2004-06-09 Paolo Bonzini <bonzini@gnu.org>
[binutils.git] / gas / config / tc-i386.c
blobf37c259249fc029c20aa64a8039a56087a908349
1 /* i386.c -- Assemble code for the Intel 80386
2 Copyright 1989, 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
3 2000, 2001, 2002, 2003, 2004
4 Free Software Foundation, Inc.
6 This file is part of GAS, the GNU Assembler.
8 GAS is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2, or (at your option)
11 any later version.
13 GAS is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with GAS; see the file COPYING. If not, write to the Free
20 Software Foundation, 59 Temple Place - Suite 330, Boston, MA
21 02111-1307, USA. */
23 /* Intel 80386 machine specific gas.
24 Written by Eliot Dresselhaus (eliot@mgm.mit.edu).
25 x86_64 support by Jan Hubicka (jh@suse.cz)
26 VIA PadLock support by Michal Ludvig (mludvig@suse.cz)
27 Bugs & suggestions are completely welcome. This is free software.
28 Please help us make it better. */
30 #include "as.h"
31 #include "safe-ctype.h"
32 #include "subsegs.h"
33 #include "dwarf2dbg.h"
34 #include "dw2gencfi.h"
35 #include "opcode/i386.h"
37 #ifndef REGISTER_WARNINGS
38 #define REGISTER_WARNINGS 1
39 #endif
41 #ifndef INFER_ADDR_PREFIX
42 #define INFER_ADDR_PREFIX 1
43 #endif
45 #ifndef SCALE1_WHEN_NO_INDEX
46 /* Specifying a scale factor besides 1 when there is no index is
47 futile. eg. `mov (%ebx,2),%al' does exactly the same as
48 `mov (%ebx),%al'. To slavishly follow what the programmer
49 specified, set SCALE1_WHEN_NO_INDEX to 0. */
50 #define SCALE1_WHEN_NO_INDEX 1
51 #endif
53 #ifndef DEFAULT_ARCH
54 #define DEFAULT_ARCH "i386"
55 #endif
57 #ifndef INLINE
58 #if __GNUC__ >= 2
59 #define INLINE __inline__
60 #else
61 #define INLINE
62 #endif
63 #endif
65 static INLINE unsigned int mode_from_disp_size PARAMS ((unsigned int));
66 static INLINE int fits_in_signed_byte PARAMS ((offsetT));
67 static INLINE int fits_in_unsigned_byte PARAMS ((offsetT));
68 static INLINE int fits_in_unsigned_word PARAMS ((offsetT));
69 static INLINE int fits_in_signed_word PARAMS ((offsetT));
70 static INLINE int fits_in_unsigned_long PARAMS ((offsetT));
71 static INLINE int fits_in_signed_long PARAMS ((offsetT));
72 static int smallest_imm_type PARAMS ((offsetT));
73 static offsetT offset_in_range PARAMS ((offsetT, int));
74 static int add_prefix PARAMS ((unsigned int));
75 static void set_code_flag PARAMS ((int));
76 static void set_16bit_gcc_code_flag PARAMS ((int));
77 static void set_intel_syntax PARAMS ((int));
78 static void set_cpu_arch PARAMS ((int));
79 #ifdef TE_PE
80 static void pe_directive_secrel PARAMS ((int));
81 #endif
82 static char *output_invalid PARAMS ((int c));
83 static int i386_operand PARAMS ((char *operand_string));
84 static int i386_intel_operand PARAMS ((char *operand_string, int got_a_float));
85 static const reg_entry *parse_register PARAMS ((char *reg_string,
86 char **end_op));
87 static char *parse_insn PARAMS ((char *, char *));
88 static char *parse_operands PARAMS ((char *, const char *));
89 static void swap_operands PARAMS ((void));
90 static void optimize_imm PARAMS ((void));
91 static void optimize_disp PARAMS ((void));
92 static int match_template PARAMS ((void));
93 static int check_string PARAMS ((void));
94 static int process_suffix PARAMS ((void));
95 static int check_byte_reg PARAMS ((void));
96 static int check_long_reg PARAMS ((void));
97 static int check_qword_reg PARAMS ((void));
98 static int check_word_reg PARAMS ((void));
99 static int finalize_imm PARAMS ((void));
100 static int process_operands PARAMS ((void));
101 static const seg_entry *build_modrm_byte PARAMS ((void));
102 static void output_insn PARAMS ((void));
103 static void output_branch PARAMS ((void));
104 static void output_jump PARAMS ((void));
105 static void output_interseg_jump PARAMS ((void));
106 static void output_imm PARAMS ((fragS *insn_start_frag,
107 offsetT insn_start_off));
108 static void output_disp PARAMS ((fragS *insn_start_frag,
109 offsetT insn_start_off));
110 #ifndef I386COFF
111 static void s_bss PARAMS ((int));
112 #endif
114 static const char *default_arch = DEFAULT_ARCH;
116 /* 'md_assemble ()' gathers together information and puts it into a
117 i386_insn. */
119 union i386_op
121 expressionS *disps;
122 expressionS *imms;
123 const reg_entry *regs;
126 struct _i386_insn
128 /* TM holds the template for the insn were currently assembling. */
129 template tm;
131 /* SUFFIX holds the instruction mnemonic suffix if given.
132 (e.g. 'l' for 'movl') */
133 char suffix;
135 /* OPERANDS gives the number of given operands. */
136 unsigned int operands;
138 /* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number
139 of given register, displacement, memory operands and immediate
140 operands. */
141 unsigned int reg_operands, disp_operands, mem_operands, imm_operands;
143 /* TYPES [i] is the type (see above #defines) which tells us how to
144 use OP[i] for the corresponding operand. */
145 unsigned int types[MAX_OPERANDS];
147 /* Displacement expression, immediate expression, or register for each
148 operand. */
149 union i386_op op[MAX_OPERANDS];
151 /* Flags for operands. */
152 unsigned int flags[MAX_OPERANDS];
153 #define Operand_PCrel 1
155 /* Relocation type for operand */
156 enum bfd_reloc_code_real reloc[MAX_OPERANDS];
158 /* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode
159 the base index byte below. */
160 const reg_entry *base_reg;
161 const reg_entry *index_reg;
162 unsigned int log2_scale_factor;
164 /* SEG gives the seg_entries of this insn. They are zero unless
165 explicit segment overrides are given. */
166 const seg_entry *seg[2];
168 /* PREFIX holds all the given prefix opcodes (usually null).
169 PREFIXES is the number of prefix opcodes. */
170 unsigned int prefixes;
171 unsigned char prefix[MAX_PREFIXES];
173 /* RM and SIB are the modrm byte and the sib byte where the
174 addressing modes of this insn are encoded. */
176 modrm_byte rm;
177 rex_byte rex;
178 sib_byte sib;
181 typedef struct _i386_insn i386_insn;
183 /* List of chars besides those in app.c:symbol_chars that can start an
184 operand. Used to prevent the scrubber eating vital white-space. */
185 #ifdef LEX_AT
186 const char extra_symbol_chars[] = "*%-(@[";
187 #else
188 const char extra_symbol_chars[] = "*%-([";
189 #endif
191 #if (defined (TE_I386AIX) \
192 || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) \
193 && !defined (TE_LINUX) \
194 && !defined (TE_FreeBSD) \
195 && !defined (TE_NetBSD)))
196 /* This array holds the chars that always start a comment. If the
197 pre-processor is disabled, these aren't very useful. */
198 const char comment_chars[] = "#/";
199 #define PREFIX_SEPARATOR '\\'
201 /* This array holds the chars that only start a comment at the beginning of
202 a line. If the line seems to have the form '# 123 filename'
203 .line and .file directives will appear in the pre-processed output.
204 Note that input_file.c hand checks for '#' at the beginning of the
205 first line of the input file. This is because the compiler outputs
206 #NO_APP at the beginning of its output.
207 Also note that comments started like this one will always work if
208 '/' isn't otherwise defined. */
209 const char line_comment_chars[] = "#";
211 #else
212 /* Putting '/' here makes it impossible to use the divide operator.
213 However, we need it for compatibility with SVR4 systems. */
214 const char comment_chars[] = "#";
215 #define PREFIX_SEPARATOR '/'
217 const char line_comment_chars[] = "/#";
218 #endif
220 const char line_separator_chars[] = ";";
222 /* Chars that can be used to separate mant from exp in floating point
223 nums. */
224 const char EXP_CHARS[] = "eE";
226 /* Chars that mean this number is a floating point constant
227 As in 0f12.456
228 or 0d1.2345e12. */
229 const char FLT_CHARS[] = "fFdDxX";
231 /* Tables for lexical analysis. */
232 static char mnemonic_chars[256];
233 static char register_chars[256];
234 static char operand_chars[256];
235 static char identifier_chars[256];
236 static char digit_chars[256];
238 /* Lexical macros. */
239 #define is_mnemonic_char(x) (mnemonic_chars[(unsigned char) x])
240 #define is_operand_char(x) (operand_chars[(unsigned char) x])
241 #define is_register_char(x) (register_chars[(unsigned char) x])
242 #define is_space_char(x) ((x) == ' ')
243 #define is_identifier_char(x) (identifier_chars[(unsigned char) x])
244 #define is_digit_char(x) (digit_chars[(unsigned char) x])
246 /* All non-digit non-letter characters that may occur in an operand. */
247 static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
249 /* md_assemble() always leaves the strings it's passed unaltered. To
250 effect this we maintain a stack of saved characters that we've smashed
251 with '\0's (indicating end of strings for various sub-fields of the
252 assembler instruction). */
253 static char save_stack[32];
254 static char *save_stack_p;
255 #define END_STRING_AND_SAVE(s) \
256 do { *save_stack_p++ = *(s); *(s) = '\0'; } while (0)
257 #define RESTORE_END_STRING(s) \
258 do { *(s) = *--save_stack_p; } while (0)
260 /* The instruction we're assembling. */
261 static i386_insn i;
263 /* Possible templates for current insn. */
264 static const templates *current_templates;
266 /* Per instruction expressionS buffers: 2 displacements & 2 immediate max. */
267 static expressionS disp_expressions[2], im_expressions[2];
269 /* Current operand we are working on. */
270 static int this_operand;
272 /* We support four different modes. FLAG_CODE variable is used to distinguish
273 these. */
275 enum flag_code {
276 CODE_32BIT,
277 CODE_16BIT,
278 CODE_64BIT };
279 #define NUM_FLAG_CODE ((int) CODE_64BIT + 1)
281 static enum flag_code flag_code;
282 static int use_rela_relocations = 0;
284 /* The names used to print error messages. */
285 static const char *flag_code_names[] =
287 "32",
288 "16",
289 "64"
292 /* 1 for intel syntax,
293 0 if att syntax. */
294 static int intel_syntax = 0;
296 /* 1 if register prefix % not required. */
297 static int allow_naked_reg = 0;
299 /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
300 leave, push, and pop instructions so that gcc has the same stack
301 frame as in 32 bit mode. */
302 static char stackop_size = '\0';
304 /* Non-zero to optimize code alignment. */
305 int optimize_align_code = 1;
307 /* Non-zero to quieten some warnings. */
308 static int quiet_warnings = 0;
310 /* CPU name. */
311 static const char *cpu_arch_name = NULL;
313 /* CPU feature flags. */
314 static unsigned int cpu_arch_flags = CpuUnknownFlags | CpuNo64;
316 /* If set, conditional jumps are not automatically promoted to handle
317 larger than a byte offset. */
318 static unsigned int no_cond_jump_promotion = 0;
320 /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
321 symbolS *GOT_symbol;
323 /* The dwarf2 return column, adjusted for 32 or 64 bit. */
324 unsigned int x86_dwarf2_return_column;
326 /* The dwarf2 data alignment, adjusted for 32 or 64 bit. */
327 int x86_cie_data_alignment;
329 /* Interface to relax_segment.
330 There are 3 major relax states for 386 jump insns because the
331 different types of jumps add different sizes to frags when we're
332 figuring out what sort of jump to choose to reach a given label. */
334 /* Types. */
335 #define UNCOND_JUMP 0
336 #define COND_JUMP 1
337 #define COND_JUMP86 2
339 /* Sizes. */
340 #define CODE16 1
341 #define SMALL 0
342 #define SMALL16 (SMALL | CODE16)
343 #define BIG 2
344 #define BIG16 (BIG | CODE16)
346 #ifndef INLINE
347 #ifdef __GNUC__
348 #define INLINE __inline__
349 #else
350 #define INLINE
351 #endif
352 #endif
354 #define ENCODE_RELAX_STATE(type, size) \
355 ((relax_substateT) (((type) << 2) | (size)))
356 #define TYPE_FROM_RELAX_STATE(s) \
357 ((s) >> 2)
358 #define DISP_SIZE_FROM_RELAX_STATE(s) \
359 ((((s) & 3) == BIG ? 4 : (((s) & 3) == BIG16 ? 2 : 1)))
361 /* This table is used by relax_frag to promote short jumps to long
362 ones where necessary. SMALL (short) jumps may be promoted to BIG
363 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
364 don't allow a short jump in a 32 bit code segment to be promoted to
365 a 16 bit offset jump because it's slower (requires data size
366 prefix), and doesn't work, unless the destination is in the bottom
367 64k of the code segment (The top 16 bits of eip are zeroed). */
369 const relax_typeS md_relax_table[] =
371 /* The fields are:
372 1) most positive reach of this state,
373 2) most negative reach of this state,
374 3) how many bytes this mode will have in the variable part of the frag
375 4) which index into the table to try if we can't fit into this one. */
377 /* UNCOND_JUMP states. */
378 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
379 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
380 /* dword jmp adds 4 bytes to frag:
381 0 extra opcode bytes, 4 displacement bytes. */
382 {0, 0, 4, 0},
383 /* word jmp adds 2 byte2 to frag:
384 0 extra opcode bytes, 2 displacement bytes. */
385 {0, 0, 2, 0},
387 /* COND_JUMP states. */
388 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
389 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
390 /* dword conditionals adds 5 bytes to frag:
391 1 extra opcode byte, 4 displacement bytes. */
392 {0, 0, 5, 0},
393 /* word conditionals add 3 bytes to frag:
394 1 extra opcode byte, 2 displacement bytes. */
395 {0, 0, 3, 0},
397 /* COND_JUMP86 states. */
398 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG)},
399 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG16)},
400 /* dword conditionals adds 5 bytes to frag:
401 1 extra opcode byte, 4 displacement bytes. */
402 {0, 0, 5, 0},
403 /* word conditionals add 4 bytes to frag:
404 1 displacement byte and a 3 byte long branch insn. */
405 {0, 0, 4, 0}
408 static const arch_entry cpu_arch[] = {
409 {"i8086", Cpu086 },
410 {"i186", Cpu086|Cpu186 },
411 {"i286", Cpu086|Cpu186|Cpu286 },
412 {"i386", Cpu086|Cpu186|Cpu286|Cpu386 },
413 {"i486", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486 },
414 {"i586", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
415 {"i686", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
416 {"pentium", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
417 {"pentiumpro",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
418 {"pentium4", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX|CpuSSE|CpuSSE2 },
419 {"k6", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow },
420 {"athlon", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuMMX|Cpu3dnow },
421 {"sledgehammer",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuSledgehammer|CpuMMX|Cpu3dnow|CpuSSE|CpuSSE2 },
422 {NULL, 0 }
425 const pseudo_typeS md_pseudo_table[] =
427 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
428 {"align", s_align_bytes, 0},
429 #else
430 {"align", s_align_ptwo, 0},
431 #endif
432 {"arch", set_cpu_arch, 0},
433 #ifndef I386COFF
434 {"bss", s_bss, 0},
435 #endif
436 {"ffloat", float_cons, 'f'},
437 {"dfloat", float_cons, 'd'},
438 {"tfloat", float_cons, 'x'},
439 {"value", cons, 2},
440 {"noopt", s_ignore, 0},
441 {"optim", s_ignore, 0},
442 {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
443 {"code16", set_code_flag, CODE_16BIT},
444 {"code32", set_code_flag, CODE_32BIT},
445 {"code64", set_code_flag, CODE_64BIT},
446 {"intel_syntax", set_intel_syntax, 1},
447 {"att_syntax", set_intel_syntax, 0},
448 {"file", (void (*) PARAMS ((int))) dwarf2_directive_file, 0},
449 {"loc", dwarf2_directive_loc, 0},
450 #ifdef TE_PE
451 {"secrel32", pe_directive_secrel, 0},
452 #endif
453 {0, 0, 0}
456 /* For interface with expression (). */
457 extern char *input_line_pointer;
459 /* Hash table for instruction mnemonic lookup. */
460 static struct hash_control *op_hash;
462 /* Hash table for register lookup. */
463 static struct hash_control *reg_hash;
465 void
466 i386_align_code (fragP, count)
467 fragS *fragP;
468 int count;
470 /* Various efficient no-op patterns for aligning code labels.
