* config-gas.com: Mark vax-vms as obsolete.
[deliverable/binutils-gdb.git] / gas / config / tc-i386.c
1 /* i386.c -- Assemble code for the Intel 80386
2 Copyright 1989, 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
3 2000, 2001, 2002, 2003, 2004, 2005
4 Free Software Foundation, Inc.
5
6 This file is part of GAS, the GNU Assembler.
7
8 GAS is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2, or (at your option)
11 any later version.
12
13 GAS is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
17
18 You should have received a copy of the GNU General Public License
19 along with GAS; see the file COPYING. If not, write to the Free
20 Software Foundation, 59 Temple Place - Suite 330, Boston, MA
21 02111-1307, USA. */
22
23 /* Intel 80386 machine specific gas.
24 Written by Eliot Dresselhaus (eliot@mgm.mit.edu).
25 x86_64 support by Jan Hubicka (jh@suse.cz)
26 VIA PadLock support by Michal Ludvig (mludvig@suse.cz)
27 Bugs & suggestions are completely welcome. This is free software.
28 Please help us make it better. */
29
30 #include "as.h"
31 #include "safe-ctype.h"
32 #include "subsegs.h"
33 #include "dwarf2dbg.h"
34 #include "dw2gencfi.h"
35 #include "opcode/i386.h"
36 #include "elf/x86-64.h"
37
38 #ifndef REGISTER_WARNINGS
39 #define REGISTER_WARNINGS 1
40 #endif
41
42 #ifndef INFER_ADDR_PREFIX
43 #define INFER_ADDR_PREFIX 1
44 #endif
45
46 #ifndef SCALE1_WHEN_NO_INDEX
47 /* Specifying a scale factor besides 1 when there is no index is
48 futile. eg. `mov (%ebx,2),%al' does exactly the same as
49 `mov (%ebx),%al'. To slavishly follow what the programmer
50 specified, set SCALE1_WHEN_NO_INDEX to 0. */
51 #define SCALE1_WHEN_NO_INDEX 1
52 #endif
53
54 #ifndef DEFAULT_ARCH
55 #define DEFAULT_ARCH "i386"
56 #endif
57
58 #ifndef INLINE
59 #if __GNUC__ >= 2
60 #define INLINE __inline__
61 #else
62 #define INLINE
63 #endif
64 #endif
65
66 static INLINE unsigned int mode_from_disp_size PARAMS ((unsigned int));
67 static INLINE int fits_in_signed_byte PARAMS ((offsetT));
68 static INLINE int fits_in_unsigned_byte PARAMS ((offsetT));
69 static INLINE int fits_in_unsigned_word PARAMS ((offsetT));
70 static INLINE int fits_in_signed_word PARAMS ((offsetT));
71 static INLINE int fits_in_unsigned_long PARAMS ((offsetT));
72 static INLINE int fits_in_signed_long PARAMS ((offsetT));
73 static int smallest_imm_type PARAMS ((offsetT));
74 static offsetT offset_in_range PARAMS ((offsetT, int));
75 static int add_prefix PARAMS ((unsigned int));
76 static void set_code_flag PARAMS ((int));
77 static void set_16bit_gcc_code_flag PARAMS ((int));
78 static void set_intel_syntax PARAMS ((int));
79 static void set_cpu_arch PARAMS ((int));
80 #ifdef TE_PE
81 static void pe_directive_secrel PARAMS ((int));
82 #endif
83 static char *output_invalid PARAMS ((int c));
84 static int i386_operand PARAMS ((char *operand_string));
85 static int i386_intel_operand PARAMS ((char *operand_string, int got_a_float));
86 static const reg_entry *parse_register PARAMS ((char *reg_string,
87 char **end_op));
88 static char *parse_insn PARAMS ((char *, char *));
89 static char *parse_operands PARAMS ((char *, const char *));
90 static void swap_operands PARAMS ((void));
91 static void optimize_imm PARAMS ((void));
92 static void optimize_disp PARAMS ((void));
93 static int match_template PARAMS ((void));
94 static int check_string PARAMS ((void));
95 static int process_suffix PARAMS ((void));
96 static int check_byte_reg PARAMS ((void));
97 static int check_long_reg PARAMS ((void));
98 static int check_qword_reg PARAMS ((void));
99 static int check_word_reg PARAMS ((void));
100 static int finalize_imm PARAMS ((void));
101 static int process_operands PARAMS ((void));
102 static const seg_entry *build_modrm_byte PARAMS ((void));
103 static void output_insn PARAMS ((void));
104 static void output_branch PARAMS ((void));
105 static void output_jump PARAMS ((void));
106 static void output_interseg_jump PARAMS ((void));
107 static void output_imm PARAMS ((fragS *insn_start_frag,
108 offsetT insn_start_off));
109 static void output_disp PARAMS ((fragS *insn_start_frag,
110 offsetT insn_start_off));
111 #ifndef I386COFF
112 static void s_bss PARAMS ((int));
113 #endif
114
115 static const char *default_arch = DEFAULT_ARCH;
116
117 /* 'md_assemble ()' gathers together information and puts it into a
118 i386_insn. */
119
120 union i386_op
121 {
122 expressionS *disps;
123 expressionS *imms;
124 const reg_entry *regs;
125 };
126
127 struct _i386_insn
128 {
129 /* TM holds the template for the insn were currently assembling. */
130 template tm;
131
132 /* SUFFIX holds the instruction mnemonic suffix if given.
133 (e.g. 'l' for 'movl') */
134 char suffix;
135
136 /* OPERANDS gives the number of given operands. */
137 unsigned int operands;
138
139 /* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number
140 of given register, displacement, memory operands and immediate
141 operands. */
142 unsigned int reg_operands, disp_operands, mem_operands, imm_operands;
143
144 /* TYPES [i] is the type (see above #defines) which tells us how to
145 use OP[i] for the corresponding operand. */
146 unsigned int types[MAX_OPERANDS];
147
148 /* Displacement expression, immediate expression, or register for each
149 operand. */
150 union i386_op op[MAX_OPERANDS];
151
152 /* Flags for operands. */
153 unsigned int flags[MAX_OPERANDS];
154 #define Operand_PCrel 1
155
156 /* Relocation type for operand */
157 enum bfd_reloc_code_real reloc[MAX_OPERANDS];
158
159 /* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode
160 the base index byte below. */
161 const reg_entry *base_reg;
162 const reg_entry *index_reg;
163 unsigned int log2_scale_factor;
164
165 /* SEG gives the seg_entries of this insn. They are zero unless
166 explicit segment overrides are given. */
167 const seg_entry *seg[2];
168
169 /* PREFIX holds all the given prefix opcodes (usually null).
170 PREFIXES is the number of prefix opcodes. */
171 unsigned int prefixes;
172 unsigned char prefix[MAX_PREFIXES];
173
174 /* RM and SIB are the modrm byte and the sib byte where the
175 addressing modes of this insn are encoded. */
176
177 modrm_byte rm;
178 rex_byte rex;
179 sib_byte sib;
180 };
181
182 typedef struct _i386_insn i386_insn;
183
184 /* List of chars besides those in app.c:symbol_chars that can start an
185 operand. Used to prevent the scrubber eating vital white-space. */
186 const char extra_symbol_chars[] = "*%-(["
187 #ifdef LEX_AT
188 "@"
189 #endif
190 #ifdef LEX_QM
191 "?"
192 #endif
193 ;
194
195 #if (defined (TE_I386AIX) \
196 || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) \
197 && !defined (TE_LINUX) \
198 && !defined (TE_NETWARE) \
199 && !defined (TE_FreeBSD) \
200 && !defined (TE_NetBSD)))
201 /* This array holds the chars that always start a comment. If the
202 pre-processor is disabled, these aren't very useful. */
203 const char comment_chars[] = "#/";
204 #define PREFIX_SEPARATOR '\\'
205
206 /* This array holds the chars that only start a comment at the beginning of
207 a line. If the line seems to have the form '# 123 filename'
208 .line and .file directives will appear in the pre-processed output.
209 Note that input_file.c hand checks for '#' at the beginning of the
210 first line of the input file. This is because the compiler outputs
211 #NO_APP at the beginning of its output.
212 Also note that comments started like this one will always work if
213 '/' isn't otherwise defined. */
214 const char line_comment_chars[] = "#";
215
216 #else
217 /* Putting '/' here makes it impossible to use the divide operator.
218 However, we need it for compatibility with SVR4 systems. */
219 const char comment_chars[] = "#";
220 #define PREFIX_SEPARATOR '/'
221
222 const char line_comment_chars[] = "/#";
223 #endif
224
225 const char line_separator_chars[] = ";";
226
227 /* Chars that can be used to separate mant from exp in floating point
228 nums. */
229 const char EXP_CHARS[] = "eE";
230
231 /* Chars that mean this number is a floating point constant
232 As in 0f12.456
233 or 0d1.2345e12. */
234 const char FLT_CHARS[] = "fFdDxX";
235
236 /* Tables for lexical analysis. */
237 static char mnemonic_chars[256];
238 static char register_chars[256];
239 static char operand_chars[256];
240 static char identifier_chars[256];
241 static char digit_chars[256];
242
243 /* Lexical macros. */
244 #define is_mnemonic_char(x) (mnemonic_chars[(unsigned char) x])
245 #define is_operand_char(x) (operand_chars[(unsigned char) x])
246 #define is_register_char(x) (register_chars[(unsigned char) x])
247 #define is_space_char(x) ((x) == ' ')
248 #define is_identifier_char(x) (identifier_chars[(unsigned char) x])
249 #define is_digit_char(x) (digit_chars[(unsigned char) x])
250
251 /* All non-digit non-letter characters that may occur in an operand. */
252 static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
253
254 /* md_assemble() always leaves the strings it's passed unaltered. To
255 effect this we maintain a stack of saved characters that we've smashed
256 with '\0's (indicating end of strings for various sub-fields of the
257 assembler instruction). */
258 static char save_stack[32];
259 static char *save_stack_p;
260 #define END_STRING_AND_SAVE(s) \
261 do { *save_stack_p++ = *(s); *(s) = '\0'; } while (0)
262 #define RESTORE_END_STRING(s) \
263 do { *(s) = *--save_stack_p; } while (0)
264
265 /* The instruction we're assembling. */
266 static i386_insn i;
267
268 /* Possible templates for current insn. */
269 static const templates *current_templates;
270
271 /* Per instruction expressionS buffers: 2 displacements & 2 immediate max. */
272 static expressionS disp_expressions[2], im_expressions[2];
273
274 /* Current operand we are working on. */
275 static int this_operand;
276
277 /* We support four different modes. FLAG_CODE variable is used to distinguish
278 these. */
279
280 enum flag_code {
281 CODE_32BIT,
282 CODE_16BIT,
283 CODE_64BIT };
284 #define NUM_FLAG_CODE ((int) CODE_64BIT + 1)
285
286 static enum flag_code flag_code;
287 static int use_rela_relocations = 0;
288
289 /* The names used to print error messages. */
290 static const char *flag_code_names[] =
291 {
292 "32",
293 "16",
294 "64"
295 };
296
297 /* 1 for intel syntax,
298 0 if att syntax. */
299 static int intel_syntax = 0;
300
301 /* 1 if register prefix % not required. */
302 static int allow_naked_reg = 0;
303
304 /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
305 leave, push, and pop instructions so that gcc has the same stack
306 frame as in 32 bit mode. */
307 static char stackop_size = '\0';
308
309 /* Non-zero to optimize code alignment. */
310 int optimize_align_code = 1;
311
312 /* Non-zero to quieten some warnings. */
313 static int quiet_warnings = 0;
314
315 /* CPU name. */
316 static const char *cpu_arch_name = NULL;
317 static const char *cpu_sub_arch_name = NULL;
318
319 /* CPU feature flags. */
320 static unsigned int cpu_arch_flags = CpuUnknownFlags | CpuNo64;
321
322 /* If set, conditional jumps are not automatically promoted to handle
323 larger than a byte offset. */
324 static unsigned int no_cond_jump_promotion = 0;
325
326 /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
327 symbolS *GOT_symbol;
328
329 /* The dwarf2 return column, adjusted for 32 or 64 bit. */
330 unsigned int x86_dwarf2_return_column;
331
332 /* The dwarf2 data alignment, adjusted for 32 or 64 bit. */
333 int x86_cie_data_alignment;
334
335 /* Interface to relax_segment.
336 There are 3 major relax states for 386 jump insns because the
337 different types of jumps add different sizes to frags when we're
338 figuring out what sort of jump to choose to reach a given label. */
339
340 /* Types. */
341 #define UNCOND_JUMP 0
342 #define COND_JUMP 1
343 #define COND_JUMP86 2
344
345 /* Sizes. */
346 #define CODE16 1
347 #define SMALL 0
348 #define SMALL16 (SMALL | CODE16)
349 #define BIG 2
350 #define BIG16 (BIG | CODE16)
351
352 #ifndef INLINE
353 #ifdef __GNUC__
354 #define INLINE __inline__
355 #else
356 #define INLINE
357 #endif
358 #endif
359
360 #define ENCODE_RELAX_STATE(type, size) \
361 ((relax_substateT) (((type) << 2) | (size)))
362 #define TYPE_FROM_RELAX_STATE(s) \
363 ((s) >> 2)
364 #define DISP_SIZE_FROM_RELAX_STATE(s) \
365 ((((s) & 3) == BIG ? 4 : (((s) & 3) == BIG16 ? 2 : 1)))
366
367 /* This table is used by relax_frag to promote short jumps to long
368 ones where necessary. SMALL (short) jumps may be promoted to BIG
369 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
370 don't allow a short jump in a 32 bit code segment to be promoted to
371 a 16 bit offset jump because it's slower (requires data size
372 prefix), and doesn't work, unless the destination is in the bottom
373 64k of the code segment (The top 16 bits of eip are zeroed). */
374
375 const relax_typeS md_relax_table[] =
376 {
377 /* The fields are:
378 1) most positive reach of this state,
379 2) most negative reach of this state,
380 3) how many bytes this mode will have in the variable part of the frag
381 4) which index into the table to try if we can't fit into this one. */
382
383 /* UNCOND_JUMP states. */
384 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
385 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
386 /* dword jmp adds 4 bytes to frag:
387 0 extra opcode bytes, 4 displacement bytes. */
388 {0, 0, 4, 0},
389 /* word jmp adds 2 byte2 to frag:
390 0 extra opcode bytes, 2 displacement bytes. */
391 {0, 0, 2, 0},
392
393 /* COND_JUMP states. */
394 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
395 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
396 /* dword conditionals adds 5 bytes to frag:
397 1 extra opcode byte, 4 displacement bytes. */
398 {0, 0, 5, 0},
399 /* word conditionals add 3 bytes to frag:
400 1 extra opcode byte, 2 displacement bytes. */
401 {0, 0, 3, 0},
402
403 /* COND_JUMP86 states. */
404 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG)},
405 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG16)},
406 /* dword conditionals adds 5 bytes to frag:
407 1 extra opcode byte, 4 displacement bytes. */
408 {0, 0, 5, 0},
409 /* word conditionals add 4 bytes to frag:
410 1 displacement byte and a 3 byte long branch insn. */
411 {0, 0, 4, 0}
412 };
413
414 static const arch_entry cpu_arch[] = {
415 {"i8086", Cpu086 },
416 {"i186", Cpu086|Cpu186 },
417 {"i286", Cpu086|Cpu186|Cpu286 },
418 {"i386", Cpu086|Cpu186|Cpu286|Cpu386 },
419 {"i486", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486 },
420 {"i586", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586 },
421 {"i686", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686 },
422 {"pentium", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586 },
423 {"pentiumpro",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686 },
424 {"pentiumii", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX },
425 {"pentiumiii",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuMMX2|CpuSSE },
426 {"pentium4", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2|CpuSSE|CpuSSE2 },
427 {"prescott", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuPNI },
428 {"k6", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX },
429 {"k6_2", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow },
430 {"athlon", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA },
431 {"sledgehammer",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2 },
432 {".mmx", CpuMMX },
433 {".sse", CpuMMX|CpuMMX2|CpuSSE },
434 {".sse2", CpuMMX|CpuMMX2|CpuSSE|CpuSSE2 },
435 {".3dnow", CpuMMX|Cpu3dnow },
436 {".3dnowa", CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA },
437 {".padlock", CpuPadLock },
438 {NULL, 0 }
439 };
440
441 const pseudo_typeS md_pseudo_table[] =
442 {
443 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
444 {"align", s_align_bytes, 0},
445 #else
446 {"align", s_align_ptwo, 0},
447 #endif
448 {"arch", set_cpu_arch, 0},
449 #ifndef I386COFF
450 {"bss", s_bss, 0},
451 #endif
452 {"ffloat", float_cons, 'f'},
453 {"dfloat", float_cons, 'd'},
454 {"tfloat", float_cons, 'x'},
455 {"value", cons, 2},
456 {"noopt", s_ignore, 0},
457 {"optim", s_ignore, 0},
458 {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
459 {"code16", set_code_flag, CODE_16BIT},
460 {"code32", set_code_flag, CODE_32BIT},
461 {"code64", set_code_flag, CODE_64BIT},
462 {"intel_syntax", set_intel_syntax, 1},
463 {"att_syntax", set_intel_syntax, 0},
464 {"file", (void (*) PARAMS ((int))) dwarf2_directive_file, 0},
465 {"loc", dwarf2_directive_loc, 0},
466 #ifdef TE_PE
467 {"secrel32", pe_directive_secrel, 0},
468 #endif
469 {0, 0, 0}
470 };
471
472 /* For interface with expression (). */
473 extern char *input_line_pointer;
474
475 /* Hash table for instruction mnemonic lookup. */
476 static struct hash_control *op_hash;
477
478 /* Hash table for register lookup. */
479 static struct hash_control *reg_hash;
480 \f
481 void
482 i386_align_code (fragP, count)
483 fragS *fragP;
484 int count;
485 {
486 /* Various efficient no-op patterns for aligning code labels.
487 Note: Don't try to assemble the instructions in the comments.
488 0L and 0w are not legal. */
489 static const char f32_1[] =
490 {0x90}; /* nop */
491 static const char f32_2[] =
492 {0x89,0xf6}; /* movl %esi,%esi */
493 static const char f32_3[] =
494 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
495 static const char f32_4[] =
496 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
497 static const char f32_5[] =
498 {0x90, /* nop */
499 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
500 static const char f32_6[] =
501 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
502 static const char f32_7[] =
503 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
504 static const char f32_8[] =
505 {0x90, /* nop */
506 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
507 static const char f32_9[] =
508 {0x89,0xf6, /* movl %esi,%esi */
509 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
510 static const char f32_10[] =
511 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
512 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
513 static const char f32_11[] =
514 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
515 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
516 static const char f32_12[] =
517 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
518 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
519 static const char f32_13[] =
520 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
521 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
522 static const char f32_14[] =
523 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
524 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
525 static const char f32_15[] =
526 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
527 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
528 static const char f16_3[] =
529 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
530 static const char f16_4[] =
531 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
532 static const char f16_5[] =
533 {0x90, /* nop */
534 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
535 static const char f16_6[] =
536 {0x89,0xf6, /* mov %si,%si */
537 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
538 static const char f16_7[] =
539 {0x8d,0x74,0x00, /* lea 0(%si),%si */
540 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
541 static const char f16_8[] =
542 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
543 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
544 static const char *const f32_patt[] = {
545 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
546 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
547 };
548 static const char *const f16_patt[] = {
549 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
550 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
551 };
552
553 if (count <= 0 || count > 15)
554 return;
555
556 /* The recommended way to pad 64bit code is to use NOPs preceded by
557 maximally four 0x66 prefixes. Balance the size of nops. */
558 if (flag_code == CODE_64BIT)
559 {
560 int i;
561 int nnops = (count + 3) / 4;
562 int len = count / nnops;
563 int remains = count - nnops * len;
564 int pos = 0;
565
566 for (i = 0; i < remains; i++)
567 {
568 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len);
569 fragP->fr_literal[fragP->fr_fix + pos + len] = 0x90;
570 pos += len + 1;
571 }
572 for (; i < nnops; i++)
573 {
574 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len - 1);
575 fragP->fr_literal[fragP->fr_fix + pos + len - 1] = 0x90;
576 pos += len;
577 }
578 }
579 else
580 if (flag_code == CODE_16BIT)
581 {
582 memcpy (fragP->fr_literal + fragP->fr_fix,
583 f16_patt[count - 1], count);
584 if (count > 8)
585 /* Adjust jump offset. */
586 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
587 }
588 else
589 memcpy (fragP->fr_literal + fragP->fr_fix,
590 f32_patt[count - 1], count);
591 fragP->fr_var = count;
592 }
593
594 static INLINE unsigned int
595 mode_from_disp_size (t)
596 unsigned int t;
597 {
598 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
599 }
600
601 static INLINE int
602 fits_in_signed_byte (num)
603 offsetT num;
604 {
605 return (num >= -128) && (num <= 127);
606 }
607
608 static INLINE int
609 fits_in_unsigned_byte (num)
610 offsetT num;
611 {
612 return (num & 0xff) == num;
613 }
614
615 static INLINE int
616 fits_in_unsigned_word (num)
617 offsetT num;
618 {
619 return (num & 0xffff) == num;
620 }
621
622 static INLINE int
623 fits_in_signed_word (num)
624 offsetT num;
625 {
626 return (-32768 <= num) && (num <= 32767);
627 }
628 static INLINE int
629 fits_in_signed_long (num)
630 offsetT num ATTRIBUTE_UNUSED;
631 {
632 #ifndef BFD64
633 return 1;
634 #else
635 return (!(((offsetT) -1 << 31) & num)
636 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
637 #endif
638 } /* fits_in_signed_long() */
639 static INLINE int
640 fits_in_unsigned_long (num)
641 offsetT num ATTRIBUTE_UNUSED;
642 {
643 #ifndef BFD64
644 return 1;
645 #else
646 return (num & (((offsetT) 2 << 31) - 1)) == num;
647 #endif
648 } /* fits_in_unsigned_long() */
649
650 static int
651 smallest_imm_type (num)
652 offsetT num;
653 {
654 if (cpu_arch_flags != (Cpu086 | Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64))
655 {
656 /* This code is disabled on the 486 because all the Imm1 forms
657 in the opcode table are slower on the i486. They're the
658 versions with the implicitly specified single-position
659 displacement, which has another syntax if you really want to
660 use that form. */
661 if (num == 1)
662 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
663 }
664 return (fits_in_signed_byte (num)
665 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
666 : fits_in_unsigned_byte (num)
667 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
668 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
669 ? (Imm16 | Imm32 | Imm32S | Imm64)
670 : fits_in_signed_long (num)
671 ? (Imm32 | Imm32S | Imm64)
672 : fits_in_unsigned_long (num)
673 ? (Imm32 | Imm64)
674 : Imm64);
675 }
676
677 static offsetT
678 offset_in_range (val, size)
679 offsetT val;
680 int size;
681 {
682 addressT mask;
683
684 switch (size)
685 {
686 case 1: mask = ((addressT) 1 << 8) - 1; break;
687 case 2: mask = ((addressT) 1 << 16) - 1; break;
688 case 4: mask = ((addressT) 2 << 31) - 1; break;
689 #ifdef BFD64
690 case 8: mask = ((addressT) 2 << 63) - 1; break;
691 #endif
692 default: abort ();
693 }
694
695 /* If BFD64, sign extend val. */
696 if (!use_rela_relocations)
697 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
698 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
699
700 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
701 {
702 char buf1[40], buf2[40];
703
704 sprint_value (buf1, val);
705 sprint_value (buf2, val & mask);
706 as_warn (_("%s shortened to %s"), buf1, buf2);
707 }
708 return val & mask;
709 }
710
711 /* Returns 0 if attempting to add a prefix where one from the same
712 class already exists, 1 if non rep/repne added, 2 if rep/repne
713 added. */
714 static int
715 add_prefix (prefix)
716 unsigned int prefix;
717 {
718 int ret = 1;
719 int q;
720
721 if (prefix >= REX_OPCODE && prefix < REX_OPCODE + 16
722 && flag_code == CODE_64BIT)
723 q = REX_PREFIX;
724 else
725 switch (prefix)
726 {
727 default:
728 abort ();
729
730 case CS_PREFIX_OPCODE:
731 case DS_PREFIX_OPCODE:
732 case ES_PREFIX_OPCODE:
733 case FS_PREFIX_OPCODE:
734 case GS_PREFIX_OPCODE:
735 case SS_PREFIX_OPCODE:
736 q = SEG_PREFIX;
737 break;
738
739 case REPNE_PREFIX_OPCODE:
740 case REPE_PREFIX_OPCODE:
741 ret = 2;
742 /* fall thru */
743 case LOCK_PREFIX_OPCODE:
744 q = LOCKREP_PREFIX;
745 break;
746
747 case FWAIT_OPCODE:
748 q = WAIT_PREFIX;
749 break;
750
751 case ADDR_PREFIX_OPCODE:
752 q = ADDR_PREFIX;
753 break;
754
755 case DATA_PREFIX_OPCODE:
756 q = DATA_PREFIX;
757 break;
758 }
759
760 if (i.prefix[q] != 0)
761 {
762 as_bad (_("same type of prefix used twice"));
763 return 0;
764 }
765
766 i.prefixes += 1;
767 i.prefix[q] = prefix;
768 return ret;
769 }
770
771 static void
772 set_code_flag (value)
773 int value;
774 {
775 flag_code = value;
776 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
777 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
778 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
779 {
780 as_bad (_("64bit mode not supported on this CPU."));
781 }
782 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
783 {
784 as_bad (_("32bit mode not supported on this CPU."));
785 }
786 stackop_size = '\0';
787 }
788
789 static void
790 set_16bit_gcc_code_flag (new_code_flag)
791 int new_code_flag;
792 {
793 flag_code = new_code_flag;
794 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
795 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
796 stackop_size = LONG_MNEM_SUFFIX;
797 }
798
799 static void
800 set_intel_syntax (syntax_flag)
801 int syntax_flag;
802 {
803 /* Find out if register prefixing is specified. */
804 int ask_naked_reg = 0;
805
806 SKIP_WHITESPACE ();
807 if (!is_end_of_line[(unsigned char) *input_line_pointer])
808 {
809 char *string = input_line_pointer;
810 int e = get_symbol_end ();
811
812 if (strcmp (string, "prefix") == 0)
813 ask_naked_reg = 1;
814 else if (strcmp (string, "noprefix") == 0)
815 ask_naked_reg = -1;
816 else
817 as_bad (_("bad argument to syntax directive."));
818 *input_line_pointer = e;
819 }
820 demand_empty_rest_of_line ();
821
822 intel_syntax = syntax_flag;
823
824 if (ask_naked_reg == 0)
825 allow_naked_reg = (intel_syntax
826 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
827 else
828 allow_naked_reg = (ask_naked_reg < 0);
829
830 identifier_chars['%'] = intel_syntax && allow_naked_reg ? '%' : 0;
831 identifier_chars['$'] = intel_syntax ? '$' : 0;
832 }
833
834 static void
835 set_cpu_arch (dummy)
836 int dummy ATTRIBUTE_UNUSED;
837 {
838 SKIP_WHITESPACE ();
839
840 if (!is_end_of_line[(unsigned char) *input_line_pointer])
841 {
842 char *string = input_line_pointer;
843 int e = get_symbol_end ();
844 int i;
845
846 for (i = 0; cpu_arch[i].name; i++)
847 {
848 if (strcmp (string, cpu_arch[i].name) == 0)
849 {
850 if (*string != '.')
