* config/tc-i386.h (REX_OPCODE): Define.
[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
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 Bugs & suggestions are completely welcome. This is free software.
27 Please help us make it better. */
28
29 #include "as.h"
30 #include "safe-ctype.h"
31 #include "subsegs.h"
32 #include "dwarf2dbg.h"
33 #include "opcode/i386.h"
34
35 #ifndef REGISTER_WARNINGS
36 #define REGISTER_WARNINGS 1
37 #endif
38
39 #ifndef INFER_ADDR_PREFIX
40 #define INFER_ADDR_PREFIX 1
41 #endif
42
43 #ifndef SCALE1_WHEN_NO_INDEX
44 /* Specifying a scale factor besides 1 when there is no index is
45 futile. eg. `mov (%ebx,2),%al' does exactly the same as
46 `mov (%ebx),%al'. To slavishly follow what the programmer
47 specified, set SCALE1_WHEN_NO_INDEX to 0. */
48 #define SCALE1_WHEN_NO_INDEX 1
49 #endif
50
51 #ifdef BFD_ASSEMBLER
52 #define RELOC_ENUM enum bfd_reloc_code_real
53 #else
54 #define RELOC_ENUM int
55 #endif
56
57 #ifndef DEFAULT_ARCH
58 #define DEFAULT_ARCH "i386"
59 #endif
60
61 static INLINE unsigned int mode_from_disp_size PARAMS ((unsigned int));
62 static INLINE int fits_in_signed_byte PARAMS ((offsetT));
63 static INLINE int fits_in_unsigned_byte PARAMS ((offsetT));
64 static INLINE int fits_in_unsigned_word PARAMS ((offsetT));
65 static INLINE int fits_in_signed_word PARAMS ((offsetT));
66 static INLINE int fits_in_unsigned_long PARAMS ((offsetT));
67 static INLINE int fits_in_signed_long PARAMS ((offsetT));
68 static int smallest_imm_type PARAMS ((offsetT));
69 static offsetT offset_in_range PARAMS ((offsetT, int));
70 static int add_prefix PARAMS ((unsigned int));
71 static void set_code_flag PARAMS ((int));
72 static void set_16bit_gcc_code_flag PARAMS ((int));
73 static void set_intel_syntax PARAMS ((int));
74 static void set_cpu_arch PARAMS ((int));
75 static char *output_invalid PARAMS ((int c));
76 static int i386_operand PARAMS ((char *operand_string));
77 static int i386_intel_operand PARAMS ((char *operand_string, int got_a_float));
78 static const reg_entry *parse_register PARAMS ((char *reg_string,
79 char **end_op));
80 static char *parse_insn PARAMS ((char *, char *));
81 static char *parse_operands PARAMS ((char *, const char *));
82 static void swap_operands PARAMS ((void));
83 static void optimize_imm PARAMS ((void));
84 static void optimize_disp PARAMS ((void));
85 static int match_template PARAMS ((void));
86 static int check_string PARAMS ((void));
87 static int process_suffix PARAMS ((void));
88 static int check_byte_reg PARAMS ((void));
89 static int check_long_reg PARAMS ((void));
90 static int check_qword_reg PARAMS ((void));
91 static int check_word_reg PARAMS ((void));
92 static int finalize_imm PARAMS ((void));
93 static int process_operands PARAMS ((void));
94 static const seg_entry *build_modrm_byte PARAMS ((void));
95 static void output_insn PARAMS ((void));
96 static void output_branch PARAMS ((void));
97 static void output_jump PARAMS ((void));
98 static void output_interseg_jump PARAMS ((void));
99 static void output_imm PARAMS ((void));
100 static void output_disp PARAMS ((void));
101 #ifndef I386COFF
102 static void s_bss PARAMS ((int));
103 #endif
104
105 static const char *default_arch = DEFAULT_ARCH;
106
107 /* 'md_assemble ()' gathers together information and puts it into a
108 i386_insn. */
109
110 union i386_op
111 {
112 expressionS *disps;
113 expressionS *imms;
114 const reg_entry *regs;
115 };
116
117 struct _i386_insn
118 {
119 /* TM holds the template for the insn were currently assembling. */
120 template tm;
121
122 /* SUFFIX holds the instruction mnemonic suffix if given.
123 (e.g. 'l' for 'movl') */
124 char suffix;
125
126 /* OPERANDS gives the number of given operands. */
127 unsigned int operands;
128
129 /* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number
130 of given register, displacement, memory operands and immediate
131 operands. */
132 unsigned int reg_operands, disp_operands, mem_operands, imm_operands;
133
134 /* TYPES [i] is the type (see above #defines) which tells us how to
135 use OP[i] for the corresponding operand. */
136 unsigned int types[MAX_OPERANDS];
137
138 /* Displacement expression, immediate expression, or register for each
139 operand. */
140 union i386_op op[MAX_OPERANDS];
141
142 /* Flags for operands. */
143 unsigned int flags[MAX_OPERANDS];
144 #define Operand_PCrel 1
145
146 /* Relocation type for operand */
147 RELOC_ENUM reloc[MAX_OPERANDS];
148
149 /* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode
150 the base index byte below. */
151 const reg_entry *base_reg;
152 const reg_entry *index_reg;
153 unsigned int log2_scale_factor;
154
155 /* SEG gives the seg_entries of this insn. They are zero unless
156 explicit segment overrides are given. */
157 const seg_entry *seg[2];
158
159 /* PREFIX holds all the given prefix opcodes (usually null).
160 PREFIXES is the number of prefix opcodes. */
161 unsigned int prefixes;
162 unsigned char prefix[MAX_PREFIXES];
163
164 /* RM and SIB are the modrm byte and the sib byte where the
165 addressing modes of this insn are encoded. */
166
167 modrm_byte rm;
168 rex_byte rex;
169 sib_byte sib;
170 };
171
172 typedef struct _i386_insn i386_insn;
173
174 /* List of chars besides those in app.c:symbol_chars that can start an
175 operand. Used to prevent the scrubber eating vital white-space. */
176 #ifdef LEX_AT
177 const char extra_symbol_chars[] = "*%-(@";
178 #else
179 const char extra_symbol_chars[] = "*%-(";
180 #endif
181
182 #if (defined (TE_I386AIX) \
183 || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) \
184 && !defined (TE_LINUX) \
185 && !defined (TE_FreeBSD) \
186 && !defined (TE_NetBSD)))
187 /* This array holds the chars that always start a comment. If the
188 pre-processor is disabled, these aren't very useful. */
189 const char comment_chars[] = "#/";
190 #define PREFIX_SEPARATOR '\\'
191
192 /* This array holds the chars that only start a comment at the beginning of
193 a line. If the line seems to have the form '# 123 filename'
194 .line and .file directives will appear in the pre-processed output.
195 Note that input_file.c hand checks for '#' at the beginning of the
196 first line of the input file. This is because the compiler outputs
197 #NO_APP at the beginning of its output.
198 Also note that comments started like this one will always work if
199 '/' isn't otherwise defined. */
200 const char line_comment_chars[] = "";
201
202 #else
203 /* Putting '/' here makes it impossible to use the divide operator.
204 However, we need it for compatibility with SVR4 systems. */
205 const char comment_chars[] = "#";
206 #define PREFIX_SEPARATOR '/'
207
208 const char line_comment_chars[] = "/";
209 #endif
210
211 const char line_separator_chars[] = ";";
212
213 /* Chars that can be used to separate mant from exp in floating point
214 nums. */
215 const char EXP_CHARS[] = "eE";
216
217 /* Chars that mean this number is a floating point constant
218 As in 0f12.456
219 or 0d1.2345e12. */
220 const char FLT_CHARS[] = "fFdDxX";
221
222 /* Tables for lexical analysis. */
223 static char mnemonic_chars[256];
224 static char register_chars[256];
225 static char operand_chars[256];
226 static char identifier_chars[256];
227 static char digit_chars[256];
228
229 /* Lexical macros. */
230 #define is_mnemonic_char(x) (mnemonic_chars[(unsigned char) x])
231 #define is_operand_char(x) (operand_chars[(unsigned char) x])
232 #define is_register_char(x) (register_chars[(unsigned char) x])
233 #define is_space_char(x) ((x) == ' ')
234 #define is_identifier_char(x) (identifier_chars[(unsigned char) x])
235 #define is_digit_char(x) (digit_chars[(unsigned char) x])
236
237 /* All non-digit non-letter charcters that may occur in an operand. */
238 static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
239
240 /* md_assemble() always leaves the strings it's passed unaltered. To
241 effect this we maintain a stack of saved characters that we've smashed
242 with '\0's (indicating end of strings for various sub-fields of the
243 assembler instruction). */
244 static char save_stack[32];
245 static char *save_stack_p;
246 #define END_STRING_AND_SAVE(s) \
247 do { *save_stack_p++ = *(s); *(s) = '\0'; } while (0)
248 #define RESTORE_END_STRING(s) \
249 do { *(s) = *--save_stack_p; } while (0)
250
251 /* The instruction we're assembling. */
252 static i386_insn i;
253
254 /* Possible templates for current insn. */
255 static const templates *current_templates;
256
257 /* Per instruction expressionS buffers: 2 displacements & 2 immediate max. */
258 static expressionS disp_expressions[2], im_expressions[2];
259
260 /* Current operand we are working on. */
261 static int this_operand;
262
263 /* We support four different modes. FLAG_CODE variable is used to distinguish
264 these. */
265
266 enum flag_code {
267 CODE_32BIT,
268 CODE_16BIT,
269 CODE_64BIT };
270 #define NUM_FLAG_CODE ((int) CODE_64BIT + 1)
271
272 static enum flag_code flag_code;
273 static int use_rela_relocations = 0;
274
275 /* The names used to print error messages. */
276 static const char *flag_code_names[] =
277 {
278 "32",
279 "16",
280 "64"
281 };
282
283 /* 1 for intel syntax,
284 0 if att syntax. */
285 static int intel_syntax = 0;
286
287 /* 1 if register prefix % not required. */
288 static int allow_naked_reg = 0;
289
290 /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
291 leave, push, and pop instructions so that gcc has the same stack
292 frame as in 32 bit mode. */
293 static char stackop_size = '\0';
294
295 /* Non-zero to quieten some warnings. */
296 static int quiet_warnings = 0;
297
298 /* CPU name. */
299 static const char *cpu_arch_name = NULL;
300
301 /* CPU feature flags. */
302 static unsigned int cpu_arch_flags = CpuUnknownFlags | CpuNo64;
303
304 /* If set, conditional jumps are not automatically promoted to handle
305 larger than a byte offset. */
306 static unsigned int no_cond_jump_promotion = 0;
307
308 /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
309 symbolS *GOT_symbol;
310
311 /* Interface to relax_segment.
312 There are 3 major relax states for 386 jump insns because the
313 different types of jumps add different sizes to frags when we're
314 figuring out what sort of jump to choose to reach a given label. */
315
316 /* Types. */
317 #define UNCOND_JUMP 0
318 #define COND_JUMP 1
319 #define COND_JUMP86 2
320
321 /* Sizes. */
322 #define CODE16 1
323 #define SMALL 0
324 #define SMALL16 (SMALL | CODE16)
325 #define BIG 2
326 #define BIG16 (BIG | CODE16)
327
328 #ifndef INLINE
329 #ifdef __GNUC__
330 #define INLINE __inline__
331 #else
332 #define INLINE
333 #endif
334 #endif
335
336 #define ENCODE_RELAX_STATE(type, size) \
337 ((relax_substateT) (((type) << 2) | (size)))
338 #define TYPE_FROM_RELAX_STATE(s) \
339 ((s) >> 2)
340 #define DISP_SIZE_FROM_RELAX_STATE(s) \
341 ((((s) & 3) == BIG ? 4 : (((s) & 3) == BIG16 ? 2 : 1)))
342
343 /* This table is used by relax_frag to promote short jumps to long
344 ones where necessary. SMALL (short) jumps may be promoted to BIG
345 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
346 don't allow a short jump in a 32 bit code segment to be promoted to
347 a 16 bit offset jump because it's slower (requires data size
348 prefix), and doesn't work, unless the destination is in the bottom
349 64k of the code segment (The top 16 bits of eip are zeroed). */
350
351 const relax_typeS md_relax_table[] =
352 {
353 /* The fields are:
354 1) most positive reach of this state,
355 2) most negative reach of this state,
356 3) how many bytes this mode will have in the variable part of the frag
357 4) which index into the table to try if we can't fit into this one. */
358
359 /* UNCOND_JUMP states. */
360 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
361 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
362 /* dword jmp adds 4 bytes to frag:
363 0 extra opcode bytes, 4 displacement bytes. */
364 {0, 0, 4, 0},
365 /* word jmp adds 2 byte2 to frag:
366 0 extra opcode bytes, 2 displacement bytes. */
367 {0, 0, 2, 0},
368
369 /* COND_JUMP states. */
370 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
371 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
372 /* dword conditionals adds 5 bytes to frag:
373 1 extra opcode byte, 4 displacement bytes. */
374 {0, 0, 5, 0},
375 /* word conditionals add 3 bytes to frag:
376 1 extra opcode byte, 2 displacement bytes. */
377 {0, 0, 3, 0},
378
379 /* COND_JUMP86 states. */
380 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG)},
381 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG16)},
382 /* dword conditionals adds 5 bytes to frag:
383 1 extra opcode byte, 4 displacement bytes. */
384 {0, 0, 5, 0},
385 /* word conditionals add 4 bytes to frag:
386 1 displacement byte and a 3 byte long branch insn. */
387 {0, 0, 4, 0}
388 };
389
390 static const arch_entry cpu_arch[] = {
391 {"i8086", Cpu086 },
392 {"i186", Cpu086|Cpu186 },
393 {"i286", Cpu086|Cpu186|Cpu286 },
394 {"i386", Cpu086|Cpu186|Cpu286|Cpu386 },
395 {"i486", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486 },
396 {"i586", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
397 {"i686", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
398 {"pentium", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
399 {"pentiumpro",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
400 {"pentium4", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX|CpuSSE|CpuSSE2 },
401 {"k6", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow },
402 {"athlon", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuMMX|Cpu3dnow },
403 {"sledgehammer",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuSledgehammer|CpuMMX|Cpu3dnow|CpuSSE|CpuSSE2 },
404 {NULL, 0 }
405 };
406
407 const pseudo_typeS md_pseudo_table[] =
408 {
409 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
410 {"align", s_align_bytes, 0},
411 #else
412 {"align", s_align_ptwo, 0},
413 #endif
414 {"arch", set_cpu_arch, 0},
415 #ifndef I386COFF
416 {"bss", s_bss, 0},
417 #endif
418 {"ffloat", float_cons, 'f'},
419 {"dfloat", float_cons, 'd'},
420 {"tfloat", float_cons, 'x'},
421 {"value", cons, 2},
422 {"noopt", s_ignore, 0},
423 {"optim", s_ignore, 0},
424 {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
425 {"code16", set_code_flag, CODE_16BIT},
426 {"code32", set_code_flag, CODE_32BIT},
427 {"code64", set_code_flag, CODE_64BIT},
428 {"intel_syntax", set_intel_syntax, 1},
429 {"att_syntax", set_intel_syntax, 0},
430 {"file", dwarf2_directive_file, 0},
431 {"loc", dwarf2_directive_loc, 0},
432 {0, 0, 0}
433 };
434
435 /* For interface with expression (). */
436 extern char *input_line_pointer;
437
438 /* Hash table for instruction mnemonic lookup. */
439 static struct hash_control *op_hash;
440
441 /* Hash table for register lookup. */
442 static struct hash_control *reg_hash;
443 \f
444 void
445 i386_align_code (fragP, count)
446 fragS *fragP;
447 int count;
448 {
449 /* Various efficient no-op patterns for aligning code labels.
450 Note: Don't try to assemble the instructions in the comments.
451 0L and 0w are not legal. */
452 static const char f32_1[] =
453 {0x90}; /* nop */
454 static const char f32_2[] =
455 {0x89,0xf6}; /* movl %esi,%esi */
456 static const char f32_3[] =
457 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
458 static const char f32_4[] =
459 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
460 static const char f32_5[] =
461 {0x90, /* nop */
462 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
463 static const char f32_6[] =
464 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
465 static const char f32_7[] =
466 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
467 static const char f32_8[] =
468 {0x90, /* nop */
469 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
470 static const char f32_9[] =
471 {0x89,0xf6, /* movl %esi,%esi */
472 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
473 static const char f32_10[] =
474 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
475 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
476 static const char f32_11[] =
477 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
478 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
479 static const char f32_12[] =
480 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
481 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
482 static const char f32_13[] =
483 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
484 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
485 static const char f32_14[] =
486 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
487 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
488 static const char f32_15[] =
489 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
490 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
491 static const char f16_3[] =
492 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
493 static const char f16_4[] =
494 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
495 static const char f16_5[] =
496 {0x90, /* nop */
497 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
498 static const char f16_6[] =
499 {0x89,0xf6, /* mov %si,%si */
500 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
501 static const char f16_7[] =
502 {0x8d,0x74,0x00, /* lea 0(%si),%si */
503 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
504 static const char f16_8[] =
505 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
506 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
507 static const char *const f32_patt[] = {
508 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
509 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
510 };
511 static const char *const f16_patt[] = {
512 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
513 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
514 };
515
516 /* ??? We can't use these fillers for x86_64, since they often kills the
517 upper halves. Solve later. */
518 if (flag_code == CODE_64BIT)
519 count = 1;
520
521 if (count > 0 && count <= 15)
522 {
523 if (flag_code == CODE_16BIT)
524 {
525 memcpy (fragP->fr_literal + fragP->fr_fix,
526 f16_patt[count - 1], count);
527 if (count > 8)
528 /* Adjust jump offset. */
529 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
530 }
531 else
532 memcpy (fragP->fr_literal + fragP->fr_fix,
533 f32_patt[count - 1], count);
534 fragP->fr_var = count;
535 }
536 }
537
538 static INLINE unsigned int
539 mode_from_disp_size (t)
540 unsigned int t;
541 {
542 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
543 }
544
545 static INLINE int
546 fits_in_signed_byte (num)
547 offsetT num;
548 {
549 return (num >= -128) && (num <= 127);
550 }
551
552 static INLINE int
553 fits_in_unsigned_byte (num)
554 offsetT num;
555 {
556 return (num & 0xff) == num;
557 }
558
559 static INLINE int
560 fits_in_unsigned_word (num)
561 offsetT num;
562 {
563 return (num & 0xffff) == num;
564 }
565
566 static INLINE int
567 fits_in_signed_word (num)
568 offsetT num;
569 {
570 return (-32768 <= num) && (num <= 32767);
571 }
572 static INLINE int
573 fits_in_signed_long (num)
574 offsetT num ATTRIBUTE_UNUSED;
575 {
576 #ifndef BFD64
577 return 1;
578 #else
579 return (!(((offsetT) -1 << 31) & num)
580 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
581 #endif
582 } /* fits_in_signed_long() */
583 static INLINE int
584 fits_in_unsigned_long (num)
585 offsetT num ATTRIBUTE_UNUSED;
586 {
587 #ifndef BFD64
588 return 1;
589 #else
590 return (num & (((offsetT) 2 << 31) - 1)) == num;
591 #endif
592 } /* fits_in_unsigned_long() */
593
594 static int
595 smallest_imm_type (num)
596 offsetT num;
597 {
598 if (cpu_arch_flags != (Cpu086 | Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64))
599 {
600 /* This code is disabled on the 486 because all the Imm1 forms
601 in the opcode table are slower on the i486. They're the
602 versions with the implicitly specified single-position
603 displacement, which has another syntax if you really want to
604 use that form. */
605 if (num == 1)
606 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
607 }
608 return (fits_in_signed_byte (num)
609 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
610 : fits_in_unsigned_byte (num)
611 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
612 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
613 ? (Imm16 | Imm32 | Imm32S | Imm64)
614 : fits_in_signed_long (num)
615 ? (Imm32 | Imm32S | Imm64)
616 : fits_in_unsigned_long (num)
617 ? (Imm32 | Imm64)
618 : Imm64);
619 }
620
621 static offsetT
622 offset_in_range (val, size)
623 offsetT val;
624 int size;
625 {
626 addressT mask;
627
628 switch (size)
629 {
630 case 1: mask = ((addressT) 1 << 8) - 1; break;
631 case 2: mask = ((addressT) 1 << 16) - 1; break;
632 case 4: mask = ((addressT) 2 << 31) - 1; break;
633 #ifdef BFD64
634 case 8: mask = ((addressT) 2 << 63) - 1; break;
635 #endif
636 default: abort ();
637 }
638
639 /* If BFD64, sign extend val. */
640 if (!use_rela_relocations)
641 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
642 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
643
644 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
645 {
646 char buf1[40], buf2[40];
647
648 sprint_value (buf1, val);
649 sprint_value (buf2, val & mask);
650 as_warn (_("%s shortened to %s"), buf1, buf2);
651 }
652 return val & mask;
653 }
654
655 /* Returns 0 if attempting to add a prefix where one from the same
656 class already exists, 1 if non rep/repne added, 2 if rep/repne
657 added. */
658 static int
659 add_prefix (prefix)
660 unsigned int prefix;
661 {
662 int ret = 1;
663 int q;
664
665 if (prefix >= REX_OPCODE && prefix < REX_OPCODE + 16
666 && flag_code == CODE_64BIT)
667 q = REX_PREFIX;
668 else
669 switch (prefix)
670 {
671 default:
672 abort ();
673
674 case CS_PREFIX_OPCODE:
675 case DS_PREFIX_OPCODE:
676 case ES_PREFIX_OPCODE:
677 case FS_PREFIX_OPCODE:
678 case GS_PREFIX_OPCODE:
679 case SS_PREFIX_OPCODE:
680 q = SEG_PREFIX;
681 break;
682
683 case REPNE_PREFIX_OPCODE:
684 case REPE_PREFIX_OPCODE:
685 ret = 2;
686 /* fall thru */
687 case LOCK_PREFIX_OPCODE:
688 q = LOCKREP_PREFIX;
689 break;
690
691 case FWAIT_OPCODE:
692 q = WAIT_PREFIX;
693 break;
694
695 case ADDR_PREFIX_OPCODE:
696 q = ADDR_PREFIX;
697 break;
698
699 case DATA_PREFIX_OPCODE:
700 q = DATA_PREFIX;
701 break;
702 }
703
704 if (i.prefix[q] != 0)
705 {
706 as_bad (_("same type of prefix used twice"));
707 return 0;
708 }
709
710 i.prefixes += 1;
711 i.prefix[q] = prefix;
712 return ret;
713 }
714
715 static void
716 set_code_flag (value)
