2007-03-21 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 "elf/x86-64.h"
36
37 #ifndef REGISTER_WARNINGS
38 #define REGISTER_WARNINGS 1
39 #endif
40
41 #ifndef INFER_ADDR_PREFIX
42 #define INFER_ADDR_PREFIX 1
43 #endif
44
45 #ifndef SCALE1_WHEN_NO_INDEX
46 /* Specifying a scale factor besides 1 when there is no index is
47 futile. eg. `mov (%ebx,2),%al' does exactly the same as
48 `mov (%ebx),%al'. To slavishly follow what the programmer
49 specified, set SCALE1_WHEN_NO_INDEX to 0. */
50 #define SCALE1_WHEN_NO_INDEX 1
51 #endif
52
53 #ifndef DEFAULT_ARCH
54 #define DEFAULT_ARCH "i386"
55 #endif
56
57 #ifndef INLINE
58 #if __GNUC__ >= 2
59 #define INLINE __inline__
60 #else
61 #define INLINE
62 #endif
63 #endif
64
65 static void set_code_flag (int);
66 static void set_16bit_gcc_code_flag (int);
67 static void set_intel_syntax (int);
68 static void set_cpu_arch (int);
69 #ifdef TE_PE
70 static void pe_directive_secrel (int);
71 #endif
72 static void signed_cons (int);
73 static char *output_invalid (int c);
74 static int i386_operand (char *);
75 static int i386_intel_operand (char *, int);
76 static const reg_entry *parse_register (char *, char **);
77 static char *parse_insn (char *, char *);
78 static char *parse_operands (char *, const char *);
79 static void swap_operands (void);
80 static void swap_2_operands (int, int);
81 static void optimize_imm (void);
82 static void optimize_disp (void);
83 static int match_template (void);
84 static int check_string (void);
85 static int process_suffix (void);
86 static int check_byte_reg (void);
87 static int check_long_reg (void);
88 static int check_qword_reg (void);
89 static int check_word_reg (void);
90 static int finalize_imm (void);
91 static int process_operands (void);
92 static const seg_entry *build_modrm_byte (void);
93 static void output_insn (void);
94 static void output_imm (fragS *, offsetT);
95 static void output_disp (fragS *, offsetT);
96 #ifndef I386COFF
97 static void s_bss (int);
98 #endif
99 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
100 static void handle_large_common (int small ATTRIBUTE_UNUSED);
101 #endif
102
103 static const char *default_arch = DEFAULT_ARCH;
104
105 /* 'md_assemble ()' gathers together information and puts it into a
106 i386_insn. */
107
108 union i386_op
109 {
110 expressionS *disps;
111 expressionS *imms;
112 const reg_entry *regs;
113 };
114
115 struct _i386_insn
116 {
117 /* TM holds the template for the insn were currently assembling. */
118 template tm;
119
120 /* SUFFIX holds the instruction mnemonic suffix if given.
121 (e.g. 'l' for 'movl') */
122 char suffix;
123
124 /* OPERANDS gives the number of given operands. */
125 unsigned int operands;
126
127 /* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number
128 of given register, displacement, memory operands and immediate
129 operands. */
130 unsigned int reg_operands, disp_operands, mem_operands, imm_operands;
131
132 /* TYPES [i] is the type (see above #defines) which tells us how to
133 use OP[i] for the corresponding operand. */
134 unsigned int types[MAX_OPERANDS];
135
136 /* Displacement expression, immediate expression, or register for each
137 operand. */
138 union i386_op op[MAX_OPERANDS];
139
140 /* Flags for operands. */
141 unsigned int flags[MAX_OPERANDS];
142 #define Operand_PCrel 1
143
144 /* Relocation type for operand */
145 enum bfd_reloc_code_real reloc[MAX_OPERANDS];
146
147 /* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode
148 the base index byte below. */
149 const reg_entry *base_reg;
150 const reg_entry *index_reg;
151 unsigned int log2_scale_factor;
152
153 /* SEG gives the seg_entries of this insn. They are zero unless
154 explicit segment overrides are given. */
155 const seg_entry *seg[2];
156
157 /* PREFIX holds all the given prefix opcodes (usually null).
158 PREFIXES is the number of prefix opcodes. */
159 unsigned int prefixes;
160 unsigned char prefix[MAX_PREFIXES];
161
162 /* RM and SIB are the modrm byte and the sib byte where the
163 addressing modes of this insn are encoded. */
164
165 modrm_byte rm;
166 rex_byte rex;
167 sib_byte sib;
168 };
169
170 typedef struct _i386_insn i386_insn;
171
172 /* List of chars besides those in app.c:symbol_chars that can start an
173 operand. Used to prevent the scrubber eating vital white-space. */
174 const char extra_symbol_chars[] = "*%-(["
175 #ifdef LEX_AT
176 "@"
177 #endif
178 #ifdef LEX_QM
179 "?"
180 #endif
181 ;
182
183 #if (defined (TE_I386AIX) \
184 || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) \
185 && !defined (TE_GNU) \
186 && !defined (TE_LINUX) \
187 && !defined (TE_NETWARE) \
188 && !defined (TE_FreeBSD) \
189 && !defined (TE_NetBSD)))
190 /* This array holds the chars that always start a comment. If the
191 pre-processor is disabled, these aren't very useful. The option
192 --divide will remove '/' from this list. */
193 const char *i386_comment_chars = "#/";
194 #define SVR4_COMMENT_CHARS 1
195 #define PREFIX_SEPARATOR '\\'
196
197 #else
198 const char *i386_comment_chars = "#";
199 #define PREFIX_SEPARATOR '/'
200 #endif
201
202 /* This array holds the chars that only start a comment at the beginning of
203 a line. If the line seems to have the form '# 123 filename'
204 .line and .file directives will appear in the pre-processed output.
205 Note that input_file.c hand checks for '#' at the beginning of the
206 first line of the input file. This is because the compiler outputs
207 #NO_APP at the beginning of its output.
208 Also note that comments started like this one will always work if
209 '/' isn't otherwise defined. */
210 const char line_comment_chars[] = "#/";
211
212 const char line_separator_chars[] = ";";
213
214 /* Chars that can be used to separate mant from exp in floating point
215 nums. */
216 const char EXP_CHARS[] = "eE";
217
218 /* Chars that mean this number is a floating point constant
219 As in 0f12.456
220 or 0d1.2345e12. */
221 const char FLT_CHARS[] = "fFdDxX";
222
223 /* Tables for lexical analysis. */
224 static char mnemonic_chars[256];
225 static char register_chars[256];
226 static char operand_chars[256];
227 static char identifier_chars[256];
228 static char digit_chars[256];
229
230 /* Lexical macros. */
231 #define is_mnemonic_char(x) (mnemonic_chars[(unsigned char) x])
232 #define is_operand_char(x) (operand_chars[(unsigned char) x])
233 #define is_register_char(x) (register_chars[(unsigned char) x])
234 #define is_space_char(x) ((x) == ' ')
235 #define is_identifier_char(x) (identifier_chars[(unsigned char) x])
236 #define is_digit_char(x) (digit_chars[(unsigned char) x])
237
238 /* All non-digit non-letter characters that may occur in an operand. */
239 static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
240
241 /* md_assemble() always leaves the strings it's passed unaltered. To
242 effect this we maintain a stack of saved characters that we've smashed
243 with '\0's (indicating end of strings for various sub-fields of the
244 assembler instruction). */
245 static char save_stack[32];
246 static char *save_stack_p;
247 #define END_STRING_AND_SAVE(s) \
248 do { *save_stack_p++ = *(s); *(s) = '\0'; } while (0)
249 #define RESTORE_END_STRING(s) \
250 do { *(s) = *--save_stack_p; } while (0)
251
252 /* The instruction we're assembling. */
253 static i386_insn i;
254
255 /* Possible templates for current insn. */
256 static const templates *current_templates;
257
258 /* Per instruction expressionS buffers: max displacements & immediates. */
259 static expressionS disp_expressions[MAX_MEMORY_OPERANDS];
260 static expressionS im_expressions[MAX_IMMEDIATE_OPERANDS];
261
262 /* Current operand we are working on. */
263 static int this_operand;
264
265 /* We support four different modes. FLAG_CODE variable is used to distinguish
266 these. */
267
268 enum flag_code {
269 CODE_32BIT,
270 CODE_16BIT,
271 CODE_64BIT };
272 #define NUM_FLAG_CODE ((int) CODE_64BIT + 1)
273
274 static enum flag_code flag_code;
275 static unsigned int object_64bit;
276 static int use_rela_relocations = 0;
277
278 /* The names used to print error messages. */
279 static const char *flag_code_names[] =
280 {
281 "32",
282 "16",
283 "64"
284 };
285
286 /* 1 for intel syntax,
287 0 if att syntax. */
288 static int intel_syntax = 0;
289
290 /* 1 if register prefix % not required. */
291 static int allow_naked_reg = 0;
292
293 /* Register prefix used for error message. */
294 static const char *register_prefix = "%";
295
296 /* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
297 leave, push, and pop instructions so that gcc has the same stack
298 frame as in 32 bit mode. */
299 static char stackop_size = '\0';
300
301 /* Non-zero to optimize code alignment. */
302 int optimize_align_code = 1;
303
304 /* Non-zero to quieten some warnings. */
305 static int quiet_warnings = 0;
306
307 /* CPU name. */
308 static const char *cpu_arch_name = NULL;
309 static const char *cpu_sub_arch_name = NULL;
310
311 /* CPU feature flags. */
312 static unsigned int cpu_arch_flags = CpuUnknownFlags | CpuNo64;
313
314 /* If we have selected a cpu we are generating instructions for. */
315 static int cpu_arch_tune_set = 0;
316
317 /* Cpu we are generating instructions for. */
318 static enum processor_type cpu_arch_tune = PROCESSOR_UNKNOWN;
319
320 /* CPU feature flags of cpu we are generating instructions for. */
321 static unsigned int cpu_arch_tune_flags = 0;
322
323 /* CPU instruction set architecture used. */
324 static enum processor_type cpu_arch_isa = PROCESSOR_UNKNOWN;
325
326 /* CPU feature flags of instruction set architecture used. */
327 static unsigned int cpu_arch_isa_flags = 0;
328
329 /* If set, conditional jumps are not automatically promoted to handle
330 larger than a byte offset. */
331 static unsigned int no_cond_jump_promotion = 0;
332
333 /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
334 static symbolS *GOT_symbol;
335
336 /* The dwarf2 return column, adjusted for 32 or 64 bit. */
337 unsigned int x86_dwarf2_return_column;
338
339 /* The dwarf2 data alignment, adjusted for 32 or 64 bit. */
340 int x86_cie_data_alignment;
341
342 /* Interface to relax_segment.
343 There are 3 major relax states for 386 jump insns because the
344 different types of jumps add different sizes to frags when we're
345 figuring out what sort of jump to choose to reach a given label. */
346
347 /* Types. */
348 #define UNCOND_JUMP 0
349 #define COND_JUMP 1
350 #define COND_JUMP86 2
351
352 /* Sizes. */
353 #define CODE16 1
354 #define SMALL 0
355 #define SMALL16 (SMALL | CODE16)
356 #define BIG 2
357 #define BIG16 (BIG | CODE16)
358
359 #ifndef INLINE
360 #ifdef __GNUC__
361 #define INLINE __inline__
362 #else
363 #define INLINE
364 #endif
365 #endif
366
367 #define ENCODE_RELAX_STATE(type, size) \
368 ((relax_substateT) (((type) << 2) | (size)))
369 #define TYPE_FROM_RELAX_STATE(s) \
370 ((s) >> 2)
371 #define DISP_SIZE_FROM_RELAX_STATE(s) \
372 ((((s) & 3) == BIG ? 4 : (((s) & 3) == BIG16 ? 2 : 1)))
373
374 /* This table is used by relax_frag to promote short jumps to long
375 ones where necessary. SMALL (short) jumps may be promoted to BIG
376 (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
377 don't allow a short jump in a 32 bit code segment to be promoted to
378 a 16 bit offset jump because it's slower (requires data size
379 prefix), and doesn't work, unless the destination is in the bottom
380 64k of the code segment (The top 16 bits of eip are zeroed). */
381
382 const relax_typeS md_relax_table[] =
383 {
384 /* The fields are:
385 1) most positive reach of this state,
386 2) most negative reach of this state,
387 3) how many bytes this mode will have in the variable part of the frag
388 4) which index into the table to try if we can't fit into this one. */
389
390 /* UNCOND_JUMP states. */
391 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
392 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
393 /* dword jmp adds 4 bytes to frag:
394 0 extra opcode bytes, 4 displacement bytes. */
395 {0, 0, 4, 0},
396 /* word jmp adds 2 byte2 to frag:
397 0 extra opcode bytes, 2 displacement bytes. */
398 {0, 0, 2, 0},
399
400 /* COND_JUMP states. */
401 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
402 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
403 /* dword conditionals adds 5 bytes to frag:
404 1 extra opcode byte, 4 displacement bytes. */
405 {0, 0, 5, 0},
406 /* word conditionals add 3 bytes to frag:
407 1 extra opcode byte, 2 displacement bytes. */
408 {0, 0, 3, 0},
409
410 /* COND_JUMP86 states. */
411 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG)},
412 {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG16)},
413 /* dword conditionals adds 5 bytes to frag:
414 1 extra opcode byte, 4 displacement bytes. */
415 {0, 0, 5, 0},
416 /* word conditionals add 4 bytes to frag:
417 1 displacement byte and a 3 byte long branch insn. */
418 {0, 0, 4, 0}
419 };
420
421 static const arch_entry cpu_arch[] =
422 {
423 {"generic32", PROCESSOR_GENERIC32,
424 Cpu186|Cpu286|Cpu386},
425 {"generic64", PROCESSOR_GENERIC64,
426 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
427 |CpuMMX2|CpuSSE|CpuSSE2},
428 {"i8086", PROCESSOR_UNKNOWN,
429 0},
430 {"i186", PROCESSOR_UNKNOWN,
431 Cpu186},
432 {"i286", PROCESSOR_UNKNOWN,
433 Cpu186|Cpu286},
434 {"i386", PROCESSOR_GENERIC32,
435 Cpu186|Cpu286|Cpu386},
436 {"i486", PROCESSOR_I486,
437 Cpu186|Cpu286|Cpu386|Cpu486},
438 {"i586", PROCESSOR_PENTIUM,
439 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586},
440 {"i686", PROCESSOR_PENTIUMPRO,
441 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686},
442 {"pentium", PROCESSOR_PENTIUM,
443 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586},
444 {"pentiumpro",PROCESSOR_PENTIUMPRO,
445 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686},
446 {"pentiumii", PROCESSOR_PENTIUMPRO,
447 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX},
448 {"pentiumiii",PROCESSOR_PENTIUMPRO,
449 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuMMX2|CpuSSE},
450 {"pentium4", PROCESSOR_PENTIUM4,
451 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
452 |CpuMMX2|CpuSSE|CpuSSE2},
453 {"prescott", PROCESSOR_NOCONA,
454 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
455 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
456 {"nocona", PROCESSOR_NOCONA,
457 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
458 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
459 {"yonah", PROCESSOR_CORE,
460 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
461 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
462 {"core", PROCESSOR_CORE,
463 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
464 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
465 {"merom", PROCESSOR_CORE2,
466 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
467 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3},
468 {"core2", PROCESSOR_CORE2,
469 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
470 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3},
471 {"k6", PROCESSOR_K6,
472 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX},
473 {"k6_2", PROCESSOR_K6,
474 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow},
475 {"athlon", PROCESSOR_ATHLON,
476 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
477 |CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA},
478 {"sledgehammer", PROCESSOR_K8,
479 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
480 |CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2},
481 {"opteron", PROCESSOR_K8,
482 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
483 |CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2},
484 {"k8", PROCESSOR_K8,
485 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
486 |CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2},
487 {"amdfam10", PROCESSOR_AMDFAM10,
488 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuSledgehammer
489 |CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2|CpuSSE3|CpuSSE4a
490 |CpuABM},
491 {".mmx", PROCESSOR_UNKNOWN,
492 CpuMMX},
493 {".sse", PROCESSOR_UNKNOWN,
494 CpuMMX|CpuMMX2|CpuSSE},
495 {".sse2", PROCESSOR_UNKNOWN,
496 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2},
497 {".sse3", PROCESSOR_UNKNOWN,
498 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
499 {".ssse3", PROCESSOR_UNKNOWN,
500 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3},
501 {".3dnow", PROCESSOR_UNKNOWN,
502 CpuMMX|Cpu3dnow},
503 {".3dnowa", PROCESSOR_UNKNOWN,
504 CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA},
505 {".padlock", PROCESSOR_UNKNOWN,
506 CpuPadLock},
507 {".pacifica", PROCESSOR_UNKNOWN,
508 CpuSVME},
509 {".svme", PROCESSOR_UNKNOWN,
510 CpuSVME},
511 {".sse4a", PROCESSOR_UNKNOWN,
512 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSE4a},
513 {".abm", PROCESSOR_UNKNOWN,
514 CpuABM}
515 };
516
517 const pseudo_typeS md_pseudo_table[] =
518 {
519 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
520 {"align", s_align_bytes, 0},
521 #else
522 {"align", s_align_ptwo, 0},
523 #endif
524 {"arch", set_cpu_arch, 0},
525 #ifndef I386COFF
526 {"bss", s_bss, 0},
527 #endif
528 {"ffloat", float_cons, 'f'},
529 {"dfloat", float_cons, 'd'},
530 {"tfloat", float_cons, 'x'},
531 {"value", cons, 2},
532 {"slong", signed_cons, 4},
533 {"noopt", s_ignore, 0},
534 {"optim", s_ignore, 0},
535 {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
536 {"code16", set_code_flag, CODE_16BIT},
537 {"code32", set_code_flag, CODE_32BIT},
538 {"code64", set_code_flag, CODE_64BIT},
539 {"intel_syntax", set_intel_syntax, 1},
540 {"att_syntax", set_intel_syntax, 0},
541 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
542 {"largecomm", handle_large_common, 0},
543 #else
544 {"file", (void (*) (int)) dwarf2_directive_file, 0},
545 {"loc", dwarf2_directive_loc, 0},
546 {"loc_mark_labels", dwarf2_directive_loc_mark_labels, 0},
547 #endif
548 #ifdef TE_PE
549 {"secrel32", pe_directive_secrel, 0},
550 #endif
551 {0, 0, 0}
552 };
553
554 /* For interface with expression (). */
555 extern char *input_line_pointer;
556
557 /* Hash table for instruction mnemonic lookup. */
558 static struct hash_control *op_hash;
559
560 /* Hash table for register lookup. */
561 static struct hash_control *reg_hash;
562 \f
563 void
564 i386_align_code (fragS *fragP, int count)
565 {
566 /* Various efficient no-op patterns for aligning code labels.
567 Note: Don't try to assemble the instructions in the comments.
568 0L and 0w are not legal. */
569 static const char f32_1[] =
570 {0x90}; /* nop */
571 static const char f32_2[] =
572 {0x66,0x90}; /* xchg %ax,%ax */
573 static const char f32_3[] =
574 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
575 static const char f32_4[] =
576 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
577 static const char f32_5[] =
578 {0x90, /* nop */
579 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
580 static const char f32_6[] =
581 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
582 static const char f32_7[] =
583 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
584 static const char f32_8[] =
585 {0x90, /* nop */
586 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
587 static const char f32_9[] =
588 {0x89,0xf6, /* movl %esi,%esi */
589 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
590 static const char f32_10[] =
591 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
592 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
593 static const char f32_11[] =
594 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
595 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
596 static const char f32_12[] =
597 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
598 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
599 static const char f32_13[] =
600 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
601 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
602 static const char f32_14[] =
603 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
604 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
605 static const char f32_15[] =
606 {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
607 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
608 static const char f16_3[] =
609 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
610 static const char f16_4[] =
611 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
612 static const char f16_5[] =
613 {0x90, /* nop */
614 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
615 static const char f16_6[] =
616 {0x89,0xf6, /* mov %si,%si */
617 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
618 static const char f16_7[] =
619 {0x8d,0x74,0x00, /* lea 0(%si),%si */
620 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
621 static const char f16_8[] =
622 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
623 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
624 static const char *const f32_patt[] = {
625 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
626 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
627 };
628 static const char *const f16_patt[] = {
629 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
630 f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
631 };
632 /* nopl (%[re]ax) */
633 static const char alt_3[] =
634 {0x0f,0x1f,0x00};
635 /* nopl 0(%[re]ax) */
636 static const char alt_4[] =
637 {0x0f,0x1f,0x40,0x00};
638 /* nopl 0(%[re]ax,%[re]ax,1) */
639 static const char alt_5[] =
640 {0x0f,0x1f,0x44,0x00,0x00};
641 /* nopw 0(%[re]ax,%[re]ax,1) */
642 static const char alt_6[] =
643 {0x66,0x0f,0x1f,0x44,0x00,0x00};
644 /* nopl 0L(%[re]ax) */
645 static const char alt_7[] =
646 {0x0f,0x1f,0x80,0x00,0x00,0x00,0x00};
647 /* nopl 0L(%[re]ax,%[re]ax,1) */
648 static const char alt_8[] =
649 {0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
650 /* nopw 0L(%[re]ax,%[re]ax,1) */
651 static const char alt_9[] =
652 {0x66,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
653 /* nopw %cs:0L(%[re]ax,%[re]ax,1) */
654 static const char alt_10[] =
655 {0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
656 /* data16
657 nopw %cs:0L(%[re]ax,%[re]ax,1) */
658 static const char alt_long_11[] =
659 {0x66,
660 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
661 /* data16
662 data16
663 nopw %cs:0L(%[re]ax,%[re]ax,1) */
664 static const char alt_long_12[] =
665 {0x66,
666 0x66,
667 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
668 /* data16
669 data16
670 data16
671 nopw %cs:0L(%[re]ax,%[re]ax,1) */
672 static const char alt_long_13[] =
673 {0x66,
674 0x66,
675 0x66,
676 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
677 /* data16
678 data16
679 data16
680 data16
681 nopw %cs:0L(%[re]ax,%[re]ax,1) */
682 static const char alt_long_14[] =
683 {0x66,
684 0x66,
685 0x66,
686 0x66,
687 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
688 /* data16
689 data16
690 data16
691 data16
692 data16
693 nopw %cs:0L(%[re]ax,%[re]ax,1) */
694 static const char alt_long_15[] =
695 {0x66,
696 0x66,
697 0x66,
698 0x66,
699 0x66,
700 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
701 /* nopl 0(%[re]ax,%[re]ax,1)
702 nopw 0(%[re]ax,%[re]ax,1) */
703 static const char alt_short_11[] =
704 {0x0f,0x1f,0x44,0x00,0x00,
705 0x66,0x0f,0x1f,0x44,0x00,0x00};
706 /* nopw 0(%[re]ax,%[re]ax,1)
707 nopw 0(%[re]ax,%[re]ax,1) */
708 static const char alt_short_12[] =
709 {0x66,0x0f,0x1f,0x44,0x00,0x00,
710 0x66,0x0f,0x1f,0x44,0x00,0x00};
711 /* nopw 0(%[re]ax,%[re]ax,1)
712 nopl 0L(%[re]ax) */
713 static const char alt_short_13[] =
714 {0x66,0x0f,0x1f,0x44,0x00,0x00,
715 0x0f,0x1f,0x80,0x00,0x00,0x00,0x00};
716 /* nopl 0L(%[re]ax)
717 nopl 0L(%[re]ax) */
718 static const char alt_short_14[] =
719 {0x0f,0x1f,0x80,0x00,0x00,0x00,0x00,
720 0x0f,0x1f,0x80,0x00,0x00,0x00,0x00};
721 /* nopl 0L(%[re]ax)
722 nopl 0L(%[re]ax,%[re]ax,1) */
723 static const char alt_short_15[] =
724 {0x0f,0x1f,0x80,0x00,0x00,0x00,0x00,
725 0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
726 static const char *const alt_short_patt[] = {
727 f32_1, f32_2, alt_3, alt_4, alt_5, alt_6, alt_7, alt_8,
728 alt_9, alt_10, alt_short_11, alt_short_12, alt_short_13,
729 alt_short_14, alt_short_15
730 };
731 static const char *const alt_long_patt[] = {
732 f32_1, f32_2, alt_3, alt_4, alt_5, alt_6, alt_7, alt_8,
733 alt_9, alt_10, alt_long_11, alt_long_12, alt_long_13,
734 alt_long_14, alt_long_15
735 };
736
737 if (count <= 0 || count > 15)
738 return;
739
740 /* We need to decide which NOP sequence to use for 32bit and
741 64bit. When -mtune= is used:
742
743 1. For PROCESSOR_I486, PROCESSOR_PENTIUM and PROCESSOR_GENERIC32,
744 f32_patt will be used.
745 2. For PROCESSOR_K8 and PROCESSOR_AMDFAM10 in 64bit, NOPs with
746 0x66 prefix will be used.
747 3. For PROCESSOR_CORE2, alt_long_patt will be used.
748 4. For PROCESSOR_PENTIUMPRO, PROCESSOR_PENTIUM4, PROCESSOR_NOCONA,
749 PROCESSOR_CORE, PROCESSOR_CORE2, PROCESSOR_K6, PROCESSOR_ATHLON
750 and PROCESSOR_GENERIC64, alt_short_patt will be used.
751
752 When -mtune= isn't used, alt_short_patt will be used if
753 cpu_arch_isa_flags has Cpu686. Otherwise, f32_patt will be used.
754
755 When -march= or .arch is used, we can't use anything beyond
756 cpu_arch_isa_flags. */
757
758 if (flag_code == CODE_16BIT)
759 {
760 memcpy (fragP->fr_literal + fragP->fr_fix,
761 f16_patt[count - 1], count);
762 if (count > 8)
763 /* Adjust jump offset. */
764 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
765 }
766 else if (flag_code == CODE_64BIT && cpu_arch_tune == PROCESSOR_K8)
767 {
768 int i;
769 int nnops = (count + 3) / 4;
770 int len = count / nnops;
771 int remains = count - nnops * len;
772 int pos = 0;
773
774 /* The recommended way to pad 64bit code is to use NOPs preceded
775 by maximally four 0x66 prefixes. Balance the size of nops. */
776 for (i = 0; i < remains; i++)
777 {
778 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len);
779 fragP->fr_literal[fragP->fr_fix + pos + len] = 0x90;
780 pos += len + 1;
781 }
782 for (; i < nnops; i++)
783 {
784 memset (fragP->fr_literal + fragP->fr_fix + pos, 0x66, len - 1);
785 fragP->fr_literal[fragP->fr_fix + pos + len - 1] = 0x90;
786 pos += len;
787 }
788 }
789 else
790 {
791 const char *const *patt = NULL;
792
793 if (cpu_arch_isa == PROCESSOR_UNKNOWN)
794 {
795 /* PROCESSOR_UNKNOWN means that all ISAs may be used. */
796 switch (cpu_arch_tune)
797 {
798 case PROCESSOR_UNKNOWN:
799 /* We use cpu_arch_isa_flags to check if we SHOULD
800 optimize for Cpu686. */
801 if ((cpu_arch_isa_flags & Cpu686) != 0)
802 patt = alt_short_patt;
803 else
804 patt = f32_patt;
805 break;
806 case PROCESSOR_CORE2:
807 patt = alt_long_patt;
808 break;
809 case PROCESSOR_PENTIUMPRO:
810 case PROCESSOR_PENTIUM4:
811 case PROCESSOR_NOCONA:
812 case PROCESSOR_CORE:
813 case PROCESSOR_K6:
814 case PROCESSOR_ATHLON:
815 case PROCESSOR_K8:
816 case PROCESSOR_GENERIC64:
817 case PROCESSOR_AMDFAM10:
818 patt = alt_short_patt;
819 break;
820 case PROCESSOR_I486:
821 case PROCESSOR_PENTIUM:
822 case PROCESSOR_GENERIC32:
823 patt = f32_patt;
824 break;
825 }
826 }
827 else
828 {
829 switch (cpu_arch_tune)
830 {
831 case PROCESSOR_UNKNOWN:
832 /* When cpu_arch_isa is net, cpu_arch_tune shouldn't be
833 PROCESSOR_UNKNOWN. */
834 abort ();
835 break;
836
837 case PROCESSOR_I486:
838 case PROCESSOR_PENTIUM:
839 case PROCESSOR_PENTIUMPRO:
840 case PROCESSOR_PENTIUM4:
841 case PROCESSOR_NOCONA:
842 case PROCESSOR_CORE:
843 case PROCESSOR_K6:
844 case PROCESSOR_ATHLON:
845 case PROCESSOR_K8:
846 case PROCESSOR_AMDFAM10:
847 case PROCESSOR_GENERIC32:
848 /* We use cpu_arch_isa_flags to check if we CAN optimize
849 for Cpu686. */
850 if ((cpu_arch_isa_flags & Cpu686) != 0)
851 patt = alt_short_patt;
852 else
853 patt = f32_patt;
854 break;
855 case PROCESSOR_CORE2:
856 if ((cpu_arch_isa_flags & Cpu686) != 0)
857 patt = alt_long_patt;
858 else
859 patt = f32_patt;
860 break;
861 case PROCESSOR_GENERIC64:
862 patt = alt_short_patt;
863 break;
864 }
865 }
866
867 memcpy (fragP->fr_literal + fragP->fr_fix,
868 patt[count - 1], count);
869 }
870 fragP->fr_var = count;
871 }
872
873 static INLINE unsigned int
874 mode_from_disp_size (unsigned int t)
875 {
876 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
877 }
878
879 static INLINE int
880 fits_in_signed_byte (offsetT num)
881 {
882 return (num >= -128) && (num <= 127);
883 }
884
885 static INLINE int
886 fits_in_unsigned_byte (offsetT num)
887 {
888 return (num & 0xff) == num;
889 }
890
891 static INLINE int
892 fits_in_unsigned_word (offsetT num)
893 {
894 return (num & 0xffff) == num;
895 }
896
897 static INLINE int
898 fits_in_signed_word (offsetT num)
899 {
900 return (-32768 <= num) && (num <= 32767);
901 }
902
903 static INLINE int
904 fits_in_signed_long (offsetT num ATTRIBUTE_UNUSED)
905 {
906 #ifndef BFD64
907 return 1;
908 #else
909 return (!(((offsetT) -1 << 31) & num)
910 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
911 #endif
912 } /* fits_in_signed_long() */
913
914 static INLINE int
915 fits_in_unsigned_long (offsetT num ATTRIBUTE_UNUSED)
916 {
917 #ifndef BFD64
918 return 1;
919 #else
920 return (num & (((offsetT) 2 << 31) - 1)) == num;
921 #endif
922 } /* fits_in_unsigned_long() */
923
924 static unsigned int
925 smallest_imm_type (offsetT num)
926 {
927 if (cpu_arch_flags != (Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64))
928 {
929 /* This code is disabled on the 486 because all the Imm1 forms
930 in the opcode table are slower on the i486. They're the
931 versions with the implicitly specified single-position
932 displacement, which has another syntax if you really want to
933 use that form. */
934 if (num == 1)
935 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
936 }
937 return (fits_in_signed_byte (num)
938 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
939 : fits_in_unsigned_byte (num)
940 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
941 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
942 ? (Imm16 | Imm32 | Imm32S | Imm64)
943 : fits_in_signed_long (num)
944 ? (Imm32 | Imm32S | Imm64)
945 : fits_in_unsigned_long (num)
946 ? (Imm32 | Imm64)
947 : Imm64);
948 }
949
950 static offsetT
951 offset_in_range (offsetT val, int size)
952 {
953 addressT mask;
954
955 switch (size)
956 {
957 case 1: mask = ((addressT) 1 << 8) - 1; break;
958 case 2: mask = ((addressT) 1 << 16) - 1; break;
959 case 4: mask = ((addressT) 2 << 31) - 1; break;
960 #ifdef BFD64
961 case 8: mask = ((addressT) 2 << 63) - 1; break;
962 #endif
963 default: abort ();
964 }
965
966 /* If BFD64, sign extend val. */
967 if (!use_rela_relocations)
968 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
969 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
970
971 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
972 {
973 char buf1[40], buf2[40];
974
975 sprint_value (buf1, val);
976 sprint_value (buf2, val & mask);
977 as_warn (_("%s shortened to %s"), buf1, buf2);
978 }
979 return val & mask;
980 }
981
982 /* Returns 0 if attempting to add a prefix where one from the same
983 class already exists, 1 if non rep/repne added, 2 if rep/repne
984 added. */
985 static int
986 add_prefix (unsigned int prefix)
987 {
988 int ret = 1;
989 unsigned int q;
990
991 if (prefix >= REX_OPCODE && prefix < REX_OPCODE + 16
992 && flag_code == CODE_64BIT)
993 {
994 if ((i.prefix[REX_PREFIX] & prefix & REX_MODE64)
995 || ((i.prefix[REX_PREFIX] & (REX_EXTX | REX_EXTY | REX_EXTZ))
996 && (prefix & (REX_EXTX | REX_EXTY | REX_EXTZ))))
997 ret = 0;
998 q = REX_PREFIX;
999 }
1000 else
1001 {
1002 switch (prefix)
1003 {
1004 default:
1005 abort ();
1006
1007 case CS_PREFIX_OPCODE:
1008 case DS_PREFIX_OPCODE:
1009 case ES_PREFIX_OPCODE:
1010 case FS_PREFIX_OPCODE:
1011 case GS_PREFIX_OPCODE:
1012 case SS_PREFIX_OPCODE:
1013 q = SEG_PREFIX;
1014 break;
1015
1016 case REPNE_PREFIX_OPCODE:
1017 case REPE_PREFIX_OPCODE:
1018 ret = 2;
1019 /* fall thru */
1020 case LOCK_PREFIX_OPCODE:
1021 q = LOCKREP_PREFIX;
1022 break;
1023
1024 case FWAIT_OPCODE:
1025 q = WAIT_PREFIX;
1026 break;
1027
1028 case ADDR_PREFIX_OPCODE:
1029 q = ADDR_PREFIX;
1030 break;
1031
1032 case DATA_PREFIX_OPCODE:
1033 q = DATA_PREFIX;
1034 break;
1035 }
1036 if (i.prefix[q] != 0)
1037 ret = 0;
1038 }
1039
1040 if (ret)
1041 {
1042 if (!i.prefix[q])
1043 ++i.prefixes;
1044 i.prefix[q] |= prefix;
1045 }
1046 else
1047 as_bad (_("same type of prefix used twice"));
1048
1049 return ret;
1050 }
1051
1052 static void
1053 set_code_flag (int value)
1054 {
1055 flag_code = value;
1056 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
1057 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
1058 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
1059 {
1060 as_bad (_("64bit mode not supported on this CPU."));
1061 }
1062 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
1063 {
1064 as_bad (_("32bit mode not supported on this CPU."));
1065 }
1066 stackop_size = '\0';
1067 }
1068
1069 static void
1070 set_16bit_gcc_code_flag (int new_code_flag)
1071 {
1072 flag_code = new_code_flag;
1073 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
1074 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
1075 stackop_size = LONG_MNEM_SUFFIX;
1076 }
1077
1078 static void
1079 set_intel_syntax (int syntax_flag)
1080 {
1081 /* Find out if register prefixing is specified. */
1082 int ask_naked_reg = 0;
1083
1084 SKIP_WHITESPACE ();
1085 if (!is_end_of_line[(unsigned char) *input_line_pointer])
1086 {
1087 char *string = input_line_pointer;
1088 int e = get_symbol_end ();
1089
1090 if (strcmp (string, "prefix") == 0)
1091 ask_naked_reg = 1;
1092 else if (strcmp (string, "noprefix") == 0)
1093 ask_naked_reg = -1;
1094 else
1095 as_bad (_("bad argument to syntax directive."));
1096 *input_line_pointer = e;
1097 }
1098 demand_empty_rest_of_line ();
1099
1100 intel_syntax = syntax_flag;
1101
1102 if (ask_naked_reg == 0)
1103 allow_naked_reg = (intel_syntax
1104 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
1105 else
1106 allow_naked_reg = (ask_naked_reg < 0);
1107
1108 identifier_chars['%'] = intel_syntax && allow_naked_reg ? '%' : 0;
1109 identifier_chars['$'] = intel_syntax ? '$' : 0;
1110 register_prefix = allow_naked_reg ? "" : "%";
1111 }
1112
1113 static void
1114 set_cpu_arch (int dummy ATTRIBUTE_UNUSED)
1115 {
1116 SKIP_WHITESPACE ();
1117
1118 if (!is_end_of_line[(unsigned char) *input_line_pointer])
1119 {
1120 char *string = input_line_pointer;
1121 int e = get_symbol_end ();
1122 unsigned int i;
1123
1124 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
1125 {
1126 if (strcmp (string, cpu_arch[i].name) == 0)
1127 {
1128 if (*string != '.')
