* mmo.c (File Layout): Correct place of misplaced blurb about
[binutils.git] / gas / config / tc-i386.c
blobb2f42dc876e5fb8eb5ef179762c28254bd01f3bf
1 /* i386.c -- Assemble code for the Intel 80386
2 Copyright 1989, 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
3 2000, 2001, 2002, 2003, 2004
4 Free Software Foundation, Inc.
6 This file is part of GAS, the GNU Assembler.
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.
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.
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. */
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 VIA PadLock support by Michal Ludvig (mludvig@suse.cz)
27 Bugs & suggestions are completely welcome. This is free software.
28 Please help us make it better. */
30 #include "as.h"
31 #include "safe-ctype.h"
32 #include "subsegs.h"
33 #include "dwarf2dbg.h"
34 #include "dw2gencfi.h"
35 #include "opcode/i386.h"
36 #include "elf/x86-64.h"
38 #ifndef REGISTER_WARNINGS
39 #define REGISTER_WARNINGS 1
40 #endif
42 #ifndef INFER_ADDR_PREFIX
43 #define INFER_ADDR_PREFIX 1
44 #endif
46 #ifndef SCALE1_WHEN_NO_INDEX
47 /* Specifying a scale factor besides 1 when there is no index is
48 futile. eg. `mov (%ebx,2),%al' does exactly the same as
49 `mov (%ebx),%al'. To slavishly follow what the programmer
50 specified, set SCALE1_WHEN_NO_INDEX to 0. */
51 #define SCALE1_WHEN_NO_INDEX 1
52 #endif
54 #ifndef DEFAULT_ARCH
55 #define DEFAULT_ARCH "i386"
56 #endif
58 #ifndef INLINE
59 #if __GNUC__ >= 2
60 #define INLINE __inline__
61 #else
62 #define INLINE
63 #endif
64 #endif
66 static INLINE unsigned int mode_from_disp_size PARAMS ((unsigned int));
67 static INLINE int fits_in_signed_byte PARAMS ((offsetT));
68 static INLINE int fits_in_unsigned_byte PARAMS ((offsetT));
69 static INLINE int fits_in_unsigned_word PARAMS ((offsetT));
70 static INLINE int fits_in_signed_word PARAMS ((offsetT));
71 static INLINE int fits_in_unsigned_long PARAMS ((offsetT));
72 static INLINE int fits_in_signed_long PARAMS ((offsetT));
73 static int smallest_imm_type PARAMS ((offsetT));
74 static offsetT offset_in_range PARAMS ((offsetT, int));
75 static int add_prefix PARAMS ((unsigned int));
76 static void set_code_flag PARAMS ((int));
77 static void set_16bit_gcc_code_flag PARAMS ((int));
78 static void set_intel_syntax PARAMS ((int));
79 static void set_cpu_arch PARAMS ((int));
80 #ifdef TE_PE
81 static void pe_directive_secrel PARAMS ((int));
82 #endif
83 static char *output_invalid PARAMS ((int c));
84 static int i386_operand PARAMS ((char *operand_string));
85 static int i386_intel_operand PARAMS ((char *operand_string, int got_a_float));
86 static const reg_entry *parse_register PARAMS ((char *reg_string,
87 char **end_op));
88 static char *parse_insn PARAMS ((char *, char *));
89 static char *parse_operands PARAMS ((char *, const char *));
90 static void swap_operands PARAMS ((void));
91 static void optimize_imm PARAMS ((void));
92 static void optimize_disp PARAMS ((void));
93 static int match_template PARAMS ((void));
94 static int check_string PARAMS ((void));
95 static int process_suffix PARAMS ((void));
96 static int check_byte_reg PARAMS ((void));
97 static int check_long_reg PARAMS ((void));
98 static int check_qword_reg PARAMS ((void));
99 static int check_word_reg PARAMS ((void));
100 static int finalize_imm PARAMS ((void));
101 static int process_operands PARAMS ((void));
102 static const seg_entry *build_modrm_byte PARAMS ((void));
103 static void output_insn PARAMS ((void));
104 static void output_branch PARAMS ((void));
105 static void output_jump PARAMS ((void));
106 static void output_interseg_jump PARAMS ((void));
107 static void output_imm PARAMS ((fragS *insn_start_frag,
108 offsetT insn_start_off));
109 static void output_disp PARAMS ((fragS *insn_start_frag,
110 offsetT insn_start_off));
111 #ifndef I386COFF
112 static void s_bss PARAMS ((int));
113 #endif
115 static const char *default_arch = DEFAULT_ARCH;
117 /* 'md_assemble ()' gathers together information and puts it into a
118 i386_insn. */
120 union i386_op
122 expressionS *disps;
123 expressionS *imms;
124 const reg_entry *regs;
127 struct _i386_insn
129 /* TM holds the template for the insn were currently assembling. */
130 template tm;
132 /* SUFFIX holds the instruction mnemonic suffix if given.
133 (e.g. 'l' for 'movl') */
134 char suffix;
136 /* OPERANDS gives the number of given operands. */
137 unsigned int operands;
139 /* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number
140 of given register, displacement, memory operands and immediate
141 operands. */
142 unsigned int reg_operands, disp_operands, mem_operands, imm_operands;
144 /* TYPES [i] is the type (see above #defines) which tells us how to
145 use OP[i] for the corresponding operand. */
146 unsigned int types[MAX_OPERANDS];
148 /* Displacement expression, immediate expression, or register for each
149 operand. */
150 union i386_op op[MAX_OPERANDS];
152 /* Flags for operands. */
153 unsigned int flags[MAX_OPERANDS];
154 #define Operand_PCrel 1
156 /* Relocation type for operand */
157 enum bfd_reloc_code_real reloc[MAX_OPERANDS];
159 /* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode
160 the base index byte below. */
161 const reg_entry *base_reg;
162 const reg_entry *index_reg;
163 unsigned int log2_scale_factor;
165 /* SEG gives the seg_entries of this insn. They are zero unless
166 explicit segment overrides are given. */
167 const seg_entry *seg[2];
169 /* PREFIX holds all the given prefix opcodes (usually null).
170 PREFIXES is the number of prefix opcodes. */
171 unsigned int prefixes;
172 unsigned char prefix[MAX_PREFIXES];
174 /* RM and SIB are the modrm byte and the sib byte where the
175 addressing modes of this insn are encoded. */
177 modrm_byte rm;
178 rex_byte rex;
179 sib_byte sib;
182 typedef struct _i386_insn i386_insn;
184 /* List of chars besides those in app.c:symbol_chars that can start an
185 operand. Used to prevent the scrubber eating vital white-space. */
186 const char extra_symbol_chars[] = "*%-(["
187 #ifdef LEX_AT
189 #endif
190 #ifdef LEX_QM
192 #endif
195 #if (defined (TE_I386AIX) \
196 || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) \
197 && !defined (TE_LINUX) \
198 && !defined (TE_NETWARE) \
199 && !defined (TE_FreeBSD) \
200 && !defined (TE_NetBSD)))
201 /* This array holds the chars that always start a comment. If the
202 pre-processor is disabled, these aren't very useful. */
203 const char comment_chars[] = "#/";
204 #define PREFIX_SEPARATOR '\\'
206 /* This array holds the chars that only start a comment at the beginning of
207 a line. If the line seems to have the form '# 123 filename'
208 .line and .file directives will appear in the pre-processed output.
209 Note that input_file.c hand checks for '#' at the beginning of the
210 first line of the input file. This is because the compiler outputs
211 #NO_APP at the beginning of its output.
212 Also note that comments started like this one will always work if
213 '/' isn't otherwise defined. */
214 const char line_comment_chars[] = "#";
216 #else
217 /* Putting '/' here makes it impossible to use the divide operator.
218 However, we need it for compatibility with SVR4 systems. */
219 const char comment_chars[] = "#";
220 #define PREFIX_SEPARATOR '/'
222 const char line_comment_chars[] = "/#";
223 #endif
225 const char line_separator_chars[] = ";";
227 /* Chars that can be used to separate mant from exp in floating point
228 nums. */
229 const char EXP_CHARS[] = "eE";
231 /* Chars that mean this number is a floating point constant
232 As in 0f12.456
233 or 0d1.2345e12. */
234 const char FLT_CHARS[] = "fFdDxX";
236 /* Tables for lexical analysis. */
237 static char mnemonic_chars[256];
238 static char register_chars[256];
239 static char operand_chars[256];
240 static char identifier_chars[256];
241 static char digit_chars[256];
243 /* Lexical macros. */
244 #define is_mnemonic_char(x) (mnemonic_chars[(unsigned char) x])
245 #define is_operand_char(x) (operand_chars[(unsigned char) x])
246 #define is_register_char(x) (register_chars[(unsigned char) x])
247 #define is_space_char(x) ((x) == ' ')
248 #define is_identifier_char(x) (identifier_chars[(unsigned char) x])
249 #define is_digit_char(x) (digit_chars[(unsigned char) x])
251 /* All non-digit non-letter characters that may occur in an operand. */
252 static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
254 /* md_assemble() always leaves the strings it's passed unaltered. To
255 effect this we maintain a stack of saved characters that we've smashed
256 with '\0's (indicating end of strings for various sub-fields of the
257 assembler instruction). */
258 static char save_stack[32];
259 static char *save_stack_p;
260 #define END_STRING_AND_SAVE(s) \
261 do { *save_stack_p++ = *(s); *(s) = '\0'; } while (0)
262 #define RESTORE_END_STRING(s) \
263 do { *(s) = *--save_stack_p; } while (0)
265 /* The instruction we're assembling. */
266 static i386_insn i;
268 /* Possible templates for current insn. */
269 static const templates *current_templates;
271 /* Per instruction expressionS buffers: 2 displacements & 2 immediate max. */
272 static expressionS disp_expressions[2], im_expressions[2];
274 /* Current operand we are working on. */
275 static int this_operand;
277 /* We support four different modes. FLAG_CODE variable is used to distinguish
278 these. */
280 enum flag_code {
281 CODE_32BIT,
282 CODE_16BIT,
283 CODE_64BIT };
284 #define NUM_FLAG_CODE ((int) CODE_64BIT + 1)
286 static enum flag_code flag_code;
287 static int use_rela_relocations = 0;
289 /* The names used to print error messages. */
290 static const char *flag_code_names[] =
292 "32",
293 "16",
294 "64"
297 /* 1 for intel syntax,
298 0 if att syntax. */
299 static int intel_syntax = 0;
301 /* 1 if register prefix % not required. */
302 static int allow_naked_reg = 0;
304 /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
305 leave, push, and pop instructions so that gcc has the same stack
306 frame as in 32 bit mode. */
307 static char stackop_size = '\0';
309 /* Non-zero to optimize code alignment. */
310 int optimize_align_code = 1;
312 /* Non-zero to quieten some warnings. */
313 static int quiet_warnings = 0;
315 /* CPU name. */
316 static const char *cpu_arch_name = NULL;
318 /* CPU feature flags. */
319 static unsigned int cpu_arch_flags = CpuUnknownFlags | CpuNo64;
321 /* If set, conditional jumps are not automatically promoted to handle
322 larger than a byte offset. */
323 static unsigned int no_cond_jump_promotion = 0;
325 /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
326 symbolS *GOT_symbol;
328 /* The dwarf2 return column, adjusted for 32 or 64 bit. */
329 unsigned int x86_dwarf2_return_column;
331 /* The dwarf2 data alignment, adjusted for 32 or 64 bit. */
332 int x86_cie_data_alignment;
334 /* Interface to relax_segment.
335 There are 3 major relax states for 386 jump insns because the
336 different types of jumps add different sizes to frags when we're
337 figuring out what sort of jump to choose to reach a given label. */
339 /* Types. */
340 #define UNCOND_JUMP 0
341 #define COND_JUMP 1
342 #define COND_JUMP86 2
344 /* Sizes. */
345 #define CODE16 1
346 #define SMALL 0
347 #define SMALL16 (SMALL | CODE16)
348 #define BIG 2
349 #define BIG16 (BIG | CODE16)
351 #ifndef INLINE
352 #ifdef __GNUC__
353 #define INLINE __inline__
354 #else
355 #define INLINE
356 #endif
357 #endif
359 #define ENCODE_RELAX_STATE(type, size) \
360 ((relax_substateT) (((type) << 2) | (size)))
361 #define TYPE_FROM_RELAX_STATE(s) \
362 ((s) >> 2)
363 #define DISP_SIZE_FROM_RELAX_STATE(s) \
364 ((((s) & 3) == BIG ? 4 : (((s) & 3) == BIG16 ? 2 : 1)))
366 /* This table is used by relax_frag to promote short jumps to long
367 ones where necessary. SMALL (short) jumps may be promoted to BIG
368 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
369 don't allow a short jump in a 32 bit code segment to be promoted to
370 a 16 bit offset jump because it's slower (requires data size
371 prefix), and doesn't work, unless the destination is in the bottom
372 64k of the code segment (The top 16 bits of eip are zeroed). */
374 const relax_typeS md_relax_table[] =
376 /* The fields are:
377 1) most positive reach of this state,
378 2) most negative reach of this state,
379 3) how many bytes this mode will have in the variable part of the frag
380 4) which index into the table to try if we can't fit into this one. */
382 /* UNCOND_JUMP states. */
383 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
384 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
385 /* dword jmp adds 4 bytes to frag:
386 0 extra opcode bytes, 4 displacement bytes. */
387 {0, 0, 4, 0},
388 /* word jmp adds 2 byte2 to frag:
389 0 extra opcode bytes, 2 displacement bytes. */
390 {0, 0, 2, 0},
392 /* COND_JUMP states. */
393 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
394 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
395 /* dword conditionals adds 5 bytes to frag:
396 1 extra opcode byte, 4 displacement bytes. */
397 {0, 0, 5, 0},
398 /* word conditionals add 3 bytes to frag:
399 1 extra opcode byte, 2 displacement bytes. */
400 {0, 0, 3, 0},
402 /* COND_JUMP86 states. */
403 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG)},
404 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG16)},
405 /* dword conditionals adds 5 bytes to frag:
406 1 extra opcode byte, 4 displacement bytes. */
407 {0, 0, 5, 0},
408 /* word conditionals add 4 bytes to frag:
409 1 displacement byte and a 3 byte long branch insn. */
410 {0, 0, 4, 0}
413 static const arch_entry cpu_arch[] = {
414 {"i8086", Cpu086 },
415 {"i186", Cpu086|Cpu186 },
416 {"i286", Cpu086|Cpu186|Cpu286 },
417 {"i386", Cpu086|Cpu186|Cpu286|Cpu386 },
418 {"i486", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486 },
419 {"i586", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
420 {"i686", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
421 {"pentium", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
422 {"pentiumpro",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
423 {"pentium4", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX|CpuSSE|CpuSSE2 },
424 {"k6", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow },
425 {"athlon", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuMMX|Cpu3dnow },
426 {"sledgehammer",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuSledgehammer|CpuMMX|Cpu3dnow|CpuSSE|CpuSSE2 },
427 {NULL, 0 }
430 const pseudo_typeS md_pseudo_table[] =
432 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
433 {"align", s_align_bytes, 0},
434 #else
435 {"align", s_align_ptwo, 0},
436 #endif
437 {"arch", set_cpu_arch, 0},
438 #ifndef I386COFF
439 {"bss", s_bss, 0},
440 #endif
441 {"ffloat", float_cons, 'f'},
442 {"dfloat", float_cons, 'd'},
443 {"tfloat", float_cons, 'x'},
444 {"value", cons, 2},
445 {"noopt", s_ignore, 0},
446 {"optim", s_ignore, 0},
447 {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
448 {"code16", set_code_flag, CODE_16BIT},
449 {"code32", set_code_flag, CODE_32BIT},
450 {"code64", set_code_flag, CODE_64BIT},
451 {"intel_syntax", set_intel_syntax, 1},
452 {"att_syntax", set_intel_syntax, 0},
453 {"file", (void (*) PARAMS ((int))) dwarf2_directive_file, 0},
454 {"loc", dwarf2_directive_loc, 0},
455 #ifdef TE_PE
456 {"secrel32", pe_directive_secrel, 0},
457 #endif
458 {0, 0, 0}
461 /* For interface with expression (). */
462 extern char *input_line_pointer;
464 /* Hash table for instruction mnemonic lookup. */
465 static struct hash_control *op_hash;
467 /* Hash table for register lookup. */
468 static struct hash_control *reg_hash;
470 void
471 i386_align_code (fragP, count)
472 fragS *fragP;
473 int count;
475 /* Various efficient no-op patterns for aligning code labels.
