* config/tc-i386.c (md_assemble): Handle Pentium4 branch hints.
[deliverable/binutils-gdb.git] / gas / config / tc-i386.c
1 /* i386.c -- Assemble code for the Intel 80386
2 Copyright 1989, 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
3 2000, 2001
4 Free Software Foundation, Inc.
5
6 This file is part of GAS, the GNU Assembler.
7
8 GAS is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2, or (at your option)
11 any later version.
12
13 GAS is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
17
18 You should have received a copy of the GNU General Public License
19 along with GAS; see the file COPYING. If not, write to the Free
20 Software Foundation, 59 Temple Place - Suite 330, Boston, MA
21 02111-1307, USA. */
22
23 /* Intel 80386 machine specific gas.
24 Written by Eliot Dresselhaus (eliot@mgm.mit.edu).
25 x86_64 support by Jan Hubicka (jh@suse.cz)
26 Bugs & suggestions are completely welcome. This is free software.
27 Please help us make it better. */
28
29 #include <ctype.h>
30
31 #include "as.h"
32 #include "subsegs.h"
33 #include "dwarf2dbg.h"
34 #include "opcode/i386.h"
35
36 #ifndef REGISTER_WARNINGS
37 #define REGISTER_WARNINGS 1
38 #endif
39
40 #ifndef INFER_ADDR_PREFIX
41 #define INFER_ADDR_PREFIX 1
42 #endif
43
44 #ifndef SCALE1_WHEN_NO_INDEX
45 /* Specifying a scale factor besides 1 when there is no index is
46 futile. eg. `mov (%ebx,2),%al' does exactly the same as
47 `mov (%ebx),%al'. To slavishly follow what the programmer
48 specified, set SCALE1_WHEN_NO_INDEX to 0. */
49 #define SCALE1_WHEN_NO_INDEX 1
50 #endif
51
52 #define true 1
53 #define false 0
54
55 static unsigned int mode_from_disp_size PARAMS ((unsigned int));
56 static int fits_in_signed_byte PARAMS ((offsetT));
57 static int fits_in_unsigned_byte PARAMS ((offsetT));
58 static int fits_in_unsigned_word PARAMS ((offsetT));
59 static int fits_in_signed_word PARAMS ((offsetT));
60 static int fits_in_unsigned_long PARAMS ((offsetT));
61 static int fits_in_signed_long PARAMS ((offsetT));
62 static int smallest_imm_type PARAMS ((offsetT));
63 static offsetT offset_in_range PARAMS ((offsetT, int));
64 static int add_prefix PARAMS ((unsigned int));
65 static void set_code_flag PARAMS ((int));
66 static void set_16bit_gcc_code_flag PARAMS ((int));
67 static void set_intel_syntax PARAMS ((int));
68 static void set_cpu_arch PARAMS ((int));
69
70 #ifdef BFD_ASSEMBLER
71 static bfd_reloc_code_real_type reloc
72 PARAMS ((int, int, int, bfd_reloc_code_real_type));
73 #define RELOC_ENUM enum bfd_reloc_code_real
74 #else
75 #define RELOC_ENUM int
76 #endif
77
78 #ifndef DEFAULT_ARCH
79 #define DEFAULT_ARCH "i386"
80 #endif
81 static char *default_arch = DEFAULT_ARCH;
82
83 /* 'md_assemble ()' gathers together information and puts it into a
84 i386_insn. */
85
86 union i386_op
87 {
88 expressionS *disps;
89 expressionS *imms;
90 const reg_entry *regs;
91 };
92
93 struct _i386_insn
94 {
95 /* TM holds the template for the insn were currently assembling. */
96 template tm;
97
98 /* SUFFIX holds the instruction mnemonic suffix if given.
99 (e.g. 'l' for 'movl') */
100 char suffix;
101
102 /* OPERANDS gives the number of given operands. */
103 unsigned int operands;
104
105 /* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number
106 of given register, displacement, memory operands and immediate
107 operands. */
108 unsigned int reg_operands, disp_operands, mem_operands, imm_operands;
109
110 /* TYPES [i] is the type (see above #defines) which tells us how to
111 use OP[i] for the corresponding operand. */
112 unsigned int types[MAX_OPERANDS];
113
114 /* Displacement expression, immediate expression, or register for each
115 operand. */
116 union i386_op op[MAX_OPERANDS];
117
118 /* Flags for operands. */
119 unsigned int flags[MAX_OPERANDS];
120 #define Operand_PCrel 1
121
122 /* Relocation type for operand */
123 RELOC_ENUM reloc[MAX_OPERANDS];
124
125 /* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode
126 the base index byte below. */
127 const reg_entry *base_reg;
128 const reg_entry *index_reg;
129 unsigned int log2_scale_factor;
130
131 /* SEG gives the seg_entries of this insn. They are zero unless
132 explicit segment overrides are given. */
133 const seg_entry *seg[2];
134
135 /* PREFIX holds all the given prefix opcodes (usually null).
136 PREFIXES is the number of prefix opcodes. */
137 unsigned int prefixes;
138 unsigned char prefix[MAX_PREFIXES];
139
140 /* RM and SIB are the modrm byte and the sib byte where the
141 addressing modes of this insn are encoded. */
142
143 modrm_byte rm;
144 rex_byte rex;
145 sib_byte sib;
146 };
147
148 typedef struct _i386_insn i386_insn;
149
150 /* List of chars besides those in app.c:symbol_chars that can start an
151 operand. Used to prevent the scrubber eating vital white-space. */
152 #ifdef LEX_AT
153 const char extra_symbol_chars[] = "*%-(@";
154 #else
155 const char extra_symbol_chars[] = "*%-(";
156 #endif
157
158 /* This array holds the chars that always start a comment. If the
159 pre-processor is disabled, these aren't very useful. */
160 #if defined (TE_I386AIX) || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) && ! defined (TE_LINUX) && !defined(TE_FreeBSD))
161 /* Putting '/' here makes it impossible to use the divide operator.
162 However, we need it for compatibility with SVR4 systems. */
163 const char comment_chars[] = "#/";
164 #define PREFIX_SEPARATOR '\\'
165 #else
166 const char comment_chars[] = "#";
167 #define PREFIX_SEPARATOR '/'
168 #endif
169
170 /* This array holds the chars that only start a comment at the beginning of
171 a line. If the line seems to have the form '# 123 filename'
172 .line and .file directives will appear in the pre-processed output.
173 Note that input_file.c hand checks for '#' at the beginning of the
174 first line of the input file. This is because the compiler outputs
175 #NO_APP at the beginning of its output.
176 Also note that comments started like this one will always work if
177 '/' isn't otherwise defined. */
178 #if defined (TE_I386AIX) || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) && ! defined (TE_LINUX) && !defined(TE_FreeBSD))
179 const char line_comment_chars[] = "";
180 #else
181 const char line_comment_chars[] = "/";
182 #endif
183
184 const char line_separator_chars[] = ";";
185
186 /* Chars that can be used to separate mant from exp in floating point
187 nums. */
188 const char EXP_CHARS[] = "eE";
189
190 /* Chars that mean this number is a floating point constant
191 As in 0f12.456
192 or 0d1.2345e12. */
193 const char FLT_CHARS[] = "fFdDxX";
194
195 /* Tables for lexical analysis. */
196 static char mnemonic_chars[256];
197 static char register_chars[256];
198 static char operand_chars[256];
199 static char identifier_chars[256];
200 static char digit_chars[256];
201
202 /* Lexical macros. */
203 #define is_mnemonic_char(x) (mnemonic_chars[(unsigned char) x])
204 #define is_operand_char(x) (operand_chars[(unsigned char) x])
205 #define is_register_char(x) (register_chars[(unsigned char) x])
206 #define is_space_char(x) ((x) == ' ')
207 #define is_identifier_char(x) (identifier_chars[(unsigned char) x])
208 #define is_digit_char(x) (digit_chars[(unsigned char) x])
209
210 /* All non-digit non-letter charcters that may occur in an operand. */
211 static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
212
213 /* md_assemble() always leaves the strings it's passed unaltered. To
214 effect this we maintain a stack of saved characters that we've smashed
215 with '\0's (indicating end of strings for various sub-fields of the
216 assembler instruction). */
217 static char save_stack[32];
218 static char *save_stack_p;
219 #define END_STRING_AND_SAVE(s) \
220 do { *save_stack_p++ = *(s); *(s) = '\0'; } while (0)
221 #define RESTORE_END_STRING(s) \
222 do { *(s) = *--save_stack_p; } while (0)
223
224 /* The instruction we're assembling. */
225 static i386_insn i;
226
227 /* Possible templates for current insn. */
228 static const templates *current_templates;
229
230 /* Per instruction expressionS buffers: 2 displacements & 2 immediate max. */
231 static expressionS disp_expressions[2], im_expressions[2];
232
233 /* Current operand we are working on. */
234 static int this_operand;
235
236 /* We support four different modes. FLAG_CODE variable is used to distinguish
237 these. */
238
239 enum flag_code {
240 CODE_32BIT,
241 CODE_16BIT,
242 CODE_64BIT };
243 #define NUM_FLAG_CODE ((int) CODE_64BIT + 1)
244
245 static enum flag_code flag_code;
246 static int use_rela_relocations = 0;
247
248 /* The names used to print error messages. */
249 static const char *flag_code_names[] =
250 {
251 "32",
252 "16",
253 "64"
254 };
255
256 /* 1 for intel syntax,
257 0 if att syntax. */
258 static int intel_syntax = 0;
259
260 /* 1 if register prefix % not required. */
261 static int allow_naked_reg = 0;
262
263 /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
264 leave, push, and pop instructions so that gcc has the same stack
265 frame as in 32 bit mode. */
266 static char stackop_size = '\0';
267
268 /* Non-zero to quieten some warnings. */
269 static int quiet_warnings = 0;
270
271 /* CPU name. */
272 static const char *cpu_arch_name = NULL;
273
274 /* CPU feature flags. */
275 static unsigned int cpu_arch_flags = CpuUnknownFlags|CpuNo64;
276
277 /* If set, conditional jumps are not automatically promoted to handle
278 larger than a byte offset. */
279 static unsigned int no_cond_jump_promotion = 0;
280
281 /* Interface to relax_segment.
282 There are 3 major relax states for 386 jump insns because the
283 different types of jumps add different sizes to frags when we're
284 figuring out what sort of jump to choose to reach a given label. */
285
286 /* Types. */
287 #define UNCOND_JUMP 0
288 #define COND_JUMP 1
289 #define COND_JUMP86 2
290
291 /* Sizes. */
292 #define CODE16 1
293 #define SMALL 0
294 #define SMALL16 (SMALL|CODE16)
295 #define BIG 2
296 #define BIG16 (BIG|CODE16)
297
298 #ifndef INLINE
299 #ifdef __GNUC__
300 #define INLINE __inline__
301 #else
302 #define INLINE
303 #endif
304 #endif
305
306 #define ENCODE_RELAX_STATE(type, size) \
307 ((relax_substateT) (((type) << 2) | (size)))
308 #define TYPE_FROM_RELAX_STATE(s) \
309 ((s) >> 2)
310 #define DISP_SIZE_FROM_RELAX_STATE(s) \
311 ((((s) & 3) == BIG ? 4 : (((s) & 3) == BIG16 ? 2 : 1)))
312
313 /* This table is used by relax_frag to promote short jumps to long
314 ones where necessary. SMALL (short) jumps may be promoted to BIG
315 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
316 don't allow a short jump in a 32 bit code segment to be promoted to
317 a 16 bit offset jump because it's slower (requires data size
318 prefix), and doesn't work, unless the destination is in the bottom
319 64k of the code segment (The top 16 bits of eip are zeroed). */
320
321 const relax_typeS md_relax_table[] =
322 {
323 /* The fields are:
324 1) most positive reach of this state,
325 2) most negative reach of this state,
326 3) how many bytes this mode will have in the variable part of the frag
327 4) which index into the table to try if we can't fit into this one. */
328
329 /* UNCOND_JUMP states. */
330 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
331 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
332 /* dword jmp adds 4 bytes to frag:
333 0 extra opcode bytes, 4 displacement bytes. */
334 {0, 0, 4, 0},
335 /* word jmp adds 2 byte2 to frag:
336 0 extra opcode bytes, 2 displacement bytes. */
337 {0, 0, 2, 0},
338
339 /* COND_JUMP states. */
340 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
341 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
342 /* dword conditionals adds 5 bytes to frag:
343 1 extra opcode byte, 4 displacement bytes. */
344 {0, 0, 5, 0},
345 /* word conditionals add 3 bytes to frag:
346 1 extra opcode byte, 2 displacement bytes. */
347 {0, 0, 3, 0},
348
349 /* COND_JUMP86 states. */
350 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG)},
351 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG16)},
352 /* dword conditionals adds 5 bytes to frag:
353 1 extra opcode byte, 4 displacement bytes. */
354 {0, 0, 5, 0},
355 /* word conditionals add 4 bytes to frag:
356 1 displacement byte and a 3 byte long branch insn. */
357 {0, 0, 4, 0}
358 };
359
360 static const arch_entry cpu_arch[] = {
361 {"i8086", Cpu086 },
362 {"i186", Cpu086|Cpu186 },
363 {"i286", Cpu086|Cpu186|Cpu286 },
364 {"i386", Cpu086|Cpu186|Cpu286|Cpu386 },
365 {"i486", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486 },
366 {"i586", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
367 {"i686", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
368 {"pentium", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
369 {"pentiumpro",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
370 {"pentium4", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX|CpuSSE|CpuSSE2 },
371 {"k6", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow },
372 {"athlon", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuMMX|Cpu3dnow },
373 {"sledgehammer",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuSledgehammer|CpuMMX|Cpu3dnow|CpuSSE|CpuSSE2 },
374 {NULL, 0 }
375 };
376
377 void
378 i386_align_code (fragP, count)
379 fragS *fragP;
380 int count;
381 {
382 /* Various efficient no-op patterns for aligning code labels.
383 Note: Don't try to assemble the instructions in the comments.
384 0L and 0w are not legal. */
385 static const char f32_1[] =
386 {0x90}; /* nop */
387 static const char f32_2[] =
388 {0x89,0xf6}; /* movl %esi,%esi */
389 static const char f32_3[] =
390 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
391 static const char f32_4[] =
392 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
393 static const char f32_5[] =
394 {0x90, /* nop */
395 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
396 static const char f32_6[] =
397 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
398 static const char f32_7[] =
399 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
400 static const char f32_8[] =
401 {0x90, /* nop */
402 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
403 static const char f32_9[] =
404 {0x89,0xf6, /* movl %esi,%esi */
405 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
406 static const char f32_10[] =
407 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
408 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
409 static const char f32_11[] =
410 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
411 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
412 static const char f32_12[] =
413 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
414 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
415 static const char f32_13[] =
416 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
417 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
418 static const char f32_14[] =
419 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
420 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
421 static const char f32_15[] =
422 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
423 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
424 static const char f16_3[] =
425 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
426 static const char f16_4[] =
427 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
428 static const char f16_5[] =
429 {0x90, /* nop */
430 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
431 static const char f16_6[] =
432 {0x89,0xf6, /* mov %si,%si */
433 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
434 static const char f16_7[] =
435 {0x8d,0x74,0x00, /* lea 0(%si),%si */
436 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
437 static const char f16_8[] =
438 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
439 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
440 static const char *const f32_patt[] = {
441 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
442 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
443 };
444 static const char *const f16_patt[] = {
445 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
446 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
447 };
448
449 /* ??? We can't use these fillers for x86_64, since they often kills the
450 upper halves. Solve later. */
451 if (flag_code == CODE_64BIT)
452 count = 1;
453
454 if (count > 0 && count <= 15)
455 {
456 if (flag_code == CODE_16BIT)
457 {
458 memcpy (fragP->fr_literal + fragP->fr_fix,
459 f16_patt[count - 1], count);
460 if (count > 8)
461 /* Adjust jump offset. */
462 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
463 }
464 else
465 memcpy (fragP->fr_literal + fragP->fr_fix,
466 f32_patt[count - 1], count);
467 fragP->fr_var = count;
468 }
469 }
470
471 static char *output_invalid PARAMS ((int c));
472 static int i386_operand PARAMS ((char *operand_string));
473 static int i386_intel_operand PARAMS ((char *operand_string, int got_a_float));
474 static const reg_entry *parse_register PARAMS ((char *reg_string,
475 char **end_op));
476
477 #ifndef I386COFF
478 static void s_bss PARAMS ((int));
479 #endif
480
481 symbolS *GOT_symbol; /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
482
483 static INLINE unsigned int
484 mode_from_disp_size (t)
485 unsigned int t;
486 {
487 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
488 }
489
490 static INLINE int
491 fits_in_signed_byte (num)
492 offsetT num;
493 {
494 return (num >= -128) && (num <= 127);
495 }
496
497 static INLINE int
498 fits_in_unsigned_byte (num)
499 offsetT num;
500 {
501 return (num & 0xff) == num;
502 }
503
504 static INLINE int
505 fits_in_unsigned_word (num)
506 offsetT num;
507 {
508 return (num & 0xffff) == num;
509 }
510
511 static INLINE int
512 fits_in_signed_word (num)
513 offsetT num;
514 {
515 return (-32768 <= num) && (num <= 32767);
516 }
517 static INLINE int
518 fits_in_signed_long (num)
519 offsetT num ATTRIBUTE_UNUSED;
520 {
521 #ifndef BFD64
522 return 1;
523 #else
524 return (!(((offsetT) -1 << 31) & num)
525 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
526 #endif
527 } /* fits_in_signed_long() */
528 static INLINE int
529 fits_in_unsigned_long (num)
530 offsetT num ATTRIBUTE_UNUSED;
531 {
532 #ifndef BFD64
533 return 1;
534 #else
535 return (num & (((offsetT) 2 << 31) - 1)) == num;
536 #endif
537 } /* fits_in_unsigned_long() */
538
539 static int
540 smallest_imm_type (num)
541 offsetT num;
542 {
543 if (cpu_arch_flags != (Cpu086 | Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64)
544 && !(cpu_arch_flags & (CpuUnknown)))
545 {
546 /* This code is disabled on the 486 because all the Imm1 forms
547 in the opcode table are slower on the i486. They're the
548 versions with the implicitly specified single-position
549 displacement, which has another syntax if you really want to
550 use that form. */
551 if (num == 1)
552 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
553 }
554 return (fits_in_signed_byte (num)
555 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
556 : fits_in_unsigned_byte (num)
557 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
558 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
559 ? (Imm16 | Imm32 | Imm32S | Imm64)
560 : fits_in_signed_long (num)
561 ? (Imm32 | Imm32S | Imm64)
562 : fits_in_unsigned_long (num)
563 ? (Imm32 | Imm64)
564 : Imm64);
565 }
566
567 static offsetT
568 offset_in_range (val, size)
569 offsetT val;
570 int size;
571 {
572 addressT mask;
573
574 switch (size)
575 {
576 case 1: mask = ((addressT) 1 << 8) - 1; break;
577 case 2: mask = ((addressT) 1 << 16) - 1; break;
578 case 4: mask = ((addressT) 2 << 31) - 1; break;
579 #ifdef BFD64
580 case 8: mask = ((addressT) 2 << 63) - 1; break;
581 #endif
582 default: abort ();
583 }
584
585 /* If BFD64, sign extend val. */
586 if (!use_rela_relocations)
587 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
588 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
589
590 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
591 {
592 char buf1[40], buf2[40];
593
594 sprint_value (buf1, val);
595 sprint_value (buf2, val & mask);
596 as_warn (_("%s shortened to %s"), buf1, buf2);
597 }
598 return val & mask;
599 }
600
601 /* Returns 0 if attempting to add a prefix where one from the same
602 class already exists, 1 if non rep/repne added, 2 if rep/repne
603 added. */
604 static int
605 add_prefix (prefix)
606 unsigned int prefix;
607 {
608 int ret = 1;
609 int q;
610
611 if (prefix >= 0x40 && prefix < 0x50 && flag_code == CODE_64BIT)
612 q = REX_PREFIX;
613 else
614 switch (prefix)
615 {
616 default:
617 abort ();
618
619 case CS_PREFIX_OPCODE:
620 case DS_PREFIX_OPCODE:
621 case ES_PREFIX_OPCODE:
622 case FS_PREFIX_OPCODE:
623 case GS_PREFIX_OPCODE:
624 case SS_PREFIX_OPCODE:
625 q = SEG_PREFIX;
626 break;
627
628 case REPNE_PREFIX_OPCODE:
629 case REPE_PREFIX_OPCODE:
630 ret = 2;
631 /* fall thru */
632 case LOCK_PREFIX_OPCODE:
633 q = LOCKREP_PREFIX;
634 break;
635
636 case FWAIT_OPCODE:
637 q = WAIT_PREFIX;
638 break;
639
640 case ADDR_PREFIX_OPCODE:
641 q = ADDR_PREFIX;
642 break;
643
644 case DATA_PREFIX_OPCODE:
645 q = DATA_PREFIX;
646 break;
647 }
648
649 if (i.prefix[q])
650 {
651 as_bad (_("same type of prefix used twice"));
652 return 0;
653 }
654
655 i.prefixes += 1;
656 i.prefix[q] = prefix;
657 return ret;
658 }
659
660 static void
661 set_code_flag (value)
662 int value;
663 {
664 flag_code = value;
665 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
