* expr.c (expr): Move code setting "retval" to the end of the loop,
[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) && !defined(TE_NetBSD))
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) && !defined(TE_NetBSD))
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 && *l != ',')
1274 {
1275 as_bad (_("invalid character %s in mnemonic"),
1276 output_invalid (*l));
1277 return;
1278 }
1279 if (token_start == l)
1280 {
1281 if (*l == PREFIX_SEPARATOR)
1282 as_bad (_("expecting prefix; got nothing"));
1283 else
1284 as_bad (_("expecting mnemonic; got nothing"));
1285 return;
1286 }
1287
1288 /* Look up instruction (or prefix) via hash table. */
1289 current_templates = hash_find (op_hash, mnemonic);
1290
1291 if (*l != END_OF_INSN
1292 && (! is_space_char (*l) || l[1] != END_OF_INSN)
1293 && current_templates
1294 && (current_templates->start->opcode_modifier & IsPrefix))
1295 {
1296 /* If we are in 16-bit mode, do not allow addr16 or data16.
1297 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1298 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1299 && (((current_templates->start->opcode_modifier & Size32) != 0)
1300 ^ (flag_code == CODE_16BIT)))
1301 {
1302 as_bad (_("redundant %s prefix"),
1303 current_templates->start->name);
1304 return;
1305 }
1306 /* Add prefix, checking for repeated prefixes. */
1307 switch (add_prefix (current_templates->start->base_opcode))
1308 {
1309 case 0:
1310 return;
1311 case 2:
1312 expecting_string_instruction = current_templates->start->name;
1313 break;
1314 }
1315 /* Skip past PREFIX_SEPARATOR and reset token_start. */
1316 token_start = ++l;
1317 }
1318 else
1319 break;
1320 }
1321
1322 if (!current_templates)
1323 {
1324 /* See if we can get a match by trimming off a suffix. */
1325 switch (mnem_p[-1])
1326 {
1327 case WORD_MNEM_SUFFIX:
1328 case BYTE_MNEM_SUFFIX:
1329 case QWORD_MNEM_SUFFIX:
1330 i.suffix = mnem_p[-1];
1331 mnem_p[-1] = '\0';
1332 current_templates = hash_find (op_hash, mnemonic);
1333 break;
1334 case SHORT_MNEM_SUFFIX:
1335 case LONG_MNEM_SUFFIX:
1336 if (!intel_syntax)
1337 {
1338 i.suffix = mnem_p[-1];
1339 mnem_p[-1] = '\0';
1340 current_templates = hash_find (op_hash, mnemonic);
1341 }
1342 break;
1343
1344 /* Intel Syntax. */
1345 case 'd':
1346 if (intel_syntax)
1347 {
1348 if (intel_float_operand (mnemonic))
1349 i.suffix = SHORT_MNEM_SUFFIX;
1350 else
1351 i.suffix = LONG_MNEM_SUFFIX;
1352 mnem_p[-1] = '\0';
1353 current_templates = hash_find (op_hash, mnemonic);
1354 }
1355 break;
1356 }
1357 if (!current_templates)
1358 {
1359 as_bad (_("no such instruction: `%s'"), token_start);
1360 return;
1361 }
1362 }
1363
1364 if (current_templates->start->opcode_modifier & (Jump | JumpByte))
1365 {
1366 /* Check for a branch hint. We allow ",pt" and ",pn" for
1367 predict taken and predict not taken respectively.
1368 I'm not sure that branch hints actually do anything on loop
1369 and jcxz insns (JumpByte) for current Pentium4 chips. They
1370 may work in the future and it doesn't hurt to accept them
1371 now. */
1372 if (l[0] == ',' && l[1] == 'p')
1373 {
1374 if (l[2] == 't')
1375 {
1376 if (! add_prefix (DS_PREFIX_OPCODE))
1377 return;
1378 l += 3;
1379 }
1380 else if (l[2] == 'n')
1381 {
1382 if (! add_prefix (CS_PREFIX_OPCODE))
1383 return;
1384 l += 3;
1385 }
1386 }
1387 }
1388 /* Any other comma loses. */
1389 if (*l == ',')
1390 {
1391 as_bad (_("invalid character %s in mnemonic"),
1392 output_invalid (*l));
1393 return;
1394 }
1395
1396 /* Check if instruction is supported on specified architecture. */
1397 if (cpu_arch_flags != 0)
1398 {
1399 if ((current_templates->start->cpu_flags & ~(Cpu64 | CpuNo64))
1400 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64)))
1401 {
1402 as_warn (_("`%s' is not supported on `%s'"),
1403 current_templates->start->name, cpu_arch_name);
1404 }
1405 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
1406 {
1407 as_warn (_("use .code16 to ensure correct addressing mode"));
1408 }
1409 }
1410
1411 /* Check for rep/repne without a string instruction. */
1412 if (expecting_string_instruction
1413 && !(current_templates->start->opcode_modifier & IsString))
1414 {
1415 as_bad (_("expecting string instruction after `%s'"),
1416 expecting_string_instruction);
1417 return;
1418 }
1419
1420 /* There may be operands to parse. */
1421 if (*l != END_OF_INSN)
1422 {
1423 /* 1 if operand is pending after ','. */
1424 unsigned int expecting_operand = 0;
1425
1426 /* Non-zero if operand parens not balanced. */
1427 unsigned int paren_not_balanced;
1428
1429 do
1430 {
1431 /* Skip optional white space before operand. */
1432 if (is_space_char (*l))
1433 ++l;
1434 if (!is_operand_char (*l) && *l != END_OF_INSN)
1435 {
1436 as_bad (_("invalid character %s before operand %d"),
1437 output_invalid (*l),
1438 i.operands + 1);
1439 return;
1440 }
1441 token_start = l; /* after white space */
1442 paren_not_balanced = 0;
1443 while (paren_not_balanced || *l != ',')
1444 {
1445 if (*l == END_OF_INSN)
1446 {
1447 if (paren_not_balanced)
1448 {
1449 if (!intel_syntax)
1450 as_bad (_("unbalanced parenthesis in operand %d."),
1451 i.operands + 1);
1452 else
1453 as_bad (_("unbalanced brackets in operand %d."),
1454 i.operands + 1);
1455 return;
1456 }
1457 else
1458 break; /* we are done */
1459 }
1460 else if (!is_operand_char (*l) && !is_space_char (*l))
1461 {
1462 as_bad (_("invalid character %s in operand %d"),
1463 output_invalid (*l),
1464 i.operands + 1);
1465 return;
1466 }
1467 if (!intel_syntax)
1468 {
1469 if (*l == '(')
1470 ++paren_not_balanced;
1471 if (*l == ')')
1472 --paren_not_balanced;
1473 }
1474 else
1475 {
1476 if (*l == '[')
1477 ++paren_not_balanced;
1478 if (*l == ']')
1479 --paren_not_balanced;
1480 }
1481 l++;
1482 }
1483 if (l != token_start)
1484 { /* Yes, we've read in another operand. */
1485 unsigned int operand_ok;
1486 this_operand = i.operands++;
1487 if (i.operands > MAX_OPERANDS)
1488 {
1489 as_bad (_("spurious operands; (%d operands/instruction max)"),
1490 MAX_OPERANDS);
1491 return;
1492 }
1493 /* Now parse operand adding info to 'i' as we go along. */
1494 END_STRING_AND_SAVE (l);
1495
1496 if (intel_syntax)
1497 operand_ok =
1498 i386_intel_operand (token_start,
1499 intel_float_operand (mnemonic));
1500 else
1501 operand_ok = i386_operand (token_start);
1502
1503 RESTORE_END_STRING (l);
1504 if (!operand_ok)
1505 return;
1506 }
1507 else
1508 {
1509 if (expecting_operand)
1510 {
1511 expecting_operand_after_comma:
1512 as_bad (_("expecting operand after ','; got nothing"));
1513 return;
1514 }
1515 if (*l == ',')
1516 {
1517 as_bad (_("expecting operand before ','; got nothing"));
1518 return;
1519 }
1520 }
1521
1522 /* Now *l must be either ',' or END_OF_INSN. */
1523 if (*l == ',')
1524 {
1525 if (*++l == END_OF_INSN)
1526 {
1527 /* Just skip it, if it's \n complain. */
1528 goto expecting_operand_after_comma;
1529 }
1530 expecting_operand = 1;
1531 }
1532 }
1533 while (*l != END_OF_INSN);
1534 }
1535 }
1536
1537 /* Now we've parsed the mnemonic into a set of templates, and have the
1538 operands at hand.
1539
1540 Next, we find a template that matches the given insn,
1541 making sure the overlap of the given operands types is consistent
1542 with the template operand types. */
1543
1544 #define MATCH(overlap, given, template) \
1545 ((overlap & ~JumpAbsolute) \
1546 && ((given) & (BaseIndex|JumpAbsolute)) == ((overlap) & (BaseIndex|JumpAbsolute)))
1547
1548 /* If given types r0 and r1 are registers they must be of the same type
1549 unless the expected operand type register overlap is null.
1550 Note that Acc in a template matches every size of reg. */
1551 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
1552 ( ((g0) & Reg) == 0 || ((g1) & Reg) == 0 || \
1553 ((g0) & Reg) == ((g1) & Reg) || \
1554 ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
1555
1556 {
1557 register unsigned int overlap0, overlap1;
1558 unsigned int overlap2;
1559 unsigned int found_reverse_match;
1560 int suffix_check;
1561
1562 /* All intel opcodes have reversed operands except for "bound" and
1563 "enter". We also don't reverse intersegment "jmp" and "call"
1564 instructions with 2 immediate operands so that the immediate segment
1565 precedes the offset, as it does when in AT&T mode. "enter" and the
1566 intersegment "jmp" and "call" instructions are the only ones that
1567 have two immediate operands. */
1568 if (intel_syntax && i.operands > 1
1569 && (strcmp (mnemonic, "bound") != 0)
1570 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1571 {
1572 union i386_op temp_op;
1573 unsigned int temp_type;
1574 RELOC_ENUM temp_reloc;
1575 int xchg1 = 0;
1576 int xchg2 = 0;
1577
1578 if (i.operands == 2)
1579 {
1580 xchg1 = 0;
1581 xchg2 = 1;
1582 }
1583 else if (i.operands == 3)
1584 {
1585 xchg1 = 0;
1586 xchg2 = 2;
1587 }
1588 temp_type = i.types[xchg2];
1589 i.types[xchg2] = i.types[xchg1];
1590 i.types[xchg1] = temp_type;
1591 temp_op = i.op[xchg2];
1592 i.op[xchg2] = i.op[xchg1];
1593 i.op[xchg1] = temp_op;
1594 temp_reloc = i.reloc[xchg2];
1595 i.reloc[xchg2] = i.reloc[xchg1];
1596 i.reloc[xchg1] = temp_reloc;
1597
1598 if (i.mem_operands == 2)
1599 {
1600 const seg_entry *temp_seg;
1601 temp_seg = i.seg[0];
1602 i.seg[0] = i.seg[1];
1603 i.seg[1] = temp_seg;
1604 }
1605 }
1606
1607 if (i.imm_operands)
1608 {
1609 /* Try to ensure constant immediates are represented in the smallest
1610 opcode possible. */
1611 char guess_suffix = 0;
1612 int op;
1613
1614 if (i.suffix)
1615 guess_suffix = i.suffix;
1616 else if (i.reg_operands)
1617 {
1618 /* Figure out a suffix from the last register operand specified.
1619 We can't do this properly yet, ie. excluding InOutPortReg,
1620 but the following works for instructions with immediates.
1621 In any case, we can't set i.suffix yet. */
1622 for (op = i.operands; --op >= 0;)
1623 if (i.types[op] & Reg)
1624 {
1625 if (i.types[op] & Reg8)
1626 guess_suffix = BYTE_MNEM_SUFFIX;
1627 else if (i.types[op] & Reg16)
1628 guess_suffix = WORD_MNEM_SUFFIX;
1629 else if (i.types[op] & Reg32)
1630 guess_suffix = LONG_MNEM_SUFFIX;
1631 else if (i.types[op] & Reg64)
1632 guess_suffix = QWORD_MNEM_SUFFIX;
1633 break;
1634 }
1635 }
1636 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
1637 guess_suffix = WORD_MNEM_SUFFIX;
1638
1639 for (op = i.operands; --op >= 0;)
1640 if (i.types[op] & Imm)
1641 {
1642 switch (i.op[op].imms->X_op)
1643 {
1644 case O_constant:
1645 /* If a suffix is given, this operand may be shortened. */
1646 switch (guess_suffix)
1647 {
1648 case LONG_MNEM_SUFFIX:
1649 i.types[op] |= Imm32 | Imm64;
1650 break;
1651 case WORD_MNEM_SUFFIX:
1652 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
1653 break;
1654 case BYTE_MNEM_SUFFIX:
1655 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
1656 break;
1657 }
1658
1659 /* If this operand is at most 16 bits, convert it
1660 to a signed 16 bit number before trying to see
1661 whether it will fit in an even smaller size.
1662 This allows a 16-bit operand such as $0xffe0 to
1663 be recognised as within Imm8S range. */
1664 if ((i.types[op] & Imm16)
1665 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
1666 {
1667 i.op[op].imms->X_add_number =
1668 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
1669 }
1670 if ((i.types[op] & Imm32)
1671 && (i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1)) == 0)
1672 {
1673 i.op[op].imms->X_add_number =
1674 (i.op[op].imms->X_add_number ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1675 }
1676 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
1677 /* We must avoid matching of Imm32 templates when 64bit only immediate is available. */
1678 if (guess_suffix == QWORD_MNEM_SUFFIX)
1679 i.types[op] &= ~Imm32;
1680 break;
1681 case O_absent:
1682 case O_register:
1683 abort ();
1684 /* Symbols and expressions. */
1685 default:
1686 /* Convert symbolic operand to proper sizes for matching. */
1687 switch (guess_suffix)
1688 {
1689 case QWORD_MNEM_SUFFIX:
1690 i.types[op] = Imm64 | Imm32S;
1691 break;
1692 case LONG_MNEM_SUFFIX:
1693 i.types[op] = Imm32 | Imm64;
1694 break;
1695 case WORD_MNEM_SUFFIX:
1696 i.types[op] = Imm16 | Imm32 | Imm64;
1697 break;
1698 break;
1699 case BYTE_MNEM_SUFFIX:
1700 i.types[op] = Imm8 | Imm8S | Imm16 | Imm32S | Imm32;
1701 break;
1702 break;
1703 }
1704 break;
1705 }
1706 }
1707 }
1708
1709 if (i.disp_operands)
1710 {
1711 /* Try to use the smallest displacement type too. */
1712 int op;
1713
1714 for (op = i.operands; --op >= 0;)
1715 if ((i.types[op] & Disp)
1716 && i.op[op].disps->X_op == O_constant)
1717 {
1718 offsetT disp = i.op[op].disps->X_add_number;
1719
1720 if (i.types[op] & Disp16)
1721 {
1722 /* We know this operand is at most 16 bits, so
1723 convert to a signed 16 bit number before trying
1724 to see whether it will fit in an even smaller
1725 size. */
1726
1727 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
1728 }
1729 else if (i.types[op] & Disp32)
1730 {
1731 /* We know this operand is at most 32 bits, so convert to a
1732 signed 32 bit number before trying to see whether it will
1733 fit in an even smaller size. */
1734 disp &= (((offsetT) 2 << 31) - 1);
1735 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1736 }
1737 if (flag_code == CODE_64BIT)
1738 {
1739 if (fits_in_signed_long (disp))
1740 i.types[op] |= Disp32S;
1741 if (fits_in_unsigned_long (disp))
1742 i.types[op] |= Disp32;
1743 }
1744 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
1745 && fits_in_signed_byte (disp))
1746 i.types[op] |= Disp8;
1747 }
1748 }
1749
1750 overlap0 = 0;
1751 overlap1 = 0;
1752 overlap2 = 0;
1753 found_reverse_match = 0;
1754 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
1755 ? No_bSuf
1756 : (i.suffix == WORD_MNEM_SUFFIX
1757 ? No_wSuf
1758 : (i.suffix == SHORT_MNEM_SUFFIX
1759 ? No_sSuf
1760 : (i.suffix == LONG_MNEM_SUFFIX
1761 ? No_lSuf
1762 : (i.suffix == QWORD_MNEM_SUFFIX
1763 ? No_qSuf
1764 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX ? No_xSuf : 0))))));
1765
1766 for (t = current_templates->start;
1767 t < current_templates->end;
1768 t++)
1769 {
1770 /* Must have right number of operands. */
1771 if (i.operands != t->operands)
1772 continue;
1773
1774 /* Check the suffix, except for some instructions in intel mode. */
1775 if ((t->opcode_modifier & suffix_check)
1776 && !(intel_syntax
1777 && (t->opcode_modifier & IgnoreSize))
1778 && !(intel_syntax
1779 && t->base_opcode == 0xd9
1780 && (t->extension_opcode == 5 /* 0xd9,5 "fldcw" */
1781 || t->extension_opcode == 7))) /* 0xd9,7 "f{n}stcw" */
1782 continue;
1783
1784 /* Do not verify operands when there are none. */
1785 else if (!t->operands)
1786 {
1787 if (t->cpu_flags & ~cpu_arch_flags)
1788 continue;
1789 /* We've found a match; break out of loop. */
1790 break;
1791 }
1792
1793 overlap0 = i.types[0] & t->operand_types[0];
1794 switch (t->operands)
1795 {
1796 case 1:
1797 if (!MATCH (overlap0, i.types[0], t->operand_types[0]))
1798 continue;
1799 break;
1800 case 2:
1801 case 3:
1802 overlap1 = i.types[1] & t->operand_types[1];
1803 if (!MATCH (overlap0, i.types[0], t->operand_types[0])
1804 || !MATCH (overlap1, i.types[1], t->operand_types[1])
1805 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
1806 t->operand_types[0],
1807 overlap1, i.types[1],
1808 t->operand_types[1]))
1809 {
1810 /* Check if other direction is valid ... */
1811 if ((t->opcode_modifier & (D|FloatD)) == 0)
1812 continue;
1813
1814 /* Try reversing direction of operands. */
1815 overlap0 = i.types[0] & t->operand_types[1];
1816 overlap1 = i.types[1] & t->operand_types[0];
1817 if (!MATCH (overlap0, i.types[0], t->operand_types[1])
1818 || !MATCH (overlap1, i.types[1], t->operand_types[0])
1819 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
1820 t->operand_types[1],
1821 overlap1, i.types[1],
1822 t->operand_types[0]))
1823 {
1824 /* Does not match either direction. */
1825 continue;
1826 }
1827 /* found_reverse_match holds which of D or FloatDR
1828 we've found. */
1829 found_reverse_match = t->opcode_modifier & (D|FloatDR);
1830 }
1831 /* Found a forward 2 operand match here. */
1832 else if (t->operands == 3)
1833 {
1834 /* Here we make use of the fact that there are no
1835 reverse match 3 operand instructions, and all 3
1836 operand instructions only need to be checked for
1837 register consistency between operands 2 and 3. */
1838 overlap2 = i.types[2] & t->operand_types[2];
1839 if (!MATCH (overlap2, i.types[2], t->operand_types[2])
1840 || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1],
1841 t->operand_types[1],
1842 overlap2, i.types[2],
1843 t->operand_types[2]))
1844
1845 continue;
1846 }
1847 /* Found either forward/reverse 2 or 3 operand match here:
1848 slip through to break. */
1849 }
1850 if (t->cpu_flags & ~cpu_arch_flags)
1851 {
1852 found_reverse_match = 0;
1853 continue;
1854 }
1855 /* We've found a match; break out of loop. */
1856 break;
1857 }
1858 if (t == current_templates->end)
1859 {
1860 /* We found no match. */
1861 as_bad (_("suffix or operands invalid for `%s'"),
1862 current_templates->start->name);
1863 return;
1864 }
1865
1866 if (!quiet_warnings)
1867 {
1868 if (!intel_syntax
1869 && ((i.types[0] & JumpAbsolute)
1870 != (t->operand_types[0] & JumpAbsolute)))
1871 {
1872 as_warn (_("indirect %s without `*'"), t->name);
1873 }
1874
1875 if ((t->opcode_modifier & (IsPrefix|IgnoreSize))
1876 == (IsPrefix|IgnoreSize))
1877 {
1878 /* Warn them that a data or address size prefix doesn't
1879 affect assembly of the next line of code. */
1880 as_warn (_("stand-alone `%s' prefix"), t->name);
1881 }
1882 }
1883
1884 /* Copy the template we found. */
1885 i.tm = *t;
1886 if (found_reverse_match)
1887 {
1888 /* If we found a reverse match we must alter the opcode
1889 direction bit. found_reverse_match holds bits to change
1890 (different for int & float insns). */
1891
1892 i.tm.base_opcode ^= found_reverse_match;
1893
1894 i.tm.operand_types[0] = t->operand_types[1];
1895 i.tm.operand_types[1] = t->operand_types[0];
1896 }
1897
1898 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1899 if (SYSV386_COMPAT
1900 && intel_syntax
1901 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1902 i.tm.base_opcode ^= FloatR;
1903
1904 if (i.tm.opcode_modifier & FWait)
1905 if (! add_prefix (FWAIT_OPCODE))
1906 return;
1907
1908 /* Check string instruction segment overrides. */
1909 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1910 {
1911 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
1912 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
1913 {
1914 if (i.seg[0] != NULL && i.seg[0] != &es)
1915 {
1916 as_bad (_("`%s' operand %d must use `%%es' segment"),
1917 i.tm.name,
1918 mem_op + 1);
1919 return;
1920 }
1921 /* There's only ever one segment override allowed per instruction.
