2005-04-05 Paolo Bonzini <bonzini@gnu.org>
[binutils.git] / gas / config / tc-i386.c
blob486b0cf65faaa1ac7b5442128a0e7300716b912f
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, 2005
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;
317 static const char *cpu_sub_arch_name = NULL;
319 /* CPU feature flags. */
320 static unsigned int cpu_arch_flags = CpuUnknownFlags | CpuNo64;
322 /* If set, conditional jumps are not automatically promoted to handle
323 larger than a byte offset. */
324 static unsigned int no_cond_jump_promotion = 0;
326 /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
327 symbolS *GOT_symbol;
329 /* The dwarf2 return column, adjusted for 32 or 64 bit. */
330 unsigned int x86_dwarf2_return_column;
332 /* The dwarf2 data alignment, adjusted for 32 or 64 bit. */
333 int x86_cie_data_alignment;
335 /* Interface to relax_segment.
336 There are 3 major relax states for 386 jump insns because the
337 different types of jumps add different sizes to frags when we're
338 figuring out what sort of jump to choose to reach a given label. */
340 /* Types. */
341 #define UNCOND_JUMP 0
342 #define COND_JUMP 1
343 #define COND_JUMP86 2
345 /* Sizes. */
346 #define CODE16 1
347 #define SMALL 0
348 #define SMALL16 (SMALL | CODE16)
349 #define BIG 2
350 #define BIG16 (BIG | CODE16)
352 #ifndef INLINE
353 #ifdef __GNUC__
354 #define INLINE __inline__
355 #else
356 #define INLINE
357 #endif
358 #endif
360 #define ENCODE_RELAX_STATE(type, size) \
361 ((relax_substateT) (((type) << 2) | (size)))
362 #define TYPE_FROM_RELAX_STATE(s) \
363 ((s) >> 2)
364 #define DISP_SIZE_FROM_RELAX_STATE(s) \
365 ((((s) & 3) == BIG ? 4 : (((s) & 3) == BIG16 ? 2 : 1)))
367 /* This table is used by relax_frag to promote short jumps to long
368 ones where necessary. SMALL (short) jumps may be promoted to BIG
369 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
370 don't allow a short jump in a 32 bit code segment to be promoted to
371 a 16 bit offset jump because it's slower (requires data size
372 prefix), and doesn't work, unless the destination is in the bottom
373 64k of the code segment (The top 16 bits of eip are zeroed). */
375 const relax_typeS md_relax_table[] =
377 /* The fields are:
378 1) most positive reach of this state,
379 2) most negative reach of this state,
380 3) how many bytes this mode will have in the variable part of the frag
381 4) which index into the table to try if we can't fit into this one. */
383 /* UNCOND_JUMP states. */
384 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
385 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
386 /* dword jmp adds 4 bytes to frag:
387 0 extra opcode bytes, 4 displacement bytes. */
388 {0, 0, 4, 0},
389 /* word jmp adds 2 byte2 to frag:
390 0 extra opcode bytes, 2 displacement bytes. */
391 {0, 0, 2, 0},
393 /* COND_JUMP states. */
394 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
395 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
396 /* dword conditionals adds 5 bytes to frag:
397 1 extra opcode byte, 4 displacement bytes. */
398 {0, 0, 5, 0},
399 /* word conditionals add 3 bytes to frag:
400 1 extra opcode byte, 2 displacement bytes. */
401 {0, 0, 3, 0},
403 /* COND_JUMP86 states. */
404 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG)},
405 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG16)},
406 /* dword conditionals adds 5 bytes to frag:
407 1 extra opcode byte, 4 displacement bytes. */
408 {0, 0, 5, 0},
409 /* word conditionals add 4 bytes to frag:
410 1 displacement byte and a 3 byte long branch insn. */
411 {0, 0, 4, 0}
414 static const arch_entry cpu_arch[] = {
415 {"i8086", Cpu086 },
416 {"i186", Cpu086|Cpu186 },
417 {"i286", Cpu086|Cpu186|Cpu286 },
418 {"i386", Cpu086|Cpu186|Cpu286|Cpu386 },
419 {"i486", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486 },
420 {"i586", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586 },
421 {"i686", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686 },
422 {"pentium", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586 },
423 {"pentiumpro",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686 },
424 {"pentiumii", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX },
425 {"pentiumiii",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuMMX2|CpuSSE },
426 {"pentium4", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2|CpuSSE|CpuSSE2 },
427 {"prescott", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuPNI },
428 {"k6", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX },
429 {"k6_2", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow },
430 {"athlon", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA },
431 {"sledgehammer",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2 },
432 {".mmx", CpuMMX },
433 {".sse", CpuMMX|CpuMMX2|CpuSSE },
434 {".sse2", CpuMMX|CpuMMX2|CpuSSE|CpuSSE2 },
435 {".3dnow", CpuMMX|Cpu3dnow },
436 {".3dnowa", CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA },
437 {".padlock", CpuPadLock },
438 {NULL, 0 }
441 const pseudo_typeS md_pseudo_table[] =
443 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
444 {"align", s_align_bytes, 0},
445 #else
446 {"align", s_align_ptwo, 0},
447 #endif
448 {"arch", set_cpu_arch, 0},
449 #ifndef I386COFF
450 {"bss", s_bss, 0},
451 #endif
452 {"ffloat", float_cons, 'f'},
453 {"dfloat", float_cons, 'd'},
454 {"tfloat", float_cons, 'x'},
455 {"value", cons, 2},
456 {"noopt", s_ignore, 0},
457 {"optim", s_ignore, 0},
458 {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
459 {"code16", set_code_flag, CODE_16BIT},
460 {"code32", set_code_flag, CODE_32BIT},
461 {"code64", set_code_flag, CODE_64BIT},
462 {"intel_syntax", set_intel_syntax, 1},
463 {"att_syntax", set_intel_syntax, 0},
464 {"file", (void (*) PARAMS ((int))) dwarf2_directive_file, 0},
465 {"loc", dwarf2_directive_loc, 0},
466 #ifdef TE_PE
467 {"secrel32", pe_directive_secrel, 0},
468 #endif
469 {0, 0, 0}
472 /* For interface with expression (). */
473 extern char *input_line_pointer;
475 /* Hash table for instruction mnemonic lookup. */
476 static struct hash_control *op_hash;
478 /* Hash table for register lookup. */
479 static struct hash_control *reg_hash;
481 void
482 i386_align_code (fragP, count)
483 fragS *fragP;
484 int count;
486 /* Various efficient no-op patterns for aligning code labels.
487 Note: Don't try to assemble the instructions in the comments.
488 0L and 0w are not legal. */
489 static const char f32_1[] =
490 {0x90}; /* nop */
491 static const char f32_2[] =
492 {0x89,0xf6}; /* movl %esi,%esi */
493 static const char f32_3[] =
494 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
495 static const char f32_4[] =
496 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
497 static const char f32_5[] =
498 {0x90, /* nop */
499 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
500 static const char f32_6[] =
501 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
502 static const char f32_7[] =
503 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
504 static const char f32_8[] =
505 {0x90, /* nop */
506 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
507 static const char f32_9[] =
508 {0x89,0xf6, /* movl %esi,%esi */
509 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
510 static const char f32_10[] =
511 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
512 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
513 static const char f32_11[] =
514 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
515 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
516 static const char f32_12[] =
517 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
518 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
519 static const char f32_13[] =
520 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
521 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
522 static const char f32_14[] =
523 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
524 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
525 static const char f32_15[] =
526 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
527 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
528 static const char f16_3[] =
529 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
530 static const char f16_4[] =
531 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
532 static const char f16_5[] =
533 {0x90, /* nop */
534 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
535 static const char f16_6[] =
536 {0x89,0xf6, /* mov %si,%si */
537 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
538 static const char f16_7[] =
539 {0x8d,0x74,0x00, /* lea 0(%si),%si */
540 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
541 static const char f16_8[] =
542 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
543 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
544 static const char *const f32_patt[] = {
545 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
546 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
548 static const char *const f16_patt[] = {
549 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
550 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
553 if (count <= 0 || count > 15)
554 return;
556 /* The recommended way to pad 64bit code is to use NOPs preceded by
557 maximally four 0x66 prefixes. Balance the size of nops. */
558 if (flag_code == CODE_64BIT)
560 int i;
561 int nnops = (count + 3) / 4;
562 int len = count / nnops;
563 int remains = count - nnops * len;
564 int pos = 0;
566 for (i = 0; i < remains; i++)
568 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len);
569 fragP->fr_literal[fragP->fr_fix + pos + len] = 0x90;
570 pos += len + 1;
572 for (; i < nnops; i++)
574 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len - 1);
575 fragP->fr_literal[fragP->fr_fix + pos + len - 1] = 0x90;
576 pos += len;
579 else
580 if (flag_code == CODE_16BIT)
582 memcpy (fragP->fr_literal + fragP->fr_fix,
583 f16_patt[count - 1], count);
584 if (count > 8)
585 /* Adjust jump offset. */
586 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
588 else
589 memcpy (fragP->fr_literal + fragP->fr_fix,
590 f32_patt[count - 1], count);
591 fragP->fr_var = count;
594 static INLINE unsigned int
595 mode_from_disp_size (t)
596 unsigned int t;
598 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
601 static INLINE int
602 fits_in_signed_byte (num)
603 offsetT num;
605 return (num >= -128) && (num <= 127);
608 static INLINE int
609 fits_in_unsigned_byte (num)
610 offsetT num;
612 return (num & 0xff) == num;
615 static INLINE int
616 fits_in_unsigned_word (num)
617 offsetT num;
619 return (num & 0xffff) == num;
622 static INLINE int
623 fits_in_signed_word (num)
624 offsetT num;
626 return (-32768 <= num) && (num <= 32767);
628 static INLINE int
629 fits_in_signed_long (num)
630 offsetT num ATTRIBUTE_UNUSED;
632 #ifndef BFD64
633 return 1;
634 #else
635 return (!(((offsetT) -1 << 31) & num)
636 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
637 #endif
638 } /* fits_in_signed_long() */
639 static INLINE int
640 fits_in_unsigned_long (num)
641 offsetT num ATTRIBUTE_UNUSED;
643 #ifndef BFD64
644 return 1;
645 #else
646 return (num & (((offsetT) 2 << 31) - 1)) == num;
647 #endif
648 } /* fits_in_unsigned_long() */
650 static int
651 smallest_imm_type (num)
652 offsetT num;
654 if (cpu_arch_flags != (Cpu086 | Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64))
656 /* This code is disabled on the 486 because all the Imm1 forms
657 in the opcode table are slower on the i486. They're the
658 versions with the implicitly specified single-position
659 displacement, which has another syntax if you really want to
660 use that form. */
661 if (num == 1)
662 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
664 return (fits_in_signed_byte (num)
665 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
666 : fits_in_unsigned_byte (num)
667 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
668 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
669 ? (Imm16 | Imm32 | Imm32S | Imm64)
670 : fits_in_signed_long (num)
671 ? (Imm32 | Imm32S | Imm64)
672 : fits_in_unsigned_long (num)
673 ? (Imm32 | Imm64)
674 : Imm64);
677 static offsetT
678 offset_in_range (val, size)
679 offsetT val;
680 int size;
682 addressT mask;
684 switch (size)
686 case 1: mask = ((addressT) 1 << 8) - 1; break;
687 case 2: mask = ((addressT) 1 << 16) - 1; break;
688 case 4: mask = ((addressT) 2 << 31) - 1; break;
689 #ifdef BFD64
690 case 8: mask = ((addressT) 2 << 63) - 1; break;
691 #endif
692 default: abort ();
695 /* If BFD64, sign extend val. */
696 if (!use_rela_relocations)
697 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
698 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
700 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
702 char buf1[40], buf2[40];
704 sprint_value (buf1, val);
705 sprint_value (buf2, val & mask);
706 as_warn (_("%s shortened to %s"), buf1, buf2);
708 return val & mask;
711 /* Returns 0 if attempting to add a prefix where one from the same
712 class already exists, 1 if non rep/repne added, 2 if rep/repne
713 added. */
714 static int
715 add_prefix (prefix)
716 unsigned int prefix;
718 int ret = 1;
719 int q;
721 if (prefix >= REX_OPCODE && prefix < REX_OPCODE + 16
722 && flag_code == CODE_64BIT)
723 q = REX_PREFIX;
724 else
725 switch (prefix)
727 default:
728 abort ();
730 case CS_PREFIX_OPCODE:
731 case DS_PREFIX_OPCODE:
732 case ES_PREFIX_OPCODE:
733 case FS_PREFIX_OPCODE:
734 case GS_PREFIX_OPCODE:
735 case SS_PREFIX_OPCODE:
736 q = SEG_PREFIX;
737 break;
739 case REPNE_PREFIX_OPCODE:
740 case REPE_PREFIX_OPCODE:
741 ret = 2;
742 /* fall thru */
743 case LOCK_PREFIX_OPCODE:
744 q = LOCKREP_PREFIX;
745 break;
747 case FWAIT_OPCODE:
748 q = WAIT_PREFIX;
749 break;
751 case ADDR_PREFIX_OPCODE:
752 q = ADDR_PREFIX;
753 break;
755 case DATA_PREFIX_OPCODE:
756 q = DATA_PREFIX;
757 break;
760 if (i.prefix[q] != 0)
762 as_bad (_("same type of prefix used twice"));
763 return 0;
766 i.prefixes += 1;
767 i.prefix[q] = prefix;
768 return ret;
771 static void
772 set_code_flag (value)
773 int value;
775 flag_code = value;
776 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
777 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
778 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
780 as_bad (_("64bit mode not supported on this CPU."));
782 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
784 as_bad (_("32bit mode not supported on this CPU."));
786 stackop_size = '\0';
789 static void
790 set_16bit_gcc_code_flag (new_code_flag)
791 int new_code_flag;
793 flag_code = new_code_flag;
794 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
795 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
796 stackop_size = LONG_MNEM_SUFFIX;
799 static void
800 set_intel_syntax (syntax_flag)
801 int syntax_flag;
803 /* Find out if register prefixing is specified. */
804 int ask_naked_reg = 0;
806 SKIP_WHITESPACE ();
807 if (!is_end_of_line[(unsigned char) *input_line_pointer])
809 char *string = input_line_pointer;
810 int e = get_symbol_end ();
812 if (strcmp (string, "prefix") == 0)
813 ask_naked_reg = 1;
814 else if (strcmp (string, "noprefix") == 0)
815 ask_naked_reg = -1;
816 else
817 as_bad (_("bad argument to syntax directive."));
818 *input_line_pointer = e;
820 demand_empty_rest_of_line ();
822 intel_syntax = syntax_flag;
824 if (ask_naked_reg == 0)
825 allow_naked_reg = (intel_syntax
826 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
827 else
828 allow_naked_reg = (ask_naked_reg < 0);
830 identifier_chars['%'] = intel_syntax && allow_naked_reg ? '%' : 0;
831 identifier_chars['$'] = intel_syntax ? '$' : 0;
834 static void
835 set_cpu_arch (dummy)
836 int dummy ATTRIBUTE_UNUSED;
838 SKIP_WHITESPACE ();
840 if (!is_end_of_line[(unsigned char) *input_line_pointer])
842 char *string = input_line_pointer;
843 int e = get_symbol_end ();
844 int i;
846 for (i = 0; cpu_arch[i].name; i++)
848 if (strcmp (string, cpu_arch[i].name) == 0)
850 if (*string != '.')
852 cpu_arch_name = cpu_arch[i].name;
853 cpu_sub_arch_name = NULL;
854 cpu_arch_flags = (cpu_arch[i].flags
855 | (flag_code == CODE_64BIT ? Cpu64 : CpuNo64));
856 break;
858 if ((cpu_arch_flags | cpu_arch[i].flags) != cpu_arch_flags)
860 cpu_sub_arch_name = cpu_arch[i].name;
861 cpu_arch_flags |= cpu_arch[i].flags;
863 *input_line_pointer = e;
864 demand_empty_rest_of_line ();
865 return;
868 if (!cpu_arch[i].name)
869 as_bad (_("no such architecture: `%s'"), string);
871 *input_line_pointer = e;
873 else
874 as_bad (_("missing cpu architecture"));
876 no_cond_jump_promotion = 0;
877 if (*input_line_pointer == ','
878 && !is_end_of_line[(unsigned char) input_line_pointer[1]])
880 char *string = ++input_line_pointer;
881 int e = get_symbol_end ();
883 if (strcmp (string, "nojumps") == 0)
884 no_cond_jump_promotion = 1;
885 else if (strcmp (string, "jumps") == 0)
887 else
888 as_bad (_("no such architecture modifier: `%s'"), string);
890 *input_line_pointer = e;
893 demand_empty_rest_of_line ();
896 unsigned long
897 i386_mach ()
899 if (!strcmp (default_arch, "x86_64"))
900 return bfd_mach_x86_64;
901 else if (!strcmp (default_arch, "i386"))
902 return bfd_mach_i386_i386;
903 else
904 as_fatal (_("Unknown architecture"));
907 void
908 md_begin ()
910 const char *hash_err;
912 /* Initialize op_hash hash table. */
913 op_hash = hash_new ();
916 const template *optab;
917 templates *core_optab;
919 /* Setup for loop. */
920 optab = i386_optab;
921 core_optab = (templates *) xmalloc (sizeof (templates));
922 core_optab->start = optab;
924 while (1)
926 ++optab;
927 if (optab->name == NULL
928 || strcmp (optab->name, (optab - 1)->name) != 0)
930 /* different name --> ship out current template list;
931 add to hash table; & begin anew. */
932 core_optab->end = optab;
933 hash_err = hash_insert (op_hash,
934 (optab - 1)->name,
935 (PTR) core_optab);
936 if (hash_err)
938 as_fatal (_("Internal Error: Can't hash %s: %s"),
939 (optab - 1)->name,
940 hash_err);
942 if (optab->name == NULL)
943 break;
944 core_optab = (templates *) xmalloc (sizeof (templates));
945 core_optab->start = optab;
950 /* Initialize reg_hash hash table. */
951 reg_hash = hash_new ();
953 const reg_entry *regtab;
955 for (regtab = i386_regtab;
956 regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
957 regtab++)
959 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
960 if (hash_err)
961 as_fatal (_("Internal Error: Can't hash %s: %s"),
962 regtab->reg_name,
963 hash_err);
967 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
969 int c;
970 char *p;
972 for (c = 0; c < 256; c++)
974 if (ISDIGIT (c))
976 digit_chars[c] = c;
977 mnemonic_chars[c] = c;
978 register_chars[c] = c;
979 operand_chars[c] = c;
981 else if (ISLOWER (c))
983 mnemonic_chars[c] = c;
984 register_chars[c] = c;
985 operand_chars[c] = c;
987 else if (ISUPPER (c))
989 mnemonic_chars[c] = TOLOWER (c);
990 register_chars[c] = mnemonic_chars[c];
991 operand_chars[c] = c;
994 if (ISALPHA (c) || ISDIGIT (c))
995 identifier_chars[c] = c;
996 else if (c >= 128)
998 identifier_chars[c] = c;
999 operand_chars[c] = c;
1003 #ifdef LEX_AT
1004 identifier_chars['@'] = '@';
1005 #endif
1006 #ifdef LEX_QM
1007 identifier_chars['?'] = '?';
1008 operand_chars['?'] = '?';
1009 #endif
1010 digit_chars['-'] = '-';
1011 identifier_chars['_'] = '_';
1012 identifier_chars['.'] = '.';
1014 for (p = operand_special_chars; *p != '\0'; p++)
1015 operand_chars[(unsigned char) *p] = *p;
1018 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1019 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
1021 record_alignment (text_section, 2);
1022 record_alignment (data_section, 2);
1023 record_alignment (bss_section, 2);
1025 #endif
1027 if (flag_code == CODE_64BIT)
1029 x86_dwarf2_return_column = 16;
1030 x86_cie_data_alignment = -8;
1032 else
1034 x86_dwarf2_return_column = 8;
1035 x86_cie_data_alignment = -4;
1039 void
1040 i386_print_statistics (file)
1041 FILE *file;
1043 hash_print_statistics (file, "i386 opcode", op_hash);
1044 hash_print_statistics (file, "i386 register", reg_hash);
1047 #ifdef DEBUG386
1049 /* Debugging routines for md_assemble. */
1050 static void pi PARAMS ((char *, i386_insn *));
1051 static void pte PARAMS ((template *));
1052 static void pt PARAMS ((unsigned int));
1053 static void pe PARAMS ((expressionS *));
1054 static void ps PARAMS ((symbolS *));
1056 static void
1057 pi (line, x)
1058 char *line;
1059 i386_insn *x;
1061 unsigned int i;
1063 fprintf (stdout, "%s: template ", line);
1064 pte (&x->tm);
1065 fprintf (stdout, " address: base %s index %s scale %x\n",
1066 x->base_reg ? x->base_reg->reg_name : "none",
1067 x->index_reg ? x->index_reg->reg_name : "none",
1068 x->log2_scale_factor);
1069 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
1070 x->rm.mode, x->rm.reg, x->rm.regmem);
1071 fprintf (stdout, " sib: base %x index %x scale %x\n",
1072 x->sib.base, x->sib.index, x->sib.scale);
1073 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
1074 (x->rex & REX_MODE64) != 0,
1075 (x->rex & REX_EXTX) != 0,
1076 (x->rex & REX_EXTY) != 0,
1077 (x->rex & REX_EXTZ) != 0);
1078 for (i = 0; i < x->operands; i++)
1080 fprintf (stdout, " #%d: ", i + 1);
1081 pt (x->types[i]);
1082 fprintf (stdout, "\n");
1083 if (x->types[i]
1084 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
1085 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
1086 if (x->types[i] & Imm)
1087 pe (x->op[i].imms);
1088 if (x->types[i] & Disp)
1089 pe (x->op[i].disps);
1093 static void
1094 pte (t)
1095 template *t;
1097 unsigned int i;
1098 fprintf (stdout, " %d operands ", t->operands);
1099 fprintf (stdout, "opcode %x ", t->base_opcode);
1100 if (t->extension_opcode != None)
1101 fprintf (stdout, "ext %x ", t->extension_opcode);
1102 if (t->opcode_modifier & D)
1103 fprintf (stdout, "D");
1104 if (t->opcode_modifier & W)
1105 fprintf (stdout, "W");
1106 fprintf (stdout, "\n");
1107 for (i = 0; i < t->operands; i++)
1109 fprintf (stdout, " #%d type ", i + 1);
1110 pt (t->operand_types[i]);
1111 fprintf (stdout, "\n");
1115 static void
1116 pe (e)
1117 expressionS *e;
1119 fprintf (stdout, " operation %d\n", e->X_op);
1120 fprintf (stdout, " add_number %ld (%lx)\n",
1121 (long) e->X_add_number, (long) e->X_add_number);
1122 if (e->X_add_symbol)
1124 fprintf (stdout, " add_symbol ");
1125 ps (e->X_add_symbol);
1126 fprintf (stdout, "\n");
1128 if (e->X_op_symbol)
1130 fprintf (stdout, " op_symbol ");
1131 ps (e->X_op_symbol);
1132 fprintf (stdout, "\n");
1136 static void
1137 ps (s)
1138 symbolS *s;
1140 fprintf (stdout, "%s type %s%s",
1141 S_GET_NAME (s),
1142 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
1143 segment_name (S_GET_SEGMENT (s)));
1146 struct type_name
1148 unsigned int mask;
1149 char *tname;
1152 static const type_names[] =
1154 { Reg8, "r8" },
1155 { Reg16, "r16" },
1156 { Reg32, "r32" },
1157 { Reg64, "r64" },
1158 { Imm8, "i8" },
1159 { Imm8S, "i8s" },
1160 { Imm16, "i16" },
1161 { Imm32, "i32" },
1162 { Imm32S, "i32s" },
1163 { Imm64, "i64" },
1164 { Imm1, "i1" },
1165 { BaseIndex, "BaseIndex" },
1166 { Disp8, "d8" },
1167 { Disp16, "d16" },
1168 { Disp32, "d32" },
1169 { Disp32S, "d32s" },
1170 { Disp64, "d64" },
1171 { InOutPortReg, "InOutPortReg" },
1172 { ShiftCount, "ShiftCount" },
1173 { Control, "control reg" },
1174 { Test, "test reg" },
1175 { Debug, "debug reg" },
1176 { FloatReg, "FReg" },
1177 { FloatAcc, "FAcc" },
1178 { SReg2, "SReg2" },
1179 { SReg3, "SReg3" },
1180 { Acc, "Acc" },
1181 { JumpAbsolute, "Jump Absolute" },
1182 { RegMMX, "rMMX" },
1183 { RegXMM, "rXMM" },
1184 { EsSeg, "es" },
1185 { 0, "" }
1188 static void
1189 pt (t)
1190 unsigned int t;
1192 const struct type_name *ty;
1194 for (ty = type_names; ty->mask; ty++)
1195 if (t & ty->mask)
1196 fprintf (stdout, "%s, ", ty->tname);
1197 fflush (stdout);
1200 #endif /* DEBUG386 */
1202 static bfd_reloc_code_real_type reloc
1203 PARAMS ((int, int, int, bfd_reloc_code_real_type));
1205 static bfd_reloc_code_real_type
1206 reloc (size, pcrel, sign, other)
1207 int size;
1208 int pcrel;
1209 int sign;
1210 bfd_reloc_code_real_type other;
1212 if (other != NO_RELOC)
1213 return other;
1215 if (pcrel)
1217 if (!sign)
1218 as_bad (_("There are no unsigned pc-relative relocations"));
1219 switch (size)
1221 case 1: return BFD_RELOC_8_PCREL;
1222 case 2: return BFD_RELOC_16_PCREL;
1223 case 4: return BFD_RELOC_32_PCREL;
1225 as_bad (_("can not do %d byte pc-relative relocation"), size);
1227 else
1229 if (sign)
1230 switch (size)
1232 case 4: return BFD_RELOC_X86_64_32S;
1234 else
1235 switch (size)
1237 case 1: return BFD_RELOC_8;
1238 case 2: return BFD_RELOC_16;
1239 case 4: return BFD_RELOC_32;
1240 case 8: return BFD_RELOC_64;
1242 as_bad (_("can not do %s %d byte relocation"),
1243 sign ? "signed" : "unsigned", size);
1246 abort ();
1247 return BFD_RELOC_NONE;
1250 /* Here we decide which fixups can be adjusted to make them relative to
1251 the beginning of the section instead of the symbol. Basically we need
1252 to make sure that the dynamic relocations are done correctly, so in
1253 some cases we force the original symbol to be used. */
1256 tc_i386_fix_adjustable (fixP)
1257 fixS *fixP ATTRIBUTE_UNUSED;
1259 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1260 if (OUTPUT_FLAVOR != bfd_target_elf_flavour)
1261 return 1;
1263 /* Don't adjust pc-relative references to merge sections in 64-bit
1264 mode. */
1265 if (use_rela_relocations
1266 && (S_GET_SEGMENT (fixP->fx_addsy)->flags & SEC_MERGE) != 0
1267 && fixP->fx_pcrel)
1268 return 0;
1270 /* The x86_64 GOTPCREL are represented as 32bit PCrel relocations
1271 and changed later by validate_fix. */
1272 if (GOT_symbol && fixP->fx_subsy == GOT_symbol
1273 && fixP->fx_r_type == BFD_RELOC_32_PCREL)
1274 return 0;
1276 /* adjust_reloc_syms doesn't know about the GOT. */
1277 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1278 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1279 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1280 || fixP->fx_r_type == BFD_RELOC_386_TLS_GD
1281 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDM
1282 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDO_32
1283 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE_32
1284 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE
1285 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTIE
1286 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE_32
1287 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE
1288 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1289 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1290 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
1291 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSGD
1292 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSLD
1293 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF32
1294 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTTPOFF
1295 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF32
1296 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1297 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1298 return 0;
1299 #endif
1300 return 1;
1303 static int intel_float_operand PARAMS ((const char *mnemonic));
1305 static int
1306 intel_float_operand (mnemonic)
1307 const char *mnemonic;
1309 /* Note that the value returned is meaningful only for opcodes with (memory)
1310 operands, hence the code here is free to improperly handle opcodes that
1311 have no operands (for better performance and smaller code). */
1313 if (mnemonic[0] != 'f')
1314 return 0; /* non-math */
1316 switch (mnemonic[1])
1318 /* fclex, fdecstp, fdisi, femms, feni, fincstp, finit, fsetpm, and
1319 the fs segment override prefix not currently handled because no
1320 call path can make opcodes without operands get here */
1321 case 'i':
1322 return 2 /* integer op */;
1323 case 'l':
1324 if (mnemonic[2] == 'd' && (mnemonic[3] == 'c' || mnemonic[3] == 'e'))
1325 return 3; /* fldcw/fldenv */
1326 break;
1327 case 'n':
1328 if (mnemonic[2] != 'o' /* fnop */)
1329 return 3; /* non-waiting control op */
1330 break;
1331 case 'r':
1332 if (mnemonic[2] == 's')
1333 return 3; /* frstor/frstpm */
1334 break;
1335 case 's':
1336 if (mnemonic[2] == 'a')
1337 return 3; /* fsave */
1338 if (mnemonic[2] == 't')
1340 switch (mnemonic[3])
1342 case 'c': /* fstcw */
1343 case 'd': /* fstdw */
1344 case 'e': /* fstenv */
1345 case 's': /* fsts[gw] */
1346 return 3;
1349 break;
1350 case 'x':
1351 if (mnemonic[2] == 'r' || mnemonic[2] == 's')
1352 return 0; /* fxsave/fxrstor are not really math ops */
1353 break;
1356 return 1;
1359 /* This is the guts of the machine-dependent assembler. LINE points to a
1360 machine dependent instruction. This function is supposed to emit
1361 the frags/bytes it assembles to. */
1363 void
1364 md_assemble (line)
1365 char *line;
1367 int j;
1368 char mnemonic[MAX_MNEM_SIZE];
1370 /* Initialize globals. */
1371 memset (&i, '\0', sizeof (i));
1372 for (j = 0; j < MAX_OPERANDS; j++)
1373 i.reloc[j] = NO_RELOC;
1374 memset (disp_expressions, '\0', sizeof (disp_expressions));
1375 memset (im_expressions, '\0', sizeof (im_expressions));
1376 save_stack_p = save_stack;
1378 /* First parse an instruction mnemonic & call i386_operand for the operands.