476 Note: Don't try to assemble the instructions in the comments.
477 0L and 0w are not legal. */
478 static const char f32_1[] =
479 {0x90}; /* nop */
480 static const char f32_2[] =
481 {0x89,0xf6}; /* movl %esi,%esi */
482 static const char f32_3[] =
483 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
484 static const char f32_4[] =
485 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
486 static const char f32_5[] =
487 {0x90, /* nop */
488 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
489 static const char f32_6[] =
490 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
491 static const char f32_7[] =
492 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
493 static const char f32_8[] =
494 {0x90, /* nop */
495 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
496 static const char f32_9[] =
497 {0x89,0xf6, /* movl %esi,%esi */
498 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
499 static const char f32_10[] =
500 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
501 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
502 static const char f32_11[] =
503 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
504 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
505 static const char f32_12[] =
506 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
507 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
508 static const char f32_13[] =
509 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
510 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
511 static const char f32_14[] =
512 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
513 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
514 static const char f32_15[] =
515 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
516 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
517 static const char f16_3[] =
518 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
519 static const char f16_4[] =
520 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
521 static const char f16_5[] =
522 {0x90, /* nop */
523 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
524 static const char f16_6[] =
525 {0x89,0xf6, /* mov %si,%si */
526 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
527 static const char f16_7[] =
528 {0x8d,0x74,0x00, /* lea 0(%si),%si */
529 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
530 static const char f16_8[] =
531 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
532 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
533 static const char *const f32_patt[] = {
534 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
535 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
537 static const char *const f16_patt[] = {
538 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
539 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
542 if (count <= 0 || count > 15)
543 return;
545 /* The recommended way to pad 64bit code is to use NOPs preceded by
546 maximally four 0x66 prefixes. Balance the size of nops. */
547 if (flag_code == CODE_64BIT)
549 int i;
550 int nnops = (count + 3) / 4;
551 int len = count / nnops;
552 int remains = count - nnops * len;
553 int pos = 0;
555 for (i = 0; i < remains; i++)
557 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len);
558 fragP->fr_literal[fragP->fr_fix + pos + len] = 0x90;
559 pos += len + 1;
561 for (; i < nnops; i++)
563 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len - 1);
564 fragP->fr_literal[fragP->fr_fix + pos + len - 1] = 0x90;
565 pos += len;
568 else
569 if (flag_code == CODE_16BIT)
571 memcpy (fragP->fr_literal + fragP->fr_fix,
572 f16_patt[count - 1], count);
573 if (count > 8)
574 /* Adjust jump offset. */
575 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
577 else
578 memcpy (fragP->fr_literal + fragP->fr_fix,
579 f32_patt[count - 1], count);
580 fragP->fr_var = count;
583 static INLINE unsigned int
584 mode_from_disp_size (t)
585 unsigned int t;
587 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
590 static INLINE int
591 fits_in_signed_byte (num)
592 offsetT num;
594 return (num >= -128) && (num <= 127);
597 static INLINE int
598 fits_in_unsigned_byte (num)
599 offsetT num;
601 return (num & 0xff) == num;
604 static INLINE int
605 fits_in_unsigned_word (num)
606 offsetT num;
608 return (num & 0xffff) == num;
611 static INLINE int
612 fits_in_signed_word (num)
613 offsetT num;
615 return (-32768 <= num) && (num <= 32767);
617 static INLINE int
618 fits_in_signed_long (num)
619 offsetT num ATTRIBUTE_UNUSED;
621 #ifndef BFD64
622 return 1;
623 #else
624 return (!(((offsetT) -1 << 31) & num)
625 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
626 #endif
627 } /* fits_in_signed_long() */
628 static INLINE int
629 fits_in_unsigned_long (num)
630 offsetT num ATTRIBUTE_UNUSED;
632 #ifndef BFD64
633 return 1;
634 #else
635 return (num & (((offsetT) 2 << 31) - 1)) == num;
636 #endif
637 } /* fits_in_unsigned_long() */
639 static int
640 smallest_imm_type (num)
641 offsetT num;
643 if (cpu_arch_flags != (Cpu086 | Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64))
645 /* This code is disabled on the 486 because all the Imm1 forms
646 in the opcode table are slower on the i486. They're the
647 versions with the implicitly specified single-position
648 displacement, which has another syntax if you really want to
649 use that form. */
650 if (num == 1)
651 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
653 return (fits_in_signed_byte (num)
654 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
655 : fits_in_unsigned_byte (num)
656 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
657 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
658 ? (Imm16 | Imm32 | Imm32S | Imm64)
659 : fits_in_signed_long (num)
660 ? (Imm32 | Imm32S | Imm64)
661 : fits_in_unsigned_long (num)
662 ? (Imm32 | Imm64)
663 : Imm64);
666 static offsetT
667 offset_in_range (val, size)
668 offsetT val;
669 int size;
671 addressT mask;
673 switch (size)
675 case 1: mask = ((addressT) 1 << 8) - 1; break;
676 case 2: mask = ((addressT) 1 << 16) - 1; break;
677 case 4: mask = ((addressT) 2 << 31) - 1; break;
678 #ifdef BFD64
679 case 8: mask = ((addressT) 2 << 63) - 1; break;
680 #endif
681 default: abort ();
684 /* If BFD64, sign extend val. */
685 if (!use_rela_relocations)
686 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
687 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
689 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
691 char buf1[40], buf2[40];
693 sprint_value (buf1, val);
694 sprint_value (buf2, val & mask);
695 as_warn (_("%s shortened to %s"), buf1, buf2);
697 return val & mask;
700 /* Returns 0 if attempting to add a prefix where one from the same
701 class already exists, 1 if non rep/repne added, 2 if rep/repne
702 added. */
703 static int
704 add_prefix (prefix)
705 unsigned int prefix;
707 int ret = 1;
708 int q;
710 if (prefix >= REX_OPCODE && prefix < REX_OPCODE + 16
711 && flag_code == CODE_64BIT)
712 q = REX_PREFIX;
713 else
714 switch (prefix)
716 default:
717 abort ();
719 case CS_PREFIX_OPCODE:
720 case DS_PREFIX_OPCODE:
721 case ES_PREFIX_OPCODE:
722 case FS_PREFIX_OPCODE:
723 case GS_PREFIX_OPCODE:
724 case SS_PREFIX_OPCODE:
725 q = SEG_PREFIX;
726 break;
728 case REPNE_PREFIX_OPCODE:
729 case REPE_PREFIX_OPCODE:
730 ret = 2;
731 /* fall thru */
732 case LOCK_PREFIX_OPCODE:
733 q = LOCKREP_PREFIX;
734 break;
736 case FWAIT_OPCODE:
737 q = WAIT_PREFIX;
738 break;
740 case ADDR_PREFIX_OPCODE:
741 q = ADDR_PREFIX;
742 break;
744 case DATA_PREFIX_OPCODE:
745 q = DATA_PREFIX;
746 break;
749 if (i.prefix[q] != 0)
751 as_bad (_("same type of prefix used twice"));
752 return 0;
755 i.prefixes += 1;
756 i.prefix[q] = prefix;
757 return ret;
760 static void
761 set_code_flag (value)
762 int value;
764 flag_code = value;
765 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
766 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
767 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
769 as_bad (_("64bit mode not supported on this CPU."));
771 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
773 as_bad (_("32bit mode not supported on this CPU."));
775 stackop_size = '\0';
778 static void
779 set_16bit_gcc_code_flag (new_code_flag)
780 int new_code_flag;
782 flag_code = new_code_flag;
783 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
784 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
785 stackop_size = 'l';
788 static void
789 set_intel_syntax (syntax_flag)
790 int syntax_flag;
792 /* Find out if register prefixing is specified. */
793 int ask_naked_reg = 0;
795 SKIP_WHITESPACE ();
796 if (!is_end_of_line[(unsigned char) *input_line_pointer])
798 char *string = input_line_pointer;
799 int e = get_symbol_end ();
801 if (strcmp (string, "prefix") == 0)
802 ask_naked_reg = 1;
803 else if (strcmp (string, "noprefix") == 0)
804 ask_naked_reg = -1;
805 else
806 as_bad (_("bad argument to syntax directive."));
807 *input_line_pointer = e;
809 demand_empty_rest_of_line ();
811 intel_syntax = syntax_flag;
813 if (ask_naked_reg == 0)
814 allow_naked_reg = (intel_syntax
815 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
816 else
817 allow_naked_reg = (ask_naked_reg < 0);
820 static void
821 set_cpu_arch (dummy)
822 int dummy ATTRIBUTE_UNUSED;
824 SKIP_WHITESPACE ();
826 if (!is_end_of_line[(unsigned char) *input_line_pointer])
828 char *string = input_line_pointer;
829 int e = get_symbol_end ();
830 int i;
832 for (i = 0; cpu_arch[i].name; i++)
834 if (strcmp (string, cpu_arch[i].name) == 0)
836 cpu_arch_name = cpu_arch[i].name;
837 cpu_arch_flags = (cpu_arch[i].flags
838 | (flag_code == CODE_64BIT ? Cpu64 : CpuNo64));
839 break;
842 if (!cpu_arch[i].name)
843 as_bad (_("no such architecture: `%s'"), string);
845 *input_line_pointer = e;
847 else
848 as_bad (_("missing cpu architecture"));
850 no_cond_jump_promotion = 0;
851 if (*input_line_pointer == ','
852 && !is_end_of_line[(unsigned char) input_line_pointer[1]])
854 char *string = ++input_line_pointer;
855 int e = get_symbol_end ();
857 if (strcmp (string, "nojumps") == 0)
858 no_cond_jump_promotion = 1;
859 else if (strcmp (string, "jumps") == 0)
861 else
862 as_bad (_("no such architecture modifier: `%s'"), string);
864 *input_line_pointer = e;
867 demand_empty_rest_of_line ();
870 unsigned long
871 i386_mach ()
873 if (!strcmp (default_arch, "x86_64"))
874 return bfd_mach_x86_64;
875 else if (!strcmp (default_arch, "i386"))
876 return bfd_mach_i386_i386;
877 else
878 as_fatal (_("Unknown architecture"));
881 void
882 md_begin ()
884 const char *hash_err;
886 /* Initialize op_hash hash table. */
887 op_hash = hash_new ();
890 const template *optab;
891 templates *core_optab;
893 /* Setup for loop. */
894 optab = i386_optab;
895 core_optab = (templates *) xmalloc (sizeof (templates));
896 core_optab->start = optab;
898 while (1)
900 ++optab;
901 if (optab->name == NULL
902 || strcmp (optab->name, (optab - 1)->name) != 0)
904 /* different name --> ship out current template list;
905 add to hash table; & begin anew. */
906 core_optab->end = optab;
907 hash_err = hash_insert (op_hash,
908 (optab - 1)->name,
909 (PTR) core_optab);
910 if (hash_err)
912 as_fatal (_("Internal Error: Can't hash %s: %s"),
913 (optab - 1)->name,
914 hash_err);
916 if (optab->name == NULL)
917 break;
918 core_optab = (templates *) xmalloc (sizeof (templates));
919 core_optab->start = optab;
924 /* Initialize reg_hash hash table. */
925 reg_hash = hash_new ();
927 const reg_entry *regtab;
929 for (regtab = i386_regtab;
930 regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
931 regtab++)
933 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
934 if (hash_err)
935 as_fatal (_("Internal Error: Can't hash %s: %s"),
936 regtab->reg_name,
937 hash_err);
941 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
943 int c;
944 char *p;
946 for (c = 0; c < 256; c++)
948 if (ISDIGIT (c))
950 digit_chars[c] = c;
951 mnemonic_chars[c] = c;
952 register_chars[c] = c;
953 operand_chars[c] = c;
955 else if (ISLOWER (c))
957 mnemonic_chars[c] = c;
958 register_chars[c] = c;
959 operand_chars[c] = c;
961 else if (ISUPPER (c))
963 mnemonic_chars[c] = TOLOWER (c);
964 register_chars[c] = mnemonic_chars[c];
965 operand_chars[c] = c;
968 if (ISALPHA (c) || ISDIGIT (c))
969 identifier_chars[c] = c;
970 else if (c >= 128)
972 identifier_chars[c] = c;
973 operand_chars[c] = c;
977 #ifdef LEX_AT
978 identifier_chars['@'] = '@';
979 #endif
980 #ifdef LEX_QM
981 identifier_chars['?'] = '?';
982 operand_chars['?'] = '?';
983 #endif
984 digit_chars['-'] = '-';
985 identifier_chars['_'] = '_';
986 identifier_chars['.'] = '.';
988 for (p = operand_special_chars; *p != '\0'; p++)
989 operand_chars[(unsigned char) *p] = *p;
992 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
993 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
995 record_alignment (text_section, 2);
996 record_alignment (data_section, 2);
997 record_alignment (bss_section, 2);
999 #endif
1001 if (flag_code == CODE_64BIT)
1003 x86_dwarf2_return_column = 16;
1004 x86_cie_data_alignment = -8;
1006 else
1008 x86_dwarf2_return_column = 8;
1009 x86_cie_data_alignment = -4;
1013 void
1014 i386_print_statistics (file)
1015 FILE *file;
1017 hash_print_statistics (file, "i386 opcode", op_hash);
1018 hash_print_statistics (file, "i386 register", reg_hash);
1021 #ifdef DEBUG386
1023 /* Debugging routines for md_assemble. */
1024 static void pi PARAMS ((char *, i386_insn *));
1025 static void pte PARAMS ((template *));
1026 static void pt PARAMS ((unsigned int));
1027 static void pe PARAMS ((expressionS *));
1028 static void ps PARAMS ((symbolS *));
1030 static void
1031 pi (line, x)
1032 char *line;
1033 i386_insn *x;
1035 unsigned int i;
1037 fprintf (stdout, "%s: template ", line);
1038 pte (&x->tm);
1039 fprintf (stdout, " address: base %s index %s scale %x\n",
1040 x->base_reg ? x->base_reg->reg_name : "none",
1041 x->index_reg ? x->index_reg->reg_name : "none",
1042 x->log2_scale_factor);
1043 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
1044 x->rm.mode, x->rm.reg, x->rm.regmem);
1045 fprintf (stdout, " sib: base %x index %x scale %x\n",
1046 x->sib.base, x->sib.index, x->sib.scale);
1047 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
1048 (x->rex & REX_MODE64) != 0,
1049 (x->rex & REX_EXTX) != 0,
1050 (x->rex & REX_EXTY) != 0,
1051 (x->rex & REX_EXTZ) != 0);
1052 for (i = 0; i < x->operands; i++)
1054 fprintf (stdout, " #%d: ", i + 1);
1055 pt (x->types[i]);
1056 fprintf (stdout, "\n");
1057 if (x->types[i]
1058 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
1059 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
1060 if (x->types[i] & Imm)
1061 pe (x->op[i].imms);
1062 if (x->types[i] & Disp)
1063 pe (x->op[i].disps);
1067 static void
1068 pte (t)
1069 template *t;
1071 unsigned int i;
1072 fprintf (stdout, " %d operands ", t->operands);
1073 fprintf (stdout, "opcode %x ", t->base_opcode);
1074 if (t->extension_opcode != None)
1075 fprintf (stdout, "ext %x ", t->extension_opcode);
1076 if (t->opcode_modifier & D)
1077 fprintf (stdout, "D");
1078 if (t->opcode_modifier & W)
1079 fprintf (stdout, "W");
1080 fprintf (stdout, "\n");
1081 for (i = 0; i < t->operands; i++)
1083 fprintf (stdout, " #%d type ", i + 1);
1084 pt (t->operand_types[i]);
1085 fprintf (stdout, "\n");
1089 static void
1090 pe (e)
1091 expressionS *e;
1093 fprintf (stdout, " operation %d\n", e->X_op);
1094 fprintf (stdout, " add_number %ld (%lx)\n",
1095 (long) e->X_add_number, (long) e->X_add_number);
1096 if (e->X_add_symbol)
1098 fprintf (stdout, " add_symbol ");
1099 ps (e->X_add_symbol);
1100 fprintf (stdout, "\n");
1102 if (e->X_op_symbol)
1104 fprintf (stdout, " op_symbol ");
1105 ps (e->X_op_symbol);
1106 fprintf (stdout, "\n");
1110 static void
1111 ps (s)
1112 symbolS *s;
1114 fprintf (stdout, "%s type %s%s",
1115 S_GET_NAME (s),
1116 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
1117 segment_name (S_GET_SEGMENT (s)));
1120 struct type_name
1122 unsigned int mask;
1123 char *tname;
1126 static const type_names[] =
1128 { Reg8, "r8" },
1129 { Reg16, "r16" },
1130 { Reg32, "r32" },
1131 { Reg64, "r64" },
1132 { Imm8, "i8" },
1133 { Imm8S, "i8s" },
1134 { Imm16, "i16" },
1135 { Imm32, "i32" },
1136 { Imm32S, "i32s" },
1137 { Imm64, "i64" },
1138 { Imm1, "i1" },
1139 { BaseIndex, "BaseIndex" },
1140 { Disp8, "d8" },
1141 { Disp16, "d16" },
1142 { Disp32, "d32" },
1143 { Disp32S, "d32s" },
1144 { Disp64, "d64" },
1145 { InOutPortReg, "InOutPortReg" },
1146 { ShiftCount, "ShiftCount" },
1147 { Control, "control reg" },
1148 { Test, "test reg" },
1149 { Debug, "debug reg" },
1150 { FloatReg, "FReg" },
1151 { FloatAcc, "FAcc" },
1152 { SReg2, "SReg2" },
1153 { SReg3, "SReg3" },
1154 { Acc, "Acc" },
1155 { JumpAbsolute, "Jump Absolute" },
1156 { RegMMX, "rMMX" },
1157 { RegXMM, "rXMM" },
1158 { EsSeg, "es" },
1159 { 0, "" }
1162 static void
1163 pt (t)
1164 unsigned int t;
1166 const struct type_name *ty;
1168 for (ty = type_names; ty->mask; ty++)
1169 if (t & ty->mask)
1170 fprintf (stdout, "%s, ", ty->tname);
1171 fflush (stdout);
1174 #endif /* DEBUG386 */
1176 static bfd_reloc_code_real_type reloc
1177 PARAMS ((int, int, int, bfd_reloc_code_real_type));
1179 static bfd_reloc_code_real_type
1180 reloc (size, pcrel, sign, other)
1181 int size;
1182 int pcrel;
1183 int sign;
1184 bfd_reloc_code_real_type other;
1186 if (other != NO_RELOC)
1187 return other;
1189 if (pcrel)
1191 if (!sign)
1192 as_bad (_("There are no unsigned pc-relative relocations"));
1193 switch (size)
1195 case 1: return BFD_RELOC_8_PCREL;
1196 case 2: return BFD_RELOC_16_PCREL;
1197 case 4: return BFD_RELOC_32_PCREL;
1199 as_bad (_("can not do %d byte pc-relative relocation"), size);
1201 else
1203 if (sign)
1204 switch (size)
1206 case 4: return BFD_RELOC_X86_64_32S;
1208 else
1209 switch (size)
1211 case 1: return BFD_RELOC_8;
1212 case 2: return BFD_RELOC_16;
1213 case 4: return BFD_RELOC_32;
1214 case 8: return BFD_RELOC_64;
1216 as_bad (_("can not do %s %d byte relocation"),
1217 sign ? "signed" : "unsigned", size);
1220 abort ();
1221 return BFD_RELOC_NONE;
1224 /* Here we decide which fixups can be adjusted to make them relative to
1225 the beginning of the section instead of the symbol. Basically we need
1226 to make sure that the dynamic relocations are done correctly, so in
1227 some cases we force the original symbol to be used. */
1230 tc_i386_fix_adjustable (fixP)
1231 fixS *fixP ATTRIBUTE_UNUSED;
1233 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1234 if (OUTPUT_FLAVOR != bfd_target_elf_flavour)
1235 return 1;
1237 /* Don't adjust pc-relative references to merge sections in 64-bit
1238 mode. */
1239 if (use_rela_relocations
1240 && (S_GET_SEGMENT (fixP->fx_addsy)->flags & SEC_MERGE) != 0
1241 && fixP->fx_pcrel)
1242 return 0;
1244 /* The x86_64 GOTPCREL are represented as 32bit PCrel relocations
1245 and changed later by validate_fix. */
1246 if (GOT_symbol && fixP->fx_subsy == GOT_symbol
1247 && fixP->fx_r_type == BFD_RELOC_32_PCREL)
1248 return 0;
1250 /* adjust_reloc_syms doesn't know about the GOT. */
1251 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1252 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1253 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1254 || fixP->fx_r_type == BFD_RELOC_386_TLS_GD
1255 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDM
1256 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDO_32
1257 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE_32
1258 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE
1259 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTIE
1260 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE_32
1261 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE
1262 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1263 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1264 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
1265 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSGD
1266 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSLD
1267 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF32
1268 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTTPOFF
1269 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF32
1270 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1271 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1272 return 0;
1273 #endif
1274 return 1;
1277 static int intel_float_operand PARAMS ((const char *mnemonic));
1279 static int
1280 intel_float_operand (mnemonic)
1281 const char *mnemonic;
1283 if (mnemonic[0] == 'f' && mnemonic[1] == 'i')
1284 return 2;
1286 if (mnemonic[0] == 'f')
1287 return 1;
1289 return 0;
1292 /* This is the guts of the machine-dependent assembler. LINE points to a
1293 machine dependent instruction. This function is supposed to emit
1294 the frags/bytes it assembles to. */
1296 void
1297 md_assemble (line)
1298 char *line;
1300 int j;
1301 char mnemonic[MAX_MNEM_SIZE];
1303 /* Initialize globals. */
1304 memset (&i, '\0', sizeof (i));
1305 for (j = 0; j < MAX_OPERANDS; j++)
1306 i.reloc[j] = NO_RELOC;
1307 memset (disp_expressions, '\0', sizeof (disp_expressions));
1308 memset (im_expressions, '\0', sizeof (im_expressions));
1309 save_stack_p = save_stack;
1311 /* First parse an instruction mnemonic & call i386_operand for the operands.
1312 We assume that the scrubber has arranged it so that line[0] is the valid
1313 start of a (possibly prefixed) mnemonic. */
1315 line = parse_insn (line, mnemonic);
1316 if (line == NULL)
1317 return;
1319 line = parse_operands (line, mnemonic);
1320 if (line == NULL)
1321 return;
1323 /* Now we've parsed the mnemonic into a set of templates, and have the
1324 operands at hand. */
1326 /* All intel opcodes have reversed operands except for "bound" and
1327 "enter". We also don't reverse intersegment "jmp" and "call"
1328 instructions with 2 immediate operands so that the immediate segment
1329 precedes the offset, as it does when in AT&T mode. "enter" and the
1330 intersegment "jmp" and "call" instructions are the only ones that
1331 have two immediate operands. */
1332 if (intel_syntax && i.operands > 1
1333 && (strcmp (mnemonic, "bound") != 0)
1334 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1335 swap_operands ();
1337 if (i.imm_operands)
1338 optimize_imm ();
1340 if (i.disp_operands)
1341 optimize_disp ();
1343 /* Next, we find a template that matches the given insn,
1344 making sure the overlap of the given operands types is consistent
1345 with the template operand types. */
1347 if (!match_template ())
1348 return;
1350 if (intel_syntax)
1352 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1353 if (SYSV386_COMPAT
1354 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1355 i.tm.base_opcode ^= FloatR;
1357 /* Zap movzx and movsx suffix. The suffix may have been set from
1358 "word ptr" or "byte ptr" on the source operand, but we'll use
1359 the suffix later to choose the destination register. */
1360 if ((i.tm.base_opcode & ~9) == 0x0fb6)
1361 i.suffix = 0;
1364 if (i.tm.opcode_modifier & FWait)
1365 if (!add_prefix (FWAIT_OPCODE))
1366 return;
1368 /* Check string instruction segment overrides. */
1369 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1371 if (!check_string ())
1372 return;
1375 if (!process_suffix ())
1376 return;
1378 /* Make still unresolved immediate matches conform to size of immediate
1379 given in i.suffix. */
1380 if (!finalize_imm ())
1381 return;
1383 if (i.types[0] & Imm1)
1384 i.imm_operands = 0; /* kludge for shift insns. */
1385 if (i.types[0] & ImplicitRegister)
1386 i.reg_operands--;
1387 if (i.types[1] & ImplicitRegister)
1388 i.reg_operands--;
1389 if (i.types[2] & ImplicitRegister)
1390 i.reg_operands--;
1392 if (i.tm.opcode_modifier & ImmExt)
1394 expressionS *exp;
1396 if ((i.tm.cpu_flags & CpuPNI) && i.operands > 0)
1398 /* These Intel Prescott New Instructions have the fixed
1399 operands with an opcode suffix which is coded in the same
1400 place as an 8-bit immediate field would be. Here we check
1401 those operands and remove them afterwards. */
1402 unsigned int x;
1404 for (x = 0; x < i.operands; x++)
1405 if (i.op[x].regs->reg_num != x)
1406 as_bad (_("can't use register '%%%s' as operand %d in '%s'."),
1407 i.op[x].regs->reg_name, x + 1, i.tm.name);
1408 i.operands = 0;
1411 /* These AMD 3DNow! and Intel Katmai New Instructions have an
1412 opcode suffix which is coded in the same place as an 8-bit
1413 immediate field would be. Here we fake an 8-bit immediate
1414 operand from the opcode suffix stored in tm.extension_opcode. */
1416 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
1418 exp = &im_expressions[i.imm_operands++];
1419 i.op[i.operands].imms = exp;
1420 i.types[i.operands++] = Imm8;
1421 exp->X_op = O_constant;
1422 exp->X_add_number = i.tm.extension_opcode;
1423 i.tm.extension_opcode = None;
1426 /* For insns with operands there are more diddles to do to the opcode. */
1427 if (i.operands)
1429 if (!process_operands ())
1430 return;
1432 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
1434 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
1435 as_warn (_("translating to `%sp'"), i.tm.name);
1438 /* Handle conversion of 'int $3' --> special int3 insn. */
1439 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
1441 i.tm.base_opcode = INT3_OPCODE;
1442 i.imm_operands = 0;
1445 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
1446 && i.op[0].disps->X_op == O_constant)
1448 /* Convert "jmp constant" (and "call constant") to a jump (call) to
1449 the absolute address given by the constant. Since ix86 jumps and
1450 calls are pc relative, we need to generate a reloc. */
1451 i.op[0].disps->X_add_symbol = &abs_symbol;
1452 i.op[0].disps->X_op = O_symbol;
1455 if ((i.tm.opcode_modifier & Rex64) != 0)
1456 i.rex |= REX_MODE64;
1458 /* For 8 bit registers we need an empty rex prefix. Also if the
1459 instruction already has a prefix, we need to convert old
1460 registers to new ones. */
1462 if (((i.types[0] & Reg8) != 0
1463 && (i.op[0].regs->reg_flags & RegRex64) != 0)
1464 || ((i.types[1] & Reg8) != 0
1465 && (i.op[1].regs->reg_flags & RegRex64) != 0)
1466 || (((i.types[0] & Reg8) != 0 || (i.types[1] & Reg8) != 0)
1467 && i.rex != 0))
1469 int x;
1471 i.rex |= REX_OPCODE;
1472 for (x = 0; x < 2; x++)
1474 /* Look for 8 bit operand that uses old registers. */
1475 if ((i.types[x] & Reg8) != 0
1476 && (i.op[x].regs->reg_flags & RegRex64) == 0)
1478 /* In case it is "hi" register, give up. */
1479 if (i.op[x].regs->reg_num > 3)
1480 as_bad (_("can't encode register '%%%s' in an instruction requiring REX prefix."),
1481 i.op[x].regs->reg_name);
1483 /* Otherwise it is equivalent to the extended register.
1484 Since the encoding doesn't change this is merely
1485 cosmetic cleanup for debug output. */
1487 i.op[x].regs = i.op[x].regs + 8;
1492 if (i.rex != 0)
1493 add_prefix (REX_OPCODE | i.rex);
1495 /* We are ready to output the insn. */
1496 output_insn ();
1499 static char *
1500 parse_insn (line, mnemonic)
1501 char *line;
1502 char *mnemonic;
1504 char *l = line;
1505 char *token_start = l;
1506 char *mnem_p;
1508 /* Non-zero if we found a prefix only acceptable with string insns. */
1509 const char *expecting_string_instruction = NULL;
1511 while (1)
1513 mnem_p = mnemonic;
1514 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1516 mnem_p++;
1517 if (mnem_p >= mnemonic + MAX_MNEM_SIZE)
1519 as_bad (_("no such instruction: `%s'"), token_start);
1520 return NULL;
1522 l++;
1524 if (!is_space_char (*l)
1525 && *l != END_OF_INSN
1526 && *l != PREFIX_SEPARATOR
1527 && *l != ',')
1529 as_bad (_("invalid character %s in mnemonic"),
1530 output_invalid (*l));
1531 return NULL;
1533 if (token_start == l)
1535 if (*l == PREFIX_SEPARATOR)
1536 as_bad (_("expecting prefix; got nothing"));
1537 else
1538 as_bad (_("expecting mnemonic; got nothing"));
1539 return NULL;
1542 /* Look up instruction (or prefix) via hash table. */
1543 current_templates = hash_find (op_hash, mnemonic);
1545 if (*l != END_OF_INSN
1546 && (!is_space_char (*l) || l[1] != END_OF_INSN)
1547 && current_templates
1548 && (current_templates->start->opcode_modifier & IsPrefix))
1550 /* If we are in 16-bit mode, do not allow addr16 or data16.