1922 This instruction possibly has a legal segment override on the
1923 second operand, so copy the segment to where non-string
1924 instructions store it, allowing common code. */
1925 i.seg[0] = i.seg[1];
1926 }
1927 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
1928 {
1929 if (i.seg[1] != NULL && i.seg[1] != &es)
1930 {
1931 as_bad (_("`%s' operand %d must use `%%es' segment"),
1932 i.tm.name,
1933 mem_op + 2);
1934 return;
1935 }
1936 }
1937 }
1938
1939 if (i.reg_operands && flag_code < CODE_64BIT)
1940 {
1941 int op;
1942 for (op = i.operands; --op >= 0;)
1943 if ((i.types[op] & Reg)
1944 && (i.op[op].regs->reg_flags & (RegRex64|RegRex)))
1945 {
1946 as_bad (_("Extended register `%%%s' available only in 64bit mode."),
1947 i.op[op].regs->reg_name);
1948 return;
1949 }
1950 }
1951
1952 /* If matched instruction specifies an explicit instruction mnemonic
1953 suffix, use it. */
1954 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
1955 {
1956 if (i.tm.opcode_modifier & Size16)
1957 i.suffix = WORD_MNEM_SUFFIX;
1958 else if (i.tm.opcode_modifier & Size64)
1959 i.suffix = QWORD_MNEM_SUFFIX;
1960 else
1961 i.suffix = LONG_MNEM_SUFFIX;
1962 }
1963 else if (i.reg_operands)
1964 {
1965 /* If there's no instruction mnemonic suffix we try to invent one
1966 based on register operands. */
1967 if (!i.suffix)
1968 {
1969 /* We take i.suffix from the last register operand specified,
1970 Destination register type is more significant than source
1971 register type. */
1972 int op;
1973 for (op = i.operands; --op >= 0;)
1974 if ((i.types[op] & Reg)
1975 && !(i.tm.operand_types[op] & InOutPortReg))
1976 {
1977 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
1978 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
1979 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
1980 LONG_MNEM_SUFFIX);
1981 break;
1982 }
1983 }
1984 else if (i.suffix == BYTE_MNEM_SUFFIX)
1985 {
1986 int op;
1987 for (op = i.operands; --op >= 0;)
1988 {
1989 /* If this is an eight bit register, it's OK. If it's
1990 the 16 or 32 bit version of an eight bit register,
1991 we will just use the low portion, and that's OK too. */
1992 if (i.types[op] & Reg8)
1993 continue;
1994
1995 /* movzx and movsx should not generate this warning. */
1996 if (intel_syntax
1997 && (i.tm.base_opcode == 0xfb7
1998 || i.tm.base_opcode == 0xfb6
1999 || i.tm.base_opcode == 0x63
2000 || i.tm.base_opcode == 0xfbe
2001 || i.tm.base_opcode == 0xfbf))
2002 continue;
2003
2004 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4
2005 #if 0
2006 /* Check that the template allows eight bit regs
2007 This kills insns such as `orb $1,%edx', which
2008 maybe should be allowed. */
2009 && (i.tm.operand_types[op] & (Reg8|InOutPortReg))
2010 #endif
2011 )
2012 {
2013 /* Prohibit these changes in the 64bit mode, since
2014 the lowering is more complicated. */
2015 if (flag_code == CODE_64BIT
2016 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2017 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2018 i.op[op].regs->reg_name,
2019 i.suffix);
2020 #if REGISTER_WARNINGS
2021 if (!quiet_warnings
2022 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2023 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2024 (i.op[op].regs
2025 + (i.types[op] & Reg16
2026 ? REGNAM_AL - REGNAM_AX
2027 : REGNAM_AL - REGNAM_EAX))->reg_name,
2028 i.op[op].regs->reg_name,
2029 i.suffix);
2030 #endif
2031 continue;
2032 }
2033 /* Any other register is bad. */
2034 if (i.types[op] & (Reg | RegMMX | RegXMM
2035 | SReg2 | SReg3
2036 | Control | Debug | Test
2037 | FloatReg | FloatAcc))
2038 {
2039 as_bad (_("`%%%s' not allowed with `%s%c'"),
2040 i.op[op].regs->reg_name,
2041 i.tm.name,
2042 i.suffix);
2043 return;
2044 }
2045 }
2046 }
2047 else if (i.suffix == LONG_MNEM_SUFFIX)
2048 {
2049 int op;
2050
2051 for (op = i.operands; --op >= 0;)
2052 /* Reject eight bit registers, except where the template
2053 requires them. (eg. movzb) */
2054 if ((i.types[op] & Reg8) != 0
2055 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2056 {
2057 as_bad (_("`%%%s' not allowed with `%s%c'"),
2058 i.op[op].regs->reg_name,
2059 i.tm.name,
2060 i.suffix);
2061 return;
2062 }
2063 /* Warn if the e prefix on a general reg is missing. */
2064 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2065 && (i.types[op] & Reg16) != 0
2066 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
2067 {
2068 /* Prohibit these changes in the 64bit mode, since
2069 the lowering is more complicated. */
2070 if (flag_code == CODE_64BIT)
2071 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2072 i.op[op].regs->reg_name,
2073 i.suffix);
2074 #if REGISTER_WARNINGS
2075 else
2076 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2077 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
2078 i.op[op].regs->reg_name,
2079 i.suffix);
2080 #endif
2081 }
2082 /* Warn if the r prefix on a general reg is missing. */
2083 else if ((i.types[op] & Reg64) != 0
2084 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
2085 {
2086 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2087 i.op[op].regs->reg_name,
2088 i.suffix);
2089 }
2090 }
2091 else if (i.suffix == QWORD_MNEM_SUFFIX)
2092 {
2093 int op;
2094
2095 for (op = i.operands; --op >= 0; )
2096 /* Reject eight bit registers, except where the template
2097 requires them. (eg. movzb) */
2098 if ((i.types[op] & Reg8) != 0
2099 && (i.tm.operand_types[op] & (Reg16|Reg32|Acc)) != 0)
2100 {
2101 as_bad (_("`%%%s' not allowed with `%s%c'"),
2102 i.op[op].regs->reg_name,
2103 i.tm.name,
2104 i.suffix);
2105 return;
2106 }
2107 /* Warn if the e prefix on a general reg is missing. */
2108 else if (((i.types[op] & Reg16) != 0
2109 || (i.types[op] & Reg32) != 0)
2110 && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
2111 {
2112 /* Prohibit these changes in the 64bit mode, since
2113 the lowering is more complicated. */
2114 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2115 i.op[op].regs->reg_name,
2116 i.suffix);
2117 }
2118 }
2119 else if (i.suffix == WORD_MNEM_SUFFIX)
2120 {
2121 int op;
2122 for (op = i.operands; --op >= 0;)
2123 /* Reject eight bit registers, except where the template
2124 requires them. (eg. movzb) */
2125 if ((i.types[op] & Reg8) != 0
2126 && (i.tm.operand_types[op] & (Reg16|Reg32|Acc)) != 0)
2127 {
2128 as_bad (_("`%%%s' not allowed with `%s%c'"),
2129 i.op[op].regs->reg_name,
2130 i.tm.name,
2131 i.suffix);
2132 return;
2133 }
2134 /* Warn if the e prefix on a general reg is present. */
2135 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2136 && (i.types[op] & Reg32) != 0
2137 && (i.tm.operand_types[op] & (Reg16|Acc)) != 0)
2138 {
2139 /* Prohibit these changes in the 64bit mode, since
2140 the lowering is more complicated. */
2141 if (flag_code == CODE_64BIT)
2142 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2143 i.op[op].regs->reg_name,
2144 i.suffix);
2145 else
2146 #if REGISTER_WARNINGS
2147 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2148 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
2149 i.op[op].regs->reg_name,
2150 i.suffix);
2151 #endif
2152 }
2153 }
2154 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2155 /* Do nothing if the instruction is going to ignore the prefix. */
2156 ;
2157 else
2158 abort ();
2159 }
2160 else if ((i.tm.opcode_modifier & DefaultSize) && !i.suffix)
2161 {
2162 i.suffix = stackop_size;
2163 }
2164 /* Make still unresolved immediate matches conform to size of immediate
2165 given in i.suffix. Note: overlap2 cannot be an immediate! */
2166 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S))
2167 && overlap0 != Imm8 && overlap0 != Imm8S
2168 && overlap0 != Imm16 && overlap0 != Imm32S
2169 && overlap0 != Imm32 && overlap0 != Imm64)
2170 {
2171 if (i.suffix)
2172 {
2173 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
2174 (i.suffix == WORD_MNEM_SUFFIX ? Imm16 :
2175 (i.suffix == QWORD_MNEM_SUFFIX ? Imm64 | Imm32S : Imm32)));
2176 }
2177 else if (overlap0 == (Imm16 | Imm32S | Imm32)
2178 || overlap0 == (Imm16 | Imm32)
2179 || overlap0 == (Imm16 | Imm32S))
2180 {
2181 overlap0 =
2182 ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32S;
2183 }
2184 if (overlap0 != Imm8 && overlap0 != Imm8S
2185 && overlap0 != Imm16 && overlap0 != Imm32S
2186 && overlap0 != Imm32 && overlap0 != Imm64)
2187 {
2188 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
2189 return;
2190 }
2191 }
2192 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32))
2193 && overlap1 != Imm8 && overlap1 != Imm8S
2194 && overlap1 != Imm16 && overlap1 != Imm32S
2195 && overlap1 != Imm32 && overlap1 != Imm64)
2196 {
2197 if (i.suffix)
2198 {
2199 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
2200 (i.suffix == WORD_MNEM_SUFFIX ? Imm16 :
2201 (i.suffix == QWORD_MNEM_SUFFIX ? Imm64 | Imm32S : Imm32)));
2202 }
2203 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
2204 || overlap1 == (Imm16 | Imm32)
2205 || overlap1 == (Imm16 | Imm32S))
2206 {
2207 overlap1 =
2208 ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32S;
2209 }
2210 if (overlap1 != Imm8 && overlap1 != Imm8S
2211 && overlap1 != Imm16 && overlap1 != Imm32S
2212 && overlap1 != Imm32 && overlap1 != Imm64)
2213 {
2214 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
2215 return;
2216 }
2217 }
2218 assert ((overlap2 & Imm) == 0);
2219
2220 i.types[0] = overlap0;
2221 if (overlap0 & ImplicitRegister)
2222 i.reg_operands--;
2223 if (overlap0 & Imm1)
2224 i.imm_operands = 0; /* kludge for shift insns. */
2225
2226 i.types[1] = overlap1;
2227 if (overlap1 & ImplicitRegister)
2228 i.reg_operands--;
2229
2230 i.types[2] = overlap2;
2231 if (overlap2 & ImplicitRegister)
2232 i.reg_operands--;
2233
2234 /* Finalize opcode. First, we change the opcode based on the operand
2235 size given by i.suffix: We need not change things for byte insns. */
2236
2237 if (!i.suffix && (i.tm.opcode_modifier & W))
2238 {
2239 as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
2240 return;
2241 }
2242
2243 /* For movzx and movsx, need to check the register type. */
2244 if (intel_syntax
2245 && (i.tm.base_opcode == 0xfb6 || i.tm.base_opcode == 0xfbe))
2246 if (i.suffix && i.suffix == BYTE_MNEM_SUFFIX)
2247 {
2248 unsigned int prefix = DATA_PREFIX_OPCODE;
2249
2250 if ((i.op[1].regs->reg_type & Reg16) != 0)
2251 if (!add_prefix (prefix))
2252 return;
2253 }
2254
2255 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2256 {
2257 /* It's not a byte, select word/dword operation. */
2258 if (i.tm.opcode_modifier & W)
2259 {
2260 if (i.tm.opcode_modifier & ShortForm)
2261 i.tm.base_opcode |= 8;
2262 else
2263 i.tm.base_opcode |= 1;
2264 }
2265 /* Now select between word & dword operations via the operand
2266 size prefix, except for instructions that will ignore this
2267 prefix anyway. */
2268 if (i.suffix != QWORD_MNEM_SUFFIX
2269 && (i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2270 && !(i.tm.opcode_modifier & IgnoreSize))
2271 {
2272 unsigned int prefix = DATA_PREFIX_OPCODE;
2273 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2274 prefix = ADDR_PREFIX_OPCODE;
2275
2276 if (! add_prefix (prefix))
2277 return;
2278 }
2279
2280 /* Set mode64 for an operand. */
2281 if (i.suffix == QWORD_MNEM_SUFFIX
2282 && !(i.tm.opcode_modifier & NoRex64))
2283 {
2284 i.rex.mode64 = 1;
2285 if (flag_code < CODE_64BIT)
2286 {
2287 as_bad (_("64bit operations available only in 64bit modes."));
2288 return;
2289 }
2290 }
2291
2292 /* Size floating point instruction. */
2293 if (i.suffix == LONG_MNEM_SUFFIX)
2294 {
2295 if (i.tm.opcode_modifier & FloatMF)
2296 i.tm.base_opcode ^= 4;
2297 }
2298 }
2299
2300 if (i.tm.opcode_modifier & ImmExt)
2301 {
2302 /* These AMD 3DNow! and Intel Katmai New Instructions have an
2303 opcode suffix which is coded in the same place as an 8-bit
2304 immediate field would be. Here we fake an 8-bit immediate
2305 operand from the opcode suffix stored in tm.extension_opcode. */
2306
2307 expressionS *exp;
2308
2309 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
2310
2311 exp = &im_expressions[i.imm_operands++];
2312 i.op[i.operands].imms = exp;
2313 i.types[i.operands++] = Imm8;
2314 exp->X_op = O_constant;
2315 exp->X_add_number = i.tm.extension_opcode;
2316 i.tm.extension_opcode = None;
2317 }
2318
2319 /* For insns with operands there are more diddles to do to the opcode. */
2320 if (i.operands)
2321 {
2322 /* Default segment register this instruction will use
2323 for memory accesses. 0 means unknown.
2324 This is only for optimizing out unnecessary segment overrides. */
2325 const seg_entry *default_seg = 0;
2326
2327 /* The imul $imm, %reg instruction is converted into
2328 imul $imm, %reg, %reg, and the clr %reg instruction
2329 is converted into xor %reg, %reg. */
2330 if (i.tm.opcode_modifier & regKludge)
2331 {
2332 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
2333 /* Pretend we saw the extra register operand. */
2334 assert (i.op[first_reg_op + 1].regs == 0);
2335 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
2336 i.types[first_reg_op + 1] = i.types[first_reg_op];
2337 i.reg_operands = 2;
2338 }
2339
2340 if (i.tm.opcode_modifier & ShortForm)
2341 {
2342 /* The register or float register operand is in operand 0 or 1. */
2343 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
2344 /* Register goes in low 3 bits of opcode. */
2345 i.tm.base_opcode |= i.op[op].regs->reg_num;
2346 if (i.op[op].regs->reg_flags & RegRex)
2347 i.rex.extZ = 1;
2348 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2349 {
2350 /* Warn about some common errors, but press on regardless.
2351 The first case can be generated by gcc (<= 2.8.1). */
2352 if (i.operands == 2)
2353 {
2354 /* Reversed arguments on faddp, fsubp, etc. */
2355 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
2356 i.op[1].regs->reg_name,
2357 i.op[0].regs->reg_name);
2358 }
2359 else
2360 {
2361 /* Extraneous `l' suffix on fp insn. */
2362 as_warn (_("translating to `%s %%%s'"), i.tm.name,
2363 i.op[0].regs->reg_name);
2364 }
2365 }
2366 }
2367 else if (i.tm.opcode_modifier & Modrm)
2368 {
2369 /* The opcode is completed (modulo i.tm.extension_opcode which
2370 must be put into the modrm byte).
