include/opcode/
[deliverable/binutils-gdb.git] / gas / config / tc-i386.c
1 /* tc-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, 2006, 2007
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, 51 Franklin Street - Fifth Floor, Boston, MA
21 02110-1301, 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 "elf/x86-64.h"
36
37 #ifndef REGISTER_WARNINGS
38 #define REGISTER_WARNINGS 1
39 #endif
40
41 #ifndef INFER_ADDR_PREFIX
42 #define INFER_ADDR_PREFIX 1
43 #endif
44
45 #ifndef SCALE1_WHEN_NO_INDEX
46 /* Specifying a scale factor besides 1 when there is no index is
47 futile. eg. `mov (%ebx,2),%al' does exactly the same as
48 `mov (%ebx),%al'. To slavishly follow what the programmer
49 specified, set SCALE1_WHEN_NO_INDEX to 0. */
50 #define SCALE1_WHEN_NO_INDEX 1
51 #endif
52
53 #ifndef DEFAULT_ARCH
54 #define DEFAULT_ARCH "i386"
55 #endif
56
57 #ifndef INLINE
58 #if __GNUC__ >= 2
59 #define INLINE __inline__
60 #else
61 #define INLINE
62 #endif
63 #endif
64
65 static void set_code_flag (int);
66 static void set_16bit_gcc_code_flag (int);
67 static void set_intel_syntax (int);
68 static void set_cpu_arch (int);
69 #ifdef TE_PE
70 static void pe_directive_secrel (int);
71 #endif
72 static void signed_cons (int);
73 static char *output_invalid (int c);
74 static int i386_operand (char *);
75 static int i386_intel_operand (char *, int);
76 static const reg_entry *parse_register (char *, char **);
77 static char *parse_insn (char *, char *);
78 static char *parse_operands (char *, const char *);
79 static void swap_operands (void);
80 static void swap_2_operands (int, int);
81 static void optimize_imm (void);
82 static void optimize_disp (void);
83 static int match_template (void);
84 static int check_string (void);
85 static int process_suffix (void);
86 static int check_byte_reg (void);
87 static int check_long_reg (void);
88 static int check_qword_reg (void);
89 static int check_word_reg (void);
90 static int finalize_imm (void);
91 static int process_operands (void);
92 static const seg_entry *build_modrm_byte (void);
93 static void output_insn (void);
94 static void output_imm (fragS *, offsetT);
95 static void output_disp (fragS *, offsetT);
96 #ifndef I386COFF
97 static void s_bss (int);
98 #endif
99 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
100 static void handle_large_common (int small ATTRIBUTE_UNUSED);
101 #endif
102
103 static const char *default_arch = DEFAULT_ARCH;
104
105 /* 'md_assemble ()' gathers together information and puts it into a
106 i386_insn. */
107
108 union i386_op
109 {
110 expressionS *disps;
111 expressionS *imms;
112 const reg_entry *regs;
113 };
114
115 struct _i386_insn
116 {
117 /* TM holds the template for the insn were currently assembling. */
118 template tm;
119
120 /* SUFFIX holds the instruction mnemonic suffix if given.
121 (e.g. 'l' for 'movl') */
122 char suffix;
123
124 /* OPERANDS gives the number of given operands. */
125 unsigned int operands;
126
127 /* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number
128 of given register, displacement, memory operands and immediate
129 operands. */
130 unsigned int reg_operands, disp_operands, mem_operands, imm_operands;
131
132 /* TYPES [i] is the type (see above #defines) which tells us how to
133 use OP[i] for the corresponding operand. */
134 unsigned int types[MAX_OPERANDS];
135
136 /* Displacement expression, immediate expression, or register for each
137 operand. */
138 union i386_op op[MAX_OPERANDS];
139
140 /* Flags for operands. */
141 unsigned int flags[MAX_OPERANDS];
142 #define Operand_PCrel 1
143
144 /* Relocation type for operand */
145 enum bfd_reloc_code_real reloc[MAX_OPERANDS];
146
147 /* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode
148 the base index byte below. */
149 const reg_entry *base_reg;
150 const reg_entry *index_reg;
151 unsigned int log2_scale_factor;
152
153 /* SEG gives the seg_entries of this insn. They are zero unless
154 explicit segment overrides are given. */
155 const seg_entry *seg[2];
156
157 /* PREFIX holds all the given prefix opcodes (usually null).
158 PREFIXES is the number of prefix opcodes. */
159 unsigned int prefixes;
160 unsigned char prefix[MAX_PREFIXES];
161
162 /* RM and SIB are the modrm byte and the sib byte where the
163 addressing modes of this insn are encoded. */
164
165 modrm_byte rm;
166 rex_byte rex;
167 sib_byte sib;
168 };
169
170 typedef struct _i386_insn i386_insn;
171
172 /* List of chars besides those in app.c:symbol_chars that can start an
173 operand. Used to prevent the scrubber eating vital white-space. */
174 const char extra_symbol_chars[] = "*%-(["
175 #ifdef LEX_AT
176 "@"
177 #endif
178 #ifdef LEX_QM
179 "?"
180 #endif
181 ;
182
183 #if (defined (TE_I386AIX) \
184 || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) \
185 && !defined (TE_GNU) \
186 && !defined (TE_LINUX) \
187 && !defined (TE_NETWARE) \
188 && !defined (TE_FreeBSD) \
189 && !defined (TE_NetBSD)))
190 /* This array holds the chars that always start a comment. If the
191 pre-processor is disabled, these aren't very useful. The option
192 --divide will remove '/' from this list. */
193 const char *i386_comment_chars = "#/";
194 #define SVR4_COMMENT_CHARS 1
195 #define PREFIX_SEPARATOR '\\'
196
197 #else
198 const char *i386_comment_chars = "#";
199 #define PREFIX_SEPARATOR '/'
200 #endif
201
202 /* This array holds the chars that only start a comment at the beginning of
203 a line. If the line seems to have the form '# 123 filename'
204 .line and .file directives will appear in the pre-processed output.
205 Note that input_file.c hand checks for '#' at the beginning of the
206 first line of the input file. This is because the compiler outputs
207 #NO_APP at the beginning of its output.
208 Also note that comments started like this one will always work if
209 '/' isn't otherwise defined. */
210 const char line_comment_chars[] = "#/";
211
212 const char line_separator_chars[] = ";";
213
214 /* Chars that can be used to separate mant from exp in floating point
215 nums. */
216 const char EXP_CHARS[] = "eE";
217
218 /* Chars that mean this number is a floating point constant
219 As in 0f12.456
220 or 0d1.2345e12. */
221 const char FLT_CHARS[] = "fFdDxX";
222
223 /* Tables for lexical analysis. */
224 static char mnemonic_chars[256];
225 static char register_chars[256];
226 static char operand_chars[256];
227 static char identifier_chars[256];
228 static char digit_chars[256];
229
230 /* Lexical macros. */
231 #define is_mnemonic_char(x) (mnemonic_chars[(unsigned char) x])
232 #define is_operand_char(x) (operand_chars[(unsigned char) x])
233 #define is_register_char(x) (register_chars[(unsigned char) x])
234 #define is_space_char(x) ((x) == ' ')
235 #define is_identifier_char(x) (identifier_chars[(unsigned char) x])
236 #define is_digit_char(x) (digit_chars[(unsigned char) x])
237
238 /* All non-digit non-letter characters that may occur in an operand. */
239 static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
240
241 /* md_assemble() always leaves the strings it's passed unaltered. To
242 effect this we maintain a stack of saved characters that we've smashed
243 with '\0's (indicating end of strings for various sub-fields of the
244 assembler instruction). */
245 static char save_stack[32];
246 static char *save_stack_p;
247 #define END_STRING_AND_SAVE(s) \
248 do { *save_stack_p++ = *(s); *(s) = '\0'; } while (0)
249 #define RESTORE_END_STRING(s) \
250 do { *(s) = *--save_stack_p; } while (0)
251
252 /* The instruction we're assembling. */
253 static i386_insn i;
254
255 /* Possible templates for current insn. */
256 static const templates *current_templates;
257
258 /* Per instruction expressionS buffers: max displacements & immediates. */
259 static expressionS disp_expressions[MAX_MEMORY_OPERANDS];
260 static expressionS im_expressions[MAX_IMMEDIATE_OPERANDS];
261
262 /* Current operand we are working on. */
263 static int this_operand;
264
265 /* We support four different modes. FLAG_CODE variable is used to distinguish
266 these. */
267
268 enum flag_code {
269 CODE_32BIT,
270 CODE_16BIT,
271 CODE_64BIT };
272 #define NUM_FLAG_CODE ((int) CODE_64BIT + 1)
273
274 static enum flag_code flag_code;
275 static unsigned int object_64bit;
276 static int use_rela_relocations = 0;
277
278 /* The names used to print error messages. */
279 static const char *flag_code_names[] =
280 {
281 "32",
282 "16",
283 "64"
284 };
285
286 /* 1 for intel syntax,
287 0 if att syntax. */
288 static int intel_syntax = 0;
289
290 /* 1 if register prefix % not required. */
291 static int allow_naked_reg = 0;
292
293 /* Register prefix used for error message. */
294 static const char *register_prefix = "%";
295
296 /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
297 leave, push, and pop instructions so that gcc has the same stack
298 frame as in 32 bit mode. */
299 static char stackop_size = '\0';
300
301 /* Non-zero to optimize code alignment. */
302 int optimize_align_code = 1;
303
304 /* Non-zero to quieten some warnings. */
305 static int quiet_warnings = 0;
306
307 /* CPU name. */
308 static const char *cpu_arch_name = NULL;
309 static const char *cpu_sub_arch_name = NULL;
310
311 /* CPU feature flags. */
312 static unsigned int cpu_arch_flags = CpuUnknownFlags | CpuNo64;
313
314 /* If we have selected a cpu we are generating instructions for. */
315 static int cpu_arch_tune_set = 0;
316
317 /* Cpu we are generating instructions for. */
318 static enum processor_type cpu_arch_tune = PROCESSOR_UNKNOWN;
319
320 /* CPU feature flags of cpu we are generating instructions for. */
321 static unsigned int cpu_arch_tune_flags = 0;
322
323 /* CPU instruction set architecture used. */
324 static enum processor_type cpu_arch_isa = PROCESSOR_UNKNOWN;
325
326 /* CPU feature flags of instruction set architecture used. */
327 static unsigned int cpu_arch_isa_flags = 0;
328
329 /* If set, conditional jumps are not automatically promoted to handle
330 larger than a byte offset. */
331 static unsigned int no_cond_jump_promotion = 0;
332
333 /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
334 static symbolS *GOT_symbol;
335
336 /* The dwarf2 return column, adjusted for 32 or 64 bit. */
337 unsigned int x86_dwarf2_return_column;
338
339 /* The dwarf2 data alignment, adjusted for 32 or 64 bit. */
340 int x86_cie_data_alignment;
341
342 /* Interface to relax_segment.
343 There are 3 major relax states for 386 jump insns because the
344 different types of jumps add different sizes to frags when we're
345 figuring out what sort of jump to choose to reach a given label. */
346
347 /* Types. */
348 #define UNCOND_JUMP 0
349 #define COND_JUMP 1
350 #define COND_JUMP86 2
351
352 /* Sizes. */
353 #define CODE16 1
354 #define SMALL 0
355 #define SMALL16 (SMALL | CODE16)
356 #define BIG 2
357 #define BIG16 (BIG | CODE16)
358
359 #ifndef INLINE
360 #ifdef __GNUC__
361 #define INLINE __inline__
362 #else
363 #define INLINE
364 #endif
365 #endif
366
367 #define ENCODE_RELAX_STATE(type, size) \
368 ((relax_substateT) (((type) << 2) | (size)))
369 #define TYPE_FROM_RELAX_STATE(s) \
370 ((s) >> 2)
371 #define DISP_SIZE_FROM_RELAX_STATE(s) \
372 ((((s) & 3) == BIG ? 4 : (((s) & 3) == BIG16 ? 2 : 1)))
373
374 /* This table is used by relax_frag to promote short jumps to long
375 ones where necessary. SMALL (short) jumps may be promoted to BIG
376 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
377 don't allow a short jump in a 32 bit code segment to be promoted to
378 a 16 bit offset jump because it's slower (requires data size
379 prefix), and doesn't work, unless the destination is in the bottom
380 64k of the code segment (The top 16 bits of eip are zeroed). */
381
382 const relax_typeS md_relax_table[] =
383 {
384 /* The fields are:
385 1) most positive reach of this state,
386 2) most negative reach of this state,
387 3) how many bytes this mode will have in the variable part of the frag
388 4) which index into the table to try if we can't fit into this one. */
389
390 /* UNCOND_JUMP states. */
391 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
392 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
393 /* dword jmp adds 4 bytes to frag:
394 0 extra opcode bytes, 4 displacement bytes. */
395 {0, 0, 4, 0},
396 /* word jmp adds 2 byte2 to frag:
397 0 extra opcode bytes, 2 displacement bytes. */
398 {0, 0, 2, 0},
399
400 /* COND_JUMP states. */
401 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
402 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
403 /* dword conditionals adds 5 bytes to frag:
404 1 extra opcode byte, 4 displacement bytes. */
405 {0, 0, 5, 0},
406 /* word conditionals add 3 bytes to frag:
407 1 extra opcode byte, 2 displacement bytes. */
408 {0, 0, 3, 0},
409
410 /* COND_JUMP86 states. */
411 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG)},
412 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG16)},
413 /* dword conditionals adds 5 bytes to frag:
414 1 extra opcode byte, 4 displacement bytes. */
415 {0, 0, 5, 0},
416 /* word conditionals add 4 bytes to frag:
417 1 displacement byte and a 3 byte long branch insn. */
418 {0, 0, 4, 0}
419 };
420
421 static const arch_entry cpu_arch[] =
422 {
423 {"generic32", PROCESSOR_GENERIC32,
424 Cpu186|Cpu286|Cpu386},
425 {"generic64", PROCESSOR_GENERIC64,
426 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
427 |CpuMMX2|CpuSSE|CpuSSE2},
428 {"i8086", PROCESSOR_UNKNOWN,
429 0},
430 {"i186", PROCESSOR_UNKNOWN,
431 Cpu186},
432 {"i286", PROCESSOR_UNKNOWN,
433 Cpu186|Cpu286},
434 {"i386", PROCESSOR_GENERIC32,
435 Cpu186|Cpu286|Cpu386},
436 {"i486", PROCESSOR_I486,
437 Cpu186|Cpu286|Cpu386|Cpu486},
438 {"i586", PROCESSOR_PENTIUM,
439 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586},
440 {"i686", PROCESSOR_PENTIUMPRO,
441 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686},
442 {"pentium", PROCESSOR_PENTIUM,
443 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586},
444 {"pentiumpro",PROCESSOR_PENTIUMPRO,
445 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686},
446 {"pentiumii", PROCESSOR_PENTIUMPRO,
447 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX},
448 {"pentiumiii",PROCESSOR_PENTIUMPRO,
449 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuMMX2|CpuSSE},
450 {"pentium4", PROCESSOR_PENTIUM4,
451 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
452 |CpuMMX2|CpuSSE|CpuSSE2},
453 {"prescott", PROCESSOR_NOCONA,
454 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
455 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
456 {"nocona", PROCESSOR_NOCONA,
457 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
458 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
459 {"yonah", PROCESSOR_CORE,
460 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
461 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
462 {"core", PROCESSOR_CORE,
463 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
464 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
465 {"merom", PROCESSOR_CORE2,
466 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
467 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3},
468 {"core2", PROCESSOR_CORE2,
469 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
470 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3},
471 {"k6", PROCESSOR_K6,
472 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX},
473 {"k6_2", PROCESSOR_K6,
474 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow},
475 {"athlon", PROCESSOR_ATHLON,
476 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
477 |CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA},
478 {"sledgehammer", PROCESSOR_K8,
479 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
480 |CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2},
481 {"opteron", PROCESSOR_K8,
482 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
483 |CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2},
484 {"k8", PROCESSOR_K8,
485 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
486 |CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2},
487 {"amdfam10", PROCESSOR_AMDFAM10,
488 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuSledgehammer
489 |CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2|CpuSSE3|CpuSSE4a
490 |CpuABM},
491 {".mmx", PROCESSOR_UNKNOWN,
492 CpuMMX},
493 {".sse", PROCESSOR_UNKNOWN,
494 CpuMMX|CpuMMX2|CpuSSE},
495 {".sse2", PROCESSOR_UNKNOWN,
496 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2},
497 {".sse3", PROCESSOR_UNKNOWN,
498 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
499 {".ssse3", PROCESSOR_UNKNOWN,
500 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3},
501 {".sse4.1", PROCESSOR_UNKNOWN,
502 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3|CpuSSE4_1},
503 {".sse4.2", PROCESSOR_UNKNOWN,
504 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3|CpuSSE4},
505 {".sse4", PROCESSOR_UNKNOWN,
506 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3|CpuSSE4},
507 {".3dnow", PROCESSOR_UNKNOWN,
508 CpuMMX|Cpu3dnow},
509 {".3dnowa", PROCESSOR_UNKNOWN,
510 CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA},
511 {".padlock", PROCESSOR_UNKNOWN,
512 CpuPadLock},
513 {".pacifica", PROCESSOR_UNKNOWN,
514 CpuSVME},
515 {".svme", PROCESSOR_UNKNOWN,
516 CpuSVME},
517 {".sse4a", PROCESSOR_UNKNOWN,
518 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSE4a},
519 {".abm", PROCESSOR_UNKNOWN,
520 CpuABM}
521 };
522
523 const pseudo_typeS md_pseudo_table[] =
524 {
525 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
526 {"align", s_align_bytes, 0},
527 #else
528 {"align", s_align_ptwo, 0},
529 #endif
530 {"arch", set_cpu_arch, 0},
531 #ifndef I386COFF
532 {"bss", s_bss, 0},
533 #endif
534 {"ffloat", float_cons, 'f'},
535 {"dfloat", float_cons, 'd'},
536 {"tfloat", float_cons, 'x'},
537 {"value", cons, 2},
538 {"slong", signed_cons, 4},
539 {"noopt", s_ignore, 0},
540 {"optim", s_ignore, 0},
541 {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
542 {"code16", set_code_flag, CODE_16BIT},
543 {"code32", set_code_flag, CODE_32BIT},
544 {"code64", set_code_flag, CODE_64BIT},
545 {"intel_syntax", set_intel_syntax, 1},
546 {"att_syntax", set_intel_syntax, 0},
547 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
548 {"largecomm", handle_large_common, 0},
549 #else
550 {"file", (void (*) (int)) dwarf2_directive_file, 0},
551 {"loc", dwarf2_directive_loc, 0},
552 {"loc_mark_labels", dwarf2_directive_loc_mark_labels, 0},
553 #endif
554 #ifdef TE_PE
555 {"secrel32", pe_directive_secrel, 0},
556 #endif
557 {0, 0, 0}
558 };
559
560 /* For interface with expression (). */
561 extern char *input_line_pointer;
562
563 /* Hash table for instruction mnemonic lookup. */
564 static struct hash_control *op_hash;
565
566 /* Hash table for register lookup. */
567 static struct hash_control *reg_hash;
568 \f
569 void
570 i386_align_code (fragS *fragP, int count)
571 {
572 /* Various efficient no-op patterns for aligning code labels.
573 Note: Don't try to assemble the instructions in the comments.
574 0L and 0w are not legal. */
575 static const char f32_1[] =
576 {0x90}; /* nop */
577 static const char f32_2[] =
578 {0x66,0x90}; /* xchg %ax,%ax */
579 static const char f32_3[] =
580 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
581 static const char f32_4[] =
582 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
583 static const char f32_5[] =
584 {0x90, /* nop */
585 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
586 static const char f32_6[] =
587 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
588 static const char f32_7[] =
589 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
590 static const char f32_8[] =
591 {0x90, /* nop */
592 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
593 static const char f32_9[] =
594 {0x89,0xf6, /* movl %esi,%esi */
595 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
596 static const char f32_10[] =
597 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
598 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
599 static const char f32_11[] =
600 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
601 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
602 static const char f32_12[] =
603 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
604 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
605 static const char f32_13[] =
606 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
607 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
608 static const char f32_14[] =
609 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
610 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
611 static const char f32_15[] =
612 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
613 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
614 static const char f16_3[] =
615 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
616 static const char f16_4[] =
617 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
618 static const char f16_5[] =
619 {0x90, /* nop */
620 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
621 static const char f16_6[] =
622 {0x89,0xf6, /* mov %si,%si */
623 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
624 static const char f16_7[] =
625 {0x8d,0x74,0x00, /* lea 0(%si),%si */
626 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
627 static const char f16_8[] =
628 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
629 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
630 static const char *const f32_patt[] = {
631 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
632 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
633 };
634 static const char *const f16_patt[] = {
635 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
636 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
637 };
638 /* nopl (%[re]ax) */
639 static const char alt_3[] =
640 {0x0f,0x1f,0x00};
641 /* nopl 0(%[re]ax) */
642 static const char alt_4[] =
643 {0x0f,0x1f,0x40,0x00};
644 /* nopl 0(%[re]ax,%[re]ax,1) */
645 static const char alt_5[] =
646 {0x0f,0x1f,0x44,0x00,0x00};
647 /* nopw 0(%[re]ax,%[re]ax,1) */
648 static const char alt_6[] =
649 {0x66,0x0f,0x1f,0x44,0x00,0x00};
650 /* nopl 0L(%[re]ax) */
651 static const char alt_7[] =
652 {0x0f,0x1f,0x80,0x00,0x00,0x00,0x00};
653 /* nopl 0L(%[re]ax,%[re]ax,1) */
654 static const char alt_8[] =
655 {0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
656 /* nopw 0L(%[re]ax,%[re]ax,1) */
657 static const char alt_9[] =
658 {0x66,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
659 /* nopw %cs:0L(%[re]ax,%[re]ax,1) */
660 static const char alt_10[] =
661 {0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
662 /* data16
663 nopw %cs:0L(%[re]ax,%[re]ax,1) */
664 static const char alt_long_11[] =
665 {0x66,
666 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
667 /* data16
668 data16
669 nopw %cs:0L(%[re]ax,%[re]ax,1) */
670 static const char alt_long_12[] =
671 {0x66,
672 0x66,
673 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
674 /* data16
675 data16
676 data16
677 nopw %cs:0L(%[re]ax,%[re]ax,1) */
678 static const char alt_long_13[] =
679 {0x66,
680 0x66,
681 0x66,
682 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
683 /* data16
684 data16
685 data16
686 data16
687 nopw %cs:0L(%[re]ax,%[re]ax,1) */
688 static const char alt_long_14[] =
689 {0x66,
690 0x66,
691 0x66,
692 0x66,
693 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
694 /* data16
695 data16
696 data16
697 data16
698 data16
699 nopw %cs:0L(%[re]ax,%[re]ax,1) */
700 static const char alt_long_15[] =
701 {0x66,
702 0x66,
703 0x66,
704 0x66,
705 0x66,
706 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
707 /* nopl 0(%[re]ax,%[re]ax,1)
708 nopw 0(%[re]ax,%[re]ax,1) */
709 static const char alt_short_11[] =
710 {0x0f,0x1f,0x44,0x00,0x00,
711 0x66,0x0f,0x1f,0x44,0x00,0x00};
712 /* nopw 0(%[re]ax,%[re]ax,1)
713 nopw 0(%[re]ax,%[re]ax,1) */
714 static const char alt_short_12[] =
715 {0x66,0x0f,0x1f,0x44,0x00,0x00,
716 0x66,0x0f,0x1f,0x44,0x00,0x00};
717 /* nopw 0(%[re]ax,%[re]ax,1)
718 nopl 0L(%[re]ax) */
719 static const char alt_short_13[] =
720 {0x66,0x0f,0x1f,0x44,0x00,0x00,
721 0x0f,0x1f,0x80,0x00,0x00,0x00,0x00};
722 /* nopl 0L(%[re]ax)
723 nopl 0L(%[re]ax) */
724 static const char alt_short_14[] =
725 {0x0f,0x1f,0x80,0x00,0x00,0x00,0x00,
726 0x0f,0x1f,0x80,0x00,0x00,0x00,0x00};
727 /* nopl 0L(%[re]ax)
728 nopl 0L(%[re]ax,%[re]ax,1) */
729 static const char alt_short_15[] =
730 {0x0f,0x1f,0x80,0x00,0x00,0x00,0x00,
731 0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
732 static const char *const alt_short_patt[] = {
733 f32_1, f32_2, alt_3, alt_4, alt_5, alt_6, alt_7, alt_8,
734 alt_9, alt_10, alt_short_11, alt_short_12, alt_short_13,
735 alt_short_14, alt_short_15
736 };
737 static const char *const alt_long_patt[] = {
738 f32_1, f32_2, alt_3, alt_4, alt_5, alt_6, alt_7, alt_8,
739 alt_9, alt_10, alt_long_11, alt_long_12, alt_long_13,
740 alt_long_14, alt_long_15
741 };
742
743 if (count <= 0 || count > 15)
744 return;
745
746 /* We need to decide which NOP sequence to use for 32bit and
747 64bit. When -mtune= is used:
748
749 1. For PROCESSOR_I486, PROCESSOR_PENTIUM and PROCESSOR_GENERIC32,
750 f32_patt will be used.
751 2. For PROCESSOR_K8 and PROCESSOR_AMDFAM10 in 64bit, NOPs with
752 0x66 prefix will be used.
753 3. For PROCESSOR_CORE2, alt_long_patt will be used.
754 4. For PROCESSOR_PENTIUMPRO, PROCESSOR_PENTIUM4, PROCESSOR_NOCONA,
755 PROCESSOR_CORE, PROCESSOR_CORE2, PROCESSOR_K6, PROCESSOR_ATHLON
756 and PROCESSOR_GENERIC64, alt_short_patt will be used.
757
758 When -mtune= isn't used, alt_short_patt will be used if
759 cpu_arch_isa_flags has Cpu686. Otherwise, f32_patt will be used.
760
761 When -march= or .arch is used, we can't use anything beyond
762 cpu_arch_isa_flags. */
763
764 if (flag_code == CODE_16BIT)
765 {
766 memcpy (fragP->fr_literal + fragP->fr_fix,
767 f16_patt[count - 1], count);
768 if (count > 8)
769 /* Adjust jump offset. */
770 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
771 }
772 else if (flag_code == CODE_64BIT && cpu_arch_tune == PROCESSOR_K8)
773 {
774 int i;
775 int nnops = (count + 3) / 4;
776 int len = count / nnops;
777 int remains = count - nnops * len;
778 int pos = 0;
779
780 /* The recommended way to pad 64bit code is to use NOPs preceded
781 by maximally four 0x66 prefixes. Balance the size of nops. */
782 for (i = 0; i < remains; i++)
783 {
784 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len);
785 fragP->fr_literal[fragP->fr_fix + pos + len] = 0x90;
786 pos += len + 1;
787 }
788 for (; i < nnops; i++)
789 {
790 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len - 1);
791 fragP->fr_literal[fragP->fr_fix + pos + len - 1] = 0x90;
792 pos += len;
793 }
794 }
795 else
796 {
797 const char *const *patt = NULL;
798
799 if (cpu_arch_isa == PROCESSOR_UNKNOWN)
800 {
801 /* PROCESSOR_UNKNOWN means that all ISAs may be used. */
802 switch (cpu_arch_tune)
803 {
804 case PROCESSOR_UNKNOWN:
805 /* We use cpu_arch_isa_flags to check if we SHOULD
806 optimize for Cpu686. */
807 if ((cpu_arch_isa_flags & Cpu686) != 0)
808 patt = alt_short_patt;
809 else
810 patt = f32_patt;
811 break;
812 case PROCESSOR_CORE2:
813 patt = alt_long_patt;
814 break;
815 case PROCESSOR_PENTIUMPRO:
816 case PROCESSOR_PENTIUM4:
817 case PROCESSOR_NOCONA:
818 case PROCESSOR_CORE:
819 case PROCESSOR_K6:
820 case PROCESSOR_ATHLON:
821 case PROCESSOR_K8:
822 case PROCESSOR_GENERIC64:
823 case PROCESSOR_AMDFAM10:
824 patt = alt_short_patt;
825 break;
826 case PROCESSOR_I486:
827 case PROCESSOR_PENTIUM:
828 case PROCESSOR_GENERIC32:
829 patt = f32_patt;
830 break;
831 }
832 }
833 else
834 {
835 switch (cpu_arch_tune)
836 {
837 case PROCESSOR_UNKNOWN:
838 /* When cpu_arch_isa is net, cpu_arch_tune shouldn't be
839 PROCESSOR_UNKNOWN. */
840 abort ();
841 break;
842
843 case PROCESSOR_I486:
844 case PROCESSOR_PENTIUM:
845 case PROCESSOR_PENTIUMPRO:
846 case PROCESSOR_PENTIUM4:
847 case PROCESSOR_NOCONA:
848 case PROCESSOR_CORE:
849 case PROCESSOR_K6:
850 case PROCESSOR_ATHLON:
851 case PROCESSOR_K8:
852 case PROCESSOR_AMDFAM10:
853 case PROCESSOR_GENERIC32:
854 /* We use cpu_arch_isa_flags to check if we CAN optimize
855 for Cpu686. */
856 if ((cpu_arch_isa_flags & Cpu686) != 0)
857 patt = alt_short_patt;
858 else
859 patt = f32_patt;
860 break;
861 case PROCESSOR_CORE2:
862 if ((cpu_arch_isa_flags & Cpu686) != 0)
863 patt = alt_long_patt;
864 else
865 patt = f32_patt;
866 break;
867 case PROCESSOR_GENERIC64:
868 patt = alt_short_patt;
869 break;
870 }
871 }
872
873 memcpy (fragP->fr_literal + fragP->fr_fix,
874 patt[count - 1], count);
875 }
876 fragP->fr_var = count;
877 }
878
879 static INLINE unsigned int
880 mode_from_disp_size (unsigned int t)
881 {
882 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
883 }
884
885 static INLINE int
886 fits_in_signed_byte (offsetT num)
887 {
888 return (num >= -128) && (num <= 127);
889 }
890
891 static INLINE int
892 fits_in_unsigned_byte (offsetT num)
893 {
894 return (num & 0xff) == num;
895 }
896
897 static INLINE int
898 fits_in_unsigned_word (offsetT num)
899 {
900 return (num & 0xffff) == num;
901 }
902
903 static INLINE int
904 fits_in_signed_word (offsetT num)
905 {
906 return (-32768 <= num) && (num <= 32767);
907 }
908
909 static INLINE int
910 fits_in_signed_long (offsetT num ATTRIBUTE_UNUSED)
911 {
912 #ifndef BFD64
913 return 1;
914 #else
915 return (!(((offsetT) -1 << 31) & num)
916 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
917 #endif
918 } /* fits_in_signed_long() */
919
920 static INLINE int
921 fits_in_unsigned_long (offsetT num ATTRIBUTE_UNUSED)
922 {
923 #ifndef BFD64
924 return 1;
925 #else
926 return (num & (((offsetT) 2 << 31) - 1)) == num;
927 #endif
928 } /* fits_in_unsigned_long() */
929
930 static unsigned int
931 smallest_imm_type (offsetT num)
932 {
933 if (cpu_arch_flags != (Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64))
934 {
935 /* This code is disabled on the 486 because all the Imm1 forms
936 in the opcode table are slower on the i486. They're the
937 versions with the implicitly specified single-position
938 displacement, which has another syntax if you really want to
939 use that form. */
940 if (num == 1)
941 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
942 }
943 return (fits_in_signed_byte (num)
944 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
945 : fits_in_unsigned_byte (num)
946 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
947 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
948 ? (Imm16 | Imm32 | Imm32S | Imm64)
949 : fits_in_signed_long (num)
950 ? (Imm32 | Imm32S | Imm64)
951 : fits_in_unsigned_long (num)
952 ? (Imm32 | Imm64)
953 : Imm64);
954 }
955
956 static offsetT
957 offset_in_range (offsetT val, int size)
958 {
959 addressT mask;
960
961 switch (size)
962 {
963 case 1: mask = ((addressT) 1 << 8) - 1; break;
964 case 2: mask = ((addressT) 1 << 16) - 1; break;
965 case 4: mask = ((addressT) 2 << 31) - 1; break;
966 #ifdef BFD64
967 case 8: mask = ((addressT) 2 << 63) - 1; break;
968 #endif
969 default: abort ();
970 }
971
972 /* If BFD64, sign extend val. */
973 if (!use_rela_relocations)
974 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
975 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
976
977 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
978 {
979 char buf1[40], buf2[40];
980
981 sprint_value (buf1, val);
982 sprint_value (buf2, val & mask);
983 as_warn (_("%s shortened to %s"), buf1, buf2);
984 }
985 return val & mask;
986 }
987
988 /* Returns 0 if attempting to add a prefix where one from the same
989 class already exists, 1 if non rep/repne added, 2 if rep/repne
990 added. */
991 static int
992 add_prefix (unsigned int prefix)
993 {
994 int ret = 1;
995 unsigned int q;
996
997 if (prefix >= REX_OPCODE && prefix < REX_OPCODE + 16
998 && flag_code == CODE_64BIT)
999 {
1000 if ((i.prefix[REX_PREFIX] & prefix & REX_W)
1001 || ((i.prefix[REX_PREFIX] & (REX_R | REX_X | REX_B))
1002 && (prefix & (REX_R | REX_X | REX_B))))
1003 ret = 0;
1004 q = REX_PREFIX;
1005 }
1006 else
1007 {
1008 switch (prefix)
1009 {
1010 default:
1011 abort ();
1012
1013 case CS_PREFIX_OPCODE:
1014 case DS_PREFIX_OPCODE:
1015 case ES_PREFIX_OPCODE:
1016 case FS_PREFIX_OPCODE:
1017 case GS_PREFIX_OPCODE:
1018 case SS_PREFIX_OPCODE:
1019 q = SEG_PREFIX;
1020 break;
1021
1022 case REPNE_PREFIX_OPCODE:
1023 case REPE_PREFIX_OPCODE:
1024 ret = 2;
1025 /* fall thru */
1026 case LOCK_PREFIX_OPCODE:
1027 q = LOCKREP_PREFIX;
1028 break;
1029
1030 case FWAIT_OPCODE:
1031 q = WAIT_PREFIX;
1032 break;
1033
1034 case ADDR_PREFIX_OPCODE:
1035 q = ADDR_PREFIX;
1036 break;
1037
1038 case DATA_PREFIX_OPCODE:
1039 q = DATA_PREFIX;
1040 break;
1041 }
1042 if (i.prefix[q] != 0)
1043 ret = 0;
1044 }
1045
1046 if (ret)
1047 {
1048 if (!i.prefix[q])
1049 ++i.prefixes;
1050 i.prefix[q] |= prefix;
1051 }
1052 else
1053 as_bad (_("same type of prefix used twice"));
1054
1055 return ret;
1056 }
1057
1058 static void
1059 set_code_flag (int value)
1060 {
1061 flag_code = value;
1062 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
1063 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
1064 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
1065 {
1066 as_bad (_("64bit mode not supported on this CPU."));
1067 }
1068 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
1069 {
1070 as_bad (_("32bit mode not supported on this CPU."));
1071 }
1072 stackop_size = '\0';
1073 }
1074
1075 static void
1076 set_16bit_gcc_code_flag (int new_code_flag)
1077 {
1078 flag_code = new_code_flag;
1079 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
1080 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
1081 stackop_size = LONG_MNEM_SUFFIX;
1082 }
1083
1084 static void
1085 set_intel_syntax (int syntax_flag)
1086 {
1087 /* Find out if register prefixing is specified. */
1088 int ask_naked_reg = 0;
1089
1090 SKIP_WHITESPACE ();
1091 if (!is_end_of_line[(unsigned char) *input_line_pointer])
1092 {
1093 char *string = input_line_pointer;
1094 int e = get_symbol_end ();
1095
1096 if (strcmp (string, "prefix") == 0)
1097 ask_naked_reg = 1;
1098 else if (strcmp (string, "noprefix") == 0)
1099 ask_naked_reg = -1;
1100 else
1101 as_bad (_("bad argument to syntax directive."));
1102 *input_line_pointer = e;
1103 }
1104 demand_empty_rest_of_line ();
1105
1106 intel_syntax = syntax_flag;
1107
1108 if (ask_naked_reg == 0)
1109 allow_naked_reg = (intel_syntax
1110 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
1111 else
1112 allow_naked_reg = (ask_naked_reg < 0);
1113
1114 identifier_chars['%'] = intel_syntax && allow_naked_reg ? '%' : 0;
1115 identifier_chars['$'] = intel_syntax ? '$' : 0;
1116 register_prefix = allow_naked_reg ? "" : "%";
1117 }
1118
1119 static void
1120 set_cpu_arch (int dummy ATTRIBUTE_UNUSED)
1121 {
1122 SKIP_WHITESPACE ();
1123
1124 if (!is_end_of_line[(unsigned char) *input_line_pointer])
1125 {
1126 char *string = input_line_pointer;
1127 int e = get_symbol_end ();
1128 unsigned int i;
1129
1130 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
1131 {
1132 if (strcmp (string, cpu_arch[i].name) == 0)
1133 {
1134 if (*string != '.')