666 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
667 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
668 {
669 as_bad (_("64bit mode not supported on this CPU."));
670 }
671 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
672 {
673 as_bad (_("32bit mode not supported on this CPU."));
674 }
675 stackop_size = '\0';
676 }
677
678 static void
679 set_16bit_gcc_code_flag (new_code_flag)
680 int new_code_flag;
681 {
682 flag_code = new_code_flag;
683 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
684 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
685 stackop_size = 'l';
686 }
687
688 static void
689 set_intel_syntax (syntax_flag)
690 int syntax_flag;
691 {
692 /* Find out if register prefixing is specified. */
693 int ask_naked_reg = 0;
694
695 SKIP_WHITESPACE ();
696 if (! is_end_of_line[(unsigned char) *input_line_pointer])
697 {
698 char *string = input_line_pointer;
699 int e = get_symbol_end ();
700
701 if (strcmp (string, "prefix") == 0)
702 ask_naked_reg = 1;
703 else if (strcmp (string, "noprefix") == 0)
704 ask_naked_reg = -1;
705 else
706 as_bad (_("bad argument to syntax directive."));
707 *input_line_pointer = e;
708 }
709 demand_empty_rest_of_line ();
710
711 intel_syntax = syntax_flag;
712
713 if (ask_naked_reg == 0)
714 {
715 #ifdef BFD_ASSEMBLER
716 allow_naked_reg = (intel_syntax
717 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
718 #else
719 /* Conservative default. */
720 allow_naked_reg = 0;
721 #endif
722 }
723 else
724 allow_naked_reg = (ask_naked_reg < 0);
725 }
726
727 static void
728 set_cpu_arch (dummy)
729 int dummy ATTRIBUTE_UNUSED;
730 {
731 SKIP_WHITESPACE ();
732
733 if (! is_end_of_line[(unsigned char) *input_line_pointer])
734 {
735 char *string = input_line_pointer;
736 int e = get_symbol_end ();
737 int i;
738
739 for (i = 0; cpu_arch[i].name; i++)
740 {
741 if (strcmp (string, cpu_arch[i].name) == 0)
742 {
743 cpu_arch_name = cpu_arch[i].name;
744 cpu_arch_flags = (cpu_arch[i].flags
745 | (flag_code == CODE_64BIT ? Cpu64 : CpuNo64));
746 break;
747 }
748 }
749 if (!cpu_arch[i].name)
750 as_bad (_("no such architecture: `%s'"), string);
751
752 *input_line_pointer = e;
753 }
754 else
755 as_bad (_("missing cpu architecture"));
756
757 no_cond_jump_promotion = 0;
758 if (*input_line_pointer == ','
759 && ! is_end_of_line[(unsigned char) input_line_pointer[1]])
760 {
761 char *string = ++input_line_pointer;
762 int e = get_symbol_end ();
763
764 if (strcmp (string, "nojumps") == 0)
765 no_cond_jump_promotion = 1;
766 else if (strcmp (string, "jumps") == 0)
767 ;
768 else
769 as_bad (_("no such architecture modifier: `%s'"), string);
770
771 *input_line_pointer = e;
772 }
773
774 demand_empty_rest_of_line ();
775 }
776
777 const pseudo_typeS md_pseudo_table[] =
778 {
779 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
780 {"align", s_align_bytes, 0},
781 #else
782 {"align", s_align_ptwo, 0},
783 #endif
784 {"arch", set_cpu_arch, 0},
785 #ifndef I386COFF
786 {"bss", s_bss, 0},
787 #endif
788 {"ffloat", float_cons, 'f'},
789 {"dfloat", float_cons, 'd'},
790 {"tfloat", float_cons, 'x'},
791 {"value", cons, 2},
792 {"noopt", s_ignore, 0},
793 {"optim", s_ignore, 0},
794 {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
795 {"code16", set_code_flag, CODE_16BIT},
796 {"code32", set_code_flag, CODE_32BIT},
797 {"code64", set_code_flag, CODE_64BIT},
798 {"intel_syntax", set_intel_syntax, 1},
799 {"att_syntax", set_intel_syntax, 0},
800 {"file", dwarf2_directive_file, 0},
801 {"loc", dwarf2_directive_loc, 0},
802 {0, 0, 0}
803 };
804
805 /* For interface with expression (). */
806 extern char *input_line_pointer;
807
808 /* Hash table for instruction mnemonic lookup. */
809 static struct hash_control *op_hash;
810
811 /* Hash table for register lookup. */
812 static struct hash_control *reg_hash;
813 \f
814 #ifdef BFD_ASSEMBLER
815 unsigned long
816 i386_mach ()
817 {
818 if (!strcmp (default_arch, "x86_64"))
819 return bfd_mach_x86_64;
820 else if (!strcmp (default_arch, "i386"))
821 return bfd_mach_i386_i386;
822 else
823 as_fatal (_("Unknown architecture"));
824 }
825 #endif
826 \f
827 void
828 md_begin ()
829 {
830 const char *hash_err;
831
832 /* Initialize op_hash hash table. */
833 op_hash = hash_new ();
834
835 {
836 register const template *optab;
837 register templates *core_optab;
838
839 /* Setup for loop. */
840 optab = i386_optab;
841 core_optab = (templates *) xmalloc (sizeof (templates));
842 core_optab->start = optab;
843
844 while (1)
845 {
846 ++optab;
847 if (optab->name == NULL
848 || strcmp (optab->name, (optab - 1)->name) != 0)
849 {
850 /* different name --> ship out current template list;
851 add to hash table; & begin anew. */
852 core_optab->end = optab;
853 hash_err = hash_insert (op_hash,
854 (optab - 1)->name,
855 (PTR) core_optab);
856 if (hash_err)
857 {
858 as_fatal (_("Internal Error: Can't hash %s: %s"),
859 (optab - 1)->name,
860 hash_err);
861 }
862 if (optab->name == NULL)
863 break;
864 core_optab = (templates *) xmalloc (sizeof (templates));
865 core_optab->start = optab;
866 }
867 }
868 }
869
870 /* Initialize reg_hash hash table. */
871 reg_hash = hash_new ();
872 {
873 register const reg_entry *regtab;
874
875 for (regtab = i386_regtab;
876 regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
877 regtab++)
878 {
879 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
880 if (hash_err)
881 as_fatal (_("Internal Error: Can't hash %s: %s"),
882 regtab->reg_name,
883 hash_err);
884 }
885 }
886
887 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
888 {
889 register int c;
890 register char *p;
891
892 for (c = 0; c < 256; c++)
893 {
894 if (isdigit (c))
895 {
896 digit_chars[c] = c;
897 mnemonic_chars[c] = c;
898 register_chars[c] = c;
899 operand_chars[c] = c;
900 }
901 else if (islower (c))
902 {
903 mnemonic_chars[c] = c;
904 register_chars[c] = c;
905 operand_chars[c] = c;
906 }
907 else if (isupper (c))
908 {
909 mnemonic_chars[c] = tolower (c);
910 register_chars[c] = mnemonic_chars[c];
911 operand_chars[c] = c;
912 }
913
914 if (isalpha (c) || isdigit (c))
915 identifier_chars[c] = c;
916 else if (c >= 128)
917 {
918 identifier_chars[c] = c;
919 operand_chars[c] = c;
920 }
921 }
922
923 #ifdef LEX_AT
924 identifier_chars['@'] = '@';
925 #endif
926 digit_chars['-'] = '-';
927 identifier_chars['_'] = '_';
928 identifier_chars['.'] = '.';
929
930 for (p = operand_special_chars; *p != '\0'; p++)
931 operand_chars[(unsigned char) *p] = *p;
932 }
933
934 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
935 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
936 {
937 record_alignment (text_section, 2);
938 record_alignment (data_section, 2);
939 record_alignment (bss_section, 2);
940 }
941 #endif
942 }
943
944 void
945 i386_print_statistics (file)
946 FILE *file;
947 {
948 hash_print_statistics (file, "i386 opcode", op_hash);
949 hash_print_statistics (file, "i386 register", reg_hash);
950 }
951 \f
952 #ifdef DEBUG386
953
954 /* Debugging routines for md_assemble. */
955 static void pi PARAMS ((char *, i386_insn *));
956 static void pte PARAMS ((template *));
957 static void pt PARAMS ((unsigned int));
958 static void pe PARAMS ((expressionS *));
959 static void ps PARAMS ((symbolS *));
960
961 static void
962 pi (line, x)
963 char *line;
964 i386_insn *x;
965 {
966 unsigned int i;
967
968 fprintf (stdout, "%s: template ", line);
969 pte (&x->tm);
970 fprintf (stdout, " address: base %s index %s scale %x\n",
971 x->base_reg ? x->base_reg->reg_name : "none",
972 x->index_reg ? x->index_reg->reg_name : "none",
973 x->log2_scale_factor);
974 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
975 x->rm.mode, x->rm.reg, x->rm.regmem);
976 fprintf (stdout, " sib: base %x index %x scale %x\n",
977 x->sib.base, x->sib.index, x->sib.scale);
978 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
979 x->rex.mode64, x->rex.extX, x->rex.extY, x->rex.extZ);
980 for (i = 0; i < x->operands; i++)
981 {
982 fprintf (stdout, " #%d: ", i + 1);
983 pt (x->types[i]);
984 fprintf (stdout, "\n");
985 if (x->types[i]
986 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
987 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
988 if (x->types[i] & Imm)
989 pe (x->op[i].imms);
990 if (x->types[i] & Disp)
991 pe (x->op[i].disps);
992 }
993 }
994
995 static void
996 pte (t)
997 template *t;
998 {
999 unsigned int i;
1000 fprintf (stdout, " %d operands ", t->operands);
1001 fprintf (stdout, "opcode %x ", t->base_opcode);
1002 if (t->extension_opcode != None)
1003 fprintf (stdout, "ext %x ", t->extension_opcode);
1004 if (t->opcode_modifier & D)
1005 fprintf (stdout, "D");
1006 if (t->opcode_modifier & W)
1007 fprintf (stdout, "W");
1008 fprintf (stdout, "\n");
1009 for (i = 0; i < t->operands; i++)
1010 {
1011 fprintf (stdout, " #%d type ", i + 1);
1012 pt (t->operand_types[i]);
1013 fprintf (stdout, "\n");
1014 }
1015 }
1016
1017 static void
1018 pe (e)
1019 expressionS *e;
1020 {
1021 fprintf (stdout, " operation %d\n", e->X_op);
1022 fprintf (stdout, " add_number %ld (%lx)\n",
1023 (long) e->X_add_number, (long) e->X_add_number);
1024 if (e->X_add_symbol)
1025 {
1026 fprintf (stdout, " add_symbol ");
1027 ps (e->X_add_symbol);
1028 fprintf (stdout, "\n");
1029 }
1030 if (e->X_op_symbol)
1031 {
1032 fprintf (stdout, " op_symbol ");
1033 ps (e->X_op_symbol);
1034 fprintf (stdout, "\n");
1035 }
1036 }
1037
1038 static void
1039 ps (s)
1040 symbolS *s;
1041 {
1042 fprintf (stdout, "%s type %s%s",
1043 S_GET_NAME (s),
1044 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
1045 segment_name (S_GET_SEGMENT (s)));
1046 }
1047
1048 struct type_name
1049 {
1050 unsigned int mask;
1051 char *tname;
1052 }
1053
1054 type_names[] =
1055 {
1056 { Reg8, "r8" },
1057 { Reg16, "r16" },
1058 { Reg32, "r32" },
1059 { Reg64, "r64" },
1060 { Imm8, "i8" },
1061 { Imm8S, "i8s" },
1062 { Imm16, "i16" },
1063 { Imm32, "i32" },
1064 { Imm32S, "i32s" },
1065 { Imm64, "i64" },
1066 { Imm1, "i1" },
1067 { BaseIndex, "BaseIndex" },
1068 { Disp8, "d8" },
1069 { Disp16, "d16" },
1070 { Disp32, "d32" },
1071 { Disp32S, "d32s" },
1072 { Disp64, "d64" },
1073 { InOutPortReg, "InOutPortReg" },
1074 { ShiftCount, "ShiftCount" },
1075 { Control, "control reg" },
1076 { Test, "test reg" },
1077 { Debug, "debug reg" },
1078 { FloatReg, "FReg" },
1079 { FloatAcc, "FAcc" },
1080 { SReg2, "SReg2" },
1081 { SReg3, "SReg3" },
1082 { Acc, "Acc" },
1083 { JumpAbsolute, "Jump Absolute" },
1084 { RegMMX, "rMMX" },
1085 { RegXMM, "rXMM" },
1086 { EsSeg, "es" },
1087 { 0, "" }
1088 };
1089
1090 static void
1091 pt (t)
1092 unsigned int t;
1093 {
1094 register struct type_name *ty;
1095
1096 for (ty = type_names; ty->mask; ty++)
1097 if (t & ty->mask)
1098 fprintf (stdout, "%s, ", ty->tname);
1099 fflush (stdout);
1100 }
1101
1102 #endif /* DEBUG386 */
1103 \f
1104 int
1105 tc_i386_force_relocation (fixp)
1106 struct fix *fixp;
1107 {
1108 #ifdef BFD_ASSEMBLER
1109 if (fixp->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1110 || fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1111 return 1;
1112 return 0;
1113 #else
1114 /* For COFF. */
1115 return fixp->fx_r_type == 7;
1116 #endif
1117 }
1118
1119 #ifdef BFD_ASSEMBLER
1120
1121 static bfd_reloc_code_real_type
1122 reloc (size, pcrel, sign, other)
1123 int size;
1124 int pcrel;
1125 int sign;
1126 bfd_reloc_code_real_type other;
1127 {
1128 if (other != NO_RELOC)
1129 return other;
1130
1131 if (pcrel)
1132 {
1133 if (!sign)
1134 as_bad (_("There are no unsigned pc-relative relocations"));
1135 switch (size)
1136 {
1137 case 1: return BFD_RELOC_8_PCREL;
1138 case 2: return BFD_RELOC_16_PCREL;
1139 case 4: return BFD_RELOC_32_PCREL;
1140 }
1141 as_bad (_("can not do %d byte pc-relative relocation"), size);
1142 }
1143 else
1144 {
1145 if (sign)
1146 switch (size)
1147 {
1148 case 4: return BFD_RELOC_X86_64_32S;
1149 }
1150 else
1151 switch (size)
1152 {
1153 case 1: return BFD_RELOC_8;
1154 case 2: return BFD_RELOC_16;
1155 case 4: return BFD_RELOC_32;
1156 case 8: return BFD_RELOC_64;
1157 }
1158 as_bad (_("can not do %s %d byte relocation"),
1159 sign ? "signed" : "unsigned", size);
1160 }
1161
1162 abort ();
1163 return BFD_RELOC_NONE;
1164 }
1165
1166 /* Here we decide which fixups can be adjusted to make them relative to
1167 the beginning of the section instead of the symbol. Basically we need
1168 to make sure that the dynamic relocations are done correctly, so in
1169 some cases we force the original symbol to be used. */
1170
1171 int
1172 tc_i386_fix_adjustable (fixP)
1173 fixS *fixP;
1174 {
1175 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1176 /* Prevent all adjustments to global symbols, or else dynamic
1177 linking will not work correctly. */
1178 if (S_IS_EXTERNAL (fixP->fx_addsy)
1179 || S_IS_WEAK (fixP->fx_addsy))
1180 return 0;
1181 #endif
1182 /* adjust_reloc_syms doesn't know about the GOT. */
1183 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1184 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1185 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1186 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1187 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1188 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
1189 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1190 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1191 return 0;
1192 return 1;
1193 }
1194 #else
1195 #define reloc(SIZE,PCREL,SIGN,OTHER) 0
1196 #define BFD_RELOC_16 0
1197 #define BFD_RELOC_32 0
1198 #define BFD_RELOC_16_PCREL 0
1199 #define BFD_RELOC_32_PCREL 0
1200 #define BFD_RELOC_386_PLT32 0
1201 #define BFD_RELOC_386_GOT32 0
1202 #define BFD_RELOC_386_GOTOFF 0
1203 #define BFD_RELOC_X86_64_PLT32 0
1204 #define BFD_RELOC_X86_64_GOT32 0
1205 #define BFD_RELOC_X86_64_GOTPCREL 0
1206 #endif
1207
1208 static int intel_float_operand PARAMS ((char *mnemonic));
1209
1210 static int
1211 intel_float_operand (mnemonic)
1212 char *mnemonic;
1213 {
1214 if (mnemonic[0] == 'f' && mnemonic[1] == 'i')
1215 return 2;
1216
1217 if (mnemonic[0] == 'f')
1218 return 1;
1219
1220 return 0;
1221 }
1222
1223 /* This is the guts of the machine-dependent assembler. LINE points to a
1224 machine dependent instruction. This function is supposed to emit
1225 the frags/bytes it assembles to. */
1226
1227 void
1228 md_assemble (line)
1229 char *line;
1230 {
1231 /* Points to template once we've found it. */
1232 const template *t;
1233
1234 int j;
1235
1236 char mnemonic[MAX_MNEM_SIZE];
1237
1238 /* Initialize globals. */
1239 memset (&i, '\0', sizeof (i));
1240 for (j = 0; j < MAX_OPERANDS; j++)
1241 i.reloc[j] = NO_RELOC;
1242 memset (disp_expressions, '\0', sizeof (disp_expressions));
1243 memset (im_expressions, '\0', sizeof (im_expressions));
1244 save_stack_p = save_stack;
1245
1246 /* First parse an instruction mnemonic & call i386_operand for the operands.
1247 We assume that the scrubber has arranged it so that line[0] is the valid
1248 start of a (possibly prefixed) mnemonic. */
1249 {
1250 char *l = line;
1251 char *token_start = l;
1252 char *mnem_p;
1253
1254 /* Non-zero if we found a prefix only acceptable with string insns. */
1255 const char *expecting_string_instruction = NULL;
1256
1257 while (1)
1258 {
1259 mnem_p = mnemonic;
1260 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1261 {
1262 mnem_p++;
1263 if (mnem_p >= mnemonic + sizeof (mnemonic))
1264 {
1265 as_bad (_("no such instruction: `%s'"), token_start);
1266 return;
1267 }
1268 l++;
1269 }
1270 if (!is_space_char (*l)
1271 && *l != END_OF_INSN
1272 && *l != PREFIX_SEPARATOR)
1273 {
1274 as_bad (_("invalid character %s in mnemonic"),
1275 output_invalid (*l));
1276 return;
1277 }
1278 if (token_start == l)
1279 {
1280 if (*l == PREFIX_SEPARATOR)
1281 as_bad (_("expecting prefix; got nothing"));
1282 else
1283 as_bad (_("expecting mnemonic; got nothing"));
1284 return;
1285 }
1286
1287 /* Look up instruction (or prefix) via hash table. */
1288 current_templates = hash_find (op_hash, mnemonic);
1289
1290 if (*l != END_OF_INSN
1291 && (! is_space_char (*l) || l[1] != END_OF_INSN)
1292 && current_templates
1293 && (current_templates->start->opcode_modifier & IsPrefix))
1294 {
1295 /* If we are in 16-bit mode, do not allow addr16 or data16.
1296 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1297 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1298 && (((current_templates->start->opcode_modifier & Size32) != 0)
1299 ^ (flag_code == CODE_16BIT)))
1300 {
1301 as_bad (_("redundant %s prefix"),
1302 current_templates->start->name);
1303 return;
1304 }
1305 /* Add prefix, checking for repeated prefixes. */
1306 switch (add_prefix (current_templates->start->base_opcode))
1307 {
1308 case 0:
1309 return;
1310 case 2:
1311 expecting_string_instruction = current_templates->start->name;
1312 break;
1313 }
1314 /* Skip past PREFIX_SEPARATOR and reset token_start. */
1315 token_start = ++l;
1316 }
1317 else
1318 break;
1319 }
1320
1321 if (!current_templates)
1322 {
1323 /* See if we can get a match by trimming off a suffix. */
1324 switch (mnem_p[-1])
1325 {
1326 case WORD_MNEM_SUFFIX:
1327 case BYTE_MNEM_SUFFIX:
1328 case QWORD_MNEM_SUFFIX:
1329 i.suffix = mnem_p[-1];
1330 mnem_p[-1] = '\0';
1331 current_templates = hash_find (op_hash, mnemonic);
1332 break;
1333 case SHORT_MNEM_SUFFIX:
1334 case LONG_MNEM_SUFFIX:
1335 if (!intel_syntax)
1336 {
1337 i.suffix = mnem_p[-1];
1338 mnem_p[-1] = '\0';
1339 current_templates = hash_find (op_hash, mnemonic);
1340 }
1341 break;
1342
1343 /* Intel Syntax. */
1344 case 'd':
1345 if (intel_syntax)
1346 {
1347 if (intel_float_operand (mnemonic))
1348 i.suffix = SHORT_MNEM_SUFFIX;
1349 else
1350 i.suffix = LONG_MNEM_SUFFIX;
1351 mnem_p[-1] = '\0';
1352 current_templates = hash_find (op_hash, mnemonic);
1353 }
1354 break;
1355 }
1356 if (!current_templates)
1357 {
1358 as_bad (_("no such instruction: `%s'"), token_start);
1359 return;
1360 }
1361 }
1362
1363 /* Check if instruction is supported on specified architecture. */
1364 if (cpu_arch_flags != 0)
1365 {
1366 if ((current_templates->start->cpu_flags & ~(Cpu64 | CpuNo64))
1367 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64)))
1368 {
1369 as_warn (_("`%s' is not supported on `%s'"),
1370 current_templates->start->name, cpu_arch_name);
1371 }
1372 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
1373 {
1374 as_warn (_("use .code16 to ensure correct addressing mode"));
1375 }
1376 }
1377
1378 /* Check for rep/repne without a string instruction. */
1379 if (expecting_string_instruction
1380 && !(current_templates->start->opcode_modifier & IsString))
1381 {
1382 as_bad (_("expecting string instruction after `%s'"),
1383 expecting_string_instruction);
1384 return;
1385 }
1386
1387 /* There may be operands to parse. */
1388 if (*l != END_OF_INSN)
1389 {
1390 /* 1 if operand is pending after ','. */
1391 unsigned int expecting_operand = 0;
1392
1393 /* Non-zero if operand parens not balanced. */
1394 unsigned int paren_not_balanced;
1395
1396 do
1397 {
1398 /* Skip optional white space before operand. */
1399 if (is_space_char (*l))
1400 ++l;
1401 if (!is_operand_char (*l) && *l != END_OF_INSN)
1402 {
1403 as_bad (_("invalid character %s before operand %d"),
1404 output_invalid (*l),
1405 i.operands + 1);
1406 return;
1407 }
1408 token_start = l; /* after white space */
1409 paren_not_balanced = 0;
1410 while (paren_not_balanced || *l != ',')
1411 {
1412 if (*l == END_OF_INSN)
1413 {
1414 if (paren_not_balanced)
1415 {
1416 if (!intel_syntax)
1417 as_bad (_("unbalanced parenthesis in operand %d."),
1418 i.operands + 1);
1419 else
1420 as_bad (_("unbalanced brackets in operand %d."),
1421 i.operands + 1);
1422 return;
1423 }
1424 else
1425 break; /* we are done */
1426 }
1427 else if (!is_operand_char (*l) && !is_space_char (*l))
1428 {
1429 as_bad (_("invalid character %s in operand %d"),
1430 output_invalid (*l),
1431 i.operands + 1);
1432 return;
1433 }
1434 if (!intel_syntax)
1435 {
1436 if (*l == '(')
1437 ++paren_not_balanced;
1438 if (*l == ')')
1439 --paren_not_balanced;
1440 }
1441 else
1442 {
1443 if (*l == '[')
1444 ++paren_not_balanced;
1445 if (*l == ']')
1446 --paren_not_balanced;
1447 }
1448 l++;
1449 }
1450 if (l != token_start)
1451 { /* Yes, we've read in another operand. */
1452 unsigned int operand_ok;
1453 this_operand = i.operands++;
1454 if (i.operands > MAX_OPERANDS)
1455 {
1456 as_bad (_("spurious operands; (%d operands/instruction max)"),
1457 MAX_OPERANDS);
1458 return;
1459 }
1460 /* Now parse operand adding info to 'i' as we go along. */
1461 END_STRING_AND_SAVE (l);
1462
1463 if (intel_syntax)
1464 operand_ok =
1465 i386_intel_operand (token_start,
1466 intel_float_operand (mnemonic));
1467 else
1468 operand_ok = i386_operand (token_start);
1469
1470 RESTORE_END_STRING (l);
1471 if (!operand_ok)
1472 return;
1473 }
1474 else
1475 {
1476 if (expecting_operand)
1477 {
1478 expecting_operand_after_comma:
1479 as_bad (_("expecting operand after ','; got nothing"));
1480 return;
1481 }
1482 if (*l == ',')
1483 {
1484 as_bad (_("expecting operand before ','; got nothing"));
1485 return;
1486 }
1487 }
1488
1489 /* Now *l must be either ',' or END_OF_INSN. */
1490 if (*l == ',')
1491 {
1492 if (*++l == END_OF_INSN)
1493 {
1494 /* Just skip it, if it's \n complain. */
1495 goto expecting_operand_after_comma;
1496 }
1497 expecting_operand = 1;
1498 }
1499 }
1500 while (*l != END_OF_INSN);
1501 }
1502 }
1503
1504 /* Now we've parsed the mnemonic into a set of templates, and have the
1505 operands at hand.
