For DefaultSize instructions, don't guess a 'q' suffix if the instruction
[binutils-gdb.git] / gas / config / tc-i386.c
1 /* i386.c -- Assemble code for the Intel 80386
2 Copyright 1989, 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
3 2000, 2001, 2002, 2003, 2004
4 Free Software Foundation, Inc.
5
6 This file is part of GAS, the GNU Assembler.
7
8 GAS is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2, or (at your option)
11 any later version.
12
13 GAS is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
17
18 You should have received a copy of the GNU General Public License
19 along with GAS; see the file COPYING. If not, write to the Free
20 Software Foundation, 59 Temple Place - Suite 330, Boston, MA
21 02111-1307, USA. */
22
23 /* Intel 80386 machine specific gas.
24 Written by Eliot Dresselhaus (eliot@mgm.mit.edu).
25 x86_64 support by Jan Hubicka (jh@suse.cz)
26 VIA PadLock support by Michal Ludvig (mludvig@suse.cz)
27 Bugs & suggestions are completely welcome. This is free software.
28 Please help us make it better. */
29
30 #include "as.h"
31 #include "safe-ctype.h"
32 #include "subsegs.h"
33 #include "dwarf2dbg.h"
34 #include "dw2gencfi.h"
35 #include "opcode/i386.h"
36
37 #ifndef REGISTER_WARNINGS
38 #define REGISTER_WARNINGS 1
39 #endif
40
41 #ifndef INFER_ADDR_PREFIX
42 #define INFER_ADDR_PREFIX 1
43 #endif
44
45 #ifndef SCALE1_WHEN_NO_INDEX
46 /* Specifying a scale factor besides 1 when there is no index is
47 futile. eg. `mov (%ebx,2),%al' does exactly the same as
48 `mov (%ebx),%al'. To slavishly follow what the programmer
49 specified, set SCALE1_WHEN_NO_INDEX to 0. */
50 #define SCALE1_WHEN_NO_INDEX 1
51 #endif
52
53 #ifndef DEFAULT_ARCH
54 #define DEFAULT_ARCH "i386"
55 #endif
56
57 #ifndef INLINE
58 #if __GNUC__ >= 2
59 #define INLINE __inline__
60 #else
61 #define INLINE
62 #endif
63 #endif
64
65 static INLINE unsigned int mode_from_disp_size PARAMS ((unsigned int));
66 static INLINE int fits_in_signed_byte PARAMS ((offsetT));
67 static INLINE int fits_in_unsigned_byte PARAMS ((offsetT));
68 static INLINE int fits_in_unsigned_word PARAMS ((offsetT));
69 static INLINE int fits_in_signed_word PARAMS ((offsetT));
70 static INLINE int fits_in_unsigned_long PARAMS ((offsetT));
71 static INLINE int fits_in_signed_long PARAMS ((offsetT));
72 static int smallest_imm_type PARAMS ((offsetT));
73 static offsetT offset_in_range PARAMS ((offsetT, int));
74 static int add_prefix PARAMS ((unsigned int));
75 static void set_code_flag PARAMS ((int));
76 static void set_16bit_gcc_code_flag PARAMS ((int));
77 static void set_intel_syntax PARAMS ((int));
78 static void set_cpu_arch PARAMS ((int));
79 #ifdef TE_PE
80 static void pe_directive_secrel PARAMS ((int));
81 #endif
82 static char *output_invalid PARAMS ((int c));
83 static int i386_operand PARAMS ((char *operand_string));
84 static int i386_intel_operand PARAMS ((char *operand_string, int got_a_float));
85 static const reg_entry *parse_register PARAMS ((char *reg_string,
86 char **end_op));
87 static char *parse_insn PARAMS ((char *, char *));
88 static char *parse_operands PARAMS ((char *, const char *));
89 static void swap_operands PARAMS ((void));
90 static void optimize_imm PARAMS ((void));
91 static void optimize_disp PARAMS ((void));
92 static int match_template PARAMS ((void));
93 static int check_string PARAMS ((void));
94 static int process_suffix PARAMS ((void));
95 static int check_byte_reg PARAMS ((void));
96 static int check_long_reg PARAMS ((void));
97 static int check_qword_reg PARAMS ((void));
98 static int check_word_reg PARAMS ((void));
99 static int finalize_imm PARAMS ((void));
100 static int process_operands PARAMS ((void));
101 static const seg_entry *build_modrm_byte PARAMS ((void));
102 static void output_insn PARAMS ((void));
103 static void output_branch PARAMS ((void));
104 static void output_jump PARAMS ((void));
105 static void output_interseg_jump PARAMS ((void));
106 static void output_imm PARAMS ((fragS *insn_start_frag,
107 offsetT insn_start_off));
108 static void output_disp PARAMS ((fragS *insn_start_frag,
109 offsetT insn_start_off));
110 #ifndef I386COFF
111 static void s_bss PARAMS ((int));
112 #endif
113
114 static const char *default_arch = DEFAULT_ARCH;
115
116 /* 'md_assemble ()' gathers together information and puts it into a
117 i386_insn. */
118
119 union i386_op
120 {
121 expressionS *disps;
122 expressionS *imms;
123 const reg_entry *regs;
124 };
125
126 struct _i386_insn
127 {
128 /* TM holds the template for the insn were currently assembling. */
129 template tm;
130
131 /* SUFFIX holds the instruction mnemonic suffix if given.
132 (e.g. 'l' for 'movl') */
133 char suffix;
134
135 /* OPERANDS gives the number of given operands. */
136 unsigned int operands;
137
138 /* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number
139 of given register, displacement, memory operands and immediate
140 operands. */
141 unsigned int reg_operands, disp_operands, mem_operands, imm_operands;
142
143 /* TYPES [i] is the type (see above #defines) which tells us how to
144 use OP[i] for the corresponding operand. */
145 unsigned int types[MAX_OPERANDS];
146
147 /* Displacement expression, immediate expression, or register for each
148 operand. */
149 union i386_op op[MAX_OPERANDS];
150
151 /* Flags for operands. */
152 unsigned int flags[MAX_OPERANDS];
153 #define Operand_PCrel 1
154
155 /* Relocation type for operand */
156 enum bfd_reloc_code_real reloc[MAX_OPERANDS];
157
158 /* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode
159 the base index byte below. */
160 const reg_entry *base_reg;
161 const reg_entry *index_reg;
162 unsigned int log2_scale_factor;
163
164 /* SEG gives the seg_entries of this insn. They are zero unless
165 explicit segment overrides are given. */
166 const seg_entry *seg[2];
167
168 /* PREFIX holds all the given prefix opcodes (usually null).
169 PREFIXES is the number of prefix opcodes. */
170 unsigned int prefixes;
171 unsigned char prefix[MAX_PREFIXES];
172
173 /* RM and SIB are the modrm byte and the sib byte where the
174 addressing modes of this insn are encoded. */
175
176 modrm_byte rm;
177 rex_byte rex;
178 sib_byte sib;
179 };
180
181 typedef struct _i386_insn i386_insn;
182
183 /* List of chars besides those in app.c:symbol_chars that can start an
184 operand. Used to prevent the scrubber eating vital white-space. */
185 const char extra_symbol_chars[] = "*%-(["
186 #ifdef LEX_AT
187 "@"
188 #endif
189 #ifdef LEX_QM
190 "?"
191 #endif
192 ;
193
194 #if (defined (TE_I386AIX) \
195 || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) \
196 && !defined (TE_LINUX) \
197 && !defined (TE_NETWARE) \
198 && !defined (TE_FreeBSD) \
199 && !defined (TE_NetBSD)))
200 /* This array holds the chars that always start a comment. If the
201 pre-processor is disabled, these aren't very useful. */
202 const char comment_chars[] = "#/";
203 #define PREFIX_SEPARATOR '\\'
204
205 /* This array holds the chars that only start a comment at the beginning of
206 a line. If the line seems to have the form '# 123 filename'
207 .line and .file directives will appear in the pre-processed output.
208 Note that input_file.c hand checks for '#' at the beginning of the
209 first line of the input file. This is because the compiler outputs
210 #NO_APP at the beginning of its output.
211 Also note that comments started like this one will always work if
212 '/' isn't otherwise defined. */
213 const char line_comment_chars[] = "#";
214
215 #else
216 /* Putting '/' here makes it impossible to use the divide operator.
217 However, we need it for compatibility with SVR4 systems. */
218 const char comment_chars[] = "#";
219 #define PREFIX_SEPARATOR '/'
220
221 const char line_comment_chars[] = "/#";
222 #endif
223
224 const char line_separator_chars[] = ";";
225
226 /* Chars that can be used to separate mant from exp in floating point
227 nums. */
228 const char EXP_CHARS[] = "eE";
229
230 /* Chars that mean this number is a floating point constant
231 As in 0f12.456
232 or 0d1.2345e12. */
233 const char FLT_CHARS[] = "fFdDxX";
234
235 /* Tables for lexical analysis. */
236 static char mnemonic_chars[256];
237 static char register_chars[256];
238 static char operand_chars[256];
239 static char identifier_chars[256];
240 static char digit_chars[256];
241
242 /* Lexical macros. */
243 #define is_mnemonic_char(x) (mnemonic_chars[(unsigned char) x])
244 #define is_operand_char(x) (operand_chars[(unsigned char) x])
245 #define is_register_char(x) (register_chars[(unsigned char) x])
246 #define is_space_char(x) ((x) == ' ')
247 #define is_identifier_char(x) (identifier_chars[(unsigned char) x])
248 #define is_digit_char(x) (digit_chars[(unsigned char) x])
249
250 /* All non-digit non-letter characters that may occur in an operand. */
251 static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
252
253 /* md_assemble() always leaves the strings it's passed unaltered. To
254 effect this we maintain a stack of saved characters that we've smashed
255 with '\0's (indicating end of strings for various sub-fields of the
256 assembler instruction). */
257 static char save_stack[32];
258 static char *save_stack_p;
259 #define END_STRING_AND_SAVE(s) \
260 do { *save_stack_p++ = *(s); *(s) = '\0'; } while (0)
261 #define RESTORE_END_STRING(s) \
262 do { *(s) = *--save_stack_p; } while (0)
263
264 /* The instruction we're assembling. */
265 static i386_insn i;
266
267 /* Possible templates for current insn. */
268 static const templates *current_templates;
269
270 /* Per instruction expressionS buffers: 2 displacements & 2 immediate max. */
271 static expressionS disp_expressions[2], im_expressions[2];
272
273 /* Current operand we are working on. */
274 static int this_operand;
275
276 /* We support four different modes. FLAG_CODE variable is used to distinguish
277 these. */
278
279 enum flag_code {
280 CODE_32BIT,
281 CODE_16BIT,
282 CODE_64BIT };
283 #define NUM_FLAG_CODE ((int) CODE_64BIT + 1)
284
285 static enum flag_code flag_code;
286 static int use_rela_relocations = 0;
287
288 /* The names used to print error messages. */
289 static const char *flag_code_names[] =
290 {
291 "32",
292 "16",
293 "64"
294 };
295
296 /* 1 for intel syntax,
297 0 if att syntax. */
298 static int intel_syntax = 0;
299
300 /* 1 if register prefix % not required. */
301 static int allow_naked_reg = 0;
302
303 /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
304 leave, push, and pop instructions so that gcc has the same stack
305 frame as in 32 bit mode. */
306 static char stackop_size = '\0';
307
308 /* Non-zero to optimize code alignment. */
309 int optimize_align_code = 1;
310
311 /* Non-zero to quieten some warnings. */
312 static int quiet_warnings = 0;
313
314 /* CPU name. */
315 static const char *cpu_arch_name = NULL;
316
317 /* CPU feature flags. */
318 static unsigned int cpu_arch_flags = CpuUnknownFlags | CpuNo64;
319
320 /* If set, conditional jumps are not automatically promoted to handle
321 larger than a byte offset. */
322 static unsigned int no_cond_jump_promotion = 0;
323
324 /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
325 symbolS *GOT_symbol;
326
327 /* The dwarf2 return column, adjusted for 32 or 64 bit. */
328 unsigned int x86_dwarf2_return_column;
329
330 /* The dwarf2 data alignment, adjusted for 32 or 64 bit. */
331 int x86_cie_data_alignment;
332
333 /* Interface to relax_segment.
334 There are 3 major relax states for 386 jump insns because the
335 different types of jumps add different sizes to frags when we're
336 figuring out what sort of jump to choose to reach a given label. */
337
338 /* Types. */
339 #define UNCOND_JUMP 0
340 #define COND_JUMP 1
341 #define COND_JUMP86 2
342
343 /* Sizes. */
344 #define CODE16 1
345 #define SMALL 0
346 #define SMALL16 (SMALL | CODE16)
347 #define BIG 2
348 #define BIG16 (BIG | CODE16)
349
350 #ifndef INLINE
351 #ifdef __GNUC__
352 #define INLINE __inline__
353 #else
354 #define INLINE
355 #endif
356 #endif
357
358 #define ENCODE_RELAX_STATE(type, size) \
359 ((relax_substateT) (((type) << 2) | (size)))
360 #define TYPE_FROM_RELAX_STATE(s) \
361 ((s) >> 2)
362 #define DISP_SIZE_FROM_RELAX_STATE(s) \
363 ((((s) & 3) == BIG ? 4 : (((s) & 3) == BIG16 ? 2 : 1)))
364
365 /* This table is used by relax_frag to promote short jumps to long
366 ones where necessary. SMALL (short) jumps may be promoted to BIG
367 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
368 don't allow a short jump in a 32 bit code segment to be promoted to
369 a 16 bit offset jump because it's slower (requires data size
370 prefix), and doesn't work, unless the destination is in the bottom
371 64k of the code segment (The top 16 bits of eip are zeroed). */
372
373 const relax_typeS md_relax_table[] =
374 {
375 /* The fields are:
376 1) most positive reach of this state,
377 2) most negative reach of this state,
378 3) how many bytes this mode will have in the variable part of the frag
379 4) which index into the table to try if we can't fit into this one. */
380
381 /* UNCOND_JUMP states. */
382 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
383 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
384 /* dword jmp adds 4 bytes to frag:
385 0 extra opcode bytes, 4 displacement bytes. */
386 {0, 0, 4, 0},
387 /* word jmp adds 2 byte2 to frag:
388 0 extra opcode bytes, 2 displacement bytes. */
389 {0, 0, 2, 0},
390
391 /* COND_JUMP states. */
392 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
393 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
394 /* dword conditionals adds 5 bytes to frag:
395 1 extra opcode byte, 4 displacement bytes. */
396 {0, 0, 5, 0},
397 /* word conditionals add 3 bytes to frag:
398 1 extra opcode byte, 2 displacement bytes. */
399 {0, 0, 3, 0},
400
401 /* COND_JUMP86 states. */
402 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG)},
403 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG16)},
404 /* dword conditionals adds 5 bytes to frag:
405 1 extra opcode byte, 4 displacement bytes. */
406 {0, 0, 5, 0},
407 /* word conditionals add 4 bytes to frag:
408 1 displacement byte and a 3 byte long branch insn. */
409 {0, 0, 4, 0}
410 };
411
412 static const arch_entry cpu_arch[] = {
413 {"i8086", Cpu086 },
414 {"i186", Cpu086|Cpu186 },
415 {"i286", Cpu086|Cpu186|Cpu286 },
416 {"i386", Cpu086|Cpu186|Cpu286|Cpu386 },
417 {"i486", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486 },
418 {"i586", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
419 {"i686", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
420 {"pentium", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
421 {"pentiumpro",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
422 {"pentium4", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX|CpuSSE|CpuSSE2 },
423 {"k6", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow },
424 {"athlon", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuMMX|Cpu3dnow },
425 {"sledgehammer",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuSledgehammer|CpuMMX|Cpu3dnow|CpuSSE|CpuSSE2 },
426 {NULL, 0 }
427 };
428
429 const pseudo_typeS md_pseudo_table[] =
430 {
431 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
432 {"align", s_align_bytes, 0},
433 #else
434 {"align", s_align_ptwo, 0},
435 #endif
436 {"arch", set_cpu_arch, 0},
437 #ifndef I386COFF
438 {"bss", s_bss, 0},
439 #endif
440 {"ffloat", float_cons, 'f'},
441 {"dfloat", float_cons, 'd'},
442 {"tfloat", float_cons, 'x'},
443 {"value", cons, 2},
444 {"noopt", s_ignore, 0},
445 {"optim", s_ignore, 0},
446 {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
447 {"code16", set_code_flag, CODE_16BIT},
448 {"code32", set_code_flag, CODE_32BIT},
449 {"code64", set_code_flag, CODE_64BIT},
450 {"intel_syntax", set_intel_syntax, 1},
451 {"att_syntax", set_intel_syntax, 0},
452 {"file", (void (*) PARAMS ((int))) dwarf2_directive_file, 0},
453 {"loc", dwarf2_directive_loc, 0},
454 #ifdef TE_PE
455 {"secrel32", pe_directive_secrel, 0},
456 #endif
457 {0, 0, 0}
458 };
459
460 /* For interface with expression (). */
461 extern char *input_line_pointer;
462
463 /* Hash table for instruction mnemonic lookup. */
464 static struct hash_control *op_hash;
465
466 /* Hash table for register lookup. */
467 static struct hash_control *reg_hash;
468 \f
469 void
470 i386_align_code (fragP, count)
471 fragS *fragP;
472 int count;
473 {
474 /* Various efficient no-op patterns for aligning code labels.
475 Note: Don't try to assemble the instructions in the comments.
476 0L and 0w are not legal. */
477 static const char f32_1[] =
478 {0x90}; /* nop */
479 static const char f32_2[] =
480 {0x89,0xf6}; /* movl %esi,%esi */
481 static const char f32_3[] =
482 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
483 static const char f32_4[] =
484 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
485 static const char f32_5[] =
486 {0x90, /* nop */
487 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
488 static const char f32_6[] =
489 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
490 static const char f32_7[] =
491 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
492 static const char f32_8[] =
493 {0x90, /* nop */
494 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
495 static const char f32_9[] =
496 {0x89,0xf6, /* movl %esi,%esi */
497 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
498 static const char f32_10[] =
499 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
500 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
501 static const char f32_11[] =
502 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
503 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
504 static const char f32_12[] =
505 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
506 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
507 static const char f32_13[] =
508 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
509 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
510 static const char f32_14[] =
511 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
512 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
513 static const char f32_15[] =
514 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
515 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
516 static const char f16_3[] =
517 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
518 static const char f16_4[] =
519 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
520 static const char f16_5[] =
521 {0x90, /* nop */
522 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
523 static const char f16_6[] =
524 {0x89,0xf6, /* mov %si,%si */
525 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
526 static const char f16_7[] =
527 {0x8d,0x74,0x00, /* lea 0(%si),%si */
528 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
529 static const char f16_8[] =
530 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
531 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
532 static const char *const f32_patt[] = {
533 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
534 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
535 };
536 static const char *const f16_patt[] = {
537 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
538 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
539 };
540
541 if (count <= 0 || count > 15)
542 return;
543
544 /* The recommended way to pad 64bit code is to use NOPs preceded by
545 maximally four 0x66 prefixes. Balance the size of nops. */
546 if (flag_code == CODE_64BIT)
547 {
548 int i;
549 int nnops = (count + 3) / 4;
550 int len = count / nnops;
551 int remains = count - nnops * len;
552 int pos = 0;
553
554 for (i = 0; i < remains; i++)
555 {
556 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len);
557 fragP->fr_literal[fragP->fr_fix + pos + len] = 0x90;
558 pos += len + 1;
559 }
560 for (; i < nnops; i++)
561 {
562 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len - 1);
563 fragP->fr_literal[fragP->fr_fix + pos + len - 1] = 0x90;
564 pos += len;
565 }
566 }
567 else
568 if (flag_code == CODE_16BIT)
569 {
570 memcpy (fragP->fr_literal + fragP->fr_fix,
571 f16_patt[count - 1], count);
572 if (count > 8)
573 /* Adjust jump offset. */
574 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
575 }
576 else
577 memcpy (fragP->fr_literal + fragP->fr_fix,
578 f32_patt[count - 1], count);
579 fragP->fr_var = count;
580 }
581
582 static INLINE unsigned int
583 mode_from_disp_size (t)
584 unsigned int t;
585 {
586 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
587 }
588
589 static INLINE int
590 fits_in_signed_byte (num)
591 offsetT num;
592 {
593 return (num >= -128) && (num <= 127);
594 }
595
596 static INLINE int
597 fits_in_unsigned_byte (num)
598 offsetT num;
599 {
600 return (num & 0xff) == num;
601 }
602
603 static INLINE int
604 fits_in_unsigned_word (num)
605 offsetT num;
606 {
607 return (num & 0xffff) == num;
608 }
609
610 static INLINE int
611 fits_in_signed_word (num)
612 offsetT num;
613 {
614 return (-32768 <= num) && (num <= 32767);
615 }
616 static INLINE int
617 fits_in_signed_long (num)
618 offsetT num ATTRIBUTE_UNUSED;
619 {
620 #ifndef BFD64
621 return 1;
622 #else
623 return (!(((offsetT) -1 << 31) & num)
624 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
625 #endif
626 } /* fits_in_signed_long() */
627 static INLINE int
628 fits_in_unsigned_long (num)
629 offsetT num ATTRIBUTE_UNUSED;
630 {
631 #ifndef BFD64
632 return 1;
633 #else
634 return (num & (((offsetT) 2 << 31) - 1)) == num;
635 #endif
636 } /* fits_in_unsigned_long() */
637
638 static int
639 smallest_imm_type (num)
640 offsetT num;
641 {
642 if (cpu_arch_flags != (Cpu086 | Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64))
643 {
644 /* This code is disabled on the 486 because all the Imm1 forms
645 in the opcode table are slower on the i486. They're the
646 versions with the implicitly specified single-position
647 displacement, which has another syntax if you really want to
648 use that form. */
649 if (num == 1)
650 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
651 }
652 return (fits_in_signed_byte (num)
653 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
654 : fits_in_unsigned_byte (num)
655 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
656 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
657 ? (Imm16 | Imm32 | Imm32S | Imm64)
658 : fits_in_signed_long (num)
659 ? (Imm32 | Imm32S | Imm64)
660 : fits_in_unsigned_long (num)
661 ? (Imm32 | Imm64)
662 : Imm64);
663 }
664
665 static offsetT
666 offset_in_range (val, size)
667 offsetT val;
668 int size;
669 {
670 addressT mask;
671
672 switch (size)
673 {
674 case 1: mask = ((addressT) 1 << 8) - 1; break;
675 case 2: mask = ((addressT) 1 << 16) - 1; break;
676 case 4: mask = ((addressT) 2 << 31) - 1; break;
677 #ifdef BFD64
678 case 8: mask = ((addressT) 2 << 63) - 1; break;
679 #endif
680 default: abort ();
681 }
682
683 /* If BFD64, sign extend val. */
684 if (!use_rela_relocations)
685 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
686 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
687
688 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
689 {
690 char buf1[40], buf2[40];