1135 {
1136 cpu_arch_name = cpu_arch[i].name;
1137 cpu_sub_arch_name = NULL;
1138 cpu_arch_flags = (cpu_arch[i].flags
1139 | (flag_code == CODE_64BIT
1140 ? Cpu64 : CpuNo64));
1141 cpu_arch_isa = cpu_arch[i].type;
1142 cpu_arch_isa_flags = cpu_arch[i].flags;
1143 if (!cpu_arch_tune_set)
1144 {
1145 cpu_arch_tune = cpu_arch_isa;
1146 cpu_arch_tune_flags = cpu_arch_isa_flags;
1147 }
1148 break;
1149 }
1150 if ((cpu_arch_flags | cpu_arch[i].flags) != cpu_arch_flags)
1151 {
1152 cpu_sub_arch_name = cpu_arch[i].name;
1153 cpu_arch_flags |= cpu_arch[i].flags;
1154 }
1155 *input_line_pointer = e;
1156 demand_empty_rest_of_line ();
1157 return;
1158 }
1159 }
1160 if (i >= ARRAY_SIZE (cpu_arch))
1161 as_bad (_("no such architecture: `%s'"), string);
1162
1163 *input_line_pointer = e;
1164 }
1165 else
1166 as_bad (_("missing cpu architecture"));
1167
1168 no_cond_jump_promotion = 0;
1169 if (*input_line_pointer == ','
1170 && !is_end_of_line[(unsigned char) input_line_pointer[1]])
1171 {
1172 char *string = ++input_line_pointer;
1173 int e = get_symbol_end ();
1174
1175 if (strcmp (string, "nojumps") == 0)
1176 no_cond_jump_promotion = 1;
1177 else if (strcmp (string, "jumps") == 0)
1178 ;
1179 else
1180 as_bad (_("no such architecture modifier: `%s'"), string);
1181
1182 *input_line_pointer = e;
1183 }
1184
1185 demand_empty_rest_of_line ();
1186 }
1187
1188 unsigned long
1189 i386_mach ()
1190 {
1191 if (!strcmp (default_arch, "x86_64"))
1192 return bfd_mach_x86_64;
1193 else if (!strcmp (default_arch, "i386"))
1194 return bfd_mach_i386_i386;
1195 else
1196 as_fatal (_("Unknown architecture"));
1197 }
1198 \f
1199 void
1200 md_begin ()
1201 {
1202 const char *hash_err;
1203
1204 /* Initialize op_hash hash table. */
1205 op_hash = hash_new ();
1206
1207 {
1208 const template *optab;
1209 templates *core_optab;
1210
1211 /* Setup for loop. */
1212 optab = i386_optab;
1213 core_optab = (templates *) xmalloc (sizeof (templates));
1214 core_optab->start = optab;
1215
1216 while (1)
1217 {
1218 ++optab;
1219 if (optab->name == NULL
1220 || strcmp (optab->name, (optab - 1)->name) != 0)
1221 {
1222 /* different name --> ship out current template list;
1223 add to hash table; & begin anew. */
1224 core_optab->end = optab;
1225 hash_err = hash_insert (op_hash,
1226 (optab - 1)->name,
1227 (PTR) core_optab);
1228 if (hash_err)
1229 {
1230 as_fatal (_("Internal Error: Can't hash %s: %s"),
1231 (optab - 1)->name,
1232 hash_err);
1233 }
1234 if (optab->name == NULL)
1235 break;
1236 core_optab = (templates *) xmalloc (sizeof (templates));
1237 core_optab->start = optab;
1238 }
1239 }
1240 }
1241
1242 /* Initialize reg_hash hash table. */
1243 reg_hash = hash_new ();
1244 {
1245 const reg_entry *regtab;
1246 unsigned int regtab_size = i386_regtab_size;
1247
1248 for (regtab = i386_regtab; regtab_size--; regtab++)
1249 {
1250 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
1251 if (hash_err)
1252 as_fatal (_("Internal Error: Can't hash %s: %s"),
1253 regtab->reg_name,
1254 hash_err);
1255 }
1256 }
1257
1258 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
1259 {
1260 int c;
1261 char *p;
1262
1263 for (c = 0; c < 256; c++)
1264 {
1265 if (ISDIGIT (c))
1266 {
1267 digit_chars[c] = c;
1268 mnemonic_chars[c] = c;
1269 register_chars[c] = c;
1270 operand_chars[c] = c;
1271 }
1272 else if (ISLOWER (c))
1273 {
1274 mnemonic_chars[c] = c;
1275 register_chars[c] = c;
1276 operand_chars[c] = c;
1277 }
1278 else if (ISUPPER (c))
1279 {
1280 mnemonic_chars[c] = TOLOWER (c);
1281 register_chars[c] = mnemonic_chars[c];
1282 operand_chars[c] = c;
1283 }
1284
1285 if (ISALPHA (c) || ISDIGIT (c))
1286 identifier_chars[c] = c;
1287 else if (c >= 128)
1288 {
1289 identifier_chars[c] = c;
1290 operand_chars[c] = c;
1291 }
1292 }
1293
1294 #ifdef LEX_AT
1295 identifier_chars['@'] = '@';
1296 #endif
1297 #ifdef LEX_QM
1298 identifier_chars['?'] = '?';
1299 operand_chars['?'] = '?';
1300 #endif
1301 digit_chars['-'] = '-';
1302 mnemonic_chars['-'] = '-';
1303 mnemonic_chars['.'] = '.';
1304 identifier_chars['_'] = '_';
1305 identifier_chars['.'] = '.';
1306
1307 for (p = operand_special_chars; *p != '\0'; p++)
1308 operand_chars[(unsigned char) *p] = *p;
1309 }
1310
1311 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1312 if (IS_ELF)
1313 {
1314 record_alignment (text_section, 2);
1315 record_alignment (data_section, 2);
1316 record_alignment (bss_section, 2);
1317 }
1318 #endif
1319
1320 if (flag_code == CODE_64BIT)
1321 {
1322 x86_dwarf2_return_column = 16;
1323 x86_cie_data_alignment = -8;
1324 }
1325 else
1326 {
1327 x86_dwarf2_return_column = 8;
1328 x86_cie_data_alignment = -4;
1329 }
1330 }
1331
1332 void
1333 i386_print_statistics (FILE *file)
1334 {
1335 hash_print_statistics (file, "i386 opcode", op_hash);
1336 hash_print_statistics (file, "i386 register", reg_hash);
1337 }
1338 \f
1339 #ifdef DEBUG386
1340
1341 /* Debugging routines for md_assemble. */
1342 static void pte (template *);
1343 static void pt (unsigned int);
1344 static void pe (expressionS *);
1345 static void ps (symbolS *);
1346
1347 static void
1348 pi (char *line, i386_insn *x)
1349 {
1350 unsigned int i;
1351
1352 fprintf (stdout, "%s: template ", line);
1353 pte (&x->tm);
1354 fprintf (stdout, " address: base %s index %s scale %x\n",
1355 x->base_reg ? x->base_reg->reg_name : "none",
1356 x->index_reg ? x->index_reg->reg_name : "none",
1357 x->log2_scale_factor);
1358 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
1359 x->rm.mode, x->rm.reg, x->rm.regmem);
1360 fprintf (stdout, " sib: base %x index %x scale %x\n",
1361 x->sib.base, x->sib.index, x->sib.scale);
1362 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
1363 (x->rex & REX_W) != 0,
1364 (x->rex & REX_R) != 0,
1365 (x->rex & REX_X) != 0,
1366 (x->rex & REX_B) != 0);
1367 for (i = 0; i < x->operands; i++)
1368 {
1369 fprintf (stdout, " #%d: ", i + 1);
1370 pt (x->types[i]);
1371 fprintf (stdout, "\n");
1372 if (x->types[i]
1373 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
1374 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
1375 if (x->types[i] & Imm)
1376 pe (x->op[i].imms);
1377 if (x->types[i] & Disp)
1378 pe (x->op[i].disps);
1379 }
1380 }
1381
1382 static void
1383 pte (template *t)
1384 {
1385 unsigned int i;
1386 fprintf (stdout, " %d operands ", t->operands);
1387 fprintf (stdout, "opcode %x ", t->base_opcode);
1388 if (t->extension_opcode != None)
1389 fprintf (stdout, "ext %x ", t->extension_opcode);
1390 if (t->opcode_modifier & D)
1391 fprintf (stdout, "D");
1392 if (t->opcode_modifier & W)
1393 fprintf (stdout, "W");
1394 fprintf (stdout, "\n");
1395 for (i = 0; i < t->operands; i++)
1396 {
1397 fprintf (stdout, " #%d type ", i + 1);
1398 pt (t->operand_types[i]);
1399 fprintf (stdout, "\n");
1400 }
1401 }
1402
1403 static void
1404 pe (expressionS *e)
1405 {
1406 fprintf (stdout, " operation %d\n", e->X_op);
1407 fprintf (stdout, " add_number %ld (%lx)\n",
1408 (long) e->X_add_number, (long) e->X_add_number);
1409 if (e->X_add_symbol)
1410 {
1411 fprintf (stdout, " add_symbol ");
1412 ps (e->X_add_symbol);
1413 fprintf (stdout, "\n");
1414 }
1415 if (e->X_op_symbol)
1416 {
1417 fprintf (stdout, " op_symbol ");
1418 ps (e->X_op_symbol);
1419 fprintf (stdout, "\n");
1420 }
1421 }
1422
1423 static void
1424 ps (symbolS *s)
1425 {
1426 fprintf (stdout, "%s type %s%s",
1427 S_GET_NAME (s),
1428 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
1429 segment_name (S_GET_SEGMENT (s)));
1430 }
1431
1432 static struct type_name
1433 {
1434 unsigned int mask;
1435 char *tname;
1436 }
1437 const type_names[] =
1438 {
1439 { Reg8, "r8" },
1440 { Reg16, "r16" },
1441 { Reg32, "r32" },
1442 { Reg64, "r64" },
1443 { Imm8, "i8" },
1444 { Imm8S, "i8s" },
1445 { Imm16, "i16" },
1446 { Imm32, "i32" },
1447 { Imm32S, "i32s" },
1448 { Imm64, "i64" },
1449 { Imm1, "i1" },
1450 { BaseIndex, "BaseIndex" },
1451 { Disp8, "d8" },
1452 { Disp16, "d16" },
1453 { Disp32, "d32" },
1454 { Disp32S, "d32s" },
1455 { Disp64, "d64" },
1456 { InOutPortReg, "InOutPortReg" },
1457 { ShiftCount, "ShiftCount" },
1458 { Control, "control reg" },
1459 { Test, "test reg" },
1460 { Debug, "debug reg" },
1461 { FloatReg, "FReg" },
1462 { FloatAcc, "FAcc" },
1463 { SReg2, "SReg2" },
1464 { SReg3, "SReg3" },
1465 { Acc, "Acc" },
1466 { JumpAbsolute, "Jump Absolute" },
1467 { RegMMX, "rMMX" },
1468 { RegXMM, "rXMM" },
1469 { EsSeg, "es" },
1470 { 0, "" }
1471 };
1472
1473 static void
1474 pt (t)
1475 unsigned int t;
1476 {
1477 const struct type_name *ty;
1478
1479 for (ty = type_names; ty->mask; ty++)
1480 if (t & ty->mask)
1481 fprintf (stdout, "%s, ", ty->tname);
1482 fflush (stdout);
1483 }
1484
1485 #endif /* DEBUG386 */
1486 \f
1487 static bfd_reloc_code_real_type
1488 reloc (unsigned int size,
1489 int pcrel,
1490 int sign,
1491 bfd_reloc_code_real_type other)
1492 {
1493 if (other != NO_RELOC)
1494 {
1495 reloc_howto_type *reloc;
1496
1497 if (size == 8)
1498 switch (other)
1499 {
1500 case BFD_RELOC_X86_64_GOT32:
1501 return BFD_RELOC_X86_64_GOT64;
1502 break;
1503 case BFD_RELOC_X86_64_PLTOFF64:
1504 return BFD_RELOC_X86_64_PLTOFF64;
1505 break;
1506 case BFD_RELOC_X86_64_GOTPC32:
1507 other = BFD_RELOC_X86_64_GOTPC64;
1508 break;
1509 case BFD_RELOC_X86_64_GOTPCREL:
1510 other = BFD_RELOC_X86_64_GOTPCREL64;
1511 break;
1512 case BFD_RELOC_X86_64_TPOFF32:
1513 other = BFD_RELOC_X86_64_TPOFF64;
1514 break;
1515 case BFD_RELOC_X86_64_DTPOFF32:
1516 other = BFD_RELOC_X86_64_DTPOFF64;
1517 break;
1518 default:
1519 break;
1520 }
1521
1522 /* Sign-checking 4-byte relocations in 16-/32-bit code is pointless. */
1523 if (size == 4 && flag_code != CODE_64BIT)
1524 sign = -1;
1525
1526 reloc = bfd_reloc_type_lookup (stdoutput, other);
1527 if (!reloc)
1528 as_bad (_("unknown relocation (%u)"), other);
1529 else if (size != bfd_get_reloc_size (reloc))
1530 as_bad (_("%u-byte relocation cannot be applied to %u-byte field"),
1531 bfd_get_reloc_size (reloc),
1532 size);
1533 else if (pcrel && !reloc->pc_relative)
1534 as_bad (_("non-pc-relative relocation for pc-relative field"));
1535 else if ((reloc->complain_on_overflow == complain_overflow_signed
1536 && !sign)
1537 || (reloc->complain_on_overflow == complain_overflow_unsigned
1538 && sign > 0))
1539 as_bad (_("relocated field and relocation type differ in signedness"));
1540 else
1541 return other;
1542 return NO_RELOC;
1543 }
1544
1545 if (pcrel)
1546 {
1547 if (!sign)
1548 as_bad (_("there are no unsigned pc-relative relocations"));
1549 switch (size)
1550 {
1551 case 1: return BFD_RELOC_8_PCREL;
1552 case 2: return BFD_RELOC_16_PCREL;
1553 case 4: return BFD_RELOC_32_PCREL;
1554 case 8: return BFD_RELOC_64_PCREL;
1555 }
1556 as_bad (_("cannot do %u byte pc-relative relocation"), size);
1557 }
1558 else
1559 {
1560 if (sign > 0)
1561 switch (size)
1562 {
1563 case 4: return BFD_RELOC_X86_64_32S;
1564 }
1565 else
1566 switch (size)
1567 {
1568 case 1: return BFD_RELOC_8;
1569 case 2: return BFD_RELOC_16;
1570 case 4: return BFD_RELOC_32;
1571 case 8: return BFD_RELOC_64;
1572 }
1573 as_bad (_("cannot do %s %u byte relocation"),
1574 sign > 0 ? "signed" : "unsigned", size);
1575 }
1576
1577 abort ();
1578 return BFD_RELOC_NONE;
1579 }
1580
1581 /* Here we decide which fixups can be adjusted to make them relative to
1582 the beginning of the section instead of the symbol. Basically we need
1583 to make sure that the dynamic relocations are done correctly, so in
1584 some cases we force the original symbol to be used. */
1585
1586 int
1587 tc_i386_fix_adjustable (fixS *fixP ATTRIBUTE_UNUSED)
1588 {
1589 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1590 if (!IS_ELF)
1591 return 1;
1592
1593 /* Don't adjust pc-relative references to merge sections in 64-bit
1594 mode. */
1595 if (use_rela_relocations
1596 && (S_GET_SEGMENT (fixP->fx_addsy)->flags & SEC_MERGE) != 0
1597 && fixP->fx_pcrel)
1598 return 0;
1599
1600 /* The x86_64 GOTPCREL are represented as 32bit PCrel relocations
1601 and changed later by validate_fix. */
1602 if (GOT_symbol && fixP->fx_subsy == GOT_symbol
1603 && fixP->fx_r_type == BFD_RELOC_32_PCREL)
1604 return 0;
1605
1606 /* adjust_reloc_syms doesn't know about the GOT. */
1607 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1608 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1609 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1610 || fixP->fx_r_type == BFD_RELOC_386_TLS_GD
1611 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDM
1612 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDO_32
1613 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE_32
1614 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE
1615 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTIE
1616 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE_32
1617 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE
1618 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTDESC
1619 || fixP->fx_r_type == BFD_RELOC_386_TLS_DESC_CALL
1620 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1621 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1622 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
1623 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSGD
1624 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSLD
1625 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF32
1626 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF64
1627 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTTPOFF
1628 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF32
1629 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF64
1630 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTOFF64
1631 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPC32_TLSDESC
1632 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSDESC_CALL
1633 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1634 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1635 return 0;
1636 #endif
1637 return 1;
1638 }
1639
1640 static int
1641 intel_float_operand (const char *mnemonic)
1642 {
1643 /* Note that the value returned is meaningful only for opcodes with (memory)
1644 operands, hence the code here is free to improperly handle opcodes that
1645 have no operands (for better performance and smaller code). */
1646
1647 if (mnemonic[0] != 'f')
1648 return 0; /* non-math */
1649
1650 switch (mnemonic[1])
1651 {
1652 /* fclex, fdecstp, fdisi, femms, feni, fincstp, finit, fsetpm, and
1653 the fs segment override prefix not currently handled because no
1654 call path can make opcodes without operands get here */
1655 case 'i':
1656 return 2 /* integer op */;
1657 case 'l':
1658 if (mnemonic[2] == 'd' && (mnemonic[3] == 'c' || mnemonic[3] == 'e'))
1659 return 3; /* fldcw/fldenv */
1660 break;
1661 case 'n':
1662 if (mnemonic[2] != 'o' /* fnop */)
1663 return 3; /* non-waiting control op */
1664 break;
1665 case 'r':
1666 if (mnemonic[2] == 's')
1667 return 3; /* frstor/frstpm */
1668 break;
1669 case 's':
1670 if (mnemonic[2] == 'a')
1671 return 3; /* fsave */
1672 if (mnemonic[2] == 't')
1673 {
1674 switch (mnemonic[3])
1675 {
1676 case 'c': /* fstcw */
1677 case 'd': /* fstdw */
1678 case 'e': /* fstenv */
1679 case 's': /* fsts[gw] */
1680 return 3;
1681 }
1682 }
1683 break;
1684 case 'x':
1685 if (mnemonic[2] == 'r' || mnemonic[2] == 's')
1686 return 0; /* fxsave/fxrstor are not really math ops */
1687 break;
1688 }
1689
1690 return 1;
1691 }
1692
1693 /* This is the guts of the machine-dependent assembler. LINE points to a
1694 machine dependent instruction. This function is supposed to emit
1695 the frags/bytes it assembles to. */
1696
1697 void
1698 md_assemble (line)
1699 char *line;
1700 {
1701 int j;
1702 char mnemonic[MAX_MNEM_SIZE];
1703
1704 /* Initialize globals. */
1705 memset (&i, '\0', sizeof (i));
1706 for (j = 0; j < MAX_OPERANDS; j++)
1707 i.reloc[j] = NO_RELOC;
1708 memset (disp_expressions, '\0', sizeof (disp_expressions));
1709 memset (im_expressions, '\0', sizeof (im_expressions));
1710 save_stack_p = save_stack;
1711
1712 /* First parse an instruction mnemonic & call i386_operand for the operands.
1713 We assume that the scrubber has arranged it so that line[0] is the valid
1714 start of a (possibly prefixed) mnemonic. */
1715
1716 line = parse_insn (line, mnemonic);
1717 if (line == NULL)
1718 return;
1719
1720 line = parse_operands (line, mnemonic);
1721 if (line == NULL)
1722 return;
1723
1724 /* The order of the immediates should be reversed
1725 for 2 immediates extrq and insertq instructions */
1726 if ((i.imm_operands == 2)
1727 && ((strcmp (mnemonic, "extrq") == 0)
1728 || (strcmp (mnemonic, "insertq") == 0)))
1729 {
1730 swap_2_operands (0, 1);
1731 /* "extrq" and insertq" are the only two instructions whose operands
1732 have to be reversed even though they have two immediate operands.
1733 */
1734 if (intel_syntax)
1735 swap_operands ();
1736 }
1737
1738 /* Now we've parsed the mnemonic into a set of templates, and have the
1739 operands at hand. */
1740
1741 /* All intel opcodes have reversed operands except for "bound" and
1742 "enter". We also don't reverse intersegment "jmp" and "call"
1743 instructions with 2 immediate operands so that the immediate segment
1744 precedes the offset, as it does when in AT&T mode. */
1745 if (intel_syntax
1746 && i.operands > 1
1747 && (strcmp (mnemonic, "bound") != 0)
1748 && (strcmp (mnemonic, "invlpga") != 0)
1749 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1750 swap_operands ();
1751
1752 if (i.imm_operands)
1753 optimize_imm ();
1754
1755 /* Don't optimize displacement for movabs since it only takes 64bit
1756 displacement. */
1757 if (i.disp_operands
1758 && (flag_code != CODE_64BIT
1759 || strcmp (mnemonic, "movabs") != 0))
1760 optimize_disp ();
1761
1762 /* Next, we find a template that matches the given insn,
1763 making sure the overlap of the given operands types is consistent
1764 with the template operand types. */
1765
1766 if (!match_template ())
1767 return;
1768
1769 if (intel_syntax)
1770 {
1771 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1772 if (SYSV386_COMPAT
1773 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1774 i.tm.base_opcode ^= Opcode_FloatR;
1775
1776 /* Zap movzx and movsx suffix. The suffix may have been set from
1777 "word ptr" or "byte ptr" on the source operand, but we'll use
1778 the suffix later to choose the destination register. */
1779 if ((i.tm.base_opcode & ~9) == 0x0fb6)
1780 {
1781 if (i.reg_operands < 2
1782 && !i.suffix
1783 && (~i.tm.opcode_modifier
1784 & (No_bSuf
1785 | No_wSuf
1786 | No_lSuf
1787 | No_sSuf
1788 | No_xSuf
1789 | No_qSuf)))
1790 as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
1791
1792 i.suffix = 0;
1793 }
1794 }
1795
1796 if (i.tm.opcode_modifier & FWait)
1797 if (!add_prefix (FWAIT_OPCODE))
1798 return;
1799
1800 /* Check string instruction segment overrides. */
1801 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1802 {
1803 if (!check_string ())
1804 return;
1805 }
1806
1807 if (!process_suffix ())
1808 return;
1809
1810 /* Make still unresolved immediate matches conform to size of immediate
1811 given in i.suffix. */
1812 if (!finalize_imm ())
1813 return;
1814
1815 if (i.types[0] & Imm1)
1816 i.imm_operands = 0; /* kludge for shift insns. */
1817 if (i.types[0] & ImplicitRegister)
1818 i.reg_operands--;
1819 if (i.types[1] & ImplicitRegister)
1820 i.reg_operands--;
1821 if (i.types[2] & ImplicitRegister)
1822 i.reg_operands--;
1823
1824 if (i.tm.opcode_modifier & ImmExt)
1825 {
1826 expressionS *exp;
1827
1828 if ((i.tm.cpu_flags & CpuSSE3) && i.operands > 0)
1829 {
1830 /* Streaming SIMD extensions 3 Instructions have the fixed
1831 operands with an opcode suffix which is coded in the same
1832 place as an 8-bit immediate field would be. Here we check
1833 those operands and remove them afterwards. */
1834 unsigned int x;
1835
1836 for (x = 0; x < i.operands; x++)
1837 if (i.op[x].regs->reg_num != x)
1838 as_bad (_("can't use register '%%%s' as operand %d in '%s'."),
1839 i.op[x].regs->reg_name, x + 1, i.tm.name);
1840 i.operands = 0;
1841 }
1842
1843 /* These AMD 3DNow! and Intel Katmai New Instructions have an
1844 opcode suffix which is coded in the same place as an 8-bit
1845 immediate field would be. Here we fake an 8-bit immediate
1846 operand from the opcode suffix stored in tm.extension_opcode. */
1847
1848 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
1849
1850 exp = &im_expressions[i.imm_operands++];
1851 i.op[i.operands].imms = exp;
1852 i.types[i.operands++] = Imm8;
1853 exp->X_op = O_constant;
1854 exp->X_add_number = i.tm.extension_opcode;
1855 i.tm.extension_opcode = None;
1856 }
1857
1858 /* For insns with operands there are more diddles to do to the opcode. */
1859 if (i.operands)
1860 {
1861 if (!process_operands ())
1862 return;
1863 }
1864 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
1865 {
1866 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
1867 as_warn (_("translating to `%sp'"), i.tm.name);
1868 }
1869
1870 /* Handle conversion of 'int $3' --> special int3 insn. */
1871 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
1872 {
1873 i.tm.base_opcode = INT3_OPCODE;
1874 i.imm_operands = 0;
1875 }
1876
1877 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
1878 && i.op[0].disps->X_op == O_constant)
1879 {
1880 /* Convert "jmp constant" (and "call constant") to a jump (call) to
1881 the absolute address given by the constant. Since ix86 jumps and
1882 calls are pc relative, we need to generate a reloc. */
1883 i.op[0].disps->X_add_symbol = &abs_symbol;
1884 i.op[0].disps->X_op = O_symbol;
1885 }
1886
1887 if ((i.tm.opcode_modifier & Rex64) != 0)
1888 i.rex |= REX_W;
1889
1890 /* For 8 bit registers we need an empty rex prefix. Also if the
1891 instruction already has a prefix, we need to convert old
1892 registers to new ones. */
1893
1894 if (((i.types[0] & Reg8) != 0
1895 && (i.op[0].regs->reg_flags & RegRex64) != 0)
1896 || ((i.types[1] & Reg8) != 0
1897 && (i.op[1].regs->reg_flags & RegRex64) != 0)
1898 || (((i.types[0] & Reg8) != 0 || (i.types[1] & Reg8) != 0)
1899 && i.rex != 0))
1900 {
1901 int x;
1902
1903 i.rex |= REX_OPCODE;
1904 for (x = 0; x < 2; x++)
1905 {
1906 /* Look for 8 bit operand that uses old registers. */
1907 if ((i.types[x] & Reg8) != 0
1908 && (i.op[x].regs->reg_flags & RegRex64) == 0)
1909 {
1910 /* In case it is "hi" register, give up. */
1911 if (i.op[x].regs->reg_num > 3)
1912 as_bad (_("can't encode register '%%%s' in an "
1913 "instruction requiring REX prefix."),
1914 i.op[x].regs->reg_name);
1915
1916 /* Otherwise it is equivalent to the extended register.
1917 Since the encoding doesn't change this is merely
1918 cosmetic cleanup for debug output. */
1919
1920 i.op[x].regs = i.op[x].regs + 8;
1921 }
1922 }
1923 }
1924
1925 if (i.rex != 0)
1926 add_prefix (REX_OPCODE | i.rex);
1927
1928 /* We are ready to output the insn. */
1929 output_insn ();
1930 }
1931
1932 static char *
1933 parse_insn (char *line, char *mnemonic)
1934 {
1935 char *l = line;
1936 char *token_start = l;
1937 char *mnem_p;
1938 int supported;
1939 const template *t;
1940
1941 /* Non-zero if we found a prefix only acceptable with string insns. */
1942 const char *expecting_string_instruction = NULL;
1943
1944 while (1)
1945 {
1946 mnem_p = mnemonic;
1947 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1948 {
1949 mnem_p++;
1950 if (mnem_p >= mnemonic + MAX_MNEM_SIZE)
1951 {
1952 as_bad (_("no such instruction: `%s'"), token_start);
1953 return NULL;
1954 }
1955 l++;
1956 }
1957 if (!is_space_char (*l)
1958 && *l != END_OF_INSN
1959 && (intel_syntax
1960 || (*l != PREFIX_SEPARATOR
1961 && *l != ',')))
1962 {
1963 as_bad (_("invalid character %s in mnemonic"),
1964 output_invalid (*l));
1965 return NULL;
1966 }
1967 if (token_start == l)
1968 {
1969 if (!intel_syntax && *l == PREFIX_SEPARATOR)
1970 as_bad (_("expecting prefix; got nothing"));
1971 else
1972 as_bad (_("expecting mnemonic; got nothing"));
1973 return NULL;
1974 }
1975
1976 /* Look up instruction (or prefix) via hash table. */
1977 current_templates = hash_find (op_hash, mnemonic);
1978
1979 if (*l != END_OF_INSN
1980 && (!is_space_char (*l) || l[1] != END_OF_INSN)
1981 && current_templates
1982 && (current_templates->start->opcode_modifier & IsPrefix))
1983 {
1984 if (current_templates->start->cpu_flags
1985 & (flag_code != CODE_64BIT ? Cpu64 : CpuNo64))
1986 {
1987 as_bad ((flag_code != CODE_64BIT
1988 ? _("`%s' is only supported in 64-bit mode")
1989 : _("`%s' is not supported in 64-bit mode")),
1990 current_templates->start->name);
1991 return NULL;
1992 }
1993 /* If we are in 16-bit mode, do not allow addr16 or data16.
