* config/tc-m68k.c: Convert to C90. Remove redundant
[deliverable/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.\n"),
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 | Imm64;
1926 break;
1927 case WORD_MNEM_SUFFIX:
1928 i.types[op] = Imm16 | Imm32 | Imm64;
1929 break;
1930 break;
1931 case BYTE_MNEM_SUFFIX:
1932 i.types[op] = Imm8 | Imm8S | Imm16 | Imm32S | Imm32;
1933 break;
1934 break;
1935 }
1936 break;
1937 }
1938 }
1939 }
1940
1941 /* Try to use the smallest displacement type too. */
1942 static void
1943 optimize_disp ()
1944 {
1945 int op;
1946
1947 for (op = i.operands; --op >= 0;)
1948 if ((i.types[op] & Disp) && i.op[op].disps->X_op == O_constant)
1949 {
1950 offsetT disp = i.op[op].disps->X_add_number;
1951
1952 if (i.types[op] & Disp16)
1953 {
1954 /* We know this operand is at most 16 bits, so
1955 convert to a signed 16 bit number before trying
1956 to see whether it will fit in an even smaller
1957 size. */
1958
1959 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
1960 }
1961 else if (i.types[op] & Disp32)
1962 {
1963 /* We know this operand is at most 32 bits, so convert to a
1964 signed 32 bit number before trying to see whether it will
1965 fit in an even smaller size. */
1966 disp &= (((offsetT) 2 << 31) - 1);
1967 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1968 }
1969 if (flag_code == CODE_64BIT)
1970 {
1971 if (fits_in_signed_long (disp))
1972 i.types[op] |= Disp32S;
1973 if (fits_in_unsigned_long (disp))
1974 i.types[op] |= Disp32;
1975 }
1976 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
1977 && fits_in_signed_byte (disp))
1978 i.types[op] |= Disp8;
1979 }
1980 }
1981
1982 static int
1983 match_template ()
1984 {
1985 /* Points to template once we've found it. */
1986 const template *t;
1987 unsigned int overlap0, overlap1, overlap2;
1988 unsigned int found_reverse_match;
1989 int suffix_check;
1990
1991 #define MATCH(overlap, given, template) \
1992 ((overlap & ~JumpAbsolute) \
1993 && (((given) & (BaseIndex | JumpAbsolute)) \
1994 == ((overlap) & (BaseIndex | JumpAbsolute))))
1995
1996 /* If given types r0 and r1 are registers they must be of the same type
1997 unless the expected operand type register overlap is null.
1998 Note that Acc in a template matches every size of reg. */
1999 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
2000 (((g0) & Reg) == 0 || ((g1) & Reg) == 0 \
2001 || ((g0) & Reg) == ((g1) & Reg) \
2002 || ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
2003
2004 overlap0 = 0;
2005 overlap1 = 0;
2006 overlap2 = 0;
2007 found_reverse_match = 0;
2008 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
2009 ? No_bSuf
2010 : (i.suffix == WORD_MNEM_SUFFIX
2011 ? No_wSuf
2012 : (i.suffix == SHORT_MNEM_SUFFIX
2013 ? No_sSuf
2014 : (i.suffix == LONG_MNEM_SUFFIX
2015 ? No_lSuf
2016 : (i.suffix == QWORD_MNEM_SUFFIX
2017 ? No_qSuf
2018 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX
2019 ? No_xSuf : 0))))));
2020
2021 for (t = current_templates->start;
2022 t < current_templates->end;
2023 t++)
2024 {
2025 /* Must have right number of operands. */
2026 if (i.operands != t->operands)
2027 continue;
2028
2029 /* Check the suffix, except for some instructions in intel mode. */
2030 if ((t->opcode_modifier & suffix_check)
2031 && !(intel_syntax
2032 && (t->opcode_modifier & IgnoreSize))
2033 && !(intel_syntax
2034 && t->base_opcode == 0xd9
2035 && (t->extension_opcode == 5 /* 0xd9,5 "fldcw" */
2036 || t->extension_opcode == 7))) /* 0xd9,7 "f{n}stcw" */
2037 continue;
2038
2039 /* Do not verify operands when there are none. */
2040 else if (!t->operands)
2041 {
2042 if (t->cpu_flags & ~cpu_arch_flags)
2043 continue;
2044 /* We've found a match; break out of loop. */
2045 break;
2046 }
2047
2048 overlap0 = i.types[0] & t->operand_types[0];
2049 switch (t->operands)
2050 {
2051 case 1:
2052 if (!MATCH (overlap0, i.types[0], t->operand_types[0]))
2053 continue;
2054 break;
2055 case 2:
2056 case 3:
2057 overlap1 = i.types[1] & t->operand_types[1];
2058 if (!MATCH (overlap0, i.types[0], t->operand_types[0])
2059 || !MATCH (overlap1, i.types[1], t->operand_types[1])
2060 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2061 t->operand_types[0],
2062 overlap1, i.types[1],
2063 t->operand_types[1]))
2064 {
2065 /* Check if other direction is valid ... */
2066 if ((t->opcode_modifier & (D | FloatD)) == 0)
2067 continue;
2068
2069 /* Try reversing direction of operands. */
2070 overlap0 = i.types[0] & t->operand_types[1];
2071 overlap1 = i.types[1] & t->operand_types[0];
2072 if (!MATCH (overlap0, i.types[0], t->operand_types[1])
2073 || !MATCH (overlap1, i.types[1], t->operand_types[0])
2074 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2075 t->operand_types[1],
2076 overlap1, i.types[1],
2077 t->operand_types[0]))
2078 {
2079 /* Does not match either direction. */
2080 continue;
2081 }
2082 /* found_reverse_match holds which of D or FloatDR
2083 we've found. */
2084 found_reverse_match = t->opcode_modifier & (D | FloatDR);
2085 }
2086 /* Found a forward 2 operand match here. */
2087 else if (t->operands == 3)
2088 {
2089 /* Here we make use of the fact that there are no
2090 reverse match 3 operand instructions, and all 3
2091 operand instructions only need to be checked for
2092 register consistency between operands 2 and 3. */
2093 overlap2 = i.types[2] & t->operand_types[2];
2094 if (!MATCH (overlap2, i.types[2], t->operand_types[2])
2095 || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1],
2096 t->operand_types[1],
2097 overlap2, i.types[2],
2098 t->operand_types[2]))
2099
2100 continue;
2101 }
2102 /* Found either forward/reverse 2 or 3 operand match here:
2103 slip through to break. */
2104 }
2105 if (t->cpu_flags & ~cpu_arch_flags)
2106 {
2107 found_reverse_match = 0;
2108 continue;
2109 }
2110 /* We've found a match; break out of loop. */
2111 break;
2112 }
2113
2114 if (t == current_templates->end)
2115 {
2116 /* We found no match. */
2117 as_bad (_("suffix or operands invalid for `%s'"),
2118 current_templates->start->name);
2119 return 0;
2120 }
2121
2122 if (!quiet_warnings)
2123 {
2124 if (!intel_syntax
2125 && ((i.types[0] & JumpAbsolute)
2126 != (t->operand_types[0] & JumpAbsolute)))
2127 {
2128 as_warn (_("indirect %s without `*'"), t->name);
2129 }
2130
2131 if ((t->opcode_modifier & (IsPrefix | IgnoreSize))
2132 == (IsPrefix | IgnoreSize))
2133 {
2134 /* Warn them that a data or address size prefix doesn't
2135 affect assembly of the next line of code. */
2136 as_warn (_("stand-alone `%s' prefix"), t->name);
2137 }
2138 }
2139
2140 /* Copy the template we found. */
2141 i.tm = *t;
2142 if (found_reverse_match)
2143 {
2144 /* If we found a reverse match we must alter the opcode
2145 direction bit. found_reverse_match holds bits to change
2146 (different for int & float insns). */
2147
2148 i.tm.base_opcode ^= found_reverse_match;
2149
2150 i.tm.operand_types[0] = t->operand_types[1];
2151 i.tm.operand_types[1] = t->operand_types[0];
2152 }
2153
2154 return 1;
2155 }
2156
2157 static int
2158 check_string ()
2159 {
2160 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
2161 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
2162 {
2163 if (i.seg[0] != NULL && i.seg[0] != &es)
2164 {
2165 as_bad (_("`%s' operand %d must use `%%es' segment"),
2166 i.tm.name,
2167 mem_op + 1);
2168 return 0;
2169 }
2170 /* There's only ever one segment override allowed per instruction.
2171 This instruction possibly has a legal segment override on the
2172 second operand, so copy the segment to where non-string
2173 instructions store it, allowing common code. */
2174 i.seg[0] = i.seg[1];
2175 }
2176 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
2177 {
2178 if (i.seg[1] != NULL && i.seg[1] != &es)
2179 {
2180 as_bad (_("`%s' operand %d must use `%%es' segment"),
2181 i.tm.name,
2182 mem_op + 2);
2183 return 0;
2184 }
2185 }
2186 return 1;
2187 }
2188
2189 static int
2190 process_suffix ()
2191 {
2192 /* If matched instruction specifies an explicit instruction mnemonic
2193 suffix, use it. */
2194 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
2195 {
2196 if (i.tm.opcode_modifier & Size16)
2197 i.suffix = WORD_MNEM_SUFFIX;
2198 else if (i.tm.opcode_modifier & Size64)
2199 i.suffix = QWORD_MNEM_SUFFIX;
2200 else
2201 i.suffix = LONG_MNEM_SUFFIX;
2202 }
2203 else if (i.reg_operands)
2204 {
2205 /* If there's no instruction mnemonic suffix we try to invent one
2206 based on register operands. */
2207 if (!i.suffix)
2208 {
2209 /* We take i.suffix from the last register operand specified,
2210 Destination register type is more significant than source
2211 register type. */
2212 int op;
2213 for (op = i.operands; --op >= 0;)
2214 if ((i.types[op] & Reg)
2215 && !(i.tm.operand_types[op] & InOutPortReg))
2216 {
2217 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
2218 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
2219 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
2220 LONG_MNEM_SUFFIX);
2221 break;
2222 }
2223 }
2224 else if (i.suffix == BYTE_MNEM_SUFFIX)
2225 {
2226 if (!check_byte_reg ())
2227 return 0;
2228 }
2229 else if (i.suffix == LONG_MNEM_SUFFIX)
2230 {
2231 if (!check_long_reg ())
2232 return 0;
2233 }
2234 else if (i.suffix == QWORD_MNEM_SUFFIX)
2235 {
2236 if (!check_qword_reg ())
2237 return 0;
2238 }
2239 else if (i.suffix == WORD_MNEM_SUFFIX)
2240 {
2241 if (!check_word_reg ())
2242 return 0;
2243 }
2244 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2245 /* Do nothing if the instruction is going to ignore the prefix. */
2246 ;
2247 else
2248 abort ();
2249 }
2250 else if ((i.tm.opcode_modifier & DefaultSize) && !i.suffix)
2251 {
2252 i.suffix = stackop_size;
2253 }
2254
2255 /* Change the opcode based on the operand size given by i.suffix;
2256 We need not change things for byte insns. */
2257
2258 if (!i.suffix && (i.tm.opcode_modifier & W))
2259 {
2260 as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
2261 return 0;
2262 }
2263
2264 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2265 {
2266 /* It's not a byte, select word/dword operation. */
2267 if (i.tm.opcode_modifier & W)
2268 {
2269 if (i.tm.opcode_modifier & ShortForm)
2270 i.tm.base_opcode |= 8;
2271 else
2272 i.tm.base_opcode |= 1;
2273 }
2274
2275 /* Now select between word & dword operations via the operand
2276 size prefix, except for instructions that will ignore this
2277 prefix anyway. */
2278 if (i.suffix != QWORD_MNEM_SUFFIX
2279 && !(i.tm.opcode_modifier & IgnoreSize)
2280 && ((i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2281 || (flag_code == CODE_64BIT
2282 && (i.tm.opcode_modifier & JumpByte))))
2283 {
2284 unsigned int prefix = DATA_PREFIX_OPCODE;
2285 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2286 prefix = ADDR_PREFIX_OPCODE;
2287
2288 if (!add_prefix (prefix))
2289 return 0;
2290 }
2291
2292 /* Set mode64 for an operand. */
2293 if (i.suffix == QWORD_MNEM_SUFFIX
2294 && flag_code == CODE_64BIT
2295 && (i.tm.opcode_modifier & NoRex64) == 0)
2296 i.rex |= REX_MODE64;
2297
2298 /* Size floating point instruction. */
2299 if (i.suffix == LONG_MNEM_SUFFIX)
2300 {
2301 if (i.tm.opcode_modifier & FloatMF)
2302 i.tm.base_opcode ^= 4;
2303 }
2304 }
2305
2306 return 1;
2307 }
2308
2309 static int
2310 check_byte_reg ()
2311 {
2312 int op;
2313 for (op = i.operands; --op >= 0;)
2314 {
2315 /* If this is an eight bit register, it's OK. If it's the 16 or
2316 32 bit version of an eight bit register, we will just use the
2317 low portion, and that's OK too. */
2318 if (i.types[op] & Reg8)
2319 continue;
2320
2321 /* movzx and movsx should not generate this warning. */
2322 if (intel_syntax
2323 && (i.tm.base_opcode == 0xfb7
2324 || i.tm.base_opcode == 0xfb6
2325 || i.tm.base_opcode == 0x63
2326 || i.tm.base_opcode == 0xfbe
2327 || i.tm.base_opcode == 0xfbf))
2328 continue;
2329
2330 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4
2331 #if 0
2332 /* Check that the template allows eight bit regs. This
2333 kills insns such as `orb $1,%edx', which maybe should be
2334 allowed. */
2335 && (i.tm.operand_types[op] & (Reg8 | InOutPortReg))
2336 #endif
2337 )
2338 {
2339 /* Prohibit these changes in the 64bit mode, since the
2340 lowering is more complicated. */
2341 if (flag_code == CODE_64BIT
2342 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2343 {
2344 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2345 i.op[op].regs->reg_name,
2346 i.suffix);
2347 return 0;
2348 }
2349 #if REGISTER_WARNINGS
2350 if (!quiet_warnings
2351 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2352 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2353 (i.op[op].regs + (i.types[op] & Reg16
2354 ? REGNAM_AL - REGNAM_AX
2355 : REGNAM_AL - REGNAM_EAX))->reg_name,
2356 i.op[op].regs->reg_name,
2357 i.suffix);
2358 #endif
2359 continue;
2360 }
2361 /* Any other register is bad. */
2362 if (i.types[op] & (Reg | RegMMX | RegXMM
2363 | SReg2 | SReg3
2364 | Control | Debug | Test
2365 | FloatReg | FloatAcc))
2366 {
2367 as_bad (_("`%%%s' not allowed with `%s%c'"),
2368 i.op[op].regs->reg_name,
2369 i.tm.name,
2370 i.suffix);
2371 return 0;
2372 }
2373 }
2374 return 1;
2375 }
2376
2377 static int
2378 check_long_reg ()
2379 {
2380 int op;
2381
2382 for (op = i.operands; --op >= 0;)
2383 /* Reject eight bit registers, except where the template requires
2384 them. (eg. movzb) */
2385 if ((i.types[op] & Reg8) != 0
2386 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2387 {
2388 as_bad (_("`%%%s' not allowed with `%s%c'"),
2389 i.op[op].regs->reg_name,
2390 i.tm.name,
2391 i.suffix);
2392 return 0;
2393 }
2394 /* Warn if the e prefix on a general reg is missing. */
2395 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2396 && (i.types[op] & Reg16) != 0
2397 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2398 {
2399 /* Prohibit these changes in the 64bit mode, since the
2400 lowering is more complicated. */
2401 if (flag_code == CODE_64BIT)
2402 {
2403 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2404 i.op[op].regs->reg_name,
2405 i.suffix);
2406 return 0;
2407 }
2408 #if REGISTER_WARNINGS
2409 else
2410 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2411 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
2412 i.op[op].regs->reg_name,
2413 i.suffix);
2414 #endif
2415 }
2416 /* Warn if the r prefix on a general reg is missing. */
2417 else if ((i.types[op] & Reg64) != 0
2418 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2419 {
2420 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2421 i.op[op].regs->reg_name,
2422 i.suffix);
2423 return 0;
2424 }
2425 return 1;
2426 }
2427
2428 static int
2429 check_qword_reg ()
2430 {
2431 int op;
2432
2433 for (op = i.operands; --op >= 0; )
2434 /* Reject eight bit registers, except where the template requires
2435 them. (eg. movzb) */
2436 if ((i.types[op] & Reg8) != 0
2437 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2438 {
2439 as_bad (_("`%%%s' not allowed with `%s%c'"),
2440 i.op[op].regs->reg_name,
2441 i.tm.name,
2442 i.suffix);
2443 return 0;
2444 }
2445 /* Warn if the e prefix on a general reg is missing. */
2446 else if (((i.types[op] & Reg16) != 0
2447 || (i.types[op] & Reg32) != 0)
2448 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2449 {
2450 /* Prohibit these changes in the 64bit mode, since the
2451 lowering is more complicated. */
2452 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2453 i.op[op].regs->reg_name,
2454 i.suffix);
2455 return 0;
2456 }
2457 return 1;
2458 }
2459
2460 static int
2461 check_word_reg ()
2462 {
2463 int op;
2464 for (op = i.operands; --op >= 0;)
2465 /* Reject eight bit registers, except where the template requires
2466 them. (eg. movzb) */
2467 if ((i.types[op] & Reg8) != 0
2468 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2469 {
2470 as_bad (_("`%%%s' not allowed with `%s%c'"),
2471 i.op[op].regs->reg_name,
2472 i.tm.name,
2473 i.suffix);
2474 return 0;
2475 }
2476 /* Warn if the e prefix on a general reg is present. */
2477 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2478 && (i.types[op] & Reg32) != 0
2479 && (i.tm.operand_types[op] & (Reg16 | Acc)) != 0)
2480 {
2481 /* Prohibit these changes in the 64bit mode, since the
2482 lowering is more complicated. */
2483 if (flag_code == CODE_64BIT)
2484 {
2485 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2486 i.op[op].regs->reg_name,
2487 i.suffix);
2488 return 0;
2489 }
2490 else
2491 #if REGISTER_WARNINGS
2492 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2493 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
2494 i.op[op].regs->reg_name,
2495 i.suffix);
2496 #endif
2497 }
2498 return 1;
2499 }
2500
2501 static int
2502 finalize_imm ()
2503 {
2504 unsigned int overlap0, overlap1, overlap2;
2505
2506 overlap0 = i.types[0] & i.tm.operand_types[0];
2507 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S))
2508 && overlap0 != Imm8 && overlap0 != Imm8S
2509 && overlap0 != Imm16 && overlap0 != Imm32S
2510 && overlap0 != Imm32 && overlap0 != Imm64)
2511 {
2512 if (i.suffix)
2513 {
2514 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX
2515 ? Imm8 | Imm8S
2516 : (i.suffix == WORD_MNEM_SUFFIX
2517 ? Imm16
2518 : (i.suffix == QWORD_MNEM_SUFFIX
2519 ? Imm64 | Imm32S
2520 : Imm32)));
2521 }
2522 else if (overlap0 == (Imm16 | Imm32S | Imm32)
2523 || overlap0 == (Imm16 | Imm32)
2524 || overlap0 == (Imm16 | Imm32S))
2525 {
2526 overlap0 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
2527 ? Imm16 : Imm32S);
2528 }
2529 if (overlap0 != Imm8 && overlap0 != Imm8S
2530 && overlap0 != Imm16 && overlap0 != Imm32S
2531 && overlap0 != Imm32 && overlap0 != Imm64)
2532 {
2533 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
2534 return 0;
2535 }
2536 }
2537 i.types[0] = overlap0;
2538
2539 overlap1 = i.types[1] & i.tm.operand_types[1];
2540 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32))
2541 && overlap1 != Imm8 && overlap1 != Imm8S
2542 && overlap1 != Imm16 && overlap1 != Imm32S
2543 && overlap1 != Imm32 && overlap1 != Imm64)
2544 {
2545 if (i.suffix)
2546 {
2547 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX
2548 ? Imm8 | Imm8S
2549 : (i.suffix == WORD_MNEM_SUFFIX
2550 ? Imm16
2551 : (i.suffix == QWORD_MNEM_SUFFIX
2552 ? Imm64 | Imm32S
2553 : Imm32)));
2554 }
2555 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
2556 || overlap1 == (Imm16 | Imm32)
2557 || overlap1 == (Imm16 | Imm32S))
2558 {
2559 overlap1 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
2560 ? Imm16 : Imm32S);
2561 }
2562 if (overlap1 != Imm8 && overlap1 != Imm8S
2563 && overlap1 != Imm16 && overlap1 != Imm32S
2564 && overlap1 != Imm32 && overlap1 != Imm64)
2565 {
2566 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
2567 return 0;
2568 }
2569 }
2570 i.types[1] = overlap1;
2571
2572 overlap2 = i.types[2] & i.tm.operand_types[2];
2573 assert ((overlap2 & Imm) == 0);
2574 i.types[2] = overlap2;
2575
2576 return 1;
2577 }
2578
2579 static int
2580 process_operands ()
2581 {
2582 /* Default segment register this instruction will use for memory
2583 accesses. 0 means unknown. This is only for optimizing out
2584 unnecessary segment overrides. */
2585 const seg_entry *default_seg = 0;
2586
2587 /* The imul $imm, %reg instruction is converted into
2588 imul $imm, %reg, %reg, and the clr %reg instruction
2589 is converted into xor %reg, %reg. */
2590 if (i.tm.opcode_modifier & regKludge)
2591 {
2592 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
2593 /* Pretend we saw the extra register operand. */
2594 assert (i.op[first_reg_op + 1].regs == 0);
2595 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
2596 i.types[first_reg_op + 1] = i.types[first_reg_op];
2597 i.reg_operands = 2;
2598 }
2599
2600 if (i.tm.opcode_modifier & ShortForm)
2601 {
2602 /* The register or float register operand is in operand 0 or 1. */
2603 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
2604 /* Register goes in low 3 bits of opcode. */
2605 i.tm.base_opcode |= i.op[op].regs->reg_num;
2606 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2607 i.rex |= REX_EXTZ;
2608 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2609 {
2610 /* Warn about some common errors, but press on regardless.