1379 We assume that the scrubber has arranged it so that line[0] is the valid
1380 start of a (possibly prefixed) mnemonic. */
1382 line = parse_insn (line, mnemonic);
1383 if (line == NULL)
1384 return;
1386 line = parse_operands (line, mnemonic);
1387 if (line == NULL)
1388 return;
1390 /* Now we've parsed the mnemonic into a set of templates, and have the
1391 operands at hand. */
1393 /* All intel opcodes have reversed operands except for "bound" and
1394 "enter". We also don't reverse intersegment "jmp" and "call"
1395 instructions with 2 immediate operands so that the immediate segment
1396 precedes the offset, as it does when in AT&T mode. "enter" and the
1397 intersegment "jmp" and "call" instructions are the only ones that
1398 have two immediate operands. */
1399 if (intel_syntax && i.operands > 1
1400 && (strcmp (mnemonic, "bound") != 0)
1401 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1402 swap_operands ();
1404 if (i.imm_operands)
1405 optimize_imm ();
1407 if (i.disp_operands)
1408 optimize_disp ();
1410 /* Next, we find a template that matches the given insn,
1411 making sure the overlap of the given operands types is consistent
1412 with the template operand types. */
1414 if (!match_template ())
1415 return;
1417 if (intel_syntax)
1419 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1420 if (SYSV386_COMPAT
1421 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1422 i.tm.base_opcode ^= FloatR;
1424 /* Zap movzx and movsx suffix. The suffix may have been set from
1425 "word ptr" or "byte ptr" on the source operand, but we'll use
1426 the suffix later to choose the destination register. */
1427 if ((i.tm.base_opcode & ~9) == 0x0fb6)
1429 if (i.reg_operands < 2
1430 && !i.suffix
1431 && (~i.tm.opcode_modifier
1432 & (No_bSuf
1433 | No_wSuf
1434 | No_lSuf
1435 | No_sSuf
1436 | No_xSuf
1437 | No_qSuf)))
1438 as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
1440 i.suffix = 0;
1444 if (i.tm.opcode_modifier & FWait)
1445 if (!add_prefix (FWAIT_OPCODE))
1446 return;
1448 /* Check string instruction segment overrides. */
1449 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1451 if (!check_string ())
1452 return;
1455 if (!process_suffix ())
1456 return;
1458 /* Make still unresolved immediate matches conform to size of immediate
1459 given in i.suffix. */
1460 if (!finalize_imm ())
1461 return;
1463 if (i.types[0] & Imm1)
1464 i.imm_operands = 0; /* kludge for shift insns. */
1465 if (i.types[0] & ImplicitRegister)
1466 i.reg_operands--;
1467 if (i.types[1] & ImplicitRegister)
1468 i.reg_operands--;
1469 if (i.types[2] & ImplicitRegister)
1470 i.reg_operands--;
1472 if (i.tm.opcode_modifier & ImmExt)
1474 expressionS *exp;
1476 if ((i.tm.cpu_flags & CpuPNI) && i.operands > 0)
1478 /* These Intel Prescott New Instructions have the fixed
1479 operands with an opcode suffix which is coded in the same
1480 place as an 8-bit immediate field would be. Here we check
1481 those operands and remove them afterwards. */
1482 unsigned int x;
1484 for (x = 0; x < i.operands; x++)
1485 if (i.op[x].regs->reg_num != x)
1486 as_bad (_("can't use register '%%%s' as operand %d in '%s'."),
1487 i.op[x].regs->reg_name, x + 1, i.tm.name);
1488 i.operands = 0;
1491 /* These AMD 3DNow! and Intel Katmai New Instructions have an
1492 opcode suffix which is coded in the same place as an 8-bit
1493 immediate field would be. Here we fake an 8-bit immediate
1494 operand from the opcode suffix stored in tm.extension_opcode. */
1496 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
1498 exp = &im_expressions[i.imm_operands++];
1499 i.op[i.operands].imms = exp;
1500 i.types[i.operands++] = Imm8;
1501 exp->X_op = O_constant;
1502 exp->X_add_number = i.tm.extension_opcode;
1503 i.tm.extension_opcode = None;
1506 /* For insns with operands there are more diddles to do to the opcode. */
1507 if (i.operands)
1509 if (!process_operands ())
1510 return;
1512 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
1514 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
1515 as_warn (_("translating to `%sp'"), i.tm.name);
1518 /* Handle conversion of 'int $3' --> special int3 insn. */
1519 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
1521 i.tm.base_opcode = INT3_OPCODE;
1522 i.imm_operands = 0;
1525 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
1526 && i.op[0].disps->X_op == O_constant)
1528 /* Convert "jmp constant" (and "call constant") to a jump (call) to
1529 the absolute address given by the constant. Since ix86 jumps and
1530 calls are pc relative, we need to generate a reloc. */
1531 i.op[0].disps->X_add_symbol = &abs_symbol;
1532 i.op[0].disps->X_op = O_symbol;
1535 if ((i.tm.opcode_modifier & Rex64) != 0)
1536 i.rex |= REX_MODE64;
1538 /* For 8 bit registers we need an empty rex prefix. Also if the
1539 instruction already has a prefix, we need to convert old
1540 registers to new ones. */
1542 if (((i.types[0] & Reg8) != 0
1543 && (i.op[0].regs->reg_flags & RegRex64) != 0)
1544 || ((i.types[1] & Reg8) != 0
1545 && (i.op[1].regs->reg_flags & RegRex64) != 0)
1546 || (((i.types[0] & Reg8) != 0 || (i.types[1] & Reg8) != 0)
1547 && i.rex != 0))
1549 int x;
1551 i.rex |= REX_OPCODE;
1552 for (x = 0; x < 2; x++)
1554 /* Look for 8 bit operand that uses old registers. */
1555 if ((i.types[x] & Reg8) != 0
1556 && (i.op[x].regs->reg_flags & RegRex64) == 0)
1558 /* In case it is "hi" register, give up. */
1559 if (i.op[x].regs->reg_num > 3)
1560 as_bad (_("can't encode register '%%%s' in an instruction requiring REX prefix."),
1561 i.op[x].regs->reg_name);
1563 /* Otherwise it is equivalent to the extended register.
1564 Since the encoding doesn't change this is merely
1565 cosmetic cleanup for debug output. */
1567 i.op[x].regs = i.op[x].regs + 8;
1572 if (i.rex != 0)
1573 add_prefix (REX_OPCODE | i.rex);
1575 /* We are ready to output the insn. */
1576 output_insn ();
1579 static char *
1580 parse_insn (line, mnemonic)
1581 char *line;
1582 char *mnemonic;
1584 char *l = line;
1585 char *token_start = l;
1586 char *mnem_p;
1587 int supported;
1588 const template *t;
1590 /* Non-zero if we found a prefix only acceptable with string insns. */
1591 const char *expecting_string_instruction = NULL;
1593 while (1)
1595 mnem_p = mnemonic;
1596 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1598 mnem_p++;
1599 if (mnem_p >= mnemonic + MAX_MNEM_SIZE)
1601 as_bad (_("no such instruction: `%s'"), token_start);
1602 return NULL;
1604 l++;
1606 if (!is_space_char (*l)
1607 && *l != END_OF_INSN
1608 && *l != PREFIX_SEPARATOR
1609 && *l != ',')
1611 as_bad (_("invalid character %s in mnemonic"),
1612 output_invalid (*l));
1613 return NULL;
1615 if (token_start == l)
1617 if (*l == PREFIX_SEPARATOR)
1618 as_bad (_("expecting prefix; got nothing"));
1619 else
1620 as_bad (_("expecting mnemonic; got nothing"));
1621 return NULL;
1624 /* Look up instruction (or prefix) via hash table. */
1625 current_templates = hash_find (op_hash, mnemonic);
1627 if (*l != END_OF_INSN
1628 && (!is_space_char (*l) || l[1] != END_OF_INSN)
1629 && current_templates
1630 && (current_templates->start->opcode_modifier & IsPrefix))
1632 /* If we are in 16-bit mode, do not allow addr16 or data16.
1633 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1634 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1635 && flag_code != CODE_64BIT
1636 && (((current_templates->start->opcode_modifier & Size32) != 0)
1637 ^ (flag_code == CODE_16BIT)))
1639 as_bad (_("redundant %s prefix"),
1640 current_templates->start->name);
1641 return NULL;
1643 /* Add prefix, checking for repeated prefixes. */
1644 switch (add_prefix (current_templates->start->base_opcode))
1646 case 0:
1647 return NULL;
1648 case 2:
1649 expecting_string_instruction = current_templates->start->name;
1650 break;
1652 /* Skip past PREFIX_SEPARATOR and reset token_start. */
1653 token_start = ++l;
1655 else
1656 break;
1659 if (!current_templates)
1661 /* See if we can get a match by trimming off a suffix. */
1662 switch (mnem_p[-1])
1664 case WORD_MNEM_SUFFIX:
1665 if (intel_syntax && (intel_float_operand (mnemonic) & 2))
1666 i.suffix = SHORT_MNEM_SUFFIX;
1667 else
1668 case BYTE_MNEM_SUFFIX:
1669 case QWORD_MNEM_SUFFIX:
1670 i.suffix = mnem_p[-1];
1671 mnem_p[-1] = '\0';
1672 current_templates = hash_find (op_hash, mnemonic);
1673 break;
1674 case SHORT_MNEM_SUFFIX:
1675 case LONG_MNEM_SUFFIX:
1676 if (!intel_syntax)
1678 i.suffix = mnem_p[-1];
1679 mnem_p[-1] = '\0';
1680 current_templates = hash_find (op_hash, mnemonic);
1682 break;
1684 /* Intel Syntax. */
1685 case 'd':
1686 if (intel_syntax)
1688 if (intel_float_operand (mnemonic) == 1)
1689 i.suffix = SHORT_MNEM_SUFFIX;
1690 else
1691 i.suffix = LONG_MNEM_SUFFIX;
1692 mnem_p[-1] = '\0';
1693 current_templates = hash_find (op_hash, mnemonic);
1695 break;
1697 if (!current_templates)
1699 as_bad (_("no such instruction: `%s'"), token_start);
1700 return NULL;
1704 if (current_templates->start->opcode_modifier & (Jump | JumpByte))
1706 /* Check for a branch hint. We allow ",pt" and ",pn" for
1707 predict taken and predict not taken respectively.
1708 I'm not sure that branch hints actually do anything on loop
1709 and jcxz insns (JumpByte) for current Pentium4 chips. They
1710 may work in the future and it doesn't hurt to accept them
1711 now. */
1712 if (l[0] == ',' && l[1] == 'p')
1714 if (l[2] == 't')
1716 if (!add_prefix (DS_PREFIX_OPCODE))
1717 return NULL;
1718 l += 3;
1720 else if (l[2] == 'n')
1722 if (!add_prefix (CS_PREFIX_OPCODE))
1723 return NULL;
1724 l += 3;
1728 /* Any other comma loses. */
1729 if (*l == ',')
1731 as_bad (_("invalid character %s in mnemonic"),
1732 output_invalid (*l));
1733 return NULL;
1736 /* Check if instruction is supported on specified architecture. */
1737 supported = 0;
1738 for (t = current_templates->start; t < current_templates->end; ++t)
1740 if (!((t->cpu_flags & ~(Cpu64 | CpuNo64))
1741 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64))))
1742 supported |= 1;
1743 if (!(t->cpu_flags & (flag_code == CODE_64BIT ? CpuNo64 : Cpu64)))
1744 supported |= 2;
1746 if (!(supported & 2))
1748 as_bad (flag_code == CODE_64BIT
1749 ? _("`%s' is not supported in 64-bit mode")
1750 : _("`%s' is only supported in 64-bit mode"),
1751 current_templates->start->name);
1752 return NULL;
1754 if (!(supported & 1))
1756 as_warn (_("`%s' is not supported on `%s%s'"),
1757 current_templates->start->name,
1758 cpu_arch_name,
1759 cpu_sub_arch_name ? cpu_sub_arch_name : "");
1761 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
1763 as_warn (_("use .code16 to ensure correct addressing mode"));
1766 /* Check for rep/repne without a string instruction. */
1767 if (expecting_string_instruction
1768 && !(current_templates->start->opcode_modifier & IsString))
1770 as_bad (_("expecting string instruction after `%s'"),
1771 expecting_string_instruction);
1772 return NULL;
1775 return l;
1778 static char *
1779 parse_operands (l, mnemonic)
1780 char *l;
1781 const char *mnemonic;
1783 char *token_start;
1785 /* 1 if operand is pending after ','. */
1786 unsigned int expecting_operand = 0;
1788 /* Non-zero if operand parens not balanced. */
1789 unsigned int paren_not_balanced;
1791 while (*l != END_OF_INSN)
1793 /* Skip optional white space before operand. */
1794 if (is_space_char (*l))
1795 ++l;
1796 if (!is_operand_char (*l) && *l != END_OF_INSN)
1798 as_bad (_("invalid character %s before operand %d"),
1799 output_invalid (*l),
1800 i.operands + 1);
1801 return NULL;
1803 token_start = l; /* after white space */
1804 paren_not_balanced = 0;
1805 while (paren_not_balanced || *l != ',')
1807 if (*l == END_OF_INSN)
1809 if (paren_not_balanced)
1811 if (!intel_syntax)
1812 as_bad (_("unbalanced parenthesis in operand %d."),
1813 i.operands + 1);
1814 else
1815 as_bad (_("unbalanced brackets in operand %d."),
1816 i.operands + 1);
1817 return NULL;
1819 else
1820 break; /* we are done */
1822 else if (!is_operand_char (*l) && !is_space_char (*l))
1824 as_bad (_("invalid character %s in operand %d"),
1825 output_invalid (*l),
1826 i.operands + 1);
1827 return NULL;
1829 if (!intel_syntax)
1831 if (*l == '(')
1832 ++paren_not_balanced;
1833 if (*l == ')')
1834 --paren_not_balanced;
1836 else
1838 if (*l == '[')
1839 ++paren_not_balanced;
1840 if (*l == ']')
1841 --paren_not_balanced;
1843 l++;
1845 if (l != token_start)
1846 { /* Yes, we've read in another operand. */
1847 unsigned int operand_ok;
1848 this_operand = i.operands++;
1849 if (i.operands > MAX_OPERANDS)
1851 as_bad (_("spurious operands; (%d operands/instruction max)"),
1852 MAX_OPERANDS);
1853 return NULL;
1855 /* Now parse operand adding info to 'i' as we go along. */
1856 END_STRING_AND_SAVE (l);
1858 if (intel_syntax)
1859 operand_ok =
1860 i386_intel_operand (token_start,
1861 intel_float_operand (mnemonic));
1862 else
1863 operand_ok = i386_operand (token_start);
1865 RESTORE_END_STRING (l);
1866 if (!operand_ok)
1867 return NULL;
1869 else
1871 if (expecting_operand)
1873 expecting_operand_after_comma:
1874 as_bad (_("expecting operand after ','; got nothing"));
1875 return NULL;
1877 if (*l == ',')
1879 as_bad (_("expecting operand before ','; got nothing"));
1880 return NULL;
1884 /* Now *l must be either ',' or END_OF_INSN. */
1885 if (*l == ',')
1887 if (*++l == END_OF_INSN)
1889 /* Just skip it, if it's \n complain. */
1890 goto expecting_operand_after_comma;
1892 expecting_operand = 1;
1895 return l;
1898 static void
1899 swap_operands ()
1901 union i386_op temp_op;
1902 unsigned int temp_type;
1903 enum bfd_reloc_code_real temp_reloc;
1904 int xchg1 = 0;
1905 int xchg2 = 0;
1907 if (i.operands == 2)
1909 xchg1 = 0;
1910 xchg2 = 1;
1912 else if (i.operands == 3)
1914 xchg1 = 0;
1915 xchg2 = 2;
1917 temp_type = i.types[xchg2];
1918 i.types[xchg2] = i.types[xchg1];
1919 i.types[xchg1] = temp_type;
1920 temp_op = i.op[xchg2];
1921 i.op[xchg2] = i.op[xchg1];
1922 i.op[xchg1] = temp_op;
1923 temp_reloc = i.reloc[xchg2];
1924 i.reloc[xchg2] = i.reloc[xchg1];
1925 i.reloc[xchg1] = temp_reloc;
1927 if (i.mem_operands == 2)
1929 const seg_entry *temp_seg;
1930 temp_seg = i.seg[0];
1931 i.seg[0] = i.seg[1];
1932 i.seg[1] = temp_seg;
1936 /* Try to ensure constant immediates are represented in the smallest
1937 opcode possible. */
1938 static void
1939 optimize_imm ()
1941 char guess_suffix = 0;
1942 int op;
1944 if (i.suffix)
1945 guess_suffix = i.suffix;
1946 else if (i.reg_operands)
1948 /* Figure out a suffix from the last register operand specified.
1949 We can't do this properly yet, ie. excluding InOutPortReg,
1950 but the following works for instructions with immediates.