1994 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1995 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1996 && flag_code != CODE_64BIT
1997 && (((current_templates->start->opcode_modifier & Size32) != 0)
1998 ^ (flag_code == CODE_16BIT)))
1999 {
2000 as_bad (_("redundant %s prefix"),
2001 current_templates->start->name);
2002 return NULL;
2003 }
2004 /* Add prefix, checking for repeated prefixes. */
2005 switch (add_prefix (current_templates->start->base_opcode))
2006 {
2007 case 0:
2008 return NULL;
2009 case 2:
2010 expecting_string_instruction = current_templates->start->name;
2011 break;
2012 }
2013 /* Skip past PREFIX_SEPARATOR and reset token_start. */
2014 token_start = ++l;
2015 }
2016 else
2017 break;
2018 }
2019
2020 if (!current_templates)
2021 {
2022 /* See if we can get a match by trimming off a suffix. */
2023 switch (mnem_p[-1])
2024 {
2025 case WORD_MNEM_SUFFIX:
2026 if (intel_syntax && (intel_float_operand (mnemonic) & 2))
2027 i.suffix = SHORT_MNEM_SUFFIX;
2028 else
2029 case BYTE_MNEM_SUFFIX:
2030 case QWORD_MNEM_SUFFIX:
2031 i.suffix = mnem_p[-1];
2032 mnem_p[-1] = '\0';
2033 current_templates = hash_find (op_hash, mnemonic);
2034 break;
2035 case SHORT_MNEM_SUFFIX:
2036 case LONG_MNEM_SUFFIX:
2037 if (!intel_syntax)
2038 {
2039 i.suffix = mnem_p[-1];
2040 mnem_p[-1] = '\0';
2041 current_templates = hash_find (op_hash, mnemonic);
2042 }
2043 break;
2044
2045 /* Intel Syntax. */
2046 case 'd':
2047 if (intel_syntax)
2048 {
2049 if (intel_float_operand (mnemonic) == 1)
2050 i.suffix = SHORT_MNEM_SUFFIX;
2051 else
2052 i.suffix = LONG_MNEM_SUFFIX;
2053 mnem_p[-1] = '\0';
2054 current_templates = hash_find (op_hash, mnemonic);
2055 }
2056 break;
2057 }
2058 if (!current_templates)
2059 {
2060 as_bad (_("no such instruction: `%s'"), token_start);
2061 return NULL;
2062 }
2063 }
2064
2065 if (current_templates->start->opcode_modifier & (Jump | JumpByte))
2066 {
2067 /* Check for a branch hint. We allow ",pt" and ",pn" for
2068 predict taken and predict not taken respectively.
2069 I'm not sure that branch hints actually do anything on loop
2070 and jcxz insns (JumpByte) for current Pentium4 chips. They
2071 may work in the future and it doesn't hurt to accept them
2072 now. */
2073 if (l[0] == ',' && l[1] == 'p')
2074 {
2075 if (l[2] == 't')
2076 {
2077 if (!add_prefix (DS_PREFIX_OPCODE))
2078 return NULL;
2079 l += 3;
2080 }
2081 else if (l[2] == 'n')
2082 {
2083 if (!add_prefix (CS_PREFIX_OPCODE))
2084 return NULL;
2085 l += 3;
2086 }
2087 }
2088 }
2089 /* Any other comma loses. */
2090 if (*l == ',')
2091 {
2092 as_bad (_("invalid character %s in mnemonic"),
2093 output_invalid (*l));
2094 return NULL;
2095 }
2096
2097 /* Check if instruction is supported on specified architecture. */
2098 supported = 0;
2099 for (t = current_templates->start; t < current_templates->end; ++t)
2100 {
2101 if (!((t->cpu_flags & ~(Cpu64 | CpuNo64))
2102 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64))))
2103 supported |= 1;
2104 if (!(t->cpu_flags & (flag_code == CODE_64BIT ? CpuNo64 : Cpu64)))
2105 supported |= 2;
2106 }
2107 if (!(supported & 2))
2108 {
2109 as_bad (flag_code == CODE_64BIT
2110 ? _("`%s' is not supported in 64-bit mode")
2111 : _("`%s' is only supported in 64-bit mode"),
2112 current_templates->start->name);
2113 return NULL;
2114 }
2115 if (!(supported & 1))
2116 {
2117 as_warn (_("`%s' is not supported on `%s%s'"),
2118 current_templates->start->name,
2119 cpu_arch_name,
2120 cpu_sub_arch_name ? cpu_sub_arch_name : "");
2121 }
2122 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
2123 {
2124 as_warn (_("use .code16 to ensure correct addressing mode"));
2125 }
2126
2127 /* Check for rep/repne without a string instruction. */
2128 if (expecting_string_instruction)
2129 {
2130 static templates override;
2131
2132 for (t = current_templates->start; t < current_templates->end; ++t)
2133 if (t->opcode_modifier & IsString)
2134 break;
2135 if (t >= current_templates->end)
2136 {
2137 as_bad (_("expecting string instruction after `%s'"),
2138 expecting_string_instruction);
2139 return NULL;
2140 }
2141 for (override.start = t; t < current_templates->end; ++t)
2142 if (!(t->opcode_modifier & IsString))
2143 break;
2144 override.end = t;
2145 current_templates = &override;
2146 }
2147
2148 return l;
2149 }
2150
2151 static char *
2152 parse_operands (char *l, const char *mnemonic)
2153 {
2154 char *token_start;
2155
2156 /* 1 if operand is pending after ','. */
2157 unsigned int expecting_operand = 0;
2158
2159 /* Non-zero if operand parens not balanced. */
2160 unsigned int paren_not_balanced;
2161
2162 while (*l != END_OF_INSN)
2163 {
2164 /* Skip optional white space before operand. */
2165 if (is_space_char (*l))
2166 ++l;
2167 if (!is_operand_char (*l) && *l != END_OF_INSN)
2168 {
2169 as_bad (_("invalid character %s before operand %d"),
2170 output_invalid (*l),
2171 i.operands + 1);
2172 return NULL;
2173 }
2174 token_start = l; /* after white space */
2175 paren_not_balanced = 0;
2176 while (paren_not_balanced || *l != ',')
2177 {
2178 if (*l == END_OF_INSN)
2179 {
2180 if (paren_not_balanced)
2181 {
2182 if (!intel_syntax)
2183 as_bad (_("unbalanced parenthesis in operand %d."),
2184 i.operands + 1);
2185 else
2186 as_bad (_("unbalanced brackets in operand %d."),
2187 i.operands + 1);
2188 return NULL;
2189 }
2190 else
2191 break; /* we are done */
2192 }
2193 else if (!is_operand_char (*l) && !is_space_char (*l))
2194 {
2195 as_bad (_("invalid character %s in operand %d"),
2196 output_invalid (*l),
2197 i.operands + 1);
2198 return NULL;
2199 }
2200 if (!intel_syntax)
2201 {
2202 if (*l == '(')
2203 ++paren_not_balanced;
2204 if (*l == ')')
2205 --paren_not_balanced;
2206 }
2207 else
2208 {
2209 if (*l == '[')
2210 ++paren_not_balanced;
2211 if (*l == ']')
2212 --paren_not_balanced;
2213 }
2214 l++;
2215 }
2216 if (l != token_start)
2217 { /* Yes, we've read in another operand. */
2218 unsigned int operand_ok;
2219 this_operand = i.operands++;
2220 if (i.operands > MAX_OPERANDS)
2221 {
2222 as_bad (_("spurious operands; (%d operands/instruction max)"),
2223 MAX_OPERANDS);
2224 return NULL;
2225 }
2226 /* Now parse operand adding info to 'i' as we go along. */
2227 END_STRING_AND_SAVE (l);
2228
2229 if (intel_syntax)
2230 operand_ok =
2231 i386_intel_operand (token_start,
2232 intel_float_operand (mnemonic));
2233 else
2234 operand_ok = i386_operand (token_start);
2235
2236 RESTORE_END_STRING (l);
2237 if (!operand_ok)
2238 return NULL;
2239 }
2240 else
2241 {
2242 if (expecting_operand)
2243 {
2244 expecting_operand_after_comma:
2245 as_bad (_("expecting operand after ','; got nothing"));
2246 return NULL;
2247 }
2248 if (*l == ',')
2249 {
2250 as_bad (_("expecting operand before ','; got nothing"));
2251 return NULL;
2252 }
2253 }
2254
2255 /* Now *l must be either ',' or END_OF_INSN. */
2256 if (*l == ',')
2257 {
2258 if (*++l == END_OF_INSN)
2259 {
2260 /* Just skip it, if it's \n complain. */
2261 goto expecting_operand_after_comma;
2262 }
2263 expecting_operand = 1;
2264 }
2265 }
2266 return l;
2267 }
2268
2269 static void
2270 swap_2_operands (int xchg1, int xchg2)
2271 {
2272 union i386_op temp_op;
2273 unsigned int temp_type;
2274 enum bfd_reloc_code_real temp_reloc;
2275
2276 temp_type = i.types[xchg2];
2277 i.types[xchg2] = i.types[xchg1];
2278 i.types[xchg1] = temp_type;
2279 temp_op = i.op[xchg2];
2280 i.op[xchg2] = i.op[xchg1];
2281 i.op[xchg1] = temp_op;
2282 temp_reloc = i.reloc[xchg2];
2283 i.reloc[xchg2] = i.reloc[xchg1];
2284 i.reloc[xchg1] = temp_reloc;
2285 }
2286
2287 static void
2288 swap_operands (void)
2289 {
2290 switch (i.operands)
2291 {
2292 case 4:
2293 swap_2_operands (1, i.operands - 2);
2294 case 3:
2295 case 2:
2296 swap_2_operands (0, i.operands - 1);
2297 break;
2298 default:
2299 abort ();
2300 }
2301
2302 if (i.mem_operands == 2)
2303 {
2304 const seg_entry *temp_seg;
2305 temp_seg = i.seg[0];
2306 i.seg[0] = i.seg[1];
2307 i.seg[1] = temp_seg;
2308 }
2309 }
2310
2311 /* Try to ensure constant immediates are represented in the smallest
2312 opcode possible. */
2313 static void
2314 optimize_imm (void)
2315 {
2316 char guess_suffix = 0;
2317 int op;
2318
2319 if (i.suffix)
2320 guess_suffix = i.suffix;
2321 else if (i.reg_operands)
2322 {
2323 /* Figure out a suffix from the last register operand specified.
2324 We can't do this properly yet, ie. excluding InOutPortReg,
2325 but the following works for instructions with immediates.
2326 In any case, we can't set i.suffix yet. */
2327 for (op = i.operands; --op >= 0;)
2328 if (i.types[op] & Reg)
2329 {
2330 if (i.types[op] & Reg8)
2331 guess_suffix = BYTE_MNEM_SUFFIX;
2332 else if (i.types[op] & Reg16)
2333 guess_suffix = WORD_MNEM_SUFFIX;
2334 else if (i.types[op] & Reg32)
2335 guess_suffix = LONG_MNEM_SUFFIX;
2336 else if (i.types[op] & Reg64)
2337 guess_suffix = QWORD_MNEM_SUFFIX;
2338 break;
2339 }
2340 }
2341 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
2342 guess_suffix = WORD_MNEM_SUFFIX;
2343
2344 for (op = i.operands; --op >= 0;)
2345 if (i.types[op] & Imm)
2346 {
2347 switch (i.op[op].imms->X_op)
2348 {
2349 case O_constant:
2350 /* If a suffix is given, this operand may be shortened. */
2351 switch (guess_suffix)
2352 {
2353 case LONG_MNEM_SUFFIX:
2354 i.types[op] |= Imm32 | Imm64;
2355 break;
2356 case WORD_MNEM_SUFFIX:
2357 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
2358 break;
2359 case BYTE_MNEM_SUFFIX:
2360 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
2361 break;
2362 }
2363
2364 /* If this operand is at most 16 bits, convert it
2365 to a signed 16 bit number before trying to see
2366 whether it will fit in an even smaller size.
2367 This allows a 16-bit operand such as $0xffe0 to
2368 be recognised as within Imm8S range. */
2369 if ((i.types[op] & Imm16)
2370 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
2371 {
2372 i.op[op].imms->X_add_number =
2373 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
2374 }
2375 if ((i.types[op] & Imm32)
2376 && ((i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1))
2377 == 0))
2378 {
2379 i.op[op].imms->X_add_number = ((i.op[op].imms->X_add_number
2380 ^ ((offsetT) 1 << 31))
2381 - ((offsetT) 1 << 31));
2382 }
2383 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
2384
2385 /* We must avoid matching of Imm32 templates when 64bit
2386 only immediate is available. */
2387 if (guess_suffix == QWORD_MNEM_SUFFIX)
2388 i.types[op] &= ~Imm32;
2389 break;
2390
2391 case O_absent:
2392 case O_register:
2393 abort ();
2394
2395 /* Symbols and expressions. */
2396 default:
2397 /* Convert symbolic operand to proper sizes for matching, but don't
2398 prevent matching a set of insns that only supports sizes other
2399 than those matching the insn suffix. */
2400 {
2401 unsigned int mask, allowed = 0;
2402 const template *t;
2403
2404 for (t = current_templates->start;
2405 t < current_templates->end;
2406 ++t)
2407 allowed |= t->operand_types[op];
2408 switch (guess_suffix)
2409 {
2410 case QWORD_MNEM_SUFFIX:
2411 mask = Imm64 | Imm32S;
2412 break;
2413 case LONG_MNEM_SUFFIX:
2414 mask = Imm32;
2415 break;
2416 case WORD_MNEM_SUFFIX:
2417 mask = Imm16;
2418 break;
2419 case BYTE_MNEM_SUFFIX:
2420 mask = Imm8;
2421 break;
2422 default:
2423 mask = 0;
2424 break;
2425 }
2426 if (mask & allowed)
2427 i.types[op] &= mask;
2428 }
2429 break;
2430 }
2431 }
2432 }
2433
2434 /* Try to use the smallest displacement type too. */
2435 static void
2436 optimize_disp (void)
2437 {
2438 int op;
2439
2440 for (op = i.operands; --op >= 0;)
2441 if (i.types[op] & Disp)
2442 {
2443 if (i.op[op].disps->X_op == O_constant)
2444 {
2445 offsetT disp = i.op[op].disps->X_add_number;
2446
2447 if ((i.types[op] & Disp16)
2448 && (disp & ~(offsetT) 0xffff) == 0)
2449 {
2450 /* If this operand is at most 16 bits, convert
2451 to a signed 16 bit number and don't use 64bit
2452 displacement. */
2453 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
2454 i.types[op] &= ~Disp64;
2455 }
2456 if ((i.types[op] & Disp32)
2457 && (disp & ~(((offsetT) 2 << 31) - 1)) == 0)
2458 {
2459 /* If this operand is at most 32 bits, convert
2460 to a signed 32 bit number and don't use 64bit
2461 displacement. */
2462 disp &= (((offsetT) 2 << 31) - 1);
2463 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
2464 i.types[op] &= ~Disp64;
2465 }
2466 if (!disp && (i.types[op] & BaseIndex))
2467 {
2468 i.types[op] &= ~Disp;
2469 i.op[op].disps = 0;
2470 i.disp_operands--;
2471 }
2472 else if (flag_code == CODE_64BIT)
2473 {
2474 if (fits_in_signed_long (disp))
2475 {
2476 i.types[op] &= ~Disp64;
2477 i.types[op] |= Disp32S;
2478 }
2479 if (fits_in_unsigned_long (disp))
2480 i.types[op] |= Disp32;
2481 }
2482 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
2483 && fits_in_signed_byte (disp))
2484 i.types[op] |= Disp8;
2485 }
2486 else if (i.reloc[op] == BFD_RELOC_386_TLS_DESC_CALL
2487 || i.reloc[op] == BFD_RELOC_X86_64_TLSDESC_CALL)
2488 {
2489 fix_new_exp (frag_now, frag_more (0) - frag_now->fr_literal, 0,
2490 i.op[op].disps, 0, i.reloc[op]);
2491 i.types[op] &= ~Disp;
2492 }
2493 else
2494 /* We only support 64bit displacement on constants. */
2495 i.types[op] &= ~Disp64;
2496 }
2497 }
2498
2499 static int
2500 match_template (void)
2501 {
2502 /* Points to template once we've found it. */
2503 const template *t;
2504 unsigned int overlap0, overlap1, overlap2, overlap3;
2505 unsigned int found_reverse_match;
2506 int suffix_check;
2507 unsigned int operand_types [MAX_OPERANDS];
2508 int addr_prefix_disp;
2509 unsigned int j;
2510
2511 #if MAX_OPERANDS != 4
2512 # error "MAX_OPERANDS must be 4."
2513 #endif
2514
2515 #define MATCH(overlap, given, template) \
2516 ((overlap & ~JumpAbsolute) \
2517 && (((given) & (BaseIndex | JumpAbsolute)) \
2518 == ((overlap) & (BaseIndex | JumpAbsolute))))
2519
2520 /* If given types r0 and r1 are registers they must be of the same type
2521 unless the expected operand type register overlap is null.
2522 Note that Acc in a template matches every size of reg. */
2523 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
2524 (((g0) & Reg) == 0 || ((g1) & Reg) == 0 \
2525 || ((g0) & Reg) == ((g1) & Reg) \
2526 || ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
2527
2528 overlap0 = 0;
2529 overlap1 = 0;
2530 overlap2 = 0;
2531 overlap3 = 0;
2532 found_reverse_match = 0;
2533 for (j = 0; j < MAX_OPERANDS; j++)
2534 operand_types [j] = 0;
2535 addr_prefix_disp = -1;
2536 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
2537 ? No_bSuf
2538 : (i.suffix == WORD_MNEM_SUFFIX
2539 ? No_wSuf
2540 : (i.suffix == SHORT_MNEM_SUFFIX
2541 ? No_sSuf
2542 : (i.suffix == LONG_MNEM_SUFFIX
2543 ? No_lSuf
2544 : (i.suffix == QWORD_MNEM_SUFFIX
2545 ? No_qSuf
2546 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX
2547 ? No_xSuf : 0))))));
2548
2549 for (t = current_templates->start; t < current_templates->end; t++)
2550 {
2551 addr_prefix_disp = -1;
2552
2553 /* Must have right number of operands. */
2554 if (i.operands != t->operands)
2555 continue;
2556
2557 /* Check the suffix, except for some instructions in intel mode. */
2558 if ((t->opcode_modifier & suffix_check)
2559 && !(intel_syntax
2560 && (t->opcode_modifier & IgnoreSize)))
2561 continue;
2562
2563 for (j = 0; j < MAX_OPERANDS; j++)
2564 operand_types [j] = t->operand_types [j];
2565
2566 /* In general, don't allow 64-bit operands in 32-bit mode. */
2567 if (i.suffix == QWORD_MNEM_SUFFIX
2568 && flag_code != CODE_64BIT
2569 && (intel_syntax
2570 ? (!(t->opcode_modifier & IgnoreSize)
2571 && !intel_float_operand (t->name))
2572 : intel_float_operand (t->name) != 2)
2573 && (!(operand_types[0] & (RegMMX | RegXMM))
2574 || !(operand_types[t->operands > 1] & (RegMMX | RegXMM)))
2575 && (t->base_opcode != 0x0fc7
2576 || t->extension_opcode != 1 /* cmpxchg8b */))
2577 continue;
2578
2579 /* Do not verify operands when there are none. */
2580 else if (!t->operands)
2581 {
2582 if (t->cpu_flags & ~cpu_arch_flags)
2583 continue;
2584 /* We've found a match; break out of loop. */
2585 break;
2586 }
2587
2588 /* Address size prefix will turn Disp64/Disp32/Disp16 operand
2589 into Disp32/Disp16/Disp32 operand. */
2590 if (i.prefix[ADDR_PREFIX] != 0)
2591 {
2592 unsigned int DispOn = 0, DispOff = 0;
2593
2594 switch (flag_code)
2595 {
2596 case CODE_16BIT:
2597 DispOn = Disp32;
2598 DispOff = Disp16;
2599 break;
2600 case CODE_32BIT:
2601 DispOn = Disp16;
2602 DispOff = Disp32;
2603 break;
2604 case CODE_64BIT:
2605 DispOn = Disp32;
2606 DispOff = Disp64;
2607 break;
2608 }
2609
2610 for (j = 0; j < MAX_OPERANDS; j++)
2611 {
2612 /* There should be only one Disp operand. */
2613 if ((operand_types[j] & DispOff))
2614 {
2615 addr_prefix_disp = j;
2616 operand_types[j] |= DispOn;
2617 operand_types[j] &= ~DispOff;
2618 break;
2619 }
2620 }
2621 }
2622
2623 overlap0 = i.types[0] & operand_types[0];
2624 switch (t->operands)
2625 {
2626 case 1:
2627 if (!MATCH (overlap0, i.types[0], operand_types[0]))
2628 continue;
2629 break;
2630 case 2:
2631 /* xchg %eax, %eax is a special case. It is an aliase for nop
2632 only in 32bit mode and we can use opcode 0x90. In 64bit
2633 mode, we can't use 0x90 for xchg %eax, %eax since it should
2634 zero-extend %eax to %rax. */
2635 if (flag_code == CODE_64BIT
2636 && t->base_opcode == 0x90
2637 && i.types [0] == (Acc | Reg32)
2638 && i.types [1] == (Acc | Reg32))
2639 continue;
2640 case 3:
2641 case 4:
2642 overlap1 = i.types[1] & operand_types[1];
2643 if (!MATCH (overlap0, i.types[0], operand_types[0])
2644 || !MATCH (overlap1, i.types[1], operand_types[1])
2645 /* monitor in SSE3 is a very special case. The first
2646 register and the second register may have different
2647 sizes. The same applies to crc32 in SSE4.2. */
2648 || !((t->base_opcode == 0x0f01
2649 && t->extension_opcode == 0xc8)
2650 || t->base_opcode == 0xf20f38f1
2651 || CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2652 operand_types[0],
2653 overlap1, i.types[1],
2654 operand_types[1])))
2655 {
2656 /* Check if other direction is valid ... */
2657 if ((t->opcode_modifier & (D | FloatD)) == 0)
2658 continue;
2659
2660 /* Try reversing direction of operands. */
2661 overlap0 = i.types[0] & operand_types[1];
2662 overlap1 = i.types[1] & operand_types[0];
2663 if (!MATCH (overlap0, i.types[0], operand_types[1])
2664 || !MATCH (overlap1, i.types[1], operand_types[0])
2665 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2666 operand_types[1],
2667 overlap1, i.types[1],
2668 operand_types[0]))
2669 {
2670 /* Does not match either direction. */
2671 continue;
2672 }
2673 /* found_reverse_match holds which of D or FloatDR
2674 we've found. */
2675 if ((t->opcode_modifier & D))
2676 found_reverse_match = Opcode_D;
2677 else if ((t->opcode_modifier & FloatD))
2678 found_reverse_match = Opcode_FloatD;
2679 else
2680 found_reverse_match = 0;
2681 if ((t->opcode_modifier & FloatR))
2682 found_reverse_match |= Opcode_FloatR;
2683 }
2684 else
2685 {
2686 /* Found a forward 2 operand match here. */
2687 switch (t->operands)
2688 {
2689 case 4:
2690 overlap3 = i.types[3] & operand_types[3];
2691 case 3:
2692 overlap2 = i.types[2] & operand_types[2];
2693 break;
2694 }
2695
2696 switch (t->operands)
2697 {
2698 case 4:
2699 if (!MATCH (overlap3, i.types[3], operand_types[3])
2700 || !CONSISTENT_REGISTER_MATCH (overlap2,
2701 i.types[2],
2702 operand_types[2],
2703 overlap3,
2704 i.types[3],
2705 operand_types[3]))
2706 continue;
2707 case 3:
2708 /* Here we make use of the fact that there are no
2709 reverse match 3 operand instructions, and all 3
2710 operand instructions only need to be checked for
2711 register consistency between operands 2 and 3. */
2712 if (!MATCH (overlap2, i.types[2], operand_types[2])
2713 || !CONSISTENT_REGISTER_MATCH (overlap1,
2714 i.types[1],
2715 operand_types[1],
2716 overlap2,
2717 i.types[2],
2718 operand_types[2]))
2719 continue;
2720 break;
2721 }
2722 }
2723 /* Found either forward/reverse 2, 3 or 4 operand match here:
2724 slip through to break. */
2725 }
2726 if (t->cpu_flags & ~cpu_arch_flags)
2727 {
2728 found_reverse_match = 0;
2729 continue;
2730 }
2731 /* We've found a match; break out of loop. */
2732 break;
2733 }
2734
2735 if (t == current_templates->end)
2736 {
2737 /* We found no match. */
2738 as_bad (_("suffix or operands invalid for `%s'"),
2739 current_templates->start->name);
2740 return 0;
2741 }
2742
2743 if (!quiet_warnings)
2744 {
2745 if (!intel_syntax
2746 && ((i.types[0] & JumpAbsolute)
2747 != (operand_types[0] & JumpAbsolute)))
2748 {
2749 as_warn (_("indirect %s without `*'"), t->name);
2750 }
2751
2752 if ((t->opcode_modifier & (IsPrefix | IgnoreSize))
2753 == (IsPrefix | IgnoreSize))
2754 {
2755 /* Warn them that a data or address size prefix doesn't
2756 affect assembly of the next line of code. */
2757 as_warn (_("stand-alone `%s' prefix"), t->name);
2758 }
2759 }
2760
2761 /* Copy the template we found. */
2762 i.tm = *t;
2763
2764 if (addr_prefix_disp != -1)
2765 i.tm.operand_types[addr_prefix_disp]
2766 = operand_types[addr_prefix_disp];
2767
2768 if (found_reverse_match)
2769 {
2770 /* If we found a reverse match we must alter the opcode
2771 direction bit. found_reverse_match holds bits to change
2772 (different for int & float insns). */
2773
2774 i.tm.base_opcode ^= found_reverse_match;
2775
2776 i.tm.operand_types[0] = operand_types[1];
2777 i.tm.operand_types[1] = operand_types[0];
2778 }
2779
2780 return 1;
2781 }
2782
2783 static int
2784 check_string (void)
2785 {
2786 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
2787 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
2788 {
2789 if (i.seg[0] != NULL && i.seg[0] != &es)
2790 {
2791 as_bad (_("`%s' operand %d must use `%%es' segment"),
2792 i.tm.name,
2793 mem_op + 1);
2794 return 0;
2795 }
2796 /* There's only ever one segment override allowed per instruction.