1506
1507 Next, we find a template that matches the given insn,
1508 making sure the overlap of the given operands types is consistent
1509 with the template operand types. */
1510
1511 #define MATCH(overlap, given, template) \
1512 ((overlap & ~JumpAbsolute) \
1513 && ((given) & (BaseIndex|JumpAbsolute)) == ((overlap) & (BaseIndex|JumpAbsolute)))
1514
1515 /* If given types r0 and r1 are registers they must be of the same type
1516 unless the expected operand type register overlap is null.
1517 Note that Acc in a template matches every size of reg. */
1518 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
1519 ( ((g0) & Reg) == 0 || ((g1) & Reg) == 0 || \
1520 ((g0) & Reg) == ((g1) & Reg) || \
1521 ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
1522
1523 {
1524 register unsigned int overlap0, overlap1;
1525 unsigned int overlap2;
1526 unsigned int found_reverse_match;
1527 int suffix_check;
1528
1529 /* All intel opcodes have reversed operands except for "bound" and
1530 "enter". We also don't reverse intersegment "jmp" and "call"
1531 instructions with 2 immediate operands so that the immediate segment
1532 precedes the offset, as it does when in AT&T mode. "enter" and the
1533 intersegment "jmp" and "call" instructions are the only ones that
1534 have two immediate operands. */
1535 if (intel_syntax && i.operands > 1
1536 && (strcmp (mnemonic, "bound") != 0)
1537 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1538 {
1539 union i386_op temp_op;
1540 unsigned int temp_type;
1541 RELOC_ENUM temp_reloc;
1542 int xchg1 = 0;
1543 int xchg2 = 0;
1544
1545 if (i.operands == 2)
1546 {
1547 xchg1 = 0;
1548 xchg2 = 1;
1549 }
1550 else if (i.operands == 3)
1551 {
1552 xchg1 = 0;
1553 xchg2 = 2;
1554 }
1555 temp_type = i.types[xchg2];
1556 i.types[xchg2] = i.types[xchg1];
1557 i.types[xchg1] = temp_type;
1558 temp_op = i.op[xchg2];
1559 i.op[xchg2] = i.op[xchg1];
1560 i.op[xchg1] = temp_op;
1561 temp_reloc = i.reloc[xchg2];
1562 i.reloc[xchg2] = i.reloc[xchg1];
1563 i.reloc[xchg1] = temp_reloc;
1564
1565 if (i.mem_operands == 2)
1566 {
1567 const seg_entry *temp_seg;
1568 temp_seg = i.seg[0];
1569 i.seg[0] = i.seg[1];
1570 i.seg[1] = temp_seg;
1571 }
1572 }
1573
1574 if (i.imm_operands)
1575 {
1576 /* Try to ensure constant immediates are represented in the smallest
1577 opcode possible. */
1578 char guess_suffix = 0;
1579 int op;
1580
1581 if (i.suffix)
1582 guess_suffix = i.suffix;
1583 else if (i.reg_operands)
1584 {
1585 /* Figure out a suffix from the last register operand specified.
1586 We can't do this properly yet, ie. excluding InOutPortReg,
1587 but the following works for instructions with immediates.
1588 In any case, we can't set i.suffix yet. */
1589 for (op = i.operands; --op >= 0;)
1590 if (i.types[op] & Reg)
1591 {
1592 if (i.types[op] & Reg8)
1593 guess_suffix = BYTE_MNEM_SUFFIX;
1594 else if (i.types[op] & Reg16)
1595 guess_suffix = WORD_MNEM_SUFFIX;
1596 else if (i.types[op] & Reg32)
1597 guess_suffix = LONG_MNEM_SUFFIX;
1598 else if (i.types[op] & Reg64)
1599 guess_suffix = QWORD_MNEM_SUFFIX;
1600 break;
1601 }
1602 }
1603 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
1604 guess_suffix = WORD_MNEM_SUFFIX;
1605
1606 for (op = i.operands; --op >= 0;)
1607 if (i.types[op] & Imm)
1608 {
1609 switch (i.op[op].imms->X_op)
1610 {
1611 case O_constant:
1612 /* If a suffix is given, this operand may be shortened. */
1613 switch (guess_suffix)
1614 {
1615 case LONG_MNEM_SUFFIX:
1616 i.types[op] |= Imm32 | Imm64;
1617 break;
1618 case WORD_MNEM_SUFFIX:
1619 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
1620 break;
1621 case BYTE_MNEM_SUFFIX:
1622 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
1623 break;
1624 }
1625
1626 /* If this operand is at most 16 bits, convert it
1627 to a signed 16 bit number before trying to see
1628 whether it will fit in an even smaller size.
1629 This allows a 16-bit operand such as $0xffe0 to
1630 be recognised as within Imm8S range. */
1631 if ((i.types[op] & Imm16)
1632 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
1633 {
1634 i.op[op].imms->X_add_number =
1635 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
1636 }
1637 if ((i.types[op] & Imm32)
1638 && (i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1)) == 0)
1639 {
1640 i.op[op].imms->X_add_number =
1641 (i.op[op].imms->X_add_number ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1642 }
1643 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
1644 /* We must avoid matching of Imm32 templates when 64bit only immediate is available. */
1645 if (guess_suffix == QWORD_MNEM_SUFFIX)
1646 i.types[op] &= ~Imm32;
1647 break;
1648 case O_absent:
1649 case O_register:
1650 abort ();
1651 /* Symbols and expressions. */
1652 default:
1653 /* Convert symbolic operand to proper sizes for matching. */
1654 switch (guess_suffix)
1655 {
1656 case QWORD_MNEM_SUFFIX:
1657 i.types[op] = Imm64 | Imm32S;
1658 break;
1659 case LONG_MNEM_SUFFIX:
1660 i.types[op] = Imm32 | Imm64;
1661 break;
1662 case WORD_MNEM_SUFFIX:
1663 i.types[op] = Imm16 | Imm32 | Imm64;
1664 break;
1665 break;
1666 case BYTE_MNEM_SUFFIX:
1667 i.types[op] = Imm8 | Imm8S | Imm16 | Imm32S | Imm32;
1668 break;
1669 break;
1670 }
1671 break;
1672 }
1673 }
1674 }
1675
1676 if (i.disp_operands)
1677 {
1678 /* Try to use the smallest displacement type too. */
1679 int op;
1680
1681 for (op = i.operands; --op >= 0;)
1682 if ((i.types[op] & Disp)
1683 && i.op[op].disps->X_op == O_constant)
1684 {
1685 offsetT disp = i.op[op].disps->X_add_number;
1686
1687 if (i.types[op] & Disp16)
1688 {
1689 /* We know this operand is at most 16 bits, so
1690 convert to a signed 16 bit number before trying
1691 to see whether it will fit in an even smaller
1692 size. */
1693
1694 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
1695 }
1696 else if (i.types[op] & Disp32)
1697 {
1698 /* We know this operand is at most 32 bits, so convert to a
1699 signed 32 bit number before trying to see whether it will
1700 fit in an even smaller size. */
1701 disp &= (((offsetT) 2 << 31) - 1);
1702 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1703 }
1704 if (flag_code == CODE_64BIT)
1705 {
1706 if (fits_in_signed_long (disp))
1707 i.types[op] |= Disp32S;
1708 if (fits_in_unsigned_long (disp))
1709 i.types[op] |= Disp32;
1710 }
1711 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
1712 && fits_in_signed_byte (disp))
1713 i.types[op] |= Disp8;
1714 }
1715 }
1716
1717 overlap0 = 0;
1718 overlap1 = 0;
1719 overlap2 = 0;
1720 found_reverse_match = 0;
1721 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
1722 ? No_bSuf
1723 : (i.suffix == WORD_MNEM_SUFFIX
1724 ? No_wSuf
1725 : (i.suffix == SHORT_MNEM_SUFFIX
1726 ? No_sSuf
1727 : (i.suffix == LONG_MNEM_SUFFIX
1728 ? No_lSuf
1729 : (i.suffix == QWORD_MNEM_SUFFIX
1730 ? No_qSuf
1731 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX ? No_xSuf : 0))))));
1732
1733 for (t = current_templates->start;
1734 t < current_templates->end;
1735 t++)
1736 {
1737 /* Must have right number of operands. */
1738 if (i.operands != t->operands)
1739 continue;
1740
1741 /* Check the suffix, except for some instructions in intel mode. */
1742 if ((t->opcode_modifier & suffix_check)
1743 && !(intel_syntax
1744 && (t->opcode_modifier & IgnoreSize))
1745 && !(intel_syntax
1746 && t->base_opcode == 0xd9
1747 && (t->extension_opcode == 5 /* 0xd9,5 "fldcw" */
1748 || t->extension_opcode == 7))) /* 0xd9,7 "f{n}stcw" */
1749 continue;
1750
1751 /* Do not verify operands when there are none. */
1752 else if (!t->operands)
1753 {
1754 if (t->cpu_flags & ~cpu_arch_flags)
1755 continue;
1756 /* We've found a match; break out of loop. */
1757 break;
1758 }
1759
1760 overlap0 = i.types[0] & t->operand_types[0];
1761 switch (t->operands)
1762 {
1763 case 1:
1764 if (!MATCH (overlap0, i.types[0], t->operand_types[0]))
1765 continue;
1766 break;
1767 case 2:
1768 case 3:
1769 overlap1 = i.types[1] & t->operand_types[1];
1770 if (!MATCH (overlap0, i.types[0], t->operand_types[0])
1771 || !MATCH (overlap1, i.types[1], t->operand_types[1])
1772 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
1773 t->operand_types[0],
1774 overlap1, i.types[1],
1775 t->operand_types[1]))
1776 {
1777 /* Check if other direction is valid ... */
1778 if ((t->opcode_modifier & (D|FloatD)) == 0)
1779 continue;
1780
1781 /* Try reversing direction of operands. */
1782 overlap0 = i.types[0] & t->operand_types[1];
1783 overlap1 = i.types[1] & t->operand_types[0];
1784 if (!MATCH (overlap0, i.types[0], t->operand_types[1])
1785 || !MATCH (overlap1, i.types[1], t->operand_types[0])
1786 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
1787 t->operand_types[1],
1788 overlap1, i.types[1],
1789 t->operand_types[0]))
1790 {
1791 /* Does not match either direction. */
1792 continue;
1793 }
1794 /* found_reverse_match holds which of D or FloatDR
1795 we've found. */
1796 found_reverse_match = t->opcode_modifier & (D|FloatDR);
1797 }
1798 /* Found a forward 2 operand match here. */
1799 else if (t->operands == 3)
1800 {
1801 /* Here we make use of the fact that there are no
1802 reverse match 3 operand instructions, and all 3
1803 operand instructions only need to be checked for
1804 register consistency between operands 2 and 3. */
1805 overlap2 = i.types[2] & t->operand_types[2];
1806 if (!MATCH (overlap2, i.types[2], t->operand_types[2])
1807 || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1],
1808 t->operand_types[1],
1809 overlap2, i.types[2],
1810 t->operand_types[2]))
1811
1812 continue;
1813 }
1814 /* Found either forward/reverse 2 or 3 operand match here:
1815 slip through to break. */
1816 }
1817 if (t->cpu_flags & ~cpu_arch_flags)
1818 {
1819 found_reverse_match = 0;
1820 continue;
1821 }
1822 /* We've found a match; break out of loop. */
1823 break;
1824 }
1825 if (t == current_templates->end)
1826 {
1827 /* We found no match. */
1828 as_bad (_("suffix or operands invalid for `%s'"),
1829 current_templates->start->name);
1830 return;
1831 }
1832
1833 if (!quiet_warnings)
1834 {
1835 if (!intel_syntax
1836 && ((i.types[0] & JumpAbsolute)
1837 != (t->operand_types[0] & JumpAbsolute)))
1838 {
1839 as_warn (_("indirect %s without `*'"), t->name);
1840 }
1841
1842 if ((t->opcode_modifier & (IsPrefix|IgnoreSize))
1843 == (IsPrefix|IgnoreSize))
1844 {
1845 /* Warn them that a data or address size prefix doesn't
1846 affect assembly of the next line of code. */
1847 as_warn (_("stand-alone `%s' prefix"), t->name);
1848 }
1849 }
1850
1851 /* Copy the template we found. */
1852 i.tm = *t;
1853 if (found_reverse_match)
1854 {
1855 /* If we found a reverse match we must alter the opcode
1856 direction bit. found_reverse_match holds bits to change
1857 (different for int & float insns). */
1858
1859 i.tm.base_opcode ^= found_reverse_match;
1860
1861 i.tm.operand_types[0] = t->operand_types[1];
1862 i.tm.operand_types[1] = t->operand_types[0];
1863 }
1864
1865 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1866 if (SYSV386_COMPAT
1867 && intel_syntax
1868 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1869 i.tm.base_opcode ^= FloatR;
1870
1871 if (i.tm.opcode_modifier & FWait)
1872 if (! add_prefix (FWAIT_OPCODE))
1873 return;
1874
1875 /* Check string instruction segment overrides. */
1876 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1877 {
1878 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
1879 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
1880 {
1881 if (i.seg[0] != NULL && i.seg[0] != &es)
1882 {
1883 as_bad (_("`%s' operand %d must use `%%es' segment"),
1884 i.tm.name,
1885 mem_op + 1);
1886 return;
1887 }
1888 /* There's only ever one segment override allowed per instruction.
1889 This instruction possibly has a legal segment override on the
1890 second operand, so copy the segment to where non-string
1891 instructions store it, allowing common code. */
1892 i.seg[0] = i.seg[1];
1893 }
1894 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
1895 {
1896 if (i.seg[1] != NULL && i.seg[1] != &es)
1897 {
1898 as_bad (_("`%s' operand %d must use `%%es' segment"),
1899 i.tm.name,
1900 mem_op + 2);
1901 return;
1902 }
1903 }
1904 }
1905
1906 if (i.reg_operands && flag_code < CODE_64BIT)
1907 {
1908 int op;
1909 for (op = i.operands; --op >= 0;)
1910 if ((i.types[op] & Reg)
1911 && (i.op[op].regs->reg_flags & (RegRex64|RegRex)))
1912 {
1913 as_bad (_("Extended register `%%%s' available only in 64bit mode."),
1914 i.op[op].regs->reg_name);
1915 return;
1916 }
1917 }
1918
1919 /* If matched instruction specifies an explicit instruction mnemonic
1920 suffix, use it. */
1921 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
1922 {
1923 if (i.tm.opcode_modifier & Size16)
1924 i.suffix = WORD_MNEM_SUFFIX;
1925 else if (i.tm.opcode_modifier & Size64)
1926 i.suffix = QWORD_MNEM_SUFFIX;
1927 else
1928 i.suffix = LONG_MNEM_SUFFIX;
1929 }
1930 else if (i.reg_operands)
1931 {
1932 /* If there's no instruction mnemonic suffix we try to invent one
1933 based on register operands. */
1934 if (!i.suffix)
1935 {
1936 /* We take i.suffix from the last register operand specified,
1937 Destination register type is more significant than source
1938 register type. */
1939 int op;
1940 for (op = i.operands; --op >= 0;)
1941 if ((i.types[op] & Reg)
1942 && !(i.tm.operand_types[op] & InOutPortReg))
1943 {
1944 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
1945 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
1946 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
1947 LONG_MNEM_SUFFIX);
1948 break;
1949 }
1950 }
1951 else if (i.suffix == BYTE_MNEM_SUFFIX)
1952 {
1953 int op;
1954 for (op = i.operands; --op >= 0;)
1955 {
1956 /* If this is an eight bit register, it's OK. If it's
1957 the 16 or 32 bit version of an eight bit register,
1958 we will just use the low portion, and that's OK too. */
1959 if (i.types[op] & Reg8)
1960 continue;
1961
1962 /* movzx and movsx should not generate this warning. */
1963 if (intel_syntax
1964 && (i.tm.base_opcode == 0xfb7
1965 || i.tm.base_opcode == 0xfb6
1966 || i.tm.base_opcode == 0x63
1967 || i.tm.base_opcode == 0xfbe
1968 || i.tm.base_opcode == 0xfbf))
1969 continue;
1970
1971 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4
1972 #if 0
1973 /* Check that the template allows eight bit regs
1974 This kills insns such as `orb $1,%edx', which
1975 maybe should be allowed. */
1976 && (i.tm.operand_types[op] & (Reg8|InOutPortReg))
1977 #endif
1978 )
1979 {
1980 /* Prohibit these changes in the 64bit mode, since
1981 the lowering is more complicated. */
1982 if (flag_code == CODE_64BIT
1983 && (i.tm.operand_types[op] & InOutPortReg) == 0)
1984 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
1985 i.op[op].regs->reg_name,
1986 i.suffix);
1987 #if REGISTER_WARNINGS
1988 if (!quiet_warnings
1989 && (i.tm.operand_types[op] & InOutPortReg) == 0)
1990 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
1991 (i.op[op].regs
1992 + (i.types[op] & Reg16
1993 ? REGNAM_AL - REGNAM_AX
1994 : REGNAM_AL - REGNAM_EAX))->reg_name,
1995 i.op[op].regs->reg_name,
1996 i.suffix);
1997 #endif
1998 continue;
1999 }
2000 /* Any other register is bad. */
2001 if (i.types[op] & (Reg | RegMMX | RegXMM
2002 | SReg2 | SReg3
2003 | Control | Debug | Test
2004 | FloatReg | FloatAcc))
2005 {
2006 as_bad (_("`%%%s' not allowed with `%s%c'"),
2007 i.op[op].regs->reg_name,
2008 i.tm.name,
2009 i.suffix);
2010 return;
2011 }
2012 }
2013 }
2014 else if (i.suffix == LONG_MNEM_SUFFIX)
2015 {
2016 int op;
2017
2018 for (op = i.operands; --op >= 0;)
2019 /* Reject eight bit registers, except where the template
2020 requires them. (eg. movzb) */
2021 if ((i.types[op] & Reg8) != 0
2022 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2023 {
2024 as_bad (_("`%%%s' not allowed with `%s%c'"),
2025 i.op[op].regs->reg_name,
2026 i.tm.name,
2027 i.suffix);
2028 return;
2029 }
2030 /* Warn if the e prefix on a general reg is missing. */
2031 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2032 && (i.types[op] & Reg16) != 0
2033 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
2034 {
2035 /* Prohibit these changes in the 64bit mode, since
2036 the lowering is more complicated. */
2037 if (flag_code == CODE_64BIT)
2038 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2039 i.op[op].regs->reg_name,
2040 i.suffix);
2041 #if REGISTER_WARNINGS
2042 else
2043 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2044 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
2045 i.op[op].regs->reg_name,
2046 i.suffix);
2047 #endif
2048 }
2049 /* Warn if the r prefix on a general reg is missing. */
2050 else if ((i.types[op] & Reg64) != 0
2051 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
2052 {
2053 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2054 i.op[op].regs->reg_name,
2055 i.suffix);
2056 }
2057 }
2058 else if (i.suffix == QWORD_MNEM_SUFFIX)
2059 {
2060 int op;
2061
2062 for (op = i.operands; --op >= 0; )
2063 /* Reject eight bit registers, except where the template
2064 requires them. (eg. movzb) */
2065 if ((i.types[op] & Reg8) != 0
2066 && (i.tm.operand_types[op] & (Reg16|Reg32|Acc)) != 0)
2067 {
2068 as_bad (_("`%%%s' not allowed with `%s%c'"),
2069 i.op[op].regs->reg_name,
2070 i.tm.name,
2071 i.suffix);
2072 return;
2073 }
2074 /* Warn if the e prefix on a general reg is missing. */
2075 else if (((i.types[op] & Reg16) != 0
2076 || (i.types[op] & Reg32) != 0)
2077 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
2078 {
2079 /* Prohibit these changes in the 64bit mode, since
2080 the lowering is more complicated. */
2081 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2082 i.op[op].regs->reg_name,
2083 i.suffix);
2084 }
2085 }
2086 else if (i.suffix == WORD_MNEM_SUFFIX)
2087 {
2088 int op;
2089 for (op = i.operands; --op >= 0;)
2090 /* Reject eight bit registers, except where the template
2091 requires them. (eg. movzb) */
2092 if ((i.types[op] & Reg8) != 0
2093 && (i.tm.operand_types[op] & (Reg16|Reg32|Acc)) != 0)
2094 {
2095 as_bad (_("`%%%s' not allowed with `%s%c'"),
2096 i.op[op].regs->reg_name,
2097 i.tm.name,
2098 i.suffix);
2099 return;
2100 }
2101 /* Warn if the e prefix on a general reg is present. */
2102 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2103 && (i.types[op] & Reg32) != 0
2104 && (i.tm.operand_types[op] & (Reg16|Acc)) != 0)
2105 {
2106 /* Prohibit these changes in the 64bit mode, since
2107 the lowering is more complicated. */
2108 if (flag_code == CODE_64BIT)
2109 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2110 i.op[op].regs->reg_name,
2111 i.suffix);
2112 else
2113 #if REGISTER_WARNINGS
2114 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2115 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
2116 i.op[op].regs->reg_name,
2117 i.suffix);
2118 #endif
2119 }
2120 }
2121 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2122 /* Do nothing if the instruction is going to ignore the prefix. */
2123 ;
2124 else
2125 abort ();
2126 }
2127 else if ((i.tm.opcode_modifier & DefaultSize) && !i.suffix)
2128 {
2129 i.suffix = stackop_size;
2130 }
2131 /* Make still unresolved immediate matches conform to size of immediate
2132 given in i.suffix. Note: overlap2 cannot be an immediate! */
2133 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S))
2134 && overlap0 != Imm8 && overlap0 != Imm8S
2135 && overlap0 != Imm16 && overlap0 != Imm32S
2136 && overlap0 != Imm32 && overlap0 != Imm64)
2137 {
2138 if (i.suffix)
2139 {
2140 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
2141 (i.suffix == WORD_MNEM_SUFFIX ? Imm16 :
2142 (i.suffix == QWORD_MNEM_SUFFIX ? Imm64 | Imm32S : Imm32)));
2143 }
2144 else if (overlap0 == (Imm16 | Imm32S | Imm32)
2145 || overlap0 == (Imm16 | Imm32)
2146 || overlap0 == (Imm16 | Imm32S))
2147 {
2148 overlap0 =
2149 ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32S;
2150 }
2151 if (overlap0 != Imm8 && overlap0 != Imm8S
2152 && overlap0 != Imm16 && overlap0 != Imm32S
2153 && overlap0 != Imm32 && overlap0 != Imm64)
2154 {
2155 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
2156 return;
2157 }
2158 }
2159 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32))
2160 && overlap1 != Imm8 && overlap1 != Imm8S
2161 && overlap1 != Imm16 && overlap1 != Imm32S
2162 && overlap1 != Imm32 && overlap1 != Imm64)
2163 {
2164 if (i.suffix)
2165 {
2166 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
2167 (i.suffix == WORD_MNEM_SUFFIX ? Imm16 :
2168 (i.suffix == QWORD_MNEM_SUFFIX ? Imm64 | Imm32S : Imm32)));
2169 }
2170 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
2171 || overlap1 == (Imm16 | Imm32)
2172 || overlap1 == (Imm16 | Imm32S))
2173 {
2174 overlap1 =
2175 ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32S;
2176 }
2177 if (overlap1 != Imm8 && overlap1 != Imm8S
2178 && overlap1 != Imm16 && overlap1 != Imm32S
2179 && overlap1 != Imm32 && overlap1 != Imm64)
2180 {
2181 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
2182 return;
2183 }
2184 }
2185 assert ((overlap2 & Imm) == 0);
2186
2187 i.types[0] = overlap0;
2188 if (overlap0 & ImplicitRegister)
2189 i.reg_operands--;
2190 if (overlap0 & Imm1)
2191 i.imm_operands = 0; /* kludge for shift insns. */
2192
2193 i.types[1] = overlap1;
2194 if (overlap1 & ImplicitRegister)
2195 i.reg_operands--;
2196
2197 i.types[2] = overlap2;
2198 if (overlap2 & ImplicitRegister)
2199 i.reg_operands--;
2200
2201 /* Finalize opcode. First, we change the opcode based on the operand
2202 size given by i.suffix: We need not change things for byte insns. */
2203
2204 if (!i.suffix && (i.tm.opcode_modifier & W))
2205 {
2206 as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
2207 return;
2208 }
2209
2210 /* For movzx and movsx, need to check the register type. */
2211 if (intel_syntax
2212 && (i.tm.base_opcode == 0xfb6 || i.tm.base_opcode == 0xfbe))
2213 if (i.suffix && i.suffix == BYTE_MNEM_SUFFIX)
2214 {
2215 unsigned int prefix = DATA_PREFIX_OPCODE;
2216
2217 if ((i.op[1].regs->reg_type & Reg16) != 0)
2218 if (!add_prefix (prefix))
2219 return;
2220 }
2221
2222 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2223 {
2224 /* It's not a byte, select word/dword operation. */
2225 if (i.tm.opcode_modifier & W)
2226 {
2227 if (i.tm.opcode_modifier & ShortForm)
2228 i.tm.base_opcode |= 8;
2229 else
2230 i.tm.base_opcode |= 1;
2231 }
2232 /* Now select between word & dword operations via the operand
2233 size prefix, except for instructions that will ignore this
2234 prefix anyway. */
2235 if (i.suffix != QWORD_MNEM_SUFFIX
2236 && (i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2237 && !(i.tm.opcode_modifier & IgnoreSize))
2238 {
2239 unsigned int prefix = DATA_PREFIX_OPCODE;
2240 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2241 prefix = ADDR_PREFIX_OPCODE;
2242
2243 if (! add_prefix (prefix))
2244 return;
2245 }
2246
2247 /* Set mode64 for an operand. */
2248 if (i.suffix == QWORD_MNEM_SUFFIX
2249 && !(i.tm.opcode_modifier & NoRex64))
2250 {
2251 i.rex.mode64 = 1;
2252 if (flag_code < CODE_64BIT)
2253 {
2254 as_bad (_("64bit operations available only in 64bit modes."));
2255 return;
2256 }
2257 }
2258
2259 /* Size floating point instruction. */
2260 if (i.suffix == LONG_MNEM_SUFFIX)
2261 {
2262 if (i.tm.opcode_modifier & FloatMF)
2263 i.tm.base_opcode ^= 4;
2264 }
2265 }
2266
2267 if (i.tm.opcode_modifier & ImmExt)
2268 {
2269 /* These AMD 3DNow! and Intel Katmai New Instructions have an
2270 opcode suffix which is coded in the same place as an 8-bit
2271 immediate field would be. Here we fake an 8-bit immediate
2272 operand from the opcode suffix stored in tm.extension_opcode. */
2273
2274 expressionS *exp;
2275
2276 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
2277
2278 exp = &im_expressions[i.imm_operands++];
2279 i.op[i.operands].imms = exp;
2280 i.types[i.operands++] = Imm8;
2281 exp->X_op = O_constant;
2282 exp->X_add_number = i.tm.extension_opcode;
2283 i.tm.extension_opcode = None;
2284 }
2285
2286 /* For insns with operands there are more diddles to do to the opcode. */
2287 if (i.operands)
2288 {
2289 /* Default segment register this instruction will use
2290 for memory accesses. 0 means unknown.
