daily update
[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 3, 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_I386,
435 Cpu186|Cpu286|Cpu386},
436 {"i486", PROCESSOR_I486,
437 Cpu186|Cpu286|Cpu386|Cpu486},
438 {"i586", PROCESSOR_PENTIUM,
439 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586},
440 {"i686", PROCESSOR_PENTIUMPRO,
441 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686},
442 {"pentium", PROCESSOR_PENTIUM,
443 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586},
444 {"pentiumpro",PROCESSOR_PENTIUMPRO,
445 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686},
446 {"pentiumii", PROCESSOR_PENTIUMPRO,
447 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX},
448 {"pentiumiii",PROCESSOR_PENTIUMPRO,
449 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuMMX2|CpuSSE},
450 {"pentium4", PROCESSOR_PENTIUM4,
451 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
452 |CpuMMX2|CpuSSE|CpuSSE2},
453 {"prescott", PROCESSOR_NOCONA,
454 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
455 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
456 {"nocona", PROCESSOR_NOCONA,
457 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
458 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
459 {"yonah", PROCESSOR_CORE,
460 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
461 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
462 {"core", PROCESSOR_CORE,
463 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
464 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
465 {"merom", PROCESSOR_CORE2,
466 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
467 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3},
468 {"core2", PROCESSOR_CORE2,
469 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX
470 |CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3},
471 {"k6", PROCESSOR_K6,
472 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX},
473 {"k6_2", PROCESSOR_K6,
474 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow},
475 {"athlon", PROCESSOR_ATHLON,
476 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
477 |CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA},
478 {"sledgehammer", PROCESSOR_K8,
479 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
480 |CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2},
481 {"opteron", PROCESSOR_K8,
482 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
483 |CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2},
484 {"k8", PROCESSOR_K8,
485 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6
486 |CpuSledgehammer|CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2},
487 {"amdfam10", PROCESSOR_AMDFAM10,
488 Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuSledgehammer
489 |CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA|CpuSSE|CpuSSE2|CpuSSE3|CpuSSE4a
490 |CpuABM},
491 {".mmx", PROCESSOR_UNKNOWN,
492 CpuMMX},
493 {".sse", PROCESSOR_UNKNOWN,
494 CpuMMX|CpuMMX2|CpuSSE},
495 {".sse2", PROCESSOR_UNKNOWN,
496 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2},
497 {".sse3", PROCESSOR_UNKNOWN,
498 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3},
499 {".ssse3", PROCESSOR_UNKNOWN,
500 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3},
501 {".sse4.1", PROCESSOR_UNKNOWN,
502 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3|CpuSSE4_1},
503 {".sse4.2", PROCESSOR_UNKNOWN,
504 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3|CpuSSE4},
505 {".sse4", PROCESSOR_UNKNOWN,
506 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSSE3|CpuSSE4},
507 {".3dnow", PROCESSOR_UNKNOWN,
508 CpuMMX|Cpu3dnow},
509 {".3dnowa", PROCESSOR_UNKNOWN,
510 CpuMMX|CpuMMX2|Cpu3dnow|Cpu3dnowA},
511 {".padlock", PROCESSOR_UNKNOWN,
512 CpuPadLock},
513 {".pacifica", PROCESSOR_UNKNOWN,
514 CpuSVME},
515 {".svme", PROCESSOR_UNKNOWN,
516 CpuSVME},
517 {".sse4a", PROCESSOR_UNKNOWN,
518 CpuMMX|CpuMMX2|CpuSSE|CpuSSE2|CpuSSE3|CpuSSE4a},
519 {".abm", PROCESSOR_UNKNOWN,
520 CpuABM}
521 };
522
523 const pseudo_typeS md_pseudo_table[] =
524 {
525 #if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
526 {"align", s_align_bytes, 0},
527 #else
528 {"align", s_align_ptwo, 0},
529 #endif
530 {"arch", set_cpu_arch, 0},
531 #ifndef I386COFF
532 {"bss", s_bss, 0},
533 #endif
534 {"ffloat", float_cons, 'f'},
535 {"dfloat", float_cons, 'd'},
536 {"tfloat", float_cons, 'x'},
537 {"value", cons, 2},
538 {"slong", signed_cons, 4},
539 {"noopt", s_ignore, 0},
540 {"optim", s_ignore, 0},
541 {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
542 {"code16", set_code_flag, CODE_16BIT},
543 {"code32", set_code_flag, CODE_32BIT},
544 {"code64", set_code_flag, CODE_64BIT},
545 {"intel_syntax", set_intel_syntax, 1},
546 {"att_syntax", set_intel_syntax, 0},
547 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
548 {"largecomm", handle_large_common, 0},
549 #else
550 {"file", (void (*) (int)) dwarf2_directive_file, 0},
551 {"loc", dwarf2_directive_loc, 0},
552 {"loc_mark_labels", dwarf2_directive_loc_mark_labels, 0},
553 #endif
554 #ifdef TE_PE
555 {"secrel32", pe_directive_secrel, 0},
556 #endif
557 {0, 0, 0}
558 };
559
560 /* For interface with expression (). */
561 extern char *input_line_pointer;
562
563 /* Hash table for instruction mnemonic lookup. */
564 static struct hash_control *op_hash;
565
566 /* Hash table for register lookup. */
567 static struct hash_control *reg_hash;
568 \f
569 void
570 i386_align_code (fragS *fragP, int count)
571 {
572 /* Various efficient no-op patterns for aligning code labels.
573 Note: Don't try to assemble the instructions in the comments.
574 0L and 0w are not legal. */
575 static const char f32_1[] =
576 {0x90}; /* nop */
577 static const char f32_2[] =
578 {0x66,0x90}; /* xchg %ax,%ax */
579 static const char f32_3[] =
580 {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
581 static const char f32_4[] =
582 {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
583 static const char f32_5[] =
584 {0x90, /* nop */
585 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
586 static const char f32_6[] =
587 {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
588 static const char f32_7[] =
589 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
590 static const char f32_8[] =
591 {0x90, /* nop */
592 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
593 static const char f32_9[] =
594 {0x89,0xf6, /* movl %esi,%esi */
595 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
596 static const char f32_10[] =
597 {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
598 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
599 static const char f32_11[] =
600 {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
601 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
602 static const char f32_12[] =
603 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
604 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
605 static const char f32_13[] =
606 {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
607 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
608 static const char f32_14[] =
609 {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
610 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
611 static const char f16_3[] =
612 {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
613 static const char f16_4[] =
614 {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
615 static const char f16_5[] =
616 {0x90, /* nop */
617 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
618 static const char f16_6[] =
619 {0x89,0xf6, /* mov %si,%si */
620 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
621 static const char f16_7[] =
622 {0x8d,0x74,0x00, /* lea 0(%si),%si */
623 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
624 static const char f16_8[] =
625 {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
626 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
627 static const char jump_31[] =
628 {0xeb,0x1d,0x90,0x90,0x90,0x90,0x90, /* jmp .+31; lotsa nops */
629 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90,
630 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90,
631 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
632 static const char *const f32_patt[] = {
633 f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
634 f32_9, f32_10, f32_11, f32_12, f32_13, f32_14
635 };
636 static const char *const f16_patt[] = {
637 f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8
638 };
639 /* nopl (%[re]ax) */
640 static const char alt_3[] =
641 {0x0f,0x1f,0x00};
642 /* nopl 0(%[re]ax) */
643 static const char alt_4[] =
644 {0x0f,0x1f,0x40,0x00};
645 /* nopl 0(%[re]ax,%[re]ax,1) */
646 static const char alt_5[] =
647 {0x0f,0x1f,0x44,0x00,0x00};
648 /* nopw 0(%[re]ax,%[re]ax,1) */
649 static const char alt_6[] =
650 {0x66,0x0f,0x1f,0x44,0x00,0x00};
651 /* nopl 0L(%[re]ax) */
652 static const char alt_7[] =
653 {0x0f,0x1f,0x80,0x00,0x00,0x00,0x00};
654 /* nopl 0L(%[re]ax,%[re]ax,1) */
655 static const char alt_8[] =
656 {0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
657 /* nopw 0L(%[re]ax,%[re]ax,1) */
658 static const char alt_9[] =
659 {0x66,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
660 /* nopw %cs:0L(%[re]ax,%[re]ax,1) */
661 static const char alt_10[] =
662 {0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
663 /* data16
664 nopw %cs:0L(%[re]ax,%[re]ax,1) */
665 static const char alt_long_11[] =
666 {0x66,
667 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
668 /* data16
669 data16
670 nopw %cs:0L(%[re]ax,%[re]ax,1) */
671 static const char alt_long_12[] =
672 {0x66,
673 0x66,
674 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
675 /* data16
676 data16
677 data16
678 nopw %cs:0L(%[re]ax,%[re]ax,1) */
679 static const char alt_long_13[] =
680 {0x66,
681 0x66,
682 0x66,
683 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
684 /* data16
685 data16
686 data16
687 data16
688 nopw %cs:0L(%[re]ax,%[re]ax,1) */
689 static const char alt_long_14[] =
690 {0x66,
691 0x66,
692 0x66,
693 0x66,
694 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
695 /* data16
696 data16
697 data16
698 data16
699 data16
700 nopw %cs:0L(%[re]ax,%[re]ax,1) */
701 static const char alt_long_15[] =
702 {0x66,
703 0x66,
704 0x66,
705 0x66,
706 0x66,
707 0x66,0x2e,0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
708 /* nopl 0(%[re]ax,%[re]ax,1)
709 nopw 0(%[re]ax,%[re]ax,1) */
710 static const char alt_short_11[] =
711 {0x0f,0x1f,0x44,0x00,0x00,
712 0x66,0x0f,0x1f,0x44,0x00,0x00};
713 /* nopw 0(%[re]ax,%[re]ax,1)
714 nopw 0(%[re]ax,%[re]ax,1) */
715 static const char alt_short_12[] =
716 {0x66,0x0f,0x1f,0x44,0x00,0x00,
717 0x66,0x0f,0x1f,0x44,0x00,0x00};
718 /* nopw 0(%[re]ax,%[re]ax,1)
719 nopl 0L(%[re]ax) */
720 static const char alt_short_13[] =
721 {0x66,0x0f,0x1f,0x44,0x00,0x00,
722 0x0f,0x1f,0x80,0x00,0x00,0x00,0x00};
723 /* nopl 0L(%[re]ax)
724 nopl 0L(%[re]ax) */
725 static const char alt_short_14[] =
726 {0x0f,0x1f,0x80,0x00,0x00,0x00,0x00,
727 0x0f,0x1f,0x80,0x00,0x00,0x00,0x00};
728 /* nopl 0L(%[re]ax)
729 nopl 0L(%[re]ax,%[re]ax,1) */
730 static const char alt_short_15[] =
731 {0x0f,0x1f,0x80,0x00,0x00,0x00,0x00,
732 0x0f,0x1f,0x84,0x00,0x00,0x00,0x00,0x00};
733 static const char *const alt_short_patt[] = {
734 f32_1, f32_2, alt_3, alt_4, alt_5, alt_6, alt_7, alt_8,
735 alt_9, alt_10, alt_short_11, alt_short_12, alt_short_13,
736 alt_short_14, alt_short_15
737 };
738 static const char *const alt_long_patt[] = {
739 f32_1, f32_2, alt_3, alt_4, alt_5, alt_6, alt_7, alt_8,
740 alt_9, alt_10, alt_long_11, alt_long_12, alt_long_13,
741 alt_long_14, alt_long_15
742 };
743
744 /* Only align for at least a positive non-zero boundary. */
745 if (count <= 0 || count > MAX_MEM_FOR_RS_ALIGN_CODE)
746 return;
747
748 /* We need to decide which NOP sequence to use for 32bit and
749 64bit. When -mtune= is used:
750
751 1. For PROCESSOR_I386, PROCESSOR_I486, PROCESSOR_PENTIUM and
752 PROCESSOR_GENERIC32, f32_patt will be used.
753 2. For PROCESSOR_PENTIUMPRO, PROCESSOR_PENTIUM4, PROCESSOR_NOCONA,
754 PROCESSOR_CORE, PROCESSOR_CORE2, and PROCESSOR_GENERIC64,
755 alt_long_patt will be used.
756 3. For PROCESSOR_ATHLON, PROCESSOR_K6, PROCESSOR_K8 and
757 PROCESSOR_AMDFAM10, alt_short_patt will be used.
758
759 When -mtune= isn't used, alt_long_patt will be used if
760 cpu_arch_isa_flags has Cpu686. Otherwise, f32_patt will
761 be used.
762
763 When -march= or .arch is used, we can't use anything beyond
764 cpu_arch_isa_flags. */
765
766 if (flag_code == CODE_16BIT)
767 {
768 if (count > 8)
769 {
770 memcpy (fragP->fr_literal + fragP->fr_fix,
771 jump_31, count);
772 /* Adjust jump offset. */
773 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
774 }
775 else
776 memcpy (fragP->fr_literal + fragP->fr_fix,
777 f16_patt[count - 1], count);
778 }
779 else
780 {
781 const char *const *patt = NULL;
782
783 if (cpu_arch_isa == PROCESSOR_UNKNOWN)
784 {
785 /* PROCESSOR_UNKNOWN means that all ISAs may be used. */
786 switch (cpu_arch_tune)
787 {
788 case PROCESSOR_UNKNOWN:
789 /* We use cpu_arch_isa_flags to check if we SHOULD
790 optimize for Cpu686. */
791 if ((cpu_arch_isa_flags & Cpu686) != 0)
792 patt = alt_long_patt;
793 else
794 patt = f32_patt;
795 break;
796 case PROCESSOR_PENTIUMPRO:
797 case PROCESSOR_PENTIUM4:
798 case PROCESSOR_NOCONA:
799 case PROCESSOR_CORE:
800 case PROCESSOR_CORE2:
801 case PROCESSOR_GENERIC64:
802 patt = alt_long_patt;
803 break;
804 case PROCESSOR_K6:
805 case PROCESSOR_ATHLON:
806 case PROCESSOR_K8:
807 case PROCESSOR_AMDFAM10:
808 patt = alt_short_patt;
809 break;
810 case PROCESSOR_I386:
811 case PROCESSOR_I486:
812 case PROCESSOR_PENTIUM:
813 case PROCESSOR_GENERIC32:
814 patt = f32_patt;
815 break;
816 }
817 }
818 else
819 {
820 switch (cpu_arch_tune)
821 {
822 case PROCESSOR_UNKNOWN:
823 /* When cpu_arch_isa is net, cpu_arch_tune shouldn't be
824 PROCESSOR_UNKNOWN. */
825 abort ();
826 break;
827
828 case PROCESSOR_I386:
829 case PROCESSOR_I486:
830 case PROCESSOR_PENTIUM:
831 case PROCESSOR_K6:
832 case PROCESSOR_ATHLON:
833 case PROCESSOR_K8:
834 case PROCESSOR_AMDFAM10:
835 case PROCESSOR_GENERIC32:
836 /* We use cpu_arch_isa_flags to check if we CAN optimize
837 for Cpu686. */
838 if ((cpu_arch_isa_flags & Cpu686) != 0)
839 patt = alt_short_patt;
840 else
841 patt = f32_patt;
842 break;
843 case PROCESSOR_PENTIUMPRO:
844 case PROCESSOR_PENTIUM4:
845 case PROCESSOR_NOCONA:
846 case PROCESSOR_CORE:
847 case PROCESSOR_CORE2:
848 if ((cpu_arch_isa_flags & Cpu686) != 0)
849 patt = alt_long_patt;
850 else
851 patt = f32_patt;
852 break;
853 case PROCESSOR_GENERIC64:
854 patt = alt_long_patt;
855 break;
856 }
857 }
858
859 if (patt == f32_patt)
860 {
861 /* If the padding is less than 15 bytes, we use the normal
862 ones. Otherwise, we use a jump instruction and adjust
863 its offset. */
864 if (count < 15)
865 memcpy (fragP->fr_literal + fragP->fr_fix,
866 patt[count - 1], count);
867 else
868 {
869 memcpy (fragP->fr_literal + fragP->fr_fix,
870 jump_31, count);
871 /* Adjust jump offset. */
872 fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
873 }
874 }
875 else
876 {
877 /* Maximum length of an instruction is 15 byte. If the
878 padding is greater than 15 bytes and we don't use jump,
879 we have to break it into smaller pieces. */
880 int padding = count;
881 while (padding > 15)
882 {
883 padding -= 15;
884 memcpy (fragP->fr_literal + fragP->fr_fix + padding,
885 patt [14], 15);
886 }
887
888 if (padding)
889 memcpy (fragP->fr_literal + fragP->fr_fix,
890 patt [padding - 1], padding);
891 }
892 }
893 fragP->fr_var = count;
894 }
895
896 static INLINE unsigned int
897 mode_from_disp_size (unsigned int t)
898 {
899 return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
900 }
901
902 static INLINE int
903 fits_in_signed_byte (offsetT num)
904 {
905 return (num >= -128) && (num <= 127);
906 }
907
908 static INLINE int
909 fits_in_unsigned_byte (offsetT num)
910 {
911 return (num & 0xff) == num;
912 }
913
914 static INLINE int
915 fits_in_unsigned_word (offsetT num)
916 {
917 return (num & 0xffff) == num;
918 }
919
920 static INLINE int
921 fits_in_signed_word (offsetT num)
922 {
923 return (-32768 <= num) && (num <= 32767);
924 }
925
926 static INLINE int
927 fits_in_signed_long (offsetT num ATTRIBUTE_UNUSED)
928 {
929 #ifndef BFD64
930 return 1;
931 #else
932 return (!(((offsetT) -1 << 31) & num)
933 || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
934 #endif
935 } /* fits_in_signed_long() */
936
937 static INLINE int
938 fits_in_unsigned_long (offsetT num ATTRIBUTE_UNUSED)
939 {
940 #ifndef BFD64
941 return 1;
942 #else
943 return (num & (((offsetT) 2 << 31) - 1)) == num;
944 #endif
945 } /* fits_in_unsigned_long() */
946
947 static unsigned int
948 smallest_imm_type (offsetT num)
949 {
950 if (cpu_arch_flags != (Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64))
951 {
952 /* This code is disabled on the 486 because all the Imm1 forms
953 in the opcode table are slower on the i486. They're the
954 versions with the implicitly specified single-position
955 displacement, which has another syntax if you really want to
956 use that form. */
957 if (num == 1)
958 return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
959 }
960 return (fits_in_signed_byte (num)
961 ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
962 : fits_in_unsigned_byte (num)
963 ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
964 : (fits_in_signed_word (num) || fits_in_unsigned_word (num))
965 ? (Imm16 | Imm32 | Imm32S | Imm64)
966 : fits_in_signed_long (num)
967 ? (Imm32 | Imm32S | Imm64)
968 : fits_in_unsigned_long (num)
969 ? (Imm32 | Imm64)
970 : Imm64);
971 }
972
973 static offsetT
974 offset_in_range (offsetT val, int size)
975 {
976 addressT mask;
977
978 switch (size)
979 {
980 case 1: mask = ((addressT) 1 << 8) - 1; break;
981 case 2: mask = ((addressT) 1 << 16) - 1; break;
982 case 4: mask = ((addressT) 2 << 31) - 1; break;
983 #ifdef BFD64
984 case 8: mask = ((addressT) 2 << 63) - 1; break;
985 #endif
986 default: abort ();
987 }
988
989 /* If BFD64, sign extend val. */
990 if (!use_rela_relocations)
991 if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
992 val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
993
994 if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
995 {
996 char buf1[40], buf2[40];
997
998 sprint_value (buf1, val);
999 sprint_value (buf2, val & mask);
1000 as_warn (_("%s shortened to %s"), buf1, buf2);
1001 }
1002 return val & mask;
1003 }
1004
1005 /* Returns 0 if attempting to add a prefix where one from the same
1006 class already exists, 1 if non rep/repne added, 2 if rep/repne
1007 added. */
1008 static int
1009 add_prefix (unsigned int prefix)
1010 {
1011 int ret = 1;
1012 unsigned int q;
1013
1014 if (prefix >= REX_OPCODE && prefix < REX_OPCODE + 16
1015 && flag_code == CODE_64BIT)
1016 {
1017 if ((i.prefix[REX_PREFIX] & prefix & REX_W)
1018 || ((i.prefix[REX_PREFIX] & (REX_R | REX_X | REX_B))
1019 && (prefix & (REX_R | REX_X | REX_B))))
1020 ret = 0;
1021 q = REX_PREFIX;
1022 }
1023 else
1024 {
1025 switch (prefix)
1026 {
1027 default:
1028 abort ();
1029
1030 case CS_PREFIX_OPCODE:
1031 case DS_PREFIX_OPCODE:
1032 case ES_PREFIX_OPCODE:
1033 case FS_PREFIX_OPCODE:
1034 case GS_PREFIX_OPCODE:
1035 case SS_PREFIX_OPCODE:
1036 q = SEG_PREFIX;
1037 break;
1038
1039 case REPNE_PREFIX_OPCODE:
1040 case REPE_PREFIX_OPCODE:
1041 ret = 2;
1042 /* fall thru */
1043 case LOCK_PREFIX_OPCODE:
1044 q = LOCKREP_PREFIX;
1045 break;
1046
1047 case FWAIT_OPCODE:
1048 q = WAIT_PREFIX;
1049 break;
1050
1051 case ADDR_PREFIX_OPCODE:
1052 q = ADDR_PREFIX;
1053 break;
1054
1055 case DATA_PREFIX_OPCODE:
1056 q = DATA_PREFIX;
1057 break;
1058 }
1059 if (i.prefix[q] != 0)
1060 ret = 0;
1061 }
1062
1063 if (ret)
1064 {
1065 if (!i.prefix[q])
1066 ++i.prefixes;
1067 i.prefix[q] |= prefix;
1068 }
1069 else
1070 as_bad (_("same type of prefix used twice"));
1071
1072 return ret;
1073 }
1074
1075 static void
1076 set_code_flag (int value)
1077 {
1078 flag_code = value;
1079 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
1080 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
1081 if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
1082 {
1083 as_bad (_("64bit mode not supported on this CPU."));
1084 }
1085 if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
1086 {
1087 as_bad (_("32bit mode not supported on this CPU."));
1088 }
1089 stackop_size = '\0';
1090 }
1091
1092 static void
1093 set_16bit_gcc_code_flag (int new_code_flag)
1094 {
1095 flag_code = new_code_flag;
1096 cpu_arch_flags &= ~(Cpu64 | CpuNo64);
1097 cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
1098 stackop_size = LONG_MNEM_SUFFIX;
1099 }
1100
1101 static void
1102 set_intel_syntax (int syntax_flag)
1103 {
1104 /* Find out if register prefixing is specified. */
1105 int ask_naked_reg = 0;
1106
1107 SKIP_WHITESPACE ();
1108 if (!is_end_of_line[(unsigned char) *input_line_pointer])
1109 {
1110 char *string = input_line_pointer;
1111 int e = get_symbol_end ();
1112
1113 if (strcmp (string, "prefix") == 0)
1114 ask_naked_reg = 1;
1115 else if (strcmp (string, "noprefix") == 0)
1116 ask_naked_reg = -1;
1117 else
1118 as_bad (_("bad argument to syntax directive."));
1119 *input_line_pointer = e;
1120 }
1121 demand_empty_rest_of_line ();
1122
1123 intel_syntax = syntax_flag;
1124
1125 if (ask_naked_reg == 0)
1126 allow_naked_reg = (intel_syntax
1127 && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
1128 else
1129 allow_naked_reg = (ask_naked_reg < 0);
1130
1131 identifier_chars['%'] = intel_syntax && allow_naked_reg ? '%' : 0;
1132 identifier_chars['$'] = intel_syntax ? '$' : 0;
1133 register_prefix = allow_naked_reg ? "" : "%";
1134 }
1135
1136 static void
1137 set_cpu_arch (int dummy ATTRIBUTE_UNUSED)
1138 {
1139 SKIP_WHITESPACE ();
1140
1141 if (!is_end_of_line[(unsigned char) *input_line_pointer])
1142 {
1143 char *string = input_line_pointer;
1144 int e = get_symbol_end ();
1145 unsigned int i;
1146
1147 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
1148 {
1149 if (strcmp (string, cpu_arch[i].name) == 0)
1150 {
1151 if (*string != '.')