1551 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1552 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1553 && flag_code != CODE_64BIT
1554 && (((current_templates->start->opcode_modifier & Size32) != 0)
1555 ^ (flag_code == CODE_16BIT)))
1557 as_bad (_("redundant %s prefix"),
1558 current_templates->start->name);
1559 return NULL;
1561 /* Add prefix, checking for repeated prefixes. */
1562 switch (add_prefix (current_templates->start->base_opcode))
1564 case 0:
1565 return NULL;
1566 case 2:
1567 expecting_string_instruction = current_templates->start->name;
1568 break;
1570 /* Skip past PREFIX_SEPARATOR and reset token_start. */
1571 token_start = ++l;
1573 else
1574 break;
1577 if (!current_templates)
1579 /* See if we can get a match by trimming off a suffix. */
1580 switch (mnem_p[-1])
1582 case WORD_MNEM_SUFFIX:
1583 case BYTE_MNEM_SUFFIX:
1584 case QWORD_MNEM_SUFFIX:
1585 i.suffix = mnem_p[-1];
1586 mnem_p[-1] = '\0';
1587 current_templates = hash_find (op_hash, mnemonic);
1588 break;
1589 case SHORT_MNEM_SUFFIX:
1590 case LONG_MNEM_SUFFIX:
1591 if (!intel_syntax)
1593 i.suffix = mnem_p[-1];
1594 mnem_p[-1] = '\0';
1595 current_templates = hash_find (op_hash, mnemonic);
1597 break;
1599 /* Intel Syntax. */
1600 case 'd':
1601 if (intel_syntax)
1603 if (intel_float_operand (mnemonic))
1604 i.suffix = SHORT_MNEM_SUFFIX;
1605 else
1606 i.suffix = LONG_MNEM_SUFFIX;
1607 mnem_p[-1] = '\0';
1608 current_templates = hash_find (op_hash, mnemonic);
1610 break;
1612 if (!current_templates)
1614 as_bad (_("no such instruction: `%s'"), token_start);
1615 return NULL;
1619 if (current_templates->start->opcode_modifier & (Jump | JumpByte))
1621 /* Check for a branch hint. We allow ",pt" and ",pn" for
1622 predict taken and predict not taken respectively.
1623 I'm not sure that branch hints actually do anything on loop
1624 and jcxz insns (JumpByte) for current Pentium4 chips. They
1625 may work in the future and it doesn't hurt to accept them
1626 now. */
1627 if (l[0] == ',' && l[1] == 'p')
1629 if (l[2] == 't')
1631 if (!add_prefix (DS_PREFIX_OPCODE))
1632 return NULL;
1633 l += 3;
1635 else if (l[2] == 'n')
1637 if (!add_prefix (CS_PREFIX_OPCODE))
1638 return NULL;
1639 l += 3;
1643 /* Any other comma loses. */
1644 if (*l == ',')
1646 as_bad (_("invalid character %s in mnemonic"),
1647 output_invalid (*l));
1648 return NULL;
1651 /* Check if instruction is supported on specified architecture. */
1652 if ((current_templates->start->cpu_flags & ~(Cpu64 | CpuNo64))
1653 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64)))
1655 as_warn (_("`%s' is not supported on `%s'"),
1656 current_templates->start->name, cpu_arch_name);
1658 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
1660 as_warn (_("use .code16 to ensure correct addressing mode"));
1663 /* Check for rep/repne without a string instruction. */
1664 if (expecting_string_instruction
1665 && !(current_templates->start->opcode_modifier & IsString))
1667 as_bad (_("expecting string instruction after `%s'"),
1668 expecting_string_instruction);
1669 return NULL;
1672 return l;
1675 static char *
1676 parse_operands (l, mnemonic)
1677 char *l;
1678 const char *mnemonic;
1680 char *token_start;
1682 /* 1 if operand is pending after ','. */
1683 unsigned int expecting_operand = 0;
1685 /* Non-zero if operand parens not balanced. */
1686 unsigned int paren_not_balanced;
1688 while (*l != END_OF_INSN)
1690 /* Skip optional white space before operand. */
1691 if (is_space_char (*l))
1692 ++l;
1693 if (!is_operand_char (*l) && *l != END_OF_INSN)
1695 as_bad (_("invalid character %s before operand %d"),
1696 output_invalid (*l),
1697 i.operands + 1);
1698 return NULL;
1700 token_start = l; /* after white space */
1701 paren_not_balanced = 0;
1702 while (paren_not_balanced || *l != ',')
1704 if (*l == END_OF_INSN)
1706 if (paren_not_balanced)
1708 if (!intel_syntax)
1709 as_bad (_("unbalanced parenthesis in operand %d."),
1710 i.operands + 1);
1711 else
1712 as_bad (_("unbalanced brackets in operand %d."),
1713 i.operands + 1);
1714 return NULL;
1716 else
1717 break; /* we are done */
1719 else if (!is_operand_char (*l) && !is_space_char (*l))
1721 as_bad (_("invalid character %s in operand %d"),
1722 output_invalid (*l),
1723 i.operands + 1);
1724 return NULL;
1726 if (!intel_syntax)
1728 if (*l == '(')
1729 ++paren_not_balanced;
1730 if (*l == ')')
1731 --paren_not_balanced;
1733 else
1735 if (*l == '[')
1736 ++paren_not_balanced;
1737 if (*l == ']')
1738 --paren_not_balanced;
1740 l++;
1742 if (l != token_start)
1743 { /* Yes, we've read in another operand. */
1744 unsigned int operand_ok;
1745 this_operand = i.operands++;
1746 if (i.operands > MAX_OPERANDS)
1748 as_bad (_("spurious operands; (%d operands/instruction max)"),
1749 MAX_OPERANDS);
1750 return NULL;
1752 /* Now parse operand adding info to 'i' as we go along. */
1753 END_STRING_AND_SAVE (l);
1755 if (intel_syntax)
1756 operand_ok =
1757 i386_intel_operand (token_start,
1758 intel_float_operand (mnemonic));
1759 else
1760 operand_ok = i386_operand (token_start);
1762 RESTORE_END_STRING (l);
1763 if (!operand_ok)
1764 return NULL;
1766 else
1768 if (expecting_operand)
1770 expecting_operand_after_comma:
1771 as_bad (_("expecting operand after ','; got nothing"));
1772 return NULL;
1774 if (*l == ',')
1776 as_bad (_("expecting operand before ','; got nothing"));
1777 return NULL;
1781 /* Now *l must be either ',' or END_OF_INSN. */
1782 if (*l == ',')
1784 if (*++l == END_OF_INSN)
1786 /* Just skip it, if it's \n complain. */
1787 goto expecting_operand_after_comma;
1789 expecting_operand = 1;
1792 return l;
1795 static void
1796 swap_operands ()
1798 union i386_op temp_op;
1799 unsigned int temp_type;
1800 enum bfd_reloc_code_real temp_reloc;
1801 int xchg1 = 0;
1802 int xchg2 = 0;
1804 if (i.operands == 2)
1806 xchg1 = 0;
1807 xchg2 = 1;
1809 else if (i.operands == 3)
1811 xchg1 = 0;
1812 xchg2 = 2;
1814 temp_type = i.types[xchg2];
1815 i.types[xchg2] = i.types[xchg1];
1816 i.types[xchg1] = temp_type;
1817 temp_op = i.op[xchg2];
1818 i.op[xchg2] = i.op[xchg1];
1819 i.op[xchg1] = temp_op;
1820 temp_reloc = i.reloc[xchg2];
1821 i.reloc[xchg2] = i.reloc[xchg1];
1822 i.reloc[xchg1] = temp_reloc;
1824 if (i.mem_operands == 2)
1826 const seg_entry *temp_seg;
1827 temp_seg = i.seg[0];
1828 i.seg[0] = i.seg[1];
1829 i.seg[1] = temp_seg;
1833 /* Try to ensure constant immediates are represented in the smallest
1834 opcode possible. */
1835 static void
1836 optimize_imm ()
1838 char guess_suffix = 0;
1839 int op;
1841 if (i.suffix)
1842 guess_suffix = i.suffix;
1843 else if (i.reg_operands)
1845 /* Figure out a suffix from the last register operand specified.
1846 We can't do this properly yet, ie. excluding InOutPortReg,
1847 but the following works for instructions with immediates.
1848 In any case, we can't set i.suffix yet. */
1849 for (op = i.operands; --op >= 0;)
1850 if (i.types[op] & Reg)
1852 if (i.types[op] & Reg8)
1853 guess_suffix = BYTE_MNEM_SUFFIX;
1854 else if (i.types[op] & Reg16)
1855 guess_suffix = WORD_MNEM_SUFFIX;
1856 else if (i.types[op] & Reg32)
1857 guess_suffix = LONG_MNEM_SUFFIX;
1858 else if (i.types[op] & Reg64)
1859 guess_suffix = QWORD_MNEM_SUFFIX;
1860 break;
1863 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
1864 guess_suffix = WORD_MNEM_SUFFIX;
1866 for (op = i.operands; --op >= 0;)
1867 if (i.types[op] & Imm)
1869 switch (i.op[op].imms->X_op)
1871 case O_constant:
1872 /* If a suffix is given, this operand may be shortened. */
1873 switch (guess_suffix)
1875 case LONG_MNEM_SUFFIX:
1876 i.types[op] |= Imm32 | Imm64;
1877 break;
1878 case WORD_MNEM_SUFFIX:
1879 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
1880 break;
1881 case BYTE_MNEM_SUFFIX:
1882 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
1883 break;
1886 /* If this operand is at most 16 bits, convert it
1887 to a signed 16 bit number before trying to see
1888 whether it will fit in an even smaller size.
1889 This allows a 16-bit operand such as $0xffe0 to
1890 be recognised as within Imm8S range. */
1891 if ((i.types[op] & Imm16)
1892 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
1894 i.op[op].imms->X_add_number =
1895 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
1897 if ((i.types[op] & Imm32)
1898 && ((i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1))
1899 == 0))
1901 i.op[op].imms->X_add_number = ((i.op[op].imms->X_add_number
1902 ^ ((offsetT) 1 << 31))
1903 - ((offsetT) 1 << 31));
1905 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
1907 /* We must avoid matching of Imm32 templates when 64bit
1908 only immediate is available. */
1909 if (guess_suffix == QWORD_MNEM_SUFFIX)
1910 i.types[op] &= ~Imm32;
1911 break;
1913 case O_absent:
1914 case O_register:
1915 abort ();
1917 /* Symbols and expressions. */
1918 default:
1919 /* Convert symbolic operand to proper sizes for matching. */
1920 switch (guess_suffix)
1922 case QWORD_MNEM_SUFFIX:
1923 i.types[op] = Imm64 | Imm32S;
1924 break;
1925 case LONG_MNEM_SUFFIX:
1926 i.types[op] = Imm32;
1927 break;
1928 case WORD_MNEM_SUFFIX:
1929 i.types[op] = Imm16;
1930 break;
1931 case BYTE_MNEM_SUFFIX:
1932 i.types[op] = Imm8 | Imm8S;
1933 break;
1935 break;
1940 /* Try to use the smallest displacement type too. */
1941 static void
1942 optimize_disp ()
1944 int op;
1946 for (op = i.operands; --op >= 0;)
1947 if ((i.types[op] & Disp) && i.op[op].disps->X_op == O_constant)
1949 offsetT disp = i.op[op].disps->X_add_number;
1951 if (i.types[op] & Disp16)
1953 /* We know this operand is at most 16 bits, so
1954 convert to a signed 16 bit number before trying
1955 to see whether it will fit in an even smaller
1956 size. */
1958 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
1960 else if (i.types[op] & Disp32)
1962 /* We know this operand is at most 32 bits, so convert to a
1963 signed 32 bit number before trying to see whether it will
1964 fit in an even smaller size. */
1965 disp &= (((offsetT) 2 << 31) - 1);
1966 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1968 if (flag_code == CODE_64BIT)
1970 if (fits_in_signed_long (disp))
1971 i.types[op] |= Disp32S;
1972 if (fits_in_unsigned_long (disp))
1973 i.types[op] |= Disp32;
1975 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
1976 && fits_in_signed_byte (disp))
1977 i.types[op] |= Disp8;
1981 static int
1982 match_template ()
1984 /* Points to template once we've found it. */
1985 const template *t;
1986 unsigned int overlap0, overlap1, overlap2;
1987 unsigned int found_reverse_match;
1988 int suffix_check;
1990 #define MATCH(overlap, given, template) \
1991 ((overlap & ~JumpAbsolute) \
1992 && (((given) & (BaseIndex | JumpAbsolute)) \
1993 == ((overlap) & (BaseIndex | JumpAbsolute))))
1995 /* If given types r0 and r1 are registers they must be of the same type
1996 unless the expected operand type register overlap is null.
1997 Note that Acc in a template matches every size of reg. */
1998 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
1999 (((g0) & Reg) == 0 || ((g1) & Reg) == 0 \
2000 || ((g0) & Reg) == ((g1) & Reg) \
2001 || ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
2003 overlap0 = 0;
2004 overlap1 = 0;
2005 overlap2 = 0;
2006 found_reverse_match = 0;
2007 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
2008 ? No_bSuf
2009 : (i.suffix == WORD_MNEM_SUFFIX
2010 ? No_wSuf
2011 : (i.suffix == SHORT_MNEM_SUFFIX
2012 ? No_sSuf
2013 : (i.suffix == LONG_MNEM_SUFFIX
2014 ? No_lSuf
2015 : (i.suffix == QWORD_MNEM_SUFFIX
2016 ? No_qSuf
2017 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX
2018 ? No_xSuf : 0))))));
2020 t = current_templates->start;
2021 if (i.suffix == QWORD_MNEM_SUFFIX
2022 && flag_code != CODE_64BIT
2023 && (!intel_syntax
2024 || (!(t->opcode_modifier & IgnoreSize)
2025 && ! intel_float_operand (t->name)))
2026 && (!(t->operand_types[0] & (RegMMX | RegXMM))
2027 || !(t->operand_types[t->operands > 1] & (RegMMX | RegXMM)))
2028 && (t->base_opcode != 0x0fc7
2029 || t->extension_opcode != 1 /* cmpxchg8b */))
2030 t = current_templates->end;
2031 for (; t < current_templates->end; t++)
2033 /* Must have right number of operands. */
2034 if (i.operands != t->operands)
2035 continue;
2037 /* Check the suffix, except for some instructions in intel mode. */
2038 if ((t->opcode_modifier & suffix_check)
2039 && !(intel_syntax
2040 && (t->opcode_modifier & IgnoreSize))
2041 && !(intel_syntax
2042 && t->base_opcode == 0xd9
2043 && (t->extension_opcode == 5 /* 0xd9,5 "fldcw" */
2044 || t->extension_opcode == 7))) /* 0xd9,7 "f{n}stcw" */
2045 continue;
2047 /* Do not verify operands when there are none. */
2048 else if (!t->operands)
2050 if (t->cpu_flags & ~cpu_arch_flags)
2051 continue;
2052 /* We've found a match; break out of loop. */
2053 break;
2056 overlap0 = i.types[0] & t->operand_types[0];
2057 switch (t->operands)
2059 case 1:
2060 if (!MATCH (overlap0, i.types[0], t->operand_types[0]))
2061 continue;
2062 break;
2063 case 2:
2064 case 3:
2065 overlap1 = i.types[1] & t->operand_types[1];
2066 if (!MATCH (overlap0, i.types[0], t->operand_types[0])
2067 || !MATCH (overlap1, i.types[1], t->operand_types[1])
2068 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2069 t->operand_types[0],
2070 overlap1, i.types[1],
2071 t->operand_types[1]))
2073 /* Check if other direction is valid ... */
2074 if ((t->opcode_modifier & (D | FloatD)) == 0)
2075 continue;
2077 /* Try reversing direction of operands. */
2078 overlap0 = i.types[0] & t->operand_types[1];
2079 overlap1 = i.types[1] & t->operand_types[0];
2080 if (!MATCH (overlap0, i.types[0], t->operand_types[1])
2081 || !MATCH (overlap1, i.types[1], t->operand_types[0])
2082 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2083 t->operand_types[1],
2084 overlap1, i.types[1],
2085 t->operand_types[0]))
2087 /* Does not match either direction. */
2088 continue;
2090 /* found_reverse_match holds which of D or FloatDR
2091 we've found. */
2092 found_reverse_match = t->opcode_modifier & (D | FloatDR);
2094 /* Found a forward 2 operand match here. */
2095 else if (t->operands == 3)
2097 /* Here we make use of the fact that there are no
2098 reverse match 3 operand instructions, and all 3
2099 operand instructions only need to be checked for
2100 register consistency between operands 2 and 3. */
2101 overlap2 = i.types[2] & t->operand_types[2];
2102 if (!MATCH (overlap2, i.types[2], t->operand_types[2])
2103 || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1],
2104 t->operand_types[1],
2105 overlap2, i.types[2],
2106 t->operand_types[2]))
2108 continue;
2110 /* Found either forward/reverse 2 or 3 operand match here:
2111 slip through to break. */
2113 if (t->cpu_flags & ~cpu_arch_flags)
2115 found_reverse_match = 0;
2116 continue;
2118 /* We've found a match; break out of loop. */
2119 break;
2122 if (t == current_templates->end)
2124 /* We found no match. */
2125 as_bad (_("suffix or operands invalid for `%s'"),
2126 current_templates->start->name);
2127 return 0;
2130 if (!quiet_warnings)
2132 if (!intel_syntax
2133 && ((i.types[0] & JumpAbsolute)
2134 != (t->operand_types[0] & JumpAbsolute)))
2136 as_warn (_("indirect %s without `*'"), t->name);
2139 if ((t->opcode_modifier & (IsPrefix | IgnoreSize))
2140 == (IsPrefix | IgnoreSize))
2142 /* Warn them that a data or address size prefix doesn't
2143 affect assembly of the next line of code. */
2144 as_warn (_("stand-alone `%s' prefix"), t->name);
2148 /* Copy the template we found. */
2149 i.tm = *t;
2150 if (found_reverse_match)
2152 /* If we found a reverse match we must alter the opcode
2153 direction bit. found_reverse_match holds bits to change
2154 (different for int & float insns). */
2156 i.tm.base_opcode ^= found_reverse_match;
2158 i.tm.operand_types[0] = t->operand_types[1];
2159 i.tm.operand_types[1] = t->operand_types[0];
2162 return 1;
2165 static int
2166 check_string ()
2168 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
2169 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
2171 if (i.seg[0] != NULL && i.seg[0] != &es)
2173 as_bad (_("`%s' operand %d must use `%%es' segment"),
2174 i.tm.name,
2175 mem_op + 1);
2176 return 0;
2178 /* There's only ever one segment override allowed per instruction.