717 int value;
718 {
719 flag_code = value;
720 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
721 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
722 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
723 {
724 as_bad (_("64bit mode not supported on this CPU."));
725 }
726 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
727 {
728 as_bad (_("32bit mode not supported on this CPU."));
729 }
730 stackop_size = '\0';
731 }
732
733 static void
734 set_16bit_gcc_code_flag (new_code_flag)
735 int new_code_flag;
736 {
737 flag_code = new_code_flag;
738 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
739 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
740 stackop_size = 'l';
741 }
742
743 static void
744 set_intel_syntax (syntax_flag)
745 int syntax_flag;
746 {
747 /* Find out if register prefixing is specified. */
748 int ask_naked_reg = 0;
749
750 SKIP_WHITESPACE ();
751 if (!is_end_of_line[(unsigned char) *input_line_pointer])
752 {
753 char *string = input_line_pointer;
754 int e = get_symbol_end ();
755
756 if (strcmp (string, "prefix") == 0)
757 ask_naked_reg = 1;
758 else if (strcmp (string, "noprefix") == 0)
759 ask_naked_reg = -1;
760 else
761 as_bad (_("bad argument to syntax directive."));
762 *input_line_pointer = e;
763 }
764 demand_empty_rest_of_line ();
765
766 intel_syntax = syntax_flag;
767
768 if (ask_naked_reg == 0)
769 {
770 #ifdef BFD_ASSEMBLER
771 allow_naked_reg = (intel_syntax
772 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
773 #else
774 /* Conservative default. */
775 allow_naked_reg = 0;
776 #endif
777 }
778 else
779 allow_naked_reg = (ask_naked_reg < 0);
780 }
781
782 static void
783 set_cpu_arch (dummy)
784 int dummy ATTRIBUTE_UNUSED;
785 {
786 SKIP_WHITESPACE ();
787
788 if (!is_end_of_line[(unsigned char) *input_line_pointer])
789 {
790 char *string = input_line_pointer;
791 int e = get_symbol_end ();
792 int i;
793
794 for (i = 0; cpu_arch[i].name; i++)
795 {
796 if (strcmp (string, cpu_arch[i].name) == 0)
797 {
798 cpu_arch_name = cpu_arch[i].name;
799 cpu_arch_flags = (cpu_arch[i].flags
800 | (flag_code == CODE_64BIT ? Cpu64 : CpuNo64));
801 break;
802 }
803 }
804 if (!cpu_arch[i].name)
805 as_bad (_("no such architecture: `%s'"), string);
806
807 *input_line_pointer = e;
808 }
809 else
810 as_bad (_("missing cpu architecture"));
811
812 no_cond_jump_promotion = 0;
813 if (*input_line_pointer == ','
814 && !is_end_of_line[(unsigned char) input_line_pointer[1]])
815 {
816 char *string = ++input_line_pointer;
817 int e = get_symbol_end ();
818
819 if (strcmp (string, "nojumps") == 0)
820 no_cond_jump_promotion = 1;
821 else if (strcmp (string, "jumps") == 0)
822 ;
823 else
824 as_bad (_("no such architecture modifier: `%s'"), string);
825
826 *input_line_pointer = e;
827 }
828
829 demand_empty_rest_of_line ();
830 }
831
832 #ifdef BFD_ASSEMBLER
833 unsigned long
834 i386_mach ()
835 {
836 if (!strcmp (default_arch, "x86_64"))
837 return bfd_mach_x86_64;
838 else if (!strcmp (default_arch, "i386"))
839 return bfd_mach_i386_i386;
840 else
841 as_fatal (_("Unknown architecture"));
842 }
843 #endif
844 \f
845 void
846 md_begin ()
847 {
848 const char *hash_err;
849
850 /* Initialize op_hash hash table. */
851 op_hash = hash_new ();
852
853 {
854 const template *optab;
855 templates *core_optab;
856
857 /* Setup for loop. */
858 optab = i386_optab;
859 core_optab = (templates *) xmalloc (sizeof (templates));
860 core_optab->start = optab;
861
862 while (1)
863 {
864 ++optab;
865 if (optab->name == NULL
866 || strcmp (optab->name, (optab - 1)->name) != 0)
867 {
868 /* different name --> ship out current template list;
869 add to hash table; & begin anew. */
870 core_optab->end = optab;
871 hash_err = hash_insert (op_hash,
872 (optab - 1)->name,
873 (PTR) core_optab);
874 if (hash_err)
875 {
876 as_fatal (_("Internal Error: Can't hash %s: %s"),
877 (optab - 1)->name,
878 hash_err);
879 }
880 if (optab->name == NULL)
881 break;
882 core_optab = (templates *) xmalloc (sizeof (templates));
883 core_optab->start = optab;
884 }
885 }
886 }
887
888 /* Initialize reg_hash hash table. */
889 reg_hash = hash_new ();
890 {
891 const reg_entry *regtab;
892
893 for (regtab = i386_regtab;
894 regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
895 regtab++)
896 {
897 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
898 if (hash_err)
899 as_fatal (_("Internal Error: Can't hash %s: %s"),
900 regtab->reg_name,
901 hash_err);
902 }
903 }
904
905 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
906 {
907 int c;
908 char *p;
909
910 for (c = 0; c < 256; c++)
911 {
912 if (ISDIGIT (c))
913 {
914 digit_chars[c] = c;
915 mnemonic_chars[c] = c;
916 register_chars[c] = c;
917 operand_chars[c] = c;
918 }
919 else if (ISLOWER (c))
920 {
921 mnemonic_chars[c] = c;
922 register_chars[c] = c;
923 operand_chars[c] = c;
924 }
925 else if (ISUPPER (c))
926 {
927 mnemonic_chars[c] = TOLOWER (c);
928 register_chars[c] = mnemonic_chars[c];
929 operand_chars[c] = c;
930 }
931
932 if (ISALPHA (c) || ISDIGIT (c))
933 identifier_chars[c] = c;
934 else if (c >= 128)
935 {
936 identifier_chars[c] = c;
937 operand_chars[c] = c;
938 }
939 }
940
941 #ifdef LEX_AT
942 identifier_chars['@'] = '@';
943 #endif
944 digit_chars['-'] = '-';
945 identifier_chars['_'] = '_';
946 identifier_chars['.'] = '.';
947
948 for (p = operand_special_chars; *p != '\0'; p++)
949 operand_chars[(unsigned char) *p] = *p;
950 }
951
952 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
953 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
954 {
955 record_alignment (text_section, 2);
956 record_alignment (data_section, 2);
957 record_alignment (bss_section, 2);
958 }
959 #endif
960 }
961
962 void
963 i386_print_statistics (file)
964 FILE *file;
965 {
966 hash_print_statistics (file, "i386 opcode", op_hash);
967 hash_print_statistics (file, "i386 register", reg_hash);
968 }
969 \f
970 #ifdef DEBUG386
971
972 /* Debugging routines for md_assemble. */
973 static void pi PARAMS ((char *, i386_insn *));
974 static void pte PARAMS ((template *));
975 static void pt PARAMS ((unsigned int));
976 static void pe PARAMS ((expressionS *));
977 static void ps PARAMS ((symbolS *));
978
979 static void
980 pi (line, x)
981 char *line;
982 i386_insn *x;
983 {
984 unsigned int i;
985
986 fprintf (stdout, "%s: template ", line);
987 pte (&x->tm);
988 fprintf (stdout, " address: base %s index %s scale %x\n",
989 x->base_reg ? x->base_reg->reg_name : "none",
990 x->index_reg ? x->index_reg->reg_name : "none",
991 x->log2_scale_factor);
992 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
993 x->rm.mode, x->rm.reg, x->rm.regmem);
994 fprintf (stdout, " sib: base %x index %x scale %x\n",
995 x->sib.base, x->sib.index, x->sib.scale);
996 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
997 (x->rex & REX_MODE64) != 0,
998 (x->rex & REX_EXTX) != 0,
999 (x->rex & REX_EXTY) != 0,
1000 (x->rex & REX_EXTZ) != 0);
1001 for (i = 0; i < x->operands; i++)
1002 {
1003 fprintf (stdout, " #%d: ", i + 1);
1004 pt (x->types[i]);
1005 fprintf (stdout, "\n");
1006 if (x->types[i]
1007 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
1008 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
1009 if (x->types[i] & Imm)
1010 pe (x->op[i].imms);
1011 if (x->types[i] & Disp)
1012 pe (x->op[i].disps);
1013 }
1014 }
1015
1016 static void
1017 pte (t)
1018 template *t;
1019 {
1020 unsigned int i;
1021 fprintf (stdout, " %d operands ", t->operands);
1022 fprintf (stdout, "opcode %x ", t->base_opcode);
1023 if (t->extension_opcode != None)
1024 fprintf (stdout, "ext %x ", t->extension_opcode);
1025 if (t->opcode_modifier & D)
1026 fprintf (stdout, "D");
1027 if (t->opcode_modifier & W)
1028 fprintf (stdout, "W");
1029 fprintf (stdout, "\n");
1030 for (i = 0; i < t->operands; i++)
1031 {
1032 fprintf (stdout, " #%d type ", i + 1);
1033 pt (t->operand_types[i]);
1034 fprintf (stdout, "\n");
1035 }
1036 }
1037
1038 static void
1039 pe (e)
1040 expressionS *e;
1041 {
1042 fprintf (stdout, " operation %d\n", e->X_op);
1043 fprintf (stdout, " add_number %ld (%lx)\n",
1044 (long) e->X_add_number, (long) e->X_add_number);
1045 if (e->X_add_symbol)
1046 {
1047 fprintf (stdout, " add_symbol ");
1048 ps (e->X_add_symbol);
1049 fprintf (stdout, "\n");
1050 }
1051 if (e->X_op_symbol)
1052 {
1053 fprintf (stdout, " op_symbol ");
1054 ps (e->X_op_symbol);
1055 fprintf (stdout, "\n");
1056 }
1057 }
1058
1059 static void
1060 ps (s)
1061 symbolS *s;
1062 {
1063 fprintf (stdout, "%s type %s%s",
1064 S_GET_NAME (s),
1065 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
1066 segment_name (S_GET_SEGMENT (s)));
1067 }
1068
1069 struct type_name
1070 {
1071 unsigned int mask;
1072 char *tname;
1073 }
1074
1075 static const type_names[] =
1076 {
1077 { Reg8, "r8" },
1078 { Reg16, "r16" },
1079 { Reg32, "r32" },
1080 { Reg64, "r64" },
1081 { Imm8, "i8" },
1082 { Imm8S, "i8s" },
1083 { Imm16, "i16" },
1084 { Imm32, "i32" },
1085 { Imm32S, "i32s" },
1086 { Imm64, "i64" },
1087 { Imm1, "i1" },
1088 { BaseIndex, "BaseIndex" },
1089 { Disp8, "d8" },
1090 { Disp16, "d16" },
1091 { Disp32, "d32" },
1092 { Disp32S, "d32s" },
1093 { Disp64, "d64" },
1094 { InOutPortReg, "InOutPortReg" },
1095 { ShiftCount, "ShiftCount" },
1096 { Control, "control reg" },
1097 { Test, "test reg" },
1098 { Debug, "debug reg" },
1099 { FloatReg, "FReg" },
1100 { FloatAcc, "FAcc" },
1101 { SReg2, "SReg2" },
1102 { SReg3, "SReg3" },
1103 { Acc, "Acc" },
1104 { JumpAbsolute, "Jump Absolute" },
1105 { RegMMX, "rMMX" },
1106 { RegXMM, "rXMM" },
1107 { EsSeg, "es" },
1108 { 0, "" }
1109 };
1110
1111 static void
1112 pt (t)
1113 unsigned int t;
1114 {
1115 const struct type_name *ty;
1116
1117 for (ty = type_names; ty->mask; ty++)
1118 if (t & ty->mask)
1119 fprintf (stdout, "%s, ", ty->tname);
1120 fflush (stdout);
1121 }
1122
1123 #endif /* DEBUG386 */
1124 \f
1125 int
1126 tc_i386_force_relocation (fixp)
1127 struct fix *fixp;
1128 {
1129 #ifdef BFD_ASSEMBLER
1130 if (fixp->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1131 || fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1132 return 1;
1133 return 0;
1134 #else
1135 /* For COFF. */
1136 return fixp->fx_r_type == 7;
1137 #endif
1138 }
1139
1140 #ifdef BFD_ASSEMBLER
1141 static bfd_reloc_code_real_type reloc
1142 PARAMS ((int, int, int, bfd_reloc_code_real_type));
1143
1144 static bfd_reloc_code_real_type
1145 reloc (size, pcrel, sign, other)
1146 int size;
1147 int pcrel;
1148 int sign;
1149 bfd_reloc_code_real_type other;
1150 {
1151 if (other != NO_RELOC)
1152 return other;
1153
1154 if (pcrel)
1155 {
1156 if (!sign)
1157 as_bad (_("There are no unsigned pc-relative relocations"));
1158 switch (size)
1159 {
1160 case 1: return BFD_RELOC_8_PCREL;
1161 case 2: return BFD_RELOC_16_PCREL;
1162 case 4: return BFD_RELOC_32_PCREL;
1163 }
1164 as_bad (_("can not do %d byte pc-relative relocation"), size);
1165 }
1166 else
1167 {
1168 if (sign)
1169 switch (size)
1170 {
1171 case 4: return BFD_RELOC_X86_64_32S;
1172 }
1173 else
1174 switch (size)
1175 {
1176 case 1: return BFD_RELOC_8;
1177 case 2: return BFD_RELOC_16;
1178 case 4: return BFD_RELOC_32;
1179 case 8: return BFD_RELOC_64;
1180 }
1181 as_bad (_("can not do %s %d byte relocation"),
1182 sign ? "signed" : "unsigned", size);
1183 }
1184
1185 abort ();
1186 return BFD_RELOC_NONE;
1187 }
1188
1189 /* Here we decide which fixups can be adjusted to make them relative to
1190 the beginning of the section instead of the symbol. Basically we need
1191 to make sure that the dynamic relocations are done correctly, so in
1192 some cases we force the original symbol to be used. */
1193
1194 int
1195 tc_i386_fix_adjustable (fixP)
1196 fixS *fixP;
1197 {
1198 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1199 /* Prevent all adjustments to global symbols, or else dynamic
1200 linking will not work correctly. */
1201 if (S_IS_EXTERNAL (fixP->fx_addsy)
1202 || S_IS_WEAK (fixP->fx_addsy)
1203 /* Don't adjust pc-relative references to merge sections in 64-bit
1204 mode. */
1205 || (use_rela_relocations
1206 && (S_GET_SEGMENT (fixP->fx_addsy)->flags & SEC_MERGE) != 0
1207 && fixP->fx_pcrel))
1208 return 0;
1209 #endif
1210 /* adjust_reloc_syms doesn't know about the GOT. */
1211 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1212 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1213 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1214 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1215 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1216 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
1217 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1218 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1219 return 0;
1220 return 1;
1221 }
1222 #else
1223 #define reloc(SIZE,PCREL,SIGN,OTHER) 0
1224 #define BFD_RELOC_16 0
1225 #define BFD_RELOC_32 0
1226 #define BFD_RELOC_16_PCREL 0
1227 #define BFD_RELOC_32_PCREL 0
1228 #define BFD_RELOC_386_PLT32 0
1229 #define BFD_RELOC_386_GOT32 0
1230 #define BFD_RELOC_386_GOTOFF 0
1231 #define BFD_RELOC_X86_64_PLT32 0
1232 #define BFD_RELOC_X86_64_GOT32 0
1233 #define BFD_RELOC_X86_64_GOTPCREL 0
1234 #endif
1235
1236 static int intel_float_operand PARAMS ((const char *mnemonic));
1237
1238 static int
1239 intel_float_operand (mnemonic)
1240 const char *mnemonic;
1241 {
1242 if (mnemonic[0] == 'f' && mnemonic[1] == 'i')
1243 return 2;
1244
1245 if (mnemonic[0] == 'f')
1246 return 1;
1247
1248 return 0;
1249 }
1250
1251 /* This is the guts of the machine-dependent assembler. LINE points to a
1252 machine dependent instruction. This function is supposed to emit
1253 the frags/bytes it assembles to. */
1254
1255 void
1256 md_assemble (line)
1257 char *line;
1258 {
1259 int j;
1260 char mnemonic[MAX_MNEM_SIZE];
1261
1262 /* Initialize globals. */
1263 memset (&i, '\0', sizeof (i));
1264 for (j = 0; j < MAX_OPERANDS; j++)
1265 i.reloc[j] = NO_RELOC;
1266 memset (disp_expressions, '\0', sizeof (disp_expressions));
1267 memset (im_expressions, '\0', sizeof (im_expressions));
1268 save_stack_p = save_stack;
1269
1270 /* First parse an instruction mnemonic & call i386_operand for the operands.
1271 We assume that the scrubber has arranged it so that line[0] is the valid
1272 start of a (possibly prefixed) mnemonic. */
1273
1274 line = parse_insn (line, mnemonic);
1275 if (line == NULL)
1276 return;
1277
1278 line = parse_operands (line, mnemonic);
1279 if (line == NULL)
1280 return;
1281
1282 /* Now we've parsed the mnemonic into a set of templates, and have the
1283 operands at hand. */
1284
1285 /* All intel opcodes have reversed operands except for "bound" and
1286 "enter". We also don't reverse intersegment "jmp" and "call"
1287 instructions with 2 immediate operands so that the immediate segment
1288 precedes the offset, as it does when in AT&T mode. "enter" and the
1289 intersegment "jmp" and "call" instructions are the only ones that
1290 have two immediate operands. */
1291 if (intel_syntax && i.operands > 1
1292 && (strcmp (mnemonic, "bound") != 0)
1293 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1294 swap_operands ();
1295
1296 if (i.imm_operands)
1297 optimize_imm ();
1298
1299 if (i.disp_operands)
1300 optimize_disp ();
1301
1302 /* Next, we find a template that matches the given insn,
1303 making sure the overlap of the given operands types is consistent
1304 with the template operand types. */
1305
1306 if (!match_template ())
1307 return;
1308
1309 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1310 if (SYSV386_COMPAT
1311 && intel_syntax
1312 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1313 i.tm.base_opcode ^= FloatR;
1314
1315 if (i.tm.opcode_modifier & FWait)
1316 if (!add_prefix (FWAIT_OPCODE))
1317 return;
1318
1319 /* Check string instruction segment overrides. */
1320 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1321 {
1322 if (!check_string ())
1323 return;
1324 }
1325
1326 if (!process_suffix ())
1327 return;
1328
1329 /* Make still unresolved immediate matches conform to size of immediate
1330 given in i.suffix. */
1331 if (!finalize_imm ())
1332 return;
1333
1334 if (i.types[0] & Imm1)
1335 i.imm_operands = 0; /* kludge for shift insns. */
1336 if (i.types[0] & ImplicitRegister)
1337 i.reg_operands--;
1338 if (i.types[1] & ImplicitRegister)
1339 i.reg_operands--;
1340 if (i.types[2] & ImplicitRegister)
1341 i.reg_operands--;
1342
1343 if (i.tm.opcode_modifier & ImmExt)
1344 {
1345 /* These AMD 3DNow! and Intel Katmai New Instructions have an
1346 opcode suffix which is coded in the same place as an 8-bit
1347 immediate field would be. Here we fake an 8-bit immediate
1348 operand from the opcode suffix stored in tm.extension_opcode. */
1349
1350 expressionS *exp;
1351
1352 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
1353
1354 exp = &im_expressions[i.imm_operands++];
1355 i.op[i.operands].imms = exp;
1356 i.types[i.operands++] = Imm8;
1357 exp->X_op = O_constant;
1358 exp->X_add_number = i.tm.extension_opcode;
1359 i.tm.extension_opcode = None;
1360 }
1361
1362 /* For insns with operands there are more diddles to do to the opcode. */
1363 if (i.operands)
1364 {
1365 if (!process_operands ())
1366 return;
1367 }
1368 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
1369 {
1370 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
1371 as_warn (_("translating to `%sp'"), i.tm.name);
1372 }
1373
1374 /* Handle conversion of 'int $3' --> special int3 insn. */
1375 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
1376 {
1377 i.tm.base_opcode = INT3_OPCODE;
1378 i.imm_operands = 0;
1379 }
1380
1381 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
1382 && i.op[0].disps->X_op == O_constant)
1383 {
1384 /* Convert "jmp constant" (and "call constant") to a jump (call) to
1385 the absolute address given by the constant. Since ix86 jumps and
1386 calls are pc relative, we need to generate a reloc. */
1387 i.op[0].disps->X_add_symbol = &abs_symbol;
1388 i.op[0].disps->X_op = O_symbol;
1389 }
1390
1391 if ((i.tm.opcode_modifier & Rex64) != 0)
1392 i.rex |= REX_MODE64;
1393
1394 /* For 8 bit registers we need an empty rex prefix. Also if the
1395 instruction already has a prefix, we need to convert old
1396 registers to new ones. */
1397
1398 if (((i.types[0] & Reg8) != 0
1399 && (i.op[0].regs->reg_flags & RegRex64) != 0)
1400 || ((i.types[1] & Reg8) != 0
1401 && (i.op[1].regs->reg_flags & RegRex64) != 0)
1402 || (((i.types[0] & Reg8) != 0 || (i.types[1] & Reg8) != 0)
1403 && i.rex != 0))
1404 {
1405 int x;
1406
1407 i.rex |= REX_OPCODE;
1408 for (x = 0; x < 2; x++)
1409 {
1410 /* Look for 8 bit operand that uses old registers. */
1411 if ((i.types[x] & Reg8) != 0
1412 && (i.op[x].regs->reg_flags & RegRex64) == 0)
1413 {
1414 /* In case it is "hi" register, give up. */
1415 if (i.op[x].regs->reg_num > 3)
1416 as_bad (_("can't encode register '%%%s' in an instruction requiring REX prefix.\n"),
1417 i.op[x].regs->reg_name);
1418
1419 /* Otherwise it is equivalent to the extended register.
1420 Since the encoding doesn't change this is merely
1421 cosmetic cleanup for debug output. */
1422
1423 i.op[x].regs = i.op[x].regs + 8;
1424 }
1425 }
1426 }
1427
1428 if (i.rex != 0)
1429 add_prefix (REX_OPCODE | i.rex);
1430
1431 /* We are ready to output the insn. */
1432 output_insn ();
1433 }
1434
1435 static char *
1436 parse_insn (line, mnemonic)
1437 char *line;
1438 char *mnemonic;
1439 {
1440 char *l = line;
1441 char *token_start = l;
1442 char *mnem_p;
1443
1444 /* Non-zero if we found a prefix only acceptable with string insns. */
1445 const char *expecting_string_instruction = NULL;
1446
1447 while (1)
1448 {
1449 mnem_p = mnemonic;
1450 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1451 {
1452 mnem_p++;
1453 if (mnem_p >= mnemonic + MAX_MNEM_SIZE)
1454 {
1455 as_bad (_("no such instruction: `%s'"), token_start);
1456 return NULL;
1457 }
1458 l++;
1459 }
1460 if (!is_space_char (*l)
1461 && *l != END_OF_INSN
1462 && *l != PREFIX_SEPARATOR
1463 && *l != ',')
1464 {
1465 as_bad (_("invalid character %s in mnemonic"),
1466 output_invalid (*l));
1467 return NULL;
1468 }
1469 if (token_start == l)
1470 {
1471 if (*l == PREFIX_SEPARATOR)
1472 as_bad (_("expecting prefix; got nothing"));
1473 else
1474 as_bad (_("expecting mnemonic; got nothing"));
1475 return NULL;
1476 }
1477
1478 /* Look up instruction (or prefix) via hash table. */
1479 current_templates = hash_find (op_hash, mnemonic);
1480
1481 if (*l != END_OF_INSN
1482 && (!is_space_char (*l) || l[1] != END_OF_INSN)
1483 && current_templates
1484 && (current_templates->start->opcode_modifier & IsPrefix))
1485 {
1486 /* If we are in 16-bit mode, do not allow addr16 or data16.
1487 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1488 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1489 && flag_code != CODE_64BIT
1490 && (((current_templates->start->opcode_modifier & Size32) != 0)
1491 ^ (flag_code == CODE_16BIT)))
1492 {
1493 as_bad (_("redundant %s prefix"),
1494 current_templates->start->name);
1495 return NULL;
1496 }
1497 /* Add prefix, checking for repeated prefixes. */
1498 switch (add_prefix (current_templates->start->base_opcode))
1499 {
1500 case 0:
1501 return NULL;
1502 case 2:
1503 expecting_string_instruction = current_templates->start->name;
1504 break;
1505 }
1506 /* Skip past PREFIX_SEPARATOR and reset token_start. */
1507 token_start = ++l;
1508 }
1509 else
1510 break;
1511 }
1512
1513 if (!current_templates)
1514 {
1515 /* See if we can get a match by trimming off a suffix. */
1516 switch (mnem_p[-1])
1517 {
1518 case WORD_MNEM_SUFFIX:
1519 case BYTE_MNEM_SUFFIX:
1520 case QWORD_MNEM_SUFFIX:
1521 i.suffix = mnem_p[-1];
1522 mnem_p[-1] = '\0';
1523 current_templates = hash_find (op_hash, mnemonic);
1524 break;
1525 case SHORT_MNEM_SUFFIX:
1526 case LONG_MNEM_SUFFIX:
1527 if (!intel_syntax)
1528 {
1529 i.suffix = mnem_p[-1];
1530 mnem_p[-1] = '\0';
1531 current_templates = hash_find (op_hash, mnemonic);
1532 }
1533 break;
1534
1535 /* Intel Syntax. */
1536 case 'd':
1537 if (intel_syntax)
1538 {
1539 if (intel_float_operand (mnemonic))
1540 i.suffix = SHORT_MNEM_SUFFIX;
1541 else
1542 i.suffix = LONG_MNEM_SUFFIX;
1543 mnem_p[-1] = '\0';
1544 current_templates = hash_find (op_hash, mnemonic);
1545 }
1546 break;
1547 }
1548 if (!current_templates)
1549 {
1550 as_bad (_("no such instruction: `%s'"), token_start);
1551 return NULL;
1552 }
1553 }
1554
1555 if (current_templates->start->opcode_modifier & (Jump | JumpByte))
1556 {
1557 /* Check for a branch hint. We allow ",pt" and ",pn" for
1558 predict taken and predict not taken respectively.
