2007-03-12 H.J. Lu <hongjiu.lu@intel.com>
[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 "opcode/i386.h"
36 #include "elf/x86-64.h"
37
38 #ifndef REGISTER_WARNINGS
39 #define REGISTER_WARNINGS 1
40 #endif
41
42 #ifndef INFER_ADDR_PREFIX
43 #define INFER_ADDR_PREFIX 1
44 #endif
45
46 #ifndef SCALE1_WHEN_NO_INDEX
47 /* Specifying a scale factor besides 1 when there is no index is
48 futile. eg. `mov (%ebx,2),%al' does exactly the same as
49 `mov (%ebx),%al'. To slavishly follow what the programmer
50 specified, set SCALE1_WHEN_NO_INDEX to 0. */
51 #define SCALE1_WHEN_NO_INDEX 1
52 #endif
53
54 #ifndef DEFAULT_ARCH
55 #define DEFAULT_ARCH "i386"
56 #endif
57
58 #ifndef INLINE
59 #if __GNUC__ >= 2
60 #define INLINE __inline__
61 #else
62 #define INLINE
63 #endif
64 #endif
65
66 static void set_code_flag (int);
67 static void set_16bit_gcc_code_flag (int);
68 static void set_intel_syntax (int);
69 static void set_cpu_arch (int);
70 #ifdef TE_PE
71 static void pe_directive_secrel (int);
72 #endif
73 static void signed_cons (int);
74 static char *output_invalid (int c);
75 static int i386_operand (char *);
76 static int i386_intel_operand (char *, int);
77 static const reg_entry *parse_register (char *, char **);
78 static char *parse_insn (char *, char *);
79 static char *parse_operands (char *, const char *);
80 static void swap_operands (void);
81 static void swap_2_operands (int, int);
82 static void optimize_imm (void);
83 static void optimize_disp (void);
84 static int match_template (void);
85 static int check_string (void);
86 static int process_suffix (void);
87 static int check_byte_reg (void);
88 static int check_long_reg (void);
89 static int check_qword_reg (void);
90 static int check_word_reg (void);
91 static int finalize_imm (void);
92 static int process_operands (void);
93 static const seg_entry *build_modrm_byte (void);
94 static void output_insn (void);
95 static void output_imm (fragS *, offsetT);
96 static void output_disp (fragS *, offsetT);
97 #ifndef I386COFF
98 static void s_bss (int);
99 #endif
100 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
101 static void handle_large_common (int small ATTRIBUTE_UNUSED);
102 #endif
103
104 static const char *default_arch = DEFAULT_ARCH;
105
106 /* 'md_assemble ()' gathers together information and puts it into a
107 i386_insn. */
108
109 union i386_op
110 {
111 expressionS *disps;
112 expressionS *imms;
113 const reg_entry *regs;
114 };
115
116 struct _i386_insn
117 {
118 /* TM holds the template for the insn were currently assembling. */
119 template tm;
120
121 /* SUFFIX holds the instruction mnemonic suffix if given.
122 (e.g. 'l' for 'movl') */
123 char suffix;
124
125 /* OPERANDS gives the number of given operands. */
126 unsigned int operands;
127
128 /* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number
129 of given register, displacement, memory operands and immediate
130 operands. */
131 unsigned int reg_operands, disp_operands, mem_operands, imm_operands;
132
133 /* TYPES [i] is the type (see above #defines) which tells us how to
134 use OP[i] for the corresponding operand. */
135 unsigned int types[MAX_OPERANDS];
136
137 /* Displacement expression, immediate expression, or register for each
138 operand. */
139 union i386_op op[MAX_OPERANDS];
140
141 /* Flags for operands. */
142 unsigned int flags[MAX_OPERANDS];
143 #define Operand_PCrel 1
144
145 /* Relocation type for operand */
146 enum bfd_reloc_code_real reloc[MAX_OPERANDS];
147
148 /* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode
149 the base index byte below. */
150 const reg_entry *base_reg;
151 const reg_entry *index_reg;
152 unsigned int log2_scale_factor;
153
154 /* SEG gives the seg_entries of this insn. They are zero unless
155 explicit segment overrides are given. */
156 const seg_entry *seg[2];
157
158 /* PREFIX holds all the given prefix opcodes (usually null).
159 PREFIXES is the number of prefix opcodes. */
160 unsigned int prefixes;
161 unsigned char prefix[MAX_PREFIXES];
162
163 /* RM and SIB are the modrm byte and the sib byte where the
164 addressing modes of this insn are encoded. */
165
166 modrm_byte rm;
167 rex_byte rex;
168 sib_byte sib;
169 };
170
171 typedef struct _i386_insn i386_insn;
172
173 /* List of chars besides those in app.c:symbol_chars that can start an
174 operand. Used to prevent the scrubber eating vital white-space. */
175 const char extra_symbol_chars[] = "*%-(["
176 #ifdef LEX_AT
177 "@"
178 #endif
179 #ifdef LEX_QM
180 "?"
181 #endif
182 ;
183
184 #if (defined (TE_I386AIX) \
185 || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) \
186 && !defined (TE_GNU) \
187 && !defined (TE_LINUX) \
188 && !defined (TE_NETWARE) \
189 && !defined (TE_FreeBSD) \
190 && !defined (TE_NetBSD)))
191 /* This array holds the chars that always start a comment. If the
192 pre-processor is disabled, these aren't very useful. The option
193 --divide will remove '/' from this list. */
194 const char *i386_comment_chars = "#/";
195 #define SVR4_COMMENT_CHARS 1
196 #define PREFIX_SEPARATOR '\\'
197
198 #else
199 const char *i386_comment_chars = "#";
200 #define PREFIX_SEPARATOR '/'
201 #endif
202
203 /* This array holds the chars that only start a comment at the beginning of
204 a line. If the line seems to have the form '# 123 filename'
205 .line and .file directives will appear in the pre-processed output.
206 Note that input_file.c hand checks for '#' at the beginning of the
207 first line of the input file. This is because the compiler outputs
208 #NO_APP at the beginning of its output.
209 Also note that comments started like this one will always work if
210 '/' isn't otherwise defined. */
211 const char line_comment_chars[] = "#/";
212
213 const char line_separator_chars[] = ";";
214
215 /* Chars that can be used to separate mant from exp in floating point
216 nums. */
217 const char EXP_CHARS[] = "eE";
218
219 /* Chars that mean this number is a floating point constant
220 As in 0f12.456
221 or 0d1.2345e12. */
222 const char FLT_CHARS[] = "fFdDxX";
223
224 /* Tables for lexical analysis. */
225 static char mnemonic_chars[256];
226 static char register_chars[256];
227 static char operand_chars[256];
228 static char identifier_chars[256];
229 static char digit_chars[256];
230
231 /* Lexical macros. */
232 #define is_mnemonic_char(x) (mnemonic_chars[(unsigned char) x])
233 #define is_operand_char(x) (operand_chars[(unsigned char) x])
234 #define is_register_char(x) (register_chars[(unsigned char) x])
235 #define is_space_char(x) ((x) == ' ')
236 #define is_identifier_char(x) (identifier_chars[(unsigned char) x])
237 #define is_digit_char(x) (digit_chars[(unsigned char) x])
238
239 /* All non-digit non-letter characters that may occur in an operand. */
240 static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
241
242 /* md_assemble() always leaves the strings it's passed unaltered. To
243 effect this we maintain a stack of saved characters that we've smashed
244 with '\0's (indicating end of strings for various sub-fields of the
245 assembler instruction). */
246 static char save_stack[32];
247 static char *save_stack_p;
248 #define END_STRING_AND_SAVE(s) \
249 do { *save_stack_p++ = *(s); *(s) = '\0'; } while (0)
250 #define RESTORE_END_STRING(s) \
251 do { *(s) = *--save_stack_p; } while (0)
252
253 /* The instruction we're assembling. */
254 static i386_insn i;
255
256 /* Possible templates for current insn. */
257 static const templates *current_templates;
258
259 /* Per instruction expressionS buffers: max displacements & immediates. */
260 static expressionS disp_expressions[MAX_MEMORY_OPERANDS];
261 static expressionS im_expressions[MAX_IMMEDIATE_OPERANDS];
262
263 /* Current operand we are working on. */
264 static int this_operand;
265
266 /* We support four different modes. FLAG_CODE variable is used to distinguish
267 these. */
268
269 enum flag_code {
270 CODE_32BIT,
271 CODE_16BIT,
272 CODE_64BIT };
273 #define NUM_FLAG_CODE ((int) CODE_64BIT + 1)
274
275 static enum flag_code flag_code;
276 static unsigned int object_64bit;
277 static int use_rela_relocations = 0;
278
279 /* The names used to print error messages. */
280 static const char *flag_code_names[] =
281 {
282 "32",
283 "16",
284 "64"
285 };
286
287 /* 1 for intel syntax,
288 0 if att syntax. */
289 static int intel_syntax = 0;
290
291 /* 1 if register prefix % not required. */
292 static int allow_naked_reg = 0;
293
294 /* Register prefix used for error message. */
295 static const char *register_prefix = "%";
296
297 /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
298 leave, push, and pop instructions so that gcc has the same stack
299 frame as in 32 bit mode. */
300 static char stackop_size = '\0';
301
302 /* Non-zero to optimize code alignment. */
303 int optimize_align_code = 1;
304
305 /* Non-zero to quieten some warnings. */
306 static int quiet_warnings = 0;
307
308 /* CPU name. */
309 static const char *cpu_arch_name = NULL;
310 static const char *cpu_sub_arch_name = NULL;
311
312 /* CPU feature flags. */
313 static unsigned int cpu_arch_flags = CpuUnknownFlags | CpuNo64;
314
315 /* If we have selected a cpu we are generating instructions for. */
316 static int cpu_arch_tune_set = 0;
317
318 /* Cpu we are generating instructions for. */
319 static enum processor_type cpu_arch_tune = PROCESSOR_UNKNOWN;
320
321 /* CPU feature flags of cpu we are generating instructions for. */
322 static unsigned int cpu_arch_tune_flags = 0;
323
324 /* CPU instruction set architecture used. */
325 static enum processor_type cpu_arch_isa = PROCESSOR_UNKNOWN;
326
327 /* CPU feature flags of instruction set architecture used. */
328 static unsigned int cpu_arch_isa_flags = 0;
329
330 /* If set, conditional jumps are not automatically promoted to handle
331 larger than a byte offset. */
332 static unsigned int no_cond_jump_promotion = 0;
333
334 /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
335 static symbolS *GOT_symbol;
336
337 /* The dwarf2 return column, adjusted for 32 or 64 bit. */
338 unsigned int x86_dwarf2_return_column;
339
340 /* The dwarf2 data alignment, adjusted for 32 or 64 bit. */
341 int x86_cie_data_alignment;
342
343 /* Interface to relax_segment.
344 There are 3 major relax states for 386 jump insns because the
345 different types of jumps add different sizes to frags when we're
346 figuring out what sort of jump to choose to reach a given label. */
347
348 /* Types. */
349 #define UNCOND_JUMP 0
350 #define COND_JUMP 1
351 #define COND_JUMP86 2
352
353 /* Sizes. */
354 #define CODE16 1
355 #define SMALL 0
356 #define SMALL16 (SMALL | CODE16)
357 #define BIG 2
358 #define BIG16 (BIG | CODE16)
359
360 #ifndef INLINE
361 #ifdef __GNUC__
362 #define INLINE __inline__
363 #else
364 #define INLINE
365 #endif
366 #endif
367
368 #define ENCODE_RELAX_STATE(type, size) \
369 ((relax_substateT) (((type) << 2) | (size)))
370 #define TYPE_FROM_RELAX_STATE(s) \
371 ((s) >> 2)
372 #define DISP_SIZE_FROM_RELAX_STATE(s) \
373 ((((s) & 3) == BIG ? 4 : (((s) & 3) == BIG16 ? 2 : 1)))
374
375 /* This table is used by relax_frag to promote short jumps to long
376 ones where necessary. SMALL (short) jumps may be promoted to BIG
377 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
378 don't allow a short jump in a 32 bit code segment to be promoted to
379 a 16 bit offset jump because it's slower (requires data size
380 prefix), and doesn't work, unless the destination is in the bottom
381 64k of the code segment (The top 16 bits of eip are zeroed). */
382
383 const relax_typeS md_relax_table[] =
384 {
385 /* The fields are:
386 1) most positive reach of this state,
387 2) most negative reach of this state,
388 3) how many bytes this mode will have in the variable part of the frag
389 4) which index into the table to try if we can't fit into this one. */
390
391 /* UNCOND_JUMP states. */
392 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
393 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
394 /* dword jmp adds 4 bytes to frag:
395 0 extra opcode bytes, 4 displacement bytes. */
396 {0, 0, 4, 0},
397 /* word jmp adds 2 byte2 to frag:
398 0 extra opcode bytes, 2 displacement bytes. */
399 {0, 0, 2, 0},
400
401 /* COND_JUMP states. */
402 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
403 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
404 /* dword conditionals adds 5 bytes to frag:
405 1 extra opcode byte, 4 displacement bytes. */
406 {0, 0, 5, 0},
407 /* word conditionals add 3 bytes to frag:
408 1 extra opcode byte, 2 displacement bytes. */
409 {0, 0, 3, 0},
410
411 /* COND_JUMP86 states. */
412 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG)},
413 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG16)},
414 /* dword conditionals adds 5 bytes to frag:
415 1 extra opcode byte, 4 displacement bytes. */
416 {0, 0, 5, 0},
417 /* word conditionals add 4 bytes to frag:
418 1 displacement byte and a 3 byte long branch insn. */
419 {0, 0, 4, 0}
420 };
421
422 static const arch_entry cpu_arch[] =
423 {
424 {"generic32", PROCESSOR_GENERIC32,
425 Cpu186|Cpu286|Cpu386},
426 {"generic64", PROCESSOR_GENERIC64,
427 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
428 |CpuMMX2|CpuSSE|CpuSSE2},
429 {"i8086", PROCESSOR_UNKNOWN,
430 0},
431 {"i186", PROCESSOR_UNKNOWN,
432 Cpu186},
433 {"i286", PROCESSOR_UNKNOWN,
434 Cpu186|Cpu286},
435 {"i386", PROCESSOR_GENERIC32,
436 Cpu186|Cpu286|Cpu386},
437 {"i486", PROCESSOR_I486,
438 Cpu186|Cpu286|Cpu386|Cpu486},
439 {"i586", PROCESSOR_PENTIUM,
440 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586},
441 {"i686", PROCESSOR_PENTIUMPRO,
442 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686},
443 {"pentium", PROCESSOR_PENTIUM,
444 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586},
445 {"pentiumpro",PROCESSOR_PENTIUMPRO,
446 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686},
447 {"pentiumii", PROCESSOR_PENTIUMPRO,
448 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX},
449 {"pentiumiii",PROCESSOR_PENTIUMPRO,
450 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuMMX2|CpuSSE},
451 {"pentium4", PROCESSOR_PENTIUM4,
452 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
453 |CpuMMX2|CpuSSE|CpuSSE2},
454 {"prescott", PROCESSOR_NOCONA,
455 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
456 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
457 {"nocona", PROCESSOR_NOCONA,
458 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
459 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
460 {"yonah", PROCESSOR_CORE,
461 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
462 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
463 {"core", PROCESSOR_CORE,
464 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
465 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
466 {"merom", PROCESSOR_CORE2,
467 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
468 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3},
469 {"core2", PROCESSOR_CORE2,
470 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
471 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3},
472 {"k6", PROCESSOR_K6,
473 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX},
474 {"k6_2", PROCESSOR_K6,
475 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow},
476 {"athlon", PROCESSOR_ATHLON,
477 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
478 |CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA},
479 {"sledgehammer", PROCESSOR_K8,
480 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
481 |CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2},
482 {"opteron", PROCESSOR_K8,
483 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
484 |CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2},
485 {"k8", PROCESSOR_K8,
486 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
487 |CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2},
488 {"amdfam10", PROCESSOR_AMDFAM10,
489 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuSledgehammer
490 |CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2|CpuSSE3|CpuSSE4a
491 |CpuABM},
492 {".mmx", PROCESSOR_UNKNOWN,
493 CpuMMX},
494 {".sse", PROCESSOR_UNKNOWN,
495 CpuMMX|CpuMMX2|CpuSSE},
496 {".sse2", PROCESSOR_UNKNOWN,
497 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2},
498 {".sse3", PROCESSOR_UNKNOWN,
499 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
500 {".ssse3", PROCESSOR_UNKNOWN,
501 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3},
502 {".3dnow", PROCESSOR_UNKNOWN,
503 CpuMMX|Cpu3dnow},
504 {".3dnowa", PROCESSOR_UNKNOWN,
505 CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA},
506 {".padlock", PROCESSOR_UNKNOWN,
507 CpuPadLock},
508 {".pacifica", PROCESSOR_UNKNOWN,
509 CpuSVME},
510 {".svme", PROCESSOR_UNKNOWN,
511 CpuSVME},
512 {".sse4a", PROCESSOR_UNKNOWN,
513 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSE4a},
514 {".abm", PROCESSOR_UNKNOWN,
515 CpuABM}
516 };
517
518 const pseudo_typeS md_pseudo_table[] =
519 {
520 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
521 {"align", s_align_bytes, 0},
522 #else
523 {"align", s_align_ptwo, 0},
524 #endif
525 {"arch", set_cpu_arch, 0},
526 #ifndef I386COFF
527 {"bss", s_bss, 0},
528 #endif
529 {"ffloat", float_cons, 'f'},
530 {"dfloat", float_cons, 'd'},
531 {"tfloat", float_cons, 'x'},
532 {"value", cons, 2},
533 {"slong", signed_cons, 4},
534 {"noopt", s_ignore, 0},
535 {"optim", s_ignore, 0},
536 {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
537 {"code16", set_code_flag, CODE_16BIT},
538 {"code32", set_code_flag, CODE_32BIT},
539 {"code64", set_code_flag, CODE_64BIT},
540 {"intel_syntax", set_intel_syntax, 1},
541 {"att_syntax", set_intel_syntax, 0},
542 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
543 {"largecomm", handle_large_common, 0},
544 #else
545 {"file", (void (*) (int)) dwarf2_directive_file, 0},
546 {"loc", dwarf2_directive_loc, 0},
547 {"loc_mark_labels", dwarf2_directive_loc_mark_labels, 0},
548 #endif
549 #ifdef TE_PE
550 {"secrel32", pe_directive_secrel, 0},
551 #endif
552 {0, 0, 0}
553 };
554
555 /* For interface with expression (). */
556 extern char *input_line_pointer;
557
558 /* Hash table for instruction mnemonic lookup. */
559 static struct hash_control *op_hash;
560
561 /* Hash table for register lookup. */
562 static struct hash_control *reg_hash;
563 \f
564 void
565 i386_align_code (fragS *fragP, int count)
566 {
567 /* Various efficient no-op patterns for aligning code labels.
568 Note: Don't try to assemble the instructions in the comments.
569 0L and 0w are not legal. */
570 static const char f32_1[] =
571 {0x90}; /* nop */
572 static const char f32_2[] =
573 {0x66,0x90}; /* xchg %ax,%ax */
574 static const char f32_3[] =
575 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
576 static const char f32_4[] =
577 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
578 static const char f32_5[] =
579 {0x90, /* nop */
580 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
581 static const char f32_6[] =
582 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
583 static const char f32_7[] =
584 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
585 static const char f32_8[] =
586 {0x90, /* nop */
587 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
588 static const char f32_9[] =
589 {0x89,0xf6, /* movl %esi,%esi */
590 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
591 static const char f32_10[] =
592 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
593 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
594 static const char f32_11[] =
595 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
596 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
597 static const char f32_12[] =
598 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
599 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
600 static const char f32_13[] =
601 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
602 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
603 static const char f32_14[] =
604 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
605 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
606 static const char f32_15[] =
607 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
608 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
609 static const char f16_3[] =
610 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
611 static const char f16_4[] =
612 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
613 static const char f16_5[] =
614 {0x90, /* nop */
615 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
616 static const char f16_6[] =
617 {0x89,0xf6, /* mov %si,%si */
618 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
619 static const char f16_7[] =
620 {0x8d,0x74,0x00, /* lea 0(%si),%si */
621 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
622 static const char f16_8[] =
623 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
624 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
625 static const char *const f32_patt[] = {
626 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
627 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
628 };
629 static const char *const f16_patt[] = {
630 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
631 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
632 };
633 /* nopl (%[re]ax) */
634 static const char alt_3[] =
635 {0x0f,0x1f,0x00};
636 /* nopl 0(%[re]ax) */
637 static const char alt_4[] =
638 {0x0f,0x1f,0x40,0x00};
639 /* nopl 0(%[re]ax,%[re]ax,1) */
640 static const char alt_5[] =
641 {0x0f,0x1f,0x44,0x00,0x00};
642 /* nopw 0(%[re]ax,%[re]ax,1) */
643 static const char alt_6[] =
644 {0x66,0x0f,0x1f,0x44,0x00,0x00};
645 /* nopl 0L(%[re]ax) */
646 static const char alt_7[] =
647 {0x0f,0x1f,0x80,0x00,0x00,0x00,0x00};
648 /* nopl 0L(%[re]ax,%[re]ax,1) */
649 static const char alt_8[] =
650 {0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
651 /* nopw 0L(%[re]ax,%[re]ax,1) */
652 static const char alt_9[] =
653 {0x66,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
654 /* nopw %cs:0L(%[re]ax,%[re]ax,1) */
655 static const char alt_10[] =
656 {0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
657 /* data16
658 nopw %cs:0L(%[re]ax,%[re]ax,1) */
659 static const char alt_long_11[] =
660 {0x66,
661 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
662 /* data16
663 data16
664 nopw %cs:0L(%[re]ax,%[re]ax,1) */
665 static const char alt_long_12[] =
666 {0x66,
667 0x66,
668 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
669 /* data16
670 data16
671 data16
672 nopw %cs:0L(%[re]ax,%[re]ax,1) */
673 static const char alt_long_13[] =
674 {0x66,
675 0x66,
676 0x66,
677 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
678 /* data16
679 data16
680 data16
681 data16
682 nopw %cs:0L(%[re]ax,%[re]ax,1) */
683 static const char alt_long_14[] =
684 {0x66,
685 0x66,
686 0x66,
687 0x66,
688 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
689 /* data16
690 data16
691 data16
692 data16
693 data16
694 nopw %cs:0L(%[re]ax,%[re]ax,1) */
695 static const char alt_long_15[] =
696 {0x66,
697 0x66,
698 0x66,
699 0x66,
700 0x66,
701 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
702 /* nopl 0(%[re]ax,%[re]ax,1)
703 nopw 0(%[re]ax,%[re]ax,1) */
704 static const char alt_short_11[] =
705 {0x0f,0x1f,0x44,0x00,0x00,
706 0x66,0x0f,0x1f,0x44,0x00,0x00};
707 /* nopw 0(%[re]ax,%[re]ax,1)
708 nopw 0(%[re]ax,%[re]ax,1) */
709 static const char alt_short_12[] =
710 {0x66,0x0f,0x1f,0x44,0x00,0x00,
711 0x66,0x0f,0x1f,0x44,0x00,0x00};
712 /* nopw 0(%[re]ax,%[re]ax,1)
713 nopl 0L(%[re]ax) */
714 static const char alt_short_13[] =
715 {0x66,0x0f,0x1f,0x44,0x00,0x00,
716 0x0f,0x1f,0x80,0x00,0x00,0x00,0x00};
717 /* nopl 0L(%[re]ax)
718 nopl 0L(%[re]ax) */
719 static const char alt_short_14[] =
720 {0x0f,0x1f,0x80,0x00,0x00,0x00,0x00,
721 0x0f,0x1f,0x80,0x00,0x00,0x00,0x00};
722 /* nopl 0L(%[re]ax)
723 nopl 0L(%[re]ax,%[re]ax,1) */
724 static const char alt_short_15[] =
725 {0x0f,0x1f,0x80,0x00,0x00,0x00,0x00,
726 0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
727 static const char *const alt_short_patt[] = {
728 f32_1, f32_2, alt_3, alt_4, alt_5, alt_6, alt_7, alt_8,
729 alt_9, alt_10, alt_short_11, alt_short_12, alt_short_13,
730 alt_short_14, alt_short_15
731 };
732 static const char *const alt_long_patt[] = {
733 f32_1, f32_2, alt_3, alt_4, alt_5, alt_6, alt_7, alt_8,
734 alt_9, alt_10, alt_long_11, alt_long_12, alt_long_13,
735 alt_long_14, alt_long_15
736 };
737
738 if (count <= 0 || count > 15)
739 return;
740
741 /* We need to decide which NOP sequence to use for 32bit and
742 64bit. When -mtune= is used:
743
744 1. For PROCESSOR_I486, PROCESSOR_PENTIUM and PROCESSOR_GENERIC32,
745 f32_patt will be used.
746 2. For PROCESSOR_K8 and PROCESSOR_AMDFAM10 in 64bit, NOPs with
747 0x66 prefix will be used.
748 3. For PROCESSOR_CORE2, alt_long_patt will be used.
749 4. For PROCESSOR_PENTIUMPRO, PROCESSOR_PENTIUM4, PROCESSOR_NOCONA,
750 PROCESSOR_CORE, PROCESSOR_CORE2, PROCESSOR_K6, PROCESSOR_ATHLON
751 and PROCESSOR_GENERIC64, alt_short_patt will be used.
752
753 When -mtune= isn't used, alt_short_patt will be used if
754 cpu_arch_isa_flags has Cpu686. Otherwise, f32_patt will be used.
755
756 When -march= or .arch is used, we can't use anything beyond
757 cpu_arch_isa_flags. */
758
759 if (flag_code == CODE_16BIT)
760 {
761 memcpy (fragP->fr_literal + fragP->fr_fix,
762 f16_patt[count - 1], count);
763 if (count > 8)
764 /* Adjust jump offset. */
765 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
766 }
767 else if (flag_code == CODE_64BIT && cpu_arch_tune == PROCESSOR_K8)
768 {
769 int i;
770 int nnops = (count + 3) / 4;
771 int len = count / nnops;
772 int remains = count - nnops * len;
773 int pos = 0;
774
775 /* The recommended way to pad 64bit code is to use NOPs preceded
776 by maximally four 0x66 prefixes. Balance the size of nops. */
777 for (i = 0; i < remains; i++)
778 {
779 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len);
780 fragP->fr_literal[fragP->fr_fix + pos + len] = 0x90;
781 pos += len + 1;
782 }
783 for (; i < nnops; i++)
784 {
785 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len - 1);
786 fragP->fr_literal[fragP->fr_fix + pos + len - 1] = 0x90;
787 pos += len;
788 }
789 }
790 else
791 {
792 const char *const *patt = NULL;
793
794 if (cpu_arch_isa == PROCESSOR_UNKNOWN)
795 {
796 /* PROCESSOR_UNKNOWN means that all ISAs may be used. */
797 switch (cpu_arch_tune)
798 {
799 case PROCESSOR_UNKNOWN:
800 /* We use cpu_arch_isa_flags to check if we SHOULD
801 optimize for Cpu686. */
802 if ((cpu_arch_isa_flags & Cpu686) != 0)
803 patt = alt_short_patt;
804 else
805 patt = f32_patt;
806 break;
807 case PROCESSOR_CORE2:
808 patt = alt_long_patt;
809 break;
810 case PROCESSOR_PENTIUMPRO:
811 case PROCESSOR_PENTIUM4:
812 case PROCESSOR_NOCONA:
813 case PROCESSOR_CORE:
814 case PROCESSOR_K6:
815 case PROCESSOR_ATHLON:
816 case PROCESSOR_K8:
817 case PROCESSOR_GENERIC64:
818 case PROCESSOR_AMDFAM10:
819 patt = alt_short_patt;
820 break;
821 case PROCESSOR_I486:
822 case PROCESSOR_PENTIUM:
823 case PROCESSOR_GENERIC32:
824 patt = f32_patt;
825 break;
826 }
827 }
828 else
829 {
830 switch (cpu_arch_tune)
831 {
832 case PROCESSOR_UNKNOWN:
833 /* When cpu_arch_isa is net, cpu_arch_tune shouldn't be
834 PROCESSOR_UNKNOWN. */
835 abort ();
836 break;
837
838 case PROCESSOR_I486:
839 case PROCESSOR_PENTIUM:
840 case PROCESSOR_PENTIUMPRO:
841 case PROCESSOR_PENTIUM4:
842 case PROCESSOR_NOCONA:
843 case PROCESSOR_CORE:
844 case PROCESSOR_K6:
845 case PROCESSOR_ATHLON:
846 case PROCESSOR_K8:
847 case PROCESSOR_AMDFAM10:
848 case PROCESSOR_GENERIC32:
849 /* We use cpu_arch_isa_flags to check if we CAN optimize
850 for Cpu686. */
851 if ((cpu_arch_isa_flags & Cpu686) != 0)
852 patt = alt_short_patt;
853 else
854 patt = f32_patt;
855 break;
856 case PROCESSOR_CORE2:
857 if ((cpu_arch_isa_flags & Cpu686) != 0)
858 patt = alt_long_patt;
859 else
860 patt = f32_patt;
861 break;
862 case PROCESSOR_GENERIC64:
863 patt = alt_short_patt;
864 break;
865 }
866 }
867
868 memcpy (fragP->fr_literal + fragP->fr_fix,
869 patt[count - 1], count);
870 }
871 fragP->fr_var = count;
872 }
873
874 static INLINE unsigned int
875 mode_from_disp_size (unsigned int t)
876 {
877 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
878 }
879
880 static INLINE int
881 fits_in_signed_byte (offsetT num)
882 {
883 return (num >= -128) && (num <= 127);
884 }
885
886 static INLINE int
887 fits_in_unsigned_byte (offsetT num)
888 {
889 return (num & 0xff) == num;
890 }
891
892 static INLINE int
893 fits_in_unsigned_word (offsetT num)
894 {
895 return (num & 0xffff) == num;
896 }
897
898 static INLINE int
899 fits_in_signed_word (offsetT num)
900 {
901 return (-32768 <= num) && (num <= 32767);
902 }
903
904 static INLINE int
905 fits_in_signed_long (offsetT num ATTRIBUTE_UNUSED)
906 {
907 #ifndef BFD64
908 return 1;
909 #else
910 return (!(((offsetT) -1 << 31) & num)
911 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
912 #endif
913 } /* fits_in_signed_long() */
914
915 static INLINE int
916 fits_in_unsigned_long (offsetT num ATTRIBUTE_UNUSED)
917 {
918 #ifndef BFD64
919 return 1;
920 #else
921 return (num & (((offsetT) 2 << 31) - 1)) == num;
922 #endif
923 } /* fits_in_unsigned_long() */
924
925 static unsigned int
926 smallest_imm_type (offsetT num)
927 {
928 if (cpu_arch_flags != (Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64))
929 {
930 /* This code is disabled on the 486 because all the Imm1 forms
931 in the opcode table are slower on the i486. They're the
932 versions with the implicitly specified single-position
933 displacement, which has another syntax if you really want to
934 use that form. */
935 if (num == 1)
936 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
937 }
938 return (fits_in_signed_byte (num)
939 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
940 : fits_in_unsigned_byte (num)
941 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
942 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
943 ? (Imm16 | Imm32 | Imm32S | Imm64)
944 : fits_in_signed_long (num)
945 ? (Imm32 | Imm32S | Imm64)
946 : fits_in_unsigned_long (num)
947 ? (Imm32 | Imm64)
948 : Imm64);
949 }
950
951 static offsetT
952 offset_in_range (offsetT val, int size)
953 {
954 addressT mask;
955
956 switch (size)
957 {
958 case 1: mask = ((addressT) 1 << 8) - 1; break;
959 case 2: mask = ((addressT) 1 << 16) - 1; break;
960 case 4: mask = ((addressT) 2 << 31) - 1; break;
961 #ifdef BFD64
962 case 8: mask = ((addressT) 2 << 63) - 1; break;
963 #endif
964 default: abort ();
965 }
966
967 /* If BFD64, sign extend val. */
968 if (!use_rela_relocations)
969 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
970 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
971
972 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
973 {
974 char buf1[40], buf2[40];
975
976 sprint_value (buf1, val);
977 sprint_value (buf2, val & mask);
978 as_warn (_("%s shortened to %s"), buf1, buf2);
979 }
980 return val & mask;
981 }
982
983 /* Returns 0 if attempting to add a prefix where one from the same
984 class already exists, 1 if non rep/repne added, 2 if rep/repne
985 added. */
986 static int
987 add_prefix (unsigned int prefix)
988 {
989 int ret = 1;
990 unsigned int q;
991
992 if (prefix >= REX_OPCODE && prefix < REX_OPCODE + 16
993 && flag_code == CODE_64BIT)
994 {
995 if ((i.prefix[REX_PREFIX] & prefix & REX_MODE64)
996 || ((i.prefix[REX_PREFIX] & (REX_EXTX | REX_EXTY | REX_EXTZ))
997 && (prefix & (REX_EXTX | REX_EXTY | REX_EXTZ))))
998 ret = 0;
999 q = REX_PREFIX;
1000 }
1001 else
1002 {
1003 switch (prefix)
1004 {
1005 default:
1006 abort ();
1007
1008 case CS_PREFIX_OPCODE:
1009 case DS_PREFIX_OPCODE:
1010 case ES_PREFIX_OPCODE:
1011 case FS_PREFIX_OPCODE:
1012 case GS_PREFIX_OPCODE:
1013 case SS_PREFIX_OPCODE:
1014 q = SEG_PREFIX;
1015 break;
1016
1017 case REPNE_PREFIX_OPCODE:
1018 case REPE_PREFIX_OPCODE:
1019 ret = 2;
1020 /* fall thru */
1021 case LOCK_PREFIX_OPCODE:
1022 q = LOCKREP_PREFIX;
1023 break;
1024
1025 case FWAIT_OPCODE:
1026 q = WAIT_PREFIX;
1027 break;
1028
1029 case ADDR_PREFIX_OPCODE:
1030 q = ADDR_PREFIX;
1031 break;
1032
1033 case DATA_PREFIX_OPCODE:
1034 q = DATA_PREFIX;
1035 break;
1036 }
1037 if (i.prefix[q] != 0)
1038 ret = 0;
1039 }
1040
1041 if (ret)
1042 {
1043 if (!i.prefix[q])
1044 ++i.prefixes;
1045 i.prefix[q] |= prefix;
1046 }
1047 else
1048 as_bad (_("same type of prefix used twice"));
1049
1050 return ret;
1051 }
1052
1053 static void
1054 set_code_flag (int value)
1055 {
1056 flag_code = value;
1057 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
1058 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
1059 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
1060 {
1061 as_bad (_("64bit mode not supported on this CPU."));
1062 }
1063 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
1064 {
1065 as_bad (_("32bit mode not supported on this CPU."));
1066 }
1067 stackop_size = '\0';
1068 }
1069
1070 static void
1071 set_16bit_gcc_code_flag (int new_code_flag)
1072 {
1073 flag_code = new_code_flag;
1074 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
1075 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
1076 stackop_size = LONG_MNEM_SUFFIX;
1077 }
1078
1079 static void
1080 set_intel_syntax (int syntax_flag)
1081 {
1082 /* Find out if register prefixing is specified. */
1083 int ask_naked_reg = 0;
1084
1085 SKIP_WHITESPACE ();
1086 if (!is_end_of_line[(unsigned char) *input_line_pointer])
1087 {
1088 char *string = input_line_pointer;
1089 int e = get_symbol_end ();
1090
1091 if (strcmp (string, "prefix") == 0)
1092 ask_naked_reg = 1;
1093 else if (strcmp (string, "noprefix") == 0)
1094 ask_naked_reg = -1;
1095 else
1096 as_bad (_("bad argument to syntax directive."));
1097 *input_line_pointer = e;
1098 }
1099 demand_empty_rest_of_line ();
1100
1101 intel_syntax = syntax_flag;
1102
1103 if (ask_naked_reg == 0)
1104 allow_naked_reg = (intel_syntax
1105 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
1106 else
1107 allow_naked_reg = (ask_naked_reg < 0);
1108
1109 identifier_chars['%'] = intel_syntax && allow_naked_reg ? '%' : 0;
1110 identifier_chars['$'] = intel_syntax ? '$' : 0;
1111 register_prefix = allow_naked_reg ? "" : "%";
1112 }
1113
1114 static void
1115 set_cpu_arch (int dummy ATTRIBUTE_UNUSED)
1116 {
1117 SKIP_WHITESPACE ();
1118
1119 if (!is_end_of_line[(unsigned char) *input_line_pointer])
1120 {
1121 char *string = input_line_pointer;
1122 int e = get_symbol_end ();
1123 unsigned int i;
1124
1125 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
1126 {
1127 if (strcmp (string, cpu_arch[i].name) == 0)
1128 {
1129 if (*string != '.')