2179 This instruction possibly has a legal segment override on the
2180 second operand, so copy the segment to where non-string
2181 instructions store it, allowing common code. */
2182 i.seg[0] = i.seg[1];
2184 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
2186 if (i.seg[1] != NULL && i.seg[1] != &es)
2188 as_bad (_("`%s' operand %d must use `%%es' segment"),
2189 i.tm.name,
2190 mem_op + 2);
2191 return 0;
2194 return 1;
2197 static int
2198 process_suffix (void)
2200 /* If matched instruction specifies an explicit instruction mnemonic
2201 suffix, use it. */
2202 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
2204 if (i.tm.opcode_modifier & Size16)
2205 i.suffix = WORD_MNEM_SUFFIX;
2206 else if (i.tm.opcode_modifier & Size64)
2207 i.suffix = QWORD_MNEM_SUFFIX;
2208 else
2209 i.suffix = LONG_MNEM_SUFFIX;
2211 else if (i.reg_operands)
2213 /* If there's no instruction mnemonic suffix we try to invent one
2214 based on register operands. */
2215 if (!i.suffix)
2217 /* We take i.suffix from the last register operand specified,
2218 Destination register type is more significant than source
2219 register type. */
2220 int op;
2222 for (op = i.operands; --op >= 0;)
2223 if ((i.types[op] & Reg)
2224 && !(i.tm.operand_types[op] & InOutPortReg))
2226 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
2227 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
2228 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
2229 LONG_MNEM_SUFFIX);
2230 break;
2233 else if (i.suffix == BYTE_MNEM_SUFFIX)
2235 if (!check_byte_reg ())
2236 return 0;
2238 else if (i.suffix == LONG_MNEM_SUFFIX)
2240 if (!check_long_reg ())
2241 return 0;
2243 else if (i.suffix == QWORD_MNEM_SUFFIX)
2245 if (!check_qword_reg ())
2246 return 0;
2248 else if (i.suffix == WORD_MNEM_SUFFIX)
2250 if (!check_word_reg ())
2251 return 0;
2253 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2254 /* Do nothing if the instruction is going to ignore the prefix. */
2256 else
2257 abort ();
2259 else if ((i.tm.opcode_modifier & DefaultSize) && !i.suffix)
2261 i.suffix = stackop_size;
2262 if (i.suffix == QWORD_MNEM_SUFFIX
2263 && (i.tm.opcode_modifier & No_qSuf))
2264 i.suffix = LONG_MNEM_SUFFIX;
2267 /* Change the opcode based on the operand size given by i.suffix;
2268 We need not change things for byte insns. */
2269 if (!i.suffix && (i.tm.opcode_modifier & W))
2271 as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
2272 return 0;
2275 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2277 /* It's not a byte, select word/dword operation. */
2278 if (i.tm.opcode_modifier & W)
2280 if (i.tm.opcode_modifier & ShortForm)
2281 i.tm.base_opcode |= 8;
2282 else
2283 i.tm.base_opcode |= 1;
2286 /* Now select between word & dword operations via the operand
2287 size prefix, except for instructions that will ignore this
2288 prefix anyway. */
2289 if (i.suffix != QWORD_MNEM_SUFFIX
2290 && !(i.tm.opcode_modifier & IgnoreSize)
2291 && ((i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2292 || (flag_code == CODE_64BIT
2293 && (i.tm.opcode_modifier & JumpByte))))
2295 unsigned int prefix = DATA_PREFIX_OPCODE;
2297 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2298 prefix = ADDR_PREFIX_OPCODE;
2300 if (!add_prefix (prefix))
2301 return 0;
2304 /* Set mode64 for an operand. */
2305 if (i.suffix == QWORD_MNEM_SUFFIX
2306 && flag_code == CODE_64BIT
2307 && (i.tm.opcode_modifier & NoRex64) == 0)
2308 i.rex |= REX_MODE64;
2310 /* Size floating point instruction. */
2311 if (i.suffix == LONG_MNEM_SUFFIX)
2312 if (i.tm.opcode_modifier & FloatMF)
2313 i.tm.base_opcode ^= 4;
2316 return 1;
2319 static int
2320 check_byte_reg (void)
2322 int op;
2324 for (op = i.operands; --op >= 0;)
2326 /* If this is an eight bit register, it's OK. If it's the 16 or
2327 32 bit version of an eight bit register, we will just use the
2328 low portion, and that's OK too. */
2329 if (i.types[op] & Reg8)
2330 continue;
2332 /* movzx and movsx should not generate this warning. */
2333 if (intel_syntax
2334 && (i.tm.base_opcode == 0xfb7
2335 || i.tm.base_opcode == 0xfb6
2336 || i.tm.base_opcode == 0x63
2337 || i.tm.base_opcode == 0xfbe
2338 || i.tm.base_opcode == 0xfbf))
2339 continue;
2341 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4
2342 #if 0
2343 /* Check that the template allows eight bit regs. This
2344 kills insns such as `orb $1,%edx', which maybe should be
2345 allowed. */
2346 && (i.tm.operand_types[op] & (Reg8 | InOutPortReg))
2347 #endif
2350 /* Prohibit these changes in the 64bit mode, since the
2351 lowering is more complicated. */
2352 if (flag_code == CODE_64BIT
2353 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2355 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2356 i.op[op].regs->reg_name,
2357 i.suffix);
2358 return 0;
2360 #if REGISTER_WARNINGS
2361 if (!quiet_warnings
2362 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2363 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2364 (i.op[op].regs + (i.types[op] & Reg16
2365 ? REGNAM_AL - REGNAM_AX
2366 : REGNAM_AL - REGNAM_EAX))->reg_name,
2367 i.op[op].regs->reg_name,
2368 i.suffix);
2369 #endif
2370 continue;
2372 /* Any other register is bad. */
2373 if (i.types[op] & (Reg | RegMMX | RegXMM
2374 | SReg2 | SReg3
2375 | Control | Debug | Test
2376 | FloatReg | FloatAcc))
2378 as_bad (_("`%%%s' not allowed with `%s%c'"),
2379 i.op[op].regs->reg_name,
2380 i.tm.name,
2381 i.suffix);
2382 return 0;
2385 return 1;
2388 static int
2389 check_long_reg ()
2391 int op;
2393 for (op = i.operands; --op >= 0;)
2394 /* Reject eight bit registers, except where the template requires
2395 them. (eg. movzb) */
2396 if ((i.types[op] & Reg8) != 0
2397 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2399 as_bad (_("`%%%s' not allowed with `%s%c'"),
2400 i.op[op].regs->reg_name,
2401 i.tm.name,
2402 i.suffix);
2403 return 0;
2405 /* Warn if the e prefix on a general reg is missing. */
2406 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2407 && (i.types[op] & Reg16) != 0
2408 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2410 /* Prohibit these changes in the 64bit mode, since the
2411 lowering is more complicated. */
2412 if (flag_code == CODE_64BIT)
2414 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2415 i.op[op].regs->reg_name,
2416 i.suffix);
2417 return 0;
2419 #if REGISTER_WARNINGS
2420 else
2421 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2422 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
2423 i.op[op].regs->reg_name,
2424 i.suffix);
2425 #endif
2427 /* Warn if the r prefix on a general reg is missing. */
2428 else if ((i.types[op] & Reg64) != 0
2429 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2431 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2432 i.op[op].regs->reg_name,
2433 i.suffix);
2434 return 0;
2436 return 1;
2439 static int
2440 check_qword_reg ()
2442 int op;
2444 for (op = i.operands; --op >= 0; )
2445 /* Reject eight bit registers, except where the template requires
2446 them. (eg. movzb) */
2447 if ((i.types[op] & Reg8) != 0
2448 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2450 as_bad (_("`%%%s' not allowed with `%s%c'"),
2451 i.op[op].regs->reg_name,
2452 i.tm.name,
2453 i.suffix);
2454 return 0;
2456 /* Warn if the e prefix on a general reg is missing. */
2457 else if (((i.types[op] & Reg16) != 0
2458 || (i.types[op] & Reg32) != 0)
2459 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2461 /* Prohibit these changes in the 64bit mode, since the
2462 lowering is more complicated. */
2463 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2464 i.op[op].regs->reg_name,
2465 i.suffix);
2466 return 0;
2468 return 1;
2471 static int
2472 check_word_reg ()
2474 int op;
2475 for (op = i.operands; --op >= 0;)
2476 /* Reject eight bit registers, except where the template requires
2477 them. (eg. movzb) */
2478 if ((i.types[op] & Reg8) != 0
2479 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2481 as_bad (_("`%%%s' not allowed with `%s%c'"),
2482 i.op[op].regs->reg_name,
2483 i.tm.name,
2484 i.suffix);
2485 return 0;
2487 /* Warn if the e prefix on a general reg is present. */
2488 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2489 && (i.types[op] & Reg32) != 0
2490 && (i.tm.operand_types[op] & (Reg16 | Acc)) != 0)
2492 /* Prohibit these changes in the 64bit mode, since the
2493 lowering is more complicated. */
2494 if (flag_code == CODE_64BIT)
2496 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2497 i.op[op].regs->reg_name,
2498 i.suffix);
2499 return 0;
2501 else
2502 #if REGISTER_WARNINGS
2503 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2504 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
2505 i.op[op].regs->reg_name,
2506 i.suffix);
2507 #endif
2509 return 1;
2512 static int
2513 finalize_imm ()
2515 unsigned int overlap0, overlap1, overlap2;
2517 overlap0 = i.types[0] & i.tm.operand_types[0];
2518 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64))
2519 && overlap0 != Imm8 && overlap0 != Imm8S
2520 && overlap0 != Imm16 && overlap0 != Imm32S
2521 && overlap0 != Imm32 && overlap0 != Imm64)
2523 if (i.suffix)
2525 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX
2526 ? Imm8 | Imm8S
2527 : (i.suffix == WORD_MNEM_SUFFIX
2528 ? Imm16
2529 : (i.suffix == QWORD_MNEM_SUFFIX
2530 ? Imm64 | Imm32S
2531 : Imm32)));
2533 else if (overlap0 == (Imm16 | Imm32S | Imm32)
2534 || overlap0 == (Imm16 | Imm32)
2535 || overlap0 == (Imm16 | Imm32S))
2537 overlap0 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
2538 ? Imm16 : Imm32S);
2540 if (overlap0 != Imm8 && overlap0 != Imm8S
2541 && overlap0 != Imm16 && overlap0 != Imm32S
2542 && overlap0 != Imm32 && overlap0 != Imm64)
2544 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
2545 return 0;
2548 i.types[0] = overlap0;
2550 overlap1 = i.types[1] & i.tm.operand_types[1];
2551 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32))
2552 && overlap1 != Imm8 && overlap1 != Imm8S
2553 && overlap1 != Imm16 && overlap1 != Imm32S
2554 && overlap1 != Imm32 && overlap1 != Imm64)
2556 if (i.suffix)
2558 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX
2559 ? Imm8 | Imm8S
2560 : (i.suffix == WORD_MNEM_SUFFIX
2561 ? Imm16
2562 : (i.suffix == QWORD_MNEM_SUFFIX
2563 ? Imm64 | Imm32S
2564 : Imm32)));
2566 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
2567 || overlap1 == (Imm16 | Imm32)
2568 || overlap1 == (Imm16 | Imm32S))
2570 overlap1 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
2571 ? Imm16 : Imm32S);
2573 if (overlap1 != Imm8 && overlap1 != Imm8S
2574 && overlap1 != Imm16 && overlap1 != Imm32S
2575 && overlap1 != Imm32 && overlap1 != Imm64)
2577 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
2578 return 0;
2581 i.types[1] = overlap1;
2583 overlap2 = i.types[2] & i.tm.operand_types[2];
2584 assert ((overlap2 & Imm) == 0);
2585 i.types[2] = overlap2;
2587 return 1;
2590 static int
2591 process_operands ()
2593 /* Default segment register this instruction will use for memory
2594 accesses. 0 means unknown. This is only for optimizing out
2595 unnecessary segment overrides. */
2596 const seg_entry *default_seg = 0;
2598 /* The imul $imm, %reg instruction is converted into
2599 imul $imm, %reg, %reg, and the clr %reg instruction
2600 is converted into xor %reg, %reg. */
2601 if (i.tm.opcode_modifier & regKludge)
2603 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
2604 /* Pretend we saw the extra register operand. */
2605 assert (i.op[first_reg_op + 1].regs == 0);
2606 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
2607 i.types[first_reg_op + 1] = i.types[first_reg_op];
2608 i.reg_operands = 2;
2611 if (i.tm.opcode_modifier & ShortForm)
2613 /* The register or float register operand is in operand 0 or 1. */
2614 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
2615 /* Register goes in low 3 bits of opcode. */
2616 i.tm.base_opcode |= i.op[op].regs->reg_num;
2617 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2618 i.rex |= REX_EXTZ;
2619 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2621 /* Warn about some common errors, but press on regardless.
2622 The first case can be generated by gcc (<= 2.8.1). */
2623 if (i.operands == 2)
2625 /* Reversed arguments on faddp, fsubp, etc. */
2626 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
2627 i.op[1].regs->reg_name,
2628 i.op[0].regs->reg_name);
2630 else
2632 /* Extraneous `l' suffix on fp insn. */
2633 as_warn (_("translating to `%s %%%s'"), i.tm.name,
2634 i.op[0].regs->reg_name);
2638 else if (i.tm.opcode_modifier & Modrm)
2640 /* The opcode is completed (modulo i.tm.extension_opcode which
2641 must be put into the modrm byte). Now, we make the modrm and
2642 index base bytes based on all the info we've collected. */
2644 default_seg = build_modrm_byte ();
2646 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
2648 if (i.tm.base_opcode == POP_SEG_SHORT
2649 && i.op[0].regs->reg_num == 1)
2651 as_bad (_("you can't `pop %%cs'"));
2652 return 0;
2654 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
2655 if ((i.op[0].regs->reg_flags & RegRex) != 0)
2656 i.rex |= REX_EXTZ;
2658 else if ((i.tm.base_opcode & ~(D | W)) == MOV_AX_DISP32)
2660 default_seg = &ds;
2662 else if ((i.tm.opcode_modifier & IsString) != 0)
2664 /* For the string instructions that allow a segment override
2665 on one of their operands, the default segment is ds. */
2666 default_seg = &ds;
2669 if (i.tm.base_opcode == 0x8d /* lea */ && i.seg[0] && !quiet_warnings)
2670 as_warn (_("segment override on `lea' is ineffectual"));
2672 /* If a segment was explicitly specified, and the specified segment
2673 is not the default, use an opcode prefix to select it. If we
2674 never figured out what the default segment is, then default_seg
2675 will be zero at this point, and the specified segment prefix will
2676 always be used. */
2677 if ((i.seg[0]) && (i.seg[0] != default_seg))
2679 if (!add_prefix (i.seg[0]->seg_prefix))
2680 return 0;
2682 return 1;
2685 static const seg_entry *
2686 build_modrm_byte ()
2688 const seg_entry *default_seg = 0;
2690 /* i.reg_operands MUST be the number of real register operands;
2691 implicit registers do not count. */
2692 if (i.reg_operands == 2)
2694 unsigned int source, dest;
2695 source = ((i.types[0]
2696 & (Reg | RegMMX | RegXMM
2697 | SReg2 | SReg3
2698 | Control | Debug | Test))
2699 ? 0 : 1);
2700 dest = source + 1;
2702 i.rm.mode = 3;
2703 /* One of the register operands will be encoded in the i.tm.reg
2704 field, the other in the combined i.tm.mode and i.tm.regmem
2705 fields. If no form of this instruction supports a memory
2706 destination operand, then we assume the source operand may
2707 sometimes be a memory operand and so we need to store the
2708 destination in the i.rm.reg field. */
2709 if ((i.tm.operand_types[dest] & AnyMem) == 0)
2711 i.rm.reg = i.op[dest].regs->reg_num;
2712 i.rm.regmem = i.op[source].regs->reg_num;
2713 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
2714 i.rex |= REX_EXTX;
2715 if ((i.op[source].regs->reg_flags & RegRex) != 0)
2716 i.rex |= REX_EXTZ;
2718 else
2720 i.rm.reg = i.op[source].regs->reg_num;
2721 i.rm.regmem = i.op[dest].regs->reg_num;
2722 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
2723 i.rex |= REX_EXTZ;
2724 if ((i.op[source].regs->reg_flags & RegRex) != 0)
2725 i.rex |= REX_EXTX;
2728 else
2729 { /* If it's not 2 reg operands... */
2730 if (i.mem_operands)
2732 unsigned int fake_zero_displacement = 0;
2733 unsigned int op = ((i.types[0] & AnyMem)
2735 : (i.types[1] & AnyMem) ? 1 : 2);
2737 default_seg = &ds;
2739 if (i.base_reg == 0)
2741 i.rm.mode = 0;
2742 if (!i.disp_operands)
2743 fake_zero_displacement = 1;
2744 if (i.index_reg == 0)
2746 /* Operand is just <disp> */
2747 if (flag_code == CODE_64BIT)
2749 /* 64bit mode overwrites the 32bit absolute
2750 addressing by RIP relative addressing and
2751 absolute addressing is encoded by one of the
2752 redundant SIB forms. */
2753 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2754 i.sib.base = NO_BASE_REGISTER;
2755 i.sib.index = NO_INDEX_REGISTER;
2756 i.types[op] = ((i.prefix[ADDR_PREFIX] == 0) ? Disp32S : Disp32);
2758 else if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
2760 i.rm.regmem = NO_BASE_REGISTER_16;
2761 i.types[op] = Disp16;
2763 else
2765 i.rm.regmem = NO_BASE_REGISTER;
2766 i.types[op] = Disp32;
2769 else /* !i.base_reg && i.index_reg */
2771 i.sib.index = i.index_reg->reg_num;
2772 i.sib.base = NO_BASE_REGISTER;
2773 i.sib.scale = i.log2_scale_factor;
2774 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2775 i.types[op] &= ~Disp;
2776 if (flag_code != CODE_64BIT)
2777 i.types[op] |= Disp32; /* Must be 32 bit */
2778 else
2779 i.types[op] |= Disp32S;
2780 if ((i.index_reg->reg_flags & RegRex) != 0)
2781 i.rex |= REX_EXTY;
2784 /* RIP addressing for 64bit mode. */
2785 else if (i.base_reg->reg_type == BaseIndex)
2787 i.rm.regmem = NO_BASE_REGISTER;
2788 i.types[op] &= ~ Disp;
2789 i.types[op] |= Disp32S;
2790 i.flags[op] = Operand_PCrel;
2791 if (! i.disp_operands)
2792 fake_zero_displacement = 1;
2794 else if (i.base_reg->reg_type & Reg16)
2796 switch (i.base_reg->reg_num)
2798 case 3: /* (%bx) */
2799 if (i.index_reg == 0)
2800 i.rm.regmem = 7;
2801 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
2802 i.rm.regmem = i.index_reg->reg_num - 6;
2803 break;
2804 case 5: /* (%bp) */
2805 default_seg = &ss;
2806 if (i.index_reg == 0)
2808 i.rm.regmem = 6;
2809 if ((i.types[op] & Disp) == 0)
2811 /* fake (%bp) into 0(%bp) */
2812 i.types[op] |= Disp8;
2813 fake_zero_displacement = 1;
2816 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
2817 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
2818 break;
2819 default: /* (%si) -> 4 or (%di) -> 5 */
2820 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
2822 i.rm.mode = mode_from_disp_size (i.types[op]);
2824 else /* i.base_reg and 32/64 bit mode */
2826 if (flag_code == CODE_64BIT
2827 && (i.types[op] & Disp))
2828 i.types[op] = (i.types[op] & Disp8) | (i.prefix[ADDR_PREFIX] == 0 ? Disp32S : Disp32);
2830 i.rm.regmem = i.base_reg->reg_num;
2831 if ((i.base_reg->reg_flags & RegRex) != 0)
2832 i.rex |= REX_EXTZ;
2833 i.sib.base = i.base_reg->reg_num;
2834 /* x86-64 ignores REX prefix bit here to avoid decoder
2835 complications. */
2836 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
2838 default_seg = &ss;
2839 if (i.disp_operands == 0)
2841 fake_zero_displacement = 1;
2842 i.types[op] |= Disp8;
2845 else if (i.base_reg->reg_num == ESP_REG_NUM)
2847 default_seg = &ss;
2849 i.sib.scale = i.log2_scale_factor;
2850 if (i.index_reg == 0)
2852 /* <disp>(%esp) becomes two byte modrm with no index
2853 register. We've already stored the code for esp
2854 in i.rm.regmem ie. ESCAPE_TO_TWO_BYTE_ADDRESSING.
2855 Any base register besides %esp will not use the
2856 extra modrm byte. */
2857 i.sib.index = NO_INDEX_REGISTER;
2858 #if !SCALE1_WHEN_NO_INDEX
2859 /* Another case where we force the second modrm byte. */
2860 if (i.log2_scale_factor)
2861 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2862 #endif
2864 else
2866 i.sib.index = i.index_reg->reg_num;
2867 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2868 if ((i.index_reg->reg_flags & RegRex) != 0)
2869 i.rex |= REX_EXTY;
2871 i.rm.mode = mode_from_disp_size (i.types[op]);
2874 if (fake_zero_displacement)
2876 /* Fakes a zero displacement assuming that i.types[op]
2877 holds the correct displacement size. */
2878 expressionS *exp;
2880 assert (i.op[op].disps == 0);
2881 exp = &disp_expressions[i.disp_operands++];
2882 i.op[op].disps = exp;
2883 exp->X_op = O_constant;
2884 exp->X_add_number = 0;
2885 exp->X_add_symbol = (symbolS *) 0;
2886 exp->X_op_symbol = (symbolS *) 0;
2890 /* Fill in i.rm.reg or i.rm.regmem field with register operand
2891 (if any) based on i.tm.extension_opcode. Again, we must be
2892 careful to make sure that segment/control/debug/test/MMX
2893 registers are coded into the i.rm.reg field. */
2894 if (i.reg_operands)
2896 unsigned int op =
2897 ((i.types[0]
2898 & (Reg | RegMMX | RegXMM
2899 | SReg2 | SReg3
2900 | Control | Debug | Test))
2902 : ((i.types[1]
2903 & (Reg | RegMMX | RegXMM
2904 | SReg2 | SReg3
2905 | Control | Debug | Test))
2907 : 2));
2908 /* If there is an extension opcode to put here, the register
2909 number must be put into the regmem field. */
2910 if (i.tm.extension_opcode != None)
2912 i.rm.regmem = i.op[op].regs->reg_num;
2913 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2914 i.rex |= REX_EXTZ;
2916 else
2918 i.rm.reg = i.op[op].regs->reg_num;
2919 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2920 i.rex |= REX_EXTX;
2923 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2 we
2924 must set it to 3 to indicate this is a register operand
2925 in the regmem field. */
2926 if (!i.mem_operands)
2927 i.rm.mode = 3;
2930 /* Fill in i.rm.reg field with extension opcode (if any). */
2931 if (i.tm.extension_opcode != None)
2932 i.rm.reg = i.tm.extension_opcode;
2934 return default_seg;
2937 static void
2938 output_branch ()
2940 char *p;
2941 int code16;
2942 int prefix;
2943 relax_substateT subtype;
2944 symbolS *sym;
2945 offsetT off;
2947 code16 = 0;
2948 if (flag_code == CODE_16BIT)
2949 code16 = CODE16;
2951 prefix = 0;
2952 if (i.prefix[DATA_PREFIX] != 0)
2954 prefix = 1;
2955 i.prefixes -= 1;
2956 code16 ^= CODE16;
2958 /* Pentium4 branch hints. */
2959 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
2960 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
2962 prefix++;
2963 i.prefixes--;
2965 if (i.prefix[REX_PREFIX] != 0)
2967 prefix++;
2968 i.prefixes--;
2971 if (i.prefixes != 0 && !intel_syntax)
2972 as_warn (_("skipping prefixes on this instruction"));
2974 /* It's always a symbol; End frag & setup for relax.
2975 Make sure there is enough room in this frag for the largest
2976 instruction we may generate in md_convert_frag. This is 2
2977 bytes for the opcode and room for the prefix and largest
2978 displacement. */
2979 frag_grow (prefix + 2 + 4);
2980 /* Prefix and 1 opcode byte go in fr_fix. */
2981 p = frag_more (prefix + 1);
2982 if (i.prefix[DATA_PREFIX] != 0)
2983 *p++ = DATA_PREFIX_OPCODE;
2984 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
2985 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
2986 *p++ = i.prefix[SEG_PREFIX];
2987 if (i.prefix[REX_PREFIX] != 0)
2988 *p++ = i.prefix[REX_PREFIX];
2989 *p = i.tm.base_opcode;
2991 if ((unsigned char) *p == JUMP_PC_RELATIVE)
2992 subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
2993 else if ((cpu_arch_flags & Cpu386) != 0)
2994 subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
2995 else
2996 subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
2997 subtype |= code16;
2999 sym = i.op[0].disps->X_add_symbol;
3000 off = i.op[0].disps->X_add_number;
3002 if (i.op[0].disps->X_op != O_constant
3003 && i.op[0].disps->X_op != O_symbol)
3005 /* Handle complex expressions. */
3006 sym = make_expr_symbol (i.op[0].disps);
3007 off = 0;
3010 /* 1 possible extra opcode + 4 byte displacement go in var part.
3011 Pass reloc in fr_var. */
3012 frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
3015 static void
3016 output_jump ()
3018 char *p;
3019 int size;
3020 fixS *fixP;
3022 if (i.tm.opcode_modifier & JumpByte)
3024 /* This is a loop or jecxz type instruction. */
3025 size = 1;
3026 if (i.prefix[ADDR_PREFIX] != 0)
3028 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
3029 i.prefixes -= 1;
3031 /* Pentium4 branch hints. */
3032 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3033 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3035 FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
3036 i.prefixes--;
3039 else
3041 int code16;
3043 code16 = 0;
3044 if (flag_code == CODE_16BIT)
3045 code16 = CODE16;
3047 if (i.prefix[DATA_PREFIX] != 0)
3049 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
3050 i.prefixes -= 1;
3051 code16 ^= CODE16;
3054 size = 4;
3055 if (code16)
3056 size = 2;
3059 if (i.prefix[REX_PREFIX] != 0)
3061 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
3062 i.prefixes -= 1;
3065 if (i.prefixes != 0 && !intel_syntax)
3066 as_warn (_("skipping prefixes on this instruction"));
3068 p = frag_more (1 + size);
3069 *p++ = i.tm.base_opcode;
3071 fixP = fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3072 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
3074 /* All jumps handled here are signed, but don't use a signed limit
3075 check for 32 and 16 bit jumps as we want to allow wrap around at
3076 4G and 64k respectively. */
3077 if (size == 1)
3078 fixP->fx_signed = 1;
3081 static void
3082 output_interseg_jump ()
3084 char *p;
3085 int size;
3086 int prefix;
3087 int code16;
3089 code16 = 0;
3090 if (flag_code == CODE_16BIT)
3091 code16 = CODE16;
3093 prefix = 0;
3094 if (i.prefix[DATA_PREFIX] != 0)
3096 prefix = 1;
3097 i.prefixes -= 1;
3098 code16 ^= CODE16;
3100 if (i.prefix[REX_PREFIX] != 0)
3102 prefix++;
3103 i.prefixes -= 1;
3106 size = 4;
3107 if (code16)
3108 size = 2;
3110 if (i.prefixes != 0 && !intel_syntax)
3111 as_warn (_("skipping prefixes on this instruction"));
3113 /* 1 opcode; 2 segment; offset */
3114 p = frag_more (prefix + 1 + 2 + size);
3116 if (i.prefix[DATA_PREFIX] != 0)
3117 *p++ = DATA_PREFIX_OPCODE;
3119 if (i.prefix[REX_PREFIX] != 0)
3120 *p++ = i.prefix[REX_PREFIX];
3122 *p++ = i.tm.base_opcode;
3123 if (i.op[1].imms->X_op == O_constant)
3125 offsetT n = i.op[1].imms->X_add_number;
3127 if (size == 2
3128 && !fits_in_unsigned_word (n)
3129 && !fits_in_signed_word (n))
3131 as_bad (_("16-bit jump out of range"));
3132 return;
3134 md_number_to_chars (p, n, size);
3136 else
3137 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3138 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
3139 if (i.op[0].imms->X_op != O_constant)
3140 as_bad (_("can't handle non absolute segment in `%s'"),
3141 i.tm.name);
3142 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
3145 static void
3146 output_insn ()
3148 fragS *insn_start_frag;
3149 offsetT insn_start_off;
3151 /* Tie dwarf2 debug info to the address at the start of the insn.