471 Note: Don't try to assemble the instructions in the comments.
472 0L and 0w are not legal. */
473 static const char f32_1[] =
474 {0x90}; /* nop */
475 static const char f32_2[] =
476 {0x89,0xf6}; /* movl %esi,%esi */
477 static const char f32_3[] =
478 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
479 static const char f32_4[] =
480 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
481 static const char f32_5[] =
482 {0x90, /* nop */
483 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
484 static const char f32_6[] =
485 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
486 static const char f32_7[] =
487 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
488 static const char f32_8[] =
489 {0x90, /* nop */
490 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
491 static const char f32_9[] =
492 {0x89,0xf6, /* movl %esi,%esi */
493 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
494 static const char f32_10[] =
495 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
496 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
497 static const char f32_11[] =
498 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
499 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
500 static const char f32_12[] =
501 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
502 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
503 static const char f32_13[] =
504 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
505 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
506 static const char f32_14[] =
507 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
508 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
509 static const char f32_15[] =
510 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
511 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
512 static const char f16_3[] =
513 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
514 static const char f16_4[] =
515 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
516 static const char f16_5[] =
517 {0x90, /* nop */
518 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
519 static const char f16_6[] =
520 {0x89,0xf6, /* mov %si,%si */
521 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
522 static const char f16_7[] =
523 {0x8d,0x74,0x00, /* lea 0(%si),%si */
524 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
525 static const char f16_8[] =
526 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
527 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
528 static const char *const f32_patt[] = {
529 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
530 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
532 static const char *const f16_patt[] = {
533 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
534 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
537 if (count <= 0 || count > 15)
538 return;
540 /* The recommended way to pad 64bit code is to use NOPs preceded by
541 maximally four 0x66 prefixes. Balance the size of nops. */
542 if (flag_code == CODE_64BIT)
544 int i;
545 int nnops = (count + 3) / 4;
546 int len = count / nnops;
547 int remains = count - nnops * len;
548 int pos = 0;
550 for (i = 0; i < remains; i++)
552 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len);
553 fragP->fr_literal[fragP->fr_fix + pos + len] = 0x90;
554 pos += len + 1;
556 for (; i < nnops; i++)
558 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len - 1);
559 fragP->fr_literal[fragP->fr_fix + pos + len - 1] = 0x90;
560 pos += len;
563 else
564 if (flag_code == CODE_16BIT)
566 memcpy (fragP->fr_literal + fragP->fr_fix,
567 f16_patt[count - 1], count);
568 if (count > 8)
569 /* Adjust jump offset. */
570 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
572 else
573 memcpy (fragP->fr_literal + fragP->fr_fix,
574 f32_patt[count - 1], count);
575 fragP->fr_var = count;
578 static INLINE unsigned int
579 mode_from_disp_size (t)
580 unsigned int t;
582 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
585 static INLINE int
586 fits_in_signed_byte (num)
587 offsetT num;
589 return (num >= -128) && (num <= 127);
592 static INLINE int
593 fits_in_unsigned_byte (num)
594 offsetT num;
596 return (num & 0xff) == num;
599 static INLINE int
600 fits_in_unsigned_word (num)
601 offsetT num;
603 return (num & 0xffff) == num;
606 static INLINE int
607 fits_in_signed_word (num)
608 offsetT num;
610 return (-32768 <= num) && (num <= 32767);
612 static INLINE int
613 fits_in_signed_long (num)
614 offsetT num ATTRIBUTE_UNUSED;
616 #ifndef BFD64
617 return 1;
618 #else
619 return (!(((offsetT) -1 << 31) & num)
620 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
621 #endif
622 } /* fits_in_signed_long() */
623 static INLINE int
624 fits_in_unsigned_long (num)
625 offsetT num ATTRIBUTE_UNUSED;
627 #ifndef BFD64
628 return 1;
629 #else
630 return (num & (((offsetT) 2 << 31) - 1)) == num;
631 #endif
632 } /* fits_in_unsigned_long() */
634 static int
635 smallest_imm_type (num)
636 offsetT num;
638 if (cpu_arch_flags != (Cpu086 | Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64))
640 /* This code is disabled on the 486 because all the Imm1 forms
641 in the opcode table are slower on the i486. They're the
642 versions with the implicitly specified single-position
643 displacement, which has another syntax if you really want to
644 use that form. */
645 if (num == 1)
646 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
648 return (fits_in_signed_byte (num)
649 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
650 : fits_in_unsigned_byte (num)
651 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
652 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
653 ? (Imm16 | Imm32 | Imm32S | Imm64)
654 : fits_in_signed_long (num)
655 ? (Imm32 | Imm32S | Imm64)
656 : fits_in_unsigned_long (num)
657 ? (Imm32 | Imm64)
658 : Imm64);
661 static offsetT
662 offset_in_range (val, size)
663 offsetT val;
664 int size;
666 addressT mask;
668 switch (size)
670 case 1: mask = ((addressT) 1 << 8) - 1; break;
671 case 2: mask = ((addressT) 1 << 16) - 1; break;
672 case 4: mask = ((addressT) 2 << 31) - 1; break;
673 #ifdef BFD64
674 case 8: mask = ((addressT) 2 << 63) - 1; break;
675 #endif
676 default: abort ();
679 /* If BFD64, sign extend val. */
680 if (!use_rela_relocations)
681 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
682 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
684 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
686 char buf1[40], buf2[40];
688 sprint_value (buf1, val);
689 sprint_value (buf2, val & mask);
690 as_warn (_("%s shortened to %s"), buf1, buf2);
692 return val & mask;
695 /* Returns 0 if attempting to add a prefix where one from the same
696 class already exists, 1 if non rep/repne added, 2 if rep/repne
697 added. */
698 static int
699 add_prefix (prefix)
700 unsigned int prefix;
702 int ret = 1;
703 int q;
705 if (prefix >= REX_OPCODE && prefix < REX_OPCODE + 16
706 && flag_code == CODE_64BIT)
707 q = REX_PREFIX;
708 else
709 switch (prefix)
711 default:
712 abort ();
714 case CS_PREFIX_OPCODE:
715 case DS_PREFIX_OPCODE:
716 case ES_PREFIX_OPCODE:
717 case FS_PREFIX_OPCODE:
718 case GS_PREFIX_OPCODE:
719 case SS_PREFIX_OPCODE:
720 q = SEG_PREFIX;
721 break;
723 case REPNE_PREFIX_OPCODE:
724 case REPE_PREFIX_OPCODE:
725 ret = 2;
726 /* fall thru */
727 case LOCK_PREFIX_OPCODE:
728 q = LOCKREP_PREFIX;
729 break;
731 case FWAIT_OPCODE:
732 q = WAIT_PREFIX;
733 break;
735 case ADDR_PREFIX_OPCODE:
736 q = ADDR_PREFIX;
737 break;
739 case DATA_PREFIX_OPCODE:
740 q = DATA_PREFIX;
741 break;
744 if (i.prefix[q] != 0)
746 as_bad (_("same type of prefix used twice"));
747 return 0;
750 i.prefixes += 1;
751 i.prefix[q] = prefix;
752 return ret;
755 static void
756 set_code_flag (value)
757 int value;
759 flag_code = value;
760 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
761 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
762 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
764 as_bad (_("64bit mode not supported on this CPU."));
766 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
768 as_bad (_("32bit mode not supported on this CPU."));
770 stackop_size = '\0';
773 static void
774 set_16bit_gcc_code_flag (new_code_flag)
775 int new_code_flag;
777 flag_code = new_code_flag;
778 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
779 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
780 stackop_size = 'l';
783 static void
784 set_intel_syntax (syntax_flag)
785 int syntax_flag;
787 /* Find out if register prefixing is specified. */
788 int ask_naked_reg = 0;
790 SKIP_WHITESPACE ();
791 if (!is_end_of_line[(unsigned char) *input_line_pointer])
793 char *string = input_line_pointer;
794 int e = get_symbol_end ();
796 if (strcmp (string, "prefix") == 0)
797 ask_naked_reg = 1;
798 else if (strcmp (string, "noprefix") == 0)
799 ask_naked_reg = -1;
800 else
801 as_bad (_("bad argument to syntax directive."));
802 *input_line_pointer = e;
804 demand_empty_rest_of_line ();
806 intel_syntax = syntax_flag;
808 if (ask_naked_reg == 0)
809 allow_naked_reg = (intel_syntax
810 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
811 else
812 allow_naked_reg = (ask_naked_reg < 0);
815 static void
816 set_cpu_arch (dummy)
817 int dummy ATTRIBUTE_UNUSED;
819 SKIP_WHITESPACE ();
821 if (!is_end_of_line[(unsigned char) *input_line_pointer])
823 char *string = input_line_pointer;
824 int e = get_symbol_end ();
825 int i;
827 for (i = 0; cpu_arch[i].name; i++)
829 if (strcmp (string, cpu_arch[i].name) == 0)
831 cpu_arch_name = cpu_arch[i].name;
832 cpu_arch_flags = (cpu_arch[i].flags
833 | (flag_code == CODE_64BIT ? Cpu64 : CpuNo64));
834 break;
837 if (!cpu_arch[i].name)
838 as_bad (_("no such architecture: `%s'"), string);
840 *input_line_pointer = e;
842 else
843 as_bad (_("missing cpu architecture"));
845 no_cond_jump_promotion = 0;
846 if (*input_line_pointer == ','
847 && !is_end_of_line[(unsigned char) input_line_pointer[1]])
849 char *string = ++input_line_pointer;
850 int e = get_symbol_end ();
852 if (strcmp (string, "nojumps") == 0)
853 no_cond_jump_promotion = 1;
854 else if (strcmp (string, "jumps") == 0)
856 else
857 as_bad (_("no such architecture modifier: `%s'"), string);
859 *input_line_pointer = e;
862 demand_empty_rest_of_line ();
865 unsigned long
866 i386_mach ()
868 if (!strcmp (default_arch, "x86_64"))
869 return bfd_mach_x86_64;
870 else if (!strcmp (default_arch, "i386"))
871 return bfd_mach_i386_i386;
872 else
873 as_fatal (_("Unknown architecture"));
876 void
877 md_begin ()
879 const char *hash_err;
881 /* Initialize op_hash hash table. */
882 op_hash = hash_new ();
885 const template *optab;
886 templates *core_optab;
888 /* Setup for loop. */
889 optab = i386_optab;
890 core_optab = (templates *) xmalloc (sizeof (templates));
891 core_optab->start = optab;
893 while (1)
895 ++optab;
896 if (optab->name == NULL
897 || strcmp (optab->name, (optab - 1)->name) != 0)
899 /* different name --> ship out current template list;
900 add to hash table; & begin anew. */
901 core_optab->end = optab;
902 hash_err = hash_insert (op_hash,
903 (optab - 1)->name,
904 (PTR) core_optab);
905 if (hash_err)
907 as_fatal (_("Internal Error: Can't hash %s: %s"),
908 (optab - 1)->name,
909 hash_err);
911 if (optab->name == NULL)
912 break;
913 core_optab = (templates *) xmalloc (sizeof (templates));
914 core_optab->start = optab;
919 /* Initialize reg_hash hash table. */
920 reg_hash = hash_new ();
922 const reg_entry *regtab;
924 for (regtab = i386_regtab;
925 regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
926 regtab++)
928 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
929 if (hash_err)
930 as_fatal (_("Internal Error: Can't hash %s: %s"),
931 regtab->reg_name,
932 hash_err);
936 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
938 int c;
939 char *p;
941 for (c = 0; c < 256; c++)
943 if (ISDIGIT (c))
945 digit_chars[c] = c;
946 mnemonic_chars[c] = c;
947 register_chars[c] = c;
948 operand_chars[c] = c;
950 else if (ISLOWER (c))
952 mnemonic_chars[c] = c;
953 register_chars[c] = c;
954 operand_chars[c] = c;
956 else if (ISUPPER (c))
958 mnemonic_chars[c] = TOLOWER (c);
959 register_chars[c] = mnemonic_chars[c];
960 operand_chars[c] = c;
963 if (ISALPHA (c) || ISDIGIT (c))
964 identifier_chars[c] = c;
965 else if (c >= 128)
967 identifier_chars[c] = c;
968 operand_chars[c] = c;
972 #ifdef LEX_AT
973 identifier_chars['@'] = '@';
974 #endif
975 digit_chars['-'] = '-';
976 identifier_chars['_'] = '_';
977 identifier_chars['.'] = '.';
979 for (p = operand_special_chars; *p != '\0'; p++)
980 operand_chars[(unsigned char) *p] = *p;
983 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
984 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
986 record_alignment (text_section, 2);
987 record_alignment (data_section, 2);
988 record_alignment (bss_section, 2);
990 #endif
992 if (flag_code == CODE_64BIT)
994 x86_dwarf2_return_column = 16;
995 x86_cie_data_alignment = -8;
997 else
999 x86_dwarf2_return_column = 8;
1000 x86_cie_data_alignment = -4;
1004 void
1005 i386_print_statistics (file)
1006 FILE *file;
1008 hash_print_statistics (file, "i386 opcode", op_hash);
1009 hash_print_statistics (file, "i386 register", reg_hash);
1012 #ifdef DEBUG386
1014 /* Debugging routines for md_assemble. */
1015 static void pi PARAMS ((char *, i386_insn *));
1016 static void pte PARAMS ((template *));
1017 static void pt PARAMS ((unsigned int));
1018 static void pe PARAMS ((expressionS *));
1019 static void ps PARAMS ((symbolS *));
1021 static void
1022 pi (line, x)
1023 char *line;
1024 i386_insn *x;
1026 unsigned int i;
1028 fprintf (stdout, "%s: template ", line);
1029 pte (&x->tm);
1030 fprintf (stdout, " address: base %s index %s scale %x\n",
1031 x->base_reg ? x->base_reg->reg_name : "none",
1032 x->index_reg ? x->index_reg->reg_name : "none",
1033 x->log2_scale_factor);
1034 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
1035 x->rm.mode, x->rm.reg, x->rm.regmem);
1036 fprintf (stdout, " sib: base %x index %x scale %x\n",
1037 x->sib.base, x->sib.index, x->sib.scale);
1038 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
1039 (x->rex & REX_MODE64) != 0,
1040 (x->rex & REX_EXTX) != 0,
1041 (x->rex & REX_EXTY) != 0,
1042 (x->rex & REX_EXTZ) != 0);
1043 for (i = 0; i < x->operands; i++)
1045 fprintf (stdout, " #%d: ", i + 1);
1046 pt (x->types[i]);
1047 fprintf (stdout, "\n");
1048 if (x->types[i]
1049 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
1050 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
1051 if (x->types[i] & Imm)
1052 pe (x->op[i].imms);
1053 if (x->types[i] & Disp)
1054 pe (x->op[i].disps);
1058 static void
1059 pte (t)
1060 template *t;
1062 unsigned int i;
1063 fprintf (stdout, " %d operands ", t->operands);
1064 fprintf (stdout, "opcode %x ", t->base_opcode);
1065 if (t->extension_opcode != None)
1066 fprintf (stdout, "ext %x ", t->extension_opcode);
1067 if (t->opcode_modifier & D)
1068 fprintf (stdout, "D");
1069 if (t->opcode_modifier & W)
1070 fprintf (stdout, "W");
1071 fprintf (stdout, "\n");
1072 for (i = 0; i < t->operands; i++)
1074 fprintf (stdout, " #%d type ", i + 1);
1075 pt (t->operand_types[i]);
1076 fprintf (stdout, "\n");
1080 static void
1081 pe (e)
1082 expressionS *e;
1084 fprintf (stdout, " operation %d\n", e->X_op);
1085 fprintf (stdout, " add_number %ld (%lx)\n",
1086 (long) e->X_add_number, (long) e->X_add_number);
1087 if (e->X_add_symbol)
1089 fprintf (stdout, " add_symbol ");
1090 ps (e->X_add_symbol);
1091 fprintf (stdout, "\n");
1093 if (e->X_op_symbol)
1095 fprintf (stdout, " op_symbol ");
1096 ps (e->X_op_symbol);
1097 fprintf (stdout, "\n");
1101 static void
1102 ps (s)
1103 symbolS *s;
1105 fprintf (stdout, "%s type %s%s",
1106 S_GET_NAME (s),
1107 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
1108 segment_name (S_GET_SEGMENT (s)));
1111 struct type_name
1113 unsigned int mask;
1114 char *tname;
1117 static const type_names[] =
1119 { Reg8, "r8" },
1120 { Reg16, "r16" },
1121 { Reg32, "r32" },
1122 { Reg64, "r64" },
1123 { Imm8, "i8" },
1124 { Imm8S, "i8s" },
1125 { Imm16, "i16" },
1126 { Imm32, "i32" },
1127 { Imm32S, "i32s" },
1128 { Imm64, "i64" },
1129 { Imm1, "i1" },
1130 { BaseIndex, "BaseIndex" },
1131 { Disp8, "d8" },
1132 { Disp16, "d16" },
1133 { Disp32, "d32" },
1134 { Disp32S, "d32s" },
1135 { Disp64, "d64" },
1136 { InOutPortReg, "InOutPortReg" },
1137 { ShiftCount, "ShiftCount" },
1138 { Control, "control reg" },
1139 { Test, "test reg" },
1140 { Debug, "debug reg" },
1141 { FloatReg, "FReg" },
1142 { FloatAcc, "FAcc" },
1143 { SReg2, "SReg2" },
1144 { SReg3, "SReg3" },
1145 { Acc, "Acc" },
1146 { JumpAbsolute, "Jump Absolute" },
1147 { RegMMX, "rMMX" },
1148 { RegXMM, "rXMM" },
1149 { EsSeg, "es" },
1150 { 0, "" }
1153 static void
1154 pt (t)
1155 unsigned int t;
1157 const struct type_name *ty;
1159 for (ty = type_names; ty->mask; ty++)
1160 if (t & ty->mask)
1161 fprintf (stdout, "%s, ", ty->tname);
1162 fflush (stdout);
1165 #endif /* DEBUG386 */
1167 static bfd_reloc_code_real_type reloc
1168 PARAMS ((int, int, int, bfd_reloc_code_real_type));
1170 static bfd_reloc_code_real_type
1171 reloc (size, pcrel, sign, other)
1172 int size;
1173 int pcrel;
1174 int sign;
1175 bfd_reloc_code_real_type other;
1177 if (other != NO_RELOC)
1178 return other;
1180 if (pcrel)
1182 if (!sign)
1183 as_bad (_("There are no unsigned pc-relative relocations"));
1184 switch (size)
1186 case 1: return BFD_RELOC_8_PCREL;
1187 case 2: return BFD_RELOC_16_PCREL;
1188 case 4: return BFD_RELOC_32_PCREL;
1190 as_bad (_("can not do %d byte pc-relative relocation"), size);
1192 else
1194 if (sign)
1195 switch (size)
1197 case 4: return BFD_RELOC_X86_64_32S;
1199 else
1200 switch (size)
1202 case 1: return BFD_RELOC_8;
1203 case 2: return BFD_RELOC_16;
1204 case 4: return BFD_RELOC_32;
1205 case 8: return BFD_RELOC_64;
1207 as_bad (_("can not do %s %d byte relocation"),
1208 sign ? "signed" : "unsigned", size);
1211 abort ();
1212 return BFD_RELOC_NONE;
1215 /* Here we decide which fixups can be adjusted to make them relative to
1216 the beginning of the section instead of the symbol. Basically we need
1217 to make sure that the dynamic relocations are done correctly, so in
1218 some cases we force the original symbol to be used. */
1221 tc_i386_fix_adjustable (fixP)
1222 fixS *fixP ATTRIBUTE_UNUSED;
1224 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1225 if (OUTPUT_FLAVOR != bfd_target_elf_flavour)
1226 return 1;
1228 /* Don't adjust pc-relative references to merge sections in 64-bit
1229 mode. */
1230 if (use_rela_relocations
1231 && (S_GET_SEGMENT (fixP->fx_addsy)->flags & SEC_MERGE) != 0
1232 && fixP->fx_pcrel)
1233 return 0;
1235 /* The x86_64 GOTPCREL are represented as 32bit PCrel relocations
1236 and changed later by validate_fix. */
1237 if (GOT_symbol && fixP->fx_subsy == GOT_symbol
1238 && fixP->fx_r_type == BFD_RELOC_32_PCREL)
1239 return 0;
1241 /* adjust_reloc_syms doesn't know about the GOT. */
1242 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1243 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1244 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1245 || fixP->fx_r_type == BFD_RELOC_386_TLS_GD
1246 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDM
1247 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDO_32
1248 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE_32
1249 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE
1250 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTIE
1251 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE_32
1252 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE
1253 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1254 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1255 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
1256 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSGD
1257 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSLD
1258 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF32
1259 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTTPOFF
1260 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF32
1261 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1262 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1263 return 0;
1264 #endif
1265 return 1;
1268 static int intel_float_operand PARAMS ((const char *mnemonic));
1270 static int
1271 intel_float_operand (mnemonic)
1272 const char *mnemonic;
1274 if (mnemonic[0] == 'f' && mnemonic[1] == 'i')
1275 return 2;
1277 if (mnemonic[0] == 'f')
1278 return 1;
1280 return 0;
1283 /* This is the guts of the machine-dependent assembler. LINE points to a
1284 machine dependent instruction. This function is supposed to emit
1285 the frags/bytes it assembles to. */
1287 void
1288 md_assemble (line)
1289 char *line;
1291 int j;
1292 char mnemonic[MAX_MNEM_SIZE];
1294 /* Initialize globals. */
1295 memset (&i, '\0', sizeof (i));
1296 for (j = 0; j < MAX_OPERANDS; j++)
1297 i.reloc[j] = NO_RELOC;
1298 memset (disp_expressions, '\0', sizeof (disp_expressions));
1299 memset (im_expressions, '\0', sizeof (im_expressions));
1300 save_stack_p = save_stack;
1302 /* First parse an instruction mnemonic & call i386_operand for the operands.
1303 We assume that the scrubber has arranged it so that line[0] is the valid
1304 start of a (possibly prefixed) mnemonic. */
1306 line = parse_insn (line, mnemonic);
1307 if (line == NULL)
1308 return;
1310 line = parse_operands (line, mnemonic);
1311 if (line == NULL)
1312 return;
1314 /* Now we've parsed the mnemonic into a set of templates, and have the
1315 operands at hand. */
1317 /* All intel opcodes have reversed operands except for "bound" and
1318 "enter". We also don't reverse intersegment "jmp" and "call"
1319 instructions with 2 immediate operands so that the immediate segment
1320 precedes the offset, as it does when in AT&T mode. "enter" and the
1321 intersegment "jmp" and "call" instructions are the only ones that
1322 have two immediate operands. */
1323 if (intel_syntax && i.operands > 1
1324 && (strcmp (mnemonic, "bound") != 0)
1325 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1326 swap_operands ();
1328 if (i.imm_operands)
1329 optimize_imm ();
1331 if (i.disp_operands)
1332 optimize_disp ();
1334 /* Next, we find a template that matches the given insn,
1335 making sure the overlap of the given operands types is consistent
1336 with the template operand types. */
1338 if (!match_template ())
1339 return;
1341 if (intel_syntax)
1343 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1344 if (SYSV386_COMPAT
1345 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1346 i.tm.base_opcode ^= FloatR;
1348 /* Zap movzx and movsx suffix. The suffix may have been set from
1349 "word ptr" or "byte ptr" on the source operand, but we'll use
1350 the suffix later to choose the destination register. */
1351 if ((i.tm.base_opcode & ~9) == 0x0fb6)
1352 i.suffix = 0;
1355 if (i.tm.opcode_modifier & FWait)
1356 if (!add_prefix (FWAIT_OPCODE))
1357 return;
1359 /* Check string instruction segment overrides. */
1360 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1362 if (!check_string ())
1363 return;
1366 if (!process_suffix ())
1367 return;
1369 /* Make still unresolved immediate matches conform to size of immediate
1370 given in i.suffix. */
1371 if (!finalize_imm ())
1372 return;
1374 if (i.types[0] & Imm1)
1375 i.imm_operands = 0; /* kludge for shift insns. */
1376 if (i.types[0] & ImplicitRegister)
1377 i.reg_operands--;
1378 if (i.types[1] & ImplicitRegister)
1379 i.reg_operands--;
1380 if (i.types[2] & ImplicitRegister)
1381 i.reg_operands--;
1383 if (i.tm.opcode_modifier & ImmExt)
1385 expressionS *exp;
1387 if ((i.tm.cpu_flags & CpuPNI) && i.operands > 0)
1389 /* These Intel Prescott New Instructions have the fixed
1390 operands with an opcode suffix which is coded in the same
1391 place as an 8-bit immediate field would be. Here we check
1392 those operands and remove them afterwards. */
1393 unsigned int x;
1395 for (x = 0; x < i.operands; x++)
1396 if (i.op[x].regs->reg_num != x)
1397 as_bad (_("can't use register '%%%s' as operand %d in '%s'."),
1398 i.op[x].regs->reg_name, x + 1, i.tm.name);
1399 i.operands = 0;
1402 /* These AMD 3DNow! and Intel Katmai New Instructions have an
1403 opcode suffix which is coded in the same place as an 8-bit
1404 immediate field would be. Here we fake an 8-bit immediate
1405 operand from the opcode suffix stored in tm.extension_opcode. */
1407 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
1409 exp = &im_expressions[i.imm_operands++];
1410 i.op[i.operands].imms = exp;
1411 i.types[i.operands++] = Imm8;
1412 exp->X_op = O_constant;
1413 exp->X_add_number = i.tm.extension_opcode;
1414 i.tm.extension_opcode = None;
1417 /* For insns with operands there are more diddles to do to the opcode. */
1418 if (i.operands)
1420 if (!process_operands ())
1421 return;
1423 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
1425 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
1426 as_warn (_("translating to `%sp'"), i.tm.name);
1429 /* Handle conversion of 'int $3' --> special int3 insn. */
1430 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
1432 i.tm.base_opcode = INT3_OPCODE;
1433 i.imm_operands = 0;
1436 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
1437 && i.op[0].disps->X_op == O_constant)
1439 /* Convert "jmp constant" (and "call constant") to a jump (call) to
1440 the absolute address given by the constant. Since ix86 jumps and
1441 calls are pc relative, we need to generate a reloc. */
1442 i.op[0].disps->X_add_symbol = &abs_symbol;
1443 i.op[0].disps->X_op = O_symbol;
1446 if ((i.tm.opcode_modifier & Rex64) != 0)
1447 i.rex |= REX_MODE64;
1449 /* For 8 bit registers we need an empty rex prefix. Also if the
1450 instruction already has a prefix, we need to convert old
1451 registers to new ones. */
1453 if (((i.types[0] & Reg8) != 0
1454 && (i.op[0].regs->reg_flags & RegRex64) != 0)
1455 || ((i.types[1] & Reg8) != 0
1456 && (i.op[1].regs->reg_flags & RegRex64) != 0)
1457 || (((i.types[0] & Reg8) != 0 || (i.types[1] & Reg8) != 0)
1458 && i.rex != 0))
1460 int x;
1462 i.rex |= REX_OPCODE;
1463 for (x = 0; x < 2; x++)
1465 /* Look for 8 bit operand that uses old registers. */
1466 if ((i.types[x] & Reg8) != 0
1467 && (i.op[x].regs->reg_flags & RegRex64) == 0)
1469 /* In case it is "hi" register, give up. */
1470 if (i.op[x].regs->reg_num > 3)
1471 as_bad (_("can't encode register '%%%s' in an instruction requiring REX prefix.\n"),
1472 i.op[x].regs->reg_name);
1474 /* Otherwise it is equivalent to the extended register.
1475 Since the encoding doesn't change this is merely
1476 cosmetic cleanup for debug output. */
1478 i.op[x].regs = i.op[x].regs + 8;
1483 if (i.rex != 0)
1484 add_prefix (REX_OPCODE | i.rex);
1486 /* We are ready to output the insn. */
1487 output_insn ();
1490 static char *
1491 parse_insn (line, mnemonic)
1492 char *line;
1493 char *mnemonic;
1495 char *l = line;
1496 char *token_start = l;
1497 char *mnem_p;
1499 /* Non-zero if we found a prefix only acceptable with string insns. */
1500 const char *expecting_string_instruction = NULL;
1502 while (1)
1504 mnem_p = mnemonic;
1505 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1507 mnem_p++;
1508 if (mnem_p >= mnemonic + MAX_MNEM_SIZE)
1510 as_bad (_("no such instruction: `%s'"), token_start);
1511 return NULL;
1513 l++;
1515 if (!is_space_char (*l)
1516 && *l != END_OF_INSN
1517 && *l != PREFIX_SEPARATOR
1518 && *l != ',')
1520 as_bad (_("invalid character %s in mnemonic"),
1521 output_invalid (*l));
1522 return NULL;
1524 if (token_start == l)
1526 if (*l == PREFIX_SEPARATOR)
1527 as_bad (_("expecting prefix; got nothing"));
1528 else
1529 as_bad (_("expecting mnemonic; got nothing"));
1530 return NULL;
1533 /* Look up instruction (or prefix) via hash table. */
1534 current_templates = hash_find (op_hash, mnemonic);
1536 if (*l != END_OF_INSN
1537 && (!is_space_char (*l) || l[1] != END_OF_INSN)
1538 && current_templates
1539 && (current_templates->start->opcode_modifier & IsPrefix))
1541 /* If we are in 16-bit mode, do not allow addr16 or data16.