1951 In any case, we can't set i.suffix yet. */
1952 for (op = i.operands; --op >= 0;)
1953 if (i.types[op] & Reg)
1955 if (i.types[op] & Reg8)
1956 guess_suffix = BYTE_MNEM_SUFFIX;
1957 else if (i.types[op] & Reg16)
1958 guess_suffix = WORD_MNEM_SUFFIX;
1959 else if (i.types[op] & Reg32)
1960 guess_suffix = LONG_MNEM_SUFFIX;
1961 else if (i.types[op] & Reg64)
1962 guess_suffix = QWORD_MNEM_SUFFIX;
1963 break;
1966 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
1967 guess_suffix = WORD_MNEM_SUFFIX;
1969 for (op = i.operands; --op >= 0;)
1970 if (i.types[op] & Imm)
1972 switch (i.op[op].imms->X_op)
1974 case O_constant:
1975 /* If a suffix is given, this operand may be shortened. */
1976 switch (guess_suffix)
1978 case LONG_MNEM_SUFFIX:
1979 i.types[op] |= Imm32 | Imm64;
1980 break;
1981 case WORD_MNEM_SUFFIX:
1982 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
1983 break;
1984 case BYTE_MNEM_SUFFIX:
1985 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
1986 break;
1989 /* If this operand is at most 16 bits, convert it
1990 to a signed 16 bit number before trying to see
1991 whether it will fit in an even smaller size.
1992 This allows a 16-bit operand such as $0xffe0 to
1993 be recognised as within Imm8S range. */
1994 if ((i.types[op] & Imm16)
1995 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
1997 i.op[op].imms->X_add_number =
1998 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
2000 if ((i.types[op] & Imm32)
2001 && ((i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1))
2002 == 0))
2004 i.op[op].imms->X_add_number = ((i.op[op].imms->X_add_number
2005 ^ ((offsetT) 1 << 31))
2006 - ((offsetT) 1 << 31));
2008 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
2010 /* We must avoid matching of Imm32 templates when 64bit
2011 only immediate is available. */
2012 if (guess_suffix == QWORD_MNEM_SUFFIX)
2013 i.types[op] &= ~Imm32;
2014 break;
2016 case O_absent:
2017 case O_register:
2018 abort ();
2020 /* Symbols and expressions. */
2021 default:
2022 /* Convert symbolic operand to proper sizes for matching. */
2023 switch (guess_suffix)
2025 case QWORD_MNEM_SUFFIX:
2026 i.types[op] = Imm64 | Imm32S;
2027 break;
2028 case LONG_MNEM_SUFFIX:
2029 i.types[op] = Imm32;
2030 break;
2031 case WORD_MNEM_SUFFIX:
2032 i.types[op] = Imm16;
2033 break;
2034 case BYTE_MNEM_SUFFIX:
2035 i.types[op] = Imm8 | Imm8S;
2036 break;
2038 break;
2043 /* Try to use the smallest displacement type too. */
2044 static void
2045 optimize_disp ()
2047 int op;
2049 for (op = i.operands; --op >= 0;)
2050 if ((i.types[op] & Disp) && i.op[op].disps->X_op == O_constant)
2052 offsetT disp = i.op[op].disps->X_add_number;
2054 if (i.types[op] & Disp16)
2056 /* We know this operand is at most 16 bits, so
2057 convert to a signed 16 bit number before trying
2058 to see whether it will fit in an even smaller
2059 size. */
2061 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
2063 else if (i.types[op] & Disp32)
2065 /* We know this operand is at most 32 bits, so convert to a
2066 signed 32 bit number before trying to see whether it will
2067 fit in an even smaller size. */
2068 disp &= (((offsetT) 2 << 31) - 1);
2069 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
2071 if (flag_code == CODE_64BIT)
2073 if (fits_in_signed_long (disp))
2074 i.types[op] |= Disp32S;
2075 if (fits_in_unsigned_long (disp))
2076 i.types[op] |= Disp32;
2078 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
2079 && fits_in_signed_byte (disp))
2080 i.types[op] |= Disp8;
2084 static int
2085 match_template ()
2087 /* Points to template once we've found it. */
2088 const template *t;
2089 unsigned int overlap0, overlap1, overlap2;
2090 unsigned int found_reverse_match;
2091 int suffix_check;
2093 #define MATCH(overlap, given, template) \
2094 ((overlap & ~JumpAbsolute) \
2095 && (((given) & (BaseIndex | JumpAbsolute)) \
2096 == ((overlap) & (BaseIndex | JumpAbsolute))))
2098 /* If given types r0 and r1 are registers they must be of the same type
2099 unless the expected operand type register overlap is null.
2100 Note that Acc in a template matches every size of reg. */
2101 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
2102 (((g0) & Reg) == 0 || ((g1) & Reg) == 0 \
2103 || ((g0) & Reg) == ((g1) & Reg) \
2104 || ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
2106 overlap0 = 0;
2107 overlap1 = 0;
2108 overlap2 = 0;
2109 found_reverse_match = 0;
2110 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
2111 ? No_bSuf
2112 : (i.suffix == WORD_MNEM_SUFFIX
2113 ? No_wSuf
2114 : (i.suffix == SHORT_MNEM_SUFFIX
2115 ? No_sSuf
2116 : (i.suffix == LONG_MNEM_SUFFIX
2117 ? No_lSuf
2118 : (i.suffix == QWORD_MNEM_SUFFIX
2119 ? No_qSuf
2120 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX
2121 ? No_xSuf : 0))))));
2123 t = current_templates->start;
2124 if (i.suffix == QWORD_MNEM_SUFFIX
2125 && flag_code != CODE_64BIT
2126 && (intel_syntax
2127 ? !(t->opcode_modifier & IgnoreSize)
2128 && !intel_float_operand (t->name)
2129 : intel_float_operand (t->name) != 2)
2130 && (!(t->operand_types[0] & (RegMMX | RegXMM))
2131 || !(t->operand_types[t->operands > 1] & (RegMMX | RegXMM)))
2132 && (t->base_opcode != 0x0fc7
2133 || t->extension_opcode != 1 /* cmpxchg8b */))
2134 t = current_templates->end;
2135 for (; t < current_templates->end; t++)
2137 /* Must have right number of operands. */
2138 if (i.operands != t->operands)
2139 continue;
2141 /* Check the suffix, except for some instructions in intel mode. */
2142 if ((t->opcode_modifier & suffix_check)
2143 && !(intel_syntax
2144 && (t->opcode_modifier & IgnoreSize)))
2145 continue;
2147 /* Do not verify operands when there are none. */
2148 else if (!t->operands)
2150 if (t->cpu_flags & ~cpu_arch_flags)
2151 continue;
2152 /* We've found a match; break out of loop. */
2153 break;
2156 overlap0 = i.types[0] & t->operand_types[0];
2157 switch (t->operands)
2159 case 1:
2160 if (!MATCH (overlap0, i.types[0], t->operand_types[0]))
2161 continue;
2162 break;
2163 case 2:
2164 case 3:
2165 overlap1 = i.types[1] & t->operand_types[1];
2166 if (!MATCH (overlap0, i.types[0], t->operand_types[0])
2167 || !MATCH (overlap1, i.types[1], t->operand_types[1])
2168 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2169 t->operand_types[0],
2170 overlap1, i.types[1],
2171 t->operand_types[1]))
2173 /* Check if other direction is valid ... */
2174 if ((t->opcode_modifier & (D | FloatD)) == 0)
2175 continue;
2177 /* Try reversing direction of operands. */
2178 overlap0 = i.types[0] & t->operand_types[1];
2179 overlap1 = i.types[1] & t->operand_types[0];
2180 if (!MATCH (overlap0, i.types[0], t->operand_types[1])
2181 || !MATCH (overlap1, i.types[1], t->operand_types[0])
2182 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2183 t->operand_types[1],
2184 overlap1, i.types[1],
2185 t->operand_types[0]))
2187 /* Does not match either direction. */
2188 continue;
2190 /* found_reverse_match holds which of D or FloatDR
2191 we've found. */
2192 found_reverse_match = t->opcode_modifier & (D | FloatDR);
2194 /* Found a forward 2 operand match here. */
2195 else if (t->operands == 3)
2197 /* Here we make use of the fact that there are no
2198 reverse match 3 operand instructions, and all 3
2199 operand instructions only need to be checked for
2200 register consistency between operands 2 and 3. */
2201 overlap2 = i.types[2] & t->operand_types[2];
2202 if (!MATCH (overlap2, i.types[2], t->operand_types[2])
2203 || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1],
2204 t->operand_types[1],
2205 overlap2, i.types[2],
2206 t->operand_types[2]))
2208 continue;
2210 /* Found either forward/reverse 2 or 3 operand match here:
2211 slip through to break. */
2213 if (t->cpu_flags & ~cpu_arch_flags)
2215 found_reverse_match = 0;
2216 continue;
2218 /* We've found a match; break out of loop. */
2219 break;
2222 if (t == current_templates->end)
2224 /* We found no match. */
2225 as_bad (_("suffix or operands invalid for `%s'"),
2226 current_templates->start->name);
2227 return 0;
2230 if (!quiet_warnings)
2232 if (!intel_syntax
2233 && ((i.types[0] & JumpAbsolute)
2234 != (t->operand_types[0] & JumpAbsolute)))
2236 as_warn (_("indirect %s without `*'"), t->name);
2239 if ((t->opcode_modifier & (IsPrefix | IgnoreSize))
2240 == (IsPrefix | IgnoreSize))
2242 /* Warn them that a data or address size prefix doesn't
2243 affect assembly of the next line of code. */
2244 as_warn (_("stand-alone `%s' prefix"), t->name);
2248 /* Copy the template we found. */
2249 i.tm = *t;
2250 if (found_reverse_match)
2252 /* If we found a reverse match we must alter the opcode
2253 direction bit. found_reverse_match holds bits to change
2254 (different for int & float insns). */
2256 i.tm.base_opcode ^= found_reverse_match;
2258 i.tm.operand_types[0] = t->operand_types[1];
2259 i.tm.operand_types[1] = t->operand_types[0];
2262 return 1;
2265 static int
2266 check_string ()
2268 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
2269 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
2271 if (i.seg[0] != NULL && i.seg[0] != &es)
2273 as_bad (_("`%s' operand %d must use `%%es' segment"),
2274 i.tm.name,
2275 mem_op + 1);
2276 return 0;
2278 /* There's only ever one segment override allowed per instruction.
2279 This instruction possibly has a legal segment override on the
2280 second operand, so copy the segment to where non-string
2281 instructions store it, allowing common code. */
2282 i.seg[0] = i.seg[1];
2284 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
2286 if (i.seg[1] != NULL && i.seg[1] != &es)
2288 as_bad (_("`%s' operand %d must use `%%es' segment"),
2289 i.tm.name,
2290 mem_op + 2);
2291 return 0;
2294 return 1;
2297 static int
2298 process_suffix (void)
2300 /* If matched instruction specifies an explicit instruction mnemonic
2301 suffix, use it. */
2302 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
2304 if (i.tm.opcode_modifier & Size16)
2305 i.suffix = WORD_MNEM_SUFFIX;
2306 else if (i.tm.opcode_modifier & Size64)
2307 i.suffix = QWORD_MNEM_SUFFIX;
2308 else
2309 i.suffix = LONG_MNEM_SUFFIX;
2311 else if (i.reg_operands)
2313 /* If there's no instruction mnemonic suffix we try to invent one
2314 based on register operands. */
2315 if (!i.suffix)
2317 /* We take i.suffix from the last register operand specified,
2318 Destination register type is more significant than source
2319 register type. */
2320 int op;
2322 for (op = i.operands; --op >= 0;)
2323 if ((i.types[op] & Reg)
2324 && !(i.tm.operand_types[op] & InOutPortReg))
2326 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
2327 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
2328 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
2329 LONG_MNEM_SUFFIX);
2330 break;
2333 else if (i.suffix == BYTE_MNEM_SUFFIX)
2335 if (!check_byte_reg ())
2336 return 0;
2338 else if (i.suffix == LONG_MNEM_SUFFIX)
2340 if (!check_long_reg ())
2341 return 0;
2343 else if (i.suffix == QWORD_MNEM_SUFFIX)
2345 if (!check_qword_reg ())
2346 return 0;
2348 else if (i.suffix == WORD_MNEM_SUFFIX)
2350 if (!check_word_reg ())
2351 return 0;
2353 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2354 /* Do nothing if the instruction is going to ignore the prefix. */
2356 else
2357 abort ();
2359 else if ((i.tm.opcode_modifier & DefaultSize)
2360 && !i.suffix
2361 /* exclude fldenv/frstor/fsave/fstenv */
2362 && (i.tm.opcode_modifier & No_sSuf))
2364 i.suffix = stackop_size;
2366 else if (intel_syntax
2367 && !i.suffix
2368 && ((i.tm.operand_types[0] & JumpAbsolute)
2369 || (i.tm.opcode_modifier & (JumpByte|JumpInterSegment))
2370 || (i.tm.base_opcode == 0x0f01 /* [ls][gi]dt */
2371 && i.tm.extension_opcode <= 3)))
2373 switch (flag_code)
2375 case CODE_64BIT:
2376 if (!(i.tm.opcode_modifier & No_qSuf))
2378 i.suffix = QWORD_MNEM_SUFFIX;
2379 break;
2381 case CODE_32BIT:
2382 if (!(i.tm.opcode_modifier & No_lSuf))
2383 i.suffix = LONG_MNEM_SUFFIX;
2384 break;
2385 case CODE_16BIT:
2386 if (!(i.tm.opcode_modifier & No_wSuf))
2387 i.suffix = WORD_MNEM_SUFFIX;
2388 break;
2392 if (!i.suffix)
2394 if (!intel_syntax)
2396 if (i.tm.opcode_modifier & W)
2398 as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
2399 return 0;
2402 else
2404 unsigned int suffixes = ~i.tm.opcode_modifier
2405 & (No_bSuf
2406 | No_wSuf
2407 | No_lSuf
2408 | No_sSuf
2409 | No_xSuf
2410 | No_qSuf);
2412 if ((i.tm.opcode_modifier & W)
2413 || ((suffixes & (suffixes - 1))
2414 && !(i.tm.opcode_modifier & (DefaultSize | IgnoreSize))))
2416 as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
2417 return 0;
2422 /* Change the opcode based on the operand size given by i.suffix;
2423 We don't need to change things for byte insns. */
2425 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2427 /* It's not a byte, select word/dword operation. */
2428 if (i.tm.opcode_modifier & W)
2430 if (i.tm.opcode_modifier & ShortForm)
2431 i.tm.base_opcode |= 8;
2432 else
2433 i.tm.base_opcode |= 1;
2436 /* Now select between word & dword operations via the operand
2437 size prefix, except for instructions that will ignore this
2438 prefix anyway. */
2439 if (i.suffix != QWORD_MNEM_SUFFIX
2440 && i.suffix != LONG_DOUBLE_MNEM_SUFFIX
2441 && !(i.tm.opcode_modifier & (IgnoreSize | FloatMF))
2442 && ((i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2443 || (flag_code == CODE_64BIT
2444 && (i.tm.opcode_modifier & JumpByte))))
2446 unsigned int prefix = DATA_PREFIX_OPCODE;
2448 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2449 prefix = ADDR_PREFIX_OPCODE;
2451 if (!add_prefix (prefix))
2452 return 0;
2455 /* Set mode64 for an operand. */
2456 if (i.suffix == QWORD_MNEM_SUFFIX
2457 && flag_code == CODE_64BIT
2458 && (i.tm.opcode_modifier & NoRex64) == 0)
2459 i.rex |= REX_MODE64;
2461 /* Size floating point instruction. */
2462 if (i.suffix == LONG_MNEM_SUFFIX)
2463 if (i.tm.opcode_modifier & FloatMF)
2464 i.tm.base_opcode ^= 4;
2467 return 1;
2470 static int
2471 check_byte_reg (void)
2473 int op;
2475 for (op = i.operands; --op >= 0;)
2477 /* If this is an eight bit register, it's OK. If it's the 16 or
2478 32 bit version of an eight bit register, we will just use the
2479 low portion, and that's OK too. */
2480 if (i.types[op] & Reg8)
2481 continue;
2483 /* movzx and movsx should not generate this warning. */
2484 if (intel_syntax
2485 && (i.tm.base_opcode == 0xfb7
2486 || i.tm.base_opcode == 0xfb6
2487 || i.tm.base_opcode == 0x63
2488 || i.tm.base_opcode == 0xfbe
2489 || i.tm.base_opcode == 0xfbf))
2490 continue;
2492 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4)
2494 /* Prohibit these changes in the 64bit mode, since the
2495 lowering is more complicated. */
2496 if (flag_code == CODE_64BIT
2497 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2499 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2500 i.op[op].regs->reg_name,
2501 i.suffix);
2502 return 0;
2504 #if REGISTER_WARNINGS
2505 if (!quiet_warnings
2506 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2507 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2508 (i.op[op].regs + (i.types[op] & Reg16
2509 ? REGNAM_AL - REGNAM_AX
2510 : REGNAM_AL - REGNAM_EAX))->reg_name,
2511 i.op[op].regs->reg_name,
2512 i.suffix);
2513 #endif
2514 continue;
2516 /* Any other register is bad. */
2517 if (i.types[op] & (Reg | RegMMX | RegXMM
2518 | SReg2 | SReg3
2519 | Control | Debug | Test
2520 | FloatReg | FloatAcc))
2522 as_bad (_("`%%%s' not allowed with `%s%c'"),
2523 i.op[op].regs->reg_name,
2524 i.tm.name,
2525 i.suffix);
2526 return 0;
2529 return 1;
2532 static int
2533 check_long_reg ()
2535 int op;
2537 for (op = i.operands; --op >= 0;)
2538 /* Reject eight bit registers, except where the template requires
2539 them. (eg. movzb) */
2540 if ((i.types[op] & Reg8) != 0
2541 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2543 as_bad (_("`%%%s' not allowed with `%s%c'"),
2544 i.op[op].regs->reg_name,
2545 i.tm.name,
2546 i.suffix);
2547 return 0;
2549 /* Warn if the e prefix on a general reg is missing. */
2550 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2551 && (i.types[op] & Reg16) != 0
2552 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2554 /* Prohibit these changes in the 64bit mode, since the
2555 lowering is more complicated. */
2556 if (flag_code == CODE_64BIT)
2558 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2559 i.op[op].regs->reg_name,
2560 i.suffix);
2561 return 0;
2563 #if REGISTER_WARNINGS
2564 else
2565 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2566 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
2567 i.op[op].regs->reg_name,
2568 i.suffix);
2569 #endif
2571 /* Warn if the r prefix on a general reg is missing. */
2572 else if ((i.types[op] & Reg64) != 0
2573 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2575 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2576 i.op[op].regs->reg_name,
2577 i.suffix);
2578 return 0;
2580 return 1;
2583 static int
2584 check_qword_reg ()
2586 int op;
2588 for (op = i.operands; --op >= 0; )
2589 /* Reject eight bit registers, except where the template requires
2590 them. (eg. movzb) */
2591 if ((i.types[op] & Reg8) != 0
2592 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2594 as_bad (_("`%%%s' not allowed with `%s%c'"),
2595 i.op[op].regs->reg_name,
2596 i.tm.name,
2597 i.suffix);
2598 return 0;
2600 /* Warn if the e prefix on a general reg is missing. */
2601 else if (((i.types[op] & Reg16) != 0
2602 || (i.types[op] & Reg32) != 0)
2603 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2605 /* Prohibit these changes in the 64bit mode, since the
2606 lowering is more complicated. */
2607 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2608 i.op[op].regs->reg_name,
2609 i.suffix);
2610 return 0;
2612 return 1;
2615 static int
2616 check_word_reg ()
2618 int op;
2619 for (op = i.operands; --op >= 0;)
2620 /* Reject eight bit registers, except where the template requires
2621 them. (eg. movzb) */
2622 if ((i.types[op] & Reg8) != 0
2623 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2625 as_bad (_("`%%%s' not allowed with `%s%c'"),
2626 i.op[op].regs->reg_name,
2627 i.tm.name,
2628 i.suffix);
2629 return 0;
2631 /* Warn if the e prefix on a general reg is present. */
2632 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2633 && (i.types[op] & Reg32) != 0
2634 && (i.tm.operand_types[op] & (Reg16 | Acc)) != 0)
2636 /* Prohibit these changes in the 64bit mode, since the
2637 lowering is more complicated. */
2638 if (flag_code == CODE_64BIT)
2640 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2641 i.op[op].regs->reg_name,
2642 i.suffix);
2643 return 0;
2645 else
2646 #if REGISTER_WARNINGS
2647 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2648 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
2649 i.op[op].regs->reg_name,
2650 i.suffix);
2651 #endif
2653 return 1;
2656 static int
2657 finalize_imm ()
2659 unsigned int overlap0, overlap1, overlap2;
2661 overlap0 = i.types[0] & i.tm.operand_types[0];
2662 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64))
2663 && overlap0 != Imm8 && overlap0 != Imm8S
2664 && overlap0 != Imm16 && overlap0 != Imm32S
2665 && overlap0 != Imm32 && overlap0 != Imm64)
2667 if (i.suffix)
2669 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX
2670 ? Imm8 | Imm8S
2671 : (i.suffix == WORD_MNEM_SUFFIX
2672 ? Imm16
2673 : (i.suffix == QWORD_MNEM_SUFFIX
2674 ? Imm64 | Imm32S
2675 : Imm32)));
2677 else if (overlap0 == (Imm16 | Imm32S | Imm32)
2678 || overlap0 == (Imm16 | Imm32)
2679 || overlap0 == (Imm16 | Imm32S))
2681 overlap0 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
2682 ? Imm16 : Imm32S);
2684 if (overlap0 != Imm8 && overlap0 != Imm8S
2685 && overlap0 != Imm16 && overlap0 != Imm32S
2686 && overlap0 != Imm32 && overlap0 != Imm64)
2688 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
2689 return 0;
2692 i.types[0] = overlap0;
2694 overlap1 = i.types[1] & i.tm.operand_types[1];
2695 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32 | Imm64))
2696 && overlap1 != Imm8 && overlap1 != Imm8S
2697 && overlap1 != Imm16 && overlap1 != Imm32S
2698 && overlap1 != Imm32 && overlap1 != Imm64)
2700 if (i.suffix)
2702 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX
2703 ? Imm8 | Imm8S
2704 : (i.suffix == WORD_MNEM_SUFFIX
2705 ? Imm16
2706 : (i.suffix == QWORD_MNEM_SUFFIX
2707 ? Imm64 | Imm32S
2708 : Imm32)));
2710 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
2711 || overlap1 == (Imm16 | Imm32)
2712 || overlap1 == (Imm16 | Imm32S))
2714 overlap1 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
2715 ? Imm16 : Imm32S);
2717 if (overlap1 != Imm8 && overlap1 != Imm8S
2718 && overlap1 != Imm16 && overlap1 != Imm32S
2719 && overlap1 != Imm32 && overlap1 != Imm64)
2721 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
2722 return 0;
2725 i.types[1] = overlap1;
2727 overlap2 = i.types[2] & i.tm.operand_types[2];
2728 assert ((overlap2 & Imm) == 0);
2729 i.types[2] = overlap2;
2731 return 1;
2734 static int
2735 process_operands ()
2737 /* Default segment register this instruction will use for memory
2738 accesses. 0 means unknown. This is only for optimizing out
2739 unnecessary segment overrides. */
2740 const seg_entry *default_seg = 0;
2742 /* The imul $imm, %reg instruction is converted into
2743 imul $imm, %reg, %reg, and the clr %reg instruction
2744 is converted into xor %reg, %reg. */
2745 if (i.tm.opcode_modifier & regKludge)
2747 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
2748 /* Pretend we saw the extra register operand. */
2749 assert (i.op[first_reg_op + 1].regs == 0);
2750 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
2751 i.types[first_reg_op + 1] = i.types[first_reg_op];
2752 i.reg_operands = 2;
2755 if (i.tm.opcode_modifier & ShortForm)
2757 /* The register or float register operand is in operand 0 or 1. */
2758 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
2759 /* Register goes in low 3 bits of opcode. */
2760 i.tm.base_opcode |= i.op[op].regs->reg_num;
2761 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2762 i.rex |= REX_EXTZ;
2763 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2765 /* Warn about some common errors, but press on regardless.
