bfd/
[binutils.git] / gas / config / tc-i386.c
blob8bc88f77ef40da8b6486a3c55ca74994c559989b
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
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 Bugs & suggestions are completely welcome. This is free software.
27 Please help us make it better. */
29 #include "as.h"
30 #include "safe-ctype.h"
31 #include "subsegs.h"
32 #include "dwarf2dbg.h"
33 #include "opcode/i386.h"
35 #ifndef REGISTER_WARNINGS
36 #define REGISTER_WARNINGS 1
37 #endif
39 #ifndef INFER_ADDR_PREFIX
40 #define INFER_ADDR_PREFIX 1
41 #endif
43 #ifndef SCALE1_WHEN_NO_INDEX
44 /* Specifying a scale factor besides 1 when there is no index is
45 futile. eg. `mov (%ebx,2),%al' does exactly the same as
46 `mov (%ebx),%al'. To slavishly follow what the programmer
47 specified, set SCALE1_WHEN_NO_INDEX to 0. */
48 #define SCALE1_WHEN_NO_INDEX 1
49 #endif
51 #ifdef BFD_ASSEMBLER
52 #define RELOC_ENUM enum bfd_reloc_code_real
53 #else
54 #define RELOC_ENUM int
55 #endif
57 #ifndef DEFAULT_ARCH
58 #define DEFAULT_ARCH "i386"
59 #endif
61 #ifndef INLINE
62 #if __GNUC__ >= 2
63 #define INLINE __inline__
64 #else
65 #define INLINE
66 #endif
67 #endif
69 static INLINE unsigned int mode_from_disp_size PARAMS ((unsigned int));
70 static INLINE int fits_in_signed_byte PARAMS ((offsetT));
71 static INLINE int fits_in_unsigned_byte PARAMS ((offsetT));
72 static INLINE int fits_in_unsigned_word PARAMS ((offsetT));
73 static INLINE int fits_in_signed_word PARAMS ((offsetT));
74 static INLINE int fits_in_unsigned_long PARAMS ((offsetT));
75 static INLINE int fits_in_signed_long PARAMS ((offsetT));
76 static int smallest_imm_type PARAMS ((offsetT));
77 static offsetT offset_in_range PARAMS ((offsetT, int));
78 static int add_prefix PARAMS ((unsigned int));
79 static void set_code_flag PARAMS ((int));
80 static void set_16bit_gcc_code_flag PARAMS ((int));
81 static void set_intel_syntax PARAMS ((int));
82 static void set_cpu_arch PARAMS ((int));
83 static char *output_invalid PARAMS ((int c));
84 static int i386_operand PARAMS ((char *operand_string));
85 static int i386_intel_operand PARAMS ((char *operand_string, int got_a_float));
86 static const reg_entry *parse_register PARAMS ((char *reg_string,
87 char **end_op));
88 static char *parse_insn PARAMS ((char *, char *));
89 static char *parse_operands PARAMS ((char *, const char *));
90 static void swap_operands PARAMS ((void));
91 static void optimize_imm PARAMS ((void));
92 static void optimize_disp PARAMS ((void));
93 static int match_template PARAMS ((void));
94 static int check_string PARAMS ((void));
95 static int process_suffix PARAMS ((void));
96 static int check_byte_reg PARAMS ((void));
97 static int check_long_reg PARAMS ((void));
98 static int check_qword_reg PARAMS ((void));
99 static int check_word_reg PARAMS ((void));
100 static int finalize_imm PARAMS ((void));
101 static int process_operands PARAMS ((void));
102 static const seg_entry *build_modrm_byte PARAMS ((void));
103 static void output_insn PARAMS ((void));
104 static void output_branch PARAMS ((void));
105 static void output_jump PARAMS ((void));
106 static void output_interseg_jump PARAMS ((void));
107 static void output_imm PARAMS ((fragS *insn_start_frag,
108 offsetT insn_start_off));
109 static void output_disp PARAMS ((fragS *insn_start_frag,
110 offsetT insn_start_off));
111 #ifndef I386COFF
112 static void s_bss PARAMS ((int));
113 #endif
115 static const char *default_arch = DEFAULT_ARCH;
117 /* 'md_assemble ()' gathers together information and puts it into a
118 i386_insn. */
120 union i386_op
122 expressionS *disps;
123 expressionS *imms;
124 const reg_entry *regs;
127 struct _i386_insn
129 /* TM holds the template for the insn were currently assembling. */
130 template tm;
132 /* SUFFIX holds the instruction mnemonic suffix if given.
133 (e.g. 'l' for 'movl') */
134 char suffix;
136 /* OPERANDS gives the number of given operands. */
137 unsigned int operands;
139 /* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number
140 of given register, displacement, memory operands and immediate
141 operands. */
142 unsigned int reg_operands, disp_operands, mem_operands, imm_operands;
144 /* TYPES [i] is the type (see above #defines) which tells us how to
145 use OP[i] for the corresponding operand. */
146 unsigned int types[MAX_OPERANDS];
148 /* Displacement expression, immediate expression, or register for each
149 operand. */
150 union i386_op op[MAX_OPERANDS];
152 /* Flags for operands. */
153 unsigned int flags[MAX_OPERANDS];
154 #define Operand_PCrel 1
156 /* Relocation type for operand */
157 RELOC_ENUM reloc[MAX_OPERANDS];
159 /* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode
160 the base index byte below. */
161 const reg_entry *base_reg;
162 const reg_entry *index_reg;
163 unsigned int log2_scale_factor;
165 /* SEG gives the seg_entries of this insn. They are zero unless
166 explicit segment overrides are given. */
167 const seg_entry *seg[2];
169 /* PREFIX holds all the given prefix opcodes (usually null).
170 PREFIXES is the number of prefix opcodes. */
171 unsigned int prefixes;
172 unsigned char prefix[MAX_PREFIXES];
174 /* RM and SIB are the modrm byte and the sib byte where the
175 addressing modes of this insn are encoded. */
177 modrm_byte rm;
178 rex_byte rex;
179 sib_byte sib;
182 typedef struct _i386_insn i386_insn;
184 /* List of chars besides those in app.c:symbol_chars that can start an
185 operand. Used to prevent the scrubber eating vital white-space. */
186 #ifdef LEX_AT
187 const char extra_symbol_chars[] = "*%-(@[";
188 #else
189 const char extra_symbol_chars[] = "*%-([";
190 #endif
192 #if (defined (TE_I386AIX) \
193 || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) \
194 && !defined (TE_LINUX) \
195 && !defined (TE_FreeBSD) \
196 && !defined (TE_NetBSD)))
197 /* This array holds the chars that always start a comment. If the
198 pre-processor is disabled, these aren't very useful. */
199 const char comment_chars[] = "#/";
200 #define PREFIX_SEPARATOR '\\'
202 /* This array holds the chars that only start a comment at the beginning of
203 a line. If the line seems to have the form '# 123 filename'
204 .line and .file directives will appear in the pre-processed output.
205 Note that input_file.c hand checks for '#' at the beginning of the
206 first line of the input file. This is because the compiler outputs
207 #NO_APP at the beginning of its output.
208 Also note that comments started like this one will always work if
209 '/' isn't otherwise defined. */
210 const char line_comment_chars[] = "";
212 #else
213 /* Putting '/' here makes it impossible to use the divide operator.
214 However, we need it for compatibility with SVR4 systems. */
215 const char comment_chars[] = "#";
216 #define PREFIX_SEPARATOR '/'
218 const char line_comment_chars[] = "/";
219 #endif
221 const char line_separator_chars[] = ";";
223 /* Chars that can be used to separate mant from exp in floating point
224 nums. */
225 const char EXP_CHARS[] = "eE";
227 /* Chars that mean this number is a floating point constant
228 As in 0f12.456
229 or 0d1.2345e12. */
230 const char FLT_CHARS[] = "fFdDxX";
232 /* Tables for lexical analysis. */
233 static char mnemonic_chars[256];
234 static char register_chars[256];
235 static char operand_chars[256];
236 static char identifier_chars[256];
237 static char digit_chars[256];
239 /* Lexical macros. */
240 #define is_mnemonic_char(x) (mnemonic_chars[(unsigned char) x])
241 #define is_operand_char(x) (operand_chars[(unsigned char) x])
242 #define is_register_char(x) (register_chars[(unsigned char) x])
243 #define is_space_char(x) ((x) == ' ')
244 #define is_identifier_char(x) (identifier_chars[(unsigned char) x])
245 #define is_digit_char(x) (digit_chars[(unsigned char) x])
247 /* All non-digit non-letter charcters that may occur in an operand. */
248 static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
250 /* md_assemble() always leaves the strings it's passed unaltered. To
251 effect this we maintain a stack of saved characters that we've smashed
252 with '\0's (indicating end of strings for various sub-fields of the
253 assembler instruction). */
254 static char save_stack[32];
255 static char *save_stack_p;
256 #define END_STRING_AND_SAVE(s) \
257 do { *save_stack_p++ = *(s); *(s) = '\0'; } while (0)
258 #define RESTORE_END_STRING(s) \
259 do { *(s) = *--save_stack_p; } while (0)
261 /* The instruction we're assembling. */
262 static i386_insn i;
264 /* Possible templates for current insn. */
265 static const templates *current_templates;
267 /* Per instruction expressionS buffers: 2 displacements & 2 immediate max. */
268 static expressionS disp_expressions[2], im_expressions[2];
270 /* Current operand we are working on. */
271 static int this_operand;
273 /* We support four different modes. FLAG_CODE variable is used to distinguish
274 these. */
276 enum flag_code {
277 CODE_32BIT,
278 CODE_16BIT,
279 CODE_64BIT };
280 #define NUM_FLAG_CODE ((int) CODE_64BIT + 1)
282 static enum flag_code flag_code;
283 static int use_rela_relocations = 0;
285 /* The names used to print error messages. */
286 static const char *flag_code_names[] =
288 "32",
289 "16",
290 "64"
293 /* 1 for intel syntax,
294 0 if att syntax. */
295 static int intel_syntax = 0;
297 /* 1 if register prefix % not required. */
298 static int allow_naked_reg = 0;
300 /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
301 leave, push, and pop instructions so that gcc has the same stack
302 frame as in 32 bit mode. */
303 static char stackop_size = '\0';
305 /* Non-zero to quieten some warnings. */
306 static int quiet_warnings = 0;
308 /* CPU name. */
309 static const char *cpu_arch_name = NULL;
311 /* CPU feature flags. */
312 static unsigned int cpu_arch_flags = CpuUnknownFlags | CpuNo64;
314 /* If set, conditional jumps are not automatically promoted to handle
315 larger than a byte offset. */
316 static unsigned int no_cond_jump_promotion = 0;
318 /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
319 symbolS *GOT_symbol;
321 /* Interface to relax_segment.
322 There are 3 major relax states for 386 jump insns because the
323 different types of jumps add different sizes to frags when we're
324 figuring out what sort of jump to choose to reach a given label. */
326 /* Types. */
327 #define UNCOND_JUMP 0
328 #define COND_JUMP 1
329 #define COND_JUMP86 2
331 /* Sizes. */
332 #define CODE16 1
333 #define SMALL 0
334 #define SMALL16 (SMALL | CODE16)
335 #define BIG 2
336 #define BIG16 (BIG | CODE16)
338 #ifndef INLINE
339 #ifdef __GNUC__
340 #define INLINE __inline__
341 #else
342 #define INLINE
343 #endif
344 #endif
346 #define ENCODE_RELAX_STATE(type, size) \
347 ((relax_substateT) (((type) << 2) | (size)))
348 #define TYPE_FROM_RELAX_STATE(s) \
349 ((s) >> 2)
350 #define DISP_SIZE_FROM_RELAX_STATE(s) \
351 ((((s) & 3) == BIG ? 4 : (((s) & 3) == BIG16 ? 2 : 1)))
353 /* This table is used by relax_frag to promote short jumps to long
354 ones where necessary. SMALL (short) jumps may be promoted to BIG
355 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
356 don't allow a short jump in a 32 bit code segment to be promoted to
357 a 16 bit offset jump because it's slower (requires data size
358 prefix), and doesn't work, unless the destination is in the bottom
359 64k of the code segment (The top 16 bits of eip are zeroed). */
361 const relax_typeS md_relax_table[] =
363 /* The fields are:
364 1) most positive reach of this state,
365 2) most negative reach of this state,
366 3) how many bytes this mode will have in the variable part of the frag
367 4) which index into the table to try if we can't fit into this one. */
369 /* UNCOND_JUMP states. */
370 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
371 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
372 /* dword jmp adds 4 bytes to frag:
373 0 extra opcode bytes, 4 displacement bytes. */
374 {0, 0, 4, 0},
375 /* word jmp adds 2 byte2 to frag:
376 0 extra opcode bytes, 2 displacement bytes. */
377 {0, 0, 2, 0},
379 /* COND_JUMP states. */
380 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
381 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
382 /* dword conditionals adds 5 bytes to frag:
383 1 extra opcode byte, 4 displacement bytes. */
384 {0, 0, 5, 0},
385 /* word conditionals add 3 bytes to frag:
386 1 extra opcode byte, 2 displacement bytes. */
387 {0, 0, 3, 0},
389 /* COND_JUMP86 states. */
390 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG)},
391 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG16)},
392 /* dword conditionals adds 5 bytes to frag:
393 1 extra opcode byte, 4 displacement bytes. */
394 {0, 0, 5, 0},
395 /* word conditionals add 4 bytes to frag:
396 1 displacement byte and a 3 byte long branch insn. */
397 {0, 0, 4, 0}
400 static const arch_entry cpu_arch[] = {
401 {"i8086", Cpu086 },
402 {"i186", Cpu086|Cpu186 },
403 {"i286", Cpu086|Cpu186|Cpu286 },
404 {"i386", Cpu086|Cpu186|Cpu286|Cpu386 },
405 {"i486", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486 },
406 {"i586", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
407 {"i686", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
408 {"pentium", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
409 {"pentiumpro",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
410 {"pentium4", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX|CpuSSE|CpuSSE2 },
411 {"k6", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow },
412 {"athlon", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuMMX|Cpu3dnow },
413 {"sledgehammer",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuSledgehammer|CpuMMX|Cpu3dnow|CpuSSE|CpuSSE2 },
414 {NULL, 0 }
417 const pseudo_typeS md_pseudo_table[] =
419 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
420 {"align", s_align_bytes, 0},
421 #else
422 {"align", s_align_ptwo, 0},
423 #endif
424 {"arch", set_cpu_arch, 0},
425 #ifndef I386COFF
426 {"bss", s_bss, 0},
427 #endif
428 {"ffloat", float_cons, 'f'},
429 {"dfloat", float_cons, 'd'},
430 {"tfloat", float_cons, 'x'},
431 {"value", cons, 2},
432 {"noopt", s_ignore, 0},
433 {"optim", s_ignore, 0},
434 {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
435 {"code16", set_code_flag, CODE_16BIT},
436 {"code32", set_code_flag, CODE_32BIT},
437 {"code64", set_code_flag, CODE_64BIT},
438 {"intel_syntax", set_intel_syntax, 1},
439 {"att_syntax", set_intel_syntax, 0},
440 {"file", (void (*) PARAMS ((int))) dwarf2_directive_file, 0},
441 {"loc", dwarf2_directive_loc, 0},
442 {0, 0, 0}
445 /* For interface with expression (). */
446 extern char *input_line_pointer;
448 /* Hash table for instruction mnemonic lookup. */
449 static struct hash_control *op_hash;
451 /* Hash table for register lookup. */
452 static struct hash_control *reg_hash;
454 void
455 i386_align_code (fragP, count)
456 fragS *fragP;
457 int count;
459 /* Various efficient no-op patterns for aligning code labels.