2611 The first case can be generated by gcc (<= 2.8.1). */
2612 if (i.operands == 2)
2613 {
2614 /* Reversed arguments on faddp, fsubp, etc. */
2615 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
2616 i.op[1].regs->reg_name,
2617 i.op[0].regs->reg_name);
2618 }
2619 else
2620 {
2621 /* Extraneous `l' suffix on fp insn. */
2622 as_warn (_("translating to `%s %%%s'"), i.tm.name,
2623 i.op[0].regs->reg_name);
2624 }
2625 }
2626 }
2627 else if (i.tm.opcode_modifier & Modrm)
2628 {
2629 /* The opcode is completed (modulo i.tm.extension_opcode which
2630 must be put into the modrm byte). Now, we make the modrm and
2631 index base bytes based on all the info we've collected. */
2632
2633 default_seg = build_modrm_byte ();
2634 }
2635 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
2636 {
2637 if (i.tm.base_opcode == POP_SEG_SHORT
2638 && i.op[0].regs->reg_num == 1)
2639 {
2640 as_bad (_("you can't `pop %%cs'"));
2641 return 0;
2642 }
2643 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
2644 if ((i.op[0].regs->reg_flags & RegRex) != 0)
2645 i.rex |= REX_EXTZ;
2646 }
2647 else if ((i.tm.base_opcode & ~(D | W)) == MOV_AX_DISP32)
2648 {
2649 default_seg = &ds;
2650 }
2651 else if ((i.tm.opcode_modifier & IsString) != 0)
2652 {
2653 /* For the string instructions that allow a segment override
2654 on one of their operands, the default segment is ds. */
2655 default_seg = &ds;
2656 }
2657
2658 if (i.tm.base_opcode == 0x8d /* lea */ && i.seg[0] && !quiet_warnings)
2659 as_warn (_("segment override on `lea' is ineffectual"));
2660
2661 /* If a segment was explicitly specified, and the specified segment
2662 is not the default, use an opcode prefix to select it. If we
2663 never figured out what the default segment is, then default_seg
2664 will be zero at this point, and the specified segment prefix will
2665 always be used. */
2666 if ((i.seg[0]) && (i.seg[0] != default_seg))
2667 {
2668 if (!add_prefix (i.seg[0]->seg_prefix))
2669 return 0;
2670 }
2671 return 1;
2672 }
2673
2674 static const seg_entry *
2675 build_modrm_byte ()
2676 {
2677 const seg_entry *default_seg = 0;
2678
2679 /* i.reg_operands MUST be the number of real register operands;
2680 implicit registers do not count. */
2681 if (i.reg_operands == 2)
2682 {
2683 unsigned int source, dest;
2684 source = ((i.types[0]
2685 & (Reg | RegMMX | RegXMM
2686 | SReg2 | SReg3
2687 | Control | Debug | Test))
2688 ? 0 : 1);
2689 dest = source + 1;
2690
2691 i.rm.mode = 3;
2692 /* One of the register operands will be encoded in the i.tm.reg
2693 field, the other in the combined i.tm.mode and i.tm.regmem
2694 fields. If no form of this instruction supports a memory
2695 destination operand, then we assume the source operand may
2696 sometimes be a memory operand and so we need to store the
2697 destination in the i.rm.reg field. */
2698 if ((i.tm.operand_types[dest] & AnyMem) == 0)
2699 {
2700 i.rm.reg = i.op[dest].regs->reg_num;
2701 i.rm.regmem = i.op[source].regs->reg_num;
2702 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
2703 i.rex |= REX_EXTX;
2704 if ((i.op[source].regs->reg_flags & RegRex) != 0)
2705 i.rex |= REX_EXTZ;
2706 }
2707 else
2708 {
2709 i.rm.reg = i.op[source].regs->reg_num;
2710 i.rm.regmem = i.op[dest].regs->reg_num;
2711 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
2712 i.rex |= REX_EXTZ;
2713 if ((i.op[source].regs->reg_flags & RegRex) != 0)
2714 i.rex |= REX_EXTX;
2715 }
2716 }
2717 else
2718 { /* If it's not 2 reg operands... */
2719 if (i.mem_operands)
2720 {
2721 unsigned int fake_zero_displacement = 0;
2722 unsigned int op = ((i.types[0] & AnyMem)
2723 ? 0
2724 : (i.types[1] & AnyMem) ? 1 : 2);
2725
2726 default_seg = &ds;
2727
2728 if (i.base_reg == 0)
2729 {
2730 i.rm.mode = 0;
2731 if (!i.disp_operands)
2732 fake_zero_displacement = 1;
2733 if (i.index_reg == 0)
2734 {
2735 /* Operand is just <disp> */
2736 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)
2737 && (flag_code != CODE_64BIT))
2738 {
2739 i.rm.regmem = NO_BASE_REGISTER_16;
2740 i.types[op] &= ~Disp;
2741 i.types[op] |= Disp16;
2742 }
2743 else if (flag_code != CODE_64BIT
2744 || (i.prefix[ADDR_PREFIX] != 0))
2745 {
2746 i.rm.regmem = NO_BASE_REGISTER;
2747 i.types[op] &= ~Disp;
2748 i.types[op] |= Disp32;
2749 }
2750 else
2751 {
2752 /* 64bit mode overwrites the 32bit absolute
2753 addressing by RIP relative addressing and
2754 absolute addressing is encoded by one of the
2755 redundant SIB forms. */
2756 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2757 i.sib.base = NO_BASE_REGISTER;
2758 i.sib.index = NO_INDEX_REGISTER;
2759 i.types[op] &= ~Disp;
2760 i.types[op] |= Disp32S;
2761 }
2762 }
2763 else /* !i.base_reg && i.index_reg */
2764 {
2765 i.sib.index = i.index_reg->reg_num;
2766 i.sib.base = NO_BASE_REGISTER;
2767 i.sib.scale = i.log2_scale_factor;
2768 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2769 i.types[op] &= ~Disp;
2770 if (flag_code != CODE_64BIT)
2771 i.types[op] |= Disp32; /* Must be 32 bit */
2772 else
2773 i.types[op] |= Disp32S;
2774 if ((i.index_reg->reg_flags & RegRex) != 0)
2775 i.rex |= REX_EXTY;
2776 }
2777 }
2778 /* RIP addressing for 64bit mode. */
2779 else if (i.base_reg->reg_type == BaseIndex)
2780 {
2781 i.rm.regmem = NO_BASE_REGISTER;
2782 i.types[op] &= ~Disp;
2783 i.types[op] |= Disp32S;
2784 i.flags[op] = Operand_PCrel;
2785 }
2786 else if (i.base_reg->reg_type & Reg16)
2787 {
2788 switch (i.base_reg->reg_num)
2789 {
2790 case 3: /* (%bx) */
2791 if (i.index_reg == 0)
2792 i.rm.regmem = 7;
2793 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
2794 i.rm.regmem = i.index_reg->reg_num - 6;
2795 break;
2796 case 5: /* (%bp) */
2797 default_seg = &ss;
2798 if (i.index_reg == 0)
2799 {
2800 i.rm.regmem = 6;
2801 if ((i.types[op] & Disp) == 0)
2802 {
2803 /* fake (%bp) into 0(%bp) */
2804 i.types[op] |= Disp8;
2805 fake_zero_displacement = 1;
2806 }
2807 }
2808 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
2809 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
2810 break;
2811 default: /* (%si) -> 4 or (%di) -> 5 */
2812 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
2813 }
2814 i.rm.mode = mode_from_disp_size (i.types[op]);
2815 }
2816 else /* i.base_reg and 32/64 bit mode */
2817 {
2818 if (flag_code == CODE_64BIT
2819 && (i.types[op] & Disp))
2820 {
2821 if (i.types[op] & Disp8)
2822 i.types[op] = Disp8 | Disp32S;
2823 else
2824 i.types[op] = Disp32S;
2825 }
2826 i.rm.regmem = i.base_reg->reg_num;
2827 if ((i.base_reg->reg_flags & RegRex) != 0)
2828 i.rex |= REX_EXTZ;
2829 i.sib.base = i.base_reg->reg_num;
2830 /* x86-64 ignores REX prefix bit here to avoid decoder
2831 complications. */
2832 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
2833 {
2834 default_seg = &ss;
2835 if (i.disp_operands == 0)
2836 {
2837 fake_zero_displacement = 1;
2838 i.types[op] |= Disp8;
2839 }
2840 }
2841 else if (i.base_reg->reg_num == ESP_REG_NUM)
2842 {
2843 default_seg = &ss;
2844 }
2845 i.sib.scale = i.log2_scale_factor;
2846 if (i.index_reg == 0)
2847 {
2848 /* <disp>(%esp) becomes two byte modrm with no index
2849 register. We've already stored the code for esp
2850 in i.rm.regmem ie. ESCAPE_TO_TWO_BYTE_ADDRESSING.
2851 Any base register besides %esp will not use the
2852 extra modrm byte. */
2853 i.sib.index = NO_INDEX_REGISTER;
2854 #if !SCALE1_WHEN_NO_INDEX
2855 /* Another case where we force the second modrm byte. */
2856 if (i.log2_scale_factor)
2857 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2858 #endif
2859 }
2860 else
2861 {
2862 i.sib.index = i.index_reg->reg_num;
2863 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2864 if ((i.index_reg->reg_flags & RegRex) != 0)
2865 i.rex |= REX_EXTY;
2866 }
2867 i.rm.mode = mode_from_disp_size (i.types[op]);
2868 }
2869
2870 if (fake_zero_displacement)
2871 {
2872 /* Fakes a zero displacement assuming that i.types[op]
2873 holds the correct displacement size. */
2874 expressionS *exp;
2875
2876 assert (i.op[op].disps == 0);
2877 exp = &disp_expressions[i.disp_operands++];
2878 i.op[op].disps = exp;
2879 exp->X_op = O_constant;
2880 exp->X_add_number = 0;
2881 exp->X_add_symbol = (symbolS *) 0;
2882 exp->X_op_symbol = (symbolS *) 0;
2883 }
2884 }
2885
2886 /* Fill in i.rm.reg or i.rm.regmem field with register operand
2887 (if any) based on i.tm.extension_opcode. Again, we must be
2888 careful to make sure that segment/control/debug/test/MMX
2889 registers are coded into the i.rm.reg field. */
2890 if (i.reg_operands)
2891 {
2892 unsigned int op =
2893 ((i.types[0]
2894 & (Reg | RegMMX | RegXMM
2895 | SReg2 | SReg3
2896 | Control | Debug | Test))
2897 ? 0
2898 : ((i.types[1]
2899 & (Reg | RegMMX | RegXMM
2900 | SReg2 | SReg3
2901 | Control | Debug | Test))
2902 ? 1
2903 : 2));
2904 /* If there is an extension opcode to put here, the register
2905 number must be put into the regmem field. */
2906 if (i.tm.extension_opcode != None)
2907 {
2908 i.rm.regmem = i.op[op].regs->reg_num;
2909 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2910 i.rex |= REX_EXTZ;
2911 }
2912 else
2913 {
2914 i.rm.reg = i.op[op].regs->reg_num;
2915 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2916 i.rex |= REX_EXTX;
2917 }
2918
2919 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2 we
2920 must set it to 3 to indicate this is a register operand
2921 in the regmem field. */
2922 if (!i.mem_operands)
2923 i.rm.mode = 3;
2924 }
2925
2926 /* Fill in i.rm.reg field with extension opcode (if any). */
2927 if (i.tm.extension_opcode != None)
2928 i.rm.reg = i.tm.extension_opcode;
2929 }
2930 return default_seg;
2931 }
2932
2933 static void
2934 output_branch ()
2935 {
2936 char *p;
2937 int code16;
2938 int prefix;
2939 relax_substateT subtype;
2940 symbolS *sym;
2941 offsetT off;
2942
2943 code16 = 0;
2944 if (flag_code == CODE_16BIT)
2945 code16 = CODE16;
2946
2947 prefix = 0;
2948 if (i.prefix[DATA_PREFIX] != 0)
2949 {
2950 prefix = 1;
2951 i.prefixes -= 1;
2952 code16 ^= CODE16;
2953 }
2954 /* Pentium4 branch hints. */
2955 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
2956 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
2957 {
2958 prefix++;
2959 i.prefixes--;
2960 }
2961 if (i.prefix[REX_PREFIX] != 0)
2962 {
2963 prefix++;
2964 i.prefixes--;
2965 }
2966
2967 if (i.prefixes != 0 && !intel_syntax)
2968 as_warn (_("skipping prefixes on this instruction"));
2969
2970 /* It's always a symbol; End frag & setup for relax.