2371 Now, we make the modrm & index base bytes based on all the
2372 info we've collected. */
2373
2374 /* i.reg_operands MUST be the number of real register operands;
2375 implicit registers do not count. */
2376 if (i.reg_operands == 2)
2377 {
2378 unsigned int source, dest;
2379 source = ((i.types[0]
2380 & (Reg | RegMMX | RegXMM
2381 | SReg2 | SReg3
2382 | Control | Debug | Test))
2383 ? 0 : 1);
2384 dest = source + 1;
2385
2386 i.rm.mode = 3;
2387 /* One of the register operands will be encoded in the
2388 i.tm.reg field, the other in the combined i.tm.mode
2389 and i.tm.regmem fields. If no form of this
2390 instruction supports a memory destination operand,
2391 then we assume the source operand may sometimes be
2392 a memory operand and so we need to store the
2393 destination in the i.rm.reg field. */
2394 if ((i.tm.operand_types[dest] & AnyMem) == 0)
2395 {
2396 i.rm.reg = i.op[dest].regs->reg_num;
2397 i.rm.regmem = i.op[source].regs->reg_num;
2398 if (i.op[dest].regs->reg_flags & RegRex)
2399 i.rex.extX = 1;
2400 if (i.op[source].regs->reg_flags & RegRex)
2401 i.rex.extZ = 1;
2402 }
2403 else
2404 {
2405 i.rm.reg = i.op[source].regs->reg_num;
2406 i.rm.regmem = i.op[dest].regs->reg_num;
2407 if (i.op[dest].regs->reg_flags & RegRex)
2408 i.rex.extZ = 1;
2409 if (i.op[source].regs->reg_flags & RegRex)
2410 i.rex.extX = 1;
2411 }
2412 }
2413 else
2414 { /* If it's not 2 reg operands... */
2415 if (i.mem_operands)
2416 {
2417 unsigned int fake_zero_displacement = 0;
2418 unsigned int op = ((i.types[0] & AnyMem)
2419 ? 0
2420 : (i.types[1] & AnyMem) ? 1 : 2);
2421
2422 default_seg = &ds;
2423
2424 if (! i.base_reg)
2425 {
2426 i.rm.mode = 0;
2427 if (! i.disp_operands)
2428 fake_zero_displacement = 1;
2429 if (! i.index_reg)
2430 {
2431 /* Operand is just <disp> */
2432 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
2433 {
2434 i.rm.regmem = NO_BASE_REGISTER_16;
2435 i.types[op] &= ~Disp;
2436 i.types[op] |= Disp16;
2437 }
2438 else if (flag_code != CODE_64BIT)
2439 {
2440 i.rm.regmem = NO_BASE_REGISTER;
2441 i.types[op] &= ~Disp;
2442 i.types[op] |= Disp32;
2443 }
2444 else
2445 {
2446 /* 64bit mode overwrites the 32bit
2447 absolute addressing by RIP relative
2448 addressing and absolute addressing
2449 is encoded by one of the redundant
2450 SIB forms. */
2451
2452 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2453 i.sib.base = NO_BASE_REGISTER;
2454 i.sib.index = NO_INDEX_REGISTER;
2455 i.types[op] &= ~Disp;
2456 i.types[op] |= Disp32S;
2457 }
2458 }
2459 else /* ! i.base_reg && i.index_reg */
2460 {
2461 i.sib.index = i.index_reg->reg_num;
2462 i.sib.base = NO_BASE_REGISTER;
2463 i.sib.scale = i.log2_scale_factor;
2464 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2465 i.types[op] &= ~Disp;
2466 if (flag_code != CODE_64BIT)
2467 i.types[op] |= Disp32; /* Must be 32 bit */
2468 else
2469 i.types[op] |= Disp32S;
2470 if (i.index_reg->reg_flags & RegRex)
2471 i.rex.extY = 1;
2472 }
2473 }
2474 /* RIP addressing for 64bit mode. */
2475 else if (i.base_reg->reg_type == BaseIndex)
2476 {
2477 i.rm.regmem = NO_BASE_REGISTER;
2478 i.types[op] &= ~Disp;
2479 i.types[op] |= Disp32S;
2480 i.flags[op] = Operand_PCrel;
2481 }
2482 else if (i.base_reg->reg_type & Reg16)
2483 {
2484 switch (i.base_reg->reg_num)
2485 {
2486 case 3: /* (%bx) */
2487 if (! i.index_reg)
2488 i.rm.regmem = 7;
2489 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
2490 i.rm.regmem = i.index_reg->reg_num - 6;
2491 break;
2492 case 5: /* (%bp) */
2493 default_seg = &ss;
2494 if (! i.index_reg)
2495 {
2496 i.rm.regmem = 6;
2497 if ((i.types[op] & Disp) == 0)
2498 {
2499 /* fake (%bp) into 0(%bp) */
2500 i.types[op] |= Disp8;
2501 fake_zero_displacement = 1;
2502 }
2503 }
2504 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
2505 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
2506 break;
2507 default: /* (%si) -> 4 or (%di) -> 5 */
2508 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
2509 }
2510 i.rm.mode = mode_from_disp_size (i.types[op]);
2511 }
2512 else /* i.base_reg and 32/64 bit mode */
2513 {
2514 if (flag_code == CODE_64BIT
2515 && (i.types[op] & Disp))
2516 {
2517 if (i.types[op] & Disp8)
2518 i.types[op] = Disp8 | Disp32S;
2519 else
2520 i.types[op] = Disp32S;
2521 }
2522 i.rm.regmem = i.base_reg->reg_num;
2523 if (i.base_reg->reg_flags & RegRex)
2524 i.rex.extZ = 1;
2525 i.sib.base = i.base_reg->reg_num;
2526 /* x86-64 ignores REX prefix bit here to avoid
2527 decoder complications. */
2528 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
2529 {
2530 default_seg = &ss;
2531 if (i.disp_operands == 0)
2532 {
2533 fake_zero_displacement = 1;
2534 i.types[op] |= Disp8;
2535 }
2536 }
2537 else if (i.base_reg->reg_num == ESP_REG_NUM)
2538 {
2539 default_seg = &ss;
2540 }
2541 i.sib.scale = i.log2_scale_factor;
2542 if (! i.index_reg)
2543 {
2544 /* <disp>(%esp) becomes two byte modrm
2545 with no index register. We've already
2546 stored the code for esp in i.rm.regmem
2547 ie. ESCAPE_TO_TWO_BYTE_ADDRESSING. Any
2548 base register besides %esp will not use
2549 the extra modrm byte. */
2550 i.sib.index = NO_INDEX_REGISTER;
2551 #if ! SCALE1_WHEN_NO_INDEX
2552 /* Another case where we force the second
2553 modrm byte. */
2554 if (i.log2_scale_factor)
2555 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2556 #endif
2557 }
2558 else
2559 {
2560 i.sib.index = i.index_reg->reg_num;
2561 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2562 if (i.index_reg->reg_flags & RegRex)
2563 i.rex.extY = 1;
2564 }
2565 i.rm.mode = mode_from_disp_size (i.types[op]);
2566 }
2567
2568 if (fake_zero_displacement)
2569 {
2570 /* Fakes a zero displacement assuming that i.types[op]
2571 holds the correct displacement size. */
2572 expressionS *exp;
2573
2574 assert (i.op[op].disps == 0);
2575 exp = &disp_expressions[i.disp_operands++];
2576 i.op[op].disps = exp;
2577 exp->X_op = O_constant;
2578 exp->X_add_number = 0;
2579 exp->X_add_symbol = (symbolS *) 0;
2580 exp->X_op_symbol = (symbolS *) 0;
2581 }
2582 }
2583
2584 /* Fill in i.rm.reg or i.rm.regmem field with register
2585 operand (if any) based on i.tm.extension_opcode.
2586 Again, we must be careful to make sure that
2587 segment/control/debug/test/MMX registers are coded
2588 into the i.rm.reg field. */
2589 if (i.reg_operands)
2590 {
2591 unsigned int op =
2592 ((i.types[0]
2593 & (Reg | RegMMX | RegXMM
2594 | SReg2 | SReg3
2595 | Control | Debug | Test))
2596 ? 0
2597 : ((i.types[1]
2598 & (Reg | RegMMX | RegXMM
2599 | SReg2 | SReg3
2600 | Control | Debug | Test))
2601 ? 1
2602 : 2));
2603 /* If there is an extension opcode to put here, the
2604 register number must be put into the regmem field. */
2605 if (i.tm.extension_opcode != None)
2606 {
2607 i.rm.regmem = i.op[op].regs->reg_num;
2608 if (i.op[op].regs->reg_flags & RegRex)
2609 i.rex.extZ = 1;
2610 }
2611 else
2612 {
2613 i.rm.reg = i.op[op].regs->reg_num;
2614 if (i.op[op].regs->reg_flags & RegRex)
2615 i.rex.extX = 1;
2616 }
2617
2618 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2
2619 we must set it to 3 to indicate this is a register
2620 operand in the regmem field. */
2621 if (!i.mem_operands)
2622 i.rm.mode = 3;
2623 }
2624
2625 /* Fill in i.rm.reg field with extension opcode (if any). */
2626 if (i.tm.extension_opcode != None)
2627 i.rm.reg = i.tm.extension_opcode;
2628 }
2629 }
2630 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
2631 {
2632 if (i.tm.base_opcode == POP_SEG_SHORT
2633 && i.op[0].regs->reg_num == 1)
2634 {
2635 as_bad (_("you can't `pop %%cs'"));
2636 return;
2637 }
2638 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
2639 if (i.op[0].regs->reg_flags & RegRex)
2640 i.rex.extZ = 1;
2641 }
2642 else if ((i.tm.base_opcode & ~(D|W)) == MOV_AX_DISP32)
2643 {
2644 default_seg = &ds;
2645 }
2646 else if ((i.tm.opcode_modifier & IsString) != 0)
2647 {
2648 /* For the string instructions that allow a segment override
2649 on one of their operands, the default segment is ds. */
2650 default_seg = &ds;
2651 }
2652
2653 /* If a segment was explicitly specified,
2654 and the specified segment is not the default,
2655 use an opcode prefix to select it.
2656 If we never figured out what the default segment is,
2657 then default_seg will be zero at this point,
2658 and the specified segment prefix will always be used. */
2659 if ((i.seg[0]) && (i.seg[0] != default_seg))
2660 {
2661 if (! add_prefix (i.seg[0]->seg_prefix))
2662 return;
2663 }
2664 }
2665 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2666 {
2667 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
2668 as_warn (_("translating to `%sp'"), i.tm.name);
2669 }
2670 }
2671
2672 /* Handle conversion of 'int $3' --> special int3 insn. */
2673 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
2674 {
2675 i.tm.base_opcode = INT3_OPCODE;
2676 i.imm_operands = 0;
2677 }
2678
2679 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
2680 && i.op[0].disps->X_op == O_constant)
2681 {
2682 /* Convert "jmp constant" (and "call constant") to a jump (call) to
2683 the absolute address given by the constant. Since ix86 jumps and
2684 calls are pc relative, we need to generate a reloc. */
2685 i.op[0].disps->X_add_symbol = &abs_symbol;
2686 i.op[0].disps->X_op = O_symbol;
2687 }
2688
2689 if (i.tm.opcode_modifier & Rex64)
2690 i.rex.mode64 = 1;
2691
2692 /* For 8bit registers we would need an empty rex prefix.
2693 Also in the case instruction is already having prefix,
2694 we need to convert old registers to new ones. */
2695
2696 if (((i.types[0] & Reg8) && (i.op[0].regs->reg_flags & RegRex64))
2697 || ((i.types[1] & Reg8) && (i.op[1].regs->reg_flags & RegRex64))
2698 || ((i.rex.mode64 || i.rex.extX || i.rex.extY || i.rex.extZ || i.rex.empty)
2699 && ((i.types[0] & Reg8) || (i.types[1] & Reg8))))
2700 {
2701 int x;
2702 i.rex.empty = 1;
2703 for (x = 0; x < 2; x++)
2704 {
2705 /* Look for 8bit operand that does use old registers. */
2706 if (i.types[x] & Reg8
2707 && !(i.op[x].regs->reg_flags & RegRex64))
2708 {
2709 /* In case it is "hi" register, give up. */
2710 if (i.op[x].regs->reg_num > 3)
2711 as_bad (_("Can't encode registers '%%%s' in the instruction requiring REX prefix.\n"),
2712 i.op[x].regs->reg_name);
2713
2714 /* Otherwise it is equivalent to the extended register.
2715 Since the encoding don't change this is merely cosmetical
2716 cleanup for debug output. */
2717
2718 i.op[x].regs = i.op[x].regs + 8;
2719 }
2720 }
2721 }
2722
2723 if (i.rex.mode64 || i.rex.extX || i.rex.extY || i.rex.extZ || i.rex.empty)
2724 add_prefix (0x40
2725 | (i.rex.mode64 ? 8 : 0)
2726 | (i.rex.extX ? 4 : 0)
2727 | (i.rex.extY ? 2 : 0)
2728 | (i.rex.extZ ? 1 : 0));
2729
2730 /* We are ready to output the insn. */
2731 {
2732 register char *p;
2733
2734 /* Tie dwarf2 debug info to the address at the start of the insn.
2735 We can't do this after the insn has been output as the current
2736 frag may have been closed off. eg. by frag_var. */
2737 dwarf2_emit_insn (0);
2738
2739 /* Output jumps. */
2740 if (i.tm.opcode_modifier & Jump)
2741 {
2742 int code16;
2743 int prefix;
2744 relax_substateT subtype;
2745 symbolS *sym;
2746 offsetT off;
2747
2748 code16 = 0;
2749 if (flag_code == CODE_16BIT)
2750 code16 = CODE16;
2751
2752 prefix = 0;
2753 if (i.prefix[DATA_PREFIX])
2754 {
2755 prefix = 1;
2756 i.prefixes -= 1;
2757 code16 ^= CODE16;
2758 }
2759 /* Pentium4 branch hints. */
2760 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
2761 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
2762 {
2763 prefix++;
2764 i.prefixes--;
2765 }
2766 if (i.prefix[REX_PREFIX])
2767 {
2768 prefix++;
2769 i.prefixes--;
2770 }
2771
2772 if (i.prefixes != 0 && !intel_syntax)
2773 as_warn (_("skipping prefixes on this instruction"));
2774
2775 /* It's always a symbol; End frag & setup for relax.
2776 Make sure there is enough room in this frag for the largest
2777 instruction we may generate in md_convert_frag. This is 2
2778 bytes for the opcode and room for the prefix and largest
2779 displacement. */
2780 frag_grow (prefix + 2 + 4);
2781 /* Prefix and 1 opcode byte go in fr_fix. */
2782 p = frag_more (prefix + 1);
2783 if (i.prefix[DATA_PREFIX])
2784 *p++ = DATA_PREFIX_OPCODE;
2785 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
2786 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
2787 *p++ = i.prefix[SEG_PREFIX];
2788 if (i.prefix[REX_PREFIX])
2789 *p++ = i.prefix[REX_PREFIX];
2790 *p = i.tm.base_opcode;
2791
2792 if ((unsigned char) *p == JUMP_PC_RELATIVE)
2793 subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
2794 else if ((cpu_arch_flags & Cpu386) != 0)
2795 subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
2796 else
2797 subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
2798 subtype |= code16;
2799
2800 sym = i.op[0].disps->X_add_symbol;
2801 off = i.op[0].disps->X_add_number;
2802
2803 if (i.op[0].disps->X_op != O_constant
2804 && i.op[0].disps->X_op != O_symbol)
2805 {
2806 /* Handle complex expressions. */
2807 sym = make_expr_symbol (i.op[0].disps);
2808 off = 0;
2809 }
2810
2811 /* 1 possible extra opcode + 4 byte displacement go in var part.