691
692 sprint_value (buf1, val);
693 sprint_value (buf2, val & mask);
694 as_warn (_("%s shortened to %s"), buf1, buf2);
695 }
696 return val & mask;
697 }
698
699 /* Returns 0 if attempting to add a prefix where one from the same
700 class already exists, 1 if non rep/repne added, 2 if rep/repne
701 added. */
702 static int
703 add_prefix (prefix)
704 unsigned int prefix;
705 {
706 int ret = 1;
707 int q;
708
709 if (prefix >= REX_OPCODE && prefix < REX_OPCODE + 16
710 && flag_code == CODE_64BIT)
711 q = REX_PREFIX;
712 else
713 switch (prefix)
714 {
715 default:
716 abort ();
717
718 case CS_PREFIX_OPCODE:
719 case DS_PREFIX_OPCODE:
720 case ES_PREFIX_OPCODE:
721 case FS_PREFIX_OPCODE:
722 case GS_PREFIX_OPCODE:
723 case SS_PREFIX_OPCODE:
724 q = SEG_PREFIX;
725 break;
726
727 case REPNE_PREFIX_OPCODE:
728 case REPE_PREFIX_OPCODE:
729 ret = 2;
730 /* fall thru */
731 case LOCK_PREFIX_OPCODE:
732 q = LOCKREP_PREFIX;
733 break;
734
735 case FWAIT_OPCODE:
736 q = WAIT_PREFIX;
737 break;
738
739 case ADDR_PREFIX_OPCODE:
740 q = ADDR_PREFIX;
741 break;
742
743 case DATA_PREFIX_OPCODE:
744 q = DATA_PREFIX;
745 break;
746 }
747
748 if (i.prefix[q] != 0)
749 {
750 as_bad (_("same type of prefix used twice"));
751 return 0;
752 }
753
754 i.prefixes += 1;
755 i.prefix[q] = prefix;
756 return ret;
757 }
758
759 static void
760 set_code_flag (value)
761 int value;
762 {
763 flag_code = value;
764 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
765 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
766 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
767 {
768 as_bad (_("64bit mode not supported on this CPU."));
769 }
770 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
771 {
772 as_bad (_("32bit mode not supported on this CPU."));
773 }
774 stackop_size = '\0';
775 }
776
777 static void
778 set_16bit_gcc_code_flag (new_code_flag)
779 int new_code_flag;
780 {
781 flag_code = new_code_flag;
782 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
783 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
784 stackop_size = 'l';
785 }
786
787 static void
788 set_intel_syntax (syntax_flag)
789 int syntax_flag;
790 {
791 /* Find out if register prefixing is specified. */
792 int ask_naked_reg = 0;
793
794 SKIP_WHITESPACE ();
795 if (!is_end_of_line[(unsigned char) *input_line_pointer])
796 {
797 char *string = input_line_pointer;
798 int e = get_symbol_end ();
799
800 if (strcmp (string, "prefix") == 0)
801 ask_naked_reg = 1;
802 else if (strcmp (string, "noprefix") == 0)
803 ask_naked_reg = -1;
804 else
805 as_bad (_("bad argument to syntax directive."));
806 *input_line_pointer = e;
807 }
808 demand_empty_rest_of_line ();
809
810 intel_syntax = syntax_flag;
811
812 if (ask_naked_reg == 0)
813 allow_naked_reg = (intel_syntax
814 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
815 else
816 allow_naked_reg = (ask_naked_reg < 0);
817 }
818
819 static void
820 set_cpu_arch (dummy)
821 int dummy ATTRIBUTE_UNUSED;
822 {
823 SKIP_WHITESPACE ();
824
825 if (!is_end_of_line[(unsigned char) *input_line_pointer])
826 {
827 char *string = input_line_pointer;
828 int e = get_symbol_end ();
829 int i;
830
831 for (i = 0; cpu_arch[i].name; i++)
832 {
833 if (strcmp (string, cpu_arch[i].name) == 0)
834 {
835 cpu_arch_name = cpu_arch[i].name;
836 cpu_arch_flags = (cpu_arch[i].flags
837 | (flag_code == CODE_64BIT ? Cpu64 : CpuNo64));
838 break;
839 }
840 }
841 if (!cpu_arch[i].name)
842 as_bad (_("no such architecture: `%s'"), string);
843
844 *input_line_pointer = e;
845 }
846 else
847 as_bad (_("missing cpu architecture"));
848
849 no_cond_jump_promotion = 0;
850 if (*input_line_pointer == ','
851 && !is_end_of_line[(unsigned char) input_line_pointer[1]])
852 {
853 char *string = ++input_line_pointer;
854 int e = get_symbol_end ();
855
856 if (strcmp (string, "nojumps") == 0)
857 no_cond_jump_promotion = 1;
858 else if (strcmp (string, "jumps") == 0)
859 ;
860 else
861 as_bad (_("no such architecture modifier: `%s'"), string);
862
863 *input_line_pointer = e;
864 }
865
866 demand_empty_rest_of_line ();
867 }
868
869 unsigned long
870 i386_mach ()
871 {
872 if (!strcmp (default_arch, "x86_64"))
873 return bfd_mach_x86_64;
874 else if (!strcmp (default_arch, "i386"))
875 return bfd_mach_i386_i386;
876 else
877 as_fatal (_("Unknown architecture"));
878 }
879 \f
880 void
881 md_begin ()
882 {
883 const char *hash_err;
884
885 /* Initialize op_hash hash table. */
886 op_hash = hash_new ();
887
888 {
889 const template *optab;
890 templates *core_optab;
891
892 /* Setup for loop. */
893 optab = i386_optab;
894 core_optab = (templates *) xmalloc (sizeof (templates));
895 core_optab->start = optab;
896
897 while (1)
898 {
899 ++optab;
900 if (optab->name == NULL
901 || strcmp (optab->name, (optab - 1)->name) != 0)
902 {
903 /* different name --> ship out current template list;
904 add to hash table; & begin anew. */
905 core_optab->end = optab;
906 hash_err = hash_insert (op_hash,
907 (optab - 1)->name,
908 (PTR) core_optab);
909 if (hash_err)
910 {
911 as_fatal (_("Internal Error: Can't hash %s: %s"),
912 (optab - 1)->name,
913 hash_err);
914 }
915 if (optab->name == NULL)
916 break;
917 core_optab = (templates *) xmalloc (sizeof (templates));
918 core_optab->start = optab;
919 }
920 }
921 }
922
923 /* Initialize reg_hash hash table. */
924 reg_hash = hash_new ();
925 {
926 const reg_entry *regtab;
927
928 for (regtab = i386_regtab;
929 regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
930 regtab++)
931 {
932 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
933 if (hash_err)
934 as_fatal (_("Internal Error: Can't hash %s: %s"),
935 regtab->reg_name,
936 hash_err);
937 }
938 }
939
940 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
941 {
942 int c;
943 char *p;
944
945 for (c = 0; c < 256; c++)
946 {
947 if (ISDIGIT (c))
948 {
949 digit_chars[c] = c;
950 mnemonic_chars[c] = c;
951 register_chars[c] = c;
952 operand_chars[c] = c;
953 }
954 else if (ISLOWER (c))
955 {
956 mnemonic_chars[c] = c;
957 register_chars[c] = c;
958 operand_chars[c] = c;
959 }
960 else if (ISUPPER (c))
961 {
962 mnemonic_chars[c] = TOLOWER (c);
963 register_chars[c] = mnemonic_chars[c];
964 operand_chars[c] = c;
965 }
966
967 if (ISALPHA (c) || ISDIGIT (c))
968 identifier_chars[c] = c;
969 else if (c >= 128)
970 {
971 identifier_chars[c] = c;
972 operand_chars[c] = c;
973 }
974 }
975
976 #ifdef LEX_AT
977 identifier_chars['@'] = '@';
978 #endif
979 #ifdef LEX_QM
980 identifier_chars['?'] = '?';
981 operand_chars['?'] = '?';
982 #endif
983 digit_chars['-'] = '-';
984 identifier_chars['_'] = '_';
985 identifier_chars['.'] = '.';
986
987 for (p = operand_special_chars; *p != '\0'; p++)
988 operand_chars[(unsigned char) *p] = *p;
989 }
990
991 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
992 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
993 {
994 record_alignment (text_section, 2);
995 record_alignment (data_section, 2);
996 record_alignment (bss_section, 2);
997 }
998 #endif
999
1000 if (flag_code == CODE_64BIT)
1001 {
1002 x86_dwarf2_return_column = 16;
1003 x86_cie_data_alignment = -8;
1004 }
1005 else
1006 {
1007 x86_dwarf2_return_column = 8;
1008 x86_cie_data_alignment = -4;
1009 }
1010 }
1011
1012 void
1013 i386_print_statistics (file)
1014 FILE *file;
1015 {
1016 hash_print_statistics (file, "i386 opcode", op_hash);
1017 hash_print_statistics (file, "i386 register", reg_hash);
1018 }
1019 \f
1020 #ifdef DEBUG386
1021
1022 /* Debugging routines for md_assemble. */
1023 static void pi PARAMS ((char *, i386_insn *));
1024 static void pte PARAMS ((template *));
1025 static void pt PARAMS ((unsigned int));
1026 static void pe PARAMS ((expressionS *));
1027 static void ps PARAMS ((symbolS *));
1028
1029 static void
1030 pi (line, x)
1031 char *line;
1032 i386_insn *x;
1033 {
1034 unsigned int i;
1035
1036 fprintf (stdout, "%s: template ", line);
1037 pte (&x->tm);
1038 fprintf (stdout, " address: base %s index %s scale %x\n",
1039 x->base_reg ? x->base_reg->reg_name : "none",
1040 x->index_reg ? x->index_reg->reg_name : "none",
1041 x->log2_scale_factor);
1042 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
1043 x->rm.mode, x->rm.reg, x->rm.regmem);
1044 fprintf (stdout, " sib: base %x index %x scale %x\n",
1045 x->sib.base, x->sib.index, x->sib.scale);
1046 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
1047 (x->rex & REX_MODE64) != 0,
1048 (x->rex & REX_EXTX) != 0,
1049 (x->rex & REX_EXTY) != 0,
1050 (x->rex & REX_EXTZ) != 0);
1051 for (i = 0; i < x->operands; i++)
1052 {
1053 fprintf (stdout, " #%d: ", i + 1);
1054 pt (x->types[i]);
1055 fprintf (stdout, "\n");
1056 if (x->types[i]
1057 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
1058 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
1059 if (x->types[i] & Imm)
1060 pe (x->op[i].imms);
1061 if (x->types[i] & Disp)
1062 pe (x->op[i].disps);
1063 }
1064 }
1065
1066 static void
1067 pte (t)
1068 template *t;
1069 {
1070 unsigned int i;
1071 fprintf (stdout, " %d operands ", t->operands);
1072 fprintf (stdout, "opcode %x ", t->base_opcode);
1073 if (t->extension_opcode != None)
1074 fprintf (stdout, "ext %x ", t->extension_opcode);
1075 if (t->opcode_modifier & D)
1076 fprintf (stdout, "D");
1077 if (t->opcode_modifier & W)
1078 fprintf (stdout, "W");
1079 fprintf (stdout, "\n");
1080 for (i = 0; i < t->operands; i++)
1081 {
1082 fprintf (stdout, " #%d type ", i + 1);
1083 pt (t->operand_types[i]);
1084 fprintf (stdout, "\n");
1085 }
1086 }
1087
1088 static void
1089 pe (e)
1090 expressionS *e;
1091 {
1092 fprintf (stdout, " operation %d\n", e->X_op);
1093 fprintf (stdout, " add_number %ld (%lx)\n",
1094 (long) e->X_add_number, (long) e->X_add_number);
1095 if (e->X_add_symbol)
1096 {
1097 fprintf (stdout, " add_symbol ");
1098 ps (e->X_add_symbol);
1099 fprintf (stdout, "\n");
1100 }
1101 if (e->X_op_symbol)
1102 {
1103 fprintf (stdout, " op_symbol ");
1104 ps (e->X_op_symbol);
1105 fprintf (stdout, "\n");
1106 }
1107 }
1108
1109 static void
1110 ps (s)
1111 symbolS *s;
1112 {
1113 fprintf (stdout, "%s type %s%s",
1114 S_GET_NAME (s),
1115 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
1116 segment_name (S_GET_SEGMENT (s)));
1117 }
1118
1119 struct type_name
1120 {
1121 unsigned int mask;
1122 char *tname;
1123 }
1124
1125 static const type_names[] =
1126 {
1127 { Reg8, "r8" },
1128 { Reg16, "r16" },
1129 { Reg32, "r32" },
1130 { Reg64, "r64" },
1131 { Imm8, "i8" },
1132 { Imm8S, "i8s" },
1133 { Imm16, "i16" },
1134 { Imm32, "i32" },
1135 { Imm32S, "i32s" },
1136 { Imm64, "i64" },
1137 { Imm1, "i1" },
1138 { BaseIndex, "BaseIndex" },
1139 { Disp8, "d8" },
1140 { Disp16, "d16" },
1141 { Disp32, "d32" },
1142 { Disp32S, "d32s" },
1143 { Disp64, "d64" },
1144 { InOutPortReg, "InOutPortReg" },
1145 { ShiftCount, "ShiftCount" },
1146 { Control, "control reg" },
1147 { Test, "test reg" },
1148 { Debug, "debug reg" },
1149 { FloatReg, "FReg" },
1150 { FloatAcc, "FAcc" },
1151 { SReg2, "SReg2" },
1152 { SReg3, "SReg3" },
1153 { Acc, "Acc" },
1154 { JumpAbsolute, "Jump Absolute" },
1155 { RegMMX, "rMMX" },
1156 { RegXMM, "rXMM" },
1157 { EsSeg, "es" },
1158 { 0, "" }
1159 };
1160
1161 static void
1162 pt (t)
1163 unsigned int t;
1164 {
1165 const struct type_name *ty;
1166
1167 for (ty = type_names; ty->mask; ty++)
1168 if (t & ty->mask)
1169 fprintf (stdout, "%s, ", ty->tname);
1170 fflush (stdout);
1171 }
1172
1173 #endif /* DEBUG386 */
1174 \f
1175 static bfd_reloc_code_real_type reloc
1176 PARAMS ((int, int, int, bfd_reloc_code_real_type));
1177
1178 static bfd_reloc_code_real_type
1179 reloc (size, pcrel, sign, other)
1180 int size;
1181 int pcrel;
1182 int sign;
1183 bfd_reloc_code_real_type other;
1184 {
1185 if (other != NO_RELOC)
1186 return other;
1187
1188 if (pcrel)
1189 {
1190 if (!sign)
1191 as_bad (_("There are no unsigned pc-relative relocations"));
1192 switch (size)
1193 {
1194 case 1: return BFD_RELOC_8_PCREL;
1195 case 2: return BFD_RELOC_16_PCREL;
1196 case 4: return BFD_RELOC_32_PCREL;
1197 }
1198 as_bad (_("can not do %d byte pc-relative relocation"), size);
1199 }
1200 else
1201 {
1202 if (sign)
1203 switch (size)
1204 {
1205 case 4: return BFD_RELOC_X86_64_32S;
1206 }
1207 else
1208 switch (size)
1209 {
1210 case 1: return BFD_RELOC_8;
1211 case 2: return BFD_RELOC_16;
1212 case 4: return BFD_RELOC_32;
1213 case 8: return BFD_RELOC_64;
1214 }
1215 as_bad (_("can not do %s %d byte relocation"),
1216 sign ? "signed" : "unsigned", size);
1217 }
1218
1219 abort ();
1220 return BFD_RELOC_NONE;
1221 }
1222
1223 /* Here we decide which fixups can be adjusted to make them relative to
1224 the beginning of the section instead of the symbol. Basically we need
1225 to make sure that the dynamic relocations are done correctly, so in
1226 some cases we force the original symbol to be used. */
1227
1228 int
1229 tc_i386_fix_adjustable (fixP)
1230 fixS *fixP ATTRIBUTE_UNUSED;
1231 {
1232 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1233 if (OUTPUT_FLAVOR != bfd_target_elf_flavour)
1234 return 1;
1235
1236 /* Don't adjust pc-relative references to merge sections in 64-bit
1237 mode. */
1238 if (use_rela_relocations
1239 && (S_GET_SEGMENT (fixP->fx_addsy)->flags & SEC_MERGE) != 0
1240 && fixP->fx_pcrel)
1241 return 0;
1242
1243 /* The x86_64 GOTPCREL are represented as 32bit PCrel relocations
1244 and changed later by validate_fix. */
1245 if (GOT_symbol && fixP->fx_subsy == GOT_symbol
1246 && fixP->fx_r_type == BFD_RELOC_32_PCREL)
1247 return 0;
1248
1249 /* adjust_reloc_syms doesn't know about the GOT. */
1250 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1251 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1252 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1253 || fixP->fx_r_type == BFD_RELOC_386_TLS_GD
1254 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDM
1255 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDO_32
1256 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE_32
1257 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE
1258 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTIE
1259 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE_32
1260 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE
1261 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1262 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1263 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
1264 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSGD
1265 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSLD
1266 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF32
1267 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTTPOFF
1268 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF32
1269 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1270 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1271 return 0;
1272 #endif
1273 return 1;
1274 }
1275
1276 static int intel_float_operand PARAMS ((const char *mnemonic));
1277
1278 static int
1279 intel_float_operand (mnemonic)
1280 const char *mnemonic;
1281 {
1282 if (mnemonic[0] == 'f' && mnemonic[1] == 'i')
1283 return 2;
1284
1285 if (mnemonic[0] == 'f')
1286 return 1;
1287
1288 return 0;
1289 }
1290
1291 /* This is the guts of the machine-dependent assembler. LINE points to a
1292 machine dependent instruction. This function is supposed to emit
1293 the frags/bytes it assembles to. */
1294
1295 void
1296 md_assemble (line)
1297 char *line;
1298 {
1299 int j;
1300 char mnemonic[MAX_MNEM_SIZE];
1301
1302 /* Initialize globals. */
1303 memset (&i, '\0', sizeof (i));
1304 for (j = 0; j < MAX_OPERANDS; j++)
1305 i.reloc[j] = NO_RELOC;
1306 memset (disp_expressions, '\0', sizeof (disp_expressions));
1307 memset (im_expressions, '\0', sizeof (im_expressions));
1308 save_stack_p = save_stack;
1309
1310 /* First parse an instruction mnemonic & call i386_operand for the operands.
1311 We assume that the scrubber has arranged it so that line[0] is the valid
1312 start of a (possibly prefixed) mnemonic. */
1313
1314 line = parse_insn (line, mnemonic);
1315 if (line == NULL)
1316 return;
1317
1318 line = parse_operands (line, mnemonic);
1319 if (line == NULL)
1320 return;
1321
1322 /* Now we've parsed the mnemonic into a set of templates, and have the
1323 operands at hand. */
1324
1325 /* All intel opcodes have reversed operands except for "bound" and
1326 "enter". We also don't reverse intersegment "jmp" and "call"
1327 instructions with 2 immediate operands so that the immediate segment
1328 precedes the offset, as it does when in AT&T mode. "enter" and the
1329 intersegment "jmp" and "call" instructions are the only ones that
1330 have two immediate operands. */
1331 if (intel_syntax && i.operands > 1
1332 && (strcmp (mnemonic, "bound") != 0)
1333 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1334 swap_operands ();
1335
1336 if (i.imm_operands)
1337 optimize_imm ();
1338
1339 if (i.disp_operands)
1340 optimize_disp ();
1341
1342 /* Next, we find a template that matches the given insn,
1343 making sure the overlap of the given operands types is consistent
1344 with the template operand types. */
1345
1346 if (!match_template ())
1347 return;
1348
1349 if (intel_syntax)
1350 {
1351 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1352 if (SYSV386_COMPAT
1353 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1354 i.tm.base_opcode ^= FloatR;
1355
1356 /* Zap movzx and movsx suffix. The suffix may have been set from
1357 "word ptr" or "byte ptr" on the source operand, but we'll use
1358 the suffix later to choose the destination register. */
1359 if ((i.tm.base_opcode & ~9) == 0x0fb6)
1360 i.suffix = 0;
1361 }
1362
1363 if (i.tm.opcode_modifier & FWait)
1364 if (!add_prefix (FWAIT_OPCODE))
1365 return;
1366
1367 /* Check string instruction segment overrides. */
1368 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1369 {
1370 if (!check_string ())
1371 return;
1372 }
1373
1374 if (!process_suffix ())
1375 return;
1376
1377 /* Make still unresolved immediate matches conform to size of immediate
1378 given in i.suffix. */
1379 if (!finalize_imm ())
1380 return;
1381
1382 if (i.types[0] & Imm1)
1383 i.imm_operands = 0; /* kludge for shift insns. */
1384 if (i.types[0] & ImplicitRegister)
1385 i.reg_operands--;
1386 if (i.types[1] & ImplicitRegister)
1387 i.reg_operands--;
1388 if (i.types[2] & ImplicitRegister)
1389 i.reg_operands--;
1390
1391 if (i.tm.opcode_modifier & ImmExt)
1392 {
1393 expressionS *exp;
1394
1395 if ((i.tm.cpu_flags & CpuPNI) && i.operands > 0)
1396 {
1397 /* These Intel Prescott New Instructions have the fixed
1398 operands with an opcode suffix which is coded in the same
1399 place as an 8-bit immediate field would be. Here we check
1400 those operands and remove them afterwards. */
1401 unsigned int x;
1402
1403 for (x = 0; x < i.operands; x++)
1404 if (i.op[x].regs->reg_num != x)
1405 as_bad (_("can't use register '%%%s' as operand %d in '%s'."),
1406 i.op[x].regs->reg_name, x + 1, i.tm.name);
1407 i.operands = 0;
1408 }
1409
1410 /* These AMD 3DNow! and Intel Katmai New Instructions have an
1411 opcode suffix which is coded in the same place as an 8-bit
1412 immediate field would be. Here we fake an 8-bit immediate
1413 operand from the opcode suffix stored in tm.extension_opcode. */
1414
1415 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
1416
1417 exp = &im_expressions[i.imm_operands++];
1418 i.op[i.operands].imms = exp;
1419 i.types[i.operands++] = Imm8;
1420 exp->X_op = O_constant;
1421 exp->X_add_number = i.tm.extension_opcode;
1422 i.tm.extension_opcode = None;
1423 }
1424
1425 /* For insns with operands there are more diddles to do to the opcode. */
1426 if (i.operands)
1427 {
1428 if (!process_operands ())
1429 return;
1430 }
1431 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
1432 {
1433 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
1434 as_warn (_("translating to `%sp'"), i.tm.name);
1435 }
1436
1437 /* Handle conversion of 'int $3' --> special int3 insn. */
1438 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
1439 {
1440 i.tm.base_opcode = INT3_OPCODE;
1441 i.imm_operands = 0;
1442 }
1443
1444 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
1445 && i.op[0].disps->X_op == O_constant)
1446 {
1447 /* Convert "jmp constant" (and "call constant") to a jump (call) to
1448 the absolute address given by the constant. Since ix86 jumps and
1449 calls are pc relative, we need to generate a reloc. */
1450 i.op[0].disps->X_add_symbol = &abs_symbol;
1451 i.op[0].disps->X_op = O_symbol;
1452 }
1453
1454 if ((i.tm.opcode_modifier & Rex64) != 0)
1455 i.rex |= REX_MODE64;
1456
1457 /* For 8 bit registers we need an empty rex prefix. Also if the
1458 instruction already has a prefix, we need to convert old
1459 registers to new ones. */
1460
1461 if (((i.types[0] & Reg8) != 0
1462 && (i.op[0].regs->reg_flags & RegRex64) != 0)
1463 || ((i.types[1] & Reg8) != 0
1464 && (i.op[1].regs->reg_flags & RegRex64) != 0)
1465 || (((i.types[0] & Reg8) != 0 || (i.types[1] & Reg8) != 0)
1466 && i.rex != 0))
1467 {
1468 int x;
1469
1470 i.rex |= REX_OPCODE;
1471 for (x = 0; x < 2; x++)
1472 {
1473 /* Look for 8 bit operand that uses old registers. */
1474 if ((i.types[x] & Reg8) != 0
1475 && (i.op[x].regs->reg_flags & RegRex64) == 0)
1476 {
1477 /* In case it is "hi" register, give up. */
1478 if (i.op[x].regs->reg_num > 3)
1479 as_bad (_("can't encode register '%%%s' in an instruction requiring REX prefix."),
1480 i.op[x].regs->reg_name);
1481
1482 /* Otherwise it is equivalent to the extended register.
1483 Since the encoding doesn't change this is merely
1484 cosmetic cleanup for debug output. */
1485
1486 i.op[x].regs = i.op[x].regs + 8;
1487 }
1488 }
1489 }
1490
1491 if (i.rex != 0)
1492 add_prefix (REX_OPCODE | i.rex);
1493
1494 /* We are ready to output the insn. */
1495 output_insn ();
1496 }
1497
1498 static char *
1499 parse_insn (line, mnemonic)
1500 char *line;
1501 char *mnemonic;
1502 {
1503 char *l = line;
1504 char *token_start = l;
1505 char *mnem_p;
1506
1507 /* Non-zero if we found a prefix only acceptable with string insns. */
1508 const char *expecting_string_instruction = NULL;
1509
1510 while (1)
1511 {
1512 mnem_p = mnemonic;
1513 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1514 {
1515 mnem_p++;
1516 if (mnem_p >= mnemonic + MAX_MNEM_SIZE)
1517 {
1518 as_bad (_("no such instruction: `%s'"), token_start);
1519 return NULL;
1520 }
1521 l++;
1522 }
1523 if (!is_space_char (*l)
1524 && *l != END_OF_INSN
1525 && *l != PREFIX_SEPARATOR
1526 && *l != ',')
1527 {
1528 as_bad (_("invalid character %s in mnemonic"),
1529 output_invalid (*l));
1530 return NULL;
1531 }
1532 if (token_start == l)
1533 {
1534 if (*l == PREFIX_SEPARATOR)
1535 as_bad (_("expecting prefix; got nothing"));
1536 else
1537 as_bad (_("expecting mnemonic; got nothing"));
1538 return NULL;
1539 }
1540
1541 /* Look up instruction (or prefix) via hash table. */
1542 current_templates = hash_find (op_hash, mnemonic);
1543
1544 if (*l != END_OF_INSN
1545 && (!is_space_char (*l) || l[1] != END_OF_INSN)
1546 && current_templates
1547 && (current_templates->start->opcode_modifier & IsPrefix))
1548 {
1549 /* If we are in 16-bit mode, do not allow addr16 or data16.