2971 Make sure there is enough room in this frag for the largest
2972 instruction we may generate in md_convert_frag. This is 2
2973 bytes for the opcode and room for the prefix and largest
2974 displacement. */
2975 frag_grow (prefix + 2 + 4);
2976 /* Prefix and 1 opcode byte go in fr_fix. */
2977 p = frag_more (prefix + 1);
2978 if (i.prefix[DATA_PREFIX] != 0)
2979 *p++ = DATA_PREFIX_OPCODE;
2980 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
2981 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
2982 *p++ = i.prefix[SEG_PREFIX];
2983 if (i.prefix[REX_PREFIX] != 0)
2984 *p++ = i.prefix[REX_PREFIX];
2985 *p = i.tm.base_opcode;
2986
2987 if ((unsigned char) *p == JUMP_PC_RELATIVE)
2988 subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
2989 else if ((cpu_arch_flags & Cpu386) != 0)
2990 subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
2991 else
2992 subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
2993 subtype |= code16;
2994
2995 sym = i.op[0].disps->X_add_symbol;
2996 off = i.op[0].disps->X_add_number;
2997
2998 if (i.op[0].disps->X_op != O_constant
2999 && i.op[0].disps->X_op != O_symbol)
3000 {
3001 /* Handle complex expressions. */
3002 sym = make_expr_symbol (i.op[0].disps);
3003 off = 0;
3004 }
3005
3006 /* 1 possible extra opcode + 4 byte displacement go in var part.
3007 Pass reloc in fr_var. */
3008 frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
3009 }
3010
3011 static void
3012 output_jump ()
3013 {
3014 char *p;
3015 int size;
3016 fixS *fixP;
3017
3018 if (i.tm.opcode_modifier & JumpByte)
3019 {
3020 /* This is a loop or jecxz type instruction. */
3021 size = 1;
3022 if (i.prefix[ADDR_PREFIX] != 0)
3023 {
3024 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
3025 i.prefixes -= 1;
3026 }
3027 /* Pentium4 branch hints. */
3028 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3029 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3030 {
3031 FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
3032 i.prefixes--;
3033 }
3034 }
3035 else
3036 {
3037 int code16;
3038
3039 code16 = 0;
3040 if (flag_code == CODE_16BIT)
3041 code16 = CODE16;
3042
3043 if (i.prefix[DATA_PREFIX] != 0)
3044 {
3045 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
3046 i.prefixes -= 1;
3047 code16 ^= CODE16;
3048 }
3049
3050 size = 4;
3051 if (code16)
3052 size = 2;
3053 }
3054
3055 if (i.prefix[REX_PREFIX] != 0)
3056 {
3057 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
3058 i.prefixes -= 1;
3059 }
3060
3061 if (i.prefixes != 0 && !intel_syntax)
3062 as_warn (_("skipping prefixes on this instruction"));
3063
3064 p = frag_more (1 + size);
3065 *p++ = i.tm.base_opcode;
3066
3067 fixP = fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3068 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
3069
3070 /* All jumps handled here are signed, but don't use a signed limit
3071 check for 32 and 16 bit jumps as we want to allow wrap around at
3072 4G and 64k respectively. */
3073 if (size == 1)
3074 fixP->fx_signed = 1;
3075 }
3076
3077 static void
3078 output_interseg_jump ()
3079 {
3080 char *p;
3081 int size;
3082 int prefix;
3083 int code16;
3084
3085 code16 = 0;
3086 if (flag_code == CODE_16BIT)
3087 code16 = CODE16;
3088
3089 prefix = 0;
3090 if (i.prefix[DATA_PREFIX] != 0)
3091 {
3092 prefix = 1;
3093 i.prefixes -= 1;
3094 code16 ^= CODE16;
3095 }
3096 if (i.prefix[REX_PREFIX] != 0)
3097 {
3098 prefix++;
3099 i.prefixes -= 1;
3100 }
3101
3102 size = 4;
3103 if (code16)
3104 size = 2;
3105
3106 if (i.prefixes != 0 && !intel_syntax)
3107 as_warn (_("skipping prefixes on this instruction"));
3108
3109 /* 1 opcode; 2 segment; offset */
3110 p = frag_more (prefix + 1 + 2 + size);
3111
3112 if (i.prefix[DATA_PREFIX] != 0)
3113 *p++ = DATA_PREFIX_OPCODE;
3114
3115 if (i.prefix[REX_PREFIX] != 0)
3116 *p++ = i.prefix[REX_PREFIX];
3117
3118 *p++ = i.tm.base_opcode;
3119 if (i.op[1].imms->X_op == O_constant)
3120 {
3121 offsetT n = i.op[1].imms->X_add_number;
3122
3123 if (size == 2
3124 && !fits_in_unsigned_word (n)
3125 && !fits_in_signed_word (n))
3126 {
3127 as_bad (_("16-bit jump out of range"));
3128 return;
3129 }
3130 md_number_to_chars (p, n, size);
3131 }
3132 else
3133 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3134 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
3135 if (i.op[0].imms->X_op != O_constant)
3136 as_bad (_("can't handle non absolute segment in `%s'"),
3137 i.tm.name);
3138 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
3139 }
3140
3141 static void
3142 output_insn ()
3143 {
3144 fragS *insn_start_frag;
3145 offsetT insn_start_off;
3146
3147 /* Tie dwarf2 debug info to the address at the start of the insn.
3148 We can't do this after the insn has been output as the current
3149 frag may have been closed off. eg. by frag_var. */
3150 dwarf2_emit_insn (0);
3151
3152 insn_start_frag = frag_now;
3153 insn_start_off = frag_now_fix ();
3154
3155 /* Output jumps. */
3156 if (i.tm.opcode_modifier & Jump)
3157 output_branch ();
3158 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
3159 output_jump ();
3160 else if (i.tm.opcode_modifier & JumpInterSegment)
3161 output_interseg_jump ();
3162 else
3163 {
3164 /* Output normal instructions here. */
3165 char *p;
3166 unsigned char *q;
3167
3168 /* All opcodes on i386 have either 1 or 2 bytes, PadLock instructions
3169 have 3 bytes. We may use one more higher byte to specify a prefix
3170 the instruction requires. */
3171 if ((i.tm.cpu_flags & CpuPadLock) != 0
3172 && (i.tm.base_opcode & 0xff000000) != 0)
3173 {
3174 unsigned int prefix;
3175 prefix = (i.tm.base_opcode >> 24) & 0xff;
3176
3177 if (prefix != REPE_PREFIX_OPCODE
3178 || i.prefix[LOCKREP_PREFIX] != REPE_PREFIX_OPCODE)
3179 add_prefix (prefix);
3180 }
3181 else
3182 if ((i.tm.cpu_flags & CpuPadLock) == 0
3183 && (i.tm.base_opcode & 0xff0000) != 0)
3184 add_prefix ((i.tm.base_opcode >> 16) & 0xff);
3185
3186 /* The prefix bytes. */
3187 for (q = i.prefix;
3188 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
3189 q++)
3190 {
3191 if (*q)
3192 {
3193 p = frag_more (1);
3194 md_number_to_chars (p, (valueT) *q, 1);
3195 }
3196 }
3197
3198 /* Now the opcode; be careful about word order here! */
3199 if (fits_in_unsigned_byte (i.tm.base_opcode))
3200 {
3201 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
3202 }
3203 else
3204 {
3205 if ((i.tm.cpu_flags & CpuPadLock) != 0)
3206 {
3207 p = frag_more (3);
3208 *p++ = (i.tm.base_opcode >> 16) & 0xff;
3209 }
3210 else
3211 p = frag_more (2);
3212
3213 /* Put out high byte first: can't use md_number_to_chars! */
3214 *p++ = (i.tm.base_opcode >> 8) & 0xff;
3215 *p = i.tm.base_opcode & 0xff;
3216 }
3217
3218 /* Now the modrm byte and sib byte (if present). */
3219 if (i.tm.opcode_modifier & Modrm)
3220 {
3221 p = frag_more (1);
3222 md_number_to_chars (p,
3223 (valueT) (i.rm.regmem << 0
3224 | i.rm.reg << 3
3225 | i.rm.mode << 6),
3226 1);
3227 /* If i.rm.regmem == ESP (4)
3228 && i.rm.mode != (Register mode)
3229 && not 16 bit
3230 ==> need second modrm byte. */
3231 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
3232 && i.rm.mode != 3
3233 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
3234 {
3235 p = frag_more (1);
3236 md_number_to_chars (p,
3237 (valueT) (i.sib.base << 0
3238 | i.sib.index << 3
3239 | i.sib.scale << 6),
3240 1);
3241 }
3242 }
3243
3244 if (i.disp_operands)
3245 output_disp (insn_start_frag, insn_start_off);
3246
3247 if (i.imm_operands)
3248 output_imm (insn_start_frag, insn_start_off);
3249 }
3250
3251 #ifdef DEBUG386
3252 if (flag_debug)
3253 {
3254 pi (line, &i);
3255 }
3256 #endif /* DEBUG386 */
3257 }
3258
3259 static void
3260 output_disp (insn_start_frag, insn_start_off)
3261 fragS *insn_start_frag;
3262 offsetT insn_start_off;
3263 {
3264 char *p;
3265 unsigned int n;
3266
3267 for (n = 0; n < i.operands; n++)
3268 {
3269 if (i.types[n] & Disp)
3270 {
3271 if (i.op[n].disps->X_op == O_constant)
3272 {
3273 int size;
3274 offsetT val;
3275
3276 size = 4;
3277 if (i.types[n] & (Disp8 | Disp16 | Disp64))
3278 {
3279 size = 2;
3280 if (i.types[n] & Disp8)
3281 size = 1;
3282 if (i.types[n] & Disp64)
3283 size = 8;
3284 }
3285 val = offset_in_range (i.op[n].disps->X_add_number,
3286 size);
3287 p = frag_more (size);
3288 md_number_to_chars (p, val, size);
3289 }
3290 else
3291 {
3292 enum bfd_reloc_code_real reloc_type;
3293 int size = 4;
3294 int sign = 0;
3295 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
3296
3297 /* The PC relative address is computed relative
3298 to the instruction boundary, so in case immediate
3299 fields follows, we need to adjust the value. */
3300 if (pcrel && i.imm_operands)
3301 {
3302 int imm_size = 4;
3303 unsigned int n1;
3304
3305 for (n1 = 0; n1 < i.operands; n1++)
3306 if (i.types[n1] & Imm)
3307 {
3308 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
3309 {
3310 imm_size = 2;
3311 if (i.types[n1] & (Imm8 | Imm8S))
3312 imm_size = 1;
3313 if (i.types[n1] & Imm64)
3314 imm_size = 8;
3315 }
3316 break;
3317 }
3318 /* We should find the immediate. */
3319 if (n1 == i.operands)
3320 abort ();
3321 i.op[n].disps->X_add_number -= imm_size;
3322 }
3323
3324 if (i.types[n] & Disp32S)
3325 sign = 1;
3326
3327 if (i.types[n] & (Disp16 | Disp64))
3328 {
3329 size = 2;
3330 if (i.types[n] & Disp64)
3331 size = 8;
3332 }
3333
3334 p = frag_more (size);
3335 reloc_type = reloc (size, pcrel, sign, i.reloc[n]);
3336 if (reloc_type == BFD_RELOC_32
3337 && GOT_symbol
3338 && GOT_symbol == i.op[n].disps->X_add_symbol
3339 && (i.op[n].disps->X_op == O_symbol
3340 || (i.op[n].disps->X_op == O_add
3341 && ((symbol_get_value_expression
3342 (i.op[n].disps->X_op_symbol)->X_op)
3343 == O_subtract))))
3344 {
3345 offsetT add;
3346
3347 if (insn_start_frag == frag_now)
3348 add = (p - frag_now->fr_literal) - insn_start_off;
3349 else
3350 {
3351 fragS *fr;
3352
3353 add = insn_start_frag->fr_fix - insn_start_off;
3354 for (fr = insn_start_frag->fr_next;
3355 fr && fr != frag_now; fr = fr->fr_next)
3356 add += fr->fr_fix;
3357 add += p - frag_now->fr_literal;
3358 }
3359
3360 /* We don't support dynamic linking on x86-64 yet. */
3361 if (flag_code == CODE_64BIT)
3362 abort ();
3363 reloc_type = BFD_RELOC_386_GOTPC;
3364 i.op[n].disps->X_add_number += add;
3365 }
3366 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3367 i.op[n].disps, pcrel, reloc_type);
3368 }
3369 }
3370 }
3371 }
3372
3373 static void
3374 output_imm (insn_start_frag, insn_start_off)
3375 fragS *insn_start_frag;
3376 offsetT insn_start_off;
3377 {
3378 char *p;
3379 unsigned int n;
3380
3381 for (n = 0; n < i.operands; n++)
3382 {
3383 if (i.types[n] & Imm)
3384 {
3385 if (i.op[n].imms->X_op == O_constant)
3386 {
3387 int size;
3388 offsetT val;
3389
3390 size = 4;
3391 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3392 {
3393 size = 2;
3394 if (i.types[n] & (Imm8 | Imm8S))
3395 size = 1;
3396 else if (i.types[n] & Imm64)
3397 size = 8;
3398 }
3399 val = offset_in_range (i.op[n].imms->X_add_number,
3400 size);
3401 p = frag_more (size);
3402 md_number_to_chars (p, val, size);
3403 }
3404 else
3405 {
3406 /* Not absolute_section.
3407 Need a 32-bit fixup (don't support 8bit
3408 non-absolute imms). Try to support other
3409 sizes ... */
3410 enum bfd_reloc_code_real reloc_type;
3411 int size = 4;
3412 int sign = 0;
3413
3414 if ((i.types[n] & (Imm32S))
3415 && i.suffix == QWORD_MNEM_SUFFIX)
3416 sign = 1;
3417 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3418 {
3419 size = 2;
3420 if (i.types[n] & (Imm8 | Imm8S))
3421 size = 1;
3422 if (i.types[n] & Imm64)
3423 size = 8;
3424 }
3425
3426 p = frag_more (size);
3427 reloc_type = reloc (size, 0, sign, i.reloc[n]);
3428
3429 /* This is tough to explain. We end up with this one if we
3430 * have operands that look like
3431 * "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal here is to
3432 * obtain the absolute address of the GOT, and it is strongly
3433 * preferable from a performance point of view to avoid using
3434 * a runtime relocation for this. The actual sequence of
3435 * instructions often look something like:
3436 *
3437 * call .L66
3438 * .L66:
3439 * popl %ebx
3440 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
3441 *
3442 * The call and pop essentially return the absolute address
3443 * of the label .L66 and store it in %ebx. The linker itself
3444 * will ultimately change the first operand of the addl so
3445 * that %ebx points to the GOT, but to keep things simple, the
3446 * .o file must have this operand set so that it generates not
3447 * the absolute address of .L66, but the absolute address of
3448 * itself. This allows the linker itself simply treat a GOTPC
3449 * relocation as asking for a pcrel offset to the GOT to be
3450 * added in, and the addend of the relocation is stored in the
3451 * operand field for the instruction itself.
3452 *
3453 * Our job here is to fix the operand so that it would add
3454 * the correct offset so that %ebx would point to itself. The
3455 * thing that is tricky is that .-.L66 will point to the
3456 * beginning of the instruction, so we need to further modify
3457 * the operand so that it will point to itself. There are
3458 * other cases where you have something like:
3459 *
3460 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
3461 *
3462 * and here no correction would be required. Internally in
3463 * the assembler we treat operands of this form as not being
3464 * pcrel since the '.' is explicitly mentioned, and I wonder
3465 * whether it would simplify matters to do it this way. Who
3466 * knows. In earlier versions of the PIC patches, the
3467 * pcrel_adjust field was used to store the correction, but
3468 * since the expression is not pcrel, I felt it would be
3469 * confusing to do it this way. */
3470
3471 if (reloc_type == BFD_RELOC_32
3472 && GOT_symbol
3473 && GOT_symbol == i.op[n].imms->X_add_symbol
3474 && (i.op[n].imms->X_op == O_symbol
3475 || (i.op[n].imms->X_op == O_add
3476 && ((symbol_get_value_expression
3477 (i.op[n].imms->X_op_symbol)->X_op)
3478 == O_subtract))))
3479 {
3480 offsetT add;
3481
3482 if (insn_start_frag == frag_now)
3483 add = (p - frag_now->fr_literal) - insn_start_off;
3484 else
3485 {
3486 fragS *fr;
3487
3488 add = insn_start_frag->fr_fix - insn_start_off;
3489 for (fr = insn_start_frag->fr_next;
3490 fr && fr != frag_now; fr = fr->fr_next)
3491 add += fr->fr_fix;
3492 add += p - frag_now->fr_literal;
3493 }
3494
3495 /* We don't support dynamic linking on x86-64 yet. */
3496 if (flag_code == CODE_64BIT)
3497 abort ();
3498 reloc_type = BFD_RELOC_386_GOTPC;
3499 i.op[n].imms->X_add_number += add;
3500 }
3501 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3502 i.op[n].imms, 0, reloc_type);
3503 }
3504 }
3505 }
3506 }
3507 \f
3508 #ifndef LEX_AT
3509 static char *lex_got PARAMS ((enum bfd_reloc_code_real *, int *));
3510
3511 /* Parse operands of the form
3512 <symbol>@GOTOFF+<nnn>
3513 and similar .plt or .got references.