1559 I'm not sure that branch hints actually do anything on loop
1560 and jcxz insns (JumpByte) for current Pentium4 chips. They
1561 may work in the future and it doesn't hurt to accept them
1562 now. */
1563 if (l[0] == ',' && l[1] == 'p')
1564 {
1565 if (l[2] == 't')
1566 {
1567 if (!add_prefix (DS_PREFIX_OPCODE))
1568 return NULL;
1569 l += 3;
1570 }
1571 else if (l[2] == 'n')
1572 {
1573 if (!add_prefix (CS_PREFIX_OPCODE))
1574 return NULL;
1575 l += 3;
1576 }
1577 }
1578 }
1579 /* Any other comma loses. */
1580 if (*l == ',')
1581 {
1582 as_bad (_("invalid character %s in mnemonic"),
1583 output_invalid (*l));
1584 return NULL;
1585 }
1586
1587 /* Check if instruction is supported on specified architecture. */
1588 if ((current_templates->start->cpu_flags & ~(Cpu64 | CpuNo64))
1589 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64)))
1590 {
1591 as_warn (_("`%s' is not supported on `%s'"),
1592 current_templates->start->name, cpu_arch_name);
1593 }
1594 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
1595 {
1596 as_warn (_("use .code16 to ensure correct addressing mode"));
1597 }
1598
1599 /* Check for rep/repne without a string instruction. */
1600 if (expecting_string_instruction
1601 && !(current_templates->start->opcode_modifier & IsString))
1602 {
1603 as_bad (_("expecting string instruction after `%s'"),
1604 expecting_string_instruction);
1605 return NULL;
1606 }
1607
1608 return l;
1609 }
1610
1611 static char *
1612 parse_operands (l, mnemonic)
1613 char *l;
1614 const char *mnemonic;
1615 {
1616 char *token_start;
1617
1618 /* 1 if operand is pending after ','. */
1619 unsigned int expecting_operand = 0;
1620
1621 /* Non-zero if operand parens not balanced. */
1622 unsigned int paren_not_balanced;
1623
1624 while (*l != END_OF_INSN)
1625 {
1626 /* Skip optional white space before operand. */
1627 if (is_space_char (*l))
1628 ++l;
1629 if (!is_operand_char (*l) && *l != END_OF_INSN)
1630 {
1631 as_bad (_("invalid character %s before operand %d"),
1632 output_invalid (*l),
1633 i.operands + 1);
1634 return NULL;
1635 }
1636 token_start = l; /* after white space */
1637 paren_not_balanced = 0;
1638 while (paren_not_balanced || *l != ',')
1639 {
1640 if (*l == END_OF_INSN)
1641 {
1642 if (paren_not_balanced)
1643 {
1644 if (!intel_syntax)
1645 as_bad (_("unbalanced parenthesis in operand %d."),
1646 i.operands + 1);
1647 else
1648 as_bad (_("unbalanced brackets in operand %d."),
1649 i.operands + 1);
1650 return NULL;
1651 }
1652 else
1653 break; /* we are done */
1654 }
1655 else if (!is_operand_char (*l) && !is_space_char (*l))
1656 {
1657 as_bad (_("invalid character %s in operand %d"),
1658 output_invalid (*l),
1659 i.operands + 1);
1660 return NULL;
1661 }
1662 if (!intel_syntax)
1663 {
1664 if (*l == '(')
1665 ++paren_not_balanced;
1666 if (*l == ')')
1667 --paren_not_balanced;
1668 }
1669 else
1670 {
1671 if (*l == '[')
1672 ++paren_not_balanced;
1673 if (*l == ']')
1674 --paren_not_balanced;
1675 }
1676 l++;
1677 }
1678 if (l != token_start)
1679 { /* Yes, we've read in another operand. */
1680 unsigned int operand_ok;
1681 this_operand = i.operands++;
1682 if (i.operands > MAX_OPERANDS)
1683 {
1684 as_bad (_("spurious operands; (%d operands/instruction max)"),
1685 MAX_OPERANDS);
1686 return NULL;
1687 }
1688 /* Now parse operand adding info to 'i' as we go along. */
1689 END_STRING_AND_SAVE (l);
1690
1691 if (intel_syntax)
1692 operand_ok =
1693 i386_intel_operand (token_start,
1694 intel_float_operand (mnemonic));
1695 else
1696 operand_ok = i386_operand (token_start);
1697
1698 RESTORE_END_STRING (l);
1699 if (!operand_ok)
1700 return NULL;
1701 }
1702 else
1703 {
1704 if (expecting_operand)
1705 {
1706 expecting_operand_after_comma:
1707 as_bad (_("expecting operand after ','; got nothing"));
1708 return NULL;
1709 }
1710 if (*l == ',')
1711 {
1712 as_bad (_("expecting operand before ','; got nothing"));
1713 return NULL;
1714 }
1715 }
1716
1717 /* Now *l must be either ',' or END_OF_INSN. */
1718 if (*l == ',')
1719 {
1720 if (*++l == END_OF_INSN)
1721 {
1722 /* Just skip it, if it's \n complain. */
1723 goto expecting_operand_after_comma;
1724 }
1725 expecting_operand = 1;
1726 }
1727 }
1728 return l;
1729 }
1730
1731 static void
1732 swap_operands ()
1733 {
1734 union i386_op temp_op;
1735 unsigned int temp_type;
1736 RELOC_ENUM temp_reloc;
1737 int xchg1 = 0;
1738 int xchg2 = 0;
1739
1740 if (i.operands == 2)
1741 {
1742 xchg1 = 0;
1743 xchg2 = 1;
1744 }
1745 else if (i.operands == 3)
1746 {
1747 xchg1 = 0;
1748 xchg2 = 2;
1749 }
1750 temp_type = i.types[xchg2];
1751 i.types[xchg2] = i.types[xchg1];
1752 i.types[xchg1] = temp_type;
1753 temp_op = i.op[xchg2];
1754 i.op[xchg2] = i.op[xchg1];
1755 i.op[xchg1] = temp_op;
1756 temp_reloc = i.reloc[xchg2];
1757 i.reloc[xchg2] = i.reloc[xchg1];
1758 i.reloc[xchg1] = temp_reloc;
1759
1760 if (i.mem_operands == 2)
1761 {
1762 const seg_entry *temp_seg;
1763 temp_seg = i.seg[0];
1764 i.seg[0] = i.seg[1];
1765 i.seg[1] = temp_seg;
1766 }
1767 }
1768
1769 /* Try to ensure constant immediates are represented in the smallest
1770 opcode possible. */
1771 static void
1772 optimize_imm ()
1773 {
1774 char guess_suffix = 0;
1775 int op;
1776
1777 if (i.suffix)
1778 guess_suffix = i.suffix;
1779 else if (i.reg_operands)
1780 {
1781 /* Figure out a suffix from the last register operand specified.
1782 We can't do this properly yet, ie. excluding InOutPortReg,
1783 but the following works for instructions with immediates.
1784 In any case, we can't set i.suffix yet. */
1785 for (op = i.operands; --op >= 0;)
1786 if (i.types[op] & Reg)
1787 {
1788 if (i.types[op] & Reg8)
1789 guess_suffix = BYTE_MNEM_SUFFIX;
1790 else if (i.types[op] & Reg16)
1791 guess_suffix = WORD_MNEM_SUFFIX;
1792 else if (i.types[op] & Reg32)
1793 guess_suffix = LONG_MNEM_SUFFIX;
1794 else if (i.types[op] & Reg64)
1795 guess_suffix = QWORD_MNEM_SUFFIX;
1796 break;
1797 }
1798 }
1799 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
1800 guess_suffix = WORD_MNEM_SUFFIX;
1801
1802 for (op = i.operands; --op >= 0;)
1803 if (i.types[op] & Imm)
1804 {
1805 switch (i.op[op].imms->X_op)
1806 {
1807 case O_constant:
1808 /* If a suffix is given, this operand may be shortened. */
1809 switch (guess_suffix)
1810 {
1811 case LONG_MNEM_SUFFIX:
1812 i.types[op] |= Imm32 | Imm64;
1813 break;
1814 case WORD_MNEM_SUFFIX:
1815 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
1816 break;
1817 case BYTE_MNEM_SUFFIX:
1818 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
1819 break;
1820 }
1821
1822 /* If this operand is at most 16 bits, convert it
1823 to a signed 16 bit number before trying to see
1824 whether it will fit in an even smaller size.
1825 This allows a 16-bit operand such as $0xffe0 to
1826 be recognised as within Imm8S range. */
1827 if ((i.types[op] & Imm16)
1828 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
1829 {
1830 i.op[op].imms->X_add_number =
1831 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
1832 }
1833 if ((i.types[op] & Imm32)
1834 && ((i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1))
1835 == 0))
1836 {
1837 i.op[op].imms->X_add_number = ((i.op[op].imms->X_add_number
1838 ^ ((offsetT) 1 << 31))
1839 - ((offsetT) 1 << 31));
1840 }
1841 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
1842
1843 /* We must avoid matching of Imm32 templates when 64bit
1844 only immediate is available. */
1845 if (guess_suffix == QWORD_MNEM_SUFFIX)
1846 i.types[op] &= ~Imm32;
1847 break;
1848
1849 case O_absent:
1850 case O_register:
1851 abort ();
1852
1853 /* Symbols and expressions. */
1854 default:
1855 /* Convert symbolic operand to proper sizes for matching. */
1856 switch (guess_suffix)
1857 {
1858 case QWORD_MNEM_SUFFIX:
1859 i.types[op] = Imm64 | Imm32S;
1860 break;
1861 case LONG_MNEM_SUFFIX:
1862 i.types[op] = Imm32 | Imm64;
1863 break;
1864 case WORD_MNEM_SUFFIX:
1865 i.types[op] = Imm16 | Imm32 | Imm64;
1866 break;
1867 break;
1868 case BYTE_MNEM_SUFFIX:
1869 i.types[op] = Imm8 | Imm8S | Imm16 | Imm32S | Imm32;
1870 break;
1871 break;
1872 }
1873 break;
1874 }
1875 }
1876 }
1877
1878 /* Try to use the smallest displacement type too. */
1879 static void
1880 optimize_disp ()
1881 {
1882 int op;
1883
1884 for (op = i.operands; --op >= 0;)
1885 if ((i.types[op] & Disp) && i.op[op].disps->X_op == O_constant)
1886 {
1887 offsetT disp = i.op[op].disps->X_add_number;
1888
1889 if (i.types[op] & Disp16)
1890 {
1891 /* We know this operand is at most 16 bits, so
1892 convert to a signed 16 bit number before trying
1893 to see whether it will fit in an even smaller
1894 size. */
1895
1896 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
1897 }
1898 else if (i.types[op] & Disp32)
1899 {
1900 /* We know this operand is at most 32 bits, so convert to a
1901 signed 32 bit number before trying to see whether it will
1902 fit in an even smaller size. */
1903 disp &= (((offsetT) 2 << 31) - 1);
1904 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
1905 }
1906 if (flag_code == CODE_64BIT)
1907 {
1908 if (fits_in_signed_long (disp))
1909 i.types[op] |= Disp32S;
1910 if (fits_in_unsigned_long (disp))
1911 i.types[op] |= Disp32;
1912 }
1913 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
1914 && fits_in_signed_byte (disp))
1915 i.types[op] |= Disp8;
1916 }
1917 }
1918
1919 static int
1920 match_template ()
1921 {
1922 /* Points to template once we've found it. */
1923 const template *t;
1924 unsigned int overlap0, overlap1, overlap2;
1925 unsigned int found_reverse_match;
1926 int suffix_check;
1927
1928 #define MATCH(overlap, given, template) \
1929 ((overlap & ~JumpAbsolute) \
1930 && (((given) & (BaseIndex | JumpAbsolute)) \
1931 == ((overlap) & (BaseIndex | JumpAbsolute))))
1932
1933 /* If given types r0 and r1 are registers they must be of the same type
1934 unless the expected operand type register overlap is null.
1935 Note that Acc in a template matches every size of reg. */
1936 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
1937 (((g0) & Reg) == 0 || ((g1) & Reg) == 0 \
1938 || ((g0) & Reg) == ((g1) & Reg) \
1939 || ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
1940
1941 overlap0 = 0;
1942 overlap1 = 0;
1943 overlap2 = 0;
1944 found_reverse_match = 0;
1945 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
1946 ? No_bSuf
1947 : (i.suffix == WORD_MNEM_SUFFIX
1948 ? No_wSuf
1949 : (i.suffix == SHORT_MNEM_SUFFIX
1950 ? No_sSuf
1951 : (i.suffix == LONG_MNEM_SUFFIX
1952 ? No_lSuf
1953 : (i.suffix == QWORD_MNEM_SUFFIX
1954 ? No_qSuf
1955 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX
1956 ? No_xSuf : 0))))));
1957
1958 for (t = current_templates->start;
1959 t < current_templates->end;
1960 t++)
1961 {
1962 /* Must have right number of operands. */
1963 if (i.operands != t->operands)
1964 continue;
1965
1966 /* Check the suffix, except for some instructions in intel mode. */
1967 if ((t->opcode_modifier & suffix_check)
1968 && !(intel_syntax
1969 && (t->opcode_modifier & IgnoreSize))
1970 && !(intel_syntax
1971 && t->base_opcode == 0xd9
1972 && (t->extension_opcode == 5 /* 0xd9,5 "fldcw" */
1973 || t->extension_opcode == 7))) /* 0xd9,7 "f{n}stcw" */
1974 continue;
1975
1976 /* Do not verify operands when there are none. */
1977 else if (!t->operands)
1978 {
1979 if (t->cpu_flags & ~cpu_arch_flags)
1980 continue;
1981 /* We've found a match; break out of loop. */
1982 break;
1983 }
1984
1985 overlap0 = i.types[0] & t->operand_types[0];
1986 switch (t->operands)
1987 {
1988 case 1:
1989 if (!MATCH (overlap0, i.types[0], t->operand_types[0]))
1990 continue;
1991 break;
1992 case 2:
1993 case 3:
1994 overlap1 = i.types[1] & t->operand_types[1];
1995 if (!MATCH (overlap0, i.types[0], t->operand_types[0])
1996 || !MATCH (overlap1, i.types[1], t->operand_types[1])
1997 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
1998 t->operand_types[0],
1999 overlap1, i.types[1],
2000 t->operand_types[1]))
2001 {
2002 /* Check if other direction is valid ... */
2003 if ((t->opcode_modifier & (D | FloatD)) == 0)
2004 continue;
2005
2006 /* Try reversing direction of operands. */
2007 overlap0 = i.types[0] & t->operand_types[1];
2008 overlap1 = i.types[1] & t->operand_types[0];
2009 if (!MATCH (overlap0, i.types[0], t->operand_types[1])
2010 || !MATCH (overlap1, i.types[1], t->operand_types[0])
2011 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2012 t->operand_types[1],
2013 overlap1, i.types[1],
2014 t->operand_types[0]))
2015 {
2016 /* Does not match either direction. */
2017 continue;
2018 }
2019 /* found_reverse_match holds which of D or FloatDR
2020 we've found. */
2021 found_reverse_match = t->opcode_modifier & (D | FloatDR);
2022 }
2023 /* Found a forward 2 operand match here. */
2024 else if (t->operands == 3)
2025 {
2026 /* Here we make use of the fact that there are no
2027 reverse match 3 operand instructions, and all 3
2028 operand instructions only need to be checked for
2029 register consistency between operands 2 and 3. */
2030 overlap2 = i.types[2] & t->operand_types[2];
2031 if (!MATCH (overlap2, i.types[2], t->operand_types[2])
2032 || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1],
2033 t->operand_types[1],
2034 overlap2, i.types[2],
2035 t->operand_types[2]))
2036
2037 continue;
2038 }
2039 /* Found either forward/reverse 2 or 3 operand match here:
2040 slip through to break. */
2041 }
2042 if (t->cpu_flags & ~cpu_arch_flags)
2043 {
2044 found_reverse_match = 0;
2045 continue;
2046 }
2047 /* We've found a match; break out of loop. */
2048 break;
2049 }
2050
2051 if (t == current_templates->end)
2052 {
2053 /* We found no match. */
2054 as_bad (_("suffix or operands invalid for `%s'"),
2055 current_templates->start->name);
2056 return 0;
2057 }
2058
2059 if (!quiet_warnings)
2060 {
2061 if (!intel_syntax
2062 && ((i.types[0] & JumpAbsolute)
2063 != (t->operand_types[0] & JumpAbsolute)))
2064 {
2065 as_warn (_("indirect %s without `*'"), t->name);
2066 }
2067
2068 if ((t->opcode_modifier & (IsPrefix | IgnoreSize))
2069 == (IsPrefix | IgnoreSize))
2070 {
2071 /* Warn them that a data or address size prefix doesn't
2072 affect assembly of the next line of code. */
2073 as_warn (_("stand-alone `%s' prefix"), t->name);
2074 }
2075 }
2076
2077 /* Copy the template we found. */
2078 i.tm = *t;
2079 if (found_reverse_match)
2080 {
2081 /* If we found a reverse match we must alter the opcode
2082 direction bit. found_reverse_match holds bits to change
2083 (different for int & float insns). */
2084
2085 i.tm.base_opcode ^= found_reverse_match;
2086
2087 i.tm.operand_types[0] = t->operand_types[1];
2088 i.tm.operand_types[1] = t->operand_types[0];
2089 }
2090
2091 return 1;
2092 }
2093
2094 static int
2095 check_string ()
2096 {
2097 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
2098 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
2099 {
2100 if (i.seg[0] != NULL && i.seg[0] != &es)
2101 {
2102 as_bad (_("`%s' operand %d must use `%%es' segment"),
2103 i.tm.name,
2104 mem_op + 1);
2105 return 0;
2106 }
2107 /* There's only ever one segment override allowed per instruction.