1542 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1543 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1544 && flag_code != CODE_64BIT
1545 && (((current_templates->start->opcode_modifier & Size32) != 0)
1546 ^ (flag_code == CODE_16BIT)))
1548 as_bad (_("redundant %s prefix"),
1549 current_templates->start->name);
1550 return NULL;
1552 /* Add prefix, checking for repeated prefixes. */
1553 switch (add_prefix (current_templates->start->base_opcode))
1555 case 0:
1556 return NULL;
1557 case 2:
1558 expecting_string_instruction = current_templates->start->name;
1559 break;
1561 /* Skip past PREFIX_SEPARATOR and reset token_start. */
1562 token_start = ++l;
1564 else
1565 break;
1568 if (!current_templates)
1570 /* See if we can get a match by trimming off a suffix. */
1571 switch (mnem_p[-1])
1573 case WORD_MNEM_SUFFIX:
1574 case BYTE_MNEM_SUFFIX:
1575 case QWORD_MNEM_SUFFIX:
1576 i.suffix = mnem_p[-1];
1577 mnem_p[-1] = '\0';
1578 current_templates = hash_find (op_hash, mnemonic);
1579 break;
1580 case SHORT_MNEM_SUFFIX:
1581 case LONG_MNEM_SUFFIX:
1582 if (!intel_syntax)
1584 i.suffix = mnem_p[-1];
1585 mnem_p[-1] = '\0';
1586 current_templates = hash_find (op_hash, mnemonic);
1588 break;
1590 /* Intel Syntax. */
1591 case 'd':
1592 if (intel_syntax)
1594 if (intel_float_operand (mnemonic))
1595 i.suffix = SHORT_MNEM_SUFFIX;
1596 else
1597 i.suffix = LONG_MNEM_SUFFIX;
1598 mnem_p[-1] = '\0';
1599 current_templates = hash_find (op_hash, mnemonic);
1601 break;
1603 if (!current_templates)
1605 as_bad (_("no such instruction: `%s'"), token_start);
1606 return NULL;
1610 if (current_templates->start->opcode_modifier & (Jump | JumpByte))
1612 /* Check for a branch hint. We allow ",pt" and ",pn" for
1613 predict taken and predict not taken respectively.
1614 I'm not sure that branch hints actually do anything on loop
1615 and jcxz insns (JumpByte) for current Pentium4 chips. They
1616 may work in the future and it doesn't hurt to accept them
1617 now. */
1618 if (l[0] == ',' && l[1] == 'p')
1620 if (l[2] == 't')
1622 if (!add_prefix (DS_PREFIX_OPCODE))
1623 return NULL;
1624 l += 3;
1626 else if (l[2] == 'n')
1628 if (!add_prefix (CS_PREFIX_OPCODE))
1629 return NULL;
1630 l += 3;
1634 /* Any other comma loses. */
1635 if (*l == ',')
1637 as_bad (_("invalid character %s in mnemonic"),
1638 output_invalid (*l));
1639 return NULL;
1642 /* Check if instruction is supported on specified architecture. */
1643 if ((current_templates->start->cpu_flags & ~(Cpu64 | CpuNo64))
1644 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64)))
1646 as_warn (_("`%s' is not supported on `%s'"),
1647 current_templates->start->name, cpu_arch_name);
1649 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
1651 as_warn (_("use .code16 to ensure correct addressing mode"));
1654 /* Check for rep/repne without a string instruction. */
1655 if (expecting_string_instruction
1656 && !(current_templates->start->opcode_modifier & IsString))
1658 as_bad (_("expecting string instruction after `%s'"),
1659 expecting_string_instruction);
1660 return NULL;
1663 return l;
1666 static char *
1667 parse_operands (l, mnemonic)
1668 char *l;
1669 const char *mnemonic;
1671 char *token_start;
1673 /* 1 if operand is pending after ','. */
1674 unsigned int expecting_operand = 0;
1676 /* Non-zero if operand parens not balanced. */
1677 unsigned int paren_not_balanced;
1679 while (*l != END_OF_INSN)
1681 /* Skip optional white space before operand. */
1682 if (is_space_char (*l))
1683 ++l;
1684 if (!is_operand_char (*l) && *l != END_OF_INSN)
1686 as_bad (_("invalid character %s before operand %d"),
1687 output_invalid (*l),
1688 i.operands + 1);
1689 return NULL;
1691 token_start = l; /* after white space */
1692 paren_not_balanced = 0;
1693 while (paren_not_balanced || *l != ',')
1695 if (*l == END_OF_INSN)
1697 if (paren_not_balanced)
1699 if (!intel_syntax)
1700 as_bad (_("unbalanced parenthesis in operand %d."),
1701 i.operands + 1);
1702 else
1703 as_bad (_("unbalanced brackets in operand %d."),
1704 i.operands + 1);
1705 return NULL;
1707 else
1708 break; /* we are done */
1710 else if (!is_operand_char (*l) && !is_space_char (*l))
1712 as_bad (_("invalid character %s in operand %d"),
1713 output_invalid (*l),
1714 i.operands + 1);
1715 return NULL;
1717 if (!intel_syntax)
1719 if (*l == '(')
1720 ++paren_not_balanced;
1721 if (*l == ')')
1722 --paren_not_balanced;
1724 else
1726 if (*l == '[')
1727 ++paren_not_balanced;
1728 if (*l == ']')
1729 --paren_not_balanced;
1731 l++;
1733 if (l != token_start)
1734 { /* Yes, we've read in another operand. */
1735 unsigned int operand_ok;
1736 this_operand = i.operands++;
1737 if (i.operands > MAX_OPERANDS)
1739 as_bad (_("spurious operands; (%d operands/instruction max)"),
1740 MAX_OPERANDS);
1741 return NULL;
1743 /* Now parse operand adding info to 'i' as we go along. */
1744 END_STRING_AND_SAVE (l);
1746 if (intel_syntax)
1747 operand_ok =
1748 i386_intel_operand (token_start,
1749 intel_float_operand (mnemonic));
1750 else
1751 operand_ok = i386_operand (token_start);
1753 RESTORE_END_STRING (l);
1754 if (!operand_ok)
1755 return NULL;
1757 else
1759 if (expecting_operand)
1761 expecting_operand_after_comma:
1762 as_bad (_("expecting operand after ','; got nothing"));
1763 return NULL;
1765 if (*l == ',')
1767 as_bad (_("expecting operand before ','; got nothing"));
1768 return NULL;
1772 /* Now *l must be either ',' or END_OF_INSN. */
1773 if (*l == ',')
1775 if (*++l == END_OF_INSN)
1777 /* Just skip it, if it's \n complain. */
1778 goto expecting_operand_after_comma;
1780 expecting_operand = 1;
1783 return l;
1786 static void
1787 swap_operands ()
1789 union i386_op temp_op;
1790 unsigned int temp_type;
1791 enum bfd_reloc_code_real temp_reloc;
1792 int xchg1 = 0;
1793 int xchg2 = 0;
1795 if (i.operands == 2)
1797 xchg1 = 0;
1798 xchg2 = 1;
1800 else if (i.operands == 3)
1802 xchg1 = 0;
1803 xchg2 = 2;
1805 temp_type = i.types[xchg2];
1806 i.types[xchg2] = i.types[xchg1];
1807 i.types[xchg1] = temp_type;
1808 temp_op = i.op[xchg2];
1809 i.op[xchg2] = i.op[xchg1];
1810 i.op[xchg1] = temp_op;
1811 temp_reloc = i.reloc[xchg2];
1812 i.reloc[xchg2] = i.reloc[xchg1];
1813 i.reloc[xchg1] = temp_reloc;
1815 if (i.mem_operands == 2)
1817 const seg_entry *temp_seg;
1818 temp_seg = i.seg[0];
1819 i.seg[0] = i.seg[1];
1820 i.seg[1] = temp_seg;
1824 /* Try to ensure constant immediates are represented in the smallest
1825 opcode possible. */
1826 static void
1827 optimize_imm ()
1829 char guess_suffix = 0;
1830 int op;
1832 if (i.suffix)
1833 guess_suffix = i.suffix;
1834 else if (i.reg_operands)
1836 /* Figure out a suffix from the last register operand specified.
1837 We can't do this properly yet, ie. excluding InOutPortReg,
1838 but the following works for instructions with immediates.
1839 In any case, we can't set i.suffix yet. */
1840 for (op = i.operands; --op >= 0;)
1841 if (i.types[op] & Reg)
1843 if (i.types[op] & Reg8)
1844 guess_suffix = BYTE_MNEM_SUFFIX;
1845 else if (i.types[op] & Reg16)
1846 guess_suffix = WORD_MNEM_SUFFIX;
1847 else if (i.types[op] & Reg32)
1848 guess_suffix = LONG_MNEM_SUFFIX;
1849 else if (i.types[op] & Reg64)
1850 guess_suffix = QWORD_MNEM_SUFFIX;
1851 break;
1854 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
1855 guess_suffix = WORD_MNEM_SUFFIX;
1857 for (op = i.operands; --op >= 0;)
1858 if (i.types[op] & Imm)
1860 switch (i.op[op].imms->X_op)
1862 case O_constant:
1863 /* If a suffix is given, this operand may be shortened. */
1864 switch (guess_suffix)
1866 case LONG_MNEM_SUFFIX:
1867 i.types[op] |= Imm32 | Imm64;
1868 break;
1869 case WORD_MNEM_SUFFIX:
1870 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
1871 break;
1872 case BYTE_MNEM_SUFFIX:
1873 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
1874 break;
1877 /* If this operand is at most 16 bits, convert it
1878 to a signed 16 bit number before trying to see
1879 whether it will fit in an even smaller size.
1880 This allows a 16-bit operand such as $0xffe0 to
1881 be recognised as within Imm8S range. */
1882 if ((i.types[op] & Imm16)
1883 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
1885 i.op[op].imms->X_add_number =
1886 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
1888 if ((i.types[op] & Imm32)
1889 && ((i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1))
1890 == 0))
1892 i.op[op].imms->X_add_number = ((i.op[op].imms->X_add_number
1893 ^ ((offsetT) 1 << 31))
1894 - ((offsetT) 1 << 31));
1896 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
1898 /* We must avoid matching of Imm32 templates when 64bit
1899 only immediate is available. */
1900 if (guess_suffix == QWORD_MNEM_SUFFIX)
1901 i.types[op] &= ~Imm32;
1902 break;
1904 case O_absent:
1905 case O_register:
1906 abort ();
1908 /* Symbols and expressions. */
1909 default:
1910 /* Convert symbolic operand to proper sizes for matching. */
1911 switch (guess_suffix)
1913 case QWORD_MNEM_SUFFIX:
1914 i.types[op] = Imm64 | Imm32S;
1915 break;
1916 case LONG_MNEM_SUFFIX:
1917 i.types[op] = Imm32 | Imm64;
1918 break;
1919 case WORD_MNEM_SUFFIX:
1920 i.types[op] = Imm16 | Imm32 | Imm64;
1921 break;
1922 break;
1923 case BYTE_MNEM_SUFFIX:
1924 i.types[op] = Imm8 | Imm8S | Imm16 | Imm32S | Imm32;
1925 break;
1926 break;
1928 break;
1933 /* Try to use the smallest displacement type too. */
1934 static void
1935 optimize_disp ()
1937 int op;
1939 for (op = i.operands; --op >= 0;)
1940 if ((i.types[op] & Disp) && i.op[op].disps->X_op == O_constant)
1942 offsetT disp = i.op[op].disps->X_add_number;
1944 if (i.types[op] & Disp16)
1946 /* We know this operand is at most 16 bits, so
1947 convert to a signed 16 bit number before trying
1948 to see whether it will fit in an even smaller
1949 size. */
1951 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
1953 else if (i.types[op] & Disp32)
1955 /* We know this operand is at most 32 bits, so convert to a
1956 signed 32 bit number before trying to see whether it will
1957 fit in an even smaller size. */
1958 disp &= (((offsetT) 2 << 31) - 1);
1959 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1961 if (flag_code == CODE_64BIT)
1963 if (fits_in_signed_long (disp))
1964 i.types[op] |= Disp32S;
1965 if (fits_in_unsigned_long (disp))
1966 i.types[op] |= Disp32;
1968 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
1969 && fits_in_signed_byte (disp))
1970 i.types[op] |= Disp8;
1974 static int
1975 match_template ()
1977 /* Points to template once we've found it. */
1978 const template *t;
1979 unsigned int overlap0, overlap1, overlap2;
1980 unsigned int found_reverse_match;
1981 int suffix_check;
1983 #define MATCH(overlap, given, template) \
1984 ((overlap & ~JumpAbsolute) \
1985 && (((given) & (BaseIndex | JumpAbsolute)) \
1986 == ((overlap) & (BaseIndex | JumpAbsolute))))
1988 /* If given types r0 and r1 are registers they must be of the same type
1989 unless the expected operand type register overlap is null.
1990 Note that Acc in a template matches every size of reg. */
1991 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
1992 (((g0) & Reg) == 0 || ((g1) & Reg) == 0 \
1993 || ((g0) & Reg) == ((g1) & Reg) \
1994 || ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
1996 overlap0 = 0;
1997 overlap1 = 0;
1998 overlap2 = 0;
1999 found_reverse_match = 0;
2000 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
2001 ? No_bSuf
2002 : (i.suffix == WORD_MNEM_SUFFIX
2003 ? No_wSuf
2004 : (i.suffix == SHORT_MNEM_SUFFIX
2005 ? No_sSuf
2006 : (i.suffix == LONG_MNEM_SUFFIX
2007 ? No_lSuf
2008 : (i.suffix == QWORD_MNEM_SUFFIX
2009 ? No_qSuf
2010 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX
2011 ? No_xSuf : 0))))));
2013 for (t = current_templates->start;
2014 t < current_templates->end;
2015 t++)
2017 /* Must have right number of operands. */
2018 if (i.operands != t->operands)
2019 continue;
2021 /* Check the suffix, except for some instructions in intel mode. */
2022 if ((t->opcode_modifier & suffix_check)
2023 && !(intel_syntax
2024 && (t->opcode_modifier & IgnoreSize))
2025 && !(intel_syntax
2026 && t->base_opcode == 0xd9
2027 && (t->extension_opcode == 5 /* 0xd9,5 "fldcw" */
2028 || t->extension_opcode == 7))) /* 0xd9,7 "f{n}stcw" */
2029 continue;
2031 /* Do not verify operands when there are none. */
2032 else if (!t->operands)
2034 if (t->cpu_flags & ~cpu_arch_flags)
2035 continue;
2036 /* We've found a match; break out of loop. */
2037 break;
2040 overlap0 = i.types[0] & t->operand_types[0];
2041 switch (t->operands)
2043 case 1:
2044 if (!MATCH (overlap0, i.types[0], t->operand_types[0]))
2045 continue;
2046 break;
2047 case 2:
2048 case 3:
2049 overlap1 = i.types[1] & t->operand_types[1];
2050 if (!MATCH (overlap0, i.types[0], t->operand_types[0])
2051 || !MATCH (overlap1, i.types[1], t->operand_types[1])
2052 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2053 t->operand_types[0],
2054 overlap1, i.types[1],
2055 t->operand_types[1]))
2057 /* Check if other direction is valid ... */
2058 if ((t->opcode_modifier & (D | FloatD)) == 0)
2059 continue;
2061 /* Try reversing direction of operands. */
2062 overlap0 = i.types[0] & t->operand_types[1];
2063 overlap1 = i.types[1] & t->operand_types[0];
2064 if (!MATCH (overlap0, i.types[0], t->operand_types[1])
2065 || !MATCH (overlap1, i.types[1], t->operand_types[0])
2066 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2067 t->operand_types[1],
2068 overlap1, i.types[1],
2069 t->operand_types[0]))
2071 /* Does not match either direction. */
2072 continue;
2074 /* found_reverse_match holds which of D or FloatDR
2075 we've found. */
2076 found_reverse_match = t->opcode_modifier & (D | FloatDR);
2078 /* Found a forward 2 operand match here. */
2079 else if (t->operands == 3)
2081 /* Here we make use of the fact that there are no
2082 reverse match 3 operand instructions, and all 3
2083 operand instructions only need to be checked for
2084 register consistency between operands 2 and 3. */
2085 overlap2 = i.types[2] & t->operand_types[2];
2086 if (!MATCH (overlap2, i.types[2], t->operand_types[2])
2087 || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1],
2088 t->operand_types[1],
2089 overlap2, i.types[2],
2090 t->operand_types[2]))
2092 continue;
2094 /* Found either forward/reverse 2 or 3 operand match here:
2095 slip through to break. */
2097 if (t->cpu_flags & ~cpu_arch_flags)
2099 found_reverse_match = 0;
2100 continue;
2102 /* We've found a match; break out of loop. */
2103 break;
2106 if (t == current_templates->end)
2108 /* We found no match. */
2109 as_bad (_("suffix or operands invalid for `%s'"),
2110 current_templates->start->name);
2111 return 0;
2114 if (!quiet_warnings)
2116 if (!intel_syntax
2117 && ((i.types[0] & JumpAbsolute)
2118 != (t->operand_types[0] & JumpAbsolute)))
2120 as_warn (_("indirect %s without `*'"), t->name);
2123 if ((t->opcode_modifier & (IsPrefix | IgnoreSize))
2124 == (IsPrefix | IgnoreSize))
2126 /* Warn them that a data or address size prefix doesn't
2127 affect assembly of the next line of code. */
2128 as_warn (_("stand-alone `%s' prefix"), t->name);
2132 /* Copy the template we found. */
2133 i.tm = *t;
2134 if (found_reverse_match)
2136 /* If we found a reverse match we must alter the opcode
2137 direction bit. found_reverse_match holds bits to change
2138 (different for int & float insns). */
2140 i.tm.base_opcode ^= found_reverse_match;
2142 i.tm.operand_types[0] = t->operand_types[1];
2143 i.tm.operand_types[1] = t->operand_types[0];
2146 return 1;
2149 static int
2150 check_string ()
2152 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
2153 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
2155 if (i.seg[0] != NULL && i.seg[0] != &es)
2157 as_bad (_("`%s' operand %d must use `%%es' segment"),
2158 i.tm.name,
2159 mem_op + 1);
2160 return 0;
2162 /* There's only ever one segment override allowed per instruction.