2291 This is only for optimizing out unnecessary segment overrides. */
2292 const seg_entry *default_seg = 0;
2293
2294 /* The imul $imm, %reg instruction is converted into
2295 imul $imm, %reg, %reg, and the clr %reg instruction
2296 is converted into xor %reg, %reg. */
2297 if (i.tm.opcode_modifier & regKludge)
2298 {
2299 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
2300 /* Pretend we saw the extra register operand. */
2301 assert (i.op[first_reg_op + 1].regs == 0);
2302 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
2303 i.types[first_reg_op + 1] = i.types[first_reg_op];
2304 i.reg_operands = 2;
2305 }
2306
2307 if (i.tm.opcode_modifier & ShortForm)
2308 {
2309 /* The register or float register operand is in operand 0 or 1. */
2310 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
2311 /* Register goes in low 3 bits of opcode. */
2312 i.tm.base_opcode |= i.op[op].regs->reg_num;
2313 if (i.op[op].regs->reg_flags & RegRex)
2314 i.rex.extZ = 1;
2315 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2316 {
2317 /* Warn about some common errors, but press on regardless.
2318 The first case can be generated by gcc (<= 2.8.1). */
2319 if (i.operands == 2)
2320 {
2321 /* Reversed arguments on faddp, fsubp, etc. */
2322 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
2323 i.op[1].regs->reg_name,
2324 i.op[0].regs->reg_name);
2325 }
2326 else
2327 {
2328 /* Extraneous `l' suffix on fp insn. */
2329 as_warn (_("translating to `%s %%%s'"), i.tm.name,
2330 i.op[0].regs->reg_name);
2331 }
2332 }
2333 }
2334 else if (i.tm.opcode_modifier & Modrm)
2335 {
2336 /* The opcode is completed (modulo i.tm.extension_opcode which
2337 must be put into the modrm byte).
2338 Now, we make the modrm & index base bytes based on all the
2339 info we've collected. */
2340
2341 /* i.reg_operands MUST be the number of real register operands;
2342 implicit registers do not count. */
2343 if (i.reg_operands == 2)
2344 {
2345 unsigned int source, dest;
2346 source = ((i.types[0]
2347 & (Reg | RegMMX | RegXMM
2348 | SReg2 | SReg3
2349 | Control | Debug | Test))
2350 ? 0 : 1);
2351 dest = source + 1;
2352
2353 i.rm.mode = 3;
2354 /* One of the register operands will be encoded in the
2355 i.tm.reg field, the other in the combined i.tm.mode
2356 and i.tm.regmem fields. If no form of this
2357 instruction supports a memory destination operand,
2358 then we assume the source operand may sometimes be
2359 a memory operand and so we need to store the
2360 destination in the i.rm.reg field. */
2361 if ((i.tm.operand_types[dest] & AnyMem) == 0)
2362 {
2363 i.rm.reg = i.op[dest].regs->reg_num;
2364 i.rm.regmem = i.op[source].regs->reg_num;
2365 if (i.op[dest].regs->reg_flags & RegRex)
2366 i.rex.extX = 1;
2367 if (i.op[source].regs->reg_flags & RegRex)
2368 i.rex.extZ = 1;
2369 }
2370 else
2371 {
2372 i.rm.reg = i.op[source].regs->reg_num;
2373 i.rm.regmem = i.op[dest].regs->reg_num;
2374 if (i.op[dest].regs->reg_flags & RegRex)
2375 i.rex.extZ = 1;
2376 if (i.op[source].regs->reg_flags & RegRex)
2377 i.rex.extX = 1;
2378 }
2379 }
2380 else
2381 { /* If it's not 2 reg operands... */
2382 if (i.mem_operands)
2383 {
2384 unsigned int fake_zero_displacement = 0;
2385 unsigned int op = ((i.types[0] & AnyMem)
2386 ? 0
2387 : (i.types[1] & AnyMem) ? 1 : 2);
2388
2389 default_seg = &ds;
2390
2391 if (! i.base_reg)
2392 {
2393 i.rm.mode = 0;
2394 if (! i.disp_operands)
2395 fake_zero_displacement = 1;
2396 if (! i.index_reg)
2397 {
2398 /* Operand is just <disp> */
2399 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
2400 {
2401 i.rm.regmem = NO_BASE_REGISTER_16;
2402 i.types[op] &= ~Disp;
2403 i.types[op] |= Disp16;
2404 }
2405 else if (flag_code != CODE_64BIT)
2406 {
2407 i.rm.regmem = NO_BASE_REGISTER;
2408 i.types[op] &= ~Disp;
2409 i.types[op] |= Disp32;
2410 }
2411 else
2412 {
2413 /* 64bit mode overwrites the 32bit
2414 absolute addressing by RIP relative
2415 addressing and absolute addressing
2416 is encoded by one of the redundant
2417 SIB forms. */
2418
2419 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2420 i.sib.base = NO_BASE_REGISTER;
2421 i.sib.index = NO_INDEX_REGISTER;
2422 i.types[op] &= ~Disp;
2423 i.types[op] |= Disp32S;
2424 }
2425 }
2426 else /* ! i.base_reg && i.index_reg */
2427 {
2428 i.sib.index = i.index_reg->reg_num;
2429 i.sib.base = NO_BASE_REGISTER;
2430 i.sib.scale = i.log2_scale_factor;
2431 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2432 i.types[op] &= ~Disp;
2433 if (flag_code != CODE_64BIT)
2434 i.types[op] |= Disp32; /* Must be 32 bit */
2435 else
2436 i.types[op] |= Disp32S;
2437 if (i.index_reg->reg_flags & RegRex)
2438 i.rex.extY = 1;
2439 }
2440 }
2441 /* RIP addressing for 64bit mode. */
2442 else if (i.base_reg->reg_type == BaseIndex)
2443 {
2444 i.rm.regmem = NO_BASE_REGISTER;
2445 i.types[op] &= ~Disp;
2446 i.types[op] |= Disp32S;
2447 i.flags[op] = Operand_PCrel;
2448 }
2449 else if (i.base_reg->reg_type & Reg16)
2450 {
2451 switch (i.base_reg->reg_num)
2452 {
2453 case 3: /* (%bx) */
2454 if (! i.index_reg)
2455 i.rm.regmem = 7;
2456 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
2457 i.rm.regmem = i.index_reg->reg_num - 6;
2458 break;
2459 case 5: /* (%bp) */
2460 default_seg = &ss;
2461 if (! i.index_reg)
2462 {
2463 i.rm.regmem = 6;
2464 if ((i.types[op] & Disp) == 0)
2465 {
2466 /* fake (%bp) into 0(%bp) */
2467 i.types[op] |= Disp8;
2468 fake_zero_displacement = 1;
2469 }
2470 }
2471 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
2472 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
2473 break;
2474 default: /* (%si) -> 4 or (%di) -> 5 */
2475 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
2476 }
2477 i.rm.mode = mode_from_disp_size (i.types[op]);
2478 }
2479 else /* i.base_reg and 32/64 bit mode */
2480 {
2481 if (flag_code == CODE_64BIT
2482 && (i.types[op] & Disp))
2483 {
2484 if (i.types[op] & Disp8)
2485 i.types[op] = Disp8 | Disp32S;
2486 else
2487 i.types[op] = Disp32S;
2488 }
2489 i.rm.regmem = i.base_reg->reg_num;
2490 if (i.base_reg->reg_flags & RegRex)
2491 i.rex.extZ = 1;
2492 i.sib.base = i.base_reg->reg_num;
2493 /* x86-64 ignores REX prefix bit here to avoid
2494 decoder complications. */
2495 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
2496 {
2497 default_seg = &ss;
2498 if (i.disp_operands == 0)
2499 {
2500 fake_zero_displacement = 1;
2501 i.types[op] |= Disp8;
2502 }
2503 }
2504 else if (i.base_reg->reg_num == ESP_REG_NUM)
2505 {
2506 default_seg = &ss;
2507 }
2508 i.sib.scale = i.log2_scale_factor;
2509 if (! i.index_reg)
2510 {
2511 /* <disp>(%esp) becomes two byte modrm
2512 with no index register. We've already
2513 stored the code for esp in i.rm.regmem
2514 ie. ESCAPE_TO_TWO_BYTE_ADDRESSING. Any
2515 base register besides %esp will not use
2516 the extra modrm byte. */
2517 i.sib.index = NO_INDEX_REGISTER;
2518 #if ! SCALE1_WHEN_NO_INDEX
2519 /* Another case where we force the second
2520 modrm byte. */
2521 if (i.log2_scale_factor)
2522 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2523 #endif
2524 }
2525 else
2526 {
2527 i.sib.index = i.index_reg->reg_num;
2528 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2529 if (i.index_reg->reg_flags & RegRex)
2530 i.rex.extY = 1;
2531 }
2532 i.rm.mode = mode_from_disp_size (i.types[op]);
2533 }
2534
2535 if (fake_zero_displacement)
2536 {
2537 /* Fakes a zero displacement assuming that i.types[op]
2538 holds the correct displacement size. */
2539 expressionS *exp;
2540
2541 assert (i.op[op].disps == 0);
2542 exp = &disp_expressions[i.disp_operands++];
2543 i.op[op].disps = exp;
2544 exp->X_op = O_constant;
2545 exp->X_add_number = 0;
2546 exp->X_add_symbol = (symbolS *) 0;
2547 exp->X_op_symbol = (symbolS *) 0;
2548 }
2549 }
2550
2551 /* Fill in i.rm.reg or i.rm.regmem field with register
2552 operand (if any) based on i.tm.extension_opcode.
2553 Again, we must be careful to make sure that
2554 segment/control/debug/test/MMX registers are coded
2555 into the i.rm.reg field. */
2556 if (i.reg_operands)
2557 {
2558 unsigned int op =
2559 ((i.types[0]
2560 & (Reg | RegMMX | RegXMM
2561 | SReg2 | SReg3
2562 | Control | Debug | Test))
2563 ? 0
2564 : ((i.types[1]
2565 & (Reg | RegMMX | RegXMM
2566 | SReg2 | SReg3
2567 | Control | Debug | Test))
2568 ? 1
2569 : 2));
2570 /* If there is an extension opcode to put here, the
2571 register number must be put into the regmem field. */
2572 if (i.tm.extension_opcode != None)
2573 {
2574 i.rm.regmem = i.op[op].regs->reg_num;
2575 if (i.op[op].regs->reg_flags & RegRex)
2576 i.rex.extZ = 1;
2577 }
2578 else
2579 {
2580 i.rm.reg = i.op[op].regs->reg_num;
2581 if (i.op[op].regs->reg_flags & RegRex)
2582 i.rex.extX = 1;
2583 }
2584
2585 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2
2586 we must set it to 3 to indicate this is a register
2587 operand in the regmem field. */
2588 if (!i.mem_operands)
2589 i.rm.mode = 3;
2590 }
2591
2592 /* Fill in i.rm.reg field with extension opcode (if any). */
2593 if (i.tm.extension_opcode != None)
2594 i.rm.reg = i.tm.extension_opcode;
2595 }
2596 }
2597 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
2598 {
2599 if (i.tm.base_opcode == POP_SEG_SHORT
2600 && i.op[0].regs->reg_num == 1)
2601 {
2602 as_bad (_("you can't `pop %%cs'"));
2603 return;
2604 }
2605 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
2606 if (i.op[0].regs->reg_flags & RegRex)
2607 i.rex.extZ = 1;
2608 }
2609 else if ((i.tm.base_opcode & ~(D|W)) == MOV_AX_DISP32)
2610 {
2611 default_seg = &ds;
2612 }
2613 else if ((i.tm.opcode_modifier & IsString) != 0)
2614 {
2615 /* For the string instructions that allow a segment override
2616 on one of their operands, the default segment is ds. */
2617 default_seg = &ds;
2618 }
2619
2620 /* If a segment was explicitly specified,
2621 and the specified segment is not the default,
2622 use an opcode prefix to select it.
2623 If we never figured out what the default segment is,
2624 then default_seg will be zero at this point,
2625 and the specified segment prefix will always be used. */
2626 if ((i.seg[0]) && (i.seg[0] != default_seg))
2627 {
2628 if (! add_prefix (i.seg[0]->seg_prefix))
2629 return;
2630 }
2631 }
2632 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2633 {
2634 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
2635 as_warn (_("translating to `%sp'"), i.tm.name);
2636 }
2637 }
2638
2639 /* Handle conversion of 'int $3' --> special int3 insn. */
2640 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
2641 {
2642 i.tm.base_opcode = INT3_OPCODE;
2643 i.imm_operands = 0;
2644 }
2645
2646 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
2647 && i.op[0].disps->X_op == O_constant)
2648 {
2649 /* Convert "jmp constant" (and "call constant") to a jump (call) to
2650 the absolute address given by the constant. Since ix86 jumps and
2651 calls are pc relative, we need to generate a reloc. */
2652 i.op[0].disps->X_add_symbol = &abs_symbol;
2653 i.op[0].disps->X_op = O_symbol;
2654 }
2655
2656 if (i.tm.opcode_modifier & Rex64)
2657 i.rex.mode64 = 1;
2658
2659 /* For 8bit registers we would need an empty rex prefix.
2660 Also in the case instruction is already having prefix,
2661 we need to convert old registers to new ones. */
2662
2663 if (((i.types[0] & Reg8) && (i.op[0].regs->reg_flags & RegRex64))
2664 || ((i.types[1] & Reg8) && (i.op[1].regs->reg_flags & RegRex64))
2665 || ((i.rex.mode64 || i.rex.extX || i.rex.extY || i.rex.extZ || i.rex.empty)
2666 && ((i.types[0] & Reg8) || (i.types[1] & Reg8))))
2667 {
2668 int x;
2669 i.rex.empty = 1;
2670 for (x = 0; x < 2; x++)
2671 {
2672 /* Look for 8bit operand that does use old registers. */
2673 if (i.types[x] & Reg8
2674 && !(i.op[x].regs->reg_flags & RegRex64))
2675 {
2676 /* In case it is "hi" register, give up. */
2677 if (i.op[x].regs->reg_num > 3)
2678 as_bad (_("Can't encode registers '%%%s' in the instruction requiring REX prefix.\n"),
2679 i.op[x].regs->reg_name);
2680
2681 /* Otherwise it is equivalent to the extended register.
2682 Since the encoding don't change this is merely cosmetical
2683 cleanup for debug output. */
2684
2685 i.op[x].regs = i.op[x].regs + 8;
2686 }
2687 }
2688 }
2689
2690 if (i.rex.mode64 || i.rex.extX || i.rex.extY || i.rex.extZ || i.rex.empty)
2691 add_prefix (0x40
2692 | (i.rex.mode64 ? 8 : 0)
2693 | (i.rex.extX ? 4 : 0)
2694 | (i.rex.extY ? 2 : 0)
2695 | (i.rex.extZ ? 1 : 0));
2696
2697 /* We are ready to output the insn. */
2698 {
2699 register char *p;
2700
2701 /* Tie dwarf2 debug info to the address at the start of the insn.
2702 We can't do this after the insn has been output as the current
2703 frag may have been closed off. eg. by frag_var. */
2704 dwarf2_emit_insn (0);
2705
2706 /* Output jumps. */
2707 if (i.tm.opcode_modifier & Jump)
2708 {
2709 int code16;
2710 int prefix;
2711
2712 code16 = 0;
2713 if (flag_code == CODE_16BIT)
2714 code16 = CODE16;
2715
2716 prefix = 0;
2717 if (i.prefix[DATA_PREFIX])
2718 {
2719 prefix = 1;
2720 i.prefixes -= 1;
2721 code16 ^= CODE16;
2722 }
2723 /* Pentium4 branch hints. */
2724 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
2725 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
2726 {
2727 prefix++;
2728 i.prefixes--;
2729 }
2730 if (i.prefix[REX_PREFIX])
2731 {
2732 prefix++;
2733 i.prefixes--;
2734 }
2735
2736 if (i.prefixes != 0 && !intel_syntax)
2737 as_warn (_("skipping prefixes on this instruction"));
2738
2739 /* It's always a symbol; End frag & setup for relax.
2740 Make sure there is enough room in this frag for the largest
2741 instruction we may generate in md_convert_frag. This is 2
2742 bytes for the opcode and room for the prefix and largest
2743 displacement. */
2744 frag_grow (prefix + 2 + 4);
2745 /* Prefix and 1 opcode byte go in fr_fix. */
2746 p = frag_more (prefix + 1);
2747 if (i.prefix[DATA_PREFIX])
2748 *p++ = DATA_PREFIX_OPCODE;
2749 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
2750 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
2751 *p++ = i.prefix[SEG_PREFIX];
2752 if (i.prefix[REX_PREFIX])
2753 *p++ = i.prefix[REX_PREFIX];
2754 *p = i.tm.base_opcode;
2755 /* 1 possible extra opcode + displacement go in var part.