851 {
852 cpu_arch_name = cpu_arch[i].name;
853 cpu_sub_arch_name = NULL;
854 cpu_arch_flags = (cpu_arch[i].flags
855 | (flag_code == CODE_64BIT ? Cpu64 : CpuNo64));
856 break;
857 }
858 if ((cpu_arch_flags | cpu_arch[i].flags) != cpu_arch_flags)
859 {
860 cpu_sub_arch_name = cpu_arch[i].name;
861 cpu_arch_flags |= cpu_arch[i].flags;
862 }
863 *input_line_pointer = e;
864 demand_empty_rest_of_line ();
865 return;
866 }
867 }
868 if (!cpu_arch[i].name)
869 as_bad (_("no such architecture: `%s'"), string);
870
871 *input_line_pointer = e;
872 }
873 else
874 as_bad (_("missing cpu architecture"));
875
876 no_cond_jump_promotion = 0;
877 if (*input_line_pointer == ','
878 && !is_end_of_line[(unsigned char) input_line_pointer[1]])
879 {
880 char *string = ++input_line_pointer;
881 int e = get_symbol_end ();
882
883 if (strcmp (string, "nojumps") == 0)
884 no_cond_jump_promotion = 1;
885 else if (strcmp (string, "jumps") == 0)
886 ;
887 else
888 as_bad (_("no such architecture modifier: `%s'"), string);
889
890 *input_line_pointer = e;
891 }
892
893 demand_empty_rest_of_line ();
894 }
895
896 unsigned long
897 i386_mach ()
898 {
899 if (!strcmp (default_arch, "x86_64"))
900 return bfd_mach_x86_64;
901 else if (!strcmp (default_arch, "i386"))
902 return bfd_mach_i386_i386;
903 else
904 as_fatal (_("Unknown architecture"));
905 }
906 \f
907 void
908 md_begin ()
909 {
910 const char *hash_err;
911
912 /* Initialize op_hash hash table. */
913 op_hash = hash_new ();
914
915 {
916 const template *optab;
917 templates *core_optab;
918
919 /* Setup for loop. */
920 optab = i386_optab;
921 core_optab = (templates *) xmalloc (sizeof (templates));
922 core_optab->start = optab;
923
924 while (1)
925 {
926 ++optab;
927 if (optab->name == NULL
928 || strcmp (optab->name, (optab - 1)->name) != 0)
929 {
930 /* different name --> ship out current template list;
931 add to hash table; & begin anew. */
932 core_optab->end = optab;
933 hash_err = hash_insert (op_hash,
934 (optab - 1)->name,
935 (PTR) core_optab);
936 if (hash_err)
937 {
938 as_fatal (_("Internal Error: Can't hash %s: %s"),
939 (optab - 1)->name,
940 hash_err);
941 }
942 if (optab->name == NULL)
943 break;
944 core_optab = (templates *) xmalloc (sizeof (templates));
945 core_optab->start = optab;
946 }
947 }
948 }
949
950 /* Initialize reg_hash hash table. */
951 reg_hash = hash_new ();
952 {
953 const reg_entry *regtab;
954
955 for (regtab = i386_regtab;
956 regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
957 regtab++)
958 {
959 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
960 if (hash_err)
961 as_fatal (_("Internal Error: Can't hash %s: %s"),
962 regtab->reg_name,
963 hash_err);
964 }
965 }
966
967 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
968 {
969 int c;
970 char *p;
971
972 for (c = 0; c < 256; c++)
973 {
974 if (ISDIGIT (c))
975 {
976 digit_chars[c] = c;
977 mnemonic_chars[c] = c;
978 register_chars[c] = c;
979 operand_chars[c] = c;
980 }
981 else if (ISLOWER (c))
982 {
983 mnemonic_chars[c] = c;
984 register_chars[c] = c;
985 operand_chars[c] = c;
986 }
987 else if (ISUPPER (c))
988 {
989 mnemonic_chars[c] = TOLOWER (c);
990 register_chars[c] = mnemonic_chars[c];
991 operand_chars[c] = c;
992 }
993
994 if (ISALPHA (c) || ISDIGIT (c))
995 identifier_chars[c] = c;
996 else if (c >= 128)
997 {
998 identifier_chars[c] = c;
999 operand_chars[c] = c;
1000 }
1001 }
1002
1003 #ifdef LEX_AT
1004 identifier_chars['@'] = '@';
1005 #endif
1006 #ifdef LEX_QM
1007 identifier_chars['?'] = '?';
1008 operand_chars['?'] = '?';
1009 #endif
1010 digit_chars['-'] = '-';
1011 identifier_chars['_'] = '_';
1012 identifier_chars['.'] = '.';
1013
1014 for (p = operand_special_chars; *p != '\0'; p++)
1015 operand_chars[(unsigned char) *p] = *p;
1016 }
1017
1018 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1019 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
1020 {
1021 record_alignment (text_section, 2);
1022 record_alignment (data_section, 2);
1023 record_alignment (bss_section, 2);
1024 }
1025 #endif
1026
1027 if (flag_code == CODE_64BIT)
1028 {
1029 x86_dwarf2_return_column = 16;
1030 x86_cie_data_alignment = -8;
1031 }
1032 else
1033 {
1034 x86_dwarf2_return_column = 8;
1035 x86_cie_data_alignment = -4;
1036 }
1037 }
1038
1039 void
1040 i386_print_statistics (file)
1041 FILE *file;
1042 {
1043 hash_print_statistics (file, "i386 opcode", op_hash);
1044 hash_print_statistics (file, "i386 register", reg_hash);
1045 }
1046 \f
1047 #ifdef DEBUG386
1048
1049 /* Debugging routines for md_assemble. */
1050 static void pi PARAMS ((char *, i386_insn *));
1051 static void pte PARAMS ((template *));
1052 static void pt PARAMS ((unsigned int));
1053 static void pe PARAMS ((expressionS *));
1054 static void ps PARAMS ((symbolS *));
1055
1056 static void
1057 pi (line, x)
1058 char *line;
1059 i386_insn *x;
1060 {
1061 unsigned int i;
1062
1063 fprintf (stdout, "%s: template ", line);
1064 pte (&x->tm);
1065 fprintf (stdout, " address: base %s index %s scale %x\n",
1066 x->base_reg ? x->base_reg->reg_name : "none",
1067 x->index_reg ? x->index_reg->reg_name : "none",
1068 x->log2_scale_factor);
1069 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
1070 x->rm.mode, x->rm.reg, x->rm.regmem);
1071 fprintf (stdout, " sib: base %x index %x scale %x\n",
1072 x->sib.base, x->sib.index, x->sib.scale);
1073 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
1074 (x->rex & REX_MODE64) != 0,
1075 (x->rex & REX_EXTX) != 0,
1076 (x->rex & REX_EXTY) != 0,
1077 (x->rex & REX_EXTZ) != 0);
1078 for (i = 0; i < x->operands; i++)
1079 {
1080 fprintf (stdout, " #%d: ", i + 1);
1081 pt (x->types[i]);
1082 fprintf (stdout, "\n");
1083 if (x->types[i]
1084 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
1085 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
1086 if (x->types[i] & Imm)
1087 pe (x->op[i].imms);
1088 if (x->types[i] & Disp)
1089 pe (x->op[i].disps);
1090 }
1091 }
1092
1093 static void
1094 pte (t)
1095 template *t;
1096 {
1097 unsigned int i;
1098 fprintf (stdout, " %d operands ", t->operands);
1099 fprintf (stdout, "opcode %x ", t->base_opcode);
1100 if (t->extension_opcode != None)
1101 fprintf (stdout, "ext %x ", t->extension_opcode);
1102 if (t->opcode_modifier & D)
1103 fprintf (stdout, "D");
1104 if (t->opcode_modifier & W)
1105 fprintf (stdout, "W");
1106 fprintf (stdout, "\n");
1107 for (i = 0; i < t->operands; i++)
1108 {
1109 fprintf (stdout, " #%d type ", i + 1);
1110 pt (t->operand_types[i]);
1111 fprintf (stdout, "\n");
1112 }
1113 }
1114
1115 static void
1116 pe (e)
1117 expressionS *e;
1118 {
1119 fprintf (stdout, " operation %d\n", e->X_op);
1120 fprintf (stdout, " add_number %ld (%lx)\n",
1121 (long) e->X_add_number, (long) e->X_add_number);
1122 if (e->X_add_symbol)
1123 {
1124 fprintf (stdout, " add_symbol ");
1125 ps (e->X_add_symbol);
1126 fprintf (stdout, "\n");
1127 }
1128 if (e->X_op_symbol)
1129 {
1130 fprintf (stdout, " op_symbol ");
1131 ps (e->X_op_symbol);
1132 fprintf (stdout, "\n");
1133 }
1134 }
1135
1136 static void
1137 ps (s)
1138 symbolS *s;
1139 {
1140 fprintf (stdout, "%s type %s%s",
1141 S_GET_NAME (s),
1142 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
1143 segment_name (S_GET_SEGMENT (s)));
1144 }
1145
1146 struct type_name
1147 {
1148 unsigned int mask;
1149 char *tname;
1150 }
1151
1152 static const type_names[] =
1153 {
1154 { Reg8, "r8" },
1155 { Reg16, "r16" },
1156 { Reg32, "r32" },
1157 { Reg64, "r64" },
1158 { Imm8, "i8" },
1159 { Imm8S, "i8s" },
1160 { Imm16, "i16" },
1161 { Imm32, "i32" },
1162 { Imm32S, "i32s" },
1163 { Imm64, "i64" },
1164 { Imm1, "i1" },
1165 { BaseIndex, "BaseIndex" },
1166 { Disp8, "d8" },
1167 { Disp16, "d16" },
1168 { Disp32, "d32" },
1169 { Disp32S, "d32s" },
1170 { Disp64, "d64" },
1171 { InOutPortReg, "InOutPortReg" },
1172 { ShiftCount, "ShiftCount" },
1173 { Control, "control reg" },
1174 { Test, "test reg" },
1175 { Debug, "debug reg" },
1176 { FloatReg, "FReg" },
1177 { FloatAcc, "FAcc" },
1178 { SReg2, "SReg2" },
1179 { SReg3, "SReg3" },
1180 { Acc, "Acc" },
1181 { JumpAbsolute, "Jump Absolute" },
1182 { RegMMX, "rMMX" },
1183 { RegXMM, "rXMM" },
1184 { EsSeg, "es" },
1185 { 0, "" }
1186 };
1187
1188 static void
1189 pt (t)
1190 unsigned int t;
1191 {
1192 const struct type_name *ty;
1193
1194 for (ty = type_names; ty->mask; ty++)
1195 if (t & ty->mask)
1196 fprintf (stdout, "%s, ", ty->tname);
1197 fflush (stdout);
1198 }
1199
1200 #endif /* DEBUG386 */
1201 \f
1202 static bfd_reloc_code_real_type reloc
1203 PARAMS ((int, int, int, bfd_reloc_code_real_type));
1204
1205 static bfd_reloc_code_real_type
1206 reloc (size, pcrel, sign, other)
1207 int size;
1208 int pcrel;
1209 int sign;
1210 bfd_reloc_code_real_type other;
1211 {
1212 if (other != NO_RELOC)
1213 return other;
1214
1215 if (pcrel)
1216 {
1217 if (!sign)
1218 as_bad (_("There are no unsigned pc-relative relocations"));
1219 switch (size)
1220 {
1221 case 1: return BFD_RELOC_8_PCREL;
1222 case 2: return BFD_RELOC_16_PCREL;
1223 case 4: return BFD_RELOC_32_PCREL;
1224 }
1225 as_bad (_("can not do %d byte pc-relative relocation"), size);
1226 }
1227 else
1228 {
1229 if (sign)
1230 switch (size)
1231 {
1232 case 4: return BFD_RELOC_X86_64_32S;
1233 }
1234 else
1235 switch (size)
1236 {
1237 case 1: return BFD_RELOC_8;
1238 case 2: return BFD_RELOC_16;
1239 case 4: return BFD_RELOC_32;
1240 case 8: return BFD_RELOC_64;
1241 }
1242 as_bad (_("can not do %s %d byte relocation"),
1243 sign ? "signed" : "unsigned", size);
1244 }
1245
1246 abort ();
1247 return BFD_RELOC_NONE;
1248 }
1249
1250 /* Here we decide which fixups can be adjusted to make them relative to
1251 the beginning of the section instead of the symbol. Basically we need
1252 to make sure that the dynamic relocations are done correctly, so in
1253 some cases we force the original symbol to be used. */
1254
1255 int
1256 tc_i386_fix_adjustable (fixP)
1257 fixS *fixP ATTRIBUTE_UNUSED;
1258 {
1259 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1260 if (OUTPUT_FLAVOR != bfd_target_elf_flavour)
1261 return 1;
1262
1263 /* Don't adjust pc-relative references to merge sections in 64-bit
1264 mode. */
1265 if (use_rela_relocations
1266 && (S_GET_SEGMENT (fixP->fx_addsy)->flags & SEC_MERGE) != 0
1267 && fixP->fx_pcrel)
1268 return 0;
1269
1270 /* The x86_64 GOTPCREL are represented as 32bit PCrel relocations
1271 and changed later by validate_fix. */
1272 if (GOT_symbol && fixP->fx_subsy == GOT_symbol
1273 && fixP->fx_r_type == BFD_RELOC_32_PCREL)
1274 return 0;
1275
1276 /* adjust_reloc_syms doesn't know about the GOT. */
1277 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1278 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1279 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1280 || fixP->fx_r_type == BFD_RELOC_386_TLS_GD
1281 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDM
1282 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDO_32
1283 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE_32
1284 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE
1285 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTIE
1286 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE_32
1287 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE
1288 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1289 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1290 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
1291 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSGD
1292 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSLD
1293 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF32
1294 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTTPOFF
1295 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF32
1296 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1297 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1298 return 0;
1299 #endif
1300 return 1;
1301 }
1302
1303 static int intel_float_operand PARAMS ((const char *mnemonic));
1304
1305 static int
1306 intel_float_operand (mnemonic)
1307 const char *mnemonic;
1308 {
1309 /* Note that the value returned is meaningful only for opcodes with (memory)
1310 operands, hence the code here is free to improperly handle opcodes that
1311 have no operands (for better performance and smaller code). */
1312
1313 if (mnemonic[0] != 'f')
1314 return 0; /* non-math */
1315
1316 switch (mnemonic[1])
1317 {
1318 /* fclex, fdecstp, fdisi, femms, feni, fincstp, finit, fsetpm, and
1319 the fs segment override prefix not currently handled because no
1320 call path can make opcodes without operands get here */
1321 case 'i':
1322 return 2 /* integer op */;
1323 case 'l':
1324 if (mnemonic[2] == 'd' && (mnemonic[3] == 'c' || mnemonic[3] == 'e'))
1325 return 3; /* fldcw/fldenv */
1326 break;
1327 case 'n':
1328 if (mnemonic[2] != 'o' /* fnop */)
1329 return 3; /* non-waiting control op */
1330 break;
1331 case 'r':
1332 if (mnemonic[2] == 's')
1333 return 3; /* frstor/frstpm */
1334 break;
1335 case 's':
1336 if (mnemonic[2] == 'a')
1337 return 3; /* fsave */
1338 if (mnemonic[2] == 't')
1339 {
1340 switch (mnemonic[3])
1341 {
1342 case 'c': /* fstcw */
1343 case 'd': /* fstdw */
1344 case 'e': /* fstenv */
1345 case 's': /* fsts[gw] */
1346 return 3;
1347 }
1348 }
1349 break;
1350 case 'x':
1351 if (mnemonic[2] == 'r' || mnemonic[2] == 's')
1352 return 0; /* fxsave/fxrstor are not really math ops */
1353 break;
1354 }
1355
1356 return 1;
1357 }
1358
1359 /* This is the guts of the machine-dependent assembler. LINE points to a
1360 machine dependent instruction. This function is supposed to emit
1361 the frags/bytes it assembles to. */
1362
1363 void
1364 md_assemble (line)
1365 char *line;
1366 {
1367 int j;
1368 char mnemonic[MAX_MNEM_SIZE];
1369
1370 /* Initialize globals. */
1371 memset (&i, '\0', sizeof (i));
1372 for (j = 0; j < MAX_OPERANDS; j++)
1373 i.reloc[j] = NO_RELOC;
1374 memset (disp_expressions, '\0', sizeof (disp_expressions));
1375 memset (im_expressions, '\0', sizeof (im_expressions));
1376 save_stack_p = save_stack;
1377
1378 /* First parse an instruction mnemonic & call i386_operand for the operands.
1379 We assume that the scrubber has arranged it so that line[0] is the valid
1380 start of a (possibly prefixed) mnemonic. */
1381
1382 line = parse_insn (line, mnemonic);
1383 if (line == NULL)
1384 return;
1385
1386 line = parse_operands (line, mnemonic);
1387 if (line == NULL)
1388 return;
1389
1390 /* Now we've parsed the mnemonic into a set of templates, and have the
1391 operands at hand. */
1392
1393 /* All intel opcodes have reversed operands except for "bound" and
1394 "enter". We also don't reverse intersegment "jmp" and "call"
1395 instructions with 2 immediate operands so that the immediate segment
1396 precedes the offset, as it does when in AT&T mode. "enter" and the
1397 intersegment "jmp" and "call" instructions are the only ones that
1398 have two immediate operands. */
1399 if (intel_syntax && i.operands > 1
1400 && (strcmp (mnemonic, "bound") != 0)
1401 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1402 swap_operands ();
1403
1404 if (i.imm_operands)
1405 optimize_imm ();
1406
1407 if (i.disp_operands)
1408 optimize_disp ();
1409
1410 /* Next, we find a template that matches the given insn,
1411 making sure the overlap of the given operands types is consistent
1412 with the template operand types. */
1413
1414 if (!match_template ())
1415 return;
1416
1417 if (intel_syntax)
1418 {
1419 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1420 if (SYSV386_COMPAT
1421 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1422 i.tm.base_opcode ^= FloatR;
1423
1424 /* Zap movzx and movsx suffix. The suffix may have been set from
1425 "word ptr" or "byte ptr" on the source operand, but we'll use
1426 the suffix later to choose the destination register. */
1427 if ((i.tm.base_opcode & ~9) == 0x0fb6)
1428 {
1429 if (i.reg_operands < 2
1430 && !i.suffix
1431 && (~i.tm.opcode_modifier
1432 & (No_bSuf
1433 | No_wSuf
1434 | No_lSuf
1435 | No_sSuf
1436 | No_xSuf
1437 | No_qSuf)))
1438 as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
1439
1440 i.suffix = 0;
1441 }
1442 }
1443
1444 if (i.tm.opcode_modifier & FWait)
1445 if (!add_prefix (FWAIT_OPCODE))
1446 return;
1447
1448 /* Check string instruction segment overrides. */
1449 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1450 {
1451 if (!check_string ())
1452 return;
1453 }
1454
1455 if (!process_suffix ())
1456 return;
1457
1458 /* Make still unresolved immediate matches conform to size of immediate
1459 given in i.suffix. */
1460 if (!finalize_imm ())
1461 return;
1462
1463 if (i.types[0] & Imm1)
1464 i.imm_operands = 0; /* kludge for shift insns. */
1465 if (i.types[0] & ImplicitRegister)
1466 i.reg_operands--;
1467 if (i.types[1] & ImplicitRegister)
1468 i.reg_operands--;
1469 if (i.types[2] & ImplicitRegister)
1470 i.reg_operands--;
1471
1472 if (i.tm.opcode_modifier & ImmExt)
1473 {
1474 expressionS *exp;
1475
1476 if ((i.tm.cpu_flags & CpuPNI) && i.operands > 0)
1477 {
1478 /* These Intel Prescott New Instructions have the fixed
1479 operands with an opcode suffix which is coded in the same
1480 place as an 8-bit immediate field would be. Here we check
1481 those operands and remove them afterwards. */
1482 unsigned int x;
1483
1484 for (x = 0; x < i.operands; x++)
1485 if (i.op[x].regs->reg_num != x)
1486 as_bad (_("can't use register '%%%s' as operand %d in '%s'."),
1487 i.op[x].regs->reg_name, x + 1, i.tm.name);
1488 i.operands = 0;
1489 }
1490
1491 /* These AMD 3DNow! and Intel Katmai New Instructions have an
1492 opcode suffix which is coded in the same place as an 8-bit
1493 immediate field would be. Here we fake an 8-bit immediate
1494 operand from the opcode suffix stored in tm.extension_opcode. */
1495
1496 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
1497
1498 exp = &im_expressions[i.imm_operands++];
1499 i.op[i.operands].imms = exp;
1500 i.types[i.operands++] = Imm8;
1501 exp->X_op = O_constant;
1502 exp->X_add_number = i.tm.extension_opcode;
1503 i.tm.extension_opcode = None;
1504 }
1505
1506 /* For insns with operands there are more diddles to do to the opcode. */
1507 if (i.operands)
1508 {
1509 if (!process_operands ())
1510 return;
1511 }
1512 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
1513 {
1514 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
1515 as_warn (_("translating to `%sp'"), i.tm.name);
1516 }
1517
1518 /* Handle conversion of 'int $3' --> special int3 insn. */
1519 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
1520 {
1521 i.tm.base_opcode = INT3_OPCODE;
1522 i.imm_operands = 0;
1523 }
1524
1525 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
1526 && i.op[0].disps->X_op == O_constant)
1527 {
1528 /* Convert "jmp constant" (and "call constant") to a jump (call) to
1529 the absolute address given by the constant. Since ix86 jumps and
1530 calls are pc relative, we need to generate a reloc. */
1531 i.op[0].disps->X_add_symbol = &abs_symbol;
1532 i.op[0].disps->X_op = O_symbol;
1533 }
1534
1535 if ((i.tm.opcode_modifier & Rex64) != 0)
1536 i.rex |= REX_MODE64;
1537
1538 /* For 8 bit registers we need an empty rex prefix. Also if the
1539 instruction already has a prefix, we need to convert old
1540 registers to new ones. */
1541
1542 if (((i.types[0] & Reg8) != 0
1543 && (i.op[0].regs->reg_flags & RegRex64) != 0)
1544 || ((i.types[1] & Reg8) != 0
1545 && (i.op[1].regs->reg_flags & RegRex64) != 0)
1546 || (((i.types[0] & Reg8) != 0 || (i.types[1] & Reg8) != 0)
1547 && i.rex != 0))
1548 {
1549 int x;
1550
1551 i.rex |= REX_OPCODE;
1552 for (x = 0; x < 2; x++)
1553 {
1554 /* Look for 8 bit operand that uses old registers. */
1555 if ((i.types[x] & Reg8) != 0
1556 && (i.op[x].regs->reg_flags & RegRex64) == 0)
1557 {
1558 /* In case it is "hi" register, give up. */
1559 if (i.op[x].regs->reg_num > 3)
1560 as_bad (_("can't encode register '%%%s' in an instruction requiring REX prefix."),
1561 i.op[x].regs->reg_name);
1562
1563 /* Otherwise it is equivalent to the extended register.
1564 Since the encoding doesn't change this is merely
1565 cosmetic cleanup for debug output. */
1566
1567 i.op[x].regs = i.op[x].regs + 8;
1568 }
1569 }
1570 }
1571
1572 if (i.rex != 0)
1573 add_prefix (REX_OPCODE | i.rex);
1574
1575 /* We are ready to output the insn. */
1576 output_insn ();
1577 }
1578
1579 static char *
1580 parse_insn (line, mnemonic)
1581 char *line;
1582 char *mnemonic;
1583 {
1584 char *l = line;
1585 char *token_start = l;
1586 char *mnem_p;
1587 int supported;
1588 const template *t;
1589
1590 /* Non-zero if we found a prefix only acceptable with string insns. */
1591 const char *expecting_string_instruction = NULL;
1592
1593 while (1)
1594 {
1595 mnem_p = mnemonic;
1596 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1597 {
1598 mnem_p++;
1599 if (mnem_p >= mnemonic + MAX_MNEM_SIZE)
1600 {
1601 as_bad (_("no such instruction: `%s'"), token_start);
1602 return NULL;
1603 }
1604 l++;
1605 }
1606 if (!is_space_char (*l)
1607 && *l != END_OF_INSN
1608 && *l != PREFIX_SEPARATOR
1609 && *l != ',')
1610 {
1611 as_bad (_("invalid character %s in mnemonic"),
1612 output_invalid (*l));
1613 return NULL;
1614 }
1615 if (token_start == l)
1616 {
1617 if (*l == PREFIX_SEPARATOR)
1618 as_bad (_("expecting prefix; got nothing"));
1619 else
1620 as_bad (_("expecting mnemonic; got nothing"));
1621 return NULL;
1622 }
1623
1624 /* Look up instruction (or prefix) via hash table. */
1625 current_templates = hash_find (op_hash, mnemonic);
1626
1627 if (*l != END_OF_INSN
1628 && (!is_space_char (*l) || l[1] != END_OF_INSN)
1629 && current_templates
1630 && (current_templates->start->opcode_modifier & IsPrefix))
1631 {
1632 /* If we are in 16-bit mode, do not allow addr16 or data16.
1633 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1634 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1635 && flag_code != CODE_64BIT
1636 && (((current_templates->start->opcode_modifier & Size32) != 0)
1637 ^ (flag_code == CODE_16BIT)))
1638 {
1639 as_bad (_("redundant %s prefix"),
1640 current_templates->start->name);
1641 return NULL;
1642 }
1643 /* Add prefix, checking for repeated prefixes. */
1644 switch (add_prefix (current_templates->start->base_opcode))
1645 {
1646 case 0:
1647 return NULL;
1648 case 2:
1649 expecting_string_instruction = current_templates->start->name;
1650 break;
1651 }
1652 /* Skip past PREFIX_SEPARATOR and reset token_start. */
1653 token_start = ++l;
1654 }
1655 else
1656 break;
1657 }
1658
1659 if (!current_templates)
1660 {
1661 /* See if we can get a match by trimming off a suffix. */
1662 switch (mnem_p[-1])
1663 {
1664 case WORD_MNEM_SUFFIX:
1665 if (intel_syntax && (intel_float_operand (mnemonic) & 2))
1666 i.suffix = SHORT_MNEM_SUFFIX;
1667 else
1668 case BYTE_MNEM_SUFFIX:
1669 case QWORD_MNEM_SUFFIX:
1670 i.suffix = mnem_p[-1];
1671 mnem_p[-1] = '\0';
1672 current_templates = hash_find (op_hash, mnemonic);
1673 break;
1674 case SHORT_MNEM_SUFFIX:
1675 case LONG_MNEM_SUFFIX:
1676 if (!intel_syntax)
1677 {
1678 i.suffix = mnem_p[-1];
1679 mnem_p[-1] = '\0';
1680 current_templates = hash_find (op_hash, mnemonic);
1681 }
1682 break;
1683
1684 /* Intel Syntax. */
1685 case 'd':
1686 if (intel_syntax)
1687 {
1688 if (intel_float_operand (mnemonic) == 1)
1689 i.suffix = SHORT_MNEM_SUFFIX;
1690 else
1691 i.suffix = LONG_MNEM_SUFFIX;
1692 mnem_p[-1] = '\0';
1693 current_templates = hash_find (op_hash, mnemonic);
1694 }
1695 break;
1696 }
1697 if (!current_templates)
1698 {
1699 as_bad (_("no such instruction: `%s'"), token_start);
1700 return NULL;
1701 }
1702 }
1703
1704 if (current_templates->start->opcode_modifier & (Jump | JumpByte))
1705 {
1706 /* Check for a branch hint. We allow ",pt" and ",pn" for
1707 predict taken and predict not taken respectively.