2766 The first case can be generated by gcc (<= 2.8.1). */
2767 if (i.operands == 2)
2769 /* Reversed arguments on faddp, fsubp, etc. */
2770 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
2771 i.op[1].regs->reg_name,
2772 i.op[0].regs->reg_name);
2774 else
2776 /* Extraneous `l' suffix on fp insn. */
2777 as_warn (_("translating to `%s %%%s'"), i.tm.name,
2778 i.op[0].regs->reg_name);
2782 else if (i.tm.opcode_modifier & Modrm)
2784 /* The opcode is completed (modulo i.tm.extension_opcode which
2785 must be put into the modrm byte). Now, we make the modrm and
2786 index base bytes based on all the info we've collected. */
2788 default_seg = build_modrm_byte ();
2790 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
2792 if (i.tm.base_opcode == POP_SEG_SHORT
2793 && i.op[0].regs->reg_num == 1)
2795 as_bad (_("you can't `pop %%cs'"));
2796 return 0;
2798 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
2799 if ((i.op[0].regs->reg_flags & RegRex) != 0)
2800 i.rex |= REX_EXTZ;
2802 else if ((i.tm.base_opcode & ~(D | W)) == MOV_AX_DISP32)
2804 default_seg = &ds;
2806 else if ((i.tm.opcode_modifier & IsString) != 0)
2808 /* For the string instructions that allow a segment override
2809 on one of their operands, the default segment is ds. */
2810 default_seg = &ds;
2813 if (i.tm.base_opcode == 0x8d /* lea */ && i.seg[0] && !quiet_warnings)
2814 as_warn (_("segment override on `lea' is ineffectual"));
2816 /* If a segment was explicitly specified, and the specified segment
2817 is not the default, use an opcode prefix to select it. If we
2818 never figured out what the default segment is, then default_seg
2819 will be zero at this point, and the specified segment prefix will
2820 always be used. */
2821 if ((i.seg[0]) && (i.seg[0] != default_seg))
2823 if (!add_prefix (i.seg[0]->seg_prefix))
2824 return 0;
2826 return 1;
2829 static const seg_entry *
2830 build_modrm_byte ()
2832 const seg_entry *default_seg = 0;
2834 /* i.reg_operands MUST be the number of real register operands;
2835 implicit registers do not count. */
2836 if (i.reg_operands == 2)
2838 unsigned int source, dest;
2839 source = ((i.types[0]
2840 & (Reg | RegMMX | RegXMM
2841 | SReg2 | SReg3
2842 | Control | Debug | Test))
2843 ? 0 : 1);
2844 dest = source + 1;
2846 i.rm.mode = 3;
2847 /* One of the register operands will be encoded in the i.tm.reg
2848 field, the other in the combined i.tm.mode and i.tm.regmem
2849 fields. If no form of this instruction supports a memory
2850 destination operand, then we assume the source operand may
2851 sometimes be a memory operand and so we need to store the
2852 destination in the i.rm.reg field. */
2853 if ((i.tm.operand_types[dest] & AnyMem) == 0)
2855 i.rm.reg = i.op[dest].regs->reg_num;
2856 i.rm.regmem = i.op[source].regs->reg_num;
2857 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
2858 i.rex |= REX_EXTX;
2859 if ((i.op[source].regs->reg_flags & RegRex) != 0)
2860 i.rex |= REX_EXTZ;
2862 else
2864 i.rm.reg = i.op[source].regs->reg_num;
2865 i.rm.regmem = i.op[dest].regs->reg_num;
2866 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
2867 i.rex |= REX_EXTZ;
2868 if ((i.op[source].regs->reg_flags & RegRex) != 0)
2869 i.rex |= REX_EXTX;
2871 if (flag_code != CODE_64BIT && (i.rex & (REX_EXTX | REX_EXTZ)))
2873 if (!((i.types[0] | i.types[1]) & Control))
2874 abort ();
2875 i.rex &= ~(REX_EXTX | REX_EXTZ);
2876 add_prefix (LOCK_PREFIX_OPCODE);
2879 else
2880 { /* If it's not 2 reg operands... */
2881 if (i.mem_operands)
2883 unsigned int fake_zero_displacement = 0;
2884 unsigned int op = ((i.types[0] & AnyMem)
2886 : (i.types[1] & AnyMem) ? 1 : 2);
2888 default_seg = &ds;
2890 if (i.base_reg == 0)
2892 i.rm.mode = 0;
2893 if (!i.disp_operands)
2894 fake_zero_displacement = 1;
2895 if (i.index_reg == 0)
2897 /* Operand is just <disp> */
2898 if (flag_code == CODE_64BIT)
2900 /* 64bit mode overwrites the 32bit absolute
2901 addressing by RIP relative addressing and
2902 absolute addressing is encoded by one of the
2903 redundant SIB forms. */
2904 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2905 i.sib.base = NO_BASE_REGISTER;
2906 i.sib.index = NO_INDEX_REGISTER;
2907 i.types[op] = ((i.prefix[ADDR_PREFIX] == 0) ? Disp32S : Disp32);
2909 else if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
2911 i.rm.regmem = NO_BASE_REGISTER_16;
2912 i.types[op] = Disp16;
2914 else
2916 i.rm.regmem = NO_BASE_REGISTER;
2917 i.types[op] = Disp32;
2920 else /* !i.base_reg && i.index_reg */
2922 i.sib.index = i.index_reg->reg_num;
2923 i.sib.base = NO_BASE_REGISTER;
2924 i.sib.scale = i.log2_scale_factor;
2925 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2926 i.types[op] &= ~Disp;
2927 if (flag_code != CODE_64BIT)
2928 i.types[op] |= Disp32; /* Must be 32 bit */
2929 else
2930 i.types[op] |= Disp32S;
2931 if ((i.index_reg->reg_flags & RegRex) != 0)
2932 i.rex |= REX_EXTY;
2935 /* RIP addressing for 64bit mode. */
2936 else if (i.base_reg->reg_type == BaseIndex)
2938 i.rm.regmem = NO_BASE_REGISTER;
2939 i.types[op] &= ~ Disp;
2940 i.types[op] |= Disp32S;
2941 i.flags[op] = Operand_PCrel;
2942 if (! i.disp_operands)
2943 fake_zero_displacement = 1;
2945 else if (i.base_reg->reg_type & Reg16)
2947 switch (i.base_reg->reg_num)
2949 case 3: /* (%bx) */
2950 if (i.index_reg == 0)
2951 i.rm.regmem = 7;
2952 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
2953 i.rm.regmem = i.index_reg->reg_num - 6;
2954 break;
2955 case 5: /* (%bp) */
2956 default_seg = &ss;
2957 if (i.index_reg == 0)
2959 i.rm.regmem = 6;
2960 if ((i.types[op] & Disp) == 0)
2962 /* fake (%bp) into 0(%bp) */
2963 i.types[op] |= Disp8;
2964 fake_zero_displacement = 1;
2967 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
2968 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
2969 break;
2970 default: /* (%si) -> 4 or (%di) -> 5 */
2971 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
2973 i.rm.mode = mode_from_disp_size (i.types[op]);
2975 else /* i.base_reg and 32/64 bit mode */
2977 if (flag_code == CODE_64BIT
2978 && (i.types[op] & Disp))
2979 i.types[op] = (i.types[op] & Disp8) | (i.prefix[ADDR_PREFIX] == 0 ? Disp32S : Disp32);
2981 i.rm.regmem = i.base_reg->reg_num;
2982 if ((i.base_reg->reg_flags & RegRex) != 0)
2983 i.rex |= REX_EXTZ;
2984 i.sib.base = i.base_reg->reg_num;
2985 /* x86-64 ignores REX prefix bit here to avoid decoder
2986 complications. */
2987 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
2989 default_seg = &ss;
2990 if (i.disp_operands == 0)
2992 fake_zero_displacement = 1;
2993 i.types[op] |= Disp8;
2996 else if (i.base_reg->reg_num == ESP_REG_NUM)
2998 default_seg = &ss;
3000 i.sib.scale = i.log2_scale_factor;
3001 if (i.index_reg == 0)
3003 /* <disp>(%esp) becomes two byte modrm with no index
3004 register. We've already stored the code for esp
3005 in i.rm.regmem ie. ESCAPE_TO_TWO_BYTE_ADDRESSING.
3006 Any base register besides %esp will not use the
3007 extra modrm byte. */
3008 i.sib.index = NO_INDEX_REGISTER;
3009 #if !SCALE1_WHEN_NO_INDEX
3010 /* Another case where we force the second modrm byte. */
3011 if (i.log2_scale_factor)
3012 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3013 #endif
3015 else
3017 i.sib.index = i.index_reg->reg_num;
3018 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3019 if ((i.index_reg->reg_flags & RegRex) != 0)
3020 i.rex |= REX_EXTY;
3022 i.rm.mode = mode_from_disp_size (i.types[op]);
3025 if (fake_zero_displacement)
3027 /* Fakes a zero displacement assuming that i.types[op]
3028 holds the correct displacement size. */
3029 expressionS *exp;
3031 assert (i.op[op].disps == 0);
3032 exp = &disp_expressions[i.disp_operands++];
3033 i.op[op].disps = exp;
3034 exp->X_op = O_constant;
3035 exp->X_add_number = 0;
3036 exp->X_add_symbol = (symbolS *) 0;
3037 exp->X_op_symbol = (symbolS *) 0;
3041 /* Fill in i.rm.reg or i.rm.regmem field with register operand
3042 (if any) based on i.tm.extension_opcode. Again, we must be
3043 careful to make sure that segment/control/debug/test/MMX
3044 registers are coded into the i.rm.reg field. */
3045 if (i.reg_operands)
3047 unsigned int op =
3048 ((i.types[0]
3049 & (Reg | RegMMX | RegXMM
3050 | SReg2 | SReg3
3051 | Control | Debug | Test))
3053 : ((i.types[1]
3054 & (Reg | RegMMX | RegXMM
3055 | SReg2 | SReg3
3056 | Control | Debug | Test))
3058 : 2));
3059 /* If there is an extension opcode to put here, the register
3060 number must be put into the regmem field. */
3061 if (i.tm.extension_opcode != None)
3063 i.rm.regmem = i.op[op].regs->reg_num;
3064 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3065 i.rex |= REX_EXTZ;
3067 else
3069 i.rm.reg = i.op[op].regs->reg_num;
3070 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3071 i.rex |= REX_EXTX;
3074 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2 we
3075 must set it to 3 to indicate this is a register operand
3076 in the regmem field. */
3077 if (!i.mem_operands)
3078 i.rm.mode = 3;
3081 /* Fill in i.rm.reg field with extension opcode (if any). */
3082 if (i.tm.extension_opcode != None)
3083 i.rm.reg = i.tm.extension_opcode;
3085 return default_seg;
3088 static void
3089 output_branch ()
3091 char *p;
3092 int code16;
3093 int prefix;
3094 relax_substateT subtype;
3095 symbolS *sym;
3096 offsetT off;
3098 code16 = 0;
3099 if (flag_code == CODE_16BIT)
3100 code16 = CODE16;
3102 prefix = 0;
3103 if (i.prefix[DATA_PREFIX] != 0)
3105 prefix = 1;
3106 i.prefixes -= 1;
3107 code16 ^= CODE16;
3109 /* Pentium4 branch hints. */
3110 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3111 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3113 prefix++;
3114 i.prefixes--;
3116 if (i.prefix[REX_PREFIX] != 0)
3118 prefix++;
3119 i.prefixes--;
3122 if (i.prefixes != 0 && !intel_syntax)
3123 as_warn (_("skipping prefixes on this instruction"));
3125 /* It's always a symbol; End frag & setup for relax.
3126 Make sure there is enough room in this frag for the largest
3127 instruction we may generate in md_convert_frag. This is 2
3128 bytes for the opcode and room for the prefix and largest
3129 displacement. */
3130 frag_grow (prefix + 2 + 4);
3131 /* Prefix and 1 opcode byte go in fr_fix. */
3132 p = frag_more (prefix + 1);
3133 if (i.prefix[DATA_PREFIX] != 0)
3134 *p++ = DATA_PREFIX_OPCODE;
3135 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
3136 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
3137 *p++ = i.prefix[SEG_PREFIX];
3138 if (i.prefix[REX_PREFIX] != 0)
3139 *p++ = i.prefix[REX_PREFIX];
3140 *p = i.tm.base_opcode;
3142 if ((unsigned char) *p == JUMP_PC_RELATIVE)
3143 subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
3144 else if ((cpu_arch_flags & Cpu386) != 0)
3145 subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
3146 else
3147 subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
3148 subtype |= code16;
3150 sym = i.op[0].disps->X_add_symbol;
3151 off = i.op[0].disps->X_add_number;
3153 if (i.op[0].disps->X_op != O_constant
3154 && i.op[0].disps->X_op != O_symbol)
3156 /* Handle complex expressions. */
3157 sym = make_expr_symbol (i.op[0].disps);
3158 off = 0;
3161 /* 1 possible extra opcode + 4 byte displacement go in var part.
3162 Pass reloc in fr_var. */
3163 frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
3166 static void
3167 output_jump ()
3169 char *p;
3170 int size;
3171 fixS *fixP;
3173 if (i.tm.opcode_modifier & JumpByte)
3175 /* This is a loop or jecxz type instruction. */
3176 size = 1;
3177 if (i.prefix[ADDR_PREFIX] != 0)
3179 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
3180 i.prefixes -= 1;
3182 /* Pentium4 branch hints. */
3183 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3184 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3186 FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
3187 i.prefixes--;
3190 else
3192 int code16;
3194 code16 = 0;
3195 if (flag_code == CODE_16BIT)
3196 code16 = CODE16;
3198 if (i.prefix[DATA_PREFIX] != 0)
3200 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
3201 i.prefixes -= 1;
3202 code16 ^= CODE16;
3205 size = 4;
3206 if (code16)
3207 size = 2;
3210 if (i.prefix[REX_PREFIX] != 0)
3212 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
3213 i.prefixes -= 1;
3216 if (i.prefixes != 0 && !intel_syntax)
3217 as_warn (_("skipping prefixes on this instruction"));
3219 p = frag_more (1 + size);
3220 *p++ = i.tm.base_opcode;
3222 fixP = fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3223 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
3225 /* All jumps handled here are signed, but don't use a signed limit
3226 check for 32 and 16 bit jumps as we want to allow wrap around at
3227 4G and 64k respectively. */
3228 if (size == 1)
3229 fixP->fx_signed = 1;
3232 static void
3233 output_interseg_jump ()
3235 char *p;
3236 int size;
3237 int prefix;
3238 int code16;
3240 code16 = 0;
3241 if (flag_code == CODE_16BIT)
3242 code16 = CODE16;
3244 prefix = 0;
3245 if (i.prefix[DATA_PREFIX] != 0)
3247 prefix = 1;
3248 i.prefixes -= 1;
3249 code16 ^= CODE16;
3251 if (i.prefix[REX_PREFIX] != 0)
3253 prefix++;
3254 i.prefixes -= 1;
3257 size = 4;
3258 if (code16)
3259 size = 2;
3261 if (i.prefixes != 0 && !intel_syntax)
3262 as_warn (_("skipping prefixes on this instruction"));
3264 /* 1 opcode; 2 segment; offset */
3265 p = frag_more (prefix + 1 + 2 + size);
3267 if (i.prefix[DATA_PREFIX] != 0)
3268 *p++ = DATA_PREFIX_OPCODE;
3270 if (i.prefix[REX_PREFIX] != 0)
3271 *p++ = i.prefix[REX_PREFIX];
3273 *p++ = i.tm.base_opcode;
3274 if (i.op[1].imms->X_op == O_constant)
3276 offsetT n = i.op[1].imms->X_add_number;
3278 if (size == 2
3279 && !fits_in_unsigned_word (n)
3280 && !fits_in_signed_word (n))
3282 as_bad (_("16-bit jump out of range"));
3283 return;
3285 md_number_to_chars (p, n, size);
3287 else
3288 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3289 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
3290 if (i.op[0].imms->X_op != O_constant)
3291 as_bad (_("can't handle non absolute segment in `%s'"),
3292 i.tm.name);
3293 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
3296 static void
3297 output_insn ()
3299 fragS *insn_start_frag;
3300 offsetT insn_start_off;
3302 /* Tie dwarf2 debug info to the address at the start of the insn.
3303 We can't do this after the insn has been output as the current
3304 frag may have been closed off. eg. by frag_var. */
3305 dwarf2_emit_insn (0);
3307 insn_start_frag = frag_now;
3308 insn_start_off = frag_now_fix ();
3310 /* Output jumps. */
3311 if (i.tm.opcode_modifier & Jump)
3312 output_branch ();
3313 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
3314 output_jump ();
3315 else if (i.tm.opcode_modifier & JumpInterSegment)
3316 output_interseg_jump ();
3317 else
3319 /* Output normal instructions here. */
3320 char *p;
3321 unsigned char *q;
3323 /* All opcodes on i386 have either 1 or 2 bytes, PadLock instructions
3324 have 3 bytes. We may use one more higher byte to specify a prefix
3325 the instruction requires. */
3326 if ((i.tm.cpu_flags & CpuPadLock) != 0
3327 && (i.tm.base_opcode & 0xff000000) != 0)
3329 unsigned int prefix;
3330 prefix = (i.tm.base_opcode >> 24) & 0xff;
3332 if (prefix != REPE_PREFIX_OPCODE
3333 || i.prefix[LOCKREP_PREFIX] != REPE_PREFIX_OPCODE)
3334 add_prefix (prefix);
3336 else
3337 if ((i.tm.cpu_flags & CpuPadLock) == 0
3338 && (i.tm.base_opcode & 0xff0000) != 0)
3339 add_prefix ((i.tm.base_opcode >> 16) & 0xff);
3341 /* The prefix bytes. */
3342 for (q = i.prefix;
3343 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
3344 q++)
3346 if (*q)
3348 p = frag_more (1);
3349 md_number_to_chars (p, (valueT) *q, 1);
3353 /* Now the opcode; be careful about word order here! */
3354 if (fits_in_unsigned_byte (i.tm.base_opcode))
3356 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
3358 else
3360 if ((i.tm.cpu_flags & CpuPadLock) != 0)
3362 p = frag_more (3);
3363 *p++ = (i.tm.base_opcode >> 16) & 0xff;
3365 else
3366 p = frag_more (2);
3368 /* Put out high byte first: can't use md_number_to_chars! */
3369 *p++ = (i.tm.base_opcode >> 8) & 0xff;
3370 *p = i.tm.base_opcode & 0xff;
3373 /* Now the modrm byte and sib byte (if present). */
3374 if (i.tm.opcode_modifier & Modrm)
3376 p = frag_more (1);
3377 md_number_to_chars (p,
3378 (valueT) (i.rm.regmem << 0
3379 | i.rm.reg << 3
3380 | i.rm.mode << 6),
3382 /* If i.rm.regmem == ESP (4)
3383 && i.rm.mode != (Register mode)
3384 && not 16 bit
3385 ==> need second modrm byte. */
3386 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
3387 && i.rm.mode != 3
3388 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
3390 p = frag_more (1);
3391 md_number_to_chars (p,
3392 (valueT) (i.sib.base << 0
3393 | i.sib.index << 3
3394 | i.sib.scale << 6),
3399 if (i.disp_operands)
3400 output_disp (insn_start_frag, insn_start_off);
3402 if (i.imm_operands)
3403 output_imm (insn_start_frag, insn_start_off);
3406 #ifdef DEBUG386
3407 if (flag_debug)
3409 pi (line, &i);
3411 #endif /* DEBUG386 */
3414 static void
3415 output_disp (insn_start_frag, insn_start_off)
3416 fragS *insn_start_frag;
3417 offsetT insn_start_off;
3419 char *p;
3420 unsigned int n;
3422 for (n = 0; n < i.operands; n++)
3424 if (i.types[n] & Disp)
3426 if (i.op[n].disps->X_op == O_constant)
3428 int size;
3429 offsetT val;
3431 size = 4;
3432 if (i.types[n] & (Disp8 | Disp16 | Disp64))
3434 size = 2;
3435 if (i.types[n] & Disp8)
3436 size = 1;
3437 if (i.types[n] & Disp64)
3438 size = 8;
3440 val = offset_in_range (i.op[n].disps->X_add_number,
3441 size);
3442 p = frag_more (size);
3443 md_number_to_chars (p, val, size);
3445 else
3447 enum bfd_reloc_code_real reloc_type;
3448 int size = 4;
3449 int sign = 0;
3450 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
3452 /* The PC relative address is computed relative
3453 to the instruction boundary, so in case immediate
3454 fields follows, we need to adjust the value. */
3455 if (pcrel && i.imm_operands)
3457 int imm_size = 4;
3458 unsigned int n1;
3460 for (n1 = 0; n1 < i.operands; n1++)
3461 if (i.types[n1] & Imm)
3463 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
3465 imm_size = 2;
3466 if (i.types[n1] & (Imm8 | Imm8S))
3467 imm_size = 1;
3468 if (i.types[n1] & Imm64)
3469 imm_size = 8;
3471 break;
3473 /* We should find the immediate. */
3474 if (n1 == i.operands)
3475 abort ();
3476 i.op[n].disps->X_add_number -= imm_size;
3479 if (i.types[n] & Disp32S)
3480 sign = 1;
3482 if (i.types[n] & (Disp16 | Disp64))
3484 size = 2;
3485 if (i.types[n] & Disp64)
3486 size = 8;
3489 p = frag_more (size);
3490 reloc_type = reloc (size, pcrel, sign, i.reloc[n]);
3491 if (reloc_type == BFD_RELOC_32
3492 && GOT_symbol
3493 && GOT_symbol == i.op[n].disps->X_add_symbol
3494 && (i.op[n].disps->X_op == O_symbol
3495 || (i.op[n].disps->X_op == O_add
3496 && ((symbol_get_value_expression
3497 (i.op[n].disps->X_op_symbol)->X_op)
3498 == O_subtract))))
3500 offsetT add;
3502 if (insn_start_frag == frag_now)
3503 add = (p - frag_now->fr_literal) - insn_start_off;
3504 else
3506 fragS *fr;
3508 add = insn_start_frag->fr_fix - insn_start_off;
3509 for (fr = insn_start_frag->fr_next;
3510 fr && fr != frag_now; fr = fr->fr_next)
3511 add += fr->fr_fix;
3512 add += p - frag_now->fr_literal;
3515 /* We don't support dynamic linking on x86-64 yet. */
3516 if (flag_code == CODE_64BIT)
3517 abort ();
3518 reloc_type = BFD_RELOC_386_GOTPC;
3519 i.op[n].disps->X_add_number += add;
3521 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3522 i.op[n].disps, pcrel, reloc_type);
3528 static void
3529 output_imm (insn_start_frag, insn_start_off)
3530 fragS *insn_start_frag;
3531 offsetT insn_start_off;
3533 char *p;
3534 unsigned int n;
3536 for (n = 0; n < i.operands; n++)
3538 if (i.types[n] & Imm)
3540 if (i.op[n].imms->X_op == O_constant)
3542 int size;
3543 offsetT val;
3545 size = 4;
3546 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3548 size = 2;
3549 if (i.types[n] & (Imm8 | Imm8S))
3550 size = 1;
3551 else if (i.types[n] & Imm64)
3552 size = 8;
3554 val = offset_in_range (i.op[n].imms->X_add_number,
3555 size);
3556 p = frag_more (size);
3557 md_number_to_chars (p, val, size);
3559 else
3561 /* Not absolute_section.