2812 Pass reloc in fr_var. */
2813 frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
2814 }
2815 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
2816 {
2817 int size;
2818
2819 if (i.tm.opcode_modifier & JumpByte)
2820 {
2821 /* This is a loop or jecxz type instruction. */
2822 size = 1;
2823 if (i.prefix[ADDR_PREFIX])
2824 {
2825 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
2826 i.prefixes -= 1;
2827 }
2828 /* Pentium4 branch hints. */
2829 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
2830 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
2831 {
2832 FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
2833 i.prefixes--;
2834 }
2835 }
2836 else
2837 {
2838 int code16;
2839
2840 code16 = 0;
2841 if (flag_code == CODE_16BIT)
2842 code16 = CODE16;
2843
2844 if (i.prefix[DATA_PREFIX])
2845 {
2846 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
2847 i.prefixes -= 1;
2848 code16 ^= CODE16;
2849 }
2850
2851 size = 4;
2852 if (code16)
2853 size = 2;
2854 }
2855
2856 if (i.prefix[REX_PREFIX])
2857 {
2858 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
2859 i.prefixes -= 1;
2860 }
2861
2862 if (i.prefixes != 0 && !intel_syntax)
2863 as_warn (_("skipping prefixes on this instruction"));
2864
2865 p = frag_more (1 + size);
2866 *p++ = i.tm.base_opcode;
2867
2868 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2869 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
2870 }
2871 else if (i.tm.opcode_modifier & JumpInterSegment)
2872 {
2873 int size;
2874 int prefix;
2875 int code16;
2876
2877 code16 = 0;
2878 if (flag_code == CODE_16BIT)
2879 code16 = CODE16;
2880
2881 prefix = 0;
2882 if (i.prefix[DATA_PREFIX])
2883 {
2884 prefix = 1;
2885 i.prefixes -= 1;
2886 code16 ^= CODE16;
2887 }
2888 if (i.prefix[REX_PREFIX])
2889 {
2890 prefix++;
2891 i.prefixes -= 1;
2892 }
2893
2894 size = 4;
2895 if (code16)
2896 size = 2;
2897
2898 if (i.prefixes != 0 && !intel_syntax)
2899 as_warn (_("skipping prefixes on this instruction"));
2900
2901 /* 1 opcode; 2 segment; offset */
2902 p = frag_more (prefix + 1 + 2 + size);
2903
2904 if (i.prefix[DATA_PREFIX])
2905 *p++ = DATA_PREFIX_OPCODE;
2906
2907 if (i.prefix[REX_PREFIX])
2908 *p++ = i.prefix[REX_PREFIX];
2909
2910 *p++ = i.tm.base_opcode;
2911 if (i.op[1].imms->X_op == O_constant)
2912 {
2913 offsetT n = i.op[1].imms->X_add_number;
2914
2915 if (size == 2
2916 && !fits_in_unsigned_word (n)
2917 && !fits_in_signed_word (n))
2918 {
2919 as_bad (_("16-bit jump out of range"));
2920 return;
2921 }
2922 md_number_to_chars (p, n, size);
2923 }
2924 else
2925 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
2926 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
2927 if (i.op[0].imms->X_op != O_constant)
2928 as_bad (_("can't handle non absolute segment in `%s'"),
2929 i.tm.name);
2930 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
2931 }
2932 else
2933 {
2934 /* Output normal instructions here. */
2935 unsigned char *q;
2936
2937 /* All opcodes on i386 have eighter 1 or 2 bytes. We may use third
2938 byte for the SSE instructions to specify prefix they require. */
2939 if (i.tm.base_opcode & 0xff0000)
2940 add_prefix ((i.tm.base_opcode >> 16) & 0xff);
2941
2942 /* The prefix bytes. */
2943 for (q = i.prefix;
2944 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
2945 q++)
2946 {
2947 if (*q)
2948 {
2949 p = frag_more (1);
2950 md_number_to_chars (p, (valueT) *q, 1);
2951 }
2952 }
2953
2954 /* Now the opcode; be careful about word order here! */
2955 if (fits_in_unsigned_byte (i.tm.base_opcode))
2956 {
2957 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
2958 }
2959 else
2960 {
2961 p = frag_more (2);
2962 /* Put out high byte first: can't use md_number_to_chars! */
2963 *p++ = (i.tm.base_opcode >> 8) & 0xff;
2964 *p = i.tm.base_opcode & 0xff;
2965 }
2966
2967 /* Now the modrm byte and sib byte (if present). */
2968 if (i.tm.opcode_modifier & Modrm)
2969 {
2970 p = frag_more (1);
2971 md_number_to_chars (p,
2972 (valueT) (i.rm.regmem << 0
2973 | i.rm.reg << 3
2974 | i.rm.mode << 6),
2975 1);
2976 /* If i.rm.regmem == ESP (4)
2977 && i.rm.mode != (Register mode)
2978 && not 16 bit
2979 ==> need second modrm byte. */
2980 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
2981 && i.rm.mode != 3
2982 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
2983 {
2984 p = frag_more (1);
2985 md_number_to_chars (p,
2986 (valueT) (i.sib.base << 0
2987 | i.sib.index << 3
2988 | i.sib.scale << 6),
2989 1);
2990 }
2991 }
2992
2993 if (i.disp_operands)
2994 {
2995 register unsigned int n;
2996
2997 for (n = 0; n < i.operands; n++)
2998 {
2999 if (i.types[n] & Disp)
3000 {
3001 if (i.op[n].disps->X_op == O_constant)
3002 {
3003 int size;
3004 offsetT val;
3005
3006 size = 4;
3007 if (i.types[n] & (Disp8 | Disp16 | Disp64))
3008 {
3009 size = 2;
3010 if (i.types[n] & Disp8)
3011 size = 1;
3012 if (i.types[n] & Disp64)
3013 size = 8;
3014 }
3015 val = offset_in_range (i.op[n].disps->X_add_number,
3016 size);
3017 p = frag_more (size);
3018 md_number_to_chars (p, val, size);
3019 }
3020 else
3021 {
3022 int size = 4;
3023 int sign = 0;
3024 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
3025
3026 /* The PC relative address is computed relative
3027 to the instruction boundary, so in case immediate
3028 fields follows, we need to adjust the value. */
3029 if (pcrel && i.imm_operands)
3030 {
3031 int imm_size = 4;
3032 register unsigned int n1;
3033
3034 for (n1 = 0; n1 < i.operands; n1++)
3035 if (i.types[n1] & Imm)
3036 {
3037 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
3038 {
3039 imm_size = 2;
3040 if (i.types[n1] & (Imm8 | Imm8S))
3041 imm_size = 1;
3042 if (i.types[n1] & Imm64)
3043 imm_size = 8;
3044 }
3045 break;
3046 }
3047 /* We should find the immediate. */
3048 if (n1 == i.operands)
3049 abort ();
3050 i.op[n].disps->X_add_number -= imm_size;
3051 }
3052
3053 if (i.types[n] & Disp32S)
3054 sign = 1;
3055
3056 if (i.types[n] & (Disp16 | Disp64))
3057 {
3058 size = 2;
3059 if (i.types[n] & Disp64)
3060 size = 8;
3061 }
3062
3063 p = frag_more (size);
3064 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3065 i.op[n].disps, pcrel,
3066 reloc (size, pcrel, sign, i.reloc[n]));
3067 }
3068 }
3069 }
3070 }
3071
3072 /* Output immediate. */
3073 if (i.imm_operands)
3074 {
3075 register unsigned int n;
3076
3077 for (n = 0; n < i.operands; n++)
3078 {
3079 if (i.types[n] & Imm)
3080 {
3081 if (i.op[n].imms->X_op == O_constant)
3082 {
3083 int size;
3084 offsetT val;
3085
3086 size = 4;
3087 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3088 {
3089 size = 2;
3090 if (i.types[n] & (Imm8 | Imm8S))
3091 size = 1;
3092 else if (i.types[n] & Imm64)
3093 size = 8;
3094 }
3095 val = offset_in_range (i.op[n].imms->X_add_number,
3096 size);
3097 p = frag_more (size);
3098 md_number_to_chars (p, val, size);
3099 }
3100 else
3101 {
3102 /* Not absolute_section.
3103 Need a 32-bit fixup (don't support 8bit
3104 non-absolute imms). Try to support other
3105 sizes ... */
3106 RELOC_ENUM reloc_type;
3107 int size = 4;
3108 int sign = 0;
3109
3110 if ((i.types[n] & (Imm32S))
3111 && i.suffix == QWORD_MNEM_SUFFIX)
3112 sign = 1;
3113 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3114 {
3115 size = 2;
3116 if (i.types[n] & (Imm8 | Imm8S))
3117 size = 1;
3118 if (i.types[n] & Imm64)
3119 size = 8;
3120 }
3121
3122 p = frag_more (size);
3123 reloc_type = reloc (size, 0, sign, i.reloc[n]);
3124 #ifdef BFD_ASSEMBLER
3125 if (reloc_type == BFD_RELOC_32
3126 && GOT_symbol
3127 && GOT_symbol == i.op[n].imms->X_add_symbol
3128 && (i.op[n].imms->X_op == O_symbol
3129 || (i.op[n].imms->X_op == O_add
3130 && ((symbol_get_value_expression
3131 (i.op[n].imms->X_op_symbol)->X_op)
3132 == O_subtract))))
3133 {
3134 /* We don't support dynamic linking on x86-64 yet. */
3135 if (flag_code == CODE_64BIT)
3136 abort ();
3137 reloc_type = BFD_RELOC_386_GOTPC;
3138 i.op[n].imms->X_add_number += 3;
3139 }
3140 #endif
3141 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3142 i.op[n].imms, 0, reloc_type);
3143 }
3144 }
3145 }
3146 }
3147 }
3148
3149 #ifdef DEBUG386
3150 if (flag_debug)
3151 {
3152 pi (line, &i);
3153 }
3154 #endif /* DEBUG386 */
3155 }
3156 }
3157 \f
3158 #ifndef LEX_AT
3159 static char *lex_got PARAMS ((RELOC_ENUM *, int *));
3160
3161 /* Parse operands of the form
3162 <symbol>@GOTOFF+<nnn>
3163 and similar .plt or .got references.
3164
3165 If we find one, set up the correct relocation in RELOC and copy the
3166 input string, minus the `@GOTOFF' into a malloc'd buffer for
3167 parsing by the calling routine. Return this buffer, and if ADJUST
3168 is non-null set it to the length of the string we removed from the
3169 input line. Otherwise return NULL. */
3170 static char *
3171 lex_got (reloc, adjust)
3172 RELOC_ENUM *reloc;
3173 int *adjust;
3174 {
3175 static const char * const mode_name[NUM_FLAG_CODE] = { "32", "16", "64" };
3176 static const struct {
3177 const char *str;
3178 const RELOC_ENUM rel[NUM_FLAG_CODE];
3179 } gotrel[] = {
3180 { "PLT", { BFD_RELOC_386_PLT32, 0, BFD_RELOC_X86_64_PLT32 } },
3181 { "GOTOFF", { BFD_RELOC_386_GOTOFF, 0, 0 } },
3182 { "GOTPCREL", { 0, 0, BFD_RELOC_X86_64_GOTPCREL } },
3183 { "GOT", { BFD_RELOC_386_GOT32, 0, BFD_RELOC_X86_64_GOT32 } }
3184 };
3185 char *cp;
3186 unsigned int j;
3187
3188 for (cp = input_line_pointer; *cp != '@'; cp++)
3189 if (is_end_of_line[(unsigned char) *cp])
3190 return NULL;
3191
3192 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
3193 {
3194 int len;
3195
3196 len = strlen (gotrel[j].str);
3197 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
3198 {
3199 if (gotrel[j].rel[(unsigned int) flag_code] != 0)
3200 {
3201 int first, second;
3202 char *tmpbuf, *past_reloc;
3203
3204 *reloc = gotrel[j].rel[(unsigned int) flag_code];
3205 if (adjust)
3206 *adjust = len;
3207
3208 if (GOT_symbol == NULL)
3209 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3210
3211 /* Replace the relocation token with ' ', so that
3212 errors like foo@GOTOFF1 will be detected. */
3213
3214 /* The length of the first part of our input line. */
3215 first = cp - input_line_pointer;
3216
3217 /* The second part goes from after the reloc token until
3218 (and including) an end_of_line char. Don't use strlen
3219 here as the end_of_line char may not be a NUL. */
3220 past_reloc = cp + 1 + len;
3221 for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
3222 ;
3223 second = cp - past_reloc;
3224
3225 /* Allocate and copy string. The trailing NUL shouldn't
3226 be necessary, but be safe. */
3227 tmpbuf = xmalloc (first + second + 2);
3228 memcpy (tmpbuf, input_line_pointer, first);
3229 tmpbuf[first] = ' ';
3230 memcpy (tmpbuf + first + 1, past_reloc, second);
3231 tmpbuf[first + second + 1] = '\0';
3232 return tmpbuf;
3233 }
3234
3235 as_bad (_("@%s reloc is not supported in %s bit mode"),
3236 gotrel[j].str, mode_name[(unsigned int) flag_code]);
3237 return NULL;
3238 }
3239 }
3240
3241 /* Might be a symbol version string. Don't as_bad here. */
3242 return NULL;
3243 }
3244
3245 /* x86_cons_fix_new is called via the expression parsing code when a
3246 reloc is needed. We use this hook to get the correct .got reloc. */
3247 static RELOC_ENUM got_reloc = NO_RELOC;
3248
3249 void
3250 x86_cons_fix_new (frag, off, len, exp)
3251 fragS *frag;
3252 unsigned int off;
3253 unsigned int len;
3254 expressionS *exp;
3255 {
3256 RELOC_ENUM r = reloc (len, 0, 0, got_reloc);
3257 got_reloc = NO_RELOC;
3258 fix_new_exp (frag, off, len, exp, 0, r);
3259 }
3260
3261 void
3262 x86_cons (exp, size)
3263 expressionS *exp;
3264 int size;
3265 {
3266 if (size == 4)
3267 {
3268 /* Handle @GOTOFF and the like in an expression. */
3269 char *save;
3270 char *gotfree_input_line;
3271 int adjust;
3272
3273 save = input_line_pointer;
3274 gotfree_input_line = lex_got (&got_reloc, &adjust);
3275 if (gotfree_input_line)
3276 input_line_pointer = gotfree_input_line;
3277
3278 expression (exp);
3279
3280 if (gotfree_input_line)
3281 {
3282 /* expression () has merrily parsed up to the end of line,
3283 or a comma - in the wrong buffer. Transfer how far
3284 input_line_pointer has moved to the right buffer. */
3285 input_line_pointer = (save
3286 + (input_line_pointer - gotfree_input_line)
3287 + adjust);
3288 free (gotfree_input_line);
3289 }
3290 }
3291 else
3292 expression (exp);
3293 }
3294 #endif
3295
3296 static int i386_immediate PARAMS ((char *));
3297
3298 static int
3299 i386_immediate (imm_start)
3300 char *imm_start;
3301 {
3302 char *save_input_line_pointer;
3303 #ifndef LEX_AT
3304 char *gotfree_input_line;
3305 #endif
3306 segT exp_seg = 0;
3307 expressionS *exp;
3308
3309 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
3310 {
3311 as_bad (_("only 1 or 2 immediate operands are allowed"));
3312 return 0;
3313 }
3314
3315 exp = &im_expressions[i.imm_operands++];
3316 i.op[this_operand].imms = exp;
3317
3318 if (is_space_char (*imm_start))
3319 ++imm_start;
3320
3321 save_input_line_pointer = input_line_pointer;
3322 input_line_pointer = imm_start;
3323
3324 #ifndef LEX_AT
3325 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3326 if (gotfree_input_line)
3327 input_line_pointer = gotfree_input_line;
3328 #endif
3329
3330 exp_seg = expression (exp);
3331
3332 SKIP_WHITESPACE ();
3333 if (*input_line_pointer)
3334 as_bad (_("junk `%s' after expression"), input_line_pointer);
3335
3336 input_line_pointer = save_input_line_pointer;
3337 #ifndef LEX_AT
3338 if (gotfree_input_line)
3339 free (gotfree_input_line);
3340 #endif
3341
3342 if (exp->X_op == O_absent || exp->X_op == O_big)
3343 {
3344 /* Missing or bad expr becomes absolute 0. */
3345 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
3346 imm_start);
3347 exp->X_op = O_constant;
3348 exp->X_add_number = 0;
3349 exp->X_add_symbol = (symbolS *) 0;
3350 exp->X_op_symbol = (symbolS *) 0;
3351 }
3352 else if (exp->X_op == O_constant)
3353 {
3354 /* Size it properly later. */
3355 i.types[this_operand] |= Imm64;
3356 /* If BFD64, sign extend val. */
3357 if (!use_rela_relocations)
3358 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
3359 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
3360 }
3361 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3362 else if (1
3363 #ifdef BFD_ASSEMBLER
3364 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3365 #endif
3366 && exp_seg != text_section
3367 && exp_seg != data_section
3368 && exp_seg != bss_section
3369 && exp_seg != undefined_section
3370 #ifdef BFD_ASSEMBLER
3371 && !bfd_is_com_section (exp_seg)
3372 #endif
3373 )
3374 {
3375 #ifdef BFD_ASSEMBLER
3376 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3377 #else
3378 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3379 #endif
3380 return 0;
3381 }
3382 #endif
3383 else
3384 {
3385 /* This is an address. The size of the address will be
3386 determined later, depending on destination register,
3387 suffix, or the default for the section. */
3388 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
3389 }
3390
3391 return 1;
3392 }
3393
3394 static char *i386_scale PARAMS ((char *));
3395
3396 static char *
3397 i386_scale (scale)
3398 char *scale;
3399 {
3400 offsetT val;
3401 char *save = input_line_pointer;
3402
3403 input_line_pointer = scale;
3404 val = get_absolute_expression ();
3405
3406 switch (val)
3407 {
3408 case 0:
3409 case 1:
3410 i.log2_scale_factor = 0;
3411 break;
3412 case 2:
3413 i.log2_scale_factor = 1;
3414 break;
3415 case 4:
3416 i.log2_scale_factor = 2;
3417 break;
3418 case 8:
3419 i.log2_scale_factor = 3;
3420 break;
3421 default:
3422 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
3423 scale);
3424 input_line_pointer = save;
3425 return NULL;
3426 }
3427 if (i.log2_scale_factor != 0 && ! i.index_reg)
3428 {
3429 as_warn (_("scale factor of %d without an index register"),
3430 1 << i.log2_scale_factor);
3431 #if SCALE1_WHEN_NO_INDEX
3432 i.log2_scale_factor = 0;
3433 #endif
3434 }
3435 scale = input_line_pointer;
3436 input_line_pointer = save;
3437 return scale;
3438 }
3439
3440 static int i386_displacement PARAMS ((char *, char *));
3441
3442 static int
3443 i386_displacement (disp_start, disp_end)
3444 char *disp_start;
3445 char *disp_end;
3446 {
3447 register expressionS *exp;
3448 segT exp_seg = 0;
3449 char *save_input_line_pointer;
3450 #ifndef LEX_AT
3451 char *gotfree_input_line;
3452 #endif
3453 int bigdisp = Disp32;
3454
3455 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3456 bigdisp = Disp16;
3457 if (flag_code == CODE_64BIT)
3458 bigdisp = Disp64;
3459 i.types[this_operand] |= bigdisp;
3460
3461 exp = &disp_expressions[i.disp_operands];
3462 i.op[this_operand].disps = exp;
3463 i.disp_operands++;
3464 save_input_line_pointer = input_line_pointer;
3465 input_line_pointer = disp_start;
3466 END_STRING_AND_SAVE (disp_end);
3467
3468 #ifndef GCC_ASM_O_HACK
3469 #define GCC_ASM_O_HACK 0
3470 #endif
3471 #if GCC_ASM_O_HACK
3472 END_STRING_AND_SAVE (disp_end + 1);
3473 if ((i.types[this_operand] & BaseIndex) != 0
3474 && displacement_string_end[-1] == '+')
3475 {
3476 /* This hack is to avoid a warning when using the "o"
3477 constraint within gcc asm statements.
