)
1488 {
1489 as_bad (_("spurious operands; (%d operands/instruction max)"),
1490 MAX_OPERANDS);
1491 return;
1492 }
1493 /* Now parse operand adding info to 'i' as we go along. */
1494 END_STRING_AND_SAVE (l);
1495
1496 if (intel_syntax)
1497 operand_ok =
1498 i386_intel_operand (token_start,
1499 intel_float_operand (mnemonic));
1500 else
1501 operand_ok = i386_operand (token_start);
1502
1503 RESTORE_END_STRING (l);
1504 if (!operand_ok)
1505 return;
1506 }
1507 else
1508 {
1509 if (expecting_operand)
1510 {
1511 expecting_operand_after_comma:
1512 as_bad (_("expecting operand after ','; got nothing"));
1513 return;
1514 }
1515 if (*l == ',')
1516 {
1517 as_bad (_("expecting operand before ','; got nothing"));
1518 return;
1519 }
1520 }
1521
1522 /* Now *l must be either ',' or END_OF_INSN. */
1523 if (*l == ',')
1524 {
1525 if (*++l == END_OF_INSN)
1526 {
1527 /* Just skip it, if it's \n complain. */
1528 goto expecting_operand_after_comma;
1529 }
1530 expecting_operand = 1;
1531 }
1532 }
1533 while (*l != END_OF_INSN);
1534 }
1535 }
1536
1537 /* Now we've parsed the mnemonic into a set of templates, and have the
1538 operands at hand.
1539
1540 Next, we find a template that matches the given insn,
1541 making sure the overlap of the given operands types is consistent
1542 with the template operand types. */
1543
1544 #define MATCH(overlap, given, template) \
1545 ((overlap & ~JumpAbsolute) \
1546 && ((given) & (BaseIndex|JumpAbsolute)) == ((overlap) & (BaseIndex|JumpAbsolute)))
1547
1548 /* If given types r0 and r1 are registers they must be of the same type
1549 unless the expected operand type register overlap is null.
1550 Note that Acc in a template matches every size of reg. */
1551 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
1552 ( ((g0) & Reg) == 0 || ((g1) & Reg) == 0 || \
1553 ((g0) & Reg) == ((g1) & Reg) || \
1554 ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
1555
1556 {
1557 register unsigned int overlap0, overlap1;
1558 unsigned int overlap2;
1559 unsigned int found_reverse_match;
1560 int suffix_check;
1561
1562 /* All intel opcodes have reversed operands except for "bound" and
1563 "enter". We also don't reverse intersegment "jmp" and "call"
1564 instructions with 2 immediate operands so that the immediate segment
1565 precedes the offset, as it does when in AT&T mode. "enter" and the
1566 intersegment "jmp" and "call" instructions are the only ones that
1567 have two immediate operands. */
1568 if (intel_syntax && i.operands > 1
1569 && (strcmp (mnemonic, "bound") != 0)
1570 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1571 {
1572 union i386_op temp_op;
1573 unsigned int temp_type;
1574 RELOC_ENUM temp_reloc;
1575 int xchg1 = 0;
1576 int xchg2 = 0;
1577
1578 if (i.operands == 2)
1579 {
1580 xchg1 = 0;
1581 xchg2 = 1;
1582 }
1583 else if (i.operands == 3)
1584 {
1585 xchg1 = 0;
1586 xchg2 = 2;
1587 }
1588 temp_type = i.types[xchg2];
1589 i.types[xchg2] = i.types[xchg1];
1590 i.types[xchg1] = temp_type;
1591 temp_op = i.op[xchg2];
1592 i.op[xchg2] = i.op[xchg1];
1593 i.op[xchg1] = temp_op;
1594 temp_reloc = i.reloc[xchg2];
1595 i.reloc[xchg2] = i.reloc[xchg1];
1596 i.reloc[xchg1] = temp_reloc;
1597
1598 if (i.mem_operands == 2)
1599 {
1600 const seg_entry *temp_seg;
1601 temp_seg = i.seg[0];
1602 i.seg[0] = i.seg[1];
1603 i.seg[1] = temp_seg;
1604 }
1605 }
1606
1607 if (i.imm_operands)
1608 {
1609 /* Try to ensure constant immediates are represented in the smallest
1610 opcode possible. */
1611 char guess_suffix = 0;
1612 int op;
1613
1614 if (i.suffix)
1615 guess_suffix = i.suffix;
1616 else if (i.reg_operands)
1617 {
1618 /* Figure out a suffix from the last register operand specified.
1619 We can't do this properly yet, ie. excluding InOutPortReg,
1620 but the following works for instructions with immediates.
1621 In any case, we can't set i.suffix yet. */
1622 for (op = i.operands; --op >= 0;)
1623 if (i.types[op] & Reg)
1624 {
1625 if (i.types[op] & Reg8)
1626 guess_suffix = BYTE_MNEM_SUFFIX;
1627 else if (i.types[op] & Reg16)
1628 guess_suffix = WORD_MNEM_SUFFIX;
1629 else if (i.types[op] & Reg32)
1630 guess_suffix = LONG_MNEM_SUFFIX;
1631 else if (i.types[op] & Reg64)
1632 guess_suffix = QWORD_MNEM_SUFFIX;
1633 break;
1634 }
1635 }
1636 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
1637 guess_suffix = WORD_MNEM_SUFFIX;
1638
1639 for (op = i.operands; --op >= 0;)
1640 if (i.types[op] & Imm)
1641 {
1642 switch (i.op[op].imms->X_op)
1643 {
1644 case O_constant:
1645 /* If a suffix is given, this operand may be shortened. */
1646 switch (guess_suffix)
1647 {
1648 case LONG_MNEM_SUFFIX:
1649 i.types[op] |= Imm32 | Imm64;
1650 break;
1651 case WORD_MNEM_SUFFIX:
1652 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
1653 break;
1654 case BYTE_MNEM_SUFFIX:
1655 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
1656 break;
1657 }
1658
1659 /* If this operand is at most 16 bits, convert it
1660 to a signed 16 bit number before trying to see
1661 whether it will fit in an even smaller size.
1662 This allows a 16-bit operand such as $0xffe0 to
1663 be recognised as within Imm8S range. */
1664 if ((i.types[op] & Imm16)
1665 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
1666 {
1667 i.op[op].imms->X_add_number =
1668 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
1669 }
1670 if ((i.types[op] & Imm32)
1671 && (i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1)) == 0)
1672 {
1673 i.op[op].imms->X_add_number =
1674 (i.op[op].imms->X_add_number ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1675 }
1676 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
1677 /* We must avoid matching of Imm32 templates when 64bit only immediate is available. */
1678 if (guess_suffix == QWORD_MNEM_SUFFIX)
1679 i.types[op] &= ~Imm32;
1680 break;
1681 case O_absent:
1682 case O_register:
1683 abort ();
1684 /* Symbols and expressions. */
1685 default:
1686 /* Convert symbolic operand to proper sizes for matching. */
1687 switch (guess_suffix)
1688 {
1689 case QWORD_MNEM_SUFFIX:
1690 i.types[op] = Imm64 | Imm32S;
1691 break;
1692 case LONG_MNEM_SUFFIX:
1693 i.types[op] = Imm32 | Imm64;
1694 break;
1695 case WORD_MNEM_SUFFIX:
1696 i.types[op] = Imm16 | Imm32 | Imm64;
1697 break;
1698 break;
1699 case BYTE_MNEM_SUFFIX:
1700 i.types[op] = Imm8 | Imm8S | Imm16 | Imm32S | Imm32;
1701 break;
1702 break;
1703 }
1704 break;
1705 }
1706 }
1707 }
1708
1709 if (i.disp_operands)
1710 {
1711 /* Try to use the smallest displacement type too. */
1712 int op;
1713
1714 for (op = i.operands; --op >= 0;)
1715 if ((i.types[op] & Disp)
1716 && i.op[op].disps->X_op == O_constant)
1717 {
1718 offsetT disp = i.op[op].disps->X_add_number;
1719
1720 if (i.types[op] & Disp16)
1721 {
1722 /* We know this operand is at most 16 bits, so
1723 convert to a signed 16 bit number before trying
1724 to see whether it will fit in an even smaller
1725 size. */
1726
1727 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
1728 }
1729 else if (i.types[op] & Disp32)
1730 {
1731 /* We know this operand is at most 32 bits, so convert to a
1732 signed 32 bit number before trying to see whether it will
1733 fit in an even smaller size. */
1734 disp &= (((offsetT) 2 << 31) - 1);
1735 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1736 }
1737 if (flag_code == CODE_64BIT)
1738 {
1739 if (fits_in_signed_long (disp))
1740 i.types[op] |= Disp32S;
1741 if (fits_in_unsigned_long (disp))
1742 i.types[op] |= Disp32;
1743 }
1744 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
1745 && fits_in_signed_byte (disp))
1746 i.types[op] |= Disp8;
1747 }
1748 }
1749
1750 overlap0 = 0;
1751 overlap1 = 0;
1752 overlap2 = 0;
1753 found_reverse_match = 0;
1754 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
1755 ? No_bSuf
1756 : (i.suffix == WORD_MNEM_SUFFIX
1757 ? No_wSuf
1758 : (i.suffix == SHORT_MNEM_SUFFIX
1759 ? No_sSuf
1760 : (i.suffix == LONG_MNEM_SUFFIX
1761 ? No_lSuf
1762 : (i.suffix == QWORD_MNEM_SUFFIX
1763 ? No_qSuf
1764 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX ? No_xSuf : 0))))));
1765
1766 for (t = current_templates->start;
1767 t < current_templates->end;
1768 t++)
1769 {
1770 /* Must have right number of operands. */
1771 if (i.operands != t->operands)
1772 continue;
1773
1774 /* Check the suffix, except for some instructions in intel mode. */
1775 if ((t->opcode_modifier & suffix_check)
1776 && !(intel_syntax
1777 && (t->opcode_modifier & IgnoreSize))
1778 && !(intel_syntax
1779 && t->base_opcode == 0xd9
1780 && (t->extension_opcode == 5 /* 0xd9,5 "fldcw" */
1781 || t->extension_opcode == 7))) /* 0xd9,7 "f{n}stcw" */
1782 continue;
1783
1784 /* Do not verify operands when there are none. */
1785 else if (!t->operands)
1786 {
1787 if (t->cpu_flags & ~cpu_arch_flags)
1788 continue;
1789 /* We've found a match; break out of loop. */
1790 break;
1791 }
1792
1793 overlap0 = i.types[0] & t->operand_types[0];
1794 switch (t->operands)
1795 {
1796 case 1:
1797 if (!MATCH (overlap0, i.types[0], t->operand_types[0]))
1798 continue;
1799 break;
1800 case 2:
1801 case 3:
1802 overlap1 = i.types[1] & t->operand_types[1];
1803 if (!MATCH (overlap0, i.types[0], t->operand_types[0])
1804 || !MATCH (overlap1, i.types[1], t->operand_types[1])
1805 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
1806 t->operand_types[0],
1807 overlap1, i.types[1],
1808 t->operand_types[1]))
1809 {
1810 /* Check if other direction is valid ... */
1811 if ((t->opcode_modifier & (D|FloatD)) == 0)
1812 continue;
1813
1814 /* Try reversing direction of operands. */
1815 overlap0 = i.types[0] & t->operand_types[1];
1816 overlap1 = i.types[1] & t->operand_types[0];
1817 if (!MATCH (overlap0, i.types[0], t->operand_types[1])
1818 || !MATCH (overlap1, i.types[1], t->operand_types[0])
1819 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
1820 t->operand_types[1],
1821 overlap1, i.types[1],
1822 t->operand_types[0]))
1823 {
1824 /* Does not match either direction. */
1825 continue;
1826 }
1827 /* found_reverse_match holds which of D or FloatDR
1828 we've found. */
1829 found_reverse_match = t->opcode_modifier & (D|FloatDR);
1830 }
1831 /* Found a forward 2 operand match here. */
1832 else if (t->operands == 3)
1833 {
1834 /* Here we make use of the fact that there are no
1835 reverse match 3 operand instructions, and all 3
1836 operand instructions only need to be checked for
1837 register consistency between operands 2 and 3. */
1838 overlap2 = i.types[2] & t->operand_types[2];
1839 if (!MATCH (overlap2, i.types[2], t->operand_types[2])
1840 || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1],
1841 t->operand_types[1],
1842 overlap2, i.types[2],
1843 t->operand_types[2]))
1844
1845 continue;
1846 }
1847 /* Found either forward/reverse 2 or 3 operand match here:
1848 slip through to break. */
1849 }
1850 if (t->cpu_flags & ~cpu_arch_flags)
1851 {
1852 found_reverse_match = 0;
1853 continue;
1854 }
1855 /* We've found a match; break out of loop. */
1856 break;
1857 }
1858 if (t == current_templates->end)
1859 {
1860 /* We found no match. */
1861 as_bad (_("suffix or operands invalid for `%s'"),
1862 current_templates->start->name);
1863 return;
1864 }
1865
1866 if (!quiet_warnings)
1867 {
1868 if (!intel_syntax
1869 && ((i.types[0] & JumpAbsolute)
1870 != (t->operand_types[0] & JumpAbsolute)))
1871 {
1872 as_warn (_("indirect %s without `*'"), t->name);
1873 }
1874
1875 if ((t->opcode_modifier & (IsPrefix|IgnoreSize))
1876 == (IsPrefix|IgnoreSize))
1877 {
1878 /* Warn them that a data or address size prefix doesn't
1879 affect assembly of the next line of code. */
1880 as_warn (_("stand-alone `%s' prefix"), t->name);
1881 }
1882 }
1883
1884 /* Copy the template we found. */
1885 i.tm = *t;
1886 if (found_reverse_match)
1887 {
1888 /* If we found a reverse match we must alter the opcode
1889 direction bit. found_reverse_match holds bits to change
1890 (different for int & float insns). */
1891
1892 i.tm.base_opcode ^= found_reverse_match;
1893
1894 i.tm.operand_types[0] = t->operand_types[1];
1895 i.tm.operand_types[1] = t->operand_types[0];
1896 }
1897
1898 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1899 if (SYSV386_COMPAT
1900 && intel_syntax
1901 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1902 i.tm.base_opcode ^= FloatR;
1903
1904 if (i.tm.opcode_modifier & FWait)
1905 if (! add_prefix (FWAIT_OPCODE))
1906 return;
1907
1908 /* Check string instruction segment overrides. */
1909 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1910 {
1911 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
1912 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
1913 {
1914 if (i.seg[0] != NULL && i.seg[0] != &es)
1915 {
1916 as_bad (_("`%s' operand %d must use `%%es' segment"),
1917 i.tm.name,
1918 mem_op + 1);
1919 return;
1920 }
1921 /* There's only ever one segment override allowed per instruction.