1129 {
1130 cpu_arch_name = cpu_arch[i].name;
1131 cpu_sub_arch_name = NULL;
1132 cpu_arch_flags = (cpu_arch[i].flags
1133 | (flag_code == CODE_64BIT
1134 ? Cpu64 : CpuNo64));
1135 cpu_arch_isa = cpu_arch[i].type;
1136 cpu_arch_isa_flags = cpu_arch[i].flags;
1137 if (!cpu_arch_tune_set)
1138 {
1139 cpu_arch_tune = cpu_arch_isa;
1140 cpu_arch_tune_flags = cpu_arch_isa_flags;
1141 }
1142 break;
1143 }
1144 if ((cpu_arch_flags | cpu_arch[i].flags) != cpu_arch_flags)
1145 {
1146 cpu_sub_arch_name = cpu_arch[i].name;
1147 cpu_arch_flags |= cpu_arch[i].flags;
1148 }
1149 *input_line_pointer = e;
1150 demand_empty_rest_of_line ();
1151 return;
1152 }
1153 }
1154 if (i >= ARRAY_SIZE (cpu_arch))
1155 as_bad (_("no such architecture: `%s'"), string);
1156
1157 *input_line_pointer = e;
1158 }
1159 else
1160 as_bad (_("missing cpu architecture"));
1161
1162 no_cond_jump_promotion = 0;
1163 if (*input_line_pointer == ','
1164 && !is_end_of_line[(unsigned char) input_line_pointer[1]])
1165 {
1166 char *string = ++input_line_pointer;
1167 int e = get_symbol_end ();
1168
1169 if (strcmp (string, "nojumps") == 0)
1170 no_cond_jump_promotion = 1;
1171 else if (strcmp (string, "jumps") == 0)
1172 ;
1173 else
1174 as_bad (_("no such architecture modifier: `%s'"), string);
1175
1176 *input_line_pointer = e;
1177 }
1178
1179 demand_empty_rest_of_line ();
1180 }
1181
1182 unsigned long
1183 i386_mach ()
1184 {
1185 if (!strcmp (default_arch, "x86_64"))
1186 return bfd_mach_x86_64;
1187 else if (!strcmp (default_arch, "i386"))
1188 return bfd_mach_i386_i386;
1189 else
1190 as_fatal (_("Unknown architecture"));
1191 }
1192 \f
1193 void
1194 md_begin ()
1195 {
1196 const char *hash_err;
1197
1198 /* Initialize op_hash hash table. */
1199 op_hash = hash_new ();
1200
1201 {
1202 const template *optab;
1203 templates *core_optab;
1204
1205 /* Setup for loop. */
1206 optab = i386_optab;
1207 core_optab = (templates *) xmalloc (sizeof (templates));
1208 core_optab->start = optab;
1209
1210 while (1)
1211 {
1212 ++optab;
1213 if (optab->name == NULL
1214 || strcmp (optab->name, (optab - 1)->name) != 0)
1215 {
1216 /* different name --> ship out current template list;
1217 add to hash table; & begin anew. */
1218 core_optab->end = optab;
1219 hash_err = hash_insert (op_hash,
1220 (optab - 1)->name,
1221 (PTR) core_optab);
1222 if (hash_err)
1223 {
1224 as_fatal (_("Internal Error: Can't hash %s: %s"),
1225 (optab - 1)->name,
1226 hash_err);
1227 }
1228 if (optab->name == NULL)
1229 break;
1230 core_optab = (templates *) xmalloc (sizeof (templates));
1231 core_optab->start = optab;
1232 }
1233 }
1234 }
1235
1236 /* Initialize reg_hash hash table. */
1237 reg_hash = hash_new ();
1238 {
1239 const reg_entry *regtab;
1240 unsigned int regtab_size = i386_regtab_size;
1241
1242 for (regtab = i386_regtab; regtab_size--; regtab++)
1243 {
1244 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
1245 if (hash_err)
1246 as_fatal (_("Internal Error: Can't hash %s: %s"),
1247 regtab->reg_name,
1248 hash_err);
1249 }
1250 }
1251
1252 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
1253 {
1254 int c;
1255 char *p;
1256
1257 for (c = 0; c < 256; c++)
1258 {
1259 if (ISDIGIT (c))
1260 {
1261 digit_chars[c] = c;
1262 mnemonic_chars[c] = c;
1263 register_chars[c] = c;
1264 operand_chars[c] = c;
1265 }
1266 else if (ISLOWER (c))
1267 {
1268 mnemonic_chars[c] = c;
1269 register_chars[c] = c;
1270 operand_chars[c] = c;
1271 }
1272 else if (ISUPPER (c))
1273 {
1274 mnemonic_chars[c] = TOLOWER (c);
1275 register_chars[c] = mnemonic_chars[c];
1276 operand_chars[c] = c;
1277 }
1278
1279 if (ISALPHA (c) || ISDIGIT (c))
1280 identifier_chars[c] = c;
1281 else if (c >= 128)
1282 {
1283 identifier_chars[c] = c;
1284 operand_chars[c] = c;
1285 }
1286 }
1287
1288 #ifdef LEX_AT
1289 identifier_chars['@'] = '@';
1290 #endif
1291 #ifdef LEX_QM
1292 identifier_chars['?'] = '?';
1293 operand_chars['?'] = '?';
1294 #endif
1295 digit_chars['-'] = '-';
1296 mnemonic_chars['-'] = '-';
1297 identifier_chars['_'] = '_';
1298 identifier_chars['.'] = '.';
1299
1300 for (p = operand_special_chars; *p != '\0'; p++)
1301 operand_chars[(unsigned char) *p] = *p;
1302 }
1303
1304 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1305 if (IS_ELF)
1306 {
1307 record_alignment (text_section, 2);
1308 record_alignment (data_section, 2);
1309 record_alignment (bss_section, 2);
1310 }
1311 #endif
1312
1313 if (flag_code == CODE_64BIT)
1314 {
1315 x86_dwarf2_return_column = 16;
1316 x86_cie_data_alignment = -8;
1317 }
1318 else
1319 {
1320 x86_dwarf2_return_column = 8;
1321 x86_cie_data_alignment = -4;
1322 }
1323 }
1324
1325 void
1326 i386_print_statistics (FILE *file)
1327 {
1328 hash_print_statistics (file, "i386 opcode", op_hash);
1329 hash_print_statistics (file, "i386 register", reg_hash);
1330 }
1331 \f
1332 #ifdef DEBUG386
1333
1334 /* Debugging routines for md_assemble. */
1335 static void pte (template *);
1336 static void pt (unsigned int);
1337 static void pe (expressionS *);
1338 static void ps (symbolS *);
1339
1340 static void
1341 pi (char *line, i386_insn *x)
1342 {
1343 unsigned int i;
1344
1345 fprintf (stdout, "%s: template ", line);
1346 pte (&x->tm);
1347 fprintf (stdout, " address: base %s index %s scale %x\n",
1348 x->base_reg ? x->base_reg->reg_name : "none",
1349 x->index_reg ? x->index_reg->reg_name : "none",
1350 x->log2_scale_factor);
1351 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
1352 x->rm.mode, x->rm.reg, x->rm.regmem);
1353 fprintf (stdout, " sib: base %x index %x scale %x\n",
1354 x->sib.base, x->sib.index, x->sib.scale);
1355 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
1356 (x->rex & REX_MODE64) != 0,
1357 (x->rex & REX_EXTX) != 0,
1358 (x->rex & REX_EXTY) != 0,
1359 (x->rex & REX_EXTZ) != 0);
1360 for (i = 0; i < x->operands; i++)
1361 {
1362 fprintf (stdout, " #%d: ", i + 1);
1363 pt (x->types[i]);
1364 fprintf (stdout, "\n");
1365 if (x->types[i]
1366 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
1367 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
1368 if (x->types[i] & Imm)
1369 pe (x->op[i].imms);
1370 if (x->types[i] & Disp)
1371 pe (x->op[i].disps);
1372 }
1373 }
1374
1375 static void
1376 pte (template *t)
1377 {
1378 unsigned int i;
1379 fprintf (stdout, " %d operands ", t->operands);
1380 fprintf (stdout, "opcode %x ", t->base_opcode);
1381 if (t->extension_opcode != None)
1382 fprintf (stdout, "ext %x ", t->extension_opcode);
1383 if (t->opcode_modifier & D)
1384 fprintf (stdout, "D");
1385 if (t->opcode_modifier & W)
1386 fprintf (stdout, "W");
1387 fprintf (stdout, "\n");
1388 for (i = 0; i < t->operands; i++)
1389 {
1390 fprintf (stdout, " #%d type ", i + 1);
1391 pt (t->operand_types[i]);
1392 fprintf (stdout, "\n");
1393 }
1394 }
1395
1396 static void
1397 pe (expressionS *e)
1398 {
1399 fprintf (stdout, " operation %d\n", e->X_op);
1400 fprintf (stdout, " add_number %ld (%lx)\n",
1401 (long) e->X_add_number, (long) e->X_add_number);
1402 if (e->X_add_symbol)
1403 {
1404 fprintf (stdout, " add_symbol ");
1405 ps (e->X_add_symbol);
1406 fprintf (stdout, "\n");
1407 }
1408 if (e->X_op_symbol)
1409 {
1410 fprintf (stdout, " op_symbol ");
1411 ps (e->X_op_symbol);
1412 fprintf (stdout, "\n");
1413 }
1414 }
1415
1416 static void
1417 ps (symbolS *s)
1418 {
1419 fprintf (stdout, "%s type %s%s",
1420 S_GET_NAME (s),
1421 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
1422 segment_name (S_GET_SEGMENT (s)));
1423 }
1424
1425 static struct type_name
1426 {
1427 unsigned int mask;
1428 char *tname;
1429 }
1430 const type_names[] =
1431 {
1432 { Reg8, "r8" },
1433 { Reg16, "r16" },
1434 { Reg32, "r32" },
1435 { Reg64, "r64" },
1436 { Imm8, "i8" },
1437 { Imm8S, "i8s" },
1438 { Imm16, "i16" },
1439 { Imm32, "i32" },
1440 { Imm32S, "i32s" },
1441 { Imm64, "i64" },
1442 { Imm1, "i1" },
1443 { BaseIndex, "BaseIndex" },
1444 { Disp8, "d8" },
1445 { Disp16, "d16" },
1446 { Disp32, "d32" },
1447 { Disp32S, "d32s" },
1448 { Disp64, "d64" },
1449 { InOutPortReg, "InOutPortReg" },
1450 { ShiftCount, "ShiftCount" },
1451 { Control, "control reg" },
1452 { Test, "test reg" },
1453 { Debug, "debug reg" },
1454 { FloatReg, "FReg" },
1455 { FloatAcc, "FAcc" },
1456 { SReg2, "SReg2" },
1457 { SReg3, "SReg3" },
1458 { Acc, "Acc" },
1459 { JumpAbsolute, "Jump Absolute" },
1460 { RegMMX, "rMMX" },
1461 { RegXMM, "rXMM" },
1462 { EsSeg, "es" },
1463 { 0, "" }
1464 };
1465
1466 static void
1467 pt (t)
1468 unsigned int t;
1469 {
1470 const struct type_name *ty;
1471
1472 for (ty = type_names; ty->mask; ty++)
1473 if (t & ty->mask)
1474 fprintf (stdout, "%s, ", ty->tname);
1475 fflush (stdout);
1476 }
1477
1478 #endif /* DEBUG386 */
1479 \f
1480 static bfd_reloc_code_real_type
1481 reloc (unsigned int size,
1482 int pcrel,
1483 int sign,
1484 bfd_reloc_code_real_type other)
1485 {
1486 if (other != NO_RELOC)
1487 {
1488 reloc_howto_type *reloc;
1489
1490 if (size == 8)
1491 switch (other)
1492 {
1493 case BFD_RELOC_X86_64_GOT32:
1494 return BFD_RELOC_X86_64_GOT64;
1495 break;
1496 case BFD_RELOC_X86_64_PLTOFF64:
1497 return BFD_RELOC_X86_64_PLTOFF64;
1498 break;
1499 case BFD_RELOC_X86_64_GOTPC32:
1500 other = BFD_RELOC_X86_64_GOTPC64;
1501 break;
1502 case BFD_RELOC_X86_64_GOTPCREL:
1503 other = BFD_RELOC_X86_64_GOTPCREL64;
1504 break;
1505 case BFD_RELOC_X86_64_TPOFF32:
1506 other = BFD_RELOC_X86_64_TPOFF64;
1507 break;
1508 case BFD_RELOC_X86_64_DTPOFF32:
1509 other = BFD_RELOC_X86_64_DTPOFF64;
1510 break;
1511 default:
1512 break;
1513 }
1514
1515 /* Sign-checking 4-byte relocations in 16-/32-bit code is pointless. */
1516 if (size == 4 && flag_code != CODE_64BIT)
1517 sign = -1;
1518
1519 reloc = bfd_reloc_type_lookup (stdoutput, other);
1520 if (!reloc)
1521 as_bad (_("unknown relocation (%u)"), other);
1522 else if (size != bfd_get_reloc_size (reloc))
1523 as_bad (_("%u-byte relocation cannot be applied to %u-byte field"),
1524 bfd_get_reloc_size (reloc),
1525 size);
1526 else if (pcrel && !reloc->pc_relative)
1527 as_bad (_("non-pc-relative relocation for pc-relative field"));
1528 else if ((reloc->complain_on_overflow == complain_overflow_signed
1529 && !sign)
1530 || (reloc->complain_on_overflow == complain_overflow_unsigned
1531 && sign > 0))
1532 as_bad (_("relocated field and relocation type differ in signedness"));
1533 else
1534 return other;
1535 return NO_RELOC;
1536 }
1537
1538 if (pcrel)
1539 {
1540 if (!sign)
1541 as_bad (_("there are no unsigned pc-relative relocations"));
1542 switch (size)
1543 {
1544 case 1: return BFD_RELOC_8_PCREL;
1545 case 2: return BFD_RELOC_16_PCREL;
1546 case 4: return BFD_RELOC_32_PCREL;
1547 case 8: return BFD_RELOC_64_PCREL;
1548 }
1549 as_bad (_("cannot do %u byte pc-relative relocation"), size);
1550 }
1551 else
1552 {
1553 if (sign > 0)
1554 switch (size)
1555 {
1556 case 4: return BFD_RELOC_X86_64_32S;
1557 }
1558 else
1559 switch (size)
1560 {
1561 case 1: return BFD_RELOC_8;
1562 case 2: return BFD_RELOC_16;
1563 case 4: return BFD_RELOC_32;
1564 case 8: return BFD_RELOC_64;
1565 }
1566 as_bad (_("cannot do %s %u byte relocation"),
1567 sign > 0 ? "signed" : "unsigned", size);
1568 }
1569
1570 abort ();
1571 return BFD_RELOC_NONE;
1572 }
1573
1574 /* Here we decide which fixups can be adjusted to make them relative to
1575 the beginning of the section instead of the symbol. Basically we need
1576 to make sure that the dynamic relocations are done correctly, so in
1577 some cases we force the original symbol to be used. */
1578
1579 int
1580 tc_i386_fix_adjustable (fixS *fixP ATTRIBUTE_UNUSED)
1581 {
1582 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1583 if (!IS_ELF)
1584 return 1;
1585
1586 /* Don't adjust pc-relative references to merge sections in 64-bit
1587 mode. */
1588 if (use_rela_relocations
1589 && (S_GET_SEGMENT (fixP->fx_addsy)->flags & SEC_MERGE) != 0
1590 && fixP->fx_pcrel)
1591 return 0;
1592
1593 /* The x86_64 GOTPCREL are represented as 32bit PCrel relocations
1594 and changed later by validate_fix. */
1595 if (GOT_symbol && fixP->fx_subsy == GOT_symbol
1596 && fixP->fx_r_type == BFD_RELOC_32_PCREL)
1597 return 0;
1598
1599 /* adjust_reloc_syms doesn't know about the GOT. */
1600 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1601 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1602 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1603 || fixP->fx_r_type == BFD_RELOC_386_TLS_GD
1604 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDM
1605 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDO_32
1606 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE_32
1607 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE
1608 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTIE
1609 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE_32
1610 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE
1611 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTDESC
1612 || fixP->fx_r_type == BFD_RELOC_386_TLS_DESC_CALL
1613 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1614 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1615 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
1616 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSGD
1617 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSLD
1618 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF32
1619 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF64
1620 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTTPOFF
1621 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF32
1622 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF64
1623 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTOFF64
1624 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPC32_TLSDESC
1625 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSDESC_CALL
1626 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1627 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1628 return 0;
1629 #endif
1630 return 1;
1631 }
1632
1633 static int
1634 intel_float_operand (const char *mnemonic)
1635 {
1636 /* Note that the value returned is meaningful only for opcodes with (memory)
1637 operands, hence the code here is free to improperly handle opcodes that
1638 have no operands (for better performance and smaller code). */
1639
1640 if (mnemonic[0] != 'f')
1641 return 0; /* non-math */
1642
1643 switch (mnemonic[1])
1644 {
1645 /* fclex, fdecstp, fdisi, femms, feni, fincstp, finit, fsetpm, and
1646 the fs segment override prefix not currently handled because no
1647 call path can make opcodes without operands get here */
1648 case 'i':
1649 return 2 /* integer op */;
1650 case 'l':
1651 if (mnemonic[2] == 'd' && (mnemonic[3] == 'c' || mnemonic[3] == 'e'))
1652 return 3; /* fldcw/fldenv */
1653 break;
1654 case 'n':
1655 if (mnemonic[2] != 'o' /* fnop */)
1656 return 3; /* non-waiting control op */
1657 break;
1658 case 'r':
1659 if (mnemonic[2] == 's')
1660 return 3; /* frstor/frstpm */
1661 break;
1662 case 's':
1663 if (mnemonic[2] == 'a')
1664 return 3; /* fsave */
1665 if (mnemonic[2] == 't')
1666 {
1667 switch (mnemonic[3])
1668 {
1669 case 'c': /* fstcw */
1670 case 'd': /* fstdw */
1671 case 'e': /* fstenv */
1672 case 's': /* fsts[gw] */
1673 return 3;
1674 }
1675 }
1676 break;
1677 case 'x':
1678 if (mnemonic[2] == 'r' || mnemonic[2] == 's')
1679 return 0; /* fxsave/fxrstor are not really math ops */
1680 break;
1681 }
1682
1683 return 1;
1684 }
1685
1686 /* This is the guts of the machine-dependent assembler. LINE points to a
1687 machine dependent instruction. This function is supposed to emit
1688 the frags/bytes it assembles to. */
1689
1690 void
1691 md_assemble (line)
1692 char *line;
1693 {
1694 int j;
1695 char mnemonic[MAX_MNEM_SIZE];
1696
1697 /* Initialize globals. */
1698 memset (&i, '\0', sizeof (i));
1699 for (j = 0; j < MAX_OPERANDS; j++)
1700 i.reloc[j] = NO_RELOC;
1701 memset (disp_expressions, '\0', sizeof (disp_expressions));
1702 memset (im_expressions, '\0', sizeof (im_expressions));
1703 save_stack_p = save_stack;
1704
1705 /* First parse an instruction mnemonic & call i386_operand for the operands.
1706 We assume that the scrubber has arranged it so that line[0] is the valid
1707 start of a (possibly prefixed) mnemonic. */
1708
1709 line = parse_insn (line, mnemonic);
1710 if (line == NULL)
1711 return;
1712
1713 line = parse_operands (line, mnemonic);
1714 if (line == NULL)
1715 return;
1716
1717 /* The order of the immediates should be reversed
1718 for 2 immediates extrq and insertq instructions */
1719 if ((i.imm_operands == 2)
1720 && ((strcmp (mnemonic, "extrq") == 0)
1721 || (strcmp (mnemonic, "insertq") == 0)))
1722 {
1723 swap_2_operands (0, 1);
1724 /* "extrq" and insertq" are the only two instructions whose operands
1725 have to be reversed even though they have two immediate operands.
1726 */
1727 if (intel_syntax)
1728 swap_operands ();
1729 }
1730
1731 /* Now we've parsed the mnemonic into a set of templates, and have the
1732 operands at hand. */
1733
1734 /* All intel opcodes have reversed operands except for "bound" and
1735 "enter". We also don't reverse intersegment "jmp" and "call"
1736 instructions with 2 immediate operands so that the immediate segment
1737 precedes the offset, as it does when in AT&T mode. */
1738 if (intel_syntax
1739 && i.operands > 1
1740 && (strcmp (mnemonic, "bound") != 0)
1741 && (strcmp (mnemonic, "invlpga") != 0)
1742 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1743 swap_operands ();
1744
1745 if (i.imm_operands)
1746 optimize_imm ();
1747
1748 /* Don't optimize displacement for movabs since it only takes 64bit
1749 displacement. */
1750 if (i.disp_operands
1751 && (flag_code != CODE_64BIT
1752 || strcmp (mnemonic, "movabs") != 0))
1753 optimize_disp ();
1754
1755 /* Next, we find a template that matches the given insn,
1756 making sure the overlap of the given operands types is consistent
1757 with the template operand types. */
1758
1759 if (!match_template ())
1760 return;
1761
1762 if (intel_syntax)
1763 {
1764 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1765 if (SYSV386_COMPAT
1766 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1767 i.tm.base_opcode ^= Opcode_FloatR;
1768
1769 /* Zap movzx and movsx suffix. The suffix may have been set from
1770 "word ptr" or "byte ptr" on the source operand, but we'll use
1771 the suffix later to choose the destination register. */
1772 if ((i.tm.base_opcode & ~9) == 0x0fb6)
1773 {
1774 if (i.reg_operands < 2
1775 && !i.suffix
1776 && (~i.tm.opcode_modifier
1777 & (No_bSuf
1778 | No_wSuf
1779 | No_lSuf
1780 | No_sSuf
1781 | No_xSuf
1782 | No_qSuf)))
1783 as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
1784
1785 i.suffix = 0;
1786 }
1787 }
1788
1789 if (i.tm.opcode_modifier & FWait)
1790 if (!add_prefix (FWAIT_OPCODE))
1791 return;
1792
1793 /* Check string instruction segment overrides. */
1794 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1795 {
1796 if (!check_string ())
1797 return;
1798 }
1799
1800 if (!process_suffix ())
1801 return;
1802
1803 /* Make still unresolved immediate matches conform to size of immediate
1804 given in i.suffix. */
1805 if (!finalize_imm ())
1806 return;
1807
1808 if (i.types[0] & Imm1)
1809 i.imm_operands = 0; /* kludge for shift insns. */
1810 if (i.types[0] & ImplicitRegister)
1811 i.reg_operands--;
1812 if (i.types[1] & ImplicitRegister)
1813 i.reg_operands--;
1814 if (i.types[2] & ImplicitRegister)
1815 i.reg_operands--;
1816
1817 if (i.tm.opcode_modifier & ImmExt)
1818 {
1819 expressionS *exp;
1820
1821 if ((i.tm.cpu_flags & CpuSSE3) && i.operands > 0)
1822 {
1823 /* Streaming SIMD extensions 3 Instructions have the fixed
1824 operands with an opcode suffix which is coded in the same
1825 place as an 8-bit immediate field would be. Here we check
1826 those operands and remove them afterwards. */
1827 unsigned int x;
1828
1829 for (x = 0; x < i.operands; x++)
1830 if (i.op[x].regs->reg_num != x)
1831 as_bad (_("can't use register '%%%s' as operand %d in '%s'."),
1832 i.op[x].regs->reg_name, x + 1, i.tm.name);
1833 i.operands = 0;
1834 }
1835
1836 /* These AMD 3DNow! and Intel Katmai New Instructions have an
1837 opcode suffix which is coded in the same place as an 8-bit
1838 immediate field would be. Here we fake an 8-bit immediate
1839 operand from the opcode suffix stored in tm.extension_opcode. */
1840
1841 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
1842
1843 exp = &im_expressions[i.imm_operands++];
1844 i.op[i.operands].imms = exp;
1845 i.types[i.operands++] = Imm8;
1846 exp->X_op = O_constant;
1847 exp->X_add_number = i.tm.extension_opcode;
1848 i.tm.extension_opcode = None;
1849 }
1850
1851 /* For insns with operands there are more diddles to do to the opcode. */
1852 if (i.operands)
1853 {
1854 if (!process_operands ())
1855 return;
1856 }
1857 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
1858 {
1859 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
1860 as_warn (_("translating to `%sp'"), i.tm.name);
1861 }
1862
1863 /* Handle conversion of 'int $3' --> special int3 insn. */
1864 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
1865 {
1866 i.tm.base_opcode = INT3_OPCODE;
1867 i.imm_operands = 0;
1868 }
1869
1870 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
1871 && i.op[0].disps->X_op == O_constant)
1872 {
1873 /* Convert "jmp constant" (and "call constant") to a jump (call) to
1874 the absolute address given by the constant. Since ix86 jumps and
1875 calls are pc relative, we need to generate a reloc. */
1876 i.op[0].disps->X_add_symbol = &abs_symbol;
1877 i.op[0].disps->X_op = O_symbol;
1878 }
1879
1880 if ((i.tm.opcode_modifier & Rex64) != 0)
1881 i.rex |= REX_MODE64;
1882
1883 /* For 8 bit registers we need an empty rex prefix. Also if the
1884 instruction already has a prefix, we need to convert old
1885 registers to new ones. */
1886
1887 if (((i.types[0] & Reg8) != 0
1888 && (i.op[0].regs->reg_flags & RegRex64) != 0)
1889 || ((i.types[1] & Reg8) != 0
1890 && (i.op[1].regs->reg_flags & RegRex64) != 0)
1891 || (((i.types[0] & Reg8) != 0 || (i.types[1] & Reg8) != 0)
1892 && i.rex != 0))
1893 {
1894 int x;
1895
1896 i.rex |= REX_OPCODE;
1897 for (x = 0; x < 2; x++)
1898 {
1899 /* Look for 8 bit operand that uses old registers. */
1900 if ((i.types[x] & Reg8) != 0
1901 && (i.op[x].regs->reg_flags & RegRex64) == 0)
1902 {
1903 /* In case it is "hi" register, give up. */
1904 if (i.op[x].regs->reg_num > 3)
1905 as_bad (_("can't encode register '%%%s' in an "
1906 "instruction requiring REX prefix."),
1907 i.op[x].regs->reg_name);
1908
1909 /* Otherwise it is equivalent to the extended register.
1910 Since the encoding doesn't change this is merely
1911 cosmetic cleanup for debug output. */
1912
1913 i.op[x].regs = i.op[x].regs + 8;
1914 }
1915 }
1916 }
1917
1918 if (i.rex != 0)
1919 add_prefix (REX_OPCODE | i.rex);
1920
1921 /* We are ready to output the insn. */
1922 output_insn ();
1923 }
1924
1925 static char *
1926 parse_insn (char *line, char *mnemonic)
1927 {
1928 char *l = line;
1929 char *token_start = l;
1930 char *mnem_p;
1931 int supported;
1932 const template *t;
1933
1934 /* Non-zero if we found a prefix only acceptable with string insns. */
1935 const char *expecting_string_instruction = NULL;
1936
1937 while (1)
1938 {
1939 mnem_p = mnemonic;
1940 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1941 {
1942 mnem_p++;
1943 if (mnem_p >= mnemonic + MAX_MNEM_SIZE)
1944 {
1945 as_bad (_("no such instruction: `%s'"), token_start);
1946 return NULL;
1947 }
1948 l++;
1949 }
1950 if (!is_space_char (*l)
1951 && *l != END_OF_INSN
1952 && (intel_syntax
1953 || (*l != PREFIX_SEPARATOR
1954 && *l != ',')))
1955 {
1956 as_bad (_("invalid character %s in mnemonic"),
1957 output_invalid (*l));
1958 return NULL;
1959 }
1960 if (token_start == l)
1961 {
1962 if (!intel_syntax && *l == PREFIX_SEPARATOR)
1963 as_bad (_("expecting prefix; got nothing"));
1964 else
1965 as_bad (_("expecting mnemonic; got nothing"));
1966 return NULL;
1967 }
1968
1969 /* Look up instruction (or prefix) via hash table. */
1970 current_templates = hash_find (op_hash, mnemonic);
1971
1972 if (*l != END_OF_INSN
1973 && (!is_space_char (*l) || l[1] != END_OF_INSN)
1974 && current_templates
1975 && (current_templates->start->opcode_modifier & IsPrefix))
1976 {
1977 if (current_templates->start->cpu_flags
1978 & (flag_code != CODE_64BIT ? Cpu64 : CpuNo64))
1979 {
1980 as_bad ((flag_code != CODE_64BIT
1981 ? _("`%s' is only supported in 64-bit mode")
1982 : _("`%s' is not supported in 64-bit mode")),
1983 current_templates->start->name);
1984 return NULL;
1985 }
1986 /* If we are in 16-bit mode, do not allow addr16 or data16.