3514
3515 If we find one, set up the correct relocation in RELOC and copy the
3516 input string, minus the `@GOTOFF' into a malloc'd buffer for
3517 parsing by the calling routine. Return this buffer, and if ADJUST
3518 is non-null set it to the length of the string we removed from the
3519 input line. Otherwise return NULL. */
3520 static char *
3521 lex_got (reloc, adjust)
3522 enum bfd_reloc_code_real *reloc;
3523 int *adjust;
3524 {
3525 static const char * const mode_name[NUM_FLAG_CODE] = { "32", "16", "64" };
3526 static const struct {
3527 const char *str;
3528 const enum bfd_reloc_code_real rel[NUM_FLAG_CODE];
3529 } gotrel[] = {
3530 { "PLT", { BFD_RELOC_386_PLT32, 0, BFD_RELOC_X86_64_PLT32 } },
3531 { "GOTOFF", { BFD_RELOC_386_GOTOFF, 0, 0 } },
3532 { "GOTPCREL", { 0, 0, BFD_RELOC_X86_64_GOTPCREL } },
3533 { "TLSGD", { BFD_RELOC_386_TLS_GD, 0, BFD_RELOC_X86_64_TLSGD } },
3534 { "TLSLDM", { BFD_RELOC_386_TLS_LDM, 0, 0 } },
3535 { "TLSLD", { 0, 0, BFD_RELOC_X86_64_TLSLD } },
3536 { "GOTTPOFF", { BFD_RELOC_386_TLS_IE_32, 0, BFD_RELOC_X86_64_GOTTPOFF } },
3537 { "TPOFF", { BFD_RELOC_386_TLS_LE_32, 0, BFD_RELOC_X86_64_TPOFF32 } },
3538 { "NTPOFF", { BFD_RELOC_386_TLS_LE, 0, 0 } },
3539 { "DTPOFF", { BFD_RELOC_386_TLS_LDO_32, 0, BFD_RELOC_X86_64_DTPOFF32 } },
3540 { "GOTNTPOFF",{ BFD_RELOC_386_TLS_GOTIE, 0, 0 } },
3541 { "INDNTPOFF",{ BFD_RELOC_386_TLS_IE, 0, 0 } },
3542 { "GOT", { BFD_RELOC_386_GOT32, 0, BFD_RELOC_X86_64_GOT32 } }
3543 };
3544 char *cp;
3545 unsigned int j;
3546
3547 for (cp = input_line_pointer; *cp != '@'; cp++)
3548 if (is_end_of_line[(unsigned char) *cp])
3549 return NULL;
3550
3551 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
3552 {
3553 int len;
3554
3555 len = strlen (gotrel[j].str);
3556 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
3557 {
3558 if (gotrel[j].rel[(unsigned int) flag_code] != 0)
3559 {
3560 int first, second;
3561 char *tmpbuf, *past_reloc;
3562
3563 *reloc = gotrel[j].rel[(unsigned int) flag_code];
3564 if (adjust)
3565 *adjust = len;
3566
3567 if (GOT_symbol == NULL)
3568 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3569
3570 /* Replace the relocation token with ' ', so that
3571 errors like foo@GOTOFF1 will be detected. */
3572
3573 /* The length of the first part of our input line. */
3574 first = cp - input_line_pointer;
3575
3576 /* The second part goes from after the reloc token until
3577 (and including) an end_of_line char. Don't use strlen
3578 here as the end_of_line char may not be a NUL. */
3579 past_reloc = cp + 1 + len;
3580 for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
3581 ;
3582 second = cp - past_reloc;
3583
3584 /* Allocate and copy string. The trailing NUL shouldn't
3585 be necessary, but be safe. */
3586 tmpbuf = xmalloc (first + second + 2);
3587 memcpy (tmpbuf, input_line_pointer, first);
3588 tmpbuf[first] = ' ';
3589 memcpy (tmpbuf + first + 1, past_reloc, second);
3590 tmpbuf[first + second + 1] = '\0';
3591 return tmpbuf;
3592 }
3593
3594 as_bad (_("@%s reloc is not supported in %s bit mode"),
3595 gotrel[j].str, mode_name[(unsigned int) flag_code]);
3596 return NULL;
3597 }
3598 }
3599
3600 /* Might be a symbol version string. Don't as_bad here. */
3601 return NULL;
3602 }
3603
3604 /* x86_cons_fix_new is called via the expression parsing code when a
3605 reloc is needed. We use this hook to get the correct .got reloc. */
3606 static enum bfd_reloc_code_real got_reloc = NO_RELOC;
3607
3608 void
3609 x86_cons_fix_new (frag, off, len, exp)
3610 fragS *frag;
3611 unsigned int off;
3612 unsigned int len;
3613 expressionS *exp;
3614 {
3615 enum bfd_reloc_code_real r = reloc (len, 0, 0, got_reloc);
3616 got_reloc = NO_RELOC;
3617 fix_new_exp (frag, off, len, exp, 0, r);
3618 }
3619
3620 void
3621 x86_cons (exp, size)
3622 expressionS *exp;
3623 int size;
3624 {
3625 if (size == 4)
3626 {
3627 /* Handle @GOTOFF and the like in an expression. */
3628 char *save;
3629 char *gotfree_input_line;
3630 int adjust;
3631
3632 save = input_line_pointer;
3633 gotfree_input_line = lex_got (&got_reloc, &adjust);
3634 if (gotfree_input_line)
3635 input_line_pointer = gotfree_input_line;
3636
3637 expression (exp);
3638
3639 if (gotfree_input_line)
3640 {
3641 /* expression () has merrily parsed up to the end of line,
3642 or a comma - in the wrong buffer. Transfer how far
3643 input_line_pointer has moved to the right buffer. */
3644 input_line_pointer = (save
3645 + (input_line_pointer - gotfree_input_line)
3646 + adjust);
3647 free (gotfree_input_line);
3648 }
3649 }
3650 else
3651 expression (exp);
3652 }
3653 #endif
3654
3655 #ifdef TE_PE
3656
3657 #define O_secrel (O_max + 1)
3658
3659 void
3660 x86_pe_cons_fix_new (frag, off, len, exp)
3661 fragS *frag;
3662 unsigned int off;
3663 unsigned int len;
3664 expressionS *exp;
3665 {
3666 enum bfd_reloc_code_real r = reloc (len, 0, 0, NO_RELOC);
3667
3668 if (exp->X_op == O_secrel)
3669 {
3670 exp->X_op = O_symbol;
3671 r = BFD_RELOC_32_SECREL;
3672 }
3673
3674 fix_new_exp (frag, off, len, exp, 0, r);
3675 }
3676
3677 static void
3678 pe_directive_secrel (dummy)
3679 int dummy ATTRIBUTE_UNUSED;
3680 {
3681 expressionS exp;
3682
3683 do
3684 {
3685 expression (&exp);
3686 if (exp.X_op == O_symbol)
3687 exp.X_op = O_secrel;
3688
3689 emit_expr (&exp, 4);
3690 }
3691 while (*input_line_pointer++ == ',');
3692
3693 input_line_pointer--;
3694 demand_empty_rest_of_line ();
3695 }
3696
3697 #endif
3698
3699 static int i386_immediate PARAMS ((char *));
3700
3701 static int
3702 i386_immediate (imm_start)
3703 char *imm_start;
3704 {
3705 char *save_input_line_pointer;
3706 #ifndef LEX_AT
3707 char *gotfree_input_line;
3708 #endif
3709 segT exp_seg = 0;
3710 expressionS *exp;
3711
3712 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
3713 {
3714 as_bad (_("only 1 or 2 immediate operands are allowed"));
3715 return 0;
3716 }
3717
3718 exp = &im_expressions[i.imm_operands++];
3719 i.op[this_operand].imms = exp;
3720
3721 if (is_space_char (*imm_start))
3722 ++imm_start;
3723
3724 save_input_line_pointer = input_line_pointer;
3725 input_line_pointer = imm_start;
3726
3727 #ifndef LEX_AT
3728 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3729 if (gotfree_input_line)
3730 input_line_pointer = gotfree_input_line;
3731 #endif
3732
3733 exp_seg = expression (exp);
3734
3735 SKIP_WHITESPACE ();
3736 if (*input_line_pointer)
3737 as_bad (_("junk `%s' after expression"), input_line_pointer);
3738
3739 input_line_pointer = save_input_line_pointer;
3740 #ifndef LEX_AT
3741 if (gotfree_input_line)
3742 free (gotfree_input_line);
3743 #endif
3744
3745 if (exp->X_op == O_absent || exp->X_op == O_big)
3746 {
3747 /* Missing or bad expr becomes absolute 0. */
3748 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
3749 imm_start);
3750 exp->X_op = O_constant;
3751 exp->X_add_number = 0;
3752 exp->X_add_symbol = (symbolS *) 0;
3753 exp->X_op_symbol = (symbolS *) 0;
3754 }
3755 else if (exp->X_op == O_constant)
3756 {
3757 /* Size it properly later. */
3758 i.types[this_operand] |= Imm64;
3759 /* If BFD64, sign extend val. */
3760 if (!use_rela_relocations)
3761 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
3762 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
3763 }
3764 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3765 else if (OUTPUT_FLAVOR == bfd_target_aout_flavour
3766 && exp_seg != absolute_section
3767 && exp_seg != text_section
3768 && exp_seg != data_section
3769 && exp_seg != bss_section
3770 && exp_seg != undefined_section
3771 && !bfd_is_com_section (exp_seg))
3772 {
3773 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3774 return 0;
3775 }
3776 #endif
3777 else
3778 {
3779 /* This is an address. The size of the address will be
3780 determined later, depending on destination register,
3781 suffix, or the default for the section. */
3782 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
3783 }
3784
3785 return 1;
3786 }
3787
3788 static char *i386_scale PARAMS ((char *));
3789
3790 static char *
3791 i386_scale (scale)
3792 char *scale;
3793 {
3794 offsetT val;
3795 char *save = input_line_pointer;
3796
3797 input_line_pointer = scale;
3798 val = get_absolute_expression ();
3799
3800 switch (val)
3801 {
3802 case 0:
3803 case 1:
3804 i.log2_scale_factor = 0;
3805 break;
3806 case 2:
3807 i.log2_scale_factor = 1;
3808 break;
3809 case 4:
3810 i.log2_scale_factor = 2;
3811 break;
3812 case 8:
3813 i.log2_scale_factor = 3;
3814 break;
3815 default:
3816 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
3817 scale);
3818 input_line_pointer = save;
3819 return NULL;
3820 }
3821 if (i.log2_scale_factor != 0 && i.index_reg == 0)
3822 {
3823 as_warn (_("scale factor of %d without an index register"),
3824 1 << i.log2_scale_factor);
3825 #if SCALE1_WHEN_NO_INDEX
3826 i.log2_scale_factor = 0;
3827 #endif
3828 }
3829 scale = input_line_pointer;
3830 input_line_pointer = save;
3831 return scale;
3832 }
3833
3834 static int i386_displacement PARAMS ((char *, char *));
3835
3836 static int
3837 i386_displacement (disp_start, disp_end)
3838 char *disp_start;
3839 char *disp_end;
3840 {
3841 expressionS *exp;
3842 segT exp_seg = 0;
3843 char *save_input_line_pointer;
3844 #ifndef LEX_AT
3845 char *gotfree_input_line;
3846 #endif
3847 int bigdisp = Disp32;
3848
3849 if (flag_code == CODE_64BIT)
3850 {
3851 if (i.prefix[ADDR_PREFIX] == 0)
3852 bigdisp = Disp64;
3853 }
3854 else if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3855 bigdisp = Disp16;
3856 i.types[this_operand] |= bigdisp;
3857
3858 exp = &disp_expressions[i.disp_operands];
3859 i.op[this_operand].disps = exp;
3860 i.disp_operands++;
3861 save_input_line_pointer = input_line_pointer;
3862 input_line_pointer = disp_start;
3863 END_STRING_AND_SAVE (disp_end);
3864
3865 #ifndef GCC_ASM_O_HACK
3866 #define GCC_ASM_O_HACK 0
3867 #endif
3868 #if GCC_ASM_O_HACK
3869 END_STRING_AND_SAVE (disp_end + 1);
3870 if ((i.types[this_operand] & BaseIndex) != 0
3871 && displacement_string_end[-1] == '+')
3872 {
3873 /* This hack is to avoid a warning when using the "o"
3874 constraint within gcc asm statements.
3875 For instance:
3876
3877 #define _set_tssldt_desc(n,addr,limit,type) \
3878 __asm__ __volatile__ ( \
3879 "movw %w2,%0\n\t" \
3880 "movw %w1,2+%0\n\t" \
3881 "rorl $16,%1\n\t" \
3882 "movb %b1,4+%0\n\t" \
3883 "movb %4,5+%0\n\t" \
3884 "movb $0,6+%0\n\t" \
3885 "movb %h1,7+%0\n\t" \
3886 "rorl $16,%1" \
3887 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
3888
3889 This works great except that the output assembler ends
3890 up looking a bit weird if it turns out that there is
3891 no offset. You end up producing code that looks like:
3892
3893 #APP
3894 movw $235,(%eax)
3895 movw %dx,2+(%eax)
3896 rorl $16,%edx
3897 movb %dl,4+(%eax)
3898 movb $137,5+(%eax)
3899 movb $0,6+(%eax)
3900 movb %dh,7+(%eax)
3901 rorl $16,%edx
3902 #NO_APP
3903
3904 So here we provide the missing zero. */
3905
3906 *displacement_string_end = '0';
3907 }
3908 #endif
3909 #ifndef LEX_AT
3910 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3911 if (gotfree_input_line)
3912 input_line_pointer = gotfree_input_line;
3913 #endif
3914
3915 exp_seg = expression (exp);
3916
3917 SKIP_WHITESPACE ();
3918 if (*input_line_pointer)
3919 as_bad (_("junk `%s' after expression"), input_line_pointer);
3920 #if GCC_ASM_O_HACK
3921 RESTORE_END_STRING (disp_end + 1);
3922 #endif
3923 RESTORE_END_STRING (disp_end);
3924 input_line_pointer = save_input_line_pointer;
3925 #ifndef LEX_AT
3926 if (gotfree_input_line)
3927 free (gotfree_input_line);
3928 #endif
3929
3930 /* We do this to make sure that the section symbol is in
3931 the symbol table. We will ultimately change the relocation
3932 to be relative to the beginning of the section. */
3933 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
3934 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3935 {
3936 if (exp->X_op != O_symbol)
3937 {
3938 as_bad (_("bad expression used with @%s"),
3939 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
3940 ? "GOTPCREL"
3941 : "GOTOFF"));
3942 return 0;
3943 }
3944
3945 if (S_IS_LOCAL (exp->X_add_symbol)
3946 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
3947 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
3948 exp->X_op = O_subtract;
3949 exp->X_op_symbol = GOT_symbol;
3950 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3951 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
3952 else
3953 i.reloc[this_operand] = BFD_RELOC_32;
3954 }
3955
3956 if (exp->X_op == O_absent || exp->X_op == O_big)
3957 {
3958 /* Missing or bad expr becomes absolute 0. */
3959 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
3960 disp_start);
3961 exp->X_op = O_constant;
3962 exp->X_add_number = 0;
3963 exp->X_add_symbol = (symbolS *) 0;
3964 exp->X_op_symbol = (symbolS *) 0;
3965 }
3966
3967 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3968 if (exp->X_op != O_constant
3969 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3970 && exp_seg != absolute_section
3971 && exp_seg != text_section
3972 && exp_seg != data_section
3973 && exp_seg != bss_section
3974 && exp_seg != undefined_section
3975 && !bfd_is_com_section (exp_seg))
3976 {
3977 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3978 return 0;
3979 }
3980 #endif
3981 else if (flag_code == CODE_64BIT)
3982 i.types[this_operand] |= Disp32S | Disp32;
3983 return 1;
3984 }
3985
3986 static int i386_index_check PARAMS ((const char *));
3987
3988 /* Make sure the memory operand we've been dealt is valid.