1152 {
1153 cpu_arch_name = cpu_arch[i].name;
1154 cpu_sub_arch_name = NULL;
1155 cpu_arch_flags = (cpu_arch[i].flags
1156 | (flag_code == CODE_64BIT
1157 ? Cpu64 : CpuNo64));
1158 cpu_arch_isa = cpu_arch[i].type;
1159 cpu_arch_isa_flags = cpu_arch[i].flags;
1160 if (!cpu_arch_tune_set)
1161 {
1162 cpu_arch_tune = cpu_arch_isa;
1163 cpu_arch_tune_flags = cpu_arch_isa_flags;
1164 }
1165 break;
1166 }
1167 if ((cpu_arch_flags | cpu_arch[i].flags) != cpu_arch_flags)
1168 {
1169 cpu_sub_arch_name = cpu_arch[i].name;
1170 cpu_arch_flags |= cpu_arch[i].flags;
1171 }
1172 *input_line_pointer = e;
1173 demand_empty_rest_of_line ();
1174 return;
1175 }
1176 }
1177 if (i >= ARRAY_SIZE (cpu_arch))
1178 as_bad (_("no such architecture: `%s'"), string);
1179
1180 *input_line_pointer = e;
1181 }
1182 else
1183 as_bad (_("missing cpu architecture"));
1184
1185 no_cond_jump_promotion = 0;
1186 if (*input_line_pointer == ','
1187 && !is_end_of_line[(unsigned char) input_line_pointer[1]])
1188 {
1189 char *string = ++input_line_pointer;
1190 int e = get_symbol_end ();
1191
1192 if (strcmp (string, "nojumps") == 0)
1193 no_cond_jump_promotion = 1;
1194 else if (strcmp (string, "jumps") == 0)
1195 ;
1196 else
1197 as_bad (_("no such architecture modifier: `%s'"), string);
1198
1199 *input_line_pointer = e;
1200 }
1201
1202 demand_empty_rest_of_line ();
1203 }
1204
1205 unsigned long
1206 i386_mach ()
1207 {
1208 if (!strcmp (default_arch, "x86_64"))
1209 return bfd_mach_x86_64;
1210 else if (!strcmp (default_arch, "i386"))
1211 return bfd_mach_i386_i386;
1212 else
1213 as_fatal (_("Unknown architecture"));
1214 }
1215 \f
1216 void
1217 md_begin ()
1218 {
1219 const char *hash_err;
1220
1221 /* Initialize op_hash hash table. */
1222 op_hash = hash_new ();
1223
1224 {
1225 const template *optab;
1226 templates *core_optab;
1227
1228 /* Setup for loop. */
1229 optab = i386_optab;
1230 core_optab = (templates *) xmalloc (sizeof (templates));
1231 core_optab->start = optab;
1232
1233 while (1)
1234 {
1235 ++optab;
1236 if (optab->name == NULL
1237 || strcmp (optab->name, (optab - 1)->name) != 0)
1238 {
1239 /* different name --> ship out current template list;
1240 add to hash table; & begin anew. */
1241 core_optab->end = optab;
1242 hash_err = hash_insert (op_hash,
1243 (optab - 1)->name,
1244 (PTR) core_optab);
1245 if (hash_err)
1246 {
1247 as_fatal (_("Internal Error: Can't hash %s: %s"),
1248 (optab - 1)->name,
1249 hash_err);
1250 }
1251 if (optab->name == NULL)
1252 break;
1253 core_optab = (templates *) xmalloc (sizeof (templates));
1254 core_optab->start = optab;
1255 }
1256 }
1257 }
1258
1259 /* Initialize reg_hash hash table. */
1260 reg_hash = hash_new ();
1261 {
1262 const reg_entry *regtab;
1263 unsigned int regtab_size = i386_regtab_size;
1264
1265 for (regtab = i386_regtab; regtab_size--; regtab++)
1266 {
1267 hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
1268 if (hash_err)
1269 as_fatal (_("Internal Error: Can't hash %s: %s"),
1270 regtab->reg_name,
1271 hash_err);
1272 }
1273 }
1274
1275 /* Fill in lexical tables: mnemonic_chars, operand_chars. */
1276 {
1277 int c;
1278 char *p;
1279
1280 for (c = 0; c < 256; c++)
1281 {
1282 if (ISDIGIT (c))
1283 {
1284 digit_chars[c] = c;
1285 mnemonic_chars[c] = c;
1286 register_chars[c] = c;
1287 operand_chars[c] = c;
1288 }
1289 else if (ISLOWER (c))
1290 {
1291 mnemonic_chars[c] = c;
1292 register_chars[c] = c;
1293 operand_chars[c] = c;
1294 }
1295 else if (ISUPPER (c))
1296 {
1297 mnemonic_chars[c] = TOLOWER (c);
1298 register_chars[c] = mnemonic_chars[c];
1299 operand_chars[c] = c;
1300 }
1301
1302 if (ISALPHA (c) || ISDIGIT (c))
1303 identifier_chars[c] = c;
1304 else if (c >= 128)
1305 {
1306 identifier_chars[c] = c;
1307 operand_chars[c] = c;
1308 }
1309 }
1310
1311 #ifdef LEX_AT
1312 identifier_chars['@'] = '@';
1313 #endif
1314 #ifdef LEX_QM
1315 identifier_chars['?'] = '?';
1316 operand_chars['?'] = '?';
1317 #endif
1318 digit_chars['-'] = '-';
1319 mnemonic_chars['-'] = '-';
1320 mnemonic_chars['.'] = '.';
1321 identifier_chars['_'] = '_';
1322 identifier_chars['.'] = '.';
1323
1324 for (p = operand_special_chars; *p != '\0'; p++)
1325 operand_chars[(unsigned char) *p] = *p;
1326 }
1327
1328 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1329 if (IS_ELF)
1330 {
1331 record_alignment (text_section, 2);
1332 record_alignment (data_section, 2);
1333 record_alignment (bss_section, 2);
1334 }
1335 #endif
1336
1337 if (flag_code == CODE_64BIT)
1338 {
1339 x86_dwarf2_return_column = 16;
1340 x86_cie_data_alignment = -8;
1341 }
1342 else
1343 {
1344 x86_dwarf2_return_column = 8;
1345 x86_cie_data_alignment = -4;
1346 }
1347 }
1348
1349 void
1350 i386_print_statistics (FILE *file)
1351 {
1352 hash_print_statistics (file, "i386 opcode", op_hash);
1353 hash_print_statistics (file, "i386 register", reg_hash);
1354 }
1355 \f
1356 #ifdef DEBUG386
1357
1358 /* Debugging routines for md_assemble. */
1359 static void pte (template *);
1360 static void pt (unsigned int);
1361 static void pe (expressionS *);
1362 static void ps (symbolS *);
1363
1364 static void
1365 pi (char *line, i386_insn *x)
1366 {
1367 unsigned int i;
1368
1369 fprintf (stdout, "%s: template ", line);
1370 pte (&x->tm);
1371 fprintf (stdout, " address: base %s index %s scale %x\n",
1372 x->base_reg ? x->base_reg->reg_name : "none",
1373 x->index_reg ? x->index_reg->reg_name : "none",
1374 x->log2_scale_factor);
1375 fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
1376 x->rm.mode, x->rm.reg, x->rm.regmem);
1377 fprintf (stdout, " sib: base %x index %x scale %x\n",
1378 x->sib.base, x->sib.index, x->sib.scale);
1379 fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
1380 (x->rex & REX_W) != 0,
1381 (x->rex & REX_R) != 0,
1382 (x->rex & REX_X) != 0,
1383 (x->rex & REX_B) != 0);
1384 for (i = 0; i < x->operands; i++)
1385 {
1386 fprintf (stdout, " #%d: ", i + 1);
1387 pt (x->types[i]);
1388 fprintf (stdout, "\n");
1389 if (x->types[i]
1390 & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
1391 fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
1392 if (x->types[i] & Imm)
1393 pe (x->op[i].imms);
1394 if (x->types[i] & Disp)
1395 pe (x->op[i].disps);
1396 }
1397 }
1398
1399 static void
1400 pte (template *t)
1401 {
1402 unsigned int i;
1403 fprintf (stdout, " %d operands ", t->operands);
1404 fprintf (stdout, "opcode %x ", t->base_opcode);
1405 if (t->extension_opcode != None)
1406 fprintf (stdout, "ext %x ", t->extension_opcode);
1407 if (t->opcode_modifier & D)
1408 fprintf (stdout, "D");
1409 if (t->opcode_modifier & W)
1410 fprintf (stdout, "W");
1411 fprintf (stdout, "\n");
1412 for (i = 0; i < t->operands; i++)
1413 {
1414 fprintf (stdout, " #%d type ", i + 1);
1415 pt (t->operand_types[i]);
1416 fprintf (stdout, "\n");
1417 }
1418 }
1419
1420 static void
1421 pe (expressionS *e)
1422 {
1423 fprintf (stdout, " operation %d\n", e->X_op);
1424 fprintf (stdout, " add_number %ld (%lx)\n",
1425 (long) e->X_add_number, (long) e->X_add_number);
1426 if (e->X_add_symbol)
1427 {
1428 fprintf (stdout, " add_symbol ");
1429 ps (e->X_add_symbol);
1430 fprintf (stdout, "\n");
1431 }
1432 if (e->X_op_symbol)
1433 {
1434 fprintf (stdout, " op_symbol ");
1435 ps (e->X_op_symbol);
1436 fprintf (stdout, "\n");
1437 }
1438 }
1439
1440 static void
1441 ps (symbolS *s)
1442 {
1443 fprintf (stdout, "%s type %s%s",
1444 S_GET_NAME (s),
1445 S_IS_EXTERNAL (s) ? "EXTERNAL " : "",
1446 segment_name (S_GET_SEGMENT (s)));
1447 }
1448
1449 static struct type_name
1450 {
1451 unsigned int mask;
1452 char *tname;
1453 }
1454 const type_names[] =
1455 {
1456 { Reg8, "r8" },
1457 { Reg16, "r16" },
1458 { Reg32, "r32" },
1459 { Reg64, "r64" },
1460 { Imm8, "i8" },
1461 { Imm8S, "i8s" },
1462 { Imm16, "i16" },
1463 { Imm32, "i32" },
1464 { Imm32S, "i32s" },
1465 { Imm64, "i64" },
1466 { Imm1, "i1" },
1467 { BaseIndex, "BaseIndex" },
1468 { Disp8, "d8" },
1469 { Disp16, "d16" },
1470 { Disp32, "d32" },
1471 { Disp32S, "d32s" },
1472 { Disp64, "d64" },
1473 { InOutPortReg, "InOutPortReg" },
1474 { ShiftCount, "ShiftCount" },
1475 { Control, "control reg" },
1476 { Test, "test reg" },
1477 { Debug, "debug reg" },
1478 { FloatReg, "FReg" },
1479 { FloatAcc, "FAcc" },
1480 { SReg2, "SReg2" },
1481 { SReg3, "SReg3" },
1482 { Acc, "Acc" },
1483 { JumpAbsolute, "Jump Absolute" },
1484 { RegMMX, "rMMX" },
1485 { RegXMM, "rXMM" },
1486 { EsSeg, "es" },
1487 { 0, "" }
1488 };
1489
1490 static void
1491 pt (t)
1492 unsigned int t;
1493 {
1494 const struct type_name *ty;
1495
1496 for (ty = type_names; ty->mask; ty++)
1497 if (t & ty->mask)
1498 fprintf (stdout, "%s, ", ty->tname);
1499 fflush (stdout);
1500 }
1501
1502 #endif /* DEBUG386 */
1503 \f
1504 static bfd_reloc_code_real_type
1505 reloc (unsigned int size,
1506 int pcrel,
1507 int sign,
1508 bfd_reloc_code_real_type other)
1509 {
1510 if (other != NO_RELOC)
1511 {
1512 reloc_howto_type *reloc;
1513
1514 if (size == 8)
1515 switch (other)
1516 {
1517 case BFD_RELOC_X86_64_GOT32:
1518 return BFD_RELOC_X86_64_GOT64;
1519 break;
1520 case BFD_RELOC_X86_64_PLTOFF64:
1521 return BFD_RELOC_X86_64_PLTOFF64;
1522 break;
1523 case BFD_RELOC_X86_64_GOTPC32:
1524 other = BFD_RELOC_X86_64_GOTPC64;
1525 break;
1526 case BFD_RELOC_X86_64_GOTPCREL:
1527 other = BFD_RELOC_X86_64_GOTPCREL64;
1528 break;
1529 case BFD_RELOC_X86_64_TPOFF32:
1530 other = BFD_RELOC_X86_64_TPOFF64;
1531 break;
1532 case BFD_RELOC_X86_64_DTPOFF32:
1533 other = BFD_RELOC_X86_64_DTPOFF64;
1534 break;
1535 default:
1536 break;
1537 }
1538
1539 /* Sign-checking 4-byte relocations in 16-/32-bit code is pointless. */
1540 if (size == 4 && flag_code != CODE_64BIT)
1541 sign = -1;
1542
1543 reloc = bfd_reloc_type_lookup (stdoutput, other);
1544 if (!reloc)
1545 as_bad (_("unknown relocation (%u)"), other);
1546 else if (size != bfd_get_reloc_size (reloc))
1547 as_bad (_("%u-byte relocation cannot be applied to %u-byte field"),
1548 bfd_get_reloc_size (reloc),
1549 size);
1550 else if (pcrel && !reloc->pc_relative)
1551 as_bad (_("non-pc-relative relocation for pc-relative field"));
1552 else if ((reloc->complain_on_overflow == complain_overflow_signed
1553 && !sign)
1554 || (reloc->complain_on_overflow == complain_overflow_unsigned
1555 && sign > 0))
1556 as_bad (_("relocated field and relocation type differ in signedness"));
1557 else
1558 return other;
1559 return NO_RELOC;
1560 }
1561
1562 if (pcrel)
1563 {
1564 if (!sign)
1565 as_bad (_("there are no unsigned pc-relative relocations"));
1566 switch (size)
1567 {
1568 case 1: return BFD_RELOC_8_PCREL;
1569 case 2: return BFD_RELOC_16_PCREL;
1570 case 4: return BFD_RELOC_32_PCREL;
1571 case 8: return BFD_RELOC_64_PCREL;
1572 }
1573 as_bad (_("cannot do %u byte pc-relative relocation"), size);
1574 }
1575 else
1576 {
1577 if (sign > 0)
1578 switch (size)
1579 {
1580 case 4: return BFD_RELOC_X86_64_32S;
1581 }
1582 else
1583 switch (size)
1584 {
1585 case 1: return BFD_RELOC_8;
1586 case 2: return BFD_RELOC_16;
1587 case 4: return BFD_RELOC_32;
1588 case 8: return BFD_RELOC_64;
1589 }
1590 as_bad (_("cannot do %s %u byte relocation"),
1591 sign > 0 ? "signed" : "unsigned", size);
1592 }
1593
1594 abort ();
1595 return BFD_RELOC_NONE;
1596 }
1597
1598 /* Here we decide which fixups can be adjusted to make them relative to
1599 the beginning of the section instead of the symbol. Basically we need
1600 to make sure that the dynamic relocations are done correctly, so in
1601 some cases we force the original symbol to be used. */
1602
1603 int
1604 tc_i386_fix_adjustable (fixS *fixP ATTRIBUTE_UNUSED)
1605 {
1606 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
1607 if (!IS_ELF)
1608 return 1;
1609
1610 /* Don't adjust pc-relative references to merge sections in 64-bit
1611 mode. */
1612 if (use_rela_relocations
1613 && (S_GET_SEGMENT (fixP->fx_addsy)->flags & SEC_MERGE) != 0
1614 && fixP->fx_pcrel)
1615 return 0;
1616
1617 /* The x86_64 GOTPCREL are represented as 32bit PCrel relocations
1618 and changed later by validate_fix. */
1619 if (GOT_symbol && fixP->fx_subsy == GOT_symbol
1620 && fixP->fx_r_type == BFD_RELOC_32_PCREL)
1621 return 0;
1622
1623 /* adjust_reloc_syms doesn't know about the GOT. */
1624 if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
1625 || fixP->fx_r_type == BFD_RELOC_386_PLT32
1626 || fixP->fx_r_type == BFD_RELOC_386_GOT32
1627 || fixP->fx_r_type == BFD_RELOC_386_TLS_GD
1628 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDM
1629 || fixP->fx_r_type == BFD_RELOC_386_TLS_LDO_32
1630 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE_32
1631 || fixP->fx_r_type == BFD_RELOC_386_TLS_IE
1632 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTIE
1633 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE_32
1634 || fixP->fx_r_type == BFD_RELOC_386_TLS_LE
1635 || fixP->fx_r_type == BFD_RELOC_386_TLS_GOTDESC
1636 || fixP->fx_r_type == BFD_RELOC_386_TLS_DESC_CALL
1637 || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
1638 || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
1639 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
1640 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSGD
1641 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSLD
1642 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF32
1643 || fixP->fx_r_type == BFD_RELOC_X86_64_DTPOFF64
1644 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTTPOFF
1645 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF32
1646 || fixP->fx_r_type == BFD_RELOC_X86_64_TPOFF64
1647 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTOFF64
1648 || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPC32_TLSDESC
1649 || fixP->fx_r_type == BFD_RELOC_X86_64_TLSDESC_CALL
1650 || fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
1651 || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
1652 return 0;
1653 #endif
1654 return 1;
1655 }
1656
1657 static int
1658 intel_float_operand (const char *mnemonic)
1659 {
1660 /* Note that the value returned is meaningful only for opcodes with (memory)
1661 operands, hence the code here is free to improperly handle opcodes that
1662 have no operands (for better performance and smaller code). */
1663
1664 if (mnemonic[0] != 'f')
1665 return 0; /* non-math */
1666
1667 switch (mnemonic[1])
1668 {
1669 /* fclex, fdecstp, fdisi, femms, feni, fincstp, finit, fsetpm, and
1670 the fs segment override prefix not currently handled because no
1671 call path can make opcodes without operands get here */
1672 case 'i':
1673 return 2 /* integer op */;
1674 case 'l':
1675 if (mnemonic[2] == 'd' && (mnemonic[3] == 'c' || mnemonic[3] == 'e'))
1676 return 3; /* fldcw/fldenv */
1677 break;
1678 case 'n':
1679 if (mnemonic[2] != 'o' /* fnop */)
1680 return 3; /* non-waiting control op */
1681 break;
1682 case 'r':
1683 if (mnemonic[2] == 's')
1684 return 3; /* frstor/frstpm */
1685 break;
1686 case 's':
1687 if (mnemonic[2] == 'a')
1688 return 3; /* fsave */
1689 if (mnemonic[2] == 't')
1690 {
1691 switch (mnemonic[3])
1692 {
1693 case 'c': /* fstcw */
1694 case 'd': /* fstdw */
1695 case 'e': /* fstenv */
1696 case 's': /* fsts[gw] */
1697 return 3;
1698 }
1699 }
1700 break;
1701 case 'x':
1702 if (mnemonic[2] == 'r' || mnemonic[2] == 's')
1703 return 0; /* fxsave/fxrstor are not really math ops */
1704 break;
1705 }
1706
1707 return 1;
1708 }
1709
1710 /* This is the guts of the machine-dependent assembler. LINE points to a
1711 machine dependent instruction. This function is supposed to emit
1712 the frags/bytes it assembles to. */
1713
1714 void
1715 md_assemble (line)
1716 char *line;
1717 {
1718 int j;
1719 char mnemonic[MAX_MNEM_SIZE];
1720
1721 /* Initialize globals. */
1722 memset (&i, '\0', sizeof (i));
1723 for (j = 0; j < MAX_OPERANDS; j++)
1724 i.reloc[j] = NO_RELOC;
1725 memset (disp_expressions, '\0', sizeof (disp_expressions));
1726 memset (im_expressions, '\0', sizeof (im_expressions));
1727 save_stack_p = save_stack;
1728
1729 /* First parse an instruction mnemonic & call i386_operand for the operands.
1730 We assume that the scrubber has arranged it so that line[0] is the valid
1731 start of a (possibly prefixed) mnemonic. */
1732
1733 line = parse_insn (line, mnemonic);
1734 if (line == NULL)
1735 return;
1736
1737 line = parse_operands (line, mnemonic);
1738 if (line == NULL)
1739 return;
1740
1741 /* The order of the immediates should be reversed
1742 for 2 immediates extrq and insertq instructions */
1743 if ((i.imm_operands == 2)
1744 && ((strcmp (mnemonic, "extrq") == 0)
1745 || (strcmp (mnemonic, "insertq") == 0)))
1746 {
1747 swap_2_operands (0, 1);
1748 /* "extrq" and insertq" are the only two instructions whose operands
1749 have to be reversed even though they have two immediate operands.
1750 */
1751 if (intel_syntax)
1752 swap_operands ();
1753 }
1754
1755 /* Now we've parsed the mnemonic into a set of templates, and have the
1756 operands at hand. */
1757
1758 /* All intel opcodes have reversed operands except for "bound" and
1759 "enter". We also don't reverse intersegment "jmp" and "call"
1760 instructions with 2 immediate operands so that the immediate segment
1761 precedes the offset, as it does when in AT&T mode. */
1762 if (intel_syntax
1763 && i.operands > 1
1764 && (strcmp (mnemonic, "bound") != 0)
1765 && (strcmp (mnemonic, "invlpga") != 0)
1766 && !((i.types[0] & Imm) && (i.types[1] & Imm)))
1767 swap_operands ();
1768
1769 if (i.imm_operands)
1770 optimize_imm ();
1771
1772 /* Don't optimize displacement for movabs since it only takes 64bit
1773 displacement. */
1774 if (i.disp_operands
1775 && (flag_code != CODE_64BIT
1776 || strcmp (mnemonic, "movabs") != 0))
1777 optimize_disp ();
1778
1779 /* Next, we find a template that matches the given insn,
1780 making sure the overlap of the given operands types is consistent
1781 with the template operand types. */
1782
1783 if (!match_template ())
1784 return;
1785
1786 if (intel_syntax)
1787 {
1788 /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
1789 if (SYSV386_COMPAT
1790 && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
1791 i.tm.base_opcode ^= Opcode_FloatR;
1792
1793 /* Zap movzx and movsx suffix. The suffix may have been set from
1794 "word ptr" or "byte ptr" on the source operand, but we'll use
1795 the suffix later to choose the destination register. */
1796 if ((i.tm.base_opcode & ~9) == 0x0fb6)
1797 {
1798 if (i.reg_operands < 2
1799 && !i.suffix
1800 && (~i.tm.opcode_modifier
1801 & (No_bSuf
1802 | No_wSuf
1803 | No_lSuf
1804 | No_sSuf
1805 | No_xSuf
1806 | No_qSuf)))
1807 as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
1808
1809 i.suffix = 0;
1810 }
1811 }
1812
1813 if (i.tm.opcode_modifier & FWait)
1814 if (!add_prefix (FWAIT_OPCODE))
1815 return;
1816
1817 /* Check string instruction segment overrides. */
1818 if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
1819 {
1820 if (!check_string ())
1821 return;
1822 }
1823
1824 if (!process_suffix ())
1825 return;
1826
1827 /* Make still unresolved immediate matches conform to size of immediate
1828 given in i.suffix. */
1829 if (!finalize_imm ())
1830 return;
1831
1832 if (i.types[0] & Imm1)
1833 i.imm_operands = 0; /* kludge for shift insns. */
1834 if (i.types[0] & ImplicitRegister)
1835 i.reg_operands--;
1836 if (i.types[1] & ImplicitRegister)
1837 i.reg_operands--;
1838 if (i.types[2] & ImplicitRegister)
1839 i.reg_operands--;
1840
1841 if (i.tm.opcode_modifier & ImmExt)
1842 {
1843 expressionS *exp;
1844
1845 if ((i.tm.cpu_flags & CpuSSE3) && i.operands > 0)
1846 {
1847 /* Streaming SIMD extensions 3 Instructions have the fixed
1848 operands with an opcode suffix which is coded in the same
1849 place as an 8-bit immediate field would be. Here we check
1850 those operands and remove them afterwards. */
1851 unsigned int x;
1852
1853 for (x = 0; x < i.operands; x++)
1854 if (i.op[x].regs->reg_num != x)
1855 as_bad (_("can't use register '%s%s' as operand %d in '%s'."),
1856 register_prefix,
1857 i.op[x].regs->reg_name,
1858 x + 1,
1859 i.tm.name);
1860 i.operands = 0;
1861 }
1862
1863 /* These AMD 3DNow! and Intel Katmai New Instructions have an
1864 opcode suffix which is coded in the same place as an 8-bit
1865 immediate field would be. Here we fake an 8-bit immediate
1866 operand from the opcode suffix stored in tm.extension_opcode. */
1867
1868 assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
1869
1870 exp = &im_expressions[i.imm_operands++];
1871 i.op[i.operands].imms = exp;
1872 i.types[i.operands++] = Imm8;
1873 exp->X_op = O_constant;
1874 exp->X_add_number = i.tm.extension_opcode;
1875 i.tm.extension_opcode = None;
1876 }
1877
1878 /* For insns with operands there are more diddles to do to the opcode. */
1879 if (i.operands)
1880 {
1881 if (!process_operands ())
1882 return;
1883 }
1884 else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
1885 {
1886 /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
1887 as_warn (_("translating to `%sp'"), i.tm.name);
1888 }
1889
1890 /* Handle conversion of 'int $3' --> special int3 insn. */
1891 if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
1892 {
1893 i.tm.base_opcode = INT3_OPCODE;
1894 i.imm_operands = 0;
1895 }
1896
1897 if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
1898 && i.op[0].disps->X_op == O_constant)
1899 {
1900 /* Convert "jmp constant" (and "call constant") to a jump (call) to
1901 the absolute address given by the constant. Since ix86 jumps and
1902 calls are pc relative, we need to generate a reloc. */
1903 i.op[0].disps->X_add_symbol = &abs_symbol;
1904 i.op[0].disps->X_op = O_symbol;
1905 }
1906
1907 if ((i.tm.opcode_modifier & Rex64) != 0)
1908 i.rex |= REX_W;
1909
1910 /* For 8 bit registers we need an empty rex prefix. Also if the
1911 instruction already has a prefix, we need to convert old
1912 registers to new ones. */
1913
1914 if (((i.types[0] & Reg8) != 0
1915 && (i.op[0].regs->reg_flags & RegRex64) != 0)
1916 || ((i.types[1] & Reg8) != 0
1917 && (i.op[1].regs->reg_flags & RegRex64) != 0)
1918 || (((i.types[0] & Reg8) != 0 || (i.types[1] & Reg8) != 0)
1919 && i.rex != 0))
1920 {
1921 int x;
1922
1923 i.rex |= REX_OPCODE;
1924 for (x = 0; x < 2; x++)
1925 {
1926 /* Look for 8 bit operand that uses old registers. */
1927 if ((i.types[x] & Reg8) != 0
1928 && (i.op[x].regs->reg_flags & RegRex64) == 0)
1929 {
1930 /* In case it is "hi" register, give up. */
1931 if (i.op[x].regs->reg_num > 3)
1932 as_bad (_("can't encode register '%s%s' in an "
1933 "instruction requiring REX prefix."),
1934 register_prefix, i.op[x].regs->reg_name);
1935
1936 /* Otherwise it is equivalent to the extended register.
1937 Since the encoding doesn't change this is merely
1938 cosmetic cleanup for debug output. */
1939
1940 i.op[x].regs = i.op[x].regs + 8;
1941 }
1942 }
1943 }
1944
1945 if (i.rex != 0)
1946 add_prefix (REX_OPCODE | i.rex);
1947
1948 /* We are ready to output the insn. */
1949 output_insn ();
1950 }
1951
1952 static char *
1953 parse_insn (char *line, char *mnemonic)
1954 {
1955 char *l = line;
1956 char *token_start = l;
1957 char *mnem_p;
1958 int supported;
1959 const template *t;
1960
1961 /* Non-zero if we found a prefix only acceptable with string insns. */
1962 const char *expecting_string_instruction = NULL;
1963
1964 while (1)
1965 {
1966 mnem_p = mnemonic;
1967 while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
1968 {
1969 mnem_p++;
1970 if (mnem_p >= mnemonic + MAX_MNEM_SIZE)
1971 {
1972 as_bad (_("no such instruction: `%s'"), token_start);
1973 return NULL;
1974 }
1975 l++;
1976 }
1977 if (!is_space_char (*l)
1978 && *l != END_OF_INSN
1979 && (intel_syntax
1980 || (*l != PREFIX_SEPARATOR
1981 && *l != ',')))
1982 {
1983 as_bad (_("invalid character %s in mnemonic"),
1984 output_invalid (*l));
1985 return NULL;
1986 }
1987 if (token_start == l)
1988 {
1989 if (!intel_syntax && *l == PREFIX_SEPARATOR)
1990 as_bad (_("expecting prefix; got nothing"));
1991 else
1992 as_bad (_("expecting mnemonic; got nothing"));
1993 return NULL;
1994 }
1995
1996 /* Look up instruction (or prefix) via hash table. */
1997 current_templates = hash_find (op_hash, mnemonic);
1998
1999 if (*l != END_OF_INSN
2000 && (!is_space_char (*l) || l[1] != END_OF_INSN)
2001 && current_templates
2002 && (current_templates->start->opcode_modifier & IsPrefix))
2003 {
2004 if (current_templates->start->cpu_flags
2005 & (flag_code != CODE_64BIT ? Cpu64 : CpuNo64))
2006 {
2007 as_bad ((flag_code != CODE_64BIT
2008 ? _("`%s' is only supported in 64-bit mode")
2009 : _("`%s' is not supported in 64-bit mode")),
2010 current_templates->start->name);
2011 return NULL;
2012 }
2013 /* If we are in 16-bit mode, do not allow addr16 or data16.
2014 Similarly, in 32-bit mode, do not allow addr32 or data32. */
2015 if ((current_templates->start->opcode_modifier & (Size16 | Size32))
2016 && flag_code != CODE_64BIT
2017 && (((current_templates->start->opcode_modifier & Size32) != 0)
2018 ^ (flag_code == CODE_16BIT)))
2019 {
2020 as_bad (_("redundant %s prefix"),
2021 current_templates->start->name);
2022 return NULL;
2023 }
2024 /* Add prefix, checking for repeated prefixes. */
2025 switch (add_prefix (current_templates->start->base_opcode))
2026 {
2027 case 0:
2028 return NULL;
2029 case 2:
2030 expecting_string_instruction = current_templates->start->name;
2031 break;
2032 }
2033 /* Skip past PREFIX_SEPARATOR and reset token_start. */
2034 token_start = ++l;
2035 }
2036 else
2037 break;
2038 }
2039
2040 if (!current_templates)
2041 {
2042 /* See if we can get a match by trimming off a suffix. */
2043 switch (mnem_p[-1])
2044 {
2045 case WORD_MNEM_SUFFIX:
2046 if (intel_syntax && (intel_float_operand (mnemonic) & 2))
2047 i.suffix = SHORT_MNEM_SUFFIX;
2048 else
2049 case BYTE_MNEM_SUFFIX:
2050 case QWORD_MNEM_SUFFIX:
2051 i.suffix = mnem_p[-1];
2052 mnem_p[-1] = '\0';
2053 current_templates = hash_find (op_hash, mnemonic);
2054 break;
2055 case SHORT_MNEM_SUFFIX:
2056 case LONG_MNEM_SUFFIX:
2057 if (!intel_syntax)
2058 {
2059 i.suffix = mnem_p[-1];
2060 mnem_p[-1] = '\0';
2061 current_templates = hash_find (op_hash, mnemonic);
2062 }
2063 break;
2064
2065 /* Intel Syntax. */
2066 case 'd':
2067 if (intel_syntax)
2068 {
2069 if (intel_float_operand (mnemonic) == 1)
2070 i.suffix = SHORT_MNEM_SUFFIX;
2071 else
2072 i.suffix = LONG_MNEM_SUFFIX;
2073 mnem_p[-1] = '\0';
2074 current_templates = hash_find (op_hash, mnemonic);
2075 }
2076 break;
2077 }
2078 if (!current_templates)
2079 {
2080 as_bad (_("no such instruction: `%s'"), token_start);
2081 return NULL;
2082 }
2083 }
2084
2085 if (current_templates->start->opcode_modifier & (Jump | JumpByte))
2086 {
2087 /* Check for a branch hint. We allow ",pt" and ",pn" for
2088 predict taken and predict not taken respectively.
2089 I'm not sure that branch hints actually do anything on loop
2090 and jcxz insns (JumpByte) for current Pentium4 chips. They
2091 may work in the future and it doesn't hurt to accept them
2092 now. */
2093 if (l[0] == ',' && l[1] == 'p')
2094 {
2095 if (l[2] == 't')
2096 {
2097 if (!add_prefix (DS_PREFIX_OPCODE))
2098 return NULL;
2099 l += 3;
2100 }
2101 else if (l[2] == 'n')
2102 {
2103 if (!add_prefix (CS_PREFIX_OPCODE))
2104 return NULL;
2105 l += 3;
2106 }
2107 }
2108 }
2109 /* Any other comma loses. */
2110 if (*l == ',')
2111 {
2112 as_bad (_("invalid character %s in mnemonic"),
2113 output_invalid (*l));
2114 return NULL;
2115 }
2116
2117 /* Check if instruction is supported on specified architecture. */
2118 supported = 0;
2119 for (t = current_templates->start; t < current_templates->end; ++t)
2120 {
2121 if (!((t->cpu_flags & ~(Cpu64 | CpuNo64))
2122 & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64))))
2123 supported |= 1;
2124 if (!(t->cpu_flags & (flag_code == CODE_64BIT ? CpuNo64 : Cpu64)))
2125 supported |= 2;
2126 }
2127 if (!(supported & 2))
2128 {
2129 as_bad (flag_code == CODE_64BIT
2130 ? _("`%s' is not supported in 64-bit mode")
2131 : _("`%s' is only supported in 64-bit mode"),
2132 current_templates->start->name);
2133 return NULL;
2134 }
2135 if (!(supported & 1))
2136 {
2137 as_warn (_("`%s' is not supported on `%s%s'"),
2138 current_templates->start->name,
2139 cpu_arch_name,
2140 cpu_sub_arch_name ? cpu_sub_arch_name : "");
2141 }
2142 else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
2143 {
2144 as_warn (_("use .code16 to ensure correct addressing mode"));
2145 }
2146
2147 /* Check for rep/repne without a string instruction. */
2148 if (expecting_string_instruction)
2149 {
2150 static templates override;
2151
2152 for (t = current_templates->start; t < current_templates->end; ++t)
2153 if (t->opcode_modifier & IsString)
2154 break;
2155 if (t >= current_templates->end)
2156 {
2157 as_bad (_("expecting string instruction after `%s'"),
2158 expecting_string_instruction);
2159 return NULL;
2160 }
2161 for (override.start = t; t < current_templates->end; ++t)
2162 if (!(t->opcode_modifier & IsString))
2163 break;
2164 override.end = t;
2165 current_templates = &override;
2166 }
2167
2168 return l;
2169 }
2170
2171 static char *
2172 parse_operands (char *l, const char *mnemonic)
2173 {
2174 char *token_start;
2175
2176 /* 1 if operand is pending after ','. */
2177 unsigned int expecting_operand = 0;
2178
2179 /* Non-zero if operand parens not balanced. */
2180 unsigned int paren_not_balanced;
2181
2182 while (*l != END_OF_INSN)
2183 {
2184 /* Skip optional white space before operand. */
2185 if (is_space_char (*l))
2186 ++l;
2187 if (!is_operand_char (*l) && *l != END_OF_INSN)
2188 {
2189 as_bad (_("invalid character %s before operand %d"),
2190 output_invalid (*l),
2191 i.operands + 1);
2192 return NULL;
2193 }
2194 token_start = l; /* after white space */
2195 paren_not_balanced = 0;
2196 while (paren_not_balanced || *l != ',')
2197 {
2198 if (*l == END_OF_INSN)
2199 {
2200 if (paren_not_balanced)
2201 {
2202 if (!intel_syntax)
2203 as_bad (_("unbalanced parenthesis in operand %d."),
2204 i.operands + 1);
2205 else
2206 as_bad (_("unbalanced brackets in operand %d."),
2207 i.operands + 1);
2208 return NULL;
2209 }
2210 else
2211 break; /* we are done */
2212 }
2213 else if (!is_operand_char (*l) && !is_space_char (*l))
2214 {
2215 as_bad (_("invalid character %s in operand %d"),
2216 output_invalid (*l),
2217 i.operands + 1);
2218 return NULL;
2219 }
2220 if (!intel_syntax)
2221 {
2222 if (*l == '(')
2223 ++paren_not_balanced;
2224 if (*l == ')')
2225 --paren_not_balanced;
2226 }
2227 else
2228 {
2229 if (*l == '[')
2230 ++paren_not_balanced;
2231 if (*l == ']')
2232 --paren_not_balanced;
2233 }
2234 l++;
2235 }
2236 if (l != token_start)
2237 { /* Yes, we've read in another operand. */
2238 unsigned int operand_ok;
2239 this_operand = i.operands++;
2240 if (i.operands > MAX_OPERANDS)
2241 {
2242 as_bad (_("spurious operands; (%d operands/instruction max)"),
2243 MAX_OPERANDS);
2244 return NULL;
2245 }
2246 /* Now parse operand adding info to 'i' as we go along. */
2247 END_STRING_AND_SAVE (l);
2248
2249 if (intel_syntax)
2250 operand_ok =
2251 i386_intel_operand (token_start,
2252 intel_float_operand (mnemonic));
2253 else
2254 operand_ok = i386_operand (token_start);
2255
2256 RESTORE_END_STRING (l);
2257 if (!operand_ok)
2258 return NULL;
2259 }
2260 else
2261 {
2262 if (expecting_operand)
2263 {
2264 expecting_operand_after_comma:
2265 as_bad (_("expecting operand after ','; got nothing"));
2266 return NULL;
2267 }
2268 if (*l == ',')
2269 {
2270 as_bad (_("expecting operand before ','; got nothing"));
2271 return NULL;
2272 }
2273 }
2274
2275 /* Now *l must be either ',' or END_OF_INSN. */
2276 if (*l == ',')
2277 {
2278 if (*++l == END_OF_INSN)
2279 {
2280 /* Just skip it, if it's \n complain. */
2281 goto expecting_operand_after_comma;
2282 }
2283 expecting_operand = 1;
2284 }
2285 }
2286 return l;
2287 }
2288
2289 static void
2290 swap_2_operands (int xchg1, int xchg2)
2291 {
2292 union i386_op temp_op;
2293 unsigned int temp_type;
2294 enum bfd_reloc_code_real temp_reloc;
2295
2296 temp_type = i.types[xchg2];
2297 i.types[xchg2] = i.types[xchg1];
2298 i.types[xchg1] = temp_type;
2299 temp_op = i.op[xchg2];
2300 i.op[xchg2] = i.op[xchg1];
2301 i.op[xchg1] = temp_op;
2302 temp_reloc = i.reloc[xchg2];
2303 i.reloc[xchg2] = i.reloc[xchg1];
2304 i.reloc[xchg1] = temp_reloc;
2305 }
2306
2307 static void
2308 swap_operands (void)
2309 {
2310 switch (i.operands)
2311 {
2312 case 4:
2313 swap_2_operands (1, i.operands - 2);
2314 case 3:
2315 case 2:
2316 swap_2_operands (0, i.operands - 1);
2317 break;
2318 default:
2319 abort ();
2320 }
2321
2322 if (i.mem_operands == 2)
2323 {
2324 const seg_entry *temp_seg;
2325 temp_seg = i.seg[0];
2326 i.seg[0] = i.seg[1];
2327 i.seg[1] = temp_seg;
2328 }
2329 }
2330
2331 /* Try to ensure constant immediates are represented in the smallest
2332 opcode possible. */
2333 static void
2334 optimize_imm (void)
2335 {
2336 char guess_suffix = 0;
2337 int op;
2338
2339 if (i.suffix)
2340 guess_suffix = i.suffix;
2341 else if (i.reg_operands)
2342 {
2343 /* Figure out a suffix from the last register operand specified.