1130 {
1131 cpu_arch_name = cpu_arch[i].name;
1132 cpu_sub_arch_name = NULL;
1133 cpu_arch_flags = (cpu_arch[i].flags
1134 | (flag_code == CODE_64BIT
1135 ? Cpu64 : CpuNo64));
1136 cpu_arch_isa = cpu_arch[i].type;
1137 cpu_arch_isa_flags = cpu_arch[i].flags;
1138 if (!cpu_arch_tune_set)
1139 {
1140 cpu_arch_tune = cpu_arch_isa;
1141 cpu_arch_tune_flags = cpu_arch_isa_flags;
1142 }
1143 break;
1144 }
1145 if ((cpu_arch_flags | cpu_arch[i].flags) != cpu_arch_flags)
1146 {
1147 cpu_sub_arch_name = cpu_arch[i].name;
1148 cpu_arch_flags |= cpu_arch[i].flags;
1149 }
1150 *input_line_pointer = e;
1151 demand_empty_rest_of_line ();
1152 return;
1153 }
1154 }
1155 if (i >= ARRAY_SIZE (cpu_arch))
1156 as_bad (_("no such architecture: `%s'"), string);
1157
1158 *input_line_pointer = e;
1159 }
1160 else
1161 as_bad (_("missing cpu architecture"));
1162
1163 no_cond_jump_promotion = 0;
1164 if (*input_line_pointer == ','
1165 && !is_end_of_line[(unsigned char) input_line_pointer[1]])
1166 {
1167 char *string = ++input_line_pointer;
1168 int e = get_symbol_end ();
1169
1170 if (strcmp (string, "nojumps") == 0)
1171 no_cond_jump_promotion = 1;
1172 else if (strcmp (string, "jumps") == 0)
1173 ;
1174 else
1175 as_bad (_("no such architecture modifier: `%s'"), string);
1176
1177 *input_line_pointer = e;
1178 }
1179
1180 demand_empty_rest_of_line ();
1181 }
1182
1183 unsigned long
1184 i386_mach ()
1185 {
1186 if (!strcmp (default_arch, "x86_64"))
1187 return bfd_mach_x86_64;
1188 else if (!strcmp (default_arch, "i386"))
1189 return bfd_mach_i386_i386;
1190 else
1191 as_fatal (_("Unknown architecture"));
1192 }
1193 \f
1194 void
1195 md_begin ()
1196 {
1197 const char *hash_err;
1198
1199 /* Initialize op_hash hash table. */
1200 op_hash = hash_new ();
1201
1202 {
1203 const template *optab;
1204 templates *core_optab;
1205
1206 /* Setup for loop. */
1207 optab = i386_optab;
1208 core_optab = (templates *) xmalloc (sizeof (templates));
1209 core_optab->start = optab;
1210
1211 while (1)
1212 {
1213 ++optab;
1214 if (optab->name == NULL
1215 || strcmp (optab->name, (optab - 1)->name) != 0)
1216 {
1217 /* different name --> ship out current template list;
1218 add to hash table; & begin anew. */
1219 core_optab->end = optab;
1220 hash_err = hash_insert (op_hash,
1221 (optab - 1)->name,
1222 (PTR) core_optab);
1223 if (hash_err)
1224 {
1225 as_fatal (_("Internal Error: Can't hash %s: %s"),
1226 (optab - 1)->name,
1227 hash_err);
1228 }
1229 if (optab->name == NULL)
1230 break;
1231 core_optab = (templates *) xmalloc (sizeof (templates));
1232 core_optab->start = optab;
1233 }
1234 }
1235 }
1236
1237 /* Initialize reg_hash hash table. */
1238 reg_hash = hash_new ();
1239 {
1240 const reg_entry *regtab;
1241
1242 for (regtab = i386_regtab;
1243 regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
1244 regtab++)
1245 {
1246 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
1247 if (hash_err)
1248 as_fatal (_("Internal Error: Can't hash %s: %s"),
1249 regtab->reg_name,
1250 hash_err);
1251 }
1252 }
1253
1254 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
1255 {
1256 int c;
1257 char *p;
1258
1259 for (c = 0; c < 256; c++)
1260 {
1261 if (ISDIGIT (c))
1262 {
1263 digit_chars[c] = c;
1264 mnemonic_chars[c] = c;
1265 register_chars[c] = c;
1266 operand_chars[c] = c;
1267 }
1268 else if (ISLOWER (c))
1269 {
1270 mnemonic_chars[c] = c;
1271 register_chars[c] = c;
1272 operand_chars[c] = c;
1273 }
1274 else if (ISUPPER (c))
1275 {
1276 mnemonic_chars[c] = TOLOWER (c);
1277 register_chars[c] = mnemonic_chars[c];
1278 operand_chars[c] = c;
1279 }
1280
1281 if (ISALPHA (c) || ISDIGIT (c))
1282 identifier_chars[c] = c;
1283 else if (c >= 128)
1284 {
1285 identifier_chars[c] = c;
1286 operand_chars[c] = c;
1287 }
1288 }
1289
1290 #ifdef LEX_AT
1291 identifier_chars['@'] = '@';
1292 #endif
1293 #ifdef LEX_QM
1294 identifier_chars['?'] = '?';
1295 operand_chars['?'] = '?';
1296 #endif
1297 digit_chars['-'] = '-';
1298 mnemonic_chars['-'] = '-';
1299 identifier_chars['_'] = '_';
1300 identifier_chars['.'] = '.';
1301
1302 for (p = operand_special_chars; *p != '\0'; p++)
1303 operand_chars[(unsigned char) *p] = *p;
1304 }
1305
1306 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1307 if (IS_ELF)
1308 {
1309 record_alignment (text_section, 2);
1310 record_alignment (data_section, 2);
1311 record_alignment (bss_section, 2);
1312 }
1313 #endif
1314
1315 if (flag_code == CODE_64BIT)
1316 {
1317 x86_dwarf2_return_column = 16;
1318 x86_cie_data_alignment = -8;
1319 }
1320 else
1321 {
1322 x86_dwarf2_return_column = 8;
1323 x86_cie_data_alignment = -4;
1324 }
1325 }
1326
1327 void
1328 i386_print_statistics (FILE *file)
1329 {
1330 hash_print_statistics (file, "i386 opcode", op_hash);
1331 hash_print_statistics (file, "i386 register", reg_hash);
1332 }
1333 \f
1334 #ifdef DEBUG386
1335
1336 /* Debugging routines for md_assemble. */
1337 static void pte (template *);
1338 static void pt (unsigned int);
1339 static void pe (expressionS *);
1340 static void ps (symbolS *);
1341
1342 static void
1343 pi (char *line, i386_insn *x)
1344 {
1345 unsigned int i;
1346
1347 fprintf (stdout, "%s: template ", line);
1348 pte (&x->tm);
1349 fprintf (stdout, " address: base %s index %s scale %x\n",
1350 x->base_reg ? x->base_reg->reg_name : "none",
1351 x->index_reg ? x->index_reg->reg_name : "none",
1352 x->log2_scale_factor);
1353 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
1354 x->rm.mode, x->rm.reg, x->rm.regmem);
1355 fprintf (stdout, " sib: base %x index %x scale %x\n",
1356 x->sib.base, x->sib.index, x->sib.scale);
1357 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
1358 (x->rex & REX_MODE64) != 0,
1359 (x->rex & REX_EXTX) != 0,
1360 (x->rex & REX_EXTY) != 0,
1361 (x->rex & REX_EXTZ) != 0);
1362 for (i = 0; i < x->operands; i++)
1363 {
1364 fprintf (stdout, " #%d: ", i + 1);
1365 pt (x->types[i]);
1366 fprintf (stdout, "\n");
1367 if (x->types[i]
1368 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
1369 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
1370 if (x->types[i] & Imm)
1371 pe (x->op[i].imms);
1372 if (x->types[i] & Disp)
1373 pe (x->op[i].disps);
1374 }
1375 }
1376
1377 static void
1378 pte (template *t)
1379 {
1380 unsigned int i;
1381 fprintf (stdout, " %d operands ", t->operands);
1382 fprintf (stdout, "opcode %x ", t->base_opcode);
1383 if (t->extension_opcode != None)
1384 fprintf (stdout, "ext %x ", t->extension_opcode);
1385 if (t->opcode_modifier & D)
1386 fprintf (stdout, "D");
1387 if (t->opcode_modifier & W)
1388 fprintf (stdout, "W");
1389 fprintf (stdout, "\n");
1390 for (i = 0; i < t->operands; i++)
1391 {
1392 fprintf (stdout, " #%d type ", i + 1);
1393 pt (t->operand_types[i]);
1394 fprintf (stdout, "\n");
1395 }
1396 }
1397
1398 static void
1399 pe (expressionS *e)
1400 {
1401 fprintf (stdout, " operation %d\n", e->X_op);
1402 fprintf (stdout, " add_number %ld (%lx)\n",
1403 (long) e->X_add_number, (long) e->X_add_number);
1404 if (e->X_add_symbol)
1405 {
1406 fprintf (stdout, " add_symbol ");
1407 ps (e->X_add_symbol);
1408 fprintf (stdout, "\n");
1409 }
1410 if (e->X_op_symbol)
1411 {
1412 fprintf (stdout, " op_symbol ");
1413 ps (e->X_op_symbol);
1414 fprintf (stdout, "\n");
1415 }
1416 }
1417
1418 static void
1419 ps (symbolS *s)
1420 {
1421 fprintf (stdout, "%s type %s%s",
1422 S_GET_NAME (s),
1423 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
1424 segment_name (S_GET_SEGMENT (s)));
1425 }
1426
1427 static struct type_name
1428 {
1429 unsigned int mask;
1430 char *tname;
1431 }
1432 const type_names[] =
1433 {
1434 { Reg8, "r8" },
1435 { Reg16, "r16" },
1436 { Reg32, "r32" },
1437 { Reg64, "r64" },
1438 { Imm8, "i8" },
1439 { Imm8S, "i8s" },
1440 { Imm16, "i16" },
1441 { Imm32, "i32" },
1442 { Imm32S, "i32s" },
1443 { Imm64, "i64" },
1444 { Imm1, "i1" },
1445 { BaseIndex, "BaseIndex" },
1446 { Disp8, "d8" },
1447 { Disp16, "d16" },
1448 { Disp32, "d32" },
1449 { Disp32S, "d32s" },
1450 { Disp64, "d64" },
1451 { InOutPortReg, "InOutPortReg" },
1452 { ShiftCount, "ShiftCount" },
1453 { Control, "control reg" },
1454 { Test, "test reg" },
1455 { Debug, "debug reg" },
1456 { FloatReg, "FReg" },
1457 { FloatAcc, "FAcc" },
1458 { SReg2, "SReg2" },
1459 { SReg3, "SReg3" },
1460 { Acc, "Acc" },
1461 { JumpAbsolute, "Jump Absolute" },
1462 { RegMMX, "rMMX" },
1463 { RegXMM, "rXMM" },
1464 { EsSeg, "es" },
1465 { 0, "" }
1466 };
1467
1468 static void
1469 pt (t)
1470 unsigned int t;
1471 {
1472 const struct type_name *ty;
1473
1474 for (ty = type_names; ty->mask; ty++)
1475 if (t & ty->mask)
1476 fprintf (stdout, "%s, ", ty->tname);
1477 fflush (stdout);
1478 }
1479
1480 #endif /* DEBUG386 */
1481 \f
1482 static bfd_reloc_code_real_type
1483 reloc (unsigned int size,
1484 int pcrel,
1485 int sign,
1486 bfd_reloc_code_real_type other)
1487 {
1488 if (other != NO_RELOC)
1489 {
1490 reloc_howto_type *reloc;
1491
1492 if (size == 8)
1493 switch (other)
1494 {
1495 case BFD_RELOC_X86_64_GOT32:
1496 return BFD_RELOC_X86_64_GOT64;
1497 break;
1498 case BFD_RELOC_X86_64_PLTOFF64:
1499 return BFD_RELOC_X86_64_PLTOFF64;
1500 break;
1501 case BFD_RELOC_X86_64_GOTPC32:
1502 other = BFD_RELOC_X86_64_GOTPC64;
1503 break;
1504 case BFD_RELOC_X86_64_GOTPCREL:
1505 other = BFD_RELOC_X86_64_GOTPCREL64;
1506 break;
1507 case BFD_RELOC_X86_64_TPOFF32:
1508 other = BFD_RELOC_X86_64_TPOFF64;
1509 break;
1510 case BFD_RELOC_X86_64_DTPOFF32:
1511 other = BFD_RELOC_X86_64_DTPOFF64;
1512 break;
1513 default:
1514 break;
1515 }
1516
1517 /* Sign-checking 4-byte relocations in 16-/32-bit code is pointless. */
1518 if (size == 4 && flag_code != CODE_64BIT)
1519 sign = -1;
1520
1521 reloc = bfd_reloc_type_lookup (stdoutput, other);
1522 if (!reloc)
1523 as_bad (_("unknown relocation (%u)"), other);
1524 else if (size != bfd_get_reloc_size (reloc))
1525 as_bad (_("%u-byte relocation cannot be applied to %u-byte field"),
1526 bfd_get_reloc_size (reloc),
1527 size);
1528 else if (pcrel && !reloc->pc_relative)
1529 as_bad (_("non-pc-relative relocation for pc-relative field"));
1530 else if ((reloc->complain_on_overflow == complain_overflow_signed
1531 && !sign)
1532 || (reloc->complain_on_overflow == complain_overflow_unsigned
1533 && sign > 0))
1534 as_bad (_("relocated field and relocation type differ in signedness"));
1535 else
1536 return other;
1537 return NO_RELOC;
1538 }
1539
1540 if (pcrel)
1541 {
1542 if (!sign)
1543 as_bad (_("there are no unsigned pc-relative relocations"));
1544 switch (size)
1545 {
1546 case 1: return BFD_RELOC_8_PCREL;
1547 case 2: return BFD_RELOC_16_PCREL;
1548 case 4: return BFD_RELOC_32_PCREL;
1549 case 8: return BFD_RELOC_64_PCREL;
1550 }
1551 as_bad (_("cannot do %u byte pc-relative relocation"), size);
1552 }
1553 else
1554 {
1555 if (sign > 0)
1556 switch (size)
1557 {
1558 case 4: return BFD_RELOC_X86_64_32S;
1559 }
1560 else
1561 switch (size)
1562 {
1563 case 1: return BFD_RELOC_8;
1564 case 2: return BFD_RELOC_16;
1565 case 4: return BFD_RELOC_32;
1566 case 8: return BFD_RELOC_64;
1567 }
1568 as_bad (_("cannot do %s %u byte relocation"),
1569 sign > 0 ? "signed" : "unsigned", size);
1570 }
1571
1572 abort ();
1573 return BFD_RELOC_NONE;
1574 }
1575
1576 /* Here we decide which fixups can be adjusted to make them relative to
1577 the beginning of the section instead of the symbol. Basically we need
1578 to make sure that the dynamic relocations are done correctly, so in
1579 some cases we force the original symbol to be used. */
1580
1581 int
1582 tc_i386_fix_adjustable (fixS *fixP ATTRIBUTE_UNUSED)
1583 {
1584 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1585 if (!IS_ELF)
1586 return 1;
1587
1588 /* Don't adjust pc-relative references to merge sections in 64-bit
1589 mode. */
1590 if (use_rela_relocations
1591 && (S_GET_SEGMENT (fixP->fx_addsy)->flags & SEC_MERGE) != 0
1592 && fixP->fx_pcrel)
1593 return 0;
1594
1595 /* The x86_64 GOTPCREL are represented as 32bit PCrel relocations
1596 and changed later by validate_fix. */
1597 if (GOT_symbol && fixP->fx_subsy == GOT_symbol
1598 && fixP->fx_r_type == BFD_RELOC_32_PCREL)
1599 return 0;
1600
1601 /* adjust_reloc_syms doesn't know about the GOT. */
1602 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1603 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1604 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1605 || fixP->fx_r_type == BFD_RELOC_386_TLS_GD
1606 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDM
1607 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDO_32
1608 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE_32
1609 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE
1610 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTIE
1611 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE_32
1612 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE
1613 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTDESC
1614 || fixP->fx_r_type == BFD_RELOC_386_TLS_DESC_CALL
1615 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1616 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1617 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
1618 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSGD
1619 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSLD
1620 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF32
1621 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF64
1622 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTTPOFF
1623 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF32
1624 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF64
1625 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTOFF64
1626 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPC32_TLSDESC
1627 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSDESC_CALL
1628 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1629 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1630 return 0;
1631 #endif
1632 return 1;
1633 }
1634
1635 static int
1636 intel_float_operand (const char *mnemonic)
1637 {
1638 /* Note that the value returned is meaningful only for opcodes with (memory)
1639 operands, hence the code here is free to improperly handle opcodes that
1640 have no operands (for better performance and smaller code). */
1641
1642 if (mnemonic[0] != 'f')
1643 return 0; /* non-math */
1644
1645 switch (mnemonic[1])
1646 {
1647 /* fclex, fdecstp, fdisi, femms, feni, fincstp, finit, fsetpm, and
1648 the fs segment override prefix not currently handled because no
1649 call path can make opcodes without operands get here */
1650 case 'i':
1651 return 2 /* integer op */;
1652 case 'l':
1653 if (mnemonic[2] == 'd' && (mnemonic[3] == 'c' || mnemonic[3] == 'e'))
1654 return 3; /* fldcw/fldenv */
1655 break;
1656 case 'n':
1657 if (mnemonic[2] != 'o' /* fnop */)
1658 return 3; /* non-waiting control op */
1659 break;
1660 case 'r':
1661 if (mnemonic[2] == 's')
1662 return 3; /* frstor/frstpm */
1663 break;
1664 case 's':
1665 if (mnemonic[2] == 'a')
1666 return 3; /* fsave */
1667 if (mnemonic[2] == 't')
1668 {
1669 switch (mnemonic[3])
1670 {
1671 case 'c': /* fstcw */
1672 case 'd': /* fstdw */
1673 case 'e': /* fstenv */
1674 case 's': /* fsts[gw] */
1675 return 3;
1676 }
1677 }
1678 break;
1679 case 'x':
1680 if (mnemonic[2] == 'r' || mnemonic[2] == 's')
1681 return 0; /* fxsave/fxrstor are not really math ops */
1682 break;
1683 }
1684
1685 return 1;
1686 }
1687
1688 /* This is the guts of the machine-dependent assembler. LINE points to a
1689 machine dependent instruction. This function is supposed to emit
1690 the frags/bytes it assembles to. */
1691
1692 void
1693 md_assemble (line)
1694 char *line;
1695 {
1696 int j;
1697 char mnemonic[MAX_MNEM_SIZE];
1698
1699 /* Initialize globals. */
1700 memset (&i, '\0', sizeof (i));
1701 for (j = 0; j < MAX_OPERANDS; j++)
1702 i.reloc[j] = NO_RELOC;
1703 memset (disp_expressions, '\0', sizeof (disp_expressions));
1704 memset (im_expressions, '\0', sizeof (im_expressions));
1705 save_stack_p = save_stack;
1706
1707 /* First parse an instruction mnemonic & call i386_operand for the operands.
1708 We assume that the scrubber has arranged it so that line[0] is the valid
1709 start of a (possibly prefixed) mnemonic. */
1710
1711 line = parse_insn (line, mnemonic);
1712 if (line == NULL)
1713 return;
1714
1715 line = parse_operands (line, mnemonic);
1716 if (line == NULL)
1717 return;
1718
1719 /* The order of the immediates should be reversed
1720 for 2 immediates extrq and insertq instructions */
1721 if ((i.imm_operands == 2)
1722 && ((strcmp (mnemonic, "extrq") == 0)
1723 || (strcmp (mnemonic, "insertq") == 0)))
1724 {
1725 swap_2_operands (0, 1);
1726 /* "extrq" and insertq" are the only two instructions whose operands
1727 have to be reversed even though they have two immediate operands.
1728 */
1729 if (intel_syntax)
1730 swap_operands ();
1731 }
1732
1733 /* Now we've parsed the mnemonic into a set of templates, and have the
1734 operands at hand. */
1735
1736 /* All intel opcodes have reversed operands except for "bound" and
1737 "enter". We also don't reverse intersegment "jmp" and "call"
1738 instructions with 2 immediate operands so that the immediate segment
1739 precedes the offset, as it does when in AT&T mode. */
1740 if (intel_syntax
1741 && i.operands > 1
1742 && (strcmp (mnemonic, "bound") != 0)
1743 && (strcmp (mnemonic, "invlpga") != 0)
1744 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1745 swap_operands ();
1746
1747 if (i.imm_operands)
1748 optimize_imm ();
1749
1750 /* Don't optimize displacement for movabs since it only takes 64bit
1751 displacement. */
1752 if (i.disp_operands
1753 && (flag_code != CODE_64BIT
1754 || strcmp (mnemonic, "movabs") != 0))
1755 optimize_disp ();
1756
1757 /* Next, we find a template that matches the given insn,
1758 making sure the overlap of the given operands types is consistent
1759 with the template operand types. */
1760
1761 if (!match_template ())
1762 return;
1763
1764 if (intel_syntax)
1765 {
1766 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1767 if (SYSV386_COMPAT
1768 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1769 i.tm.base_opcode ^= Opcode_FloatR;
1770
1771 /* Zap movzx and movsx suffix. The suffix may have been set from
1772 "word ptr" or "byte ptr" on the source operand, but we'll use
1773 the suffix later to choose the destination register. */
1774 if ((i.tm.base_opcode & ~9) == 0x0fb6)
1775 {
1776 if (i.reg_operands < 2
1777 && !i.suffix
1778 && (~i.tm.opcode_modifier
1779 & (No_bSuf
1780 | No_wSuf
1781 | No_lSuf
1782 | No_sSuf
1783 | No_xSuf
1784 | No_qSuf)))
1785 as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
1786
1787 i.suffix = 0;
1788 }
1789 }
1790
1791 if (i.tm.opcode_modifier & FWait)
1792 if (!add_prefix (FWAIT_OPCODE))
1793 return;
1794
1795 /* Check string instruction segment overrides. */
1796 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1797 {
1798 if (!check_string ())
1799 return;
1800 }
1801
1802 if (!process_suffix ())
1803 return;
1804
1805 /* Make still unresolved immediate matches conform to size of immediate
1806 given in i.suffix. */
1807 if (!finalize_imm ())
1808 return;
1809
1810 if (i.types[0] & Imm1)
1811 i.imm_operands = 0; /* kludge for shift insns. */
1812 if (i.types[0] & ImplicitRegister)
1813 i.reg_operands--;
1814 if (i.types[1] & ImplicitRegister)
1815 i.reg_operands--;
1816 if (i.types[2] & ImplicitRegister)
1817 i.reg_operands--;
1818
1819 if (i.tm.opcode_modifier & ImmExt)
1820 {
1821 expressionS *exp;
1822
1823 if ((i.tm.cpu_flags & CpuSSE3) && i.operands > 0)
1824 {
1825 /* Streaming SIMD extensions 3 Instructions have the fixed
1826 operands with an opcode suffix which is coded in the same
1827 place as an 8-bit immediate field would be. Here we check
1828 those operands and remove them afterwards. */
1829 unsigned int x;
1830
1831 for (x = 0; x < i.operands; x++)
1832 if (i.op[x].regs->reg_num != x)
1833 as_bad (_("can't use register '%%%s' as operand %d in '%s'."),
1834 i.op[x].regs->reg_name, x + 1, i.tm.name);
1835 i.operands = 0;
1836 }
1837
1838 /* These AMD 3DNow! and Intel Katmai New Instructions have an
1839 opcode suffix which is coded in the same place as an 8-bit
1840 immediate field would be. Here we fake an 8-bit immediate
1841 operand from the opcode suffix stored in tm.extension_opcode. */
1842
1843 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
1844
1845 exp = &im_expressions[i.imm_operands++];
1846 i.op[i.operands].imms = exp;
1847 i.types[i.operands++] = Imm8;
1848 exp->X_op = O_constant;
1849 exp->X_add_number = i.tm.extension_opcode;
1850 i.tm.extension_opcode = None;
1851 }
1852
1853 /* For insns with operands there are more diddles to do to the opcode. */
1854 if (i.operands)
1855 {
1856 if (!process_operands ())
1857 return;
1858 }
1859 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
1860 {
1861 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
1862 as_warn (_("translating to `%sp'"), i.tm.name);
1863 }
1864
1865 /* Handle conversion of 'int $3' --> special int3 insn. */
1866 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
1867 {
1868 i.tm.base_opcode = INT3_OPCODE;
1869 i.imm_operands = 0;
1870 }
1871
1872 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
1873 && i.op[0].disps->X_op == O_constant)
1874 {
1875 /* Convert "jmp constant" (and "call constant") to a jump (call) to
1876 the absolute address given by the constant. Since ix86 jumps and
1877 calls are pc relative, we need to generate a reloc. */
1878 i.op[0].disps->X_add_symbol = &abs_symbol;
1879 i.op[0].disps->X_op = O_symbol;
1880 }
1881
1882 if ((i.tm.opcode_modifier & Rex64) != 0)
1883 i.rex |= REX_MODE64;
1884
1885 /* For 8 bit registers we need an empty rex prefix. Also if the
1886 instruction already has a prefix, we need to convert old
1887 registers to new ones. */
1888
1889 if (((i.types[0] & Reg8) != 0
1890 && (i.op[0].regs->reg_flags & RegRex64) != 0)
1891 || ((i.types[1] & Reg8) != 0
1892 && (i.op[1].regs->reg_flags & RegRex64) != 0)
1893 || (((i.types[0] & Reg8) != 0 || (i.types[1] & Reg8) != 0)
1894 && i.rex != 0))
1895 {
1896 int x;
1897
1898 i.rex |= REX_OPCODE;
1899 for (x = 0; x < 2; x++)
1900 {
1901 /* Look for 8 bit operand that uses old registers. */
1902 if ((i.types[x] & Reg8) != 0
1903 && (i.op[x].regs->reg_flags & RegRex64) == 0)
1904 {
1905 /* In case it is "hi" register, give up. */
1906 if (i.op[x].regs->reg_num > 3)
1907 as_bad (_("can't encode register '%%%s' in an "
1908 "instruction requiring REX prefix."),
1909 i.op[x].regs->reg_name);
1910
1911 /* Otherwise it is equivalent to the extended register.
1912 Since the encoding doesn't change this is merely
1913 cosmetic cleanup for debug output. */
1914
1915 i.op[x].regs = i.op[x].regs + 8;
1916 }
1917 }
1918 }
1919
1920 if (i.rex != 0)
1921 add_prefix (REX_OPCODE | i.rex);
1922
1923 /* We are ready to output the insn. */
1924 output_insn ();
1925 }
1926
1927 static char *
1928 parse_insn (char *line, char *mnemonic)
1929 {
1930 char *l = line;
1931 char *token_start = l;
1932 char *mnem_p;
1933 int supported;
1934 const template *t;
1935
1936 /* Non-zero if we found a prefix only acceptable with string insns. */
1937 const char *expecting_string_instruction = NULL;
1938
1939 while (1)
1940 {
1941 mnem_p = mnemonic;
1942 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1943 {
1944 mnem_p++;
1945 if (mnem_p >= mnemonic + MAX_MNEM_SIZE)
1946 {
1947 as_bad (_("no such instruction: `%s'"), token_start);
1948 return NULL;
1949 }
1950 l++;
1951 }
1952 if (!is_space_char (*l)
1953 && *l != END_OF_INSN
1954 && (intel_syntax
1955 || (*l != PREFIX_SEPARATOR
1956 && *l != ',')))
1957 {
1958 as_bad (_("invalid character %s in mnemonic"),
1959 output_invalid (*l));
1960 return NULL;
1961 }
1962 if (token_start == l)
1963 {
1964 if (!intel_syntax && *l == PREFIX_SEPARATOR)
1965 as_bad (_("expecting prefix; got nothing"));
1966 else
1967 as_bad (_("expecting mnemonic; got nothing"));
1968 return NULL;
1969 }
1970
1971 /* Look up instruction (or prefix) via hash table. */
1972 current_templates = hash_find (op_hash, mnemonic);
1973
1974 if (*l != END_OF_INSN
1975 && (!is_space_char (*l) || l[1] != END_OF_INSN)
1976 && current_templates
1977 && (current_templates->start->opcode_modifier & IsPrefix))
1978 {
1979 if (current_templates->start->cpu_flags
1980 & (flag_code != CODE_64BIT ? Cpu64 : CpuNo64))
1981 {
1982 as_bad ((flag_code != CODE_64BIT
1983 ? _("`%s' is only supported in 64-bit mode")
1984 : _("`%s' is not supported in 64-bit mode")),
1985 current_templates->start->name);
1986 return NULL;
1987 }
1988 /* If we are in 16-bit mode, do not allow addr16 or data16.