2797 This instruction possibly has a legal segment override on the
2798 second operand, so copy the segment to where non-string
2799 instructions store it, allowing common code. */
2800 i.seg[0] = i.seg[1];
2801 }
2802 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
2803 {
2804 if (i.seg[1] != NULL && i.seg[1] != &es)
2805 {
2806 as_bad (_("`%s' operand %d must use `%%es' segment"),
2807 i.tm.name,
2808 mem_op + 2);
2809 return 0;
2810 }
2811 }
2812 return 1;
2813 }
2814
2815 static int
2816 process_suffix (void)
2817 {
2818 /* If matched instruction specifies an explicit instruction mnemonic
2819 suffix, use it. */
2820 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
2821 {
2822 if (i.tm.opcode_modifier & Size16)
2823 i.suffix = WORD_MNEM_SUFFIX;
2824 else if (i.tm.opcode_modifier & Size64)
2825 i.suffix = QWORD_MNEM_SUFFIX;
2826 else
2827 i.suffix = LONG_MNEM_SUFFIX;
2828 }
2829 else if (i.reg_operands)
2830 {
2831 /* If there's no instruction mnemonic suffix we try to invent one
2832 based on register operands. */
2833 if (!i.suffix)
2834 {
2835 /* We take i.suffix from the last register operand specified,
2836 Destination register type is more significant than source
2837 register type. crc32 in SSE4.2 prefers source register
2838 type. */
2839 if (i.tm.base_opcode == 0xf20f38f1)
2840 {
2841 if ((i.types[0] & Reg))
2842 i.suffix = ((i.types[0] & Reg16) ? WORD_MNEM_SUFFIX :
2843 LONG_MNEM_SUFFIX);
2844 }
2845
2846 if (!i.suffix)
2847 {
2848 int op;
2849
2850 for (op = i.operands; --op >= 0;)
2851 if ((i.types[op] & Reg)
2852 && !(i.tm.operand_types[op] & InOutPortReg))
2853 {
2854 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
2855 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
2856 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
2857 LONG_MNEM_SUFFIX);
2858 break;
2859 }
2860 }
2861 }
2862 else if (i.suffix == BYTE_MNEM_SUFFIX)
2863 {
2864 if (!check_byte_reg ())
2865 return 0;
2866 }
2867 else if (i.suffix == LONG_MNEM_SUFFIX)
2868 {
2869 if (!check_long_reg ())
2870 return 0;
2871 }
2872 else if (i.suffix == QWORD_MNEM_SUFFIX)
2873 {
2874 if (!check_qword_reg ())
2875 return 0;
2876 }
2877 else if (i.suffix == WORD_MNEM_SUFFIX)
2878 {
2879 if (!check_word_reg ())
2880 return 0;
2881 }
2882 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2883 /* Do nothing if the instruction is going to ignore the prefix. */
2884 ;
2885 else
2886 abort ();
2887 }
2888 else if ((i.tm.opcode_modifier & DefaultSize)
2889 && !i.suffix
2890 /* exclude fldenv/frstor/fsave/fstenv */
2891 && (i.tm.opcode_modifier & No_sSuf))
2892 {
2893 i.suffix = stackop_size;
2894 }
2895 else if (intel_syntax
2896 && !i.suffix
2897 && ((i.tm.operand_types[0] & JumpAbsolute)
2898 || (i.tm.opcode_modifier & (JumpByte|JumpInterSegment))
2899 || (i.tm.base_opcode == 0x0f01 /* [ls][gi]dt */
2900 && i.tm.extension_opcode <= 3)))
2901 {
2902 switch (flag_code)
2903 {
2904 case CODE_64BIT:
2905 if (!(i.tm.opcode_modifier & No_qSuf))
2906 {
2907 i.suffix = QWORD_MNEM_SUFFIX;
2908 break;
2909 }
2910 case CODE_32BIT:
2911 if (!(i.tm.opcode_modifier & No_lSuf))
2912 i.suffix = LONG_MNEM_SUFFIX;
2913 break;
2914 case CODE_16BIT:
2915 if (!(i.tm.opcode_modifier & No_wSuf))
2916 i.suffix = WORD_MNEM_SUFFIX;
2917 break;
2918 }
2919 }
2920
2921 if (!i.suffix)
2922 {
2923 if (!intel_syntax)
2924 {
2925 if (i.tm.opcode_modifier & W)
2926 {
2927 as_bad (_("no instruction mnemonic suffix given and "
2928 "no register operands; can't size instruction"));
2929 return 0;
2930 }
2931 }
2932 else
2933 {
2934 unsigned int suffixes = (~i.tm.opcode_modifier
2935 & (No_bSuf
2936 | No_wSuf
2937 | No_lSuf
2938 | No_sSuf
2939 | No_xSuf
2940 | No_qSuf));
2941
2942 if ((i.tm.opcode_modifier & W)
2943 || ((suffixes & (suffixes - 1))
2944 && !(i.tm.opcode_modifier & (DefaultSize | IgnoreSize))))
2945 {
2946 as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
2947 return 0;
2948 }
2949 }
2950 }
2951
2952 /* Change the opcode based on the operand size given by i.suffix;
2953 We don't need to change things for byte insns. */
2954
2955 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2956 {
2957 /* It's not a byte, select word/dword operation. */
2958 if (i.tm.opcode_modifier & W)
2959 {
2960 if (i.tm.opcode_modifier & ShortForm)
2961 i.tm.base_opcode |= 8;
2962 else
2963 i.tm.base_opcode |= 1;
2964 }
2965
2966 /* Now select between word & dword operations via the operand
2967 size prefix, except for instructions that will ignore this
2968 prefix anyway. */
2969 if (i.tm.base_opcode == 0x0f01 && i.tm.extension_opcode == 0xc8)
2970 {
2971 /* monitor in SSE3 is a very special case. The default size
2972 of AX is the size of mode. The address size override
2973 prefix will change the size of AX. */
2974 if (i.op->regs[0].reg_type &
2975 (flag_code == CODE_32BIT ? Reg16 : Reg32))
2976 if (!add_prefix (ADDR_PREFIX_OPCODE))
2977 return 0;
2978 }
2979 else if (i.suffix != QWORD_MNEM_SUFFIX
2980 && i.suffix != LONG_DOUBLE_MNEM_SUFFIX
2981 && !(i.tm.opcode_modifier & (IgnoreSize | FloatMF))
2982 && ((i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2983 || (flag_code == CODE_64BIT
2984 && (i.tm.opcode_modifier & JumpByte))))
2985 {
2986 unsigned int prefix = DATA_PREFIX_OPCODE;
2987
2988 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2989 prefix = ADDR_PREFIX_OPCODE;
2990
2991 if (!add_prefix (prefix))
2992 return 0;
2993 }
2994
2995 /* Set mode64 for an operand. */
2996 if (i.suffix == QWORD_MNEM_SUFFIX
2997 && flag_code == CODE_64BIT
2998 && (i.tm.opcode_modifier & NoRex64) == 0)
2999 {
3000 /* Special case for xchg %rax,%rax. It is NOP and doesn't
3001 need rex64. */
3002 if (i.operands != 2
3003 || i.types [0] != (Acc | Reg64)
3004 || i.types [1] != (Acc | Reg64)
3005 || i.tm.base_opcode != 0x90)
3006 i.rex |= REX_W;
3007 }
3008
3009 /* Size floating point instruction. */
3010 if (i.suffix == LONG_MNEM_SUFFIX)
3011 if (i.tm.opcode_modifier & FloatMF)
3012 i.tm.base_opcode ^= 4;
3013 }
3014
3015 return 1;
3016 }
3017
3018 static int
3019 check_byte_reg (void)
3020 {
3021 int op;
3022
3023 for (op = i.operands; --op >= 0;)
3024 {
3025 /* If this is an eight bit register, it's OK. If it's the 16 or
3026 32 bit version of an eight bit register, we will just use the
3027 low portion, and that's OK too. */
3028 if (i.types[op] & Reg8)
3029 continue;
3030
3031 /* movzx and movsx should not generate this warning. */
3032 if (intel_syntax
3033 && (i.tm.base_opcode == 0xfb7
3034 || i.tm.base_opcode == 0xfb6
3035 || i.tm.base_opcode == 0x63
3036 || i.tm.base_opcode == 0xfbe
3037 || i.tm.base_opcode == 0xfbf))
3038 continue;
3039
3040 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4)
3041 {
3042 /* Prohibit these changes in the 64bit mode, since the
3043 lowering is more complicated. */
3044 if (flag_code == CODE_64BIT
3045 && (i.tm.operand_types[op] & InOutPortReg) == 0)
3046 {
3047 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3048 register_prefix, i.op[op].regs->reg_name,
3049 i.suffix);
3050 return 0;
3051 }
3052 #if REGISTER_WARNINGS
3053 if (!quiet_warnings
3054 && (i.tm.operand_types[op] & InOutPortReg) == 0)
3055 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
3056 (i.op[op].regs + (i.types[op] & Reg16
3057 ? REGNAM_AL - REGNAM_AX
3058 : REGNAM_AL - REGNAM_EAX))->reg_name,
3059 i.op[op].regs->reg_name,
3060 i.suffix);
3061 #endif
3062 continue;
3063 }
3064 /* Any other register is bad. */
3065 if (i.types[op] & (Reg | RegMMX | RegXMM
3066 | SReg2 | SReg3
3067 | Control | Debug | Test
3068 | FloatReg | FloatAcc))
3069 {
3070 as_bad (_("`%%%s' not allowed with `%s%c'"),
3071 i.op[op].regs->reg_name,
3072 i.tm.name,
3073 i.suffix);
3074 return 0;
3075 }
3076 }
3077 return 1;
3078 }
3079
3080 static int
3081 check_long_reg (void)
3082 {
3083 int op;
3084
3085 for (op = i.operands; --op >= 0;)
3086 /* Reject eight bit registers, except where the template requires
3087 them. (eg. movzb) */
3088 if ((i.types[op] & Reg8) != 0
3089 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3090 {
3091 as_bad (_("`%%%s' not allowed with `%s%c'"),
3092 i.op[op].regs->reg_name,
3093 i.tm.name,
3094 i.suffix);
3095 return 0;
3096 }
3097 /* Warn if the e prefix on a general reg is missing. */
3098 else if ((!quiet_warnings || flag_code == CODE_64BIT)
3099 && (i.types[op] & Reg16) != 0
3100 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3101 {
3102 /* Prohibit these changes in the 64bit mode, since the
3103 lowering is more complicated. */
3104 if (flag_code == CODE_64BIT)
3105 {
3106 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3107 register_prefix, i.op[op].regs->reg_name,
3108 i.suffix);
3109 return 0;
3110 }
3111 #if REGISTER_WARNINGS
3112 else
3113 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
3114 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
3115 i.op[op].regs->reg_name,
3116 i.suffix);
3117 #endif
3118 }
3119 /* Warn if the r prefix on a general reg is missing. */
3120 else if ((i.types[op] & Reg64) != 0
3121 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3122 {
3123 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3124 register_prefix, i.op[op].regs->reg_name,
3125 i.suffix);
3126 return 0;
3127 }
3128 return 1;
3129 }
3130
3131 static int
3132 check_qword_reg (void)
3133 {
3134 int op;
3135
3136 for (op = i.operands; --op >= 0; )
3137 /* Reject eight bit registers, except where the template requires
3138 them. (eg. movzb) */
3139 if ((i.types[op] & Reg8) != 0
3140 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3141 {
3142 as_bad (_("`%%%s' not allowed with `%s%c'"),
3143 i.op[op].regs->reg_name,
3144 i.tm.name,
3145 i.suffix);
3146 return 0;
3147 }
3148 /* Warn if the e prefix on a general reg is missing. */
3149 else if (((i.types[op] & Reg16) != 0
3150 || (i.types[op] & Reg32) != 0)
3151 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3152 {
3153 /* Prohibit these changes in the 64bit mode, since the
3154 lowering is more complicated. */
3155 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3156 register_prefix, i.op[op].regs->reg_name,
3157 i.suffix);
3158 return 0;
3159 }
3160 return 1;
3161 }
3162
3163 static int
3164 check_word_reg (void)
3165 {
3166 int op;
3167 for (op = i.operands; --op >= 0;)
3168 /* Reject eight bit registers, except where the template requires
3169 them. (eg. movzb) */
3170 if ((i.types[op] & Reg8) != 0
3171 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3172 {
3173 as_bad (_("`%%%s' not allowed with `%s%c'"),
3174 i.op[op].regs->reg_name,
3175 i.tm.name,
3176 i.suffix);
3177 return 0;
3178 }
3179 /* Warn if the e prefix on a general reg is present. */
3180 else if ((!quiet_warnings || flag_code == CODE_64BIT)
3181 && (i.types[op] & Reg32) != 0
3182 && (i.tm.operand_types[op] & (Reg16 | Acc)) != 0)
3183 {
3184 /* Prohibit these changes in the 64bit mode, since the
3185 lowering is more complicated. */
3186 if (flag_code == CODE_64BIT)
3187 {
3188 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3189 register_prefix, i.op[op].regs->reg_name,
3190 i.suffix);
3191 return 0;
3192 }
3193 else
3194 #if REGISTER_WARNINGS
3195 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
3196 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
3197 i.op[op].regs->reg_name,
3198 i.suffix);
3199 #endif
3200 }
3201 return 1;
3202 }
3203
3204 static int
3205 finalize_imm (void)
3206 {
3207 unsigned int overlap0, overlap1, overlap2;
3208
3209 overlap0 = i.types[0] & i.tm.operand_types[0];
3210 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64))
3211 && overlap0 != Imm8 && overlap0 != Imm8S
3212 && overlap0 != Imm16 && overlap0 != Imm32S
3213 && overlap0 != Imm32 && overlap0 != Imm64)
3214 {
3215 if (i.suffix)
3216 {
3217 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX
3218 ? Imm8 | Imm8S
3219 : (i.suffix == WORD_MNEM_SUFFIX
3220 ? Imm16
3221 : (i.suffix == QWORD_MNEM_SUFFIX
3222 ? Imm64 | Imm32S
3223 : Imm32)));
3224 }
3225 else if (overlap0 == (Imm16 | Imm32S | Imm32)
3226 || overlap0 == (Imm16 | Imm32)
3227 || overlap0 == (Imm16 | Imm32S))
3228 {
3229 overlap0 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
3230 ? Imm16 : Imm32S);
3231 }
3232 if (overlap0 != Imm8 && overlap0 != Imm8S
3233 && overlap0 != Imm16 && overlap0 != Imm32S
3234 && overlap0 != Imm32 && overlap0 != Imm64)
3235 {
3236 as_bad (_("no instruction mnemonic suffix given; "
3237 "can't determine immediate size"));
3238 return 0;
3239 }
3240 }
3241 i.types[0] = overlap0;
3242
3243 overlap1 = i.types[1] & i.tm.operand_types[1];
3244 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32 | Imm64))
3245 && overlap1 != Imm8 && overlap1 != Imm8S
3246 && overlap1 != Imm16 && overlap1 != Imm32S
3247 && overlap1 != Imm32 && overlap1 != Imm64)
3248 {
3249 if (i.suffix)
3250 {
3251 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX
3252 ? Imm8 | Imm8S
3253 : (i.suffix == WORD_MNEM_SUFFIX
3254 ? Imm16
3255 : (i.suffix == QWORD_MNEM_SUFFIX
3256 ? Imm64 | Imm32S
3257 : Imm32)));
3258 }
3259 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
3260 || overlap1 == (Imm16 | Imm32)
3261 || overlap1 == (Imm16 | Imm32S))
3262 {
3263 overlap1 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
3264 ? Imm16 : Imm32S);
3265 }
3266 if (overlap1 != Imm8 && overlap1 != Imm8S
3267 && overlap1 != Imm16 && overlap1 != Imm32S
3268 && overlap1 != Imm32 && overlap1 != Imm64)
3269 {
3270 as_bad (_("no instruction mnemonic suffix given; "
3271 "can't determine immediate size %x %c"),
3272 overlap1, i.suffix);
3273 return 0;
3274 }
3275 }
3276 i.types[1] = overlap1;
3277
3278 overlap2 = i.types[2] & i.tm.operand_types[2];
3279 assert ((overlap2 & Imm) == 0);
3280 i.types[2] = overlap2;
3281
3282 return 1;
3283 }
3284
3285 static int
3286 process_operands (void)
3287 {
3288 /* Default segment register this instruction will use for memory
3289 accesses. 0 means unknown. This is only for optimizing out
3290 unnecessary segment overrides. */
3291 const seg_entry *default_seg = 0;
3292
3293 /* The imul $imm, %reg instruction is converted into
3294 imul $imm, %reg, %reg, and the clr %reg instruction
3295 is converted into xor %reg, %reg. */
3296 if (i.tm.opcode_modifier & regKludge)
3297 {
3298 if ((i.tm.cpu_flags & CpuSSE4_1))
3299 {
3300 /* The first operand in instruction blendvpd, blendvps and
3301 pblendvb in SSE4.1 is implicit and must be xmm0. */
3302 assert (i.operands == 3
3303 && i.reg_operands >= 2
3304 && i.types[0] == RegXMM);
3305 if (i.op[0].regs->reg_num != 0)
3306 {
3307 if (intel_syntax)
3308 as_bad (_("the last operand of `%s' must be `%sxmm0'"),
3309 i.tm.name, register_prefix);
3310 else
3311 as_bad (_("the first operand of `%s' must be `%sxmm0'"),
3312 i.tm.name, register_prefix);
3313 return 0;
3314 }
3315 i.op[0] = i.op[1];
3316 i.op[1] = i.op[2];
3317 i.types[0] = i.types[1];
3318 i.types[1] = i.types[2];
3319 i.operands--;
3320 i.reg_operands--;
3321
3322 /* We need to adjust fields in i.tm since they are used by
3323 build_modrm_byte. */
3324 i.tm.operand_types [0] = i.tm.operand_types [1];
3325 i.tm.operand_types [1] = i.tm.operand_types [2];
3326 i.tm.operands--;
3327 }
3328 else
3329 {
3330 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
3331 /* Pretend we saw the extra register operand. */
3332 assert (i.reg_operands == 1
3333 && i.op[first_reg_op + 1].regs == 0);
3334 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
3335 i.types[first_reg_op + 1] = i.types[first_reg_op];
3336 i.operands++;
3337 i.reg_operands++;
3338 }
3339 }
3340
3341 if (i.tm.opcode_modifier & ShortForm)
3342 {
3343 if (i.types[0] & (SReg2 | SReg3))
3344 {
3345 if (i.tm.base_opcode == POP_SEG_SHORT
3346 && i.op[0].regs->reg_num == 1)
3347 {
3348 as_bad (_("you can't `pop %%cs'"));
3349 return 0;
3350 }
3351 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
3352 if ((i.op[0].regs->reg_flags & RegRex) != 0)
3353 i.rex |= REX_B;
3354 }
3355 else
3356 {
3357 /* The register or float register operand is in operand 0 or 1. */
3358 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
3359 /* Register goes in low 3 bits of opcode. */
3360 i.tm.base_opcode |= i.op[op].regs->reg_num;
3361 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3362 i.rex |= REX_B;
3363 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
3364 {
3365 /* Warn about some common errors, but press on regardless.
3366 The first case can be generated by gcc (<= 2.8.1). */
3367 if (i.operands == 2)
3368 {
3369 /* Reversed arguments on faddp, fsubp, etc. */
3370 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
3371 i.op[1].regs->reg_name,
3372 i.op[0].regs->reg_name);
3373 }
3374 else
3375 {
3376 /* Extraneous `l' suffix on fp insn. */
3377 as_warn (_("translating to `%s %%%s'"), i.tm.name,
3378 i.op[0].regs->reg_name);
3379 }
3380 }
3381 }
3382 }
3383 else if (i.tm.opcode_modifier & Modrm)
3384 {
3385 /* The opcode is completed (modulo i.tm.extension_opcode which
3386 must be put into the modrm byte). Now, we make the modrm and
3387 index base bytes based on all the info we've collected. */
3388
3389 default_seg = build_modrm_byte ();
3390 }
3391 else if ((i.tm.base_opcode & ~0x3) == MOV_AX_DISP32)
3392 {
3393 default_seg = &ds;
3394 }
3395 else if ((i.tm.opcode_modifier & IsString) != 0)
3396 {
3397 /* For the string instructions that allow a segment override
3398 on one of their operands, the default segment is ds. */
3399 default_seg = &ds;
3400 }
3401
3402 if ((i.tm.base_opcode == 0x8d /* lea */
3403 || (i.tm.cpu_flags & CpuSVME))
3404 && i.seg[0] && !quiet_warnings)
3405 as_warn (_("segment override on `%s' is ineffectual"), i.tm.name);
3406
3407 /* If a segment was explicitly specified, and the specified segment
3408 is not the default, use an opcode prefix to select it. If we
3409 never figured out what the default segment is, then default_seg
3410 will be zero at this point, and the specified segment prefix will
3411 always be used. */
3412 if ((i.seg[0]) && (i.seg[0] != default_seg))
3413 {
3414 if (!add_prefix (i.seg[0]->seg_prefix))
3415 return 0;
3416 }
3417 return 1;
3418 }
3419
3420 static const seg_entry *
3421 build_modrm_byte (void)
3422 {
3423 const seg_entry *default_seg = 0;
3424
3425 /* i.reg_operands MUST be the number of real register operands;
3426 implicit registers do not count. */
3427 if (i.reg_operands == 2)
3428 {
3429 unsigned int source, dest;
3430
3431 switch (i.operands)
3432 {
3433 case 2:
3434 source = 0;
3435 break;
3436 case 3:
3437 /* When there are 3 operands, one of them may be immediate,
3438 which may be the first or the last operand. Otherwise,
3439 the first operand must be shift count register (cl). */
3440 assert (i.imm_operands == 1
3441 || (i.imm_operands == 0
3442 && (i.types[0] & ShiftCount)));
3443 source = (i.types[0] & (Imm | ShiftCount)) ? 1 : 0;
3444 break;
3445 case 4:
3446 /* When there are 4 operands, the first two must be immediate
3447 operands. The source operand will be the 3rd one. */
3448 assert (i.imm_operands == 2
3449 && (i.types[0] & Imm)
3450 && (i.types[1] & Imm));
3451 source = 2;
3452 break;
3453 default:
3454 abort ();
3455 }
3456
3457 dest = source + 1;
3458
3459 i.rm.mode = 3;
3460 /* One of the register operands will be encoded in the i.tm.reg
3461 field, the other in the combined i.tm.mode and i.tm.regmem
3462 fields. If no form of this instruction supports a memory
3463 destination operand, then we assume the source operand may
3464 sometimes be a memory operand and so we need to store the
3465 destination in the i.rm.reg field. */
3466 if ((i.tm.operand_types[dest] & (AnyMem | RegMem)) == 0)
3467 {
3468 i.rm.reg = i.op[dest].regs->reg_num;
3469 i.rm.regmem = i.op[source].regs->reg_num;
3470 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
3471 i.rex |= REX_R;
3472 if ((i.op[source].regs->reg_flags & RegRex) != 0)
3473 i.rex |= REX_B;
3474 }
3475 else
3476 {
3477 i.rm.reg = i.op[source].regs->reg_num;
3478 i.rm.regmem = i.op[dest].regs->reg_num;
3479 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
3480 i.rex |= REX_B;
3481 if ((i.op[source].regs->reg_flags & RegRex) != 0)
3482 i.rex |= REX_R;
3483 }
3484 if (flag_code != CODE_64BIT && (i.rex & (REX_R | REX_B)))
3485 {
3486 if (!((i.types[0] | i.types[1]) & Control))
3487 abort ();
3488 i.rex &= ~(REX_R | REX_B);
3489 add_prefix (LOCK_PREFIX_OPCODE);
3490 }
3491 }
3492 else
3493 { /* If it's not 2 reg operands... */
3494 if (i.mem_operands)
3495 {
3496 unsigned int fake_zero_displacement = 0;
3497 unsigned int op;
3498
3499 for (op = 0; op < i.operands; op++)
3500 if ((i.types[op] & AnyMem))
3501 break;
3502 assert (op < i.operands);
3503
3504 default_seg = &ds;
3505
3506 if (i.base_reg == 0)
3507 {
3508 i.rm.mode = 0;
3509 if (!i.disp_operands)
3510 fake_zero_displacement = 1;
3511 if (i.index_reg == 0)
3512 {
3513 /* Operand is just <disp> */
3514 if (flag_code == CODE_64BIT)
3515 {
3516 /* 64bit mode overwrites the 32bit absolute
3517 addressing by RIP relative addressing and
3518 absolute addressing is encoded by one of the
3519 redundant SIB forms. */
3520 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3521 i.sib.base = NO_BASE_REGISTER;
3522 i.sib.index = NO_INDEX_REGISTER;
3523 i.types[op] = ((i.prefix[ADDR_PREFIX] == 0)
3524 ? Disp32S : Disp32);
3525 }
3526 else if ((flag_code == CODE_16BIT)
3527 ^ (i.prefix[ADDR_PREFIX] != 0))
3528 {
3529 i.rm.regmem = NO_BASE_REGISTER_16;
3530 i.types[op] = Disp16;
3531 }
3532 else
3533 {
3534 i.rm.regmem = NO_BASE_REGISTER;
3535 i.types[op] = Disp32;
3536 }
3537 }
3538 else /* !i.base_reg && i.index_reg */
3539 {
3540 i.sib.index = i.index_reg->reg_num;
3541 i.sib.base = NO_BASE_REGISTER;
3542 i.sib.scale = i.log2_scale_factor;
3543 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3544 i.types[op] &= ~Disp;
3545 if (flag_code != CODE_64BIT)
3546 i.types[op] |= Disp32; /* Must be 32 bit */
3547 else
3548 i.types[op] |= Disp32S;
3549 if ((i.index_reg->reg_flags & RegRex) != 0)
3550 i.rex |= REX_X;
3551 }
3552 }
3553 /* RIP addressing for 64bit mode. */
3554 else if (i.base_reg->reg_type == BaseIndex)
3555 {
3556 i.rm.regmem = NO_BASE_REGISTER;
3557 i.types[op] &= ~ Disp;
3558 i.types[op] |= Disp32S;
3559 i.flags[op] |= Operand_PCrel;
3560 if (! i.disp_operands)
3561 fake_zero_displacement = 1;
3562 }
3563 else if (i.base_reg->reg_type & Reg16)
3564 {
3565 switch (i.base_reg->reg_num)
3566 {
3567 case 3: /* (%bx) */
3568 if (i.index_reg == 0)
3569 i.rm.regmem = 7;
3570 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
3571 i.rm.regmem = i.index_reg->reg_num - 6;
3572 break;
3573 case 5: /* (%bp) */
3574 default_seg = &ss;
3575 if (i.index_reg == 0)
3576 {
3577 i.rm.regmem = 6;
3578 if ((i.types[op] & Disp) == 0)
3579 {
3580 /* fake (%bp) into 0(%bp) */
3581 i.types[op] |= Disp8;
3582 fake_zero_displacement = 1;
3583 }
3584 }
3585 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
3586 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
3587 break;
3588 default: /* (%si) -> 4 or (%di) -> 5 */
3589 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
3590 }
3591 i.rm.mode = mode_from_disp_size (i.types[op]);
3592 }
3593 else /* i.base_reg and 32/64 bit mode */
3594 {
3595 if (flag_code == CODE_64BIT
3596 && (i.types[op] & Disp))
3597 i.types[op] = ((i.types[op] & Disp8)
3598 | (i.prefix[ADDR_PREFIX] == 0
3599 ? Disp32S : Disp32));
3600
3601 i.rm.regmem = i.base_reg->reg_num;
3602 if ((i.base_reg->reg_flags & RegRex) != 0)
3603 i.rex |= REX_B;
3604 i.sib.base = i.base_reg->reg_num;
3605 /* x86-64 ignores REX prefix bit here to avoid decoder
3606 complications. */
3607 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
3608 {
3609 default_seg = &ss;
3610 if (i.disp_operands == 0)
3611 {
3612 fake_zero_displacement = 1;
3613 i.types[op] |= Disp8;
3614 }
3615 }
3616 else if (i.base_reg->reg_num == ESP_REG_NUM)
3617 {
3618 default_seg = &ss;
3619 }
3620 i.sib.scale = i.log2_scale_factor;
3621 if (i.index_reg == 0)
3622 {
3623 /* <disp>(%esp) becomes two byte modrm with no index
3624 register. We've already stored the code for esp
3625 in i.rm.regmem ie. ESCAPE_TO_TWO_BYTE_ADDRESSING.
3626 Any base register besides %esp will not use the
3627 extra modrm byte. */
3628 i.sib.index = NO_INDEX_REGISTER;
3629 #if !SCALE1_WHEN_NO_INDEX
3630 /* Another case where we force the second modrm byte. */
3631 if (i.log2_scale_factor)
3632 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3633 #endif
3634 }
3635 else
3636 {
3637 i.sib.index = i.index_reg->reg_num;
3638 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3639 if ((i.index_reg->reg_flags & RegRex) != 0)
3640 i.rex |= REX_X;
3641 }
3642
3643 if (i.disp_operands
3644 && (i.reloc[op] == BFD_RELOC_386_TLS_DESC_CALL
3645 || i.reloc[op] == BFD_RELOC_X86_64_TLSDESC_CALL))
3646 i.rm.mode = 0;
3647 else
3648 i.rm.mode = mode_from_disp_size (i.types[op]);
3649 }
3650
3651 if (fake_zero_displacement)
3652 {
3653 /* Fakes a zero displacement assuming that i.types[op]
3654 holds the correct displacement size. */
3655 expressionS *exp;
3656
3657 assert (i.op[op].disps == 0);
3658 exp = &disp_expressions[i.disp_operands++];
3659 i.op[op].disps = exp;
3660 exp->X_op = O_constant;
3661 exp->X_add_number = 0;
3662 exp->X_add_symbol = (symbolS *) 0;
3663 exp->X_op_symbol = (symbolS *) 0;
3664 }
3665 }
3666
3667 /* Fill in i.rm.reg or i.rm.regmem field with register operand
3668 (if any) based on i.tm.extension_opcode. Again, we must be
3669 careful to make sure that segment/control/debug/test/MMX
3670 registers are coded into the i.rm.reg field. */
3671 if (i.reg_operands)
3672 {
3673 unsigned int op;
3674
3675 for (op = 0; op < i.operands; op++)
3676 if ((i.types[op] & (Reg | RegMMX | RegXMM
3677 | SReg2 | SReg3
3678 | Control | Debug | Test)))
3679 break;
3680 assert (op < i.operands);
3681
3682 /* If there is an extension opcode to put here, the register
3683 number must be put into the regmem field. */
3684 if (i.tm.extension_opcode != None)
3685 {
3686 i.rm.regmem = i.op[op].regs->reg_num;
3687 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3688 i.rex |= REX_B;
3689 }
3690 else
3691 {
3692 i.rm.reg = i.op[op].regs->reg_num;
3693 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3694 i.rex |= REX_R;
3695 }
3696
3697 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2 we
3698 must set it to 3 to indicate this is a register operand
3699 in the regmem field. */
3700 if (!i.mem_operands)
3701 i.rm.mode = 3;
3702 }
3703
3704 /* Fill in i.rm.reg field with extension opcode (if any). */
3705 if (i.tm.extension_opcode != None)
3706 i.rm.reg = i.tm.extension_opcode;
3707 }
3708 return default_seg;
3709 }
3710
3711 static void
3712 output_branch (void)
3713 {
3714 char *p;
3715 int code16;
3716 int prefix;
3717 relax_substateT subtype;
3718 symbolS *sym;
3719 offsetT off;
3720
3721 code16 = 0;
3722 if (flag_code == CODE_16BIT)
3723 code16 = CODE16;
3724
3725 prefix = 0;
3726 if (i.prefix[DATA_PREFIX] != 0)
3727 {
3728 prefix = 1;
3729 i.prefixes -= 1;
3730 code16 ^= CODE16;
3731 }
3732 /* Pentium4 branch hints. */
3733 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3734 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3735 {
3736 prefix++;
3737 i.prefixes--;
3738 }
3739 if (i.prefix[REX_PREFIX] != 0)
3740 {
3741 prefix++;
3742 i.prefixes--;
3743 }
3744
3745 if (i.prefixes != 0 && !intel_syntax)
3746 as_warn (_("skipping prefixes on this instruction"));
3747
3748 /* It's always a symbol; End frag & setup for relax.
3749 Make sure there is enough room in this frag for the largest
3750 instruction we may generate in md_convert_frag. This is 2
3751 bytes for the opcode and room for the prefix and largest
3752 displacement. */
3753 frag_grow (prefix + 2 + 4);
3754 /* Prefix and 1 opcode byte go in fr_fix. */
3755 p = frag_more (prefix + 1);
3756 if (i.prefix[DATA_PREFIX] != 0)
3757 *p++ = DATA_PREFIX_OPCODE;
3758 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
3759 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
3760 *p++ = i.prefix[SEG_PREFIX];
3761 if (i.prefix[REX_PREFIX] != 0)
3762 *p++ = i.prefix[REX_PREFIX];
3763 *p = i.tm.base_opcode;
3764
3765 if ((unsigned char) *p == JUMP_PC_RELATIVE)
3766 subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
3767 else if ((cpu_arch_flags & Cpu386) != 0)
3768 subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
3769 else
3770 subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
3771 subtype |= code16;
3772
3773 sym = i.op[0].disps->X_add_symbol;
3774 off = i.op[0].disps->X_add_number;
3775
3776 if (i.op[0].disps->X_op != O_constant
3777 && i.op[0].disps->X_op != O_symbol)
3778 {
3779 /* Handle complex expressions. */
3780 sym = make_expr_symbol (i.op[0].disps);
3781 off = 0;
3782 }
3783
3784 /* 1 possible extra opcode + 4 byte displacement go in var part.
3785 Pass reloc in fr_var. */
3786 frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
3787 }
3788
3789 static void
3790 output_jump (void)
3791 {
3792 char *p;
3793 int size;
3794 fixS *fixP;
3795
3796 if (i.tm.opcode_modifier & JumpByte)
3797 {
3798 /* This is a loop or jecxz type instruction. */
3799 size = 1;
3800 if (i.prefix[ADDR_PREFIX] != 0)
3801 {
3802 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
3803 i.prefixes -= 1;
3804 }
3805 /* Pentium4 branch hints. */
3806 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3807 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3808 {
3809 FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
3810 i.prefixes--;
3811 }
3812 }
3813 else
3814 {
3815 int code16;
3816
3817 code16 = 0;
3818 if (flag_code == CODE_16BIT)
3819 code16 = CODE16;
3820
3821 if (i.prefix[DATA_PREFIX] != 0)
3822 {
3823 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
3824 i.prefixes -= 1;
3825 code16 ^= CODE16;
3826 }
3827
3828 size = 4;
3829 if (code16)
3830 size = 2;
3831 }
3832
3833 if (i.prefix[REX_PREFIX] != 0)
3834 {
3835 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
3836 i.prefixes -= 1;
3837 }
3838
3839 if (i.prefixes != 0 && !intel_syntax)
3840 as_warn (_("skipping prefixes on this instruction"));
3841
3842 p = frag_more (1 + size);
3843 *p++ = i.tm.base_opcode;
3844
3845 fixP = fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3846 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
3847
3848 /* All jumps handled here are signed, but don't use a signed limit
3849 check for 32 and 16 bit jumps as we want to allow wrap around at
3850 4G and 64k respectively. */
3851 if (size == 1)
3852 fixP->fx_signed = 1;
3853 }
3854
3855 static void
3856 output_interseg_jump (void)
3857 {
3858 char *p;
3859 int size;
3860 int prefix;
3861 int code16;
3862
3863 code16 = 0;
3864 if (flag_code == CODE_16BIT)
3865 code16 = CODE16;
3866
3867 prefix = 0;
3868 if (i.prefix[DATA_PREFIX] != 0)
3869 {
3870 prefix = 1;
3871 i.prefixes -= 1;
3872 code16 ^= CODE16;
3873 }
3874 if (i.prefix[REX_PREFIX] != 0)
3875 {
3876 prefix++;
3877 i.prefixes -= 1;
3878 }
3879
3880 size = 4;
3881 if (code16)
3882 size = 2;
3883
3884 if (i.prefixes != 0 && !intel_syntax)
3885 as_warn (_("skipping prefixes on this instruction"));
3886
3887 /* 1 opcode; 2 segment; offset */
3888 p = frag_more (prefix + 1 + 2 + size);
3889
3890 if (i.prefix[DATA_PREFIX] != 0)
3891 *p++ = DATA_PREFIX_OPCODE;
3892
3893 if (i.prefix[REX_PREFIX] != 0)
3894 *p++ = i.prefix[REX_PREFIX];
3895
3896 *p++ = i.tm.base_opcode;
3897 if (i.op[1].imms->X_op == O_constant)
3898 {
3899 offsetT n = i.op[1].imms->X_add_number;
3900
3901 if (size == 2
3902 && !fits_in_unsigned_word (n)
3903 && !fits_in_signed_word (n))
3904 {
3905 as_bad (_("16-bit jump out of range"));
3906 return;
3907 }
3908 md_number_to_chars (p, n, size);
3909 }
3910 else
3911 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3912 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
3913 if (i.op[0].imms->X_op != O_constant)
3914 as_bad (_("can't handle non absolute segment in `%s'"),
3915 i.tm.name);
3916 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
3917 }
3918
3919 static void
3920 output_insn (void)
3921 {
3922 fragS *insn_start_frag;
3923 offsetT insn_start_off;
3924
3925 /* Tie dwarf2 debug info to the address at the start of the insn.