3989 Return 1 on success, 0 on a failure. */
3990
3991 static int
3992 i386_index_check (operand_string)
3993 const char *operand_string;
3994 {
3995 int ok;
3996 #if INFER_ADDR_PREFIX
3997 int fudged = 0;
3998
3999 tryprefix:
4000 #endif
4001 ok = 1;
4002 if (flag_code == CODE_64BIT)
4003 {
4004 if (i.prefix[ADDR_PREFIX] == 0)
4005 {
4006 /* 64bit checks. */
4007 if ((i.base_reg
4008 && ((i.base_reg->reg_type & Reg64) == 0)
4009 && (i.base_reg->reg_type != BaseIndex
4010 || i.index_reg))
4011 || (i.index_reg
4012 && ((i.index_reg->reg_type & (Reg64 | BaseIndex))
4013 != (Reg64 | BaseIndex))))
4014 ok = 0;
4015 }
4016 else
4017 {
4018 /* 32bit checks. */
4019 if ((i.base_reg
4020 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
4021 || (i.index_reg
4022 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
4023 != (Reg32 | BaseIndex))))
4024 ok = 0;
4025 }
4026 }
4027 else
4028 {
4029 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
4030 {
4031 /* 16bit checks. */
4032 if ((i.base_reg
4033 && ((i.base_reg->reg_type & (Reg16 | BaseIndex | RegRex))
4034 != (Reg16 | BaseIndex)))
4035 || (i.index_reg
4036 && (((i.index_reg->reg_type & (Reg16 | BaseIndex))
4037 != (Reg16 | BaseIndex))
4038 || !(i.base_reg
4039 && i.base_reg->reg_num < 6
4040 && i.index_reg->reg_num >= 6
4041 && i.log2_scale_factor == 0))))
4042 ok = 0;
4043 }
4044 else
4045 {
4046 /* 32bit checks. */
4047 if ((i.base_reg
4048 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
4049 || (i.index_reg
4050 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
4051 != (Reg32 | BaseIndex))))
4052 ok = 0;
4053 }
4054 }
4055 if (!ok)
4056 {
4057 #if INFER_ADDR_PREFIX
4058 if (flag_code != CODE_64BIT
4059 && i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0')
4060 {
4061 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
4062 i.prefixes += 1;
4063 /* Change the size of any displacement too. At most one of
4064 Disp16 or Disp32 is set.
4065 FIXME. There doesn't seem to be any real need for separate
4066 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
4067 Removing them would probably clean up the code quite a lot. */
4068 if (i.types[this_operand] & (Disp16 | Disp32))
4069 i.types[this_operand] ^= (Disp16 | Disp32);
4070 fudged = 1;
4071 goto tryprefix;
4072 }
4073 if (fudged)
4074 as_bad (_("`%s' is not a valid base/index expression"),
4075 operand_string);
4076 else
4077 #endif
4078 as_bad (_("`%s' is not a valid %s bit base/index expression"),
4079 operand_string,
4080 flag_code_names[flag_code]);
4081 return 0;
4082 }
4083 return 1;
4084 }
4085
4086 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
4087 on error. */
4088
4089 static int
4090 i386_operand (operand_string)
4091 char *operand_string;
4092 {
4093 const reg_entry *r;
4094 char *end_op;
4095 char *op_string = operand_string;
4096
4097 if (is_space_char (*op_string))
4098 ++op_string;
4099
4100 /* We check for an absolute prefix (differentiating,
4101 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
4102 if (*op_string == ABSOLUTE_PREFIX)
4103 {
4104 ++op_string;
4105 if (is_space_char (*op_string))
4106 ++op_string;
4107 i.types[this_operand] |= JumpAbsolute;
4108 }
4109
4110 /* Check if operand is a register. */
4111 if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
4112 && (r = parse_register (op_string, &end_op)) != NULL)
4113 {
4114 /* Check for a segment override by searching for ':' after a
4115 segment register. */
4116 op_string = end_op;
4117 if (is_space_char (*op_string))
4118 ++op_string;
4119 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
4120 {
4121 switch (r->reg_num)
4122 {
4123 case 0:
4124 i.seg[i.mem_operands] = &es;
4125 break;
4126 case 1:
4127 i.seg[i.mem_operands] = &cs;
4128 break;
4129 case 2:
4130 i.seg[i.mem_operands] = &ss;
4131 break;
4132 case 3:
4133 i.seg[i.mem_operands] = &ds;
4134 break;
4135 case 4:
4136 i.seg[i.mem_operands] = &fs;
4137 break;
4138 case 5:
4139 i.seg[i.mem_operands] = &gs;
4140 break;
4141 }
4142
4143 /* Skip the ':' and whitespace. */
4144 ++op_string;
4145 if (is_space_char (*op_string))
4146 ++op_string;
4147
4148 if (!is_digit_char (*op_string)
4149 && !is_identifier_char (*op_string)
4150 && *op_string != '('
4151 && *op_string != ABSOLUTE_PREFIX)
4152 {
4153 as_bad (_("bad memory operand `%s'"), op_string);
4154 return 0;
4155 }
4156 /* Handle case of %es:*foo. */
4157 if (*op_string == ABSOLUTE_PREFIX)
4158 {
4159 ++op_string;
4160 if (is_space_char (*op_string))
4161 ++op_string;
4162 i.types[this_operand] |= JumpAbsolute;
4163 }
4164 goto do_memory_reference;
4165 }
4166 if (*op_string)
4167 {
4168 as_bad (_("junk `%s' after register"), op_string);
4169 return 0;
4170 }
4171 i.types[this_operand] |= r->reg_type & ~BaseIndex;
4172 i.op[this_operand].regs = r;
4173 i.reg_operands++;
4174 }
4175 else if (*op_string == REGISTER_PREFIX)
4176 {
4177 as_bad (_("bad register name `%s'"), op_string);
4178 return 0;
4179 }
4180 else if (*op_string == IMMEDIATE_PREFIX)
4181 {
4182 ++op_string;
4183 if (i.types[this_operand] & JumpAbsolute)
4184 {
4185 as_bad (_("immediate operand illegal with absolute jump"));
4186 return 0;
4187 }
4188 if (!i386_immediate (op_string))
4189 return 0;
4190 }
4191 else if (is_digit_char (*op_string)
4192 || is_identifier_char (*op_string)
4193 || *op_string == '(')
4194 {
4195 /* This is a memory reference of some sort. */
4196 char *base_string;
4197
4198 /* Start and end of displacement string expression (if found). */
4199 char *displacement_string_start;
4200 char *displacement_string_end;
4201
4202 do_memory_reference:
4203 if ((i.mem_operands == 1
4204 && (current_templates->start->opcode_modifier & IsString) == 0)
4205 || i.mem_operands == 2)
4206 {
4207 as_bad (_("too many memory references for `%s'"),
4208 current_templates->start->name);
4209 return 0;
4210 }
4211
4212 /* Check for base index form. We detect the base index form by
4213 looking for an ')' at the end of the operand, searching
4214 for the '(' matching it, and finding a REGISTER_PREFIX or ','
4215 after the '('. */
4216 base_string = op_string + strlen (op_string);
4217
4218 --base_string;
4219 if (is_space_char (*base_string))
4220 --base_string;
4221
4222 /* If we only have a displacement, set-up for it to be parsed later. */
4223 displacement_string_start = op_string;
4224 displacement_string_end = base_string + 1;
4225
4226 if (*base_string == ')')
4227 {
4228 char *temp_string;
4229 unsigned int parens_balanced = 1;
4230 /* We've already checked that the number of left & right ()'s are
4231 equal, so this loop will not be infinite. */
4232 do
4233 {
4234 base_string--;
4235 if (*base_string == ')')
4236 parens_balanced++;
4237 if (*base_string == '(')
4238 parens_balanced--;
4239 }
4240 while (parens_balanced);
4241
4242 temp_string = base_string;
4243
4244 /* Skip past '(' and whitespace. */
4245 ++base_string;
4246 if (is_space_char (*base_string))
4247 ++base_string;
4248
4249 if (*base_string == ','
4250 || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
4251 && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
4252 {
4253 displacement_string_end = temp_string;
4254
4255 i.types[this_operand] |= BaseIndex;
4256
4257 if (i.base_reg)
4258 {
4259 base_string = end_op;
4260 if (is_space_char (*base_string))
4261 ++base_string;
4262 }
4263
4264 /* There may be an index reg or scale factor here. */
4265 if (*base_string == ',')
4266 {
4267 ++base_string;
4268 if (is_space_char (*base_string))
4269 ++base_string;
4270
4271 if ((*base_string == REGISTER_PREFIX || allow_naked_reg)
4272 && (i.index_reg = parse_register (base_string, &end_op)) != NULL)
4273 {
4274 base_string = end_op;
4275 if (is_space_char (*base_string))
4276 ++base_string;
4277 if (*base_string == ',')
4278 {
4279 ++base_string;
4280 if (is_space_char (*base_string))
4281 ++base_string;
4282 }
4283 else if (*base_string != ')')
4284 {
4285 as_bad (_("expecting `,' or `)' after index register in `%s'"),
4286 operand_string);
4287 return 0;
4288 }
4289 }
4290 else if (*base_string == REGISTER_PREFIX)
4291 {
4292 as_bad (_("bad register name `%s'"), base_string);
4293 return 0;
4294 }
4295
4296 /* Check for scale factor. */
4297 if (*base_string != ')')
4298 {
4299 char *end_scale = i386_scale (base_string);
4300
4301 if (!end_scale)
4302 return 0;
4303
4304 base_string = end_scale;
4305 if (is_space_char (*base_string))
4306 ++base_string;
4307 if (*base_string != ')')
4308 {
4309 as_bad (_("expecting `)' after scale factor in `%s'"),
4310 operand_string);
4311 return 0;
4312 }
4313 }
4314 else if (!i.index_reg)
4315 {
4316 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
4317 *base_string);
4318 return 0;
4319 }
4320 }
4321 else if (*base_string != ')')
4322 {
4323 as_bad (_("expecting `,' or `)' after base register in `%s'"),
4324 operand_string);
4325 return 0;
4326 }
4327 }
4328 else if (*base_string == REGISTER_PREFIX)
4329 {
4330 as_bad (_("bad register name `%s'"), base_string);
4331 return 0;
4332 }
4333 }
4334
4335 /* If there's an expression beginning the operand, parse it,
4336 assuming displacement_string_start and
4337 displacement_string_end are meaningful. */
4338 if (displacement_string_start != displacement_string_end)
4339 {
4340 if (!i386_displacement (displacement_string_start,
4341 displacement_string_end))
4342 return 0;
4343 }
4344
4345 /* Special case for (%dx) while doing input/output op. */
4346 if (i.base_reg
4347 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
4348 && i.index_reg == 0
4349 && i.log2_scale_factor == 0
4350 && i.seg[i.mem_operands] == 0
4351 && (i.types[this_operand] & Disp) == 0)
4352 {
4353 i.types[this_operand] = InOutPortReg;
4354 return 1;
4355 }
4356
4357 if (i386_index_check (operand_string) == 0)
4358 return 0;
4359 i.mem_operands++;
4360 }
4361 else
4362 {
4363 /* It's not a memory operand; argh! */
4364 as_bad (_("invalid char %s beginning operand %d `%s'"),
4365 output_invalid (*op_string),
4366 this_operand + 1,
4367 op_string);
4368 return 0;
4369 }
4370 return 1; /* Normal return. */
4371 }
4372 \f
4373 /* md_estimate_size_before_relax()
4374
4375 Called just before relax() for rs_machine_dependent frags. The x86
4376 assembler uses these frags to handle variable size jump
4377 instructions.
4378
4379 Any symbol that is now undefined will not become defined.
4380 Return the correct fr_subtype in the frag.
4381 Return the initial "guess for variable size of frag" to caller.
4382 The guess is actually the growth beyond the fixed part. Whatever
4383 we do to grow the fixed or variable part contributes to our
4384 returned value. */
4385
4386 int
4387 md_estimate_size_before_relax (fragP, segment)
4388 fragS *fragP;
4389 segT segment;
4390 {
4391 /* We've already got fragP->fr_subtype right; all we have to do is
4392 check for un-relaxable symbols. On an ELF system, we can't relax
4393 an externally visible symbol, because it may be overridden by a
4394 shared library. */
4395 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
4396 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4397 || (OUTPUT_FLAVOR == bfd_target_elf_flavour
4398 && (S_IS_EXTERNAL (fragP->fr_symbol)
4399 || S_IS_WEAK (fragP->fr_symbol)))
4400 #endif
4401 )
4402 {
4403 /* Symbol is undefined in this segment, or we need to keep a
4404 reloc so that weak symbols can be overridden. */
4405 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
4406 enum bfd_reloc_code_real reloc_type;
4407 unsigned char *opcode;
4408 int old_fr_fix;
4409
4410 if (fragP->fr_var != NO_RELOC)
4411 reloc_type = fragP->fr_var;
4412 else if (size == 2)
4413 reloc_type = BFD_RELOC_16_PCREL;
4414 else
4415 reloc_type = BFD_RELOC_32_PCREL;
4416
4417 old_fr_fix = fragP->fr_fix;
4418 opcode = (unsigned char *) fragP->fr_opcode;
4419
4420 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
4421 {
4422 case UNCOND_JUMP:
4423 /* Make jmp (0xeb) a (d)word displacement jump. */
4424 opcode[0] = 0xe9;
4425 fragP->fr_fix += size;
4426 fix_new (fragP, old_fr_fix, size,
4427 fragP->fr_symbol,
4428 fragP->fr_offset, 1,
4429 reloc_type);
4430 break;
4431
4432 case COND_JUMP86:
4433 if (size == 2
4434 && (!no_cond_jump_promotion || fragP->fr_var != NO_RELOC))
4435 {
4436 /* Negate the condition, and branch past an
4437 unconditional jump. */
4438 opcode[0] ^= 1;
4439 opcode[1] = 3;
4440 /* Insert an unconditional jump. */
4441 opcode[2] = 0xe9;
4442 /* We added two extra opcode bytes, and have a two byte
4443 offset. */
4444 fragP->fr_fix += 2 + 2;
4445 fix_new (fragP, old_fr_fix + 2, 2,
4446 fragP->fr_symbol,
4447 fragP->fr_offset, 1,
4448 reloc_type);
4449 break;
4450 }
4451 /* Fall through. */
4452
4453 case COND_JUMP:
4454 if (no_cond_jump_promotion && fragP->fr_var == NO_RELOC)
4455 {
4456 fixS *fixP;
4457
4458 fragP->fr_fix += 1;
4459 fixP = fix_new (fragP, old_fr_fix, 1,
4460 fragP->fr_symbol,
4461 fragP->fr_offset, 1,
4462 BFD_RELOC_8_PCREL);
4463 fixP->fx_signed = 1;
4464 break;
4465 }
4466
4467 /* This changes the byte-displacement jump 0x7N
4468 to the (d)word-displacement jump 0x0f,0x8N. */
4469 opcode[1] = opcode[0] + 0x10;
4470 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4471 /* We've added an opcode byte. */
4472 fragP->fr_fix += 1 + size;
4473 fix_new (fragP, old_fr_fix + 1, size,
4474 fragP->fr_symbol,
4475 fragP->fr_offset, 1,
4476 reloc_type);
4477 break;
4478
4479 default:
4480 BAD_CASE (fragP->fr_subtype);
4481 break;
4482 }
4483 frag_wane (fragP);
4484 return fragP->fr_fix - old_fr_fix;
4485 }
4486
4487 /* Guess size depending on current relax state. Initially the relax
4488 state will correspond to a short jump and we return 1, because
4489 the variable part of the frag (the branch offset) is one byte
4490 long. However, we can relax a section more than once and in that
4491 case we must either set fr_subtype back to the unrelaxed state,
4492 or return the value for the appropriate branch. */
4493 return md_relax_table[fragP->fr_subtype].rlx_length;
4494 }
4495
4496 /* Called after relax() is finished.
4497
4498 In: Address of frag.
4499 fr_type == rs_machine_dependent.
4500 fr_subtype is what the address relaxed to.
4501
4502 Out: Any fixSs and constants are set up.
4503 Caller will turn frag into a ".space 0". */
4504
4505 void
4506 md_convert_frag (abfd, sec, fragP)
4507 bfd *abfd ATTRIBUTE_UNUSED;
4508 segT sec ATTRIBUTE_UNUSED;
4509 fragS *fragP;
4510 {
4511 unsigned char *opcode;
4512 unsigned char *where_to_put_displacement = NULL;
4513 offsetT target_address;
4514 offsetT opcode_address;
4515 unsigned int extension = 0;
4516 offsetT displacement_from_opcode_start;
4517
4518 opcode = (unsigned char *) fragP->fr_opcode;
4519
4520 /* Address we want to reach in file space. */
4521 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
4522
4523 /* Address opcode resides at in file space. */
4524 opcode_address = fragP->fr_address + fragP->fr_fix;
4525
4526 /* Displacement from opcode start to fill into instruction. */
4527 displacement_from_opcode_start = target_address - opcode_address;
4528
4529 if ((fragP->fr_subtype & BIG) == 0)
4530 {
4531 /* Don't have to change opcode. */
4532 extension = 1; /* 1 opcode + 1 displacement */
4533 where_to_put_displacement = &opcode[1];
4534 }
4535 else
4536 {
4537 if (no_cond_jump_promotion
4538 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
4539 as_warn_where (fragP->fr_file, fragP->fr_line, _("long jump required"));
4540
4541 switch (fragP->fr_subtype)
4542 {
4543 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
4544 extension = 4; /* 1 opcode + 4 displacement */
4545 opcode[0] = 0xe9;
4546 where_to_put_displacement = &opcode[1];
4547 break;
4548
4549 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
4550 extension = 2; /* 1 opcode + 2 displacement */
4551 opcode[0] = 0xe9;
4552 where_to_put_displacement = &opcode[1];
4553 break;
4554
4555 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
4556 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
4557 extension = 5; /* 2 opcode + 4 displacement */
4558 opcode[1] = opcode[0] + 0x10;
4559 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4560 where_to_put_displacement = &opcode[2];
4561 break;
4562
4563 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
4564 extension = 3; /* 2 opcode + 2 displacement */
4565 opcode[1] = opcode[0] + 0x10;
4566 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4567 where_to_put_displacement = &opcode[2];
4568 break;
4569
4570 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
4571 extension = 4;
4572 opcode[0] ^= 1;
4573 opcode[1] = 3;
4574 opcode[2] = 0xe9;
4575 where_to_put_displacement = &opcode[3];
4576 break;
4577
4578 default:
4579 BAD_CASE (fragP->fr_subtype);
4580 break;
4581 }
4582 }
4583
4584 /* Now put displacement after opcode. */
4585 md_number_to_chars ((char *) where_to_put_displacement,
4586 (valueT) (displacement_from_opcode_start - extension),
4587 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
4588 fragP->fr_fix += extension;
4589 }
4590 \f
4591 /* Size of byte displacement jmp. */
4592 int md_short_jump_size = 2;
4593
4594 /* Size of dword displacement jmp. */
4595 int md_long_jump_size = 5;
4596
4597 /* Size of relocation record. */
4598 const int md_reloc_size = 8;
4599
4600 void
4601 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
4602 char *ptr;
4603 addressT from_addr, to_addr;
4604 fragS *frag ATTRIBUTE_UNUSED;
4605 symbolS *to_symbol ATTRIBUTE_UNUSED;
4606 {
4607 offsetT offset;
4608
4609 offset = to_addr - (from_addr + 2);
4610 /* Opcode for byte-disp jump. */
4611 md_number_to_chars (ptr, (valueT) 0xeb, 1);
4612 md_number_to_chars (ptr + 1, (valueT) offset, 1);
4613 }
4614
4615 void
4616 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
4617 char *ptr;
4618 addressT from_addr, to_addr;
4619 fragS *frag ATTRIBUTE_UNUSED;
4620 symbolS *to_symbol ATTRIBUTE_UNUSED;
4621 {
4622 offsetT offset;
4623
4624 offset = to_addr - (from_addr + 5);
4625 md_number_to_chars (ptr, (valueT) 0xe9, 1);
4626 md_number_to_chars (ptr + 1, (valueT) offset, 4);
4627 }
4628 \f
4629 /* Apply a fixup (fixS) to segment data, once it has been determined
4630 by our caller that we have all the info we need to fix it up.