1989 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1990 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1991 && flag_code != CODE_64BIT
1992 && (((current_templates->start->opcode_modifier & Size32) != 0)
1993 ^ (flag_code == CODE_16BIT)))
1994 {
1995 as_bad (_("redundant %s prefix"),
1996 current_templates->start->name);
1997 return NULL;
1998 }
1999 /* Add prefix, checking for repeated prefixes. */
2000 switch (add_prefix (current_templates->start->base_opcode))
2001 {
2002 case 0:
2003 return NULL;
2004 case 2:
2005 expecting_string_instruction = current_templates->start->name;
2006 break;
2007 }
2008 /* Skip past PREFIX_SEPARATOR and reset token_start. */
2009 token_start = ++l;
2010 }
2011 else
2012 break;
2013 }
2014
2015 if (!current_templates)
2016 {
2017 /* See if we can get a match by trimming off a suffix. */
2018 switch (mnem_p[-1])
2019 {
2020 case WORD_MNEM_SUFFIX:
2021 if (intel_syntax && (intel_float_operand (mnemonic) & 2))
2022 i.suffix = SHORT_MNEM_SUFFIX;
2023 else
2024 case BYTE_MNEM_SUFFIX:
2025 case QWORD_MNEM_SUFFIX:
2026 i.suffix = mnem_p[-1];
2027 mnem_p[-1] = '\0';
2028 current_templates = hash_find (op_hash, mnemonic);
2029 break;
2030 case SHORT_MNEM_SUFFIX:
2031 case LONG_MNEM_SUFFIX:
2032 if (!intel_syntax)
2033 {
2034 i.suffix = mnem_p[-1];
2035 mnem_p[-1] = '\0';
2036 current_templates = hash_find (op_hash, mnemonic);
2037 }
2038 break;
2039
2040 /* Intel Syntax. */
2041 case 'd':
2042 if (intel_syntax)
2043 {
2044 if (intel_float_operand (mnemonic) == 1)
2045 i.suffix = SHORT_MNEM_SUFFIX;
2046 else
2047 i.suffix = LONG_MNEM_SUFFIX;
2048 mnem_p[-1] = '\0';
2049 current_templates = hash_find (op_hash, mnemonic);
2050 }
2051 break;
2052 }
2053 if (!current_templates)
2054 {
2055 as_bad (_("no such instruction: `%s'"), token_start);
2056 return NULL;
2057 }
2058 }
2059
2060 if (current_templates->start->opcode_modifier & (Jump | JumpByte))
2061 {
2062 /* Check for a branch hint. We allow ",pt" and ",pn" for
2063 predict taken and predict not taken respectively.
2064 I'm not sure that branch hints actually do anything on loop
2065 and jcxz insns (JumpByte) for current Pentium4 chips. They
2066 may work in the future and it doesn't hurt to accept them
2067 now. */
2068 if (l[0] == ',' && l[1] == 'p')
2069 {
2070 if (l[2] == 't')
2071 {
2072 if (!add_prefix (DS_PREFIX_OPCODE))
2073 return NULL;
2074 l += 3;
2075 }
2076 else if (l[2] == 'n')
2077 {
2078 if (!add_prefix (CS_PREFIX_OPCODE))
2079 return NULL;
2080 l += 3;
2081 }
2082 }
2083 }
2084 /* Any other comma loses. */
2085 if (*l == ',')
2086 {
2087 as_bad (_("invalid character %s in mnemonic"),
2088 output_invalid (*l));
2089 return NULL;
2090 }
2091
2092 /* Check if instruction is supported on specified architecture. */
2093 supported = 0;
2094 for (t = current_templates->start; t < current_templates->end; ++t)
2095 {
2096 if (!((t->cpu_flags & ~(Cpu64 | CpuNo64))
2097 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64))))
2098 supported |= 1;
2099 if (!(t->cpu_flags & (flag_code == CODE_64BIT ? CpuNo64 : Cpu64)))
2100 supported |= 2;
2101 }
2102 if (!(supported & 2))
2103 {
2104 as_bad (flag_code == CODE_64BIT
2105 ? _("`%s' is not supported in 64-bit mode")
2106 : _("`%s' is only supported in 64-bit mode"),
2107 current_templates->start->name);
2108 return NULL;
2109 }
2110 if (!(supported & 1))
2111 {
2112 as_warn (_("`%s' is not supported on `%s%s'"),
2113 current_templates->start->name,
2114 cpu_arch_name,
2115 cpu_sub_arch_name ? cpu_sub_arch_name : "");
2116 }
2117 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
2118 {
2119 as_warn (_("use .code16 to ensure correct addressing mode"));
2120 }
2121
2122 /* Check for rep/repne without a string instruction. */
2123 if (expecting_string_instruction)
2124 {
2125 static templates override;
2126
2127 for (t = current_templates->start; t < current_templates->end; ++t)
2128 if (t->opcode_modifier & IsString)
2129 break;
2130 if (t >= current_templates->end)
2131 {
2132 as_bad (_("expecting string instruction after `%s'"),
2133 expecting_string_instruction);
2134 return NULL;
2135 }
2136 for (override.start = t; t < current_templates->end; ++t)
2137 if (!(t->opcode_modifier & IsString))
2138 break;
2139 override.end = t;
2140 current_templates = &override;
2141 }
2142
2143 return l;
2144 }
2145
2146 static char *
2147 parse_operands (char *l, const char *mnemonic)
2148 {
2149 char *token_start;
2150
2151 /* 1 if operand is pending after ','. */
2152 unsigned int expecting_operand = 0;
2153
2154 /* Non-zero if operand parens not balanced. */
2155 unsigned int paren_not_balanced;
2156
2157 while (*l != END_OF_INSN)
2158 {
2159 /* Skip optional white space before operand. */
2160 if (is_space_char (*l))
2161 ++l;
2162 if (!is_operand_char (*l) && *l != END_OF_INSN)
2163 {
2164 as_bad (_("invalid character %s before operand %d"),
2165 output_invalid (*l),
2166 i.operands + 1);
2167 return NULL;
2168 }
2169 token_start = l; /* after white space */
2170 paren_not_balanced = 0;
2171 while (paren_not_balanced || *l != ',')
2172 {
2173 if (*l == END_OF_INSN)
2174 {
2175 if (paren_not_balanced)
2176 {
2177 if (!intel_syntax)
2178 as_bad (_("unbalanced parenthesis in operand %d."),
2179 i.operands + 1);
2180 else
2181 as_bad (_("unbalanced brackets in operand %d."),
2182 i.operands + 1);
2183 return NULL;
2184 }
2185 else
2186 break; /* we are done */
2187 }
2188 else if (!is_operand_char (*l) && !is_space_char (*l))
2189 {
2190 as_bad (_("invalid character %s in operand %d"),
2191 output_invalid (*l),
2192 i.operands + 1);
2193 return NULL;
2194 }
2195 if (!intel_syntax)
2196 {
2197 if (*l == '(')
2198 ++paren_not_balanced;
2199 if (*l == ')')
2200 --paren_not_balanced;
2201 }
2202 else
2203 {
2204 if (*l == '[')
2205 ++paren_not_balanced;
2206 if (*l == ']')
2207 --paren_not_balanced;
2208 }
2209 l++;
2210 }
2211 if (l != token_start)
2212 { /* Yes, we've read in another operand. */
2213 unsigned int operand_ok;
2214 this_operand = i.operands++;
2215 if (i.operands > MAX_OPERANDS)
2216 {
2217 as_bad (_("spurious operands; (%d operands/instruction max)"),
2218 MAX_OPERANDS);
2219 return NULL;
2220 }
2221 /* Now parse operand adding info to 'i' as we go along. */
2222 END_STRING_AND_SAVE (l);
2223
2224 if (intel_syntax)
2225 operand_ok =
2226 i386_intel_operand (token_start,
2227 intel_float_operand (mnemonic));
2228 else
2229 operand_ok = i386_operand (token_start);
2230
2231 RESTORE_END_STRING (l);
2232 if (!operand_ok)
2233 return NULL;
2234 }
2235 else
2236 {
2237 if (expecting_operand)
2238 {
2239 expecting_operand_after_comma:
2240 as_bad (_("expecting operand after ','; got nothing"));
2241 return NULL;
2242 }
2243 if (*l == ',')
2244 {
2245 as_bad (_("expecting operand before ','; got nothing"));
2246 return NULL;
2247 }
2248 }
2249
2250 /* Now *l must be either ',' or END_OF_INSN. */
2251 if (*l == ',')
2252 {
2253 if (*++l == END_OF_INSN)
2254 {
2255 /* Just skip it, if it's \n complain. */
2256 goto expecting_operand_after_comma;
2257 }
2258 expecting_operand = 1;
2259 }
2260 }
2261 return l;
2262 }
2263
2264 static void
2265 swap_2_operands (int xchg1, int xchg2)
2266 {
2267 union i386_op temp_op;
2268 unsigned int temp_type;
2269 enum bfd_reloc_code_real temp_reloc;
2270
2271 temp_type = i.types[xchg2];
2272 i.types[xchg2] = i.types[xchg1];
2273 i.types[xchg1] = temp_type;
2274 temp_op = i.op[xchg2];
2275 i.op[xchg2] = i.op[xchg1];
2276 i.op[xchg1] = temp_op;
2277 temp_reloc = i.reloc[xchg2];
2278 i.reloc[xchg2] = i.reloc[xchg1];
2279 i.reloc[xchg1] = temp_reloc;
2280 }
2281
2282 static void
2283 swap_operands (void)
2284 {
2285 switch (i.operands)
2286 {
2287 case 4:
2288 swap_2_operands (1, i.operands - 2);
2289 case 3:
2290 case 2:
2291 swap_2_operands (0, i.operands - 1);
2292 break;
2293 default:
2294 abort ();
2295 }
2296
2297 if (i.mem_operands == 2)
2298 {
2299 const seg_entry *temp_seg;
2300 temp_seg = i.seg[0];
2301 i.seg[0] = i.seg[1];
2302 i.seg[1] = temp_seg;
2303 }
2304 }
2305
2306 /* Try to ensure constant immediates are represented in the smallest
2307 opcode possible. */
2308 static void
2309 optimize_imm (void)
2310 {
2311 char guess_suffix = 0;
2312 int op;
2313
2314 if (i.suffix)
2315 guess_suffix = i.suffix;
2316 else if (i.reg_operands)
2317 {
2318 /* Figure out a suffix from the last register operand specified.
2319 We can't do this properly yet, ie. excluding InOutPortReg,
2320 but the following works for instructions with immediates.
2321 In any case, we can't set i.suffix yet. */
2322 for (op = i.operands; --op >= 0;)
2323 if (i.types[op] & Reg)
2324 {
2325 if (i.types[op] & Reg8)
2326 guess_suffix = BYTE_MNEM_SUFFIX;
2327 else if (i.types[op] & Reg16)
2328 guess_suffix = WORD_MNEM_SUFFIX;
2329 else if (i.types[op] & Reg32)
2330 guess_suffix = LONG_MNEM_SUFFIX;
2331 else if (i.types[op] & Reg64)
2332 guess_suffix = QWORD_MNEM_SUFFIX;
2333 break;
2334 }
2335 }
2336 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
2337 guess_suffix = WORD_MNEM_SUFFIX;
2338
2339 for (op = i.operands; --op >= 0;)
2340 if (i.types[op] & Imm)
2341 {
2342 switch (i.op[op].imms->X_op)
2343 {
2344 case O_constant:
2345 /* If a suffix is given, this operand may be shortened. */
2346 switch (guess_suffix)
2347 {
2348 case LONG_MNEM_SUFFIX:
2349 i.types[op] |= Imm32 | Imm64;
2350 break;
2351 case WORD_MNEM_SUFFIX:
2352 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
2353 break;
2354 case BYTE_MNEM_SUFFIX:
2355 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
2356 break;
2357 }
2358
2359 /* If this operand is at most 16 bits, convert it
2360 to a signed 16 bit number before trying to see
2361 whether it will fit in an even smaller size.
2362 This allows a 16-bit operand such as $0xffe0 to
2363 be recognised as within Imm8S range. */
2364 if ((i.types[op] & Imm16)
2365 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
2366 {
2367 i.op[op].imms->X_add_number =
2368 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
2369 }
2370 if ((i.types[op] & Imm32)
2371 && ((i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1))
2372 == 0))
2373 {
2374 i.op[op].imms->X_add_number = ((i.op[op].imms->X_add_number
2375 ^ ((offsetT) 1 << 31))
2376 - ((offsetT) 1 << 31));
2377 }
2378 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
2379
2380 /* We must avoid matching of Imm32 templates when 64bit
2381 only immediate is available. */
2382 if (guess_suffix == QWORD_MNEM_SUFFIX)
2383 i.types[op] &= ~Imm32;
2384 break;
2385
2386 case O_absent:
2387 case O_register:
2388 abort ();
2389
2390 /* Symbols and expressions. */
2391 default:
2392 /* Convert symbolic operand to proper sizes for matching, but don't
2393 prevent matching a set of insns that only supports sizes other
2394 than those matching the insn suffix. */
2395 {
2396 unsigned int mask, allowed = 0;
2397 const template *t;
2398
2399 for (t = current_templates->start;
2400 t < current_templates->end;
2401 ++t)
2402 allowed |= t->operand_types[op];
2403 switch (guess_suffix)
2404 {
2405 case QWORD_MNEM_SUFFIX:
2406 mask = Imm64 | Imm32S;
2407 break;
2408 case LONG_MNEM_SUFFIX:
2409 mask = Imm32;
2410 break;
2411 case WORD_MNEM_SUFFIX:
2412 mask = Imm16;
2413 break;
2414 case BYTE_MNEM_SUFFIX:
2415 mask = Imm8;
2416 break;
2417 default:
2418 mask = 0;
2419 break;
2420 }
2421 if (mask & allowed)
2422 i.types[op] &= mask;
2423 }
2424 break;
2425 }
2426 }
2427 }
2428
2429 /* Try to use the smallest displacement type too. */
2430 static void
2431 optimize_disp (void)
2432 {
2433 int op;
2434
2435 for (op = i.operands; --op >= 0;)
2436 if (i.types[op] & Disp)
2437 {
2438 if (i.op[op].disps->X_op == O_constant)
2439 {
2440 offsetT disp = i.op[op].disps->X_add_number;
2441
2442 if ((i.types[op] & Disp16)
2443 && (disp & ~(offsetT) 0xffff) == 0)
2444 {
2445 /* If this operand is at most 16 bits, convert
2446 to a signed 16 bit number and don't use 64bit
2447 displacement. */
2448 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
2449 i.types[op] &= ~Disp64;
2450 }
2451 if ((i.types[op] & Disp32)
2452 && (disp & ~(((offsetT) 2 << 31) - 1)) == 0)
2453 {
2454 /* If this operand is at most 32 bits, convert
2455 to a signed 32 bit number and don't use 64bit
2456 displacement. */
2457 disp &= (((offsetT) 2 << 31) - 1);
2458 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
2459 i.types[op] &= ~Disp64;
2460 }
2461 if (!disp && (i.types[op] & BaseIndex))
2462 {
2463 i.types[op] &= ~Disp;
2464 i.op[op].disps = 0;
2465 i.disp_operands--;
2466 }
2467 else if (flag_code == CODE_64BIT)
2468 {
2469 if (fits_in_signed_long (disp))
2470 {
2471 i.types[op] &= ~Disp64;
2472 i.types[op] |= Disp32S;
2473 }
2474 if (fits_in_unsigned_long (disp))
2475 i.types[op] |= Disp32;
2476 }
2477 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
2478 && fits_in_signed_byte (disp))
2479 i.types[op] |= Disp8;
2480 }
2481 else if (i.reloc[op] == BFD_RELOC_386_TLS_DESC_CALL
2482 || i.reloc[op] == BFD_RELOC_X86_64_TLSDESC_CALL)
2483 {
2484 fix_new_exp (frag_now, frag_more (0) - frag_now->fr_literal, 0,
2485 i.op[op].disps, 0, i.reloc[op]);
2486 i.types[op] &= ~Disp;
2487 }
2488 else
2489 /* We only support 64bit displacement on constants. */
2490 i.types[op] &= ~Disp64;
2491 }
2492 }
2493
2494 static int
2495 match_template (void)
2496 {
2497 /* Points to template once we've found it. */
2498 const template *t;
2499 unsigned int overlap0, overlap1, overlap2, overlap3;
2500 unsigned int found_reverse_match;
2501 int suffix_check;
2502 unsigned int operand_types [MAX_OPERANDS];
2503 int addr_prefix_disp;
2504 unsigned int j;
2505
2506 #if MAX_OPERANDS != 4
2507 # error "MAX_OPERANDS must be 4."
2508 #endif
2509
2510 #define MATCH(overlap, given, template) \
2511 ((overlap & ~JumpAbsolute) \
2512 && (((given) & (BaseIndex | JumpAbsolute)) \
2513 == ((overlap) & (BaseIndex | JumpAbsolute))))
2514
2515 /* If given types r0 and r1 are registers they must be of the same type
2516 unless the expected operand type register overlap is null.
2517 Note that Acc in a template matches every size of reg. */
2518 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
2519 (((g0) & Reg) == 0 || ((g1) & Reg) == 0 \
2520 || ((g0) & Reg) == ((g1) & Reg) \
2521 || ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
2522
2523 overlap0 = 0;
2524 overlap1 = 0;
2525 overlap2 = 0;
2526 overlap3 = 0;
2527 found_reverse_match = 0;
2528 for (j = 0; j < MAX_OPERANDS; j++)
2529 operand_types [j] = 0;
2530 addr_prefix_disp = -1;
2531 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
2532 ? No_bSuf
2533 : (i.suffix == WORD_MNEM_SUFFIX
2534 ? No_wSuf
2535 : (i.suffix == SHORT_MNEM_SUFFIX
2536 ? No_sSuf
2537 : (i.suffix == LONG_MNEM_SUFFIX
2538 ? No_lSuf
2539 : (i.suffix == QWORD_MNEM_SUFFIX
2540 ? No_qSuf
2541 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX
2542 ? No_xSuf : 0))))));
2543
2544 for (t = current_templates->start; t < current_templates->end; t++)
2545 {
2546 addr_prefix_disp = -1;
2547
2548 /* Must have right number of operands. */
2549 if (i.operands != t->operands)
2550 continue;
2551
2552 /* Check the suffix, except for some instructions in intel mode. */
2553 if ((t->opcode_modifier & suffix_check)
2554 && !(intel_syntax
2555 && (t->opcode_modifier & IgnoreSize)))
2556 continue;
2557
2558 for (j = 0; j < MAX_OPERANDS; j++)
2559 operand_types [j] = t->operand_types [j];
2560
2561 /* In general, don't allow 64-bit operands in 32-bit mode. */
2562 if (i.suffix == QWORD_MNEM_SUFFIX
2563 && flag_code != CODE_64BIT
2564 && (intel_syntax
2565 ? (!(t->opcode_modifier & IgnoreSize)
2566 && !intel_float_operand (t->name))
2567 : intel_float_operand (t->name) != 2)
2568 && (!(operand_types[0] & (RegMMX | RegXMM))
2569 || !(operand_types[t->operands > 1] & (RegMMX | RegXMM)))
2570 && (t->base_opcode != 0x0fc7
2571 || t->extension_opcode != 1 /* cmpxchg8b */))
2572 continue;
2573
2574 /* Do not verify operands when there are none. */
2575 else if (!t->operands)
2576 {
2577 if (t->cpu_flags & ~cpu_arch_flags)
2578 continue;
2579 /* We've found a match; break out of loop. */
2580 break;
2581 }
2582
2583 /* Address size prefix will turn Disp64/Disp32/Disp16 operand
2584 into Disp32/Disp16/Disp32 operand. */
2585 if (i.prefix[ADDR_PREFIX] != 0)
2586 {
2587 unsigned int DispOn = 0, DispOff = 0;
2588
2589 switch (flag_code)
2590 {
2591 case CODE_16BIT:
2592 DispOn = Disp32;
2593 DispOff = Disp16;
2594 break;
2595 case CODE_32BIT:
2596 DispOn = Disp16;
2597 DispOff = Disp32;
2598 break;
2599 case CODE_64BIT:
2600 DispOn = Disp32;
2601 DispOff = Disp64;
2602 break;
2603 }
2604
2605 for (j = 0; j < MAX_OPERANDS; j++)
2606 {
2607 /* There should be only one Disp operand. */
2608 if ((operand_types[j] & DispOff))
2609 {
2610 addr_prefix_disp = j;
2611 operand_types[j] |= DispOn;
2612 operand_types[j] &= ~DispOff;
2613 break;
2614 }
2615 }
2616 }
2617
2618 overlap0 = i.types[0] & operand_types[0];
2619 switch (t->operands)
2620 {
2621 case 1:
2622 if (!MATCH (overlap0, i.types[0], operand_types[0]))
2623 continue;
2624 break;
2625 case 2:
2626 case 3:
2627 case 4:
2628 overlap1 = i.types[1] & operand_types[1];
2629 if (!MATCH (overlap0, i.types[0], operand_types[0])
2630 || !MATCH (overlap1, i.types[1], operand_types[1])
2631 /* monitor in SSE3 is a very special case. The first
2632 register and the second register may have different
2633 sizes. */
2634 || !((t->base_opcode == 0x0f01
2635 && t->extension_opcode == 0xc8)
2636 || CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2637 operand_types[0],
2638 overlap1, i.types[1],
2639 operand_types[1])))
2640 {
2641 /* Check if other direction is valid ... */
2642 if ((t->opcode_modifier & (D | FloatD)) == 0)
2643 continue;
2644
2645 /* Try reversing direction of operands. */
2646 overlap0 = i.types[0] & operand_types[1];
2647 overlap1 = i.types[1] & operand_types[0];
2648 if (!MATCH (overlap0, i.types[0], operand_types[1])
2649 || !MATCH (overlap1, i.types[1], operand_types[0])
2650 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2651 operand_types[1],
2652 overlap1, i.types[1],
2653 operand_types[0]))
2654 {
2655 /* Does not match either direction. */
2656 continue;
2657 }
2658 /* found_reverse_match holds which of D or FloatDR
2659 we've found. */
2660 if ((t->opcode_modifier & D))
2661 found_reverse_match = Opcode_D;
2662 else if ((t->opcode_modifier & FloatD))
2663 found_reverse_match = Opcode_FloatD;
2664 else
2665 found_reverse_match = 0;
2666 if ((t->opcode_modifier & FloatR))
2667 found_reverse_match |= Opcode_FloatR;
2668 }
2669 else
2670 {
2671 /* Found a forward 2 operand match here. */
2672 switch (t->operands)
2673 {
2674 case 4:
2675 overlap3 = i.types[3] & operand_types[3];
2676 case 3:
2677 overlap2 = i.types[2] & operand_types[2];
2678 break;
2679 }
2680
2681 switch (t->operands)
2682 {
2683 case 4:
2684 if (!MATCH (overlap3, i.types[3], operand_types[3])
2685 || !CONSISTENT_REGISTER_MATCH (overlap2,
2686 i.types[2],
2687 operand_types[2],
2688 overlap3,
2689 i.types[3],
2690 operand_types[3]))
2691 continue;
2692 case 3:
2693 /* Here we make use of the fact that there are no
2694 reverse match 3 operand instructions, and all 3
2695 operand instructions only need to be checked for
2696 register consistency between operands 2 and 3. */
2697 if (!MATCH (overlap2, i.types[2], operand_types[2])
2698 || !CONSISTENT_REGISTER_MATCH (overlap1,
2699 i.types[1],
2700 operand_types[1],
2701 overlap2,
2702 i.types[2],
2703 operand_types[2]))
2704 continue;
2705 break;
2706 }
2707 }
2708 /* Found either forward/reverse 2, 3 or 4 operand match here:
2709 slip through to break. */
2710 }
2711 if (t->cpu_flags & ~cpu_arch_flags)
2712 {
2713 found_reverse_match = 0;
2714 continue;
2715 }
2716 /* We've found a match; break out of loop. */
2717 break;
2718 }
2719
2720 if (t == current_templates->end)
2721 {
2722 /* We found no match. */
2723 as_bad (_("suffix or operands invalid for `%s'"),
2724 current_templates->start->name);
2725 return 0;
2726 }
2727
2728 if (!quiet_warnings)
2729 {
2730 if (!intel_syntax
2731 && ((i.types[0] & JumpAbsolute)
2732 != (operand_types[0] & JumpAbsolute)))
2733 {
2734 as_warn (_("indirect %s without `*'"), t->name);
2735 }
2736
2737 if ((t->opcode_modifier & (IsPrefix | IgnoreSize))
2738 == (IsPrefix | IgnoreSize))
2739 {
2740 /* Warn them that a data or address size prefix doesn't
2741 affect assembly of the next line of code. */
2742 as_warn (_("stand-alone `%s' prefix"), t->name);
2743 }
2744 }
2745
2746 /* Copy the template we found. */
2747 i.tm = *t;
2748
2749 if (addr_prefix_disp != -1)
2750 i.tm.operand_types[addr_prefix_disp]
2751 = operand_types[addr_prefix_disp];
2752
2753 if (found_reverse_match)
2754 {
2755 /* If we found a reverse match we must alter the opcode
2756 direction bit. found_reverse_match holds bits to change
2757 (different for int & float insns). */
2758
2759 i.tm.base_opcode ^= found_reverse_match;
2760
2761 i.tm.operand_types[0] = operand_types[1];
2762 i.tm.operand_types[1] = operand_types[0];
2763 }
2764
2765 return 1;
2766 }
2767
2768 static int
2769 check_string (void)
2770 {
2771 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
2772 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
2773 {
2774 if (i.seg[0] != NULL && i.seg[0] != &es)
2775 {
2776 as_bad (_("`%s' operand %d must use `%%es' segment"),
2777 i.tm.name,
2778 mem_op + 1);
2779 return 0;
2780 }
2781 /* There's only ever one segment override allowed per instruction.