3562 Need a 32-bit fixup (don't support 8bit
3563 non-absolute imms). Try to support other
3564 sizes ... */
3565 enum bfd_reloc_code_real reloc_type;
3566 int size = 4;
3567 int sign = 0;
3569 if ((i.types[n] & (Imm32S))
3570 && (i.suffix == QWORD_MNEM_SUFFIX
3571 || (!i.suffix && (i.tm.opcode_modifier & No_lSuf))))
3572 sign = 1;
3573 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3575 size = 2;
3576 if (i.types[n] & (Imm8 | Imm8S))
3577 size = 1;
3578 if (i.types[n] & Imm64)
3579 size = 8;
3582 p = frag_more (size);
3583 reloc_type = reloc (size, 0, sign, i.reloc[n]);
3585 /* This is tough to explain. We end up with this one if we
3586 * have operands that look like
3587 * "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal here is to
3588 * obtain the absolute address of the GOT, and it is strongly
3589 * preferable from a performance point of view to avoid using
3590 * a runtime relocation for this. The actual sequence of
3591 * instructions often look something like:
3593 * call .L66
3594 * .L66:
3595 * popl %ebx
3596 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
3598 * The call and pop essentially return the absolute address
3599 * of the label .L66 and store it in %ebx. The linker itself
3600 * will ultimately change the first operand of the addl so
3601 * that %ebx points to the GOT, but to keep things simple, the
3602 * .o file must have this operand set so that it generates not
3603 * the absolute address of .L66, but the absolute address of
3604 * itself. This allows the linker itself simply treat a GOTPC
3605 * relocation as asking for a pcrel offset to the GOT to be
3606 * added in, and the addend of the relocation is stored in the
3607 * operand field for the instruction itself.
3609 * Our job here is to fix the operand so that it would add
3610 * the correct offset so that %ebx would point to itself. The
3611 * thing that is tricky is that .-.L66 will point to the
3612 * beginning of the instruction, so we need to further modify
3613 * the operand so that it will point to itself. There are
3614 * other cases where you have something like:
3616 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
3618 * and here no correction would be required. Internally in
3619 * the assembler we treat operands of this form as not being
3620 * pcrel since the '.' is explicitly mentioned, and I wonder
3621 * whether it would simplify matters to do it this way. Who
3622 * knows. In earlier versions of the PIC patches, the
3623 * pcrel_adjust field was used to store the correction, but
3624 * since the expression is not pcrel, I felt it would be
3625 * confusing to do it this way. */
3627 if (reloc_type == BFD_RELOC_32
3628 && GOT_symbol
3629 && GOT_symbol == i.op[n].imms->X_add_symbol
3630 && (i.op[n].imms->X_op == O_symbol
3631 || (i.op[n].imms->X_op == O_add
3632 && ((symbol_get_value_expression
3633 (i.op[n].imms->X_op_symbol)->X_op)
3634 == O_subtract))))
3636 offsetT add;
3638 if (insn_start_frag == frag_now)
3639 add = (p - frag_now->fr_literal) - insn_start_off;
3640 else
3642 fragS *fr;
3644 add = insn_start_frag->fr_fix - insn_start_off;
3645 for (fr = insn_start_frag->fr_next;
3646 fr && fr != frag_now; fr = fr->fr_next)
3647 add += fr->fr_fix;
3648 add += p - frag_now->fr_literal;
3651 /* We don't support dynamic linking on x86-64 yet. */
3652 if (flag_code == CODE_64BIT)
3653 abort ();
3654 reloc_type = BFD_RELOC_386_GOTPC;
3655 i.op[n].imms->X_add_number += add;
3657 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3658 i.op[n].imms, 0, reloc_type);
3664 #ifndef LEX_AT
3665 static char *lex_got PARAMS ((enum bfd_reloc_code_real *, int *));
3667 /* Parse operands of the form
3668 <symbol>@GOTOFF+<nnn>
3669 and similar .plt or .got references.
3671 If we find one, set up the correct relocation in RELOC and copy the
3672 input string, minus the `@GOTOFF' into a malloc'd buffer for
3673 parsing by the calling routine. Return this buffer, and if ADJUST
3674 is non-null set it to the length of the string we removed from the
3675 input line. Otherwise return NULL. */
3676 static char *
3677 lex_got (reloc, adjust)
3678 enum bfd_reloc_code_real *reloc;
3679 int *adjust;
3681 static const char * const mode_name[NUM_FLAG_CODE] = { "32", "16", "64" };
3682 static const struct {
3683 const char *str;
3684 const enum bfd_reloc_code_real rel[NUM_FLAG_CODE];
3685 } gotrel[] = {
3686 { "PLT", { BFD_RELOC_386_PLT32, 0, BFD_RELOC_X86_64_PLT32 } },
3687 { "GOTOFF", { BFD_RELOC_386_GOTOFF, 0, 0 } },
3688 { "GOTPCREL", { 0, 0, BFD_RELOC_X86_64_GOTPCREL } },
3689 { "TLSGD", { BFD_RELOC_386_TLS_GD, 0, BFD_RELOC_X86_64_TLSGD } },
3690 { "TLSLDM", { BFD_RELOC_386_TLS_LDM, 0, 0 } },
3691 { "TLSLD", { 0, 0, BFD_RELOC_X86_64_TLSLD } },
3692 { "GOTTPOFF", { BFD_RELOC_386_TLS_IE_32, 0, BFD_RELOC_X86_64_GOTTPOFF } },
3693 { "TPOFF", { BFD_RELOC_386_TLS_LE_32, 0, BFD_RELOC_X86_64_TPOFF32 } },
3694 { "NTPOFF", { BFD_RELOC_386_TLS_LE, 0, 0 } },
3695 { "DTPOFF", { BFD_RELOC_386_TLS_LDO_32, 0, BFD_RELOC_X86_64_DTPOFF32 } },
3696 { "GOTNTPOFF",{ BFD_RELOC_386_TLS_GOTIE, 0, 0 } },
3697 { "INDNTPOFF",{ BFD_RELOC_386_TLS_IE, 0, 0 } },
3698 { "GOT", { BFD_RELOC_386_GOT32, 0, BFD_RELOC_X86_64_GOT32 } }
3700 char *cp;
3701 unsigned int j;
3703 for (cp = input_line_pointer; *cp != '@'; cp++)
3704 if (is_end_of_line[(unsigned char) *cp])
3705 return NULL;
3707 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
3709 int len;
3711 len = strlen (gotrel[j].str);
3712 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
3714 if (gotrel[j].rel[(unsigned int) flag_code] != 0)
3716 int first, second;
3717 char *tmpbuf, *past_reloc;
3719 *reloc = gotrel[j].rel[(unsigned int) flag_code];
3720 if (adjust)
3721 *adjust = len;
3723 if (GOT_symbol == NULL)
3724 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3726 /* Replace the relocation token with ' ', so that
3727 errors like foo@GOTOFF1 will be detected. */
3729 /* The length of the first part of our input line. */
3730 first = cp - input_line_pointer;
3732 /* The second part goes from after the reloc token until
3733 (and including) an end_of_line char. Don't use strlen
3734 here as the end_of_line char may not be a NUL. */
3735 past_reloc = cp + 1 + len;
3736 for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
3738 second = cp - past_reloc;
3740 /* Allocate and copy string. The trailing NUL shouldn't
3741 be necessary, but be safe. */
3742 tmpbuf = xmalloc (first + second + 2);
3743 memcpy (tmpbuf, input_line_pointer, first);
3744 tmpbuf[first] = ' ';
3745 memcpy (tmpbuf + first + 1, past_reloc, second);
3746 tmpbuf[first + second + 1] = '\0';
3747 return tmpbuf;
3750 as_bad (_("@%s reloc is not supported in %s bit mode"),
3751 gotrel[j].str, mode_name[(unsigned int) flag_code]);
3752 return NULL;
3756 /* Might be a symbol version string. Don't as_bad here. */
3757 return NULL;
3760 /* x86_cons_fix_new is called via the expression parsing code when a
3761 reloc is needed. We use this hook to get the correct .got reloc. */
3762 static enum bfd_reloc_code_real got_reloc = NO_RELOC;
3764 void
3765 x86_cons_fix_new (frag, off, len, exp)
3766 fragS *frag;
3767 unsigned int off;
3768 unsigned int len;
3769 expressionS *exp;
3771 enum bfd_reloc_code_real r = reloc (len, 0, 0, got_reloc);
3772 got_reloc = NO_RELOC;
3773 fix_new_exp (frag, off, len, exp, 0, r);
3776 void
3777 x86_cons (exp, size)
3778 expressionS *exp;
3779 int size;
3781 if (size == 4)
3783 /* Handle @GOTOFF and the like in an expression. */
3784 char *save;
3785 char *gotfree_input_line;
3786 int adjust;
3788 save = input_line_pointer;
3789 gotfree_input_line = lex_got (&got_reloc, &adjust);
3790 if (gotfree_input_line)
3791 input_line_pointer = gotfree_input_line;
3793 expression (exp);
3795 if (gotfree_input_line)
3797 /* expression () has merrily parsed up to the end of line,
3798 or a comma - in the wrong buffer. Transfer how far
3799 input_line_pointer has moved to the right buffer. */
3800 input_line_pointer = (save
3801 + (input_line_pointer - gotfree_input_line)
3802 + adjust);
3803 free (gotfree_input_line);
3806 else
3807 expression (exp);
3809 #endif
3811 #ifdef TE_PE
3813 void
3814 x86_pe_cons_fix_new (frag, off, len, exp)
3815 fragS *frag;
3816 unsigned int off;
3817 unsigned int len;
3818 expressionS *exp;
3820 enum bfd_reloc_code_real r = reloc (len, 0, 0, NO_RELOC);
3822 if (exp->X_op == O_secrel)
3824 exp->X_op = O_symbol;
3825 r = BFD_RELOC_32_SECREL;
3828 fix_new_exp (frag, off, len, exp, 0, r);
3831 static void
3832 pe_directive_secrel (dummy)
3833 int dummy ATTRIBUTE_UNUSED;
3835 expressionS exp;
3839 expression (&exp);
3840 if (exp.X_op == O_symbol)
3841 exp.X_op = O_secrel;
3843 emit_expr (&exp, 4);
3845 while (*input_line_pointer++ == ',');
3847 input_line_pointer--;
3848 demand_empty_rest_of_line ();
3851 #endif
3853 static int i386_immediate PARAMS ((char *));
3855 static int
3856 i386_immediate (imm_start)
3857 char *imm_start;
3859 char *save_input_line_pointer;
3860 #ifndef LEX_AT
3861 char *gotfree_input_line;
3862 #endif
3863 segT exp_seg = 0;
3864 expressionS *exp;
3866 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
3868 as_bad (_("only 1 or 2 immediate operands are allowed"));
3869 return 0;
3872 exp = &im_expressions[i.imm_operands++];
3873 i.op[this_operand].imms = exp;
3875 if (is_space_char (*imm_start))
3876 ++imm_start;
3878 save_input_line_pointer = input_line_pointer;
3879 input_line_pointer = imm_start;
3881 #ifndef LEX_AT
3882 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3883 if (gotfree_input_line)
3884 input_line_pointer = gotfree_input_line;
3885 #endif
3887 exp_seg = expression (exp);
3889 SKIP_WHITESPACE ();
3890 if (*input_line_pointer)
3891 as_bad (_("junk `%s' after expression"), input_line_pointer);
3893 input_line_pointer = save_input_line_pointer;
3894 #ifndef LEX_AT
3895 if (gotfree_input_line)
3896 free (gotfree_input_line);
3897 #endif
3899 if (exp->X_op == O_absent || exp->X_op == O_big)
3901 /* Missing or bad expr becomes absolute 0. */
3902 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
3903 imm_start);
3904 exp->X_op = O_constant;
3905 exp->X_add_number = 0;
3906 exp->X_add_symbol = (symbolS *) 0;
3907 exp->X_op_symbol = (symbolS *) 0;
3909 else if (exp->X_op == O_constant)
3911 /* Size it properly later. */
3912 i.types[this_operand] |= Imm64;
3913 /* If BFD64, sign extend val. */
3914 if (!use_rela_relocations)
3915 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
3916 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
3918 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3919 else if (OUTPUT_FLAVOR == bfd_target_aout_flavour
3920 && exp_seg != absolute_section
3921 && exp_seg != text_section
3922 && exp_seg != data_section
3923 && exp_seg != bss_section
3924 && exp_seg != undefined_section
3925 && !bfd_is_com_section (exp_seg))
3927 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3928 return 0;
3930 #endif
3931 else
3933 /* This is an address. The size of the address will be
3934 determined later, depending on destination register,
3935 suffix, or the default for the section. */
3936 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
3939 return 1;
3942 static char *i386_scale PARAMS ((char *));
3944 static char *
3945 i386_scale (scale)
3946 char *scale;
3948 offsetT val;
3949 char *save = input_line_pointer;
3951 input_line_pointer = scale;
3952 val = get_absolute_expression ();
3954 switch (val)
3956 case 1:
3957 i.log2_scale_factor = 0;
3958 break;
3959 case 2:
3960 i.log2_scale_factor = 1;
3961 break;
3962 case 4:
3963 i.log2_scale_factor = 2;
3964 break;
3965 case 8:
3966 i.log2_scale_factor = 3;
3967 break;
3968 default:
3970 char sep = *input_line_pointer;
3972 *input_line_pointer = '\0';
3973 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
3974 scale);
3975 *input_line_pointer = sep;
3976 input_line_pointer = save;
3977 return NULL;
3980 if (i.log2_scale_factor != 0 && i.index_reg == 0)
3982 as_warn (_("scale factor of %d without an index register"),
3983 1 << i.log2_scale_factor);
3984 #if SCALE1_WHEN_NO_INDEX
3985 i.log2_scale_factor = 0;
3986 #endif
3988 scale = input_line_pointer;
3989 input_line_pointer = save;
3990 return scale;
3993 static int i386_displacement PARAMS ((char *, char *));
3995 static int
3996 i386_displacement (disp_start, disp_end)
3997 char *disp_start;
3998 char *disp_end;
4000 expressionS *exp;
4001 segT exp_seg = 0;
4002 char *save_input_line_pointer;
4003 #ifndef LEX_AT
4004 char *gotfree_input_line;
4005 #endif
4006 int bigdisp = Disp32;
4008 if (flag_code == CODE_64BIT)
4010 if (i.prefix[ADDR_PREFIX] == 0)
4011 bigdisp = Disp64;
4013 else if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
4014 bigdisp = Disp16;
4015 i.types[this_operand] |= bigdisp;
4017 exp = &disp_expressions[i.disp_operands];
4018 i.op[this_operand].disps = exp;
4019 i.disp_operands++;
4020 save_input_line_pointer = input_line_pointer;
4021 input_line_pointer = disp_start;
4022 END_STRING_AND_SAVE (disp_end);
4024 #ifndef GCC_ASM_O_HACK
4025 #define GCC_ASM_O_HACK 0
4026 #endif
4027 #if GCC_ASM_O_HACK
4028 END_STRING_AND_SAVE (disp_end + 1);
4029 if ((i.types[this_operand] & BaseIndex) != 0
4030 && displacement_string_end[-1] == '+')
4032 /* This hack is to avoid a warning when using the "o"
4033 constraint within gcc asm statements.
4034 For instance:
4036 #define _set_tssldt_desc(n,addr,limit,type) \
4037 __asm__ __volatile__ ( \
4038 "movw %w2,%0\n\t" \
4039 "movw %w1,2+%0\n\t" \
4040 "rorl $16,%1\n\t" \
4041 "movb %b1,4+%0\n\t" \
4042 "movb %4,5+%0\n\t" \
4043 "movb $0,6+%0\n\t" \
4044 "movb %h1,7+%0\n\t" \
4045 "rorl $16,%1" \
4046 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
4048 This works great except that the output assembler ends
4049 up looking a bit weird if it turns out that there is
4050 no offset. You end up producing code that looks like:
4052 #APP
4053 movw $235,(%eax)
4054 movw %dx,2+(%eax)
4055 rorl $16,%edx
4056 movb %dl,4+(%eax)
4057 movb $137,5+(%eax)
4058 movb $0,6+(%eax)
4059 movb %dh,7+(%eax)
4060 rorl $16,%edx
4061 #NO_APP
4063 So here we provide the missing zero. */
4065 *displacement_string_end = '0';
4067 #endif
4068 #ifndef LEX_AT
4069 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
4070 if (gotfree_input_line)
4071 input_line_pointer = gotfree_input_line;
4072 #endif
4074 exp_seg = expression (exp);
4076 SKIP_WHITESPACE ();
4077 if (*input_line_pointer)
4078 as_bad (_("junk `%s' after expression"), input_line_pointer);
4079 #if GCC_ASM_O_HACK
4080 RESTORE_END_STRING (disp_end + 1);
4081 #endif
4082 RESTORE_END_STRING (disp_end);
4083 input_line_pointer = save_input_line_pointer;
4084 #ifndef LEX_AT
4085 if (gotfree_input_line)
4086 free (gotfree_input_line);
4087 #endif
4089 /* We do this to make sure that the section symbol is in
4090 the symbol table. We will ultimately change the relocation
4091 to be relative to the beginning of the section. */
4092 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
4093 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
4095 if (exp->X_op != O_symbol)
4097 as_bad (_("bad expression used with @%s"),
4098 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
4099 ? "GOTPCREL"
4100 : "GOTOFF"));
4101 return 0;
4104 if (S_IS_LOCAL (exp->X_add_symbol)
4105 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
4106 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
4107 exp->X_op = O_subtract;
4108 exp->X_op_symbol = GOT_symbol;
4109 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
4110 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
4111 else
4112 i.reloc[this_operand] = BFD_RELOC_32;
4115 if (exp->X_op == O_absent || exp->X_op == O_big)
4117 /* Missing or bad expr becomes absolute 0. */
4118 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
4119 disp_start);
4120 exp->X_op = O_constant;
4121 exp->X_add_number = 0;
4122 exp->X_add_symbol = (symbolS *) 0;
4123 exp->X_op_symbol = (symbolS *) 0;
4126 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4127 if (exp->X_op != O_constant
4128 && OUTPUT_FLAVOR == bfd_target_aout_flavour
4129 && exp_seg != absolute_section
4130 && exp_seg != text_section
4131 && exp_seg != data_section
4132 && exp_seg != bss_section
4133 && exp_seg != undefined_section
4134 && !bfd_is_com_section (exp_seg))
4136 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
4137 return 0;
4139 #endif
4140 else if (flag_code == CODE_64BIT)
4141 i.types[this_operand] |= Disp32S | Disp32;
4142 return 1;
4145 static int i386_index_check PARAMS ((const char *));
4147 /* Make sure the memory operand we've been dealt is valid.
4148 Return 1 on success, 0 on a failure. */
4150 static int
4151 i386_index_check (operand_string)
4152 const char *operand_string;
4154 int ok;
4155 #if INFER_ADDR_PREFIX
4156 int fudged = 0;
4158 tryprefix:
4159 #endif
4160 ok = 1;
4161 if (flag_code == CODE_64BIT)
4163 unsigned RegXX = (i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32);
4165 if ((i.base_reg
4166 && ((i.base_reg->reg_type & RegXX) == 0)
4167 && (i.base_reg->reg_type != BaseIndex
4168 || i.index_reg))
4169 || (i.index_reg
4170 && ((i.index_reg->reg_type & (RegXX | BaseIndex))
4171 != (RegXX | BaseIndex))))
4172 ok = 0;
4174 else
4176 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
4178 /* 16bit checks. */
4179 if ((i.base_reg
4180 && ((i.base_reg->reg_type & (Reg16 | BaseIndex | RegRex))
4181 != (Reg16 | BaseIndex)))
4182 || (i.index_reg
4183 && (((i.index_reg->reg_type & (Reg16 | BaseIndex))
4184 != (Reg16 | BaseIndex))
4185 || !(i.base_reg
4186 && i.base_reg->reg_num < 6
4187 && i.index_reg->reg_num >= 6
4188 && i.log2_scale_factor == 0))))
4189 ok = 0;
4191 else
4193 /* 32bit checks. */
4194 if ((i.base_reg
4195 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
4196 || (i.index_reg
4197 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
4198 != (Reg32 | BaseIndex))))
4199 ok = 0;
4202 if (!ok)
4204 #if INFER_ADDR_PREFIX
4205 if (i.prefix[ADDR_PREFIX] == 0)
4207 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
4208 i.prefixes += 1;