1922 This instruction possibly has a legal segment override on the
1923 second operand, so copy the segment to where non-string
1924 instructions store it, allowing common code. */
1925 i.seg[0] = i.seg[1];
1926 }
1927 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
1928 {
1929 if (i.seg[1] != NULL && i.seg[1] != &es)
1930 {
1931 as_bad (_("`%s' operand %d must use `%%es' segment"),
1932 i.tm.name,
1933 mem_op + 2);
1934 return;
1935 }
1936 }
1937 }
1938
1939 /* If matched instruction specifies an explicit instruction mnemonic
1940 suffix, use it. */
1941 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
1942 {
1943 if (i.tm.opcode_modifier & Size16)
1944 i.suffix = WORD_MNEM_SUFFIX;
1945 else if (i.tm.opcode_modifier & Size64)
1946 i.suffix = QWORD_MNEM_SUFFIX;
1947 else
1948 i.suffix = LONG_MNEM_SUFFIX;
1949 }
1950 else if (i.reg_operands)
1951 {
1952 /* If there's no instruction mnemonic suffix we try to invent one
1953 based on register operands. */
1954 if (!i.suffix)
1955 {
1956 /* We take i.suffix from the last register operand specified,
1957 Destination register type is more significant than source
1958 register type. */
1959 int op;
1960 for (op = i.operands; --op >= 0;)
1961 if ((i.types[op] & Reg)
1962 && !(i.tm.operand_types[op] & InOutPortReg))
1963 {
1964 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
1965 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
1966 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
1967 LONG_MNEM_SUFFIX);
1968 break;
1969 }
1970 }
1971 else if (i.suffix == BYTE_MNEM_SUFFIX)
1972 {
1973 int op;
1974 for (op = i.operands; --op >= 0;)
1975 {
1976 /* If this is an eight bit register, it's OK. If it's
1977 the 16 or 32 bit version of an eight bit register,
1978 we will just use the low portion, and that's OK too. */
1979 if (i.types[op] & Reg8)
1980 continue;
1981
1982 /* movzx and movsx should not generate this warning. */
1983 if (intel_syntax
1984 && (i.tm.base_opcode == 0xfb7
1985 || i.tm.base_opcode == 0xfb6
1986 || i.tm.base_opcode == 0x63
1987 || i.tm.base_opcode == 0xfbe
1988 || i.tm.base_opcode == 0xfbf))
1989 continue;
1990
1991 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4
1992 #if 0
1993 /* Check that the template allows eight bit regs
1994 This kills insns such as `orb $1,%edx', which
1995 maybe should be allowed. */
1996 && (i.tm.operand_types[op] & (Reg8|InOutPortReg))
1997 #endif
1998 )
1999 {
2000 /* Prohibit these changes in the 64bit mode, since
2001 the lowering is more complicated. */
2002 if (flag_code == CODE_64BIT
2003 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2004 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2005 i.op[op].regs->reg_name,
2006 i.suffix);
2007 #if REGISTER_WARNINGS
2008 if (!quiet_warnings
2009 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2010 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2011 (i.op[op].regs
2012 + (i.types[op] & Reg16
2013 ? REGNAM_AL - REGNAM_AX
2014 : REGNAM_AL - REGNAM_EAX))->reg_name,
2015 i.op[op].regs->reg_name,
2016 i.suffix);
2017 #endif
2018 continue;
2019 }
2020 /* Any other register is bad. */
2021 if (i.types[op] & (Reg | RegMMX | RegXMM
2022 | SReg2 | SReg3
2023 | Control | Debug | Test
2024 | FloatReg | FloatAcc))
2025 {
2026 as_bad (_("`%%%s' not allowed with `%s%c'"),
2027 i.op[op].regs->reg_name,
2028 i.tm.name,
2029 i.suffix);
2030 return;
2031 }
2032 }
2033 }
2034 else if (i.suffix == LONG_MNEM_SUFFIX)
2035 {
2036 int op;
2037
2038 for (op = i.operands; --op >= 0;)
2039 /* Reject eight bit registers, except where the template
2040 requires them. (eg. movzb) */
2041 if ((i.types[op] & Reg8) != 0
2042 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2043 {
2044 as_bad (_("`%%%s' not allowed with `%s%c'"),
2045 i.op[op].regs->reg_name,
2046 i.tm.name,
2047 i.suffix);
2048 return;
2049 }
2050 /* Warn if the e prefix on a general reg is missing. */
2051 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2052 && (i.types[op] & Reg16) != 0
2053 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
2054 {
2055 /* Prohibit these changes in the 64bit mode, since
2056 the lowering is more complicated. */
2057 if (flag_code == CODE_64BIT)
2058 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2059 i.op[op].regs->reg_name,
2060 i.suffix);
2061 #if REGISTER_WARNINGS
2062 else
2063 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2064 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
2065 i.op[op].regs->reg_name,
2066 i.suffix);
2067 #endif
2068 }
2069 /* Warn if the r prefix on a general reg is missing. */
2070 else if ((i.types[op] & Reg64) != 0
2071 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
2072 {
2073 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2074 i.op[op].regs->reg_name,
2075 i.suffix);
2076 }
2077 }
2078 else if (i.suffix == QWORD_MNEM_SUFFIX)
2079 {
2080 int op;
2081
2082 for (op = i.operands; --op >= 0; )
2083 /* Reject eight bit registers, except where the template
2084 requires them. (eg. movzb) */
2085 if ((i.types[op] & Reg8) != 0
2086 && (i.tm.operand_types[op] & (Reg16|Reg32|Acc)) != 0)
2087 {
2088 as_bad (_("`%%%s' not allowed with `%s%c'"),
2089 i.op[op].regs->reg_name,
2090 i.tm.name,
2091 i.suffix);
2092 return;
2093 }
2094 /* Warn if the e prefix on a general reg is missing. */
2095 else if (((i.types[op] & Reg16) != 0
2096 || (i.types[op] & Reg32) != 0)
2097 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
2098 {
2099 /* Prohibit these changes in the 64bit mode, since
2100 the lowering is more complicated. */
2101 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2102 i.op[op].regs->reg_name,
2103 i.suffix);
2104 }
2105 }
2106 else if (i.suffix == WORD_MNEM_SUFFIX)
2107 {
2108 int op;
2109 for (op = i.operands; --op >= 0;)
2110 /* Reject eight bit registers, except where the template
2111 requires them. (eg. movzb) */
2112 if ((i.types[op] & Reg8) != 0
2113 && (i.tm.operand_types[op] & (Reg16|Reg32|Acc)) != 0)
2114 {
2115 as_bad (_("`%%%s' not allowed with `%s%c'"),
2116 i.op[op].regs->reg_name,
2117 i.tm.name,
2118 i.suffix);
2119 return;
2120 }
2121 /* Warn if the e prefix on a general reg is present. */
2122 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2123 && (i.types[op] & Reg32) != 0
2124 && (i.tm.operand_types[op] & (Reg16|Acc)) != 0)
2125 {
2126 /* Prohibit these changes in the 64bit mode, since
2127 the lowering is more complicated. */
2128 if (flag_code == CODE_64BIT)
2129 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2130 i.op[op].regs->reg_name,
2131 i.suffix);
2132 else
2133 #if REGISTER_WARNINGS
2134 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2135 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
2136 i.op[op].regs->reg_name,
2137 i.suffix);
2138 #endif
2139 }
2140 }
2141 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2142 /* Do nothing if the instruction is going to ignore the prefix. */
2143 ;
2144 else
2145 abort ();
2146 }
2147 else if ((i.tm.opcode_modifier & DefaultSize) && !i.suffix)
2148 {
2149 i.suffix = stackop_size;
2150 }
2151 /* Make still unresolved immediate matches conform to size of immediate
2152 given in i.suffix. Note: overlap2 cannot be an immediate! */
2153 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S))
2154 && overlap0 != Imm8 && overlap0 != Imm8S
2155 && overlap0 != Imm16 && overlap0 != Imm32S
2156 && overlap0 != Imm32 && overlap0 != Imm64)
2157 {
2158 if (i.suffix)
2159 {
2160 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
2161 (i.suffix == WORD_MNEM_SUFFIX ? Imm16 :
2162 (i.suffix == QWORD_MNEM_SUFFIX ? Imm64 | Imm32S : Imm32)));
2163 }
2164 else if (overlap0 == (Imm16 | Imm32S | Imm32)
2165 || overlap0 == (Imm16 | Imm32)
2166 || overlap0 == (Imm16 | Imm32S))
2167 {
2168 overlap0 =
2169 ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32S;
2170 }
2171 if (overlap0 != Imm8 && overlap0 != Imm8S
2172 && overlap0 != Imm16 && overlap0 != Imm32S
2173 && overlap0 != Imm32 && overlap0 != Imm64)
2174 {
2175 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
2176 return;
2177 }
2178 }
2179 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32))
2180 && overlap1 != Imm8 && overlap1 != Imm8S
2181 && overlap1 != Imm16 && overlap1 != Imm32S
2182 && overlap1 != Imm32 && overlap1 != Imm64)
2183 {
2184 if (i.suffix)
2185 {
2186 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
2187 (i.suffix == WORD_MNEM_SUFFIX ? Imm16 :
2188 (i.suffix == QWORD_MNEM_SUFFIX ? Imm64 | Imm32S : Imm32)));
2189 }
2190 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
2191 || overlap1 == (Imm16 | Imm32)
2192 || overlap1 == (Imm16 | Imm32S))
2193 {
2194 overlap1 =
2195 ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32S;
2196 }
2197 if (overlap1 != Imm8 && overlap1 != Imm8S
2198 && overlap1 != Imm16 && overlap1 != Imm32S
2199 && overlap1 != Imm32 && overlap1 != Imm64)
2200 {
2201 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
2202 return;
2203 }
2204 }
2205 assert ((overlap2 & Imm) == 0);
2206
2207 i.types[0] = overlap0;
2208 if (overlap0 & ImplicitRegister)
2209 i.reg_operands--;
2210 if (overlap0 & Imm1)
2211 i.imm_operands = 0; /* kludge for shift insns. */
2212
2213 i.types[1] = overlap1;
2214 if (overlap1 & ImplicitRegister)
2215 i.reg_operands--;
2216
2217 i.types[2] = overlap2;
2218 if (overlap2 & ImplicitRegister)
2219 i.reg_operands--;
2220
2221 /* Finalize opcode. First, we change the opcode based on the operand
2222 size given by i.suffix: We need not change things for byte insns. */
2223
2224 if (!i.suffix && (i.tm.opcode_modifier & W))
2225 {
2226 as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
2227 return;
2228 }
2229
2230 /* For movzx and movsx, need to check the register type. */
2231 if (intel_syntax
2232 && (i.tm.base_opcode == 0xfb6 || i.tm.base_opcode == 0xfbe))
2233 if (i.suffix && i.suffix == BYTE_MNEM_SUFFIX)
2234 {
2235 unsigned int prefix = DATA_PREFIX_OPCODE;
2236
2237 if ((i.op[1].regs->reg_type & Reg16) != 0)
2238 if (!add_prefix (prefix))
2239 return;
2240 }
2241
2242 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2243 {
2244 /* It's not a byte, select word/dword operation. */
2245 if (i.tm.opcode_modifier & W)
2246 {
2247 if (i.tm.opcode_modifier & ShortForm)
2248 i.tm.base_opcode |= 8;
2249 else
2250 i.tm.base_opcode |= 1;
2251 }
2252 /* Now select between word & dword operations via the operand
2253 size prefix, except for instructions that will ignore this
2254 prefix anyway. */
2255 if (i.suffix != QWORD_MNEM_SUFFIX
2256 && (i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2257 && !(i.tm.opcode_modifier & IgnoreSize))
2258 {
2259 unsigned int prefix = DATA_PREFIX_OPCODE;
2260 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2261 prefix = ADDR_PREFIX_OPCODE;
2262
2263 if (! add_prefix (prefix))
2264 return;
2265 }
2266
2267 if (i.suffix != QWORD_MNEM_SUFFIX && (flag_code == CODE_64BIT)
2268 && !(i.tm.opcode_modifier & IgnoreSize)
2269 && (i.tm.opcode_modifier & JumpByte))
2270 {
2271 if (! add_prefix (ADDR_PREFIX_OPCODE))
2272 return;
2273 }
2274
2275 /* Set mode64 for an operand. */
2276 if (i.suffix == QWORD_MNEM_SUFFIX
2277 && !(i.tm.opcode_modifier & NoRex64))
2278 {
2279 i.rex.mode64 = 1;
2280 if (flag_code < CODE_64BIT)
2281 {
2282 as_bad (_("64bit operations available only in 64bit modes."));
2283 return;
2284 }
2285 }
2286
2287 /* Size floating point instruction. */
2288 if (i.suffix == LONG_MNEM_SUFFIX)
2289 {
2290 if (i.tm.opcode_modifier & FloatMF)
2291 i.tm.base_opcode ^= 4;
2292 }
2293 }
2294
2295 if (i.tm.opcode_modifier & ImmExt)
2296 {
2297 /* These AMD 3DNow! and Intel Katmai New Instructions have an
2298 opcode suffix which is coded in the same place as an 8-bit
2299 immediate field would be. Here we fake an 8-bit immediate
2300 operand from the opcode suffix stored in tm.extension_opcode. */
2301
2302 expressionS *exp;
2303
2304 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
2305
2306 exp = &im_expressions[i.imm_operands++];
2307 i.op[i.operands].imms = exp;
2308 i.types[i.operands++] = Imm8;
2309 exp->X_op = O_constant;
2310 exp->X_add_number = i.tm.extension_opcode;
2311 i.tm.extension_opcode = None;
2312 }
2313
2314 /* For insns with operands there are more diddles to do to the opcode. */
2315 if (i.operands)
2316 {
2317 /* Default segment register this instruction will use
2318 for memory accesses. 0 means unknown.
2319 This is only for optimizing out unnecessary segment overrides. */
2320 const seg_entry *default_seg = 0;
2321
2322 /* The imul $imm, %reg instruction is converted into
2323 imul $imm, %reg, %reg, and the clr %reg instruction
2324 is converted into xor %reg, %reg. */
2325 if (i.tm.opcode_modifier & regKludge)
2326 {
2327 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
2328 /* Pretend we saw the extra register operand. */
2329 assert (i.op[first_reg_op + 1].regs == 0);
2330 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
2331 i.types[first_reg_op + 1] = i.types[first_reg_op];
2332 i.reg_operands = 2;
2333 }
2334
2335 if (i.tm.opcode_modifier & ShortForm)
2336 {
2337 /* The register or float register operand is in operand 0 or 1. */
2338 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
2339 /* Register goes in low 3 bits of opcode. */
2340 i.tm.base_opcode |= i.op[op].regs->reg_num;
2341 if (i.op[op].regs->reg_flags & RegRex)
2342 i.rex.extZ = 1;
2343 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2344 {
2345 /* Warn about some common errors, but press on regardless.
2346 The first case can be generated by gcc (<= 2.8.1). */
2347 if (i.operands == 2)
2348 {
2349 /* Reversed arguments on faddp, fsubp, etc. */
2350 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
2351 i.op[1].regs->reg_name,
2352 i.op[0].regs->reg_name);
2353 }
2354 else
2355 {
2356 /* Extraneous `l' suffix on fp insn. */
2357 as_warn (_("translating to `%s %%%s'"), i.tm.name,
2358 i.op[0].regs->reg_name);
2359 }
2360 }
2361 }
2362 else if (i.tm.opcode_modifier & Modrm)
2363 {
2364 /* The opcode is completed (modulo i.tm.extension_opcode which
2365 must be put into the modrm byte).
2366 Now, we make the modrm & index base bytes based on all the
2367 info we've collected. */
2368
2369 /* i.reg_operands MUST be the number of real register operands;
2370 implicit registers do not count. */
2371 if (i.reg_operands == 2)
2372 {
2373 unsigned int source, dest;
2374 source = ((i.types[0]
2375 & (Reg | RegMMX | RegXMM
2376 | SReg2 | SReg3
2377 | Control | Debug | Test))
2378 ? 0 : 1);
2379 dest = source + 1;
2380
2381 i.rm.mode = 3;
2382 /* One of the register operands will be encoded in the
2383 i.tm.reg field, the other in the combined i.tm.mode
2384 and i.tm.regmem fields. If no form of this
2385 instruction supports a memory destination operand,
2386 then we assume the source operand may sometimes be
2387 a memory operand and so we need to store the
2388 destination in the i.rm.reg field. */
2389 if ((i.tm.operand_types[dest] & AnyMem) == 0)
2390 {
2391 i.rm.reg = i.op[dest].regs->reg_num;
2392 i.rm.regmem = i.op[source].regs->reg_num;
2393 if (i.op[dest].regs->reg_flags & RegRex)
2394 i.rex.extX = 1;
2395 if (i.op[source].regs->reg_flags & RegRex)
2396 i.rex.extZ = 1;
2397 }
2398 else
2399 {
2400 i.rm.reg = i.op[source].regs->reg_num;
2401 i.rm.regmem = i.op[dest].regs->reg_num;
2402 if (i.op[dest].regs->reg_flags & RegRex)
2403 i.rex.extZ = 1;
2404 if (i.op[source].regs->reg_flags & RegRex)
2405 i.rex.extX = 1;
2406 }
2407 }
2408 else
2409 { /* If it's not 2 reg operands... */
2410 if (i.mem_operands)
2411 {
2412 unsigned int fake_zero_displacement = 0;
2413 unsigned int op = ((i.types[0] & AnyMem)
2414 ? 0
2415 : (i.types[1] & AnyMem) ? 1 : 2);
2416
2417 default_seg = &ds;
2418
2419 if (! i.base_reg)
2420 {
2421 i.rm.mode = 0;
2422 if (! i.disp_operands)
2423 fake_zero_displacement = 1;
2424 if (! i.index_reg)
2425 {
2426 /* Operand is just <disp> */
2427 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)
2428 && (flag_code != CODE_64BIT))
2429 {
2430 i.rm.regmem = NO_BASE_REGISTER_16;
2431 i.types[op] &= ~Disp;
2432 i.types[op] |= Disp16;
2433 }
2434 else if (flag_code != CODE_64BIT
2435 || (i.prefix[ADDR_PREFIX] != 0))
2436 {
2437 i.rm.regmem = NO_BASE_REGISTER;
2438 i.types[op] &= ~Disp;
2439 i.types[op] |= Disp32;
2440 }
2441 else
2442 {
2443 /* 64bit mode overwrites the 32bit
2444 absolute addressing by RIP relative
2445 addressing and absolute addressing
2446 is encoded by one of the redundant
2447 SIB forms. */
2448
2449 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2450 i.sib.base = NO_BASE_REGISTER;
2451 i.sib.index = NO_INDEX_REGISTER;
2452 i.types[op] &= ~Disp;
2453 i.types[op] |= Disp32S;
2454 }
2455 }
2456 else /* ! i.base_reg && i.index_reg */
2457 {
2458 i.sib.index = i.index_reg->reg_num;
2459 i.sib.base = NO_BASE_REGISTER;
2460 i.sib.scale = i.log2_scale_factor;
2461 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2462 i.types[op] &= ~Disp;
2463 if (flag_code != CODE_64BIT)
2464 i.types[op] |= Disp32; /* Must be 32 bit */
2465 else
2466 i.types[op] |= Disp32S;
2467 if (i.index_reg->reg_flags & RegRex)
2468 i.rex.extY = 1;
2469 }
2470 }
2471 /* RIP addressing for 64bit mode. */
2472 else if (i.base_reg->reg_type == BaseIndex)
2473 {
2474 i.rm.regmem = NO_BASE_REGISTER;
2475 i.types[op] &= ~Disp;
2476 i.types[op] |= Disp32S;
2477 i.flags[op] = Operand_PCrel;
2478 }
2479 else if (i.base_reg->reg_type & Reg16)
2480 {
2481 switch (i.base_reg->reg_num)
2482 {
2483 case 3: /* (%bx) */
2484 if (! i.index_reg)
2485 i.rm.regmem = 7;
2486 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
2487 i.rm.regmem = i.index_reg->reg_num - 6;
2488 break;
2489 case 5: /* (%bp) */
2490 default_seg = &ss;
2491 if (! i.index_reg)
2492 {
2493 i.rm.regmem = 6;
2494 if ((i.types[op] & Disp) == 0)
2495 {
2496 /* fake (%bp) into 0(%bp) */
2497 i.types[op] |= Disp8;
2498 fake_zero_displacement = 1;
2499 }
2500 }
2501 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
2502 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
2503 break;
2504 default: /* (%si) -> 4 or (%di) -> 5 */
2505 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
2506 }
2507 i.rm.mode = mode_from_disp_size (i.types[op]);
2508 }
2509 else /* i.base_reg and 32/64 bit mode */
2510 {
2511 if (flag_code == CODE_64BIT
2512 && (i.types[op] & Disp))
2513 {
2514 if (i.types[op] & Disp8)
2515 i.types[op] = Disp8 | Disp32S;
2516 else
2517 i.types[op] = Disp32S;
2518 }
2519 i.rm.regmem = i.base_reg->reg_num;
2520 if (i.base_reg->reg_flags & RegRex)
2521 i.rex.extZ = 1;
2522 i.sib.base = i.base_reg->reg_num;
2523 /* x86-64 ignores REX prefix bit here to avoid
2524 decoder complications. */
2525 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
2526 {
2527 default_seg = &ss;
2528 if (i.disp_operands == 0)
2529 {
2530 fake_zero_displacement = 1;
2531 i.types[op] |= Disp8;
2532 }
2533 }
2534 else if (i.base_reg->reg_num == ESP_REG_NUM)
2535 {
2536 default_seg = &ss;
2537 }
2538 i.sib.scale = i.log2_scale_factor;
2539 if (! i.index_reg)
2540 {
2541 /* <disp>(%esp) becomes two byte modrm
2542 with no index register. We've already
2543 stored the code for esp in i.rm.regmem
2544 ie. ESCAPE_TO_TWO_BYTE_ADDRESSING. Any
2545 base register besides %esp will not use
2546 the extra modrm byte. */
2547 i.sib.index = NO_INDEX_REGISTER;
2548 #if ! SCALE1_WHEN_NO_INDEX
2549 /* Another case where we force the second
2550 modrm byte. */
2551 if (i.log2_scale_factor)
2552 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2553 #endif
2554 }
2555 else
2556 {
2557 i.sib.index = i.index_reg->reg_num;
2558 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2559 if (i.index_reg->reg_flags & RegRex)
2560 i.rex.extY = 1;
2561 }
2562 i.rm.mode = mode_from_disp_size (i.types[op]);
2563 }
2564
2565 if (fake_zero_displacement)
2566 {
2567 /* Fakes a zero displacement assuming that i.types[op]
2568 holds the correct displacement size. */
2569 expressionS *exp;
2570
2571 assert (i.op[op].disps == 0);
2572 exp = &disp_expressions[i.disp_operands++];
2573 i.op[op].disps = exp;
2574 exp->X_op = O_constant;
2575 exp->X_add_number = 0;
2576 exp->X_add_symbol = (symbolS *) 0;
2577 exp->X_op_symbol = (symbolS *) 0;
2578 }
2579 }
2580
2581 /* Fill in i.rm.reg or i.rm.regmem field with register
2582 operand (if any) based on i.tm.extension_opcode.