460 Note: Don't try to assemble the instructions in the comments.
461 0L and 0w are not legal. */
462 static const char f32_1[] =
463 {0x90}; /* nop */
464 static const char f32_2[] =
465 {0x89,0xf6}; /* movl %esi,%esi */
466 static const char f32_3[] =
467 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
468 static const char f32_4[] =
469 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
470 static const char f32_5[] =
471 {0x90, /* nop */
472 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
473 static const char f32_6[] =
474 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
475 static const char f32_7[] =
476 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
477 static const char f32_8[] =
478 {0x90, /* nop */
479 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
480 static const char f32_9[] =
481 {0x89,0xf6, /* movl %esi,%esi */
482 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
483 static const char f32_10[] =
484 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
485 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
486 static const char f32_11[] =
487 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
488 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
489 static const char f32_12[] =
490 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
491 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
492 static const char f32_13[] =
493 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
494 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
495 static const char f32_14[] =
496 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
497 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
498 static const char f32_15[] =
499 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
500 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
501 static const char f16_3[] =
502 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
503 static const char f16_4[] =
504 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
505 static const char f16_5[] =
506 {0x90, /* nop */
507 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
508 static const char f16_6[] =
509 {0x89,0xf6, /* mov %si,%si */
510 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
511 static const char f16_7[] =
512 {0x8d,0x74,0x00, /* lea 0(%si),%si */
513 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
514 static const char f16_8[] =
515 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
516 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
517 static const char *const f32_patt[] = {
518 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
519 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
521 static const char *const f16_patt[] = {
522 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
523 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
526 if (count <= 0 || count > 15)
527 return;
529 /* The recommended way to pad 64bit code is to use NOPs preceded by
530 maximally four 0x66 prefixes. Balance the size of nops. */
531 if (flag_code == CODE_64BIT)
533 int i;
534 int nnops = (count + 3) / 4;
535 int len = count / nnops;
536 int remains = count - nnops * len;
537 int pos = 0;
539 for (i = 0; i < remains; i++)
541 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len);
542 fragP->fr_literal[fragP->fr_fix + pos + len] = 0x90;
543 pos += len + 1;
545 for (; i < nnops; i++)
547 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len - 1);
548 fragP->fr_literal[fragP->fr_fix + pos + len - 1] = 0x90;
549 pos += len;
552 else
553 if (flag_code == CODE_16BIT)
555 memcpy (fragP->fr_literal + fragP->fr_fix,
556 f16_patt[count - 1], count);
557 if (count > 8)
558 /* Adjust jump offset. */
559 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
561 else
562 memcpy (fragP->fr_literal + fragP->fr_fix,
563 f32_patt[count - 1], count);
564 fragP->fr_var = count;
567 static INLINE unsigned int
568 mode_from_disp_size (t)
569 unsigned int t;
571 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
574 static INLINE int
575 fits_in_signed_byte (num)
576 offsetT num;
578 return (num >= -128) && (num <= 127);
581 static INLINE int
582 fits_in_unsigned_byte (num)
583 offsetT num;
585 return (num & 0xff) == num;
588 static INLINE int
589 fits_in_unsigned_word (num)
590 offsetT num;
592 return (num & 0xffff) == num;
595 static INLINE int
596 fits_in_signed_word (num)
597 offsetT num;
599 return (-32768 <= num) && (num <= 32767);
601 static INLINE int
602 fits_in_signed_long (num)
603 offsetT num ATTRIBUTE_UNUSED;
605 #ifndef BFD64
606 return 1;
607 #else
608 return (!(((offsetT) -1 << 31) & num)
609 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
610 #endif
611 } /* fits_in_signed_long() */
612 static INLINE int
613 fits_in_unsigned_long (num)
614 offsetT num ATTRIBUTE_UNUSED;
616 #ifndef BFD64
617 return 1;
618 #else
619 return (num & (((offsetT) 2 << 31) - 1)) == num;
620 #endif
621 } /* fits_in_unsigned_long() */
623 static int
624 smallest_imm_type (num)
625 offsetT num;
627 if (cpu_arch_flags != (Cpu086 | Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64))
629 /* This code is disabled on the 486 because all the Imm1 forms
630 in the opcode table are slower on the i486. They're the
631 versions with the implicitly specified single-position
632 displacement, which has another syntax if you really want to
633 use that form. */
634 if (num == 1)
635 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
637 return (fits_in_signed_byte (num)
638 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
639 : fits_in_unsigned_byte (num)
640 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
641 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
642 ? (Imm16 | Imm32 | Imm32S | Imm64)
643 : fits_in_signed_long (num)
644 ? (Imm32 | Imm32S | Imm64)
645 : fits_in_unsigned_long (num)
646 ? (Imm32 | Imm64)
647 : Imm64);
650 static offsetT
651 offset_in_range (val, size)
652 offsetT val;
653 int size;
655 addressT mask;
657 switch (size)
659 case 1: mask = ((addressT) 1 << 8) - 1; break;
660 case 2: mask = ((addressT) 1 << 16) - 1; break;
661 case 4: mask = ((addressT) 2 << 31) - 1; break;
662 #ifdef BFD64
663 case 8: mask = ((addressT) 2 << 63) - 1; break;
664 #endif
665 default: abort ();
668 /* If BFD64, sign extend val. */
669 if (!use_rela_relocations)
670 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
671 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
673 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
675 char buf1[40], buf2[40];
677 sprint_value (buf1, val);
678 sprint_value (buf2, val & mask);
679 as_warn (_("%s shortened to %s"), buf1, buf2);
681 return val & mask;
684 /* Returns 0 if attempting to add a prefix where one from the same
685 class already exists, 1 if non rep/repne added, 2 if rep/repne
686 added. */
687 static int
688 add_prefix (prefix)
689 unsigned int prefix;
691 int ret = 1;
692 int q;
694 if (prefix >= REX_OPCODE && prefix < REX_OPCODE + 16
695 && flag_code == CODE_64BIT)
696 q = REX_PREFIX;
697 else
698 switch (prefix)
700 default:
701 abort ();
703 case CS_PREFIX_OPCODE:
704 case DS_PREFIX_OPCODE:
705 case ES_PREFIX_OPCODE:
706 case FS_PREFIX_OPCODE:
707 case GS_PREFIX_OPCODE:
708 case SS_PREFIX_OPCODE:
709 q = SEG_PREFIX;
710 break;
712 case REPNE_PREFIX_OPCODE:
713 case REPE_PREFIX_OPCODE:
714 ret = 2;
715 /* fall thru */
716 case LOCK_PREFIX_OPCODE:
717 q = LOCKREP_PREFIX;
718 break;
720 case FWAIT_OPCODE:
721 q = WAIT_PREFIX;
722 break;
724 case ADDR_PREFIX_OPCODE:
725 q = ADDR_PREFIX;
726 break;
728 case DATA_PREFIX_OPCODE:
729 q = DATA_PREFIX;
730 break;
733 if (i.prefix[q] != 0)
735 as_bad (_("same type of prefix used twice"));
736 return 0;
739 i.prefixes += 1;
740 i.prefix[q] = prefix;
741 return ret;
744 static void
745 set_code_flag (value)
746 int value;
748 flag_code = value;
749 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
750 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
751 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
753 as_bad (_("64bit mode not supported on this CPU."));
755 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
757 as_bad (_("32bit mode not supported on this CPU."));
759 stackop_size = '\0';
762 static void
763 set_16bit_gcc_code_flag (new_code_flag)
764 int new_code_flag;
766 flag_code = new_code_flag;
767 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
768 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
769 stackop_size = 'l';
772 static void
773 set_intel_syntax (syntax_flag)
774 int syntax_flag;
776 /* Find out if register prefixing is specified. */
777 int ask_naked_reg = 0;
779 SKIP_WHITESPACE ();
780 if (!is_end_of_line[(unsigned char) *input_line_pointer])
782 char *string = input_line_pointer;
783 int e = get_symbol_end ();
785 if (strcmp (string, "prefix") == 0)
786 ask_naked_reg = 1;
787 else if (strcmp (string, "noprefix") == 0)
788 ask_naked_reg = -1;
789 else
790 as_bad (_("bad argument to syntax directive."));
791 *input_line_pointer = e;
793 demand_empty_rest_of_line ();
795 intel_syntax = syntax_flag;
797 if (ask_naked_reg == 0)
799 #ifdef BFD_ASSEMBLER
800 allow_naked_reg = (intel_syntax
801 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
802 #else
803 /* Conservative default. */
804 allow_naked_reg = 0;
805 #endif
807 else
808 allow_naked_reg = (ask_naked_reg < 0);
811 static void
812 set_cpu_arch (dummy)
813 int dummy ATTRIBUTE_UNUSED;
815 SKIP_WHITESPACE ();
817 if (!is_end_of_line[(unsigned char) *input_line_pointer])
819 char *string = input_line_pointer;
820 int e = get_symbol_end ();
821 int i;
823 for (i = 0; cpu_arch[i].name; i++)
825 if (strcmp (string, cpu_arch[i].name) == 0)
827 cpu_arch_name = cpu_arch[i].name;
828 cpu_arch_flags = (cpu_arch[i].flags
829 | (flag_code == CODE_64BIT ? Cpu64 : CpuNo64));
830 break;
833 if (!cpu_arch[i].name)
834 as_bad (_("no such architecture: `%s'"), string);
836 *input_line_pointer = e;
838 else
839 as_bad (_("missing cpu architecture"));
841 no_cond_jump_promotion = 0;
842 if (*input_line_pointer == ','
843 && !is_end_of_line[(unsigned char) input_line_pointer[1]])
845 char *string = ++input_line_pointer;
846 int e = get_symbol_end ();
848 if (strcmp (string, "nojumps") == 0)
849 no_cond_jump_promotion = 1;
850 else if (strcmp (string, "jumps") == 0)
852 else
853 as_bad (_("no such architecture modifier: `%s'"), string);
855 *input_line_pointer = e;
858 demand_empty_rest_of_line ();
861 #ifdef BFD_ASSEMBLER
862 unsigned long
863 i386_mach ()
865 if (!strcmp (default_arch, "x86_64"))
866 return bfd_mach_x86_64;
867 else if (!strcmp (default_arch, "i386"))
868 return bfd_mach_i386_i386;
869 else
870 as_fatal (_("Unknown architecture"));
872 #endif
874 void
875 md_begin ()
877 const char *hash_err;
879 /* Initialize op_hash hash table. */
880 op_hash = hash_new ();
883 const template *optab;
884 templates *core_optab;
886 /* Setup for loop. */
887 optab = i386_optab;
888 core_optab = (templates *) xmalloc (sizeof (templates));
889 core_optab->start = optab;
891 while (1)
893 ++optab;
894 if (optab->name == NULL
895 || strcmp (optab->name, (optab - 1)->name) != 0)
897 /* different name --> ship out current template list;
898 add to hash table; & begin anew. */
899 core_optab->end = optab;
900 hash_err = hash_insert (op_hash,
901 (optab - 1)->name,
902 (PTR) core_optab);
903 if (hash_err)
905 as_fatal (_("Internal Error: Can't hash %s: %s"),
906 (optab - 1)->name,
907 hash_err);
909 if (optab->name == NULL)
910 break;
911 core_optab = (templates *) xmalloc (sizeof (templates));
912 core_optab->start = optab;
917 /* Initialize reg_hash hash table. */
918 reg_hash = hash_new ();
920 const reg_entry *regtab;
922 for (regtab = i386_regtab;
923 regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
924 regtab++)
926 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
927 if (hash_err)
928 as_fatal (_("Internal Error: Can't hash %s: %s"),
929 regtab->reg_name,
930 hash_err);
934 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
936 int c;
937 char *p;
939 for (c = 0; c < 256; c++)
941 if (ISDIGIT (c))
943 digit_chars[c] = c;
944 mnemonic_chars[c] = c;
945 register_chars[c] = c;
946 operand_chars[c] = c;
948 else if (ISLOWER (c))
950 mnemonic_chars[c] = c;
951 register_chars[c] = c;
952 operand_chars[c] = c;
954 else if (ISUPPER (c))
956 mnemonic_chars[c] = TOLOWER (c);
957 register_chars[c] = mnemonic_chars[c];
958 operand_chars[c] = c;
961 if (ISALPHA (c) || ISDIGIT (c))
962 identifier_chars[c] = c;
963 else if (c >= 128)
965 identifier_chars[c] = c;
966 operand_chars[c] = c;
970 #ifdef LEX_AT
971 identifier_chars['@'] = '@';
972 #endif
973 digit_chars['-'] = '-';
974 identifier_chars['_'] = '_';
975 identifier_chars['.'] = '.';
977 for (p = operand_special_chars; *p != '\0'; p++)
978 operand_chars[(unsigned char) *p] = *p;
981 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
982 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
984 record_alignment (text_section, 2);
985 record_alignment (data_section, 2);
986 record_alignment (bss_section, 2);
988 #endif
991 void
992 i386_print_statistics (file)
993 FILE *file;
995 hash_print_statistics (file, "i386 opcode", op_hash);
996 hash_print_statistics (file, "i386 register", reg_hash);
999 #ifdef DEBUG386
1001 /* Debugging routines for md_assemble. */
1002 static void pi PARAMS ((char *, i386_insn *));
1003 static void pte PARAMS ((template *));
1004 static void pt PARAMS ((unsigned int));
1005 static void pe PARAMS ((expressionS *));
1006 static void ps PARAMS ((symbolS *));
1008 static void
1009 pi (line, x)
1010 char *line;
1011 i386_insn *x;
1013 unsigned int i;
1015 fprintf (stdout, "%s: template ", line);
1016 pte (&x->tm);
1017 fprintf (stdout, " address: base %s index %s scale %x\n",
1018 x->base_reg ? x->base_reg->reg_name : "none",
1019 x->index_reg ? x->index_reg->reg_name : "none",
1020 x->log2_scale_factor);
1021 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
1022 x->rm.mode, x->rm.reg, x->rm.regmem);
1023 fprintf (stdout, " sib: base %x index %x scale %x\n",
1024 x->sib.base, x->sib.index, x->sib.scale);
1025 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
1026 (x->rex & REX_MODE64) != 0,
1027 (x->rex & REX_EXTX) != 0,
1028 (x->rex & REX_EXTY) != 0,
1029 (x->rex & REX_EXTZ) != 0);
1030 for (i = 0; i < x->operands; i++)
1032 fprintf (stdout, " #%d: ", i + 1);
1033 pt (x->types[i]);
1034 fprintf (stdout, "\n");
1035 if (x->types[i]
1036 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
1037 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
1038 if (x->types[i] & Imm)
1039 pe (x->op[i].imms);
1040 if (x->types[i] & Disp)
1041 pe (x->op[i].disps);
1045 static void
1046 pte (t)
1047 template *t;
1049 unsigned int i;
1050 fprintf (stdout, " %d operands ", t->operands);
1051 fprintf (stdout, "opcode %x ", t->base_opcode);
1052 if (t->extension_opcode != None)
1053 fprintf (stdout, "ext %x ", t->extension_opcode);
1054 if (t->opcode_modifier & D)
1055 fprintf (stdout, "D");
1056 if (t->opcode_modifier & W)
1057 fprintf (stdout, "W");
1058 fprintf (stdout, "\n");
1059 for (i = 0; i < t->operands; i++)
1061 fprintf (stdout, " #%d type ", i + 1);
1062 pt (t->operand_types[i]);
1063 fprintf (stdout, "\n");
1067 static void
1068 pe (e)
1069 expressionS *e;
1071 fprintf (stdout, " operation %d\n", e->X_op);
1072 fprintf (stdout, " add_number %ld (%lx)\n",
1073 (long) e->X_add_number, (long) e->X_add_number);
1074 if (e->X_add_symbol)
1076 fprintf (stdout, " add_symbol ");
1077 ps (e->X_add_symbol);
1078 fprintf (stdout, "\n");
1080 if (e->X_op_symbol)
1082 fprintf (stdout, " op_symbol ");
1083 ps (e->X_op_symbol);
1084 fprintf (stdout, "\n");
1088 static void
1089 ps (s)
1090 symbolS *s;
1092 fprintf (stdout, "%s type %s%s",
1093 S_GET_NAME (s),
1094 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
1095 segment_name (S_GET_SEGMENT (s)));
1098 struct type_name
1100 unsigned int mask;
1101 char *tname;
1104 static const type_names[] =
1106 { Reg8, "r8" },
1107 { Reg16, "r16" },
1108 { Reg32, "r32" },
1109 { Reg64, "r64" },
1110 { Imm8, "i8" },
1111 { Imm8S, "i8s" },
1112 { Imm16, "i16" },
1113 { Imm32, "i32" },
1114 { Imm32S, "i32s" },
1115 { Imm64, "i64" },
1116 { Imm1, "i1" },
1117 { BaseIndex, "BaseIndex" },
1118 { Disp8, "d8" },
1119 { Disp16, "d16" },
1120 { Disp32, "d32" },
1121 { Disp32S, "d32s" },
1122 { Disp64, "d64" },
1123 { InOutPortReg, "InOutPortReg" },
1124 { ShiftCount, "ShiftCount" },
1125 { Control, "control reg" },
1126 { Test, "test reg" },
1127 { Debug, "debug reg" },
1128 { FloatReg, "FReg" },
1129 { FloatAcc, "FAcc" },
1130 { SReg2, "SReg2" },
1131 { SReg3, "SReg3" },
1132 { Acc, "Acc" },
1133 { JumpAbsolute, "Jump Absolute" },
1134 { RegMMX, "rMMX" },
1135 { RegXMM, "rXMM" },
1136 { EsSeg, "es" },
1137 { 0, "" }
1140 static void
1141 pt (t)
1142 unsigned int t;
1144 const struct type_name *ty;
1146 for (ty = type_names; ty->mask; ty++)
1147 if (t & ty->mask)
1148 fprintf (stdout, "%s, ", ty->tname);
1149 fflush (stdout);
1152 #endif /* DEBUG386 */
1154 #ifdef BFD_ASSEMBLER
1155 static bfd_reloc_code_real_type reloc
1156 PARAMS ((int, int, int, bfd_reloc_code_real_type));
1158 static bfd_reloc_code_real_type
1159 reloc (size, pcrel, sign, other)
1160 int size;
1161 int pcrel;
1162 int sign;
1163 bfd_reloc_code_real_type other;
1165 if (other != NO_RELOC)
1166 return other;
1168 if (pcrel)
1170 if (!sign)
1171 as_bad (_("There are no unsigned pc-relative relocations"));
1172 switch (size)
1174 case 1: return BFD_RELOC_8_PCREL;
1175 case 2: return BFD_RELOC_16_PCREL;
1176 case 4: return BFD_RELOC_32_PCREL;
1178 as_bad (_("can not do %d byte pc-relative relocation"), size);
1180 else
1182 if (sign)
1183 switch (size)
1185 case 4: return BFD_RELOC_X86_64_32S;
1187 else
1188 switch (size)
1190 case 1: return BFD_RELOC_8;
1191 case 2: return BFD_RELOC_16;
1192 case 4: return BFD_RELOC_32;
1193 case 8: return BFD_RELOC_64;
1195 as_bad (_("can not do %s %d byte relocation"),
1196 sign ? "signed" : "unsigned", size);
1199 abort ();
1200 return BFD_RELOC_NONE;
1203 /* Here we decide which fixups can be adjusted to make them relative to
1204 the beginning of the section instead of the symbol. Basically we need
1205 to make sure that the dynamic relocations are done correctly, so in
1206 some cases we force the original symbol to be used. */
1209 tc_i386_fix_adjustable (fixP)
1210 fixS *fixP ATTRIBUTE_UNUSED;
1212 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1213 if (OUTPUT_FLAVOR != bfd_target_elf_flavour)
1214 return 1;
1216 /* Don't adjust pc-relative references to merge sections in 64-bit
1217 mode. */
1218 if (use_rela_relocations
1219 && (S_GET_SEGMENT (fixP->fx_addsy)->flags & SEC_MERGE) != 0
1220 && fixP->fx_pcrel)
1221 return 0;
1223 /* adjust_reloc_syms doesn't know about the GOT. */
1224 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1225 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1226 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1227 || fixP->fx_r_type == BFD_RELOC_386_TLS_GD
1228 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDM
1229 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDO_32
1230 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE_32
1231 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE
1232 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTIE
1233 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE_32
1234 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE
1235 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1236 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1237 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
1238 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1239 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1240 return 0;
1241 #endif
1242 return 1;
1244 #else
1245 #define reloc(SIZE,PCREL,SIGN,OTHER) 0
1246 #define BFD_RELOC_8 0
1247 #define BFD_RELOC_16 0
1248 #define BFD_RELOC_32 0
1249 #define BFD_RELOC_8_PCREL 0
1250 #define BFD_RELOC_16_PCREL 0
1251 #define BFD_RELOC_32_PCREL 0
1252 #define BFD_RELOC_386_PLT32 0
1253 #define BFD_RELOC_386_GOT32 0
1254 #define BFD_RELOC_386_GOTOFF 0
1255 #define BFD_RELOC_386_TLS_GD 0
1256 #define BFD_RELOC_386_TLS_LDM 0
1257 #define BFD_RELOC_386_TLS_LDO_32 0
1258 #define BFD_RELOC_386_TLS_IE_32 0
1259 #define BFD_RELOC_386_TLS_IE 0
1260 #define BFD_RELOC_386_TLS_GOTIE 0
1261 #define BFD_RELOC_386_TLS_LE_32 0
1262 #define BFD_RELOC_386_TLS_LE 0
1263 #define BFD_RELOC_X86_64_PLT32 0
1264 #define BFD_RELOC_X86_64_GOT32 0
1265 #define BFD_RELOC_X86_64_GOTPCREL 0
1266 #endif
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 /* These AMD 3DNow! and Intel Katmai New Instructions have an
1386 opcode suffix which is coded in the same place as an 8-bit
1387 immediate field would be. Here we fake an 8-bit immediate
1388 operand from the opcode suffix stored in tm.extension_opcode. */
1390 expressionS *exp;
1392 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
1394 exp = &im_expressions[i.imm_operands++];
1395 i.op[i.operands].imms = exp;
1396 i.types[i.operands++] = Imm8;
1397 exp->X_op = O_constant;
1398 exp->X_add_number = i.tm.extension_opcode;
1399 i.tm.extension_opcode = None;
1402 /* For insns with operands there are more diddles to do to the opcode. */
1403 if (i.operands)
1405 if (!process_operands ())
1406 return;
1408 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
1410 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
1411 as_warn (_("translating to `%sp'"), i.tm.name);
1414 /* Handle conversion of 'int $3' --> special int3 insn. */
1415 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
1417 i.tm.base_opcode = INT3_OPCODE;
1418 i.imm_operands = 0;
1421 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
1422 && i.op[0].disps->X_op == O_constant)
1424 /* Convert "jmp constant" (and "call constant") to a jump (call) to
1425 the absolute address given by the constant. Since ix86 jumps and
1426 calls are pc relative, we need to generate a reloc. */
1427 i.op[0].disps->X_add_symbol = &abs_symbol;
1428 i.op[0].disps->X_op = O_symbol;
1431 if ((i.tm.opcode_modifier & Rex64) != 0)
1432 i.rex |= REX_MODE64;
1434 /* For 8 bit registers we need an empty rex prefix. Also if the
1435 instruction already has a prefix, we need to convert old
1436 registers to new ones. */
1438 if (((i.types[0] & Reg8) != 0
1439 && (i.op[0].regs->reg_flags & RegRex64) != 0)
1440 || ((i.types[1] & Reg8) != 0
1441 && (i.op[1].regs->reg_flags & RegRex64) != 0)
1442 || (((i.types[0] & Reg8) != 0 || (i.types[1] & Reg8) != 0)
1443 && i.rex != 0))
1445 int x;
1447 i.rex |= REX_OPCODE;
1448 for (x = 0; x < 2; x++)
1450 /* Look for 8 bit operand that uses old registers. */
1451 if ((i.types[x] & Reg8) != 0
1452 && (i.op[x].regs->reg_flags & RegRex64) == 0)
1454 /* In case it is "hi" register, give up. */
1455 if (i.op[x].regs->reg_num > 3)
1456 as_bad (_("can't encode register '%%%s' in an instruction requiring REX prefix.\n"),
1457 i.op[x].regs->reg_name);
1459 /* Otherwise it is equivalent to the extended register.
1460 Since the encoding doesn't change this is merely
1461 cosmetic cleanup for debug output. */
1463 i.op[x].regs = i.op[x].regs + 8;
1468 if (i.rex != 0)
1469 add_prefix (REX_OPCODE | i.rex);
1471 /* We are ready to output the insn. */
1472 output_insn ();
1475 static char *
1476 parse_insn (line, mnemonic)
1477 char *line;
1478 char *mnemonic;
1480 char *l = line;
1481 char *token_start = l;
1482 char *mnem_p;
1484 /* Non-zero if we found a prefix only acceptable with string insns. */
1485 const char *expecting_string_instruction = NULL;
1487 while (1)
1489 mnem_p = mnemonic;
1490 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1492 mnem_p++;
1493 if (mnem_p >= mnemonic + MAX_MNEM_SIZE)
1495 as_bad (_("no such instruction: `%s'"), token_start);
1496 return NULL;
1498 l++;
1500 if (!is_space_char (*l)
1501 && *l != END_OF_INSN
1502 && *l != PREFIX_SEPARATOR
1503 && *l != ',')
1505 as_bad (_("invalid character %s in mnemonic"),
1506 output_invalid (*l));
1507 return NULL;
1509 if (token_start == l)
1511 if (*l == PREFIX_SEPARATOR)
1512 as_bad (_("expecting prefix; got nothing"));
1513 else
1514 as_bad (_("expecting mnemonic; got nothing"));
1515 return NULL;
1518 /* Look up instruction (or prefix) via hash table. */
1519 current_templates = hash_find (op_hash, mnemonic);
1521 if (*l != END_OF_INSN
1522 && (!is_space_char (*l) || l[1] != END_OF_INSN)
1523 && current_templates
1524 && (current_templates->start->opcode_modifier & IsPrefix))
1526 /* If we are in 16-bit mode, do not allow addr16 or data16.
1527 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1528 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1529 && flag_code != CODE_64BIT
1530 && (((current_templates->start->opcode_modifier & Size32) != 0)
1531 ^ (flag_code == CODE_16BIT)))
1533 as_bad (_("redundant %s prefix"),
1534 current_templates->start->name);
1535 return NULL;
1537 /* Add prefix, checking for repeated prefixes. */
1538 switch (add_prefix (current_templates->start->base_opcode))
1540 case 0:
1541 return NULL;
1542 case 2:
1543 expecting_string_instruction = current_templates->start->name;
1544 break;
1546 /* Skip past PREFIX_SEPARATOR and reset token_start. */
1547 token_start = ++l;
1549 else
1550 break;
1553 if (!current_templates)
1555 /* See if we can get a match by trimming off a suffix. */
1556 switch (mnem_p[-1])
1558 case WORD_MNEM_SUFFIX:
1559 case BYTE_MNEM_SUFFIX:
1560 case QWORD_MNEM_SUFFIX:
1561 i.suffix = mnem_p[-1];
1562 mnem_p[-1] = '\0';
1563 current_templates = hash_find (op_hash, mnemonic);
1564 break;
1565 case SHORT_MNEM_SUFFIX:
1566 case LONG_MNEM_SUFFIX:
1567 if (!intel_syntax)
1569 i.suffix = mnem_p[-1];
1570 mnem_p[-1] = '\0';
1571 current_templates = hash_find (op_hash, mnemonic);
1573 break;
1575 /* Intel Syntax. */
1576 case 'd':
1577 if (intel_syntax)
1579 if (intel_float_operand (mnemonic))
1580 i.suffix = SHORT_MNEM_SUFFIX;
1581 else
1582 i.suffix = LONG_MNEM_SUFFIX;
1583 mnem_p[-1] = '\0';
1584 current_templates = hash_find (op_hash, mnemonic);
1586 break;
1588 if (!current_templates)
1590 as_bad (_("no such instruction: `%s'"), token_start);
1591 return NULL;
1595 if (current_templates->start->opcode_modifier & (Jump | JumpByte))
1597 /* Check for a branch hint. We allow ",pt" and ",pn" for
1598 predict taken and predict not taken respectively.