2756 Pass reloc in fr_var. */
2757 frag_var (rs_machine_dependent,
2758 1 + 4,
2759 i.reloc[0],
2760 ((unsigned char) *p == JUMP_PC_RELATIVE
2761 ? ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL) | code16
2762 : ((cpu_arch_flags & Cpu386) != 0
2763 ? ENCODE_RELAX_STATE (COND_JUMP, SMALL) | code16
2764 : ENCODE_RELAX_STATE (COND_JUMP86, SMALL) | code16)),
2765 i.op[0].disps->X_add_symbol,
2766 i.op[0].disps->X_add_number,
2767 p);
2768 }
2769 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
2770 {
2771 int size;
2772
2773 if (i.tm.opcode_modifier & JumpByte)
2774 {
2775 /* This is a loop or jecxz type instruction. */
2776 size = 1;
2777 if (i.prefix[ADDR_PREFIX])
2778 {
2779 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
2780 i.prefixes -= 1;
2781 }
2782 /* Pentium4 branch hints. */
2783 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
2784 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
2785 {
2786 FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
2787 i.prefixes--;
2788 }
2789 }
2790 else
2791 {
2792 int code16;
2793
2794 code16 = 0;
2795 if (flag_code == CODE_16BIT)
2796 code16 = CODE16;
2797
2798 if (i.prefix[DATA_PREFIX])
2799 {
2800 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
2801 i.prefixes -= 1;
2802 code16 ^= CODE16;
2803 }
2804
2805 size = 4;
2806 if (code16)
2807 size = 2;
2808 }
2809
2810 if (i.prefix[REX_PREFIX])
2811 {
2812 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
2813 i.prefixes -= 1;
2814 }
2815
2816 if (i.prefixes != 0 && !intel_syntax)
2817 as_warn (_("skipping prefixes on this instruction"));
2818
2819 p = frag_more (1 + size);
2820 *p++ = i.tm.base_opcode;
2821
2822 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2823 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
2824 }
2825 else if (i.tm.opcode_modifier & JumpInterSegment)
2826 {
2827 int size;
2828 int prefix;
2829 int code16;
2830
2831 code16 = 0;
2832 if (flag_code == CODE_16BIT)
2833 code16 = CODE16;
2834
2835 prefix = 0;
2836 if (i.prefix[DATA_PREFIX])
2837 {
2838 prefix = 1;
2839 i.prefixes -= 1;
2840 code16 ^= CODE16;
2841 }
2842 if (i.prefix[REX_PREFIX])
2843 {
2844 prefix++;
2845 i.prefixes -= 1;
2846 }
2847
2848 size = 4;
2849 if (code16)
2850 size = 2;
2851
2852 if (i.prefixes != 0 && !intel_syntax)
2853 as_warn (_("skipping prefixes on this instruction"));
2854
2855 /* 1 opcode; 2 segment; offset */
2856 p = frag_more (prefix + 1 + 2 + size);
2857
2858 if (i.prefix[DATA_PREFIX])
2859 *p++ = DATA_PREFIX_OPCODE;
2860
2861 if (i.prefix[REX_PREFIX])
2862 *p++ = i.prefix[REX_PREFIX];
2863
2864 *p++ = i.tm.base_opcode;
2865 if (i.op[1].imms->X_op == O_constant)
2866 {
2867 offsetT n = i.op[1].imms->X_add_number;
2868
2869 if (size == 2
2870 && !fits_in_unsigned_word (n)
2871 && !fits_in_signed_word (n))
2872 {
2873 as_bad (_("16-bit jump out of range"));
2874 return;
2875 }
2876 md_number_to_chars (p, n, size);
2877 }
2878 else
2879 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2880 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
2881 if (i.op[0].imms->X_op != O_constant)
2882 as_bad (_("can't handle non absolute segment in `%s'"),
2883 i.tm.name);
2884 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
2885 }
2886 else
2887 {
2888 /* Output normal instructions here. */
2889 unsigned char *q;
2890
2891 /* All opcodes on i386 have eighter 1 or 2 bytes. We may use third
2892 byte for the SSE instructions to specify prefix they require. */
2893 if (i.tm.base_opcode & 0xff0000)
2894 add_prefix ((i.tm.base_opcode >> 16) & 0xff);
2895
2896 /* The prefix bytes. */
2897 for (q = i.prefix;
2898 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
2899 q++)
2900 {
2901 if (*q)
2902 {
2903 p = frag_more (1);
2904 md_number_to_chars (p, (valueT) *q, 1);
2905 }
2906 }
2907
2908 /* Now the opcode; be careful about word order here! */
2909 if (fits_in_unsigned_byte (i.tm.base_opcode))
2910 {
2911 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
2912 }
2913 else
2914 {
2915 p = frag_more (2);
2916 /* Put out high byte first: can't use md_number_to_chars! */
2917 *p++ = (i.tm.base_opcode >> 8) & 0xff;
2918 *p = i.tm.base_opcode & 0xff;
2919 }
2920
2921 /* Now the modrm byte and sib byte (if present). */
2922 if (i.tm.opcode_modifier & Modrm)
2923 {
2924 p = frag_more (1);
2925 md_number_to_chars (p,
2926 (valueT) (i.rm.regmem << 0
2927 | i.rm.reg << 3
2928 | i.rm.mode << 6),
2929 1);
2930 /* If i.rm.regmem == ESP (4)
2931 && i.rm.mode != (Register mode)
2932 && not 16 bit
2933 ==> need second modrm byte. */
2934 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
2935 && i.rm.mode != 3
2936 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
2937 {
2938 p = frag_more (1);
2939 md_number_to_chars (p,
2940 (valueT) (i.sib.base << 0
2941 | i.sib.index << 3
2942 | i.sib.scale << 6),
2943 1);
2944 }
2945 }
2946
2947 if (i.disp_operands)
2948 {
2949 register unsigned int n;
2950
2951 for (n = 0; n < i.operands; n++)
2952 {
2953 if (i.types[n] & Disp)
2954 {
2955 if (i.op[n].disps->X_op == O_constant)
2956 {
2957 int size;
2958 offsetT val;
2959
2960 size = 4;
2961 if (i.types[n] & (Disp8 | Disp16 | Disp64))
2962 {
2963 size = 2;
2964 if (i.types[n] & Disp8)
2965 size = 1;
2966 if (i.types[n] & Disp64)
2967 size = 8;
2968 }
2969 val = offset_in_range (i.op[n].disps->X_add_number,
2970 size);
2971 p = frag_more (size);
2972 md_number_to_chars (p, val, size);
2973 }
2974 else
2975 {
2976 int size = 4;
2977 int sign = 0;
2978 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
2979
2980 /* The PC relative address is computed relative
2981 to the instruction boundary, so in case immediate
2982 fields follows, we need to adjust the value. */
2983 if (pcrel && i.imm_operands)
2984 {
2985 int imm_size = 4;
2986 register unsigned int n1;
2987
2988 for (n1 = 0; n1 < i.operands; n1++)
2989 if (i.types[n1] & Imm)
2990 {
2991 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
2992 {
2993 imm_size = 2;
2994 if (i.types[n1] & (Imm8 | Imm8S))
2995 imm_size = 1;
2996 if (i.types[n1] & Imm64)
2997 imm_size = 8;
2998 }
2999 break;
3000 }
3001 /* We should find the immediate. */
3002 if (n1 == i.operands)
3003 abort ();
3004 i.op[n].disps->X_add_number -= imm_size;
3005 }
3006
3007 if (i.types[n] & Disp32S)
3008 sign = 1;
3009
3010 if (i.types[n] & (Disp16 | Disp64))
3011 {
3012 size = 2;
3013 if (i.types[n] & Disp64)
3014 size = 8;
3015 }
3016
3017 p = frag_more (size);
3018 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3019 i.op[n].disps, pcrel,
3020 reloc (size, pcrel, sign, i.reloc[n]));
3021 }
3022 }
3023 }
3024 }
3025
3026 /* Output immediate. */
3027 if (i.imm_operands)
3028 {
3029 register unsigned int n;
3030
3031 for (n = 0; n < i.operands; n++)
3032 {
3033 if (i.types[n] & Imm)
3034 {
3035 if (i.op[n].imms->X_op == O_constant)
3036 {
3037 int size;
3038 offsetT val;
3039
3040 size = 4;
3041 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3042 {
3043 size = 2;
3044 if (i.types[n] & (Imm8 | Imm8S))
3045 size = 1;
3046 else if (i.types[n] & Imm64)
3047 size = 8;
3048 }
3049 val = offset_in_range (i.op[n].imms->X_add_number,
3050 size);
3051 p = frag_more (size);
3052 md_number_to_chars (p, val, size);
3053 }
3054 else
3055 {
3056 /* Not absolute_section.
3057 Need a 32-bit fixup (don't support 8bit
3058 non-absolute imms). Try to support other
3059 sizes ... */
3060 RELOC_ENUM reloc_type;
3061 int size = 4;
3062 int sign = 0;
3063
3064 if ((i.types[n] & (Imm32S))
3065 && i.suffix == QWORD_MNEM_SUFFIX)
3066 sign = 1;
3067 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3068 {
3069 size = 2;
3070 if (i.types[n] & (Imm8 | Imm8S))
3071 size = 1;
3072 if (i.types[n] & Imm64)
3073 size = 8;
3074 }
3075
3076 p = frag_more (size);
3077 reloc_type = reloc (size, 0, sign, i.reloc[n]);
3078 #ifdef BFD_ASSEMBLER
3079 if (reloc_type == BFD_RELOC_32
3080 && GOT_symbol
3081 && GOT_symbol == i.op[n].imms->X_add_symbol
3082 && (i.op[n].imms->X_op == O_symbol
3083 || (i.op[n].imms->X_op == O_add
3084 && ((symbol_get_value_expression
3085 (i.op[n].imms->X_op_symbol)->X_op)
3086 == O_subtract))))
3087 {
3088 /* We don't support dynamic linking on x86-64 yet. */
3089 if (flag_code == CODE_64BIT)
3090 abort ();
3091 reloc_type = BFD_RELOC_386_GOTPC;
3092 i.op[n].imms->X_add_number += 3;
3093 }
3094 #endif
3095 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3096 i.op[n].imms, 0, reloc_type);
3097 }
3098 }
3099 }
3100 }
3101 }
3102
3103 #ifdef DEBUG386
3104 if (flag_debug)
3105 {
3106 pi (line, &i);
3107 }
3108 #endif /* DEBUG386 */
3109 }
3110 }
3111 \f
3112 #ifndef LEX_AT
3113 static char *lex_got PARAMS ((RELOC_ENUM *, int *));
3114
3115 /* Parse operands of the form
3116 <symbol>@GOTOFF+<nnn>
3117 and similar .plt or .got references.
3118
3119 If we find one, set up the correct relocation in RELOC and copy the
3120 input string, minus the `@GOTOFF' into a malloc'd buffer for
3121 parsing by the calling routine. Return this buffer, and if ADJUST
3122 is non-null set it to the length of the string we removed from the
3123 input line. Otherwise return NULL. */
3124 static char *
3125 lex_got (reloc, adjust)
3126 RELOC_ENUM *reloc;
3127 int *adjust;
3128 {
3129 static const char * const mode_name[NUM_FLAG_CODE] = { "32", "16", "64" };
3130 static const struct {
3131 const char *str;
3132 const RELOC_ENUM rel[NUM_FLAG_CODE];
3133 } gotrel[] = {
3134 { "PLT", { BFD_RELOC_386_PLT32, 0, BFD_RELOC_X86_64_PLT32 } },
3135 { "GOTOFF", { BFD_RELOC_386_GOTOFF, 0, 0 } },
3136 { "GOTPCREL", { 0, 0, BFD_RELOC_X86_64_GOTPCREL } },
3137 { "GOT", { BFD_RELOC_386_GOT32, 0, BFD_RELOC_X86_64_GOT32 } }
3138 };
3139 char *cp;
3140 unsigned int j;
3141
3142 for (cp = input_line_pointer; *cp != '@'; cp++)
3143 if (is_end_of_line[(unsigned char) *cp])
3144 return NULL;
3145
3146 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
3147 {
3148 int len;
3149
3150 len = strlen (gotrel[j].str);
3151 if (strncmp (cp + 1, gotrel[j].str, len) == 0)
3152 {
3153 if (gotrel[j].rel[(unsigned int) flag_code] != 0)
3154 {
3155 int first;
3156 char *tmpbuf;
3157
3158 *reloc = gotrel[j].rel[(unsigned int) flag_code];
3159
3160 if (GOT_symbol == NULL)
3161 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3162
3163 /* Replace the relocation token with ' ', so that
3164 errors like foo@GOTOFF1 will be detected. */
3165 first = cp - input_line_pointer;
3166 tmpbuf = xmalloc (strlen (input_line_pointer));
3167 memcpy (tmpbuf, input_line_pointer, first);
3168 tmpbuf[first] = ' ';
3169 strcpy (tmpbuf + first + 1, cp + 1 + len);
3170 if (adjust)
3171 *adjust = len;
3172 return tmpbuf;
3173 }
3174
3175 as_bad (_("@%s reloc is not supported in %s bit mode"),
3176 gotrel[j].str, mode_name[(unsigned int) flag_code]);
3177 return NULL;
3178 }
3179 }
3180
3181 /* Might be a symbol version string. Don't as_bad here. */
3182 return NULL;
3183 }
3184
3185 /* x86_cons_fix_new is called via the expression parsing code when a
3186 reloc is needed. We use this hook to get the correct .got reloc. */
3187 static RELOC_ENUM got_reloc = NO_RELOC;
3188
3189 void
3190 x86_cons_fix_new (frag, off, len, exp)
3191 fragS *frag;
3192 unsigned int off;
3193 unsigned int len;
3194 expressionS *exp;
3195 {
3196 RELOC_ENUM r = reloc (len, 0, 0, got_reloc);
3197 got_reloc = NO_RELOC;
3198 fix_new_exp (frag, off, len, exp, 0, r);
3199 }
3200
3201 void
3202 x86_cons (exp, size)
3203 expressionS *exp;
3204 int size;
3205 {
3206 if (size == 4)
3207 {
3208 /* Handle @GOTOFF and the like in an expression. */
3209 char *save;
3210 char *gotfree_input_line;
3211 int adjust;
3212
3213 save = input_line_pointer;
3214 gotfree_input_line = lex_got (&got_reloc, &adjust);
3215 if (gotfree_input_line)
3216 input_line_pointer = gotfree_input_line;
3217
3218 expression (exp);
3219
3220 if (gotfree_input_line)
3221 {
3222 /* expression () has merrily parsed up to the end of line,
3223 or a comma - in the wrong buffer. Transfer how far
3224 input_line_pointer has moved to the right buffer. */
3225 input_line_pointer = (save
3226 + (input_line_pointer - gotfree_input_line)
3227 + adjust);
3228 free (gotfree_input_line);
3229 }
3230 }
3231 else
3232 expression (exp);
3233 }
3234 #endif
3235
3236 static int i386_immediate PARAMS ((char *));
3237
3238 static int
3239 i386_immediate (imm_start)
3240 char *imm_start;
3241 {
3242 char *save_input_line_pointer;
3243 #ifndef LEX_AT
3244 char *gotfree_input_line;
3245 #endif
3246 segT exp_seg = 0;
3247 expressionS *exp;
3248
3249 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
3250 {
3251 as_bad (_("only 1 or 2 immediate operands are allowed"));
3252 return 0;
3253 }
3254
3255 exp = &im_expressions[i.imm_operands++];
3256 i.op[this_operand].imms = exp;
3257
3258 if (is_space_char (*imm_start))
3259 ++imm_start;
3260
3261 save_input_line_pointer = input_line_pointer;
3262 input_line_pointer = imm_start;
3263
3264 #ifndef LEX_AT
3265 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3266 if (gotfree_input_line)
3267 input_line_pointer = gotfree_input_line;
3268 #endif
3269
3270 exp_seg = expression (exp);
3271
3272 SKIP_WHITESPACE ();
3273 if (*input_line_pointer)
3274 as_bad (_("junk `%s' after expression"), input_line_pointer);
3275
3276 input_line_pointer = save_input_line_pointer;
3277 #ifndef LEX_AT
3278 if (gotfree_input_line)
3279 free (gotfree_input_line);
3280 #endif
3281
3282 if (exp->X_op == O_absent || exp->X_op == O_big)
3283 {
3284 /* Missing or bad expr becomes absolute 0. */
3285 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
3286 imm_start);
3287 exp->X_op = O_constant;
3288 exp->X_add_number = 0;
3289 exp->X_add_symbol = (symbolS *) 0;
3290 exp->X_op_symbol = (symbolS *) 0;
3291 }
3292 else if (exp->X_op == O_constant)
3293 {
3294 /* Size it properly later. */
3295 i.types[this_operand] |= Imm64;
3296 /* If BFD64, sign extend val. */
3297 if (!use_rela_relocations)
3298 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
3299 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
3300 }
3301 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3302 else if (1
3303 #ifdef BFD_ASSEMBLER
3304 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3305 #endif
3306 && exp_seg != text_section
3307 && exp_seg != data_section
3308 && exp_seg != bss_section
3309 && exp_seg != undefined_section
3310 #ifdef BFD_ASSEMBLER
3311 && !bfd_is_com_section (exp_seg)
3312 #endif
3313 )
3314 {
3315 #ifdef BFD_ASSEMBLER
3316 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3317 #else
3318 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3319 #endif
3320 return 0;
3321 }
3322 #endif
3323 else
3324 {
3325 /* This is an address. The size of the address will be
3326 determined later, depending on destination register,
3327 suffix, or the default for the section. */
3328 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
3329 }
3330
3331 return 1;
3332 }
3333
3334 static char *i386_scale PARAMS ((char *));
3335
3336 static char *
3337 i386_scale (scale)
3338 char *scale;
3339 {
3340 offsetT val;
3341 char *save = input_line_pointer;
3342
3343 input_line_pointer = scale;
3344 val = get_absolute_expression ();
3345
3346 switch (val)
3347 {
3348 case 0:
3349 case 1:
3350 i.log2_scale_factor = 0;
3351 break;
3352 case 2:
3353 i.log2_scale_factor = 1;
3354 break;
3355 case 4:
3356 i.log2_scale_factor = 2;
3357 break;
3358 case 8:
3359 i.log2_scale_factor = 3;
3360 break;
3361 default:
3362 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
3363 scale);
3364 input_line_pointer = save;
3365 return NULL;
3366 }
3367 if (i.log2_scale_factor != 0 && ! i.index_reg)
3368 {
3369 as_warn (_("scale factor of %d without an index register"),
3370 1 << i.log2_scale_factor);
3371 #if SCALE1_WHEN_NO_INDEX
3372 i.log2_scale_factor = 0;
3373 #endif
3374 }
3375 scale = input_line_pointer;
3376 input_line_pointer = save;
3377 return scale;
3378 }
3379
3380 static int i386_displacement PARAMS ((char *, char *));
3381
3382 static int
3383 i386_displacement (disp_start, disp_end)
3384 char *disp_start;
3385 char *disp_end;
3386 {
3387 register expressionS *exp;
3388 segT exp_seg = 0;
3389 char *save_input_line_pointer;
3390 #ifndef LEX_AT
3391 char *gotfree_input_line;
3392 #endif
3393 int bigdisp = Disp32;
3394
3395 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3396 bigdisp = Disp16;
3397 if (flag_code == CODE_64BIT)
3398 bigdisp = Disp64;
3399 i.types[this_operand] |= bigdisp;
3400
3401 exp = &disp_expressions[i.disp_operands];
3402 i.op[this_operand].disps = exp;
3403 i.disp_operands++;
3404 save_input_line_pointer = input_line_pointer;
3405 input_line_pointer = disp_start;
3406 END_STRING_AND_SAVE (disp_end);
3407
3408 #ifndef GCC_ASM_O_HACK
3409 #define GCC_ASM_O_HACK 0
3410 #endif
3411 #if GCC_ASM_O_HACK
3412 END_STRING_AND_SAVE (disp_end + 1);
3413 if ((i.types[this_operand] & BaseIndex) != 0
3414 && displacement_string_end[-1] == '+')
3415 {
3416 /* This hack is to avoid a warning when using the "o"
3417 constraint within gcc asm statements.
3418 For instance:
3419
3420 #define _set_tssldt_desc(n,addr,limit,type) \
3421 __asm__ __volatile__ ( \
3422 "movw %w2,%0\n\t" \
3423 "movw %w1,2+%0\n\t" \
3424 "rorl $16,%1\n\t" \
3425 "movb %b1,4+%0\n\t" \
3426 "movb %4,5+%0\n\t" \
3427 "movb $0,6+%0\n\t" \
3428 "movb %h1,7+%0\n\t" \
3429 "rorl $16,%1" \
3430 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
3431
3432 This works great except that the output assembler ends
3433 up looking a bit weird if it turns out that there is
3434 no offset. You end up producing code that looks like:
3435
3436 #APP
3437 movw $235,(%eax)
3438 movw %dx,2+(%eax)
3439 rorl $16,%edx
3440 movb %dl,4+(%eax)
3441 movb $137,5+(%eax)
3442 movb $0,6+(%eax)
3443 movb %dh,7+(%eax)
3444 rorl $16,%edx
3445 #NO_APP
3446
3447 So here we provide the missing zero. */
3448
3449 *displacement_string_end = '0';
3450 }
3451 #endif
3452 #ifndef LEX_AT
3453 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3454 if (gotfree_input_line)
3455 input_line_pointer = gotfree_input_line;
3456 #endif
3457
3458 exp_seg = expression (exp);
3459
3460 SKIP_WHITESPACE ();
3461 if (*input_line_pointer)
3462 as_bad (_("junk `%s' after expression"), input_line_pointer);
3463 #if GCC_ASM_O_HACK
3464 RESTORE_END_STRING (disp_end + 1);
3465 #endif
3466 RESTORE_END_STRING (disp_end);
3467 input_line_pointer = save_input_line_pointer;
3468 #ifndef LEX_AT
3469 if (gotfree_input_line)
3470 free (gotfree_input_line);
3471 #endif
3472
3473 #ifdef BFD_ASSEMBLER
3474 /* We do this to make sure that the section symbol is in
3475 the symbol table. We will ultimately change the relocation
3476 to be relative to the beginning of the section. */
3477 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
3478 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3479 {
3480 if (exp->X_op != O_symbol)
3481 {
3482 as_bad (_("bad expression used with @%s"),
3483 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
3484 ? "GOTPCREL"
3485 : "GOTOFF"));
3486 return 0;
3487 }
3488
3489 if (S_IS_LOCAL (exp->X_add_symbol)
3490 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
3491 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
3492 exp->X_op = O_subtract;
3493 exp->X_op_symbol = GOT_symbol;
3494 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3495 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
3496 else
3497 i.reloc[this_operand] = BFD_RELOC_32;
3498 }
3499 #endif
3500
3501 if (exp->X_op == O_absent || exp->X_op == O_big)
3502 {
3503 /* Missing or bad expr becomes absolute 0. */
3504 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
3505 disp_start);
3506 exp->X_op = O_constant;
3507 exp->X_add_number = 0;
3508 exp->X_add_symbol = (symbolS *) 0;
3509 exp->X_op_symbol = (symbolS *) 0;
3510 }
3511
3512 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3513 if (exp->X_op != O_constant
3514 #ifdef BFD_ASSEMBLER
3515 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3516 #endif
3517 && exp_seg != text_section
3518 && exp_seg != data_section
3519 && exp_seg != bss_section
3520 && exp_seg != undefined_section)
3521 {
3522 #ifdef BFD_ASSEMBLER
3523 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3524 #else
3525 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3526 #endif
3527 return 0;
3528 }
3529 #endif
3530 else if (flag_code == CODE_64BIT)
3531 i.types[this_operand] |= Disp32S | Disp32;
3532 return 1;
3533 }
3534
3535 static int i386_index_check PARAMS ((const char *));
3536
3537 /* Make sure the memory operand we've been dealt is valid.