2583 Again, we must be careful to make sure that
2584 segment/control/debug/test/MMX registers are coded
2585 into the i.rm.reg field. */
2586 if (i.reg_operands)
2587 {
2588 unsigned int op =
2589 ((i.types[0]
2590 & (Reg | RegMMX | RegXMM
2591 | SReg2 | SReg3
2592 | Control | Debug | Test))
2593 ? 0
2594 : ((i.types[1]
2595 & (Reg | RegMMX | RegXMM
2596 | SReg2 | SReg3
2597 | Control | Debug | Test))
2598 ? 1
2599 : 2));
2600 /* If there is an extension opcode to put here, the
2601 register number must be put into the regmem field. */
2602 if (i.tm.extension_opcode != None)
2603 {
2604 i.rm.regmem = i.op[op].regs->reg_num;
2605 if (i.op[op].regs->reg_flags & RegRex)
2606 i.rex.extZ = 1;
2607 }
2608 else
2609 {
2610 i.rm.reg = i.op[op].regs->reg_num;
2611 if (i.op[op].regs->reg_flags & RegRex)
2612 i.rex.extX = 1;
2613 }
2614
2615 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2
2616 we must set it to 3 to indicate this is a register
2617 operand in the regmem field. */
2618 if (!i.mem_operands)
2619 i.rm.mode = 3;
2620 }
2621
2622 /* Fill in i.rm.reg field with extension opcode (if any). */
2623 if (i.tm.extension_opcode != None)
2624 i.rm.reg = i.tm.extension_opcode;
2625 }
2626 }
2627 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
2628 {
2629 if (i.tm.base_opcode == POP_SEG_SHORT
2630 && i.op[0].regs->reg_num == 1)
2631 {
2632 as_bad (_("you can't `pop %%cs'"));
2633 return;
2634 }
2635 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
2636 if (i.op[0].regs->reg_flags & RegRex)
2637 i.rex.extZ = 1;
2638 }
2639 else if ((i.tm.base_opcode & ~(D|W)) == MOV_AX_DISP32)
2640 {
2641 default_seg = &ds;
2642 }
2643 else if ((i.tm.opcode_modifier & IsString) != 0)
2644 {
2645 /* For the string instructions that allow a segment override
2646 on one of their operands, the default segment is ds. */
2647 default_seg = &ds;
2648 }
2649
2650 /* If a segment was explicitly specified,
2651 and the specified segment is not the default,
2652 use an opcode prefix to select it.
2653 If we never figured out what the default segment is,
2654 then default_seg will be zero at this point,
2655 and the specified segment prefix will always be used. */
2656 if ((i.seg[0]) && (i.seg[0] != default_seg))
2657 {
2658 if (! add_prefix (i.seg[0]->seg_prefix))
2659 return;
2660 }
2661 }
2662 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2663 {
2664 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
2665 as_warn (_("translating to `%sp'"), i.tm.name);
2666 }
2667 }
2668
2669 /* Handle conversion of 'int $3' --> special int3 insn. */
2670 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
2671 {
2672 i.tm.base_opcode = INT3_OPCODE;
2673 i.imm_operands = 0;
2674 }
2675
2676 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
2677 && i.op[0].disps->X_op == O_constant)
2678 {
2679 /* Convert "jmp constant" (and "call constant") to a jump (call) to
2680 the absolute address given by the constant. Since ix86 jumps and
2681 calls are pc relative, we need to generate a reloc. */
2682 i.op[0].disps->X_add_symbol = &abs_symbol;
2683 i.op[0].disps->X_op = O_symbol;
2684 }
2685
2686 if (i.tm.opcode_modifier & Rex64)
2687 i.rex.mode64 = 1;
2688
2689 /* For 8bit registers we would need an empty rex prefix.
2690 Also in the case instruction is already having prefix,
2691 we need to convert old registers to new ones. */
2692
2693 if (((i.types[0] & Reg8) && (i.op[0].regs->reg_flags & RegRex64))
2694 || ((i.types[1] & Reg8) && (i.op[1].regs->reg_flags & RegRex64))
2695 || ((i.rex.mode64 || i.rex.extX || i.rex.extY || i.rex.extZ || i.rex.empty)
2696 && ((i.types[0] & Reg8) || (i.types[1] & Reg8))))
2697 {
2698 int x;
2699 i.rex.empty = 1;
2700 for (x = 0; x < 2; x++)
2701 {
2702 /* Look for 8bit operand that does use old registers. */
2703 if (i.types[x] & Reg8
2704 && !(i.op[x].regs->reg_flags & RegRex64))
2705 {
2706 /* In case it is "hi" register, give up. */
2707 if (i.op[x].regs->reg_num > 3)
2708 as_bad (_("Can't encode registers '%%%s' in the instruction requiring REX prefix.\n"),
2709 i.op[x].regs->reg_name);
2710
2711 /* Otherwise it is equivalent to the extended register.
2712 Since the encoding don't change this is merely cosmetical
2713 cleanup for debug output. */
2714
2715 i.op[x].regs = i.op[x].regs + 8;
2716 }
2717 }
2718 }
2719
2720 if (i.rex.mode64 || i.rex.extX || i.rex.extY || i.rex.extZ || i.rex.empty)
2721 add_prefix (0x40
2722 | (i.rex.mode64 ? 8 : 0)
2723 | (i.rex.extX ? 4 : 0)
2724 | (i.rex.extY ? 2 : 0)
2725 | (i.rex.extZ ? 1 : 0));
2726
2727 /* We are ready to output the insn. */
2728 {
2729 register char *p;
2730
2731 /* Tie dwarf2 debug info to the address at the start of the insn.
2732 We can't do this after the insn has been output as the current
2733 frag may have been closed off. eg. by frag_var. */
2734 dwarf2_emit_insn (0);
2735
2736 /* Output jumps. */
2737 if (i.tm.opcode_modifier & Jump)
2738 {
2739 int code16;
2740 int prefix;
2741 relax_substateT subtype;
2742 symbolS *sym;
2743 offsetT off;
2744
2745 code16 = 0;
2746 if (flag_code == CODE_16BIT)
2747 code16 = CODE16;
2748
2749 prefix = 0;
2750 if (i.prefix[DATA_PREFIX])
2751 {
2752 prefix = 1;
2753 i.prefixes -= 1;
2754 code16 ^= CODE16;
2755 }
2756 /* Pentium4 branch hints. */
2757 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
2758 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
2759 {
2760 prefix++;
2761 i.prefixes--;
2762 }
2763 if (i.prefix[REX_PREFIX])
2764 {
2765 prefix++;
2766 i.prefixes--;
2767 }
2768
2769 if (i.prefixes != 0 && !intel_syntax)
2770 as_warn (_("skipping prefixes on this instruction"));
2771
2772 /* It's always a symbol; End frag & setup for relax.
2773 Make sure there is enough room in this frag for the largest
2774 instruction we may generate in md_convert_frag. This is 2
2775 bytes for the opcode and room for the prefix and largest
2776 displacement. */
2777 frag_grow (prefix + 2 + 4);
2778 /* Prefix and 1 opcode byte go in fr_fix. */
2779 p = frag_more (prefix + 1);
2780 if (i.prefix[DATA_PREFIX])
2781 *p++ = DATA_PREFIX_OPCODE;
2782 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
2783 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
2784 *p++ = i.prefix[SEG_PREFIX];
2785 if (i.prefix[REX_PREFIX])
2786 *p++ = i.prefix[REX_PREFIX];
2787 *p = i.tm.base_opcode;
2788
2789 if ((unsigned char) *p == JUMP_PC_RELATIVE)
2790 subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
2791 else if ((cpu_arch_flags & Cpu386) != 0)
2792 subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
2793 else
2794 subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
2795 subtype |= code16;
2796
2797 sym = i.op[0].disps->X_add_symbol;
2798 off = i.op[0].disps->X_add_number;
2799
2800 if (i.op[0].disps->X_op != O_constant
2801 && i.op[0].disps->X_op != O_symbol)
2802 {
2803 /* Handle complex expressions. */
2804 sym = make_expr_symbol (i.op[0].disps);
2805 off = 0;
2806 }
2807
2808 /* 1 possible extra opcode + 4 byte displacement go in var part.
2809 Pass reloc in fr_var. */
2810 frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
2811 }
2812 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
2813 {
2814 int size;
2815
2816 if (i.tm.opcode_modifier & JumpByte)
2817 {
2818 /* This is a loop or jecxz type instruction. */
2819 size = 1;
2820 if (i.prefix[ADDR_PREFIX])
2821 {
2822 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
2823 i.prefixes -= 1;
2824 }
2825 /* Pentium4 branch hints. */
2826 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
2827 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
2828 {
2829 FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
2830 i.prefixes--;
2831 }
2832 }
2833 else
2834 {
2835 int code16;
2836
2837 code16 = 0;
2838 if (flag_code == CODE_16BIT)
2839 code16 = CODE16;
2840
2841 if (i.prefix[DATA_PREFIX])
2842 {
2843 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
2844 i.prefixes -= 1;
2845 code16 ^= CODE16;
2846 }
2847
2848 size = 4;
2849 if (code16)
2850 size = 2;
2851 }
2852
2853 if (i.prefix[REX_PREFIX])
2854 {
2855 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
2856 i.prefixes -= 1;
2857 }
2858
2859 if (i.prefixes != 0 && !intel_syntax)
2860 as_warn (_("skipping prefixes on this instruction"));
2861
2862 p = frag_more (1 + size);
2863 *p++ = i.tm.base_opcode;
2864
2865 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2866 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
2867 }
2868 else if (i.tm.opcode_modifier & JumpInterSegment)
2869 {
2870 int size;
2871 int prefix;
2872 int code16;
2873
2874 code16 = 0;
2875 if (flag_code == CODE_16BIT)
2876 code16 = CODE16;
2877
2878 prefix = 0;
2879 if (i.prefix[DATA_PREFIX])
2880 {
2881 prefix = 1;
2882 i.prefixes -= 1;
2883 code16 ^= CODE16;
2884 }
2885 if (i.prefix[REX_PREFIX])
2886 {
2887 prefix++;
2888 i.prefixes -= 1;
2889 }
2890
2891 size = 4;
2892 if (code16)
2893 size = 2;
2894
2895 if (i.prefixes != 0 && !intel_syntax)
2896 as_warn (_("skipping prefixes on this instruction"));
2897
2898 /* 1 opcode; 2 segment; offset */
2899 p = frag_more (prefix + 1 + 2 + size);
2900
2901 if (i.prefix[DATA_PREFIX])
2902 *p++ = DATA_PREFIX_OPCODE;
2903
2904 if (i.prefix[REX_PREFIX])
2905 *p++ = i.prefix[REX_PREFIX];
2906
2907 *p++ = i.tm.base_opcode;
2908 if (i.op[1].imms->X_op == O_constant)
2909 {
2910 offsetT n = i.op[1].imms->X_add_number;
2911
2912 if (size == 2
2913 && !fits_in_unsigned_word (n)
2914 && !fits_in_signed_word (n))
2915 {
2916 as_bad (_("16-bit jump out of range"));
2917 return;
2918 }
2919 md_number_to_chars (p, n, size);
2920 }
2921 else
2922 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2923 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
2924 if (i.op[0].imms->X_op != O_constant)
2925 as_bad (_("can't handle non absolute segment in `%s'"),
2926 i.tm.name);
2927 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
2928 }
2929 else
2930 {
2931 /* Output normal instructions here. */
2932 unsigned char *q;
2933
2934 /* All opcodes on i386 have eighter 1 or 2 bytes. We may use third
2935 byte for the SSE instructions to specify prefix they require. */
2936 if (i.tm.base_opcode & 0xff0000)
2937 add_prefix ((i.tm.base_opcode >> 16) & 0xff);
2938
2939 /* The prefix bytes. */
2940 for (q = i.prefix;
2941 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
2942 q++)
2943 {
2944 if (*q)
2945 {
2946 p = frag_more (1);
2947 md_number_to_chars (p, (valueT) *q, 1);
2948 }
2949 }
2950
2951 /* Now the opcode; be careful about word order here! */
2952 if (fits_in_unsigned_byte (i.tm.base_opcode))
2953 {
2954 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
2955 }
2956 else
2957 {
2958 p = frag_more (2);
2959 /* Put out high byte first: can't use md_number_to_chars! */
2960 *p++ = (i.tm.base_opcode >> 8) & 0xff;
2961 *p = i.tm.base_opcode & 0xff;
2962 }
2963
2964 /* Now the modrm byte and sib byte (if present). */
2965 if (i.tm.opcode_modifier & Modrm)
2966 {
2967 p = frag_more (1);
2968 md_number_to_chars (p,
2969 (valueT) (i.rm.regmem << 0
2970 | i.rm.reg << 3
2971 | i.rm.mode << 6),
2972 1);
2973 /* If i.rm.regmem == ESP (4)
2974 && i.rm.mode != (Register mode)
2975 && not 16 bit
2976 ==> need second modrm byte. */
2977 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
2978 && i.rm.mode != 3
2979 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
2980 {
2981 p = frag_more (1);
2982 md_number_to_chars (p,
2983 (valueT) (i.sib.base << 0
2984 | i.sib.index << 3
2985 | i.sib.scale << 6),
2986 1);
2987 }
2988 }
2989
2990 if (i.disp_operands)
2991 {
2992 register unsigned int n;
2993
2994 for (n = 0; n < i.operands; n++)
2995 {
2996 if (i.types[n] & Disp)
2997 {
2998 if (i.op[n].disps->X_op == O_constant)
2999 {
3000 int size;
3001 offsetT val;
3002
3003 size = 4;
3004 if (i.types[n] & (Disp8 | Disp16 | Disp64))
3005 {
3006 size = 2;
3007 if (i.types[n] & Disp8)
3008 size = 1;
3009 if (i.types[n] & Disp64)
3010 size = 8;
3011 }
3012 val = offset_in_range (i.op[n].disps->X_add_number,
3013 size);
3014 p = frag_more (size);
3015 md_number_to_chars (p, val, size);
3016 }
3017 else
3018 {
3019 int size = 4;
3020 int sign = 0;
3021 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
3022
3023 /* The PC relative address is computed relative
3024 to the instruction boundary, so in case immediate
3025 fields follows, we need to adjust the value. */
3026 if (pcrel && i.imm_operands)
3027 {
3028 int imm_size = 4;
3029 register unsigned int n1;
3030
3031 for (n1 = 0; n1 < i.operands; n1++)
3032 if (i.types[n1] & Imm)
3033 {
3034 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
3035 {
3036 imm_size = 2;
3037 if (i.types[n1] & (Imm8 | Imm8S))
3038 imm_size = 1;
3039 if (i.types[n1] & Imm64)
3040 imm_size = 8;
3041 }
3042 break;
3043 }
3044 /* We should find the immediate. */
3045 if (n1 == i.operands)
3046 abort ();
3047 i.op[n].disps->X_add_number -= imm_size;
3048 }
3049
3050 if (i.types[n] & Disp32S)
3051 sign = 1;
3052
3053 if (i.types[n] & (Disp16 | Disp64))
3054 {
3055 size = 2;
3056 if (i.types[n] & Disp64)
3057 size = 8;
3058 }
3059
3060 p = frag_more (size);
3061 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3062 i.op[n].disps, pcrel,
3063 reloc (size, pcrel, sign, i.reloc[n]));
3064 }
3065 }
3066 }
3067 }
3068
3069 /* Output immediate. */
3070 if (i.imm_operands)
3071 {
3072 register unsigned int n;
3073
3074 for (n = 0; n < i.operands; n++)
3075 {
3076 if (i.types[n] & Imm)
3077 {
3078 if (i.op[n].imms->X_op == O_constant)
3079 {
3080 int size;
3081 offsetT val;
3082
3083 size = 4;
3084 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3085 {
3086 size = 2;
3087 if (i.types[n] & (Imm8 | Imm8S))
3088 size = 1;
3089 else if (i.types[n] & Imm64)
3090 size = 8;
3091 }
3092 val = offset_in_range (i.op[n].imms->X_add_number,
3093 size);
3094 p = frag_more (size);
3095 md_number_to_chars (p, val, size);
3096 }
3097 else
3098 {
3099 /* Not absolute_section.