1708 I'm not sure that branch hints actually do anything on loop
1709 and jcxz insns (JumpByte) for current Pentium4 chips. They
1710 may work in the future and it doesn't hurt to accept them
1711 now. */
1712 if (l[0] == ',' && l[1] == 'p')
1713 {
1714 if (l[2] == 't')
1715 {
1716 if (!add_prefix (DS_PREFIX_OPCODE))
1717 return NULL;
1718 l += 3;
1719 }
1720 else if (l[2] == 'n')
1721 {
1722 if (!add_prefix (CS_PREFIX_OPCODE))
1723 return NULL;
1724 l += 3;
1725 }
1726 }
1727 }
1728 /* Any other comma loses. */
1729 if (*l == ',')
1730 {
1731 as_bad (_("invalid character %s in mnemonic"),
1732 output_invalid (*l));
1733 return NULL;
1734 }
1735
1736 /* Check if instruction is supported on specified architecture. */
1737 supported = 0;
1738 for (t = current_templates->start; t < current_templates->end; ++t)
1739 {
1740 if (!((t->cpu_flags & ~(Cpu64 | CpuNo64))
1741 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64))))
1742 supported |= 1;
1743 if (!(t->cpu_flags & (flag_code == CODE_64BIT ? CpuNo64 : Cpu64)))
1744 supported |= 2;
1745 }
1746 if (!(supported & 2))
1747 {
1748 as_bad (flag_code == CODE_64BIT
1749 ? _("`%s' is not supported in 64-bit mode")
1750 : _("`%s' is only supported in 64-bit mode"),
1751 current_templates->start->name);
1752 return NULL;
1753 }
1754 if (!(supported & 1))
1755 {
1756 as_warn (_("`%s' is not supported on `%s%s'"),
1757 current_templates->start->name,
1758 cpu_arch_name,
1759 cpu_sub_arch_name ? cpu_sub_arch_name : "");
1760 }
1761 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
1762 {
1763 as_warn (_("use .code16 to ensure correct addressing mode"));
1764 }
1765
1766 /* Check for rep/repne without a string instruction. */
1767 if (expecting_string_instruction
1768 && !(current_templates->start->opcode_modifier & IsString))
1769 {
1770 as_bad (_("expecting string instruction after `%s'"),
1771 expecting_string_instruction);
1772 return NULL;
1773 }
1774
1775 return l;
1776 }
1777
1778 static char *
1779 parse_operands (l, mnemonic)
1780 char *l;
1781 const char *mnemonic;
1782 {
1783 char *token_start;
1784
1785 /* 1 if operand is pending after ','. */
1786 unsigned int expecting_operand = 0;
1787
1788 /* Non-zero if operand parens not balanced. */
1789 unsigned int paren_not_balanced;
1790
1791 while (*l != END_OF_INSN)
1792 {
1793 /* Skip optional white space before operand. */
1794 if (is_space_char (*l))
1795 ++l;
1796 if (!is_operand_char (*l) && *l != END_OF_INSN)
1797 {
1798 as_bad (_("invalid character %s before operand %d"),
1799 output_invalid (*l),
1800 i.operands + 1);
1801 return NULL;
1802 }
1803 token_start = l; /* after white space */
1804 paren_not_balanced = 0;
1805 while (paren_not_balanced || *l != ',')
1806 {
1807 if (*l == END_OF_INSN)
1808 {
1809 if (paren_not_balanced)
1810 {
1811 if (!intel_syntax)
1812 as_bad (_("unbalanced parenthesis in operand %d."),
1813 i.operands + 1);
1814 else
1815 as_bad (_("unbalanced brackets in operand %d."),
1816 i.operands + 1);
1817 return NULL;
1818 }
1819 else
1820 break; /* we are done */
1821 }
1822 else if (!is_operand_char (*l) && !is_space_char (*l))
1823 {
1824 as_bad (_("invalid character %s in operand %d"),
1825 output_invalid (*l),
1826 i.operands + 1);
1827 return NULL;
1828 }
1829 if (!intel_syntax)
1830 {
1831 if (*l == '(')
1832 ++paren_not_balanced;
1833 if (*l == ')')
1834 --paren_not_balanced;
1835 }
1836 else
1837 {
1838 if (*l == '[')
1839 ++paren_not_balanced;
1840 if (*l == ']')
1841 --paren_not_balanced;
1842 }
1843 l++;
1844 }
1845 if (l != token_start)
1846 { /* Yes, we've read in another operand. */
1847 unsigned int operand_ok;
1848 this_operand = i.operands++;
1849 if (i.operands > MAX_OPERANDS)
1850 {
1851 as_bad (_("spurious operands; (%d operands/instruction max)"),
1852 MAX_OPERANDS);
1853 return NULL;
1854 }
1855 /* Now parse operand adding info to 'i' as we go along. */
1856 END_STRING_AND_SAVE (l);
1857
1858 if (intel_syntax)
1859 operand_ok =
1860 i386_intel_operand (token_start,
1861 intel_float_operand (mnemonic));
1862 else
1863 operand_ok = i386_operand (token_start);
1864
1865 RESTORE_END_STRING (l);
1866 if (!operand_ok)
1867 return NULL;
1868 }
1869 else
1870 {
1871 if (expecting_operand)
1872 {
1873 expecting_operand_after_comma:
1874 as_bad (_("expecting operand after ','; got nothing"));
1875 return NULL;
1876 }
1877 if (*l == ',')
1878 {
1879 as_bad (_("expecting operand before ','; got nothing"));
1880 return NULL;
1881 }
1882 }
1883
1884 /* Now *l must be either ',' or END_OF_INSN. */
1885 if (*l == ',')
1886 {
1887 if (*++l == END_OF_INSN)
1888 {
1889 /* Just skip it, if it's \n complain. */
1890 goto expecting_operand_after_comma;
1891 }
1892 expecting_operand = 1;
1893 }
1894 }
1895 return l;
1896 }
1897
1898 static void
1899 swap_operands ()
1900 {
1901 union i386_op temp_op;
1902 unsigned int temp_type;
1903 enum bfd_reloc_code_real temp_reloc;
1904 int xchg1 = 0;
1905 int xchg2 = 0;
1906
1907 if (i.operands == 2)
1908 {
1909 xchg1 = 0;
1910 xchg2 = 1;
1911 }
1912 else if (i.operands == 3)
1913 {
1914 xchg1 = 0;
1915 xchg2 = 2;
1916 }
1917 temp_type = i.types[xchg2];
1918 i.types[xchg2] = i.types[xchg1];
1919 i.types[xchg1] = temp_type;
1920 temp_op = i.op[xchg2];
1921 i.op[xchg2] = i.op[xchg1];
1922 i.op[xchg1] = temp_op;
1923 temp_reloc = i.reloc[xchg2];
1924 i.reloc[xchg2] = i.reloc[xchg1];
1925 i.reloc[xchg1] = temp_reloc;
1926
1927 if (i.mem_operands == 2)
1928 {
1929 const seg_entry *temp_seg;
1930 temp_seg = i.seg[0];
1931 i.seg[0] = i.seg[1];
1932 i.seg[1] = temp_seg;
1933 }
1934 }
1935
1936 /* Try to ensure constant immediates are represented in the smallest
1937 opcode possible. */
1938 static void
1939 optimize_imm ()
1940 {
1941 char guess_suffix = 0;
1942 int op;
1943
1944 if (i.suffix)
1945 guess_suffix = i.suffix;
1946 else if (i.reg_operands)
1947 {
1948 /* Figure out a suffix from the last register operand specified.
1949 We can't do this properly yet, ie. excluding InOutPortReg,
1950 but the following works for instructions with immediates.
1951 In any case, we can't set i.suffix yet. */
1952 for (op = i.operands; --op >= 0;)
1953 if (i.types[op] & Reg)
1954 {
1955 if (i.types[op] & Reg8)
1956 guess_suffix = BYTE_MNEM_SUFFIX;
1957 else if (i.types[op] & Reg16)
1958 guess_suffix = WORD_MNEM_SUFFIX;
1959 else if (i.types[op] & Reg32)
1960 guess_suffix = LONG_MNEM_SUFFIX;
1961 else if (i.types[op] & Reg64)
1962 guess_suffix = QWORD_MNEM_SUFFIX;
1963 break;
1964 }
1965 }
1966 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
1967 guess_suffix = WORD_MNEM_SUFFIX;
1968
1969 for (op = i.operands; --op >= 0;)
1970 if (i.types[op] & Imm)
1971 {
1972 switch (i.op[op].imms->X_op)
1973 {
1974 case O_constant:
1975 /* If a suffix is given, this operand may be shortened. */
1976 switch (guess_suffix)
1977 {
1978 case LONG_MNEM_SUFFIX:
1979 i.types[op] |= Imm32 | Imm64;
1980 break;
1981 case WORD_MNEM_SUFFIX:
1982 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
1983 break;
1984 case BYTE_MNEM_SUFFIX:
1985 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
1986 break;
1987 }
1988
1989 /* If this operand is at most 16 bits, convert it
1990 to a signed 16 bit number before trying to see
1991 whether it will fit in an even smaller size.
1992 This allows a 16-bit operand such as $0xffe0 to
1993 be recognised as within Imm8S range. */
1994 if ((i.types[op] & Imm16)
1995 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
1996 {
1997 i.op[op].imms->X_add_number =
1998 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
1999 }
2000 if ((i.types[op] & Imm32)
2001 && ((i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1))
2002 == 0))
2003 {
2004 i.op[op].imms->X_add_number = ((i.op[op].imms->X_add_number
2005 ^ ((offsetT) 1 << 31))
2006 - ((offsetT) 1 << 31));
2007 }
2008 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
2009
2010 /* We must avoid matching of Imm32 templates when 64bit
2011 only immediate is available. */
2012 if (guess_suffix == QWORD_MNEM_SUFFIX)
2013 i.types[op] &= ~Imm32;
2014 break;
2015
2016 case O_absent:
2017 case O_register:
2018 abort ();
2019
2020 /* Symbols and expressions. */
2021 default:
2022 /* Convert symbolic operand to proper sizes for matching. */
2023 switch (guess_suffix)
2024 {
2025 case QWORD_MNEM_SUFFIX:
2026 i.types[op] = Imm64 | Imm32S;
2027 break;
2028 case LONG_MNEM_SUFFIX:
2029 i.types[op] = Imm32;
2030 break;
2031 case WORD_MNEM_SUFFIX:
2032 i.types[op] = Imm16;
2033 break;
2034 case BYTE_MNEM_SUFFIX:
2035 i.types[op] = Imm8 | Imm8S;
2036 break;
2037 }
2038 break;
2039 }
2040 }
2041 }
2042
2043 /* Try to use the smallest displacement type too. */
2044 static void
2045 optimize_disp ()
2046 {
2047 int op;
2048
2049 for (op = i.operands; --op >= 0;)
2050 if ((i.types[op] & Disp) && i.op[op].disps->X_op == O_constant)
2051 {
2052 offsetT disp = i.op[op].disps->X_add_number;
2053
2054 if (i.types[op] & Disp16)
2055 {
2056 /* We know this operand is at most 16 bits, so
2057 convert to a signed 16 bit number before trying
2058 to see whether it will fit in an even smaller
2059 size. */
2060
2061 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
2062 }
2063 else if (i.types[op] & Disp32)
2064 {
2065 /* We know this operand is at most 32 bits, so convert to a
2066 signed 32 bit number before trying to see whether it will
2067 fit in an even smaller size. */
2068 disp &= (((offsetT) 2 << 31) - 1);
2069 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
2070 }
2071 if (flag_code == CODE_64BIT)
2072 {
2073 if (fits_in_signed_long (disp))
2074 i.types[op] |= Disp32S;
2075 if (fits_in_unsigned_long (disp))
2076 i.types[op] |= Disp32;
2077 }
2078 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
2079 && fits_in_signed_byte (disp))
2080 i.types[op] |= Disp8;
2081 }
2082 }
2083
2084 static int
2085 match_template ()
2086 {
2087 /* Points to template once we've found it. */
2088 const template *t;
2089 unsigned int overlap0, overlap1, overlap2;
2090 unsigned int found_reverse_match;
2091 int suffix_check;
2092
2093 #define MATCH(overlap, given, template) \
2094 ((overlap & ~JumpAbsolute) \
2095 && (((given) & (BaseIndex | JumpAbsolute)) \
2096 == ((overlap) & (BaseIndex | JumpAbsolute))))
2097
2098 /* If given types r0 and r1 are registers they must be of the same type
2099 unless the expected operand type register overlap is null.
2100 Note that Acc in a template matches every size of reg. */
2101 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
2102 (((g0) & Reg) == 0 || ((g1) & Reg) == 0 \
2103 || ((g0) & Reg) == ((g1) & Reg) \
2104 || ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
2105
2106 overlap0 = 0;
2107 overlap1 = 0;
2108 overlap2 = 0;
2109 found_reverse_match = 0;
2110 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
2111 ? No_bSuf
2112 : (i.suffix == WORD_MNEM_SUFFIX
2113 ? No_wSuf
2114 : (i.suffix == SHORT_MNEM_SUFFIX
2115 ? No_sSuf
2116 : (i.suffix == LONG_MNEM_SUFFIX
2117 ? No_lSuf
2118 : (i.suffix == QWORD_MNEM_SUFFIX
2119 ? No_qSuf
2120 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX
2121 ? No_xSuf : 0))))));
2122
2123 t = current_templates->start;
2124 if (i.suffix == QWORD_MNEM_SUFFIX
2125 && flag_code != CODE_64BIT
2126 && (intel_syntax
2127 ? !(t->opcode_modifier & IgnoreSize)
2128 && !intel_float_operand (t->name)
2129 : intel_float_operand (t->name) != 2)
2130 && (!(t->operand_types[0] & (RegMMX | RegXMM))
2131 || !(t->operand_types[t->operands > 1] & (RegMMX | RegXMM)))
2132 && (t->base_opcode != 0x0fc7
2133 || t->extension_opcode != 1 /* cmpxchg8b */))
2134 t = current_templates->end;
2135 for (; t < current_templates->end; t++)
2136 {
2137 /* Must have right number of operands. */
2138 if (i.operands != t->operands)
2139 continue;
2140
2141 /* Check the suffix, except for some instructions in intel mode. */
2142 if ((t->opcode_modifier & suffix_check)
2143 && !(intel_syntax
2144 && (t->opcode_modifier & IgnoreSize)))
2145 continue;
2146
2147 /* Do not verify operands when there are none. */
2148 else if (!t->operands)
2149 {
2150 if (t->cpu_flags & ~cpu_arch_flags)
2151 continue;
2152 /* We've found a match; break out of loop. */
2153 break;
2154 }
2155
2156 overlap0 = i.types[0] & t->operand_types[0];
2157 switch (t->operands)
2158 {
2159 case 1:
2160 if (!MATCH (overlap0, i.types[0], t->operand_types[0]))
2161 continue;
2162 break;
2163 case 2:
2164 case 3:
2165 overlap1 = i.types[1] & t->operand_types[1];
2166 if (!MATCH (overlap0, i.types[0], t->operand_types[0])
2167 || !MATCH (overlap1, i.types[1], t->operand_types[1])
2168 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2169 t->operand_types[0],
2170 overlap1, i.types[1],
2171 t->operand_types[1]))
2172 {
2173 /* Check if other direction is valid ... */
2174 if ((t->opcode_modifier & (D | FloatD)) == 0)
2175 continue;
2176
2177 /* Try reversing direction of operands. */
2178 overlap0 = i.types[0] & t->operand_types[1];
2179 overlap1 = i.types[1] & t->operand_types[0];
2180 if (!MATCH (overlap0, i.types[0], t->operand_types[1])
2181 || !MATCH (overlap1, i.types[1], t->operand_types[0])
2182 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2183 t->operand_types[1],
2184 overlap1, i.types[1],
2185 t->operand_types[0]))
2186 {
2187 /* Does not match either direction. */
2188 continue;
2189 }
2190 /* found_reverse_match holds which of D or FloatDR
2191 we've found. */
2192 found_reverse_match = t->opcode_modifier & (D | FloatDR);
2193 }
2194 /* Found a forward 2 operand match here. */
2195 else if (t->operands == 3)
2196 {
2197 /* Here we make use of the fact that there are no
2198 reverse match 3 operand instructions, and all 3
2199 operand instructions only need to be checked for
2200 register consistency between operands 2 and 3. */
2201 overlap2 = i.types[2] & t->operand_types[2];
2202 if (!MATCH (overlap2, i.types[2], t->operand_types[2])
2203 || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1],
2204 t->operand_types[1],
2205 overlap2, i.types[2],
2206 t->operand_types[2]))
2207
2208 continue;
2209 }
2210 /* Found either forward/reverse 2 or 3 operand match here:
2211 slip through to break. */
2212 }
2213 if (t->cpu_flags & ~cpu_arch_flags)
2214 {
2215 found_reverse_match = 0;
2216 continue;
2217 }
2218 /* We've found a match; break out of loop. */
2219 break;
2220 }
2221
2222 if (t == current_templates->end)
2223 {
2224 /* We found no match. */
2225 as_bad (_("suffix or operands invalid for `%s'"),
2226 current_templates->start->name);
2227 return 0;
2228 }
2229
2230 if (!quiet_warnings)
2231 {
2232 if (!intel_syntax
2233 && ((i.types[0] & JumpAbsolute)
2234 != (t->operand_types[0] & JumpAbsolute)))
2235 {
2236 as_warn (_("indirect %s without `*'"), t->name);
2237 }
2238
2239 if ((t->opcode_modifier & (IsPrefix | IgnoreSize))
2240 == (IsPrefix | IgnoreSize))
2241 {
2242 /* Warn them that a data or address size prefix doesn't
2243 affect assembly of the next line of code. */
2244 as_warn (_("stand-alone `%s' prefix"), t->name);
2245 }
2246 }
2247
2248 /* Copy the template we found. */
2249 i.tm = *t;
2250 if (found_reverse_match)
2251 {
2252 /* If we found a reverse match we must alter the opcode
2253 direction bit. found_reverse_match holds bits to change
2254 (different for int & float insns). */
2255
2256 i.tm.base_opcode ^= found_reverse_match;
2257
2258 i.tm.operand_types[0] = t->operand_types[1];
2259 i.tm.operand_types[1] = t->operand_types[0];
2260 }
2261
2262 return 1;
2263 }
2264
2265 static int
2266 check_string ()
2267 {
2268 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
2269 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
2270 {
2271 if (i.seg[0] != NULL && i.seg[0] != &es)
2272 {
2273 as_bad (_("`%s' operand %d must use `%%es' segment"),
2274 i.tm.name,
2275 mem_op + 1);
2276 return 0;
2277 }
2278 /* There's only ever one segment override allowed per instruction.