3538 Return 1 on success, 0 on a failure. */
3539
3540 static int
3541 i386_index_check (operand_string)
3542 const char *operand_string;
3543 {
3544 int ok;
3545 #if INFER_ADDR_PREFIX
3546 int fudged = 0;
3547
3548 tryprefix:
3549 #endif
3550 ok = 1;
3551 if (flag_code == CODE_64BIT)
3552 {
3553 /* 64bit checks. */
3554 if ((i.base_reg
3555 && ((i.base_reg->reg_type & Reg64) == 0)
3556 && (i.base_reg->reg_type != BaseIndex
3557 || i.index_reg))
3558 || (i.index_reg
3559 && ((i.index_reg->reg_type & (Reg64|BaseIndex))
3560 != (Reg64|BaseIndex))))
3561 ok = 0;
3562 }
3563 else
3564 {
3565 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3566 {
3567 /* 16bit checks. */
3568 if ((i.base_reg
3569 && ((i.base_reg->reg_type & (Reg16|BaseIndex|RegRex))
3570 != (Reg16|BaseIndex)))
3571 || (i.index_reg
3572 && (((i.index_reg->reg_type & (Reg16|BaseIndex))
3573 != (Reg16|BaseIndex))
3574 || ! (i.base_reg
3575 && i.base_reg->reg_num < 6
3576 && i.index_reg->reg_num >= 6
3577 && i.log2_scale_factor == 0))))
3578 ok = 0;
3579 }
3580 else
3581 {
3582 /* 32bit checks. */
3583 if ((i.base_reg
3584 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
3585 || (i.index_reg
3586 && ((i.index_reg->reg_type & (Reg32|BaseIndex|RegRex))
3587 != (Reg32|BaseIndex))))
3588 ok = 0;
3589 }
3590 }
3591 if (!ok)
3592 {
3593 #if INFER_ADDR_PREFIX
3594 if (flag_code != CODE_64BIT
3595 && i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0')
3596 {
3597 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
3598 i.prefixes += 1;
3599 /* Change the size of any displacement too. At most one of
3600 Disp16 or Disp32 is set.
3601 FIXME. There doesn't seem to be any real need for separate
3602 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
3603 Removing them would probably clean up the code quite a lot. */
3604 if (i.types[this_operand] & (Disp16|Disp32))
3605 i.types[this_operand] ^= (Disp16|Disp32);
3606 fudged = 1;
3607 goto tryprefix;
3608 }
3609 if (fudged)
3610 as_bad (_("`%s' is not a valid base/index expression"),
3611 operand_string);
3612 else
3613 #endif
3614 as_bad (_("`%s' is not a valid %s bit base/index expression"),
3615 operand_string,
3616 flag_code_names[flag_code]);
3617 return 0;
3618 }
3619 return 1;
3620 }
3621
3622 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
3623 on error. */
3624
3625 static int
3626 i386_operand (operand_string)
3627 char *operand_string;
3628 {
3629 const reg_entry *r;
3630 char *end_op;
3631 char *op_string = operand_string;
3632
3633 if (is_space_char (*op_string))
3634 ++op_string;
3635
3636 /* We check for an absolute prefix (differentiating,
3637 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
3638 if (*op_string == ABSOLUTE_PREFIX)
3639 {
3640 ++op_string;
3641 if (is_space_char (*op_string))
3642 ++op_string;
3643 i.types[this_operand] |= JumpAbsolute;
3644 }
3645
3646 /* Check if operand is a register. */
3647 if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
3648 && (r = parse_register (op_string, &end_op)) != NULL)
3649 {
3650 /* Check for a segment override by searching for ':' after a
3651 segment register. */
3652 op_string = end_op;
3653 if (is_space_char (*op_string))
3654 ++op_string;
3655 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
3656 {
3657 switch (r->reg_num)
3658 {
3659 case 0:
3660 i.seg[i.mem_operands] = &es;
3661 break;
3662 case 1:
3663 i.seg[i.mem_operands] = &cs;
3664 break;
3665 case 2:
3666 i.seg[i.mem_operands] = &ss;
3667 break;
3668 case 3:
3669 i.seg[i.mem_operands] = &ds;
3670 break;
3671 case 4:
3672 i.seg[i.mem_operands] = &fs;
3673 break;
3674 case 5:
3675 i.seg[i.mem_operands] = &gs;
3676 break;
3677 }
3678
3679 /* Skip the ':' and whitespace. */
3680 ++op_string;
3681 if (is_space_char (*op_string))
3682 ++op_string;
3683
3684 if (!is_digit_char (*op_string)
3685 && !is_identifier_char (*op_string)
3686 && *op_string != '('
3687 && *op_string != ABSOLUTE_PREFIX)
3688 {
3689 as_bad (_("bad memory operand `%s'"), op_string);
3690 return 0;
3691 }
3692 /* Handle case of %es:*foo. */
3693 if (*op_string == ABSOLUTE_PREFIX)
3694 {
3695 ++op_string;
3696 if (is_space_char (*op_string))
3697 ++op_string;
3698 i.types[this_operand] |= JumpAbsolute;
3699 }
3700 goto do_memory_reference;
3701 }
3702 if (*op_string)
3703 {
3704 as_bad (_("junk `%s' after register"), op_string);
3705 return 0;
3706 }
3707 i.types[this_operand] |= r->reg_type & ~BaseIndex;
3708 i.op[this_operand].regs = r;
3709 i.reg_operands++;
3710 }
3711 else if (*op_string == REGISTER_PREFIX)
3712 {
3713 as_bad (_("bad register name `%s'"), op_string);
3714 return 0;
3715 }
3716 else if (*op_string == IMMEDIATE_PREFIX)
3717 {
3718 ++op_string;
3719 if (i.types[this_operand] & JumpAbsolute)
3720 {
3721 as_bad (_("immediate operand illegal with absolute jump"));
3722 return 0;
3723 }
3724 if (!i386_immediate (op_string))
3725 return 0;
3726 }
3727 else if (is_digit_char (*op_string)
3728 || is_identifier_char (*op_string)
3729 || *op_string == '(')
3730 {
3731 /* This is a memory reference of some sort. */
3732 char *base_string;
3733
3734 /* Start and end of displacement string expression (if found). */
3735 char *displacement_string_start;
3736 char *displacement_string_end;
3737
3738 do_memory_reference:
3739 if ((i.mem_operands == 1
3740 && (current_templates->start->opcode_modifier & IsString) == 0)
3741 || i.mem_operands == 2)
3742 {
3743 as_bad (_("too many memory references for `%s'"),
3744 current_templates->start->name);
3745 return 0;
3746 }
3747
3748 /* Check for base index form. We detect the base index form by
3749 looking for an ')' at the end of the operand, searching
3750 for the '(' matching it, and finding a REGISTER_PREFIX or ','
3751 after the '('. */
3752 base_string = op_string + strlen (op_string);
3753
3754 --base_string;
3755 if (is_space_char (*base_string))
3756 --base_string;
3757
3758 /* If we only have a displacement, set-up for it to be parsed later. */
3759 displacement_string_start = op_string;
3760 displacement_string_end = base_string + 1;
3761
3762 if (*base_string == ')')
3763 {
3764 char *temp_string;
3765 unsigned int parens_balanced = 1;
3766 /* We've already checked that the number of left & right ()'s are
3767 equal, so this loop will not be infinite. */
3768 do
3769 {
3770 base_string--;
3771 if (*base_string == ')')
3772 parens_balanced++;
3773 if (*base_string == '(')
3774 parens_balanced--;
3775 }
3776 while (parens_balanced);
3777
3778 temp_string = base_string;
3779
3780 /* Skip past '(' and whitespace. */
3781 ++base_string;
3782 if (is_space_char (*base_string))
3783 ++base_string;
3784
3785 if (*base_string == ','
3786 || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3787 && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
3788 {
3789 displacement_string_end = temp_string;
3790
3791 i.types[this_operand] |= BaseIndex;
3792
3793 if (i.base_reg)
3794 {
3795 base_string = end_op;
3796 if (is_space_char (*base_string))
3797 ++base_string;
3798 }
3799
3800 /* There may be an index reg or scale factor here. */
3801 if (*base_string == ',')
3802 {
3803 ++base_string;
3804 if (is_space_char (*base_string))
3805 ++base_string;
3806
3807 if ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3808 && (i.index_reg = parse_register (base_string, &end_op)) != NULL)
3809 {
3810 base_string = end_op;
3811 if (is_space_char (*base_string))
3812 ++base_string;
3813 if (*base_string == ',')
3814 {
3815 ++base_string;
3816 if (is_space_char (*base_string))
3817 ++base_string;
3818 }
3819 else if (*base_string != ')')
3820 {
3821 as_bad (_("expecting `,' or `)' after index register in `%s'"),
3822 operand_string);
3823 return 0;
3824 }
3825 }
3826 else if (*base_string == REGISTER_PREFIX)
3827 {
3828 as_bad (_("bad register name `%s'"), base_string);
3829 return 0;
3830 }
3831
3832 /* Check for scale factor. */
3833 if (*base_string != ')')
3834 {
3835 char *end_scale = i386_scale (base_string);
3836
3837 if (!end_scale)
3838 return 0;
3839
3840 base_string = end_scale;
3841 if (is_space_char (*base_string))
3842 ++base_string;
3843 if (*base_string != ')')
3844 {
3845 as_bad (_("expecting `)' after scale factor in `%s'"),
3846 operand_string);
3847 return 0;
3848 }
3849 }
3850 else if (!i.index_reg)
3851 {
3852 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
3853 *base_string);
3854 return 0;
3855 }
3856 }
3857 else if (*base_string != ')')
3858 {
3859 as_bad (_("expecting `,' or `)' after base register in `%s'"),
3860 operand_string);
3861 return 0;
3862 }
3863 }
3864 else if (*base_string == REGISTER_PREFIX)
3865 {
3866 as_bad (_("bad register name `%s'"), base_string);
3867 return 0;
3868 }
3869 }
3870
3871 /* If there's an expression beginning the operand, parse it,
3872 assuming displacement_string_start and
3873 displacement_string_end are meaningful. */
3874 if (displacement_string_start != displacement_string_end)
3875 {
3876 if (!i386_displacement (displacement_string_start,
3877 displacement_string_end))
3878 return 0;
3879 }
3880
3881 /* Special case for (%dx) while doing input/output op. */
3882 if (i.base_reg
3883 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
3884 && i.index_reg == 0
3885 && i.log2_scale_factor == 0
3886 && i.seg[i.mem_operands] == 0
3887 && (i.types[this_operand] & Disp) == 0)
3888 {
3889 i.types[this_operand] = InOutPortReg;
3890 return 1;
3891 }
3892
3893 if (i386_index_check (operand_string) == 0)
3894 return 0;
3895 i.mem_operands++;
3896 }
3897 else
3898 {
3899 /* It's not a memory operand; argh! */
3900 as_bad (_("invalid char %s beginning operand %d `%s'"),
3901 output_invalid (*op_string),
3902 this_operand + 1,
3903 op_string);
3904 return 0;
3905 }
3906 return 1; /* Normal return. */
3907 }
3908 \f
3909 /* md_estimate_size_before_relax()
3910
3911 Called just before relax() for rs_machine_dependent frags. The x86
3912 assembler uses these frags to handle variable size jump
3913 instructions.
3914
3915 Any symbol that is now undefined will not become defined.
3916 Return the correct fr_subtype in the frag.
3917 Return the initial "guess for variable size of frag" to caller.
3918 The guess is actually the growth beyond the fixed part. Whatever
3919 we do to grow the fixed or variable part contributes to our
3920 returned value. */
3921
3922 int
3923 md_estimate_size_before_relax (fragP, segment)
3924 register fragS *fragP;
3925 register segT segment;
3926 {
3927 /* We've already got fragP->fr_subtype right; all we have to do is
3928 check for un-relaxable symbols. On an ELF system, we can't relax
3929 an externally visible symbol, because it may be overridden by a
3930 shared library. */
3931 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
3932 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
3933 || S_IS_EXTERNAL (fragP->fr_symbol)
3934 || S_IS_WEAK (fragP->fr_symbol)
3935 #endif
3936 )
3937 {
3938 /* Symbol is undefined in this segment, or we need to keep a
3939 reloc so that weak symbols can be overridden. */
3940 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
3941 RELOC_ENUM reloc_type;
3942 unsigned char *opcode;
3943 int old_fr_fix;
3944
3945 if (fragP->fr_var != NO_RELOC)
3946 reloc_type = fragP->fr_var;
3947 else if (size == 2)
3948 reloc_type = BFD_RELOC_16_PCREL;
3949 else
3950 reloc_type = BFD_RELOC_32_PCREL;
3951
3952 old_fr_fix = fragP->fr_fix;
3953 opcode = (unsigned char *) fragP->fr_opcode;
3954
3955 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
3956 {
3957 case UNCOND_JUMP:
3958 /* Make jmp (0xeb) a (d)word displacement jump. */
3959 opcode[0] = 0xe9;
3960 fragP->fr_fix += size;
3961 fix_new (fragP, old_fr_fix, size,
3962 fragP->fr_symbol,
3963 fragP->fr_offset, 1,
3964 reloc_type);
3965 break;
3966
3967 case COND_JUMP86:
3968 if (no_cond_jump_promotion)
3969 goto relax_guess;
3970
3971 if (size == 2)
3972 {
3973 /* Negate the condition, and branch past an
3974 unconditional jump. */
3975 opcode[0] ^= 1;
3976 opcode[1] = 3;
3977 /* Insert an unconditional jump. */
3978 opcode[2] = 0xe9;
3979 /* We added two extra opcode bytes, and have a two byte
3980 offset. */
3981 fragP->fr_fix += 2 + 2;
3982 fix_new (fragP, old_fr_fix + 2, 2,
3983 fragP->fr_symbol,
3984 fragP->fr_offset, 1,
3985 reloc_type);
3986 break;
3987 }
3988 /* Fall through. */
3989
3990 case COND_JUMP:
3991 if (no_cond_jump_promotion)
3992 goto relax_guess;
3993
3994 /* This changes the byte-displacement jump 0x7N
3995 to the (d)word-displacement jump 0x0f,0x8N. */
3996 opcode[1] = opcode[0] + 0x10;
3997 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
3998 /* We've added an opcode byte. */
3999 fragP->fr_fix += 1 + size;
4000 fix_new (fragP, old_fr_fix + 1, size,
4001 fragP->fr_symbol,
4002 fragP->fr_offset, 1,
4003 reloc_type);
4004 break;
4005
4006 default:
4007 BAD_CASE (fragP->fr_subtype);
4008 break;
4009 }
4010 frag_wane (fragP);
4011 return fragP->fr_fix - old_fr_fix;
4012 }
4013
4014 relax_guess:
4015 /* Guess size depending on current relax state. Initially the relax
4016 state will correspond to a short jump and we return 1, because
4017 the variable part of the frag (the branch offset) is one byte
4018 long. However, we can relax a section more than once and in that
4019 case we must either set fr_subtype back to the unrelaxed state,
4020 or return the value for the appropriate branch. */
4021 return md_relax_table[fragP->fr_subtype].rlx_length;
4022 }
4023
4024 /* Called after relax() is finished.
4025
4026 In: Address of frag.
4027 fr_type == rs_machine_dependent.
4028 fr_subtype is what the address relaxed to.
4029
4030 Out: Any fixSs and constants are set up.
4031 Caller will turn frag into a ".space 0". */
4032
4033 #ifndef BFD_ASSEMBLER
4034 void
4035 md_convert_frag (headers, sec, fragP)
4036 object_headers *headers ATTRIBUTE_UNUSED;
4037 segT sec ATTRIBUTE_UNUSED;
4038 register fragS *fragP;
4039 #else
4040 void
4041 md_convert_frag (abfd, sec, fragP)
4042 bfd *abfd ATTRIBUTE_UNUSED;
4043 segT sec ATTRIBUTE_UNUSED;
4044 register fragS *fragP;
4045 #endif
4046 {
4047 register unsigned char *opcode;
4048 unsigned char *where_to_put_displacement = NULL;
4049 offsetT target_address;
4050 offsetT opcode_address;
4051 unsigned int extension = 0;
4052 offsetT displacement_from_opcode_start;
4053
4054 opcode = (unsigned char *) fragP->fr_opcode;
4055
4056 /* Address we want to reach in file space. */
4057 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
4058 #ifdef BFD_ASSEMBLER
4059 /* Not needed otherwise? */
4060 {
4061 /* Local symbols which have already been resolved have a NULL frag. */
4062 fragS *sym_frag = symbol_get_frag (fragP->fr_symbol);
4063 if (sym_frag)
4064 target_address += sym_frag->fr_address;
4065 }
4066 #endif
4067
4068 /* Address opcode resides at in file space. */
4069 opcode_address = fragP->fr_address + fragP->fr_fix;
4070
4071 /* Displacement from opcode start to fill into instruction. */
4072 displacement_from_opcode_start = target_address - opcode_address;
4073
4074 if ((fragP->fr_subtype & BIG) == 0)
4075 {
4076 /* Don't have to change opcode. */
4077 extension = 1; /* 1 opcode + 1 displacement */
4078 where_to_put_displacement = &opcode[1];
4079 }
4080 else
4081 {
4082 if (no_cond_jump_promotion
4083 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
4084 as_warn_where (fragP->fr_file, fragP->fr_line, _("long jump required"));
4085
4086 switch (fragP->fr_subtype)
4087 {
4088 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
4089 extension = 4; /* 1 opcode + 4 displacement */
4090 opcode[0] = 0xe9;
4091 where_to_put_displacement = &opcode[1];
4092 break;
4093
4094 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
4095 extension = 2; /* 1 opcode + 2 displacement */
4096 opcode[0] = 0xe9;
4097 where_to_put_displacement = &opcode[1];
4098 break;
4099
4100 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
4101 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
4102 extension = 5; /* 2 opcode + 4 displacement */
4103 opcode[1] = opcode[0] + 0x10;
4104 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4105 where_to_put_displacement = &opcode[2];
4106 break;
4107
4108 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
4109 extension = 3; /* 2 opcode + 2 displacement */
4110 opcode[1] = opcode[0] + 0x10;
4111 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4112 where_to_put_displacement = &opcode[2];
4113 break;
4114
4115 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
4116 extension = 4;
4117 opcode[0] ^= 1;
4118 opcode[1] = 3;
4119 opcode[2] = 0xe9;
4120 where_to_put_displacement = &opcode[3];
4121 break;
4122
4123 default:
4124 BAD_CASE (fragP->fr_subtype);
4125 break;
4126 }
4127 }
4128
4129 /* Now put displacement after opcode. */
4130 md_number_to_chars ((char *) where_to_put_displacement,
4131 (valueT) (displacement_from_opcode_start - extension),
4132 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
4133 fragP->fr_fix += extension;
4134 }
4135 \f
4136 /* Size of byte displacement jmp. */
4137 int md_short_jump_size = 2;
4138
4139 /* Size of dword displacement jmp. */
4140 int md_long_jump_size = 5;
4141
4142 /* Size of relocation record. */
4143 const int md_reloc_size = 8;
4144
4145 void
4146 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
4147 char *ptr;
4148 addressT from_addr, to_addr;
4149 fragS *frag ATTRIBUTE_UNUSED;
4150 symbolS *to_symbol ATTRIBUTE_UNUSED;
4151 {
4152 offsetT offset;
4153
4154 offset = to_addr - (from_addr + 2);
4155 /* Opcode for byte-disp jump. */
4156 md_number_to_chars (ptr, (valueT) 0xeb, 1);
4157 md_number_to_chars (ptr + 1, (valueT) offset, 1);
4158 }
4159
4160 void
4161 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
4162 char *ptr;
4163 addressT from_addr, to_addr;
4164 fragS *frag ATTRIBUTE_UNUSED;
4165 symbolS *to_symbol ATTRIBUTE_UNUSED;
4166 {
4167 offsetT offset;
4168
4169 offset = to_addr - (from_addr + 5);
4170 md_number_to_chars (ptr, (valueT) 0xe9, 1);
4171 md_number_to_chars (ptr + 1, (valueT) offset, 4);
4172 }
4173 \f
4174 /* Apply a fixup (fixS) to segment data, once it has been determined
4175 by our caller that we have all the info we need to fix it up.
4176
4177 On the 386, immediates, displacements, and data pointers are all in
4178 the same (little-endian) format, so we don't need to care about which
4179 we are handling. */
4180
4181 int
4182 md_apply_fix3 (fixP, valp, seg)
4183 /* The fix we're to put in. */
4184 fixS *fixP;
4185
4186 /* Pointer to the value of the bits. */
4187 valueT *valp;
4188
4189 /* Segment fix is from. */
4190 segT seg ATTRIBUTE_UNUSED;
4191 {
4192 register char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
4193 valueT value = *valp;
4194
4195 #if defined (BFD_ASSEMBLER) && !defined (TE_Mach)
4196 if (fixP->fx_pcrel)
4197 {
4198 switch (fixP->fx_r_type)
4199 {
4200 default:
4201 break;
4202
4203 case BFD_RELOC_32:
4204 fixP->fx_r_type = BFD_RELOC_32_PCREL;
4205 break;
4206 case BFD_RELOC_16:
4207 fixP->fx_r_type = BFD_RELOC_16_PCREL;
4208 break;
4209 case BFD_RELOC_8:
4210 fixP->fx_r_type = BFD_RELOC_8_PCREL;
4211 break;
4212 }
4213 }
4214
4215 /* This is a hack. There should be a better way to handle this.
4216 This covers for the fact that bfd_install_relocation will
4217 subtract the current location (for partial_inplace, PC relative
4218 relocations); see more below. */
4219 if ((fixP->fx_r_type == BFD_RELOC_32_PCREL
4220 || fixP->fx_r_type == BFD_RELOC_16_PCREL
4221 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
4222 && fixP->fx_addsy && !use_rela_relocations)
4223 {
4224 #ifndef OBJ_AOUT
4225 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4226 #ifdef TE_PE
4227 || OUTPUT_FLAVOR == bfd_target_coff_flavour
4228 #endif
4229 )
4230 value += fixP->fx_where + fixP->fx_frag->fr_address;
4231 #endif
4232 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4233 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
4234 {
4235 segT fseg = S_GET_SEGMENT (fixP->fx_addsy);
4236
4237 if ((fseg == seg
4238 || (symbol_section_p (fixP->fx_addsy)
4239 && fseg != absolute_section))
4240 && ! S_IS_EXTERNAL (fixP->fx_addsy)
4241 && ! S_IS_WEAK (fixP->fx_addsy)
4242 && S_IS_DEFINED (fixP->fx_addsy)
4243 && ! S_IS_COMMON (fixP->fx_addsy))
4244 {
4245 /* Yes, we add the values in twice. This is because
4246 bfd_perform_relocation subtracts them out again. I think
4247 bfd_perform_relocation is broken, but I don't dare change
4248 it. FIXME. */
4249 value += fixP->fx_where + fixP->fx_frag->fr_address;
4250 }
4251 }
4252 #endif
4253 #if defined (OBJ_COFF) && defined (TE_PE)
4254 /* For some reason, the PE format does not store a section
4255 address offset for a PC relative symbol. */
4256 if (S_GET_SEGMENT (fixP->fx_addsy) != seg)
4257 value += md_pcrel_from (fixP);
4258 #endif
4259 }
4260
4261 /* Fix a few things - the dynamic linker expects certain values here,
4262 and we must not dissappoint it. */
4263 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4264 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4265 && fixP->fx_addsy)
4266 switch (fixP->fx_r_type)
4267 {
4268 case BFD_RELOC_386_PLT32:
4269 case BFD_RELOC_X86_64_PLT32:
4270 /* Make the jump instruction point to the address of the operand. At
4271 runtime we merely add the offset to the actual PLT entry. */
4272 value = -4;
4273 break;
4274 case BFD_RELOC_386_GOTPC:
4275
4276 /* This is tough to explain. We end up with this one if we have
4277 * operands that look like "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal
4278 * here is to obtain the absolute address of the GOT, and it is strongly
4279 * preferable from a performance point of view to avoid using a runtime
4280 * relocation for this. The actual sequence of instructions often look
4281 * something like:
4282 *
4283 * call .L66
4284 * .L66:
4285 * popl %ebx
4286 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
4287 *
4288 * The call and pop essentially return the absolute address of
4289 * the label .L66 and store it in %ebx. The linker itself will
4290 * ultimately change the first operand of the addl so that %ebx points to
4291 * the GOT, but to keep things simple, the .o file must have this operand
4292 * set so that it generates not the absolute address of .L66, but the
4293 * absolute address of itself. This allows the linker itself simply
4294 * treat a GOTPC relocation as asking for a pcrel offset to the GOT to be
4295 * added in, and the addend of the relocation is stored in the operand
4296 * field for the instruction itself.