2163 This instruction possibly has a legal segment override on the
2164 second operand, so copy the segment to where non-string
2165 instructions store it, allowing common code. */
2166 i.seg[0] = i.seg[1];
2168 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
2170 if (i.seg[1] != NULL && i.seg[1] != &es)
2172 as_bad (_("`%s' operand %d must use `%%es' segment"),
2173 i.tm.name,
2174 mem_op + 2);
2175 return 0;
2178 return 1;
2181 static int
2182 process_suffix ()
2184 /* If matched instruction specifies an explicit instruction mnemonic
2185 suffix, use it. */
2186 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
2188 if (i.tm.opcode_modifier & Size16)
2189 i.suffix = WORD_MNEM_SUFFIX;
2190 else if (i.tm.opcode_modifier & Size64)
2191 i.suffix = QWORD_MNEM_SUFFIX;
2192 else
2193 i.suffix = LONG_MNEM_SUFFIX;
2195 else if (i.reg_operands)
2197 /* If there's no instruction mnemonic suffix we try to invent one
2198 based on register operands. */
2199 if (!i.suffix)
2201 /* We take i.suffix from the last register operand specified,
2202 Destination register type is more significant than source
2203 register type. */
2204 int op;
2205 for (op = i.operands; --op >= 0;)
2206 if ((i.types[op] & Reg)
2207 && !(i.tm.operand_types[op] & InOutPortReg))
2209 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
2210 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
2211 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
2212 LONG_MNEM_SUFFIX);
2213 break;
2216 else if (i.suffix == BYTE_MNEM_SUFFIX)
2218 if (!check_byte_reg ())
2219 return 0;
2221 else if (i.suffix == LONG_MNEM_SUFFIX)
2223 if (!check_long_reg ())
2224 return 0;
2226 else if (i.suffix == QWORD_MNEM_SUFFIX)
2228 if (!check_qword_reg ())
2229 return 0;
2231 else if (i.suffix == WORD_MNEM_SUFFIX)
2233 if (!check_word_reg ())
2234 return 0;
2236 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2237 /* Do nothing if the instruction is going to ignore the prefix. */
2239 else
2240 abort ();
2242 else if ((i.tm.opcode_modifier & DefaultSize) && !i.suffix)
2244 i.suffix = stackop_size;
2247 /* Change the opcode based on the operand size given by i.suffix;
2248 We need not change things for byte insns. */
2250 if (!i.suffix && (i.tm.opcode_modifier & W))
2252 as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
2253 return 0;
2256 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2258 /* It's not a byte, select word/dword operation. */
2259 if (i.tm.opcode_modifier & W)
2261 if (i.tm.opcode_modifier & ShortForm)
2262 i.tm.base_opcode |= 8;
2263 else
2264 i.tm.base_opcode |= 1;
2267 /* Now select between word & dword operations via the operand
2268 size prefix, except for instructions that will ignore this
2269 prefix anyway. */
2270 if (i.suffix != QWORD_MNEM_SUFFIX
2271 && !(i.tm.opcode_modifier & IgnoreSize)
2272 && ((i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2273 || (flag_code == CODE_64BIT
2274 && (i.tm.opcode_modifier & JumpByte))))
2276 unsigned int prefix = DATA_PREFIX_OPCODE;
2277 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2278 prefix = ADDR_PREFIX_OPCODE;
2280 if (!add_prefix (prefix))
2281 return 0;
2284 /* Set mode64 for an operand. */
2285 if (i.suffix == QWORD_MNEM_SUFFIX
2286 && flag_code == CODE_64BIT
2287 && (i.tm.opcode_modifier & NoRex64) == 0)
2288 i.rex |= REX_MODE64;
2290 /* Size floating point instruction. */
2291 if (i.suffix == LONG_MNEM_SUFFIX)
2293 if (i.tm.opcode_modifier & FloatMF)
2294 i.tm.base_opcode ^= 4;
2298 return 1;
2301 static int
2302 check_byte_reg ()
2304 int op;
2305 for (op = i.operands; --op >= 0;)
2307 /* If this is an eight bit register, it's OK. If it's the 16 or
2308 32 bit version of an eight bit register, we will just use the
2309 low portion, and that's OK too. */
2310 if (i.types[op] & Reg8)
2311 continue;
2313 /* movzx and movsx should not generate this warning. */
2314 if (intel_syntax
2315 && (i.tm.base_opcode == 0xfb7
2316 || i.tm.base_opcode == 0xfb6
2317 || i.tm.base_opcode == 0x63
2318 || i.tm.base_opcode == 0xfbe
2319 || i.tm.base_opcode == 0xfbf))
2320 continue;
2322 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4
2323 #if 0
2324 /* Check that the template allows eight bit regs. This
2325 kills insns such as `orb $1,%edx', which maybe should be
2326 allowed. */
2327 && (i.tm.operand_types[op] & (Reg8 | InOutPortReg))
2328 #endif
2331 /* Prohibit these changes in the 64bit mode, since the
2332 lowering is more complicated. */
2333 if (flag_code == CODE_64BIT
2334 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2336 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2337 i.op[op].regs->reg_name,
2338 i.suffix);
2339 return 0;
2341 #if REGISTER_WARNINGS
2342 if (!quiet_warnings
2343 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2344 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2345 (i.op[op].regs + (i.types[op] & Reg16
2346 ? REGNAM_AL - REGNAM_AX
2347 : REGNAM_AL - REGNAM_EAX))->reg_name,
2348 i.op[op].regs->reg_name,
2349 i.suffix);
2350 #endif
2351 continue;
2353 /* Any other register is bad. */
2354 if (i.types[op] & (Reg | RegMMX | RegXMM
2355 | SReg2 | SReg3
2356 | Control | Debug | Test
2357 | FloatReg | FloatAcc))
2359 as_bad (_("`%%%s' not allowed with `%s%c'"),
2360 i.op[op].regs->reg_name,
2361 i.tm.name,
2362 i.suffix);
2363 return 0;
2366 return 1;
2369 static int
2370 check_long_reg ()
2372 int op;
2374 for (op = i.operands; --op >= 0;)
2375 /* Reject eight bit registers, except where the template requires
2376 them. (eg. movzb) */
2377 if ((i.types[op] & Reg8) != 0
2378 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2380 as_bad (_("`%%%s' not allowed with `%s%c'"),
2381 i.op[op].regs->reg_name,
2382 i.tm.name,
2383 i.suffix);
2384 return 0;
2386 /* Warn if the e prefix on a general reg is missing. */
2387 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2388 && (i.types[op] & Reg16) != 0
2389 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2391 /* Prohibit these changes in the 64bit mode, since the
2392 lowering is more complicated. */
2393 if (flag_code == CODE_64BIT)
2395 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2396 i.op[op].regs->reg_name,
2397 i.suffix);
2398 return 0;
2400 #if REGISTER_WARNINGS
2401 else
2402 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2403 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
2404 i.op[op].regs->reg_name,
2405 i.suffix);
2406 #endif
2408 /* Warn if the r prefix on a general reg is missing. */
2409 else if ((i.types[op] & Reg64) != 0
2410 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2412 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2413 i.op[op].regs->reg_name,
2414 i.suffix);
2415 return 0;
2417 return 1;
2420 static int
2421 check_qword_reg ()
2423 int op;
2425 for (op = i.operands; --op >= 0; )
2426 /* Reject eight bit registers, except where the template requires
2427 them. (eg. movzb) */
2428 if ((i.types[op] & Reg8) != 0
2429 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2431 as_bad (_("`%%%s' not allowed with `%s%c'"),
2432 i.op[op].regs->reg_name,
2433 i.tm.name,
2434 i.suffix);
2435 return 0;
2437 /* Warn if the e prefix on a general reg is missing. */
2438 else if (((i.types[op] & Reg16) != 0
2439 || (i.types[op] & Reg32) != 0)
2440 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2442 /* Prohibit these changes in the 64bit mode, since the
2443 lowering is more complicated. */
2444 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2445 i.op[op].regs->reg_name,
2446 i.suffix);
2447 return 0;
2449 return 1;
2452 static int
2453 check_word_reg ()
2455 int op;
2456 for (op = i.operands; --op >= 0;)
2457 /* Reject eight bit registers, except where the template requires
2458 them. (eg. movzb) */
2459 if ((i.types[op] & Reg8) != 0
2460 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2462 as_bad (_("`%%%s' not allowed with `%s%c'"),
2463 i.op[op].regs->reg_name,
2464 i.tm.name,
2465 i.suffix);
2466 return 0;
2468 /* Warn if the e prefix on a general reg is present. */
2469 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2470 && (i.types[op] & Reg32) != 0
2471 && (i.tm.operand_types[op] & (Reg16 | Acc)) != 0)
2473 /* Prohibit these changes in the 64bit mode, since the
2474 lowering is more complicated. */
2475 if (flag_code == CODE_64BIT)
2477 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2478 i.op[op].regs->reg_name,
2479 i.suffix);
2480 return 0;
2482 else
2483 #if REGISTER_WARNINGS
2484 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2485 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
2486 i.op[op].regs->reg_name,
2487 i.suffix);
2488 #endif
2490 return 1;
2493 static int
2494 finalize_imm ()
2496 unsigned int overlap0, overlap1, overlap2;
2498 overlap0 = i.types[0] & i.tm.operand_types[0];
2499 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S))
2500 && overlap0 != Imm8 && overlap0 != Imm8S
2501 && overlap0 != Imm16 && overlap0 != Imm32S
2502 && overlap0 != Imm32 && overlap0 != Imm64)
2504 if (i.suffix)
2506 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX
2507 ? Imm8 | Imm8S
2508 : (i.suffix == WORD_MNEM_SUFFIX
2509 ? Imm16
2510 : (i.suffix == QWORD_MNEM_SUFFIX
2511 ? Imm64 | Imm32S
2512 : Imm32)));
2514 else if (overlap0 == (Imm16 | Imm32S | Imm32)
2515 || overlap0 == (Imm16 | Imm32)
2516 || overlap0 == (Imm16 | Imm32S))
2518 overlap0 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
2519 ? Imm16 : Imm32S);
2521 if (overlap0 != Imm8 && overlap0 != Imm8S
2522 && overlap0 != Imm16 && overlap0 != Imm32S
2523 && overlap0 != Imm32 && overlap0 != Imm64)
2525 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
2526 return 0;
2529 i.types[0] = overlap0;
2531 overlap1 = i.types[1] & i.tm.operand_types[1];
2532 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32))
2533 && overlap1 != Imm8 && overlap1 != Imm8S
2534 && overlap1 != Imm16 && overlap1 != Imm32S
2535 && overlap1 != Imm32 && overlap1 != Imm64)
2537 if (i.suffix)
2539 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX
2540 ? Imm8 | Imm8S
2541 : (i.suffix == WORD_MNEM_SUFFIX
2542 ? Imm16
2543 : (i.suffix == QWORD_MNEM_SUFFIX
2544 ? Imm64 | Imm32S
2545 : Imm32)));
2547 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
2548 || overlap1 == (Imm16 | Imm32)
2549 || overlap1 == (Imm16 | Imm32S))
2551 overlap1 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
2552 ? Imm16 : Imm32S);
2554 if (overlap1 != Imm8 && overlap1 != Imm8S
2555 && overlap1 != Imm16 && overlap1 != Imm32S
2556 && overlap1 != Imm32 && overlap1 != Imm64)
2558 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
2559 return 0;
2562 i.types[1] = overlap1;
2564 overlap2 = i.types[2] & i.tm.operand_types[2];
2565 assert ((overlap2 & Imm) == 0);
2566 i.types[2] = overlap2;
2568 return 1;
2571 static int
2572 process_operands ()
2574 /* Default segment register this instruction will use for memory
2575 accesses. 0 means unknown. This is only for optimizing out
2576 unnecessary segment overrides. */
2577 const seg_entry *default_seg = 0;
2579 /* The imul $imm, %reg instruction is converted into
2580 imul $imm, %reg, %reg, and the clr %reg instruction
2581 is converted into xor %reg, %reg. */
2582 if (i.tm.opcode_modifier & regKludge)
2584 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
2585 /* Pretend we saw the extra register operand. */
2586 assert (i.op[first_reg_op + 1].regs == 0);
2587 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
2588 i.types[first_reg_op + 1] = i.types[first_reg_op];
2589 i.reg_operands = 2;
2592 if (i.tm.opcode_modifier & ShortForm)
2594 /* The register or float register operand is in operand 0 or 1. */
2595 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
2596 /* Register goes in low 3 bits of opcode. */
2597 i.tm.base_opcode |= i.op[op].regs->reg_num;
2598 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2599 i.rex |= REX_EXTZ;
2600 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2602 /* Warn about some common errors, but press on regardless.
2603 The first case can be generated by gcc (<= 2.8.1). */
2604 if (i.operands == 2)
2606 /* Reversed arguments on faddp, fsubp, etc. */
2607 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
2608 i.op[1].regs->reg_name,
2609 i.op[0].regs->reg_name);
2611 else
2613 /* Extraneous `l' suffix on fp insn. */
2614 as_warn (_("translating to `%s %%%s'"), i.tm.name,
2615 i.op[0].regs->reg_name);
2619 else if (i.tm.opcode_modifier & Modrm)
2621 /* The opcode is completed (modulo i.tm.extension_opcode which
2622 must be put into the modrm byte). Now, we make the modrm and
2623 index base bytes based on all the info we've collected. */
2625 default_seg = build_modrm_byte ();
2627 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
2629 if (i.tm.base_opcode == POP_SEG_SHORT
2630 && i.op[0].regs->reg_num == 1)
2632 as_bad (_("you can't `pop %%cs'"));
2633 return 0;
2635 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
2636 if ((i.op[0].regs->reg_flags & RegRex) != 0)
2637 i.rex |= REX_EXTZ;
2639 else if ((i.tm.base_opcode & ~(D | W)) == MOV_AX_DISP32)
2641 default_seg = &ds;
2643 else if ((i.tm.opcode_modifier & IsString) != 0)
2645 /* For the string instructions that allow a segment override
2646 on one of their operands, the default segment is ds. */
2647 default_seg = &ds;
2650 if (i.tm.base_opcode == 0x8d /* lea */ && i.seg[0] && !quiet_warnings)
2651 as_warn (_("segment override on `lea' is ineffectual"));
2653 /* If a segment was explicitly specified, and the specified segment
2654 is not the default, use an opcode prefix to select it. If we
2655 never figured out what the default segment is, then default_seg
2656 will be zero at this point, and the specified segment prefix will
2657 always be used. */
2658 if ((i.seg[0]) && (i.seg[0] != default_seg))
2660 if (!add_prefix (i.seg[0]->seg_prefix))
2661 return 0;
2663 return 1;
2666 static const seg_entry *
2667 build_modrm_byte ()
2669 const seg_entry *default_seg = 0;
2671 /* i.reg_operands MUST be the number of real register operands;
2672 implicit registers do not count. */
2673 if (i.reg_operands == 2)
2675 unsigned int source, dest;
2676 source = ((i.types[0]
2677 & (Reg | RegMMX | RegXMM
2678 | SReg2 | SReg3
2679 | Control | Debug | Test))
2680 ? 0 : 1);
2681 dest = source + 1;
2683 i.rm.mode = 3;
2684 /* One of the register operands will be encoded in the i.tm.reg
2685 field, the other in the combined i.tm.mode and i.tm.regmem
2686 fields. If no form of this instruction supports a memory
2687 destination operand, then we assume the source operand may
2688 sometimes be a memory operand and so we need to store the
2689 destination in the i.rm.reg field. */
2690 if ((i.tm.operand_types[dest] & AnyMem) == 0)
2692 i.rm.reg = i.op[dest].regs->reg_num;
2693 i.rm.regmem = i.op[source].regs->reg_num;
2694 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
2695 i.rex |= REX_EXTX;
2696 if ((i.op[source].regs->reg_flags & RegRex) != 0)
2697 i.rex |= REX_EXTZ;
2699 else
2701 i.rm.reg = i.op[source].regs->reg_num;
2702 i.rm.regmem = i.op[dest].regs->reg_num;
2703 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
2704 i.rex |= REX_EXTZ;
2705 if ((i.op[source].regs->reg_flags & RegRex) != 0)
2706 i.rex |= REX_EXTX;
2709 else
2710 { /* If it's not 2 reg operands... */
2711 if (i.mem_operands)
2713 unsigned int fake_zero_displacement = 0;
2714 unsigned int op = ((i.types[0] & AnyMem)
2716 : (i.types[1] & AnyMem) ? 1 : 2);
2718 default_seg = &ds;
2720 if (i.base_reg == 0)
2722 i.rm.mode = 0;
2723 if (!i.disp_operands)
2724 fake_zero_displacement = 1;
2725 if (i.index_reg == 0)
2727 /* Operand is just <disp> */
2728 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)
2729 && (flag_code != CODE_64BIT))
2731 i.rm.regmem = NO_BASE_REGISTER_16;
2732 i.types[op] &= ~Disp;
2733 i.types[op] |= Disp16;
2735 else if (flag_code != CODE_64BIT
2736 || (i.prefix[ADDR_PREFIX] != 0))
2738 i.rm.regmem = NO_BASE_REGISTER;
2739 i.types[op] &= ~Disp;
2740 i.types[op] |= Disp32;
2742 else
2744 /* 64bit mode overwrites the 32bit absolute
2745 addressing by RIP relative addressing and
2746 absolute addressing is encoded by one of the
2747 redundant SIB forms. */
2748 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2749 i.sib.base = NO_BASE_REGISTER;
2750 i.sib.index = NO_INDEX_REGISTER;
2751 i.types[op] &= ~Disp;
2752 i.types[op] |= Disp32S;
2755 else /* !i.base_reg && i.index_reg */
2757 i.sib.index = i.index_reg->reg_num;
2758 i.sib.base = NO_BASE_REGISTER;
2759 i.sib.scale = i.log2_scale_factor;
2760 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2761 i.types[op] &= ~Disp;
2762 if (flag_code != CODE_64BIT)
2763 i.types[op] |= Disp32; /* Must be 32 bit */
2764 else
2765 i.types[op] |= Disp32S;
2766 if ((i.index_reg->reg_flags & RegRex) != 0)
2767 i.rex |= REX_EXTY;
2770 /* RIP addressing for 64bit mode. */
2771 else if (i.base_reg->reg_type == BaseIndex)
2773 i.rm.regmem = NO_BASE_REGISTER;
2774 i.types[op] &= ~Disp;
2775 i.types[op] |= Disp32S;
2776 i.flags[op] = Operand_PCrel;
2778 else if (i.base_reg->reg_type & Reg16)
2780 switch (i.base_reg->reg_num)
2782 case 3: /* (%bx) */
2783 if (i.index_reg == 0)
2784 i.rm.regmem = 7;
2785 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
2786 i.rm.regmem = i.index_reg->reg_num - 6;
2787 break;
2788 case 5: /* (%bp) */
2789 default_seg = &ss;
2790 if (i.index_reg == 0)
2792 i.rm.regmem = 6;
2793 if ((i.types[op] & Disp) == 0)
2795 /* fake (%bp) into 0(%bp) */
2796 i.types[op] |= Disp8;
2797 fake_zero_displacement = 1;
2800 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
2801 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
2802 break;
2803 default: /* (%si) -> 4 or (%di) -> 5 */
2804 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
2806 i.rm.mode = mode_from_disp_size (i.types[op]);
2808 else /* i.base_reg and 32/64 bit mode */
2810 if (flag_code == CODE_64BIT
2811 && (i.types[op] & Disp))
2813 if (i.types[op] & Disp8)
2814 i.types[op] = Disp8 | Disp32S;
2815 else
2816 i.types[op] = Disp32S;
2818 i.rm.regmem = i.base_reg->reg_num;
2819 if ((i.base_reg->reg_flags & RegRex) != 0)
2820 i.rex |= REX_EXTZ;
2821 i.sib.base = i.base_reg->reg_num;
2822 /* x86-64 ignores REX prefix bit here to avoid decoder
2823 complications. */
2824 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
2826 default_seg = &ss;
2827 if (i.disp_operands == 0)
2829 fake_zero_displacement = 1;
2830 i.types[op] |= Disp8;
2833 else if (i.base_reg->reg_num == ESP_REG_NUM)
2835 default_seg = &ss;
2837 i.sib.scale = i.log2_scale_factor;
2838 if (i.index_reg == 0)
2840 /* <disp>(%esp) becomes two byte modrm with no index
2841 register. We've already stored the code for esp
2842 in i.rm.regmem ie. ESCAPE_TO_TWO_BYTE_ADDRESSING.
2843 Any base register besides %esp will not use the
2844 extra modrm byte. */
2845 i.sib.index = NO_INDEX_REGISTER;
2846 #if !SCALE1_WHEN_NO_INDEX
2847 /* Another case where we force the second modrm byte. */
2848 if (i.log2_scale_factor)
2849 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2850 #endif
2852 else
2854 i.sib.index = i.index_reg->reg_num;
2855 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2856 if ((i.index_reg->reg_flags & RegRex) != 0)
2857 i.rex |= REX_EXTY;
2859 i.rm.mode = mode_from_disp_size (i.types[op]);
2862 if (fake_zero_displacement)
2864 /* Fakes a zero displacement assuming that i.types[op]
2865 holds the correct displacement size. */
2866 expressionS *exp;
2868 assert (i.op[op].disps == 0);
2869 exp = &disp_expressions[i.disp_operands++];
2870 i.op[op].disps = exp;
2871 exp->X_op = O_constant;
2872 exp->X_add_number = 0;
2873 exp->X_add_symbol = (symbolS *) 0;
2874 exp->X_op_symbol = (symbolS *) 0;
2878 /* Fill in i.rm.reg or i.rm.regmem field with register operand
2879 (if any) based on i.tm.extension_opcode. Again, we must be
2880 careful to make sure that segment/control/debug/test/MMX
2881 registers are coded into the i.rm.reg field. */
2882 if (i.reg_operands)
2884 unsigned int op =
2885 ((i.types[0]
2886 & (Reg | RegMMX | RegXMM
2887 | SReg2 | SReg3
2888 | Control | Debug | Test))
2890 : ((i.types[1]
2891 & (Reg | RegMMX | RegXMM
2892 | SReg2 | SReg3
2893 | Control | Debug | Test))
2895 : 2));
2896 /* If there is an extension opcode to put here, the register
2897 number must be put into the regmem field. */
2898 if (i.tm.extension_opcode != None)
2900 i.rm.regmem = i.op[op].regs->reg_num;
2901 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2902 i.rex |= REX_EXTZ;
2904 else
2906 i.rm.reg = i.op[op].regs->reg_num;
2907 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2908 i.rex |= REX_EXTX;
2911 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2 we
2912 must set it to 3 to indicate this is a register operand
2913 in the regmem field. */
2914 if (!i.mem_operands)
2915 i.rm.mode = 3;
2918 /* Fill in i.rm.reg field with extension opcode (if any). */
2919 if (i.tm.extension_opcode != None)
2920 i.rm.reg = i.tm.extension_opcode;
2922 return default_seg;
2925 static void
2926 output_branch ()
2928 char *p;
2929 int code16;
2930 int prefix;
2931 relax_substateT subtype;
2932 symbolS *sym;
2933 offsetT off;
2935 code16 = 0;
2936 if (flag_code == CODE_16BIT)
2937 code16 = CODE16;
2939 prefix = 0;
2940 if (i.prefix[DATA_PREFIX] != 0)
2942 prefix = 1;
2943 i.prefixes -= 1;
2944 code16 ^= CODE16;
2946 /* Pentium4 branch hints. */
2947 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
2948 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
2950 prefix++;
2951 i.prefixes--;
2953 if (i.prefix[REX_PREFIX] != 0)
2955 prefix++;
2956 i.prefixes--;
2959 if (i.prefixes != 0 && !intel_syntax)
2960 as_warn (_("skipping prefixes on this instruction"));
2962 /* It's always a symbol; End frag & setup for relax.
2963 Make sure there is enough room in this frag for the largest
2964 instruction we may generate in md_convert_frag. This is 2
2965 bytes for the opcode and room for the prefix and largest
2966 displacement. */
2967 frag_grow (prefix + 2 + 4);
2968 /* Prefix and 1 opcode byte go in fr_fix. */
2969 p = frag_more (prefix + 1);
2970 if (i.prefix[DATA_PREFIX] != 0)
2971 *p++ = DATA_PREFIX_OPCODE;
2972 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
2973 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
2974 *p++ = i.prefix[SEG_PREFIX];
2975 if (i.prefix[REX_PREFIX] != 0)
2976 *p++ = i.prefix[REX_PREFIX];
2977 *p = i.tm.base_opcode;
2979 if ((unsigned char) *p == JUMP_PC_RELATIVE)
2980 subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
2981 else if ((cpu_arch_flags & Cpu386) != 0)
2982 subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
2983 else
2984 subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
2985 subtype |= code16;
2987 sym = i.op[0].disps->X_add_symbol;
2988 off = i.op[0].disps->X_add_number;
2990 if (i.op[0].disps->X_op != O_constant
2991 && i.op[0].disps->X_op != O_symbol)
2993 /* Handle complex expressions. */
2994 sym = make_expr_symbol (i.op[0].disps);
2995 off = 0;
2998 /* 1 possible extra opcode + 4 byte displacement go in var part.
2999 Pass reloc in fr_var. */
3000 frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
3003 static void
3004 output_jump ()
3006 char *p;
3007 int size;
3008 fixS *fixP;
3010 if (i.tm.opcode_modifier & JumpByte)
3012 /* This is a loop or jecxz type instruction. */
3013 size = 1;
3014 if (i.prefix[ADDR_PREFIX] != 0)
3016 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
3017 i.prefixes -= 1;
3019 /* Pentium4 branch hints. */
3020 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3021 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3023 FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
3024 i.prefixes--;
3027 else
3029 int code16;
3031 code16 = 0;
3032 if (flag_code == CODE_16BIT)
3033 code16 = CODE16;
3035 if (i.prefix[DATA_PREFIX] != 0)
3037 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
3038 i.prefixes -= 1;
3039 code16 ^= CODE16;
3042 size = 4;
3043 if (code16)
3044 size = 2;
3047 if (i.prefix[REX_PREFIX] != 0)
3049 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
3050 i.prefixes -= 1;
3053 if (i.prefixes != 0 && !intel_syntax)
3054 as_warn (_("skipping prefixes on this instruction"));
3056 p = frag_more (1 + size);
3057 *p++ = i.tm.base_opcode;
3059 fixP = fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3060 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
3062 /* All jumps handled here are signed, but don't use a signed limit
3063 check for 32 and 16 bit jumps as we want to allow wrap around at
3064 4G and 64k respectively. */
3065 if (size == 1)
3066 fixP->fx_signed = 1;
3069 static void
3070 output_interseg_jump ()
3072 char *p;
3073 int size;
3074 int prefix;
3075 int code16;
3077 code16 = 0;
3078 if (flag_code == CODE_16BIT)
3079 code16 = CODE16;
3081 prefix = 0;
3082 if (i.prefix[DATA_PREFIX] != 0)
3084 prefix = 1;
3085 i.prefixes -= 1;
3086 code16 ^= CODE16;
3088 if (i.prefix[REX_PREFIX] != 0)
3090 prefix++;
3091 i.prefixes -= 1;
3094 size = 4;
3095 if (code16)
3096 size = 2;
3098 if (i.prefixes != 0 && !intel_syntax)
3099 as_warn (_("skipping prefixes on this instruction"));
3101 /* 1 opcode; 2 segment; offset */
3102 p = frag_more (prefix + 1 + 2 + size);
3104 if (i.prefix[DATA_PREFIX] != 0)
3105 *p++ = DATA_PREFIX_OPCODE;
3107 if (i.prefix[REX_PREFIX] != 0)
3108 *p++ = i.prefix[REX_PREFIX];
3110 *p++ = i.tm.base_opcode;
3111 if (i.op[1].imms->X_op == O_constant)
3113 offsetT n = i.op[1].imms->X_add_number;
3115 if (size == 2
3116 && !fits_in_unsigned_word (n)
3117 && !fits_in_signed_word (n))
3119 as_bad (_("16-bit jump out of range"));
3120 return;
3122 md_number_to_chars (p, n, size);
3124 else
3125 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3126 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
3127 if (i.op[0].imms->X_op != O_constant)
3128 as_bad (_("can't handle non absolute segment in `%s'"),
3129 i.tm.name);
3130 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
3133 static void
3134 output_insn ()
3136 fragS *insn_start_frag;
3137 offsetT insn_start_off;
3139 /* Tie dwarf2 debug info to the address at the start of the insn.