2279 This instruction possibly has a legal segment override on the
2280 second operand, so copy the segment to where non-string
2281 instructions store it, allowing common code. */
2282 i.seg[0] = i.seg[1];
2283 }
2284 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
2285 {
2286 if (i.seg[1] != NULL && i.seg[1] != &es)
2287 {
2288 as_bad (_("`%s' operand %d must use `%%es' segment"),
2289 i.tm.name,
2290 mem_op + 2);
2291 return 0;
2292 }
2293 }
2294 return 1;
2295 }
2296
2297 static int
2298 process_suffix (void)
2299 {
2300 /* If matched instruction specifies an explicit instruction mnemonic
2301 suffix, use it. */
2302 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
2303 {
2304 if (i.tm.opcode_modifier & Size16)
2305 i.suffix = WORD_MNEM_SUFFIX;
2306 else if (i.tm.opcode_modifier & Size64)
2307 i.suffix = QWORD_MNEM_SUFFIX;
2308 else
2309 i.suffix = LONG_MNEM_SUFFIX;
2310 }
2311 else if (i.reg_operands)
2312 {
2313 /* If there's no instruction mnemonic suffix we try to invent one
2314 based on register operands. */
2315 if (!i.suffix)
2316 {
2317 /* We take i.suffix from the last register operand specified,
2318 Destination register type is more significant than source
2319 register type. */
2320 int op;
2321
2322 for (op = i.operands; --op >= 0;)
2323 if ((i.types[op] & Reg)
2324 && !(i.tm.operand_types[op] & InOutPortReg))
2325 {
2326 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
2327 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
2328 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
2329 LONG_MNEM_SUFFIX);
2330 break;
2331 }
2332 }
2333 else if (i.suffix == BYTE_MNEM_SUFFIX)
2334 {
2335 if (!check_byte_reg ())
2336 return 0;
2337 }
2338 else if (i.suffix == LONG_MNEM_SUFFIX)
2339 {
2340 if (!check_long_reg ())
2341 return 0;
2342 }
2343 else if (i.suffix == QWORD_MNEM_SUFFIX)
2344 {
2345 if (!check_qword_reg ())
2346 return 0;
2347 }
2348 else if (i.suffix == WORD_MNEM_SUFFIX)
2349 {
2350 if (!check_word_reg ())
2351 return 0;
2352 }
2353 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2354 /* Do nothing if the instruction is going to ignore the prefix. */
2355 ;
2356 else
2357 abort ();
2358 }
2359 else if ((i.tm.opcode_modifier & DefaultSize)
2360 && !i.suffix
2361 /* exclude fldenv/frstor/fsave/fstenv */
2362 && (i.tm.opcode_modifier & No_sSuf))
2363 {
2364 i.suffix = stackop_size;
2365 }
2366 else if (intel_syntax
2367 && !i.suffix
2368 && ((i.tm.operand_types[0] & JumpAbsolute)
2369 || (i.tm.opcode_modifier & (JumpByte|JumpInterSegment))
2370 || (i.tm.base_opcode == 0x0f01 /* [ls][gi]dt */
2371 && i.tm.extension_opcode <= 3)))
2372 {
2373 switch (flag_code)
2374 {
2375 case CODE_64BIT:
2376 if (!(i.tm.opcode_modifier & No_qSuf))
2377 {
2378 i.suffix = QWORD_MNEM_SUFFIX;
2379 break;
2380 }
2381 case CODE_32BIT:
2382 if (!(i.tm.opcode_modifier & No_lSuf))
2383 i.suffix = LONG_MNEM_SUFFIX;
2384 break;
2385 case CODE_16BIT:
2386 if (!(i.tm.opcode_modifier & No_wSuf))
2387 i.suffix = WORD_MNEM_SUFFIX;
2388 break;
2389 }
2390 }
2391
2392 if (!i.suffix)
2393 {
2394 if (!intel_syntax)
2395 {
2396 if (i.tm.opcode_modifier & W)
2397 {
2398 as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
2399 return 0;
2400 }
2401 }
2402 else
2403 {
2404 unsigned int suffixes = ~i.tm.opcode_modifier
2405 & (No_bSuf
2406 | No_wSuf
2407 | No_lSuf
2408 | No_sSuf
2409 | No_xSuf
2410 | No_qSuf);
2411
2412 if ((i.tm.opcode_modifier & W)
2413 || ((suffixes & (suffixes - 1))
2414 && !(i.tm.opcode_modifier & (DefaultSize | IgnoreSize))))
2415 {
2416 as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
2417 return 0;
2418 }
2419 }
2420 }
2421
2422 /* Change the opcode based on the operand size given by i.suffix;
2423 We don't need to change things for byte insns. */
2424
2425 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2426 {
2427 /* It's not a byte, select word/dword operation. */
2428 if (i.tm.opcode_modifier & W)
2429 {
2430 if (i.tm.opcode_modifier & ShortForm)
2431 i.tm.base_opcode |= 8;
2432 else
2433 i.tm.base_opcode |= 1;
2434 }
2435
2436 /* Now select between word & dword operations via the operand
2437 size prefix, except for instructions that will ignore this
2438 prefix anyway. */
2439 if (i.suffix != QWORD_MNEM_SUFFIX
2440 && i.suffix != LONG_DOUBLE_MNEM_SUFFIX
2441 && !(i.tm.opcode_modifier & (IgnoreSize | FloatMF))
2442 && ((i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2443 || (flag_code == CODE_64BIT
2444 && (i.tm.opcode_modifier & JumpByte))))
2445 {
2446 unsigned int prefix = DATA_PREFIX_OPCODE;
2447
2448 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2449 prefix = ADDR_PREFIX_OPCODE;
2450
2451 if (!add_prefix (prefix))
2452 return 0;
2453 }
2454
2455 /* Set mode64 for an operand. */
2456 if (i.suffix == QWORD_MNEM_SUFFIX
2457 && flag_code == CODE_64BIT
2458 && (i.tm.opcode_modifier & NoRex64) == 0)
2459 i.rex |= REX_MODE64;
2460
2461 /* Size floating point instruction. */
2462 if (i.suffix == LONG_MNEM_SUFFIX)
2463 if (i.tm.opcode_modifier & FloatMF)
2464 i.tm.base_opcode ^= 4;
2465 }
2466
2467 return 1;
2468 }
2469
2470 static int
2471 check_byte_reg (void)
2472 {
2473 int op;
2474
2475 for (op = i.operands; --op >= 0;)
2476 {
2477 /* If this is an eight bit register, it's OK. If it's the 16 or
2478 32 bit version of an eight bit register, we will just use the
2479 low portion, and that's OK too. */
2480 if (i.types[op] & Reg8)
2481 continue;
2482
2483 /* movzx and movsx should not generate this warning. */
2484 if (intel_syntax
2485 && (i.tm.base_opcode == 0xfb7
2486 || i.tm.base_opcode == 0xfb6
2487 || i.tm.base_opcode == 0x63
2488 || i.tm.base_opcode == 0xfbe
2489 || i.tm.base_opcode == 0xfbf))
2490 continue;
2491
2492 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4)
2493 {
2494 /* Prohibit these changes in the 64bit mode, since the
2495 lowering is more complicated. */
2496 if (flag_code == CODE_64BIT
2497 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2498 {
2499 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2500 i.op[op].regs->reg_name,
2501 i.suffix);
2502 return 0;
2503 }
2504 #if REGISTER_WARNINGS
2505 if (!quiet_warnings
2506 && (i.tm.operand_types[op] & InOutPortReg) == 0)
2507 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2508 (i.op[op].regs + (i.types[op] & Reg16
2509 ? REGNAM_AL - REGNAM_AX
2510 : REGNAM_AL - REGNAM_EAX))->reg_name,
2511 i.op[op].regs->reg_name,
2512 i.suffix);
2513 #endif
2514 continue;
2515 }
2516 /* Any other register is bad. */
2517 if (i.types[op] & (Reg | RegMMX | RegXMM
2518 | SReg2 | SReg3
2519 | Control | Debug | Test
2520 | FloatReg | FloatAcc))
2521 {
2522 as_bad (_("`%%%s' not allowed with `%s%c'"),
2523 i.op[op].regs->reg_name,
2524 i.tm.name,
2525 i.suffix);
2526 return 0;
2527 }
2528 }
2529 return 1;
2530 }
2531
2532 static int
2533 check_long_reg ()
2534 {
2535 int op;
2536
2537 for (op = i.operands; --op >= 0;)
2538 /* Reject eight bit registers, except where the template requires
2539 them. (eg. movzb) */
2540 if ((i.types[op] & Reg8) != 0
2541 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2542 {
2543 as_bad (_("`%%%s' not allowed with `%s%c'"),
2544 i.op[op].regs->reg_name,
2545 i.tm.name,
2546 i.suffix);
2547 return 0;
2548 }
2549 /* Warn if the e prefix on a general reg is missing. */
2550 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2551 && (i.types[op] & Reg16) != 0
2552 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2553 {
2554 /* Prohibit these changes in the 64bit mode, since the
2555 lowering is more complicated. */
2556 if (flag_code == CODE_64BIT)
2557 {
2558 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2559 i.op[op].regs->reg_name,
2560 i.suffix);
2561 return 0;
2562 }
2563 #if REGISTER_WARNINGS
2564 else
2565 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2566 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
2567 i.op[op].regs->reg_name,
2568 i.suffix);
2569 #endif
2570 }
2571 /* Warn if the r prefix on a general reg is missing. */
2572 else if ((i.types[op] & Reg64) != 0
2573 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2574 {
2575 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2576 i.op[op].regs->reg_name,
2577 i.suffix);
2578 return 0;
2579 }
2580 return 1;
2581 }
2582
2583 static int
2584 check_qword_reg ()
2585 {
2586 int op;
2587
2588 for (op = i.operands; --op >= 0; )
2589 /* Reject eight bit registers, except where the template requires
2590 them. (eg. movzb) */
2591 if ((i.types[op] & Reg8) != 0
2592 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2593 {
2594 as_bad (_("`%%%s' not allowed with `%s%c'"),
2595 i.op[op].regs->reg_name,
2596 i.tm.name,
2597 i.suffix);
2598 return 0;
2599 }
2600 /* Warn if the e prefix on a general reg is missing. */
2601 else if (((i.types[op] & Reg16) != 0
2602 || (i.types[op] & Reg32) != 0)
2603 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
2604 {
2605 /* Prohibit these changes in the 64bit mode, since the
2606 lowering is more complicated. */
2607 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2608 i.op[op].regs->reg_name,
2609 i.suffix);
2610 return 0;
2611 }
2612 return 1;
2613 }
2614
2615 static int
2616 check_word_reg ()
2617 {
2618 int op;
2619 for (op = i.operands; --op >= 0;)
2620 /* Reject eight bit registers, except where the template requires
2621 them. (eg. movzb) */
2622 if ((i.types[op] & Reg8) != 0
2623 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
2624 {
2625 as_bad (_("`%%%s' not allowed with `%s%c'"),
2626 i.op[op].regs->reg_name,
2627 i.tm.name,
2628 i.suffix);
2629 return 0;
2630 }
2631 /* Warn if the e prefix on a general reg is present. */
2632 else if ((!quiet_warnings || flag_code == CODE_64BIT)
2633 && (i.types[op] & Reg32) != 0
2634 && (i.tm.operand_types[op] & (Reg16 | Acc)) != 0)
2635 {
2636 /* Prohibit these changes in the 64bit mode, since the
2637 lowering is more complicated. */
2638 if (flag_code == CODE_64BIT)
2639 {
2640 as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
2641 i.op[op].regs->reg_name,
2642 i.suffix);
2643 return 0;
2644 }
2645 else
2646 #if REGISTER_WARNINGS
2647 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
2648 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
2649 i.op[op].regs->reg_name,
2650 i.suffix);
2651 #endif
2652 }
2653 return 1;
2654 }
2655
2656 static int
2657 finalize_imm ()
2658 {
2659 unsigned int overlap0, overlap1, overlap2;
2660
2661 overlap0 = i.types[0] & i.tm.operand_types[0];
2662 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64))
2663 && overlap0 != Imm8 && overlap0 != Imm8S
2664 && overlap0 != Imm16 && overlap0 != Imm32S
2665 && overlap0 != Imm32 && overlap0 != Imm64)
2666 {
2667 if (i.suffix)
2668 {
2669 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX
2670 ? Imm8 | Imm8S
2671 : (i.suffix == WORD_MNEM_SUFFIX
2672 ? Imm16
2673 : (i.suffix == QWORD_MNEM_SUFFIX
2674 ? Imm64 | Imm32S
2675 : Imm32)));
2676 }
2677 else if (overlap0 == (Imm16 | Imm32S | Imm32)
2678 || overlap0 == (Imm16 | Imm32)
2679 || overlap0 == (Imm16 | Imm32S))
2680 {
2681 overlap0 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
2682 ? Imm16 : Imm32S);
2683 }
2684 if (overlap0 != Imm8 && overlap0 != Imm8S
2685 && overlap0 != Imm16 && overlap0 != Imm32S
2686 && overlap0 != Imm32 && overlap0 != Imm64)
2687 {
2688 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
2689 return 0;
2690 }
2691 }
2692 i.types[0] = overlap0;
2693
2694 overlap1 = i.types[1] & i.tm.operand_types[1];
2695 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32 | Imm64))
2696 && overlap1 != Imm8 && overlap1 != Imm8S
2697 && overlap1 != Imm16 && overlap1 != Imm32S
2698 && overlap1 != Imm32 && overlap1 != Imm64)
2699 {
2700 if (i.suffix)
2701 {
2702 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX
2703 ? Imm8 | Imm8S
2704 : (i.suffix == WORD_MNEM_SUFFIX
2705 ? Imm16
2706 : (i.suffix == QWORD_MNEM_SUFFIX
2707 ? Imm64 | Imm32S
2708 : Imm32)));
2709 }
2710 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
2711 || overlap1 == (Imm16 | Imm32)
2712 || overlap1 == (Imm16 | Imm32S))
2713 {
2714 overlap1 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
2715 ? Imm16 : Imm32S);
2716 }
2717 if (overlap1 != Imm8 && overlap1 != Imm8S
2718 && overlap1 != Imm16 && overlap1 != Imm32S
2719 && overlap1 != Imm32 && overlap1 != Imm64)
2720 {
2721 as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
2722 return 0;
2723 }
2724 }
2725 i.types[1] = overlap1;
2726
2727 overlap2 = i.types[2] & i.tm.operand_types[2];
2728 assert ((overlap2 & Imm) == 0);
2729 i.types[2] = overlap2;
2730
2731 return 1;
2732 }
2733
2734 static int
2735 process_operands ()
2736 {
2737 /* Default segment register this instruction will use for memory
2738 accesses. 0 means unknown. This is only for optimizing out
2739 unnecessary segment overrides. */
2740 const seg_entry *default_seg = 0;
2741
2742 /* The imul $imm, %reg instruction is converted into
2743 imul $imm, %reg, %reg, and the clr %reg instruction
2744 is converted into xor %reg, %reg. */
2745 if (i.tm.opcode_modifier & regKludge)
2746 {
2747 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
2748 /* Pretend we saw the extra register operand. */
2749 assert (i.op[first_reg_op + 1].regs == 0);
2750 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
2751 i.types[first_reg_op + 1] = i.types[first_reg_op];
2752 i.reg_operands = 2;
2753 }
2754
2755 if (i.tm.opcode_modifier & ShortForm)
2756 {
2757 /* The register or float register operand is in operand 0 or 1. */
2758 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
2759 /* Register goes in low 3 bits of opcode. */
2760 i.tm.base_opcode |= i.op[op].regs->reg_num;
2761 if ((i.op[op].regs->reg_flags & RegRex) != 0)
2762 i.rex |= REX_EXTZ;
2763 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
2764 {
2765 /* Warn about some common errors, but press on regardless.
2766 The first case can be generated by gcc (<= 2.8.1). */
2767 if (i.operands == 2)
2768 {
2769 /* Reversed arguments on faddp, fsubp, etc. */
2770 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
2771 i.op[1].regs->reg_name,
2772 i.op[0].regs->reg_name);
2773 }
2774 else
2775 {
2776 /* Extraneous `l' suffix on fp insn. */
2777 as_warn (_("translating to `%s %%%s'"), i.tm.name,
2778 i.op[0].regs->reg_name);
2779 }
2780 }
2781 }
2782 else if (i.tm.opcode_modifier & Modrm)
2783 {
2784 /* The opcode is completed (modulo i.tm.extension_opcode which
2785 must be put into the modrm byte). Now, we make the modrm and
2786 index base bytes based on all the info we've collected. */
2787
2788 default_seg = build_modrm_byte ();
2789 }
2790 else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
2791 {
2792 if (i.tm.base_opcode == POP_SEG_SHORT
2793 && i.op[0].regs->reg_num == 1)
2794 {
2795 as_bad (_("you can't `pop %%cs'"));
2796 return 0;
2797 }
2798 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
2799 if ((i.op[0].regs->reg_flags & RegRex) != 0)
2800 i.rex |= REX_EXTZ;
2801 }
2802 else if ((i.tm.base_opcode & ~(D | W)) == MOV_AX_DISP32)
2803 {
2804 default_seg = &ds;
2805 }
2806 else if ((i.tm.opcode_modifier & IsString) != 0)
2807 {
2808 /* For the string instructions that allow a segment override
2809 on one of their operands, the default segment is ds. */
2810 default_seg = &ds;
2811 }
2812
2813 if (i.tm.base_opcode == 0x8d /* lea */ && i.seg[0] && !quiet_warnings)
2814 as_warn (_("segment override on `lea' is ineffectual"));
2815
2816 /* If a segment was explicitly specified, and the specified segment
2817 is not the default, use an opcode prefix to select it. If we
2818 never figured out what the default segment is, then default_seg
2819 will be zero at this point, and the specified segment prefix will
2820 always be used. */
2821 if ((i.seg[0]) && (i.seg[0] != default_seg))
2822 {
2823 if (!add_prefix (i.seg[0]->seg_prefix))
2824 return 0;
2825 }
2826 return 1;
2827 }
2828
2829 static const seg_entry *
2830 build_modrm_byte ()
2831 {
2832 const seg_entry *default_seg = 0;
2833
2834 /* i.reg_operands MUST be the number of real register operands;
2835 implicit registers do not count. */
2836 if (i.reg_operands == 2)
2837 {
2838 unsigned int source, dest;
2839 source = ((i.types[0]
2840 & (Reg | RegMMX | RegXMM
2841 | SReg2 | SReg3
2842 | Control | Debug | Test))
2843 ? 0 : 1);
2844 dest = source + 1;
2845
2846 i.rm.mode = 3;
2847 /* One of the register operands will be encoded in the i.tm.reg
2848 field, the other in the combined i.tm.mode and i.tm.regmem
2849 fields. If no form of this instruction supports a memory
2850 destination operand, then we assume the source operand may
2851 sometimes be a memory operand and so we need to store the
2852 destination in the i.rm.reg field. */
2853 if ((i.tm.operand_types[dest] & AnyMem) == 0)
2854 {
2855 i.rm.reg = i.op[dest].regs->reg_num;
2856 i.rm.regmem = i.op[source].regs->reg_num;
2857 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
2858 i.rex |= REX_EXTX;
2859 if ((i.op[source].regs->reg_flags & RegRex) != 0)
2860 i.rex |= REX_EXTZ;
2861 }
2862 else
2863 {
2864 i.rm.reg = i.op[source].regs->reg_num;
2865 i.rm.regmem = i.op[dest].regs->reg_num;
2866 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
2867 i.rex |= REX_EXTZ;
2868 if ((i.op[source].regs->reg_flags & RegRex) != 0)
2869 i.rex |= REX_EXTX;
2870 }
2871 if (flag_code != CODE_64BIT && (i.rex & (REX_EXTX | REX_EXTZ)))
2872 {
2873 if (!((i.types[0] | i.types[1]) & Control))
2874 abort ();
2875 i.rex &= ~(REX_EXTX | REX_EXTZ);
2876 add_prefix (LOCK_PREFIX_OPCODE);
2877 }
2878 }
2879 else
2880 { /* If it's not 2 reg operands... */
2881 if (i.mem_operands)
2882 {
2883 unsigned int fake_zero_displacement = 0;
2884 unsigned int op = ((i.types[0] & AnyMem)
2885 ? 0
2886 : (i.types[1] & AnyMem) ? 1 : 2);
2887
2888 default_seg = &ds;
2889
2890 if (i.base_reg == 0)
2891 {
2892 i.rm.mode = 0;
2893 if (!i.disp_operands)
2894 fake_zero_displacement = 1;
2895 if (i.index_reg == 0)
2896 {
2897 /* Operand is just <disp> */
2898 if (flag_code == CODE_64BIT)
2899 {
2900 /* 64bit mode overwrites the 32bit absolute
2901 addressing by RIP relative addressing and
2902 absolute addressing is encoded by one of the
2903 redundant SIB forms. */
2904 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2905 i.sib.base = NO_BASE_REGISTER;
2906 i.sib.index = NO_INDEX_REGISTER;
2907 i.types[op] = ((i.prefix[ADDR_PREFIX] == 0) ? Disp32S : Disp32);
2908 }
2909 else if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
2910 {
2911 i.rm.regmem = NO_BASE_REGISTER_16;
2912 i.types[op] = Disp16;
2913 }
2914 else
2915 {
2916 i.rm.regmem = NO_BASE_REGISTER;
2917 i.types[op] = Disp32;
2918 }
2919 }
2920 else /* !i.base_reg && i.index_reg */
2921 {
2922 i.sib.index = i.index_reg->reg_num;
2923 i.sib.base = NO_BASE_REGISTER;
2924 i.sib.scale = i.log2_scale_factor;
2925 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
2926 i.types[op] &= ~Disp;
2927 if (flag_code != CODE_64BIT)
2928 i.types[op] |= Disp32; /* Must be 32 bit */
2929 else
2930 i.types[op] |= Disp32S;
2931 if ((i.index_reg->reg_flags & RegRex) != 0)
2932 i.rex |= REX_EXTY;
2933 }
2934 }
2935 /* RIP addressing for 64bit mode. */
2936 else if (i.base_reg->reg_type == BaseIndex)
2937 {
2938 i.rm.regmem = NO_BASE_REGISTER;
2939 i.types[op] &= ~ Disp;
2940 i.types[op] |= Disp32S;
2941 i.flags[op] = Operand_PCrel;
2942 if (! i.disp_operands)
2943 fake_zero_displacement = 1;
2944 }
2945 else if (i.base_reg->reg_type & Reg16)
2946 {
2947 switch (i.base_reg->reg_num)
2948 {
2949 case 3: /* (%bx) */
2950 if (i.index_reg == 0)
2951 i.rm.regmem = 7;
2952 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
2953 i.rm.regmem = i.index_reg->reg_num - 6;
2954 break;
2955 case 5: /* (%bp) */
2956 default_seg = &ss;
2957 if (i.index_reg == 0)
2958 {
2959 i.rm.regmem = 6;
2960 if ((i.types[op] & Disp) == 0)
2961 {
2962 /* fake (%bp) into 0(%bp) */
2963 i.types[op] |= Disp8;
2964 fake_zero_displacement = 1;
2965 }
2966 }
2967 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
2968 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
2969 break;
2970 default: /* (%si) -> 4 or (%di) -> 5 */
2971 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
2972 }
2973 i.rm.mode = mode_from_disp_size (i.types[op]);
2974 }
2975 else /* i.base_reg and 32/64 bit mode */
2976 {
2977 if (flag_code == CODE_64BIT
2978 && (i.types[op] & Disp))
2979 i.types[op] = (i.types[op] & Disp8) | (i.prefix[ADDR_PREFIX] == 0 ? Disp32S : Disp32);
2980
2981 i.rm.regmem = i.base_reg->reg_num;
2982 if ((i.base_reg->reg_flags & RegRex) != 0)
2983 i.rex |= REX_EXTZ;
2984 i.sib.base = i.base_reg->reg_num;
2985 /* x86-64 ignores REX prefix bit here to avoid decoder
2986 complications. */
2987 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
2988 {
2989 default_seg = &ss;
2990 if (i.disp_operands == 0)
2991 {
2992 fake_zero_displacement = 1;
2993 i.types[op] |= Disp8;
2994 }
2995 }
2996 else if (i.base_reg->reg_num == ESP_REG_NUM)
2997 {
2998 default_seg = &ss;
2999 }
3000 i.sib.scale = i.log2_scale_factor;
3001 if (i.index_reg == 0)
3002 {
3003 /* <disp>(%esp) becomes two byte modrm with no index
3004 register. We've already stored the code for esp
3005 in i.rm.regmem ie. ESCAPE_TO_TWO_BYTE_ADDRESSING.
3006 Any base register besides %esp will not use the
3007 extra modrm byte. */
3008 i.sib.index = NO_INDEX_REGISTER;
3009 #if !SCALE1_WHEN_NO_INDEX
3010 /* Another case where we force the second modrm byte. */
3011 if (i.log2_scale_factor)
3012 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3013 #endif
3014 }
3015 else
3016 {
3017 i.sib.index = i.index_reg->reg_num;
3018 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3019 if ((i.index_reg->reg_flags & RegRex) != 0)
3020 i.rex |= REX_EXTY;
3021 }
3022 i.rm.mode = mode_from_disp_size (i.types[op]);
3023 }
3024
3025 if (fake_zero_displacement)
3026 {
3027 /* Fakes a zero displacement assuming that i.types[op]
3028 holds the correct displacement size. */
3029 expressionS *exp;
3030
3031 assert (i.op[op].disps == 0);
3032 exp = &disp_expressions[i.disp_operands++];
3033 i.op[op].disps = exp;
3034 exp->X_op = O_constant;
3035 exp->X_add_number = 0;
3036 exp->X_add_symbol = (symbolS *) 0;
3037 exp->X_op_symbol = (symbolS *) 0;
3038 }
3039 }
3040
3041 /* Fill in i.rm.reg or i.rm.regmem field with register operand
3042 (if any) based on i.tm.extension_opcode. Again, we must be
3043 careful to make sure that segment/control/debug/test/MMX
3044 registers are coded into the i.rm.reg field. */
3045 if (i.reg_operands)
3046 {
3047 unsigned int op =
3048 ((i.types[0]
3049 & (Reg | RegMMX | RegXMM
3050 | SReg2 | SReg3
3051 | Control | Debug | Test))
3052 ? 0
3053 : ((i.types[1]
3054 & (Reg | RegMMX | RegXMM
3055 | SReg2 | SReg3
3056 | Control | Debug | Test))
3057 ? 1
3058 : 2));
3059 /* If there is an extension opcode to put here, the register
3060 number must be put into the regmem field. */
3061 if (i.tm.extension_opcode != None)
3062 {
3063 i.rm.regmem = i.op[op].regs->reg_num;
3064 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3065 i.rex |= REX_EXTZ;
3066 }
3067 else
3068 {
3069 i.rm.reg = i.op[op].regs->reg_num;
3070 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3071 i.rex |= REX_EXTX;
3072 }
3073
3074 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2 we
3075 must set it to 3 to indicate this is a register operand
3076 in the regmem field. */
3077 if (!i.mem_operands)
3078 i.rm.mode = 3;
3079 }
3080
3081 /* Fill in i.rm.reg field with extension opcode (if any). */
3082 if (i.tm.extension_opcode != None)
3083 i.rm.reg = i.tm.extension_opcode;
3084 }
3085 return default_seg;
3086 }
3087
3088 static void
3089 output_branch ()
3090 {
3091 char *p;
3092 int code16;
3093 int prefix;
3094 relax_substateT subtype;
3095 symbolS *sym;
3096 offsetT off;
3097
3098 code16 = 0;
3099 if (flag_code == CODE_16BIT)
3100 code16 = CODE16;
3101
3102 prefix = 0;
3103 if (i.prefix[DATA_PREFIX] != 0)
3104 {
3105 prefix = 1;
3106 i.prefixes -= 1;
3107 code16 ^= CODE16;
3108 }
3109 /* Pentium4 branch hints. */
3110 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3111 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3112 {
3113 prefix++;
3114 i.prefixes--;
3115 }
3116 if (i.prefix[REX_PREFIX] != 0)
3117 {
3118 prefix++;
3119 i.prefixes--;
3120 }
3121
3122 if (i.prefixes != 0 && !intel_syntax)
3123 as_warn (_("skipping prefixes on this instruction"));
3124
3125 /* It's always a symbol; End frag & setup for relax.
3126 Make sure there is enough room in this frag for the largest
3127 instruction we may generate in md_convert_frag. This is 2
3128 bytes for the opcode and room for the prefix and largest
3129 displacement. */
3130 frag_grow (prefix + 2 + 4);
3131 /* Prefix and 1 opcode byte go in fr_fix. */
3132 p = frag_more (prefix + 1);
3133 if (i.prefix[DATA_PREFIX] != 0)
3134 *p++ = DATA_PREFIX_OPCODE;
3135 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
3136 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
3137 *p++ = i.prefix[SEG_PREFIX];
3138 if (i.prefix[REX_PREFIX] != 0)
3139 *p++ = i.prefix[REX_PREFIX];
3140 *p = i.tm.base_opcode;
3141
3142 if ((unsigned char) *p == JUMP_PC_RELATIVE)
3143 subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
3144 else if ((cpu_arch_flags & Cpu386) != 0)
3145 subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
3146 else
3147 subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
3148 subtype |= code16;
3149
3150 sym = i.op[0].disps->X_add_symbol;
3151 off = i.op[0].disps->X_add_number;
3152
3153 if (i.op[0].disps->X_op != O_constant
3154 && i.op[0].disps->X_op != O_symbol)
3155 {
3156 /* Handle complex expressions. */
3157 sym = make_expr_symbol (i.op[0].disps);
3158 off = 0;
3159 }
3160
3161 /* 1 possible extra opcode + 4 byte displacement go in var part.
3162 Pass reloc in fr_var. */
3163 frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
3164 }
3165
3166 static void
3167 output_jump ()
3168 {
3169 char *p;
3170 int size;
3171 fixS *fixP;
3172
3173 if (i.tm.opcode_modifier & JumpByte)
3174 {
3175 /* This is a loop or jecxz type instruction. */
3176 size = 1;
3177 if (i.prefix[ADDR_PREFIX] != 0)
3178 {
3179 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
3180 i.prefixes -= 1;
3181 }
3182 /* Pentium4 branch hints. */
3183 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3184 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3185 {
3186 FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
3187 i.prefixes--;
3188 }
3189 }
3190 else
3191 {
3192 int code16;
3193
3194 code16 = 0;
3195 if (flag_code == CODE_16BIT)
3196 code16 = CODE16;
3197
3198 if (i.prefix[DATA_PREFIX] != 0)
3199 {
3200 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
3201 i.prefixes -= 1;
3202 code16 ^= CODE16;
3203 }
3204
3205 size = 4;
3206 if (code16)
3207 size = 2;
3208 }
3209
3210 if (i.prefix[REX_PREFIX] != 0)
3211 {
3212 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
3213 i.prefixes -= 1;
3214 }
3215
3216 if (i.prefixes != 0 && !intel_syntax)
3217 as_warn (_("skipping prefixes on this instruction"));
3218
3219 p = frag_more (1 + size);
3220 *p++ = i.tm.base_opcode;
3221
3222 fixP = fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3223 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
3224
3225 /* All jumps handled here are signed, but don't use a signed limit
3226 check for 32 and 16 bit jumps as we want to allow wrap around at
3227 4G and 64k respectively. */
3228 if (size == 1)
3229 fixP->fx_signed = 1;
3230 }
3231
3232 static void
3233 output_interseg_jump ()
3234 {
3235 char *p;
3236 int size;
3237 int prefix;
3238 int code16;
3239
3240 code16 = 0;
3241 if (flag_code == CODE_16BIT)
3242 code16 = CODE16;
3243
3244 prefix = 0;
3245 if (i.prefix[DATA_PREFIX] != 0)
3246 {
3247 prefix = 1;
3248 i.prefixes -= 1;
3249 code16 ^= CODE16;
3250 }
3251 if (i.prefix[REX_PREFIX] != 0)
3252 {
3253 prefix++;
3254 i.prefixes -= 1;
3255 }
3256
3257 size = 4;
3258 if (code16)
3259 size = 2;
3260
3261 if (i.prefixes != 0 && !intel_syntax)
3262 as_warn (_("skipping prefixes on this instruction"));
3263
3264 /* 1 opcode; 2 segment; offset */
3265 p = frag_more (prefix + 1 + 2 + size);
3266
3267 if (i.prefix[DATA_PREFIX] != 0)
3268 *p++ = DATA_PREFIX_OPCODE;
3269
3270 if (i.prefix[REX_PREFIX] != 0)
3271 *p++ = i.prefix[REX_PREFIX];
3272
3273 *p++ = i.tm.base_opcode;
3274 if (i.op[1].imms->X_op == O_constant)
3275 {
3276 offsetT n = i.op[1].imms->X_add_number;
3277
3278 if (size == 2
3279 && !fits_in_unsigned_word (n)
3280 && !fits_in_signed_word (n))
3281 {
3282 as_bad (_("16-bit jump out of range"));
3283 return;
3284 }
3285 md_number_to_chars (p, n, size);
3286 }
3287 else
3288 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3289 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
3290 if (i.op[0].imms->X_op != O_constant)
3291 as_bad (_("can't handle non absolute segment in `%s'"),
3292 i.tm.name);
3293 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
3294 }
3295
3296 static void
3297 output_insn ()
3298 {
3299 fragS *insn_start_frag;
3300 offsetT insn_start_off;
3301
3302 /* Tie dwarf2 debug info to the address at the start of the insn.