2344 We can't do this properly yet, ie. excluding InOutPortReg,
2345 but the following works for instructions with immediates.
2346 In any case, we can't set i.suffix yet. */
2347 for (op = i.operands; --op >= 0;)
2348 if (i.types[op] & Reg)
2349 {
2350 if (i.types[op] & Reg8)
2351 guess_suffix = BYTE_MNEM_SUFFIX;
2352 else if (i.types[op] & Reg16)
2353 guess_suffix = WORD_MNEM_SUFFIX;
2354 else if (i.types[op] & Reg32)
2355 guess_suffix = LONG_MNEM_SUFFIX;
2356 else if (i.types[op] & Reg64)
2357 guess_suffix = QWORD_MNEM_SUFFIX;
2358 break;
2359 }
2360 }
2361 else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
2362 guess_suffix = WORD_MNEM_SUFFIX;
2363
2364 for (op = i.operands; --op >= 0;)
2365 if (i.types[op] & Imm)
2366 {
2367 switch (i.op[op].imms->X_op)
2368 {
2369 case O_constant:
2370 /* If a suffix is given, this operand may be shortened. */
2371 switch (guess_suffix)
2372 {
2373 case LONG_MNEM_SUFFIX:
2374 i.types[op] |= Imm32 | Imm64;
2375 break;
2376 case WORD_MNEM_SUFFIX:
2377 i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
2378 break;
2379 case BYTE_MNEM_SUFFIX:
2380 i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
2381 break;
2382 }
2383
2384 /* If this operand is at most 16 bits, convert it
2385 to a signed 16 bit number before trying to see
2386 whether it will fit in an even smaller size.
2387 This allows a 16-bit operand such as $0xffe0 to
2388 be recognised as within Imm8S range. */
2389 if ((i.types[op] & Imm16)
2390 && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0)
2391 {
2392 i.op[op].imms->X_add_number =
2393 (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
2394 }
2395 if ((i.types[op] & Imm32)
2396 && ((i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1))
2397 == 0))
2398 {
2399 i.op[op].imms->X_add_number = ((i.op[op].imms->X_add_number
2400 ^ ((offsetT) 1 << 31))
2401 - ((offsetT) 1 << 31));
2402 }
2403 i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
2404
2405 /* We must avoid matching of Imm32 templates when 64bit
2406 only immediate is available. */
2407 if (guess_suffix == QWORD_MNEM_SUFFIX)
2408 i.types[op] &= ~Imm32;
2409 break;
2410
2411 case O_absent:
2412 case O_register:
2413 abort ();
2414
2415 /* Symbols and expressions. */
2416 default:
2417 /* Convert symbolic operand to proper sizes for matching, but don't
2418 prevent matching a set of insns that only supports sizes other
2419 than those matching the insn suffix. */
2420 {
2421 unsigned int mask, allowed = 0;
2422 const template *t;
2423
2424 for (t = current_templates->start;
2425 t < current_templates->end;
2426 ++t)
2427 allowed |= t->operand_types[op];
2428 switch (guess_suffix)
2429 {
2430 case QWORD_MNEM_SUFFIX:
2431 mask = Imm64 | Imm32S;
2432 break;
2433 case LONG_MNEM_SUFFIX:
2434 mask = Imm32;
2435 break;
2436 case WORD_MNEM_SUFFIX:
2437 mask = Imm16;
2438 break;
2439 case BYTE_MNEM_SUFFIX:
2440 mask = Imm8;
2441 break;
2442 default:
2443 mask = 0;
2444 break;
2445 }
2446 if (mask & allowed)
2447 i.types[op] &= mask;
2448 }
2449 break;
2450 }
2451 }
2452 }
2453
2454 /* Try to use the smallest displacement type too. */
2455 static void
2456 optimize_disp (void)
2457 {
2458 int op;
2459
2460 for (op = i.operands; --op >= 0;)
2461 if (i.types[op] & Disp)
2462 {
2463 if (i.op[op].disps->X_op == O_constant)
2464 {
2465 offsetT disp = i.op[op].disps->X_add_number;
2466
2467 if ((i.types[op] & Disp16)
2468 && (disp & ~(offsetT) 0xffff) == 0)
2469 {
2470 /* If this operand is at most 16 bits, convert
2471 to a signed 16 bit number and don't use 64bit
2472 displacement. */
2473 disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
2474 i.types[op] &= ~Disp64;
2475 }
2476 if ((i.types[op] & Disp32)
2477 && (disp & ~(((offsetT) 2 << 31) - 1)) == 0)
2478 {
2479 /* If this operand is at most 32 bits, convert
2480 to a signed 32 bit number and don't use 64bit
2481 displacement. */
2482 disp &= (((offsetT) 2 << 31) - 1);
2483 disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
2484 i.types[op] &= ~Disp64;
2485 }
2486 if (!disp && (i.types[op] & BaseIndex))
2487 {
2488 i.types[op] &= ~Disp;
2489 i.op[op].disps = 0;
2490 i.disp_operands--;
2491 }
2492 else if (flag_code == CODE_64BIT)
2493 {
2494 if (fits_in_signed_long (disp))
2495 {
2496 i.types[op] &= ~Disp64;
2497 i.types[op] |= Disp32S;
2498 }
2499 if (fits_in_unsigned_long (disp))
2500 i.types[op] |= Disp32;
2501 }
2502 if ((i.types[op] & (Disp32 | Disp32S | Disp16))
2503 && fits_in_signed_byte (disp))
2504 i.types[op] |= Disp8;
2505 }
2506 else if (i.reloc[op] == BFD_RELOC_386_TLS_DESC_CALL
2507 || i.reloc[op] == BFD_RELOC_X86_64_TLSDESC_CALL)
2508 {
2509 fix_new_exp (frag_now, frag_more (0) - frag_now->fr_literal, 0,
2510 i.op[op].disps, 0, i.reloc[op]);
2511 i.types[op] &= ~Disp;
2512 }
2513 else
2514 /* We only support 64bit displacement on constants. */
2515 i.types[op] &= ~Disp64;
2516 }
2517 }
2518
2519 static int
2520 match_template (void)
2521 {
2522 /* Points to template once we've found it. */
2523 const template *t;
2524 unsigned int overlap0, overlap1, overlap2, overlap3;
2525 unsigned int found_reverse_match;
2526 int suffix_check;
2527 unsigned int operand_types [MAX_OPERANDS];
2528 int addr_prefix_disp;
2529 unsigned int j;
2530
2531 #if MAX_OPERANDS != 4
2532 # error "MAX_OPERANDS must be 4."
2533 #endif
2534
2535 #define MATCH(overlap, given, template) \
2536 ((overlap & ~JumpAbsolute) \
2537 && (((given) & (BaseIndex | JumpAbsolute)) \
2538 == ((overlap) & (BaseIndex | JumpAbsolute))))
2539
2540 /* If given types r0 and r1 are registers they must be of the same type
2541 unless the expected operand type register overlap is null.
2542 Note that Acc in a template matches every size of reg. */
2543 #define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
2544 (((g0) & Reg) == 0 || ((g1) & Reg) == 0 \
2545 || ((g0) & Reg) == ((g1) & Reg) \
2546 || ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
2547
2548 overlap0 = 0;
2549 overlap1 = 0;
2550 overlap2 = 0;
2551 overlap3 = 0;
2552 found_reverse_match = 0;
2553 for (j = 0; j < MAX_OPERANDS; j++)
2554 operand_types [j] = 0;
2555 addr_prefix_disp = -1;
2556 suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
2557 ? No_bSuf
2558 : (i.suffix == WORD_MNEM_SUFFIX
2559 ? No_wSuf
2560 : (i.suffix == SHORT_MNEM_SUFFIX
2561 ? No_sSuf
2562 : (i.suffix == LONG_MNEM_SUFFIX
2563 ? No_lSuf
2564 : (i.suffix == QWORD_MNEM_SUFFIX
2565 ? No_qSuf
2566 : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX
2567 ? No_xSuf : 0))))));
2568
2569 for (t = current_templates->start; t < current_templates->end; t++)
2570 {
2571 addr_prefix_disp = -1;
2572
2573 /* Must have right number of operands. */
2574 if (i.operands != t->operands)
2575 continue;
2576
2577 /* Check the suffix, except for some instructions in intel mode. */
2578 if ((t->opcode_modifier & suffix_check)
2579 && !(intel_syntax
2580 && (t->opcode_modifier & IgnoreSize)))
2581 continue;
2582
2583 for (j = 0; j < MAX_OPERANDS; j++)
2584 operand_types [j] = t->operand_types [j];
2585
2586 /* In general, don't allow 64-bit operands in 32-bit mode. */
2587 if (i.suffix == QWORD_MNEM_SUFFIX
2588 && flag_code != CODE_64BIT
2589 && (intel_syntax
2590 ? (!(t->opcode_modifier & IgnoreSize)
2591 && !intel_float_operand (t->name))
2592 : intel_float_operand (t->name) != 2)
2593 && (!(operand_types[0] & (RegMMX | RegXMM))
2594 || !(operand_types[t->operands > 1] & (RegMMX | RegXMM)))
2595 && (t->base_opcode != 0x0fc7
2596 || t->extension_opcode != 1 /* cmpxchg8b */))
2597 continue;
2598
2599 /* Do not verify operands when there are none. */
2600 else if (!t->operands)
2601 {
2602 if (t->cpu_flags & ~cpu_arch_flags)
2603 continue;
2604 /* We've found a match; break out of loop. */
2605 break;
2606 }
2607
2608 /* Address size prefix will turn Disp64/Disp32/Disp16 operand
2609 into Disp32/Disp16/Disp32 operand. */
2610 if (i.prefix[ADDR_PREFIX] != 0)
2611 {
2612 unsigned int DispOn = 0, DispOff = 0;
2613
2614 switch (flag_code)
2615 {
2616 case CODE_16BIT:
2617 DispOn = Disp32;
2618 DispOff = Disp16;
2619 break;
2620 case CODE_32BIT:
2621 DispOn = Disp16;
2622 DispOff = Disp32;
2623 break;
2624 case CODE_64BIT:
2625 DispOn = Disp32;
2626 DispOff = Disp64;
2627 break;
2628 }
2629
2630 for (j = 0; j < MAX_OPERANDS; j++)
2631 {
2632 /* There should be only one Disp operand. */
2633 if ((operand_types[j] & DispOff))
2634 {
2635 addr_prefix_disp = j;
2636 operand_types[j] |= DispOn;
2637 operand_types[j] &= ~DispOff;
2638 break;
2639 }
2640 }
2641 }
2642
2643 overlap0 = i.types[0] & operand_types[0];
2644 switch (t->operands)
2645 {
2646 case 1:
2647 if (!MATCH (overlap0, i.types[0], operand_types[0]))
2648 continue;
2649 break;
2650 case 2:
2651 /* xchg %eax, %eax is a special case. It is an aliase for nop
2652 only in 32bit mode and we can use opcode 0x90. In 64bit
2653 mode, we can't use 0x90 for xchg %eax, %eax since it should
2654 zero-extend %eax to %rax. */
2655 if (flag_code == CODE_64BIT
2656 && t->base_opcode == 0x90
2657 && i.types [0] == (Acc | Reg32)
2658 && i.types [1] == (Acc | Reg32))
2659 continue;
2660 case 3:
2661 case 4:
2662 overlap1 = i.types[1] & operand_types[1];
2663 if (!MATCH (overlap0, i.types[0], operand_types[0])
2664 || !MATCH (overlap1, i.types[1], operand_types[1])
2665 /* monitor in SSE3 is a very special case. The first
2666 register and the second register may have different
2667 sizes. The same applies to crc32 in SSE4.2. */
2668 || !((t->base_opcode == 0x0f01
2669 && t->extension_opcode == 0xc8)
2670 || t->base_opcode == 0xf20f38f1
2671 || CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2672 operand_types[0],
2673 overlap1, i.types[1],
2674 operand_types[1])))
2675 {
2676 /* Check if other direction is valid ... */
2677 if ((t->opcode_modifier & (D | FloatD)) == 0)
2678 continue;
2679
2680 /* Try reversing direction of operands. */
2681 overlap0 = i.types[0] & operand_types[1];
2682 overlap1 = i.types[1] & operand_types[0];
2683 if (!MATCH (overlap0, i.types[0], operand_types[1])
2684 || !MATCH (overlap1, i.types[1], operand_types[0])
2685 || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
2686 operand_types[1],
2687 overlap1, i.types[1],
2688 operand_types[0]))
2689 {
2690 /* Does not match either direction. */
2691 continue;
2692 }
2693 /* found_reverse_match holds which of D or FloatDR
2694 we've found. */
2695 if ((t->opcode_modifier & D))
2696 found_reverse_match = Opcode_D;
2697 else if ((t->opcode_modifier & FloatD))
2698 found_reverse_match = Opcode_FloatD;
2699 else
2700 found_reverse_match = 0;
2701 if ((t->opcode_modifier & FloatR))
2702 found_reverse_match |= Opcode_FloatR;
2703 }
2704 else
2705 {
2706 /* Found a forward 2 operand match here. */
2707 switch (t->operands)
2708 {
2709 case 4:
2710 overlap3 = i.types[3] & operand_types[3];
2711 case 3:
2712 overlap2 = i.types[2] & operand_types[2];
2713 break;
2714 }
2715
2716 switch (t->operands)
2717 {
2718 case 4:
2719 if (!MATCH (overlap3, i.types[3], operand_types[3])
2720 || !CONSISTENT_REGISTER_MATCH (overlap2,
2721 i.types[2],
2722 operand_types[2],
2723 overlap3,
2724 i.types[3],
2725 operand_types[3]))
2726 continue;
2727 case 3:
2728 /* Here we make use of the fact that there are no
2729 reverse match 3 operand instructions, and all 3
2730 operand instructions only need to be checked for
2731 register consistency between operands 2 and 3. */
2732 if (!MATCH (overlap2, i.types[2], operand_types[2])
2733 || !CONSISTENT_REGISTER_MATCH (overlap1,
2734 i.types[1],
2735 operand_types[1],
2736 overlap2,
2737 i.types[2],
2738 operand_types[2]))
2739 continue;
2740 break;
2741 }
2742 }
2743 /* Found either forward/reverse 2, 3 or 4 operand match here:
2744 slip through to break. */
2745 }
2746 if (t->cpu_flags & ~cpu_arch_flags)
2747 {
2748 found_reverse_match = 0;
2749 continue;
2750 }
2751 /* We've found a match; break out of loop. */
2752 break;
2753 }
2754
2755 if (t == current_templates->end)
2756 {
2757 /* We found no match. */
2758 as_bad (_("suffix or operands invalid for `%s'"),
2759 current_templates->start->name);
2760 return 0;
2761 }
2762
2763 if (!quiet_warnings)
2764 {
2765 if (!intel_syntax
2766 && ((i.types[0] & JumpAbsolute)
2767 != (operand_types[0] & JumpAbsolute)))
2768 {
2769 as_warn (_("indirect %s without `*'"), t->name);
2770 }
2771
2772 if ((t->opcode_modifier & (IsPrefix | IgnoreSize))
2773 == (IsPrefix | IgnoreSize))
2774 {
2775 /* Warn them that a data or address size prefix doesn't
2776 affect assembly of the next line of code. */
2777 as_warn (_("stand-alone `%s' prefix"), t->name);
2778 }
2779 }
2780
2781 /* Copy the template we found. */
2782 i.tm = *t;
2783
2784 if (addr_prefix_disp != -1)
2785 i.tm.operand_types[addr_prefix_disp]
2786 = operand_types[addr_prefix_disp];
2787
2788 if (found_reverse_match)
2789 {
2790 /* If we found a reverse match we must alter the opcode
2791 direction bit. found_reverse_match holds bits to change
2792 (different for int & float insns). */
2793
2794 i.tm.base_opcode ^= found_reverse_match;
2795
2796 i.tm.operand_types[0] = operand_types[1];
2797 i.tm.operand_types[1] = operand_types[0];
2798 }
2799
2800 return 1;
2801 }
2802
2803 static int
2804 check_string (void)
2805 {
2806 int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
2807 if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
2808 {
2809 if (i.seg[0] != NULL && i.seg[0] != &es)
2810 {
2811 as_bad (_("`%s' operand %d must use `%%es' segment"),
2812 i.tm.name,
2813 mem_op + 1);
2814 return 0;
2815 }
2816 /* There's only ever one segment override allowed per instruction.
2817 This instruction possibly has a legal segment override on the
2818 second operand, so copy the segment to where non-string
2819 instructions store it, allowing common code. */
2820 i.seg[0] = i.seg[1];
2821 }
2822 else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
2823 {
2824 if (i.seg[1] != NULL && i.seg[1] != &es)
2825 {
2826 as_bad (_("`%s' operand %d must use `%%es' segment"),
2827 i.tm.name,
2828 mem_op + 2);
2829 return 0;
2830 }
2831 }
2832 return 1;
2833 }
2834
2835 static int
2836 process_suffix (void)
2837 {
2838 /* If matched instruction specifies an explicit instruction mnemonic
2839 suffix, use it. */
2840 if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
2841 {
2842 if (i.tm.opcode_modifier & Size16)
2843 i.suffix = WORD_MNEM_SUFFIX;
2844 else if (i.tm.opcode_modifier & Size64)
2845 i.suffix = QWORD_MNEM_SUFFIX;
2846 else
2847 i.suffix = LONG_MNEM_SUFFIX;
2848 }
2849 else if (i.reg_operands)
2850 {
2851 /* If there's no instruction mnemonic suffix we try to invent one
2852 based on register operands. */
2853 if (!i.suffix)
2854 {
2855 /* We take i.suffix from the last register operand specified,
2856 Destination register type is more significant than source
2857 register type. crc32 in SSE4.2 prefers source register
2858 type. */
2859 if (i.tm.base_opcode == 0xf20f38f1)
2860 {
2861 if ((i.types[0] & Reg))
2862 i.suffix = ((i.types[0] & Reg16) ? WORD_MNEM_SUFFIX :
2863 LONG_MNEM_SUFFIX);
2864 }
2865 else if (i.tm.base_opcode == 0xf20f38f0)
2866 {
2867 if ((i.types[0] & Reg8))
2868 i.suffix = BYTE_MNEM_SUFFIX;
2869 }
2870
2871 if (!i.suffix)
2872 {
2873 int op;
2874
2875 if (i.tm.base_opcode == 0xf20f38f1
2876 || i.tm.base_opcode == 0xf20f38f0)
2877 {
2878 /* We have to know the operand size for crc32. */
2879 as_bad (_("ambiguous memory operand size for `%s`"),
2880 i.tm.name);
2881 return 0;
2882 }
2883
2884 for (op = i.operands; --op >= 0;)
2885 if ((i.types[op] & Reg)
2886 && !(i.tm.operand_types[op] & InOutPortReg))
2887 {
2888 i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
2889 (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
2890 (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
2891 LONG_MNEM_SUFFIX);
2892 break;
2893 }
2894 }
2895 }
2896 else if (i.suffix == BYTE_MNEM_SUFFIX)
2897 {
2898 if (!check_byte_reg ())
2899 return 0;
2900 }
2901 else if (i.suffix == LONG_MNEM_SUFFIX)
2902 {
2903 if (!check_long_reg ())
2904 return 0;
2905 }
2906 else if (i.suffix == QWORD_MNEM_SUFFIX)
2907 {
2908 if (!check_qword_reg ())
2909 return 0;
2910 }
2911 else if (i.suffix == WORD_MNEM_SUFFIX)
2912 {
2913 if (!check_word_reg ())
2914 return 0;
2915 }
2916 else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
2917 /* Do nothing if the instruction is going to ignore the prefix. */
2918 ;
2919 else
2920 abort ();
2921 }
2922 else if ((i.tm.opcode_modifier & DefaultSize)
2923 && !i.suffix
2924 /* exclude fldenv/frstor/fsave/fstenv */
2925 && (i.tm.opcode_modifier & No_sSuf))
2926 {
2927 i.suffix = stackop_size;
2928 }
2929 else if (intel_syntax
2930 && !i.suffix
2931 && ((i.tm.operand_types[0] & JumpAbsolute)
2932 || (i.tm.opcode_modifier & (JumpByte|JumpInterSegment))
2933 || (i.tm.base_opcode == 0x0f01 /* [ls][gi]dt */
2934 && i.tm.extension_opcode <= 3)))
2935 {
2936 switch (flag_code)
2937 {
2938 case CODE_64BIT:
2939 if (!(i.tm.opcode_modifier & No_qSuf))
2940 {
2941 i.suffix = QWORD_MNEM_SUFFIX;
2942 break;
2943 }
2944 case CODE_32BIT:
2945 if (!(i.tm.opcode_modifier & No_lSuf))
2946 i.suffix = LONG_MNEM_SUFFIX;
2947 break;
2948 case CODE_16BIT:
2949 if (!(i.tm.opcode_modifier & No_wSuf))
2950 i.suffix = WORD_MNEM_SUFFIX;
2951 break;
2952 }
2953 }
2954
2955 if (!i.suffix)
2956 {
2957 if (!intel_syntax)
2958 {
2959 if (i.tm.opcode_modifier & W)
2960 {
2961 as_bad (_("no instruction mnemonic suffix given and "
2962 "no register operands; can't size instruction"));
2963 return 0;
2964 }
2965 }
2966 else
2967 {
2968 unsigned int suffixes = (~i.tm.opcode_modifier
2969 & (No_bSuf
2970 | No_wSuf
2971 | No_lSuf
2972 | No_sSuf
2973 | No_xSuf
2974 | No_qSuf));
2975
2976 if ((i.tm.opcode_modifier & W)
2977 || ((suffixes & (suffixes - 1))
2978 && !(i.tm.opcode_modifier & (DefaultSize | IgnoreSize))))
2979 {
2980 as_bad (_("ambiguous operand size for `%s'"), i.tm.name);
2981 return 0;
2982 }
2983 }
2984 }
2985
2986 /* Change the opcode based on the operand size given by i.suffix;
2987 We don't need to change things for byte insns. */
2988
2989 if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
2990 {
2991 /* It's not a byte, select word/dword operation. */
2992 if (i.tm.opcode_modifier & W)
2993 {
2994 if (i.tm.opcode_modifier & ShortForm)
2995 i.tm.base_opcode |= 8;
2996 else
2997 i.tm.base_opcode |= 1;
2998 }
2999
3000 /* Now select between word & dword operations via the operand
3001 size prefix, except for instructions that will ignore this
3002 prefix anyway. */
3003 if (i.tm.base_opcode == 0x0f01 && i.tm.extension_opcode == 0xc8)
3004 {
3005 /* monitor in SSE3 is a very special case. The default size
3006 of AX is the size of mode. The address size override
3007 prefix will change the size of AX. */
3008 if (i.op->regs[0].reg_type &
3009 (flag_code == CODE_32BIT ? Reg16 : Reg32))
3010 if (!add_prefix (ADDR_PREFIX_OPCODE))
3011 return 0;
3012 }
3013 else if (i.suffix != QWORD_MNEM_SUFFIX
3014 && i.suffix != LONG_DOUBLE_MNEM_SUFFIX
3015 && !(i.tm.opcode_modifier & (IgnoreSize | FloatMF))
3016 && ((i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
3017 || (flag_code == CODE_64BIT
3018 && (i.tm.opcode_modifier & JumpByte))))
3019 {
3020 unsigned int prefix = DATA_PREFIX_OPCODE;
3021
3022 if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
3023 prefix = ADDR_PREFIX_OPCODE;
3024
3025 if (!add_prefix (prefix))
3026 return 0;
3027 }
3028
3029 /* Set mode64 for an operand. */
3030 if (i.suffix == QWORD_MNEM_SUFFIX
3031 && flag_code == CODE_64BIT
3032 && (i.tm.opcode_modifier & NoRex64) == 0)
3033 {
3034 /* Special case for xchg %rax,%rax. It is NOP and doesn't
3035 need rex64. */
3036 if (i.operands != 2
3037 || i.types [0] != (Acc | Reg64)
3038 || i.types [1] != (Acc | Reg64)
3039 || i.tm.base_opcode != 0x90)
3040 i.rex |= REX_W;
3041 }
3042
3043 /* Size floating point instruction. */
3044 if (i.suffix == LONG_MNEM_SUFFIX)
3045 if (i.tm.opcode_modifier & FloatMF)
3046 i.tm.base_opcode ^= 4;
3047 }
3048
3049 return 1;
3050 }
3051
3052 static int
3053 check_byte_reg (void)
3054 {
3055 int op;
3056
3057 for (op = i.operands; --op >= 0;)
3058 {
3059 /* If this is an eight bit register, it's OK. If it's the 16 or
3060 32 bit version of an eight bit register, we will just use the
3061 low portion, and that's OK too. */
3062 if (i.types[op] & Reg8)
3063 continue;
3064
3065 /* movzx and movsx should not generate this warning. */
3066 if (intel_syntax
3067 && (i.tm.base_opcode == 0xfb7
3068 || i.tm.base_opcode == 0xfb6
3069 || i.tm.base_opcode == 0x63
3070 || i.tm.base_opcode == 0xfbe
3071 || i.tm.base_opcode == 0xfbf))
3072 continue;
3073
3074 /* crc32 doesn't generate this warning. */
3075 if (i.tm.base_opcode == 0xf20f38f0)
3076 continue;
3077
3078 if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4)
3079 {
3080 /* Prohibit these changes in the 64bit mode, since the
3081 lowering is more complicated. */
3082 if (flag_code == CODE_64BIT
3083 && (i.tm.operand_types[op] & InOutPortReg) == 0)
3084 {
3085 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3086 register_prefix, i.op[op].regs->reg_name,
3087 i.suffix);
3088 return 0;
3089 }
3090 #if REGISTER_WARNINGS
3091 if (!quiet_warnings
3092 && (i.tm.operand_types[op] & InOutPortReg) == 0)
3093 as_warn (_("using `%s%s' instead of `%s%s' due to `%c' suffix"),
3094 register_prefix,
3095 (i.op[op].regs + (i.types[op] & Reg16
3096 ? REGNAM_AL - REGNAM_AX
3097 : REGNAM_AL - REGNAM_EAX))->reg_name,
3098 register_prefix,
3099 i.op[op].regs->reg_name,
3100 i.suffix);
3101 #endif
3102 continue;
3103 }
3104 /* Any other register is bad. */
3105 if (i.types[op] & (Reg | RegMMX | RegXMM
3106 | SReg2 | SReg3
3107 | Control | Debug | Test
3108 | FloatReg | FloatAcc))
3109 {
3110 as_bad (_("`%s%s' not allowed with `%s%c'"),
3111 register_prefix,
3112 i.op[op].regs->reg_name,
3113 i.tm.name,
3114 i.suffix);
3115 return 0;
3116 }
3117 }
3118 return 1;
3119 }
3120
3121 static int
3122 check_long_reg (void)
3123 {
3124 int op;
3125
3126 for (op = i.operands; --op >= 0;)
3127 /* Reject eight bit registers, except where the template requires
3128 them. (eg. movzb) */
3129 if ((i.types[op] & Reg8) != 0
3130 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3131 {
3132 as_bad (_("`%s%s' not allowed with `%s%c'"),
3133 register_prefix,
3134 i.op[op].regs->reg_name,
3135 i.tm.name,
3136 i.suffix);
3137 return 0;
3138 }
3139 /* Warn if the e prefix on a general reg is missing. */
3140 else if ((!quiet_warnings || flag_code == CODE_64BIT)
3141 && (i.types[op] & Reg16) != 0
3142 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3143 {
3144 /* Prohibit these changes in the 64bit mode, since the
3145 lowering is more complicated. */
3146 if (flag_code == CODE_64BIT)
3147 {
3148 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3149 register_prefix, i.op[op].regs->reg_name,
3150 i.suffix);
3151 return 0;
3152 }
3153 #if REGISTER_WARNINGS
3154 else
3155 as_warn (_("using `%s%s' instead of `%s%s' due to `%c' suffix"),
3156 register_prefix,
3157 (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
3158 register_prefix,
3159 i.op[op].regs->reg_name,
3160 i.suffix);
3161 #endif
3162 }
3163 /* Warn if the r prefix on a general reg is missing. */
3164 else if ((i.types[op] & Reg64) != 0
3165 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3166 {
3167 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3168 register_prefix, i.op[op].regs->reg_name,
3169 i.suffix);
3170 return 0;
3171 }
3172 return 1;
3173 }
3174
3175 static int
3176 check_qword_reg (void)
3177 {
3178 int op;
3179
3180 for (op = i.operands; --op >= 0; )
3181 /* Reject eight bit registers, except where the template requires
3182 them. (eg. movzb) */
3183 if ((i.types[op] & Reg8) != 0
3184 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3185 {
3186 as_bad (_("`%s%s' not allowed with `%s%c'"),
3187 register_prefix,
3188 i.op[op].regs->reg_name,
3189 i.tm.name,
3190 i.suffix);
3191 return 0;
3192 }
3193 /* Warn if the e prefix on a general reg is missing. */
3194 else if (((i.types[op] & Reg16) != 0
3195 || (i.types[op] & Reg32) != 0)
3196 && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
3197 {
3198 /* Prohibit these changes in the 64bit mode, since the
3199 lowering is more complicated. */
3200 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3201 register_prefix, i.op[op].regs->reg_name,
3202 i.suffix);
3203 return 0;
3204 }
3205 return 1;
3206 }
3207
3208 static int
3209 check_word_reg (void)
3210 {
3211 int op;
3212 for (op = i.operands; --op >= 0;)
3213 /* Reject eight bit registers, except where the template requires
3214 them. (eg. movzb) */
3215 if ((i.types[op] & Reg8) != 0
3216 && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0)
3217 {
3218 as_bad (_("`%s%s' not allowed with `%s%c'"),
3219 register_prefix,
3220 i.op[op].regs->reg_name,
3221 i.tm.name,
3222 i.suffix);
3223 return 0;
3224 }
3225 /* Warn if the e prefix on a general reg is present. */
3226 else if ((!quiet_warnings || flag_code == CODE_64BIT)
3227 && (i.types[op] & Reg32) != 0
3228 && (i.tm.operand_types[op] & (Reg16 | Acc)) != 0)
3229 {
3230 /* Prohibit these changes in the 64bit mode, since the
3231 lowering is more complicated. */
3232 if (flag_code == CODE_64BIT)
3233 {
3234 as_bad (_("Incorrect register `%s%s' used with `%c' suffix"),
3235 register_prefix, i.op[op].regs->reg_name,
3236 i.suffix);
3237 return 0;
3238 }
3239 else
3240 #if REGISTER_WARNINGS
3241 as_warn (_("using `%s%s' instead of `%s%s' due to `%c' suffix"),
3242 register_prefix,
3243 (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
3244 register_prefix,
3245 i.op[op].regs->reg_name,
3246 i.suffix);
3247 #endif
3248 }
3249 return 1;
3250 }
3251
3252 static int
3253 finalize_imm (void)
3254 {
3255 unsigned int overlap0, overlap1, overlap2;
3256
3257 overlap0 = i.types[0] & i.tm.operand_types[0];
3258 if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64))
3259 && overlap0 != Imm8 && overlap0 != Imm8S
3260 && overlap0 != Imm16 && overlap0 != Imm32S
3261 && overlap0 != Imm32 && overlap0 != Imm64)
3262 {
3263 if (i.suffix)
3264 {
3265 overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX
3266 ? Imm8 | Imm8S
3267 : (i.suffix == WORD_MNEM_SUFFIX
3268 ? Imm16
3269 : (i.suffix == QWORD_MNEM_SUFFIX
3270 ? Imm64 | Imm32S
3271 : Imm32)));
3272 }
3273 else if (overlap0 == (Imm16 | Imm32S | Imm32)
3274 || overlap0 == (Imm16 | Imm32)
3275 || overlap0 == (Imm16 | Imm32S))
3276 {
3277 overlap0 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
3278 ? Imm16 : Imm32S);
3279 }
3280 if (overlap0 != Imm8 && overlap0 != Imm8S
3281 && overlap0 != Imm16 && overlap0 != Imm32S
3282 && overlap0 != Imm32 && overlap0 != Imm64)
3283 {
3284 as_bad (_("no instruction mnemonic suffix given; "
3285 "can't determine immediate size"));
3286 return 0;
3287 }
3288 }
3289 i.types[0] = overlap0;
3290
3291 overlap1 = i.types[1] & i.tm.operand_types[1];
3292 if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32 | Imm64))
3293 && overlap1 != Imm8 && overlap1 != Imm8S
3294 && overlap1 != Imm16 && overlap1 != Imm32S
3295 && overlap1 != Imm32 && overlap1 != Imm64)
3296 {
3297 if (i.suffix)
3298 {
3299 overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX
3300 ? Imm8 | Imm8S
3301 : (i.suffix == WORD_MNEM_SUFFIX
3302 ? Imm16
3303 : (i.suffix == QWORD_MNEM_SUFFIX
3304 ? Imm64 | Imm32S
3305 : Imm32)));
3306 }
3307 else if (overlap1 == (Imm16 | Imm32 | Imm32S)
3308 || overlap1 == (Imm16 | Imm32)
3309 || overlap1 == (Imm16 | Imm32S))
3310 {
3311 overlap1 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
3312 ? Imm16 : Imm32S);
3313 }
3314 if (overlap1 != Imm8 && overlap1 != Imm8S
3315 && overlap1 != Imm16 && overlap1 != Imm32S
3316 && overlap1 != Imm32 && overlap1 != Imm64)
3317 {
3318 as_bad (_("no instruction mnemonic suffix given; "
3319 "can't determine immediate size %x %c"),
3320 overlap1, i.suffix);
3321 return 0;
3322 }
3323 }
3324 i.types[1] = overlap1;
3325
3326 overlap2 = i.types[2] & i.tm.operand_types[2];
3327 assert ((overlap2 & Imm) == 0);
3328 i.types[2] = overlap2;
3329
3330 return 1;
3331 }
3332
3333 static int
3334 process_operands (void)
3335 {
3336 /* Default segment register this instruction will use for memory
3337 accesses. 0 means unknown. This is only for optimizing out
3338 unnecessary segment overrides. */
3339 const seg_entry *default_seg = 0;
3340
3341 /* The imul $imm, %reg instruction is converted into
3342 imul $imm, %reg, %reg, and the clr %reg instruction
3343 is converted into xor %reg, %reg. */
3344 if (i.tm.opcode_modifier & RegKludge)
3345 {
3346 if ((i.tm.cpu_flags & CpuSSE4_1))
3347 {
3348 /* The first operand in instruction blendvpd, blendvps and
3349 pblendvb in SSE4.1 is implicit and must be xmm0. */
3350 assert (i.operands == 3
3351 && i.reg_operands >= 2
3352 && i.types[0] == RegXMM);
3353 if (i.op[0].regs->reg_num != 0)
3354 {
3355 if (intel_syntax)
3356 as_bad (_("the last operand of `%s' must be `%sxmm0'"),
3357 i.tm.name, register_prefix);
3358 else
3359 as_bad (_("the first operand of `%s' must be `%sxmm0'"),
3360 i.tm.name, register_prefix);
3361 return 0;
3362 }
3363 i.op[0] = i.op[1];
3364 i.op[1] = i.op[2];
3365 i.types[0] = i.types[1];
3366 i.types[1] = i.types[2];
3367 i.operands--;
3368 i.reg_operands--;
3369
3370 /* We need to adjust fields in i.tm since they are used by
3371 build_modrm_byte. */
3372 i.tm.operand_types [0] = i.tm.operand_types [1];
3373 i.tm.operand_types [1] = i.tm.operand_types [2];
3374 i.tm.operands--;
3375 }
3376 else
3377 {
3378 unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
3379 /* Pretend we saw the extra register operand. */
3380 assert (i.reg_operands == 1
3381 && i.op[first_reg_op + 1].regs == 0);
3382 i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
3383 i.types[first_reg_op + 1] = i.types[first_reg_op];
3384 i.operands++;
3385 i.reg_operands++;
3386 }
3387 }
3388
3389 if (i.tm.opcode_modifier & ShortForm)
3390 {
3391 if (i.types[0] & (SReg2 | SReg3))
3392 {
3393 if (i.tm.base_opcode == POP_SEG_SHORT
3394 && i.op[0].regs->reg_num == 1)
3395 {
3396 as_bad (_("you can't `pop %%cs'"));
3397 return 0;
3398 }
3399 i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
3400 if ((i.op[0].regs->reg_flags & RegRex) != 0)
3401 i.rex |= REX_B;
3402 }
3403 else
3404 {
3405 /* The register or float register operand is in operand 0 or 1. */
3406 unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
3407 /* Register goes in low 3 bits of opcode. */
3408 i.tm.base_opcode |= i.op[op].regs->reg_num;
3409 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3410 i.rex |= REX_B;
3411 if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
3412 {
3413 /* Warn about some common errors, but press on regardless.