3140 We can't do this after the insn has been output as the current
3141 frag may have been closed off. eg. by frag_var. */
3142 dwarf2_emit_insn (0);
3144 insn_start_frag = frag_now;
3145 insn_start_off = frag_now_fix ();
3147 /* Output jumps. */
3148 if (i.tm.opcode_modifier & Jump)
3149 output_branch ();
3150 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
3151 output_jump ();
3152 else if (i.tm.opcode_modifier & JumpInterSegment)
3153 output_interseg_jump ();
3154 else
3156 /* Output normal instructions here. */
3157 char *p;
3158 unsigned char *q;
3160 /* All opcodes on i386 have either 1 or 2 bytes, PadLock instructions
3161 have 3 bytes. We may use one more higher byte to specify a prefix
3162 the instruction requires. */
3163 if ((i.tm.cpu_flags & CpuPadLock) != 0
3164 && (i.tm.base_opcode & 0xff000000) != 0)
3166 unsigned int prefix;
3167 prefix = (i.tm.base_opcode >> 24) & 0xff;
3169 if (prefix != REPE_PREFIX_OPCODE
3170 || i.prefix[LOCKREP_PREFIX] != REPE_PREFIX_OPCODE)
3171 add_prefix (prefix);
3173 else
3174 if ((i.tm.cpu_flags & CpuPadLock) == 0
3175 && (i.tm.base_opcode & 0xff0000) != 0)
3176 add_prefix ((i.tm.base_opcode >> 16) & 0xff);
3178 /* The prefix bytes. */
3179 for (q = i.prefix;
3180 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
3181 q++)
3183 if (*q)
3185 p = frag_more (1);
3186 md_number_to_chars (p, (valueT) *q, 1);
3190 /* Now the opcode; be careful about word order here! */
3191 if (fits_in_unsigned_byte (i.tm.base_opcode))
3193 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
3195 else
3197 if ((i.tm.cpu_flags & CpuPadLock) != 0)
3199 p = frag_more (3);
3200 *p++ = (i.tm.base_opcode >> 16) & 0xff;
3202 else
3203 p = frag_more (2);
3205 /* Put out high byte first: can't use md_number_to_chars! */
3206 *p++ = (i.tm.base_opcode >> 8) & 0xff;
3207 *p = i.tm.base_opcode & 0xff;
3210 /* Now the modrm byte and sib byte (if present). */
3211 if (i.tm.opcode_modifier & Modrm)
3213 p = frag_more (1);
3214 md_number_to_chars (p,
3215 (valueT) (i.rm.regmem << 0
3216 | i.rm.reg << 3
3217 | i.rm.mode << 6),
3219 /* If i.rm.regmem == ESP (4)
3220 && i.rm.mode != (Register mode)
3221 && not 16 bit
3222 ==> need second modrm byte. */
3223 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
3224 && i.rm.mode != 3
3225 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
3227 p = frag_more (1);
3228 md_number_to_chars (p,
3229 (valueT) (i.sib.base << 0
3230 | i.sib.index << 3
3231 | i.sib.scale << 6),
3236 if (i.disp_operands)
3237 output_disp (insn_start_frag, insn_start_off);
3239 if (i.imm_operands)
3240 output_imm (insn_start_frag, insn_start_off);
3243 #ifdef DEBUG386
3244 if (flag_debug)
3246 pi (line, &i);
3248 #endif /* DEBUG386 */
3251 static void
3252 output_disp (insn_start_frag, insn_start_off)
3253 fragS *insn_start_frag;
3254 offsetT insn_start_off;
3256 char *p;
3257 unsigned int n;
3259 for (n = 0; n < i.operands; n++)
3261 if (i.types[n] & Disp)
3263 if (i.op[n].disps->X_op == O_constant)
3265 int size;
3266 offsetT val;
3268 size = 4;
3269 if (i.types[n] & (Disp8 | Disp16 | Disp64))
3271 size = 2;
3272 if (i.types[n] & Disp8)
3273 size = 1;
3274 if (i.types[n] & Disp64)
3275 size = 8;
3277 val = offset_in_range (i.op[n].disps->X_add_number,
3278 size);
3279 p = frag_more (size);
3280 md_number_to_chars (p, val, size);
3282 else
3284 enum bfd_reloc_code_real reloc_type;
3285 int size = 4;
3286 int sign = 0;
3287 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
3289 /* The PC relative address is computed relative
3290 to the instruction boundary, so in case immediate
3291 fields follows, we need to adjust the value. */
3292 if (pcrel && i.imm_operands)
3294 int imm_size = 4;
3295 unsigned int n1;
3297 for (n1 = 0; n1 < i.operands; n1++)
3298 if (i.types[n1] & Imm)
3300 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
3302 imm_size = 2;
3303 if (i.types[n1] & (Imm8 | Imm8S))
3304 imm_size = 1;
3305 if (i.types[n1] & Imm64)
3306 imm_size = 8;
3308 break;
3310 /* We should find the immediate. */
3311 if (n1 == i.operands)
3312 abort ();
3313 i.op[n].disps->X_add_number -= imm_size;
3316 if (i.types[n] & Disp32S)
3317 sign = 1;
3319 if (i.types[n] & (Disp16 | Disp64))
3321 size = 2;
3322 if (i.types[n] & Disp64)
3323 size = 8;
3326 p = frag_more (size);
3327 reloc_type = reloc (size, pcrel, sign, i.reloc[n]);
3328 if (reloc_type == BFD_RELOC_32
3329 && GOT_symbol
3330 && GOT_symbol == i.op[n].disps->X_add_symbol
3331 && (i.op[n].disps->X_op == O_symbol
3332 || (i.op[n].disps->X_op == O_add
3333 && ((symbol_get_value_expression
3334 (i.op[n].disps->X_op_symbol)->X_op)
3335 == O_subtract))))
3337 offsetT add;
3339 if (insn_start_frag == frag_now)
3340 add = (p - frag_now->fr_literal) - insn_start_off;
3341 else
3343 fragS *fr;
3345 add = insn_start_frag->fr_fix - insn_start_off;
3346 for (fr = insn_start_frag->fr_next;
3347 fr && fr != frag_now; fr = fr->fr_next)
3348 add += fr->fr_fix;
3349 add += p - frag_now->fr_literal;
3352 /* We don't support dynamic linking on x86-64 yet. */
3353 if (flag_code == CODE_64BIT)
3354 abort ();
3355 reloc_type = BFD_RELOC_386_GOTPC;
3356 i.op[n].disps->X_add_number += add;
3358 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3359 i.op[n].disps, pcrel, reloc_type);
3365 static void
3366 output_imm (insn_start_frag, insn_start_off)
3367 fragS *insn_start_frag;
3368 offsetT insn_start_off;
3370 char *p;
3371 unsigned int n;
3373 for (n = 0; n < i.operands; n++)
3375 if (i.types[n] & Imm)
3377 if (i.op[n].imms->X_op == O_constant)
3379 int size;
3380 offsetT val;
3382 size = 4;
3383 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3385 size = 2;
3386 if (i.types[n] & (Imm8 | Imm8S))
3387 size = 1;
3388 else if (i.types[n] & Imm64)
3389 size = 8;
3391 val = offset_in_range (i.op[n].imms->X_add_number,
3392 size);
3393 p = frag_more (size);
3394 md_number_to_chars (p, val, size);
3396 else
3398 /* Not absolute_section.
3399 Need a 32-bit fixup (don't support 8bit
3400 non-absolute imms). Try to support other
3401 sizes ... */
3402 enum bfd_reloc_code_real reloc_type;
3403 int size = 4;
3404 int sign = 0;
3406 if ((i.types[n] & (Imm32S))
3407 && i.suffix == QWORD_MNEM_SUFFIX)
3408 sign = 1;
3409 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3411 size = 2;
3412 if (i.types[n] & (Imm8 | Imm8S))
3413 size = 1;
3414 if (i.types[n] & Imm64)
3415 size = 8;
3418 p = frag_more (size);
3419 reloc_type = reloc (size, 0, sign, i.reloc[n]);
3421 /* This is tough to explain. We end up with this one if we
3422 * have operands that look like
3423 * "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal here is to
3424 * obtain the absolute address of the GOT, and it is strongly
3425 * preferable from a performance point of view to avoid using
3426 * a runtime relocation for this. The actual sequence of
3427 * instructions often look something like:
3429 * call .L66
3430 * .L66:
3431 * popl %ebx
3432 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
3434 * The call and pop essentially return the absolute address
3435 * of the label .L66 and store it in %ebx. The linker itself
3436 * will ultimately change the first operand of the addl so
3437 * that %ebx points to the GOT, but to keep things simple, the
3438 * .o file must have this operand set so that it generates not
3439 * the absolute address of .L66, but the absolute address of
3440 * itself. This allows the linker itself simply treat a GOTPC
3441 * relocation as asking for a pcrel offset to the GOT to be
3442 * added in, and the addend of the relocation is stored in the
3443 * operand field for the instruction itself.
3445 * Our job here is to fix the operand so that it would add
3446 * the correct offset so that %ebx would point to itself. The
3447 * thing that is tricky is that .-.L66 will point to the
3448 * beginning of the instruction, so we need to further modify
3449 * the operand so that it will point to itself. There are
3450 * other cases where you have something like:
3452 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
3454 * and here no correction would be required. Internally in
3455 * the assembler we treat operands of this form as not being
3456 * pcrel since the '.' is explicitly mentioned, and I wonder
3457 * whether it would simplify matters to do it this way. Who
3458 * knows. In earlier versions of the PIC patches, the
3459 * pcrel_adjust field was used to store the correction, but
3460 * since the expression is not pcrel, I felt it would be
3461 * confusing to do it this way. */
3463 if (reloc_type == BFD_RELOC_32
3464 && GOT_symbol
3465 && GOT_symbol == i.op[n].imms->X_add_symbol
3466 && (i.op[n].imms->X_op == O_symbol
3467 || (i.op[n].imms->X_op == O_add
3468 && ((symbol_get_value_expression
3469 (i.op[n].imms->X_op_symbol)->X_op)
3470 == O_subtract))))
3472 offsetT add;
3474 if (insn_start_frag == frag_now)
3475 add = (p - frag_now->fr_literal) - insn_start_off;
3476 else
3478 fragS *fr;
3480 add = insn_start_frag->fr_fix - insn_start_off;
3481 for (fr = insn_start_frag->fr_next;
3482 fr && fr != frag_now; fr = fr->fr_next)
3483 add += fr->fr_fix;
3484 add += p - frag_now->fr_literal;
3487 /* We don't support dynamic linking on x86-64 yet. */
3488 if (flag_code == CODE_64BIT)
3489 abort ();
3490 reloc_type = BFD_RELOC_386_GOTPC;
3491 i.op[n].imms->X_add_number += add;
3493 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3494 i.op[n].imms, 0, reloc_type);
3500 #ifndef LEX_AT
3501 static char *lex_got PARAMS ((enum bfd_reloc_code_real *, int *));
3503 /* Parse operands of the form
3504 <symbol>@GOTOFF+<nnn>
3505 and similar .plt or .got references.
3507 If we find one, set up the correct relocation in RELOC and copy the
3508 input string, minus the `@GOTOFF' into a malloc'd buffer for
3509 parsing by the calling routine. Return this buffer, and if ADJUST
3510 is non-null set it to the length of the string we removed from the
3511 input line. Otherwise return NULL. */
3512 static char *
3513 lex_got (reloc, adjust)
3514 enum bfd_reloc_code_real *reloc;
3515 int *adjust;
3517 static const char * const mode_name[NUM_FLAG_CODE] = { "32", "16", "64" };
3518 static const struct {
3519 const char *str;
3520 const enum bfd_reloc_code_real rel[NUM_FLAG_CODE];
3521 } gotrel[] = {
3522 { "PLT", { BFD_RELOC_386_PLT32, 0, BFD_RELOC_X86_64_PLT32 } },
3523 { "GOTOFF", { BFD_RELOC_386_GOTOFF, 0, 0 } },
3524 { "GOTPCREL", { 0, 0, BFD_RELOC_X86_64_GOTPCREL } },
3525 { "TLSGD", { BFD_RELOC_386_TLS_GD, 0, BFD_RELOC_X86_64_TLSGD } },
3526 { "TLSLDM", { BFD_RELOC_386_TLS_LDM, 0, 0 } },
3527 { "TLSLD", { 0, 0, BFD_RELOC_X86_64_TLSLD } },
3528 { "GOTTPOFF", { BFD_RELOC_386_TLS_IE_32, 0, BFD_RELOC_X86_64_GOTTPOFF } },
3529 { "TPOFF", { BFD_RELOC_386_TLS_LE_32, 0, BFD_RELOC_X86_64_TPOFF32 } },
3530 { "NTPOFF", { BFD_RELOC_386_TLS_LE, 0, 0 } },
3531 { "DTPOFF", { BFD_RELOC_386_TLS_LDO_32, 0, BFD_RELOC_X86_64_DTPOFF32 } },
3532 { "GOTNTPOFF",{ BFD_RELOC_386_TLS_GOTIE, 0, 0 } },
3533 { "INDNTPOFF",{ BFD_RELOC_386_TLS_IE, 0, 0 } },
3534 { "GOT", { BFD_RELOC_386_GOT32, 0, BFD_RELOC_X86_64_GOT32 } }
3536 char *cp;
3537 unsigned int j;
3539 for (cp = input_line_pointer; *cp != '@'; cp++)
3540 if (is_end_of_line[(unsigned char) *cp])
3541 return NULL;
3543 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
3545 int len;
3547 len = strlen (gotrel[j].str);
3548 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
3550 if (gotrel[j].rel[(unsigned int) flag_code] != 0)
3552 int first, second;
3553 char *tmpbuf, *past_reloc;
3555 *reloc = gotrel[j].rel[(unsigned int) flag_code];
3556 if (adjust)
3557 *adjust = len;
3559 if (GOT_symbol == NULL)
3560 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3562 /* Replace the relocation token with ' ', so that
3563 errors like foo@GOTOFF1 will be detected. */
3565 /* The length of the first part of our input line. */
3566 first = cp - input_line_pointer;
3568 /* The second part goes from after the reloc token until
3569 (and including) an end_of_line char. Don't use strlen
3570 here as the end_of_line char may not be a NUL. */
3571 past_reloc = cp + 1 + len;
3572 for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
3574 second = cp - past_reloc;
3576 /* Allocate and copy string. The trailing NUL shouldn't
3577 be necessary, but be safe. */
3578 tmpbuf = xmalloc (first + second + 2);
3579 memcpy (tmpbuf, input_line_pointer, first);
3580 tmpbuf[first] = ' ';
3581 memcpy (tmpbuf + first + 1, past_reloc, second);
3582 tmpbuf[first + second + 1] = '\0';
3583 return tmpbuf;
3586 as_bad (_("@%s reloc is not supported in %s bit mode"),
3587 gotrel[j].str, mode_name[(unsigned int) flag_code]);
3588 return NULL;
3592 /* Might be a symbol version string. Don't as_bad here. */
3593 return NULL;
3596 /* x86_cons_fix_new is called via the expression parsing code when a
3597 reloc is needed. We use this hook to get the correct .got reloc. */
3598 static enum bfd_reloc_code_real got_reloc = NO_RELOC;
3600 void
3601 x86_cons_fix_new (frag, off, len, exp)
3602 fragS *frag;
3603 unsigned int off;
3604 unsigned int len;
3605 expressionS *exp;
3607 enum bfd_reloc_code_real r = reloc (len, 0, 0, got_reloc);
3608 got_reloc = NO_RELOC;
3609 fix_new_exp (frag, off, len, exp, 0, r);
3612 void
3613 x86_cons (exp, size)
3614 expressionS *exp;
3615 int size;
3617 if (size == 4)
3619 /* Handle @GOTOFF and the like in an expression. */
3620 char *save;
3621 char *gotfree_input_line;
3622 int adjust;
3624 save = input_line_pointer;
3625 gotfree_input_line = lex_got (&got_reloc, &adjust);
3626 if (gotfree_input_line)
3627 input_line_pointer = gotfree_input_line;
3629 expression (exp);
3631 if (gotfree_input_line)
3633 /* expression () has merrily parsed up to the end of line,
3634 or a comma - in the wrong buffer. Transfer how far
3635 input_line_pointer has moved to the right buffer. */
3636 input_line_pointer = (save
3637 + (input_line_pointer - gotfree_input_line)
3638 + adjust);
3639 free (gotfree_input_line);
3642 else
3643 expression (exp);
3645 #endif
3647 #ifdef TE_PE
3649 #define O_secrel (O_max + 1)
3651 void
3652 x86_pe_cons_fix_new (frag, off, len, exp)
3653 fragS *frag;
3654 unsigned int off;
3655 unsigned int len;
3656 expressionS *exp;
3658 enum bfd_reloc_code_real r = reloc (len, 0, 0, NO_RELOC);
3660 if (exp->X_op == O_secrel)
3662 exp->X_op = O_symbol;
3663 r = BFD_RELOC_32_SECREL;
3666 fix_new_exp (frag, off, len, exp, 0, r);
3669 static void
3670 pe_directive_secrel (dummy)
3671 int dummy ATTRIBUTE_UNUSED;
3673 expressionS exp;
3677 expression (&exp);
3678 if (exp.X_op == O_symbol)
3679 exp.X_op = O_secrel;
3681 emit_expr (&exp, 4);
3683 while (*input_line_pointer++ == ',');
3685 input_line_pointer--;
3686 demand_empty_rest_of_line ();
3689 #endif
3691 static int i386_immediate PARAMS ((char *));
3693 static int
3694 i386_immediate (imm_start)
3695 char *imm_start;
3697 char *save_input_line_pointer;
3698 #ifndef LEX_AT
3699 char *gotfree_input_line;
3700 #endif
3701 segT exp_seg = 0;
3702 expressionS *exp;
3704 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
3706 as_bad (_("only 1 or 2 immediate operands are allowed"));
3707 return 0;
3710 exp = &im_expressions[i.imm_operands++];
3711 i.op[this_operand].imms = exp;
3713 if (is_space_char (*imm_start))
3714 ++imm_start;
3716 save_input_line_pointer = input_line_pointer;
3717 input_line_pointer = imm_start;
3719 #ifndef LEX_AT
3720 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3721 if (gotfree_input_line)
3722 input_line_pointer = gotfree_input_line;
3723 #endif
3725 exp_seg = expression (exp);
3727 SKIP_WHITESPACE ();
3728 if (*input_line_pointer)
3729 as_bad (_("junk `%s' after expression"), input_line_pointer);
3731 input_line_pointer = save_input_line_pointer;
3732 #ifndef LEX_AT
3733 if (gotfree_input_line)
3734 free (gotfree_input_line);
3735 #endif
3737 if (exp->X_op == O_absent || exp->X_op == O_big)
3739 /* Missing or bad expr becomes absolute 0. */
3740 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
3741 imm_start);
3742 exp->X_op = O_constant;
3743 exp->X_add_number = 0;
3744 exp->X_add_symbol = (symbolS *) 0;
3745 exp->X_op_symbol = (symbolS *) 0;
3747 else if (exp->X_op == O_constant)
3749 /* Size it properly later. */
3750 i.types[this_operand] |= Imm64;
3751 /* If BFD64, sign extend val. */
3752 if (!use_rela_relocations)
3753 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
3754 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
3756 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3757 else if (OUTPUT_FLAVOR == bfd_target_aout_flavour
3758 && exp_seg != absolute_section
3759 && exp_seg != text_section
3760 && exp_seg != data_section
3761 && exp_seg != bss_section
3762 && exp_seg != undefined_section
3763 && !bfd_is_com_section (exp_seg))
3765 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3766 return 0;
3768 #endif
3769 else
3771 /* This is an address. The size of the address will be
3772 determined later, depending on destination register,
3773 suffix, or the default for the section. */
3774 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
3777 return 1;
3780 static char *i386_scale PARAMS ((char *));
3782 static char *
3783 i386_scale (scale)
3784 char *scale;
3786 offsetT val;
3787 char *save = input_line_pointer;
3789 input_line_pointer = scale;
3790 val = get_absolute_expression ();
3792 switch (val)
3794 case 0:
3795 case 1:
3796 i.log2_scale_factor = 0;
3797 break;
3798 case 2:
3799 i.log2_scale_factor = 1;
3800 break;
3801 case 4:
3802 i.log2_scale_factor = 2;
3803 break;
3804 case 8:
3805 i.log2_scale_factor = 3;
3806 break;
3807 default:
3808 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
3809 scale);
3810 input_line_pointer = save;
3811 return NULL;
3813 if (i.log2_scale_factor != 0 && i.index_reg == 0)
3815 as_warn (_("scale factor of %d without an index register"),
3816 1 << i.log2_scale_factor);
3817 #if SCALE1_WHEN_NO_INDEX
3818 i.log2_scale_factor = 0;
3819 #endif
3821 scale = input_line_pointer;
3822 input_line_pointer = save;
3823 return scale;
3826 static int i386_displacement PARAMS ((char *, char *));
3828 static int
3829 i386_displacement (disp_start, disp_end)
3830 char *disp_start;
3831 char *disp_end;
3833 expressionS *exp;
3834 segT exp_seg = 0;
3835 char *save_input_line_pointer;
3836 #ifndef LEX_AT
3837 char *gotfree_input_line;
3838 #endif
3839 int bigdisp = Disp32;
3841 if (flag_code == CODE_64BIT)
3843 if (i.prefix[ADDR_PREFIX] == 0)
3844 bigdisp = Disp64;
3846 else if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3847 bigdisp = Disp16;
3848 i.types[this_operand] |= bigdisp;
3850 exp = &disp_expressions[i.disp_operands];
3851 i.op[this_operand].disps = exp;
3852 i.disp_operands++;
3853 save_input_line_pointer = input_line_pointer;
3854 input_line_pointer = disp_start;
3855 END_STRING_AND_SAVE (disp_end);
3857 #ifndef GCC_ASM_O_HACK
3858 #define GCC_ASM_O_HACK 0
3859 #endif
3860 #if GCC_ASM_O_HACK
3861 END_STRING_AND_SAVE (disp_end + 1);
3862 if ((i.types[this_operand] & BaseIndex) != 0
3863 && displacement_string_end[-1] == '+')
3865 /* This hack is to avoid a warning when using the "o"
3866 constraint within gcc asm statements.