3303 We can't do this after the insn has been output as the current
3304 frag may have been closed off. eg. by frag_var. */
3305 dwarf2_emit_insn (0);
3306
3307 insn_start_frag = frag_now;
3308 insn_start_off = frag_now_fix ();
3309
3310 /* Output jumps. */
3311 if (i.tm.opcode_modifier & Jump)
3312 output_branch ();
3313 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
3314 output_jump ();
3315 else if (i.tm.opcode_modifier & JumpInterSegment)
3316 output_interseg_jump ();
3317 else
3318 {
3319 /* Output normal instructions here. */
3320 char *p;
3321 unsigned char *q;
3322
3323 /* All opcodes on i386 have either 1 or 2 bytes, PadLock instructions
3324 have 3 bytes. We may use one more higher byte to specify a prefix
3325 the instruction requires. */
3326 if ((i.tm.cpu_flags & CpuPadLock) != 0
3327 && (i.tm.base_opcode & 0xff000000) != 0)
3328 {
3329 unsigned int prefix;
3330 prefix = (i.tm.base_opcode >> 24) & 0xff;
3331
3332 if (prefix != REPE_PREFIX_OPCODE
3333 || i.prefix[LOCKREP_PREFIX] != REPE_PREFIX_OPCODE)
3334 add_prefix (prefix);
3335 }
3336 else
3337 if ((i.tm.cpu_flags & CpuPadLock) == 0
3338 && (i.tm.base_opcode & 0xff0000) != 0)
3339 add_prefix ((i.tm.base_opcode >> 16) & 0xff);
3340
3341 /* The prefix bytes. */
3342 for (q = i.prefix;
3343 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
3344 q++)
3345 {
3346 if (*q)
3347 {
3348 p = frag_more (1);
3349 md_number_to_chars (p, (valueT) *q, 1);
3350 }
3351 }
3352
3353 /* Now the opcode; be careful about word order here! */
3354 if (fits_in_unsigned_byte (i.tm.base_opcode))
3355 {
3356 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
3357 }
3358 else
3359 {
3360 if ((i.tm.cpu_flags & CpuPadLock) != 0)
3361 {
3362 p = frag_more (3);
3363 *p++ = (i.tm.base_opcode >> 16) & 0xff;
3364 }
3365 else
3366 p = frag_more (2);
3367
3368 /* Put out high byte first: can't use md_number_to_chars! */
3369 *p++ = (i.tm.base_opcode >> 8) & 0xff;
3370 *p = i.tm.base_opcode & 0xff;
3371 }
3372
3373 /* Now the modrm byte and sib byte (if present). */
3374 if (i.tm.opcode_modifier & Modrm)
3375 {
3376 p = frag_more (1);
3377 md_number_to_chars (p,
3378 (valueT) (i.rm.regmem << 0
3379 | i.rm.reg << 3
3380 | i.rm.mode << 6),
3381 1);
3382 /* If i.rm.regmem == ESP (4)
3383 && i.rm.mode != (Register mode)
3384 && not 16 bit
3385 ==> need second modrm byte. */
3386 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
3387 && i.rm.mode != 3
3388 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
3389 {
3390 p = frag_more (1);
3391 md_number_to_chars (p,
3392 (valueT) (i.sib.base << 0
3393 | i.sib.index << 3
3394 | i.sib.scale << 6),
3395 1);
3396 }
3397 }
3398
3399 if (i.disp_operands)
3400 output_disp (insn_start_frag, insn_start_off);
3401
3402 if (i.imm_operands)
3403 output_imm (insn_start_frag, insn_start_off);
3404 }
3405
3406 #ifdef DEBUG386
3407 if (flag_debug)
3408 {
3409 pi (line, &i);
3410 }
3411 #endif /* DEBUG386 */
3412 }
3413
3414 static void
3415 output_disp (insn_start_frag, insn_start_off)
3416 fragS *insn_start_frag;
3417 offsetT insn_start_off;
3418 {
3419 char *p;
3420 unsigned int n;
3421
3422 for (n = 0; n < i.operands; n++)
3423 {
3424 if (i.types[n] & Disp)
3425 {
3426 if (i.op[n].disps->X_op == O_constant)
3427 {
3428 int size;
3429 offsetT val;
3430
3431 size = 4;
3432 if (i.types[n] & (Disp8 | Disp16 | Disp64))
3433 {
3434 size = 2;
3435 if (i.types[n] & Disp8)
3436 size = 1;
3437 if (i.types[n] & Disp64)
3438 size = 8;
3439 }
3440 val = offset_in_range (i.op[n].disps->X_add_number,
3441 size);
3442 p = frag_more (size);
3443 md_number_to_chars (p, val, size);
3444 }
3445 else
3446 {
3447 enum bfd_reloc_code_real reloc_type;
3448 int size = 4;
3449 int sign = 0;
3450 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
3451
3452 /* The PC relative address is computed relative
3453 to the instruction boundary, so in case immediate
3454 fields follows, we need to adjust the value. */
3455 if (pcrel && i.imm_operands)
3456 {
3457 int imm_size = 4;
3458 unsigned int n1;
3459
3460 for (n1 = 0; n1 < i.operands; n1++)
3461 if (i.types[n1] & Imm)
3462 {
3463 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
3464 {
3465 imm_size = 2;
3466 if (i.types[n1] & (Imm8 | Imm8S))
3467 imm_size = 1;
3468 if (i.types[n1] & Imm64)
3469 imm_size = 8;
3470 }
3471 break;
3472 }
3473 /* We should find the immediate. */
3474 if (n1 == i.operands)
3475 abort ();
3476 i.op[n].disps->X_add_number -= imm_size;
3477 }
3478
3479 if (i.types[n] & Disp32S)
3480 sign = 1;
3481
3482 if (i.types[n] & (Disp16 | Disp64))
3483 {
3484 size = 2;
3485 if (i.types[n] & Disp64)
3486 size = 8;
3487 }
3488
3489 p = frag_more (size);
3490 reloc_type = reloc (size, pcrel, sign, i.reloc[n]);
3491 if (reloc_type == BFD_RELOC_32
3492 && GOT_symbol
3493 && GOT_symbol == i.op[n].disps->X_add_symbol
3494 && (i.op[n].disps->X_op == O_symbol
3495 || (i.op[n].disps->X_op == O_add
3496 && ((symbol_get_value_expression
3497 (i.op[n].disps->X_op_symbol)->X_op)
3498 == O_subtract))))
3499 {
3500 offsetT add;
3501
3502 if (insn_start_frag == frag_now)
3503 add = (p - frag_now->fr_literal) - insn_start_off;
3504 else
3505 {
3506 fragS *fr;
3507
3508 add = insn_start_frag->fr_fix - insn_start_off;
3509 for (fr = insn_start_frag->fr_next;
3510 fr && fr != frag_now; fr = fr->fr_next)
3511 add += fr->fr_fix;
3512 add += p - frag_now->fr_literal;
3513 }
3514
3515 /* We don't support dynamic linking on x86-64 yet. */
3516 if (flag_code == CODE_64BIT)
3517 abort ();
3518 reloc_type = BFD_RELOC_386_GOTPC;
3519 i.op[n].disps->X_add_number += add;
3520 }
3521 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3522 i.op[n].disps, pcrel, reloc_type);
3523 }
3524 }
3525 }
3526 }
3527
3528 static void
3529 output_imm (insn_start_frag, insn_start_off)
3530 fragS *insn_start_frag;
3531 offsetT insn_start_off;
3532 {
3533 char *p;
3534 unsigned int n;
3535
3536 for (n = 0; n < i.operands; n++)
3537 {
3538 if (i.types[n] & Imm)
3539 {
3540 if (i.op[n].imms->X_op == O_constant)
3541 {
3542 int size;
3543 offsetT val;
3544
3545 size = 4;
3546 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3547 {
3548 size = 2;
3549 if (i.types[n] & (Imm8 | Imm8S))
3550 size = 1;
3551 else if (i.types[n] & Imm64)
3552 size = 8;
3553 }
3554 val = offset_in_range (i.op[n].imms->X_add_number,
3555 size);
3556 p = frag_more (size);
3557 md_number_to_chars (p, val, size);
3558 }
3559 else
3560 {
3561 /* Not absolute_section.
3562 Need a 32-bit fixup (don't support 8bit
3563 non-absolute imms). Try to support other
3564 sizes ... */
3565 enum bfd_reloc_code_real reloc_type;
3566 int size = 4;
3567 int sign = 0;
3568
3569 if ((i.types[n] & (Imm32S))
3570 && i.suffix == QWORD_MNEM_SUFFIX)
3571 sign = 1;
3572 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
3573 {
3574 size = 2;
3575 if (i.types[n] & (Imm8 | Imm8S))
3576 size = 1;
3577 if (i.types[n] & Imm64)
3578 size = 8;
3579 }
3580
3581 p = frag_more (size);
3582 reloc_type = reloc (size, 0, sign, i.reloc[n]);
3583
3584 /* This is tough to explain. We end up with this one if we
3585 * have operands that look like
3586 * "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal here is to
3587 * obtain the absolute address of the GOT, and it is strongly
3588 * preferable from a performance point of view to avoid using
3589 * a runtime relocation for this. The actual sequence of
3590 * instructions often look something like:
3591 *
3592 * call .L66
3593 * .L66:
3594 * popl %ebx
3595 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
3596 *
3597 * The call and pop essentially return the absolute address
3598 * of the label .L66 and store it in %ebx. The linker itself
3599 * will ultimately change the first operand of the addl so
3600 * that %ebx points to the GOT, but to keep things simple, the
3601 * .o file must have this operand set so that it generates not
3602 * the absolute address of .L66, but the absolute address of
3603 * itself. This allows the linker itself simply treat a GOTPC
3604 * relocation as asking for a pcrel offset to the GOT to be
3605 * added in, and the addend of the relocation is stored in the
3606 * operand field for the instruction itself.
3607 *
3608 * Our job here is to fix the operand so that it would add
3609 * the correct offset so that %ebx would point to itself. The
3610 * thing that is tricky is that .-.L66 will point to the
3611 * beginning of the instruction, so we need to further modify
3612 * the operand so that it will point to itself. There are
3613 * other cases where you have something like:
3614 *
3615 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
3616 *
3617 * and here no correction would be required. Internally in
3618 * the assembler we treat operands of this form as not being
3619 * pcrel since the '.' is explicitly mentioned, and I wonder
3620 * whether it would simplify matters to do it this way. Who
3621 * knows. In earlier versions of the PIC patches, the
3622 * pcrel_adjust field was used to store the correction, but
3623 * since the expression is not pcrel, I felt it would be
3624 * confusing to do it this way. */
3625
3626 if (reloc_type == BFD_RELOC_32
3627 && GOT_symbol
3628 && GOT_symbol == i.op[n].imms->X_add_symbol
3629 && (i.op[n].imms->X_op == O_symbol
3630 || (i.op[n].imms->X_op == O_add
3631 && ((symbol_get_value_expression
3632 (i.op[n].imms->X_op_symbol)->X_op)
3633 == O_subtract))))
3634 {
3635 offsetT add;
3636
3637 if (insn_start_frag == frag_now)
3638 add = (p - frag_now->fr_literal) - insn_start_off;
3639 else
3640 {
3641 fragS *fr;
3642
3643 add = insn_start_frag->fr_fix - insn_start_off;
3644 for (fr = insn_start_frag->fr_next;
3645 fr && fr != frag_now; fr = fr->fr_next)
3646 add += fr->fr_fix;
3647 add += p - frag_now->fr_literal;
3648 }
3649
3650 /* We don't support dynamic linking on x86-64 yet. */
3651 if (flag_code == CODE_64BIT)
3652 abort ();
3653 reloc_type = BFD_RELOC_386_GOTPC;
3654 i.op[n].imms->X_add_number += add;
3655 }
3656 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3657 i.op[n].imms, 0, reloc_type);
3658 }
3659 }
3660 }
3661 }
3662 \f
3663 #ifndef LEX_AT
3664 static char *lex_got PARAMS ((enum bfd_reloc_code_real *, int *));
3665
3666 /* Parse operands of the form
3667 <symbol>@GOTOFF+<nnn>
3668 and similar .plt or .got references.
3669
3670 If we find one, set up the correct relocation in RELOC and copy the
3671 input string, minus the `@GOTOFF' into a malloc'd buffer for
3672 parsing by the calling routine. Return this buffer, and if ADJUST
3673 is non-null set it to the length of the string we removed from the
3674 input line. Otherwise return NULL. */
3675 static char *
3676 lex_got (reloc, adjust)
3677 enum bfd_reloc_code_real *reloc;
3678 int *adjust;
3679 {
3680 static const char * const mode_name[NUM_FLAG_CODE] = { "32", "16", "64" };
3681 static const struct {
3682 const char *str;
3683 const enum bfd_reloc_code_real rel[NUM_FLAG_CODE];
3684 } gotrel[] = {
3685 { "PLT", { BFD_RELOC_386_PLT32, 0, BFD_RELOC_X86_64_PLT32 } },
3686 { "GOTOFF", { BFD_RELOC_386_GOTOFF, 0, 0 } },
3687 { "GOTPCREL", { 0, 0, BFD_RELOC_X86_64_GOTPCREL } },
3688 { "TLSGD", { BFD_RELOC_386_TLS_GD, 0, BFD_RELOC_X86_64_TLSGD } },
3689 { "TLSLDM", { BFD_RELOC_386_TLS_LDM, 0, 0 } },
3690 { "TLSLD", { 0, 0, BFD_RELOC_X86_64_TLSLD } },
3691 { "GOTTPOFF", { BFD_RELOC_386_TLS_IE_32, 0, BFD_RELOC_X86_64_GOTTPOFF } },
3692 { "TPOFF", { BFD_RELOC_386_TLS_LE_32, 0, BFD_RELOC_X86_64_TPOFF32 } },
3693 { "NTPOFF", { BFD_RELOC_386_TLS_LE, 0, 0 } },
3694 { "DTPOFF", { BFD_RELOC_386_TLS_LDO_32, 0, BFD_RELOC_X86_64_DTPOFF32 } },
3695 { "GOTNTPOFF",{ BFD_RELOC_386_TLS_GOTIE, 0, 0 } },
3696 { "INDNTPOFF",{ BFD_RELOC_386_TLS_IE, 0, 0 } },
3697 { "GOT", { BFD_RELOC_386_GOT32, 0, BFD_RELOC_X86_64_GOT32 } }
3698 };
3699 char *cp;
3700 unsigned int j;
3701
3702 for (cp = input_line_pointer; *cp != '@'; cp++)
3703 if (is_end_of_line[(unsigned char) *cp])
3704 return NULL;
3705
3706 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
3707 {
3708 int len;
3709
3710 len = strlen (gotrel[j].str);
3711 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
3712 {
3713 if (gotrel[j].rel[(unsigned int) flag_code] != 0)
3714 {
3715 int first, second;
3716 char *tmpbuf, *past_reloc;
3717
3718 *reloc = gotrel[j].rel[(unsigned int) flag_code];
3719 if (adjust)
3720 *adjust = len;
3721
3722 if (GOT_symbol == NULL)
3723 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
3724
3725 /* Replace the relocation token with ' ', so that
3726 errors like foo@GOTOFF1 will be detected. */
3727
3728 /* The length of the first part of our input line. */
3729 first = cp - input_line_pointer;
3730
3731 /* The second part goes from after the reloc token until
3732 (and including) an end_of_line char. Don't use strlen
3733 here as the end_of_line char may not be a NUL. */
3734 past_reloc = cp + 1 + len;
3735 for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
3736 ;
3737 second = cp - past_reloc;
3738
3739 /* Allocate and copy string. The trailing NUL shouldn't
3740 be necessary, but be safe. */
3741 tmpbuf = xmalloc (first + second + 2);
3742 memcpy (tmpbuf, input_line_pointer, first);
3743 tmpbuf[first] = ' ';
3744 memcpy (tmpbuf + first + 1, past_reloc, second);
3745 tmpbuf[first + second + 1] = '\0';
3746 return tmpbuf;
3747 }
3748
3749 as_bad (_("@%s reloc is not supported in %s bit mode"),
3750 gotrel[j].str, mode_name[(unsigned int) flag_code]);
3751 return NULL;
3752 }
3753 }
3754
3755 /* Might be a symbol version string. Don't as_bad here. */
3756 return NULL;
3757 }
3758
3759 /* x86_cons_fix_new is called via the expression parsing code when a
3760 reloc is needed. We use this hook to get the correct .got reloc. */
3761 static enum bfd_reloc_code_real got_reloc = NO_RELOC;
3762
3763 void
3764 x86_cons_fix_new (frag, off, len, exp)
3765 fragS *frag;
3766 unsigned int off;
3767 unsigned int len;
3768 expressionS *exp;
3769 {
3770 enum bfd_reloc_code_real r = reloc (len, 0, 0, got_reloc);
3771 got_reloc = NO_RELOC;
3772 fix_new_exp (frag, off, len, exp, 0, r);
3773 }
3774
3775 void
3776 x86_cons (exp, size)
3777 expressionS *exp;
3778 int size;
3779 {
3780 if (size == 4)
3781 {
3782 /* Handle @GOTOFF and the like in an expression. */
3783 char *save;
3784 char *gotfree_input_line;
3785 int adjust;
3786
3787 save = input_line_pointer;
3788 gotfree_input_line = lex_got (&got_reloc, &adjust);
3789 if (gotfree_input_line)
3790 input_line_pointer = gotfree_input_line;
3791
3792 expression (exp);
3793
3794 if (gotfree_input_line)
3795 {
3796 /* expression () has merrily parsed up to the end of line,
3797 or a comma - in the wrong buffer. Transfer how far
3798 input_line_pointer has moved to the right buffer. */
3799 input_line_pointer = (save
3800 + (input_line_pointer - gotfree_input_line)
3801 + adjust);
3802 free (gotfree_input_line);
3803 }
3804 }
3805 else
3806 expression (exp);
3807 }
3808 #endif
3809
3810 #ifdef TE_PE
3811
3812 void
3813 x86_pe_cons_fix_new (frag, off, len, exp)
3814 fragS *frag;
3815 unsigned int off;
3816 unsigned int len;
3817 expressionS *exp;
3818 {
3819 enum bfd_reloc_code_real r = reloc (len, 0, 0, NO_RELOC);
3820
3821 if (exp->X_op == O_secrel)
3822 {
3823 exp->X_op = O_symbol;
3824 r = BFD_RELOC_32_SECREL;
3825 }
3826
3827 fix_new_exp (frag, off, len, exp, 0, r);
3828 }
3829
3830 static void
3831 pe_directive_secrel (dummy)
3832 int dummy ATTRIBUTE_UNUSED;
3833 {
3834 expressionS exp;
3835
3836 do
3837 {
3838 expression (&exp);
3839 if (exp.X_op == O_symbol)
3840 exp.X_op = O_secrel;
3841
3842 emit_expr (&exp, 4);
3843 }
3844 while (*input_line_pointer++ == ',');
3845
3846 input_line_pointer--;
3847 demand_empty_rest_of_line ();
3848 }
3849
3850 #endif
3851
3852 static int i386_immediate PARAMS ((char *));
3853
3854 static int
3855 i386_immediate (imm_start)
3856 char *imm_start;
3857 {
3858 char *save_input_line_pointer;
3859 #ifndef LEX_AT
3860 char *gotfree_input_line;
3861 #endif
3862 segT exp_seg = 0;
3863 expressionS *exp;
3864
3865 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
3866 {
3867 as_bad (_("only 1 or 2 immediate operands are allowed"));
3868 return 0;
3869 }
3870
3871 exp = &im_expressions[i.imm_operands++];
3872 i.op[this_operand].imms = exp;
3873
3874 if (is_space_char (*imm_start))
3875 ++imm_start;
3876
3877 save_input_line_pointer = input_line_pointer;
3878 input_line_pointer = imm_start;
3879
3880 #ifndef LEX_AT
3881 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
3882 if (gotfree_input_line)
3883 input_line_pointer = gotfree_input_line;
3884 #endif
3885
3886 exp_seg = expression (exp);
3887
3888 SKIP_WHITESPACE ();
3889 if (*input_line_pointer)
3890 as_bad (_("junk `%s' after expression"), input_line_pointer);
3891
3892 input_line_pointer = save_input_line_pointer;
3893 #ifndef LEX_AT
3894 if (gotfree_input_line)
3895 free (gotfree_input_line);
3896 #endif
3897
3898 if (exp->X_op == O_absent || exp->X_op == O_big)
3899 {
3900 /* Missing or bad expr becomes absolute 0. */
3901 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
3902 imm_start);
3903 exp->X_op = O_constant;
3904 exp->X_add_number = 0;
3905 exp->X_add_symbol = (symbolS *) 0;
3906 exp->X_op_symbol = (symbolS *) 0;
3907 }
3908 else if (exp->X_op == O_constant)
3909 {
3910 /* Size it properly later. */
3911 i.types[this_operand] |= Imm64;
3912 /* If BFD64, sign extend val. */
3913 if (!use_rela_relocations)
3914 if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
3915 exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
3916 }
3917 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
3918 else if (OUTPUT_FLAVOR == bfd_target_aout_flavour
3919 && exp_seg != absolute_section
3920 && exp_seg != text_section
3921 && exp_seg != data_section
3922 && exp_seg != bss_section
3923 && exp_seg != undefined_section
3924 && !bfd_is_com_section (exp_seg))
3925 {
3926 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
3927 return 0;
3928 }
3929 #endif
3930 else
3931 {
3932 /* This is an address. The size of the address will be
3933 determined later, depending on destination register,
3934 suffix, or the default for the section. */
3935 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
3936 }
3937
3938 return 1;
3939 }
3940
3941 static char *i386_scale PARAMS ((char *));
3942
3943 static char *
3944 i386_scale (scale)
3945 char *scale;
3946 {
3947 offsetT val;
3948 char *save = input_line_pointer;
3949
3950 input_line_pointer = scale;
3951 val = get_absolute_expression ();
3952
3953 switch (val)
3954 {
3955 case 1:
3956 i.log2_scale_factor = 0;
3957 break;
3958 case 2:
3959 i.log2_scale_factor = 1;
3960 break;
3961 case 4:
3962 i.log2_scale_factor = 2;
3963 break;
3964 case 8:
3965 i.log2_scale_factor = 3;
3966 break;
3967 default:
3968 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
3969 scale);
3970 input_line_pointer = save;
3971 return NULL;
3972 }
3973 if (i.log2_scale_factor != 0 && i.index_reg == 0)
3974 {
3975 as_warn (_("scale factor of %d without an index register"),
3976 1 << i.log2_scale_factor);
3977 #if SCALE1_WHEN_NO_INDEX
3978 i.log2_scale_factor = 0;
3979 #endif
3980 }
3981 scale = input_line_pointer;
3982 input_line_pointer = save;
3983 return scale;
3984 }
3985
3986 static int i386_displacement PARAMS ((char *, char *));
3987
3988 static int
3989 i386_displacement (disp_start, disp_end)
3990 char *disp_start;
3991 char *disp_end;
3992 {
3993 expressionS *exp;
3994 segT exp_seg = 0;
3995 char *save_input_line_pointer;
3996 #ifndef LEX_AT
3997 char *gotfree_input_line;
3998 #endif
3999 int bigdisp = Disp32;
4000
4001 if (flag_code == CODE_64BIT)
4002 {
4003 if (i.prefix[ADDR_PREFIX] == 0)
4004 bigdisp = Disp64;
4005 }
4006 else if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
4007 bigdisp = Disp16;
4008 i.types[this_operand] |= bigdisp;
4009
4010 exp = &disp_expressions[i.disp_operands];
4011 i.op[this_operand].disps = exp;
4012 i.disp_operands++;
4013 save_input_line_pointer = input_line_pointer;
4014 input_line_pointer = disp_start;
4015 END_STRING_AND_SAVE (disp_end);
4016
4017 #ifndef GCC_ASM_O_HACK
4018 #define GCC_ASM_O_HACK 0
4019 #endif
4020 #if GCC_ASM_O_HACK
4021 END_STRING_AND_SAVE (disp_end + 1);
4022 if ((i.types[this_operand] & BaseIndex) != 0
4023 && displacement_string_end[-1] == '+')
4024 {
4025 /* This hack is to avoid a warning when using the "o"
4026 constraint within gcc asm statements.
4027 For instance:
4028
4029 #define _set_tssldt_desc(n,addr,limit,type) \
4030 __asm__ __volatile__ ( \
4031 "movw %w2,%0\n\t" \
4032 "movw %w1,2+%0\n\t" \
4033 "rorl $16,%1\n\t" \
4034 "movb %b1,4+%0\n\t" \
4035 "movb %4,5+%0\n\t" \
4036 "movb $0,6+%0\n\t" \
4037 "movb %h1,7+%0\n\t" \
4038 "rorl $16,%1" \
4039 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
4040
4041 This works great except that the output assembler ends
4042 up looking a bit weird if it turns out that there is
4043 no offset. You end up producing code that looks like:
4044
4045 #APP
4046 movw $235,(%eax)
4047 movw %dx,2+(%eax)
4048 rorl $16,%edx
4049 movb %dl,4+(%eax)
4050 movb $137,5+(%eax)
4051 movb $0,6+(%eax)
4052 movb %dh,7+(%eax)
4053 rorl $16,%edx
4054 #NO_APP
4055
4056 So here we provide the missing zero. */
4057
4058 *displacement_string_end = '0';
4059 }
4060 #endif
4061 #ifndef LEX_AT
4062 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
4063 if (gotfree_input_line)
4064 input_line_pointer = gotfree_input_line;
4065 #endif
4066
4067 exp_seg = expression (exp);
4068
4069 SKIP_WHITESPACE ();
4070 if (*input_line_pointer)
4071 as_bad (_("junk `%s' after expression"), input_line_pointer);
4072 #if GCC_ASM_O_HACK
4073 RESTORE_END_STRING (disp_end + 1);
4074 #endif
4075 RESTORE_END_STRING (disp_end);
4076 input_line_pointer = save_input_line_pointer;
4077 #ifndef LEX_AT
4078 if (gotfree_input_line)
4079 free (gotfree_input_line);
4080 #endif
4081
4082 /* We do this to make sure that the section symbol is in
4083 the symbol table. We will ultimately change the relocation
4084 to be relative to the beginning of the section. */
4085 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
4086 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
4087 {
4088 if (exp->X_op != O_symbol)
4089 {
4090 as_bad (_("bad expression used with @%s"),
4091 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
4092 ? "GOTPCREL"
4093 : "GOTOFF"));
4094 return 0;
4095 }
4096
4097 if (S_IS_LOCAL (exp->X_add_symbol)
4098 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
4099 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
4100 exp->X_op = O_subtract;
4101 exp->X_op_symbol = GOT_symbol;
4102 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
4103 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
4104 else
4105 i.reloc[this_operand] = BFD_RELOC_32;
4106 }
4107
4108 if (exp->X_op == O_absent || exp->X_op == O_big)
4109 {
4110 /* Missing or bad expr becomes absolute 0. */
4111 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
4112 disp_start);
4113 exp->X_op = O_constant;
4114 exp->X_add_number = 0;
4115 exp->X_add_symbol = (symbolS *) 0;
4116 exp->X_op_symbol = (symbolS *) 0;
4117 }
4118
4119 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4120 if (exp->X_op != O_constant
4121 && OUTPUT_FLAVOR == bfd_target_aout_flavour
4122 && exp_seg != absolute_section
4123 && exp_seg != text_section
4124 && exp_seg != data_section
4125 && exp_seg != bss_section
4126 && exp_seg != undefined_section
4127 && !bfd_is_com_section (exp_seg))
4128 {
4129 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
4130 return 0;
4131 }
4132 #endif
4133 else if (flag_code == CODE_64BIT)
4134 i.types[this_operand] |= Disp32S | Disp32;
4135 return 1;
4136 }
4137
4138 static int i386_index_check PARAMS ((const char *));
4139
4140 /* Make sure the memory operand we've been dealt is valid.