2782 This instruction possibly has a legal segment override on the
2783 second operand, so copy the segment to where non-string
2784 instructions store it, allowing common code. */
2785 i.seg[0] = i.seg[1];
2786 }
2787 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
2788 {
2789 if (i.seg[1] != NULL && i.seg[1] != &es)
2790 {
2791 as_bad (_("`%s' operand %d must use `%%es' segment"),
2792 i.tm.name,
2793 mem_op + 2);
2794 return 0;
2795 }
2796 }
2797 return 1;
2798 }
2799
2800 static int
2801 process_suffix (void)
2802 {
2803 /* If matched instruction specifies an explicit instruction mnemonic
2804 suffix, use it. */
2805 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
2806 {
2807 if (i.tm.opcode_modifier & Size16)
2808 i.suffix = WORD_MNEM_SUFFIX;
2809 else if (i.tm.opcode_modifier & Size64)
2810 i.suffix = QWORD_MNEM_SUFFIX;
2811 else
2812 i.suffix = LONG_MNEM_SUFFIX;
2813 }
2814 else if (i.reg_operands)
2815 {
2816 /* If there's no instruction mnemonic suffix we try to invent one
2817 based on register operands. */
2818 if (!i.suffix)
2819 {
2820 /* We take i.suffix from the last register operand specified,
2821 Destination register type is more significant than source
2822 register type. */
2823 int op;
2824
2825 for (op = i.operands; --op >= 0;)
2826 if ((i.types[op] & Reg)
2827 && !(i.tm.operand_types[op] & InOutPortReg))
2828 {
2829 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
2830 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
2831 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
2832 LONG_MNEM_SUFFIX);
2833 break;
2834 }
2835 }
2836 else if (i.suffix == BYTE_MNEM_SUFFIX)
2837 {
2838 if (!check_byte_reg ())
2839 return 0;
2840 }
2841 else if (i.suffix == LONG_MNEM_SUFFIX)
2842 {
2843 if (!check_long_reg ())
2844 return 0;
2845 }
2846 else if (i.suffix == QWORD_MNEM_SUFFIX)
2847 {
2848 if (!check_qword_reg ())
2849 return 0;
2850 }
2851 else if (i.suffix == WORD_MNEM_SUFFIX)
2852 {
2853 if (!check_word_reg ())
2854 return 0;
2855 }
2856 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2857 /* Do nothing if the instruction is going to ignore the prefix. */
2858 ;
2859 else
2860 abort ();
2861 }
2862 else if ((i.tm.opcode_modifier & DefaultSize)
2863 && !i.suffix
2864 /* exclude fldenv/frstor/fsave/fstenv */
2865 && (i.tm.opcode_modifier & No_sSuf))
2866 {
2867 i.suffix = stackop_size;
2868 }
2869 else if (intel_syntax
2870 && !i.suffix
2871 && ((i.tm.operand_types[0] & JumpAbsolute)
2872 || (i.tm.opcode_modifier & (JumpByte|JumpInterSegment))
2873 || (i.tm.base_opcode == 0x0f01 /* [ls][gi]dt */
2874 && i.tm.extension_opcode <= 3)))
2875 {
2876 switch (flag_code)
2877 {
2878 case CODE_64BIT:
2879 if (!(i.tm.opcode_modifier & No_qSuf))
2880 {
2881 i.suffix = QWORD_MNEM_SUFFIX;
2882 break;
2883 }
2884 case CODE_32BIT:
2885 if (!(i.tm.opcode_modifier & No_lSuf))
2886 i.suffix = LONG_MNEM_SUFFIX;
2887 break;
2888 case CODE_16BIT:
2889 if (!(i.tm.opcode_modifier & No_wSuf))
2890 i.suffix = WORD_MNEM_SUFFIX;
2891 break;
2892 }
2893 }
2894
2895 if (!i.suffix)
2896 {
2897 if (!intel_syntax)
2898 {
2899 if (i.tm.opcode_modifier & W)
2900 {
2901 as_bad (_("no instruction mnemonic suffix given and "
2902 "no register operands; can't size instruction"));
2903 return 0;
2904 }
2905 }
2906 else
2907 {
2908 unsigned int suffixes = (~i.tm.opcode_modifier
2909 & (No_bSuf
2910 | No_wSuf
2911 | No_lSuf
2912 | No_sSuf
2913 | No_xSuf
2914 | No_qSuf));
2915
2916 if ((i.tm.opcode_modifier & W)
2917 || ((suffixes & (suffixes - 1))
2918 && !(i.tm.opcode_modifier & (DefaultSize | IgnoreSize))))
2919 {
2920 as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
2921 return 0;
2922 }
2923 }
2924 }
2925
2926 /* Change the opcode based on the operand size given by i.suffix;
2927 We don't need to change things for byte insns. */
2928
2929 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2930 {
2931 /* It's not a byte, select word/dword operation. */
2932 if (i.tm.opcode_modifier & W)
2933 {
2934 if (i.tm.opcode_modifier & ShortForm)
2935 i.tm.base_opcode |= 8;
2936 else
2937 i.tm.base_opcode |= 1;
2938 }
2939
2940 /* Now select between word & dword operations via the operand
2941 size prefix, except for instructions that will ignore this
2942 prefix anyway. */
2943 if (i.tm.base_opcode == 0x0f01 && i.tm.extension_opcode == 0xc8)
2944 {
2945 /* monitor in SSE3 is a very special case. The default size
2946 of AX is the size of mode. The address size override
2947 prefix will change the size of AX. */
2948 if (i.op->regs[0].reg_type &
2949 (flag_code == CODE_32BIT ? Reg16 : Reg32))
2950 if (!add_prefix (ADDR_PREFIX_OPCODE))
2951 return 0;
2952 }
2953 else if (i.suffix != QWORD_MNEM_SUFFIX
2954 && i.suffix != LONG_DOUBLE_MNEM_SUFFIX
2955 && !(i.tm.opcode_modifier & (IgnoreSize | FloatMF))
2956 && ((i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2957 || (flag_code == CODE_64BIT
2958 && (i.tm.opcode_modifier & JumpByte))))
2959 {
2960 unsigned int prefix = DATA_PREFIX_OPCODE;
2961
2962 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2963 prefix = ADDR_PREFIX_OPCODE;
2964
2965 if (!add_prefix (prefix))
2966 return 0;
2967 }
2968
2969 /* Set mode64 for an operand. */
2970 if (i.suffix == QWORD_MNEM_SUFFIX
2971 && flag_code == CODE_64BIT
2972 && (i.tm.opcode_modifier & NoRex64) == 0)
2973 {
2974 /* Special case for xchg %rax,%rax. It is NOP and doesn't
2975 need rex64. */
2976 if (i.operands != 2
2977 || i.types [0] != (Acc | Reg64)
2978 || i.types [1] != (Acc | Reg64)
2979 || strcmp (i.tm.name, "xchg") != 0)
2980 i.rex |= REX_MODE64;
2981 }
2982
2983 /* Size floating point instruction. */
2984 if (i.suffix == LONG_MNEM_SUFFIX)
2985 if (i.tm.opcode_modifier & FloatMF)
2986 i.tm.base_opcode ^= 4;
2987 }
2988
2989 return 1;
2990 }
2991
2992 static int
2993 check_byte_reg (void)
2994 {
2995 int op;
2996
2997 for (op = i.operands; --op >= 0;)
2998 {
2999 /* If this is an eight bit register, it's OK. If it's the 16 or
3000 32 bit version of an eight bit register, we will just use the
3001 low portion, and that's OK too. */
3002 if (i.types[op] & Reg8)
3003 continue;
3004
3005 /* movzx and movsx should not generate this warning. */
3006 if (intel_syntax
3007 && (i.tm.base_opcode == 0xfb7
3008 || i.tm.base_opcode == 0xfb6
3009 || i.tm.base_opcode == 0x63
3010 || i.tm.base_opcode == 0xfbe
3011 || i.tm.base_opcode == 0xfbf))
3012 continue;
3013
3014 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4)
3015 {
3016 /* Prohibit these changes in the 64bit mode, since the
3017 lowering is more complicated. */
3018 if (flag_code == CODE_64BIT
3019 && (i.tm.operand_types[op] & InOutPortReg) == 0)
3020 {
3021 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3022 register_prefix, i.op[op].regs->reg_name,
3023 i.suffix);
3024 return 0;
3025 }
3026 #if REGISTER_WARNINGS
3027 if (!quiet_warnings
3028 && (i.tm.operand_types[op] & InOutPortReg) == 0)
3029 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
3030 (i.op[op].regs + (i.types[op] & Reg16
3031 ? REGNAM_AL - REGNAM_AX
3032 : REGNAM_AL - REGNAM_EAX))->reg_name,
3033 i.op[op].regs->reg_name,
3034 i.suffix);
3035 #endif
3036 continue;
3037 }
3038 /* Any other register is bad. */
3039 if (i.types[op] & (Reg | RegMMX | RegXMM
3040 | SReg2 | SReg3
3041 | Control | Debug | Test
3042 | FloatReg | FloatAcc))
3043 {
3044 as_bad (_("`%%%s' not allowed with `%s%c'"),
3045 i.op[op].regs->reg_name,
3046 i.tm.name,
3047 i.suffix);
3048 return 0;
3049 }
3050 }
3051 return 1;
3052 }
3053
3054 static int
3055 check_long_reg (void)
3056 {
3057 int op;
3058
3059 for (op = i.operands; --op >= 0;)
3060 /* Reject eight bit registers, except where the template requires
3061 them. (eg. movzb) */
3062 if ((i.types[op] & Reg8) != 0
3063 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3064 {
3065 as_bad (_("`%%%s' not allowed with `%s%c'"),
3066 i.op[op].regs->reg_name,
3067 i.tm.name,
3068 i.suffix);
3069 return 0;
3070 }
3071 /* Warn if the e prefix on a general reg is missing. */
3072 else if ((!quiet_warnings || flag_code == CODE_64BIT)
3073 && (i.types[op] & Reg16) != 0
3074 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3075 {
3076 /* Prohibit these changes in the 64bit mode, since the
3077 lowering is more complicated. */
3078 if (flag_code == CODE_64BIT)
3079 {
3080 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3081 register_prefix, i.op[op].regs->reg_name,
3082 i.suffix);
3083 return 0;
3084 }
3085 #if REGISTER_WARNINGS
3086 else
3087 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
3088 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
3089 i.op[op].regs->reg_name,
3090 i.suffix);
3091 #endif
3092 }
3093 /* Warn if the r prefix on a general reg is missing. */
3094 else if ((i.types[op] & Reg64) != 0
3095 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3096 {
3097 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3098 register_prefix, i.op[op].regs->reg_name,
3099 i.suffix);
3100 return 0;
3101 }
3102 return 1;
3103 }
3104
3105 static int
3106 check_qword_reg (void)
3107 {
3108 int op;
3109
3110 for (op = i.operands; --op >= 0; )
3111 /* Reject eight bit registers, except where the template requires
3112 them. (eg. movzb) */
3113 if ((i.types[op] & Reg8) != 0
3114 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3115 {
3116 as_bad (_("`%%%s' not allowed with `%s%c'"),
3117 i.op[op].regs->reg_name,
3118 i.tm.name,
3119 i.suffix);
3120 return 0;
3121 }
3122 /* Warn if the e prefix on a general reg is missing. */
3123 else if (((i.types[op] & Reg16) != 0
3124 || (i.types[op] & Reg32) != 0)
3125 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3126 {
3127 /* Prohibit these changes in the 64bit mode, since the
3128 lowering is more complicated. */
3129 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3130 register_prefix, i.op[op].regs->reg_name,
3131 i.suffix);
3132 return 0;
3133 }
3134 return 1;
3135 }
3136
3137 static int
3138 check_word_reg (void)
3139 {
3140 int op;
3141 for (op = i.operands; --op >= 0;)
3142 /* Reject eight bit registers, except where the template requires
3143 them. (eg. movzb) */
3144 if ((i.types[op] & Reg8) != 0
3145 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3146 {
3147 as_bad (_("`%%%s' not allowed with `%s%c'"),
3148 i.op[op].regs->reg_name,
3149 i.tm.name,
3150 i.suffix);
3151 return 0;
3152 }
3153 /* Warn if the e prefix on a general reg is present. */
3154 else if ((!quiet_warnings || flag_code == CODE_64BIT)
3155 && (i.types[op] & Reg32) != 0
3156 && (i.tm.operand_types[op] & (Reg16 | Acc)) != 0)
3157 {
3158 /* Prohibit these changes in the 64bit mode, since the
3159 lowering is more complicated. */
3160 if (flag_code == CODE_64BIT)
3161 {
3162 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3163 register_prefix, i.op[op].regs->reg_name,
3164 i.suffix);
3165 return 0;
3166 }
3167 else
3168 #if REGISTER_WARNINGS
3169 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
3170 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
3171 i.op[op].regs->reg_name,
3172 i.suffix);
3173 #endif
3174 }
3175 return 1;
3176 }
3177
3178 static int
3179 finalize_imm (void)
3180 {
3181 unsigned int overlap0, overlap1, overlap2;
3182
3183 overlap0 = i.types[0] & i.tm.operand_types[0];
3184 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64))
3185 && overlap0 != Imm8 && overlap0 != Imm8S
3186 && overlap0 != Imm16 && overlap0 != Imm32S
3187 && overlap0 != Imm32 && overlap0 != Imm64)
3188 {
3189 if (i.suffix)
3190 {
3191 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX
3192 ? Imm8 | Imm8S
3193 : (i.suffix == WORD_MNEM_SUFFIX
3194 ? Imm16
3195 : (i.suffix == QWORD_MNEM_SUFFIX
3196 ? Imm64 | Imm32S
3197 : Imm32)));
3198 }
3199 else if (overlap0 == (Imm16 | Imm32S | Imm32)
3200 || overlap0 == (Imm16 | Imm32)
3201 || overlap0 == (Imm16 | Imm32S))
3202 {
3203 overlap0 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
3204 ? Imm16 : Imm32S);
3205 }
3206 if (overlap0 != Imm8 && overlap0 != Imm8S
3207 && overlap0 != Imm16 && overlap0 != Imm32S
3208 && overlap0 != Imm32 && overlap0 != Imm64)
3209 {
3210 as_bad (_("no instruction mnemonic suffix given; "
3211 "can't determine immediate size"));
3212 return 0;
3213 }
3214 }
3215 i.types[0] = overlap0;
3216
3217 overlap1 = i.types[1] & i.tm.operand_types[1];
3218 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32 | Imm64))
3219 && overlap1 != Imm8 && overlap1 != Imm8S
3220 && overlap1 != Imm16 && overlap1 != Imm32S
3221 && overlap1 != Imm32 && overlap1 != Imm64)
3222 {
3223 if (i.suffix)
3224 {
3225 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX
3226 ? Imm8 | Imm8S
3227 : (i.suffix == WORD_MNEM_SUFFIX
3228 ? Imm16
3229 : (i.suffix == QWORD_MNEM_SUFFIX
3230 ? Imm64 | Imm32S
3231 : Imm32)));
3232 }
3233 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
3234 || overlap1 == (Imm16 | Imm32)
3235 || overlap1 == (Imm16 | Imm32S))
3236 {
3237 overlap1 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
3238 ? Imm16 : Imm32S);
3239 }
3240 if (overlap1 != Imm8 && overlap1 != Imm8S
3241 && overlap1 != Imm16 && overlap1 != Imm32S
3242 && overlap1 != Imm32 && overlap1 != Imm64)
3243 {
3244 as_bad (_("no instruction mnemonic suffix given; "
3245 "can't determine immediate size %x %c"),
3246 overlap1, i.suffix);
3247 return 0;
3248 }
3249 }
3250 i.types[1] = overlap1;
3251
3252 overlap2 = i.types[2] & i.tm.operand_types[2];
3253 assert ((overlap2 & Imm) == 0);
3254 i.types[2] = overlap2;
3255
3256 return 1;
3257 }
3258
3259 static int
3260 process_operands (void)
3261 {
3262 /* Default segment register this instruction will use for memory
3263 accesses. 0 means unknown. This is only for optimizing out
3264 unnecessary segment overrides. */
3265 const seg_entry *default_seg = 0;
3266
3267 /* The imul $imm, %reg instruction is converted into
3268 imul $imm, %reg, %reg, and the clr %reg instruction
3269 is converted into xor %reg, %reg. */
3270 if (i.tm.opcode_modifier & regKludge)
3271 {
3272 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
3273 /* Pretend we saw the extra register operand. */
3274 assert (i.reg_operands == 1
3275 && i.op[first_reg_op + 1].regs == 0);
3276 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
3277 i.types[first_reg_op + 1] = i.types[first_reg_op];
3278 i.operands++;
3279 i.reg_operands++;
3280 }
3281
3282 if (i.tm.opcode_modifier & ShortForm)
3283 {
3284 if (i.types[0] & (SReg2 | SReg3))
3285 {
3286 if (i.tm.base_opcode == POP_SEG_SHORT
3287 && i.op[0].regs->reg_num == 1)
3288 {
3289 as_bad (_("you can't `pop %%cs'"));
3290 return 0;
3291 }
3292 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
3293 if ((i.op[0].regs->reg_flags & RegRex) != 0)
3294 i.rex |= REX_EXTZ;
3295 }
3296 else
3297 {
3298 /* The register or float register operand is in operand 0 or 1. */
3299 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
3300 /* Register goes in low 3 bits of opcode. */
3301 i.tm.base_opcode |= i.op[op].regs->reg_num;
3302 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3303 i.rex |= REX_EXTZ;
3304 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
3305 {
3306 /* Warn about some common errors, but press on regardless.
3307 The first case can be generated by gcc (<= 2.8.1). */
3308 if (i.operands == 2)
3309 {
3310 /* Reversed arguments on faddp, fsubp, etc. */
3311 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
3312 i.op[1].regs->reg_name,
3313 i.op[0].regs->reg_name);
3314 }
3315 else
3316 {
3317 /* Extraneous `l' suffix on fp insn. */
3318 as_warn (_("translating to `%s %%%s'"), i.tm.name,
3319 i.op[0].regs->reg_name);
3320 }
3321 }
3322 }
3323 }
3324 else if (i.tm.opcode_modifier & Modrm)
3325 {
3326 /* The opcode is completed (modulo i.tm.extension_opcode which
3327 must be put into the modrm byte). Now, we make the modrm and
3328 index base bytes based on all the info we've collected. */
3329
3330 default_seg = build_modrm_byte ();
3331 }
3332 else if ((i.tm.base_opcode & ~0x3) == MOV_AX_DISP32)
3333 {
3334 default_seg = &ds;
3335 }
3336 else if ((i.tm.opcode_modifier & IsString) != 0)
3337 {
3338 /* For the string instructions that allow a segment override
3339 on one of their operands, the default segment is ds. */
3340 default_seg = &ds;
3341 }
3342
3343 if ((i.tm.base_opcode == 0x8d /* lea */
3344 || (i.tm.cpu_flags & CpuSVME))
3345 && i.seg[0] && !quiet_warnings)
3346 as_warn (_("segment override on `%s' is ineffectual"), i.tm.name);
3347
3348 /* If a segment was explicitly specified, and the specified segment
3349 is not the default, use an opcode prefix to select it. If we
3350 never figured out what the default segment is, then default_seg
3351 will be zero at this point, and the specified segment prefix will
3352 always be used. */
3353 if ((i.seg[0]) && (i.seg[0] != default_seg))
3354 {
3355 if (!add_prefix (i.seg[0]->seg_prefix))
3356 return 0;
3357 }
3358 return 1;
3359 }
3360
3361 static const seg_entry *
3362 build_modrm_byte (void)
3363 {
3364 const seg_entry *default_seg = 0;
3365
3366 /* i.reg_operands MUST be the number of real register operands;
3367 implicit registers do not count. */
3368 if (i.reg_operands == 2)
3369 {
3370 unsigned int source, dest;
3371
3372 switch (i.operands)
3373 {
3374 case 2:
3375 source = 0;
3376 break;
3377 case 3:
3378 /* When there are 3 operands, one of them may be immediate,
3379 which may be the first or the last operand. Otherwise,
3380 the first operand must be shift count register (cl). */
3381 assert (i.imm_operands == 1
3382 || (i.imm_operands == 0
3383 && (i.types[0] & ShiftCount)));
3384 source = (i.types[0] & (Imm | ShiftCount)) ? 1 : 0;
3385 break;
3386 case 4:
3387 /* When there are 4 operands, the first two must be immediate
3388 operands. The source operand will be the 3rd one. */
3389 assert (i.imm_operands == 2
3390 && (i.types[0] & Imm)
3391 && (i.types[1] & Imm));
3392 source = 2;
3393 break;
3394 default:
3395 abort ();
3396 }
3397
3398 dest = source + 1;
3399
3400 i.rm.mode = 3;
3401 /* One of the register operands will be encoded in the i.tm.reg
3402 field, the other in the combined i.tm.mode and i.tm.regmem
3403 fields. If no form of this instruction supports a memory
3404 destination operand, then we assume the source operand may
3405 sometimes be a memory operand and so we need to store the
3406 destination in the i.rm.reg field. */
3407 if ((i.tm.operand_types[dest] & AnyMem) == 0)
3408 {
3409 i.rm.reg = i.op[dest].regs->reg_num;
3410 i.rm.regmem = i.op[source].regs->reg_num;
3411 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
3412 i.rex |= REX_EXTX;
3413 if ((i.op[source].regs->reg_flags & RegRex) != 0)
3414 i.rex |= REX_EXTZ;
3415 }
3416 else
3417 {
3418 i.rm.reg = i.op[source].regs->reg_num;
3419 i.rm.regmem = i.op[dest].regs->reg_num;
3420 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
3421 i.rex |= REX_EXTZ;
3422 if ((i.op[source].regs->reg_flags & RegRex) != 0)
3423 i.rex |= REX_EXTX;
3424 }
3425 if (flag_code != CODE_64BIT && (i.rex & (REX_EXTX | REX_EXTZ)))
3426 {
3427 if (!((i.types[0] | i.types[1]) & Control))
3428 abort ();
3429 i.rex &= ~(REX_EXTX | REX_EXTZ);
3430 add_prefix (LOCK_PREFIX_OPCODE);
3431 }
3432 }
3433 else
3434 { /* If it's not 2 reg operands... */
3435 if (i.mem_operands)
3436 {
3437 unsigned int fake_zero_displacement = 0;
3438 unsigned int op;
3439
3440 for (op = 0; op < i.operands; op++)
3441 if ((i.types[op] & AnyMem))
3442 break;
3443 assert (op < i.operands);
3444
3445 default_seg = &ds;
3446
3447 if (i.base_reg == 0)
3448 {
3449 i.rm.mode = 0;
3450 if (!i.disp_operands)
3451 fake_zero_displacement = 1;
3452 if (i.index_reg == 0)
3453 {
3454 /* Operand is just <disp> */
3455 if (flag_code == CODE_64BIT)
3456 {
3457 /* 64bit mode overwrites the 32bit absolute
3458 addressing by RIP relative addressing and
3459 absolute addressing is encoded by one of the
3460 redundant SIB forms. */
3461 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3462 i.sib.base = NO_BASE_REGISTER;
3463 i.sib.index = NO_INDEX_REGISTER;
3464 i.types[op] = ((i.prefix[ADDR_PREFIX] == 0)
3465 ? Disp32S : Disp32);
3466 }
3467 else if ((flag_code == CODE_16BIT)
3468 ^ (i.prefix[ADDR_PREFIX] != 0))
3469 {
3470 i.rm.regmem = NO_BASE_REGISTER_16;
3471 i.types[op] = Disp16;
3472 }
3473 else
3474 {
3475 i.rm.regmem = NO_BASE_REGISTER;
3476 i.types[op] = Disp32;
3477 }
3478 }
3479 else /* !i.base_reg && i.index_reg */
3480 {
3481 i.sib.index = i.index_reg->reg_num;
3482 i.sib.base = NO_BASE_REGISTER;
3483 i.sib.scale = i.log2_scale_factor;
3484 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3485 i.types[op] &= ~Disp;
3486 if (flag_code != CODE_64BIT)
3487 i.types[op] |= Disp32; /* Must be 32 bit */
3488 else
3489 i.types[op] |= Disp32S;
3490 if ((i.index_reg->reg_flags & RegRex) != 0)
3491 i.rex |= REX_EXTY;
3492 }
3493 }
3494 /* RIP addressing for 64bit mode. */
3495 else if (i.base_reg->reg_type == BaseIndex)
3496 {
3497 i.rm.regmem = NO_BASE_REGISTER;
3498 i.types[op] &= ~ Disp;
3499 i.types[op] |= Disp32S;
3500 i.flags[op] |= Operand_PCrel;
3501 if (! i.disp_operands)
3502 fake_zero_displacement = 1;
3503 }
3504 else if (i.base_reg->reg_type & Reg16)
3505 {
3506 switch (i.base_reg->reg_num)
3507 {
3508 case 3: /* (%bx) */
3509 if (i.index_reg == 0)
3510 i.rm.regmem = 7;
3511 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
3512 i.rm.regmem = i.index_reg->reg_num - 6;
3513 break;
3514 case 5: /* (%bp) */
3515 default_seg = &ss;
3516 if (i.index_reg == 0)
3517 {
3518 i.rm.regmem = 6;
3519 if ((i.types[op] & Disp) == 0)
3520 {
3521 /* fake (%bp) into 0(%bp) */
3522 i.types[op] |= Disp8;
3523 fake_zero_displacement = 1;
3524 }
3525 }
3526 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
3527 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
3528 break;
3529 default: /* (%si) -> 4 or (%di) -> 5 */
3530 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
3531 }
3532 i.rm.mode = mode_from_disp_size (i.types[op]);
3533 }
3534 else /* i.base_reg and 32/64 bit mode */
3535 {
3536 if (flag_code == CODE_64BIT
3537 && (i.types[op] & Disp))
3538 i.types[op] = ((i.types[op] & Disp8)
3539 | (i.prefix[ADDR_PREFIX] == 0
3540 ? Disp32S : Disp32));
3541
3542 i.rm.regmem = i.base_reg->reg_num;
3543 if ((i.base_reg->reg_flags & RegRex) != 0)
3544 i.rex |= REX_EXTZ;
3545 i.sib.base = i.base_reg->reg_num;
3546 /* x86-64 ignores REX prefix bit here to avoid decoder
3547 complications. */
3548 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
3549 {
3550 default_seg = &ss;
3551 if (i.disp_operands == 0)
3552 {
3553 fake_zero_displacement = 1;
3554 i.types[op] |= Disp8;
3555 }
3556 }
3557 else if (i.base_reg->reg_num == ESP_REG_NUM)
3558 {
3559 default_seg = &ss;
3560 }
3561 i.sib.scale = i.log2_scale_factor;
3562 if (i.index_reg == 0)
3563 {
3564 /* <disp>(%esp) becomes two byte modrm with no index
3565 register. We've already stored the code for esp
3566 in i.rm.regmem ie. ESCAPE_TO_TWO_BYTE_ADDRESSING.
3567 Any base register besides %esp will not use the
3568 extra modrm byte. */
3569 i.sib.index = NO_INDEX_REGISTER;
3570 #if !SCALE1_WHEN_NO_INDEX
3571 /* Another case where we force the second modrm byte. */
3572 if (i.log2_scale_factor)
3573 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3574 #endif
3575 }
3576 else
3577 {
3578 i.sib.index = i.index_reg->reg_num;
3579 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3580 if ((i.index_reg->reg_flags & RegRex) != 0)
3581 i.rex |= REX_EXTY;
3582 }
3583
3584 if (i.disp_operands
3585 && (i.reloc[op] == BFD_RELOC_386_TLS_DESC_CALL
3586 || i.reloc[op] == BFD_RELOC_X86_64_TLSDESC_CALL))
3587 i.rm.mode = 0;
3588 else
3589 i.rm.mode = mode_from_disp_size (i.types[op]);
3590 }
3591
3592 if (fake_zero_displacement)
3593 {
3594 /* Fakes a zero displacement assuming that i.types[op]
3595 holds the correct displacement size. */
3596 expressionS *exp;
3597
3598 assert (i.op[op].disps == 0);
3599 exp = &disp_expressions[i.disp_operands++];
3600 i.op[op].disps = exp;
3601 exp->X_op = O_constant;
3602 exp->X_add_number = 0;
3603 exp->X_add_symbol = (symbolS *) 0;
3604 exp->X_op_symbol = (symbolS *) 0;
3605 }
3606 }
3607
3608 /* Fill in i.rm.reg or i.rm.regmem field with register operand
3609 (if any) based on i.tm.extension_opcode. Again, we must be
3610 careful to make sure that segment/control/debug/test/MMX
3611 registers are coded into the i.rm.reg field. */
3612 if (i.reg_operands)
3613 {
3614 unsigned int op;
3615
3616 for (op = 0; op < i.operands; op++)
3617 if ((i.types[op] & (Reg | RegMMX | RegXMM
3618 | SReg2 | SReg3
3619 | Control | Debug | Test)))
3620 break;
3621 assert (op < i.operands);
3622
3623 /* If there is an extension opcode to put here, the register
3624 number must be put into the regmem field. */
3625 if (i.tm.extension_opcode != None)
3626 {
3627 i.rm.regmem = i.op[op].regs->reg_num;
3628 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3629 i.rex |= REX_EXTZ;
3630 }
3631 else
3632 {
3633 i.rm.reg = i.op[op].regs->reg_num;
3634 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3635 i.rex |= REX_EXTX;
3636 }
3637
3638 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2 we
3639 must set it to 3 to indicate this is a register operand
3640 in the regmem field. */
3641 if (!i.mem_operands)
3642 i.rm.mode = 3;
3643 }
3644
3645 /* Fill in i.rm.reg field with extension opcode (if any). */
3646 if (i.tm.extension_opcode != None)
3647 i.rm.reg = i.tm.extension_opcode;
3648 }
3649 return default_seg;
3650 }
3651
3652 static void
3653 output_branch (void)
3654 {
3655 char *p;
3656 int code16;
3657 int prefix;
3658 relax_substateT subtype;
3659 symbolS *sym;
3660 offsetT off;
3661
3662 code16 = 0;
3663 if (flag_code == CODE_16BIT)
3664 code16 = CODE16;
3665
3666 prefix = 0;
3667 if (i.prefix[DATA_PREFIX] != 0)
3668 {
3669 prefix = 1;
3670 i.prefixes -= 1;
3671 code16 ^= CODE16;
3672 }
3673 /* Pentium4 branch hints. */
3674 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3675 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3676 {
3677 prefix++;
3678 i.prefixes--;
3679 }
3680 if (i.prefix[REX_PREFIX] != 0)
3681 {
3682 prefix++;
3683 i.prefixes--;
3684 }
3685
3686 if (i.prefixes != 0 && !intel_syntax)
3687 as_warn (_("skipping prefixes on this instruction"));
3688
3689 /* It's always a symbol; End frag & setup for relax.
3690 Make sure there is enough room in this frag for the largest
3691 instruction we may generate in md_convert_frag. This is 2
3692 bytes for the opcode and room for the prefix and largest
3693 displacement. */
3694 frag_grow (prefix + 2 + 4);
3695 /* Prefix and 1 opcode byte go in fr_fix. */
3696 p = frag_more (prefix + 1);
3697 if (i.prefix[DATA_PREFIX] != 0)
3698 *p++ = DATA_PREFIX_OPCODE;
3699 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
3700 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
3701 *p++ = i.prefix[SEG_PREFIX];
3702 if (i.prefix[REX_PREFIX] != 0)
3703 *p++ = i.prefix[REX_PREFIX];
3704 *p = i.tm.base_opcode;
3705
3706 if ((unsigned char) *p == JUMP_PC_RELATIVE)
3707 subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
3708 else if ((cpu_arch_flags & Cpu386) != 0)
3709 subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
3710 else
3711 subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
3712 subtype |= code16;
3713
3714 sym = i.op[0].disps->X_add_symbol;
3715 off = i.op[0].disps->X_add_number;
3716
3717 if (i.op[0].disps->X_op != O_constant
3718 && i.op[0].disps->X_op != O_symbol)
3719 {
3720 /* Handle complex expressions. */
3721 sym = make_expr_symbol (i.op[0].disps);
3722 off = 0;
3723 }
3724
3725 /* 1 possible extra opcode + 4 byte displacement go in var part.
3726 Pass reloc in fr_var. */
3727 frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
3728 }
3729
3730 static void
3731 output_jump (void)
3732 {
3733 char *p;
3734 int size;
3735 fixS *fixP;
3736
3737 if (i.tm.opcode_modifier & JumpByte)
3738 {
3739 /* This is a loop or jecxz type instruction. */
3740 size = 1;
3741 if (i.prefix[ADDR_PREFIX] != 0)
3742 {
3743 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
3744 i.prefixes -= 1;
3745 }
3746 /* Pentium4 branch hints. */
3747 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3748 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3749 {
3750 FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
3751 i.prefixes--;
3752 }
3753 }
3754 else
3755 {
3756 int code16;
3757
3758 code16 = 0;
3759 if (flag_code == CODE_16BIT)
3760 code16 = CODE16;
3761
3762 if (i.prefix[DATA_PREFIX] != 0)
3763 {
3764 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
3765 i.prefixes -= 1;
3766 code16 ^= CODE16;
3767 }
3768
3769 size = 4;
3770 if (code16)
3771 size = 2;
3772 }
3773
3774 if (i.prefix[REX_PREFIX] != 0)
3775 {
3776 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
3777 i.prefixes -= 1;
3778 }
3779
3780 if (i.prefixes != 0 && !intel_syntax)
3781 as_warn (_("skipping prefixes on this instruction"));
3782
3783 p = frag_more (1 + size);
3784 *p++ = i.tm.base_opcode;
3785
3786 fixP = fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3787 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
3788
3789 /* All jumps handled here are signed, but don't use a signed limit
3790 check for 32 and 16 bit jumps as we want to allow wrap around at
3791 4G and 64k respectively. */
3792 if (size == 1)
3793 fixP->fx_signed = 1;
3794 }
3795
3796 static void
3797 output_interseg_jump (void)
3798 {
3799 char *p;
3800 int size;
3801 int prefix;
3802 int code16;
3803
3804 code16 = 0;
3805 if (flag_code == CODE_16BIT)
3806 code16 = CODE16;
3807
3808 prefix = 0;
3809 if (i.prefix[DATA_PREFIX] != 0)
3810 {
3811 prefix = 1;
3812 i.prefixes -= 1;
3813 code16 ^= CODE16;
3814 }
3815 if (i.prefix[REX_PREFIX] != 0)
3816 {
3817 prefix++;
3818 i.prefixes -= 1;
3819 }
3820
3821 size = 4;
3822 if (code16)
3823 size = 2;
3824
3825 if (i.prefixes != 0 && !intel_syntax)
3826 as_warn (_("skipping prefixes on this instruction"));
3827
3828 /* 1 opcode; 2 segment; offset */
3829 p = frag_more (prefix + 1 + 2 + size);
3830
3831 if (i.prefix[DATA_PREFIX] != 0)
3832 *p++ = DATA_PREFIX_OPCODE;
3833
3834 if (i.prefix[REX_PREFIX] != 0)
3835 *p++ = i.prefix[REX_PREFIX];
3836
3837 *p++ = i.tm.base_opcode;
3838 if (i.op[1].imms->X_op == O_constant)
3839 {
3840 offsetT n = i.op[1].imms->X_add_number;
3841
3842 if (size == 2
3843 && !fits_in_unsigned_word (n)
3844 && !fits_in_signed_word (n))
3845 {
3846 as_bad (_("16-bit jump out of range"));
3847 return;
3848 }
3849 md_number_to_chars (p, n, size);
3850 }
3851 else
3852 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3853 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
3854 if (i.op[0].imms->X_op != O_constant)
3855 as_bad (_("can't handle non absolute segment in `%s'"),
3856 i.tm.name);
3857 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
3858 }
3859
3860 static void
3861 output_insn (void)
3862 {
3863 fragS *insn_start_frag;
3864 offsetT insn_start_off;
3865
3866 /* Tie dwarf2 debug info to the address at the start of the insn.