3926 We can't do this after the insn has been output as the current
3927 frag may have been closed off. eg. by frag_var. */
3928 dwarf2_emit_insn (0);
3929
3930 insn_start_frag = frag_now;
3931 insn_start_off = frag_now_fix ();
3932
3933 /* Output jumps. */
3934 if (i.tm.opcode_modifier & Jump)
3935 output_branch ();
3936 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
3937 output_jump ();
3938 else if (i.tm.opcode_modifier & JumpInterSegment)
3939 output_interseg_jump ();
3940 else
3941 {
3942 /* Output normal instructions here. */
3943 char *p;
3944 unsigned char *q;
3945 unsigned int prefix;
3946
3947 /* All opcodes on i386 have either 1 or 2 bytes. SSSE3 and
3948 SSE4 instructions have 3 bytes. We may use one more higher
3949 byte to specify a prefix the instruction requires. Exclude
3950 instructions which are in both SSE4 and ABM. */
3951 if ((i.tm.cpu_flags & (CpuSSSE3 | CpuSSE4)) != 0
3952 && (i.tm.cpu_flags & CpuABM) == 0)
3953 {
3954 if (i.tm.base_opcode & 0xff000000)
3955 {
3956 prefix = (i.tm.base_opcode >> 24) & 0xff;
3957 goto check_prefix;
3958 }
3959 }
3960 else if ((i.tm.base_opcode & 0xff0000) != 0)
3961 {
3962 prefix = (i.tm.base_opcode >> 16) & 0xff;
3963 if ((i.tm.cpu_flags & CpuPadLock) != 0)
3964 {
3965 check_prefix:
3966 if (prefix != REPE_PREFIX_OPCODE
3967 || i.prefix[LOCKREP_PREFIX] != REPE_PREFIX_OPCODE)
3968 add_prefix (prefix);
3969 }
3970 else
3971 add_prefix (prefix);
3972 }
3973
3974 /* The prefix bytes. */
3975 for (q = i.prefix;
3976 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
3977 q++)
3978 {
3979 if (*q)
3980 {
3981 p = frag_more (1);
3982 md_number_to_chars (p, (valueT) *q, 1);
3983 }
3984 }
3985
3986 /* Now the opcode; be careful about word order here! */
3987 if (fits_in_unsigned_byte (i.tm.base_opcode))
3988 {
3989 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
3990 }
3991 else
3992 {
3993 if ((i.tm.cpu_flags & (CpuSSSE3 | CpuSSE4)) != 0
3994 && (i.tm.cpu_flags & CpuABM) == 0)
3995 {
3996 p = frag_more (3);
3997 *p++ = (i.tm.base_opcode >> 16) & 0xff;
3998 }
3999 else
4000 p = frag_more (2);
4001
4002 /* Put out high byte first: can't use md_number_to_chars! */
4003 *p++ = (i.tm.base_opcode >> 8) & 0xff;
4004 *p = i.tm.base_opcode & 0xff;
4005 }
4006
4007 /* Now the modrm byte and sib byte (if present). */
4008 if (i.tm.opcode_modifier & Modrm)
4009 {
4010 p = frag_more (1);
4011 md_number_to_chars (p,
4012 (valueT) (i.rm.regmem << 0
4013 | i.rm.reg << 3
4014 | i.rm.mode << 6),
4015 1);
4016 /* If i.rm.regmem == ESP (4)
4017 && i.rm.mode != (Register mode)
4018 && not 16 bit
4019 ==> need second modrm byte. */
4020 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
4021 && i.rm.mode != 3
4022 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
4023 {
4024 p = frag_more (1);
4025 md_number_to_chars (p,
4026 (valueT) (i.sib.base << 0
4027 | i.sib.index << 3
4028 | i.sib.scale << 6),
4029 1);
4030 }
4031 }
4032
4033 if (i.disp_operands)
4034 output_disp (insn_start_frag, insn_start_off);
4035
4036 if (i.imm_operands)
4037 output_imm (insn_start_frag, insn_start_off);
4038 }
4039
4040 #ifdef DEBUG386
4041 if (flag_debug)
4042 {
4043 pi ("" /*line*/, &i);
4044 }
4045 #endif /* DEBUG386 */
4046 }
4047
4048 static void
4049 output_disp (fragS *insn_start_frag, offsetT insn_start_off)
4050 {
4051 char *p;
4052 unsigned int n;
4053
4054 for (n = 0; n < i.operands; n++)
4055 {
4056 if (i.types[n] & Disp)
4057 {
4058 if (i.op[n].disps->X_op == O_constant)
4059 {
4060 int size;
4061 offsetT val;
4062
4063 size = 4;
4064 if (i.types[n] & (Disp8 | Disp16 | Disp64))
4065 {
4066 size = 2;
4067 if (i.types[n] & Disp8)
4068 size = 1;
4069 if (i.types[n] & Disp64)
4070 size = 8;
4071 }
4072 val = offset_in_range (i.op[n].disps->X_add_number,
4073 size);
4074 p = frag_more (size);
4075 md_number_to_chars (p, val, size);
4076 }
4077 else
4078 {
4079 enum bfd_reloc_code_real reloc_type;
4080 int size = 4;
4081 int sign = 0;
4082 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
4083
4084 /* The PC relative address is computed relative
4085 to the instruction boundary, so in case immediate
4086 fields follows, we need to adjust the value. */
4087 if (pcrel && i.imm_operands)
4088 {
4089 int imm_size = 4;
4090 unsigned int n1;
4091
4092 for (n1 = 0; n1 < i.operands; n1++)
4093 if (i.types[n1] & Imm)
4094 {
4095 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
4096 {
4097 imm_size = 2;
4098 if (i.types[n1] & (Imm8 | Imm8S))
4099 imm_size = 1;
4100 if (i.types[n1] & Imm64)
4101 imm_size = 8;
4102 }
4103 break;
4104 }
4105 /* We should find the immediate. */
4106 if (n1 == i.operands)
4107 abort ();
4108 i.op[n].disps->X_add_number -= imm_size;
4109 }
4110
4111 if (i.types[n] & Disp32S)
4112 sign = 1;
4113
4114 if (i.types[n] & (Disp16 | Disp64))
4115 {
4116 size = 2;
4117 if (i.types[n] & Disp64)
4118 size = 8;
4119 }
4120
4121 p = frag_more (size);
4122 reloc_type = reloc (size, pcrel, sign, i.reloc[n]);
4123 if (GOT_symbol
4124 && GOT_symbol == i.op[n].disps->X_add_symbol
4125 && (((reloc_type == BFD_RELOC_32
4126 || reloc_type == BFD_RELOC_X86_64_32S
4127 || (reloc_type == BFD_RELOC_64
4128 && object_64bit))
4129 && (i.op[n].disps->X_op == O_symbol
4130 || (i.op[n].disps->X_op == O_add
4131 && ((symbol_get_value_expression
4132 (i.op[n].disps->X_op_symbol)->X_op)
4133 == O_subtract))))
4134 || reloc_type == BFD_RELOC_32_PCREL))
4135 {
4136 offsetT add;
4137
4138 if (insn_start_frag == frag_now)
4139 add = (p - frag_now->fr_literal) - insn_start_off;
4140 else
4141 {
4142 fragS *fr;
4143
4144 add = insn_start_frag->fr_fix - insn_start_off;
4145 for (fr = insn_start_frag->fr_next;
4146 fr && fr != frag_now; fr = fr->fr_next)
4147 add += fr->fr_fix;
4148 add += p - frag_now->fr_literal;
4149 }
4150
4151 if (!object_64bit)
4152 {
4153 reloc_type = BFD_RELOC_386_GOTPC;
4154 i.op[n].imms->X_add_number += add;
4155 }
4156 else if (reloc_type == BFD_RELOC_64)
4157 reloc_type = BFD_RELOC_X86_64_GOTPC64;
4158 else
4159 /* Don't do the adjustment for x86-64, as there
4160 the pcrel addressing is relative to the _next_
4161 insn, and that is taken care of in other code. */
4162 reloc_type = BFD_RELOC_X86_64_GOTPC32;
4163 }
4164 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
4165 i.op[n].disps, pcrel, reloc_type);
4166 }
4167 }
4168 }
4169 }
4170
4171 static void
4172 output_imm (fragS *insn_start_frag, offsetT insn_start_off)
4173 {
4174 char *p;
4175 unsigned int n;
4176
4177 for (n = 0; n < i.operands; n++)
4178 {
4179 if (i.types[n] & Imm)
4180 {
4181 if (i.op[n].imms->X_op == O_constant)
4182 {
4183 int size;
4184 offsetT val;
4185
4186 size = 4;
4187 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
4188 {
4189 size = 2;
4190 if (i.types[n] & (Imm8 | Imm8S))
4191 size = 1;
4192 else if (i.types[n] & Imm64)
4193 size = 8;
4194 }
4195 val = offset_in_range (i.op[n].imms->X_add_number,
4196 size);
4197 p = frag_more (size);
4198 md_number_to_chars (p, val, size);
4199 }
4200 else
4201 {
4202 /* Not absolute_section.
4203 Need a 32-bit fixup (don't support 8bit
4204 non-absolute imms). Try to support other
4205 sizes ... */
4206 enum bfd_reloc_code_real reloc_type;
4207 int size = 4;
4208 int sign = 0;
4209
4210 if ((i.types[n] & (Imm32S))
4211 && (i.suffix == QWORD_MNEM_SUFFIX
4212 || (!i.suffix && (i.tm.opcode_modifier & No_lSuf))))
4213 sign = 1;
4214 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
4215 {
4216 size = 2;
4217 if (i.types[n] & (Imm8 | Imm8S))
4218 size = 1;
4219 if (i.types[n] & Imm64)
4220 size = 8;
4221 }
4222
4223 p = frag_more (size);
4224 reloc_type = reloc (size, 0, sign, i.reloc[n]);
4225
4226 /* This is tough to explain. We end up with this one if we
4227 * have operands that look like
4228 * "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal here is to
4229 * obtain the absolute address of the GOT, and it is strongly
4230 * preferable from a performance point of view to avoid using
4231 * a runtime relocation for this. The actual sequence of
4232 * instructions often look something like:
4233 *
4234 * call .L66
4235 * .L66:
4236 * popl %ebx
4237 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
4238 *
4239 * The call and pop essentially return the absolute address
4240 * of the label .L66 and store it in %ebx. The linker itself
4241 * will ultimately change the first operand of the addl so
4242 * that %ebx points to the GOT, but to keep things simple, the
4243 * .o file must have this operand set so that it generates not
4244 * the absolute address of .L66, but the absolute address of
4245 * itself. This allows the linker itself simply treat a GOTPC
4246 * relocation as asking for a pcrel offset to the GOT to be
4247 * added in, and the addend of the relocation is stored in the
4248 * operand field for the instruction itself.
4249 *
4250 * Our job here is to fix the operand so that it would add
4251 * the correct offset so that %ebx would point to itself. The
4252 * thing that is tricky is that .-.L66 will point to the
4253 * beginning of the instruction, so we need to further modify
4254 * the operand so that it will point to itself. There are
4255 * other cases where you have something like:
4256 *
4257 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
4258 *
4259 * and here no correction would be required. Internally in
4260 * the assembler we treat operands of this form as not being
4261 * pcrel since the '.' is explicitly mentioned, and I wonder
4262 * whether it would simplify matters to do it this way. Who
4263 * knows. In earlier versions of the PIC patches, the
4264 * pcrel_adjust field was used to store the correction, but
4265 * since the expression is not pcrel, I felt it would be
4266 * confusing to do it this way. */
4267
4268 if ((reloc_type == BFD_RELOC_32
4269 || reloc_type == BFD_RELOC_X86_64_32S
4270 || reloc_type == BFD_RELOC_64)
4271 && GOT_symbol
4272 && GOT_symbol == i.op[n].imms->X_add_symbol
4273 && (i.op[n].imms->X_op == O_symbol
4274 || (i.op[n].imms->X_op == O_add
4275 && ((symbol_get_value_expression
4276 (i.op[n].imms->X_op_symbol)->X_op)
4277 == O_subtract))))
4278 {
4279 offsetT add;
4280
4281 if (insn_start_frag == frag_now)
4282 add = (p - frag_now->fr_literal) - insn_start_off;
4283 else
4284 {
4285 fragS *fr;
4286
4287 add = insn_start_frag->fr_fix - insn_start_off;
4288 for (fr = insn_start_frag->fr_next;
4289 fr && fr != frag_now; fr = fr->fr_next)
4290 add += fr->fr_fix;
4291 add += p - frag_now->fr_literal;
4292 }
4293
4294 if (!object_64bit)
4295 reloc_type = BFD_RELOC_386_GOTPC;
4296 else if (size == 4)
4297 reloc_type = BFD_RELOC_X86_64_GOTPC32;
4298 else if (size == 8)
4299 reloc_type = BFD_RELOC_X86_64_GOTPC64;
4300 i.op[n].imms->X_add_number += add;
4301 }
4302 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
4303 i.op[n].imms, 0, reloc_type);
4304 }
4305 }
4306 }
4307 }
4308 \f
4309 /* x86_cons_fix_new is called via the expression parsing code when a
4310 reloc is needed. We use this hook to get the correct .got reloc. */
4311 static enum bfd_reloc_code_real got_reloc = NO_RELOC;
4312 static int cons_sign = -1;
4313
4314 void
4315 x86_cons_fix_new (fragS *frag, unsigned int off, unsigned int len,
4316 expressionS *exp)
4317 {
4318 enum bfd_reloc_code_real r = reloc (len, 0, cons_sign, got_reloc);
4319
4320 got_reloc = NO_RELOC;
4321
4322 #ifdef TE_PE
4323 if (exp->X_op == O_secrel)
4324 {
4325 exp->X_op = O_symbol;
4326 r = BFD_RELOC_32_SECREL;
4327 }
4328 #endif
4329
4330 fix_new_exp (frag, off, len, exp, 0, r);
4331 }
4332
4333 #if (!defined (OBJ_ELF) && !defined (OBJ_MAYBE_ELF)) || defined (LEX_AT)
4334 # define lex_got(reloc, adjust, types) NULL
4335 #else
4336 /* Parse operands of the form
4337 <symbol>@GOTOFF+<nnn>
4338 and similar .plt or .got references.
4339
4340 If we find one, set up the correct relocation in RELOC and copy the
4341 input string, minus the `@GOTOFF' into a malloc'd buffer for
4342 parsing by the calling routine. Return this buffer, and if ADJUST
4343 is non-null set it to the length of the string we removed from the
4344 input line. Otherwise return NULL. */
4345 static char *
4346 lex_got (enum bfd_reloc_code_real *reloc,
4347 int *adjust,
4348 unsigned int *types)
4349 {
4350 /* Some of the relocations depend on the size of what field is to
4351 be relocated. But in our callers i386_immediate and i386_displacement
4352 we don't yet know the operand size (this will be set by insn
4353 matching). Hence we record the word32 relocation here,
4354 and adjust the reloc according to the real size in reloc(). */
4355 static const struct {
4356 const char *str;
4357 const enum bfd_reloc_code_real rel[2];
4358 const unsigned int types64;
4359 } gotrel[] = {
4360 { "PLTOFF", { 0,
4361 BFD_RELOC_X86_64_PLTOFF64 },
4362 Imm64 },
4363 { "PLT", { BFD_RELOC_386_PLT32,
4364 BFD_RELOC_X86_64_PLT32 },
4365 Imm32 | Imm32S | Disp32 },
4366 { "GOTPLT", { 0,
4367 BFD_RELOC_X86_64_GOTPLT64 },
4368 Imm64 | Disp64 },
4369 { "GOTOFF", { BFD_RELOC_386_GOTOFF,
4370 BFD_RELOC_X86_64_GOTOFF64 },
4371 Imm64 | Disp64 },
4372 { "GOTPCREL", { 0,
4373 BFD_RELOC_X86_64_GOTPCREL },
4374 Imm32 | Imm32S | Disp32 },
4375 { "TLSGD", { BFD_RELOC_386_TLS_GD,
4376 BFD_RELOC_X86_64_TLSGD },
4377 Imm32 | Imm32S | Disp32 },
4378 { "TLSLDM", { BFD_RELOC_386_TLS_LDM,
4379 0 },
4380 0 },
4381 { "TLSLD", { 0,
4382 BFD_RELOC_X86_64_TLSLD },
4383 Imm32 | Imm32S | Disp32 },
4384 { "GOTTPOFF", { BFD_RELOC_386_TLS_IE_32,
4385 BFD_RELOC_X86_64_GOTTPOFF },
4386 Imm32 | Imm32S | Disp32 },
4387 { "TPOFF", { BFD_RELOC_386_TLS_LE_32,
4388 BFD_RELOC_X86_64_TPOFF32 },
4389 Imm32 | Imm32S | Imm64 | Disp32 | Disp64 },
4390 { "NTPOFF", { BFD_RELOC_386_TLS_LE,
4391 0 },
4392 0 },
4393 { "DTPOFF", { BFD_RELOC_386_TLS_LDO_32,
4394 BFD_RELOC_X86_64_DTPOFF32 },
4395 Imm32 | Imm32S | Imm64 | Disp32 | Disp64 },
4396 { "GOTNTPOFF",{ BFD_RELOC_386_TLS_GOTIE,
4397 0 },
4398 0 },
4399 { "INDNTPOFF",{ BFD_RELOC_386_TLS_IE,
4400 0 },
4401 0 },
4402 { "GOT", { BFD_RELOC_386_GOT32,
4403 BFD_RELOC_X86_64_GOT32 },
4404 Imm32 | Imm32S | Disp32 | Imm64 },
4405 { "TLSDESC", { BFD_RELOC_386_TLS_GOTDESC,
4406 BFD_RELOC_X86_64_GOTPC32_TLSDESC },
4407 Imm32 | Imm32S | Disp32 },
4408 { "TLSCALL", { BFD_RELOC_386_TLS_DESC_CALL,
4409 BFD_RELOC_X86_64_TLSDESC_CALL },
4410 Imm32 | Imm32S | Disp32 }
4411 };
4412 char *cp;
4413 unsigned int j;
4414
4415 if (!IS_ELF)
4416 return NULL;
4417
4418 for (cp = input_line_pointer; *cp != '@'; cp++)
4419 if (is_end_of_line[(unsigned char) *cp])
4420 return NULL;
4421
4422 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
4423 {
4424 int len;
4425
4426 len = strlen (gotrel[j].str);
4427 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
4428 {
4429 if (gotrel[j].rel[object_64bit] != 0)
4430 {
4431 int first, second;
4432 char *tmpbuf, *past_reloc;
4433
4434 *reloc = gotrel[j].rel[object_64bit];
4435 if (adjust)
4436 *adjust = len;
4437
4438 if (types)
4439 {
4440 if (flag_code != CODE_64BIT)
4441 *types = Imm32 | Disp32;
4442 else
4443 *types = gotrel[j].types64;
4444 }
4445
4446 if (GOT_symbol == NULL)
4447 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
4448
4449 /* Replace the relocation token with ' ', so that
4450 errors like foo@GOTOFF1 will be detected. */
4451
4452 /* The length of the first part of our input line. */
4453 first = cp - input_line_pointer;
4454
4455 /* The second part goes from after the reloc token until
4456 (and including) an end_of_line char. Don't use strlen
4457 here as the end_of_line char may not be a NUL. */
4458 past_reloc = cp + 1 + len;
4459 for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
4460 ;
4461 second = cp - past_reloc;
4462
4463 /* Allocate and copy string. The trailing NUL shouldn't
4464 be necessary, but be safe. */
4465 tmpbuf = xmalloc (first + second + 2);
4466 memcpy (tmpbuf, input_line_pointer, first);
4467 tmpbuf[first] = ' ';
4468 memcpy (tmpbuf + first + 1, past_reloc, second);
4469 tmpbuf[first + second + 1] = '\0';
4470 return tmpbuf;
4471 }
4472
4473 as_bad (_("@%s reloc is not supported with %d-bit output format"),
4474 gotrel[j].str, 1 << (5 + object_64bit));
4475 return NULL;
4476 }
4477 }
4478
4479 /* Might be a symbol version string. Don't as_bad here. */
4480 return NULL;
4481 }
4482
4483 void
4484 x86_cons (expressionS *exp, int size)
4485 {
4486 if (size == 4 || (object_64bit && size == 8))
4487 {
4488 /* Handle @GOTOFF and the like in an expression. */
4489 char *save;
4490 char *gotfree_input_line;
4491 int adjust;
4492
4493 save = input_line_pointer;
4494 gotfree_input_line = lex_got (&got_reloc, &adjust, NULL);
4495 if (gotfree_input_line)
4496 input_line_pointer = gotfree_input_line;
4497
4498 expression (exp);
4499
4500 if (gotfree_input_line)
4501 {
4502 /* expression () has merrily parsed up to the end of line,
4503 or a comma - in the wrong buffer. Transfer how far
4504 input_line_pointer has moved to the right buffer. */
4505 input_line_pointer = (save
4506 + (input_line_pointer - gotfree_input_line)
4507 + adjust);
4508 free (gotfree_input_line);
4509 }
4510 }
4511 else
4512 expression (exp);
4513 }
4514 #endif
4515
4516 static void signed_cons (int size)
4517 {
4518 if (flag_code == CODE_64BIT)
4519 cons_sign = 1;
4520 cons (size);
4521 cons_sign = -1;
4522 }
4523
4524 #ifdef TE_PE
4525 static void
4526 pe_directive_secrel (dummy)
4527 int dummy ATTRIBUTE_UNUSED;
4528 {
4529 expressionS exp;
4530
4531 do
4532 {
4533 expression (&exp);
4534 if (exp.X_op == O_symbol)
4535 exp.X_op = O_secrel;
4536
4537 emit_expr (&exp, 4);
4538 }
4539 while (*input_line_pointer++ == ',');
4540
4541 input_line_pointer--;
4542 demand_empty_rest_of_line ();
4543 }
4544 #endif
4545
4546 static int
4547 i386_immediate (char *imm_start)
4548 {
4549 char *save_input_line_pointer;
4550 char *gotfree_input_line;
4551 segT exp_seg = 0;
4552 expressionS *exp;
4553 unsigned int types = ~0U;
4554
4555 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
4556 {
4557 as_bad (_("at most %d immediate operands are allowed"),
4558 MAX_IMMEDIATE_OPERANDS);
4559 return 0;
4560 }
4561
4562 exp = &im_expressions[i.imm_operands++];
4563 i.op[this_operand].imms = exp;
4564
4565 if (is_space_char (*imm_start))
4566 ++imm_start;
4567
4568 save_input_line_pointer = input_line_pointer;
4569 input_line_pointer = imm_start;
4570
4571 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
4572 if (gotfree_input_line)
4573 input_line_pointer = gotfree_input_line;
4574
4575 exp_seg = expression (exp);
4576
4577 SKIP_WHITESPACE ();
4578 if (*input_line_pointer)
4579 as_bad (_("junk `%s' after expression"), input_line_pointer);
4580
4581 input_line_pointer = save_input_line_pointer;
4582 if (gotfree_input_line)
4583 free (gotfree_input_line);
4584
4585 if (exp->X_op == O_absent || exp->X_op == O_big)
4586 {
4587 /* Missing or bad expr becomes absolute 0. */
4588 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
4589 imm_start);
4590 exp->X_op = O_constant;
4591 exp->X_add_number = 0;
4592 exp->X_add_symbol = (symbolS *) 0;
4593 exp->X_op_symbol = (symbolS *) 0;
4594 }
4595 else if (exp->X_op == O_constant)
4596 {
4597 /* Size it properly later. */
4598 i.types[this_operand] |= Imm64;
4599 /* If BFD64, sign extend val. */
4600 if (!use_rela_relocations
4601 && (exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
4602 exp->X_add_number
4603 = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
4604 }
4605 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4606 else if (OUTPUT_FLAVOR == bfd_target_aout_flavour
4607 && exp_seg != absolute_section
4608 && exp_seg != text_section
4609 && exp_seg != data_section
4610 && exp_seg != bss_section
4611 && exp_seg != undefined_section
4612 && !bfd_is_com_section (exp_seg))
4613 {
4614 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
4615 return 0;
4616 }
4617 #endif
4618 else if (!intel_syntax && exp->X_op == O_register)
4619 {
4620 as_bad (_("illegal immediate register operand %s"), imm_start);
4621 return 0;
4622 }
4623 else
4624 {
4625 /* This is an address. The size of the address will be
4626 determined later, depending on destination register,
4627 suffix, or the default for the section. */
4628 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
4629 i.types[this_operand] &= types;
4630 }
4631
4632 return 1;
4633 }
4634
4635 static char *
4636 i386_scale (char *scale)
4637 {
4638 offsetT val;
4639 char *save = input_line_pointer;
4640
4641 input_line_pointer = scale;
4642 val = get_absolute_expression ();
4643
4644 switch (val)
4645 {
4646 case 1:
4647 i.log2_scale_factor = 0;
4648 break;
4649 case 2:
4650 i.log2_scale_factor = 1;
4651 break;
4652 case 4:
4653 i.log2_scale_factor = 2;
4654 break;
4655 case 8:
4656 i.log2_scale_factor = 3;
4657 break;
4658 default:
4659 {
4660 char sep = *input_line_pointer;
4661
4662 *input_line_pointer = '\0';
4663 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
4664 scale);
4665 *input_line_pointer = sep;
4666 input_line_pointer = save;
4667 return NULL;
4668 }
4669 }
4670 if (i.log2_scale_factor != 0 && i.index_reg == 0)
4671 {
4672 as_warn (_("scale factor of %d without an index register"),
4673 1 << i.log2_scale_factor);
4674 #if SCALE1_WHEN_NO_INDEX
4675 i.log2_scale_factor = 0;
4676 #endif
4677 }
4678 scale = input_line_pointer;
4679 input_line_pointer = save;
4680 return scale;
4681 }
4682
4683 static int
4684 i386_displacement (char *disp_start, char *disp_end)
4685 {
4686 expressionS *exp;
4687 segT exp_seg = 0;
4688 char *save_input_line_pointer;
4689 char *gotfree_input_line;
4690 int bigdisp, override;
4691 unsigned int types = Disp;
4692
4693 if (i.disp_operands == MAX_MEMORY_OPERANDS)
4694 {
4695 as_bad (_("at most %d displacement operands are allowed"),
4696 MAX_MEMORY_OPERANDS);
4697 return 0;
4698 }
4699
4700 if ((i.types[this_operand] & JumpAbsolute)
4701 || !(current_templates->start->opcode_modifier & (Jump | JumpDword)))
4702 {
4703 bigdisp = Disp32;
4704 override = (i.prefix[ADDR_PREFIX] != 0);
4705 }
4706 else
4707 {
4708 /* For PC-relative branches, the width of the displacement
4709 is dependent upon data size, not address size. */
4710 bigdisp = 0;
4711 override = (i.prefix[DATA_PREFIX] != 0);
4712 }
4713 if (flag_code == CODE_64BIT)
4714 {
4715 if (!bigdisp)
4716 bigdisp = ((override || i.suffix == WORD_MNEM_SUFFIX)
4717 ? Disp16
4718 : Disp32S | Disp32);
4719 else if (!override)
4720 bigdisp = Disp64 | Disp32S | Disp32;
4721 }
4722 else
4723 {
4724 if (!bigdisp)
4725 {
4726 if (!override)
4727 override = (i.suffix == (flag_code != CODE_16BIT
4728 ? WORD_MNEM_SUFFIX
4729 : LONG_MNEM_SUFFIX));
4730 bigdisp = Disp32;
4731 }
4732 if ((flag_code == CODE_16BIT) ^ override)
4733 bigdisp = Disp16;
4734 }
4735 i.types[this_operand] |= bigdisp;
4736
4737 exp = &disp_expressions[i.disp_operands];
4738 i.op[this_operand].disps = exp;
4739 i.disp_operands++;
4740 save_input_line_pointer = input_line_pointer;
4741 input_line_pointer = disp_start;
4742 END_STRING_AND_SAVE (disp_end);
4743
4744 #ifndef GCC_ASM_O_HACK
4745 #define GCC_ASM_O_HACK 0
4746 #endif
4747 #if GCC_ASM_O_HACK
4748 END_STRING_AND_SAVE (disp_end + 1);
4749 if ((i.types[this_operand] & BaseIndex) != 0
4750 && displacement_string_end[-1] == '+')
4751 {
4752 /* This hack is to avoid a warning when using the "o"
4753 constraint within gcc asm statements.
4754 For instance:
4755
4756 #define _set_tssldt_desc(n,addr,limit,type) \
4757 __asm__ __volatile__ ( \
4758 "movw %w2,%0\n\t" \
4759 "movw %w1,2+%0\n\t" \
4760 "rorl $16,%1\n\t" \
4761 "movb %b1,4+%0\n\t" \
4762 "movb %4,5+%0\n\t" \
4763 "movb $0,6+%0\n\t" \
4764 "movb %h1,7+%0\n\t" \
4765 "rorl $16,%1" \
4766 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
4767
4768 This works great except that the output assembler ends
4769 up looking a bit weird if it turns out that there is
4770 no offset. You end up producing code that looks like:
4771
4772 #APP
4773 movw $235,(%eax)
4774 movw %dx,2+(%eax)
4775 rorl $16,%edx
4776 movb %dl,4+(%eax)
4777 movb $137,5+(%eax)
4778 movb $0,6+(%eax)
4779 movb %dh,7+(%eax)
4780 rorl $16,%edx
4781 #NO_APP
4782
4783 So here we provide the missing zero. */
4784
4785 *displacement_string_end = '0';
4786 }
4787 #endif
4788 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
4789 if (gotfree_input_line)
4790 input_line_pointer = gotfree_input_line;
4791
4792 exp_seg = expression (exp);
4793
4794 SKIP_WHITESPACE ();
4795 if (*input_line_pointer)
4796 as_bad (_("junk `%s' after expression"), input_line_pointer);
4797 #if GCC_ASM_O_HACK
4798 RESTORE_END_STRING (disp_end + 1);
4799 #endif
4800 RESTORE_END_STRING (disp_end);
4801 input_line_pointer = save_input_line_pointer;
4802 if (gotfree_input_line)
4803 free (gotfree_input_line);
4804
4805 /* We do this to make sure that the section symbol is in
4806 the symbol table. We will ultimately change the relocation
4807 to be relative to the beginning of the section. */
4808 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
4809 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
4810 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTOFF64)
4811 {
4812 if (exp->X_op != O_symbol)
4813 {
4814 as_bad (_("bad expression used with @%s"),
4815 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
4816 ? "GOTPCREL"
4817 : "GOTOFF"));
4818 return 0;
4819 }
4820
4821 if (S_IS_LOCAL (exp->X_add_symbol)
4822 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
4823 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
4824 exp->X_op = O_subtract;
4825 exp->X_op_symbol = GOT_symbol;
4826 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
4827 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
4828 else if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTOFF64)
4829 i.reloc[this_operand] = BFD_RELOC_64;
4830 else
4831 i.reloc[this_operand] = BFD_RELOC_32;
4832 }
4833
4834 if (exp->X_op == O_absent || exp->X_op == O_big)
4835 {
4836 /* Missing or bad expr becomes absolute 0. */
4837 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
4838 disp_start);
4839 exp->X_op = O_constant;
4840 exp->X_add_number = 0;
4841 exp->X_add_symbol = (symbolS *) 0;
4842 exp->X_op_symbol = (symbolS *) 0;
4843 }
4844
4845 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4846 if (exp->X_op != O_constant
4847 && OUTPUT_FLAVOR == bfd_target_aout_flavour
4848 && exp_seg != absolute_section
4849 && exp_seg != text_section
4850 && exp_seg != data_section
4851 && exp_seg != bss_section
4852 && exp_seg != undefined_section
4853 && !bfd_is_com_section (exp_seg))
4854 {
4855 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
4856 return 0;
4857 }
4858 #endif
4859
4860 if (!(i.types[this_operand] & ~Disp))
4861 i.types[this_operand] &= types;
4862
4863 return 1;
4864 }
4865
4866 /* Make sure the memory operand we've been dealt is valid.