3867 For instance:
3869 #define _set_tssldt_desc(n,addr,limit,type) \
3870 __asm__ __volatile__ ( \
3871 "movw %w2,%0\n\t" \
3872 "movw %w1,2+%0\n\t" \
3873 "rorl $16,%1\n\t" \
3874 "movb %b1,4+%0\n\t" \
3875 "movb %4,5+%0\n\t" \
3876 "movb $0,6+%0\n\t" \
3877 "movb %h1,7+%0\n\t" \
3878 "rorl $16,%1" \
3879 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
3881 This works great except that the output assembler ends
3882 up looking a bit weird if it turns out that there is
3883 no offset. You end up producing code that looks like:
3885 #APP
3886 movw $235,(%eax)
3887 movw %dx,2+(%eax)
3888 rorl $16,%edx
3889 movb %dl,4+(%eax)
3890 movb $137,5+(%eax)
3891 movb $0,6+(%eax)
3892 movb %dh,7+(%eax)
3893 rorl $16,%edx
3894 #NO_APP
3896 So here we provide the missing zero. */
3898 *displacement_string_end = '0';
3900 #endif
3901 #ifndef LEX_AT
3902 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3903 if (gotfree_input_line)
3904 input_line_pointer = gotfree_input_line;
3905 #endif
3907 exp_seg = expression (exp);
3909 SKIP_WHITESPACE ();
3910 if (*input_line_pointer)
3911 as_bad (_("junk `%s' after expression"), input_line_pointer);
3912 #if GCC_ASM_O_HACK
3913 RESTORE_END_STRING (disp_end + 1);
3914 #endif
3915 RESTORE_END_STRING (disp_end);
3916 input_line_pointer = save_input_line_pointer;
3917 #ifndef LEX_AT
3918 if (gotfree_input_line)
3919 free (gotfree_input_line);
3920 #endif
3922 /* We do this to make sure that the section symbol is in
3923 the symbol table. We will ultimately change the relocation
3924 to be relative to the beginning of the section. */
3925 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
3926 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3928 if (exp->X_op != O_symbol)
3930 as_bad (_("bad expression used with @%s"),
3931 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
3932 ? "GOTPCREL"
3933 : "GOTOFF"));
3934 return 0;
3937 if (S_IS_LOCAL (exp->X_add_symbol)
3938 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
3939 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
3940 exp->X_op = O_subtract;
3941 exp->X_op_symbol = GOT_symbol;
3942 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3943 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
3944 else
3945 i.reloc[this_operand] = BFD_RELOC_32;
3948 if (exp->X_op == O_absent || exp->X_op == O_big)
3950 /* Missing or bad expr becomes absolute 0. */
3951 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
3952 disp_start);
3953 exp->X_op = O_constant;
3954 exp->X_add_number = 0;
3955 exp->X_add_symbol = (symbolS *) 0;
3956 exp->X_op_symbol = (symbolS *) 0;
3959 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3960 if (exp->X_op != O_constant
3961 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3962 && exp_seg != absolute_section
3963 && exp_seg != text_section
3964 && exp_seg != data_section
3965 && exp_seg != bss_section
3966 && exp_seg != undefined_section
3967 && !bfd_is_com_section (exp_seg))
3969 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3970 return 0;
3972 #endif
3973 else if (flag_code == CODE_64BIT)
3974 i.types[this_operand] |= Disp32S | Disp32;
3975 return 1;
3978 static int i386_index_check PARAMS ((const char *));
3980 /* Make sure the memory operand we've been dealt is valid.
3981 Return 1 on success, 0 on a failure. */
3983 static int
3984 i386_index_check (operand_string)
3985 const char *operand_string;
3987 int ok;
3988 #if INFER_ADDR_PREFIX
3989 int fudged = 0;
3991 tryprefix:
3992 #endif
3993 ok = 1;
3994 if (flag_code == CODE_64BIT)
3996 if (i.prefix[ADDR_PREFIX] == 0)
3998 /* 64bit checks. */
3999 if ((i.base_reg
4000 && ((i.base_reg->reg_type & Reg64) == 0)
4001 && (i.base_reg->reg_type != BaseIndex
4002 || i.index_reg))
4003 || (i.index_reg
4004 && ((i.index_reg->reg_type & (Reg64 | BaseIndex))
4005 != (Reg64 | BaseIndex))))
4006 ok = 0;
4008 else
4010 /* 32bit checks. */
4011 if ((i.base_reg
4012 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
4013 || (i.index_reg
4014 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
4015 != (Reg32 | BaseIndex))))
4016 ok = 0;
4019 else
4021 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
4023 /* 16bit checks. */
4024 if ((i.base_reg
4025 && ((i.base_reg->reg_type & (Reg16 | BaseIndex | RegRex))
4026 != (Reg16 | BaseIndex)))
4027 || (i.index_reg
4028 && (((i.index_reg->reg_type & (Reg16 | BaseIndex))
4029 != (Reg16 | BaseIndex))
4030 || !(i.base_reg
4031 && i.base_reg->reg_num < 6
4032 && i.index_reg->reg_num >= 6
4033 && i.log2_scale_factor == 0))))
4034 ok = 0;
4036 else
4038 /* 32bit checks. */
4039 if ((i.base_reg
4040 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
4041 || (i.index_reg
4042 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
4043 != (Reg32 | BaseIndex))))
4044 ok = 0;
4047 if (!ok)
4049 #if INFER_ADDR_PREFIX
4050 if (flag_code != CODE_64BIT
4051 && i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0')
4053 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
4054 i.prefixes += 1;
4055 /* Change the size of any displacement too. At most one of
4056 Disp16 or Disp32 is set.
4057 FIXME. There doesn't seem to be any real need for separate
4058 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
4059 Removing them would probably clean up the code quite a lot. */
4060 if (i.types[this_operand] & (Disp16 | Disp32))
4061 i.types[this_operand] ^= (Disp16 | Disp32);
4062 fudged = 1;
4063 goto tryprefix;
4065 if (fudged)
4066 as_bad (_("`%s' is not a valid base/index expression"),
4067 operand_string);
4068 else
4069 #endif
4070 as_bad (_("`%s' is not a valid %s bit base/index expression"),
4071 operand_string,
4072 flag_code_names[flag_code]);
4073 return 0;
4075 return 1;
4078 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
4079 on error. */
4081 static int
4082 i386_operand (operand_string)
4083 char *operand_string;
4085 const reg_entry *r;
4086 char *end_op;
4087 char *op_string = operand_string;
4089 if (is_space_char (*op_string))
4090 ++op_string;
4092 /* We check for an absolute prefix (differentiating,
4093 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
4094 if (*op_string == ABSOLUTE_PREFIX)
4096 ++op_string;
4097 if (is_space_char (*op_string))
4098 ++op_string;
4099 i.types[this_operand] |= JumpAbsolute;
4102 /* Check if operand is a register. */
4103 if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
4104 && (r = parse_register (op_string, &end_op)) != NULL)
4106 /* Check for a segment override by searching for ':' after a
4107 segment register. */
4108 op_string = end_op;
4109 if (is_space_char (*op_string))
4110 ++op_string;
4111 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
4113 switch (r->reg_num)
4115 case 0:
4116 i.seg[i.mem_operands] = &es;
4117 break;
4118 case 1:
4119 i.seg[i.mem_operands] = &cs;
4120 break;
4121 case 2:
4122 i.seg[i.mem_operands] = &ss;
4123 break;
4124 case 3:
4125 i.seg[i.mem_operands] = &ds;
4126 break;
4127 case 4:
4128 i.seg[i.mem_operands] = &fs;
4129 break;
4130 case 5:
4131 i.seg[i.mem_operands] = &gs;
4132 break;
4135 /* Skip the ':' and whitespace. */
4136 ++op_string;
4137 if (is_space_char (*op_string))
4138 ++op_string;
4140 if (!is_digit_char (*op_string)
4141 && !is_identifier_char (*op_string)
4142 && *op_string != '('
4143 && *op_string != ABSOLUTE_PREFIX)
4145 as_bad (_("bad memory operand `%s'"), op_string);
4146 return 0;
4148 /* Handle case of %es:*foo. */
4149 if (*op_string == ABSOLUTE_PREFIX)
4151 ++op_string;
4152 if (is_space_char (*op_string))
4153 ++op_string;
4154 i.types[this_operand] |= JumpAbsolute;
4156 goto do_memory_reference;
4158 if (*op_string)
4160 as_bad (_("junk `%s' after register"), op_string);
4161 return 0;
4163 i.types[this_operand] |= r->reg_type & ~BaseIndex;
4164 i.op[this_operand].regs = r;
4165 i.reg_operands++;
4167 else if (*op_string == REGISTER_PREFIX)
4169 as_bad (_("bad register name `%s'"), op_string);
4170 return 0;
4172 else if (*op_string == IMMEDIATE_PREFIX)
4174 ++op_string;
4175 if (i.types[this_operand] & JumpAbsolute)
4177 as_bad (_("immediate operand illegal with absolute jump"));
4178 return 0;
4180 if (!i386_immediate (op_string))
4181 return 0;
4183 else if (is_digit_char (*op_string)
4184 || is_identifier_char (*op_string)
4185 || *op_string == '(')
4187 /* This is a memory reference of some sort. */
4188 char *base_string;
4190 /* Start and end of displacement string expression (if found). */
4191 char *displacement_string_start;
4192 char *displacement_string_end;
4194 do_memory_reference:
4195 if ((i.mem_operands == 1
4196 && (current_templates->start->opcode_modifier & IsString) == 0)
4197 || i.mem_operands == 2)
4199 as_bad (_("too many memory references for `%s'"),
4200 current_templates->start->name);
4201 return 0;
4204 /* Check for base index form. We detect the base index form by
4205 looking for an ')' at the end of the operand, searching
4206 for the '(' matching it, and finding a REGISTER_PREFIX or ','
4207 after the '('. */
4208 base_string = op_string + strlen (op_string);
4210 --base_string;
4211 if (is_space_char (*base_string))
4212 --base_string;
4214 /* If we only have a displacement, set-up for it to be parsed later. */
4215 displacement_string_start = op_string;
4216 displacement_string_end = base_string + 1;
4218 if (*base_string == ')')
4220 char *temp_string;
4221 unsigned int parens_balanced = 1;
4222 /* We've already checked that the number of left & right ()'s are
4223 equal, so this loop will not be infinite. */
4226 base_string--;
4227 if (*base_string == ')')
4228 parens_balanced++;
4229 if (*base_string == '(')
4230 parens_balanced--;
4232 while (parens_balanced);
4234 temp_string = base_string;
4236 /* Skip past '(' and whitespace. */
4237 ++base_string;
4238 if (is_space_char (*base_string))
4239 ++base_string;
4241 if (*base_string == ','
4242 || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
4243 && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
4245 displacement_string_end = temp_string;
4247 i.types[this_operand] |= BaseIndex;
4249 if (i.base_reg)
4251 base_string = end_op;
4252 if (is_space_char (*base_string))
4253 ++base_string;
4256 /* There may be an index reg or scale factor here. */
4257 if (*base_string == ',')
4259 ++base_string;
4260 if (is_space_char (*base_string))
4261 ++base_string;
4263 if ((*base_string == REGISTER_PREFIX || allow_naked_reg)
4264 && (i.index_reg = parse_register (base_string, &end_op)) != NULL)
4266 base_string = end_op;
4267 if (is_space_char (*base_string))
4268 ++base_string;
4269 if (*base_string == ',')
4271 ++base_string;
4272 if (is_space_char (*base_string))
4273 ++base_string;
4275 else if (*base_string != ')')
4277 as_bad (_("expecting `,' or `)' after index register in `%s'"),
4278 operand_string);
4279 return 0;
4282 else if (*base_string == REGISTER_PREFIX)
4284 as_bad (_("bad register name `%s'"), base_string);
4285 return 0;
4288 /* Check for scale factor. */
4289 if (*base_string != ')')
4291 char *end_scale = i386_scale (base_string);
4293 if (!end_scale)
4294 return 0;
4296 base_string = end_scale;
4297 if (is_space_char (*base_string))
4298 ++base_string;
4299 if (*base_string != ')')
4301 as_bad (_("expecting `)' after scale factor in `%s'"),
4302 operand_string);
4303 return 0;
4306 else if (!i.index_reg)
4308 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
4309 *base_string);
4310 return 0;
4313 else if (*base_string != ')')
4315 as_bad (_("expecting `,' or `)' after base register in `%s'"),
4316 operand_string);
4317 return 0;
4320 else if (*base_string == REGISTER_PREFIX)
4322 as_bad (_("bad register name `%s'"), base_string);
4323 return 0;
4327 /* If there's an expression beginning the operand, parse it,
4328 assuming displacement_string_start and
4329 displacement_string_end are meaningful. */
4330 if (displacement_string_start != displacement_string_end)
4332 if (!i386_displacement (displacement_string_start,
4333 displacement_string_end))
4334 return 0;
4337 /* Special case for (%dx) while doing input/output op. */
4338 if (i.base_reg
4339 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
4340 && i.index_reg == 0
4341 && i.log2_scale_factor == 0
4342 && i.seg[i.mem_operands] == 0
4343 && (i.types[this_operand] & Disp) == 0)
4345 i.types[this_operand] = InOutPortReg;
4346 return 1;
4349 if (i386_index_check (operand_string) == 0)
4350 return 0;
4351 i.mem_operands++;
4353 else
4355 /* It's not a memory operand; argh! */
4356 as_bad (_("invalid char %s beginning operand %d `%s'"),
4357 output_invalid (*op_string),
4358 this_operand + 1,
4359 op_string);
4360 return 0;
4362 return 1; /* Normal return. */
4365 /* md_estimate_size_before_relax()
4367 Called just before relax() for rs_machine_dependent frags. The x86
4368 assembler uses these frags to handle variable size jump
4369 instructions.
4371 Any symbol that is now undefined will not become defined.
4372 Return the correct fr_subtype in the frag.
4373 Return the initial "guess for variable size of frag" to caller.
4374 The guess is actually the growth beyond the fixed part. Whatever
4375 we do to grow the fixed or variable part contributes to our
4376 returned value. */
4379 md_estimate_size_before_relax (fragP, segment)
4380 fragS *fragP;
4381 segT segment;
4383 /* We've already got fragP->fr_subtype right; all we have to do is
4384 check for un-relaxable symbols. On an ELF system, we can't relax
4385 an externally visible symbol, because it may be overridden by a
4386 shared library. */
4387 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
4388 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4389 || (OUTPUT_FLAVOR == bfd_target_elf_flavour
4390 && (S_IS_EXTERNAL (fragP->fr_symbol)
4391 || S_IS_WEAK (fragP->fr_symbol)))
4392 #endif
4395 /* Symbol is undefined in this segment, or we need to keep a
4396 reloc so that weak symbols can be overridden. */
4397 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
4398 enum bfd_reloc_code_real reloc_type;
4399 unsigned char *opcode;
4400 int old_fr_fix;
4402 if (fragP->fr_var != NO_RELOC)
4403 reloc_type = fragP->fr_var;
4404 else if (size == 2)
4405 reloc_type = BFD_RELOC_16_PCREL;
4406 else
4407 reloc_type = BFD_RELOC_32_PCREL;
4409 old_fr_fix = fragP->fr_fix;
4410 opcode = (unsigned char *) fragP->fr_opcode;
4412 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
4414 case UNCOND_JUMP:
4415 /* Make jmp (0xeb) a (d)word displacement jump. */
4416 opcode[0] = 0xe9;
4417 fragP->fr_fix += size;
4418 fix_new (fragP, old_fr_fix, size,
4419 fragP->fr_symbol,
4420 fragP->fr_offset, 1,
4421 reloc_type);
4422 break;
4424 case COND_JUMP86:
4425 if (size == 2
4426 && (!no_cond_jump_promotion || fragP->fr_var != NO_RELOC))
4428 /* Negate the condition, and branch past an
4429 unconditional jump. */
4430 opcode[0] ^= 1;
4431 opcode[1] = 3;
4432 /* Insert an unconditional jump. */
4433 opcode[2] = 0xe9;
4434 /* We added two extra opcode bytes, and have a two byte
4435 offset. */
4436 fragP->fr_fix += 2 + 2;
4437 fix_new (fragP, old_fr_fix + 2, 2,
4438 fragP->fr_symbol,
4439 fragP->fr_offset, 1,
4440 reloc_type);
4441 break;
4443 /* Fall through. */
4445 case COND_JUMP:
4446 if (no_cond_jump_promotion && fragP->fr_var == NO_RELOC)
4448 fixS *fixP;
4450 fragP->fr_fix += 1;
4451 fixP = fix_new (fragP, old_fr_fix, 1,
4452 fragP->fr_symbol,
4453 fragP->fr_offset, 1,
4454 BFD_RELOC_8_PCREL);
4455 fixP->fx_signed = 1;
4456 break;
4459 /* This changes the byte-displacement jump 0x7N
4460 to the (d)word-displacement jump 0x0f,0x8N. */
4461 opcode[1] = opcode[0] + 0x10;
4462 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4463 /* We've added an opcode byte. */
4464 fragP->fr_fix += 1 + size;
4465 fix_new (fragP, old_fr_fix + 1, size,
4466 fragP->fr_symbol,
4467 fragP->fr_offset, 1,
4468 reloc_type);
4469 break;
4471 default:
4472 BAD_CASE (fragP->fr_subtype);
4473 break;
4475 frag_wane (fragP);
4476 return fragP->fr_fix - old_fr_fix;
4479 /* Guess size depending on current relax state. Initially the relax
4480 state will correspond to a short jump and we return 1, because
4481 the variable part of the frag (the branch offset) is one byte
4482 long. However, we can relax a section more than once and in that
4483 case we must either set fr_subtype back to the unrelaxed state,
4484 or return the value for the appropriate branch. */
4485 return md_relax_table[fragP->fr_subtype].rlx_length;
4488 /* Called after relax() is finished.
4490 In: Address of frag.
4491 fr_type == rs_machine_dependent.
4492 fr_subtype is what the address relaxed to.
4494 Out: Any fixSs and constants are set up.
4495 Caller will turn frag into a ".space 0". */
4497 void
4498 md_convert_frag (abfd, sec, fragP)
4499 bfd *abfd ATTRIBUTE_UNUSED;
4500 segT sec ATTRIBUTE_UNUSED;
4501 fragS *fragP;
4503 unsigned char *opcode;
4504 unsigned char *where_to_put_displacement = NULL;
4505 offsetT target_address;
4506 offsetT opcode_address;
4507 unsigned int extension = 0;
4508 offsetT displacement_from_opcode_start;
4510 opcode = (unsigned char *) fragP->fr_opcode;
4512 /* Address we want to reach in file space. */
4513 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
4515 /* Address opcode resides at in file space. */
4516 opcode_address = fragP->fr_address + fragP->fr_fix;
4518 /* Displacement from opcode start to fill into instruction. */
4519 displacement_from_opcode_start = target_address - opcode_address;
4521 if ((fragP->fr_subtype & BIG) == 0)
4523 /* Don't have to change opcode. */
4524 extension = 1; /* 1 opcode + 1 displacement */
4525 where_to_put_displacement = &opcode[1];
4527 else
4529 if (no_cond_jump_promotion
4530 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
4531 as_warn_where (fragP->fr_file, fragP->fr_line, _("long jump required"));
4533 switch (fragP->fr_subtype)
4535 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
4536 extension = 4; /* 1 opcode + 4 displacement */
4537 opcode[0] = 0xe9;
4538 where_to_put_displacement = &opcode[1];
4539 break;
4541 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
4542 extension = 2; /* 1 opcode + 2 displacement */
4543 opcode[0] = 0xe9;
4544 where_to_put_displacement = &opcode[1];
4545 break;
4547 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
4548 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
4549 extension = 5; /* 2 opcode + 4 displacement */
4550 opcode[1] = opcode[0] + 0x10;
4551 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4552 where_to_put_displacement = &opcode[2];
4553 break;
4555 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
4556 extension = 3; /* 2 opcode + 2 displacement */
4557 opcode[1] = opcode[0] + 0x10;
4558 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4559 where_to_put_displacement = &opcode[2];
4560 break;
4562 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
4563 extension = 4;
4564 opcode[0] ^= 1;
4565 opcode[1] = 3;
4566 opcode[2] = 0xe9;
4567 where_to_put_displacement = &opcode[3];
4568 break;
4570 default:
4571 BAD_CASE (fragP->fr_subtype);
4572 break;
4576 /* Now put displacement after opcode. */
4577 md_number_to_chars ((char *) where_to_put_displacement,
4578 (valueT) (displacement_from_opcode_start - extension),
4579 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
4580 fragP->fr_fix += extension;
4583 /* Size of byte displacement jmp. */
4584 int md_short_jump_size = 2;
4586 /* Size of dword displacement jmp. */
4587 int md_long_jump_size = 5;
4589 /* Size of relocation record. */
4590 const int md_reloc_size = 8;
4592 void
4593 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
4594 char *ptr;
4595 addressT from_addr, to_addr;
4596 fragS *frag ATTRIBUTE_UNUSED;
4597 symbolS *to_symbol ATTRIBUTE_UNUSED;
4599 offsetT offset;
4601 offset = to_addr - (from_addr + 2);
4602 /* Opcode for byte-disp jump. */
4603 md_number_to_chars (ptr, (valueT) 0xeb, 1);
4604 md_number_to_chars (ptr + 1, (valueT) offset, 1);
4607 void
4608 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
4609 char *ptr;
4610 addressT from_addr, to_addr;
4611 fragS *frag ATTRIBUTE_UNUSED;
4612 symbolS *to_symbol ATTRIBUTE_UNUSED;
4614 offsetT offset;
4616 offset = to_addr - (from_addr + 5);
4617 md_number_to_chars (ptr, (valueT) 0xe9, 1);
4618 md_number_to_chars (ptr + 1, (valueT) offset, 4);
4621 /* Apply a fixup (fixS) to segment data, once it has been determined
4622 by our caller that we have all the info we need to fix it up.
4624 On the 386, immediates, displacements, and data pointers are all in
4625 the same (little-endian) format, so we don't need to care about which
4626 we are handling. */
4628 void
4629 md_apply_fix3 (fixP, valP, seg)
4630 /* The fix we're to put in. */
4631 fixS *fixP;
4632 /* Pointer to the value of the bits. */
4633 valueT *valP;
4634 /* Segment fix is from. */
4635 segT seg ATTRIBUTE_UNUSED;
4637 char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
4638 valueT value = *valP;
4640 #if !defined (TE_Mach)
4641 if (fixP->fx_pcrel)
4643 switch (fixP->fx_r_type)
4645 default:
4646 break;
4648 case BFD_RELOC_32:
4649 fixP->fx_r_type = BFD_RELOC_32_PCREL;
4650 break;
4651 case BFD_RELOC_16:
4652 fixP->fx_r_type = BFD_RELOC_16_PCREL;
4653 break;
4654 case BFD_RELOC_8:
4655 fixP->fx_r_type = BFD_RELOC_8_PCREL;
4656 break;
4660 if (fixP->fx_addsy != NULL
4661 && (fixP->fx_r_type == BFD_RELOC_32_PCREL
4662 || fixP->fx_r_type == BFD_RELOC_16_PCREL
4663 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
4664 && !use_rela_relocations)
4666 /* This is a hack. There should be a better way to handle this.