2108 This instruction possibly has a legal segment override on the
2109 second operand, so copy the segment to where non-string
2110 instructions store it, allowing common code. */
2111 i.seg[0] = i.seg[1];
2112 }
2113 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
2114 {
2115 if (i.seg[1] != NULL && i.seg[1] != &es)
2116 {
2117 as_bad (_("`%s' operand %d must use `%%es' segment"),
2118 i.tm.name,
2119 mem_op + 2);
2120 return 0;
2121 }
2122 }
2123 return 1;
2124 }
2125
2126 static int
2127 process_suffix ()
2128 {
2129 /* If matched instruction specifies an explicit instruction mnemonic
2130 suffix, use it. */
2131 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
2132 {
2133 if (i.tm.opcode_modifier & Size16)
2134 i.suffix = WORD_MNEM_SUFFIX;
2135 else if (i.tm.opcode_modifier & Size64)
2136 i.suffix = QWORD_MNEM_SUFFIX;
2137 else
2138 i.suffix = LONG_MNEM_SUFFIX;
2139 }
2140 else if (i.reg_operands)
2141 {
2142 /* If there's no instruction mnemonic suffix we try to invent one
2143 based on register operands. */
2144 if (!i.suffix)
2145 {
2146 /* We take i.suffix from the last register operand specified,
2147 Destination register type is more significant than source
2148 register type. */
2149 int op;
2150 for (op = i.operands; --op >= 0;)
2151 if ((i.types[op] & Reg)
2152 && !(i.tm.operand_types[op] & InOutPortReg))
2153 {
2154 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
2155 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
2156 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
2157 LONG_MNEM_SUFFIX);
2158 break;
2159 }
2160 }
2161 else if (i.suffix == BYTE_MNEM_SUFFIX)
2162 {
2163 if (!check_byte_reg ())
2164 return 0;
2165 }
2166 else if (i.suffix == LONG_MNEM_SUFFIX)
2167 {
2168 if (!check_long_reg ())
2169 return 0;
2170 }
2171 else if (i.suffix == QWORD_MNEM_SUFFIX)
2172 {
2173 if (!check_qword_reg ())
2174 return 0;
2175 }
2176 else if (i.suffix == WORD_MNEM_SUFFIX)
2177 {
2178 if (!check_word_reg ())
2179 return 0;
2180 }
2181 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2182 /* Do nothing if the instruction is going to ignore the prefix. */
2183 ;
2184 else
2185 abort ();
2186 }
2187 else if ((i.tm.opcode_modifier & DefaultSize) && !i.suffix)
2188 {
2189 i.suffix = stackop_size;
2190 }
2191
2192 /* Change the opcode based on the operand size given by i.suffix;
2193 We need not change things for byte insns. */
2194
2195 if (!i.suffix && (i.tm.opcode_modifier & W))
2196 {
2197 as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
2198 return 0;
2199 }
2200
2201 /* For movzx and movsx, need to check the register type. */
2202 if (intel_syntax
2203 && (i.tm.base_opcode == 0xfb6 || i.tm.base_opcode == 0xfbe))
2204 if (i.suffix && i.suffix == BYTE_MNEM_SUFFIX)
2205 {
2206 unsigned int prefix = DATA_PREFIX_OPCODE;
2207
2208 if ((i.op[1].regs->reg_type & Reg16) != 0)
2209 if (!add_prefix (prefix))
2210 return 0;
2211 }
2212
2213 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2214 {
2215 /* It's not a byte, select word/dword operation. */
2216 if (i.tm.opcode_modifier & W)
2217 {
2218 if (i.tm.opcode_modifier & ShortForm)
2219 i.tm.base_opcode |= 8;
2220 else
2221 i.tm.base_opcode |= 1;
2222 }
2223 /* Now select between word & dword operations via the operand
2224 size prefix, except for instructions that will ignore this
2225 prefix anyway. */
2226 if (i.suffix != QWORD_MNEM_SUFFIX
2227 && (i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2228 && !(i.tm.opcode_modifier & IgnoreSize))
2229 {
2230 unsigned int prefix = DATA_PREFIX_OPCODE;
2231 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2232 prefix = ADDR_PREFIX_OPCODE;
2233
2234 if (!add_prefix (prefix))
2235 return 0;
2236 }
2237
2238 if (i.suffix != QWORD_MNEM_SUFFIX && (flag_code == CODE_64BIT)
2239 && !(i.tm.opcode_modifier & IgnoreSize)
2240 && (i.tm.opcode_modifier & JumpByte))
2241 {
2242 if (!add_prefix (ADDR_PREFIX_OPCODE))
2243 return 0;
2244 }
2245
2246 /* Set mode64 for an operand. */
2247 if (i.suffix == QWORD_MNEM_SUFFIX
2248 && (i.tm.opcode_modifier & NoRex64) == 0)
2249 {
2250 i.rex |= REX_MODE64;
2251 if (flag_code < CODE_64BIT)
2252 {
2253 as_bad (_("64bit operations available only in 64bit modes."));
2254 return 0;
2255 }
2256 }
2257
2258 /* Size floating point instruction. */
2259 if (i.suffix == LONG_MNEM_SUFFIX)
2260 {
2261 if (i.tm.opcode_modifier & FloatMF)
2262 i.tm.base_opcode ^= 4;
2263 }
2264 }
2265
2266 return 1;
2267 }
2268
2269 static int
2270 check_byte_reg ()
2271 {
2272 int op;
2273 for (op = i.operands; --op >= 0;)
2274 {
2275 /* If this is an eight bit register, it's OK. If it's the 16 or
2276 32 bit version of an eight bit register, we will just use the
2277 low portion, and that's OK too. */
2278 if (i.types[op] & Reg8)
2279 continue;
2280
2281 /* movzx and movsx should not generate this warning. */
2282 if (intel_syntax
2283 && (i.tm.base_opcode == 0xfb7
2284 || i.tm.base_opcode == 0xfb6
2285 || i.tm.base_opcode == 0x63
2286 || i.tm.base_opcode == 0xfbe
2287 || i.tm.base_opcode == 0xfbf))
2288 continue;
2289
2290 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4
2291 #if 0
2292 /* Check that the template allows eight bit regs. This
2293 kills insns such as `orb $1,%edx', which maybe should be
2294 allowed. */
2295 && (i.tm.operand_types[op] & (Reg8 | InOutPortReg))
2296 #endif
2297 )
2298 {
2299 /* Prohibit these changes in the 64bit mode, since the
2300 lowering is more complicated. */
2301 if (flag_code == CODE_64BIT
2302 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2303 {
2304 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2305 i.op[op].regs->reg_name,
2306 i.suffix);
2307 return 0;
2308 }
2309 #if REGISTER_WARNINGS
2310 if (!quiet_warnings
2311 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2312 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2313 (i.op[op].regs + (i.types[op] & Reg16
2314 ? REGNAM_AL - REGNAM_AX
2315 : REGNAM_AL - REGNAM_EAX))->reg_name,
2316 i.op[op].regs->reg_name,
2317 i.suffix);
2318 #endif
2319 continue;
2320 }
2321 /* Any other register is bad. */
2322 if (i.types[op] & (Reg | RegMMX | RegXMM
2323 | SReg2 | SReg3
2324 | Control | Debug | Test
2325 | FloatReg | FloatAcc))
2326 {
2327 as_bad (_("`%%%s' not allowed with `%s%c'"),
2328 i.op[op].regs->reg_name,
2329 i.tm.name,
2330 i.suffix);
2331 return 0;
2332 }
2333 }
2334 return 1;
2335 }
2336
2337 static int
2338 check_long_reg ()
2339 {
2340 int op;
2341
2342 for (op = i.operands; --op >= 0;)
2343 /* Reject eight bit registers, except where the template requires
2344 them. (eg. movzb) */
2345 if ((i.types[op] & Reg8) != 0
2346 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2347 {
2348 as_bad (_("`%%%s' not allowed with `%s%c'"),
2349 i.op[op].regs->reg_name,
2350 i.tm.name,
2351 i.suffix);
2352 return 0;
2353 }
2354 /* Warn if the e prefix on a general reg is missing. */
2355 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2356 && (i.types[op] & Reg16) != 0
2357 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2358 {
2359 /* Prohibit these changes in the 64bit mode, since the
2360 lowering is more complicated. */
2361 if (flag_code == CODE_64BIT)
2362 {
2363 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2364 i.op[op].regs->reg_name,
2365 i.suffix);
2366 return 0;
2367 }
2368 #if REGISTER_WARNINGS
2369 else
2370 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2371 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
2372 i.op[op].regs->reg_name,
2373 i.suffix);
2374 #endif
2375 }
2376 /* Warn if the r prefix on a general reg is missing. */
2377 else if ((i.types[op] & Reg64) != 0
2378 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2379 {
2380 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2381 i.op[op].regs->reg_name,
2382 i.suffix);
2383 return 0;
2384 }
2385 return 1;
2386 }
2387
2388 static int
2389 check_qword_reg ()
2390 {
2391 int op;
2392
2393 for (op = i.operands; --op >= 0; )
2394 /* Reject eight bit registers, except where the template requires
2395 them. (eg. movzb) */
2396 if ((i.types[op] & Reg8) != 0
2397 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2398 {
2399 as_bad (_("`%%%s' not allowed with `%s%c'"),
2400 i.op[op].regs->reg_name,
2401 i.tm.name,
2402 i.suffix);
2403 return 0;
2404 }
2405 /* Warn if the e prefix on a general reg is missing. */
2406 else if (((i.types[op] & Reg16) != 0
2407 || (i.types[op] & Reg32) != 0)
2408 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2409 {
2410 /* Prohibit these changes in the 64bit mode, since the
2411 lowering is more complicated. */
2412 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2413 i.op[op].regs->reg_name,
2414 i.suffix);
2415 return 0;
2416 }
2417 return 1;
2418 }
2419
2420 static int
2421 check_word_reg ()
2422 {
2423 int op;
2424 for (op = i.operands; --op >= 0;)
2425 /* Reject eight bit registers, except where the template requires
2426 them. (eg. movzb) */
2427 if ((i.types[op] & Reg8) != 0
2428 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2429 {
2430 as_bad (_("`%%%s' not allowed with `%s%c'"),
2431 i.op[op].regs->reg_name,
2432 i.tm.name,
2433 i.suffix);
2434 return 0;
2435 }
2436 /* Warn if the e prefix on a general reg is present. */
2437 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2438 && (i.types[op] & Reg32) != 0
2439 && (i.tm.operand_types[op] & (Reg16 | Acc)) != 0)
2440 {
2441 /* Prohibit these changes in the 64bit mode, since the
2442 lowering is more complicated. */
2443 if (flag_code == CODE_64BIT)
2444 {
2445 as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
2446 i.op[op].regs->reg_name,
2447 i.suffix);
2448 return 0;
2449 }
2450 else
2451 #if REGISTER_WARNINGS
2452 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2453 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
2454 i.op[op].regs->reg_name,
2455 i.suffix);
2456 #endif
2457 }
2458 return 1;
2459 }
2460
2461 static int
2462 finalize_imm ()
2463 {
2464 unsigned int overlap0, overlap1, overlap2;
2465
2466 overlap0 = i.types[0] & i.tm.operand_types[0];
2467 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S))
2468 && overlap0 != Imm8 && overlap0 != Imm8S
2469 && overlap0 != Imm16 && overlap0 != Imm32S
2470 && overlap0 != Imm32 && overlap0 != Imm64)
2471 {
2472 if (i.suffix)
2473 {
2474 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX
2475 ? Imm8 | Imm8S
2476 : (i.suffix == WORD_MNEM_SUFFIX
2477 ? Imm16
2478 : (i.suffix == QWORD_MNEM_SUFFIX
2479 ? Imm64 | Imm32S
2480 : Imm32)));
2481 }
2482 else if (overlap0 == (Imm16 | Imm32S | Imm32)
2483 || overlap0 == (Imm16 | Imm32)
2484 || overlap0 == (Imm16 | Imm32S))
2485 {
2486 overlap0 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
2487 ? Imm16 : Imm32S);
2488 }
2489 if (overlap0 != Imm8 && overlap0 != Imm8S
2490 && overlap0 != Imm16 && overlap0 != Imm32S
2491 && overlap0 != Imm32 && overlap0 != Imm64)
2492 {
2493 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
2494 return 0;
2495 }
2496 }
2497 i.types[0] = overlap0;
2498
2499 overlap1 = i.types[1] & i.tm.operand_types[1];
2500 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32))
2501 && overlap1 != Imm8 && overlap1 != Imm8S
2502 && overlap1 != Imm16 && overlap1 != Imm32S
2503 && overlap1 != Imm32 && overlap1 != Imm64)
2504 {
2505 if (i.suffix)
2506 {
2507 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX
2508 ? Imm8 | Imm8S
2509 : (i.suffix == WORD_MNEM_SUFFIX
2510 ? Imm16
2511 : (i.suffix == QWORD_MNEM_SUFFIX
2512 ? Imm64 | Imm32S
2513 : Imm32)));
2514 }
2515 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
2516 || overlap1 == (Imm16 | Imm32)
2517 || overlap1 == (Imm16 | Imm32S))
2518 {
2519 overlap1 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
2520 ? Imm16 : Imm32S);
2521 }
2522 if (overlap1 != Imm8 && overlap1 != Imm8S
2523 && overlap1 != Imm16 && overlap1 != Imm32S
2524 && overlap1 != Imm32 && overlap1 != Imm64)
2525 {
2526 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
2527 return 0;
2528 }
2529 }
2530 i.types[1] = overlap1;
2531
2532 overlap2 = i.types[2] & i.tm.operand_types[2];
2533 assert ((overlap2 & Imm) == 0);
2534 i.types[2] = overlap2;
2535
2536 return 1;
2537 }
2538
2539 static int
2540 process_operands ()
2541 {
2542 /* Default segment register this instruction will use for memory
2543 accesses. 0 means unknown. This is only for optimizing out
2544 unnecessary segment overrides. */
2545 const seg_entry *default_seg = 0;
2546
2547 /* The imul $imm, %reg instruction is converted into
2548 imul $imm, %reg, %reg, and the clr %reg instruction
2549 is converted into xor %reg, %reg. */
2550 if (i.tm.opcode_modifier & regKludge)
2551 {
2552 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
2553 /* Pretend we saw the extra register operand. */
2554 assert (i.op[first_reg_op + 1].regs == 0);
2555 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
2556 i.types[first_reg_op + 1] = i.types[first_reg_op];
2557 i.reg_operands = 2;
2558 }
2559
2560 if (i.tm.opcode_modifier & ShortForm)
2561 {
2562 /* The register or float register operand is in operand 0 or 1. */
2563 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
2564 /* Register goes in low 3 bits of opcode. */
2565 i.tm.base_opcode |= i.op[op].regs->reg_num;
2566 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2567 i.rex |= REX_EXTZ;
2568 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2569 {
2570 /* Warn about some common errors, but press on regardless.
2571 The first case can be generated by gcc (<= 2.8.1). */
2572 if (i.operands == 2)
2573 {
2574 /* Reversed arguments on faddp, fsubp, etc. */
2575 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
2576 i.op[1].regs->reg_name,
2577 i.op[0].regs->reg_name);
2578 }
2579 else
2580 {
2581 /* Extraneous `l' suffix on fp insn. */
2582 as_warn (_("translating to `%s %%%s'"), i.tm.name,
2583 i.op[0].regs->reg_name);
2584 }
2585 }
2586 }
2587 else if (i.tm.opcode_modifier & Modrm)
2588 {
2589 /* The opcode is completed (modulo i.tm.extension_opcode which
2590 must be put into the modrm byte).
2591 Now, we make the modrm & index base bytes based on all the
2592 info we've collected. */
2593
2594 default_seg = build_modrm_byte ();
2595 }
2596 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
2597 {
2598 if (i.tm.base_opcode == POP_SEG_SHORT
2599 && i.op[0].regs->reg_num == 1)
2600 {
2601 as_bad (_("you can't `pop %%cs'"));
2602 return 0;
2603 }
2604 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
2605 if ((i.op[0].regs->reg_flags & RegRex) != 0)
2606 i.rex |= REX_EXTZ;
2607 }
2608 else if ((i.tm.base_opcode & ~(D | W)) == MOV_AX_DISP32)
2609 {
2610 default_seg = &ds;
2611 }
2612 else if ((i.tm.opcode_modifier & IsString) != 0)
2613 {
2614 /* For the string instructions that allow a segment override
2615 on one of their operands, the default segment is ds. */
2616 default_seg = &ds;
2617 }
2618
2619 /* If a segment was explicitly specified,
2620 and the specified segment is not the default,
2621 use an opcode prefix to select it.
2622 If we never figured out what the default segment is,
2623 then default_seg will be zero at this point,
2624 and the specified segment prefix will always be used. */
2625 if ((i.seg[0]) && (i.seg[0] != default_seg))
2626 {
2627 if (!add_prefix (i.seg[0]->seg_prefix))
2628 return 0;
2629 }
2630 return 1;
2631 }
2632
2633 static const seg_entry *
2634 build_modrm_byte ()
2635 {
2636 const seg_entry *default_seg = 0;
2637
2638 /* i.reg_operands MUST be the number of real register operands;
2639 implicit registers do not count. */
2640 if (i.reg_operands == 2)
2641 {
2642 unsigned int source, dest;
2643 source = ((i.types[0]
2644 & (Reg | RegMMX | RegXMM
2645 | SReg2 | SReg3
2646 | Control | Debug | Test))
2647 ? 0 : 1);
2648 dest = source + 1;
2649
2650 i.rm.mode = 3;
2651 /* One of the register operands will be encoded in the i.tm.reg
2652 field, the other in the combined i.tm.mode and i.tm.regmem
2653 fields. If no form of this instruction supports a memory
2654 destination operand, then we assume the source operand may
2655 sometimes be a memory operand and so we need to store the
2656 destination in the i.rm.reg field. */
2657 if ((i.tm.operand_types[dest] & AnyMem) == 0)
2658 {
2659 i.rm.reg = i.op[dest].regs->reg_num;
2660 i.rm.regmem = i.op[source].regs->reg_num;
2661 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
2662 i.rex |= REX_EXTX;
2663 if ((i.op[source].regs->reg_flags & RegRex) != 0)
2664 i.rex |= REX_EXTZ;
2665 }
2666 else
2667 {
2668 i.rm.reg = i.op[source].regs->reg_num;
2669 i.rm.regmem = i.op[dest].regs->reg_num;
2670 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
2671 i.rex |= REX_EXTZ;
2672 if ((i.op[source].regs->reg_flags & RegRex) != 0)
2673 i.rex |= REX_EXTX;
2674 }
2675 }
2676 else
2677 { /* If it's not 2 reg operands... */
2678 if (i.mem_operands)
2679 {
2680 unsigned int fake_zero_displacement = 0;
2681 unsigned int op = ((i.types[0] & AnyMem)
2682 ? 0
2683 : (i.types[1] & AnyMem) ? 1 : 2);
2684
2685 default_seg = &ds;
2686
2687 if (i.base_reg == 0)
2688 {
2689 i.rm.mode = 0;
2690 if (!i.disp_operands)
2691 fake_zero_displacement = 1;
2692 if (i.index_reg == 0)
2693 {
2694 /* Operand is just <disp> */
2695 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)
2696 && (flag_code != CODE_64BIT))
2697 {
2698 i.rm.regmem = NO_BASE_REGISTER_16;
2699 i.types[op] &= ~Disp;
2700 i.types[op] |= Disp16;
2701 }
2702 else if (flag_code != CODE_64BIT
2703 || (i.prefix[ADDR_PREFIX] != 0))
2704 {
2705 i.rm.regmem = NO_BASE_REGISTER;
2706 i.types[op] &= ~Disp;
2707 i.types[op] |= Disp32;
2708 }
2709 else
2710 {
2711 /* 64bit mode overwrites the 32bit absolute
2712 addressing by RIP relative addressing and
2713 absolute addressing is encoded by one of the
2714 redundant SIB forms. */
2715 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2716 i.sib.base = NO_BASE_REGISTER;
2717 i.sib.index = NO_INDEX_REGISTER;
2718 i.types[op] &= ~Disp;
2719 i.types[op] |= Disp32S;
2720 }
2721 }
2722 else /* !i.base_reg && i.index_reg */
2723 {
2724 i.sib.index = i.index_reg->reg_num;
2725 i.sib.base = NO_BASE_REGISTER;
2726 i.sib.scale = i.log2_scale_factor;
2727 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2728 i.types[op] &= ~Disp;
2729 if (flag_code != CODE_64BIT)
2730 i.types[op] |= Disp32; /* Must be 32 bit */
2731 else
2732 i.types[op] |= Disp32S;
2733 if ((i.index_reg->reg_flags & RegRex) != 0)
2734 i.rex |= REX_EXTY;
2735 }
2736 }
2737 /* RIP addressing for 64bit mode. */
2738 else if (i.base_reg->reg_type == BaseIndex)
2739 {
2740 i.rm.regmem = NO_BASE_REGISTER;
2741 i.types[op] &= ~Disp;
2742 i.types[op] |= Disp32S;
2743 i.flags[op] = Operand_PCrel;
2744 }
2745 else if (i.base_reg->reg_type & Reg16)
2746 {
2747 switch (i.base_reg->reg_num)
2748 {
2749 case 3: /* (%bx) */
2750 if (i.index_reg == 0)
2751 i.rm.regmem = 7;
2752 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
2753 i.rm.regmem = i.index_reg->reg_num - 6;
2754 break;
2755 case 5: /* (%bp) */
2756 default_seg = &ss;
2757 if (i.index_reg == 0)
2758 {
2759 i.rm.regmem = 6;
2760 if ((i.types[op] & Disp) == 0)
2761 {
2762 /* fake (%bp) into 0(%bp) */
2763 i.types[op] |= Disp8;
2764 fake_zero_displacement = 1;
2765 }
2766 }
2767 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
2768 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
2769 break;
2770 default: /* (%si) -> 4 or (%di) -> 5 */
2771 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
2772 }
2773 i.rm.mode = mode_from_disp_size (i.types[op]);
2774 }
2775 else /* i.base_reg and 32/64 bit mode */
2776 {
2777 if (flag_code == CODE_64BIT
2778 && (i.types[op] & Disp))
2779 {
2780 if (i.types[op] & Disp8)
2781 i.types[op] = Disp8 | Disp32S;
2782 else
2783 i.types[op] = Disp32S;
2784 }
2785 i.rm.regmem = i.base_reg->reg_num;
2786 if ((i.base_reg->reg_flags & RegRex) != 0)
2787 i.rex |= REX_EXTZ;
2788 i.sib.base = i.base_reg->reg_num;
2789 /* x86-64 ignores REX prefix bit here to avoid decoder
2790 complications. */
2791 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
2792 {
2793 default_seg = &ss;
2794 if (i.disp_operands == 0)
2795 {
2796 fake_zero_displacement = 1;
2797 i.types[op] |= Disp8;
2798 }
2799 }
2800 else if (i.base_reg->reg_num == ESP_REG_NUM)
2801 {
2802 default_seg = &ss;
2803 }
2804 i.sib.scale = i.log2_scale_factor;
2805 if (i.index_reg == 0)
2806 {
2807 /* <disp>(%esp) becomes two byte modrm with no index
2808 register. We've already stored the code for esp
2809 in i.rm.regmem ie. ESCAPE_TO_TWO_BYTE_ADDRESSING.
2810 Any base register besides %esp will not use the
2811 extra modrm byte. */
2812 i.sib.index = NO_INDEX_REGISTER;
2813 #if !SCALE1_WHEN_NO_INDEX
2814 /* Another case where we force the second modrm byte. */
2815 if (i.log2_scale_factor)
2816 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2817 #endif
2818 }
2819 else
2820 {
2821 i.sib.index = i.index_reg->reg_num;
2822 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2823 if ((i.index_reg->reg_flags & RegRex) != 0)
2824 i.rex |= REX_EXTY;
2825 }
2826 i.rm.mode = mode_from_disp_size (i.types[op]);
2827 }
2828
2829 if (fake_zero_displacement)
2830 {
2831 /* Fakes a zero displacement assuming that i.types[op]
2832 holds the correct displacement size. */
2833 expressionS *exp;
2834
2835 assert (i.op[op].disps == 0);
2836 exp = &disp_expressions[i.disp_operands++];
2837 i.op[op].disps = exp;
2838 exp->X_op = O_constant;
2839 exp->X_add_number = 0;
2840 exp->X_add_symbol = (symbolS *) 0;
2841 exp->X_op_symbol = (symbolS *) 0;
2842 }
2843 }
2844
2845 /* Fill in i.rm.reg or i.rm.regmem field with register operand
2846 (if any) based on i.tm.extension_opcode. Again, we must be
2847 careful to make sure that segment/control/debug/test/MMX
2848 registers are coded into the i.rm.reg field. */
2849 if (i.reg_operands)
2850 {
2851 unsigned int op =
2852 ((i.types[0]
2853 & (Reg | RegMMX | RegXMM
2854 | SReg2 | SReg3
2855 | Control | Debug | Test))
2856 ? 0
2857 : ((i.types[1]
2858 & (Reg | RegMMX | RegXMM
2859 | SReg2 | SReg3
2860 | Control | Debug | Test))
2861 ? 1
2862 : 2));
2863 /* If there is an extension opcode to put here, the register
2864 number must be put into the regmem field. */
2865 if (i.tm.extension_opcode != None)
2866 {
2867 i.rm.regmem = i.op[op].regs->reg_num;
2868 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2869 i.rex |= REX_EXTZ;
2870 }
2871 else
2872 {
2873 i.rm.reg = i.op[op].regs->reg_num;
2874 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2875 i.rex |= REX_EXTX;
2876 }
2877
2878 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2 we
2879 must set it to 3 to indicate this is a register operand
2880 in the regmem field. */
2881 if (!i.mem_operands)
2882 i.rm.mode = 3;
2883 }
2884
2885 /* Fill in i.rm.reg field with extension opcode (if any). */
2886 if (i.tm.extension_opcode != None)
2887 i.rm.reg = i.tm.extension_opcode;
2888 }
2889 return default_seg;
2890 }
2891
2892 static void
2893 output_branch ()
2894 {
2895 char *p;
2896 int code16;
2897 int prefix;
2898 relax_substateT subtype;
2899 symbolS *sym;
2900 offsetT off;
2901
2902 code16 = 0;
2903 if (flag_code == CODE_16BIT)
2904 code16 = CODE16;
2905
2906 prefix = 0;
2907 if (i.prefix[DATA_PREFIX] != 0)
2908 {
2909 prefix = 1;
2910 i.prefixes -= 1;
2911 code16 ^= CODE16;
2912 }
2913 /* Pentium4 branch hints. */
2914 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
2915 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
2916 {
2917 prefix++;
2918 i.prefixes--;
2919 }
2920 if (i.prefix[REX_PREFIX] != 0)
2921 {
2922 prefix++;
2923 i.prefixes--;
2924 }
2925
2926 if (i.prefixes != 0 && !intel_syntax)
2927 as_warn (_("skipping prefixes on this instruction"));
2928
2929 /* It's always a symbol; End frag & setup for relax.
2930 Make sure there is enough room in this frag for the largest
2931 instruction we may generate in md_convert_frag. This is 2
2932 bytes for the opcode and room for the prefix and largest
2933 displacement. */
2934 frag_grow (prefix + 2 + 4);
2935 /* Prefix and 1 opcode byte go in fr_fix. */
2936 p = frag_more (prefix + 1);
2937 if (i.prefix[DATA_PREFIX] != 0)
2938 *p++ = DATA_PREFIX_OPCODE;
2939 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
2940 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
2941 *p++ = i.prefix[SEG_PREFIX];
2942 if (i.prefix[REX_PREFIX] != 0)
2943 *p++ = i.prefix[REX_PREFIX];
2944 *p = i.tm.base_opcode;
2945
2946 if ((unsigned char) *p == JUMP_PC_RELATIVE)
2947 subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
2948 else if ((cpu_arch_flags & Cpu386) != 0)
2949 subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
2950 else
2951 subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
2952 subtype |= code16;
2953
2954 sym = i.op[0].disps->X_add_symbol;
2955 off = i.op[0].disps->X_add_number;
2956
2957 if (i.op[0].disps->X_op != O_constant
2958 && i.op[0].disps->X_op != O_symbol)
2959 {
2960 /* Handle complex expressions. */
2961 sym = make_expr_symbol (i.op[0].disps);
2962 off = 0;
2963 }
2964
2965 /* 1 possible extra opcode + 4 byte displacement go in var part.