1599 I'm not sure that branch hints actually do anything on loop
1600 and jcxz insns (JumpByte) for current Pentium4 chips. They
1601 may work in the future and it doesn't hurt to accept them
1602 now. */
1603 if (l[0] == ',' && l[1] == 'p')
1605 if (l[2] == 't')
1607 if (!add_prefix (DS_PREFIX_OPCODE))
1608 return NULL;
1609 l += 3;
1611 else if (l[2] == 'n')
1613 if (!add_prefix (CS_PREFIX_OPCODE))
1614 return NULL;
1615 l += 3;
1619 /* Any other comma loses. */
1620 if (*l == ',')
1622 as_bad (_("invalid character %s in mnemonic"),
1623 output_invalid (*l));
1624 return NULL;
1627 /* Check if instruction is supported on specified architecture. */
1628 if ((current_templates->start->cpu_flags & ~(Cpu64 | CpuNo64))
1629 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64)))
1631 as_warn (_("`%s' is not supported on `%s'"),
1632 current_templates->start->name, cpu_arch_name);
1634 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
1636 as_warn (_("use .code16 to ensure correct addressing mode"));
1639 /* Check for rep/repne without a string instruction. */
1640 if (expecting_string_instruction
1641 && !(current_templates->start->opcode_modifier & IsString))
1643 as_bad (_("expecting string instruction after `%s'"),
1644 expecting_string_instruction);
1645 return NULL;
1648 return l;
1651 static char *
1652 parse_operands (l, mnemonic)
1653 char *l;
1654 const char *mnemonic;
1656 char *token_start;
1658 /* 1 if operand is pending after ','. */
1659 unsigned int expecting_operand = 0;
1661 /* Non-zero if operand parens not balanced. */
1662 unsigned int paren_not_balanced;
1664 while (*l != END_OF_INSN)
1666 /* Skip optional white space before operand. */
1667 if (is_space_char (*l))
1668 ++l;
1669 if (!is_operand_char (*l) && *l != END_OF_INSN)
1671 as_bad (_("invalid character %s before operand %d"),
1672 output_invalid (*l),
1673 i.operands + 1);
1674 return NULL;
1676 token_start = l; /* after white space */
1677 paren_not_balanced = 0;
1678 while (paren_not_balanced || *l != ',')
1680 if (*l == END_OF_INSN)
1682 if (paren_not_balanced)
1684 if (!intel_syntax)
1685 as_bad (_("unbalanced parenthesis in operand %d."),
1686 i.operands + 1);
1687 else
1688 as_bad (_("unbalanced brackets in operand %d."),
1689 i.operands + 1);
1690 return NULL;
1692 else
1693 break; /* we are done */
1695 else if (!is_operand_char (*l) && !is_space_char (*l))
1697 as_bad (_("invalid character %s in operand %d"),
1698 output_invalid (*l),
1699 i.operands + 1);
1700 return NULL;
1702 if (!intel_syntax)
1704 if (*l == '(')
1705 ++paren_not_balanced;
1706 if (*l == ')')
1707 --paren_not_balanced;
1709 else
1711 if (*l == '[')
1712 ++paren_not_balanced;
1713 if (*l == ']')
1714 --paren_not_balanced;
1716 l++;
1718 if (l != token_start)
1719 { /* Yes, we've read in another operand. */
1720 unsigned int operand_ok;
1721 this_operand = i.operands++;
1722 if (i.operands > MAX_OPERANDS)
1724 as_bad (_("spurious operands; (%d operands/instruction max)"),
1725 MAX_OPERANDS);
1726 return NULL;
1728 /* Now parse operand adding info to 'i' as we go along. */
1729 END_STRING_AND_SAVE (l);
1731 if (intel_syntax)
1732 operand_ok =
1733 i386_intel_operand (token_start,
1734 intel_float_operand (mnemonic));
1735 else
1736 operand_ok = i386_operand (token_start);
1738 RESTORE_END_STRING (l);
1739 if (!operand_ok)
1740 return NULL;
1742 else
1744 if (expecting_operand)
1746 expecting_operand_after_comma:
1747 as_bad (_("expecting operand after ','; got nothing"));
1748 return NULL;
1750 if (*l == ',')
1752 as_bad (_("expecting operand before ','; got nothing"));
1753 return NULL;
1757 /* Now *l must be either ',' or END_OF_INSN. */
1758 if (*l == ',')
1760 if (*++l == END_OF_INSN)
1762 /* Just skip it, if it's \n complain. */
1763 goto expecting_operand_after_comma;
1765 expecting_operand = 1;
1768 return l;
1771 static void
1772 swap_operands ()
1774 union i386_op temp_op;
1775 unsigned int temp_type;
1776 RELOC_ENUM temp_reloc;
1777 int xchg1 = 0;
1778 int xchg2 = 0;
1780 if (i.operands == 2)
1782 xchg1 = 0;
1783 xchg2 = 1;
1785 else if (i.operands == 3)
1787 xchg1 = 0;
1788 xchg2 = 2;
1790 temp_type = i.types[xchg2];
1791 i.types[xchg2] = i.types[xchg1];
1792 i.types[xchg1] = temp_type;
1793 temp_op = i.op[xchg2];
1794 i.op[xchg2] = i.op[xchg1];
1795 i.op[xchg1] = temp_op;
1796 temp_reloc = i.reloc[xchg2];
1797 i.reloc[xchg2] = i.reloc[xchg1];
1798 i.reloc[xchg1] = temp_reloc;
1800 if (i.mem_operands == 2)
1802 const seg_entry *temp_seg;
1803 temp_seg = i.seg[0];
1804 i.seg[0] = i.seg[1];
1805 i.seg[1] = temp_seg;
1809 /* Try to ensure constant immediates are represented in the smallest
1810 opcode possible. */
1811 static void
1812 optimize_imm ()
1814 char guess_suffix = 0;
1815 int op;
1817 if (i.suffix)
1818 guess_suffix = i.suffix;
1819 else if (i.reg_operands)
1821 /* Figure out a suffix from the last register operand specified.
1822 We can't do this properly yet, ie. excluding InOutPortReg,
1823 but the following works for instructions with immediates.
1824 In any case, we can't set i.suffix yet. */
1825 for (op = i.operands; --op >= 0;)
1826 if (i.types[op] & Reg)
1828 if (i.types[op] & Reg8)
1829 guess_suffix = BYTE_MNEM_SUFFIX;
1830 else if (i.types[op] & Reg16)
1831 guess_suffix = WORD_MNEM_SUFFIX;
1832 else if (i.types[op] & Reg32)
1833 guess_suffix = LONG_MNEM_SUFFIX;
1834 else if (i.types[op] & Reg64)
1835 guess_suffix = QWORD_MNEM_SUFFIX;
1836 break;
1839 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
1840 guess_suffix = WORD_MNEM_SUFFIX;
1842 for (op = i.operands; --op >= 0;)
1843 if (i.types[op] & Imm)
1845 switch (i.op[op].imms->X_op)
1847 case O_constant:
1848 /* If a suffix is given, this operand may be shortened. */
1849 switch (guess_suffix)
1851 case LONG_MNEM_SUFFIX:
1852 i.types[op] |= Imm32 | Imm64;
1853 break;
1854 case WORD_MNEM_SUFFIX:
1855 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
1856 break;
1857 case BYTE_MNEM_SUFFIX:
1858 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
1859 break;
1862 /* If this operand is at most 16 bits, convert it
1863 to a signed 16 bit number before trying to see
1864 whether it will fit in an even smaller size.
1865 This allows a 16-bit operand such as $0xffe0 to
1866 be recognised as within Imm8S range. */
1867 if ((i.types[op] & Imm16)
1868 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
1870 i.op[op].imms->X_add_number =
1871 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
1873 if ((i.types[op] & Imm32)
1874 && ((i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1))
1875 == 0))
1877 i.op[op].imms->X_add_number = ((i.op[op].imms->X_add_number
1878 ^ ((offsetT) 1 << 31))
1879 - ((offsetT) 1 << 31));
1881 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
1883 /* We must avoid matching of Imm32 templates when 64bit
1884 only immediate is available. */
1885 if (guess_suffix == QWORD_MNEM_SUFFIX)
1886 i.types[op] &= ~Imm32;
1887 break;
1889 case O_absent:
1890 case O_register:
1891 abort ();
1893 /* Symbols and expressions. */
1894 default:
1895 /* Convert symbolic operand to proper sizes for matching. */
1896 switch (guess_suffix)
1898 case QWORD_MNEM_SUFFIX:
1899 i.types[op] = Imm64 | Imm32S;
1900 break;
1901 case LONG_MNEM_SUFFIX:
1902 i.types[op] = Imm32 | Imm64;
1903 break;
1904 case WORD_MNEM_SUFFIX:
1905 i.types[op] = Imm16 | Imm32 | Imm64;
1906 break;
1907 break;
1908 case BYTE_MNEM_SUFFIX:
1909 i.types[op] = Imm8 | Imm8S | Imm16 | Imm32S | Imm32;
1910 break;
1911 break;
1913 break;
1918 /* Try to use the smallest displacement type too. */
1919 static void
1920 optimize_disp ()
1922 int op;
1924 for (op = i.operands; --op >= 0;)
1925 if ((i.types[op] & Disp) && i.op[op].disps->X_op == O_constant)
1927 offsetT disp = i.op[op].disps->X_add_number;
1929 if (i.types[op] & Disp16)
1931 /* We know this operand is at most 16 bits, so
1932 convert to a signed 16 bit number before trying
1933 to see whether it will fit in an even smaller
1934 size. */
1936 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
1938 else if (i.types[op] & Disp32)
1940 /* We know this operand is at most 32 bits, so convert to a
1941 signed 32 bit number before trying to see whether it will
1942 fit in an even smaller size. */
1943 disp &= (((offsetT) 2 << 31) - 1);
1944 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1946 if (flag_code == CODE_64BIT)
1948 if (fits_in_signed_long (disp))
1949 i.types[op] |= Disp32S;
1950 if (fits_in_unsigned_long (disp))
1951 i.types[op] |= Disp32;
1953 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
1954 && fits_in_signed_byte (disp))
1955 i.types[op] |= Disp8;
1959 static int
1960 match_template ()
1962 /* Points to template once we've found it. */
1963 const template *t;
1964 unsigned int overlap0, overlap1, overlap2;
1965 unsigned int found_reverse_match;
1966 int suffix_check;
1968 #define MATCH(overlap, given, template) \
1969 ((overlap & ~JumpAbsolute) \
1970 && (((given) & (BaseIndex | JumpAbsolute)) \
1971 == ((overlap) & (BaseIndex | JumpAbsolute))))
1973 /* If given types r0 and r1 are registers they must be of the same type
1974 unless the expected operand type register overlap is null.
1975 Note that Acc in a template matches every size of reg. */
1976 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
1977 (((g0) & Reg) == 0 || ((g1) & Reg) == 0 \
1978 || ((g0) & Reg) == ((g1) & Reg) \
1979 || ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
1981 overlap0 = 0;
1982 overlap1 = 0;
1983 overlap2 = 0;
1984 found_reverse_match = 0;
1985 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
1986 ? No_bSuf
1987 : (i.suffix == WORD_MNEM_SUFFIX
1988 ? No_wSuf
1989 : (i.suffix == SHORT_MNEM_SUFFIX
1990 ? No_sSuf
1991 : (i.suffix == LONG_MNEM_SUFFIX
1992 ? No_lSuf
1993 : (i.suffix == QWORD_MNEM_SUFFIX
1994 ? No_qSuf
1995 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX
1996 ? No_xSuf : 0))))));
1998 for (t = current_templates->start;
1999 t < current_templates->end;
2000 t++)
2002 /* Must have right number of operands. */
2003 if (i.operands != t->operands)
2004 continue;
2006 /* Check the suffix, except for some instructions in intel mode. */
2007 if ((t->opcode_modifier & suffix_check)
2008 && !(intel_syntax
2009 && (t->opcode_modifier & IgnoreSize))
2010 && !(intel_syntax
2011 && t->base_opcode == 0xd9
2012 && (t->extension_opcode == 5 /* 0xd9,5 "fldcw" */
2013 || t->extension_opcode == 7))) /* 0xd9,7 "f{n}stcw" */
2014 continue;
2016 /* Do not verify operands when there are none. */
2017 else if (!t->operands)
2019 if (t->cpu_flags & ~cpu_arch_flags)
2020 continue;
2021 /* We've found a match; break out of loop. */
2022 break;
2025 overlap0 = i.types[0] & t->operand_types[0];
2026 switch (t->operands)
2028 case 1:
2029 if (!MATCH (overlap0, i.types[0], t->operand_types[0]))
2030 continue;
2031 break;
2032 case 2:
2033 case 3:
2034 overlap1 = i.types[1] & t->operand_types[1];
2035 if (!MATCH (overlap0, i.types[0], t->operand_types[0])
2036 || !MATCH (overlap1, i.types[1], t->operand_types[1])
2037 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2038 t->operand_types[0],
2039 overlap1, i.types[1],
2040 t->operand_types[1]))
2042 /* Check if other direction is valid ... */
2043 if ((t->opcode_modifier & (D | FloatD)) == 0)
2044 continue;
2046 /* Try reversing direction of operands. */
2047 overlap0 = i.types[0] & t->operand_types[1];
2048 overlap1 = i.types[1] & t->operand_types[0];
2049 if (!MATCH (overlap0, i.types[0], t->operand_types[1])
2050 || !MATCH (overlap1, i.types[1], t->operand_types[0])
2051 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2052 t->operand_types[1],
2053 overlap1, i.types[1],
2054 t->operand_types[0]))
2056 /* Does not match either direction. */
2057 continue;
2059 /* found_reverse_match holds which of D or FloatDR
2060 we've found. */
2061 found_reverse_match = t->opcode_modifier & (D | FloatDR);
2063 /* Found a forward 2 operand match here. */
2064 else if (t->operands == 3)
2066 /* Here we make use of the fact that there are no
2067 reverse match 3 operand instructions, and all 3
2068 operand instructions only need to be checked for
2069 register consistency between operands 2 and 3. */
2070 overlap2 = i.types[2] & t->operand_types[2];
2071 if (!MATCH (overlap2, i.types[2], t->operand_types[2])
2072 || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1],
2073 t->operand_types[1],
2074 overlap2, i.types[2],
2075 t->operand_types[2]))
2077 continue;
2079 /* Found either forward/reverse 2 or 3 operand match here:
2080 slip through to break. */
2082 if (t->cpu_flags & ~cpu_arch_flags)
2084 found_reverse_match = 0;
2085 continue;
2087 /* We've found a match; break out of loop. */
2088 break;
2091 if (t == current_templates->end)
2093 /* We found no match. */
2094 as_bad (_("suffix or operands invalid for `%s'"),
2095 current_templates->start->name);
2096 return 0;
2099 if (!quiet_warnings)
2101 if (!intel_syntax
2102 && ((i.types[0] & JumpAbsolute)
2103 != (t->operand_types[0] & JumpAbsolute)))
2105 as_warn (_("indirect %s without `*'"), t->name);
2108 if ((t->opcode_modifier & (IsPrefix | IgnoreSize))
2109 == (IsPrefix | IgnoreSize))
2111 /* Warn them that a data or address size prefix doesn't
2112 affect assembly of the next line of code. */
2113 as_warn (_("stand-alone `%s' prefix"), t->name);
2117 /* Copy the template we found. */
2118 i.tm = *t;
2119 if (found_reverse_match)
2121 /* If we found a reverse match we must alter the opcode
2122 direction bit. found_reverse_match holds bits to change
2123 (different for int & float insns). */
2125 i.tm.base_opcode ^= found_reverse_match;
2127 i.tm.operand_types[0] = t->operand_types[1];
2128 i.tm.operand_types[1] = t->operand_types[0];
2131 return 1;
2134 static int
2135 check_string ()
2137 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
2138 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
2140 if (i.seg[0] != NULL && i.seg[0] != &es)
2142 as_bad (_("`%s' operand %d must use `%%es' segment"),
2143 i.tm.name,
2144 mem_op + 1);
2145 return 0;
2147 /* There's only ever one segment override allowed per instruction.