1987 Similarly, in 32-bit mode, do not allow addr32 or data32. */
1988 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
1989 && flag_code != CODE_64BIT
1990 && (((current_templates->start->opcode_modifier & Size32) != 0)
1991 ^ (flag_code == CODE_16BIT)))
1992 {
1993 as_bad (_("redundant %s prefix"),
1994 current_templates->start->name);
1995 return NULL;
1996 }
1997 /* Add prefix, checking for repeated prefixes. */
1998 switch (add_prefix (current_templates->start->base_opcode))
1999 {
2000 case 0:
2001 return NULL;
2002 case 2:
2003 expecting_string_instruction = current_templates->start->name;
2004 break;
2005 }
2006 /* Skip past PREFIX_SEPARATOR and reset token_start. */
2007 token_start = ++l;
2008 }
2009 else
2010 break;
2011 }
2012
2013 if (!current_templates)
2014 {
2015 /* See if we can get a match by trimming off a suffix. */
2016 switch (mnem_p[-1])
2017 {
2018 case WORD_MNEM_SUFFIX:
2019 if (intel_syntax && (intel_float_operand (mnemonic) & 2))
2020 i.suffix = SHORT_MNEM_SUFFIX;
2021 else
2022 case BYTE_MNEM_SUFFIX:
2023 case QWORD_MNEM_SUFFIX:
2024 i.suffix = mnem_p[-1];
2025 mnem_p[-1] = '\0';
2026 current_templates = hash_find (op_hash, mnemonic);
2027 break;
2028 case SHORT_MNEM_SUFFIX:
2029 case LONG_MNEM_SUFFIX:
2030 if (!intel_syntax)
2031 {
2032 i.suffix = mnem_p[-1];
2033 mnem_p[-1] = '\0';
2034 current_templates = hash_find (op_hash, mnemonic);
2035 }
2036 break;
2037
2038 /* Intel Syntax. */
2039 case 'd':
2040 if (intel_syntax)
2041 {
2042 if (intel_float_operand (mnemonic) == 1)
2043 i.suffix = SHORT_MNEM_SUFFIX;
2044 else
2045 i.suffix = LONG_MNEM_SUFFIX;
2046 mnem_p[-1] = '\0';
2047 current_templates = hash_find (op_hash, mnemonic);
2048 }
2049 break;
2050 }
2051 if (!current_templates)
2052 {
2053 as_bad (_("no such instruction: `%s'"), token_start);
2054 return NULL;
2055 }
2056 }
2057
2058 if (current_templates->start->opcode_modifier & (Jump | JumpByte))
2059 {
2060 /* Check for a branch hint. We allow ",pt" and ",pn" for
2061 predict taken and predict not taken respectively.
2062 I'm not sure that branch hints actually do anything on loop
2063 and jcxz insns (JumpByte) for current Pentium4 chips. They
2064 may work in the future and it doesn't hurt to accept them
2065 now. */
2066 if (l[0] == ',' && l[1] == 'p')
2067 {
2068 if (l[2] == 't')
2069 {
2070 if (!add_prefix (DS_PREFIX_OPCODE))
2071 return NULL;
2072 l += 3;
2073 }
2074 else if (l[2] == 'n')
2075 {
2076 if (!add_prefix (CS_PREFIX_OPCODE))
2077 return NULL;
2078 l += 3;
2079 }
2080 }
2081 }
2082 /* Any other comma loses. */
2083 if (*l == ',')
2084 {
2085 as_bad (_("invalid character %s in mnemonic"),
2086 output_invalid (*l));
2087 return NULL;
2088 }
2089
2090 /* Check if instruction is supported on specified architecture. */
2091 supported = 0;
2092 for (t = current_templates->start; t < current_templates->end; ++t)
2093 {
2094 if (!((t->cpu_flags & ~(Cpu64 | CpuNo64))
2095 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64))))
2096 supported |= 1;
2097 if (!(t->cpu_flags & (flag_code == CODE_64BIT ? CpuNo64 : Cpu64)))
2098 supported |= 2;
2099 }
2100 if (!(supported & 2))
2101 {
2102 as_bad (flag_code == CODE_64BIT
2103 ? _("`%s' is not supported in 64-bit mode")
2104 : _("`%s' is only supported in 64-bit mode"),
2105 current_templates->start->name);
2106 return NULL;
2107 }
2108 if (!(supported & 1))
2109 {
2110 as_warn (_("`%s' is not supported on `%s%s'"),
2111 current_templates->start->name,
2112 cpu_arch_name,
2113 cpu_sub_arch_name ? cpu_sub_arch_name : "");
2114 }
2115 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
2116 {
2117 as_warn (_("use .code16 to ensure correct addressing mode"));
2118 }
2119
2120 /* Check for rep/repne without a string instruction. */
2121 if (expecting_string_instruction)
2122 {
2123 static templates override;
2124
2125 for (t = current_templates->start; t < current_templates->end; ++t)
2126 if (t->opcode_modifier & IsString)
2127 break;
2128 if (t >= current_templates->end)
2129 {
2130 as_bad (_("expecting string instruction after `%s'"),
2131 expecting_string_instruction);
2132 return NULL;
2133 }
2134 for (override.start = t; t < current_templates->end; ++t)
2135 if (!(t->opcode_modifier & IsString))
2136 break;
2137 override.end = t;
2138 current_templates = &override;
2139 }
2140
2141 return l;
2142 }
2143
2144 static char *
2145 parse_operands (char *l, const char *mnemonic)
2146 {
2147 char *token_start;
2148
2149 /* 1 if operand is pending after ','. */
2150 unsigned int expecting_operand = 0;
2151
2152 /* Non-zero if operand parens not balanced. */
2153 unsigned int paren_not_balanced;
2154
2155 while (*l != END_OF_INSN)
2156 {
2157 /* Skip optional white space before operand. */
2158 if (is_space_char (*l))
2159 ++l;
2160 if (!is_operand_char (*l) && *l != END_OF_INSN)
2161 {
2162 as_bad (_("invalid character %s before operand %d"),
2163 output_invalid (*l),
2164 i.operands + 1);
2165 return NULL;
2166 }
2167 token_start = l; /* after white space */
2168 paren_not_balanced = 0;
2169 while (paren_not_balanced || *l != ',')
2170 {
2171 if (*l == END_OF_INSN)
2172 {
2173 if (paren_not_balanced)
2174 {
2175 if (!intel_syntax)
2176 as_bad (_("unbalanced parenthesis in operand %d."),
2177 i.operands + 1);
2178 else
2179 as_bad (_("unbalanced brackets in operand %d."),
2180 i.operands + 1);
2181 return NULL;
2182 }
2183 else
2184 break; /* we are done */
2185 }
2186 else if (!is_operand_char (*l) && !is_space_char (*l))
2187 {
2188 as_bad (_("invalid character %s in operand %d"),
2189 output_invalid (*l),
2190 i.operands + 1);
2191 return NULL;
2192 }
2193 if (!intel_syntax)
2194 {
2195 if (*l == '(')
2196 ++paren_not_balanced;
2197 if (*l == ')')
2198 --paren_not_balanced;
2199 }
2200 else
2201 {
2202 if (*l == '[')
2203 ++paren_not_balanced;
2204 if (*l == ']')
2205 --paren_not_balanced;
2206 }
2207 l++;
2208 }
2209 if (l != token_start)
2210 { /* Yes, we've read in another operand. */
2211 unsigned int operand_ok;
2212 this_operand = i.operands++;
2213 if (i.operands > MAX_OPERANDS)
2214 {
2215 as_bad (_("spurious operands; (%d operands/instruction max)"),
2216 MAX_OPERANDS);
2217 return NULL;
2218 }
2219 /* Now parse operand adding info to 'i' as we go along. */
2220 END_STRING_AND_SAVE (l);
2221
2222 if (intel_syntax)
2223 operand_ok =
2224 i386_intel_operand (token_start,
2225 intel_float_operand (mnemonic));
2226 else
2227 operand_ok = i386_operand (token_start);
2228
2229 RESTORE_END_STRING (l);
2230 if (!operand_ok)
2231 return NULL;
2232 }
2233 else
2234 {
2235 if (expecting_operand)
2236 {
2237 expecting_operand_after_comma:
2238 as_bad (_("expecting operand after ','; got nothing"));
2239 return NULL;
2240 }
2241 if (*l == ',')
2242 {
2243 as_bad (_("expecting operand before ','; got nothing"));
2244 return NULL;
2245 }
2246 }
2247
2248 /* Now *l must be either ',' or END_OF_INSN. */
2249 if (*l == ',')
2250 {
2251 if (*++l == END_OF_INSN)
2252 {
2253 /* Just skip it, if it's \n complain. */
2254 goto expecting_operand_after_comma;
2255 }
2256 expecting_operand = 1;
2257 }
2258 }
2259 return l;
2260 }
2261
2262 static void
2263 swap_2_operands (int xchg1, int xchg2)
2264 {
2265 union i386_op temp_op;
2266 unsigned int temp_type;
2267 enum bfd_reloc_code_real temp_reloc;
2268
2269 temp_type = i.types[xchg2];
2270 i.types[xchg2] = i.types[xchg1];
2271 i.types[xchg1] = temp_type;
2272 temp_op = i.op[xchg2];
2273 i.op[xchg2] = i.op[xchg1];
2274 i.op[xchg1] = temp_op;
2275 temp_reloc = i.reloc[xchg2];
2276 i.reloc[xchg2] = i.reloc[xchg1];
2277 i.reloc[xchg1] = temp_reloc;
2278 }
2279
2280 static void
2281 swap_operands (void)
2282 {
2283 switch (i.operands)
2284 {
2285 case 4:
2286 swap_2_operands (1, i.operands - 2);
2287 case 3:
2288 case 2:
2289 swap_2_operands (0, i.operands - 1);
2290 break;
2291 default:
2292 abort ();
2293 }
2294
2295 if (i.mem_operands == 2)
2296 {
2297 const seg_entry *temp_seg;
2298 temp_seg = i.seg[0];
2299 i.seg[0] = i.seg[1];
2300 i.seg[1] = temp_seg;
2301 }
2302 }
2303
2304 /* Try to ensure constant immediates are represented in the smallest
2305 opcode possible. */
2306 static void
2307 optimize_imm (void)
2308 {
2309 char guess_suffix = 0;
2310 int op;
2311
2312 if (i.suffix)
2313 guess_suffix = i.suffix;
2314 else if (i.reg_operands)
2315 {
2316 /* Figure out a suffix from the last register operand specified.
2317 We can't do this properly yet, ie. excluding InOutPortReg,
2318 but the following works for instructions with immediates.
2319 In any case, we can't set i.suffix yet. */
2320 for (op = i.operands; --op >= 0;)
2321 if (i.types[op] & Reg)
2322 {
2323 if (i.types[op] & Reg8)
2324 guess_suffix = BYTE_MNEM_SUFFIX;
2325 else if (i.types[op] & Reg16)
2326 guess_suffix = WORD_MNEM_SUFFIX;
2327 else if (i.types[op] & Reg32)
2328 guess_suffix = LONG_MNEM_SUFFIX;
2329 else if (i.types[op] & Reg64)
2330 guess_suffix = QWORD_MNEM_SUFFIX;
2331 break;
2332 }
2333 }
2334 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
2335 guess_suffix = WORD_MNEM_SUFFIX;
2336
2337 for (op = i.operands; --op >= 0;)
2338 if (i.types[op] & Imm)
2339 {
2340 switch (i.op[op].imms->X_op)
2341 {
2342 case O_constant:
2343 /* If a suffix is given, this operand may be shortened. */
2344 switch (guess_suffix)
2345 {
2346 case LONG_MNEM_SUFFIX:
2347 i.types[op] |= Imm32 | Imm64;
2348 break;
2349 case WORD_MNEM_SUFFIX:
2350 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
2351 break;
2352 case BYTE_MNEM_SUFFIX:
2353 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
2354 break;
2355 }
2356
2357 /* If this operand is at most 16 bits, convert it
2358 to a signed 16 bit number before trying to see
2359 whether it will fit in an even smaller size.
2360 This allows a 16-bit operand such as $0xffe0 to
2361 be recognised as within Imm8S range. */
2362 if ((i.types[op] & Imm16)
2363 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
2364 {
2365 i.op[op].imms->X_add_number =
2366 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
2367 }
2368 if ((i.types[op] & Imm32)
2369 && ((i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1))
2370 == 0))
2371 {
2372 i.op[op].imms->X_add_number = ((i.op[op].imms->X_add_number
2373 ^ ((offsetT) 1 << 31))
2374 - ((offsetT) 1 << 31));
2375 }
2376 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
2377
2378 /* We must avoid matching of Imm32 templates when 64bit
2379 only immediate is available. */
2380 if (guess_suffix == QWORD_MNEM_SUFFIX)
2381 i.types[op] &= ~Imm32;
2382 break;
2383
2384 case O_absent:
2385 case O_register:
2386 abort ();
2387
2388 /* Symbols and expressions. */
2389 default:
2390 /* Convert symbolic operand to proper sizes for matching, but don't
2391 prevent matching a set of insns that only supports sizes other
2392 than those matching the insn suffix. */
2393 {
2394 unsigned int mask, allowed = 0;
2395 const template *t;
2396
2397 for (t = current_templates->start;
2398 t < current_templates->end;
2399 ++t)
2400 allowed |= t->operand_types[op];
2401 switch (guess_suffix)
2402 {
2403 case QWORD_MNEM_SUFFIX:
2404 mask = Imm64 | Imm32S;
2405 break;
2406 case LONG_MNEM_SUFFIX:
2407 mask = Imm32;
2408 break;
2409 case WORD_MNEM_SUFFIX:
2410 mask = Imm16;
2411 break;
2412 case BYTE_MNEM_SUFFIX:
2413 mask = Imm8;
2414 break;
2415 default:
2416 mask = 0;
2417 break;
2418 }
2419 if (mask & allowed)
2420 i.types[op] &= mask;
2421 }
2422 break;
2423 }
2424 }
2425 }
2426
2427 /* Try to use the smallest displacement type too. */
2428 static void
2429 optimize_disp (void)
2430 {
2431 int op;
2432
2433 for (op = i.operands; --op >= 0;)
2434 if (i.types[op] & Disp)
2435 {
2436 if (i.op[op].disps->X_op == O_constant)
2437 {
2438 offsetT disp = i.op[op].disps->X_add_number;
2439
2440 if ((i.types[op] & Disp16)
2441 && (disp & ~(offsetT) 0xffff) == 0)
2442 {
2443 /* If this operand is at most 16 bits, convert
2444 to a signed 16 bit number and don't use 64bit
2445 displacement. */
2446 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
2447 i.types[op] &= ~Disp64;
2448 }
2449 if ((i.types[op] & Disp32)
2450 && (disp & ~(((offsetT) 2 << 31) - 1)) == 0)
2451 {
2452 /* If this operand is at most 32 bits, convert
2453 to a signed 32 bit number and don't use 64bit
2454 displacement. */
2455 disp &= (((offsetT) 2 << 31) - 1);
2456 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
2457 i.types[op] &= ~Disp64;
2458 }
2459 if (!disp && (i.types[op] & BaseIndex))
2460 {
2461 i.types[op] &= ~Disp;
2462 i.op[op].disps = 0;
2463 i.disp_operands--;
2464 }
2465 else if (flag_code == CODE_64BIT)
2466 {
2467 if (fits_in_signed_long (disp))
2468 {
2469 i.types[op] &= ~Disp64;
2470 i.types[op] |= Disp32S;
2471 }
2472 if (fits_in_unsigned_long (disp))
2473 i.types[op] |= Disp32;
2474 }
2475 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
2476 && fits_in_signed_byte (disp))
2477 i.types[op] |= Disp8;
2478 }
2479 else if (i.reloc[op] == BFD_RELOC_386_TLS_DESC_CALL
2480 || i.reloc[op] == BFD_RELOC_X86_64_TLSDESC_CALL)
2481 {
2482 fix_new_exp (frag_now, frag_more (0) - frag_now->fr_literal, 0,
2483 i.op[op].disps, 0, i.reloc[op]);
2484 i.types[op] &= ~Disp;
2485 }
2486 else
2487 /* We only support 64bit displacement on constants. */
2488 i.types[op] &= ~Disp64;
2489 }
2490 }
2491
2492 static int
2493 match_template (void)
2494 {
2495 /* Points to template once we've found it. */
2496 const template *t;
2497 unsigned int overlap0, overlap1, overlap2, overlap3;
2498 unsigned int found_reverse_match;
2499 int suffix_check;
2500 unsigned int operand_types [MAX_OPERANDS];
2501 int addr_prefix_disp;
2502 unsigned int j;
2503
2504 #if MAX_OPERANDS != 4
2505 # error "MAX_OPERANDS must be 4."
2506 #endif
2507
2508 #define MATCH(overlap, given, template) \
2509 ((overlap & ~JumpAbsolute) \
2510 && (((given) & (BaseIndex | JumpAbsolute)) \
2511 == ((overlap) & (BaseIndex | JumpAbsolute))))
2512
2513 /* If given types r0 and r1 are registers they must be of the same type
2514 unless the expected operand type register overlap is null.
2515 Note that Acc in a template matches every size of reg. */
2516 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
2517 (((g0) & Reg) == 0 || ((g1) & Reg) == 0 \
2518 || ((g0) & Reg) == ((g1) & Reg) \
2519 || ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
2520
2521 overlap0 = 0;
2522 overlap1 = 0;
2523 overlap2 = 0;
2524 overlap3 = 0;
2525 found_reverse_match = 0;
2526 for (j = 0; j < MAX_OPERANDS; j++)
2527 operand_types [j] = 0;
2528 addr_prefix_disp = -1;
2529 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
2530 ? No_bSuf
2531 : (i.suffix == WORD_MNEM_SUFFIX
2532 ? No_wSuf
2533 : (i.suffix == SHORT_MNEM_SUFFIX
2534 ? No_sSuf
2535 : (i.suffix == LONG_MNEM_SUFFIX
2536 ? No_lSuf
2537 : (i.suffix == QWORD_MNEM_SUFFIX
2538 ? No_qSuf
2539 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX
2540 ? No_xSuf : 0))))));
2541
2542 for (t = current_templates->start; t < current_templates->end; t++)
2543 {
2544 addr_prefix_disp = -1;
2545
2546 /* Must have right number of operands. */
2547 if (i.operands != t->operands)
2548 continue;
2549
2550 /* Check the suffix, except for some instructions in intel mode. */
2551 if ((t->opcode_modifier & suffix_check)
2552 && !(intel_syntax
2553 && (t->opcode_modifier & IgnoreSize)))
2554 continue;
2555
2556 for (j = 0; j < MAX_OPERANDS; j++)
2557 operand_types [j] = t->operand_types [j];
2558
2559 /* In general, don't allow 64-bit operands in 32-bit mode. */
2560 if (i.suffix == QWORD_MNEM_SUFFIX
2561 && flag_code != CODE_64BIT
2562 && (intel_syntax
2563 ? (!(t->opcode_modifier & IgnoreSize)
2564 && !intel_float_operand (t->name))
2565 : intel_float_operand (t->name) != 2)
2566 && (!(operand_types[0] & (RegMMX | RegXMM))
2567 || !(operand_types[t->operands > 1] & (RegMMX | RegXMM)))
2568 && (t->base_opcode != 0x0fc7
2569 || t->extension_opcode != 1 /* cmpxchg8b */))
2570 continue;
2571
2572 /* Do not verify operands when there are none. */
2573 else if (!t->operands)
2574 {
2575 if (t->cpu_flags & ~cpu_arch_flags)
2576 continue;
2577 /* We've found a match; break out of loop. */
2578 break;
2579 }
2580
2581 /* Address size prefix will turn Disp64/Disp32/Disp16 operand
2582 into Disp32/Disp16/Disp32 operand. */
2583 if (i.prefix[ADDR_PREFIX] != 0)
2584 {
2585 unsigned int DispOn = 0, DispOff = 0;
2586
2587 switch (flag_code)
2588 {
2589 case CODE_16BIT:
2590 DispOn = Disp32;
2591 DispOff = Disp16;
2592 break;
2593 case CODE_32BIT:
2594 DispOn = Disp16;
2595 DispOff = Disp32;
2596 break;
2597 case CODE_64BIT:
2598 DispOn = Disp32;
2599 DispOff = Disp64;
2600 break;
2601 }
2602
2603 for (j = 0; j < MAX_OPERANDS; j++)
2604 {
2605 /* There should be only one Disp operand. */
2606 if ((operand_types[j] & DispOff))
2607 {
2608 addr_prefix_disp = j;
2609 operand_types[j] |= DispOn;
2610 operand_types[j] &= ~DispOff;
2611 break;
2612 }
2613 }
2614 }
2615
2616 overlap0 = i.types[0] & operand_types[0];
2617 switch (t->operands)
2618 {
2619 case 1:
2620 if (!MATCH (overlap0, i.types[0], operand_types[0]))
2621 continue;
2622 break;
2623 case 2:
2624 case 3:
2625 case 4:
2626 overlap1 = i.types[1] & operand_types[1];
2627 if (!MATCH (overlap0, i.types[0], operand_types[0])
2628 || !MATCH (overlap1, i.types[1], operand_types[1])
2629 /* monitor in SSE3 is a very special case. The first
2630 register and the second register may have different
2631 sizes. */
2632 || !((t->base_opcode == 0x0f01
2633 && t->extension_opcode == 0xc8)
2634 || CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2635 operand_types[0],
2636 overlap1, i.types[1],
2637 operand_types[1])))
2638 {
2639 /* Check if other direction is valid ... */
2640 if ((t->opcode_modifier & (D | FloatD)) == 0)
2641 continue;
2642
2643 /* Try reversing direction of operands. */
2644 overlap0 = i.types[0] & operand_types[1];
2645 overlap1 = i.types[1] & operand_types[0];
2646 if (!MATCH (overlap0, i.types[0], operand_types[1])
2647 || !MATCH (overlap1, i.types[1], operand_types[0])
2648 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2649 operand_types[1],
2650 overlap1, i.types[1],
2651 operand_types[0]))
2652 {
2653 /* Does not match either direction. */
2654 continue;
2655 }
2656 /* found_reverse_match holds which of D or FloatDR
2657 we've found. */
2658 if ((t->opcode_modifier & D))
2659 found_reverse_match = Opcode_D;
2660 else if ((t->opcode_modifier & FloatD))
2661 found_reverse_match = Opcode_FloatD;
2662 else
2663 found_reverse_match = 0;
2664 if ((t->opcode_modifier & FloatR))
2665 found_reverse_match |= Opcode_FloatR;
2666 }
2667 else
2668 {
2669 /* Found a forward 2 operand match here. */
2670 switch (t->operands)
2671 {
2672 case 4:
2673 overlap3 = i.types[3] & operand_types[3];
2674 case 3:
2675 overlap2 = i.types[2] & operand_types[2];
2676 break;
2677 }
2678
2679 switch (t->operands)
2680 {
2681 case 4:
2682 if (!MATCH (overlap3, i.types[3], operand_types[3])
2683 || !CONSISTENT_REGISTER_MATCH (overlap2,
2684 i.types[2],
2685 operand_types[2],
2686 overlap3,
2687 i.types[3],
2688 operand_types[3]))
2689 continue;
2690 case 3:
2691 /* Here we make use of the fact that there are no
2692 reverse match 3 operand instructions, and all 3
2693 operand instructions only need to be checked for
2694 register consistency between operands 2 and 3. */
2695 if (!MATCH (overlap2, i.types[2], operand_types[2])
2696 || !CONSISTENT_REGISTER_MATCH (overlap1,
2697 i.types[1],
2698 operand_types[1],
2699 overlap2,
2700 i.types[2],
2701 operand_types[2]))
2702 continue;
2703 break;
2704 }
2705 }
2706 /* Found either forward/reverse 2, 3 or 4 operand match here:
2707 slip through to break. */
2708 }
2709 if (t->cpu_flags & ~cpu_arch_flags)
2710 {
2711 found_reverse_match = 0;
2712 continue;
2713 }
2714 /* We've found a match; break out of loop. */
2715 break;
2716 }
2717
2718 if (t == current_templates->end)
2719 {
2720 /* We found no match. */
2721 as_bad (_("suffix or operands invalid for `%s'"),
2722 current_templates->start->name);
2723 return 0;
2724 }
2725
2726 if (!quiet_warnings)
2727 {
2728 if (!intel_syntax
2729 && ((i.types[0] & JumpAbsolute)
2730 != (operand_types[0] & JumpAbsolute)))
2731 {
2732 as_warn (_("indirect %s without `*'"), t->name);
2733 }
2734
2735 if ((t->opcode_modifier & (IsPrefix | IgnoreSize))
2736 == (IsPrefix | IgnoreSize))
2737 {
2738 /* Warn them that a data or address size prefix doesn't
2739 affect assembly of the next line of code. */
2740 as_warn (_("stand-alone `%s' prefix"), t->name);
2741 }
2742 }
2743
2744 /* Copy the template we found. */
2745 i.tm = *t;
2746
2747 if (addr_prefix_disp != -1)
2748 i.tm.operand_types[addr_prefix_disp]
2749 = operand_types[addr_prefix_disp];
2750
2751 if (found_reverse_match)
2752 {
2753 /* If we found a reverse match we must alter the opcode
2754 direction bit. found_reverse_match holds bits to change
2755 (different for int & float insns). */
2756
2757 i.tm.base_opcode ^= found_reverse_match;
2758
2759 i.tm.operand_types[0] = operand_types[1];
2760 i.tm.operand_types[1] = operand_types[0];
2761 }
2762
2763 return 1;
2764 }
2765
2766 static int
2767 check_string (void)
2768 {
2769 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
2770 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
2771 {
2772 if (i.seg[0] != NULL && i.seg[0] != &es)
2773 {
2774 as_bad (_("`%s' operand %d must use `%%es' segment"),
2775 i.tm.name,
2776 mem_op + 1);
2777 return 0;
2778 }
2779 /* There's only ever one segment override allowed per instruction.