1550 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1551 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1552 && flag_code != CODE_64BIT
1553 && (((current_templates->start->opcode_modifier & Size32) != 0)
1554 ^ (flag_code == CODE_16BIT)))
1555 {
1556 as_bad (_("redundant %s prefix"),
1557 current_templates->start->name);
1558 return NULL;
1559 }
1560 /* Add prefix, checking for repeated prefixes. */
1561 switch (add_prefix (current_templates->start->base_opcode))
1562 {
1563 case 0:
1564 return NULL;
1565 case 2:
1566 expecting_string_instruction = current_templates->start->name;
1567 break;
1568 }
1569 /* Skip past PREFIX_SEPARATOR and reset token_start. */
1570 token_start = ++l;
1571 }
1572 else
1573 break;
1574 }
1575
1576 if (!current_templates)
1577 {
1578 /* See if we can get a match by trimming off a suffix. */
1579 switch (mnem_p[-1])
1580 {
1581 case WORD_MNEM_SUFFIX:
1582 case BYTE_MNEM_SUFFIX:
1583 case QWORD_MNEM_SUFFIX:
1584 i.suffix = mnem_p[-1];
1585 mnem_p[-1] = '\0';
1586 current_templates = hash_find (op_hash, mnemonic);
1587 break;
1588 case SHORT_MNEM_SUFFIX:
1589 case LONG_MNEM_SUFFIX:
1590 if (!intel_syntax)
1591 {
1592 i.suffix = mnem_p[-1];
1593 mnem_p[-1] = '\0';
1594 current_templates = hash_find (op_hash, mnemonic);
1595 }
1596 break;
1597
1598 /* Intel Syntax. */
1599 case 'd':
1600 if (intel_syntax)
1601 {
1602 if (intel_float_operand (mnemonic))
1603 i.suffix = SHORT_MNEM_SUFFIX;
1604 else
1605 i.suffix = LONG_MNEM_SUFFIX;
1606 mnem_p[-1] = '\0';
1607 current_templates = hash_find (op_hash, mnemonic);
1608 }
1609 break;
1610 }
1611 if (!current_templates)
1612 {
1613 as_bad (_("no such instruction: `%s'"), token_start);
1614 return NULL;
1615 }
1616 }
1617
1618 if (current_templates->start->opcode_modifier & (Jump | JumpByte))
1619 {
1620 /* Check for a branch hint. We allow ",pt" and ",pn" for
1621 predict taken and predict not taken respectively.
1622 I'm not sure that branch hints actually do anything on loop
1623 and jcxz insns (JumpByte) for current Pentium4 chips. They
1624 may work in the future and it doesn't hurt to accept them
1625 now. */
1626 if (l[0] == ',' && l[1] == 'p')
1627 {
1628 if (l[2] == 't')
1629 {
1630 if (!add_prefix (DS_PREFIX_OPCODE))
1631 return NULL;
1632 l += 3;
1633 }
1634 else if (l[2] == 'n')
1635 {
1636 if (!add_prefix (CS_PREFIX_OPCODE))
1637 return NULL;
1638 l += 3;
1639 }
1640 }
1641 }
1642 /* Any other comma loses. */
1643 if (*l == ',')
1644 {
1645 as_bad (_("invalid character %s in mnemonic"),
1646 output_invalid (*l));
1647 return NULL;
1648 }
1649
1650 /* Check if instruction is supported on specified architecture. */
1651 if ((current_templates->start->cpu_flags & ~(Cpu64 | CpuNo64))
1652 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64)))
1653 {
1654 as_warn (_("`%s' is not supported on `%s'"),
1655 current_templates->start->name, cpu_arch_name);
1656 }
1657 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
1658 {
1659 as_warn (_("use .code16 to ensure correct addressing mode"));
1660 }
1661
1662 /* Check for rep/repne without a string instruction. */
1663 if (expecting_string_instruction
1664 && !(current_templates->start->opcode_modifier & IsString))
1665 {
1666 as_bad (_("expecting string instruction after `%s'"),
1667 expecting_string_instruction);
1668 return NULL;
1669 }
1670
1671 return l;
1672 }
1673
1674 static char *
1675 parse_operands (l, mnemonic)
1676 char *l;
1677 const char *mnemonic;
1678 {
1679 char *token_start;
1680
1681 /* 1 if operand is pending after ','. */
1682 unsigned int expecting_operand = 0;
1683
1684 /* Non-zero if operand parens not balanced. */
1685 unsigned int paren_not_balanced;
1686
1687 while (*l != END_OF_INSN)
1688 {
1689 /* Skip optional white space before operand. */
1690 if (is_space_char (*l))
1691 ++l;
1692 if (!is_operand_char (*l) && *l != END_OF_INSN)
1693 {
1694 as_bad (_("invalid character %s before operand %d"),
1695 output_invalid (*l),
1696 i.operands + 1);
1697 return NULL;
1698 }
1699 token_start = l; /* after white space */
1700 paren_not_balanced = 0;
1701 while (paren_not_balanced || *l != ',')
1702 {
1703 if (*l == END_OF_INSN)
1704 {
1705 if (paren_not_balanced)
1706 {
1707 if (!intel_syntax)
1708 as_bad (_("unbalanced parenthesis in operand %d."),
1709 i.operands + 1);
1710 else
1711 as_bad (_("unbalanced brackets in operand %d."),
1712 i.operands + 1);
1713 return NULL;
1714 }
1715 else
1716 break; /* we are done */
1717 }
1718 else if (!is_operand_char (*l) && !is_space_char (*l))
1719 {
1720 as_bad (_("invalid character %s in operand %d"),
1721 output_invalid (*l),
1722 i.operands + 1);
1723 return NULL;
1724 }
1725 if (!intel_syntax)
1726 {
1727 if (*l == '(')
1728 ++paren_not_balanced;
1729 if (*l == ')')
1730 --paren_not_balanced;
1731 }
1732 else
1733 {
1734 if (*l == '[')
1735 ++paren_not_balanced;
1736 if (*l == ']')
1737 --paren_not_balanced;
1738 }
1739 l++;
1740 }
1741 if (l != token_start)
1742 { /* Yes, we've read in another operand. */
1743 unsigned int operand_ok;
1744 this_operand = i.operands++;
1745 if (i.operands > MAX_OPERANDS)
1746 {
1747 as_bad (_("spurious operands; (%d operands/instruction max)"),
1748 MAX_OPERANDS);
1749 return NULL;
1750 }
1751 /* Now parse operand adding info to 'i' as we go along. */
1752 END_STRING_AND_SAVE (l);
1753
1754 if (intel_syntax)
1755 operand_ok =
1756 i386_intel_operand (token_start,
1757 intel_float_operand (mnemonic));
1758 else
1759 operand_ok = i386_operand (token_start);
1760
1761 RESTORE_END_STRING (l);
1762 if (!operand_ok)
1763 return NULL;
1764 }
1765 else
1766 {
1767 if (expecting_operand)
1768 {
1769 expecting_operand_after_comma:
1770 as_bad (_("expecting operand after ','; got nothing"));
1771 return NULL;
1772 }
1773 if (*l == ',')
1774 {
1775 as_bad (_("expecting operand before ','; got nothing"));
1776 return NULL;
1777 }
1778 }
1779
1780 /* Now *l must be either ',' or END_OF_INSN. */
1781 if (*l == ',')
1782 {
1783 if (*++l == END_OF_INSN)
1784 {
1785 /* Just skip it, if it's \n complain. */
1786 goto expecting_operand_after_comma;
1787 }
1788 expecting_operand = 1;
1789 }
1790 }
1791 return l;
1792 }
1793
1794 static void
1795 swap_operands ()
1796 {
1797 union i386_op temp_op;
1798 unsigned int temp_type;
1799 enum bfd_reloc_code_real temp_reloc;
1800 int xchg1 = 0;
1801 int xchg2 = 0;
1802
1803 if (i.operands == 2)
1804 {
1805 xchg1 = 0;
1806 xchg2 = 1;
1807 }
1808 else if (i.operands == 3)
1809 {
1810 xchg1 = 0;
1811 xchg2 = 2;
1812 }
1813 temp_type = i.types[xchg2];
1814 i.types[xchg2] = i.types[xchg1];
1815 i.types[xchg1] = temp_type;
1816 temp_op = i.op[xchg2];
1817 i.op[xchg2] = i.op[xchg1];
1818 i.op[xchg1] = temp_op;
1819 temp_reloc = i.reloc[xchg2];
1820 i.reloc[xchg2] = i.reloc[xchg1];
1821 i.reloc[xchg1] = temp_reloc;
1822
1823 if (i.mem_operands == 2)
1824 {
1825 const seg_entry *temp_seg;
1826 temp_seg = i.seg[0];
1827 i.seg[0] = i.seg[1];
1828 i.seg[1] = temp_seg;
1829 }
1830 }
1831
1832 /* Try to ensure constant immediates are represented in the smallest
1833 opcode possible. */
1834 static void
1835 optimize_imm ()
1836 {
1837 char guess_suffix = 0;
1838 int op;
1839
1840 if (i.suffix)
1841 guess_suffix = i.suffix;
1842 else if (i.reg_operands)
1843 {
1844 /* Figure out a suffix from the last register operand specified.
1845 We can't do this properly yet, ie. excluding InOutPortReg,
1846 but the following works for instructions with immediates.
1847 In any case, we can't set i.suffix yet. */
1848 for (op = i.operands; --op >= 0;)
1849 if (i.types[op] & Reg)
1850 {
1851 if (i.types[op] & Reg8)
1852 guess_suffix = BYTE_MNEM_SUFFIX;
1853 else if (i.types[op] & Reg16)
1854 guess_suffix = WORD_MNEM_SUFFIX;
1855 else if (i.types[op] & Reg32)
1856 guess_suffix = LONG_MNEM_SUFFIX;
1857 else if (i.types[op] & Reg64)
1858 guess_suffix = QWORD_MNEM_SUFFIX;
1859 break;
1860 }
1861 }
1862 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
1863 guess_suffix = WORD_MNEM_SUFFIX;
1864
1865 for (op = i.operands; --op >= 0;)
1866 if (i.types[op] & Imm)
1867 {
1868 switch (i.op[op].imms->X_op)
1869 {
1870 case O_constant:
1871 /* If a suffix is given, this operand may be shortened. */
1872 switch (guess_suffix)
1873 {
1874 case LONG_MNEM_SUFFIX:
1875 i.types[op] |= Imm32 | Imm64;
1876 break;
1877 case WORD_MNEM_SUFFIX:
1878 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
1879 break;
1880 case BYTE_MNEM_SUFFIX:
1881 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
1882 break;
1883 }
1884
1885 /* If this operand is at most 16 bits, convert it
1886 to a signed 16 bit number before trying to see
1887 whether it will fit in an even smaller size.
1888 This allows a 16-bit operand such as $0xffe0 to
1889 be recognised as within Imm8S range. */
1890 if ((i.types[op] & Imm16)
1891 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
1892 {
1893 i.op[op].imms->X_add_number =
1894 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
1895 }
1896 if ((i.types[op] & Imm32)
1897 && ((i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1))
1898 == 0))
1899 {
1900 i.op[op].imms->X_add_number = ((i.op[op].imms->X_add_number
1901 ^ ((offsetT) 1 << 31))
1902 - ((offsetT) 1 << 31));
1903 }
1904 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
1905
1906 /* We must avoid matching of Imm32 templates when 64bit
1907 only immediate is available. */
1908 if (guess_suffix == QWORD_MNEM_SUFFIX)
1909 i.types[op] &= ~Imm32;
1910 break;
1911
1912 case O_absent:
1913 case O_register:
1914 abort ();
1915
1916 /* Symbols and expressions. */
1917 default:
1918 /* Convert symbolic operand to proper sizes for matching. */
1919 switch (guess_suffix)
1920 {
1921 case QWORD_MNEM_SUFFIX:
1922 i.types[op] = Imm64 | Imm32S;
1923 break;
1924 case LONG_MNEM_SUFFIX:
1925 i.types[op] = Imm32;
1926 break;
1927 case WORD_MNEM_SUFFIX:
1928 i.types[op] = Imm16;
1929 break;
1930 case BYTE_MNEM_SUFFIX:
1931 i.types[op] = Imm8 | Imm8S;
1932 break;
1933 }
1934 break;
1935 }
1936 }
1937 }
1938
1939 /* Try to use the smallest displacement type too. */
1940 static void
1941 optimize_disp ()
1942 {
1943 int op;
1944
1945 for (op = i.operands; --op >= 0;)
1946 if ((i.types[op] & Disp) && i.op[op].disps->X_op == O_constant)
1947 {
1948 offsetT disp = i.op[op].disps->X_add_number;
1949
1950 if (i.types[op] & Disp16)
1951 {
1952 /* We know this operand is at most 16 bits, so
1953 convert to a signed 16 bit number before trying
1954 to see whether it will fit in an even smaller
1955 size. */
1956
1957 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
1958 }
1959 else if (i.types[op] & Disp32)
1960 {
1961 /* We know this operand is at most 32 bits, so convert to a
1962 signed 32 bit number before trying to see whether it will
1963 fit in an even smaller size. */
1964 disp &= (((offsetT) 2 << 31) - 1);
1965 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1966 }
1967 if (flag_code == CODE_64BIT)
1968 {
1969 if (fits_in_signed_long (disp))
1970 i.types[op] |= Disp32S;
1971 if (fits_in_unsigned_long (disp))
1972 i.types[op] |= Disp32;
1973 }
1974 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
1975 && fits_in_signed_byte (disp))
1976 i.types[op] |= Disp8;
1977 }
1978 }
1979
1980 static int
1981 match_template ()
1982 {
1983 /* Points to template once we've found it. */
1984 const template *t;
1985 unsigned int overlap0, overlap1, overlap2;
1986 unsigned int found_reverse_match;
1987 int suffix_check;
1988
1989 #define MATCH(overlap, given, template) \
1990 ((overlap & ~JumpAbsolute) \
1991 && (((given) & (BaseIndex | JumpAbsolute)) \
1992 == ((overlap) & (BaseIndex | JumpAbsolute))))
1993
1994 /* If given types r0 and r1 are registers they must be of the same type
1995 unless the expected operand type register overlap is null.
1996 Note that Acc in a template matches every size of reg. */
1997 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
1998 (((g0) & Reg) == 0 || ((g1) & Reg) == 0 \
1999 || ((g0) & Reg) == ((g1) & Reg) \
2000 || ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
2001
2002 overlap0 = 0;
2003 overlap1 = 0;
2004 overlap2 = 0;
2005 found_reverse_match = 0;
2006 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
2007 ? No_bSuf
2008 : (i.suffix == WORD_MNEM_SUFFIX
2009 ? No_wSuf
2010 : (i.suffix == SHORT_MNEM_SUFFIX
2011 ? No_sSuf
2012 : (i.suffix == LONG_MNEM_SUFFIX
2013 ? No_lSuf
2014 : (i.suffix == QWORD_MNEM_SUFFIX
2015 ? No_qSuf
2016 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX
2017 ? No_xSuf : 0))))));
2018
2019 t = current_templates->start;
2020 if (i.suffix == QWORD_MNEM_SUFFIX
2021 && flag_code != CODE_64BIT
2022 && (!intel_syntax
2023 || (!(t->opcode_modifier & IgnoreSize)
2024 && ! intel_float_operand (t->name)))
2025 && (!(t->operand_types[0] & (RegMMX | RegXMM))
2026 || !(t->operand_types[t->operands > 1] & (RegMMX | RegXMM)))
2027 && (t->base_opcode != 0x0fc7
2028 || t->extension_opcode != 1 /* cmpxchg8b */))
2029 t = current_templates->end;
2030 for (; t < current_templates->end; t++)
2031 {
2032 /* Must have right number of operands. */
2033 if (i.operands != t->operands)
2034 continue;
2035
2036 /* Check the suffix, except for some instructions in intel mode. */
2037 if ((t->opcode_modifier & suffix_check)
2038 && !(intel_syntax
2039 && (t->opcode_modifier & IgnoreSize))
2040 && !(intel_syntax
2041 && t->base_opcode == 0xd9
2042 && (t->extension_opcode == 5 /* 0xd9,5 "fldcw" */
2043 || t->extension_opcode == 7))) /* 0xd9,7 "f{n}stcw" */
2044 continue;
2045
2046 /* Do not verify operands when there are none. */
2047 else if (!t->operands)
2048 {
2049 if (t->cpu_flags & ~cpu_arch_flags)
2050 continue;
2051 /* We've found a match; break out of loop. */
2052 break;
2053 }
2054
2055 overlap0 = i.types[0] & t->operand_types[0];
2056 switch (t->operands)
2057 {
2058 case 1:
2059 if (!MATCH (overlap0, i.types[0], t->operand_types[0]))
2060 continue;
2061 break;
2062 case 2:
2063 case 3:
2064 overlap1 = i.types[1] & t->operand_types[1];
2065 if (!MATCH (overlap0, i.types[0], t->operand_types[0])
2066 || !MATCH (overlap1, i.types[1], t->operand_types[1])
2067 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2068 t->operand_types[0],
2069 overlap1, i.types[1],
2070 t->operand_types[1]))
2071 {
2072 /* Check if other direction is valid ... */
2073 if ((t->opcode_modifier & (D | FloatD)) == 0)
2074 continue;
2075
2076 /* Try reversing direction of operands. */
2077 overlap0 = i.types[0] & t->operand_types[1];
2078 overlap1 = i.types[1] & t->operand_types[0];
2079 if (!MATCH (overlap0, i.types[0], t->operand_types[1])
2080 || !MATCH (overlap1, i.types[1], t->operand_types[0])
2081 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2082 t->operand_types[1],
2083 overlap1, i.types[1],
2084 t->operand_types[0]))
2085 {
2086 /* Does not match either direction. */
2087 continue;
2088 }
2089 /* found_reverse_match holds which of D or FloatDR
2090 we've found. */
2091 found_reverse_match = t->opcode_modifier & (D | FloatDR);
2092 }
2093 /* Found a forward 2 operand match here. */
2094 else if (t->operands == 3)
2095 {
2096 /* Here we make use of the fact that there are no
2097 reverse match 3 operand instructions, and all 3
2098 operand instructions only need to be checked for
2099 register consistency between operands 2 and 3. */
2100 overlap2 = i.types[2] & t->operand_types[2];
2101 if (!MATCH (overlap2, i.types[2], t->operand_types[2])
2102 || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1],
2103 t->operand_types[1],
2104 overlap2, i.types[2],
2105 t->operand_types[2]))
2106
2107 continue;
2108 }
2109 /* Found either forward/reverse 2 or 3 operand match here:
2110 slip through to break. */
2111 }
2112 if (t->cpu_flags & ~cpu_arch_flags)
2113 {
2114 found_reverse_match = 0;
2115 continue;
2116 }
2117 /* We've found a match; break out of loop. */
2118 break;
2119 }
2120
2121 if (t == current_templates->end)
2122 {
2123 /* We found no match. */
2124 as_bad (_("suffix or operands invalid for `%s'"),
2125 current_templates->start->name);
2126 return 0;
2127 }
2128
2129 if (!quiet_warnings)
2130 {
2131 if (!intel_syntax
2132 && ((i.types[0] & JumpAbsolute)
2133 != (t->operand_types[0] & JumpAbsolute)))
2134 {
2135 as_warn (_("indirect %s without `*'"), t->name);
2136 }
2137
2138 if ((t->opcode_modifier & (IsPrefix | IgnoreSize))
2139 == (IsPrefix | IgnoreSize))
2140 {
2141 /* Warn them that a data or address size prefix doesn't
2142 affect assembly of the next line of code. */
2143 as_warn (_("stand-alone `%s' prefix"), t->name);
2144 }
2145 }
2146
2147 /* Copy the template we found. */
2148 i.tm = *t;
2149 if (found_reverse_match)
2150 {
2151 /* If we found a reverse match we must alter the opcode
2152 direction bit. found_reverse_match holds bits to change
2153 (different for int & float insns). */
2154
2155 i.tm.base_opcode ^= found_reverse_match;
2156
2157 i.tm.operand_types[0] = t->operand_types[1];
2158 i.tm.operand_types[1] = t->operand_types[0];
2159 }
2160
2161 return 1;
2162 }
2163
2164 static int
2165 check_string ()
2166 {
2167 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
2168 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
2169 {
2170 if (i.seg[0] != NULL && i.seg[0] != &es)
2171 {
2172 as_bad (_("`%s' operand %d must use `%%es' segment"),
2173 i.tm.name,
2174 mem_op + 1);
2175 return 0;
2176 }
2177 /* There's only ever one segment override allowed per instruction.