3100 Need a 32-bit fixup (don't support 8bit
3101 non-absolute imms). Try to support other
3102 sizes ... */
3103 RELOC_ENUM reloc_type;
3104 int size = 4;
3105 int sign = 0;
3106
3107 if ((i.types[n] & (Imm32S))
3108 && i.suffix == QWORD_MNEM_SUFFIX)
3109 sign = 1;
3110 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3111 {
3112 size = 2;
3113 if (i.types[n] & (Imm8 | Imm8S))
3114 size = 1;
3115 if (i.types[n] & Imm64)
3116 size = 8;
3117 }
3118
3119 p = frag_more (size);
3120 reloc_type = reloc (size, 0, sign, i.reloc[n]);
3121 #ifdef BFD_ASSEMBLER
3122 if (reloc_type == BFD_RELOC_32
3123 && GOT_symbol
3124 && GOT_symbol == i.op[n].imms->X_add_symbol
3125 && (i.op[n].imms->X_op == O_symbol
3126 || (i.op[n].imms->X_op == O_add
3127 && ((symbol_get_value_expression
3128 (i.op[n].imms->X_op_symbol)->X_op)
3129 == O_subtract))))
3130 {
3131 /* We don't support dynamic linking on x86-64 yet. */
3132 if (flag_code == CODE_64BIT)
3133 abort ();
3134 reloc_type = BFD_RELOC_386_GOTPC;
3135 i.op[n].imms->X_add_number += 3;
3136 }
3137 #endif
3138 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3139 i.op[n].imms, 0, reloc_type);
3140 }
3141 }
3142 }
3143 }
3144 }
3145
3146 #ifdef DEBUG386
3147 if (flag_debug)
3148 {
3149 pi (line, &i);
3150 }
3151 #endif /* DEBUG386 */
3152 }
3153 }
3154 \f
3155 #ifndef LEX_AT
3156 static char *lex_got PARAMS ((RELOC_ENUM *, int *));
3157
3158 /* Parse operands of the form
3159 <symbol>@GOTOFF+<nnn>
3160 and similar .plt or .got references.
3161
3162 If we find one, set up the correct relocation in RELOC and copy the
3163 input string, minus the `@GOTOFF' into a malloc'd buffer for
3164 parsing by the calling routine. Return this buffer, and if ADJUST
3165 is non-null set it to the length of the string we removed from the
3166 input line. Otherwise return NULL. */
3167 static char *
3168 lex_got (reloc, adjust)
3169 RELOC_ENUM *reloc;
3170 int *adjust;
3171 {
3172 static const char * const mode_name[NUM_FLAG_CODE] = { "32", "16", "64" };
3173 static const struct {
3174 const char *str;
3175 const RELOC_ENUM rel[NUM_FLAG_CODE];
3176 } gotrel[] = {
3177 { "PLT", { BFD_RELOC_386_PLT32, 0, BFD_RELOC_X86_64_PLT32 } },
3178 { "GOTOFF", { BFD_RELOC_386_GOTOFF, 0, 0 } },
3179 { "GOTPCREL", { 0, 0, BFD_RELOC_X86_64_GOTPCREL } },
3180 { "GOT", { BFD_RELOC_386_GOT32, 0, BFD_RELOC_X86_64_GOT32 } }
3181 };
3182 char *cp;
3183 unsigned int j;
3184
3185 for (cp = input_line_pointer; *cp != '@'; cp++)
3186 if (is_end_of_line[(unsigned char) *cp])
3187 return NULL;
3188
3189 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
3190 {
3191 int len;
3192
3193 len = strlen (gotrel[j].str);
3194 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
3195 {
3196 if (gotrel[j].rel[(unsigned int) flag_code] != 0)
3197 {
3198 int first, second;
3199 char *tmpbuf, *past_reloc;
3200
3201 *reloc = gotrel[j].rel[(unsigned int) flag_code];
3202 if (adjust)
3203 *adjust = len;
3204
3205 if (GOT_symbol == NULL)
3206 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3207
3208 /* Replace the relocation token with ' ', so that
3209 errors like foo@GOTOFF1 will be detected. */
3210
3211 /* The length of the first part of our input line. */
3212 first = cp - input_line_pointer;
3213
3214 /* The second part goes from after the reloc token until
3215 (and including) an end_of_line char. Don't use strlen
3216 here as the end_of_line char may not be a NUL. */
3217 past_reloc = cp + 1 + len;
3218 for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
3219 ;
3220 second = cp - past_reloc;
3221
3222 /* Allocate and copy string. The trailing NUL shouldn't
3223 be necessary, but be safe. */
3224 tmpbuf = xmalloc (first + second + 2);
3225 memcpy (tmpbuf, input_line_pointer, first);
3226 tmpbuf[first] = ' ';
3227 memcpy (tmpbuf + first + 1, past_reloc, second);
3228 tmpbuf[first + second + 1] = '\0';
3229 return tmpbuf;
3230 }
3231
3232 as_bad (_("@%s reloc is not supported in %s bit mode"),
3233 gotrel[j].str, mode_name[(unsigned int) flag_code]);
3234 return NULL;
3235 }
3236 }
3237
3238 /* Might be a symbol version string. Don't as_bad here. */
3239 return NULL;
3240 }
3241
3242 /* x86_cons_fix_new is called via the expression parsing code when a
3243 reloc is needed. We use this hook to get the correct .got reloc. */
3244 static RELOC_ENUM got_reloc = NO_RELOC;
3245
3246 void
3247 x86_cons_fix_new (frag, off, len, exp)
3248 fragS *frag;
3249 unsigned int off;
3250 unsigned int len;
3251 expressionS *exp;
3252 {
3253 RELOC_ENUM r = reloc (len, 0, 0, got_reloc);
3254 got_reloc = NO_RELOC;
3255 fix_new_exp (frag, off, len, exp, 0, r);
3256 }
3257
3258 void
3259 x86_cons (exp, size)
3260 expressionS *exp;
3261 int size;
3262 {
3263 if (size == 4)
3264 {
3265 /* Handle @GOTOFF and the like in an expression. */
3266 char *save;
3267 char *gotfree_input_line;
3268 int adjust;
3269
3270 save = input_line_pointer;
3271 gotfree_input_line = lex_got (&got_reloc, &adjust);
3272 if (gotfree_input_line)
3273 input_line_pointer = gotfree_input_line;
3274
3275 expression (exp);
3276
3277 if (gotfree_input_line)
3278 {
3279 /* expression () has merrily parsed up to the end of line,
3280 or a comma - in the wrong buffer. Transfer how far
3281 input_line_pointer has moved to the right buffer. */
3282 input_line_pointer = (save
3283 + (input_line_pointer - gotfree_input_line)
3284 + adjust);
3285 free (gotfree_input_line);
3286 }
3287 }
3288 else
3289 expression (exp);
3290 }
3291 #endif
3292
3293 static int i386_immediate PARAMS ((char *));
3294
3295 static int
3296 i386_immediate (imm_start)
3297 char *imm_start;
3298 {
3299 char *save_input_line_pointer;
3300 #ifndef LEX_AT
3301 char *gotfree_input_line;
3302 #endif
3303 segT exp_seg = 0;
3304 expressionS *exp;
3305
3306 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
3307 {
3308 as_bad (_("only 1 or 2 immediate operands are allowed"));
3309 return 0;
3310 }
3311
3312 exp = &im_expressions[i.imm_operands++];
3313 i.op[this_operand].imms = exp;
3314
3315 if (is_space_char (*imm_start))
3316 ++imm_start;
3317
3318 save_input_line_pointer = input_line_pointer;
3319 input_line_pointer = imm_start;
3320
3321 #ifndef LEX_AT
3322 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3323 if (gotfree_input_line)
3324 input_line_pointer = gotfree_input_line;
3325 #endif
3326
3327 exp_seg = expression (exp);
3328
3329 SKIP_WHITESPACE ();
3330 if (*input_line_pointer)
3331 as_bad (_("junk `%s' after expression"), input_line_pointer);
3332
3333 input_line_pointer = save_input_line_pointer;
3334 #ifndef LEX_AT
3335 if (gotfree_input_line)
3336 free (gotfree_input_line);
3337 #endif
3338
3339 if (exp->X_op == O_absent || exp->X_op == O_big)
3340 {
3341 /* Missing or bad expr becomes absolute 0. */
3342 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
3343 imm_start);
3344 exp->X_op = O_constant;
3345 exp->X_add_number = 0;
3346 exp->X_add_symbol = (symbolS *) 0;
3347 exp->X_op_symbol = (symbolS *) 0;
3348 }
3349 else if (exp->X_op == O_constant)
3350 {
3351 /* Size it properly later. */
3352 i.types[this_operand] |= Imm64;
3353 /* If BFD64, sign extend val. */
3354 if (!use_rela_relocations)
3355 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
3356 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
3357 }
3358 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3359 else if (1
3360 #ifdef BFD_ASSEMBLER
3361 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3362 #endif
3363 && exp_seg != text_section
3364 && exp_seg != data_section
3365 && exp_seg != bss_section
3366 && exp_seg != undefined_section
3367 #ifdef BFD_ASSEMBLER
3368 && !bfd_is_com_section (exp_seg)
3369 #endif
3370 )
3371 {
3372 #ifdef BFD_ASSEMBLER
3373 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3374 #else
3375 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3376 #endif
3377 return 0;
3378 }
3379 #endif
3380 else
3381 {
3382 /* This is an address. The size of the address will be
3383 determined later, depending on destination register,
3384 suffix, or the default for the section. */
3385 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
3386 }
3387
3388 return 1;
3389 }
3390
3391 static char *i386_scale PARAMS ((char *));
3392
3393 static char *
3394 i386_scale (scale)
3395 char *scale;
3396 {
3397 offsetT val;
3398 char *save = input_line_pointer;
3399
3400 input_line_pointer = scale;
3401 val = get_absolute_expression ();
3402
3403 switch (val)
3404 {
3405 case 0:
3406 case 1:
3407 i.log2_scale_factor = 0;
3408 break;
3409 case 2:
3410 i.log2_scale_factor = 1;
3411 break;
3412 case 4:
3413 i.log2_scale_factor = 2;
3414 break;
3415 case 8:
3416 i.log2_scale_factor = 3;
3417 break;
3418 default:
3419 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
3420 scale);
3421 input_line_pointer = save;
3422 return NULL;
3423 }
3424 if (i.log2_scale_factor != 0 && ! i.index_reg)
3425 {
3426 as_warn (_("scale factor of %d without an index register"),
3427 1 << i.log2_scale_factor);
3428 #if SCALE1_WHEN_NO_INDEX
3429 i.log2_scale_factor = 0;
3430 #endif
3431 }
3432 scale = input_line_pointer;
3433 input_line_pointer = save;
3434 return scale;
3435 }
3436
3437 static int i386_displacement PARAMS ((char *, char *));
3438
3439 static int
3440 i386_displacement (disp_start, disp_end)
3441 char *disp_start;
3442 char *disp_end;
3443 {
3444 register expressionS *exp;
3445 segT exp_seg = 0;
3446 char *save_input_line_pointer;
3447 #ifndef LEX_AT
3448 char *gotfree_input_line;
3449 #endif
3450 int bigdisp = Disp32;
3451
3452 if (flag_code == CODE_64BIT)
3453 {
3454 if (!i.prefix[ADDR_PREFIX])
3455 bigdisp = Disp64;
3456 }
3457 else if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3458 bigdisp = Disp16;
3459 i.types[this_operand] |= bigdisp;
3460
3461 exp = &disp_expressions[i.disp_operands];
3462 i.op[this_operand].disps = exp;
3463 i.disp_operands++;
3464 save_input_line_pointer = input_line_pointer;
3465 input_line_pointer = disp_start;
3466 END_STRING_AND_SAVE (disp_end);
3467
3468 #ifndef GCC_ASM_O_HACK
3469 #define GCC_ASM_O_HACK 0
3470 #endif
3471 #if GCC_ASM_O_HACK
3472 END_STRING_AND_SAVE (disp_end + 1);
3473 if ((i.types[this_operand] & BaseIndex) != 0
3474 && displacement_string_end[-1] == '+')
3475 {
3476 /* This hack is to avoid a warning when using the "o"
3477 constraint within gcc asm statements.
3478 For instance:
3479
3480 #define _set_tssldt_desc(n,addr,limit,type) \
3481 __asm__ __volatile__ ( \
3482 "movw %w2,%0\n\t" \
3483 "movw %w1,2+%0\n\t" \
3484 "rorl $16,%1\n\t" \
3485 "movb %b1,4+%0\n\t" \
3486 "movb %4,5+%0\n\t" \
3487 "movb $0,6+%0\n\t" \
3488 "movb %h1,7+%0\n\t" \
3489 "rorl $16,%1" \
3490 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
3491
3492 This works great except that the output assembler ends
3493 up looking a bit weird if it turns out that there is
3494 no offset. You end up producing code that looks like:
3495
3496 #APP
3497 movw $235,(%eax)
3498 movw %dx,2+(%eax)
3499 rorl $16,%edx
3500 movb %dl,4+(%eax)
3501 movb $137,5+(%eax)
3502 movb $0,6+(%eax)
3503 movb %dh,7+(%eax)
3504 rorl $16,%edx
3505 #NO_APP
3506
3507 So here we provide the missing zero. */
3508
3509 *displacement_string_end = '0';
3510 }
3511 #endif
3512 #ifndef LEX_AT
3513 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3514 if (gotfree_input_line)
3515 input_line_pointer = gotfree_input_line;
3516 #endif
3517
3518 exp_seg = expression (exp);
3519
3520 SKIP_WHITESPACE ();
3521 if (*input_line_pointer)
3522 as_bad (_("junk `%s' after expression"), input_line_pointer);
3523 #if GCC_ASM_O_HACK
3524 RESTORE_END_STRING (disp_end + 1);
3525 #endif
3526 RESTORE_END_STRING (disp_end);
3527 input_line_pointer = save_input_line_pointer;
3528 #ifndef LEX_AT
3529 if (gotfree_input_line)
3530 free (gotfree_input_line);
3531 #endif
3532
3533 #ifdef BFD_ASSEMBLER
3534 /* We do this to make sure that the section symbol is in
3535 the symbol table. We will ultimately change the relocation
3536 to be relative to the beginning of the section. */
3537 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
3538 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3539 {
3540 if (exp->X_op != O_symbol)
3541 {
3542 as_bad (_("bad expression used with @%s"),
3543 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
3544 ? "GOTPCREL"
3545 : "GOTOFF"));
3546 return 0;
3547 }
3548
3549 if (S_IS_LOCAL (exp->X_add_symbol)
3550 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
3551 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
3552 exp->X_op = O_subtract;
3553 exp->X_op_symbol = GOT_symbol;
3554 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3555 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
3556 else
3557 i.reloc[this_operand] = BFD_RELOC_32;
3558 }
3559 #endif
3560
3561 if (exp->X_op == O_absent || exp->X_op == O_big)
3562 {
3563 /* Missing or bad expr becomes absolute 0. */
3564 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
3565 disp_start);
3566 exp->X_op = O_constant;
3567 exp->X_add_number = 0;
3568 exp->X_add_symbol = (symbolS *) 0;
3569 exp->X_op_symbol = (symbolS *) 0;
3570 }
3571
3572 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3573 if (exp->X_op != O_constant
3574 #ifdef BFD_ASSEMBLER
3575 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3576 #endif
3577 && exp_seg != text_section
3578 && exp_seg != data_section
3579 && exp_seg != bss_section
3580 && exp_seg != undefined_section)
3581 {
3582 #ifdef BFD_ASSEMBLER
3583 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3584 #else
3585 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3586 #endif
3587 return 0;
3588 }
3589 #endif
3590 else if (flag_code == CODE_64BIT)
3591 i.types[this_operand] |= Disp32S | Disp32;
3592 return 1;
3593 }
3594
3595 static int i386_index_check PARAMS ((const char *));
3596
3597 /* Make sure the memory operand we've been dealt is valid.