3867 We can't do this after the insn has been output as the current
3868 frag may have been closed off. eg. by frag_var. */
3869 dwarf2_emit_insn (0);
3870
3871 insn_start_frag = frag_now;
3872 insn_start_off = frag_now_fix ();
3873
3874 /* Output jumps. */
3875 if (i.tm.opcode_modifier & Jump)
3876 output_branch ();
3877 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
3878 output_jump ();
3879 else if (i.tm.opcode_modifier & JumpInterSegment)
3880 output_interseg_jump ();
3881 else
3882 {
3883 /* Output normal instructions here. */
3884 char *p;
3885 unsigned char *q;
3886 unsigned int prefix;
3887
3888 /* All opcodes on i386 have either 1 or 2 bytes. Supplemental
3889 Streaming SIMD extensions 3 Instructions have 3 bytes. We may
3890 use one more higher byte to specify a prefix the instruction
3891 requires. */
3892 if ((i.tm.cpu_flags & CpuSSSE3) != 0)
3893 {
3894 if (i.tm.base_opcode & 0xff000000)
3895 {
3896 prefix = (i.tm.base_opcode >> 24) & 0xff;
3897 goto check_prefix;
3898 }
3899 }
3900 else if ((i.tm.base_opcode & 0xff0000) != 0)
3901 {
3902 prefix = (i.tm.base_opcode >> 16) & 0xff;
3903 if ((i.tm.cpu_flags & CpuPadLock) != 0)
3904 {
3905 check_prefix:
3906 if (prefix != REPE_PREFIX_OPCODE
3907 || i.prefix[LOCKREP_PREFIX] != REPE_PREFIX_OPCODE)
3908 add_prefix (prefix);
3909 }
3910 else
3911 add_prefix (prefix);
3912 }
3913
3914 /* The prefix bytes. */
3915 for (q = i.prefix;
3916 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
3917 q++)
3918 {
3919 if (*q)
3920 {
3921 p = frag_more (1);
3922 md_number_to_chars (p, (valueT) *q, 1);
3923 }
3924 }
3925
3926 /* Now the opcode; be careful about word order here! */
3927 if (fits_in_unsigned_byte (i.tm.base_opcode))
3928 {
3929 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
3930 }
3931 else
3932 {
3933 if ((i.tm.cpu_flags & CpuSSSE3) != 0)
3934 {
3935 p = frag_more (3);
3936 *p++ = (i.tm.base_opcode >> 16) & 0xff;
3937 }
3938 else
3939 p = frag_more (2);
3940
3941 /* Put out high byte first: can't use md_number_to_chars! */
3942 *p++ = (i.tm.base_opcode >> 8) & 0xff;
3943 *p = i.tm.base_opcode & 0xff;
3944 }
3945
3946 /* Now the modrm byte and sib byte (if present). */
3947 if (i.tm.opcode_modifier & Modrm)
3948 {
3949 p = frag_more (1);
3950 md_number_to_chars (p,
3951 (valueT) (i.rm.regmem << 0
3952 | i.rm.reg << 3
3953 | i.rm.mode << 6),
3954 1);
3955 /* If i.rm.regmem == ESP (4)
3956 && i.rm.mode != (Register mode)
3957 && not 16 bit
3958 ==> need second modrm byte. */
3959 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
3960 && i.rm.mode != 3
3961 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
3962 {
3963 p = frag_more (1);
3964 md_number_to_chars (p,
3965 (valueT) (i.sib.base << 0
3966 | i.sib.index << 3
3967 | i.sib.scale << 6),
3968 1);
3969 }
3970 }
3971
3972 if (i.disp_operands)
3973 output_disp (insn_start_frag, insn_start_off);
3974
3975 if (i.imm_operands)
3976 output_imm (insn_start_frag, insn_start_off);
3977 }
3978
3979 #ifdef DEBUG386
3980 if (flag_debug)
3981 {
3982 pi ("" /*line*/, &i);
3983 }
3984 #endif /* DEBUG386 */
3985 }
3986
3987 static void
3988 output_disp (fragS *insn_start_frag, offsetT insn_start_off)
3989 {
3990 char *p;
3991 unsigned int n;
3992
3993 for (n = 0; n < i.operands; n++)
3994 {
3995 if (i.types[n] & Disp)
3996 {
3997 if (i.op[n].disps->X_op == O_constant)
3998 {
3999 int size;
4000 offsetT val;
4001
4002 size = 4;
4003 if (i.types[n] & (Disp8 | Disp16 | Disp64))
4004 {
4005 size = 2;
4006 if (i.types[n] & Disp8)
4007 size = 1;
4008 if (i.types[n] & Disp64)
4009 size = 8;
4010 }
4011 val = offset_in_range (i.op[n].disps->X_add_number,
4012 size);
4013 p = frag_more (size);
4014 md_number_to_chars (p, val, size);
4015 }
4016 else
4017 {
4018 enum bfd_reloc_code_real reloc_type;
4019 int size = 4;
4020 int sign = 0;
4021 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
4022
4023 /* The PC relative address is computed relative
4024 to the instruction boundary, so in case immediate
4025 fields follows, we need to adjust the value. */
4026 if (pcrel && i.imm_operands)
4027 {
4028 int imm_size = 4;
4029 unsigned int n1;
4030
4031 for (n1 = 0; n1 < i.operands; n1++)
4032 if (i.types[n1] & Imm)
4033 {
4034 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
4035 {
4036 imm_size = 2;
4037 if (i.types[n1] & (Imm8 | Imm8S))
4038 imm_size = 1;
4039 if (i.types[n1] & Imm64)
4040 imm_size = 8;
4041 }
4042 break;
4043 }
4044 /* We should find the immediate. */
4045 if (n1 == i.operands)
4046 abort ();
4047 i.op[n].disps->X_add_number -= imm_size;
4048 }
4049
4050 if (i.types[n] & Disp32S)
4051 sign = 1;
4052
4053 if (i.types[n] & (Disp16 | Disp64))
4054 {
4055 size = 2;
4056 if (i.types[n] & Disp64)
4057 size = 8;
4058 }
4059
4060 p = frag_more (size);
4061 reloc_type = reloc (size, pcrel, sign, i.reloc[n]);
4062 if (GOT_symbol
4063 && GOT_symbol == i.op[n].disps->X_add_symbol
4064 && (((reloc_type == BFD_RELOC_32
4065 || reloc_type == BFD_RELOC_X86_64_32S
4066 || (reloc_type == BFD_RELOC_64
4067 && object_64bit))
4068 && (i.op[n].disps->X_op == O_symbol
4069 || (i.op[n].disps->X_op == O_add
4070 && ((symbol_get_value_expression
4071 (i.op[n].disps->X_op_symbol)->X_op)
4072 == O_subtract))))
4073 || reloc_type == BFD_RELOC_32_PCREL))
4074 {
4075 offsetT add;
4076
4077 if (insn_start_frag == frag_now)
4078 add = (p - frag_now->fr_literal) - insn_start_off;
4079 else
4080 {
4081 fragS *fr;
4082
4083 add = insn_start_frag->fr_fix - insn_start_off;
4084 for (fr = insn_start_frag->fr_next;
4085 fr && fr != frag_now; fr = fr->fr_next)
4086 add += fr->fr_fix;
4087 add += p - frag_now->fr_literal;
4088 }
4089
4090 if (!object_64bit)
4091 {
4092 reloc_type = BFD_RELOC_386_GOTPC;
4093 i.op[n].imms->X_add_number += add;
4094 }
4095 else if (reloc_type == BFD_RELOC_64)
4096 reloc_type = BFD_RELOC_X86_64_GOTPC64;
4097 else
4098 /* Don't do the adjustment for x86-64, as there
4099 the pcrel addressing is relative to the _next_
4100 insn, and that is taken care of in other code. */
4101 reloc_type = BFD_RELOC_X86_64_GOTPC32;
4102 }
4103 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
4104 i.op[n].disps, pcrel, reloc_type);
4105 }
4106 }
4107 }
4108 }
4109
4110 static void
4111 output_imm (fragS *insn_start_frag, offsetT insn_start_off)
4112 {
4113 char *p;
4114 unsigned int n;
4115
4116 for (n = 0; n < i.operands; n++)
4117 {
4118 if (i.types[n] & Imm)
4119 {
4120 if (i.op[n].imms->X_op == O_constant)
4121 {
4122 int size;
4123 offsetT val;
4124
4125 size = 4;
4126 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
4127 {
4128 size = 2;
4129 if (i.types[n] & (Imm8 | Imm8S))
4130 size = 1;
4131 else if (i.types[n] & Imm64)
4132 size = 8;
4133 }
4134 val = offset_in_range (i.op[n].imms->X_add_number,
4135 size);
4136 p = frag_more (size);
4137 md_number_to_chars (p, val, size);
4138 }
4139 else
4140 {
4141 /* Not absolute_section.
4142 Need a 32-bit fixup (don't support 8bit
4143 non-absolute imms). Try to support other
4144 sizes ... */
4145 enum bfd_reloc_code_real reloc_type;
4146 int size = 4;
4147 int sign = 0;
4148
4149 if ((i.types[n] & (Imm32S))
4150 && (i.suffix == QWORD_MNEM_SUFFIX
4151 || (!i.suffix && (i.tm.opcode_modifier & No_lSuf))))
4152 sign = 1;
4153 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
4154 {
4155 size = 2;
4156 if (i.types[n] & (Imm8 | Imm8S))
4157 size = 1;
4158 if (i.types[n] & Imm64)
4159 size = 8;
4160 }
4161
4162 p = frag_more (size);
4163 reloc_type = reloc (size, 0, sign, i.reloc[n]);
4164
4165 /* This is tough to explain. We end up with this one if we
4166 * have operands that look like
4167 * "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal here is to
4168 * obtain the absolute address of the GOT, and it is strongly
4169 * preferable from a performance point of view to avoid using
4170 * a runtime relocation for this. The actual sequence of
4171 * instructions often look something like:
4172 *
4173 * call .L66
4174 * .L66:
4175 * popl %ebx
4176 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
4177 *
4178 * The call and pop essentially return the absolute address
4179 * of the label .L66 and store it in %ebx. The linker itself
4180 * will ultimately change the first operand of the addl so
4181 * that %ebx points to the GOT, but to keep things simple, the
4182 * .o file must have this operand set so that it generates not
4183 * the absolute address of .L66, but the absolute address of
4184 * itself. This allows the linker itself simply treat a GOTPC
4185 * relocation as asking for a pcrel offset to the GOT to be
4186 * added in, and the addend of the relocation is stored in the
4187 * operand field for the instruction itself.
4188 *
4189 * Our job here is to fix the operand so that it would add
4190 * the correct offset so that %ebx would point to itself. The
4191 * thing that is tricky is that .-.L66 will point to the
4192 * beginning of the instruction, so we need to further modify
4193 * the operand so that it will point to itself. There are
4194 * other cases where you have something like:
4195 *
4196 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
4197 *
4198 * and here no correction would be required. Internally in
4199 * the assembler we treat operands of this form as not being
4200 * pcrel since the '.' is explicitly mentioned, and I wonder
4201 * whether it would simplify matters to do it this way. Who
4202 * knows. In earlier versions of the PIC patches, the
4203 * pcrel_adjust field was used to store the correction, but
4204 * since the expression is not pcrel, I felt it would be
4205 * confusing to do it this way. */
4206
4207 if ((reloc_type == BFD_RELOC_32
4208 || reloc_type == BFD_RELOC_X86_64_32S
4209 || reloc_type == BFD_RELOC_64)
4210 && GOT_symbol
4211 && GOT_symbol == i.op[n].imms->X_add_symbol
4212 && (i.op[n].imms->X_op == O_symbol
4213 || (i.op[n].imms->X_op == O_add
4214 && ((symbol_get_value_expression
4215 (i.op[n].imms->X_op_symbol)->X_op)
4216 == O_subtract))))
4217 {
4218 offsetT add;
4219
4220 if (insn_start_frag == frag_now)
4221 add = (p - frag_now->fr_literal) - insn_start_off;
4222 else
4223 {
4224 fragS *fr;
4225
4226 add = insn_start_frag->fr_fix - insn_start_off;
4227 for (fr = insn_start_frag->fr_next;
4228 fr && fr != frag_now; fr = fr->fr_next)
4229 add += fr->fr_fix;
4230 add += p - frag_now->fr_literal;
4231 }
4232
4233 if (!object_64bit)
4234 reloc_type = BFD_RELOC_386_GOTPC;
4235 else if (size == 4)
4236 reloc_type = BFD_RELOC_X86_64_GOTPC32;
4237 else if (size == 8)
4238 reloc_type = BFD_RELOC_X86_64_GOTPC64;
4239 i.op[n].imms->X_add_number += add;
4240 }
4241 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
4242 i.op[n].imms, 0, reloc_type);
4243 }
4244 }
4245 }
4246 }
4247 \f
4248 /* x86_cons_fix_new is called via the expression parsing code when a
4249 reloc is needed. We use this hook to get the correct .got reloc. */
4250 static enum bfd_reloc_code_real got_reloc = NO_RELOC;
4251 static int cons_sign = -1;
4252
4253 void
4254 x86_cons_fix_new (fragS *frag, unsigned int off, unsigned int len,
4255 expressionS *exp)
4256 {
4257 enum bfd_reloc_code_real r = reloc (len, 0, cons_sign, got_reloc);
4258
4259 got_reloc = NO_RELOC;
4260
4261 #ifdef TE_PE
4262 if (exp->X_op == O_secrel)
4263 {
4264 exp->X_op = O_symbol;
4265 r = BFD_RELOC_32_SECREL;
4266 }
4267 #endif
4268
4269 fix_new_exp (frag, off, len, exp, 0, r);
4270 }
4271
4272 #if (!defined (OBJ_ELF) && !defined (OBJ_MAYBE_ELF)) || defined (LEX_AT)
4273 # define lex_got(reloc, adjust, types) NULL
4274 #else
4275 /* Parse operands of the form
4276 <symbol>@GOTOFF+<nnn>
4277 and similar .plt or .got references.
4278
4279 If we find one, set up the correct relocation in RELOC and copy the
4280 input string, minus the `@GOTOFF' into a malloc'd buffer for
4281 parsing by the calling routine. Return this buffer, and if ADJUST
4282 is non-null set it to the length of the string we removed from the
4283 input line. Otherwise return NULL. */
4284 static char *
4285 lex_got (enum bfd_reloc_code_real *reloc,
4286 int *adjust,
4287 unsigned int *types)
4288 {
4289 /* Some of the relocations depend on the size of what field is to
4290 be relocated. But in our callers i386_immediate and i386_displacement
4291 we don't yet know the operand size (this will be set by insn
4292 matching). Hence we record the word32 relocation here,
4293 and adjust the reloc according to the real size in reloc(). */
4294 static const struct {
4295 const char *str;
4296 const enum bfd_reloc_code_real rel[2];
4297 const unsigned int types64;
4298 } gotrel[] = {
4299 { "PLTOFF", { 0,
4300 BFD_RELOC_X86_64_PLTOFF64 },
4301 Imm64 },
4302 { "PLT", { BFD_RELOC_386_PLT32,
4303 BFD_RELOC_X86_64_PLT32 },
4304 Imm32 | Imm32S | Disp32 },
4305 { "GOTPLT", { 0,
4306 BFD_RELOC_X86_64_GOTPLT64 },
4307 Imm64 | Disp64 },
4308 { "GOTOFF", { BFD_RELOC_386_GOTOFF,
4309 BFD_RELOC_X86_64_GOTOFF64 },
4310 Imm64 | Disp64 },
4311 { "GOTPCREL", { 0,
4312 BFD_RELOC_X86_64_GOTPCREL },
4313 Imm32 | Imm32S | Disp32 },
4314 { "TLSGD", { BFD_RELOC_386_TLS_GD,
4315 BFD_RELOC_X86_64_TLSGD },
4316 Imm32 | Imm32S | Disp32 },
4317 { "TLSLDM", { BFD_RELOC_386_TLS_LDM,
4318 0 },
4319 0 },
4320 { "TLSLD", { 0,
4321 BFD_RELOC_X86_64_TLSLD },
4322 Imm32 | Imm32S | Disp32 },
4323 { "GOTTPOFF", { BFD_RELOC_386_TLS_IE_32,
4324 BFD_RELOC_X86_64_GOTTPOFF },
4325 Imm32 | Imm32S | Disp32 },
4326 { "TPOFF", { BFD_RELOC_386_TLS_LE_32,
4327 BFD_RELOC_X86_64_TPOFF32 },
4328 Imm32 | Imm32S | Imm64 | Disp32 | Disp64 },
4329 { "NTPOFF", { BFD_RELOC_386_TLS_LE,
4330 0 },
4331 0 },
4332 { "DTPOFF", { BFD_RELOC_386_TLS_LDO_32,
4333 BFD_RELOC_X86_64_DTPOFF32 },
4334 Imm32 | Imm32S | Imm64 | Disp32 | Disp64 },
4335 { "GOTNTPOFF",{ BFD_RELOC_386_TLS_GOTIE,
4336 0 },
4337 0 },
4338 { "INDNTPOFF",{ BFD_RELOC_386_TLS_IE,
4339 0 },
4340 0 },
4341 { "GOT", { BFD_RELOC_386_GOT32,
4342 BFD_RELOC_X86_64_GOT32 },
4343 Imm32 | Imm32S | Disp32 | Imm64 },
4344 { "TLSDESC", { BFD_RELOC_386_TLS_GOTDESC,
4345 BFD_RELOC_X86_64_GOTPC32_TLSDESC },
4346 Imm32 | Imm32S | Disp32 },
4347 { "TLSCALL", { BFD_RELOC_386_TLS_DESC_CALL,
4348 BFD_RELOC_X86_64_TLSDESC_CALL },
4349 Imm32 | Imm32S | Disp32 }
4350 };
4351 char *cp;
4352 unsigned int j;
4353
4354 if (!IS_ELF)
4355 return NULL;
4356
4357 for (cp = input_line_pointer; *cp != '@'; cp++)
4358 if (is_end_of_line[(unsigned char) *cp])
4359 return NULL;
4360
4361 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
4362 {
4363 int len;
4364
4365 len = strlen (gotrel[j].str);
4366 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
4367 {
4368 if (gotrel[j].rel[object_64bit] != 0)
4369 {
4370 int first, second;
4371 char *tmpbuf, *past_reloc;
4372
4373 *reloc = gotrel[j].rel[object_64bit];
4374 if (adjust)
4375 *adjust = len;
4376
4377 if (types)
4378 {
4379 if (flag_code != CODE_64BIT)
4380 *types = Imm32 | Disp32;
4381 else
4382 *types = gotrel[j].types64;
4383 }
4384
4385 if (GOT_symbol == NULL)
4386 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
4387
4388 /* Replace the relocation token with ' ', so that
4389 errors like foo@GOTOFF1 will be detected. */
4390
4391 /* The length of the first part of our input line. */
4392 first = cp - input_line_pointer;
4393
4394 /* The second part goes from after the reloc token until
4395 (and including) an end_of_line char. Don't use strlen
4396 here as the end_of_line char may not be a NUL. */
4397 past_reloc = cp + 1 + len;
4398 for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
4399 ;
4400 second = cp - past_reloc;
4401
4402 /* Allocate and copy string. The trailing NUL shouldn't
4403 be necessary, but be safe. */
4404 tmpbuf = xmalloc (first + second + 2);
4405 memcpy (tmpbuf, input_line_pointer, first);
4406 tmpbuf[first] = ' ';
4407 memcpy (tmpbuf + first + 1, past_reloc, second);
4408 tmpbuf[first + second + 1] = '\0';
4409 return tmpbuf;
4410 }
4411
4412 as_bad (_("@%s reloc is not supported with %d-bit output format"),
4413 gotrel[j].str, 1 << (5 + object_64bit));
4414 return NULL;
4415 }
4416 }
4417
4418 /* Might be a symbol version string. Don't as_bad here. */
4419 return NULL;
4420 }
4421
4422 void
4423 x86_cons (expressionS *exp, int size)
4424 {
4425 if (size == 4 || (object_64bit && size == 8))
4426 {
4427 /* Handle @GOTOFF and the like in an expression. */
4428 char *save;
4429 char *gotfree_input_line;
4430 int adjust;
4431
4432 save = input_line_pointer;
4433 gotfree_input_line = lex_got (&got_reloc, &adjust, NULL);
4434 if (gotfree_input_line)
4435 input_line_pointer = gotfree_input_line;
4436
4437 expression (exp);
4438
4439 if (gotfree_input_line)
4440 {
4441 /* expression () has merrily parsed up to the end of line,
4442 or a comma - in the wrong buffer. Transfer how far
4443 input_line_pointer has moved to the right buffer. */
4444 input_line_pointer = (save
4445 + (input_line_pointer - gotfree_input_line)
4446 + adjust);
4447 free (gotfree_input_line);
4448 }
4449 }
4450 else
4451 expression (exp);
4452 }
4453 #endif
4454
4455 static void signed_cons (int size)
4456 {
4457 if (flag_code == CODE_64BIT)
4458 cons_sign = 1;
4459 cons (size);
4460 cons_sign = -1;
4461 }
4462
4463 #ifdef TE_PE
4464 static void
4465 pe_directive_secrel (dummy)
4466 int dummy ATTRIBUTE_UNUSED;
4467 {
4468 expressionS exp;
4469
4470 do
4471 {
4472 expression (&exp);
4473 if (exp.X_op == O_symbol)
4474 exp.X_op = O_secrel;
4475
4476 emit_expr (&exp, 4);
4477 }
4478 while (*input_line_pointer++ == ',');
4479
4480 input_line_pointer--;
4481 demand_empty_rest_of_line ();
4482 }
4483 #endif
4484
4485 static int
4486 i386_immediate (char *imm_start)
4487 {
4488 char *save_input_line_pointer;
4489 char *gotfree_input_line;
4490 segT exp_seg = 0;
4491 expressionS *exp;
4492 unsigned int types = ~0U;
4493
4494 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
4495 {
4496 as_bad (_("at most %d immediate operands are allowed"),
4497 MAX_IMMEDIATE_OPERANDS);
4498 return 0;
4499 }
4500
4501 exp = &im_expressions[i.imm_operands++];
4502 i.op[this_operand].imms = exp;
4503
4504 if (is_space_char (*imm_start))
4505 ++imm_start;
4506
4507 save_input_line_pointer = input_line_pointer;
4508 input_line_pointer = imm_start;
4509
4510 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
4511 if (gotfree_input_line)
4512 input_line_pointer = gotfree_input_line;
4513
4514 exp_seg = expression (exp);
4515
4516 SKIP_WHITESPACE ();
4517 if (*input_line_pointer)
4518 as_bad (_("junk `%s' after expression"), input_line_pointer);
4519
4520 input_line_pointer = save_input_line_pointer;
4521 if (gotfree_input_line)
4522 free (gotfree_input_line);
4523
4524 if (exp->X_op == O_absent || exp->X_op == O_big)
4525 {
4526 /* Missing or bad expr becomes absolute 0. */
4527 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
4528 imm_start);
4529 exp->X_op = O_constant;
4530 exp->X_add_number = 0;
4531 exp->X_add_symbol = (symbolS *) 0;
4532 exp->X_op_symbol = (symbolS *) 0;
4533 }
4534 else if (exp->X_op == O_constant)
4535 {
4536 /* Size it properly later. */
4537 i.types[this_operand] |= Imm64;
4538 /* If BFD64, sign extend val. */
4539 if (!use_rela_relocations
4540 && (exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
4541 exp->X_add_number
4542 = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
4543 }
4544 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4545 else if (OUTPUT_FLAVOR == bfd_target_aout_flavour
4546 && exp_seg != absolute_section
4547 && exp_seg != text_section
4548 && exp_seg != data_section
4549 && exp_seg != bss_section
4550 && exp_seg != undefined_section
4551 && !bfd_is_com_section (exp_seg))
4552 {
4553 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
4554 return 0;
4555 }
4556 #endif
4557 else if (!intel_syntax && exp->X_op == O_register)
4558 {
4559 as_bad (_("illegal immediate register operand %s"), imm_start);
4560 return 0;
4561 }
4562 else
4563 {
4564 /* This is an address. The size of the address will be
4565 determined later, depending on destination register,
4566 suffix, or the default for the section. */
4567 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
4568 i.types[this_operand] &= types;
4569 }
4570
4571 return 1;
4572 }
4573
4574 static char *
4575 i386_scale (char *scale)
4576 {
4577 offsetT val;
4578 char *save = input_line_pointer;
4579
4580 input_line_pointer = scale;
4581 val = get_absolute_expression ();
4582
4583 switch (val)
4584 {
4585 case 1:
4586 i.log2_scale_factor = 0;
4587 break;
4588 case 2:
4589 i.log2_scale_factor = 1;
4590 break;
4591 case 4:
4592 i.log2_scale_factor = 2;
4593 break;
4594 case 8:
4595 i.log2_scale_factor = 3;
4596 break;
4597 default:
4598 {
4599 char sep = *input_line_pointer;
4600
4601 *input_line_pointer = '\0';
4602 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
4603 scale);
4604 *input_line_pointer = sep;
4605 input_line_pointer = save;
4606 return NULL;
4607 }
4608 }
4609 if (i.log2_scale_factor != 0 && i.index_reg == 0)
4610 {
4611 as_warn (_("scale factor of %d without an index register"),
4612 1 << i.log2_scale_factor);
4613 #if SCALE1_WHEN_NO_INDEX
4614 i.log2_scale_factor = 0;
4615 #endif
4616 }
4617 scale = input_line_pointer;
4618 input_line_pointer = save;
4619 return scale;
4620 }
4621
4622 static int
4623 i386_displacement (char *disp_start, char *disp_end)
4624 {
4625 expressionS *exp;
4626 segT exp_seg = 0;
4627 char *save_input_line_pointer;
4628 char *gotfree_input_line;
4629 int bigdisp, override;
4630 unsigned int types = Disp;
4631
4632 if (i.disp_operands == MAX_MEMORY_OPERANDS)
4633 {
4634 as_bad (_("at most %d displacement operands are allowed"),
4635 MAX_MEMORY_OPERANDS);
4636 return 0;
4637 }
4638
4639 if ((i.types[this_operand] & JumpAbsolute)
4640 || !(current_templates->start->opcode_modifier & (Jump | JumpDword)))
4641 {
4642 bigdisp = Disp32;
4643 override = (i.prefix[ADDR_PREFIX] != 0);
4644 }
4645 else
4646 {
4647 /* For PC-relative branches, the width of the displacement
4648 is dependent upon data size, not address size. */
4649 bigdisp = 0;
4650 override = (i.prefix[DATA_PREFIX] != 0);
4651 }
4652 if (flag_code == CODE_64BIT)
4653 {
4654 if (!bigdisp)
4655 bigdisp = ((override || i.suffix == WORD_MNEM_SUFFIX)
4656 ? Disp16
4657 : Disp32S | Disp32);
4658 else if (!override)
4659 bigdisp = Disp64 | Disp32S | Disp32;
4660 }
4661 else
4662 {
4663 if (!bigdisp)
4664 {
4665 if (!override)
4666 override = (i.suffix == (flag_code != CODE_16BIT
4667 ? WORD_MNEM_SUFFIX
4668 : LONG_MNEM_SUFFIX));
4669 bigdisp = Disp32;
4670 }
4671 if ((flag_code == CODE_16BIT) ^ override)
4672 bigdisp = Disp16;
4673 }
4674 i.types[this_operand] |= bigdisp;
4675
4676 exp = &disp_expressions[i.disp_operands];
4677 i.op[this_operand].disps = exp;
4678 i.disp_operands++;
4679 save_input_line_pointer = input_line_pointer;
4680 input_line_pointer = disp_start;
4681 END_STRING_AND_SAVE (disp_end);
4682
4683 #ifndef GCC_ASM_O_HACK
4684 #define GCC_ASM_O_HACK 0
4685 #endif
4686 #if GCC_ASM_O_HACK
4687 END_STRING_AND_SAVE (disp_end + 1);
4688 if ((i.types[this_operand] & BaseIndex) != 0
4689 && displacement_string_end[-1] == '+')
4690 {
4691 /* This hack is to avoid a warning when using the "o"
4692 constraint within gcc asm statements.
4693 For instance:
4694
4695 #define _set_tssldt_desc(n,addr,limit,type) \
4696 __asm__ __volatile__ ( \
4697 "movw %w2,%0\n\t" \
4698 "movw %w1,2+%0\n\t" \
4699 "rorl $16,%1\n\t" \
4700 "movb %b1,4+%0\n\t" \
4701 "movb %4,5+%0\n\t" \
4702 "movb $0,6+%0\n\t" \
4703 "movb %h1,7+%0\n\t" \
4704 "rorl $16,%1" \
4705 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
4706
4707 This works great except that the output assembler ends
4708 up looking a bit weird if it turns out that there is
4709 no offset. You end up producing code that looks like:
4710
4711 #APP
4712 movw $235,(%eax)
4713 movw %dx,2+(%eax)
4714 rorl $16,%edx
4715 movb %dl,4+(%eax)
4716 movb $137,5+(%eax)
4717 movb $0,6+(%eax)
4718 movb %dh,7+(%eax)
4719 rorl $16,%edx
4720 #NO_APP
4721
4722 So here we provide the missing zero. */
4723
4724 *displacement_string_end = '0';
4725 }
4726 #endif
4727 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
4728 if (gotfree_input_line)
4729 input_line_pointer = gotfree_input_line;
4730
4731 exp_seg = expression (exp);
4732
4733 SKIP_WHITESPACE ();
4734 if (*input_line_pointer)
4735 as_bad (_("junk `%s' after expression"), input_line_pointer);
4736 #if GCC_ASM_O_HACK
4737 RESTORE_END_STRING (disp_end + 1);
4738 #endif
4739 RESTORE_END_STRING (disp_end);
4740 input_line_pointer = save_input_line_pointer;
4741 if (gotfree_input_line)
4742 free (gotfree_input_line);
4743
4744 /* We do this to make sure that the section symbol is in
4745 the symbol table. We will ultimately change the relocation
4746 to be relative to the beginning of the section. */
4747 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
4748 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
4749 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTOFF64)
4750 {
4751 if (exp->X_op != O_symbol)
4752 {
4753 as_bad (_("bad expression used with @%s"),
4754 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
4755 ? "GOTPCREL"
4756 : "GOTOFF"));
4757 return 0;
4758 }
4759
4760 if (S_IS_LOCAL (exp->X_add_symbol)
4761 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
4762 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
4763 exp->X_op = O_subtract;
4764 exp->X_op_symbol = GOT_symbol;
4765 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
4766 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
4767 else if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTOFF64)
4768 i.reloc[this_operand] = BFD_RELOC_64;
4769 else
4770 i.reloc[this_operand] = BFD_RELOC_32;
4771 }
4772
4773 if (exp->X_op == O_absent || exp->X_op == O_big)
4774 {
4775 /* Missing or bad expr becomes absolute 0. */
4776 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
4777 disp_start);
4778 exp->X_op = O_constant;
4779 exp->X_add_number = 0;
4780 exp->X_add_symbol = (symbolS *) 0;
4781 exp->X_op_symbol = (symbolS *) 0;
4782 }
4783
4784 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4785 if (exp->X_op != O_constant
4786 && OUTPUT_FLAVOR == bfd_target_aout_flavour
4787 && exp_seg != absolute_section
4788 && exp_seg != text_section
4789 && exp_seg != data_section
4790 && exp_seg != bss_section
4791 && exp_seg != undefined_section
4792 && !bfd_is_com_section (exp_seg))
4793 {
4794 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
4795 return 0;
4796 }
4797 #endif
4798
4799 if (!(i.types[this_operand] & ~Disp))
4800 i.types[this_operand] &= types;
4801
4802 return 1;
4803 }
4804
4805 /* Make sure the memory operand we've been dealt is valid.