2780 This instruction possibly has a legal segment override on the
2781 second operand, so copy the segment to where non-string
2782 instructions store it, allowing common code. */
2783 i.seg[0] = i.seg[1];
2784 }
2785 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
2786 {
2787 if (i.seg[1] != NULL && i.seg[1] != &es)
2788 {
2789 as_bad (_("`%s' operand %d must use `%%es' segment"),
2790 i.tm.name,
2791 mem_op + 2);
2792 return 0;
2793 }
2794 }
2795 return 1;
2796 }
2797
2798 static int
2799 process_suffix (void)
2800 {
2801 /* If matched instruction specifies an explicit instruction mnemonic
2802 suffix, use it. */
2803 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
2804 {
2805 if (i.tm.opcode_modifier & Size16)
2806 i.suffix = WORD_MNEM_SUFFIX;
2807 else if (i.tm.opcode_modifier & Size64)
2808 i.suffix = QWORD_MNEM_SUFFIX;
2809 else
2810 i.suffix = LONG_MNEM_SUFFIX;
2811 }
2812 else if (i.reg_operands)
2813 {
2814 /* If there's no instruction mnemonic suffix we try to invent one
2815 based on register operands. */
2816 if (!i.suffix)
2817 {
2818 /* We take i.suffix from the last register operand specified,
2819 Destination register type is more significant than source
2820 register type. */
2821 int op;
2822
2823 for (op = i.operands; --op >= 0;)
2824 if ((i.types[op] & Reg)
2825 && !(i.tm.operand_types[op] & InOutPortReg))
2826 {
2827 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
2828 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
2829 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
2830 LONG_MNEM_SUFFIX);
2831 break;
2832 }
2833 }
2834 else if (i.suffix == BYTE_MNEM_SUFFIX)
2835 {
2836 if (!check_byte_reg ())
2837 return 0;
2838 }
2839 else if (i.suffix == LONG_MNEM_SUFFIX)
2840 {
2841 if (!check_long_reg ())
2842 return 0;
2843 }
2844 else if (i.suffix == QWORD_MNEM_SUFFIX)
2845 {
2846 if (!check_qword_reg ())
2847 return 0;
2848 }
2849 else if (i.suffix == WORD_MNEM_SUFFIX)
2850 {
2851 if (!check_word_reg ())
2852 return 0;
2853 }
2854 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2855 /* Do nothing if the instruction is going to ignore the prefix. */
2856 ;
2857 else
2858 abort ();
2859 }
2860 else if ((i.tm.opcode_modifier & DefaultSize)
2861 && !i.suffix
2862 /* exclude fldenv/frstor/fsave/fstenv */
2863 && (i.tm.opcode_modifier & No_sSuf))
2864 {
2865 i.suffix = stackop_size;
2866 }
2867 else if (intel_syntax
2868 && !i.suffix
2869 && ((i.tm.operand_types[0] & JumpAbsolute)
2870 || (i.tm.opcode_modifier & (JumpByte|JumpInterSegment))
2871 || (i.tm.base_opcode == 0x0f01 /* [ls][gi]dt */
2872 && i.tm.extension_opcode <= 3)))
2873 {
2874 switch (flag_code)
2875 {
2876 case CODE_64BIT:
2877 if (!(i.tm.opcode_modifier & No_qSuf))
2878 {
2879 i.suffix = QWORD_MNEM_SUFFIX;
2880 break;
2881 }
2882 case CODE_32BIT:
2883 if (!(i.tm.opcode_modifier & No_lSuf))
2884 i.suffix = LONG_MNEM_SUFFIX;
2885 break;
2886 case CODE_16BIT:
2887 if (!(i.tm.opcode_modifier & No_wSuf))
2888 i.suffix = WORD_MNEM_SUFFIX;
2889 break;
2890 }
2891 }
2892
2893 if (!i.suffix)
2894 {
2895 if (!intel_syntax)
2896 {
2897 if (i.tm.opcode_modifier & W)
2898 {
2899 as_bad (_("no instruction mnemonic suffix given and "
2900 "no register operands; can't size instruction"));
2901 return 0;
2902 }
2903 }
2904 else
2905 {
2906 unsigned int suffixes = (~i.tm.opcode_modifier
2907 & (No_bSuf
2908 | No_wSuf
2909 | No_lSuf
2910 | No_sSuf
2911 | No_xSuf
2912 | No_qSuf));
2913
2914 if ((i.tm.opcode_modifier & W)
2915 || ((suffixes & (suffixes - 1))
2916 && !(i.tm.opcode_modifier & (DefaultSize | IgnoreSize))))
2917 {
2918 as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
2919 return 0;
2920 }
2921 }
2922 }
2923
2924 /* Change the opcode based on the operand size given by i.suffix;
2925 We don't need to change things for byte insns. */
2926
2927 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2928 {
2929 /* It's not a byte, select word/dword operation. */
2930 if (i.tm.opcode_modifier & W)
2931 {
2932 if (i.tm.opcode_modifier & ShortForm)
2933 i.tm.base_opcode |= 8;
2934 else
2935 i.tm.base_opcode |= 1;
2936 }
2937
2938 /* Now select between word & dword operations via the operand
2939 size prefix, except for instructions that will ignore this
2940 prefix anyway. */
2941 if (i.tm.base_opcode == 0x0f01 && i.tm.extension_opcode == 0xc8)
2942 {
2943 /* monitor in SSE3 is a very special case. The default size
2944 of AX is the size of mode. The address size override
2945 prefix will change the size of AX. */
2946 if (i.op->regs[0].reg_type &
2947 (flag_code == CODE_32BIT ? Reg16 : Reg32))
2948 if (!add_prefix (ADDR_PREFIX_OPCODE))
2949 return 0;
2950 }
2951 else if (i.suffix != QWORD_MNEM_SUFFIX
2952 && i.suffix != LONG_DOUBLE_MNEM_SUFFIX
2953 && !(i.tm.opcode_modifier & (IgnoreSize | FloatMF))
2954 && ((i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
2955 || (flag_code == CODE_64BIT
2956 && (i.tm.opcode_modifier & JumpByte))))
2957 {
2958 unsigned int prefix = DATA_PREFIX_OPCODE;
2959
2960 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
2961 prefix = ADDR_PREFIX_OPCODE;
2962
2963 if (!add_prefix (prefix))
2964 return 0;
2965 }
2966
2967 /* Set mode64 for an operand. */
2968 if (i.suffix == QWORD_MNEM_SUFFIX
2969 && flag_code == CODE_64BIT
2970 && (i.tm.opcode_modifier & NoRex64) == 0)
2971 {
2972 /* Special case for xchg %rax,%rax. It is NOP and doesn't
2973 need rex64. */
2974 if (i.operands != 2
2975 || i.types [0] != (Acc | Reg64)
2976 || i.types [1] != (Acc | Reg64)
2977 || strcmp (i.tm.name, "xchg") != 0)
2978 i.rex |= REX_MODE64;
2979 }
2980
2981 /* Size floating point instruction. */
2982 if (i.suffix == LONG_MNEM_SUFFIX)
2983 if (i.tm.opcode_modifier & FloatMF)
2984 i.tm.base_opcode ^= 4;
2985 }
2986
2987 return 1;
2988 }
2989
2990 static int
2991 check_byte_reg (void)
2992 {
2993 int op;
2994
2995 for (op = i.operands; --op >= 0;)
2996 {
2997 /* If this is an eight bit register, it's OK. If it's the 16 or
2998 32 bit version of an eight bit register, we will just use the
2999 low portion, and that's OK too. */
3000 if (i.types[op] & Reg8)
3001 continue;
3002
3003 /* movzx and movsx should not generate this warning. */
3004 if (intel_syntax
3005 && (i.tm.base_opcode == 0xfb7
3006 || i.tm.base_opcode == 0xfb6
3007 || i.tm.base_opcode == 0x63
3008 || i.tm.base_opcode == 0xfbe
3009 || i.tm.base_opcode == 0xfbf))
3010 continue;
3011
3012 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4)
3013 {
3014 /* Prohibit these changes in the 64bit mode, since the
3015 lowering is more complicated. */
3016 if (flag_code == CODE_64BIT
3017 && (i.tm.operand_types[op] & InOutPortReg) == 0)
3018 {
3019 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3020 register_prefix, i.op[op].regs->reg_name,
3021 i.suffix);
3022 return 0;
3023 }
3024 #if REGISTER_WARNINGS
3025 if (!quiet_warnings
3026 && (i.tm.operand_types[op] & InOutPortReg) == 0)
3027 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
3028 (i.op[op].regs + (i.types[op] & Reg16
3029 ? REGNAM_AL - REGNAM_AX
3030 : REGNAM_AL - REGNAM_EAX))->reg_name,
3031 i.op[op].regs->reg_name,
3032 i.suffix);
3033 #endif
3034 continue;
3035 }
3036 /* Any other register is bad. */
3037 if (i.types[op] & (Reg | RegMMX | RegXMM
3038 | SReg2 | SReg3
3039 | Control | Debug | Test
3040 | FloatReg | FloatAcc))
3041 {
3042 as_bad (_("`%%%s' not allowed with `%s%c'"),
3043 i.op[op].regs->reg_name,
3044 i.tm.name,
3045 i.suffix);
3046 return 0;
3047 }
3048 }
3049 return 1;
3050 }
3051
3052 static int
3053 check_long_reg (void)
3054 {
3055 int op;
3056
3057 for (op = i.operands; --op >= 0;)
3058 /* Reject eight bit registers, except where the template requires
3059 them. (eg. movzb) */
3060 if ((i.types[op] & Reg8) != 0
3061 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3062 {
3063 as_bad (_("`%%%s' not allowed with `%s%c'"),
3064 i.op[op].regs->reg_name,
3065 i.tm.name,
3066 i.suffix);
3067 return 0;
3068 }
3069 /* Warn if the e prefix on a general reg is missing. */
3070 else if ((!quiet_warnings || flag_code == CODE_64BIT)
3071 && (i.types[op] & Reg16) != 0
3072 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3073 {
3074 /* Prohibit these changes in the 64bit mode, since the
3075 lowering is more complicated. */
3076 if (flag_code == CODE_64BIT)
3077 {
3078 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3079 register_prefix, i.op[op].regs->reg_name,
3080 i.suffix);
3081 return 0;
3082 }
3083 #if REGISTER_WARNINGS
3084 else
3085 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
3086 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
3087 i.op[op].regs->reg_name,
3088 i.suffix);
3089 #endif
3090 }
3091 /* Warn if the r prefix on a general reg is missing. */
3092 else if ((i.types[op] & Reg64) != 0
3093 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3094 {
3095 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3096 register_prefix, i.op[op].regs->reg_name,
3097 i.suffix);
3098 return 0;
3099 }
3100 return 1;
3101 }
3102
3103 static int
3104 check_qword_reg (void)
3105 {
3106 int op;
3107
3108 for (op = i.operands; --op >= 0; )
3109 /* Reject eight bit registers, except where the template requires
3110 them. (eg. movzb) */
3111 if ((i.types[op] & Reg8) != 0
3112 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3113 {
3114 as_bad (_("`%%%s' not allowed with `%s%c'"),
3115 i.op[op].regs->reg_name,
3116 i.tm.name,
3117 i.suffix);
3118 return 0;
3119 }
3120 /* Warn if the e prefix on a general reg is missing. */
3121 else if (((i.types[op] & Reg16) != 0
3122 || (i.types[op] & Reg32) != 0)
3123 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3124 {
3125 /* Prohibit these changes in the 64bit mode, since the
3126 lowering is more complicated. */
3127 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3128 register_prefix, i.op[op].regs->reg_name,
3129 i.suffix);
3130 return 0;
3131 }
3132 return 1;
3133 }
3134
3135 static int
3136 check_word_reg (void)
3137 {
3138 int op;
3139 for (op = i.operands; --op >= 0;)
3140 /* Reject eight bit registers, except where the template requires
3141 them. (eg. movzb) */
3142 if ((i.types[op] & Reg8) != 0
3143 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3144 {
3145 as_bad (_("`%%%s' not allowed with `%s%c'"),
3146 i.op[op].regs->reg_name,
3147 i.tm.name,
3148 i.suffix);
3149 return 0;
3150 }
3151 /* Warn if the e prefix on a general reg is present. */
3152 else if ((!quiet_warnings || flag_code == CODE_64BIT)
3153 && (i.types[op] & Reg32) != 0
3154 && (i.tm.operand_types[op] & (Reg16 | Acc)) != 0)
3155 {
3156 /* Prohibit these changes in the 64bit mode, since the
3157 lowering is more complicated. */
3158 if (flag_code == CODE_64BIT)
3159 {
3160 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3161 register_prefix, i.op[op].regs->reg_name,
3162 i.suffix);
3163 return 0;
3164 }
3165 else
3166 #if REGISTER_WARNINGS
3167 as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
3168 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
3169 i.op[op].regs->reg_name,
3170 i.suffix);
3171 #endif
3172 }
3173 return 1;
3174 }
3175
3176 static int
3177 finalize_imm (void)
3178 {
3179 unsigned int overlap0, overlap1, overlap2;
3180
3181 overlap0 = i.types[0] & i.tm.operand_types[0];
3182 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64))
3183 && overlap0 != Imm8 && overlap0 != Imm8S
3184 && overlap0 != Imm16 && overlap0 != Imm32S
3185 && overlap0 != Imm32 && overlap0 != Imm64)
3186 {
3187 if (i.suffix)
3188 {
3189 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX
3190 ? Imm8 | Imm8S
3191 : (i.suffix == WORD_MNEM_SUFFIX
3192 ? Imm16
3193 : (i.suffix == QWORD_MNEM_SUFFIX
3194 ? Imm64 | Imm32S
3195 : Imm32)));
3196 }
3197 else if (overlap0 == (Imm16 | Imm32S | Imm32)
3198 || overlap0 == (Imm16 | Imm32)
3199 || overlap0 == (Imm16 | Imm32S))
3200 {
3201 overlap0 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
3202 ? Imm16 : Imm32S);
3203 }
3204 if (overlap0 != Imm8 && overlap0 != Imm8S
3205 && overlap0 != Imm16 && overlap0 != Imm32S
3206 && overlap0 != Imm32 && overlap0 != Imm64)
3207 {
3208 as_bad (_("no instruction mnemonic suffix given; "
3209 "can't determine immediate size"));
3210 return 0;
3211 }
3212 }
3213 i.types[0] = overlap0;
3214
3215 overlap1 = i.types[1] & i.tm.operand_types[1];
3216 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32 | Imm64))
3217 && overlap1 != Imm8 && overlap1 != Imm8S
3218 && overlap1 != Imm16 && overlap1 != Imm32S
3219 && overlap1 != Imm32 && overlap1 != Imm64)
3220 {
3221 if (i.suffix)
3222 {
3223 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX
3224 ? Imm8 | Imm8S
3225 : (i.suffix == WORD_MNEM_SUFFIX
3226 ? Imm16
3227 : (i.suffix == QWORD_MNEM_SUFFIX
3228 ? Imm64 | Imm32S
3229 : Imm32)));
3230 }
3231 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
3232 || overlap1 == (Imm16 | Imm32)
3233 || overlap1 == (Imm16 | Imm32S))
3234 {
3235 overlap1 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
3236 ? Imm16 : Imm32S);
3237 }
3238 if (overlap1 != Imm8 && overlap1 != Imm8S
3239 && overlap1 != Imm16 && overlap1 != Imm32S
3240 && overlap1 != Imm32 && overlap1 != Imm64)
3241 {
3242 as_bad (_("no instruction mnemonic suffix given; "
3243 "can't determine immediate size %x %c"),
3244 overlap1, i.suffix);
3245 return 0;
3246 }
3247 }
3248 i.types[1] = overlap1;
3249
3250 overlap2 = i.types[2] & i.tm.operand_types[2];
3251 assert ((overlap2 & Imm) == 0);
3252 i.types[2] = overlap2;
3253
3254 return 1;
3255 }
3256
3257 static int
3258 process_operands (void)
3259 {
3260 /* Default segment register this instruction will use for memory
3261 accesses. 0 means unknown. This is only for optimizing out
3262 unnecessary segment overrides. */
3263 const seg_entry *default_seg = 0;
3264
3265 /* The imul $imm, %reg instruction is converted into
3266 imul $imm, %reg, %reg, and the clr %reg instruction
3267 is converted into xor %reg, %reg. */
3268 if (i.tm.opcode_modifier & regKludge)
3269 {
3270 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
3271 /* Pretend we saw the extra register operand. */
3272 assert (i.reg_operands == 1
3273 && i.op[first_reg_op + 1].regs == 0);
3274 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
3275 i.types[first_reg_op + 1] = i.types[first_reg_op];
3276 i.operands++;
3277 i.reg_operands++;
3278 }
3279
3280 if (i.tm.opcode_modifier & ShortForm)
3281 {
3282 if (i.types[0] & (SReg2 | SReg3))
3283 {
3284 if (i.tm.base_opcode == POP_SEG_SHORT
3285 && i.op[0].regs->reg_num == 1)
3286 {
3287 as_bad (_("you can't `pop %%cs'"));
3288 return 0;
3289 }
3290 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
3291 if ((i.op[0].regs->reg_flags & RegRex) != 0)
3292 i.rex |= REX_EXTZ;
3293 }
3294 else
3295 {
3296 /* The register or float register operand is in operand 0 or 1. */
3297 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
3298 /* Register goes in low 3 bits of opcode. */
3299 i.tm.base_opcode |= i.op[op].regs->reg_num;
3300 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3301 i.rex |= REX_EXTZ;
3302 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
3303 {
3304 /* Warn about some common errors, but press on regardless.
3305 The first case can be generated by gcc (<= 2.8.1). */
3306 if (i.operands == 2)
3307 {
3308 /* Reversed arguments on faddp, fsubp, etc. */
3309 as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
3310 i.op[1].regs->reg_name,
3311 i.op[0].regs->reg_name);
3312 }
3313 else
3314 {
3315 /* Extraneous `l' suffix on fp insn. */
3316 as_warn (_("translating to `%s %%%s'"), i.tm.name,
3317 i.op[0].regs->reg_name);
3318 }
3319 }
3320 }
3321 }
3322 else if (i.tm.opcode_modifier & Modrm)
3323 {
3324 /* The opcode is completed (modulo i.tm.extension_opcode which
3325 must be put into the modrm byte). Now, we make the modrm and
3326 index base bytes based on all the info we've collected. */
3327
3328 default_seg = build_modrm_byte ();
3329 }
3330 else if ((i.tm.base_opcode & ~0x3) == MOV_AX_DISP32)
3331 {
3332 default_seg = &ds;
3333 }
3334 else if ((i.tm.opcode_modifier & IsString) != 0)
3335 {
3336 /* For the string instructions that allow a segment override
3337 on one of their operands, the default segment is ds. */
3338 default_seg = &ds;
3339 }
3340
3341 if ((i.tm.base_opcode == 0x8d /* lea */
3342 || (i.tm.cpu_flags & CpuSVME))
3343 && i.seg[0] && !quiet_warnings)
3344 as_warn (_("segment override on `%s' is ineffectual"), i.tm.name);
3345
3346 /* If a segment was explicitly specified, and the specified segment
3347 is not the default, use an opcode prefix to select it. If we
3348 never figured out what the default segment is, then default_seg
3349 will be zero at this point, and the specified segment prefix will
3350 always be used. */
3351 if ((i.seg[0]) && (i.seg[0] != default_seg))
3352 {
3353 if (!add_prefix (i.seg[0]->seg_prefix))
3354 return 0;
3355 }
3356 return 1;
3357 }
3358
3359 static const seg_entry *
3360 build_modrm_byte (void)
3361 {
3362 const seg_entry *default_seg = 0;
3363
3364 /* i.reg_operands MUST be the number of real register operands;
3365 implicit registers do not count. */
3366 if (i.reg_operands == 2)
3367 {
3368 unsigned int source, dest;
3369
3370 switch (i.operands)
3371 {
3372 case 2:
3373 source = 0;
3374 break;
3375 case 3:
3376 /* When there are 3 operands, one of them may be immediate,
3377 which may be the first or the last operand. Otherwise,
3378 the first operand must be shift count register (cl). */
3379 assert (i.imm_operands == 1
3380 || (i.imm_operands == 0
3381 && (i.types[0] & ShiftCount)));
3382 source = (i.types[0] & (Imm | ShiftCount)) ? 1 : 0;
3383 break;
3384 case 4:
3385 /* When there are 4 operands, the first two must be immediate
3386 operands. The source operand will be the 3rd one. */
3387 assert (i.imm_operands == 2
3388 && (i.types[0] & Imm)
3389 && (i.types[1] & Imm));
3390 source = 2;
3391 break;
3392 default:
3393 abort ();
3394 }
3395
3396 dest = source + 1;
3397
3398 i.rm.mode = 3;
3399 /* One of the register operands will be encoded in the i.tm.reg
3400 field, the other in the combined i.tm.mode and i.tm.regmem
3401 fields. If no form of this instruction supports a memory
3402 destination operand, then we assume the source operand may
3403 sometimes be a memory operand and so we need to store the
3404 destination in the i.rm.reg field. */
3405 if ((i.tm.operand_types[dest] & AnyMem) == 0)
3406 {
3407 i.rm.reg = i.op[dest].regs->reg_num;
3408 i.rm.regmem = i.op[source].regs->reg_num;
3409 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
3410 i.rex |= REX_EXTX;
3411 if ((i.op[source].regs->reg_flags & RegRex) != 0)
3412 i.rex |= REX_EXTZ;
3413 }
3414 else
3415 {
3416 i.rm.reg = i.op[source].regs->reg_num;
3417 i.rm.regmem = i.op[dest].regs->reg_num;
3418 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
3419 i.rex |= REX_EXTZ;
3420 if ((i.op[source].regs->reg_flags & RegRex) != 0)
3421 i.rex |= REX_EXTX;
3422 }
3423 if (flag_code != CODE_64BIT && (i.rex & (REX_EXTX | REX_EXTZ)))
3424 {
3425 if (!((i.types[0] | i.types[1]) & Control))
3426 abort ();
3427 i.rex &= ~(REX_EXTX | REX_EXTZ);
3428 add_prefix (LOCK_PREFIX_OPCODE);
3429 }
3430 }
3431 else
3432 { /* If it's not 2 reg operands... */
3433 if (i.mem_operands)
3434 {
3435 unsigned int fake_zero_displacement = 0;
3436 unsigned int op;
3437
3438 for (op = 0; op < i.operands; op++)
3439 if ((i.types[op] & AnyMem))
3440 break;
3441 assert (op < i.operands);
3442
3443 default_seg = &ds;
3444
3445 if (i.base_reg == 0)
3446 {
3447 i.rm.mode = 0;
3448 if (!i.disp_operands)
3449 fake_zero_displacement = 1;
3450 if (i.index_reg == 0)
3451 {
3452 /* Operand is just <disp> */
3453 if (flag_code == CODE_64BIT)
3454 {
3455 /* 64bit mode overwrites the 32bit absolute
3456 addressing by RIP relative addressing and
3457 absolute addressing is encoded by one of the
3458 redundant SIB forms. */
3459 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3460 i.sib.base = NO_BASE_REGISTER;
3461 i.sib.index = NO_INDEX_REGISTER;
3462 i.types[op] = ((i.prefix[ADDR_PREFIX] == 0)
3463 ? Disp32S : Disp32);
3464 }
3465 else if ((flag_code == CODE_16BIT)
3466 ^ (i.prefix[ADDR_PREFIX] != 0))
3467 {
3468 i.rm.regmem = NO_BASE_REGISTER_16;
3469 i.types[op] = Disp16;
3470 }
3471 else
3472 {
3473 i.rm.regmem = NO_BASE_REGISTER;
3474 i.types[op] = Disp32;
3475 }
3476 }
3477 else /* !i.base_reg && i.index_reg */
3478 {
3479 i.sib.index = i.index_reg->reg_num;
3480 i.sib.base = NO_BASE_REGISTER;
3481 i.sib.scale = i.log2_scale_factor;
3482 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3483 i.types[op] &= ~Disp;
3484 if (flag_code != CODE_64BIT)
3485 i.types[op] |= Disp32; /* Must be 32 bit */
3486 else
3487 i.types[op] |= Disp32S;
3488 if ((i.index_reg->reg_flags & RegRex) != 0)
3489 i.rex |= REX_EXTY;
3490 }
3491 }
3492 /* RIP addressing for 64bit mode. */
3493 else if (i.base_reg->reg_type == BaseIndex)
3494 {
3495 i.rm.regmem = NO_BASE_REGISTER;
3496 i.types[op] &= ~ Disp;
3497 i.types[op] |= Disp32S;
3498 i.flags[op] |= Operand_PCrel;
3499 if (! i.disp_operands)
3500 fake_zero_displacement = 1;
3501 }
3502 else if (i.base_reg->reg_type & Reg16)
3503 {
3504 switch (i.base_reg->reg_num)
3505 {
3506 case 3: /* (%bx) */
3507 if (i.index_reg == 0)
3508 i.rm.regmem = 7;
3509 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
3510 i.rm.regmem = i.index_reg->reg_num - 6;
3511 break;
3512 case 5: /* (%bp) */
3513 default_seg = &ss;
3514 if (i.index_reg == 0)
3515 {
3516 i.rm.regmem = 6;
3517 if ((i.types[op] & Disp) == 0)
3518 {
3519 /* fake (%bp) into 0(%bp) */
3520 i.types[op] |= Disp8;
3521 fake_zero_displacement = 1;
3522 }
3523 }
3524 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
3525 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
3526 break;
3527 default: /* (%si) -> 4 or (%di) -> 5 */
3528 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
3529 }
3530 i.rm.mode = mode_from_disp_size (i.types[op]);
3531 }
3532 else /* i.base_reg and 32/64 bit mode */
3533 {
3534 if (flag_code == CODE_64BIT
3535 && (i.types[op] & Disp))
3536 i.types[op] = ((i.types[op] & Disp8)
3537 | (i.prefix[ADDR_PREFIX] == 0
3538 ? Disp32S : Disp32));
3539
3540 i.rm.regmem = i.base_reg->reg_num;
3541 if ((i.base_reg->reg_flags & RegRex) != 0)
3542 i.rex |= REX_EXTZ;
3543 i.sib.base = i.base_reg->reg_num;
3544 /* x86-64 ignores REX prefix bit here to avoid decoder
3545 complications. */
3546 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
3547 {
3548 default_seg = &ss;
3549 if (i.disp_operands == 0)
3550 {
3551 fake_zero_displacement = 1;
3552 i.types[op] |= Disp8;
3553 }
3554 }
3555 else if (i.base_reg->reg_num == ESP_REG_NUM)
3556 {
3557 default_seg = &ss;
3558 }
3559 i.sib.scale = i.log2_scale_factor;
3560 if (i.index_reg == 0)
3561 {
3562 /* <disp>(%esp) becomes two byte modrm with no index
3563 register. We've already stored the code for esp
3564 in i.rm.regmem ie. ESCAPE_TO_TWO_BYTE_ADDRESSING.
3565 Any base register besides %esp will not use the
3566 extra modrm byte. */
3567 i.sib.index = NO_INDEX_REGISTER;
3568 #if !SCALE1_WHEN_NO_INDEX
3569 /* Another case where we force the second modrm byte. */
3570 if (i.log2_scale_factor)
3571 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3572 #endif
3573 }
3574 else
3575 {
3576 i.sib.index = i.index_reg->reg_num;
3577 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3578 if ((i.index_reg->reg_flags & RegRex) != 0)
3579 i.rex |= REX_EXTY;
3580 }
3581
3582 if (i.disp_operands
3583 && (i.reloc[op] == BFD_RELOC_386_TLS_DESC_CALL
3584 || i.reloc[op] == BFD_RELOC_X86_64_TLSDESC_CALL))
3585 i.rm.mode = 0;
3586 else
3587 i.rm.mode = mode_from_disp_size (i.types[op]);
3588 }
3589
3590 if (fake_zero_displacement)
3591 {
3592 /* Fakes a zero displacement assuming that i.types[op]
3593 holds the correct displacement size. */
3594 expressionS *exp;
3595
3596 assert (i.op[op].disps == 0);
3597 exp = &disp_expressions[i.disp_operands++];
3598 i.op[op].disps = exp;
3599 exp->X_op = O_constant;
3600 exp->X_add_number = 0;
3601 exp->X_add_symbol = (symbolS *) 0;
3602 exp->X_op_symbol = (symbolS *) 0;
3603 }
3604 }
3605
3606 /* Fill in i.rm.reg or i.rm.regmem field with register operand
3607 (if any) based on i.tm.extension_opcode. Again, we must be
3608 careful to make sure that segment/control/debug/test/MMX
3609 registers are coded into the i.rm.reg field. */
3610 if (i.reg_operands)
3611 {
3612 unsigned int op;
3613
3614 for (op = 0; op < i.operands; op++)
3615 if ((i.types[op] & (Reg | RegMMX | RegXMM
3616 | SReg2 | SReg3
3617 | Control | Debug | Test)))
3618 break;
3619 assert (op < i.operands);
3620
3621 /* If there is an extension opcode to put here, the register
3622 number must be put into the regmem field. */
3623 if (i.tm.extension_opcode != None)
3624 {
3625 i.rm.regmem = i.op[op].regs->reg_num;
3626 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3627 i.rex |= REX_EXTZ;
3628 }
3629 else
3630 {
3631 i.rm.reg = i.op[op].regs->reg_num;
3632 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3633 i.rex |= REX_EXTX;
3634 }
3635
3636 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2 we
3637 must set it to 3 to indicate this is a register operand
3638 in the regmem field. */
3639 if (!i.mem_operands)
3640 i.rm.mode = 3;
3641 }
3642
3643 /* Fill in i.rm.reg field with extension opcode (if any). */
3644 if (i.tm.extension_opcode != None)
3645 i.rm.reg = i.tm.extension_opcode;
3646 }
3647 return default_seg;
3648 }
3649
3650 static void
3651 output_branch (void)
3652 {
3653 char *p;
3654 int code16;
3655 int prefix;
3656 relax_substateT subtype;
3657 symbolS *sym;
3658 offsetT off;
3659
3660 code16 = 0;
3661 if (flag_code == CODE_16BIT)
3662 code16 = CODE16;
3663
3664 prefix = 0;
3665 if (i.prefix[DATA_PREFIX] != 0)
3666 {
3667 prefix = 1;
3668 i.prefixes -= 1;
3669 code16 ^= CODE16;
3670 }
3671 /* Pentium4 branch hints. */
3672 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3673 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3674 {
3675 prefix++;
3676 i.prefixes--;
3677 }
3678 if (i.prefix[REX_PREFIX] != 0)
3679 {
3680 prefix++;
3681 i.prefixes--;
3682 }
3683
3684 if (i.prefixes != 0 && !intel_syntax)
3685 as_warn (_("skipping prefixes on this instruction"));
3686
3687 /* It's always a symbol; End frag & setup for relax.
3688 Make sure there is enough room in this frag for the largest
3689 instruction we may generate in md_convert_frag. This is 2
3690 bytes for the opcode and room for the prefix and largest
3691 displacement. */
3692 frag_grow (prefix + 2 + 4);
3693 /* Prefix and 1 opcode byte go in fr_fix. */
3694 p = frag_more (prefix + 1);
3695 if (i.prefix[DATA_PREFIX] != 0)
3696 *p++ = DATA_PREFIX_OPCODE;
3697 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
3698 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
3699 *p++ = i.prefix[SEG_PREFIX];
3700 if (i.prefix[REX_PREFIX] != 0)
3701 *p++ = i.prefix[REX_PREFIX];
3702 *p = i.tm.base_opcode;
3703
3704 if ((unsigned char) *p == JUMP_PC_RELATIVE)
3705 subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
3706 else if ((cpu_arch_flags & Cpu386) != 0)
3707 subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
3708 else
3709 subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
3710 subtype |= code16;
3711
3712 sym = i.op[0].disps->X_add_symbol;
3713 off = i.op[0].disps->X_add_number;
3714
3715 if (i.op[0].disps->X_op != O_constant
3716 && i.op[0].disps->X_op != O_symbol)
3717 {
3718 /* Handle complex expressions. */
3719 sym = make_expr_symbol (i.op[0].disps);
3720 off = 0;
3721 }
3722
3723 /* 1 possible extra opcode + 4 byte displacement go in var part.
3724 Pass reloc in fr_var. */
3725 frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
3726 }
3727
3728 static void
3729 output_jump (void)
3730 {
3731 char *p;
3732 int size;
3733 fixS *fixP;
3734
3735 if (i.tm.opcode_modifier & JumpByte)
3736 {
3737 /* This is a loop or jecxz type instruction. */
3738 size = 1;
3739 if (i.prefix[ADDR_PREFIX] != 0)
3740 {
3741 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
3742 i.prefixes -= 1;
3743 }
3744 /* Pentium4 branch hints. */
3745 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3746 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3747 {
3748 FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
3749 i.prefixes--;
3750 }
3751 }
3752 else
3753 {
3754 int code16;
3755
3756 code16 = 0;
3757 if (flag_code == CODE_16BIT)
3758 code16 = CODE16;
3759
3760 if (i.prefix[DATA_PREFIX] != 0)
3761 {
3762 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
3763 i.prefixes -= 1;
3764 code16 ^= CODE16;
3765 }
3766
3767 size = 4;
3768 if (code16)
3769 size = 2;
3770 }
3771
3772 if (i.prefix[REX_PREFIX] != 0)
3773 {
3774 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
3775 i.prefixes -= 1;
3776 }
3777
3778 if (i.prefixes != 0 && !intel_syntax)
3779 as_warn (_("skipping prefixes on this instruction"));
3780
3781 p = frag_more (1 + size);
3782 *p++ = i.tm.base_opcode;
3783
3784 fixP = fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3785 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
3786
3787 /* All jumps handled here are signed, but don't use a signed limit
3788 check for 32 and 16 bit jumps as we want to allow wrap around at
3789 4G and 64k respectively. */
3790 if (size == 1)
3791 fixP->fx_signed = 1;
3792 }
3793
3794 static void
3795 output_interseg_jump (void)
3796 {
3797 char *p;
3798 int size;
3799 int prefix;
3800 int code16;
3801
3802 code16 = 0;
3803 if (flag_code == CODE_16BIT)
3804 code16 = CODE16;
3805
3806 prefix = 0;
3807 if (i.prefix[DATA_PREFIX] != 0)
3808 {
3809 prefix = 1;
3810 i.prefixes -= 1;
3811 code16 ^= CODE16;
3812 }
3813 if (i.prefix[REX_PREFIX] != 0)
3814 {
3815 prefix++;
3816 i.prefixes -= 1;
3817 }
3818
3819 size = 4;
3820 if (code16)
3821 size = 2;
3822
3823 if (i.prefixes != 0 && !intel_syntax)
3824 as_warn (_("skipping prefixes on this instruction"));
3825
3826 /* 1 opcode; 2 segment; offset */
3827 p = frag_more (prefix + 1 + 2 + size);
3828
3829 if (i.prefix[DATA_PREFIX] != 0)
3830 *p++ = DATA_PREFIX_OPCODE;
3831
3832 if (i.prefix[REX_PREFIX] != 0)
3833 *p++ = i.prefix[REX_PREFIX];
3834
3835 *p++ = i.tm.base_opcode;
3836 if (i.op[1].imms->X_op == O_constant)
3837 {
3838 offsetT n = i.op[1].imms->X_add_number;
3839
3840 if (size == 2
3841 && !fits_in_unsigned_word (n)
3842 && !fits_in_signed_word (n))
3843 {
3844 as_bad (_("16-bit jump out of range"));
3845 return;
3846 }
3847 md_number_to_chars (p, n, size);
3848 }
3849 else
3850 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3851 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
3852 if (i.op[0].imms->X_op != O_constant)
3853 as_bad (_("can't handle non absolute segment in `%s'"),
3854 i.tm.name);
3855 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
3856 }
3857
3858 static void
3859 output_insn (void)
3860 {
3861 fragS *insn_start_frag;
3862 offsetT insn_start_off;
3863
3864 /* Tie dwarf2 debug info to the address at the start of the insn.