2148 This instruction possibly has a legal segment override on the
2149 second operand, so copy the segment to where non-string
2150 instructions store it, allowing common code. */
2151 i.seg[0] = i.seg[1];
2153 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
2155 if (i.seg[1] != NULL && i.seg[1] != &es)
2157 as_bad (_("`%s' operand %d must use `%%es' segment"),
2158 i.tm.name,
2159 mem_op + 2);
2160 return 0;
2163 return 1;
2166 static int
2167 process_suffix ()
2169 /* If matched instruction specifies an explicit instruction mnemonic
2170 suffix, use it. */
2171 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
2173 if (i.tm.opcode_modifier & Size16)
2174 i.suffix = WORD_MNEM_SUFFIX;
2175 else if (i.tm.opcode_modifier & Size64)
2176 i.suffix = QWORD_MNEM_SUFFIX;
2177 else
2178 i.suffix = LONG_MNEM_SUFFIX;
2180 else if (i.reg_operands)
2182 /* If there's no instruction mnemonic suffix we try to invent one
2183 based on register operands. */
2184 if (!i.suffix)
2186 /* We take i.suffix from the last register operand specified,
2187 Destination register type is more significant than source
2188 register type. */
2189 int op;
2190 for (op = i.operands; --op >= 0;)
2191 if ((i.types[op] & Reg)
2192 && !(i.tm.operand_types[op] & InOutPortReg))
2194 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
2195 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
2196 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
2197 LONG_MNEM_SUFFIX);
2198 break;
2201 else if (i.suffix == BYTE_MNEM_SUFFIX)
2203 if (!check_byte_reg ())
2204 return 0;
2206 else if (i.suffix == LONG_MNEM_SUFFIX)
2208 if (!check_long_reg ())
2209 return 0;
2211 else if (i.suffix == QWORD_MNEM_SUFFIX)
2213 if (!check_qword_reg ())
2214 return 0;
2216 else if (i.suffix == WORD_MNEM_SUFFIX)
2218 if (!check_word_reg ())
2219 return 0;
2221 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2222 /* Do nothing if the instruction is going to ignore the prefix. */
2224 else
2225 abort ();
2227 else if ((i.tm.opcode_modifier & DefaultSize) && !i.suffix)
2229 i.suffix = stackop_size;
2232 /* Change the opcode based on the operand size given by i.suffix;
2233 We need not change things for byte insns. */
2235 if (!i.suffix && (i.tm.opcode_modifier & W))
2237 as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
2238 return 0;
2241 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2243 /* It's not a byte, select word/dword operation. */
2244 if (i.tm.opcode_modifier & W)
2246 if (i.tm.opcode_modifier & ShortForm)
2247 i.tm.base_opcode |= 8;
2248 else
2249 i.tm.base_opcode |= 1;
2252 /* Now select between word & dword operations via the operand
2253 size prefix, except for instructions that will ignore this
2254 prefix anyway. */
2255 if (i.suffix != QWORD_MNEM_SUFFIX
2256 && !(i.tm.opcode_modifier & IgnoreSize)
2257 && ((i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2258 || (flag_code == CODE_64BIT
2259 && (i.tm.opcode_modifier & JumpByte))))
2261 unsigned int prefix = DATA_PREFIX_OPCODE;
2262 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2263 prefix = ADDR_PREFIX_OPCODE;
2265 if (!add_prefix (prefix))
2266 return 0;
2269 /* Set mode64 for an operand. */
2270 if (i.suffix == QWORD_MNEM_SUFFIX
2271 && flag_code == CODE_64BIT
2272 && (i.tm.opcode_modifier & NoRex64) == 0)
2273 i.rex |= REX_MODE64;
2275 /* Size floating point instruction. */
2276 if (i.suffix == LONG_MNEM_SUFFIX)
2278 if (i.tm.opcode_modifier & FloatMF)
2279 i.tm.base_opcode ^= 4;
2283 return 1;
2286 static int
2287 check_byte_reg ()
2289 int op;
2290 for (op = i.operands; --op >= 0;)
2292 /* If this is an eight bit register, it's OK. If it's the 16 or
2293 32 bit version of an eight bit register, we will just use the
2294 low portion, and that's OK too. */
2295 if (i.types[op] & Reg8)
2296 continue;
2298 /* movzx and movsx should not generate this warning. */
2299 if (intel_syntax
2300 && (i.tm.base_opcode == 0xfb7
2301 || i.tm.base_opcode == 0xfb6
2302 || i.tm.base_opcode == 0x63
2303 || i.tm.base_opcode == 0xfbe
2304 || i.tm.base_opcode == 0xfbf))
2305 continue;
2307 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4
2308 #if 0
2309 /* Check that the template allows eight bit regs. This
2310 kills insns such as `orb $1,%edx', which maybe should be
2311 allowed. */
2312 && (i.tm.operand_types[op] & (Reg8 | InOutPortReg))
2313 #endif
2316 /* Prohibit these changes in the 64bit mode, since the
2317 lowering is more complicated. */
2318 if (flag_code == CODE_64BIT
2319 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2321 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2322 i.op[op].regs->reg_name,
2323 i.suffix);
2324 return 0;
2326 #if REGISTER_WARNINGS
2327 if (!quiet_warnings
2328 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2329 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2330 (i.op[op].regs + (i.types[op] & Reg16
2331 ? REGNAM_AL - REGNAM_AX
2332 : REGNAM_AL - REGNAM_EAX))->reg_name,
2333 i.op[op].regs->reg_name,
2334 i.suffix);
2335 #endif
2336 continue;
2338 /* Any other register is bad. */
2339 if (i.types[op] & (Reg | RegMMX | RegXMM
2340 | SReg2 | SReg3
2341 | Control | Debug | Test
2342 | FloatReg | FloatAcc))
2344 as_bad (_("`%%%s' not allowed with `%s%c'"),
2345 i.op[op].regs->reg_name,
2346 i.tm.name,
2347 i.suffix);
2348 return 0;
2351 return 1;
2354 static int
2355 check_long_reg ()
2357 int op;
2359 for (op = i.operands; --op >= 0;)
2360 /* Reject eight bit registers, except where the template requires
2361 them. (eg. movzb) */
2362 if ((i.types[op] & Reg8) != 0
2363 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2365 as_bad (_("`%%%s' not allowed with `%s%c'"),
2366 i.op[op].regs->reg_name,
2367 i.tm.name,
2368 i.suffix);
2369 return 0;
2371 /* Warn if the e prefix on a general reg is missing. */
2372 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2373 && (i.types[op] & Reg16) != 0
2374 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2376 /* Prohibit these changes in the 64bit mode, since the
2377 lowering is more complicated. */
2378 if (flag_code == CODE_64BIT)
2380 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2381 i.op[op].regs->reg_name,
2382 i.suffix);
2383 return 0;
2385 #if REGISTER_WARNINGS
2386 else
2387 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2388 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
2389 i.op[op].regs->reg_name,
2390 i.suffix);
2391 #endif
2393 /* Warn if the r prefix on a general reg is missing. */
2394 else if ((i.types[op] & Reg64) != 0
2395 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2397 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2398 i.op[op].regs->reg_name,
2399 i.suffix);
2400 return 0;
2402 return 1;
2405 static int
2406 check_qword_reg ()
2408 int op;
2410 for (op = i.operands; --op >= 0; )
2411 /* Reject eight bit registers, except where the template requires
2412 them. (eg. movzb) */
2413 if ((i.types[op] & Reg8) != 0
2414 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2416 as_bad (_("`%%%s' not allowed with `%s%c'"),
2417 i.op[op].regs->reg_name,
2418 i.tm.name,
2419 i.suffix);
2420 return 0;
2422 /* Warn if the e prefix on a general reg is missing. */
2423 else if (((i.types[op] & Reg16) != 0
2424 || (i.types[op] & Reg32) != 0)
2425 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2427 /* Prohibit these changes in the 64bit mode, since the
2428 lowering is more complicated. */
2429 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2430 i.op[op].regs->reg_name,
2431 i.suffix);
2432 return 0;
2434 return 1;
2437 static int
2438 check_word_reg ()
2440 int op;
2441 for (op = i.operands; --op >= 0;)
2442 /* Reject eight bit registers, except where the template requires
2443 them. (eg. movzb) */
2444 if ((i.types[op] & Reg8) != 0
2445 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2447 as_bad (_("`%%%s' not allowed with `%s%c'"),
2448 i.op[op].regs->reg_name,
2449 i.tm.name,
2450 i.suffix);
2451 return 0;
2453 /* Warn if the e prefix on a general reg is present. */
2454 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2455 && (i.types[op] & Reg32) != 0
2456 && (i.tm.operand_types[op] & (Reg16 | Acc)) != 0)
2458 /* Prohibit these changes in the 64bit mode, since the
2459 lowering is more complicated. */
2460 if (flag_code == CODE_64BIT)
2462 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2463 i.op[op].regs->reg_name,
2464 i.suffix);
2465 return 0;
2467 else
2468 #if REGISTER_WARNINGS
2469 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2470 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
2471 i.op[op].regs->reg_name,
2472 i.suffix);
2473 #endif
2475 return 1;
2478 static int
2479 finalize_imm ()
2481 unsigned int overlap0, overlap1, overlap2;
2483 overlap0 = i.types[0] & i.tm.operand_types[0];
2484 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S))
2485 && overlap0 != Imm8 && overlap0 != Imm8S
2486 && overlap0 != Imm16 && overlap0 != Imm32S
2487 && overlap0 != Imm32 && overlap0 != Imm64)
2489 if (i.suffix)
2491 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX
2492 ? Imm8 | Imm8S
2493 : (i.suffix == WORD_MNEM_SUFFIX
2494 ? Imm16
2495 : (i.suffix == QWORD_MNEM_SUFFIX
2496 ? Imm64 | Imm32S
2497 : Imm32)));
2499 else if (overlap0 == (Imm16 | Imm32S | Imm32)
2500 || overlap0 == (Imm16 | Imm32)
2501 || overlap0 == (Imm16 | Imm32S))
2503 overlap0 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
2504 ? Imm16 : Imm32S);
2506 if (overlap0 != Imm8 && overlap0 != Imm8S
2507 && overlap0 != Imm16 && overlap0 != Imm32S
2508 && overlap0 != Imm32 && overlap0 != Imm64)
2510 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
2511 return 0;
2514 i.types[0] = overlap0;
2516 overlap1 = i.types[1] & i.tm.operand_types[1];
2517 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32))
2518 && overlap1 != Imm8 && overlap1 != Imm8S
2519 && overlap1 != Imm16 && overlap1 != Imm32S
2520 && overlap1 != Imm32 && overlap1 != Imm64)
2522 if (i.suffix)
2524 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX
2525 ? Imm8 | Imm8S
2526 : (i.suffix == WORD_MNEM_SUFFIX
2527 ? Imm16
2528 : (i.suffix == QWORD_MNEM_SUFFIX
2529 ? Imm64 | Imm32S
2530 : Imm32)));
2532 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
2533 || overlap1 == (Imm16 | Imm32)
2534 || overlap1 == (Imm16 | Imm32S))
2536 overlap1 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
2537 ? Imm16 : Imm32S);
2539 if (overlap1 != Imm8 && overlap1 != Imm8S
2540 && overlap1 != Imm16 && overlap1 != Imm32S
2541 && overlap1 != Imm32 && overlap1 != Imm64)
2543 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
2544 return 0;
2547 i.types[1] = overlap1;
2549 overlap2 = i.types[2] & i.tm.operand_types[2];
2550 assert ((overlap2 & Imm) == 0);
2551 i.types[2] = overlap2;
2553 return 1;
2556 static int
2557 process_operands ()
2559 /* Default segment register this instruction will use for memory
2560 accesses. 0 means unknown. This is only for optimizing out
2561 unnecessary segment overrides. */
2562 const seg_entry *default_seg = 0;
2564 /* The imul $imm, %reg instruction is converted into
2565 imul $imm, %reg, %reg, and the clr %reg instruction
2566 is converted into xor %reg, %reg. */
2567 if (i.tm.opcode_modifier & regKludge)
2569 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
2570 /* Pretend we saw the extra register operand. */
2571 assert (i.op[first_reg_op + 1].regs == 0);
2572 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
2573 i.types[first_reg_op + 1] = i.types[first_reg_op];
2574 i.reg_operands = 2;
2577 if (i.tm.opcode_modifier & ShortForm)
2579 /* The register or float register operand is in operand 0 or 1. */
2580 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
2581 /* Register goes in low 3 bits of opcode. */
2582 i.tm.base_opcode |= i.op[op].regs->reg_num;
2583 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2584 i.rex |= REX_EXTZ;
2585 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2587 /* Warn about some common errors, but press on regardless.
2588 The first case can be generated by gcc (<= 2.8.1). */
2589 if (i.operands == 2)
2591 /* Reversed arguments on faddp, fsubp, etc. */
2592 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
2593 i.op[1].regs->reg_name,
2594 i.op[0].regs->reg_name);
2596 else
2598 /* Extraneous `l' suffix on fp insn. */
2599 as_warn (_("translating to `%s %%%s'"), i.tm.name,
2600 i.op[0].regs->reg_name);
2604 else if (i.tm.opcode_modifier & Modrm)
2606 /* The opcode is completed (modulo i.tm.extension_opcode which
2607 must be put into the modrm byte).
2608 Now, we make the modrm & index base bytes based on all the
2609 info we've collected. */
2611 default_seg = build_modrm_byte ();
2613 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
2615 if (i.tm.base_opcode == POP_SEG_SHORT
2616 && i.op[0].regs->reg_num == 1)
2618 as_bad (_("you can't `pop %%cs'"));
2619 return 0;
2621 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
2622 if ((i.op[0].regs->reg_flags & RegRex) != 0)
2623 i.rex |= REX_EXTZ;
2625 else if ((i.tm.base_opcode & ~(D | W)) == MOV_AX_DISP32)
2627 default_seg = &ds;
2629 else if ((i.tm.opcode_modifier & IsString) != 0)
2631 /* For the string instructions that allow a segment override
2632 on one of their operands, the default segment is ds. */
2633 default_seg = &ds;
2636 /* If a segment was explicitly specified,
2637 and the specified segment is not the default,
2638 use an opcode prefix to select it.
2639 If we never figured out what the default segment is,
2640 then default_seg will be zero at this point,
2641 and the specified segment prefix will always be used. */
2642 if ((i.seg[0]) && (i.seg[0] != default_seg))
2644 if (!add_prefix (i.seg[0]->seg_prefix))
2645 return 0;
2647 return 1;
2650 static const seg_entry *
2651 build_modrm_byte ()
2653 const seg_entry *default_seg = 0;
2655 /* i.reg_operands MUST be the number of real register operands;
2656 implicit registers do not count. */
2657 if (i.reg_operands == 2)
2659 unsigned int source, dest;
2660 source = ((i.types[0]
2661 & (Reg | RegMMX | RegXMM
2662 | SReg2 | SReg3
2663 | Control | Debug | Test))
2664 ? 0 : 1);
2665 dest = source + 1;
2667 i.rm.mode = 3;
2668 /* One of the register operands will be encoded in the i.tm.reg
2669 field, the other in the combined i.tm.mode and i.tm.regmem
2670 fields. If no form of this instruction supports a memory
2671 destination operand, then we assume the source operand may
2672 sometimes be a memory operand and so we need to store the
2673 destination in the i.rm.reg field. */
2674 if ((i.tm.operand_types[dest] & AnyMem) == 0)
2676 i.rm.reg = i.op[dest].regs->reg_num;
2677 i.rm.regmem = i.op[source].regs->reg_num;
2678 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
2679 i.rex |= REX_EXTX;
2680 if ((i.op[source].regs->reg_flags & RegRex) != 0)
2681 i.rex |= REX_EXTZ;
2683 else
2685 i.rm.reg = i.op[source].regs->reg_num;
2686 i.rm.regmem = i.op[dest].regs->reg_num;
2687 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
2688 i.rex |= REX_EXTZ;
2689 if ((i.op[source].regs->reg_flags & RegRex) != 0)
2690 i.rex |= REX_EXTX;
2693 else
2694 { /* If it's not 2 reg operands... */
2695 if (i.mem_operands)
2697 unsigned int fake_zero_displacement = 0;
2698 unsigned int op = ((i.types[0] & AnyMem)
2700 : (i.types[1] & AnyMem) ? 1 : 2);
2702 default_seg = &ds;
2704 if (i.base_reg == 0)
2706 i.rm.mode = 0;
2707 if (!i.disp_operands)
2708 fake_zero_displacement = 1;
2709 if (i.index_reg == 0)
2711 /* Operand is just <disp> */
2712 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)
2713 && (flag_code != CODE_64BIT))
2715 i.rm.regmem = NO_BASE_REGISTER_16;
2716 i.types[op] &= ~Disp;
2717 i.types[op] |= Disp16;
2719 else if (flag_code != CODE_64BIT
2720 || (i.prefix[ADDR_PREFIX] != 0))
2722 i.rm.regmem = NO_BASE_REGISTER;
2723 i.types[op] &= ~Disp;
2724 i.types[op] |= Disp32;
2726 else
2728 /* 64bit mode overwrites the 32bit absolute
2729 addressing by RIP relative addressing and
2730 absolute addressing is encoded by one of the
2731 redundant SIB forms. */
2732 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2733 i.sib.base = NO_BASE_REGISTER;
2734 i.sib.index = NO_INDEX_REGISTER;
2735 i.types[op] &= ~Disp;
2736 i.types[op] |= Disp32S;
2739 else /* !i.base_reg && i.index_reg */
2741 i.sib.index = i.index_reg->reg_num;
2742 i.sib.base = NO_BASE_REGISTER;
2743 i.sib.scale = i.log2_scale_factor;
2744 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2745 i.types[op] &= ~Disp;
2746 if (flag_code != CODE_64BIT)
2747 i.types[op] |= Disp32; /* Must be 32 bit */
2748 else
2749 i.types[op] |= Disp32S;
2750 if ((i.index_reg->reg_flags & RegRex) != 0)
2751 i.rex |= REX_EXTY;
2754 /* RIP addressing for 64bit mode. */
2755 else if (i.base_reg->reg_type == BaseIndex)
2757 i.rm.regmem = NO_BASE_REGISTER;
2758 i.types[op] &= ~Disp;
2759 i.types[op] |= Disp32S;
2760 i.flags[op] = Operand_PCrel;
2762 else if (i.base_reg->reg_type & Reg16)
2764 switch (i.base_reg->reg_num)
2766 case 3: /* (%bx) */
2767 if (i.index_reg == 0)
2768 i.rm.regmem = 7;
2769 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
2770 i.rm.regmem = i.index_reg->reg_num - 6;
2771 break;
2772 case 5: /* (%bp) */
2773 default_seg = &ss;
2774 if (i.index_reg == 0)
2776 i.rm.regmem = 6;
2777 if ((i.types[op] & Disp) == 0)
2779 /* fake (%bp) into 0(%bp) */
2780 i.types[op] |= Disp8;
2781 fake_zero_displacement = 1;
2784 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
2785 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
2786 break;
2787 default: /* (%si) -> 4 or (%di) -> 5 */
2788 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
2790 i.rm.mode = mode_from_disp_size (i.types[op]);
2792 else /* i.base_reg and 32/64 bit mode */
2794 if (flag_code == CODE_64BIT
2795 && (i.types[op] & Disp))
2797 if (i.types[op] & Disp8)
2798 i.types[op] = Disp8 | Disp32S;
2799 else
2800 i.types[op] = Disp32S;
2802 i.rm.regmem = i.base_reg->reg_num;
2803 if ((i.base_reg->reg_flags & RegRex) != 0)
2804 i.rex |= REX_EXTZ;
2805 i.sib.base = i.base_reg->reg_num;
2806 /* x86-64 ignores REX prefix bit here to avoid decoder
2807 complications. */
2808 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
2810 default_seg = &ss;
2811 if (i.disp_operands == 0)
2813 fake_zero_displacement = 1;
2814 i.types[op] |= Disp8;
2817 else if (i.base_reg->reg_num == ESP_REG_NUM)
2819 default_seg = &ss;
2821 i.sib.scale = i.log2_scale_factor;
2822 if (i.index_reg == 0)
2824 /* <disp>(%esp) becomes two byte modrm with no index
2825 register. We've already stored the code for esp
2826 in i.rm.regmem ie. ESCAPE_TO_TWO_BYTE_ADDRESSING.
2827 Any base register besides %esp will not use the
2828 extra modrm byte. */
2829 i.sib.index = NO_INDEX_REGISTER;
2830 #if !SCALE1_WHEN_NO_INDEX
2831 /* Another case where we force the second modrm byte. */
2832 if (i.log2_scale_factor)
2833 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2834 #endif
2836 else
2838 i.sib.index = i.index_reg->reg_num;
2839 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2840 if ((i.index_reg->reg_flags & RegRex) != 0)
2841 i.rex |= REX_EXTY;
2843 i.rm.mode = mode_from_disp_size (i.types[op]);
2846 if (fake_zero_displacement)
2848 /* Fakes a zero displacement assuming that i.types[op]
2849 holds the correct displacement size. */
2850 expressionS *exp;
2852 assert (i.op[op].disps == 0);
2853 exp = &disp_expressions[i.disp_operands++];
2854 i.op[op].disps = exp;
2855 exp->X_op = O_constant;
2856 exp->X_add_number = 0;
2857 exp->X_add_symbol = (symbolS *) 0;
2858 exp->X_op_symbol = (symbolS *) 0;
2862 /* Fill in i.rm.reg or i.rm.regmem field with register operand
2863 (if any) based on i.tm.extension_opcode. Again, we must be
2864 careful to make sure that segment/control/debug/test/MMX
2865 registers are coded into the i.rm.reg field. */
2866 if (i.reg_operands)
2868 unsigned int op =
2869 ((i.types[0]
2870 & (Reg | RegMMX | RegXMM
2871 | SReg2 | SReg3
2872 | Control | Debug | Test))
2874 : ((i.types[1]
2875 & (Reg | RegMMX | RegXMM
2876 | SReg2 | SReg3
2877 | Control | Debug | Test))
2879 : 2));
2880 /* If there is an extension opcode to put here, the register
2881 number must be put into the regmem field. */
2882 if (i.tm.extension_opcode != None)
2884 i.rm.regmem = i.op[op].regs->reg_num;
2885 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2886 i.rex |= REX_EXTZ;
2888 else
2890 i.rm.reg = i.op[op].regs->reg_num;
2891 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2892 i.rex |= REX_EXTX;
2895 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2 we
2896 must set it to 3 to indicate this is a register operand
2897 in the regmem field. */
2898 if (!i.mem_operands)
2899 i.rm.mode = 3;
2902 /* Fill in i.rm.reg field with extension opcode (if any). */
2903 if (i.tm.extension_opcode != None)
2904 i.rm.reg = i.tm.extension_opcode;
2906 return default_seg;
2909 static void
2910 output_branch ()
2912 char *p;
2913 int code16;
2914 int prefix;
2915 relax_substateT subtype;
2916 symbolS *sym;
2917 offsetT off;
2919 code16 = 0;
2920 if (flag_code == CODE_16BIT)
2921 code16 = CODE16;
2923 prefix = 0;
2924 if (i.prefix[DATA_PREFIX] != 0)
2926 prefix = 1;
2927 i.prefixes -= 1;
2928 code16 ^= CODE16;
2930 /* Pentium4 branch hints. */
2931 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
2932 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
2934 prefix++;
2935 i.prefixes--;
2937 if (i.prefix[REX_PREFIX] != 0)
2939 prefix++;
2940 i.prefixes--;
2943 if (i.prefixes != 0 && !intel_syntax)
2944 as_warn (_("skipping prefixes on this instruction"));
2946 /* It's always a symbol; End frag & setup for relax.
2947 Make sure there is enough room in this frag for the largest
2948 instruction we may generate in md_convert_frag. This is 2
2949 bytes for the opcode and room for the prefix and largest
2950 displacement. */
2951 frag_grow (prefix + 2 + 4);
2952 /* Prefix and 1 opcode byte go in fr_fix. */
2953 p = frag_more (prefix + 1);
2954 if (i.prefix[DATA_PREFIX] != 0)
2955 *p++ = DATA_PREFIX_OPCODE;
2956 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
2957 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
2958 *p++ = i.prefix[SEG_PREFIX];
2959 if (i.prefix[REX_PREFIX] != 0)
2960 *p++ = i.prefix[REX_PREFIX];
2961 *p = i.tm.base_opcode;
2963 if ((unsigned char) *p == JUMP_PC_RELATIVE)
2964 subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
2965 else if ((cpu_arch_flags & Cpu386) != 0)
2966 subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
2967 else
2968 subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
2969 subtype |= code16;
2971 sym = i.op[0].disps->X_add_symbol;
2972 off = i.op[0].disps->X_add_number;
2974 if (i.op[0].disps->X_op != O_constant
2975 && i.op[0].disps->X_op != O_symbol)
2977 /* Handle complex expressions. */
2978 sym = make_expr_symbol (i.op[0].disps);
2979 off = 0;
2982 /* 1 possible extra opcode + 4 byte displacement go in var part.