4141 Return 1 on success, 0 on a failure. */
4142
4143 static int
4144 i386_index_check (operand_string)
4145 const char *operand_string;
4146 {
4147 int ok;
4148 #if INFER_ADDR_PREFIX
4149 int fudged = 0;
4150
4151 tryprefix:
4152 #endif
4153 ok = 1;
4154 if (flag_code == CODE_64BIT)
4155 {
4156 unsigned RegXX = (i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32);
4157
4158 if ((i.base_reg
4159 && ((i.base_reg->reg_type & RegXX) == 0)
4160 && (i.base_reg->reg_type != BaseIndex
4161 || i.index_reg))
4162 || (i.index_reg
4163 && ((i.index_reg->reg_type & (RegXX | BaseIndex))
4164 != (RegXX | BaseIndex))))
4165 ok = 0;
4166 }
4167 else
4168 {
4169 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
4170 {
4171 /* 16bit checks. */
4172 if ((i.base_reg
4173 && ((i.base_reg->reg_type & (Reg16 | BaseIndex | RegRex))
4174 != (Reg16 | BaseIndex)))
4175 || (i.index_reg
4176 && (((i.index_reg->reg_type & (Reg16 | BaseIndex))
4177 != (Reg16 | BaseIndex))
4178 || !(i.base_reg
4179 && i.base_reg->reg_num < 6
4180 && i.index_reg->reg_num >= 6
4181 && i.log2_scale_factor == 0))))
4182 ok = 0;
4183 }
4184 else
4185 {
4186 /* 32bit checks. */
4187 if ((i.base_reg
4188 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
4189 || (i.index_reg
4190 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
4191 != (Reg32 | BaseIndex))))
4192 ok = 0;
4193 }
4194 }
4195 if (!ok)
4196 {
4197 #if INFER_ADDR_PREFIX
4198 if (i.prefix[ADDR_PREFIX] == 0)
4199 {
4200 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
4201 i.prefixes += 1;
4202 /* Change the size of any displacement too. At most one of
4203 Disp16 or Disp32 is set.
4204 FIXME. There doesn't seem to be any real need for separate
4205 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
4206 Removing them would probably clean up the code quite a lot. */
4207 if (flag_code != CODE_64BIT && (i.types[this_operand] & (Disp16 | Disp32)))
4208 i.types[this_operand] ^= (Disp16 | Disp32);
4209 fudged = 1;
4210 goto tryprefix;
4211 }
4212 if (fudged)
4213 as_bad (_("`%s' is not a valid base/index expression"),
4214 operand_string);
4215 else
4216 #endif
4217 as_bad (_("`%s' is not a valid %s bit base/index expression"),
4218 operand_string,
4219 flag_code_names[flag_code]);
4220 }
4221 return ok;
4222 }
4223
4224 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
4225 on error. */
4226
4227 static int
4228 i386_operand (operand_string)
4229 char *operand_string;
4230 {
4231 const reg_entry *r;
4232 char *end_op;
4233 char *op_string = operand_string;
4234
4235 if (is_space_char (*op_string))
4236 ++op_string;
4237
4238 /* We check for an absolute prefix (differentiating,
4239 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
4240 if (*op_string == ABSOLUTE_PREFIX)
4241 {
4242 ++op_string;
4243 if (is_space_char (*op_string))
4244 ++op_string;
4245 i.types[this_operand] |= JumpAbsolute;
4246 }
4247
4248 /* Check if operand is a register. */
4249 if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
4250 && (r = parse_register (op_string, &end_op)) != NULL)
4251 {
4252 /* Check for a segment override by searching for ':' after a
4253 segment register. */
4254 op_string = end_op;
4255 if (is_space_char (*op_string))
4256 ++op_string;
4257 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
4258 {
4259 switch (r->reg_num)
4260 {
4261 case 0:
4262 i.seg[i.mem_operands] = &es;
4263 break;
4264 case 1:
4265 i.seg[i.mem_operands] = &cs;
4266 break;
4267 case 2:
4268 i.seg[i.mem_operands] = &ss;
4269 break;
4270 case 3:
4271 i.seg[i.mem_operands] = &ds;
4272 break;
4273 case 4:
4274 i.seg[i.mem_operands] = &fs;
4275 break;
4276 case 5:
4277 i.seg[i.mem_operands] = &gs;
4278 break;
4279 }
4280
4281 /* Skip the ':' and whitespace. */
4282 ++op_string;
4283 if (is_space_char (*op_string))
4284 ++op_string;
4285
4286 if (!is_digit_char (*op_string)
4287 && !is_identifier_char (*op_string)
4288 && *op_string != '('
4289 && *op_string != ABSOLUTE_PREFIX)
4290 {
4291 as_bad (_("bad memory operand `%s'"), op_string);
4292 return 0;
4293 }
4294 /* Handle case of %es:*foo. */
4295 if (*op_string == ABSOLUTE_PREFIX)
4296 {
4297 ++op_string;
4298 if (is_space_char (*op_string))
4299 ++op_string;
4300 i.types[this_operand] |= JumpAbsolute;
4301 }
4302 goto do_memory_reference;
4303 }
4304 if (*op_string)
4305 {
4306 as_bad (_("junk `%s' after register"), op_string);
4307 return 0;
4308 }
4309 i.types[this_operand] |= r->reg_type & ~BaseIndex;
4310 i.op[this_operand].regs = r;
4311 i.reg_operands++;
4312 }
4313 else if (*op_string == REGISTER_PREFIX)
4314 {
4315 as_bad (_("bad register name `%s'"), op_string);
4316 return 0;
4317 }
4318 else if (*op_string == IMMEDIATE_PREFIX)
4319 {
4320 ++op_string;
4321 if (i.types[this_operand] & JumpAbsolute)
4322 {
4323 as_bad (_("immediate operand illegal with absolute jump"));
4324 return 0;
4325 }
4326 if (!i386_immediate (op_string))
4327 return 0;
4328 }
4329 else if (is_digit_char (*op_string)
4330 || is_identifier_char (*op_string)
4331 || *op_string == '(')
4332 {
4333 /* This is a memory reference of some sort. */
4334 char *base_string;
4335
4336 /* Start and end of displacement string expression (if found). */
4337 char *displacement_string_start;
4338 char *displacement_string_end;
4339
4340 do_memory_reference:
4341 if ((i.mem_operands == 1
4342 && (current_templates->start->opcode_modifier & IsString) == 0)
4343 || i.mem_operands == 2)
4344 {
4345 as_bad (_("too many memory references for `%s'"),
4346 current_templates->start->name);
4347 return 0;
4348 }
4349
4350 /* Check for base index form. We detect the base index form by
4351 looking for an ')' at the end of the operand, searching
4352 for the '(' matching it, and finding a REGISTER_PREFIX or ','
4353 after the '('. */
4354 base_string = op_string + strlen (op_string);
4355
4356 --base_string;
4357 if (is_space_char (*base_string))
4358 --base_string;
4359
4360 /* If we only have a displacement, set-up for it to be parsed later. */
4361 displacement_string_start = op_string;
4362 displacement_string_end = base_string + 1;
4363
4364 if (*base_string == ')')
4365 {
4366 char *temp_string;
4367 unsigned int parens_balanced = 1;
4368 /* We've already checked that the number of left & right ()'s are
4369 equal, so this loop will not be infinite. */
4370 do
4371 {
4372 base_string--;
4373 if (*base_string == ')')
4374 parens_balanced++;
4375 if (*base_string == '(')
4376 parens_balanced--;
4377 }
4378 while (parens_balanced);
4379
4380 temp_string = base_string;
4381
4382 /* Skip past '(' and whitespace. */
4383 ++base_string;
4384 if (is_space_char (*base_string))
4385 ++base_string;
4386
4387 if (*base_string == ','
4388 || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
4389 && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
4390 {
4391 displacement_string_end = temp_string;
4392
4393 i.types[this_operand] |= BaseIndex;
4394
4395 if (i.base_reg)
4396 {
4397 base_string = end_op;
4398 if (is_space_char (*base_string))
4399 ++base_string;
4400 }
4401
4402 /* There may be an index reg or scale factor here. */
4403 if (*base_string == ',')
4404 {
4405 ++base_string;
4406 if (is_space_char (*base_string))
4407 ++base_string;
4408
4409 if ((*base_string == REGISTER_PREFIX || allow_naked_reg)
4410 && (i.index_reg = parse_register (base_string, &end_op)) != NULL)
4411 {
4412 base_string = end_op;
4413 if (is_space_char (*base_string))
4414 ++base_string;
4415 if (*base_string == ',')
4416 {
4417 ++base_string;
4418 if (is_space_char (*base_string))
4419 ++base_string;
4420 }
4421 else if (*base_string != ')')
4422 {
4423 as_bad (_("expecting `,' or `)' after index register in `%s'"),
4424 operand_string);
4425 return 0;
4426 }
4427 }
4428 else if (*base_string == REGISTER_PREFIX)
4429 {
4430 as_bad (_("bad register name `%s'"), base_string);
4431 return 0;
4432 }
4433
4434 /* Check for scale factor. */
4435 if (*base_string != ')')
4436 {
4437 char *end_scale = i386_scale (base_string);
4438
4439 if (!end_scale)
4440 return 0;
4441
4442 base_string = end_scale;
4443 if (is_space_char (*base_string))
4444 ++base_string;
4445 if (*base_string != ')')
4446 {
4447 as_bad (_("expecting `)' after scale factor in `%s'"),
4448 operand_string);
4449 return 0;
4450 }
4451 }
4452 else if (!i.index_reg)
4453 {
4454 as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
4455 *base_string);
4456 return 0;
4457 }
4458 }
4459 else if (*base_string != ')')
4460 {
4461 as_bad (_("expecting `,' or `)' after base register in `%s'"),
4462 operand_string);
4463 return 0;
4464 }
4465 }
4466 else if (*base_string == REGISTER_PREFIX)
4467 {
4468 as_bad (_("bad register name `%s'"), base_string);
4469 return 0;
4470 }
4471 }
4472
4473 /* If there's an expression beginning the operand, parse it,
4474 assuming displacement_string_start and
4475 displacement_string_end are meaningful. */
4476 if (displacement_string_start != displacement_string_end)
4477 {
4478 if (!i386_displacement (displacement_string_start,
4479 displacement_string_end))
4480 return 0;
4481 }
4482
4483 /* Special case for (%dx) while doing input/output op. */
4484 if (i.base_reg
4485 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
4486 && i.index_reg == 0
4487 && i.log2_scale_factor == 0
4488 && i.seg[i.mem_operands] == 0
4489 && (i.types[this_operand] & Disp) == 0)
4490 {
4491 i.types[this_operand] = InOutPortReg;
4492 return 1;
4493 }
4494
4495 if (i386_index_check (operand_string) == 0)
4496 return 0;
4497 i.mem_operands++;
4498 }
4499 else
4500 {
4501 /* It's not a memory operand; argh! */
4502 as_bad (_("invalid char %s beginning operand %d `%s'"),
4503 output_invalid (*op_string),
4504 this_operand + 1,
4505 op_string);
4506 return 0;
4507 }
4508 return 1; /* Normal return. */
4509 }
4510 \f
4511 /* md_estimate_size_before_relax()
4512
4513 Called just before relax() for rs_machine_dependent frags. The x86
4514 assembler uses these frags to handle variable size jump
4515 instructions.
4516
4517 Any symbol that is now undefined will not become defined.
4518 Return the correct fr_subtype in the frag.
4519 Return the initial "guess for variable size of frag" to caller.
4520 The guess is actually the growth beyond the fixed part. Whatever
4521 we do to grow the fixed or variable part contributes to our
4522 returned value. */
4523
4524 int
4525 md_estimate_size_before_relax (fragP, segment)
4526 fragS *fragP;
4527 segT segment;
4528 {
4529 /* We've already got fragP->fr_subtype right; all we have to do is
4530 check for un-relaxable symbols. On an ELF system, we can't relax
4531 an externally visible symbol, because it may be overridden by a
4532 shared library. */
4533 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
4534 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4535 || (OUTPUT_FLAVOR == bfd_target_elf_flavour
4536 && (S_IS_EXTERNAL (fragP->fr_symbol)
4537 || S_IS_WEAK (fragP->fr_symbol)))
4538 #endif
4539 )
4540 {
4541 /* Symbol is undefined in this segment, or we need to keep a
4542 reloc so that weak symbols can be overridden. */
4543 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
4544 enum bfd_reloc_code_real reloc_type;
4545 unsigned char *opcode;
4546 int old_fr_fix;
4547
4548 if (fragP->fr_var != NO_RELOC)
4549 reloc_type = fragP->fr_var;
4550 else if (size == 2)
4551 reloc_type = BFD_RELOC_16_PCREL;
4552 else
4553 reloc_type = BFD_RELOC_32_PCREL;
4554
4555 old_fr_fix = fragP->fr_fix;
4556 opcode = (unsigned char *) fragP->fr_opcode;
4557
4558 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
4559 {
4560 case UNCOND_JUMP:
4561 /* Make jmp (0xeb) a (d)word displacement jump. */
4562 opcode[0] = 0xe9;
4563 fragP->fr_fix += size;
4564 fix_new (fragP, old_fr_fix, size,
4565 fragP->fr_symbol,
4566 fragP->fr_offset, 1,
4567 reloc_type);
4568 break;
4569
4570 case COND_JUMP86:
4571 if (size == 2
4572 && (!no_cond_jump_promotion || fragP->fr_var != NO_RELOC))
4573 {
4574 /* Negate the condition, and branch past an
4575 unconditional jump. */
4576 opcode[0] ^= 1;
4577 opcode[1] = 3;
4578 /* Insert an unconditional jump. */
4579 opcode[2] = 0xe9;
4580 /* We added two extra opcode bytes, and have a two byte
4581 offset. */
4582 fragP->fr_fix += 2 + 2;
4583 fix_new (fragP, old_fr_fix + 2, 2,
4584 fragP->fr_symbol,
4585 fragP->fr_offset, 1,
4586 reloc_type);
4587 break;
4588 }
4589 /* Fall through. */
4590
4591 case COND_JUMP:
4592 if (no_cond_jump_promotion && fragP->fr_var == NO_RELOC)
4593 {
4594 fixS *fixP;
4595
4596 fragP->fr_fix += 1;
4597 fixP = fix_new (fragP, old_fr_fix, 1,
4598 fragP->fr_symbol,
4599 fragP->fr_offset, 1,
4600 BFD_RELOC_8_PCREL);
4601 fixP->fx_signed = 1;
4602 break;
4603 }
4604
4605 /* This changes the byte-displacement jump 0x7N
4606 to the (d)word-displacement jump 0x0f,0x8N. */
4607 opcode[1] = opcode[0] + 0x10;
4608 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4609 /* We've added an opcode byte. */
4610 fragP->fr_fix += 1 + size;
4611 fix_new (fragP, old_fr_fix + 1, size,
4612 fragP->fr_symbol,
4613 fragP->fr_offset, 1,
4614 reloc_type);
4615 break;
4616
4617 default:
4618 BAD_CASE (fragP->fr_subtype);
4619 break;
4620 }
4621 frag_wane (fragP);
4622 return fragP->fr_fix - old_fr_fix;
4623 }
4624
4625 /* Guess size depending on current relax state. Initially the relax
4626 state will correspond to a short jump and we return 1, because
4627 the variable part of the frag (the branch offset) is one byte
4628 long. However, we can relax a section more than once and in that
4629 case we must either set fr_subtype back to the unrelaxed state,
4630 or return the value for the appropriate branch. */
4631 return md_relax_table[fragP->fr_subtype].rlx_length;
4632 }
4633
4634 /* Called after relax() is finished.
4635
4636 In: Address of frag.
4637 fr_type == rs_machine_dependent.
4638 fr_subtype is what the address relaxed to.
4639
4640 Out: Any fixSs and constants are set up.
4641 Caller will turn frag into a ".space 0". */
4642
4643 void
4644 md_convert_frag (abfd, sec, fragP)
4645 bfd *abfd ATTRIBUTE_UNUSED;
4646 segT sec ATTRIBUTE_UNUSED;
4647 fragS *fragP;
4648 {
4649 unsigned char *opcode;
4650 unsigned char *where_to_put_displacement = NULL;
4651 offsetT target_address;
4652 offsetT opcode_address;
4653 unsigned int extension = 0;
4654 offsetT displacement_from_opcode_start;
4655
4656 opcode = (unsigned char *) fragP->fr_opcode;
4657
4658 /* Address we want to reach in file space. */
4659 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
4660
4661 /* Address opcode resides at in file space. */
4662 opcode_address = fragP->fr_address + fragP->fr_fix;
4663
4664 /* Displacement from opcode start to fill into instruction. */
4665 displacement_from_opcode_start = target_address - opcode_address;
4666
4667 if ((fragP->fr_subtype & BIG) == 0)
4668 {
4669 /* Don't have to change opcode. */
4670 extension = 1; /* 1 opcode + 1 displacement */
4671 where_to_put_displacement = &opcode[1];
4672 }
4673 else
4674 {
4675 if (no_cond_jump_promotion
4676 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
4677 as_warn_where (fragP->fr_file, fragP->fr_line, _("long jump required"));
4678
4679 switch (fragP->fr_subtype)
4680 {
4681 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
4682 extension = 4; /* 1 opcode + 4 displacement */
4683 opcode[0] = 0xe9;
4684 where_to_put_displacement = &opcode[1];
4685 break;
4686
4687 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
4688 extension = 2; /* 1 opcode + 2 displacement */
4689 opcode[0] = 0xe9;
4690 where_to_put_displacement = &opcode[1];
4691 break;
4692
4693 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
4694 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
4695 extension = 5; /* 2 opcode + 4 displacement */
4696 opcode[1] = opcode[0] + 0x10;
4697 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4698 where_to_put_displacement = &opcode[2];
4699 break;
4700
4701 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
4702 extension = 3; /* 2 opcode + 2 displacement */
4703 opcode[1] = opcode[0] + 0x10;
4704 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
4705 where_to_put_displacement = &opcode[2];
4706 break;
4707
4708 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
4709 extension = 4;
4710 opcode[0] ^= 1;
4711 opcode[1] = 3;
4712 opcode[2] = 0xe9;
4713 where_to_put_displacement = &opcode[3];
4714 break;
4715
4716 default:
4717 BAD_CASE (fragP->fr_subtype);
4718 break;
4719 }
4720 }
4721
4722 /* Now put displacement after opcode. */
4723 md_number_to_chars ((char *) where_to_put_displacement,
4724 (valueT) (displacement_from_opcode_start - extension),
4725 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
4726 fragP->fr_fix += extension;
4727 }
4728 \f
4729 /* Size of byte displacement jmp. */
4730 int md_short_jump_size = 2;
4731
4732 /* Size of dword displacement jmp. */
4733 int md_long_jump_size = 5;
4734
4735 /* Size of relocation record. */
4736 const int md_reloc_size = 8;
4737
4738 void
4739 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
4740 char *ptr;
4741 addressT from_addr, to_addr;
4742 fragS *frag ATTRIBUTE_UNUSED;
4743 symbolS *to_symbol ATTRIBUTE_UNUSED;
4744 {
4745 offsetT offset;
4746
4747 offset = to_addr - (from_addr + 2);
4748 /* Opcode for byte-disp jump. */
4749 md_number_to_chars (ptr, (valueT) 0xeb, 1);
4750 md_number_to_chars (ptr + 1, (valueT) offset, 1);
4751 }
4752
4753 void
4754 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
4755 char *ptr;
4756 addressT from_addr, to_addr;
4757 fragS *frag ATTRIBUTE_UNUSED;
4758 symbolS *to_symbol ATTRIBUTE_UNUSED;
4759 {
4760 offsetT offset;
4761
4762 offset = to_addr - (from_addr + 5);
4763 md_number_to_chars (ptr, (valueT) 0xe9, 1);
4764 md_number_to_chars (ptr + 1, (valueT) offset, 4);
4765 }
4766 \f
4767 /* Apply a fixup (fixS) to segment data, once it has been determined
4768 by our caller that we have all the info we need to fix it up.
4769
4770 On the 386, immediates, displacements, and data pointers are all in
4771 the same (little-endian) format, so we don't need to care about which
4772 we are handling. */
4773
4774 void
4775 md_apply_fix3 (fixP, valP, seg)
4776 /* The fix we're to put in. */
4777 fixS *fixP;
4778 /* Pointer to the value of the bits. */
4779 valueT *valP;
4780 /* Segment fix is from. */
4781 segT seg ATTRIBUTE_UNUSED;
4782 {
4783 char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
4784 valueT value = *valP;
4785
4786 #if !defined (TE_Mach)
4787 if (fixP->fx_pcrel)
4788 {
4789 switch (fixP->fx_r_type)
4790 {
4791 default:
4792 break;
4793
4794 case BFD_RELOC_32:
4795 fixP->fx_r_type = BFD_RELOC_32_PCREL;
4796 break;
4797 case BFD_RELOC_16:
4798 fixP->fx_r_type = BFD_RELOC_16_PCREL;
4799 break;
4800 case BFD_RELOC_8:
4801 fixP->fx_r_type = BFD_RELOC_8_PCREL;
4802 break;
4803 }
4804 }
4805
4806 if (fixP->fx_addsy != NULL
4807 && (fixP->fx_r_type == BFD_RELOC_32_PCREL
4808 || fixP->fx_r_type == BFD_RELOC_16_PCREL
4809 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
4810 && !use_rela_relocations)
4811 {
4812 /* This is a hack. There should be a better way to handle this.
4813 This covers for the fact that bfd_install_relocation will
4814 subtract the current location (for partial_inplace, PC relative
4815 relocations); see more below. */
4816 #ifndef OBJ_AOUT
4817 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4818 #ifdef TE_PE
4819 || OUTPUT_FLAVOR == bfd_target_coff_flavour
4820 #endif
4821 )
4822 value += fixP->fx_where + fixP->fx_frag->fr_address;
4823 #endif
4824 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4825 if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
4826 {
4827 segT sym_seg = S_GET_SEGMENT (fixP->fx_addsy);
4828
4829 if ((sym_seg == seg
4830 || (symbol_section_p (fixP->fx_addsy)
4831 && sym_seg != absolute_section))
4832 && !generic_force_reloc (fixP))
4833 {
4834 /* Yes, we add the values in twice. This is because
4835 bfd_install_relocation subtracts them out again. I think
4836 bfd_install_relocation is broken, but I don't dare change
4837 it. FIXME. */
4838 value += fixP->fx_where + fixP->fx_frag->fr_address;
4839 }
4840 }
4841 #endif
4842 #if defined (OBJ_COFF) && defined (TE_PE)
4843 /* For some reason, the PE format does not store a
4844 section address offset for a PC relative symbol. */
4845 if (S_GET_SEGMENT (fixP->fx_addsy) != seg
4846 #if defined(BFD_ASSEMBLER) || defined(S_IS_WEAK)
4847 || S_IS_WEAK (fixP->fx_addsy)
4848 #endif
4849 )
4850 value += md_pcrel_from (fixP);
4851 #endif
4852 }
4853
4854 /* Fix a few things - the dynamic linker expects certain values here,
4855 and we must not disappoint it. */
4856 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
4857 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
4858 && fixP->fx_addsy)
4859 switch (fixP->fx_r_type)
4860 {
4861 case BFD_RELOC_386_PLT32:
4862 case BFD_RELOC_X86_64_PLT32:
4863 /* Make the jump instruction point to the address of the operand. At
4864 runtime we merely add the offset to the actual PLT entry. */
4865 value = -4;
4866 break;
4867
4868 case BFD_RELOC_386_TLS_GD:
4869 case BFD_RELOC_386_TLS_LDM:
4870 case BFD_RELOC_386_TLS_IE_32:
4871 case BFD_RELOC_386_TLS_IE:
4872 case BFD_RELOC_386_TLS_GOTIE:
4873 case BFD_RELOC_X86_64_TLSGD:
4874 case BFD_RELOC_X86_64_TLSLD:
4875 case BFD_RELOC_X86_64_GOTTPOFF:
4876 value = 0; /* Fully resolved at runtime. No addend. */
4877 /* Fallthrough */
4878 case BFD_RELOC_386_TLS_LE:
4879 case BFD_RELOC_386_TLS_LDO_32:
4880 case BFD_RELOC_386_TLS_LE_32:
4881 case BFD_RELOC_X86_64_DTPOFF32:
4882 case BFD_RELOC_X86_64_TPOFF32:
4883 S_SET_THREAD_LOCAL (fixP->fx_addsy);
4884 break;
4885
4886 case BFD_RELOC_386_GOT32:
4887 case BFD_RELOC_X86_64_GOT32:
4888 value = 0; /* Fully resolved at runtime. No addend. */
4889 break;
4890
4891 case BFD_RELOC_VTABLE_INHERIT:
4892 case BFD_RELOC_VTABLE_ENTRY:
4893 fixP->fx_done = 0;
4894 return;
4895
4896 default:
4897 break;
4898 }
4899 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
4900 *valP = value;
4901 #endif /* !defined (TE_Mach) */
4902
4903 /* Are we finished with this relocation now? */
4904 if (fixP->fx_addsy == NULL)
4905 fixP->fx_done = 1;
4906 else if (use_rela_relocations)
4907 {
4908 fixP->fx_no_overflow = 1;
4909 /* Remember value for tc_gen_reloc. */
4910 fixP->fx_addnumber = value;
4911 value = 0;
4912 }
4913
4914 md_number_to_chars (p, value, fixP->fx_size);
4915 }
4916 \f
4917 #define MAX_LITTLENUMS 6
4918
4919 /* Turn the string pointed to by litP into a floating point constant
4920 of type TYPE, and emit the appropriate bytes. The number of
4921 LITTLENUMS emitted is stored in *SIZEP. An error message is
4922 returned, or NULL on OK. */
4923
4924 char *
4925 md_atof (type, litP, sizeP)
4926 int type;
4927 char *litP;
4928 int *sizeP;
4929 {
4930 int prec;
4931 LITTLENUM_TYPE words[MAX_LITTLENUMS];
4932 LITTLENUM_TYPE *wordP;
4933 char *t;
4934
4935 switch (type)
4936 {
4937 case 'f':
4938 case 'F':
4939 prec = 2;
4940 break;
4941
4942 case 'd':
4943 case 'D':
4944 prec = 4;
4945 break;
4946
4947 case 'x':
4948 case 'X':
4949 prec = 5;
4950 break;
4951
4952 default:
4953 *sizeP = 0;
4954 return _("Bad call to md_atof ()");
4955 }
4956 t = atof_ieee (input_line_pointer, type, words);
4957 if (t)
4958 input_line_pointer = t;
4959
4960 *sizeP = prec * sizeof (LITTLENUM_TYPE);
4961 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
4962 the bigendian 386. */
4963 for (wordP = words + prec - 1; prec--;)
4964 {
4965 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
4966 litP += sizeof (LITTLENUM_TYPE);
4967 }
4968 return 0;
4969 }
4970 \f
4971 char output_invalid_buf[8];
4972
4973 static char *
4974 output_invalid (c)
4975 int c;
4976 {
4977 if (ISPRINT (c))
4978 sprintf (output_invalid_buf, "'%c'", c);
4979 else
4980 sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
4981 return output_invalid_buf;
4982 }
4983
4984 /* REG_STRING starts *before* REGISTER_PREFIX. */
4985
4986 static const reg_entry *
4987 parse_register (reg_string, end_op)
4988 char *reg_string;
4989 char **end_op;
4990 {
4991 char *s = reg_string;
4992 char *p;
4993 char reg_name_given[MAX_REG_NAME_SIZE + 1];
4994 const reg_entry *r;
4995
4996 /* Skip possible REGISTER_PREFIX and possible whitespace. */
4997 if (*s == REGISTER_PREFIX)
4998 ++s;
4999
5000 if (is_space_char (*s))
5001 ++s;
5002
5003 p = reg_name_given;
5004 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
5005 {
5006 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
5007 return (const reg_entry *) NULL;
5008 s++;
5009 }
5010
5011 /* For naked regs, make sure that we are not dealing with an identifier.