4806 Return 1 on success, 0 on a failure. */
4807
4808 static int
4809 i386_index_check (const char *operand_string)
4810 {
4811 int ok;
4812 #if INFER_ADDR_PREFIX
4813 int fudged = 0;
4814
4815 tryprefix:
4816 #endif
4817 ok = 1;
4818 if ((current_templates->start->cpu_flags & CpuSVME)
4819 && current_templates->end[-1].operand_types[0] == AnyMem)
4820 {
4821 /* Memory operands of SVME insns are special in that they only allow
4822 rAX as their memory address and ignore any segment override. */
4823 unsigned RegXX;
4824
4825 /* SKINIT is even more restrictive: it always requires EAX. */
4826 if (strcmp (current_templates->start->name, "skinit") == 0)
4827 RegXX = Reg32;
4828 else if (flag_code == CODE_64BIT)
4829 RegXX = i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32;
4830 else
4831 RegXX = ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)
4832 ? Reg16
4833 : Reg32);
4834 if (!i.base_reg
4835 || !(i.base_reg->reg_type & Acc)
4836 || !(i.base_reg->reg_type & RegXX)
4837 || i.index_reg
4838 || (i.types[0] & Disp))
4839 ok = 0;
4840 }
4841 else if (flag_code == CODE_64BIT)
4842 {
4843 unsigned RegXX = (i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32);
4844
4845 if ((i.base_reg
4846 && ((i.base_reg->reg_type & RegXX) == 0)
4847 && (i.base_reg->reg_type != BaseIndex
4848 || i.index_reg))
4849 || (i.index_reg
4850 && ((i.index_reg->reg_type & (RegXX | BaseIndex))
4851 != (RegXX | BaseIndex))))
4852 ok = 0;
4853 }
4854 else
4855 {
4856 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
4857 {
4858 /* 16bit checks. */
4859 if ((i.base_reg
4860 && ((i.base_reg->reg_type & (Reg16 | BaseIndex | RegRex))
4861 != (Reg16 | BaseIndex)))
4862 || (i.index_reg
4863 && (((i.index_reg->reg_type & (Reg16 | BaseIndex))
4864 != (Reg16 | BaseIndex))
4865 || !(i.base_reg
4866 && i.base_reg->reg_num < 6
4867 && i.index_reg->reg_num >= 6
4868 && i.log2_scale_factor == 0))))
4869 ok = 0;
4870 }
4871 else
4872 {
4873 /* 32bit checks. */
4874 if ((i.base_reg
4875 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
4876 || (i.index_reg
4877 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
4878 != (Reg32 | BaseIndex))))
4879 ok = 0;
4880 }
4881 }
4882 if (!ok)
4883 {
4884 #if INFER_ADDR_PREFIX
4885 if (i.prefix[ADDR_PREFIX] == 0)
4886 {
4887 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
4888 i.prefixes += 1;
4889 /* Change the size of any displacement too. At most one of
4890 Disp16 or Disp32 is set.
4891 FIXME. There doesn't seem to be any real need for separate
4892 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
4893 Removing them would probably clean up the code quite a lot. */
4894 if (flag_code != CODE_64BIT
4895 && (i.types[this_operand] & (Disp16 | Disp32)))
4896 i.types[this_operand] ^= (Disp16 | Disp32);
4897 fudged = 1;
4898 goto tryprefix;
4899 }
4900 if (fudged)
4901 as_bad (_("`%s' is not a valid base/index expression"),
4902 operand_string);
4903 else
4904 #endif
4905 as_bad (_("`%s' is not a valid %s bit base/index expression"),
4906 operand_string,
4907 flag_code_names[flag_code]);
4908 }
4909 return ok;
4910 }
4911
4912 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
4913 on error. */
4914
4915 static int
4916 i386_operand (char *operand_string)
4917 {
4918 const reg_entry *r;
4919 char *end_op;
4920 char *op_string = operand_string;
4921
4922 if (is_space_char (*op_string))
4923 ++op_string;
4924
4925 /* We check for an absolute prefix (differentiating,
4926 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
4927 if (*op_string == ABSOLUTE_PREFIX)
4928 {
4929 ++op_string;
4930 if (is_space_char (*op_string))
4931 ++op_string;
4932 i.types[this_operand] |= JumpAbsolute;
4933 }
4934
4935 /* Check if operand is a register. */
4936 if ((r = parse_register (op_string, &end_op)) != NULL)
4937 {
4938 /* Check for a segment override by searching for ':' after a
4939 segment register. */
4940 op_string = end_op;
4941 if (is_space_char (*op_string))
4942 ++op_string;
4943 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
4944 {
4945 switch (r->reg_num)
4946 {
4947 case 0:
4948 i.seg[i.mem_operands] = &es;
4949 break;
4950 case 1:
4951 i.seg[i.mem_operands] = &cs;
4952 break;
4953 case 2:
4954 i.seg[i.mem_operands] = &ss;
4955 break;
4956 case 3:
4957 i.seg[i.mem_operands] = &ds;
4958 break;
4959 case 4:
4960 i.seg[i.mem_operands] = &fs;
4961 break;
4962 case 5:
4963 i.seg[i.mem_operands] = &gs;
4964 break;
4965 }
4966
4967 /* Skip the ':' and whitespace. */
4968 ++op_string;
4969 if (is_space_char (*op_string))
4970 ++op_string;
4971
4972 if (!is_digit_char (*op_string)
4973 && !is_identifier_char (*op_string)
4974 && *op_string != '('
4975 && *op_string != ABSOLUTE_PREFIX)
4976 {
4977 as_bad (_("bad memory operand `%s'"), op_string);
4978 return 0;
4979 }
4980 /* Handle case of %es:*foo. */
4981 if (*op_string == ABSOLUTE_PREFIX)
4982 {
4983 ++op_string;
4984 if (is_space_char (*op_string))
4985 ++op_string;
4986 i.types[this_operand] |= JumpAbsolute;
4987 }
4988 goto do_memory_reference;
4989 }
4990 if (*op_string)
4991 {
4992 as_bad (_("junk `%s' after register"), op_string);
4993 return 0;
4994 }
4995 i.types[this_operand] |= r->reg_type & ~BaseIndex;
4996 i.op[this_operand].regs = r;
4997 i.reg_operands++;
4998 }
4999 else if (*op_string == REGISTER_PREFIX)
5000 {
5001 as_bad (_("bad register name `%s'"), op_string);
5002 return 0;
5003 }
5004 else if (*op_string == IMMEDIATE_PREFIX)
5005 {
5006 ++op_string;
5007 if (i.types[this_operand] & JumpAbsolute)
5008 {
5009 as_bad (_("immediate operand illegal with absolute jump"));
5010 return 0;
5011 }
5012 if (!i386_immediate (op_string))
5013 return 0;
5014 }
5015 else if (is_digit_char (*op_string)
5016 || is_identifier_char (*op_string)
5017 || *op_string == '(')
5018 {
5019 /* This is a memory reference of some sort. */
5020 char *base_string;
5021
5022 /* Start and end of displacement string expression (if found). */
5023 char *displacement_string_start;
5024 char *displacement_string_end;
5025
5026 do_memory_reference:
5027 if ((i.mem_operands == 1
5028 && (current_templates->start->opcode_modifier & IsString) == 0)
5029 || i.mem_operands == 2)
5030 {
5031 as_bad (_("too many memory references for `%s'"),
5032 current_templates->start->name);
5033 return 0;
5034 }
5035
5036 /* Check for base index form. We detect the base index form by
5037 looking for an ')' at the end of the operand, searching
5038 for the '(' matching it, and finding a REGISTER_PREFIX or ','
5039 after the '('. */
5040 base_string = op_string + strlen (op_string);
5041
5042 --base_string;
5043 if (is_space_char (*base_string))
5044 --base_string;
5045
5046 /* If we only have a displacement, set-up for it to be parsed later. */
5047 displacement_string_start = op_string;
5048 displacement_string_end = base_string + 1;
5049
5050 if (*base_string == ')')
5051 {
5052 char *temp_string;
5053 unsigned int parens_balanced = 1;
5054 /* We've already checked that the number of left & right ()'s are
5055 equal, so this loop will not be infinite. */
5056 do
5057 {
5058 base_string--;
5059 if (*base_string == ')')
5060 parens_balanced++;
5061 if (*base_string == '(')
5062 parens_balanced--;
5063 }
5064 while (parens_balanced);
5065
5066 temp_string = base_string;
5067
5068 /* Skip past '(' and whitespace. */
5069 ++base_string;
5070 if (is_space_char (*base_string))
5071 ++base_string;
5072
5073 if (*base_string == ','
5074 || ((i.base_reg = parse_register (base_string, &end_op))
5075 != NULL))
5076 {
5077 displacement_string_end = temp_string;
5078
5079 i.types[this_operand] |= BaseIndex;
5080
5081 if (i.base_reg)
5082 {
5083 base_string = end_op;
5084 if (is_space_char (*base_string))
5085 ++base_string;
5086 }
5087
5088 /* There may be an index reg or scale factor here. */
5089 if (*base_string == ',')
5090 {
5091 ++base_string;
5092 if (is_space_char (*base_string))
5093 ++base_string;
5094
5095 if ((i.index_reg = parse_register (base_string, &end_op))
5096 != NULL)
5097 {
5098 base_string = end_op;
5099 if (is_space_char (*base_string))
5100 ++base_string;
5101 if (*base_string == ',')
5102 {
5103 ++base_string;
5104 if (is_space_char (*base_string))
5105 ++base_string;
5106 }
5107 else if (*base_string != ')')
5108 {
5109 as_bad (_("expecting `,' or `)' "
5110 "after index register in `%s'"),
5111 operand_string);
5112 return 0;
5113 }
5114 }
5115 else if (*base_string == REGISTER_PREFIX)
5116 {
5117 as_bad (_("bad register name `%s'"), base_string);
5118 return 0;
5119 }
5120
5121 /* Check for scale factor. */
5122 if (*base_string != ')')
5123 {
5124 char *end_scale = i386_scale (base_string);
5125
5126 if (!end_scale)
5127 return 0;
5128
5129 base_string = end_scale;
5130 if (is_space_char (*base_string))
5131 ++base_string;
5132 if (*base_string != ')')
5133 {
5134 as_bad (_("expecting `)' "
5135 "after scale factor in `%s'"),
5136 operand_string);
5137 return 0;
5138 }
5139 }
5140 else if (!i.index_reg)
5141 {
5142 as_bad (_("expecting index register or scale factor "
5143 "after `,'; got '%c'"),
5144 *base_string);
5145 return 0;
5146 }
5147 }
5148 else if (*base_string != ')')
5149 {
5150 as_bad (_("expecting `,' or `)' "
5151 "after base register in `%s'"),
5152 operand_string);
5153 return 0;
5154 }
5155 }
5156 else if (*base_string == REGISTER_PREFIX)
5157 {
5158 as_bad (_("bad register name `%s'"), base_string);
5159 return 0;
5160 }
5161 }
5162
5163 /* If there's an expression beginning the operand, parse it,
5164 assuming displacement_string_start and
5165 displacement_string_end are meaningful. */
5166 if (displacement_string_start != displacement_string_end)
5167 {
5168 if (!i386_displacement (displacement_string_start,
5169 displacement_string_end))
5170 return 0;
5171 }
5172
5173 /* Special case for (%dx) while doing input/output op. */
5174 if (i.base_reg
5175 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
5176 && i.index_reg == 0
5177 && i.log2_scale_factor == 0
5178 && i.seg[i.mem_operands] == 0
5179 && (i.types[this_operand] & Disp) == 0)
5180 {
5181 i.types[this_operand] = InOutPortReg;
5182 return 1;
5183 }
5184
5185 if (i386_index_check (operand_string) == 0)
5186 return 0;
5187 i.mem_operands++;
5188 }
5189 else
5190 {
5191 /* It's not a memory operand; argh! */
5192 as_bad (_("invalid char %s beginning operand %d `%s'"),
5193 output_invalid (*op_string),
5194 this_operand + 1,
5195 op_string);
5196 return 0;
5197 }
5198 return 1; /* Normal return. */
5199 }
5200 \f
5201 /* md_estimate_size_before_relax()
5202
5203 Called just before relax() for rs_machine_dependent frags. The x86
5204 assembler uses these frags to handle variable size jump
5205 instructions.
5206
5207 Any symbol that is now undefined will not become defined.
5208 Return the correct fr_subtype in the frag.
5209 Return the initial "guess for variable size of frag" to caller.
5210 The guess is actually the growth beyond the fixed part. Whatever
5211 we do to grow the fixed or variable part contributes to our
5212 returned value. */
5213
5214 int
5215 md_estimate_size_before_relax (fragP, segment)
5216 fragS *fragP;
5217 segT segment;
5218 {
5219 /* We've already got fragP->fr_subtype right; all we have to do is
5220 check for un-relaxable symbols. On an ELF system, we can't relax
5221 an externally visible symbol, because it may be overridden by a
5222 shared library. */
5223 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
5224 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5225 || (IS_ELF
5226 && (S_IS_EXTERNAL (fragP->fr_symbol)
5227 || S_IS_WEAK (fragP->fr_symbol)))
5228 #endif
5229 )
5230 {
5231 /* Symbol is undefined in this segment, or we need to keep a
5232 reloc so that weak symbols can be overridden. */
5233 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
5234 enum bfd_reloc_code_real reloc_type;
5235 unsigned char *opcode;
5236 int old_fr_fix;
5237
5238 if (fragP->fr_var != NO_RELOC)
5239 reloc_type = fragP->fr_var;
5240 else if (size == 2)
5241 reloc_type = BFD_RELOC_16_PCREL;
5242 else
5243 reloc_type = BFD_RELOC_32_PCREL;
5244
5245 old_fr_fix = fragP->fr_fix;
5246 opcode = (unsigned char *) fragP->fr_opcode;
5247
5248 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
5249 {
5250 case UNCOND_JUMP:
5251 /* Make jmp (0xeb) a (d)word displacement jump. */
5252 opcode[0] = 0xe9;
5253 fragP->fr_fix += size;
5254 fix_new (fragP, old_fr_fix, size,
5255 fragP->fr_symbol,
5256 fragP->fr_offset, 1,
5257 reloc_type);
5258 break;
5259
5260 case COND_JUMP86:
5261 if (size == 2
5262 && (!no_cond_jump_promotion || fragP->fr_var != NO_RELOC))
5263 {
5264 /* Negate the condition, and branch past an
5265 unconditional jump. */
5266 opcode[0] ^= 1;
5267 opcode[1] = 3;
5268 /* Insert an unconditional jump. */
5269 opcode[2] = 0xe9;
5270 /* We added two extra opcode bytes, and have a two byte
5271 offset. */
5272 fragP->fr_fix += 2 + 2;
5273 fix_new (fragP, old_fr_fix + 2, 2,
5274 fragP->fr_symbol,
5275 fragP->fr_offset, 1,
5276 reloc_type);
5277 break;
5278 }
5279 /* Fall through. */
5280
5281 case COND_JUMP:
5282 if (no_cond_jump_promotion && fragP->fr_var == NO_RELOC)
5283 {
5284 fixS *fixP;
5285
5286 fragP->fr_fix += 1;
5287 fixP = fix_new (fragP, old_fr_fix, 1,
5288 fragP->fr_symbol,
5289 fragP->fr_offset, 1,
5290 BFD_RELOC_8_PCREL);
5291 fixP->fx_signed = 1;
5292 break;
5293 }
5294
5295 /* This changes the byte-displacement jump 0x7N
5296 to the (d)word-displacement jump 0x0f,0x8N. */
5297 opcode[1] = opcode[0] + 0x10;
5298 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5299 /* We've added an opcode byte. */
5300 fragP->fr_fix += 1 + size;
5301 fix_new (fragP, old_fr_fix + 1, size,
5302 fragP->fr_symbol,
5303 fragP->fr_offset, 1,
5304 reloc_type);
5305 break;
5306
5307 default:
5308 BAD_CASE (fragP->fr_subtype);
5309 break;
5310 }
5311 frag_wane (fragP);
5312 return fragP->fr_fix - old_fr_fix;
5313 }
5314
5315 /* Guess size depending on current relax state. Initially the relax
5316 state will correspond to a short jump and we return 1, because
5317 the variable part of the frag (the branch offset) is one byte
5318 long. However, we can relax a section more than once and in that
5319 case we must either set fr_subtype back to the unrelaxed state,
5320 or return the value for the appropriate branch. */
5321 return md_relax_table[fragP->fr_subtype].rlx_length;
5322 }
5323
5324 /* Called after relax() is finished.
5325
5326 In: Address of frag.
5327 fr_type == rs_machine_dependent.
5328 fr_subtype is what the address relaxed to.
5329
5330 Out: Any fixSs and constants are set up.
5331 Caller will turn frag into a ".space 0". */
5332
5333 void
5334 md_convert_frag (abfd, sec, fragP)
5335 bfd *abfd ATTRIBUTE_UNUSED;
5336 segT sec ATTRIBUTE_UNUSED;
5337 fragS *fragP;
5338 {
5339 unsigned char *opcode;
5340 unsigned char *where_to_put_displacement = NULL;
5341 offsetT target_address;
5342 offsetT opcode_address;
5343 unsigned int extension = 0;
5344 offsetT displacement_from_opcode_start;
5345
5346 opcode = (unsigned char *) fragP->fr_opcode;
5347
5348 /* Address we want to reach in file space. */
5349 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
5350
5351 /* Address opcode resides at in file space. */
5352 opcode_address = fragP->fr_address + fragP->fr_fix;
5353
5354 /* Displacement from opcode start to fill into instruction. */
5355 displacement_from_opcode_start = target_address - opcode_address;
5356
5357 if ((fragP->fr_subtype & BIG) == 0)
5358 {
5359 /* Don't have to change opcode. */
5360 extension = 1; /* 1 opcode + 1 displacement */
5361 where_to_put_displacement = &opcode[1];
5362 }
5363 else
5364 {
5365 if (no_cond_jump_promotion
5366 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
5367 as_warn_where (fragP->fr_file, fragP->fr_line,
5368 _("long jump required"));
5369
5370 switch (fragP->fr_subtype)
5371 {
5372 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
5373 extension = 4; /* 1 opcode + 4 displacement */
5374 opcode[0] = 0xe9;
5375 where_to_put_displacement = &opcode[1];
5376 break;
5377
5378 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
5379 extension = 2; /* 1 opcode + 2 displacement */
5380 opcode[0] = 0xe9;
5381 where_to_put_displacement = &opcode[1];
5382 break;
5383
5384 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
5385 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
5386 extension = 5; /* 2 opcode + 4 displacement */
5387 opcode[1] = opcode[0] + 0x10;
5388 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5389 where_to_put_displacement = &opcode[2];
5390 break;
5391
5392 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
5393 extension = 3; /* 2 opcode + 2 displacement */
5394 opcode[1] = opcode[0] + 0x10;
5395 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5396 where_to_put_displacement = &opcode[2];
5397 break;
5398
5399 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
5400 extension = 4;
5401 opcode[0] ^= 1;
5402 opcode[1] = 3;
5403 opcode[2] = 0xe9;
5404 where_to_put_displacement = &opcode[3];
5405 break;
5406
5407 default:
5408 BAD_CASE (fragP->fr_subtype);
5409 break;
5410 }
5411 }
5412
5413 /* If size if less then four we are sure that the operand fits,
5414 but if it's 4, then it could be that the displacement is larger
5415 then -/+ 2GB. */
5416 if (DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype) == 4
5417 && object_64bit
5418 && ((addressT) (displacement_from_opcode_start - extension
5419 + ((addressT) 1 << 31))
5420 > (((addressT) 2 << 31) - 1)))
5421 {
5422 as_bad_where (fragP->fr_file, fragP->fr_line,
5423 _("jump target out of range"));
5424 /* Make us emit 0. */
5425 displacement_from_opcode_start = extension;
5426 }
5427 /* Now put displacement after opcode. */
5428 md_number_to_chars ((char *) where_to_put_displacement,
5429 (valueT) (displacement_from_opcode_start - extension),
5430 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
5431 fragP->fr_fix += extension;
5432 }
5433 \f
5434 /* Size of byte displacement jmp. */
5435 int md_short_jump_size = 2;
5436
5437 /* Size of dword displacement jmp. */
5438 int md_long_jump_size = 5;
5439
5440 void
5441 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
5442 char *ptr;
5443 addressT from_addr, to_addr;
5444 fragS *frag ATTRIBUTE_UNUSED;
5445 symbolS *to_symbol ATTRIBUTE_UNUSED;
5446 {
5447 offsetT offset;
5448
5449 offset = to_addr - (from_addr + 2);
5450 /* Opcode for byte-disp jump. */
5451 md_number_to_chars (ptr, (valueT) 0xeb, 1);
5452 md_number_to_chars (ptr + 1, (valueT) offset, 1);
5453 }
5454
5455 void
5456 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
5457 char *ptr;
5458 addressT from_addr, to_addr;
5459 fragS *frag ATTRIBUTE_UNUSED;
5460 symbolS *to_symbol ATTRIBUTE_UNUSED;
5461 {
5462 offsetT offset;
5463
5464 offset = to_addr - (from_addr + 5);
5465 md_number_to_chars (ptr, (valueT) 0xe9, 1);
5466 md_number_to_chars (ptr + 1, (valueT) offset, 4);
5467 }
5468 \f
5469 /* Apply a fixup (fixS) to segment data, once it has been determined
5470 by our caller that we have all the info we need to fix it up.
5471
5472 On the 386, immediates, displacements, and data pointers are all in
5473 the same (little-endian) format, so we don't need to care about which
5474 we are handling. */
5475
5476 void
5477 md_apply_fix (fixP, valP, seg)
5478 /* The fix we're to put in. */
5479 fixS *fixP;
5480 /* Pointer to the value of the bits. */
5481 valueT *valP;
5482 /* Segment fix is from. */
5483 segT seg ATTRIBUTE_UNUSED;
5484 {
5485 char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
5486 valueT value = *valP;
5487
5488 #if !defined (TE_Mach)
5489 if (fixP->fx_pcrel)
5490 {
5491 switch (fixP->fx_r_type)
5492 {
5493 default:
5494 break;
5495
5496 case BFD_RELOC_64:
5497 fixP->fx_r_type = BFD_RELOC_64_PCREL;
5498 break;
5499 case BFD_RELOC_32:
5500 case BFD_RELOC_X86_64_32S:
5501 fixP->fx_r_type = BFD_RELOC_32_PCREL;
5502 break;
5503 case BFD_RELOC_16:
5504 fixP->fx_r_type = BFD_RELOC_16_PCREL;
5505 break;
5506 case BFD_RELOC_8:
5507 fixP->fx_r_type = BFD_RELOC_8_PCREL;
5508 break;
5509 }
5510 }
5511
5512 if (fixP->fx_addsy != NULL
5513 && (fixP->fx_r_type == BFD_RELOC_32_PCREL
5514 || fixP->fx_r_type == BFD_RELOC_64_PCREL
5515 || fixP->fx_r_type == BFD_RELOC_16_PCREL
5516 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
5517 && !use_rela_relocations)
5518 {
5519 /* This is a hack. There should be a better way to handle this.
5520 This covers for the fact that bfd_install_relocation will
5521 subtract the current location (for partial_inplace, PC relative
5522 relocations); see more below. */
5523 #ifndef OBJ_AOUT
5524 if (IS_ELF
5525 #ifdef TE_PE
5526 || OUTPUT_FLAVOR == bfd_target_coff_flavour
5527 #endif
5528 )
5529 value += fixP->fx_where + fixP->fx_frag->fr_address;
5530 #endif
5531 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5532 if (IS_ELF)
5533 {
5534 segT sym_seg = S_GET_SEGMENT (fixP->fx_addsy);
5535
5536 if ((sym_seg == seg
5537 || (symbol_section_p (fixP->fx_addsy)
5538 && sym_seg != absolute_section))
5539 && !generic_force_reloc (fixP))
5540 {
5541 /* Yes, we add the values in twice. This is because
5542 bfd_install_relocation subtracts them out again. I think
5543 bfd_install_relocation is broken, but I don't dare change
5544 it. FIXME. */
5545 value += fixP->fx_where + fixP->fx_frag->fr_address;
5546 }
5547 }
5548 #endif
5549 #if defined (OBJ_COFF) && defined (TE_PE)
5550 /* For some reason, the PE format does not store a
5551 section address offset for a PC relative symbol. */
5552 if (S_GET_SEGMENT (fixP->fx_addsy) != seg
5553 || S_IS_WEAK (fixP->fx_addsy))
5554 value += md_pcrel_from (fixP);
5555 #endif
5556 }
5557
5558 /* Fix a few things - the dynamic linker expects certain values here,
5559 and we must not disappoint it. */
5560 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5561 if (IS_ELF && fixP->fx_addsy)
5562 switch (fixP->fx_r_type)
5563 {
5564 case BFD_RELOC_386_PLT32:
5565 case BFD_RELOC_X86_64_PLT32:
5566 /* Make the jump instruction point to the address of the operand. At
5567 runtime we merely add the offset to the actual PLT entry. */
5568 value = -4;
5569 break;
5570
5571 case BFD_RELOC_386_TLS_GD:
5572 case BFD_RELOC_386_TLS_LDM:
5573 case BFD_RELOC_386_TLS_IE_32:
5574 case BFD_RELOC_386_TLS_IE:
5575 case BFD_RELOC_386_TLS_GOTIE:
5576 case BFD_RELOC_386_TLS_GOTDESC:
5577 case BFD_RELOC_X86_64_TLSGD:
5578 case BFD_RELOC_X86_64_TLSLD:
5579 case BFD_RELOC_X86_64_GOTTPOFF:
5580 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
5581 value = 0; /* Fully resolved at runtime. No addend. */
5582 /* Fallthrough */
5583 case BFD_RELOC_386_TLS_LE:
5584 case BFD_RELOC_386_TLS_LDO_32:
5585 case BFD_RELOC_386_TLS_LE_32:
5586 case BFD_RELOC_X86_64_DTPOFF32:
5587 case BFD_RELOC_X86_64_DTPOFF64:
5588 case BFD_RELOC_X86_64_TPOFF32:
5589 case BFD_RELOC_X86_64_TPOFF64:
5590 S_SET_THREAD_LOCAL (fixP->fx_addsy);
5591 break;
5592
5593 case BFD_RELOC_386_TLS_DESC_CALL:
5594 case BFD_RELOC_X86_64_TLSDESC_CALL:
5595 value = 0; /* Fully resolved at runtime. No addend. */
5596 S_SET_THREAD_LOCAL (fixP->fx_addsy);
5597 fixP->fx_done = 0;
5598 return;
5599
5600 case BFD_RELOC_386_GOT32:
5601 case BFD_RELOC_X86_64_GOT32:
5602 value = 0; /* Fully resolved at runtime. No addend. */
5603 break;
5604
5605 case BFD_RELOC_VTABLE_INHERIT:
5606 case BFD_RELOC_VTABLE_ENTRY:
5607 fixP->fx_done = 0;
5608 return;
5609
5610 default:
5611 break;
5612 }
5613 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
5614 *valP = value;
5615 #endif /* !defined (TE_Mach) */
5616
5617 /* Are we finished with this relocation now? */
5618 if (fixP->fx_addsy == NULL)
5619 fixP->fx_done = 1;
5620 else if (use_rela_relocations)
5621 {
5622 fixP->fx_no_overflow = 1;
5623 /* Remember value for tc_gen_reloc. */
5624 fixP->fx_addnumber = value;
5625 value = 0;
5626 }
5627
5628 md_number_to_chars (p, value, fixP->fx_size);
5629 }
5630 \f
5631 #define MAX_LITTLENUMS 6
5632
5633 /* Turn the string pointed to by litP into a floating point constant
5634 of type TYPE, and emit the appropriate bytes. The number of
5635 LITTLENUMS emitted is stored in *SIZEP. An error message is
5636 returned, or NULL on OK. */
5637
5638 char *
5639 md_atof (type, litP, sizeP)
5640 int type;
5641 char *litP;
5642 int *sizeP;
5643 {
5644 int prec;
5645 LITTLENUM_TYPE words[MAX_LITTLENUMS];
5646 LITTLENUM_TYPE *wordP;
5647 char *t;
5648
5649 switch (type)
5650 {
5651 case 'f':
5652 case 'F':
5653 prec = 2;
5654 break;
5655
5656 case 'd':
5657 case 'D':
5658 prec = 4;
5659 break;
5660
5661 case 'x':
5662 case 'X':
5663 prec = 5;
5664 break;
5665
5666 default:
5667 *sizeP = 0;
5668 return _("Bad call to md_atof ()");
5669 }
5670 t = atof_ieee (input_line_pointer, type, words);
5671 if (t)
5672 input_line_pointer = t;
5673
5674 *sizeP = prec * sizeof (LITTLENUM_TYPE);
5675 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
5676 the bigendian 386. */
5677 for (wordP = words + prec - 1; prec--;)
5678 {
5679 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
5680 litP += sizeof (LITTLENUM_TYPE);
5681 }
5682 return 0;
5683 }
5684 \f
5685 static char output_invalid_buf[sizeof (unsigned char) * 2 + 6];
5686
5687 static char *
5688 output_invalid (int c)
5689 {
5690 if (ISPRINT (c))
5691 snprintf (output_invalid_buf, sizeof (output_invalid_buf),
5692 "'%c'", c);
5693 else
5694 snprintf (output_invalid_buf, sizeof (output_invalid_buf),
5695 "(0x%x)", (unsigned char) c);
5696 return output_invalid_buf;
5697 }
5698
5699 /* REG_STRING starts *before* REGISTER_PREFIX. */
5700
5701 static const reg_entry *
5702 parse_real_register (char *reg_string, char **end_op)
5703 {
5704 char *s = reg_string;
5705 char *p;
5706 char reg_name_given[MAX_REG_NAME_SIZE + 1];
5707 const reg_entry *r;
5708
5709 /* Skip possible REGISTER_PREFIX and possible whitespace. */
5710 if (*s == REGISTER_PREFIX)
5711 ++s;
5712
5713 if (is_space_char (*s))
5714 ++s;
5715
5716 p = reg_name_given;
5717 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
5718 {
5719 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
5720 return (const reg_entry *) NULL;
5721 s++;
5722 }
5723
5724 /* For naked regs, make sure that we are not dealing with an identifier.
5725 This prevents confusing an identifier like `eax_var' with register
5726 `eax'. */
5727 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
5728 return (const reg_entry *) NULL;
5729
5730 *end_op = s;
5731
5732 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
5733
5734 /* Handle floating point regs, allowing spaces in the (i) part. */
5735 if (r == i386_regtab /* %st is first entry of table */)
5736 {
5737 if (is_space_char (*s))
5738 ++s;
5739 if (*s == '(')
5740 {
5741 ++s;
5742 if (is_space_char (*s))
5743 ++s;
5744 if (*s >= '0' && *s <= '7')
5745 {
5746 r = &i386_float_regtab[*s - '0'];
5747 ++s;
5748 if (is_space_char (*s))
5749 ++s;
5750 if (*s == ')')
5751 {
5752 *end_op = s + 1;
5753 return r;
5754 }
5755 }
5756 /* We have "%st(" then garbage. */
5757 return (const reg_entry *) NULL;
5758 }
5759 }
5760
5761 if (r != NULL
5762 && ((r->reg_flags & (RegRex64 | RegRex)) | (r->reg_type & Reg64)) != 0
5763 && (r->reg_type != Control || !(cpu_arch_flags & CpuSledgehammer))
5764 && flag_code != CODE_64BIT)
5765 return (const reg_entry *) NULL;
5766
5767 return r;
5768 }
5769
5770 /* REG_STRING starts *before* REGISTER_PREFIX. */
5771
5772 static const reg_entry *
5773 parse_register (char *reg_string, char **end_op)
5774 {
5775 const reg_entry *r;
5776
5777 if (*reg_string == REGISTER_PREFIX || allow_naked_reg)
5778 r = parse_real_register (reg_string, end_op);
5779 else
5780 r = NULL;
5781 if (!r)
5782 {
5783 char *save = input_line_pointer;
5784 char c;
5785 symbolS *symbolP;
5786
5787 input_line_pointer = reg_string;
5788 c = get_symbol_end ();
5789 symbolP = symbol_find (reg_string);
5790 if (symbolP && S_GET_SEGMENT (symbolP) == reg_section)
5791 {
5792 const expressionS *e = symbol_get_value_expression (symbolP);
5793
5794 know (e->X_op == O_register);
5795 know (e->X_add_number >= 0
5796 && (valueT) e->X_add_number < ARRAY_SIZE (i386_regtab));
5797 r = i386_regtab + e->X_add_number;
5798 *end_op = input_line_pointer;
5799 }
5800 *input_line_pointer = c;
5801 input_line_pointer = save;
5802 }
5803 return r;
5804 }
5805
5806 int
5807 i386_parse_name (char *name, expressionS *e, char *nextcharP)
5808 {
5809 const reg_entry *r;
5810 char *end = input_line_pointer;
5811
5812 *end = *nextcharP;
5813 r = parse_register (name, &input_line_pointer);
5814 if (r && end <= input_line_pointer)
5815 {
5816 *nextcharP = *input_line_pointer;
5817 *input_line_pointer = 0;
5818 e->X_op = O_register;
5819 e->X_add_number = r - i386_regtab;
5820 return 1;
5821 }
5822 input_line_pointer = end;
5823 *end = 0;
5824 return 0;
5825 }
5826
5827 void
5828 md_operand (expressionS *e)
5829 {
5830 if (*input_line_pointer == REGISTER_PREFIX)
5831 {
5832 char *end;
5833 const reg_entry *r = parse_real_register (input_line_pointer, &end);
5834
5835 if (r)
5836 {
5837 e->X_op = O_register;
5838 e->X_add_number = r - i386_regtab;
5839 input_line_pointer = end;
5840 }
5841 }
5842 }
5843
5844 \f
5845 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5846 const char *md_shortopts = "kVQ:sqn";
5847 #else
5848 const char *md_shortopts = "qn";
5849 #endif
5850
5851 #define OPTION_32 (OPTION_MD_BASE + 0)
5852 #define OPTION_64 (OPTION_MD_BASE + 1)
5853 #define OPTION_DIVIDE (OPTION_MD_BASE + 2)
5854 #define OPTION_MARCH (OPTION_MD_BASE + 3)
5855 #define OPTION_MTUNE (OPTION_MD_BASE + 4)
5856
5857 struct option md_longopts[] =
5858 {
5859 {"32", no_argument, NULL, OPTION_32},
5860 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
5861 {"64", no_argument, NULL, OPTION_64},
5862 #endif
5863 {"divide", no_argument, NULL, OPTION_DIVIDE},
5864 {"march", required_argument, NULL, OPTION_MARCH},
5865 {"mtune", required_argument, NULL, OPTION_MTUNE},
5866 {NULL, no_argument, NULL, 0}
5867 };
5868 size_t md_longopts_size = sizeof (md_longopts);
5869
5870 int
5871 md_parse_option (int c, char *arg)
5872 {
5873 unsigned int i;
5874
5875 switch (c)
5876 {
5877 case 'n':
5878 optimize_align_code = 0;
5879 break;
5880
5881 case 'q':
5882 quiet_warnings = 1;
5883 break;
5884
5885 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5886 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
5887 should be emitted or not. FIXME: Not implemented. */
5888 case 'Q':
5889 break;
5890
5891 /* -V: SVR4 argument to print version ID. */
5892 case 'V':
5893 print_version_id ();
5894 break;
5895
5896 /* -k: Ignore for FreeBSD compatibility. */
5897 case 'k':
5898 break;
5899
5900 case 's':
5901 /* -s: On i386 Solaris, this tells the native assembler to use
5902 .stab instead of .stab.excl. We always use .stab anyhow. */
5903 break;
5904 #endif
5905 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
5906 case OPTION_64:
5907 {
5908 const char **list, **l;
5909
5910 list = bfd_target_list ();
5911 for (l = list; *l != NULL; l++)
5912 if (CONST_STRNEQ (*l, "elf64-x86-64")
5913 || strcmp (*l, "coff-x86-64") == 0
5914 || strcmp (*l, "pe-x86-64") == 0
5915 || strcmp (*l, "pei-x86-64") == 0)
5916 {
5917 default_arch = "x86_64";
5918 break;
5919 }
5920 if (*l == NULL)
5921 as_fatal (_("No compiled in support for x86_64"));
5922 free (list);
5923 }
5924 break;
5925 #endif
5926
5927 case OPTION_32:
5928 default_arch = "i386";
5929 break;
5930
5931 case OPTION_DIVIDE:
5932 #ifdef SVR4_COMMENT_CHARS
5933 {
5934 char *n, *t;
5935 const char *s;
5936
5937 n = (char *) xmalloc (strlen (i386_comment_chars) + 1);
5938 t = n;
5939 for (s = i386_comment_chars; *s != '\0'; s++)
5940 if (*s != '/')
5941 *t++ = *s;
5942 *t = '\0';
5943 i386_comment_chars = n;
5944 }
5945 #endif
5946 break;
5947
5948 case OPTION_MARCH:
5949 if (*arg == '.')