3414 The first case can be generated by gcc (<= 2.8.1). */
3415 if (i.operands == 2)
3416 {
3417 /* Reversed arguments on faddp, fsubp, etc. */
3418 as_warn (_("translating to `%s %s%s,%s%s'"), i.tm.name,
3419 register_prefix, i.op[1].regs->reg_name,
3420 register_prefix, i.op[0].regs->reg_name);
3421 }
3422 else
3423 {
3424 /* Extraneous `l' suffix on fp insn. */
3425 as_warn (_("translating to `%s %s%s'"), i.tm.name,
3426 register_prefix, i.op[0].regs->reg_name);
3427 }
3428 }
3429 }
3430 }
3431 else if (i.tm.opcode_modifier & Modrm)
3432 {
3433 /* The opcode is completed (modulo i.tm.extension_opcode which
3434 must be put into the modrm byte). Now, we make the modrm and
3435 index base bytes based on all the info we've collected. */
3436
3437 default_seg = build_modrm_byte ();
3438 }
3439 else if ((i.tm.base_opcode & ~0x3) == MOV_AX_DISP32)
3440 {
3441 default_seg = &ds;
3442 }
3443 else if ((i.tm.opcode_modifier & IsString) != 0)
3444 {
3445 /* For the string instructions that allow a segment override
3446 on one of their operands, the default segment is ds. */
3447 default_seg = &ds;
3448 }
3449
3450 if ((i.tm.base_opcode == 0x8d /* lea */
3451 || (i.tm.cpu_flags & CpuSVME))
3452 && i.seg[0] && !quiet_warnings)
3453 as_warn (_("segment override on `%s' is ineffectual"), i.tm.name);
3454
3455 /* If a segment was explicitly specified, and the specified segment
3456 is not the default, use an opcode prefix to select it. If we
3457 never figured out what the default segment is, then default_seg
3458 will be zero at this point, and the specified segment prefix will
3459 always be used. */
3460 if ((i.seg[0]) && (i.seg[0] != default_seg))
3461 {
3462 if (!add_prefix (i.seg[0]->seg_prefix))
3463 return 0;
3464 }
3465 return 1;
3466 }
3467
3468 static const seg_entry *
3469 build_modrm_byte (void)
3470 {
3471 const seg_entry *default_seg = 0;
3472
3473 /* i.reg_operands MUST be the number of real register operands;
3474 implicit registers do not count. */
3475 if (i.reg_operands == 2)
3476 {
3477 unsigned int source, dest;
3478
3479 switch (i.operands)
3480 {
3481 case 2:
3482 source = 0;
3483 break;
3484 case 3:
3485 /* When there are 3 operands, one of them may be immediate,
3486 which may be the first or the last operand. Otherwise,
3487 the first operand must be shift count register (cl). */
3488 assert (i.imm_operands == 1
3489 || (i.imm_operands == 0
3490 && (i.types[0] & ShiftCount)));
3491 source = (i.types[0] & (Imm | ShiftCount)) ? 1 : 0;
3492 break;
3493 case 4:
3494 /* When there are 4 operands, the first two must be immediate
3495 operands. The source operand will be the 3rd one. */
3496 assert (i.imm_operands == 2
3497 && (i.types[0] & Imm)
3498 && (i.types[1] & Imm));
3499 source = 2;
3500 break;
3501 default:
3502 abort ();
3503 }
3504
3505 dest = source + 1;
3506
3507 i.rm.mode = 3;
3508 /* One of the register operands will be encoded in the i.tm.reg
3509 field, the other in the combined i.tm.mode and i.tm.regmem
3510 fields. If no form of this instruction supports a memory
3511 destination operand, then we assume the source operand may
3512 sometimes be a memory operand and so we need to store the
3513 destination in the i.rm.reg field. */
3514 if ((i.tm.operand_types[dest] & (AnyMem | RegMem)) == 0)
3515 {
3516 i.rm.reg = i.op[dest].regs->reg_num;
3517 i.rm.regmem = i.op[source].regs->reg_num;
3518 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
3519 i.rex |= REX_R;
3520 if ((i.op[source].regs->reg_flags & RegRex) != 0)
3521 i.rex |= REX_B;
3522 }
3523 else
3524 {
3525 i.rm.reg = i.op[source].regs->reg_num;
3526 i.rm.regmem = i.op[dest].regs->reg_num;
3527 if ((i.op[dest].regs->reg_flags & RegRex) != 0)
3528 i.rex |= REX_B;
3529 if ((i.op[source].regs->reg_flags & RegRex) != 0)
3530 i.rex |= REX_R;
3531 }
3532 if (flag_code != CODE_64BIT && (i.rex & (REX_R | REX_B)))
3533 {
3534 if (!((i.types[0] | i.types[1]) & Control))
3535 abort ();
3536 i.rex &= ~(REX_R | REX_B);
3537 add_prefix (LOCK_PREFIX_OPCODE);
3538 }
3539 }
3540 else
3541 { /* If it's not 2 reg operands... */
3542 if (i.mem_operands)
3543 {
3544 unsigned int fake_zero_displacement = 0;
3545 unsigned int op;
3546
3547 for (op = 0; op < i.operands; op++)
3548 if ((i.types[op] & AnyMem))
3549 break;
3550 assert (op < i.operands);
3551
3552 default_seg = &ds;
3553
3554 if (i.base_reg == 0)
3555 {
3556 i.rm.mode = 0;
3557 if (!i.disp_operands)
3558 fake_zero_displacement = 1;
3559 if (i.index_reg == 0)
3560 {
3561 /* Operand is just <disp> */
3562 if (flag_code == CODE_64BIT)
3563 {
3564 /* 64bit mode overwrites the 32bit absolute
3565 addressing by RIP relative addressing and
3566 absolute addressing is encoded by one of the
3567 redundant SIB forms. */
3568 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3569 i.sib.base = NO_BASE_REGISTER;
3570 i.sib.index = NO_INDEX_REGISTER;
3571 i.types[op] = ((i.prefix[ADDR_PREFIX] == 0)
3572 ? Disp32S : Disp32);
3573 }
3574 else if ((flag_code == CODE_16BIT)
3575 ^ (i.prefix[ADDR_PREFIX] != 0))
3576 {
3577 i.rm.regmem = NO_BASE_REGISTER_16;
3578 i.types[op] = Disp16;
3579 }
3580 else
3581 {
3582 i.rm.regmem = NO_BASE_REGISTER;
3583 i.types[op] = Disp32;
3584 }
3585 }
3586 else /* !i.base_reg && i.index_reg */
3587 {
3588 i.sib.index = i.index_reg->reg_num;
3589 i.sib.base = NO_BASE_REGISTER;
3590 i.sib.scale = i.log2_scale_factor;
3591 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3592 i.types[op] &= ~Disp;
3593 if (flag_code != CODE_64BIT)
3594 i.types[op] |= Disp32; /* Must be 32 bit */
3595 else
3596 i.types[op] |= Disp32S;
3597 if ((i.index_reg->reg_flags & RegRex) != 0)
3598 i.rex |= REX_X;
3599 }
3600 }
3601 /* RIP addressing for 64bit mode. */
3602 else if (i.base_reg->reg_type == BaseIndex)
3603 {
3604 i.rm.regmem = NO_BASE_REGISTER;
3605 i.types[op] &= ~ Disp;
3606 i.types[op] |= Disp32S;
3607 i.flags[op] |= Operand_PCrel;
3608 if (! i.disp_operands)
3609 fake_zero_displacement = 1;
3610 }
3611 else if (i.base_reg->reg_type & Reg16)
3612 {
3613 switch (i.base_reg->reg_num)
3614 {
3615 case 3: /* (%bx) */
3616 if (i.index_reg == 0)
3617 i.rm.regmem = 7;
3618 else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
3619 i.rm.regmem = i.index_reg->reg_num - 6;
3620 break;
3621 case 5: /* (%bp) */
3622 default_seg = &ss;
3623 if (i.index_reg == 0)
3624 {
3625 i.rm.regmem = 6;
3626 if ((i.types[op] & Disp) == 0)
3627 {
3628 /* fake (%bp) into 0(%bp) */
3629 i.types[op] |= Disp8;
3630 fake_zero_displacement = 1;
3631 }
3632 }
3633 else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
3634 i.rm.regmem = i.index_reg->reg_num - 6 + 2;
3635 break;
3636 default: /* (%si) -> 4 or (%di) -> 5 */
3637 i.rm.regmem = i.base_reg->reg_num - 6 + 4;
3638 }
3639 i.rm.mode = mode_from_disp_size (i.types[op]);
3640 }
3641 else /* i.base_reg and 32/64 bit mode */
3642 {
3643 if (flag_code == CODE_64BIT
3644 && (i.types[op] & Disp))
3645 i.types[op] = ((i.types[op] & Disp8)
3646 | (i.prefix[ADDR_PREFIX] == 0
3647 ? Disp32S : Disp32));
3648
3649 i.rm.regmem = i.base_reg->reg_num;
3650 if ((i.base_reg->reg_flags & RegRex) != 0)
3651 i.rex |= REX_B;
3652 i.sib.base = i.base_reg->reg_num;
3653 /* x86-64 ignores REX prefix bit here to avoid decoder
3654 complications. */
3655 if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
3656 {
3657 default_seg = &ss;
3658 if (i.disp_operands == 0)
3659 {
3660 fake_zero_displacement = 1;
3661 i.types[op] |= Disp8;
3662 }
3663 }
3664 else if (i.base_reg->reg_num == ESP_REG_NUM)
3665 {
3666 default_seg = &ss;
3667 }
3668 i.sib.scale = i.log2_scale_factor;
3669 if (i.index_reg == 0)
3670 {
3671 /* <disp>(%esp) becomes two byte modrm with no index
3672 register. We've already stored the code for esp
3673 in i.rm.regmem ie. ESCAPE_TO_TWO_BYTE_ADDRESSING.
3674 Any base register besides %esp will not use the
3675 extra modrm byte. */
3676 i.sib.index = NO_INDEX_REGISTER;
3677 #if !SCALE1_WHEN_NO_INDEX
3678 /* Another case where we force the second modrm byte. */
3679 if (i.log2_scale_factor)
3680 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3681 #endif
3682 }
3683 else
3684 {
3685 i.sib.index = i.index_reg->reg_num;
3686 i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
3687 if ((i.index_reg->reg_flags & RegRex) != 0)
3688 i.rex |= REX_X;
3689 }
3690
3691 if (i.disp_operands
3692 && (i.reloc[op] == BFD_RELOC_386_TLS_DESC_CALL
3693 || i.reloc[op] == BFD_RELOC_X86_64_TLSDESC_CALL))
3694 i.rm.mode = 0;
3695 else
3696 i.rm.mode = mode_from_disp_size (i.types[op]);
3697 }
3698
3699 if (fake_zero_displacement)
3700 {
3701 /* Fakes a zero displacement assuming that i.types[op]
3702 holds the correct displacement size. */
3703 expressionS *exp;
3704
3705 assert (i.op[op].disps == 0);
3706 exp = &disp_expressions[i.disp_operands++];
3707 i.op[op].disps = exp;
3708 exp->X_op = O_constant;
3709 exp->X_add_number = 0;
3710 exp->X_add_symbol = (symbolS *) 0;
3711 exp->X_op_symbol = (symbolS *) 0;
3712 }
3713 }
3714
3715 /* Fill in i.rm.reg or i.rm.regmem field with register operand
3716 (if any) based on i.tm.extension_opcode. Again, we must be
3717 careful to make sure that segment/control/debug/test/MMX
3718 registers are coded into the i.rm.reg field. */
3719 if (i.reg_operands)
3720 {
3721 unsigned int op;
3722
3723 for (op = 0; op < i.operands; op++)
3724 if ((i.types[op] & (Reg | RegMMX | RegXMM
3725 | SReg2 | SReg3
3726 | Control | Debug | Test)))
3727 break;
3728 assert (op < i.operands);
3729
3730 /* If there is an extension opcode to put here, the register
3731 number must be put into the regmem field. */
3732 if (i.tm.extension_opcode != None)
3733 {
3734 i.rm.regmem = i.op[op].regs->reg_num;
3735 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3736 i.rex |= REX_B;
3737 }
3738 else
3739 {
3740 i.rm.reg = i.op[op].regs->reg_num;
3741 if ((i.op[op].regs->reg_flags & RegRex) != 0)
3742 i.rex |= REX_R;
3743 }
3744
3745 /* Now, if no memory operand has set i.rm.mode = 0, 1, 2 we
3746 must set it to 3 to indicate this is a register operand
3747 in the regmem field. */
3748 if (!i.mem_operands)
3749 i.rm.mode = 3;
3750 }
3751
3752 /* Fill in i.rm.reg field with extension opcode (if any). */
3753 if (i.tm.extension_opcode != None)
3754 i.rm.reg = i.tm.extension_opcode;
3755 }
3756 return default_seg;
3757 }
3758
3759 static void
3760 output_branch (void)
3761 {
3762 char *p;
3763 int code16;
3764 int prefix;
3765 relax_substateT subtype;
3766 symbolS *sym;
3767 offsetT off;
3768
3769 code16 = 0;
3770 if (flag_code == CODE_16BIT)
3771 code16 = CODE16;
3772
3773 prefix = 0;
3774 if (i.prefix[DATA_PREFIX] != 0)
3775 {
3776 prefix = 1;
3777 i.prefixes -= 1;
3778 code16 ^= CODE16;
3779 }
3780 /* Pentium4 branch hints. */
3781 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3782 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3783 {
3784 prefix++;
3785 i.prefixes--;
3786 }
3787 if (i.prefix[REX_PREFIX] != 0)
3788 {
3789 prefix++;
3790 i.prefixes--;
3791 }
3792
3793 if (i.prefixes != 0 && !intel_syntax)
3794 as_warn (_("skipping prefixes on this instruction"));
3795
3796 /* It's always a symbol; End frag & setup for relax.
3797 Make sure there is enough room in this frag for the largest
3798 instruction we may generate in md_convert_frag. This is 2
3799 bytes for the opcode and room for the prefix and largest
3800 displacement. */
3801 frag_grow (prefix + 2 + 4);
3802 /* Prefix and 1 opcode byte go in fr_fix. */
3803 p = frag_more (prefix + 1);
3804 if (i.prefix[DATA_PREFIX] != 0)
3805 *p++ = DATA_PREFIX_OPCODE;
3806 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
3807 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
3808 *p++ = i.prefix[SEG_PREFIX];
3809 if (i.prefix[REX_PREFIX] != 0)
3810 *p++ = i.prefix[REX_PREFIX];
3811 *p = i.tm.base_opcode;
3812
3813 if ((unsigned char) *p == JUMP_PC_RELATIVE)
3814 subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
3815 else if ((cpu_arch_flags & Cpu386) != 0)
3816 subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
3817 else
3818 subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
3819 subtype |= code16;
3820
3821 sym = i.op[0].disps->X_add_symbol;
3822 off = i.op[0].disps->X_add_number;
3823
3824 if (i.op[0].disps->X_op != O_constant
3825 && i.op[0].disps->X_op != O_symbol)
3826 {
3827 /* Handle complex expressions. */
3828 sym = make_expr_symbol (i.op[0].disps);
3829 off = 0;
3830 }
3831
3832 /* 1 possible extra opcode + 4 byte displacement go in var part.
3833 Pass reloc in fr_var. */
3834 frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
3835 }
3836
3837 static void
3838 output_jump (void)
3839 {
3840 char *p;
3841 int size;
3842 fixS *fixP;
3843
3844 if (i.tm.opcode_modifier & JumpByte)
3845 {
3846 /* This is a loop or jecxz type instruction. */
3847 size = 1;
3848 if (i.prefix[ADDR_PREFIX] != 0)
3849 {
3850 FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
3851 i.prefixes -= 1;
3852 }
3853 /* Pentium4 branch hints. */
3854 if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
3855 || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
3856 {
3857 FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
3858 i.prefixes--;
3859 }
3860 }
3861 else
3862 {
3863 int code16;
3864
3865 code16 = 0;
3866 if (flag_code == CODE_16BIT)
3867 code16 = CODE16;
3868
3869 if (i.prefix[DATA_PREFIX] != 0)
3870 {
3871 FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
3872 i.prefixes -= 1;
3873 code16 ^= CODE16;
3874 }
3875
3876 size = 4;
3877 if (code16)
3878 size = 2;
3879 }
3880
3881 if (i.prefix[REX_PREFIX] != 0)
3882 {
3883 FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
3884 i.prefixes -= 1;
3885 }
3886
3887 if (i.prefixes != 0 && !intel_syntax)
3888 as_warn (_("skipping prefixes on this instruction"));
3889
3890 p = frag_more (1 + size);
3891 *p++ = i.tm.base_opcode;
3892
3893 fixP = fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3894 i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
3895
3896 /* All jumps handled here are signed, but don't use a signed limit
3897 check for 32 and 16 bit jumps as we want to allow wrap around at
3898 4G and 64k respectively. */
3899 if (size == 1)
3900 fixP->fx_signed = 1;
3901 }
3902
3903 static void
3904 output_interseg_jump (void)
3905 {
3906 char *p;
3907 int size;
3908 int prefix;
3909 int code16;
3910
3911 code16 = 0;
3912 if (flag_code == CODE_16BIT)
3913 code16 = CODE16;
3914
3915 prefix = 0;
3916 if (i.prefix[DATA_PREFIX] != 0)
3917 {
3918 prefix = 1;
3919 i.prefixes -= 1;
3920 code16 ^= CODE16;
3921 }
3922 if (i.prefix[REX_PREFIX] != 0)
3923 {
3924 prefix++;
3925 i.prefixes -= 1;
3926 }
3927
3928 size = 4;
3929 if (code16)
3930 size = 2;
3931
3932 if (i.prefixes != 0 && !intel_syntax)
3933 as_warn (_("skipping prefixes on this instruction"));
3934
3935 /* 1 opcode; 2 segment; offset */
3936 p = frag_more (prefix + 1 + 2 + size);
3937
3938 if (i.prefix[DATA_PREFIX] != 0)
3939 *p++ = DATA_PREFIX_OPCODE;
3940
3941 if (i.prefix[REX_PREFIX] != 0)
3942 *p++ = i.prefix[REX_PREFIX];
3943
3944 *p++ = i.tm.base_opcode;
3945 if (i.op[1].imms->X_op == O_constant)
3946 {
3947 offsetT n = i.op[1].imms->X_add_number;
3948
3949 if (size == 2
3950 && !fits_in_unsigned_word (n)
3951 && !fits_in_signed_word (n))
3952 {
3953 as_bad (_("16-bit jump out of range"));
3954 return;
3955 }
3956 md_number_to_chars (p, n, size);
3957 }
3958 else
3959 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
3960 i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
3961 if (i.op[0].imms->X_op != O_constant)
3962 as_bad (_("can't handle non absolute segment in `%s'"),
3963 i.tm.name);
3964 md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
3965 }
3966
3967 static void
3968 output_insn (void)
3969 {
3970 fragS *insn_start_frag;
3971 offsetT insn_start_off;
3972
3973 /* Tie dwarf2 debug info to the address at the start of the insn.
3974 We can't do this after the insn has been output as the current
3975 frag may have been closed off. eg. by frag_var. */
3976 dwarf2_emit_insn (0);
3977
3978 insn_start_frag = frag_now;
3979 insn_start_off = frag_now_fix ();
3980
3981 /* Output jumps. */
3982 if (i.tm.opcode_modifier & Jump)
3983 output_branch ();
3984 else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
3985 output_jump ();
3986 else if (i.tm.opcode_modifier & JumpInterSegment)
3987 output_interseg_jump ();
3988 else
3989 {
3990 /* Output normal instructions here. */
3991 char *p;
3992 unsigned char *q;
3993 unsigned int prefix;
3994
3995 /* All opcodes on i386 have either 1 or 2 bytes. SSSE3 and
3996 SSE4 instructions have 3 bytes. We may use one more higher
3997 byte to specify a prefix the instruction requires. Exclude
3998 instructions which are in both SSE4 and ABM. */
3999 if ((i.tm.cpu_flags & (CpuSSSE3 | CpuSSE4)) != 0
4000 && (i.tm.cpu_flags & CpuABM) == 0)
4001 {
4002 if (i.tm.base_opcode & 0xff000000)
4003 {
4004 prefix = (i.tm.base_opcode >> 24) & 0xff;
4005 goto check_prefix;
4006 }
4007 }
4008 else if ((i.tm.base_opcode & 0xff0000) != 0)
4009 {
4010 prefix = (i.tm.base_opcode >> 16) & 0xff;
4011 if ((i.tm.cpu_flags & CpuPadLock) != 0)
4012 {
4013 check_prefix:
4014 if (prefix != REPE_PREFIX_OPCODE
4015 || i.prefix[LOCKREP_PREFIX] != REPE_PREFIX_OPCODE)
4016 add_prefix (prefix);
4017 }
4018 else
4019 add_prefix (prefix);
4020 }
4021
4022 /* The prefix bytes. */
4023 for (q = i.prefix;
4024 q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
4025 q++)
4026 {
4027 if (*q)
4028 {
4029 p = frag_more (1);
4030 md_number_to_chars (p, (valueT) *q, 1);
4031 }
4032 }
4033
4034 /* Now the opcode; be careful about word order here! */
4035 if (fits_in_unsigned_byte (i.tm.base_opcode))
4036 {
4037 FRAG_APPEND_1_CHAR (i.tm.base_opcode);
4038 }
4039 else
4040 {
4041 if ((i.tm.cpu_flags & (CpuSSSE3 | CpuSSE4)) != 0
4042 && (i.tm.cpu_flags & CpuABM) == 0)
4043 {
4044 p = frag_more (3);
4045 *p++ = (i.tm.base_opcode >> 16) & 0xff;
4046 }
4047 else
4048 p = frag_more (2);
4049
4050 /* Put out high byte first: can't use md_number_to_chars! */
4051 *p++ = (i.tm.base_opcode >> 8) & 0xff;
4052 *p = i.tm.base_opcode & 0xff;
4053 }
4054
4055 /* Now the modrm byte and sib byte (if present). */
4056 if (i.tm.opcode_modifier & Modrm)
4057 {
4058 p = frag_more (1);
4059 md_number_to_chars (p,
4060 (valueT) (i.rm.regmem << 0
4061 | i.rm.reg << 3
4062 | i.rm.mode << 6),
4063 1);
4064 /* If i.rm.regmem == ESP (4)
4065 && i.rm.mode != (Register mode)
4066 && not 16 bit
4067 ==> need second modrm byte. */
4068 if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
4069 && i.rm.mode != 3
4070 && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
4071 {
4072 p = frag_more (1);
4073 md_number_to_chars (p,
4074 (valueT) (i.sib.base << 0
4075 | i.sib.index << 3
4076 | i.sib.scale << 6),
4077 1);
4078 }
4079 }
4080
4081 if (i.disp_operands)
4082 output_disp (insn_start_frag, insn_start_off);
4083
4084 if (i.imm_operands)
4085 output_imm (insn_start_frag, insn_start_off);
4086 }
4087
4088 #ifdef DEBUG386
4089 if (flag_debug)
4090 {
4091 pi ("" /*line*/, &i);
4092 }
4093 #endif /* DEBUG386 */
4094 }
4095
4096 /* Return the size of the displacement operand N. */
4097
4098 static int
4099 disp_size (unsigned int n)
4100 {
4101 int size = 4;
4102 if (i.types[n] & (Disp8 | Disp16 | Disp64))
4103 {
4104 size = 2;
4105 if (i.types[n] & Disp8)
4106 size = 1;
4107 if (i.types[n] & Disp64)
4108 size = 8;
4109 }
4110 return size;
4111 }
4112
4113 /* Return the size of the immediate operand N. */
4114
4115 static int
4116 imm_size (unsigned int n)
4117 {
4118 int size = 4;
4119 if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
4120 {
4121 size = 2;
4122 if (i.types[n] & (Imm8 | Imm8S))
4123 size = 1;
4124 if (i.types[n] & Imm64)
4125 size = 8;
4126 }
4127 return size;
4128 }
4129
4130 static void
4131 output_disp (fragS *insn_start_frag, offsetT insn_start_off)
4132 {
4133 char *p;
4134 unsigned int n;
4135
4136 for (n = 0; n < i.operands; n++)
4137 {
4138 if (i.types[n] & Disp)
4139 {
4140 if (i.op[n].disps->X_op == O_constant)
4141 {
4142 int size = disp_size (n);
4143 offsetT val;
4144
4145 val = offset_in_range (i.op[n].disps->X_add_number,
4146 size);
4147 p = frag_more (size);
4148 md_number_to_chars (p, val, size);
4149 }
4150 else
4151 {
4152 enum bfd_reloc_code_real reloc_type;
4153 int size = disp_size (n);
4154 int sign = (i.types[n] & Disp32S) != 0;
4155 int pcrel = (i.flags[n] & Operand_PCrel) != 0;
4156
4157 /* We can't have 8 bit displacement here. */
4158 assert ((i.types[n] & Disp8) == 0);
4159
4160 /* The PC relative address is computed relative
4161 to the instruction boundary, so in case immediate
4162 fields follows, we need to adjust the value. */
4163 if (pcrel && i.imm_operands)
4164 {
4165 unsigned int n1;
4166 int sz = 0;
4167
4168 for (n1 = 0; n1 < i.operands; n1++)
4169 if (i.types[n1] & Imm)
4170 {
4171 /* Only one immediate is allowed for PC
4172 relative address. */
4173 assert (sz == 0);
4174 sz = imm_size (n1);
4175 i.op[n].disps->X_add_number -= sz;
4176 }
4177 /* We should find the immediate. */
4178 assert (sz != 0);
4179 }
4180
4181 p = frag_more (size);
4182 reloc_type = reloc (size, pcrel, sign, i.reloc[n]);
4183 if (GOT_symbol
4184 && GOT_symbol == i.op[n].disps->X_add_symbol
4185 && (((reloc_type == BFD_RELOC_32
4186 || reloc_type == BFD_RELOC_X86_64_32S
4187 || (reloc_type == BFD_RELOC_64
4188 && object_64bit))
4189 && (i.op[n].disps->X_op == O_symbol
4190 || (i.op[n].disps->X_op == O_add
4191 && ((symbol_get_value_expression
4192 (i.op[n].disps->X_op_symbol)->X_op)
4193 == O_subtract))))
4194 || reloc_type == BFD_RELOC_32_PCREL))
4195 {
4196 offsetT add;
4197
4198 if (insn_start_frag == frag_now)
4199 add = (p - frag_now->fr_literal) - insn_start_off;
4200 else
4201 {
4202 fragS *fr;
4203
4204 add = insn_start_frag->fr_fix - insn_start_off;
4205 for (fr = insn_start_frag->fr_next;
4206 fr && fr != frag_now; fr = fr->fr_next)
4207 add += fr->fr_fix;
4208 add += p - frag_now->fr_literal;
4209 }
4210
4211 if (!object_64bit)
4212 {
4213 reloc_type = BFD_RELOC_386_GOTPC;
4214 i.op[n].imms->X_add_number += add;
4215 }
4216 else if (reloc_type == BFD_RELOC_64)
4217 reloc_type = BFD_RELOC_X86_64_GOTPC64;
4218 else
4219 /* Don't do the adjustment for x86-64, as there
4220 the pcrel addressing is relative to the _next_
4221 insn, and that is taken care of in other code. */
4222 reloc_type = BFD_RELOC_X86_64_GOTPC32;
4223 }
4224 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
4225 i.op[n].disps, pcrel, reloc_type);
4226 }
4227 }
4228 }
4229 }
4230
4231 static void
4232 output_imm (fragS *insn_start_frag, offsetT insn_start_off)
4233 {
4234 char *p;
4235 unsigned int n;
4236
4237 for (n = 0; n < i.operands; n++)
4238 {
4239 if (i.types[n] & Imm)
4240 {
4241 if (i.op[n].imms->X_op == O_constant)
4242 {
4243 int size = imm_size (n);
4244 offsetT val;
4245
4246 val = offset_in_range (i.op[n].imms->X_add_number,
4247 size);
4248 p = frag_more (size);
4249 md_number_to_chars (p, val, size);
4250 }
4251 else
4252 {
4253 /* Not absolute_section.