4867 Return 1 on success, 0 on a failure. */
4868
4869 static int
4870 i386_index_check (const char *operand_string)
4871 {
4872 int ok;
4873 #if INFER_ADDR_PREFIX
4874 int fudged = 0;
4875
4876 tryprefix:
4877 #endif
4878 ok = 1;
4879 if ((current_templates->start->cpu_flags & CpuSVME)
4880 && current_templates->end[-1].operand_types[0] == AnyMem)
4881 {
4882 /* Memory operands of SVME insns are special in that they only allow
4883 rAX as their memory address and ignore any segment override. */
4884 unsigned RegXX;
4885
4886 /* SKINIT is even more restrictive: it always requires EAX. */
4887 if (strcmp (current_templates->start->name, "skinit") == 0)
4888 RegXX = Reg32;
4889 else if (flag_code == CODE_64BIT)
4890 RegXX = i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32;
4891 else
4892 RegXX = ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)
4893 ? Reg16
4894 : Reg32);
4895 if (!i.base_reg
4896 || !(i.base_reg->reg_type & Acc)
4897 || !(i.base_reg->reg_type & RegXX)
4898 || i.index_reg
4899 || (i.types[0] & Disp))
4900 ok = 0;
4901 }
4902 else if (flag_code == CODE_64BIT)
4903 {
4904 unsigned RegXX = (i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32);
4905
4906 if ((i.base_reg
4907 && ((i.base_reg->reg_type & RegXX) == 0)
4908 && (i.base_reg->reg_type != BaseIndex
4909 || i.index_reg))
4910 || (i.index_reg
4911 && ((i.index_reg->reg_type & (RegXX | BaseIndex))
4912 != (RegXX | BaseIndex))))
4913 ok = 0;
4914 }
4915 else
4916 {
4917 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
4918 {
4919 /* 16bit checks. */
4920 if ((i.base_reg
4921 && ((i.base_reg->reg_type & (Reg16 | BaseIndex | RegRex))
4922 != (Reg16 | BaseIndex)))
4923 || (i.index_reg
4924 && (((i.index_reg->reg_type & (Reg16 | BaseIndex))
4925 != (Reg16 | BaseIndex))
4926 || !(i.base_reg
4927 && i.base_reg->reg_num < 6
4928 && i.index_reg->reg_num >= 6
4929 && i.log2_scale_factor == 0))))
4930 ok = 0;
4931 }
4932 else
4933 {
4934 /* 32bit checks. */
4935 if ((i.base_reg
4936 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
4937 || (i.index_reg
4938 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
4939 != (Reg32 | BaseIndex))))
4940 ok = 0;
4941 }
4942 }
4943 if (!ok)
4944 {
4945 #if INFER_ADDR_PREFIX
4946 if (i.prefix[ADDR_PREFIX] == 0)
4947 {
4948 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
4949 i.prefixes += 1;
4950 /* Change the size of any displacement too. At most one of
4951 Disp16 or Disp32 is set.
4952 FIXME. There doesn't seem to be any real need for separate
4953 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
4954 Removing them would probably clean up the code quite a lot. */
4955 if (flag_code != CODE_64BIT
4956 && (i.types[this_operand] & (Disp16 | Disp32)))
4957 i.types[this_operand] ^= (Disp16 | Disp32);
4958 fudged = 1;
4959 goto tryprefix;
4960 }
4961 if (fudged)
4962 as_bad (_("`%s' is not a valid base/index expression"),
4963 operand_string);
4964 else
4965 #endif
4966 as_bad (_("`%s' is not a valid %s bit base/index expression"),
4967 operand_string,
4968 flag_code_names[flag_code]);
4969 }
4970 return ok;
4971 }
4972
4973 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
4974 on error. */
4975
4976 static int
4977 i386_operand (char *operand_string)
4978 {
4979 const reg_entry *r;
4980 char *end_op;
4981 char *op_string = operand_string;
4982
4983 if (is_space_char (*op_string))
4984 ++op_string;
4985
4986 /* We check for an absolute prefix (differentiating,
4987 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
4988 if (*op_string == ABSOLUTE_PREFIX)
4989 {
4990 ++op_string;
4991 if (is_space_char (*op_string))
4992 ++op_string;
4993 i.types[this_operand] |= JumpAbsolute;
4994 }
4995
4996 /* Check if operand is a register. */
4997 if ((r = parse_register (op_string, &end_op)) != NULL)
4998 {
4999 /* Check for a segment override by searching for ':' after a
5000 segment register. */
5001 op_string = end_op;
5002 if (is_space_char (*op_string))
5003 ++op_string;
5004 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
5005 {
5006 switch (r->reg_num)
5007 {
5008 case 0:
5009 i.seg[i.mem_operands] = &es;
5010 break;
5011 case 1:
5012 i.seg[i.mem_operands] = &cs;
5013 break;
5014 case 2:
5015 i.seg[i.mem_operands] = &ss;
5016 break;
5017 case 3:
5018 i.seg[i.mem_operands] = &ds;
5019 break;
5020 case 4:
5021 i.seg[i.mem_operands] = &fs;
5022 break;
5023 case 5:
5024 i.seg[i.mem_operands] = &gs;
5025 break;
5026 }
5027
5028 /* Skip the ':' and whitespace. */
5029 ++op_string;
5030 if (is_space_char (*op_string))
5031 ++op_string;
5032
5033 if (!is_digit_char (*op_string)
5034 && !is_identifier_char (*op_string)
5035 && *op_string != '('
5036 && *op_string != ABSOLUTE_PREFIX)
5037 {
5038 as_bad (_("bad memory operand `%s'"), op_string);
5039 return 0;
5040 }
5041 /* Handle case of %es:*foo. */
5042 if (*op_string == ABSOLUTE_PREFIX)
5043 {
5044 ++op_string;
5045 if (is_space_char (*op_string))
5046 ++op_string;
5047 i.types[this_operand] |= JumpAbsolute;
5048 }
5049 goto do_memory_reference;
5050 }
5051 if (*op_string)
5052 {
5053 as_bad (_("junk `%s' after register"), op_string);
5054 return 0;
5055 }
5056 i.types[this_operand] |= r->reg_type & ~BaseIndex;
5057 i.op[this_operand].regs = r;
5058 i.reg_operands++;
5059 }
5060 else if (*op_string == REGISTER_PREFIX)
5061 {
5062 as_bad (_("bad register name `%s'"), op_string);
5063 return 0;
5064 }
5065 else if (*op_string == IMMEDIATE_PREFIX)
5066 {
5067 ++op_string;
5068 if (i.types[this_operand] & JumpAbsolute)
5069 {
5070 as_bad (_("immediate operand illegal with absolute jump"));
5071 return 0;
5072 }
5073 if (!i386_immediate (op_string))
5074 return 0;
5075 }
5076 else if (is_digit_char (*op_string)
5077 || is_identifier_char (*op_string)
5078 || *op_string == '(')
5079 {
5080 /* This is a memory reference of some sort. */
5081 char *base_string;
5082
5083 /* Start and end of displacement string expression (if found). */
5084 char *displacement_string_start;
5085 char *displacement_string_end;
5086
5087 do_memory_reference:
5088 if ((i.mem_operands == 1
5089 && (current_templates->start->opcode_modifier & IsString) == 0)
5090 || i.mem_operands == 2)
5091 {
5092 as_bad (_("too many memory references for `%s'"),
5093 current_templates->start->name);
5094 return 0;
5095 }
5096
5097 /* Check for base index form. We detect the base index form by
5098 looking for an ')' at the end of the operand, searching
5099 for the '(' matching it, and finding a REGISTER_PREFIX or ','
5100 after the '('. */
5101 base_string = op_string + strlen (op_string);
5102
5103 --base_string;
5104 if (is_space_char (*base_string))
5105 --base_string;
5106
5107 /* If we only have a displacement, set-up for it to be parsed later. */
5108 displacement_string_start = op_string;
5109 displacement_string_end = base_string + 1;
5110
5111 if (*base_string == ')')
5112 {
5113 char *temp_string;
5114 unsigned int parens_balanced = 1;
5115 /* We've already checked that the number of left & right ()'s are
5116 equal, so this loop will not be infinite. */
5117 do
5118 {
5119 base_string--;
5120 if (*base_string == ')')
5121 parens_balanced++;
5122 if (*base_string == '(')
5123 parens_balanced--;
5124 }
5125 while (parens_balanced);
5126
5127 temp_string = base_string;
5128
5129 /* Skip past '(' and whitespace. */
5130 ++base_string;
5131 if (is_space_char (*base_string))
5132 ++base_string;
5133
5134 if (*base_string == ','
5135 || ((i.base_reg = parse_register (base_string, &end_op))
5136 != NULL))
5137 {
5138 displacement_string_end = temp_string;
5139
5140 i.types[this_operand] |= BaseIndex;
5141
5142 if (i.base_reg)
5143 {
5144 base_string = end_op;
5145 if (is_space_char (*base_string))
5146 ++base_string;
5147 }
5148
5149 /* There may be an index reg or scale factor here. */
5150 if (*base_string == ',')
5151 {
5152 ++base_string;
5153 if (is_space_char (*base_string))
5154 ++base_string;
5155
5156 if ((i.index_reg = parse_register (base_string, &end_op))
5157 != NULL)
5158 {
5159 base_string = end_op;
5160 if (is_space_char (*base_string))
5161 ++base_string;
5162 if (*base_string == ',')
5163 {
5164 ++base_string;
5165 if (is_space_char (*base_string))
5166 ++base_string;
5167 }
5168 else if (*base_string != ')')
5169 {
5170 as_bad (_("expecting `,' or `)' "
5171 "after index register in `%s'"),
5172 operand_string);
5173 return 0;
5174 }
5175 }
5176 else if (*base_string == REGISTER_PREFIX)
5177 {
5178 as_bad (_("bad register name `%s'"), base_string);
5179 return 0;
5180 }
5181
5182 /* Check for scale factor. */
5183 if (*base_string != ')')
5184 {
5185 char *end_scale = i386_scale (base_string);
5186
5187 if (!end_scale)
5188 return 0;
5189
5190 base_string = end_scale;
5191 if (is_space_char (*base_string))
5192 ++base_string;
5193 if (*base_string != ')')
5194 {
5195 as_bad (_("expecting `)' "
5196 "after scale factor in `%s'"),
5197 operand_string);
5198 return 0;
5199 }
5200 }
5201 else if (!i.index_reg)
5202 {
5203 as_bad (_("expecting index register or scale factor "
5204 "after `,'; got '%c'"),
5205 *base_string);
5206 return 0;
5207 }
5208 }
5209 else if (*base_string != ')')
5210 {
5211 as_bad (_("expecting `,' or `)' "
5212 "after base register in `%s'"),
5213 operand_string);
5214 return 0;
5215 }
5216 }
5217 else if (*base_string == REGISTER_PREFIX)
5218 {
5219 as_bad (_("bad register name `%s'"), base_string);
5220 return 0;
5221 }
5222 }
5223
5224 /* If there's an expression beginning the operand, parse it,
5225 assuming displacement_string_start and
5226 displacement_string_end are meaningful. */
5227 if (displacement_string_start != displacement_string_end)
5228 {
5229 if (!i386_displacement (displacement_string_start,
5230 displacement_string_end))
5231 return 0;
5232 }
5233
5234 /* Special case for (%dx) while doing input/output op. */
5235 if (i.base_reg
5236 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
5237 && i.index_reg == 0
5238 && i.log2_scale_factor == 0
5239 && i.seg[i.mem_operands] == 0
5240 && (i.types[this_operand] & Disp) == 0)
5241 {
5242 i.types[this_operand] = InOutPortReg;
5243 return 1;
5244 }
5245
5246 if (i386_index_check (operand_string) == 0)
5247 return 0;
5248 i.mem_operands++;
5249 }
5250 else
5251 {
5252 /* It's not a memory operand; argh! */
5253 as_bad (_("invalid char %s beginning operand %d `%s'"),
5254 output_invalid (*op_string),
5255 this_operand + 1,
5256 op_string);
5257 return 0;
5258 }
5259 return 1; /* Normal return. */
5260 }
5261 \f
5262 /* md_estimate_size_before_relax()
5263
5264 Called just before relax() for rs_machine_dependent frags. The x86
5265 assembler uses these frags to handle variable size jump
5266 instructions.
5267
5268 Any symbol that is now undefined will not become defined.
5269 Return the correct fr_subtype in the frag.
5270 Return the initial "guess for variable size of frag" to caller.
5271 The guess is actually the growth beyond the fixed part. Whatever
5272 we do to grow the fixed or variable part contributes to our
5273 returned value. */
5274
5275 int
5276 md_estimate_size_before_relax (fragP, segment)
5277 fragS *fragP;
5278 segT segment;
5279 {
5280 /* We've already got fragP->fr_subtype right; all we have to do is
5281 check for un-relaxable symbols. On an ELF system, we can't relax
5282 an externally visible symbol, because it may be overridden by a
5283 shared library. */
5284 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
5285 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5286 || (IS_ELF
5287 && (S_IS_EXTERNAL (fragP->fr_symbol)
5288 || S_IS_WEAK (fragP->fr_symbol)))
5289 #endif
5290 )
5291 {
5292 /* Symbol is undefined in this segment, or we need to keep a
5293 reloc so that weak symbols can be overridden. */
5294 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
5295 enum bfd_reloc_code_real reloc_type;
5296 unsigned char *opcode;
5297 int old_fr_fix;
5298
5299 if (fragP->fr_var != NO_RELOC)
5300 reloc_type = fragP->fr_var;
5301 else if (size == 2)
5302 reloc_type = BFD_RELOC_16_PCREL;
5303 else
5304 reloc_type = BFD_RELOC_32_PCREL;
5305
5306 old_fr_fix = fragP->fr_fix;
5307 opcode = (unsigned char *) fragP->fr_opcode;
5308
5309 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
5310 {
5311 case UNCOND_JUMP:
5312 /* Make jmp (0xeb) a (d)word displacement jump. */
5313 opcode[0] = 0xe9;
5314 fragP->fr_fix += size;
5315 fix_new (fragP, old_fr_fix, size,
5316 fragP->fr_symbol,
5317 fragP->fr_offset, 1,
5318 reloc_type);
5319 break;
5320
5321 case COND_JUMP86:
5322 if (size == 2
5323 && (!no_cond_jump_promotion || fragP->fr_var != NO_RELOC))
5324 {
5325 /* Negate the condition, and branch past an
5326 unconditional jump. */
5327 opcode[0] ^= 1;
5328 opcode[1] = 3;
5329 /* Insert an unconditional jump. */
5330 opcode[2] = 0xe9;
5331 /* We added two extra opcode bytes, and have a two byte
5332 offset. */
5333 fragP->fr_fix += 2 + 2;
5334 fix_new (fragP, old_fr_fix + 2, 2,
5335 fragP->fr_symbol,
5336 fragP->fr_offset, 1,
5337 reloc_type);
5338 break;
5339 }
5340 /* Fall through. */
5341
5342 case COND_JUMP:
5343 if (no_cond_jump_promotion && fragP->fr_var == NO_RELOC)
5344 {
5345 fixS *fixP;
5346
5347 fragP->fr_fix += 1;
5348 fixP = fix_new (fragP, old_fr_fix, 1,
5349 fragP->fr_symbol,
5350 fragP->fr_offset, 1,
5351 BFD_RELOC_8_PCREL);
5352 fixP->fx_signed = 1;
5353 break;
5354 }
5355
5356 /* This changes the byte-displacement jump 0x7N
5357 to the (d)word-displacement jump 0x0f,0x8N. */
5358 opcode[1] = opcode[0] + 0x10;
5359 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5360 /* We've added an opcode byte. */
5361 fragP->fr_fix += 1 + size;
5362 fix_new (fragP, old_fr_fix + 1, size,
5363 fragP->fr_symbol,
5364 fragP->fr_offset, 1,
5365 reloc_type);
5366 break;
5367
5368 default:
5369 BAD_CASE (fragP->fr_subtype);
5370 break;
5371 }
5372 frag_wane (fragP);
5373 return fragP->fr_fix - old_fr_fix;
5374 }
5375
5376 /* Guess size depending on current relax state. Initially the relax
5377 state will correspond to a short jump and we return 1, because
5378 the variable part of the frag (the branch offset) is one byte
5379 long. However, we can relax a section more than once and in that
5380 case we must either set fr_subtype back to the unrelaxed state,
5381 or return the value for the appropriate branch. */
5382 return md_relax_table[fragP->fr_subtype].rlx_length;
5383 }
5384
5385 /* Called after relax() is finished.
5386
5387 In: Address of frag.
5388 fr_type == rs_machine_dependent.
5389 fr_subtype is what the address relaxed to.
5390
5391 Out: Any fixSs and constants are set up.
5392 Caller will turn frag into a ".space 0". */
5393
5394 void
5395 md_convert_frag (abfd, sec, fragP)
5396 bfd *abfd ATTRIBUTE_UNUSED;
5397 segT sec ATTRIBUTE_UNUSED;
5398 fragS *fragP;
5399 {
5400 unsigned char *opcode;
5401 unsigned char *where_to_put_displacement = NULL;
5402 offsetT target_address;
5403 offsetT opcode_address;
5404 unsigned int extension = 0;
5405 offsetT displacement_from_opcode_start;
5406
5407 opcode = (unsigned char *) fragP->fr_opcode;
5408
5409 /* Address we want to reach in file space. */
5410 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
5411
5412 /* Address opcode resides at in file space. */
5413 opcode_address = fragP->fr_address + fragP->fr_fix;
5414
5415 /* Displacement from opcode start to fill into instruction. */
5416 displacement_from_opcode_start = target_address - opcode_address;
5417
5418 if ((fragP->fr_subtype & BIG) == 0)
5419 {
5420 /* Don't have to change opcode. */
5421 extension = 1; /* 1 opcode + 1 displacement */
5422 where_to_put_displacement = &opcode[1];
5423 }
5424 else
5425 {
5426 if (no_cond_jump_promotion
5427 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
5428 as_warn_where (fragP->fr_file, fragP->fr_line,
5429 _("long jump required"));
5430
5431 switch (fragP->fr_subtype)
5432 {
5433 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
5434 extension = 4; /* 1 opcode + 4 displacement */
5435 opcode[0] = 0xe9;
5436 where_to_put_displacement = &opcode[1];
5437 break;
5438
5439 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
5440 extension = 2; /* 1 opcode + 2 displacement */
5441 opcode[0] = 0xe9;
5442 where_to_put_displacement = &opcode[1];
5443 break;
5444
5445 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
5446 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
5447 extension = 5; /* 2 opcode + 4 displacement */
5448 opcode[1] = opcode[0] + 0x10;
5449 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5450 where_to_put_displacement = &opcode[2];
5451 break;
5452
5453 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
5454 extension = 3; /* 2 opcode + 2 displacement */
5455 opcode[1] = opcode[0] + 0x10;
5456 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5457 where_to_put_displacement = &opcode[2];
5458 break;
5459
5460 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
5461 extension = 4;
5462 opcode[0] ^= 1;
5463 opcode[1] = 3;
5464 opcode[2] = 0xe9;
5465 where_to_put_displacement = &opcode[3];
5466 break;
5467
5468 default:
5469 BAD_CASE (fragP->fr_subtype);
5470 break;
5471 }
5472 }
5473
5474 /* If size if less then four we are sure that the operand fits,
5475 but if it's 4, then it could be that the displacement is larger
5476 then -/+ 2GB. */
5477 if (DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype) == 4
5478 && object_64bit
5479 && ((addressT) (displacement_from_opcode_start - extension
5480 + ((addressT) 1 << 31))
5481 > (((addressT) 2 << 31) - 1)))
5482 {
5483 as_bad_where (fragP->fr_file, fragP->fr_line,
5484 _("jump target out of range"));
5485 /* Make us emit 0. */
5486 displacement_from_opcode_start = extension;
5487 }
5488 /* Now put displacement after opcode. */
5489 md_number_to_chars ((char *) where_to_put_displacement,
5490 (valueT) (displacement_from_opcode_start - extension),
5491 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
5492 fragP->fr_fix += extension;
5493 }
5494 \f
5495 /* Size of byte displacement jmp. */
5496 int md_short_jump_size = 2;
5497
5498 /* Size of dword displacement jmp. */
5499 int md_long_jump_size = 5;
5500
5501 void
5502 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
5503 char *ptr;
5504 addressT from_addr, to_addr;
5505 fragS *frag ATTRIBUTE_UNUSED;
5506 symbolS *to_symbol ATTRIBUTE_UNUSED;
5507 {
5508 offsetT offset;
5509
5510 offset = to_addr - (from_addr + 2);
5511 /* Opcode for byte-disp jump. */
5512 md_number_to_chars (ptr, (valueT) 0xeb, 1);
5513 md_number_to_chars (ptr + 1, (valueT) offset, 1);
5514 }
5515
5516 void
5517 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
5518 char *ptr;
5519 addressT from_addr, to_addr;
5520 fragS *frag ATTRIBUTE_UNUSED;
5521 symbolS *to_symbol ATTRIBUTE_UNUSED;
5522 {
5523 offsetT offset;
5524
5525 offset = to_addr - (from_addr + 5);
5526 md_number_to_chars (ptr, (valueT) 0xe9, 1);
5527 md_number_to_chars (ptr + 1, (valueT) offset, 4);
5528 }
5529 \f
5530 /* Apply a fixup (fixS) to segment data, once it has been determined
5531 by our caller that we have all the info we need to fix it up.
5532
5533 On the 386, immediates, displacements, and data pointers are all in
5534 the same (little-endian) format, so we don't need to care about which
5535 we are handling. */
5536
5537 void
5538 md_apply_fix (fixP, valP, seg)
5539 /* The fix we're to put in. */
5540 fixS *fixP;
5541 /* Pointer to the value of the bits. */
5542 valueT *valP;
5543 /* Segment fix is from. */
5544 segT seg ATTRIBUTE_UNUSED;
5545 {
5546 char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
5547 valueT value = *valP;
5548
5549 #if !defined (TE_Mach)
5550 if (fixP->fx_pcrel)
5551 {
5552 switch (fixP->fx_r_type)
5553 {
5554 default:
5555 break;
5556
5557 case BFD_RELOC_64:
5558 fixP->fx_r_type = BFD_RELOC_64_PCREL;
5559 break;
5560 case BFD_RELOC_32:
5561 case BFD_RELOC_X86_64_32S:
5562 fixP->fx_r_type = BFD_RELOC_32_PCREL;
5563 break;
5564 case BFD_RELOC_16:
5565 fixP->fx_r_type = BFD_RELOC_16_PCREL;
5566 break;
5567 case BFD_RELOC_8:
5568 fixP->fx_r_type = BFD_RELOC_8_PCREL;
5569 break;
5570 }
5571 }
5572
5573 if (fixP->fx_addsy != NULL
5574 && (fixP->fx_r_type == BFD_RELOC_32_PCREL
5575 || fixP->fx_r_type == BFD_RELOC_64_PCREL
5576 || fixP->fx_r_type == BFD_RELOC_16_PCREL
5577 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
5578 && !use_rela_relocations)
5579 {
5580 /* This is a hack. There should be a better way to handle this.
5581 This covers for the fact that bfd_install_relocation will
5582 subtract the current location (for partial_inplace, PC relative
5583 relocations); see more below. */
5584 #ifndef OBJ_AOUT
5585 if (IS_ELF
5586 #ifdef TE_PE
5587 || OUTPUT_FLAVOR == bfd_target_coff_flavour
5588 #endif
5589 )
5590 value += fixP->fx_where + fixP->fx_frag->fr_address;
5591 #endif
5592 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5593 if (IS_ELF)
5594 {
5595 segT sym_seg = S_GET_SEGMENT (fixP->fx_addsy);
5596
5597 if ((sym_seg == seg
5598 || (symbol_section_p (fixP->fx_addsy)
5599 && sym_seg != absolute_section))
5600 && !generic_force_reloc (fixP))
5601 {
5602 /* Yes, we add the values in twice. This is because
5603 bfd_install_relocation subtracts them out again. I think
5604 bfd_install_relocation is broken, but I don't dare change
5605 it. FIXME. */
5606 value += fixP->fx_where + fixP->fx_frag->fr_address;
5607 }
5608 }
5609 #endif
5610 #if defined (OBJ_COFF) && defined (TE_PE)
5611 /* For some reason, the PE format does not store a
5612 section address offset for a PC relative symbol. */
5613 if (S_GET_SEGMENT (fixP->fx_addsy) != seg
5614 || S_IS_WEAK (fixP->fx_addsy))
5615 value += md_pcrel_from (fixP);
5616 #endif
5617 }
5618
5619 /* Fix a few things - the dynamic linker expects certain values here,
5620 and we must not disappoint it. */
5621 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5622 if (IS_ELF && fixP->fx_addsy)
5623 switch (fixP->fx_r_type)
5624 {
5625 case BFD_RELOC_386_PLT32:
5626 case BFD_RELOC_X86_64_PLT32:
5627 /* Make the jump instruction point to the address of the operand. At
5628 runtime we merely add the offset to the actual PLT entry. */
5629 value = -4;
5630 break;
5631
5632 case BFD_RELOC_386_TLS_GD:
5633 case BFD_RELOC_386_TLS_LDM:
5634 case BFD_RELOC_386_TLS_IE_32:
5635 case BFD_RELOC_386_TLS_IE:
5636 case BFD_RELOC_386_TLS_GOTIE:
5637 case BFD_RELOC_386_TLS_GOTDESC:
5638 case BFD_RELOC_X86_64_TLSGD:
5639 case BFD_RELOC_X86_64_TLSLD:
5640 case BFD_RELOC_X86_64_GOTTPOFF:
5641 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
5642 value = 0; /* Fully resolved at runtime. No addend. */
5643 /* Fallthrough */
5644 case BFD_RELOC_386_TLS_LE:
5645 case BFD_RELOC_386_TLS_LDO_32:
5646 case BFD_RELOC_386_TLS_LE_32:
5647 case BFD_RELOC_X86_64_DTPOFF32:
5648 case BFD_RELOC_X86_64_DTPOFF64:
5649 case BFD_RELOC_X86_64_TPOFF32:
5650 case BFD_RELOC_X86_64_TPOFF64:
5651 S_SET_THREAD_LOCAL (fixP->fx_addsy);
5652 break;
5653
5654 case BFD_RELOC_386_TLS_DESC_CALL:
5655 case BFD_RELOC_X86_64_TLSDESC_CALL:
5656 value = 0; /* Fully resolved at runtime. No addend. */
5657 S_SET_THREAD_LOCAL (fixP->fx_addsy);
5658 fixP->fx_done = 0;
5659 return;
5660
5661 case BFD_RELOC_386_GOT32:
5662 case BFD_RELOC_X86_64_GOT32:
5663 value = 0; /* Fully resolved at runtime. No addend. */
5664 break;
5665
5666 case BFD_RELOC_VTABLE_INHERIT:
5667 case BFD_RELOC_VTABLE_ENTRY:
5668 fixP->fx_done = 0;
5669 return;
5670
5671 default:
5672 break;
5673 }
5674 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
5675 *valP = value;
5676 #endif /* !defined (TE_Mach) */
5677
5678 /* Are we finished with this relocation now? */
5679 if (fixP->fx_addsy == NULL)
5680 fixP->fx_done = 1;
5681 else if (use_rela_relocations)
5682 {
5683 fixP->fx_no_overflow = 1;
5684 /* Remember value for tc_gen_reloc. */
5685 fixP->fx_addnumber = value;
5686 value = 0;
5687 }
5688
5689 md_number_to_chars (p, value, fixP->fx_size);
5690 }
5691 \f
5692 #define MAX_LITTLENUMS 6
5693
5694 /* Turn the string pointed to by litP into a floating point constant
5695 of type TYPE, and emit the appropriate bytes. The number of
5696 LITTLENUMS emitted is stored in *SIZEP. An error message is
5697 returned, or NULL on OK. */
5698
5699 char *
5700 md_atof (type, litP, sizeP)
5701 int type;
5702 char *litP;
5703 int *sizeP;
5704 {
5705 int prec;
5706 LITTLENUM_TYPE words[MAX_LITTLENUMS];
5707 LITTLENUM_TYPE *wordP;
5708 char *t;
5709
5710 switch (type)
5711 {
5712 case 'f':
5713 case 'F':
5714 prec = 2;
5715 break;
5716
5717 case 'd':
5718 case 'D':
5719 prec = 4;
5720 break;
5721
5722 case 'x':
5723 case 'X':
5724 prec = 5;
5725 break;
5726
5727 default:
5728 *sizeP = 0;
5729 return _("Bad call to md_atof ()");
5730 }
5731 t = atof_ieee (input_line_pointer, type, words);
5732 if (t)
5733 input_line_pointer = t;
5734
5735 *sizeP = prec * sizeof (LITTLENUM_TYPE);
5736 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
5737 the bigendian 386. */
5738 for (wordP = words + prec - 1; prec--;)
5739 {
5740 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
5741 litP += sizeof (LITTLENUM_TYPE);
5742 }
5743 return 0;
5744 }
5745 \f
5746 static char output_invalid_buf[sizeof (unsigned char) * 2 + 6];
5747
5748 static char *
5749 output_invalid (int c)
5750 {
5751 if (ISPRINT (c))
5752 snprintf (output_invalid_buf, sizeof (output_invalid_buf),
5753 "'%c'", c);
5754 else
5755 snprintf (output_invalid_buf, sizeof (output_invalid_buf),
5756 "(0x%x)", (unsigned char) c);
5757 return output_invalid_buf;
5758 }
5759
5760 /* REG_STRING starts *before* REGISTER_PREFIX. */
5761
5762 static const reg_entry *
5763 parse_real_register (char *reg_string, char **end_op)
5764 {
5765 char *s = reg_string;
5766 char *p;
5767 char reg_name_given[MAX_REG_NAME_SIZE + 1];
5768 const reg_entry *r;
5769
5770 /* Skip possible REGISTER_PREFIX and possible whitespace. */
5771 if (*s == REGISTER_PREFIX)
5772 ++s;
5773
5774 if (is_space_char (*s))
5775 ++s;
5776
5777 p = reg_name_given;
5778 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
5779 {
5780 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
5781 return (const reg_entry *) NULL;
5782 s++;
5783 }
5784
5785 /* For naked regs, make sure that we are not dealing with an identifier.