4209 /* Change the size of any displacement too. At most one of
4210 Disp16 or Disp32 is set.
4211 FIXME. There doesn't seem to be any real need for separate
4212 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
4213 Removing them would probably clean up the code quite a lot. */
4214 if (flag_code != CODE_64BIT && (i.types[this_operand] & (Disp16 | Disp32)))
4215 i.types[this_operand] ^= (Disp16 | Disp32);
4216 fudged = 1;
4217 goto tryprefix;
4219 if (fudged)
4220 as_bad (_("`%s' is not a valid base/index expression"),
4221 operand_string);
4222 else
4223 #endif
4224 as_bad (_("`%s' is not a valid %s bit base/index expression"),
4225 operand_string,
4226 flag_code_names[flag_code]);
4228 return ok;
4231 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
4232 on error. */
4234 static int
4235 i386_operand (operand_string)
4236 char *operand_string;
4238 const reg_entry *r;
4239 char *end_op;
4240 char *op_string = operand_string;
4242 if (is_space_char (*op_string))
4243 ++op_string;
4245 /* We check for an absolute prefix (differentiating,
4246 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
4247 if (*op_string == ABSOLUTE_PREFIX)
4249 ++op_string;
4250 if (is_space_char (*op_string))
4251 ++op_string;
4252 i.types[this_operand] |= JumpAbsolute;
4255 /* Check if operand is a register. */
4256 if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
4257 && (r = parse_register (op_string, &end_op)) != NULL)
4259 /* Check for a segment override by searching for ':' after a
4260 segment register. */
4261 op_string = end_op;
4262 if (is_space_char (*op_string))
4263 ++op_string;
4264 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
4266 switch (r->reg_num)
4268 case 0:
4269 i.seg[i.mem_operands] = &es;
4270 break;
4271 case 1:
4272 i.seg[i.mem_operands] = &cs;
4273 break;
4274 case 2:
4275 i.seg[i.mem_operands] = &ss;
4276 break;
4277 case 3:
4278 i.seg[i.mem_operands] = &ds;
4279 break;
4280 case 4:
4281 i.seg[i.mem_operands] = &fs;
4282 break;
4283 case 5:
4284 i.seg[i.mem_operands] = &gs;
4285 break;
4288 /* Skip the ':' and whitespace. */
4289 ++op_string;
4290 if (is_space_char (*op_string))
4291 ++op_string;
4293 if (!is_digit_char (*op_string)
4294 && !is_identifier_char (*op_string)
4295 && *op_string != '('
4296 && *op_string != ABSOLUTE_PREFIX)
4298 as_bad (_("bad memory operand `%s'"), op_string);
4299 return 0;
4301 /* Handle case of %es:*foo. */
4302 if (*op_string == ABSOLUTE_PREFIX)
4304 ++op_string;
4305 if (is_space_char (*op_string))
4306 ++op_string;
4307 i.types[this_operand] |= JumpAbsolute;
4309 goto do_memory_reference;
4311 if (*op_string)
4313 as_bad (_("junk `%s' after register"), op_string);
4314 return 0;
4316 i.types[this_operand] |= r->reg_type & ~BaseIndex;
4317 i.op[this_operand].regs = r;
4318 i.reg_operands++;
4320 else if (*op_string == REGISTER_PREFIX)
4322 as_bad (_("bad register name `%s'"), op_string);
4323 return 0;
4325 else if (*op_string == IMMEDIATE_PREFIX)
4327 ++op_string;
4328 if (i.types[this_operand] & JumpAbsolute)
4330 as_bad (_("immediate operand illegal with absolute jump"));
4331 return 0;
4333 if (!i386_immediate (op_string))
4334 return 0;
4336 else if (is_digit_char (*op_string)
4337 || is_identifier_char (*op_string)
4338 || *op_string == '(')
4340 /* This is a memory reference of some sort. */
4341 char *base_string;
4343 /* Start and end of displacement string expression (if found). */
4344 char *displacement_string_start;
4345 char *displacement_string_end;
4347 do_memory_reference:
4348 if ((i.mem_operands == 1
4349 && (current_templates->start->opcode_modifier & IsString) == 0)
4350 || i.mem_operands == 2)
4352 as_bad (_("too many memory references for `%s'"),
4353 current_templates->start->name);
4354 return 0;
4357 /* Check for base index form. We detect the base index form by
4358 looking for an ')' at the end of the operand, searching
4359 for the '(' matching it, and finding a REGISTER_PREFIX or ','
4360 after the '('. */
4361 base_string = op_string + strlen (op_string);
4363 --base_string;
4364 if (is_space_char (*base_string))
4365 --base_string;
4367 /* If we only have a displacement, set-up for it to be parsed later. */
4368 displacement_string_start = op_string;
4369 displacement_string_end = base_string + 1;
4371 if (*base_string == ')')
4373 char *temp_string;
4374 unsigned int parens_balanced = 1;
4375 /* We've already checked that the number of left & right ()'s are
4376 equal, so this loop will not be infinite. */
4379 base_string--;
4380 if (*base_string == ')')
4381 parens_balanced++;
4382 if (*base_string == '(')
4383 parens_balanced--;
4385 while (parens_balanced);
4387 temp_string = base_string;
4389 /* Skip past '(' and whitespace. */
4390 ++base_string;
4391 if (is_space_char (*base_string))
4392 ++base_string;
4394 if (*base_string == ','
4395 || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
4396 && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
4398 displacement_string_end = temp_string;
4400 i.types[this_operand] |= BaseIndex;
4402 if (i.base_reg)
4404 base_string = end_op;
4405 if (is_space_char (*base_string))
4406 ++base_string;
4409 /* There may be an index reg or scale factor here. */
4410 if (*base_string == ',')
4412 ++base_string;
4413 if (is_space_char (*base_string))
4414 ++base_string;
4416 if ((*base_string == REGISTER_PREFIX || allow_naked_reg)
4417 && (i.index_reg = parse_register (base_string, &end_op)) != NULL)
4419 base_string = end_op;
4420 if (is_space_char (*base_string))
4421 ++base_string;
4422 if (*base_string == ',')
4424 ++base_string;
4425 if (is_space_char (*base_string))
4426 ++base_string;
4428 else if (*base_string != ')')
4430 as_bad (_("expecting `,' or `)' after index register in `%s'"),
4431 operand_string);
4432 return 0;
4435 else if (*base_string == REGISTER_PREFIX)
4437 as_bad (_("bad register name `%s'"), base_string);
4438 return 0;
4441 /* Check for scale factor. */
4442 if (*base_string != ')')
4444 char *end_scale = i386_scale (base_string);
4446 if (!end_scale)
4447 return 0;
4449 base_string = end_scale;
4450 if (is_space_char (*base_string))
4451 ++base_string;
4452 if (*base_string != ')')
4454 as_bad (_("expecting `)' after scale factor in `%s'"),
4455 operand_string);
4456 return 0;
4459 else if (!i.index_reg)
4461 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
4462 *base_string);
4463 return 0;
4466 else if (*base_string != ')')
4468 as_bad (_("expecting `,' or `)' after base register in `%s'"),
4469 operand_string);
4470 return 0;
4473 else if (*base_string == REGISTER_PREFIX)
4475 as_bad (_("bad register name `%s'"), base_string);
4476 return 0;
4480 /* If there's an expression beginning the operand, parse it,
4481 assuming displacement_string_start and
4482 displacement_string_end are meaningful. */
4483 if (displacement_string_start != displacement_string_end)
4485 if (!i386_displacement (displacement_string_start,
4486 displacement_string_end))
4487 return 0;
4490 /* Special case for (%dx) while doing input/output op. */
4491 if (i.base_reg
4492 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
4493 && i.index_reg == 0
4494 && i.log2_scale_factor == 0
4495 && i.seg[i.mem_operands] == 0
4496 && (i.types[this_operand] & Disp) == 0)
4498 i.types[this_operand] = InOutPortReg;
4499 return 1;
4502 if (i386_index_check (operand_string) == 0)
4503 return 0;
4504 i.mem_operands++;
4506 else
4508 /* It's not a memory operand; argh! */
4509 as_bad (_("invalid char %s beginning operand %d `%s'"),
4510 output_invalid (*op_string),
4511 this_operand + 1,
4512 op_string);
4513 return 0;
4515 return 1; /* Normal return. */
4518 /* md_estimate_size_before_relax()
4520 Called just before relax() for rs_machine_dependent frags. The x86
4521 assembler uses these frags to handle variable size jump
4522 instructions.
4524 Any symbol that is now undefined will not become defined.
4525 Return the correct fr_subtype in the frag.
4526 Return the initial "guess for variable size of frag" to caller.
4527 The guess is actually the growth beyond the fixed part. Whatever
4528 we do to grow the fixed or variable part contributes to our
4529 returned value. */
4532 md_estimate_size_before_relax (fragP, segment)
4533 fragS *fragP;
4534 segT segment;
4536 /* We've already got fragP->fr_subtype right; all we have to do is
4537 check for un-relaxable symbols. On an ELF system, we can't relax
4538 an externally visible symbol, because it may be overridden by a
4539 shared library. */
4540 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
4541 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4542 || (OUTPUT_FLAVOR == bfd_target_elf_flavour
4543 && (S_IS_EXTERNAL (fragP->fr_symbol)
4544 || S_IS_WEAK (fragP->fr_symbol)))
4545 #endif
4548 /* Symbol is undefined in this segment, or we need to keep a
4549 reloc so that weak symbols can be overridden. */
4550 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
4551 enum bfd_reloc_code_real reloc_type;
4552 unsigned char *opcode;
4553 int old_fr_fix;
4555 if (fragP->fr_var != NO_RELOC)
4556 reloc_type = fragP->fr_var;
4557 else if (size == 2)
4558 reloc_type = BFD_RELOC_16_PCREL;
4559 else
4560 reloc_type = BFD_RELOC_32_PCREL;
4562 old_fr_fix = fragP->fr_fix;
4563 opcode = (unsigned char *) fragP->fr_opcode;
4565 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
4567 case UNCOND_JUMP:
4568 /* Make jmp (0xeb) a (d)word displacement jump. */
4569 opcode[0] = 0xe9;
4570 fragP->fr_fix += size;
4571 fix_new (fragP, old_fr_fix, size,
4572 fragP->fr_symbol,
4573 fragP->fr_offset, 1,
4574 reloc_type);
4575 break;
4577 case COND_JUMP86:
4578 if (size == 2
4579 && (!no_cond_jump_promotion || fragP->fr_var != NO_RELOC))
4581 /* Negate the condition, and branch past an
4582 unconditional jump. */
4583 opcode[0] ^= 1;
4584 opcode[1] = 3;
4585 /* Insert an unconditional jump. */
4586 opcode[2] = 0xe9;
4587 /* We added two extra opcode bytes, and have a two byte
4588 offset. */
4589 fragP->fr_fix += 2 + 2;
4590 fix_new (fragP, old_fr_fix + 2, 2,
4591 fragP->fr_symbol,
4592 fragP->fr_offset, 1,
4593 reloc_type);
4594 break;
4596 /* Fall through. */
4598 case COND_JUMP:
4599 if (no_cond_jump_promotion && fragP->fr_var == NO_RELOC)
4601 fixS *fixP;
4603 fragP->fr_fix += 1;
4604 fixP = fix_new (fragP, old_fr_fix, 1,
4605 fragP->fr_symbol,
4606 fragP->fr_offset, 1,
4607 BFD_RELOC_8_PCREL);
4608 fixP->fx_signed = 1;
4609 break;
4612 /* This changes the byte-displacement jump 0x7N
4613 to the (d)word-displacement jump 0x0f,0x8N. */
4614 opcode[1] = opcode[0] + 0x10;
4615 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4616 /* We've added an opcode byte. */
4617 fragP->fr_fix += 1 + size;
4618 fix_new (fragP, old_fr_fix + 1, size,
4619 fragP->fr_symbol,
4620 fragP->fr_offset, 1,
4621 reloc_type);
4622 break;
4624 default:
4625 BAD_CASE (fragP->fr_subtype);
4626 break;
4628 frag_wane (fragP);
4629 return fragP->fr_fix - old_fr_fix;
4632 /* Guess size depending on current relax state. Initially the relax
4633 state will correspond to a short jump and we return 1, because
4634 the variable part of the frag (the branch offset) is one byte
4635 long. However, we can relax a section more than once and in that
4636 case we must either set fr_subtype back to the unrelaxed state,
4637 or return the value for the appropriate branch. */
4638 return md_relax_table[fragP->fr_subtype].rlx_length;
4641 /* Called after relax() is finished.
4643 In: Address of frag.
4644 fr_type == rs_machine_dependent.
4645 fr_subtype is what the address relaxed to.
4647 Out: Any fixSs and constants are set up.
4648 Caller will turn frag into a ".space 0". */
4650 void
4651 md_convert_frag (abfd, sec, fragP)
4652 bfd *abfd ATTRIBUTE_UNUSED;
4653 segT sec ATTRIBUTE_UNUSED;
4654 fragS *fragP;
4656 unsigned char *opcode;
4657 unsigned char *where_to_put_displacement = NULL;
4658 offsetT target_address;
4659 offsetT opcode_address;
4660 unsigned int extension = 0;
4661 offsetT displacement_from_opcode_start;
4663 opcode = (unsigned char *) fragP->fr_opcode;
4665 /* Address we want to reach in file space. */
4666 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
4668 /* Address opcode resides at in file space. */
4669 opcode_address = fragP->fr_address + fragP->fr_fix;
4671 /* Displacement from opcode start to fill into instruction. */
4672 displacement_from_opcode_start = target_address - opcode_address;
4674 if ((fragP->fr_subtype & BIG) == 0)
4676 /* Don't have to change opcode. */
4677 extension = 1; /* 1 opcode + 1 displacement */
4678 where_to_put_displacement = &opcode[1];
4680 else
4682 if (no_cond_jump_promotion
4683 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
4684 as_warn_where (fragP->fr_file, fragP->fr_line, _("long jump required"));
4686 switch (fragP->fr_subtype)
4688 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
4689 extension = 4; /* 1 opcode + 4 displacement */
4690 opcode[0] = 0xe9;
4691 where_to_put_displacement = &opcode[1];
4692 break;
4694 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
4695 extension = 2; /* 1 opcode + 2 displacement */
4696 opcode[0] = 0xe9;
4697 where_to_put_displacement = &opcode[1];
4698 break;
4700 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
4701 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
4702 extension = 5; /* 2 opcode + 4 displacement */
4703 opcode[1] = opcode[0] + 0x10;
4704 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4705 where_to_put_displacement = &opcode[2];
4706 break;
4708 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
4709 extension = 3; /* 2 opcode + 2 displacement */
4710 opcode[1] = opcode[0] + 0x10;
4711 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4712 where_to_put_displacement = &opcode[2];
4713 break;
4715 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
4716 extension = 4;
4717 opcode[0] ^= 1;
4718 opcode[1] = 3;
4719 opcode[2] = 0xe9;
4720 where_to_put_displacement = &opcode[3];
4721 break;
4723 default:
4724 BAD_CASE (fragP->fr_subtype);
4725 break;
4729 /* Now put displacement after opcode. */
4730 md_number_to_chars ((char *) where_to_put_displacement,
4731 (valueT) (displacement_from_opcode_start - extension),
4732 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
4733 fragP->fr_fix += extension;
4736 /* Size of byte displacement jmp. */
4737 int md_short_jump_size = 2;
4739 /* Size of dword displacement jmp. */
4740 int md_long_jump_size = 5;
4742 /* Size of relocation record. */
4743 const int md_reloc_size = 8;
4745 void
4746 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
4747 char *ptr;
4748 addressT from_addr, to_addr;
4749 fragS *frag ATTRIBUTE_UNUSED;
4750 symbolS *to_symbol ATTRIBUTE_UNUSED;
4752 offsetT offset;
4754 offset = to_addr - (from_addr + 2);
4755 /* Opcode for byte-disp jump. */
4756 md_number_to_chars (ptr, (valueT) 0xeb, 1);
4757 md_number_to_chars (ptr + 1, (valueT) offset, 1);
4760 void
4761 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
4762 char *ptr;
4763 addressT from_addr, to_addr;
4764 fragS *frag ATTRIBUTE_UNUSED;
4765 symbolS *to_symbol ATTRIBUTE_UNUSED;
4767 offsetT offset;
4769 offset = to_addr - (from_addr + 5);
4770 md_number_to_chars (ptr, (valueT) 0xe9, 1);
4771 md_number_to_chars (ptr + 1, (valueT) offset, 4);
4774 /* Apply a fixup (fixS) to segment data, once it has been determined
4775 by our caller that we have all the info we need to fix it up.
4777 On the 386, immediates, displacements, and data pointers are all in
4778 the same (little-endian) format, so we don't need to care about which
4779 we are handling. */
4781 void
4782 md_apply_fix3 (fixP, valP, seg)
4783 /* The fix we're to put in. */
4784 fixS *fixP;
4785 /* Pointer to the value of the bits. */
4786 valueT *valP;
4787 /* Segment fix is from. */
4788 segT seg ATTRIBUTE_UNUSED;
4790 char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
4791 valueT value = *valP;
4793 #if !defined (TE_Mach)
4794 if (fixP->fx_pcrel)
4796 switch (fixP->fx_r_type)
4798 default:
4799 break;
4801 case BFD_RELOC_32:
4802 case BFD_RELOC_X86_64_32S:
4803 fixP->fx_r_type = BFD_RELOC_32_PCREL;
4804 break;
4805 case BFD_RELOC_16:
4806 fixP->fx_r_type = BFD_RELOC_16_PCREL;
4807 break;
4808 case BFD_RELOC_8:
4809 fixP->fx_r_type = BFD_RELOC_8_PCREL;
4810 break;
4814 if (fixP->fx_addsy != NULL
4815 && (fixP->fx_r_type == BFD_RELOC_32_PCREL
4816 || fixP->fx_r_type == BFD_RELOC_16_PCREL
4817 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
4818 && !use_rela_relocations)
4820 /* This is a hack. There should be a better way to handle this.
4821 This covers for the fact that bfd_install_relocation will
4822 subtract the current location (for partial_inplace, PC relative
4823 relocations); see more below. */
4824 #ifndef OBJ_AOUT
4825 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4826 #ifdef TE_PE
4827 || OUTPUT_FLAVOR == bfd_target_coff_flavour
4828 #endif
4830 value += fixP->fx_where + fixP->fx_frag->fr_address;
4831 #endif
4832 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4833 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
4835 segT sym_seg = S_GET_SEGMENT (fixP->fx_addsy);
4837 if ((sym_seg == seg
4838 || (symbol_section_p (fixP->fx_addsy)
4839 && sym_seg != absolute_section))
4840 && !generic_force_reloc (fixP))
4842 /* Yes, we add the values in twice. This is because
4843 bfd_install_relocation subtracts them out again. I think
4844 bfd_install_relocation is broken, but I don't dare change
4845 it. FIXME. */
4846 value += fixP->fx_where + fixP->fx_frag->fr_address;
4849 #endif
4850 #if defined (OBJ_COFF) && defined (TE_PE)
4851 /* For some reason, the PE format does not store a
4852 section address offset for a PC relative symbol. */
4853 if (S_GET_SEGMENT (fixP->fx_addsy) != seg
4854 #if defined(BFD_ASSEMBLER) || defined(S_IS_WEAK)
4855 || S_IS_WEAK (fixP->fx_addsy)
4856 #endif
4858 value += md_pcrel_from (fixP);
4859 #endif
4862 /* Fix a few things - the dynamic linker expects certain values here,
4863 and we must not disappoint it. */
4864 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4865 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4866 && fixP->fx_addsy)
4867 switch (fixP->fx_r_type)
4869 case BFD_RELOC_386_PLT32:
4870 case BFD_RELOC_X86_64_PLT32:
4871 /* Make the jump instruction point to the address of the operand. At
4872 runtime we merely add the offset to the actual PLT entry. */
4873 value = -4;
4874 break;
4876 case BFD_RELOC_386_TLS_GD:
4877 case BFD_RELOC_386_TLS_LDM:
4878 case BFD_RELOC_386_TLS_IE_32:
4879 case BFD_RELOC_386_TLS_IE:
4880 case BFD_RELOC_386_TLS_GOTIE:
4881 case BFD_RELOC_X86_64_TLSGD:
4882 case BFD_RELOC_X86_64_TLSLD:
4883 case BFD_RELOC_X86_64_GOTTPOFF:
4884 value = 0; /* Fully resolved at runtime. No addend. */
4885 /* Fallthrough */
4886 case BFD_RELOC_386_TLS_LE:
4887 case BFD_RELOC_386_TLS_LDO_32:
4888 case BFD_RELOC_386_TLS_LE_32:
4889 case BFD_RELOC_X86_64_DTPOFF32:
4890 case BFD_RELOC_X86_64_TPOFF32:
4891 S_SET_THREAD_LOCAL (fixP->fx_addsy);
4892 break;
4894 case BFD_RELOC_386_GOT32:
4895 case BFD_RELOC_X86_64_GOT32:
4896 value = 0; /* Fully resolved at runtime. No addend. */
4897 break;
4899 case BFD_RELOC_VTABLE_INHERIT:
4900 case BFD_RELOC_VTABLE_ENTRY:
4901 fixP->fx_done = 0;
4902 return;
4904 default:
4905 break;
4907 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
4908 *valP = value;
4909 #endif /* !defined (TE_Mach) */
4911 /* Are we finished with this relocation now? */
4912 if (fixP->fx_addsy == NULL)
4913 fixP->fx_done = 1;
4914 else if (use_rela_relocations)
4916 fixP->fx_no_overflow = 1;
4917 /* Remember value for tc_gen_reloc. */
4918 fixP->fx_addnumber = value;
4919 value = 0;
4922 md_number_to_chars (p, value, fixP->fx_size);
4925 #define MAX_LITTLENUMS 6
4927 /* Turn the string pointed to by litP into a floating point constant
4928 of type TYPE, and emit the appropriate bytes. The number of
4929 LITTLENUMS emitted is stored in *SIZEP. An error message is
4930 returned, or NULL on OK. */
4932 char *
4933 md_atof (type, litP, sizeP)
4934 int type;
4935 char *litP;
4936 int *sizeP;
4938 int prec;
4939 LITTLENUM_TYPE words[MAX_LITTLENUMS];
4940 LITTLENUM_TYPE *wordP;
4941 char *t;
4943 switch (type)
4945 case 'f':
4946 case 'F':
4947 prec = 2;
4948 break;
4950 case 'd':
4951 case 'D':
4952 prec = 4;
4953 break;
4955 case 'x':
4956 case 'X':
4957 prec = 5;
4958 break;
4960 default:
4961 *sizeP = 0;
4962 return _("Bad call to md_atof ()");
4964 t = atof_ieee (input_line_pointer, type, words);
4965 if (t)
4966 input_line_pointer = t;
4968 *sizeP = prec * sizeof (LITTLENUM_TYPE);
4969 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
4970 the bigendian 386. */
4971 for (wordP = words + prec - 1; prec--;)
4973 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
4974 litP += sizeof (LITTLENUM_TYPE);
4976 return 0;
4979 char output_invalid_buf[8];
4981 static char *
4982 output_invalid (c)
4983 int c;
4985 if (ISPRINT (c))
4986 sprintf (output_invalid_buf, "'%c'", c);
4987 else
4988 sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
4989 return output_invalid_buf;
4992 /* REG_STRING starts *before* REGISTER_PREFIX. */
4994 static const reg_entry *
4995 parse_register (reg_string, end_op)
4996 char *reg_string;
4997 char **end_op;
4999 char *s = reg_string;
5000 char *p;
5001 char reg_name_given[MAX_REG_NAME_SIZE + 1];
5002 const reg_entry *r;
5004 /* Skip possible REGISTER_PREFIX and possible whitespace. */
5005 if (*s == REGISTER_PREFIX)
5006 ++s;
5008 if (is_space_char (*s))
5009 ++s;
5011 p = reg_name_given;
5012 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
5014 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
5015 return (const reg_entry *) NULL;
5016 s++;
5019 /* For naked regs, make sure that we are not dealing with an identifier.