2178 This instruction possibly has a legal segment override on the
2179 second operand, so copy the segment to where non-string
2180 instructions store it, allowing common code. */
2181 i.seg[0] = i.seg[1];
2182 }
2183 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
2184 {
2185 if (i.seg[1] != NULL && i.seg[1] != &es)
2186 {
2187 as_bad (_("`%s' operand %d must use `%%es' segment"),
2188 i.tm.name,
2189 mem_op + 2);
2190 return 0;
2191 }
2192 }
2193 return 1;
2194 }
2195
2196 static int
2197 process_suffix (void)
2198 {
2199 /* If matched instruction specifies an explicit instruction mnemonic
2200 suffix, use it. */
2201 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
2202 {
2203 if (i.tm.opcode_modifier & Size16)
2204 i.suffix = WORD_MNEM_SUFFIX;
2205 else if (i.tm.opcode_modifier & Size64)
2206 i.suffix = QWORD_MNEM_SUFFIX;
2207 else
2208 i.suffix = LONG_MNEM_SUFFIX;
2209 }
2210 else if (i.reg_operands)
2211 {
2212 /* If there's no instruction mnemonic suffix we try to invent one
2213 based on register operands. */
2214 if (!i.suffix)
2215 {
2216 /* We take i.suffix from the last register operand specified,
2217 Destination register type is more significant than source
2218 register type. */
2219 int op;
2220
2221 for (op = i.operands; --op >= 0;)
2222 if ((i.types[op] & Reg)
2223 && !(i.tm.operand_types[op] & InOutPortReg))
2224 {
2225 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
2226 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
2227 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
2228 LONG_MNEM_SUFFIX);
2229 break;
2230 }
2231 }
2232 else if (i.suffix == BYTE_MNEM_SUFFIX)
2233 {
2234 if (!check_byte_reg ())
2235 return 0;
2236 }
2237 else if (i.suffix == LONG_MNEM_SUFFIX)
2238 {
2239 if (!check_long_reg ())
2240 return 0;
2241 }
2242 else if (i.suffix == QWORD_MNEM_SUFFIX)
2243 {
2244 if (!check_qword_reg ())
2245 return 0;
2246 }
2247 else if (i.suffix == WORD_MNEM_SUFFIX)
2248 {
2249 if (!check_word_reg ())
2250 return 0;
2251 }
2252 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2253 /* Do nothing if the instruction is going to ignore the prefix. */
2254 ;
2255 else
2256 abort ();
2257 }
2258 else if ((i.tm.opcode_modifier & DefaultSize) && !i.suffix)
2259 {
2260 i.suffix = stackop_size;
2261 if (i.suffix == QWORD_MNEM_SUFFIX
2262 && (i.tm.opcode_modifier & No_qSuf))
2263 i.suffix = LONG_MNEM_SUFFIX;
2264 }
2265
2266 /* Change the opcode based on the operand size given by i.suffix;
2267 We need not change things for byte insns. */
2268 if (!i.suffix && (i.tm.opcode_modifier & W))
2269 {
2270 as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
2271 return 0;
2272 }
2273
2274 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2275 {
2276 /* It's not a byte, select word/dword operation. */
2277 if (i.tm.opcode_modifier & W)
2278 {
2279 if (i.tm.opcode_modifier & ShortForm)
2280 i.tm.base_opcode |= 8;
2281 else
2282 i.tm.base_opcode |= 1;
2283 }
2284
2285 /* Now select between word & dword operations via the operand
2286 size prefix, except for instructions that will ignore this
2287 prefix anyway. */
2288 if (i.suffix != QWORD_MNEM_SUFFIX
2289 && !(i.tm.opcode_modifier & IgnoreSize)
2290 && ((i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2291 || (flag_code == CODE_64BIT
2292 && (i.tm.opcode_modifier & JumpByte))))
2293 {
2294 unsigned int prefix = DATA_PREFIX_OPCODE;
2295
2296 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2297 prefix = ADDR_PREFIX_OPCODE;
2298
2299 if (!add_prefix (prefix))
2300 return 0;
2301 }
2302
2303 /* Set mode64 for an operand. */
2304 if (i.suffix == QWORD_MNEM_SUFFIX
2305 && flag_code == CODE_64BIT
2306 && (i.tm.opcode_modifier & NoRex64) == 0)
2307 i.rex |= REX_MODE64;
2308
2309 /* Size floating point instruction. */
2310 if (i.suffix == LONG_MNEM_SUFFIX)
2311 if (i.tm.opcode_modifier & FloatMF)
2312 i.tm.base_opcode ^= 4;
2313 }
2314
2315 return 1;
2316 }
2317
2318 static int
2319 check_byte_reg (void)
2320 {
2321 int op;
2322
2323 for (op = i.operands; --op >= 0;)
2324 {
2325 /* If this is an eight bit register, it's OK. If it's the 16 or
2326 32 bit version of an eight bit register, we will just use the
2327 low portion, and that's OK too. */
2328 if (i.types[op] & Reg8)
2329 continue;
2330
2331 /* movzx and movsx should not generate this warning. */
2332 if (intel_syntax
2333 && (i.tm.base_opcode == 0xfb7
2334 || i.tm.base_opcode == 0xfb6
2335 || i.tm.base_opcode == 0x63
2336 || i.tm.base_opcode == 0xfbe
2337 || i.tm.base_opcode == 0xfbf))
2338 continue;
2339
2340 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4
2341 #if 0
2342 /* Check that the template allows eight bit regs. This
2343 kills insns such as `orb $1,%edx', which maybe should be
2344 allowed. */
2345 && (i.tm.operand_types[op] & (Reg8 | InOutPortReg))
2346 #endif
2347 )
2348 {
2349 /* Prohibit these changes in the 64bit mode, since the
2350 lowering is more complicated. */
2351 if (flag_code == CODE_64BIT
2352 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2353 {
2354 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2355 i.op[op].regs->reg_name,
2356 i.suffix);
2357 return 0;
2358 }
2359 #if REGISTER_WARNINGS
2360 if (!quiet_warnings
2361 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2362 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2363 (i.op[op].regs + (i.types[op] & Reg16
2364 ? REGNAM_AL - REGNAM_AX
2365 : REGNAM_AL - REGNAM_EAX))->reg_name,
2366 i.op[op].regs->reg_name,
2367 i.suffix);
2368 #endif
2369 continue;
2370 }
2371 /* Any other register is bad. */
2372 if (i.types[op] & (Reg | RegMMX | RegXMM
2373 | SReg2 | SReg3
2374 | Control | Debug | Test
2375 | FloatReg | FloatAcc))
2376 {
2377 as_bad (_("`%%%s' not allowed with `%s%c'"),
2378 i.op[op].regs->reg_name,
2379 i.tm.name,
2380 i.suffix);
2381 return 0;
2382 }
2383 }
2384 return 1;
2385 }
2386
2387 static int
2388 check_long_reg ()
2389 {
2390 int op;
2391
2392 for (op = i.operands; --op >= 0;)
2393 /* Reject eight bit registers, except where the template requires
2394 them. (eg. movzb) */
2395 if ((i.types[op] & Reg8) != 0
2396 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2397 {
2398 as_bad (_("`%%%s' not allowed with `%s%c'"),
2399 i.op[op].regs->reg_name,
2400 i.tm.name,
2401 i.suffix);
2402 return 0;
2403 }
2404 /* Warn if the e prefix on a general reg is missing. */
2405 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2406 && (i.types[op] & Reg16) != 0
2407 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2408 {
2409 /* Prohibit these changes in the 64bit mode, since the
2410 lowering is more complicated. */
2411 if (flag_code == CODE_64BIT)
2412 {
2413 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2414 i.op[op].regs->reg_name,
2415 i.suffix);
2416 return 0;
2417 }
2418 #if REGISTER_WARNINGS
2419 else
2420 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2421 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
2422 i.op[op].regs->reg_name,
2423 i.suffix);
2424 #endif
2425 }
2426 /* Warn if the r prefix on a general reg is missing. */
2427 else if ((i.types[op] & Reg64) != 0
2428 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2429 {
2430 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2431 i.op[op].regs->reg_name,
2432 i.suffix);
2433 return 0;
2434 }
2435 return 1;
2436 }
2437
2438 static int
2439 check_qword_reg ()
2440 {
2441 int op;
2442
2443 for (op = i.operands; --op >= 0; )
2444 /* Reject eight bit registers, except where the template requires
2445 them. (eg. movzb) */
2446 if ((i.types[op] & Reg8) != 0
2447 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2448 {
2449 as_bad (_("`%%%s' not allowed with `%s%c'"),
2450 i.op[op].regs->reg_name,
2451 i.tm.name,
2452 i.suffix);
2453 return 0;
2454 }
2455 /* Warn if the e prefix on a general reg is missing. */
2456 else if (((i.types[op] & Reg16) != 0
2457 || (i.types[op] & Reg32) != 0)
2458 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2459 {
2460 /* Prohibit these changes in the 64bit mode, since the
2461 lowering is more complicated. */
2462 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2463 i.op[op].regs->reg_name,
2464 i.suffix);
2465 return 0;
2466 }
2467 return 1;
2468 }
2469
2470 static int
2471 check_word_reg ()
2472 {
2473 int op;
2474 for (op = i.operands; --op >= 0;)
2475 /* Reject eight bit registers, except where the template requires
2476 them. (eg. movzb) */
2477 if ((i.types[op] & Reg8) != 0
2478 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2479 {
2480 as_bad (_("`%%%s' not allowed with `%s%c'"),
2481 i.op[op].regs->reg_name,
2482 i.tm.name,
2483 i.suffix);
2484 return 0;
2485 }
2486 /* Warn if the e prefix on a general reg is present. */
2487 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2488 && (i.types[op] & Reg32) != 0
2489 && (i.tm.operand_types[op] & (Reg16 | Acc)) != 0)
2490 {
2491 /* Prohibit these changes in the 64bit mode, since the
2492 lowering is more complicated. */
2493 if (flag_code == CODE_64BIT)
2494 {
2495 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2496 i.op[op].regs->reg_name,
2497 i.suffix);
2498 return 0;
2499 }
2500 else
2501 #if REGISTER_WARNINGS
2502 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2503 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
2504 i.op[op].regs->reg_name,
2505 i.suffix);
2506 #endif
2507 }
2508 return 1;
2509 }
2510
2511 static int
2512 finalize_imm ()
2513 {
2514 unsigned int overlap0, overlap1, overlap2;
2515
2516 overlap0 = i.types[0] & i.tm.operand_types[0];
2517 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64))
2518 && overlap0 != Imm8 && overlap0 != Imm8S
2519 && overlap0 != Imm16 && overlap0 != Imm32S
2520 && overlap0 != Imm32 && overlap0 != Imm64)
2521 {
2522 if (i.suffix)
2523 {
2524 overlap0 &= (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)));
2531 }
2532 else if (overlap0 == (Imm16 | Imm32S | Imm32)
2533 || overlap0 == (Imm16 | Imm32)
2534 || overlap0 == (Imm16 | Imm32S))
2535 {
2536 overlap0 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
2537 ? Imm16 : Imm32S);
2538 }
2539 if (overlap0 != Imm8 && overlap0 != Imm8S
2540 && overlap0 != Imm16 && overlap0 != Imm32S
2541 && overlap0 != Imm32 && overlap0 != Imm64)
2542 {
2543 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
2544 return 0;
2545 }
2546 }
2547 i.types[0] = overlap0;
2548
2549 overlap1 = i.types[1] & i.tm.operand_types[1];
2550 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32))
2551 && overlap1 != Imm8 && overlap1 != Imm8S
2552 && overlap1 != Imm16 && overlap1 != Imm32S
2553 && overlap1 != Imm32 && overlap1 != Imm64)
2554 {
2555 if (i.suffix)
2556 {
2557 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX
2558 ? Imm8 | Imm8S
2559 : (i.suffix == WORD_MNEM_SUFFIX
2560 ? Imm16
2561 : (i.suffix == QWORD_MNEM_SUFFIX
2562 ? Imm64 | Imm32S
2563 : Imm32)));
2564 }
2565 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
2566 || overlap1 == (Imm16 | Imm32)
2567 || overlap1 == (Imm16 | Imm32S))
2568 {
2569 overlap1 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
2570 ? Imm16 : Imm32S);
2571 }
2572 if (overlap1 != Imm8 && overlap1 != Imm8S
2573 && overlap1 != Imm16 && overlap1 != Imm32S
2574 && overlap1 != Imm32 && overlap1 != Imm64)
2575 {
2576 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
2577 return 0;
2578 }
2579 }
2580 i.types[1] = overlap1;
2581
2582 overlap2 = i.types[2] & i.tm.operand_types[2];
2583 assert ((overlap2 & Imm) == 0);
2584 i.types[2] = overlap2;
2585
2586 return 1;
2587 }
2588
2589 static int
2590 process_operands ()
2591 {
2592 /* Default segment register this instruction will use for memory
2593 accesses. 0 means unknown. This is only for optimizing out
2594 unnecessary segment overrides. */
2595 const seg_entry *default_seg = 0;
2596
2597 /* The imul $imm, %reg instruction is converted into
2598 imul $imm, %reg, %reg, and the clr %reg instruction
2599 is converted into xor %reg, %reg. */
2600 if (i.tm.opcode_modifier & regKludge)
2601 {
2602 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
2603 /* Pretend we saw the extra register operand. */
2604 assert (i.op[first_reg_op + 1].regs == 0);
2605 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
2606 i.types[first_reg_op + 1] = i.types[first_reg_op];
2607 i.reg_operands = 2;
2608 }
2609
2610 if (i.tm.opcode_modifier & ShortForm)
2611 {
2612 /* The register or float register operand is in operand 0 or 1. */
2613 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
2614 /* Register goes in low 3 bits of opcode. */
2615 i.tm.base_opcode |= i.op[op].regs->reg_num;
2616 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2617 i.rex |= REX_EXTZ;
2618 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2619 {
2620 /* Warn about some common errors, but press on regardless.
2621 The first case can be generated by gcc (<= 2.8.1). */
2622 if (i.operands == 2)
2623 {
2624 /* Reversed arguments on faddp, fsubp, etc. */
2625 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
2626 i.op[1].regs->reg_name,
2627 i.op[0].regs->reg_name);
2628 }
2629 else
2630 {
2631 /* Extraneous `l' suffix on fp insn. */
2632 as_warn (_("translating to `%s %%%s'"), i.tm.name,
2633 i.op[0].regs->reg_name);
2634 }
2635 }
2636 }
2637 else if (i.tm.opcode_modifier & Modrm)
2638 {
2639 /* The opcode is completed (modulo i.tm.extension_opcode which
2640 must be put into the modrm byte). Now, we make the modrm and
2641 index base bytes based on all the info we've collected. */
2642
2643 default_seg = build_modrm_byte ();
2644 }
2645 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
2646 {
2647 if (i.tm.base_opcode == POP_SEG_SHORT
2648 && i.op[0].regs->reg_num == 1)
2649 {
2650 as_bad (_("you can't `pop %%cs'"));
2651 return 0;
2652 }
2653 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
2654 if ((i.op[0].regs->reg_flags & RegRex) != 0)
2655 i.rex |= REX_EXTZ;
2656 }
2657 else if ((i.tm.base_opcode & ~(D | W)) == MOV_AX_DISP32)
2658 {
2659 default_seg = &ds;
2660 }
2661 else if ((i.tm.opcode_modifier & IsString) != 0)
2662 {
2663 /* For the string instructions that allow a segment override
2664 on one of their operands, the default segment is ds. */
2665 default_seg = &ds;
2666 }
2667
2668 if (i.tm.base_opcode == 0x8d /* lea */ && i.seg[0] && !quiet_warnings)
2669 as_warn (_("segment override on `lea' is ineffectual"));
2670
2671 /* If a segment was explicitly specified, and the specified segment
2672 is not the default, use an opcode prefix to select it. If we
2673 never figured out what the default segment is, then default_seg
2674 will be zero at this point, and the specified segment prefix will
2675 always be used. */
2676 if ((i.seg[0]) && (i.seg[0] != default_seg))
2677 {
2678 if (!add_prefix (i.seg[0]->seg_prefix))
2679 return 0;
2680 }
2681 return 1;
2682 }
2683
2684 static const seg_entry *
2685 build_modrm_byte ()
2686 {
2687 const seg_entry *default_seg = 0;
2688
2689 /* i.reg_operands MUST be the number of real register operands;
2690 implicit registers do not count. */
2691 if (i.reg_operands == 2)
2692 {
2693 unsigned int source, dest;
2694 source = ((i.types[0]
2695 & (Reg | RegMMX | RegXMM
2696 | SReg2 | SReg3
2697 | Control | Debug | Test))
2698 ? 0 : 1);
2699 dest = source + 1;
2700
2701 i.rm.mode = 3;
2702 /* One of the register operands will be encoded in the i.tm.reg
2703 field, the other in the combined i.tm.mode and i.tm.regmem
2704 fields. If no form of this instruction supports a memory
2705 destination operand, then we assume the source operand may
2706 sometimes be a memory operand and so we need to store the
2707 destination in the i.rm.reg field. */
2708 if ((i.tm.operand_types[dest] & AnyMem) == 0)
2709 {
2710 i.rm.reg = i.op[dest].regs->reg_num;
2711 i.rm.regmem = i.op[source].regs->reg_num;
2712 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
2713 i.rex |= REX_EXTX;
2714 if ((i.op[source].regs->reg_flags & RegRex) != 0)
2715 i.rex |= REX_EXTZ;
2716 }
2717 else
2718 {
2719 i.rm.reg = i.op[source].regs->reg_num;
2720 i.rm.regmem = i.op[dest].regs->reg_num;
2721 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
2722 i.rex |= REX_EXTZ;
2723 if ((i.op[source].regs->reg_flags & RegRex) != 0)
2724 i.rex |= REX_EXTX;
2725 }
2726 }
2727 else
2728 { /* If it's not 2 reg operands... */
2729 if (i.mem_operands)
2730 {
2731 unsigned int fake_zero_displacement = 0;
2732 unsigned int op = ((i.types[0] & AnyMem)
2733 ? 0
2734 : (i.types[1] & AnyMem) ? 1 : 2);
2735
2736 default_seg = &ds;
2737
2738 if (i.base_reg == 0)
2739 {
2740 i.rm.mode = 0;
2741 if (!i.disp_operands)
2742 fake_zero_displacement = 1;
2743 if (i.index_reg == 0)
2744 {
2745 /* Operand is just <disp> */
2746 if (flag_code == CODE_64BIT)
2747 {
2748 /* 64bit mode overwrites the 32bit absolute
2749 addressing by RIP relative addressing and
2750 absolute addressing is encoded by one of the
2751 redundant SIB forms. */
2752 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2753 i.sib.base = NO_BASE_REGISTER;
2754 i.sib.index = NO_INDEX_REGISTER;
2755 i.types[op] = ((i.prefix[ADDR_PREFIX] == 0) ? Disp32S : Disp32);
2756 }
2757 else if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
2758 {
2759 i.rm.regmem = NO_BASE_REGISTER_16;
2760 i.types[op] = Disp16;
2761 }
2762 else
2763 {
2764 i.rm.regmem = NO_BASE_REGISTER;
2765 i.types[op] = Disp32;
2766 }
2767 }
2768 else /* !i.base_reg && i.index_reg */
2769 {
2770 i.sib.index = i.index_reg->reg_num;
2771 i.sib.base = NO_BASE_REGISTER;
2772 i.sib.scale = i.log2_scale_factor;
2773 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2774 i.types[op] &= ~Disp;
2775 if (flag_code != CODE_64BIT)
2776 i.types[op] |= Disp32; /* Must be 32 bit */
2777 else
2778 i.types[op] |= Disp32S;
2779 if ((i.index_reg->reg_flags & RegRex) != 0)
2780 i.rex |= REX_EXTY;
2781 }
2782 }
2783 /* RIP addressing for 64bit mode. */
2784 else if (i.base_reg->reg_type == BaseIndex)
2785 {
2786 i.rm.regmem = NO_BASE_REGISTER;
2787 i.types[op] &= ~ Disp;
2788 i.types[op] |= Disp32S;
2789 i.flags[op] = Operand_PCrel;
2790 if (! i.disp_operands)
2791 fake_zero_displacement = 1;
2792 }
2793 else if (i.base_reg->reg_type & Reg16)
2794 {
2795 switch (i.base_reg->reg_num)
2796 {
2797 case 3: /* (%bx) */
2798 if (i.index_reg == 0)
2799 i.rm.regmem = 7;
2800 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
2801 i.rm.regmem = i.index_reg->reg_num - 6;
2802 break;
2803 case 5: /* (%bp) */
2804 default_seg = &ss;
2805 if (i.index_reg == 0)
2806 {
2807 i.rm.regmem = 6;
2808 if ((i.types[op] & Disp) == 0)
2809 {
2810 /* fake (%bp) into 0(%bp) */
2811 i.types[op] |= Disp8;
2812 fake_zero_displacement = 1;
2813 }
2814 }
2815 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
2816 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
2817 break;
2818 default: /* (%si) -> 4 or (%di) -> 5 */
2819 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
2820 }
2821 i.rm.mode = mode_from_disp_size (i.types[op]);
2822 }
2823 else /* i.base_reg and 32/64 bit mode */
2824 {
2825 if (flag_code == CODE_64BIT
2826 && (i.types[op] & Disp))
2827 i.types[op] = (i.types[op] & Disp8) | (i.prefix[ADDR_PREFIX] == 0 ? Disp32S : Disp32);
2828
2829 i.rm.regmem = i.base_reg->reg_num;
2830 if ((i.base_reg->reg_flags & RegRex) != 0)
2831 i.rex |= REX_EXTZ;
2832 i.sib.base = i.base_reg->reg_num;
2833 /* x86-64 ignores REX prefix bit here to avoid decoder
2834 complications. */
2835 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
2836 {
2837 default_seg = &ss;
2838 if (i.disp_operands == 0)
2839 {
2840 fake_zero_displacement = 1;
2841 i.types[op] |= Disp8;
2842 }
2843 }
2844 else if (i.base_reg->reg_num == ESP_REG_NUM)
2845 {
2846 default_seg = &ss;
2847 }
2848 i.sib.scale = i.log2_scale_factor;
2849 if (i.index_reg == 0)
2850 {
2851 /* <disp>(%esp) becomes two byte modrm with no index
2852 register. We've already stored the code for esp
2853 in i.rm.regmem ie. ESCAPE_TO_TWO_BYTE_ADDRESSING.
2854 Any base register besides %esp will not use the
2855 extra modrm byte. */
2856 i.sib.index = NO_INDEX_REGISTER;
2857 #if !SCALE1_WHEN_NO_INDEX
2858 /* Another case where we force the second modrm byte. */
2859 if (i.log2_scale_factor)
2860 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2861 #endif
2862 }
2863 else
2864 {
2865 i.sib.index = i.index_reg->reg_num;
2866 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2867 if ((i.index_reg->reg_flags & RegRex) != 0)
2868 i.rex |= REX_EXTY;
2869 }
2870 i.rm.mode = mode_from_disp_size (i.types[op]);
2871 }
2872
2873 if (fake_zero_displacement)
2874 {
2875 /* Fakes a zero displacement assuming that i.types[op]
2876 holds the correct displacement size. */
2877 expressionS *exp;
2878
2879 assert (i.op[op].disps == 0);
2880 exp = &disp_expressions[i.disp_operands++];
2881 i.op[op].disps = exp;
2882 exp->X_op = O_constant;
2883 exp->X_add_number = 0;
2884 exp->X_add_symbol = (symbolS *) 0;
2885 exp->X_op_symbol = (symbolS *) 0;
2886 }
2887 }
2888
2889 /* Fill in i.rm.reg or i.rm.regmem field with register operand
2890 (if any) based on i.tm.extension_opcode. Again, we must be
2891 careful to make sure that segment/control/debug/test/MMX
2892 registers are coded into the i.rm.reg field. */
2893 if (i.reg_operands)
2894 {
2895 unsigned int op =
2896 ((i.types[0]
2897 & (Reg | RegMMX | RegXMM
2898 | SReg2 | SReg3
2899 | Control | Debug | Test))
2900 ? 0
2901 : ((i.types[1]
2902 & (Reg | RegMMX | RegXMM
2903 | SReg2 | SReg3
2904 | Control | Debug | Test))
2905 ? 1
2906 : 2));
2907 /* If there is an extension opcode to put here, the register
2908 number must be put into the regmem field. */
2909 if (i.tm.extension_opcode != None)
2910 {
2911 i.rm.regmem = i.op[op].regs->reg_num;
2912 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2913 i.rex |= REX_EXTZ;
2914 }
2915 else
2916 {
2917 i.rm.reg = i.op[op].regs->reg_num;
2918 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2919 i.rex |= REX_EXTX;
2920 }
2921
2922 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2 we
2923 must set it to 3 to indicate this is a register operand
2924 in the regmem field. */
2925 if (!i.mem_operands)
2926 i.rm.mode = 3;
2927 }
2928
2929 /* Fill in i.rm.reg field with extension opcode (if any). */
2930 if (i.tm.extension_opcode != None)
2931 i.rm.reg = i.tm.extension_opcode;
2932 }
2933 return default_seg;
2934 }
2935
2936 static void
2937 output_branch ()
2938 {
2939 char *p;
2940 int code16;
2941 int prefix;
2942 relax_substateT subtype;
2943 symbolS *sym;
2944 offsetT off;
2945
2946 code16 = 0;
2947 if (flag_code == CODE_16BIT)
2948 code16 = CODE16;
2949
2950 prefix = 0;
2951 if (i.prefix[DATA_PREFIX] != 0)
2952 {
2953 prefix = 1;
2954 i.prefixes -= 1;
2955 code16 ^= CODE16;
2956 }
2957 /* Pentium4 branch hints. */
2958 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
2959 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
2960 {
2961 prefix++;
2962 i.prefixes--;
2963 }
2964 if (i.prefix[REX_PREFIX] != 0)
2965 {
2966 prefix++;
2967 i.prefixes--;
2968 }
2969
2970 if (i.prefixes != 0 && !intel_syntax)
2971 as_warn (_("skipping prefixes on this instruction"));
2972
2973 /* It's always a symbol; End frag & setup for relax.
2974 Make sure there is enough room in this frag for the largest
2975 instruction we may generate in md_convert_frag. This is 2
2976 bytes for the opcode and room for the prefix and largest
2977 displacement. */
2978 frag_grow (prefix + 2 + 4);
2979 /* Prefix and 1 opcode byte go in fr_fix. */
2980 p = frag_more (prefix + 1);
2981 if (i.prefix[DATA_PREFIX] != 0)
2982 *p++ = DATA_PREFIX_OPCODE;
2983 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
2984 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
2985 *p++ = i.prefix[SEG_PREFIX];
2986 if (i.prefix[REX_PREFIX] != 0)
2987 *p++ = i.prefix[REX_PREFIX];
2988 *p = i.tm.base_opcode;
2989
2990 if ((unsigned char) *p == JUMP_PC_RELATIVE)
2991 subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
2992 else if ((cpu_arch_flags & Cpu386) != 0)
2993 subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
2994 else
2995 subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
2996 subtype |= code16;
2997
2998 sym = i.op[0].disps->X_add_symbol;
2999 off = i.op[0].disps->X_add_number;
3000
3001 if (i.op[0].disps->X_op != O_constant
3002 && i.op[0].disps->X_op != O_symbol)
3003 {
3004 /* Handle complex expressions. */
3005 sym = make_expr_symbol (i.op[0].disps);
3006 off = 0;
3007 }
3008
3009 /* 1 possible extra opcode + 4 byte displacement go in var part.
3010 Pass reloc in fr_var. */
3011 frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
3012 }
3013
3014 static void
3015 output_jump ()
3016 {
3017 char *p;
3018 int size;
3019 fixS *fixP;
3020
3021 if (i.tm.opcode_modifier & JumpByte)
3022 {
3023 /* This is a loop or jecxz type instruction. */
3024 size = 1;
3025 if (i.prefix[ADDR_PREFIX] != 0)
3026 {
3027 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
3028 i.prefixes -= 1;
3029 }
3030 /* Pentium4 branch hints. */
3031 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3032 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3033 {
3034 FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
3035 i.prefixes--;
3036 }
3037 }
3038 else
3039 {
3040 int code16;
3041
3042 code16 = 0;
3043 if (flag_code == CODE_16BIT)
3044 code16 = CODE16;
3045
3046 if (i.prefix[DATA_PREFIX] != 0)
3047 {
3048 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
3049 i.prefixes -= 1;
3050 code16 ^= CODE16;
3051 }
3052
3053 size = 4;
3054 if (code16)
3055 size = 2;
3056 }
3057
3058 if (i.prefix[REX_PREFIX] != 0)
3059 {
3060 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
3061 i.prefixes -= 1;
3062 }
3063
3064 if (i.prefixes != 0 && !intel_syntax)
3065 as_warn (_("skipping prefixes on this instruction"));
3066
3067 p = frag_more (1 + size);
3068 *p++ = i.tm.base_opcode;
3069
3070 fixP = fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3071 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
3072
3073 /* All jumps handled here are signed, but don't use a signed limit
3074 check for 32 and 16 bit jumps as we want to allow wrap around at
3075 4G and 64k respectively. */
3076 if (size == 1)
3077 fixP->fx_signed = 1;
3078 }
3079
3080 static void
3081 output_interseg_jump ()
3082 {
3083 char *p;
3084 int size;
3085 int prefix;
3086 int code16;
3087
3088 code16 = 0;
3089 if (flag_code == CODE_16BIT)
3090 code16 = CODE16;
3091
3092 prefix = 0;
3093 if (i.prefix[DATA_PREFIX] != 0)
3094 {
3095 prefix = 1;
3096 i.prefixes -= 1;
3097 code16 ^= CODE16;
3098 }
3099 if (i.prefix[REX_PREFIX] != 0)
3100 {
3101 prefix++;
3102 i.prefixes -= 1;
3103 }
3104
3105 size = 4;
3106 if (code16)
3107 size = 2;
3108
3109 if (i.prefixes != 0 && !intel_syntax)
3110 as_warn (_("skipping prefixes on this instruction"));
3111
3112 /* 1 opcode; 2 segment; offset */
3113 p = frag_more (prefix + 1 + 2 + size);
3114
3115 if (i.prefix[DATA_PREFIX] != 0)
3116 *p++ = DATA_PREFIX_OPCODE;
3117
3118 if (i.prefix[REX_PREFIX] != 0)
3119 *p++ = i.prefix[REX_PREFIX];
3120
3121 *p++ = i.tm.base_opcode;
3122 if (i.op[1].imms->X_op == O_constant)
3123 {
3124 offsetT n = i.op[1].imms->X_add_number;
3125
3126 if (size == 2
3127 && !fits_in_unsigned_word (n)
3128 && !fits_in_signed_word (n))
3129 {
3130 as_bad (_("16-bit jump out of range"));
3131 return;
3132 }
3133 md_number_to_chars (p, n, size);
3134 }
3135 else
3136 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3137 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
3138 if (i.op[0].imms->X_op != O_constant)
3139 as_bad (_("can't handle non absolute segment in `%s'"),
3140 i.tm.name);
3141 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
3142 }
3143
3144 static void
3145 output_insn ()
3146 {
3147 fragS *insn_start_frag;
3148 offsetT insn_start_off;
3149
3150 /* Tie dwarf2 debug info to the address at the start of the insn.