5786 This prevents confusing an identifier like `eax_var' with register
5787 `eax'. */
5788 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
5789 return (const reg_entry *) NULL;
5790
5791 *end_op = s;
5792
5793 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
5794
5795 /* Handle floating point regs, allowing spaces in the (i) part. */
5796 if (r == i386_regtab /* %st is first entry of table */)
5797 {
5798 if (is_space_char (*s))
5799 ++s;
5800 if (*s == '(')
5801 {
5802 ++s;
5803 if (is_space_char (*s))
5804 ++s;
5805 if (*s >= '0' && *s <= '7')
5806 {
5807 r = &i386_float_regtab[*s - '0'];
5808 ++s;
5809 if (is_space_char (*s))
5810 ++s;
5811 if (*s == ')')
5812 {
5813 *end_op = s + 1;
5814 return r;
5815 }
5816 }
5817 /* We have "%st(" then garbage. */
5818 return (const reg_entry *) NULL;
5819 }
5820 }
5821
5822 if (r != NULL
5823 && ((r->reg_flags & (RegRex64 | RegRex)) | (r->reg_type & Reg64)) != 0
5824 && (r->reg_type != Control || !(cpu_arch_flags & CpuSledgehammer))
5825 && flag_code != CODE_64BIT)
5826 return (const reg_entry *) NULL;
5827
5828 return r;
5829 }
5830
5831 /* REG_STRING starts *before* REGISTER_PREFIX. */
5832
5833 static const reg_entry *
5834 parse_register (char *reg_string, char **end_op)
5835 {
5836 const reg_entry *r;
5837
5838 if (*reg_string == REGISTER_PREFIX || allow_naked_reg)
5839 r = parse_real_register (reg_string, end_op);
5840 else
5841 r = NULL;
5842 if (!r)
5843 {
5844 char *save = input_line_pointer;
5845 char c;
5846 symbolS *symbolP;
5847
5848 input_line_pointer = reg_string;
5849 c = get_symbol_end ();
5850 symbolP = symbol_find (reg_string);
5851 if (symbolP && S_GET_SEGMENT (symbolP) == reg_section)
5852 {
5853 const expressionS *e = symbol_get_value_expression (symbolP);
5854
5855 know (e->X_op == O_register);
5856 know (e->X_add_number >= 0
5857 && (valueT) e->X_add_number < i386_regtab_size);
5858 r = i386_regtab + e->X_add_number;
5859 *end_op = input_line_pointer;
5860 }
5861 *input_line_pointer = c;
5862 input_line_pointer = save;
5863 }
5864 return r;
5865 }
5866
5867 int
5868 i386_parse_name (char *name, expressionS *e, char *nextcharP)
5869 {
5870 const reg_entry *r;
5871 char *end = input_line_pointer;
5872
5873 *end = *nextcharP;
5874 r = parse_register (name, &input_line_pointer);
5875 if (r && end <= input_line_pointer)
5876 {
5877 *nextcharP = *input_line_pointer;
5878 *input_line_pointer = 0;
5879 e->X_op = O_register;
5880 e->X_add_number = r - i386_regtab;
5881 return 1;
5882 }
5883 input_line_pointer = end;
5884 *end = 0;
5885 return 0;
5886 }
5887
5888 void
5889 md_operand (expressionS *e)
5890 {
5891 if (*input_line_pointer == REGISTER_PREFIX)
5892 {
5893 char *end;
5894 const reg_entry *r = parse_real_register (input_line_pointer, &end);
5895
5896 if (r)
5897 {
5898 e->X_op = O_register;
5899 e->X_add_number = r - i386_regtab;
5900 input_line_pointer = end;
5901 }
5902 }
5903 }
5904
5905 \f
5906 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5907 const char *md_shortopts = "kVQ:sqn";
5908 #else
5909 const char *md_shortopts = "qn";
5910 #endif
5911
5912 #define OPTION_32 (OPTION_MD_BASE + 0)
5913 #define OPTION_64 (OPTION_MD_BASE + 1)
5914 #define OPTION_DIVIDE (OPTION_MD_BASE + 2)
5915 #define OPTION_MARCH (OPTION_MD_BASE + 3)
5916 #define OPTION_MTUNE (OPTION_MD_BASE + 4)
5917
5918 struct option md_longopts[] =
5919 {
5920 {"32", no_argument, NULL, OPTION_32},
5921 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
5922 {"64", no_argument, NULL, OPTION_64},
5923 #endif
5924 {"divide", no_argument, NULL, OPTION_DIVIDE},
5925 {"march", required_argument, NULL, OPTION_MARCH},
5926 {"mtune", required_argument, NULL, OPTION_MTUNE},
5927 {NULL, no_argument, NULL, 0}
5928 };
5929 size_t md_longopts_size = sizeof (md_longopts);
5930
5931 int
5932 md_parse_option (int c, char *arg)
5933 {
5934 unsigned int i;
5935
5936 switch (c)
5937 {
5938 case 'n':
5939 optimize_align_code = 0;
5940 break;
5941
5942 case 'q':
5943 quiet_warnings = 1;
5944 break;
5945
5946 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5947 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
5948 should be emitted or not. FIXME: Not implemented. */
5949 case 'Q':
5950 break;
5951
5952 /* -V: SVR4 argument to print version ID. */
5953 case 'V':
5954 print_version_id ();
5955 break;
5956
5957 /* -k: Ignore for FreeBSD compatibility. */
5958 case 'k':
5959 break;
5960
5961 case 's':
5962 /* -s: On i386 Solaris, this tells the native assembler to use
5963 .stab instead of .stab.excl. We always use .stab anyhow. */
5964 break;
5965 #endif
5966 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
5967 case OPTION_64:
5968 {
5969 const char **list, **l;
5970
5971 list = bfd_target_list ();
5972 for (l = list; *l != NULL; l++)
5973 if (CONST_STRNEQ (*l, "elf64-x86-64")
5974 || strcmp (*l, "coff-x86-64") == 0
5975 || strcmp (*l, "pe-x86-64") == 0
5976 || strcmp (*l, "pei-x86-64") == 0)
5977 {
5978 default_arch = "x86_64";
5979 break;
5980 }
5981 if (*l == NULL)
5982 as_fatal (_("No compiled in support for x86_64"));
5983 free (list);
5984 }
5985 break;
5986 #endif
5987
5988 case OPTION_32:
5989 default_arch = "i386";
5990 break;
5991
5992 case OPTION_DIVIDE:
5993 #ifdef SVR4_COMMENT_CHARS
5994 {
5995 char *n, *t;
5996 const char *s;
5997
5998 n = (char *) xmalloc (strlen (i386_comment_chars) + 1);
5999 t = n;
6000 for (s = i386_comment_chars; *s != '\0'; s++)
6001 if (*s != '/')
6002 *t++ = *s;
6003 *t = '\0';
6004 i386_comment_chars = n;
6005 }
6006 #endif
6007 break;
6008
6009 case OPTION_MARCH:
6010 if (*arg == '.')
6011 as_fatal (_("Invalid -march= option: `%s'"), arg);
6012 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
6013 {
6014 if (strcmp (arg, cpu_arch [i].name) == 0)
6015 {
6016 cpu_arch_isa = cpu_arch[i].type;
6017 cpu_arch_isa_flags = cpu_arch[i].flags;
6018 if (!cpu_arch_tune_set)
6019 {
6020 cpu_arch_tune = cpu_arch_isa;
6021 cpu_arch_tune_flags = cpu_arch_isa_flags;
6022 }
6023 break;
6024 }
6025 }
6026 if (i >= ARRAY_SIZE (cpu_arch))
6027 as_fatal (_("Invalid -march= option: `%s'"), arg);
6028 break;
6029
6030 case OPTION_MTUNE:
6031 if (*arg == '.')
6032 as_fatal (_("Invalid -mtune= option: `%s'"), arg);
6033 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
6034 {
6035 if (strcmp (arg, cpu_arch [i].name) == 0)
6036 {
6037 cpu_arch_tune_set = 1;
6038 cpu_arch_tune = cpu_arch [i].type;
6039 cpu_arch_tune_flags = cpu_arch[i].flags;
6040 break;
6041 }
6042 }
6043 if (i >= ARRAY_SIZE (cpu_arch))
6044 as_fatal (_("Invalid -mtune= option: `%s'"), arg);
6045 break;
6046
6047 default:
6048 return 0;
6049 }
6050 return 1;
6051 }
6052
6053 void
6054 md_show_usage (stream)
6055 FILE *stream;
6056 {
6057 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
6058 fprintf (stream, _("\
6059 -Q ignored\n\
6060 -V print assembler version number\n\
6061 -k ignored\n"));
6062 #endif
6063 fprintf (stream, _("\
6064 -n Do not optimize code alignment\n\
6065 -q quieten some warnings\n"));
6066 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
6067 fprintf (stream, _("\
6068 -s ignored\n"));
6069 #endif
6070 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
6071 fprintf (stream, _("\
6072 --32/--64 generate 32bit/64bit code\n"));
6073 #endif
6074 #ifdef SVR4_COMMENT_CHARS
6075 fprintf (stream, _("\
6076 --divide do not treat `/' as a comment character\n"));
6077 #else
6078 fprintf (stream, _("\
6079 --divide ignored\n"));
6080 #endif
6081 fprintf (stream, _("\
6082 -march=CPU/-mtune=CPU generate code/optimize for CPU, where CPU is one of:\n\
6083 i386, i486, pentium, pentiumpro, pentium4, nocona,\n\
6084 core, core2, k6, athlon, k8, generic32, generic64\n"));
6085
6086 }
6087
6088 #if defined(TE_PEP)
6089 const char *
6090 x86_64_target_format (void)
6091 {
6092 if (strcmp (default_arch, "x86_64") == 0)
6093 {
6094 set_code_flag (CODE_64BIT);
6095 return COFF_TARGET_FORMAT;
6096 }
6097 else if (strcmp (default_arch, "i386") == 0)
6098 {
6099 set_code_flag (CODE_32BIT);
6100 return "coff-i386";
6101 }
6102
6103 as_fatal (_("Unknown architecture"));
6104 return NULL;
6105 }
6106 #endif
6107
6108 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
6109 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
6110
6111 /* Pick the target format to use. */
6112
6113 const char *
6114 i386_target_format (void)
6115 {
6116 if (!strcmp (default_arch, "x86_64"))
6117 {
6118 set_code_flag (CODE_64BIT);
6119 if (cpu_arch_isa_flags == 0)
6120 cpu_arch_isa_flags = Cpu186|Cpu286|Cpu386|Cpu486
6121 |Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2
6122 |CpuSSE|CpuSSE2;
6123 if (cpu_arch_tune_flags == 0)
6124 cpu_arch_tune_flags = Cpu186|Cpu286|Cpu386|Cpu486
6125 |Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2
6126 |CpuSSE|CpuSSE2;
6127 }
6128 else if (!strcmp (default_arch, "i386"))
6129 {
6130 set_code_flag (CODE_32BIT);
6131 if (cpu_arch_isa_flags == 0)
6132 cpu_arch_isa_flags = Cpu186|Cpu286|Cpu386;
6133 if (cpu_arch_tune_flags == 0)
6134 cpu_arch_tune_flags = Cpu186|Cpu286|Cpu386;
6135 }
6136 else
6137 as_fatal (_("Unknown architecture"));
6138 switch (OUTPUT_FLAVOR)
6139 {
6140 #ifdef OBJ_MAYBE_AOUT
6141 case bfd_target_aout_flavour:
6142 return AOUT_TARGET_FORMAT;
6143 #endif
6144 #ifdef OBJ_MAYBE_COFF
6145 case bfd_target_coff_flavour:
6146 return "coff-i386";
6147 #endif
6148 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
6149 case bfd_target_elf_flavour:
6150 {
6151 if (flag_code == CODE_64BIT)
6152 {
6153 object_64bit = 1;
6154 use_rela_relocations = 1;
6155 }
6156 return flag_code == CODE_64BIT ? ELF_TARGET_FORMAT64 : ELF_TARGET_FORMAT;
6157 }
6158 #endif
6159 default:
6160 abort ();
6161 return NULL;
6162 }
6163 }
6164
6165 #endif /* OBJ_MAYBE_ more than one */
6166
6167 #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
6168 void
6169 i386_elf_emit_arch_note (void)
6170 {
6171 if (IS_ELF && cpu_arch_name != NULL)
6172 {
6173 char *p;
6174 asection *seg = now_seg;
6175 subsegT subseg = now_subseg;
6176 Elf_Internal_Note i_note;
6177 Elf_External_Note e_note;
6178 asection *note_secp;
6179 int len;
6180
6181 /* Create the .note section. */
6182 note_secp = subseg_new (".note", 0);
6183 bfd_set_section_flags (stdoutput,
6184 note_secp,
6185 SEC_HAS_CONTENTS | SEC_READONLY);
6186
6187 /* Process the arch string. */
6188 len = strlen (cpu_arch_name);
6189
6190 i_note.namesz = len + 1;
6191 i_note.descsz = 0;
6192 i_note.type = NT_ARCH;
6193 p = frag_more (sizeof (e_note.namesz));
6194 md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
6195 p = frag_more (sizeof (e_note.descsz));
6196 md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
6197 p = frag_more (sizeof (e_note.type));
6198 md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
6199 p = frag_more (len + 1);
6200 strcpy (p, cpu_arch_name);
6201
6202 frag_align (2, 0, 0);
6203
6204 subseg_set (seg, subseg);
6205 }
6206 }
6207 #endif
6208 \f
6209 symbolS *
6210 md_undefined_symbol (name)
6211 char *name;
6212 {
6213 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
6214 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
6215 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
6216 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
6217 {
6218 if (!GOT_symbol)
6219 {
6220 if (symbol_find (name))
6221 as_bad (_("GOT already in symbol table"));
6222 GOT_symbol = symbol_new (name, undefined_section,
6223 (valueT) 0, &zero_address_frag);
6224 };
6225 return GOT_symbol;
6226 }
6227 return 0;
6228 }
6229
6230 /* Round up a section size to the appropriate boundary. */
6231
6232 valueT
6233 md_section_align (segment, size)
6234 segT segment ATTRIBUTE_UNUSED;
6235 valueT size;
6236 {
6237 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
6238 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
6239 {
6240 /* For a.out, force the section size to be aligned. If we don't do
6241 this, BFD will align it for us, but it will not write out the
6242 final bytes of the section. This may be a bug in BFD, but it is
6243 easier to fix it here since that is how the other a.out targets
6244 work. */
6245 int align;
6246
6247 align = bfd_get_section_alignment (stdoutput, segment);
6248 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
6249 }
6250 #endif
6251
6252 return size;
6253 }
6254
6255 /* On the i386, PC-relative offsets are relative to the start of the
6256 next instruction. That is, the address of the offset, plus its
6257 size, since the offset is always the last part of the insn. */
6258
6259 long
6260 md_pcrel_from (fixS *fixP)
6261 {
6262 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
6263 }
6264
6265 #ifndef I386COFF
6266
6267 static void
6268 s_bss (int ignore ATTRIBUTE_UNUSED)
6269 {
6270 int temp;
6271
6272 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
6273 if (IS_ELF)
6274 obj_elf_section_change_hook ();
6275 #endif
6276 temp = get_absolute_expression ();
6277 subseg_set (bss_section, (subsegT) temp);
6278 demand_empty_rest_of_line ();
6279 }
6280
6281 #endif
6282
6283 void
6284 i386_validate_fix (fixS *fixp)
6285 {
6286 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
6287 {
6288 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
6289 {
6290 if (!object_64bit)
6291 abort ();
6292 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
6293 }
6294 else
6295 {
6296 if (!object_64bit)
6297 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
6298 else
6299 fixp->fx_r_type = BFD_RELOC_X86_64_GOTOFF64;
6300 }
6301 fixp->fx_subsy = 0;
6302 }
6303 }
6304
6305 arelent *
6306 tc_gen_reloc (section, fixp)
6307 asection *section ATTRIBUTE_UNUSED;
6308 fixS *fixp;
6309 {
6310 arelent *rel;
6311 bfd_reloc_code_real_type code;
6312
6313 switch (fixp->fx_r_type)
6314 {
6315 case BFD_RELOC_X86_64_PLT32:
6316 case BFD_RELOC_X86_64_GOT32:
6317 case BFD_RELOC_X86_64_GOTPCREL:
6318 case BFD_RELOC_386_PLT32:
6319 case BFD_RELOC_386_GOT32:
6320 case BFD_RELOC_386_GOTOFF:
6321 case BFD_RELOC_386_GOTPC:
6322 case BFD_RELOC_386_TLS_GD:
6323 case BFD_RELOC_386_TLS_LDM:
6324 case BFD_RELOC_386_TLS_LDO_32:
6325 case BFD_RELOC_386_TLS_IE_32:
6326 case BFD_RELOC_386_TLS_IE:
6327 case BFD_RELOC_386_TLS_GOTIE:
6328 case BFD_RELOC_386_TLS_LE_32:
6329 case BFD_RELOC_386_TLS_LE:
6330 case BFD_RELOC_386_TLS_GOTDESC:
6331 case BFD_RELOC_386_TLS_DESC_CALL:
6332 case BFD_RELOC_X86_64_TLSGD:
6333 case BFD_RELOC_X86_64_TLSLD:
6334 case BFD_RELOC_X86_64_DTPOFF32:
6335 case BFD_RELOC_X86_64_DTPOFF64:
6336 case BFD_RELOC_X86_64_GOTTPOFF:
6337 case BFD_RELOC_X86_64_TPOFF32:
6338 case BFD_RELOC_X86_64_TPOFF64:
6339 case BFD_RELOC_X86_64_GOTOFF64:
6340 case BFD_RELOC_X86_64_GOTPC32:
6341 case BFD_RELOC_X86_64_GOT64:
6342 case BFD_RELOC_X86_64_GOTPCREL64:
6343 case BFD_RELOC_X86_64_GOTPC64:
6344 case BFD_RELOC_X86_64_GOTPLT64:
6345 case BFD_RELOC_X86_64_PLTOFF64:
6346 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
6347 case BFD_RELOC_X86_64_TLSDESC_CALL:
6348 case BFD_RELOC_RVA:
6349 case BFD_RELOC_VTABLE_ENTRY:
6350 case BFD_RELOC_VTABLE_INHERIT:
6351 #ifdef TE_PE
6352 case BFD_RELOC_32_SECREL:
6353 #endif
6354 code = fixp->fx_r_type;
6355 break;
6356 case BFD_RELOC_X86_64_32S:
6357 if (!fixp->fx_pcrel)
6358 {
6359 /* Don't turn BFD_RELOC_X86_64_32S into BFD_RELOC_32. */
6360 code = fixp->fx_r_type;
6361 break;
6362 }
6363 default:
6364 if (fixp->fx_pcrel)
6365 {
6366 switch (fixp->fx_size)
6367 {
6368 default:
6369 as_bad_where (fixp->fx_file, fixp->fx_line,
6370 _("can not do %d byte pc-relative relocation"),
6371 fixp->fx_size);
6372 code = BFD_RELOC_32_PCREL;
6373 break;
6374 case 1: code = BFD_RELOC_8_PCREL; break;
6375 case 2: code = BFD_RELOC_16_PCREL; break;
6376 case 4: code = BFD_RELOC_32_PCREL; break;
6377 #ifdef BFD64
6378 case 8: code = BFD_RELOC_64_PCREL; break;
6379 #endif
6380 }
6381 }
6382 else
6383 {
6384 switch (fixp->fx_size)
6385 {
6386 default:
6387 as_bad_where (fixp->fx_file, fixp->fx_line,
6388 _("can not do %d byte relocation"),
6389 fixp->fx_size);
6390 code = BFD_RELOC_32;
6391 break;
6392 case 1: code = BFD_RELOC_8; break;
6393 case 2: code = BFD_RELOC_16; break;
6394 case 4: code = BFD_RELOC_32; break;
6395 #ifdef BFD64
6396 case 8: code = BFD_RELOC_64; break;
6397 #endif
6398 }
6399 }
6400 break;
6401 }
6402
6403 if ((code == BFD_RELOC_32
6404 || code == BFD_RELOC_32_PCREL
6405 || code == BFD_RELOC_X86_64_32S)
6406 && GOT_symbol
6407 && fixp->fx_addsy == GOT_symbol)
6408 {
6409 if (!object_64bit)
6410 code = BFD_RELOC_386_GOTPC;
6411 else
6412 code = BFD_RELOC_X86_64_GOTPC32;
6413 }
6414 if ((code == BFD_RELOC_64 || code == BFD_RELOC_64_PCREL)
6415 && GOT_symbol
6416 && fixp->fx_addsy == GOT_symbol)
6417 {
6418 code = BFD_RELOC_X86_64_GOTPC64;
6419 }
6420
6421 rel = (arelent *) xmalloc (sizeof (arelent));
6422 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
6423 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
6424
6425 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
6426
6427 if (!use_rela_relocations)
6428 {
6429 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
6430 vtable entry to be used in the relocation's section offset. */
6431 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
6432 rel->address = fixp->fx_offset;
6433
6434 rel->addend = 0;
6435 }
6436 /* Use the rela in 64bit mode. */
6437 else
6438 {
6439 if (!fixp->fx_pcrel)
6440 rel->addend = fixp->fx_offset;
6441 else
6442 switch (code)
6443 {
6444 case BFD_RELOC_X86_64_PLT32:
6445 case BFD_RELOC_X86_64_GOT32:
6446 case BFD_RELOC_X86_64_GOTPCREL:
6447 case BFD_RELOC_X86_64_TLSGD:
6448 case BFD_RELOC_X86_64_TLSLD:
6449 case BFD_RELOC_X86_64_GOTTPOFF:
6450 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
6451 case BFD_RELOC_X86_64_TLSDESC_CALL:
6452 rel->addend = fixp->fx_offset - fixp->fx_size;
6453 break;
6454 default:
6455 rel->addend = (section->vma
6456 - fixp->fx_size
6457 + fixp->fx_addnumber
6458 + md_pcrel_from (fixp));
6459 break;
6460 }
6461 }
6462
6463 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
6464 if (rel->howto == NULL)
6465 {
6466 as_bad_where (fixp->fx_file, fixp->fx_line,
6467 _("cannot represent relocation type %s"),
6468 bfd_get_reloc_code_name (code));
6469 /* Set howto to a garbage value so that we can keep going. */
6470 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
6471 assert (rel->howto != NULL);
6472 }
6473
6474 return rel;
6475 }
6476
6477 \f
6478 /* Parse operands using Intel syntax. This implements a recursive descent
6479 parser based on the BNF grammar published in Appendix B of the MASM 6.1
6480 Programmer's Guide.
6481
6482 FIXME: We do not recognize the full operand grammar defined in the MASM
6483 documentation. In particular, all the structure/union and
6484 high-level macro operands are missing.
6485
6486 Uppercase words are terminals, lower case words are non-terminals.
6487 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
6488 bars '|' denote choices. Most grammar productions are implemented in
6489 functions called 'intel_<production>'.
6490
6491 Initial production is 'expr'.
6492
6493 addOp + | -
6494
6495 alpha [a-zA-Z]
6496
6497 binOp & | AND | \| | OR | ^ | XOR
6498
6499 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
6500
6501 constant digits [[ radixOverride ]]
6502
6503 dataType BYTE | WORD | DWORD | FWORD | QWORD | TBYTE | OWORD | XMMWORD
6504
6505 digits decdigit
6506 | digits decdigit
6507 | digits hexdigit
6508
6509 decdigit [0-9]
6510
6511 e04 e04 addOp e05
6512 | e05
6513
6514 e05 e05 binOp e06
6515 | e06
6516
6517 e06 e06 mulOp e09
6518 | e09
6519
6520 e09 OFFSET e10
6521 | SHORT e10
6522 | + e10
6523 | - e10
6524 | ~ e10
6525 | NOT e10
6526 | e09 PTR e10
6527 | e09 : e10
6528 | e10
6529
6530 e10 e10 [ expr ]
6531 | e11
6532
6533 e11 ( expr )
6534 | [ expr ]
6535 | constant
6536 | dataType
6537 | id
6538 | $
6539 | register
6540
6541 => expr expr cmpOp e04
6542 | e04
6543
6544 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
6545 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
6546
6547 hexdigit a | b | c | d | e | f
6548 | A | B | C | D | E | F
6549
6550 id alpha
6551 | id alpha
6552 | id decdigit
6553
6554 mulOp * | / | % | MOD | << | SHL | >> | SHR
6555
6556 quote " | '
6557
6558 register specialRegister
6559 | gpRegister
6560 | byteRegister
6561
6562 segmentRegister CS | DS | ES | FS | GS | SS
6563
6564 specialRegister CR0 | CR2 | CR3 | CR4
6565 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
6566 | TR3 | TR4 | TR5 | TR6 | TR7
6567
6568 We simplify the grammar in obvious places (e.g., register parsing is
6569 done by calling parse_register) and eliminate immediate left recursion
6570 to implement a recursive-descent parser.
6571
6572 expr e04 expr'
6573
6574 expr' cmpOp e04 expr'
6575 | Empty
6576
6577 e04 e05 e04'
6578
6579 e04' addOp e05 e04'
6580 | Empty
6581
6582 e05 e06 e05'
6583
6584 e05' binOp e06 e05'
6585 | Empty
6586
6587 e06 e09 e06'
6588
6589 e06' mulOp e09 e06'
6590 | Empty
6591
6592 e09 OFFSET e10 e09'
6593 | SHORT e10'
6594 | + e10'
6595 | - e10'
6596 | ~ e10'
6597 | NOT e10'
6598 | e10 e09'
6599
6600 e09' PTR e10 e09'
6601 | : e10 e09'
6602 | Empty
6603
6604 e10 e11 e10'
6605
6606 e10' [ expr ] e10'
6607 | Empty
6608
6609 e11 ( expr )
6610 | [ expr ]
6611 | BYTE
6612 | WORD
6613 | DWORD
6614 | FWORD
6615 | QWORD
6616 | TBYTE
6617 | OWORD
6618 | XMMWORD
6619 | .
6620 | $
6621 | register
6622 | id
6623 | constant */
6624
6625 /* Parsing structure for the intel syntax parser. Used to implement the
6626 semantic actions for the operand grammar. */
6627 struct intel_parser_s
6628 {
6629 char *op_string; /* The string being parsed. */
6630 int got_a_float; /* Whether the operand is a float. */
6631 int op_modifier; /* Operand modifier. */
6632 int is_mem; /* 1 if operand is memory reference. */
6633 int in_offset; /* >=1 if parsing operand of offset. */
6634 int in_bracket; /* >=1 if parsing operand in brackets. */
6635 const reg_entry *reg; /* Last register reference found. */
6636 char *disp; /* Displacement string being built. */
6637 char *next_operand; /* Resume point when splitting operands. */
6638 };
6639
6640 static struct intel_parser_s intel_parser;
6641
6642 /* Token structure for parsing intel syntax. */
6643 struct intel_token
6644 {
6645 int code; /* Token code. */
6646 const reg_entry *reg; /* Register entry for register tokens. */
6647 char *str; /* String representation. */
6648 };
6649
6650 static struct intel_token cur_token, prev_token;
6651
6652 /* Token codes for the intel parser. Since T_SHORT is already used
6653 by COFF, undefine it first to prevent a warning. */
6654 #define T_NIL -1
6655 #define T_CONST 1
6656 #define T_REG 2
6657 #define T_BYTE 3
6658 #define T_WORD 4
6659 #define T_DWORD 5
6660 #define T_FWORD 6
6661 #define T_QWORD 7
6662 #define T_TBYTE 8
6663 #define T_XMMWORD 9
6664 #undef T_SHORT
6665 #define T_SHORT 10
6666 #define T_OFFSET 11
6667 #define T_PTR 12
6668 #define T_ID 13
6669 #define T_SHL 14
6670 #define T_SHR 15
6671
6672 /* Prototypes for intel parser functions. */
6673 static int intel_match_token (int);
6674 static void intel_putback_token (void);
6675 static void intel_get_token (void);
6676 static int intel_expr (void);
6677 static int intel_e04 (void);
6678 static int intel_e05 (void);
6679 static int intel_e06 (void);
6680 static int intel_e09 (void);
6681 static int intel_e10 (void);
6682 static int intel_e11 (void);
6683
6684 static int
6685 i386_intel_operand (char *operand_string, int got_a_float)
6686 {
6687 int ret;
6688 char *p;
6689
6690 p = intel_parser.op_string = xstrdup (operand_string);
6691 intel_parser.disp = (char *) xmalloc (strlen (operand_string) + 1);
6692
6693 for (;;)
6694 {
6695 /* Initialize token holders. */
6696 cur_token.code = prev_token.code = T_NIL;
6697 cur_token.reg = prev_token.reg = NULL;
6698 cur_token.str = prev_token.str = NULL;
6699
6700 /* Initialize parser structure. */
6701 intel_parser.got_a_float = got_a_float;
6702 intel_parser.op_modifier = 0;
6703 intel_parser.is_mem = 0;
6704 intel_parser.in_offset = 0;
6705 intel_parser.in_bracket = 0;
6706 intel_parser.reg = NULL;
6707 intel_parser.disp[0] = '\0';
6708 intel_parser.next_operand = NULL;
6709
6710 /* Read the first token and start the parser. */
6711 intel_get_token ();
6712 ret = intel_expr ();
6713
6714 if (!ret)
6715 break;
6716
6717 if (cur_token.code != T_NIL)
6718 {
6719 as_bad (_("invalid operand for '%s' ('%s' unexpected)"),
6720 current_templates->start->name, cur_token.str);
6721 ret = 0;
6722 }
6723 /* If we found a memory reference, hand it over to i386_displacement
6724 to fill in the rest of the operand fields. */
6725 else if (intel_parser.is_mem)
6726 {
6727 if ((i.mem_operands == 1
6728 && (current_templates->start->opcode_modifier & IsString) == 0)
6729 || i.mem_operands == 2)
6730 {
6731 as_bad (_("too many memory references for '%s'"),
6732 current_templates->start->name);
6733 ret = 0;
6734 }
6735 else
6736 {
6737 char *s = intel_parser.disp;
6738 i.mem_operands++;
6739
6740 if (!quiet_warnings && intel_parser.is_mem < 0)
6741 /* See the comments in intel_bracket_expr. */
6742 as_warn (_("Treating `%s' as memory reference"), operand_string);
6743
6744 /* Add the displacement expression. */
6745 if (*s != '\0')
6746 ret = i386_displacement (s, s + strlen (s));
6747 if (ret)
6748 {
6749 /* Swap base and index in 16-bit memory operands like
6750 [si+bx]. Since i386_index_check is also used in AT&T
6751 mode we have to do that here. */
6752 if (i.base_reg
6753 && i.index_reg
6754 && (i.base_reg->reg_type & Reg16)
6755 && (i.index_reg->reg_type & Reg16)
6756 && i.base_reg->reg_num >= 6
6757 && i.index_reg->reg_num < 6)
6758 {
6759 const reg_entry *base = i.index_reg;
6760
6761 i.index_reg = i.base_reg;
6762 i.base_reg = base;
6763 }
6764 ret = i386_index_check (operand_string);
6765 }
6766 }
6767 }
6768
6769 /* Constant and OFFSET expressions are handled by i386_immediate. */
6770 else if ((intel_parser.op_modifier & (1 << T_OFFSET))
6771 || intel_parser.reg == NULL)
6772 ret = i386_immediate (intel_parser.disp);
6773
6774 if (intel_parser.next_operand && this_operand >= MAX_OPERANDS - 1)
6775 ret = 0;
6776 if (!ret || !intel_parser.next_operand)
6777 break;
6778 intel_parser.op_string = intel_parser.next_operand;
6779 this_operand = i.operands++;
6780 }
6781
6782 free (p);
6783 free (intel_parser.disp);
6784
6785 return ret;
6786 }
6787
6788 #define NUM_ADDRESS_REGS (!!i.base_reg + !!i.index_reg)
6789
6790 /* expr e04 expr'
6791
6792 expr' cmpOp e04 expr'
6793 | Empty */
6794 static int
6795 intel_expr (void)
6796 {
6797 /* XXX Implement the comparison operators. */
6798 return intel_e04 ();
6799 }
6800
6801 /* e04 e05 e04'
6802
6803 e04' addOp e05 e04'
6804 | Empty */
6805 static int
6806 intel_e04 (void)
6807 {
6808 int nregs = -1;
6809
6810 for (;;)
6811 {
6812 if (!intel_e05())
6813 return 0;
6814
6815 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6816 i.base_reg = i386_regtab + REGNAM_AL; /* al is invalid as base */
6817
6818 if (cur_token.code == '+')
6819 nregs = -1;
6820 else if (cur_token.code == '-')
6821 nregs = NUM_ADDRESS_REGS;
6822 else
6823 return 1;
6824
6825 strcat (intel_parser.disp, cur_token.str);
6826 intel_match_token (cur_token.code);
6827 }
6828 }
6829
6830 /* e05 e06 e05'
6831
6832 e05' binOp e06 e05'
6833 | Empty */
6834 static int
6835 intel_e05 (void)
6836 {
6837 int nregs = ~NUM_ADDRESS_REGS;
6838
6839 for (;;)
6840 {
6841 if (!intel_e06())
6842 return 0;
6843
6844 if (cur_token.code == '&'
6845 || cur_token.code == '|'
6846 || cur_token.code == '^')
6847 {
6848 char str[2];
6849
6850 str[0] = cur_token.code;
6851 str[1] = 0;
6852 strcat (intel_parser.disp, str);
6853 }
6854 else
6855 break;
6856
6857 intel_match_token (cur_token.code);
6858
6859 if (nregs < 0)
6860 nregs = ~nregs;
6861 }
6862 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6863 i.base_reg = i386_regtab + REGNAM_AL + 1; /* cl is invalid as base */
6864 return 1;
6865 }
6866
6867 /* e06 e09 e06'
6868
6869 e06' mulOp e09 e06'
6870 | Empty */
6871 static int
6872 intel_e06 (void)
6873 {
6874 int nregs = ~NUM_ADDRESS_REGS;
6875
6876 for (;;)
6877 {
6878 if (!intel_e09())
6879 return 0;
6880
6881 if (cur_token.code == '*'
6882 || cur_token.code == '/'
6883 || cur_token.code == '%')
6884 {
6885 char str[2];
6886
6887 str[0] = cur_token.code;
6888 str[1] = 0;
6889 strcat (intel_parser.disp, str);
6890 }
6891 else if (cur_token.code == T_SHL)
6892 strcat (intel_parser.disp, "<<");
6893 else if (cur_token.code == T_SHR)
6894 strcat (intel_parser.disp, ">>");
6895 else
6896 break;
6897
6898 intel_match_token (cur_token.code);
6899
6900 if (nregs < 0)
6901 nregs = ~nregs;
6902 }
6903 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6904 i.base_reg = i386_regtab + REGNAM_AL + 2; /* dl is invalid as base */
6905 return 1;
6906 }
6907
6908 /* e09 OFFSET e09
6909 | SHORT e09
6910 | + e09
6911 | - e09
6912 | ~ e09
6913 | NOT e09
6914 | e10 e09'
6915
6916 e09' PTR e10 e09'
6917 | : e10 e09'
6918 | Empty */
6919 static int
6920 intel_e09 (void)
6921 {
6922 int nregs = ~NUM_ADDRESS_REGS;
6923 int in_offset = 0;
6924
6925 for (;;)
6926 {
6927 /* Don't consume constants here. */
6928 if (cur_token.code == '+' || cur_token.code == '-')
6929 {
6930 /* Need to look one token ahead - if the next token
6931 is a constant, the current token is its sign. */
6932 int next_code;
6933
6934 intel_match_token (cur_token.code);
6935 next_code = cur_token.code;
6936 intel_putback_token ();
6937 if (next_code == T_CONST)
6938 break;
6939 }
6940
6941 /* e09 OFFSET e09 */
6942 if (cur_token.code == T_OFFSET)
6943 {
6944 if (!in_offset++)
6945 ++intel_parser.in_offset;
6946 }
6947
6948 /* e09 SHORT e09 */
6949 else if (cur_token.code == T_SHORT)
6950 intel_parser.op_modifier |= 1 << T_SHORT;
6951
6952 /* e09 + e09 */
6953 else if (cur_token.code == '+')
6954 strcat (intel_parser.disp, "+");
6955
6956 /* e09 - e09
6957 | ~ e09
6958 | NOT e09 */
6959 else if (cur_token.code == '-' || cur_token.code == '~')
6960 {
6961 char str[2];
6962
6963 if (nregs < 0)
6964 nregs = ~nregs;
6965 str[0] = cur_token.code;
6966 str[1] = 0;
6967 strcat (intel_parser.disp, str);
6968 }
6969
6970 /* e09 e10 e09' */
6971 else
6972 break;
6973
6974 intel_match_token (cur_token.code);
6975 }
6976
6977 for (;;)
6978 {
6979 if (!intel_e10 ())
6980 return 0;
6981
6982 /* e09' PTR e10 e09' */
6983 if (cur_token.code == T_PTR)
6984 {
6985 char suffix;
6986
6987 if (prev_token.code == T_BYTE)
6988 suffix = BYTE_MNEM_SUFFIX;
6989
6990 else if (prev_token.code == T_WORD)
6991 {
6992 if (current_templates->start->name[0] == 'l'
6993 && current_templates->start->name[2] == 's'
6994 && current_templates->start->name[3] == 0)
6995 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
6996 else if (intel_parser.got_a_float == 2) /* "fi..." */
6997 suffix = SHORT_MNEM_SUFFIX;
6998 else
6999 suffix = WORD_MNEM_SUFFIX;
7000 }
7001
7002 else if (prev_token.code == T_DWORD)
7003 {
7004 if (current_templates->start->name[0] == 'l'
7005 && current_templates->start->name[2] == 's'
7006 && current_templates->start->name[3] == 0)
7007 suffix = WORD_MNEM_SUFFIX;
7008 else if (flag_code == CODE_16BIT
7009 && (current_templates->start->opcode_modifier
7010 & (Jump | JumpDword)))
7011 suffix = LONG_DOUBLE_MNEM_SUFFIX;
7012 else if (intel_parser.got_a_float == 1) /* "f..." */
7013 suffix = SHORT_MNEM_SUFFIX;
7014 else
7015 suffix = LONG_MNEM_SUFFIX;
7016 }
7017
7018 else if (prev_token.code == T_FWORD)
7019 {
7020 if (current_templates->start->name[0] == 'l'
7021 && current_templates->start->name[2] == 's'
7022 && current_templates->start->name[3] == 0)
7023 suffix = LONG_MNEM_SUFFIX;
7024 else if (!intel_parser.got_a_float)
7025 {
7026 if (flag_code == CODE_16BIT)
7027 add_prefix (DATA_PREFIX_OPCODE);
7028 suffix = LONG_DOUBLE_MNEM_SUFFIX;
7029 }
7030 else
7031 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
7032 }
7033
7034 else if (prev_token.code == T_QWORD)
7035 {
7036 if (intel_parser.got_a_float == 1) /* "f..." */
7037 suffix = LONG_MNEM_SUFFIX;
7038 else
7039 suffix = QWORD_MNEM_SUFFIX;
7040 }
7041
7042 else if (prev_token.code == T_TBYTE)
7043 {
7044 if (intel_parser.got_a_float == 1)
7045 suffix = LONG_DOUBLE_MNEM_SUFFIX;
7046 else
7047 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
7048 }
7049
7050 else if (prev_token.code == T_XMMWORD)
7051 {
7052 /* XXX ignored for now, but accepted since gcc uses it */
7053 suffix = 0;
7054 }
7055
7056 else
7057 {
7058 as_bad (_("Unknown operand modifier `%s'"), prev_token.str);
7059 return 0;
7060 }
7061
7062 /* Operands for jump/call using 'ptr' notation denote absolute
7063 addresses. */
7064 if (current_templates->start->opcode_modifier & (Jump | JumpDword))
7065 i.types[this_operand] |= JumpAbsolute;
7066
7067 if (current_templates->start->base_opcode == 0x8d /* lea */)
7068 ;
7069 else if (!i.suffix)
7070 i.suffix = suffix;
7071 else if (i.suffix != suffix)
7072 {
7073 as_bad (_("Conflicting operand modifiers"));
7074 return 0;
7075 }
7076
7077 }
7078
7079 /* e09' : e10 e09' */
7080 else if (cur_token.code == ':')
7081 {
7082 if (prev_token.code != T_REG)
7083 {
7084 /* While {call,jmp} SSSS:OOOO is MASM syntax only when SSSS is a
7085 segment/group identifier (which we don't have), using comma
7086 as the operand separator there is even less consistent, since
7087 there all branches only have a single operand. */
7088 if (this_operand != 0
7089 || intel_parser.in_offset
7090 || intel_parser.in_bracket
7091 || (!(current_templates->start->opcode_modifier
7092 & (Jump|JumpDword|JumpInterSegment))
7093 && !(current_templates->start->operand_types[0]
7094 & JumpAbsolute)))
7095 return intel_match_token (T_NIL);
7096 /* Remember the start of the 2nd operand and terminate 1st
7097 operand here.