4667 This covers for the fact that bfd_install_relocation will
4668 subtract the current location (for partial_inplace, PC relative
4669 relocations); see more below. */
4670 #ifndef OBJ_AOUT
4671 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4672 #ifdef TE_PE
4673 || OUTPUT_FLAVOR == bfd_target_coff_flavour
4674 #endif
4676 value += fixP->fx_where + fixP->fx_frag->fr_address;
4677 #endif
4678 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4679 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
4681 segT sym_seg = S_GET_SEGMENT (fixP->fx_addsy);
4683 if ((sym_seg == seg
4684 || (symbol_section_p (fixP->fx_addsy)
4685 && sym_seg != absolute_section))
4686 && !generic_force_reloc (fixP))
4688 /* Yes, we add the values in twice. This is because
4689 bfd_install_relocation subtracts them out again. I think
4690 bfd_install_relocation is broken, but I don't dare change
4691 it. FIXME. */
4692 value += fixP->fx_where + fixP->fx_frag->fr_address;
4695 #endif
4696 #if defined (OBJ_COFF) && defined (TE_PE)
4697 /* For some reason, the PE format does not store a section
4698 address offset for a PC relative symbol. */
4699 if (S_GET_SEGMENT (fixP->fx_addsy) != seg)
4700 value += md_pcrel_from (fixP);
4701 #endif
4704 /* Fix a few things - the dynamic linker expects certain values here,
4705 and we must not disappoint it. */
4706 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4707 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4708 && fixP->fx_addsy)
4709 switch (fixP->fx_r_type)
4711 case BFD_RELOC_386_PLT32:
4712 case BFD_RELOC_X86_64_PLT32:
4713 /* Make the jump instruction point to the address of the operand. At
4714 runtime we merely add the offset to the actual PLT entry. */
4715 value = -4;
4716 break;
4718 case BFD_RELOC_386_TLS_GD:
4719 case BFD_RELOC_386_TLS_LDM:
4720 case BFD_RELOC_386_TLS_IE_32:
4721 case BFD_RELOC_386_TLS_IE:
4722 case BFD_RELOC_386_TLS_GOTIE:
4723 case BFD_RELOC_X86_64_TLSGD:
4724 case BFD_RELOC_X86_64_TLSLD:
4725 case BFD_RELOC_X86_64_GOTTPOFF:
4726 value = 0; /* Fully resolved at runtime. No addend. */
4727 /* Fallthrough */
4728 case BFD_RELOC_386_TLS_LE:
4729 case BFD_RELOC_386_TLS_LDO_32:
4730 case BFD_RELOC_386_TLS_LE_32:
4731 case BFD_RELOC_X86_64_DTPOFF32:
4732 case BFD_RELOC_X86_64_TPOFF32:
4733 S_SET_THREAD_LOCAL (fixP->fx_addsy);
4734 break;
4736 case BFD_RELOC_386_GOT32:
4737 case BFD_RELOC_X86_64_GOT32:
4738 value = 0; /* Fully resolved at runtime. No addend. */
4739 break;
4741 case BFD_RELOC_VTABLE_INHERIT:
4742 case BFD_RELOC_VTABLE_ENTRY:
4743 fixP->fx_done = 0;
4744 return;
4746 default:
4747 break;
4749 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
4750 *valP = value;
4751 #endif /* !defined (TE_Mach) */
4753 /* Are we finished with this relocation now? */
4754 if (fixP->fx_addsy == NULL)
4755 fixP->fx_done = 1;
4756 else if (use_rela_relocations)
4758 fixP->fx_no_overflow = 1;
4759 /* Remember value for tc_gen_reloc. */
4760 fixP->fx_addnumber = value;
4761 value = 0;
4764 md_number_to_chars (p, value, fixP->fx_size);
4767 #define MAX_LITTLENUMS 6
4769 /* Turn the string pointed to by litP into a floating point constant
4770 of type TYPE, and emit the appropriate bytes. The number of
4771 LITTLENUMS emitted is stored in *SIZEP. An error message is
4772 returned, or NULL on OK. */
4774 char *
4775 md_atof (type, litP, sizeP)
4776 int type;
4777 char *litP;
4778 int *sizeP;
4780 int prec;
4781 LITTLENUM_TYPE words[MAX_LITTLENUMS];
4782 LITTLENUM_TYPE *wordP;
4783 char *t;
4785 switch (type)
4787 case 'f':
4788 case 'F':
4789 prec = 2;
4790 break;
4792 case 'd':
4793 case 'D':
4794 prec = 4;
4795 break;
4797 case 'x':
4798 case 'X':
4799 prec = 5;
4800 break;
4802 default:
4803 *sizeP = 0;
4804 return _("Bad call to md_atof ()");
4806 t = atof_ieee (input_line_pointer, type, words);
4807 if (t)
4808 input_line_pointer = t;
4810 *sizeP = prec * sizeof (LITTLENUM_TYPE);
4811 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
4812 the bigendian 386. */
4813 for (wordP = words + prec - 1; prec--;)
4815 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
4816 litP += sizeof (LITTLENUM_TYPE);
4818 return 0;
4821 char output_invalid_buf[8];
4823 static char *
4824 output_invalid (c)
4825 int c;
4827 if (ISPRINT (c))
4828 sprintf (output_invalid_buf, "'%c'", c);
4829 else
4830 sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
4831 return output_invalid_buf;
4834 /* REG_STRING starts *before* REGISTER_PREFIX. */
4836 static const reg_entry *
4837 parse_register (reg_string, end_op)
4838 char *reg_string;
4839 char **end_op;
4841 char *s = reg_string;
4842 char *p;
4843 char reg_name_given[MAX_REG_NAME_SIZE + 1];
4844 const reg_entry *r;
4846 /* Skip possible REGISTER_PREFIX and possible whitespace. */
4847 if (*s == REGISTER_PREFIX)
4848 ++s;
4850 if (is_space_char (*s))
4851 ++s;
4853 p = reg_name_given;
4854 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
4856 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
4857 return (const reg_entry *) NULL;
4858 s++;
4861 /* For naked regs, make sure that we are not dealing with an identifier.
4862 This prevents confusing an identifier like `eax_var' with register
4863 `eax'. */
4864 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
4865 return (const reg_entry *) NULL;
4867 *end_op = s;
4869 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
4871 /* Handle floating point regs, allowing spaces in the (i) part. */
4872 if (r == i386_regtab /* %st is first entry of table */)
4874 if (is_space_char (*s))
4875 ++s;
4876 if (*s == '(')
4878 ++s;
4879 if (is_space_char (*s))
4880 ++s;
4881 if (*s >= '0' && *s <= '7')
4883 r = &i386_float_regtab[*s - '0'];
4884 ++s;
4885 if (is_space_char (*s))
4886 ++s;
4887 if (*s == ')')
4889 *end_op = s + 1;
4890 return r;
4893 /* We have "%st(" then garbage. */
4894 return (const reg_entry *) NULL;
4898 if (r != NULL
4899 && (r->reg_flags & (RegRex64 | RegRex)) != 0
4900 && flag_code != CODE_64BIT)
4902 return (const reg_entry *) NULL;
4905 return r;
4908 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4909 const char *md_shortopts = "kVQ:sqn";
4910 #else
4911 const char *md_shortopts = "qn";
4912 #endif
4914 struct option md_longopts[] = {
4915 #define OPTION_32 (OPTION_MD_BASE + 0)
4916 {"32", no_argument, NULL, OPTION_32},
4917 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4918 #define OPTION_64 (OPTION_MD_BASE + 1)
4919 {"64", no_argument, NULL, OPTION_64},
4920 #endif
4921 {NULL, no_argument, NULL, 0}
4923 size_t md_longopts_size = sizeof (md_longopts);
4926 md_parse_option (c, arg)
4927 int c;
4928 char *arg ATTRIBUTE_UNUSED;
4930 switch (c)
4932 case 'n':
4933 optimize_align_code = 0;
4934 break;
4936 case 'q':
4937 quiet_warnings = 1;
4938 break;
4940 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4941 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
4942 should be emitted or not. FIXME: Not implemented. */
4943 case 'Q':
4944 break;
4946 /* -V: SVR4 argument to print version ID. */
4947 case 'V':
4948 print_version_id ();
4949 break;
4951 /* -k: Ignore for FreeBSD compatibility. */
4952 case 'k':
4953 break;
4955 case 's':
4956 /* -s: On i386 Solaris, this tells the native assembler to use
4957 .stab instead of .stab.excl. We always use .stab anyhow. */
4958 break;
4960 case OPTION_64:
4962 const char **list, **l;
4964 list = bfd_target_list ();
4965 for (l = list; *l != NULL; l++)
4966 if (strcmp (*l, "elf64-x86-64") == 0)
4968 default_arch = "x86_64";
4969 break;
4971 if (*l == NULL)
4972 as_fatal (_("No compiled in support for x86_64"));
4973 free (list);
4975 break;
4976 #endif
4978 case OPTION_32:
4979 default_arch = "i386";
4980 break;
4982 default:
4983 return 0;
4985 return 1;
4988 void
4989 md_show_usage (stream)
4990 FILE *stream;
4992 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4993 fprintf (stream, _("\
4994 -Q ignored\n\
4995 -V print assembler version number\n\
4996 -k ignored\n\
4997 -n Do not optimize code alignment\n\
4998 -q quieten some warnings\n\
4999 -s ignored\n"));
5000 #else
5001 fprintf (stream, _("\
5002 -n Do not optimize code alignment\n\
5003 -q quieten some warnings\n"));
5004 #endif
5007 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
5008 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
5010 /* Pick the target format to use. */
5012 const char *
5013 i386_target_format ()
5015 if (!strcmp (default_arch, "x86_64"))
5016 set_code_flag (CODE_64BIT);
5017 else if (!strcmp (default_arch, "i386"))
5018 set_code_flag (CODE_32BIT);
5019 else
5020 as_fatal (_("Unknown architecture"));
5021 switch (OUTPUT_FLAVOR)
5023 #ifdef OBJ_MAYBE_AOUT
5024 case bfd_target_aout_flavour:
5025 return AOUT_TARGET_FORMAT;
5026 #endif
5027 #ifdef OBJ_MAYBE_COFF
5028 case bfd_target_coff_flavour:
5029 return "coff-i386";
5030 #endif
5031 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
5032 case bfd_target_elf_flavour:
5034 if (flag_code == CODE_64BIT)
5035 use_rela_relocations = 1;
5036 return flag_code == CODE_64BIT ? "elf64-x86-64" : ELF_TARGET_FORMAT;
5038 #endif
5039 default:
5040 abort ();
5041 return NULL;
5045 #endif /* OBJ_MAYBE_ more than one */
5047 #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
5048 void i386_elf_emit_arch_note ()
5050 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
5051 && cpu_arch_name != NULL)
5053 char *p;
5054 asection *seg = now_seg;
5055 subsegT subseg = now_subseg;
5056 Elf_Internal_Note i_note;
5057 Elf_External_Note e_note;
5058 asection *note_secp;
5059 int len;
5061 /* Create the .note section. */
5062 note_secp = subseg_new (".note", 0);
5063 bfd_set_section_flags (stdoutput,
5064 note_secp,
5065 SEC_HAS_CONTENTS | SEC_READONLY);
5067 /* Process the arch string. */
5068 len = strlen (cpu_arch_name);
5070 i_note.namesz = len + 1;
5071 i_note.descsz = 0;
5072 i_note.type = NT_ARCH;
5073 p = frag_more (sizeof (e_note.namesz));
5074 md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
5075 p = frag_more (sizeof (e_note.descsz));
5076 md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
5077 p = frag_more (sizeof (e_note.type));
5078 md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
5079 p = frag_more (len + 1);
5080 strcpy (p, cpu_arch_name);
5082 frag_align (2, 0, 0);
5084 subseg_set (seg, subseg);
5087 #endif
5089 symbolS *
5090 md_undefined_symbol (name)
5091 char *name;
5093 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
5094 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
5095 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
5096 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
5098 if (!GOT_symbol)
5100 if (symbol_find (name))
5101 as_bad (_("GOT already in symbol table"));
5102 GOT_symbol = symbol_new (name, undefined_section,
5103 (valueT) 0, &zero_address_frag);
5105 return GOT_symbol;
5107 return 0;
5110 /* Round up a section size to the appropriate boundary. */
5112 valueT
5113 md_section_align (segment, size)
5114 segT segment ATTRIBUTE_UNUSED;
5115 valueT size;
5117 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
5118 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
5120 /* For a.out, force the section size to be aligned. If we don't do
5121 this, BFD will align it for us, but it will not write out the
5122 final bytes of the section. This may be a bug in BFD, but it is
5123 easier to fix it here since that is how the other a.out targets
5124 work. */
5125 int align;
5127 align = bfd_get_section_alignment (stdoutput, segment);
5128 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
5130 #endif
5132 return size;
5135 /* On the i386, PC-relative offsets are relative to the start of the
5136 next instruction. That is, the address of the offset, plus its
5137 size, since the offset is always the last part of the insn. */
5139 long
5140 md_pcrel_from (fixP)
5141 fixS *fixP;
5143 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
5146 #ifndef I386COFF
5148 static void
5149 s_bss (ignore)
5150 int ignore ATTRIBUTE_UNUSED;
5152 int temp;
5154 temp = get_absolute_expression ();
5155 subseg_set (bss_section, (subsegT) temp);
5156 demand_empty_rest_of_line ();
5159 #endif
5161 void
5162 i386_validate_fix (fixp)
5163 fixS *fixp;
5165 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
5167 /* GOTOFF relocation are nonsense in 64bit mode. */
5168 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
5170 if (flag_code != CODE_64BIT)
5171 abort ();
5172 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
5174 else
5176 if (flag_code == CODE_64BIT)
5177 abort ();
5178 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
5180 fixp->fx_subsy = 0;
5184 arelent *
5185 tc_gen_reloc (section, fixp)
5186 asection *section ATTRIBUTE_UNUSED;
5187 fixS *fixp;
5189 arelent *rel;
5190 bfd_reloc_code_real_type code;
5192 switch (fixp->fx_r_type)
5194 case BFD_RELOC_X86_64_PLT32:
5195 case BFD_RELOC_X86_64_GOT32:
5196 case BFD_RELOC_X86_64_GOTPCREL:
5197 case BFD_RELOC_386_PLT32:
5198 case BFD_RELOC_386_GOT32:
5199 case BFD_RELOC_386_GOTOFF:
5200 case BFD_RELOC_386_GOTPC:
5201 case BFD_RELOC_386_TLS_GD:
5202 case BFD_RELOC_386_TLS_LDM:
5203 case BFD_RELOC_386_TLS_LDO_32:
5204 case BFD_RELOC_386_TLS_IE_32:
5205 case BFD_RELOC_386_TLS_IE:
5206 case BFD_RELOC_386_TLS_GOTIE:
5207 case BFD_RELOC_386_TLS_LE_32:
5208 case BFD_RELOC_386_TLS_LE:
5209 case BFD_RELOC_X86_64_32S:
5210 case BFD_RELOC_X86_64_TLSGD:
5211 case BFD_RELOC_X86_64_TLSLD:
5212 case BFD_RELOC_X86_64_DTPOFF32:
5213 case BFD_RELOC_X86_64_GOTTPOFF:
5214 case BFD_RELOC_X86_64_TPOFF32:
5215 case BFD_RELOC_RVA:
5216 case BFD_RELOC_VTABLE_ENTRY:
5217 case BFD_RELOC_VTABLE_INHERIT:
5218 #ifdef TE_PE
5219 case BFD_RELOC_32_SECREL:
5220 #endif
5221 code = fixp->fx_r_type;
5222 break;
5223 default:
5224 if (fixp->fx_pcrel)
5226 switch (fixp->fx_size)
5228 default:
5229 as_bad_where (fixp->fx_file, fixp->fx_line,
5230 _("can not do %d byte pc-relative relocation"),
5231 fixp->fx_size);
5232 code = BFD_RELOC_32_PCREL;
5233 break;
5234 case 1: code = BFD_RELOC_8_PCREL; break;
5235 case 2: code = BFD_RELOC_16_PCREL; break;
5236 case 4: code = BFD_RELOC_32_PCREL; break;
5239 else
5241 switch (fixp->fx_size)
5243 default:
5244 as_bad_where (fixp->fx_file, fixp->fx_line,
5245 _("can not do %d byte relocation"),
5246 fixp->fx_size);
5247 code = BFD_RELOC_32;
5248 break;
5249 case 1: code = BFD_RELOC_8; break;
5250 case 2: code = BFD_RELOC_16; break;
5251 case 4: code = BFD_RELOC_32; break;
5252 #ifdef BFD64
5253 case 8: code = BFD_RELOC_64; break;
5254 #endif
5257 break;
5260 if (code == BFD_RELOC_32
5261 && GOT_symbol
5262 && fixp->fx_addsy == GOT_symbol)
5264 /* We don't support GOTPC on 64bit targets. */
5265 if (flag_code == CODE_64BIT)
5266 abort ();
5267 code = BFD_RELOC_386_GOTPC;
5270 rel = (arelent *) xmalloc (sizeof (arelent));
5271 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
5272 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
5274 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
5275 if (!use_rela_relocations)
5277 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
5278 vtable entry to be used in the relocation's section offset. */
5279 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
5280 rel->address = fixp->fx_offset;
5282 rel->addend = 0;
5284 /* Use the rela in 64bit mode. */
5285 else
5287 if (!fixp->fx_pcrel)
5288 rel->addend = fixp->fx_offset;
5289 else
5290 switch (code)
5292 case BFD_RELOC_X86_64_PLT32:
5293 case BFD_RELOC_X86_64_GOT32:
5294 case BFD_RELOC_X86_64_GOTPCREL:
5295 case BFD_RELOC_X86_64_TLSGD:
5296 case BFD_RELOC_X86_64_TLSLD:
5297 case BFD_RELOC_X86_64_GOTTPOFF:
5298 rel->addend = fixp->fx_offset - fixp->fx_size;
5299 break;
5300 default:
5301 rel->addend = (section->vma
5302 - fixp->fx_size
5303 + fixp->fx_addnumber
5304 + md_pcrel_from (fixp));
5305 break;
5309 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
5310 if (rel->howto == NULL)
5312 as_bad_where (fixp->fx_file, fixp->fx_line,
5313 _("cannot represent relocation type %s"),
5314 bfd_get_reloc_code_name (code));
5315 /* Set howto to a garbage value so that we can keep going. */
5316 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
5317 assert (rel->howto != NULL);
5320 return rel;
5324 /* Parse operands using Intel syntax. This implements a recursive descent
5325 parser based on the BNF grammar published in Appendix B of the MASM 6.1
5326 Programmer's Guide.
5328 FIXME: We do not recognize the full operand grammar defined in the MASM
5329 documentation. In particular, all the structure/union and
5330 high-level macro operands are missing.
5332 Uppercase words are terminals, lower case words are non-terminals.
5333 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
5334 bars '|' denote choices. Most grammar productions are implemented in
5335 functions called 'intel_<production>'.
5337 Initial production is 'expr'.
5339 addOp + | -
5341 alpha [a-zA-Z]
5343 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
5345 constant digits [[ radixOverride ]]
5347 dataType BYTE | WORD | DWORD | QWORD | XWORD
5349 digits decdigit
5350 | digits decdigit
5351 | digits hexdigit
5353 decdigit [0-9]
5355 e05 e05 addOp e06
5356 | e06
5358 e06 e06 mulOp e09
5359 | e09
5361 e09 OFFSET e10
5362 | e09 PTR e10
5363 | e09 : e10
5364 | e10
5366 e10 e10 [ expr ]
5367 | e11
5369 e11 ( expr )
5370 | [ expr ]
5371 | constant
5372 | dataType
5373 | id
5375 | register
5377 => expr SHORT e05
5378 | e05
5380 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
5381 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
5383 hexdigit a | b | c | d | e | f
5384 | A | B | C | D | E | F
5386 id alpha
5387 | id alpha
5388 | id decdigit
5390 mulOp * | / | MOD
5392 quote " | '
5394 register specialRegister
5395 | gpRegister
5396 | byteRegister
5398 segmentRegister CS | DS | ES | FS | GS | SS
5400 specialRegister CR0 | CR2 | CR3
5401 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
5402 | TR3 | TR4 | TR5 | TR6 | TR7
5404 We simplify the grammar in obvious places (e.g., register parsing is
5405 done by calling parse_register) and eliminate immediate left recursion
5406 to implement a recursive-descent parser.