3478 For instance:
3479
3480 #define _set_tssldt_desc(n,addr,limit,type) \
3481 __asm__ __volatile__ ( \
3482 "movw %w2,%0\n\t" \
3483 "movw %w1,2+%0\n\t" \
3484 "rorl $16,%1\n\t" \
3485 "movb %b1,4+%0\n\t" \
3486 "movb %4,5+%0\n\t" \
3487 "movb $0,6+%0\n\t" \
3488 "movb %h1,7+%0\n\t" \
3489 "rorl $16,%1" \
3490 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
3491
3492 This works great except that the output assembler ends
3493 up looking a bit weird if it turns out that there is
3494 no offset. You end up producing code that looks like:
3495
3496 #APP
3497 movw $235,(%eax)
3498 movw %dx,2+(%eax)
3499 rorl $16,%edx
3500 movb %dl,4+(%eax)
3501 movb $137,5+(%eax)
3502 movb $0,6+(%eax)
3503 movb %dh,7+(%eax)
3504 rorl $16,%edx
3505 #NO_APP
3506
3507 So here we provide the missing zero. */
3508
3509 *displacement_string_end = '0';
3510 }
3511 #endif
3512 #ifndef LEX_AT
3513 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3514 if (gotfree_input_line)
3515 input_line_pointer = gotfree_input_line;
3516 #endif
3517
3518 exp_seg = expression (exp);
3519
3520 SKIP_WHITESPACE ();
3521 if (*input_line_pointer)
3522 as_bad (_("junk `%s' after expression"), input_line_pointer);
3523 #if GCC_ASM_O_HACK
3524 RESTORE_END_STRING (disp_end + 1);
3525 #endif
3526 RESTORE_END_STRING (disp_end);
3527 input_line_pointer = save_input_line_pointer;
3528 #ifndef LEX_AT
3529 if (gotfree_input_line)
3530 free (gotfree_input_line);
3531 #endif
3532
3533 #ifdef BFD_ASSEMBLER
3534 /* We do this to make sure that the section symbol is in
3535 the symbol table. We will ultimately change the relocation
3536 to be relative to the beginning of the section. */
3537 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
3538 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3539 {
3540 if (exp->X_op != O_symbol)
3541 {
3542 as_bad (_("bad expression used with @%s"),
3543 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
3544 ? "GOTPCREL"
3545 : "GOTOFF"));
3546 return 0;
3547 }
3548
3549 if (S_IS_LOCAL (exp->X_add_symbol)
3550 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
3551 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
3552 exp->X_op = O_subtract;
3553 exp->X_op_symbol = GOT_symbol;
3554 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3555 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
3556 else
3557 i.reloc[this_operand] = BFD_RELOC_32;
3558 }
3559 #endif
3560
3561 if (exp->X_op == O_absent || exp->X_op == O_big)
3562 {
3563 /* Missing or bad expr becomes absolute 0. */
3564 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
3565 disp_start);
3566 exp->X_op = O_constant;
3567 exp->X_add_number = 0;
3568 exp->X_add_symbol = (symbolS *) 0;
3569 exp->X_op_symbol = (symbolS *) 0;
3570 }
3571
3572 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3573 if (exp->X_op != O_constant
3574 #ifdef BFD_ASSEMBLER
3575 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3576 #endif
3577 && exp_seg != text_section
3578 && exp_seg != data_section
3579 && exp_seg != bss_section
3580 && exp_seg != undefined_section)
3581 {
3582 #ifdef BFD_ASSEMBLER
3583 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3584 #else
3585 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3586 #endif
3587 return 0;
3588 }
3589 #endif
3590 else if (flag_code == CODE_64BIT)
3591 i.types[this_operand] |= Disp32S | Disp32;
3592 return 1;
3593 }
3594
3595 static int i386_index_check PARAMS ((const char *));
3596
3597 /* Make sure the memory operand we've been dealt is valid.
3598 Return 1 on success, 0 on a failure. */
3599
3600 static int
3601 i386_index_check (operand_string)
3602 const char *operand_string;
3603 {
3604 int ok;
3605 #if INFER_ADDR_PREFIX
3606 int fudged = 0;
3607
3608 tryprefix:
3609 #endif
3610 ok = 1;
3611 if (flag_code == CODE_64BIT)
3612 {
3613 /* 64bit checks. */
3614 if ((i.base_reg
3615 && ((i.base_reg->reg_type & Reg64) == 0)
3616 && (i.base_reg->reg_type != BaseIndex
3617 || i.index_reg))
3618 || (i.index_reg
3619 && ((i.index_reg->reg_type & (Reg64|BaseIndex))
3620 != (Reg64|BaseIndex))))
3621 ok = 0;
3622 }
3623 else
3624 {
3625 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3626 {
3627 /* 16bit checks. */
3628 if ((i.base_reg
3629 && ((i.base_reg->reg_type & (Reg16|BaseIndex|RegRex))
3630 != (Reg16|BaseIndex)))
3631 || (i.index_reg
3632 && (((i.index_reg->reg_type & (Reg16|BaseIndex))
3633 != (Reg16|BaseIndex))
3634 || ! (i.base_reg
3635 && i.base_reg->reg_num < 6
3636 && i.index_reg->reg_num >= 6
3637 && i.log2_scale_factor == 0))))
3638 ok = 0;
3639 }
3640 else
3641 {
3642 /* 32bit checks. */
3643 if ((i.base_reg
3644 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
3645 || (i.index_reg
3646 && ((i.index_reg->reg_type & (Reg32|BaseIndex|RegRex))
3647 != (Reg32|BaseIndex))))
3648 ok = 0;
3649 }
3650 }
3651 if (!ok)
3652 {
3653 #if INFER_ADDR_PREFIX
3654 if (flag_code != CODE_64BIT
3655 && i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0')
3656 {
3657 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
3658 i.prefixes += 1;
3659 /* Change the size of any displacement too. At most one of
3660 Disp16 or Disp32 is set.
3661 FIXME. There doesn't seem to be any real need for separate
3662 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
3663 Removing them would probably clean up the code quite a lot. */
3664 if (i.types[this_operand] & (Disp16|Disp32))
3665 i.types[this_operand] ^= (Disp16|Disp32);
3666 fudged = 1;
3667 goto tryprefix;
3668 }
3669 if (fudged)
3670 as_bad (_("`%s' is not a valid base/index expression"),
3671 operand_string);
3672 else
3673 #endif
3674 as_bad (_("`%s' is not a valid %s bit base/index expression"),
3675 operand_string,
3676 flag_code_names[flag_code]);
3677 return 0;
3678 }
3679 return 1;
3680 }
3681
3682 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
3683 on error. */
3684
3685 static int
3686 i386_operand (operand_string)
3687 char *operand_string;
3688 {
3689 const reg_entry *r;
3690 char *end_op;
3691 char *op_string = operand_string;
3692
3693 if (is_space_char (*op_string))
3694 ++op_string;
3695
3696 /* We check for an absolute prefix (differentiating,
3697 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
3698 if (*op_string == ABSOLUTE_PREFIX)
3699 {
3700 ++op_string;
3701 if (is_space_char (*op_string))
3702 ++op_string;
3703 i.types[this_operand] |= JumpAbsolute;
3704 }
3705
3706 /* Check if operand is a register. */
3707 if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
3708 && (r = parse_register (op_string, &end_op)) != NULL)
3709 {
3710 /* Check for a segment override by searching for ':' after a
3711 segment register. */
3712 op_string = end_op;
3713 if (is_space_char (*op_string))
3714 ++op_string;
3715 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
3716 {
3717 switch (r->reg_num)
3718 {
3719 case 0:
3720 i.seg[i.mem_operands] = &es;
3721 break;
3722 case 1:
3723 i.seg[i.mem_operands] = &cs;
3724 break;
3725 case 2:
3726 i.seg[i.mem_operands] = &ss;
3727 break;
3728 case 3:
3729 i.seg[i.mem_operands] = &ds;
3730 break;
3731 case 4:
3732 i.seg[i.mem_operands] = &fs;
3733 break;
3734 case 5:
3735 i.seg[i.mem_operands] = &gs;
3736 break;
3737 }
3738
3739 /* Skip the ':' and whitespace. */
3740 ++op_string;
3741 if (is_space_char (*op_string))
3742 ++op_string;
3743
3744 if (!is_digit_char (*op_string)
3745 && !is_identifier_char (*op_string)
3746 && *op_string != '('
3747 && *op_string != ABSOLUTE_PREFIX)
3748 {
3749 as_bad (_("bad memory operand `%s'"), op_string);
3750 return 0;
3751 }
3752 /* Handle case of %es:*foo. */
3753 if (*op_string == ABSOLUTE_PREFIX)
3754 {
3755 ++op_string;
3756 if (is_space_char (*op_string))
3757 ++op_string;
3758 i.types[this_operand] |= JumpAbsolute;
3759 }
3760 goto do_memory_reference;
3761 }
3762 if (*op_string)
3763 {
3764 as_bad (_("junk `%s' after register"), op_string);
3765 return 0;
3766 }
3767 i.types[this_operand] |= r->reg_type & ~BaseIndex;
3768 i.op[this_operand].regs = r;
3769 i.reg_operands++;
3770 }
3771 else if (*op_string == REGISTER_PREFIX)
3772 {
3773 as_bad (_("bad register name `%s'"), op_string);
3774 return 0;
3775 }
3776 else if (*op_string == IMMEDIATE_PREFIX)
3777 {
3778 ++op_string;
3779 if (i.types[this_operand] & JumpAbsolute)
3780 {
3781 as_bad (_("immediate operand illegal with absolute jump"));
3782 return 0;
3783 }
3784 if (!i386_immediate (op_string))
3785 return 0;
3786 }
3787 else if (is_digit_char (*op_string)
3788 || is_identifier_char (*op_string)
3789 || *op_string == '(')
3790 {
3791 /* This is a memory reference of some sort. */
3792 char *base_string;
3793
3794 /* Start and end of displacement string expression (if found). */
3795 char *displacement_string_start;
3796 char *displacement_string_end;
3797
3798 do_memory_reference:
3799 if ((i.mem_operands == 1
3800 && (current_templates->start->opcode_modifier & IsString) == 0)
3801 || i.mem_operands == 2)
3802 {
3803 as_bad (_("too many memory references for `%s'"),
3804 current_templates->start->name);
3805 return 0;
3806 }
3807
3808 /* Check for base index form. We detect the base index form by
3809 looking for an ')' at the end of the operand, searching
3810 for the '(' matching it, and finding a REGISTER_PREFIX or ','
3811 after the '('. */
3812 base_string = op_string + strlen (op_string);
3813
3814 --base_string;
3815 if (is_space_char (*base_string))
3816 --base_string;
3817
3818 /* If we only have a displacement, set-up for it to be parsed later. */
3819 displacement_string_start = op_string;
3820 displacement_string_end = base_string + 1;
3821
3822 if (*base_string == ')')
3823 {
3824 char *temp_string;
3825 unsigned int parens_balanced = 1;
3826 /* We've already checked that the number of left & right ()'s are
3827 equal, so this loop will not be infinite. */
3828 do
3829 {
3830 base_string--;
3831 if (*base_string == ')')
3832 parens_balanced++;
3833 if (*base_string == '(')
3834 parens_balanced--;
3835 }
3836 while (parens_balanced);
3837
3838 temp_string = base_string;
3839
3840 /* Skip past '(' and whitespace. */
3841 ++base_string;
3842 if (is_space_char (*base_string))
3843 ++base_string;
3844
3845 if (*base_string == ','
3846 || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3847 && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
3848 {
3849 displacement_string_end = temp_string;
3850
3851 i.types[this_operand] |= BaseIndex;
3852
3853 if (i.base_reg)
3854 {
3855 base_string = end_op;
3856 if (is_space_char (*base_string))
3857 ++base_string;
3858 }
3859
3860 /* There may be an index reg or scale factor here. */
3861 if (*base_string == ',')
3862 {
3863 ++base_string;
3864 if (is_space_char (*base_string))
3865 ++base_string;
3866
3867 if ((*base_string == REGISTER_PREFIX || allow_naked_reg)
3868 && (i.index_reg = parse_register (base_string, &end_op)) != NULL)
3869 {
3870 base_string = end_op;
3871 if (is_space_char (*base_string))
3872 ++base_string;
3873 if (*base_string == ',')
3874 {
3875 ++base_string;
3876 if (is_space_char (*base_string))
3877 ++base_string;
3878 }
3879 else if (*base_string != ')')
3880 {
3881 as_bad (_("expecting `,' or `)' after index register in `%s'"),
3882 operand_string);
3883 return 0;
3884 }
3885 }
3886 else if (*base_string == REGISTER_PREFIX)
3887 {
3888 as_bad (_("bad register name `%s'"), base_string);
3889 return 0;
3890 }
3891
3892 /* Check for scale factor. */
3893 if (*base_string != ')')
3894 {
3895 char *end_scale = i386_scale (base_string);
3896
3897 if (!end_scale)
3898 return 0;
3899
3900 base_string = end_scale;
3901 if (is_space_char (*base_string))
3902 ++base_string;
3903 if (*base_string != ')')
3904 {
3905 as_bad (_("expecting `)' after scale factor in `%s'"),
3906 operand_string);
3907 return 0;
3908 }
3909 }
3910 else if (!i.index_reg)
3911 {
3912 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
3913 *base_string);
3914 return 0;
3915 }
3916 }
3917 else if (*base_string != ')')
3918 {
3919 as_bad (_("expecting `,' or `)' after base register in `%s'"),
3920 operand_string);
3921 return 0;
3922 }
3923 }
3924 else if (*base_string == REGISTER_PREFIX)
3925 {
3926 as_bad (_("bad register name `%s'"), base_string);
3927 return 0;
3928 }
3929 }
3930
3931 /* If there's an expression beginning the operand, parse it,
3932 assuming displacement_string_start and
3933 displacement_string_end are meaningful. */
3934 if (displacement_string_start != displacement_string_end)
3935 {
3936 if (!i386_displacement (displacement_string_start,
3937 displacement_string_end))
3938 return 0;
3939 }
3940
3941 /* Special case for (%dx) while doing input/output op. */
3942 if (i.base_reg
3943 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
3944 && i.index_reg == 0
3945 && i.log2_scale_factor == 0
3946 && i.seg[i.mem_operands] == 0
3947 && (i.types[this_operand] & Disp) == 0)
3948 {
3949 i.types[this_operand] = InOutPortReg;
3950 return 1;
3951 }
3952
3953 if (i386_index_check (operand_string) == 0)
3954 return 0;
3955 i.mem_operands++;
3956 }
3957 else
3958 {
3959 /* It's not a memory operand; argh! */
3960 as_bad (_("invalid char %s beginning operand %d `%s'"),
3961 output_invalid (*op_string),
3962 this_operand + 1,
3963 op_string);
3964 return 0;
3965 }
3966 return 1; /* Normal return. */
3967 }
3968 \f
3969 /* md_estimate_size_before_relax()
3970
3971 Called just before relax() for rs_machine_dependent frags. The x86
3972 assembler uses these frags to handle variable size jump
3973 instructions.
3974
3975 Any symbol that is now undefined will not become defined.
3976 Return the correct fr_subtype in the frag.
3977 Return the initial "guess for variable size of frag" to caller.
3978 The guess is actually the growth beyond the fixed part. Whatever
3979 we do to grow the fixed or variable part contributes to our
3980 returned value. */
3981
3982 int
3983 md_estimate_size_before_relax (fragP, segment)
3984 register fragS *fragP;
3985 register segT segment;
3986 {
3987 /* We've already got fragP->fr_subtype right; all we have to do is
3988 check for un-relaxable symbols. On an ELF system, we can't relax
3989 an externally visible symbol, because it may be overridden by a
3990 shared library. */
3991 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
3992 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
3993 || S_IS_EXTERNAL (fragP->fr_symbol)
3994 || S_IS_WEAK (fragP->fr_symbol)
3995 #endif
3996 )
3997 {
3998 /* Symbol is undefined in this segment, or we need to keep a
3999 reloc so that weak symbols can be overridden. */
4000 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
4001 RELOC_ENUM reloc_type;
4002 unsigned char *opcode;
4003 int old_fr_fix;
4004
4005 if (fragP->fr_var != NO_RELOC)
4006 reloc_type = fragP->fr_var;
4007 else if (size == 2)
4008 reloc_type = BFD_RELOC_16_PCREL;
4009 else
4010 reloc_type = BFD_RELOC_32_PCREL;
4011
4012 old_fr_fix = fragP->fr_fix;
4013 opcode = (unsigned char *) fragP->fr_opcode;
4014
4015 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
4016 {
4017 case UNCOND_JUMP:
4018 /* Make jmp (0xeb) a (d)word displacement jump. */
4019 opcode[0] = 0xe9;
4020 fragP->fr_fix += size;
4021 fix_new (fragP, old_fr_fix, size,
4022 fragP->fr_symbol,
4023 fragP->fr_offset, 1,
4024 reloc_type);
4025 break;
4026
4027 case COND_JUMP86:
4028 if (no_cond_jump_promotion)
4029 goto relax_guess;
4030
4031 if (size == 2)
4032 {
4033 /* Negate the condition, and branch past an
4034 unconditional jump. */
4035 opcode[0] ^= 1;
4036 opcode[1] = 3;
4037 /* Insert an unconditional jump. */
4038 opcode[2] = 0xe9;
4039 /* We added two extra opcode bytes, and have a two byte
4040 offset. */
4041 fragP->fr_fix += 2 + 2;
4042 fix_new (fragP, old_fr_fix + 2, 2,
4043 fragP->fr_symbol,
4044 fragP->fr_offset, 1,
4045 reloc_type);
4046 break;
4047 }
4048 /* Fall through. */
4049
4050 case COND_JUMP:
4051 if (no_cond_jump_promotion)
4052 goto relax_guess;
4053
4054 /* This changes the byte-displacement jump 0x7N
4055 to the (d)word-displacement jump 0x0f,0x8N. */
4056 opcode[1] = opcode[0] + 0x10;
4057 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4058 /* We've added an opcode byte. */
4059 fragP->fr_fix += 1 + size;
4060 fix_new (fragP, old_fr_fix + 1, size,
4061 fragP->fr_symbol,
4062 fragP->fr_offset, 1,
4063 reloc_type);
4064 break;
4065
4066 default:
4067 BAD_CASE (fragP->fr_subtype);
4068 break;
4069 }
4070 frag_wane (fragP);
4071 return fragP->fr_fix - old_fr_fix;
4072 }
4073
4074 relax_guess:
4075 /* Guess size depending on current relax state. Initially the relax
4076 state will correspond to a short jump and we return 1, because
4077 the variable part of the frag (the branch offset) is one byte
4078 long. However, we can relax a section more than once and in that
4079 case we must either set fr_subtype back to the unrelaxed state,
4080 or return the value for the appropriate branch. */
4081 return md_relax_table[fragP->fr_subtype].rlx_length;
4082 }
4083
4084 /* Called after relax() is finished.
4085
4086 In: Address of frag.
4087 fr_type == rs_machine_dependent.
4088 fr_subtype is what the address relaxed to.