4631
4632 On the 386, immediates, displacements, and data pointers are all in
4633 the same (little-endian) format, so we don't need to care about which
4634 we are handling. */
4635
4636 void
4637 md_apply_fix3 (fixP, valP, seg)
4638 /* The fix we're to put in. */
4639 fixS *fixP;
4640 /* Pointer to the value of the bits. */
4641 valueT *valP;
4642 /* Segment fix is from. */
4643 segT seg ATTRIBUTE_UNUSED;
4644 {
4645 char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
4646 valueT value = *valP;
4647
4648 #if !defined (TE_Mach)
4649 if (fixP->fx_pcrel)
4650 {
4651 switch (fixP->fx_r_type)
4652 {
4653 default:
4654 break;
4655
4656 case BFD_RELOC_32:
4657 fixP->fx_r_type = BFD_RELOC_32_PCREL;
4658 break;
4659 case BFD_RELOC_16:
4660 fixP->fx_r_type = BFD_RELOC_16_PCREL;
4661 break;
4662 case BFD_RELOC_8:
4663 fixP->fx_r_type = BFD_RELOC_8_PCREL;
4664 break;
4665 }
4666 }
4667
4668 if (fixP->fx_addsy != NULL
4669 && (fixP->fx_r_type == BFD_RELOC_32_PCREL
4670 || fixP->fx_r_type == BFD_RELOC_16_PCREL
4671 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
4672 && !use_rela_relocations)
4673 {
4674 /* This is a hack. There should be a better way to handle this.
4675 This covers for the fact that bfd_install_relocation will
4676 subtract the current location (for partial_inplace, PC relative
4677 relocations); see more below. */
4678 #ifndef OBJ_AOUT
4679 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4680 #ifdef TE_PE
4681 || OUTPUT_FLAVOR == bfd_target_coff_flavour
4682 #endif
4683 )
4684 value += fixP->fx_where + fixP->fx_frag->fr_address;
4685 #endif
4686 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4687 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
4688 {
4689 segT sym_seg = S_GET_SEGMENT (fixP->fx_addsy);
4690
4691 if ((sym_seg == seg
4692 || (symbol_section_p (fixP->fx_addsy)
4693 && sym_seg != absolute_section))
4694 && !generic_force_reloc (fixP))
4695 {
4696 /* Yes, we add the values in twice. This is because
4697 bfd_install_relocation subtracts them out again. I think
4698 bfd_install_relocation is broken, but I don't dare change
4699 it. FIXME. */
4700 value += fixP->fx_where + fixP->fx_frag->fr_address;
4701 }
4702 }
4703 #endif
4704 #if defined (OBJ_COFF) && defined (TE_PE)
4705 /* For some reason, the PE format does not store a section
4706 address offset for a PC relative symbol. */
4707 if (S_GET_SEGMENT (fixP->fx_addsy) != seg)
4708 value += md_pcrel_from (fixP);
4709 #endif
4710 }
4711
4712 /* Fix a few things - the dynamic linker expects certain values here,
4713 and we must not disappoint it. */
4714 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4715 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4716 && fixP->fx_addsy)
4717 switch (fixP->fx_r_type)
4718 {
4719 case BFD_RELOC_386_PLT32:
4720 case BFD_RELOC_X86_64_PLT32:
4721 /* Make the jump instruction point to the address of the operand. At
4722 runtime we merely add the offset to the actual PLT entry. */
4723 value = -4;
4724 break;
4725
4726 case BFD_RELOC_386_TLS_GD:
4727 case BFD_RELOC_386_TLS_LDM:
4728 case BFD_RELOC_386_TLS_IE_32:
4729 case BFD_RELOC_386_TLS_IE:
4730 case BFD_RELOC_386_TLS_GOTIE:
4731 case BFD_RELOC_X86_64_TLSGD:
4732 case BFD_RELOC_X86_64_TLSLD:
4733 case BFD_RELOC_X86_64_GOTTPOFF:
4734 value = 0; /* Fully resolved at runtime. No addend. */
4735 /* Fallthrough */
4736 case BFD_RELOC_386_TLS_LE:
4737 case BFD_RELOC_386_TLS_LDO_32:
4738 case BFD_RELOC_386_TLS_LE_32:
4739 case BFD_RELOC_X86_64_DTPOFF32:
4740 case BFD_RELOC_X86_64_TPOFF32:
4741 S_SET_THREAD_LOCAL (fixP->fx_addsy);
4742 break;
4743
4744 case BFD_RELOC_386_GOT32:
4745 case BFD_RELOC_X86_64_GOT32:
4746 value = 0; /* Fully resolved at runtime. No addend. */
4747 break;
4748
4749 case BFD_RELOC_VTABLE_INHERIT:
4750 case BFD_RELOC_VTABLE_ENTRY:
4751 fixP->fx_done = 0;
4752 return;
4753
4754 default:
4755 break;
4756 }
4757 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
4758 *valP = value;
4759 #endif /* !defined (TE_Mach) */
4760
4761 /* Are we finished with this relocation now? */
4762 if (fixP->fx_addsy == NULL)
4763 fixP->fx_done = 1;
4764 else if (use_rela_relocations)
4765 {
4766 fixP->fx_no_overflow = 1;
4767 /* Remember value for tc_gen_reloc. */
4768 fixP->fx_addnumber = value;
4769 value = 0;
4770 }
4771
4772 md_number_to_chars (p, value, fixP->fx_size);
4773 }
4774 \f
4775 #define MAX_LITTLENUMS 6
4776
4777 /* Turn the string pointed to by litP into a floating point constant
4778 of type TYPE, and emit the appropriate bytes. The number of
4779 LITTLENUMS emitted is stored in *SIZEP. An error message is
4780 returned, or NULL on OK. */
4781
4782 char *
4783 md_atof (type, litP, sizeP)
4784 int type;
4785 char *litP;
4786 int *sizeP;
4787 {
4788 int prec;
4789 LITTLENUM_TYPE words[MAX_LITTLENUMS];
4790 LITTLENUM_TYPE *wordP;
4791 char *t;
4792
4793 switch (type)
4794 {
4795 case 'f':
4796 case 'F':
4797 prec = 2;
4798 break;
4799
4800 case 'd':
4801 case 'D':
4802 prec = 4;
4803 break;
4804
4805 case 'x':
4806 case 'X':
4807 prec = 5;
4808 break;
4809
4810 default:
4811 *sizeP = 0;
4812 return _("Bad call to md_atof ()");
4813 }
4814 t = atof_ieee (input_line_pointer, type, words);
4815 if (t)
4816 input_line_pointer = t;
4817
4818 *sizeP = prec * sizeof (LITTLENUM_TYPE);
4819 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
4820 the bigendian 386. */
4821 for (wordP = words + prec - 1; prec--;)
4822 {
4823 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
4824 litP += sizeof (LITTLENUM_TYPE);
4825 }
4826 return 0;
4827 }
4828 \f
4829 char output_invalid_buf[8];
4830
4831 static char *
4832 output_invalid (c)
4833 int c;
4834 {
4835 if (ISPRINT (c))
4836 sprintf (output_invalid_buf, "'%c'", c);
4837 else
4838 sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
4839 return output_invalid_buf;
4840 }
4841
4842 /* REG_STRING starts *before* REGISTER_PREFIX. */
4843
4844 static const reg_entry *
4845 parse_register (reg_string, end_op)
4846 char *reg_string;
4847 char **end_op;
4848 {
4849 char *s = reg_string;
4850 char *p;
4851 char reg_name_given[MAX_REG_NAME_SIZE + 1];
4852 const reg_entry *r;
4853
4854 /* Skip possible REGISTER_PREFIX and possible whitespace. */
4855 if (*s == REGISTER_PREFIX)
4856 ++s;
4857
4858 if (is_space_char (*s))
4859 ++s;
4860
4861 p = reg_name_given;
4862 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
4863 {
4864 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
4865 return (const reg_entry *) NULL;
4866 s++;
4867 }
4868
4869 /* For naked regs, make sure that we are not dealing with an identifier.
4870 This prevents confusing an identifier like `eax_var' with register
4871 `eax'. */
4872 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
4873 return (const reg_entry *) NULL;
4874
4875 *end_op = s;
4876
4877 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
4878
4879 /* Handle floating point regs, allowing spaces in the (i) part. */
4880 if (r == i386_regtab /* %st is first entry of table */)
4881 {
4882 if (is_space_char (*s))
4883 ++s;
4884 if (*s == '(')
4885 {
4886 ++s;
4887 if (is_space_char (*s))
4888 ++s;
4889 if (*s >= '0' && *s <= '7')
4890 {
4891 r = &i386_float_regtab[*s - '0'];
4892 ++s;
4893 if (is_space_char (*s))
4894 ++s;
4895 if (*s == ')')
4896 {
4897 *end_op = s + 1;
4898 return r;
4899 }
4900 }
4901 /* We have "%st(" then garbage. */
4902 return (const reg_entry *) NULL;
4903 }
4904 }
4905
4906 if (r != NULL
4907 && (r->reg_flags & (RegRex64 | RegRex)) != 0
4908 && flag_code != CODE_64BIT)
4909 {
4910 return (const reg_entry *) NULL;
4911 }
4912
4913 return r;
4914 }
4915 \f
4916 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4917 const char *md_shortopts = "kVQ:sqn";
4918 #else
4919 const char *md_shortopts = "qn";
4920 #endif
4921
4922 struct option md_longopts[] = {
4923 #define OPTION_32 (OPTION_MD_BASE + 0)
4924 {"32", no_argument, NULL, OPTION_32},
4925 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4926 #define OPTION_64 (OPTION_MD_BASE + 1)
4927 {"64", no_argument, NULL, OPTION_64},
4928 #endif
4929 {NULL, no_argument, NULL, 0}
4930 };
4931 size_t md_longopts_size = sizeof (md_longopts);
4932
4933 int
4934 md_parse_option (c, arg)
4935 int c;
4936 char *arg ATTRIBUTE_UNUSED;
4937 {
4938 switch (c)
4939 {
4940 case 'n':
4941 optimize_align_code = 0;
4942 break;
4943
4944 case 'q':
4945 quiet_warnings = 1;
4946 break;
4947
4948 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4949 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
4950 should be emitted or not. FIXME: Not implemented. */
4951 case 'Q':
4952 break;
4953
4954 /* -V: SVR4 argument to print version ID. */
4955 case 'V':
4956 print_version_id ();
4957 break;
4958
4959 /* -k: Ignore for FreeBSD compatibility. */
4960 case 'k':
4961 break;
4962
4963 case 's':
4964 /* -s: On i386 Solaris, this tells the native assembler to use
4965 .stab instead of .stab.excl. We always use .stab anyhow. */
4966 break;
4967
4968 case OPTION_64:
4969 {
4970 const char **list, **l;
4971
4972 list = bfd_target_list ();
4973 for (l = list; *l != NULL; l++)
4974 if (strcmp (*l, "elf64-x86-64") == 0)
4975 {
4976 default_arch = "x86_64";
4977 break;
4978 }
4979 if (*l == NULL)
4980 as_fatal (_("No compiled in support for x86_64"));
4981 free (list);
4982 }
4983 break;
4984 #endif
4985
4986 case OPTION_32:
4987 default_arch = "i386";
4988 break;
4989
4990 default:
4991 return 0;
4992 }
4993 return 1;
4994 }
4995
4996 void
4997 md_show_usage (stream)
4998 FILE *stream;
4999 {
5000 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5001 fprintf (stream, _("\
5002 -Q ignored\n\
5003 -V print assembler version number\n\
5004 -k ignored\n\
5005 -n Do not optimize code alignment\n\
5006 -q quieten some warnings\n\
5007 -s ignored\n"));
5008 #else
5009 fprintf (stream, _("\
5010 -n Do not optimize code alignment\n\
5011 -q quieten some warnings\n"));
5012 #endif
5013 }
5014
5015 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
5016 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
5017
5018 /* Pick the target format to use. */
5019
5020 const char *
5021 i386_target_format ()
5022 {
5023 if (!strcmp (default_arch, "x86_64"))
5024 set_code_flag (CODE_64BIT);
5025 else if (!strcmp (default_arch, "i386"))
5026 set_code_flag (CODE_32BIT);
5027 else
5028 as_fatal (_("Unknown architecture"));
5029 switch (OUTPUT_FLAVOR)
5030 {
5031 #ifdef OBJ_MAYBE_AOUT
5032 case bfd_target_aout_flavour:
5033 return AOUT_TARGET_FORMAT;
5034 #endif
5035 #ifdef OBJ_MAYBE_COFF
5036 case bfd_target_coff_flavour:
5037 return "coff-i386";
5038 #endif
5039 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
5040 case bfd_target_elf_flavour:
5041 {
5042 if (flag_code == CODE_64BIT)
5043 use_rela_relocations = 1;
5044 return flag_code == CODE_64BIT ? "elf64-x86-64" : ELF_TARGET_FORMAT;
5045 }
5046 #endif
5047 default:
5048 abort ();
5049 return NULL;
5050 }
5051 }
5052
5053 #endif /* OBJ_MAYBE_ more than one */
5054
5055 #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
5056 void i386_elf_emit_arch_note ()
5057 {
5058 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
5059 && cpu_arch_name != NULL)
5060 {
5061 char *p;
5062 asection *seg = now_seg;
5063 subsegT subseg = now_subseg;
5064 Elf_Internal_Note i_note;
5065 Elf_External_Note e_note;
5066 asection *note_secp;
5067 int len;
5068
5069 /* Create the .note section. */
5070 note_secp = subseg_new (".note", 0);
5071 bfd_set_section_flags (stdoutput,
5072 note_secp,
5073 SEC_HAS_CONTENTS | SEC_READONLY);
5074
5075 /* Process the arch string. */
5076 len = strlen (cpu_arch_name);
5077
5078 i_note.namesz = len + 1;
5079 i_note.descsz = 0;
5080 i_note.type = NT_ARCH;
5081 p = frag_more (sizeof (e_note.namesz));
5082 md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
5083 p = frag_more (sizeof (e_note.descsz));
5084 md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
5085 p = frag_more (sizeof (e_note.type));
5086 md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
5087 p = frag_more (len + 1);
5088 strcpy (p, cpu_arch_name);
5089
5090 frag_align (2, 0, 0);
5091
5092 subseg_set (seg, subseg);
5093 }
5094 }
5095 #endif
5096 \f
5097 symbolS *
5098 md_undefined_symbol (name)
5099 char *name;
5100 {
5101 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
5102 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
5103 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
5104 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
5105 {
5106 if (!GOT_symbol)
5107 {
5108 if (symbol_find (name))
5109 as_bad (_("GOT already in symbol table"));
5110 GOT_symbol = symbol_new (name, undefined_section,
5111 (valueT) 0, &zero_address_frag);
5112 };
5113 return GOT_symbol;
5114 }
5115 return 0;
5116 }
5117
5118 /* Round up a section size to the appropriate boundary. */
5119
5120 valueT
5121 md_section_align (segment, size)
5122 segT segment ATTRIBUTE_UNUSED;
5123 valueT size;
5124 {
5125 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
5126 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
5127 {
5128 /* For a.out, force the section size to be aligned. If we don't do
5129 this, BFD will align it for us, but it will not write out the
5130 final bytes of the section. This may be a bug in BFD, but it is
5131 easier to fix it here since that is how the other a.out targets
5132 work. */
5133 int align;
5134
5135 align = bfd_get_section_alignment (stdoutput, segment);
5136 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
5137 }
5138 #endif
5139
5140 return size;
5141 }
5142
5143 /* On the i386, PC-relative offsets are relative to the start of the
5144 next instruction. That is, the address of the offset, plus its
5145 size, since the offset is always the last part of the insn. */
5146
5147 long
5148 md_pcrel_from (fixP)
5149 fixS *fixP;
5150 {
5151 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
5152 }
5153
5154 #ifndef I386COFF
5155
5156 static void
5157 s_bss (ignore)
5158 int ignore ATTRIBUTE_UNUSED;
5159 {
5160 int temp;
5161
5162 temp = get_absolute_expression ();
5163 subseg_set (bss_section, (subsegT) temp);
5164 demand_empty_rest_of_line ();
5165 }
5166
5167 #endif
5168
5169 void
5170 i386_validate_fix (fixp)
5171 fixS *fixp;
5172 {
5173 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
5174 {
5175 /* GOTOFF relocation are nonsense in 64bit mode. */
5176 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
5177 {
5178 if (flag_code != CODE_64BIT)
5179 abort ();
5180 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
5181 }
5182 else
5183 {
5184 if (flag_code == CODE_64BIT)
5185 abort ();
5186 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
5187 }
5188 fixp->fx_subsy = 0;
5189 }
5190 }
5191
5192 arelent *
5193 tc_gen_reloc (section, fixp)
5194 asection *section ATTRIBUTE_UNUSED;
5195 fixS *fixp;
5196 {
5197 arelent *rel;
5198 bfd_reloc_code_real_type code;
5199
5200 switch (fixp->fx_r_type)
5201 {
5202 case BFD_RELOC_X86_64_PLT32:
5203 case BFD_RELOC_X86_64_GOT32:
5204 case BFD_RELOC_X86_64_GOTPCREL:
5205 case BFD_RELOC_386_PLT32:
5206 case BFD_RELOC_386_GOT32:
5207 case BFD_RELOC_386_GOTOFF:
5208 case BFD_RELOC_386_GOTPC:
5209 case BFD_RELOC_386_TLS_GD:
5210 case BFD_RELOC_386_TLS_LDM:
5211 case BFD_RELOC_386_TLS_LDO_32:
5212 case BFD_RELOC_386_TLS_IE_32:
5213 case BFD_RELOC_386_TLS_IE:
5214 case BFD_RELOC_386_TLS_GOTIE:
5215 case BFD_RELOC_386_TLS_LE_32:
5216 case BFD_RELOC_386_TLS_LE:
5217 case BFD_RELOC_X86_64_32S:
5218 case BFD_RELOC_X86_64_TLSGD:
5219 case BFD_RELOC_X86_64_TLSLD:
5220 case BFD_RELOC_X86_64_DTPOFF32:
5221 case BFD_RELOC_X86_64_GOTTPOFF:
5222 case BFD_RELOC_X86_64_TPOFF32:
5223 case BFD_RELOC_RVA:
5224 case BFD_RELOC_VTABLE_ENTRY:
5225 case BFD_RELOC_VTABLE_INHERIT:
5226 #ifdef TE_PE
5227 case BFD_RELOC_32_SECREL:
5228 #endif
5229 code = fixp->fx_r_type;
5230 break;
5231 default:
5232 if (fixp->fx_pcrel)
5233 {
5234 switch (fixp->fx_size)
5235 {
5236 default:
5237 as_bad_where (fixp->fx_file, fixp->fx_line,
5238 _("can not do %d byte pc-relative relocation"),
5239 fixp->fx_size);
5240 code = BFD_RELOC_32_PCREL;
5241 break;
5242 case 1: code = BFD_RELOC_8_PCREL; break;
5243 case 2: code = BFD_RELOC_16_PCREL; break;
5244 case 4: code = BFD_RELOC_32_PCREL; break;
5245 }
5246 }
5247 else
5248 {
5249 switch (fixp->fx_size)
5250 {
5251 default:
5252 as_bad_where (fixp->fx_file, fixp->fx_line,
5253 _("can not do %d byte relocation"),
5254 fixp->fx_size);
5255 code = BFD_RELOC_32;
5256 break;
5257 case 1: code = BFD_RELOC_8; break;
5258 case 2: code = BFD_RELOC_16; break;
5259 case 4: code = BFD_RELOC_32; break;
5260 #ifdef BFD64
5261 case 8: code = BFD_RELOC_64; break;
5262 #endif
5263 }
5264 }
5265 break;
5266 }
5267
5268 if (code == BFD_RELOC_32
5269 && GOT_symbol
5270 && fixp->fx_addsy == GOT_symbol)
5271 {
5272 /* We don't support GOTPC on 64bit targets. */
5273 if (flag_code == CODE_64BIT)
5274 abort ();
5275 code = BFD_RELOC_386_GOTPC;
5276 }
5277
5278 rel = (arelent *) xmalloc (sizeof (arelent));
5279 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
5280 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
5281
5282 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
5283
5284 #ifdef TE_PE
5285 if (S_IS_WEAK (fixp->fx_addsy))
5286 rel->addend = rel->address - (*rel->sym_ptr_ptr)->value + 4;
5287 else
5288 #endif
5289 if (!use_rela_relocations)
5290 {
5291 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
5292 vtable entry to be used in the relocation's section offset. */
5293 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
5294 rel->address = fixp->fx_offset;
5295
5296 rel->addend = 0;
5297 }
5298 /* Use the rela in 64bit mode. */
5299 else
5300 {
5301 if (!fixp->fx_pcrel)
5302 rel->addend = fixp->fx_offset;
5303 else
5304 switch (code)
5305 {
5306 case BFD_RELOC_X86_64_PLT32:
5307 case BFD_RELOC_X86_64_GOT32:
5308 case BFD_RELOC_X86_64_GOTPCREL:
5309 case BFD_RELOC_X86_64_TLSGD:
5310 case BFD_RELOC_X86_64_TLSLD:
5311 case BFD_RELOC_X86_64_GOTTPOFF:
5312 rel->addend = fixp->fx_offset - fixp->fx_size;
5313 break;
5314 default:
5315 rel->addend = (section->vma
5316 - fixp->fx_size
5317 + fixp->fx_addnumber
5318 + md_pcrel_from (fixp));
5319 break;
5320 }
5321 }
5322
5323 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
5324 if (rel->howto == NULL)
5325 {
5326 as_bad_where (fixp->fx_file, fixp->fx_line,
5327 _("cannot represent relocation type %s"),
5328 bfd_get_reloc_code_name (code));
5329 /* Set howto to a garbage value so that we can keep going. */
5330 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
5331 assert (rel->howto != NULL);
5332 }
5333
5334 return rel;