5020 This prevents confusing an identifier like `eax_var' with register
5021 `eax'. */
5022 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
5023 return (const reg_entry *) NULL;
5025 *end_op = s;
5027 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
5029 /* Handle floating point regs, allowing spaces in the (i) part. */
5030 if (r == i386_regtab /* %st is first entry of table */)
5032 if (is_space_char (*s))
5033 ++s;
5034 if (*s == '(')
5036 ++s;
5037 if (is_space_char (*s))
5038 ++s;
5039 if (*s >= '0' && *s <= '7')
5041 r = &i386_float_regtab[*s - '0'];
5042 ++s;
5043 if (is_space_char (*s))
5044 ++s;
5045 if (*s == ')')
5047 *end_op = s + 1;
5048 return r;
5051 /* We have "%st(" then garbage. */
5052 return (const reg_entry *) NULL;
5056 if (r != NULL
5057 && ((r->reg_flags & (RegRex64 | RegRex)) | (r->reg_type & Reg64)) != 0
5058 && (r->reg_type != Control || !(cpu_arch_flags & CpuSledgehammer))
5059 && flag_code != CODE_64BIT)
5060 return (const reg_entry *) NULL;
5062 return r;
5065 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5066 const char *md_shortopts = "kVQ:sqn";
5067 #else
5068 const char *md_shortopts = "qn";
5069 #endif
5071 struct option md_longopts[] = {
5072 #define OPTION_32 (OPTION_MD_BASE + 0)
5073 {"32", no_argument, NULL, OPTION_32},
5074 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5075 #define OPTION_64 (OPTION_MD_BASE + 1)
5076 {"64", no_argument, NULL, OPTION_64},
5077 #endif
5078 {NULL, no_argument, NULL, 0}
5080 size_t md_longopts_size = sizeof (md_longopts);
5083 md_parse_option (c, arg)
5084 int c;
5085 char *arg ATTRIBUTE_UNUSED;
5087 switch (c)
5089 case 'n':
5090 optimize_align_code = 0;
5091 break;
5093 case 'q':
5094 quiet_warnings = 1;
5095 break;
5097 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5098 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
5099 should be emitted or not. FIXME: Not implemented. */
5100 case 'Q':
5101 break;
5103 /* -V: SVR4 argument to print version ID. */
5104 case 'V':
5105 print_version_id ();
5106 break;
5108 /* -k: Ignore for FreeBSD compatibility. */
5109 case 'k':
5110 break;
5112 case 's':
5113 /* -s: On i386 Solaris, this tells the native assembler to use
5114 .stab instead of .stab.excl. We always use .stab anyhow. */
5115 break;
5117 case OPTION_64:
5119 const char **list, **l;
5121 list = bfd_target_list ();
5122 for (l = list; *l != NULL; l++)
5123 if (strcmp (*l, "elf64-x86-64") == 0)
5125 default_arch = "x86_64";
5126 break;
5128 if (*l == NULL)
5129 as_fatal (_("No compiled in support for x86_64"));
5130 free (list);
5132 break;
5133 #endif
5135 case OPTION_32:
5136 default_arch = "i386";
5137 break;
5139 default:
5140 return 0;
5142 return 1;
5145 void
5146 md_show_usage (stream)
5147 FILE *stream;
5149 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5150 fprintf (stream, _("\
5151 -Q ignored\n\
5152 -V print assembler version number\n\
5153 -k ignored\n\
5154 -n Do not optimize code alignment\n\
5155 -q quieten some warnings\n\
5156 -s ignored\n"));
5157 #else
5158 fprintf (stream, _("\
5159 -n Do not optimize code alignment\n\
5160 -q quieten some warnings\n"));
5161 #endif
5164 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
5165 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
5167 /* Pick the target format to use. */
5169 const char *
5170 i386_target_format ()
5172 if (!strcmp (default_arch, "x86_64"))
5173 set_code_flag (CODE_64BIT);
5174 else if (!strcmp (default_arch, "i386"))
5175 set_code_flag (CODE_32BIT);
5176 else
5177 as_fatal (_("Unknown architecture"));
5178 switch (OUTPUT_FLAVOR)
5180 #ifdef OBJ_MAYBE_AOUT
5181 case bfd_target_aout_flavour:
5182 return AOUT_TARGET_FORMAT;
5183 #endif
5184 #ifdef OBJ_MAYBE_COFF
5185 case bfd_target_coff_flavour:
5186 return "coff-i386";
5187 #endif
5188 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
5189 case bfd_target_elf_flavour:
5191 if (flag_code == CODE_64BIT)
5192 use_rela_relocations = 1;
5193 return flag_code == CODE_64BIT ? "elf64-x86-64" : ELF_TARGET_FORMAT;
5195 #endif
5196 default:
5197 abort ();
5198 return NULL;
5202 #endif /* OBJ_MAYBE_ more than one */
5204 #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
5205 void i386_elf_emit_arch_note ()
5207 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
5208 && cpu_arch_name != NULL)
5210 char *p;
5211 asection *seg = now_seg;
5212 subsegT subseg = now_subseg;
5213 Elf_Internal_Note i_note;
5214 Elf_External_Note e_note;
5215 asection *note_secp;
5216 int len;
5218 /* Create the .note section. */
5219 note_secp = subseg_new (".note", 0);
5220 bfd_set_section_flags (stdoutput,
5221 note_secp,
5222 SEC_HAS_CONTENTS | SEC_READONLY);
5224 /* Process the arch string. */
5225 len = strlen (cpu_arch_name);
5227 i_note.namesz = len + 1;
5228 i_note.descsz = 0;
5229 i_note.type = NT_ARCH;
5230 p = frag_more (sizeof (e_note.namesz));
5231 md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
5232 p = frag_more (sizeof (e_note.descsz));
5233 md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
5234 p = frag_more (sizeof (e_note.type));
5235 md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
5236 p = frag_more (len + 1);
5237 strcpy (p, cpu_arch_name);
5239 frag_align (2, 0, 0);
5241 subseg_set (seg, subseg);
5244 #endif
5246 symbolS *
5247 md_undefined_symbol (name)
5248 char *name;
5250 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
5251 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
5252 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
5253 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
5255 if (!GOT_symbol)
5257 if (symbol_find (name))
5258 as_bad (_("GOT already in symbol table"));
5259 GOT_symbol = symbol_new (name, undefined_section,
5260 (valueT) 0, &zero_address_frag);
5262 return GOT_symbol;
5264 return 0;
5267 /* Round up a section size to the appropriate boundary. */
5269 valueT
5270 md_section_align (segment, size)
5271 segT segment ATTRIBUTE_UNUSED;
5272 valueT size;
5274 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
5275 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
5277 /* For a.out, force the section size to be aligned. If we don't do
5278 this, BFD will align it for us, but it will not write out the
5279 final bytes of the section. This may be a bug in BFD, but it is
5280 easier to fix it here since that is how the other a.out targets
5281 work. */
5282 int align;
5284 align = bfd_get_section_alignment (stdoutput, segment);
5285 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
5287 #endif
5289 return size;
5292 /* On the i386, PC-relative offsets are relative to the start of the
5293 next instruction. That is, the address of the offset, plus its
5294 size, since the offset is always the last part of the insn. */
5296 long
5297 md_pcrel_from (fixP)
5298 fixS *fixP;
5300 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
5303 #ifndef I386COFF
5305 static void
5306 s_bss (ignore)
5307 int ignore ATTRIBUTE_UNUSED;
5309 int temp;
5311 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5312 if (IS_ELF)
5313 obj_elf_section_change_hook ();
5314 #endif
5315 temp = get_absolute_expression ();
5316 subseg_set (bss_section, (subsegT) temp);
5317 demand_empty_rest_of_line ();
5320 #endif
5322 void
5323 i386_validate_fix (fixp)
5324 fixS *fixp;
5326 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
5328 /* GOTOFF relocation are nonsense in 64bit mode. */
5329 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
5331 if (flag_code != CODE_64BIT)
5332 abort ();
5333 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
5335 else
5337 if (flag_code == CODE_64BIT)
5338 abort ();
5339 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
5341 fixp->fx_subsy = 0;
5345 arelent *
5346 tc_gen_reloc (section, fixp)
5347 asection *section ATTRIBUTE_UNUSED;
5348 fixS *fixp;
5350 arelent *rel;
5351 bfd_reloc_code_real_type code;
5353 switch (fixp->fx_r_type)
5355 case BFD_RELOC_X86_64_PLT32:
5356 case BFD_RELOC_X86_64_GOT32:
5357 case BFD_RELOC_X86_64_GOTPCREL:
5358 case BFD_RELOC_386_PLT32:
5359 case BFD_RELOC_386_GOT32:
5360 case BFD_RELOC_386_GOTOFF:
5361 case BFD_RELOC_386_GOTPC:
5362 case BFD_RELOC_386_TLS_GD:
5363 case BFD_RELOC_386_TLS_LDM:
5364 case BFD_RELOC_386_TLS_LDO_32:
5365 case BFD_RELOC_386_TLS_IE_32:
5366 case BFD_RELOC_386_TLS_IE:
5367 case BFD_RELOC_386_TLS_GOTIE:
5368 case BFD_RELOC_386_TLS_LE_32:
5369 case BFD_RELOC_386_TLS_LE:
5370 case BFD_RELOC_X86_64_TLSGD:
5371 case BFD_RELOC_X86_64_TLSLD:
5372 case BFD_RELOC_X86_64_DTPOFF32:
5373 case BFD_RELOC_X86_64_GOTTPOFF:
5374 case BFD_RELOC_X86_64_TPOFF32:
5375 case BFD_RELOC_RVA:
5376 case BFD_RELOC_VTABLE_ENTRY:
5377 case BFD_RELOC_VTABLE_INHERIT:
5378 #ifdef TE_PE
5379 case BFD_RELOC_32_SECREL:
5380 #endif
5381 code = fixp->fx_r_type;
5382 break;
5383 default:
5384 if (fixp->fx_pcrel)
5386 switch (fixp->fx_size)
5388 default:
5389 as_bad_where (fixp->fx_file, fixp->fx_line,
5390 _("can not do %d byte pc-relative relocation"),
5391 fixp->fx_size);
5392 code = BFD_RELOC_32_PCREL;
5393 break;
5394 case 1: code = BFD_RELOC_8_PCREL; break;
5395 case 2: code = BFD_RELOC_16_PCREL; break;
5396 case 4: code = BFD_RELOC_32_PCREL; break;
5399 else
5401 switch (fixp->fx_size)
5403 default:
5404 as_bad_where (fixp->fx_file, fixp->fx_line,
5405 _("can not do %d byte relocation"),
5406 fixp->fx_size);
5407 code = BFD_RELOC_32;
5408 break;
5409 case 1: code = BFD_RELOC_8; break;
5410 case 2: code = BFD_RELOC_16; break;
5411 case 4: code = BFD_RELOC_32; break;
5412 #ifdef BFD64
5413 case 8: code = BFD_RELOC_64; break;
5414 #endif
5417 break;
5420 if (code == BFD_RELOC_32
5421 && GOT_symbol
5422 && fixp->fx_addsy == GOT_symbol)
5424 /* We don't support GOTPC on 64bit targets. */
5425 if (flag_code == CODE_64BIT)
5426 abort ();
5427 code = BFD_RELOC_386_GOTPC;
5430 rel = (arelent *) xmalloc (sizeof (arelent));
5431 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
5432 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
5434 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
5436 if (!use_rela_relocations)
5438 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
5439 vtable entry to be used in the relocation's section offset. */
5440 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
5441 rel->address = fixp->fx_offset;
5443 rel->addend = 0;
5445 /* Use the rela in 64bit mode. */
5446 else
5448 if (!fixp->fx_pcrel)
5449 rel->addend = fixp->fx_offset;
5450 else
5451 switch (code)
5453 case BFD_RELOC_X86_64_PLT32:
5454 case BFD_RELOC_X86_64_GOT32:
5455 case BFD_RELOC_X86_64_GOTPCREL:
5456 case BFD_RELOC_X86_64_TLSGD:
5457 case BFD_RELOC_X86_64_TLSLD:
5458 case BFD_RELOC_X86_64_GOTTPOFF:
5459 rel->addend = fixp->fx_offset - fixp->fx_size;
5460 break;
5461 default:
5462 rel->addend = (section->vma
5463 - fixp->fx_size
5464 + fixp->fx_addnumber
5465 + md_pcrel_from (fixp));
5466 break;
5470 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
5471 if (rel->howto == NULL)
5473 as_bad_where (fixp->fx_file, fixp->fx_line,
5474 _("cannot represent relocation type %s"),
5475 bfd_get_reloc_code_name (code));
5476 /* Set howto to a garbage value so that we can keep going. */
5477 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
5478 assert (rel->howto != NULL);
5481 return rel;
5485 /* Parse operands using Intel syntax. This implements a recursive descent
5486 parser based on the BNF grammar published in Appendix B of the MASM 6.1
5487 Programmer's Guide.
5489 FIXME: We do not recognize the full operand grammar defined in the MASM
5490 documentation. In particular, all the structure/union and
5491 high-level macro operands are missing.
5493 Uppercase words are terminals, lower case words are non-terminals.
5494 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
5495 bars '|' denote choices. Most grammar productions are implemented in
5496 functions called 'intel_<production>'.
5498 Initial production is 'expr'.
5500 addOp + | -
5502 alpha [a-zA-Z]
5504 binOp & | AND | \| | OR | ^ | XOR
5506 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
5508 constant digits [[ radixOverride ]]
5510 dataType BYTE | WORD | DWORD | FWORD | QWORD | TBYTE | OWORD | XMMWORD
5512 digits decdigit
5513 | digits decdigit
5514 | digits hexdigit
5516 decdigit [0-9]
5518 e04 e04 addOp e05
5519 | e05
5521 e05 e05 binOp e06
5522 | e06
5524 e06 e06 mulOp e09
5525 | e09
5527 e09 OFFSET e10
5528 | SHORT e10
5529 | + e10
5530 | - e10
5531 | ~ e10
5532 | NOT e10
5533 | e09 PTR e10
5534 | e09 : e10
5535 | e10
5537 e10 e10 [ expr ]
5538 | e11
5540 e11 ( expr )
5541 | [ expr ]
5542 | constant
5543 | dataType
5544 | id
5546 | register
5548 => expr expr cmpOp e04
5549 | e04
5551 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
5552 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
5554 hexdigit a | b | c | d | e | f
5555 | A | B | C | D | E | F
5557 id alpha
5558 | id alpha
5559 | id decdigit
5561 mulOp * | / | % | MOD | << | SHL | >> | SHR
5563 quote " | '
5565 register specialRegister
5566 | gpRegister
5567 | byteRegister
5569 segmentRegister CS | DS | ES | FS | GS | SS
5571 specialRegister CR0 | CR2 | CR3 | CR4
5572 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
5573 | TR3 | TR4 | TR5 | TR6 | TR7
5575 We simplify the grammar in obvious places (e.g., register parsing is
5576 done by calling parse_register) and eliminate immediate left recursion
5577 to implement a recursive-descent parser.
5579 expr e04 expr'
5581 expr' cmpOp e04 expr'
5582 | Empty
5584 e04 e05 e04'
5586 e04' addOp e05 e04'
5587 | Empty
5589 e05 e06 e05'
5591 e05' binOp e06 e05'
5592 | Empty
5594 e06 e09 e06'
5596 e06' mulOp e09 e06'
5597 | Empty
5599 e09 OFFSET e10 e09'
5600 | SHORT e10'
5601 | + e10'
5602 | - e10'
5603 | ~ e10'
5604 | NOT e10'
5605 | e10 e09'
5607 e09' PTR e10 e09'
5608 | : e10 e09'
5609 | Empty
5611 e10 e11 e10'
5613 e10' [ expr ] e10'
5614 | Empty
5616 e11 ( expr )
5617 | [ expr ]
5618 | BYTE
5619 | WORD
5620 | DWORD
5621 | FWORD
5622 | QWORD
5623 | TBYTE
5624 | OWORD
5625 | XMMWORD
5628 | register
5629 | id
5630 | constant */
5632 /* Parsing structure for the intel syntax parser. Used to implement the
5633 semantic actions for the operand grammar. */
5634 struct intel_parser_s
5636 char *op_string; /* The string being parsed. */
5637 int got_a_float; /* Whether the operand is a float. */
5638 int op_modifier; /* Operand modifier. */
5639 int is_mem; /* 1 if operand is memory reference. */
5640 int in_offset; /* >=1 if parsing operand of offset. */
5641 int in_bracket; /* >=1 if parsing operand in brackets. */
5642 const reg_entry *reg; /* Last register reference found. */
5643 char *disp; /* Displacement string being built. */
5644 char *next_operand; /* Resume point when splitting operands. */
5647 static struct intel_parser_s intel_parser;
5649 /* Token structure for parsing intel syntax. */
5650 struct intel_token
5652 int code; /* Token code. */
5653 const reg_entry *reg; /* Register entry for register tokens. */
5654 char *str; /* String representation. */
5657 static struct intel_token cur_token, prev_token;
5659 /* Token codes for the intel parser. Since T_SHORT is already used
5660 by COFF, undefine it first to prevent a warning. */
5661 #define T_NIL -1
5662 #define T_CONST 1
5663 #define T_REG 2
5664 #define T_BYTE 3
5665 #define T_WORD 4
5666 #define T_DWORD 5
5667 #define T_FWORD 6
5668 #define T_QWORD 7
5669 #define T_TBYTE 8
5670 #define T_XMMWORD 9
5671 #undef T_SHORT
5672 #define T_SHORT 10
5673 #define T_OFFSET 11
5674 #define T_PTR 12
5675 #define T_ID 13
5676 #define T_SHL 14
5677 #define T_SHR 15
5679 /* Prototypes for intel parser functions. */
5680 static int intel_match_token PARAMS ((int code));
5681 static void intel_get_token PARAMS ((void));
5682 static void intel_putback_token PARAMS ((void));
5683 static int intel_expr PARAMS ((void));
5684 static int intel_e04 PARAMS ((void));
5685 static int intel_e05 PARAMS ((void));
5686 static int intel_e06 PARAMS ((void));
5687 static int intel_e09 PARAMS ((void));
5688 static int intel_bracket_expr PARAMS ((void));
5689 static int intel_e10 PARAMS ((void));
5690 static int intel_e11 PARAMS ((void));
5692 static int
5693 i386_intel_operand (operand_string, got_a_float)
5694 char *operand_string;
5695 int got_a_float;
5697 int ret;
5698 char *p;
5700 p = intel_parser.op_string = xstrdup (operand_string);
5701 intel_parser.disp = (char *) xmalloc (strlen (operand_string) + 1);
5703 for (;;)
5705 /* Initialize token holders. */
5706 cur_token.code = prev_token.code = T_NIL;
5707 cur_token.reg = prev_token.reg = NULL;
5708 cur_token.str = prev_token.str = NULL;
5710 /* Initialize parser structure. */
5711 intel_parser.got_a_float = got_a_float;
5712 intel_parser.op_modifier = 0;
5713 intel_parser.is_mem = 0;
5714 intel_parser.in_offset = 0;
5715 intel_parser.in_bracket = 0;
5716 intel_parser.reg = NULL;
5717 intel_parser.disp[0] = '\0';
5718 intel_parser.next_operand = NULL;
5720 /* Read the first token and start the parser. */
5721 intel_get_token ();
5722 ret = intel_expr ();
5724 if (!ret)
5725 break;
5727 if (cur_token.code != T_NIL)
5729 as_bad (_("invalid operand for '%s' ('%s' unexpected)"),
5730 current_templates->start->name, cur_token.str);
5731 ret = 0;
5733 /* If we found a memory reference, hand it over to i386_displacement
5734 to fill in the rest of the operand fields. */
5735 else if (intel_parser.is_mem)
5737 if ((i.mem_operands == 1
5738 && (current_templates->start->opcode_modifier & IsString) == 0)
5739 || i.mem_operands == 2)
5741 as_bad (_("too many memory references for '%s'"),
5742 current_templates->start->name);
5743 ret = 0;
5745 else
5747 char *s = intel_parser.disp;
5748 i.mem_operands++;
5750 if (!quiet_warnings && intel_parser.is_mem < 0)
5751 /* See the comments in intel_bracket_expr. */
5752 as_warn (_("Treating `%s' as memory reference"), operand_string);
5754 /* Add the displacement expression. */
5755 if (*s != '\0')
5756 ret = i386_displacement (s, s + strlen (s));
5757 if (ret)
5759 /* Swap base and index in 16-bit memory operands like
5760 [si+bx]. Since i386_index_check is also used in AT&T
5761 mode we have to do that here. */
5762 if (i.base_reg
5763 && i.index_reg
5764 && (i.base_reg->reg_type & Reg16)
5765 && (i.index_reg->reg_type & Reg16)
5766 && i.base_reg->reg_num >= 6
5767 && i.index_reg->reg_num < 6)
5769 const reg_entry *base = i.index_reg;
5771 i.index_reg = i.base_reg;
5772 i.base_reg = base;
5774 ret = i386_index_check (operand_string);
5779 /* Constant and OFFSET expressions are handled by i386_immediate. */
5780 else if ((intel_parser.op_modifier & (1 << T_OFFSET))
5781 || intel_parser.reg == NULL)
5782 ret = i386_immediate (intel_parser.disp);
5784 if (intel_parser.next_operand && this_operand >= MAX_OPERANDS - 1)
5785 ret = 0;
5786 if (!ret || !intel_parser.next_operand)
5787 break;
5788 intel_parser.op_string = intel_parser.next_operand;
5789 this_operand = i.operands++;
5792 free (p);
5793 free (intel_parser.disp);
5795 return ret;
5798 #define NUM_ADDRESS_REGS (!!i.base_reg + !!i.index_reg)
5800 /* expr e04 expr'
5802 expr' cmpOp e04 expr'
5803 | Empty */
5804 static int
5805 intel_expr ()
5807 /* XXX Implement the comparison operators. */
5808 return intel_e04 ();
5811 /* e04 e05 e04'
5813 e04' addOp e05 e04'
5814 | Empty */
5815 static int
5816 intel_e04 ()
5818 int nregs = -1;
5820 for (;;)
5822 if (!intel_e05())
5823 return 0;
5825 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
5826 i.base_reg = i386_regtab + REGNAM_AL; /* al is invalid as base */
5828 if (cur_token.code == '+')
5829 nregs = -1;
5830 else if (cur_token.code == '-')
5831 nregs = NUM_ADDRESS_REGS;
5832 else
5833 return 1;
5835 strcat (intel_parser.disp, cur_token.str);
5836 intel_match_token (cur_token.code);
5840 /* e05 e06 e05'
5842 e05' binOp e06 e05'
5843 | Empty */
5844 static int
5845 intel_e05 ()
5847 int nregs = ~NUM_ADDRESS_REGS;
5849 for (;;)
5851 if (!intel_e06())
5852 return 0;
5854 if (cur_token.code == '&' || cur_token.code == '|' || cur_token.code == '^')
5856 char str[2];
5858 str[0] = cur_token.code;
5859 str[1] = 0;
5860 strcat (intel_parser.disp, str);
5862 else
5863 break;
5865 intel_match_token (cur_token.code);
5867 if (nregs < 0)
5868 nregs = ~nregs;
5870 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
5871 i.base_reg = i386_regtab + REGNAM_AL + 1; /* cl is invalid as base */
5872 return 1;
5875 /* e06 e09 e06'
5877 e06' mulOp e09 e06'
5878 | Empty */
5879 static int
5880 intel_e06 ()
5882 int nregs = ~NUM_ADDRESS_REGS;
5884 for (;;)
5886 if (!intel_e09())
5887 return 0;
5889 if (cur_token.code == '*' || cur_token.code == '/' || cur_token.code == '%')
5891 char str[2];
5893 str[0] = cur_token.code;
5894 str[1] = 0;
5895 strcat (intel_parser.disp, str);
5897 else if (cur_token.code == T_SHL)
5898 strcat (intel_parser.disp, "<<");
5899 else if (cur_token.code == T_SHR)
5900 strcat (intel_parser.disp, ">>");
5901 else
5902 break;
5904 intel_match_token (cur_token.code);
5906 if (nregs < 0)
5907 nregs = ~nregs;
5909 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
5910 i.base_reg = i386_regtab + REGNAM_AL + 2; /* dl is invalid as base */
5911 return 1;
5914 /* e09 OFFSET e09
5915 | SHORT e09
5916 | + e09
5917 | - e09
5918 | ~ e09
5919 | NOT e09
5920 | e10 e09'
5922 e09' PTR e10 e09'
5923 | : e10 e09'
5924 | Empty */
5925 static int
5926 intel_e09 ()
5928 int nregs = ~NUM_ADDRESS_REGS;
5929 int in_offset = 0;
5931 for (;;)
5933 /* Don't consume constants here. */
5934 if (cur_token.code == '+' || cur_token.code == '-')
5936 /* Need to look one token ahead - if the next token
5937 is a constant, the current token is its sign. */
5938 int next_code;
5940 intel_match_token (cur_token.code);
5941 next_code = cur_token.code;
5942 intel_putback_token ();
5943 if (next_code == T_CONST)
5944 break;
5947 /* e09 OFFSET e09 */
5948 if (cur_token.code == T_OFFSET)
5950 if (!in_offset++)
5951 ++intel_parser.in_offset;
5954 /* e09 SHORT e09 */
5955 else if (cur_token.code == T_SHORT)
5956 intel_parser.op_modifier |= 1 << T_SHORT;
5958 /* e09 + e09 */
5959 else if (cur_token.code == '+')
5960 strcat (intel_parser.disp, "+");
5962 /* e09 - e09
5963 | ~ e09
5964 | NOT e09 */
5965 else if (cur_token.code == '-' || cur_token.code == '~')
5967 char str[2];
5969 if (nregs < 0)
5970 nregs = ~nregs;
5971 str[0] = cur_token.code;
5972 str[1] = 0;
5973 strcat (intel_parser.disp, str);
5976 /* e09 e10 e09' */
5977 else
5978 break;
5980 intel_match_token (cur_token.code);
5983 for (;;)
5985 if (!intel_e10 ())
5986 return 0;
5988 /* e09' PTR e10 e09' */
5989 if (cur_token.code == T_PTR)
5991 char suffix;
5993 if (prev_token.code == T_BYTE)
5994 suffix = BYTE_MNEM_SUFFIX;
5996 else if (prev_token.code == T_WORD)
5998 if (current_templates->start->name[0] == 'l'
5999 && current_templates->start->name[2] == 's'
6000 && current_templates->start->name[3] == 0)
6001 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
6002 else if (intel_parser.got_a_float == 2) /* "fi..." */
6003 suffix = SHORT_MNEM_SUFFIX;
6004 else
6005 suffix = WORD_MNEM_SUFFIX;
6008 else if (prev_token.code == T_DWORD)
6010 if (current_templates->start->name[0] == 'l'
6011 && current_templates->start->name[2] == 's'
6012 && current_templates->start->name[3] == 0)
6013 suffix = WORD_MNEM_SUFFIX;
6014 else if (flag_code == CODE_16BIT
6015 && (current_templates->start->opcode_modifier
6016 & (Jump|JumpDword|JumpInterSegment)))
6017 suffix = LONG_DOUBLE_MNEM_SUFFIX;
6018 else if (intel_parser.got_a_float == 1) /* "f..." */
6019 suffix = SHORT_MNEM_SUFFIX;
6020 else
6021 suffix = LONG_MNEM_SUFFIX;
6024 else if (prev_token.code == T_FWORD)
6026 if (current_templates->start->name[0] == 'l'
6027 && current_templates->start->name[2] == 's'
6028 && current_templates->start->name[3] == 0)
6029 suffix = LONG_MNEM_SUFFIX;
6030 else if (!intel_parser.got_a_float)
6032 if (flag_code == CODE_16BIT)
6033 add_prefix (DATA_PREFIX_OPCODE);
6034 suffix = LONG_DOUBLE_MNEM_SUFFIX;
6036 else
6037 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
6040 else if (prev_token.code == T_QWORD)
6042 if (intel_parser.got_a_float == 1) /* "f..." */
6043 suffix = LONG_MNEM_SUFFIX;
6044 else
6045 suffix = QWORD_MNEM_SUFFIX;
6048 else if (prev_token.code == T_TBYTE)
6050 if (intel_parser.got_a_float == 1)
6051 suffix = LONG_DOUBLE_MNEM_SUFFIX;
6052 else
6053 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
6056 else if (prev_token.code == T_XMMWORD)
6058 /* XXX ignored for now, but accepted since gcc uses it */
6059 suffix = 0;
6062 else
6064 as_bad (_("Unknown operand modifier `%s'"), prev_token.str);
6065 return 0;
6068 if (current_templates->start->base_opcode == 0x8d /* lea */)
6070 else if (!i.suffix)
6071 i.suffix = suffix;
6072 else if (i.suffix != suffix)
6074 as_bad (_("Conflicting operand modifiers"));
6075 return 0;
6080 /* e09' : e10 e09' */
6081 else if (cur_token.code == ':')
6083 if (prev_token.code != T_REG)
6085 /* While {call,jmp} SSSS:OOOO is MASM syntax only when SSSS is a
6086 segment/group identifier (which we don't have), using comma
6087 as the operand separator there is even less consistent, since
6088 there all branches only have a single operand. */
6089 if (this_operand != 0
6090 || intel_parser.in_offset
6091 || intel_parser.in_bracket
6092 || (!(current_templates->start->opcode_modifier
6093 & (Jump|JumpDword|JumpInterSegment))
6094 && !(current_templates->start->operand_types[0]
6095 & JumpAbsolute)))
6096 return intel_match_token (T_NIL);
6097 /* Remember the start of the 2nd operand and terminate 1st
6098 operand here.