3152 We can't do this after the insn has been output as the current
3153 frag may have been closed off. eg. by frag_var. */
3154 dwarf2_emit_insn (0);
3156 insn_start_frag = frag_now;
3157 insn_start_off = frag_now_fix ();
3159 /* Output jumps. */
3160 if (i.tm.opcode_modifier & Jump)
3161 output_branch ();
3162 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
3163 output_jump ();
3164 else if (i.tm.opcode_modifier & JumpInterSegment)
3165 output_interseg_jump ();
3166 else
3168 /* Output normal instructions here. */
3169 char *p;
3170 unsigned char *q;
3172 /* All opcodes on i386 have either 1 or 2 bytes, PadLock instructions
3173 have 3 bytes. We may use one more higher byte to specify a prefix
3174 the instruction requires. */
3175 if ((i.tm.cpu_flags & CpuPadLock) != 0
3176 && (i.tm.base_opcode & 0xff000000) != 0)
3178 unsigned int prefix;
3179 prefix = (i.tm.base_opcode >> 24) & 0xff;
3181 if (prefix != REPE_PREFIX_OPCODE
3182 || i.prefix[LOCKREP_PREFIX] != REPE_PREFIX_OPCODE)
3183 add_prefix (prefix);
3185 else
3186 if ((i.tm.cpu_flags & CpuPadLock) == 0
3187 && (i.tm.base_opcode & 0xff0000) != 0)
3188 add_prefix ((i.tm.base_opcode >> 16) & 0xff);
3190 /* The prefix bytes. */
3191 for (q = i.prefix;
3192 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
3193 q++)
3195 if (*q)
3197 p = frag_more (1);
3198 md_number_to_chars (p, (valueT) *q, 1);
3202 /* Now the opcode; be careful about word order here! */
3203 if (fits_in_unsigned_byte (i.tm.base_opcode))
3205 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
3207 else
3209 if ((i.tm.cpu_flags & CpuPadLock) != 0)
3211 p = frag_more (3);
3212 *p++ = (i.tm.base_opcode >> 16) & 0xff;
3214 else
3215 p = frag_more (2);
3217 /* Put out high byte first: can't use md_number_to_chars! */
3218 *p++ = (i.tm.base_opcode >> 8) & 0xff;
3219 *p = i.tm.base_opcode & 0xff;
3222 /* Now the modrm byte and sib byte (if present). */
3223 if (i.tm.opcode_modifier & Modrm)
3225 p = frag_more (1);
3226 md_number_to_chars (p,
3227 (valueT) (i.rm.regmem << 0
3228 | i.rm.reg << 3
3229 | i.rm.mode << 6),
3231 /* If i.rm.regmem == ESP (4)
3232 && i.rm.mode != (Register mode)
3233 && not 16 bit
3234 ==> need second modrm byte. */
3235 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
3236 && i.rm.mode != 3
3237 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
3239 p = frag_more (1);
3240 md_number_to_chars (p,
3241 (valueT) (i.sib.base << 0
3242 | i.sib.index << 3
3243 | i.sib.scale << 6),
3248 if (i.disp_operands)
3249 output_disp (insn_start_frag, insn_start_off);
3251 if (i.imm_operands)
3252 output_imm (insn_start_frag, insn_start_off);
3255 #ifdef DEBUG386
3256 if (flag_debug)
3258 pi (line, &i);
3260 #endif /* DEBUG386 */
3263 static void
3264 output_disp (insn_start_frag, insn_start_off)
3265 fragS *insn_start_frag;
3266 offsetT insn_start_off;
3268 char *p;
3269 unsigned int n;
3271 for (n = 0; n < i.operands; n++)
3273 if (i.types[n] & Disp)
3275 if (i.op[n].disps->X_op == O_constant)
3277 int size;
3278 offsetT val;
3280 size = 4;
3281 if (i.types[n] & (Disp8 | Disp16 | Disp64))
3283 size = 2;
3284 if (i.types[n] & Disp8)
3285 size = 1;
3286 if (i.types[n] & Disp64)
3287 size = 8;
3289 val = offset_in_range (i.op[n].disps->X_add_number,
3290 size);
3291 p = frag_more (size);
3292 md_number_to_chars (p, val, size);
3294 else
3296 enum bfd_reloc_code_real reloc_type;
3297 int size = 4;
3298 int sign = 0;
3299 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
3301 /* The PC relative address is computed relative
3302 to the instruction boundary, so in case immediate
3303 fields follows, we need to adjust the value. */
3304 if (pcrel && i.imm_operands)
3306 int imm_size = 4;
3307 unsigned int n1;
3309 for (n1 = 0; n1 < i.operands; n1++)
3310 if (i.types[n1] & Imm)
3312 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
3314 imm_size = 2;
3315 if (i.types[n1] & (Imm8 | Imm8S))
3316 imm_size = 1;
3317 if (i.types[n1] & Imm64)
3318 imm_size = 8;
3320 break;
3322 /* We should find the immediate. */
3323 if (n1 == i.operands)
3324 abort ();
3325 i.op[n].disps->X_add_number -= imm_size;
3328 if (i.types[n] & Disp32S)
3329 sign = 1;
3331 if (i.types[n] & (Disp16 | Disp64))
3333 size = 2;
3334 if (i.types[n] & Disp64)
3335 size = 8;
3338 p = frag_more (size);
3339 reloc_type = reloc (size, pcrel, sign, i.reloc[n]);
3340 if (reloc_type == BFD_RELOC_32
3341 && GOT_symbol
3342 && GOT_symbol == i.op[n].disps->X_add_symbol
3343 && (i.op[n].disps->X_op == O_symbol
3344 || (i.op[n].disps->X_op == O_add
3345 && ((symbol_get_value_expression
3346 (i.op[n].disps->X_op_symbol)->X_op)
3347 == O_subtract))))
3349 offsetT add;
3351 if (insn_start_frag == frag_now)
3352 add = (p - frag_now->fr_literal) - insn_start_off;
3353 else
3355 fragS *fr;
3357 add = insn_start_frag->fr_fix - insn_start_off;
3358 for (fr = insn_start_frag->fr_next;
3359 fr && fr != frag_now; fr = fr->fr_next)
3360 add += fr->fr_fix;
3361 add += p - frag_now->fr_literal;
3364 /* We don't support dynamic linking on x86-64 yet. */
3365 if (flag_code == CODE_64BIT)
3366 abort ();
3367 reloc_type = BFD_RELOC_386_GOTPC;
3368 i.op[n].disps->X_add_number += add;
3370 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3371 i.op[n].disps, pcrel, reloc_type);
3377 static void
3378 output_imm (insn_start_frag, insn_start_off)
3379 fragS *insn_start_frag;
3380 offsetT insn_start_off;
3382 char *p;
3383 unsigned int n;
3385 for (n = 0; n < i.operands; n++)
3387 if (i.types[n] & Imm)
3389 if (i.op[n].imms->X_op == O_constant)
3391 int size;
3392 offsetT val;
3394 size = 4;
3395 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3397 size = 2;
3398 if (i.types[n] & (Imm8 | Imm8S))
3399 size = 1;
3400 else if (i.types[n] & Imm64)
3401 size = 8;
3403 val = offset_in_range (i.op[n].imms->X_add_number,
3404 size);
3405 p = frag_more (size);
3406 md_number_to_chars (p, val, size);
3408 else
3410 /* Not absolute_section.
3411 Need a 32-bit fixup (don't support 8bit
3412 non-absolute imms). Try to support other
3413 sizes ... */
3414 enum bfd_reloc_code_real reloc_type;
3415 int size = 4;
3416 int sign = 0;
3418 if ((i.types[n] & (Imm32S))
3419 && i.suffix == QWORD_MNEM_SUFFIX)
3420 sign = 1;
3421 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3423 size = 2;
3424 if (i.types[n] & (Imm8 | Imm8S))
3425 size = 1;
3426 if (i.types[n] & Imm64)
3427 size = 8;
3430 p = frag_more (size);
3431 reloc_type = reloc (size, 0, sign, i.reloc[n]);
3433 /* This is tough to explain. We end up with this one if we
3434 * have operands that look like
3435 * "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal here is to
3436 * obtain the absolute address of the GOT, and it is strongly
3437 * preferable from a performance point of view to avoid using
3438 * a runtime relocation for this. The actual sequence of
3439 * instructions often look something like:
3441 * call .L66
3442 * .L66:
3443 * popl %ebx
3444 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
3446 * The call and pop essentially return the absolute address
3447 * of the label .L66 and store it in %ebx. The linker itself
3448 * will ultimately change the first operand of the addl so
3449 * that %ebx points to the GOT, but to keep things simple, the
3450 * .o file must have this operand set so that it generates not
3451 * the absolute address of .L66, but the absolute address of
3452 * itself. This allows the linker itself simply treat a GOTPC
3453 * relocation as asking for a pcrel offset to the GOT to be
3454 * added in, and the addend of the relocation is stored in the
3455 * operand field for the instruction itself.
3457 * Our job here is to fix the operand so that it would add
3458 * the correct offset so that %ebx would point to itself. The
3459 * thing that is tricky is that .-.L66 will point to the
3460 * beginning of the instruction, so we need to further modify
3461 * the operand so that it will point to itself. There are
3462 * other cases where you have something like:
3464 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
3466 * and here no correction would be required. Internally in
3467 * the assembler we treat operands of this form as not being
3468 * pcrel since the '.' is explicitly mentioned, and I wonder
3469 * whether it would simplify matters to do it this way. Who
3470 * knows. In earlier versions of the PIC patches, the
3471 * pcrel_adjust field was used to store the correction, but
3472 * since the expression is not pcrel, I felt it would be
3473 * confusing to do it this way. */
3475 if (reloc_type == BFD_RELOC_32
3476 && GOT_symbol
3477 && GOT_symbol == i.op[n].imms->X_add_symbol
3478 && (i.op[n].imms->X_op == O_symbol
3479 || (i.op[n].imms->X_op == O_add
3480 && ((symbol_get_value_expression
3481 (i.op[n].imms->X_op_symbol)->X_op)
3482 == O_subtract))))
3484 offsetT add;
3486 if (insn_start_frag == frag_now)
3487 add = (p - frag_now->fr_literal) - insn_start_off;
3488 else
3490 fragS *fr;
3492 add = insn_start_frag->fr_fix - insn_start_off;
3493 for (fr = insn_start_frag->fr_next;
3494 fr && fr != frag_now; fr = fr->fr_next)
3495 add += fr->fr_fix;
3496 add += p - frag_now->fr_literal;
3499 /* We don't support dynamic linking on x86-64 yet. */
3500 if (flag_code == CODE_64BIT)
3501 abort ();
3502 reloc_type = BFD_RELOC_386_GOTPC;
3503 i.op[n].imms->X_add_number += add;
3505 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3506 i.op[n].imms, 0, reloc_type);
3512 #ifndef LEX_AT
3513 static char *lex_got PARAMS ((enum bfd_reloc_code_real *, int *));
3515 /* Parse operands of the form
3516 <symbol>@GOTOFF+<nnn>
3517 and similar .plt or .got references.
3519 If we find one, set up the correct relocation in RELOC and copy the
3520 input string, minus the `@GOTOFF' into a malloc'd buffer for
3521 parsing by the calling routine. Return this buffer, and if ADJUST
3522 is non-null set it to the length of the string we removed from the
3523 input line. Otherwise return NULL. */
3524 static char *
3525 lex_got (reloc, adjust)
3526 enum bfd_reloc_code_real *reloc;
3527 int *adjust;
3529 static const char * const mode_name[NUM_FLAG_CODE] = { "32", "16", "64" };
3530 static const struct {
3531 const char *str;
3532 const enum bfd_reloc_code_real rel[NUM_FLAG_CODE];
3533 } gotrel[] = {
3534 { "PLT", { BFD_RELOC_386_PLT32, 0, BFD_RELOC_X86_64_PLT32 } },
3535 { "GOTOFF", { BFD_RELOC_386_GOTOFF, 0, 0 } },
3536 { "GOTPCREL", { 0, 0, BFD_RELOC_X86_64_GOTPCREL } },
3537 { "TLSGD", { BFD_RELOC_386_TLS_GD, 0, BFD_RELOC_X86_64_TLSGD } },
3538 { "TLSLDM", { BFD_RELOC_386_TLS_LDM, 0, 0 } },
3539 { "TLSLD", { 0, 0, BFD_RELOC_X86_64_TLSLD } },
3540 { "GOTTPOFF", { BFD_RELOC_386_TLS_IE_32, 0, BFD_RELOC_X86_64_GOTTPOFF } },
3541 { "TPOFF", { BFD_RELOC_386_TLS_LE_32, 0, BFD_RELOC_X86_64_TPOFF32 } },
3542 { "NTPOFF", { BFD_RELOC_386_TLS_LE, 0, 0 } },
3543 { "DTPOFF", { BFD_RELOC_386_TLS_LDO_32, 0, BFD_RELOC_X86_64_DTPOFF32 } },
3544 { "GOTNTPOFF",{ BFD_RELOC_386_TLS_GOTIE, 0, 0 } },
3545 { "INDNTPOFF",{ BFD_RELOC_386_TLS_IE, 0, 0 } },
3546 { "GOT", { BFD_RELOC_386_GOT32, 0, BFD_RELOC_X86_64_GOT32 } }
3548 char *cp;
3549 unsigned int j;
3551 for (cp = input_line_pointer; *cp != '@'; cp++)
3552 if (is_end_of_line[(unsigned char) *cp])
3553 return NULL;
3555 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
3557 int len;
3559 len = strlen (gotrel[j].str);
3560 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
3562 if (gotrel[j].rel[(unsigned int) flag_code] != 0)
3564 int first, second;
3565 char *tmpbuf, *past_reloc;
3567 *reloc = gotrel[j].rel[(unsigned int) flag_code];
3568 if (adjust)
3569 *adjust = len;
3571 if (GOT_symbol == NULL)
3572 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3574 /* Replace the relocation token with ' ', so that
3575 errors like foo@GOTOFF1 will be detected. */
3577 /* The length of the first part of our input line. */
3578 first = cp - input_line_pointer;
3580 /* The second part goes from after the reloc token until
3581 (and including) an end_of_line char. Don't use strlen
3582 here as the end_of_line char may not be a NUL. */
3583 past_reloc = cp + 1 + len;
3584 for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
3586 second = cp - past_reloc;
3588 /* Allocate and copy string. The trailing NUL shouldn't
3589 be necessary, but be safe. */
3590 tmpbuf = xmalloc (first + second + 2);
3591 memcpy (tmpbuf, input_line_pointer, first);
3592 tmpbuf[first] = ' ';
3593 memcpy (tmpbuf + first + 1, past_reloc, second);
3594 tmpbuf[first + second + 1] = '\0';
3595 return tmpbuf;
3598 as_bad (_("@%s reloc is not supported in %s bit mode"),
3599 gotrel[j].str, mode_name[(unsigned int) flag_code]);
3600 return NULL;
3604 /* Might be a symbol version string. Don't as_bad here. */
3605 return NULL;
3608 /* x86_cons_fix_new is called via the expression parsing code when a
3609 reloc is needed. We use this hook to get the correct .got reloc. */
3610 static enum bfd_reloc_code_real got_reloc = NO_RELOC;
3612 void
3613 x86_cons_fix_new (frag, off, len, exp)
3614 fragS *frag;
3615 unsigned int off;
3616 unsigned int len;
3617 expressionS *exp;
3619 enum bfd_reloc_code_real r = reloc (len, 0, 0, got_reloc);
3620 got_reloc = NO_RELOC;
3621 fix_new_exp (frag, off, len, exp, 0, r);
3624 void
3625 x86_cons (exp, size)
3626 expressionS *exp;
3627 int size;
3629 if (size == 4)
3631 /* Handle @GOTOFF and the like in an expression. */
3632 char *save;
3633 char *gotfree_input_line;
3634 int adjust;
3636 save = input_line_pointer;
3637 gotfree_input_line = lex_got (&got_reloc, &adjust);
3638 if (gotfree_input_line)
3639 input_line_pointer = gotfree_input_line;
3641 expression (exp);
3643 if (gotfree_input_line)
3645 /* expression () has merrily parsed up to the end of line,
3646 or a comma - in the wrong buffer. Transfer how far
3647 input_line_pointer has moved to the right buffer. */
3648 input_line_pointer = (save
3649 + (input_line_pointer - gotfree_input_line)
3650 + adjust);
3651 free (gotfree_input_line);
3654 else
3655 expression (exp);
3657 #endif
3659 #ifdef TE_PE
3661 void
3662 x86_pe_cons_fix_new (frag, off, len, exp)
3663 fragS *frag;
3664 unsigned int off;
3665 unsigned int len;
3666 expressionS *exp;
3668 enum bfd_reloc_code_real r = reloc (len, 0, 0, NO_RELOC);
3670 if (exp->X_op == O_secrel)
3672 exp->X_op = O_symbol;
3673 r = BFD_RELOC_32_SECREL;
3676 fix_new_exp (frag, off, len, exp, 0, r);
3679 static void
3680 pe_directive_secrel (dummy)
3681 int dummy ATTRIBUTE_UNUSED;
3683 expressionS exp;
3687 expression (&exp);
3688 if (exp.X_op == O_symbol)
3689 exp.X_op = O_secrel;
3691 emit_expr (&exp, 4);
3693 while (*input_line_pointer++ == ',');
3695 input_line_pointer--;
3696 demand_empty_rest_of_line ();
3699 #endif
3701 static int i386_immediate PARAMS ((char *));
3703 static int
3704 i386_immediate (imm_start)
3705 char *imm_start;
3707 char *save_input_line_pointer;
3708 #ifndef LEX_AT
3709 char *gotfree_input_line;
3710 #endif
3711 segT exp_seg = 0;
3712 expressionS *exp;
3714 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
3716 as_bad (_("only 1 or 2 immediate operands are allowed"));
3717 return 0;
3720 exp = &im_expressions[i.imm_operands++];
3721 i.op[this_operand].imms = exp;
3723 if (is_space_char (*imm_start))
3724 ++imm_start;
3726 save_input_line_pointer = input_line_pointer;
3727 input_line_pointer = imm_start;
3729 #ifndef LEX_AT
3730 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3731 if (gotfree_input_line)
3732 input_line_pointer = gotfree_input_line;
3733 #endif
3735 exp_seg = expression (exp);
3737 SKIP_WHITESPACE ();
3738 if (*input_line_pointer)
3739 as_bad (_("junk `%s' after expression"), input_line_pointer);
3741 input_line_pointer = save_input_line_pointer;
3742 #ifndef LEX_AT
3743 if (gotfree_input_line)
3744 free (gotfree_input_line);
3745 #endif
3747 if (exp->X_op == O_absent || exp->X_op == O_big)
3749 /* Missing or bad expr becomes absolute 0. */
3750 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
3751 imm_start);
3752 exp->X_op = O_constant;
3753 exp->X_add_number = 0;
3754 exp->X_add_symbol = (symbolS *) 0;
3755 exp->X_op_symbol = (symbolS *) 0;
3757 else if (exp->X_op == O_constant)
3759 /* Size it properly later. */
3760 i.types[this_operand] |= Imm64;
3761 /* If BFD64, sign extend val. */
3762 if (!use_rela_relocations)
3763 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
3764 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
3766 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3767 else if (OUTPUT_FLAVOR == bfd_target_aout_flavour
3768 && exp_seg != absolute_section
3769 && exp_seg != text_section
3770 && exp_seg != data_section
3771 && exp_seg != bss_section
3772 && exp_seg != undefined_section
3773 && !bfd_is_com_section (exp_seg))
3775 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3776 return 0;
3778 #endif
3779 else
3781 /* This is an address. The size of the address will be
3782 determined later, depending on destination register,
3783 suffix, or the default for the section. */
3784 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
3787 return 1;
3790 static char *i386_scale PARAMS ((char *));
3792 static char *
3793 i386_scale (scale)
3794 char *scale;
3796 offsetT val;
3797 char *save = input_line_pointer;
3799 input_line_pointer = scale;
3800 val = get_absolute_expression ();
3802 switch (val)
3804 case 0:
3805 case 1:
3806 i.log2_scale_factor = 0;
3807 break;
3808 case 2:
3809 i.log2_scale_factor = 1;
3810 break;
3811 case 4:
3812 i.log2_scale_factor = 2;
3813 break;
3814 case 8:
3815 i.log2_scale_factor = 3;
3816 break;
3817 default:
3818 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
3819 scale);
3820 input_line_pointer = save;
3821 return NULL;
3823 if (i.log2_scale_factor != 0 && i.index_reg == 0)
3825 as_warn (_("scale factor of %d without an index register"),
3826 1 << i.log2_scale_factor);
3827 #if SCALE1_WHEN_NO_INDEX
3828 i.log2_scale_factor = 0;
3829 #endif
3831 scale = input_line_pointer;
3832 input_line_pointer = save;
3833 return scale;
3836 static int i386_displacement PARAMS ((char *, char *));
3838 static int
3839 i386_displacement (disp_start, disp_end)
3840 char *disp_start;
3841 char *disp_end;
3843 expressionS *exp;
3844 segT exp_seg = 0;
3845 char *save_input_line_pointer;
3846 #ifndef LEX_AT
3847 char *gotfree_input_line;
3848 #endif
3849 int bigdisp = Disp32;
3851 if (flag_code == CODE_64BIT)
3853 if (i.prefix[ADDR_PREFIX] == 0)
3854 bigdisp = Disp64;
3856 else if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3857 bigdisp = Disp16;
3858 i.types[this_operand] |= bigdisp;
3860 exp = &disp_expressions[i.disp_operands];
3861 i.op[this_operand].disps = exp;
3862 i.disp_operands++;
3863 save_input_line_pointer = input_line_pointer;
3864 input_line_pointer = disp_start;
3865 END_STRING_AND_SAVE (disp_end);
3867 #ifndef GCC_ASM_O_HACK
3868 #define GCC_ASM_O_HACK 0
3869 #endif
3870 #if GCC_ASM_O_HACK
3871 END_STRING_AND_SAVE (disp_end + 1);
3872 if ((i.types[this_operand] & BaseIndex) != 0
3873 && displacement_string_end[-1] == '+')
3875 /* This hack is to avoid a warning when using the "o"
3876 constraint within gcc asm statements.