5950 as_fatal (_("Invalid -march= option: `%s'"), arg);
5951 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
5952 {
5953 if (strcmp (arg, cpu_arch [i].name) == 0)
5954 {
5955 cpu_arch_isa = cpu_arch[i].type;
5956 cpu_arch_isa_flags = cpu_arch[i].flags;
5957 if (!cpu_arch_tune_set)
5958 {
5959 cpu_arch_tune = cpu_arch_isa;
5960 cpu_arch_tune_flags = cpu_arch_isa_flags;
5961 }
5962 break;
5963 }
5964 }
5965 if (i >= ARRAY_SIZE (cpu_arch))
5966 as_fatal (_("Invalid -march= option: `%s'"), arg);
5967 break;
5968
5969 case OPTION_MTUNE:
5970 if (*arg == '.')
5971 as_fatal (_("Invalid -mtune= option: `%s'"), arg);
5972 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
5973 {
5974 if (strcmp (arg, cpu_arch [i].name) == 0)
5975 {
5976 cpu_arch_tune_set = 1;
5977 cpu_arch_tune = cpu_arch [i].type;
5978 cpu_arch_tune_flags = cpu_arch[i].flags;
5979 break;
5980 }
5981 }
5982 if (i >= ARRAY_SIZE (cpu_arch))
5983 as_fatal (_("Invalid -mtune= option: `%s'"), arg);
5984 break;
5985
5986 default:
5987 return 0;
5988 }
5989 return 1;
5990 }
5991
5992 void
5993 md_show_usage (stream)
5994 FILE *stream;
5995 {
5996 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5997 fprintf (stream, _("\
5998 -Q ignored\n\
5999 -V print assembler version number\n\
6000 -k ignored\n"));
6001 #endif
6002 fprintf (stream, _("\
6003 -n Do not optimize code alignment\n\
6004 -q quieten some warnings\n"));
6005 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
6006 fprintf (stream, _("\
6007 -s ignored\n"));
6008 #endif
6009 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
6010 fprintf (stream, _("\
6011 --32/--64 generate 32bit/64bit code\n"));
6012 #endif
6013 #ifdef SVR4_COMMENT_CHARS
6014 fprintf (stream, _("\
6015 --divide do not treat `/' as a comment character\n"));
6016 #else
6017 fprintf (stream, _("\
6018 --divide ignored\n"));
6019 #endif
6020 fprintf (stream, _("\
6021 -march=CPU/-mtune=CPU generate code/optimize for CPU, where CPU is one of:\n\
6022 i386, i486, pentium, pentiumpro, pentium4, nocona,\n\
6023 core, core2, k6, athlon, k8, generic32, generic64\n"));
6024
6025 }
6026
6027 #if defined(TE_PEP)
6028 const char *
6029 x86_64_target_format (void)
6030 {
6031 if (strcmp (default_arch, "x86_64") == 0)
6032 {
6033 set_code_flag (CODE_64BIT);
6034 return COFF_TARGET_FORMAT;
6035 }
6036 else if (strcmp (default_arch, "i386") == 0)
6037 {
6038 set_code_flag (CODE_32BIT);
6039 return "coff-i386";
6040 }
6041
6042 as_fatal (_("Unknown architecture"));
6043 return NULL;
6044 }
6045 #endif
6046
6047 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
6048 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
6049
6050 /* Pick the target format to use. */
6051
6052 const char *
6053 i386_target_format (void)
6054 {
6055 if (!strcmp (default_arch, "x86_64"))
6056 {
6057 set_code_flag (CODE_64BIT);
6058 if (cpu_arch_isa_flags == 0)
6059 cpu_arch_isa_flags = Cpu186|Cpu286|Cpu386|Cpu486
6060 |Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2
6061 |CpuSSE|CpuSSE2;
6062 if (cpu_arch_tune_flags == 0)
6063 cpu_arch_tune_flags = Cpu186|Cpu286|Cpu386|Cpu486
6064 |Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2
6065 |CpuSSE|CpuSSE2;
6066 }
6067 else if (!strcmp (default_arch, "i386"))
6068 {
6069 set_code_flag (CODE_32BIT);
6070 if (cpu_arch_isa_flags == 0)
6071 cpu_arch_isa_flags = Cpu186|Cpu286|Cpu386;
6072 if (cpu_arch_tune_flags == 0)
6073 cpu_arch_tune_flags = Cpu186|Cpu286|Cpu386;
6074 }
6075 else
6076 as_fatal (_("Unknown architecture"));
6077 switch (OUTPUT_FLAVOR)
6078 {
6079 #ifdef OBJ_MAYBE_AOUT
6080 case bfd_target_aout_flavour:
6081 return AOUT_TARGET_FORMAT;
6082 #endif
6083 #ifdef OBJ_MAYBE_COFF
6084 case bfd_target_coff_flavour:
6085 return "coff-i386";
6086 #endif
6087 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
6088 case bfd_target_elf_flavour:
6089 {
6090 if (flag_code == CODE_64BIT)
6091 {
6092 object_64bit = 1;
6093 use_rela_relocations = 1;
6094 }
6095 return flag_code == CODE_64BIT ? ELF_TARGET_FORMAT64 : ELF_TARGET_FORMAT;
6096 }
6097 #endif
6098 default:
6099 abort ();
6100 return NULL;
6101 }
6102 }
6103
6104 #endif /* OBJ_MAYBE_ more than one */
6105
6106 #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
6107 void
6108 i386_elf_emit_arch_note (void)
6109 {
6110 if (IS_ELF && cpu_arch_name != NULL)
6111 {
6112 char *p;
6113 asection *seg = now_seg;
6114 subsegT subseg = now_subseg;
6115 Elf_Internal_Note i_note;
6116 Elf_External_Note e_note;
6117 asection *note_secp;
6118 int len;
6119
6120 /* Create the .note section. */
6121 note_secp = subseg_new (".note", 0);
6122 bfd_set_section_flags (stdoutput,
6123 note_secp,
6124 SEC_HAS_CONTENTS | SEC_READONLY);
6125
6126 /* Process the arch string. */
6127 len = strlen (cpu_arch_name);
6128
6129 i_note.namesz = len + 1;
6130 i_note.descsz = 0;
6131 i_note.type = NT_ARCH;
6132 p = frag_more (sizeof (e_note.namesz));
6133 md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
6134 p = frag_more (sizeof (e_note.descsz));
6135 md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
6136 p = frag_more (sizeof (e_note.type));
6137 md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
6138 p = frag_more (len + 1);
6139 strcpy (p, cpu_arch_name);
6140
6141 frag_align (2, 0, 0);
6142
6143 subseg_set (seg, subseg);
6144 }
6145 }
6146 #endif
6147 \f
6148 symbolS *
6149 md_undefined_symbol (name)
6150 char *name;
6151 {
6152 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
6153 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
6154 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
6155 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
6156 {
6157 if (!GOT_symbol)
6158 {
6159 if (symbol_find (name))
6160 as_bad (_("GOT already in symbol table"));
6161 GOT_symbol = symbol_new (name, undefined_section,
6162 (valueT) 0, &zero_address_frag);
6163 };
6164 return GOT_symbol;
6165 }
6166 return 0;
6167 }
6168
6169 /* Round up a section size to the appropriate boundary. */
6170
6171 valueT
6172 md_section_align (segment, size)
6173 segT segment ATTRIBUTE_UNUSED;
6174 valueT size;
6175 {
6176 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
6177 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
6178 {
6179 /* For a.out, force the section size to be aligned. If we don't do
6180 this, BFD will align it for us, but it will not write out the
6181 final bytes of the section. This may be a bug in BFD, but it is
6182 easier to fix it here since that is how the other a.out targets
6183 work. */
6184 int align;
6185
6186 align = bfd_get_section_alignment (stdoutput, segment);
6187 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
6188 }
6189 #endif
6190
6191 return size;
6192 }
6193
6194 /* On the i386, PC-relative offsets are relative to the start of the
6195 next instruction. That is, the address of the offset, plus its
6196 size, since the offset is always the last part of the insn. */
6197
6198 long
6199 md_pcrel_from (fixS *fixP)
6200 {
6201 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
6202 }
6203
6204 #ifndef I386COFF
6205
6206 static void
6207 s_bss (int ignore ATTRIBUTE_UNUSED)
6208 {
6209 int temp;
6210
6211 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
6212 if (IS_ELF)
6213 obj_elf_section_change_hook ();
6214 #endif
6215 temp = get_absolute_expression ();
6216 subseg_set (bss_section, (subsegT) temp);
6217 demand_empty_rest_of_line ();
6218 }
6219
6220 #endif
6221
6222 void
6223 i386_validate_fix (fixS *fixp)
6224 {
6225 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
6226 {
6227 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
6228 {
6229 if (!object_64bit)
6230 abort ();
6231 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
6232 }
6233 else
6234 {
6235 if (!object_64bit)
6236 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
6237 else
6238 fixp->fx_r_type = BFD_RELOC_X86_64_GOTOFF64;
6239 }
6240 fixp->fx_subsy = 0;
6241 }
6242 }
6243
6244 arelent *
6245 tc_gen_reloc (section, fixp)
6246 asection *section ATTRIBUTE_UNUSED;
6247 fixS *fixp;
6248 {
6249 arelent *rel;
6250 bfd_reloc_code_real_type code;
6251
6252 switch (fixp->fx_r_type)
6253 {
6254 case BFD_RELOC_X86_64_PLT32:
6255 case BFD_RELOC_X86_64_GOT32:
6256 case BFD_RELOC_X86_64_GOTPCREL:
6257 case BFD_RELOC_386_PLT32:
6258 case BFD_RELOC_386_GOT32:
6259 case BFD_RELOC_386_GOTOFF:
6260 case BFD_RELOC_386_GOTPC:
6261 case BFD_RELOC_386_TLS_GD:
6262 case BFD_RELOC_386_TLS_LDM:
6263 case BFD_RELOC_386_TLS_LDO_32:
6264 case BFD_RELOC_386_TLS_IE_32:
6265 case BFD_RELOC_386_TLS_IE:
6266 case BFD_RELOC_386_TLS_GOTIE:
6267 case BFD_RELOC_386_TLS_LE_32:
6268 case BFD_RELOC_386_TLS_LE:
6269 case BFD_RELOC_386_TLS_GOTDESC:
6270 case BFD_RELOC_386_TLS_DESC_CALL:
6271 case BFD_RELOC_X86_64_TLSGD:
6272 case BFD_RELOC_X86_64_TLSLD:
6273 case BFD_RELOC_X86_64_DTPOFF32:
6274 case BFD_RELOC_X86_64_DTPOFF64:
6275 case BFD_RELOC_X86_64_GOTTPOFF:
6276 case BFD_RELOC_X86_64_TPOFF32:
6277 case BFD_RELOC_X86_64_TPOFF64:
6278 case BFD_RELOC_X86_64_GOTOFF64:
6279 case BFD_RELOC_X86_64_GOTPC32:
6280 case BFD_RELOC_X86_64_GOT64:
6281 case BFD_RELOC_X86_64_GOTPCREL64:
6282 case BFD_RELOC_X86_64_GOTPC64:
6283 case BFD_RELOC_X86_64_GOTPLT64:
6284 case BFD_RELOC_X86_64_PLTOFF64:
6285 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
6286 case BFD_RELOC_X86_64_TLSDESC_CALL:
6287 case BFD_RELOC_RVA:
6288 case BFD_RELOC_VTABLE_ENTRY:
6289 case BFD_RELOC_VTABLE_INHERIT:
6290 #ifdef TE_PE
6291 case BFD_RELOC_32_SECREL:
6292 #endif
6293 code = fixp->fx_r_type;
6294 break;
6295 case BFD_RELOC_X86_64_32S:
6296 if (!fixp->fx_pcrel)
6297 {
6298 /* Don't turn BFD_RELOC_X86_64_32S into BFD_RELOC_32. */
6299 code = fixp->fx_r_type;
6300 break;
6301 }
6302 default:
6303 if (fixp->fx_pcrel)
6304 {
6305 switch (fixp->fx_size)
6306 {
6307 default:
6308 as_bad_where (fixp->fx_file, fixp->fx_line,
6309 _("can not do %d byte pc-relative relocation"),
6310 fixp->fx_size);
6311 code = BFD_RELOC_32_PCREL;
6312 break;
6313 case 1: code = BFD_RELOC_8_PCREL; break;
6314 case 2: code = BFD_RELOC_16_PCREL; break;
6315 case 4: code = BFD_RELOC_32_PCREL; break;
6316 #ifdef BFD64
6317 case 8: code = BFD_RELOC_64_PCREL; break;
6318 #endif
6319 }
6320 }
6321 else
6322 {
6323 switch (fixp->fx_size)
6324 {
6325 default:
6326 as_bad_where (fixp->fx_file, fixp->fx_line,
6327 _("can not do %d byte relocation"),
6328 fixp->fx_size);
6329 code = BFD_RELOC_32;
6330 break;
6331 case 1: code = BFD_RELOC_8; break;
6332 case 2: code = BFD_RELOC_16; break;
6333 case 4: code = BFD_RELOC_32; break;
6334 #ifdef BFD64
6335 case 8: code = BFD_RELOC_64; break;
6336 #endif
6337 }
6338 }
6339 break;
6340 }
6341
6342 if ((code == BFD_RELOC_32
6343 || code == BFD_RELOC_32_PCREL
6344 || code == BFD_RELOC_X86_64_32S)
6345 && GOT_symbol
6346 && fixp->fx_addsy == GOT_symbol)
6347 {
6348 if (!object_64bit)
6349 code = BFD_RELOC_386_GOTPC;
6350 else
6351 code = BFD_RELOC_X86_64_GOTPC32;
6352 }
6353 if ((code == BFD_RELOC_64 || code == BFD_RELOC_64_PCREL)
6354 && GOT_symbol
6355 && fixp->fx_addsy == GOT_symbol)
6356 {
6357 code = BFD_RELOC_X86_64_GOTPC64;
6358 }
6359
6360 rel = (arelent *) xmalloc (sizeof (arelent));
6361 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
6362 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
6363
6364 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
6365
6366 if (!use_rela_relocations)
6367 {
6368 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
6369 vtable entry to be used in the relocation's section offset. */
6370 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
6371 rel->address = fixp->fx_offset;
6372
6373 rel->addend = 0;
6374 }
6375 /* Use the rela in 64bit mode. */
6376 else
6377 {
6378 if (!fixp->fx_pcrel)
6379 rel->addend = fixp->fx_offset;
6380 else
6381 switch (code)
6382 {
6383 case BFD_RELOC_X86_64_PLT32:
6384 case BFD_RELOC_X86_64_GOT32:
6385 case BFD_RELOC_X86_64_GOTPCREL:
6386 case BFD_RELOC_X86_64_TLSGD:
6387 case BFD_RELOC_X86_64_TLSLD:
6388 case BFD_RELOC_X86_64_GOTTPOFF:
6389 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
6390 case BFD_RELOC_X86_64_TLSDESC_CALL:
6391 rel->addend = fixp->fx_offset - fixp->fx_size;
6392 break;
6393 default:
6394 rel->addend = (section->vma
6395 - fixp->fx_size
6396 + fixp->fx_addnumber
6397 + md_pcrel_from (fixp));
6398 break;
6399 }
6400 }
6401
6402 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
6403 if (rel->howto == NULL)
6404 {
6405 as_bad_where (fixp->fx_file, fixp->fx_line,
6406 _("cannot represent relocation type %s"),
6407 bfd_get_reloc_code_name (code));
6408 /* Set howto to a garbage value so that we can keep going. */
6409 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
6410 assert (rel->howto != NULL);
6411 }
6412
6413 return rel;
6414 }
6415
6416 \f
6417 /* Parse operands using Intel syntax. This implements a recursive descent
6418 parser based on the BNF grammar published in Appendix B of the MASM 6.1
6419 Programmer's Guide.
6420
6421 FIXME: We do not recognize the full operand grammar defined in the MASM
6422 documentation. In particular, all the structure/union and
6423 high-level macro operands are missing.
6424
6425 Uppercase words are terminals, lower case words are non-terminals.
6426 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
6427 bars '|' denote choices. Most grammar productions are implemented in
6428 functions called 'intel_<production>'.
6429
6430 Initial production is 'expr'.
6431
6432 addOp + | -
6433
6434 alpha [a-zA-Z]
6435
6436 binOp & | AND | \| | OR | ^ | XOR
6437
6438 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
6439
6440 constant digits [[ radixOverride ]]
6441
6442 dataType BYTE | WORD | DWORD | FWORD | QWORD | TBYTE | OWORD | XMMWORD
6443
6444 digits decdigit
6445 | digits decdigit
6446 | digits hexdigit
6447
6448 decdigit [0-9]
6449
6450 e04 e04 addOp e05
6451 | e05
6452
6453 e05 e05 binOp e06
6454 | e06
6455
6456 e06 e06 mulOp e09
6457 | e09
6458
6459 e09 OFFSET e10
6460 | SHORT e10
6461 | + e10
6462 | - e10
6463 | ~ e10
6464 | NOT e10
6465 | e09 PTR e10
6466 | e09 : e10
6467 | e10
6468
6469 e10 e10 [ expr ]
6470 | e11
6471
6472 e11 ( expr )
6473 | [ expr ]
6474 | constant
6475 | dataType
6476 | id
6477 | $
6478 | register
6479
6480 => expr expr cmpOp e04
6481 | e04
6482
6483 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
6484 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
6485
6486 hexdigit a | b | c | d | e | f
6487 | A | B | C | D | E | F
6488
6489 id alpha
6490 | id alpha
6491 | id decdigit
6492
6493 mulOp * | / | % | MOD | << | SHL | >> | SHR
6494
6495 quote " | '
6496
6497 register specialRegister
6498 | gpRegister
6499 | byteRegister
6500
6501 segmentRegister CS | DS | ES | FS | GS | SS
6502
6503 specialRegister CR0 | CR2 | CR3 | CR4
6504 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
6505 | TR3 | TR4 | TR5 | TR6 | TR7
6506
6507 We simplify the grammar in obvious places (e.g., register parsing is
6508 done by calling parse_register) and eliminate immediate left recursion
6509 to implement a recursive-descent parser.
6510
6511 expr e04 expr'
6512
6513 expr' cmpOp e04 expr'
6514 | Empty
6515
6516 e04 e05 e04'
6517
6518 e04' addOp e05 e04'
6519 | Empty
6520
6521 e05 e06 e05'
6522
6523 e05' binOp e06 e05'
6524 | Empty
6525
6526 e06 e09 e06'
6527
6528 e06' mulOp e09 e06'
6529 | Empty
6530
6531 e09 OFFSET e10 e09'
6532 | SHORT e10'
6533 | + e10'
6534 | - e10'
6535 | ~ e10'
6536 | NOT e10'
6537 | e10 e09'
6538
6539 e09' PTR e10 e09'
6540 | : e10 e09'
6541 | Empty
6542
6543 e10 e11 e10'
6544
6545 e10' [ expr ] e10'
6546 | Empty
6547
6548 e11 ( expr )
6549 | [ expr ]
6550 | BYTE
6551 | WORD
6552 | DWORD
6553 | FWORD
6554 | QWORD
6555 | TBYTE
6556 | OWORD
6557 | XMMWORD
6558 | .
6559 | $
6560 | register
6561 | id
6562 | constant */
6563
6564 /* Parsing structure for the intel syntax parser. Used to implement the
6565 semantic actions for the operand grammar. */
6566 struct intel_parser_s
6567 {
6568 char *op_string; /* The string being parsed. */
6569 int got_a_float; /* Whether the operand is a float. */
6570 int op_modifier; /* Operand modifier. */
6571 int is_mem; /* 1 if operand is memory reference. */
6572 int in_offset; /* >=1 if parsing operand of offset. */
6573 int in_bracket; /* >=1 if parsing operand in brackets. */
6574 const reg_entry *reg; /* Last register reference found. */
6575 char *disp; /* Displacement string being built. */
6576 char *next_operand; /* Resume point when splitting operands. */
6577 };
6578
6579 static struct intel_parser_s intel_parser;
6580
6581 /* Token structure for parsing intel syntax. */
6582 struct intel_token
6583 {
6584 int code; /* Token code. */
6585 const reg_entry *reg; /* Register entry for register tokens. */
6586 char *str; /* String representation. */
6587 };
6588
6589 static struct intel_token cur_token, prev_token;
6590
6591 /* Token codes for the intel parser. Since T_SHORT is already used
6592 by COFF, undefine it first to prevent a warning. */
6593 #define T_NIL -1
6594 #define T_CONST 1
6595 #define T_REG 2
6596 #define T_BYTE 3
6597 #define T_WORD 4
6598 #define T_DWORD 5
6599 #define T_FWORD 6
6600 #define T_QWORD 7
6601 #define T_TBYTE 8
6602 #define T_XMMWORD 9
6603 #undef T_SHORT
6604 #define T_SHORT 10
6605 #define T_OFFSET 11
6606 #define T_PTR 12
6607 #define T_ID 13
6608 #define T_SHL 14
6609 #define T_SHR 15
6610
6611 /* Prototypes for intel parser functions. */
6612 static int intel_match_token (int);
6613 static void intel_putback_token (void);
6614 static void intel_get_token (void);
6615 static int intel_expr (void);
6616 static int intel_e04 (void);
6617 static int intel_e05 (void);
6618 static int intel_e06 (void);
6619 static int intel_e09 (void);
6620 static int intel_e10 (void);
6621 static int intel_e11 (void);
6622
6623 static int
6624 i386_intel_operand (char *operand_string, int got_a_float)
6625 {
6626 int ret;
6627 char *p;
6628
6629 p = intel_parser.op_string = xstrdup (operand_string);
6630 intel_parser.disp = (char *) xmalloc (strlen (operand_string) + 1);
6631
6632 for (;;)
6633 {
6634 /* Initialize token holders. */
6635 cur_token.code = prev_token.code = T_NIL;
6636 cur_token.reg = prev_token.reg = NULL;
6637 cur_token.str = prev_token.str = NULL;
6638
6639 /* Initialize parser structure. */
6640 intel_parser.got_a_float = got_a_float;
6641 intel_parser.op_modifier = 0;
6642 intel_parser.is_mem = 0;
6643 intel_parser.in_offset = 0;
6644 intel_parser.in_bracket = 0;
6645 intel_parser.reg = NULL;
6646 intel_parser.disp[0] = '\0';
6647 intel_parser.next_operand = NULL;
6648
6649 /* Read the first token and start the parser. */
6650 intel_get_token ();
6651 ret = intel_expr ();
6652
6653 if (!ret)
6654 break;
6655
6656 if (cur_token.code != T_NIL)
6657 {
6658 as_bad (_("invalid operand for '%s' ('%s' unexpected)"),
6659 current_templates->start->name, cur_token.str);
6660 ret = 0;
6661 }
6662 /* If we found a memory reference, hand it over to i386_displacement
6663 to fill in the rest of the operand fields. */
6664 else if (intel_parser.is_mem)
6665 {
6666 if ((i.mem_operands == 1
6667 && (current_templates->start->opcode_modifier & IsString) == 0)
6668 || i.mem_operands == 2)
6669 {
6670 as_bad (_("too many memory references for '%s'"),
6671 current_templates->start->name);
6672 ret = 0;
6673 }
6674 else
6675 {
6676 char *s = intel_parser.disp;
6677 i.mem_operands++;
6678
6679 if (!quiet_warnings && intel_parser.is_mem < 0)
6680 /* See the comments in intel_bracket_expr. */
6681 as_warn (_("Treating `%s' as memory reference"), operand_string);
6682
6683 /* Add the displacement expression. */
6684 if (*s != '\0')
6685 ret = i386_displacement (s, s + strlen (s));
6686 if (ret)
6687 {
6688 /* Swap base and index in 16-bit memory operands like
6689 [si+bx]. Since i386_index_check is also used in AT&T
6690 mode we have to do that here. */
6691 if (i.base_reg
6692 && i.index_reg
6693 && (i.base_reg->reg_type & Reg16)
6694 && (i.index_reg->reg_type & Reg16)
6695 && i.base_reg->reg_num >= 6
6696 && i.index_reg->reg_num < 6)
6697 {
6698 const reg_entry *base = i.index_reg;
6699
6700 i.index_reg = i.base_reg;
6701 i.base_reg = base;
6702 }
6703 ret = i386_index_check (operand_string);
6704 }
6705 }
6706 }
6707
6708 /* Constant and OFFSET expressions are handled by i386_immediate. */
6709 else if ((intel_parser.op_modifier & (1 << T_OFFSET))
6710 || intel_parser.reg == NULL)
6711 ret = i386_immediate (intel_parser.disp);
6712
6713 if (intel_parser.next_operand && this_operand >= MAX_OPERANDS - 1)
6714 ret = 0;
6715 if (!ret || !intel_parser.next_operand)
6716 break;
6717 intel_parser.op_string = intel_parser.next_operand;
6718 this_operand = i.operands++;
6719 }
6720
6721 free (p);
6722 free (intel_parser.disp);
6723
6724 return ret;
6725 }
6726
6727 #define NUM_ADDRESS_REGS (!!i.base_reg + !!i.index_reg)
6728
6729 /* expr e04 expr'
6730
6731 expr' cmpOp e04 expr'
6732 | Empty */
6733 static int
6734 intel_expr (void)
6735 {
6736 /* XXX Implement the comparison operators. */
6737 return intel_e04 ();
6738 }
6739
6740 /* e04 e05 e04'
6741
6742 e04' addOp e05 e04'
6743 | Empty */
6744 static int
6745 intel_e04 (void)
6746 {
6747 int nregs = -1;
6748
6749 for (;;)
6750 {
6751 if (!intel_e05())
6752 return 0;
6753
6754 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6755 i.base_reg = i386_regtab + REGNAM_AL; /* al is invalid as base */
6756
6757 if (cur_token.code == '+')
6758 nregs = -1;
6759 else if (cur_token.code == '-')
6760 nregs = NUM_ADDRESS_REGS;
6761 else
6762 return 1;
6763
6764 strcat (intel_parser.disp, cur_token.str);
6765 intel_match_token (cur_token.code);
6766 }
6767 }
6768
6769 /* e05 e06 e05'
6770
6771 e05' binOp e06 e05'
6772 | Empty */
6773 static int
6774 intel_e05 (void)
6775 {
6776 int nregs = ~NUM_ADDRESS_REGS;
6777
6778 for (;;)
6779 {
6780 if (!intel_e06())
6781 return 0;
6782
6783 if (cur_token.code == '&'
6784 || cur_token.code == '|'
6785 || cur_token.code == '^')
6786 {
6787 char str[2];
6788
6789 str[0] = cur_token.code;
6790 str[1] = 0;
6791 strcat (intel_parser.disp, str);
6792 }
6793 else
6794 break;
6795
6796 intel_match_token (cur_token.code);
6797
6798 if (nregs < 0)
6799 nregs = ~nregs;
6800 }
6801 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6802 i.base_reg = i386_regtab + REGNAM_AL + 1; /* cl is invalid as base */
6803 return 1;
6804 }
6805
6806 /* e06 e09 e06'
6807
6808 e06' mulOp e09 e06'
6809 | Empty */
6810 static int
6811 intel_e06 (void)
6812 {
6813 int nregs = ~NUM_ADDRESS_REGS;
6814
6815 for (;;)
6816 {
6817 if (!intel_e09())
6818 return 0;
6819
6820 if (cur_token.code == '*'
6821 || cur_token.code == '/'
6822 || cur_token.code == '%')
6823 {
6824 char str[2];
6825
6826 str[0] = cur_token.code;
6827 str[1] = 0;
6828 strcat (intel_parser.disp, str);
6829 }
6830 else if (cur_token.code == T_SHL)
6831 strcat (intel_parser.disp, "<<");
6832 else if (cur_token.code == T_SHR)
6833 strcat (intel_parser.disp, ">>");
6834 else
6835 break;
6836
6837 intel_match_token (cur_token.code);
6838
6839 if (nregs < 0)
6840 nregs = ~nregs;
6841 }
6842 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6843 i.base_reg = i386_regtab + REGNAM_AL + 2; /* dl is invalid as base */
6844 return 1;
6845 }
6846
6847 /* e09 OFFSET e09
6848 | SHORT e09
6849 | + e09
6850 | - e09
6851 | ~ e09
6852 | NOT e09
6853 | e10 e09'
6854
6855 e09' PTR e10 e09'
6856 | : e10 e09'
6857 | Empty */
6858 static int
6859 intel_e09 (void)
6860 {
6861 int nregs = ~NUM_ADDRESS_REGS;
6862 int in_offset = 0;
6863
6864 for (;;)
6865 {
6866 /* Don't consume constants here. */
6867 if (cur_token.code == '+' || cur_token.code == '-')
6868 {
6869 /* Need to look one token ahead - if the next token
6870 is a constant, the current token is its sign. */
6871 int next_code;
6872
6873 intel_match_token (cur_token.code);
6874 next_code = cur_token.code;
6875 intel_putback_token ();
6876 if (next_code == T_CONST)
6877 break;
6878 }
6879
6880 /* e09 OFFSET e09 */
6881 if (cur_token.code == T_OFFSET)
6882 {
6883 if (!in_offset++)
6884 ++intel_parser.in_offset;
6885 }
6886
6887 /* e09 SHORT e09 */
6888 else if (cur_token.code == T_SHORT)
6889 intel_parser.op_modifier |= 1 << T_SHORT;
6890
6891 /* e09 + e09 */
6892 else if (cur_token.code == '+')
6893 strcat (intel_parser.disp, "+");
6894
6895 /* e09 - e09
6896 | ~ e09
6897 | NOT e09 */
6898 else if (cur_token.code == '-' || cur_token.code == '~')
6899 {
6900 char str[2];
6901
6902 if (nregs < 0)
6903 nregs = ~nregs;
6904 str[0] = cur_token.code;
6905 str[1] = 0;
6906 strcat (intel_parser.disp, str);
6907 }
6908
6909 /* e09 e10 e09' */
6910 else
6911 break;
6912
6913 intel_match_token (cur_token.code);
6914 }
6915
6916 for (;;)
6917 {
6918 if (!intel_e10 ())
6919 return 0;
6920
6921 /* e09' PTR e10 e09' */
6922 if (cur_token.code == T_PTR)
6923 {
6924 char suffix;
6925
6926 if (prev_token.code == T_BYTE)
6927 suffix = BYTE_MNEM_SUFFIX;
6928
6929 else if (prev_token.code == T_WORD)
6930 {
6931 if (current_templates->start->name[0] == 'l'
6932 && current_templates->start->name[2] == 's'
6933 && current_templates->start->name[3] == 0)
6934 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
6935 else if (intel_parser.got_a_float == 2) /* "fi..." */
6936 suffix = SHORT_MNEM_SUFFIX;
6937 else
6938 suffix = WORD_MNEM_SUFFIX;
6939 }
6940
6941 else if (prev_token.code == T_DWORD)
6942 {
6943 if (current_templates->start->name[0] == 'l'
6944 && current_templates->start->name[2] == 's'
6945 && current_templates->start->name[3] == 0)
6946 suffix = WORD_MNEM_SUFFIX;
6947 else if (flag_code == CODE_16BIT
6948 && (current_templates->start->opcode_modifier
6949 & (Jump | JumpDword)))
6950 suffix = LONG_DOUBLE_MNEM_SUFFIX;
6951 else if (intel_parser.got_a_float == 1) /* "f..." */
6952 suffix = SHORT_MNEM_SUFFIX;
6953 else
6954 suffix = LONG_MNEM_SUFFIX;
6955 }
6956
6957 else if (prev_token.code == T_FWORD)
6958 {
6959 if (current_templates->start->name[0] == 'l'
6960 && current_templates->start->name[2] == 's'
6961 && current_templates->start->name[3] == 0)
6962 suffix = LONG_MNEM_SUFFIX;
6963 else if (!intel_parser.got_a_float)
6964 {
6965 if (flag_code == CODE_16BIT)
6966 add_prefix (DATA_PREFIX_OPCODE);
6967 suffix = LONG_DOUBLE_MNEM_SUFFIX;
6968 }
6969 else
6970 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
6971 }
6972
6973 else if (prev_token.code == T_QWORD)
6974 {
6975 if (intel_parser.got_a_float == 1) /* "f..." */
6976 suffix = LONG_MNEM_SUFFIX;
6977 else
6978 suffix = QWORD_MNEM_SUFFIX;
6979 }
6980
6981 else if (prev_token.code == T_TBYTE)
6982 {
6983 if (intel_parser.got_a_float == 1)
6984 suffix = LONG_DOUBLE_MNEM_SUFFIX;
6985 else
6986 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
6987 }
6988
6989 else if (prev_token.code == T_XMMWORD)
6990 {
6991 /* XXX ignored for now, but accepted since gcc uses it */
6992 suffix = 0;
6993 }
6994
6995 else
6996 {
6997 as_bad (_("Unknown operand modifier `%s'"), prev_token.str);
6998 return 0;
6999 }
7000
7001 /* Operands for jump/call using 'ptr' notation denote absolute
7002 addresses. */
7003 if (current_templates->start->opcode_modifier & (Jump | JumpDword))
7004 i.types[this_operand] |= JumpAbsolute;
7005
7006 if (current_templates->start->base_opcode == 0x8d /* lea */)
7007 ;
7008 else if (!i.suffix)
7009 i.suffix = suffix;
7010 else if (i.suffix != suffix)
7011 {
7012 as_bad (_("Conflicting operand modifiers"));
7013 return 0;
7014 }
7015
7016 }
7017
7018 /* e09' : e10 e09' */
7019 else if (cur_token.code == ':')
7020 {
7021 if (prev_token.code != T_REG)
7022 {
7023 /* While {call,jmp} SSSS:OOOO is MASM syntax only when SSSS is a
7024 segment/group identifier (which we don't have), using comma
7025 as the operand separator there is even less consistent, since
7026 there all branches only have a single operand. */
7027 if (this_operand != 0
7028 || intel_parser.in_offset
7029 || intel_parser.in_bracket
7030 || (!(current_templates->start->opcode_modifier
7031 & (Jump|JumpDword|JumpInterSegment))
7032 && !(current_templates->start->operand_types[0]
7033 & JumpAbsolute)))
7034 return intel_match_token (T_NIL);
7035 /* Remember the start of the 2nd operand and terminate 1st
7036 operand here.