7098 XXX This isn't right, yet (when SSSS:OOOO is right operand of
7099 another expression), but it gets at least the simplest case
7100 (a plain number or symbol on the left side) right. */
7101 intel_parser.next_operand = intel_parser.op_string;
7102 *--intel_parser.op_string = '\0';
7103 return intel_match_token (':');
7104 }
7105 }
7106
7107 /* e09' Empty */
7108 else
7109 break;
7110
7111 intel_match_token (cur_token.code);
7112
7113 }
7114
7115 if (in_offset)
7116 {
7117 --intel_parser.in_offset;
7118 if (nregs < 0)
7119 nregs = ~nregs;
7120 if (NUM_ADDRESS_REGS > nregs)
7121 {
7122 as_bad (_("Invalid operand to `OFFSET'"));
7123 return 0;
7124 }
7125 intel_parser.op_modifier |= 1 << T_OFFSET;
7126 }
7127
7128 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
7129 i.base_reg = i386_regtab + REGNAM_AL + 3; /* bl is invalid as base */
7130 return 1;
7131 }
7132
7133 static int
7134 intel_bracket_expr (void)
7135 {
7136 int was_offset = intel_parser.op_modifier & (1 << T_OFFSET);
7137 const char *start = intel_parser.op_string;
7138 int len;
7139
7140 if (i.op[this_operand].regs)
7141 return intel_match_token (T_NIL);
7142
7143 intel_match_token ('[');
7144
7145 /* Mark as a memory operand only if it's not already known to be an
7146 offset expression. If it's an offset expression, we need to keep
7147 the brace in. */
7148 if (!intel_parser.in_offset)
7149 {
7150 ++intel_parser.in_bracket;
7151
7152 /* Operands for jump/call inside brackets denote absolute addresses. */
7153 if (current_templates->start->opcode_modifier & (Jump | JumpDword))
7154 i.types[this_operand] |= JumpAbsolute;
7155
7156 /* Unfortunately gas always diverged from MASM in a respect that can't
7157 be easily fixed without risking to break code sequences likely to be
7158 encountered (the testsuite even check for this): MASM doesn't consider
7159 an expression inside brackets unconditionally as a memory reference.
7160 When that is e.g. a constant, an offset expression, or the sum of the
7161 two, this is still taken as a constant load. gas, however, always
7162 treated these as memory references. As a compromise, we'll try to make
7163 offset expressions inside brackets work the MASM way (since that's
7164 less likely to be found in real world code), but make constants alone
7165 continue to work the traditional gas way. In either case, issue a
7166 warning. */
7167 intel_parser.op_modifier &= ~was_offset;
7168 }
7169 else
7170 strcat (intel_parser.disp, "[");
7171
7172 /* Add a '+' to the displacement string if necessary. */
7173 if (*intel_parser.disp != '\0'
7174 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
7175 strcat (intel_parser.disp, "+");
7176
7177 if (intel_expr ()
7178 && (len = intel_parser.op_string - start - 1,
7179 intel_match_token (']')))
7180 {
7181 /* Preserve brackets when the operand is an offset expression. */
7182 if (intel_parser.in_offset)
7183 strcat (intel_parser.disp, "]");
7184 else
7185 {
7186 --intel_parser.in_bracket;
7187 if (i.base_reg || i.index_reg)
7188 intel_parser.is_mem = 1;
7189 if (!intel_parser.is_mem)
7190 {
7191 if (!(intel_parser.op_modifier & (1 << T_OFFSET)))
7192 /* Defer the warning until all of the operand was parsed. */
7193 intel_parser.is_mem = -1;
7194 else if (!quiet_warnings)
7195 as_warn (_("`[%.*s]' taken to mean just `%.*s'"),
7196 len, start, len, start);
7197 }
7198 }
7199 intel_parser.op_modifier |= was_offset;
7200
7201 return 1;
7202 }
7203 return 0;
7204 }
7205
7206 /* e10 e11 e10'
7207
7208 e10' [ expr ] e10'
7209 | Empty */
7210 static int
7211 intel_e10 (void)
7212 {
7213 if (!intel_e11 ())
7214 return 0;
7215
7216 while (cur_token.code == '[')
7217 {
7218 if (!intel_bracket_expr ())
7219 return 0;
7220 }
7221
7222 return 1;
7223 }
7224
7225 /* e11 ( expr )
7226 | [ expr ]
7227 | BYTE
7228 | WORD
7229 | DWORD
7230 | FWORD
7231 | QWORD
7232 | TBYTE
7233 | OWORD
7234 | XMMWORD
7235 | $
7236 | .
7237 | register
7238 | id
7239 | constant */
7240 static int
7241 intel_e11 (void)
7242 {
7243 switch (cur_token.code)
7244 {
7245 /* e11 ( expr ) */
7246 case '(':
7247 intel_match_token ('(');
7248 strcat (intel_parser.disp, "(");
7249
7250 if (intel_expr () && intel_match_token (')'))
7251 {
7252 strcat (intel_parser.disp, ")");
7253 return 1;
7254 }
7255 return 0;
7256
7257 /* e11 [ expr ] */
7258 case '[':
7259 return intel_bracket_expr ();
7260
7261 /* e11 $
7262 | . */
7263 case '.':
7264 strcat (intel_parser.disp, cur_token.str);
7265 intel_match_token (cur_token.code);
7266
7267 /* Mark as a memory operand only if it's not already known to be an
7268 offset expression. */
7269 if (!intel_parser.in_offset)
7270 intel_parser.is_mem = 1;
7271
7272 return 1;
7273
7274 /* e11 register */
7275 case T_REG:
7276 {
7277 const reg_entry *reg = intel_parser.reg = cur_token.reg;
7278
7279 intel_match_token (T_REG);
7280
7281 /* Check for segment change. */
7282 if (cur_token.code == ':')
7283 {
7284 if (!(reg->reg_type & (SReg2 | SReg3)))
7285 {
7286 as_bad (_("`%s' is not a valid segment register"),
7287 reg->reg_name);
7288 return 0;
7289 }
7290 else if (i.seg[i.mem_operands])
7291 as_warn (_("Extra segment override ignored"));
7292 else
7293 {
7294 if (!intel_parser.in_offset)
7295 intel_parser.is_mem = 1;
7296 switch (reg->reg_num)
7297 {
7298 case 0:
7299 i.seg[i.mem_operands] = &es;
7300 break;
7301 case 1:
7302 i.seg[i.mem_operands] = &cs;
7303 break;
7304 case 2:
7305 i.seg[i.mem_operands] = &ss;
7306 break;
7307 case 3:
7308 i.seg[i.mem_operands] = &ds;
7309 break;
7310 case 4:
7311 i.seg[i.mem_operands] = &fs;
7312 break;
7313 case 5:
7314 i.seg[i.mem_operands] = &gs;
7315 break;
7316 }
7317 }
7318 }
7319
7320 /* Not a segment register. Check for register scaling. */
7321 else if (cur_token.code == '*')
7322 {
7323 if (!intel_parser.in_bracket)
7324 {
7325 as_bad (_("Register scaling only allowed in memory operands"));
7326 return 0;
7327 }
7328
7329 if (reg->reg_type & Reg16) /* Disallow things like [si*1]. */
7330 reg = i386_regtab + REGNAM_AX + 4; /* sp is invalid as index */
7331 else if (i.index_reg)
7332 reg = i386_regtab + REGNAM_EAX + 4; /* esp is invalid as index */
7333
7334 /* What follows must be a valid scale. */
7335 intel_match_token ('*');
7336 i.index_reg = reg;
7337 i.types[this_operand] |= BaseIndex;
7338
7339 /* Set the scale after setting the register (otherwise,
7340 i386_scale will complain) */
7341 if (cur_token.code == '+' || cur_token.code == '-')
7342 {
7343 char *str, sign = cur_token.code;
7344 intel_match_token (cur_token.code);
7345 if (cur_token.code != T_CONST)
7346 {
7347 as_bad (_("Syntax error: Expecting a constant, got `%s'"),
7348 cur_token.str);
7349 return 0;
7350 }
7351 str = (char *) xmalloc (strlen (cur_token.str) + 2);
7352 strcpy (str + 1, cur_token.str);
7353 *str = sign;
7354 if (!i386_scale (str))
7355 return 0;
7356 free (str);
7357 }
7358 else if (!i386_scale (cur_token.str))
7359 return 0;
7360 intel_match_token (cur_token.code);
7361 }
7362
7363 /* No scaling. If this is a memory operand, the register is either a
7364 base register (first occurrence) or an index register (second
7365 occurrence). */
7366 else if (intel_parser.in_bracket)
7367 {
7368
7369 if (!i.base_reg)
7370 i.base_reg = reg;
7371 else if (!i.index_reg)
7372 i.index_reg = reg;
7373 else
7374 {
7375 as_bad (_("Too many register references in memory operand"));
7376 return 0;
7377 }
7378
7379 i.types[this_operand] |= BaseIndex;
7380 }
7381
7382 /* It's neither base nor index. */
7383 else if (!intel_parser.in_offset && !intel_parser.is_mem)
7384 {
7385 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
7386 i.op[this_operand].regs = reg;
7387 i.reg_operands++;
7388 }
7389 else
7390 {
7391 as_bad (_("Invalid use of register"));
7392 return 0;
7393 }
7394
7395 /* Since registers are not part of the displacement string (except
7396 when we're parsing offset operands), we may need to remove any
7397 preceding '+' from the displacement string. */
7398 if (*intel_parser.disp != '\0'
7399 && !intel_parser.in_offset)
7400 {
7401 char *s = intel_parser.disp;
7402 s += strlen (s) - 1;
7403 if (*s == '+')
7404 *s = '\0';
7405 }
7406
7407 return 1;
7408 }
7409
7410 /* e11 BYTE
7411 | WORD
7412 | DWORD
7413 | FWORD
7414 | QWORD
7415 | TBYTE
7416 | OWORD
7417 | XMMWORD */
7418 case T_BYTE:
7419 case T_WORD:
7420 case T_DWORD:
7421 case T_FWORD:
7422 case T_QWORD:
7423 case T_TBYTE:
7424 case T_XMMWORD:
7425 intel_match_token (cur_token.code);
7426
7427 if (cur_token.code == T_PTR)
7428 return 1;
7429
7430 /* It must have been an identifier. */
7431 intel_putback_token ();
7432 cur_token.code = T_ID;
7433 /* FALLTHRU */
7434
7435 /* e11 id
7436 | constant */
7437 case T_ID:
7438 if (!intel_parser.in_offset && intel_parser.is_mem <= 0)
7439 {
7440 symbolS *symbolP;
7441
7442 /* The identifier represents a memory reference only if it's not
7443 preceded by an offset modifier and if it's not an equate. */
7444 symbolP = symbol_find(cur_token.str);
7445 if (!symbolP || S_GET_SEGMENT(symbolP) != absolute_section)
7446 intel_parser.is_mem = 1;
7447 }
7448 /* FALLTHRU */
7449
7450 case T_CONST:
7451 case '-':
7452 case '+':
7453 {
7454 char *save_str, sign = 0;
7455
7456 /* Allow constants that start with `+' or `-'. */
7457 if (cur_token.code == '-' || cur_token.code == '+')
7458 {
7459 sign = cur_token.code;
7460 intel_match_token (cur_token.code);
7461 if (cur_token.code != T_CONST)
7462 {
7463 as_bad (_("Syntax error: Expecting a constant, got `%s'"),
7464 cur_token.str);
7465 return 0;
7466 }
7467 }
7468
7469 save_str = (char *) xmalloc (strlen (cur_token.str) + 2);
7470 strcpy (save_str + !!sign, cur_token.str);
7471 if (sign)
7472 *save_str = sign;
7473
7474 /* Get the next token to check for register scaling. */
7475 intel_match_token (cur_token.code);
7476
7477 /* Check if this constant is a scaling factor for an
7478 index register. */
7479 if (cur_token.code == '*')
7480 {
7481 if (intel_match_token ('*') && cur_token.code == T_REG)
7482 {
7483 const reg_entry *reg = cur_token.reg;
7484
7485 if (!intel_parser.in_bracket)
7486 {
7487 as_bad (_("Register scaling only allowed "
7488 "in memory operands"));
7489 return 0;
7490 }
7491
7492 /* Disallow things like [1*si].
7493 sp and esp are invalid as index. */
7494 if (reg->reg_type & Reg16)
7495 reg = i386_regtab + REGNAM_AX + 4;
7496 else if (i.index_reg)
7497 reg = i386_regtab + REGNAM_EAX + 4;
7498
7499 /* The constant is followed by `* reg', so it must be
7500 a valid scale. */
7501 i.index_reg = reg;
7502 i.types[this_operand] |= BaseIndex;
7503
7504 /* Set the scale after setting the register (otherwise,
7505 i386_scale will complain) */
7506 if (!i386_scale (save_str))
7507 return 0;
7508 intel_match_token (T_REG);
7509
7510 /* Since registers are not part of the displacement
7511 string, we may need to remove any preceding '+' from
7512 the displacement string. */
7513 if (*intel_parser.disp != '\0')
7514 {
7515 char *s = intel_parser.disp;
7516 s += strlen (s) - 1;
7517 if (*s == '+')
7518 *s = '\0';
7519 }
7520
7521 free (save_str);
7522
7523 return 1;
7524 }
7525
7526 /* The constant was not used for register scaling. Since we have
7527 already consumed the token following `*' we now need to put it
7528 back in the stream. */
7529 intel_putback_token ();
7530 }
7531
7532 /* Add the constant to the displacement string. */
7533 strcat (intel_parser.disp, save_str);
7534 free (save_str);
7535
7536 return 1;
7537 }
7538 }
7539
7540 as_bad (_("Unrecognized token '%s'"), cur_token.str);
7541 return 0;
7542 }
7543
7544 /* Match the given token against cur_token. If they match, read the next
7545 token from the operand string. */
7546 static int
7547 intel_match_token (int code)
7548 {
7549 if (cur_token.code == code)
7550 {
7551 intel_get_token ();
7552 return 1;
7553 }
7554 else
7555 {
7556 as_bad (_("Unexpected token `%s'"), cur_token.str);
7557 return 0;
7558 }
7559 }
7560
7561 /* Read a new token from intel_parser.op_string and store it in cur_token. */
7562 static void
7563 intel_get_token (void)
7564 {
7565 char *end_op;
7566 const reg_entry *reg;
7567 struct intel_token new_token;
7568
7569 new_token.code = T_NIL;
7570 new_token.reg = NULL;
7571 new_token.str = NULL;
7572
7573 /* Free the memory allocated to the previous token and move
7574 cur_token to prev_token. */
7575 if (prev_token.str)
7576 free (prev_token.str);
7577
7578 prev_token = cur_token;
7579
7580 /* Skip whitespace. */
7581 while (is_space_char (*intel_parser.op_string))
7582 intel_parser.op_string++;
7583
7584 /* Return an empty token if we find nothing else on the line. */
7585 if (*intel_parser.op_string == '\0')
7586 {
7587 cur_token = new_token;
7588 return;
7589 }
7590
7591 /* The new token cannot be larger than the remainder of the operand
7592 string. */
7593 new_token.str = (char *) xmalloc (strlen (intel_parser.op_string) + 1);
7594 new_token.str[0] = '\0';
7595
7596 if (strchr ("0123456789", *intel_parser.op_string))
7597 {
7598 char *p = new_token.str;
7599 char *q = intel_parser.op_string;
7600 new_token.code = T_CONST;
7601
7602 /* Allow any kind of identifier char to encompass floating point and
7603 hexadecimal numbers. */
7604 while (is_identifier_char (*q))
7605 *p++ = *q++;
7606 *p = '\0';
7607
7608 /* Recognize special symbol names [0-9][bf]. */
7609 if (strlen (intel_parser.op_string) == 2
7610 && (intel_parser.op_string[1] == 'b'
7611 || intel_parser.op_string[1] == 'f'))
7612 new_token.code = T_ID;
7613 }
7614
7615 else if ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL)
7616 {
7617 size_t len = end_op - intel_parser.op_string;
7618
7619 new_token.code = T_REG;
7620 new_token.reg = reg;
7621
7622 memcpy (new_token.str, intel_parser.op_string, len);
7623 new_token.str[len] = '\0';
7624 }
7625
7626 else if (is_identifier_char (*intel_parser.op_string))
7627 {
7628 char *p = new_token.str;
7629 char *q = intel_parser.op_string;
7630
7631 /* A '.' or '$' followed by an identifier char is an identifier.
7632 Otherwise, it's operator '.' followed by an expression. */
7633 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
7634 {
7635 new_token.code = '.';
7636 new_token.str[0] = '.';
7637 new_token.str[1] = '\0';
7638 }
7639 else
7640 {
7641 while (is_identifier_char (*q) || *q == '@')
7642 *p++ = *q++;
7643 *p = '\0';
7644
7645 if (strcasecmp (new_token.str, "NOT") == 0)
7646 new_token.code = '~';
7647
7648 else if (strcasecmp (new_token.str, "MOD") == 0)
7649 new_token.code = '%';
7650
7651 else if (strcasecmp (new_token.str, "AND") == 0)
7652 new_token.code = '&';
7653
7654 else if (strcasecmp (new_token.str, "OR") == 0)
7655 new_token.code = '|';
7656
7657 else if (strcasecmp (new_token.str, "XOR") == 0)
7658 new_token.code = '^';
7659
7660 else if (strcasecmp (new_token.str, "SHL") == 0)
7661 new_token.code = T_SHL;
7662
7663 else if (strcasecmp (new_token.str, "SHR") == 0)
7664 new_token.code = T_SHR;
7665
7666 else if (strcasecmp (new_token.str, "BYTE") == 0)
7667 new_token.code = T_BYTE;
7668
7669 else if (strcasecmp (new_token.str, "WORD") == 0)
7670 new_token.code = T_WORD;
7671
7672 else if (strcasecmp (new_token.str, "DWORD") == 0)
7673 new_token.code = T_DWORD;
7674
7675 else if (strcasecmp (new_token.str, "FWORD") == 0)
7676 new_token.code = T_FWORD;
7677
7678 else if (strcasecmp (new_token.str, "QWORD") == 0)
7679 new_token.code = T_QWORD;
7680
7681 else if (strcasecmp (new_token.str, "TBYTE") == 0
7682 /* XXX remove (gcc still uses it) */
7683 || strcasecmp (new_token.str, "XWORD") == 0)
7684 new_token.code = T_TBYTE;
7685
7686 else if (strcasecmp (new_token.str, "XMMWORD") == 0
7687 || strcasecmp (new_token.str, "OWORD") == 0)
7688 new_token.code = T_XMMWORD;
7689
7690 else if (strcasecmp (new_token.str, "PTR") == 0)
7691 new_token.code = T_PTR;
7692
7693 else if (strcasecmp (new_token.str, "SHORT") == 0)
7694 new_token.code = T_SHORT;
7695
7696 else if (strcasecmp (new_token.str, "OFFSET") == 0)
7697 {
7698 new_token.code = T_OFFSET;
7699
7700 /* ??? This is not mentioned in the MASM grammar but gcc
7701 makes use of it with -mintel-syntax. OFFSET may be
7702 followed by FLAT: */
7703 if (strncasecmp (q, " FLAT:", 6) == 0)
7704 strcat (new_token.str, " FLAT:");
7705 }
7706
7707 /* ??? This is not mentioned in the MASM grammar. */
7708 else if (strcasecmp (new_token.str, "FLAT") == 0)
7709 {
7710 new_token.code = T_OFFSET;
7711 if (*q == ':')
7712 strcat (new_token.str, ":");
7713 else
7714 as_bad (_("`:' expected"));
7715 }
7716
7717 else
7718 new_token.code = T_ID;
7719 }
7720 }
7721
7722 else if (strchr ("+-/*%|&^:[]()~", *intel_parser.op_string))
7723 {
7724 new_token.code = *intel_parser.op_string;
7725 new_token.str[0] = *intel_parser.op_string;
7726 new_token.str[1] = '\0';
7727 }
7728
7729 else if (strchr ("<>", *intel_parser.op_string)
7730 && *intel_parser.op_string == *(intel_parser.op_string + 1))
7731 {
7732 new_token.code = *intel_parser.op_string == '<' ? T_SHL : T_SHR;
7733 new_token.str[0] = *intel_parser.op_string;
7734 new_token.str[1] = *intel_parser.op_string;
7735 new_token.str[2] = '\0';
7736 }
7737
7738 else
7739 as_bad (_("Unrecognized token `%s'"), intel_parser.op_string);
7740
7741 intel_parser.op_string += strlen (new_token.str);
7742 cur_token = new_token;
7743 }
7744
7745 /* Put cur_token back into the token stream and make cur_token point to
7746 prev_token. */
7747 static void
7748 intel_putback_token (void)
7749 {
7750 if (cur_token.code != T_NIL)
7751 {
7752 intel_parser.op_string -= strlen (cur_token.str);
7753 free (cur_token.str);
7754 }
7755 cur_token = prev_token;
7756
7757 /* Forget prev_token. */
7758 prev_token.code = T_NIL;
7759 prev_token.reg = NULL;
7760 prev_token.str = NULL;
7761 }
7762
7763 int
7764 tc_x86_regname_to_dw2regnum (char *regname)
7765 {
7766 unsigned int regnum;
7767 unsigned int regnames_count;
7768 static const char *const regnames_32[] =
7769 {
7770 "eax", "ecx", "edx", "ebx",
7771 "esp", "ebp", "esi", "edi",
7772 "eip", "eflags", NULL,
7773 "st0", "st1", "st2", "st3",
7774 "st4", "st5", "st6", "st7",
7775 NULL, NULL,
7776 "xmm0", "xmm1", "xmm2", "xmm3",
7777 "xmm4", "xmm5", "xmm6", "xmm7",
7778 "mm0", "mm1", "mm2", "mm3",
7779 "mm4", "mm5", "mm6", "mm7",
7780 "fcw", "fsw", "mxcsr",
7781 "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL,
7782 "tr", "ldtr"
7783 };
7784 static const char *const regnames_64[] =
7785 {
7786 "rax", "rdx", "rcx", "rbx",
7787 "rsi", "rdi", "rbp", "rsp",
7788 "r8", "r9", "r10", "r11",
7789 "r12", "r13", "r14", "r15",
7790 "rip",
7791 "xmm0", "xmm1", "xmm2", "xmm3",
7792 "xmm4", "xmm5", "xmm6", "xmm7",
7793 "xmm8", "xmm9", "xmm10", "xmm11",
7794 "xmm12", "xmm13", "xmm14", "xmm15",
7795 "st0", "st1", "st2", "st3",
7796 "st4", "st5", "st6", "st7",
7797 "mm0", "mm1", "mm2", "mm3",
7798 "mm4", "mm5", "mm6", "mm7",
7799 "rflags",
7800 "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL,
7801 "fs.base", "gs.base", NULL, NULL,
7802 "tr", "ldtr",
7803 "mxcsr", "fcw", "fsw"
7804 };
7805 const char *const *regnames;
7806
7807 if (flag_code == CODE_64BIT)
7808 {
7809 regnames = regnames_64;
7810 regnames_count = ARRAY_SIZE (regnames_64);
7811 }
7812 else
7813 {
7814 regnames = regnames_32;
7815 regnames_count = ARRAY_SIZE (regnames_32);
7816 }
7817
7818 for (regnum = 0; regnum < regnames_count; regnum++)
7819 if (regnames[regnum] != NULL
7820 && strcmp (regname, regnames[regnum]) == 0)
7821 return regnum;
7822
7823 return -1;
7824 }
7825
7826 void
7827 tc_x86_frame_initial_instructions (void)
7828 {
7829 static unsigned int sp_regno;
7830
7831 if (!sp_regno)
7832 sp_regno = tc_x86_regname_to_dw2regnum (flag_code == CODE_64BIT
7833 ? "rsp" : "esp");
7834
7835 cfi_add_CFA_def_cfa (sp_regno, -x86_cie_data_alignment);
7836 cfi_add_CFA_offset (x86_dwarf2_return_column, x86_cie_data_alignment);
7837 }
7838
7839 int
7840 i386_elf_section_type (const char *str, size_t len)
7841 {
7842 if (flag_code == CODE_64BIT
7843 && len == sizeof ("unwind") - 1
7844 && strncmp (str, "unwind", 6) == 0)
7845 return SHT_X86_64_UNWIND;
7846
7847 return -1;
7848 }
7849
7850 #ifdef TE_PE
7851 void
7852 tc_pe_dwarf2_emit_offset (symbolS *symbol, unsigned int size)
7853 {
7854 expressionS expr;
7855
7856 expr.X_op = O_secrel;
7857 expr.X_add_symbol = symbol;
7858 expr.X_add_number = 0;
7859 emit_expr (&expr, size);
7860 }
7861 #endif
7862
7863 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
7864 /* For ELF on x86-64, add support for SHF_X86_64_LARGE. */
7865
7866 int
7867 x86_64_section_letter (int letter, char **ptr_msg)
7868 {
7869 if (flag_code == CODE_64BIT)
7870 {
7871 if (letter == 'l')
7872 return SHF_X86_64_LARGE;
7873
7874 *ptr_msg = _("Bad .section directive: want a,l,w,x,M,S,G,T in string");
7875 }
7876 else
7877 *ptr_msg = _("Bad .section directive: want a,w,x,M,S,G,T in string");
7878 return -1;
7879 }
7880
7881 int
7882 x86_64_section_word (char *str, size_t len)
7883 {
7884 if (len == 5 && flag_code == CODE_64BIT && CONST_STRNEQ (str, "large"))
7885 return SHF_X86_64_LARGE;
7886
7887 return -1;
7888 }
7889
7890 static void
7891 handle_large_common (int small ATTRIBUTE_UNUSED)
7892 {
7893 if (flag_code != CODE_64BIT)
7894 {
7895 s_comm_internal (0, elf_common_parse);
7896 as_warn (_(".largecomm supported only in 64bit mode, producing .comm"));
7897 }
7898 else
7899 {
7900 static segT lbss_section;
7901 asection *saved_com_section_ptr = elf_com_section_ptr;
7902 asection *saved_bss_section = bss_section;
7903
7904 if (lbss_section == NULL)
7905 {
7906 flagword applicable;
7907 segT seg = now_seg;
7908 subsegT subseg = now_subseg;
7909
7910 /* The .lbss section is for local .largecomm symbols. */
7911 lbss_section = subseg_new (".lbss", 0);
7912 applicable = bfd_applicable_section_flags (stdoutput);
7913 bfd_set_section_flags (stdoutput, lbss_section,
7914 applicable & SEC_ALLOC);
7915 seg_info (lbss_section)->bss = 1;
7916
7917 subseg_set (seg, subseg);
7918 }
7919
7920 elf_com_section_ptr = &_bfd_elf_large_com_section;
7921 bss_section = lbss_section;
7922
7923 s_comm_internal (0, elf_common_parse);
7924
7925 elf_com_section_ptr = saved_com_section_ptr;
7926 bss_section = saved_bss_section;
7927 }
7928 }
7929 #endif /* OBJ_ELF || OBJ_MAYBE_ELF */
This page took 0.196801 seconds and 4 git commands to generate.