4089
4090 Out: Any fixSs and constants are set up.
4091 Caller will turn frag into a ".space 0". */
4092
4093 #ifndef BFD_ASSEMBLER
4094 void
4095 md_convert_frag (headers, sec, fragP)
4096 object_headers *headers ATTRIBUTE_UNUSED;
4097 segT sec ATTRIBUTE_UNUSED;
4098 register fragS *fragP;
4099 #else
4100 void
4101 md_convert_frag (abfd, sec, fragP)
4102 bfd *abfd ATTRIBUTE_UNUSED;
4103 segT sec ATTRIBUTE_UNUSED;
4104 register fragS *fragP;
4105 #endif
4106 {
4107 register unsigned char *opcode;
4108 unsigned char *where_to_put_displacement = NULL;
4109 offsetT target_address;
4110 offsetT opcode_address;
4111 unsigned int extension = 0;
4112 offsetT displacement_from_opcode_start;
4113
4114 opcode = (unsigned char *) fragP->fr_opcode;
4115
4116 /* Address we want to reach in file space. */
4117 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
4118
4119 /* Address opcode resides at in file space. */
4120 opcode_address = fragP->fr_address + fragP->fr_fix;
4121
4122 /* Displacement from opcode start to fill into instruction. */
4123 displacement_from_opcode_start = target_address - opcode_address;
4124
4125 if ((fragP->fr_subtype & BIG) == 0)
4126 {
4127 /* Don't have to change opcode. */
4128 extension = 1; /* 1 opcode + 1 displacement */
4129 where_to_put_displacement = &opcode[1];
4130 }
4131 else
4132 {
4133 if (no_cond_jump_promotion
4134 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
4135 as_warn_where (fragP->fr_file, fragP->fr_line, _("long jump required"));
4136
4137 switch (fragP->fr_subtype)
4138 {
4139 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
4140 extension = 4; /* 1 opcode + 4 displacement */
4141 opcode[0] = 0xe9;
4142 where_to_put_displacement = &opcode[1];
4143 break;
4144
4145 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
4146 extension = 2; /* 1 opcode + 2 displacement */
4147 opcode[0] = 0xe9;
4148 where_to_put_displacement = &opcode[1];
4149 break;
4150
4151 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
4152 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
4153 extension = 5; /* 2 opcode + 4 displacement */
4154 opcode[1] = opcode[0] + 0x10;
4155 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4156 where_to_put_displacement = &opcode[2];
4157 break;
4158
4159 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
4160 extension = 3; /* 2 opcode + 2 displacement */
4161 opcode[1] = opcode[0] + 0x10;
4162 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4163 where_to_put_displacement = &opcode[2];
4164 break;
4165
4166 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
4167 extension = 4;
4168 opcode[0] ^= 1;
4169 opcode[1] = 3;
4170 opcode[2] = 0xe9;
4171 where_to_put_displacement = &opcode[3];
4172 break;
4173
4174 default:
4175 BAD_CASE (fragP->fr_subtype);
4176 break;
4177 }
4178 }
4179
4180 /* Now put displacement after opcode. */
4181 md_number_to_chars ((char *) where_to_put_displacement,
4182 (valueT) (displacement_from_opcode_start - extension),
4183 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
4184 fragP->fr_fix += extension;
4185 }
4186 \f
4187 /* Size of byte displacement jmp. */
4188 int md_short_jump_size = 2;
4189
4190 /* Size of dword displacement jmp. */
4191 int md_long_jump_size = 5;
4192
4193 /* Size of relocation record. */
4194 const int md_reloc_size = 8;
4195
4196 void
4197 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
4198 char *ptr;
4199 addressT from_addr, to_addr;
4200 fragS *frag ATTRIBUTE_UNUSED;
4201 symbolS *to_symbol ATTRIBUTE_UNUSED;
4202 {
4203 offsetT offset;
4204
4205 offset = to_addr - (from_addr + 2);
4206 /* Opcode for byte-disp jump. */
4207 md_number_to_chars (ptr, (valueT) 0xeb, 1);
4208 md_number_to_chars (ptr + 1, (valueT) offset, 1);
4209 }
4210
4211 void
4212 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
4213 char *ptr;
4214 addressT from_addr, to_addr;
4215 fragS *frag ATTRIBUTE_UNUSED;
4216 symbolS *to_symbol ATTRIBUTE_UNUSED;
4217 {
4218 offsetT offset;
4219
4220 offset = to_addr - (from_addr + 5);
4221 md_number_to_chars (ptr, (valueT) 0xe9, 1);
4222 md_number_to_chars (ptr + 1, (valueT) offset, 4);
4223 }
4224 \f
4225 /* Apply a fixup (fixS) to segment data, once it has been determined
4226 by our caller that we have all the info we need to fix it up.
4227
4228 On the 386, immediates, displacements, and data pointers are all in
4229 the same (little-endian) format, so we don't need to care about which
4230 we are handling. */
4231
4232 int
4233 md_apply_fix3 (fixP, valp, seg)
4234 /* The fix we're to put in. */
4235 fixS *fixP;
4236
4237 /* Pointer to the value of the bits. */
4238 valueT *valp;
4239
4240 /* Segment fix is from. */
4241 segT seg ATTRIBUTE_UNUSED;
4242 {
4243 register char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
4244 valueT value = *valp;
4245
4246 #if defined (BFD_ASSEMBLER) && !defined (TE_Mach)
4247 if (fixP->fx_pcrel)
4248 {
4249 switch (fixP->fx_r_type)
4250 {
4251 default:
4252 break;
4253
4254 case BFD_RELOC_32:
4255 fixP->fx_r_type = BFD_RELOC_32_PCREL;
4256 break;
4257 case BFD_RELOC_16:
4258 fixP->fx_r_type = BFD_RELOC_16_PCREL;
4259 break;
4260 case BFD_RELOC_8:
4261 fixP->fx_r_type = BFD_RELOC_8_PCREL;
4262 break;
4263 }
4264 }
4265
4266 /* This is a hack. There should be a better way to handle this.
4267 This covers for the fact that bfd_install_relocation will
4268 subtract the current location (for partial_inplace, PC relative
4269 relocations); see more below. */
4270 if ((fixP->fx_r_type == BFD_RELOC_32_PCREL
4271 || fixP->fx_r_type == BFD_RELOC_16_PCREL
4272 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
4273 && fixP->fx_addsy && !use_rela_relocations)
4274 {
4275 #ifndef OBJ_AOUT
4276 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4277 #ifdef TE_PE
4278 || OUTPUT_FLAVOR == bfd_target_coff_flavour
4279 #endif
4280 )
4281 value += fixP->fx_where + fixP->fx_frag->fr_address;
4282 #endif
4283 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4284 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
4285 {
4286 segT fseg = S_GET_SEGMENT (fixP->fx_addsy);
4287
4288 if ((fseg == seg
4289 || (symbol_section_p (fixP->fx_addsy)
4290 && fseg != absolute_section))
4291 && ! S_IS_EXTERNAL (fixP->fx_addsy)
4292 && ! S_IS_WEAK (fixP->fx_addsy)
4293 && S_IS_DEFINED (fixP->fx_addsy)
4294 && ! S_IS_COMMON (fixP->fx_addsy))
4295 {
4296 /* Yes, we add the values in twice. This is because
4297 bfd_perform_relocation subtracts them out again. I think
4298 bfd_perform_relocation is broken, but I don't dare change
4299 it. FIXME. */
4300 value += fixP->fx_where + fixP->fx_frag->fr_address;
4301 }
4302 }
4303 #endif
4304 #if defined (OBJ_COFF) && defined (TE_PE)
4305 /* For some reason, the PE format does not store a section
4306 address offset for a PC relative symbol. */
4307 if (S_GET_SEGMENT (fixP->fx_addsy) != seg)
4308 value += md_pcrel_from (fixP);
4309 #endif
4310 }
4311
4312 /* Fix a few things - the dynamic linker expects certain values here,
4313 and we must not dissappoint it. */
4314 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4315 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4316 && fixP->fx_addsy)
4317 switch (fixP->fx_r_type)
4318 {
4319 case BFD_RELOC_386_PLT32:
4320 case BFD_RELOC_X86_64_PLT32:
4321 /* Make the jump instruction point to the address of the operand. At
4322 runtime we merely add the offset to the actual PLT entry. */
4323 value = -4;
4324 break;
4325 case BFD_RELOC_386_GOTPC:
4326
4327 /* This is tough to explain. We end up with this one if we have
4328 * operands that look like "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal
4329 * here is to obtain the absolute address of the GOT, and it is strongly
4330 * preferable from a performance point of view to avoid using a runtime
4331 * relocation for this. The actual sequence of instructions often look
4332 * something like:
4333 *
4334 * call .L66
4335 * .L66:
4336 * popl %ebx
4337 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
4338 *
4339 * The call and pop essentially return the absolute address of
4340 * the label .L66 and store it in %ebx. The linker itself will
4341 * ultimately change the first operand of the addl so that %ebx points to
4342 * the GOT, but to keep things simple, the .o file must have this operand
4343 * set so that it generates not the absolute address of .L66, but the
4344 * absolute address of itself. This allows the linker itself simply
4345 * treat a GOTPC relocation as asking for a pcrel offset to the GOT to be
4346 * added in, and the addend of the relocation is stored in the operand
4347 * field for the instruction itself.
4348 *
4349 * Our job here is to fix the operand so that it would add the correct
4350 * offset so that %ebx would point to itself. The thing that is tricky is
4351 * that .-.L66 will point to the beginning of the instruction, so we need
4352 * to further modify the operand so that it will point to itself.
4353 * There are other cases where you have something like:
4354 *
4355 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
4356 *
4357 * and here no correction would be required. Internally in the assembler
4358 * we treat operands of this form as not being pcrel since the '.' is
4359 * explicitly mentioned, and I wonder whether it would simplify matters
4360 * to do it this way. Who knows. In earlier versions of the PIC patches,
4361 * the pcrel_adjust field was used to store the correction, but since the
4362 * expression is not pcrel, I felt it would be confusing to do it this
4363 * way. */
4364
4365 value -= 1;
4366 break;
4367 case BFD_RELOC_386_GOT32:
4368 case BFD_RELOC_X86_64_GOT32:
4369 value = 0; /* Fully resolved at runtime. No addend. */
4370 break;
4371 case BFD_RELOC_386_GOTOFF:
4372 case BFD_RELOC_X86_64_GOTPCREL:
4373 break;
4374
4375 case BFD_RELOC_VTABLE_INHERIT:
4376 case BFD_RELOC_VTABLE_ENTRY:
4377 fixP->fx_done = 0;
4378 return 1;
4379
4380 default:
4381 break;
4382 }
4383 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
4384 *valp = value;
4385 #endif /* defined (BFD_ASSEMBLER) && !defined (TE_Mach) */
4386
4387 #ifndef BFD_ASSEMBLER
4388 md_number_to_chars (p, value, fixP->fx_size);
4389 #else
4390 /* Are we finished with this relocation now? */
4391 if (fixP->fx_addsy == 0 && fixP->fx_pcrel == 0)
4392 fixP->fx_done = 1;
4393 else if (use_rela_relocations)
4394 {
4395 fixP->fx_no_overflow = 1;
4396 value = 0;
4397 }
4398 md_number_to_chars (p, value, fixP->fx_size);
4399 #endif
4400
4401 return 1;
4402 }
4403 \f
4404 #define MAX_LITTLENUMS 6
4405
4406 /* Turn the string pointed to by litP into a floating point constant
4407 of type TYPE, and emit the appropriate bytes. The number of
4408 LITTLENUMS emitted is stored in *SIZEP. An error message is
4409 returned, or NULL on OK. */
4410
4411 char *
4412 md_atof (type, litP, sizeP)
4413 int type;
4414 char *litP;
4415 int *sizeP;
4416 {
4417 int prec;
4418 LITTLENUM_TYPE words[MAX_LITTLENUMS];
4419 LITTLENUM_TYPE *wordP;
4420 char *t;
4421
4422 switch (type)
4423 {
4424 case 'f':
4425 case 'F':
4426 prec = 2;
4427 break;
4428
4429 case 'd':
4430 case 'D':
4431 prec = 4;
4432 break;
4433
4434 case 'x':
4435 case 'X':
4436 prec = 5;
4437 break;
4438
4439 default:
4440 *sizeP = 0;
4441 return _("Bad call to md_atof ()");
4442 }
4443 t = atof_ieee (input_line_pointer, type, words);
4444 if (t)
4445 input_line_pointer = t;
4446
4447 *sizeP = prec * sizeof (LITTLENUM_TYPE);
4448 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
4449 the bigendian 386. */
4450 for (wordP = words + prec - 1; prec--;)
4451 {
4452 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
4453 litP += sizeof (LITTLENUM_TYPE);
4454 }
4455 return 0;
4456 }
4457 \f
4458 char output_invalid_buf[8];
4459
4460 static char *
4461 output_invalid (c)
4462 int c;
4463 {
4464 if (isprint (c))
4465 sprintf (output_invalid_buf, "'%c'", c);
4466 else
4467 sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
4468 return output_invalid_buf;
4469 }
4470
4471 /* REG_STRING starts *before* REGISTER_PREFIX. */
4472
4473 static const reg_entry *
4474 parse_register (reg_string, end_op)
4475 char *reg_string;
4476 char **end_op;
4477 {
4478 char *s = reg_string;
4479 char *p;
4480 char reg_name_given[MAX_REG_NAME_SIZE + 1];
4481 const reg_entry *r;
4482
4483 /* Skip possible REGISTER_PREFIX and possible whitespace. */
4484 if (*s == REGISTER_PREFIX)
4485 ++s;
4486
4487 if (is_space_char (*s))
4488 ++s;
4489
4490 p = reg_name_given;
4491 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
4492 {
4493 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
4494 return (const reg_entry *) NULL;
4495 s++;
4496 }
4497
4498 /* For naked regs, make sure that we are not dealing with an identifier.