6099 XXX This isn't right, yet (when SSSS:OOOO is right operand of
6100 another expression), but it gets at least the simplest case
6101 (a plain number or symbol on the left side) right. */
6102 intel_parser.next_operand = intel_parser.op_string;
6103 *--intel_parser.op_string = '\0';
6104 return intel_match_token (':');
6108 /* e09' Empty */
6109 else
6110 break;
6112 intel_match_token (cur_token.code);
6116 if (in_offset)
6118 --intel_parser.in_offset;
6119 if (nregs < 0)
6120 nregs = ~nregs;
6121 if (NUM_ADDRESS_REGS > nregs)
6123 as_bad (_("Invalid operand to `OFFSET'"));
6124 return 0;
6126 intel_parser.op_modifier |= 1 << T_OFFSET;
6129 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6130 i.base_reg = i386_regtab + REGNAM_AL + 3; /* bl is invalid as base */
6131 return 1;
6134 static int
6135 intel_bracket_expr ()
6137 int was_offset = intel_parser.op_modifier & (1 << T_OFFSET);
6138 const char *start = intel_parser.op_string;
6139 int len;
6141 if (i.op[this_operand].regs)
6142 return intel_match_token (T_NIL);
6144 intel_match_token ('[');
6146 /* Mark as a memory operand only if it's not already known to be an
6147 offset expression. If it's an offset expression, we need to keep
6148 the brace in. */
6149 if (!intel_parser.in_offset)
6151 ++intel_parser.in_bracket;
6152 /* Unfortunately gas always diverged from MASM in a respect that can't
6153 be easily fixed without risking to break code sequences likely to be
6154 encountered (the testsuite even check for this): MASM doesn't consider
6155 an expression inside brackets unconditionally as a memory reference.
6156 When that is e.g. a constant, an offset expression, or the sum of the
6157 two, this is still taken as a constant load. gas, however, always
6158 treated these as memory references. As a compromise, we'll try to make
6159 offset expressions inside brackets work the MASM way (since that's
6160 less likely to be found in real world code), but make constants alone
6161 continue to work the traditional gas way. In either case, issue a
6162 warning. */
6163 intel_parser.op_modifier &= ~was_offset;
6165 else
6166 strcat (intel_parser.disp, "[");
6168 /* Add a '+' to the displacement string if necessary. */
6169 if (*intel_parser.disp != '\0'
6170 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
6171 strcat (intel_parser.disp, "+");
6173 if (intel_expr ()
6174 && (len = intel_parser.op_string - start - 1,
6175 intel_match_token (']')))
6177 /* Preserve brackets when the operand is an offset expression. */
6178 if (intel_parser.in_offset)
6179 strcat (intel_parser.disp, "]");
6180 else
6182 --intel_parser.in_bracket;
6183 if (i.base_reg || i.index_reg)
6184 intel_parser.is_mem = 1;
6185 if (!intel_parser.is_mem)
6187 if (!(intel_parser.op_modifier & (1 << T_OFFSET)))
6188 /* Defer the warning until all of the operand was parsed. */
6189 intel_parser.is_mem = -1;
6190 else if (!quiet_warnings)
6191 as_warn (_("`[%.*s]' taken to mean just `%.*s'"), len, start, len, start);
6194 intel_parser.op_modifier |= was_offset;
6196 return 1;
6198 return 0;
6201 /* e10 e11 e10'
6203 e10' [ expr ] e10'
6204 | Empty */
6205 static int
6206 intel_e10 ()
6208 if (!intel_e11 ())
6209 return 0;
6211 while (cur_token.code == '[')
6213 if (!intel_bracket_expr ())
6214 return 0;
6217 return 1;
6220 /* e11 ( expr )
6221 | [ expr ]
6222 | BYTE
6223 | WORD
6224 | DWORD
6225 | FWORD
6226 | QWORD
6227 | TBYTE
6228 | OWORD
6229 | XMMWORD
6232 | register
6233 | id
6234 | constant */
6235 static int
6236 intel_e11 ()
6238 switch (cur_token.code)
6240 /* e11 ( expr ) */
6241 case '(':
6242 intel_match_token ('(');
6243 strcat (intel_parser.disp, "(");
6245 if (intel_expr () && intel_match_token (')'))
6247 strcat (intel_parser.disp, ")");
6248 return 1;
6250 return 0;
6252 /* e11 [ expr ] */
6253 case '[':
6254 /* Operands for jump/call inside brackets denote absolute addresses.
6255 XXX This shouldn't be needed anymore (or if it should rather live
6256 in intel_bracket_expr). */
6257 if (current_templates->start->opcode_modifier
6258 & (Jump|JumpDword|JumpByte|JumpInterSegment))
6259 i.types[this_operand] |= JumpAbsolute;
6261 return intel_bracket_expr ();
6263 /* e11 $
6264 | . */
6265 case '.':
6266 strcat (intel_parser.disp, cur_token.str);
6267 intel_match_token (cur_token.code);
6269 /* Mark as a memory operand only if it's not already known to be an
6270 offset expression. */
6271 if (!intel_parser.in_offset)
6272 intel_parser.is_mem = 1;
6274 return 1;
6276 /* e11 register */
6277 case T_REG:
6279 const reg_entry *reg = intel_parser.reg = cur_token.reg;
6281 intel_match_token (T_REG);
6283 /* Check for segment change. */
6284 if (cur_token.code == ':')
6286 if (!(reg->reg_type & (SReg2 | SReg3)))
6288 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
6289 return 0;
6291 else if (i.seg[i.mem_operands])
6292 as_warn (_("Extra segment override ignored"));
6293 else
6295 if (!intel_parser.in_offset)
6296 intel_parser.is_mem = 1;
6297 switch (reg->reg_num)
6299 case 0:
6300 i.seg[i.mem_operands] = &es;
6301 break;
6302 case 1:
6303 i.seg[i.mem_operands] = &cs;
6304 break;
6305 case 2:
6306 i.seg[i.mem_operands] = &ss;
6307 break;
6308 case 3:
6309 i.seg[i.mem_operands] = &ds;
6310 break;
6311 case 4:
6312 i.seg[i.mem_operands] = &fs;
6313 break;
6314 case 5:
6315 i.seg[i.mem_operands] = &gs;
6316 break;
6321 /* Not a segment register. Check for register scaling. */
6322 else if (cur_token.code == '*')
6324 if (!intel_parser.in_bracket)
6326 as_bad (_("Register scaling only allowed in memory operands"));
6327 return 0;
6330 if (reg->reg_type & Reg16) /* Disallow things like [si*1]. */
6331 reg = i386_regtab + REGNAM_AX + 4; /* sp is invalid as index */
6332 else if (i.index_reg)
6333 reg = i386_regtab + REGNAM_EAX + 4; /* esp is invalid as index */
6335 /* What follows must be a valid scale. */
6336 intel_match_token ('*');
6337 i.index_reg = reg;
6338 i.types[this_operand] |= BaseIndex;
6340 /* Set the scale after setting the register (otherwise,
6341 i386_scale will complain) */
6342 if (cur_token.code == '+' || cur_token.code == '-')
6344 char *str, sign = cur_token.code;
6345 intel_match_token (cur_token.code);
6346 if (cur_token.code != T_CONST)
6348 as_bad (_("Syntax error: Expecting a constant, got `%s'"),
6349 cur_token.str);
6350 return 0;
6352 str = (char *) xmalloc (strlen (cur_token.str) + 2);
6353 strcpy (str + 1, cur_token.str);
6354 *str = sign;
6355 if (!i386_scale (str))
6356 return 0;
6357 free (str);
6359 else if (!i386_scale (cur_token.str))
6360 return 0;
6361 intel_match_token (cur_token.code);
6364 /* No scaling. If this is a memory operand, the register is either a
6365 base register (first occurrence) or an index register (second
6366 occurrence). */
6367 else if (intel_parser.in_bracket && !(reg->reg_type & (SReg2 | SReg3)))
6370 if (!i.base_reg)
6371 i.base_reg = reg;
6372 else if (!i.index_reg)
6373 i.index_reg = reg;
6374 else
6376 as_bad (_("Too many register references in memory operand"));
6377 return 0;
6380 i.types[this_operand] |= BaseIndex;
6383 /* Offset modifier. Add the register to the displacement string to be
6384 parsed as an immediate expression after we're done. */
6385 else if (intel_parser.in_offset)
6387 as_warn (_("Using register names in OFFSET expressions is deprecated"));
6388 strcat (intel_parser.disp, reg->reg_name);
6391 /* It's neither base nor index nor offset. */
6392 else if (!intel_parser.is_mem)
6394 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
6395 i.op[this_operand].regs = reg;
6396 i.reg_operands++;
6398 else
6400 as_bad (_("Invalid use of register"));
6401 return 0;
6404 /* Since registers are not part of the displacement string (except
6405 when we're parsing offset operands), we may need to remove any
6406 preceding '+' from the displacement string. */
6407 if (*intel_parser.disp != '\0'
6408 && !intel_parser.in_offset)
6410 char *s = intel_parser.disp;
6411 s += strlen (s) - 1;
6412 if (*s == '+')
6413 *s = '\0';
6416 return 1;
6419 /* e11 BYTE
6420 | WORD
6421 | DWORD
6422 | FWORD
6423 | QWORD
6424 | TBYTE
6425 | OWORD
6426 | XMMWORD */
6427 case T_BYTE:
6428 case T_WORD:
6429 case T_DWORD:
6430 case T_FWORD:
6431 case T_QWORD:
6432 case T_TBYTE:
6433 case T_XMMWORD:
6434 intel_match_token (cur_token.code);
6436 if (cur_token.code == T_PTR)
6437 return 1;
6439 /* It must have been an identifier. */
6440 intel_putback_token ();
6441 cur_token.code = T_ID;
6442 /* FALLTHRU */
6444 /* e11 id
6445 | constant */
6446 case T_ID:
6447 if (!intel_parser.in_offset && intel_parser.is_mem <= 0)
6449 symbolS *symbolP;
6451 /* The identifier represents a memory reference only if it's not
6452 preceded by an offset modifier and if it's not an equate. */
6453 symbolP = symbol_find(cur_token.str);
6454 if (!symbolP || S_GET_SEGMENT(symbolP) != absolute_section)
6455 intel_parser.is_mem = 1;
6457 /* FALLTHRU */
6459 case T_CONST:
6460 case '-':
6461 case '+':
6463 char *save_str, sign = 0;
6465 /* Allow constants that start with `+' or `-'. */
6466 if (cur_token.code == '-' || cur_token.code == '+')
6468 sign = cur_token.code;
6469 intel_match_token (cur_token.code);
6470 if (cur_token.code != T_CONST)
6472 as_bad (_("Syntax error: Expecting a constant, got `%s'"),
6473 cur_token.str);
6474 return 0;
6478 save_str = (char *) xmalloc (strlen (cur_token.str) + 2);
6479 strcpy (save_str + !!sign, cur_token.str);
6480 if (sign)
6481 *save_str = sign;
6483 /* Get the next token to check for register scaling. */
6484 intel_match_token (cur_token.code);
6486 /* Check if this constant is a scaling factor for an index register. */
6487 if (cur_token.code == '*')
6489 if (intel_match_token ('*') && cur_token.code == T_REG)
6491 const reg_entry *reg = cur_token.reg;
6493 if (!intel_parser.in_bracket)
6495 as_bad (_("Register scaling only allowed in memory operands"));
6496 return 0;
6499 if (reg->reg_type & Reg16) /* Disallow things like [1*si]. */
6500 reg = i386_regtab + REGNAM_AX + 4; /* sp is invalid as index */
6501 else if (i.index_reg)
6502 reg = i386_regtab + REGNAM_EAX + 4; /* esp is invalid as index */
6504 /* The constant is followed by `* reg', so it must be
6505 a valid scale. */
6506 i.index_reg = reg;
6507 i.types[this_operand] |= BaseIndex;
6509 /* Set the scale after setting the register (otherwise,
6510 i386_scale will complain) */
6511 if (!i386_scale (save_str))
6512 return 0;
6513 intel_match_token (T_REG);
6515 /* Since registers are not part of the displacement
6516 string, we may need to remove any preceding '+' from
6517 the displacement string. */
6518 if (*intel_parser.disp != '\0')
6520 char *s = intel_parser.disp;
6521 s += strlen (s) - 1;
6522 if (*s == '+')
6523 *s = '\0';
6526 free (save_str);
6528 return 1;
6531 /* The constant was not used for register scaling. Since we have
6532 already consumed the token following `*' we now need to put it
6533 back in the stream. */
6534 intel_putback_token ();
6537 /* Add the constant to the displacement string. */
6538 strcat (intel_parser.disp, save_str);
6539 free (save_str);
6541 return 1;
6545 as_bad (_("Unrecognized token '%s'"), cur_token.str);
6546 return 0;
6549 /* Match the given token against cur_token. If they match, read the next
6550 token from the operand string. */
6551 static int
6552 intel_match_token (code)
6553 int code;
6555 if (cur_token.code == code)
6557 intel_get_token ();
6558 return 1;
6560 else
6562 as_bad (_("Unexpected token `%s'"), cur_token.str);
6563 return 0;
6567 /* Read a new token from intel_parser.op_string and store it in cur_token. */
6568 static void
6569 intel_get_token ()
6571 char *end_op;
6572 const reg_entry *reg;
6573 struct intel_token new_token;
6575 new_token.code = T_NIL;
6576 new_token.reg = NULL;
6577 new_token.str = NULL;
6579 /* Free the memory allocated to the previous token and move
6580 cur_token to prev_token. */
6581 if (prev_token.str)
6582 free (prev_token.str);
6584 prev_token = cur_token;
6586 /* Skip whitespace. */
6587 while (is_space_char (*intel_parser.op_string))
6588 intel_parser.op_string++;
6590 /* Return an empty token if we find nothing else on the line. */
6591 if (*intel_parser.op_string == '\0')
6593 cur_token = new_token;
6594 return;
6597 /* The new token cannot be larger than the remainder of the operand
6598 string. */
6599 new_token.str = (char *) xmalloc (strlen (intel_parser.op_string) + 1);
6600 new_token.str[0] = '\0';
6602 if (strchr ("0123456789", *intel_parser.op_string))
6604 char *p = new_token.str;
6605 char *q = intel_parser.op_string;
6606 new_token.code = T_CONST;
6608 /* Allow any kind of identifier char to encompass floating point and
6609 hexadecimal numbers. */
6610 while (is_identifier_char (*q))
6611 *p++ = *q++;
6612 *p = '\0';
6614 /* Recognize special symbol names [0-9][bf]. */
6615 if (strlen (intel_parser.op_string) == 2
6616 && (intel_parser.op_string[1] == 'b'
6617 || intel_parser.op_string[1] == 'f'))
6618 new_token.code = T_ID;
6621 else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
6622 && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
6624 new_token.code = T_REG;
6625 new_token.reg = reg;
6627 if (*intel_parser.op_string == REGISTER_PREFIX)
6629 new_token.str[0] = REGISTER_PREFIX;
6630 new_token.str[1] = '\0';
6633 strcat (new_token.str, reg->reg_name);
6636 else if (is_identifier_char (*intel_parser.op_string))
6638 char *p = new_token.str;
6639 char *q = intel_parser.op_string;
6641 /* A '.' or '$' followed by an identifier char is an identifier.
6642 Otherwise, it's operator '.' followed by an expression. */
6643 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
6645 new_token.code = '.';
6646 new_token.str[0] = '.';
6647 new_token.str[1] = '\0';
6649 else
6651 while (is_identifier_char (*q) || *q == '@')
6652 *p++ = *q++;
6653 *p = '\0';
6655 if (strcasecmp (new_token.str, "NOT") == 0)
6656 new_token.code = '~';
6658 else if (strcasecmp (new_token.str, "MOD") == 0)
6659 new_token.code = '%';
6661 else if (strcasecmp (new_token.str, "AND") == 0)
6662 new_token.code = '&';
6664 else if (strcasecmp (new_token.str, "OR") == 0)
6665 new_token.code = '|';
6667 else if (strcasecmp (new_token.str, "XOR") == 0)
6668 new_token.code = '^';
6670 else if (strcasecmp (new_token.str, "SHL") == 0)
6671 new_token.code = T_SHL;
6673 else if (strcasecmp (new_token.str, "SHR") == 0)
6674 new_token.code = T_SHR;
6676 else if (strcasecmp (new_token.str, "BYTE") == 0)
6677 new_token.code = T_BYTE;
6679 else if (strcasecmp (new_token.str, "WORD") == 0)
6680 new_token.code = T_WORD;
6682 else if (strcasecmp (new_token.str, "DWORD") == 0)
6683 new_token.code = T_DWORD;
6685 else if (strcasecmp (new_token.str, "FWORD") == 0)
6686 new_token.code = T_FWORD;
6688 else if (strcasecmp (new_token.str, "QWORD") == 0)
6689 new_token.code = T_QWORD;
6691 else if (strcasecmp (new_token.str, "TBYTE") == 0
6692 /* XXX remove (gcc still uses it) */
6693 || strcasecmp (new_token.str, "XWORD") == 0)
6694 new_token.code = T_TBYTE;
6696 else if (strcasecmp (new_token.str, "XMMWORD") == 0
6697 || strcasecmp (new_token.str, "OWORD") == 0)
6698 new_token.code = T_XMMWORD;
6700 else if (strcasecmp (new_token.str, "PTR") == 0)
6701 new_token.code = T_PTR;
6703 else if (strcasecmp (new_token.str, "SHORT") == 0)
6704 new_token.code = T_SHORT;
6706 else if (strcasecmp (new_token.str, "OFFSET") == 0)
6708 new_token.code = T_OFFSET;
6710 /* ??? This is not mentioned in the MASM grammar but gcc
6711 makes use of it with -mintel-syntax. OFFSET may be
6712 followed by FLAT: */
6713 if (strncasecmp (q, " FLAT:", 6) == 0)
6714 strcat (new_token.str, " FLAT:");
6717 /* ??? This is not mentioned in the MASM grammar. */
6718 else if (strcasecmp (new_token.str, "FLAT") == 0)
6720 new_token.code = T_OFFSET;
6721 if (*q == ':')
6722 strcat (new_token.str, ":");
6723 else
6724 as_bad (_("`:' expected"));
6727 else
6728 new_token.code = T_ID;
6732 else if (strchr ("+-/*%|&^:[]()~", *intel_parser.op_string))
6734 new_token.code = *intel_parser.op_string;
6735 new_token.str[0] = *intel_parser.op_string;
6736 new_token.str[1] = '\0';
6739 else if (strchr ("<>", *intel_parser.op_string)
6740 && *intel_parser.op_string == *(intel_parser.op_string + 1))
6742 new_token.code = *intel_parser.op_string == '<' ? T_SHL : T_SHR;
6743 new_token.str[0] = *intel_parser.op_string;
6744 new_token.str[1] = *intel_parser.op_string;
6745 new_token.str[2] = '\0';
6748 else
6749 as_bad (_("Unrecognized token `%s'"), intel_parser.op_string);
6751 intel_parser.op_string += strlen (new_token.str);
6752 cur_token = new_token;
6755 /* Put cur_token back into the token stream and make cur_token point to
6756 prev_token. */
6757 static void
6758 intel_putback_token ()
6760 if (cur_token.code != T_NIL)
6762 intel_parser.op_string -= strlen (cur_token.str);
6763 free (cur_token.str);
6765 cur_token = prev_token;
6767 /* Forget prev_token. */
6768 prev_token.code = T_NIL;
6769 prev_token.reg = NULL;
6770 prev_token.str = NULL;
6774 tc_x86_regname_to_dw2regnum (const char *regname)
6776 unsigned int regnum;
6777 unsigned int regnames_count;
6778 char *regnames_32[] =
6780 "eax", "ecx", "edx", "ebx",
6781 "esp", "ebp", "esi", "edi",
6782 "eip"
6784 char *regnames_64[] =
6786 "rax", "rbx", "rcx", "rdx",
6787 "rdi", "rsi", "rbp", "rsp",
6788 "r8", "r9", "r10", "r11",
6789 "r12", "r13", "r14", "r15",
6790 "rip"
6792 char **regnames;
6794 if (flag_code == CODE_64BIT)
6796 regnames = regnames_64;
6797 regnames_count = ARRAY_SIZE (regnames_64);
6799 else
6801 regnames = regnames_32;
6802 regnames_count = ARRAY_SIZE (regnames_32);
6805 for (regnum = 0; regnum < regnames_count; regnum++)
6806 if (strcmp (regname, regnames[regnum]) == 0)
6807 return regnum;
6809 return -1;
6812 void
6813 tc_x86_frame_initial_instructions (void)
6815 static unsigned int sp_regno;
6817 if (!sp_regno)
6818 sp_regno = tc_x86_regname_to_dw2regnum (flag_code == CODE_64BIT
6819 ? "rsp" : "esp");
6821 cfi_add_CFA_def_cfa (sp_regno, -x86_cie_data_alignment);
6822 cfi_add_CFA_offset (x86_dwarf2_return_column, x86_cie_data_alignment);
6826 i386_elf_section_type (const char *str, size_t len)
6828 if (flag_code == CODE_64BIT
6829 && len == sizeof ("unwind") - 1
6830 && strncmp (str, "unwind", 6) == 0)
6831 return SHT_X86_64_UNWIND;
6833 return -1;
6836 #ifdef TE_PE
6837 void
6838 tc_pe_dwarf2_emit_offset (symbolS *symbol, unsigned int size)
6840 expressionS expr;
6842 expr.X_op = O_secrel;
6843 expr.X_add_symbol = symbol;
6844 expr.X_add_number = 0;
6845 emit_expr (&expr, size);
6847 #endif