3598 Return 1 on success, 0 on a failure. */
3599
3600 static int
3601 i386_index_check (operand_string)
3602 const char *operand_string;
3603 {
3604 int ok;
3605 #if INFER_ADDR_PREFIX
3606 int fudged = 0;
3607
3608 tryprefix:
3609 #endif
3610 ok = 1;
3611 if (flag_code == CODE_64BIT)
3612 {
3613 if (i.prefix[ADDR_PREFIX] == 0)
3614 {
3615 /* 64bit checks. */
3616 if ((i.base_reg
3617 && ((i.base_reg->reg_type & Reg64) == 0)
3618 && (i.base_reg->reg_type != BaseIndex
3619 || i.index_reg))
3620 || (i.index_reg
3621 && ((i.index_reg->reg_type & (Reg64|BaseIndex))
3622 != (Reg64|BaseIndex))))
3623 ok = 0;
3624 }
3625 else
3626 {
3627 /* 32bit checks. */
3628 if ((i.base_reg
3629 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
3630 || (i.index_reg
3631 && ((i.index_reg->reg_type & (Reg32|BaseIndex|RegRex))
3632 != (Reg32|BaseIndex))))
3633 ok = 0;
3634 }
3635 }
3636 else
3637 {
3638 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3639 {
3640 /* 16bit checks. */
3641 if ((i.base_reg
3642 && ((i.base_reg->reg_type & (Reg16|BaseIndex|RegRex))
3643 != (Reg16|BaseIndex)))
3644 || (i.index_reg
3645 && (((i.index_reg->reg_type & (Reg16|BaseIndex))
3646 != (Reg16|BaseIndex))
3647 || ! (i.base_reg
3648 && i.base_reg->reg_num < 6
3649 && i.index_reg->reg_num >= 6
3650 && i.log2_scale_factor == 0))))
3651 ok = 0;
3652 }
3653 else
3654 {
3655 /* 32bit checks. */
3656 if ((i.base_reg
3657 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
3658 || (i.index_reg
3659 && ((i.index_reg->reg_type & (Reg32|BaseIndex|RegRex))
3660 != (Reg32|BaseIndex))))
3661 ok = 0;
3662 }
3663 }
3664 if (!ok)
3665 {
3666 #if INFER_ADDR_PREFIX
3667 if (flag_code != CODE_64BIT
3668 && i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0')
3669 {
3670 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
3671 i.prefixes += 1;
3672 /* Change the size of any displacement too. At most one of
3673 Disp16 or Disp32 is set.
3674 FIXME. There doesn't seem to be any real need for separate
3675 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
3676 Removing them would probably clean up the code quite a lot. */
3677 if (i.types[this_operand] & (Disp16|Disp32))
3678 i.types[this_operand] ^= (Disp16|Disp32);
3679 fudged = 1;
3680 goto tryprefix;
3681 }
3682 if (fudged)
3683 as_bad (_("`%s' is not a valid base/index expression"),
3684 operand_string);
3685 else
3686 #endif
3687 as_bad (_("`%s' is not a valid %s bit base/index expression"),
3688 operand_string,
3689 flag_code_names[flag_code]);
3690 return 0;
3691 }
3692 return 1;
3693 }
3694
3695 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
3696 on error. */
3697
3698 static int
3699 i386_operand (operand_string)
3700 char *operand_string;
3701 {
3702 const reg_entry *r;
3703 char *end_op;
3704 char *op_string = operand_string;
3705
3706 if (is_space_char (*op_string))
3707 ++op_string;
3708
3709 /* We check for an absolute prefix (differentiating,
3710 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
3711 if (*op_string == ABSOLUTE_PREFIX)
3712 {
3713 ++op_string;
3714 if (is_space_char (*op_string))
3715 ++op_string;
3716 i.types[this_operand] |= JumpAbsolute;
3717 }
3718
3719 /* Check if operand is a register. */
3720 if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
3721 && (r = parse_register (op_string, &end_op)) != NULL)
3722 {
3723 /* Check for a segment override by searching for ':' after a
3724 segment register. */
3725 op_string = end_op;
3726 if (is_space_char (*op_string))
3727 ++op_string;
3728 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
3729 {
3730 switch (r->reg_num)
3731 {
3732 case 0:
3733 i.seg[i.mem_operands] = &es;
3734 break;
3735 case 1:
3736 i.seg[i.mem_operands] = &cs;
3737 break;
3738 case 2:
3739 i.seg[i.mem_operands] = &ss;
3740 break;
3741 case 3:
3742 i.seg[i.mem_operands] = &ds;
3743 break;
3744 case 4:
3745 i.seg[i.mem_operands] = &fs;
3746 break;
3747 case 5:
3748 i.seg[i.mem_operands] = &gs;
3749 break;
3750 }
3751
3752 /* Skip the ':' and whitespace. */
3753 ++op_string;
3754 if (is_space_char (*op_string))
3755 ++op_string;
3756
3757 if (!is_digit_char (*op_string)
3758 && !is_identifier_char (*op_string)
3759 && *op_string != '('
3760 && *op_string != ABSOLUTE_PREFIX)
3761 {
3762 as_bad (_("bad memory operand `%s'"), op_string);
3763 return 0;
3764 }
3765 /* Handle case of %es:*foo. */
3766 if (*op_string == ABSOLUTE_PREFIX)
3767 {
3768 ++op_string;
3769 if (is_space_char (*op_string))
3770 ++op_string;
3771 i.types[this_operand] |= JumpAbsolute;
3772 }
3773 goto do_memory_reference;
3774 }
3775 if (*op_string)
3776 {
3777 as_bad (_("junk `%s' after register"), op_string);
3778 return 0;
3779 }
3780 i.types[this_operand] |= r->reg_type & ~BaseIndex;
3781 i.op[this_operand].regs = r;
3782 i.reg_operands++;
3783 }
3784 else if (*op_string == REGISTER_PREFIX)
3785 {
3786 as_bad (_("bad register name `%s'"), op_string);
3787 return 0;
3788 }
3789 else if (*op_string == IMMEDIATE_PREFIX)
3790 {
3791 ++op_string;
3792 if (i.types[this_operand] & JumpAbsolute)
3793 {
3794 as_bad (_("immediate operand illegal with absolute jump"));
3795 return 0;
3796 }
3797 if (!i386_immediate (op_string))
3798 return 0;
3799 }
3800 else if (is_digit_char (*op_string)
3801 || is_identifier_char (*op_string)
3802 || *op_string == '(')
3803 {
3804 /* This is a memory reference of some sort. */
3805 char *base_string;
3806
3807 /* Start and end of displacement string expression (if found). */
3808 char *displacement_string_start;
3809 char *displacement_string_end;
3810
3811 do_memory_reference:
3812 if ((i.mem_operands == 1
3813 && (current_templates->start->opcode_modifier & IsString) == 0)
3814 || i.mem_operands == 2)
3815 {
3816 as_bad (_("too many memory references for `%s'"),
3817 current_templates->start->name);
3818 return 0;
3819 }
3820
3821 /* Check for base index form. We detect the base index form by
3822 looking for an ')' at the end of the operand, searching
3823 for the '(' matching it, and finding a REGISTER_PREFIX or ','
3824 after the '('. */
3825 base_string = op_string + strlen (op_string);
3826
3827 --base_string;
3828 if (is_space_char (*base_string))
3829 --base_string;
3830
3831 /* If we only have a displacement, set-up for it to be parsed later. */
3832 displacement_string_start = op_string;
3833 displacement_string_end = base_string + 1;
3834
3835 if (*base_string == ')')
3836 {
3837 char *temp_string;
3838 unsigned int parens_balanced = 1;
3839 /* We've already checked that the number of left & right ()'s are
3840 equal, so this loop will not be infinite. */
3841 do
3842 {
3843 base_string--;
3844 if (*base_string == ')')
3845 parens_balanced++;
3846 if (*base_string == '(')
3847 parens_balanced--;
3848 }
3849 while (parens_balanced);
3850
3851 temp_string = base_string;
3852
3853 /* Skip past '(' and whitespace. */
3854 ++base_string;
3855 if (is_space_char (*base_string))
3856 ++base_string;
3857
3858 if (*base_string == ','
3859 || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3860 && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
3861 {
3862 displacement_string_end = temp_string;
3863
3864 i.types[this_operand] |= BaseIndex;
3865
3866 if (i.base_reg)
3867 {
3868 base_string = end_op;
3869 if (is_space_char (*base_string))
3870 ++base_string;
3871 }
3872
3873 /* There may be an index reg or scale factor here. */
3874 if (*base_string == ',')
3875 {
3876 ++base_string;
3877 if (is_space_char (*base_string))
3878 ++base_string;
3879
3880 if ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3881 && (i.index_reg = parse_register (base_string, &end_op)) != NULL)
3882 {
3883 base_string = end_op;
3884 if (is_space_char (*base_string))
3885 ++base_string;
3886 if (*base_string == ',')
3887 {
3888 ++base_string;
3889 if (is_space_char (*base_string))
3890 ++base_string;
3891 }
3892 else if (*base_string != ')')
3893 {
3894 as_bad (_("expecting `,' or `)' after index register in `%s'"),
3895 operand_string);
3896 return 0;
3897 }
3898 }
3899 else if (*base_string == REGISTER_PREFIX)
3900 {
3901 as_bad (_("bad register name `%s'"), base_string);
3902 return 0;
3903 }
3904
3905 /* Check for scale factor. */
3906 if (*base_string != ')')
3907 {
3908 char *end_scale = i386_scale (base_string);
3909
3910 if (!end_scale)
3911 return 0;
3912
3913 base_string = end_scale;
3914 if (is_space_char (*base_string))
3915 ++base_string;
3916 if (*base_string != ')')
3917 {
3918 as_bad (_("expecting `)' after scale factor in `%s'"),
3919 operand_string);
3920 return 0;
3921 }
3922 }
3923 else if (!i.index_reg)
3924 {
3925 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
3926 *base_string);
3927 return 0;
3928 }
3929 }
3930 else if (*base_string != ')')
3931 {
3932 as_bad (_("expecting `,' or `)' after base register in `%s'"),
3933 operand_string);
3934 return 0;
3935 }
3936 }
3937 else if (*base_string == REGISTER_PREFIX)
3938 {
3939 as_bad (_("bad register name `%s'"), base_string);
3940 return 0;
3941 }
3942 }
3943
3944 /* If there's an expression beginning the operand, parse it,
3945 assuming displacement_string_start and
3946 displacement_string_end are meaningful. */
3947 if (displacement_string_start != displacement_string_end)
3948 {
3949 if (!i386_displacement (displacement_string_start,
3950 displacement_string_end))
3951 return 0;
3952 }
3953
3954 /* Special case for (%dx) while doing input/output op. */
3955 if (i.base_reg
3956 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
3957 && i.index_reg == 0
3958 && i.log2_scale_factor == 0
3959 && i.seg[i.mem_operands] == 0
3960 && (i.types[this_operand] & Disp) == 0)
3961 {
3962 i.types[this_operand] = InOutPortReg;
3963 return 1;
3964 }
3965
3966 if (i386_index_check (operand_string) == 0)
3967 return 0;
3968 i.mem_operands++;
3969 }
3970 else
3971 {
3972 /* It's not a memory operand; argh! */
3973 as_bad (_("invalid char %s beginning operand %d `%s'"),
3974 output_invalid (*op_string),
3975 this_operand + 1,
3976 op_string);
3977 return 0;
3978 }
3979 return 1; /* Normal return. */
3980 }
3981 \f
3982 /* md_estimate_size_before_relax()
3983
3984 Called just before relax() for rs_machine_dependent frags. The x86
3985 assembler uses these frags to handle variable size jump
3986 instructions.
3987
3988 Any symbol that is now undefined will not become defined.
3989 Return the correct fr_subtype in the frag.
3990 Return the initial "guess for variable size of frag" to caller.
3991 The guess is actually the growth beyond the fixed part. Whatever
3992 we do to grow the fixed or variable part contributes to our
3993 returned value. */
3994
3995 int
3996 md_estimate_size_before_relax (fragP, segment)
3997 register fragS *fragP;
3998 register segT segment;
3999 {
4000 /* We've already got fragP->fr_subtype right; all we have to do is
4001 check for un-relaxable symbols. On an ELF system, we can't relax
4002 an externally visible symbol, because it may be overridden by a
4003 shared library. */
4004 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
4005 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4006 || S_IS_EXTERNAL (fragP->fr_symbol)
4007 || S_IS_WEAK (fragP->fr_symbol)
4008 #endif
4009 )
4010 {
4011 /* Symbol is undefined in this segment, or we need to keep a
4012 reloc so that weak symbols can be overridden. */
4013 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
4014 RELOC_ENUM reloc_type;
4015 unsigned char *opcode;
4016 int old_fr_fix;
4017
4018 if (fragP->fr_var != NO_RELOC)
4019 reloc_type = fragP->fr_var;
4020 else if (size == 2)
4021 reloc_type = BFD_RELOC_16_PCREL;
4022 else
4023 reloc_type = BFD_RELOC_32_PCREL;
4024
4025 old_fr_fix = fragP->fr_fix;
4026 opcode = (unsigned char *) fragP->fr_opcode;
4027
4028 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
4029 {
4030 case UNCOND_JUMP:
4031 /* Make jmp (0xeb) a (d)word displacement jump. */
4032 opcode[0] = 0xe9;
4033 fragP->fr_fix += size;
4034 fix_new (fragP, old_fr_fix, size,
4035 fragP->fr_symbol,
4036 fragP->fr_offset, 1,
4037 reloc_type);
4038 break;
4039
4040 case COND_JUMP86:
4041 if (no_cond_jump_promotion)
4042 goto relax_guess;
4043
4044 if (size == 2)
4045 {
4046 /* Negate the condition, and branch past an
4047 unconditional jump. */
4048 opcode[0] ^= 1;
4049 opcode[1] = 3;
4050 /* Insert an unconditional jump. */
4051 opcode[2] = 0xe9;
4052 /* We added two extra opcode bytes, and have a two byte
4053 offset. */
4054 fragP->fr_fix += 2 + 2;
4055 fix_new (fragP, old_fr_fix + 2, 2,
4056 fragP->fr_symbol,
4057 fragP->fr_offset, 1,
4058 reloc_type);
4059 break;
4060 }
4061 /* Fall through. */
4062
4063 case COND_JUMP:
4064 if (no_cond_jump_promotion)
4065 goto relax_guess;
4066
4067 /* This changes the byte-displacement jump 0x7N
4068 to the (d)word-displacement jump 0x0f,0x8N. */
4069 opcode[1] = opcode[0] + 0x10;
4070 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4071 /* We've added an opcode byte. */
4072 fragP->fr_fix += 1 + size;
4073 fix_new (fragP, old_fr_fix + 1, size,
4074 fragP->fr_symbol,
4075 fragP->fr_offset, 1,
4076 reloc_type);
4077 break;
4078
4079 default:
4080 BAD_CASE (fragP->fr_subtype);
4081 break;
4082 }
4083 frag_wane (fragP);
4084 return fragP->fr_fix - old_fr_fix;
4085 }
4086
4087 relax_guess:
4088 /* Guess size depending on current relax state. Initially the relax
4089 state will correspond to a short jump and we return 1, because
4090 the variable part of the frag (the branch offset) is one byte
4091 long. However, we can relax a section more than once and in that
4092 case we must either set fr_subtype back to the unrelaxed state,
4093 or return the value for the appropriate branch. */
4094 return md_relax_table[fragP->fr_subtype].rlx_length;
4095 }
4096
4097 /* Called after relax() is finished.
4098
4099 In: Address of frag.
4100 fr_type == rs_machine_dependent.
4101 fr_subtype is what the address relaxed to.
4102
4103 Out: Any fixSs and constants are set up.
4104 Caller will turn frag into a ".space 0". */
4105
4106 #ifndef BFD_ASSEMBLER
4107 void
4108 md_convert_frag (headers, sec, fragP)
4109 object_headers *headers ATTRIBUTE_UNUSED;
4110 segT sec ATTRIBUTE_UNUSED;
4111 register fragS *fragP;
4112 #else
4113 void
4114 md_convert_frag (abfd, sec, fragP)
4115 bfd *abfd ATTRIBUTE_UNUSED;
4116 segT sec ATTRIBUTE_UNUSED;
4117 register fragS *fragP;
4118 #endif
4119 {
4120 register unsigned char *opcode;
4121 unsigned char *where_to_put_displacement = NULL;
4122 offsetT target_address;
4123 offsetT opcode_address;
4124 unsigned int extension = 0;
4125 offsetT displacement_from_opcode_start;
4126
4127 opcode = (unsigned char *) fragP->fr_opcode;
4128
4129 /* Address we want to reach in file space. */
4130 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
4131
4132 /* Address opcode resides at in file space. */
4133 opcode_address = fragP->fr_address + fragP->fr_fix;
4134
4135 /* Displacement from opcode start to fill into instruction. */
4136 displacement_from_opcode_start = target_address - opcode_address;
4137
4138 if ((fragP->fr_subtype & BIG) == 0)
4139 {
4140 /* Don't have to change opcode. */
4141 extension = 1; /* 1 opcode + 1 displacement */
4142 where_to_put_displacement = &opcode[1];
4143 }
4144 else
4145 {
4146 if (no_cond_jump_promotion
4147 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
4148 as_warn_where (fragP->fr_file, fragP->fr_line, _("long jump required"));
4149
4150 switch (fragP->fr_subtype)
4151 {
4152 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
4153 extension = 4; /* 1 opcode + 4 displacement */
4154 opcode[0] = 0xe9;
4155 where_to_put_displacement = &opcode[1];
4156 break;
4157
4158 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
4159 extension = 2; /* 1 opcode + 2 displacement */
4160 opcode[0] = 0xe9;
4161 where_to_put_displacement = &opcode[1];
4162 break;
4163
4164 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
4165 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
4166 extension = 5; /* 2 opcode + 4 displacement */
4167 opcode[1] = opcode[0] + 0x10;
4168 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4169 where_to_put_displacement = &opcode[2];
4170 break;
4171
4172 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
4173 extension = 3; /* 2 opcode + 2 displacement */
4174 opcode[1] = opcode[0] + 0x10;
4175 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4176 where_to_put_displacement = &opcode[2];
4177 break;
4178
4179 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
4180 extension = 4;
4181 opcode[0] ^= 1;
4182 opcode[1] = 3;
4183 opcode[2] = 0xe9;
4184 where_to_put_displacement = &opcode[3];
4185 break;
4186
4187 default:
4188 BAD_CASE (fragP->fr_subtype);
4189 break;
4190 }
4191 }
4192
4193 /* Now put displacement after opcode. */
4194 md_number_to_chars ((char *) where_to_put_displacement,
4195 (valueT) (displacement_from_opcode_start - extension),
4196 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
4197 fragP->fr_fix += extension;
4198 }
4199 \f
4200 /* Size of byte displacement jmp. */
4201 int md_short_jump_size = 2;
4202
4203 /* Size of dword displacement jmp. */
4204 int md_long_jump_size = 5;
4205
4206 /* Size of relocation record. */
4207 const int md_reloc_size = 8;
4208
4209 void
4210 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
4211 char *ptr;
4212 addressT from_addr, to_addr;
4213 fragS *frag ATTRIBUTE_UNUSED;
4214 symbolS *to_symbol ATTRIBUTE_UNUSED;
4215 {
4216 offsetT offset;
4217
4218 offset = to_addr - (from_addr + 2);
4219 /* Opcode for byte-disp jump. */
4220 md_number_to_chars (ptr, (valueT) 0xeb, 1);
4221 md_number_to_chars (ptr + 1, (valueT) offset, 1);
4222 }
4223
4224 void
4225 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
4226 char *ptr;
4227 addressT from_addr, to_addr;
4228 fragS *frag ATTRIBUTE_UNUSED;
4229 symbolS *to_symbol ATTRIBUTE_UNUSED;
4230 {
4231 offsetT offset;
4232
4233 offset = to_addr - (from_addr + 5);
4234 md_number_to_chars (ptr, (valueT) 0xe9, 1);
4235 md_number_to_chars (ptr + 1, (valueT) offset, 4);
4236 }
4237 \f
4238 /* Apply a fixup (fixS) to segment data, once it has been determined
4239 by our caller that we have all the info we need to fix it up.