4254 Need a 32-bit fixup (don't support 8bit
4255 non-absolute imms). Try to support other
4256 sizes ... */
4257 enum bfd_reloc_code_real reloc_type;
4258 int size = imm_size (n);
4259 int sign;
4260
4261 if ((i.types[n] & (Imm32S))
4262 && (i.suffix == QWORD_MNEM_SUFFIX
4263 || (!i.suffix && (i.tm.opcode_modifier & No_lSuf))))
4264 sign = 1;
4265 else
4266 sign = 0;
4267
4268 p = frag_more (size);
4269 reloc_type = reloc (size, 0, sign, i.reloc[n]);
4270
4271 /* This is tough to explain. We end up with this one if we
4272 * have operands that look like
4273 * "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal here is to
4274 * obtain the absolute address of the GOT, and it is strongly
4275 * preferable from a performance point of view to avoid using
4276 * a runtime relocation for this. The actual sequence of
4277 * instructions often look something like:
4278 *
4279 * call .L66
4280 * .L66:
4281 * popl %ebx
4282 * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
4283 *
4284 * The call and pop essentially return the absolute address
4285 * of the label .L66 and store it in %ebx. The linker itself
4286 * will ultimately change the first operand of the addl so
4287 * that %ebx points to the GOT, but to keep things simple, the
4288 * .o file must have this operand set so that it generates not
4289 * the absolute address of .L66, but the absolute address of
4290 * itself. This allows the linker itself simply treat a GOTPC
4291 * relocation as asking for a pcrel offset to the GOT to be
4292 * added in, and the addend of the relocation is stored in the
4293 * operand field for the instruction itself.
4294 *
4295 * Our job here is to fix the operand so that it would add
4296 * the correct offset so that %ebx would point to itself. The
4297 * thing that is tricky is that .-.L66 will point to the
4298 * beginning of the instruction, so we need to further modify
4299 * the operand so that it will point to itself. There are
4300 * other cases where you have something like:
4301 *
4302 * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
4303 *
4304 * and here no correction would be required. Internally in
4305 * the assembler we treat operands of this form as not being
4306 * pcrel since the '.' is explicitly mentioned, and I wonder
4307 * whether it would simplify matters to do it this way. Who
4308 * knows. In earlier versions of the PIC patches, the
4309 * pcrel_adjust field was used to store the correction, but
4310 * since the expression is not pcrel, I felt it would be
4311 * confusing to do it this way. */
4312
4313 if ((reloc_type == BFD_RELOC_32
4314 || reloc_type == BFD_RELOC_X86_64_32S
4315 || reloc_type == BFD_RELOC_64)
4316 && GOT_symbol
4317 && GOT_symbol == i.op[n].imms->X_add_symbol
4318 && (i.op[n].imms->X_op == O_symbol
4319 || (i.op[n].imms->X_op == O_add
4320 && ((symbol_get_value_expression
4321 (i.op[n].imms->X_op_symbol)->X_op)
4322 == O_subtract))))
4323 {
4324 offsetT add;
4325
4326 if (insn_start_frag == frag_now)
4327 add = (p - frag_now->fr_literal) - insn_start_off;
4328 else
4329 {
4330 fragS *fr;
4331
4332 add = insn_start_frag->fr_fix - insn_start_off;
4333 for (fr = insn_start_frag->fr_next;
4334 fr && fr != frag_now; fr = fr->fr_next)
4335 add += fr->fr_fix;
4336 add += p - frag_now->fr_literal;
4337 }
4338
4339 if (!object_64bit)
4340 reloc_type = BFD_RELOC_386_GOTPC;
4341 else if (size == 4)
4342 reloc_type = BFD_RELOC_X86_64_GOTPC32;
4343 else if (size == 8)
4344 reloc_type = BFD_RELOC_X86_64_GOTPC64;
4345 i.op[n].imms->X_add_number += add;
4346 }
4347 fix_new_exp (frag_now, p - frag_now->fr_literal, size,
4348 i.op[n].imms, 0, reloc_type);
4349 }
4350 }
4351 }
4352 }
4353 \f
4354 /* x86_cons_fix_new is called via the expression parsing code when a
4355 reloc is needed. We use this hook to get the correct .got reloc. */
4356 static enum bfd_reloc_code_real got_reloc = NO_RELOC;
4357 static int cons_sign = -1;
4358
4359 void
4360 x86_cons_fix_new (fragS *frag, unsigned int off, unsigned int len,
4361 expressionS *exp)
4362 {
4363 enum bfd_reloc_code_real r = reloc (len, 0, cons_sign, got_reloc);
4364
4365 got_reloc = NO_RELOC;
4366
4367 #ifdef TE_PE
4368 if (exp->X_op == O_secrel)
4369 {
4370 exp->X_op = O_symbol;
4371 r = BFD_RELOC_32_SECREL;
4372 }
4373 #endif
4374
4375 fix_new_exp (frag, off, len, exp, 0, r);
4376 }
4377
4378 #if (!defined (OBJ_ELF) && !defined (OBJ_MAYBE_ELF)) || defined (LEX_AT)
4379 # define lex_got(reloc, adjust, types) NULL
4380 #else
4381 /* Parse operands of the form
4382 <symbol>@GOTOFF+<nnn>
4383 and similar .plt or .got references.
4384
4385 If we find one, set up the correct relocation in RELOC and copy the
4386 input string, minus the `@GOTOFF' into a malloc'd buffer for
4387 parsing by the calling routine. Return this buffer, and if ADJUST
4388 is non-null set it to the length of the string we removed from the
4389 input line. Otherwise return NULL. */
4390 static char *
4391 lex_got (enum bfd_reloc_code_real *reloc,
4392 int *adjust,
4393 unsigned int *types)
4394 {
4395 /* Some of the relocations depend on the size of what field is to
4396 be relocated. But in our callers i386_immediate and i386_displacement
4397 we don't yet know the operand size (this will be set by insn
4398 matching). Hence we record the word32 relocation here,
4399 and adjust the reloc according to the real size in reloc(). */
4400 static const struct {
4401 const char *str;
4402 const enum bfd_reloc_code_real rel[2];
4403 const unsigned int types64;
4404 } gotrel[] = {
4405 { "PLTOFF", { 0,
4406 BFD_RELOC_X86_64_PLTOFF64 },
4407 Imm64 },
4408 { "PLT", { BFD_RELOC_386_PLT32,
4409 BFD_RELOC_X86_64_PLT32 },
4410 Imm32 | Imm32S | Disp32 },
4411 { "GOTPLT", { 0,
4412 BFD_RELOC_X86_64_GOTPLT64 },
4413 Imm64 | Disp64 },
4414 { "GOTOFF", { BFD_RELOC_386_GOTOFF,
4415 BFD_RELOC_X86_64_GOTOFF64 },
4416 Imm64 | Disp64 },
4417 { "GOTPCREL", { 0,
4418 BFD_RELOC_X86_64_GOTPCREL },
4419 Imm32 | Imm32S | Disp32 },
4420 { "TLSGD", { BFD_RELOC_386_TLS_GD,
4421 BFD_RELOC_X86_64_TLSGD },
4422 Imm32 | Imm32S | Disp32 },
4423 { "TLSLDM", { BFD_RELOC_386_TLS_LDM,
4424 0 },
4425 0 },
4426 { "TLSLD", { 0,
4427 BFD_RELOC_X86_64_TLSLD },
4428 Imm32 | Imm32S | Disp32 },
4429 { "GOTTPOFF", { BFD_RELOC_386_TLS_IE_32,
4430 BFD_RELOC_X86_64_GOTTPOFF },
4431 Imm32 | Imm32S | Disp32 },
4432 { "TPOFF", { BFD_RELOC_386_TLS_LE_32,
4433 BFD_RELOC_X86_64_TPOFF32 },
4434 Imm32 | Imm32S | Imm64 | Disp32 | Disp64 },
4435 { "NTPOFF", { BFD_RELOC_386_TLS_LE,
4436 0 },
4437 0 },
4438 { "DTPOFF", { BFD_RELOC_386_TLS_LDO_32,
4439 BFD_RELOC_X86_64_DTPOFF32 },
4440 Imm32 | Imm32S | Imm64 | Disp32 | Disp64 },
4441 { "GOTNTPOFF",{ BFD_RELOC_386_TLS_GOTIE,
4442 0 },
4443 0 },
4444 { "INDNTPOFF",{ BFD_RELOC_386_TLS_IE,
4445 0 },
4446 0 },
4447 { "GOT", { BFD_RELOC_386_GOT32,
4448 BFD_RELOC_X86_64_GOT32 },
4449 Imm32 | Imm32S | Disp32 | Imm64 },
4450 { "TLSDESC", { BFD_RELOC_386_TLS_GOTDESC,
4451 BFD_RELOC_X86_64_GOTPC32_TLSDESC },
4452 Imm32 | Imm32S | Disp32 },
4453 { "TLSCALL", { BFD_RELOC_386_TLS_DESC_CALL,
4454 BFD_RELOC_X86_64_TLSDESC_CALL },
4455 Imm32 | Imm32S | Disp32 }
4456 };
4457 char *cp;
4458 unsigned int j;
4459
4460 if (!IS_ELF)
4461 return NULL;
4462
4463 for (cp = input_line_pointer; *cp != '@'; cp++)
4464 if (is_end_of_line[(unsigned char) *cp])
4465 return NULL;
4466
4467 for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
4468 {
4469 int len;
4470
4471 len = strlen (gotrel[j].str);
4472 if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
4473 {
4474 if (gotrel[j].rel[object_64bit] != 0)
4475 {
4476 int first, second;
4477 char *tmpbuf, *past_reloc;
4478
4479 *reloc = gotrel[j].rel[object_64bit];
4480 if (adjust)
4481 *adjust = len;
4482
4483 if (types)
4484 {
4485 if (flag_code != CODE_64BIT)
4486 *types = Imm32 | Disp32;
4487 else
4488 *types = gotrel[j].types64;
4489 }
4490
4491 if (GOT_symbol == NULL)
4492 GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
4493
4494 /* The length of the first part of our input line. */
4495 first = cp - input_line_pointer;
4496
4497 /* The second part goes from after the reloc token until
4498 (and including) an end_of_line char. Don't use strlen
4499 here as the end_of_line char may not be a NUL. */
4500 past_reloc = cp + 1 + len;
4501 for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
4502 ;
4503 second = cp - past_reloc;
4504
4505 /* Allocate and copy string. The trailing NUL shouldn't
4506 be necessary, but be safe. */
4507 tmpbuf = xmalloc (first + second + 2);
4508 memcpy (tmpbuf, input_line_pointer, first);
4509 if (second != 0 && *past_reloc != ' ')
4510 /* Replace the relocation token with ' ', so that
4511 errors like foo@GOTOFF1 will be detected. */
4512 tmpbuf[first++] = ' ';
4513 memcpy (tmpbuf + first, past_reloc, second);
4514 tmpbuf[first + second] = '\0';
4515 return tmpbuf;
4516 }
4517
4518 as_bad (_("@%s reloc is not supported with %d-bit output format"),
4519 gotrel[j].str, 1 << (5 + object_64bit));
4520 return NULL;
4521 }
4522 }
4523
4524 /* Might be a symbol version string. Don't as_bad here. */
4525 return NULL;
4526 }
4527
4528 void
4529 x86_cons (expressionS *exp, int size)
4530 {
4531 if (size == 4 || (object_64bit && size == 8))
4532 {
4533 /* Handle @GOTOFF and the like in an expression. */
4534 char *save;
4535 char *gotfree_input_line;
4536 int adjust;
4537
4538 save = input_line_pointer;
4539 gotfree_input_line = lex_got (&got_reloc, &adjust, NULL);
4540 if (gotfree_input_line)
4541 input_line_pointer = gotfree_input_line;
4542
4543 expression (exp);
4544
4545 if (gotfree_input_line)
4546 {
4547 /* expression () has merrily parsed up to the end of line,
4548 or a comma - in the wrong buffer. Transfer how far
4549 input_line_pointer has moved to the right buffer. */
4550 input_line_pointer = (save
4551 + (input_line_pointer - gotfree_input_line)
4552 + adjust);
4553 free (gotfree_input_line);
4554 }
4555 }
4556 else
4557 expression (exp);
4558 }
4559 #endif
4560
4561 static void signed_cons (int size)
4562 {
4563 if (flag_code == CODE_64BIT)
4564 cons_sign = 1;
4565 cons (size);
4566 cons_sign = -1;
4567 }
4568
4569 #ifdef TE_PE
4570 static void
4571 pe_directive_secrel (dummy)
4572 int dummy ATTRIBUTE_UNUSED;
4573 {
4574 expressionS exp;
4575
4576 do
4577 {
4578 expression (&exp);
4579 if (exp.X_op == O_symbol)
4580 exp.X_op = O_secrel;
4581
4582 emit_expr (&exp, 4);
4583 }
4584 while (*input_line_pointer++ == ',');
4585
4586 input_line_pointer--;
4587 demand_empty_rest_of_line ();
4588 }
4589 #endif
4590
4591 static int
4592 i386_immediate (char *imm_start)
4593 {
4594 char *save_input_line_pointer;
4595 char *gotfree_input_line;
4596 segT exp_seg = 0;
4597 expressionS *exp;
4598 unsigned int types = ~0U;
4599
4600 if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
4601 {
4602 as_bad (_("at most %d immediate operands are allowed"),
4603 MAX_IMMEDIATE_OPERANDS);
4604 return 0;
4605 }
4606
4607 exp = &im_expressions[i.imm_operands++];
4608 i.op[this_operand].imms = exp;
4609
4610 if (is_space_char (*imm_start))
4611 ++imm_start;
4612
4613 save_input_line_pointer = input_line_pointer;
4614 input_line_pointer = imm_start;
4615
4616 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
4617 if (gotfree_input_line)
4618 input_line_pointer = gotfree_input_line;
4619
4620 exp_seg = expression (exp);
4621
4622 SKIP_WHITESPACE ();
4623 if (*input_line_pointer)
4624 as_bad (_("junk `%s' after expression"), input_line_pointer);
4625
4626 input_line_pointer = save_input_line_pointer;
4627 if (gotfree_input_line)
4628 free (gotfree_input_line);
4629
4630 if (exp->X_op == O_absent || exp->X_op == O_big)
4631 {
4632 /* Missing or bad expr becomes absolute 0. */
4633 as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
4634 imm_start);
4635 exp->X_op = O_constant;
4636 exp->X_add_number = 0;
4637 exp->X_add_symbol = (symbolS *) 0;
4638 exp->X_op_symbol = (symbolS *) 0;
4639 }
4640 else if (exp->X_op == O_constant)
4641 {
4642 /* Size it properly later. */
4643 i.types[this_operand] |= Imm64;
4644 /* If BFD64, sign extend val. */
4645 if (!use_rela_relocations
4646 && (exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
4647 exp->X_add_number
4648 = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
4649 }
4650 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4651 else if (OUTPUT_FLAVOR == bfd_target_aout_flavour
4652 && exp_seg != absolute_section
4653 && exp_seg != text_section
4654 && exp_seg != data_section
4655 && exp_seg != bss_section
4656 && exp_seg != undefined_section
4657 && !bfd_is_com_section (exp_seg))
4658 {
4659 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
4660 return 0;
4661 }
4662 #endif
4663 else if (!intel_syntax && exp->X_op == O_register)
4664 {
4665 as_bad (_("illegal immediate register operand %s"), imm_start);
4666 return 0;
4667 }
4668 else
4669 {
4670 /* This is an address. The size of the address will be
4671 determined later, depending on destination register,
4672 suffix, or the default for the section. */
4673 i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
4674 i.types[this_operand] &= types;
4675 }
4676
4677 return 1;
4678 }
4679
4680 static char *
4681 i386_scale (char *scale)
4682 {
4683 offsetT val;
4684 char *save = input_line_pointer;
4685
4686 input_line_pointer = scale;
4687 val = get_absolute_expression ();
4688
4689 switch (val)
4690 {
4691 case 1:
4692 i.log2_scale_factor = 0;
4693 break;
4694 case 2:
4695 i.log2_scale_factor = 1;
4696 break;
4697 case 4:
4698 i.log2_scale_factor = 2;
4699 break;
4700 case 8:
4701 i.log2_scale_factor = 3;
4702 break;
4703 default:
4704 {
4705 char sep = *input_line_pointer;
4706
4707 *input_line_pointer = '\0';
4708 as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
4709 scale);
4710 *input_line_pointer = sep;
4711 input_line_pointer = save;
4712 return NULL;
4713 }
4714 }
4715 if (i.log2_scale_factor != 0 && i.index_reg == 0)
4716 {
4717 as_warn (_("scale factor of %d without an index register"),
4718 1 << i.log2_scale_factor);
4719 #if SCALE1_WHEN_NO_INDEX
4720 i.log2_scale_factor = 0;
4721 #endif
4722 }
4723 scale = input_line_pointer;
4724 input_line_pointer = save;
4725 return scale;
4726 }
4727
4728 static int
4729 i386_displacement (char *disp_start, char *disp_end)
4730 {
4731 expressionS *exp;
4732 segT exp_seg = 0;
4733 char *save_input_line_pointer;
4734 char *gotfree_input_line;
4735 int bigdisp, override;
4736 unsigned int types = Disp;
4737
4738 if (i.disp_operands == MAX_MEMORY_OPERANDS)
4739 {
4740 as_bad (_("at most %d displacement operands are allowed"),
4741 MAX_MEMORY_OPERANDS);
4742 return 0;
4743 }
4744
4745 if ((i.types[this_operand] & JumpAbsolute)
4746 || !(current_templates->start->opcode_modifier & (Jump | JumpDword)))
4747 {
4748 bigdisp = Disp32;
4749 override = (i.prefix[ADDR_PREFIX] != 0);
4750 }
4751 else
4752 {
4753 /* For PC-relative branches, the width of the displacement
4754 is dependent upon data size, not address size. */
4755 bigdisp = 0;
4756 override = (i.prefix[DATA_PREFIX] != 0);
4757 }
4758 if (flag_code == CODE_64BIT)
4759 {
4760 if (!bigdisp)
4761 bigdisp = ((override || i.suffix == WORD_MNEM_SUFFIX)
4762 ? Disp16
4763 : Disp32S | Disp32);
4764 else if (!override)
4765 bigdisp = Disp64 | Disp32S | Disp32;
4766 }
4767 else
4768 {
4769 if (!bigdisp)
4770 {
4771 if (!override)
4772 override = (i.suffix == (flag_code != CODE_16BIT
4773 ? WORD_MNEM_SUFFIX
4774 : LONG_MNEM_SUFFIX));
4775 bigdisp = Disp32;
4776 }
4777 if ((flag_code == CODE_16BIT) ^ override)
4778 bigdisp = Disp16;
4779 }
4780 i.types[this_operand] |= bigdisp;
4781
4782 exp = &disp_expressions[i.disp_operands];
4783 i.op[this_operand].disps = exp;
4784 i.disp_operands++;
4785 save_input_line_pointer = input_line_pointer;
4786 input_line_pointer = disp_start;
4787 END_STRING_AND_SAVE (disp_end);
4788
4789 #ifndef GCC_ASM_O_HACK
4790 #define GCC_ASM_O_HACK 0
4791 #endif
4792 #if GCC_ASM_O_HACK
4793 END_STRING_AND_SAVE (disp_end + 1);
4794 if ((i.types[this_operand] & BaseIndex) != 0
4795 && displacement_string_end[-1] == '+')
4796 {
4797 /* This hack is to avoid a warning when using the "o"
4798 constraint within gcc asm statements.
4799 For instance:
4800
4801 #define _set_tssldt_desc(n,addr,limit,type) \
4802 __asm__ __volatile__ ( \
4803 "movw %w2,%0\n\t" \
4804 "movw %w1,2+%0\n\t" \
4805 "rorl $16,%1\n\t" \
4806 "movb %b1,4+%0\n\t" \
4807 "movb %4,5+%0\n\t" \
4808 "movb $0,6+%0\n\t" \
4809 "movb %h1,7+%0\n\t" \
4810 "rorl $16,%1" \
4811 : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
4812
4813 This works great except that the output assembler ends
4814 up looking a bit weird if it turns out that there is
4815 no offset. You end up producing code that looks like:
4816
4817 #APP
4818 movw $235,(%eax)
4819 movw %dx,2+(%eax)
4820 rorl $16,%edx
4821 movb %dl,4+(%eax)
4822 movb $137,5+(%eax)
4823 movb $0,6+(%eax)
4824 movb %dh,7+(%eax)
4825 rorl $16,%edx
4826 #NO_APP
4827
4828 So here we provide the missing zero. */
4829
4830 *displacement_string_end = '0';
4831 }
4832 #endif
4833 gotfree_input_line = lex_got (&i.reloc[this_operand], NULL, &types);
4834 if (gotfree_input_line)
4835 input_line_pointer = gotfree_input_line;
4836
4837 exp_seg = expression (exp);
4838
4839 SKIP_WHITESPACE ();
4840 if (*input_line_pointer)
4841 as_bad (_("junk `%s' after expression"), input_line_pointer);
4842 #if GCC_ASM_O_HACK
4843 RESTORE_END_STRING (disp_end + 1);
4844 #endif
4845 RESTORE_END_STRING (disp_end);
4846 input_line_pointer = save_input_line_pointer;
4847 if (gotfree_input_line)
4848 free (gotfree_input_line);
4849
4850 /* We do this to make sure that the section symbol is in
4851 the symbol table. We will ultimately change the relocation
4852 to be relative to the beginning of the section. */
4853 if (i.reloc[this_operand] == BFD_RELOC_386_GOTOFF
4854 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
4855 || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTOFF64)
4856 {
4857 if (exp->X_op != O_symbol)
4858 {
4859 as_bad (_("bad expression used with @%s"),
4860 (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
4861 ? "GOTPCREL"
4862 : "GOTOFF"));
4863 return 0;
4864 }
4865
4866 if (S_IS_LOCAL (exp->X_add_symbol)
4867 && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
4868 section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
4869 exp->X_op = O_subtract;
4870 exp->X_op_symbol = GOT_symbol;
4871 if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
4872 i.reloc[this_operand] = BFD_RELOC_32_PCREL;
4873 else if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTOFF64)
4874 i.reloc[this_operand] = BFD_RELOC_64;
4875 else
4876 i.reloc[this_operand] = BFD_RELOC_32;
4877 }
4878
4879 if (exp->X_op == O_absent || exp->X_op == O_big)
4880 {
4881 /* Missing or bad expr becomes absolute 0. */
4882 as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
4883 disp_start);
4884 exp->X_op = O_constant;
4885 exp->X_add_number = 0;
4886 exp->X_add_symbol = (symbolS *) 0;
4887 exp->X_op_symbol = (symbolS *) 0;
4888 }
4889
4890 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
4891 if (exp->X_op != O_constant
4892 && OUTPUT_FLAVOR == bfd_target_aout_flavour
4893 && exp_seg != absolute_section
4894 && exp_seg != text_section
4895 && exp_seg != data_section
4896 && exp_seg != bss_section
4897 && exp_seg != undefined_section
4898 && !bfd_is_com_section (exp_seg))
4899 {
4900 as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
4901 return 0;
4902 }
4903 #endif
4904
4905 if (!(i.types[this_operand] & ~Disp))
4906 i.types[this_operand] &= types;
4907
4908 return 1;
4909 }
4910
4911 /* Make sure the memory operand we've been dealt is valid.
4912 Return 1 on success, 0 on a failure. */
4913
4914 static int
4915 i386_index_check (const char *operand_string)
4916 {
4917 int ok;
4918 #if INFER_ADDR_PREFIX
4919 int fudged = 0;
4920
4921 tryprefix:
4922 #endif
4923 ok = 1;
4924 if ((current_templates->start->cpu_flags & CpuSVME)
4925 && current_templates->end[-1].operand_types[0] == AnyMem)
4926 {
4927 /* Memory operands of SVME insns are special in that they only allow
4928 rAX as their memory address and ignore any segment override. */
4929 unsigned RegXX;
4930
4931 /* SKINIT is even more restrictive: it always requires EAX. */
4932 if (strcmp (current_templates->start->name, "skinit") == 0)
4933 RegXX = Reg32;
4934 else if (flag_code == CODE_64BIT)
4935 RegXX = i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32;
4936 else
4937 RegXX = ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)
4938 ? Reg16
4939 : Reg32);
4940 if (!i.base_reg
4941 || !(i.base_reg->reg_type & Acc)
4942 || !(i.base_reg->reg_type & RegXX)
4943 || i.index_reg
4944 || (i.types[0] & Disp))
4945 ok = 0;
4946 }
4947 else if (flag_code == CODE_64BIT)
4948 {
4949 unsigned RegXX = (i.prefix[ADDR_PREFIX] == 0 ? Reg64 : Reg32);
4950
4951 if ((i.base_reg
4952 && ((i.base_reg->reg_type & RegXX) == 0)
4953 && (i.base_reg->reg_type != BaseIndex
4954 || i.index_reg))
4955 || (i.index_reg
4956 && ((i.index_reg->reg_type & (RegXX | BaseIndex))
4957 != (RegXX | BaseIndex))))
4958 ok = 0;
4959 }
4960 else
4961 {
4962 if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
4963 {
4964 /* 16bit checks. */
4965 if ((i.base_reg
4966 && ((i.base_reg->reg_type & (Reg16 | BaseIndex | RegRex))
4967 != (Reg16 | BaseIndex)))
4968 || (i.index_reg
4969 && (((i.index_reg->reg_type & (Reg16 | BaseIndex))
4970 != (Reg16 | BaseIndex))
4971 || !(i.base_reg
4972 && i.base_reg->reg_num < 6
4973 && i.index_reg->reg_num >= 6
4974 && i.log2_scale_factor == 0))))
4975 ok = 0;
4976 }
4977 else
4978 {
4979 /* 32bit checks. */
4980 if ((i.base_reg
4981 && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
4982 || (i.index_reg
4983 && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
4984 != (Reg32 | BaseIndex))))
4985 ok = 0;
4986 }
4987 }
4988 if (!ok)
4989 {
4990 #if INFER_ADDR_PREFIX
4991 if (i.prefix[ADDR_PREFIX] == 0)
4992 {
4993 i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
4994 i.prefixes += 1;