3151 We can't do this after the insn has been output as the current
3152 frag may have been closed off. eg. by frag_var. */
3153 dwarf2_emit_insn (0);
3154
3155 insn_start_frag = frag_now;
3156 insn_start_off = frag_now_fix ();
3157
3158 /* Output jumps. */
3159 if (i.tm.opcode_modifier & Jump)
3160 output_branch ();
3161 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
3162 output_jump ();
3163 else if (i.tm.opcode_modifier & JumpInterSegment)
3164 output_interseg_jump ();
3165 else
3166 {
3167 /* Output normal instructions here. */
3168 char *p;
3169 unsigned char *q;
3170
3171 /* All opcodes on i386 have either 1 or 2 bytes, PadLock instructions
3172 have 3 bytes. We may use one more higher byte to specify a prefix
3173 the instruction requires. */
3174 if ((i.tm.cpu_flags & CpuPadLock) != 0
3175 && (i.tm.base_opcode & 0xff000000) != 0)
3176 {
3177 unsigned int prefix;
3178 prefix = (i.tm.base_opcode >> 24) & 0xff;
3179
3180 if (prefix != REPE_PREFIX_OPCODE
3181 || i.prefix[LOCKREP_PREFIX] != REPE_PREFIX_OPCODE)
3182 add_prefix (prefix);
3183 }
3184 else
3185 if ((i.tm.cpu_flags & CpuPadLock) == 0
3186 && (i.tm.base_opcode & 0xff0000) != 0)
3187 add_prefix ((i.tm.base_opcode >> 16) & 0xff);
3188
3189 /* The prefix bytes. */
3190 for (q = i.prefix;
3191 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
3192 q++)
3193 {
3194 if (*q)
3195 {
3196 p = frag_more (1);
3197 md_number_to_chars (p, (valueT) *q, 1);
3198 }
3199 }
3200
3201 /* Now the opcode; be careful about word order here! */
3202 if (fits_in_unsigned_byte (i.tm.base_opcode))
3203 {
3204 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
3205 }
3206 else
3207 {
3208 if ((i.tm.cpu_flags & CpuPadLock) != 0)
3209 {
3210 p = frag_more (3);
3211 *p++ = (i.tm.base_opcode >> 16) & 0xff;
3212 }
3213 else
3214 p = frag_more (2);
3215
3216 /* Put out high byte first: can't use md_number_to_chars! */
3217 *p++ = (i.tm.base_opcode >> 8) & 0xff;
3218 *p = i.tm.base_opcode & 0xff;
3219 }
3220
3221 /* Now the modrm byte and sib byte (if present). */
3222 if (i.tm.opcode_modifier & Modrm)
3223 {
3224 p = frag_more (1);
3225 md_number_to_chars (p,
3226 (valueT) (i.rm.regmem << 0
3227 | i.rm.reg << 3
3228 | i.rm.mode << 6),
3229 1);
3230 /* If i.rm.regmem == ESP (4)
3231 && i.rm.mode != (Register mode)
3232 && not 16 bit
3233 ==> need second modrm byte. */
3234 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
3235 && i.rm.mode != 3
3236 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
3237 {
3238 p = frag_more (1);
3239 md_number_to_chars (p,
3240 (valueT) (i.sib.base << 0
3241 | i.sib.index << 3
3242 | i.sib.scale << 6),
3243 1);
3244 }
3245 }
3246
3247 if (i.disp_operands)
3248 output_disp (insn_start_frag, insn_start_off);
3249
3250 if (i.imm_operands)
3251 output_imm (insn_start_frag, insn_start_off);
3252 }
3253
3254 #ifdef DEBUG386
3255 if (flag_debug)
3256 {
3257 pi (line, &i);
3258 }
3259 #endif /* DEBUG386 */
3260 }
3261
3262 static void
3263 output_disp (insn_start_frag, insn_start_off)
3264 fragS *insn_start_frag;
3265 offsetT insn_start_off;
3266 {
3267 char *p;
3268 unsigned int n;
3269
3270 for (n = 0; n < i.operands; n++)
3271 {
3272 if (i.types[n] & Disp)
3273 {
3274 if (i.op[n].disps->X_op == O_constant)
3275 {
3276 int size;
3277 offsetT val;
3278
3279 size = 4;
3280 if (i.types[n] & (Disp8 | Disp16 | Disp64))
3281 {
3282 size = 2;
3283 if (i.types[n] & Disp8)
3284 size = 1;
3285 if (i.types[n] & Disp64)
3286 size = 8;
3287 }
3288 val = offset_in_range (i.op[n].disps->X_add_number,
3289 size);
3290 p = frag_more (size);
3291 md_number_to_chars (p, val, size);
3292 }
3293 else
3294 {
3295 enum bfd_reloc_code_real reloc_type;
3296 int size = 4;
3297 int sign = 0;
3298 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
3299
3300 /* The PC relative address is computed relative
3301 to the instruction boundary, so in case immediate
3302 fields follows, we need to adjust the value. */
3303 if (pcrel && i.imm_operands)
3304 {
3305 int imm_size = 4;
3306 unsigned int n1;
3307
3308 for (n1 = 0; n1 < i.operands; n1++)
3309 if (i.types[n1] & Imm)
3310 {
3311 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
3312 {
3313 imm_size = 2;
3314 if (i.types[n1] & (Imm8 | Imm8S))
3315 imm_size = 1;
3316 if (i.types[n1] & Imm64)
3317 imm_size = 8;
3318 }
3319 break;
3320 }
3321 /* We should find the immediate. */
3322 if (n1 == i.operands)
3323 abort ();
3324 i.op[n].disps->X_add_number -= imm_size;
3325 }
3326
3327 if (i.types[n] & Disp32S)
3328 sign = 1;
3329
3330 if (i.types[n] & (Disp16 | Disp64))
3331 {
3332 size = 2;
3333 if (i.types[n] & Disp64)
3334 size = 8;
3335 }
3336
3337 p = frag_more (size);
3338 reloc_type = reloc (size, pcrel, sign, i.reloc[n]);
3339 if (reloc_type == BFD_RELOC_32
3340 && GOT_symbol
3341 && GOT_symbol == i.op[n].disps->X_add_symbol
3342 && (i.op[n].disps->X_op == O_symbol
3343 || (i.op[n].disps->X_op == O_add
3344 && ((symbol_get_value_expression
3345 (i.op[n].disps->X_op_symbol)->X_op)
3346 == O_subtract))))
3347 {
3348 offsetT add;
3349
3350 if (insn_start_frag == frag_now)
3351 add = (p - frag_now->fr_literal) - insn_start_off;
3352 else
3353 {
3354 fragS *fr;
3355
3356 add = insn_start_frag->fr_fix - insn_start_off;
3357 for (fr = insn_start_frag->fr_next;
3358 fr && fr != frag_now; fr = fr->fr_next)
3359 add += fr->fr_fix;
3360 add += p - frag_now->fr_literal;
3361 }
3362
3363 /* We don't support dynamic linking on x86-64 yet. */
3364 if (flag_code == CODE_64BIT)
3365 abort ();
3366 reloc_type = BFD_RELOC_386_GOTPC;
3367 i.op[n].disps->X_add_number += add;
3368 }
3369 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3370 i.op[n].disps, pcrel, reloc_type);
3371 }
3372 }
3373 }
3374 }
3375
3376 static void
3377 output_imm (insn_start_frag, insn_start_off)
3378 fragS *insn_start_frag;
3379 offsetT insn_start_off;
3380 {
3381 char *p;
3382 unsigned int n;
3383
3384 for (n = 0; n < i.operands; n++)
3385 {
3386 if (i.types[n] & Imm)
3387 {
3388 if (i.op[n].imms->X_op == O_constant)
3389 {
3390 int size;
3391 offsetT val;
3392
3393 size = 4;
3394 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3395 {
3396 size = 2;
3397 if (i.types[n] & (Imm8 | Imm8S))
3398 size = 1;
3399 else if (i.types[n] & Imm64)
3400 size = 8;
3401 }
3402 val = offset_in_range (i.op[n].imms->X_add_number,
3403 size);
3404 p = frag_more (size);
3405 md_number_to_chars (p, val, size);
3406 }
3407 else
3408 {
3409 /* Not absolute_section.
3410 Need a 32-bit fixup (don't support 8bit
3411 non-absolute imms). Try to support other
3412 sizes ... */
3413 enum bfd_reloc_code_real reloc_type;
3414 int size = 4;
3415 int sign = 0;
3416
3417 if ((i.types[n] & (Imm32S))
3418 && i.suffix == QWORD_MNEM_SUFFIX)
3419 sign = 1;
3420 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3421 {
3422 size = 2;
3423 if (i.types[n] & (Imm8 | Imm8S))
3424 size = 1;
3425 if (i.types[n] & Imm64)
3426 size = 8;
3427 }
3428
3429 p = frag_more (size);
3430 reloc_type = reloc (size, 0, sign, i.reloc[n]);
3431
3432 /* This is tough to explain. We end up with this one if we
3433 * have operands that look like
3434 * "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal here is to
3435 * obtain the absolute address of the GOT, and it is strongly
3436 * preferable from a performance point of view to avoid using
3437 * a runtime relocation for this. The actual sequence of
3438 * instructions often look something like:
3439 *
3440 * call .L66
3441 * .L66:
3442 * popl %ebx
3443 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
3444 *
3445 * The call and pop essentially return the absolute address
3446 * of the label .L66 and store it in %ebx. The linker itself
3447 * will ultimately change the first operand of the addl so
3448 * that %ebx points to the GOT, but to keep things simple, the
3449 * .o file must have this operand set so that it generates not
3450 * the absolute address of .L66, but the absolute address of
3451 * itself. This allows the linker itself simply treat a GOTPC
3452 * relocation as asking for a pcrel offset to the GOT to be
3453 * added in, and the addend of the relocation is stored in the
3454 * operand field for the instruction itself.
3455 *
3456 * Our job here is to fix the operand so that it would add
3457 * the correct offset so that %ebx would point to itself. The
3458 * thing that is tricky is that .-.L66 will point to the
3459 * beginning of the instruction, so we need to further modify
3460 * the operand so that it will point to itself. There are
3461 * other cases where you have something like:
3462 *
3463 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
3464 *
3465 * and here no correction would be required. Internally in
3466 * the assembler we treat operands of this form as not being
3467 * pcrel since the '.' is explicitly mentioned, and I wonder
3468 * whether it would simplify matters to do it this way. Who
3469 * knows. In earlier versions of the PIC patches, the
3470 * pcrel_adjust field was used to store the correction, but
3471 * since the expression is not pcrel, I felt it would be
3472 * confusing to do it this way. */
3473
3474 if (reloc_type == BFD_RELOC_32
3475 && GOT_symbol
3476 && GOT_symbol == i.op[n].imms->X_add_symbol
3477 && (i.op[n].imms->X_op == O_symbol
3478 || (i.op[n].imms->X_op == O_add
3479 && ((symbol_get_value_expression
3480 (i.op[n].imms->X_op_symbol)->X_op)
3481 == O_subtract))))
3482 {
3483 offsetT add;
3484
3485 if (insn_start_frag == frag_now)
3486 add = (p - frag_now->fr_literal) - insn_start_off;
3487 else
3488 {
3489 fragS *fr;
3490
3491 add = insn_start_frag->fr_fix - insn_start_off;
3492 for (fr = insn_start_frag->fr_next;
3493 fr && fr != frag_now; fr = fr->fr_next)
3494 add += fr->fr_fix;
3495 add += p - frag_now->fr_literal;
3496 }
3497
3498 /* We don't support dynamic linking on x86-64 yet. */
3499 if (flag_code == CODE_64BIT)
3500 abort ();
3501 reloc_type = BFD_RELOC_386_GOTPC;
3502 i.op[n].imms->X_add_number += add;
3503 }
3504 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3505 i.op[n].imms, 0, reloc_type);
3506 }
3507 }
3508 }
3509 }
3510 \f
3511 #ifndef LEX_AT
3512 static char *lex_got PARAMS ((enum bfd_reloc_code_real *, int *));
3513
3514 /* Parse operands of the form
3515 <symbol>@GOTOFF+<nnn>
3516 and similar .plt or .got references.
3517
3518 If we find one, set up the correct relocation in RELOC and copy the
3519 input string, minus the `@GOTOFF' into a malloc'd buffer for
3520 parsing by the calling routine. Return this buffer, and if ADJUST
3521 is non-null set it to the length of the string we removed from the
3522 input line. Otherwise return NULL. */
3523 static char *
3524 lex_got (reloc, adjust)
3525 enum bfd_reloc_code_real *reloc;
3526 int *adjust;
3527 {
3528 static const char * const mode_name[NUM_FLAG_CODE] = { "32", "16", "64" };
3529 static const struct {
3530 const char *str;
3531 const enum bfd_reloc_code_real rel[NUM_FLAG_CODE];
3532 } gotrel[] = {
3533 { "PLT", { BFD_RELOC_386_PLT32, 0, BFD_RELOC_X86_64_PLT32 } },
3534 { "GOTOFF", { BFD_RELOC_386_GOTOFF, 0, 0 } },
3535 { "GOTPCREL", { 0, 0, BFD_RELOC_X86_64_GOTPCREL } },
3536 { "TLSGD", { BFD_RELOC_386_TLS_GD, 0, BFD_RELOC_X86_64_TLSGD } },
3537 { "TLSLDM", { BFD_RELOC_386_TLS_LDM, 0, 0 } },
3538 { "TLSLD", { 0, 0, BFD_RELOC_X86_64_TLSLD } },
3539 { "GOTTPOFF", { BFD_RELOC_386_TLS_IE_32, 0, BFD_RELOC_X86_64_GOTTPOFF } },
3540 { "TPOFF", { BFD_RELOC_386_TLS_LE_32, 0, BFD_RELOC_X86_64_TPOFF32 } },
3541 { "NTPOFF", { BFD_RELOC_386_TLS_LE, 0, 0 } },
3542 { "DTPOFF", { BFD_RELOC_386_TLS_LDO_32, 0, BFD_RELOC_X86_64_DTPOFF32 } },
3543 { "GOTNTPOFF",{ BFD_RELOC_386_TLS_GOTIE, 0, 0 } },
3544 { "INDNTPOFF",{ BFD_RELOC_386_TLS_IE, 0, 0 } },
3545 { "GOT", { BFD_RELOC_386_GOT32, 0, BFD_RELOC_X86_64_GOT32 } }
3546 };
3547 char *cp;
3548 unsigned int j;
3549
3550 for (cp = input_line_pointer; *cp != '@'; cp++)
3551 if (is_end_of_line[(unsigned char) *cp])
3552 return NULL;
3553
3554 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
3555 {
3556 int len;
3557
3558 len = strlen (gotrel[j].str);
3559 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
3560 {
3561 if (gotrel[j].rel[(unsigned int) flag_code] != 0)
3562 {
3563 int first, second;
3564 char *tmpbuf, *past_reloc;
3565
3566 *reloc = gotrel[j].rel[(unsigned int) flag_code];
3567 if (adjust)
3568 *adjust = len;
3569
3570 if (GOT_symbol == NULL)
3571 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3572
3573 /* Replace the relocation token with ' ', so that
3574 errors like foo@GOTOFF1 will be detected. */
3575
3576 /* The length of the first part of our input line. */
3577 first = cp - input_line_pointer;
3578
3579 /* The second part goes from after the reloc token until
3580 (and including) an end_of_line char. Don't use strlen
3581 here as the end_of_line char may not be a NUL. */
3582 past_reloc = cp + 1 + len;
3583 for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
3584 ;
3585 second = cp - past_reloc;
3586
3587 /* Allocate and copy string. The trailing NUL shouldn't
3588 be necessary, but be safe. */
3589 tmpbuf = xmalloc (first + second + 2);
3590 memcpy (tmpbuf, input_line_pointer, first);
3591 tmpbuf[first] = ' ';
3592 memcpy (tmpbuf + first + 1, past_reloc, second);
3593 tmpbuf[first + second + 1] = '\0';
3594 return tmpbuf;
3595 }
3596
3597 as_bad (_("@%s reloc is not supported in %s bit mode"),
3598 gotrel[j].str, mode_name[(unsigned int) flag_code]);
3599 return NULL;
3600 }
3601 }
3602
3603 /* Might be a symbol version string. Don't as_bad here. */
3604 return NULL;
3605 }
3606
3607 /* x86_cons_fix_new is called via the expression parsing code when a
3608 reloc is needed. We use this hook to get the correct .got reloc. */
3609 static enum bfd_reloc_code_real got_reloc = NO_RELOC;
3610
3611 void
3612 x86_cons_fix_new (frag, off, len, exp)
3613 fragS *frag;
3614 unsigned int off;
3615 unsigned int len;
3616 expressionS *exp;
3617 {
3618 enum bfd_reloc_code_real r = reloc (len, 0, 0, got_reloc);
3619 got_reloc = NO_RELOC;
3620 fix_new_exp (frag, off, len, exp, 0, r);
3621 }
3622
3623 void
3624 x86_cons (exp, size)
3625 expressionS *exp;
3626 int size;
3627 {
3628 if (size == 4)
3629 {
3630 /* Handle @GOTOFF and the like in an expression. */
3631 char *save;
3632 char *gotfree_input_line;
3633 int adjust;
3634
3635 save = input_line_pointer;
3636 gotfree_input_line = lex_got (&got_reloc, &adjust);
3637 if (gotfree_input_line)
3638 input_line_pointer = gotfree_input_line;
3639
3640 expression (exp);
3641
3642 if (gotfree_input_line)
3643 {
3644 /* expression () has merrily parsed up to the end of line,
3645 or a comma - in the wrong buffer. Transfer how far
3646 input_line_pointer has moved to the right buffer. */
3647 input_line_pointer = (save
3648 + (input_line_pointer - gotfree_input_line)
3649 + adjust);
3650 free (gotfree_input_line);
3651 }
3652 }
3653 else
3654 expression (exp);
3655 }
3656 #endif
3657
3658 #ifdef TE_PE
3659
3660 #define O_secrel (O_max + 1)
3661
3662 void
3663 x86_pe_cons_fix_new (frag, off, len, exp)
3664 fragS *frag;
3665 unsigned int off;
3666 unsigned int len;
3667 expressionS *exp;
3668 {
3669 enum bfd_reloc_code_real r = reloc (len, 0, 0, NO_RELOC);
3670
3671 if (exp->X_op == O_secrel)
3672 {
3673 exp->X_op = O_symbol;
3674 r = BFD_RELOC_32_SECREL;
3675 }
3676
3677 fix_new_exp (frag, off, len, exp, 0, r);
3678 }
3679
3680 static void
3681 pe_directive_secrel (dummy)
3682 int dummy ATTRIBUTE_UNUSED;
3683 {
3684 expressionS exp;
3685
3686 do
3687 {
3688 expression (&exp);
3689 if (exp.X_op == O_symbol)
3690 exp.X_op = O_secrel;
3691
3692 emit_expr (&exp, 4);
3693 }
3694 while (*input_line_pointer++ == ',');
3695
3696 input_line_pointer--;
3697 demand_empty_rest_of_line ();
3698 }
3699
3700 #endif
3701
3702 static int i386_immediate PARAMS ((char *));
3703
3704 static int
3705 i386_immediate (imm_start)
3706 char *imm_start;
3707 {
3708 char *save_input_line_pointer;
3709 #ifndef LEX_AT
3710 char *gotfree_input_line;
3711 #endif
3712 segT exp_seg = 0;
3713 expressionS *exp;
3714
3715 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
3716 {
3717 as_bad (_("only 1 or 2 immediate operands are allowed"));
3718 return 0;
3719 }
3720
3721 exp = &im_expressions[i.imm_operands++];
3722 i.op[this_operand].imms = exp;
3723
3724 if (is_space_char (*imm_start))
3725 ++imm_start;
3726
3727 save_input_line_pointer = input_line_pointer;
3728 input_line_pointer = imm_start;
3729
3730 #ifndef LEX_AT
3731 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3732 if (gotfree_input_line)
3733 input_line_pointer = gotfree_input_line;
3734 #endif
3735
3736 exp_seg = expression (exp);
3737
3738 SKIP_WHITESPACE ();
3739 if (*input_line_pointer)
3740 as_bad (_("junk `%s' after expression"), input_line_pointer);
3741
3742 input_line_pointer = save_input_line_pointer;
3743 #ifndef LEX_AT
3744 if (gotfree_input_line)
3745 free (gotfree_input_line);
3746 #endif
3747
3748 if (exp->X_op == O_absent || exp->X_op == O_big)
3749 {
3750 /* Missing or bad expr becomes absolute 0. */
3751 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
3752 imm_start);
3753 exp->X_op = O_constant;
3754 exp->X_add_number = 0;
3755 exp->X_add_symbol = (symbolS *) 0;
3756 exp->X_op_symbol = (symbolS *) 0;
3757 }
3758 else if (exp->X_op == O_constant)
3759 {
3760 /* Size it properly later. */
3761 i.types[this_operand] |= Imm64;
3762 /* If BFD64, sign extend val. */
3763 if (!use_rela_relocations)
3764 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
3765 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
3766 }
3767 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3768 else if (OUTPUT_FLAVOR == bfd_target_aout_flavour
3769 && exp_seg != absolute_section
3770 && exp_seg != text_section
3771 && exp_seg != data_section
3772 && exp_seg != bss_section
3773 && exp_seg != undefined_section
3774 && !bfd_is_com_section (exp_seg))
3775 {
3776 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3777 return 0;
3778 }
3779 #endif
3780 else
3781 {
3782 /* This is an address. The size of the address will be
3783 determined later, depending on destination register,
3784 suffix, or the default for the section. */
3785 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
3786 }
3787
3788 return 1;
3789 }
3790
3791 static char *i386_scale PARAMS ((char *));
3792
3793 static char *
3794 i386_scale (scale)
3795 char *scale;
3796 {
3797 offsetT val;
3798 char *save = input_line_pointer;
3799
3800 input_line_pointer = scale;
3801 val = get_absolute_expression ();
3802
3803 switch (val)
3804 {
3805 case 0:
3806 case 1:
3807 i.log2_scale_factor = 0;
3808 break;
3809 case 2:
3810 i.log2_scale_factor = 1;
3811 break;
3812 case 4:
3813 i.log2_scale_factor = 2;
3814 break;
3815 case 8:
3816 i.log2_scale_factor = 3;
3817 break;
3818 default:
3819 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
3820 scale);
3821 input_line_pointer = save;
3822 return NULL;
3823 }
3824 if (i.log2_scale_factor != 0 && i.index_reg == 0)
3825 {
3826 as_warn (_("scale factor of %d without an index register"),
3827 1 << i.log2_scale_factor);
3828 #if SCALE1_WHEN_NO_INDEX
3829 i.log2_scale_factor = 0;
3830 #endif
3831 }
3832 scale = input_line_pointer;
3833 input_line_pointer = save;
3834 return scale;
3835 }
3836
3837 static int i386_displacement PARAMS ((char *, char *));
3838
3839 static int
3840 i386_displacement (disp_start, disp_end)
3841 char *disp_start;
3842 char *disp_end;
3843 {
3844 expressionS *exp;
3845 segT exp_seg = 0;
3846 char *save_input_line_pointer;
3847 #ifndef LEX_AT
3848 char *gotfree_input_line;
3849 #endif
3850 int bigdisp = Disp32;
3851
3852 if (flag_code == CODE_64BIT)
3853 {
3854 if (i.prefix[ADDR_PREFIX] == 0)
3855 bigdisp = Disp64;
3856 }
3857 else if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3858 bigdisp = Disp16;
3859 i.types[this_operand] |= bigdisp;
3860
3861 exp = &disp_expressions[i.disp_operands];
3862 i.op[this_operand].disps = exp;
3863 i.disp_operands++;
3864 save_input_line_pointer = input_line_pointer;
3865 input_line_pointer = disp_start;
3866 END_STRING_AND_SAVE (disp_end);
3867
3868 #ifndef GCC_ASM_O_HACK
3869 #define GCC_ASM_O_HACK 0
3870 #endif
3871 #if GCC_ASM_O_HACK
3872 END_STRING_AND_SAVE (disp_end + 1);
3873 if ((i.types[this_operand] & BaseIndex) != 0
3874 && displacement_string_end[-1] == '+')
3875 {
3876 /* This hack is to avoid a warning when using the "o"
3877 constraint within gcc asm statements.