3865 We can't do this after the insn has been output as the current
3866 frag may have been closed off. eg. by frag_var. */
3867 dwarf2_emit_insn (0);
3868
3869 insn_start_frag = frag_now;
3870 insn_start_off = frag_now_fix ();
3871
3872 /* Output jumps. */
3873 if (i.tm.opcode_modifier & Jump)
3874 output_branch ();
3875 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
3876 output_jump ();
3877 else if (i.tm.opcode_modifier & JumpInterSegment)
3878 output_interseg_jump ();
3879 else
3880 {
3881 /* Output normal instructions here. */
3882 char *p;
3883 unsigned char *q;
3884 unsigned int prefix;
3885
3886 /* All opcodes on i386 have either 1 or 2 bytes. Supplemental
3887 Streaming SIMD extensions 3 Instructions have 3 bytes. We may
3888 use one more higher byte to specify a prefix the instruction
3889 requires. */
3890 if ((i.tm.cpu_flags & CpuSSSE3) != 0)
3891 {
3892 if (i.tm.base_opcode & 0xff000000)
3893 {
3894 prefix = (i.tm.base_opcode >> 24) & 0xff;
3895 goto check_prefix;
3896 }
3897 }
3898 else if ((i.tm.base_opcode & 0xff0000) != 0)
3899 {
3900 prefix = (i.tm.base_opcode >> 16) & 0xff;
3901 if ((i.tm.cpu_flags & CpuPadLock) != 0)
3902 {
3903 check_prefix:
3904 if (prefix != REPE_PREFIX_OPCODE
3905 || i.prefix[LOCKREP_PREFIX] != REPE_PREFIX_OPCODE)
3906 add_prefix (prefix);
3907 }
3908 else
3909 add_prefix (prefix);
3910 }
3911
3912 /* The prefix bytes. */
3913 for (q = i.prefix;
3914 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
3915 q++)
3916 {
3917 if (*q)
3918 {
3919 p = frag_more (1);
3920 md_number_to_chars (p, (valueT) *q, 1);
3921 }
3922 }
3923
3924 /* Now the opcode; be careful about word order here! */
3925 if (fits_in_unsigned_byte (i.tm.base_opcode))
3926 {
3927 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
3928 }
3929 else
3930 {
3931 if ((i.tm.cpu_flags & CpuSSSE3) != 0)
3932 {
3933 p = frag_more (3);
3934 *p++ = (i.tm.base_opcode >> 16) & 0xff;
3935 }
3936 else
3937 p = frag_more (2);
3938
3939 /* Put out high byte first: can't use md_number_to_chars! */
3940 *p++ = (i.tm.base_opcode >> 8) & 0xff;
3941 *p = i.tm.base_opcode & 0xff;
3942 }
3943
3944 /* Now the modrm byte and sib byte (if present). */
3945 if (i.tm.opcode_modifier & Modrm)
3946 {
3947 p = frag_more (1);
3948 md_number_to_chars (p,
3949 (valueT) (i.rm.regmem << 0
3950 | i.rm.reg << 3
3951 | i.rm.mode << 6),
3952 1);
3953 /* If i.rm.regmem == ESP (4)
3954 && i.rm.mode != (Register mode)
3955 && not 16 bit
3956 ==> need second modrm byte. */
3957 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
3958 && i.rm.mode != 3
3959 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
3960 {
3961 p = frag_more (1);
3962 md_number_to_chars (p,
3963 (valueT) (i.sib.base << 0
3964 | i.sib.index << 3
3965 | i.sib.scale << 6),
3966 1);
3967 }
3968 }
3969
3970 if (i.disp_operands)
3971 output_disp (insn_start_frag, insn_start_off);
3972
3973 if (i.imm_operands)
3974 output_imm (insn_start_frag, insn_start_off);
3975 }
3976
3977 #ifdef DEBUG386
3978 if (flag_debug)
3979 {
3980 pi ("" /*line*/, &i);
3981 }
3982 #endif /* DEBUG386 */
3983 }
3984
3985 static void
3986 output_disp (fragS *insn_start_frag, offsetT insn_start_off)
3987 {
3988 char *p;
3989 unsigned int n;
3990
3991 for (n = 0; n < i.operands; n++)
3992 {
3993 if (i.types[n] & Disp)
3994 {
3995 if (i.op[n].disps->X_op == O_constant)
3996 {
3997 int size;
3998 offsetT val;
3999
4000 size = 4;
4001 if (i.types[n] & (Disp8 | Disp16 | Disp64))
4002 {
4003 size = 2;
4004 if (i.types[n] & Disp8)
4005 size = 1;
4006 if (i.types[n] & Disp64)
4007 size = 8;
4008 }
4009 val = offset_in_range (i.op[n].disps->X_add_number,
4010 size);
4011 p = frag_more (size);
4012 md_number_to_chars (p, val, size);
4013 }
4014 else
4015 {
4016 enum bfd_reloc_code_real reloc_type;
4017 int size = 4;
4018 int sign = 0;
4019 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
4020
4021 /* The PC relative address is computed relative
4022 to the instruction boundary, so in case immediate
4023 fields follows, we need to adjust the value. */
4024 if (pcrel && i.imm_operands)
4025 {
4026 int imm_size = 4;
4027 unsigned int n1;
4028
4029 for (n1 = 0; n1 < i.operands; n1++)
4030 if (i.types[n1] & Imm)
4031 {
4032 if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
4033 {
4034 imm_size = 2;
4035 if (i.types[n1] & (Imm8 | Imm8S))
4036 imm_size = 1;
4037 if (i.types[n1] & Imm64)
4038 imm_size = 8;
4039 }
4040 break;
4041 }
4042 /* We should find the immediate. */
4043 if (n1 == i.operands)
4044 abort ();
4045 i.op[n].disps->X_add_number -= imm_size;
4046 }
4047
4048 if (i.types[n] & Disp32S)
4049 sign = 1;
4050
4051 if (i.types[n] & (Disp16 | Disp64))
4052 {
4053 size = 2;
4054 if (i.types[n] & Disp64)
4055 size = 8;
4056 }
4057
4058 p = frag_more (size);
4059 reloc_type = reloc (size, pcrel, sign, i.reloc[n]);
4060 if (GOT_symbol
4061 && GOT_symbol == i.op[n].disps->X_add_symbol
4062 && (((reloc_type == BFD_RELOC_32
4063 || reloc_type == BFD_RELOC_X86_64_32S
4064 || (reloc_type == BFD_RELOC_64
4065 && object_64bit))
4066 && (i.op[n].disps->X_op == O_symbol
4067 || (i.op[n].disps->X_op == O_add
4068 && ((symbol_get_value_expression
4069 (i.op[n].disps->X_op_symbol)->X_op)
4070 == O_subtract))))
4071 || reloc_type == BFD_RELOC_32_PCREL))
4072 {
4073 offsetT add;
4074
4075 if (insn_start_frag == frag_now)
4076 add = (p - frag_now->fr_literal) - insn_start_off;
4077 else
4078 {
4079 fragS *fr;
4080
4081 add = insn_start_frag->fr_fix - insn_start_off;
4082 for (fr = insn_start_frag->fr_next;
4083 fr && fr != frag_now; fr = fr->fr_next)
4084 add += fr->fr_fix;
4085 add += p - frag_now->fr_literal;
4086 }
4087
4088 if (!object_64bit)
4089 {
4090 reloc_type = BFD_RELOC_386_GOTPC;
4091 i.op[n].imms->X_add_number += add;
4092 }
4093 else if (reloc_type == BFD_RELOC_64)
4094 reloc_type = BFD_RELOC_X86_64_GOTPC64;
4095 else
4096 /* Don't do the adjustment for x86-64, as there
4097 the pcrel addressing is relative to the _next_
4098 insn, and that is taken care of in other code. */
4099 reloc_type = BFD_RELOC_X86_64_GOTPC32;
4100 }
4101 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
4102 i.op[n].disps, pcrel, reloc_type);
4103 }
4104 }
4105 }
4106 }
4107
4108 static void
4109 output_imm (fragS *insn_start_frag, offsetT insn_start_off)
4110 {
4111 char *p;
4112 unsigned int n;
4113
4114 for (n = 0; n < i.operands; n++)
4115 {
4116 if (i.types[n] & Imm)
4117 {
4118 if (i.op[n].imms->X_op == O_constant)
4119 {
4120 int size;
4121 offsetT val;
4122
4123 size = 4;
4124 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
4125 {
4126 size = 2;
4127 if (i.types[n] & (Imm8 | Imm8S))
4128 size = 1;
4129 else if (i.types[n] & Imm64)
4130 size = 8;
4131 }
4132 val = offset_in_range (i.op[n].imms->X_add_number,
4133 size);
4134 p = frag_more (size);
4135 md_number_to_chars (p, val, size);
4136 }
4137 else
4138 {
4139 /* Not absolute_section.
4140 Need a 32-bit fixup (don't support 8bit
4141 non-absolute imms). Try to support other
4142 sizes ... */
4143 enum bfd_reloc_code_real reloc_type;
4144 int size = 4;
4145 int sign = 0;
4146
4147 if ((i.types[n] & (Imm32S))
4148 && (i.suffix == QWORD_MNEM_SUFFIX
4149 || (!i.suffix && (i.tm.opcode_modifier & No_lSuf))))
4150 sign = 1;
4151 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
4152 {
4153 size = 2;
4154 if (i.types[n] & (Imm8 | Imm8S))
4155 size = 1;
4156 if (i.types[n] & Imm64)
4157 size = 8;
4158 }
4159
4160 p = frag_more (size);
4161 reloc_type = reloc (size, 0, sign, i.reloc[n]);
4162
4163 /* This is tough to explain. We end up with this one if we
4164 * have operands that look like
4165 * "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal here is to
4166 * obtain the absolute address of the GOT, and it is strongly
4167 * preferable from a performance point of view to avoid using
4168 * a runtime relocation for this. The actual sequence of
4169 * instructions often look something like:
4170 *
4171 * call .L66
4172 * .L66:
4173 * popl %ebx
4174 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
4175 *
4176 * The call and pop essentially return the absolute address
4177 * of the label .L66 and store it in %ebx. The linker itself
4178 * will ultimately change the first operand of the addl so
4179 * that %ebx points to the GOT, but to keep things simple, the
4180 * .o file must have this operand set so that it generates not
4181 * the absolute address of .L66, but the absolute address of
4182 * itself. This allows the linker itself simply treat a GOTPC
4183 * relocation as asking for a pcrel offset to the GOT to be
4184 * added in, and the addend of the relocation is stored in the
4185 * operand field for the instruction itself.
4186 *
4187 * Our job here is to fix the operand so that it would add
4188 * the correct offset so that %ebx would point to itself. The
4189 * thing that is tricky is that .-.L66 will point to the
4190 * beginning of the instruction, so we need to further modify
4191 * the operand so that it will point to itself. There are
4192 * other cases where you have something like:
4193 *
4194 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
4195 *
4196 * and here no correction would be required. Internally in
4197 * the assembler we treat operands of this form as not being
4198 * pcrel since the '.' is explicitly mentioned, and I wonder
4199 * whether it would simplify matters to do it this way. Who
4200 * knows. In earlier versions of the PIC patches, the
4201 * pcrel_adjust field was used to store the correction, but
4202 * since the expression is not pcrel, I felt it would be
4203 * confusing to do it this way. */
4204
4205 if ((reloc_type == BFD_RELOC_32
4206 || reloc_type == BFD_RELOC_X86_64_32S
4207 || reloc_type == BFD_RELOC_64)
4208 && GOT_symbol
4209 && GOT_symbol == i.op[n].imms->X_add_symbol
4210 && (i.op[n].imms->X_op == O_symbol
4211 || (i.op[n].imms->X_op == O_add
4212 && ((symbol_get_value_expression
4213 (i.op[n].imms->X_op_symbol)->X_op)
4214 == O_subtract))))
4215 {
4216 offsetT add;
4217
4218 if (insn_start_frag == frag_now)
4219 add = (p - frag_now->fr_literal) - insn_start_off;
4220 else
4221 {
4222 fragS *fr;
4223
4224 add = insn_start_frag->fr_fix - insn_start_off;
4225 for (fr = insn_start_frag->fr_next;
4226 fr && fr != frag_now; fr = fr->fr_next)
4227 add += fr->fr_fix;
4228 add += p - frag_now->fr_literal;
4229 }
4230
4231 if (!object_64bit)
4232 reloc_type = BFD_RELOC_386_GOTPC;
4233 else if (size == 4)
4234 reloc_type = BFD_RELOC_X86_64_GOTPC32;
4235 else if (size == 8)
4236 reloc_type = BFD_RELOC_X86_64_GOTPC64;
4237 i.op[n].imms->X_add_number += add;
4238 }
4239 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
4240 i.op[n].imms, 0, reloc_type);
4241 }
4242 }
4243 }
4244 }
4245 \f
4246 /* x86_cons_fix_new is called via the expression parsing code when a
4247 reloc is needed. We use this hook to get the correct .got reloc. */
4248 static enum bfd_reloc_code_real got_reloc = NO_RELOC;
4249 static int cons_sign = -1;
4250
4251 void
4252 x86_cons_fix_new (fragS *frag, unsigned int off, unsigned int len,
4253 expressionS *exp)
4254 {
4255 enum bfd_reloc_code_real r = reloc (len, 0, cons_sign, got_reloc);
4256
4257 got_reloc = NO_RELOC;
4258
4259 #ifdef TE_PE
4260 if (exp->X_op == O_secrel)
4261 {
4262 exp->X_op = O_symbol;
4263 r = BFD_RELOC_32_SECREL;
4264 }
4265 #endif
4266
4267 fix_new_exp (frag, off, len, exp, 0, r);
4268 }
4269
4270 #if (!defined (OBJ_ELF) && !defined (OBJ_MAYBE_ELF)) || defined (LEX_AT)
4271 # define lex_got(reloc, adjust, types) NULL
4272 #else
4273 /* Parse operands of the form
4274 <symbol>@GOTOFF+<nnn>
4275 and similar .plt or .got references.
4276
4277 If we find one, set up the correct relocation in RELOC and copy the
4278 input string, minus the `@GOTOFF' into a malloc'd buffer for
4279 parsing by the calling routine. Return this buffer, and if ADJUST
4280 is non-null set it to the length of the string we removed from the
4281 input line. Otherwise return NULL. */
4282 static char *
4283 lex_got (enum bfd_reloc_code_real *reloc,
4284 int *adjust,
4285 unsigned int *types)
4286 {
4287 /* Some of the relocations depend on the size of what field is to
4288 be relocated. But in our callers i386_immediate and i386_displacement
4289 we don't yet know the operand size (this will be set by insn
4290 matching). Hence we record the word32 relocation here,
4291 and adjust the reloc according to the real size in reloc(). */
4292 static const struct {
4293 const char *str;
4294 const enum bfd_reloc_code_real rel[2];
4295 const unsigned int types64;
4296 } gotrel[] = {
4297 { "PLTOFF", { 0,
4298 BFD_RELOC_X86_64_PLTOFF64 },
4299 Imm64 },
4300 { "PLT", { BFD_RELOC_386_PLT32,
4301 BFD_RELOC_X86_64_PLT32 },
4302 Imm32 | Imm32S | Disp32 },
4303 { "GOTPLT", { 0,
4304 BFD_RELOC_X86_64_GOTPLT64 },
4305 Imm64 | Disp64 },
4306 { "GOTOFF", { BFD_RELOC_386_GOTOFF,
4307 BFD_RELOC_X86_64_GOTOFF64 },
4308 Imm64 | Disp64 },
4309 { "GOTPCREL", { 0,
4310 BFD_RELOC_X86_64_GOTPCREL },
4311 Imm32 | Imm32S | Disp32 },
4312 { "TLSGD", { BFD_RELOC_386_TLS_GD,
4313 BFD_RELOC_X86_64_TLSGD },
4314 Imm32 | Imm32S | Disp32 },
4315 { "TLSLDM", { BFD_RELOC_386_TLS_LDM,
4316 0 },
4317 0 },
4318 { "TLSLD", { 0,
4319 BFD_RELOC_X86_64_TLSLD },
4320 Imm32 | Imm32S | Disp32 },
4321 { "GOTTPOFF", { BFD_RELOC_386_TLS_IE_32,
4322 BFD_RELOC_X86_64_GOTTPOFF },
4323 Imm32 | Imm32S | Disp32 },
4324 { "TPOFF", { BFD_RELOC_386_TLS_LE_32,
4325 BFD_RELOC_X86_64_TPOFF32 },
4326 Imm32 | Imm32S | Imm64 | Disp32 | Disp64 },
4327 { "NTPOFF", { BFD_RELOC_386_TLS_LE,
4328 0 },
4329 0 },
4330 { "DTPOFF", { BFD_RELOC_386_TLS_LDO_32,
4331 BFD_RELOC_X86_64_DTPOFF32 },
4332 Imm32 | Imm32S | Imm64 | Disp32 | Disp64 },
4333 { "GOTNTPOFF",{ BFD_RELOC_386_TLS_GOTIE,
4334 0 },
4335 0 },
4336 { "INDNTPOFF",{ BFD_RELOC_386_TLS_IE,
4337 0 },
4338 0 },
4339 { "GOT", { BFD_RELOC_386_GOT32,
4340 BFD_RELOC_X86_64_GOT32 },
4341 Imm32 | Imm32S | Disp32 | Imm64 },
4342 { "TLSDESC", { BFD_RELOC_386_TLS_GOTDESC,
4343 BFD_RELOC_X86_64_GOTPC32_TLSDESC },
4344 Imm32 | Imm32S | Disp32 },
4345 { "TLSCALL", { BFD_RELOC_386_TLS_DESC_CALL,
4346 BFD_RELOC_X86_64_TLSDESC_CALL },
4347 Imm32 | Imm32S | Disp32 }
4348 };
4349 char *cp;
4350 unsigned int j;
4351
4352 if (!IS_ELF)
4353 return NULL;
4354
4355 for (cp = input_line_pointer; *cp != '@'; cp++)
4356 if (is_end_of_line[(unsigned char) *cp])
4357 return NULL;
4358
4359 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
4360 {
4361 int len;
4362
4363 len = strlen (gotrel[j].str);
4364 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
4365 {
4366 if (gotrel[j].rel[object_64bit] != 0)
4367 {
4368 int first, second;
4369 char *tmpbuf, *past_reloc;
4370
4371 *reloc = gotrel[j].rel[object_64bit];
4372 if (adjust)
4373 *adjust = len;
4374
4375 if (types)
4376 {
4377 if (flag_code != CODE_64BIT)
4378 *types = Imm32 | Disp32;
4379 else
4380 *types = gotrel[j].types64;
4381 }
4382
4383 if (GOT_symbol == NULL)
4384 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
4385
4386 /* Replace the relocation token with ' ', so that
4387 errors like foo@GOTOFF1 will be detected. */
4388
4389 /* The length of the first part of our input line. */
4390 first = cp - input_line_pointer;
4391
4392 /* The second part goes from after the reloc token until
4393 (and including) an end_of_line char. Don't use strlen
4394 here as the end_of_line char may not be a NUL. */
4395 past_reloc = cp + 1 + len;
4396 for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
4397 ;
4398 second = cp - past_reloc;
4399
4400 /* Allocate and copy string. The trailing NUL shouldn't
4401 be necessary, but be safe. */
4402 tmpbuf = xmalloc (first + second + 2);
4403 memcpy (tmpbuf, input_line_pointer, first);
4404 tmpbuf[first] = ' ';
4405 memcpy (tmpbuf + first + 1, past_reloc, second);
4406 tmpbuf[first + second + 1] = '\0';
4407 return tmpbuf;
4408 }
4409
4410 as_bad (_("@%s reloc is not supported with %d-bit output format"),
4411 gotrel[j].str, 1 << (5 + object_64bit));
4412 return NULL;
4413 }
4414 }
4415
4416 /* Might be a symbol version string. Don't as_bad here. */
4417 return NULL;
4418 }
4419
4420 void
4421 x86_cons (expressionS *exp, int size)
4422 {
4423 if (size == 4 || (object_64bit && size == 8))
4424 {
4425 /* Handle @GOTOFF and the like in an expression. */
4426 char *save;
4427 char *gotfree_input_line;
4428 int adjust;
4429
4430 save = input_line_pointer;
4431 gotfree_input_line = lex_got (&got_reloc, &adjust, NULL);
4432 if (gotfree_input_line)
4433 input_line_pointer = gotfree_input_line;
4434
4435 expression (exp);
4436
4437 if (gotfree_input_line)
4438 {
4439 /* expression () has merrily parsed up to the end of line,
4440 or a comma - in the wrong buffer. Transfer how far
4441 input_line_pointer has moved to the right buffer. */
4442 input_line_pointer = (save
4443 + (input_line_pointer - gotfree_input_line)
4444 + adjust);
4445 free (gotfree_input_line);
4446 }
4447 }
4448 else
4449 expression (exp);
4450 }
4451 #endif
4452
4453 static void signed_cons (int size)
4454 {
4455 if (flag_code == CODE_64BIT)
4456 cons_sign = 1;
4457 cons (size);
4458 cons_sign = -1;
4459 }
4460
4461 #ifdef TE_PE
4462 static void
4463 pe_directive_secrel (dummy)
4464 int dummy ATTRIBUTE_UNUSED;
4465 {
4466 expressionS exp;
4467
4468 do
4469 {
4470 expression (&exp);
4471 if (exp.X_op == O_symbol)
4472 exp.X_op = O_secrel;
4473
4474 emit_expr (&exp, 4);
4475 }
4476 while (*input_line_pointer++ == ',');
4477
4478 input_line_pointer--;
4479 demand_empty_rest_of_line ();
4480 }
4481 #endif
4482
4483 static int
4484 i386_immediate (char *imm_start)
4485 {
4486 char *save_input_line_pointer;
4487 char *gotfree_input_line;
4488 segT exp_seg = 0;
4489 expressionS *exp;
4490 unsigned int types = ~0U;
4491
4492 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
4493 {
4494 as_bad (_("at most %d immediate operands are allowed"),
4495 MAX_IMMEDIATE_OPERANDS);
4496 return 0;
4497 }
4498
4499 exp = &im_expressions[i.imm_operands++];
4500 i.op[this_operand].imms = exp;
4501
4502 if (is_space_char (*imm_start))
4503 ++imm_start;
4504
4505 save_input_line_pointer = input_line_pointer;
4506 input_line_pointer = imm_start;
4507
4508 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
4509 if (gotfree_input_line)
4510 input_line_pointer = gotfree_input_line;
4511
4512 exp_seg = expression (exp);
4513
4514 SKIP_WHITESPACE ();
4515 if (*input_line_pointer)
4516 as_bad (_("junk `%s' after expression"), input_line_pointer);
4517
4518 input_line_pointer = save_input_line_pointer;
4519 if (gotfree_input_line)
4520 free (gotfree_input_line);
4521
4522 if (exp->X_op == O_absent || exp->X_op == O_big)
4523 {
4524 /* Missing or bad expr becomes absolute 0. */
4525 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
4526 imm_start);
4527 exp->X_op = O_constant;
4528 exp->X_add_number = 0;
4529 exp->X_add_symbol = (symbolS *) 0;
4530 exp->X_op_symbol = (symbolS *) 0;
4531 }
4532 else if (exp->X_op == O_constant)
4533 {
4534 /* Size it properly later. */
4535 i.types[this_operand] |= Imm64;
4536 /* If BFD64, sign extend val. */
4537 if (!use_rela_relocations
4538 && (exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
4539 exp->X_add_number
4540 = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
4541 }
4542 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4543 else if (OUTPUT_FLAVOR == bfd_target_aout_flavour
4544 && exp_seg != absolute_section
4545 && exp_seg != text_section
4546 && exp_seg != data_section
4547 && exp_seg != bss_section
4548 && exp_seg != undefined_section
4549 && !bfd_is_com_section (exp_seg))
4550 {
4551 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
4552 return 0;
4553 }
4554 #endif
4555 else if (!intel_syntax && exp->X_op == O_register)
4556 {
4557 as_bad (_("illegal immediate register operand %s"), imm_start);
4558 return 0;
4559 }
4560 else
4561 {
4562 /* This is an address. The size of the address will be
4563 determined later, depending on destination register,
4564 suffix, or the default for the section. */
4565 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
4566 i.types[this_operand] &= types;
4567 }
4568
4569 return 1;
4570 }
4571
4572 static char *
4573 i386_scale (char *scale)
4574 {
4575 offsetT val;
4576 char *save = input_line_pointer;
4577
4578 input_line_pointer = scale;
4579 val = get_absolute_expression ();
4580
4581 switch (val)
4582 {
4583 case 1:
4584 i.log2_scale_factor = 0;
4585 break;
4586 case 2:
4587 i.log2_scale_factor = 1;
4588 break;
4589 case 4:
4590 i.log2_scale_factor = 2;
4591 break;
4592 case 8:
4593 i.log2_scale_factor = 3;
4594 break;
4595 default:
4596 {
4597 char sep = *input_line_pointer;
4598
4599 *input_line_pointer = '\0';
4600 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
4601 scale);
4602 *input_line_pointer = sep;
4603 input_line_pointer = save;
4604 return NULL;
4605 }
4606 }
4607 if (i.log2_scale_factor != 0 && i.index_reg == 0)
4608 {
4609 as_warn (_("scale factor of %d without an index register"),
4610 1 << i.log2_scale_factor);
4611 #if SCALE1_WHEN_NO_INDEX
4612 i.log2_scale_factor = 0;
4613 #endif
4614 }
4615 scale = input_line_pointer;
4616 input_line_pointer = save;
4617 return scale;
4618 }
4619
4620 static int
4621 i386_displacement (char *disp_start, char *disp_end)
4622 {
4623 expressionS *exp;
4624 segT exp_seg = 0;
4625 char *save_input_line_pointer;
4626 char *gotfree_input_line;
4627 int bigdisp, override;
4628 unsigned int types = Disp;
4629
4630 if (i.disp_operands == MAX_MEMORY_OPERANDS)
4631 {
4632 as_bad (_("at most %d displacement operands are allowed"),
4633 MAX_MEMORY_OPERANDS);
4634 return 0;
4635 }
4636
4637 if ((i.types[this_operand] & JumpAbsolute)
4638 || !(current_templates->start->opcode_modifier & (Jump | JumpDword)))
4639 {
4640 bigdisp = Disp32;
4641 override = (i.prefix[ADDR_PREFIX] != 0);
4642 }
4643 else
4644 {
4645 /* For PC-relative branches, the width of the displacement
4646 is dependent upon data size, not address size. */
4647 bigdisp = 0;
4648 override = (i.prefix[DATA_PREFIX] != 0);
4649 }
4650 if (flag_code == CODE_64BIT)
4651 {
4652 if (!bigdisp)
4653 bigdisp = ((override || i.suffix == WORD_MNEM_SUFFIX)
4654 ? Disp16
4655 : Disp32S | Disp32);
4656 else if (!override)
4657 bigdisp = Disp64 | Disp32S | Disp32;
4658 }
4659 else
4660 {
4661 if (!bigdisp)
4662 {
4663 if (!override)
4664 override = (i.suffix == (flag_code != CODE_16BIT
4665 ? WORD_MNEM_SUFFIX
4666 : LONG_MNEM_SUFFIX));
4667 bigdisp = Disp32;
4668 }
4669 if ((flag_code == CODE_16BIT) ^ override)
4670 bigdisp = Disp16;
4671 }
4672 i.types[this_operand] |= bigdisp;
4673
4674 exp = &disp_expressions[i.disp_operands];
4675 i.op[this_operand].disps = exp;
4676 i.disp_operands++;
4677 save_input_line_pointer = input_line_pointer;
4678 input_line_pointer = disp_start;
4679 END_STRING_AND_SAVE (disp_end);
4680
4681 #ifndef GCC_ASM_O_HACK
4682 #define GCC_ASM_O_HACK 0
4683 #endif
4684 #if GCC_ASM_O_HACK
4685 END_STRING_AND_SAVE (disp_end + 1);
4686 if ((i.types[this_operand] & BaseIndex) != 0
4687 && displacement_string_end[-1] == '+')
4688 {
4689 /* This hack is to avoid a warning when using the "o"
4690 constraint within gcc asm statements.
4691 For instance:
4692
4693 #define _set_tssldt_desc(n,addr,limit,type) \
4694 __asm__ __volatile__ ( \
4695 "movw %w2,%0\n\t" \
4696 "movw %w1,2+%0\n\t" \
4697 "rorl $16,%1\n\t" \
4698 "movb %b1,4+%0\n\t" \
4699 "movb %4,5+%0\n\t" \
4700 "movb $0,6+%0\n\t" \
4701 "movb %h1,7+%0\n\t" \
4702 "rorl $16,%1" \
4703 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
4704
4705 This works great except that the output assembler ends
4706 up looking a bit weird if it turns out that there is
4707 no offset. You end up producing code that looks like:
4708
4709 #APP
4710 movw $235,(%eax)
4711 movw %dx,2+(%eax)
4712 rorl $16,%edx
4713 movb %dl,4+(%eax)
4714 movb $137,5+(%eax)
4715 movb $0,6+(%eax)
4716 movb %dh,7+(%eax)
4717 rorl $16,%edx
4718 #NO_APP
4719
4720 So here we provide the missing zero. */
4721
4722 *displacement_string_end = '0';
4723 }
4724 #endif
4725 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
4726 if (gotfree_input_line)
4727 input_line_pointer = gotfree_input_line;
4728
4729 exp_seg = expression (exp);
4730
4731 SKIP_WHITESPACE ();
4732 if (*input_line_pointer)
4733 as_bad (_("junk `%s' after expression"), input_line_pointer);
4734 #if GCC_ASM_O_HACK
4735 RESTORE_END_STRING (disp_end + 1);
4736 #endif
4737 RESTORE_END_STRING (disp_end);
4738 input_line_pointer = save_input_line_pointer;
4739 if (gotfree_input_line)
4740 free (gotfree_input_line);
4741
4742 /* We do this to make sure that the section symbol is in
4743 the symbol table. We will ultimately change the relocation
4744 to be relative to the beginning of the section. */
4745 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
4746 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
4747 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTOFF64)
4748 {
4749 if (exp->X_op != O_symbol)
4750 {
4751 as_bad (_("bad expression used with @%s"),
4752 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
4753 ? "GOTPCREL"
4754 : "GOTOFF"));
4755 return 0;
4756 }
4757
4758 if (S_IS_LOCAL (exp->X_add_symbol)
4759 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
4760 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
4761 exp->X_op = O_subtract;
4762 exp->X_op_symbol = GOT_symbol;
4763 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
4764 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
4765 else if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTOFF64)
4766 i.reloc[this_operand] = BFD_RELOC_64;
4767 else
4768 i.reloc[this_operand] = BFD_RELOC_32;
4769 }
4770
4771 if (exp->X_op == O_absent || exp->X_op == O_big)
4772 {
4773 /* Missing or bad expr becomes absolute 0. */
4774 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
4775 disp_start);
4776 exp->X_op = O_constant;
4777 exp->X_add_number = 0;
4778 exp->X_add_symbol = (symbolS *) 0;
4779 exp->X_op_symbol = (symbolS *) 0;
4780 }
4781
4782 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4783 if (exp->X_op != O_constant
4784 && OUTPUT_FLAVOR == bfd_target_aout_flavour
4785 && exp_seg != absolute_section
4786 && exp_seg != text_section
4787 && exp_seg != data_section
4788 && exp_seg != bss_section
4789 && exp_seg != undefined_section
4790 && !bfd_is_com_section (exp_seg))
4791 {
4792 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
4793 return 0;
4794 }
4795 #endif
4796
4797 if (!(i.types[this_operand] & ~Disp))
4798 i.types[this_operand] &= types;
4799
4800 return 1;
4801 }
4802
4803 /* Make sure the memory operand we've been dealt is valid.