2966 Pass reloc in fr_var. */
2967 frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
2968 }
2969
2970 static void
2971 output_jump ()
2972 {
2973 char *p;
2974 int size;
2975
2976 if (i.tm.opcode_modifier & JumpByte)
2977 {
2978 /* This is a loop or jecxz type instruction. */
2979 size = 1;
2980 if (i.prefix[ADDR_PREFIX] != 0)
2981 {
2982 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
2983 i.prefixes -= 1;
2984 }
2985 /* Pentium4 branch hints. */
2986 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
2987 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
2988 {
2989 FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
2990 i.prefixes--;
2991 }
2992 }
2993 else
2994 {
2995 int code16;
2996
2997 code16 = 0;
2998 if (flag_code == CODE_16BIT)
2999 code16 = CODE16;
3000
3001 if (i.prefix[DATA_PREFIX] != 0)
3002 {
3003 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
3004 i.prefixes -= 1;
3005 code16 ^= CODE16;
3006 }
3007
3008 size = 4;
3009 if (code16)
3010 size = 2;
3011 }
3012
3013 if (i.prefix[REX_PREFIX] != 0)
3014 {
3015 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
3016 i.prefixes -= 1;
3017 }
3018
3019 if (i.prefixes != 0 && !intel_syntax)
3020 as_warn (_("skipping prefixes on this instruction"));
3021
3022 p = frag_more (1 + size);
3023 *p++ = i.tm.base_opcode;
3024
3025 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3026 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
3027 }
3028
3029 static void
3030 output_interseg_jump ()
3031 {
3032 char *p;
3033 int size;
3034 int prefix;
3035 int code16;
3036
3037 code16 = 0;
3038 if (flag_code == CODE_16BIT)
3039 code16 = CODE16;
3040
3041 prefix = 0;
3042 if (i.prefix[DATA_PREFIX] != 0)
3043 {
3044 prefix = 1;
3045 i.prefixes -= 1;
3046 code16 ^= CODE16;
3047 }
3048 if (i.prefix[REX_PREFIX] != 0)
3049 {
3050 prefix++;
3051 i.prefixes -= 1;
3052 }
3053
3054 size = 4;
3055 if (code16)
3056 size = 2;
3057
3058 if (i.prefixes != 0 && !intel_syntax)
3059 as_warn (_("skipping prefixes on this instruction"));
3060
3061 /* 1 opcode; 2 segment; offset */
3062 p = frag_more (prefix + 1 + 2 + size);
3063
3064 if (i.prefix[DATA_PREFIX] != 0)
3065 *p++ = DATA_PREFIX_OPCODE;
3066
3067 if (i.prefix[REX_PREFIX] != 0)
3068 *p++ = i.prefix[REX_PREFIX];
3069
3070 *p++ = i.tm.base_opcode;
3071 if (i.op[1].imms->X_op == O_constant)
3072 {
3073 offsetT n = i.op[1].imms->X_add_number;
3074
3075 if (size == 2
3076 && !fits_in_unsigned_word (n)
3077 && !fits_in_signed_word (n))
3078 {
3079 as_bad (_("16-bit jump out of range"));
3080 return;
3081 }
3082 md_number_to_chars (p, n, size);
3083 }
3084 else
3085 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3086 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
3087 if (i.op[0].imms->X_op != O_constant)
3088 as_bad (_("can't handle non absolute segment in `%s'"),
3089 i.tm.name);
3090 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
3091 }
3092
3093 static void
3094 output_insn ()
3095 {
3096 /* Tie dwarf2 debug info to the address at the start of the insn.
3097 We can't do this after the insn has been output as the current
3098 frag may have been closed off. eg. by frag_var. */
3099 dwarf2_emit_insn (0);
3100
3101 /* Output jumps. */
3102 if (i.tm.opcode_modifier & Jump)
3103 output_branch ();
3104 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
3105 output_jump ();
3106 else if (i.tm.opcode_modifier & JumpInterSegment)
3107 output_interseg_jump ();
3108 else
3109 {
3110 /* Output normal instructions here. */
3111 char *p;
3112 unsigned char *q;
3113
3114 /* All opcodes on i386 have either 1 or 2 bytes. We may use third
3115 byte for the SSE instructions to specify a prefix they require. */
3116 if (i.tm.base_opcode & 0xff0000)
3117 add_prefix ((i.tm.base_opcode >> 16) & 0xff);
3118
3119 /* The prefix bytes. */
3120 for (q = i.prefix;
3121 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
3122 q++)
3123 {
3124 if (*q)
3125 {
3126 p = frag_more (1);
3127 md_number_to_chars (p, (valueT) *q, 1);
3128 }
3129 }
3130
3131 /* Now the opcode; be careful about word order here! */
3132 if (fits_in_unsigned_byte (i.tm.base_opcode))
3133 {
3134 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
3135 }
3136 else
3137 {
3138 p = frag_more (2);
3139 /* Put out high byte first: can't use md_number_to_chars! */
3140 *p++ = (i.tm.base_opcode >> 8) & 0xff;
3141 *p = i.tm.base_opcode & 0xff;
3142 }
3143
3144 /* Now the modrm byte and sib byte (if present). */
3145 if (i.tm.opcode_modifier & Modrm)
3146 {
3147 p = frag_more (1);
3148 md_number_to_chars (p,
3149 (valueT) (i.rm.regmem << 0
3150 | i.rm.reg << 3
3151 | i.rm.mode << 6),
3152 1);
3153 /* If i.rm.regmem == ESP (4)
3154 && i.rm.mode != (Register mode)
3155 && not 16 bit
3156 ==> need second modrm byte. */
3157 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
3158 && i.rm.mode != 3
3159 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
3160 {
3161 p = frag_more (1);
3162 md_number_to_chars (p,
3163 (valueT) (i.sib.base << 0
3164 | i.sib.index << 3
3165 | i.sib.scale << 6),
3166 1);
3167 }
3168 }
3169
3170 if (i.disp_operands)
3171 output_disp ();
3172
3173 if (i.imm_operands)
3174 output_imm ();
3175 }
3176
3177 #ifdef DEBUG386
3178 if (flag_debug)
3179 {
3180 pi (line, &i);
3181 }
3182 #endif /* DEBUG386 */
3183 }
3184
3185 static void
3186 output_disp ()
3187 {
3188 char *p;
3189 unsigned int n;
3190
3191 for (n = 0; n < i.operands; n++)
3192 {
3193 if (i.types[n] & Disp)
3194 {
3195 if (i.op[n].disps->X_op == O_constant)
3196 {
3197 int size;
3198 offsetT val;
3199
3200 size = 4;
3201 if (i.types[n] & (Disp8 | Disp16 | Disp64))
3202 {
3203 size = 2;
3204 if (i.types[n] & Disp8)
3205 size = 1;
3206 if (i.types[n] & Disp64)
3207 size = 8;
3208 }
3209 val = offset_in_range (i.op[n].disps->X_add_number,
3210 size);
3211 p = frag_more (size);
3212 md_number_to_chars (p, val, size);
3213 }
3214 else
3215 {
3216 int size = 4;
3217 int sign = 0;
3218 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
3219
3220 /* The PC relative address is computed relative
3221 to the instruction boundary, so in case immediate
3222 fields follows, we need to adjust the value. */
3223 if (pcrel && i.imm_operands)
3224 {
3225 int imm_size = 4;
3226 unsigned int n1;
3227
3228 for (n1 = 0; n1 < i.operands; n1++)
3229 if (i.types[n1] & Imm)
3230 {
3231 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
3232 {
3233 imm_size = 2;
3234 if (i.types[n1] & (Imm8 | Imm8S))
3235 imm_size = 1;
3236 if (i.types[n1] & Imm64)
3237 imm_size = 8;
3238 }
3239 break;
3240 }
3241 /* We should find the immediate. */
3242 if (n1 == i.operands)
3243 abort ();
3244 i.op[n].disps->X_add_number -= imm_size;
3245 }
3246
3247 if (i.types[n] & Disp32S)
3248 sign = 1;
3249
3250 if (i.types[n] & (Disp16 | Disp64))
3251 {
3252 size = 2;
3253 if (i.types[n] & Disp64)
3254 size = 8;
3255 }
3256
3257 p = frag_more (size);
3258 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3259 i.op[n].disps, pcrel,
3260 reloc (size, pcrel, sign, i.reloc[n]));
3261 }
3262 }
3263 }
3264 }
3265
3266 static void
3267 output_imm ()
3268 {
3269 char *p;
3270 unsigned int n;
3271
3272 for (n = 0; n < i.operands; n++)
3273 {
3274 if (i.types[n] & Imm)
3275 {
3276 if (i.op[n].imms->X_op == O_constant)
3277 {
3278 int size;
3279 offsetT val;
3280
3281 size = 4;
3282 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3283 {
3284 size = 2;
3285 if (i.types[n] & (Imm8 | Imm8S))
3286 size = 1;
3287 else if (i.types[n] & Imm64)
3288 size = 8;
3289 }
3290 val = offset_in_range (i.op[n].imms->X_add_number,
3291 size);
3292 p = frag_more (size);
3293 md_number_to_chars (p, val, size);
3294 }
3295 else
3296 {
3297 /* Not absolute_section.
3298 Need a 32-bit fixup (don't support 8bit
3299 non-absolute imms). Try to support other
3300 sizes ... */
3301 RELOC_ENUM reloc_type;
3302 int size = 4;
3303 int sign = 0;
3304
3305 if ((i.types[n] & (Imm32S))
3306 && i.suffix == QWORD_MNEM_SUFFIX)
3307 sign = 1;
3308 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3309 {
3310 size = 2;
3311 if (i.types[n] & (Imm8 | Imm8S))
3312 size = 1;
3313 if (i.types[n] & Imm64)
3314 size = 8;
3315 }
3316
3317 p = frag_more (size);
3318 reloc_type = reloc (size, 0, sign, i.reloc[n]);
3319 #ifdef BFD_ASSEMBLER
3320 if (reloc_type == BFD_RELOC_32
3321 && GOT_symbol
3322 && GOT_symbol == i.op[n].imms->X_add_symbol
3323 && (i.op[n].imms->X_op == O_symbol
3324 || (i.op[n].imms->X_op == O_add
3325 && ((symbol_get_value_expression
3326 (i.op[n].imms->X_op_symbol)->X_op)
3327 == O_subtract))))
3328 {
3329 /* We don't support dynamic linking on x86-64 yet. */
3330 if (flag_code == CODE_64BIT)
3331 abort ();
3332 reloc_type = BFD_RELOC_386_GOTPC;
3333 i.op[n].imms->X_add_number += 3;
3334 }
3335 #endif
3336 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3337 i.op[n].imms, 0, reloc_type);
3338 }
3339 }
3340 }
3341 }
3342 \f
3343 #ifndef LEX_AT
3344 static char *lex_got PARAMS ((RELOC_ENUM *, int *));
3345
3346 /* Parse operands of the form
3347 <symbol>@GOTOFF+<nnn>
3348 and similar .plt or .got references.
3349
3350 If we find one, set up the correct relocation in RELOC and copy the
3351 input string, minus the `@GOTOFF' into a malloc'd buffer for
3352 parsing by the calling routine. Return this buffer, and if ADJUST
3353 is non-null set it to the length of the string we removed from the
3354 input line. Otherwise return NULL. */
3355 static char *
3356 lex_got (reloc, adjust)
3357 RELOC_ENUM *reloc;
3358 int *adjust;
3359 {
3360 static const char * const mode_name[NUM_FLAG_CODE] = { "32", "16", "64" };
3361 static const struct {
3362 const char *str;
3363 const RELOC_ENUM rel[NUM_FLAG_CODE];
3364 } gotrel[] = {
3365 { "PLT", { BFD_RELOC_386_PLT32, 0, BFD_RELOC_X86_64_PLT32 } },
3366 { "GOTOFF", { BFD_RELOC_386_GOTOFF, 0, 0 } },
3367 { "GOTPCREL", { 0, 0, BFD_RELOC_X86_64_GOTPCREL } },
3368 { "GOT", { BFD_RELOC_386_GOT32, 0, BFD_RELOC_X86_64_GOT32 } }
3369 };
3370 char *cp;
3371 unsigned int j;
3372
3373 for (cp = input_line_pointer; *cp != '@'; cp++)
3374 if (is_end_of_line[(unsigned char) *cp])
3375 return NULL;
3376
3377 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
3378 {
3379 int len;
3380
3381 len = strlen (gotrel[j].str);
3382 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
3383 {
3384 if (gotrel[j].rel[(unsigned int) flag_code] != 0)
3385 {
3386 int first, second;
3387 char *tmpbuf, *past_reloc;
3388
3389 *reloc = gotrel[j].rel[(unsigned int) flag_code];
3390 if (adjust)
3391 *adjust = len;
3392
3393 if (GOT_symbol == NULL)
3394 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3395
3396 /* Replace the relocation token with ' ', so that
3397 errors like foo@GOTOFF1 will be detected. */
3398
3399 /* The length of the first part of our input line. */
3400 first = cp - input_line_pointer;
3401
3402 /* The second part goes from after the reloc token until
3403 (and including) an end_of_line char. Don't use strlen
3404 here as the end_of_line char may not be a NUL. */
3405 past_reloc = cp + 1 + len;
3406 for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
3407 ;
3408 second = cp - past_reloc;
3409
3410 /* Allocate and copy string. The trailing NUL shouldn't
3411 be necessary, but be safe. */
3412 tmpbuf = xmalloc (first + second + 2);
3413 memcpy (tmpbuf, input_line_pointer, first);
3414 tmpbuf[first] = ' ';
3415 memcpy (tmpbuf + first + 1, past_reloc, second);
3416 tmpbuf[first + second + 1] = '\0';
3417 return tmpbuf;
3418 }
3419
3420 as_bad (_("@%s reloc is not supported in %s bit mode"),
3421 gotrel[j].str, mode_name[(unsigned int) flag_code]);
3422 return NULL;
3423 }
3424 }
3425
3426 /* Might be a symbol version string. Don't as_bad here. */
3427 return NULL;
3428 }
3429
3430 /* x86_cons_fix_new is called via the expression parsing code when a
3431 reloc is needed. We use this hook to get the correct .got reloc. */
3432 static RELOC_ENUM got_reloc = NO_RELOC;
3433
3434 void
3435 x86_cons_fix_new (frag, off, len, exp)
3436 fragS *frag;
3437 unsigned int off;
3438 unsigned int len;
3439 expressionS *exp;
3440 {
3441 RELOC_ENUM r = reloc (len, 0, 0, got_reloc);
3442 got_reloc = NO_RELOC;
3443 fix_new_exp (frag, off, len, exp, 0, r);
3444 }
3445
3446 void
3447 x86_cons (exp, size)
3448 expressionS *exp;
3449 int size;
3450 {
3451 if (size == 4)
3452 {
3453 /* Handle @GOTOFF and the like in an expression. */
3454 char *save;
3455 char *gotfree_input_line;
3456 int adjust;
3457
3458 save = input_line_pointer;
3459 gotfree_input_line = lex_got (&got_reloc, &adjust);
3460 if (gotfree_input_line)
3461 input_line_pointer = gotfree_input_line;
3462
3463 expression (exp);
3464
3465 if (gotfree_input_line)
3466 {
3467 /* expression () has merrily parsed up to the end of line,
3468 or a comma - in the wrong buffer. Transfer how far
3469 input_line_pointer has moved to the right buffer. */
3470 input_line_pointer = (save
3471 + (input_line_pointer - gotfree_input_line)
3472 + adjust);
3473 free (gotfree_input_line);
3474 }
3475 }
3476 else
3477 expression (exp);
3478 }
3479 #endif
3480
3481 static int i386_immediate PARAMS ((char *));
3482
3483 static int
3484 i386_immediate (imm_start)
3485 char *imm_start;
3486 {
3487 char *save_input_line_pointer;
3488 #ifndef LEX_AT
3489 char *gotfree_input_line;
3490 #endif
3491 segT exp_seg = 0;
3492 expressionS *exp;
3493
3494 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
3495 {
3496 as_bad (_("only 1 or 2 immediate operands are allowed"));
3497 return 0;
3498 }
3499
3500 exp = &im_expressions[i.imm_operands++];
3501 i.op[this_operand].imms = exp;
3502
3503 if (is_space_char (*imm_start))
3504 ++imm_start;
3505
3506 save_input_line_pointer = input_line_pointer;
3507 input_line_pointer = imm_start;
3508
3509 #ifndef LEX_AT
3510 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3511 if (gotfree_input_line)
3512 input_line_pointer = gotfree_input_line;
3513 #endif
3514
3515 exp_seg = expression (exp);
3516
3517 SKIP_WHITESPACE ();
3518 if (*input_line_pointer)
3519 as_bad (_("junk `%s' after expression"), input_line_pointer);
3520
3521 input_line_pointer = save_input_line_pointer;
3522 #ifndef LEX_AT
3523 if (gotfree_input_line)
3524 free (gotfree_input_line);
3525 #endif
3526
3527 if (exp->X_op == O_absent || exp->X_op == O_big)
3528 {
3529 /* Missing or bad expr becomes absolute 0. */
3530 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
3531 imm_start);
3532 exp->X_op = O_constant;
3533 exp->X_add_number = 0;
3534 exp->X_add_symbol = (symbolS *) 0;
3535 exp->X_op_symbol = (symbolS *) 0;
3536 }
3537 else if (exp->X_op == O_constant)
3538 {
3539 /* Size it properly later. */
3540 i.types[this_operand] |= Imm64;
3541 /* If BFD64, sign extend val. */
3542 if (!use_rela_relocations)
3543 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
3544 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
3545 }
3546 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3547 else if (1
3548 #ifdef BFD_ASSEMBLER
3549 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3550 #endif
3551 && exp_seg != text_section
3552 && exp_seg != data_section
3553 && exp_seg != bss_section
3554 && exp_seg != undefined_section
3555 #ifdef BFD_ASSEMBLER
3556 && !bfd_is_com_section (exp_seg)
3557 #endif
3558 )
3559 {
3560 #ifdef BFD_ASSEMBLER
3561 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3562 #else
3563 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3564 #endif
3565 return 0;
3566 }
3567 #endif
3568 else
3569 {
3570 /* This is an address. The size of the address will be
3571 determined later, depending on destination register,
3572 suffix, or the default for the section. */
3573 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
3574 }
3575
3576 return 1;
3577 }
3578
3579 static char *i386_scale PARAMS ((char *));
3580
3581 static char *
3582 i386_scale (scale)
3583 char *scale;
3584 {
3585 offsetT val;
3586 char *save = input_line_pointer;
3587
3588 input_line_pointer = scale;
3589 val = get_absolute_expression ();
3590
3591 switch (val)
3592 {
3593 case 0:
3594 case 1:
3595 i.log2_scale_factor = 0;
3596 break;
3597 case 2:
3598 i.log2_scale_factor = 1;
3599 break;
3600 case 4:
3601 i.log2_scale_factor = 2;
3602 break;
3603 case 8:
3604 i.log2_scale_factor = 3;
3605 break;
3606 default:
3607 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
3608 scale);
3609 input_line_pointer = save;
3610 return NULL;
3611 }
3612 if (i.log2_scale_factor != 0 && i.index_reg == 0)
3613 {
3614 as_warn (_("scale factor of %d without an index register"),
3615 1 << i.log2_scale_factor);
3616 #if SCALE1_WHEN_NO_INDEX
3617 i.log2_scale_factor = 0;
3618 #endif
3619 }
3620 scale = input_line_pointer;
3621 input_line_pointer = save;
3622 return scale;
3623 }
3624
3625 static int i386_displacement PARAMS ((char *, char *));
3626
3627 static int
3628 i386_displacement (disp_start, disp_end)
3629 char *disp_start;
3630 char *disp_end;
3631 {
3632 expressionS *exp;
3633 segT exp_seg = 0;
3634 char *save_input_line_pointer;
3635 #ifndef LEX_AT
3636 char *gotfree_input_line;
3637 #endif
3638 int bigdisp = Disp32;
3639
3640 if (flag_code == CODE_64BIT)
3641 {
3642 if (i.prefix[ADDR_PREFIX] == 0)
3643 bigdisp = Disp64;
3644 }
3645 else if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3646 bigdisp = Disp16;
3647 i.types[this_operand] |= bigdisp;
3648
3649 exp = &disp_expressions[i.disp_operands];
3650 i.op[this_operand].disps = exp;
3651 i.disp_operands++;
3652 save_input_line_pointer = input_line_pointer;
3653 input_line_pointer = disp_start;
3654 END_STRING_AND_SAVE (disp_end);
3655
3656 #ifndef GCC_ASM_O_HACK
3657 #define GCC_ASM_O_HACK 0
3658 #endif
3659 #if GCC_ASM_O_HACK
3660 END_STRING_AND_SAVE (disp_end + 1);
3661 if ((i.types[this_operand] & BaseIndex) != 0
3662 && displacement_string_end[-1] == '+')
3663 {
3664 /* This hack is to avoid a warning when using the "o"
3665 constraint within gcc asm statements.
3666 For instance:
3667
3668 #define _set_tssldt_desc(n,addr,limit,type) \
3669 __asm__ __volatile__ ( \
3670 "movw %w2,%0\n\t" \
3671 "movw %w1,2+%0\n\t" \
3672 "rorl $16,%1\n\t" \
3673 "movb %b1,4+%0\n\t" \
3674 "movb %4,5+%0\n\t" \
3675 "movb $0,6+%0\n\t" \
3676 "movb %h1,7+%0\n\t" \
3677 "rorl $16,%1" \
3678 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
3679
3680 This works great except that the output assembler ends
3681 up looking a bit weird if it turns out that there is
3682 no offset. You end up producing code that looks like:
3683
3684 #APP
3685 movw $235,(%eax)
3686 movw %dx,2+(%eax)
3687 rorl $16,%edx
3688 movb %dl,4+(%eax)
3689 movb $137,5+(%eax)
3690 movb $0,6+(%eax)
3691 movb %dh,7+(%eax)
3692 rorl $16,%edx
3693 #NO_APP
3694
3695 So here we provide the missing zero. */
3696
3697 *displacement_string_end = '0';
3698 }
3699 #endif
3700 #ifndef LEX_AT
3701 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3702 if (gotfree_input_line)
3703 input_line_pointer = gotfree_input_line;
3704 #endif
3705
3706 exp_seg = expression (exp);
3707
3708 SKIP_WHITESPACE ();
3709 if (*input_line_pointer)
3710 as_bad (_("junk `%s' after expression"), input_line_pointer);
3711 #if GCC_ASM_O_HACK
3712 RESTORE_END_STRING (disp_end + 1);
3713 #endif
3714 RESTORE_END_STRING (disp_end);
3715 input_line_pointer = save_input_line_pointer;
3716 #ifndef LEX_AT
3717 if (gotfree_input_line)
3718 free (gotfree_input_line);
3719 #endif
3720
3721 #ifdef BFD_ASSEMBLER
3722 /* We do this to make sure that the section symbol is in
3723 the symbol table. We will ultimately change the relocation
3724 to be relative to the beginning of the section. */
3725 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
3726 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3727 {
3728 if (exp->X_op != O_symbol)
3729 {
3730 as_bad (_("bad expression used with @%s"),
3731 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
3732 ? "GOTPCREL"
3733 : "GOTOFF"));
3734 return 0;
3735 }
3736
3737 if (S_IS_LOCAL (exp->X_add_symbol)
3738 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
3739 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
3740 exp->X_op = O_subtract;
3741 exp->X_op_symbol = GOT_symbol;
3742 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
3743 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
3744 else
3745 i.reloc[this_operand] = BFD_RELOC_32;
3746 }
3747 #endif
3748
3749 if (exp->X_op == O_absent || exp->X_op == O_big)
3750 {
3751 /* Missing or bad expr becomes absolute 0. */
3752 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
3753 disp_start);
3754 exp->X_op = O_constant;
3755 exp->X_add_number = 0;
3756 exp->X_add_symbol = (symbolS *) 0;
3757 exp->X_op_symbol = (symbolS *) 0;
3758 }
3759
3760 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3761 if (exp->X_op != O_constant
3762 #ifdef BFD_ASSEMBLER
3763 && OUTPUT_FLAVOR == bfd_target_aout_flavour
3764 #endif
3765 && exp_seg != text_section
3766 && exp_seg != data_section
3767 && exp_seg != bss_section
3768 && exp_seg != undefined_section)
3769 {
3770 #ifdef BFD_ASSEMBLER
3771 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3772 #else
3773 as_bad (_("unimplemented segment type %d in operand"), exp_seg);
3774 #endif
3775 return 0;
3776 }
3777 #endif
3778 else if (flag_code == CODE_64BIT)
3779 i.types[this_operand] |= Disp32S | Disp32;
3780 return 1;
3781 }
3782
3783 static int i386_index_check PARAMS ((const char *));
3784
3785 /* Make sure the memory operand we've been dealt is valid.