4995 /* Change the size of any displacement too. At most one of
4996 Disp16 or Disp32 is set.
4997 FIXME. There doesn't seem to be any real need for separate
4998 Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
4999 Removing them would probably clean up the code quite a lot. */
5000 if (flag_code != CODE_64BIT
5001 && (i.types[this_operand] & (Disp16 | Disp32)))
5002 i.types[this_operand] ^= (Disp16 | Disp32);
5003 fudged = 1;
5004 goto tryprefix;
5005 }
5006 if (fudged)
5007 as_bad (_("`%s' is not a valid base/index expression"),
5008 operand_string);
5009 else
5010 #endif
5011 as_bad (_("`%s' is not a valid %s bit base/index expression"),
5012 operand_string,
5013 flag_code_names[flag_code]);
5014 }
5015 return ok;
5016 }
5017
5018 /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
5019 on error. */
5020
5021 static int
5022 i386_operand (char *operand_string)
5023 {
5024 const reg_entry *r;
5025 char *end_op;
5026 char *op_string = operand_string;
5027
5028 if (is_space_char (*op_string))
5029 ++op_string;
5030
5031 /* We check for an absolute prefix (differentiating,
5032 for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
5033 if (*op_string == ABSOLUTE_PREFIX)
5034 {
5035 ++op_string;
5036 if (is_space_char (*op_string))
5037 ++op_string;
5038 i.types[this_operand] |= JumpAbsolute;
5039 }
5040
5041 /* Check if operand is a register. */
5042 if ((r = parse_register (op_string, &end_op)) != NULL)
5043 {
5044 /* Check for a segment override by searching for ':' after a
5045 segment register. */
5046 op_string = end_op;
5047 if (is_space_char (*op_string))
5048 ++op_string;
5049 if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
5050 {
5051 switch (r->reg_num)
5052 {
5053 case 0:
5054 i.seg[i.mem_operands] = &es;
5055 break;
5056 case 1:
5057 i.seg[i.mem_operands] = &cs;
5058 break;
5059 case 2:
5060 i.seg[i.mem_operands] = &ss;
5061 break;
5062 case 3:
5063 i.seg[i.mem_operands] = &ds;
5064 break;
5065 case 4:
5066 i.seg[i.mem_operands] = &fs;
5067 break;
5068 case 5:
5069 i.seg[i.mem_operands] = &gs;
5070 break;
5071 }
5072
5073 /* Skip the ':' and whitespace. */
5074 ++op_string;
5075 if (is_space_char (*op_string))
5076 ++op_string;
5077
5078 if (!is_digit_char (*op_string)
5079 && !is_identifier_char (*op_string)
5080 && *op_string != '('
5081 && *op_string != ABSOLUTE_PREFIX)
5082 {
5083 as_bad (_("bad memory operand `%s'"), op_string);
5084 return 0;
5085 }
5086 /* Handle case of %es:*foo. */
5087 if (*op_string == ABSOLUTE_PREFIX)
5088 {
5089 ++op_string;
5090 if (is_space_char (*op_string))
5091 ++op_string;
5092 i.types[this_operand] |= JumpAbsolute;
5093 }
5094 goto do_memory_reference;
5095 }
5096 if (*op_string)
5097 {
5098 as_bad (_("junk `%s' after register"), op_string);
5099 return 0;
5100 }
5101 i.types[this_operand] |= r->reg_type & ~BaseIndex;
5102 i.op[this_operand].regs = r;
5103 i.reg_operands++;
5104 }
5105 else if (*op_string == REGISTER_PREFIX)
5106 {
5107 as_bad (_("bad register name `%s'"), op_string);
5108 return 0;
5109 }
5110 else if (*op_string == IMMEDIATE_PREFIX)
5111 {
5112 ++op_string;
5113 if (i.types[this_operand] & JumpAbsolute)
5114 {
5115 as_bad (_("immediate operand illegal with absolute jump"));
5116 return 0;
5117 }
5118 if (!i386_immediate (op_string))
5119 return 0;
5120 }
5121 else if (is_digit_char (*op_string)
5122 || is_identifier_char (*op_string)
5123 || *op_string == '(')
5124 {
5125 /* This is a memory reference of some sort. */
5126 char *base_string;
5127
5128 /* Start and end of displacement string expression (if found). */
5129 char *displacement_string_start;
5130 char *displacement_string_end;
5131
5132 do_memory_reference:
5133 if ((i.mem_operands == 1
5134 && (current_templates->start->opcode_modifier & IsString) == 0)
5135 || i.mem_operands == 2)
5136 {
5137 as_bad (_("too many memory references for `%s'"),
5138 current_templates->start->name);
5139 return 0;
5140 }
5141
5142 /* Check for base index form. We detect the base index form by
5143 looking for an ')' at the end of the operand, searching
5144 for the '(' matching it, and finding a REGISTER_PREFIX or ','
5145 after the '('. */
5146 base_string = op_string + strlen (op_string);
5147
5148 --base_string;
5149 if (is_space_char (*base_string))
5150 --base_string;
5151
5152 /* If we only have a displacement, set-up for it to be parsed later. */
5153 displacement_string_start = op_string;
5154 displacement_string_end = base_string + 1;
5155
5156 if (*base_string == ')')
5157 {
5158 char *temp_string;
5159 unsigned int parens_balanced = 1;
5160 /* We've already checked that the number of left & right ()'s are
5161 equal, so this loop will not be infinite. */
5162 do
5163 {
5164 base_string--;
5165 if (*base_string == ')')
5166 parens_balanced++;
5167 if (*base_string == '(')
5168 parens_balanced--;
5169 }
5170 while (parens_balanced);
5171
5172 temp_string = base_string;
5173
5174 /* Skip past '(' and whitespace. */
5175 ++base_string;
5176 if (is_space_char (*base_string))
5177 ++base_string;
5178
5179 if (*base_string == ','
5180 || ((i.base_reg = parse_register (base_string, &end_op))
5181 != NULL))
5182 {
5183 displacement_string_end = temp_string;
5184
5185 i.types[this_operand] |= BaseIndex;
5186
5187 if (i.base_reg)
5188 {
5189 base_string = end_op;
5190 if (is_space_char (*base_string))
5191 ++base_string;
5192 }
5193
5194 /* There may be an index reg or scale factor here. */
5195 if (*base_string == ',')
5196 {
5197 ++base_string;
5198 if (is_space_char (*base_string))
5199 ++base_string;
5200
5201 if ((i.index_reg = parse_register (base_string, &end_op))
5202 != NULL)
5203 {
5204 base_string = end_op;
5205 if (is_space_char (*base_string))
5206 ++base_string;
5207 if (*base_string == ',')
5208 {
5209 ++base_string;
5210 if (is_space_char (*base_string))
5211 ++base_string;
5212 }
5213 else if (*base_string != ')')
5214 {
5215 as_bad (_("expecting `,' or `)' "
5216 "after index register in `%s'"),
5217 operand_string);
5218 return 0;
5219 }
5220 }
5221 else if (*base_string == REGISTER_PREFIX)
5222 {
5223 as_bad (_("bad register name `%s'"), base_string);
5224 return 0;
5225 }
5226
5227 /* Check for scale factor. */
5228 if (*base_string != ')')
5229 {
5230 char *end_scale = i386_scale (base_string);
5231
5232 if (!end_scale)
5233 return 0;
5234
5235 base_string = end_scale;
5236 if (is_space_char (*base_string))
5237 ++base_string;
5238 if (*base_string != ')')
5239 {
5240 as_bad (_("expecting `)' "
5241 "after scale factor in `%s'"),
5242 operand_string);
5243 return 0;
5244 }
5245 }
5246 else if (!i.index_reg)
5247 {
5248 as_bad (_("expecting index register or scale factor "
5249 "after `,'; got '%c'"),
5250 *base_string);
5251 return 0;
5252 }
5253 }
5254 else if (*base_string != ')')
5255 {
5256 as_bad (_("expecting `,' or `)' "
5257 "after base register in `%s'"),
5258 operand_string);
5259 return 0;
5260 }
5261 }
5262 else if (*base_string == REGISTER_PREFIX)
5263 {
5264 as_bad (_("bad register name `%s'"), base_string);
5265 return 0;
5266 }
5267 }
5268
5269 /* If there's an expression beginning the operand, parse it,
5270 assuming displacement_string_start and
5271 displacement_string_end are meaningful. */
5272 if (displacement_string_start != displacement_string_end)
5273 {
5274 if (!i386_displacement (displacement_string_start,
5275 displacement_string_end))
5276 return 0;
5277 }
5278
5279 /* Special case for (%dx) while doing input/output op. */
5280 if (i.base_reg
5281 && i.base_reg->reg_type == (Reg16 | InOutPortReg)
5282 && i.index_reg == 0
5283 && i.log2_scale_factor == 0
5284 && i.seg[i.mem_operands] == 0
5285 && (i.types[this_operand] & Disp) == 0)
5286 {
5287 i.types[this_operand] = InOutPortReg;
5288 return 1;
5289 }
5290
5291 if (i386_index_check (operand_string) == 0)
5292 return 0;
5293 i.mem_operands++;
5294 }
5295 else
5296 {
5297 /* It's not a memory operand; argh! */
5298 as_bad (_("invalid char %s beginning operand %d `%s'"),
5299 output_invalid (*op_string),
5300 this_operand + 1,
5301 op_string);
5302 return 0;
5303 }
5304 return 1; /* Normal return. */
5305 }
5306 \f
5307 /* md_estimate_size_before_relax()
5308
5309 Called just before relax() for rs_machine_dependent frags. The x86
5310 assembler uses these frags to handle variable size jump
5311 instructions.
5312
5313 Any symbol that is now undefined will not become defined.
5314 Return the correct fr_subtype in the frag.
5315 Return the initial "guess for variable size of frag" to caller.
5316 The guess is actually the growth beyond the fixed part. Whatever
5317 we do to grow the fixed or variable part contributes to our
5318 returned value. */
5319
5320 int
5321 md_estimate_size_before_relax (fragP, segment)
5322 fragS *fragP;
5323 segT segment;
5324 {
5325 /* We've already got fragP->fr_subtype right; all we have to do is
5326 check for un-relaxable symbols. On an ELF system, we can't relax
5327 an externally visible symbol, because it may be overridden by a
5328 shared library. */
5329 if (S_GET_SEGMENT (fragP->fr_symbol) != segment
5330 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5331 || (IS_ELF
5332 && (S_IS_EXTERNAL (fragP->fr_symbol)
5333 || S_IS_WEAK (fragP->fr_symbol)))
5334 #endif
5335 )
5336 {
5337 /* Symbol is undefined in this segment, or we need to keep a
5338 reloc so that weak symbols can be overridden. */
5339 int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
5340 enum bfd_reloc_code_real reloc_type;
5341 unsigned char *opcode;
5342 int old_fr_fix;
5343
5344 if (fragP->fr_var != NO_RELOC)
5345 reloc_type = fragP->fr_var;
5346 else if (size == 2)
5347 reloc_type = BFD_RELOC_16_PCREL;
5348 else
5349 reloc_type = BFD_RELOC_32_PCREL;
5350
5351 old_fr_fix = fragP->fr_fix;
5352 opcode = (unsigned char *) fragP->fr_opcode;
5353
5354 switch (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
5355 {
5356 case UNCOND_JUMP:
5357 /* Make jmp (0xeb) a (d)word displacement jump. */
5358 opcode[0] = 0xe9;
5359 fragP->fr_fix += size;
5360 fix_new (fragP, old_fr_fix, size,
5361 fragP->fr_symbol,
5362 fragP->fr_offset, 1,
5363 reloc_type);
5364 break;
5365
5366 case COND_JUMP86:
5367 if (size == 2
5368 && (!no_cond_jump_promotion || fragP->fr_var != NO_RELOC))
5369 {
5370 /* Negate the condition, and branch past an
5371 unconditional jump. */
5372 opcode[0] ^= 1;
5373 opcode[1] = 3;
5374 /* Insert an unconditional jump. */
5375 opcode[2] = 0xe9;
5376 /* We added two extra opcode bytes, and have a two byte
5377 offset. */
5378 fragP->fr_fix += 2 + 2;
5379 fix_new (fragP, old_fr_fix + 2, 2,
5380 fragP->fr_symbol,
5381 fragP->fr_offset, 1,
5382 reloc_type);
5383 break;
5384 }
5385 /* Fall through. */
5386
5387 case COND_JUMP:
5388 if (no_cond_jump_promotion && fragP->fr_var == NO_RELOC)
5389 {
5390 fixS *fixP;
5391
5392 fragP->fr_fix += 1;
5393 fixP = fix_new (fragP, old_fr_fix, 1,
5394 fragP->fr_symbol,
5395 fragP->fr_offset, 1,
5396 BFD_RELOC_8_PCREL);
5397 fixP->fx_signed = 1;
5398 break;
5399 }
5400
5401 /* This changes the byte-displacement jump 0x7N
5402 to the (d)word-displacement jump 0x0f,0x8N. */
5403 opcode[1] = opcode[0] + 0x10;
5404 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5405 /* We've added an opcode byte. */
5406 fragP->fr_fix += 1 + size;
5407 fix_new (fragP, old_fr_fix + 1, size,
5408 fragP->fr_symbol,
5409 fragP->fr_offset, 1,
5410 reloc_type);
5411 break;
5412
5413 default:
5414 BAD_CASE (fragP->fr_subtype);
5415 break;
5416 }
5417 frag_wane (fragP);
5418 return fragP->fr_fix - old_fr_fix;
5419 }
5420
5421 /* Guess size depending on current relax state. Initially the relax
5422 state will correspond to a short jump and we return 1, because
5423 the variable part of the frag (the branch offset) is one byte
5424 long. However, we can relax a section more than once and in that
5425 case we must either set fr_subtype back to the unrelaxed state,
5426 or return the value for the appropriate branch. */
5427 return md_relax_table[fragP->fr_subtype].rlx_length;
5428 }
5429
5430 /* Called after relax() is finished.
5431
5432 In: Address of frag.
5433 fr_type == rs_machine_dependent.
5434 fr_subtype is what the address relaxed to.
5435
5436 Out: Any fixSs and constants are set up.
5437 Caller will turn frag into a ".space 0". */
5438
5439 void
5440 md_convert_frag (abfd, sec, fragP)
5441 bfd *abfd ATTRIBUTE_UNUSED;
5442 segT sec ATTRIBUTE_UNUSED;
5443 fragS *fragP;
5444 {
5445 unsigned char *opcode;
5446 unsigned char *where_to_put_displacement = NULL;
5447 offsetT target_address;
5448 offsetT opcode_address;
5449 unsigned int extension = 0;
5450 offsetT displacement_from_opcode_start;
5451
5452 opcode = (unsigned char *) fragP->fr_opcode;
5453
5454 /* Address we want to reach in file space. */
5455 target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
5456
5457 /* Address opcode resides at in file space. */
5458 opcode_address = fragP->fr_address + fragP->fr_fix;
5459
5460 /* Displacement from opcode start to fill into instruction. */
5461 displacement_from_opcode_start = target_address - opcode_address;
5462
5463 if ((fragP->fr_subtype & BIG) == 0)
5464 {
5465 /* Don't have to change opcode. */
5466 extension = 1; /* 1 opcode + 1 displacement */
5467 where_to_put_displacement = &opcode[1];
5468 }
5469 else
5470 {
5471 if (no_cond_jump_promotion
5472 && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
5473 as_warn_where (fragP->fr_file, fragP->fr_line,
5474 _("long jump required"));
5475
5476 switch (fragP->fr_subtype)
5477 {
5478 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
5479 extension = 4; /* 1 opcode + 4 displacement */
5480 opcode[0] = 0xe9;
5481 where_to_put_displacement = &opcode[1];
5482 break;
5483
5484 case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
5485 extension = 2; /* 1 opcode + 2 displacement */
5486 opcode[0] = 0xe9;
5487 where_to_put_displacement = &opcode[1];
5488 break;
5489
5490 case ENCODE_RELAX_STATE (COND_JUMP, BIG):
5491 case ENCODE_RELAX_STATE (COND_JUMP86, BIG):
5492 extension = 5; /* 2 opcode + 4 displacement */
5493 opcode[1] = opcode[0] + 0x10;
5494 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5495 where_to_put_displacement = &opcode[2];
5496 break;
5497
5498 case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
5499 extension = 3; /* 2 opcode + 2 displacement */
5500 opcode[1] = opcode[0] + 0x10;
5501 opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
5502 where_to_put_displacement = &opcode[2];
5503 break;
5504
5505 case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
5506 extension = 4;
5507 opcode[0] ^= 1;
5508 opcode[1] = 3;
5509 opcode[2] = 0xe9;
5510 where_to_put_displacement = &opcode[3];
5511 break;
5512
5513 default:
5514 BAD_CASE (fragP->fr_subtype);
5515 break;
5516 }
5517 }
5518
5519 /* If size if less then four we are sure that the operand fits,
5520 but if it's 4, then it could be that the displacement is larger
5521 then -/+ 2GB. */
5522 if (DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype) == 4
5523 && object_64bit
5524 && ((addressT) (displacement_from_opcode_start - extension
5525 + ((addressT) 1 << 31))
5526 > (((addressT) 2 << 31) - 1)))
5527 {
5528 as_bad_where (fragP->fr_file, fragP->fr_line,
5529 _("jump target out of range"));
5530 /* Make us emit 0. */
5531 displacement_from_opcode_start = extension;
5532 }
5533 /* Now put displacement after opcode. */
5534 md_number_to_chars ((char *) where_to_put_displacement,
5535 (valueT) (displacement_from_opcode_start - extension),
5536 DISP_SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
5537 fragP->fr_fix += extension;
5538 }
5539 \f
5540 /* Size of byte displacement jmp. */
5541 int md_short_jump_size = 2;
5542
5543 /* Size of dword displacement jmp. */
5544 int md_long_jump_size = 5;
5545
5546 void
5547 md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
5548 char *ptr;
5549 addressT from_addr, to_addr;
5550 fragS *frag ATTRIBUTE_UNUSED;
5551 symbolS *to_symbol ATTRIBUTE_UNUSED;
5552 {
5553 offsetT offset;
5554
5555 offset = to_addr - (from_addr + 2);
5556 /* Opcode for byte-disp jump. */
5557 md_number_to_chars (ptr, (valueT) 0xeb, 1);
5558 md_number_to_chars (ptr + 1, (valueT) offset, 1);
5559 }
5560
5561 void
5562 md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
5563 char *ptr;
5564 addressT from_addr, to_addr;
5565 fragS *frag ATTRIBUTE_UNUSED;
5566 symbolS *to_symbol ATTRIBUTE_UNUSED;
5567 {
5568 offsetT offset;
5569
5570 offset = to_addr - (from_addr + 5);
5571 md_number_to_chars (ptr, (valueT) 0xe9, 1);
5572 md_number_to_chars (ptr + 1, (valueT) offset, 4);
5573 }
5574 \f
5575 /* Apply a fixup (fixS) to segment data, once it has been determined
5576 by our caller that we have all the info we need to fix it up.
5577
5578 On the 386, immediates, displacements, and data pointers are all in
5579 the same (little-endian) format, so we don't need to care about which
5580 we are handling. */
5581
5582 void
5583 md_apply_fix (fixP, valP, seg)
5584 /* The fix we're to put in. */
5585 fixS *fixP;
5586 /* Pointer to the value of the bits. */
5587 valueT *valP;
5588 /* Segment fix is from. */
5589 segT seg ATTRIBUTE_UNUSED;
5590 {
5591 char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
5592 valueT value = *valP;
5593
5594 #if !defined (TE_Mach)
5595 if (fixP->fx_pcrel)
5596 {
5597 switch (fixP->fx_r_type)
5598 {
5599 default:
5600 break;
5601
5602 case BFD_RELOC_64:
5603 fixP->fx_r_type = BFD_RELOC_64_PCREL;
5604 break;
5605 case BFD_RELOC_32:
5606 case BFD_RELOC_X86_64_32S:
5607 fixP->fx_r_type = BFD_RELOC_32_PCREL;
5608 break;
5609 case BFD_RELOC_16:
5610 fixP->fx_r_type = BFD_RELOC_16_PCREL;
5611 break;
5612 case BFD_RELOC_8:
5613 fixP->fx_r_type = BFD_RELOC_8_PCREL;
5614 break;
5615 }
5616 }
5617
5618 if (fixP->fx_addsy != NULL
5619 && (fixP->fx_r_type == BFD_RELOC_32_PCREL
5620 || fixP->fx_r_type == BFD_RELOC_64_PCREL
5621 || fixP->fx_r_type == BFD_RELOC_16_PCREL
5622 || fixP->fx_r_type == BFD_RELOC_8_PCREL)
5623 && !use_rela_relocations)
5624 {
5625 /* This is a hack. There should be a better way to handle this.
5626 This covers for the fact that bfd_install_relocation will
5627 subtract the current location (for partial_inplace, PC relative
5628 relocations); see more below. */
5629 #ifndef OBJ_AOUT
5630 if (IS_ELF
5631 #ifdef TE_PE
5632 || OUTPUT_FLAVOR == bfd_target_coff_flavour
5633 #endif
5634 )
5635 value += fixP->fx_where + fixP->fx_frag->fr_address;
5636 #endif
5637 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5638 if (IS_ELF)
5639 {
5640 segT sym_seg = S_GET_SEGMENT (fixP->fx_addsy);
5641
5642 if ((sym_seg == seg
5643 || (symbol_section_p (fixP->fx_addsy)
5644 && sym_seg != absolute_section))
5645 && !generic_force_reloc (fixP))
5646 {
5647 /* Yes, we add the values in twice. This is because
5648 bfd_install_relocation subtracts them out again. I think
5649 bfd_install_relocation is broken, but I don't dare change
5650 it. FIXME. */
5651 value += fixP->fx_where + fixP->fx_frag->fr_address;
5652 }
5653 }
5654 #endif
5655 #if defined (OBJ_COFF) && defined (TE_PE)
5656 /* For some reason, the PE format does not store a
5657 section address offset for a PC relative symbol. */
5658 if (S_GET_SEGMENT (fixP->fx_addsy) != seg
5659 || S_IS_WEAK (fixP->fx_addsy))
5660 value += md_pcrel_from (fixP);
5661 #endif
5662 }
5663
5664 /* Fix a few things - the dynamic linker expects certain values here,
5665 and we must not disappoint it. */
5666 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5667 if (IS_ELF && fixP->fx_addsy)
5668 switch (fixP->fx_r_type)
5669 {
5670 case BFD_RELOC_386_PLT32:
5671 case BFD_RELOC_X86_64_PLT32:
5672 /* Make the jump instruction point to the address of the operand. At
5673 runtime we merely add the offset to the actual PLT entry. */
5674 value = -4;
5675 break;
5676
5677 case BFD_RELOC_386_TLS_GD:
5678 case BFD_RELOC_386_TLS_LDM:
5679 case BFD_RELOC_386_TLS_IE_32:
5680 case BFD_RELOC_386_TLS_IE:
5681 case BFD_RELOC_386_TLS_GOTIE:
5682 case BFD_RELOC_386_TLS_GOTDESC:
5683 case BFD_RELOC_X86_64_TLSGD:
5684 case BFD_RELOC_X86_64_TLSLD:
5685 case BFD_RELOC_X86_64_GOTTPOFF:
5686 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
5687 value = 0; /* Fully resolved at runtime. No addend. */
5688 /* Fallthrough */
5689 case BFD_RELOC_386_TLS_LE:
5690 case BFD_RELOC_386_TLS_LDO_32:
5691 case BFD_RELOC_386_TLS_LE_32:
5692 case BFD_RELOC_X86_64_DTPOFF32:
5693 case BFD_RELOC_X86_64_DTPOFF64:
5694 case BFD_RELOC_X86_64_TPOFF32:
5695 case BFD_RELOC_X86_64_TPOFF64:
5696 S_SET_THREAD_LOCAL (fixP->fx_addsy);
5697 break;
5698
5699 case BFD_RELOC_386_TLS_DESC_CALL:
5700 case BFD_RELOC_X86_64_TLSDESC_CALL:
5701 value = 0; /* Fully resolved at runtime. No addend. */
5702 S_SET_THREAD_LOCAL (fixP->fx_addsy);
5703 fixP->fx_done = 0;
5704 return;
5705
5706 case BFD_RELOC_386_GOT32:
5707 case BFD_RELOC_X86_64_GOT32:
5708 value = 0; /* Fully resolved at runtime. No addend. */
5709 break;
5710
5711 case BFD_RELOC_VTABLE_INHERIT:
5712 case BFD_RELOC_VTABLE_ENTRY:
5713 fixP->fx_done = 0;
5714 return;
5715
5716 default:
5717 break;
5718 }
5719 #endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
5720 *valP = value;
5721 #endif /* !defined (TE_Mach) */
5722
5723 /* Are we finished with this relocation now? */
5724 if (fixP->fx_addsy == NULL)
5725 fixP->fx_done = 1;
5726 else if (use_rela_relocations)
5727 {
5728 fixP->fx_no_overflow = 1;
5729 /* Remember value for tc_gen_reloc. */
5730 fixP->fx_addnumber = value;
5731 value = 0;
5732 }
5733
5734 md_number_to_chars (p, value, fixP->fx_size);
5735 }
5736 \f
5737 #define MAX_LITTLENUMS 6
5738
5739 /* Turn the string pointed to by litP into a floating point constant
5740 of type TYPE, and emit the appropriate bytes. The number of
5741 LITTLENUMS emitted is stored in *SIZEP. An error message is
5742 returned, or NULL on OK. */
5743
5744 char *
5745 md_atof (type, litP, sizeP)
5746 int type;
5747 char *litP;
5748 int *sizeP;
5749 {
5750 int prec;
5751 LITTLENUM_TYPE words[MAX_LITTLENUMS];
5752 LITTLENUM_TYPE *wordP;
5753 char *t;
5754
5755 switch (type)
5756 {
5757 case 'f':
5758 case 'F':
5759 prec = 2;
5760 break;
5761
5762 case 'd':
5763 case 'D':
5764 prec = 4;
5765 break;
5766
5767 case 'x':
5768 case 'X':
5769 prec = 5;
5770 break;
5771
5772 default:
5773 *sizeP = 0;
5774 return _("Bad call to md_atof ()");
5775 }
5776 t = atof_ieee (input_line_pointer, type, words);
5777 if (t)
5778 input_line_pointer = t;
5779
5780 *sizeP = prec * sizeof (LITTLENUM_TYPE);
5781 /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
5782 the bigendian 386. */
5783 for (wordP = words + prec - 1; prec--;)
5784 {
5785 md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
5786 litP += sizeof (LITTLENUM_TYPE);
5787 }
5788 return 0;
5789 }
5790 \f
5791 static char output_invalid_buf[sizeof (unsigned char) * 2 + 6];
5792
5793 static char *
5794 output_invalid (int c)
5795 {
5796 if (ISPRINT (c))
5797 snprintf (output_invalid_buf, sizeof (output_invalid_buf),
5798 "'%c'", c);
5799 else
5800 snprintf (output_invalid_buf, sizeof (output_invalid_buf),
5801 "(0x%x)", (unsigned char) c);
5802 return output_invalid_buf;
5803 }
5804
5805 /* REG_STRING starts *before* REGISTER_PREFIX. */
5806
5807 static const reg_entry *
5808 parse_real_register (char *reg_string, char **end_op)
5809 {
5810 char *s = reg_string;
5811 char *p;
5812 char reg_name_given[MAX_REG_NAME_SIZE + 1];
5813 const reg_entry *r;
5814
5815 /* Skip possible REGISTER_PREFIX and possible whitespace. */
5816 if (*s == REGISTER_PREFIX)
5817 ++s;
5818
5819 if (is_space_char (*s))
5820 ++s;
5821
5822 p = reg_name_given;
5823 while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
5824 {
5825 if (p >= reg_name_given + MAX_REG_NAME_SIZE)
5826 return (const reg_entry *) NULL;
5827 s++;
5828 }
5829
5830 /* For naked regs, make sure that we are not dealing with an identifier.
5831 This prevents confusing an identifier like `eax_var' with register
5832 `eax'. */
5833 if (allow_naked_reg && identifier_chars[(unsigned char) *s])
5834 return (const reg_entry *) NULL;
5835
5836 *end_op = s;
5837
5838 r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
5839
5840 /* Handle floating point regs, allowing spaces in the (i) part. */
5841 if (r == i386_regtab /* %st is first entry of table */)
5842 {
5843 if (is_space_char (*s))
5844 ++s;
5845 if (*s == '(')
5846 {
5847 ++s;
5848 if (is_space_char (*s))
5849 ++s;
5850 if (*s >= '0' && *s <= '7')
5851 {
5852 int fpr = *s - '0';
5853 ++s;
5854 if (is_space_char (*s))
5855 ++s;
5856 if (*s == ')')
5857 {
5858 *end_op = s + 1;
5859 r = hash_find (reg_hash, "st(0)");
5860 know (r);
5861 return r + fpr;
5862 }
5863 }
5864 /* We have "%st(" then garbage. */
5865 return (const reg_entry *) NULL;
5866 }
5867 }
5868
5869 if (r != NULL
5870 && ((r->reg_flags & (RegRex64 | RegRex)) | (r->reg_type & Reg64)) != 0
5871 && (r->reg_type != Control || !(cpu_arch_flags & CpuSledgehammer))
5872 && flag_code != CODE_64BIT)
5873 return (const reg_entry *) NULL;
5874
5875 return r;
5876 }
5877
5878 /* REG_STRING starts *before* REGISTER_PREFIX. */
5879
5880 static const reg_entry *
5881 parse_register (char *reg_string, char **end_op)
5882 {
5883 const reg_entry *r;
5884
5885 if (*reg_string == REGISTER_PREFIX || allow_naked_reg)
5886 r = parse_real_register (reg_string, end_op);
5887 else
5888 r = NULL;
5889 if (!r)
5890 {
5891 char *save = input_line_pointer;
5892 char c;
5893 symbolS *symbolP;
5894
5895 input_line_pointer = reg_string;
5896 c = get_symbol_end ();
5897 symbolP = symbol_find (reg_string);
5898 if (symbolP && S_GET_SEGMENT (symbolP) == reg_section)
5899 {
5900 const expressionS *e = symbol_get_value_expression (symbolP);
5901
5902 know (e->X_op == O_register);
5903 know (e->X_add_number >= 0
5904 && (valueT) e->X_add_number < i386_regtab_size);
5905 r = i386_regtab + e->X_add_number;
5906 *end_op = input_line_pointer;
5907 }
5908 *input_line_pointer = c;
5909 input_line_pointer = save;
5910 }
5911 return r;
5912 }
5913
5914 int
5915 i386_parse_name (char *name, expressionS *e, char *nextcharP)
5916 {
5917 const reg_entry *r;
5918 char *end = input_line_pointer;
5919
5920 *end = *nextcharP;
5921 r = parse_register (name, &input_line_pointer);
5922 if (r && end <= input_line_pointer)
5923 {
5924 *nextcharP = *input_line_pointer;
5925 *input_line_pointer = 0;
5926 e->X_op = O_register;
5927 e->X_add_number = r - i386_regtab;
5928 return 1;
5929 }
5930 input_line_pointer = end;
5931 *end = 0;
5932 return 0;
5933 }
5934
5935 void
5936 md_operand (expressionS *e)
5937 {
5938 if (*input_line_pointer == REGISTER_PREFIX)
5939 {
5940 char *end;
5941 const reg_entry *r = parse_real_register (input_line_pointer, &end);
5942
5943 if (r)
5944 {
5945 e->X_op = O_register;
5946 e->X_add_number = r - i386_regtab;
5947 input_line_pointer = end;
5948 }
5949 }
5950 }
5951
5952 \f
5953 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5954 const char *md_shortopts = "kVQ:sqn";
5955 #else
5956 const char *md_shortopts = "qn";
5957 #endif
5958
5959 #define OPTION_32 (OPTION_MD_BASE + 0)
5960 #define OPTION_64 (OPTION_MD_BASE + 1)
5961 #define OPTION_DIVIDE (OPTION_MD_BASE + 2)
5962 #define OPTION_MARCH (OPTION_MD_BASE + 3)
5963 #define OPTION_MTUNE (OPTION_MD_BASE + 4)
5964
5965 struct option md_longopts[] =
5966 {
5967 {"32", no_argument, NULL, OPTION_32},
5968 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
5969 {"64", no_argument, NULL, OPTION_64},
5970 #endif
5971 {"divide", no_argument, NULL, OPTION_DIVIDE},
5972 {"march", required_argument, NULL, OPTION_MARCH},
5973 {"mtune", required_argument, NULL, OPTION_MTUNE},
5974 {NULL, no_argument, NULL, 0}
5975 };
5976 size_t md_longopts_size = sizeof (md_longopts);
5977
5978 int
5979 md_parse_option (int c, char *arg)
5980 {
5981 unsigned int i;
5982
5983 switch (c)
5984 {
5985 case 'n':
5986 optimize_align_code = 0;
5987 break;
5988
5989 case 'q':
5990 quiet_warnings = 1;
5991 break;
5992
5993 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
5994 /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
5995 should be emitted or not. FIXME: Not implemented. */
5996 case 'Q':
5997 break;
5998
5999 /* -V: SVR4 argument to print version ID. */
6000 case 'V':
6001 print_version_id ();
6002 break;
6003
6004 /* -k: Ignore for FreeBSD compatibility. */
6005 case 'k':
6006 break;
6007
6008 case 's':
6009 /* -s: On i386 Solaris, this tells the native assembler to use
6010 .stab instead of .stab.excl. We always use .stab anyhow. */
6011 break;
6012 #endif
6013 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
6014 case OPTION_64:
6015 {
6016 const char **list, **l;
6017
6018 list = bfd_target_list ();
6019 for (l = list; *l != NULL; l++)
6020 if (CONST_STRNEQ (*l, "elf64-x86-64")
6021 || strcmp (*l, "coff-x86-64") == 0
6022 || strcmp (*l, "pe-x86-64") == 0
6023 || strcmp (*l, "pei-x86-64") == 0)
6024 {
6025 default_arch = "x86_64";
6026 break;
6027 }
6028 if (*l == NULL)
6029 as_fatal (_("No compiled in support for x86_64"));
6030 free (list);
6031 }
6032 break;
6033 #endif
6034
6035 case OPTION_32:
6036 default_arch = "i386";
6037 break;
6038
6039 case OPTION_DIVIDE:
6040 #ifdef SVR4_COMMENT_CHARS
6041 {
6042 char *n, *t;
6043 const char *s;
6044
6045 n = (char *) xmalloc (strlen (i386_comment_chars) + 1);
6046 t = n;
6047 for (s = i386_comment_chars; *s != '\0'; s++)
6048 if (*s != '/')
6049 *t++ = *s;
6050 *t = '\0';
6051 i386_comment_chars = n;
6052 }
6053 #endif
6054 break;
6055
6056 case OPTION_MARCH:
6057 if (*arg == '.')