3878 For instance:
3879
3880 #define _set_tssldt_desc(n,addr,limit,type) \
3881 __asm__ __volatile__ ( \
3882 "movw %w2,%0\n\t" \
3883 "movw %w1,2+%0\n\t" \
3884 "rorl $16,%1\n\t" \
3885 "movb %b1,4+%0\n\t" \
3886 "movb %4,5+%0\n\t" \
3887 "movb $0,6+%0\n\t" \
3888 "movb %h1,7+%0\n\t" \
3889 "rorl $16,%1" \
3890 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
3891
3892 This works great except that the output assembler ends
3893 up looking a bit weird if it turns out that there is
3894 no offset. You end up producing code that looks like:
3895
3896 #APP
3897 movw $235,(%eax)
3898 movw %dx,2+(%eax)
3899 rorl $16,%edx
3900 movb %dl,4+(%eax)
3901 movb $137,5+(%eax)
3902 movb $0,6+(%eax)
3903 movb %dh,7+(%eax)
3904 rorl $16,%edx
3905 #NO_APP
3906
3907 So here we provide the missing zero. */
3908
3909 *displacement_string_end = '0';
3910 }
3911 #endif
3912 #ifndef LEX_AT
3913 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3914 if (gotfree_input_line)
3915 input_line_pointer = gotfree_input_line;
3916 #endif
3917
3918 exp_seg = expression (exp);
3919
3920 SKIP_WHITESPACE ();
3921 if (*input_line_pointer)
3922 as_bad (_("junk `%s' after expression"), input_line_pointer);
3923 #if GCC_ASM_O_HACK
3924 RESTORE_END_STRING (disp_end + 1);
3925 #endif
3926 RESTORE_END_STRING (disp_end);
3927 input_line_pointer = save_input_line_pointer;
3928 #ifndef LEX_AT
3929 if (gotfree_input_line)
3930 free (gotfree_input_line);
3931 #endif
3932
3933 /* We do this to make sure that the section symbol is in
3934 the symbol table. We will ultimately change the relocation
3935 to be relative to the beginning of the section. */
3936 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
3937 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3938 {
3939 if (exp->X_op != O_symbol)
3940 {
3941 as_bad (_("bad expression used with @%s"),
3942 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
3943 ? "GOTPCREL"
3944 : "GOTOFF"));
3945 return 0;
3946 }
3947
3948 if (S_IS_LOCAL (exp->X_add_symbol)
3949 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
3950 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
3951 exp->X_op = O_subtract;
3952 exp->X_op_symbol = GOT_symbol;
3953 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3954 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
3955 else
3956 i.reloc[this_operand] = BFD_RELOC_32;
3957 }
3958
3959 if (exp->X_op == O_absent || exp->X_op == O_big)
3960 {
3961 /* Missing or bad expr becomes absolute 0. */
3962 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
3963 disp_start);
3964 exp->X_op = O_constant;
3965 exp->X_add_number = 0;
3966 exp->X_add_symbol = (symbolS *) 0;
3967 exp->X_op_symbol = (symbolS *) 0;
3968 }
3969
3970 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3971 if (exp->X_op != O_constant
3972 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3973 && exp_seg != absolute_section
3974 && exp_seg != text_section
3975 && exp_seg != data_section
3976 && exp_seg != bss_section
3977 && exp_seg != undefined_section
3978 && !bfd_is_com_section (exp_seg))
3979 {
3980 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3981 return 0;
3982 }
3983 #endif
3984 else if (flag_code == CODE_64BIT)
3985 i.types[this_operand] |= Disp32S | Disp32;
3986 return 1;
3987 }
3988
3989 static int i386_index_check PARAMS ((const char *));
3990
3991 /* Make sure the memory operand we've been dealt is valid.
3992 Return 1 on success, 0 on a failure. */
3993
3994 static int
3995 i386_index_check (operand_string)
3996 const char *operand_string;
3997 {
3998 int ok;
3999 #if INFER_ADDR_PREFIX
4000 int fudged = 0;
4001
4002 tryprefix:
4003 #endif
4004 ok = 1;
4005 if (flag_code == CODE_64BIT)
4006 {
4007 unsigned RegXX = (i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32);
4008
4009 if ((i.base_reg
4010 && ((i.base_reg->reg_type & RegXX) == 0)
4011 && (i.base_reg->reg_type != BaseIndex
4012 || i.index_reg))
4013 || (i.index_reg
4014 && ((i.index_reg->reg_type & (RegXX | BaseIndex))
4015 != (RegXX | BaseIndex))))
4016 ok = 0;
4017 }
4018 else
4019 {
4020 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
4021 {
4022 /* 16bit checks. */
4023 if ((i.base_reg
4024 && ((i.base_reg->reg_type & (Reg16 | BaseIndex | RegRex))
4025 != (Reg16 | BaseIndex)))
4026 || (i.index_reg
4027 && (((i.index_reg->reg_type & (Reg16 | BaseIndex))
4028 != (Reg16 | BaseIndex))
4029 || !(i.base_reg
4030 && i.base_reg->reg_num < 6
4031 && i.index_reg->reg_num >= 6
4032 && i.log2_scale_factor == 0))))
4033 ok = 0;
4034 }
4035 else
4036 {
4037 /* 32bit checks. */
4038 if ((i.base_reg
4039 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
4040 || (i.index_reg
4041 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
4042 != (Reg32 | BaseIndex))))
4043 ok = 0;
4044 }
4045 }
4046 if (!ok)
4047 {
4048 #if INFER_ADDR_PREFIX
4049 if (i.prefix[ADDR_PREFIX] == 0)
4050 {
4051 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
4052 i.prefixes += 1;
4053 /* Change the size of any displacement too. At most one of
4054 Disp16 or Disp32 is set.
4055 FIXME. There doesn't seem to be any real need for separate
4056 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
4057 Removing them would probably clean up the code quite a lot. */
4058 if (flag_code != CODE_64BIT && (i.types[this_operand] & (Disp16 | Disp32)))
4059 i.types[this_operand] ^= (Disp16 | Disp32);
4060 fudged = 1;
4061 goto tryprefix;
4062 }
4063 if (fudged)
4064 as_bad (_("`%s' is not a valid base/index expression"),
4065 operand_string);
4066 else
4067 #endif
4068 as_bad (_("`%s' is not a valid %s bit base/index expression"),
4069 operand_string,
4070 flag_code_names[flag_code]);
4071 }
4072 return ok;
4073 }
4074
4075 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
4076 on error. */
4077
4078 static int
4079 i386_operand (operand_string)
4080 char *operand_string;
4081 {
4082 const reg_entry *r;
4083 char *end_op;
4084 char *op_string = operand_string;
4085
4086 if (is_space_char (*op_string))
4087 ++op_string;
4088
4089 /* We check for an absolute prefix (differentiating,
4090 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
4091 if (*op_string == ABSOLUTE_PREFIX)
4092 {
4093 ++op_string;
4094 if (is_space_char (*op_string))
4095 ++op_string;
4096 i.types[this_operand] |= JumpAbsolute;
4097 }
4098
4099 /* Check if operand is a register. */
4100 if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
4101 && (r = parse_register (op_string, &end_op)) != NULL)
4102 {
4103 /* Check for a segment override by searching for ':' after a
4104 segment register. */
4105 op_string = end_op;
4106 if (is_space_char (*op_string))
4107 ++op_string;
4108 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
4109 {
4110 switch (r->reg_num)
4111 {
4112 case 0:
4113 i.seg[i.mem_operands] = &es;
4114 break;
4115 case 1:
4116 i.seg[i.mem_operands] = &cs;
4117 break;
4118 case 2:
4119 i.seg[i.mem_operands] = &ss;
4120 break;
4121 case 3:
4122 i.seg[i.mem_operands] = &ds;
4123 break;
4124 case 4:
4125 i.seg[i.mem_operands] = &fs;
4126 break;
4127 case 5:
4128 i.seg[i.mem_operands] = &gs;
4129 break;
4130 }
4131
4132 /* Skip the ':' and whitespace. */
4133 ++op_string;
4134 if (is_space_char (*op_string))
4135 ++op_string;
4136
4137 if (!is_digit_char (*op_string)
4138 && !is_identifier_char (*op_string)
4139 && *op_string != '('
4140 && *op_string != ABSOLUTE_PREFIX)
4141 {
4142 as_bad (_("bad memory operand `%s'"), op_string);
4143 return 0;
4144 }
4145 /* Handle case of %es:*foo. */
4146 if (*op_string == ABSOLUTE_PREFIX)
4147 {
4148 ++op_string;
4149 if (is_space_char (*op_string))
4150 ++op_string;
4151 i.types[this_operand] |= JumpAbsolute;
4152 }
4153 goto do_memory_reference;
4154 }
4155 if (*op_string)
4156 {
4157 as_bad (_("junk `%s' after register"), op_string);
4158 return 0;
4159 }
4160 i.types[this_operand] |= r->reg_type & ~BaseIndex;
4161 i.op[this_operand].regs = r;
4162 i.reg_operands++;
4163 }
4164 else if (*op_string == REGISTER_PREFIX)
4165 {
4166 as_bad (_("bad register name `%s'"), op_string);
4167 return 0;
4168 }
4169 else if (*op_string == IMMEDIATE_PREFIX)
4170 {
4171 ++op_string;
4172 if (i.types[this_operand] & JumpAbsolute)
4173 {
4174 as_bad (_("immediate operand illegal with absolute jump"));
4175 return 0;
4176 }
4177 if (!i386_immediate (op_string))
4178 return 0;
4179 }
4180 else if (is_digit_char (*op_string)
4181 || is_identifier_char (*op_string)
4182 || *op_string == '(')
4183 {
4184 /* This is a memory reference of some sort. */
4185 char *base_string;
4186
4187 /* Start and end of displacement string expression (if found). */
4188 char *displacement_string_start;
4189 char *displacement_string_end;
4190
4191 do_memory_reference:
4192 if ((i.mem_operands == 1
4193 && (current_templates->start->opcode_modifier & IsString) == 0)
4194 || i.mem_operands == 2)
4195 {
4196 as_bad (_("too many memory references for `%s'"),
4197 current_templates->start->name);
4198 return 0;
4199 }
4200
4201 /* Check for base index form. We detect the base index form by
4202 looking for an ')' at the end of the operand, searching
4203 for the '(' matching it, and finding a REGISTER_PREFIX or ','
4204 after the '('. */
4205 base_string = op_string + strlen (op_string);
4206
4207 --base_string;
4208 if (is_space_char (*base_string))
4209 --base_string;
4210
4211 /* If we only have a displacement, set-up for it to be parsed later. */
4212 displacement_string_start = op_string;
4213 displacement_string_end = base_string + 1;
4214
4215 if (*base_string == ')')
4216 {
4217 char *temp_string;
4218 unsigned int parens_balanced = 1;
4219 /* We've already checked that the number of left & right ()'s are
4220 equal, so this loop will not be infinite. */
4221 do
4222 {
4223 base_string--;
4224 if (*base_string == ')')
4225 parens_balanced++;
4226 if (*base_string == '(')
4227 parens_balanced--;
4228 }
4229 while (parens_balanced);
4230
4231 temp_string = base_string;
4232
4233 /* Skip past '(' and whitespace. */
4234 ++base_string;
4235 if (is_space_char (*base_string))
4236 ++base_string;
4237
4238 if (*base_string == ','
4239 || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
4240 && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
4241 {
4242 displacement_string_end = temp_string;
4243
4244 i.types[this_operand] |= BaseIndex;
4245
4246 if (i.base_reg)
4247 {
4248 base_string = end_op;
4249 if (is_space_char (*base_string))
4250 ++base_string;
4251 }
4252
4253 /* There may be an index reg or scale factor here. */
4254 if (*base_string == ',')
4255 {
4256 ++base_string;
4257 if (is_space_char (*base_string))
4258 ++base_string;
4259
4260 if ((*base_string == REGISTER_PREFIX || allow_naked_reg)
4261 && (i.index_reg = parse_register (base_string, &end_op)) != NULL)
4262 {
4263 base_string = end_op;
4264 if (is_space_char (*base_string))
4265 ++base_string;
4266 if (*base_string == ',')
4267 {
4268 ++base_string;
4269 if (is_space_char (*base_string))
4270 ++base_string;
4271 }
4272 else if (*base_string != ')')
4273 {
4274 as_bad (_("expecting `,' or `)' after index register in `%s'"),
4275 operand_string);
4276 return 0;
4277 }
4278 }
4279 else if (*base_string == REGISTER_PREFIX)
4280 {
4281 as_bad (_("bad register name `%s'"), base_string);
4282 return 0;
4283 }
4284
4285 /* Check for scale factor. */
4286 if (*base_string != ')')
4287 {
4288 char *end_scale = i386_scale (base_string);
4289
4290 if (!end_scale)
4291 return 0;
4292
4293 base_string = end_scale;
4294 if (is_space_char (*base_string))
4295 ++base_string;
4296 if (*base_string != ')')
4297 {
4298 as_bad (_("expecting `)' after scale factor in `%s'"),
4299 operand_string);
4300 return 0;
4301 }
4302 }
4303 else if (!i.index_reg)
4304 {
4305 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
4306 *base_string);
4307 return 0;
4308 }
4309 }
4310 else if (*base_string != ')')
4311 {
4312 as_bad (_("expecting `,' or `)' after base register in `%s'"),
4313 operand_string);
4314 return 0;
4315 }
4316 }
4317 else if (*base_string == REGISTER_PREFIX)
4318 {
4319 as_bad (_("bad register name `%s'"), base_string);
4320 return 0;
4321 }
4322 }
4323
4324 /* If there's an expression beginning the operand, parse it,
4325 assuming displacement_string_start and
4326 displacement_string_end are meaningful. */
4327 if (displacement_string_start != displacement_string_end)
4328 {
4329 if (!i386_displacement (displacement_string_start,
4330 displacement_string_end))
4331 return 0;
4332 }
4333
4334 /* Special case for (%dx) while doing input/output op. */
4335 if (i.base_reg
4336 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
4337 && i.index_reg == 0
4338 && i.log2_scale_factor == 0
4339 && i.seg[i.mem_operands] == 0
4340 && (i.types[this_operand] & Disp) == 0)
4341 {
4342 i.types[this_operand] = InOutPortReg;
4343 return 1;
4344 }
4345
4346 if (i386_index_check (operand_string) == 0)
4347 return 0;
4348 i.mem_operands++;
4349 }
4350 else
4351 {
4352 /* It's not a memory operand; argh! */
4353 as_bad (_("invalid char %s beginning operand %d `%s'"),
4354 output_invalid (*op_string),
4355 this_operand + 1,
4356 op_string);
4357 return 0;
4358 }
4359 return 1; /* Normal return. */
4360 }
4361 \f
4362 /* md_estimate_size_before_relax()
4363
4364 Called just before relax() for rs_machine_dependent frags. The x86
4365 assembler uses these frags to handle variable size jump
4366 instructions.
4367
4368 Any symbol that is now undefined will not become defined.
4369 Return the correct fr_subtype in the frag.
4370 Return the initial "guess for variable size of frag" to caller.
4371 The guess is actually the growth beyond the fixed part. Whatever
4372 we do to grow the fixed or variable part contributes to our
4373 returned value. */
4374
4375 int
4376 md_estimate_size_before_relax (fragP, segment)
4377 fragS *fragP;
4378 segT segment;
4379 {
4380 /* We've already got fragP->fr_subtype right; all we have to do is
4381 check for un-relaxable symbols. On an ELF system, we can't relax
4382 an externally visible symbol, because it may be overridden by a
4383 shared library. */
4384 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
4385 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4386 || (OUTPUT_FLAVOR == bfd_target_elf_flavour
4387 && (S_IS_EXTERNAL (fragP->fr_symbol)
4388 || S_IS_WEAK (fragP->fr_symbol)))
4389 #endif
4390 )
4391 {
4392 /* Symbol is undefined in this segment, or we need to keep a
4393 reloc so that weak symbols can be overridden. */
4394 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
4395 enum bfd_reloc_code_real reloc_type;
4396 unsigned char *opcode;
4397 int old_fr_fix;
4398
4399 if (fragP->fr_var != NO_RELOC)
4400 reloc_type = fragP->fr_var;
4401 else if (size == 2)
4402 reloc_type = BFD_RELOC_16_PCREL;
4403 else
4404 reloc_type = BFD_RELOC_32_PCREL;
4405
4406 old_fr_fix = fragP->fr_fix;
4407 opcode = (unsigned char *) fragP->fr_opcode;
4408
4409 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
4410 {
4411 case UNCOND_JUMP:
4412 /* Make jmp (0xeb) a (d)word displacement jump. */
4413 opcode[0] = 0xe9;
4414 fragP->fr_fix += size;
4415 fix_new (fragP, old_fr_fix, size,
4416 fragP->fr_symbol,
4417 fragP->fr_offset, 1,
4418 reloc_type);
4419 break;
4420
4421 case COND_JUMP86:
4422 if (size == 2
4423 && (!no_cond_jump_promotion || fragP->fr_var != NO_RELOC))
4424 {
4425 /* Negate the condition, and branch past an
4426 unconditional jump. */
4427 opcode[0] ^= 1;
4428 opcode[1] = 3;
4429 /* Insert an unconditional jump. */
4430 opcode[2] = 0xe9;
4431 /* We added two extra opcode bytes, and have a two byte
4432 offset. */
4433 fragP->fr_fix += 2 + 2;
4434 fix_new (fragP, old_fr_fix + 2, 2,
4435 fragP->fr_symbol,
4436 fragP->fr_offset, 1,
4437 reloc_type);
4438 break;
4439 }
4440 /* Fall through. */
4441
4442 case COND_JUMP:
4443 if (no_cond_jump_promotion && fragP->fr_var == NO_RELOC)
4444 {
4445 fixS *fixP;
4446
4447 fragP->fr_fix += 1;
4448 fixP = fix_new (fragP, old_fr_fix, 1,
4449 fragP->fr_symbol,
4450 fragP->fr_offset, 1,
4451 BFD_RELOC_8_PCREL);
4452 fixP->fx_signed = 1;
4453 break;
4454 }
4455
4456 /* This changes the byte-displacement jump 0x7N
4457 to the (d)word-displacement jump 0x0f,0x8N. */
4458 opcode[1] = opcode[0] + 0x10;
4459 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4460 /* We've added an opcode byte. */
4461 fragP->fr_fix += 1 + size;
4462 fix_new (fragP, old_fr_fix + 1, size,
4463 fragP->fr_symbol,
4464 fragP->fr_offset, 1,
4465 reloc_type);
4466 break;
4467
4468 default:
4469 BAD_CASE (fragP->fr_subtype);
4470 break;
4471 }
4472 frag_wane (fragP);
4473 return fragP->fr_fix - old_fr_fix;
4474 }
4475
4476 /* Guess size depending on current relax state. Initially the relax
4477 state will correspond to a short jump and we return 1, because
4478 the variable part of the frag (the branch offset) is one byte
4479 long. However, we can relax a section more than once and in that
4480 case we must either set fr_subtype back to the unrelaxed state,
4481 or return the value for the appropriate branch. */
4482 return md_relax_table[fragP->fr_subtype].rlx_length;
4483 }
4484
4485 /* Called after relax() is finished.
4486
4487 In: Address of frag.
4488 fr_type == rs_machine_dependent.
4489 fr_subtype is what the address relaxed to.
4490
4491 Out: Any fixSs and constants are set up.
4492 Caller will turn frag into a ".space 0". */
4493
4494 void
4495 md_convert_frag (abfd, sec, fragP)
4496 bfd *abfd ATTRIBUTE_UNUSED;
4497 segT sec ATTRIBUTE_UNUSED;
4498 fragS *fragP;
4499 {
4500 unsigned char *opcode;
4501 unsigned char *where_to_put_displacement = NULL;
4502 offsetT target_address;
4503 offsetT opcode_address;
4504 unsigned int extension = 0;
4505 offsetT displacement_from_opcode_start;
4506
4507 opcode = (unsigned char *) fragP->fr_opcode;
4508
4509 /* Address we want to reach in file space. */
4510 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
4511
4512 /* Address opcode resides at in file space. */
4513 opcode_address = fragP->fr_address + fragP->fr_fix;
4514
4515 /* Displacement from opcode start to fill into instruction. */
4516 displacement_from_opcode_start = target_address - opcode_address;
4517
4518 if ((fragP->fr_subtype & BIG) == 0)
4519 {
4520 /* Don't have to change opcode. */
4521 extension = 1; /* 1 opcode + 1 displacement */
4522 where_to_put_displacement = &opcode[1];
4523 }
4524 else
4525 {
4526 if (no_cond_jump_promotion
4527 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
4528 as_warn_where (fragP->fr_file, fragP->fr_line, _("long jump required"));
4529
4530 switch (fragP->fr_subtype)
4531 {
4532 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
4533 extension = 4; /* 1 opcode + 4 displacement */
4534 opcode[0] = 0xe9;
4535 where_to_put_displacement = &opcode[1];
4536 break;
4537
4538 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
4539 extension = 2; /* 1 opcode + 2 displacement */
4540 opcode[0] = 0xe9;
4541 where_to_put_displacement = &opcode[1];
4542 break;
4543
4544 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
4545 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
4546 extension = 5; /* 2 opcode + 4 displacement */
4547 opcode[1] = opcode[0] + 0x10;
4548 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4549 where_to_put_displacement = &opcode[2];
4550 break;
4551
4552 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
4553 extension = 3; /* 2 opcode + 2 displacement */
4554 opcode[1] = opcode[0] + 0x10;
4555 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4556 where_to_put_displacement = &opcode[2];
4557 break;
4558
4559 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
4560 extension = 4;
4561 opcode[0] ^= 1;
4562 opcode[1] = 3;
4563 opcode[2] = 0xe9;
4564 where_to_put_displacement = &opcode[3];
4565 break;
4566
4567 default:
4568 BAD_CASE (fragP->fr_subtype);
4569 break;
4570 }
4571 }
4572
4573 /* Now put displacement after opcode. */
4574 md_number_to_chars ((char *) where_to_put_displacement,
4575 (valueT) (displacement_from_opcode_start - extension),
4576 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
4577 fragP->fr_fix += extension;
4578 }
4579 \f
4580 /* Size of byte displacement jmp. */
4581 int md_short_jump_size = 2;
4582
4583 /* Size of dword displacement jmp. */
4584 int md_long_jump_size = 5;
4585
4586 /* Size of relocation record. */
4587 const int md_reloc_size = 8;
4588
4589 void
4590 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
4591 char *ptr;
4592 addressT from_addr, to_addr;
4593 fragS *frag ATTRIBUTE_UNUSED;
4594 symbolS *to_symbol ATTRIBUTE_UNUSED;
4595 {
4596 offsetT offset;
4597
4598 offset = to_addr - (from_addr + 2);
4599 /* Opcode for byte-disp jump. */
4600 md_number_to_chars (ptr, (valueT) 0xeb, 1);
4601 md_number_to_chars (ptr + 1, (valueT) offset, 1);
4602 }
4603
4604 void
4605 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
4606 char *ptr;
4607 addressT from_addr, to_addr;
4608 fragS *frag ATTRIBUTE_UNUSED;
4609 symbolS *to_symbol ATTRIBUTE_UNUSED;
4610 {
4611 offsetT offset;
4612
4613 offset = to_addr - (from_addr + 5);
4614 md_number_to_chars (ptr, (valueT) 0xe9, 1);
4615 md_number_to_chars (ptr + 1, (valueT) offset, 4);
4616 }
4617 \f
4618 /* Apply a fixup (fixS) to segment data, once it has been determined
4619 by our caller that we have all the info we need to fix it up.
4620
4621 On the 386, immediates, displacements, and data pointers are all in
4622 the same (little-endian) format, so we don't need to care about which
4623 we are handling. */
4624
4625 void
4626 md_apply_fix3 (fixP, valP, seg)
4627 /* The fix we're to put in. */
4628 fixS *fixP;
4629 /* Pointer to the value of the bits. */
4630 valueT *valP;
4631 /* Segment fix is from. */
4632 segT seg ATTRIBUTE_UNUSED;
4633 {
4634 char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
4635 valueT value = *valP;
4636
4637 #if !defined (TE_Mach)
4638 if (fixP->fx_pcrel)
4639 {
4640 switch (fixP->fx_r_type)
4641 {
4642 default:
4643 break;
4644
4645 case BFD_RELOC_32:
4646 fixP->fx_r_type = BFD_RELOC_32_PCREL;
4647 break;
4648 case BFD_RELOC_16:
4649 fixP->fx_r_type = BFD_RELOC_16_PCREL;
4650 break;
4651 case BFD_RELOC_8:
4652 fixP->fx_r_type = BFD_RELOC_8_PCREL;
4653 break;
4654 }
4655 }
4656
4657 if (fixP->fx_addsy != NULL
4658 && (fixP->fx_r_type == BFD_RELOC_32_PCREL
4659 || fixP->fx_r_type == BFD_RELOC_16_PCREL
4660 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
4661 && !use_rela_relocations)
4662 {
4663 /* This is a hack. There should be a better way to handle this.