3786 Return 1 on success, 0 on a failure. */
3787
3788 static int
3789 i386_index_check (operand_string)
3790 const char *operand_string;
3791 {
3792 int ok;
3793 #if INFER_ADDR_PREFIX
3794 int fudged = 0;
3795
3796 tryprefix:
3797 #endif
3798 ok = 1;
3799 if (flag_code == CODE_64BIT)
3800 {
3801 if (i.prefix[ADDR_PREFIX] == 0)
3802 {
3803 /* 64bit checks. */
3804 if ((i.base_reg
3805 && ((i.base_reg->reg_type & Reg64) == 0)
3806 && (i.base_reg->reg_type != BaseIndex
3807 || i.index_reg))
3808 || (i.index_reg
3809 && ((i.index_reg->reg_type & (Reg64 | BaseIndex))
3810 != (Reg64 | BaseIndex))))
3811 ok = 0;
3812 }
3813 else
3814 {
3815 /* 32bit checks. */
3816 if ((i.base_reg
3817 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
3818 || (i.index_reg
3819 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
3820 != (Reg32 | BaseIndex))))
3821 ok = 0;
3822 }
3823 }
3824 else
3825 {
3826 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
3827 {
3828 /* 16bit checks. */
3829 if ((i.base_reg
3830 && ((i.base_reg->reg_type & (Reg16 | BaseIndex | RegRex))
3831 != (Reg16 | BaseIndex)))
3832 || (i.index_reg
3833 && (((i.index_reg->reg_type & (Reg16 | BaseIndex))
3834 != (Reg16 | BaseIndex))
3835 || !(i.base_reg
3836 && i.base_reg->reg_num < 6
3837 && i.index_reg->reg_num >= 6
3838 && i.log2_scale_factor == 0))))
3839 ok = 0;
3840 }
3841 else
3842 {
3843 /* 32bit checks. */
3844 if ((i.base_reg
3845 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
3846 || (i.index_reg
3847 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
3848 != (Reg32 | BaseIndex))))
3849 ok = 0;
3850 }
3851 }
3852 if (!ok)
3853 {
3854 #if INFER_ADDR_PREFIX
3855 if (flag_code != CODE_64BIT
3856 && i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0')
3857 {
3858 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
3859 i.prefixes += 1;
3860 /* Change the size of any displacement too. At most one of
3861 Disp16 or Disp32 is set.
3862 FIXME. There doesn't seem to be any real need for separate
3863 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
3864 Removing them would probably clean up the code quite a lot. */
3865 if (i.types[this_operand] & (Disp16 | Disp32))
3866 i.types[this_operand] ^= (Disp16 | Disp32);
3867 fudged = 1;
3868 goto tryprefix;
3869 }
3870 if (fudged)
3871 as_bad (_("`%s' is not a valid base/index expression"),
3872 operand_string);
3873 else
3874 #endif
3875 as_bad (_("`%s' is not a valid %s bit base/index expression"),
3876 operand_string,
3877 flag_code_names[flag_code]);
3878 return 0;
3879 }
3880 return 1;
3881 }
3882
3883 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
3884 on error. */
3885
3886 static int
3887 i386_operand (operand_string)
3888 char *operand_string;
3889 {
3890 const reg_entry *r;
3891 char *end_op;
3892 char *op_string = operand_string;
3893
3894 if (is_space_char (*op_string))
3895 ++op_string;
3896
3897 /* We check for an absolute prefix (differentiating,
3898 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
3899 if (*op_string == ABSOLUTE_PREFIX)
3900 {
3901 ++op_string;
3902 if (is_space_char (*op_string))
3903 ++op_string;
3904 i.types[this_operand] |= JumpAbsolute;
3905 }
3906
3907 /* Check if operand is a register. */
3908 if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
3909 && (r = parse_register (op_string, &end_op)) != NULL)
3910 {
3911 /* Check for a segment override by searching for ':' after a
3912 segment register. */
3913 op_string = end_op;
3914 if (is_space_char (*op_string))
3915 ++op_string;
3916 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
3917 {
3918 switch (r->reg_num)
3919 {
3920 case 0:
3921 i.seg[i.mem_operands] = &es;
3922 break;
3923 case 1:
3924 i.seg[i.mem_operands] = &cs;
3925 break;
3926 case 2:
3927 i.seg[i.mem_operands] = &ss;
3928 break;
3929 case 3:
3930 i.seg[i.mem_operands] = &ds;
3931 break;
3932 case 4:
3933 i.seg[i.mem_operands] = &fs;
3934 break;
3935 case 5:
3936 i.seg[i.mem_operands] = &gs;
3937 break;
3938 }
3939
3940 /* Skip the ':' and whitespace. */
3941 ++op_string;
3942 if (is_space_char (*op_string))
3943 ++op_string;
3944
3945 if (!is_digit_char (*op_string)
3946 && !is_identifier_char (*op_string)
3947 && *op_string != '('
3948 && *op_string != ABSOLUTE_PREFIX)
3949 {
3950 as_bad (_("bad memory operand `%s'"), op_string);
3951 return 0;
3952 }
3953 /* Handle case of %es:*foo. */
3954 if (*op_string == ABSOLUTE_PREFIX)
3955 {
3956 ++op_string;
3957 if (is_space_char (*op_string))
3958 ++op_string;
3959 i.types[this_operand] |= JumpAbsolute;
3960 }
3961 goto do_memory_reference;
3962 }
3963 if (*op_string)
3964 {
3965 as_bad (_("junk `%s' after register"), op_string);
3966 return 0;
3967 }
3968 i.types[this_operand] |= r->reg_type & ~BaseIndex;
3969 i.op[this_operand].regs = r;
3970 i.reg_operands++;
3971 }
3972 else if (*op_string == REGISTER_PREFIX)
3973 {
3974 as_bad (_("bad register name `%s'"), op_string);
3975 return 0;
3976 }
3977 else if (*op_string == IMMEDIATE_PREFIX)
3978 {
3979 ++op_string;
3980 if (i.types[this_operand] & JumpAbsolute)
3981 {
3982 as_bad (_("immediate operand illegal with absolute jump"));
3983 return 0;
3984 }
3985 if (!i386_immediate (op_string))
3986 return 0;
3987 }
3988 else if (is_digit_char (*op_string)
3989 || is_identifier_char (*op_string)
3990 || *op_string == '(')
3991 {
3992 /* This is a memory reference of some sort. */
3993 char *base_string;
3994
3995 /* Start and end of displacement string expression (if found). */
3996 char *displacement_string_start;
3997 char *displacement_string_end;
3998
3999 do_memory_reference:
4000 if ((i.mem_operands == 1
4001 && (current_templates->start->opcode_modifier & IsString) == 0)
4002 || i.mem_operands == 2)
4003 {
4004 as_bad (_("too many memory references for `%s'"),
4005 current_templates->start->name);
4006 return 0;
4007 }
4008
4009 /* Check for base index form. We detect the base index form by
4010 looking for an ')' at the end of the operand, searching
4011 for the '(' matching it, and finding a REGISTER_PREFIX or ','
4012 after the '('. */
4013 base_string = op_string + strlen (op_string);
4014
4015 --base_string;
4016 if (is_space_char (*base_string))
4017 --base_string;
4018
4019 /* If we only have a displacement, set-up for it to be parsed later. */
4020 displacement_string_start = op_string;
4021 displacement_string_end = base_string + 1;
4022
4023 if (*base_string == ')')
4024 {
4025 char *temp_string;
4026 unsigned int parens_balanced = 1;
4027 /* We've already checked that the number of left & right ()'s are
4028 equal, so this loop will not be infinite. */
4029 do
4030 {
4031 base_string--;
4032 if (*base_string == ')')
4033 parens_balanced++;
4034 if (*base_string == '(')
4035 parens_balanced--;
4036 }
4037 while (parens_balanced);
4038
4039 temp_string = base_string;
4040
4041 /* Skip past '(' and whitespace. */
4042 ++base_string;
4043 if (is_space_char (*base_string))
4044 ++base_string;
4045
4046 if (*base_string == ','
4047 || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
4048 && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
4049 {
4050 displacement_string_end = temp_string;
4051
4052 i.types[this_operand] |= BaseIndex;
4053
4054 if (i.base_reg)
4055 {
4056 base_string = end_op;
4057 if (is_space_char (*base_string))
4058 ++base_string;
4059 }
4060
4061 /* There may be an index reg or scale factor here. */
4062 if (*base_string == ',')
4063 {
4064 ++base_string;
4065 if (is_space_char (*base_string))
4066 ++base_string;
4067
4068 if ((*base_string == REGISTER_PREFIX || allow_naked_reg)
4069 && (i.index_reg = parse_register (base_string, &end_op)) != NULL)
4070 {
4071 base_string = end_op;
4072 if (is_space_char (*base_string))
4073 ++base_string;
4074 if (*base_string == ',')
4075 {
4076 ++base_string;
4077 if (is_space_char (*base_string))
4078 ++base_string;
4079 }
4080 else if (*base_string != ')')
4081 {
4082 as_bad (_("expecting `,' or `)' after index register in `%s'"),
4083 operand_string);
4084 return 0;
4085 }
4086 }
4087 else if (*base_string == REGISTER_PREFIX)
4088 {
4089 as_bad (_("bad register name `%s'"), base_string);
4090 return 0;
4091 }
4092
4093 /* Check for scale factor. */
4094 if (*base_string != ')')
4095 {
4096 char *end_scale = i386_scale (base_string);
4097
4098 if (!end_scale)
4099 return 0;
4100
4101 base_string = end_scale;
4102 if (is_space_char (*base_string))
4103 ++base_string;
4104 if (*base_string != ')')
4105 {
4106 as_bad (_("expecting `)' after scale factor in `%s'"),
4107 operand_string);
4108 return 0;
4109 }
4110 }
4111 else if (!i.index_reg)
4112 {
4113 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
4114 *base_string);
4115 return 0;
4116 }
4117 }
4118 else if (*base_string != ')')
4119 {
4120 as_bad (_("expecting `,' or `)' after base register in `%s'"),
4121 operand_string);
4122 return 0;
4123 }
4124 }
4125 else if (*base_string == REGISTER_PREFIX)
4126 {
4127 as_bad (_("bad register name `%s'"), base_string);
4128 return 0;
4129 }
4130 }
4131
4132 /* If there's an expression beginning the operand, parse it,
4133 assuming displacement_string_start and
4134 displacement_string_end are meaningful. */
4135 if (displacement_string_start != displacement_string_end)
4136 {
4137 if (!i386_displacement (displacement_string_start,
4138 displacement_string_end))
4139 return 0;
4140 }
4141
4142 /* Special case for (%dx) while doing input/output op. */
4143 if (i.base_reg
4144 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
4145 && i.index_reg == 0
4146 && i.log2_scale_factor == 0
4147 && i.seg[i.mem_operands] == 0
4148 && (i.types[this_operand] & Disp) == 0)
4149 {
4150 i.types[this_operand] = InOutPortReg;
4151 return 1;
4152 }
4153
4154 if (i386_index_check (operand_string) == 0)
4155 return 0;
4156 i.mem_operands++;
4157 }
4158 else
4159 {
4160 /* It's not a memory operand; argh! */
4161 as_bad (_("invalid char %s beginning operand %d `%s'"),
4162 output_invalid (*op_string),
4163 this_operand + 1,
4164 op_string);
4165 return 0;
4166 }
4167 return 1; /* Normal return. */
4168 }
4169 \f
4170 /* md_estimate_size_before_relax()
4171
4172 Called just before relax() for rs_machine_dependent frags. The x86
4173 assembler uses these frags to handle variable size jump
4174 instructions.
4175
4176 Any symbol that is now undefined will not become defined.
4177 Return the correct fr_subtype in the frag.
4178 Return the initial "guess for variable size of frag" to caller.
4179 The guess is actually the growth beyond the fixed part. Whatever
4180 we do to grow the fixed or variable part contributes to our
4181 returned value. */
4182
4183 int
4184 md_estimate_size_before_relax (fragP, segment)
4185 fragS *fragP;
4186 segT segment;
4187 {
4188 /* We've already got fragP->fr_subtype right; all we have to do is
4189 check for un-relaxable symbols. On an ELF system, we can't relax
4190 an externally visible symbol, because it may be overridden by a
4191 shared library. */
4192 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
4193 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4194 || S_IS_EXTERNAL (fragP->fr_symbol)
4195 || S_IS_WEAK (fragP->fr_symbol)
4196 #endif
4197 )
4198 {
4199 /* Symbol is undefined in this segment, or we need to keep a
4200 reloc so that weak symbols can be overridden. */
4201 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
4202 RELOC_ENUM reloc_type;
4203 unsigned char *opcode;
4204 int old_fr_fix;
4205
4206 if (fragP->fr_var != NO_RELOC)
4207 reloc_type = fragP->fr_var;
4208 else if (size == 2)
4209 reloc_type = BFD_RELOC_16_PCREL;
4210 else
4211 reloc_type = BFD_RELOC_32_PCREL;
4212
4213 old_fr_fix = fragP->fr_fix;
4214 opcode = (unsigned char *) fragP->fr_opcode;
4215
4216 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
4217 {
4218 case UNCOND_JUMP:
4219 /* Make jmp (0xeb) a (d)word displacement jump. */
4220 opcode[0] = 0xe9;
4221 fragP->fr_fix += size;
4222 fix_new (fragP, old_fr_fix, size,
4223 fragP->fr_symbol,
4224 fragP->fr_offset, 1,
4225 reloc_type);
4226 break;
4227
4228 case COND_JUMP86:
4229 if (no_cond_jump_promotion)
4230 goto relax_guess;
4231
4232 if (size == 2)
4233 {
4234 /* Negate the condition, and branch past an
4235 unconditional jump. */
4236 opcode[0] ^= 1;
4237 opcode[1] = 3;
4238 /* Insert an unconditional jump. */
4239 opcode[2] = 0xe9;
4240 /* We added two extra opcode bytes, and have a two byte
4241 offset. */
4242 fragP->fr_fix += 2 + 2;
4243 fix_new (fragP, old_fr_fix + 2, 2,
4244 fragP->fr_symbol,
4245 fragP->fr_offset, 1,
4246 reloc_type);
4247 break;
4248 }
4249 /* Fall through. */
4250
4251 case COND_JUMP:
4252 if (no_cond_jump_promotion)
4253 goto relax_guess;
4254
4255 /* This changes the byte-displacement jump 0x7N
4256 to the (d)word-displacement jump 0x0f,0x8N. */
4257 opcode[1] = opcode[0] + 0x10;
4258 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4259 /* We've added an opcode byte. */
4260 fragP->fr_fix += 1 + size;
4261 fix_new (fragP, old_fr_fix + 1, size,
4262 fragP->fr_symbol,
4263 fragP->fr_offset, 1,
4264 reloc_type);
4265 break;
4266
4267 default:
4268 BAD_CASE (fragP->fr_subtype);
4269 break;
4270 }
4271 frag_wane (fragP);
4272 return fragP->fr_fix - old_fr_fix;
4273 }
4274
4275 relax_guess:
4276 /* Guess size depending on current relax state. Initially the relax
4277 state will correspond to a short jump and we return 1, because
4278 the variable part of the frag (the branch offset) is one byte
4279 long. However, we can relax a section more than once and in that
4280 case we must either set fr_subtype back to the unrelaxed state,
4281 or return the value for the appropriate branch. */
4282 return md_relax_table[fragP->fr_subtype].rlx_length;
4283 }
4284
4285 /* Called after relax() is finished.
4286
4287 In: Address of frag.
4288 fr_type == rs_machine_dependent.
4289 fr_subtype is what the address relaxed to.
4290
4291 Out: Any fixSs and constants are set up.
4292 Caller will turn frag into a ".space 0". */
4293
4294 #ifndef BFD_ASSEMBLER
4295 void
4296 md_convert_frag (headers, sec, fragP)
4297 object_headers *headers ATTRIBUTE_UNUSED;
4298 segT sec ATTRIBUTE_UNUSED;
4299 fragS *fragP;
4300 #else
4301 void
4302 md_convert_frag (abfd, sec, fragP)
4303 bfd *abfd ATTRIBUTE_UNUSED;
4304 segT sec ATTRIBUTE_UNUSED;
4305 fragS *fragP;
4306 #endif
4307 {
4308 unsigned char *opcode;
4309 unsigned char *where_to_put_displacement = NULL;
4310 offsetT target_address;
4311 offsetT opcode_address;
4312 unsigned int extension = 0;
4313 offsetT displacement_from_opcode_start;
4314
4315 opcode = (unsigned char *) fragP->fr_opcode;
4316
4317 /* Address we want to reach in file space. */
4318 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
4319
4320 /* Address opcode resides at in file space. */
4321 opcode_address = fragP->fr_address + fragP->fr_fix;
4322
4323 /* Displacement from opcode start to fill into instruction. */
4324 displacement_from_opcode_start = target_address - opcode_address;
4325
4326 if ((fragP->fr_subtype & BIG) == 0)
4327 {
4328 /* Don't have to change opcode. */
4329 extension = 1; /* 1 opcode + 1 displacement */
4330 where_to_put_displacement = &opcode[1];
4331 }
4332 else
4333 {
4334 if (no_cond_jump_promotion
4335 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
4336 as_warn_where (fragP->fr_file, fragP->fr_line, _("long jump required"));
4337
4338 switch (fragP->fr_subtype)
4339 {
4340 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
4341 extension = 4; /* 1 opcode + 4 displacement */
4342 opcode[0] = 0xe9;
4343 where_to_put_displacement = &opcode[1];
4344 break;
4345
4346 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
4347 extension = 2; /* 1 opcode + 2 displacement */
4348 opcode[0] = 0xe9;
4349 where_to_put_displacement = &opcode[1];
4350 break;
4351
4352 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
4353 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
4354 extension = 5; /* 2 opcode + 4 displacement */
4355 opcode[1] = opcode[0] + 0x10;
4356 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4357 where_to_put_displacement = &opcode[2];
4358 break;
4359
4360 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
4361 extension = 3; /* 2 opcode + 2 displacement */
4362 opcode[1] = opcode[0] + 0x10;
4363 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4364 where_to_put_displacement = &opcode[2];
4365 break;
4366
4367 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
4368 extension = 4;
4369 opcode[0] ^= 1;
4370 opcode[1] = 3;
4371 opcode[2] = 0xe9;
4372 where_to_put_displacement = &opcode[3];
4373 break;
4374
4375 default:
4376 BAD_CASE (fragP->fr_subtype);
4377 break;
4378 }
4379 }
4380
4381 /* Now put displacement after opcode. */
4382 md_number_to_chars ((char *) where_to_put_displacement,
4383 (valueT) (displacement_from_opcode_start - extension),
4384 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
4385 fragP->fr_fix += extension;
4386 }
4387 \f
4388 /* Size of byte displacement jmp. */
4389 int md_short_jump_size = 2;
4390
4391 /* Size of dword displacement jmp. */
4392 int md_long_jump_size = 5;
4393
4394 /* Size of relocation record. */
4395 const int md_reloc_size = 8;
4396
4397 void
4398 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
4399 char *ptr;
4400 addressT from_addr, to_addr;
4401 fragS *frag ATTRIBUTE_UNUSED;
4402 symbolS *to_symbol ATTRIBUTE_UNUSED;
4403 {
4404 offsetT offset;
4405
4406 offset = to_addr - (from_addr + 2);
4407 /* Opcode for byte-disp jump. */
4408 md_number_to_chars (ptr, (valueT) 0xeb, 1);
4409 md_number_to_chars (ptr + 1, (valueT) offset, 1);
4410 }
4411
4412 void
4413 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
4414 char *ptr;
4415 addressT from_addr, to_addr;
4416 fragS *frag ATTRIBUTE_UNUSED;
4417 symbolS *to_symbol ATTRIBUTE_UNUSED;
4418 {
4419 offsetT offset;
4420
4421 offset = to_addr - (from_addr + 5);
4422 md_number_to_chars (ptr, (valueT) 0xe9, 1);
4423 md_number_to_chars (ptr + 1, (valueT) offset, 4);
4424 }
4425 \f
4426 /* Apply a fixup (fixS) to segment data, once it has been determined
4427 by our caller that we have all the info we need to fix it up.
4428
4429 On the 386, immediates, displacements, and data pointers are all in
4430 the same (little-endian) format, so we don't need to care about which
4431 we are handling. */
4432
4433 void
4434 md_apply_fix3 (fixP, valP, seg)
4435 /* The fix we're to put in. */
4436 fixS *fixP;
4437 /* Pointer to the value of the bits. */
4438 valueT * valP;
4439 /* Segment fix is from. */
4440 segT seg ATTRIBUTE_UNUSED;
4441 {
4442 char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
4443 valueT value = * valP;
4444
4445 #if defined (BFD_ASSEMBLER) && !defined (TE_Mach)
4446 if (fixP->fx_pcrel)
4447 {
4448 switch (fixP->fx_r_type)
4449 {
4450 default:
4451 break;
4452
4453 case BFD_RELOC_32:
4454 fixP->fx_r_type = BFD_RELOC_32_PCREL;
4455 break;
4456 case BFD_RELOC_16:
4457 fixP->fx_r_type = BFD_RELOC_16_PCREL;
4458 break;
4459 case BFD_RELOC_8:
4460 fixP->fx_r_type = BFD_RELOC_8_PCREL;
4461 break;
4462 }
4463 }
4464
4465 /* This is a hack. There should be a better way to handle this.
4466 This covers for the fact that bfd_install_relocation will
4467 subtract the current location (for partial_inplace, PC relative
4468 relocations); see more below. */
4469 if ((fixP->fx_r_type == BFD_RELOC_32_PCREL
4470 || fixP->fx_r_type == BFD_RELOC_16_PCREL
4471 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
4472 && fixP->fx_addsy && !use_rela_relocations)
4473 {
4474 #ifndef OBJ_AOUT
4475 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4476 #ifdef TE_PE
4477 || OUTPUT_FLAVOR == bfd_target_coff_flavour
4478 #endif
4479 )
4480 value += fixP->fx_where + fixP->fx_frag->fr_address;
4481 #endif
4482 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4483 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
4484 {
4485 segT fseg = S_GET_SEGMENT (fixP->fx_addsy);
4486
4487 if ((fseg == seg
4488 || (symbol_section_p (fixP->fx_addsy)
4489 && fseg != absolute_section))
4490 && !S_IS_EXTERNAL (fixP->fx_addsy)
4491 && !S_IS_WEAK (fixP->fx_addsy)
4492 && S_IS_DEFINED (fixP->fx_addsy)
4493 && !S_IS_COMMON (fixP->fx_addsy))
4494 {
4495 /* Yes, we add the values in twice. This is because
4496 bfd_perform_relocation subtracts them out again. I think
4497 bfd_perform_relocation is broken, but I don't dare change
4498 it. FIXME. */
4499 value += fixP->fx_where + fixP->fx_frag->fr_address;
4500 }
4501 }
4502 #endif
4503 #if defined (OBJ_COFF) && defined (TE_PE)
4504 /* For some reason, the PE format does not store a section
4505 address offset for a PC relative symbol. */
4506 if (S_GET_SEGMENT (fixP->fx_addsy) != seg)
4507 value += md_pcrel_from (fixP);
4508 #endif
4509 }
4510
4511 /* Fix a few things - the dynamic linker expects certain values here,
4512 and we must not dissappoint it. */
4513 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4514 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4515 && fixP->fx_addsy)
4516 switch (fixP->fx_r_type)
4517 {
4518 case BFD_RELOC_386_PLT32:
4519 case BFD_RELOC_X86_64_PLT32:
4520 /* Make the jump instruction point to the address of the operand. At
4521 runtime we merely add the offset to the actual PLT entry. */
4522 value = -4;
4523 break;
4524 case BFD_RELOC_386_GOTPC:
4525
4526 /* This is tough to explain. We end up with this one if we have
4527 * operands that look like "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal
4528 * here is to obtain the absolute address of the GOT, and it is strongly
4529 * preferable from a performance point of view to avoid using a runtime
4530 * relocation for this. The actual sequence of instructions often look
4531 * something like:
4532 *
4533 * call .L66
4534 * .L66:
4535 * popl %ebx
4536 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
4537 *
4538 * The call and pop essentially return the absolute address of
4539 * the label .L66 and store it in %ebx. The linker itself will
4540 * ultimately change the first operand of the addl so that %ebx points to
4541 * the GOT, but to keep things simple, the .o file must have this operand
4542 * set so that it generates not the absolute address of .L66, but the
4543 * absolute address of itself. This allows the linker itself simply
4544 * treat a GOTPC relocation as asking for a pcrel offset to the GOT to be
4545 * added in, and the addend of the relocation is stored in the operand
4546 * field for the instruction itself.