4804 Return 1 on success, 0 on a failure. */
4805
4806 static int
4807 i386_index_check (const char *operand_string)
4808 {
4809 int ok;
4810 #if INFER_ADDR_PREFIX
4811 int fudged = 0;
4812
4813 tryprefix:
4814 #endif
4815 ok = 1;
4816 if ((current_templates->start->cpu_flags & CpuSVME)
4817 && current_templates->end[-1].operand_types[0] == AnyMem)
4818 {
4819 /* Memory operands of SVME insns are special in that they only allow
4820 rAX as their memory address and ignore any segment override. */
4821 unsigned RegXX;
4822
4823 /* SKINIT is even more restrictive: it always requires EAX. */
4824 if (strcmp (current_templates->start->name, "skinit") == 0)
4825 RegXX = Reg32;
4826 else if (flag_code == CODE_64BIT)
4827 RegXX = i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32;
4828 else
4829 RegXX = ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)
4830 ? Reg16
4831 : Reg32);
4832 if (!i.base_reg
4833 || !(i.base_reg->reg_type & Acc)
4834 || !(i.base_reg->reg_type & RegXX)
4835 || i.index_reg
4836 || (i.types[0] & Disp))
4837 ok = 0;
4838 }
4839 else if (flag_code == CODE_64BIT)
4840 {
4841 unsigned RegXX = (i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32);
4842
4843 if ((i.base_reg
4844 && ((i.base_reg->reg_type & RegXX) == 0)
4845 && (i.base_reg->reg_type != BaseIndex
4846 || i.index_reg))
4847 || (i.index_reg
4848 && ((i.index_reg->reg_type & (RegXX | BaseIndex))
4849 != (RegXX | BaseIndex))))
4850 ok = 0;
4851 }
4852 else
4853 {
4854 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
4855 {
4856 /* 16bit checks. */
4857 if ((i.base_reg
4858 && ((i.base_reg->reg_type & (Reg16 | BaseIndex | RegRex))
4859 != (Reg16 | BaseIndex)))
4860 || (i.index_reg
4861 && (((i.index_reg->reg_type & (Reg16 | BaseIndex))
4862 != (Reg16 | BaseIndex))
4863 || !(i.base_reg
4864 && i.base_reg->reg_num < 6
4865 && i.index_reg->reg_num >= 6
4866 && i.log2_scale_factor == 0))))
4867 ok = 0;
4868 }
4869 else
4870 {
4871 /* 32bit checks. */
4872 if ((i.base_reg
4873 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
4874 || (i.index_reg
4875 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
4876 != (Reg32 | BaseIndex))))
4877 ok = 0;
4878 }
4879 }
4880 if (!ok)
4881 {
4882 #if INFER_ADDR_PREFIX
4883 if (i.prefix[ADDR_PREFIX] == 0)
4884 {
4885 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
4886 i.prefixes += 1;
4887 /* Change the size of any displacement too. At most one of
4888 Disp16 or Disp32 is set.
4889 FIXME. There doesn't seem to be any real need for separate
4890 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
4891 Removing them would probably clean up the code quite a lot. */
4892 if (flag_code != CODE_64BIT
4893 && (i.types[this_operand] & (Disp16 | Disp32)))
4894 i.types[this_operand] ^= (Disp16 | Disp32);
4895 fudged = 1;
4896 goto tryprefix;
4897 }
4898 if (fudged)
4899 as_bad (_("`%s' is not a valid base/index expression"),
4900 operand_string);
4901 else
4902 #endif
4903 as_bad (_("`%s' is not a valid %s bit base/index expression"),
4904 operand_string,
4905 flag_code_names[flag_code]);
4906 }
4907 return ok;
4908 }
4909
4910 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
4911 on error. */
4912
4913 static int
4914 i386_operand (char *operand_string)
4915 {
4916 const reg_entry *r;
4917 char *end_op;
4918 char *op_string = operand_string;
4919
4920 if (is_space_char (*op_string))
4921 ++op_string;
4922
4923 /* We check for an absolute prefix (differentiating,
4924 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
4925 if (*op_string == ABSOLUTE_PREFIX)
4926 {
4927 ++op_string;
4928 if (is_space_char (*op_string))
4929 ++op_string;
4930 i.types[this_operand] |= JumpAbsolute;
4931 }
4932
4933 /* Check if operand is a register. */
4934 if ((r = parse_register (op_string, &end_op)) != NULL)
4935 {
4936 /* Check for a segment override by searching for ':' after a
4937 segment register. */
4938 op_string = end_op;
4939 if (is_space_char (*op_string))
4940 ++op_string;
4941 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
4942 {
4943 switch (r->reg_num)
4944 {
4945 case 0:
4946 i.seg[i.mem_operands] = &es;
4947 break;
4948 case 1:
4949 i.seg[i.mem_operands] = &cs;
4950 break;
4951 case 2:
4952 i.seg[i.mem_operands] = &ss;
4953 break;
4954 case 3:
4955 i.seg[i.mem_operands] = &ds;
4956 break;
4957 case 4:
4958 i.seg[i.mem_operands] = &fs;
4959 break;
4960 case 5:
4961 i.seg[i.mem_operands] = &gs;
4962 break;
4963 }
4964
4965 /* Skip the ':' and whitespace. */
4966 ++op_string;
4967 if (is_space_char (*op_string))
4968 ++op_string;
4969
4970 if (!is_digit_char (*op_string)
4971 && !is_identifier_char (*op_string)
4972 && *op_string != '('
4973 && *op_string != ABSOLUTE_PREFIX)
4974 {
4975 as_bad (_("bad memory operand `%s'"), op_string);
4976 return 0;
4977 }
4978 /* Handle case of %es:*foo. */
4979 if (*op_string == ABSOLUTE_PREFIX)
4980 {
4981 ++op_string;
4982 if (is_space_char (*op_string))
4983 ++op_string;
4984 i.types[this_operand] |= JumpAbsolute;
4985 }
4986 goto do_memory_reference;
4987 }
4988 if (*op_string)
4989 {
4990 as_bad (_("junk `%s' after register"), op_string);
4991 return 0;
4992 }
4993 i.types[this_operand] |= r->reg_type & ~BaseIndex;
4994 i.op[this_operand].regs = r;
4995 i.reg_operands++;
4996 }
4997 else if (*op_string == REGISTER_PREFIX)
4998 {
4999 as_bad (_("bad register name `%s'"), op_string);
5000 return 0;
5001 }
5002 else if (*op_string == IMMEDIATE_PREFIX)
5003 {
5004 ++op_string;
5005 if (i.types[this_operand] & JumpAbsolute)
5006 {
5007 as_bad (_("immediate operand illegal with absolute jump"));
5008 return 0;
5009 }
5010 if (!i386_immediate (op_string))
5011 return 0;
5012 }
5013 else if (is_digit_char (*op_string)
5014 || is_identifier_char (*op_string)
5015 || *op_string == '(')
5016 {
5017 /* This is a memory reference of some sort. */
5018 char *base_string;
5019
5020 /* Start and end of displacement string expression (if found). */
5021 char *displacement_string_start;
5022 char *displacement_string_end;
5023
5024 do_memory_reference:
5025 if ((i.mem_operands == 1
5026 && (current_templates->start->opcode_modifier & IsString) == 0)
5027 || i.mem_operands == 2)
5028 {
5029 as_bad (_("too many memory references for `%s'"),
5030 current_templates->start->name);
5031 return 0;
5032 }
5033
5034 /* Check for base index form. We detect the base index form by
5035 looking for an ')' at the end of the operand, searching
5036 for the '(' matching it, and finding a REGISTER_PREFIX or ','
5037 after the '('. */
5038 base_string = op_string + strlen (op_string);
5039
5040 --base_string;
5041 if (is_space_char (*base_string))
5042 --base_string;
5043
5044 /* If we only have a displacement, set-up for it to be parsed later. */
5045 displacement_string_start = op_string;
5046 displacement_string_end = base_string + 1;
5047
5048 if (*base_string == ')')
5049 {
5050 char *temp_string;
5051 unsigned int parens_balanced = 1;
5052 /* We've already checked that the number of left & right ()'s are
5053 equal, so this loop will not be infinite. */
5054 do
5055 {
5056 base_string--;
5057 if (*base_string == ')')
5058 parens_balanced++;
5059 if (*base_string == '(')
5060 parens_balanced--;
5061 }
5062 while (parens_balanced);
5063
5064 temp_string = base_string;
5065
5066 /* Skip past '(' and whitespace. */
5067 ++base_string;
5068 if (is_space_char (*base_string))
5069 ++base_string;
5070
5071 if (*base_string == ','
5072 || ((i.base_reg = parse_register (base_string, &end_op))
5073 != NULL))
5074 {
5075 displacement_string_end = temp_string;
5076
5077 i.types[this_operand] |= BaseIndex;
5078
5079 if (i.base_reg)
5080 {
5081 base_string = end_op;
5082 if (is_space_char (*base_string))
5083 ++base_string;
5084 }
5085
5086 /* There may be an index reg or scale factor here. */
5087 if (*base_string == ',')
5088 {
5089 ++base_string;
5090 if (is_space_char (*base_string))
5091 ++base_string;
5092
5093 if ((i.index_reg = parse_register (base_string, &end_op))
5094 != NULL)
5095 {
5096 base_string = end_op;
5097 if (is_space_char (*base_string))
5098 ++base_string;
5099 if (*base_string == ',')
5100 {
5101 ++base_string;
5102 if (is_space_char (*base_string))
5103 ++base_string;
5104 }
5105 else if (*base_string != ')')
5106 {
5107 as_bad (_("expecting `,' or `)' "
5108 "after index register in `%s'"),
5109 operand_string);
5110 return 0;
5111 }
5112 }
5113 else if (*base_string == REGISTER_PREFIX)
5114 {
5115 as_bad (_("bad register name `%s'"), base_string);
5116 return 0;
5117 }
5118
5119 /* Check for scale factor. */
5120 if (*base_string != ')')
5121 {
5122 char *end_scale = i386_scale (base_string);
5123
5124 if (!end_scale)
5125 return 0;
5126
5127 base_string = end_scale;
5128 if (is_space_char (*base_string))
5129 ++base_string;
5130 if (*base_string != ')')
5131 {
5132 as_bad (_("expecting `)' "
5133 "after scale factor in `%s'"),
5134 operand_string);
5135 return 0;
5136 }
5137 }
5138 else if (!i.index_reg)
5139 {
5140 as_bad (_("expecting index register or scale factor "
5141 "after `,'; got '%c'"),
5142 *base_string);
5143 return 0;
5144 }
5145 }
5146 else if (*base_string != ')')
5147 {
5148 as_bad (_("expecting `,' or `)' "
5149 "after base register in `%s'"),
5150 operand_string);
5151 return 0;
5152 }
5153 }
5154 else if (*base_string == REGISTER_PREFIX)
5155 {
5156 as_bad (_("bad register name `%s'"), base_string);
5157 return 0;
5158 }
5159 }
5160
5161 /* If there's an expression beginning the operand, parse it,
5162 assuming displacement_string_start and
5163 displacement_string_end are meaningful. */
5164 if (displacement_string_start != displacement_string_end)
5165 {
5166 if (!i386_displacement (displacement_string_start,
5167 displacement_string_end))
5168 return 0;
5169 }
5170
5171 /* Special case for (%dx) while doing input/output op. */
5172 if (i.base_reg
5173 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
5174 && i.index_reg == 0
5175 && i.log2_scale_factor == 0
5176 && i.seg[i.mem_operands] == 0
5177 && (i.types[this_operand] & Disp) == 0)
5178 {
5179 i.types[this_operand] = InOutPortReg;
5180 return 1;
5181 }
5182
5183 if (i386_index_check (operand_string) == 0)
5184 return 0;
5185 i.mem_operands++;
5186 }
5187 else
5188 {
5189 /* It's not a memory operand; argh! */
5190 as_bad (_("invalid char %s beginning operand %d `%s'"),
5191 output_invalid (*op_string),
5192 this_operand + 1,
5193 op_string);
5194 return 0;
5195 }
5196 return 1; /* Normal return. */
5197 }
5198 \f
5199 /* md_estimate_size_before_relax()
5200
5201 Called just before relax() for rs_machine_dependent frags. The x86
5202 assembler uses these frags to handle variable size jump
5203 instructions.
5204
5205 Any symbol that is now undefined will not become defined.
5206 Return the correct fr_subtype in the frag.
5207 Return the initial "guess for variable size of frag" to caller.
5208 The guess is actually the growth beyond the fixed part. Whatever
5209 we do to grow the fixed or variable part contributes to our
5210 returned value. */
5211
5212 int
5213 md_estimate_size_before_relax (fragP, segment)
5214 fragS *fragP;
5215 segT segment;
5216 {
5217 /* We've already got fragP->fr_subtype right; all we have to do is
5218 check for un-relaxable symbols. On an ELF system, we can't relax
5219 an externally visible symbol, because it may be overridden by a
5220 shared library. */
5221 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
5222 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5223 || (IS_ELF
5224 && (S_IS_EXTERNAL (fragP->fr_symbol)
5225 || S_IS_WEAK (fragP->fr_symbol)))
5226 #endif
5227 )
5228 {
5229 /* Symbol is undefined in this segment, or we need to keep a
5230 reloc so that weak symbols can be overridden. */
5231 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
5232 enum bfd_reloc_code_real reloc_type;
5233 unsigned char *opcode;
5234 int old_fr_fix;
5235
5236 if (fragP->fr_var != NO_RELOC)
5237 reloc_type = fragP->fr_var;
5238 else if (size == 2)
5239 reloc_type = BFD_RELOC_16_PCREL;
5240 else
5241 reloc_type = BFD_RELOC_32_PCREL;
5242
5243 old_fr_fix = fragP->fr_fix;
5244 opcode = (unsigned char *) fragP->fr_opcode;
5245
5246 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
5247 {
5248 case UNCOND_JUMP:
5249 /* Make jmp (0xeb) a (d)word displacement jump. */
5250 opcode[0] = 0xe9;
5251 fragP->fr_fix += size;
5252 fix_new (fragP, old_fr_fix, size,
5253 fragP->fr_symbol,
5254 fragP->fr_offset, 1,
5255 reloc_type);
5256 break;
5257
5258 case COND_JUMP86:
5259 if (size == 2
5260 && (!no_cond_jump_promotion || fragP->fr_var != NO_RELOC))
5261 {
5262 /* Negate the condition, and branch past an
5263 unconditional jump. */
5264 opcode[0] ^= 1;
5265 opcode[1] = 3;
5266 /* Insert an unconditional jump. */
5267 opcode[2] = 0xe9;
5268 /* We added two extra opcode bytes, and have a two byte
5269 offset. */
5270 fragP->fr_fix += 2 + 2;
5271 fix_new (fragP, old_fr_fix + 2, 2,
5272 fragP->fr_symbol,
5273 fragP->fr_offset, 1,
5274 reloc_type);
5275 break;
5276 }
5277 /* Fall through. */
5278
5279 case COND_JUMP:
5280 if (no_cond_jump_promotion && fragP->fr_var == NO_RELOC)
5281 {
5282 fixS *fixP;
5283
5284 fragP->fr_fix += 1;
5285 fixP = fix_new (fragP, old_fr_fix, 1,
5286 fragP->fr_symbol,
5287 fragP->fr_offset, 1,
5288 BFD_RELOC_8_PCREL);
5289 fixP->fx_signed = 1;
5290 break;
5291 }
5292
5293 /* This changes the byte-displacement jump 0x7N
5294 to the (d)word-displacement jump 0x0f,0x8N. */
5295 opcode[1] = opcode[0] + 0x10;
5296 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5297 /* We've added an opcode byte. */
5298 fragP->fr_fix += 1 + size;
5299 fix_new (fragP, old_fr_fix + 1, size,
5300 fragP->fr_symbol,
5301 fragP->fr_offset, 1,
5302 reloc_type);
5303 break;
5304
5305 default:
5306 BAD_CASE (fragP->fr_subtype);
5307 break;
5308 }
5309 frag_wane (fragP);
5310 return fragP->fr_fix - old_fr_fix;
5311 }
5312
5313 /* Guess size depending on current relax state. Initially the relax
5314 state will correspond to a short jump and we return 1, because
5315 the variable part of the frag (the branch offset) is one byte
5316 long. However, we can relax a section more than once and in that
5317 case we must either set fr_subtype back to the unrelaxed state,
5318 or return the value for the appropriate branch. */
5319 return md_relax_table[fragP->fr_subtype].rlx_length;
5320 }
5321
5322 /* Called after relax() is finished.
5323
5324 In: Address of frag.
5325 fr_type == rs_machine_dependent.
5326 fr_subtype is what the address relaxed to.
5327
5328 Out: Any fixSs and constants are set up.
5329 Caller will turn frag into a ".space 0". */
5330
5331 void
5332 md_convert_frag (abfd, sec, fragP)
5333 bfd *abfd ATTRIBUTE_UNUSED;
5334 segT sec ATTRIBUTE_UNUSED;
5335 fragS *fragP;
5336 {
5337 unsigned char *opcode;
5338 unsigned char *where_to_put_displacement = NULL;
5339 offsetT target_address;
5340 offsetT opcode_address;
5341 unsigned int extension = 0;
5342 offsetT displacement_from_opcode_start;
5343
5344 opcode = (unsigned char *) fragP->fr_opcode;
5345
5346 /* Address we want to reach in file space. */
5347 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
5348
5349 /* Address opcode resides at in file space. */
5350 opcode_address = fragP->fr_address + fragP->fr_fix;
5351
5352 /* Displacement from opcode start to fill into instruction. */
5353 displacement_from_opcode_start = target_address - opcode_address;
5354
5355 if ((fragP->fr_subtype & BIG) == 0)
5356 {
5357 /* Don't have to change opcode. */
5358 extension = 1; /* 1 opcode + 1 displacement */
5359 where_to_put_displacement = &opcode[1];
5360 }
5361 else
5362 {
5363 if (no_cond_jump_promotion
5364 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
5365 as_warn_where (fragP->fr_file, fragP->fr_line,
5366 _("long jump required"));
5367
5368 switch (fragP->fr_subtype)
5369 {
5370 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
5371 extension = 4; /* 1 opcode + 4 displacement */
5372 opcode[0] = 0xe9;
5373 where_to_put_displacement = &opcode[1];
5374 break;
5375
5376 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
5377 extension = 2; /* 1 opcode + 2 displacement */
5378 opcode[0] = 0xe9;
5379 where_to_put_displacement = &opcode[1];
5380 break;
5381
5382 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
5383 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
5384 extension = 5; /* 2 opcode + 4 displacement */
5385 opcode[1] = opcode[0] + 0x10;
5386 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5387 where_to_put_displacement = &opcode[2];
5388 break;
5389
5390 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
5391 extension = 3; /* 2 opcode + 2 displacement */
5392 opcode[1] = opcode[0] + 0x10;
5393 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5394 where_to_put_displacement = &opcode[2];
5395 break;
5396
5397 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
5398 extension = 4;
5399 opcode[0] ^= 1;
5400 opcode[1] = 3;
5401 opcode[2] = 0xe9;
5402 where_to_put_displacement = &opcode[3];
5403 break;
5404
5405 default:
5406 BAD_CASE (fragP->fr_subtype);
5407 break;
5408 }
5409 }
5410
5411 /* If size if less then four we are sure that the operand fits,
5412 but if it's 4, then it could be that the displacement is larger
5413 then -/+ 2GB. */
5414 if (DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype) == 4
5415 && object_64bit
5416 && ((addressT) (displacement_from_opcode_start - extension
5417 + ((addressT) 1 << 31))
5418 > (((addressT) 2 << 31) - 1)))
5419 {
5420 as_bad_where (fragP->fr_file, fragP->fr_line,
5421 _("jump target out of range"));
5422 /* Make us emit 0. */
5423 displacement_from_opcode_start = extension;
5424 }
5425 /* Now put displacement after opcode. */
5426 md_number_to_chars ((char *) where_to_put_displacement,
5427 (valueT) (displacement_from_opcode_start - extension),
5428 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
5429 fragP->fr_fix += extension;
5430 }
5431 \f
5432 /* Size of byte displacement jmp. */
5433 int md_short_jump_size = 2;
5434
5435 /* Size of dword displacement jmp. */
5436 int md_long_jump_size = 5;
5437
5438 void
5439 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
5440 char *ptr;
5441 addressT from_addr, to_addr;
5442 fragS *frag ATTRIBUTE_UNUSED;
5443 symbolS *to_symbol ATTRIBUTE_UNUSED;
5444 {
5445 offsetT offset;
5446
5447 offset = to_addr - (from_addr + 2);
5448 /* Opcode for byte-disp jump. */
5449 md_number_to_chars (ptr, (valueT) 0xeb, 1);
5450 md_number_to_chars (ptr + 1, (valueT) offset, 1);
5451 }
5452
5453 void
5454 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
5455 char *ptr;
5456 addressT from_addr, to_addr;
5457 fragS *frag ATTRIBUTE_UNUSED;
5458 symbolS *to_symbol ATTRIBUTE_UNUSED;
5459 {
5460 offsetT offset;
5461
5462 offset = to_addr - (from_addr + 5);
5463 md_number_to_chars (ptr, (valueT) 0xe9, 1);
5464 md_number_to_chars (ptr + 1, (valueT) offset, 4);
5465 }
5466 \f
5467 /* Apply a fixup (fixS) to segment data, once it has been determined
5468 by our caller that we have all the info we need to fix it up.
5469
5470 On the 386, immediates, displacements, and data pointers are all in
5471 the same (little-endian) format, so we don't need to care about which
5472 we are handling. */
5473
5474 void
5475 md_apply_fix (fixP, valP, seg)
5476 /* The fix we're to put in. */
5477 fixS *fixP;
5478 /* Pointer to the value of the bits. */
5479 valueT *valP;
5480 /* Segment fix is from. */
5481 segT seg ATTRIBUTE_UNUSED;
5482 {
5483 char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
5484 valueT value = *valP;
5485
5486 #if !defined (TE_Mach)
5487 if (fixP->fx_pcrel)
5488 {
5489 switch (fixP->fx_r_type)
5490 {
5491 default:
5492 break;
5493
5494 case BFD_RELOC_64:
5495 fixP->fx_r_type = BFD_RELOC_64_PCREL;
5496 break;
5497 case BFD_RELOC_32:
5498 case BFD_RELOC_X86_64_32S:
5499 fixP->fx_r_type = BFD_RELOC_32_PCREL;
5500 break;
5501 case BFD_RELOC_16:
5502 fixP->fx_r_type = BFD_RELOC_16_PCREL;
5503 break;
5504 case BFD_RELOC_8:
5505 fixP->fx_r_type = BFD_RELOC_8_PCREL;
5506 break;
5507 }
5508 }
5509
5510 if (fixP->fx_addsy != NULL
5511 && (fixP->fx_r_type == BFD_RELOC_32_PCREL
5512 || fixP->fx_r_type == BFD_RELOC_64_PCREL
5513 || fixP->fx_r_type == BFD_RELOC_16_PCREL
5514 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
5515 && !use_rela_relocations)
5516 {
5517 /* This is a hack. There should be a better way to handle this.
5518 This covers for the fact that bfd_install_relocation will
5519 subtract the current location (for partial_inplace, PC relative
5520 relocations); see more below. */
5521 #ifndef OBJ_AOUT
5522 if (IS_ELF
5523 #ifdef TE_PE
5524 || OUTPUT_FLAVOR == bfd_target_coff_flavour
5525 #endif
5526 )
5527 value += fixP->fx_where + fixP->fx_frag->fr_address;
5528 #endif
5529 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5530 if (IS_ELF)
5531 {
5532 segT sym_seg = S_GET_SEGMENT (fixP->fx_addsy);
5533
5534 if ((sym_seg == seg
5535 || (symbol_section_p (fixP->fx_addsy)
5536 && sym_seg != absolute_section))
5537 && !generic_force_reloc (fixP))
5538 {
5539 /* Yes, we add the values in twice. This is because
5540 bfd_install_relocation subtracts them out again. I think
5541 bfd_install_relocation is broken, but I don't dare change
5542 it. FIXME. */
5543 value += fixP->fx_where + fixP->fx_frag->fr_address;
5544 }
5545 }
5546 #endif
5547 #if defined (OBJ_COFF) && defined (TE_PE)
5548 /* For some reason, the PE format does not store a
5549 section address offset for a PC relative symbol. */
5550 if (S_GET_SEGMENT (fixP->fx_addsy) != seg
5551 || S_IS_WEAK (fixP->fx_addsy))
5552 value += md_pcrel_from (fixP);
5553 #endif
5554 }
5555
5556 /* Fix a few things - the dynamic linker expects certain values here,
5557 and we must not disappoint it. */
5558 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5559 if (IS_ELF && fixP->fx_addsy)
5560 switch (fixP->fx_r_type)
5561 {
5562 case BFD_RELOC_386_PLT32:
5563 case BFD_RELOC_X86_64_PLT32:
5564 /* Make the jump instruction point to the address of the operand. At
5565 runtime we merely add the offset to the actual PLT entry. */
5566 value = -4;
5567 break;
5568
5569 case BFD_RELOC_386_TLS_GD:
5570 case BFD_RELOC_386_TLS_LDM:
5571 case BFD_RELOC_386_TLS_IE_32:
5572 case BFD_RELOC_386_TLS_IE:
5573 case BFD_RELOC_386_TLS_GOTIE:
5574 case BFD_RELOC_386_TLS_GOTDESC:
5575 case BFD_RELOC_X86_64_TLSGD:
5576 case BFD_RELOC_X86_64_TLSLD:
5577 case BFD_RELOC_X86_64_GOTTPOFF:
5578 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
5579 value = 0; /* Fully resolved at runtime. No addend. */
5580 /* Fallthrough */
5581 case BFD_RELOC_386_TLS_LE:
5582 case BFD_RELOC_386_TLS_LDO_32:
5583 case BFD_RELOC_386_TLS_LE_32:
5584 case BFD_RELOC_X86_64_DTPOFF32:
5585 case BFD_RELOC_X86_64_DTPOFF64:
5586 case BFD_RELOC_X86_64_TPOFF32:
5587 case BFD_RELOC_X86_64_TPOFF64:
5588 S_SET_THREAD_LOCAL (fixP->fx_addsy);
5589 break;
5590
5591 case BFD_RELOC_386_TLS_DESC_CALL:
5592 case BFD_RELOC_X86_64_TLSDESC_CALL:
5593 value = 0; /* Fully resolved at runtime. No addend. */
5594 S_SET_THREAD_LOCAL (fixP->fx_addsy);
5595 fixP->fx_done = 0;
5596 return;
5597
5598 case BFD_RELOC_386_GOT32:
5599 case BFD_RELOC_X86_64_GOT32:
5600 value = 0; /* Fully resolved at runtime. No addend. */
5601 break;
5602
5603 case BFD_RELOC_VTABLE_INHERIT:
5604 case BFD_RELOC_VTABLE_ENTRY:
5605 fixP->fx_done = 0;
5606 return;
5607
5608 default:
5609 break;
5610 }
5611 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
5612 *valP = value;
5613 #endif /* !defined (TE_Mach) */
5614
5615 /* Are we finished with this relocation now? */
5616 if (fixP->fx_addsy == NULL)
5617 fixP->fx_done = 1;
5618 else if (use_rela_relocations)
5619 {
5620 fixP->fx_no_overflow = 1;
5621 /* Remember value for tc_gen_reloc. */
5622 fixP->fx_addnumber = value;
5623 value = 0;
5624 }
5625
5626 md_number_to_chars (p, value, fixP->fx_size);
5627 }
5628 \f
5629 #define MAX_LITTLENUMS 6
5630
5631 /* Turn the string pointed to by litP into a floating point constant
5632 of type TYPE, and emit the appropriate bytes. The number of
5633 LITTLENUMS emitted is stored in *SIZEP. An error message is
5634 returned, or NULL on OK. */
5635
5636 char *
5637 md_atof (type, litP, sizeP)
5638 int type;
5639 char *litP;
5640 int *sizeP;
5641 {
5642 int prec;
5643 LITTLENUM_TYPE words[MAX_LITTLENUMS];
5644 LITTLENUM_TYPE *wordP;
5645 char *t;
5646
5647 switch (type)
5648 {
5649 case 'f':
5650 case 'F':
5651 prec = 2;
5652 break;
5653
5654 case 'd':
5655 case 'D':
5656 prec = 4;
5657 break;
5658
5659 case 'x':
5660 case 'X':
5661 prec = 5;
5662 break;
5663
5664 default:
5665 *sizeP = 0;
5666 return _("Bad call to md_atof ()");
5667 }
5668 t = atof_ieee (input_line_pointer, type, words);
5669 if (t)
5670 input_line_pointer = t;
5671
5672 *sizeP = prec * sizeof (LITTLENUM_TYPE);
5673 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
5674 the bigendian 386. */
5675 for (wordP = words + prec - 1; prec--;)
5676 {
5677 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
5678 litP += sizeof (LITTLENUM_TYPE);
5679 }
5680 return 0;
5681 }
5682 \f
5683 static char output_invalid_buf[sizeof (unsigned char) * 2 + 6];
5684
5685 static char *
5686 output_invalid (int c)
5687 {
5688 if (ISPRINT (c))
5689 snprintf (output_invalid_buf, sizeof (output_invalid_buf),
5690 "'%c'", c);
5691 else
5692 snprintf (output_invalid_buf, sizeof (output_invalid_buf),
5693 "(0x%x)", (unsigned char) c);
5694 return output_invalid_buf;
5695 }
5696
5697 /* REG_STRING starts *before* REGISTER_PREFIX. */
5698
5699 static const reg_entry *
5700 parse_real_register (char *reg_string, char **end_op)
5701 {
5702 char *s = reg_string;
5703 char *p;
5704 char reg_name_given[MAX_REG_NAME_SIZE + 1];
5705 const reg_entry *r;
5706
5707 /* Skip possible REGISTER_PREFIX and possible whitespace. */
5708 if (*s == REGISTER_PREFIX)
5709 ++s;
5710
5711 if (is_space_char (*s))
5712 ++s;
5713
5714 p = reg_name_given;
5715 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
5716 {
5717 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
5718 return (const reg_entry *) NULL;
5719 s++;
5720 }
5721
5722 /* For naked regs, make sure that we are not dealing with an identifier.
5723 This prevents confusing an identifier like `eax_var' with register
5724 `eax'. */
5725 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
5726 return (const reg_entry *) NULL;
5727
5728 *end_op = s;
5729
5730 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
5731
5732 /* Handle floating point regs, allowing spaces in the (i) part. */
5733 if (r == i386_regtab /* %st is first entry of table */)
5734 {
5735 if (is_space_char (*s))
5736 ++s;
5737 if (*s == '(')
5738 {
5739 ++s;
5740 if (is_space_char (*s))
5741 ++s;
5742 if (*s >= '0' && *s <= '7')
5743 {
5744 r = &i386_float_regtab[*s - '0'];
5745 ++s;
5746 if (is_space_char (*s))
5747 ++s;
5748 if (*s == ')')
5749 {
5750 *end_op = s + 1;
5751 return r;
5752 }
5753 }
5754 /* We have "%st(" then garbage. */
5755 return (const reg_entry *) NULL;
5756 }
5757 }
5758
5759 if (r != NULL
5760 && ((r->reg_flags & (RegRex64 | RegRex)) | (r->reg_type & Reg64)) != 0
5761 && (r->reg_type != Control || !(cpu_arch_flags & CpuSledgehammer))
5762 && flag_code != CODE_64BIT)
5763 return (const reg_entry *) NULL;
5764
5765 return r;
5766 }
5767
5768 /* REG_STRING starts *before* REGISTER_PREFIX. */
5769
5770 static const reg_entry *
5771 parse_register (char *reg_string, char **end_op)
5772 {
5773 const reg_entry *r;
5774
5775 if (*reg_string == REGISTER_PREFIX || allow_naked_reg)
5776 r = parse_real_register (reg_string, end_op);
5777 else
5778 r = NULL;
5779 if (!r)
5780 {
5781 char *save = input_line_pointer;
5782 char c;
5783 symbolS *symbolP;
5784
5785 input_line_pointer = reg_string;
5786 c = get_symbol_end ();
5787 symbolP = symbol_find (reg_string);
5788 if (symbolP && S_GET_SEGMENT (symbolP) == reg_section)
5789 {
5790 const expressionS *e = symbol_get_value_expression (symbolP);
5791
5792 know (e->X_op == O_register);
5793 know (e->X_add_number >= 0
5794 && (valueT) e->X_add_number < i386_regtab_size);
5795 r = i386_regtab + e->X_add_number;
5796 *end_op = input_line_pointer;
5797 }
5798 *input_line_pointer = c;
5799 input_line_pointer = save;
5800 }
5801 return r;
5802 }
5803
5804 int
5805 i386_parse_name (char *name, expressionS *e, char *nextcharP)
5806 {
5807 const reg_entry *r;
5808 char *end = input_line_pointer;
5809
5810 *end = *nextcharP;
5811 r = parse_register (name, &input_line_pointer);
5812 if (r && end <= input_line_pointer)
5813 {
5814 *nextcharP = *input_line_pointer;
5815 *input_line_pointer = 0;
5816 e->X_op = O_register;
5817 e->X_add_number = r - i386_regtab;
5818 return 1;
5819 }
5820 input_line_pointer = end;
5821 *end = 0;
5822 return 0;
5823 }
5824
5825 void
5826 md_operand (expressionS *e)
5827 {
5828 if (*input_line_pointer == REGISTER_PREFIX)
5829 {
5830 char *end;
5831 const reg_entry *r = parse_real_register (input_line_pointer, &end);
5832
5833 if (r)
5834 {
5835 e->X_op = O_register;
5836 e->X_add_number = r - i386_regtab;
5837 input_line_pointer = end;
5838 }
5839 }
5840 }
5841
5842 \f
5843 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5844 const char *md_shortopts = "kVQ:sqn";
5845 #else
5846 const char *md_shortopts = "qn";
5847 #endif
5848
5849 #define OPTION_32 (OPTION_MD_BASE + 0)
5850 #define OPTION_64 (OPTION_MD_BASE + 1)
5851 #define OPTION_DIVIDE (OPTION_MD_BASE + 2)
5852 #define OPTION_MARCH (OPTION_MD_BASE + 3)
5853 #define OPTION_MTUNE (OPTION_MD_BASE + 4)
5854
5855 struct option md_longopts[] =
5856 {
5857 {"32", no_argument, NULL, OPTION_32},
5858 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
5859 {"64", no_argument, NULL, OPTION_64},
5860 #endif
5861 {"divide", no_argument, NULL, OPTION_DIVIDE},
5862 {"march", required_argument, NULL, OPTION_MARCH},
5863 {"mtune", required_argument, NULL, OPTION_MTUNE},
5864 {NULL, no_argument, NULL, 0}
5865 };
5866 size_t md_longopts_size = sizeof (md_longopts);
5867
5868 int
5869 md_parse_option (int c, char *arg)
5870 {
5871 unsigned int i;
5872
5873 switch (c)
5874 {
5875 case 'n':
5876 optimize_align_code = 0;
5877 break;
5878
5879 case 'q':
5880 quiet_warnings = 1;
5881 break;
5882
5883 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5884 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
5885 should be emitted or not. FIXME: Not implemented. */
5886 case 'Q':
5887 break;
5888
5889 /* -V: SVR4 argument to print version ID. */
5890 case 'V':
5891 print_version_id ();
5892 break;
5893
5894 /* -k: Ignore for FreeBSD compatibility. */
5895 case 'k':
5896 break;
5897
5898 case 's':
5899 /* -s: On i386 Solaris, this tells the native assembler to use
5900 .stab instead of .stab.excl. We always use .stab anyhow. */
5901 break;
5902 #endif
5903 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
5904 case OPTION_64:
5905 {
5906 const char **list, **l;
5907
5908 list = bfd_target_list ();
5909 for (l = list; *l != NULL; l++)
5910 if (CONST_STRNEQ (*l, "elf64-x86-64")
5911 || strcmp (*l, "coff-x86-64") == 0
5912 || strcmp (*l, "pe-x86-64") == 0
5913 || strcmp (*l, "pei-x86-64") == 0)
5914 {
5915 default_arch = "x86_64";
5916 break;
5917 }
5918 if (*l == NULL)
5919 as_fatal (_("No compiled in support for x86_64"));
5920 free (list);
5921 }
5922 break;
5923 #endif
5924
5925 case OPTION_32:
5926 default_arch = "i386";
5927 break;
5928
5929 case OPTION_DIVIDE:
5930 #ifdef SVR4_COMMENT_CHARS
5931 {
5932 char *n, *t;
5933 const char *s;
5934
5935 n = (char *) xmalloc (strlen (i386_comment_chars) + 1);
5936 t = n;
5937 for (s = i386_comment_chars; *s != '\0'; s++)
5938 if (*s != '/')
5939 *t++ = *s;
5940 *t = '\0';
5941 i386_comment_chars = n;
5942 }
5943 #endif
5944 break;
5945
5946 case OPTION_MARCH:
5947 if (*arg == '.')