4297 *
4298 * Our job here is to fix the operand so that it would add the correct
4299 * offset so that %ebx would point to itself. The thing that is tricky is
4300 * that .-.L66 will point to the beginning of the instruction, so we need
4301 * to further modify the operand so that it will point to itself.
4302 * There are other cases where you have something like:
4303 *
4304 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
4305 *
4306 * and here no correction would be required. Internally in the assembler
4307 * we treat operands of this form as not being pcrel since the '.' is
4308 * explicitly mentioned, and I wonder whether it would simplify matters
4309 * to do it this way. Who knows. In earlier versions of the PIC patches,
4310 * the pcrel_adjust field was used to store the correction, but since the
4311 * expression is not pcrel, I felt it would be confusing to do it this
4312 * way. */
4313
4314 value -= 1;
4315 break;
4316 case BFD_RELOC_386_GOT32:
4317 case BFD_RELOC_X86_64_GOT32:
4318 value = 0; /* Fully resolved at runtime. No addend. */
4319 break;
4320 case BFD_RELOC_386_GOTOFF:
4321 case BFD_RELOC_X86_64_GOTPCREL:
4322 break;
4323
4324 case BFD_RELOC_VTABLE_INHERIT:
4325 case BFD_RELOC_VTABLE_ENTRY:
4326 fixP->fx_done = 0;
4327 return 1;
4328
4329 default:
4330 break;
4331 }
4332 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
4333 *valp = value;
4334 #endif /* defined (BFD_ASSEMBLER) && !defined (TE_Mach) */
4335
4336 #ifndef BFD_ASSEMBLER
4337 md_number_to_chars (p, value, fixP->fx_size);
4338 #else
4339 /* Are we finished with this relocation now? */
4340 if (fixP->fx_addsy == 0 && fixP->fx_pcrel == 0)
4341 fixP->fx_done = 1;
4342 else if (use_rela_relocations)
4343 {
4344 fixP->fx_no_overflow = 1;
4345 value = 0;
4346 }
4347 md_number_to_chars (p, value, fixP->fx_size);
4348 #endif
4349
4350 return 1;
4351 }
4352 \f
4353 #define MAX_LITTLENUMS 6
4354
4355 /* Turn the string pointed to by litP into a floating point constant
4356 of type TYPE, and emit the appropriate bytes. The number of
4357 LITTLENUMS emitted is stored in *SIZEP. An error message is
4358 returned, or NULL on OK. */
4359
4360 char *
4361 md_atof (type, litP, sizeP)
4362 int type;
4363 char *litP;
4364 int *sizeP;
4365 {
4366 int prec;
4367 LITTLENUM_TYPE words[MAX_LITTLENUMS];
4368 LITTLENUM_TYPE *wordP;
4369 char *t;
4370
4371 switch (type)
4372 {
4373 case 'f':
4374 case 'F':
4375 prec = 2;
4376 break;
4377
4378 case 'd':
4379 case 'D':
4380 prec = 4;
4381 break;
4382
4383 case 'x':
4384 case 'X':
4385 prec = 5;
4386 break;
4387
4388 default:
4389 *sizeP = 0;
4390 return _("Bad call to md_atof ()");
4391 }
4392 t = atof_ieee (input_line_pointer, type, words);
4393 if (t)
4394 input_line_pointer = t;
4395
4396 *sizeP = prec * sizeof (LITTLENUM_TYPE);
4397 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
4398 the bigendian 386. */
4399 for (wordP = words + prec - 1; prec--;)
4400 {
4401 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
4402 litP += sizeof (LITTLENUM_TYPE);
4403 }
4404 return 0;
4405 }
4406 \f
4407 char output_invalid_buf[8];
4408
4409 static char *
4410 output_invalid (c)
4411 int c;
4412 {
4413 if (isprint (c))
4414 sprintf (output_invalid_buf, "'%c'", c);
4415 else
4416 sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
4417 return output_invalid_buf;
4418 }
4419
4420 /* REG_STRING starts *before* REGISTER_PREFIX. */
4421
4422 static const reg_entry *
4423 parse_register (reg_string, end_op)
4424 char *reg_string;
4425 char **end_op;
4426 {
4427 char *s = reg_string;
4428 char *p;
4429 char reg_name_given[MAX_REG_NAME_SIZE + 1];
4430 const reg_entry *r;
4431
4432 /* Skip possible REGISTER_PREFIX and possible whitespace. */
4433 if (*s == REGISTER_PREFIX)
4434 ++s;
4435
4436 if (is_space_char (*s))
4437 ++s;
4438
4439 p = reg_name_given;
4440 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
4441 {
4442 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
4443 return (const reg_entry *) NULL;
4444 s++;
4445 }
4446
4447 /* For naked regs, make sure that we are not dealing with an identifier.
4448 This prevents confusing an identifier like `eax_var' with register
4449 `eax'. */
4450 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
4451 return (const reg_entry *) NULL;
4452
4453 *end_op = s;
4454
4455 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
4456
4457 /* Handle floating point regs, allowing spaces in the (i) part. */
4458 if (r == i386_regtab /* %st is first entry of table */)
4459 {
4460 if (is_space_char (*s))
4461 ++s;
4462 if (*s == '(')
4463 {
4464 ++s;
4465 if (is_space_char (*s))
4466 ++s;
4467 if (*s >= '0' && *s <= '7')
4468 {
4469 r = &i386_float_regtab[*s - '0'];
4470 ++s;
4471 if (is_space_char (*s))
4472 ++s;
4473 if (*s == ')')
4474 {
4475 *end_op = s + 1;
4476 return r;
4477 }
4478 }
4479 /* We have "%st(" then garbage. */
4480 return (const reg_entry *) NULL;
4481 }
4482 }
4483
4484 return r;
4485 }
4486 \f
4487 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4488 const char *md_shortopts = "kVQ:sq";
4489 #else
4490 const char *md_shortopts = "q";
4491 #endif
4492
4493 struct option md_longopts[] = {
4494 #define OPTION_32 (OPTION_MD_BASE + 0)
4495 {"32", no_argument, NULL, OPTION_32},
4496 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4497 #define OPTION_64 (OPTION_MD_BASE + 1)
4498 {"64", no_argument, NULL, OPTION_64},
4499 #endif
4500 {NULL, no_argument, NULL, 0}
4501 };
4502 size_t md_longopts_size = sizeof (md_longopts);
4503
4504 int
4505 md_parse_option (c, arg)
4506 int c;
4507 char *arg ATTRIBUTE_UNUSED;
4508 {
4509 switch (c)
4510 {
4511 case 'q':
4512 quiet_warnings = 1;
4513 break;
4514
4515 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4516 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
4517 should be emitted or not. FIXME: Not implemented. */
4518 case 'Q':
4519 break;
4520
4521 /* -V: SVR4 argument to print version ID. */
4522 case 'V':
4523 print_version_id ();
4524 break;
4525
4526 /* -k: Ignore for FreeBSD compatibility. */
4527 case 'k':
4528 break;
4529
4530 case 's':
4531 /* -s: On i386 Solaris, this tells the native assembler to use
4532 .stab instead of .stab.excl. We always use .stab anyhow. */
4533 break;
4534
4535 case OPTION_64:
4536 {
4537 const char **list, **l;
4538
4539 list = bfd_target_list ();
4540 for (l = list; *l != NULL; l++)
4541 if (strcmp (*l, "elf64-x86-64") == 0)
4542 {
4543 default_arch = "x86_64";
4544 break;
4545 }
4546 if (*l == NULL)
4547 as_fatal (_("No compiled in support for x86_64"));
4548 free (list);
4549 }
4550 break;
4551 #endif
4552
4553 case OPTION_32:
4554 default_arch = "i386";
4555 break;
4556
4557 default:
4558 return 0;
4559 }
4560 return 1;
4561 }
4562
4563 void
4564 md_show_usage (stream)
4565 FILE *stream;
4566 {
4567 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4568 fprintf (stream, _("\
4569 -Q ignored\n\
4570 -V print assembler version number\n\
4571 -k ignored\n\
4572 -q quieten some warnings\n\
4573 -s ignored\n"));
4574 #else
4575 fprintf (stream, _("\
4576 -q quieten some warnings\n"));
4577 #endif
4578 }
4579
4580 #ifdef BFD_ASSEMBLER
4581 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
4582 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
4583
4584 /* Pick the target format to use. */
4585
4586 const char *
4587 i386_target_format ()
4588 {
4589 if (!strcmp (default_arch, "x86_64"))
4590 set_code_flag (CODE_64BIT);
4591 else if (!strcmp (default_arch, "i386"))
4592 set_code_flag (CODE_32BIT);
4593 else
4594 as_fatal (_("Unknown architecture"));
4595 switch (OUTPUT_FLAVOR)
4596 {
4597 #ifdef OBJ_MAYBE_AOUT
4598 case bfd_target_aout_flavour:
4599 return AOUT_TARGET_FORMAT;
4600 #endif
4601 #ifdef OBJ_MAYBE_COFF
4602 case bfd_target_coff_flavour:
4603 return "coff-i386";
4604 #endif
4605 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
4606 case bfd_target_elf_flavour:
4607 {
4608 if (flag_code == CODE_64BIT)
4609 use_rela_relocations = 1;
4610 return flag_code == CODE_64BIT ? "elf64-x86-64" : "elf32-i386";
4611 }
4612 #endif
4613 default:
4614 abort ();
4615 return NULL;
4616 }
4617 }
4618
4619 #endif /* OBJ_MAYBE_ more than one */
4620 #endif /* BFD_ASSEMBLER */
4621 \f
4622 symbolS *
4623 md_undefined_symbol (name)
4624 char *name;
4625 {
4626 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
4627 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
4628 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
4629 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
4630 {
4631 if (!GOT_symbol)
4632 {
4633 if (symbol_find (name))
4634 as_bad (_("GOT already in symbol table"));
4635 GOT_symbol = symbol_new (name, undefined_section,
4636 (valueT) 0, &zero_address_frag);
4637 };
4638 return GOT_symbol;
4639 }
4640 return 0;
4641 }
4642
4643 /* Round up a section size to the appropriate boundary. */
4644
4645 valueT
4646 md_section_align (segment, size)
4647 segT segment ATTRIBUTE_UNUSED;
4648 valueT size;
4649 {
4650 #ifdef BFD_ASSEMBLER
4651 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4652 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
4653 {
4654 /* For a.out, force the section size to be aligned. If we don't do
4655 this, BFD will align it for us, but it will not write out the
4656 final bytes of the section. This may be a bug in BFD, but it is
4657 easier to fix it here since that is how the other a.out targets
4658 work. */
4659 int align;
4660
4661 align = bfd_get_section_alignment (stdoutput, segment);
4662 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
4663 }
4664 #endif
4665 #endif
4666
4667 return size;
4668 }
4669
4670 /* On the i386, PC-relative offsets are relative to the start of the
4671 next instruction. That is, the address of the offset, plus its
4672 size, since the offset is always the last part of the insn. */
4673
4674 long
4675 md_pcrel_from (fixP)
4676 fixS *fixP;
4677 {
4678 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
4679 }
4680
4681 #ifndef I386COFF
4682
4683 static void
4684 s_bss (ignore)
4685 int ignore ATTRIBUTE_UNUSED;
4686 {
4687 register int temp;
4688
4689 temp = get_absolute_expression ();
4690 subseg_set (bss_section, (subsegT) temp);
4691 demand_empty_rest_of_line ();
4692 }
4693
4694 #endif
4695
4696 #ifdef BFD_ASSEMBLER
4697
4698 void
4699 i386_validate_fix (fixp)
4700 fixS *fixp;
4701 {
4702 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
4703 {
4704 /* GOTOFF relocation are nonsense in 64bit mode. */
4705 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
4706 {
4707 if (flag_code != CODE_64BIT)
4708 abort ();
4709 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
4710 }
4711 else
4712 {
4713 if (flag_code == CODE_64BIT)
4714 abort ();
4715 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
4716 }
4717 fixp->fx_subsy = 0;
4718 }
4719 }
4720
4721 arelent *
4722 tc_gen_reloc (section, fixp)
4723 asection *section ATTRIBUTE_UNUSED;
4724 fixS *fixp;
4725 {
4726 arelent *rel;
4727 bfd_reloc_code_real_type code;
4728
4729 switch (fixp->fx_r_type)
4730 {
4731 case BFD_RELOC_X86_64_PLT32:
4732 case BFD_RELOC_X86_64_GOT32:
4733 case BFD_RELOC_X86_64_GOTPCREL:
4734 case BFD_RELOC_386_PLT32:
4735 case BFD_RELOC_386_GOT32:
4736 case BFD_RELOC_386_GOTOFF:
4737 case BFD_RELOC_386_GOTPC:
4738 case BFD_RELOC_X86_64_32S:
4739 case BFD_RELOC_RVA:
4740 case BFD_RELOC_VTABLE_ENTRY:
4741 case BFD_RELOC_VTABLE_INHERIT:
4742 code = fixp->fx_r_type;
4743 break;
4744 default:
4745 if (fixp->fx_pcrel)
4746 {
4747 switch (fixp->fx_size)
4748 {
4749 default:
4750 as_bad (_("can not do %d byte pc-relative relocation"),
4751 fixp->fx_size);
4752 code = BFD_RELOC_32_PCREL;
4753 break;
4754 case 1: code = BFD_RELOC_8_PCREL; break;
4755 case 2: code = BFD_RELOC_16_PCREL; break;
4756 case 4: code = BFD_RELOC_32_PCREL; break;
4757 }
4758 }
4759 else
4760 {
4761 switch (fixp->fx_size)
4762 {
4763 default:
4764 as_bad (_("can not do %d byte relocation"), fixp->fx_size);
4765 code = BFD_RELOC_32;
4766 break;
4767 case 1: code = BFD_RELOC_8; break;
4768 case 2: code = BFD_RELOC_16; break;
4769 case 4: code = BFD_RELOC_32; break;
4770 case 8: code = BFD_RELOC_64; break;
4771 }
4772 }
4773 break;
4774 }
4775
4776 if (code == BFD_RELOC_32
4777 && GOT_symbol
4778 && fixp->fx_addsy == GOT_symbol)
4779 {
4780 /* We don't support GOTPC on 64bit targets. */
4781 if (flag_code == CODE_64BIT)
4782 abort ();
4783 code = BFD_RELOC_386_GOTPC;
4784 }
4785
4786 rel = (arelent *) xmalloc (sizeof (arelent));
4787 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
4788 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
4789
4790 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
4791 if (!use_rela_relocations)
4792 {
4793 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
4794 vtable entry to be used in the relocation's section offset. */
4795 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
4796 rel->address = fixp->fx_offset;
4797
4798 if (fixp->fx_pcrel)
4799 rel->addend = fixp->fx_addnumber;
4800 else
4801 rel->addend = 0;
4802 }
4803 /* Use the rela in 64bit mode. */
4804 else
4805 {
4806 rel->addend = fixp->fx_offset;
4807 if (fixp->fx_pcrel)
4808 rel->addend -= fixp->fx_size;
4809 }
4810
4811 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
4812 if (rel->howto == NULL)
4813 {
4814 as_bad_where (fixp->fx_file, fixp->fx_line,
4815 _("cannot represent relocation type %s"),
4816 bfd_get_reloc_code_name (code));
4817 /* Set howto to a garbage value so that we can keep going. */
4818 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
4819 assert (rel->howto != NULL);
4820 }
4821
4822 return rel;
4823 }
4824
4825 #else /* ! BFD_ASSEMBLER */
4826
4827 #if (defined(OBJ_AOUT) | defined(OBJ_BOUT))
4828 void
4829 tc_aout_fix_to_chars (where, fixP, segment_address_in_file)
4830 char *where;
4831 fixS *fixP;
4832 relax_addressT segment_address_in_file;
4833 {
4834 /* In: length of relocation (or of address) in chars: 1, 2 or 4.
4835 Out: GNU LD relocation length code: 0, 1, or 2. */
4836
4837 static const unsigned char nbytes_r_length[] = { 42, 0, 1, 42, 2 };
4838 long r_symbolnum;
4839
4840 know (fixP->fx_addsy != NULL);
4841
4842 md_number_to_chars (where,
4843 (valueT) (fixP->fx_frag->fr_address
4844 + fixP->fx_where - segment_address_in_file),
4845 4);
4846
4847 r_symbolnum = (S_IS_DEFINED (fixP->fx_addsy)
4848 ? S_GET_TYPE (fixP->fx_addsy)
4849 : fixP->fx_addsy->sy_number);
4850
4851 where[6] = (r_symbolnum >> 16) & 0x0ff;
4852 where[5] = (r_symbolnum >> 8) & 0x0ff;
4853 where[4] = r_symbolnum & 0x0ff;
4854 where[7] = ((((!S_IS_DEFINED (fixP->fx_addsy)) << 3) & 0x08)
4855 | ((nbytes_r_length[fixP->fx_size] << 1) & 0x06)
4856 | (((fixP->fx_pcrel << 0) & 0x01) & 0x0f));
4857 }
4858
4859 #endif /* OBJ_AOUT or OBJ_BOUT. */
4860
4861 #if defined (I386COFF)
4862
4863 short
4864 tc_coff_fix2rtype (fixP)
4865 fixS *fixP;
4866 {
4867 if (fixP->fx_r_type == R_IMAGEBASE)
4868 return R_IMAGEBASE;
4869
4870 return (fixP->fx_pcrel ?
4871 (fixP->fx_size == 1 ? R_PCRBYTE :
4872 fixP->fx_size == 2 ? R_PCRWORD :
4873 R_PCRLONG) :
4874 (fixP->fx_size == 1 ? R_RELBYTE :
4875 fixP->fx_size == 2 ? R_RELWORD :
4876 R_DIR32));
4877 }
4878
4879 int
4880 tc_coff_sizemachdep (frag)
4881 fragS *frag;
4882 {
4883 if (frag->fr_next)
4884 return (frag->fr_next->fr_address - frag->fr_address);
4885 else
4886 return 0;
4887 }
4888
4889 #endif /* I386COFF */
4890
4891 #endif /* ! BFD_ASSEMBLER */
4892 \f
4893 /* Parse operands using Intel syntax. This implements a recursive descent
4894 parser based on the BNF grammar published in Appendix B of the MASM 6.1
4895 Programmer's Guide.
4896
4897 FIXME: We do not recognize the full operand grammar defined in the MASM
4898 documentation. In particular, all the structure/union and
4899 high-level macro operands are missing.
4900
4901 Uppercase words are terminals, lower case words are non-terminals.
4902 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
4903 bars '|' denote choices. Most grammar productions are implemented in
4904 functions called 'intel_<production>'.
4905
4906 Initial production is 'expr'.
4907
4908 addOp + | -
4909
4910 alpha [a-zA-Z]
4911
4912 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
4913
4914 constant digits [[ radixOverride ]]
4915
4916 dataType BYTE | WORD | DWORD | QWORD | XWORD
4917
4918 digits decdigit
4919 | digits decdigit
4920 | digits hexdigit
4921
4922 decdigit [0-9]
4923
4924 e05 e05 addOp e06
4925 | e06
4926
4927 e06 e06 mulOp e09
4928 | e09
4929
4930 e09 OFFSET e10
4931 | e09 PTR e10
4932 | e09 : e10
4933 | e10
4934
4935 e10 e10 [ expr ]
4936 | e11
4937
4938 e11 ( expr )
4939 | [ expr ]
4940 | constant
4941 | dataType
4942 | id
4943 | $
4944 | register
4945
4946 => expr SHORT e05
4947 | e05
4948
4949 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
4950 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
4951
4952 hexdigit a | b | c | d | e | f
4953 | A | B | C | D | E | F
4954
4955 id alpha
4956 | id alpha
4957 | id decdigit
4958
4959 mulOp * | / | MOD
4960
4961 quote " | '
4962
4963 register specialRegister
4964 | gpRegister
4965 | byteRegister
4966
4967 segmentRegister CS | DS | ES | FS | GS | SS
4968
4969 specialRegister CR0 | CR2 | CR3
4970 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
4971 | TR3 | TR4 | TR5 | TR6 | TR7
4972
4973 We simplify the grammar in obvious places (e.g., register parsing is
4974 done by calling parse_register) and eliminate immediate left recursion
4975 to implement a recursive-descent parser.
4976
4977 expr SHORT e05
4978 | e05
4979
4980 e05 e06 e05'
4981
4982 e05' addOp e06 e05'
4983 | Empty
4984
4985 e06 e09 e06'
4986
4987 e06' mulOp e09 e06'
4988 | Empty
4989
4990 e09 OFFSET e10 e09'
4991 | e10 e09'
4992
4993 e09' PTR e10 e09'
4994 | : e10 e09'
4995 | Empty
4996
4997 e10 e11 e10'
4998
4999 e10' [ expr ] e10'
5000 | Empty
5001
5002 e11 ( expr )
5003 | [ expr ]
5004 | BYTE
5005 | WORD
5006 | DWORD
5007 | QWORD
5008 | XWORD
5009 | .