5012 This prevents confusing an identifier like `eax_var' with register
5013 `eax'. */
5014 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
5015 return (const reg_entry *) NULL;
5016
5017 *end_op = s;
5018
5019 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
5020
5021 /* Handle floating point regs, allowing spaces in the (i) part. */
5022 if (r == i386_regtab /* %st is first entry of table */)
5023 {
5024 if (is_space_char (*s))
5025 ++s;
5026 if (*s == '(')
5027 {
5028 ++s;
5029 if (is_space_char (*s))
5030 ++s;
5031 if (*s >= '0' && *s <= '7')
5032 {
5033 r = &i386_float_regtab[*s - '0'];
5034 ++s;
5035 if (is_space_char (*s))
5036 ++s;
5037 if (*s == ')')
5038 {
5039 *end_op = s + 1;
5040 return r;
5041 }
5042 }
5043 /* We have "%st(" then garbage. */
5044 return (const reg_entry *) NULL;
5045 }
5046 }
5047
5048 if (r != NULL
5049 && ((r->reg_flags & (RegRex64 | RegRex)) | (r->reg_type & Reg64)) != 0
5050 && (r->reg_type != Control || !(cpu_arch_flags & CpuSledgehammer))
5051 && flag_code != CODE_64BIT)
5052 return (const reg_entry *) NULL;
5053
5054 return r;
5055 }
5056 \f
5057 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5058 const char *md_shortopts = "kVQ:sqn";
5059 #else
5060 const char *md_shortopts = "qn";
5061 #endif
5062
5063 struct option md_longopts[] = {
5064 #define OPTION_32 (OPTION_MD_BASE + 0)
5065 {"32", no_argument, NULL, OPTION_32},
5066 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5067 #define OPTION_64 (OPTION_MD_BASE + 1)
5068 {"64", no_argument, NULL, OPTION_64},
5069 #endif
5070 {NULL, no_argument, NULL, 0}
5071 };
5072 size_t md_longopts_size = sizeof (md_longopts);
5073
5074 int
5075 md_parse_option (c, arg)
5076 int c;
5077 char *arg ATTRIBUTE_UNUSED;
5078 {
5079 switch (c)
5080 {
5081 case 'n':
5082 optimize_align_code = 0;
5083 break;
5084
5085 case 'q':
5086 quiet_warnings = 1;
5087 break;
5088
5089 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5090 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
5091 should be emitted or not. FIXME: Not implemented. */
5092 case 'Q':
5093 break;
5094
5095 /* -V: SVR4 argument to print version ID. */
5096 case 'V':
5097 print_version_id ();
5098 break;
5099
5100 /* -k: Ignore for FreeBSD compatibility. */
5101 case 'k':
5102 break;
5103
5104 case 's':
5105 /* -s: On i386 Solaris, this tells the native assembler to use
5106 .stab instead of .stab.excl. We always use .stab anyhow. */
5107 break;
5108
5109 case OPTION_64:
5110 {
5111 const char **list, **l;
5112
5113 list = bfd_target_list ();
5114 for (l = list; *l != NULL; l++)
5115 if (strcmp (*l, "elf64-x86-64") == 0)
5116 {
5117 default_arch = "x86_64";
5118 break;
5119 }
5120 if (*l == NULL)
5121 as_fatal (_("No compiled in support for x86_64"));
5122 free (list);
5123 }
5124 break;
5125 #endif
5126
5127 case OPTION_32:
5128 default_arch = "i386";
5129 break;
5130
5131 default:
5132 return 0;
5133 }
5134 return 1;
5135 }
5136
5137 void
5138 md_show_usage (stream)
5139 FILE *stream;
5140 {
5141 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5142 fprintf (stream, _("\
5143 -Q ignored\n\
5144 -V print assembler version number\n\
5145 -k ignored\n\
5146 -n Do not optimize code alignment\n\
5147 -q quieten some warnings\n\
5148 -s ignored\n"));
5149 #else
5150 fprintf (stream, _("\
5151 -n Do not optimize code alignment\n\
5152 -q quieten some warnings\n"));
5153 #endif
5154 }
5155
5156 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
5157 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
5158
5159 /* Pick the target format to use. */
5160
5161 const char *
5162 i386_target_format ()
5163 {
5164 if (!strcmp (default_arch, "x86_64"))
5165 set_code_flag (CODE_64BIT);
5166 else if (!strcmp (default_arch, "i386"))
5167 set_code_flag (CODE_32BIT);
5168 else
5169 as_fatal (_("Unknown architecture"));
5170 switch (OUTPUT_FLAVOR)
5171 {
5172 #ifdef OBJ_MAYBE_AOUT
5173 case bfd_target_aout_flavour:
5174 return AOUT_TARGET_FORMAT;
5175 #endif
5176 #ifdef OBJ_MAYBE_COFF
5177 case bfd_target_coff_flavour:
5178 return "coff-i386";
5179 #endif
5180 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
5181 case bfd_target_elf_flavour:
5182 {
5183 if (flag_code == CODE_64BIT)
5184 use_rela_relocations = 1;
5185 return flag_code == CODE_64BIT ? "elf64-x86-64" : ELF_TARGET_FORMAT;
5186 }
5187 #endif
5188 default:
5189 abort ();
5190 return NULL;
5191 }
5192 }
5193
5194 #endif /* OBJ_MAYBE_ more than one */
5195
5196 #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
5197 void i386_elf_emit_arch_note ()
5198 {
5199 if (OUTPUT_FLAVOR == bfd_target_elf_flavour
5200 && cpu_arch_name != NULL)
5201 {
5202 char *p;
5203 asection *seg = now_seg;
5204 subsegT subseg = now_subseg;
5205 Elf_Internal_Note i_note;
5206 Elf_External_Note e_note;
5207 asection *note_secp;
5208 int len;
5209
5210 /* Create the .note section. */
5211 note_secp = subseg_new (".note", 0);
5212 bfd_set_section_flags (stdoutput,
5213 note_secp,
5214 SEC_HAS_CONTENTS | SEC_READONLY);
5215
5216 /* Process the arch string. */
5217 len = strlen (cpu_arch_name);
5218
5219 i_note.namesz = len + 1;
5220 i_note.descsz = 0;
5221 i_note.type = NT_ARCH;
5222 p = frag_more (sizeof (e_note.namesz));
5223 md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
5224 p = frag_more (sizeof (e_note.descsz));
5225 md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
5226 p = frag_more (sizeof (e_note.type));
5227 md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
5228 p = frag_more (len + 1);
5229 strcpy (p, cpu_arch_name);
5230
5231 frag_align (2, 0, 0);
5232
5233 subseg_set (seg, subseg);
5234 }
5235 }
5236 #endif
5237 \f
5238 symbolS *
5239 md_undefined_symbol (name)
5240 char *name;
5241 {
5242 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
5243 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
5244 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
5245 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
5246 {
5247 if (!GOT_symbol)
5248 {
5249 if (symbol_find (name))
5250 as_bad (_("GOT already in symbol table"));
5251 GOT_symbol = symbol_new (name, undefined_section,
5252 (valueT) 0, &zero_address_frag);
5253 };
5254 return GOT_symbol;
5255 }
5256 return 0;
5257 }
5258
5259 /* Round up a section size to the appropriate boundary. */
5260
5261 valueT
5262 md_section_align (segment, size)
5263 segT segment ATTRIBUTE_UNUSED;
5264 valueT size;
5265 {
5266 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
5267 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
5268 {
5269 /* For a.out, force the section size to be aligned. If we don't do
5270 this, BFD will align it for us, but it will not write out the
5271 final bytes of the section. This may be a bug in BFD, but it is
5272 easier to fix it here since that is how the other a.out targets
5273 work. */
5274 int align;
5275
5276 align = bfd_get_section_alignment (stdoutput, segment);
5277 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
5278 }
5279 #endif
5280
5281 return size;
5282 }
5283
5284 /* On the i386, PC-relative offsets are relative to the start of the
5285 next instruction. That is, the address of the offset, plus its
5286 size, since the offset is always the last part of the insn. */
5287
5288 long
5289 md_pcrel_from (fixP)
5290 fixS *fixP;
5291 {
5292 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
5293 }
5294
5295 #ifndef I386COFF
5296
5297 static void
5298 s_bss (ignore)
5299 int ignore ATTRIBUTE_UNUSED;
5300 {
5301 int temp;
5302
5303 temp = get_absolute_expression ();
5304 subseg_set (bss_section, (subsegT) temp);
5305 demand_empty_rest_of_line ();
5306 }
5307
5308 #endif
5309
5310 void
5311 i386_validate_fix (fixp)
5312 fixS *fixp;
5313 {
5314 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
5315 {
5316 /* GOTOFF relocation are nonsense in 64bit mode. */
5317 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
5318 {
5319 if (flag_code != CODE_64BIT)
5320 abort ();
5321 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
5322 }
5323 else
5324 {
5325 if (flag_code == CODE_64BIT)
5326 abort ();
5327 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
5328 }
5329 fixp->fx_subsy = 0;
5330 }
5331 }
5332
5333 arelent *
5334 tc_gen_reloc (section, fixp)
5335 asection *section ATTRIBUTE_UNUSED;
5336 fixS *fixp;
5337 {
5338 arelent *rel;
5339 bfd_reloc_code_real_type code;
5340
5341 switch (fixp->fx_r_type)
5342 {
5343 case BFD_RELOC_X86_64_PLT32:
5344 case BFD_RELOC_X86_64_GOT32:
5345 case BFD_RELOC_X86_64_GOTPCREL:
5346 case BFD_RELOC_386_PLT32:
5347 case BFD_RELOC_386_GOT32:
5348 case BFD_RELOC_386_GOTOFF:
5349 case BFD_RELOC_386_GOTPC:
5350 case BFD_RELOC_386_TLS_GD:
5351 case BFD_RELOC_386_TLS_LDM:
5352 case BFD_RELOC_386_TLS_LDO_32:
5353 case BFD_RELOC_386_TLS_IE_32:
5354 case BFD_RELOC_386_TLS_IE:
5355 case BFD_RELOC_386_TLS_GOTIE:
5356 case BFD_RELOC_386_TLS_LE_32:
5357 case BFD_RELOC_386_TLS_LE:
5358 case BFD_RELOC_X86_64_32S:
5359 case BFD_RELOC_X86_64_TLSGD:
5360 case BFD_RELOC_X86_64_TLSLD:
5361 case BFD_RELOC_X86_64_DTPOFF32:
5362 case BFD_RELOC_X86_64_GOTTPOFF:
5363 case BFD_RELOC_X86_64_TPOFF32:
5364 case BFD_RELOC_RVA:
5365 case BFD_RELOC_VTABLE_ENTRY:
5366 case BFD_RELOC_VTABLE_INHERIT:
5367 #ifdef TE_PE
5368 case BFD_RELOC_32_SECREL:
5369 #endif
5370 code = fixp->fx_r_type;
5371 break;
5372 default:
5373 if (fixp->fx_pcrel)
5374 {
5375 switch (fixp->fx_size)
5376 {
5377 default:
5378 as_bad_where (fixp->fx_file, fixp->fx_line,
5379 _("can not do %d byte pc-relative relocation"),
5380 fixp->fx_size);
5381 code = BFD_RELOC_32_PCREL;
5382 break;
5383 case 1: code = BFD_RELOC_8_PCREL; break;
5384 case 2: code = BFD_RELOC_16_PCREL; break;
5385 case 4: code = BFD_RELOC_32_PCREL; break;
5386 }
5387 }
5388 else
5389 {
5390 switch (fixp->fx_size)
5391 {
5392 default:
5393 as_bad_where (fixp->fx_file, fixp->fx_line,
5394 _("can not do %d byte relocation"),
5395 fixp->fx_size);
5396 code = BFD_RELOC_32;
5397 break;
5398 case 1: code = BFD_RELOC_8; break;
5399 case 2: code = BFD_RELOC_16; break;
5400 case 4: code = BFD_RELOC_32; break;
5401 #ifdef BFD64
5402 case 8: code = BFD_RELOC_64; break;
5403 #endif
5404 }
5405 }
5406 break;
5407 }
5408
5409 if (code == BFD_RELOC_32
5410 && GOT_symbol
5411 && fixp->fx_addsy == GOT_symbol)
5412 {
5413 /* We don't support GOTPC on 64bit targets. */
5414 if (flag_code == CODE_64BIT)
5415 abort ();
5416 code = BFD_RELOC_386_GOTPC;
5417 }
5418
5419 rel = (arelent *) xmalloc (sizeof (arelent));
5420 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
5421 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
5422
5423 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
5424
5425 if (!use_rela_relocations)
5426 {
5427 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
5428 vtable entry to be used in the relocation's section offset. */
5429 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
5430 rel->address = fixp->fx_offset;
5431
5432 rel->addend = 0;
5433 }
5434 /* Use the rela in 64bit mode. */
5435 else
5436 {
5437 if (!fixp->fx_pcrel)
5438 rel->addend = fixp->fx_offset;
5439 else
5440 switch (code)
5441 {
5442 case BFD_RELOC_X86_64_PLT32:
5443 case BFD_RELOC_X86_64_GOT32:
5444 case BFD_RELOC_X86_64_GOTPCREL:
5445 case BFD_RELOC_X86_64_TLSGD:
5446 case BFD_RELOC_X86_64_TLSLD:
5447 case BFD_RELOC_X86_64_GOTTPOFF:
5448 rel->addend = fixp->fx_offset - fixp->fx_size;
5449 break;
5450 default:
5451 rel->addend = (section->vma
5452 - fixp->fx_size
5453 + fixp->fx_addnumber
5454 + md_pcrel_from (fixp));
5455 break;
5456 }
5457 }
5458
5459 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
5460 if (rel->howto == NULL)
5461 {
5462 as_bad_where (fixp->fx_file, fixp->fx_line,
5463 _("cannot represent relocation type %s"),
5464 bfd_get_reloc_code_name (code));
5465 /* Set howto to a garbage value so that we can keep going. */
5466 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
5467 assert (rel->howto != NULL);
5468 }
5469
5470 return rel;
5471 }
5472
5473 \f
5474 /* Parse operands using Intel syntax. This implements a recursive descent
5475 parser based on the BNF grammar published in Appendix B of the MASM 6.1
5476 Programmer's Guide.
5477
5478 FIXME: We do not recognize the full operand grammar defined in the MASM
5479 documentation. In particular, all the structure/union and
5480 high-level macro operands are missing.
5481
5482 Uppercase words are terminals, lower case words are non-terminals.
5483 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
5484 bars '|' denote choices. Most grammar productions are implemented in
5485 functions called 'intel_<production>'.
5486
5487 Initial production is 'expr'.
5488
5489 addOp + | -
5490
5491 alpha [a-zA-Z]
5492
5493 binOp & | AND | \| | OR | ^ | XOR
5494
5495 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
5496
5497 constant digits [[ radixOverride ]]
5498
5499 dataType BYTE | WORD | DWORD | FWORD | QWORD | TBYTE | OWORD | XMMWORD
5500
5501 digits decdigit
5502 | digits decdigit
5503 | digits hexdigit
5504
5505 decdigit [0-9]
5506
5507 e04 e04 addOp e05
5508 | e05
5509
5510 e05 e05 binOp e06
5511 | e06
5512
5513 e06 e06 mulOp e09
5514 | e09
5515
5516 e09 OFFSET e10
5517 | ~ e10
5518 | NOT e10
5519 | e09 PTR e10
5520 | e09 : e10
5521 | e10
5522
5523 e10 e10 [ expr ]
5524 | e11
5525
5526 e11 ( expr )
5527 | [ expr ]
5528 | constant
5529 | dataType
5530 | id
5531 | $
5532 | register
5533
5534 => expr SHORT e04
5535 | e04
5536
5537 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
5538 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
5539
5540 hexdigit a | b | c | d | e | f
5541 | A | B | C | D | E | F
5542
5543 id alpha
5544 | id alpha
5545 | id decdigit
5546
5547 mulOp * | / | % | MOD | << | SHL | >> | SHR
5548
5549 quote " | '
5550
5551 register specialRegister
5552 | gpRegister
5553 | byteRegister
5554
5555 segmentRegister CS | DS | ES | FS | GS | SS
5556
5557 specialRegister CR0 | CR2 | CR3 | CR4
5558 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
5559 | TR3 | TR4 | TR5 | TR6 | TR7
5560
5561 We simplify the grammar in obvious places (e.g., register parsing is
5562 done by calling parse_register) and eliminate immediate left recursion
5563 to implement a recursive-descent parser.
5564
5565 expr SHORT e04
5566 | e04
5567
5568 e04 e05 e04'
5569
5570 e04' addOp e05 e04'
5571 | Empty
5572
5573 e05 e06 e05'
5574
5575 e05' binOp e06 e05'
5576 | Empty
5577
5578 e06 e09 e06'
5579
5580 e06' mulOp e09 e06'
5581 | Empty
5582
5583 e09 OFFSET e10 e09'
5584 | ~ e10
5585 | NOT e10
5586 | e10 e09'
5587
5588 e09' PTR e10 e09'
5589 | : e10 e09'
5590 | Empty
5591
5592 e10 e11 e10'
5593
5594 e10' [ expr ] e10'
5595 | Empty
5596
5597 e11 ( expr )
5598 | [ expr ]
5599 | BYTE
5600 | WORD
5601 | DWORD
5602 | FWORD
5603 | QWORD
5604 | TBYTE
5605 | OWORD
5606 | XMMWORD
5607 | .
5608 | $
5609 | register
5610 | id
5611 | constant */
5612
5613 /* Parsing structure for the intel syntax parser. Used to implement the
5614 semantic actions for the operand grammar. */
5615 struct intel_parser_s
5616 {
5617 char *op_string; /* The string being parsed. */
5618 int got_a_float; /* Whether the operand is a float. */
5619 int op_modifier; /* Operand modifier. */
5620 int is_mem; /* 1 if operand is memory reference. */
5621 const reg_entry *reg; /* Last register reference found. */
5622 char *disp; /* Displacement string being built. */
5623 };
5624
5625 static struct intel_parser_s intel_parser;
5626
5627 /* Token structure for parsing intel syntax. */
5628 struct intel_token
5629 {
5630 int code; /* Token code. */
5631 const reg_entry *reg; /* Register entry for register tokens. */
5632 char *str; /* String representation. */
5633 };
5634
5635 static struct intel_token cur_token, prev_token;
5636
5637 /* Token codes for the intel parser. Since T_SHORT is already used
5638 by COFF, undefine it first to prevent a warning. */
5639 #define T_NIL -1
5640 #define T_CONST 1
5641 #define T_REG 2
5642 #define T_BYTE 3
5643 #define T_WORD 4
5644 #define T_DWORD 5
5645 #define T_FWORD 6
5646 #define T_QWORD 7
5647 #define T_TBYTE 8
5648 #define T_XMMWORD 9
5649 #undef T_SHORT
5650 #define T_SHORT 10
5651 #define T_OFFSET 11
5652 #define T_PTR 12
5653 #define T_ID 13
5654 #define T_SHL 14
5655 #define T_SHR 15
5656
5657 /* Prototypes for intel parser functions. */
5658 static int intel_match_token PARAMS ((int code));
5659 static void intel_get_token PARAMS ((void));
5660 static void intel_putback_token PARAMS ((void));
5661 static int intel_expr PARAMS ((void));
5662 static int intel_e04 PARAMS ((void));
5663 static int intel_e04_1 PARAMS ((void));
5664 static int intel_e05 PARAMS ((void));
5665 static int intel_e05_1 PARAMS ((void));
5666 static int intel_e06 PARAMS ((void));
5667 static int intel_e06_1 PARAMS ((void));
5668 static int intel_e09 PARAMS ((void));
5669 static int intel_e09_1 PARAMS ((void));
5670 static int intel_e10 PARAMS ((void));
5671 static int intel_e10_1 PARAMS ((void));
5672 static int intel_e11 PARAMS ((void));
5673
5674 static int
5675 i386_intel_operand (operand_string, got_a_float)
5676 char *operand_string;
5677 int got_a_float;
5678 {
5679 int ret;
5680 char *p;
5681
5682 /* Initialize token holders. */
5683 cur_token.code = prev_token.code = T_NIL;
5684 cur_token.reg = prev_token.reg = NULL;
5685 cur_token.str = prev_token.str = NULL;
5686
5687 /* Initialize parser structure. */
5688 p = intel_parser.op_string = (char *) malloc (strlen (operand_string) + 1);
5689 if (p == NULL)
5690 abort ();
5691 strcpy (intel_parser.op_string, operand_string);
5692 intel_parser.got_a_float = got_a_float;
5693 intel_parser.op_modifier = -1;
5694 intel_parser.is_mem = 0;
5695 intel_parser.reg = NULL;
5696 intel_parser.disp = (char *) malloc (strlen (operand_string) + 1);
5697 if (intel_parser.disp == NULL)
5698 abort ();
5699 intel_parser.disp[0] = '\0';
5700
5701 /* Read the first token and start the parser. */
5702 intel_get_token ();
5703 ret = intel_expr ();
5704
5705 if (ret)
5706 {
5707 if (cur_token.code != T_NIL)
5708 {
5709 as_bad (_("invalid operand for '%s' ('%s' unexpected)"),
5710 current_templates->start->name, cur_token.str);
5711 ret = 0;
5712 }
5713 /* If we found a memory reference, hand it over to i386_displacement
5714 to fill in the rest of the operand fields. */
5715 else if (intel_parser.is_mem)
5716 {
5717 if ((i.mem_operands == 1
5718 && (current_templates->start->opcode_modifier & IsString) == 0)
5719 || i.mem_operands == 2)
5720 {
5721 as_bad (_("too many memory references for '%s'"),
5722 current_templates->start->name);
5723 ret = 0;
5724 }
5725 else
5726 {
5727 char *s = intel_parser.disp;
5728 i.mem_operands++;
5729
5730 /* Add the displacement expression. */
5731 if (*s != '\0')
5732 ret = i386_displacement (s, s + strlen (s));
5733 if (ret)
5734 ret = i386_index_check (operand_string);
5735 }
5736 }
5737
5738 /* Constant and OFFSET expressions are handled by i386_immediate. */
5739 else if (intel_parser.op_modifier == T_OFFSET
5740 || intel_parser.reg == NULL)
5741 ret = i386_immediate (intel_parser.disp);
5742 }
5743
5744 free (p);
5745 free (intel_parser.disp);
5746
5747 return ret;
5748 }
5749
5750 /* expr SHORT e04
5751 | e04 */
5752 static int
5753 intel_expr ()
5754 {
5755 /* expr SHORT e04 */
5756 if (cur_token.code == T_SHORT)
5757 {
5758 intel_parser.op_modifier = T_SHORT;
5759 intel_match_token (T_SHORT);
5760
5761 return (intel_e04 ());
5762 }
5763
5764 /* expr e04 */
5765 else
5766 return intel_e04 ();
5767 }
5768
5769 /* e04 e06 e04'
5770
5771 e04' addOp e06 e04'
5772 | Empty */
5773 static int
5774 intel_e04 ()
5775 {
5776 return (intel_e05 () && intel_e04_1 ());
5777 }
5778
5779 static int
5780 intel_e04_1 ()
5781 {
5782 /* e04' addOp e05 e04' */
5783 if (cur_token.code == '+' || cur_token.code == '-')
5784 {
5785 char str[2];
5786
5787 str[0] = cur_token.code;
5788 str[1] = 0;
5789 strcat (intel_parser.disp, str);
5790 intel_match_token (cur_token.code);
5791
5792 return (intel_e05 () && intel_e04_1 ());
5793 }
5794
5795 /* e04' Empty */
5796 else
5797 return 1;
5798 }
5799
5800 /* e05 e06 e05'
5801
5802 e05' binOp e06 e05'
5803 | Empty */
5804 static int
5805 intel_e05 ()
5806 {
5807 return (intel_e06 () && intel_e05_1 ());
5808 }
5809
5810 static int
5811 intel_e05_1 ()
5812 {
5813 /* e05' binOp e06 e05' */
5814 if (cur_token.code == '&' || cur_token.code == '|' || cur_token.code == '^')
5815 {
5816 char str[2];
5817
5818 str[0] = cur_token.code;
5819 str[1] = 0;
5820 strcat (intel_parser.disp, str);
5821 intel_match_token (cur_token.code);
5822
5823 return (intel_e06 () && intel_e05_1 ());
5824 }
5825
5826 /* e05' Empty */
5827 else
5828 return 1;
5829 }
5830
5831 /* e06 e09 e06'
5832
5833 e06' mulOp e09 e06'
5834 | Empty */
5835 static int
5836 intel_e06 ()
5837 {
5838 return (intel_e09 () && intel_e06_1 ());
5839 }
5840
5841 static int
5842 intel_e06_1 ()
5843 {
5844 /* e06' mulOp e09 e06' */
5845 if (cur_token.code == '*' || cur_token.code == '/' || cur_token.code == '%')
5846 {
5847 char str[2];
5848
5849 str[0] = cur_token.code;
5850 str[1] = 0;
5851 strcat (intel_parser.disp, str);
5852 intel_match_token (cur_token.code);
5853
5854 return (intel_e09 () && intel_e06_1 ());
5855 }
5856 else if (cur_token.code == T_SHL)
5857 {
5858 strcat (intel_parser.disp, "<<");
5859 intel_match_token (cur_token.code);
5860
5861 return (intel_e09 () && intel_e06_1 ());
5862 }
5863 else if (cur_token.code == T_SHR)
5864 {
5865 strcat (intel_parser.disp, ">>");
5866 intel_match_token (cur_token.code);
5867
5868 return (intel_e09 () && intel_e06_1 ());
5869 }
5870
5871 /* e06' Empty */
5872 else
5873 return 1;
5874 }
5875
5876 /* e09 OFFSET e10 e09'
5877 | e10 e09'
5878
5879 e09 ~ e10 e09'
5880 | NOT e10 e09'
5881 | e10 e09'
5882
5883 e09' PTR e10 e09'
5884 | : e10 e09'
5885 | Empty */
5886 static int
5887 intel_e09 ()
5888 {
5889 /* e09 OFFSET e10 e09' */
5890 if (cur_token.code == T_OFFSET)
5891 {
5892 intel_parser.is_mem = 0;
5893 intel_parser.op_modifier = T_OFFSET;
5894 intel_match_token (T_OFFSET);
5895
5896 return (intel_e10 () && intel_e09_1 ());
5897 }
5898
5899 /* e09 NOT e10 e09' */
5900 else if (cur_token.code == '~')
5901 {
5902 char str[2];
5903
5904 str[0] = cur_token.code;
5905 str[1] = 0;
5906 strcat (intel_parser.disp, str);
5907 intel_match_token (cur_token.code);
5908
5909 return (intel_e10 () && intel_e09_1 ());
5910 }
5911
5912 /* e09 e10 e09' */
5913 else
5914 return (intel_e10 () && intel_e09_1 ());
5915 }
5916
5917 static int
5918 intel_e09_1 ()
5919 {
5920 /* e09' PTR e10 e09' */
5921 if (cur_token.code == T_PTR)
5922 {
5923 char suffix;
5924
5925 if (prev_token.code == T_BYTE)
5926 suffix = BYTE_MNEM_SUFFIX;
5927
5928 else if (prev_token.code == T_WORD)
5929 {
5930 if (current_templates->start->name[0] == 'l'
5931 && current_templates->start->name[2] == 's'
5932 && current_templates->start->name[3] == 0)
5933 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
5934 else if (intel_parser.got_a_float == 2) /* "fi..." */
5935 suffix = SHORT_MNEM_SUFFIX;
5936 else
5937 suffix = WORD_MNEM_SUFFIX;
5938 }
5939
5940 else if (prev_token.code == T_DWORD)
5941 {
5942 if (current_templates->start->name[0] == 'l'
5943 && current_templates->start->name[2] == 's'
5944 && current_templates->start->name[3] == 0)
5945 suffix = WORD_MNEM_SUFFIX;
5946 else if (flag_code == CODE_16BIT
5947 && (current_templates->start->opcode_modifier
5948 & (Jump|JumpDword|JumpInterSegment)))
5949 suffix = LONG_DOUBLE_MNEM_SUFFIX;
5950 else if (intel_parser.got_a_float == 1) /* "f..." */
5951 suffix = SHORT_MNEM_SUFFIX;
5952 else
5953 suffix = LONG_MNEM_SUFFIX;
5954 }
5955
5956 else if (prev_token.code == T_FWORD)
5957 {
5958 if (current_templates->start->name[0] == 'l'
5959 && current_templates->start->name[2] == 's'
5960 && current_templates->start->name[3] == 0)
5961 suffix = LONG_MNEM_SUFFIX;
5962 else if (!intel_parser.got_a_float)
5963 {
5964 if (flag_code == CODE_16BIT)
5965 add_prefix (DATA_PREFIX_OPCODE);
5966 suffix = LONG_DOUBLE_MNEM_SUFFIX;
5967 }
5968 else
5969 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
5970 }
5971
5972 else if (prev_token.code == T_QWORD)
5973 {
5974 if (intel_parser.got_a_float == 1) /* "f..." */
5975 suffix = LONG_MNEM_SUFFIX;
5976 else
5977 suffix = QWORD_MNEM_SUFFIX;
5978 }
5979
5980 else if (prev_token.code == T_TBYTE)
5981 {
5982 if (intel_parser.got_a_float == 1)
5983 suffix = LONG_DOUBLE_MNEM_SUFFIX;
5984 else
5985 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
5986 }
5987
5988 else if (prev_token.code == T_XMMWORD)
5989 {
5990 /* XXX ignored for now, but accepted since gcc uses it */
5991 suffix = 0;
5992 }
5993
5994 else
5995 {
5996 as_bad (_("Unknown operand modifier `%s'"), prev_token.str);
5997 return 0;
5998 }
5999
6000 if (current_templates->start->base_opcode == 0x8d /* lea */)
6001 ;
6002 else if (!i.suffix)
6003 i.suffix = suffix;
6004 else if (i.suffix != suffix)
6005 {
6006 as_bad (_("Conflicting operand modifiers"));
6007 return 0;
6008 }
6009
6010 intel_match_token (T_PTR);
6011
6012 return (intel_e10 () && intel_e09_1 ());
6013 }
6014
6015 /* e09 : e10 e09' */
6016 else if (cur_token.code == ':')
6017 {
6018 /* Mark as a memory operand only if it's not already known to be an
6019 offset expression. */
6020 if (intel_parser.op_modifier != T_OFFSET)
6021 intel_parser.is_mem = 1;
6022
6023 return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
6024 }
6025
6026 /* e09' Empty */
6027 else
6028 return 1;
6029 }
6030
6031 /* e10 e11 e10'
6032
6033 e10' [ expr ] e10'
6034 | Empty */
6035 static int
6036 intel_e10 ()
6037 {
6038 return (intel_e11 () && intel_e10_1 ());
6039 }
6040
6041 static int
6042 intel_e10_1 ()
6043 {
6044 /* e10' [ expr ] e10' */
6045 if (cur_token.code == '[')
6046 {
6047 intel_match_token ('[');
6048
6049 /* Mark as a memory operand only if it's not already known to be an
6050 offset expression. If it's an offset expression, we need to keep
6051 the brace in. */
6052 if (intel_parser.op_modifier != T_OFFSET)
6053 intel_parser.is_mem = 1;
6054 else
6055 strcat (intel_parser.disp, "[");
6056
6057 /* Add a '+' to the displacement string if necessary. */
6058 if (*intel_parser.disp != '\0'
6059 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
6060 strcat (intel_parser.disp, "+");
6061
6062 if (intel_expr () && intel_match_token (']'))
6063 {
6064 /* Preserve brackets when the operand is an offset expression. */
6065 if (intel_parser.op_modifier == T_OFFSET)
6066 strcat (intel_parser.disp, "]");
6067
6068 return intel_e10_1 ();
6069 }
6070 else
6071 return 0;
6072 }
6073
6074 /* e10' Empty */
6075 else
6076 return 1;
6077 }
6078
6079 /* e11 ( expr )
6080 | [ expr ]
6081 | BYTE
6082 | WORD
6083 | DWORD
6084 | FWORD
6085 | QWORD
6086 | TBYTE
6087 | OWORD
6088 | XMMWORD
6089 | $
6090 | .