4664 This covers for the fact that bfd_install_relocation will
4665 subtract the current location (for partial_inplace, PC relative
4666 relocations); see more below. */
4667 #ifndef OBJ_AOUT
4668 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4669 #ifdef TE_PE
4670 || OUTPUT_FLAVOR == bfd_target_coff_flavour
4671 #endif
4672 )
4673 value += fixP->fx_where + fixP->fx_frag->fr_address;
4674 #endif
4675 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4676 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
4677 {
4678 segT sym_seg = S_GET_SEGMENT (fixP->fx_addsy);
4679
4680 if ((sym_seg == seg
4681 || (symbol_section_p (fixP->fx_addsy)
4682 && sym_seg != absolute_section))
4683 && !generic_force_reloc (fixP))
4684 {
4685 /* Yes, we add the values in twice. This is because
4686 bfd_install_relocation subtracts them out again. I think
4687 bfd_install_relocation is broken, but I don't dare change
4688 it. FIXME. */
4689 value += fixP->fx_where + fixP->fx_frag->fr_address;
4690 }
4691 }
4692 #endif
4693 #if defined (OBJ_COFF) && defined (TE_PE)
4694 /* For some reason, the PE format does not store a section
4695 address offset for a PC relative symbol. */
4696 if (S_GET_SEGMENT (fixP->fx_addsy) != seg)
4697 value += md_pcrel_from (fixP);
4698 #endif
4699 }
4700
4701 /* Fix a few things - the dynamic linker expects certain values here,
4702 and we must not disappoint it. */
4703 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4704 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4705 && fixP->fx_addsy)
4706 switch (fixP->fx_r_type)
4707 {
4708 case BFD_RELOC_386_PLT32:
4709 case BFD_RELOC_X86_64_PLT32:
4710 /* Make the jump instruction point to the address of the operand. At
4711 runtime we merely add the offset to the actual PLT entry. */
4712 value = -4;
4713 break;
4714
4715 case BFD_RELOC_386_TLS_GD:
4716 case BFD_RELOC_386_TLS_LDM:
4717 case BFD_RELOC_386_TLS_IE_32:
4718 case BFD_RELOC_386_TLS_IE:
4719 case BFD_RELOC_386_TLS_GOTIE:
4720 case BFD_RELOC_X86_64_TLSGD:
4721 case BFD_RELOC_X86_64_TLSLD:
4722 case BFD_RELOC_X86_64_GOTTPOFF:
4723 value = 0; /* Fully resolved at runtime. No addend. */
4724 /* Fallthrough */
4725 case BFD_RELOC_386_TLS_LE:
4726 case BFD_RELOC_386_TLS_LDO_32:
4727 case BFD_RELOC_386_TLS_LE_32:
4728 case BFD_RELOC_X86_64_DTPOFF32:
4729 case BFD_RELOC_X86_64_TPOFF32:
4730 S_SET_THREAD_LOCAL (fixP->fx_addsy);
4731 break;
4732
4733 case BFD_RELOC_386_GOT32:
4734 case BFD_RELOC_X86_64_GOT32:
4735 value = 0; /* Fully resolved at runtime. No addend. */
4736 break;
4737
4738 case BFD_RELOC_VTABLE_INHERIT:
4739 case BFD_RELOC_VTABLE_ENTRY:
4740 fixP->fx_done = 0;
4741 return;
4742
4743 default:
4744 break;
4745 }
4746 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
4747 *valP = value;
4748 #endif /* !defined (TE_Mach) */
4749
4750 /* Are we finished with this relocation now? */
4751 if (fixP->fx_addsy == NULL)
4752 fixP->fx_done = 1;
4753 else if (use_rela_relocations)
4754 {
4755 fixP->fx_no_overflow = 1;
4756 /* Remember value for tc_gen_reloc. */
4757 fixP->fx_addnumber = value;
4758 value = 0;
4759 }
4760
4761 md_number_to_chars (p, value, fixP->fx_size);
4762 }
4763 \f
4764 #define MAX_LITTLENUMS 6
4765
4766 /* Turn the string pointed to by litP into a floating point constant
4767 of type TYPE, and emit the appropriate bytes. The number of
4768 LITTLENUMS emitted is stored in *SIZEP. An error message is
4769 returned, or NULL on OK. */
4770
4771 char *
4772 md_atof (type, litP, sizeP)
4773 int type;
4774 char *litP;
4775 int *sizeP;
4776 {
4777 int prec;
4778 LITTLENUM_TYPE words[MAX_LITTLENUMS];
4779 LITTLENUM_TYPE *wordP;
4780 char *t;
4781
4782 switch (type)
4783 {
4784 case 'f':
4785 case 'F':
4786 prec = 2;
4787 break;
4788
4789 case 'd':
4790 case 'D':
4791 prec = 4;
4792 break;
4793
4794 case 'x':
4795 case 'X':
4796 prec = 5;
4797 break;
4798
4799 default:
4800 *sizeP = 0;
4801 return _("Bad call to md_atof ()");
4802 }
4803 t = atof_ieee (input_line_pointer, type, words);
4804 if (t)
4805 input_line_pointer = t;
4806
4807 *sizeP = prec * sizeof (LITTLENUM_TYPE);
4808 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
4809 the bigendian 386. */
4810 for (wordP = words + prec - 1; prec--;)
4811 {
4812 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
4813 litP += sizeof (LITTLENUM_TYPE);
4814 }
4815 return 0;
4816 }
4817 \f
4818 char output_invalid_buf[8];
4819
4820 static char *
4821 output_invalid (c)
4822 int c;
4823 {
4824 if (ISPRINT (c))
4825 sprintf (output_invalid_buf, "'%c'", c);
4826 else
4827 sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
4828 return output_invalid_buf;
4829 }
4830
4831 /* REG_STRING starts *before* REGISTER_PREFIX. */
4832
4833 static const reg_entry *
4834 parse_register (reg_string, end_op)
4835 char *reg_string;
4836 char **end_op;
4837 {
4838 char *s = reg_string;
4839 char *p;
4840 char reg_name_given[MAX_REG_NAME_SIZE + 1];
4841 const reg_entry *r;
4842
4843 /* Skip possible REGISTER_PREFIX and possible whitespace. */
4844 if (*s == REGISTER_PREFIX)
4845 ++s;
4846
4847 if (is_space_char (*s))
4848 ++s;
4849
4850 p = reg_name_given;
4851 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
4852 {
4853 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
4854 return (const reg_entry *) NULL;
4855 s++;
4856 }
4857
4858 /* For naked regs, make sure that we are not dealing with an identifier.
4859 This prevents confusing an identifier like `eax_var' with register
4860 `eax'. */
4861 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
4862 return (const reg_entry *) NULL;
4863
4864 *end_op = s;
4865
4866 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
4867
4868 /* Handle floating point regs, allowing spaces in the (i) part. */
4869 if (r == i386_regtab /* %st is first entry of table */)
4870 {
4871 if (is_space_char (*s))
4872 ++s;
4873 if (*s == '(')
4874 {
4875 ++s;
4876 if (is_space_char (*s))
4877 ++s;
4878 if (*s >= '0' && *s <= '7')
4879 {
4880 r = &i386_float_regtab[*s - '0'];
4881 ++s;
4882 if (is_space_char (*s))
4883 ++s;
4884 if (*s == ')')
4885 {
4886 *end_op = s + 1;
4887 return r;
4888 }
4889 }
4890 /* We have "%st(" then garbage. */
4891 return (const reg_entry *) NULL;
4892 }
4893 }
4894
4895 if (r != NULL
4896 && ((r->reg_flags & (RegRex64 | RegRex)) | (r->reg_type & Reg64)) != 0
4897 && flag_code != CODE_64BIT)
4898 return (const reg_entry *) NULL;
4899
4900 return r;
4901 }
4902 \f
4903 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4904 const char *md_shortopts = "kVQ:sqn";
4905 #else
4906 const char *md_shortopts = "qn";
4907 #endif
4908
4909 struct option md_longopts[] = {
4910 #define OPTION_32 (OPTION_MD_BASE + 0)
4911 {"32", no_argument, NULL, OPTION_32},
4912 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4913 #define OPTION_64 (OPTION_MD_BASE + 1)
4914 {"64", no_argument, NULL, OPTION_64},
4915 #endif
4916 {NULL, no_argument, NULL, 0}
4917 };
4918 size_t md_longopts_size = sizeof (md_longopts);
4919
4920 int
4921 md_parse_option (c, arg)
4922 int c;
4923 char *arg ATTRIBUTE_UNUSED;
4924 {
4925 switch (c)
4926 {
4927 case 'n':
4928 optimize_align_code = 0;
4929 break;
4930
4931 case 'q':
4932 quiet_warnings = 1;
4933 break;
4934
4935 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4936 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
4937 should be emitted or not. FIXME: Not implemented. */
4938 case 'Q':
4939 break;
4940
4941 /* -V: SVR4 argument to print version ID. */
4942 case 'V':
4943 print_version_id ();
4944 break;
4945
4946 /* -k: Ignore for FreeBSD compatibility. */
4947 case 'k':
4948 break;
4949
4950 case 's':
4951 /* -s: On i386 Solaris, this tells the native assembler to use
4952 .stab instead of .stab.excl. We always use .stab anyhow. */
4953 break;
4954
4955 case OPTION_64:
4956 {
4957 const char **list, **l;
4958
4959 list = bfd_target_list ();
4960 for (l = list; *l != NULL; l++)
4961 if (strcmp (*l, "elf64-x86-64") == 0)
4962 {
4963 default_arch = "x86_64";
4964 break;
4965 }
4966 if (*l == NULL)
4967 as_fatal (_("No compiled in support for x86_64"));
4968 free (list);
4969 }
4970 break;
4971 #endif
4972
4973 case OPTION_32:
4974 default_arch = "i386";
4975 break;
4976
4977 default:
4978 return 0;
4979 }
4980 return 1;
4981 }
4982
4983 void
4984 md_show_usage (stream)
4985 FILE *stream;
4986 {
4987 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4988 fprintf (stream, _("\
4989 -Q ignored\n\
4990 -V print assembler version number\n\
4991 -k ignored\n\
4992 -n Do not optimize code alignment\n\
4993 -q quieten some warnings\n\
4994 -s ignored\n"));
4995 #else
4996 fprintf (stream, _("\
4997 -n Do not optimize code alignment\n\
4998 -q quieten some warnings\n"));
4999 #endif
5000 }
5001
5002 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
5003 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
5004
5005 /* Pick the target format to use. */
5006
5007 const char *
5008 i386_target_format ()
5009 {
5010 if (!strcmp (default_arch, "x86_64"))
5011 set_code_flag (CODE_64BIT);
5012 else if (!strcmp (default_arch, "i386"))
5013 set_code_flag (CODE_32BIT);
5014 else
5015 as_fatal (_("Unknown architecture"));
5016 switch (OUTPUT_FLAVOR)
5017 {
5018 #ifdef OBJ_MAYBE_AOUT
5019 case bfd_target_aout_flavour:
5020 return AOUT_TARGET_FORMAT;
5021 #endif
5022 #ifdef OBJ_MAYBE_COFF
5023 case bfd_target_coff_flavour:
5024 return "coff-i386";
5025 #endif
5026 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
5027 case bfd_target_elf_flavour:
5028 {
5029 if (flag_code == CODE_64BIT)
5030 use_rela_relocations = 1;
5031 return flag_code == CODE_64BIT ? "elf64-x86-64" : ELF_TARGET_FORMAT;
5032 }
5033 #endif
5034 default:
5035 abort ();
5036 return NULL;
5037 }
5038 }
5039
5040 #endif /* OBJ_MAYBE_ more than one */
5041
5042 #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
5043 void i386_elf_emit_arch_note ()
5044 {
5045 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
5046 && cpu_arch_name != NULL)
5047 {
5048 char *p;
5049 asection *seg = now_seg;
5050 subsegT subseg = now_subseg;
5051 Elf_Internal_Note i_note;
5052 Elf_External_Note e_note;
5053 asection *note_secp;
5054 int len;
5055
5056 /* Create the .note section. */
5057 note_secp = subseg_new (".note", 0);
5058 bfd_set_section_flags (stdoutput,
5059 note_secp,
5060 SEC_HAS_CONTENTS | SEC_READONLY);
5061
5062 /* Process the arch string. */
5063 len = strlen (cpu_arch_name);
5064
5065 i_note.namesz = len + 1;
5066 i_note.descsz = 0;
5067 i_note.type = NT_ARCH;
5068 p = frag_more (sizeof (e_note.namesz));
5069 md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
5070 p = frag_more (sizeof (e_note.descsz));
5071 md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
5072 p = frag_more (sizeof (e_note.type));
5073 md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
5074 p = frag_more (len + 1);
5075 strcpy (p, cpu_arch_name);
5076
5077 frag_align (2, 0, 0);
5078
5079 subseg_set (seg, subseg);
5080 }
5081 }
5082 #endif
5083 \f
5084 symbolS *
5085 md_undefined_symbol (name)
5086 char *name;
5087 {
5088 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
5089 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
5090 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
5091 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
5092 {
5093 if (!GOT_symbol)
5094 {
5095 if (symbol_find (name))
5096 as_bad (_("GOT already in symbol table"));
5097 GOT_symbol = symbol_new (name, undefined_section,
5098 (valueT) 0, &zero_address_frag);
5099 };
5100 return GOT_symbol;
5101 }
5102 return 0;
5103 }
5104
5105 /* Round up a section size to the appropriate boundary. */
5106
5107 valueT
5108 md_section_align (segment, size)
5109 segT segment ATTRIBUTE_UNUSED;
5110 valueT size;
5111 {
5112 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
5113 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
5114 {
5115 /* For a.out, force the section size to be aligned. If we don't do
5116 this, BFD will align it for us, but it will not write out the
5117 final bytes of the section. This may be a bug in BFD, but it is
5118 easier to fix it here since that is how the other a.out targets
5119 work. */
5120 int align;
5121
5122 align = bfd_get_section_alignment (stdoutput, segment);
5123 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
5124 }
5125 #endif
5126
5127 return size;
5128 }
5129
5130 /* On the i386, PC-relative offsets are relative to the start of the
5131 next instruction. That is, the address of the offset, plus its
5132 size, since the offset is always the last part of the insn. */
5133
5134 long
5135 md_pcrel_from (fixP)
5136 fixS *fixP;
5137 {
5138 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
5139 }
5140
5141 #ifndef I386COFF
5142
5143 static void
5144 s_bss (ignore)
5145 int ignore ATTRIBUTE_UNUSED;
5146 {
5147 int temp;
5148
5149 temp = get_absolute_expression ();
5150 subseg_set (bss_section, (subsegT) temp);
5151 demand_empty_rest_of_line ();
5152 }
5153
5154 #endif
5155
5156 void
5157 i386_validate_fix (fixp)
5158 fixS *fixp;
5159 {
5160 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
5161 {
5162 /* GOTOFF relocation are nonsense in 64bit mode. */
5163 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
5164 {
5165 if (flag_code != CODE_64BIT)
5166 abort ();
5167 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
5168 }
5169 else
5170 {
5171 if (flag_code == CODE_64BIT)
5172 abort ();
5173 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
5174 }
5175 fixp->fx_subsy = 0;
5176 }
5177 }
5178
5179 arelent *
5180 tc_gen_reloc (section, fixp)
5181 asection *section ATTRIBUTE_UNUSED;
5182 fixS *fixp;
5183 {
5184 arelent *rel;
5185 bfd_reloc_code_real_type code;
5186
5187 switch (fixp->fx_r_type)
5188 {
5189 case BFD_RELOC_X86_64_PLT32:
5190 case BFD_RELOC_X86_64_GOT32:
5191 case BFD_RELOC_X86_64_GOTPCREL:
5192 case BFD_RELOC_386_PLT32:
5193 case BFD_RELOC_386_GOT32:
5194 case BFD_RELOC_386_GOTOFF:
5195 case BFD_RELOC_386_GOTPC:
5196 case BFD_RELOC_386_TLS_GD:
5197 case BFD_RELOC_386_TLS_LDM:
5198 case BFD_RELOC_386_TLS_LDO_32:
5199 case BFD_RELOC_386_TLS_IE_32:
5200 case BFD_RELOC_386_TLS_IE:
5201 case BFD_RELOC_386_TLS_GOTIE:
5202 case BFD_RELOC_386_TLS_LE_32:
5203 case BFD_RELOC_386_TLS_LE:
5204 case BFD_RELOC_X86_64_32S:
5205 case BFD_RELOC_X86_64_TLSGD:
5206 case BFD_RELOC_X86_64_TLSLD:
5207 case BFD_RELOC_X86_64_DTPOFF32:
5208 case BFD_RELOC_X86_64_GOTTPOFF:
5209 case BFD_RELOC_X86_64_TPOFF32:
5210 case BFD_RELOC_RVA:
5211 case BFD_RELOC_VTABLE_ENTRY:
5212 case BFD_RELOC_VTABLE_INHERIT:
5213 #ifdef TE_PE
5214 case BFD_RELOC_32_SECREL:
5215 #endif
5216 code = fixp->fx_r_type;
5217 break;
5218 default:
5219 if (fixp->fx_pcrel)
5220 {
5221 switch (fixp->fx_size)
5222 {
5223 default:
5224 as_bad_where (fixp->fx_file, fixp->fx_line,
5225 _("can not do %d byte pc-relative relocation"),
5226 fixp->fx_size);
5227 code = BFD_RELOC_32_PCREL;
5228 break;
5229 case 1: code = BFD_RELOC_8_PCREL; break;
5230 case 2: code = BFD_RELOC_16_PCREL; break;
5231 case 4: code = BFD_RELOC_32_PCREL; break;
5232 }
5233 }
5234 else
5235 {
5236 switch (fixp->fx_size)
5237 {
5238 default:
5239 as_bad_where (fixp->fx_file, fixp->fx_line,
5240 _("can not do %d byte relocation"),
5241 fixp->fx_size);
5242 code = BFD_RELOC_32;
5243 break;
5244 case 1: code = BFD_RELOC_8; break;
5245 case 2: code = BFD_RELOC_16; break;
5246 case 4: code = BFD_RELOC_32; break;
5247 #ifdef BFD64
5248 case 8: code = BFD_RELOC_64; break;
5249 #endif
5250 }
5251 }
5252 break;
5253 }
5254
5255 if (code == BFD_RELOC_32
5256 && GOT_symbol
5257 && fixp->fx_addsy == GOT_symbol)
5258 {
5259 /* We don't support GOTPC on 64bit targets. */
5260 if (flag_code == CODE_64BIT)
5261 abort ();
5262 code = BFD_RELOC_386_GOTPC;
5263 }
5264
5265 rel = (arelent *) xmalloc (sizeof (arelent));
5266 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
5267 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
5268
5269 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
5270
5271 #ifdef TE_PE
5272 if (S_IS_WEAK (fixp->fx_addsy))
5273 rel->addend = rel->address - (*rel->sym_ptr_ptr)->value + 4;
5274 else
5275 #endif
5276 if (!use_rela_relocations)
5277 {
5278 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
5279 vtable entry to be used in the relocation's section offset. */
5280 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
5281 rel->address = fixp->fx_offset;
5282
5283 rel->addend = 0;
5284 }
5285 /* Use the rela in 64bit mode. */
5286 else
5287 {
5288 if (!fixp->fx_pcrel)
5289 rel->addend = fixp->fx_offset;
5290 else
5291 switch (code)
5292 {
5293 case BFD_RELOC_X86_64_PLT32:
5294 case BFD_RELOC_X86_64_GOT32:
5295 case BFD_RELOC_X86_64_GOTPCREL:
5296 case BFD_RELOC_X86_64_TLSGD:
5297 case BFD_RELOC_X86_64_TLSLD:
5298 case BFD_RELOC_X86_64_GOTTPOFF:
5299 rel->addend = fixp->fx_offset - fixp->fx_size;
5300 break;
5301 default:
5302 rel->addend = (section->vma
5303 - fixp->fx_size
5304 + fixp->fx_addnumber
5305 + md_pcrel_from (fixp));
5306 break;
5307 }
5308 }
5309
5310 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
5311 if (rel->howto == NULL)
5312 {
5313 as_bad_where (fixp->fx_file, fixp->fx_line,
5314 _("cannot represent relocation type %s"),
5315 bfd_get_reloc_code_name (code));
5316 /* Set howto to a garbage value so that we can keep going. */
5317 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
5318 assert (rel->howto != NULL);
5319 }
5320
5321 return rel;
5322 }
5323
5324 \f
5325 /* Parse operands using Intel syntax. This implements a recursive descent
5326 parser based on the BNF grammar published in Appendix B of the MASM 6.1
5327 Programmer's Guide.
5328
5329 FIXME: We do not recognize the full operand grammar defined in the MASM
5330 documentation. In particular, all the structure/union and
5331 high-level macro operands are missing.
5332
5333 Uppercase words are terminals, lower case words are non-terminals.
5334 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
5335 bars '|' denote choices. Most grammar productions are implemented in
5336 functions called 'intel_<production>'.
5337
5338 Initial production is 'expr'.
5339
5340 addOp + | - | & | \| | << | >>
5341
5342 alpha [a-zA-Z]
5343
5344 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
5345
5346 constant digits [[ radixOverride ]]
5347
5348 dataType BYTE | WORD | DWORD | QWORD | XWORD
5349
5350 digits decdigit
5351 | digits decdigit
5352 | digits hexdigit
5353
5354 decdigit [0-9]
5355
5356 e05 e05 addOp e06
5357 | e06
5358
5359 e06 e06 mulOp e09
5360 | e09
5361
5362 e09 OFFSET e10
5363 | e09 PTR e10
5364 | e09 : e10
5365 | e10
5366
5367 e10 e10 [ expr ]
5368 | e11
5369
5370 e11 ( expr )
5371 | [ expr ]
5372 | constant
5373 | dataType
5374 | id
5375 | $
5376 | register
5377 | ~
5378
5379 => expr SHORT e05
5380 | e05
5381
5382 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
5383 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
5384
5385 hexdigit a | b | c | d | e | f
5386 | A | B | C | D | E | F
5387
5388 id alpha
5389 | id alpha
5390 | id decdigit
5391
5392 mulOp * | / | MOD
5393
5394 quote " | '
5395
5396 register specialRegister
5397 | gpRegister
5398 | byteRegister
5399
5400 segmentRegister CS | DS | ES | FS | GS | SS
5401
5402 specialRegister CR0 | CR2 | CR3
5403 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
5404 | TR3 | TR4 | TR5 | TR6 | TR7
5405
5406 We simplify the grammar in obvious places (e.g., register parsing is
5407 done by calling parse_register) and eliminate immediate left recursion
5408 to implement a recursive-descent parser.
5409
5410 expr SHORT e05
5411 | e05
5412
5413 e05 e06 e05'
5414
5415 e05' addOp e06 e05'
5416 | Empty
5417
5418 e06 e09 e06'
5419
5420 e06' mulOp e09 e06'
5421 | Empty
5422
5423 e09 OFFSET e10 e09'
5424 | e10 e09'
5425
5426 e09' PTR e10 e09'
5427 | : e10 e09'
5428 | Empty
5429
5430 e10 e11 e10'
5431
5432 e10' [ expr ] e10'
5433 | Empty
5434
5435 e11 ( expr )
5436 | [ expr ]
5437 | BYTE
5438 | WORD
5439 | DWORD
5440 | QWORD
5441 | XWORD
5442 | .