4240
4241 On the 386, immediates, displacements, and data pointers are all in
4242 the same (little-endian) format, so we don't need to care about which
4243 we are handling. */
4244
4245 void
4246 md_apply_fix3 (fixP, valP, seg)
4247 /* The fix we're to put in. */
4248 fixS *fixP;
4249 /* Pointer to the value of the bits. */
4250 valueT * valP;
4251 /* Segment fix is from. */
4252 segT seg ATTRIBUTE_UNUSED;
4253 {
4254 char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
4255 valueT value = * valP;
4256
4257 #if defined (BFD_ASSEMBLER) && !defined (TE_Mach)
4258 if (fixP->fx_pcrel)
4259 {
4260 switch (fixP->fx_r_type)
4261 {
4262 default:
4263 break;
4264
4265 case BFD_RELOC_32:
4266 fixP->fx_r_type = BFD_RELOC_32_PCREL;
4267 break;
4268 case BFD_RELOC_16:
4269 fixP->fx_r_type = BFD_RELOC_16_PCREL;
4270 break;
4271 case BFD_RELOC_8:
4272 fixP->fx_r_type = BFD_RELOC_8_PCREL;
4273 break;
4274 }
4275 }
4276
4277 /* This is a hack. There should be a better way to handle this.
4278 This covers for the fact that bfd_install_relocation will
4279 subtract the current location (for partial_inplace, PC relative
4280 relocations); see more below. */
4281 if ((fixP->fx_r_type == BFD_RELOC_32_PCREL
4282 || fixP->fx_r_type == BFD_RELOC_16_PCREL
4283 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
4284 && fixP->fx_addsy && !use_rela_relocations)
4285 {
4286 #ifndef OBJ_AOUT
4287 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4288 #ifdef TE_PE
4289 || OUTPUT_FLAVOR == bfd_target_coff_flavour
4290 #endif
4291 )
4292 value += fixP->fx_where + fixP->fx_frag->fr_address;
4293 #endif
4294 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4295 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
4296 {
4297 segT fseg = S_GET_SEGMENT (fixP->fx_addsy);
4298
4299 if ((fseg == seg
4300 || (symbol_section_p (fixP->fx_addsy)
4301 && fseg != absolute_section))
4302 && ! S_IS_EXTERNAL (fixP->fx_addsy)
4303 && ! S_IS_WEAK (fixP->fx_addsy)
4304 && S_IS_DEFINED (fixP->fx_addsy)
4305 && ! S_IS_COMMON (fixP->fx_addsy))
4306 {
4307 /* Yes, we add the values in twice. This is because
4308 bfd_perform_relocation subtracts them out again. I think
4309 bfd_perform_relocation is broken, but I don't dare change
4310 it. FIXME. */
4311 value += fixP->fx_where + fixP->fx_frag->fr_address;
4312 }
4313 }
4314 #endif
4315 #if defined (OBJ_COFF) && defined (TE_PE)
4316 /* For some reason, the PE format does not store a section
4317 address offset for a PC relative symbol. */
4318 if (S_GET_SEGMENT (fixP->fx_addsy) != seg)
4319 value += md_pcrel_from (fixP);
4320 #endif
4321 }
4322
4323 /* Fix a few things - the dynamic linker expects certain values here,
4324 and we must not dissappoint it. */
4325 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4326 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4327 && fixP->fx_addsy)
4328 switch (fixP->fx_r_type)
4329 {
4330 case BFD_RELOC_386_PLT32:
4331 case BFD_RELOC_X86_64_PLT32:
4332 /* Make the jump instruction point to the address of the operand. At
4333 runtime we merely add the offset to the actual PLT entry. */
4334 value = -4;
4335 break;
4336 case BFD_RELOC_386_GOTPC:
4337
4338 /* This is tough to explain. We end up with this one if we have
4339 * operands that look like "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal
4340 * here is to obtain the absolute address of the GOT, and it is strongly
4341 * preferable from a performance point of view to avoid using a runtime
4342 * relocation for this. The actual sequence of instructions often look
4343 * something like:
4344 *
4345 * call .L66
4346 * .L66:
4347 * popl %ebx
4348 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
4349 *
4350 * The call and pop essentially return the absolute address of
4351 * the label .L66 and store it in %ebx. The linker itself will
4352 * ultimately change the first operand of the addl so that %ebx points to
4353 * the GOT, but to keep things simple, the .o file must have this operand
4354 * set so that it generates not the absolute address of .L66, but the
4355 * absolute address of itself. This allows the linker itself simply
4356 * treat a GOTPC relocation as asking for a pcrel offset to the GOT to be
4357 * added in, and the addend of the relocation is stored in the operand
4358 * field for the instruction itself.
4359 *
4360 * Our job here is to fix the operand so that it would add the correct
4361 * offset so that %ebx would point to itself. The thing that is tricky is
4362 * that .-.L66 will point to the beginning of the instruction, so we need
4363 * to further modify the operand so that it will point to itself.
4364 * There are other cases where you have something like:
4365 *
4366 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
4367 *
4368 * and here no correction would be required. Internally in the assembler
4369 * we treat operands of this form as not being pcrel since the '.' is
4370 * explicitly mentioned, and I wonder whether it would simplify matters
4371 * to do it this way. Who knows. In earlier versions of the PIC patches,
4372 * the pcrel_adjust field was used to store the correction, but since the
4373 * expression is not pcrel, I felt it would be confusing to do it this
4374 * way. */
4375
4376 value -= 1;
4377 break;
4378 case BFD_RELOC_386_GOT32:
4379 case BFD_RELOC_X86_64_GOT32:
4380 value = 0; /* Fully resolved at runtime. No addend. */
4381 break;
4382 case BFD_RELOC_386_GOTOFF:
4383 case BFD_RELOC_X86_64_GOTPCREL:
4384 break;
4385
4386 case BFD_RELOC_VTABLE_INHERIT:
4387 case BFD_RELOC_VTABLE_ENTRY:
4388 fixP->fx_done = 0;
4389 return;
4390
4391 default:
4392 break;
4393 }
4394 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
4395 * valP = value;
4396 #endif /* defined (BFD_ASSEMBLER) && !defined (TE_Mach) */
4397
4398 /* Are we finished with this relocation now? */
4399 if (fixP->fx_addsy == NULL && fixP->fx_pcrel == 0)
4400 fixP->fx_done = 1;
4401 #ifdef BFD_ASSEMBLER
4402 else if (use_rela_relocations)
4403 {
4404 fixP->fx_no_overflow = 1;
4405 value = 0;
4406 }
4407 #endif
4408 md_number_to_chars (p, value, fixP->fx_size);
4409 }
4410 \f
4411 #define MAX_LITTLENUMS 6
4412
4413 /* Turn the string pointed to by litP into a floating point constant
4414 of type TYPE, and emit the appropriate bytes. The number of
4415 LITTLENUMS emitted is stored in *SIZEP. An error message is
4416 returned, or NULL on OK. */
4417
4418 char *
4419 md_atof (type, litP, sizeP)
4420 int type;
4421 char *litP;
4422 int *sizeP;
4423 {
4424 int prec;
4425 LITTLENUM_TYPE words[MAX_LITTLENUMS];
4426 LITTLENUM_TYPE *wordP;
4427 char *t;
4428
4429 switch (type)
4430 {
4431 case 'f':
4432 case 'F':
4433 prec = 2;
4434 break;
4435
4436 case 'd':
4437 case 'D':
4438 prec = 4;
4439 break;
4440
4441 case 'x':
4442 case 'X':
4443 prec = 5;
4444 break;
4445
4446 default:
4447 *sizeP = 0;
4448 return _("Bad call to md_atof ()");
4449 }
4450 t = atof_ieee (input_line_pointer, type, words);
4451 if (t)
4452 input_line_pointer = t;
4453
4454 *sizeP = prec * sizeof (LITTLENUM_TYPE);
4455 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
4456 the bigendian 386. */
4457 for (wordP = words + prec - 1; prec--;)
4458 {
4459 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
4460 litP += sizeof (LITTLENUM_TYPE);
4461 }
4462 return 0;
4463 }
4464 \f
4465 char output_invalid_buf[8];
4466
4467 static char *
4468 output_invalid (c)
4469 int c;
4470 {
4471 if (ISPRINT (c))
4472 sprintf (output_invalid_buf, "'%c'", c);
4473 else
4474 sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
4475 return output_invalid_buf;
4476 }
4477
4478 /* REG_STRING starts *before* REGISTER_PREFIX. */
4479
4480 static const reg_entry *
4481 parse_register (reg_string, end_op)
4482 char *reg_string;
4483 char **end_op;
4484 {
4485 char *s = reg_string;
4486 char *p;
4487 char reg_name_given[MAX_REG_NAME_SIZE + 1];
4488 const reg_entry *r;
4489
4490 /* Skip possible REGISTER_PREFIX and possible whitespace. */
4491 if (*s == REGISTER_PREFIX)
4492 ++s;
4493
4494 if (is_space_char (*s))
4495 ++s;
4496
4497 p = reg_name_given;
4498 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
4499 {
4500 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
4501 return (const reg_entry *) NULL;
4502 s++;
4503 }
4504
4505 /* For naked regs, make sure that we are not dealing with an identifier.
4506 This prevents confusing an identifier like `eax_var' with register
4507 `eax'. */
4508 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
4509 return (const reg_entry *) NULL;
4510
4511 *end_op = s;
4512
4513 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
4514
4515 /* Handle floating point regs, allowing spaces in the (i) part. */
4516 if (r == i386_regtab /* %st is first entry of table */)
4517 {
4518 if (is_space_char (*s))
4519 ++s;
4520 if (*s == '(')
4521 {
4522 ++s;
4523 if (is_space_char (*s))
4524 ++s;
4525 if (*s >= '0' && *s <= '7')
4526 {
4527 r = &i386_float_regtab[*s - '0'];
4528 ++s;
4529 if (is_space_char (*s))
4530 ++s;
4531 if (*s == ')')
4532 {
4533 *end_op = s + 1;
4534 return r;
4535 }
4536 }
4537 /* We have "%st(" then garbage. */
4538 return (const reg_entry *) NULL;
4539 }
4540 }
4541
4542 if (r != NULL
4543 && r->reg_flags & (RegRex64|RegRex)
4544 && flag_code != CODE_64BIT)
4545 {
4546 return (const reg_entry *) NULL;
4547 }
4548
4549 return r;
4550 }
4551 \f
4552 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4553 const char *md_shortopts = "kVQ:sq";
4554 #else
4555 const char *md_shortopts = "q";
4556 #endif
4557
4558 struct option md_longopts[] = {
4559 #define OPTION_32 (OPTION_MD_BASE + 0)
4560 {"32", no_argument, NULL, OPTION_32},
4561 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4562 #define OPTION_64 (OPTION_MD_BASE + 1)
4563 {"64", no_argument, NULL, OPTION_64},
4564 #endif
4565 {NULL, no_argument, NULL, 0}
4566 };
4567 size_t md_longopts_size = sizeof (md_longopts);
4568
4569 int
4570 md_parse_option (c, arg)
4571 int c;
4572 char *arg ATTRIBUTE_UNUSED;
4573 {
4574 switch (c)
4575 {
4576 case 'q':
4577 quiet_warnings = 1;
4578 break;
4579
4580 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4581 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
4582 should be emitted or not. FIXME: Not implemented. */
4583 case 'Q':
4584 break;
4585
4586 /* -V: SVR4 argument to print version ID. */
4587 case 'V':
4588 print_version_id ();
4589 break;
4590
4591 /* -k: Ignore for FreeBSD compatibility. */
4592 case 'k':
4593 break;
4594
4595 case 's':
4596 /* -s: On i386 Solaris, this tells the native assembler to use
4597 .stab instead of .stab.excl. We always use .stab anyhow. */
4598 break;
4599
4600 case OPTION_64:
4601 {
4602 const char **list, **l;
4603
4604 list = bfd_target_list ();
4605 for (l = list; *l != NULL; l++)
4606 if (strcmp (*l, "elf64-x86-64") == 0)
4607 {
4608 default_arch = "x86_64";
4609 break;
4610 }
4611 if (*l == NULL)
4612 as_fatal (_("No compiled in support for x86_64"));
4613 free (list);
4614 }
4615 break;
4616 #endif
4617
4618 case OPTION_32:
4619 default_arch = "i386";
4620 break;
4621
4622 default:
4623 return 0;
4624 }
4625 return 1;
4626 }
4627
4628 void
4629 md_show_usage (stream)
4630 FILE *stream;
4631 {
4632 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4633 fprintf (stream, _("\
4634 -Q ignored\n\
4635 -V print assembler version number\n\
4636 -k ignored\n\
4637 -q quieten some warnings\n\
4638 -s ignored\n"));
4639 #else
4640 fprintf (stream, _("\
4641 -q quieten some warnings\n"));
4642 #endif
4643 }
4644
4645 #ifdef BFD_ASSEMBLER
4646 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
4647 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
4648
4649 /* Pick the target format to use. */
4650
4651 const char *
4652 i386_target_format ()
4653 {
4654 if (!strcmp (default_arch, "x86_64"))
4655 set_code_flag (CODE_64BIT);
4656 else if (!strcmp (default_arch, "i386"))
4657 set_code_flag (CODE_32BIT);
4658 else
4659 as_fatal (_("Unknown architecture"));
4660 switch (OUTPUT_FLAVOR)
4661 {
4662 #ifdef OBJ_MAYBE_AOUT
4663 case bfd_target_aout_flavour:
4664 return AOUT_TARGET_FORMAT;
4665 #endif
4666 #ifdef OBJ_MAYBE_COFF
4667 case bfd_target_coff_flavour:
4668 return "coff-i386";
4669 #endif
4670 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
4671 case bfd_target_elf_flavour:
4672 {
4673 if (flag_code == CODE_64BIT)
4674 use_rela_relocations = 1;
4675 return flag_code == CODE_64BIT ? "elf64-x86-64" : "elf32-i386";
4676 }
4677 #endif
4678 default:
4679 abort ();
4680 return NULL;
4681 }
4682 }
4683
4684 #endif /* OBJ_MAYBE_ more than one */
4685
4686 #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
4687 void i386_elf_emit_arch_note ()
4688 {
4689 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4690 && cpu_arch_name != NULL)
4691 {
4692 char *p;
4693 asection *seg = now_seg;
4694 subsegT subseg = now_subseg;
4695 Elf_Internal_Note i_note;
4696 Elf_External_Note e_note;
4697 asection *note_secp;
4698 int len;
4699
4700 /* Create the .note section. */
4701 note_secp = subseg_new (".note", 0);
4702 bfd_set_section_flags (stdoutput,
4703 note_secp,
4704 SEC_HAS_CONTENTS | SEC_READONLY);
4705
4706 /* Process the arch string. */
4707 len = strlen (cpu_arch_name);
4708
4709 i_note.namesz = len + 1;
4710 i_note.descsz = 0;
4711 i_note.type = NT_ARCH;
4712 p = frag_more (sizeof (e_note.namesz));
4713 md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
4714 p = frag_more (sizeof (e_note.descsz));
4715 md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
4716 p = frag_more (sizeof (e_note.type));
4717 md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
4718 p = frag_more (len + 1);
4719 strcpy (p, cpu_arch_name);
4720
4721 frag_align (2, 0, 0);
4722
4723 subseg_set (seg, subseg);
4724 }
4725 }
4726 #endif
4727 #endif /* BFD_ASSEMBLER */
4728 \f
4729 symbolS *
4730 md_undefined_symbol (name)
4731 char *name;
4732 {
4733 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
4734 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
4735 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
4736 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
4737 {
4738 if (!GOT_symbol)
4739 {
4740 if (symbol_find (name))
4741 as_bad (_("GOT already in symbol table"));
4742 GOT_symbol = symbol_new (name, undefined_section,
4743 (valueT) 0, &zero_address_frag);
4744 };
4745 return GOT_symbol;
4746 }
4747 return 0;
4748 }
4749
4750 /* Round up a section size to the appropriate boundary. */
4751
4752 valueT
4753 md_section_align (segment, size)
4754 segT segment ATTRIBUTE_UNUSED;
4755 valueT size;
4756 {
4757 #ifdef BFD_ASSEMBLER
4758 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4759 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
4760 {
4761 /* For a.out, force the section size to be aligned. If we don't do
4762 this, BFD will align it for us, but it will not write out the
4763 final bytes of the section. This may be a bug in BFD, but it is
4764 easier to fix it here since that is how the other a.out targets
4765 work. */
4766 int align;
4767
4768 align = bfd_get_section_alignment (stdoutput, segment);
4769 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
4770 }
4771 #endif
4772 #endif
4773
4774 return size;
4775 }
4776
4777 /* On the i386, PC-relative offsets are relative to the start of the
4778 next instruction. That is, the address of the offset, plus its
4779 size, since the offset is always the last part of the insn. */
4780
4781 long
4782 md_pcrel_from (fixP)
4783 fixS *fixP;
4784 {
4785 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
4786 }
4787
4788 #ifndef I386COFF
4789
4790 static void
4791 s_bss (ignore)
4792 int ignore ATTRIBUTE_UNUSED;
4793 {
4794 register int temp;
4795
4796 temp = get_absolute_expression ();
4797 subseg_set (bss_section, (subsegT) temp);
4798 demand_empty_rest_of_line ();
4799 }
4800
4801 #endif
4802
4803 #ifdef BFD_ASSEMBLER
4804
4805 void
4806 i386_validate_fix (fixp)
4807 fixS *fixp;
4808 {
4809 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
4810 {
4811 /* GOTOFF relocation are nonsense in 64bit mode. */
4812 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
4813 {
4814 if (flag_code != CODE_64BIT)
4815 abort ();
4816 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
4817 }
4818 else
4819 {
4820 if (flag_code == CODE_64BIT)
4821 abort ();
4822 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
4823 }
4824 fixp->fx_subsy = 0;
4825 }
4826 }
4827
4828 arelent *
4829 tc_gen_reloc (section, fixp)
4830 asection *section ATTRIBUTE_UNUSED;
4831 fixS *fixp;
4832 {
4833 arelent *rel;
4834 bfd_reloc_code_real_type code;
4835
4836 switch (fixp->fx_r_type)
4837 {
4838 case BFD_RELOC_X86_64_PLT32:
4839 case BFD_RELOC_X86_64_GOT32:
4840 case BFD_RELOC_X86_64_GOTPCREL:
4841 case BFD_RELOC_386_PLT32:
4842 case BFD_RELOC_386_GOT32:
4843 case BFD_RELOC_386_GOTOFF:
4844 case BFD_RELOC_386_GOTPC:
4845 case BFD_RELOC_X86_64_32S:
4846 case BFD_RELOC_RVA:
4847 case BFD_RELOC_VTABLE_ENTRY:
4848 case BFD_RELOC_VTABLE_INHERIT:
4849 code = fixp->fx_r_type;
4850 break;
4851 default:
4852 if (fixp->fx_pcrel)
4853 {
4854 switch (fixp->fx_size)
4855 {
4856 default:
4857 as_bad_where (fixp->fx_file, fixp->fx_line,
4858 _("can not do %d byte pc-relative relocation"),
4859 fixp->fx_size);
4860 code = BFD_RELOC_32_PCREL;
4861 break;
4862 case 1: code = BFD_RELOC_8_PCREL; break;
4863 case 2: code = BFD_RELOC_16_PCREL; break;
4864 case 4: code = BFD_RELOC_32_PCREL; break;
4865 }
4866 }
4867 else
4868 {
4869 switch (fixp->fx_size)
4870 {
4871 default:
4872 as_bad_where (fixp->fx_file, fixp->fx_line,
4873 _("can not do %d byte relocation"),
4874 fixp->fx_size);
4875 code = BFD_RELOC_32;
4876 break;
4877 case 1: code = BFD_RELOC_8; break;
4878 case 2: code = BFD_RELOC_16; break;
4879 case 4: code = BFD_RELOC_32; break;
4880 case 8: code = BFD_RELOC_64; break;
4881 }
4882 }
4883 break;
4884 }
4885
4886 if (code == BFD_RELOC_32
4887 && GOT_symbol
4888 && fixp->fx_addsy == GOT_symbol)
4889 {
4890 /* We don't support GOTPC on 64bit targets. */
4891 if (flag_code == CODE_64BIT)
4892 abort ();
4893 code = BFD_RELOC_386_GOTPC;
4894 }
4895
4896 rel = (arelent *) xmalloc (sizeof (arelent));
4897 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
4898 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
4899
4900 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
4901 if (!use_rela_relocations)
4902 {
4903 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
4904 vtable entry to be used in the relocation's section offset. */
4905 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
4906 rel->address = fixp->fx_offset;
4907
4908 if (fixp->fx_pcrel)
4909 rel->addend = fixp->fx_addnumber;
4910 else
4911 rel->addend = 0;
4912 }
4913 /* Use the rela in 64bit mode. */
4914 else
4915 {
4916 rel->addend = fixp->fx_offset;
4917 if (fixp->fx_pcrel)
4918 rel->addend -= fixp->fx_size;
4919 }
4920
4921 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
4922 if (rel->howto == NULL)
4923 {
4924 as_bad_where (fixp->fx_file, fixp->fx_line,
4925 _("cannot represent relocation type %s"),
4926 bfd_get_reloc_code_name (code));
4927 /* Set howto to a garbage value so that we can keep going. */
4928 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
4929 assert (rel->howto != NULL);
4930 }
4931
4932 return rel;
4933 }
4934
4935 #else /* ! BFD_ASSEMBLER */
4936
4937 #if (defined(OBJ_AOUT) | defined(OBJ_BOUT))
4938 void
4939 tc_aout_fix_to_chars (where, fixP, segment_address_in_file)
4940 char *where;
4941 fixS *fixP;
4942 relax_addressT segment_address_in_file;
4943 {
4944 /* In: length of relocation (or of address) in chars: 1, 2 or 4.