5335 }
5336
5337 \f
5338 /* Parse operands using Intel syntax. This implements a recursive descent
5339 parser based on the BNF grammar published in Appendix B of the MASM 6.1
5340 Programmer's Guide.
5341
5342 FIXME: We do not recognize the full operand grammar defined in the MASM
5343 documentation. In particular, all the structure/union and
5344 high-level macro operands are missing.
5345
5346 Uppercase words are terminals, lower case words are non-terminals.
5347 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
5348 bars '|' denote choices. Most grammar productions are implemented in
5349 functions called 'intel_<production>'.
5350
5351 Initial production is 'expr'.
5352
5353 addOp + | -
5354
5355 alpha [a-zA-Z]
5356
5357 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
5358
5359 constant digits [[ radixOverride ]]
5360
5361 dataType BYTE | WORD | DWORD | QWORD | XWORD
5362
5363 digits decdigit
5364 | digits decdigit
5365 | digits hexdigit
5366
5367 decdigit [0-9]
5368
5369 e05 e05 addOp e06
5370 | e06
5371
5372 e06 e06 mulOp e09
5373 | e09
5374
5375 e09 OFFSET e10
5376 | e09 PTR e10
5377 | e09 : e10
5378 | e10
5379
5380 e10 e10 [ expr ]
5381 | e11
5382
5383 e11 ( expr )
5384 | [ expr ]
5385 | constant
5386 | dataType
5387 | id
5388 | $
5389 | register
5390
5391 => expr SHORT e05
5392 | e05
5393
5394 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
5395 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
5396
5397 hexdigit a | b | c | d | e | f
5398 | A | B | C | D | E | F
5399
5400 id alpha
5401 | id alpha
5402 | id decdigit
5403
5404 mulOp * | / | MOD
5405
5406 quote " | '
5407
5408 register specialRegister
5409 | gpRegister
5410 | byteRegister
5411
5412 segmentRegister CS | DS | ES | FS | GS | SS
5413
5414 specialRegister CR0 | CR2 | CR3
5415 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
5416 | TR3 | TR4 | TR5 | TR6 | TR7
5417
5418 We simplify the grammar in obvious places (e.g., register parsing is
5419 done by calling parse_register) and eliminate immediate left recursion
5420 to implement a recursive-descent parser.
5421
5422 expr SHORT e05
5423 | e05
5424
5425 e05 e06 e05'
5426
5427 e05' addOp e06 e05'
5428 | Empty
5429
5430 e06 e09 e06'
5431
5432 e06' mulOp e09 e06'
5433 | Empty
5434
5435 e09 OFFSET e10 e09'
5436 | e10 e09'
5437
5438 e09' PTR e10 e09'
5439 | : e10 e09'
5440 | Empty
5441
5442 e10 e11 e10'
5443
5444 e10' [ expr ] e10'
5445 | Empty
5446
5447 e11 ( expr )
5448 | [ expr ]
5449 | BYTE
5450 | WORD
5451 | DWORD
5452 | QWORD
5453 | XWORD
5454 | .
5455 | $
5456 | register
5457 | id
5458 | constant */
5459
5460 /* Parsing structure for the intel syntax parser. Used to implement the
5461 semantic actions for the operand grammar. */
5462 struct intel_parser_s
5463 {
5464 char *op_string; /* The string being parsed. */
5465 int got_a_float; /* Whether the operand is a float. */
5466 int op_modifier; /* Operand modifier. */
5467 int is_mem; /* 1 if operand is memory reference. */
5468 const reg_entry *reg; /* Last register reference found. */
5469 char *disp; /* Displacement string being built. */
5470 };
5471
5472 static struct intel_parser_s intel_parser;
5473
5474 /* Token structure for parsing intel syntax. */
5475 struct intel_token
5476 {
5477 int code; /* Token code. */
5478 const reg_entry *reg; /* Register entry for register tokens. */
5479 char *str; /* String representation. */
5480 };
5481
5482 static struct intel_token cur_token, prev_token;
5483
5484 /* Token codes for the intel parser. Since T_SHORT is already used
5485 by COFF, undefine it first to prevent a warning. */
5486 #define T_NIL -1
5487 #define T_CONST 1
5488 #define T_REG 2
5489 #define T_BYTE 3
5490 #define T_WORD 4
5491 #define T_DWORD 5
5492 #define T_QWORD 6
5493 #define T_XWORD 7
5494 #undef T_SHORT
5495 #define T_SHORT 8
5496 #define T_OFFSET 9
5497 #define T_PTR 10
5498 #define T_ID 11
5499
5500 /* Prototypes for intel parser functions. */
5501 static int intel_match_token PARAMS ((int code));
5502 static void intel_get_token PARAMS ((void));
5503 static void intel_putback_token PARAMS ((void));
5504 static int intel_expr PARAMS ((void));
5505 static int intel_e05 PARAMS ((void));
5506 static int intel_e05_1 PARAMS ((void));
5507 static int intel_e06 PARAMS ((void));
5508 static int intel_e06_1 PARAMS ((void));
5509 static int intel_e09 PARAMS ((void));
5510 static int intel_e09_1 PARAMS ((void));
5511 static int intel_e10 PARAMS ((void));
5512 static int intel_e10_1 PARAMS ((void));
5513 static int intel_e11 PARAMS ((void));
5514
5515 static int
5516 i386_intel_operand (operand_string, got_a_float)
5517 char *operand_string;
5518 int got_a_float;
5519 {
5520 int ret;
5521 char *p;
5522
5523 /* Initialize token holders. */
5524 cur_token.code = prev_token.code = T_NIL;
5525 cur_token.reg = prev_token.reg = NULL;
5526 cur_token.str = prev_token.str = NULL;
5527
5528 /* Initialize parser structure. */
5529 p = intel_parser.op_string = (char *) malloc (strlen (operand_string) + 1);
5530 if (p == NULL)
5531 abort ();
5532 strcpy (intel_parser.op_string, operand_string);
5533 intel_parser.got_a_float = got_a_float;
5534 intel_parser.op_modifier = -1;
5535 intel_parser.is_mem = 0;
5536 intel_parser.reg = NULL;
5537 intel_parser.disp = (char *) malloc (strlen (operand_string) + 1);
5538 if (intel_parser.disp == NULL)
5539 abort ();
5540 intel_parser.disp[0] = '\0';
5541
5542 /* Read the first token and start the parser. */
5543 intel_get_token ();
5544 ret = intel_expr ();
5545
5546 if (ret)
5547 {
5548 /* If we found a memory reference, hand it over to i386_displacement
5549 to fill in the rest of the operand fields. */
5550 if (intel_parser.is_mem)
5551 {
5552 if ((i.mem_operands == 1
5553 && (current_templates->start->opcode_modifier & IsString) == 0)
5554 || i.mem_operands == 2)
5555 {
5556 as_bad (_("too many memory references for '%s'"),
5557 current_templates->start->name);
5558 ret = 0;
5559 }
5560 else
5561 {
5562 char *s = intel_parser.disp;
5563 i.mem_operands++;
5564
5565 /* Add the displacement expression. */
5566 if (*s != '\0')
5567 ret = i386_displacement (s, s + strlen (s));
5568 if (ret)
5569 ret = i386_index_check (operand_string);
5570 }
5571 }
5572
5573 /* Constant and OFFSET expressions are handled by i386_immediate. */
5574 else if (intel_parser.op_modifier == OFFSET_FLAT
5575 || intel_parser.reg == NULL)
5576 ret = i386_immediate (intel_parser.disp);
5577 }
5578
5579 free (p);
5580 free (intel_parser.disp);
5581
5582 return ret;
5583 }
5584
5585 /* expr SHORT e05
5586 | e05 */
5587 static int
5588 intel_expr ()
5589 {
5590 /* expr SHORT e05 */
5591 if (cur_token.code == T_SHORT)
5592 {
5593 intel_parser.op_modifier = SHORT;
5594 intel_match_token (T_SHORT);
5595
5596 return (intel_e05 ());
5597 }
5598
5599 /* expr e05 */
5600 else
5601 return intel_e05 ();
5602 }
5603
5604 /* e05 e06 e05'
5605
5606 e05' addOp e06 e05'
5607 | Empty */
5608 static int
5609 intel_e05 ()
5610 {
5611 return (intel_e06 () && intel_e05_1 ());
5612 }
5613
5614 static int
5615 intel_e05_1 ()
5616 {
5617 /* e05' addOp e06 e05' */
5618 if (cur_token.code == '+' || cur_token.code == '-')
5619 {
5620 strcat (intel_parser.disp, cur_token.str);
5621 intel_match_token (cur_token.code);
5622
5623 return (intel_e06 () && intel_e05_1 ());
5624 }
5625
5626 /* e05' Empty */
5627 else
5628 return 1;
5629 }
5630
5631 /* e06 e09 e06'
5632
5633 e06' mulOp e09 e06'
5634 | Empty */
5635 static int
5636 intel_e06 ()
5637 {
5638 return (intel_e09 () && intel_e06_1 ());
5639 }
5640
5641 static int
5642 intel_e06_1 ()
5643 {
5644 /* e06' mulOp e09 e06' */
5645 if (cur_token.code == '*' || cur_token.code == '/')
5646 {
5647 strcat (intel_parser.disp, cur_token.str);
5648 intel_match_token (cur_token.code);
5649
5650 return (intel_e09 () && intel_e06_1 ());
5651 }
5652
5653 /* e06' Empty */
5654 else
5655 return 1;
5656 }
5657
5658 /* e09 OFFSET e10 e09'
5659 | e10 e09'
5660
5661 e09' PTR e10 e09'
5662 | : e10 e09'
5663 | Empty */
5664 static int
5665 intel_e09 ()
5666 {
5667 /* e09 OFFSET e10 e09' */
5668 if (cur_token.code == T_OFFSET)
5669 {
5670 intel_parser.is_mem = 0;
5671 intel_parser.op_modifier = OFFSET_FLAT;
5672 intel_match_token (T_OFFSET);
5673
5674 return (intel_e10 () && intel_e09_1 ());
5675 }
5676
5677 /* e09 e10 e09' */
5678 else
5679 return (intel_e10 () && intel_e09_1 ());
5680 }
5681
5682 static int
5683 intel_e09_1 ()
5684 {
5685 /* e09' PTR e10 e09' */
5686 if (cur_token.code == T_PTR)
5687 {
5688 if (prev_token.code == T_BYTE)
5689 i.suffix = BYTE_MNEM_SUFFIX;
5690
5691 else if (prev_token.code == T_WORD)
5692 {
5693 if (intel_parser.got_a_float == 2) /* "fi..." */
5694 i.suffix = SHORT_MNEM_SUFFIX;
5695 else
5696 i.suffix = WORD_MNEM_SUFFIX;
5697 }
5698
5699 else if (prev_token.code == T_DWORD)
5700 {
5701 if (intel_parser.got_a_float == 1) /* "f..." */
5702 i.suffix = SHORT_MNEM_SUFFIX;
5703 else
5704 i.suffix = LONG_MNEM_SUFFIX;
5705 }
5706
5707 else if (prev_token.code == T_QWORD)
5708 {
5709 if (intel_parser.got_a_float == 1) /* "f..." */
5710 i.suffix = LONG_MNEM_SUFFIX;
5711 else
5712 i.suffix = QWORD_MNEM_SUFFIX;
5713 }
5714
5715 else if (prev_token.code == T_XWORD)
5716 i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
5717
5718 else
5719 {
5720 as_bad (_("Unknown operand modifier `%s'\n"), prev_token.str);
5721 return 0;
5722 }
5723
5724 intel_match_token (T_PTR);
5725
5726 return (intel_e10 () && intel_e09_1 ());
5727 }
5728
5729 /* e09 : e10 e09' */
5730 else if (cur_token.code == ':')
5731 {
5732 /* Mark as a memory operand only if it's not already known to be an
5733 offset expression. */
5734 if (intel_parser.op_modifier != OFFSET_FLAT)
5735 intel_parser.is_mem = 1;
5736
5737 return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
5738 }
5739
5740 /* e09' Empty */
5741 else
5742 return 1;
5743 }
5744
5745 /* e10 e11 e10'
5746
5747 e10' [ expr ] e10'
5748 | Empty */
5749 static int
5750 intel_e10 ()
5751 {
5752 return (intel_e11 () && intel_e10_1 ());
5753 }
5754
5755 static int
5756 intel_e10_1 ()
5757 {
5758 /* e10' [ expr ] e10' */
5759 if (cur_token.code == '[')
5760 {
5761 intel_match_token ('[');
5762
5763 /* Mark as a memory operand only if it's not already known to be an
5764 offset expression. If it's an offset expression, we need to keep
5765 the brace in. */
5766 if (intel_parser.op_modifier != OFFSET_FLAT)
5767 intel_parser.is_mem = 1;
5768 else
5769 strcat (intel_parser.disp, "[");
5770
5771 /* Add a '+' to the displacement string if necessary. */
5772 if (*intel_parser.disp != '\0'
5773 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5774 strcat (intel_parser.disp, "+");
5775
5776 if (intel_expr () && intel_match_token (']'))
5777 {
5778 /* Preserve brackets when the operand is an offset expression. */
5779 if (intel_parser.op_modifier == OFFSET_FLAT)
5780 strcat (intel_parser.disp, "]");
5781
5782 return intel_e10_1 ();
5783 }
5784 else
5785 return 0;
5786 }
5787
5788 /* e10' Empty */
5789 else
5790 return 1;
5791 }
5792
5793 /* e11 ( expr )
5794 | [ expr ]
5795 | BYTE
5796 | WORD
5797 | DWORD
5798 | QWORD
5799 | XWORD
5800 | $
5801 | .