6091 | register
6092 | id
6093 | constant */
6094 static int
6095 intel_e11 ()
6096 {
6097 /* e11 ( expr ) */
6098 if (cur_token.code == '(')
6099 {
6100 intel_match_token ('(');
6101 strcat (intel_parser.disp, "(");
6102
6103 if (intel_expr () && intel_match_token (')'))
6104 {
6105 strcat (intel_parser.disp, ")");
6106 return 1;
6107 }
6108 else
6109 return 0;
6110 }
6111
6112 /* e11 [ expr ] */
6113 else if (cur_token.code == '[')
6114 {
6115 intel_match_token ('[');
6116
6117 /* Mark as a memory operand only if it's not already known to be an
6118 offset expression. If it's an offset expression, we need to keep
6119 the brace in. */
6120 if (intel_parser.op_modifier != T_OFFSET)
6121 intel_parser.is_mem = 1;
6122 else
6123 strcat (intel_parser.disp, "[");
6124
6125 /* Operands for jump/call inside brackets denote absolute addresses. */
6126 if (current_templates->start->opcode_modifier
6127 & (Jump|JumpDword|JumpByte|JumpInterSegment))
6128 i.types[this_operand] |= JumpAbsolute;
6129
6130 /* Add a '+' to the displacement string if necessary. */
6131 if (*intel_parser.disp != '\0'
6132 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
6133 strcat (intel_parser.disp, "+");
6134
6135 if (intel_expr () && intel_match_token (']'))
6136 {
6137 /* Preserve brackets when the operand is an offset expression. */
6138 if (intel_parser.op_modifier == T_OFFSET)
6139 strcat (intel_parser.disp, "]");
6140
6141 return 1;
6142 }
6143 else
6144 return 0;
6145 }
6146
6147 /* e11 BYTE
6148 | WORD
6149 | DWORD
6150 | FWORD
6151 | QWORD
6152 | TBYTE
6153 | OWORD
6154 | XMMWORD */
6155 else if (cur_token.code == T_BYTE
6156 || cur_token.code == T_WORD
6157 || cur_token.code == T_DWORD
6158 || cur_token.code == T_FWORD
6159 || cur_token.code == T_QWORD
6160 || cur_token.code == T_TBYTE
6161 || cur_token.code == T_XMMWORD)
6162 {
6163 intel_match_token (cur_token.code);
6164
6165 if (cur_token.code != T_PTR)
6166 {
6167 /* It must have been an identifier; add it to the displacement string. */
6168 strcat (intel_parser.disp, prev_token.str);
6169
6170 /* The identifier represents a memory reference only if it's not
6171 preceded by an offset modifier and if it's not an equate. */
6172 if (intel_parser.op_modifier != T_OFFSET)
6173 {
6174 symbolS *symbolP;
6175
6176 symbolP = symbol_find(prev_token.str);
6177 if (!symbolP || S_GET_SEGMENT(symbolP) != absolute_section)
6178 intel_parser.is_mem = 1;
6179 }
6180 }
6181
6182 return 1;
6183 }
6184
6185 /* e11 $
6186 | . */
6187 else if (cur_token.code == '.')
6188 {
6189 strcat (intel_parser.disp, cur_token.str);
6190 intel_match_token (cur_token.code);
6191
6192 /* Mark as a memory operand only if it's not already known to be an
6193 offset expression. */
6194 if (intel_parser.op_modifier != T_OFFSET)
6195 intel_parser.is_mem = 1;
6196
6197 return 1;
6198 }
6199
6200 /* e11 register */
6201 else if (cur_token.code == T_REG)
6202 {
6203 const reg_entry *reg = intel_parser.reg = cur_token.reg;
6204
6205 intel_match_token (T_REG);
6206
6207 /* Check for segment change. */
6208 if (cur_token.code == ':')
6209 {
6210 if (reg->reg_type & (SReg2 | SReg3))
6211 {
6212 switch (reg->reg_num)
6213 {
6214 case 0:
6215 i.seg[i.mem_operands] = &es;
6216 break;
6217 case 1:
6218 i.seg[i.mem_operands] = &cs;
6219 break;
6220 case 2:
6221 i.seg[i.mem_operands] = &ss;
6222 break;
6223 case 3:
6224 i.seg[i.mem_operands] = &ds;
6225 break;
6226 case 4:
6227 i.seg[i.mem_operands] = &fs;
6228 break;
6229 case 5:
6230 i.seg[i.mem_operands] = &gs;
6231 break;
6232 }
6233 }
6234 else
6235 {
6236 as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
6237 return 0;
6238 }
6239 }
6240
6241 /* Not a segment register. Check for register scaling. */
6242 else if (cur_token.code == '*')
6243 {
6244 if (!intel_parser.is_mem)
6245 {
6246 as_bad (_("Register scaling only allowed in memory operands."));
6247 return 0;
6248 }
6249
6250 /* What follows must be a valid scale. */
6251 if (intel_match_token ('*')
6252 && strchr ("01248", *cur_token.str))
6253 {
6254 i.index_reg = reg;
6255 i.types[this_operand] |= BaseIndex;
6256
6257 /* Set the scale after setting the register (otherwise,
6258 i386_scale will complain) */
6259 i386_scale (cur_token.str);
6260 intel_match_token (T_CONST);
6261 }
6262 else
6263 {
6264 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
6265 cur_token.str);
6266 return 0;
6267 }
6268 }
6269
6270 /* No scaling. If this is a memory operand, the register is either a
6271 base register (first occurrence) or an index register (second
6272 occurrence). */
6273 else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3)))
6274 {
6275 if (i.base_reg && i.index_reg)
6276 {
6277 as_bad (_("Too many register references in memory operand."));
6278 return 0;
6279 }
6280
6281 if (i.base_reg == NULL)
6282 i.base_reg = reg;
6283 else
6284 i.index_reg = reg;
6285
6286 i.types[this_operand] |= BaseIndex;
6287 }
6288
6289 /* Offset modifier. Add the register to the displacement string to be
6290 parsed as an immediate expression after we're done. */
6291 else if (intel_parser.op_modifier == T_OFFSET)
6292 strcat (intel_parser.disp, reg->reg_name);
6293
6294 /* It's neither base nor index nor offset. */
6295 else
6296 {
6297 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
6298 i.op[this_operand].regs = reg;
6299 i.reg_operands++;
6300 }
6301
6302 /* Since registers are not part of the displacement string (except
6303 when we're parsing offset operands), we may need to remove any
6304 preceding '+' from the displacement string. */
6305 if (*intel_parser.disp != '\0'
6306 && intel_parser.op_modifier != T_OFFSET)
6307 {
6308 char *s = intel_parser.disp;
6309 s += strlen (s) - 1;
6310 if (*s == '+')
6311 *s = '\0';
6312 }
6313
6314 return 1;
6315 }
6316
6317 /* e11 id */
6318 else if (cur_token.code == T_ID)
6319 {
6320 /* Add the identifier to the displacement string. */
6321 strcat (intel_parser.disp, cur_token.str);
6322
6323 /* The identifier represents a memory reference only if it's not
6324 preceded by an offset modifier and if it's not an equate. */
6325 if (intel_parser.op_modifier != T_OFFSET)
6326 {
6327 symbolS *symbolP;
6328
6329 symbolP = symbol_find(cur_token.str);
6330 if (!symbolP || S_GET_SEGMENT(symbolP) != absolute_section)
6331 intel_parser.is_mem = 1;
6332 }
6333
6334 intel_match_token (T_ID);
6335 return 1;
6336 }
6337
6338 /* e11 constant */
6339 else if (cur_token.code == T_CONST
6340 || cur_token.code == '-'
6341 || cur_token.code == '+')
6342 {
6343 char *save_str;
6344
6345 /* Allow constants that start with `+' or `-'. */
6346 if (cur_token.code == '-' || cur_token.code == '+')
6347 {
6348 strcat (intel_parser.disp, cur_token.str);
6349 intel_match_token (cur_token.code);
6350 if (cur_token.code != T_CONST)
6351 {
6352 as_bad (_("Syntax error. Expecting a constant. Got `%s'."),
6353 cur_token.str);
6354 return 0;
6355 }
6356 }
6357
6358 save_str = (char *) malloc (strlen (cur_token.str) + 1);
6359 if (save_str == NULL)
6360 abort ();
6361 strcpy (save_str, cur_token.str);
6362
6363 /* Get the next token to check for register scaling. */
6364 intel_match_token (cur_token.code);
6365
6366 /* Check if this constant is a scaling factor for an index register. */
6367 if (cur_token.code == '*')
6368 {
6369 if (intel_match_token ('*') && cur_token.code == T_REG)
6370 {
6371 if (!intel_parser.is_mem)
6372 {
6373 as_bad (_("Register scaling only allowed in memory operands."));
6374 return 0;
6375 }
6376
6377 /* The constant is followed by `* reg', so it must be
6378 a valid scale. */
6379 if (strchr ("01248", *save_str))
6380 {
6381 i.index_reg = cur_token.reg;
6382 i.types[this_operand] |= BaseIndex;
6383
6384 /* Set the scale after setting the register (otherwise,
6385 i386_scale will complain) */
6386 i386_scale (save_str);
6387 intel_match_token (T_REG);
6388
6389 /* Since registers are not part of the displacement
6390 string, we may need to remove any preceding '+' from
6391 the displacement string. */
6392 if (*intel_parser.disp != '\0')
6393 {
6394 char *s = intel_parser.disp;
6395 s += strlen (s) - 1;
6396 if (*s == '+')
6397 *s = '\0';
6398 }
6399
6400 free (save_str);
6401
6402 return 1;
6403 }
6404 else
6405 return 0;
6406 }
6407
6408 /* The constant was not used for register scaling. Since we have
6409 already consumed the token following `*' we now need to put it
6410 back in the stream. */
6411 else
6412 intel_putback_token ();
6413 }
6414
6415 /* Add the constant to the displacement string. */
6416 strcat (intel_parser.disp, save_str);
6417 free (save_str);
6418
6419 return 1;
6420 }
6421
6422 as_bad (_("Unrecognized token '%s'"), cur_token.str);
6423 return 0;
6424 }
6425
6426 /* Match the given token against cur_token. If they match, read the next
6427 token from the operand string. */
6428 static int
6429 intel_match_token (code)
6430 int code;
6431 {
6432 if (cur_token.code == code)
6433 {
6434 intel_get_token ();
6435 return 1;
6436 }
6437 else
6438 {
6439 as_bad (_("Unexpected token `%s'"), cur_token.str);
6440 return 0;
6441 }
6442 }
6443
6444 /* Read a new token from intel_parser.op_string and store it in cur_token. */
6445 static void
6446 intel_get_token ()
6447 {
6448 char *end_op;
6449 const reg_entry *reg;
6450 struct intel_token new_token;
6451
6452 new_token.code = T_NIL;
6453 new_token.reg = NULL;
6454 new_token.str = NULL;
6455
6456 /* Free the memory allocated to the previous token and move
6457 cur_token to prev_token. */
6458 if (prev_token.str)
6459 free (prev_token.str);
6460
6461 prev_token = cur_token;
6462
6463 /* Skip whitespace. */
6464 while (is_space_char (*intel_parser.op_string))
6465 intel_parser.op_string++;
6466
6467 /* Return an empty token if we find nothing else on the line. */
6468 if (*intel_parser.op_string == '\0')
6469 {
6470 cur_token = new_token;
6471 return;
6472 }
6473
6474 /* The new token cannot be larger than the remainder of the operand
6475 string. */
6476 new_token.str = (char *) malloc (strlen (intel_parser.op_string) + 1);
6477 if (new_token.str == NULL)
6478 abort ();
6479 new_token.str[0] = '\0';
6480
6481 if (strchr ("0123456789", *intel_parser.op_string))
6482 {
6483 char *p = new_token.str;
6484 char *q = intel_parser.op_string;
6485 new_token.code = T_CONST;
6486
6487 /* Allow any kind of identifier char to encompass floating point and
6488 hexadecimal numbers. */
6489 while (is_identifier_char (*q))
6490 *p++ = *q++;
6491 *p = '\0';
6492
6493 /* Recognize special symbol names [0-9][bf]. */
6494 if (strlen (intel_parser.op_string) == 2
6495 && (intel_parser.op_string[1] == 'b'
6496 || intel_parser.op_string[1] == 'f'))
6497 new_token.code = T_ID;
6498 }
6499
6500 else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
6501 && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
6502 {
6503 new_token.code = T_REG;
6504 new_token.reg = reg;
6505
6506 if (*intel_parser.op_string == REGISTER_PREFIX)
6507 {
6508 new_token.str[0] = REGISTER_PREFIX;
6509 new_token.str[1] = '\0';
6510 }
6511
6512 strcat (new_token.str, reg->reg_name);
6513 }
6514
6515 else if (is_identifier_char (*intel_parser.op_string))
6516 {
6517 char *p = new_token.str;
6518 char *q = intel_parser.op_string;
6519
6520 /* A '.' or '$' followed by an identifier char is an identifier.
6521 Otherwise, it's operator '.' followed by an expression. */
6522 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
6523 {
6524 new_token.code = '.';
6525 new_token.str[0] = '.';
6526 new_token.str[1] = '\0';
6527 }
6528 else
6529 {
6530 while (is_identifier_char (*q) || *q == '@')
6531 *p++ = *q++;
6532 *p = '\0';
6533
6534 if (strcasecmp (new_token.str, "NOT") == 0)
6535 new_token.code = '~';
6536
6537 else if (strcasecmp (new_token.str, "MOD") == 0)
6538 new_token.code = '%';
6539
6540 else if (strcasecmp (new_token.str, "AND") == 0)
6541 new_token.code = '&';
6542
6543 else if (strcasecmp (new_token.str, "OR") == 0)
6544 new_token.code = '|';
6545
6546 else if (strcasecmp (new_token.str, "XOR") == 0)
6547 new_token.code = '^';
6548
6549 else if (strcasecmp (new_token.str, "SHL") == 0)
6550 new_token.code = T_SHL;
6551
6552 else if (strcasecmp (new_token.str, "SHR") == 0)
6553 new_token.code = T_SHR;
6554
6555 else if (strcasecmp (new_token.str, "BYTE") == 0)
6556 new_token.code = T_BYTE;
6557
6558 else if (strcasecmp (new_token.str, "WORD") == 0)
6559 new_token.code = T_WORD;
6560
6561 else if (strcasecmp (new_token.str, "DWORD") == 0)
6562 new_token.code = T_DWORD;
6563
6564 else if (strcasecmp (new_token.str, "FWORD") == 0)
6565 new_token.code = T_FWORD;
6566
6567 else if (strcasecmp (new_token.str, "QWORD") == 0)
6568 new_token.code = T_QWORD;
6569
6570 else if (strcasecmp (new_token.str, "TBYTE") == 0
6571 /* XXX remove (gcc still uses it) */
6572 || strcasecmp (new_token.str, "XWORD") == 0)
6573 new_token.code = T_TBYTE;
6574
6575 else if (strcasecmp (new_token.str, "XMMWORD") == 0
6576 || strcasecmp (new_token.str, "OWORD") == 0)
6577 new_token.code = T_XMMWORD;
6578
6579 else if (strcasecmp (new_token.str, "PTR") == 0)
6580 new_token.code = T_PTR;
6581
6582 else if (strcasecmp (new_token.str, "SHORT") == 0)
6583 new_token.code = T_SHORT;
6584
6585 else if (strcasecmp (new_token.str, "OFFSET") == 0)
6586 {
6587 new_token.code = T_OFFSET;
6588
6589 /* ??? This is not mentioned in the MASM grammar but gcc
6590 makes use of it with -mintel-syntax. OFFSET may be
6591 followed by FLAT: */
6592 if (strncasecmp (q, " FLAT:", 6) == 0)
6593 strcat (new_token.str, " FLAT:");
6594 }
6595
6596 /* ??? This is not mentioned in the MASM grammar. */
6597 else if (strcasecmp (new_token.str, "FLAT") == 0)
6598 new_token.code = T_OFFSET;
6599
6600 else
6601 new_token.code = T_ID;
6602 }
6603 }
6604
6605 else if (strchr ("+-/*%|&^:[]()~", *intel_parser.op_string))
6606 {
6607 new_token.code = *intel_parser.op_string;
6608 new_token.str[0] = *intel_parser.op_string;
6609 new_token.str[1] = '\0';
6610 }
6611
6612 else if (strchr ("<>", *intel_parser.op_string)
6613 && *intel_parser.op_string == *(intel_parser.op_string + 1))
6614 {
6615 new_token.code = *intel_parser.op_string == '<' ? T_SHL : T_SHR;
6616 new_token.str[0] = *intel_parser.op_string;
6617 new_token.str[1] = *intel_parser.op_string;
6618 new_token.str[2] = '\0';
6619 }
6620
6621 else
6622 as_bad (_("Unrecognized token `%s'"), intel_parser.op_string);
6623
6624 intel_parser.op_string += strlen (new_token.str);
6625 cur_token = new_token;
6626 }
6627
6628 /* Put cur_token back into the token stream and make cur_token point to
6629 prev_token. */
6630 static void
6631 intel_putback_token ()
6632 {
6633 intel_parser.op_string -= strlen (cur_token.str);
6634 free (cur_token.str);
6635 cur_token = prev_token;
6636
6637 /* Forget prev_token. */
6638 prev_token.code = T_NIL;
6639 prev_token.reg = NULL;
6640 prev_token.str = NULL;
6641 }
6642
6643 int
6644 tc_x86_regname_to_dw2regnum (const char *regname)
6645 {
6646 unsigned int regnum;
6647 unsigned int regnames_count;
6648 char *regnames_32[] =
6649 {
6650 "eax", "ecx", "edx", "ebx",
6651 "esp", "ebp", "esi", "edi",
6652 "eip"
6653 };
6654 char *regnames_64[] =
6655 {
6656 "rax", "rbx", "rcx", "rdx",
6657 "rdi", "rsi", "rbp", "rsp",
6658 "r8", "r9", "r10", "r11",
6659 "r12", "r13", "r14", "r15",
6660 "rip"
6661 };
6662 char **regnames;
6663
6664 if (flag_code == CODE_64BIT)
6665 {
6666 regnames = regnames_64;
6667 regnames_count = ARRAY_SIZE (regnames_64);
6668 }
6669 else
6670 {
6671 regnames = regnames_32;
6672 regnames_count = ARRAY_SIZE (regnames_32);
6673 }
6674
6675 for (regnum = 0; regnum < regnames_count; regnum++)
6676 if (strcmp (regname, regnames[regnum]) == 0)
6677 return regnum;
6678
6679 return -1;
6680 }
6681
6682 void
6683 tc_x86_frame_initial_instructions (void)
6684 {
6685 static unsigned int sp_regno;
6686
6687 if (!sp_regno)
6688 sp_regno = tc_x86_regname_to_dw2regnum (flag_code == CODE_64BIT
6689 ? "rsp" : "esp");
6690
6691 cfi_add_CFA_def_cfa (sp_regno, -x86_cie_data_alignment);
6692 cfi_add_CFA_offset (x86_dwarf2_return_column, x86_cie_data_alignment);
6693 }
6694
6695 int
6696 i386_elf_section_type (const char *str, size_t len)
6697 {
6698 if (flag_code == CODE_64BIT
6699 && len == sizeof ("unwind") - 1
6700 && strncmp (str, "unwind", 6) == 0)
6701 return SHT_X86_64_UNWIND;
6702
6703 return -1;
6704 }
6705
6706 #ifdef TE_PE
6707 void
6708 tc_pe_dwarf2_emit_offset (symbolS *symbol, unsigned int size)
6709 {
6710 expressionS expr;
6711
6712 expr.X_op = O_secrel;
6713 expr.X_add_symbol = symbol;
6714 expr.X_add_number = 0;
6715 emit_expr (&expr, size);
6716 }
6717 #endif
This page took 0.17346 seconds and 4 git commands to generate.