2983 Pass reloc in fr_var. */
2984 frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
2987 static void
2988 output_jump ()
2990 char *p;
2991 int size;
2993 if (i.tm.opcode_modifier & JumpByte)
2995 /* This is a loop or jecxz type instruction. */
2996 size = 1;
2997 if (i.prefix[ADDR_PREFIX] != 0)
2999 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
3000 i.prefixes -= 1;
3002 /* Pentium4 branch hints. */
3003 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3004 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3006 FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
3007 i.prefixes--;
3010 else
3012 int code16;
3014 code16 = 0;
3015 if (flag_code == CODE_16BIT)
3016 code16 = CODE16;
3018 if (i.prefix[DATA_PREFIX] != 0)
3020 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
3021 i.prefixes -= 1;
3022 code16 ^= CODE16;
3025 size = 4;
3026 if (code16)
3027 size = 2;
3030 if (i.prefix[REX_PREFIX] != 0)
3032 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
3033 i.prefixes -= 1;
3036 if (i.prefixes != 0 && !intel_syntax)
3037 as_warn (_("skipping prefixes on this instruction"));
3039 p = frag_more (1 + size);
3040 *p++ = i.tm.base_opcode;
3042 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3043 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
3046 static void
3047 output_interseg_jump ()
3049 char *p;
3050 int size;
3051 int prefix;
3052 int code16;
3054 code16 = 0;
3055 if (flag_code == CODE_16BIT)
3056 code16 = CODE16;
3058 prefix = 0;
3059 if (i.prefix[DATA_PREFIX] != 0)
3061 prefix = 1;
3062 i.prefixes -= 1;
3063 code16 ^= CODE16;
3065 if (i.prefix[REX_PREFIX] != 0)
3067 prefix++;
3068 i.prefixes -= 1;
3071 size = 4;
3072 if (code16)
3073 size = 2;
3075 if (i.prefixes != 0 && !intel_syntax)
3076 as_warn (_("skipping prefixes on this instruction"));
3078 /* 1 opcode; 2 segment; offset */
3079 p = frag_more (prefix + 1 + 2 + size);
3081 if (i.prefix[DATA_PREFIX] != 0)
3082 *p++ = DATA_PREFIX_OPCODE;
3084 if (i.prefix[REX_PREFIX] != 0)
3085 *p++ = i.prefix[REX_PREFIX];
3087 *p++ = i.tm.base_opcode;
3088 if (i.op[1].imms->X_op == O_constant)
3090 offsetT n = i.op[1].imms->X_add_number;
3092 if (size == 2
3093 && !fits_in_unsigned_word (n)
3094 && !fits_in_signed_word (n))
3096 as_bad (_("16-bit jump out of range"));
3097 return;
3099 md_number_to_chars (p, n, size);
3101 else
3102 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3103 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
3104 if (i.op[0].imms->X_op != O_constant)
3105 as_bad (_("can't handle non absolute segment in `%s'"),
3106 i.tm.name);
3107 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
3111 static void
3112 output_insn ()
3114 fragS *insn_start_frag;
3115 offsetT insn_start_off;
3117 /* Tie dwarf2 debug info to the address at the start of the insn.
3118 We can't do this after the insn has been output as the current
3119 frag may have been closed off. eg. by frag_var. */
3120 dwarf2_emit_insn (0);
3122 insn_start_frag = frag_now;
3123 insn_start_off = frag_now_fix ();
3125 /* Output jumps. */
3126 if (i.tm.opcode_modifier & Jump)
3127 output_branch ();
3128 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
3129 output_jump ();
3130 else if (i.tm.opcode_modifier & JumpInterSegment)
3131 output_interseg_jump ();
3132 else
3134 /* Output normal instructions here. */
3135 char *p;
3136 unsigned char *q;
3138 /* All opcodes on i386 have either 1 or 2 bytes. We may use third
3139 byte for the SSE instructions to specify a prefix they require. */
3140 if (i.tm.base_opcode & 0xff0000)
3141 add_prefix ((i.tm.base_opcode >> 16) & 0xff);
3143 /* The prefix bytes. */
3144 for (q = i.prefix;
3145 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
3146 q++)
3148 if (*q)
3150 p = frag_more (1);
3151 md_number_to_chars (p, (valueT) *q, 1);
3155 /* Now the opcode; be careful about word order here! */
3156 if (fits_in_unsigned_byte (i.tm.base_opcode))
3158 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
3160 else
3162 p = frag_more (2);
3163 /* Put out high byte first: can't use md_number_to_chars! */
3164 *p++ = (i.tm.base_opcode >> 8) & 0xff;
3165 *p = i.tm.base_opcode & 0xff;
3168 /* Now the modrm byte and sib byte (if present). */
3169 if (i.tm.opcode_modifier & Modrm)
3171 p = frag_more (1);
3172 md_number_to_chars (p,
3173 (valueT) (i.rm.regmem << 0
3174 | i.rm.reg << 3
3175 | i.rm.mode << 6),
3177 /* If i.rm.regmem == ESP (4)
3178 && i.rm.mode != (Register mode)
3179 && not 16 bit
3180 ==> need second modrm byte. */
3181 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
3182 && i.rm.mode != 3
3183 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
3185 p = frag_more (1);
3186 md_number_to_chars (p,
3187 (valueT) (i.sib.base << 0
3188 | i.sib.index << 3
3189 | i.sib.scale << 6),
3194 if (i.disp_operands)
3195 output_disp (insn_start_frag, insn_start_off);
3197 if (i.imm_operands)
3198 output_imm (insn_start_frag, insn_start_off);
3201 #ifdef DEBUG386
3202 if (flag_debug)
3204 pi (line, &i);
3206 #endif /* DEBUG386 */
3209 static void
3210 output_disp (insn_start_frag, insn_start_off)
3211 fragS *insn_start_frag;
3212 offsetT insn_start_off;
3214 char *p;
3215 unsigned int n;
3217 for (n = 0; n < i.operands; n++)
3219 if (i.types[n] & Disp)
3221 if (i.op[n].disps->X_op == O_constant)
3223 int size;
3224 offsetT val;
3226 size = 4;
3227 if (i.types[n] & (Disp8 | Disp16 | Disp64))
3229 size = 2;
3230 if (i.types[n] & Disp8)
3231 size = 1;
3232 if (i.types[n] & Disp64)
3233 size = 8;
3235 val = offset_in_range (i.op[n].disps->X_add_number,
3236 size);
3237 p = frag_more (size);
3238 md_number_to_chars (p, val, size);
3240 else
3242 RELOC_ENUM reloc_type;
3243 int size = 4;
3244 int sign = 0;
3245 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
3247 /* The PC relative address is computed relative
3248 to the instruction boundary, so in case immediate
3249 fields follows, we need to adjust the value. */
3250 if (pcrel && i.imm_operands)
3252 int imm_size = 4;
3253 unsigned int n1;
3255 for (n1 = 0; n1 < i.operands; n1++)
3256 if (i.types[n1] & Imm)
3258 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
3260 imm_size = 2;
3261 if (i.types[n1] & (Imm8 | Imm8S))
3262 imm_size = 1;
3263 if (i.types[n1] & Imm64)
3264 imm_size = 8;
3266 break;
3268 /* We should find the immediate. */
3269 if (n1 == i.operands)
3270 abort ();
3271 i.op[n].disps->X_add_number -= imm_size;
3274 if (i.types[n] & Disp32S)
3275 sign = 1;
3277 if (i.types[n] & (Disp16 | Disp64))
3279 size = 2;
3280 if (i.types[n] & Disp64)
3281 size = 8;
3284 p = frag_more (size);
3285 reloc_type = reloc (size, pcrel, sign, i.reloc[n]);
3286 #ifdef BFD_ASSEMBLER
3287 if (reloc_type == BFD_RELOC_32
3288 && GOT_symbol
3289 && GOT_symbol == i.op[n].disps->X_add_symbol
3290 && (i.op[n].disps->X_op == O_symbol
3291 || (i.op[n].disps->X_op == O_add
3292 && ((symbol_get_value_expression
3293 (i.op[n].disps->X_op_symbol)->X_op)
3294 == O_subtract))))
3296 offsetT add;
3298 if (insn_start_frag == frag_now)
3299 add = (p - frag_now->fr_literal) - insn_start_off;
3300 else
3302 fragS *fr;
3304 add = insn_start_frag->fr_fix - insn_start_off;
3305 for (fr = insn_start_frag->fr_next;
3306 fr && fr != frag_now; fr = fr->fr_next)
3307 add += fr->fr_fix;
3308 add += p - frag_now->fr_literal;
3311 /* We don't support dynamic linking on x86-64 yet. */
3312 if (flag_code == CODE_64BIT)
3313 abort ();
3314 reloc_type = BFD_RELOC_386_GOTPC;
3315 i.op[n].disps->X_add_number += add;
3317 #endif
3318 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3319 i.op[n].disps, pcrel, reloc_type);
3325 static void
3326 output_imm (insn_start_frag, insn_start_off)
3327 fragS *insn_start_frag;
3328 offsetT insn_start_off;
3330 char *p;
3331 unsigned int n;
3333 for (n = 0; n < i.operands; n++)
3335 if (i.types[n] & Imm)
3337 if (i.op[n].imms->X_op == O_constant)
3339 int size;
3340 offsetT val;
3342 size = 4;
3343 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3345 size = 2;
3346 if (i.types[n] & (Imm8 | Imm8S))
3347 size = 1;
3348 else if (i.types[n] & Imm64)
3349 size = 8;
3351 val = offset_in_range (i.op[n].imms->X_add_number,
3352 size);
3353 p = frag_more (size);
3354 md_number_to_chars (p, val, size);
3356 else
3358 /* Not absolute_section.
3359 Need a 32-bit fixup (don't support 8bit
3360 non-absolute imms). Try to support other
3361 sizes ... */
3362 RELOC_ENUM reloc_type;
3363 int size = 4;
3364 int sign = 0;
3366 if ((i.types[n] & (Imm32S))
3367 && i.suffix == QWORD_MNEM_SUFFIX)
3368 sign = 1;
3369 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3371 size = 2;
3372 if (i.types[n] & (Imm8 | Imm8S))
3373 size = 1;
3374 if (i.types[n] & Imm64)
3375 size = 8;
3378 p = frag_more (size);
3379 reloc_type = reloc (size, 0, sign, i.reloc[n]);
3380 #ifdef BFD_ASSEMBLER
3381 /* This is tough to explain. We end up with this one if we
3382 * have operands that look like
3383 * "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal here is to
3384 * obtain the absolute address of the GOT, and it is strongly
3385 * preferable from a performance point of view to avoid using
3386 * a runtime relocation for this. The actual sequence of
3387 * instructions often look something like:
3389 * call .L66
3390 * .L66:
3391 * popl %ebx
3392 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
3394 * The call and pop essentially return the absolute address
3395 * of the label .L66 and store it in %ebx. The linker itself
3396 * will ultimately change the first operand of the addl so
3397 * that %ebx points to the GOT, but to keep things simple, the
3398 * .o file must have this operand set so that it generates not
3399 * the absolute address of .L66, but the absolute address of
3400 * itself. This allows the linker itself simply treat a GOTPC
3401 * relocation as asking for a pcrel offset to the GOT to be
3402 * added in, and the addend of the relocation is stored in the
3403 * operand field for the instruction itself.
3405 * Our job here is to fix the operand so that it would add
3406 * the correct offset so that %ebx would point to itself. The
3407 * thing that is tricky is that .-.L66 will point to the
3408 * beginning of the instruction, so we need to further modify
3409 * the operand so that it will point to itself. There are
3410 * other cases where you have something like:
3412 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
3414 * and here no correction would be required. Internally in
3415 * the assembler we treat operands of this form as not being
3416 * pcrel since the '.' is explicitly mentioned, and I wonder
3417 * whether it would simplify matters to do it this way. Who
3418 * knows. In earlier versions of the PIC patches, the
3419 * pcrel_adjust field was used to store the correction, but
3420 * since the expression is not pcrel, I felt it would be
3421 * confusing to do it this way. */
3423 if (reloc_type == BFD_RELOC_32
3424 && GOT_symbol
3425 && GOT_symbol == i.op[n].imms->X_add_symbol
3426 && (i.op[n].imms->X_op == O_symbol
3427 || (i.op[n].imms->X_op == O_add
3428 && ((symbol_get_value_expression
3429 (i.op[n].imms->X_op_symbol)->X_op)
3430 == O_subtract))))
3432 offsetT add;
3434 if (insn_start_frag == frag_now)
3435 add = (p - frag_now->fr_literal) - insn_start_off;
3436 else
3438 fragS *fr;
3440 add = insn_start_frag->fr_fix - insn_start_off;
3441 for (fr = insn_start_frag->fr_next;
3442 fr && fr != frag_now; fr = fr->fr_next)
3443 add += fr->fr_fix;
3444 add += p - frag_now->fr_literal;
3447 /* We don't support dynamic linking on x86-64 yet. */
3448 if (flag_code == CODE_64BIT)
3449 abort ();
3450 reloc_type = BFD_RELOC_386_GOTPC;
3451 i.op[n].imms->X_add_number += add;
3453 #endif
3454 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3455 i.op[n].imms, 0, reloc_type);
3461 #ifndef LEX_AT
3462 static char *lex_got PARAMS ((RELOC_ENUM *, int *));
3464 /* Parse operands of the form
3465 <symbol>@GOTOFF+<nnn>
3466 and similar .plt or .got references.
3468 If we find one, set up the correct relocation in RELOC and copy the
3469 input string, minus the `@GOTOFF' into a malloc'd buffer for
3470 parsing by the calling routine. Return this buffer, and if ADJUST
3471 is non-null set it to the length of the string we removed from the
3472 input line. Otherwise return NULL. */
3473 static char *
3474 lex_got (reloc, adjust)
3475 RELOC_ENUM *reloc;
3476 int *adjust;
3478 static const char * const mode_name[NUM_FLAG_CODE] = { "32", "16", "64" };
3479 static const struct {
3480 const char *str;
3481 const RELOC_ENUM rel[NUM_FLAG_CODE];
3482 } gotrel[] = {
3483 { "PLT", { BFD_RELOC_386_PLT32, 0, BFD_RELOC_X86_64_PLT32 } },
3484 { "GOTOFF", { BFD_RELOC_386_GOTOFF, 0, 0 } },
3485 { "GOTPCREL", { 0, 0, BFD_RELOC_X86_64_GOTPCREL } },
3486 { "TLSGD", { BFD_RELOC_386_TLS_GD, 0, 0 } },
3487 { "TLSLDM", { BFD_RELOC_386_TLS_LDM, 0, 0 } },
3488 { "GOTTPOFF", { BFD_RELOC_386_TLS_IE_32, 0, 0 } },
3489 { "TPOFF", { BFD_RELOC_386_TLS_LE_32, 0, 0 } },
3490 { "NTPOFF", { BFD_RELOC_386_TLS_LE, 0, 0 } },
3491 { "DTPOFF", { BFD_RELOC_386_TLS_LDO_32, 0, 0 } },
3492 { "GOTNTPOFF",{ BFD_RELOC_386_TLS_GOTIE, 0, 0 } },
3493 { "INDNTPOFF",{ BFD_RELOC_386_TLS_IE, 0, 0 } },
3494 { "GOT", { BFD_RELOC_386_GOT32, 0, BFD_RELOC_X86_64_GOT32 } }
3496 char *cp;
3497 unsigned int j;
3499 for (cp = input_line_pointer; *cp != '@'; cp++)
3500 if (is_end_of_line[(unsigned char) *cp])
3501 return NULL;
3503 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
3505 int len;
3507 len = strlen (gotrel[j].str);
3508 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
3510 if (gotrel[j].rel[(unsigned int) flag_code] != 0)
3512 int first, second;
3513 char *tmpbuf, *past_reloc;
3515 *reloc = gotrel[j].rel[(unsigned int) flag_code];
3516 if (adjust)
3517 *adjust = len;
3519 if (GOT_symbol == NULL)
3520 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3522 /* Replace the relocation token with ' ', so that
3523 errors like foo@GOTOFF1 will be detected. */
3525 /* The length of the first part of our input line. */
3526 first = cp - input_line_pointer;
3528 /* The second part goes from after the reloc token until
3529 (and including) an end_of_line char. Don't use strlen
3530 here as the end_of_line char may not be a NUL. */
3531 past_reloc = cp + 1 + len;
3532 for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
3534 second = cp - past_reloc;
3536 /* Allocate and copy string. The trailing NUL shouldn't
3537 be necessary, but be safe. */
3538 tmpbuf = xmalloc (first + second + 2);
3539 memcpy (tmpbuf, input_line_pointer, first);
3540 tmpbuf[first] = ' ';
3541 memcpy (tmpbuf + first + 1, past_reloc, second);
3542 tmpbuf[first + second + 1] = '\0';
3543 return tmpbuf;
3546 as_bad (_("@%s reloc is not supported in %s bit mode"),
3547 gotrel[j].str, mode_name[(unsigned int) flag_code]);
3548 return NULL;
3552 /* Might be a symbol version string. Don't as_bad here. */
3553 return NULL;
3556 /* x86_cons_fix_new is called via the expression parsing code when a
3557 reloc is needed. We use this hook to get the correct .got reloc. */
3558 static RELOC_ENUM got_reloc = NO_RELOC;
3560 void
3561 x86_cons_fix_new (frag, off, len, exp)
3562 fragS *frag;
3563 unsigned int off;
3564 unsigned int len;
3565 expressionS *exp;
3567 RELOC_ENUM r = reloc (len, 0, 0, got_reloc);
3568 got_reloc = NO_RELOC;
3569 fix_new_exp (frag, off, len, exp, 0, r);
3572 void
3573 x86_cons (exp, size)
3574 expressionS *exp;
3575 int size;
3577 if (size == 4)
3579 /* Handle @GOTOFF and the like in an expression. */
3580 char *save;
3581 char *gotfree_input_line;
3582 int adjust;
3584 save = input_line_pointer;
3585 gotfree_input_line = lex_got (&got_reloc, &adjust);
3586 if (gotfree_input_line)
3587 input_line_pointer = gotfree_input_line;
3589 expression (exp);
3591 if (gotfree_input_line)
3593 /* expression () has merrily parsed up to the end of line,
3594 or a comma - in the wrong buffer. Transfer how far
3595 input_line_pointer has moved to the right buffer. */
3596 input_line_pointer = (save
3597 + (input_line_pointer - gotfree_input_line)
3598 + adjust);
3599 free (gotfree_input_line);
3602 else
3603 expression (exp);
3605 #endif
3607 static int i386_immediate PARAMS ((char *));
3609 static int
3610 i386_immediate (imm_start)
3611 char *imm_start;
3613 char *save_input_line_pointer;
3614 #ifndef LEX_AT
3615 char *gotfree_input_line;
3616 #endif
3617 segT exp_seg = 0;
3618 expressionS *exp;
3620 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
3622 as_bad (_("only 1 or 2 immediate operands are allowed"));
3623 return 0;
3626 exp = &im_expressions[i.imm_operands++];
3627 i.op[this_operand].imms = exp;
3629 if (is_space_char (*imm_start))
3630 ++imm_start;
3632 save_input_line_pointer = input_line_pointer;
3633 input_line_pointer = imm_start;
3635 #ifndef LEX_AT
3636 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3637 if (gotfree_input_line)
3638 input_line_pointer = gotfree_input_line;
3639 #endif
3641 exp_seg = expression (exp);
3643 SKIP_WHITESPACE ();
3644 if (*input_line_pointer)
3645 as_bad (_("junk `%s' after expression"), input_line_pointer);
3647 input_line_pointer = save_input_line_pointer;
3648 #ifndef LEX_AT
3649 if (gotfree_input_line)
3650 free (gotfree_input_line);
3651 #endif
3653 if (exp->X_op == O_absent || exp->X_op == O_big)
3655 /* Missing or bad expr becomes absolute 0. */
3656 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
3657 imm_start);
3658 exp->X_op = O_constant;
3659 exp->X_add_number = 0;
3660 exp->X_add_symbol = (symbolS *) 0;
3661 exp->X_op_symbol = (symbolS *) 0;
3663 else if (exp->X_op == O_constant)
3665 /* Size it properly later. */
3666 i.types[this_operand] |= Imm64;
3667 /* If BFD64, sign extend val. */
3668 if (!use_rela_relocations)
3669 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
3670 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
3672 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3673 else if (1
3674 #ifdef BFD_ASSEMBLER
3675 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3676 #endif
3677 && exp_seg != absolute_section
3678 && exp_seg != text_section
3679 && exp_seg != data_section
3680 && exp_seg != bss_section
3681 && exp_seg != undefined_section
3682 #ifdef BFD_ASSEMBLER
3683 && !bfd_is_com_section (exp_seg)
3684 #endif
3687 #ifdef BFD_ASSEMBLER
3688 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3689 #else
3690 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3691 #endif
3692 return 0;
3694 #endif
3695 else
3697 /* This is an address. The size of the address will be
3698 determined later, depending on destination register,
3699 suffix, or the default for the section. */
3700 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
3703 return 1;
3706 static char *i386_scale PARAMS ((char *));
3708 static char *
3709 i386_scale (scale)
3710 char *scale;
3712 offsetT val;
3713 char *save = input_line_pointer;
3715 input_line_pointer = scale;
3716 val = get_absolute_expression ();
3718 switch (val)
3720 case 0:
3721 case 1:
3722 i.log2_scale_factor = 0;
3723 break;
3724 case 2:
3725 i.log2_scale_factor = 1;
3726 break;
3727 case 4:
3728 i.log2_scale_factor = 2;
3729 break;
3730 case 8:
3731 i.log2_scale_factor = 3;
3732 break;
3733 default:
3734 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
3735 scale);
3736 input_line_pointer = save;
3737 return NULL;
3739 if (i.log2_scale_factor != 0 && i.index_reg == 0)
3741 as_warn (_("scale factor of %d without an index register"),
3742 1 << i.log2_scale_factor);
3743 #if SCALE1_WHEN_NO_INDEX
3744 i.log2_scale_factor = 0;
3745 #endif
3747 scale = input_line_pointer;
3748 input_line_pointer = save;
3749 return scale;
3752 static int i386_displacement PARAMS ((char *, char *));
3754 static int
3755 i386_displacement (disp_start, disp_end)
3756 char *disp_start;
3757 char *disp_end;
3759 expressionS *exp;
3760 segT exp_seg = 0;
3761 char *save_input_line_pointer;
3762 #ifndef LEX_AT
3763 char *gotfree_input_line;
3764 #endif
3765 int bigdisp = Disp32;
3767 if (flag_code == CODE_64BIT)
3769 if (i.prefix[ADDR_PREFIX] == 0)
3770 bigdisp = Disp64;
3772 else if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3773 bigdisp = Disp16;
3774 i.types[this_operand] |= bigdisp;
3776 exp = &disp_expressions[i.disp_operands];
3777 i.op[this_operand].disps = exp;
3778 i.disp_operands++;
3779 save_input_line_pointer = input_line_pointer;
3780 input_line_pointer = disp_start;
3781 END_STRING_AND_SAVE (disp_end);
3783 #ifndef GCC_ASM_O_HACK
3784 #define GCC_ASM_O_HACK 0
3785 #endif
3786 #if GCC_ASM_O_HACK
3787 END_STRING_AND_SAVE (disp_end + 1);
3788 if ((i.types[this_operand] & BaseIndex) != 0
3789 && displacement_string_end[-1] == '+')
3791 /* This hack is to avoid a warning when using the "o"
3792 constraint within gcc asm statements.