3877 For instance:
3879 #define _set_tssldt_desc(n,addr,limit,type) \
3880 __asm__ __volatile__ ( \
3881 "movw %w2,%0\n\t" \
3882 "movw %w1,2+%0\n\t" \
3883 "rorl $16,%1\n\t" \
3884 "movb %b1,4+%0\n\t" \
3885 "movb %4,5+%0\n\t" \
3886 "movb $0,6+%0\n\t" \
3887 "movb %h1,7+%0\n\t" \
3888 "rorl $16,%1" \
3889 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
3891 This works great except that the output assembler ends
3892 up looking a bit weird if it turns out that there is
3893 no offset. You end up producing code that looks like:
3895 #APP
3896 movw $235,(%eax)
3897 movw %dx,2+(%eax)
3898 rorl $16,%edx
3899 movb %dl,4+(%eax)
3900 movb $137,5+(%eax)
3901 movb $0,6+(%eax)
3902 movb %dh,7+(%eax)
3903 rorl $16,%edx
3904 #NO_APP
3906 So here we provide the missing zero. */
3908 *displacement_string_end = '0';
3910 #endif
3911 #ifndef LEX_AT
3912 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3913 if (gotfree_input_line)
3914 input_line_pointer = gotfree_input_line;
3915 #endif
3917 exp_seg = expression (exp);
3919 SKIP_WHITESPACE ();
3920 if (*input_line_pointer)
3921 as_bad (_("junk `%s' after expression"), input_line_pointer);
3922 #if GCC_ASM_O_HACK
3923 RESTORE_END_STRING (disp_end + 1);
3924 #endif
3925 RESTORE_END_STRING (disp_end);
3926 input_line_pointer = save_input_line_pointer;
3927 #ifndef LEX_AT
3928 if (gotfree_input_line)
3929 free (gotfree_input_line);
3930 #endif
3932 /* We do this to make sure that the section symbol is in
3933 the symbol table. We will ultimately change the relocation
3934 to be relative to the beginning of the section. */
3935 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
3936 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3938 if (exp->X_op != O_symbol)
3940 as_bad (_("bad expression used with @%s"),
3941 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
3942 ? "GOTPCREL"
3943 : "GOTOFF"));
3944 return 0;
3947 if (S_IS_LOCAL (exp->X_add_symbol)
3948 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
3949 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
3950 exp->X_op = O_subtract;
3951 exp->X_op_symbol = GOT_symbol;
3952 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3953 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
3954 else
3955 i.reloc[this_operand] = BFD_RELOC_32;
3958 if (exp->X_op == O_absent || exp->X_op == O_big)
3960 /* Missing or bad expr becomes absolute 0. */
3961 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
3962 disp_start);
3963 exp->X_op = O_constant;
3964 exp->X_add_number = 0;
3965 exp->X_add_symbol = (symbolS *) 0;
3966 exp->X_op_symbol = (symbolS *) 0;
3969 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3970 if (exp->X_op != O_constant
3971 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3972 && exp_seg != absolute_section
3973 && exp_seg != text_section
3974 && exp_seg != data_section
3975 && exp_seg != bss_section
3976 && exp_seg != undefined_section
3977 && !bfd_is_com_section (exp_seg))
3979 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3980 return 0;
3982 #endif
3983 else if (flag_code == CODE_64BIT)
3984 i.types[this_operand] |= Disp32S | Disp32;
3985 return 1;
3988 static int i386_index_check PARAMS ((const char *));
3990 /* Make sure the memory operand we've been dealt is valid.
3991 Return 1 on success, 0 on a failure. */
3993 static int
3994 i386_index_check (operand_string)
3995 const char *operand_string;
3997 int ok;
3998 #if INFER_ADDR_PREFIX
3999 int fudged = 0;
4001 tryprefix:
4002 #endif
4003 ok = 1;
4004 if (flag_code == CODE_64BIT)
4006 unsigned RegXX = (i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32);
4008 if ((i.base_reg
4009 && ((i.base_reg->reg_type & RegXX) == 0)
4010 && (i.base_reg->reg_type != BaseIndex
4011 || i.index_reg))
4012 || (i.index_reg
4013 && ((i.index_reg->reg_type & (RegXX | BaseIndex))
4014 != (RegXX | BaseIndex))))
4015 ok = 0;
4017 else
4019 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
4021 /* 16bit checks. */
4022 if ((i.base_reg
4023 && ((i.base_reg->reg_type & (Reg16 | BaseIndex | RegRex))
4024 != (Reg16 | BaseIndex)))
4025 || (i.index_reg
4026 && (((i.index_reg->reg_type & (Reg16 | BaseIndex))
4027 != (Reg16 | BaseIndex))
4028 || !(i.base_reg
4029 && i.base_reg->reg_num < 6
4030 && i.index_reg->reg_num >= 6
4031 && i.log2_scale_factor == 0))))
4032 ok = 0;
4034 else
4036 /* 32bit checks. */
4037 if ((i.base_reg
4038 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
4039 || (i.index_reg
4040 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
4041 != (Reg32 | BaseIndex))))
4042 ok = 0;
4045 if (!ok)
4047 #if INFER_ADDR_PREFIX
4048 if (i.prefix[ADDR_PREFIX] == 0)
4050 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
4051 i.prefixes += 1;
4052 /* Change the size of any displacement too. At most one of
4053 Disp16 or Disp32 is set.
4054 FIXME. There doesn't seem to be any real need for separate
4055 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
4056 Removing them would probably clean up the code quite a lot. */
4057 if (flag_code != CODE_64BIT && (i.types[this_operand] & (Disp16 | Disp32)))
4058 i.types[this_operand] ^= (Disp16 | Disp32);
4059 fudged = 1;
4060 goto tryprefix;
4062 if (fudged)
4063 as_bad (_("`%s' is not a valid base/index expression"),
4064 operand_string);
4065 else
4066 #endif
4067 as_bad (_("`%s' is not a valid %s bit base/index expression"),
4068 operand_string,
4069 flag_code_names[flag_code]);
4071 return ok;
4074 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
4075 on error. */
4077 static int
4078 i386_operand (operand_string)
4079 char *operand_string;
4081 const reg_entry *r;
4082 char *end_op;
4083 char *op_string = operand_string;
4085 if (is_space_char (*op_string))
4086 ++op_string;
4088 /* We check for an absolute prefix (differentiating,
4089 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
4090 if (*op_string == ABSOLUTE_PREFIX)
4092 ++op_string;
4093 if (is_space_char (*op_string))
4094 ++op_string;
4095 i.types[this_operand] |= JumpAbsolute;
4098 /* Check if operand is a register. */
4099 if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
4100 && (r = parse_register (op_string, &end_op)) != NULL)
4102 /* Check for a segment override by searching for ':' after a
4103 segment register. */
4104 op_string = end_op;
4105 if (is_space_char (*op_string))
4106 ++op_string;
4107 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
4109 switch (r->reg_num)
4111 case 0:
4112 i.seg[i.mem_operands] = &es;
4113 break;
4114 case 1:
4115 i.seg[i.mem_operands] = &cs;
4116 break;
4117 case 2:
4118 i.seg[i.mem_operands] = &ss;
4119 break;
4120 case 3:
4121 i.seg[i.mem_operands] = &ds;
4122 break;
4123 case 4:
4124 i.seg[i.mem_operands] = &fs;
4125 break;
4126 case 5:
4127 i.seg[i.mem_operands] = &gs;
4128 break;
4131 /* Skip the ':' and whitespace. */
4132 ++op_string;
4133 if (is_space_char (*op_string))
4134 ++op_string;
4136 if (!is_digit_char (*op_string)
4137 && !is_identifier_char (*op_string)
4138 && *op_string != '('
4139 && *op_string != ABSOLUTE_PREFIX)
4141 as_bad (_("bad memory operand `%s'"), op_string);
4142 return 0;
4144 /* Handle case of %es:*foo. */
4145 if (*op_string == ABSOLUTE_PREFIX)
4147 ++op_string;
4148 if (is_space_char (*op_string))
4149 ++op_string;
4150 i.types[this_operand] |= JumpAbsolute;
4152 goto do_memory_reference;
4154 if (*op_string)
4156 as_bad (_("junk `%s' after register"), op_string);
4157 return 0;
4159 i.types[this_operand] |= r->reg_type & ~BaseIndex;
4160 i.op[this_operand].regs = r;
4161 i.reg_operands++;
4163 else if (*op_string == REGISTER_PREFIX)
4165 as_bad (_("bad register name `%s'"), op_string);
4166 return 0;
4168 else if (*op_string == IMMEDIATE_PREFIX)
4170 ++op_string;
4171 if (i.types[this_operand] & JumpAbsolute)
4173 as_bad (_("immediate operand illegal with absolute jump"));
4174 return 0;
4176 if (!i386_immediate (op_string))
4177 return 0;
4179 else if (is_digit_char (*op_string)
4180 || is_identifier_char (*op_string)
4181 || *op_string == '(')
4183 /* This is a memory reference of some sort. */
4184 char *base_string;
4186 /* Start and end of displacement string expression (if found). */
4187 char *displacement_string_start;
4188 char *displacement_string_end;
4190 do_memory_reference:
4191 if ((i.mem_operands == 1
4192 && (current_templates->start->opcode_modifier & IsString) == 0)
4193 || i.mem_operands == 2)
4195 as_bad (_("too many memory references for `%s'"),
4196 current_templates->start->name);
4197 return 0;
4200 /* Check for base index form. We detect the base index form by
4201 looking for an ')' at the end of the operand, searching
4202 for the '(' matching it, and finding a REGISTER_PREFIX or ','
4203 after the '('. */
4204 base_string = op_string + strlen (op_string);
4206 --base_string;
4207 if (is_space_char (*base_string))
4208 --base_string;
4210 /* If we only have a displacement, set-up for it to be parsed later. */
4211 displacement_string_start = op_string;
4212 displacement_string_end = base_string + 1;
4214 if (*base_string == ')')
4216 char *temp_string;
4217 unsigned int parens_balanced = 1;
4218 /* We've already checked that the number of left & right ()'s are
4219 equal, so this loop will not be infinite. */
4222 base_string--;
4223 if (*base_string == ')')
4224 parens_balanced++;
4225 if (*base_string == '(')
4226 parens_balanced--;
4228 while (parens_balanced);
4230 temp_string = base_string;
4232 /* Skip past '(' and whitespace. */
4233 ++base_string;
4234 if (is_space_char (*base_string))
4235 ++base_string;
4237 if (*base_string == ','
4238 || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
4239 && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
4241 displacement_string_end = temp_string;
4243 i.types[this_operand] |= BaseIndex;
4245 if (i.base_reg)
4247 base_string = end_op;
4248 if (is_space_char (*base_string))
4249 ++base_string;
4252 /* There may be an index reg or scale factor here. */
4253 if (*base_string == ',')
4255 ++base_string;
4256 if (is_space_char (*base_string))
4257 ++base_string;
4259 if ((*base_string == REGISTER_PREFIX || allow_naked_reg)
4260 && (i.index_reg = parse_register (base_string, &end_op)) != NULL)
4262 base_string = end_op;
4263 if (is_space_char (*base_string))
4264 ++base_string;
4265 if (*base_string == ',')
4267 ++base_string;
4268 if (is_space_char (*base_string))
4269 ++base_string;
4271 else if (*base_string != ')')
4273 as_bad (_("expecting `,' or `)' after index register in `%s'"),
4274 operand_string);
4275 return 0;
4278 else if (*base_string == REGISTER_PREFIX)
4280 as_bad (_("bad register name `%s'"), base_string);
4281 return 0;
4284 /* Check for scale factor. */
4285 if (*base_string != ')')
4287 char *end_scale = i386_scale (base_string);
4289 if (!end_scale)
4290 return 0;
4292 base_string = end_scale;
4293 if (is_space_char (*base_string))
4294 ++base_string;
4295 if (*base_string != ')')
4297 as_bad (_("expecting `)' after scale factor in `%s'"),
4298 operand_string);
4299 return 0;
4302 else if (!i.index_reg)
4304 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
4305 *base_string);
4306 return 0;
4309 else if (*base_string != ')')
4311 as_bad (_("expecting `,' or `)' after base register in `%s'"),
4312 operand_string);
4313 return 0;
4316 else if (*base_string == REGISTER_PREFIX)
4318 as_bad (_("bad register name `%s'"), base_string);
4319 return 0;
4323 /* If there's an expression beginning the operand, parse it,
4324 assuming displacement_string_start and
4325 displacement_string_end are meaningful. */
4326 if (displacement_string_start != displacement_string_end)
4328 if (!i386_displacement (displacement_string_start,
4329 displacement_string_end))
4330 return 0;
4333 /* Special case for (%dx) while doing input/output op. */
4334 if (i.base_reg
4335 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
4336 && i.index_reg == 0
4337 && i.log2_scale_factor == 0
4338 && i.seg[i.mem_operands] == 0
4339 && (i.types[this_operand] & Disp) == 0)
4341 i.types[this_operand] = InOutPortReg;
4342 return 1;
4345 if (i386_index_check (operand_string) == 0)
4346 return 0;
4347 i.mem_operands++;
4349 else
4351 /* It's not a memory operand; argh! */
4352 as_bad (_("invalid char %s beginning operand %d `%s'"),
4353 output_invalid (*op_string),
4354 this_operand + 1,
4355 op_string);
4356 return 0;
4358 return 1; /* Normal return. */
4361 /* md_estimate_size_before_relax()
4363 Called just before relax() for rs_machine_dependent frags. The x86
4364 assembler uses these frags to handle variable size jump
4365 instructions.
4367 Any symbol that is now undefined will not become defined.
4368 Return the correct fr_subtype in the frag.
4369 Return the initial "guess for variable size of frag" to caller.
4370 The guess is actually the growth beyond the fixed part. Whatever
4371 we do to grow the fixed or variable part contributes to our
4372 returned value. */
4375 md_estimate_size_before_relax (fragP, segment)
4376 fragS *fragP;
4377 segT segment;
4379 /* We've already got fragP->fr_subtype right; all we have to do is
4380 check for un-relaxable symbols. On an ELF system, we can't relax
4381 an externally visible symbol, because it may be overridden by a
4382 shared library. */
4383 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
4384 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4385 || (OUTPUT_FLAVOR == bfd_target_elf_flavour
4386 && (S_IS_EXTERNAL (fragP->fr_symbol)
4387 || S_IS_WEAK (fragP->fr_symbol)))
4388 #endif
4391 /* Symbol is undefined in this segment, or we need to keep a
4392 reloc so that weak symbols can be overridden. */
4393 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
4394 enum bfd_reloc_code_real reloc_type;
4395 unsigned char *opcode;
4396 int old_fr_fix;
4398 if (fragP->fr_var != NO_RELOC)
4399 reloc_type = fragP->fr_var;
4400 else if (size == 2)
4401 reloc_type = BFD_RELOC_16_PCREL;
4402 else
4403 reloc_type = BFD_RELOC_32_PCREL;
4405 old_fr_fix = fragP->fr_fix;
4406 opcode = (unsigned char *) fragP->fr_opcode;
4408 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
4410 case UNCOND_JUMP:
4411 /* Make jmp (0xeb) a (d)word displacement jump. */
4412 opcode[0] = 0xe9;
4413 fragP->fr_fix += size;
4414 fix_new (fragP, old_fr_fix, size,
4415 fragP->fr_symbol,
4416 fragP->fr_offset, 1,
4417 reloc_type);
4418 break;
4420 case COND_JUMP86:
4421 if (size == 2
4422 && (!no_cond_jump_promotion || fragP->fr_var != NO_RELOC))
4424 /* Negate the condition, and branch past an
4425 unconditional jump. */
4426 opcode[0] ^= 1;
4427 opcode[1] = 3;
4428 /* Insert an unconditional jump. */
4429 opcode[2] = 0xe9;
4430 /* We added two extra opcode bytes, and have a two byte
4431 offset. */
4432 fragP->fr_fix += 2 + 2;
4433 fix_new (fragP, old_fr_fix + 2, 2,
4434 fragP->fr_symbol,
4435 fragP->fr_offset, 1,
4436 reloc_type);
4437 break;
4439 /* Fall through. */
4441 case COND_JUMP:
4442 if (no_cond_jump_promotion && fragP->fr_var == NO_RELOC)
4444 fixS *fixP;
4446 fragP->fr_fix += 1;
4447 fixP = fix_new (fragP, old_fr_fix, 1,
4448 fragP->fr_symbol,
4449 fragP->fr_offset, 1,
4450 BFD_RELOC_8_PCREL);
4451 fixP->fx_signed = 1;
4452 break;
4455 /* This changes the byte-displacement jump 0x7N
4456 to the (d)word-displacement jump 0x0f,0x8N. */
4457 opcode[1] = opcode[0] + 0x10;
4458 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4459 /* We've added an opcode byte. */
4460 fragP->fr_fix += 1 + size;
4461 fix_new (fragP, old_fr_fix + 1, size,
4462 fragP->fr_symbol,
4463 fragP->fr_offset, 1,
4464 reloc_type);
4465 break;
4467 default:
4468 BAD_CASE (fragP->fr_subtype);
4469 break;
4471 frag_wane (fragP);
4472 return fragP->fr_fix - old_fr_fix;
4475 /* Guess size depending on current relax state. Initially the relax
4476 state will correspond to a short jump and we return 1, because
4477 the variable part of the frag (the branch offset) is one byte
4478 long. However, we can relax a section more than once and in that
4479 case we must either set fr_subtype back to the unrelaxed state,
4480 or return the value for the appropriate branch. */
4481 return md_relax_table[fragP->fr_subtype].rlx_length;
4484 /* Called after relax() is finished.
4486 In: Address of frag.
4487 fr_type == rs_machine_dependent.
4488 fr_subtype is what the address relaxed to.
4490 Out: Any fixSs and constants are set up.
4491 Caller will turn frag into a ".space 0". */
4493 void
4494 md_convert_frag (abfd, sec, fragP)
4495 bfd *abfd ATTRIBUTE_UNUSED;
4496 segT sec ATTRIBUTE_UNUSED;
4497 fragS *fragP;
4499 unsigned char *opcode;
4500 unsigned char *where_to_put_displacement = NULL;
4501 offsetT target_address;
4502 offsetT opcode_address;
4503 unsigned int extension = 0;
4504 offsetT displacement_from_opcode_start;
4506 opcode = (unsigned char *) fragP->fr_opcode;
4508 /* Address we want to reach in file space. */
4509 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
4511 /* Address opcode resides at in file space. */
4512 opcode_address = fragP->fr_address + fragP->fr_fix;
4514 /* Displacement from opcode start to fill into instruction. */
4515 displacement_from_opcode_start = target_address - opcode_address;
4517 if ((fragP->fr_subtype & BIG) == 0)
4519 /* Don't have to change opcode. */
4520 extension = 1; /* 1 opcode + 1 displacement */
4521 where_to_put_displacement = &opcode[1];
4523 else
4525 if (no_cond_jump_promotion
4526 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
4527 as_warn_where (fragP->fr_file, fragP->fr_line, _("long jump required"));
4529 switch (fragP->fr_subtype)
4531 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
4532 extension = 4; /* 1 opcode + 4 displacement */
4533 opcode[0] = 0xe9;
4534 where_to_put_displacement = &opcode[1];
4535 break;
4537 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
4538 extension = 2; /* 1 opcode + 2 displacement */
4539 opcode[0] = 0xe9;
4540 where_to_put_displacement = &opcode[1];
4541 break;
4543 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
4544 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
4545 extension = 5; /* 2 opcode + 4 displacement */
4546 opcode[1] = opcode[0] + 0x10;
4547 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4548 where_to_put_displacement = &opcode[2];
4549 break;
4551 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
4552 extension = 3; /* 2 opcode + 2 displacement */
4553 opcode[1] = opcode[0] + 0x10;
4554 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4555 where_to_put_displacement = &opcode[2];
4556 break;
4558 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
4559 extension = 4;
4560 opcode[0] ^= 1;
4561 opcode[1] = 3;
4562 opcode[2] = 0xe9;
4563 where_to_put_displacement = &opcode[3];
4564 break;
4566 default:
4567 BAD_CASE (fragP->fr_subtype);
4568 break;
4572 /* Now put displacement after opcode. */
4573 md_number_to_chars ((char *) where_to_put_displacement,
4574 (valueT) (displacement_from_opcode_start - extension),
4575 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
4576 fragP->fr_fix += extension;
4579 /* Size of byte displacement jmp. */
4580 int md_short_jump_size = 2;
4582 /* Size of dword displacement jmp. */
4583 int md_long_jump_size = 5;
4585 /* Size of relocation record. */
4586 const int md_reloc_size = 8;
4588 void
4589 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
4590 char *ptr;
4591 addressT from_addr, to_addr;
4592 fragS *frag ATTRIBUTE_UNUSED;
4593 symbolS *to_symbol ATTRIBUTE_UNUSED;
4595 offsetT offset;
4597 offset = to_addr - (from_addr + 2);
4598 /* Opcode for byte-disp jump. */
4599 md_number_to_chars (ptr, (valueT) 0xeb, 1);
4600 md_number_to_chars (ptr + 1, (valueT) offset, 1);
4603 void
4604 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
4605 char *ptr;
4606 addressT from_addr, to_addr;
4607 fragS *frag ATTRIBUTE_UNUSED;
4608 symbolS *to_symbol ATTRIBUTE_UNUSED;
4610 offsetT offset;
4612 offset = to_addr - (from_addr + 5);
4613 md_number_to_chars (ptr, (valueT) 0xe9, 1);
4614 md_number_to_chars (ptr + 1, (valueT) offset, 4);
4617 /* Apply a fixup (fixS) to segment data, once it has been determined
4618 by our caller that we have all the info we need to fix it up.
4620 On the 386, immediates, displacements, and data pointers are all in
4621 the same (little-endian) format, so we don't need to care about which
4622 we are handling. */
4624 void
4625 md_apply_fix3 (fixP, valP, seg)
4626 /* The fix we're to put in. */
4627 fixS *fixP;
4628 /* Pointer to the value of the bits. */
4629 valueT *valP;
4630 /* Segment fix is from. */
4631 segT seg ATTRIBUTE_UNUSED;
4633 char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
4634 valueT value = *valP;
4636 #if !defined (TE_Mach)
4637 if (fixP->fx_pcrel)
4639 switch (fixP->fx_r_type)
4641 default:
4642 break;
4644 case BFD_RELOC_32:
4645 fixP->fx_r_type = BFD_RELOC_32_PCREL;
4646 break;
4647 case BFD_RELOC_16:
4648 fixP->fx_r_type = BFD_RELOC_16_PCREL;
4649 break;
4650 case BFD_RELOC_8:
4651 fixP->fx_r_type = BFD_RELOC_8_PCREL;
4652 break;
4656 if (fixP->fx_addsy != NULL
4657 && (fixP->fx_r_type == BFD_RELOC_32_PCREL
4658 || fixP->fx_r_type == BFD_RELOC_16_PCREL
4659 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
4660 && !use_rela_relocations)
4662 /* This is a hack. There should be a better way to handle this.