4945 Out: GNU LD relocation length code: 0, 1, or 2. */
4946
4947 static const unsigned char nbytes_r_length[] = { 42, 0, 1, 42, 2 };
4948 long r_symbolnum;
4949
4950 know (fixP->fx_addsy != NULL);
4951
4952 md_number_to_chars (where,
4953 (valueT) (fixP->fx_frag->fr_address
4954 + fixP->fx_where - segment_address_in_file),
4955 4);
4956
4957 r_symbolnum = (S_IS_DEFINED (fixP->fx_addsy)
4958 ? S_GET_TYPE (fixP->fx_addsy)
4959 : fixP->fx_addsy->sy_number);
4960
4961 where[6] = (r_symbolnum >> 16) & 0x0ff;
4962 where[5] = (r_symbolnum >> 8) & 0x0ff;
4963 where[4] = r_symbolnum & 0x0ff;
4964 where[7] = ((((!S_IS_DEFINED (fixP->fx_addsy)) << 3) & 0x08)
4965 | ((nbytes_r_length[fixP->fx_size] << 1) & 0x06)
4966 | (((fixP->fx_pcrel << 0) & 0x01) & 0x0f));
4967 }
4968
4969 #endif /* OBJ_AOUT or OBJ_BOUT. */
4970
4971 #if defined (I386COFF)
4972
4973 short
4974 tc_coff_fix2rtype (fixP)
4975 fixS *fixP;
4976 {
4977 if (fixP->fx_r_type == R_IMAGEBASE)
4978 return R_IMAGEBASE;
4979
4980 return (fixP->fx_pcrel ?
4981 (fixP->fx_size == 1 ? R_PCRBYTE :
4982 fixP->fx_size == 2 ? R_PCRWORD :
4983 R_PCRLONG) :
4984 (fixP->fx_size == 1 ? R_RELBYTE :
4985 fixP->fx_size == 2 ? R_RELWORD :
4986 R_DIR32));
4987 }
4988
4989 int
4990 tc_coff_sizemachdep (frag)
4991 fragS *frag;
4992 {
4993 if (frag->fr_next)
4994 return (frag->fr_next->fr_address - frag->fr_address);
4995 else
4996 return 0;
4997 }
4998
4999 #endif /* I386COFF */
5000
5001 #endif /* ! BFD_ASSEMBLER */
5002 \f
5003 /* Parse operands using Intel syntax. This implements a recursive descent
5004 parser based on the BNF grammar published in Appendix B of the MASM 6.1
5005 Programmer's Guide.
5006
5007 FIXME: We do not recognize the full operand grammar defined in the MASM
5008 documentation. In particular, all the structure/union and
5009 high-level macro operands are missing.
5010
5011 Uppercase words are terminals, lower case words are non-terminals.
5012 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
5013 bars '|' denote choices. Most grammar productions are implemented in
5014 functions called 'intel_<production>'.
5015
5016 Initial production is 'expr'.
5017
5018 addOp + | -
5019
5020 alpha [a-zA-Z]
5021
5022 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
5023
5024 constant digits [[ radixOverride ]]
5025
5026 dataType BYTE | WORD | DWORD | QWORD | XWORD
5027
5028 digits decdigit
5029 | digits decdigit
5030 | digits hexdigit
5031
5032 decdigit [0-9]
5033
5034 e05 e05 addOp e06
5035 | e06
5036
5037 e06 e06 mulOp e09
5038 | e09
5039
5040 e09 OFFSET e10
5041 | e09 PTR e10
5042 | e09 : e10
5043 | e10
5044
5045 e10 e10 [ expr ]
5046 | e11
5047
5048 e11 ( expr )
5049 | [ expr ]
5050 | constant
5051 | dataType
5052 | id
5053 | $
5054 | register
5055
5056 => expr SHORT e05
5057 | e05
5058
5059 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
5060 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
5061
5062 hexdigit a | b | c | d | e | f
5063 | A | B | C | D | E | F
5064
5065 id alpha
5066 | id alpha
5067 | id decdigit
5068
5069 mulOp * | / | MOD
5070
5071 quote " | '
5072
5073 register specialRegister
5074 | gpRegister
5075 | byteRegister
5076
5077 segmentRegister CS | DS | ES | FS | GS | SS
5078
5079 specialRegister CR0 | CR2 | CR3
5080 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
5081 | TR3 | TR4 | TR5 | TR6 | TR7
5082
5083 We simplify the grammar in obvious places (e.g., register parsing is
5084 done by calling parse_register) and eliminate immediate left recursion
5085 to implement a recursive-descent parser.
5086
5087 expr SHORT e05
5088 | e05
5089
5090 e05 e06 e05'
5091
5092 e05' addOp e06 e05'
5093 | Empty
5094
5095 e06 e09 e06'
5096
5097 e06' mulOp e09 e06'
5098 | Empty
5099
5100 e09 OFFSET e10 e09'
5101 | e10 e09'
5102
5103 e09' PTR e10 e09'
5104 | : e10 e09'
5105 | Empty
5106
5107 e10 e11 e10'
5108
5109 e10' [ expr ] e10'
5110 | Empty
5111
5112 e11 ( expr )
5113 | [ expr ]
5114 | BYTE
5115 | WORD
5116 | DWORD
5117 | QWORD
5118 | XWORD
5119 | .
5120 | $
5121 | register
5122 | id
5123 | constant */
5124
5125 /* Parsing structure for the intel syntax parser. Used to implement the
5126 semantic actions for the operand grammar. */
5127 struct intel_parser_s
5128 {
5129 char *op_string; /* The string being parsed. */
5130 int got_a_float; /* Whether the operand is a float. */
5131 int op_modifier; /* Operand modifier. */
5132 int is_mem; /* 1 if operand is memory reference. */
5133 const reg_entry *reg; /* Last register reference found. */
5134 char *disp; /* Displacement string being built. */
5135 };
5136
5137 static struct intel_parser_s intel_parser;
5138
5139 /* Token structure for parsing intel syntax. */
5140 struct intel_token
5141 {
5142 int code; /* Token code. */
5143 const reg_entry *reg; /* Register entry for register tokens. */
5144 char *str; /* String representation. */
5145 };
5146
5147 static struct intel_token cur_token, prev_token;
5148
5149 /* Token codes for the intel parser. Since T_SHORT is already used
5150 by COFF, undefine it first to prevent a warning. */
5151 #define T_NIL -1
5152 #define T_CONST 1
5153 #define T_REG 2
5154 #define T_BYTE 3
5155 #define T_WORD 4
5156 #define T_DWORD 5
5157 #define T_QWORD 6
5158 #define T_XWORD 7
5159 #undef T_SHORT
5160 #define T_SHORT 8
5161 #define T_OFFSET 9
5162 #define T_PTR 10
5163 #define T_ID 11
5164
5165 /* Prototypes for intel parser functions. */
5166 static int intel_match_token PARAMS ((int code));
5167 static void intel_get_token PARAMS ((void));
5168 static void intel_putback_token PARAMS ((void));
5169 static int intel_expr PARAMS ((void));
5170 static int intel_e05 PARAMS ((void));
5171 static int intel_e05_1 PARAMS ((void));
5172 static int intel_e06 PARAMS ((void));
5173 static int intel_e06_1 PARAMS ((void));
5174 static int intel_e09 PARAMS ((void));
5175 static int intel_e09_1 PARAMS ((void));
5176 static int intel_e10 PARAMS ((void));
5177 static int intel_e10_1 PARAMS ((void));
5178 static int intel_e11 PARAMS ((void));
5179
5180 static int
5181 i386_intel_operand (operand_string, got_a_float)
5182 char *operand_string;
5183 int got_a_float;
5184 {
5185 int ret;
5186 char *p;
5187
5188 /* Initialize token holders. */
5189 cur_token.code = prev_token.code = T_NIL;
5190 cur_token.reg = prev_token.reg = NULL;
5191 cur_token.str = prev_token.str = NULL;
5192
5193 /* Initialize parser structure. */
5194 p = intel_parser.op_string = (char *) malloc (strlen (operand_string) + 1);
5195 if (p == NULL)
5196 abort ();
5197 strcpy (intel_parser.op_string, operand_string);
5198 intel_parser.got_a_float = got_a_float;
5199 intel_parser.op_modifier = -1;
5200 intel_parser.is_mem = 0;
5201 intel_parser.reg = NULL;
5202 intel_parser.disp = (char *) malloc (strlen (operand_string) + 1);
5203 if (intel_parser.disp == NULL)
5204 abort ();
5205 intel_parser.disp[0] = '\0';
5206
5207 /* Read the first token and start the parser. */
5208 intel_get_token ();
5209 ret = intel_expr ();
5210
5211 if (ret)
5212 {
5213 /* If we found a memory reference, hand it over to i386_displacement
5214 to fill in the rest of the operand fields. */
5215 if (intel_parser.is_mem)
5216 {
5217 if ((i.mem_operands == 1
5218 && (current_templates->start->opcode_modifier & IsString) == 0)
5219 || i.mem_operands == 2)
5220 {
5221 as_bad (_("too many memory references for '%s'"),
5222 current_templates->start->name);
5223 ret = 0;
5224 }
5225 else
5226 {
5227 char *s = intel_parser.disp;
5228 i.mem_operands++;
5229
5230 /* Add the displacement expression. */
5231 if (*s != '\0')
5232 ret = i386_displacement (s, s + strlen (s))
5233 && i386_index_check (s);
5234 }
5235 }
5236
5237 /* Constant and OFFSET expressions are handled by i386_immediate. */
5238 else if (intel_parser.op_modifier == OFFSET_FLAT
5239 || intel_parser.reg == NULL)
5240 ret = i386_immediate (intel_parser.disp);
5241 }
5242
5243 free (p);
5244 free (intel_parser.disp);
5245
5246 return ret;
5247 }
5248
5249 /* expr SHORT e05
5250 | e05 */
5251 static int
5252 intel_expr ()
5253 {
5254 /* expr SHORT e05 */
5255 if (cur_token.code == T_SHORT)
5256 {
5257 intel_parser.op_modifier = SHORT;
5258 intel_match_token (T_SHORT);
5259
5260 return (intel_e05 ());
5261 }
5262
5263 /* expr e05 */
5264 else
5265 return intel_e05 ();
5266 }
5267
5268 /* e05 e06 e05'
5269
5270 e05' addOp e06 e05'
5271 | Empty */
5272 static int
5273 intel_e05 ()
5274 {
5275 return (intel_e06 () && intel_e05_1 ());
5276 }
5277
5278 static int
5279 intel_e05_1 ()
5280 {
5281 /* e05' addOp e06 e05' */
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_e06 () && intel_e05_1 ());
5288 }
5289
5290 /* e05' Empty */
5291 else
5292 return 1;
5293 }
5294
5295 /* e06 e09 e06'
5296
5297 e06' mulOp e09 e06'
5298 | Empty */
5299 static int
5300 intel_e06 ()
5301 {
5302 return (intel_e09 () && intel_e06_1 ());
5303 }
5304
5305 static int
5306 intel_e06_1 ()
5307 {
5308 /* e06' mulOp e09 e06' */
5309 if (cur_token.code == '*' || cur_token.code == '/')
5310 {
5311 strcat (intel_parser.disp, cur_token.str);
5312 intel_match_token (cur_token.code);
5313
5314 return (intel_e09 () && intel_e06_1 ());
5315 }
5316
5317 /* e06' Empty */
5318 else
5319 return 1;
5320 }
5321
5322 /* e09 OFFSET e10 e09'
5323 | e10 e09'
5324
5325 e09' PTR e10 e09'
5326 | : e10 e09'
5327 | Empty */
5328 static int
5329 intel_e09 ()
5330 {
5331 /* e09 OFFSET e10 e09' */
5332 if (cur_token.code == T_OFFSET)
5333 {
5334 intel_parser.is_mem = 0;
5335 intel_parser.op_modifier = OFFSET_FLAT;
5336 intel_match_token (T_OFFSET);
5337
5338 return (intel_e10 () && intel_e09_1 ());
5339 }
5340
5341 /* e09 e10 e09' */
5342 else
5343 return (intel_e10 () && intel_e09_1 ());
5344 }
5345
5346 static int
5347 intel_e09_1 ()
5348 {
5349 /* e09' PTR e10 e09' */
5350 if (cur_token.code == T_PTR)
5351 {
5352 if (prev_token.code == T_BYTE)
5353 i.suffix = BYTE_MNEM_SUFFIX;
5354
5355 else if (prev_token.code == T_WORD)
5356 {
5357 if (intel_parser.got_a_float == 2) /* "fi..." */
5358 i.suffix = SHORT_MNEM_SUFFIX;
5359 else
5360 i.suffix = WORD_MNEM_SUFFIX;
5361 }
5362
5363 else if (prev_token.code == T_DWORD)
5364 {
5365 if (intel_parser.got_a_float == 1) /* "f..." */
5366 i.suffix = SHORT_MNEM_SUFFIX;
5367 else
5368 i.suffix = LONG_MNEM_SUFFIX;
5369 }
5370
5371 else if (prev_token.code == T_QWORD)
5372 {
5373 if (intel_parser.got_a_float == 1) /* "f..." */
5374 i.suffix = LONG_MNEM_SUFFIX;
5375 else
5376 i.suffix = QWORD_MNEM_SUFFIX;
5377 }
5378
5379 else if (prev_token.code == T_XWORD)
5380 i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
5381
5382 else
5383 {
5384 as_bad (_("Unknown operand modifier `%s'\n"), prev_token.str);
5385 return 0;
5386 }
5387
5388 intel_match_token (T_PTR);
5389
5390 return (intel_e10 () && intel_e09_1 ());
5391 }
5392
5393 /* e09 : e10 e09' */
5394 else if (cur_token.code == ':')
5395 {
5396 /* Mark as a memory operand only if it's not already known to be an
5397 offset expression. */
5398 if (intel_parser.op_modifier != OFFSET_FLAT)
5399 intel_parser.is_mem = 1;
5400
5401 return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
5402 }
5403
5404 /* e09' Empty */
5405 else
5406 return 1;
5407 }
5408
5409 /* e10 e11 e10'
5410
5411 e10' [ expr ] e10'
5412 | Empty */
5413 static int
5414 intel_e10 ()
5415 {
5416 return (intel_e11 () && intel_e10_1 ());
5417 }
5418
5419 static int
5420 intel_e10_1 ()
5421 {
5422 /* e10' [ expr ] e10' */
5423 if (cur_token.code == '[')
5424 {
5425 intel_match_token ('[');
5426
5427 /* Mark as a memory operand only if it's not already known to be an
5428 offset expression. If it's an offset expression, we need to keep
5429 the brace in. */
5430 if (intel_parser.op_modifier != OFFSET_FLAT)
5431 intel_parser.is_mem = 1;
5432 else
5433 strcat (intel_parser.disp, "[");
5434
5435 /* Add a '+' to the displacement string if necessary. */
5436 if (*intel_parser.disp != '\0'
5437 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5438 strcat (intel_parser.disp, "+");
5439
5440 if (intel_expr () && intel_match_token (']'))
5441 {
5442 /* Preserve brackets when the operand is an offset expression. */
5443 if (intel_parser.op_modifier == OFFSET_FLAT)
5444 strcat (intel_parser.disp, "]");
5445
5446 return intel_e10_1 ();
5447 }
5448 else
5449 return 0;
5450 }
5451
5452 /* e10' Empty */
5453 else
5454 return 1;
5455 }
5456
5457 /* e11 ( expr )
5458 | [ expr ]
5459 | BYTE
5460 | WORD
5461 | DWORD
5462 | QWORD
5463 | XWORD
5464 | $
5465 | .