3793 For instance:
3795 #define _set_tssldt_desc(n,addr,limit,type) \
3796 __asm__ __volatile__ ( \
3797 "movw %w2,%0\n\t" \
3798 "movw %w1,2+%0\n\t" \
3799 "rorl $16,%1\n\t" \
3800 "movb %b1,4+%0\n\t" \
3801 "movb %4,5+%0\n\t" \
3802 "movb $0,6+%0\n\t" \
3803 "movb %h1,7+%0\n\t" \
3804 "rorl $16,%1" \
3805 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
3807 This works great except that the output assembler ends
3808 up looking a bit weird if it turns out that there is
3809 no offset. You end up producing code that looks like:
3811 #APP
3812 movw $235,(%eax)
3813 movw %dx,2+(%eax)
3814 rorl $16,%edx
3815 movb %dl,4+(%eax)
3816 movb $137,5+(%eax)
3817 movb $0,6+(%eax)
3818 movb %dh,7+(%eax)
3819 rorl $16,%edx
3820 #NO_APP
3822 So here we provide the missing zero. */
3824 *displacement_string_end = '0';
3826 #endif
3827 #ifndef LEX_AT
3828 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3829 if (gotfree_input_line)
3830 input_line_pointer = gotfree_input_line;
3831 #endif
3833 exp_seg = expression (exp);
3835 SKIP_WHITESPACE ();
3836 if (*input_line_pointer)
3837 as_bad (_("junk `%s' after expression"), input_line_pointer);
3838 #if GCC_ASM_O_HACK
3839 RESTORE_END_STRING (disp_end + 1);
3840 #endif
3841 RESTORE_END_STRING (disp_end);
3842 input_line_pointer = save_input_line_pointer;
3843 #ifndef LEX_AT
3844 if (gotfree_input_line)
3845 free (gotfree_input_line);
3846 #endif
3848 #ifdef BFD_ASSEMBLER
3849 /* We do this to make sure that the section symbol is in
3850 the symbol table. We will ultimately change the relocation
3851 to be relative to the beginning of the section. */
3852 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
3853 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3855 if (exp->X_op != O_symbol)
3857 as_bad (_("bad expression used with @%s"),
3858 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
3859 ? "GOTPCREL"
3860 : "GOTOFF"));
3861 return 0;
3864 if (S_IS_LOCAL (exp->X_add_symbol)
3865 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
3866 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
3867 exp->X_op = O_subtract;
3868 exp->X_op_symbol = GOT_symbol;
3869 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3870 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
3871 else
3872 i.reloc[this_operand] = BFD_RELOC_32;
3874 #endif
3876 if (exp->X_op == O_absent || exp->X_op == O_big)
3878 /* Missing or bad expr becomes absolute 0. */
3879 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
3880 disp_start);
3881 exp->X_op = O_constant;
3882 exp->X_add_number = 0;
3883 exp->X_add_symbol = (symbolS *) 0;
3884 exp->X_op_symbol = (symbolS *) 0;
3887 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3888 if (exp->X_op != O_constant
3889 #ifdef BFD_ASSEMBLER
3890 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3891 #endif
3892 && exp_seg != absolute_section
3893 && exp_seg != text_section
3894 && exp_seg != data_section
3895 && exp_seg != bss_section
3896 && exp_seg != undefined_section
3897 #ifdef BFD_ASSEMBLER
3898 && !bfd_is_com_section (exp_seg)
3899 #endif
3902 #ifdef BFD_ASSEMBLER
3903 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3904 #else
3905 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3906 #endif
3907 return 0;
3909 #endif
3910 else if (flag_code == CODE_64BIT)
3911 i.types[this_operand] |= Disp32S | Disp32;
3912 return 1;
3915 static int i386_index_check PARAMS ((const char *));
3917 /* Make sure the memory operand we've been dealt is valid.
3918 Return 1 on success, 0 on a failure. */
3920 static int
3921 i386_index_check (operand_string)
3922 const char *operand_string;
3924 int ok;
3925 #if INFER_ADDR_PREFIX
3926 int fudged = 0;
3928 tryprefix:
3929 #endif
3930 ok = 1;
3931 if (flag_code == CODE_64BIT)
3933 if (i.prefix[ADDR_PREFIX] == 0)
3935 /* 64bit checks. */
3936 if ((i.base_reg
3937 && ((i.base_reg->reg_type & Reg64) == 0)
3938 && (i.base_reg->reg_type != BaseIndex
3939 || i.index_reg))
3940 || (i.index_reg
3941 && ((i.index_reg->reg_type & (Reg64 | BaseIndex))
3942 != (Reg64 | BaseIndex))))
3943 ok = 0;
3945 else
3947 /* 32bit checks. */
3948 if ((i.base_reg
3949 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
3950 || (i.index_reg
3951 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
3952 != (Reg32 | BaseIndex))))
3953 ok = 0;
3956 else
3958 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3960 /* 16bit checks. */
3961 if ((i.base_reg
3962 && ((i.base_reg->reg_type & (Reg16 | BaseIndex | RegRex))
3963 != (Reg16 | BaseIndex)))
3964 || (i.index_reg
3965 && (((i.index_reg->reg_type & (Reg16 | BaseIndex))
3966 != (Reg16 | BaseIndex))
3967 || !(i.base_reg
3968 && i.base_reg->reg_num < 6
3969 && i.index_reg->reg_num >= 6
3970 && i.log2_scale_factor == 0))))
3971 ok = 0;
3973 else
3975 /* 32bit checks. */
3976 if ((i.base_reg
3977 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
3978 || (i.index_reg
3979 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
3980 != (Reg32 | BaseIndex))))
3981 ok = 0;
3984 if (!ok)
3986 #if INFER_ADDR_PREFIX
3987 if (flag_code != CODE_64BIT
3988 && i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0')
3990 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
3991 i.prefixes += 1;
3992 /* Change the size of any displacement too. At most one of
3993 Disp16 or Disp32 is set.
3994 FIXME. There doesn't seem to be any real need for separate
3995 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
3996 Removing them would probably clean up the code quite a lot. */
3997 if (i.types[this_operand] & (Disp16 | Disp32))
3998 i.types[this_operand] ^= (Disp16 | Disp32);
3999 fudged = 1;
4000 goto tryprefix;
4002 if (fudged)
4003 as_bad (_("`%s' is not a valid base/index expression"),
4004 operand_string);
4005 else
4006 #endif
4007 as_bad (_("`%s' is not a valid %s bit base/index expression"),
4008 operand_string,
4009 flag_code_names[flag_code]);
4010 return 0;
4012 return 1;
4015 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
4016 on error. */
4018 static int
4019 i386_operand (operand_string)
4020 char *operand_string;
4022 const reg_entry *r;
4023 char *end_op;
4024 char *op_string = operand_string;
4026 if (is_space_char (*op_string))
4027 ++op_string;
4029 /* We check for an absolute prefix (differentiating,
4030 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
4031 if (*op_string == ABSOLUTE_PREFIX)
4033 ++op_string;
4034 if (is_space_char (*op_string))
4035 ++op_string;
4036 i.types[this_operand] |= JumpAbsolute;
4039 /* Check if operand is a register. */
4040 if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
4041 && (r = parse_register (op_string, &end_op)) != NULL)
4043 /* Check for a segment override by searching for ':' after a
4044 segment register. */
4045 op_string = end_op;
4046 if (is_space_char (*op_string))
4047 ++op_string;
4048 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
4050 switch (r->reg_num)
4052 case 0:
4053 i.seg[i.mem_operands] = &es;
4054 break;
4055 case 1:
4056 i.seg[i.mem_operands] = &cs;
4057 break;
4058 case 2:
4059 i.seg[i.mem_operands] = &ss;
4060 break;
4061 case 3:
4062 i.seg[i.mem_operands] = &ds;
4063 break;
4064 case 4:
4065 i.seg[i.mem_operands] = &fs;
4066 break;
4067 case 5:
4068 i.seg[i.mem_operands] = &gs;
4069 break;
4072 /* Skip the ':' and whitespace. */
4073 ++op_string;
4074 if (is_space_char (*op_string))
4075 ++op_string;
4077 if (!is_digit_char (*op_string)
4078 && !is_identifier_char (*op_string)
4079 && *op_string != '('
4080 && *op_string != ABSOLUTE_PREFIX)
4082 as_bad (_("bad memory operand `%s'"), op_string);
4083 return 0;
4085 /* Handle case of %es:*foo. */
4086 if (*op_string == ABSOLUTE_PREFIX)
4088 ++op_string;
4089 if (is_space_char (*op_string))
4090 ++op_string;
4091 i.types[this_operand] |= JumpAbsolute;
4093 goto do_memory_reference;
4095 if (*op_string)
4097 as_bad (_("junk `%s' after register"), op_string);
4098 return 0;
4100 i.types[this_operand] |= r->reg_type & ~BaseIndex;
4101 i.op[this_operand].regs = r;
4102 i.reg_operands++;
4104 else if (*op_string == REGISTER_PREFIX)
4106 as_bad (_("bad register name `%s'"), op_string);
4107 return 0;
4109 else if (*op_string == IMMEDIATE_PREFIX)
4111 ++op_string;
4112 if (i.types[this_operand] & JumpAbsolute)
4114 as_bad (_("immediate operand illegal with absolute jump"));
4115 return 0;
4117 if (!i386_immediate (op_string))
4118 return 0;
4120 else if (is_digit_char (*op_string)
4121 || is_identifier_char (*op_string)
4122 || *op_string == '(')
4124 /* This is a memory reference of some sort. */
4125 char *base_string;
4127 /* Start and end of displacement string expression (if found). */
4128 char *displacement_string_start;
4129 char *displacement_string_end;
4131 do_memory_reference:
4132 if ((i.mem_operands == 1
4133 && (current_templates->start->opcode_modifier & IsString) == 0)
4134 || i.mem_operands == 2)
4136 as_bad (_("too many memory references for `%s'"),
4137 current_templates->start->name);
4138 return 0;
4141 /* Check for base index form. We detect the base index form by
4142 looking for an ')' at the end of the operand, searching
4143 for the '(' matching it, and finding a REGISTER_PREFIX or ','
4144 after the '('. */
4145 base_string = op_string + strlen (op_string);
4147 --base_string;
4148 if (is_space_char (*base_string))
4149 --base_string;
4151 /* If we only have a displacement, set-up for it to be parsed later. */
4152 displacement_string_start = op_string;
4153 displacement_string_end = base_string + 1;
4155 if (*base_string == ')')
4157 char *temp_string;
4158 unsigned int parens_balanced = 1;
4159 /* We've already checked that the number of left & right ()'s are
4160 equal, so this loop will not be infinite. */
4163 base_string--;
4164 if (*base_string == ')')
4165 parens_balanced++;
4166 if (*base_string == '(')
4167 parens_balanced--;
4169 while (parens_balanced);
4171 temp_string = base_string;
4173 /* Skip past '(' and whitespace. */
4174 ++base_string;
4175 if (is_space_char (*base_string))
4176 ++base_string;
4178 if (*base_string == ','
4179 || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
4180 && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
4182 displacement_string_end = temp_string;
4184 i.types[this_operand] |= BaseIndex;
4186 if (i.base_reg)
4188 base_string = end_op;
4189 if (is_space_char (*base_string))
4190 ++base_string;
4193 /* There may be an index reg or scale factor here. */
4194 if (*base_string == ',')
4196 ++base_string;
4197 if (is_space_char (*base_string))
4198 ++base_string;
4200 if ((*base_string == REGISTER_PREFIX || allow_naked_reg)
4201 && (i.index_reg = parse_register (base_string, &end_op)) != NULL)
4203 base_string = end_op;
4204 if (is_space_char (*base_string))
4205 ++base_string;
4206 if (*base_string == ',')
4208 ++base_string;
4209 if (is_space_char (*base_string))
4210 ++base_string;
4212 else if (*base_string != ')')
4214 as_bad (_("expecting `,' or `)' after index register in `%s'"),
4215 operand_string);
4216 return 0;
4219 else if (*base_string == REGISTER_PREFIX)
4221 as_bad (_("bad register name `%s'"), base_string);
4222 return 0;
4225 /* Check for scale factor. */
4226 if (*base_string != ')')
4228 char *end_scale = i386_scale (base_string);
4230 if (!end_scale)
4231 return 0;
4233 base_string = end_scale;
4234 if (is_space_char (*base_string))
4235 ++base_string;
4236 if (*base_string != ')')
4238 as_bad (_("expecting `)' after scale factor in `%s'"),
4239 operand_string);
4240 return 0;
4243 else if (!i.index_reg)
4245 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
4246 *base_string);
4247 return 0;
4250 else if (*base_string != ')')
4252 as_bad (_("expecting `,' or `)' after base register in `%s'"),
4253 operand_string);
4254 return 0;
4257 else if (*base_string == REGISTER_PREFIX)
4259 as_bad (_("bad register name `%s'"), base_string);
4260 return 0;
4264 /* If there's an expression beginning the operand, parse it,
4265 assuming displacement_string_start and
4266 displacement_string_end are meaningful. */
4267 if (displacement_string_start != displacement_string_end)
4269 if (!i386_displacement (displacement_string_start,
4270 displacement_string_end))
4271 return 0;
4274 /* Special case for (%dx) while doing input/output op. */
4275 if (i.base_reg
4276 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
4277 && i.index_reg == 0
4278 && i.log2_scale_factor == 0
4279 && i.seg[i.mem_operands] == 0
4280 && (i.types[this_operand] & Disp) == 0)
4282 i.types[this_operand] = InOutPortReg;
4283 return 1;
4286 if (i386_index_check (operand_string) == 0)
4287 return 0;
4288 i.mem_operands++;
4290 else
4292 /* It's not a memory operand; argh! */
4293 as_bad (_("invalid char %s beginning operand %d `%s'"),
4294 output_invalid (*op_string),
4295 this_operand + 1,
4296 op_string);
4297 return 0;
4299 return 1; /* Normal return. */
4302 /* md_estimate_size_before_relax()
4304 Called just before relax() for rs_machine_dependent frags. The x86
4305 assembler uses these frags to handle variable size jump
4306 instructions.
4308 Any symbol that is now undefined will not become defined.
4309 Return the correct fr_subtype in the frag.
4310 Return the initial "guess for variable size of frag" to caller.
4311 The guess is actually the growth beyond the fixed part. Whatever
4312 we do to grow the fixed or variable part contributes to our
4313 returned value. */
4316 md_estimate_size_before_relax (fragP, segment)
4317 fragS *fragP;
4318 segT segment;
4320 /* We've already got fragP->fr_subtype right; all we have to do is
4321 check for un-relaxable symbols. On an ELF system, we can't relax
4322 an externally visible symbol, because it may be overridden by a
4323 shared library. */
4324 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
4325 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4326 || (OUTPUT_FLAVOR == bfd_target_elf_flavour
4327 && (S_IS_EXTERNAL (fragP->fr_symbol)
4328 || S_IS_WEAK (fragP->fr_symbol)))
4329 #endif
4332 /* Symbol is undefined in this segment, or we need to keep a
4333 reloc so that weak symbols can be overridden. */
4334 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
4335 RELOC_ENUM reloc_type;
4336 unsigned char *opcode;
4337 int old_fr_fix;
4339 if (fragP->fr_var != NO_RELOC)
4340 reloc_type = fragP->fr_var;
4341 else if (size == 2)
4342 reloc_type = BFD_RELOC_16_PCREL;
4343 else
4344 reloc_type = BFD_RELOC_32_PCREL;
4346 old_fr_fix = fragP->fr_fix;
4347 opcode = (unsigned char *) fragP->fr_opcode;
4349 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
4351 case UNCOND_JUMP:
4352 /* Make jmp (0xeb) a (d)word displacement jump. */
4353 opcode[0] = 0xe9;
4354 fragP->fr_fix += size;
4355 fix_new (fragP, old_fr_fix, size,
4356 fragP->fr_symbol,
4357 fragP->fr_offset, 1,
4358 reloc_type);
4359 break;
4361 case COND_JUMP86:
4362 if (size == 2
4363 && (!no_cond_jump_promotion || fragP->fr_var != NO_RELOC))
4365 /* Negate the condition, and branch past an
4366 unconditional jump. */
4367 opcode[0] ^= 1;
4368 opcode[1] = 3;
4369 /* Insert an unconditional jump. */
4370 opcode[2] = 0xe9;
4371 /* We added two extra opcode bytes, and have a two byte
4372 offset. */
4373 fragP->fr_fix += 2 + 2;
4374 fix_new (fragP, old_fr_fix + 2, 2,
4375 fragP->fr_symbol,
4376 fragP->fr_offset, 1,
4377 reloc_type);
4378 break;
4380 /* Fall through. */
4382 case COND_JUMP:
4383 if (no_cond_jump_promotion && fragP->fr_var == NO_RELOC)
4385 fragP->fr_fix += 1;
4386 fix_new (fragP, old_fr_fix, 1,
4387 fragP->fr_symbol,
4388 fragP->fr_offset, 1,
4389 BFD_RELOC_8_PCREL);
4390 break;
4393 /* This changes the byte-displacement jump 0x7N
4394 to the (d)word-displacement jump 0x0f,0x8N. */
4395 opcode[1] = opcode[0] + 0x10;
4396 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4397 /* We've added an opcode byte. */
4398 fragP->fr_fix += 1 + size;
4399 fix_new (fragP, old_fr_fix + 1, size,
4400 fragP->fr_symbol,
4401 fragP->fr_offset, 1,
4402 reloc_type);
4403 break;
4405 default:
4406 BAD_CASE (fragP->fr_subtype);
4407 break;
4409 frag_wane (fragP);
4410 return fragP->fr_fix - old_fr_fix;
4413 /* Guess size depending on current relax state. Initially the relax
4414 state will correspond to a short jump and we return 1, because
4415 the variable part of the frag (the branch offset) is one byte
4416 long. However, we can relax a section more than once and in that
4417 case we must either set fr_subtype back to the unrelaxed state,
4418 or return the value for the appropriate branch. */
4419 return md_relax_table[fragP->fr_subtype].rlx_length;
4422 /* Called after relax() is finished.
4424 In: Address of frag.
4425 fr_type == rs_machine_dependent.
4426 fr_subtype is what the address relaxed to.