4663 This covers for the fact that bfd_install_relocation will
4664 subtract the current location (for partial_inplace, PC relative
4665 relocations); see more below. */
4666 #ifndef OBJ_AOUT
4667 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4668 #ifdef TE_PE
4669 || OUTPUT_FLAVOR == bfd_target_coff_flavour
4670 #endif
4672 value += fixP->fx_where + fixP->fx_frag->fr_address;
4673 #endif
4674 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4675 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
4677 segT sym_seg = S_GET_SEGMENT (fixP->fx_addsy);
4679 if ((sym_seg == seg
4680 || (symbol_section_p (fixP->fx_addsy)
4681 && sym_seg != absolute_section))
4682 && !generic_force_reloc (fixP))
4684 /* Yes, we add the values in twice. This is because
4685 bfd_install_relocation subtracts them out again. I think
4686 bfd_install_relocation is broken, but I don't dare change
4687 it. FIXME. */
4688 value += fixP->fx_where + fixP->fx_frag->fr_address;
4691 #endif
4692 #if defined (OBJ_COFF) && defined (TE_PE)
4693 /* For some reason, the PE format does not store a section
4694 address offset for a PC relative symbol. */
4695 if (S_GET_SEGMENT (fixP->fx_addsy) != seg)
4696 value += md_pcrel_from (fixP);
4697 #endif
4700 /* Fix a few things - the dynamic linker expects certain values here,
4701 and we must not disappoint it. */
4702 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4703 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4704 && fixP->fx_addsy)
4705 switch (fixP->fx_r_type)
4707 case BFD_RELOC_386_PLT32:
4708 case BFD_RELOC_X86_64_PLT32:
4709 /* Make the jump instruction point to the address of the operand. At
4710 runtime we merely add the offset to the actual PLT entry. */
4711 value = -4;
4712 break;
4714 case BFD_RELOC_386_TLS_GD:
4715 case BFD_RELOC_386_TLS_LDM:
4716 case BFD_RELOC_386_TLS_IE_32:
4717 case BFD_RELOC_386_TLS_IE:
4718 case BFD_RELOC_386_TLS_GOTIE:
4719 case BFD_RELOC_X86_64_TLSGD:
4720 case BFD_RELOC_X86_64_TLSLD:
4721 case BFD_RELOC_X86_64_GOTTPOFF:
4722 value = 0; /* Fully resolved at runtime. No addend. */
4723 /* Fallthrough */
4724 case BFD_RELOC_386_TLS_LE:
4725 case BFD_RELOC_386_TLS_LDO_32:
4726 case BFD_RELOC_386_TLS_LE_32:
4727 case BFD_RELOC_X86_64_DTPOFF32:
4728 case BFD_RELOC_X86_64_TPOFF32:
4729 S_SET_THREAD_LOCAL (fixP->fx_addsy);
4730 break;
4732 case BFD_RELOC_386_GOT32:
4733 case BFD_RELOC_X86_64_GOT32:
4734 value = 0; /* Fully resolved at runtime. No addend. */
4735 break;
4737 case BFD_RELOC_VTABLE_INHERIT:
4738 case BFD_RELOC_VTABLE_ENTRY:
4739 fixP->fx_done = 0;
4740 return;
4742 default:
4743 break;
4745 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
4746 *valP = value;
4747 #endif /* !defined (TE_Mach) */
4749 /* Are we finished with this relocation now? */
4750 if (fixP->fx_addsy == NULL)
4751 fixP->fx_done = 1;
4752 else if (use_rela_relocations)
4754 fixP->fx_no_overflow = 1;
4755 /* Remember value for tc_gen_reloc. */
4756 fixP->fx_addnumber = value;
4757 value = 0;
4760 md_number_to_chars (p, value, fixP->fx_size);
4763 #define MAX_LITTLENUMS 6
4765 /* Turn the string pointed to by litP into a floating point constant
4766 of type TYPE, and emit the appropriate bytes. The number of
4767 LITTLENUMS emitted is stored in *SIZEP. An error message is
4768 returned, or NULL on OK. */
4770 char *
4771 md_atof (type, litP, sizeP)
4772 int type;
4773 char *litP;
4774 int *sizeP;
4776 int prec;
4777 LITTLENUM_TYPE words[MAX_LITTLENUMS];
4778 LITTLENUM_TYPE *wordP;
4779 char *t;
4781 switch (type)
4783 case 'f':
4784 case 'F':
4785 prec = 2;
4786 break;
4788 case 'd':
4789 case 'D':
4790 prec = 4;
4791 break;
4793 case 'x':
4794 case 'X':
4795 prec = 5;
4796 break;
4798 default:
4799 *sizeP = 0;
4800 return _("Bad call to md_atof ()");
4802 t = atof_ieee (input_line_pointer, type, words);
4803 if (t)
4804 input_line_pointer = t;
4806 *sizeP = prec * sizeof (LITTLENUM_TYPE);
4807 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
4808 the bigendian 386. */
4809 for (wordP = words + prec - 1; prec--;)
4811 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
4812 litP += sizeof (LITTLENUM_TYPE);
4814 return 0;
4817 char output_invalid_buf[8];
4819 static char *
4820 output_invalid (c)
4821 int c;
4823 if (ISPRINT (c))
4824 sprintf (output_invalid_buf, "'%c'", c);
4825 else
4826 sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
4827 return output_invalid_buf;
4830 /* REG_STRING starts *before* REGISTER_PREFIX. */
4832 static const reg_entry *
4833 parse_register (reg_string, end_op)
4834 char *reg_string;
4835 char **end_op;
4837 char *s = reg_string;
4838 char *p;
4839 char reg_name_given[MAX_REG_NAME_SIZE + 1];
4840 const reg_entry *r;
4842 /* Skip possible REGISTER_PREFIX and possible whitespace. */
4843 if (*s == REGISTER_PREFIX)
4844 ++s;
4846 if (is_space_char (*s))
4847 ++s;
4849 p = reg_name_given;
4850 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
4852 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
4853 return (const reg_entry *) NULL;
4854 s++;
4857 /* For naked regs, make sure that we are not dealing with an identifier.
4858 This prevents confusing an identifier like `eax_var' with register
4859 `eax'. */
4860 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
4861 return (const reg_entry *) NULL;
4863 *end_op = s;
4865 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
4867 /* Handle floating point regs, allowing spaces in the (i) part. */
4868 if (r == i386_regtab /* %st is first entry of table */)
4870 if (is_space_char (*s))
4871 ++s;
4872 if (*s == '(')
4874 ++s;
4875 if (is_space_char (*s))
4876 ++s;
4877 if (*s >= '0' && *s <= '7')
4879 r = &i386_float_regtab[*s - '0'];
4880 ++s;
4881 if (is_space_char (*s))
4882 ++s;
4883 if (*s == ')')
4885 *end_op = s + 1;
4886 return r;
4889 /* We have "%st(" then garbage. */
4890 return (const reg_entry *) NULL;
4894 if (r != NULL
4895 && ((r->reg_flags & (RegRex64 | RegRex)) | (r->reg_type & Reg64)) != 0
4896 && flag_code != CODE_64BIT)
4897 return (const reg_entry *) NULL;
4899 return r;
4902 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4903 const char *md_shortopts = "kVQ:sqn";
4904 #else
4905 const char *md_shortopts = "qn";
4906 #endif
4908 struct option md_longopts[] = {
4909 #define OPTION_32 (OPTION_MD_BASE + 0)
4910 {"32", no_argument, NULL, OPTION_32},
4911 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4912 #define OPTION_64 (OPTION_MD_BASE + 1)
4913 {"64", no_argument, NULL, OPTION_64},
4914 #endif
4915 {NULL, no_argument, NULL, 0}
4917 size_t md_longopts_size = sizeof (md_longopts);
4920 md_parse_option (c, arg)
4921 int c;
4922 char *arg ATTRIBUTE_UNUSED;
4924 switch (c)
4926 case 'n':
4927 optimize_align_code = 0;
4928 break;
4930 case 'q':
4931 quiet_warnings = 1;
4932 break;
4934 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4935 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
4936 should be emitted or not. FIXME: Not implemented. */
4937 case 'Q':
4938 break;
4940 /* -V: SVR4 argument to print version ID. */
4941 case 'V':
4942 print_version_id ();
4943 break;
4945 /* -k: Ignore for FreeBSD compatibility. */
4946 case 'k':
4947 break;
4949 case 's':
4950 /* -s: On i386 Solaris, this tells the native assembler to use
4951 .stab instead of .stab.excl. We always use .stab anyhow. */
4952 break;
4954 case OPTION_64:
4956 const char **list, **l;
4958 list = bfd_target_list ();
4959 for (l = list; *l != NULL; l++)
4960 if (strcmp (*l, "elf64-x86-64") == 0)
4962 default_arch = "x86_64";
4963 break;
4965 if (*l == NULL)
4966 as_fatal (_("No compiled in support for x86_64"));
4967 free (list);
4969 break;
4970 #endif
4972 case OPTION_32:
4973 default_arch = "i386";
4974 break;
4976 default:
4977 return 0;
4979 return 1;
4982 void
4983 md_show_usage (stream)
4984 FILE *stream;
4986 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4987 fprintf (stream, _("\
4988 -Q ignored\n\
4989 -V print assembler version number\n\
4990 -k ignored\n\
4991 -n Do not optimize code alignment\n\
4992 -q quieten some warnings\n\
4993 -s ignored\n"));
4994 #else
4995 fprintf (stream, _("\
4996 -n Do not optimize code alignment\n\
4997 -q quieten some warnings\n"));
4998 #endif
5001 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
5002 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
5004 /* Pick the target format to use. */
5006 const char *
5007 i386_target_format ()
5009 if (!strcmp (default_arch, "x86_64"))
5010 set_code_flag (CODE_64BIT);
5011 else if (!strcmp (default_arch, "i386"))
5012 set_code_flag (CODE_32BIT);
5013 else
5014 as_fatal (_("Unknown architecture"));
5015 switch (OUTPUT_FLAVOR)
5017 #ifdef OBJ_MAYBE_AOUT
5018 case bfd_target_aout_flavour:
5019 return AOUT_TARGET_FORMAT;
5020 #endif
5021 #ifdef OBJ_MAYBE_COFF
5022 case bfd_target_coff_flavour:
5023 return "coff-i386";
5024 #endif
5025 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
5026 case bfd_target_elf_flavour:
5028 if (flag_code == CODE_64BIT)
5029 use_rela_relocations = 1;
5030 return flag_code == CODE_64BIT ? "elf64-x86-64" : ELF_TARGET_FORMAT;
5032 #endif
5033 default:
5034 abort ();
5035 return NULL;
5039 #endif /* OBJ_MAYBE_ more than one */
5041 #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
5042 void i386_elf_emit_arch_note ()
5044 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
5045 && cpu_arch_name != NULL)
5047 char *p;
5048 asection *seg = now_seg;
5049 subsegT subseg = now_subseg;
5050 Elf_Internal_Note i_note;
5051 Elf_External_Note e_note;
5052 asection *note_secp;
5053 int len;
5055 /* Create the .note section. */
5056 note_secp = subseg_new (".note", 0);
5057 bfd_set_section_flags (stdoutput,
5058 note_secp,
5059 SEC_HAS_CONTENTS | SEC_READONLY);
5061 /* Process the arch string. */
5062 len = strlen (cpu_arch_name);
5064 i_note.namesz = len + 1;
5065 i_note.descsz = 0;
5066 i_note.type = NT_ARCH;
5067 p = frag_more (sizeof (e_note.namesz));
5068 md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
5069 p = frag_more (sizeof (e_note.descsz));
5070 md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
5071 p = frag_more (sizeof (e_note.type));
5072 md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
5073 p = frag_more (len + 1);
5074 strcpy (p, cpu_arch_name);
5076 frag_align (2, 0, 0);
5078 subseg_set (seg, subseg);
5081 #endif
5083 symbolS *
5084 md_undefined_symbol (name)
5085 char *name;
5087 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
5088 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
5089 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
5090 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
5092 if (!GOT_symbol)
5094 if (symbol_find (name))
5095 as_bad (_("GOT already in symbol table"));
5096 GOT_symbol = symbol_new (name, undefined_section,
5097 (valueT) 0, &zero_address_frag);
5099 return GOT_symbol;
5101 return 0;
5104 /* Round up a section size to the appropriate boundary. */
5106 valueT
5107 md_section_align (segment, size)
5108 segT segment ATTRIBUTE_UNUSED;
5109 valueT size;
5111 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
5112 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
5114 /* For a.out, force the section size to be aligned. If we don't do
5115 this, BFD will align it for us, but it will not write out the
5116 final bytes of the section. This may be a bug in BFD, but it is
5117 easier to fix it here since that is how the other a.out targets
5118 work. */
5119 int align;
5121 align = bfd_get_section_alignment (stdoutput, segment);
5122 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
5124 #endif
5126 return size;
5129 /* On the i386, PC-relative offsets are relative to the start of the
5130 next instruction. That is, the address of the offset, plus its
5131 size, since the offset is always the last part of the insn. */
5133 long
5134 md_pcrel_from (fixP)
5135 fixS *fixP;
5137 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
5140 #ifndef I386COFF
5142 static void
5143 s_bss (ignore)
5144 int ignore ATTRIBUTE_UNUSED;
5146 int temp;
5148 temp = get_absolute_expression ();
5149 subseg_set (bss_section, (subsegT) temp);
5150 demand_empty_rest_of_line ();
5153 #endif
5155 void
5156 i386_validate_fix (fixp)
5157 fixS *fixp;
5159 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
5161 /* GOTOFF relocation are nonsense in 64bit mode. */
5162 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
5164 if (flag_code != CODE_64BIT)
5165 abort ();
5166 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
5168 else
5170 if (flag_code == CODE_64BIT)
5171 abort ();
5172 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
5174 fixp->fx_subsy = 0;
5178 arelent *
5179 tc_gen_reloc (section, fixp)
5180 asection *section ATTRIBUTE_UNUSED;
5181 fixS *fixp;
5183 arelent *rel;
5184 bfd_reloc_code_real_type code;
5186 switch (fixp->fx_r_type)
5188 case BFD_RELOC_X86_64_PLT32:
5189 case BFD_RELOC_X86_64_GOT32:
5190 case BFD_RELOC_X86_64_GOTPCREL:
5191 case BFD_RELOC_386_PLT32:
5192 case BFD_RELOC_386_GOT32:
5193 case BFD_RELOC_386_GOTOFF:
5194 case BFD_RELOC_386_GOTPC:
5195 case BFD_RELOC_386_TLS_GD:
5196 case BFD_RELOC_386_TLS_LDM:
5197 case BFD_RELOC_386_TLS_LDO_32:
5198 case BFD_RELOC_386_TLS_IE_32:
5199 case BFD_RELOC_386_TLS_IE:
5200 case BFD_RELOC_386_TLS_GOTIE:
5201 case BFD_RELOC_386_TLS_LE_32:
5202 case BFD_RELOC_386_TLS_LE:
5203 case BFD_RELOC_X86_64_32S:
5204 case BFD_RELOC_X86_64_TLSGD:
5205 case BFD_RELOC_X86_64_TLSLD:
5206 case BFD_RELOC_X86_64_DTPOFF32:
5207 case BFD_RELOC_X86_64_GOTTPOFF:
5208 case BFD_RELOC_X86_64_TPOFF32:
5209 case BFD_RELOC_RVA:
5210 case BFD_RELOC_VTABLE_ENTRY:
5211 case BFD_RELOC_VTABLE_INHERIT:
5212 #ifdef TE_PE
5213 case BFD_RELOC_32_SECREL:
5214 #endif
5215 code = fixp->fx_r_type;
5216 break;
5217 default:
5218 if (fixp->fx_pcrel)
5220 switch (fixp->fx_size)
5222 default:
5223 as_bad_where (fixp->fx_file, fixp->fx_line,
5224 _("can not do %d byte pc-relative relocation"),
5225 fixp->fx_size);
5226 code = BFD_RELOC_32_PCREL;
5227 break;
5228 case 1: code = BFD_RELOC_8_PCREL; break;
5229 case 2: code = BFD_RELOC_16_PCREL; break;
5230 case 4: code = BFD_RELOC_32_PCREL; break;
5233 else
5235 switch (fixp->fx_size)
5237 default:
5238 as_bad_where (fixp->fx_file, fixp->fx_line,
5239 _("can not do %d byte relocation"),
5240 fixp->fx_size);
5241 code = BFD_RELOC_32;
5242 break;
5243 case 1: code = BFD_RELOC_8; break;
5244 case 2: code = BFD_RELOC_16; break;
5245 case 4: code = BFD_RELOC_32; break;
5246 #ifdef BFD64
5247 case 8: code = BFD_RELOC_64; break;
5248 #endif
5251 break;
5254 if (code == BFD_RELOC_32
5255 && GOT_symbol
5256 && fixp->fx_addsy == GOT_symbol)
5258 /* We don't support GOTPC on 64bit targets. */
5259 if (flag_code == CODE_64BIT)
5260 abort ();
5261 code = BFD_RELOC_386_GOTPC;
5264 rel = (arelent *) xmalloc (sizeof (arelent));
5265 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
5266 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
5268 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
5270 #ifdef TE_PE
5271 if (S_IS_WEAK (fixp->fx_addsy))
5272 rel->addend = rel->address - (*rel->sym_ptr_ptr)->value + 4;
5273 else
5274 #endif
5275 if (!use_rela_relocations)
5277 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
5278 vtable entry to be used in the relocation's section offset. */
5279 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
5280 rel->address = fixp->fx_offset;
5282 rel->addend = 0;
5284 /* Use the rela in 64bit mode. */
5285 else
5287 if (!fixp->fx_pcrel)
5288 rel->addend = fixp->fx_offset;
5289 else
5290 switch (code)
5292 case BFD_RELOC_X86_64_PLT32:
5293 case BFD_RELOC_X86_64_GOT32:
5294 case BFD_RELOC_X86_64_GOTPCREL:
5295 case BFD_RELOC_X86_64_TLSGD:
5296 case BFD_RELOC_X86_64_TLSLD:
5297 case BFD_RELOC_X86_64_GOTTPOFF:
5298 rel->addend = fixp->fx_offset - fixp->fx_size;
5299 break;
5300 default:
5301 rel->addend = (section->vma
5302 - fixp->fx_size
5303 + fixp->fx_addnumber
5304 + md_pcrel_from (fixp));
5305 break;
5309 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
5310 if (rel->howto == NULL)
5312 as_bad_where (fixp->fx_file, fixp->fx_line,
5313 _("cannot represent relocation type %s"),
5314 bfd_get_reloc_code_name (code));
5315 /* Set howto to a garbage value so that we can keep going. */
5316 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
5317 assert (rel->howto != NULL);
5320 return rel;
5324 /* Parse operands using Intel syntax. This implements a recursive descent
5325 parser based on the BNF grammar published in Appendix B of the MASM 6.1
5326 Programmer's Guide.
5328 FIXME: We do not recognize the full operand grammar defined in the MASM
5329 documentation. In particular, all the structure/union and
5330 high-level macro operands are missing.
5332 Uppercase words are terminals, lower case words are non-terminals.
5333 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
5334 bars '|' denote choices. Most grammar productions are implemented in
5335 functions called 'intel_<production>'.
5337 Initial production is 'expr'.
5339 addOp + | - | & | \| | << | >>
5341 alpha [a-zA-Z]
5343 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
5345 constant digits [[ radixOverride ]]
5347 dataType BYTE | WORD | DWORD | QWORD | XWORD
5349 digits decdigit
5350 | digits decdigit
5351 | digits hexdigit
5353 decdigit [0-9]
5355 e05 e05 addOp e06
5356 | e06
5358 e06 e06 mulOp e09
5359 | e09
5361 e09 OFFSET e10
5362 | e09 PTR e10
5363 | e09 : e10
5364 | e10
5366 e10 e10 [ expr ]
5367 | e11
5369 e11 ( expr )
5370 | [ expr ]
5371 | constant
5372 | dataType
5373 | id
5375 | register
5378 => expr SHORT e05
5379 | e05
5381 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
5382 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
5384 hexdigit a | b | c | d | e | f
5385 | A | B | C | D | E | F
5387 id alpha
5388 | id alpha
5389 | id decdigit
5391 mulOp * | / | MOD
5393 quote " | '
5395 register specialRegister
5396 | gpRegister
5397 | byteRegister
5399 segmentRegister CS | DS | ES | FS | GS | SS
5401 specialRegister CR0 | CR2 | CR3
5402 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
5403 | TR3 | TR4 | TR5 | TR6 | TR7
5405 We simplify the grammar in obvious places (e.g., register parsing is
5406 done by calling parse_register) and eliminate immediate left recursion
5407 to implement a recursive-descent parser.