4547 *
4548 * Our job here is to fix the operand so that it would add the correct
4549 * offset so that %ebx would point to itself. The thing that is tricky is
4550 * that .-.L66 will point to the beginning of the instruction, so we need
4551 * to further modify the operand so that it will point to itself.
4552 * There are other cases where you have something like:
4553 *
4554 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
4555 *
4556 * and here no correction would be required. Internally in the assembler
4557 * we treat operands of this form as not being pcrel since the '.' is
4558 * explicitly mentioned, and I wonder whether it would simplify matters
4559 * to do it this way. Who knows. In earlier versions of the PIC patches,
4560 * the pcrel_adjust field was used to store the correction, but since the
4561 * expression is not pcrel, I felt it would be confusing to do it this
4562 * way. */
4563
4564 value -= 1;
4565 break;
4566 case BFD_RELOC_386_GOT32:
4567 case BFD_RELOC_X86_64_GOT32:
4568 value = 0; /* Fully resolved at runtime. No addend. */
4569 break;
4570 case BFD_RELOC_386_GOTOFF:
4571 case BFD_RELOC_X86_64_GOTPCREL:
4572 break;
4573
4574 case BFD_RELOC_VTABLE_INHERIT:
4575 case BFD_RELOC_VTABLE_ENTRY:
4576 fixP->fx_done = 0;
4577 return;
4578
4579 default:
4580 break;
4581 }
4582 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
4583 * valP = value;
4584 #endif /* defined (BFD_ASSEMBLER) && !defined (TE_Mach) */
4585
4586 /* Are we finished with this relocation now? */
4587 if (fixP->fx_addsy == NULL && fixP->fx_pcrel == 0)
4588 fixP->fx_done = 1;
4589 #ifdef BFD_ASSEMBLER
4590 else if (use_rela_relocations)
4591 {
4592 fixP->fx_no_overflow = 1;
4593 value = 0;
4594 }
4595 #endif
4596 md_number_to_chars (p, value, fixP->fx_size);
4597 }
4598 \f
4599 #define MAX_LITTLENUMS 6
4600
4601 /* Turn the string pointed to by litP into a floating point constant
4602 of type TYPE, and emit the appropriate bytes. The number of
4603 LITTLENUMS emitted is stored in *SIZEP. An error message is
4604 returned, or NULL on OK. */
4605
4606 char *
4607 md_atof (type, litP, sizeP)
4608 int type;
4609 char *litP;
4610 int *sizeP;
4611 {
4612 int prec;
4613 LITTLENUM_TYPE words[MAX_LITTLENUMS];
4614 LITTLENUM_TYPE *wordP;
4615 char *t;
4616
4617 switch (type)
4618 {
4619 case 'f':
4620 case 'F':
4621 prec = 2;
4622 break;
4623
4624 case 'd':
4625 case 'D':
4626 prec = 4;
4627 break;
4628
4629 case 'x':
4630 case 'X':
4631 prec = 5;
4632 break;
4633
4634 default:
4635 *sizeP = 0;
4636 return _("Bad call to md_atof ()");
4637 }
4638 t = atof_ieee (input_line_pointer, type, words);
4639 if (t)
4640 input_line_pointer = t;
4641
4642 *sizeP = prec * sizeof (LITTLENUM_TYPE);
4643 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
4644 the bigendian 386. */
4645 for (wordP = words + prec - 1; prec--;)
4646 {
4647 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
4648 litP += sizeof (LITTLENUM_TYPE);
4649 }
4650 return 0;
4651 }
4652 \f
4653 char output_invalid_buf[8];
4654
4655 static char *
4656 output_invalid (c)
4657 int c;
4658 {
4659 if (ISPRINT (c))
4660 sprintf (output_invalid_buf, "'%c'", c);
4661 else
4662 sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
4663 return output_invalid_buf;
4664 }
4665
4666 /* REG_STRING starts *before* REGISTER_PREFIX. */
4667
4668 static const reg_entry *
4669 parse_register (reg_string, end_op)
4670 char *reg_string;
4671 char **end_op;
4672 {
4673 char *s = reg_string;
4674 char *p;
4675 char reg_name_given[MAX_REG_NAME_SIZE + 1];
4676 const reg_entry *r;
4677
4678 /* Skip possible REGISTER_PREFIX and possible whitespace. */
4679 if (*s == REGISTER_PREFIX)
4680 ++s;
4681
4682 if (is_space_char (*s))
4683 ++s;
4684
4685 p = reg_name_given;
4686 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
4687 {
4688 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
4689 return (const reg_entry *) NULL;
4690 s++;
4691 }
4692
4693 /* For naked regs, make sure that we are not dealing with an identifier.
4694 This prevents confusing an identifier like `eax_var' with register
4695 `eax'. */
4696 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
4697 return (const reg_entry *) NULL;
4698
4699 *end_op = s;
4700
4701 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
4702
4703 /* Handle floating point regs, allowing spaces in the (i) part. */
4704 if (r == i386_regtab /* %st is first entry of table */)
4705 {
4706 if (is_space_char (*s))
4707 ++s;
4708 if (*s == '(')
4709 {
4710 ++s;
4711 if (is_space_char (*s))
4712 ++s;
4713 if (*s >= '0' && *s <= '7')
4714 {
4715 r = &i386_float_regtab[*s - '0'];
4716 ++s;
4717 if (is_space_char (*s))
4718 ++s;
4719 if (*s == ')')
4720 {
4721 *end_op = s + 1;
4722 return r;
4723 }
4724 }
4725 /* We have "%st(" then garbage. */
4726 return (const reg_entry *) NULL;
4727 }
4728 }
4729
4730 if (r != NULL
4731 && (r->reg_flags & (RegRex64 | RegRex)) != 0
4732 && flag_code != CODE_64BIT)
4733 {
4734 return (const reg_entry *) NULL;
4735 }
4736
4737 return r;
4738 }
4739 \f
4740 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4741 const char *md_shortopts = "kVQ:sq";
4742 #else
4743 const char *md_shortopts = "q";
4744 #endif
4745
4746 struct option md_longopts[] = {
4747 #define OPTION_32 (OPTION_MD_BASE + 0)
4748 {"32", no_argument, NULL, OPTION_32},
4749 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4750 #define OPTION_64 (OPTION_MD_BASE + 1)
4751 {"64", no_argument, NULL, OPTION_64},
4752 #endif
4753 {NULL, no_argument, NULL, 0}
4754 };
4755 size_t md_longopts_size = sizeof (md_longopts);
4756
4757 int
4758 md_parse_option (c, arg)
4759 int c;
4760 char *arg ATTRIBUTE_UNUSED;
4761 {
4762 switch (c)
4763 {
4764 case 'q':
4765 quiet_warnings = 1;
4766 break;
4767
4768 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4769 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
4770 should be emitted or not. FIXME: Not implemented. */
4771 case 'Q':
4772 break;
4773
4774 /* -V: SVR4 argument to print version ID. */
4775 case 'V':
4776 print_version_id ();
4777 break;
4778
4779 /* -k: Ignore for FreeBSD compatibility. */
4780 case 'k':
4781 break;
4782
4783 case 's':
4784 /* -s: On i386 Solaris, this tells the native assembler to use
4785 .stab instead of .stab.excl. We always use .stab anyhow. */
4786 break;
4787
4788 case OPTION_64:
4789 {
4790 const char **list, **l;
4791
4792 list = bfd_target_list ();
4793 for (l = list; *l != NULL; l++)
4794 if (strcmp (*l, "elf64-x86-64") == 0)
4795 {
4796 default_arch = "x86_64";
4797 break;
4798 }
4799 if (*l == NULL)
4800 as_fatal (_("No compiled in support for x86_64"));
4801 free (list);
4802 }
4803 break;
4804 #endif
4805
4806 case OPTION_32:
4807 default_arch = "i386";
4808 break;
4809
4810 default:
4811 return 0;
4812 }
4813 return 1;
4814 }
4815
4816 void
4817 md_show_usage (stream)
4818 FILE *stream;
4819 {
4820 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4821 fprintf (stream, _("\
4822 -Q ignored\n\
4823 -V print assembler version number\n\
4824 -k ignored\n\
4825 -q quieten some warnings\n\
4826 -s ignored\n"));
4827 #else
4828 fprintf (stream, _("\
4829 -q quieten some warnings\n"));
4830 #endif
4831 }
4832
4833 #ifdef BFD_ASSEMBLER
4834 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
4835 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
4836
4837 /* Pick the target format to use. */
4838
4839 const char *
4840 i386_target_format ()
4841 {
4842 if (!strcmp (default_arch, "x86_64"))
4843 set_code_flag (CODE_64BIT);
4844 else if (!strcmp (default_arch, "i386"))
4845 set_code_flag (CODE_32BIT);
4846 else
4847 as_fatal (_("Unknown architecture"));
4848 switch (OUTPUT_FLAVOR)
4849 {
4850 #ifdef OBJ_MAYBE_AOUT
4851 case bfd_target_aout_flavour:
4852 return AOUT_TARGET_FORMAT;
4853 #endif
4854 #ifdef OBJ_MAYBE_COFF
4855 case bfd_target_coff_flavour:
4856 return "coff-i386";
4857 #endif
4858 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
4859 case bfd_target_elf_flavour:
4860 {
4861 if (flag_code == CODE_64BIT)
4862 use_rela_relocations = 1;
4863 return flag_code == CODE_64BIT ? "elf64-x86-64" : "elf32-i386";
4864 }
4865 #endif
4866 default:
4867 abort ();
4868 return NULL;
4869 }
4870 }
4871
4872 #endif /* OBJ_MAYBE_ more than one */
4873
4874 #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
4875 void i386_elf_emit_arch_note ()
4876 {
4877 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4878 && cpu_arch_name != NULL)
4879 {
4880 char *p;
4881 asection *seg = now_seg;
4882 subsegT subseg = now_subseg;
4883 Elf_Internal_Note i_note;
4884 Elf_External_Note e_note;
4885 asection *note_secp;
4886 int len;
4887
4888 /* Create the .note section. */
4889 note_secp = subseg_new (".note", 0);
4890 bfd_set_section_flags (stdoutput,
4891 note_secp,
4892 SEC_HAS_CONTENTS | SEC_READONLY);
4893
4894 /* Process the arch string. */
4895 len = strlen (cpu_arch_name);
4896
4897 i_note.namesz = len + 1;
4898 i_note.descsz = 0;
4899 i_note.type = NT_ARCH;
4900 p = frag_more (sizeof (e_note.namesz));
4901 md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
4902 p = frag_more (sizeof (e_note.descsz));
4903 md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
4904 p = frag_more (sizeof (e_note.type));
4905 md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
4906 p = frag_more (len + 1);
4907 strcpy (p, cpu_arch_name);
4908
4909 frag_align (2, 0, 0);
4910
4911 subseg_set (seg, subseg);
4912 }
4913 }
4914 #endif
4915 #endif /* BFD_ASSEMBLER */
4916 \f
4917 symbolS *
4918 md_undefined_symbol (name)
4919 char *name;
4920 {
4921 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
4922 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
4923 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
4924 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
4925 {
4926 if (!GOT_symbol)
4927 {
4928 if (symbol_find (name))
4929 as_bad (_("GOT already in symbol table"));
4930 GOT_symbol = symbol_new (name, undefined_section,
4931 (valueT) 0, &zero_address_frag);
4932 };
4933 return GOT_symbol;
4934 }
4935 return 0;
4936 }
4937
4938 /* Round up a section size to the appropriate boundary. */
4939
4940 valueT
4941 md_section_align (segment, size)
4942 segT segment ATTRIBUTE_UNUSED;
4943 valueT size;
4944 {
4945 #ifdef BFD_ASSEMBLER
4946 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4947 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
4948 {
4949 /* For a.out, force the section size to be aligned. If we don't do
4950 this, BFD will align it for us, but it will not write out the
4951 final bytes of the section. This may be a bug in BFD, but it is
4952 easier to fix it here since that is how the other a.out targets
4953 work. */
4954 int align;
4955
4956 align = bfd_get_section_alignment (stdoutput, segment);
4957 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
4958 }
4959 #endif
4960 #endif
4961
4962 return size;
4963 }
4964
4965 /* On the i386, PC-relative offsets are relative to the start of the
4966 next instruction. That is, the address of the offset, plus its
4967 size, since the offset is always the last part of the insn. */
4968
4969 long
4970 md_pcrel_from (fixP)
4971 fixS *fixP;
4972 {
4973 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
4974 }
4975
4976 #ifndef I386COFF
4977
4978 static void
4979 s_bss (ignore)
4980 int ignore ATTRIBUTE_UNUSED;
4981 {
4982 int temp;
4983
4984 temp = get_absolute_expression ();
4985 subseg_set (bss_section, (subsegT) temp);
4986 demand_empty_rest_of_line ();
4987 }
4988
4989 #endif
4990
4991 #ifdef BFD_ASSEMBLER
4992
4993 void
4994 i386_validate_fix (fixp)
4995 fixS *fixp;
4996 {
4997 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
4998 {
4999 /* GOTOFF relocation are nonsense in 64bit mode. */
5000 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
5001 {
5002 if (flag_code != CODE_64BIT)
5003 abort ();
5004 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
5005 }
5006 else
5007 {
5008 if (flag_code == CODE_64BIT)
5009 abort ();
5010 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
5011 }
5012 fixp->fx_subsy = 0;
5013 }
5014 }
5015
5016 arelent *
5017 tc_gen_reloc (section, fixp)
5018 asection *section ATTRIBUTE_UNUSED;
5019 fixS *fixp;
5020 {
5021 arelent *rel;
5022 bfd_reloc_code_real_type code;
5023
5024 switch (fixp->fx_r_type)
5025 {
5026 case BFD_RELOC_X86_64_PLT32:
5027 case BFD_RELOC_X86_64_GOT32:
5028 case BFD_RELOC_X86_64_GOTPCREL:
5029 case BFD_RELOC_386_PLT32:
5030 case BFD_RELOC_386_GOT32:
5031 case BFD_RELOC_386_GOTOFF:
5032 case BFD_RELOC_386_GOTPC:
5033 case BFD_RELOC_X86_64_32S:
5034 case BFD_RELOC_RVA:
5035 case BFD_RELOC_VTABLE_ENTRY:
5036 case BFD_RELOC_VTABLE_INHERIT:
5037 code = fixp->fx_r_type;
5038 break;
5039 default:
5040 if (fixp->fx_pcrel)
5041 {
5042 switch (fixp->fx_size)
5043 {
5044 default:
5045 as_bad_where (fixp->fx_file, fixp->fx_line,
5046 _("can not do %d byte pc-relative relocation"),
5047 fixp->fx_size);
5048 code = BFD_RELOC_32_PCREL;
5049 break;
5050 case 1: code = BFD_RELOC_8_PCREL; break;
5051 case 2: code = BFD_RELOC_16_PCREL; break;
5052 case 4: code = BFD_RELOC_32_PCREL; break;
5053 }
5054 }
5055 else
5056 {
5057 switch (fixp->fx_size)
5058 {
5059 default:
5060 as_bad_where (fixp->fx_file, fixp->fx_line,
5061 _("can not do %d byte relocation"),
5062 fixp->fx_size);
5063 code = BFD_RELOC_32;
5064 break;
5065 case 1: code = BFD_RELOC_8; break;
5066 case 2: code = BFD_RELOC_16; break;
5067 case 4: code = BFD_RELOC_32; break;
5068 #ifdef BFD64
5069 case 8: code = BFD_RELOC_64; break;
5070 #endif
5071 }
5072 }
5073 break;
5074 }
5075
5076 if (code == BFD_RELOC_32
5077 && GOT_symbol
5078 && fixp->fx_addsy == GOT_symbol)
5079 {
5080 /* We don't support GOTPC on 64bit targets. */
5081 if (flag_code == CODE_64BIT)
5082 abort ();
5083 code = BFD_RELOC_386_GOTPC;
5084 }
5085
5086 rel = (arelent *) xmalloc (sizeof (arelent));
5087 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
5088 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
5089
5090 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
5091 if (!use_rela_relocations)
5092 {
5093 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
5094 vtable entry to be used in the relocation's section offset. */
5095 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
5096 rel->address = fixp->fx_offset;
5097
5098 if (fixp->fx_pcrel)
5099 rel->addend = fixp->fx_addnumber;
5100 else
5101 rel->addend = 0;
5102 }
5103 /* Use the rela in 64bit mode. */
5104 else
5105 {
5106 rel->addend = fixp->fx_offset;
5107 if (fixp->fx_pcrel)
5108 rel->addend -= fixp->fx_size;
5109 }
5110
5111 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
5112 if (rel->howto == NULL)
5113 {
5114 as_bad_where (fixp->fx_file, fixp->fx_line,
5115 _("cannot represent relocation type %s"),
5116 bfd_get_reloc_code_name (code));
5117 /* Set howto to a garbage value so that we can keep going. */
5118 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
5119 assert (rel->howto != NULL);
5120 }
5121
5122 return rel;
5123 }
5124
5125 #else /* !BFD_ASSEMBLER */
5126
5127 #if (defined(OBJ_AOUT) | defined(OBJ_BOUT))
5128 void
5129 tc_aout_fix_to_chars (where, fixP, segment_address_in_file)
5130 char *where;
5131 fixS *fixP;
5132 relax_addressT segment_address_in_file;
5133 {
5134 /* In: length of relocation (or of address) in chars: 1, 2 or 4.
5135 Out: GNU LD relocation length code: 0, 1, or 2. */
5136
5137 static const unsigned char nbytes_r_length[] = { 42, 0, 1, 42, 2 };
5138 long r_symbolnum;
5139
5140 know (fixP->fx_addsy != NULL);
5141
5142 md_number_to_chars (where,
5143 (valueT) (fixP->fx_frag->fr_address
5144 + fixP->fx_where - segment_address_in_file),
5145 4);
5146
5147 r_symbolnum = (S_IS_DEFINED (fixP->fx_addsy)
5148 ? S_GET_TYPE (fixP->fx_addsy)
5149 : fixP->fx_addsy->sy_number);
5150
5151 where[6] = (r_symbolnum >> 16) & 0x0ff;
5152 where[5] = (r_symbolnum >> 8) & 0x0ff;
5153 where[4] = r_symbolnum & 0x0ff;
5154 where[7] = ((((!S_IS_DEFINED (fixP->fx_addsy)) << 3) & 0x08)
5155 | ((nbytes_r_length[fixP->fx_size] << 1) & 0x06)
5156 | (((fixP->fx_pcrel << 0) & 0x01) & 0x0f));
5157 }
5158
5159 #endif /* OBJ_AOUT or OBJ_BOUT. */
5160
5161 #if defined (I386COFF)
5162
5163 short
5164 tc_coff_fix2rtype (fixP)
5165 fixS *fixP;
5166 {
5167 if (fixP->fx_r_type == R_IMAGEBASE)
5168 return R_IMAGEBASE;
5169
5170 return (fixP->fx_pcrel ?
5171 (fixP->fx_size == 1 ? R_PCRBYTE :
5172 fixP->fx_size == 2 ? R_PCRWORD :
5173 R_PCRLONG) :
5174 (fixP->fx_size == 1 ? R_RELBYTE :
5175 fixP->fx_size == 2 ? R_RELWORD :
5176 R_DIR32));
5177 }
5178
5179 int
5180 tc_coff_sizemachdep (frag)
5181 fragS *frag;
5182 {
5183 if (frag->fr_next)
5184 return (frag->fr_next->fr_address - frag->fr_address);
5185 else
5186 return 0;
5187 }
5188
5189 #endif /* I386COFF */
5190
5191 #endif /* !BFD_ASSEMBLER */
5192 \f
5193 /* Parse operands using Intel syntax. This implements a recursive descent
5194 parser based on the BNF grammar published in Appendix B of the MASM 6.1
5195 Programmer's Guide.
5196
5197 FIXME: We do not recognize the full operand grammar defined in the MASM
5198 documentation. In particular, all the structure/union and
5199 high-level macro operands are missing.
5200
5201 Uppercase words are terminals, lower case words are non-terminals.
5202 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
5203 bars '|' denote choices. Most grammar productions are implemented in
5204 functions called 'intel_<production>'.
5205
5206 Initial production is 'expr'.
5207
5208 addOp + | -
5209
5210 alpha [a-zA-Z]
5211
5212 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
5213
5214 constant digits [[ radixOverride ]]
5215
5216 dataType BYTE | WORD | DWORD | QWORD | XWORD
5217
5218 digits decdigit
5219 | digits decdigit
5220 | digits hexdigit
5221
5222 decdigit [0-9]
5223
5224 e05 e05 addOp e06
5225 | e06
5226
5227 e06 e06 mulOp e09
5228 | e09
5229
5230 e09 OFFSET e10
5231 | e09 PTR e10
5232 | e09 : e10
5233 | e10
5234
5235 e10 e10 [ expr ]
5236 | e11
5237
5238 e11 ( expr )
5239 | [ expr ]
5240 | constant
5241 | dataType
5242 | id
5243 | $
5244 | register
5245
5246 => expr SHORT e05
5247 | e05
5248
5249 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
5250 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
5251
5252 hexdigit a | b | c | d | e | f
5253 | A | B | C | D | E | F
5254
5255 id alpha
5256 | id alpha
5257 | id decdigit
5258
5259 mulOp * | / | MOD
5260
5261 quote " | '
5262
5263 register specialRegister
5264 | gpRegister
5265 | byteRegister
5266
5267 segmentRegister CS | DS | ES | FS | GS | SS
5268
5269 specialRegister CR0 | CR2 | CR3
5270 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
5271 | TR3 | TR4 | TR5 | TR6 | TR7
5272
5273 We simplify the grammar in obvious places (e.g., register parsing is
5274 done by calling parse_register) and eliminate immediate left recursion
5275 to implement a recursive-descent parser.
5276
5277 expr SHORT e05
5278 | e05
5279
5280 e05 e06 e05'
5281
5282 e05' addOp e06 e05'
5283 | Empty
5284
5285 e06 e09 e06'
5286
5287 e06' mulOp e09 e06'
5288 | Empty
5289
5290 e09 OFFSET e10 e09'
5291 | e10 e09'
5292
5293 e09' PTR e10 e09'
5294 | : e10 e09'
5295 | Empty
5296
5297 e10 e11 e10'
5298
5299 e10' [ expr ] e10'
5300 | Empty
5301
5302 e11 ( expr )
5303 | [ expr ]
5304 | BYTE
5305 | WORD
5306 | DWORD
5307 | QWORD
5308 | XWORD
5309 | .