6058 as_fatal (_("Invalid -march= option: `%s'"), arg);
6059 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
6060 {
6061 if (strcmp (arg, cpu_arch [i].name) == 0)
6062 {
6063 cpu_arch_isa = cpu_arch[i].type;
6064 cpu_arch_isa_flags = cpu_arch[i].flags;
6065 if (!cpu_arch_tune_set)
6066 {
6067 cpu_arch_tune = cpu_arch_isa;
6068 cpu_arch_tune_flags = cpu_arch_isa_flags;
6069 }
6070 break;
6071 }
6072 }
6073 if (i >= ARRAY_SIZE (cpu_arch))
6074 as_fatal (_("Invalid -march= option: `%s'"), arg);
6075 break;
6076
6077 case OPTION_MTUNE:
6078 if (*arg == '.')
6079 as_fatal (_("Invalid -mtune= option: `%s'"), arg);
6080 for (i = 0; i < ARRAY_SIZE (cpu_arch); i++)
6081 {
6082 if (strcmp (arg, cpu_arch [i].name) == 0)
6083 {
6084 cpu_arch_tune_set = 1;
6085 cpu_arch_tune = cpu_arch [i].type;
6086 cpu_arch_tune_flags = cpu_arch[i].flags;
6087 break;
6088 }
6089 }
6090 if (i >= ARRAY_SIZE (cpu_arch))
6091 as_fatal (_("Invalid -mtune= option: `%s'"), arg);
6092 break;
6093
6094 default:
6095 return 0;
6096 }
6097 return 1;
6098 }
6099
6100 void
6101 md_show_usage (stream)
6102 FILE *stream;
6103 {
6104 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
6105 fprintf (stream, _("\
6106 -Q ignored\n\
6107 -V print assembler version number\n\
6108 -k ignored\n"));
6109 #endif
6110 fprintf (stream, _("\
6111 -n Do not optimize code alignment\n\
6112 -q quieten some warnings\n"));
6113 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
6114 fprintf (stream, _("\
6115 -s ignored\n"));
6116 #endif
6117 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined(TE_PEP)
6118 fprintf (stream, _("\
6119 --32/--64 generate 32bit/64bit code\n"));
6120 #endif
6121 #ifdef SVR4_COMMENT_CHARS
6122 fprintf (stream, _("\
6123 --divide do not treat `/' as a comment character\n"));
6124 #else
6125 fprintf (stream, _("\
6126 --divide ignored\n"));
6127 #endif
6128 fprintf (stream, _("\
6129 -march=CPU/-mtune=CPU generate code/optimize for CPU, where CPU is one of:\n\
6130 i386, i486, pentium, pentiumpro, pentium4, nocona,\n\
6131 core, core2, k6, athlon, k8, generic32, generic64\n"));
6132
6133 }
6134
6135 #if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
6136 || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || defined (TE_PEP))
6137
6138 /* Pick the target format to use. */
6139
6140 const char *
6141 i386_target_format (void)
6142 {
6143 if (!strcmp (default_arch, "x86_64"))
6144 {
6145 set_code_flag (CODE_64BIT);
6146 if (cpu_arch_isa_flags == 0)
6147 cpu_arch_isa_flags = Cpu186|Cpu286|Cpu386|Cpu486
6148 |Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2
6149 |CpuSSE|CpuSSE2;
6150 if (cpu_arch_tune_flags == 0)
6151 cpu_arch_tune_flags = Cpu186|Cpu286|Cpu386|Cpu486
6152 |Cpu586|Cpu686|CpuP4|CpuMMX|CpuMMX2
6153 |CpuSSE|CpuSSE2;
6154 }
6155 else if (!strcmp (default_arch, "i386"))
6156 {
6157 set_code_flag (CODE_32BIT);
6158 if (cpu_arch_isa_flags == 0)
6159 cpu_arch_isa_flags = Cpu186|Cpu286|Cpu386;
6160 if (cpu_arch_tune_flags == 0)
6161 cpu_arch_tune_flags = Cpu186|Cpu286|Cpu386;
6162 }
6163 else
6164 as_fatal (_("Unknown architecture"));
6165 switch (OUTPUT_FLAVOR)
6166 {
6167 #ifdef TE_PEP
6168 case bfd_target_coff_flavour:
6169 return flag_code == CODE_64BIT ? COFF_TARGET_FORMAT : "coff-i386";
6170 break;
6171 #endif
6172 #ifdef OBJ_MAYBE_AOUT
6173 case bfd_target_aout_flavour:
6174 return AOUT_TARGET_FORMAT;
6175 #endif
6176 #ifdef OBJ_MAYBE_COFF
6177 case bfd_target_coff_flavour:
6178 return "coff-i386";
6179 #endif
6180 #if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
6181 case bfd_target_elf_flavour:
6182 {
6183 if (flag_code == CODE_64BIT)
6184 {
6185 object_64bit = 1;
6186 use_rela_relocations = 1;
6187 }
6188 return flag_code == CODE_64BIT ? ELF_TARGET_FORMAT64 : ELF_TARGET_FORMAT;
6189 }
6190 #endif
6191 default:
6192 abort ();
6193 return NULL;
6194 }
6195 }
6196
6197 #endif /* OBJ_MAYBE_ more than one */
6198
6199 #if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
6200 void
6201 i386_elf_emit_arch_note (void)
6202 {
6203 if (IS_ELF && cpu_arch_name != NULL)
6204 {
6205 char *p;
6206 asection *seg = now_seg;
6207 subsegT subseg = now_subseg;
6208 Elf_Internal_Note i_note;
6209 Elf_External_Note e_note;
6210 asection *note_secp;
6211 int len;
6212
6213 /* Create the .note section. */
6214 note_secp = subseg_new (".note", 0);
6215 bfd_set_section_flags (stdoutput,
6216 note_secp,
6217 SEC_HAS_CONTENTS | SEC_READONLY);
6218
6219 /* Process the arch string. */
6220 len = strlen (cpu_arch_name);
6221
6222 i_note.namesz = len + 1;
6223 i_note.descsz = 0;
6224 i_note.type = NT_ARCH;
6225 p = frag_more (sizeof (e_note.namesz));
6226 md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
6227 p = frag_more (sizeof (e_note.descsz));
6228 md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
6229 p = frag_more (sizeof (e_note.type));
6230 md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
6231 p = frag_more (len + 1);
6232 strcpy (p, cpu_arch_name);
6233
6234 frag_align (2, 0, 0);
6235
6236 subseg_set (seg, subseg);
6237 }
6238 }
6239 #endif
6240 \f
6241 symbolS *
6242 md_undefined_symbol (name)
6243 char *name;
6244 {
6245 if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
6246 && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
6247 && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
6248 && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
6249 {
6250 if (!GOT_symbol)
6251 {
6252 if (symbol_find (name))
6253 as_bad (_("GOT already in symbol table"));
6254 GOT_symbol = symbol_new (name, undefined_section,
6255 (valueT) 0, &zero_address_frag);
6256 };
6257 return GOT_symbol;
6258 }
6259 return 0;
6260 }
6261
6262 /* Round up a section size to the appropriate boundary. */
6263
6264 valueT
6265 md_section_align (segment, size)
6266 segT segment ATTRIBUTE_UNUSED;
6267 valueT size;
6268 {
6269 #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
6270 if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
6271 {
6272 /* For a.out, force the section size to be aligned. If we don't do
6273 this, BFD will align it for us, but it will not write out the
6274 final bytes of the section. This may be a bug in BFD, but it is
6275 easier to fix it here since that is how the other a.out targets
6276 work. */
6277 int align;
6278
6279 align = bfd_get_section_alignment (stdoutput, segment);
6280 size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
6281 }
6282 #endif
6283
6284 return size;
6285 }
6286
6287 /* On the i386, PC-relative offsets are relative to the start of the
6288 next instruction. That is, the address of the offset, plus its
6289 size, since the offset is always the last part of the insn. */
6290
6291 long
6292 md_pcrel_from (fixS *fixP)
6293 {
6294 return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
6295 }
6296
6297 #ifndef I386COFF
6298
6299 static void
6300 s_bss (int ignore ATTRIBUTE_UNUSED)
6301 {
6302 int temp;
6303
6304 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
6305 if (IS_ELF)
6306 obj_elf_section_change_hook ();
6307 #endif
6308 temp = get_absolute_expression ();
6309 subseg_set (bss_section, (subsegT) temp);
6310 demand_empty_rest_of_line ();
6311 }
6312
6313 #endif
6314
6315 void
6316 i386_validate_fix (fixS *fixp)
6317 {
6318 if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
6319 {
6320 if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
6321 {
6322 if (!object_64bit)
6323 abort ();
6324 fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
6325 }
6326 else
6327 {
6328 if (!object_64bit)
6329 fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
6330 else
6331 fixp->fx_r_type = BFD_RELOC_X86_64_GOTOFF64;
6332 }
6333 fixp->fx_subsy = 0;
6334 }
6335 }
6336
6337 arelent *
6338 tc_gen_reloc (section, fixp)
6339 asection *section ATTRIBUTE_UNUSED;
6340 fixS *fixp;
6341 {
6342 arelent *rel;
6343 bfd_reloc_code_real_type code;
6344
6345 switch (fixp->fx_r_type)
6346 {
6347 case BFD_RELOC_X86_64_PLT32:
6348 case BFD_RELOC_X86_64_GOT32:
6349 case BFD_RELOC_X86_64_GOTPCREL:
6350 case BFD_RELOC_386_PLT32:
6351 case BFD_RELOC_386_GOT32:
6352 case BFD_RELOC_386_GOTOFF:
6353 case BFD_RELOC_386_GOTPC:
6354 case BFD_RELOC_386_TLS_GD:
6355 case BFD_RELOC_386_TLS_LDM:
6356 case BFD_RELOC_386_TLS_LDO_32:
6357 case BFD_RELOC_386_TLS_IE_32:
6358 case BFD_RELOC_386_TLS_IE:
6359 case BFD_RELOC_386_TLS_GOTIE:
6360 case BFD_RELOC_386_TLS_LE_32:
6361 case BFD_RELOC_386_TLS_LE:
6362 case BFD_RELOC_386_TLS_GOTDESC:
6363 case BFD_RELOC_386_TLS_DESC_CALL:
6364 case BFD_RELOC_X86_64_TLSGD:
6365 case BFD_RELOC_X86_64_TLSLD:
6366 case BFD_RELOC_X86_64_DTPOFF32:
6367 case BFD_RELOC_X86_64_DTPOFF64:
6368 case BFD_RELOC_X86_64_GOTTPOFF:
6369 case BFD_RELOC_X86_64_TPOFF32:
6370 case BFD_RELOC_X86_64_TPOFF64:
6371 case BFD_RELOC_X86_64_GOTOFF64:
6372 case BFD_RELOC_X86_64_GOTPC32:
6373 case BFD_RELOC_X86_64_GOT64:
6374 case BFD_RELOC_X86_64_GOTPCREL64:
6375 case BFD_RELOC_X86_64_GOTPC64:
6376 case BFD_RELOC_X86_64_GOTPLT64:
6377 case BFD_RELOC_X86_64_PLTOFF64:
6378 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
6379 case BFD_RELOC_X86_64_TLSDESC_CALL:
6380 case BFD_RELOC_RVA:
6381 case BFD_RELOC_VTABLE_ENTRY:
6382 case BFD_RELOC_VTABLE_INHERIT:
6383 #ifdef TE_PE
6384 case BFD_RELOC_32_SECREL:
6385 #endif
6386 code = fixp->fx_r_type;
6387 break;
6388 case BFD_RELOC_X86_64_32S:
6389 if (!fixp->fx_pcrel)
6390 {
6391 /* Don't turn BFD_RELOC_X86_64_32S into BFD_RELOC_32. */
6392 code = fixp->fx_r_type;
6393 break;
6394 }
6395 default:
6396 if (fixp->fx_pcrel)
6397 {
6398 switch (fixp->fx_size)
6399 {
6400 default:
6401 as_bad_where (fixp->fx_file, fixp->fx_line,
6402 _("can not do %d byte pc-relative relocation"),
6403 fixp->fx_size);
6404 code = BFD_RELOC_32_PCREL;
6405 break;
6406 case 1: code = BFD_RELOC_8_PCREL; break;
6407 case 2: code = BFD_RELOC_16_PCREL; break;
6408 case 4: code = BFD_RELOC_32_PCREL; break;
6409 #ifdef BFD64
6410 case 8: code = BFD_RELOC_64_PCREL; break;
6411 #endif
6412 }
6413 }
6414 else
6415 {
6416 switch (fixp->fx_size)
6417 {
6418 default:
6419 as_bad_where (fixp->fx_file, fixp->fx_line,
6420 _("can not do %d byte relocation"),
6421 fixp->fx_size);
6422 code = BFD_RELOC_32;
6423 break;
6424 case 1: code = BFD_RELOC_8; break;
6425 case 2: code = BFD_RELOC_16; break;
6426 case 4: code = BFD_RELOC_32; break;
6427 #ifdef BFD64
6428 case 8: code = BFD_RELOC_64; break;
6429 #endif
6430 }
6431 }
6432 break;
6433 }
6434
6435 if ((code == BFD_RELOC_32
6436 || code == BFD_RELOC_32_PCREL
6437 || code == BFD_RELOC_X86_64_32S)
6438 && GOT_symbol
6439 && fixp->fx_addsy == GOT_symbol)
6440 {
6441 if (!object_64bit)
6442 code = BFD_RELOC_386_GOTPC;
6443 else
6444 code = BFD_RELOC_X86_64_GOTPC32;
6445 }
6446 if ((code == BFD_RELOC_64 || code == BFD_RELOC_64_PCREL)
6447 && GOT_symbol
6448 && fixp->fx_addsy == GOT_symbol)
6449 {
6450 code = BFD_RELOC_X86_64_GOTPC64;
6451 }
6452
6453 rel = (arelent *) xmalloc (sizeof (arelent));
6454 rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
6455 *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
6456
6457 rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
6458
6459 if (!use_rela_relocations)
6460 {
6461 /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
6462 vtable entry to be used in the relocation's section offset. */
6463 if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
6464 rel->address = fixp->fx_offset;
6465
6466 rel->addend = 0;
6467 }
6468 /* Use the rela in 64bit mode. */
6469 else
6470 {
6471 if (!fixp->fx_pcrel)
6472 rel->addend = fixp->fx_offset;
6473 else
6474 switch (code)
6475 {
6476 case BFD_RELOC_X86_64_PLT32:
6477 case BFD_RELOC_X86_64_GOT32:
6478 case BFD_RELOC_X86_64_GOTPCREL:
6479 case BFD_RELOC_X86_64_TLSGD:
6480 case BFD_RELOC_X86_64_TLSLD:
6481 case BFD_RELOC_X86_64_GOTTPOFF:
6482 case BFD_RELOC_X86_64_GOTPC32_TLSDESC:
6483 case BFD_RELOC_X86_64_TLSDESC_CALL:
6484 rel->addend = fixp->fx_offset - fixp->fx_size;
6485 break;
6486 default:
6487 rel->addend = (section->vma
6488 - fixp->fx_size
6489 + fixp->fx_addnumber
6490 + md_pcrel_from (fixp));
6491 break;
6492 }
6493 }
6494
6495 rel->howto = bfd_reloc_type_lookup (stdoutput, code);
6496 if (rel->howto == NULL)
6497 {
6498 as_bad_where (fixp->fx_file, fixp->fx_line,
6499 _("cannot represent relocation type %s"),
6500 bfd_get_reloc_code_name (code));
6501 /* Set howto to a garbage value so that we can keep going. */
6502 rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
6503 assert (rel->howto != NULL);
6504 }
6505
6506 return rel;
6507 }
6508
6509 \f
6510 /* Parse operands using Intel syntax. This implements a recursive descent
6511 parser based on the BNF grammar published in Appendix B of the MASM 6.1
6512 Programmer's Guide.
6513
6514 FIXME: We do not recognize the full operand grammar defined in the MASM
6515 documentation. In particular, all the structure/union and
6516 high-level macro operands are missing.
6517
6518 Uppercase words are terminals, lower case words are non-terminals.
6519 Objects surrounded by double brackets '[[' ']]' are optional. Vertical
6520 bars '|' denote choices. Most grammar productions are implemented in
6521 functions called 'intel_<production>'.
6522
6523 Initial production is 'expr'.
6524
6525 addOp + | -
6526
6527 alpha [a-zA-Z]
6528
6529 binOp & | AND | \| | OR | ^ | XOR
6530
6531 byteRegister AL | AH | BL | BH | CL | CH | DL | DH
6532
6533 constant digits [[ radixOverride ]]
6534
6535 dataType BYTE | WORD | DWORD | FWORD | QWORD | TBYTE | OWORD | XMMWORD
6536
6537 digits decdigit
6538 | digits decdigit
6539 | digits hexdigit
6540
6541 decdigit [0-9]
6542
6543 e04 e04 addOp e05
6544 | e05
6545
6546 e05 e05 binOp e06
6547 | e06
6548
6549 e06 e06 mulOp e09
6550 | e09
6551
6552 e09 OFFSET e10
6553 | SHORT e10
6554 | + e10
6555 | - e10
6556 | ~ e10
6557 | NOT e10
6558 | e09 PTR e10
6559 | e09 : e10
6560 | e10
6561
6562 e10 e10 [ expr ]
6563 | e11
6564
6565 e11 ( expr )
6566 | [ expr ]
6567 | constant
6568 | dataType
6569 | id
6570 | $
6571 | register
6572
6573 => expr expr cmpOp e04
6574 | e04
6575
6576 gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
6577 | BP | EBP | SP | ESP | DI | EDI | SI | ESI
6578
6579 hexdigit a | b | c | d | e | f
6580 | A | B | C | D | E | F
6581
6582 id alpha
6583 | id alpha
6584 | id decdigit
6585
6586 mulOp * | / | % | MOD | << | SHL | >> | SHR
6587
6588 quote " | '
6589
6590 register specialRegister
6591 | gpRegister
6592 | byteRegister
6593
6594 segmentRegister CS | DS | ES | FS | GS | SS
6595
6596 specialRegister CR0 | CR2 | CR3 | CR4
6597 | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
6598 | TR3 | TR4 | TR5 | TR6 | TR7
6599
6600 We simplify the grammar in obvious places (e.g., register parsing is
6601 done by calling parse_register) and eliminate immediate left recursion
6602 to implement a recursive-descent parser.
6603
6604 expr e04 expr'
6605
6606 expr' cmpOp e04 expr'
6607 | Empty
6608
6609 e04 e05 e04'
6610
6611 e04' addOp e05 e04'
6612 | Empty
6613
6614 e05 e06 e05'
6615
6616 e05' binOp e06 e05'
6617 | Empty
6618
6619 e06 e09 e06'
6620
6621 e06' mulOp e09 e06'
6622 | Empty
6623
6624 e09 OFFSET e10 e09'
6625 | SHORT e10'
6626 | + e10'
6627 | - e10'
6628 | ~ e10'
6629 | NOT e10'
6630 | e10 e09'
6631
6632 e09' PTR e10 e09'
6633 | : e10 e09'
6634 | Empty
6635
6636 e10 e11 e10'
6637
6638 e10' [ expr ] e10'
6639 | Empty
6640
6641 e11 ( expr )
6642 | [ expr ]
6643 | BYTE
6644 | WORD
6645 | DWORD
6646 | FWORD
6647 | QWORD
6648 | TBYTE
6649 | OWORD
6650 | XMMWORD
6651 | .
6652 | $
6653 | register
6654 | id
6655 | constant */
6656
6657 /* Parsing structure for the intel syntax parser. Used to implement the
6658 semantic actions for the operand grammar. */
6659 struct intel_parser_s
6660 {
6661 char *op_string; /* The string being parsed. */
6662 int got_a_float; /* Whether the operand is a float. */
6663 int op_modifier; /* Operand modifier. */
6664 int is_mem; /* 1 if operand is memory reference. */
6665 int in_offset; /* >=1 if parsing operand of offset. */
6666 int in_bracket; /* >=1 if parsing operand in brackets. */
6667 const reg_entry *reg; /* Last register reference found. */
6668 char *disp; /* Displacement string being built. */
6669 char *next_operand; /* Resume point when splitting operands. */
6670 };
6671
6672 static struct intel_parser_s intel_parser;
6673
6674 /* Token structure for parsing intel syntax. */
6675 struct intel_token
6676 {
6677 int code; /* Token code. */
6678 const reg_entry *reg; /* Register entry for register tokens. */
6679 char *str; /* String representation. */
6680 };
6681
6682 static struct intel_token cur_token, prev_token;
6683
6684 /* Token codes for the intel parser. Since T_SHORT is already used
6685 by COFF, undefine it first to prevent a warning. */
6686 #define T_NIL -1
6687 #define T_CONST 1
6688 #define T_REG 2
6689 #define T_BYTE 3
6690 #define T_WORD 4
6691 #define T_DWORD 5
6692 #define T_FWORD 6
6693 #define T_QWORD 7
6694 #define T_TBYTE 8
6695 #define T_XMMWORD 9
6696 #undef T_SHORT
6697 #define T_SHORT 10
6698 #define T_OFFSET 11
6699 #define T_PTR 12
6700 #define T_ID 13
6701 #define T_SHL 14
6702 #define T_SHR 15
6703
6704 /* Prototypes for intel parser functions. */
6705 static int intel_match_token (int);
6706 static void intel_putback_token (void);
6707 static void intel_get_token (void);
6708 static int intel_expr (void);
6709 static int intel_e04 (void);
6710 static int intel_e05 (void);
6711 static int intel_e06 (void);
6712 static int intel_e09 (void);
6713 static int intel_e10 (void);
6714 static int intel_e11 (void);
6715
6716 static int
6717 i386_intel_operand (char *operand_string, int got_a_float)
6718 {
6719 int ret;
6720 char *p;
6721
6722 p = intel_parser.op_string = xstrdup (operand_string);
6723 intel_parser.disp = (char *) xmalloc (strlen (operand_string) + 1);
6724
6725 for (;;)
6726 {
6727 /* Initialize token holders. */
6728 cur_token.code = prev_token.code = T_NIL;
6729 cur_token.reg = prev_token.reg = NULL;
6730 cur_token.str = prev_token.str = NULL;
6731
6732 /* Initialize parser structure. */
6733 intel_parser.got_a_float = got_a_float;
6734 intel_parser.op_modifier = 0;
6735 intel_parser.is_mem = 0;
6736 intel_parser.in_offset = 0;
6737 intel_parser.in_bracket = 0;
6738 intel_parser.reg = NULL;
6739 intel_parser.disp[0] = '\0';
6740 intel_parser.next_operand = NULL;
6741
6742 /* Read the first token and start the parser. */
6743 intel_get_token ();
6744 ret = intel_expr ();
6745
6746 if (!ret)
6747 break;
6748
6749 if (cur_token.code != T_NIL)
6750 {
6751 as_bad (_("invalid operand for '%s' ('%s' unexpected)"),
6752 current_templates->start->name, cur_token.str);
6753 ret = 0;
6754 }
6755 /* If we found a memory reference, hand it over to i386_displacement
6756 to fill in the rest of the operand fields. */
6757 else if (intel_parser.is_mem)
6758 {
6759 if ((i.mem_operands == 1
6760 && (current_templates->start->opcode_modifier & IsString) == 0)
6761 || i.mem_operands == 2)
6762 {
6763 as_bad (_("too many memory references for '%s'"),
6764 current_templates->start->name);
6765 ret = 0;
6766 }
6767 else
6768 {
6769 char *s = intel_parser.disp;
6770 i.mem_operands++;
6771
6772 if (!quiet_warnings && intel_parser.is_mem < 0)
6773 /* See the comments in intel_bracket_expr. */
6774 as_warn (_("Treating `%s' as memory reference"), operand_string);
6775
6776 /* Add the displacement expression. */
6777 if (*s != '\0')
6778 ret = i386_displacement (s, s + strlen (s));
6779 if (ret)
6780 {
6781 /* Swap base and index in 16-bit memory operands like
6782 [si+bx]. Since i386_index_check is also used in AT&T
6783 mode we have to do that here. */
6784 if (i.base_reg
6785 && i.index_reg
6786 && (i.base_reg->reg_type & Reg16)
6787 && (i.index_reg->reg_type & Reg16)
6788 && i.base_reg->reg_num >= 6
6789 && i.index_reg->reg_num < 6)
6790 {
6791 const reg_entry *base = i.index_reg;
6792
6793 i.index_reg = i.base_reg;
6794 i.base_reg = base;
6795 }
6796 ret = i386_index_check (operand_string);
6797 }
6798 }
6799 }
6800
6801 /* Constant and OFFSET expressions are handled by i386_immediate. */
6802 else if ((intel_parser.op_modifier & (1 << T_OFFSET))
6803 || intel_parser.reg == NULL)
6804 ret = i386_immediate (intel_parser.disp);
6805
6806 if (intel_parser.next_operand && this_operand >= MAX_OPERANDS - 1)
6807 ret = 0;
6808 if (!ret || !intel_parser.next_operand)
6809 break;
6810 intel_parser.op_string = intel_parser.next_operand;
6811 this_operand = i.operands++;
6812 }
6813
6814 free (p);
6815 free (intel_parser.disp);
6816
6817 return ret;
6818 }
6819
6820 #define NUM_ADDRESS_REGS (!!i.base_reg + !!i.index_reg)
6821
6822 /* expr e04 expr'
6823
6824 expr' cmpOp e04 expr'
6825 | Empty */
6826 static int
6827 intel_expr (void)
6828 {
6829 /* XXX Implement the comparison operators. */
6830 return intel_e04 ();
6831 }
6832
6833 /* e04 e05 e04'
6834
6835 e04' addOp e05 e04'
6836 | Empty */
6837 static int
6838 intel_e04 (void)
6839 {
6840 int nregs = -1;
6841
6842 for (;;)
6843 {
6844 if (!intel_e05())
6845 return 0;
6846
6847 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6848 i.base_reg = i386_regtab + REGNAM_AL; /* al is invalid as base */
6849
6850 if (cur_token.code == '+')
6851 nregs = -1;
6852 else if (cur_token.code == '-')
6853 nregs = NUM_ADDRESS_REGS;
6854 else
6855 return 1;
6856
6857 strcat (intel_parser.disp, cur_token.str);
6858 intel_match_token (cur_token.code);
6859 }
6860 }
6861
6862 /* e05 e06 e05'
6863
6864 e05' binOp e06 e05'
6865 | Empty */
6866 static int
6867 intel_e05 (void)
6868 {
6869 int nregs = ~NUM_ADDRESS_REGS;
6870
6871 for (;;)
6872 {
6873 if (!intel_e06())
6874 return 0;
6875
6876 if (cur_token.code == '&'
6877 || cur_token.code == '|'
6878 || cur_token.code == '^')
6879 {
6880 char str[2];
6881
6882 str[0] = cur_token.code;
6883 str[1] = 0;
6884 strcat (intel_parser.disp, str);
6885 }
6886 else
6887 break;
6888
6889 intel_match_token (cur_token.code);
6890
6891 if (nregs < 0)
6892 nregs = ~nregs;
6893 }
6894 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6895 i.base_reg = i386_regtab + REGNAM_AL + 1; /* cl is invalid as base */
6896 return 1;
6897 }
6898
6899 /* e06 e09 e06'
6900
6901 e06' mulOp e09 e06'
6902 | Empty */
6903 static int
6904 intel_e06 (void)
6905 {
6906 int nregs = ~NUM_ADDRESS_REGS;
6907
6908 for (;;)
6909 {
6910 if (!intel_e09())
6911 return 0;
6912
6913 if (cur_token.code == '*'
6914 || cur_token.code == '/'
6915 || cur_token.code == '%')
6916 {
6917 char str[2];
6918
6919 str[0] = cur_token.code;
6920 str[1] = 0;
6921 strcat (intel_parser.disp, str);
6922 }
6923 else if (cur_token.code == T_SHL)
6924 strcat (intel_parser.disp, "<<");
6925 else if (cur_token.code == T_SHR)
6926 strcat (intel_parser.disp, ">>");
6927 else
6928 break;
6929
6930 intel_match_token (cur_token.code);
6931
6932 if (nregs < 0)
6933 nregs = ~nregs;
6934 }
6935 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
6936 i.base_reg = i386_regtab + REGNAM_AL + 2; /* dl is invalid as base */
6937 return 1;
6938 }
6939
6940 /* e09 OFFSET e09
6941 | SHORT e09
6942 | + e09
6943 | - e09
6944 | ~ e09
6945 | NOT e09
6946 | e10 e09'
6947
6948 e09' PTR e10 e09'
6949 | : e10 e09'
6950 | Empty */
6951 static int
6952 intel_e09 (void)
6953 {
6954 int nregs = ~NUM_ADDRESS_REGS;
6955 int in_offset = 0;
6956
6957 for (;;)
6958 {
6959 /* Don't consume constants here. */
6960 if (cur_token.code == '+' || cur_token.code == '-')
6961 {
6962 /* Need to look one token ahead - if the next token
6963 is a constant, the current token is its sign. */
6964 int next_code;
6965
6966 intel_match_token (cur_token.code);
6967 next_code = cur_token.code;
6968 intel_putback_token ();
6969 if (next_code == T_CONST)
6970 break;
6971 }
6972
6973 /* e09 OFFSET e09 */
6974 if (cur_token.code == T_OFFSET)
6975 {
6976 if (!in_offset++)
6977 ++intel_parser.in_offset;
6978 }
6979
6980 /* e09 SHORT e09 */
6981 else if (cur_token.code == T_SHORT)
6982 intel_parser.op_modifier |= 1 << T_SHORT;
6983
6984 /* e09 + e09 */
6985 else if (cur_token.code == '+')
6986 strcat (intel_parser.disp, "+");
6987
6988 /* e09 - e09
6989 | ~ e09
6990 | NOT e09 */
6991 else if (cur_token.code == '-' || cur_token.code == '~')
6992 {
6993 char str[2];
6994
6995 if (nregs < 0)
6996 nregs = ~nregs;
6997 str[0] = cur_token.code;
6998 str[1] = 0;
6999 strcat (intel_parser.disp, str);
7000 }
7001
7002 /* e09 e10 e09' */
7003 else
7004 break;
7005
7006 intel_match_token (cur_token.code);
7007 }
7008
7009 for (;;)
7010 {
7011 if (!intel_e10 ())
7012 return 0;
7013
7014 /* e09' PTR e10 e09' */
7015 if (cur_token.code == T_PTR)
7016 {
7017 char suffix;
7018
7019 if (prev_token.code == T_BYTE)
7020 suffix = BYTE_MNEM_SUFFIX;
7021
7022 else if (prev_token.code == T_WORD)
7023 {
7024 if (current_templates->start->name[0] == 'l'
7025 && current_templates->start->name[2] == 's'
7026 && current_templates->start->name[3] == 0)
7027 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
7028 else if (intel_parser.got_a_float == 2) /* "fi..." */
7029 suffix = SHORT_MNEM_SUFFIX;
7030 else
7031 suffix = WORD_MNEM_SUFFIX;
7032 }
7033
7034 else if (prev_token.code == T_DWORD)
7035 {
7036 if (current_templates->start->name[0] == 'l'
7037 && current_templates->start->name[2] == 's'
7038 && current_templates->start->name[3] == 0)
7039 suffix = WORD_MNEM_SUFFIX;
7040 else if (flag_code == CODE_16BIT
7041 && (current_templates->start->opcode_modifier
7042 & (Jump | JumpDword)))
7043 suffix = LONG_DOUBLE_MNEM_SUFFIX;
7044 else if (intel_parser.got_a_float == 1) /* "f..." */
7045 suffix = SHORT_MNEM_SUFFIX;
7046 else
7047 suffix = LONG_MNEM_SUFFIX;
7048 }
7049
7050 else if (prev_token.code == T_FWORD)
7051 {
7052 if (current_templates->start->name[0] == 'l'
7053 && current_templates->start->name[2] == 's'
7054 && current_templates->start->name[3] == 0)
7055 suffix = LONG_MNEM_SUFFIX;
7056 else if (!intel_parser.got_a_float)
7057 {
7058 if (flag_code == CODE_16BIT)
7059 add_prefix (DATA_PREFIX_OPCODE);
7060 suffix = LONG_DOUBLE_MNEM_SUFFIX;
7061 }
7062 else
7063 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
7064 }
7065
7066 else if (prev_token.code == T_QWORD)
7067 {
7068 if (intel_parser.got_a_float == 1) /* "f..." */
7069 suffix = LONG_MNEM_SUFFIX;
7070 else
7071 suffix = QWORD_MNEM_SUFFIX;
7072 }
7073
7074 else if (prev_token.code == T_TBYTE)
7075 {
7076 if (intel_parser.got_a_float == 1)
7077 suffix = LONG_DOUBLE_MNEM_SUFFIX;
7078 else
7079 suffix = BYTE_MNEM_SUFFIX; /* so it will cause an error */
7080 }
7081
7082 else if (prev_token.code == T_XMMWORD)
7083 {
7084 /* XXX ignored for now, but accepted since gcc uses it */
7085 suffix = 0;
7086 }
7087
7088 else
7089 {
7090 as_bad (_("Unknown operand modifier `%s'"), prev_token.str);
7091 return 0;
7092 }
7093
7094 /* Operands for jump/call using 'ptr' notation denote absolute
7095 addresses. */
7096 if (current_templates->start->opcode_modifier & (Jump | JumpDword))
7097 i.types[this_operand] |= JumpAbsolute;
7098
7099 if (current_templates->start->base_opcode == 0x8d /* lea */)
7100 ;
7101 else if (!i.suffix)
7102 i.suffix = suffix;
7103 else if (i.suffix != suffix)
7104 {
7105 as_bad (_("Conflicting operand modifiers"));
7106 return 0;
7107 }
7108
7109 }
7110
7111 /* e09' : e10 e09' */
7112 else if (cur_token.code == ':')
7113 {
7114 if (prev_token.code != T_REG)
7115 {
7116 /* While {call,jmp} SSSS:OOOO is MASM syntax only when SSSS is a
7117 segment/group identifier (which we don't have), using comma
7118 as the operand separator there is even less consistent, since
7119 there all branches only have a single operand. */
7120 if (this_operand != 0
7121 || intel_parser.in_offset
7122 || intel_parser.in_bracket
7123 || (!(current_templates->start->opcode_modifier
7124 & (Jump|JumpDword|JumpInterSegment))
7125 && !(current_templates->start->operand_types[0]
7126 & JumpAbsolute)))
7127 return intel_match_token (T_NIL);
7128 /* Remember the start of the 2nd operand and terminate 1st
7129 operand here.