5409 expr SHORT e05
5410 | e05
5412 e05 e06 e05'
5414 e05' addOp e06 e05'
5415 | Empty
5417 e06 e09 e06'
5419 e06' mulOp e09 e06'
5420 | Empty
5422 e09 OFFSET e10 e09'
5423 | e10 e09'
5425 e09' PTR e10 e09'
5426 | : e10 e09'
5427 | Empty
5429 e10 e11 e10'
5431 e10' [ expr ] e10'
5432 | Empty
5434 e11 ( expr )
5435 | [ expr ]
5436 | BYTE
5437 | WORD
5438 | DWORD
5439 | QWORD
5440 | XWORD
5443 | register
5444 | id
5446 | constant */
5448 /* Parsing structure for the intel syntax parser. Used to implement the
5449 semantic actions for the operand grammar. */
5450 struct intel_parser_s
5452 char *op_string; /* The string being parsed. */
5453 int got_a_float; /* Whether the operand is a float. */
5454 int op_modifier; /* Operand modifier. */
5455 int is_mem; /* 1 if operand is memory reference. */
5456 const reg_entry *reg; /* Last register reference found. */
5457 char *disp; /* Displacement string being built. */
5460 static struct intel_parser_s intel_parser;
5462 /* Token structure for parsing intel syntax. */
5463 struct intel_token
5465 int code; /* Token code. */
5466 const reg_entry *reg; /* Register entry for register tokens. */
5467 char *str; /* String representation. */
5470 static struct intel_token cur_token, prev_token;
5472 /* Token codes for the intel parser. Since T_SHORT is already used
5473 by COFF, undefine it first to prevent a warning. */
5474 #define T_NIL -1
5475 #define T_CONST 1
5476 #define T_REG 2
5477 #define T_BYTE 3
5478 #define T_WORD 4
5479 #define T_DWORD 5
5480 #define T_QWORD 6
5481 #define T_XWORD 7
5482 #undef T_SHORT
5483 #define T_SHORT 8
5484 #define T_OFFSET 9
5485 #define T_PTR 10
5486 #define T_ID 11
5487 #define T_SHIFTOP 12
5489 /* Prototypes for intel parser functions. */
5490 static int intel_match_token PARAMS ((int code));
5491 static void intel_get_token PARAMS ((void));
5492 static void intel_putback_token PARAMS ((void));
5493 static int intel_expr PARAMS ((void));
5494 static int intel_e05 PARAMS ((void));
5495 static int intel_e05_1 PARAMS ((void));
5496 static int intel_e06 PARAMS ((void));
5497 static int intel_e06_1 PARAMS ((void));
5498 static int intel_e09 PARAMS ((void));
5499 static int intel_e09_1 PARAMS ((void));
5500 static int intel_e10 PARAMS ((void));
5501 static int intel_e10_1 PARAMS ((void));
5502 static int intel_e11 PARAMS ((void));
5504 static int
5505 i386_intel_operand (operand_string, got_a_float)
5506 char *operand_string;
5507 int got_a_float;
5509 int ret;
5510 char *p;
5512 /* Initialize token holders. */
5513 cur_token.code = prev_token.code = T_NIL;
5514 cur_token.reg = prev_token.reg = NULL;
5515 cur_token.str = prev_token.str = NULL;
5517 /* Initialize parser structure. */
5518 p = intel_parser.op_string = (char *) malloc (strlen (operand_string) + 1);
5519 if (p == NULL)
5520 abort ();
5521 strcpy (intel_parser.op_string, operand_string);
5522 intel_parser.got_a_float = got_a_float;
5523 intel_parser.op_modifier = -1;
5524 intel_parser.is_mem = 0;
5525 intel_parser.reg = NULL;
5526 intel_parser.disp = (char *) malloc (strlen (operand_string) + 1);
5527 if (intel_parser.disp == NULL)
5528 abort ();
5529 intel_parser.disp[0] = '\0';
5531 /* Read the first token and start the parser. */
5532 intel_get_token ();
5533 ret = intel_expr ();
5535 if (ret)
5537 /* If we found a memory reference, hand it over to i386_displacement
5538 to fill in the rest of the operand fields. */
5539 if (intel_parser.is_mem)
5541 if ((i.mem_operands == 1
5542 && (current_templates->start->opcode_modifier & IsString) == 0)
5543 || i.mem_operands == 2)
5545 as_bad (_("too many memory references for '%s'"),
5546 current_templates->start->name);
5547 ret = 0;
5549 else
5551 char *s = intel_parser.disp;
5552 i.mem_operands++;
5554 /* Add the displacement expression. */
5555 if (*s != '\0')
5556 ret = i386_displacement (s, s + strlen (s));
5557 if (ret)
5558 ret = i386_index_check (operand_string);
5562 /* Constant and OFFSET expressions are handled by i386_immediate. */
5563 else if (intel_parser.op_modifier == OFFSET_FLAT
5564 || intel_parser.reg == NULL)
5565 ret = i386_immediate (intel_parser.disp);
5568 free (p);
5569 free (intel_parser.disp);
5571 return ret;
5574 /* expr SHORT e05
5575 | e05 */
5576 static int
5577 intel_expr ()
5579 /* expr SHORT e05 */
5580 if (cur_token.code == T_SHORT)
5582 intel_parser.op_modifier = SHORT;
5583 intel_match_token (T_SHORT);
5585 return (intel_e05 ());
5588 /* expr e05 */
5589 else
5590 return intel_e05 ();
5593 /* e05 e06 e05'
5595 e05' addOp e06 e05'
5596 | Empty */
5597 static int
5598 intel_e05 ()
5600 return (intel_e06 () && intel_e05_1 ());
5603 static int
5604 intel_e05_1 ()
5606 /* e05' addOp e06 e05' */
5607 if (cur_token.code == '+' || cur_token.code == '-'
5608 || cur_token.code == '&' || cur_token.code == '|'
5609 || cur_token.code == T_SHIFTOP)
5611 strcat (intel_parser.disp, cur_token.str);
5612 intel_match_token (cur_token.code);
5614 return (intel_e06 () && intel_e05_1 ());
5617 /* e05' Empty */
5618 else
5619 return 1;
5622 /* e06 e09 e06'
5624 e06' mulOp e09 e06'
5625 | Empty */
5626 static int
5627 intel_e06 ()
5629 return (intel_e09 () && intel_e06_1 ());
5632 static int
5633 intel_e06_1 ()
5635 /* e06' mulOp e09 e06' */
5636 if (cur_token.code == '*' || cur_token.code == '/')
5638 strcat (intel_parser.disp, cur_token.str);
5639 intel_match_token (cur_token.code);
5641 return (intel_e09 () && intel_e06_1 ());
5644 /* e06' Empty */
5645 else
5646 return 1;
5649 /* e09 OFFSET e10 e09'
5650 | e10 e09'
5652 e09' PTR e10 e09'
5653 | : e10 e09'
5654 | Empty */
5655 static int
5656 intel_e09 ()
5658 /* e09 OFFSET e10 e09' */
5659 if (cur_token.code == T_OFFSET)
5661 intel_parser.is_mem = 0;
5662 intel_parser.op_modifier = OFFSET_FLAT;
5663 intel_match_token (T_OFFSET);
5665 return (intel_e10 () && intel_e09_1 ());
5668 /* e09 e10 e09' */
5669 else
5670 return (intel_e10 () && intel_e09_1 ());
5673 static int
5674 intel_e09_1 ()
5676 /* e09' PTR e10 e09' */
5677 if (cur_token.code == T_PTR)
5679 if (prev_token.code == T_BYTE)
5680 i.suffix = BYTE_MNEM_SUFFIX;
5682 else if (prev_token.code == T_WORD)
5684 if (intel_parser.got_a_float == 2) /* "fi..." */
5685 i.suffix = SHORT_MNEM_SUFFIX;
5686 else
5687 i.suffix = WORD_MNEM_SUFFIX;
5690 else if (prev_token.code == T_DWORD)
5692 if (intel_parser.got_a_float == 1) /* "f..." */
5693 i.suffix = SHORT_MNEM_SUFFIX;
5694 else
5695 i.suffix = LONG_MNEM_SUFFIX;
5698 else if (prev_token.code == T_QWORD)
5700 if (intel_parser.got_a_float == 1) /* "f..." */
5701 i.suffix = LONG_MNEM_SUFFIX;
5702 else
5703 i.suffix = QWORD_MNEM_SUFFIX;
5706 else if (prev_token.code == T_XWORD)
5707 i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
5709 else
5711 as_bad (_("Unknown operand modifier `%s'"), prev_token.str);
5712 return 0;
5715 intel_match_token (T_PTR);
5717 return (intel_e10 () && intel_e09_1 ());
5720 /* e09 : e10 e09' */
5721 else if (cur_token.code == ':')
5723 /* Mark as a memory operand only if it's not already known to be an
5724 offset expression. */
5725 if (intel_parser.op_modifier != OFFSET_FLAT)
5726 intel_parser.is_mem = 1;
5728 return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
5731 /* e09' Empty */
5732 else
5733 return 1;
5736 /* e10 e11 e10'
5738 e10' [ expr ] e10'
5739 | Empty */
5740 static int
5741 intel_e10 ()
5743 return (intel_e11 () && intel_e10_1 ());
5746 static int
5747 intel_e10_1 ()
5749 /* e10' [ expr ] e10' */
5750 if (cur_token.code == '[')
5752 intel_match_token ('[');
5754 /* Mark as a memory operand only if it's not already known to be an
5755 offset expression. If it's an offset expression, we need to keep
5756 the brace in. */
5757 if (intel_parser.op_modifier != OFFSET_FLAT)
5758 intel_parser.is_mem = 1;
5759 else
5760 strcat (intel_parser.disp, "[");
5762 /* Add a '+' to the displacement string if necessary. */
5763 if (*intel_parser.disp != '\0'
5764 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5765 strcat (intel_parser.disp, "+");
5767 if (intel_expr () && intel_match_token (']'))
5769 /* Preserve brackets when the operand is an offset expression. */
5770 if (intel_parser.op_modifier == OFFSET_FLAT)
5771 strcat (intel_parser.disp, "]");
5773 return intel_e10_1 ();
5775 else
5776 return 0;
5779 /* e10' Empty */
5780 else
5781 return 1;
5784 /* e11 ( expr )
5785 | [ expr ]
5786 | BYTE
5787 | WORD
5788 | DWORD
5789 | QWORD
5790 | XWORD
5793 | register
5794 | id
5796 | constant */
5797 static int
5798 intel_e11 ()
5800 /* e11 ( expr ) */
5801 if (cur_token.code == '(')
5803 intel_match_token ('(');
5804 strcat (intel_parser.disp, "(");
5806 if (intel_expr () && intel_match_token (')'))
5808 strcat (intel_parser.disp, ")");
5809 return 1;
5811 else
5812 return 0;
5815 /* e11 ~ expr */
5816 else if (cur_token.code == '~')
5818 strcat (intel_parser.disp, "~");
5819 intel_match_token ('~');
5821 return (intel_e11 ());
5824 /* e11 [ expr ] */
5825 else if (cur_token.code == '[')
5827 intel_match_token ('[');
5829 /* Mark as a memory operand only if it's not already known to be an
5830 offset expression. If it's an offset expression, we need to keep
5831 the brace in. */
5832 if (intel_parser.op_modifier != OFFSET_FLAT)
5833 intel_parser.is_mem = 1;
5834 else
5835 strcat (intel_parser.disp, "[");
5837 /* Operands for jump/call inside brackets denote absolute addresses. */
5838 if (current_templates->start->opcode_modifier & Jump
5839 || current_templates->start->opcode_modifier & JumpDword
5840 || current_templates->start->opcode_modifier & JumpByte
5841 || current_templates->start->opcode_modifier & JumpInterSegment)
5842 i.types[this_operand] |= JumpAbsolute;
5844 /* Add a '+' to the displacement string if necessary. */
5845 if (*intel_parser.disp != '\0'
5846 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5847 strcat (intel_parser.disp, "+");
5849 if (intel_expr () && intel_match_token (']'))
5851 /* Preserve brackets when the operand is an offset expression. */
5852 if (intel_parser.op_modifier == OFFSET_FLAT)
5853 strcat (intel_parser.disp, "]");
5855 return 1;
5857 else
5858 return 0;
5861 /* e11 BYTE
5862 | WORD
5863 | DWORD
5864 | QWORD
5865 | XWORD */
5866 else if (cur_token.code == T_BYTE
5867 || cur_token.code == T_WORD
5868 || cur_token.code == T_DWORD
5869 || cur_token.code == T_QWORD
5870 || cur_token.code == T_XWORD)
5872 intel_match_token (cur_token.code);
5874 return 1;
5877 /* e11 $
5878 | . */
5879 else if (cur_token.code == '$' || cur_token.code == '.')
5881 strcat (intel_parser.disp, cur_token.str);
5882 intel_match_token (cur_token.code);
5884 /* Mark as a memory operand only if it's not already known to be an
5885 offset expression. */
5886 if (intel_parser.op_modifier != OFFSET_FLAT)
5887 intel_parser.is_mem = 1;
5889 return 1;
5892 /* e11 register */
5893 else if (cur_token.code == T_REG)
5895 const reg_entry *reg = intel_parser.reg = cur_token.reg;
5897 intel_match_token (T_REG);
5899 /* Check for segment change. */
5900 if (cur_token.code == ':')
5902 if (reg->reg_type & (SReg2 | SReg3))
5904 switch (reg->reg_num)
5906 case 0:
5907 i.seg[i.mem_operands] = &es;
5908 break;
5909 case 1:
5910 i.seg[i.mem_operands] = &cs;
5911 break;
5912 case 2:
5913 i.seg[i.mem_operands] = &ss;
5914 break;
5915 case 3:
5916 i.seg[i.mem_operands] = &ds;
5917 break;
5918 case 4:
5919 i.seg[i.mem_operands] = &fs;
5920 break;
5921 case 5:
5922 i.seg[i.mem_operands] = &gs;
5923 break;
5926 else
5928 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
5929 return 0;
5933 /* Not a segment register. Check for register scaling. */
5934 else if (cur_token.code == '*')
5936 if (!intel_parser.is_mem)
5938 as_bad (_("Register scaling only allowed in memory operands."));
5939 return 0;
5942 /* What follows must be a valid scale. */
5943 if (intel_match_token ('*')
5944 && strchr ("01248", *cur_token.str))
5946 i.index_reg = reg;
5947 i.types[this_operand] |= BaseIndex;
5949 /* Set the scale after setting the register (otherwise,
5950 i386_scale will complain) */
5951 i386_scale (cur_token.str);
5952 intel_match_token (T_CONST);
5954 else
5956 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
5957 cur_token.str);
5958 return 0;
5962 /* No scaling. If this is a memory operand, the register is either a
5963 base register (first occurrence) or an index register (second
5964 occurrence). */
5965 else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3)))
5967 if (i.base_reg && i.index_reg)
5969 as_bad (_("Too many register references in memory operand."));
5970 return 0;
5973 if (i.base_reg == NULL)
5974 i.base_reg = reg;
5975 else
5976 i.index_reg = reg;
5978 i.types[this_operand] |= BaseIndex;
5981 /* Offset modifier. Add the register to the displacement string to be
5982 parsed as an immediate expression after we're done. */
5983 else if (intel_parser.op_modifier == OFFSET_FLAT)
5984 strcat (intel_parser.disp, reg->reg_name);
5986 /* It's neither base nor index nor offset. */
5987 else
5989 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
5990 i.op[this_operand].regs = reg;
5991 i.reg_operands++;
5994 /* Since registers are not part of the displacement string (except
5995 when we're parsing offset operands), we may need to remove any
5996 preceding '+' from the displacement string. */
5997 if (*intel_parser.disp != '\0'
5998 && intel_parser.op_modifier != OFFSET_FLAT)
6000 char *s = intel_parser.disp;
6001 s += strlen (s) - 1;
6002 if (*s == '+')
6003 *s = '\0';
6006 return 1;
6009 /* e11 id */
6010 else if (cur_token.code == T_ID)
6012 /* Add the identifier to the displacement string. */
6013 strcat (intel_parser.disp, cur_token.str);
6014 intel_match_token (T_ID);
6016 /* The identifier represents a memory reference only if it's not
6017 preceded by an offset modifier. */
6018 if (intel_parser.op_modifier != OFFSET_FLAT)
6019 intel_parser.is_mem = 1;
6021 return 1;
6024 /* e11 constant */
6025 else if (cur_token.code == T_CONST
6026 || cur_token.code == '-'
6027 || cur_token.code == '+')
6029 char *save_str;
6031 /* Allow constants that start with `+' or `-'. */
6032 if (cur_token.code == '-' || cur_token.code == '+')
6034 strcat (intel_parser.disp, cur_token.str);
6035 intel_match_token (cur_token.code);
6036 if (cur_token.code != T_CONST)
6038 as_bad (_("Syntax error. Expecting a constant. Got `%s'."),
6039 cur_token.str);
6040 return 0;
6044 save_str = (char *) malloc (strlen (cur_token.str) + 1);
6045 if (save_str == NULL)
6046 abort ();
6047 strcpy (save_str, cur_token.str);
6049 /* Get the next token to check for register scaling. */
6050 intel_match_token (cur_token.code);
6052 /* Check if this constant is a scaling factor for an index register. */
6053 if (cur_token.code == '*')
6055 if (intel_match_token ('*') && cur_token.code == T_REG)
6057 if (!intel_parser.is_mem)
6059 as_bad (_("Register scaling only allowed in memory operands."));
6060 return 0;
6063 /* The constant is followed by `* reg', so it must be
6064 a valid scale. */
6065 if (strchr ("01248", *save_str))
6067 i.index_reg = cur_token.reg;
6068 i.types[this_operand] |= BaseIndex;
6070 /* Set the scale after setting the register (otherwise,
6071 i386_scale will complain) */
6072 i386_scale (save_str);
6073 intel_match_token (T_REG);
6075 /* Since registers are not part of the displacement
6076 string, we may need to remove any preceding '+' from
6077 the displacement string. */
6078 if (*intel_parser.disp != '\0')
6080 char *s = intel_parser.disp;
6081 s += strlen (s) - 1;
6082 if (*s == '+')
6083 *s = '\0';
6086 free (save_str);
6088 return 1;
6090 else
6091 return 0;
6094 /* The constant was not used for register scaling. Since we have
6095 already consumed the token following `*' we now need to put it
6096 back in the stream. */
6097 else
6098 intel_putback_token ();
6101 /* Add the constant to the displacement string. */
6102 strcat (intel_parser.disp, save_str);
6103 free (save_str);
6105 return 1;
6108 as_bad (_("Unrecognized token '%s'"), cur_token.str);
6109 return 0;
6112 /* Match the given token against cur_token. If they match, read the next
6113 token from the operand string. */
6114 static int
6115 intel_match_token (code)
6116 int code;
6118 if (cur_token.code == code)
6120 intel_get_token ();
6121 return 1;
6123 else
6125 as_bad (_("Unexpected token `%s'"), cur_token.str);
6126 return 0;
6130 /* Read a new token from intel_parser.op_string and store it in cur_token. */
6131 static void
6132 intel_get_token ()
6134 char *end_op;
6135 const reg_entry *reg;
6136 struct intel_token new_token;
6138 new_token.code = T_NIL;
6139 new_token.reg = NULL;
6140 new_token.str = NULL;
6142 /* Free the memory allocated to the previous token and move
6143 cur_token to prev_token. */
6144 if (prev_token.str)
6145 free (prev_token.str);
6147 prev_token = cur_token;
6149 /* Skip whitespace. */
6150 while (is_space_char (*intel_parser.op_string))
6151 intel_parser.op_string++;
6153 /* Return an empty token if we find nothing else on the line. */
6154 if (*intel_parser.op_string == '\0')
6156 cur_token = new_token;
6157 return;
6160 /* The new token cannot be larger than the remainder of the operand
6161 string. */
6162 new_token.str = (char *) malloc (strlen (intel_parser.op_string) + 1);
6163 if (new_token.str == NULL)
6164 abort ();
6165 new_token.str[0] = '\0';
6167 if (strchr ("0123456789", *intel_parser.op_string))
6169 char *p = new_token.str;
6170 char *q = intel_parser.op_string;
6171 new_token.code = T_CONST;
6173 /* Allow any kind of identifier char to encompass floating point and
6174 hexadecimal numbers. */
6175 while (is_identifier_char (*q))
6176 *p++ = *q++;
6177 *p = '\0';
6179 /* Recognize special symbol names [0-9][bf]. */
6180 if (strlen (intel_parser.op_string) == 2
6181 && (intel_parser.op_string[1] == 'b'
6182 || intel_parser.op_string[1] == 'f'))
6183 new_token.code = T_ID;
6186 else if (strchr ("<>", *intel_parser.op_string)
6187 && *intel_parser.op_string == *(intel_parser.op_string + 1))
6189 new_token.code = T_SHIFTOP;
6190 new_token.str[0] = *intel_parser.op_string;
6191 new_token.str[1] = *intel_parser.op_string;
6192 new_token.str[2] = '\0';
6195 else if (strchr ("+-/*&|:[]()~", *intel_parser.op_string))
6197 new_token.code = *intel_parser.op_string;
6198 new_token.str[0] = *intel_parser.op_string;
6199 new_token.str[1] = '\0';
6202 else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
6203 && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
6205 new_token.code = T_REG;
6206 new_token.reg = reg;
6208 if (*intel_parser.op_string == REGISTER_PREFIX)
6210 new_token.str[0] = REGISTER_PREFIX;
6211 new_token.str[1] = '\0';
6214 strcat (new_token.str, reg->reg_name);
6217 else if (is_identifier_char (*intel_parser.op_string))
6219 char *p = new_token.str;
6220 char *q = intel_parser.op_string;
6222 /* A '.' or '$' followed by an identifier char is an identifier.
6223 Otherwise, it's operator '.' followed by an expression. */
6224 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
6226 new_token.code = *q;
6227 new_token.str[0] = *q;
6228 new_token.str[1] = '\0';
6230 else
6232 while (is_identifier_char (*q) || *q == '@')
6233 *p++ = *q++;
6234 *p = '\0';
6236 if (strcasecmp (new_token.str, "BYTE") == 0)
6237 new_token.code = T_BYTE;
6239 else if (strcasecmp (new_token.str, "WORD") == 0)
6240 new_token.code = T_WORD;
6242 else if (strcasecmp (new_token.str, "DWORD") == 0)
6243 new_token.code = T_DWORD;
6245 else if (strcasecmp (new_token.str, "QWORD") == 0)
6246 new_token.code = T_QWORD;
6248 else if (strcasecmp (new_token.str, "XWORD") == 0)
6249 new_token.code = T_XWORD;
6251 else if (strcasecmp (new_token.str, "PTR") == 0)
6252 new_token.code = T_PTR;
6254 else if (strcasecmp (new_token.str, "SHORT") == 0)
6255 new_token.code = T_SHORT;
6257 else if (strcasecmp (new_token.str, "OFFSET") == 0)
6259 new_token.code = T_OFFSET;
6261 /* ??? This is not mentioned in the MASM grammar but gcc
6262 makes use of it with -mintel-syntax. OFFSET may be
6263 followed by FLAT: */
6264 if (strncasecmp (q, " FLAT:", 6) == 0)
6265 strcat (new_token.str, " FLAT:");
6268 /* ??? This is not mentioned in the MASM grammar. */
6269 else if (strcasecmp (new_token.str, "FLAT") == 0)
6270 new_token.code = T_OFFSET;
6272 else
6273 new_token.code = T_ID;
6277 else
6278 as_bad (_("Unrecognized token `%s'"), intel_parser.op_string);
6280 intel_parser.op_string += strlen (new_token.str);
6281 cur_token = new_token;
6284 /* Put cur_token back into the token stream and make cur_token point to
6285 prev_token. */
6286 static void
6287 intel_putback_token ()
6289 intel_parser.op_string -= strlen (cur_token.str);
6290 free (cur_token.str);
6291 cur_token = prev_token;
6293 /* Forget prev_token. */
6294 prev_token.code = T_NIL;
6295 prev_token.reg = NULL;
6296 prev_token.str = NULL;
6300 tc_x86_regname_to_dw2regnum (const char *regname)
6302 unsigned int regnum;
6303 unsigned int regnames_count;
6304 char *regnames_32[] =
6306 "eax", "ecx", "edx", "ebx",
6307 "esp", "ebp", "esi", "edi",
6308 "eip"
6310 char *regnames_64[] =
6312 "rax", "rbx", "rcx", "rdx",
6313 "rdi", "rsi", "rbp", "rsp",
6314 "r8", "r9", "r10", "r11",
6315 "r12", "r13", "r14", "r15",
6316 "rip"
6318 char **regnames;
6320 if (flag_code == CODE_64BIT)
6322 regnames = regnames_64;
6323 regnames_count = ARRAY_SIZE (regnames_64);
6325 else
6327 regnames = regnames_32;
6328 regnames_count = ARRAY_SIZE (regnames_32);
6331 for (regnum = 0; regnum < regnames_count; regnum++)
6332 if (strcmp (regname, regnames[regnum]) == 0)
6333 return regnum;
6335 return -1;
6338 void
6339 tc_x86_frame_initial_instructions (void)
6341 static unsigned int sp_regno;
6343 if (!sp_regno)
6344 sp_regno = tc_x86_regname_to_dw2regnum (flag_code == CODE_64BIT
6345 ? "rsp" : "esp");
6347 cfi_add_CFA_def_cfa (sp_regno, -x86_cie_data_alignment);
6348 cfi_add_CFA_offset (x86_dwarf2_return_column, x86_cie_data_alignment);
6352 i386_elf_section_type (const char *str, size_t len)
6354 if (flag_code == CODE_64BIT
6355 && len == sizeof ("unwind") - 1
6356 && strncmp (str, "unwind", 6) == 0)
6357 return SHT_X86_64_UNWIND;
6359 return -1;
6362 #ifdef TE_PE
6363 void
6364 tc_pe_dwarf2_emit_offset (symbolS *symbol, unsigned int size)
6366 expressionS expr;
6368 expr.X_op = O_secrel;
6369 expr.X_add_symbol = symbol;
6370 expr.X_add_number = 0;
6371 emit_expr (&expr, size);
6373 #endif