5443 | $
5444 | register
5445 | id
5446 | ~
5447 | constant */
5448
5449 /* Parsing structure for the intel syntax parser. Used to implement the
5450 semantic actions for the operand grammar. */
5451 struct intel_parser_s
5452 {
5453 char *op_string; /* The string being parsed. */
5454 int got_a_float; /* Whether the operand is a float. */
5455 int op_modifier; /* Operand modifier. */
5456 int is_mem; /* 1 if operand is memory reference. */
5457 const reg_entry *reg; /* Last register reference found. */
5458 char *disp; /* Displacement string being built. */
5459 };
5460
5461 static struct intel_parser_s intel_parser;
5462
5463 /* Token structure for parsing intel syntax. */
5464 struct intel_token
5465 {
5466 int code; /* Token code. */
5467 const reg_entry *reg; /* Register entry for register tokens. */
5468 char *str; /* String representation. */
5469 };
5470
5471 static struct intel_token cur_token, prev_token;
5472
5473 /* Token codes for the intel parser. Since T_SHORT is already used
5474 by COFF, undefine it first to prevent a warning. */
5475 #define T_NIL -1
5476 #define T_CONST 1
5477 #define T_REG 2
5478 #define T_BYTE 3
5479 #define T_WORD 4
5480 #define T_DWORD 5
5481 #define T_QWORD 6
5482 #define T_XWORD 7
5483 #undef T_SHORT
5484 #define T_SHORT 8
5485 #define T_OFFSET 9
5486 #define T_PTR 10
5487 #define T_ID 11
5488 #define T_SHIFTOP 12
5489
5490 /* Prototypes for intel parser functions. */
5491 static int intel_match_token PARAMS ((int code));
5492 static void intel_get_token PARAMS ((void));
5493 static void intel_putback_token PARAMS ((void));
5494 static int intel_expr PARAMS ((void));
5495 static int intel_e05 PARAMS ((void));
5496 static int intel_e05_1 PARAMS ((void));
5497 static int intel_e06 PARAMS ((void));
5498 static int intel_e06_1 PARAMS ((void));
5499 static int intel_e09 PARAMS ((void));
5500 static int intel_e09_1 PARAMS ((void));
5501 static int intel_e10 PARAMS ((void));
5502 static int intel_e10_1 PARAMS ((void));
5503 static int intel_e11 PARAMS ((void));
5504
5505 static int
5506 i386_intel_operand (operand_string, got_a_float)
5507 char *operand_string;
5508 int got_a_float;
5509 {
5510 int ret;
5511 char *p;
5512
5513 /* Initialize token holders. */
5514 cur_token.code = prev_token.code = T_NIL;
5515 cur_token.reg = prev_token.reg = NULL;
5516 cur_token.str = prev_token.str = NULL;
5517
5518 /* Initialize parser structure. */
5519 p = intel_parser.op_string = (char *) malloc (strlen (operand_string) + 1);
5520 if (p == NULL)
5521 abort ();
5522 strcpy (intel_parser.op_string, operand_string);
5523 intel_parser.got_a_float = got_a_float;
5524 intel_parser.op_modifier = -1;
5525 intel_parser.is_mem = 0;
5526 intel_parser.reg = NULL;
5527 intel_parser.disp = (char *) malloc (strlen (operand_string) + 1);
5528 if (intel_parser.disp == NULL)
5529 abort ();
5530 intel_parser.disp[0] = '\0';
5531
5532 /* Read the first token and start the parser. */
5533 intel_get_token ();
5534 ret = intel_expr ();
5535
5536 if (ret)
5537 {
5538 /* If we found a memory reference, hand it over to i386_displacement
5539 to fill in the rest of the operand fields. */
5540 if (intel_parser.is_mem)
5541 {
5542 if ((i.mem_operands == 1
5543 && (current_templates->start->opcode_modifier & IsString) == 0)
5544 || i.mem_operands == 2)
5545 {
5546 as_bad (_("too many memory references for '%s'"),
5547 current_templates->start->name);
5548 ret = 0;
5549 }
5550 else
5551 {
5552 char *s = intel_parser.disp;
5553 i.mem_operands++;
5554
5555 /* Add the displacement expression. */
5556 if (*s != '\0')
5557 ret = i386_displacement (s, s + strlen (s));
5558 if (ret)
5559 ret = i386_index_check (operand_string);
5560 }
5561 }
5562
5563 /* Constant and OFFSET expressions are handled by i386_immediate. */
5564 else if (intel_parser.op_modifier == OFFSET_FLAT
5565 || intel_parser.reg == NULL)
5566 ret = i386_immediate (intel_parser.disp);
5567 }
5568
5569 free (p);
5570 free (intel_parser.disp);
5571
5572 return ret;
5573 }
5574
5575 /* expr SHORT e05
5576 | e05 */
5577 static int
5578 intel_expr ()
5579 {
5580 /* expr SHORT e05 */
5581 if (cur_token.code == T_SHORT)
5582 {
5583 intel_parser.op_modifier = SHORT;
5584 intel_match_token (T_SHORT);
5585
5586 return (intel_e05 ());
5587 }
5588
5589 /* expr e05 */
5590 else
5591 return intel_e05 ();
5592 }
5593
5594 /* e05 e06 e05'
5595
5596 e05' addOp e06 e05'
5597 | Empty */
5598 static int
5599 intel_e05 ()
5600 {
5601 return (intel_e06 () && intel_e05_1 ());
5602 }
5603
5604 static int
5605 intel_e05_1 ()
5606 {
5607 /* e05' addOp e06 e05' */
5608 if (cur_token.code == '+' || cur_token.code == '-'
5609 || cur_token.code == '&' || cur_token.code == '|'
5610 || cur_token.code == T_SHIFTOP)
5611 {
5612 strcat (intel_parser.disp, cur_token.str);
5613 intel_match_token (cur_token.code);
5614
5615 return (intel_e06 () && intel_e05_1 ());
5616 }
5617
5618 /* e05' Empty */
5619 else
5620 return 1;
5621 }
5622
5623 /* e06 e09 e06'
5624
5625 e06' mulOp e09 e06'
5626 | Empty */
5627 static int
5628 intel_e06 ()
5629 {
5630 return (intel_e09 () && intel_e06_1 ());
5631 }
5632
5633 static int
5634 intel_e06_1 ()
5635 {
5636 /* e06' mulOp e09 e06' */
5637 if (cur_token.code == '*' || cur_token.code == '/')
5638 {
5639 strcat (intel_parser.disp, cur_token.str);
5640 intel_match_token (cur_token.code);
5641
5642 return (intel_e09 () && intel_e06_1 ());
5643 }
5644
5645 /* e06' Empty */
5646 else
5647 return 1;
5648 }
5649
5650 /* e09 OFFSET e10 e09'
5651 | e10 e09'
5652
5653 e09' PTR e10 e09'
5654 | : e10 e09'
5655 | Empty */
5656 static int
5657 intel_e09 ()
5658 {
5659 /* e09 OFFSET e10 e09' */
5660 if (cur_token.code == T_OFFSET)
5661 {
5662 intel_parser.is_mem = 0;
5663 intel_parser.op_modifier = OFFSET_FLAT;
5664 intel_match_token (T_OFFSET);
5665
5666 return (intel_e10 () && intel_e09_1 ());
5667 }
5668
5669 /* e09 e10 e09' */
5670 else
5671 return (intel_e10 () && intel_e09_1 ());
5672 }
5673
5674 static int
5675 intel_e09_1 ()
5676 {
5677 /* e09' PTR e10 e09' */
5678 if (cur_token.code == T_PTR)
5679 {
5680 if (prev_token.code == T_BYTE)
5681 i.suffix = BYTE_MNEM_SUFFIX;
5682
5683 else if (prev_token.code == T_WORD)
5684 {
5685 if (intel_parser.got_a_float == 2) /* "fi..." */
5686 i.suffix = SHORT_MNEM_SUFFIX;
5687 else
5688 i.suffix = WORD_MNEM_SUFFIX;
5689 }
5690
5691 else if (prev_token.code == T_DWORD)
5692 {
5693 if (intel_parser.got_a_float == 1) /* "f..." */
5694 i.suffix = SHORT_MNEM_SUFFIX;
5695 else
5696 i.suffix = LONG_MNEM_SUFFIX;
5697 }
5698
5699 else if (prev_token.code == T_QWORD)
5700 {
5701 if (intel_parser.got_a_float == 1) /* "f..." */
5702 i.suffix = LONG_MNEM_SUFFIX;
5703 else
5704 i.suffix = QWORD_MNEM_SUFFIX;
5705 }
5706
5707 else if (prev_token.code == T_XWORD)
5708 i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
5709
5710 else
5711 {
5712 as_bad (_("Unknown operand modifier `%s'"), prev_token.str);
5713 return 0;
5714 }
5715
5716 intel_match_token (T_PTR);
5717
5718 return (intel_e10 () && intel_e09_1 ());
5719 }
5720
5721 /* e09 : e10 e09' */
5722 else if (cur_token.code == ':')
5723 {
5724 /* Mark as a memory operand only if it's not already known to be an
5725 offset expression. */
5726 if (intel_parser.op_modifier != OFFSET_FLAT)
5727 intel_parser.is_mem = 1;
5728
5729 return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
5730 }
5731
5732 /* e09' Empty */
5733 else
5734 return 1;
5735 }
5736
5737 /* e10 e11 e10'
5738
5739 e10' [ expr ] e10'
5740 | Empty */
5741 static int
5742 intel_e10 ()
5743 {
5744 return (intel_e11 () && intel_e10_1 ());
5745 }
5746
5747 static int
5748 intel_e10_1 ()
5749 {
5750 /* e10' [ expr ] e10' */
5751 if (cur_token.code == '[')
5752 {
5753 intel_match_token ('[');
5754
5755 /* Mark as a memory operand only if it's not already known to be an
5756 offset expression. If it's an offset expression, we need to keep
5757 the brace in. */
5758 if (intel_parser.op_modifier != OFFSET_FLAT)
5759 intel_parser.is_mem = 1;
5760 else
5761 strcat (intel_parser.disp, "[");
5762
5763 /* Add a '+' to the displacement string if necessary. */
5764 if (*intel_parser.disp != '\0'
5765 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5766 strcat (intel_parser.disp, "+");
5767
5768 if (intel_expr () && intel_match_token (']'))
5769 {
5770 /* Preserve brackets when the operand is an offset expression. */
5771 if (intel_parser.op_modifier == OFFSET_FLAT)
5772 strcat (intel_parser.disp, "]");
5773
5774 return intel_e10_1 ();
5775 }
5776 else
5777 return 0;
5778 }
5779
5780 /* e10' Empty */
5781 else
5782 return 1;
5783 }
5784
5785 /* e11 ( expr )
5786 | [ expr ]
5787 | BYTE
5788 | WORD
5789 | DWORD
5790 | QWORD
5791 | XWORD
5792 | $
5793 | .
5794 | register
5795 | id
5796 | ~
5797 | constant */
5798 static int
5799 intel_e11 ()
5800 {
5801 /* e11 ( expr ) */
5802 if (cur_token.code == '(')
5803 {
5804 intel_match_token ('(');
5805 strcat (intel_parser.disp, "(");
5806
5807 if (intel_expr () && intel_match_token (')'))
5808 {
5809 strcat (intel_parser.disp, ")");
5810 return 1;
5811 }
5812 else
5813 return 0;
5814 }
5815
5816 /* e11 ~ expr */
5817 else if (cur_token.code == '~')
5818 {
5819 strcat (intel_parser.disp, "~");
5820 intel_match_token ('~');
5821
5822 return (intel_e11 ());
5823 }
5824
5825 /* e11 [ expr ] */
5826 else if (cur_token.code == '[')
5827 {
5828 intel_match_token ('[');
5829
5830 /* Mark as a memory operand only if it's not already known to be an
5831 offset expression. If it's an offset expression, we need to keep
5832 the brace in. */
5833 if (intel_parser.op_modifier != OFFSET_FLAT)
5834 intel_parser.is_mem = 1;
5835 else
5836 strcat (intel_parser.disp, "[");
5837
5838 /* Operands for jump/call inside brackets denote absolute addresses. */
5839 if (current_templates->start->opcode_modifier & Jump
5840 || current_templates->start->opcode_modifier & JumpDword
5841 || current_templates->start->opcode_modifier & JumpByte
5842 || current_templates->start->opcode_modifier & JumpInterSegment)
5843 i.types[this_operand] |= JumpAbsolute;
5844
5845 /* Add a '+' to the displacement string if necessary. */
5846 if (*intel_parser.disp != '\0'
5847 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5848 strcat (intel_parser.disp, "+");
5849
5850 if (intel_expr () && intel_match_token (']'))
5851 {
5852 /* Preserve brackets when the operand is an offset expression. */
5853 if (intel_parser.op_modifier == OFFSET_FLAT)
5854 strcat (intel_parser.disp, "]");
5855
5856 return 1;
5857 }
5858 else
5859 return 0;
5860 }
5861
5862 /* e11 BYTE
5863 | WORD
5864 | DWORD
5865 | QWORD
5866 | XWORD */
5867 else if (cur_token.code == T_BYTE
5868 || cur_token.code == T_WORD
5869 || cur_token.code == T_DWORD
5870 || cur_token.code == T_QWORD
5871 || cur_token.code == T_XWORD)
5872 {
5873 intel_match_token (cur_token.code);
5874
5875 return 1;
5876 }
5877
5878 /* e11 $
5879 | . */
5880 else if (cur_token.code == '$' || cur_token.code == '.')
5881 {
5882 strcat (intel_parser.disp, cur_token.str);
5883 intel_match_token (cur_token.code);
5884
5885 /* Mark as a memory operand only if it's not already known to be an
5886 offset expression. */
5887 if (intel_parser.op_modifier != OFFSET_FLAT)
5888 intel_parser.is_mem = 1;
5889
5890 return 1;
5891 }
5892
5893 /* e11 register */
5894 else if (cur_token.code == T_REG)
5895 {
5896 const reg_entry *reg = intel_parser.reg = cur_token.reg;
5897
5898 intel_match_token (T_REG);
5899
5900 /* Check for segment change. */
5901 if (cur_token.code == ':')
5902 {
5903 if (reg->reg_type & (SReg2 | SReg3))
5904 {
5905 switch (reg->reg_num)
5906 {
5907 case 0:
5908 i.seg[i.mem_operands] = &es;
5909 break;
5910 case 1:
5911 i.seg[i.mem_operands] = &cs;
5912 break;
5913 case 2:
5914 i.seg[i.mem_operands] = &ss;
5915 break;
5916 case 3:
5917 i.seg[i.mem_operands] = &ds;
5918 break;
5919 case 4:
5920 i.seg[i.mem_operands] = &fs;
5921 break;
5922 case 5:
5923 i.seg[i.mem_operands] = &gs;
5924 break;
5925 }
5926 }
5927 else
5928 {
5929 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
5930 return 0;
5931 }
5932 }
5933
5934 /* Not a segment register. Check for register scaling. */
5935 else if (cur_token.code == '*')
5936 {
5937 if (!intel_parser.is_mem)
5938 {
5939 as_bad (_("Register scaling only allowed in memory operands."));
5940 return 0;
5941 }
5942
5943 /* What follows must be a valid scale. */
5944 if (intel_match_token ('*')
5945 && strchr ("01248", *cur_token.str))
5946 {
5947 i.index_reg = reg;
5948 i.types[this_operand] |= BaseIndex;
5949
5950 /* Set the scale after setting the register (otherwise,
5951 i386_scale will complain) */
5952 i386_scale (cur_token.str);
5953 intel_match_token (T_CONST);
5954 }
5955 else
5956 {
5957 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
5958 cur_token.str);
5959 return 0;
5960 }
5961 }
5962
5963 /* No scaling. If this is a memory operand, the register is either a
5964 base register (first occurrence) or an index register (second
5965 occurrence). */
5966 else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3)))
5967 {
5968 if (i.base_reg && i.index_reg)
5969 {
5970 as_bad (_("Too many register references in memory operand."));
5971 return 0;
5972 }
5973
5974 if (i.base_reg == NULL)
5975 i.base_reg = reg;
5976 else
5977 i.index_reg = reg;
5978
5979 i.types[this_operand] |= BaseIndex;
5980 }
5981
5982 /* Offset modifier. Add the register to the displacement string to be
5983 parsed as an immediate expression after we're done. */
5984 else if (intel_parser.op_modifier == OFFSET_FLAT)
5985 strcat (intel_parser.disp, reg->reg_name);
5986
5987 /* It's neither base nor index nor offset. */
5988 else
5989 {
5990 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
5991 i.op[this_operand].regs = reg;
5992 i.reg_operands++;
5993 }
5994
5995 /* Since registers are not part of the displacement string (except
5996 when we're parsing offset operands), we may need to remove any
5997 preceding '+' from the displacement string. */
5998 if (*intel_parser.disp != '\0'
5999 && intel_parser.op_modifier != OFFSET_FLAT)
6000 {
6001 char *s = intel_parser.disp;
6002 s += strlen (s) - 1;
6003 if (*s == '+')
6004 *s = '\0';
6005 }
6006
6007 return 1;
6008 }
6009
6010 /* e11 id */
6011 else if (cur_token.code == T_ID)
6012 {
6013 /* Add the identifier to the displacement string. */
6014 strcat (intel_parser.disp, cur_token.str);
6015 intel_match_token (T_ID);
6016
6017 /* The identifier represents a memory reference only if it's not
6018 preceded by an offset modifier. */
6019 if (intel_parser.op_modifier != OFFSET_FLAT)
6020 intel_parser.is_mem = 1;
6021
6022 return 1;
6023 }
6024
6025 /* e11 constant */
6026 else if (cur_token.code == T_CONST
6027 || cur_token.code == '-'
6028 || cur_token.code == '+')
6029 {
6030 char *save_str;
6031
6032 /* Allow constants that start with `+' or `-'. */
6033 if (cur_token.code == '-' || cur_token.code == '+')
6034 {
6035 strcat (intel_parser.disp, cur_token.str);
6036 intel_match_token (cur_token.code);
6037 if (cur_token.code != T_CONST)
6038 {
6039 as_bad (_("Syntax error. Expecting a constant. Got `%s'."),
6040 cur_token.str);
6041 return 0;
6042 }
6043 }
6044
6045 save_str = (char *) malloc (strlen (cur_token.str) + 1);
6046 if (save_str == NULL)
6047 abort ();
6048 strcpy (save_str, cur_token.str);
6049
6050 /* Get the next token to check for register scaling. */
6051 intel_match_token (cur_token.code);
6052
6053 /* Check if this constant is a scaling factor for an index register. */
6054 if (cur_token.code == '*')
6055 {
6056 if (intel_match_token ('*') && cur_token.code == T_REG)
6057 {
6058 if (!intel_parser.is_mem)
6059 {
6060 as_bad (_("Register scaling only allowed in memory operands."));
6061 return 0;
6062 }
6063
6064 /* The constant is followed by `* reg', so it must be
6065 a valid scale. */
6066 if (strchr ("01248", *save_str))
6067 {
6068 i.index_reg = cur_token.reg;
6069 i.types[this_operand] |= BaseIndex;
6070
6071 /* Set the scale after setting the register (otherwise,
6072 i386_scale will complain) */
6073 i386_scale (save_str);
6074 intel_match_token (T_REG);
6075
6076 /* Since registers are not part of the displacement
6077 string, we may need to remove any preceding '+' from
6078 the displacement string. */
6079 if (*intel_parser.disp != '\0')
6080 {
6081 char *s = intel_parser.disp;
6082 s += strlen (s) - 1;
6083 if (*s == '+')
6084 *s = '\0';
6085 }
6086
6087 free (save_str);
6088
6089 return 1;
6090 }
6091 else
6092 return 0;
6093 }
6094
6095 /* The constant was not used for register scaling. Since we have
6096 already consumed the token following `*' we now need to put it
6097 back in the stream. */
6098 else
6099 intel_putback_token ();
6100 }
6101
6102 /* Add the constant to the displacement string. */
6103 strcat (intel_parser.disp, save_str);
6104 free (save_str);
6105
6106 return 1;
6107 }
6108
6109 as_bad (_("Unrecognized token '%s'"), cur_token.str);
6110 return 0;
6111 }
6112
6113 /* Match the given token against cur_token. If they match, read the next
6114 token from the operand string. */
6115 static int
6116 intel_match_token (code)
6117 int code;
6118 {
6119 if (cur_token.code == code)
6120 {
6121 intel_get_token ();
6122 return 1;
6123 }
6124 else
6125 {
6126 as_bad (_("Unexpected token `%s'"), cur_token.str);
6127 return 0;
6128 }
6129 }
6130
6131 /* Read a new token from intel_parser.op_string and store it in cur_token. */
6132 static void
6133 intel_get_token ()
6134 {
6135 char *end_op;
6136 const reg_entry *reg;
6137 struct intel_token new_token;
6138
6139 new_token.code = T_NIL;
6140 new_token.reg = NULL;
6141 new_token.str = NULL;
6142
6143 /* Free the memory allocated to the previous token and move
6144 cur_token to prev_token. */
6145 if (prev_token.str)
6146 free (prev_token.str);
6147
6148 prev_token = cur_token;
6149
6150 /* Skip whitespace. */
6151 while (is_space_char (*intel_parser.op_string))
6152 intel_parser.op_string++;
6153
6154 /* Return an empty token if we find nothing else on the line. */
6155 if (*intel_parser.op_string == '\0')
6156 {
6157 cur_token = new_token;
6158 return;
6159 }
6160
6161 /* The new token cannot be larger than the remainder of the operand
6162 string. */
6163 new_token.str = (char *) malloc (strlen (intel_parser.op_string) + 1);
6164 if (new_token.str == NULL)
6165 abort ();
6166 new_token.str[0] = '\0';
6167
6168 if (strchr ("0123456789", *intel_parser.op_string))
6169 {
6170 char *p = new_token.str;
6171 char *q = intel_parser.op_string;
6172 new_token.code = T_CONST;
6173
6174 /* Allow any kind of identifier char to encompass floating point and
6175 hexadecimal numbers. */
6176 while (is_identifier_char (*q))
6177 *p++ = *q++;
6178 *p = '\0';
6179
6180 /* Recognize special symbol names [0-9][bf]. */
6181 if (strlen (intel_parser.op_string) == 2
6182 && (intel_parser.op_string[1] == 'b'
6183 || intel_parser.op_string[1] == 'f'))
6184 new_token.code = T_ID;
6185 }
6186
6187 else if (strchr ("<>", *intel_parser.op_string)
6188 && *intel_parser.op_string == *(intel_parser.op_string + 1))
6189 {
6190 new_token.code = T_SHIFTOP;
6191 new_token.str[0] = *intel_parser.op_string;
6192 new_token.str[1] = *intel_parser.op_string;
6193 new_token.str[2] = '\0';
6194 }
6195
6196 else if (strchr ("+-/*&|:[]()~", *intel_parser.op_string))
6197 {
6198 new_token.code = *intel_parser.op_string;
6199 new_token.str[0] = *intel_parser.op_string;
6200 new_token.str[1] = '\0';
6201 }
6202
6203 else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
6204 && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
6205 {
6206 new_token.code = T_REG;
6207 new_token.reg = reg;
6208
6209 if (*intel_parser.op_string == REGISTER_PREFIX)
6210 {
6211 new_token.str[0] = REGISTER_PREFIX;
6212 new_token.str[1] = '\0';
6213 }
6214
6215 strcat (new_token.str, reg->reg_name);
6216 }
6217
6218 else if (is_identifier_char (*intel_parser.op_string))
6219 {
6220 char *p = new_token.str;
6221 char *q = intel_parser.op_string;
6222
6223 /* A '.' or '$' followed by an identifier char is an identifier.
6224 Otherwise, it's operator '.' followed by an expression. */
6225 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
6226 {
6227 new_token.code = *q;
6228 new_token.str[0] = *q;
6229 new_token.str[1] = '\0';
6230 }
6231 else
6232 {
6233 while (is_identifier_char (*q) || *q == '@')
6234 *p++ = *q++;
6235 *p = '\0';
6236
6237 if (strcasecmp (new_token.str, "BYTE") == 0)
6238 new_token.code = T_BYTE;
6239
6240 else if (strcasecmp (new_token.str, "WORD") == 0)
6241 new_token.code = T_WORD;
6242
6243 else if (strcasecmp (new_token.str, "DWORD") == 0)
6244 new_token.code = T_DWORD;
6245
6246 else if (strcasecmp (new_token.str, "QWORD") == 0)
6247 new_token.code = T_QWORD;
6248
6249 else if (strcasecmp (new_token.str, "XWORD") == 0)
6250 new_token.code = T_XWORD;
6251
6252 else if (strcasecmp (new_token.str, "PTR") == 0)
6253 new_token.code = T_PTR;
6254
6255 else if (strcasecmp (new_token.str, "SHORT") == 0)
6256 new_token.code = T_SHORT;
6257
6258 else if (strcasecmp (new_token.str, "OFFSET") == 0)
6259 {
6260 new_token.code = T_OFFSET;
6261
6262 /* ??? This is not mentioned in the MASM grammar but gcc
6263 makes use of it with -mintel-syntax. OFFSET may be
6264 followed by FLAT: */
6265 if (strncasecmp (q, " FLAT:", 6) == 0)
6266 strcat (new_token.str, " FLAT:");
6267 }
6268
6269 /* ??? This is not mentioned in the MASM grammar. */
6270 else if (strcasecmp (new_token.str, "FLAT") == 0)
6271 new_token.code = T_OFFSET;
6272
6273 else
6274 new_token.code = T_ID;
6275 }
6276 }
6277
6278 else
6279 as_bad (_("Unrecognized token `%s'"), intel_parser.op_string);
6280
6281 intel_parser.op_string += strlen (new_token.str);
6282 cur_token = new_token;
6283 }
6284
6285 /* Put cur_token back into the token stream and make cur_token point to
6286 prev_token. */
6287 static void
6288 intel_putback_token ()
6289 {
6290 intel_parser.op_string -= strlen (cur_token.str);
6291 free (cur_token.str);
6292 cur_token = prev_token;
6293
6294 /* Forget prev_token. */
6295 prev_token.code = T_NIL;
6296 prev_token.reg = NULL;
6297 prev_token.str = NULL;
6298 }
6299
6300 int
6301 tc_x86_regname_to_dw2regnum (const char *regname)
6302 {
6303 unsigned int regnum;
6304 unsigned int regnames_count;
6305 char *regnames_32[] =
6306 {
6307 "eax", "ecx", "edx", "ebx",
6308 "esp", "ebp", "esi", "edi",
6309 "eip"
6310 };
6311 char *regnames_64[] =
6312 {
6313 "rax", "rbx", "rcx", "rdx",
6314 "rdi", "rsi", "rbp", "rsp",
6315 "r8", "r9", "r10", "r11",
6316 "r12", "r13", "r14", "r15",
6317 "rip"
6318 };
6319 char **regnames;
6320
6321 if (flag_code == CODE_64BIT)
6322 {
6323 regnames = regnames_64;
6324 regnames_count = ARRAY_SIZE (regnames_64);
6325 }
6326 else
6327 {
6328 regnames = regnames_32;
6329 regnames_count = ARRAY_SIZE (regnames_32);
6330 }
6331
6332 for (regnum = 0; regnum < regnames_count; regnum++)
6333 if (strcmp (regname, regnames[regnum]) == 0)
6334 return regnum;
6335
6336 return -1;
6337 }
6338
6339 void
6340 tc_x86_frame_initial_instructions (void)
6341 {
6342 static unsigned int sp_regno;
6343
6344 if (!sp_regno)
6345 sp_regno = tc_x86_regname_to_dw2regnum (flag_code == CODE_64BIT
6346 ? "rsp" : "esp");
6347
6348 cfi_add_CFA_def_cfa (sp_regno, -x86_cie_data_alignment);
6349 cfi_add_CFA_offset (x86_dwarf2_return_column, x86_cie_data_alignment);
6350 }