5802 | register
5803 | id
5804 | constant */
5805 static int
5806 intel_e11 ()
5807 {
5808 /* e11 ( expr ) */
5809 if (cur_token.code == '(')
5810 {
5811 intel_match_token ('(');
5812 strcat (intel_parser.disp, "(");
5813
5814 if (intel_expr () && intel_match_token (')'))
5815 {
5816 strcat (intel_parser.disp, ")");
5817 return 1;
5818 }
5819 else
5820 return 0;
5821 }
5822
5823 /* e11 [ expr ] */
5824 else if (cur_token.code == '[')
5825 {
5826 intel_match_token ('[');
5827
5828 /* Mark as a memory operand only if it's not already known to be an
5829 offset expression. If it's an offset expression, we need to keep
5830 the brace in. */
5831 if (intel_parser.op_modifier != OFFSET_FLAT)
5832 intel_parser.is_mem = 1;
5833 else
5834 strcat (intel_parser.disp, "[");
5835
5836 /* Operands for jump/call inside brackets denote absolute addresses. */
5837 if (current_templates->start->opcode_modifier & Jump
5838 || current_templates->start->opcode_modifier & JumpDword
5839 || current_templates->start->opcode_modifier & JumpByte
5840 || current_templates->start->opcode_modifier & JumpInterSegment)
5841 i.types[this_operand] |= JumpAbsolute;
5842
5843 /* Add a '+' to the displacement string if necessary. */
5844 if (*intel_parser.disp != '\0'
5845 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5846 strcat (intel_parser.disp, "+");
5847
5848 if (intel_expr () && intel_match_token (']'))
5849 {
5850 /* Preserve brackets when the operand is an offset expression. */
5851 if (intel_parser.op_modifier == OFFSET_FLAT)
5852 strcat (intel_parser.disp, "]");
5853
5854 return 1;
5855 }
5856 else
5857 return 0;
5858 }
5859
5860 /* e11 BYTE
5861 | WORD
5862 | DWORD
5863 | QWORD
5864 | XWORD */
5865 else if (cur_token.code == T_BYTE
5866 || cur_token.code == T_WORD
5867 || cur_token.code == T_DWORD
5868 || cur_token.code == T_QWORD
5869 || cur_token.code == T_XWORD)
5870 {
5871 intel_match_token (cur_token.code);
5872
5873 return 1;
5874 }
5875
5876 /* e11 $
5877 | . */
5878 else if (cur_token.code == '$' || cur_token.code == '.')
5879 {
5880 strcat (intel_parser.disp, cur_token.str);
5881 intel_match_token (cur_token.code);
5882
5883 /* Mark as a memory operand only if it's not already known to be an
5884 offset expression. */
5885 if (intel_parser.op_modifier != OFFSET_FLAT)
5886 intel_parser.is_mem = 1;
5887
5888 return 1;
5889 }
5890
5891 /* e11 register */
5892 else if (cur_token.code == T_REG)
5893 {
5894 const reg_entry *reg = intel_parser.reg = cur_token.reg;
5895
5896 intel_match_token (T_REG);
5897
5898 /* Check for segment change. */
5899 if (cur_token.code == ':')
5900 {
5901 if (reg->reg_type & (SReg2 | SReg3))
5902 {
5903 switch (reg->reg_num)
5904 {
5905 case 0:
5906 i.seg[i.mem_operands] = &es;
5907 break;
5908 case 1:
5909 i.seg[i.mem_operands] = &cs;
5910 break;
5911 case 2:
5912 i.seg[i.mem_operands] = &ss;
5913 break;
5914 case 3:
5915 i.seg[i.mem_operands] = &ds;
5916 break;
5917 case 4:
5918 i.seg[i.mem_operands] = &fs;
5919 break;
5920 case 5:
5921 i.seg[i.mem_operands] = &gs;
5922 break;
5923 }
5924 }
5925 else
5926 {
5927 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
5928 return 0;
5929 }
5930 }
5931
5932 /* Not a segment register. Check for register scaling. */
5933 else if (cur_token.code == '*')
5934 {
5935 if (!intel_parser.is_mem)
5936 {
5937 as_bad (_("Register scaling only allowed in memory operands."));
5938 return 0;
5939 }
5940
5941 /* What follows must be a valid scale. */
5942 if (intel_match_token ('*')
5943 && strchr ("01248", *cur_token.str))
5944 {
5945 i.index_reg = reg;
5946 i.types[this_operand] |= BaseIndex;
5947
5948 /* Set the scale after setting the register (otherwise,
5949 i386_scale will complain) */
5950 i386_scale (cur_token.str);
5951 intel_match_token (T_CONST);
5952 }
5953 else
5954 {
5955 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
5956 cur_token.str);
5957 return 0;
5958 }
5959 }
5960
5961 /* No scaling. If this is a memory operand, the register is either a
5962 base register (first occurrence) or an index register (second
5963 occurrence). */
5964 else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3)))
5965 {
5966 if (i.base_reg && i.index_reg)
5967 {
5968 as_bad (_("Too many register references in memory operand.\n"));
5969 return 0;
5970 }
5971
5972 if (i.base_reg == NULL)
5973 i.base_reg = reg;
5974 else
5975 i.index_reg = reg;
5976
5977 i.types[this_operand] |= BaseIndex;
5978 }
5979
5980 /* Offset modifier. Add the register to the displacement string to be
5981 parsed as an immediate expression after we're done. */
5982 else if (intel_parser.op_modifier == OFFSET_FLAT)
5983 strcat (intel_parser.disp, reg->reg_name);
5984
5985 /* It's neither base nor index nor offset. */
5986 else
5987 {
5988 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
5989 i.op[this_operand].regs = reg;
5990 i.reg_operands++;
5991 }
5992
5993 /* Since registers are not part of the displacement string (except
5994 when we're parsing offset operands), we may need to remove any
5995 preceding '+' from the displacement string. */
5996 if (*intel_parser.disp != '\0'
5997 && intel_parser.op_modifier != OFFSET_FLAT)
5998 {
5999 char *s = intel_parser.disp;
6000 s += strlen (s) - 1;
6001 if (*s == '+')
6002 *s = '\0';
6003 }
6004
6005 return 1;
6006 }
6007
6008 /* e11 id */
6009 else if (cur_token.code == T_ID)
6010 {
6011 /* Add the identifier to the displacement string. */
6012 strcat (intel_parser.disp, cur_token.str);
6013 intel_match_token (T_ID);
6014
6015 /* The identifier represents a memory reference only if it's not
6016 preceded by an offset modifier. */
6017 if (intel_parser.op_modifier != OFFSET_FLAT)
6018 intel_parser.is_mem = 1;
6019
6020 return 1;
6021 }
6022
6023 /* e11 constant */
6024 else if (cur_token.code == T_CONST
6025 || cur_token.code == '-'
6026 || cur_token.code == '+')
6027 {
6028 char *save_str;
6029
6030 /* Allow constants that start with `+' or `-'. */
6031 if (cur_token.code == '-' || cur_token.code == '+')
6032 {
6033 strcat (intel_parser.disp, cur_token.str);
6034 intel_match_token (cur_token.code);
6035 if (cur_token.code != T_CONST)
6036 {
6037 as_bad (_("Syntax error. Expecting a constant. Got `%s'.\n"),
6038 cur_token.str);
6039 return 0;
6040 }
6041 }
6042
6043 save_str = (char *) malloc (strlen (cur_token.str) + 1);
6044 if (save_str == NULL)
6045 abort ();
6046 strcpy (save_str, cur_token.str);
6047
6048 /* Get the next token to check for register scaling. */
6049 intel_match_token (cur_token.code);
6050
6051 /* Check if this constant is a scaling factor for an index register. */
6052 if (cur_token.code == '*')
6053 {
6054 if (intel_match_token ('*') && cur_token.code == T_REG)
6055 {
6056 if (!intel_parser.is_mem)
6057 {
6058 as_bad (_("Register scaling only allowed in memory operands."));
6059 return 0;
6060 }
6061
6062 /* The constant is followed by `* reg', so it must be
6063 a valid scale. */
6064 if (strchr ("01248", *save_str))
6065 {
6066 i.index_reg = cur_token.reg;
6067 i.types[this_operand] |= BaseIndex;
6068
6069 /* Set the scale after setting the register (otherwise,
6070 i386_scale will complain) */
6071 i386_scale (save_str);
6072 intel_match_token (T_REG);
6073
6074 /* Since registers are not part of the displacement
6075 string, we may need to remove any preceding '+' from
6076 the displacement string. */
6077 if (*intel_parser.disp != '\0')
6078 {
6079 char *s = intel_parser.disp;
6080 s += strlen (s) - 1;
6081 if (*s == '+')
6082 *s = '\0';
6083 }
6084
6085 free (save_str);
6086
6087 return 1;
6088 }
6089 else
6090 return 0;
6091 }
6092
6093 /* The constant was not used for register scaling. Since we have
6094 already consumed the token following `*' we now need to put it
6095 back in the stream. */
6096 else
6097 intel_putback_token ();
6098 }
6099
6100 /* Add the constant to the displacement string. */
6101 strcat (intel_parser.disp, save_str);
6102 free (save_str);
6103
6104 return 1;
6105 }
6106
6107 as_bad (_("Unrecognized token '%s'"), cur_token.str);
6108 return 0;
6109 }
6110
6111 /* Match the given token against cur_token. If they match, read the next
6112 token from the operand string. */
6113 static int
6114 intel_match_token (code)
6115 int code;
6116 {
6117 if (cur_token.code == code)
6118 {
6119 intel_get_token ();
6120 return 1;
6121 }
6122 else
6123 {
6124 as_bad (_("Unexpected token `%s'\n"), cur_token.str);
6125 return 0;
6126 }
6127 }
6128
6129 /* Read a new token from intel_parser.op_string and store it in cur_token. */
6130 static void
6131 intel_get_token ()
6132 {
6133 char *end_op;
6134 const reg_entry *reg;
6135 struct intel_token new_token;
6136
6137 new_token.code = T_NIL;
6138 new_token.reg = NULL;
6139 new_token.str = NULL;
6140
6141 /* Free the memory allocated to the previous token and move
6142 cur_token to prev_token. */
6143 if (prev_token.str)
6144 free (prev_token.str);
6145
6146 prev_token = cur_token;
6147
6148 /* Skip whitespace. */
6149 while (is_space_char (*intel_parser.op_string))
6150 intel_parser.op_string++;
6151
6152 /* Return an empty token if we find nothing else on the line. */
6153 if (*intel_parser.op_string == '\0')
6154 {
6155 cur_token = new_token;
6156 return;
6157 }
6158
6159 /* The new token cannot be larger than the remainder of the operand
6160 string. */
6161 new_token.str = (char *) malloc (strlen (intel_parser.op_string) + 1);
6162 if (new_token.str == NULL)
6163 abort ();
6164 new_token.str[0] = '\0';
6165
6166 if (strchr ("0123456789", *intel_parser.op_string))
6167 {
6168 char *p = new_token.str;
6169 char *q = intel_parser.op_string;
6170 new_token.code = T_CONST;
6171
6172 /* Allow any kind of identifier char to encompass floating point and
6173 hexadecimal numbers. */
6174 while (is_identifier_char (*q))
6175 *p++ = *q++;
6176 *p = '\0';
6177
6178 /* Recognize special symbol names [0-9][bf]. */
6179 if (strlen (intel_parser.op_string) == 2
6180 && (intel_parser.op_string[1] == 'b'
6181 || intel_parser.op_string[1] == 'f'))
6182 new_token.code = T_ID;
6183 }
6184
6185 else if (strchr ("+-/*:[]()", *intel_parser.op_string))
6186 {
6187 new_token.code = *intel_parser.op_string;
6188 new_token.str[0] = *intel_parser.op_string;
6189 new_token.str[1] = '\0';
6190 }
6191
6192 else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
6193 && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
6194 {
6195 new_token.code = T_REG;
6196 new_token.reg = reg;
6197
6198 if (*intel_parser.op_string == REGISTER_PREFIX)
6199 {
6200 new_token.str[0] = REGISTER_PREFIX;
6201 new_token.str[1] = '\0';
6202 }
6203
6204 strcat (new_token.str, reg->reg_name);
6205 }
6206
6207 else if (is_identifier_char (*intel_parser.op_string))
6208 {
6209 char *p = new_token.str;
6210 char *q = intel_parser.op_string;
6211
6212 /* A '.' or '$' followed by an identifier char is an identifier.
6213 Otherwise, it's operator '.' followed by an expression. */
6214 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
6215 {
6216 new_token.code = *q;
6217 new_token.str[0] = *q;
6218 new_token.str[1] = '\0';
6219 }
6220 else
6221 {
6222 while (is_identifier_char (*q) || *q == '@')
6223 *p++ = *q++;
6224 *p = '\0';
6225
6226 if (strcasecmp (new_token.str, "BYTE") == 0)
6227 new_token.code = T_BYTE;
6228
6229 else if (strcasecmp (new_token.str, "WORD") == 0)
6230 new_token.code = T_WORD;
6231
6232 else if (strcasecmp (new_token.str, "DWORD") == 0)
6233 new_token.code = T_DWORD;
6234
6235 else if (strcasecmp (new_token.str, "QWORD") == 0)
6236 new_token.code = T_QWORD;
6237
6238 else if (strcasecmp (new_token.str, "XWORD") == 0)
6239 new_token.code = T_XWORD;
6240
6241 else if (strcasecmp (new_token.str, "PTR") == 0)
6242 new_token.code = T_PTR;
6243
6244 else if (strcasecmp (new_token.str, "SHORT") == 0)
6245 new_token.code = T_SHORT;
6246
6247 else if (strcasecmp (new_token.str, "OFFSET") == 0)
6248 {
6249 new_token.code = T_OFFSET;
6250
6251 /* ??? This is not mentioned in the MASM grammar but gcc
6252 makes use of it with -mintel-syntax. OFFSET may be
6253 followed by FLAT: */
6254 if (strncasecmp (q, " FLAT:", 6) == 0)
6255 strcat (new_token.str, " FLAT:");
6256 }
6257
6258 /* ??? This is not mentioned in the MASM grammar. */
6259 else if (strcasecmp (new_token.str, "FLAT") == 0)
6260 new_token.code = T_OFFSET;
6261
6262 else
6263 new_token.code = T_ID;
6264 }
6265 }
6266
6267 else
6268 as_bad (_("Unrecognized token `%s'\n"), intel_parser.op_string);
6269
6270 intel_parser.op_string += strlen (new_token.str);
6271 cur_token = new_token;
6272 }
6273
6274 /* Put cur_token back into the token stream and make cur_token point to
6275 prev_token. */
6276 static void
6277 intel_putback_token ()
6278 {
6279 intel_parser.op_string -= strlen (cur_token.str);
6280 free (cur_token.str);
6281 cur_token = prev_token;
6282
6283 /* Forget prev_token. */
6284 prev_token.code = T_NIL;
6285 prev_token.reg = NULL;
6286 prev_token.str = NULL;
6287 }
6288
6289 int
6290 tc_x86_regname_to_dw2regnum (const char *regname)
6291 {
6292 unsigned int regnum;
6293 unsigned int regnames_count;
6294 char *regnames_32[] =
6295 {
6296 "eax", "ecx", "edx", "ebx",
6297 "esp", "ebp", "esi", "edi",
6298 "eip"
6299 };
6300 char *regnames_64[] =
6301 {
6302 "rax", "rbx", "rcx", "rdx",
6303 "rdi", "rsi", "rbp", "rsp",
6304 "r8", "r9", "r10", "r11",
6305 "r12", "r13", "r14", "r15",
6306 "rip"
6307 };
6308 char **regnames;
6309
6310 if (flag_code == CODE_64BIT)
6311 {
6312 regnames = regnames_64;
6313 regnames_count = ARRAY_SIZE (regnames_64);
6314 }
6315 else
6316 {
6317 regnames = regnames_32;
6318 regnames_count = ARRAY_SIZE (regnames_32);
6319 }
6320
6321 for (regnum = 0; regnum < regnames_count; regnum++)
6322 if (strcmp (regname, regnames[regnum]) == 0)
6323 return regnum;
6324
6325 return -1;
6326 }
6327
6328 void
6329 tc_x86_frame_initial_instructions (void)
6330 {
6331 static unsigned int sp_regno;
6332
6333 if (!sp_regno)
6334 sp_regno = tc_x86_regname_to_dw2regnum (flag_code == CODE_64BIT
6335 ? "rsp" : "esp");
6336
6337 cfi_add_CFA_def_cfa (sp_regno, -x86_cie_data_alignment);
6338 cfi_add_CFA_offset (x86_dwarf2_return_column, x86_cie_data_alignment);
6339 }
This page took 0.1611 seconds and 4 git commands to generate.