4499 This prevents confusing an identifier like `eax_var' with register
4500 `eax'. */
4501 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
4502 return (const reg_entry *) NULL;
4503
4504 *end_op = s;
4505
4506 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
4507
4508 /* Handle floating point regs, allowing spaces in the (i) part. */
4509 if (r == i386_regtab /* %st is first entry of table */)
4510 {
4511 if (is_space_char (*s))
4512 ++s;
4513 if (*s == '(')
4514 {
4515 ++s;
4516 if (is_space_char (*s))
4517 ++s;
4518 if (*s >= '0' && *s <= '7')
4519 {
4520 r = &i386_float_regtab[*s - '0'];
4521 ++s;
4522 if (is_space_char (*s))
4523 ++s;
4524 if (*s == ')')
4525 {
4526 *end_op = s + 1;
4527 return r;
4528 }
4529 }
4530 /* We have "%st(" then garbage. */
4531 return (const reg_entry *) NULL;
4532 }
4533 }
4534
4535 return r;
4536 }
4537 \f
4538 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4539 const char *md_shortopts = "kVQ:sq";
4540 #else
4541 const char *md_shortopts = "q";
4542 #endif
4543
4544 struct option md_longopts[] = {
4545 #define OPTION_32 (OPTION_MD_BASE + 0)
4546 {"32", no_argument, NULL, OPTION_32},
4547 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4548 #define OPTION_64 (OPTION_MD_BASE + 1)
4549 {"64", no_argument, NULL, OPTION_64},
4550 #endif
4551 {NULL, no_argument, NULL, 0}
4552 };
4553 size_t md_longopts_size = sizeof (md_longopts);
4554
4555 int
4556 md_parse_option (c, arg)
4557 int c;
4558 char *arg ATTRIBUTE_UNUSED;
4559 {
4560 switch (c)
4561 {
4562 case 'q':
4563 quiet_warnings = 1;
4564 break;
4565
4566 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4567 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
4568 should be emitted or not. FIXME: Not implemented. */
4569 case 'Q':
4570 break;
4571
4572 /* -V: SVR4 argument to print version ID. */
4573 case 'V':
4574 print_version_id ();
4575 break;
4576
4577 /* -k: Ignore for FreeBSD compatibility. */
4578 case 'k':
4579 break;
4580
4581 case 's':
4582 /* -s: On i386 Solaris, this tells the native assembler to use
4583 .stab instead of .stab.excl. We always use .stab anyhow. */
4584 break;
4585
4586 case OPTION_64:
4587 {
4588 const char **list, **l;
4589
4590 list = bfd_target_list ();
4591 for (l = list; *l != NULL; l++)
4592 if (strcmp (*l, "elf64-x86-64") == 0)
4593 {
4594 default_arch = "x86_64";
4595 break;
4596 }
4597 if (*l == NULL)
4598 as_fatal (_("No compiled in support for x86_64"));
4599 free (list);
4600 }
4601 break;
4602 #endif
4603
4604 case OPTION_32:
4605 default_arch = "i386";
4606 break;
4607
4608 default:
4609 return 0;
4610 }
4611 return 1;
4612 }
4613
4614 void
4615 md_show_usage (stream)
4616 FILE *stream;
4617 {
4618 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4619 fprintf (stream, _("\
4620 -Q ignored\n\
4621 -V print assembler version number\n\
4622 -k ignored\n\
4623 -q quieten some warnings\n\
4624 -s ignored\n"));
4625 #else
4626 fprintf (stream, _("\
4627 -q quieten some warnings\n"));
4628 #endif
4629 }
4630
4631 #ifdef BFD_ASSEMBLER
4632 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
4633 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
4634
4635 /* Pick the target format to use. */
4636
4637 const char *
4638 i386_target_format ()
4639 {
4640 if (!strcmp (default_arch, "x86_64"))
4641 set_code_flag (CODE_64BIT);
4642 else if (!strcmp (default_arch, "i386"))
4643 set_code_flag (CODE_32BIT);
4644 else
4645 as_fatal (_("Unknown architecture"));
4646 switch (OUTPUT_FLAVOR)
4647 {
4648 #ifdef OBJ_MAYBE_AOUT
4649 case bfd_target_aout_flavour:
4650 return AOUT_TARGET_FORMAT;
4651 #endif
4652 #ifdef OBJ_MAYBE_COFF
4653 case bfd_target_coff_flavour:
4654 return "coff-i386";
4655 #endif
4656 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
4657 case bfd_target_elf_flavour:
4658 {
4659 if (flag_code == CODE_64BIT)
4660 use_rela_relocations = 1;
4661 return flag_code == CODE_64BIT ? "elf64-x86-64" : "elf32-i386";
4662 }
4663 #endif
4664 default:
4665 abort ();
4666 return NULL;
4667 }
4668 }
4669
4670 #endif /* OBJ_MAYBE_ more than one */
4671 #endif /* BFD_ASSEMBLER */
4672 \f
4673 symbolS *
4674 md_undefined_symbol (name)
4675 char *name;
4676 {
4677 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
4678 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
4679 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
4680 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
4681 {
4682 if (!GOT_symbol)
4683 {
4684 if (symbol_find (name))
4685 as_bad (_("GOT already in symbol table"));
4686 GOT_symbol = symbol_new (name, undefined_section,
4687 (valueT) 0, &zero_address_frag);
4688 };
4689 return GOT_symbol;
4690 }
4691 return 0;
4692 }
4693
4694 /* Round up a section size to the appropriate boundary. */
4695
4696 valueT
4697 md_section_align (segment, size)
4698 segT segment ATTRIBUTE_UNUSED;
4699 valueT size;
4700 {
4701 #ifdef BFD_ASSEMBLER
4702 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4703 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
4704 {
4705 /* For a.out, force the section size to be aligned. If we don't do
4706 this, BFD will align it for us, but it will not write out the
4707 final bytes of the section. This may be a bug in BFD, but it is
4708 easier to fix it here since that is how the other a.out targets
4709 work. */
4710 int align;
4711
4712 align = bfd_get_section_alignment (stdoutput, segment);
4713 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
4714 }
4715 #endif
4716 #endif
4717
4718 return size;
4719 }
4720
4721 /* On the i386, PC-relative offsets are relative to the start of the
4722 next instruction. That is, the address of the offset, plus its
4723 size, since the offset is always the last part of the insn. */
4724
4725 long
4726 md_pcrel_from (fixP)
4727 fixS *fixP;
4728 {
4729 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
4730 }
4731
4732 #ifndef I386COFF
4733
4734 static void
4735 s_bss (ignore)
4736 int ignore ATTRIBUTE_UNUSED;
4737 {
4738 register int temp;
4739
4740 temp = get_absolute_expression ();
4741 subseg_set (bss_section, (subsegT) temp);
4742 demand_empty_rest_of_line ();
4743 }
4744
4745 #endif
4746
4747 #ifdef BFD_ASSEMBLER
4748
4749 void
4750 i386_validate_fix (fixp)
4751 fixS *fixp;
4752 {
4753 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
4754 {
4755 /* GOTOFF relocation are nonsense in 64bit mode. */
4756 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
4757 {
4758 if (flag_code != CODE_64BIT)
4759 abort ();
4760 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
4761 }
4762 else
4763 {
4764 if (flag_code == CODE_64BIT)
4765 abort ();
4766 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
4767 }
4768 fixp->fx_subsy = 0;
4769 }
4770 }
4771
4772 arelent *
4773 tc_gen_reloc (section, fixp)
4774 asection *section ATTRIBUTE_UNUSED;
4775 fixS *fixp;
4776 {
4777 arelent *rel;
4778 bfd_reloc_code_real_type code;
4779
4780 switch (fixp->fx_r_type)
4781 {
4782 case BFD_RELOC_X86_64_PLT32:
4783 case BFD_RELOC_X86_64_GOT32:
4784 case BFD_RELOC_X86_64_GOTPCREL:
4785 case BFD_RELOC_386_PLT32:
4786 case BFD_RELOC_386_GOT32:
4787 case BFD_RELOC_386_GOTOFF:
4788 case BFD_RELOC_386_GOTPC:
4789 case BFD_RELOC_X86_64_32S:
4790 case BFD_RELOC_RVA:
4791 case BFD_RELOC_VTABLE_ENTRY:
4792 case BFD_RELOC_VTABLE_INHERIT:
4793 code = fixp->fx_r_type;
4794 break;
4795 default:
4796 if (fixp->fx_pcrel)
4797 {
4798 switch (fixp->fx_size)
4799 {
4800 default:
4801 as_bad_where (fixp->fx_file, fixp->fx_line,
4802 _("can not do %d byte pc-relative relocation"),
4803 fixp->fx_size);
4804 code = BFD_RELOC_32_PCREL;
4805 break;
4806 case 1: code = BFD_RELOC_8_PCREL; break;
4807 case 2: code = BFD_RELOC_16_PCREL; break;
4808 case 4: code = BFD_RELOC_32_PCREL; break;
4809 }
4810 }
4811 else
4812 {
4813 switch (fixp->fx_size)
4814 {
4815 default:
4816 as_bad_where (fixp->fx_file, fixp->fx_line,
4817 _("can not do %d byte relocation"),
4818 fixp->fx_size);
4819 code = BFD_RELOC_32;
4820 break;
4821 case 1: code = BFD_RELOC_8; break;
4822 case 2: code = BFD_RELOC_16; break;
4823 case 4: code = BFD_RELOC_32; break;
4824 case 8: code = BFD_RELOC_64; break;
4825 }
4826 }
4827 break;
4828 }
4829
4830 if (code == BFD_RELOC_32
4831 && GOT_symbol
4832 && fixp->fx_addsy == GOT_symbol)
4833 {
4834 /* We don't support GOTPC on 64bit targets. */
4835 if (flag_code == CODE_64BIT)
4836 abort ();
4837 code = BFD_RELOC_386_GOTPC;
4838 }
4839
4840 rel = (arelent *) xmalloc (sizeof (arelent));
4841 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
4842 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
4843
4844 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
4845 if (!use_rela_relocations)
4846 {
4847 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
4848 vtable entry to be used in the relocation's section offset. */
4849 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
4850 rel->address = fixp->fx_offset;
4851
4852 if (fixp->fx_pcrel)
4853 rel->addend = fixp->fx_addnumber;
4854 else
4855 rel->addend = 0;
4856 }
4857 /* Use the rela in 64bit mode. */
4858 else
4859 {
4860 rel->addend = fixp->fx_offset;
4861 if (fixp->fx_pcrel)
4862 rel->addend -= fixp->fx_size;
4863 }
4864
4865 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
4866 if (rel->howto == NULL)
4867 {
4868 as_bad_where (fixp->fx_file, fixp->fx_line,
4869 _("cannot represent relocation type %s"),
4870 bfd_get_reloc_code_name (code));
4871 /* Set howto to a garbage value so that we can keep going. */
4872 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
4873 assert (rel->howto != NULL);
4874 }
4875
4876 return rel;
4877 }
4878
4879 #else /* ! BFD_ASSEMBLER */
4880
4881 #if (defined(OBJ_AOUT) | defined(OBJ_BOUT))
4882 void
4883 tc_aout_fix_to_chars (where, fixP, segment_address_in_file)
4884 char *where;
4885 fixS *fixP;
4886 relax_addressT segment_address_in_file;
4887 {
4888 /* In: length of relocation (or of address) in chars: 1, 2 or 4.
4889 Out: GNU LD relocation length code: 0, 1, or 2. */
4890
4891 static const unsigned char nbytes_r_length[] = { 42, 0, 1, 42, 2 };
4892 long r_symbolnum;
4893
4894 know (fixP->fx_addsy != NULL);
4895
4896 md_number_to_chars (where,
4897 (valueT) (fixP->fx_frag->fr_address
4898 + fixP->fx_where - segment_address_in_file),
4899 4);
4900
4901 r_symbolnum = (S_IS_DEFINED (fixP->fx_addsy)
4902 ? S_GET_TYPE (fixP->fx_addsy)
4903 : fixP->fx_addsy->sy_number);
4904
4905 where[6] = (r_symbolnum >> 16) & 0x0ff;
4906 where[5] = (r_symbolnum >> 8) & 0x0ff;
4907 where[4] = r_symbolnum & 0x0ff;
4908 where[7] = ((((!S_IS_DEFINED (fixP->fx_addsy)) << 3) & 0x08)
4909 | ((nbytes_r_length[fixP->fx_size] << 1) & 0x06)
4910 | (((fixP->fx_pcrel << 0) & 0x01) & 0x0f));
4911 }
4912
4913 #endif /* OBJ_AOUT or OBJ_BOUT. */
4914
4915 #if defined (I386COFF)
4916
4917 short
4918 tc_coff_fix2rtype (fixP)
4919 fixS *fixP;
4920 {
4921 if (fixP->fx_r_type == R_IMAGEBASE)
4922 return R_IMAGEBASE;
4923
4924 return (fixP->fx_pcrel ?
4925 (fixP->fx_size == 1 ? R_PCRBYTE :
4926 fixP->fx_size == 2 ? R_PCRWORD :
4927 R_PCRLONG) :
4928 (fixP->fx_size == 1 ? R_RELBYTE :
4929 fixP->fx_size == 2 ? R_RELWORD :
4930 R_DIR32));
4931 }
4932
4933 int
4934 tc_coff_sizemachdep (frag)
4935 fragS *frag;
4936 {
4937 if (frag->fr_next)
4938 return (frag->fr_next->fr_address - frag->fr_address);
4939 else
4940 return 0;
4941 }
4942
4943 #endif /* I386COFF */
4944
4945 #endif /* ! BFD_ASSEMBLER */
4946 \f
4947 /* Parse operands using Intel syntax. This implements a recursive descent
4948 parser based on the BNF grammar published in Appendix B of the MASM 6.1
4949 Programmer's Guide.
4950
4951 FIXME: We do not recognize the full operand grammar defined in the MASM
4952 documentation. In particular, all the structure/union and
4953 high-level macro operands are missing.
4954
4955 Uppercase words are terminals, lower case words are non-terminals.
4956 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
4957 bars '|' denote choices. Most grammar productions are implemented in
4958 functions called 'intel_<production>'.
4959
4960 Initial production is 'expr'.
4961
4962 addOp + | -
4963
4964 alpha [a-zA-Z]
4965
4966 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
4967
4968 constant digits [[ radixOverride ]]
4969
4970 dataType BYTE | WORD | DWORD | QWORD | XWORD
4971
4972 digits decdigit
4973 | digits decdigit
4974 | digits hexdigit
4975
4976 decdigit [0-9]
4977
4978 e05 e05 addOp e06
4979 | e06
4980
4981 e06 e06 mulOp e09
4982 | e09
4983
4984 e09 OFFSET e10
4985 | e09 PTR e10
4986 | e09 : e10
4987 | e10
4988
4989 e10 e10 [ expr ]
4990 | e11
4991
4992 e11 ( expr )
4993 | [ expr ]
4994 | constant
4995 | dataType
4996 | id
4997 | $
4998 | register
4999
5000 => expr SHORT e05
5001 | e05
5002
5003 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
5004 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
5005
5006 hexdigit a | b | c | d | e | f
5007 | A | B | C | D | E | F
5008
5009 id alpha
5010 | id alpha
5011 | id decdigit
5012
5013 mulOp * | / | MOD
5014
5015 quote " | '
5016
5017 register specialRegister
5018 | gpRegister
5019 | byteRegister
5020
5021 segmentRegister CS | DS | ES | FS | GS | SS
5022
5023 specialRegister CR0 | CR2 | CR3
5024 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
5025 | TR3 | TR4 | TR5 | TR6 | TR7
5026
5027 We simplify the grammar in obvious places (e.g., register parsing is
5028 done by calling parse_register) and eliminate immediate left recursion
5029 to implement a recursive-descent parser.
5030
5031 expr SHORT e05
5032 | e05
5033
5034 e05 e06 e05'
5035
5036 e05' addOp e06 e05'
5037 | Empty
5038
5039 e06 e09 e06'
5040
5041 e06' mulOp e09 e06'
5042 | Empty
5043
5044 e09 OFFSET e10 e09'
5045 | e10 e09'
5046
5047 e09' PTR e10 e09'
5048 | : e10 e09'
5049 | Empty
5050
5051 e10 e11 e10'
5052
5053 e10' [ expr ] e10'
5054 | Empty
5055
5056 e11 ( expr )
5057 | [ expr ]
5058 | BYTE
5059 | WORD
5060 | DWORD
5061 | QWORD
5062 | XWORD
5063 | .
5064 | $
5065 | register
5066 | id
5067 | constant */
5068
5069 /* Parsing structure for the intel syntax parser. Used to implement the
5070 semantic actions for the operand grammar. */
5071 struct intel_parser_s
5072 {
5073 char *op_string; /* The string being parsed. */
5074 int got_a_float; /* Whether the operand is a float. */
5075 int op_modifier; /* Operand modifier. */
5076 int is_mem; /* 1 if operand is memory reference. */
5077 const reg_entry *reg; /* Last register reference found. */
5078 char *disp; /* Displacement string being built. */
5079 };
5080
5081 static struct intel_parser_s intel_parser;
5082
5083 /* Token structure for parsing intel syntax. */
5084 struct intel_token
5085 {
5086 int code; /* Token code. */
5087 const reg_entry *reg; /* Register entry for register tokens. */
5088 char *str; /* String representation. */
5089 };
5090
5091 static struct intel_token cur_token, prev_token;
5092
5093 /* Token codes for the intel parser. Since T_SHORT is already used
5094 by COFF, undefine it first to prevent a warning. */
5095 #define T_NIL -1
5096 #define T_CONST 1
5097 #define T_REG 2
5098 #define T_BYTE 3
5099 #define T_WORD 4
5100 #define T_DWORD 5
5101 #define T_QWORD 6
5102 #define T_XWORD 7
5103 #undef T_SHORT
5104 #define T_SHORT 8
5105 #define T_OFFSET 9
5106 #define T_PTR 10
5107 #define T_ID 11
5108
5109 /* Prototypes for intel parser functions. */
5110 static int intel_match_token PARAMS ((int code));
5111 static void intel_get_token PARAMS ((void));
5112 static void intel_putback_token PARAMS ((void));
5113 static int intel_expr PARAMS ((void));
5114 static int intel_e05 PARAMS ((void));
5115 static int intel_e05_1 PARAMS ((void));
5116 static int intel_e06 PARAMS ((void));
5117 static int intel_e06_1 PARAMS ((void));
5118 static int intel_e09 PARAMS ((void));
5119 static int intel_e09_1 PARAMS ((void));
5120 static int intel_e10 PARAMS ((void));
5121 static int intel_e10_1 PARAMS ((void));
5122 static int intel_e11 PARAMS ((void));
5123
5124 static int
5125 i386_intel_operand (operand_string, got_a_float)
5126 char *operand_string;
5127 int got_a_float;
5128 {
5129 int ret;
5130 char *p;
5131
5132 /* Initialize token holders. */
5133 cur_token.code = prev_token.code = T_NIL;
5134 cur_token.reg = prev_token.reg = NULL;
5135 cur_token.str = prev_token.str = NULL;
5136
5137 /* Initialize parser structure. */
5138 p = intel_parser.op_string = (char *) malloc (strlen (operand_string) + 1);
5139 if (p == NULL)
5140 abort ();
5141 strcpy (intel_parser.op_string, operand_string);
5142 intel_parser.got_a_float = got_a_float;
5143 intel_parser.op_modifier = -1;
5144 intel_parser.is_mem = 0;
5145 intel_parser.reg = NULL;
5146 intel_parser.disp = (char *) malloc (strlen (operand_string) + 1);
5147 if (intel_parser.disp == NULL)
5148 abort ();
5149 intel_parser.disp[0] = '\0';
5150
5151 /* Read the first token and start the parser. */
5152 intel_get_token ();
5153 ret = intel_expr ();
5154
5155 if (ret)
5156 {
5157 /* If we found a memory reference, hand it over to i386_displacement
5158 to fill in the rest of the operand fields. */
5159 if (intel_parser.is_mem)
5160 {
5161 if ((i.mem_operands == 1
5162 && (current_templates->start->opcode_modifier & IsString) == 0)
5163 || i.mem_operands == 2)
5164 {
5165 as_bad (_("too many memory references for '%s'"),
5166 current_templates->start->name);
5167 ret = 0;
5168 }
5169 else
5170 {
5171 char *s = intel_parser.disp;
5172 i.mem_operands++;
5173
5174 /* Add the displacement expression. */
5175 if (*s != '\0')
5176 ret = i386_displacement (s, s + strlen (s))
5177 && i386_index_check (s);
5178 }
5179 }
5180
5181 /* Constant and OFFSET expressions are handled by i386_immediate. */
5182 else if (intel_parser.op_modifier == OFFSET_FLAT
5183 || intel_parser.reg == NULL)
5184 ret = i386_immediate (intel_parser.disp);
5185 }
5186
5187 free (p);
5188 free (intel_parser.disp);
5189
5190 return ret;
5191 }
5192
5193 /* expr SHORT e05
5194 | e05 */
5195 static int
5196 intel_expr ()
5197 {
5198 /* expr SHORT e05 */
5199 if (cur_token.code == T_SHORT)
5200 {
5201 intel_parser.op_modifier = SHORT;
5202 intel_match_token (T_SHORT);
5203
5204 return (intel_e05 ());
5205 }
5206
5207 /* expr e05 */
5208 else
5209 return intel_e05 ();
5210 }
5211
5212 /* e05 e06 e05'
5213
5214 e05' addOp e06 e05'
5215 | Empty */
5216 static int
5217 intel_e05 ()
5218 {
5219 return (intel_e06 () && intel_e05_1 ());
5220 }
5221
5222 static int
5223 intel_e05_1 ()
5224 {
5225 /* e05' addOp e06 e05' */
5226 if (cur_token.code == '+' || cur_token.code == '-')
5227 {
5228 strcat (intel_parser.disp, cur_token.str);
5229 intel_match_token (cur_token.code);
5230
5231 return (intel_e06 () && intel_e05_1 ());
5232 }
5233
5234 /* e05' Empty */
5235 else
5236 return 1;
5237 }
5238
5239 /* e06 e09 e06'
5240
5241 e06' mulOp e09 e06'
5242 | Empty */
5243 static int
5244 intel_e06 ()
5245 {
5246 return (intel_e09 () && intel_e06_1 ());
5247 }
5248
5249 static int
5250 intel_e06_1 ()
5251 {
5252 /* e06' mulOp e09 e06' */
5253 if (cur_token.code == '*' || cur_token.code == '/')
5254 {
5255 strcat (intel_parser.disp, cur_token.str);
5256 intel_match_token (cur_token.code);
5257
5258 return (intel_e09 () && intel_e06_1 ());
5259 }
5260
5261 /* e06' Empty */
5262 else
5263 return 1;
5264 }
5265
5266 /* e09 OFFSET e10 e09'
5267 | e10 e09'
5268
5269 e09' PTR e10 e09'
5270 | : e10 e09'
5271 | Empty */
5272 static int
5273 intel_e09 ()
5274 {
5275 /* e09 OFFSET e10 e09' */
5276 if (cur_token.code == T_OFFSET)
5277 {
5278 intel_parser.is_mem = 0;
5279 intel_parser.op_modifier = OFFSET_FLAT;
5280 intel_match_token (T_OFFSET);
5281
5282 return (intel_e10 () && intel_e09_1 ());
5283 }
5284
5285 /* e09 e10 e09' */
5286 else
5287 return (intel_e10 () && intel_e09_1 ());
5288 }
5289
5290 static int
5291 intel_e09_1 ()
5292 {
5293 /* e09' PTR e10 e09' */
5294 if (cur_token.code == T_PTR)
5295 {
5296 if (prev_token.code == T_BYTE)
5297 i.suffix = BYTE_MNEM_SUFFIX;
5298
5299 else if (prev_token.code == T_WORD)
5300 {
5301 if (intel_parser.got_a_float == 2) /* "fi..." */
5302 i.suffix = SHORT_MNEM_SUFFIX;
5303 else
5304 i.suffix = WORD_MNEM_SUFFIX;
5305 }
5306
5307 else if (prev_token.code == T_DWORD)
5308 {
5309 if (intel_parser.got_a_float == 1) /* "f..." */
5310 i.suffix = SHORT_MNEM_SUFFIX;
5311 else
5312 i.suffix = LONG_MNEM_SUFFIX;
5313 }
5314
5315 else if (prev_token.code == T_QWORD)
5316 {
5317 if (intel_parser.got_a_float == 1) /* "f..." */
5318 i.suffix = LONG_MNEM_SUFFIX;
5319 else
5320 i.suffix = QWORD_MNEM_SUFFIX;
5321 }
5322
5323 else if (prev_token.code == T_XWORD)
5324 i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
5325
5326 else
5327 {
5328 as_bad (_("Unknown operand modifier `%s'\n"), prev_token.str);
5329 return 0;
5330 }
5331
5332 intel_match_token (T_PTR);
5333
5334 return (intel_e10 () && intel_e09_1 ());
5335 }
5336
5337 /* e09 : e10 e09' */
5338 else if (cur_token.code == ':')
5339 {
5340 /* Mark as a memory operand only if it's not already known to be an
5341 offset expression. */
5342 if (intel_parser.op_modifier != OFFSET_FLAT)
5343 intel_parser.is_mem = 1;
5344
5345 return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
5346 }
5347
5348 /* e09' Empty */
5349 else
5350 return 1;
5351 }
5352
5353 /* e10 e11 e10'
5354
5355 e10' [ expr ] e10'
5356 | Empty */
5357 static int
5358 intel_e10 ()
5359 {
5360 return (intel_e11 () && intel_e10_1 ());
5361 }
5362
5363 static int
5364 intel_e10_1 ()
5365 {
5366 /* e10' [ expr ] e10' */
5367 if (cur_token.code == '[')
5368 {
5369 intel_match_token ('[');
5370
5371 /* Mark as a memory operand only if it's not already known to be an
5372 offset expression. If it's an offset expression, we need to keep
5373 the brace in. */
5374 if (intel_parser.op_modifier != OFFSET_FLAT)
5375 intel_parser.is_mem = 1;
5376 else
5377 strcat (intel_parser.disp, "[");
5378
5379 /* Add a '+' to the displacement string if necessary. */
5380 if (*intel_parser.disp != '\0'
5381 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5382 strcat (intel_parser.disp, "+");
5383
5384 if (intel_expr () && intel_match_token (']'))
5385 {
5386 /* Preserve brackets when the operand is an offset expression. */
5387 if (intel_parser.op_modifier == OFFSET_FLAT)
5388 strcat (intel_parser.disp, "]");
5389
5390 return intel_e10_1 ();
5391 }
5392 else
5393 return 0;
5394 }
5395
5396 /* e10' Empty */
5397 else
5398 return 1;
5399 }
5400
5401 /* e11 ( expr )
5402 | [ expr ]
5403 | BYTE
5404 | WORD
5405 | DWORD
5406 | QWORD
5407 | XWORD
5408 | $
5409 | .