7037 XXX This isn't right, yet (when SSSS:OOOO is right operand of
7038 another expression), but it gets at least the simplest case
7039 (a plain number or symbol on the left side) right. */
7040 intel_parser.next_operand = intel_parser.op_string;
7041 *--intel_parser.op_string = '\0';
7042 return intel_match_token (':');
7043 }
7044 }
7045
7046 /* e09' Empty */
7047 else
7048 break;
7049
7050 intel_match_token (cur_token.code);
7051
7052 }
7053
7054 if (in_offset)
7055 {
7056 --intel_parser.in_offset;
7057 if (nregs < 0)
7058 nregs = ~nregs;
7059 if (NUM_ADDRESS_REGS > nregs)
7060 {
7061 as_bad (_("Invalid operand to `OFFSET'"));
7062 return 0;
7063 }
7064 intel_parser.op_modifier |= 1 << T_OFFSET;
7065 }
7066
7067 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
7068 i.base_reg = i386_regtab + REGNAM_AL + 3; /* bl is invalid as base */
7069 return 1;
7070 }
7071
7072 static int
7073 intel_bracket_expr (void)
7074 {
7075 int was_offset = intel_parser.op_modifier & (1 << T_OFFSET);
7076 const char *start = intel_parser.op_string;
7077 int len;
7078
7079 if (i.op[this_operand].regs)
7080 return intel_match_token (T_NIL);
7081
7082 intel_match_token ('[');
7083
7084 /* Mark as a memory operand only if it's not already known to be an
7085 offset expression. If it's an offset expression, we need to keep
7086 the brace in. */
7087 if (!intel_parser.in_offset)
7088 {
7089 ++intel_parser.in_bracket;
7090
7091 /* Operands for jump/call inside brackets denote absolute addresses. */
7092 if (current_templates->start->opcode_modifier & (Jump | JumpDword))
7093 i.types[this_operand] |= JumpAbsolute;
7094
7095 /* Unfortunately gas always diverged from MASM in a respect that can't
7096 be easily fixed without risking to break code sequences likely to be
7097 encountered (the testsuite even check for this): MASM doesn't consider
7098 an expression inside brackets unconditionally as a memory reference.
7099 When that is e.g. a constant, an offset expression, or the sum of the
7100 two, this is still taken as a constant load. gas, however, always
7101 treated these as memory references. As a compromise, we'll try to make
7102 offset expressions inside brackets work the MASM way (since that's
7103 less likely to be found in real world code), but make constants alone
7104 continue to work the traditional gas way. In either case, issue a
7105 warning. */
7106 intel_parser.op_modifier &= ~was_offset;
7107 }
7108 else
7109 strcat (intel_parser.disp, "[");
7110
7111 /* Add a '+' to the displacement string if necessary. */
7112 if (*intel_parser.disp != '\0'
7113 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
7114 strcat (intel_parser.disp, "+");
7115
7116 if (intel_expr ()
7117 && (len = intel_parser.op_string - start - 1,
7118 intel_match_token (']')))
7119 {
7120 /* Preserve brackets when the operand is an offset expression. */
7121 if (intel_parser.in_offset)
7122 strcat (intel_parser.disp, "]");
7123 else
7124 {
7125 --intel_parser.in_bracket;
7126 if (i.base_reg || i.index_reg)
7127 intel_parser.is_mem = 1;
7128 if (!intel_parser.is_mem)
7129 {
7130 if (!(intel_parser.op_modifier & (1 << T_OFFSET)))
7131 /* Defer the warning until all of the operand was parsed. */
7132 intel_parser.is_mem = -1;
7133 else if (!quiet_warnings)
7134 as_warn (_("`[%.*s]' taken to mean just `%.*s'"),
7135 len, start, len, start);
7136 }
7137 }
7138 intel_parser.op_modifier |= was_offset;
7139
7140 return 1;
7141 }
7142 return 0;
7143 }
7144
7145 /* e10 e11 e10'
7146
7147 e10' [ expr ] e10'
7148 | Empty */
7149 static int
7150 intel_e10 (void)
7151 {
7152 if (!intel_e11 ())
7153 return 0;
7154
7155 while (cur_token.code == '[')
7156 {
7157 if (!intel_bracket_expr ())
7158 return 0;
7159 }
7160
7161 return 1;
7162 }
7163
7164 /* e11 ( expr )
7165 | [ expr ]
7166 | BYTE
7167 | WORD
7168 | DWORD
7169 | FWORD
7170 | QWORD
7171 | TBYTE
7172 | OWORD
7173 | XMMWORD
7174 | $
7175 | .
7176 | register
7177 | id
7178 | constant */
7179 static int
7180 intel_e11 (void)
7181 {
7182 switch (cur_token.code)
7183 {
7184 /* e11 ( expr ) */
7185 case '(':
7186 intel_match_token ('(');
7187 strcat (intel_parser.disp, "(");
7188
7189 if (intel_expr () && intel_match_token (')'))
7190 {
7191 strcat (intel_parser.disp, ")");
7192 return 1;
7193 }
7194 return 0;
7195
7196 /* e11 [ expr ] */
7197 case '[':
7198 return intel_bracket_expr ();
7199
7200 /* e11 $
7201 | . */
7202 case '.':
7203 strcat (intel_parser.disp, cur_token.str);
7204 intel_match_token (cur_token.code);
7205
7206 /* Mark as a memory operand only if it's not already known to be an
7207 offset expression. */
7208 if (!intel_parser.in_offset)
7209 intel_parser.is_mem = 1;
7210
7211 return 1;
7212
7213 /* e11 register */
7214 case T_REG:
7215 {
7216 const reg_entry *reg = intel_parser.reg = cur_token.reg;
7217
7218 intel_match_token (T_REG);
7219
7220 /* Check for segment change. */
7221 if (cur_token.code == ':')
7222 {
7223 if (!(reg->reg_type & (SReg2 | SReg3)))
7224 {
7225 as_bad (_("`%s' is not a valid segment register"),
7226 reg->reg_name);
7227 return 0;
7228 }
7229 else if (i.seg[i.mem_operands])
7230 as_warn (_("Extra segment override ignored"));
7231 else
7232 {
7233 if (!intel_parser.in_offset)
7234 intel_parser.is_mem = 1;
7235 switch (reg->reg_num)
7236 {
7237 case 0:
7238 i.seg[i.mem_operands] = &es;
7239 break;
7240 case 1:
7241 i.seg[i.mem_operands] = &cs;
7242 break;
7243 case 2:
7244 i.seg[i.mem_operands] = &ss;
7245 break;
7246 case 3:
7247 i.seg[i.mem_operands] = &ds;
7248 break;
7249 case 4:
7250 i.seg[i.mem_operands] = &fs;
7251 break;
7252 case 5:
7253 i.seg[i.mem_operands] = &gs;
7254 break;
7255 }
7256 }
7257 }
7258
7259 /* Not a segment register. Check for register scaling. */
7260 else if (cur_token.code == '*')
7261 {
7262 if (!intel_parser.in_bracket)
7263 {
7264 as_bad (_("Register scaling only allowed in memory operands"));
7265 return 0;
7266 }
7267
7268 if (reg->reg_type & Reg16) /* Disallow things like [si*1]. */
7269 reg = i386_regtab + REGNAM_AX + 4; /* sp is invalid as index */
7270 else if (i.index_reg)
7271 reg = i386_regtab + REGNAM_EAX + 4; /* esp is invalid as index */
7272
7273 /* What follows must be a valid scale. */
7274 intel_match_token ('*');
7275 i.index_reg = reg;
7276 i.types[this_operand] |= BaseIndex;
7277
7278 /* Set the scale after setting the register (otherwise,
7279 i386_scale will complain) */
7280 if (cur_token.code == '+' || cur_token.code == '-')
7281 {
7282 char *str, sign = cur_token.code;
7283 intel_match_token (cur_token.code);
7284 if (cur_token.code != T_CONST)
7285 {
7286 as_bad (_("Syntax error: Expecting a constant, got `%s'"),
7287 cur_token.str);
7288 return 0;
7289 }
7290 str = (char *) xmalloc (strlen (cur_token.str) + 2);
7291 strcpy (str + 1, cur_token.str);
7292 *str = sign;
7293 if (!i386_scale (str))
7294 return 0;
7295 free (str);
7296 }
7297 else if (!i386_scale (cur_token.str))
7298 return 0;
7299 intel_match_token (cur_token.code);
7300 }
7301
7302 /* No scaling. If this is a memory operand, the register is either a
7303 base register (first occurrence) or an index register (second
7304 occurrence). */
7305 else if (intel_parser.in_bracket)
7306 {
7307
7308 if (!i.base_reg)
7309 i.base_reg = reg;
7310 else if (!i.index_reg)
7311 i.index_reg = reg;
7312 else
7313 {
7314 as_bad (_("Too many register references in memory operand"));
7315 return 0;
7316 }
7317
7318 i.types[this_operand] |= BaseIndex;
7319 }
7320
7321 /* It's neither base nor index. */
7322 else if (!intel_parser.in_offset && !intel_parser.is_mem)
7323 {
7324 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
7325 i.op[this_operand].regs = reg;
7326 i.reg_operands++;
7327 }
7328 else
7329 {
7330 as_bad (_("Invalid use of register"));
7331 return 0;
7332 }
7333
7334 /* Since registers are not part of the displacement string (except
7335 when we're parsing offset operands), we may need to remove any
7336 preceding '+' from the displacement string. */
7337 if (*intel_parser.disp != '\0'
7338 && !intel_parser.in_offset)
7339 {
7340 char *s = intel_parser.disp;
7341 s += strlen (s) - 1;
7342 if (*s == '+')
7343 *s = '\0';
7344 }
7345
7346 return 1;
7347 }
7348
7349 /* e11 BYTE
7350 | WORD
7351 | DWORD
7352 | FWORD
7353 | QWORD
7354 | TBYTE
7355 | OWORD
7356 | XMMWORD */
7357 case T_BYTE:
7358 case T_WORD:
7359 case T_DWORD:
7360 case T_FWORD:
7361 case T_QWORD:
7362 case T_TBYTE:
7363 case T_XMMWORD:
7364 intel_match_token (cur_token.code);
7365
7366 if (cur_token.code == T_PTR)
7367 return 1;
7368
7369 /* It must have been an identifier. */
7370 intel_putback_token ();
7371 cur_token.code = T_ID;
7372 /* FALLTHRU */
7373
7374 /* e11 id
7375 | constant */
7376 case T_ID:
7377 if (!intel_parser.in_offset && intel_parser.is_mem <= 0)
7378 {
7379 symbolS *symbolP;
7380
7381 /* The identifier represents a memory reference only if it's not
7382 preceded by an offset modifier and if it's not an equate. */
7383 symbolP = symbol_find(cur_token.str);
7384 if (!symbolP || S_GET_SEGMENT(symbolP) != absolute_section)
7385 intel_parser.is_mem = 1;
7386 }
7387 /* FALLTHRU */
7388
7389 case T_CONST:
7390 case '-':
7391 case '+':
7392 {
7393 char *save_str, sign = 0;
7394
7395 /* Allow constants that start with `+' or `-'. */
7396 if (cur_token.code == '-' || cur_token.code == '+')
7397 {
7398 sign = cur_token.code;
7399 intel_match_token (cur_token.code);
7400 if (cur_token.code != T_CONST)
7401 {
7402 as_bad (_("Syntax error: Expecting a constant, got `%s'"),
7403 cur_token.str);
7404 return 0;
7405 }
7406 }
7407
7408 save_str = (char *) xmalloc (strlen (cur_token.str) + 2);
7409 strcpy (save_str + !!sign, cur_token.str);
7410 if (sign)
7411 *save_str = sign;
7412
7413 /* Get the next token to check for register scaling. */
7414 intel_match_token (cur_token.code);
7415
7416 /* Check if this constant is a scaling factor for an
7417 index register. */
7418 if (cur_token.code == '*')
7419 {
7420 if (intel_match_token ('*') && cur_token.code == T_REG)
7421 {
7422 const reg_entry *reg = cur_token.reg;
7423
7424 if (!intel_parser.in_bracket)
7425 {
7426 as_bad (_("Register scaling only allowed "
7427 "in memory operands"));
7428 return 0;
7429 }
7430
7431 /* Disallow things like [1*si].
7432 sp and esp are invalid as index. */
7433 if (reg->reg_type & Reg16)
7434 reg = i386_regtab + REGNAM_AX + 4;
7435 else if (i.index_reg)
7436 reg = i386_regtab + REGNAM_EAX + 4;
7437
7438 /* The constant is followed by `* reg', so it must be
7439 a valid scale. */
7440 i.index_reg = reg;
7441 i.types[this_operand] |= BaseIndex;
7442
7443 /* Set the scale after setting the register (otherwise,
7444 i386_scale will complain) */
7445 if (!i386_scale (save_str))
7446 return 0;
7447 intel_match_token (T_REG);
7448
7449 /* Since registers are not part of the displacement
7450 string, we may need to remove any preceding '+' from
7451 the displacement string. */
7452 if (*intel_parser.disp != '\0')
7453 {
7454 char *s = intel_parser.disp;
7455 s += strlen (s) - 1;
7456 if (*s == '+')
7457 *s = '\0';
7458 }
7459
7460 free (save_str);
7461
7462 return 1;
7463 }
7464
7465 /* The constant was not used for register scaling. Since we have
7466 already consumed the token following `*' we now need to put it
7467 back in the stream. */
7468 intel_putback_token ();
7469 }
7470
7471 /* Add the constant to the displacement string. */
7472 strcat (intel_parser.disp, save_str);
7473 free (save_str);
7474
7475 return 1;
7476 }
7477 }
7478
7479 as_bad (_("Unrecognized token '%s'"), cur_token.str);
7480 return 0;
7481 }
7482
7483 /* Match the given token against cur_token. If they match, read the next
7484 token from the operand string. */
7485 static int
7486 intel_match_token (int code)
7487 {
7488 if (cur_token.code == code)
7489 {
7490 intel_get_token ();
7491 return 1;
7492 }
7493 else
7494 {
7495 as_bad (_("Unexpected token `%s'"), cur_token.str);
7496 return 0;
7497 }
7498 }
7499
7500 /* Read a new token from intel_parser.op_string and store it in cur_token. */
7501 static void
7502 intel_get_token (void)
7503 {
7504 char *end_op;
7505 const reg_entry *reg;
7506 struct intel_token new_token;
7507
7508 new_token.code = T_NIL;
7509 new_token.reg = NULL;
7510 new_token.str = NULL;
7511
7512 /* Free the memory allocated to the previous token and move
7513 cur_token to prev_token. */
7514 if (prev_token.str)
7515 free (prev_token.str);
7516
7517 prev_token = cur_token;
7518
7519 /* Skip whitespace. */
7520 while (is_space_char (*intel_parser.op_string))
7521 intel_parser.op_string++;
7522
7523 /* Return an empty token if we find nothing else on the line. */
7524 if (*intel_parser.op_string == '\0')
7525 {
7526 cur_token = new_token;
7527 return;
7528 }
7529
7530 /* The new token cannot be larger than the remainder of the operand
7531 string. */
7532 new_token.str = (char *) xmalloc (strlen (intel_parser.op_string) + 1);
7533 new_token.str[0] = '\0';
7534
7535 if (strchr ("0123456789", *intel_parser.op_string))
7536 {
7537 char *p = new_token.str;
7538 char *q = intel_parser.op_string;
7539 new_token.code = T_CONST;
7540
7541 /* Allow any kind of identifier char to encompass floating point and
7542 hexadecimal numbers. */
7543 while (is_identifier_char (*q))
7544 *p++ = *q++;
7545 *p = '\0';
7546
7547 /* Recognize special symbol names [0-9][bf]. */
7548 if (strlen (intel_parser.op_string) == 2
7549 && (intel_parser.op_string[1] == 'b'
7550 || intel_parser.op_string[1] == 'f'))
7551 new_token.code = T_ID;
7552 }
7553
7554 else if ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL)
7555 {
7556 size_t len = end_op - intel_parser.op_string;
7557
7558 new_token.code = T_REG;
7559 new_token.reg = reg;
7560
7561 memcpy (new_token.str, intel_parser.op_string, len);
7562 new_token.str[len] = '\0';
7563 }
7564
7565 else if (is_identifier_char (*intel_parser.op_string))
7566 {
7567 char *p = new_token.str;
7568 char *q = intel_parser.op_string;
7569
7570 /* A '.' or '$' followed by an identifier char is an identifier.
7571 Otherwise, it's operator '.' followed by an expression. */
7572 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
7573 {
7574 new_token.code = '.';
7575 new_token.str[0] = '.';
7576 new_token.str[1] = '\0';
7577 }
7578 else
7579 {
7580 while (is_identifier_char (*q) || *q == '@')
7581 *p++ = *q++;
7582 *p = '\0';
7583
7584 if (strcasecmp (new_token.str, "NOT") == 0)
7585 new_token.code = '~';
7586
7587 else if (strcasecmp (new_token.str, "MOD") == 0)
7588 new_token.code = '%';
7589
7590 else if (strcasecmp (new_token.str, "AND") == 0)
7591 new_token.code = '&';
7592
7593 else if (strcasecmp (new_token.str, "OR") == 0)
7594 new_token.code = '|';
7595
7596 else if (strcasecmp (new_token.str, "XOR") == 0)
7597 new_token.code = '^';
7598
7599 else if (strcasecmp (new_token.str, "SHL") == 0)
7600 new_token.code = T_SHL;
7601
7602 else if (strcasecmp (new_token.str, "SHR") == 0)
7603 new_token.code = T_SHR;
7604
7605 else if (strcasecmp (new_token.str, "BYTE") == 0)
7606 new_token.code = T_BYTE;
7607
7608 else if (strcasecmp (new_token.str, "WORD") == 0)
7609 new_token.code = T_WORD;
7610
7611 else if (strcasecmp (new_token.str, "DWORD") == 0)
7612 new_token.code = T_DWORD;
7613
7614 else if (strcasecmp (new_token.str, "FWORD") == 0)
7615 new_token.code = T_FWORD;
7616
7617 else if (strcasecmp (new_token.str, "QWORD") == 0)
7618 new_token.code = T_QWORD;
7619
7620 else if (strcasecmp (new_token.str, "TBYTE") == 0
7621 /* XXX remove (gcc still uses it) */
7622 || strcasecmp (new_token.str, "XWORD") == 0)
7623 new_token.code = T_TBYTE;
7624
7625 else if (strcasecmp (new_token.str, "XMMWORD") == 0
7626 || strcasecmp (new_token.str, "OWORD") == 0)
7627 new_token.code = T_XMMWORD;
7628
7629 else if (strcasecmp (new_token.str, "PTR") == 0)
7630 new_token.code = T_PTR;
7631
7632 else if (strcasecmp (new_token.str, "SHORT") == 0)
7633 new_token.code = T_SHORT;
7634
7635 else if (strcasecmp (new_token.str, "OFFSET") == 0)
7636 {
7637 new_token.code = T_OFFSET;
7638
7639 /* ??? This is not mentioned in the MASM grammar but gcc
7640 makes use of it with -mintel-syntax. OFFSET may be
7641 followed by FLAT: */
7642 if (strncasecmp (q, " FLAT:", 6) == 0)
7643 strcat (new_token.str, " FLAT:");
7644 }
7645
7646 /* ??? This is not mentioned in the MASM grammar. */
7647 else if (strcasecmp (new_token.str, "FLAT") == 0)
7648 {
7649 new_token.code = T_OFFSET;
7650 if (*q == ':')
7651 strcat (new_token.str, ":");
7652 else
7653 as_bad (_("`:' expected"));
7654 }
7655
7656 else
7657 new_token.code = T_ID;
7658 }
7659 }
7660
7661 else if (strchr ("+-/*%|&^:[]()~", *intel_parser.op_string))
7662 {
7663 new_token.code = *intel_parser.op_string;
7664 new_token.str[0] = *intel_parser.op_string;
7665 new_token.str[1] = '\0';
7666 }
7667
7668 else if (strchr ("<>", *intel_parser.op_string)
7669 && *intel_parser.op_string == *(intel_parser.op_string + 1))
7670 {
7671 new_token.code = *intel_parser.op_string == '<' ? T_SHL : T_SHR;
7672 new_token.str[0] = *intel_parser.op_string;
7673 new_token.str[1] = *intel_parser.op_string;
7674 new_token.str[2] = '\0';
7675 }
7676
7677 else
7678 as_bad (_("Unrecognized token `%s'"), intel_parser.op_string);
7679
7680 intel_parser.op_string += strlen (new_token.str);
7681 cur_token = new_token;
7682 }
7683
7684 /* Put cur_token back into the token stream and make cur_token point to
7685 prev_token. */
7686 static void
7687 intel_putback_token (void)
7688 {
7689 if (cur_token.code != T_NIL)
7690 {
7691 intel_parser.op_string -= strlen (cur_token.str);
7692 free (cur_token.str);
7693 }
7694 cur_token = prev_token;
7695
7696 /* Forget prev_token. */
7697 prev_token.code = T_NIL;
7698 prev_token.reg = NULL;
7699 prev_token.str = NULL;
7700 }
7701
7702 int
7703 tc_x86_regname_to_dw2regnum (char *regname)
7704 {
7705 unsigned int regnum;
7706 unsigned int regnames_count;
7707 static const char *const regnames_32[] =
7708 {
7709 "eax", "ecx", "edx", "ebx",
7710 "esp", "ebp", "esi", "edi",
7711 "eip", "eflags", NULL,
7712 "st0", "st1", "st2", "st3",
7713 "st4", "st5", "st6", "st7",
7714 NULL, NULL,
7715 "xmm0", "xmm1", "xmm2", "xmm3",
7716 "xmm4", "xmm5", "xmm6", "xmm7",
7717 "mm0", "mm1", "mm2", "mm3",
7718 "mm4", "mm5", "mm6", "mm7",
7719 "fcw", "fsw", "mxcsr",
7720 "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL,
7721 "tr", "ldtr"
7722 };
7723 static const char *const regnames_64[] =
7724 {
7725 "rax", "rdx", "rcx", "rbx",
7726 "rsi", "rdi", "rbp", "rsp",
7727 "r8", "r9", "r10", "r11",
7728 "r12", "r13", "r14", "r15",
7729 "rip",
7730 "xmm0", "xmm1", "xmm2", "xmm3",
7731 "xmm4", "xmm5", "xmm6", "xmm7",
7732 "xmm8", "xmm9", "xmm10", "xmm11",
7733 "xmm12", "xmm13", "xmm14", "xmm15",
7734 "st0", "st1", "st2", "st3",
7735 "st4", "st5", "st6", "st7",
7736 "mm0", "mm1", "mm2", "mm3",
7737 "mm4", "mm5", "mm6", "mm7",
7738 "rflags",
7739 "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL,
7740 "fs.base", "gs.base", NULL, NULL,
7741 "tr", "ldtr",
7742 "mxcsr", "fcw", "fsw"
7743 };
7744 const char *const *regnames;
7745
7746 if (flag_code == CODE_64BIT)
7747 {
7748 regnames = regnames_64;
7749 regnames_count = ARRAY_SIZE (regnames_64);
7750 }
7751 else
7752 {
7753 regnames = regnames_32;
7754 regnames_count = ARRAY_SIZE (regnames_32);
7755 }
7756
7757 for (regnum = 0; regnum < regnames_count; regnum++)
7758 if (regnames[regnum] != NULL
7759 && strcmp (regname, regnames[regnum]) == 0)
7760 return regnum;
7761
7762 return -1;
7763 }
7764
7765 void
7766 tc_x86_frame_initial_instructions (void)
7767 {
7768 static unsigned int sp_regno;
7769
7770 if (!sp_regno)
7771 sp_regno = tc_x86_regname_to_dw2regnum (flag_code == CODE_64BIT
7772 ? "rsp" : "esp");
7773
7774 cfi_add_CFA_def_cfa (sp_regno, -x86_cie_data_alignment);
7775 cfi_add_CFA_offset (x86_dwarf2_return_column, x86_cie_data_alignment);
7776 }
7777
7778 int
7779 i386_elf_section_type (const char *str, size_t len)
7780 {
7781 if (flag_code == CODE_64BIT
7782 && len == sizeof ("unwind") - 1
7783 && strncmp (str, "unwind", 6) == 0)
7784 return SHT_X86_64_UNWIND;
7785
7786 return -1;
7787 }
7788
7789 #ifdef TE_PE
7790 void
7791 tc_pe_dwarf2_emit_offset (symbolS *symbol, unsigned int size)
7792 {
7793 expressionS expr;
7794
7795 expr.X_op = O_secrel;
7796 expr.X_add_symbol = symbol;
7797 expr.X_add_number = 0;
7798 emit_expr (&expr, size);
7799 }
7800 #endif
7801
7802 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
7803 /* For ELF on x86-64, add support for SHF_X86_64_LARGE. */
7804
7805 int
7806 x86_64_section_letter (int letter, char **ptr_msg)
7807 {
7808 if (flag_code == CODE_64BIT)
7809 {
7810 if (letter == 'l')
7811 return SHF_X86_64_LARGE;
7812
7813 *ptr_msg = _("Bad .section directive: want a,l,w,x,M,S,G,T in string");
7814 }
7815 else
7816 *ptr_msg = _("Bad .section directive: want a,w,x,M,S,G,T in string");
7817 return -1;
7818 }
7819
7820 int
7821 x86_64_section_word (char *str, size_t len)
7822 {
7823 if (len == 5 && flag_code == CODE_64BIT && CONST_STRNEQ (str, "large"))
7824 return SHF_X86_64_LARGE;
7825
7826 return -1;
7827 }
7828
7829 static void
7830 handle_large_common (int small ATTRIBUTE_UNUSED)
7831 {
7832 if (flag_code != CODE_64BIT)
7833 {
7834 s_comm_internal (0, elf_common_parse);
7835 as_warn (_(".largecomm supported only in 64bit mode, producing .comm"));
7836 }
7837 else
7838 {
7839 static segT lbss_section;
7840 asection *saved_com_section_ptr = elf_com_section_ptr;
7841 asection *saved_bss_section = bss_section;
7842
7843 if (lbss_section == NULL)
7844 {
7845 flagword applicable;
7846 segT seg = now_seg;
7847 subsegT subseg = now_subseg;
7848
7849 /* The .lbss section is for local .largecomm symbols. */
7850 lbss_section = subseg_new (".lbss", 0);
7851 applicable = bfd_applicable_section_flags (stdoutput);
7852 bfd_set_section_flags (stdoutput, lbss_section,
7853 applicable & SEC_ALLOC);
7854 seg_info (lbss_section)->bss = 1;
7855
7856 subseg_set (seg, subseg);
7857 }
7858
7859 elf_com_section_ptr = &_bfd_elf_large_com_section;
7860 bss_section = lbss_section;
7861
7862 s_comm_internal (0, elf_common_parse);
7863
7864 elf_com_section_ptr = saved_com_section_ptr;
7865 bss_section = saved_bss_section;
7866 }
7867 }
7868 #endif /* OBJ_ELF || OBJ_MAYBE_ELF */
This page took 0.206488 seconds and 5 git commands to generate.