4428 Out: Any fixSs and constants are set up.
4429 Caller will turn frag into a ".space 0". */
4431 #ifndef BFD_ASSEMBLER
4432 void
4433 md_convert_frag (headers, sec, fragP)
4434 object_headers *headers ATTRIBUTE_UNUSED;
4435 segT sec ATTRIBUTE_UNUSED;
4436 fragS *fragP;
4437 #else
4438 void
4439 md_convert_frag (abfd, sec, fragP)
4440 bfd *abfd ATTRIBUTE_UNUSED;
4441 segT sec ATTRIBUTE_UNUSED;
4442 fragS *fragP;
4443 #endif
4445 unsigned char *opcode;
4446 unsigned char *where_to_put_displacement = NULL;
4447 offsetT target_address;
4448 offsetT opcode_address;
4449 unsigned int extension = 0;
4450 offsetT displacement_from_opcode_start;
4452 opcode = (unsigned char *) fragP->fr_opcode;
4454 /* Address we want to reach in file space. */
4455 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
4457 /* Address opcode resides at in file space. */
4458 opcode_address = fragP->fr_address + fragP->fr_fix;
4460 /* Displacement from opcode start to fill into instruction. */
4461 displacement_from_opcode_start = target_address - opcode_address;
4463 if ((fragP->fr_subtype & BIG) == 0)
4465 /* Don't have to change opcode. */
4466 extension = 1; /* 1 opcode + 1 displacement */
4467 where_to_put_displacement = &opcode[1];
4469 else
4471 if (no_cond_jump_promotion
4472 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
4473 as_warn_where (fragP->fr_file, fragP->fr_line, _("long jump required"));
4475 switch (fragP->fr_subtype)
4477 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
4478 extension = 4; /* 1 opcode + 4 displacement */
4479 opcode[0] = 0xe9;
4480 where_to_put_displacement = &opcode[1];
4481 break;
4483 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
4484 extension = 2; /* 1 opcode + 2 displacement */
4485 opcode[0] = 0xe9;
4486 where_to_put_displacement = &opcode[1];
4487 break;
4489 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
4490 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
4491 extension = 5; /* 2 opcode + 4 displacement */
4492 opcode[1] = opcode[0] + 0x10;
4493 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4494 where_to_put_displacement = &opcode[2];
4495 break;
4497 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
4498 extension = 3; /* 2 opcode + 2 displacement */
4499 opcode[1] = opcode[0] + 0x10;
4500 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4501 where_to_put_displacement = &opcode[2];
4502 break;
4504 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
4505 extension = 4;
4506 opcode[0] ^= 1;
4507 opcode[1] = 3;
4508 opcode[2] = 0xe9;
4509 where_to_put_displacement = &opcode[3];
4510 break;
4512 default:
4513 BAD_CASE (fragP->fr_subtype);
4514 break;
4518 /* Now put displacement after opcode. */
4519 md_number_to_chars ((char *) where_to_put_displacement,
4520 (valueT) (displacement_from_opcode_start - extension),
4521 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
4522 fragP->fr_fix += extension;
4525 /* Size of byte displacement jmp. */
4526 int md_short_jump_size = 2;
4528 /* Size of dword displacement jmp. */
4529 int md_long_jump_size = 5;
4531 /* Size of relocation record. */
4532 const int md_reloc_size = 8;
4534 void
4535 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
4536 char *ptr;
4537 addressT from_addr, to_addr;
4538 fragS *frag ATTRIBUTE_UNUSED;
4539 symbolS *to_symbol ATTRIBUTE_UNUSED;
4541 offsetT offset;
4543 offset = to_addr - (from_addr + 2);
4544 /* Opcode for byte-disp jump. */
4545 md_number_to_chars (ptr, (valueT) 0xeb, 1);
4546 md_number_to_chars (ptr + 1, (valueT) offset, 1);
4549 void
4550 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
4551 char *ptr;
4552 addressT from_addr, to_addr;
4553 fragS *frag ATTRIBUTE_UNUSED;
4554 symbolS *to_symbol ATTRIBUTE_UNUSED;
4556 offsetT offset;
4558 offset = to_addr - (from_addr + 5);
4559 md_number_to_chars (ptr, (valueT) 0xe9, 1);
4560 md_number_to_chars (ptr + 1, (valueT) offset, 4);
4563 /* Apply a fixup (fixS) to segment data, once it has been determined
4564 by our caller that we have all the info we need to fix it up.
4566 On the 386, immediates, displacements, and data pointers are all in
4567 the same (little-endian) format, so we don't need to care about which
4568 we are handling. */
4570 void
4571 md_apply_fix3 (fixP, valP, seg)
4572 /* The fix we're to put in. */
4573 fixS *fixP;
4574 /* Pointer to the value of the bits. */
4575 valueT *valP;
4576 /* Segment fix is from. */
4577 segT seg ATTRIBUTE_UNUSED;
4579 char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
4580 valueT value = *valP;
4582 #if defined (BFD_ASSEMBLER) && !defined (TE_Mach)
4583 if (fixP->fx_pcrel)
4585 switch (fixP->fx_r_type)
4587 default:
4588 break;
4590 case BFD_RELOC_32:
4591 fixP->fx_r_type = BFD_RELOC_32_PCREL;
4592 break;
4593 case BFD_RELOC_16:
4594 fixP->fx_r_type = BFD_RELOC_16_PCREL;
4595 break;
4596 case BFD_RELOC_8:
4597 fixP->fx_r_type = BFD_RELOC_8_PCREL;
4598 break;
4602 if (fixP->fx_addsy != NULL
4603 && (fixP->fx_r_type == BFD_RELOC_32_PCREL
4604 || fixP->fx_r_type == BFD_RELOC_16_PCREL
4605 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
4606 && !use_rela_relocations)
4608 /* This is a hack. There should be a better way to handle this.
4609 This covers for the fact that bfd_install_relocation will
4610 subtract the current location (for partial_inplace, PC relative
4611 relocations); see more below. */
4612 #ifndef OBJ_AOUT
4613 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4614 #ifdef TE_PE
4615 || OUTPUT_FLAVOR == bfd_target_coff_flavour
4616 #endif
4618 value += fixP->fx_where + fixP->fx_frag->fr_address;
4619 #endif
4620 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4621 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
4623 segT fseg = S_GET_SEGMENT (fixP->fx_addsy);
4625 if ((fseg == seg
4626 || (symbol_section_p (fixP->fx_addsy)
4627 && fseg != absolute_section))
4628 && !S_IS_EXTERNAL (fixP->fx_addsy)
4629 && !S_IS_WEAK (fixP->fx_addsy)
4630 && S_IS_DEFINED (fixP->fx_addsy)
4631 && !S_IS_COMMON (fixP->fx_addsy))
4633 /* Yes, we add the values in twice. This is because
4634 bfd_perform_relocation subtracts them out again. I think
4635 bfd_perform_relocation is broken, but I don't dare change
4636 it. FIXME. */
4637 value += fixP->fx_where + fixP->fx_frag->fr_address;
4640 #endif
4641 #if defined (OBJ_COFF) && defined (TE_PE)
4642 /* For some reason, the PE format does not store a section
4643 address offset for a PC relative symbol. */
4644 if (S_GET_SEGMENT (fixP->fx_addsy) != seg)
4645 value += md_pcrel_from (fixP);
4646 #endif
4649 /* Fix a few things - the dynamic linker expects certain values here,
4650 and we must not dissappoint it. */
4651 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4652 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4653 && fixP->fx_addsy)
4654 switch (fixP->fx_r_type)
4656 case BFD_RELOC_386_PLT32:
4657 case BFD_RELOC_X86_64_PLT32:
4658 /* Make the jump instruction point to the address of the operand. At
4659 runtime we merely add the offset to the actual PLT entry. */
4660 value = -4;
4661 break;
4663 case BFD_RELOC_386_GOT32:
4664 case BFD_RELOC_386_TLS_GD:
4665 case BFD_RELOC_386_TLS_LDM:
4666 case BFD_RELOC_386_TLS_IE_32:
4667 case BFD_RELOC_386_TLS_IE:
4668 case BFD_RELOC_386_TLS_GOTIE:
4669 case BFD_RELOC_X86_64_GOT32:
4670 value = 0; /* Fully resolved at runtime. No addend. */
4671 break;
4673 case BFD_RELOC_VTABLE_INHERIT:
4674 case BFD_RELOC_VTABLE_ENTRY:
4675 fixP->fx_done = 0;
4676 return;
4678 default:
4679 break;
4681 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
4682 *valP = value;
4683 #endif /* defined (BFD_ASSEMBLER) && !defined (TE_Mach) */
4685 /* Are we finished with this relocation now? */
4686 if (fixP->fx_addsy == NULL)
4687 fixP->fx_done = 1;
4688 #ifdef BFD_ASSEMBLER
4689 else if (use_rela_relocations)
4691 fixP->fx_no_overflow = 1;
4692 /* Remember value for tc_gen_reloc. */
4693 fixP->fx_addnumber = value;
4694 value = 0;
4696 #endif
4697 md_number_to_chars (p, value, fixP->fx_size);
4700 #define MAX_LITTLENUMS 6
4702 /* Turn the string pointed to by litP into a floating point constant
4703 of type TYPE, and emit the appropriate bytes. The number of
4704 LITTLENUMS emitted is stored in *SIZEP. An error message is
4705 returned, or NULL on OK. */
4707 char *
4708 md_atof (type, litP, sizeP)
4709 int type;
4710 char *litP;
4711 int *sizeP;
4713 int prec;
4714 LITTLENUM_TYPE words[MAX_LITTLENUMS];
4715 LITTLENUM_TYPE *wordP;
4716 char *t;
4718 switch (type)
4720 case 'f':
4721 case 'F':
4722 prec = 2;
4723 break;
4725 case 'd':
4726 case 'D':
4727 prec = 4;
4728 break;
4730 case 'x':
4731 case 'X':
4732 prec = 5;
4733 break;
4735 default:
4736 *sizeP = 0;
4737 return _("Bad call to md_atof ()");
4739 t = atof_ieee (input_line_pointer, type, words);
4740 if (t)
4741 input_line_pointer = t;
4743 *sizeP = prec * sizeof (LITTLENUM_TYPE);
4744 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
4745 the bigendian 386. */
4746 for (wordP = words + prec - 1; prec--;)
4748 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
4749 litP += sizeof (LITTLENUM_TYPE);
4751 return 0;
4754 char output_invalid_buf[8];
4756 static char *
4757 output_invalid (c)
4758 int c;
4760 if (ISPRINT (c))
4761 sprintf (output_invalid_buf, "'%c'", c);
4762 else
4763 sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
4764 return output_invalid_buf;
4767 /* REG_STRING starts *before* REGISTER_PREFIX. */
4769 static const reg_entry *
4770 parse_register (reg_string, end_op)
4771 char *reg_string;
4772 char **end_op;
4774 char *s = reg_string;
4775 char *p;
4776 char reg_name_given[MAX_REG_NAME_SIZE + 1];
4777 const reg_entry *r;
4779 /* Skip possible REGISTER_PREFIX and possible whitespace. */
4780 if (*s == REGISTER_PREFIX)
4781 ++s;
4783 if (is_space_char (*s))
4784 ++s;
4786 p = reg_name_given;
4787 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
4789 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
4790 return (const reg_entry *) NULL;
4791 s++;
4794 /* For naked regs, make sure that we are not dealing with an identifier.
4795 This prevents confusing an identifier like `eax_var' with register
4796 `eax'. */
4797 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
4798 return (const reg_entry *) NULL;
4800 *end_op = s;
4802 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
4804 /* Handle floating point regs, allowing spaces in the (i) part. */
4805 if (r == i386_regtab /* %st is first entry of table */)
4807 if (is_space_char (*s))
4808 ++s;
4809 if (*s == '(')
4811 ++s;
4812 if (is_space_char (*s))
4813 ++s;
4814 if (*s >= '0' && *s <= '7')
4816 r = &i386_float_regtab[*s - '0'];
4817 ++s;
4818 if (is_space_char (*s))
4819 ++s;
4820 if (*s == ')')
4822 *end_op = s + 1;
4823 return r;
4826 /* We have "%st(" then garbage. */
4827 return (const reg_entry *) NULL;
4831 if (r != NULL
4832 && (r->reg_flags & (RegRex64 | RegRex)) != 0
4833 && flag_code != CODE_64BIT)
4835 return (const reg_entry *) NULL;
4838 return r;
4841 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4842 const char *md_shortopts = "kVQ:sq";
4843 #else
4844 const char *md_shortopts = "q";
4845 #endif
4847 struct option md_longopts[] = {
4848 #define OPTION_32 (OPTION_MD_BASE + 0)
4849 {"32", no_argument, NULL, OPTION_32},
4850 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4851 #define OPTION_64 (OPTION_MD_BASE + 1)
4852 {"64", no_argument, NULL, OPTION_64},
4853 #endif
4854 {NULL, no_argument, NULL, 0}
4856 size_t md_longopts_size = sizeof (md_longopts);
4859 md_parse_option (c, arg)
4860 int c;
4861 char *arg ATTRIBUTE_UNUSED;
4863 switch (c)
4865 case 'q':
4866 quiet_warnings = 1;
4867 break;
4869 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4870 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
4871 should be emitted or not. FIXME: Not implemented. */
4872 case 'Q':
4873 break;
4875 /* -V: SVR4 argument to print version ID. */
4876 case 'V':
4877 print_version_id ();
4878 break;
4880 /* -k: Ignore for FreeBSD compatibility. */
4881 case 'k':
4882 break;
4884 case 's':
4885 /* -s: On i386 Solaris, this tells the native assembler to use
4886 .stab instead of .stab.excl. We always use .stab anyhow. */
4887 break;
4889 case OPTION_64:
4891 const char **list, **l;
4893 list = bfd_target_list ();
4894 for (l = list; *l != NULL; l++)
4895 if (strcmp (*l, "elf64-x86-64") == 0)
4897 default_arch = "x86_64";
4898 break;
4900 if (*l == NULL)
4901 as_fatal (_("No compiled in support for x86_64"));
4902 free (list);
4904 break;
4905 #endif
4907 case OPTION_32:
4908 default_arch = "i386";
4909 break;
4911 default:
4912 return 0;
4914 return 1;
4917 void
4918 md_show_usage (stream)
4919 FILE *stream;
4921 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4922 fprintf (stream, _("\
4923 -Q ignored\n\
4924 -V print assembler version number\n\
4925 -k ignored\n\
4926 -q quieten some warnings\n\
4927 -s ignored\n"));
4928 #else
4929 fprintf (stream, _("\
4930 -q quieten some warnings\n"));
4931 #endif
4934 #ifdef BFD_ASSEMBLER
4935 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
4936 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
4938 /* Pick the target format to use. */
4940 const char *
4941 i386_target_format ()
4943 if (!strcmp (default_arch, "x86_64"))
4944 set_code_flag (CODE_64BIT);
4945 else if (!strcmp (default_arch, "i386"))
4946 set_code_flag (CODE_32BIT);
4947 else
4948 as_fatal (_("Unknown architecture"));
4949 switch (OUTPUT_FLAVOR)
4951 #ifdef OBJ_MAYBE_AOUT
4952 case bfd_target_aout_flavour:
4953 return AOUT_TARGET_FORMAT;
4954 #endif
4955 #ifdef OBJ_MAYBE_COFF
4956 case bfd_target_coff_flavour:
4957 return "coff-i386";
4958 #endif
4959 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
4960 case bfd_target_elf_flavour:
4962 if (flag_code == CODE_64BIT)
4963 use_rela_relocations = 1;
4964 return flag_code == CODE_64BIT ? "elf64-x86-64" : ELF_TARGET_FORMAT;
4966 #endif
4967 default:
4968 abort ();
4969 return NULL;
4973 #endif /* OBJ_MAYBE_ more than one */
4975 #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
4976 void i386_elf_emit_arch_note ()
4978 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4979 && cpu_arch_name != NULL)
4981 char *p;
4982 asection *seg = now_seg;
4983 subsegT subseg = now_subseg;
4984 Elf_Internal_Note i_note;
4985 Elf_External_Note e_note;
4986 asection *note_secp;
4987 int len;
4989 /* Create the .note section. */
4990 note_secp = subseg_new (".note", 0);
4991 bfd_set_section_flags (stdoutput,
4992 note_secp,
4993 SEC_HAS_CONTENTS | SEC_READONLY);
4995 /* Process the arch string. */
4996 len = strlen (cpu_arch_name);
4998 i_note.namesz = len + 1;
4999 i_note.descsz = 0;
5000 i_note.type = NT_ARCH;
5001 p = frag_more (sizeof (e_note.namesz));
5002 md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
5003 p = frag_more (sizeof (e_note.descsz));
5004 md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
5005 p = frag_more (sizeof (e_note.type));
5006 md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
5007 p = frag_more (len + 1);
5008 strcpy (p, cpu_arch_name);
5010 frag_align (2, 0, 0);
5012 subseg_set (seg, subseg);
5015 #endif
5016 #endif /* BFD_ASSEMBLER */
5018 symbolS *
5019 md_undefined_symbol (name)
5020 char *name;
5022 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
5023 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
5024 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
5025 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
5027 if (!GOT_symbol)
5029 if (symbol_find (name))
5030 as_bad (_("GOT already in symbol table"));
5031 GOT_symbol = symbol_new (name, undefined_section,
5032 (valueT) 0, &zero_address_frag);
5034 return GOT_symbol;
5036 return 0;
5039 /* Round up a section size to the appropriate boundary. */
5041 valueT
5042 md_section_align (segment, size)
5043 segT segment ATTRIBUTE_UNUSED;
5044 valueT size;
5046 #ifdef BFD_ASSEMBLER
5047 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
5048 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
5050 /* For a.out, force the section size to be aligned. If we don't do
5051 this, BFD will align it for us, but it will not write out the
5052 final bytes of the section. This may be a bug in BFD, but it is
5053 easier to fix it here since that is how the other a.out targets
5054 work. */
5055 int align;
5057 align = bfd_get_section_alignment (stdoutput, segment);
5058 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
5060 #endif
5061 #endif
5063 return size;
5066 /* On the i386, PC-relative offsets are relative to the start of the
5067 next instruction. That is, the address of the offset, plus its
5068 size, since the offset is always the last part of the insn. */
5070 long
5071 md_pcrel_from (fixP)
5072 fixS *fixP;
5074 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
5077 #ifndef I386COFF
5079 static void
5080 s_bss (ignore)
5081 int ignore ATTRIBUTE_UNUSED;
5083 int temp;
5085 temp = get_absolute_expression ();
5086 subseg_set (bss_section, (subsegT) temp);
5087 demand_empty_rest_of_line ();
5090 #endif
5092 #ifdef BFD_ASSEMBLER
5094 void
5095 i386_validate_fix (fixp)
5096 fixS *fixp;
5098 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
5100 /* GOTOFF relocation are nonsense in 64bit mode. */
5101 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
5103 if (flag_code != CODE_64BIT)
5104 abort ();
5105 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
5107 else
5109 if (flag_code == CODE_64BIT)
5110 abort ();
5111 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
5113 fixp->fx_subsy = 0;
5117 boolean
5118 i386_force_relocation (fixp)
5119 fixS *fixp;
5121 if (fixp->fx_r_type == BFD_RELOC_VTABLE_INHERIT
5122 || fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
5123 return 1;
5125 return S_FORCE_RELOC (fixp->fx_addsy);
5128 arelent *
5129 tc_gen_reloc (section, fixp)
5130 asection *section ATTRIBUTE_UNUSED;
5131 fixS *fixp;
5133 arelent *rel;
5134 bfd_reloc_code_real_type code;
5136 switch (fixp->fx_r_type)
5138 case BFD_RELOC_X86_64_PLT32:
5139 case BFD_RELOC_X86_64_GOT32:
5140 case BFD_RELOC_X86_64_GOTPCREL:
5141 case BFD_RELOC_386_PLT32:
5142 case BFD_RELOC_386_GOT32:
5143 case BFD_RELOC_386_GOTOFF:
5144 case BFD_RELOC_386_GOTPC:
5145 case BFD_RELOC_386_TLS_GD:
5146 case BFD_RELOC_386_TLS_LDM:
5147 case BFD_RELOC_386_TLS_LDO_32:
5148 case BFD_RELOC_386_TLS_IE_32:
5149 case BFD_RELOC_386_TLS_IE:
5150 case BFD_RELOC_386_TLS_GOTIE:
5151 case BFD_RELOC_386_TLS_LE_32:
5152 case BFD_RELOC_386_TLS_LE:
5153 case BFD_RELOC_X86_64_32S:
5154 case BFD_RELOC_RVA:
5155 case BFD_RELOC_VTABLE_ENTRY:
5156 case BFD_RELOC_VTABLE_INHERIT:
5157 code = fixp->fx_r_type;
5158 break;
5159 default:
5160 if (fixp->fx_pcrel)
5162 switch (fixp->fx_size)
5164 default:
5165 as_bad_where (fixp->fx_file, fixp->fx_line,
5166 _("can not do %d byte pc-relative relocation"),
5167 fixp->fx_size);
5168 code = BFD_RELOC_32_PCREL;
5169 break;
5170 case 1: code = BFD_RELOC_8_PCREL; break;
5171 case 2: code = BFD_RELOC_16_PCREL; break;
5172 case 4: code = BFD_RELOC_32_PCREL; break;
5175 else
5177 switch (fixp->fx_size)
5179 default:
5180 as_bad_where (fixp->fx_file, fixp->fx_line,
5181 _("can not do %d byte relocation"),
5182 fixp->fx_size);
5183 code = BFD_RELOC_32;
5184 break;
5185 case 1: code = BFD_RELOC_8; break;
5186 case 2: code = BFD_RELOC_16; break;
5187 case 4: code = BFD_RELOC_32; break;
5188 #ifdef BFD64
5189 case 8: code = BFD_RELOC_64; break;
5190 #endif
5193 break;
5196 if (code == BFD_RELOC_32
5197 && GOT_symbol
5198 && fixp->fx_addsy == GOT_symbol)
5200 /* We don't support GOTPC on 64bit targets. */
5201 if (flag_code == CODE_64BIT)
5202 abort ();
5203 code = BFD_RELOC_386_GOTPC;
5206 rel = (arelent *) xmalloc (sizeof (arelent));
5207 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
5208 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
5210 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
5211 if (!use_rela_relocations)
5213 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
5214 vtable entry to be used in the relocation's section offset. */
5215 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
5216 rel->address = fixp->fx_offset;
5218 rel->addend = 0;
5220 /* Use the rela in 64bit mode. */
5221 else
5223 if (!fixp->fx_pcrel)
5224 rel->addend = fixp->fx_offset;
5225 else
5226 switch (code)
5228 case BFD_RELOC_X86_64_PLT32:
5229 case BFD_RELOC_X86_64_GOT32:
5230 case BFD_RELOC_X86_64_GOTPCREL:
5231 rel->addend = fixp->fx_offset - fixp->fx_size;
5232 break;
5233 default:
5234 rel->addend = (section->vma
5235 - fixp->fx_size
5236 + fixp->fx_addnumber
5237 + md_pcrel_from (fixp));
5238 break;
5242 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
5243 if (rel->howto == NULL)
5245 as_bad_where (fixp->fx_file, fixp->fx_line,
5246 _("cannot represent relocation type %s"),
5247 bfd_get_reloc_code_name (code));
5248 /* Set howto to a garbage value so that we can keep going. */
5249 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
5250 assert (rel->howto != NULL);
5253 return rel;
5256 #else /* !BFD_ASSEMBLER */
5258 #if (defined(OBJ_AOUT) | defined(OBJ_BOUT))
5259 void
5260 tc_aout_fix_to_chars (where, fixP, segment_address_in_file)
5261 char *where;
5262 fixS *fixP;
5263 relax_addressT segment_address_in_file;
5265 /* In: length of relocation (or of address) in chars: 1, 2 or 4.