5310 | $
5311 | register
5312 | id
5313 | constant */
5314
5315 /* Parsing structure for the intel syntax parser. Used to implement the
5316 semantic actions for the operand grammar. */
5317 struct intel_parser_s
5318 {
5319 char *op_string; /* The string being parsed. */
5320 int got_a_float; /* Whether the operand is a float. */
5321 int op_modifier; /* Operand modifier. */
5322 int is_mem; /* 1 if operand is memory reference. */
5323 const reg_entry *reg; /* Last register reference found. */
5324 char *disp; /* Displacement string being built. */
5325 };
5326
5327 static struct intel_parser_s intel_parser;
5328
5329 /* Token structure for parsing intel syntax. */
5330 struct intel_token
5331 {
5332 int code; /* Token code. */
5333 const reg_entry *reg; /* Register entry for register tokens. */
5334 char *str; /* String representation. */
5335 };
5336
5337 static struct intel_token cur_token, prev_token;
5338
5339 /* Token codes for the intel parser. Since T_SHORT is already used
5340 by COFF, undefine it first to prevent a warning. */
5341 #define T_NIL -1
5342 #define T_CONST 1
5343 #define T_REG 2
5344 #define T_BYTE 3
5345 #define T_WORD 4
5346 #define T_DWORD 5
5347 #define T_QWORD 6
5348 #define T_XWORD 7
5349 #undef T_SHORT
5350 #define T_SHORT 8
5351 #define T_OFFSET 9
5352 #define T_PTR 10
5353 #define T_ID 11
5354
5355 /* Prototypes for intel parser functions. */
5356 static int intel_match_token PARAMS ((int code));
5357 static void intel_get_token PARAMS ((void));
5358 static void intel_putback_token PARAMS ((void));
5359 static int intel_expr PARAMS ((void));
5360 static int intel_e05 PARAMS ((void));
5361 static int intel_e05_1 PARAMS ((void));
5362 static int intel_e06 PARAMS ((void));
5363 static int intel_e06_1 PARAMS ((void));
5364 static int intel_e09 PARAMS ((void));
5365 static int intel_e09_1 PARAMS ((void));
5366 static int intel_e10 PARAMS ((void));
5367 static int intel_e10_1 PARAMS ((void));
5368 static int intel_e11 PARAMS ((void));
5369
5370 static int
5371 i386_intel_operand (operand_string, got_a_float)
5372 char *operand_string;
5373 int got_a_float;
5374 {
5375 int ret;
5376 char *p;
5377
5378 /* Initialize token holders. */
5379 cur_token.code = prev_token.code = T_NIL;
5380 cur_token.reg = prev_token.reg = NULL;
5381 cur_token.str = prev_token.str = NULL;
5382
5383 /* Initialize parser structure. */
5384 p = intel_parser.op_string = (char *) malloc (strlen (operand_string) + 1);
5385 if (p == NULL)
5386 abort ();
5387 strcpy (intel_parser.op_string, operand_string);
5388 intel_parser.got_a_float = got_a_float;
5389 intel_parser.op_modifier = -1;
5390 intel_parser.is_mem = 0;
5391 intel_parser.reg = NULL;
5392 intel_parser.disp = (char *) malloc (strlen (operand_string) + 1);
5393 if (intel_parser.disp == NULL)
5394 abort ();
5395 intel_parser.disp[0] = '\0';
5396
5397 /* Read the first token and start the parser. */
5398 intel_get_token ();
5399 ret = intel_expr ();
5400
5401 if (ret)
5402 {
5403 /* If we found a memory reference, hand it over to i386_displacement
5404 to fill in the rest of the operand fields. */
5405 if (intel_parser.is_mem)
5406 {
5407 if ((i.mem_operands == 1
5408 && (current_templates->start->opcode_modifier & IsString) == 0)
5409 || i.mem_operands == 2)
5410 {
5411 as_bad (_("too many memory references for '%s'"),
5412 current_templates->start->name);
5413 ret = 0;
5414 }
5415 else
5416 {
5417 char *s = intel_parser.disp;
5418 i.mem_operands++;
5419
5420 /* Add the displacement expression. */
5421 if (*s != '\0')
5422 ret = i386_displacement (s, s + strlen (s))
5423 && i386_index_check (s);
5424 }
5425 }
5426
5427 /* Constant and OFFSET expressions are handled by i386_immediate. */
5428 else if (intel_parser.op_modifier == OFFSET_FLAT
5429 || intel_parser.reg == NULL)
5430 ret = i386_immediate (intel_parser.disp);
5431 }
5432
5433 free (p);
5434 free (intel_parser.disp);
5435
5436 return ret;
5437 }
5438
5439 /* expr SHORT e05
5440 | e05 */
5441 static int
5442 intel_expr ()
5443 {
5444 /* expr SHORT e05 */
5445 if (cur_token.code == T_SHORT)
5446 {
5447 intel_parser.op_modifier = SHORT;
5448 intel_match_token (T_SHORT);
5449
5450 return (intel_e05 ());
5451 }
5452
5453 /* expr e05 */
5454 else
5455 return intel_e05 ();
5456 }
5457
5458 /* e05 e06 e05'
5459
5460 e05' addOp e06 e05'
5461 | Empty */
5462 static int
5463 intel_e05 ()
5464 {
5465 return (intel_e06 () && intel_e05_1 ());
5466 }
5467
5468 static int
5469 intel_e05_1 ()
5470 {
5471 /* e05' addOp e06 e05' */
5472 if (cur_token.code == '+' || cur_token.code == '-')
5473 {
5474 strcat (intel_parser.disp, cur_token.str);
5475 intel_match_token (cur_token.code);
5476
5477 return (intel_e06 () && intel_e05_1 ());
5478 }
5479
5480 /* e05' Empty */
5481 else
5482 return 1;
5483 }
5484
5485 /* e06 e09 e06'
5486
5487 e06' mulOp e09 e06'
5488 | Empty */
5489 static int
5490 intel_e06 ()
5491 {
5492 return (intel_e09 () && intel_e06_1 ());
5493 }
5494
5495 static int
5496 intel_e06_1 ()
5497 {
5498 /* e06' mulOp e09 e06' */
5499 if (cur_token.code == '*' || cur_token.code == '/')
5500 {
5501 strcat (intel_parser.disp, cur_token.str);
5502 intel_match_token (cur_token.code);
5503
5504 return (intel_e09 () && intel_e06_1 ());
5505 }
5506
5507 /* e06' Empty */
5508 else
5509 return 1;
5510 }
5511
5512 /* e09 OFFSET e10 e09'
5513 | e10 e09'
5514
5515 e09' PTR e10 e09'
5516 | : e10 e09'
5517 | Empty */
5518 static int
5519 intel_e09 ()
5520 {
5521 /* e09 OFFSET e10 e09' */
5522 if (cur_token.code == T_OFFSET)
5523 {
5524 intel_parser.is_mem = 0;
5525 intel_parser.op_modifier = OFFSET_FLAT;
5526 intel_match_token (T_OFFSET);
5527
5528 return (intel_e10 () && intel_e09_1 ());
5529 }
5530
5531 /* e09 e10 e09' */
5532 else
5533 return (intel_e10 () && intel_e09_1 ());
5534 }
5535
5536 static int
5537 intel_e09_1 ()
5538 {
5539 /* e09' PTR e10 e09' */
5540 if (cur_token.code == T_PTR)
5541 {
5542 if (prev_token.code == T_BYTE)
5543 i.suffix = BYTE_MNEM_SUFFIX;
5544
5545 else if (prev_token.code == T_WORD)
5546 {
5547 if (intel_parser.got_a_float == 2) /* "fi..." */
5548 i.suffix = SHORT_MNEM_SUFFIX;
5549 else
5550 i.suffix = WORD_MNEM_SUFFIX;
5551 }
5552
5553 else if (prev_token.code == T_DWORD)
5554 {
5555 if (intel_parser.got_a_float == 1) /* "f..." */
5556 i.suffix = SHORT_MNEM_SUFFIX;
5557 else
5558 i.suffix = LONG_MNEM_SUFFIX;
5559 }
5560
5561 else if (prev_token.code == T_QWORD)
5562 {
5563 if (intel_parser.got_a_float == 1) /* "f..." */
5564 i.suffix = LONG_MNEM_SUFFIX;
5565 else
5566 i.suffix = QWORD_MNEM_SUFFIX;
5567 }
5568
5569 else if (prev_token.code == T_XWORD)
5570 i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
5571
5572 else
5573 {
5574 as_bad (_("Unknown operand modifier `%s'\n"), prev_token.str);
5575 return 0;
5576 }
5577
5578 intel_match_token (T_PTR);
5579
5580 return (intel_e10 () && intel_e09_1 ());
5581 }
5582
5583 /* e09 : e10 e09' */
5584 else if (cur_token.code == ':')
5585 {
5586 /* Mark as a memory operand only if it's not already known to be an
5587 offset expression. */
5588 if (intel_parser.op_modifier != OFFSET_FLAT)
5589 intel_parser.is_mem = 1;
5590
5591 return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
5592 }
5593
5594 /* e09' Empty */
5595 else
5596 return 1;
5597 }
5598
5599 /* e10 e11 e10'
5600
5601 e10' [ expr ] e10'
5602 | Empty */
5603 static int
5604 intel_e10 ()
5605 {
5606 return (intel_e11 () && intel_e10_1 ());
5607 }
5608
5609 static int
5610 intel_e10_1 ()
5611 {
5612 /* e10' [ expr ] e10' */
5613 if (cur_token.code == '[')
5614 {
5615 intel_match_token ('[');
5616
5617 /* Mark as a memory operand only if it's not already known to be an
5618 offset expression. If it's an offset expression, we need to keep
5619 the brace in. */
5620 if (intel_parser.op_modifier != OFFSET_FLAT)
5621 intel_parser.is_mem = 1;
5622 else
5623 strcat (intel_parser.disp, "[");
5624
5625 /* Add a '+' to the displacement string if necessary. */
5626 if (*intel_parser.disp != '\0'
5627 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5628 strcat (intel_parser.disp, "+");
5629
5630 if (intel_expr () && intel_match_token (']'))
5631 {
5632 /* Preserve brackets when the operand is an offset expression. */
5633 if (intel_parser.op_modifier == OFFSET_FLAT)
5634 strcat (intel_parser.disp, "]");
5635
5636 return intel_e10_1 ();
5637 }
5638 else
5639 return 0;
5640 }
5641
5642 /* e10' Empty */
5643 else
5644 return 1;
5645 }
5646
5647 /* e11 ( expr )
5648 | [ expr ]
5649 | BYTE
5650 | WORD
5651 | DWORD
5652 | QWORD
5653 | XWORD
5654 | $
5655 | .
5656 | register
5657 | id
5658 | constant */
5659 static int
5660 intel_e11 ()
5661 {
5662 /* e11 ( expr ) */
5663 if (cur_token.code == '(')
5664 {
5665 intel_match_token ('(');
5666 strcat (intel_parser.disp, "(");
5667
5668 if (intel_expr () && intel_match_token (')'))
5669 {
5670 strcat (intel_parser.disp, ")");
5671 return 1;
5672 }
5673 else
5674 return 0;
5675 }
5676
5677 /* e11 [ expr ] */
5678 else if (cur_token.code == '[')
5679 {
5680 intel_match_token ('[');
5681
5682 /* Mark as a memory operand only if it's not already known to be an
5683 offset expression. If it's an offset expression, we need to keep
5684 the brace in. */
5685 if (intel_parser.op_modifier != OFFSET_FLAT)
5686 intel_parser.is_mem = 1;
5687 else
5688 strcat (intel_parser.disp, "[");
5689
5690 /* Operands for jump/call inside brackets denote absolute addresses. */
5691 if (current_templates->start->opcode_modifier & Jump
5692 || current_templates->start->opcode_modifier & JumpDword
5693 || current_templates->start->opcode_modifier & JumpByte
5694 || current_templates->start->opcode_modifier & JumpInterSegment)
5695 i.types[this_operand] |= JumpAbsolute;
5696
5697 /* Add a '+' to the displacement string if necessary. */
5698 if (*intel_parser.disp != '\0'
5699 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
5700 strcat (intel_parser.disp, "+");
5701
5702 if (intel_expr () && intel_match_token (']'))
5703 {
5704 /* Preserve brackets when the operand is an offset expression. */
5705 if (intel_parser.op_modifier == OFFSET_FLAT)
5706 strcat (intel_parser.disp, "]");
5707
5708 return 1;
5709 }
5710 else
5711 return 0;
5712 }
5713
5714 /* e11 BYTE
5715 | WORD
5716 | DWORD
5717 | QWORD
5718 | XWORD */
5719 else if (cur_token.code == T_BYTE
5720 || cur_token.code == T_WORD
5721 || cur_token.code == T_DWORD
5722 || cur_token.code == T_QWORD
5723 || cur_token.code == T_XWORD)
5724 {
5725 intel_match_token (cur_token.code);
5726
5727 return 1;
5728 }
5729
5730 /* e11 $
5731 | . */
5732 else if (cur_token.code == '$' || cur_token.code == '.')
5733 {
5734 strcat (intel_parser.disp, cur_token.str);
5735 intel_match_token (cur_token.code);
5736
5737 /* Mark as a memory operand only if it's not already known to be an
5738 offset expression. */
5739 if (intel_parser.op_modifier != OFFSET_FLAT)
5740 intel_parser.is_mem = 1;
5741
5742 return 1;
5743 }
5744
5745 /* e11 register */
5746 else if (cur_token.code == T_REG)
5747 {
5748 const reg_entry *reg = intel_parser.reg = cur_token.reg;
5749
5750 intel_match_token (T_REG);
5751
5752 /* Check for segment change. */
5753 if (cur_token.code == ':')
5754 {
5755 if (reg->reg_type & (SReg2 | SReg3))
5756 {
5757 switch (reg->reg_num)
5758 {
5759 case 0:
5760 i.seg[i.mem_operands] = &es;
5761 break;
5762 case 1:
5763 i.seg[i.mem_operands] = &cs;
5764 break;
5765 case 2:
5766 i.seg[i.mem_operands] = &ss;
5767 break;
5768 case 3:
5769 i.seg[i.mem_operands] = &ds;
5770 break;
5771 case 4:
5772 i.seg[i.mem_operands] = &fs;
5773 break;
5774 case 5:
5775 i.seg[i.mem_operands] = &gs;
5776 break;
5777 }
5778 }
5779 else
5780 {
5781 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
5782 return 0;
5783 }
5784 }
5785
5786 /* Not a segment register. Check for register scaling. */
5787 else if (cur_token.code == '*')
5788 {
5789 if (!intel_parser.is_mem)
5790 {
5791 as_bad (_("Register scaling only allowed in memory operands."));
5792 return 0;
5793 }
5794
5795 /* What follows must be a valid scale. */
5796 if (intel_match_token ('*')
5797 && strchr ("01248", *cur_token.str))
5798 {
5799 i.index_reg = reg;
5800 i.types[this_operand] |= BaseIndex;
5801
5802 /* Set the scale after setting the register (otherwise,
5803 i386_scale will complain) */
5804 i386_scale (cur_token.str);
5805 intel_match_token (T_CONST);
5806 }
5807 else
5808 {
5809 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
5810 cur_token.str);
5811 return 0;
5812 }
5813 }
5814
5815 /* No scaling. If this is a memory operand, the register is either a
5816 base register (first occurrence) or an index register (second
5817 occurrence). */
5818 else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3)))
5819 {
5820 if (i.base_reg && i.index_reg)
5821 {
5822 as_bad (_("Too many register references in memory operand.\n"));
5823 return 0;
5824 }
5825
5826 if (i.base_reg == NULL)
5827 i.base_reg = reg;
5828 else
5829 i.index_reg = reg;
5830
5831 i.types[this_operand] |= BaseIndex;
5832 }
5833
5834 /* Offset modifier. Add the register to the displacement string to be
5835 parsed as an immediate expression after we're done. */
5836 else if (intel_parser.op_modifier == OFFSET_FLAT)
5837 strcat (intel_parser.disp, reg->reg_name);
5838
5839 /* It's neither base nor index nor offset. */
5840 else
5841 {
5842 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
5843 i.op[this_operand].regs = reg;
5844 i.reg_operands++;
5845 }
5846
5847 /* Since registers are not part of the displacement string (except
5848 when we're parsing offset operands), we may need to remove any
5849 preceding '+' from the displacement string. */
5850 if (*intel_parser.disp != '\0'
5851 && intel_parser.op_modifier != OFFSET_FLAT)
5852 {
5853 char *s = intel_parser.disp;
5854 s += strlen (s) - 1;
5855 if (*s == '+')
5856 *s = '\0';
5857 }
5858
5859 return 1;
5860 }
5861
5862 /* e11 id */
5863 else if (cur_token.code == T_ID)
5864 {
5865 /* Add the identifier to the displacement string. */
5866 strcat (intel_parser.disp, cur_token.str);
5867 intel_match_token (T_ID);
5868
5869 /* The identifier represents a memory reference only if it's not
5870 preceded by an offset modifier. */
5871 if (intel_parser.op_modifier != OFFSET_FLAT)
5872 intel_parser.is_mem = 1;
5873
5874 return 1;
5875 }
5876
5877 /* e11 constant */
5878 else if (cur_token.code == T_CONST
5879 || cur_token.code == '-'
5880 || cur_token.code == '+')
5881 {
5882 char *save_str;
5883
5884 /* Allow constants that start with `+' or `-'. */
5885 if (cur_token.code == '-' || cur_token.code == '+')
5886 {
5887 strcat (intel_parser.disp, cur_token.str);
5888 intel_match_token (cur_token.code);
5889 if (cur_token.code != T_CONST)
5890 {
5891 as_bad (_("Syntax error. Expecting a constant. Got `%s'.\n"),
5892 cur_token.str);
5893 return 0;
5894 }
5895 }
5896
5897 save_str = (char *) malloc (strlen (cur_token.str) + 1);
5898 if (save_str == NULL)
5899 abort ();
5900 strcpy (save_str, cur_token.str);
5901
5902 /* Get the next token to check for register scaling. */
5903 intel_match_token (cur_token.code);
5904
5905 /* Check if this constant is a scaling factor for an index register. */
5906 if (cur_token.code == '*')
5907 {
5908 if (intel_match_token ('*') && cur_token.code == T_REG)
5909 {
5910 if (!intel_parser.is_mem)
5911 {
5912 as_bad (_("Register scaling only allowed in memory operands."));
5913 return 0;
5914 }
5915
5916 /* The constant is followed by `* reg', so it must be
5917 a valid scale. */
5918 if (strchr ("01248", *save_str))
5919 {
5920 i.index_reg = cur_token.reg;
5921 i.types[this_operand] |= BaseIndex;
5922
5923 /* Set the scale after setting the register (otherwise,
5924 i386_scale will complain) */
5925 i386_scale (save_str);
5926 intel_match_token (T_REG);
5927
5928 /* Since registers are not part of the displacement
5929 string, we may need to remove any preceding '+' from
5930 the displacement string. */
5931 if (*intel_parser.disp != '\0')
5932 {
5933 char *s = intel_parser.disp;
5934 s += strlen (s) - 1;
5935 if (*s == '+')
5936 *s = '\0';
5937 }
5938
5939 free (save_str);
5940
5941 return 1;
5942 }
5943 else
5944 return 0;
5945 }
5946
5947 /* The constant was not used for register scaling. Since we have
5948 already consumed the token following `*' we now need to put it
5949 back in the stream. */
5950 else
5951 intel_putback_token ();
5952 }
5953
5954 /* Add the constant to the displacement string. */
5955 strcat (intel_parser.disp, save_str);
5956 free (save_str);
5957
5958 return 1;
5959 }
5960
5961 as_bad (_("Unrecognized token '%s'"), cur_token.str);
5962 return 0;
5963 }
5964
5965 /* Match the given token against cur_token. If they match, read the next
5966 token from the operand string. */
5967 static int
5968 intel_match_token (code)
5969 int code;
5970 {
5971 if (cur_token.code == code)
5972 {
5973 intel_get_token ();
5974 return 1;
5975 }
5976 else
5977 {
5978 as_bad (_("Unexpected token `%s'\n"), cur_token.str);
5979 return 0;
5980 }
5981 }
5982
5983 /* Read a new token from intel_parser.op_string and store it in cur_token. */
5984 static void
5985 intel_get_token ()
5986 {
5987 char *end_op;
5988 const reg_entry *reg;
5989 struct intel_token new_token;
5990
5991 new_token.code = T_NIL;
5992 new_token.reg = NULL;
5993 new_token.str = NULL;
5994
5995 /* Free the memory allocated to the previous token and move
5996 cur_token to prev_token. */
5997 if (prev_token.str)
5998 free (prev_token.str);
5999
6000 prev_token = cur_token;
6001
6002 /* Skip whitespace. */
6003 while (is_space_char (*intel_parser.op_string))
6004 intel_parser.op_string++;
6005
6006 /* Return an empty token if we find nothing else on the line. */
6007 if (*intel_parser.op_string == '\0')
6008 {
6009 cur_token = new_token;
6010 return;
6011 }
6012
6013 /* The new token cannot be larger than the remainder of the operand
6014 string. */
6015 new_token.str = (char *) malloc (strlen (intel_parser.op_string) + 1);
6016 if (new_token.str == NULL)
6017 abort ();
6018 new_token.str[0] = '\0';
6019
6020 if (strchr ("0123456789", *intel_parser.op_string))
6021 {
6022 char *p = new_token.str;
6023 char *q = intel_parser.op_string;
6024 new_token.code = T_CONST;
6025
6026 /* Allow any kind of identifier char to encompass floating point and
6027 hexadecimal numbers. */
6028 while (is_identifier_char (*q))
6029 *p++ = *q++;
6030 *p = '\0';
6031
6032 /* Recognize special symbol names [0-9][bf]. */
6033 if (strlen (intel_parser.op_string) == 2
6034 && (intel_parser.op_string[1] == 'b'
6035 || intel_parser.op_string[1] == 'f'))
6036 new_token.code = T_ID;
6037 }
6038
6039 else if (strchr ("+-/*:[]()", *intel_parser.op_string))
6040 {
6041 new_token.code = *intel_parser.op_string;
6042 new_token.str[0] = *intel_parser.op_string;
6043 new_token.str[1] = '\0';
6044 }
6045
6046 else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
6047 && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
6048 {
6049 new_token.code = T_REG;
6050 new_token.reg = reg;
6051
6052 if (*intel_parser.op_string == REGISTER_PREFIX)
6053 {
6054 new_token.str[0] = REGISTER_PREFIX;
6055 new_token.str[1] = '\0';
6056 }
6057
6058 strcat (new_token.str, reg->reg_name);
6059 }
6060
6061 else if (is_identifier_char (*intel_parser.op_string))
6062 {
6063 char *p = new_token.str;
6064 char *q = intel_parser.op_string;
6065
6066 /* A '.' or '$' followed by an identifier char is an identifier.
6067 Otherwise, it's operator '.' followed by an expression. */
6068 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
6069 {
6070 new_token.code = *q;
6071 new_token.str[0] = *q;
6072 new_token.str[1] = '\0';
6073 }
6074 else
6075 {
6076 while (is_identifier_char (*q) || *q == '@')
6077 *p++ = *q++;
6078 *p = '\0';
6079
6080 if (strcasecmp (new_token.str, "BYTE") == 0)
6081 new_token.code = T_BYTE;
6082
6083 else if (strcasecmp (new_token.str, "WORD") == 0)
6084 new_token.code = T_WORD;
6085
6086 else if (strcasecmp (new_token.str, "DWORD") == 0)
6087 new_token.code = T_DWORD;
6088
6089 else if (strcasecmp (new_token.str, "QWORD") == 0)
6090 new_token.code = T_QWORD;
6091
6092 else if (strcasecmp (new_token.str, "XWORD") == 0)
6093 new_token.code = T_XWORD;
6094
6095 else if (strcasecmp (new_token.str, "PTR") == 0)
6096 new_token.code = T_PTR;
6097
6098 else if (strcasecmp (new_token.str, "SHORT") == 0)
6099 new_token.code = T_SHORT;
6100
6101 else if (strcasecmp (new_token.str, "OFFSET") == 0)
6102 {
6103 new_token.code = T_OFFSET;
6104
6105 /* ??? This is not mentioned in the MASM grammar but gcc
6106 makes use of it with -mintel-syntax. OFFSET may be
6107 followed by FLAT: */
6108 if (strncasecmp (q, " FLAT:", 6) == 0)
6109 strcat (new_token.str, " FLAT:");
6110 }
6111
6112 /* ??? This is not mentioned in the MASM grammar. */
6113 else if (strcasecmp (new_token.str, "FLAT") == 0)
6114 new_token.code = T_OFFSET;
6115
6116 else
6117 new_token.code = T_ID;
6118 }
6119 }
6120
6121 else
6122 as_bad (_("Unrecognized token `%s'\n"), intel_parser.op_string);
6123
6124 intel_parser.op_string += strlen (new_token.str);
6125 cur_token = new_token;
6126 }
6127
6128 /* Put cur_token back into the token stream and make cur_token point to
6129 prev_token. */
6130 static void
6131 intel_putback_token ()
6132 {
6133 intel_parser.op_string -= strlen (cur_token.str);
6134 free (cur_token.str);
6135 cur_token = prev_token;
6136
6137 /* Forget prev_token. */
6138 prev_token.code = T_NIL;
6139 prev_token.reg = NULL;
6140 prev_token.str = NULL;
6141 }
This page took 0.159186 seconds and 4 git commands to generate.