5948 as_fatal (_("Invalid -march= option: `%s'"), arg);
5949 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
5950 {
5951 if (strcmp (arg, cpu_arch [i].name) == 0)
5952 {
5953 cpu_arch_isa = cpu_arch[i].type;
5954 cpu_arch_isa_flags = cpu_arch[i].flags;
5955 if (!cpu_arch_tune_set)
5956 {
5957 cpu_arch_tune = cpu_arch_isa;
5958 cpu_arch_tune_flags = cpu_arch_isa_flags;
5959 }
5960 break;
5961 }
5962 }
5963 if (i >= ARRAY_SIZE (cpu_arch))
5964 as_fatal (_("Invalid -march= option: `%s'"), arg);
5965 break;
5966
5967 case OPTION_MTUNE:
5968 if (*arg == '.')
5969 as_fatal (_("Invalid -mtune= option: `%s'"), arg);
5970 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
5971 {
5972 if (strcmp (arg, cpu_arch [i].name) == 0)
5973 {
5974 cpu_arch_tune_set = 1;
5975 cpu_arch_tune = cpu_arch [i].type;
5976 cpu_arch_tune_flags = cpu_arch[i].flags;
5977 break;
5978 }
5979 }
5980 if (i >= ARRAY_SIZE (cpu_arch))
5981 as_fatal (_("Invalid -mtune= option: `%s'"), arg);
5982 break;
5983
5984 default:
5985 return 0;
5986 }
5987 return 1;
5988 }
5989
5990 void
5991 md_show_usage (stream)
5992 FILE *stream;
5993 {
5994 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5995 fprintf (stream, _("\
5996 -Q ignored\n\
5997 -V print assembler version number\n\
5998 -k ignored\n"));
5999 #endif
6000 fprintf (stream, _("\
6001 -n Do not optimize code alignment\n\
6002 -q quieten some warnings\n"));
6003 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
6004 fprintf (stream, _("\
6005 -s ignored\n"));
6006 #endif
6007 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
6008 fprintf (stream, _("\
6009 --32/--64 generate 32bit/64bit code\n"));
6010 #endif
6011 #ifdef SVR4_COMMENT_CHARS
6012 fprintf (stream, _("\
6013 --divide do not treat `/' as a comment character\n"));
6014 #else
6015 fprintf (stream, _("\
6016 --divide ignored\n"));
6017 #endif
6018 fprintf (stream, _("\
6019 -march=CPU/-mtune=CPU generate code/optimize for CPU, where CPU is one of:\n\
6020 i386, i486, pentium, pentiumpro, pentium4, nocona,\n\
6021 core, core2, k6, athlon, k8, generic32, generic64\n"));
6022
6023 }
6024
6025 #if defined(TE_PEP)
6026 const char *
6027 x86_64_target_format (void)
6028 {
6029 if (strcmp (default_arch, "x86_64") == 0)
6030 {
6031 set_code_flag (CODE_64BIT);
6032 return COFF_TARGET_FORMAT;
6033 }
6034 else if (strcmp (default_arch, "i386") == 0)
6035 {
6036 set_code_flag (CODE_32BIT);
6037 return "coff-i386";
6038 }
6039
6040 as_fatal (_("Unknown architecture"));
6041 return NULL;
6042 }
6043 #endif
6044
6045 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
6046 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
6047
6048 /* Pick the target format to use. */
6049
6050 const char *
6051 i386_target_format (void)
6052 {
6053 if (!strcmp (default_arch, "x86_64"))
6054 {
6055 set_code_flag (CODE_64BIT);
6056 if (cpu_arch_isa_flags == 0)
6057 cpu_arch_isa_flags = Cpu186|Cpu286|Cpu386|Cpu486
6058 |Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2
6059 |CpuSSE|CpuSSE2;
6060 if (cpu_arch_tune_flags == 0)
6061 cpu_arch_tune_flags = Cpu186|Cpu286|Cpu386|Cpu486
6062 |Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2
6063 |CpuSSE|CpuSSE2;
6064 }
6065 else if (!strcmp (default_arch, "i386"))
6066 {
6067 set_code_flag (CODE_32BIT);
6068 if (cpu_arch_isa_flags == 0)
6069 cpu_arch_isa_flags = Cpu186|Cpu286|Cpu386;
6070 if (cpu_arch_tune_flags == 0)
6071 cpu_arch_tune_flags = Cpu186|Cpu286|Cpu386;
6072 }
6073 else
6074 as_fatal (_("Unknown architecture"));
6075 switch (OUTPUT_FLAVOR)
6076 {
6077 #ifdef OBJ_MAYBE_AOUT
6078 case bfd_target_aout_flavour:
6079 return AOUT_TARGET_FORMAT;
6080 #endif
6081 #ifdef OBJ_MAYBE_COFF
6082 case bfd_target_coff_flavour:
6083 return "coff-i386";
6084 #endif
6085 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
6086 case bfd_target_elf_flavour:
6087 {
6088 if (flag_code == CODE_64BIT)
6089 {
6090 object_64bit = 1;
6091 use_rela_relocations = 1;
6092 }
6093 return flag_code == CODE_64BIT ? ELF_TARGET_FORMAT64 : ELF_TARGET_FORMAT;
6094 }
6095 #endif
6096 default:
6097 abort ();
6098 return NULL;
6099 }
6100 }
6101
6102 #endif /* OBJ_MAYBE_ more than one */
6103
6104 #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
6105 void
6106 i386_elf_emit_arch_note (void)
6107 {
6108 if (IS_ELF && cpu_arch_name != NULL)
6109 {
6110 char *p;
6111 asection *seg = now_seg;
6112 subsegT subseg = now_subseg;
6113 Elf_Internal_Note i_note;
6114 Elf_External_Note e_note;
6115 asection *note_secp;
6116 int len;
6117
6118 /* Create the .note section. */
6119 note_secp = subseg_new (".note", 0);
6120 bfd_set_section_flags (stdoutput,
6121 note_secp,
6122 SEC_HAS_CONTENTS | SEC_READONLY);
6123
6124 /* Process the arch string. */
6125 len = strlen (cpu_arch_name);
6126
6127 i_note.namesz = len + 1;
6128 i_note.descsz = 0;
6129 i_note.type = NT_ARCH;
6130 p = frag_more (sizeof (e_note.namesz));
6131 md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
6132 p = frag_more (sizeof (e_note.descsz));
6133 md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
6134 p = frag_more (sizeof (e_note.type));
6135 md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
6136 p = frag_more (len + 1);
6137 strcpy (p, cpu_arch_name);
6138
6139 frag_align (2, 0, 0);
6140
6141 subseg_set (seg, subseg);
6142 }
6143 }
6144 #endif
6145 \f
6146 symbolS *
6147 md_undefined_symbol (name)
6148 char *name;
6149 {
6150 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
6151 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
6152 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
6153 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
6154 {
6155 if (!GOT_symbol)
6156 {
6157 if (symbol_find (name))
6158 as_bad (_("GOT already in symbol table"));
6159 GOT_symbol = symbol_new (name, undefined_section,
6160 (valueT) 0, &zero_address_frag);
6161 };
6162 return GOT_symbol;
6163 }
6164 return 0;
6165 }
6166
6167 /* Round up a section size to the appropriate boundary. */
6168
6169 valueT
6170 md_section_align (segment, size)
6171 segT segment ATTRIBUTE_UNUSED;
6172 valueT size;
6173 {
6174 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
6175 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
6176 {
6177 /* For a.out, force the section size to be aligned. If we don't do
6178 this, BFD will align it for us, but it will not write out the
6179 final bytes of the section. This may be a bug in BFD, but it is
6180 easier to fix it here since that is how the other a.out targets
6181 work. */
6182 int align;
6183
6184 align = bfd_get_section_alignment (stdoutput, segment);
6185 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
6186 }
6187 #endif
6188
6189 return size;
6190 }
6191
6192 /* On the i386, PC-relative offsets are relative to the start of the
6193 next instruction. That is, the address of the offset, plus its
6194 size, since the offset is always the last part of the insn. */
6195
6196 long
6197 md_pcrel_from (fixS *fixP)
6198 {
6199 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
6200 }
6201
6202 #ifndef I386COFF
6203
6204 static void
6205 s_bss (int ignore ATTRIBUTE_UNUSED)
6206 {
6207 int temp;
6208
6209 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
6210 if (IS_ELF)
6211 obj_elf_section_change_hook ();
6212 #endif
6213 temp = get_absolute_expression ();
6214 subseg_set (bss_section, (subsegT) temp);
6215 demand_empty_rest_of_line ();
6216 }
6217
6218 #endif
6219
6220 void
6221 i386_validate_fix (fixS *fixp)
6222 {
6223 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
6224 {
6225 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
6226 {
6227 if (!object_64bit)
6228 abort ();
6229 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
6230 }
6231 else
6232 {
6233 if (!object_64bit)
6234 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
6235 else
6236 fixp->fx_r_type = BFD_RELOC_X86_64_GOTOFF64;
6237 }
6238 fixp->fx_subsy = 0;
6239 }
6240 }
6241
6242 arelent *
6243 tc_gen_reloc (section, fixp)
6244 asection *section ATTRIBUTE_UNUSED;
6245 fixS *fixp;
6246 {
6247 arelent *rel;
6248 bfd_reloc_code_real_type code;
6249
6250 switch (fixp->fx_r_type)
6251 {
6252 case BFD_RELOC_X86_64_PLT32:
6253 case BFD_RELOC_X86_64_GOT32:
6254 case BFD_RELOC_X86_64_GOTPCREL:
6255 case BFD_RELOC_386_PLT32:
6256 case BFD_RELOC_386_GOT32:
6257 case BFD_RELOC_386_GOTOFF:
6258 case BFD_RELOC_386_GOTPC:
6259 case BFD_RELOC_386_TLS_GD:
6260 case BFD_RELOC_386_TLS_LDM:
6261 case BFD_RELOC_386_TLS_LDO_32:
6262 case BFD_RELOC_386_TLS_IE_32:
6263 case BFD_RELOC_386_TLS_IE:
6264 case BFD_RELOC_386_TLS_GOTIE:
6265 case BFD_RELOC_386_TLS_LE_32:
6266 case BFD_RELOC_386_TLS_LE:
6267 case BFD_RELOC_386_TLS_GOTDESC:
6268 case BFD_RELOC_386_TLS_DESC_CALL:
6269 case BFD_RELOC_X86_64_TLSGD:
6270 case BFD_RELOC_X86_64_TLSLD:
6271 case BFD_RELOC_X86_64_DTPOFF32:
6272 case BFD_RELOC_X86_64_DTPOFF64:
6273 case BFD_RELOC_X86_64_GOTTPOFF:
6274 case BFD_RELOC_X86_64_TPOFF32:
6275 case BFD_RELOC_X86_64_TPOFF64:
6276 case BFD_RELOC_X86_64_GOTOFF64:
6277 case BFD_RELOC_X86_64_GOTPC32:
6278 case BFD_RELOC_X86_64_GOT64:
6279 case BFD_RELOC_X86_64_GOTPCREL64:
6280 case BFD_RELOC_X86_64_GOTPC64:
6281 case BFD_RELOC_X86_64_GOTPLT64:
6282 case BFD_RELOC_X86_64_PLTOFF64:
6283 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
6284 case BFD_RELOC_X86_64_TLSDESC_CALL:
6285 case BFD_RELOC_RVA:
6286 case BFD_RELOC_VTABLE_ENTRY:
6287 case BFD_RELOC_VTABLE_INHERIT:
6288 #ifdef TE_PE
6289 case BFD_RELOC_32_SECREL:
6290 #endif
6291 code = fixp->fx_r_type;
6292 break;
6293 case BFD_RELOC_X86_64_32S:
6294 if (!fixp->fx_pcrel)
6295 {
6296 /* Don't turn BFD_RELOC_X86_64_32S into BFD_RELOC_32. */
6297 code = fixp->fx_r_type;
6298 break;
6299 }
6300 default:
6301 if (fixp->fx_pcrel)
6302 {
6303 switch (fixp->fx_size)
6304 {
6305 default:
6306 as_bad_where (fixp->fx_file, fixp->fx_line,
6307 _("can not do %d byte pc-relative relocation"),
6308 fixp->fx_size);
6309 code = BFD_RELOC_32_PCREL;
6310 break;
6311 case 1: code = BFD_RELOC_8_PCREL; break;
6312 case 2: code = BFD_RELOC_16_PCREL; break;
6313 case 4: code = BFD_RELOC_32_PCREL; break;
6314 #ifdef BFD64
6315 case 8: code = BFD_RELOC_64_PCREL; break;
6316 #endif
6317 }
6318 }
6319 else
6320 {
6321 switch (fixp->fx_size)
6322 {
6323 default:
6324 as_bad_where (fixp->fx_file, fixp->fx_line,
6325 _("can not do %d byte relocation"),
6326 fixp->fx_size);
6327 code = BFD_RELOC_32;
6328 break;
6329 case 1: code = BFD_RELOC_8; break;
6330 case 2: code = BFD_RELOC_16; break;
6331 case 4: code = BFD_RELOC_32; break;
6332 #ifdef BFD64
6333 case 8: code = BFD_RELOC_64; break;
6334 #endif
6335 }
6336 }
6337 break;
6338 }
6339
6340 if ((code == BFD_RELOC_32
6341 || code == BFD_RELOC_32_PCREL
6342 || code == BFD_RELOC_X86_64_32S)
6343 && GOT_symbol
6344 && fixp->fx_addsy == GOT_symbol)
6345 {
6346 if (!object_64bit)
6347 code = BFD_RELOC_386_GOTPC;
6348 else
6349 code = BFD_RELOC_X86_64_GOTPC32;
6350 }
6351 if ((code == BFD_RELOC_64 || code == BFD_RELOC_64_PCREL)
6352 && GOT_symbol
6353 && fixp->fx_addsy == GOT_symbol)
6354 {
6355 code = BFD_RELOC_X86_64_GOTPC64;
6356 }
6357
6358 rel = (arelent *) xmalloc (sizeof (arelent));
6359 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
6360 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
6361
6362 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
6363
6364 if (!use_rela_relocations)
6365 {
6366 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
6367 vtable entry to be used in the relocation's section offset. */
6368 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
6369 rel->address = fixp->fx_offset;
6370
6371 rel->addend = 0;
6372 }
6373 /* Use the rela in 64bit mode. */
6374 else
6375 {
6376 if (!fixp->fx_pcrel)
6377 rel->addend = fixp->fx_offset;
6378 else
6379 switch (code)
6380 {
6381 case BFD_RELOC_X86_64_PLT32:
6382 case BFD_RELOC_X86_64_GOT32:
6383 case BFD_RELOC_X86_64_GOTPCREL:
6384 case BFD_RELOC_X86_64_TLSGD:
6385 case BFD_RELOC_X86_64_TLSLD:
6386 case BFD_RELOC_X86_64_GOTTPOFF:
6387 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
6388 case BFD_RELOC_X86_64_TLSDESC_CALL:
6389 rel->addend = fixp->fx_offset - fixp->fx_size;
6390 break;
6391 default:
6392 rel->addend = (section->vma
6393 - fixp->fx_size
6394 + fixp->fx_addnumber
6395 + md_pcrel_from (fixp));
6396 break;
6397 }
6398 }
6399
6400 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
6401 if (rel->howto == NULL)
6402 {
6403 as_bad_where (fixp->fx_file, fixp->fx_line,
6404 _("cannot represent relocation type %s"),
6405 bfd_get_reloc_code_name (code));
6406 /* Set howto to a garbage value so that we can keep going. */
6407 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
6408 assert (rel->howto != NULL);
6409 }
6410
6411 return rel;
6412 }
6413
6414 \f
6415 /* Parse operands using Intel syntax. This implements a recursive descent
6416 parser based on the BNF grammar published in Appendix B of the MASM 6.1
6417 Programmer's Guide.
6418
6419 FIXME: We do not recognize the full operand grammar defined in the MASM
6420 documentation. In particular, all the structure/union and
6421 high-level macro operands are missing.
6422
6423 Uppercase words are terminals, lower case words are non-terminals.
6424 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
6425 bars '|' denote choices. Most grammar productions are implemented in
6426 functions called 'intel_<production>'.
6427
6428 Initial production is 'expr'.
6429
6430 addOp + | -
6431
6432 alpha [a-zA-Z]
6433
6434 binOp & | AND | \| | OR | ^ | XOR
6435
6436 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
6437
6438 constant digits [[ radixOverride ]]
6439
6440 dataType BYTE | WORD | DWORD | FWORD | QWORD | TBYTE | OWORD | XMMWORD
6441
6442 digits decdigit
6443 | digits decdigit
6444 | digits hexdigit
6445
6446 decdigit [0-9]
6447
6448 e04 e04 addOp e05
6449 | e05
6450
6451 e05 e05 binOp e06
6452 | e06
6453
6454 e06 e06 mulOp e09
6455 | e09
6456
6457 e09 OFFSET e10
6458 | SHORT e10
6459 | + e10
6460 | - e10
6461 | ~ e10
6462 | NOT e10
6463 | e09 PTR e10
6464 | e09 : e10
6465 | e10
6466
6467 e10 e10 [ expr ]
6468 | e11
6469
6470 e11 ( expr )
6471 | [ expr ]
6472 | constant
6473 | dataType
6474 | id
6475 | $
6476 | register
6477
6478 => expr expr cmpOp e04
6479 | e04
6480
6481 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
6482 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
6483
6484 hexdigit a | b | c | d | e | f
6485 | A | B | C | D | E | F
6486
6487 id alpha
6488 | id alpha
6489 | id decdigit
6490
6491 mulOp * | / | % | MOD | << | SHL | >> | SHR
6492
6493 quote " | '
6494
6495 register specialRegister
6496 | gpRegister
6497 | byteRegister
6498
6499 segmentRegister CS | DS | ES | FS | GS | SS
6500
6501 specialRegister CR0 | CR2 | CR3 | CR4
6502 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
6503 | TR3 | TR4 | TR5 | TR6 | TR7
6504
6505 We simplify the grammar in obvious places (e.g., register parsing is
6506 done by calling parse_register) and eliminate immediate left recursion
6507 to implement a recursive-descent parser.
6508
6509 expr e04 expr'
6510
6511 expr' cmpOp e04 expr'
6512 | Empty
6513
6514 e04 e05 e04'
6515
6516 e04' addOp e05 e04'
6517 | Empty
6518
6519 e05 e06 e05'
6520
6521 e05' binOp e06 e05'
6522 | Empty
6523
6524 e06 e09 e06'
6525
6526 e06' mulOp e09 e06'
6527 | Empty
6528
6529 e09 OFFSET e10 e09'
6530 | SHORT e10'
6531 | + e10'
6532 | - e10'
6533 | ~ e10'
6534 | NOT e10'
6535 | e10 e09'
6536
6537 e09' PTR e10 e09'
6538 | : e10 e09'
6539 | Empty
6540
6541 e10 e11 e10'
6542
6543 e10' [ expr ] e10'
6544 | Empty
6545
6546 e11 ( expr )
6547 | [ expr ]
6548 | BYTE
6549 | WORD
6550 | DWORD
6551 | FWORD
6552 | QWORD
6553 | TBYTE
6554 | OWORD
6555 | XMMWORD
6556 | .
6557 | $
6558 | register
6559 | id
6560 | constant */
6561
6562 /* Parsing structure for the intel syntax parser. Used to implement the
6563 semantic actions for the operand grammar. */
6564 struct intel_parser_s
6565 {
6566 char *op_string; /* The string being parsed. */
6567 int got_a_float; /* Whether the operand is a float. */
6568 int op_modifier; /* Operand modifier. */
6569 int is_mem; /* 1 if operand is memory reference. */
6570 int in_offset; /* >=1 if parsing operand of offset. */
6571 int in_bracket; /* >=1 if parsing operand in brackets. */
6572 const reg_entry *reg; /* Last register reference found. */
6573 char *disp; /* Displacement string being built. */
6574 char *next_operand; /* Resume point when splitting operands. */
6575 };
6576
6577 static struct intel_parser_s intel_parser;
6578
6579 /* Token structure for parsing intel syntax. */
6580 struct intel_token
6581 {
6582 int code; /* Token code. */
6583 const reg_entry *reg; /* Register entry for register tokens. */
6584 char *str; /* String representation. */
6585 };
6586
6587 static struct intel_token cur_token, prev_token;
6588
6589 /* Token codes for the intel parser. Since T_SHORT is already used
6590 by COFF, undefine it first to prevent a warning. */
6591 #define T_NIL -1
6592 #define T_CONST 1
6593 #define T_REG 2
6594 #define T_BYTE 3
6595 #define T_WORD 4
6596 #define T_DWORD 5
6597 #define T_FWORD 6
6598 #define T_QWORD 7
6599 #define T_TBYTE 8
6600 #define T_XMMWORD 9
6601 #undef T_SHORT
6602 #define T_SHORT 10
6603 #define T_OFFSET 11
6604 #define T_PTR 12
6605 #define T_ID 13
6606 #define T_SHL 14
6607 #define T_SHR 15
6608
6609 /* Prototypes for intel parser functions. */
6610 static int intel_match_token (int);
6611 static void intel_putback_token (void);
6612 static void intel_get_token (void);
6613 static int intel_expr (void);
6614 static int intel_e04 (void);
6615 static int intel_e05 (void);
6616 static int intel_e06 (void);
6617 static int intel_e09 (void);
6618 static int intel_e10 (void);
6619 static int intel_e11 (void);
6620
6621 static int
6622 i386_intel_operand (char *operand_string, int got_a_float)
6623 {
6624 int ret;
6625 char *p;
6626
6627 p = intel_parser.op_string = xstrdup (operand_string);
6628 intel_parser.disp = (char *) xmalloc (strlen (operand_string) + 1);
6629
6630 for (;;)
6631 {
6632 /* Initialize token holders. */
6633 cur_token.code = prev_token.code = T_NIL;
6634 cur_token.reg = prev_token.reg = NULL;
6635 cur_token.str = prev_token.str = NULL;
6636
6637 /* Initialize parser structure. */
6638 intel_parser.got_a_float = got_a_float;
6639 intel_parser.op_modifier = 0;
6640 intel_parser.is_mem = 0;
6641 intel_parser.in_offset = 0;
6642 intel_parser.in_bracket = 0;
6643 intel_parser.reg = NULL;
6644 intel_parser.disp[0] = '\0';
6645 intel_parser.next_operand = NULL;
6646
6647 /* Read the first token and start the parser. */
6648 intel_get_token ();
6649 ret = intel_expr ();
6650
6651 if (!ret)
6652 break;
6653
6654 if (cur_token.code != T_NIL)
6655 {
6656 as_bad (_("invalid operand for '%s' ('%s' unexpected)"),
6657 current_templates->start->name, cur_token.str);
6658 ret = 0;
6659 }
6660 /* If we found a memory reference, hand it over to i386_displacement
6661 to fill in the rest of the operand fields. */
6662 else if (intel_parser.is_mem)
6663 {
6664 if ((i.mem_operands == 1
6665 && (current_templates->start->opcode_modifier & IsString) == 0)
6666 || i.mem_operands == 2)
6667 {
6668 as_bad (_("too many memory references for '%s'"),
6669 current_templates->start->name);
6670 ret = 0;
6671 }
6672 else
6673 {
6674 char *s = intel_parser.disp;
6675 i.mem_operands++;
6676
6677 if (!quiet_warnings && intel_parser.is_mem < 0)
6678 /* See the comments in intel_bracket_expr. */
6679 as_warn (_("Treating `%s' as memory reference"), operand_string);
6680
6681 /* Add the displacement expression. */
6682 if (*s != '\0')
6683 ret = i386_displacement (s, s + strlen (s));
6684 if (ret)
6685 {
6686 /* Swap base and index in 16-bit memory operands like
6687 [si+bx]. Since i386_index_check is also used in AT&T
6688 mode we have to do that here. */
6689 if (i.base_reg
6690 && i.index_reg
6691 && (i.base_reg->reg_type & Reg16)
6692 && (i.index_reg->reg_type & Reg16)
6693 && i.base_reg->reg_num >= 6
6694 && i.index_reg->reg_num < 6)
6695 {
6696 const reg_entry *base = i.index_reg;
6697
6698 i.index_reg = i.base_reg;
6699 i.base_reg = base;
6700 }
6701 ret = i386_index_check (operand_string);
6702 }
6703 }
6704 }
6705
6706 /* Constant and OFFSET expressions are handled by i386_immediate. */
6707 else if ((intel_parser.op_modifier & (1 << T_OFFSET))
6708 || intel_parser.reg == NULL)
6709 ret = i386_immediate (intel_parser.disp);
6710
6711 if (intel_parser.next_operand && this_operand >= MAX_OPERANDS - 1)
6712 ret = 0;
6713 if (!ret || !intel_parser.next_operand)
6714 break;
6715 intel_parser.op_string = intel_parser.next_operand;
6716 this_operand = i.operands++;
6717 }
6718
6719 free (p);
6720 free (intel_parser.disp);
6721
6722 return ret;
6723 }
6724
6725 #define NUM_ADDRESS_REGS (!!i.base_reg + !!i.index_reg)
6726
6727 /* expr e04 expr'
6728
6729 expr' cmpOp e04 expr'
6730 | Empty */
6731 static int
6732 intel_expr (void)
6733 {
6734 /* XXX Implement the comparison operators. */
6735 return intel_e04 ();
6736 }
6737
6738 /* e04 e05 e04'
6739
6740 e04' addOp e05 e04'
6741 | Empty */
6742 static int
6743 intel_e04 (void)
6744 {
6745 int nregs = -1;
6746
6747 for (;;)
6748 {
6749 if (!intel_e05())
6750 return 0;
6751
6752 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6753 i.base_reg = i386_regtab + REGNAM_AL; /* al is invalid as base */
6754
6755 if (cur_token.code == '+')
6756 nregs = -1;
6757 else if (cur_token.code == '-')
6758 nregs = NUM_ADDRESS_REGS;
6759 else
6760 return 1;
6761
6762 strcat (intel_parser.disp, cur_token.str);
6763 intel_match_token (cur_token.code);
6764 }
6765 }
6766
6767 /* e05 e06 e05'
6768
6769 e05' binOp e06 e05'
6770 | Empty */
6771 static int
6772 intel_e05 (void)
6773 {
6774 int nregs = ~NUM_ADDRESS_REGS;
6775
6776 for (;;)
6777 {
6778 if (!intel_e06())
6779 return 0;
6780
6781 if (cur_token.code == '&'
6782 || cur_token.code == '|'
6783 || cur_token.code == '^')
6784 {
6785 char str[2];
6786
6787 str[0] = cur_token.code;
6788 str[1] = 0;
6789 strcat (intel_parser.disp, str);
6790 }
6791 else
6792 break;
6793
6794 intel_match_token (cur_token.code);
6795
6796 if (nregs < 0)
6797 nregs = ~nregs;
6798 }
6799 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6800 i.base_reg = i386_regtab + REGNAM_AL + 1; /* cl is invalid as base */
6801 return 1;
6802 }
6803
6804 /* e06 e09 e06'
6805
6806 e06' mulOp e09 e06'
6807 | Empty */
6808 static int
6809 intel_e06 (void)
6810 {
6811 int nregs = ~NUM_ADDRESS_REGS;
6812
6813 for (;;)
6814 {
6815 if (!intel_e09())
6816 return 0;
6817
6818 if (cur_token.code == '*'
6819 || cur_token.code == '/'
6820 || cur_token.code == '%')
6821 {
6822 char str[2];
6823
6824 str[0] = cur_token.code;
6825 str[1] = 0;
6826 strcat (intel_parser.disp, str);
6827 }
6828 else if (cur_token.code == T_SHL)
6829 strcat (intel_parser.disp, "<<");
6830 else if (cur_token.code == T_SHR)
6831 strcat (intel_parser.disp, ">>");
6832 else
6833 break;
6834
6835 intel_match_token (cur_token.code);
6836
6837 if (nregs < 0)
6838 nregs = ~nregs;
6839 }
6840 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6841 i.base_reg = i386_regtab + REGNAM_AL + 2; /* dl is invalid as base */
6842 return 1;
6843 }
6844
6845 /* e09 OFFSET e09
6846 | SHORT e09
6847 | + e09
6848 | - e09
6849 | ~ e09
6850 | NOT e09
6851 | e10 e09'
6852
6853 e09' PTR e10 e09'
6854 | : e10 e09'
6855 | Empty */
6856 static int
6857 intel_e09 (void)
6858 {
6859 int nregs = ~NUM_ADDRESS_REGS;
6860 int in_offset = 0;
6861
6862 for (;;)
6863 {
6864 /* Don't consume constants here. */
6865 if (cur_token.code == '+' || cur_token.code == '-')
6866 {
6867 /* Need to look one token ahead - if the next token
6868 is a constant, the current token is its sign. */
6869 int next_code;
6870
6871 intel_match_token (cur_token.code);
6872 next_code = cur_token.code;
6873 intel_putback_token ();
6874 if (next_code == T_CONST)
6875 break;
6876 }
6877
6878 /* e09 OFFSET e09 */
6879 if (cur_token.code == T_OFFSET)
6880 {
6881 if (!in_offset++)
6882 ++intel_parser.in_offset;
6883 }
6884
6885 /* e09 SHORT e09 */
6886 else if (cur_token.code == T_SHORT)
6887 intel_parser.op_modifier |= 1 << T_SHORT;
6888
6889 /* e09 + e09 */
6890 else if (cur_token.code == '+')
6891 strcat (intel_parser.disp, "+");
6892
6893 /* e09 - e09
6894 | ~ e09
6895 | NOT e09 */
6896 else if (cur_token.code == '-' || cur_token.code == '~')
6897 {
6898 char str[2];
6899
6900 if (nregs < 0)
6901 nregs = ~nregs;
6902 str[0] = cur_token.code;
6903 str[1] = 0;
6904 strcat (intel_parser.disp, str);
6905 }
6906
6907 /* e09 e10 e09' */
6908 else
6909 break;
6910
6911 intel_match_token (cur_token.code);
6912 }
6913
6914 for (;;)
6915 {
6916 if (!intel_e10 ())
6917 return 0;
6918
6919 /* e09' PTR e10 e09' */
6920 if (cur_token.code == T_PTR)
6921 {
6922 char suffix;
6923
6924 if (prev_token.code == T_BYTE)
6925 suffix = BYTE_MNEM_SUFFIX;
6926
6927 else if (prev_token.code == T_WORD)
6928 {
6929 if (current_templates->start->name[0] == 'l'
6930 && current_templates->start->name[2] == 's'
6931 && current_templates->start->name[3] == 0)
6932 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
6933 else if (intel_parser.got_a_float == 2) /* "fi..." */
6934 suffix = SHORT_MNEM_SUFFIX;
6935 else
6936 suffix = WORD_MNEM_SUFFIX;
6937 }
6938
6939 else if (prev_token.code == T_DWORD)
6940 {
6941 if (current_templates->start->name[0] == 'l'
6942 && current_templates->start->name[2] == 's'
6943 && current_templates->start->name[3] == 0)
6944 suffix = WORD_MNEM_SUFFIX;
6945 else if (flag_code == CODE_16BIT
6946 && (current_templates->start->opcode_modifier
6947 & (Jump | JumpDword)))
6948 suffix = LONG_DOUBLE_MNEM_SUFFIX;
6949 else if (intel_parser.got_a_float == 1) /* "f..." */
6950 suffix = SHORT_MNEM_SUFFIX;
6951 else
6952 suffix = LONG_MNEM_SUFFIX;
6953 }
6954
6955 else if (prev_token.code == T_FWORD)
6956 {
6957 if (current_templates->start->name[0] == 'l'
6958 && current_templates->start->name[2] == 's'
6959 && current_templates->start->name[3] == 0)
6960 suffix = LONG_MNEM_SUFFIX;
6961 else if (!intel_parser.got_a_float)
6962 {
6963 if (flag_code == CODE_16BIT)
6964 add_prefix (DATA_PREFIX_OPCODE);
6965 suffix = LONG_DOUBLE_MNEM_SUFFIX;
6966 }
6967 else
6968 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
6969 }
6970
6971 else if (prev_token.code == T_QWORD)
6972 {
6973 if (intel_parser.got_a_float == 1) /* "f..." */
6974 suffix = LONG_MNEM_SUFFIX;
6975 else
6976 suffix = QWORD_MNEM_SUFFIX;
6977 }
6978
6979 else if (prev_token.code == T_TBYTE)
6980 {
6981 if (intel_parser.got_a_float == 1)
6982 suffix = LONG_DOUBLE_MNEM_SUFFIX;
6983 else
6984 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
6985 }
6986
6987 else if (prev_token.code == T_XMMWORD)
6988 {
6989 /* XXX ignored for now, but accepted since gcc uses it */
6990 suffix = 0;
6991 }
6992
6993 else
6994 {
6995 as_bad (_("Unknown operand modifier `%s'"), prev_token.str);
6996 return 0;
6997 }
6998
6999 /* Operands for jump/call using 'ptr' notation denote absolute
7000 addresses. */
7001 if (current_templates->start->opcode_modifier & (Jump | JumpDword))
7002 i.types[this_operand] |= JumpAbsolute;
7003
7004 if (current_templates->start->base_opcode == 0x8d /* lea */)
7005 ;
7006 else if (!i.suffix)
7007 i.suffix = suffix;
7008 else if (i.suffix != suffix)
7009 {
7010 as_bad (_("Conflicting operand modifiers"));
7011 return 0;
7012 }
7013
7014 }
7015
7016 /* e09' : e10 e09' */
7017 else if (cur_token.code == ':')
7018 {
7019 if (prev_token.code != T_REG)
7020 {
7021 /* While {call,jmp} SSSS:OOOO is MASM syntax only when SSSS is a
7022 segment/group identifier (which we don't have), using comma
7023 as the operand separator there is even less consistent, since
7024 there all branches only have a single operand. */
7025 if (this_operand != 0
7026 || intel_parser.in_offset
7027 || intel_parser.in_bracket
7028 || (!(current_templates->start->opcode_modifier
7029 & (Jump|JumpDword|JumpInterSegment))
7030 && !(current_templates->start->operand_types[0]
7031 & JumpAbsolute)))
7032 return intel_match_token (T_NIL);
7033 /* Remember the start of the 2nd operand and terminate 1st
7034 operand here.