5408 expr SHORT e05
5409 | e05
5411 e05 e06 e05'
5413 e05' addOp e06 e05'
5414 | Empty
5416 e06 e09 e06'
5418 e06' mulOp e09 e06'
5419 | Empty
5421 e09 OFFSET e10 e09'
5422 | e10 e09'
5424 e09' PTR e10 e09'
5425 | : e10 e09'
5426 | Empty
5428 e10 e11 e10'
5430 e10' [ expr ] e10'
5431 | Empty
5433 e11 ( expr )
5434 | [ expr ]
5435 | BYTE
5436 | WORD
5437 | DWORD
5438 | QWORD
5439 | XWORD
5442 | register
5443 | id
5444 | constant */
5446 /* Parsing structure for the intel syntax parser. Used to implement the
5447 semantic actions for the operand grammar. */
5448 struct intel_parser_s
5450 char *op_string; /* The string being parsed. */
5451 int got_a_float; /* Whether the operand is a float. */
5452 int op_modifier; /* Operand modifier. */
5453 int is_mem; /* 1 if operand is memory reference. */
5454 const reg_entry *reg; /* Last register reference found. */
5455 char *disp; /* Displacement string being built. */
5458 static struct intel_parser_s intel_parser;
5460 /* Token structure for parsing intel syntax. */
5461 struct intel_token
5463 int code; /* Token code. */
5464 const reg_entry *reg; /* Register entry for register tokens. */
5465 char *str; /* String representation. */
5468 static struct intel_token cur_token, prev_token;
5470 /* Token codes for the intel parser. Since T_SHORT is already used
5471 by COFF, undefine it first to prevent a warning. */
5472 #define T_NIL -1
5473 #define T_CONST 1
5474 #define T_REG 2
5475 #define T_BYTE 3
5476 #define T_WORD 4
5477 #define T_DWORD 5
5478 #define T_QWORD 6
5479 #define T_XWORD 7
5480 #undef T_SHORT
5481 #define T_SHORT 8
5482 #define T_OFFSET 9
5483 #define T_PTR 10
5484 #define T_ID 11
5486 /* Prototypes for intel parser functions. */
5487 static int intel_match_token PARAMS ((int code));
5488 static void intel_get_token PARAMS ((void));
5489 static void intel_putback_token PARAMS ((void));
5490 static int intel_expr PARAMS ((void));
5491 static int intel_e05 PARAMS ((void));
5492 static int intel_e05_1 PARAMS ((void));
5493 static int intel_e06 PARAMS ((void));
5494 static int intel_e06_1 PARAMS ((void));
5495 static int intel_e09 PARAMS ((void));
5496 static int intel_e09_1 PARAMS ((void));
5497 static int intel_e10 PARAMS ((void));
5498 static int intel_e10_1 PARAMS ((void));
5499 static int intel_e11 PARAMS ((void));
5501 static int
5502 i386_intel_operand (operand_string, got_a_float)
5503 char *operand_string;
5504 int got_a_float;
5506 int ret;
5507 char *p;
5509 /* Initialize token holders. */
5510 cur_token.code = prev_token.code = T_NIL;
5511 cur_token.reg = prev_token.reg = NULL;
5512 cur_token.str = prev_token.str = NULL;
5514 /* Initialize parser structure. */
5515 p = intel_parser.op_string = (char *) malloc (strlen (operand_string) + 1);
5516 if (p == NULL)
5517 abort ();
5518 strcpy (intel_parser.op_string, operand_string);
5519 intel_parser.got_a_float = got_a_float;
5520 intel_parser.op_modifier = -1;
5521 intel_parser.is_mem = 0;
5522 intel_parser.reg = NULL;
5523 intel_parser.disp = (char *) malloc (strlen (operand_string) + 1);
5524 if (intel_parser.disp == NULL)
5525 abort ();
5526 intel_parser.disp[0] = '\0';
5528 /* Read the first token and start the parser. */
5529 intel_get_token ();
5530 ret = intel_expr ();
5532 if (ret)
5534 /* If we found a memory reference, hand it over to i386_displacement
5535 to fill in the rest of the operand fields. */
5536 if (intel_parser.is_mem)
5538 if ((i.mem_operands == 1
5539 && (current_templates->start->opcode_modifier & IsString) == 0)
5540 || i.mem_operands == 2)
5542 as_bad (_("too many memory references for '%s'"),
5543 current_templates->start->name);
5544 ret = 0;
5546 else
5548 char *s = intel_parser.disp;
5549 i.mem_operands++;
5551 /* Add the displacement expression. */
5552 if (*s != '\0')
5553 ret = i386_displacement (s, s + strlen (s));
5554 if (ret)
5555 ret = i386_index_check (operand_string);
5559 /* Constant and OFFSET expressions are handled by i386_immediate. */
5560 else if (intel_parser.op_modifier == OFFSET_FLAT
5561 || intel_parser.reg == NULL)
5562 ret = i386_immediate (intel_parser.disp);
5565 free (p);
5566 free (intel_parser.disp);
5568 return ret;
5571 /* expr SHORT e05
5572 | e05 */
5573 static int
5574 intel_expr ()
5576 /* expr SHORT e05 */
5577 if (cur_token.code == T_SHORT)
5579 intel_parser.op_modifier = SHORT;
5580 intel_match_token (T_SHORT);
5582 return (intel_e05 ());
5585 /* expr e05 */
5586 else
5587 return intel_e05 ();
5590 /* e05 e06 e05'
5592 e05' addOp e06 e05'
5593 | Empty */
5594 static int
5595 intel_e05 ()
5597 return (intel_e06 () && intel_e05_1 ());
5600 static int
5601 intel_e05_1 ()
5603 /* e05' addOp e06 e05' */
5604 if (cur_token.code == '+' || cur_token.code == '-')
5606 strcat (intel_parser.disp, cur_token.str);
5607 intel_match_token (cur_token.code);
5609 return (intel_e06 () && intel_e05_1 ());
5612 /* e05' Empty */
5613 else
5614 return 1;
5617 /* e06 e09 e06'
5619 e06' mulOp e09 e06'
5620 | Empty */
5621 static int
5622 intel_e06 ()
5624 return (intel_e09 () && intel_e06_1 ());
5627 static int
5628 intel_e06_1 ()
5630 /* e06' mulOp e09 e06' */
5631 if (cur_token.code == '*' || cur_token.code == '/')
5633 strcat (intel_parser.disp, cur_token.str);
5634 intel_match_token (cur_token.code);
5636 return (intel_e09 () && intel_e06_1 ());
5639 /* e06' Empty */
5640 else
5641 return 1;
5644 /* e09 OFFSET e10 e09'
5645 | e10 e09'
5647 e09' PTR e10 e09'
5648 | : e10 e09'
5649 | Empty */
5650 static int
5651 intel_e09 ()
5653 /* e09 OFFSET e10 e09' */
5654 if (cur_token.code == T_OFFSET)
5656 intel_parser.is_mem = 0;
5657 intel_parser.op_modifier = OFFSET_FLAT;
5658 intel_match_token (T_OFFSET);
5660 return (intel_e10 () && intel_e09_1 ());
5663 /* e09 e10 e09' */
5664 else
5665 return (intel_e10 () && intel_e09_1 ());
5668 static int
5669 intel_e09_1 ()
5671 /* e09' PTR e10 e09' */
5672 if (cur_token.code == T_PTR)
5674 if (prev_token.code == T_BYTE)
5675 i.suffix = BYTE_MNEM_SUFFIX;
5677 else if (prev_token.code == T_WORD)
5679 if (intel_parser.got_a_float == 2) /* "fi..." */
5680 i.suffix = SHORT_MNEM_SUFFIX;
5681 else
5682 i.suffix = WORD_MNEM_SUFFIX;
5685 else if (prev_token.code == T_DWORD)
5687 if (intel_parser.got_a_float == 1) /* "f..." */
5688 i.suffix = SHORT_MNEM_SUFFIX;
5689 else
5690 i.suffix = LONG_MNEM_SUFFIX;
5693 else if (prev_token.code == T_QWORD)
5695 if (intel_parser.got_a_float == 1) /* "f..." */
5696 i.suffix = LONG_MNEM_SUFFIX;
5697 else
5698 i.suffix = QWORD_MNEM_SUFFIX;
5701 else if (prev_token.code == T_XWORD)
5702 i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
5704 else
5706 as_bad (_("Unknown operand modifier `%s'\n"), prev_token.str);
5707 return 0;
5710 intel_match_token (T_PTR);
5712 return (intel_e10 () && intel_e09_1 ());
5715 /* e09 : e10 e09' */
5716 else if (cur_token.code == ':')
5718 /* Mark as a memory operand only if it's not already known to be an
5719 offset expression. */
5720 if (intel_parser.op_modifier != OFFSET_FLAT)
5721 intel_parser.is_mem = 1;
5723 return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
5726 /* e09' Empty */
5727 else
5728 return 1;
5731 /* e10 e11 e10'
5733 e10' [ expr ] e10'
5734 | Empty */
5735 static int
5736 intel_e10 ()
5738 return (intel_e11 () && intel_e10_1 ());
5741 static int
5742 intel_e10_1 ()
5744 /* e10' [ expr ] e10' */
5745 if (cur_token.code == '[')
5747 intel_match_token ('[');
5749 /* Mark as a memory operand only if it's not already known to be an
5750 offset expression. If it's an offset expression, we need to keep
5751 the brace in. */
5752 if (intel_parser.op_modifier != OFFSET_FLAT)
5753 intel_parser.is_mem = 1;
5754 else
5755 strcat (intel_parser.disp, "[");
5757 /* Add a '+' to the displacement string if necessary. */
5758 if (*intel_parser.disp != '\0'
5759 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5760 strcat (intel_parser.disp, "+");
5762 if (intel_expr () && intel_match_token (']'))
5764 /* Preserve brackets when the operand is an offset expression. */
5765 if (intel_parser.op_modifier == OFFSET_FLAT)
5766 strcat (intel_parser.disp, "]");
5768 return intel_e10_1 ();
5770 else
5771 return 0;
5774 /* e10' Empty */
5775 else
5776 return 1;
5779 /* e11 ( expr )
5780 | [ expr ]
5781 | BYTE
5782 | WORD
5783 | DWORD
5784 | QWORD
5785 | XWORD
5788 | register
5789 | id
5790 | constant */
5791 static int
5792 intel_e11 ()
5794 /* e11 ( expr ) */
5795 if (cur_token.code == '(')
5797 intel_match_token ('(');
5798 strcat (intel_parser.disp, "(");
5800 if (intel_expr () && intel_match_token (')'))
5802 strcat (intel_parser.disp, ")");
5803 return 1;
5805 else
5806 return 0;
5809 /* e11 [ expr ] */
5810 else if (cur_token.code == '[')
5812 intel_match_token ('[');
5814 /* Mark as a memory operand only if it's not already known to be an
5815 offset expression. If it's an offset expression, we need to keep
5816 the brace in. */
5817 if (intel_parser.op_modifier != OFFSET_FLAT)
5818 intel_parser.is_mem = 1;
5819 else
5820 strcat (intel_parser.disp, "[");
5822 /* Operands for jump/call inside brackets denote absolute addresses. */
5823 if (current_templates->start->opcode_modifier & Jump
5824 || current_templates->start->opcode_modifier & JumpDword
5825 || current_templates->start->opcode_modifier & JumpByte
5826 || current_templates->start->opcode_modifier & JumpInterSegment)
5827 i.types[this_operand] |= JumpAbsolute;
5829 /* Add a '+' to the displacement string if necessary. */
5830 if (*intel_parser.disp != '\0'
5831 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5832 strcat (intel_parser.disp, "+");
5834 if (intel_expr () && intel_match_token (']'))
5836 /* Preserve brackets when the operand is an offset expression. */
5837 if (intel_parser.op_modifier == OFFSET_FLAT)
5838 strcat (intel_parser.disp, "]");
5840 return 1;
5842 else
5843 return 0;
5846 /* e11 BYTE
5847 | WORD
5848 | DWORD
5849 | QWORD
5850 | XWORD */
5851 else if (cur_token.code == T_BYTE
5852 || cur_token.code == T_WORD
5853 || cur_token.code == T_DWORD
5854 || cur_token.code == T_QWORD
5855 || cur_token.code == T_XWORD)
5857 intel_match_token (cur_token.code);
5859 return 1;
5862 /* e11 $
5863 | . */
5864 else if (cur_token.code == '$' || cur_token.code == '.')
5866 strcat (intel_parser.disp, cur_token.str);
5867 intel_match_token (cur_token.code);
5869 /* Mark as a memory operand only if it's not already known to be an
5870 offset expression. */
5871 if (intel_parser.op_modifier != OFFSET_FLAT)
5872 intel_parser.is_mem = 1;
5874 return 1;
5877 /* e11 register */
5878 else if (cur_token.code == T_REG)
5880 const reg_entry *reg = intel_parser.reg = cur_token.reg;
5882 intel_match_token (T_REG);
5884 /* Check for segment change. */
5885 if (cur_token.code == ':')
5887 if (reg->reg_type & (SReg2 | SReg3))
5889 switch (reg->reg_num)
5891 case 0:
5892 i.seg[i.mem_operands] = &es;
5893 break;
5894 case 1:
5895 i.seg[i.mem_operands] = &cs;
5896 break;
5897 case 2:
5898 i.seg[i.mem_operands] = &ss;
5899 break;
5900 case 3:
5901 i.seg[i.mem_operands] = &ds;
5902 break;
5903 case 4:
5904 i.seg[i.mem_operands] = &fs;
5905 break;
5906 case 5:
5907 i.seg[i.mem_operands] = &gs;
5908 break;
5911 else
5913 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
5914 return 0;
5918 /* Not a segment register. Check for register scaling. */
5919 else if (cur_token.code == '*')
5921 if (!intel_parser.is_mem)
5923 as_bad (_("Register scaling only allowed in memory operands."));
5924 return 0;
5927 /* What follows must be a valid scale. */
5928 if (intel_match_token ('*')
5929 && strchr ("01248", *cur_token.str))
5931 i.index_reg = reg;
5932 i.types[this_operand] |= BaseIndex;
5934 /* Set the scale after setting the register (otherwise,
5935 i386_scale will complain) */
5936 i386_scale (cur_token.str);
5937 intel_match_token (T_CONST);
5939 else
5941 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
5942 cur_token.str);
5943 return 0;
5947 /* No scaling. If this is a memory operand, the register is either a
5948 base register (first occurrence) or an index register (second
5949 occurrence). */
5950 else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3)))
5952 if (i.base_reg && i.index_reg)
5954 as_bad (_("Too many register references in memory operand.\n"));
5955 return 0;
5958 if (i.base_reg == NULL)
5959 i.base_reg = reg;
5960 else
5961 i.index_reg = reg;
5963 i.types[this_operand] |= BaseIndex;
5966 /* Offset modifier. Add the register to the displacement string to be
5967 parsed as an immediate expression after we're done. */
5968 else if (intel_parser.op_modifier == OFFSET_FLAT)
5969 strcat (intel_parser.disp, reg->reg_name);
5971 /* It's neither base nor index nor offset. */
5972 else
5974 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
5975 i.op[this_operand].regs = reg;
5976 i.reg_operands++;
5979 /* Since registers are not part of the displacement string (except
5980 when we're parsing offset operands), we may need to remove any
5981 preceding '+' from the displacement string. */
5982 if (*intel_parser.disp != '\0'
5983 && intel_parser.op_modifier != OFFSET_FLAT)
5985 char *s = intel_parser.disp;
5986 s += strlen (s) - 1;
5987 if (*s == '+')
5988 *s = '\0';
5991 return 1;
5994 /* e11 id */
5995 else if (cur_token.code == T_ID)
5997 /* Add the identifier to the displacement string. */
5998 strcat (intel_parser.disp, cur_token.str);
5999 intel_match_token (T_ID);
6001 /* The identifier represents a memory reference only if it's not
6002 preceded by an offset modifier. */
6003 if (intel_parser.op_modifier != OFFSET_FLAT)
6004 intel_parser.is_mem = 1;
6006 return 1;
6009 /* e11 constant */
6010 else if (cur_token.code == T_CONST
6011 || cur_token.code == '-'
6012 || cur_token.code == '+')
6014 char *save_str;
6016 /* Allow constants that start with `+' or `-'. */
6017 if (cur_token.code == '-' || cur_token.code == '+')
6019 strcat (intel_parser.disp, cur_token.str);
6020 intel_match_token (cur_token.code);
6021 if (cur_token.code != T_CONST)
6023 as_bad (_("Syntax error. Expecting a constant. Got `%s'.\n"),
6024 cur_token.str);
6025 return 0;
6029 save_str = (char *) malloc (strlen (cur_token.str) + 1);
6030 if (save_str == NULL)
6031 abort ();
6032 strcpy (save_str, cur_token.str);
6034 /* Get the next token to check for register scaling. */
6035 intel_match_token (cur_token.code);
6037 /* Check if this constant is a scaling factor for an index register. */
6038 if (cur_token.code == '*')
6040 if (intel_match_token ('*') && cur_token.code == T_REG)
6042 if (!intel_parser.is_mem)
6044 as_bad (_("Register scaling only allowed in memory operands."));
6045 return 0;
6048 /* The constant is followed by `* reg', so it must be
6049 a valid scale. */
6050 if (strchr ("01248", *save_str))
6052 i.index_reg = cur_token.reg;
6053 i.types[this_operand] |= BaseIndex;
6055 /* Set the scale after setting the register (otherwise,
6056 i386_scale will complain) */
6057 i386_scale (save_str);
6058 intel_match_token (T_REG);
6060 /* Since registers are not part of the displacement
6061 string, we may need to remove any preceding '+' from
6062 the displacement string. */
6063 if (*intel_parser.disp != '\0')
6065 char *s = intel_parser.disp;
6066 s += strlen (s) - 1;
6067 if (*s == '+')
6068 *s = '\0';
6071 free (save_str);
6073 return 1;
6075 else
6076 return 0;
6079 /* The constant was not used for register scaling. Since we have
6080 already consumed the token following `*' we now need to put it
6081 back in the stream. */
6082 else
6083 intel_putback_token ();
6086 /* Add the constant to the displacement string. */
6087 strcat (intel_parser.disp, save_str);
6088 free (save_str);
6090 return 1;
6093 as_bad (_("Unrecognized token '%s'"), cur_token.str);
6094 return 0;
6097 /* Match the given token against cur_token. If they match, read the next
6098 token from the operand string. */
6099 static int
6100 intel_match_token (code)
6101 int code;
6103 if (cur_token.code == code)
6105 intel_get_token ();
6106 return 1;
6108 else
6110 as_bad (_("Unexpected token `%s'\n"), cur_token.str);
6111 return 0;
6115 /* Read a new token from intel_parser.op_string and store it in cur_token. */
6116 static void
6117 intel_get_token ()
6119 char *end_op;
6120 const reg_entry *reg;
6121 struct intel_token new_token;
6123 new_token.code = T_NIL;
6124 new_token.reg = NULL;
6125 new_token.str = NULL;
6127 /* Free the memory allocated to the previous token and move
6128 cur_token to prev_token. */
6129 if (prev_token.str)
6130 free (prev_token.str);
6132 prev_token = cur_token;
6134 /* Skip whitespace. */
6135 while (is_space_char (*intel_parser.op_string))
6136 intel_parser.op_string++;
6138 /* Return an empty token if we find nothing else on the line. */
6139 if (*intel_parser.op_string == '\0')
6141 cur_token = new_token;
6142 return;
6145 /* The new token cannot be larger than the remainder of the operand
6146 string. */
6147 new_token.str = (char *) malloc (strlen (intel_parser.op_string) + 1);
6148 if (new_token.str == NULL)
6149 abort ();
6150 new_token.str[0] = '\0';
6152 if (strchr ("0123456789", *intel_parser.op_string))
6154 char *p = new_token.str;
6155 char *q = intel_parser.op_string;
6156 new_token.code = T_CONST;
6158 /* Allow any kind of identifier char to encompass floating point and
6159 hexadecimal numbers. */
6160 while (is_identifier_char (*q))
6161 *p++ = *q++;
6162 *p = '\0';
6164 /* Recognize special symbol names [0-9][bf]. */
6165 if (strlen (intel_parser.op_string) == 2
6166 && (intel_parser.op_string[1] == 'b'
6167 || intel_parser.op_string[1] == 'f'))
6168 new_token.code = T_ID;
6171 else if (strchr ("+-/*:[]()", *intel_parser.op_string))
6173 new_token.code = *intel_parser.op_string;
6174 new_token.str[0] = *intel_parser.op_string;
6175 new_token.str[1] = '\0';
6178 else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
6179 && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
6181 new_token.code = T_REG;
6182 new_token.reg = reg;
6184 if (*intel_parser.op_string == REGISTER_PREFIX)
6186 new_token.str[0] = REGISTER_PREFIX;
6187 new_token.str[1] = '\0';
6190 strcat (new_token.str, reg->reg_name);
6193 else if (is_identifier_char (*intel_parser.op_string))
6195 char *p = new_token.str;
6196 char *q = intel_parser.op_string;
6198 /* A '.' or '$' followed by an identifier char is an identifier.
6199 Otherwise, it's operator '.' followed by an expression. */
6200 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
6202 new_token.code = *q;
6203 new_token.str[0] = *q;
6204 new_token.str[1] = '\0';
6206 else
6208 while (is_identifier_char (*q) || *q == '@')
6209 *p++ = *q++;
6210 *p = '\0';
6212 if (strcasecmp (new_token.str, "BYTE") == 0)
6213 new_token.code = T_BYTE;
6215 else if (strcasecmp (new_token.str, "WORD") == 0)
6216 new_token.code = T_WORD;
6218 else if (strcasecmp (new_token.str, "DWORD") == 0)
6219 new_token.code = T_DWORD;
6221 else if (strcasecmp (new_token.str, "QWORD") == 0)
6222 new_token.code = T_QWORD;
6224 else if (strcasecmp (new_token.str, "XWORD") == 0)
6225 new_token.code = T_XWORD;
6227 else if (strcasecmp (new_token.str, "PTR") == 0)
6228 new_token.code = T_PTR;
6230 else if (strcasecmp (new_token.str, "SHORT") == 0)
6231 new_token.code = T_SHORT;
6233 else if (strcasecmp (new_token.str, "OFFSET") == 0)
6235 new_token.code = T_OFFSET;
6237 /* ??? This is not mentioned in the MASM grammar but gcc
6238 makes use of it with -mintel-syntax. OFFSET may be
6239 followed by FLAT: */
6240 if (strncasecmp (q, " FLAT:", 6) == 0)
6241 strcat (new_token.str, " FLAT:");
6244 /* ??? This is not mentioned in the MASM grammar. */
6245 else if (strcasecmp (new_token.str, "FLAT") == 0)
6246 new_token.code = T_OFFSET;
6248 else
6249 new_token.code = T_ID;
6253 else
6254 as_bad (_("Unrecognized token `%s'\n"), intel_parser.op_string);
6256 intel_parser.op_string += strlen (new_token.str);
6257 cur_token = new_token;
6260 /* Put cur_token back into the token stream and make cur_token point to
6261 prev_token. */
6262 static void
6263 intel_putback_token ()
6265 intel_parser.op_string -= strlen (cur_token.str);
6266 free (cur_token.str);
6267 cur_token = prev_token;
6269 /* Forget prev_token. */
6270 prev_token.code = T_NIL;
6271 prev_token.reg = NULL;
6272 prev_token.str = NULL;
6276 tc_x86_regname_to_dw2regnum (const char *regname)
6278 unsigned int regnum;
6279 unsigned int regnames_count;
6280 char *regnames_32[] =
6282 "eax", "ecx", "edx", "ebx",
6283 "esp", "ebp", "esi", "edi",
6284 "eip"
6286 char *regnames_64[] =
6288 "rax", "rbx", "rcx", "rdx",
6289 "rdi", "rsi", "rbp", "rsp",
6290 "r8", "r9", "r10", "r11",
6291 "r12", "r13", "r14", "r15",
6292 "rip"
6294 char **regnames;
6296 if (flag_code == CODE_64BIT)
6298 regnames = regnames_64;
6299 regnames_count = ARRAY_SIZE (regnames_64);
6301 else
6303 regnames = regnames_32;
6304 regnames_count = ARRAY_SIZE (regnames_32);
6307 for (regnum = 0; regnum < regnames_count; regnum++)
6308 if (strcmp (regname, regnames[regnum]) == 0)
6309 return regnum;
6311 return -1;
6314 void
6315 tc_x86_frame_initial_instructions (void)
6317 static unsigned int sp_regno;
6319 if (!sp_regno)
6320 sp_regno = tc_x86_regname_to_dw2regnum (flag_code == CODE_64BIT
6321 ? "rsp" : "esp");
6323 cfi_add_CFA_def_cfa (sp_regno, -x86_cie_data_alignment);
6324 cfi_add_CFA_offset (x86_dwarf2_return_column, x86_cie_data_alignment);