7130 XXX This isn't right, yet (when SSSS:OOOO is right operand of
7131 another expression), but it gets at least the simplest case
7132 (a plain number or symbol on the left side) right. */
7133 intel_parser.next_operand = intel_parser.op_string;
7134 *--intel_parser.op_string = '\0';
7135 return intel_match_token (':');
7136 }
7137 }
7138
7139 /* e09' Empty */
7140 else
7141 break;
7142
7143 intel_match_token (cur_token.code);
7144
7145 }
7146
7147 if (in_offset)
7148 {
7149 --intel_parser.in_offset;
7150 if (nregs < 0)
7151 nregs = ~nregs;
7152 if (NUM_ADDRESS_REGS > nregs)
7153 {
7154 as_bad (_("Invalid operand to `OFFSET'"));
7155 return 0;
7156 }
7157 intel_parser.op_modifier |= 1 << T_OFFSET;
7158 }
7159
7160 if (nregs >= 0 && NUM_ADDRESS_REGS > nregs)
7161 i.base_reg = i386_regtab + REGNAM_AL + 3; /* bl is invalid as base */
7162 return 1;
7163 }
7164
7165 static int
7166 intel_bracket_expr (void)
7167 {
7168 int was_offset = intel_parser.op_modifier & (1 << T_OFFSET);
7169 const char *start = intel_parser.op_string;
7170 int len;
7171
7172 if (i.op[this_operand].regs)
7173 return intel_match_token (T_NIL);
7174
7175 intel_match_token ('[');
7176
7177 /* Mark as a memory operand only if it's not already known to be an
7178 offset expression. If it's an offset expression, we need to keep
7179 the brace in. */
7180 if (!intel_parser.in_offset)
7181 {
7182 ++intel_parser.in_bracket;
7183
7184 /* Operands for jump/call inside brackets denote absolute addresses. */
7185 if (current_templates->start->opcode_modifier & (Jump | JumpDword))
7186 i.types[this_operand] |= JumpAbsolute;
7187
7188 /* Unfortunately gas always diverged from MASM in a respect that can't
7189 be easily fixed without risking to break code sequences likely to be
7190 encountered (the testsuite even check for this): MASM doesn't consider
7191 an expression inside brackets unconditionally as a memory reference.
7192 When that is e.g. a constant, an offset expression, or the sum of the
7193 two, this is still taken as a constant load. gas, however, always
7194 treated these as memory references. As a compromise, we'll try to make
7195 offset expressions inside brackets work the MASM way (since that's
7196 less likely to be found in real world code), but make constants alone
7197 continue to work the traditional gas way. In either case, issue a
7198 warning. */
7199 intel_parser.op_modifier &= ~was_offset;
7200 }
7201 else
7202 strcat (intel_parser.disp, "[");
7203
7204 /* Add a '+' to the displacement string if necessary. */
7205 if (*intel_parser.disp != '\0'
7206 && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
7207 strcat (intel_parser.disp, "+");
7208
7209 if (intel_expr ()
7210 && (len = intel_parser.op_string - start - 1,
7211 intel_match_token (']')))
7212 {
7213 /* Preserve brackets when the operand is an offset expression. */
7214 if (intel_parser.in_offset)
7215 strcat (intel_parser.disp, "]");
7216 else
7217 {
7218 --intel_parser.in_bracket;
7219 if (i.base_reg || i.index_reg)
7220 intel_parser.is_mem = 1;
7221 if (!intel_parser.is_mem)
7222 {
7223 if (!(intel_parser.op_modifier & (1 << T_OFFSET)))
7224 /* Defer the warning until all of the operand was parsed. */
7225 intel_parser.is_mem = -1;
7226 else if (!quiet_warnings)
7227 as_warn (_("`[%.*s]' taken to mean just `%.*s'"),
7228 len, start, len, start);
7229 }
7230 }
7231 intel_parser.op_modifier |= was_offset;
7232
7233 return 1;
7234 }
7235 return 0;
7236 }
7237
7238 /* e10 e11 e10'
7239
7240 e10' [ expr ] e10'
7241 | Empty */
7242 static int
7243 intel_e10 (void)
7244 {
7245 if (!intel_e11 ())
7246 return 0;
7247
7248 while (cur_token.code == '[')
7249 {
7250 if (!intel_bracket_expr ())
7251 return 0;
7252 }
7253
7254 return 1;
7255 }
7256
7257 /* e11 ( expr )
7258 | [ expr ]
7259 | BYTE
7260 | WORD
7261 | DWORD
7262 | FWORD
7263 | QWORD
7264 | TBYTE
7265 | OWORD
7266 | XMMWORD
7267 | $
7268 | .
7269 | register
7270 | id
7271 | constant */
7272 static int
7273 intel_e11 (void)
7274 {
7275 switch (cur_token.code)
7276 {
7277 /* e11 ( expr ) */
7278 case '(':
7279 intel_match_token ('(');
7280 strcat (intel_parser.disp, "(");
7281
7282 if (intel_expr () && intel_match_token (')'))
7283 {
7284 strcat (intel_parser.disp, ")");
7285 return 1;
7286 }
7287 return 0;
7288
7289 /* e11 [ expr ] */
7290 case '[':
7291 return intel_bracket_expr ();
7292
7293 /* e11 $
7294 | . */
7295 case '.':
7296 strcat (intel_parser.disp, cur_token.str);
7297 intel_match_token (cur_token.code);
7298
7299 /* Mark as a memory operand only if it's not already known to be an
7300 offset expression. */
7301 if (!intel_parser.in_offset)
7302 intel_parser.is_mem = 1;
7303
7304 return 1;
7305
7306 /* e11 register */
7307 case T_REG:
7308 {
7309 const reg_entry *reg = intel_parser.reg = cur_token.reg;
7310
7311 intel_match_token (T_REG);
7312
7313 /* Check for segment change. */
7314 if (cur_token.code == ':')
7315 {
7316 if (!(reg->reg_type & (SReg2 | SReg3)))
7317 {
7318 as_bad (_("`%s' is not a valid segment register"),
7319 reg->reg_name);
7320 return 0;
7321 }
7322 else if (i.seg[i.mem_operands])
7323 as_warn (_("Extra segment override ignored"));
7324 else
7325 {
7326 if (!intel_parser.in_offset)
7327 intel_parser.is_mem = 1;
7328 switch (reg->reg_num)
7329 {
7330 case 0:
7331 i.seg[i.mem_operands] = &es;
7332 break;
7333 case 1:
7334 i.seg[i.mem_operands] = &cs;
7335 break;
7336 case 2:
7337 i.seg[i.mem_operands] = &ss;
7338 break;
7339 case 3:
7340 i.seg[i.mem_operands] = &ds;
7341 break;
7342 case 4:
7343 i.seg[i.mem_operands] = &fs;
7344 break;
7345 case 5:
7346 i.seg[i.mem_operands] = &gs;
7347 break;
7348 }
7349 }
7350 }
7351
7352 /* Not a segment register. Check for register scaling. */
7353 else if (cur_token.code == '*')
7354 {
7355 if (!intel_parser.in_bracket)
7356 {
7357 as_bad (_("Register scaling only allowed in memory operands"));
7358 return 0;
7359 }
7360
7361 if (reg->reg_type & Reg16) /* Disallow things like [si*1]. */
7362 reg = i386_regtab + REGNAM_AX + 4; /* sp is invalid as index */
7363 else if (i.index_reg)
7364 reg = i386_regtab + REGNAM_EAX + 4; /* esp is invalid as index */
7365
7366 /* What follows must be a valid scale. */
7367 intel_match_token ('*');
7368 i.index_reg = reg;
7369 i.types[this_operand] |= BaseIndex;
7370
7371 /* Set the scale after setting the register (otherwise,
7372 i386_scale will complain) */
7373 if (cur_token.code == '+' || cur_token.code == '-')
7374 {
7375 char *str, sign = cur_token.code;
7376 intel_match_token (cur_token.code);
7377 if (cur_token.code != T_CONST)
7378 {
7379 as_bad (_("Syntax error: Expecting a constant, got `%s'"),
7380 cur_token.str);
7381 return 0;
7382 }
7383 str = (char *) xmalloc (strlen (cur_token.str) + 2);
7384 strcpy (str + 1, cur_token.str);
7385 *str = sign;
7386 if (!i386_scale (str))
7387 return 0;
7388 free (str);
7389 }
7390 else if (!i386_scale (cur_token.str))
7391 return 0;
7392 intel_match_token (cur_token.code);
7393 }
7394
7395 /* No scaling. If this is a memory operand, the register is either a
7396 base register (first occurrence) or an index register (second
7397 occurrence). */
7398 else if (intel_parser.in_bracket)
7399 {
7400
7401 if (!i.base_reg)
7402 i.base_reg = reg;
7403 else if (!i.index_reg)
7404 i.index_reg = reg;
7405 else
7406 {
7407 as_bad (_("Too many register references in memory operand"));
7408 return 0;
7409 }
7410
7411 i.types[this_operand] |= BaseIndex;
7412 }
7413
7414 /* It's neither base nor index. */
7415 else if (!intel_parser.in_offset && !intel_parser.is_mem)
7416 {
7417 i.types[this_operand] |= reg->reg_type & ~BaseIndex;
7418 i.op[this_operand].regs = reg;
7419 i.reg_operands++;
7420 }
7421 else
7422 {
7423 as_bad (_("Invalid use of register"));
7424 return 0;
7425 }
7426
7427 /* Since registers are not part of the displacement string (except
7428 when we're parsing offset operands), we may need to remove any
7429 preceding '+' from the displacement string. */
7430 if (*intel_parser.disp != '\0'
7431 && !intel_parser.in_offset)
7432 {
7433 char *s = intel_parser.disp;
7434 s += strlen (s) - 1;
7435 if (*s == '+')
7436 *s = '\0';
7437 }
7438
7439 return 1;
7440 }
7441
7442 /* e11 BYTE
7443 | WORD
7444 | DWORD
7445 | FWORD
7446 | QWORD
7447 | TBYTE
7448 | OWORD
7449 | XMMWORD */
7450 case T_BYTE:
7451 case T_WORD:
7452 case T_DWORD:
7453 case T_FWORD:
7454 case T_QWORD:
7455 case T_TBYTE:
7456 case T_XMMWORD:
7457 intel_match_token (cur_token.code);
7458
7459 if (cur_token.code == T_PTR)
7460 return 1;
7461
7462 /* It must have been an identifier. */
7463 intel_putback_token ();
7464 cur_token.code = T_ID;
7465 /* FALLTHRU */
7466
7467 /* e11 id
7468 | constant */
7469 case T_ID:
7470 if (!intel_parser.in_offset && intel_parser.is_mem <= 0)
7471 {
7472 symbolS *symbolP;
7473
7474 /* The identifier represents a memory reference only if it's not
7475 preceded by an offset modifier and if it's not an equate. */
7476 symbolP = symbol_find(cur_token.str);
7477 if (!symbolP || S_GET_SEGMENT(symbolP) != absolute_section)
7478 intel_parser.is_mem = 1;
7479 }
7480 /* FALLTHRU */
7481
7482 case T_CONST:
7483 case '-':
7484 case '+':
7485 {
7486 char *save_str, sign = 0;
7487
7488 /* Allow constants that start with `+' or `-'. */
7489 if (cur_token.code == '-' || cur_token.code == '+')
7490 {
7491 sign = cur_token.code;
7492 intel_match_token (cur_token.code);
7493 if (cur_token.code != T_CONST)
7494 {
7495 as_bad (_("Syntax error: Expecting a constant, got `%s'"),
7496 cur_token.str);
7497 return 0;
7498 }
7499 }
7500
7501 save_str = (char *) xmalloc (strlen (cur_token.str) + 2);
7502 strcpy (save_str + !!sign, cur_token.str);
7503 if (sign)
7504 *save_str = sign;
7505
7506 /* Get the next token to check for register scaling. */
7507 intel_match_token (cur_token.code);
7508
7509 /* Check if this constant is a scaling factor for an
7510 index register. */
7511 if (cur_token.code == '*')
7512 {
7513 if (intel_match_token ('*') && cur_token.code == T_REG)
7514 {
7515 const reg_entry *reg = cur_token.reg;
7516
7517 if (!intel_parser.in_bracket)
7518 {
7519 as_bad (_("Register scaling only allowed "
7520 "in memory operands"));
7521 return 0;
7522 }
7523
7524 /* Disallow things like [1*si].
7525 sp and esp are invalid as index. */
7526 if (reg->reg_type & Reg16)
7527 reg = i386_regtab + REGNAM_AX + 4;
7528 else if (i.index_reg)
7529 reg = i386_regtab + REGNAM_EAX + 4;
7530
7531 /* The constant is followed by `* reg', so it must be
7532 a valid scale. */
7533 i.index_reg = reg;
7534 i.types[this_operand] |= BaseIndex;
7535
7536 /* Set the scale after setting the register (otherwise,
7537 i386_scale will complain) */
7538 if (!i386_scale (save_str))
7539 return 0;
7540 intel_match_token (T_REG);
7541
7542 /* Since registers are not part of the displacement
7543 string, we may need to remove any preceding '+' from
7544 the displacement string. */
7545 if (*intel_parser.disp != '\0')
7546 {
7547 char *s = intel_parser.disp;
7548 s += strlen (s) - 1;
7549 if (*s == '+')
7550 *s = '\0';
7551 }
7552
7553 free (save_str);
7554
7555 return 1;
7556 }
7557
7558 /* The constant was not used for register scaling. Since we have
7559 already consumed the token following `*' we now need to put it
7560 back in the stream. */
7561 intel_putback_token ();
7562 }
7563
7564 /* Add the constant to the displacement string. */
7565 strcat (intel_parser.disp, save_str);
7566 free (save_str);
7567
7568 return 1;
7569 }
7570 }
7571
7572 as_bad (_("Unrecognized token '%s'"), cur_token.str);
7573 return 0;
7574 }
7575
7576 /* Match the given token against cur_token. If they match, read the next
7577 token from the operand string. */
7578 static int
7579 intel_match_token (int code)
7580 {
7581 if (cur_token.code == code)
7582 {
7583 intel_get_token ();
7584 return 1;
7585 }
7586 else
7587 {
7588 as_bad (_("Unexpected token `%s'"), cur_token.str);
7589 return 0;
7590 }
7591 }
7592
7593 /* Read a new token from intel_parser.op_string and store it in cur_token. */
7594 static void
7595 intel_get_token (void)
7596 {
7597 char *end_op;
7598 const reg_entry *reg;
7599 struct intel_token new_token;
7600
7601 new_token.code = T_NIL;
7602 new_token.reg = NULL;
7603 new_token.str = NULL;
7604
7605 /* Free the memory allocated to the previous token and move
7606 cur_token to prev_token. */
7607 if (prev_token.str)
7608 free (prev_token.str);
7609
7610 prev_token = cur_token;
7611
7612 /* Skip whitespace. */
7613 while (is_space_char (*intel_parser.op_string))
7614 intel_parser.op_string++;
7615
7616 /* Return an empty token if we find nothing else on the line. */
7617 if (*intel_parser.op_string == '\0')
7618 {
7619 cur_token = new_token;
7620 return;
7621 }
7622
7623 /* The new token cannot be larger than the remainder of the operand
7624 string. */
7625 new_token.str = (char *) xmalloc (strlen (intel_parser.op_string) + 1);
7626 new_token.str[0] = '\0';
7627
7628 if (strchr ("0123456789", *intel_parser.op_string))
7629 {
7630 char *p = new_token.str;
7631 char *q = intel_parser.op_string;
7632 new_token.code = T_CONST;
7633
7634 /* Allow any kind of identifier char to encompass floating point and
7635 hexadecimal numbers. */
7636 while (is_identifier_char (*q))
7637 *p++ = *q++;
7638 *p = '\0';
7639
7640 /* Recognize special symbol names [0-9][bf]. */
7641 if (strlen (intel_parser.op_string) == 2
7642 && (intel_parser.op_string[1] == 'b'
7643 || intel_parser.op_string[1] == 'f'))
7644 new_token.code = T_ID;
7645 }
7646
7647 else if ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL)
7648 {
7649 size_t len = end_op - intel_parser.op_string;
7650
7651 new_token.code = T_REG;
7652 new_token.reg = reg;
7653
7654 memcpy (new_token.str, intel_parser.op_string, len);
7655 new_token.str[len] = '\0';
7656 }
7657
7658 else if (is_identifier_char (*intel_parser.op_string))
7659 {
7660 char *p = new_token.str;
7661 char *q = intel_parser.op_string;
7662
7663 /* A '.' or '$' followed by an identifier char is an identifier.
7664 Otherwise, it's operator '.' followed by an expression. */
7665 if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
7666 {
7667 new_token.code = '.';
7668 new_token.str[0] = '.';
7669 new_token.str[1] = '\0';
7670 }
7671 else
7672 {
7673 while (is_identifier_char (*q) || *q == '@')
7674 *p++ = *q++;
7675 *p = '\0';
7676
7677 if (strcasecmp (new_token.str, "NOT") == 0)
7678 new_token.code = '~';
7679
7680 else if (strcasecmp (new_token.str, "MOD") == 0)
7681 new_token.code = '%';
7682
7683 else if (strcasecmp (new_token.str, "AND") == 0)
7684 new_token.code = '&';
7685
7686 else if (strcasecmp (new_token.str, "OR") == 0)
7687 new_token.code = '|';
7688
7689 else if (strcasecmp (new_token.str, "XOR") == 0)
7690 new_token.code = '^';
7691
7692 else if (strcasecmp (new_token.str, "SHL") == 0)
7693 new_token.code = T_SHL;
7694
7695 else if (strcasecmp (new_token.str, "SHR") == 0)
7696 new_token.code = T_SHR;
7697
7698 else if (strcasecmp (new_token.str, "BYTE") == 0)
7699 new_token.code = T_BYTE;
7700
7701 else if (strcasecmp (new_token.str, "WORD") == 0)
7702 new_token.code = T_WORD;
7703
7704 else if (strcasecmp (new_token.str, "DWORD") == 0)
7705 new_token.code = T_DWORD;
7706
7707 else if (strcasecmp (new_token.str, "FWORD") == 0)
7708 new_token.code = T_FWORD;
7709
7710 else if (strcasecmp (new_token.str, "QWORD") == 0)
7711 new_token.code = T_QWORD;
7712
7713 else if (strcasecmp (new_token.str, "TBYTE") == 0
7714 /* XXX remove (gcc still uses it) */
7715 || strcasecmp (new_token.str, "XWORD") == 0)
7716 new_token.code = T_TBYTE;
7717
7718 else if (strcasecmp (new_token.str, "XMMWORD") == 0
7719 || strcasecmp (new_token.str, "OWORD") == 0)
7720 new_token.code = T_XMMWORD;
7721
7722 else if (strcasecmp (new_token.str, "PTR") == 0)
7723 new_token.code = T_PTR;
7724
7725 else if (strcasecmp (new_token.str, "SHORT") == 0)
7726 new_token.code = T_SHORT;
7727
7728 else if (strcasecmp (new_token.str, "OFFSET") == 0)
7729 {
7730 new_token.code = T_OFFSET;
7731
7732 /* ??? This is not mentioned in the MASM grammar but gcc
7733 makes use of it with -mintel-syntax. OFFSET may be
7734 followed by FLAT: */
7735 if (strncasecmp (q, " FLAT:", 6) == 0)
7736 strcat (new_token.str, " FLAT:");
7737 }
7738
7739 /* ??? This is not mentioned in the MASM grammar. */
7740 else if (strcasecmp (new_token.str, "FLAT") == 0)
7741 {
7742 new_token.code = T_OFFSET;
7743 if (*q == ':')
7744 strcat (new_token.str, ":");
7745 else
7746 as_bad (_("`:' expected"));
7747 }
7748
7749 else
7750 new_token.code = T_ID;
7751 }
7752 }
7753
7754 else if (strchr ("+-/*%|&^:[]()~", *intel_parser.op_string))
7755 {
7756 new_token.code = *intel_parser.op_string;
7757 new_token.str[0] = *intel_parser.op_string;
7758 new_token.str[1] = '\0';
7759 }
7760
7761 else if (strchr ("<>", *intel_parser.op_string)
7762 && *intel_parser.op_string == *(intel_parser.op_string + 1))
7763 {
7764 new_token.code = *intel_parser.op_string == '<' ? T_SHL : T_SHR;
7765 new_token.str[0] = *intel_parser.op_string;
7766 new_token.str[1] = *intel_parser.op_string;
7767 new_token.str[2] = '\0';
7768 }
7769
7770 else
7771 as_bad (_("Unrecognized token `%s'"), intel_parser.op_string);
7772
7773 intel_parser.op_string += strlen (new_token.str);
7774 cur_token = new_token;
7775 }
7776
7777 /* Put cur_token back into the token stream and make cur_token point to
7778 prev_token. */
7779 static void
7780 intel_putback_token (void)
7781 {
7782 if (cur_token.code != T_NIL)
7783 {
7784 intel_parser.op_string -= strlen (cur_token.str);
7785 free (cur_token.str);
7786 }
7787 cur_token = prev_token;
7788
7789 /* Forget prev_token. */
7790 prev_token.code = T_NIL;
7791 prev_token.reg = NULL;
7792 prev_token.str = NULL;
7793 }
7794
7795 int
7796 tc_x86_regname_to_dw2regnum (char *regname)
7797 {
7798 unsigned int regnum;
7799 unsigned int regnames_count;
7800 static const char *const regnames_32[] =
7801 {
7802 "eax", "ecx", "edx", "ebx",
7803 "esp", "ebp", "esi", "edi",
7804 "eip", "eflags", NULL,
7805 "st0", "st1", "st2", "st3",
7806 "st4", "st5", "st6", "st7",
7807 NULL, NULL,
7808 "xmm0", "xmm1", "xmm2", "xmm3",
7809 "xmm4", "xmm5", "xmm6", "xmm7",
7810 "mm0", "mm1", "mm2", "mm3",
7811 "mm4", "mm5", "mm6", "mm7",
7812 "fcw", "fsw", "mxcsr",
7813 "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL,
7814 "tr", "ldtr"
7815 };
7816 static const char *const regnames_64[] =
7817 {
7818 "rax", "rdx", "rcx", "rbx",
7819 "rsi", "rdi", "rbp", "rsp",
7820 "r8", "r9", "r10", "r11",
7821 "r12", "r13", "r14", "r15",
7822 "rip",
7823 "xmm0", "xmm1", "xmm2", "xmm3",
7824 "xmm4", "xmm5", "xmm6", "xmm7",
7825 "xmm8", "xmm9", "xmm10", "xmm11",
7826 "xmm12", "xmm13", "xmm14", "xmm15",
7827 "st0", "st1", "st2", "st3",
7828 "st4", "st5", "st6", "st7",
7829 "mm0", "mm1", "mm2", "mm3",
7830 "mm4", "mm5", "mm6", "mm7",
7831 "rflags",
7832 "es", "cs", "ss", "ds", "fs", "gs", NULL, NULL,
7833 "fs.base", "gs.base", NULL, NULL,
7834 "tr", "ldtr",
7835 "mxcsr", "fcw", "fsw"
7836 };
7837 const char *const *regnames;
7838
7839 if (flag_code == CODE_64BIT)
7840 {
7841 regnames = regnames_64;
7842 regnames_count = ARRAY_SIZE (regnames_64);
7843 }
7844 else
7845 {
7846 regnames = regnames_32;
7847 regnames_count = ARRAY_SIZE (regnames_32);
7848 }
7849
7850 for (regnum = 0; regnum < regnames_count; regnum++)
7851 if (regnames[regnum] != NULL
7852 && strcmp (regname, regnames[regnum]) == 0)
7853 return regnum;
7854
7855 return -1;
7856 }
7857
7858 void
7859 tc_x86_frame_initial_instructions (void)
7860 {
7861 static unsigned int sp_regno;
7862
7863 if (!sp_regno)
7864 sp_regno = tc_x86_regname_to_dw2regnum (flag_code == CODE_64BIT
7865 ? "rsp" : "esp");
7866
7867 cfi_add_CFA_def_cfa (sp_regno, -x86_cie_data_alignment);
7868 cfi_add_CFA_offset (x86_dwarf2_return_column, x86_cie_data_alignment);
7869 }
7870
7871 int
7872 i386_elf_section_type (const char *str, size_t len)
7873 {
7874 if (flag_code == CODE_64BIT
7875 && len == sizeof ("unwind") - 1
7876 && strncmp (str, "unwind", 6) == 0)
7877 return SHT_X86_64_UNWIND;
7878
7879 return -1;
7880 }
7881
7882 #ifdef TE_PE
7883 void
7884 tc_pe_dwarf2_emit_offset (symbolS *symbol, unsigned int size)
7885 {
7886 expressionS expr;
7887
7888 expr.X_op = O_secrel;
7889 expr.X_add_symbol = symbol;
7890 expr.X_add_number = 0;
7891 emit_expr (&expr, size);
7892 }
7893 #endif
7894
7895 #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
7896 /* For ELF on x86-64, add support for SHF_X86_64_LARGE. */
7897
7898 int
7899 x86_64_section_letter (int letter, char **ptr_msg)
7900 {
7901 if (flag_code == CODE_64BIT)
7902 {
7903 if (letter == 'l')
7904 return SHF_X86_64_LARGE;
7905
7906 *ptr_msg = _("Bad .section directive: want a,l,w,x,M,S,G,T in string");
7907 }
7908 else
7909 *ptr_msg = _("Bad .section directive: want a,w,x,M,S,G,T in string");
7910 return -1;
7911 }
7912
7913 int
7914 x86_64_section_word (char *str, size_t len)
7915 {
7916 if (len == 5 && flag_code == CODE_64BIT && CONST_STRNEQ (str, "large"))
7917 return SHF_X86_64_LARGE;
7918
7919 return -1;
7920 }
7921
7922 static void
7923 handle_large_common (int small ATTRIBUTE_UNUSED)
7924 {
7925 if (flag_code != CODE_64BIT)
7926 {
7927 s_comm_internal (0, elf_common_parse);
7928 as_warn (_(".largecomm supported only in 64bit mode, producing .comm"));
7929 }
7930 else
7931 {
7932 static segT lbss_section;
7933 asection *saved_com_section_ptr = elf_com_section_ptr;
7934 asection *saved_bss_section = bss_section;
7935
7936 if (lbss_section == NULL)
7937 {
7938 flagword applicable;
7939 segT seg = now_seg;
7940 subsegT subseg = now_subseg;
7941
7942 /* The .lbss section is for local .largecomm symbols. */
7943 lbss_section = subseg_new (".lbss", 0);
7944 applicable = bfd_applicable_section_flags (stdoutput);
7945 bfd_set_section_flags (stdoutput, lbss_section,
7946 applicable & SEC_ALLOC);
7947 seg_info (lbss_section)->bss = 1;
7948
7949 subseg_set (seg, subseg);
7950 }
7951
7952 elf_com_section_ptr = &_bfd_elf_large_com_section;
7953 bss_section = lbss_section;
7954
7955 s_comm_internal (0, elf_common_parse);
7956
7957 elf_com_section_ptr = saved_com_section_ptr;
7958 bss_section = saved_bss_section;
7959 }
7960 }
7961 #endif /* OBJ_ELF || OBJ_MAYBE_ELF */
This page took 0.19375 seconds and 4 git commands to generate.