7035 XXX This isn't right, yet (when SSSS:OOOO is right operand of
7036 another expression), but it gets at least the simplest case
7037 (a plain number or symbol on the left side) right. */
7038 intel_parser.next_operand = intel_parser.op_string;
7039 *--intel_parser.op_string = '\0';
7040 return intel_match_token (':');
7041 }
7042 }
7043
7044 /* e09' Empty */
7045 else
7046 break;
7047
7048 intel_match_token (cur_token.code);
7049
7050 }
7051
7052 if (in_offset)
7053 {
7054 --intel_parser.in_offset;
7055 if (nregs < 0)
7056 nregs = ~nregs;
7057 if (NUM_ADDRESS_REGS > nregs)
7058 {
7059 as_bad (_("Invalid operand to `OFFSET'"));
7060 return 0;
7061 }
7062 intel_parser.op_modifier |= 1 << T_OFFSET;
7063 }
7064
7065 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
7066 i.base_reg = i386_regtab + REGNAM_AL + 3; /* bl is invalid as base */
7067 return 1;
7068 }
7069
7070 static int
7071 intel_bracket_expr (void)
7072 {
7073 int was_offset = intel_parser.op_modifier & (1 << T_OFFSET);
7074 const char *start = intel_parser.op_string;
7075 int len;
7076
7077 if (i.op[this_operand].regs)
7078 return intel_match_token (T_NIL);
7079
7080 intel_match_token ('[');
7081
7082 /* Mark as a memory operand only if it's not already known to be an
7083 offset expression. If it's an offset expression, we need to keep
7084 the brace in. */
7085 if (!intel_parser.in_offset)
7086 {
7087 ++intel_parser.in_bracket;
7088
7089 /* Operands for jump/call inside brackets denote absolute addresses. */
7090 if (current_templates->start->opcode_modifier & (Jump | JumpDword))
7091 i.types[this_operand] |= JumpAbsolute;
7092
7093 /* Unfortunately gas always diverged from MASM in a respect that can't
7094 be easily fixed without risking to break code sequences likely to be
7095 encountered (the testsuite even check for this): MASM doesn't consider
7096 an expression inside brackets unconditionally as a memory reference.
7097 When that is e.g. a constant, an offset expression, or the sum of the
7098 two, this is still taken as a constant load. gas, however, always
7099 treated these as memory references. As a compromise, we'll try to make
7100 offset expressions inside brackets work the MASM way (since that's
7101 less likely to be found in real world code), but make constants alone
7102 continue to work the traditional gas way. In either case, issue a
7103 warning. */
7104 intel_parser.op_modifier &= ~was_offset;
7105 }
7106 else
7107 strcat (intel_parser.disp, "[");
7108
7109 /* Add a '+' to the displacement string if necessary. */
7110 if (*intel_parser.disp != '\0'
7111 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
7112 strcat (intel_parser.disp, "+");
7113
7114 if (intel_expr ()
7115 && (len = intel_parser.op_string - start - 1,
7116 intel_match_token (']')))
7117 {
7118 /* Preserve brackets when the operand is an offset expression. */
7119 if (intel_parser.in_offset)
7120 strcat (intel_parser.disp, "]");
7121 else
7122 {
7123 --intel_parser.in_bracket;
7124 if (i.base_reg || i.index_reg)
7125 intel_parser.is_mem = 1;
7126 if (!intel_parser.is_mem)
7127 {
7128 if (!(intel_parser.op_modifier & (1 << T_OFFSET)))
7129 /* Defer the warning until all of the operand was parsed. */
7130 intel_parser.is_mem = -1;
7131 else if (!quiet_warnings)
7132 as_warn (_("`[%.*s]' taken to mean just `%.*s'"),
7133 len, start, len, start);
7134 }
7135 }
7136 intel_parser.op_modifier |= was_offset;
7137
7138 return 1;
7139 }
7140 return 0;
7141 }
7142
7143 /* e10 e11 e10'
7144
7145 e10' [ expr ] e10'
7146 | Empty */
7147 static int
7148 intel_e10 (void)
7149 {
7150 if (!intel_e11 ())
7151 return 0;
7152
7153 while (cur_token.code == '[')
7154 {
7155 if (!intel_bracket_expr ())
7156 return 0;
7157 }
7158
7159 return 1;
7160 }
7161
7162 /* e11 ( expr )
7163 | [ expr ]
7164 | BYTE
7165 | WORD
7166 | DWORD
7167 | FWORD
7168 | QWORD
7169 | TBYTE
7170 | OWORD
7171 | XMMWORD
7172 | $
7173 | .
7174 | register
7175 | id
7176 | constant */
7177 static int
7178 intel_e11 (void)
7179 {
7180 switch (cur_token.code)
7181 {
7182 /* e11 ( expr ) */
7183 case '(':
7184 intel_match_token ('(');
7185 strcat (intel_parser.disp, "(");
7186
7187 if (intel_expr () && intel_match_token (')'))
7188 {
7189 strcat (intel_parser.disp, ")");
7190 return 1;
7191 }
7192 return 0;
7193
7194 /* e11 [ expr ] */
7195 case '[':
7196 return intel_bracket_expr ();
7197
7198 /* e11 $
7199 | . */
7200 case '.':
7201 strcat (intel_parser.disp, cur_token.str);
7202 intel_match_token (cur_token.code);
7203
7204 /* Mark as a memory operand only if it's not already known to be an
7205 offset expression. */
7206 if (!intel_parser.in_offset)
7207 intel_parser.is_mem = 1;
7208
7209 return 1;
7210
7211 /* e11 register */
7212 case T_REG:
7213 {
7214 const reg_entry *reg = intel_parser.reg = cur_token.reg;
7215
7216 intel_match_token (T_REG);
7217
7218 /* Check for segment change. */
7219 if (cur_token.code == ':')
7220 {
7221 if (!(reg->reg_type & (SReg2 | SReg3)))
7222 {
7223 as_bad (_("`%s' is not a valid segment register"),
7224 reg->reg_name);
7225 return 0;
7226 }
7227 else if (i.seg[i.mem_operands])
7228 as_warn (_("Extra segment override ignored"));
7229 else
7230 {
7231 if (!intel_parser.in_offset)
7232 intel_parser.is_mem = 1;
7233 switch (reg->reg_num)
7234 {
7235 case 0:
7236 i.seg[i.mem_operands] = &es;
7237 break;
7238 case 1:
7239 i.seg[i.mem_operands] = &cs;
7240 break;
7241 case 2:
7242 i.seg[i.mem_operands] = &ss;
7243 break;
7244 case 3:
7245 i.seg[i.mem_operands] = &ds;
7246 break;
7247 case 4:
7248 i.seg[i.mem_operands] = &fs;
7249 break;
7250 case 5:
7251 i.seg[i.mem_operands] = &gs;
7252 break;
7253 }
7254 }
7255 }
7256
7257 /* Not a segment register. Check for register scaling. */
7258 else if (cur_token.code == '*')
7259 {
7260 if (!intel_parser.in_bracket)
7261 {
7262 as_bad (_("Register scaling only allowed in memory operands"));
7263 return 0;
7264 }
7265
7266 if (reg->reg_type & Reg16) /* Disallow things like [si*1]. */
7267 reg = i386_regtab + REGNAM_AX + 4; /* sp is invalid as index */
7268 else if (i.index_reg)
7269 reg = i386_regtab + REGNAM_EAX + 4; /* esp is invalid as index */
7270
7271 /* What follows must be a valid scale. */
7272 intel_match_token ('*');
7273 i.index_reg = reg;
7274 i.types[this_operand] |= BaseIndex;
7275
7276 /* Set the scale after setting the register (otherwise,
7277 i386_scale will complain) */
7278 if (cur_token.code == '+' || cur_token.code == '-')
7279 {
7280 char *str, sign = cur_token.code;
7281 intel_match_token (cur_token.code);
7282 if (cur_token.code != T_CONST)
7283 {
7284 as_bad (_("Syntax error: Expecting a constant, got `%s'"),
7285 cur_token.str);
7286 return 0;
7287 }
7288 str = (char *) xmalloc (strlen (cur_token.str) + 2);
7289 strcpy (str + 1, cur_token.str);
7290 *str = sign;
7291 if (!i386_scale (str))
7292 return 0;
7293 free (str);
7294 }
7295 else if (!i386_scale (cur_token.str))
7296 return 0;
7297 intel_match_token (cur_token.code);
7298 }
7299
7300 /* No scaling. If this is a memory operand, the register is either a
7301 base register (first occurrence) or an index register (second
7302 occurrence). */
7303 else if (intel_parser.in_bracket)
7304 {
7305
7306 if (!i.base_reg)
7307 i.base_reg = reg;
7308 else if (!i.index_reg)
7309 i.index_reg = reg;
7310 else
7311 {
7312 as_bad (_("Too many register references in memory operand"));
7313 return 0;
7314 }
7315
7316 i.types[this_operand] |= BaseIndex;
7317 }
7318
7319 /* It's neither base nor index. */
7320 else if (!intel_parser.in_offset && !intel_parser.is_mem)
7321 {
7322 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
7323 i.op[this_operand].regs = reg;
7324 i.reg_operands++;
7325 }
7326 else
7327 {
7328 as_bad (_("Invalid use of register"));
7329 return 0;
7330 }
7331
7332 /* Since registers are not part of the displacement string (except
7333 when we're parsing offset operands), we may need to remove any
7334 preceding '+' from the displacement string. */
7335 if (*intel_parser.disp != '\0'
7336 && !intel_parser.in_offset)
7337 {
7338 char *s = intel_parser.disp;
7339 s += strlen (s) - 1;
7340 if (*s == '+')
7341 *s = '\0';
7342 }
7343
7344 return 1;
7345 }
7346
7347 /* e11 BYTE
7348 | WORD
7349 | DWORD
7350 | FWORD
7351 | QWORD
7352 | TBYTE
7353 | OWORD
7354 | XMMWORD */
7355 case T_BYTE:
7356 case T_WORD:
7357 case T_DWORD:
7358 case T_FWORD:
7359 case T_QWORD:
7360 case T_TBYTE:
7361 case T_XMMWORD:
7362 intel_match_token (cur_token.code);
7363
7364 if (cur_token.code == T_PTR)
7365 return 1;
7366
7367 /* It must have been an identifier. */
7368 intel_putback_token ();
7369 cur_token.code = T_ID;
7370 /* FALLTHRU */
7371
7372 /* e11 id
7373 | constant */
7374 case T_ID:
7375 if (!intel_parser.in_offset && intel_parser.is_mem <= 0)
7376 {
7377 symbolS *symbolP;
7378
7379 /* The identifier represents a memory reference only if it's not
7380 preceded by an offset modifier and if it's not an equate. */
7381 symbolP = symbol_find(cur_token.str);
7382 if (!symbolP || S_GET_SEGMENT(symbolP) != absolute_section)
7383 intel_parser.is_mem = 1;
7384 }
7385 /* FALLTHRU */
7386
7387 case T_CONST:
7388 case '-':
7389 case '+':
7390 {
7391 char *save_str, sign = 0;
7392
7393 /* Allow constants that start with `+' or `-'. */
7394 if (cur_token.code == '-' || cur_token.code == '+')
7395 {
7396 sign = cur_token.code;
7397 intel_match_token (cur_token.code);
7398 if (cur_token.code != T_CONST)
7399 {
7400 as_bad (_("Syntax error: Expecting a constant, got `%s'"),
7401 cur_token.str);
7402 return 0;
7403 }
7404 }
7405
7406 save_str = (char *) xmalloc (strlen (cur_token.str) + 2);
7407 strcpy (save_str + !!sign, cur_token.str);
7408 if (sign)
7409 *save_str = sign;
7410
7411 /* Get the next token to check for register scaling. */
7412 intel_match_token (cur_token.code);
7413
7414 /* Check if this constant is a scaling factor for an
7415 index register. */
7416 if (cur_token.code == '*')
7417 {
7418 if (intel_match_token ('*') && cur_token.code == T_REG)
7419 {
7420 const reg_entry *reg = cur_token.reg;
7421
7422 if (!intel_parser.in_bracket)
7423 {
7424 as_bad (_("Register scaling only allowed "
7425 "in memory operands"));
7426 return 0;
7427 }
7428
7429 /* Disallow things like [1*si].
7430 sp and esp are invalid as index. */
7431 if (reg->reg_type & Reg16)
7432 reg = i386_regtab + REGNAM_AX + 4;
7433 else if (i.index_reg)
7434 reg = i386_regtab + REGNAM_EAX + 4;
7435
7436 /* The constant is followed by `* reg', so it must be
7437 a valid scale. */
7438 i.index_reg = reg;
7439 i.types[this_operand] |= BaseIndex;
7440
7441 /* Set the scale after setting the register (otherwise,
7442 i386_scale will complain) */
7443 if (!i386_scale (save_str))
7444 return 0;
7445 intel_match_token (T_REG);
7446
7447 /* Since registers are not part of the displacement
7448 string, we may need to remove any preceding '+' from
7449 the displacement string. */
7450 if (*intel_parser.disp != '\0')
7451 {
7452 char *s = intel_parser.disp;
7453 s += strlen (s) - 1;
7454 if (*s == '+')
7455 *s = '\0';
7456 }
7457
7458 free (save_str);
7459
7460 return 1;
7461 }
7462
7463 /* The constant was not used for register scaling. Since we have
7464 already consumed the token following `*' we now need to put it
7465 back in the stream. */
7466 intel_putback_token ();
7467 }
7468
7469 /* Add the constant to the displacement string. */
7470 strcat (intel_parser.disp, save_str);
7471 free (save_str);
7472
7473 return 1;
7474 }
7475 }
7476
7477 as_bad (_("Unrecognized token '%s'"), cur_token.str);
7478 return 0;
7479 }
7480
7481 /* Match the given token against cur_token. If they match, read the next
7482 token from the operand string. */
7483 static int
7484 intel_match_token (int code)
7485 {
7486 if (cur_token.code == code)
7487 {
7488 intel_get_token ();
7489 return 1;
7490 }
7491 else
7492 {
7493 as_bad (_("Unexpected token `%s'"), cur_token.str);
7494 return 0;
7495 }
7496 }
7497
7498 /* Read a new token from intel_parser.op_string and store it in cur_token. */
7499 static void
7500 intel_get_token (void)
7501 {
7502 char *end_op;
7503 const reg_entry *reg;
7504 struct intel_token new_token;
7505
7506 new_token.code = T_NIL;
7507 new_token.reg = NULL;
7508 new_token.str = NULL;
7509
7510 /* Free the memory allocated to the previous token and move
7511 cur_token to prev_token. */
7512 if (prev_token.str)
7513 free (prev_token.str);
7514
7515 prev_token = cur_token;
7516
7517 /* Skip whitespace. */
7518 while (is_space_char (*intel_parser.op_string))
7519 intel_parser.op_string++;
7520
7521 /* Return an empty token if we find nothing else on the line. */
7522 if (*intel_parser.op_string == '\0')
7523 {
7524 cur_token = new_token;
7525 return;
7526 }
7527
7528 /* The new token cannot be larger than the remainder of the operand
7529 string. */
7530 new_token.str = (char *) xmalloc (strlen (intel_parser.op_string) + 1);
7531 new_token.str[0] = '\0';
7532
7533 if (strchr ("0123456789", *intel_parser.op_string))
7534 {
7535 char *p = new_token.str;
7536 char *q = intel_parser.op_string;
7537 new_token.code = T_CONST;
7538
7539 /* Allow any kind of identifier char to encompass floating point and
7540 hexadecimal numbers. */
7541 while (is_identifier_char (*q))
7542 *p++ = *q++;
7543 *p = '\0';
7544
7545 /* Recognize special symbol names [0-9][bf]. */
7546 if (strlen (intel_parser.op_string) == 2
7547 && (intel_parser.op_string[1] == 'b'
7548 || intel_parser.op_string[1] == 'f'))
7549 new_token.code = T_ID;
7550 }
7551
7552 else if ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL)
7553 {
7554 size_t len = end_op - intel_parser.op_string;
7555
7556 new_token.code = T_REG;
7557 new_token.reg = reg;
7558
7559 memcpy (new_token.str, intel_parser.op_string, len);
7560 new_token.str[len] = '\0';
7561 }
7562
7563 else if (is_identifier_char (*intel_parser.op_string))
7564 {
7565 char *p = new_token.str;
7566 char *q = intel_parser.op_string;
7567
7568 /* A '.' or '$' followed by an identifier char is an identifier.
7569 Otherwise, it's operator '.' followed by an expression. */
7570 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
7571 {
7572 new_token.code = '.';
7573 new_token.str[0] = '.';
7574 new_token.str[1] = '\0';
7575 }
7576 else
7577 {
7578 while (is_identifier_char (*q) || *q == '@')
7579 *p++ = *q++;
7580 *p = '\0';
7581
7582 if (strcasecmp (new_token.str, "NOT") == 0)
7583 new_token.code = '~';
7584
7585 else if (strcasecmp (new_token.str, "MOD") == 0)
7586 new_token.code = '%';
7587
7588 else if (strcasecmp (new_token.str, "AND") == 0)
7589 new_token.code = '&';
7590
7591 else if (strcasecmp (new_token.str, "OR") == 0)
7592 new_token.code = '|';
7593
7594 else if (strcasecmp (new_token.str, "XOR") == 0)
7595 new_token.code = '^';
7596
7597 else if (strcasecmp (new_token.str, "SHL") == 0)
7598 new_token.code = T_SHL;
7599
7600 else if (strcasecmp (new_token.str, "SHR") == 0)
7601 new_token.code = T_SHR;
7602
7603 else if (strcasecmp (new_token.str, "BYTE") == 0)
7604 new_token.code = T_BYTE;
7605
7606 else if (strcasecmp (new_token.str, "WORD") == 0)
7607 new_token.code = T_WORD;
7608
7609 else if (strcasecmp (new_token.str, "DWORD") == 0)
7610 new_token.code = T_DWORD;
7611
7612 else if (strcasecmp (new_token.str, "FWORD") == 0)
7613 new_token.code = T_FWORD;
7614
7615 else if (strcasecmp (new_token.str, "QWORD") == 0)
7616 new_token.code = T_QWORD;
7617
7618 else if (strcasecmp (new_token.str, "TBYTE") == 0
7619 /* XXX remove (gcc still uses it) */
7620 || strcasecmp (new_token.str, "XWORD") == 0)
7621 new_token.code = T_TBYTE;
7622
7623 else if (strcasecmp (new_token.str, "XMMWORD") == 0
7624 || strcasecmp (new_token.str, "OWORD") == 0)
7625 new_token.code = T_XMMWORD;
7626
7627 else if (strcasecmp (new_token.str, "PTR") == 0)
7628 new_token.code = T_PTR;
7629
7630 else if (strcasecmp (new_token.str, "SHORT") == 0)
7631 new_token.code = T_SHORT;
7632
7633 else if (strcasecmp (new_token.str, "OFFSET") == 0)
7634 {
7635 new_token.code = T_OFFSET;
7636
7637 /* ??? This is not mentioned in the MASM grammar but gcc
7638 makes use of it with -mintel-syntax. OFFSET may be
7639 followed by FLAT: */
7640 if (strncasecmp (q, " FLAT:", 6) == 0)
7641 strcat (new_token.str, " FLAT:");
7642 }
7643
7644 /* ??? This is not mentioned in the MASM grammar. */
7645 else if (strcasecmp (new_token.str, "FLAT") == 0)
7646 {
7647 new_token.code = T_OFFSET;
7648 if (*q == ':')
7649 strcat (new_token.str, ":");
7650 else
7651 as_bad (_("`:' expected"));
7652 }
7653
7654 else
7655 new_token.code = T_ID;
7656 }
7657 }
7658
7659 else if (strchr ("+-/*%|&^:[]()~", *intel_parser.op_string))
7660 {
7661 new_token.code = *intel_parser.op_string;
7662 new_token.str[0] = *intel_parser.op_string;
7663 new_token.str[1] = '\0';
7664 }
7665
7666 else if (strchr ("<>", *intel_parser.op_string)
7667 && *intel_parser.op_string == *(intel_parser.op_string + 1))
7668 {
7669 new_token.code = *intel_parser.op_string == '<' ? T_SHL : T_SHR;
7670 new_token.str[0] = *intel_parser.op_string;
7671 new_token.str[1] = *intel_parser.op_string;
7672 new_token.str[2] = '\0';
7673 }
7674
7675 else
7676 as_bad (_("Unrecognized token `%s'"), intel_parser.op_string);
7677
7678 intel_parser.op_string += strlen (new_token.str);
7679 cur_token = new_token;
7680 }
7681
7682 /* Put cur_token back into the token stream and make cur_token point to
7683 prev_token. */
7684 static void
7685 intel_putback_token (void)
7686 {
7687 if (cur_token.code != T_NIL)
7688 {
7689 intel_parser.op_string -= strlen (cur_token.str);
7690 free (cur_token.str);
7691 }
7692 cur_token = prev_token;
7693
7694 /* Forget prev_token. */
7695 prev_token.code = T_NIL;
7696 prev_token.reg = NULL;
7697 prev_token.str = NULL;
7698 }
7699
7700 int
7701 tc_x86_regname_to_dw2regnum (char *regname)
7702 {
7703 unsigned int regnum;
7704 unsigned int regnames_count;
7705 static const char *const regnames_32[] =
7706 {
7707 "eax", "ecx", "edx", "ebx",
7708 "esp", "ebp", "esi", "edi",
7709 "eip", "eflags", NULL,
7710 "st0", "st1", "st2", "st3",
7711 "st4", "st5", "st6", "st7",
7712 NULL, NULL,
7713 "xmm0", "xmm1", "xmm2", "xmm3",
7714 "xmm4", "xmm5", "xmm6", "xmm7",
7715 "mm0", "mm1", "mm2", "mm3",
7716 "mm4", "mm5", "mm6", "mm7",
7717 "fcw", "fsw", "mxcsr",
7718 "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL,
7719 "tr", "ldtr"
7720 };
7721 static const char *const regnames_64[] =
7722 {
7723 "rax", "rdx", "rcx", "rbx",
7724 "rsi", "rdi", "rbp", "rsp",
7725 "r8", "r9", "r10", "r11",
7726 "r12", "r13", "r14", "r15",
7727 "rip",
7728 "xmm0", "xmm1", "xmm2", "xmm3",
7729 "xmm4", "xmm5", "xmm6", "xmm7",
7730 "xmm8", "xmm9", "xmm10", "xmm11",
7731 "xmm12", "xmm13", "xmm14", "xmm15",
7732 "st0", "st1", "st2", "st3",
7733 "st4", "st5", "st6", "st7",
7734 "mm0", "mm1", "mm2", "mm3",
7735 "mm4", "mm5", "mm6", "mm7",
7736 "rflags",
7737 "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL,
7738 "fs.base", "gs.base", NULL, NULL,
7739 "tr", "ldtr",
7740 "mxcsr", "fcw", "fsw"
7741 };
7742 const char *const *regnames;
7743
7744 if (flag_code == CODE_64BIT)
7745 {
7746 regnames = regnames_64;
7747 regnames_count = ARRAY_SIZE (regnames_64);
7748 }
7749 else
7750 {
7751 regnames = regnames_32;
7752 regnames_count = ARRAY_SIZE (regnames_32);
7753 }
7754
7755 for (regnum = 0; regnum < regnames_count; regnum++)
7756 if (regnames[regnum] != NULL
7757 && strcmp (regname, regnames[regnum]) == 0)
7758 return regnum;
7759
7760 return -1;
7761 }
7762
7763 void
7764 tc_x86_frame_initial_instructions (void)
7765 {
7766 static unsigned int sp_regno;
7767
7768 if (!sp_regno)
7769 sp_regno = tc_x86_regname_to_dw2regnum (flag_code == CODE_64BIT
7770 ? "rsp" : "esp");
7771
7772 cfi_add_CFA_def_cfa (sp_regno, -x86_cie_data_alignment);
7773 cfi_add_CFA_offset (x86_dwarf2_return_column, x86_cie_data_alignment);
7774 }
7775
7776 int
7777 i386_elf_section_type (const char *str, size_t len)
7778 {
7779 if (flag_code == CODE_64BIT
7780 && len == sizeof ("unwind") - 1
7781 && strncmp (str, "unwind", 6) == 0)
7782 return SHT_X86_64_UNWIND;
7783
7784 return -1;
7785 }
7786
7787 #ifdef TE_PE
7788 void
7789 tc_pe_dwarf2_emit_offset (symbolS *symbol, unsigned int size)
7790 {
7791 expressionS expr;
7792
7793 expr.X_op = O_secrel;
7794 expr.X_add_symbol = symbol;
7795 expr.X_add_number = 0;
7796 emit_expr (&expr, size);
7797 }
7798 #endif
7799
7800 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
7801 /* For ELF on x86-64, add support for SHF_X86_64_LARGE. */
7802
7803 int
7804 x86_64_section_letter (int letter, char **ptr_msg)
7805 {
7806 if (flag_code == CODE_64BIT)
7807 {
7808 if (letter == 'l')
7809 return SHF_X86_64_LARGE;
7810
7811 *ptr_msg = _("Bad .section directive: want a,l,w,x,M,S,G,T in string");
7812 }
7813 else
7814 *ptr_msg = _("Bad .section directive: want a,w,x,M,S,G,T in string");
7815 return -1;
7816 }
7817
7818 int
7819 x86_64_section_word (char *str, size_t len)
7820 {
7821 if (len == 5 && flag_code == CODE_64BIT && CONST_STRNEQ (str, "large"))
7822 return SHF_X86_64_LARGE;
7823
7824 return -1;
7825 }
7826
7827 static void
7828 handle_large_common (int small ATTRIBUTE_UNUSED)
7829 {
7830 if (flag_code != CODE_64BIT)
7831 {
7832 s_comm_internal (0, elf_common_parse);
7833 as_warn (_(".largecomm supported only in 64bit mode, producing .comm"));
7834 }
7835 else
7836 {
7837 static segT lbss_section;
7838 asection *saved_com_section_ptr = elf_com_section_ptr;
7839 asection *saved_bss_section = bss_section;
7840
7841 if (lbss_section == NULL)
7842 {
7843 flagword applicable;
7844 segT seg = now_seg;
7845 subsegT subseg = now_subseg;
7846
7847 /* The .lbss section is for local .largecomm symbols. */
7848 lbss_section = subseg_new (".lbss", 0);
7849 applicable = bfd_applicable_section_flags (stdoutput);
7850 bfd_set_section_flags (stdoutput, lbss_section,
7851 applicable & SEC_ALLOC);
7852 seg_info (lbss_section)->bss = 1;
7853
7854 subseg_set (seg, subseg);
7855 }
7856
7857 elf_com_section_ptr = &_bfd_elf_large_com_section;
7858 bss_section = lbss_section;
7859
7860 s_comm_internal (0, elf_common_parse);
7861
7862 elf_com_section_ptr = saved_com_section_ptr;
7863 bss_section = saved_bss_section;
7864 }
7865 }
7866 #endif /* OBJ_ELF || OBJ_MAYBE_ELF */
This page took 0.225201 seconds and 4 git commands to generate.