5410 | register
5411 | id
5412 | constant */
5413 static int
5414 intel_e11 ()
5415 {
5416 /* e11 ( expr ) */
5417 if (cur_token.code == '(')
5418 {
5419 intel_match_token ('(');
5420 strcat (intel_parser.disp, "(");
5421
5422 if (intel_expr () && intel_match_token (')'))
5423 {
5424 strcat (intel_parser.disp, ")");
5425 return 1;
5426 }
5427 else
5428 return 0;
5429 }
5430
5431 /* e11 [ expr ] */
5432 else if (cur_token.code == '[')
5433 {
5434 intel_match_token ('[');
5435
5436 /* Mark as a memory operand only if it's not already known to be an
5437 offset expression. If it's an offset expression, we need to keep
5438 the brace in. */
5439 if (intel_parser.op_modifier != OFFSET_FLAT)
5440 intel_parser.is_mem = 1;
5441 else
5442 strcat (intel_parser.disp, "[");
5443
5444 /* Operands for jump/call inside brackets denote absolute addresses. */
5445 if (current_templates->start->opcode_modifier & Jump
5446 || current_templates->start->opcode_modifier & JumpDword
5447 || current_templates->start->opcode_modifier & JumpByte
5448 || current_templates->start->opcode_modifier & JumpInterSegment)
5449 i.types[this_operand] |= JumpAbsolute;
5450
5451 /* Add a '+' to the displacement string if necessary. */
5452 if (*intel_parser.disp != '\0'
5453 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5454 strcat (intel_parser.disp, "+");
5455
5456 if (intel_expr () && intel_match_token (']'))
5457 {
5458 /* Preserve brackets when the operand is an offset expression. */
5459 if (intel_parser.op_modifier == OFFSET_FLAT)
5460 strcat (intel_parser.disp, "]");
5461
5462 return 1;
5463 }
5464 else
5465 return 0;
5466 }
5467
5468 /* e11 BYTE
5469 | WORD
5470 | DWORD
5471 | QWORD
5472 | XWORD */
5473 else if (cur_token.code == T_BYTE
5474 || cur_token.code == T_WORD
5475 || cur_token.code == T_DWORD
5476 || cur_token.code == T_QWORD
5477 || cur_token.code == T_XWORD)
5478 {
5479 intel_match_token (cur_token.code);
5480
5481 return 1;
5482 }
5483
5484 /* e11 $
5485 | . */
5486 else if (cur_token.code == '$' || cur_token.code == '.')
5487 {
5488 strcat (intel_parser.disp, cur_token.str);
5489 intel_match_token (cur_token.code);
5490
5491 /* Mark as a memory operand only if it's not already known to be an
5492 offset expression. */
5493 if (intel_parser.op_modifier != OFFSET_FLAT)
5494 intel_parser.is_mem = 1;
5495
5496 return 1;
5497 }
5498
5499 /* e11 register */
5500 else if (cur_token.code == T_REG)
5501 {
5502 const reg_entry *reg = intel_parser.reg = cur_token.reg;
5503
5504 intel_match_token (T_REG);
5505
5506 /* Check for segment change. */
5507 if (cur_token.code == ':')
5508 {
5509 if (reg->reg_type & (SReg2 | SReg3))
5510 {
5511 switch (reg->reg_num)
5512 {
5513 case 0:
5514 i.seg[i.mem_operands] = &es;
5515 break;
5516 case 1:
5517 i.seg[i.mem_operands] = &cs;
5518 break;
5519 case 2:
5520 i.seg[i.mem_operands] = &ss;
5521 break;
5522 case 3:
5523 i.seg[i.mem_operands] = &ds;
5524 break;
5525 case 4:
5526 i.seg[i.mem_operands] = &fs;
5527 break;
5528 case 5:
5529 i.seg[i.mem_operands] = &gs;
5530 break;
5531 }
5532 }
5533 else
5534 {
5535 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
5536 return 0;
5537 }
5538 }
5539
5540 /* Not a segment register. Check for register scaling. */
5541 else if (cur_token.code == '*')
5542 {
5543 if (!intel_parser.is_mem)
5544 {
5545 as_bad (_("Register scaling only allowed in memory operands."));
5546 return 0;
5547 }
5548
5549 /* What follows must be a valid scale. */
5550 if (intel_match_token ('*')
5551 && strchr ("01248", *cur_token.str))
5552 {
5553 i.index_reg = reg;
5554 i.types[this_operand] |= BaseIndex;
5555
5556 /* Set the scale after setting the register (otherwise,
5557 i386_scale will complain) */
5558 i386_scale (cur_token.str);
5559 intel_match_token (T_CONST);
5560 }
5561 else
5562 {
5563 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
5564 cur_token.str);
5565 return 0;
5566 }
5567 }
5568
5569 /* No scaling. If this is a memory operand, the register is either a
5570 base register (first occurrence) or an index register (second
5571 occurrence). */
5572 else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3)))
5573 {
5574 if (i.base_reg && i.index_reg)
5575 {
5576 as_bad (_("Too many register references in memory operand.\n"));
5577 return 0;
5578 }
5579
5580 if (i.base_reg == NULL)
5581 i.base_reg = reg;
5582 else
5583 i.index_reg = reg;
5584
5585 i.types[this_operand] |= BaseIndex;
5586 }
5587
5588 /* Offset modifier. Add the register to the displacement string to be
5589 parsed as an immediate expression after we're done. */
5590 else if (intel_parser.op_modifier == OFFSET_FLAT)
5591 strcat (intel_parser.disp, reg->reg_name);
5592
5593 /* It's neither base nor index nor offset. */
5594 else
5595 {
5596 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
5597 i.op[this_operand].regs = reg;
5598 i.reg_operands++;
5599 }
5600
5601 /* Since registers are not part of the displacement string (except
5602 when we're parsing offset operands), we may need to remove any
5603 preceding '+' from the displacement string. */
5604 if (*intel_parser.disp != '\0'
5605 && intel_parser.op_modifier != OFFSET_FLAT)
5606 {
5607 char *s = intel_parser.disp;
5608 s += strlen (s) - 1;
5609 if (*s == '+')
5610 *s = '\0';
5611 }
5612
5613 return 1;
5614 }
5615
5616 /* e11 id */
5617 else if (cur_token.code == T_ID)
5618 {
5619 /* Add the identifier to the displacement string. */
5620 strcat (intel_parser.disp, cur_token.str);
5621 intel_match_token (T_ID);
5622
5623 /* The identifier represents a memory reference only if it's not
5624 preceded by an offset modifier. */
5625 if (intel_parser.op_modifier != OFFSET_FLAT)
5626 intel_parser.is_mem = 1;
5627
5628 return 1;
5629 }
5630
5631 /* e11 constant */
5632 else if (cur_token.code == T_CONST
5633 || cur_token.code == '-'
5634 || cur_token.code == '+')
5635 {
5636 char *save_str;
5637
5638 /* Allow constants that start with `+' or `-'. */
5639 if (cur_token.code == '-' || cur_token.code == '+')
5640 {
5641 strcat (intel_parser.disp, cur_token.str);
5642 intel_match_token (cur_token.code);
5643 if (cur_token.code != T_CONST)
5644 {
5645 as_bad (_("Syntax error. Expecting a constant. Got `%s'.\n"),
5646 cur_token.str);
5647 return 0;
5648 }
5649 }
5650
5651 save_str = (char *) malloc (strlen (cur_token.str) + 1);
5652 if (save_str == NULL)
5653 abort ();
5654 strcpy (save_str, cur_token.str);
5655
5656 /* Get the next token to check for register scaling. */
5657 intel_match_token (cur_token.code);
5658
5659 /* Check if this constant is a scaling factor for an index register. */
5660 if (cur_token.code == '*')
5661 {
5662 if (intel_match_token ('*') && cur_token.code == T_REG)
5663 {
5664 if (!intel_parser.is_mem)
5665 {
5666 as_bad (_("Register scaling only allowed in memory operands."));
5667 return 0;
5668 }
5669
5670 /* The constant is followed by `* reg', so it must be
5671 a valid scale. */
5672 if (strchr ("01248", *save_str))
5673 {
5674 i.index_reg = cur_token.reg;
5675 i.types[this_operand] |= BaseIndex;
5676
5677 /* Set the scale after setting the register (otherwise,
5678 i386_scale will complain) */
5679 i386_scale (save_str);
5680 intel_match_token (T_REG);
5681
5682 /* Since registers are not part of the displacement
5683 string, we may need to remove any preceding '+' from
5684 the displacement string. */
5685 if (*intel_parser.disp != '\0')
5686 {
5687 char *s = intel_parser.disp;
5688 s += strlen (s) - 1;
5689 if (*s == '+')
5690 *s = '\0';
5691 }
5692
5693 free (save_str);
5694
5695 return 1;
5696 }
5697 else
5698 return 0;
5699 }
5700
5701 /* The constant was not used for register scaling. Since we have
5702 already consumed the token following `*' we now need to put it
5703 back in the stream. */
5704 else
5705 intel_putback_token ();
5706 }
5707
5708 /* Add the constant to the displacement string. */
5709 strcat (intel_parser.disp, save_str);
5710 free (save_str);
5711
5712 return 1;
5713 }
5714
5715 as_bad (_("Unrecognized token '%s'"), cur_token.str);
5716 return 0;
5717 }
5718
5719 /* Match the given token against cur_token. If they match, read the next
5720 token from the operand string. */
5721 static int
5722 intel_match_token (code)
5723 int code;
5724 {
5725 if (cur_token.code == code)
5726 {
5727 intel_get_token ();
5728 return 1;
5729 }
5730 else
5731 {
5732 as_bad (_("Unexpected token `%s'\n"), cur_token.str);
5733 return 0;
5734 }
5735 }
5736
5737 /* Read a new token from intel_parser.op_string and store it in cur_token. */
5738 static void
5739 intel_get_token ()
5740 {
5741 char *end_op;
5742 const reg_entry *reg;
5743 struct intel_token new_token;
5744
5745 new_token.code = T_NIL;
5746 new_token.reg = NULL;
5747 new_token.str = NULL;
5748
5749 /* Free the memory allocated to the previous token and move
5750 cur_token to prev_token. */
5751 if (prev_token.str)
5752 free (prev_token.str);
5753
5754 prev_token = cur_token;
5755
5756 /* Skip whitespace. */
5757 while (is_space_char (*intel_parser.op_string))
5758 intel_parser.op_string++;
5759
5760 /* Return an empty token if we find nothing else on the line. */
5761 if (*intel_parser.op_string == '\0')
5762 {
5763 cur_token = new_token;
5764 return;
5765 }
5766
5767 /* The new token cannot be larger than the remainder of the operand
5768 string. */
5769 new_token.str = (char *) malloc (strlen (intel_parser.op_string) + 1);
5770 if (new_token.str == NULL)
5771 abort ();
5772 new_token.str[0] = '\0';
5773
5774 if (strchr ("0123456789", *intel_parser.op_string))
5775 {
5776 char *p = new_token.str;
5777 char *q = intel_parser.op_string;
5778 new_token.code = T_CONST;
5779
5780 /* Allow any kind of identifier char to encompass floating point and
5781 hexadecimal numbers. */
5782 while (is_identifier_char (*q))
5783 *p++ = *q++;
5784 *p = '\0';
5785
5786 /* Recognize special symbol names [0-9][bf]. */
5787 if (strlen (intel_parser.op_string) == 2
5788 && (intel_parser.op_string[1] == 'b'
5789 || intel_parser.op_string[1] == 'f'))
5790 new_token.code = T_ID;
5791 }
5792
5793 else if (strchr ("+-/*:[]()", *intel_parser.op_string))
5794 {
5795 new_token.code = *intel_parser.op_string;
5796 new_token.str[0] = *intel_parser.op_string;
5797 new_token.str[1] = '\0';
5798 }
5799
5800 else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
5801 && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
5802 {
5803 new_token.code = T_REG;
5804 new_token.reg = reg;
5805
5806 if (*intel_parser.op_string == REGISTER_PREFIX)
5807 {
5808 new_token.str[0] = REGISTER_PREFIX;
5809 new_token.str[1] = '\0';
5810 }
5811
5812 strcat (new_token.str, reg->reg_name);
5813 }
5814
5815 else if (is_identifier_char (*intel_parser.op_string))
5816 {
5817 char *p = new_token.str;
5818 char *q = intel_parser.op_string;
5819
5820 /* A '.' or '$' followed by an identifier char is an identifier.
5821 Otherwise, it's operator '.' followed by an expression. */
5822 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
5823 {
5824 new_token.code = *q;
5825 new_token.str[0] = *q;
5826 new_token.str[1] = '\0';
5827 }
5828 else
5829 {
5830 while (is_identifier_char (*q) || *q == '@')
5831 *p++ = *q++;
5832 *p = '\0';
5833
5834 if (strcasecmp (new_token.str, "BYTE") == 0)
5835 new_token.code = T_BYTE;
5836
5837 else if (strcasecmp (new_token.str, "WORD") == 0)
5838 new_token.code = T_WORD;
5839
5840 else if (strcasecmp (new_token.str, "DWORD") == 0)
5841 new_token.code = T_DWORD;
5842
5843 else if (strcasecmp (new_token.str, "QWORD") == 0)
5844 new_token.code = T_QWORD;
5845
5846 else if (strcasecmp (new_token.str, "XWORD") == 0)
5847 new_token.code = T_XWORD;
5848
5849 else if (strcasecmp (new_token.str, "PTR") == 0)
5850 new_token.code = T_PTR;
5851
5852 else if (strcasecmp (new_token.str, "SHORT") == 0)
5853 new_token.code = T_SHORT;
5854
5855 else if (strcasecmp (new_token.str, "OFFSET") == 0)
5856 {
5857 new_token.code = T_OFFSET;
5858
5859 /* ??? This is not mentioned in the MASM grammar but gcc
5860 makes use of it with -mintel-syntax. OFFSET may be
5861 followed by FLAT: */
5862 if (strncasecmp (q, " FLAT:", 6) == 0)
5863 strcat (new_token.str, " FLAT:");
5864 }
5865
5866 /* ??? This is not mentioned in the MASM grammar. */
5867 else if (strcasecmp (new_token.str, "FLAT") == 0)
5868 new_token.code = T_OFFSET;
5869
5870 else
5871 new_token.code = T_ID;
5872 }
5873 }
5874
5875 else
5876 as_bad (_("Unrecognized token `%s'\n"), intel_parser.op_string);
5877
5878 intel_parser.op_string += strlen (new_token.str);
5879 cur_token = new_token;
5880 }
5881
5882 /* Put cur_token back into the token stream and make cur_token point to
5883 prev_token. */
5884 static void
5885 intel_putback_token ()
5886 {
5887 intel_parser.op_string -= strlen (cur_token.str);
5888 free (cur_token.str);
5889 cur_token = prev_token;
5890
5891 /* Forget prev_token. */
5892 prev_token.code = T_NIL;
5893 prev_token.reg = NULL;
5894 prev_token.str = NULL;
5895 }