5266 Out: GNU LD relocation length code: 0, 1, or 2. */
5268 static const unsigned char nbytes_r_length[] = { 42, 0, 1, 42, 2 };
5269 long r_symbolnum;
5271 know (fixP->fx_addsy != NULL);
5273 md_number_to_chars (where,
5274 (valueT) (fixP->fx_frag->fr_address
5275 + fixP->fx_where - segment_address_in_file),
5278 r_symbolnum = (S_IS_DEFINED (fixP->fx_addsy)
5279 ? S_GET_TYPE (fixP->fx_addsy)
5280 : fixP->fx_addsy->sy_number);
5282 where[6] = (r_symbolnum >> 16) & 0x0ff;
5283 where[5] = (r_symbolnum >> 8) & 0x0ff;
5284 where[4] = r_symbolnum & 0x0ff;
5285 where[7] = ((((!S_IS_DEFINED (fixP->fx_addsy)) << 3) & 0x08)
5286 | ((nbytes_r_length[fixP->fx_size] << 1) & 0x06)
5287 | (((fixP->fx_pcrel << 0) & 0x01) & 0x0f));
5290 #endif /* OBJ_AOUT or OBJ_BOUT. */
5292 #if defined (I386COFF)
5294 short
5295 tc_coff_fix2rtype (fixP)
5296 fixS *fixP;
5298 if (fixP->fx_r_type == R_IMAGEBASE)
5299 return R_IMAGEBASE;
5301 return (fixP->fx_pcrel ?
5302 (fixP->fx_size == 1 ? R_PCRBYTE :
5303 fixP->fx_size == 2 ? R_PCRWORD :
5304 R_PCRLONG) :
5305 (fixP->fx_size == 1 ? R_RELBYTE :
5306 fixP->fx_size == 2 ? R_RELWORD :
5307 R_DIR32));
5311 tc_coff_sizemachdep (frag)
5312 fragS *frag;
5314 if (frag->fr_next)
5315 return (frag->fr_next->fr_address - frag->fr_address);
5316 else
5317 return 0;
5320 #endif /* I386COFF */
5322 #endif /* !BFD_ASSEMBLER */
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 && i386_index_check (s);
5558 /* Constant and OFFSET expressions are handled by i386_immediate. */
5559 else if (intel_parser.op_modifier == OFFSET_FLAT
5560 || intel_parser.reg == NULL)
5561 ret = i386_immediate (intel_parser.disp);
5564 free (p);
5565 free (intel_parser.disp);
5567 return ret;
5570 /* expr SHORT e05
5571 | e05 */
5572 static int
5573 intel_expr ()
5575 /* expr SHORT e05 */
5576 if (cur_token.code == T_SHORT)
5578 intel_parser.op_modifier = SHORT;
5579 intel_match_token (T_SHORT);
5581 return (intel_e05 ());
5584 /* expr e05 */
5585 else
5586 return intel_e05 ();
5589 /* e05 e06 e05'
5591 e05' addOp e06 e05'
5592 | Empty */
5593 static int
5594 intel_e05 ()
5596 return (intel_e06 () && intel_e05_1 ());
5599 static int
5600 intel_e05_1 ()
5602 /* e05' addOp e06 e05' */
5603 if (cur_token.code == '+' || cur_token.code == '-')
5605 strcat (intel_parser.disp, cur_token.str);
5606 intel_match_token (cur_token.code);
5608 return (intel_e06 () && intel_e05_1 ());
5611 /* e05' Empty */
5612 else
5613 return 1;
5616 /* e06 e09 e06'
5618 e06' mulOp e09 e06'
5619 | Empty */
5620 static int
5621 intel_e06 ()
5623 return (intel_e09 () && intel_e06_1 ());
5626 static int
5627 intel_e06_1 ()
5629 /* e06' mulOp e09 e06' */
5630 if (cur_token.code == '*' || cur_token.code == '/')
5632 strcat (intel_parser.disp, cur_token.str);
5633 intel_match_token (cur_token.code);
5635 return (intel_e09 () && intel_e06_1 ());
5638 /* e06' Empty */
5639 else
5640 return 1;
5643 /* e09 OFFSET e10 e09'
5644 | e10 e09'
5646 e09' PTR e10 e09'
5647 | : e10 e09'
5648 | Empty */
5649 static int
5650 intel_e09 ()
5652 /* e09 OFFSET e10 e09' */
5653 if (cur_token.code == T_OFFSET)
5655 intel_parser.is_mem = 0;
5656 intel_parser.op_modifier = OFFSET_FLAT;
5657 intel_match_token (T_OFFSET);
5659 return (intel_e10 () && intel_e09_1 ());
5662 /* e09 e10 e09' */
5663 else
5664 return (intel_e10 () && intel_e09_1 ());
5667 static int
5668 intel_e09_1 ()
5670 /* e09' PTR e10 e09' */
5671 if (cur_token.code == T_PTR)
5673 if (prev_token.code == T_BYTE)
5674 i.suffix = BYTE_MNEM_SUFFIX;
5676 else if (prev_token.code == T_WORD)
5678 if (intel_parser.got_a_float == 2) /* "fi..." */
5679 i.suffix = SHORT_MNEM_SUFFIX;
5680 else
5681 i.suffix = WORD_MNEM_SUFFIX;
5684 else if (prev_token.code == T_DWORD)
5686 if (intel_parser.got_a_float == 1) /* "f..." */
5687 i.suffix = SHORT_MNEM_SUFFIX;
5688 else
5689 i.suffix = LONG_MNEM_SUFFIX;
5692 else if (prev_token.code == T_QWORD)
5694 if (intel_parser.got_a_float == 1) /* "f..." */
5695 i.suffix = LONG_MNEM_SUFFIX;
5696 else
5697 i.suffix = QWORD_MNEM_SUFFIX;
5700 else if (prev_token.code == T_XWORD)
5701 i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
5703 else
5705 as_bad (_("Unknown operand modifier `%s'\n"), prev_token.str);
5706 return 0;
5709 intel_match_token (T_PTR);
5711 return (intel_e10 () && intel_e09_1 ());
5714 /* e09 : e10 e09' */
5715 else if (cur_token.code == ':')
5717 /* Mark as a memory operand only if it's not already known to be an
5718 offset expression. */
5719 if (intel_parser.op_modifier != OFFSET_FLAT)
5720 intel_parser.is_mem = 1;
5722 return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
5725 /* e09' Empty */
5726 else
5727 return 1;
5730 /* e10 e11 e10'
5732 e10' [ expr ] e10'
5733 | Empty */
5734 static int
5735 intel_e10 ()
5737 return (intel_e11 () && intel_e10_1 ());
5740 static int
5741 intel_e10_1 ()
5743 /* e10' [ expr ] e10' */
5744 if (cur_token.code == '[')
5746 intel_match_token ('[');
5748 /* Mark as a memory operand only if it's not already known to be an
5749 offset expression. If it's an offset expression, we need to keep
5750 the brace in. */
5751 if (intel_parser.op_modifier != OFFSET_FLAT)
5752 intel_parser.is_mem = 1;
5753 else
5754 strcat (intel_parser.disp, "[");
5756 /* Add a '+' to the displacement string if necessary. */
5757 if (*intel_parser.disp != '\0'
5758 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5759 strcat (intel_parser.disp, "+");
5761 if (intel_expr () && intel_match_token (']'))
5763 /* Preserve brackets when the operand is an offset expression. */
5764 if (intel_parser.op_modifier == OFFSET_FLAT)
5765 strcat (intel_parser.disp, "]");
5767 return intel_e10_1 ();
5769 else
5770 return 0;
5773 /* e10' Empty */
5774 else
5775 return 1;
5778 /* e11 ( expr )
5779 | [ expr ]
5780 | BYTE
5781 | WORD
5782 | DWORD
5783 | QWORD
5784 | XWORD
5787 | register
5788 | id
5789 | constant */
5790 static int
5791 intel_e11 ()
5793 /* e11 ( expr ) */
5794 if (cur_token.code == '(')
5796 intel_match_token ('(');
5797 strcat (intel_parser.disp, "(");
5799 if (intel_expr () && intel_match_token (')'))
5801 strcat (intel_parser.disp, ")");
5802 return 1;
5804 else
5805 return 0;
5808 /* e11 [ expr ] */
5809 else if (cur_token.code == '[')
5811 intel_match_token ('[');
5813 /* Mark as a memory operand only if it's not already known to be an
5814 offset expression. If it's an offset expression, we need to keep
5815 the brace in. */
5816 if (intel_parser.op_modifier != OFFSET_FLAT)
5817 intel_parser.is_mem = 1;
5818 else
5819 strcat (intel_parser.disp, "[");
5821 /* Operands for jump/call inside brackets denote absolute addresses. */
5822 if (current_templates->start->opcode_modifier & Jump
5823 || current_templates->start->opcode_modifier & JumpDword
5824 || current_templates->start->opcode_modifier & JumpByte
5825 || current_templates->start->opcode_modifier & JumpInterSegment)
5826 i.types[this_operand] |= JumpAbsolute;
5828 /* Add a '+' to the displacement string if necessary. */
5829 if (*intel_parser.disp != '\0'
5830 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5831 strcat (intel_parser.disp, "+");
5833 if (intel_expr () && intel_match_token (']'))
5835 /* Preserve brackets when the operand is an offset expression. */
5836 if (intel_parser.op_modifier == OFFSET_FLAT)
5837 strcat (intel_parser.disp, "]");
5839 return 1;
5841 else
5842 return 0;
5845 /* e11 BYTE
5846 | WORD
5847 | DWORD
5848 | QWORD
5849 | XWORD */
5850 else if (cur_token.code == T_BYTE
5851 || cur_token.code == T_WORD
5852 || cur_token.code == T_DWORD
5853 || cur_token.code == T_QWORD
5854 || cur_token.code == T_XWORD)
5856 intel_match_token (cur_token.code);
5858 return 1;
5861 /* e11 $
5862 | . */
5863 else if (cur_token.code == '$' || cur_token.code == '.')
5865 strcat (intel_parser.disp, cur_token.str);
5866 intel_match_token (cur_token.code);
5868 /* Mark as a memory operand only if it's not already known to be an
5869 offset expression. */
5870 if (intel_parser.op_modifier != OFFSET_FLAT)
5871 intel_parser.is_mem = 1;
5873 return 1;
5876 /* e11 register */
5877 else if (cur_token.code == T_REG)
5879 const reg_entry *reg = intel_parser.reg = cur_token.reg;
5881 intel_match_token (T_REG);
5883 /* Check for segment change. */
5884 if (cur_token.code == ':')
5886 if (reg->reg_type & (SReg2 | SReg3))
5888 switch (reg->reg_num)
5890 case 0:
5891 i.seg[i.mem_operands] = &es;
5892 break;
5893 case 1:
5894 i.seg[i.mem_operands] = &cs;
5895 break;
5896 case 2:
5897 i.seg[i.mem_operands] = &ss;
5898 break;
5899 case 3:
5900 i.seg[i.mem_operands] = &ds;
5901 break;
5902 case 4:
5903 i.seg[i.mem_operands] = &fs;
5904 break;
5905 case 5:
5906 i.seg[i.mem_operands] = &gs;
5907 break;
5910 else
5912 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
5913 return 0;
5917 /* Not a segment register. Check for register scaling. */
5918 else if (cur_token.code == '*')
5920 if (!intel_parser.is_mem)
5922 as_bad (_("Register scaling only allowed in memory operands."));
5923 return 0;
5926 /* What follows must be a valid scale. */
5927 if (intel_match_token ('*')
5928 && strchr ("01248", *cur_token.str))
5930 i.index_reg = reg;
5931 i.types[this_operand] |= BaseIndex;
5933 /* Set the scale after setting the register (otherwise,
5934 i386_scale will complain) */
5935 i386_scale (cur_token.str);
5936 intel_match_token (T_CONST);
5938 else
5940 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
5941 cur_token.str);
5942 return 0;
5946 /* No scaling. If this is a memory operand, the register is either a
5947 base register (first occurrence) or an index register (second
5948 occurrence). */
5949 else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3)))
5951 if (i.base_reg && i.index_reg)
5953 as_bad (_("Too many register references in memory operand.\n"));
5954 return 0;
5957 if (i.base_reg == NULL)
5958 i.base_reg = reg;
5959 else
5960 i.index_reg = reg;
5962 i.types[this_operand] |= BaseIndex;
5965 /* Offset modifier. Add the register to the displacement string to be
5966 parsed as an immediate expression after we're done. */
5967 else if (intel_parser.op_modifier == OFFSET_FLAT)
5968 strcat (intel_parser.disp, reg->reg_name);
5970 /* It's neither base nor index nor offset. */
5971 else
5973 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
5974 i.op[this_operand].regs = reg;
5975 i.reg_operands++;
5978 /* Since registers are not part of the displacement string (except
5979 when we're parsing offset operands), we may need to remove any
5980 preceding '+' from the displacement string. */
5981 if (*intel_parser.disp != '\0'
5982 && intel_parser.op_modifier != OFFSET_FLAT)
5984 char *s = intel_parser.disp;
5985 s += strlen (s) - 1;
5986 if (*s == '+')
5987 *s = '\0';
5990 return 1;
5993 /* e11 id */
5994 else if (cur_token.code == T_ID)
5996 /* Add the identifier to the displacement string. */
5997 strcat (intel_parser.disp, cur_token.str);
5998 intel_match_token (T_ID);
6000 /* The identifier represents a memory reference only if it's not
6001 preceded by an offset modifier. */
6002 if (intel_parser.op_modifier != OFFSET_FLAT)
6003 intel_parser.is_mem = 1;
6005 return 1;
6008 /* e11 constant */
6009 else if (cur_token.code == T_CONST
6010 || cur_token.code == '-'
6011 || cur_token.code == '+')
6013 char *save_str;
6015 /* Allow constants that start with `+' or `-'. */
6016 if (cur_token.code == '-' || cur_token.code == '+')
6018 strcat (intel_parser.disp, cur_token.str);
6019 intel_match_token (cur_token.code);
6020 if (cur_token.code != T_CONST)
6022 as_bad (_("Syntax error. Expecting a constant. Got `%s'.\n"),
6023 cur_token.str);
6024 return 0;
6028 save_str = (char *) malloc (strlen (cur_token.str) + 1);
6029 if (save_str == NULL)
6030 abort ();
6031 strcpy (save_str, cur_token.str);
6033 /* Get the next token to check for register scaling. */
6034 intel_match_token (cur_token.code);
6036 /* Check if this constant is a scaling factor for an index register. */
6037 if (cur_token.code == '*')
6039 if (intel_match_token ('*') && cur_token.code == T_REG)
6041 if (!intel_parser.is_mem)
6043 as_bad (_("Register scaling only allowed in memory operands."));
6044 return 0;
6047 /* The constant is followed by `* reg', so it must be
6048 a valid scale. */
6049 if (strchr ("01248", *save_str))
6051 i.index_reg = cur_token.reg;
6052 i.types[this_operand] |= BaseIndex;
6054 /* Set the scale after setting the register (otherwise,
6055 i386_scale will complain) */
6056 i386_scale (save_str);
6057 intel_match_token (T_REG);
6059 /* Since registers are not part of the displacement
6060 string, we may need to remove any preceding '+' from
6061 the displacement string. */
6062 if (*intel_parser.disp != '\0')
6064 char *s = intel_parser.disp;
6065 s += strlen (s) - 1;
6066 if (*s == '+')
6067 *s = '\0';
6070 free (save_str);
6072 return 1;
6074 else
6075 return 0;
6078 /* The constant was not used for register scaling. Since we have
6079 already consumed the token following `*' we now need to put it
6080 back in the stream. */
6081 else
6082 intel_putback_token ();
6085 /* Add the constant to the displacement string. */
6086 strcat (intel_parser.disp, save_str);
6087 free (save_str);
6089 return 1;
6092 as_bad (_("Unrecognized token '%s'"), cur_token.str);
6093 return 0;
6096 /* Match the given token against cur_token. If they match, read the next
6097 token from the operand string. */
6098 static int
6099 intel_match_token (code)
6100 int code;
6102 if (cur_token.code == code)
6104 intel_get_token ();
6105 return 1;
6107 else
6109 as_bad (_("Unexpected token `%s'\n"), cur_token.str);
6110 return 0;
6114 /* Read a new token from intel_parser.op_string and store it in cur_token. */
6115 static void
6116 intel_get_token ()
6118 char *end_op;
6119 const reg_entry *reg;
6120 struct intel_token new_token;
6122 new_token.code = T_NIL;
6123 new_token.reg = NULL;
6124 new_token.str = NULL;
6126 /* Free the memory allocated to the previous token and move
6127 cur_token to prev_token. */
6128 if (prev_token.str)
6129 free (prev_token.str);
6131 prev_token = cur_token;
6133 /* Skip whitespace. */
6134 while (is_space_char (*intel_parser.op_string))
6135 intel_parser.op_string++;
6137 /* Return an empty token if we find nothing else on the line. */
6138 if (*intel_parser.op_string == '\0')
6140 cur_token = new_token;
6141 return;
6144 /* The new token cannot be larger than the remainder of the operand
6145 string. */
6146 new_token.str = (char *) malloc (strlen (intel_parser.op_string) + 1);
6147 if (new_token.str == NULL)
6148 abort ();
6149 new_token.str[0] = '\0';
6151 if (strchr ("0123456789", *intel_parser.op_string))
6153 char *p = new_token.str;
6154 char *q = intel_parser.op_string;
6155 new_token.code = T_CONST;
6157 /* Allow any kind of identifier char to encompass floating point and
6158 hexadecimal numbers. */
6159 while (is_identifier_char (*q))
6160 *p++ = *q++;
6161 *p = '\0';
6163 /* Recognize special symbol names [0-9][bf]. */
6164 if (strlen (intel_parser.op_string) == 2
6165 && (intel_parser.op_string[1] == 'b'
6166 || intel_parser.op_string[1] == 'f'))
6167 new_token.code = T_ID;
6170 else if (strchr ("+-/*:[]()", *intel_parser.op_string))
6172 new_token.code = *intel_parser.op_string;
6173 new_token.str[0] = *intel_parser.op_string;
6174 new_token.str[1] = '\0';
6177 else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
6178 && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
6180 new_token.code = T_REG;
6181 new_token.reg = reg;
6183 if (*intel_parser.op_string == REGISTER_PREFIX)
6185 new_token.str[0] = REGISTER_PREFIX;
6186 new_token.str[1] = '\0';
6189 strcat (new_token.str, reg->reg_name);
6192 else if (is_identifier_char (*intel_parser.op_string))
6194 char *p = new_token.str;
6195 char *q = intel_parser.op_string;
6197 /* A '.' or '$' followed by an identifier char is an identifier.
6198 Otherwise, it's operator '.' followed by an expression. */
6199 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
6201 new_token.code = *q;
6202 new_token.str[0] = *q;
6203 new_token.str[1] = '\0';
6205 else
6207 while (is_identifier_char (*q) || *q == '@')
6208 *p++ = *q++;
6209 *p = '\0';
6211 if (strcasecmp (new_token.str, "BYTE") == 0)
6212 new_token.code = T_BYTE;
6214 else if (strcasecmp (new_token.str, "WORD") == 0)
6215 new_token.code = T_WORD;
6217 else if (strcasecmp (new_token.str, "DWORD") == 0)
6218 new_token.code = T_DWORD;
6220 else if (strcasecmp (new_token.str, "QWORD") == 0)
6221 new_token.code = T_QWORD;
6223 else if (strcasecmp (new_token.str, "XWORD") == 0)
6224 new_token.code = T_XWORD;
6226 else if (strcasecmp (new_token.str, "PTR") == 0)
6227 new_token.code = T_PTR;
6229 else if (strcasecmp (new_token.str, "SHORT") == 0)
6230 new_token.code = T_SHORT;
6232 else if (strcasecmp (new_token.str, "OFFSET") == 0)
6234 new_token.code = T_OFFSET;
6236 /* ??? This is not mentioned in the MASM grammar but gcc
6237 makes use of it with -mintel-syntax. OFFSET may be
6238 followed by FLAT: */
6239 if (strncasecmp (q, " FLAT:", 6) == 0)
6240 strcat (new_token.str, " FLAT:");
6243 /* ??? This is not mentioned in the MASM grammar. */
6244 else if (strcasecmp (new_token.str, "FLAT") == 0)
6245 new_token.code = T_OFFSET;
6247 else
6248 new_token.code = T_ID;
6252 else
6253 as_bad (_("Unrecognized token `%s'\n"), intel_parser.op_string);
6255 intel_parser.op_string += strlen (new_token.str);
6256 cur_token = new_token;
6259 /* Put cur_token back into the token stream and make cur_token point to
6260 prev_token. */
6261 static void
6262 intel_putback_token ()
6264 intel_parser.op_string -= strlen (cur_token.str);
6265 free (cur_token.str);
6266 cur_token = prev_token;
6268 /* Forget prev_token. */
6269 prev_token.code = T_NIL;
6270 prev_token.reg = NULL;
6271 prev_token.str = NULL;