5466 | register
5467 | id
5468 | constant */
5469 static int
5470 intel_e11 ()
5471 {
5472 /* e11 ( expr ) */
5473 if (cur_token.code == '(')
5474 {
5475 intel_match_token ('(');
5476 strcat (intel_parser.disp, "(");
5477
5478 if (intel_expr () && intel_match_token (')'))
5479 {
5480 strcat (intel_parser.disp, ")");
5481 return 1;
5482 }
5483 else
5484 return 0;
5485 }
5486
5487 /* e11 [ expr ] */
5488 else if (cur_token.code == '[')
5489 {
5490 intel_match_token ('[');
5491
5492 /* Mark as a memory operand only if it's not already known to be an
5493 offset expression. If it's an offset expression, we need to keep
5494 the brace in. */
5495 if (intel_parser.op_modifier != OFFSET_FLAT)
5496 intel_parser.is_mem = 1;
5497 else
5498 strcat (intel_parser.disp, "[");
5499
5500 /* Operands for jump/call inside brackets denote absolute addresses. */
5501 if (current_templates->start->opcode_modifier & Jump
5502 || current_templates->start->opcode_modifier & JumpDword
5503 || current_templates->start->opcode_modifier & JumpByte
5504 || current_templates->start->opcode_modifier & JumpInterSegment)
5505 i.types[this_operand] |= JumpAbsolute;
5506
5507 /* Add a '+' to the displacement string if necessary. */
5508 if (*intel_parser.disp != '\0'
5509 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5510 strcat (intel_parser.disp, "+");
5511
5512 if (intel_expr () && intel_match_token (']'))
5513 {
5514 /* Preserve brackets when the operand is an offset expression. */
5515 if (intel_parser.op_modifier == OFFSET_FLAT)
5516 strcat (intel_parser.disp, "]");
5517
5518 return 1;
5519 }
5520 else
5521 return 0;
5522 }
5523
5524 /* e11 BYTE
5525 | WORD
5526 | DWORD
5527 | QWORD
5528 | XWORD */
5529 else if (cur_token.code == T_BYTE
5530 || cur_token.code == T_WORD
5531 || cur_token.code == T_DWORD
5532 || cur_token.code == T_QWORD
5533 || cur_token.code == T_XWORD)
5534 {
5535 intel_match_token (cur_token.code);
5536
5537 return 1;
5538 }
5539
5540 /* e11 $
5541 | . */
5542 else if (cur_token.code == '$' || cur_token.code == '.')
5543 {
5544 strcat (intel_parser.disp, cur_token.str);
5545 intel_match_token (cur_token.code);
5546
5547 /* Mark as a memory operand only if it's not already known to be an
5548 offset expression. */
5549 if (intel_parser.op_modifier != OFFSET_FLAT)
5550 intel_parser.is_mem = 1;
5551
5552 return 1;
5553 }
5554
5555 /* e11 register */
5556 else if (cur_token.code == T_REG)
5557 {
5558 const reg_entry *reg = intel_parser.reg = cur_token.reg;
5559
5560 intel_match_token (T_REG);
5561
5562 /* Check for segment change. */
5563 if (cur_token.code == ':')
5564 {
5565 if (reg->reg_type & (SReg2 | SReg3))
5566 {
5567 switch (reg->reg_num)
5568 {
5569 case 0:
5570 i.seg[i.mem_operands] = &es;
5571 break;
5572 case 1:
5573 i.seg[i.mem_operands] = &cs;
5574 break;
5575 case 2:
5576 i.seg[i.mem_operands] = &ss;
5577 break;
5578 case 3:
5579 i.seg[i.mem_operands] = &ds;
5580 break;
5581 case 4:
5582 i.seg[i.mem_operands] = &fs;
5583 break;
5584 case 5:
5585 i.seg[i.mem_operands] = &gs;
5586 break;
5587 }
5588 }
5589 else
5590 {
5591 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
5592 return 0;
5593 }
5594 }
5595
5596 /* Not a segment register. Check for register scaling. */
5597 else if (cur_token.code == '*')
5598 {
5599 if (!intel_parser.is_mem)
5600 {
5601 as_bad (_("Register scaling only allowed in memory operands."));
5602 return 0;
5603 }
5604
5605 /* What follows must be a valid scale. */
5606 if (intel_match_token ('*')
5607 && strchr ("01248", *cur_token.str))
5608 {
5609 i.index_reg = reg;
5610 i.types[this_operand] |= BaseIndex;
5611
5612 /* Set the scale after setting the register (otherwise,
5613 i386_scale will complain) */
5614 i386_scale (cur_token.str);
5615 intel_match_token (T_CONST);
5616 }
5617 else
5618 {
5619 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
5620 cur_token.str);
5621 return 0;
5622 }
5623 }
5624
5625 /* No scaling. If this is a memory operand, the register is either a
5626 base register (first occurrence) or an index register (second
5627 occurrence). */
5628 else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3)))
5629 {
5630 if (i.base_reg && i.index_reg)
5631 {
5632 as_bad (_("Too many register references in memory operand.\n"));
5633 return 0;
5634 }
5635
5636 if (i.base_reg == NULL)
5637 i.base_reg = reg;
5638 else
5639 i.index_reg = reg;
5640
5641 i.types[this_operand] |= BaseIndex;
5642 }
5643
5644 /* Offset modifier. Add the register to the displacement string to be
5645 parsed as an immediate expression after we're done. */
5646 else if (intel_parser.op_modifier == OFFSET_FLAT)
5647 strcat (intel_parser.disp, reg->reg_name);
5648
5649 /* It's neither base nor index nor offset. */
5650 else
5651 {
5652 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
5653 i.op[this_operand].regs = reg;
5654 i.reg_operands++;
5655 }
5656
5657 /* Since registers are not part of the displacement string (except
5658 when we're parsing offset operands), we may need to remove any
5659 preceding '+' from the displacement string. */
5660 if (*intel_parser.disp != '\0'
5661 && intel_parser.op_modifier != OFFSET_FLAT)
5662 {
5663 char *s = intel_parser.disp;
5664 s += strlen (s) - 1;
5665 if (*s == '+')
5666 *s = '\0';
5667 }
5668
5669 return 1;
5670 }
5671
5672 /* e11 id */
5673 else if (cur_token.code == T_ID)
5674 {
5675 /* Add the identifier to the displacement string. */
5676 strcat (intel_parser.disp, cur_token.str);
5677 intel_match_token (T_ID);
5678
5679 /* The identifier represents a memory reference only if it's not
5680 preceded by an offset modifier. */
5681 if (intel_parser.op_modifier != OFFSET_FLAT)
5682 intel_parser.is_mem = 1;
5683
5684 return 1;
5685 }
5686
5687 /* e11 constant */
5688 else if (cur_token.code == T_CONST
5689 || cur_token.code == '-'
5690 || cur_token.code == '+')
5691 {
5692 char *save_str;
5693
5694 /* Allow constants that start with `+' or `-'. */
5695 if (cur_token.code == '-' || cur_token.code == '+')
5696 {
5697 strcat (intel_parser.disp, cur_token.str);
5698 intel_match_token (cur_token.code);
5699 if (cur_token.code != T_CONST)
5700 {
5701 as_bad (_("Syntax error. Expecting a constant. Got `%s'.\n"),
5702 cur_token.str);
5703 return 0;
5704 }
5705 }
5706
5707 save_str = (char *) malloc (strlen (cur_token.str) + 1);
5708 if (save_str == NULL)
5709 abort ();
5710 strcpy (save_str, cur_token.str);
5711
5712 /* Get the next token to check for register scaling. */
5713 intel_match_token (cur_token.code);
5714
5715 /* Check if this constant is a scaling factor for an index register. */
5716 if (cur_token.code == '*')
5717 {
5718 if (intel_match_token ('*') && cur_token.code == T_REG)
5719 {
5720 if (!intel_parser.is_mem)
5721 {
5722 as_bad (_("Register scaling only allowed in memory operands."));
5723 return 0;
5724 }
5725
5726 /* The constant is followed by `* reg', so it must be
5727 a valid scale. */
5728 if (strchr ("01248", *save_str))
5729 {
5730 i.index_reg = cur_token.reg;
5731 i.types[this_operand] |= BaseIndex;
5732
5733 /* Set the scale after setting the register (otherwise,
5734 i386_scale will complain) */
5735 i386_scale (save_str);
5736 intel_match_token (T_REG);
5737
5738 /* Since registers are not part of the displacement
5739 string, we may need to remove any preceding '+' from
5740 the displacement string. */
5741 if (*intel_parser.disp != '\0')
5742 {
5743 char *s = intel_parser.disp;
5744 s += strlen (s) - 1;
5745 if (*s == '+')
5746 *s = '\0';
5747 }
5748
5749 free (save_str);
5750
5751 return 1;
5752 }
5753 else
5754 return 0;
5755 }
5756
5757 /* The constant was not used for register scaling. Since we have
5758 already consumed the token following `*' we now need to put it
5759 back in the stream. */
5760 else
5761 intel_putback_token ();
5762 }
5763
5764 /* Add the constant to the displacement string. */
5765 strcat (intel_parser.disp, save_str);
5766 free (save_str);
5767
5768 return 1;
5769 }
5770
5771 as_bad (_("Unrecognized token '%s'"), cur_token.str);
5772 return 0;
5773 }
5774
5775 /* Match the given token against cur_token. If they match, read the next
5776 token from the operand string. */
5777 static int
5778 intel_match_token (code)
5779 int code;
5780 {
5781 if (cur_token.code == code)
5782 {
5783 intel_get_token ();
5784 return 1;
5785 }
5786 else
5787 {
5788 as_bad (_("Unexpected token `%s'\n"), cur_token.str);
5789 return 0;
5790 }
5791 }
5792
5793 /* Read a new token from intel_parser.op_string and store it in cur_token. */
5794 static void
5795 intel_get_token ()
5796 {
5797 char *end_op;
5798 const reg_entry *reg;
5799 struct intel_token new_token;
5800
5801 new_token.code = T_NIL;
5802 new_token.reg = NULL;
5803 new_token.str = NULL;
5804
5805 /* Free the memory allocated to the previous token and move
5806 cur_token to prev_token. */
5807 if (prev_token.str)
5808 free (prev_token.str);
5809
5810 prev_token = cur_token;
5811
5812 /* Skip whitespace. */
5813 while (is_space_char (*intel_parser.op_string))
5814 intel_parser.op_string++;
5815
5816 /* Return an empty token if we find nothing else on the line. */
5817 if (*intel_parser.op_string == '\0')
5818 {
5819 cur_token = new_token;
5820 return;
5821 }
5822
5823 /* The new token cannot be larger than the remainder of the operand
5824 string. */
5825 new_token.str = (char *) malloc (strlen (intel_parser.op_string) + 1);
5826 if (new_token.str == NULL)
5827 abort ();
5828 new_token.str[0] = '\0';
5829
5830 if (strchr ("0123456789", *intel_parser.op_string))
5831 {
5832 char *p = new_token.str;
5833 char *q = intel_parser.op_string;
5834 new_token.code = T_CONST;
5835
5836 /* Allow any kind of identifier char to encompass floating point and
5837 hexadecimal numbers. */
5838 while (is_identifier_char (*q))
5839 *p++ = *q++;
5840 *p = '\0';
5841
5842 /* Recognize special symbol names [0-9][bf]. */
5843 if (strlen (intel_parser.op_string) == 2
5844 && (intel_parser.op_string[1] == 'b'
5845 || intel_parser.op_string[1] == 'f'))
5846 new_token.code = T_ID;
5847 }
5848
5849 else if (strchr ("+-/*:[]()", *intel_parser.op_string))
5850 {
5851 new_token.code = *intel_parser.op_string;
5852 new_token.str[0] = *intel_parser.op_string;
5853 new_token.str[1] = '\0';
5854 }
5855
5856 else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
5857 && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
5858 {
5859 new_token.code = T_REG;
5860 new_token.reg = reg;
5861
5862 if (*intel_parser.op_string == REGISTER_PREFIX)
5863 {
5864 new_token.str[0] = REGISTER_PREFIX;
5865 new_token.str[1] = '\0';
5866 }
5867
5868 strcat (new_token.str, reg->reg_name);
5869 }
5870
5871 else if (is_identifier_char (*intel_parser.op_string))
5872 {
5873 char *p = new_token.str;
5874 char *q = intel_parser.op_string;
5875
5876 /* A '.' or '$' followed by an identifier char is an identifier.
5877 Otherwise, it's operator '.' followed by an expression. */
5878 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
5879 {
5880 new_token.code = *q;
5881 new_token.str[0] = *q;
5882 new_token.str[1] = '\0';
5883 }
5884 else
5885 {
5886 while (is_identifier_char (*q) || *q == '@')
5887 *p++ = *q++;
5888 *p = '\0';
5889
5890 if (strcasecmp (new_token.str, "BYTE") == 0)
5891 new_token.code = T_BYTE;
5892
5893 else if (strcasecmp (new_token.str, "WORD") == 0)
5894 new_token.code = T_WORD;
5895
5896 else if (strcasecmp (new_token.str, "DWORD") == 0)
5897 new_token.code = T_DWORD;
5898
5899 else if (strcasecmp (new_token.str, "QWORD") == 0)
5900 new_token.code = T_QWORD;
5901
5902 else if (strcasecmp (new_token.str, "XWORD") == 0)
5903 new_token.code = T_XWORD;
5904
5905 else if (strcasecmp (new_token.str, "PTR") == 0)
5906 new_token.code = T_PTR;
5907
5908 else if (strcasecmp (new_token.str, "SHORT") == 0)
5909 new_token.code = T_SHORT;
5910
5911 else if (strcasecmp (new_token.str, "OFFSET") == 0)
5912 {
5913 new_token.code = T_OFFSET;
5914
5915 /* ??? This is not mentioned in the MASM grammar but gcc
5916 makes use of it with -mintel-syntax. OFFSET may be
5917 followed by FLAT: */
5918 if (strncasecmp (q, " FLAT:", 6) == 0)
5919 strcat (new_token.str, " FLAT:");
5920 }
5921
5922 /* ??? This is not mentioned in the MASM grammar. */
5923 else if (strcasecmp (new_token.str, "FLAT") == 0)
5924 new_token.code = T_OFFSET;
5925
5926 else
5927 new_token.code = T_ID;
5928 }
5929 }
5930
5931 else
5932 as_bad (_("Unrecognized token `%s'\n"), intel_parser.op_string);
5933
5934 intel_parser.op_string += strlen (new_token.str);
5935 cur_token = new_token;
5936 }
5937
5938 /* Put cur_token back into the token stream and make cur_token point to
5939 prev_token. */
5940 static void
5941 intel_putback_token ()
5942 {
5943 intel_parser.op_string -= strlen (cur_token.str);
5944 free (cur_token.str);
5945 cur_token = prev_token;
5946
5947 /* Forget prev_token. */
5948 prev_token.code = T_NIL;
5949 prev_token.reg = NULL;
5950 prev_token.str = NULL;
5951 }