5010 | $
5011 | register
5012 | id
5013 | constant */
5014
5015 /* Parsing structure for the intel syntax parser. Used to implement the
5016 semantic actions for the operand grammar. */
5017 struct intel_parser_s
5018 {
5019 char *op_string; /* The string being parsed. */
5020 int got_a_float; /* Whether the operand is a float. */
5021 int op_modifier; /* Operand modifier. */
5022 int is_mem; /* 1 if operand is memory reference. */
5023 const reg_entry *reg; /* Last register reference found. */
5024 char *disp; /* Displacement string being built. */
5025 };
5026
5027 static struct intel_parser_s intel_parser;
5028
5029 /* Token structure for parsing intel syntax. */
5030 struct intel_token
5031 {
5032 int code; /* Token code. */
5033 const reg_entry *reg; /* Register entry for register tokens. */
5034 char *str; /* String representation. */
5035 };
5036
5037 static struct intel_token cur_token, prev_token;
5038
5039 /* Token codes for the intel parser. Since T_SHORT is already used
5040 by COFF, undefine it first to prevent a warning. */
5041 #define T_NIL -1
5042 #define T_CONST 1
5043 #define T_REG 2
5044 #define T_BYTE 3
5045 #define T_WORD 4
5046 #define T_DWORD 5
5047 #define T_QWORD 6
5048 #define T_XWORD 7
5049 #undef T_SHORT
5050 #define T_SHORT 8
5051 #define T_OFFSET 9
5052 #define T_PTR 10
5053 #define T_ID 11
5054
5055 /* Prototypes for intel parser functions. */
5056 static int intel_match_token PARAMS ((int code));
5057 static void intel_get_token PARAMS ((void));
5058 static void intel_putback_token PARAMS ((void));
5059 static int intel_expr PARAMS ((void));
5060 static int intel_e05 PARAMS ((void));
5061 static int intel_e05_1 PARAMS ((void));
5062 static int intel_e06 PARAMS ((void));
5063 static int intel_e06_1 PARAMS ((void));
5064 static int intel_e09 PARAMS ((void));
5065 static int intel_e09_1 PARAMS ((void));
5066 static int intel_e10 PARAMS ((void));
5067 static int intel_e10_1 PARAMS ((void));
5068 static int intel_e11 PARAMS ((void));
5069
5070 static int
5071 i386_intel_operand (operand_string, got_a_float)
5072 char *operand_string;
5073 int got_a_float;
5074 {
5075 int ret;
5076 char *p;
5077
5078 /* Initialize token holders. */
5079 cur_token.code = prev_token.code = T_NIL;
5080 cur_token.reg = prev_token.reg = NULL;
5081 cur_token.str = prev_token.str = NULL;
5082
5083 /* Initialize parser structure. */
5084 p = intel_parser.op_string = (char *) malloc (strlen (operand_string) + 1);
5085 if (p == NULL)
5086 abort ();
5087 strcpy (intel_parser.op_string, operand_string);
5088 intel_parser.got_a_float = got_a_float;
5089 intel_parser.op_modifier = -1;
5090 intel_parser.is_mem = 0;
5091 intel_parser.reg = NULL;
5092 intel_parser.disp = (char *) malloc (strlen (operand_string) + 1);
5093 if (intel_parser.disp == NULL)
5094 abort ();
5095 intel_parser.disp[0] = '\0';
5096
5097 /* Read the first token and start the parser. */
5098 intel_get_token ();
5099 ret = intel_expr ();
5100
5101 if (ret)
5102 {
5103 /* If we found a memory reference, hand it over to i386_displacement
5104 to fill in the rest of the operand fields. */
5105 if (intel_parser.is_mem)
5106 {
5107 if ((i.mem_operands == 1
5108 && (current_templates->start->opcode_modifier & IsString) == 0)
5109 || i.mem_operands == 2)
5110 {
5111 as_bad (_("too many memory references for '%s'"),
5112 current_templates->start->name);
5113 ret = 0;
5114 }
5115 else
5116 {
5117 char *s = intel_parser.disp;
5118 i.mem_operands++;
5119
5120 /* Add the displacement expression. */
5121 if (*s != '\0')
5122 ret = i386_displacement (s, s + strlen (s))
5123 && i386_index_check (s);
5124 }
5125 }
5126
5127 /* Constant and OFFSET expressions are handled by i386_immediate. */
5128 else if (intel_parser.op_modifier == OFFSET_FLAT
5129 || intel_parser.reg == NULL)
5130 ret = i386_immediate (intel_parser.disp);
5131 }
5132
5133 free (p);
5134 free (intel_parser.disp);
5135
5136 return ret;
5137 }
5138
5139 /* expr SHORT e05
5140 | e05 */
5141 static int
5142 intel_expr ()
5143 {
5144 /* expr SHORT e05 */
5145 if (cur_token.code == T_SHORT)
5146 {
5147 intel_parser.op_modifier = SHORT;
5148 intel_match_token (T_SHORT);
5149
5150 return (intel_e05 ());
5151 }
5152
5153 /* expr e05 */
5154 else
5155 return intel_e05 ();
5156 }
5157
5158 /* e05 e06 e05'
5159
5160 e05' addOp e06 e05'
5161 | Empty */
5162 static int
5163 intel_e05 ()
5164 {
5165 return (intel_e06 () && intel_e05_1 ());
5166 }
5167
5168 static int
5169 intel_e05_1 ()
5170 {
5171 /* e05' addOp e06 e05' */
5172 if (cur_token.code == '+' || cur_token.code == '-')
5173 {
5174 strcat (intel_parser.disp, cur_token.str);
5175 intel_match_token (cur_token.code);
5176
5177 return (intel_e06 () && intel_e05_1 ());
5178 }
5179
5180 /* e05' Empty */
5181 else
5182 return 1;
5183 }
5184
5185 /* e06 e09 e06'
5186
5187 e06' mulOp e09 e06'
5188 | Empty */
5189 static int
5190 intel_e06 ()
5191 {
5192 return (intel_e09 () && intel_e06_1 ());
5193 }
5194
5195 static int
5196 intel_e06_1 ()
5197 {
5198 /* e06' mulOp e09 e06' */
5199 if (cur_token.code == '*' || cur_token.code == '/')
5200 {
5201 strcat (intel_parser.disp, cur_token.str);
5202 intel_match_token (cur_token.code);
5203
5204 return (intel_e09 () && intel_e06_1 ());
5205 }
5206
5207 /* e06' Empty */
5208 else
5209 return 1;
5210 }
5211
5212 /* e09 OFFSET e10 e09'
5213 | e10 e09'
5214
5215 e09' PTR e10 e09'
5216 | : e10 e09'
5217 | Empty */
5218 static int
5219 intel_e09 ()
5220 {
5221 /* e09 OFFSET e10 e09' */
5222 if (cur_token.code == T_OFFSET)
5223 {
5224 intel_parser.is_mem = 0;
5225 intel_parser.op_modifier = OFFSET_FLAT;
5226 intel_match_token (T_OFFSET);
5227
5228 return (intel_e10 () && intel_e09_1 ());
5229 }
5230
5231 /* e09 e10 e09' */
5232 else
5233 return (intel_e10 () && intel_e09_1 ());
5234 }
5235
5236 static int
5237 intel_e09_1 ()
5238 {
5239 /* e09' PTR e10 e09' */
5240 if (cur_token.code == T_PTR)
5241 {
5242 if (prev_token.code == T_BYTE)
5243 i.suffix = BYTE_MNEM_SUFFIX;
5244
5245 else if (prev_token.code == T_WORD)
5246 {
5247 if (intel_parser.got_a_float == 2) /* "fi..." */
5248 i.suffix = SHORT_MNEM_SUFFIX;
5249 else
5250 i.suffix = WORD_MNEM_SUFFIX;
5251 }
5252
5253 else if (prev_token.code == T_DWORD)
5254 {
5255 if (intel_parser.got_a_float == 1) /* "f..." */
5256 i.suffix = SHORT_MNEM_SUFFIX;
5257 else
5258 i.suffix = LONG_MNEM_SUFFIX;
5259 }
5260
5261 else if (prev_token.code == T_QWORD)
5262 {
5263 if (intel_parser.got_a_float == 1) /* "f..." */
5264 i.suffix = LONG_MNEM_SUFFIX;
5265 else
5266 i.suffix = QWORD_MNEM_SUFFIX;
5267 }
5268
5269 else if (prev_token.code == T_XWORD)
5270 i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
5271
5272 else
5273 {
5274 as_bad (_("Unknown operand modifier `%s'\n"), prev_token.str);
5275 return 0;
5276 }
5277
5278 intel_match_token (T_PTR);
5279
5280 return (intel_e10 () && intel_e09_1 ());
5281 }
5282
5283 /* e09 : e10 e09' */
5284 else if (cur_token.code == ':')
5285 {
5286 /* Mark as a memory operand only if it's not already known to be an
5287 offset expression. */
5288 if (intel_parser.op_modifier != OFFSET_FLAT)
5289 intel_parser.is_mem = 1;
5290
5291 return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
5292 }
5293
5294 /* e09' Empty */
5295 else
5296 return 1;
5297 }
5298
5299 /* e10 e11 e10'
5300
5301 e10' [ expr ] e10'
5302 | Empty */
5303 static int
5304 intel_e10 ()
5305 {
5306 return (intel_e11 () && intel_e10_1 ());
5307 }
5308
5309 static int
5310 intel_e10_1 ()
5311 {
5312 /* e10' [ expr ] e10' */
5313 if (cur_token.code == '[')
5314 {
5315 intel_match_token ('[');
5316
5317 /* Mark as a memory operand only if it's not already known to be an
5318 offset expression. If it's an offset expression, we need to keep
5319 the brace in. */
5320 if (intel_parser.op_modifier != OFFSET_FLAT)
5321 intel_parser.is_mem = 1;
5322 else
5323 strcat (intel_parser.disp, "[");
5324
5325 /* Add a '+' to the displacement string if necessary. */
5326 if (*intel_parser.disp != '\0'
5327 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5328 strcat (intel_parser.disp, "+");
5329
5330 if (intel_expr () && intel_match_token (']'))
5331 {
5332 /* Preserve brackets when the operand is an offset expression. */
5333 if (intel_parser.op_modifier == OFFSET_FLAT)
5334 strcat (intel_parser.disp, "]");
5335
5336 return intel_e10_1 ();
5337 }
5338 else
5339 return 0;
5340 }
5341
5342 /* e10' Empty */
5343 else
5344 return 1;
5345 }
5346
5347 /* e11 ( expr )
5348 | [ expr ]
5349 | BYTE
5350 | WORD
5351 | DWORD
5352 | QWORD
5353 | XWORD
5354 | $
5355 | .
5356 | register
5357 | id
5358 | constant */
5359 static int
5360 intel_e11 ()
5361 {
5362 /* e11 ( expr ) */
5363 if (cur_token.code == '(')
5364 {
5365 intel_match_token ('(');
5366 strcat (intel_parser.disp, "(");
5367
5368 if (intel_expr () && intel_match_token (')'))
5369 {
5370 strcat (intel_parser.disp, ")");
5371 return 1;
5372 }
5373 else
5374 return 0;
5375 }
5376
5377 /* e11 [ expr ] */
5378 else if (cur_token.code == '[')
5379 {
5380 intel_match_token ('[');
5381
5382 /* Mark as a memory operand only if it's not already known to be an
5383 offset expression. If it's an offset expression, we need to keep
5384 the brace in. */
5385 if (intel_parser.op_modifier != OFFSET_FLAT)
5386 intel_parser.is_mem = 1;
5387 else
5388 strcat (intel_parser.disp, "[");
5389
5390 /* Operands for jump/call inside brackets denote absolute addresses. */
5391 if (current_templates->start->opcode_modifier & Jump
5392 || current_templates->start->opcode_modifier & JumpDword
5393 || current_templates->start->opcode_modifier & JumpByte
5394 || current_templates->start->opcode_modifier & JumpInterSegment)
5395 i.types[this_operand] |= JumpAbsolute;
5396
5397 /* Add a '+' to the displacement string if necessary. */
5398 if (*intel_parser.disp != '\0'
5399 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5400 strcat (intel_parser.disp, "+");
5401
5402 if (intel_expr () && intel_match_token (']'))
5403 {
5404 /* Preserve brackets when the operand is an offset expression. */
5405 if (intel_parser.op_modifier == OFFSET_FLAT)
5406 strcat (intel_parser.disp, "]");
5407
5408 return 1;
5409 }
5410 else
5411 return 0;
5412 }
5413
5414 /* e11 BYTE
5415 | WORD
5416 | DWORD
5417 | QWORD
5418 | XWORD */
5419 else if (cur_token.code == T_BYTE
5420 || cur_token.code == T_WORD
5421 || cur_token.code == T_DWORD
5422 || cur_token.code == T_QWORD
5423 || cur_token.code == T_XWORD)
5424 {
5425 intel_match_token (cur_token.code);
5426
5427 return 1;
5428 }
5429
5430 /* e11 $
5431 | . */
5432 else if (cur_token.code == '$' || cur_token.code == '.')
5433 {
5434 strcat (intel_parser.disp, cur_token.str);
5435 intel_match_token (cur_token.code);
5436
5437 /* Mark as a memory operand only if it's not already known to be an
5438 offset expression. */
5439 if (intel_parser.op_modifier != OFFSET_FLAT)
5440 intel_parser.is_mem = 1;
5441
5442 return 1;
5443 }
5444
5445 /* e11 register */
5446 else if (cur_token.code == T_REG)
5447 {
5448 const reg_entry *reg = intel_parser.reg = cur_token.reg;
5449
5450 intel_match_token (T_REG);
5451
5452 /* Check for segment change. */
5453 if (cur_token.code == ':')
5454 {
5455 if (reg->reg_type & (SReg2 | SReg3))
5456 {
5457 switch (reg->reg_num)
5458 {
5459 case 0:
5460 i.seg[i.mem_operands] = &es;
5461 break;
5462 case 1:
5463 i.seg[i.mem_operands] = &cs;
5464 break;
5465 case 2:
5466 i.seg[i.mem_operands] = &ss;
5467 break;
5468 case 3:
5469 i.seg[i.mem_operands] = &ds;
5470 break;
5471 case 4:
5472 i.seg[i.mem_operands] = &fs;
5473 break;
5474 case 5:
5475 i.seg[i.mem_operands] = &gs;
5476 break;
5477 }
5478 }
5479 else
5480 {
5481 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
5482 return 0;
5483 }
5484 }
5485
5486 /* Not a segment register. Check for register scaling. */
5487 else if (cur_token.code == '*')
5488 {
5489 if (!intel_parser.is_mem)
5490 {
5491 as_bad (_("Register scaling only allowed in memory operands."));
5492 return 0;
5493 }
5494
5495 /* What follows must be a valid scale. */
5496 if (intel_match_token ('*')
5497 && strchr ("01248", *cur_token.str))
5498 {
5499 i.index_reg = reg;
5500 i.types[this_operand] |= BaseIndex;
5501
5502 /* Set the scale after setting the register (otherwise,
5503 i386_scale will complain) */
5504 i386_scale (cur_token.str);
5505 intel_match_token (T_CONST);
5506 }
5507 else
5508 {
5509 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
5510 cur_token.str);
5511 return 0;
5512 }
5513 }
5514
5515 /* No scaling. If this is a memory operand, the register is either a
5516 base register (first occurrence) or an index register (second
5517 occurrence). */
5518 else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3)))
5519 {
5520 if (i.base_reg && i.index_reg)
5521 {
5522 as_bad (_("Too many register references in memory operand.\n"));
5523 return 0;
5524 }
5525
5526 if (i.base_reg == NULL)
5527 i.base_reg = reg;
5528 else
5529 i.index_reg = reg;
5530
5531 i.types[this_operand] |= BaseIndex;
5532 }
5533
5534 /* Offset modifier. Add the register to the displacement string to be
5535 parsed as an immediate expression after we're done. */
5536 else if (intel_parser.op_modifier == OFFSET_FLAT)
5537 strcat (intel_parser.disp, reg->reg_name);
5538
5539 /* It's neither base nor index nor offset. */
5540 else
5541 {
5542 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
5543 i.op[this_operand].regs = reg;
5544 i.reg_operands++;
5545 }
5546
5547 /* Since registers are not part of the displacement string (except
5548 when we're parsing offset operands), we may need to remove any
5549 preceding '+' from the displacement string. */
5550 if (*intel_parser.disp != '\0'
5551 && intel_parser.op_modifier != OFFSET_FLAT)
5552 {
5553 char *s = intel_parser.disp;
5554 s += strlen (s) - 1;
5555 if (*s == '+')
5556 *s = '\0';
5557 }
5558
5559 return 1;
5560 }
5561
5562 /* e11 id */
5563 else if (cur_token.code == T_ID)
5564 {
5565 /* Add the identifier to the displacement string. */
5566 strcat (intel_parser.disp, cur_token.str);
5567 intel_match_token (T_ID);
5568
5569 /* The identifier represents a memory reference only if it's not
5570 preceded by an offset modifier. */
5571 if (intel_parser.op_modifier != OFFSET_FLAT)
5572 intel_parser.is_mem = 1;
5573
5574 return 1;
5575 }
5576
5577 /* e11 constant */
5578 else if (cur_token.code == T_CONST
5579 || cur_token.code == '-'
5580 || cur_token.code == '+')
5581 {
5582 char *save_str;
5583
5584 /* Allow constants that start with `+' or `-'. */
5585 if (cur_token.code == '-' || cur_token.code == '+')
5586 {
5587 strcat (intel_parser.disp, cur_token.str);
5588 intel_match_token (cur_token.code);
5589 if (cur_token.code != T_CONST)
5590 {
5591 as_bad (_("Syntax error. Expecting a constant. Got `%s'.\n"),
5592 cur_token.str);
5593 return 0;
5594 }
5595 }
5596
5597 save_str = (char *) malloc (strlen (cur_token.str) + 1);
5598 if (save_str == NULL)
5599 abort ();
5600 strcpy (save_str, cur_token.str);
5601
5602 /* Get the next token to check for register scaling. */
5603 intel_match_token (cur_token.code);
5604
5605 /* Check if this constant is a scaling factor for an index register. */
5606 if (cur_token.code == '*')
5607 {
5608 if (intel_match_token ('*') && cur_token.code == T_REG)
5609 {
5610 if (!intel_parser.is_mem)
5611 {
5612 as_bad (_("Register scaling only allowed in memory operands."));
5613 return 0;
5614 }
5615
5616 /* The constant is followed by `* reg', so it must be
5617 a valid scale. */
5618 if (strchr ("01248", *save_str))
5619 {
5620 i.index_reg = cur_token.reg;
5621 i.types[this_operand] |= BaseIndex;
5622
5623 /* Set the scale after setting the register (otherwise,
5624 i386_scale will complain) */
5625 i386_scale (save_str);
5626 intel_match_token (T_REG);
5627
5628 /* Since registers are not part of the displacement
5629 string, we may need to remove any preceding '+' from
5630 the displacement string. */
5631 if (*intel_parser.disp != '\0')
5632 {
5633 char *s = intel_parser.disp;
5634 s += strlen (s) - 1;
5635 if (*s == '+')
5636 *s = '\0';
5637 }
5638
5639 free (save_str);
5640
5641 return 1;
5642 }
5643 else
5644 return 0;
5645 }
5646
5647 /* The constant was not used for register scaling. Since we have
5648 already consumed the token following `*' we now need to put it
5649 back in the stream. */
5650 else
5651 intel_putback_token ();
5652 }
5653
5654 /* Add the constant to the displacement string. */
5655 strcat (intel_parser.disp, save_str);
5656 free (save_str);
5657
5658 return 1;
5659 }
5660
5661 as_bad (_("Unrecognized token '%s'"), cur_token.str);
5662 return 0;
5663 }
5664
5665 /* Match the given token against cur_token. If they match, read the next
5666 token from the operand string. */
5667 static int
5668 intel_match_token (code)
5669 int code;
5670 {
5671 if (cur_token.code == code)
5672 {
5673 intel_get_token ();
5674 return 1;
5675 }
5676 else
5677 {
5678 as_bad (_("Unexpected token `%s'\n"), cur_token.str);
5679 return 0;
5680 }
5681 }
5682
5683 /* Read a new token from intel_parser.op_string and store it in cur_token. */
5684 static void
5685 intel_get_token ()
5686 {
5687 char *end_op;
5688 const reg_entry *reg;
5689 struct intel_token new_token;
5690
5691 new_token.code = T_NIL;
5692 new_token.reg = NULL;
5693 new_token.str = NULL;
5694
5695 /* Free the memory allocated to the previous token and move
5696 cur_token to prev_token. */
5697 if (prev_token.str)
5698 free (prev_token.str);
5699
5700 prev_token = cur_token;
5701
5702 /* Skip whitespace. */
5703 while (is_space_char (*intel_parser.op_string))
5704 intel_parser.op_string++;
5705
5706 /* Return an empty token if we find nothing else on the line. */
5707 if (*intel_parser.op_string == '\0')
5708 {
5709 cur_token = new_token;
5710 return;
5711 }
5712
5713 /* The new token cannot be larger than the remainder of the operand
5714 string. */
5715 new_token.str = (char *) malloc (strlen (intel_parser.op_string) + 1);
5716 if (new_token.str == NULL)
5717 abort ();
5718 new_token.str[0] = '\0';
5719
5720 if (strchr ("0123456789", *intel_parser.op_string))
5721 {
5722 char *p = new_token.str;
5723 char *q = intel_parser.op_string;
5724 new_token.code = T_CONST;
5725
5726 /* Allow any kind of identifier char to encompass floating point and
5727 hexadecimal numbers. */
5728 while (is_identifier_char (*q))
5729 *p++ = *q++;
5730 *p = '\0';
5731
5732 /* Recognize special symbol names [0-9][bf]. */
5733 if (strlen (intel_parser.op_string) == 2
5734 && (intel_parser.op_string[1] == 'b'
5735 || intel_parser.op_string[1] == 'f'))
5736 new_token.code = T_ID;
5737 }
5738
5739 else if (strchr ("+-/*:[]()", *intel_parser.op_string))
5740 {
5741 new_token.code = *intel_parser.op_string;
5742 new_token.str[0] = *intel_parser.op_string;
5743 new_token.str[1] = '\0';
5744 }
5745
5746 else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
5747 && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
5748 {
5749 new_token.code = T_REG;
5750 new_token.reg = reg;
5751
5752 if (*intel_parser.op_string == REGISTER_PREFIX)
5753 {
5754 new_token.str[0] = REGISTER_PREFIX;
5755 new_token.str[1] = '\0';
5756 }
5757
5758 strcat (new_token.str, reg->reg_name);
5759 }
5760
5761 else if (is_identifier_char (*intel_parser.op_string))
5762 {
5763 char *p = new_token.str;
5764 char *q = intel_parser.op_string;
5765
5766 /* A '.' or '$' followed by an identifier char is an identifier.
5767 Otherwise, it's operator '.' followed by an expression. */
5768 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
5769 {
5770 new_token.code = *q;
5771 new_token.str[0] = *q;
5772 new_token.str[1] = '\0';
5773 }
5774 else
5775 {
5776 while (is_identifier_char (*q) || *q == '@')
5777 *p++ = *q++;
5778 *p = '\0';
5779
5780 if (strcasecmp (new_token.str, "BYTE") == 0)
5781 new_token.code = T_BYTE;
5782
5783 else if (strcasecmp (new_token.str, "WORD") == 0)
5784 new_token.code = T_WORD;
5785
5786 else if (strcasecmp (new_token.str, "DWORD") == 0)
5787 new_token.code = T_DWORD;
5788
5789 else if (strcasecmp (new_token.str, "QWORD") == 0)
5790 new_token.code = T_QWORD;
5791
5792 else if (strcasecmp (new_token.str, "XWORD") == 0)
5793 new_token.code = T_XWORD;
5794
5795 else if (strcasecmp (new_token.str, "PTR") == 0)
5796 new_token.code = T_PTR;
5797
5798 else if (strcasecmp (new_token.str, "SHORT") == 0)
5799 new_token.code = T_SHORT;
5800
5801 else if (strcasecmp (new_token.str, "OFFSET") == 0)
5802 {
5803 new_token.code = T_OFFSET;
5804
5805 /* ??? This is not mentioned in the MASM grammar but gcc
5806 makes use of it with -mintel-syntax. OFFSET may be
5807 followed by FLAT: */
5808 if (strncasecmp (q, " FLAT:", 6) == 0)
5809 strcat (new_token.str, " FLAT:");
5810 }
5811
5812 /* ??? This is not mentioned in the MASM grammar. */
5813 else if (strcasecmp (new_token.str, "FLAT") == 0)
5814 new_token.code = T_OFFSET;
5815
5816 else
5817 new_token.code = T_ID;
5818 }
5819 }
5820
5821 else
5822 as_bad (_("Unrecognized token `%s'\n"), intel_parser.op_string);
5823
5824 intel_parser.op_string += strlen (new_token.str);
5825 cur_token = new_token;
5826 }
5827
5828 /* Put cur_token back into the token stream and make cur_token point to
5829 prev_token. */
5830 static void
5831 intel_putback_token ()
5832 {
5833 intel_parser.op_string -= strlen (cur_token.str);
5834 free (cur_token.str);
5835 cur_token = prev_token;
5836
5837 /* Forget prev_token. */
5838 prev_token.code = T_NIL;
5839 prev_token.reg = NULL;
5840 prev_token.str = NULL;
5841 }
This page took 0.143177 seconds and 5 git commands to generate.